xref: /illumos-gate/usr/src/cmd/cmd-inet/usr.sbin/in.routed/output.c (revision 1e56f352c1c208679012bca47d552e127f5b1072)
1 /*
2  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
3  * Use is subject to license terms.
4  *
5  * Copyright (c) 1983, 1988, 1993
6  *	The Regents of the University of California.  All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. All advertising materials mentioning features or use of this software
17  *    must display the following acknowledgment:
18  *	This product includes software developed by the University of
19  *	California, Berkeley and its contributors.
20  * 4. Neither the name of the University nor the names of its contributors
21  *    may be used to endorse or promote products derived from this software
22  *    without specific prior written permission.
23  *
24  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34  * SUCH DAMAGE.
35  *
36  * $FreeBSD: src/sbin/routed/output.c,v 1.7 2000/08/11 08:24:38 sheldonh Exp $
37  */
38 
39 #include "defs.h"
40 #include <md5.h>
41 #include <alloca.h>
42 
43 uint_t update_seqno;
44 
45 
46 /*
47  * walk the tree of routes with this for output
48  */
49 static	struct {
50 	struct sockaddr_in to;
51 	in_addr_t	to_mask;
52 	in_addr_t	to_net;
53 	in_addr_t	to_std_mask;
54 	in_addr_t	to_std_net;
55 	struct interface *ifp;		/* usually output interface */
56 	struct auth	*a;
57 	uint8_t		metric;		/* adjust metrics by interface */
58 	uint32_t	npackets;
59 	uint32_t	gen_limit;
60 #define	WS_GEN_LIMIT_MAX	1024
61 	uint16_t	state;
62 #define	WS_ST_FLASH	0x001	/* send only changed routes */
63 #define	WS_ST_RIP2_ALL	0x002	/* send full featured RIPv2 */
64 #define	WS_ST_AG	0x004	/* ok to aggregate subnets */
65 #define	WS_ST_SUPER_AG	0x008	/* ok to aggregate networks */
66 #define	WS_ST_QUERY	0x010	/* responding to a query */
67 #define	WS_ST_TO_ON_NET	0x020	/* sending onto one of our nets */
68 #define	WS_ST_DEFAULT	0x040	/* faking a default */
69 } ws;
70 
71 /* A buffer for what can be heard by both RIPv1 and RIPv2 listeners */
72 struct ws_buf v12buf;
73 static	union pkt_buf ripv12_buf;
74 
75 /* Another for only RIPv2 listeners */
76 static	struct ws_buf v2buf;
77 static	union pkt_buf rip_v2_buf;
78 
79 
80 
81 void
82 bufinit(void)
83 {
84 	ripv12_buf.rip.rip_cmd = RIPCMD_RESPONSE;
85 	v12buf.buf = &ripv12_buf.rip;
86 	v12buf.base = &v12buf.buf->rip_nets[0];
87 
88 	rip_v2_buf.rip.rip_cmd = RIPCMD_RESPONSE;
89 	rip_v2_buf.rip.rip_vers = RIPv2;
90 	v2buf.buf = &rip_v2_buf.rip;
91 	v2buf.base = &v2buf.buf->rip_nets[0];
92 }
93 
94 
95 /*
96  * Send the contents of the global buffer via the non-multicast socket
97  */
98 int					/* <0 on failure */
99 output(enum output_type type,
100     struct sockaddr_in *dst,		/* send to here */
101     struct interface *ifp,
102     struct rip *buf,
103     int size)			/* this many bytes */
104 {
105 	struct sockaddr_in sin;
106 	int flags;
107 	const char *msg;
108 	int res;
109 	int ifindex;
110 	struct in_addr addr;
111 
112 	sin = *dst;
113 	if (sin.sin_port == 0)
114 		sin.sin_port = htons(RIP_PORT);
115 
116 	flags = 0;
117 
118 	if (ifp == NULL && type == OUT_MULTICAST) {
119 		msglog("Cannot send RIP message to %s",
120 		    inet_ntoa(sin.sin_addr));
121 		return (-1);
122 	}
123 
124 	switch (type) {
125 	case OUT_QUERY:
126 		msg = "Answer Query";
127 		break;
128 	case OUT_UNICAST:
129 		msg = "Send";
130 		flags = MSG_DONTROUTE;
131 		break;
132 	case OUT_BROADCAST:
133 		msg = "Send bcast";
134 		break;
135 	case OUT_MULTICAST:
136 		msg = "Send mcast";
137 		break;
138 
139 	case NO_OUT_MULTICAST:
140 	case NO_OUT_RIPV2:
141 	default:
142 #ifdef DEBUG
143 		abort();
144 #endif
145 		return (-1);
146 	}
147 
148 	/*
149 	 * IP_PKTINFO overrides IP_MULTICAST_IF, so we don't set ifindex
150 	 * for multicast traffic.
151 	 */
152 	ifindex = (type != OUT_MULTICAST && type != OUT_QUERY &&
153 	    ifp != NULL && ifp->int_phys != NULL) ?
154 	    ifp->int_phys->phyi_index : 0;
155 
156 	if (rip_sock_interface != ifp) {
157 		/*
158 		 * For multicast, we have to choose the source
159 		 * address.  This is either the local address
160 		 * (non-point-to-point) or the remote address.
