Home | History | Annotate | Line # | Download | only in ldpd
socketops.c revision 1.1
      1 /* $NetBSD: socketops.c,v 1.1 2010/12/08 07:20:15 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 (at) 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 
     74 void	recv_pdu(int);
     75 void	send_hello_alarm(int);
     76 void	bail_out(int);
     77 static int get_local_addr(struct sockaddr_dl *, struct in_addr *);
     78 
     79 int
     80 create_hello_socket()
     81 {
     82 	struct ip_mreq  mcast_addr;
     83 	int             s = socket(PF_INET, SOCK_DGRAM, 17);
     84 
     85 	if (s < 0)
     86 		return s;
     87 
     88 	/*
     89 	 * RFC3036 specifies we should listen to all subnet routers multicast
     90 	 * group
     91 	 */
     92 	mcast_addr.imr_multiaddr.s_addr = inet_addr(ALL_ROUTERS);
     93 	mcast_addr.imr_interface.s_addr = htonl(INADDR_ANY);
     94 
     95 	socket_reuse_port(s);
     96 	/* Bind it to port 646 on specific address */
     97 	if (bind_socket(s, htonl(INADDR_ANY)) == -1) {
     98 		warnp("Cannot bind hello socket\n");
     99 		close(s);
    100 		return -1;
    101 	}
    102 	/* We don't need to receive back our messages */
    103 	if (setsockopt(s, IPPROTO_IP, IP_MULTICAST_LOOP, &(uint8_t){0},
    104 	    sizeof(uint8_t)) == -1) {
    105 		fatalp("setsockopt: %s", strerror(errno));
    106 		close(s);
    107 		return -1;
    108 	}
    109 	/* Finally join the group */
    110         if (setsockopt(s, IPPROTO_IP, IP_ADD_MEMBERSHIP, (char *) &mcast_addr,
    111 	    sizeof(mcast_addr)) == -1) {
    112                 fatalp("setsockopt: %s", strerror(errno));
    113                 close(s);
    114                 return -1;
    115         }
    116 	/* TTL:1, TOS: 0xc0 */
    117 	if (set_mcast_ttl(s) == -1) {
    118 		close(s);
    119 		return -1;
    120 	}
    121 	if (set_tos(s) == -1) {
    122 		fatalp("set_tos: %s", strerror(errno));
    123 		close(s);
    124 		return -1;
    125 	}
    126 	if (setsockopt(s, IPPROTO_IP, IP_RECVIF, &(uint32_t){1}, sizeof(uint32_t)) == -1) {
    127 		fatalp("Cannot set IP_RECVIF\n");
    128 		close(s);
    129 		return -1;
    130 	}
    131 	hello_socket = s;
    132 	return hello_socket;
    133 }
    134 
    135 /* Sets the TTL to 1 as we don't want to transmit outside this subnet */
    136 int
    137 set_ttl(int s)
    138 {
    139 	int             ret;
    140 	if ((ret = setsockopt(s, IPPROTO_IP, IP_TTL, &(int){1}, sizeof(int)))
    141 	    == -1)
    142 		fatalp("set_ttl: %s", strerror(errno));
    143 	return ret;
    144 }
    145 
    146 /* Sets multicast TTL to 1 */
    147 int
    148 set_mcast_ttl(int s)
    149 {
    150 	int	ret;
    151 	if ((ret = setsockopt(s, IPPROTO_IP, IP_MULTICAST_TTL, &(int){1},
    152 	    sizeof(int))) == -1)
    153 		fatalp("set_mcast_ttl: %s", strerror(errno));
    154 	return ret;
    155 }
    156 
    157 /* Sets TOS to 0xc0 aka IP Precedence 6 */
    158 int
    159 set_tos(int s)
    160 {
    161 	int             ret;
    162 	if ((ret = setsockopt(s, IPPROTO_IP, IP_TOS, &(int){0xc0},
    163 	    sizeof(int))) == -1)
    164 		fatalp("set_tos: %s", strerror(errno));
    165 	return ret;
    166 }
    167 
    168 int
    169 socket_reuse_port(int s)
    170 {
    171 	int             ret;
    172 	if ((ret = setsockopt(s, SOL_SOCKET, SO_REUSEPORT, &(int){1},
    173 	    sizeof(int))) == -1)
    174 		fatalp("socket_reuse_port: %s", strerror(errno));
    175 	return ret;
    176 }
    177 
    178 /* binds an UDP socket */
    179 int
