xref: /netbsd-src/usr.sbin/ldpd/socketops.c (revision c2f76ff004a2cb67efe5b12d97bd3ef7fe89e18d)
1 /* $NetBSD: socketops.c,v 1.5 2011/01/22 19:46:46 kefren 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 <errno.h>
41 #include <signal.h>
42 #include <stdlib.h>
43 #include <unistd.h>
44 #include <strings.h>
45 #include <stdio.h>
46 #include <ifaddrs.h>
47 #include <poll.h>
48 
49 #include "fsm.h"
50 #include "ldp.h"
51 #include "ldp_command.h"
52 #include "tlv.h"
53 #include "ldp_peer.h"
54 #include "notifications.h"
55 #include "tlv_stack.h"
56 #include "mpls_interface.h"
57 #include "label.h"
58 #include "mpls_routes.h"
59 #include "ldp_errors.h"
60 #include "socketops.h"
61 
62 int             ls;			/* TCP listening socket on port 646 */
63 int             route_socket;		/* used to see when a route is added/deleted */
64 int		hello_socket;		/* hello multicast listener - transmitter */
65 int		command_socket;		/* Listening socket for interface command */
66 int             current_msg_id = 0x233;
67 int		command_port = LDP_COMMAND_PORT;
68 extern int      replay_index;
69 extern struct rt_msg replay_rt[REPLAY_MAX];
70 extern struct com_sock	csockets[MAX_COMMAND_SOCKETS];
71 
72 int	ldp_hello_time = LDP_HELLO_TIME;
73 int	ldp_keepalive_time = LDP_KEEPALIVE_TIME;
74 int	ldp_holddown_time = LDP_HOLDTIME;
75 
76 void	recv_pdu(int);
77 void	send_hello_alarm(int);
78 void	bail_out(int);
79 static int get_local_addr(struct sockaddr_dl *, struct in_addr *);
80 
81 int
82 create_hello_socket()
83 {
84 	struct ip_mreq  mcast_addr;
85 	int             s = socket(PF_INET, SOCK_DGRAM, 17);
86 
87 	if (s < 0)
88 		return s;
89 
90 	/*
91 	 * RFC3036 specifies we should listen to all subnet routers multicast
92 	 * group
93 	 */
94 	mcast_addr.imr_multiaddr.s_addr = inet_addr(ALL_ROUTERS);
95 	mcast_addr.imr_interface.s_addr = htonl(INADDR_ANY);
96 
97 	socket_reuse_port(s);
98 	/* Bind it to port 646 on specific address */
99 	if (bind_socket(s, htonl(INADDR_ANY)) == -1) {
100 		warnp("Cannot bind hello socket\n");
101 		close(s);
102 		return -1;
103 	}
104 	/* We don't need to receive back our messages */
105 	if (setsockopt(s, IPPROTO_IP, IP_MULTICAST_LOOP, &(uint8_t){0},
106 	    sizeof(uint8_t)) == -1) {
107 		fatalp("setsockopt: %s", strerror(errno));
108 		close(s);
109 		return -1;
110 	}
111 	/* Finally join the group */
112         if (setsockopt(s, IPPROTO_IP, IP_ADD_MEMBERSHIP, (char *) &mcast_addr,
113 	    sizeof(mcast_addr)) == -1) {
114                 fatalp("setsockopt: %s", strerror(errno));
115                 close(s);
116                 return -1;
117         }
118 	/* TTL:1, TOS: 0xc0 */
119 	if (set_mcast_ttl(s) == -1) {
120 		close(s);
121 		return -1;
122 	}
123 	if (set_tos(s) == -1) {
124 		fatalp("set_tos: %s", strerror(errno));
125 		close(s);
126 		return -1;
127 	}
128 	if (setsockopt(s, IPPROTO_IP, IP_RECVIF, &(uint32_t){1}, sizeof(uint32_t)) == -1) {
129 		fatalp("Cannot set IP_RECVIF\n");
130 		close(s);
131 		return -1;
132 	}
133 	hello_socket = s;
134 	return hello_socket;
135 }
136 
137 /* Sets the TTL to 1 as we don't want to transmit outside this subnet */
138 int
139 set_ttl(int s)
140 {
141 	int             ret;
142 	if ((ret = setsockopt(s, IPPROTO_IP, IP_TTL, &(int){1}, sizeof(int)))
143 	    == -1)
144 		fatalp("set_ttl: %s", strerror(errno));
145 	return ret;
146 }
147 
148 /* Sets multicast TTL to 1 */
149 int
150 set_mcast_ttl(int s)
151 {
152 	int	ret;
153 	if ((ret = setsockopt(s, IPPROTO_IP, IP_MULTICAST_TTL, &(int){1},
154 	    sizeof(int))) == -1)
