Home | History | Annotate | Line # | Download | only in routed
      1 /*	$NetBSD: table.c,v 1.28 2018/02/06 09:33:07 mrg Exp $	*/
      2 
      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. All advertising materials mentioning features or use of this software
     16  *    must display the following acknowledgment:
     17  *	This product includes software developed by the University of
     18  *	California, Berkeley and its contributors.
     19  * 4. Neither the name of the University nor the names of its contributors
     20  *    may be used to endorse or promote products derived from this software
     21  *    without specific prior written permission.
     22  *
     23  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
     24  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
     25  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
     26  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
     27  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
     28  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
     29  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
     30  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
     31  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
     32  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
     33  * SUCH DAMAGE.
     34  */
     35 
     36 #include "defs.h"
     37 
     38 #ifdef __NetBSD__
     39 __RCSID("$NetBSD: table.c,v 1.28 2018/02/06 09:33:07 mrg Exp $");
     40 #elif defined(__FreeBSD__)
     41 __RCSID("$FreeBSD$");
     42 #else
     43 __RCSID("Revision: 2.27 ");
     44 #ident "Revision: 2.27 "
     45 #endif
     46 
     47 static struct rt_spare *rts_better(struct rt_entry *);
     48 static struct rt_spare rts_empty = {0,0,0,HOPCNT_INFINITY,0,0,0};
     49 static void  set_need_flash(void);
     50 #ifdef _HAVE_SIN_LEN
     51 static void masktrim(struct sockaddr_in *ap);
     52 #else
     53 static void masktrim(struct sockaddr_in_new *ap);
     54 #endif
     55 
     56 
     57 struct radix_node_head *rhead;		/* root of the radix tree */
     58 
     59 int	need_flash = 1;			/* flash update needed
     60 					 * start =1 to suppress the 1st
     61 					 */
     62 
     63 struct timeval age_timer;		/* next check of old routes */
     64 struct timeval need_kern = {		/* need to update kernel table */
     65 	EPOCH+MIN_WAITTIME-1, 0
     66 };
     67 
     68 int	stopint;
     69 
     70 int	total_routes;
     71 
     72 /* zap any old routes through this gateway */
     73 naddr	age_bad_gate;
     74 
     75 
     76 /* It is desirable to "aggregate" routes, to combine differing routes of
     77  * the same metric and next hop into a common route with a smaller netmask
     78  * or to suppress redundant routes, routes that add no information to
     79  * routes with smaller netmasks.
     80  *
     81  * A route is redundant if and only if any and all routes with smaller
     82  * but matching netmasks and nets are the same.  Since routes are
     83  * kept sorted in the radix tree, redundant routes always come second.
     84  *
     85  * There are two kinds of aggregations.  First, two routes of the same bit
     86  * mask and differing only in the least significant bit of the network
     87  * number can be combined into a single route with a coarser mask.
     88  *
     89  * Second, a route can be suppressed in favor of another route with a more
     90  * coarse mask provided no incompatible routes with intermediate masks
     91  * are present.  The second kind of aggregation involves suppressing routes.
     92  * A route must not be suppressed if an incompatible route exists with
     93  * an intermediate mask, since the suppressed route would be covered
     94  * by the intermediate.
     95  *
     96  * This code relies on the radix tree walk encountering routes
     97  * sorted first by address, with the smallest address first.
     98  */
     99 
    100 struct ag_info ag_slots[NUM_AG_SLOTS], *ag_avail, *ag_corsest, *ag_finest;
    101 
    102 /* #define DEBUG_AG */
    103 #ifdef DEBUG_AG
    104 #define CHECK_AG() {int acnt = 0; struct ag_info *cag;		\
    105 	for (cag = ag_avail; cag != 0; cag = cag->ag_fine)	\
    106 		acnt++;						\
    107 	for (cag = ag_corsest; cag != 0; cag = cag->ag_fine)	\
    108 		acnt++;						\
    109 	if (acnt != NUM_AG_SLOTS) {				\
    110 		(void)fflush(stderr);				\
    111 		abort();					\
    112 	}							\
    113 }
    114 #else
    115 #define CHECK_AG()
    116 #endif
    117 
    118 
    119 /* Output the contents of an aggregation table slot.
    120  *	This function must always be immediately followed with the deletion
    121  *	of the target slot.
    122  */
    123 static void
    124 ag_out(struct ag_info *ag,
    125 	 void (*out)(struct ag_info *))
    126 {
    127 	struct ag_info *ag_cors;
    128 	naddr bit;
    129 
    130 
    131 	/* Forget it if this route should not be output for split-horizon. */
    132 	if (ag->ag_state & AGS_SPLIT_HZ)
    133 		return;
    134 
    135 	/* If we output both the even and odd twins, then the immediate parent,
    136 	 * if it is present, is redundant, unless the parent manages to
    137 	 * aggregate into something coarser.
    138 	 * On successive calls, this code detects the even and odd twins,
    139 	 * and marks the parent.
    140 	 *
    141 	 * Note that the order in which the radix tree code emits routes
    142 	 * ensures that the twins are seen before the parent is emitted.
    143 	 */
    144 	ag_cors = ag->ag_cors;
    145 	if (ag_cors != 0
    146 	    && ag_cors->ag_mask == ag->ag_mask<<1
    147 	    && ag_cors->ag_dst_h == (ag->ag_dst_h & ag_cors->ag_mask)) {
    148 		ag_cors->ag_state |= ((ag_cors->ag_dst_h == ag->ag_dst_h)
    149 				      ? AGS_REDUN0
    150 				      : AGS_REDUN1);
    151 	}
    152 
    153 	/* Skip it if this route is itself redundant.
    154 	 *
    155 	 * It is ok to change the contents of the slot here, since it is
    156 	 * always deleted next.
    157 	 */
    158 	if (ag->ag_state & AGS_REDUN0) {
    159 		if (ag->ag_state & AGS_REDUN1)
    160 			return;		/* quit if fully redundant */
    161 		/* make it finer if it is half-redundant */
    162 		bit = (-ag->ag_mask) >> 1;
    163 		ag->ag_dst_h |= bit;
    164 		ag->ag_mask |= bit;
    165 
    166 	} else if (ag->ag_state & AGS_REDUN1) {
    167 		/* make it finer if it is half-redundant */
    168 		bit = (-ag->ag_mask) >> 1;
    169 		ag->ag_mask |= bit;
    170 	}
    171 	out(ag);
    172 }
    173 
    174 
    175 static void
    176 ag_del(struct ag_info *ag)
    177 {
    178 	CHECK_AG();
    179 
    180 	if (ag->ag_cors == 0)
    181 		ag_corsest = ag->ag_fine;
    182 	else
    183 		ag->ag_cors->ag_fine = ag->ag_fine;
    184 
    185 	if (ag->ag_fine == 0)
    186 		ag_finest = ag->ag_cors;
    187 	else
    188 		ag->ag_fine->ag_cors = ag->ag_cors;
    189 
    190 	ag->ag_fine = ag_avail;
    191 	ag_avail = ag;
    192 
    193 	CHECK_AG();
    194 }
    195 
    196 
    197 /* Flush routes waiting for aggregation.
    198  *	This must not suppress a route unless it is known that among all
    199  *	routes with coarser masks that match it, the one with the longest
    200  *	mask is appropriate.  This is ensured by scanning the routes
    201  *	in lexical order, and with the most restrictive mask first
    202  *	among routes to the same destination.
    203  */
    204 void
    205 ag_flush(naddr lim_dst_h,		/* flush routes to here */
    206 	 naddr lim_mask,		/* matching this mask */
    207 	 void (*out)(struct ag_info *))
    208 {
    209 	struct ag_info *ag, *ag_cors;
    210 	naddr dst_h;
    211 
    212 
    213 	for (ag = ag_finest;
    214 	     ag != 0 && ag->ag_mask >= lim_mask;
    215 	     ag = ag_cors) {
    216 		ag_cors = ag->ag_cors;
    217 
    218 		/* work on only the specified routes */
    219 		dst_h = ag->ag_dst_h;
    220 		if ((dst_h & lim_mask) != lim_dst_h)
    221 			continue;
    222 
    223 		if (!(ag->ag_state & AGS_SUPPRESS))
    224 			ag_out(ag, out);
    225 
    226 		else for ( ; ; ag_cors = ag_cors->ag_cors) {
    227 			/* Look for a route that can suppress the
    228 			 * current route */
    229 			if (ag_cors == 0) {
    230 				/* failed, so output it and look for
    231 				 * another route to work on
    232 				 */
    233 				ag_out(ag, out);
    234 				break;
    235 			}
    236 
    237 			if ((dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h) {
    238 				/* We found a route with a coarser mask that
    239 				 * aggregates the current target.
    240 				 *
    241 				 * If it has a different next hop, it
    242 				 * cannot replace the target, so output
    243 				 * the target.
    244 				 */
    245 				if (ag->ag_gate != ag_cors->ag_gate
    246 				    && !(ag->ag_state & AGS_FINE_GATE)
    247 				    && !(ag_cors->ag_state & AGS_CORS_GATE)) {
    248 					ag_out(ag, out);
    249 					break;
    250 				}
    251 
    252 				/* If the coarse route has a good enough
    253 				 * metric, it suppresses the target.
    254 				 * If the suppressed target was redundant,
    255 				 * then mark the suppressor redundant.
    256 				 */
    257 				if (ag_cors->ag_pref <= ag->ag_pref) {
    258 				    if (AG_IS_REDUN(ag->ag_state)
    259 					&& ag_cors->ag_mask==ag->ag_mask<<1) {
    260 					if (ag_cors->ag_dst_h == dst_h)
    261 					    ag_cors->ag_state |= AGS_REDUN0;
    262 					else
    263 					    ag_cors->ag_state |= AGS_REDUN1;
    264 				    }
    265 				    if (ag->ag_tag != ag_cors->ag_tag)
    266 					    ag_cors->ag_tag = 0;
    267 				    if (ag->ag_nhop != ag_cors->ag_nhop)
    268 					    ag_cors->ag_nhop = 0;
    269 				    break;
    270 				}
    271 			}
    272 		}
    273 
    274 		/* That route has either been output or suppressed */
    275 		ag_cors = ag->ag_cors;
    276 		ag_del(ag);
    277 	}
    278 
    279 	CHECK_AG();
    280 }
    281 
    282 
    283 /* Try to aggregate a route with previous routes.
    284  */
    285 void
    286 ag_check(naddr	dst,
    287 	 naddr	mask,
    288 	 naddr	gate,
    289 	 naddr	nhop,
    290 	 char	metric,
    291 	 char	pref,
    292 	 u_int	new_seqno,
    293 	 u_short tag,
    294 	 u_short state,
    295 	 void (*out)(struct ag_info *))	/* output using this */
    296 {
    297 	struct ag_info *ag, *nag, *ag_cors;
    298 	naddr xaddr;
    299 	int x;
    300 
    301 	dst = ntohl(dst);
    302 
    303 	/* Punt non-contiguous subnet masks.