161 		 */
162 		if (ifp != NULL) {
163 			addr.s_addr = (ifp->int_if_flags & IFF_POINTOPOINT) ?
164 			    ifp->int_dstaddr : ifp->int_addr;
165 			if (type == OUT_MULTICAST &&
166 			    setsockopt(rip_sock, IPPROTO_IP,
167 			    IP_MULTICAST_IF, &addr, sizeof (addr)) == -1) {
168 				LOGERR("setsockopt(rip_sock, IP_MULTICAST_IF)");
169 				return (-1);
170 			}
171 		}
172 		rip_sock_interface = ifp;
173 	}
174 
175 	trace_rip(msg, "to", &sin, ifp, buf, size);
176 
177 	res = sendtoif(rip_sock, buf, size, flags, &sin, ifindex);
178 	if (res < 0 && (ifp == NULL || !(ifp->int_state & IS_BROKE))) {
179 		writelog(LOG_WARNING, "%s sendto(%s%s%s.%d): %s", msg,
180 		    ifp != NULL ? ifp->int_name : "",
181 		    ifp != NULL ? ", " : "",
182 		    inet_ntoa(sin.sin_addr),
183 		    ntohs(sin.sin_port),
184 		    rip_strerror(errno));
185 	}
186 
187 	return (res);
188 }
189 
190 /*
191  * Semantically identical to sendto(), but sends the message through a
192  * specific interface (if ifindex is non-zero) using IP_PKTINFO.
193  */
194 int
195 sendtoif(int fd, const void *buf, uint_t bufsize, uint_t flags,
196     struct sockaddr_in *sinp, uint_t ifindex)
197 {
198 	struct iovec iov;
199 	struct msghdr msg;
200 	struct cmsghdr *cmsgp;
201 	struct in_pktinfo *ipip;
202 
203 	iov.iov_base = (void *)buf;
204 	iov.iov_len = bufsize;
205 
206 	(void) memset(&msg, 0, sizeof (struct msghdr));
207 	msg.msg_name = (struct sockaddr *)sinp;
208 	msg.msg_namelen = sizeof (struct sockaddr_in);
209 	msg.msg_iov = &iov;
210 	msg.msg_iovlen = 1;
211 
212 	if (ifindex != 0) {
213 		/*
214 		 * We can't precisely predict the alignment padding we'll
215 		 * need, so we allocate the maximum alignment and then
216 		 * use CMSG_NXTHDR() to fix it up at the end.
217 		 */
218 		msg.msg_controllen = sizeof (*cmsgp) + _MAX_ALIGNMENT +
219 		    sizeof (*ipip) + _MAX_ALIGNMENT + sizeof (*cmsgp);
220 		msg.msg_control = alloca(msg.msg_controllen);
221 
222 		cmsgp = CMSG_FIRSTHDR(&msg);
223 		ipip = (void *)CMSG_DATA(cmsgp);
224 		(void) memset(ipip, 0, sizeof (struct in_pktinfo));
225 		ipip->ipi_ifindex = ifindex;
226 		cmsgp->cmsg_len = (caddr_t)(ipip + 1) - (caddr_t)cmsgp;
227 		cmsgp->cmsg_type = IP_PKTINFO;
228 		cmsgp->cmsg_level = IPPROTO_IP;
229 
230 		/*
231 		 * Correct the control message length.
232 		 */
233 		cmsgp = CMSG_NXTHDR(&msg, cmsgp);
234 		msg.msg_controllen = (caddr_t)cmsgp - (caddr_t)msg.msg_control;
235 	}
236 
237 	return (sendmsg(fd, &msg, flags));
238 }
239 
240 /*
241  * Find the first key for a packet to send.
242  * Try for a key that is eligible and has not expired, but settle for
243  * the last key if they have all expired.
244  * If no key is ready yet, give up.
