Remove unnecessary whitespace in macro arguments in some manual pages.
[dragonfly.git] / sbin / routed / table.c
blobf821d46e820012659909063e1390ed10c1ee3f64
1 /*
2 * Copyright (c) 1983, 1988, 1993
3 * The Regents of the University of California. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
7 * are met:
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. All advertising materials mentioning features or use of this software
14 * must display the following acknowledgment:
15 * This product includes software developed by the University of
16 * California, Berkeley and its contributors.
17 * 4. Neither the name of the University nor the names of its contributors
18 * may be used to endorse or promote products derived from this software
19 * without specific prior written permission.
21 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
22 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
25 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * SUCH DAMAGE.
33 * $FreeBSD: src/sbin/routed/table.c,v 1.9.2.2 2000/08/14 17:00:04 sheldonh Exp $
36 #include "defs.h"
38 #if !defined(sgi) && !defined(__NetBSD__)
39 static char sccsid[] __attribute__((unused)) = "@(#)tables.c 8.1 (Berkeley) 6/5/93";
40 #elif defined(__NetBSD__)
41 __RCSID("$NetBSD$");
42 #endif
44 static struct rt_spare *rts_better(struct rt_entry *);
45 static struct rt_spare rts_empty = {0,0,0,HOPCNT_INFINITY,0,0,0};
46 static void set_need_flash(void);
47 #ifdef _HAVE_SIN_LEN
48 static void masktrim(struct sockaddr_in *ap);
49 #else
50 static void masktrim(struct sockaddr_in_new *ap);
51 #endif
54 struct radix_node_head *rhead; /* root of the radix tree */
56 int need_flash = 1; /* flash update needed
57 * start =1 to suppress the 1st
60 struct timeval age_timer; /* next check of old routes */
61 struct timeval need_kern = { /* need to update kernel table */
62 EPOCH+MIN_WAITTIME-1, 0
65 int stopint;
67 int total_routes;
69 /* zap any old routes through this gateway */
70 naddr age_bad_gate;
73 /* It is desirable to "aggregate" routes, to combine differing routes of
74 * the same metric and next hop into a common route with a smaller netmask
75 * or to suppress redundant routes, routes that add no information to
76 * routes with smaller netmasks.
78 * A route is redundant if and only if any and all routes with smaller
79 * but matching netmasks and nets are the same. Since routes are
80 * kept sorted in the radix tree, redundant routes always come second.
82 * There are two kinds of aggregations. First, two routes of the same bit
83 * mask and differing only in the least significant bit of the network
84 * number can be combined into a single route with a coarser mask.
86 * Second, a route can be suppressed in favor of another route with a more
87 * coarse mask provided no incompatible routes with intermediate masks
88 * are present. The second kind of aggregation involves suppressing routes.
89 * A route must not be suppressed if an incompatible route exists with
90 * an intermediate mask, since the suppressed route would be covered
91 * by the intermediate.
93 * This code relies on the radix tree walk encountering routes
94 * sorted first by address, with the smallest address first.
97 struct ag_info ag_slots[NUM_AG_SLOTS], *ag_avail, *ag_corsest, *ag_finest;
99 /* #define DEBUG_AG */
100 #ifdef DEBUG_AG
101 #define CHECK_AG() {int acnt = 0; struct ag_info *cag; \
102 for (cag = ag_avail; cag != NULL; cag = cag->ag_fine) \
103 acnt++; \
104 for (cag = ag_corsest; cag != NULL; cag = cag->ag_fine) \
105 acnt++; \
106 if (acnt != NUM_AG_SLOTS) { \
107 fflush(stderr); \
108 abort(); \
111 #else
112 #define CHECK_AG()
113 #endif
116 /* Output the contents of an aggregation table slot.
117 * This function must always be immediately followed with the deletion
118 * of the target slot.
120 static void
121 ag_out(struct ag_info *ag,
122 void (*out)(struct ag_info *))
124 struct ag_info *ag_cors;
125 naddr bit;
128 /* Forget it if this route should not be output for split-horizon. */
129 if (ag->ag_state & AGS_SPLIT_HZ)
130 return;
132 /* If we output both the even and odd twins, then the immediate parent,
133 * if it is present, is redundant, unless the parent manages to
134 * aggregate into something coarser.
135 * On successive calls, this code detects the even and odd twins,
136 * and marks the parent.
138 * Note that the order in which the radix tree code emits routes
139 * ensures that the twins are seen before the parent is emitted.
141 ag_cors = ag->ag_cors;
142 if (ag_cors != NULL
143 && ag_cors->ag_mask == ag->ag_mask<<1
144 && ag_cors->ag_dst_h == (ag->ag_dst_h & ag_cors->ag_mask)) {
145 ag_cors->ag_state |= ((ag_cors->ag_dst_h == ag->ag_dst_h)
146 ? AGS_REDUN0
147 : AGS_REDUN1);
150 /* Skip it if this route is itself redundant.
152 * It is ok to change the contents of the slot here, since it is
153 * always deleted next.
155 if (ag->ag_state & AGS_REDUN0) {
156 if (ag->ag_state & AGS_REDUN1)
157 return; /* quit if fully redundant */
158 /* make it finer if it is half-redundant */
159 bit = (-ag->ag_mask) >> 1;
160 ag->ag_dst_h |= bit;
161 ag->ag_mask |= bit;
163 } else if (ag->ag_state & AGS_REDUN1) {
164 /* make it finer if it is half-redundant */
165 bit = (-ag->ag_mask) >> 1;
166 ag->ag_mask |= bit;
168 out(ag);
172 static void
173 ag_del(struct ag_info *ag)
175 CHECK_AG();
177 if (ag->ag_cors == 0)
178 ag_corsest = ag->ag_fine;
179 else
180 ag->ag_cors->ag_fine = ag->ag_fine;
182 if (ag->ag_fine == 0)
183 ag_finest = ag->ag_cors;
184 else
185 ag->ag_fine->ag_cors = ag->ag_cors;
187 ag->ag_fine = ag_avail;
188 ag_avail = ag;
190 CHECK_AG();
194 /* Flush routes waiting for aggregation.
195 * This must not suppress a route unless it is known that among all
196 * routes with coarser masks that match it, the one with the longest
197 * mask is appropriate. This is ensured by scanning the routes
198 * in lexical order, and with the most restrictive mask first
199 * among routes to the same destination.
201 void
202 ag_flush(naddr lim_dst_h, /* flush routes to here */
203 naddr lim_mask, /* matching this mask */
204 void (*out)(struct ag_info *))
206 struct ag_info *ag, *ag_cors;
207 naddr dst_h;
210 for (ag = ag_finest;
211 ag != NULL && ag->ag_mask >= lim_mask;
212 ag = ag_cors) {
213 ag_cors = ag->ag_cors;
215 /* work on only the specified routes */
216 dst_h = ag->ag_dst_h;
217 if ((dst_h & lim_mask) != lim_dst_h)
218 continue;
220 if (!(ag->ag_state & AGS_SUPPRESS))
221 ag_out(ag, out);
223 else for ( ; ; ag_cors = ag_cors->ag_cors) {
224 /* Look for a route that can suppress the
225 * current route */
226 if (ag_cors == NULL) {
227 /* failed, so output it and look for
228 * another route to work on
230 ag_out(ag, out);
231 break;
234 if ((dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h) {
235 /* We found a route with a coarser mask that
236 * aggregates the current target.
238 * If it has a different next hop, it
239 * cannot replace the target, so output
240 * the target.
242 if (ag->ag_gate != ag_cors->ag_gate
243 && !(ag->ag_state & AGS_FINE_GATE)
244 && !(ag_cors->ag_state & AGS_CORS_GATE)) {
245 ag_out(ag, out);
246 break;
249 /* If the coarse route has a good enough
250 * metric, it suppresses the target.
251 * If the suppressed target was redundant,
252 * then mark the suppressor redundant.