155 		fatalp("set_mcast_ttl: %s", strerror(errno));
156 	return ret;
157 }
158 
159 /* Sets TOS to 0xc0 aka IP Precedence 6 */
160 int
161 set_tos(int s)
162 {
163 	int             ret;
164 	if ((ret = setsockopt(s, IPPROTO_IP, IP_TOS, &(int){0xc0},
165 	    sizeof(int))) == -1)
166 		fatalp("set_tos: %s", strerror(errno));
167 	return ret;
168 }
169 
170 int
171 socket_reuse_port(int s)
172 {
173 	int             ret;
174 	if ((ret = setsockopt(s, SOL_SOCKET, SO_REUSEPORT, &(int){1},
175 	    sizeof(int))) == -1)
176 		fatalp("socket_reuse_port: %s", strerror(errno));
177 	return ret;
178 }
179 
180 /* binds an UDP socket */
181 int
182 bind_socket(int s, uint32_t addr)
183 {
184 	struct sockaddr_in sa;
185 
186 	sa.sin_len = sizeof(sa);
187 	sa.sin_family = AF_INET;
188 	sa.sin_port = htons(LDP_PORT);
189 	sa.sin_addr.s_addr = addr;
190 	if (bind(s, (struct sockaddr *) (&sa), sizeof(sa))) {
191 		fatalp("bind_socket: %s", strerror(errno));
192 		return -1;
193 	}
194 	return 0;
195 }
196 
197 /* Create / bind the TCP socket */
198 int
199 create_listening_socket(void)
200 {
201 	struct sockaddr_in sa;
202 	int             s;
203 
204 	sa.sin_len = sizeof(sa);
205 	sa.sin_family = AF_INET;
206 	sa.sin_port = htons(LDP_PORT);
207 	sa.sin_addr.s_addr = htonl(INADDR_ANY);
208 
209 	s = socket(PF_INET, SOCK_STREAM, 6);
210 	if (s < 0)
211 		return s;
212 	if (bind(s, (struct sockaddr *) & sa, sizeof(sa))) {
213 		fatalp("bind: %s", strerror(errno));
214 		close(s);
215 		return -1;
216 	}
217 	if (listen(s, 10) == -1) {
218 		fatalp("listen: %s", strerror(errno));
219 		close(s);
220 		return -1;
221 	}
222 /*	if (set_tos(s) == -1) {
223 		fatalp("set_tos: %s", strerror(errno));
224 		close(s);
225 		return -1;
226 	}
227 */	return s;
228 }
229 
230 /*
231  * It's ugly. We need a function to pass all tlvs and create pdu but since I
232  * use UDP socket only to send hellos, I didn't bother
233  */
234 void
235 send_hello(void)
236 {
237 	struct hello_tlv *t;
238 	struct common_hello_tlv *cht;
239 	struct ldp_pdu  *spdu;
240 	struct transport_address_tlv *trtlv;
241 	void *v;
242 	struct sockaddr_in sadest;	/* Destination ALL_ROUTERS */
243 	int sb = 0;			/* sent bytes */
244 	struct ifaddrs *ifa, *ifb;
245 	struct sockaddr_in *if_sa;
246 	char lastifname[20];
247 
248 #define HELLO_MSG_SIZE (sizeof(struct ldp_pdu) + 	/* PDU */	\
249 			TLV_TYPE_LENGTH + MSGID_SIZE +	/* Hello TLV */	\
250 			/* Common Hello TLV */				\
251 			sizeof(struct common_hello_tlv) +		\
252 			/* IPv4 Transport Address */			\
253 			sizeof(struct transport_address_tlv))
254 
255 	if ((v = calloc(1, HELLO_MSG_SIZE)) == NULL) {
256 		fatalp("malloc problem in send_hello()\n");
257 		return;
258 	}
259 
260 	spdu = (struct ldp_pdu *)((char *)v);
261 	t = (struct hello_tlv *)(spdu + 1);
262 	cht = &t->ch;	/* Hello tlv struct includes CHT */
263 	trtlv = (struct transport_address_tlv *)(t + 1);
264 
265 	/* Prepare PDU envelope */
266 	spdu->version = htons(LDP_VERSION);
267 	spdu->length = htons(HELLO_MSG_SIZE - PDU_VER_LENGTH);
268 	inet_aton(LDP_ID, &spdu->ldp_id);
269 
270 	/* Prepare Hello TLV */
271 	t->type = htons(LDP_HELLO);
272 	t->length = htons(MSGID_SIZE +
273 			sizeof(struct common_hello_tlv) +
274 			sizeof(struct transport_address_tlv));
275 	/*
276 	 * I used ID 0 instead of htonl(get_message_id()) because I've
277 	 * seen hellos from a cisco router doing the same thing
278 	 */
279 	t->messageid = 0;
280 
281 	/* Prepare Common Hello attributes */
282 	cht->type = htons(TLV_COMMON_HELLO);
283 	cht->length = htons(sizeof(cht->holdtime) + sizeof(cht->res));
