xref: /netbsd-src/usr.sbin/ldpd/socketops.c (revision 5bbd2a12505d72a8177929a37b5cee489d0a1cfd)
1 /* $NetBSD: socketops.c,v 1.11 2011/08/31 13:32:38 joerg Exp $ */
2 
3 /*-
4  * Copyright (c) 2010 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Mihai Chelaru <kefren@NetBSD.org>
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 #include <sys/types.h>
33 #include <sys/stat.h>
34 #include <sys/socket.h>
35 #include <sys/ioctl.h>
36 #include <net/if.h>
37 #include <netinet/in.h>
38 #include <arpa/inet.h>
39 
40 #include <assert.h>
41 #include <errno.h>
42 #include <ifaddrs.h>
43 #include <poll.h>
44 #include <signal.h>
45 #include <stdio.h>
46 #include <stdlib.h>
47 #include <strings.h>
48 #include <unistd.h>
49 
50 #include "fsm.h"
51 #include "ldp.h"
52 #include "ldp_command.h"
53 #include "tlv.h"
54 #include "ldp_peer.h"
55 #include "notifications.h"
56 #include "tlv_stack.h"
57 #include "mpls_interface.h"
58 #include "label.h"
59 #include "mpls_routes.h"
60 #include "ldp_errors.h"
61 #include "socketops.h"
62 
63 int             ls;			/* TCP listening socket on port 646 */
64 int             route_socket;		/* used to see when a route is added/deleted */
65 int		hello_socket;		/* hello multicast listener - transmitter */
66 int		command_socket;		/* Listening socket for interface command */
67 int             current_msg_id = 0x233;
68 int		command_port = LDP_COMMAND_PORT;
69 extern int      replay_index;
70 extern struct rt_msg replay_rt[REPLAY_MAX];
71 extern struct com_sock	csockets[MAX_COMMAND_SOCKETS];
72 
73 int	ldp_hello_time = LDP_HELLO_TIME;
74 int	ldp_keepalive_time = LDP_KEEPALIVE_TIME;
75 int	ldp_holddown_time = LDP_HOLDTIME;
76 int	no_default_route = 1;
77 
78 void	recv_pdu(int);
79 void	send_hello_alarm(int);
80 __dead static void bail_out(int);
81 static int get_local_addr(struct sockaddr_dl *, struct in_addr *);
82 
83 int
84 create_hello_socket()
85 {
86 	struct ip_mreq  mcast_addr;
87 	int             s = socket(PF_INET, SOCK_DGRAM, 17);
88 
89 	if (s < 0)
90 		return s;
91 
92 	/*
93 	 * RFC3036 specifies we should listen to all subnet routers multicast
94 	 * group
95 	 */
96 	mcast_addr.imr_multiaddr.s_addr = inet_addr(ALL_ROUTERS);
97 	mcast_addr.imr_interface.s_addr = htonl(INADDR_ANY);
98 
99 	socket_reuse_port(s);
100 	/* Bind it to port 646 on specific address */
101 	if (bind_socket(s, htonl(INADDR_ANY)) == -1) {
102 		warnp("Cannot bind hello socket\n");
103 		close(s);
104 		return -1;
105 	}
106 	/* We don't need to receive back our messages */
107 	if (setsockopt(s, IPPROTO_IP, IP_MULTICAST_LOOP, &(uint8_t){0},
108 	    sizeof(uint8_t)) == -1) {
109 		fatalp("setsockopt: %s", strerror(errno));
110 		close(s);
111 		return -1;
112 	}
113 	/* Finally join the group */
114         if (setsockopt(s, IPPROTO_IP, IP_ADD_MEMBERSHIP, (char *) &mcast_addr,
115 	    sizeof(mcast_addr)) == -1) {
116                 fatalp("setsockopt: %s", strerror(errno));
117                 close(s);
118                 return -1;
119         }
120 	/* TTL:1, TOS: 0xc0 */
121 	if (set_mcast_ttl(s) == -1) {
122 		close(s);
123 		return -1;
124 	}
125 	if (set_tos(s) == -1) {
126 		fatalp("set_tos: %s", strerror(errno));
127 		close(s);
128 		return -1;
129 	}
130 	if (setsockopt(s, IPPROTO_IP, IP_RECVIF, &(uint32_t){1}, sizeof(uint32_t)) == -1) {
131 		fatalp("Cannot set IP_RECVIF\n");
132 		close(s);
133 		return -1;
134 	}
135 	hello_socket = s;
136 	return hello_socket;
137 }
138 
139 /* Sets the TTL to 1 as we don't want to transmit outside this subnet */
140 int
141 set_ttl(int s)
142 {
143 	int             ret;
144 	if ((ret = setsockopt(s, IPPROTO_IP, IP_TTL, &(int){1}, sizeof(int)))
145 	    == -1)