245  */
246 struct auth *
247 find_auth(struct interface *ifp)
248 {
249 	struct auth *ap, *res = NULL;
250 	int i;
251 
252 
253 	if (ifp == NULL)
254 		return (NULL);
255 
256 	if ((ap = ifp->int_auth) == NULL)
257 		return (NULL);
258 
259 	for (i = 0; i < MAX_AUTH_KEYS; i++, ap++) {
260 		/* stop looking after the last key */
261 		if (ap->type == RIP_AUTH_NONE)
262 			break;
263 
264 		/* ignore keys that are not ready yet */
265 		if ((ulong_t)ap->start > (ulong_t)clk.tv_sec)
266 			continue;
267 
268 		if ((ulong_t)ap->end < (ulong_t)clk.tv_sec) {
269 			/* note best expired password as a fall-back */
270 			if (res == NULL ||
271 			    (((ulong_t)ap->end > (ulong_t)res->end)) &&
272 			    ((ulong_t)res->end < (ulong_t)clk.tv_sec))
273 				res = ap;
274 			continue;
275 		}
276 
277 		/* note key with the best future */
278 		if (res == NULL || (ulong_t)res->end < (ulong_t)ap->end)
279 			res = ap;
280 	}
281 	return (res);
282 }
283 
284 
285 void
286 clr_ws_buf(struct ws_buf *wb, struct auth *ap)
287 {
288 	struct netauth *na;
289 
290 	wb->lim = wb->base + NETS_LEN;
291 	wb->n = wb->base;
292 	(void) memset(wb->n, 0, NETS_LEN*sizeof (*wb->n));
293 
294 	/*
295 	 * (start to) install authentication if appropriate
296 	 */
297 	if (ap == NULL)
298 		return;
299 
300 	na = (struct netauth *)wb->n;
301 	if (ap->type == RIP_AUTH_PW) {
302 		na->a_family = RIP_AF_AUTH;
303 		na->a_type = RIP_AUTH_PW;
304 		(void) memcpy(na->au.au_pw, ap->key, sizeof (na->au.au_pw));
305 		wb->n++;
306 
307 	} else if (ap->type ==  RIP_AUTH_MD5) {
308 		na->a_family = RIP_AF_AUTH;
309 		na->a_type = RIP_AUTH_MD5;
310 		na->au.a_md5.md5_keyid = ap->keyid;
311 		na->au.a_md5.md5_auth_len = RIP_AUTH_MD5_LEN;
312 		na->au.a_md5.md5_seqno = htonl(clk.tv_sec);
313 		wb->n++;
314 		wb->lim--;		/* make room for trailer */
315 	}
316 }
317 
318 
319 void
320 end_md5_auth(struct ws_buf *wb, struct auth *ap)
321 {
322 	struct netauth *na, *na2;
323 	MD5_CTX md5_ctx;
324 	int len;
325 
326 	na = (struct netauth *)wb->base;
327 	na2 = (struct netauth *)wb->n;
328 	len = (char *)na2-(char *)wb->buf;
329 	na2->a_family = RIP_AF_AUTH;
330 	na2->a_type = RIP_AUTH_TRAILER;
331 	na->au.a_md5.md5_pkt_len = htons(len);
332 	MD5Init(&md5_ctx);
333 	/* len+4 to include auth trailer's family/type in MD5 sum */
334 	MD5Update(&md5_ctx, (uchar_t *)wb->buf, len + 4);
335 	MD5Update(&md5_ctx, ap->key, RIP_AUTH_MD5_LEN);
336 	MD5Final(na2->au.au_pw, &md5_ctx);
337 	wb->n++;
338 }
339 
340 
341 /*
342  * Send the buffer
343  */
344 static void
345 supply_write(struct ws_buf *wb)
346 {
347 	/*
348 	 * Output multicast only if legal.
349 	 * If we would multicast and it would be illegal, then discard the
350 	 * packet.
351 	 */
352 	switch (wb->type) {
353 	case NO_OUT_MULTICAST:
354 		trace_pkt("skip multicast to %s because impossible",
355 		    naddr_ntoa(ws.to.sin_addr.s_addr));
356 		break;
357 	case NO_OUT_RIPV2:
358 		break;
359 	default:
360 		if (ws.a != NULL && ws.a->type == RIP_AUTH_MD5)
361 			end_md5_auth(wb, ws.a);
362 		if (output(wb->type, &ws.to, ws.ifp, wb->buf,
363 		    ((char *)wb->n - (char *)wb->buf)) < 0 && ws.ifp != NULL)
364 			if_sick(ws.ifp, _B_FALSE);
365 		ws.npackets++;
366 		break;
367 	}
368 
369 	clr_ws_buf(wb, ws.a);
370 }
371 
372 
373 /*
374  * Put an entry into the packet
375  */
376 static void
377 supply_out(struct ag_info *ag)
378 {
379 	uint32_t dstcount;
380 	in_addr_t mask, v1_mask, dst_h, ddst_h = 0;
381 	struct ws_buf *wb;
382 
383 
384 	/*
385 	 * Skip this route if doing a flash update and it and the routes
386 	 * it aggregates have not changed recently.
387 	 */
388 	if (ag->ag_seqno < update_seqno && (ws.state & WS_ST_FLASH))
389 		return;
390 
391 	dst_h = ag->ag_dst_h;
392 	mask = ag->ag_mask;
393 	v1_mask = ripv1_mask_host(htonl(dst_h),
394 	    (ws.state & WS_ST_TO_ON_NET) ? ws.ifp : NULL);
395 	dstcount = 0;
396 
397 	/*
398 	 * If we are sending RIPv2 packets that cannot (or must not) be
399 	 * heard by RIPv1 listeners, do not worry about sub- or supernets.
400 	 * Subnets (from other networks) can only be sent via multicast.
401 	 * A pair of subnet routes might have been promoted so that they
402 	 * are legal to send by RIPv1.
403 	 * If RIPv1 is off, use the multicast buffer.
404 	 */
405 	if ((ws.state & WS_ST_RIP2_ALL) ||
406 	    ((ag->ag_state & AGS_RIPV2) && v1_mask != mask)) {
407 		/* use the RIPv2-only buffer */
408 		wb = &v2buf;
409 
410 	} else {
411 		/*
412 		 * use the RIPv1-or-RIPv2 buffer
413 		 */
414 		wb = &v12buf;
415 
416 		/*
417 		 * Convert supernet route into corresponding set of network
418 		 * routes for RIPv1, but leave non-contiguous netmasks
419 		 * to ag_check().