254 if (ag_cors->ag_pref <= ag->ag_pref) {
255 if (ag_cors->ag_seqno > ag->ag_seqno)
256 ag_cors->ag_seqno = ag->ag_seqno;
257 if (AG_IS_REDUN(ag->ag_state)
258 && ag_cors->ag_mask==ag->ag_mask<<1) {
259 if (ag_cors->ag_dst_h == dst_h)
260 ag_cors->ag_state |= AGS_REDUN0;
261 else
262 ag_cors->ag_state |= AGS_REDUN1;
264 if (ag->ag_tag != ag_cors->ag_tag)
265 ag_cors->ag_tag = 0;
266 if (ag->ag_nhop != ag_cors->ag_nhop)
267 ag_cors->ag_nhop = 0;
268 break;
273 /* That route has either been output or suppressed */
274 ag_cors = ag->ag_cors;
275 ag_del(ag);
278 CHECK_AG();
282 /* Try to aggregate a route with previous routes.
284 void
285 ag_check(naddr dst,
286 naddr mask,
287 naddr gate,
288 naddr nhop,
289 char metric,
290 char pref,
291 u_int seqnum,
292 u_short tag,
293 u_short state,
294 void (*out)(struct ag_info *)) /* output using this */
296 struct ag_info *ag, *nag, *ag_cors;
297 naddr xaddr;
298 int x;
300 dst = ntohl(dst);
302 /* Punt non-contiguous subnet masks.
304 * (X & -X) contains a single bit if and only if X is a power of 2.
305 * (X + (X & -X)) == 0 if and only if X is a power of 2.
307 if ((mask & -mask) + mask != 0) {
308 struct ag_info nc_ag;
310 nc_ag.ag_dst_h = dst;
311 nc_ag.ag_mask = mask;
312 nc_ag.ag_gate = gate;
313 nc_ag.ag_nhop = nhop;
314 nc_ag.ag_metric = metric;
315 nc_ag.ag_pref = pref;
316 nc_ag.ag_tag = tag;
317 nc_ag.ag_state = state;
318 nc_ag.ag_seqno = seqnum;
319 out(&nc_ag);
320 return;
323 /* Search for the right slot in the aggregation table.
325 ag_cors = NULL;
326 ag = ag_corsest;
327 while (ag != NULL) {
328 if (ag->ag_mask >= mask)
329 break;
331 /* Suppress old routes (i.e. combine with compatible routes
332 * with coarser masks) as we look for the right slot in the
333 * aggregation table for the new route.
334 * A route to an address less than the current destination
335 * will not be affected by the current route or any route
336 * seen hereafter. That means it is safe to suppress it.
337 * This check keeps poor routes (e.g. with large hop counts)
338 * from preventing suppression of finer routes.
340 if (ag_cors != NULL
341 && ag->ag_dst_h < dst
342 && (ag->ag_state & AGS_SUPPRESS)
343 && ag_cors->ag_pref <= ag->ag_pref
344 && (ag->ag_dst_h & ag_cors->ag_mask) == ag_cors->ag_dst_h
345 && (ag_cors->ag_gate == ag->ag_gate
346 || (ag->ag_state & AGS_FINE_GATE)
347 || (ag_cors->ag_state & AGS_CORS_GATE))) {
348 if (ag_cors->ag_seqno > ag->ag_seqno)
349 ag_cors->ag_seqno = ag->ag_seqno;
350 /* If the suppressed target was redundant,
351 * then mark the suppressor redundant.
353 if (AG_IS_REDUN(ag->ag_state)
354 && ag_cors->ag_mask == ag->ag_mask<<1) {
355 if (ag_cors->ag_dst_h == dst)
356 ag_cors->ag_state |= AGS_REDUN0;
357 else
358 ag_cors->ag_state |= AGS_REDUN1;
360 if (ag->ag_tag != ag_cors->ag_tag)
361 ag_cors->ag_tag = 0;
362 if (ag->ag_nhop != ag_cors->ag_nhop)
363 ag_cors->ag_nhop = 0;
364 ag_del(ag);
365 CHECK_AG();
366 } else {
367 ag_cors = ag;
369 ag = ag_cors->ag_fine;
372 /* If we find the even/odd twin of the new route, and if the
373 * masks and so forth are equal, we can aggregate them.
374 * We can probably promote one of the pair.
376 * Since the routes are encountered in lexical order,
377 * the new route must be odd. However, the second or later
378 * times around this loop, it could be the even twin promoted
379 * from the even/odd pair of twins of the finer route.
381 while (ag != NULL
382 && ag->ag_mask == mask
383 && ((ag->ag_dst_h ^ dst) & (mask<<1)) == 0) {
385 /* Here we know the target route and the route in the current
386 * slot have the same netmasks and differ by at most the
387 * last bit. They are either for the same destination, or
388 * for an even/odd pair of destinations.
390 if (ag->ag_dst_h == dst) {
391 /* We have two routes to the same destination.
392 * Routes are encountered in lexical order, so a
393 * route is never promoted until the parent route is
394 * already present. So we know that the new route is
395 * a promoted (or aggregated) pair and the route
396 * already in the slot is the explicit route.
398 * Prefer the best route if their metrics differ,
399 * or the aggregated one if not, following a sort
400 * of longest-match rule.
402 if (pref <= ag->ag_pref) {
403 ag->ag_gate = gate;
404 ag->ag_nhop = nhop;
405 ag->ag_tag = tag;
406 ag->ag_metric = metric;
407 ag->ag_pref = pref;
408 x = ag->ag_state;
409 ag->ag_state = state;
410 state = x;
413 /* The sequence number controls flash updating,
414 * and should be the smaller of the two.
416 if (ag->ag_seqno > seqnum)
417 ag->ag_seqno = seqnum;
419 /* Some bits are set if they are set on either route,
420 * except when the route is for an interface.
422 if (!(ag->ag_state & AGS_IF))
423 ag->ag_state |= (state & (AGS_AGGREGATE_EITHER
424 | AGS_REDUN0
425 | AGS_REDUN1));
426 return;
429 /* If one of the routes can be promoted and the other can
430 * be suppressed, it may be possible to combine them or
431 * worthwhile to promote one.
433 * Any route that can be promoted is always
434 * marked to be eligible to be suppressed.
436 if (!((state & AGS_AGGREGATE)
437 && (ag->ag_state & AGS_SUPPRESS))
438 && !((ag->ag_state & AGS_AGGREGATE)
439 && (state & AGS_SUPPRESS)))
440 break;
442 /* A pair of even/odd twin routes can be combined
443 * if either is redundant, or if they are via the
444 * same gateway and have the same metric.
446 if (AG_IS_REDUN(ag->ag_state)
447 || AG_IS_REDUN(state)
448 || (ag->ag_gate == gate
449 && ag->ag_pref == pref
450 && (state & ag->ag_state & AGS_AGGREGATE) != 0)) {
452 /* We have both the even and odd pairs.
453 * Since the routes are encountered in order,
454 * the route in the slot must be the even twin.
456 * Combine and promote (aggregate) the pair of routes.
458 if (seqnum > ag->ag_seqno)
459 seqnum = ag->ag_seqno;
460 if (!AG_IS_REDUN(state))
461 state &= ~AGS_REDUN1;
462 if (AG_IS_REDUN(ag->ag_state))
463 state |= AGS_REDUN0;
464 else
465 state &= ~AGS_REDUN0;
466 state |= (ag->ag_state & AGS_AGGREGATE_EITHER);
467 if (ag->ag_tag != tag)
468 tag = 0;
469 if (ag->ag_nhop != nhop)
470 nhop = 0;
472 /* Get rid of the even twin that was already
473 * in the slot.
475 ag_del(ag);
477 } else if (ag->ag_pref >= pref
478 && (ag->ag_state & AGS_AGGREGATE)) {
479 /* If we cannot combine the pair, maybe the route
480 * with the worse metric can be promoted.
482 * Promote the old, even twin, by giving its slot
483 * in the table to the new, odd twin.
485 ag->ag_dst_h = dst;
487 xaddr = ag->ag_gate;
488 ag->ag_gate = gate;
489 gate = xaddr;
491 xaddr = ag->ag_nhop;
492 ag->ag_nhop = nhop;
493 nhop = xaddr;
495 x = ag->ag_tag;
496 ag->ag_tag = tag;
497 tag = x;
499 /* The promoted route is even-redundant only if the
500 * even twin was fully redundant. It is not
501 * odd-redundant because the odd-twin will still be
502 * in the table.