146 		fatalp("set_ttl: %s", strerror(errno));
147 	return ret;
148 }
149 
150 /* Sets multicast TTL to 1 */
151 int
152 set_mcast_ttl(int s)
153 {
154 	int	ret;
155 	if ((ret = setsockopt(s, IPPROTO_IP, IP_MULTICAST_TTL, &(int){1},
156 	    sizeof(int))) == -1)
157 		fatalp("set_mcast_ttl: %s", strerror(errno));
158 	return ret;
159 }
160 
161 /* Sets TOS to 0xc0 aka IP Precedence 6 */
162 int
163 set_tos(int s)
164 {
165 	int             ret;
166 	if ((ret = setsockopt(s, IPPROTO_IP, IP_TOS, &(int){0xc0},
167 	    sizeof(int))) == -1)
168 		fatalp("set_tos: %s", strerror(errno));
169 	return ret;
170 }
171 
172 int
173 socket_reuse_port(int s)
174 {
175 	int             ret;
176 	if ((ret = setsockopt(s, SOL_SOCKET, SO_REUSEPORT, &(int){1},
177 	    sizeof(int))) == -1)
178 		fatalp("socket_reuse_port: %s", strerror(errno));
179 	return ret;
180 }
181 
182 /* binds an UDP socket */
183 int
184 bind_socket(int s, uint32_t addr)
185 {
186 	struct sockaddr_in sa;
187 
188 	sa.sin_len = sizeof(sa);
189 	sa.sin_family = AF_INET;
190 	sa.sin_port = htons(LDP_PORT);
191 	sa.sin_addr.s_addr = addr;
192 	if (bind(s, (struct sockaddr *) (&sa), sizeof(sa))) {
193 		fatalp("bind_socket: %s", strerror(errno));
194 		return -1;
195 	}
196 	return 0;
197 }
198 
199 /* Create / bind the TCP socket */
200 int
201 create_listening_socket(void)
202 {
203 	struct sockaddr_in sa;
204 	int             s;
205 
206 	sa.sin_len = sizeof(sa);
207 	sa.sin_family = AF_INET;
208 	sa.sin_port = htons(LDP_PORT);
209 	sa.sin_addr.s_addr = htonl(INADDR_ANY);
210 
211 	s = socket(PF_INET, SOCK_STREAM, 6);
212 	if (s < 0)
213 		return s;
214 	if (bind(s, (struct sockaddr *) & sa, sizeof(sa))) {
215 		fatalp("bind: %s", strerror(errno));
216 		close(s);
217 		return -1;
218 	}
219 	if (listen(s, 10) == -1) {
220 		fatalp("listen: %s", strerror(errno));
221 		close(s);
222 		return -1;
223 	}
224 /*	if (set_tos(s) == -1) {
225 		fatalp("set_tos: %s", strerror(errno));
226 		close(s);
227 		return -1;
228 	}
229 */	return s;
230 }
231 
232 /*
233  * It's ugly. We need a function to pass all tlvs and create pdu but since I
234  * use UDP socket only to send hellos, I didn't bother
235  */
236 void
237 send_hello(void)
238 {
239 	struct hello_tlv *t;
240 	struct common_hello_tlv *cht;
241 	struct ldp_pdu  *spdu;
242 	struct transport_address_tlv *trtlv;
243 	void *v;
244 	struct sockaddr_in sadest;	/* Destination ALL_ROUTERS */
245 	int sb = 0;			/* sent bytes */
246 	struct ifaddrs *ifa, *ifb;
247 	struct sockaddr_in *if_sa;
248 	char lastifname[20];
249 
250 #define HELLO_MSG_SIZE (sizeof(struct ldp_pdu) + 	/* PDU */	\
251 			TLV_TYPE_LENGTH + MSGID_SIZE +	/* Hello TLV */	\
252 			/* Common Hello TLV */				\
253 			sizeof(struct common_hello_tlv) +		\
254 			/* IPv4 Transport Address */			\
255 			sizeof(struct transport_address_tlv))
256 
257 	if ((v = calloc(1, HELLO_MSG_SIZE)) == NULL) {
258 		fatalp("malloc problem in send_hello()\n");
259 		return;
260 	}
261 
262 	spdu = (struct ldp_pdu *)((char *)v);
263 	t = (struct hello_tlv *)(spdu + 1);
264 	cht = &t->ch;	/* Hello tlv struct includes CHT */
265 	trtlv = (struct transport_address_tlv *)(t + 1);
266 
267 	/* Prepare PDU envelope */
268 	spdu->version = htons(LDP_VERSION);
269 	spdu->length = htons(HELLO_MSG_SIZE - PDU_VER_LENGTH);
270 	inet_aton(LDP_ID, &spdu->ldp_id);
271 
272 	/* Prepare Hello TLV */
273 	t->type = htons(LDP_HELLO);
274 	t->length = htons(MSGID_SIZE +
275 			sizeof(struct common_hello_tlv) +
276 			sizeof(struct transport_address_tlv));
277 	/*
278 	 * I used ID 0 instead of htonl(get_message_id()) because I've
279 	 * seen hellos from a cisco router doing the same thing
280 	 */