420 		 */
421 		if (v1_mask > mask &&
422 		    mask + (mask & -mask) == 0) {
423 			ddst_h = v1_mask & -v1_mask;
424 			dstcount = (v1_mask & ~mask)/ddst_h;
425 
426 			if (dstcount > ws.gen_limit) {
427 				/*
428 				 * Punt if we would have to generate an
429 				 * unreasonable number of routes.
430 				 */
431 				if (TRACECONTENTS)
432 					trace_misc("sending %s-->%s as 1"
433 					    " instead of %d routes",
434 					    addrname(htonl(dst_h), mask, 1),
435 					    naddr_ntoa(ws.to.sin_addr.s_addr),
436 					    dstcount + 1);
437 				dstcount = 0;
438 
439 			} else {
440 				mask = v1_mask;
441 				ws.gen_limit -= dstcount;
442 			}
443 		}
444 	}
445 
446 	do {
447 		wb->n->n_family = RIP_AF_INET;
448 		wb->n->n_dst = htonl(dst_h);
449 		/*
450 		 * If the route is from router-discovery or we are
451 		 * shutting down, or this is a broken/sick interface,
452 		 * admit only a bad metric.
453 		 */
454 		wb->n->n_metric = ((stopint || ag->ag_metric < 1 ||
455 		    (ag->ag_ifp && (ag->ag_ifp->int_state &
456 		    (IS_BROKE|IS_SICK)))) ? HOPCNT_INFINITY : ag->ag_metric);
457 		wb->n->n_metric = htonl(wb->n->n_metric);
458 		/*
459 		 * Any non-zero bits in the supposedly unused RIPv1 fields
460 		 * cause the old `routed` to ignore the route.
461 		 * That means the mask and so forth cannot be sent
462 		 * in the hybrid RIPv1/RIPv2 mode.
463 		 */
464 		if (ws.state & WS_ST_RIP2_ALL) {
465 			if (ag->ag_nhop != 0 &&
466 			    ((ws.state & WS_ST_QUERY) ||
467 			    (ag->ag_nhop != ws.ifp->int_addr &&
468 			    on_net(ag->ag_nhop, ws.ifp->int_net,
469 			    ws.ifp->int_mask)) &&
470 			    ifwithaddr(ag->ag_nhop, _B_FALSE, _B_FALSE) ==
471 			    NULL))
472 				wb->n->n_nhop = ag->ag_nhop;
473 			wb->n->n_mask = htonl(mask);
474 			wb->n->n_tag = ag->ag_tag;
475 		}
476 		dst_h += ddst_h;
477 
478 		if (++wb->n >= wb->lim)
479 			supply_write(wb);
480 	} while (dstcount-- > 0);
481 }
482 
483 
484 /*
485  * Supply one route from the table
486  */
487 /* ARGSUSED */
488 static int
489 walk_supply(struct radix_node *rn, void *argp)
490 {
491 #define	RT ((struct rt_entry *)rn)
492 	ushort_t ags;
493 	uint8_t metric, pref;
494 	in_addr_t dst, nhop;
495 	struct rt_spare *rts;
496 	uint_t sparecount;
497 
498 
499 	/*
500 	 * Do not advertise external remote interfaces or passive interfaces.
501 	 */
502 	if ((RT->rt_state & RS_IF) && RT->rt_ifp != NULL &&
503 	    (RT->rt_ifp->int_state & IS_PASSIVE) &&
504 	    !(RT->rt_state & RS_MHOME))
505 		return (0);
506 	/*
507 	 * Do not advertise routes learnt from /etc/gateways.
508 	 */
509 	if (RT->rt_spares[0].rts_origin == RO_FILE)
510 		return (0);
511 
512 	/*
513 	 * Do not advertise routes which would lead to forwarding on a
514 	 * non-forwarding interface.
515 	 */
516 	if (RT->rt_state & RS_NOPROPAGATE)
517 		return (0);
518 
519 	/*
520 	 * If being quiet about our ability to forward, then
521 	 * do not say anything unless responding to a query,
522 	 * except about our main interface.
523 	 */
524 	if (!should_supply(NULL) && !(ws.state & WS_ST_QUERY) &&
525 	    !(RT->rt_state & RS_MHOME))
526 		return (0);
527 
528 	dst = RT->rt_dst;
529 
530 	/*
531 	 * do not collide with the fake default route
532 	 */
533 	if (dst == RIP_DEFAULT && (ws.state & WS_ST_DEFAULT))
534 		return (0);
535 
536 	if (RT->rt_state & RS_NET_SYN) {
537 		if (RT->rt_state & RS_NET_INT) {
538 			/*
539 			 * Do not send manual synthetic network routes
540 			 * into the subnet.
541 			 */
542 			if (on_net(ws.to.sin_addr.s_addr,
543 			    ntohl(dst), RT->rt_mask))
544 				return (0);
545 
546 		} else {
547 			/*
548 			 * Do not send automatic synthetic network routes
549 			 * if they are not needed because no RIPv1 listeners
550 			 * can hear them.
551 			 */
552 			if (ws.state & WS_ST_RIP2_ALL)
553 				return (0);
554 
555 			/*
556 			 * Do not send automatic synthetic network routes to
557 			 * the real subnet.