    180 bind_socket(int s, uint32_t addr)
    181 {
    182 	struct sockaddr_in sa;
    183 
    184 	sa.sin_len = sizeof(sa);
    185 	sa.sin_family = AF_INET;
    186 	sa.sin_port = htons(LDP_PORT);
    187 	sa.sin_addr.s_addr = addr;
    188 	if (bind(s, (struct sockaddr *) (&sa), sizeof(sa))) {
    189 		fatalp("bind_socket: %s", strerror(errno));
    190 		return -1;
    191 	}
    192 	return 0;
    193 }
    194 
    195 /* Create / bind the TCP socket */
    196 int
    197 create_listening_socket(void)
    198 {
    199 	struct sockaddr_in sa;
    200 	int             s;
    201 
    202 	sa.sin_len = sizeof(sa);
    203 	sa.sin_family = AF_INET;
    204 	sa.sin_port = htons(LDP_PORT);
    205 	sa.sin_addr.s_addr = htonl(INADDR_ANY);
    206 
    207 	s = socket(PF_INET, SOCK_STREAM, 6);
    208 	if (s < 0)
    209 		return s;
    210 	if (bind(s, (struct sockaddr *) & sa, sizeof(sa))) {
    211 		fatalp("bind: %s", strerror(errno));
    212 		close(s);
    213 		return -1;
    214 	}
    215 	if (listen(s, 10) == -1) {
    216 		fatalp("listen: %s", strerror(errno));
    217 		close(s);
    218 		return -1;
    219 	}
    220 /*	if (set_tos(s) == -1) {
    221 		fatalp("set_tos: %s", strerror(errno));
    222 		close(s);
    223 		return -1;
    224 	}
    225 */	return s;
    226 }
    227 
    228 /*
    229  * It's ugly. We need a function to pass all tlvs and create pdu but since I
    230  * use UDP socket only to send hellos, I didn't bother
    231  */
    232 void
    233 send_hello(void)
    234 {
    235 	struct hello_tlv *t;
    236 	struct common_hello_tlv *cht;
    237 	struct ldp_pdu  *spdu;
    238 	struct transport_address_tlv *trtlv;
    239 	void *v;
    240 	struct sockaddr_in sadest;	/* Destination ALL_ROUTERS */
    241 	int sb = 0;			/* sent bytes */
    242 	struct ifaddrs *ifa, *ifb;
    243 	struct sockaddr_in *if_sa;
    244 	char lastifname[20];
    245 
    246 #define HELLO_MSG_SIZE (sizeof(struct ldp_pdu) + 	/* PDU */	\
    247 			TLV_TYPE_LENGTH + MSGID_SIZE +	/* Hello TLV */	\
    248 			/* Common Hello TLV */				\
    249 			sizeof(struct common_hello_tlv) +		\
    250 			/* IPv4 Transport Address */			\
    251 			sizeof(struct transport_address_tlv))
    252 
    253 	if ((v = malloc(HELLO_MSG_SIZE)) == NULL) {
    254 		fatalp("malloc problem in send_hello()\n");
    255 		return;
    256 	}
    257 	memset(v, 0, HELLO_MSG_SIZE);
    258 
    259 	spdu = (struct ldp_pdu *)((char *)v);
    260 	t = (struct hello_tlv *)(spdu + 1);
    261 	cht = &t->ch;	/* Hello tlv struct includes CHT */
    262 	trtlv = (struct transport_address_tlv *)(t + 1);
    263 
    264 	/* Prepare PDU envelope */
    265 	spdu->version = htons(LDP_VERSION);
    266 	spdu->length = htons(HELLO_MSG_SIZE - PDU_VER_LENGTH);
    267 	inet_aton(LDP_ID, &spdu->ldp_id);
    268 
    269 	/* Prepare Hello TLV */
    270 	t->type = htons(LDP_HELLO);
    271 	t->length = htons(MSGID_SIZE +
    272 			sizeof(struct common_hello_tlv) +
    273 			sizeof(struct transport_address_tlv));
    274 	/*
    275 	 * I used ID 0 instead of htonl(get_message_id()) because I've
    276 	 * seen hellos from a cisco router doing the same thing
    277 	 */
    278 	t->messageid = 0;
    279 
    280 	/* Prepare Common Hello attributes */
    281 	cht->type = htons(TLV_COMMON_HELLO);
    282 	cht->length = htons(sizeof(cht->holdtime) + sizeof(cht->res));
    283 	cht->holdtime = htons(LDP_HOLDTIME);
    284 	cht->res = 0;
    285 
    286 	/*
    287 	 * Prepare Transport Address TLV RFC3036 says: "If this optional TLV