281 	t->messageid = 0;
282 
283 	/* Prepare Common Hello attributes */
284 	cht->type = htons(TLV_COMMON_HELLO);
285 	cht->length = htons(sizeof(cht->holdtime) + sizeof(cht->res));
286 	cht->holdtime = htons(ldp_holddown_time);
287 	cht->res = 0;
288 
289 	/*
290 	 * Prepare Transport Address TLV RFC3036 says: "If this optional TLV
291 	 * is not present the IPv4 source address for the UDP packet carrying
292 	 * the Hello should be used." But we send it because everybody seems
293 	 * to do so
294 	 */
295 	trtlv->type = htons(TLV_IPV4_TRANSPORT);
296 	trtlv->length = htons(sizeof(struct in_addr));
297 	/* trtlv->address will be set for each socket */
298 
299 	/* Destination sockaddr */
300 	memset(&sadest, 0, sizeof(sadest));
301 	sadest.sin_len = sizeof(sadest);
302 	sadest.sin_family = AF_INET;
303 	sadest.sin_port = htons(LDP_PORT);
304 	inet_aton(ALL_ROUTERS, &sadest.sin_addr);
305 
306 	if (getifaddrs(&ifa) == -1) {
307 		free(v);
308 		return;
309 	}
310 
311 	lastifname[0] = '\0';
312 	for (ifb = ifa; ifb; ifb = ifb->ifa_next) {
313 		if_sa = (struct sockaddr_in *) ifb->ifa_addr;
314 		if (if_sa->sin_family != AF_INET)
315 			continue;
316 		if (ntohl(if_sa->sin_addr.s_addr) >> 24 == IN_LOOPBACKNET ||
317 		    ntohl(if_sa->sin_addr.s_addr) >> 24 == 0)
318 			continue;
319 		/* Send only once per interface, using master address */
320 		if (strcmp(ifb->ifa_name, lastifname) == 0)
321 			continue;
322 		debugp("Sending hello on %s\n", ifb->ifa_name);
323 		if (setsockopt(hello_socket, IPPROTO_IP, IP_MULTICAST_IF,
324 		    &if_sa->sin_addr, sizeof(struct in_addr)) == -1) {
325 			warnp("setsockopt failed: %s\n", strerror(errno));
326 			continue;
327 		}
328 		trtlv->address.s_addr = if_sa->sin_addr.s_addr;
329 
330 		strlcpy(lastifname, ifb->ifa_name, sizeof(lastifname));
331 
332 		/* Send to the wire */
333 		sb = sendto(hello_socket, v, HELLO_MSG_SIZE,
334 			    0, (struct sockaddr *) & sadest, sizeof(sadest));
335 		if (sb < (int)HELLO_MSG_SIZE)
336 		    fatalp("send: %s", strerror(errno));
337 		else
338 		    debugp("Send %d bytes (PDU: %d, Hello TLV: %d, CH: %d)\n",
339 			sb, (int) (sizeof(struct ldp_pdu) - PDU_VER_LENGTH),
340 		       (int) (TLV_TYPE_LENGTH + MSGID_SIZE),
341 		       (int) (sizeof(struct common_hello_tlv)));
342 
343 	}
344 	freeifaddrs(ifa);
345 	free(v);
346 }
347 
348 int
349 get_message_id(void)
350 {
351 	current_msg_id++;
352 	return current_msg_id;
353 }
354 
355 static int
356 get_local_addr(struct sockaddr_dl *sdl, struct in_addr *sin)
357 {
358 	struct ifaddrs *ifa, *ifb;
359 	struct sockaddr_in *sinet;
360 
361 	if (sdl == NULL)
362 		return -1;
363 
364 	if (getifaddrs(&ifa) == -1)
365 		return -1;
366 	for (ifb = ifa; ifb; ifb = ifb->ifa_next)
367 		if (ifb->ifa_addr->sa_family == AF_INET) {
368 			if (if_nametoindex(ifb->ifa_name) != sdl->sdl_index)
369 				continue;
370 			sinet = (struct sockaddr_in*) ifb->ifa_addr;
371 			sin->s_addr = sinet->sin_addr.s_addr;
372 			freeifaddrs(ifa);
373 			return 0;
374 		}
375 	freeifaddrs(ifa);
376 	return -1;
377 }
378 
379 /* Receive PDUs on Multicast UDP socket */
380 void
381 recv_pdu(int sock)
382 {
383 	struct ldp_pdu  rpdu;
384 	int             c, i;
385 	struct msghdr msg;
386 	struct iovec iov[1];
387 	unsigned char recvspace[MAX_PDU_SIZE];
388 	struct hello_tlv *t;
389 	struct sockaddr_in fromsa;
390 	struct sockaddr_dl *sdl = NULL;
391 	struct in_addr my_ldp_addr, local_addr;
392 	struct cmsghdr *cmptr;
393 	union {
394 		struct cmsghdr cm;
395 		char control[1024];
396 	} control_un;
397 
398 	debugp("Entering RECV_PDU\n");
399 
400 	memset(&msg, 0, sizeof(msg));
401 	msg.msg_control = control_un.control;