558 			 */
559 			if (on_net(ws.to.sin_addr.s_addr,
560 			    ntohl(dst), RT->rt_mask))
561 				return (0);
562 		}
563 		nhop = 0;
564 
565 	} else {
566 		/*
567 		 * Advertise the next hop if this is not a route for one
568 		 * of our interfaces and the next hop is on the same
569 		 * network as the target.
570 		 * The final determination is made by supply_out().
571 		 */
572 		if (!(RT->rt_state & RS_IF) && !(RT->rt_state & RS_MHOME) &&
573 		    RT->rt_gate != loopaddr)
574 			nhop = RT->rt_gate;
575 		else
576 			nhop = 0;
577 	}
578 
579 	metric = RT->rt_metric;
580 	ags = 0;
581 
582 	if (!RT_ISHOST(RT)) {
583 		/*
584 		 * Always suppress network routes into other, existing
585 		 * network routes
586 		 */
587 		ags |= AGS_SUPPRESS;
588 
589 		/*
590 		 * Generate supernets if allowed.
591 		 * If we can be heard by RIPv1 systems, we will
592 		 * later convert back to ordinary nets.
593 		 * This unifies dealing with received supernets.
594 		 */
595 		if ((ws.state & WS_ST_AG) && ((RT->rt_state & RS_SUBNET) ||
596 		    (ws.state & WS_ST_SUPER_AG)))
597 			ags |= AGS_AGGREGATE;
598 	} else if (!(RT->rt_state & RS_MHOME)) {
599 		/*
600 		 * We should always suppress (into existing network routes)
601 		 * the host routes for the local end of our point-to-point
602 		 * links.
603 		 * If we are suppressing host routes in general, then do so.
604 		 * Avoid advertising host routes onto their own network,
605 		 * where they should be handled by proxy-ARP.
606 		 */
607 		if ((RT->rt_state & RS_LOCAL) || ridhosts ||
608 		    on_net(dst, ws.to_net, ws.to_mask))
609 			ags |= AGS_SUPPRESS;
610 
611 		/*
612 		 * Aggregate stray host routes into network routes if allowed.
613 		 * We cannot aggregate host routes into small network routes
614 		 * without confusing RIPv1 listeners into thinking the
615 		 * network routes are host routes.
616 		 */
617 		if ((ws.state & WS_ST_AG) && (ws.state & WS_ST_RIP2_ALL))
618 			ags |= AGS_AGGREGATE;
619 	}
620 
621 	/*
622 	 * Do not send RIPv1 advertisements of subnets to other
623 	 * networks. If possible, multicast them by RIPv2.
624 	 */
625 	if ((RT->rt_state & RS_SUBNET) && !(ws.state & WS_ST_RIP2_ALL) &&
626 	    !on_net(dst, ws.to_std_net, ws.to_std_mask))
627 		ags |= AGS_RIPV2 | AGS_AGGREGATE;
628 
629 
630 	/*
631 	 * Do not send a route back to where it came from, except in
632 	 * response to a query.  This is "split-horizon".  That means not
633 	 * advertising back to the same network	and so via the same interface.
634 	 *
635 	 * We want to suppress routes that might have been fragmented
636 	 * from this route by a RIPv1 router and sent back to us, and so we
637 	 * cannot forget this route here.  Let the split-horizon route
638 	 * suppress the fragmented routes and then itself be forgotten.
639 	 *
640 	 * Include the routes for both ends of point-to-point interfaces
641 	 * among those suppressed by split-horizon, since the other side
642 	 * should knows them as well as we do.
643 	 *
644 	 * Notice spare routes with the same metric that we are about to
645 	 * advertise, to split the horizon on redundant, inactive paths.
646 	 */
647 	if (ws.ifp != NULL && !(ws.state & WS_ST_QUERY) &&
648 	    (ws.state & WS_ST_TO_ON_NET) && (!(RT->rt_state & RS_IF) ||
649 	    (ws.ifp->int_if_flags & IFF_POINTOPOINT))) {
650 		for (rts = RT->rt_spares, sparecount = 0;
651 		    sparecount < RT->rt_num_spares; sparecount++, rts++) {
652 			if (rts->rts_metric > metric || rts->rts_ifp != ws.ifp)
653 				continue;
654 
655 			/*
656 			 * If we do not mark the route with AGS_SPLIT_HZ here,
657 			 * it will be poisoned-reverse, or advertised back
658 			 * toward its source with an infinite metric.
659 			 * If we have recently advertised the route with a
660 			 * better metric than we now have, then we should
661 			 * poison-reverse the route before suppressing it for
662 			 * split-horizon.
663 			 *
664 			 * In almost all cases, if there is no spare for the
665 			 * route then it is either old and dead or a brand
666 			 * new route. If it is brand new, there is no need
667 			 * for poison-reverse. If it is old and dead, it
668 			 * is already poisoned.
669 			 */
670 			if (RT->rt_poison_time < now_expire ||
671 			    RT->rt_poison_metric >= metric ||
672 			    RT->rt_spares[1].rts_gate == 0) {
673 				ags |= AGS_SPLIT_HZ;
674 				ags &= ~AGS_SUPPRESS;
675 			}
676 			metric = HOPCNT_INFINITY;
677 			break;
678 		}
679 	}
680 
681 	/*
682 	 * Keep track of the best metric with which the
683 	 * route has been advertised recently.