504 x = ag->ag_state;
505 if (!AG_IS_REDUN(x))
506 x &= ~AGS_REDUN0;
507 x &= ~AGS_REDUN1;
508 ag->ag_state = state;
509 state = x;
511 x = ag->ag_metric;
512 ag->ag_metric = metric;
513 metric = x;
515 x = ag->ag_pref;
516 ag->ag_pref = pref;
517 pref = x;
519 /* take the newest sequence number */
520 if (seqnum >= ag->ag_seqno)
521 seqnum = ag->ag_seqno;
522 else
523 ag->ag_seqno = seqnum;
525 } else {
526 if (!(state & AGS_AGGREGATE))
527 break; /* cannot promote either twin */
529 /* Promote the new, odd twin by shaving its
530 * mask and address.
531 * The promoted route is odd-redundant only if the
532 * odd twin was fully redundant. It is not
533 * even-redundant because the even twin is still in
534 * the table.
536 if (!AG_IS_REDUN(state))
537 state &= ~AGS_REDUN1;
538 state &= ~AGS_REDUN0;
539 if (seqnum > ag->ag_seqno)
540 seqnum = ag->ag_seqno;
541 else
542 ag->ag_seqno = seqnum;
545 mask <<= 1;
546 dst &= mask;
548 if (ag_cors == NULL) {
549 ag = ag_corsest;
550 break;
552 ag = ag_cors;
553 ag_cors = ag->ag_cors;
556 /* When we can no longer promote and combine routes,
557 * flush the old route in the target slot. Also flush
558 * any finer routes that we know will never be aggregated by
559 * the new route.
561 * In case we moved toward coarser masks,
562 * get back where we belong
564 if (ag != NULL
565 && ag->ag_mask < mask) {
566 ag_cors = ag;
567 ag = ag->ag_fine;
570 /* Empty the target slot
572 if (ag != NULL && ag->ag_mask == mask) {
573 ag_flush(ag->ag_dst_h, ag->ag_mask, out);
574 ag = (ag_cors == NULL) ? ag_corsest : ag_cors->ag_fine;
577 #ifdef DEBUG_AG
578 fflush(stderr);
579 if (ag == NULL && ag_cors != ag_finest)
580 abort();
581 if (ag_cors == NULL && ag != ag_corsest)
582 abort();
583 if (ag != NULL && ag->ag_cors != ag_cors)
584 abort();
585 if (ag_cors != NULL && ag_cors->ag_fine != ag)
586 abort();
587 CHECK_AG();
588 #endif
590 /* Save the new route on the end of the table.
592 nag = ag_avail;
593 ag_avail = nag->ag_fine;
595 nag->ag_dst_h = dst;
596 nag->ag_mask = mask;
597 nag->ag_gate = gate;
598 nag->ag_nhop = nhop;
599 nag->ag_metric = metric;
600 nag->ag_pref = pref;
601 nag->ag_tag = tag;
602 nag->ag_state = state;
603 nag->ag_seqno = seqnum;
605 nag->ag_fine = ag;
606 if (ag != NULL)
607 ag->ag_cors = nag;
608 else
609 ag_finest = nag;
610 nag->ag_cors = ag_cors;
611 if (ag_cors == NULL)
612 ag_corsest = nag;
613 else
614 ag_cors->ag_fine = nag;
615 CHECK_AG();
619 #define NAME0_LEN 14
620 static const char *
621 rtm_type_name(u_char type)
623 static const char *rtm_types[] = {
624 "RTM_ADD",
625 "RTM_DELETE",
626 "RTM_CHANGE",
627 "RTM_GET",
628 "RTM_LOSING",
629 "RTM_REDIRECT",
630 "RTM_MISS",
631 "RTM_LOCK",
632 "RTM_OLDADD",
633 "RTM_OLDDEL",
634 "RTM_RESOLVE",
635 "RTM_NEWADDR",
636 "RTM_DELADDR",
637 "RTM_IFINFO",
638 "RTM_NEWMADDR",
639 "RTM_DELMADDR"
641 #define NEW_RTM_PAT "RTM type %#x"
642 static char name0[sizeof(NEW_RTM_PAT)+2];
645 if (type > sizeof(rtm_types)/sizeof(rtm_types[0])
646 || type == 0) {
647 snprintf(name0, sizeof(name0), NEW_RTM_PAT, type);
648 return name0;
649 } else {
650 return rtm_types[type-1];
652 #undef NEW_RTM_PAT
656 /* Trim a mask in a sockaddr
657 * Produce a length of 0 for an address of 0.
658 * Otherwise produce the index of the first zero byte.
660 void
661 #ifdef _HAVE_SIN_LEN
662 masktrim(struct sockaddr_in *ap)
663 #else
664 masktrim(struct sockaddr_in_new *ap)
665 #endif
667 char *cp;
669 if (ap->sin_addr.s_addr == 0) {
670 ap->sin_len = 0;
671 return;
673 cp = (char *)(&ap->sin_addr.s_addr+1);
674 while (*--cp == 0)
675 continue;
676 ap->sin_len = cp - (char*)ap + 1;
680 /* Tell the kernel to add, delete or change a route
682 static void
683 rtioctl(int action, /* RTM_DELETE, etc */
684 naddr dst,
685 naddr gate,
686 naddr mask,
687 int metric,
688 int flags)
690 struct {
691 struct rt_msghdr w_rtm;
692 struct sockaddr_in w_dst;
693 struct sockaddr_in w_gate;
694 #ifdef _HAVE_SA_LEN
695 struct sockaddr_in w_mask;
696 #else
697 struct sockaddr_in_new w_mask;
698 #endif
699 } w;
700 long cc;
701 # define PAT " %-10s %s metric=%d flags=%#x"
702 # define ARGS rtm_type_name(action), rtname(dst,mask,gate), metric, flags
704 again:
705 memset(&w, 0, sizeof(w));
706 w.w_rtm.rtm_msglen = sizeof(w);
707 w.w_rtm.rtm_version = RTM_VERSION;
708 w.w_rtm.rtm_type = action;
709 w.w_rtm.rtm_flags = flags;
710 w.w_rtm.rtm_seq = ++rt_sock_seqno;
711 w.w_rtm.rtm_addrs = RTA_DST|RTA_GATEWAY;
712 if (metric != 0 || action == RTM_CHANGE) {
713 w.w_rtm.rtm_rmx.rmx_hopcount = metric;
714 w.w_rtm.rtm_inits |= RTV_HOPCOUNT;
716 w.w_dst.sin_family = AF_INET;
717 w.w_dst.sin_addr.s_addr = dst;
718 w.w_gate.sin_family = AF_INET;
719 w.w_gate.sin_addr.s_addr = gate;
720 #ifdef _HAVE_SA_LEN
721 w.w_dst.sin_len = sizeof(w.w_dst);
722 w.w_gate.sin_len = sizeof(w.w_gate);
723 #endif
724 if (mask == HOST_MASK) {
725 w.w_rtm.rtm_flags |= RTF_HOST;
726 w.w_rtm.rtm_msglen -= sizeof(w.w_mask);
727 } else {
728 w.w_rtm.rtm_addrs |= RTA_NETMASK;
729 w.w_mask.sin_addr.s_addr = htonl(mask);
730 #ifdef _HAVE_SA_LEN
731 masktrim(&w.w_mask);
732 if (w.w_mask.sin_len == 0)
733 w.w_mask.sin_len = sizeof(long);
734 w.w_rtm.rtm_msglen -= (sizeof(w.w_mask) - w.w_mask.sin_len);
735 #endif
738 #ifndef NO_INSTALL
739 cc = write(rt_sock, &w, w.w_rtm.rtm_msglen);
740 if (cc < 0) {
741 if (errno == ESRCH
742 && (action == RTM_CHANGE || action == RTM_DELETE)) {
743 trace_act("route disappeared before" PAT, ARGS);
744 if (action == RTM_CHANGE) {
745 action = RTM_ADD;
746 goto again;
748 return;
750 msglog("write(rt_sock)" PAT ": %s", ARGS, strerror(errno));
751 return;
752 } else if (cc != w.w_rtm.rtm_msglen) {
753 msglog("write(rt_sock) wrote %ld instead of %d for" PAT,
754 cc, w.w_rtm.rtm_msglen, ARGS);
755 return;
757 #endif
758 if (TRACEKERNEL)
759 trace_misc("write kernel" PAT, ARGS);
760 #undef PAT
761 #undef ARGS
765 #define KHASH_SIZE 71 /* should be prime */
766 #define KHASH(a,m) khash_bins[((a) ^ (m)) % KHASH_SIZE]
767 static struct khash {
768 struct khash *k_next;
769 naddr k_dst;
770 naddr k_mask;
771 naddr k_gate;
772 short k_metric;
773 u_short k_state;
774 #define KS_NEW 0x001
775 #define KS_DELETE 0x002 /* need to delete the route */
776 #define KS_ADD 0x004 /* add to the kernel */
777 #define KS_CHANGE 0x008 /* tell kernel to change the route */
778 #define KS_DEL_ADD 0x010 /* delete & add to change the kernel */
779 #define KS_STATIC 0x020 /* Static flag in kernel */
780 #define KS_GATEWAY 0x040 /* G flag in kernel */
781 #define KS_DYNAMIC 0x080 /* result of redirect */
782 #define KS_DELETED 0x100 /* already deleted from kernel */
783 #define KS_CHECK 0x200
784 time_t k_keep;
785 #define K_KEEP_LIM 30
786 time_t k_redirect_time; /* when redirected route 1st seen */
787 } *khash_bins[KHASH_SIZE];
790 static struct khash*
791 kern_find(naddr dst, naddr mask, struct khash ***ppk)
793 struct khash *k, **pk;
795 for (pk = &KHASH(dst,mask); (k = *pk) != NULL; pk = &k->k_next) {
796 if (k->k_dst == dst && k->k_mask == mask)
797 break;
799 if (ppk != NULL)
800 *ppk = pk;
801 return k;
805 static struct khash*
806 kern_add(naddr dst, naddr mask)
808 struct khash *k, **pk;
810 k = kern_find(dst, mask, &pk);
811 if (k != NULL)
812 return k;
814 k = (struct khash *)rtmalloc(sizeof(*k), "kern_add");
816 memset(k, 0, sizeof(*k));
817 k->k_dst = dst;
818 k->k_mask = mask;
819 k->k_state = KS_NEW;
820 k->k_keep = now.tv_sec;
821 *pk = k;
823 return k;
827 /* If a kernel route has a non-zero metric, check that it is still in the
828 * daemon table, and not deleted by interfaces coming and going.