284 	cht->holdtime = htons(ldp_holddown_time);
285 	cht->res = 0;
286 
287 	/*
288 	 * Prepare Transport Address TLV RFC3036 says: "If this optional TLV
289 	 * is not present the IPv4 source address for the UDP packet carrying
290 	 * the Hello should be used." But we send it because everybody seems
291 	 * to do so
292 	 */
293 	trtlv->type = htons(TLV_IPV4_TRANSPORT);
294 	trtlv->length = htons(sizeof(struct in_addr));
295 	/* trtlv->address will be set for each socket */
296 
297 	/* Destination sockaddr */
298 	memset(&sadest, 0, sizeof(sadest));
299 	sadest.sin_len = sizeof(sadest);
300 	sadest.sin_family = AF_INET;
301 	sadest.sin_port = htons(LDP_PORT);
302 	inet_aton(ALL_ROUTERS, &sadest.sin_addr);
303 
304 	if (getifaddrs(&ifa) == -1) {
305 		free(v);
306 		return;
307 	}
308 
309 	lastifname[0] = '\0';
310 	for (ifb = ifa; ifb; ifb = ifb->ifa_next) {
311 		if_sa = (struct sockaddr_in *) ifb->ifa_addr;
312 		if (if_sa->sin_family != AF_INET)
313 			continue;
314 		if (ntohl(if_sa->sin_addr.s_addr) >> 24 == IN_LOOPBACKNET ||
315 		    ntohl(if_sa->sin_addr.s_addr) >> 24 == 0)
316 			continue;
317 		/* Send only once per interface, using master address */
318 		if (strcmp(ifb->ifa_name, lastifname) == 0)
319 			continue;
320 		debugp("Sending hello on %s\n", ifb->ifa_name);
321 		if (setsockopt(hello_socket, IPPROTO_IP, IP_MULTICAST_IF,
322 		    &if_sa->sin_addr, sizeof(struct in_addr)) == -1) {
323 			warnp("setsockopt failed: %s\n", strerror(errno));
324 			continue;
325 		}
326 		trtlv->address.s_addr = if_sa->sin_addr.s_addr;
327 
328 		strlcpy(lastifname, ifb->ifa_name, sizeof(lastifname));
329 
330 		/* Send to the wire */
331 		sb = sendto(hello_socket, v, HELLO_MSG_SIZE,
332 			    0, (struct sockaddr *) & sadest, sizeof(sadest));
333 		if (sb < (int)HELLO_MSG_SIZE)
334 		    fatalp("send: %s", strerror(errno));
335 		else
336 		    debugp("Send %d bytes (PDU: %d, Hello TLV: %d, CH: %d)\n",
337 			sb, (int) (sizeof(struct ldp_pdu) - PDU_VER_LENGTH),
338 		       (int) (TLV_TYPE_LENGTH + MSGID_SIZE),
339 		       (int) (sizeof(struct common_hello_tlv)));
340 
341 	}
342 	freeifaddrs(ifa);
343 	free(v);
344 }
345 
346 int
347 get_message_id(void)
348 {
349 	current_msg_id++;
350 	return current_msg_id;
351 }
352 
353 static int
354 get_local_addr(struct sockaddr_dl *sdl, struct in_addr *sin)
355 {
356 	struct ifaddrs *ifa, *ifb;
357 	struct sockaddr_in *sinet;
358 
359 	if (sdl == NULL)
360 		return -1;
361 
362 	if (getifaddrs(&ifa) == -1)
363 		return -1;
364 	for (ifb = ifa; ifb; ifb = ifb->ifa_next)
365 		if (ifb->ifa_addr->sa_family == AF_INET) {
366 			if (if_nametoindex(ifb->ifa_name) != sdl->sdl_index)
367 				continue;
368 			sinet = (struct sockaddr_in*) ifb->ifa_addr;
369 			sin->s_addr = sinet->sin_addr.s_addr;
370 			freeifaddrs(ifa);
371 			return 0;
372 		}
373 	freeifaddrs(ifa);
374 	return -1;
375 }
376 
377 /* Receive PDUs on Multicast UDP socket */
378 void
379 recv_pdu(int sock)
380 {
381 	struct ldp_pdu  rpdu;
382 	int             c, i;
383 	struct msghdr msg;
384 	struct iovec iov[1];
385 	unsigned char recvspace[MAX_PDU_SIZE];
386 	struct hello_tlv *t;
387 	struct sockaddr_in fromsa;
388 	struct sockaddr_dl *sdl = NULL;
389 	struct in_addr my_ldp_addr, local_addr;
390 	struct cmsghdr *cmptr;
391 	union {
392 		struct cmsghdr cm;
393 		char control[1024];
394 	} control_un;
395 
396 	debugp("Entering RECV_PDU\n");
397 
398 	memset(&msg, 0, sizeof(msg));
399 	msg.msg_control = control_un.control;
400 	msg.msg_controllen = sizeof(control_un.control);
401 	msg.msg_flags = 0;