684 	 */
685 	if (RT->rt_poison_metric >= metric ||
686 	    RT->rt_poison_time < now_expire) {
687 		RT->rt_poison_time = now.tv_sec;
688 		RT->rt_poison_metric = metric;
689 	}
690 
691 	/*
692 	 * Adjust the outgoing metric by the cost of the link.
693 	 * Avoid aggregation when a route is counting to infinity.
694 	 */
695 	pref = RT->rt_poison_metric + ws.metric;
696 	metric += ws.metric;
697 
698 	/*
699 	 * If this is a static route pointing to the same interface
700 	 * upon which we are sending out the RIP RESPONSE
701 	 * adjust the preference so that we don't aggregate into this
702 	 * route. Note that the maximum possible hop count on a route
703 	 * per RFC 2453 is 16 (HOPCNT_INFINITY)
704 	 */
705 	if ((RT->rt_state & RS_STATIC) && (ws.ifp == RT->rt_ifp))
706 		pref = (HOPCNT_INFINITY+1);
707 
708 	/*
709 	 * Do not advertise stable routes that will be ignored,
710 	 * unless we are answering a query.
711 	 * If the route recently was advertised with a metric that
712 	 * would have been less than infinity through this interface,
713 	 * we need to continue to advertise it in order to poison it.
714 	 */
715 	if (metric >= HOPCNT_INFINITY) {
716 		if (!(ws.state & WS_ST_QUERY) && (pref >= HOPCNT_INFINITY ||
717 		    RT->rt_poison_time < now_garbage))
718 			return (0);
719 
720 		metric = HOPCNT_INFINITY;
721 	}
722 
723 	/*
724 	 * supply this route out on the wire- we only care about dest/mask
725 	 * and so can ignore all rt_spares[i] with i > 0
726 	 */
727 	ag_check(dst, RT->rt_mask, 0, RT->rt_ifp, nhop, metric, pref,
728 	    RT->rt_seqno, RT->rt_tag, ags, supply_out);
729 	return (0);
730 #undef RT
731 }
732 
733 
734 /*
735  * Supply dst with the contents of the routing tables.
736  * If this won't fit in one packet, chop it up into several.
737  */
738 void
739 supply(struct sockaddr_in *dst,
740     struct interface *ifp,	/* output interface */
741     enum output_type type,
742     int flash,			/* 1=flash update */
743     int vers,			/* RIP version */
744     boolean_t passwd_ok)	/* OK to include cleartext password */
745 {
746 	struct rt_entry *rt;
747 	uint8_t def_metric;
748 
749 
750 	ws.state = 0;
751 	ws.gen_limit = WS_GEN_LIMIT_MAX;
752 
753 	ws.to = *dst;
754 	ws.to_std_mask = std_mask(ws.to.sin_addr.s_addr);
755 	ws.to_std_net = ntohl(ws.to.sin_addr.s_addr) & ws.to_std_mask;
756 
757 	if (ifp != NULL) {
758 		ws.to_mask = ifp->int_mask;
759 		ws.to_net = ifp->int_net;
760 		if (on_net(ws.to.sin_addr.s_addr, ws.to_net, ws.to_mask) ||
761 		    type == OUT_MULTICAST)
762 			ws.state |= WS_ST_TO_ON_NET;
763 
764 	} else {
765 		ws.to_mask = ripv1_mask_net(ws.to.sin_addr.s_addr, NULL);
766 		ws.to_net = ntohl(ws.to.sin_addr.s_addr) & ws.to_mask;
767 		rt = rtfind(dst->sin_addr.s_addr);
768 		if (rt != NULL)
769 			ifp = rt->rt_ifp;
770 		else
771 			return;
772 	}
773 
774 	ws.npackets = 0;
775 	if (flash)
776 		ws.state |= WS_ST_FLASH;
777 
778 	ws.ifp = ifp;
779 
780 	/*
781 	 * Routes in the table were already adjusted by their respective
782 	 * destination interface costs (which are zero by default) on
783 	 * input.  The following is the value by which each route's metric
784 	 * will be bumped up on output.