830 static void
831 kern_check_static(struct khash *k,
832 struct interface *ifp)
834 struct rt_entry *rt;
835 struct rt_spare new;
837 if (k->k_metric == 0)
838 return;
840 memset(&new, 0, sizeof(new));
841 new.rts_ifp = ifp;
842 new.rts_gate = k->k_gate;
843 new.rts_router = (ifp != NULL) ? ifp->int_addr : loopaddr;
844 new.rts_metric = k->k_metric;
845 new.rts_time = now.tv_sec;
847 rt = rtget(k->k_dst, k->k_mask);
848 if (rt != NULL) {
849 if (!(rt->rt_state & RS_STATIC))
850 rtchange(rt, rt->rt_state | RS_STATIC, &new, 0);
851 } else {
852 rtadd(k->k_dst, k->k_mask, RS_STATIC, &new);
857 /* operate on a kernel entry
859 static void
860 kern_ioctl(struct khash *k,
861 int action, /* RTM_DELETE, etc */
862 int flags)
865 switch (action) {
866 case RTM_DELETE:
867 k->k_state &= ~KS_DYNAMIC;
868 if (k->k_state & KS_DELETED)
869 return;
870 k->k_state |= KS_DELETED;
871 break;
872 case RTM_ADD:
873 k->k_state &= ~KS_DELETED;
874 break;
875 case RTM_CHANGE:
876 if (k->k_state & KS_DELETED) {
877 action = RTM_ADD;
878 k->k_state &= ~KS_DELETED;
880 break;
883 rtioctl(action, k->k_dst, k->k_gate, k->k_mask, k->k_metric, flags);
887 /* add a route the kernel told us
889 static void
890 rtm_add(struct rt_msghdr *rtm,
891 struct rt_addrinfo *info,
892 time_t keep)
894 struct khash *k;
895 struct interface *ifp;
896 naddr mask;
899 if (rtm->rtm_flags & RTF_HOST) {
900 mask = HOST_MASK;
901 } else if (INFO_MASK(info) != 0) {
902 mask = ntohl(S_ADDR(INFO_MASK(info)));
903 } else {
904 msglog("ignore %s without mask", rtm_type_name(rtm->rtm_type));
905 return;
908 k = kern_add(S_ADDR(INFO_DST(info)), mask);
909 if (k->k_state & KS_NEW)
910 k->k_keep = now.tv_sec+keep;
911 if (INFO_GATE(info) == 0) {
912 trace_act("note %s without gateway",
913 rtm_type_name(rtm->rtm_type));
914 k->k_metric = HOPCNT_INFINITY;
915 } else if (INFO_GATE(info)->sa_family != AF_INET) {
916 trace_act("note %s with gateway AF=%d",
917 rtm_type_name(rtm->rtm_type),
918 INFO_GATE(info)->sa_family);
919 k->k_metric = HOPCNT_INFINITY;
920 } else {
921 k->k_gate = S_ADDR(INFO_GATE(info));
922 k->k_metric = rtm->rtm_rmx.rmx_hopcount;
923 if (k->k_metric < 0)
924 k->k_metric = 0;
925 else if (k->k_metric > HOPCNT_INFINITY-1)
926 k->k_metric = HOPCNT_INFINITY-1;
928 k->k_state &= ~(KS_DELETE | KS_ADD | KS_CHANGE | KS_DEL_ADD
929 | KS_DELETED | KS_GATEWAY | KS_STATIC
930 | KS_NEW | KS_CHECK);
931 if (rtm->rtm_flags & RTF_GATEWAY)
932 k->k_state |= KS_GATEWAY;
933 if (rtm->rtm_flags & RTF_STATIC)
934 k->k_state |= KS_STATIC;
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 = NULL;
942 if (supplier
943 && (ifp == NULL || !(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.
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");
964 return;
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.
970 if (!(k->k_state & KS_STATIC)) {
971 k->k_state |= KS_DELETE;
972 LIM_SEC(need_kern, k->k_keep);
973 return;
976 /* Put static routes with real metrics into the daemon table so
977 * they can be advertised.
979 * Find the interface toward the gateway.
981 ifp = iflookup(k->k_gate);
982 if (ifp == NULL)
983 msglog("static route %s --> %s impossibly lacks ifp",
984 addrname(S_ADDR(INFO_DST(info)), mask, 0),
985 naddr_ntoa(k->k_gate));
987 kern_check_static(k, ifp);
991 /* deal with packet loss
993 static void
994 rtm_lose(struct rt_msghdr *rtm,
995 struct rt_addrinfo *info)
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;
1004 if (rdisc_ok)
1005 rdisc_age(S_ADDR(INFO_GATE(info)));
1006 age(S_ADDR(INFO_GATE(info)));
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.
1014 static int
1015 get_info_gate(struct sockaddr **sap,
1016 struct sockaddr_in *in)
1018 struct sockaddr_dl *sdl = (struct sockaddr_dl *)*sap;
1019 struct interface *ifp;
1021 if (sdl == NULL)
1022 return 0;
1023 if ((sdl)->sdl_family == AF_INET)
1024 return 1;
1025 if ((sdl)->sdl_family != AF_LINK)
1026 return 0;
1028 ifp = ifwithindex(sdl->sdl_index, 1);
1029 if (ifp == NULL)
1030 return 0;
1032 in->sin_addr.s_addr = ifp->int_addr;
1033 #ifdef _HAVE_SA_LEN
1034 in->sin_len = sizeof(*in);
1035 #endif
1036 in->sin_family = AF_INET;
1037 *sap = (struct sockaddr *)in;
1039 return 1;
1043 /* Clean the kernel table by copying it to the daemon image.
1044 * Eventually the daemon will delete any extra routes.
1046 void
1047 flush_kern(void)
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;
1061 for (i = 0; i < KHASH_SIZE; i++) {
1062 for (k = khash_bins[i]; k != NULL; k = k->k_next) {
1063 k->k_state |= KS_CHECK;
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;
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.
1088 needed += 50*(sizeof(*rtm)+5*sizeof(struct sockaddr));
1089 sysctl_buf = rtmalloc(sysctl_buf_size = needed,
1090 "flush_kern sysctl(RT_DUMP)");
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;
1103 rt_xaddrs(&info,
1104 (struct sockaddr *)(rtm+1),
1105 (struct sockaddr *)(next + rtm->rtm_msglen),
1106 rtm->rtm_addrs);
1108 if (INFO_DST(&info) == 0
1109 || INFO_DST(&info)->sa_family != AF_INET)
1110 continue;
1112 /* ignore ARP table entries on systems with a merged route
1113 * and ARP table.