402 	msg.msg_controllen = sizeof(control_un.control);
403 	msg.msg_flags = 0;
404 	msg.msg_name = &fromsa;
405 	msg.msg_namelen = sizeof(fromsa);
406 	iov[0].iov_base = recvspace;
407 	iov[0].iov_len = sizeof(recvspace);
408 	msg.msg_iov = iov;
409 	msg.msg_iovlen = 1;
410 
411 	c = recvmsg(sock, &msg, MSG_WAITALL);
412 	debugp("Incoming PDU size: %d\n", c);
413 
414 	debugp("PDU from: %s\n", inet_ntoa(fromsa.sin_addr));
415 
416 	/* Check to see if this is larger than MIN_PDU_SIZE */
417 	if (c < MIN_PDU_SIZE)
418 		return;
419 
420 	/* Read the PDU */
421 	i = get_pdu(recvspace, &rpdu);
422 
423 	/* We currently understand Version 1 */
424 	if (rpdu.version != LDP_VERSION) {
425 		fatalp("recv_pdu: Version mismatch\n");
426 		return;
427 	}
428 
429 	/* Maybe it's our hello */
430 	inet_aton(LDP_ID, &my_ldp_addr);
431 	if (rpdu.ldp_id.s_addr == my_ldp_addr.s_addr) {
432 		fatalp("Received our PDU..\n");	/* it should be not looped */
433 		return;
434 	}
435 
436 	if (msg.msg_controllen < (socklen_t)sizeof(struct cmsghdr) ||
437 	    (msg.msg_flags & MSG_CTRUNC))
438 		local_addr.s_addr = my_ldp_addr.s_addr;
439 	else {
440 		for (cmptr = CMSG_FIRSTHDR(&msg); cmptr != NULL;
441 		    cmptr = CMSG_NXTHDR(&msg, cmptr))
442 			if (cmptr->cmsg_level == IPPROTO_IP &&
443 			    cmptr->cmsg_type == IP_RECVIF) {
444 				sdl = (struct sockaddr_dl *) CMSG_DATA(cmptr);
445 				break;
446 			}
447 		if (get_local_addr(sdl, &local_addr) != 0)
448 			local_addr.s_addr = my_ldp_addr.s_addr;
449 	}
450 
451 
452 	debugp("Read %d bytes from address %s Length: %.4d Version: %d\n",
453 	       c, inet_ntoa(rpdu.ldp_id), rpdu.length, rpdu.version);
454 
455 	/* Fill the TLV messages */
456 	t = get_hello_tlv(recvspace + i, c - i);
457 	run_ldp_hello(&rpdu, t, &fromsa.sin_addr, &local_addr, sock);
458 }
459 
460 void
461 send_hello_alarm(int unused)
462 {
463 	struct ldp_peer *p, *ptmp;
464 	struct hello_info *hi, *hinext;
465 	time_t          t = time(NULL);
466 	int             olderrno = errno;
467 
468 	/* Send hellos */
469 	if (!(t % ldp_hello_time))
470 		send_hello();
471 
472 	/* Timeout -- */
473 	SLIST_FOREACH(p, &ldp_peer_head, peers)
474 		p->timeout--;
475 
476 	/* Check for timeout */
477 	SLIST_FOREACH_SAFE(p, &ldp_peer_head, peers, ptmp)
478 		if (p->timeout < 1)
479 			switch (p->state) {
480 			case LDP_PEER_HOLDDOWN:
481 				debugp("LDP holddown expired for peer %s\n",
482 				       inet_ntoa(p->ldp_id));
483 				ldp_peer_delete(p);
484 				break;
485 			case LDP_PEER_ESTABLISHED:
486 			case LDP_PEER_CONNECTED:
487 				send_notification(p, 0,
488 				    NOTIF_KEEP_ALIVE_TIMER_EXPIRED);
489 				warnp("Keepalive expired for %s\n",
490 				    inet_ntoa(p->ldp_id));
491 				ldp_peer_holddown(p);
492 				break;
493 			}	/* switch */
494 
495 	/* send keepalives */
496 	if (!(t % ldp_keepalive_time)) {
497 		SLIST_FOREACH(p, &ldp_peer_head, peers)
498 		    if (p->state == LDP_PEER_ESTABLISHED) {
499 			debugp("Sending KeepAlive to %s\n",
500 			    inet_ntoa(p->ldp_id));
501 			keep_alive(p);
502 		    }
503 	}
504 
505 	/* Decrement hello info keepalives */
506 	SLIST_FOREACH(hi, &hello_info_head, infos)
507 		if (hi->keepalive != 0xFFFF)
508 			hi->keepalive--;
509 
510 	/* Check hello keepalives */
511 	SLIST_FOREACH_SAFE(hi, &hello_info_head, infos, hinext)
512 		if (hi->keepalive < 1)
513 			SLIST_REMOVE(&hello_info_head, hi, hello_info, infos);
514 
515 	/* Set the alarm again and bail out */
516 	alarm(1);
517 	errno = olderrno;
518 }
519 
520 static void
521 bail_out(int x)
522 {
523 	ldp_peer_holddown_all();
524 	flush_mpls_routes();
525 	exit(0);
526 }
527 
528 /*
529  * The big poll that catches every single event
530  * on every socket.