785 	 */
786 	ws.metric = 1;
787 
788 	ripv12_buf.rip.rip_vers = vers;
789 
790 	switch (type) {
791 	case OUT_MULTICAST:
792 		if (ifp->int_if_flags & IFF_MULTICAST)
793 			v2buf.type = OUT_MULTICAST;
794 		else
795 			v2buf.type = NO_OUT_MULTICAST;
796 		v12buf.type = OUT_BROADCAST;
797 		break;
798 
799 	case OUT_QUERY:
800 		ws.state |= WS_ST_QUERY;
801 		/* FALLTHROUGH */
802 	case OUT_BROADCAST:
803 	case OUT_UNICAST:
804 		v2buf.type = (vers == RIPv2) ? type : NO_OUT_RIPV2;
805 		v12buf.type = type;
806 		break;
807 
808 	case NO_OUT_MULTICAST:
809 	case NO_OUT_RIPV2:
810 		return;			/* no output */
811 	}
812 
813 	if (vers == RIPv2) {
814 		/* full RIPv2 only if cannot be heard by RIPv1 listeners */
815 		if (type != OUT_BROADCAST)
816 			ws.state |= WS_ST_RIP2_ALL;
817 		if ((ws.state & WS_ST_QUERY) || !(ws.state & WS_ST_TO_ON_NET)) {
818 			ws.state |= (WS_ST_AG | WS_ST_SUPER_AG);
819 		} else if (ifp == NULL || !(ifp->int_state & IS_NO_AG)) {
820 			ws.state |= WS_ST_AG;
821 			if (type != OUT_BROADCAST && (ifp == NULL ||
822 			    !(ifp->int_state & IS_NO_SUPER_AG)))
823 				ws.state |= WS_ST_SUPER_AG;
824 		}
825 
826 		/* See if this packet needs authenticating */
827 		ws.a = find_auth(ifp);
828 		if (!passwd_ok && ws.a != NULL && ws.a->type == RIP_AUTH_PW)
829 			ws.a = NULL;
830 		if (ws.a != NULL && (ulong_t)ws.a->end < (ulong_t)clk.tv_sec &&
831 		    !ws.a->warnedflag) {
832 			/*
833 			 * If the best key is an expired one, we may as
834 			 * well use it.  Log this event.
835 			 */
836 			writelog(LOG_WARNING,
837 			    "Using expired auth while transmitting to %s",
838 			    naddr_ntoa(ws.to.sin_addr.s_addr));
839 			ws.a->warnedflag = 1;
840 		}
841 	} else {
842 		ws.a = NULL;
843 	}
844 
845 	clr_ws_buf(&v12buf, ws.a);
846 	clr_ws_buf(&v2buf, ws.a);
847 
848 	/*
849 	 * Fake a default route if asked and if there is not already
850 	 * a better, real default route.
851 	 */
852 	if (should_supply(NULL) && (def_metric = ifp->int_d_metric) != 0) {
853 		if (NULL == (rt = rtget(RIP_DEFAULT, 0)) ||
854 		    rt->rt_metric+ws.metric >= def_metric) {
855 			ws.state |= WS_ST_DEFAULT;
856 			ag_check(0, 0, 0, NULL, 0, def_metric, def_metric,
857 			    0, 0, 0, supply_out);
858 		} else {
859 			def_metric = rt->rt_metric+ws.metric;
860 		}
861 
862 		/*
863 		 * If both RIPv2 and the poor-man's router discovery
864 		 * kludge are on, arrange to advertise an extra
865 		 * default route via RIPv1.
866 		 */
867 		if ((ws.state & WS_ST_RIP2_ALL) &&
868 		    (ifp->int_state & IS_PM_RDISC)) {
869 			ripv12_buf.rip.rip_vers = RIPv1;
870 			v12buf.n->n_family = RIP_AF_INET;
871 			v12buf.n->n_dst = htonl(RIP_DEFAULT);
872 			v12buf.n->n_metric = htonl(def_metric);
873 			v12buf.n++;
874 		}
875 	}
876 
877 	(void) rn_walktree(rhead, walk_supply, NULL);
878 	ag_flush(0, 0, supply_out);
879 
880 	/*
881 	 * Flush the packet buffers, provided they are not empty and
882 	 * do not contain only the password.
883 	 */
884 	if (v12buf.n != v12buf.base &&
885 	    (v12buf.n > v12buf.base+1 ||
886 	    v12buf.base->n_family != RIP_AF_AUTH))
887 		supply_write(&v12buf);
888 	if (v2buf.n != v2buf.base && (v2buf.n > v2buf.base+1 ||
889 	    v2buf.base->n_family != RIP_AF_AUTH))
890 		supply_write(&v2buf);
891 
892 	/*
893 	 * If we sent nothing and this is an answer to a query, send
894 	 * an empty buffer.
895 	 */
896 	if (ws.npackets == 0 && (ws.state & WS_ST_QUERY)) {
897 		supply_write(&v2buf);
898 		if (ws.npackets == 0)
899 			supply_write(&v12buf);
900 	}
901 }
902 
903 
904 /*
905  * send all of the routing table or just do a flash update
906  */
907 void
908 rip_bcast(int flash)
909 {
910 	static struct sockaddr_in dst = {AF_INET};
911 	struct interface *ifp;
912 	enum output_type type;
913 	int vers;
914 	struct timeval rtime;
915 
916 
917 	need_flash = _B_FALSE;
918 	intvl_random(&rtime, MIN_WAITTIME, MAX_WAITTIME);
919 	no_flash = rtime;
920 	timevaladd(&no_flash, &now);
921 
922 	if (!rip_enabled)
923 		return;
924 
925 	trace_act("send %s and inhibit dynamic updates for %.3f sec",
926 	    flash ? "dynamic update" : "all routes",
927 	    rtime.tv_sec + ((double)rtime.tv_usec)/1000000.0);
928 
929 	for (ifp = ifnet; ifp != NULL; ifp = ifp->int_next) {
930 		/*
931 		 * Skip interfaces not doing RIP or for which IP
932 		 * forwarding isn't turned on.  Skip duplicate
933 		 * interfaces, we don't want to generate duplicate
934 		 * packets.  Do try broken interfaces to see if they
935 		 * have healed.