    304 	 *
    305 	 * (X & -X) contains a single bit if and only if X is a power of 2.
    306 	 * (X + (X & -X)) == 0 if and only if X is a power of 2.
    307 	 */
    308 	if ((mask & -mask) + mask != 0) {
    309 		struct ag_info nc_ag;
    310 
    311 		nc_ag.ag_dst_h = dst;
    312 		nc_ag.ag_mask = mask;
    313 		nc_ag.ag_gate = gate;
    314 		nc_ag.ag_nhop = nhop;
    315 		nc_ag.ag_metric = metric;
    316 		nc_ag.ag_pref = pref;
    317 		nc_ag.ag_tag = tag;
    318 		nc_ag.ag_state = state;
    319 		nc_ag.ag_seqno = new_seqno;
    320 		out(&nc_ag);
    321 		return;
    322 	}
    323 
    324 	/* Search for the right slot in the aggregation table.
    325 	 */
    326 	ag_cors = 0;
    327 	ag = ag_corsest;
    328 	while (ag != 0) {
    329 		if (ag->ag_mask >= mask)
    330 			break;
    331 
    332 		/* Suppress old routes (i.e. combine with compatible routes
    333 		 * with coarser masks) as we look for the right slot in the
    334 		 * aggregation table for the new route.
    335 		 * A route to an address less than the current destination
    336 		 * will not be affected by the current route or any route
    337 		 * seen hereafter.  That means it is safe to suppress it.
    338 		 * This check keeps poor routes (e.g. with large hop counts)
    339 		 * from preventing suppression of finer routes.
    340 		 */
    341 		if (ag_cors != 0
    342 		    && ag->ag_dst_h < dst
    343 		    && (ag->ag_state & AGS_SUPPRESS)
    344 		    && ag_cors->ag_pref <= ag->ag_pref
    345 		    && (ag->ag_dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h
    346 		    && (ag_cors->ag_gate == ag->ag_gate
    347 			|| (ag->ag_state & AGS_FINE_GATE)
    348 			|| (ag_cors->ag_state & AGS_CORS_GATE))) {
    349 			/*  If the suppressed target was redundant,
    350 			 * then mark the suppressor redundant.
    351 			 */
    352 			if (AG_IS_REDUN(ag->ag_state)
    353 			    && ag_cors->ag_mask == ag->ag_mask<<1) {
    354 				if (ag_cors->ag_dst_h == dst)
    355 					ag_cors->ag_state |= AGS_REDUN0;
    356 				else
    357 					ag_cors->ag_state |= AGS_REDUN1;
    358 			}
    359 			if (ag->ag_tag != ag_cors->ag_tag)
    360 				ag_cors->ag_tag = 0;
    361 			if (ag->ag_nhop != ag_cors->ag_nhop)
    362 				ag_cors->ag_nhop = 0;
    363 			ag_del(ag);
    364 			CHECK_AG();
    365 		} else {
    366 			ag_cors = ag;
    367 		}
    368 		ag = ag_cors->ag_fine;
    369 	}
    370 
    371 	/* If we find the even/odd twin of the new route, and if the
    372 	 * masks and so forth are equal, we can aggregate them.
    373 	 * We can probably promote one of the pair.
    374 	 *
    375 	 * Since the routes are encountered in lexical order,
    376 	 * the new route must be odd.  However, the second or later
    377 	 * times around this loop, it could be the even twin promoted
    378 	 * from the even/odd pair of twins of the finer route.
    379 	 */
    380 	while (ag != 0
    381 	       && ag->ag_mask == mask
    382 	       && ((ag->ag_dst_h ^ dst) & (mask<<1)) == 0) {
    383 
    384 		/* Here we know the target route and the route in the current
    385 		 * slot have the same netmasks and differ by at most the
    386 		 * last bit.  They are either for the same destination, or
    387 		 * for an even/odd pair of destinations.
    388 		 */
    389 		if (ag->ag_dst_h == dst) {
    390 			/* We have two routes to the same destination.
    391 			 * Routes are encountered in lexical order, so a
    392 			 * route is never promoted until the parent route is
    393 			 * already present.  So we know that the new route is
    394 			 * a promoted (or aggregated) pair and the route
    395 			 * already in the slot is the explicit route.
    396 			 *
    397 			 * Prefer the best route if their metrics differ,
    398 			 * or the aggregated one if not, following a sort
    399 			 * of longest-match rule.
    400 			 */
    401 			if (pref <= ag->ag_pref) {
    402 				ag->ag_gate = gate;
    403 				ag->ag_nhop = nhop;
    404 				ag->ag_tag = tag;
    405 				ag->ag_metric = metric;
    406 				ag->ag_pref = pref;
    407 				if (ag->ag_seqno < new_seqno)
    408 					ag->ag_seqno = new_seqno;
    409 				x = ag->ag_state;
    410 				ag->ag_state = state;
    411 				state = x;
    412 			}
    413 
    414 			/* Some bits are set if they are set on either route,
    415 			 * except when the route is for an interface.
    416 			 */
    417 			if (!(ag->ag_state & AGS_IF))
    418 				ag->ag_state |= (state & (AGS_AGGREGATE_EITHER
    419 							| AGS_REDUN0
    420 							| AGS_REDUN1));
    421 			return;
    422 		}
    423 
    424 		/* If one of the routes can be promoted and the other can
    425 		 * be suppressed, it may be possible to combine them or
    426 		 * worthwhile to promote one.
    427 		 *
    428 		 * Any route that can be promoted is always
    429 		 * marked to be eligible to be suppressed.
    430 		 */
    431 		if (!((state & AGS_AGGREGATE)
    432 		      && (ag->ag_state & AGS_SUPPRESS))
    433 		    && !((ag->ag_state & AGS_AGGREGATE)
    434 			 && (state & AGS_SUPPRESS)))
    435 			break;
    436 
    437 		/* A pair of even/odd twin routes can be combined
    438 		 * if either is redundant, or if they are via the
    439 		 * same gateway and have the same metric.
    440 		 */
    441 		if (AG_IS_REDUN(ag->ag_state)
    442 		    || AG_IS_REDUN(state)
    443 		    || (ag->ag_gate == gate
    444 			&& ag->ag_pref == pref
    445 			&& (state & ag->ag_state & AGS_AGGREGATE) != 0)) {
    446 
    447 			/* We have both the even and odd pairs.
    448 			 * Since the routes are encountered in order,
    449 			 * the route in the slot must be the even twin.
    450 			 *
    451 			 * Combine and promote (aggregate) the pair of routes.
    452 			 */
    453 			if (new_seqno < ag->ag_seqno)
    454 				new_seqno = ag->ag_seqno;
    455 			if (!AG_IS_REDUN(state))
    456 				state &= ~AGS_REDUN1;
    457 			if (AG_IS_REDUN(ag->ag_state))
    458 				state |= AGS_REDUN0;
    459 			else
    460 				state &= ~AGS_REDUN0;
    461 			state |= (ag->ag_state & AGS_AGGREGATE_EITHER);
    462 			if (ag->ag_tag != tag)
    463 				tag = 0;
    464 			if (ag->ag_nhop != nhop)
    465 				nhop = 0;
    466 
    467 			/* Get rid of the even twin that was already
    468 			 * in the slot.
    469 			 */
    470 			ag_del(ag);
    471 
    472 		} else if (ag->ag_pref >= pref
    473 			   && (ag->ag_state & AGS_AGGREGATE)) {
    474 			/* If we cannot combine the pair, maybe the route
    475 			 * with the worse metric can be promoted.
    476 			 *
    477 			 * Promote the old, even twin, by giving its slot
    478 			 * in the table to the new, odd twin.
    479 			 */
    480 			ag->ag_dst_h = dst;
    481 
    482 			xaddr = ag->ag_gate;
    483 			ag->ag_gate = gate;
    484 			gate = xaddr;
    485 
    486 			xaddr = ag->ag_nhop;
    487 			ag->ag_nhop = nhop;
    488 			nhop = xaddr;
    489 
    490 			x = ag->ag_tag;
    491 			ag->ag_tag = tag;
    492 			tag = x;
    493 
    494 			/* The promoted route is even-redundant only if the
    495 			 * even twin was fully redundant.  It is not
    496 			 * odd-redundant because the odd-twin will still be
    497 			 * in the table.
    498 			 */
    499 			x = ag->ag_state;
    500 			if (!AG_IS_REDUN(x))
    501 				x &= ~AGS_REDUN0;
    502 			x &= ~AGS_REDUN1;
    503 			ag->ag_state = state;
    504 			state = x;
    505 
    506 			x = ag->ag_metric;
    507 			ag->ag_metric = metric;
    508 			metric = x;
    509 
    510 			x = ag->ag_pref;
    511 			ag->ag_pref = pref;
    512 			pref = x;
    513 
    514 			/* take the newest sequence number */
    515 			if (new_seqno <= ag->ag_seqno)
    516 				new_seqno = ag->ag_seqno;
    517 			else
    518 				ag->ag_seqno = new_seqno;
    519 
    520 		} else {
    521 			if (!(state & AGS_AGGREGATE))
    522 				break;	/* cannot promote either twin */
    523 
    524 			/* Promote the new, odd twin by shaving its
    525 			 * mask and address.
    526 			 * The promoted route is odd-redundant only if the
    527 			 * odd twin was fully redundant.  It is not
    528 			 * even-redundant because the even twin is still in
    529 			 * the table.
    530 			 */
    531 			if (!AG_IS_REDUN(state))
    532 				state &= ~AGS_REDUN1;
    533 			state &= ~AGS_REDUN0;
    534 			if (new_seqno < ag->ag_seqno)
    535 				new_seqno = ag->ag_seqno;
    536 			else
    537 				ag->ag_seqno = new_seqno;
    538 		}
    539 
    540 		mask <<= 1;
    541 		dst &= mask;
    542 
    543 		if (ag_cors == 0) {
    544 			ag = ag_corsest;
    545 			break;
    546 		}
    547 		ag = ag_cors;
    548 		ag_cors = ag->ag_cors;
    549 	}
    550 
    551 	/* When we can no longer promote and combine routes,
    552 	 * flush the old route in the target slot.  Also flush
    553 	 * any finer routes that we know will never be aggregated by
    554 	 * the new route.