1115 if (rtm->rtm_flags & RTF_LLINFO)
1116 continue;
1118 /* ignore multicast addresses
1120 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info)))))
1121 continue;
1123 if (!get_info_gate(&INFO_GATE(&info), &gate_sin))
1124 continue;
1126 /* Note static routes and interface routes, and also
1127 * preload the image of the kernel table so that
1128 * we can later clean it, as well as avoid making
1129 * unneeded changes. Keep the old kernel routes for a
1130 * few seconds to allow a RIP or router-discovery
1131 * response to be heard.
1133 rtm_add(rtm,&info,MIN_WAITTIME);
1136 for (i = 0; i < KHASH_SIZE; i++) {
1137 for (k = khash_bins[i]; k != NULL; k = k->k_next) {
1138 if (k->k_state & KS_CHECK) {
1139 msglog("%s --> %s disappeared from kernel",
1140 addrname(k->k_dst, k->k_mask, 0),
1141 naddr_ntoa(k->k_gate));
1142 del_static(k->k_dst, k->k_mask, k->k_gate, 1);
1149 /* Listen to announcements from the kernel
1151 void
1152 read_rt(void)
1154 long cc;
1155 struct interface *ifp;
1156 struct sockaddr_in gate_sin;
1157 naddr mask, gate;
1158 union {
1159 struct {
1160 struct rt_msghdr rtm;
1161 struct sockaddr addrs[RTAX_MAX];
1162 } r;
1163 struct if_msghdr ifm;
1164 } m;
1165 char str[100], *strp;
1166 struct rt_addrinfo info;
1169 for (;;) {
1170 cc = read(rt_sock, &m, sizeof(m));
1171 if (cc <= 0) {
1172 if (cc < 0 && errno != EWOULDBLOCK)
1173 LOGERR("read(rt_sock)");
1174 return;
1177 if (m.r.rtm.rtm_version != RTM_VERSION) {
1178 msglog("bogus routing message version %d",
1179 m.r.rtm.rtm_version);
1180 continue;
1183 /* Ignore our own results.
1185 if (m.r.rtm.rtm_type <= RTM_CHANGE
1186 && m.r.rtm.rtm_pid == mypid) {
1187 static int complained = 0;
1188 if (!complained) {
1189 msglog("receiving our own change messages");
1190 complained = 1;
1192 continue;
1195 if (m.r.rtm.rtm_type == RTM_IFINFO
1196 || m.r.rtm.rtm_type == RTM_NEWADDR
1197 || m.r.rtm.rtm_type == RTM_DELADDR) {
1198 ifp = ifwithindex(m.ifm.ifm_index,
1199 m.r.rtm.rtm_type != RTM_DELADDR);
1200 if (ifp == NULL)
1201 trace_act("note %s with flags %#x"
1202 " for unknown interface index #%d",
1203 rtm_type_name(m.r.rtm.rtm_type),
1204 m.ifm.ifm_flags,
1205 m.ifm.ifm_index);
1206 else
1207 trace_act("note %s with flags %#x for %s",
1208 rtm_type_name(m.r.rtm.rtm_type),
1209 m.ifm.ifm_flags,
1210 ifp->int_name);
1212 /* After being informed of a change to an interface,
1213 * check them all now if the check would otherwise
1214 * be a long time from now, if the interface is
1215 * not known, or if the interface has been turned
1216 * off or on.
1218 if (ifinit_timer.tv_sec-now.tv_sec>=CHECK_BAD_INTERVAL
1219 || ifp == NULL
1220 || ((ifp->int_if_flags ^ m.ifm.ifm_flags)
1221 & IFF_UP) != 0)
1222 ifinit_timer.tv_sec = now.tv_sec;
1223 continue;
1226 strcpy(str, rtm_type_name(m.r.rtm.rtm_type));
1227 strp = &str[strlen(str)];
1228 if (m.r.rtm.rtm_type <= RTM_CHANGE)
1229 strp += sprintf(strp," from pid %d",m.r.rtm.rtm_pid);
1231 rt_xaddrs(&info, m.r.addrs, &m.r.addrs[RTAX_MAX],
1232 m.r.rtm.rtm_addrs);
1234 if (INFO_DST(&info) == 0) {
1235 trace_act("ignore %s without dst", str);
1236 continue;
1239 if (INFO_DST(&info)->sa_family != AF_INET) {
1240 trace_act("ignore %s for AF %d", str,
1241 INFO_DST(&info)->sa_family);
1242 continue;
1245 mask = ((INFO_MASK(&info) != 0)
1246 ? ntohl(S_ADDR(INFO_MASK(&info)))
1247 : (m.r.rtm.rtm_flags & RTF_HOST)
1248 ? HOST_MASK
1249 : std_mask(S_ADDR(INFO_DST(&info))));
1251 strp += sprintf(strp, ": %s",
1252 addrname(S_ADDR(INFO_DST(&info)), mask, 0));
1254 if (IN_MULTICAST(ntohl(S_ADDR(INFO_DST(&info))))) {
1255 trace_act("ignore multicast %s", str);
1256 continue;
1259 if (m.r.rtm.rtm_flags & RTF_LLINFO) {
1260 trace_act("ignore ARP %s", str);
1261 continue;
1264 if (get_info_gate(&INFO_GATE(&info), &gate_sin)) {
1265 gate = S_ADDR(INFO_GATE(&info));
1266 strp += sprintf(strp, " --> %s", naddr_ntoa(gate));
1267 } else {
1268 gate = 0;
1271 if (INFO_AUTHOR(&info) != 0)
1272 strp += sprintf(strp, " by authority of %s",
1273 saddr_ntoa(INFO_AUTHOR(&info)));
1275 switch (m.r.rtm.rtm_type) {
1276 case RTM_ADD:
1277 case RTM_CHANGE:
1278 case RTM_REDIRECT:
1279 if (m.r.rtm.rtm_errno != 0) {
1280 trace_act("ignore %s with \"%s\" error",
1281 str, strerror(m.r.rtm.rtm_errno));
1282 } else {
1283 trace_act("%s", str);
1284 rtm_add(&m.r.rtm,&info,0);
1286 break;
1288 case RTM_DELETE:
1289 if (m.r.rtm.rtm_errno != 0
1290 && m.r.rtm.rtm_errno != ESRCH) {
1291 trace_act("ignore %s with \"%s\" error",
1292 str, strerror(m.r.rtm.rtm_errno));
1293 } else {
1294 trace_act("%s", str);
1295 del_static(S_ADDR(INFO_DST(&info)), mask,
1296 gate, 1);
1298 break;
1300 case RTM_LOSING:
1301 trace_act("%s", str);
1302 rtm_lose(&m.r.rtm,&info);
1303 break;
1305 default:
1306 trace_act("ignore %s", str);
1307 break;
1313 /* after aggregating, note routes that belong in the kernel
1315 static void
1316 kern_out(struct ag_info *ag)
1318 struct khash *k;
1321 /* Do not install bad routes if they are not already present.
1322 * This includes routes that had RS_NET_SYN for interfaces that
1323 * recently died.
1325 if (ag->ag_metric == HOPCNT_INFINITY) {
1326 k = kern_find(htonl(ag->ag_dst_h), ag->ag_mask, 0);
1327 if (k == NULL)
1328 return;
1329 } else {
1330 k = kern_add(htonl(ag->ag_dst_h), ag->ag_mask);
1333 if (k->k_state & KS_NEW) {
1334 /* will need to add new entry to the kernel table */
1335 k->k_state = KS_ADD;
1336 if (ag->ag_state & AGS_GATEWAY)
1337 k->k_state |= KS_GATEWAY;
1338 k->k_gate = ag->ag_gate;
1339 k->k_metric = ag->ag_metric;
1340 return;
1343 if (k->k_state & KS_STATIC)
1344 return;
1346 /* modify existing kernel entry if necessary */
1347 if (k->k_gate != ag->ag_gate
1348 || k->k_metric != ag->ag_metric) {
1349 /* Must delete bad interface routes etc. to change them. */
1350 if (k->k_metric == HOPCNT_INFINITY)
1351 k->k_state |= KS_DEL_ADD;
1352 k->k_gate = ag->ag_gate;
1353 k->k_metric = ag->ag_metric;
1354 k->k_state |= KS_CHANGE;
1357 /* If the daemon thinks the route should exist, forget
1358 * about any redirections.