402 	msg.msg_name = &fromsa;
403 	msg.msg_namelen = sizeof(fromsa);
404 	iov[0].iov_base = recvspace;
405 	iov[0].iov_len = sizeof(recvspace);
406 	msg.msg_iov = iov;
407 	msg.msg_iovlen = 1;
408 
409 	c = recvmsg(sock, &msg, MSG_WAITALL);
410 	debugp("Incoming PDU size: %d\n", c);
411 
412 	debugp("PDU from: %s\n", inet_ntoa(fromsa.sin_addr));
413 
414 	/* Check to see if this is larger than MIN_PDU_SIZE */
415 	if (c < MIN_PDU_SIZE)
416 		return;
417 
418 	/* Read the PDU */
419 	i = get_pdu(recvspace, &rpdu);
420 
421 	/* We currently understand Version 1 */
422 	if (rpdu.version != LDP_VERSION) {
423 		fatalp("recv_pdu: Version mismatch\n");
424 		return;
425 	}
426 
427 	/* Maybe it's our hello */
428 	inet_aton(LDP_ID, &my_ldp_addr);
429 	if (rpdu.ldp_id.s_addr == my_ldp_addr.s_addr) {
430 		fatalp("Received our PDU..\n");	/* it should be not looped */
431 		return;
432 	}
433 
434 	if (msg.msg_controllen < (socklen_t)sizeof(struct cmsghdr) ||
435 	    (msg.msg_flags & MSG_CTRUNC))
436 		local_addr.s_addr = my_ldp_addr.s_addr;
437 	else {
438 		for (cmptr = CMSG_FIRSTHDR(&msg); cmptr != NULL;
439 		    cmptr = CMSG_NXTHDR(&msg, cmptr))
440 			if (cmptr->cmsg_level == IPPROTO_IP &&
441 			    cmptr->cmsg_type == IP_RECVIF) {
442 				sdl = (struct sockaddr_dl *) CMSG_DATA(cmptr);
443 				break;
444 			}
445 		if (get_local_addr(sdl, &local_addr) != 0)
446 			local_addr.s_addr = my_ldp_addr.s_addr;
447 	}
448 
449 
450 	debugp("Read %d bytes from address %s Length: %.4d Version: %d\n",
451 	       c, inet_ntoa(rpdu.ldp_id), rpdu.length, rpdu.version);
452 
453 	/* Fill the TLV messages */
454 	t = get_hello_tlv(recvspace + i, c - i);
455 	run_ldp_hello(&rpdu, t, &fromsa.sin_addr, &local_addr, sock);
456 }
457 
458 void
459 send_hello_alarm(int unused)
460 {
461 	struct ldp_peer *p, *ptmp;
462 	struct hello_info *hi, *hinext;
463 	time_t          t = time(NULL);
464 	int             olderrno = errno;
465 
466 	/* Send hellos */
467 	if (!(t % ldp_hello_time))
468 		send_hello();
469 
470 	/* Timeout -- */
471 	SLIST_FOREACH(p, &ldp_peer_head, peers)
472 		p->timeout--;
473 
474 	/* Check for timeout */
475 	SLIST_FOREACH_SAFE(p, &ldp_peer_head, peers, ptmp)
476 		if (p->timeout < 1)
477 			switch (p->state) {
478 			case LDP_PEER_HOLDDOWN:
479 				debugp("LDP holddown expired for peer %s\n",
480 				       inet_ntoa(p->ldp_id));
481 				ldp_peer_delete(p);
482 				break;
483 			case LDP_PEER_ESTABLISHED:
484 			case LDP_PEER_CONNECTED:
485 				send_notification(p, 0,
486 				    NOTIF_KEEP_ALIVE_TIMER_EXPIRED);
487 				warnp("Keepalive expired for %s\n",
488 				    inet_ntoa(p->ldp_id));
489 				ldp_peer_holddown(p);
490 				break;
491 			}	/* switch */
492 
493 	/* send keepalives */
494 	if (!(t % ldp_keepalive_time)) {
495 		SLIST_FOREACH(p, &ldp_peer_head, peers)
496 		    if (p->state == LDP_PEER_ESTABLISHED) {
497 			debugp("Sending KeepAlive to %s\n",
498 			    inet_ntoa(p->ldp_id));
499 			keep_alive(p);
500 		    }
501 	}
502 
503 	/* Decrement hello info keepalives */
504 	SLIST_FOREACH(hi, &hello_info_head, infos)
505 		hi->keepalive--;
506 
507 	/* Check hello keepalives */
508 	SLIST_FOREACH_SAFE(hi, &hello_info_head, infos, hinext)
509 		if (hi->keepalive < 1)
510 			SLIST_REMOVE(&hello_info_head, hi, hello_info, infos);
511 
512 	/* Set the alarm again and bail out */
513 	alarm(1);
514 	errno = olderrno;
515 }
516 
517 void
518 bail_out(int x)
519 {
520 	ldp_peer_holddown_all();
521 	flush_mpls_routes();
522 	exit(0);
523 }
524 
525 /*
526  * The big poll that catches every single event
527  * on every socket.