    555 	 *
    556 	 * In case we moved toward coarser masks,
    557 	 * get back where we belong
    558 	 */
    559 	if (ag != 0
    560 	    && ag->ag_mask < mask) {
    561 		ag_cors = ag;
    562 		ag = ag->ag_fine;
    563 	}
    564 
    565 	/* Empty the target slot
    566 	 */
    567 	if (ag != 0 && ag->ag_mask == mask) {
    568 		ag_flush(ag->ag_dst_h, ag->ag_mask, out);
    569 		ag = (ag_cors == 0) ? ag_corsest : ag_cors->ag_fine;
    570 	}
    571 
    572 #ifdef DEBUG_AG
    573 	(void)fflush(stderr);
    574 	if (ag == 0 && ag_cors != ag_finest)
    575 		abort();
    576 	if (ag_cors == 0 && ag != ag_corsest)
    577 		abort();
    578 	if (ag != 0 && ag->ag_cors != ag_cors)
    579 		abort();
    580 	if (ag_cors != 0 && ag_cors->ag_fine != ag)
    581 		abort();
    582 	CHECK_AG();
    583 #endif
    584 
    585 	/* Save the new route on the end of the table.
    586 	 */
    587 	nag = ag_avail;
    588 	ag_avail = nag->ag_fine;
    589 
    590 	nag->ag_dst_h = dst;
    591 	nag->ag_mask = mask;
    592 	nag->ag_gate = gate;
    593 	nag->ag_nhop = nhop;
    594 	nag->ag_metric = metric;
    595 	nag->ag_pref = pref;
    596 	nag->ag_tag = tag;
    597 	nag->ag_state = state;
    598 	nag->ag_seqno = new_seqno;
    599 
    600 	nag->ag_fine = ag;
    601 	if (ag != 0)
    602 		ag->ag_cors = nag;
    603 	else
    604 		ag_finest = nag;
    605 	nag->ag_cors = ag_cors;
    606 	if (ag_cors == 0)
    607 		ag_corsest = nag;
    608 	else
    609 		ag_cors->ag_fine = nag;
    610 	CHECK_AG();
    611 }
    612 
    613 
    614 static const char *
    615 rtm_type_name(u_char type)
    616 {
    617 	static const char *rtm_types[] = {
    618 		"RTM_ADD",
    619 		"RTM_DELETE",
    620 		"RTM_CHANGE",
    621 		"RTM_GET",
    622 		"RTM_LOSING",
    623 		"RTM_REDIRECT",
    624 		"RTM_MISS",
    625 		"RTM_LOCK",
    626 		"RTM_OLDADD",
    627 		"RTM_OLDDEL",
    628 		"RTM_RESOLVE",
    629 		"RTM_NEWADDR",
    630 		"RTM_DELADDR",
    631 #ifdef RTM_OIFINFO
    632 		"RTM_OIFINFO",
    633 #endif
    634 		"RTM_IFINFO",
    635 		"RTM_NEWMADDR",
    636 		"RTM_DELMADDR"
    637 	};
    638 #define NEW_RTM_PAT "RTM type %#x"
    639 	static char name0[sizeof(NEW_RTM_PAT)+2];
    640 
    641 
    642 	if (type > sizeof(rtm_types)/sizeof(rtm_types[0])
    643 	    || type == 0) {
    644 		snprintf(name0, sizeof(name0), NEW_RTM_PAT, type);
    645 		return name0;
    646 	} else {
    647 		return rtm_types[type-1];
    648 	}
    649 #undef NEW_RTM_PAT
    650 }
    651 
    652 
    653 /* Trim a mask in a sockaddr
    654  *	Produce a length of 0 for an address of 0.
    655  *	Otherwise produce the index of the first zero byte.
    656  */
    657 void
    658 #ifdef _HAVE_SIN_LEN
    659 masktrim(struct sockaddr_in *ap)
    660 #else
    661 masktrim(struct sockaddr_in_new *ap)
    662 #endif
    663 {
    664 	char *cp;
    665 
    666 	if (ap->sin_addr.s_addr == 0) {
    667 		ap->sin_len = 0;
    668 		return;
    669 	}
    670 	cp = (char *)&ap->sin_addr.s_addr + sizeof(ap->sin_addr.s_addr);
    671 	while (*--cp == 0)
    672 		continue;
    673 	ap->sin_len = cp - (char*)ap + 1;
    674 }
    675 
    676 
    677 /* Tell the kernel to add, delete or change a route
    678  */
    679 static void
    680 rtioctl(int action,			/* RTM_DELETE, etc */
    681 	naddr dst,
    682 	naddr gate,
    683 	naddr mask,
    684 	int metric,
    685 	int flags)
    686 {
    687 	struct {
    688 		struct rt_msghdr w_rtm;
    689 		struct sockaddr_in w_dst;
    690 		struct sockaddr_in w_gate;
    691 #ifdef _HAVE_SA_LEN
    692 		struct sockaddr_in w_mask;
    693 #else
    694 		struct sockaddr_in_new w_mask;
    695 #endif
    696 	} w;
    697 	long cc;
    698 #   define PAT " %-10s %s metric=%d flags=%#x"
    699 #   define ARGS rtm_type_name(action), rtname(dst,mask,gate), metric, flags
    700 
    701 again:
    702 	memset(&w, 0, sizeof(w));
    703 	w.w_rtm.rtm_msglen = sizeof(w);
    704 	w.w_rtm.rtm_version = RTM_VERSION;
    705 	w.w_rtm.rtm_type = action;
    706 	w.w_rtm.rtm_flags = flags;
    707 	w.w_rtm.rtm_seq = ++rt_sock_seqno;
    708 	w.w_rtm.rtm_addrs = RTA_DST|RTA_GATEWAY;
    709 	if (metric != 0 || action == RTM_CHANGE) {
    710 		w.w_rtm.rtm_rmx.rmx_hopcount = metric;
    711 		w.w_rtm.rtm_inits |= RTV_HOPCOUNT;
    712 	}
    713 	w.w_dst.sin_family = AF_INET;
    714 	w.w_dst.sin_addr.s_addr = dst;
    715 	w.w_gate.sin_family = AF_INET;
    716 	w.w_gate.sin_addr.s_addr = gate;
    717 #ifdef _HAVE_SA_LEN
    718 	w.w_dst.sin_len = sizeof(w.w_dst);
    719 	w.w_gate.sin_len = sizeof(w.w_gate);
    720 #endif
    721 	if (mask == HOST_MASK) {
    722 		w.w_rtm.rtm_flags |= RTF_HOST;
    723 		w.w_rtm.rtm_msglen -= sizeof(w.w_mask);
    724 	} else {
    725 		w.w_rtm.rtm_addrs |= RTA_NETMASK;
    726 		w.w_mask.sin_addr.s_addr = htonl(mask);
    727 #ifdef _HAVE_SA_LEN
    728 		masktrim(&w.w_mask);
    729 		if (w.w_mask.sin_len == 0)
    730 			w.w_mask.sin_len = sizeof(long);
    731 		w.w_rtm.rtm_msglen -= (sizeof(w.w_mask) - w.w_mask.sin_len);
    732 #endif
    733 	}
    734 
    735 #ifndef NO_INSTALL
    736 	cc = write(rt_sock, &w, w.w_rtm.rtm_msglen);
    737 	if (cc < 0) {
    738 		if (errno == ESRCH
    739 		    && (action == RTM_CHANGE || action == RTM_DELETE)) {
    740 			trace_act("route disappeared before" PAT, ARGS);
    741 			if (action == RTM_CHANGE) {
    742 				action = RTM_ADD;
    743 				goto again;
    744 			}
    745 			return;
    746 		}
    747 		msglog("write(rt_sock)" PAT ": %s", ARGS, strerror(errno));
    748 		return;
    749 	} else if (cc != w.w_rtm.rtm_msglen) {
    750 		msglog("write(rt_sock) wrote %ld instead of %d for" PAT,
    751 		       cc, w.w_rtm.rtm_msglen, ARGS);
    752 		return;
    753 	}
    754 #endif
    755 	if (TRACEKERNEL)
    756 		trace_misc("write kernel" PAT, ARGS);
    757 #undef PAT
    758 #undef ARGS
    759 }
    760 
    761 
    762 #define KHASH_SIZE 71			/* should be prime */
    763 #define KHASH(a,m) khash_bins[((a) ^ (m)) % KHASH_SIZE]
    764 static struct khash {
    765 	struct khash *k_next;
    766 	naddr	k_dst;
    767 	naddr	k_mask;
    768 	naddr	k_gate;
    769 	short	k_metric;
    770 	u_short	k_state;
    771 #define	    KS_NEW	0x001
    772 #define	    KS_DELETE	0x002		/* need to delete the route */
    773 #define	    KS_ADD	0x004		/* add to the kernel */
    774 #define	    KS_CHANGE	0x008		/* tell kernel to change the route */
    775 #define	    KS_DEL_ADD	0x010		/* delete & add to change the kernel */
    776 #define	    KS_STATIC	0x020		/* Static flag in kernel */
    777 #define	    KS_GATEWAY	0x040		/* G flag in kernel */
    778 #define	    KS_DYNAMIC	0x080		/* result of redirect */
    779 #define	    KS_DELETED	0x100		/* already deleted from kernel */
    780 #define	    KS_CHECK	0x200
    781 #define	    KS_LOCAL	0x400
    782 	time_t	k_keep;
    783 #define	    K_KEEP_LIM	30
    784 	time_t	k_redirect_time;	/* when redirected route 1st seen */
    785 } *khash_bins[KHASH_SIZE];
    786 
    787 
    788 static struct khash*
    789 kern_find(naddr dst, naddr mask, struct khash ***ppk)
    790 {
    791 	struct khash *k, **pk;
    792 
    793 	for (pk = &KHASH(dst,mask); (k = *pk) != 0; pk = &k->k_next) {
    794 		if (k->k_dst == dst && k->k_mask == mask)
    795 			break;
    796 	}
    797 	if (ppk != 0)
    798 		*ppk = pk;
    799 	return k;
    800 }
    801 
    802 
    803 static struct khash*
    804 kern_add(naddr dst, naddr mask)
    805 {
    806 	struct khash *k, **pk;
    807 
    808 	k = kern_find(dst, mask, &pk);
    809 	if (k != 0)
    810 		return k;
    811 
    812 	k = (struct khash *)rtmalloc(sizeof(*k), "kern_add");
    813 
    814 	memset(k, 0, sizeof(*k));
    815 	k->k_dst = dst;
    816 	k->k_mask = mask;
    817 	k->k_state = KS_NEW;
    818 	k->k_keep = now.tv_sec;
    819 	*pk = k;
    820 
    821 	return k;
    822 }
    823 
    824 
    825 /* If a kernel route has a non-zero metric, check that it is still in the
    826  *	daemon table, and not deleted by interfaces coming and going.