531  */
532 int
533 the_big_loop(void)
534 {
535 	int		sock_error;
536 	uint32_t	i;
537 	socklen_t       sock_error_size = sizeof(int);
538 	struct ldp_peer *p;
539 	struct com_sock	*cs;
540 	struct pollfd	pfd[MAX_POLL_FDS];
541 
542 	assert(MAX_POLL_FDS > 3);
543 
544 	SLIST_INIT(&hello_info_head);
545 
546 	signal(SIGALRM, send_hello_alarm);
547 	signal(SIGPIPE, SIG_IGN);
548 	signal(SIGINT, bail_out);
549 	signal(SIGTERM, bail_out);
550 	send_hello_alarm(1);
551 
552 	route_socket = socket(PF_ROUTE, SOCK_RAW, AF_UNSPEC);
553 
554 	sock_error = bind_current_routes();
555 	if (sock_error != LDP_E_OK) {
556 		fatalp("Cannot get current routes\n");
557 		return sock_error;
558 	}
559 
560 	for (;;) {
561 		nfds_t pollsum = 4;
562 
563 		pfd[0].fd = ls;
564 		pfd[0].events = POLLRDNORM;
565 		pfd[0].revents = 0;
566 
567 		pfd[1].fd = route_socket;
568 		pfd[1].events = POLLRDNORM;
569 		pfd[1].revents = 0;
570 
571 		pfd[2].fd = command_socket;
572 		pfd[2].events = POLLRDNORM;
573 		pfd[2].revents = 0;
574 
575 		/* Hello socket */
576 		pfd[3].fd = hello_socket;
577 		pfd[3].events = POLLIN;
578 		pfd[3].revents = 0;
579 
580 		/* Command sockets */
581 		for (i=0; i < MAX_COMMAND_SOCKETS; i++)
582 			if (csockets[i].socket != -1) {
583 				if (pollsum >= MAX_POLL_FDS)
584 					break;
585 				pfd[pollsum].fd = csockets[i].socket;
586 				pfd[pollsum].events = POLLIN;
587 				pfd[pollsum].revents = 0;
588 				pollsum++;
589 			}
590 
591 		/* LDP Peer sockets */
592 		SLIST_FOREACH(p, &ldp_peer_head, peers) {
593 			if (p->socket < 1)
594 				continue;
595 			switch (p->state) {
596 			    case LDP_PEER_CONNECTED:
597 			    case LDP_PEER_ESTABLISHED:
598 				if (pollsum >= MAX_POLL_FDS)
599 					break;
600 				pfd[pollsum].fd = p->socket;
601 				pfd[pollsum].events = POLLRDNORM;
602 				pfd[pollsum].revents = 0;
603 				pollsum++;
604 				break;
605 			    case LDP_PEER_CONNECTING:
606 				if (pollsum >= MAX_POLL_FDS)
607 					break;
608 				pfd[pollsum].fd = p->socket;
609 				pfd[pollsum].events = POLLWRNORM;
610 				pfd[pollsum].revents = 0;
611 				pollsum++;
612 				break;
613 			}
614 		}
615 
616 		if (pollsum >= MAX_POLL_FDS) {
617 			fatalp("Too many sockets. Increase MAX_POLL_FDS\n");
618 			return LDP_E_TOO_MANY_FDS;
619 		}
620 		if (poll(pfd, pollsum, INFTIM) < 0) {
621 			if (errno != EINTR)
622 				fatalp("poll: %s", strerror(errno));
623 			continue;
624 		}
625 
626 		for (i = 0; i < pollsum; i++) {
627 			if ((pfd[i].revents & POLLRDNORM) ||
628 			    (pfd[i].revents & POLLIN)) {
629 				if(pfd[i].fd == ls)
630 					new_peer_connection();
631 				else if (pfd[i].fd == route_socket) {
632 					struct rt_msg xbuf;
633 					int l;
634 					do {
635 						l = read(route_socket, &xbuf,
636 						    sizeof(xbuf));
637 					} while ((l == -1) && (errno == EINTR));
638 
639 					if (l == -1)
640 						break;
641 
642 					check_route(&xbuf, l);
643 
644 				} else if (pfd[i].fd == hello_socket) {
645 					/* Receiving hello socket */
646 					recv_pdu(pfd[i].fd);
647 				} else if (pfd[i].fd == command_socket) {
648 					command_accept(command_socket);
649 				} else if ((cs = is_command_socket(pfd[i].fd))
650 						!= NULL) {
651 					command_dispatch(cs);
652 				} else {
653 					/* ldp peer socket */
654 					p = get_ldp_peer_by_socket(pfd[i].fd);
655 					if (p)
656 						recv_session_pdu(p);
657 				}
658 			} else if(pfd[i].revents & POLLWRNORM) {
659 				p = get_ldp_peer_by_socket(pfd[i].fd);
660 				if (!p)
661 					continue;
662 				if (getsockopt(pfd[i].fd, SOL_SOCKET, SO_ERROR,
663 				    &sock_error, &sock_error_size) != 0 ||
664 				    sock_error != 0) {
665 					ldp_peer_holddown(p);
666 					sock_error = 0;
667 				} else {
668 					p->state = LDP_PEER_CONNECTED;
669 					send_initialize(p);
670 				}
671 			}
672 		}
673 
674 		for (int ri = 0; ri < replay_index; ri++) {
675 			debugp("Replaying: PID %d, SEQ %d\n",
676 				replay_rt[ri].m_rtm.rtm_pid,