528  */
529 void
530 the_big_loop(void)
531 {
532 	int		sock_error;
533 	uint32_t	i;
534 	socklen_t       sock_error_size = sizeof(int);
535 	struct ldp_peer *p;
536 	struct com_sock	*cs;
537 	struct pollfd	pfd[MAX_POLL_FDS];
538 
539 	SLIST_INIT(&hello_info_head);
540 
541 	signal(SIGALRM, send_hello_alarm);
542 	signal(SIGPIPE, SIG_IGN);
543 	signal(SIGTERM, bail_out);
544 	send_hello_alarm(1);
545 
546 	route_socket = socket(PF_ROUTE, SOCK_RAW, AF_UNSPEC);
547 
548 	if (bind_current_routes() != LDP_E_OK)
549 		fatalp("Cannot get current routes\n");
550 
551 	for (;;) {
552 		nfds_t pollsum = 4;
553 
554 		pfd[0].fd = ls;
555 		pfd[0].events = POLLRDNORM;
556 		pfd[0].revents = 0;
557 
558 		pfd[1].fd = route_socket;
559 		pfd[1].events = POLLRDNORM;
560 		pfd[1].revents = 0;
561 
562 		pfd[2].fd = command_socket;
563 		pfd[2].events = POLLRDNORM;
564 		pfd[2].revents = 0;
565 
566 		/* Hello socket */
567 		pfd[3].fd = hello_socket;
568 		pfd[3].events = POLLIN;
569 		pfd[3].revents = 0;
570 
571 		/* Command sockets */
572 		for (i=0; i < MAX_COMMAND_SOCKETS; i++)
573 			if (csockets[i].socket != -1) {
574 				pfd[pollsum].fd = csockets[i].socket;
575 				pfd[pollsum].events = POLLIN;
576 				pfd[pollsum].revents = 0;
577 				pollsum++;
578 			}
579 
580 		/* LDP Peer sockets */
581 		SLIST_FOREACH(p, &ldp_peer_head, peers) {
582 			if (p->socket < 1)
583 				continue;
584 			switch (p->state) {
585 			    case LDP_PEER_CONNECTED:
586 			    case LDP_PEER_ESTABLISHED:
587 				pfd[pollsum].fd = p->socket;
588 				pfd[pollsum].events = POLLRDNORM;
589 				pfd[pollsum].revents = 0;
590 				pollsum++;
591 				break;
592 			    case LDP_PEER_CONNECTING:
593 				pfd[pollsum].fd = p->socket;
594 				pfd[pollsum].events = POLLWRNORM;
595 				pfd[pollsum].revents = 0;
596 				pollsum++;
597 				break;
598 			}
599 		}
600 
601 		if (pollsum >= MAX_POLL_FDS) {
602 			fatalp("Too many sockets. Increase MAX_POLL_FDS\n");
603 			return;
604 			}
605 		if (poll(pfd, pollsum, INFTIM) < 0) {
606 			if (errno != EINTR)
607 				fatalp("poll: %s", strerror(errno));
608 			continue;
609 			}
610 
611 		for (i = 0; i < pollsum; i++) {
612 			if ((pfd[i].revents & POLLRDNORM) ||
613 			    (pfd[i].revents & POLLIN)) {
614 				if(pfd[i].fd == ls) {
615 					new_peer_connection();
616 				} else if (pfd[i].fd == route_socket) {
617 					struct rt_msg xbuf;
618 					int l, to_read;
619 					do {
620 					    l = recv(route_socket, &xbuf,
621 					      sizeof(struct rt_msg), MSG_PEEK);
622 					} while ((l == -1) && (errno == EINTR));
623 
624 					if (l == -1)
625 						break;
626 
627 					to_read = l;
628 					l = 0;
629 					do {
630 					    l += recv(route_socket, &xbuf,
631 						to_read - l, MSG_WAITALL);
632 					} while (l != to_read);
633 
634 					check_route(&xbuf, to_read);
635 
636 				} else if (pfd[i].fd == hello_socket) {
637 					/* Receiving hello socket */
638 					recv_pdu(pfd[i].fd);
639 				} else if (pfd[i].fd == command_socket) {
640 					command_accept(command_socket);
641 				} else if ((cs = is_command_socket(pfd[i].fd))
642 						!= NULL) {
643 					command_dispatch(cs);
644 				} else {
645 					/* ldp peer socket */
646 					p = get_ldp_peer_by_socket(pfd[i].fd);
647 					if (p)
648 						recv_session_pdu(p);
649 				}
650 			} else if(pfd[i].revents & POLLWRNORM) {
651 				p = get_ldp_peer_by_socket(pfd[i].fd);
652 				if (!p)
653 					continue;
654 				if ((getsockopt(pfd[i].fd, SOL_SOCKET, SO_ERROR,
655 					&sock_error, &sock_error_size) != 0) ||
656 					    (sock_error)) {
657 						ldp_peer_holddown(p);
658 					} else {
659 						p->state = LDP_PEER_CONNECTED;
660 						send_initialize(p);
661 				}
662 			}
663 		}
664 
665 		for (int ri = 0; ri < replay_index; ri++) {
666 			debugp("Replaying: PID %d, SEQ %d\n",
667 				replay_rt[ri].m_rtm.rtm_pid,
668 				replay_rt[ri].m_rtm.rtm_seq);