    827  */
    828 static void
    829 kern_check_static(struct khash *k,
    830 		  struct interface *ifp)
    831 {
    832 	struct rt_entry *rt;
    833 	struct rt_spare new;
    834 
    835 	if (k->k_metric == 0)
    836 		return;
    837 
    838 	memset(&new, 0, sizeof(new));
    839 	new.rts_ifp = ifp;
    840 	new.rts_gate = k->k_gate;
    841 	new.rts_router = (ifp != 0) ? ifp->int_addr : loopaddr;
    842 	new.rts_metric = k->k_metric;
    843 	new.rts_time = now.tv_sec;
    844 
    845 	rt = rtget(k->k_dst, k->k_mask);
    846 	if (rt != 0) {
    847 		if (!(rt->rt_state & RS_STATIC))
    848 			rtchange(rt, rt->rt_state | RS_STATIC, &new, 0);
    849 	} else {
    850 		rtadd(k->k_dst, k->k_mask, RS_STATIC, &new);
    851 	}
    852 }
    853 
    854 
    855 /* operate on a kernel entry
    856  */
    857 static void
    858 kern_ioctl(struct khash *k,
    859 	   int action,			/* RTM_DELETE, etc */
    860 	   int flags)
    861 
    862 {
    863 	switch (action) {
    864 	case RTM_DELETE:
    865 		k->k_state &= ~KS_DYNAMIC;
    866 		if (k->k_state & KS_DELETED)
    867 			return;
    868 		k->k_state |= KS_DELETED;
    869 		break;
    870 	case RTM_ADD:
    871 		k->k_state &= ~KS_DELETED;
    872 		break;
    873 	case RTM_CHANGE:
    874 		if (k->k_state & KS_DELETED) {
    875 			action = RTM_ADD;
    876 			k->k_state &= ~KS_DELETED;
    877 		}
    878 		break;
    879 	}
    880 
    881 	rtioctl(action, k->k_dst, k->k_gate, k->k_mask, k->k_metric, flags);
    882 }
    883 
    884 
    885 /* add a route the kernel told us
    886  */
    887 static void
    888 rtm_add(struct rt_msghdr *rtm,
    889 	struct rt_addrinfo *info,
    890 	time_t keep)
    891 {
    892 	struct khash *k;
    893 	struct interface *ifp;
    894 	naddr mask;
    895 
    896 
    897 	if (rtm->rtm_flags & RTF_HOST) {
    898 		mask = HOST_MASK;
    899 	} else if (INFO_MASK(info) != 0) {
    900 		mask = ntohl(S_ADDR(INFO_MASK(info)));
    901 	} else {
    902 		msglog("ignore %s without mask", rtm_type_name(rtm->rtm_type));
    903 		return;
    904 	}
    905 
    906 	k = kern_add(S_ADDR(INFO_DST(info)), mask);
    907 	if (k->k_state & KS_NEW)
    908 		k->k_keep = now.tv_sec+keep;
    909 	if (INFO_GATE(info) == 0) {
    910 		trace_act("note %s without gateway",
    911 			  rtm_type_name(rtm->rtm_type));
    912 		k->k_metric = HOPCNT_INFINITY;
    913 	} else if (INFO_GATE(info)->sa_family != AF_INET) {
    914 		trace_act("note %s with gateway AF=%d",
    915 			  rtm_type_name(rtm->rtm_type),
    916 			  INFO_GATE(info)->sa_family);
    917 		k->k_metric = HOPCNT_INFINITY;
    918 	} else {
    919 		k->k_gate = S_ADDR(INFO_GATE(info));
    920 		k->k_metric = rtm->rtm_rmx.rmx_hopcount;
    921 		if (k->k_metric < 0)
    922 			k->k_metric = 0;
    923 		else if (k->k_metric > HOPCNT_INFINITY-1)
    924 			k->k_metric = HOPCNT_INFINITY-1;
    925 	}
    926 	k->k_state &= ~(KS_DELETE | KS_ADD | KS_CHANGE | KS_DEL_ADD
    927 			| KS_DELETED | KS_GATEWAY | KS_STATIC
    928 			| KS_NEW | KS_CHECK | KS_LOCAL);
    929 	if (rtm->rtm_flags & RTF_GATEWAY)
    930 		k->k_state |= KS_GATEWAY;
    931 	if (rtm->rtm_flags & RTF_STATIC)
    932 		k->k_state |= KS_STATIC;
    933 	if (rtm->rtm_flags & RTF_LOCAL)
    934 		k->k_state |= KS_LOCAL;
    935 
    936 	if (0 != (rtm->rtm_flags & (RTF_DYNAMIC | RTF_MODIFIED))) {
    937 		if (INFO_AUTHOR(info) != 0
    938 		    && INFO_AUTHOR(info)->sa_family == AF_INET)
    939 			ifp = iflookup(S_ADDR(INFO_AUTHOR(info)));
    940 		else
    941 			ifp = 0;
    942 		if (supplier
    943 		    && (ifp == 0 || !(ifp->int_state & IS_REDIRECT_OK))) {
    944 			/* Routers are not supposed to listen to redirects,
    945 			 * so delete it if it came via an unknown interface
    946 			 * or the interface does not have special permission.
    947 			 */
    948 			k->k_state &= ~KS_DYNAMIC;
    949 			k->k_state |= KS_DELETE;
    950 			LIM_SEC(need_kern, 0);
    951 			trace_act("mark for deletion redirected %s --> %s"
    952 				  " via %s",
    953 				  addrname(k->k_dst, k->k_mask, 0),
    954 				  naddr_ntoa(k->k_gate),
    955 				  ifp ? ifp->int_name : "unknown interface");
    956 		} else {
    957 			k->k_state |= KS_DYNAMIC;
    958 			k->k_redirect_time = now.tv_sec;
    959 			trace_act("accept redirected %s --> %s via %s",
    960 				  addrname(k->k_dst, k->k_mask, 0),
    961 				  naddr_ntoa(k->k_gate),
    962 				  ifp ? ifp->int_name : "unknown interface");
    963 		}
    964 		return;
    965 	}
    966 
    967 	/* If it is not a static route, quit until the next comparison
    968 	 * between the kernel and daemon tables, when it will be deleted.
    969 	 */
    970 	if (!(k->k_state & KS_STATIC) && !(k->k_state & KS_LOCAL)) {
    971 		k->k_state |= KS_DELETE;
    972 		LIM_SEC(need_kern, k->k_keep);
    973 		return;
    974 	}
    975 
    976 	/* Put static routes with real metrics into the daemon table so
    977 	 * they can be advertised.
    978 	 *
    979 	 * Find the interface toward the gateway.
    980 	 */
    981 	ifp = iflookup(k->k_gate);
    982 	if (ifp == 0)
    983 		msglog("static route %s --> %s impossibly lacks ifp",
    984 		       addrname(S_ADDR(INFO_DST(info)), mask, 0),
    985 		       naddr_ntoa(k->k_gate));
    986 
    987 	kern_check_static(k, ifp);
    988 }
    989 
    990 
    991 /* deal with packet loss
    992  */
    993 static void
    994 rtm_lose(struct rt_msghdr *rtm,
    995 	 struct rt_addrinfo *info)
    996 {
    997 	if (INFO_GATE(info) == 0
    998 	    || INFO_GATE(info)->sa_family != AF_INET) {
    999 		trace_act("ignore %s without gateway",
   1000 			  rtm_type_name(rtm->rtm_type));
   1001 		return;
   1002 	}
   1003 
   1004 	if (rdisc_ok)
   1005 		rdisc_age(S_ADDR(INFO_GATE(info)));
   1006 	age(S_ADDR(INFO_GATE(info)));
   1007 }
   1008 
   1009 
   1010 /* Make the gateway slot of an info structure point to something
   1011  * useful.  If it is not already useful, but it specifies an interface,
   1012  * then fill in the sockaddr_in provided and point it there.
   1013  */
   1014 static int
   1015 get_info_gate(const struct sockaddr **sap,
   1016 	      struct sockaddr_in *rsin)
   1017 {
   1018 	const struct sockaddr_dl *sdl = (const struct sockaddr_dl *)*sap;
   1019 	struct interface *ifp;
   1020 
   1021 	if (sdl == 0)
   1022 		return 0;
   1023 	if ((sdl)->sdl_family == AF_INET)
   1024 		return 1;
   1025 	if ((sdl)->sdl_family != AF_LINK)
   1026 		return 0;
   1027 
   1028 	ifp = ifwithindex(sdl->sdl_index, 1);
   1029 	if (ifp == 0)
   1030 		return 0;
   1031 
   1032 	rsin->sin_addr.s_addr = ifp->int_addr;
   1033 #ifdef _HAVE_SA_LEN
   1034 	rsin->sin_len = sizeof(*rsin);
   1035 #endif
   1036 	rsin->sin_family = AF_INET;
   1037 	*sap = (const struct sockaddr*)rsin;
   1038 
   1039 	return 1;
   1040 }
   1041 
   1042 
   1043 /* Clean the kernel table by copying it to the daemon image.
   1044  * Eventually the daemon will delete any extra routes.
   1045  */
   1046 void
   1047 flush_kern(void)
   1048 {
   1049 	static char *sysctl_buf;
   1050 	static size_t sysctl_buf_size = 0;
   1051 	size_t needed;
   1052 	int mib[6];
   1053 	char *next, *lim;
   1054 	struct rt_msghdr *rtm;
   1055 	struct sockaddr_in gate_sin;
   1056 	struct rt_addrinfo info;
   1057 	int i;
   1058 	struct khash *k;
   1059 
   1060 
   1061 	for (i = 0; i < KHASH_SIZE; i++) {
   1062 		for (k = khash_bins[i]; k != 0; k = k->k_next) {
   1063 			k->k_state |= KS_CHECK;
   1064 		}
   1065 	}
   1066 
   1067 	mib[0] = CTL_NET;
   1068 	mib[1] = PF_ROUTE;
   1069 	mib[2] = 0;		/* protocol */
   1070 	mib[3] = 0;		/* wildcard address family */
   1071 	mib[4] = NET_RT_DUMP;
   1072 	mib[5] = 0;		/* no flags */
   1073 	for (;;) {
   1074 		if ((needed = sysctl_buf_size) != 0) {
   1075 			if (sysctl(mib, 6, sysctl_buf,&needed, 0, 0) >= 0)
   1076 				break;
   1077 			if (errno != ENOMEM && errno != EFAULT)
   1078 				BADERR(1,"flush_kern: sysctl(RT_DUMP)");
   1079 			free(sysctl_buf);
   1080 			needed = 0;
   1081 		}
   1082 		if (sysctl(mib, 6, 0, &needed, 0, 0) < 0)
   1083 			BADERR(1,"flush_kern: sysctl(RT_DUMP) estimate");
   1084 		/* Kludge around the habit of some systems, such as
   1085 		 * BSD/OS 3.1, to not admit how many routes are in the
   1086 		 * kernel, or at least to be quite wrong.