677 				replay_rt[ri].m_rtm.rtm_seq);
678 			check_route(&replay_rt[ri], sizeof(struct rt_msg));
679                 }
680 		replay_index = 0;
681 	}	/* for (;;) */
682 }
683 
684 void
685 new_peer_connection()
686 {
687 	struct sockaddr_in sa, sin_me;
688 	int             s;
689 
690 	s = accept(ls, (struct sockaddr *) & sa,
691 		& (socklen_t) { sizeof(struct sockaddr_in) } );
692 	if (s < 0) {
693 		fatalp("accept: %s", strerror(errno));
694 		return;
695 	}
696 
697 	if (get_ldp_peer(&sa.sin_addr) != NULL) {
698 		close(s);
699 		return;
700 	}
701 
702 	warnp("Accepted a connection from %s\n", inet_ntoa(sa.sin_addr));
703 
704 	if (getsockname(s, (struct sockaddr *)&sin_me,
705 	    & (socklen_t) { sizeof(struct sockaddr_in) } )) {
706 		fatalp("new_peer_connection(): cannot getsockname\n");
707 		close(s);
708 		return;
709 	}
710 
711 	if (ntohl(sa.sin_addr.s_addr) < ntohl(sin_me.sin_addr.s_addr)) {
712 		fatalp("Peer %s: connect from lower ID\n",
713 		    inet_ntoa(sa.sin_addr));
714 		close(s);
715 		return;
716 	}
717 	/* XXX: sa.sin_addr ain't peer LDP ID ... */
718 	ldp_peer_new(&sa.sin_addr, &sa.sin_addr, NULL, ldp_holddown_time, s);
719 
720 }
721 
722 void
723 send_initialize(struct ldp_peer * p)
724 {
725 	struct init_tlv ti;
726 
727 	ti.type = htons(LDP_INITIALIZE);
728 	ti.length = htons(sizeof(struct init_tlv) - TLV_TYPE_LENGTH);
729 	ti.messageid = htonl(get_message_id());
730 	ti.cs_type = htons(TLV_COMMON_SESSION);
731 	ti.cs_len = htons(CS_LEN);
732 	ti.cs_version = htons(LDP_VERSION);
733 	ti.cs_keepalive = htons(2 * ldp_keepalive_time);
734 	ti.cs_adpvlim = 0;
735 	ti.cs_maxpdulen = htons(MAX_PDU_SIZE);
736 	ti.cs_peeraddress.s_addr = p->ldp_id.s_addr;
737 	ti.cs_peeraddrspace = 0;
738 
739 	send_tlv(p, (struct tlv *) (void *) &ti);
740 }
741 
742 void
743 keep_alive(struct ldp_peer * p)
744 {
745 	struct ka_tlv   kt;
746 
747 	kt.type = htons(LDP_KEEPALIVE);
748 	kt.length = htons(sizeof(kt.messageid));
749 	kt.messageid = htonl(get_message_id());
750 
751 	send_tlv(p, (struct tlv *) (void *) &kt);
752 
753 }
754 
755 void
756 recv_session_pdu(struct ldp_peer * p)
757 {
758 	struct ldp_pdu *rpdu;
759 	struct address_tlv *atlv;
760 	struct al_tlv  *altlv;
761 	struct init_tlv	*itlv;
762 	struct label_map_tlv *lmtlv;
763 	struct fec_tlv *fectlv;
764 	struct label_tlv *labeltlv;
765 	struct notification_tlv *nottlv;
766 	struct hello_info *hi;
767 
768 	int             c;
769 	int32_t         wo = 0;
770 	struct tlv     *ttmp;
771 	unsigned char   recvspace[MAX_PDU_SIZE];
772 
773 	memset(recvspace, 0, MAX_PDU_SIZE);
774 
775 	c = recv(p->socket, (void *) recvspace, MAX_PDU_SIZE, MSG_PEEK);
776 
777 	debugp("Ready to read %d bytes\n", c);
778 
779 	if (c < 1) {		/* Session closed */
780 		warnp("Error in connection with %s\n", inet_ntoa(p->ldp_id));
781 		ldp_peer_holddown(p);
782 		return;
783 	}
784 	if (c > MAX_PDU_SIZE) {
785 		debugp("Incoming PDU size exceeds MAX_PDU_SIZE !\n");
786 		return;
787 	}
788 	if (c < MIN_PDU_SIZE) {
789 		debugp("PDU too small received from peer %s\n", inet_ntoa(p->ldp_id));
790 		return;
791 	}
792 	rpdu = (struct ldp_pdu *) recvspace;
793 	/* XXX: buggy messages may crash the whole thing */
794 	c = recv(p->socket, (void *) recvspace,
795 		ntohs(rpdu->length) + PDU_VER_LENGTH, MSG_WAITALL);
796 	rpdu = (struct ldp_pdu *) recvspace;
797 
798 	/* Check if it's somehow OK... */
799 	if (check_recv_pdu(p, rpdu, c) != 0)
800 		return;
801 
802 	debugp("Read %d bytes, PDU size: %d bytes\n", c, ntohs(rpdu->length));
803 	wo = sizeof(struct ldp_pdu);
804 
805 	while (wo + TLV_TYPE_LENGTH < (uint)c) {
806 
807 		ttmp = (struct tlv *) (&recvspace[wo]);
808 
809 		if ((ntohs(ttmp->type) != LDP_KEEPALIVE) &&
810 		    (ntohs(ttmp->type) != LDP_LABEL_MAPPING)) {
811 			debugp("Got Type: 0x%.4X (Length: %d) from %s\n",
812 			    ntohs(ttmp->type), ntohs(ttmp->length),
813 			    inet_ntoa(p->ldp_id));