669 			check_route(&replay_rt[ri], sizeof(struct rt_msg));
670                 }
671 		replay_index = 0;
672 	}	/* for (;;) */
673 }
674 
675 void
676 new_peer_connection()
677 {
678 	struct sockaddr_in sa, sin_me;
679 	int             s;
680 
681 	s = accept(ls, (struct sockaddr *) & sa,
682 		& (socklen_t) { sizeof(struct sockaddr_in) } );
683 	if (s < 0) {
684 		fatalp("accept: %s", strerror(errno));
685 		return;
686 	}
687 
688 	if (get_ldp_peer(&sa.sin_addr) != NULL) {
689 		close(s);
690 		return;
691 	}
692 
693 	warnp("Accepted a connection from %s\n", inet_ntoa(sa.sin_addr));
694 
695 	if (getsockname(s, (struct sockaddr *)&sin_me,
696 	    & (socklen_t) { sizeof(struct sockaddr_in) } )) {
697 		fatalp("new_peer_connection(): cannot getsockname\n");
698 		close(s);
699 		return;
700 	}
701 
702 	if (ntohl(sa.sin_addr.s_addr) < ntohl(sin_me.sin_addr.s_addr)) {
703 		fatalp("Peer %s: connect from lower ID\n",
704 		    inet_ntoa(sa.sin_addr));
705 		close(s);
706 		return;
707 	}
708 	/* XXX: sa.sin_addr ain't peer LDP ID ... */
709 	ldp_peer_new(&sa.sin_addr, &sa.sin_addr, NULL, ldp_holddown_time, s);
710 
711 }
712 
713 void
714 send_initialize(struct ldp_peer * p)
715 {
716 	struct init_tlv ti;
717 
718 	ti.type = htons(LDP_INITIALIZE);
719 	ti.length = htons(sizeof(struct init_tlv) - TLV_TYPE_LENGTH);
720 	ti.messageid = htonl(get_message_id());
721 	ti.cs_type = htons(TLV_COMMON_SESSION);
722 	ti.cs_len = htons(CS_LEN);
723 	ti.cs_version = htons(LDP_VERSION);
724 	ti.cs_keepalive = htons(2 * ldp_keepalive_time);
725 	ti.cs_adpvlim = 0;
726 	ti.cs_maxpdulen = htons(MAX_PDU_SIZE);
727 	ti.cs_peeraddress.s_addr = p->ldp_id.s_addr;
728 	ti.cs_peeraddrspace = 0;
729 
730 	send_tlv(p, (struct tlv *) (void *) &ti);
731 }
732 
733 void
734 keep_alive(struct ldp_peer * p)
735 {
736 	struct ka_tlv   kt;
737 
738 	kt.type = htons(LDP_KEEPALIVE);
739 	kt.length = htons(sizeof(kt.messageid));
740 	kt.messageid = htonl(get_message_id());
741 
742 	send_tlv(p, (struct tlv *) (void *) &kt);
743 
744 }
745 
746 void
747 recv_session_pdu(struct ldp_peer * p)
748 {
749 	struct ldp_pdu *rpdu;
750 	struct address_tlv *atlv;
751 	struct al_tlv  *altlv;
752 	struct init_tlv	*itlv;
753 	struct label_map_tlv *lmtlv;
754 	struct fec_tlv *fectlv;
755 	struct label_tlv *__packed labeltlv;
756 	struct notification_tlv *nottlv;
757 	struct hello_info *hi;
758 
759 	int             c;
760 	int32_t         wo = 0;
761 	struct tlv     *ttmp;
762 	unsigned char   recvspace[MAX_PDU_SIZE];
763 
764 	memset(recvspace, 0, MAX_PDU_SIZE);
765 
766 	c = recv(p->socket, (void *) recvspace, MAX_PDU_SIZE, MSG_PEEK);
767 
768 	debugp("Ready to read %d bytes\n", c);
769 
770 	if (c < 1) {		/* Session closed */
771 		warnp("Error in connection with %s\n", inet_ntoa(p->ldp_id));
772 		ldp_peer_holddown(p);
773 		return;
774 	}
775 	if (c > MAX_PDU_SIZE) {
776 		debugp("Incoming PDU size exceeds MAX_PDU_SIZE !\n");
777 		return;
778 	}
779 	if (c < MIN_PDU_SIZE) {
780 		debugp("PDU too small received from peer %s\n", inet_ntoa(p->ldp_id));
781 		return;
782 	}
783 	rpdu = (struct ldp_pdu *) recvspace;
784 	/* XXX: buggy messages may crash the whole thing */
785 	c = recv(p->socket, (void *) recvspace,
786 		ntohs(rpdu->length) + PDU_VER_LENGTH, MSG_WAITALL);
787 	rpdu = (struct ldp_pdu *) recvspace;
788 
789 	/* Check if it's somehow OK... */
790 	if (check_recv_pdu(p, rpdu, c) != 0)
791 		return;
792 
793 	debugp("Read %d bytes, PDU size: %d bytes\n", c, ntohs(rpdu->length));
794 	wo = sizeof(struct ldp_pdu);
795 
796 	while (wo + TLV_TYPE_LENGTH < (uint)c) {
797 
798 		ttmp = (struct tlv *) (&recvspace[wo]);
799 
800 		if ((ntohs(ttmp->type) != LDP_KEEPALIVE) &&
801 		    (ntohs(ttmp->type) != LDP_LABEL_MAPPING)) {
802 			debugp("Got Type: 0x%.4X (Length: %d) from %s\n",
803 			    ntohs(ttmp->type), ntohs(ttmp->length),
804 			    inet_ntoa(p->ldp_id));
805 		} else