1359 * If the daemon thinks the route should exist, eventually
1360 * override manual intervention by the operator.
1362 if ((k->k_state & (KS_DYNAMIC | KS_DELETED)) != 0) {
1363 k->k_state &= ~KS_DYNAMIC;
1364 k->k_state |= (KS_ADD | KS_DEL_ADD);
1367 if ((k->k_state & KS_GATEWAY)
1368 && !(ag->ag_state & AGS_GATEWAY)) {
1369 k->k_state &= ~KS_GATEWAY;
1370 k->k_state |= (KS_ADD | KS_DEL_ADD);
1371 } else if (!(k->k_state & KS_GATEWAY)
1372 && (ag->ag_state & AGS_GATEWAY)) {
1373 k->k_state |= KS_GATEWAY;
1374 k->k_state |= (KS_ADD | KS_DEL_ADD);
1377 /* Deleting-and-adding is necessary to change aspects of a route.
1378 * Just delete instead of deleting and then adding a bad route.
1379 * Otherwise, we want to keep the route in the kernel.
1381 if (k->k_metric == HOPCNT_INFINITY
1382 && (k->k_state & KS_DEL_ADD))
1383 k->k_state |= KS_DELETE;
1384 else
1385 k->k_state &= ~KS_DELETE;
1386 #undef RT
1390 /* ARGSUSED */
1391 static int
1392 walk_kern(struct radix_node *rn,
1393 struct walkarg *argp UNUSED)
1395 #define RT ((struct rt_entry *)rn)
1396 char metric, pref;
1397 u_int ags = 0;
1400 /* Do not install synthetic routes */
1401 if (RT->rt_state & RS_NET_SYN)
1402 return 0;
1404 if (!(RT->rt_state & RS_IF)) {
1405 /* This is an ordinary route, not for an interface.
1408 /* aggregate, ordinary good routes without regard to
1409 * their metric
1411 pref = 1;
1412 ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_AGGREGATE);
1414 /* Do not install host routes directly to hosts, to avoid
1415 * interfering with ARP entries in the kernel table.
1417 if (RT_ISHOST(RT)
1418 && ntohl(RT->rt_dst) == RT->rt_gate)
1419 return 0;
1421 } else {
1422 /* This is an interface route.
1423 * Do not install routes for "external" remote interfaces.
1425 if (RT->rt_ifp != 0 && (RT->rt_ifp->int_state & IS_EXTERNAL))
1426 return 0;
1428 /* Interfaces should override received routes.
1430 pref = 0;
1431 ags |= (AGS_IF | AGS_CORS_GATE);
1433 /* If it is not an interface, or an alias for an interface,
1434 * it must be a "gateway."
1436 * If it is a "remote" interface, it is also a "gateway" to
1437 * the kernel if is not a alias.
1439 if (RT->rt_ifp == 0
1440 || (RT->rt_ifp->int_state & IS_REMOTE))
1441 ags |= (AGS_GATEWAY | AGS_SUPPRESS | AGS_AGGREGATE);
1444 /* If RIP is off and IRDP is on, let the route to the discovered
1445 * route suppress any RIP routes. Eventually the RIP routes
1446 * will time-out and be deleted. This reaches the steady-state
1447 * quicker.
1449 if ((RT->rt_state & RS_RDISC) && rip_sock < 0)
1450 ags |= AGS_CORS_GATE;
1452 metric = RT->rt_metric;
1453 if (metric == HOPCNT_INFINITY) {
1454 /* if the route is dead, so try hard to aggregate. */
1455 pref = HOPCNT_INFINITY;
1456 ags |= (AGS_FINE_GATE | AGS_SUPPRESS);
1457 ags &= ~(AGS_IF | AGS_CORS_GATE);
1460 ag_check(RT->rt_dst, RT->rt_mask, RT->rt_gate, 0,
1461 metric,pref, 0, 0, ags, kern_out);
1462 return 0;
1463 #undef RT
1467 /* Update the kernel table to match the daemon table.
1469 static void
1470 fix_kern(void)
1472 int i;
1473 struct khash *k, **pk;
1476 need_kern = age_timer;
1478 /* Walk daemon table, updating the copy of the kernel table.
1480 rn_walktree(rhead, walk_kern, 0);
1481 ag_flush(0,0,kern_out);
1483 for (i = 0; i < KHASH_SIZE; i++) {
1484 for (pk = &khash_bins[i]; (k = *pk) != NULL; ) {
1485 /* Do not touch static routes */
1486 if (k->k_state & KS_STATIC) {
1487 kern_check_static(k,0);
1488 pk = &k->k_next;
1489 continue;
1492 /* check hold on routes deleted by the operator */
1493 if (k->k_keep > now.tv_sec) {
1494 /* ensure we check when the hold is over */
1495 LIM_SEC(need_kern, k->k_keep);
1496 /* mark for the next cycle */
1497 k->k_state |= KS_DELETE;
1498 pk = &k->k_next;
1499 continue;
1502 if ((k->k_state & KS_DELETE)
1503 && !(k->k_state & KS_DYNAMIC)) {
1504 kern_ioctl(k, RTM_DELETE, 0);
1505 *pk = k->k_next;
1506 free(k);
1507 continue;
1510 if (k->k_state & KS_DEL_ADD)
1511 kern_ioctl(k, RTM_DELETE, 0);
1513 if (k->k_state & KS_ADD) {
1514 kern_ioctl(k, RTM_ADD,
1515 ((0 != (k->k_state & (KS_GATEWAY
1516 | KS_DYNAMIC)))
1517 ? RTF_GATEWAY : 0));
1518 } else if (k->k_state & KS_CHANGE) {
1519 kern_ioctl(k, RTM_CHANGE,
1520 ((0 != (k->k_state & (KS_GATEWAY
1521 | KS_DYNAMIC)))
1522 ? RTF_GATEWAY : 0));
1524 k->k_state &= ~(KS_ADD|KS_CHANGE|KS_DEL_ADD);
1526 /* Mark this route to be deleted in the next cycle.
1527 * This deletes routes that disappear from the
1528 * daemon table, since the normal aging code
1529 * will clear the bit for routes that have not
1530 * disappeared from the daemon table.
1532 k->k_state |= KS_DELETE;
1533 pk = &k->k_next;
1539 /* Delete a static route in the image of the kernel table.
1541 void
1542 del_static(naddr dst,
1543 naddr mask,
1544 naddr gate,
1545 int gone)
1547 struct khash *k;
1548 struct rt_entry *rt;
1550 /* Just mark it in the table to be deleted next time the kernel
1551 * table is updated.
1552 * If it has already been deleted, mark it as such, and set its
1553 * keep-timer so that it will not be deleted again for a while.
1554 * This lets the operator delete a route added by the daemon
1555 * and add a replacement.
1557 k = kern_find(dst, mask, 0);
1558 if (k != NULL && (gate == 0 || k->k_gate == gate)) {
1559 k->k_state &= ~(KS_STATIC | KS_DYNAMIC | KS_CHECK);
1560 k->k_state |= KS_DELETE;
1561 if (gone) {
1562 k->k_state |= KS_DELETED;
1563 k->k_keep = now.tv_sec + K_KEEP_LIM;
1567 rt = rtget(dst, mask);
1568 if (rt != NULL && (rt->rt_state & RS_STATIC))
1569 rtbad(rt);
1573 /* Delete all routes generated from ICMP Redirects that use a given gateway,
1574 * as well as old redirected routes.
1576 void
1577 del_redirects(naddr bad_gate,
1578 time_t old)
1580 int i;
1581 struct khash *k;
1584 for (i = 0; i < KHASH_SIZE; i++) {
1585 for (k = khash_bins[i]; k != NULL; k = k->k_next) {
1586 if (!(k->k_state & KS_DYNAMIC)
1587 || (k->k_state & KS_STATIC))
1588 continue;
1590 if (k->k_gate != bad_gate
1591 && k->k_redirect_time > old
1592 && !supplier)
1593 continue;
1595 k->k_state |= KS_DELETE;
1596 k->k_state &= ~KS_DYNAMIC;
1597 need_kern.tv_sec = now.tv_sec;
1598 trace_act("mark redirected %s --> %s for deletion",
1599 addrname(k->k_dst, k->k_mask, 0),
1600 naddr_ntoa(k->k_gate));
1606 /* Start the daemon tables.