   1087 		 */
   1088 		needed += 50*(sizeof(*rtm)+5*sizeof(struct sockaddr));
   1089 		sysctl_buf = rtmalloc(sysctl_buf_size = needed,
   1090 				      "flush_kern sysctl(RT_DUMP)");
   1091 	}
   1092 
   1093 	lim = sysctl_buf + needed;
   1094 	for (next = sysctl_buf; next < lim; next += rtm->rtm_msglen) {
   1095 		rtm = (struct rt_msghdr *)next;
   1096 		if (rtm->rtm_msglen == 0) {
   1097 			msglog("zero length kernel route at "
   1098 			       " %#lx in buffer %#lx before %#lx",
   1099 			       (u_long)rtm, (u_long)sysctl_buf, (u_long)lim);
   1100 			break;
   1101 		}
   1102 
   1103 		rt_xaddrs(&info,
   1104 			  (struct sockaddr *)(rtm+1),
   1105 			  (struct sockaddr *)(next + rtm->rtm_msglen),
   1106 			  rtm->rtm_addrs);
   1107 
   1108 		if (INFO_DST(&info) == 0
   1109 		    || INFO_DST(&info)->sa_family != AF_INET)
   1110 			continue;
   1111 
   1112 		/* ignore cloned routes
   1113 		 */
   1114 #if defined(RTF_CLONED) && defined(__bsdi__)
   1115 		if (rtm->rtm_flags & RTF_CLONED)
   1116 			continue;
   1117 #endif
   1118 #if defined(RTF_WASCLONED) && defined(__FreeBSD__)
   1119 		if (rtm->rtm_flags & RTF_WASCLONED)
   1120 			continue;
   1121 #endif
   1122 
   1123 
   1124 		/* ignore multicast addresses
   1125 		 */
   1126 		if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info)))))
   1127 			continue;
   1128 
   1129 		if (!get_info_gate(&INFO_GATE(&info), &gate_sin))
   1130 			continue;
   1131 
   1132 		/* Note static routes and interface routes, and also
   1133 		 * preload the image of the kernel table so that
   1134 		 * we can later clean it, as well as avoid making
   1135 		 * unneeded changes.  Keep the old kernel routes for a
   1136 		 * few seconds to allow a RIP or router-discovery
   1137 		 * response to be heard.
   1138 		 */
   1139 		rtm_add(rtm,&info,MIN_WAITTIME);
   1140 	}
   1141 
   1142 	for (i = 0; i < KHASH_SIZE; i++) {
   1143 		for (k = khash_bins[i]; k != 0; k = k->k_next) {
   1144 			if (k->k_state & KS_CHECK) {
   1145 				msglog("%s --> %s disappeared from kernel",
   1146 				       addrname(k->k_dst, k->k_mask, 0),
   1147 				       naddr_ntoa(k->k_gate));
   1148 				del_static(k->k_dst, k->k_mask, k->k_gate, 1);
   1149 			}
   1150 		}
   1151 	}
   1152 }
   1153 
   1154 
   1155 /* Listen to announcements from the kernel
   1156  */
   1157 void
   1158 read_rt(void)
   1159 {
   1160 	long cc;
   1161 	struct interface *ifp;
   1162 	struct sockaddr_in gate_sin;
   1163 	naddr mask, gate;
   1164 	union {
   1165 		struct {
   1166 			struct rt_msghdr rtm;
   1167 			struct sockaddr addrs[RTAX_MAX];
   1168 		} r;
   1169 		struct if_msghdr ifm;
   1170 	} m;
   1171 	char str[100], *strp;
   1172 	struct rt_addrinfo info;
   1173 
   1174 
   1175 	for (;;) {
   1176 		cc = read(rt_sock, &m, sizeof(m));
   1177 		if (cc <= 0) {
   1178 			if (cc < 0 && errno != EWOULDBLOCK)
   1179 				LOGERR("read(rt_sock)");
   1180 			return;
   1181 		}
   1182 
   1183 		if (m.r.rtm.rtm_version != RTM_VERSION) {
   1184 			msglog("bogus routing message version %d",
   1185 			       m.r.rtm.rtm_version);
   1186 			continue;
   1187 		}
   1188 
   1189 		/* Ignore our own results.
   1190 		 */
   1191 		if (m.r.rtm.rtm_type <= RTM_CHANGE
   1192 		    && m.r.rtm.rtm_pid == mypid) {
   1193 			static int complained = 0;
   1194 			if (!complained) {
   1195 				msglog("receiving our own change messages");
   1196 				complained = 1;
   1197 			}
   1198 			continue;
   1199 		}
   1200 
   1201 		if (m.r.rtm.rtm_type == RTM_IFINFO
   1202 		    || m.r.rtm.rtm_type == RTM_NEWADDR
   1203 		    || m.r.rtm.rtm_type == RTM_DELADDR) {
   1204 			ifp = ifwithindex(m.ifm.ifm_index,
   1205 					  m.r.rtm.rtm_type != RTM_DELADDR);
   1206 			if (ifp == 0)
   1207 				trace_act("note %s with flags %#x"
   1208 					  " for unknown interface index #%d",
   1209 					  rtm_type_name(m.r.rtm.rtm_type),
   1210 					  m.ifm.ifm_flags,
   1211 					  m.ifm.ifm_index);
   1212 			else
   1213 				trace_act("note %s with flags %#x for %s",
   1214 					  rtm_type_name(m.r.rtm.rtm_type),
   1215 					  m.ifm.ifm_flags,
   1216 					  ifp->int_name);
   1217 
   1218 			/* After being informed of a change to an interface,
   1219 			 * check them all now if the check would otherwise
   1220 			 * be a long time from now, if the interface is
   1221 			 * not known, or if the interface has been turned
   1222 			 * off or on.
   1223 			 */
   1224 			if (ifinit_timer.tv_sec-now.tv_sec>=CHECK_BAD_INTERVAL
   1225 			    || ifp == 0
   1226 			    || ((ifp->int_if_flags ^ m.ifm.ifm_flags)
   1227 				& IFF_UP) != 0)
   1228 				ifinit_timer.tv_sec = now.tv_sec;
   1229 			continue;
   1230 		}
   1231 #ifdef RTM_OIFINFO
   1232 		if (m.r.rtm.rtm_type == RTM_OIFINFO)
   1233 			continue;	/* ignore compat message */
   1234 #endif
   1235 
   1236 		strlcpy(str, rtm_type_name(m.r.rtm.rtm_type), sizeof(str));
   1237 		strp = &str[strlen(str)];
   1238 		if (m.r.rtm.rtm_type <= RTM_CHANGE) {
   1239 			snprintf(strp, str + sizeof(str) - strp,
   1240 			    " from pid %d",m.r.rtm.rtm_pid);
   1241 			strp += strlen(strp);
   1242 		}
   1243 
   1244 		rt_xaddrs(&info, m.r.addrs, &m.r.addrs[RTAX_MAX],
   1245 			  m.r.rtm.rtm_addrs);
   1246 
   1247 		if (INFO_DST(&info) == 0) {
   1248 			trace_act("ignore %s without dst", str);
   1249 			continue;
   1250 		}
   1251 
   1252 		if (INFO_DST(&info)->sa_family != AF_INET) {
   1253 			trace_act("ignore %s for AF %d", str,
   1254 				  INFO_DST(&info)->sa_family);
   1255 			continue;
   1256 		}
   1257 
   1258 		mask = ((INFO_MASK(&info) != 0)
   1259 			? ntohl(S_ADDR(INFO_MASK(&info)))
   1260 			: (m.r.rtm.rtm_flags & RTF_HOST)
   1261 			? HOST_MASK
   1262 			: std_mask(S_ADDR(INFO_DST(&info))));
   1263 
   1264 		snprintf(strp, str + sizeof(str) - strp, ": %s",
   1265 		    addrname(S_ADDR(INFO_DST(&info)), mask, 0));
   1266 		strp += strlen(strp);
   1267 
   1268 		if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) {
   1269 			trace_act("ignore multicast %s", str);
   1270 			continue;
   1271 		}
   1272 
   1273 #if defined(RTF_CLONED) && defined(__bsdi__)
   1274 		if (m.r.rtm.rtm_flags & RTF_CLONED) {
   1275 			trace_act("ignore cloned %s", str);
   1276 			continue;
   1277 		}
   1278 #endif
   1279 #if defined(RTF_WASCLONED) && defined(__FreeBSD__)
   1280 		if (m.r.rtm.rtm_flags & RTF_WASCLONED) {
   1281 			trace_act("ignore cloned %s", str);
   1282 			continue;
   1283 		}
   1284 #endif
   1285 
   1286 		if (get_info_gate(&INFO_GATE(&info), &gate_sin)) {
   1287 			gate = S_ADDR(INFO_GATE(&info));
   1288 			snprintf(strp, str + sizeof(str) - strp,
   1289 			    " --> %s", naddr_ntoa(gate));
   1290 			strp += strlen(strp);
   1291 		} else {
   1292 			gate = 0;
   1293 		}
   1294 
   1295 		if (INFO_AUTHOR(&info) != 0) {
   1296 			snprintf(strp, str + sizeof(str) - strp,
   1297 			    " by authority of %s",
   1298 			    saddr_ntoa(INFO_AUTHOR(&info)));
   1299 			strp += strlen(strp);
   1300 		}
   1301 
   1302 		switch (m.r.rtm.rtm_type) {
   1303 		case RTM_ADD:
   1304 		case RTM_CHANGE:
   1305 		case RTM_REDIRECT:
   1306 			if (m.r.rtm.rtm_errno != 0) {
   1307 				trace_act("ignore %s with \"%s\" error",
   1308 					  str, strerror(m.r.rtm.rtm_errno));
   1309 			} else {
   1310 				trace_act("%s", str);
   1311 				rtm_add(&m.r.rtm,&info,0);
   1312 			}
   1313 			break;
   1314 
   1315 		case RTM_DELETE:
   1316 			if (m.r.rtm.rtm_errno != 0
   1317 			    && m.r.rtm.rtm_errno != ESRCH) {
   1318 				trace_act("ignore %s with \"%s\" error",
   1319 					  str, strerror(m.r.rtm.rtm_errno));
   1320 			} else {
   1321 				trace_act("%s", str);
   1322 				del_static(S_ADDR(INFO_DST(&info)), mask,
   1323 					   gate, 1);
   1324 			}
   1325 			break;
   1326 
   1327 		case RTM_LOSING:
   1328 			trace_act("%s", str);
   1329 			rtm_lose(&m.r.rtm,&info);
   1330 			break;
   1331 
   1332 		default:
   1333 			trace_act("ignore %s", str);
   1334 			break;
   1335 		}
   1336 	}
   1337 }
   1338 
   1339 
   1340 /* after aggregating, note routes that belong in the kernel
   1341  */
   1342 static void
   1343 kern_out(struct ag_info *ag)
   1344 {
   1345 	struct khash *k;
   1346 
   1347 
   1348 	/* Do not install bad routes if they are not already present.
   1349 	 * This includes routes that had RS_NET_SYN for interfaces that
   1350 	 * recently died.