806 			debugp("Got Type: 0x%.4X (Length: %d) from %s\n",
807 			    ntohs(ttmp->type), ntohs(ttmp->length),
808 			    inet_ntoa(p->ldp_id));
809 
810 		/* Should we get the message ? */
811 		if (p->state != LDP_PEER_ESTABLISHED &&
812 		    ntohs(ttmp->type) != LDP_INITIALIZE &&
813 		    ntohs(ttmp->type) != LDP_KEEPALIVE)
814 			break;
815 		/* The big switch */
816 		switch (ntohs(ttmp->type)) {
817 		case LDP_INITIALIZE:
818 			itlv = (struct init_tlv *)ttmp;
819 			/* Check size */
820 			if (ntohs(itlv->length) <
821 			    sizeof(struct init_tlv) - TLV_TYPE_LENGTH) {
822 				send_notification(p, 0,
823 				    NOTIF_BAD_PDU_LEN | NOTIF_FATAL);
824 				ldp_peer_holddown(p);
825 				break;
826 			}
827 			/* Check version */
828 			if (ntohs(itlv->cs_version) != LDP_VERSION) {
829 				send_notification(p, ntohl(itlv->messageid),
830 					NOTIF_BAD_LDP_VER | NOTIF_FATAL);
831 				ldp_peer_holddown(p);
832 				break;
833 			}
834 			/* Check if we got any hello from this one */
835 			SLIST_FOREACH(hi, &hello_info_head, infos)
836 				if (hi->ldp_id.s_addr == rpdu->ldp_id.s_addr)
837 					break;
838 			if (hi == NULL) {
839 			    send_notification(p, ntohl(itlv->messageid),
840 				NOTIF_SESSION_REJECTED_NO_HELLO | NOTIF_FATAL);
841 			    ldp_peer_holddown(p);
842 			    break;
843 			}
844 
845 			if (!p->master) {
846 				keep_alive(p);
847 				send_initialize(p);
848 			} else {
849 				p->state = LDP_PEER_ESTABLISHED;
850 				p->established_t = time(NULL);
851 				keep_alive(p);
852 
853 				/*
854 				 * Recheck here ldp id because we accepted
855 				 * connection without knowing who is it for sure
856 				 */
857 				p->ldp_id.s_addr = rpdu->ldp_id.s_addr;
858 
859 				fatalp("LDP neighbour %s is UP\n",
860 				    inet_ntoa(p->ldp_id));
861 				mpls_add_ldp_peer(p);
862 				send_addresses(p);
863 				send_all_bindings(p);
864 			}
865 			break;
866 		case LDP_KEEPALIVE:
867 			if ((p->state == LDP_PEER_CONNECTED) && (!p->master)) {
868 				p->state = LDP_PEER_ESTABLISHED;
869 				p->established_t = time(NULL);
870 				fatalp("LDP neighbour %s is UP\n",
871 				    inet_ntoa(p->ldp_id));
872 				mpls_add_ldp_peer(p);
873 				send_addresses(p);
874 				send_all_bindings(p);
875 			}
876 			p->timeout = p->holdtime;
877 			break;
878 		case LDP_ADDRESS:
879 			/* Add peer addresses */
880 			atlv = (struct address_tlv *) ttmp;
881 			altlv = (struct al_tlv *) (&atlv[1]);
882 			add_ifaddresses(p, altlv);
883 			print_bounded_addresses(p);
884 			break;
885 		case LDP_ADDRESS_WITHDRAW:
886 			atlv = (struct address_tlv *) ttmp;
887 			altlv = (struct al_tlv *) (&atlv[1]);
888 			del_ifaddresses(p, altlv);
889 			break;
890 		case LDP_LABEL_MAPPING:
891 			lmtlv = (struct label_map_tlv *) ttmp;
892 			fectlv = (struct fec_tlv *) (&lmtlv[1]);
893 			labeltlv = (struct label_tlv *)((unsigned char *)fectlv
894 				+ ntohs(fectlv->length) + TLV_TYPE_LENGTH);
895 			map_label(p, fectlv, labeltlv);
896 			break;
897 		case LDP_LABEL_REQUEST:
898 			lmtlv = (struct label_map_tlv *) ttmp;
899 			fectlv = (struct fec_tlv *) (&lmtlv[1]);
900 			switch (request_respond(p, lmtlv, fectlv)) {
901 			case LDP_E_BAD_FEC:
902 				send_notification(p, ntohl(lmtlv->messageid),
903 					NOTIF_UNKNOWN_TLV);
904 				break;
905 			case LDP_E_BAD_AF:
906 				send_notification(p, ntohl(lmtlv->messageid),
907 					NOTIF_UNSUPPORTED_AF);
908 				break;
909 			case LDP_E_NO_SUCH_ROUTE:
910 				send_notification(p, ntohl(lmtlv->messageid),
911 					NOTIF_NO_ROUTE);
912 				break;
913 			}
914 			break;
915 		case LDP_LABEL_WITHDRAW:
916 			lmtlv = (struct label_map_tlv *) ttmp;
917 			fectlv = (struct fec_tlv *) (&lmtlv[1]);
918 			if (withdraw_label(p, fectlv) == LDP_E_OK) {
919 				/* Send RELEASE */
920 				prepare_release(ttmp);
921 				send_tlv(p, ttmp);
922 				}
923 			break;
924 		case LDP_LABEL_RELEASE:
925 			/*
926 			 * XXX: we need to make a timed queue...