814 		} else
815 			debugp("Got Type: 0x%.4X (Length: %d) from %s\n",
816 			    ntohs(ttmp->type), ntohs(ttmp->length),
817 			    inet_ntoa(p->ldp_id));
818 
819 		/* Should we get the message ? */
820 		if (p->state != LDP_PEER_ESTABLISHED &&
821 		    ntohs(ttmp->type) != LDP_INITIALIZE &&
822 		    ntohs(ttmp->type) != LDP_KEEPALIVE)
823 			break;
824 		/* The big switch */
825 		switch (ntohs(ttmp->type)) {
826 		case LDP_INITIALIZE:
827 			itlv = (struct init_tlv *)ttmp;
828 			/* Check size */
829 			if (ntohs(itlv->length) <
830 			    sizeof(struct init_tlv) - TLV_TYPE_LENGTH) {
831 				send_notification(p, 0,
832 				    NOTIF_BAD_PDU_LEN | NOTIF_FATAL);
833 				ldp_peer_holddown(p);
834 				break;
835 			}
836 			/* Check version */
837 			if (ntohs(itlv->cs_version) != LDP_VERSION) {
838 				send_notification(p, ntohl(itlv->messageid),
839 					NOTIF_BAD_LDP_VER | NOTIF_FATAL);
840 				ldp_peer_holddown(p);
841 				break;
842 			}
843 			/* Check if we got any hello from this one */
844 			SLIST_FOREACH(hi, &hello_info_head, infos)
845 				if (hi->ldp_id.s_addr == rpdu->ldp_id.s_addr)
846 					break;
847 			if (hi == NULL) {
848 			    send_notification(p, ntohl(itlv->messageid),
849 				NOTIF_SESSION_REJECTED_NO_HELLO | NOTIF_FATAL);
850 			    ldp_peer_holddown(p);
851 			    break;
852 			}
853 
854 			if (!p->master) {
855 				keep_alive(p);
856 				send_initialize(p);
857 			} else {
858 				p->state = LDP_PEER_ESTABLISHED;
859 				p->established_t = time(NULL);
860 				keep_alive(p);
861 
862 				/*
863 				 * Recheck here ldp id because we accepted
864 				 * connection without knowing who is it for sure
865 				 */
866 				p->ldp_id.s_addr = rpdu->ldp_id.s_addr;
867 
868 				fatalp("LDP neighbour %s is UP\n",
869 				    inet_ntoa(p->ldp_id));
870 				mpls_add_ldp_peer(p);
871 				send_addresses(p);
872 				send_all_bindings(p);
873 			}
874 			break;
875 		case LDP_KEEPALIVE:
876 			if ((p->state == LDP_PEER_CONNECTED) && (!p->master)) {
877 				p->state = LDP_PEER_ESTABLISHED;
878 				p->established_t = time(NULL);
879 				fatalp("LDP neighbour %s is UP\n",
880 				    inet_ntoa(p->ldp_id));
881 				mpls_add_ldp_peer(p);
882 				send_addresses(p);
883 				send_all_bindings(p);
884 			}
885 			p->timeout = p->holdtime;
886 			break;
887 		case LDP_ADDRESS:
888 			/* Add peer addresses */
889 			atlv = (struct address_tlv *) ttmp;
890 			altlv = (struct al_tlv *) (&atlv[1]);
891 			add_ifaddresses(p, altlv);
892 			print_bounded_addresses(p);
893 			break;
894 		case LDP_ADDRESS_WITHDRAW:
895 			atlv = (struct address_tlv *) ttmp;
896 			altlv = (struct al_tlv *) (&atlv[1]);
897 			del_ifaddresses(p, altlv);
898 			break;
899 		case LDP_LABEL_MAPPING:
900 			lmtlv = (struct label_map_tlv *) ttmp;
901 			fectlv = (struct fec_tlv *) (&lmtlv[1]);
902 			labeltlv = (struct label_tlv *)((unsigned char *)fectlv
903 				+ ntohs(fectlv->length) + TLV_TYPE_LENGTH);
904 			map_label(p, fectlv, labeltlv);
905 			break;
906 		case LDP_LABEL_REQUEST:
907 			lmtlv = (struct label_map_tlv *) ttmp;
908 			fectlv = (struct fec_tlv *) (&lmtlv[1]);
909 			switch (request_respond(p, lmtlv, fectlv)) {
910 			case LDP_E_BAD_FEC:
911 				send_notification(p, ntohl(lmtlv->messageid),
912 					NOTIF_UNKNOWN_TLV);
913 				break;
914 			case LDP_E_BAD_AF:
915 				send_notification(p, ntohl(lmtlv->messageid),
916 					NOTIF_UNSUPPORTED_AF);
917 				break;
918 			case LDP_E_NO_SUCH_ROUTE:
919 				send_notification(p, ntohl(lmtlv->messageid),
920 					NOTIF_NO_ROUTE);
921 				break;
922 			}
923 			break;
924 		case LDP_LABEL_WITHDRAW:
925 			lmtlv = (struct label_map_tlv *) ttmp;
926 			fectlv = (struct fec_tlv *) (&lmtlv[1]);
927 			if (withdraw_label(p, fectlv) == LDP_E_OK) {
928 				/* Send RELEASE */
929 				prepare_release(ttmp);
930 				send_tlv(p, ttmp);
931 				}
932 			break;
933 		case LDP_LABEL_RELEASE:
934 			/*
935 			 * XXX: we need to make a timed queue...