1608 extern int max_keylen;
1610 void
1611 rtinit(void)
1613 int i;
1614 struct ag_info *ag;
1616 /* Initialize the radix trees */
1617 max_keylen = sizeof(struct sockaddr_in);
1618 rn_init();
1619 rn_inithead(&rhead, 32);
1621 /* mark all of the slots in the table free */
1622 ag_avail = ag_slots;
1623 for (ag = ag_slots, i = 1; i < NUM_AG_SLOTS; i++) {
1624 ag->ag_fine = ag+1;
1625 ag++;
1630 #ifdef _HAVE_SIN_LEN
1631 static struct sockaddr_in dst_sock = {sizeof(dst_sock), AF_INET, 0, {0}, {0}};
1632 static struct sockaddr_in mask_sock = {sizeof(mask_sock), AF_INET, 0, {0}, {0}};
1633 #else
1634 static struct sockaddr_in_new dst_sock = {_SIN_ADDR_SIZE, AF_INET};
1635 static struct sockaddr_in_new mask_sock = {_SIN_ADDR_SIZE, AF_INET};
1636 #endif
1639 static void
1640 set_need_flash(void)
1642 if (!need_flash) {
1643 need_flash = 1;
1644 /* Do not send the flash update immediately. Wait a little
1645 * while to hear from other routers.
1647 no_flash.tv_sec = now.tv_sec + MIN_WAITTIME;
1652 /* Get a particular routing table entry
1654 struct rt_entry *
1655 rtget(naddr dst, naddr mask)
1657 struct rt_entry *rt;
1659 dst_sock.sin_addr.s_addr = dst;
1660 mask_sock.sin_addr.s_addr = htonl(mask);
1661 masktrim(&mask_sock);
1662 rt = (struct rt_entry *)rhead->rnh_lookup(&dst_sock,&mask_sock,rhead);
1663 if (!rt
1664 || rt->rt_dst != dst
1665 || rt->rt_mask != mask)
1666 return 0;
1668 return rt;
1672 /* Find a route to dst as the kernel would.
1674 struct rt_entry *
1675 rtfind(naddr dst)
1677 dst_sock.sin_addr.s_addr = dst;
1678 return (struct rt_entry *)rhead->rnh_matchaddr(&dst_sock, rhead);
1682 /* add a route to the table
1684 void
1685 rtadd(naddr dst,
1686 naddr mask,
1687 u_int state, /* rt_state for the entry */
1688 struct rt_spare *new)
1690 struct rt_entry *rt;
1691 naddr smask;
1692 int i;
1693 struct rt_spare *rts;
1695 rt = (struct rt_entry *)rtmalloc(sizeof (*rt), "rtadd");
1696 memset(rt, 0, sizeof(*rt));
1697 for (rts = rt->rt_spares, i = NUM_SPARES; i != 0; i--, rts++)
1698 rts->rts_metric = HOPCNT_INFINITY;
1700 rt->rt_nodes->rn_key = (caddr_t)&rt->rt_dst_sock;
1701 rt->rt_dst = dst;
1702 rt->rt_dst_sock.sin_family = AF_INET;
1703 #ifdef _HAVE_SIN_LEN
1704 rt->rt_dst_sock.sin_len = dst_sock.sin_len;
1705 #endif
1706 if (mask != HOST_MASK) {
1707 smask = std_mask(dst);
1708 if ((smask & ~mask) == 0 && mask > smask)
1709 state |= RS_SUBNET;
1711 mask_sock.sin_addr.s_addr = htonl(mask);
1712 masktrim(&mask_sock);
1713 rt->rt_mask = mask;
1714 rt->rt_state = state;
1715 rt->rt_spares[0] = *new;
1716 rt->rt_time = now.tv_sec;
1717 rt->rt_poison_metric = HOPCNT_INFINITY;
1718 rt->rt_seqno = update_seqno;
1720 if (++total_routes == MAX_ROUTES)
1721 msglog("have maximum (%d) routes", total_routes);
1722 if (TRACEACTIONS)
1723 trace_add_del("Add", rt);
1725 need_kern.tv_sec = now.tv_sec;
1726 set_need_flash();
1728 if (0 == rhead->rnh_addaddr(&rt->rt_dst_sock, &mask_sock,
1729 rhead, rt->rt_nodes)) {
1730 msglog("rnh_addaddr() failed for %s mask=%#lx",
1731 naddr_ntoa(dst), (u_long)mask);
1732 free(rt);
1737 /* notice a changed route
1739 void
1740 rtchange(struct rt_entry *rt,
1741 u_int state, /* new state bits */
1742 struct rt_spare *new,
1743 char *label)
1745 if (rt->rt_metric != new->rts_metric) {
1746 /* Fix the kernel immediately if it seems the route
1747 * has gone bad, since there may be a working route that
1748 * aggregates this route.
1750 if (new->rts_metric == HOPCNT_INFINITY) {
1751 need_kern.tv_sec = now.tv_sec;
1752 if (new->rts_time >= now.tv_sec - EXPIRE_TIME)
1753 new->rts_time = now.tv_sec - EXPIRE_TIME;
1755 rt->rt_seqno = update_seqno;
1756 set_need_flash();
1759 if (rt->rt_gate != new->rts_gate) {
1760 need_kern.tv_sec = now.tv_sec;
1761 rt->rt_seqno = update_seqno;
1762 set_need_flash();
1765 state |= (rt->rt_state & RS_SUBNET);
1767 /* Keep various things from deciding ageless routes are stale.
1769 if (!AGE_RT(state, new->rts_ifp))
1770 new->rts_time = now.tv_sec;
1772 if (TRACEACTIONS)
1773 trace_change(rt, state, new,
1774 label ? label : "Chg ");
1776 rt->rt_state = state;
1777 rt->rt_spares[0] = *new;
1781 /* check for a better route among the spares
1783 static struct rt_spare *
1784 rts_better(struct rt_entry *rt)
1786 struct rt_spare *rts, *rts1;
1787 int i;
1789 /* find the best alternative among the spares */
1790 rts = rt->rt_spares+1;
1791 for (i = NUM_SPARES, rts1 = rts+1; i > 2; i--, rts1++) {
1792 if (BETTER_LINK(rt,rts1,rts))
1793 rts = rts1;
1796 return rts;
1800 /* switch to a backup route
1802 void
1803 rtswitch(struct rt_entry *rt,
1804 struct rt_spare *rts)
1806 struct rt_spare swap;
1807 char label[10];
1810 /* Do not change permanent routes */
1811 if (0 != (rt->rt_state & (RS_MHOME | RS_STATIC | RS_RDISC
1812 | RS_NET_SYN | RS_IF)))
1813 return;
1815 /* find the best alternative among the spares */
1816 if (rts == NULL)
1817 rts = rts_better(rt);
1819 /* Do not bother if it is not worthwhile.
1821 if (!BETTER_LINK(rt, rts, rt->rt_spares))
1822 return;
1824 swap = rt->rt_spares[0];
1825 sprintf(label, "Use #%d", (int)(rts - rt->rt_spares));
1826 rtchange(rt, rt->rt_state & ~(RS_NET_SYN | RS_RDISC), rts, label);
1827 if (swap.rts_metric == HOPCNT_INFINITY) {
1828 *rts = rts_empty;
1829 } else {
1830 *rts = swap;
1835 void
1836 rtdelete(struct rt_entry *rt)
1838 struct khash *k;
1841 if (TRACEACTIONS)
1842 trace_add_del("Del", rt);
1844 k = kern_find(rt->rt_dst, rt->rt_mask, 0);
1845 if (k != NULL) {
1846 k->k_state |= KS_DELETE;
1847 need_kern.tv_sec = now.tv_sec;
1850 dst_sock.sin_addr.s_addr = rt->rt_dst;
1851 mask_sock.sin_addr.s_addr = htonl(rt->rt_mask);
1852 masktrim(&mask_sock);
1853 if (rt != (struct rt_entry *)rhead->rnh_deladdr(&dst_sock, &mask_sock,
1854 rhead)) {
1855 msglog("rnh_deladdr() failed");
1856 } else {
1857 free(rt);
1858 total_routes--;
1863 void
1864 rts_delete(struct rt_entry *rt,
1865 struct rt_spare *rts)
1867 trace_upslot(rt, rts, &rts_empty);
1868 *rts = rts_empty;
1872 /* Get rid of a bad route, and try to switch to a replacement.