   1351 	 */
   1352 	if (ag->ag_metric == HOPCNT_INFINITY) {
   1353 		k = kern_find(htonl(ag->ag_dst_h), ag->ag_mask, 0);
   1354 		if (k == 0)
   1355 			return;
   1356 	} else {
   1357 		k = kern_add(htonl(ag->ag_dst_h), ag->ag_mask);
   1358 	}
   1359 
   1360 	if (k->k_state & KS_NEW) {
   1361 		/* will need to add new entry to the kernel table */
   1362 		k->k_state = KS_ADD;
   1363 		if (ag->ag_state & AGS_GATEWAY)
   1364 			k->k_state |= KS_GATEWAY;
   1365 		k->k_gate = ag->ag_gate;
   1366 		k->k_metric = ag->ag_metric;
   1367 		return;
   1368 	}
   1369 
   1370 	if ((k->k_state & KS_STATIC) || (k->k_state & KS_LOCAL))
   1371 		return;
   1372 
   1373 	/* modify existing kernel entry if necessary */
   1374 	if (k->k_gate != ag->ag_gate
   1375 	    || k->k_metric != ag->ag_metric) {
   1376 		/* Must delete bad interface routes etc. to change them. */
   1377 		if (k->k_metric == HOPCNT_INFINITY)
   1378 			k->k_state |= KS_DEL_ADD;
   1379 		k->k_gate = ag->ag_gate;
   1380 		k->k_metric = ag->ag_metric;
   1381 		k->k_state |= KS_CHANGE;
   1382 	}
   1383 
   1384 	/* If the daemon thinks the route should exist, forget
   1385 	 * about any redirections.
   1386 	 * If the daemon thinks the route should exist, eventually
   1387 	 * override manual intervention by the operator.
   1388 	 */
   1389 	if ((k->k_state & (KS_DYNAMIC | KS_DELETED)) != 0) {
   1390 		k->k_state &= ~KS_DYNAMIC;
   1391 		k->k_state |= (KS_ADD | KS_DEL_ADD);
   1392 	}
   1393 
   1394 	if ((k->k_state & KS_GATEWAY)
   1395 	    && !(ag->ag_state & AGS_GATEWAY)) {
   1396 		k->k_state &= ~KS_GATEWAY;
   1397 		k->k_state |= (KS_ADD | KS_DEL_ADD);
   1398 	} else if (!(k->k_state & KS_GATEWAY)
   1399 		   && (ag->ag_state & AGS_GATEWAY)) {
   1400 		k->k_state |= KS_GATEWAY;
   1401 		k->k_state |= (KS_ADD | KS_DEL_ADD);
   1402 	}
   1403 
   1404 	/* Deleting-and-adding is necessary to change aspects of a route.
   1405 	 * Just delete instead of deleting and then adding a bad route.
   1406 	 * Otherwise, we want to keep the route in the kernel.
   1407 	 */
   1408 	if (k->k_metric == HOPCNT_INFINITY
   1409 	    && (k->k_state & KS_DEL_ADD))
   1410 		k->k_state |= KS_DELETE;
   1411 	else
   1412 		k->k_state &= ~KS_DELETE;
   1413 #undef RT
   1414 }
   1415 
   1416 
   1417 /* ARGSUSED */
   1418 static int
   1419 walk_kern(struct radix_node *rn,
   1420 	  struct walkarg *argp UNUSED)
   1421 {
   1422 #define RT ((struct rt_entry *)rn)
   1423 	char metric, pref;
   1424 	u_int ags = 0;
   1425 
   1426 
   1427 	/* Do not install synthetic routes */
   1428 	if (RT->rt_state & RS_NET_SYN)
   1429 		return 0;
   1430 
   1431 	if (!(RT->rt_state & RS_IF)) {
   1432 		/* This is an ordinary route, not for an interface.
   1433 		 */
   1434 
   1435 		/* aggregate, ordinary good routes without regard to
   1436 		 * their metric
   1437 		 */
   1438 		pref = 1;
   1439 		ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_AGGREGATE);
   1440 
   1441 		/* Do not install host routes directly to hosts, to avoid
   1442 		 * interfering with ARP entries in the kernel table.
   1443 		 */
   1444 		if (RT_ISHOST(RT)
   1445 		    && ntohl(RT->rt_dst) == RT->rt_gate)
   1446 			return 0;
   1447 
   1448 	} else {
   1449 		/* This is an interface route.
   1450 		 * Do not install routes for "external" remote interfaces.
   1451 		 */
   1452 		if (RT->rt_ifp != 0 && (RT->rt_ifp->int_state & IS_EXTERNAL))
   1453 			return 0;
   1454 
   1455 		/* Interfaces should override received routes.
   1456 		 */
   1457 		pref = 0;
   1458 		ags |= (AGS_IF | AGS_CORS_GATE);
   1459 
   1460 		/* If it is not an interface, or an alias for an interface,
   1461 		 * it must be a "gateway."
   1462 		 *
   1463 		 * If it is a "remote" interface, it is also a "gateway" to
   1464 		 * the kernel if is not a alias.
   1465 		 */
   1466 		if (RT->rt_ifp == 0
   1467 		    || (RT->rt_ifp->int_state & IS_REMOTE))
   1468 			ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_AGGREGATE);
   1469 	}
   1470 
   1471 	/* If RIP is off and IRDP is on, let the route to the discovered
   1472 	 * route suppress any RIP routes.  Eventually the RIP routes
   1473 	 * will time-out and be deleted.  This reaches the steady-state
   1474 	 * quicker.
   1475 	 */
   1476 	if ((RT->rt_state & RS_RDISC) && rip_sock < 0)
   1477 		ags |= AGS_CORS_GATE;
   1478 
   1479 	metric = RT->rt_metric;
   1480 	if (metric == HOPCNT_INFINITY) {
   1481 		/* if the route is dead, so try hard to aggregate. */
   1482 		pref = HOPCNT_INFINITY;
   1483 		ags |= (AGS_FINE_GATE | AGS_SUPPRESS);
   1484 		ags &= ~(AGS_IF | AGS_CORS_GATE);
   1485 	}
   1486 
   1487 	ag_check(RT->rt_dst, RT->rt_mask, RT->rt_gate, 0,
   1488 		 metric,pref, 0, 0, ags, kern_out);
   1489 	return 0;
   1490 #undef RT
   1491 }
   1492 
   1493 
   1494 /* Update the kernel table to match the daemon table.
   1495  */
   1496 static void
   1497 fix_kern(void)
   1498 {
   1499 	int i;
   1500 	struct khash *k, **pk;
   1501 
   1502 
   1503 	need_kern = age_timer;
   1504 
   1505 	/* Walk daemon table, updating the copy of the kernel table.
   1506 	 */
   1507 	(void)rn_walktree(rhead, walk_kern, 0);
   1508 	ag_flush(0,0,kern_out);
   1509 
   1510 	for (i = 0; i < KHASH_SIZE; i++) {
   1511 		for (pk = &khash_bins[i]; (k = *pk) != 0; ) {
   1512 			/* Do not touch static routes */
   1513 			if (k->k_state & KS_STATIC) {
   1514 				kern_check_static(k,0);
   1515 				pk = &k->k_next;
   1516 				continue;
   1517 			}
   1518 
   1519 			/* Do not touch local routes */
   1520 			if (k->k_state & KS_LOCAL) {
   1521 				pk = &k->k_next;
   1522 				continue;
   1523 			}
   1524 
   1525 			/* check hold on routes deleted by the operator */
   1526 			if (k->k_keep > now.tv_sec) {
   1527 				/* ensure we check when the hold is over */
   1528 				LIM_SEC(need_kern, k->k_keep);
   1529 				/* mark for the next cycle */
   1530 				k->k_state |= KS_DELETE;
   1531 				pk = &k->k_next;
   1532 				continue;
   1533 			}
   1534 
   1535 			if ((k->k_state & KS_DELETE)
   1536 			    && !(k->k_state & KS_DYNAMIC)) {
   1537 				kern_ioctl(k, RTM_DELETE, 0);
   1538 				*pk = k->k_next;
   1539 				free(k);
   1540 				continue;
   1541 			}
   1542 
   1543 			if (k->k_state & KS_DEL_ADD)
   1544 				kern_ioctl(k, RTM_DELETE, 0);
   1545 
   1546 			if (k->k_state & KS_ADD) {
   1547 				kern_ioctl(k, RTM_ADD,
   1548 					   ((0 != (k->k_state & (KS_GATEWAY
   1549 							| KS_DYNAMIC)))
   1550 					    ? RTF_GATEWAY : 0));
   1551 			} else if (k->k_state & KS_CHANGE) {
   1552 				kern_ioctl(k,  RTM_CHANGE,
   1553 					   ((0 != (k->k_state & (KS_GATEWAY
   1554 							| KS_DYNAMIC)))
   1555 					    ? RTF_GATEWAY : 0));
   1556 			}
   1557 			k->k_state &= ~(KS_ADD|KS_CHANGE|KS_DEL_ADD);
   1558 
   1559 			/* Mark this route to be deleted in the next cycle.
   1560 			 * This deletes routes that disappear from the
   1561 			 * daemon table, since the normal aging code
   1562 			 * will clear the bit for routes that have not
   1563 			 * disappeared from the daemon table.
   1564 			 */
   1565 			k->k_state |= KS_DELETE;
   1566 			pk = &k->k_next;
   1567 		}
   1568 	}
   1569 }
   1570 
   1571 
   1572 /* Delete a static route in the image of the kernel table.
   1573  */
   1574 void
   1575 del_static(naddr dst,
   1576 	   naddr mask,
   1577 	   naddr gate,
   1578 	   int gone)
   1579 {
   1580 	struct khash *k;
   1581 	struct rt_entry *rt;
   1582 
   1583 	/* Just mark it in the table to be deleted next time the kernel
   1584 	 * table is updated.
   1585 	 * If it has already been deleted, mark it as such, and set its
   1586 	 * keep-timer so that it will not be deleted again for a while.
   1587 	 * This lets the operator delete a route added by the daemon
   1588 	 * and add a replacement.
   1589 	 */
   1590 	k = kern_find(dst, mask, 0);
   1591 	if (k != 0 && (gate == 0 || k->k_gate == gate)) {
   1592 		k->k_state &= ~(KS_STATIC | KS_DYNAMIC | KS_CHECK);
   1593 		k->k_state |= KS_DELETE;
   1594 		if (gone) {
   1595 			k->k_state |= KS_DELETED;
   1596 			k->k_keep = now.tv_sec + K_KEEP_LIM;
   1597 		}
   1598 	}
   1599 
   1600 	rt = rtget(dst, mask);
   1601 	if (rt != 0 && (rt->rt_state & RS_STATIC))
   1602 		rtbad(rt);
   1603 }
   1604 
   1605 
   1606 /* Delete all routes generated from ICMP Redirects that use a given gateway,
   1607  * as well as old redirected routes.