936 			 * For now I just assume peers are processing messages
937 			 * correctly so I just ignore confirmations
938 			 */
939 			wo = -1;	/* Ignore rest of message */
940 			break;
941 		case LDP_LABEL_ABORT:
942 		/* XXX: For now I pretend I can process everything
943 		 * RFC 3036, Section 3.5.9.1
944 		 * If an LSR receives a Label Abort Request Message after it
945 		 * has responded to the Label Request in question with a Label
946 		 * Mapping message or a Notification message, it ignores the
947 		 * abort request.
948 		 */
949 			wo = -1;
950 			break;
951 		case LDP_NOTIFICATION:
952 			nottlv = (struct notification_tlv *) ttmp;
953 			nottlv->st_code = ntohl(nottlv->st_code);
954 			fatalp("Got notification 0x%X from peer %s\n",
955 			    nottlv->st_code, inet_ntoa(p->ldp_id));
956 			if (nottlv->st_code >> 31) {
957 				fatalp("LDP peer %s signalized %s\n",
958 				    inet_ntoa(p->ldp_id),
959 				    NOTIF_STR[(nottlv->st_code << 1) >> 1]);
960 				ldp_peer_holddown(p);
961 				wo = -1;
962 			}
963 			break;
964 		case LDP_HELLO:
965 			/* No hellos should came on tcp session */
966 			wo = -1;
967 			break;
968 		default:
969 			warnp("Unknown TLV received from %s\n",
970 			    inet_ntoa(p->ldp_id));
971 			debug_tlv(ttmp);
972 			wo = -1;/* discard the rest of the message */
973 			break;
974 		}
975 		if (wo < 0) {
976 			debugp("Discarding the rest of the message\n");
977 			break;
978 		} else {
979 			wo += ntohs(ttmp->length) + TLV_TYPE_LENGTH;
980 			debugp("WORKED ON %u bytes (Left %d)\n", wo, c - wo);
981 		}
982 	}			/* while */
983 
984 }
985 
986 /* Sends a pdu, tlv pair to a connected peer */
987 int
988 send_message(struct ldp_peer * p, struct ldp_pdu * pdu, struct tlv * t)
989 {
990 	unsigned char   sendspace[MAX_PDU_SIZE];
991 
992 	/* Check if peer is connected */
993 	switch (p->state) {
994 	case LDP_PEER_CONNECTED:
995 	case LDP_PEER_ESTABLISHED:
996 		break;
997 	default:
998 		return -1;
999 	}
1000 
1001 	/* Check length validity first */
1002 	if (ntohs(pdu->length) !=
1003 	    ntohs(t->length) + TLV_TYPE_LENGTH + PDU_PAYLOAD_LENGTH) {
1004 		fatalp("LDP: TLV - PDU incompability. Message discarded\n");
1005 		fatalp("LDP: TLV len %d - PDU len %d\n", ntohs(t->length),
1006 		    ntohs(pdu->length));
1007 		return -1;
1008 	}
1009 	if (ntohs(t->length) + PDU_VER_LENGTH > MAX_PDU_SIZE) {
1010 		fatalp("Message to large discarded\n");
1011 		return -1;
1012 	}
1013 	/* Arrange them in a buffer and send */
1014 	memcpy(sendspace, pdu, sizeof(struct ldp_pdu));
1015 	memcpy(sendspace + sizeof(struct ldp_pdu), t,
1016 	    ntohs(t->length) + TLV_TYPE_LENGTH);
1017 
1018 	/* Report keepalives only for DEBUG */
1019 	if ((ntohs(t->type) != 0x201) && (ntohs(t->type) != 0x400)) {
1020 		debugp("Sending message type 0x%.4X to %s (size: %d)\n",
1021 		    ntohs(t->type), inet_ntoa(p->ldp_id), ntohs(t->length));
1022 	} else
1023 	/* downgraded from warnp to debugp for now */
1024 		debugp("Sending message type 0x%.4X to %s (size: %d)\n",
1025 		    ntohs(t->type), inet_ntoa(p->ldp_id), ntohs(t->length));
1026 
1027 	/* Send it finally */
1028 	return send(p->socket, sendspace,
1029 		ntohs(pdu->length) + PDU_VER_LENGTH, 0);
1030 }
1031 
1032 /*
1033  * Encapsulates TLV into a PDU and sends it to a peer
1034  */
1035 int
1036 send_tlv(struct ldp_peer * p, struct tlv * t)
1037 {
1038 	struct ldp_pdu  pdu;
1039 
1040 	pdu.version = htons(LDP_VERSION);
1041 	inet_aton(LDP_ID, &pdu.ldp_id);
1042 	pdu.label_space = 0;
1043 	pdu.length = htons(ntohs(t->length) + TLV_TYPE_LENGTH +
1044 		PDU_PAYLOAD_LENGTH);
1045 
1046 	return send_message(p, &pdu, t);
1047 }
1048 
1049 
1050 int
1051 send_addresses(struct ldp_peer * p)
1052 {
1053 	struct address_list_tlv *t;
1054 	int             ret;
1055 
1056 	t = build_address_list_tlv();
1057 
1058 	ret = send_tlv(p, (struct tlv *) t);
1059 	free(t);
1060 	return ret;
1061 
1062 }
1063