1874 void
1875 rtbad(struct rt_entry *rt)
1877 struct rt_spare new;
1879 /* Poison the route */
1880 new = rt->rt_spares[0];
1881 new.rts_metric = HOPCNT_INFINITY;
1882 rtchange(rt, rt->rt_state & ~(RS_IF | RS_LOCAL | RS_STATIC), &new, 0);
1883 rtswitch(rt, 0);
1887 /* Junk a RS_NET_SYN or RS_LOCAL route,
1888 * unless it is needed by another interface.
1890 void
1891 rtbad_sub(struct rt_entry *rt)
1893 struct interface *ifp, *ifp1;
1894 struct intnet *intnetp;
1895 u_int state;
1898 ifp1 = NULL;
1899 state = 0;
1901 if (rt->rt_state & RS_LOCAL) {
1902 /* Is this the route through loopback for the interface?
1903 * If so, see if it is used by any other interfaces, such
1904 * as a point-to-point interface with the same local address.
1906 for (ifp = ifnet; ifp != NULL; ifp = ifp->int_next) {
1907 /* Retain it if another interface needs it.
1909 if (ifp->int_addr == rt->rt_ifp->int_addr) {
1910 state |= RS_LOCAL;
1911 ifp1 = ifp;
1912 break;
1918 if (!(state & RS_LOCAL)) {
1919 /* Retain RIPv1 logical network route if there is another
1920 * interface that justifies it.
1922 if (rt->rt_state & RS_NET_SYN) {
1923 for (ifp = ifnet; ifp != NULL; ifp = ifp->int_next) {
1924 if ((ifp->int_state & IS_NEED_NET_SYN)
1925 && rt->rt_mask == ifp->int_std_mask
1926 && rt->rt_dst == ifp->int_std_addr) {
1927 state |= RS_NET_SYN;
1928 ifp1 = ifp;
1929 break;
1934 /* or if there is an authority route that needs it. */
1935 for (intnetp = intnets;
1936 intnetp != NULL;
1937 intnetp = intnetp->intnet_next) {
1938 if (intnetp->intnet_addr == rt->rt_dst
1939 && intnetp->intnet_mask == rt->rt_mask) {
1940 state |= (RS_NET_SYN | RS_NET_INT);
1941 break;
1946 if (ifp1 != NULL || (state & RS_NET_SYN)) {
1947 struct rt_spare new = rt->rt_spares[0];
1948 new.rts_ifp = ifp1;
1949 rtchange(rt, ((rt->rt_state & ~(RS_NET_SYN|RS_LOCAL)) | state),
1950 &new, 0);
1951 } else {
1952 rtbad(rt);
1957 /* Called while walking the table looking for sick interfaces
1958 * or after a time change.
1960 /* ARGSUSED */
1962 walk_bad(struct radix_node *rn,
1963 struct walkarg *argp UNUSED)
1965 #define RT ((struct rt_entry *)rn)
1966 struct rt_spare *rts;
1967 int i;
1970 /* fix any spare routes through the interface
1972 rts = RT->rt_spares;
1973 for (i = NUM_SPARES; i != 1; i--) {
1974 rts++;
1975 if (rts->rts_metric < HOPCNT_INFINITY
1976 && (rts->rts_ifp == 0
1977 || (rts->rts_ifp->int_state & IS_BROKE)))
1978 rts_delete(RT, rts);
1981 /* Deal with the main route
1983 /* finished if it has been handled before or if its interface is ok
1985 if (RT->rt_ifp == 0 || !(RT->rt_ifp->int_state & IS_BROKE))
1986 return 0;
1988 /* Bad routes for other than interfaces are easy.
1990 if (0 == (RT->rt_state & (RS_IF | RS_NET_SYN | RS_LOCAL))) {
1991 rtbad(RT);
1992 return 0;
1995 rtbad_sub(RT);
1996 return 0;
1997 #undef RT
2001 /* Check the age of an individual route.
2003 /* ARGSUSED */
2004 static int
2005 walk_age(struct radix_node *rn,
2006 struct walkarg *argp UNUSED)
2008 #define RT ((struct rt_entry *)rn)
2009 struct interface *ifp;
2010 struct rt_spare *rts;
2011 int i;
2014 /* age all of the spare routes, including the primary route
2015 * currently in use
2017 rts = RT->rt_spares;
2018 for (i = NUM_SPARES; i != 0; i--, rts++) {
2020 ifp = rts->rts_ifp;
2021 if (i == NUM_SPARES) {
2022 if (!AGE_RT(RT->rt_state, ifp)) {
2023 /* Keep various things from deciding ageless
2024 * routes are stale
2026 rts->rts_time = now.tv_sec;
2027 continue;
2030 /* forget RIP routes after RIP has been turned off.
2032 if (rip_sock < 0) {
2033 rtdelete(RT);
2034 return 0;
2038 /* age failing routes
2040 if (age_bad_gate == rts->rts_gate
2041 && rts->rts_time >= now_stale) {
2042 rts->rts_time -= SUPPLY_INTERVAL;
2045 /* trash the spare routes when they go bad */
2046 if (rts->rts_metric < HOPCNT_INFINITY
2047 && now_garbage > rts->rts_time
2048 && i != NUM_SPARES)
2049 rts_delete(RT, rts);
2053 /* finished if the active route is still fresh */
2054 if (now_stale <= RT->rt_time)
2055 return 0;
2057 /* try to switch to an alternative */
2058 rtswitch(RT, 0);
2060 /* Delete a dead route after it has been publically mourned. */
2061 if (now_garbage > RT->rt_time) {
2062 rtdelete(RT);
2063 return 0;
2066 /* Start poisoning a bad route before deleting it. */
2067 if (now.tv_sec - RT->rt_time > EXPIRE_TIME) {
2068 struct rt_spare new = RT->rt_spares[0];
2069 new.rts_metric = HOPCNT_INFINITY;
2070 rtchange(RT, RT->rt_state, &new, 0);
2072 return 0;
2076 /* Watch for dead routes and interfaces.
2078 void
2079 age(naddr bad_gate)
2081 struct interface *ifp;
2082 int need_query = 0;
2084 /* If not listening to RIP, there is no need to age the routes in
2085 * the table.
2087 age_timer.tv_sec = (now.tv_sec
2088 + ((rip_sock < 0) ? NEVER : SUPPLY_INTERVAL));
2090 /* Check for dead IS_REMOTE interfaces by timing their
2091 * transmissions.
2093 for (ifp = ifnet; ifp; ifp = ifp->int_next) {
2094 if (!(ifp->int_state & IS_REMOTE))
2095 continue;
2097 /* ignore unreachable remote interfaces */
2098 if (!check_remote(ifp))
2099 continue;
2101 /* Restore remote interface that has become reachable
2103 if (ifp->int_state & IS_BROKE)
2104 if_ok(ifp, "remote ");
2106 if (ifp->int_act_time != NEVER
2107 && now.tv_sec - ifp->int_act_time > EXPIRE_TIME) {
2108 msglog("remote interface %s to %s timed out after"
2109 " %ld:%ld",
2110 ifp->int_name,
2111 naddr_ntoa(ifp->int_dstaddr),
2112 (now.tv_sec - ifp->int_act_time)/60,
2113 (now.tv_sec - ifp->int_act_time)%60);
2114 if_sick(ifp);
2117 /* If we have not heard from the other router
2118 * recently, ask it.
2120 if (now.tv_sec >= ifp->int_query_time) {
2121 ifp->int_query_time = NEVER;
2122 need_query = 1;
2126 /* Age routes. */
2127 age_bad_gate = bad_gate;
2128 rn_walktree(rhead, walk_age, 0);
2130 /* delete old redirected routes to keep the kernel table small
2131 * and prevent blackholes
2133 del_redirects(bad_gate, now.tv_sec-STALE_TIME);
2135 /* Update the kernel routing table. */
2136 fix_kern();
2138 /* poke reticent remote gateways */
2139 if (need_query)
2140 rip_query();