   1608  */
   1609 void
   1610 del_redirects(naddr bad_gate,
   1611 	      time_t old)
   1612 {
   1613 	int i;
   1614 	struct khash *k;
   1615 
   1616 
   1617 	for (i = 0; i < KHASH_SIZE; i++) {
   1618 		for (k = khash_bins[i]; k != 0; k = k->k_next) {
   1619 			if (!(k->k_state & KS_DYNAMIC)
   1620 			    || (k->k_state & KS_STATIC))
   1621 				continue;
   1622 
   1623 			if (k->k_gate != bad_gate
   1624 			    && k->k_redirect_time > old
   1625 			    && !supplier)
   1626 				continue;
   1627 
   1628 			k->k_state |= KS_DELETE;
   1629 			k->k_state &= ~KS_DYNAMIC;
   1630 			need_kern.tv_sec = now.tv_sec;
   1631 			trace_act("mark redirected %s --> %s for deletion",
   1632 				  addrname(k->k_dst, k->k_mask, 0),
   1633 				  naddr_ntoa(k->k_gate));
   1634 		}
   1635 	}
   1636 }
   1637 
   1638 
   1639 /* Start the daemon tables.
   1640  */
   1641 extern int max_keylen;
   1642 
   1643 void
   1644 rtinit(void)
   1645 {
   1646 	int i;
   1647 	struct ag_info *ag;
   1648 
   1649 	/* Initialize the radix trees */
   1650 	max_keylen = sizeof(struct sockaddr_in);
   1651 	rn_init();
   1652 	rn_inithead((void*)&rhead, 32);
   1653 
   1654 	/* mark all of the slots in the table free */
   1655 	ag_avail = ag_slots;
   1656 	for (ag = ag_slots, i = 1; i < NUM_AG_SLOTS; i++) {
   1657 		ag->ag_fine = ag+1;
   1658 		ag++;
   1659 	}
   1660 }
   1661 
   1662 
   1663 #ifdef _HAVE_SIN_LEN
   1664 static struct sockaddr_in dst_sock = {sizeof(dst_sock), AF_INET, 0, {0}, {0}};
   1665 static struct sockaddr_in mask_sock = {sizeof(mask_sock), AF_INET, 0, {0}, {0}};
   1666 #else
   1667 static struct sockaddr_in_new dst_sock = {_SIN_ADDR_SIZE, AF_INET};
   1668 static struct sockaddr_in_new mask_sock = {_SIN_ADDR_SIZE, AF_INET};
   1669 #endif
   1670 
   1671 
   1672 static void
   1673 set_need_flash(void)
   1674 {
   1675 	if (!need_flash) {
   1676 		need_flash = 1;
   1677 		/* Do not send the flash update immediately.  Wait a little
   1678 		 * while to hear from other routers.
   1679 		 */
   1680 		no_flash.tv_sec = now.tv_sec + MIN_WAITTIME;
   1681 	}
   1682 }
   1683 
   1684 
   1685 /* Get a particular routing table entry
   1686  */
   1687 struct rt_entry *
   1688 rtget(naddr dst, naddr mask)
   1689 {
   1690 	struct rt_entry *rt;
   1691 
   1692 	dst_sock.sin_addr.s_addr = dst;
   1693 	mask_sock.sin_addr.s_addr = htonl(mask);
   1694 	masktrim(&mask_sock);
   1695 	rt = (struct rt_entry *)rhead->rnh_lookup(&dst_sock,&mask_sock,rhead);
   1696 	if (!rt
   1697 	    || rt->rt_dst != dst
   1698 	    || rt->rt_mask != mask)
   1699 		return 0;
   1700 
   1701 	return rt;
   1702 }
   1703 
   1704 
   1705 /* Find a route to dst as the kernel would.
   1706  */
   1707 struct rt_entry *
   1708 rtfind(naddr dst)
   1709 {
   1710 	dst_sock.sin_addr.s_addr = dst;
   1711 	return (struct rt_entry *)rhead->rnh_matchaddr(&dst_sock, rhead);
   1712 }
   1713 
   1714 
   1715 /* add a route to the table
   1716  */
   1717 void
   1718 rtadd(naddr	dst,
   1719       naddr	mask,
   1720       u_int	state,			/* rt_state for the entry */
   1721       struct	rt_spare *new)
   1722 {
   1723 	struct rt_entry *rt;
   1724 	naddr smask;
   1725 	int i;
   1726 	struct rt_spare *rts;
   1727 
   1728 	rt = (struct rt_entry *)rtmalloc(sizeof (*rt), "rtadd");
   1729 	memset(rt, 0, sizeof(*rt));
   1730 	for (rts = rt->rt_spares, i = NUM_SPARES; i != 0; i--, rts++)
   1731 		rts->rts_metric = HOPCNT_INFINITY;
   1732 
   1733 	rt->rt_nodes->rn_key = (caddr_t)&rt->rt_dst_sock;
   1734 	rt->rt_dst = dst;
   1735 	rt->rt_dst_sock.sin_family = AF_INET;
   1736 #ifdef _HAVE_SIN_LEN
   1737 	rt->rt_dst_sock.sin_len = dst_sock.sin_len;
   1738 #endif
   1739 	if (mask != HOST_MASK) {
   1740 		smask = std_mask(dst);
   1741 		if ((smask & ~mask) == 0 && mask > smask)
   1742 			state |= RS_SUBNET;
   1743 	}
   1744 	mask_sock.sin_addr.s_addr = htonl(mask);
   1745 	masktrim(&mask_sock);
   1746 	rt->rt_mask = mask;
   1747 	rt->rt_state = state;
   1748 	rt->rt_spares[0] = *new;
   1749 	rt->rt_time = now.tv_sec;
   1750 	rt->rt_poison_metric = HOPCNT_INFINITY;
   1751 	rt->rt_seqno = update_seqno;
   1752 
   1753 	if (++total_routes == MAX_ROUTES)
   1754 		msglog("have maximum (%d) routes", total_routes);
   1755 	if (TRACEACTIONS)
   1756 		trace_add_del("Add", rt);
   1757 
   1758 	need_kern.tv_sec = now.tv_sec;
   1759 	set_need_flash();
   1760 
   1761 	if (0 == rhead->rnh_addaddr(&rt->rt_dst_sock, &mask_sock,
   1762 				    rhead, rt->rt_nodes)) {
   1763 		msglog("rnh_addaddr() failed for %s mask=%#lx",
   1764 		       naddr_ntoa(dst), (u_long)mask);
   1765 		free(rt);
   1766 	}
   1767 }
   1768 
   1769 
   1770 /* notice a changed route
   1771  */
   1772 void
   1773 rtchange(struct rt_entry *rt,
   1774 	 u_int	state,			/* new state bits */
   1775 	 struct rt_spare *new,
   1776 	 char	*label)
   1777 {
   1778 	if (rt->rt_metric != new->rts_metric) {
   1779 		/* Fix the kernel immediately if it seems the route
   1780 		 * has gone bad, since there may be a working route that
   1781 		 * aggregates this route.
   1782 		 */
   1783 		if (new->rts_metric == HOPCNT_INFINITY) {
   1784 			need_kern.tv_sec = now.tv_sec;
   1785 			if (new->rts_time >= now.tv_sec - EXPIRE_TIME)
   1786 				new->rts_time = now.tv_sec - EXPIRE_TIME;
   1787 		}
   1788 		rt->rt_seqno = update_seqno;
   1789 		set_need_flash();
   1790 	}
   1791 
   1792 	if (rt->rt_gate != new->rts_gate) {
   1793 		need_kern.tv_sec = now.tv_sec;
   1794 		rt->rt_seqno = update_seqno;
   1795 		set_need_flash();
   1796 	}
   1797 
   1798 	state |= (rt->rt_state & RS_SUBNET);
   1799 
   1800 	/* Keep various things from deciding ageless routes are stale.
   1801 	 */
   1802 	if (!AGE_RT(state, new->rts_ifp))
   1803 		new->rts_time = now.tv_sec;
   1804 
   1805 	if (TRACEACTIONS)
   1806 		trace_change(rt, state, new,
   1807 			     label ? label : "Chg   ");
   1808 
   1809 	rt->rt_state = state;
   1810 	rt->rt_spares[0] = *new;
   1811 }
   1812 
   1813 
   1814 /* check for a better route among the spares
   1815  */
   1816 static struct rt_spare *
   1817 rts_better(struct rt_entry *rt)
   1818 {
   1819 	struct rt_spare *rts, *rts1;
   1820 	int i;
   1821 
   1822 	/* find the best alternative among the spares */
   1823 	rts = rt->rt_spares+1;
   1824 	for (i = NUM_SPARES, rts1 = rts+1; i > 2; i--, rts1++) {
   1825 		if (BETTER_LINK(rt,rts1,rts))
   1826 			rts = rts1;
   1827 	}
   1828 
   1829 	return rts;
   1830 }
   1831 
   1832 
   1833 /* switch to a backup route
   1834  */
   1835 void
   1836 rtswitch(struct rt_entry *rt,
   1837 	 struct rt_spare *rts)
   1838 {
   1839 	struct rt_spare swap;
   1840 	char label[20];
   1841 
   1842 	/* Do not change permanent routes */
   1843 	if (0 != (rt->rt_state & (RS_MHOME | RS_STATIC | RS_RDISC
   1844 				  | RS_NET_SYN | RS_IF)))
   1845 		return;
   1846 
   1847 	/* find the best alternative among the spares */
   1848 	if (rts == 0)
   1849 		rts = rts_better(rt);
   1850 
   1851 	/* Do not bother if it is not worthwhile.
   1852 	 */
   1853 	if (!BETTER_LINK(rt, rts, rt->rt_spares))
   1854 		return;
   1855 
   1856 	swap = rt->rt_spares[0];
   1857 	(void)snprintf(label, sizeof(label), "Use #%d",
   1858 	    (int)(rts - rt->rt_spares));
   1859 	rtchange(rt, rt->rt_state & ~(RS_NET_SYN | RS_RDISC), rts, label);
   1860 	if (swap.rts_metric == HOPCNT_INFINITY) {
   1861 		*rts = rts_empty;
   1862 	} else {
   1863 		*rts = swap;
   1864 	}
   1865 }
   1866 
   1867 
   1868 void
   1869 rtdelete(struct rt_entry *rt)
   1870 {
   1871 	struct khash *k;
   1872 
   1873 
   1874 	if (TRACEACTIONS)
   1875 		trace_add_del("Del", rt);
   1876 
   1877 	k = kern_find(rt->rt_dst, rt->rt_mask, 0);
   1878 	if (k != 0) {
   1879 		k->k_state |= KS_DELETE;
   1880 		need_kern.tv_sec = now.tv_sec;
   1881 	}
   1882 
   1883 	dst_sock.sin_addr.s_addr = rt->rt_dst;
   1884 	mask_sock.sin_addr.s_addr = htonl(rt->rt_mask);
   1885 	masktrim(&mask_sock);
   1886 	if (rt != (struct rt_entry *)rhead->rnh_deladdr(&dst_sock, &mask_sock,
   1887 							rhead)) {
   1888 		msglog("rnh_deladdr() failed");
   1889 	} else {
   1890 		free(rt);
   1891 		total_routes--;
   1892 	}
   1893 }
   1894 
   1895 
   1896 void
   1897 rts_delete(struct rt_entry *rt,
   1898 	   struct rt_spare *rts)
   1899 {
   1900 	trace_upslot(rt, rts, &rts_empty);
   1901 	*rts = rts_empty;
   1902 }
   1903 
   1904 
   1905 /* Get rid of a bad route, and try to switch to a replacement.