936 		 */
937 		if (IS_RIP_OUT_OFF(ifp->int_state) ||
938 		    (ifp->int_state & IS_DUP) ||
939 		    !IS_IFF_ROUTING(ifp->int_if_flags))
940 			continue;
941 
942 		/* skip turned off interfaces */
943 		if (!IS_IFF_UP(ifp->int_if_flags))
944 			continue;
945 
946 		/* skip interfaces we shouldn't use */
947 		if (IS_IFF_QUIET(ifp->int_if_flags))
948 			continue;
949 
950 		vers = (ifp->int_state & IS_NO_RIPV1_OUT) ? RIPv2 : RIPv1;
951 		dst.sin_addr.s_addr = ifp->int_ripout_addr;
952 
953 		/*
954 		 * Ignore the interface if it's not broadcast,
955 		 * point-to-point, or remote.  It must be non-broadcast
956 		 * multiaccess, and therefore unsupported.
957 		 */
958 		if (!(ifp->int_if_flags & (IFF_BROADCAST | IFF_POINTOPOINT)) &&
959 		    !(ifp->int_state & IS_REMOTE))
960 			continue;
961 
962 		type = (ifp->int_if_flags & IFF_BROADCAST) ?
963 		    OUT_BROADCAST : OUT_UNICAST;
964 		if (vers == RIPv2 && (ifp->int_if_flags & IFF_MULTICAST) &&
965 		    !(ifp->int_state & IS_NO_RIP_MCAST))
966 			type = OUT_MULTICAST;
967 
968 		supply(&dst, ifp, type, flash, vers, _B_TRUE);
969 	}
970 
971 	update_seqno++;			/* all routes are up to date */
972 }
973 
974 
975 /*
976  * Ask for routes
977  * Do it only once to an interface, and not even after the interface
978  * was broken and recovered.
979  */
980 void
981 rip_query(void)
982 {
983 	static struct sockaddr_in dst = {AF_INET};
984 	struct interface *ifp;
985 	struct rip buf;
986 	enum output_type type;
987 
988 
989 	if (!rip_enabled)
990 		return;
991 
992 	(void) memset(&buf, 0, sizeof (buf));
993 
994 	for (ifp = ifnet; ifp; ifp = ifp->int_next) {
995 		/*
996 		 * Skip interfaces those already queried.  Do not ask
997 		 * via interfaces through which we don't accept input.
998 		 * Do not ask via interfaces that cannot send RIP
999 		 * packets.  Don't send queries on duplicate
1000 		 * interfaces, that would generate duplicate packets
1001 		 * on link.  Do try broken interfaces to see if they
1002 		 * have healed.
1003 		 */
1004 		if (IS_RIP_IN_OFF(ifp->int_state) ||
1005 		    (ifp->int_state & IS_DUP) ||
1006 		    ifp->int_query_time != NEVER)
1007 			continue;
1008 
1009 		/* skip turned off interfaces */
1010 		if (!IS_IFF_UP(ifp->int_if_flags))
1011 			continue;
1012 
1013 		/* skip interfaces we shouldn't use */
1014 		if (IS_IFF_QUIET(ifp->int_if_flags))
1015 			continue;
1016 
1017 		/*
1018 		 * Ignore the interface if it's not broadcast,
1019 		 * point-to-point, or remote.  It must be non-broadcast
1020 		 * multiaccess, and therefore unsupported.
1021 		 */
1022 		if (!(ifp->int_if_flags & (IFF_BROADCAST | IFF_POINTOPOINT)) &&
1023 		    !(ifp->int_state & IS_REMOTE))
1024 			continue;
1025 
1026 		buf.rip_cmd = RIPCMD_REQUEST;
1027 		buf.rip_nets[0].n_family = RIP_AF_UNSPEC;
1028 		buf.rip_nets[0].n_metric = htonl(HOPCNT_INFINITY);
1029 
1030 		/*
1031 		 * Send a RIPv1 query only if allowed and if we will
1032 		 * listen to RIPv1 routers.
1033 		 */
1034 		if ((ifp->int_state & IS_NO_RIPV1_OUT) ||
1035 		    (ifp->int_state & IS_NO_RIPV1_IN)) {
1036 			buf.rip_vers = RIPv2;
1037 		} else {
1038 			buf.rip_vers = RIPv1;
1039 		}
1040 
1041 		dst.sin_addr.s_addr = ifp->int_ripout_addr;
1042 
1043 		type = (ifp->int_if_flags & IFF_BROADCAST) ?
1044 		    OUT_BROADCAST : OUT_UNICAST;
1045 		if (buf.rip_vers == RIPv2 &&
1046 		    (ifp->int_if_flags & IFF_MULTICAST) &&
1047 		    !(ifp->int_state & IS_NO_RIP_MCAST))
1048 			type = OUT_MULTICAST;
1049 
1050 		ifp->int_query_time = now.tv_sec+SUPPLY_INTERVAL;
1051 		if (output(type, &dst, ifp, &buf, sizeof (buf)) < 0)
1052 			if_sick(ifp, _B_FALSE);
1053 	}
1054 }
1055