927 			 * For now I just assume peers are processing messages
928 			 * correctly so I just ignore confirmations
929 			 */
930 			wo = -1;	/* Ignore rest of message */
931 			break;
932 		case LDP_LABEL_ABORT:
933 		/* XXX: For now I pretend I can process everything
934 		 * RFC 3036, Section 3.5.9.1
935 		 * If an LSR receives a Label Abort Request Message after it
936 		 * has responded to the Label Request in question with a Label
937 		 * Mapping message or a Notification message, it ignores the
938 		 * abort request.
939 		 */
940 			wo = -1;
941 			break;
942 		case LDP_NOTIFICATION:
943 			nottlv = (struct notification_tlv *) ttmp;
944 			nottlv->st_code = ntohl(nottlv->st_code);
945 			fatalp("Got notification 0x%X from peer %s\n",
946 			    nottlv->st_code, inet_ntoa(p->ldp_id));
947 			if (nottlv->st_code >> 31) {
948 				fatalp("LDP peer %s signalized %s\n",
949 				    inet_ntoa(p->ldp_id),
950 				    NOTIF_STR[(nottlv->st_code << 1) >> 1]);
951 				ldp_peer_holddown(p);
952 				wo = -1;
953 			}
954 			break;
955 		case LDP_HELLO:
956 			/* No hellos should came on tcp session */
957 			wo = -1;
958 			break;
959 		default:
960 			warnp("Unknown TLV received from %s\n",
961 			    inet_ntoa(p->ldp_id));
962 			debug_tlv(ttmp);
963 			wo = -1;/* discard the rest of the message */
964 			break;
965 		}
966 		if (wo < 0) {
967 			debugp("Discarding the rest of the message\n");
968 			break;
969 		} else {
970 			wo += ntohs(ttmp->length) + TLV_TYPE_LENGTH;
971 			debugp("WORKED ON %u bytes (Left %d)\n", wo, c - wo);
972 		}
973 	}			/* while */
974 
975 }
976 
977 /* Sends a pdu, tlv pair to a connected peer */
978 int
979 send_message(struct ldp_peer * p, struct ldp_pdu * pdu, struct tlv * t)
980 {
981 	unsigned char   sendspace[MAX_PDU_SIZE];
982 
983 	/* Check if peer is connected */
984 	switch (p->state) {
985 	case LDP_PEER_CONNECTED:
986 	case LDP_PEER_ESTABLISHED:
987 		break;
988 	default:
989 		return -1;
990 	}
991 
992 	/* Check length validity first */
993 	if (ntohs(pdu->length) !=
994 	    ntohs(t->length) + TLV_TYPE_LENGTH + PDU_PAYLOAD_LENGTH) {
995 		fatalp("LDP: TLV - PDU incompability. Message discarded\n");
996 		fatalp("LDP: TLV len %d - PDU len %d\n", ntohs(t->length),
997 		    ntohs(pdu->length));
998 		return -1;
999 	}
1000 	if (ntohs(t->length) + PDU_VER_LENGTH > MAX_PDU_SIZE) {
1001 		fatalp("Message to large discarded\n");
1002 		return -1;
1003 	}
1004 	/* Arrange them in a buffer and send */
1005 	memcpy(sendspace, pdu, sizeof(struct ldp_pdu));
1006 	memcpy(sendspace + sizeof(struct ldp_pdu), t,
1007 	    ntohs(t->length) + TLV_TYPE_LENGTH);
1008 
1009 	/* Report keepalives only for DEBUG */
1010 	if ((ntohs(t->type) != 0x201) && (ntohs(t->type) != 0x400)) {
1011 		debugp("Sending message type 0x%.4X to %s (size: %d)\n",
1012 		    ntohs(t->type), inet_ntoa(p->ldp_id), ntohs(t->length));
1013 	} else
1014 	/* downgraded from warnp to debugp for now */
1015 		debugp("Sending message type 0x%.4X to %s (size: %d)\n",
1016 		    ntohs(t->type), inet_ntoa(p->ldp_id), ntohs(t->length));
1017 
1018 	/* Send it finally */
1019 	return send(p->socket, sendspace,
1020 		ntohs(pdu->length) + PDU_VER_LENGTH, 0);
1021 }
1022 
1023 /*
1024  * Encapsulates TLV into a PDU and sends it to a peer
1025  */
1026 int
1027 send_tlv(struct ldp_peer * p, struct tlv * t)
1028 {
1029 	struct ldp_pdu  pdu;
1030 
1031 	pdu.version = htons(LDP_VERSION);
1032 	inet_aton(LDP_ID, &pdu.ldp_id);
1033 	pdu.label_space = 0;
1034 	pdu.length = htons(ntohs(t->length) + TLV_TYPE_LENGTH +
1035 		PDU_PAYLOAD_LENGTH);
1036 
1037 	return send_message(p, &pdu, t);
1038 }
1039 
1040 
1041 int
1042 send_addresses(struct ldp_peer * p)
1043 {
1044 	struct address_list_tlv *t;
1045 	int             ret;
1046 
1047 	t = build_address_list_tlv();
1048 
1049 	ret = send_tlv(p, (struct tlv *) t);
1050 	free(t);
1051 	return ret;
1052 
1053 }
1054