   1906  */
   1907 void
   1908 rtbad(struct rt_entry *rt)
   1909 {
   1910 	struct rt_spare new;
   1911 
   1912 	/* Poison the route */
   1913 	new = rt->rt_spares[0];
   1914 	new.rts_metric = HOPCNT_INFINITY;
   1915 	rtchange(rt, rt->rt_state & ~(RS_IF | RS_LOCAL | RS_STATIC), &new, 0);
   1916 	rtswitch(rt, 0);
   1917 }
   1918 
   1919 
   1920 /* Junk a RS_NET_SYN or RS_LOCAL route,
   1921  *	unless it is needed by another interface.
   1922  */
   1923 void
   1924 rtbad_sub(struct rt_entry *rt)
   1925 {
   1926 	struct interface *ifp, *ifp1;
   1927 	struct intnet *intnetp;
   1928 	u_int state;
   1929 
   1930 
   1931 	ifp1 = 0;
   1932 	state = 0;
   1933 
   1934 	if (rt->rt_state & RS_LOCAL) {
   1935 		/* Is this the route through loopback for the interface?
   1936 		 * If so, see if it is used by any other interfaces, such
   1937 		 * as a point-to-point interface with the same local address.
   1938 		 */
   1939 		for (ifp = ifnet; ifp != 0; ifp = ifp->int_next) {
   1940 			/* Retain it if another interface needs it.
   1941 			 */
   1942 			if (ifp->int_addr == rt->rt_ifp->int_addr) {
   1943 				state |= RS_LOCAL;
   1944 				ifp1 = ifp;
   1945 				break;
   1946 			}
   1947 		}
   1948 
   1949 	}
   1950 
   1951 	if (!(state & RS_LOCAL)) {
   1952 		/* Retain RIPv1 logical network route if there is another
   1953 		 * interface that justifies it.
   1954 		 */
   1955 		if (rt->rt_state & RS_NET_SYN) {
   1956 			for (ifp = ifnet; ifp != 0; ifp = ifp->int_next) {
   1957 				if ((ifp->int_state & IS_NEED_NET_SYN)
   1958 				    && rt->rt_mask == ifp->int_std_mask
   1959 				    && rt->rt_dst == ifp->int_std_addr) {
   1960 					state |= RS_NET_SYN;
   1961 					ifp1 = ifp;
   1962 					break;
   1963 				}
   1964 			}
   1965 		}
   1966 
   1967 		/* or if there is an authority route that needs it. */
   1968 		for (intnetp = intnets;
   1969 		     intnetp != 0;
   1970 		     intnetp = intnetp->intnet_next) {
   1971 			if (intnetp->intnet_addr == rt->rt_dst
   1972 			    && intnetp->intnet_mask == rt->rt_mask) {
   1973 				state |= (RS_NET_SYN | RS_NET_INT);
   1974 				break;
   1975 			}
   1976 		}
   1977 	}
   1978 
   1979 	if (ifp1 != 0 || (state & RS_NET_SYN)) {
   1980 		struct rt_spare new = rt->rt_spares[0];
   1981 		new.rts_ifp = ifp1;
   1982 		rtchange(rt, ((rt->rt_state & ~(RS_NET_SYN|RS_LOCAL)) | state),
   1983 			 &new, 0);
   1984 	} else {
   1985 		rtbad(rt);
   1986 	}
   1987 }
   1988 
   1989 
   1990 /* Called while walking the table looking for sick interfaces
   1991  * or after a time change.
   1992  */
   1993 /* ARGSUSED */
   1994 int
   1995 walk_bad(struct radix_node *rn,
   1996 	 struct walkarg *argp UNUSED)
   1997 {
   1998 #define RT ((struct rt_entry *)rn)
   1999 	struct rt_spare *rts;
   2000 	int i;
   2001 
   2002 
   2003 	/* fix any spare routes through the interface
   2004 	 */
   2005 	rts = RT->rt_spares;
   2006 	for (i = NUM_SPARES; i != 1; i--) {
   2007 		rts++;
   2008 		if (rts->rts_metric < HOPCNT_INFINITY
   2009 		    && (rts->rts_ifp == 0
   2010 			|| (rts->rts_ifp->int_state & IS_BROKE)))
   2011 			rts_delete(RT, rts);
   2012 	}
   2013 
   2014 	/* Deal with the main route
   2015 	 */
   2016 	/* finished if it has been handled before or if its interface is ok
   2017 	 */
   2018 	if (RT->rt_ifp == 0 || !(RT->rt_ifp->int_state & IS_BROKE))
   2019 		return 0;
   2020 
   2021 	/* Bad routes for other than interfaces are easy.
   2022 	 */
   2023 	if (0 == (RT->rt_state & (RS_IF | RS_NET_SYN | RS_LOCAL))) {
   2024 		rtbad(RT);
   2025 		return 0;
   2026 	}
   2027 
   2028 	rtbad_sub(RT);
   2029 	return 0;
   2030 #undef RT
   2031 }
   2032 
   2033 
   2034 /* Check the age of an individual route.
   2035  */
   2036 /* ARGSUSED */
   2037 static int
   2038 walk_age(struct radix_node *rn,
   2039 	   struct walkarg *argp UNUSED)
   2040 {
   2041 #define RT ((struct rt_entry *)rn)
   2042 	struct interface *ifp;
   2043 	struct rt_spare *rts;
   2044 	int i;
   2045 
   2046 
   2047 	/* age all of the spare routes, including the primary route
   2048 	 * currently in use
   2049 	 */
   2050 	rts = RT->rt_spares;
   2051 	for (i = NUM_SPARES; i != 0; i--, rts++) {
   2052 
   2053 		ifp = rts->rts_ifp;
   2054 		if (i == NUM_SPARES) {
   2055 			if (!AGE_RT(RT->rt_state, ifp)) {
   2056 				/* Keep various things from deciding ageless
   2057 				 * routes are stale
   2058 				 */
   2059 				rts->rts_time = now.tv_sec;
   2060 				continue;
   2061 			}
   2062 
   2063 			/* forget RIP routes after RIP has been turned off.
   2064 			 */
   2065 			if (rip_sock < 0) {
   2066 				rtdelete(RT);
   2067 				return 0;
   2068 			}
   2069 		}
   2070 
   2071 		/* age failing routes
   2072 		 */
   2073 		if (age_bad_gate == rts->rts_gate
   2074 		    && rts->rts_time >= now_stale) {
   2075 			rts->rts_time -= SUPPLY_INTERVAL;
   2076 		}
   2077 
   2078 		/* trash the spare routes when they go bad */
   2079 		if (rts->rts_metric < HOPCNT_INFINITY
   2080 		    && now_garbage > rts->rts_time
   2081 		    && i != NUM_SPARES)
   2082 			rts_delete(RT, rts);
   2083 	}
   2084 
   2085 
   2086 	/* finished if the active route is still fresh */
   2087 	if (now_stale <= RT->rt_time)
   2088 		return 0;
   2089 
   2090 	/* try to switch to an alternative */
   2091 	rtswitch(RT, 0);
   2092 
   2093 	/* Delete a dead route after it has been publically mourned. */
   2094 	if (now_garbage > RT->rt_time) {
   2095 		rtdelete(RT);
   2096 		return 0;
   2097 	}
   2098 
   2099 	/* Start poisoning a bad route before deleting it. */
   2100 	if (now.tv_sec - RT->rt_time > EXPIRE_TIME) {
   2101 		struct rt_spare new = RT->rt_spares[0];
   2102 		new.rts_metric = HOPCNT_INFINITY;
   2103 		rtchange(RT, RT->rt_state, &new, 0);
   2104 	}
   2105 	return 0;
   2106 }
   2107 
   2108 
   2109 /* Watch for dead routes and interfaces.
   2110  */
   2111 void
   2112 age(naddr bad_gate)
   2113 {
   2114 	struct interface *ifp;
   2115 	int need_query = 0;
   2116 
   2117 	/* If not listening to RIP, there is no need to age the routes in
   2118 	 * the table.
   2119 	 */
   2120 	age_timer.tv_sec = (now.tv_sec
   2121 			    + ((rip_sock < 0) ? NEVER : SUPPLY_INTERVAL));
   2122 
   2123 	/* Check for dead IS_REMOTE interfaces by timing their
   2124 	 * transmissions.
   2125 	 */
   2126 	for (ifp = ifnet; ifp; ifp = ifp->int_next) {
   2127 		if (!(ifp->int_state & IS_REMOTE))
   2128 			continue;
   2129 
   2130 		/* ignore unreachable remote interfaces */
   2131 		if (!check_remote(ifp))
   2132 			continue;
   2133 
   2134 		/* Restore remote interface that has become reachable
   2135 		 */
   2136 		if (ifp->int_state & IS_BROKE)
   2137 			if_ok(ifp, "remote ");
   2138 
   2139 		if (ifp->int_act_time != NEVER
   2140 		    && now.tv_sec - ifp->int_act_time > EXPIRE_TIME) {
   2141 			msglog("remote interface %s to %s timed out after"
   2142 			       " %lld:%lld",
   2143 			       ifp->int_name,
   2144 			       naddr_ntoa(ifp->int_dstaddr),
   2145 			       (long long)(now.tv_sec - ifp->int_act_time)/60,
   2146 			       (long long)(now.tv_sec - ifp->int_act_time)%60);
   2147 			if_sick(ifp);
   2148 		}
   2149 
   2150 		/* If we have not heard from the other router
   2151 		 * recently, ask it.
   2152 		 */
   2153 		if (now.tv_sec >= ifp->int_query_time) {
   2154 			ifp->int_query_time = NEVER;
   2155 			need_query = 1;
   2156 		}
   2157 	}
   2158 
   2159 	/* Age routes. */
   2160 	age_bad_gate = bad_gate;
   2161 	(void)rn_walktree(rhead, walk_age, 0);
   2162 
   2163 	/* delete old redirected routes to keep the kernel table small
   2164 	 * and prevent blackholes
   2165 	 */
   2166 	del_redirects(bad_gate, now.tv_sec-STALE_TIME);
   2167 
   2168 	/* Update the kernel routing table. */
   2169 	fix_kern();
   2170 
   2171 	/* poke reticent remote gateways */
   2172 	if (need_query)
   2173 		rip_query();
   2174 }
   2175