    288 	 * is not present the IPv4 source address for the UDP packet carrying
    289 	 * the Hello should be used." But we send it because everybody seems
    290 	 * to do so
    291 	 */
    292 	trtlv->type = htons(TLV_IPV4_TRANSPORT);
    293 	trtlv->length = htons(sizeof(struct in_addr));
    294 	/* trtlv->address will be set for each socket */
    295 
    296 	/* Destination sockaddr */
    297 	memset(&sadest, 0, sizeof(sadest));
    298 	sadest.sin_len = sizeof(sadest);
    299 	sadest.sin_family = AF_INET;
    300 	sadest.sin_port = htons(LDP_PORT);
    301 	inet_aton(ALL_ROUTERS, &sadest.sin_addr);
    302 
    303 	if (getifaddrs(&ifa) == -1) {
    304 		free(v);
    305 		return;
    306 	}
    307 
    308 	lastifname[0] = '\0';
    309 	for (ifb = ifa; ifb; ifb = ifb->ifa_next) {
    310 		if_sa = (struct sockaddr_in *) ifb->ifa_addr;
    311 		if (if_sa->sin_family != AF_INET)
    312 			continue;
    313 		if (ntohl(if_sa->sin_addr.s_addr) >> 24 == IN_LOOPBACKNET)
    314 			continue;
    315 		/* Send only once per interface, using master address */
    316 		if (strcmp(ifb->ifa_name, lastifname) == 0)
    317 			continue;
    318 		debugp("Sending hello on %s\n", ifb->ifa_name);
    319 		if (setsockopt(hello_socket, IPPROTO_IP, IP_MULTICAST_IF,
    320 		    &if_sa->sin_addr, sizeof(struct in_addr)) == -1) {
    321 			warnp("setsockopt failed: %s\n", strerror(errno));
    322 			continue;
    323 		}
    324 		trtlv->address.s_addr = if_sa->sin_addr.s_addr;
    325 
    326 		strlcpy(lastifname, ifb->ifa_name, sizeof(lastifname));
    327 
    328 		/* Send to the wire */
    329 		sb = sendto(hello_socket, v, HELLO_MSG_SIZE,
    330 			    0, (struct sockaddr *) & sadest, sizeof(sadest));
    331 		if (sb < (int)HELLO_MSG_SIZE)
    332 		    fatalp("send: %s", strerror(errno));
    333 		else
    334 		    debugp("Send %d bytes (PDU: %d, Hello TLV: %d, CH: %d)\n",
    335 			sb, (int) (sizeof(struct ldp_pdu) - PDU_VER_LENGTH),
    336 		       (int) (TLV_TYPE_LENGTH + MSGID_SIZE),
    337 		       (int) (sizeof(struct common_hello_tlv)));
    338 
    339 	}
    340 	freeifaddrs(ifa);
    341 	free(v);
    342 }
    343 
    344 int
    345 get_message_id(void)
    346 {
    347 	current_msg_id++;
    348 	return current_msg_id;
    349 }
    350 
    351 static int
    352 get_local_addr(struct sockaddr_dl *sdl, struct in_addr *sin)
    353 {
    354 	struct ifaddrs *ifa, *ifb;
    355 	struct sockaddr_in *sinet;
    356 
    357 	if (sdl == NULL)
    358 		return -1;
    359 
    360 	if (getifaddrs(&ifa) == -1)
    361 		return -1;
    362 	for (ifb = ifa; ifb; ifb = ifb->ifa_next)
    363 		if (ifb->ifa_addr->sa_family == AF_INET) {
    364 			if (if_nametoindex(ifb->ifa_name) != sdl->sdl_index)
    365 				continue;
    366 			sinet = (struct sockaddr_in*) ifb->ifa_addr;
    367 			sin->s_addr = sinet->sin_addr.s_addr;
    368 			freeifaddrs(ifa);
    369 			return 0;
    370 		}
    371 	freeifaddrs(ifa);
    372 	return -1;
    373 }
    374 
    375 /* Receive PDUs on Multicast UDP socket */
    376 void
    377 recv_pdu(int sock)
    378 {
    379 	struct ldp_pdu  rpdu;
    380 	int             c, i;
    381 	struct msghdr msg;
    382 	struct iovec iov[1];
    383 	unsigned char recvspace[MAX_PDU_SIZE];
    384 	struct hello_tlv *t;
    385 	struct sockaddr_in fromsa;
    386 	struct sockaddr_dl *sdl = NULL;
    387 	struct in_addr my_ldp_addr, local_addr;
    388 	struct cmsghdr *cmptr;
    389 	union {
    390 		struct cmsghdr cm;
    391 		char control[1024];
    392 	} control_un;
    393 
    394 	debugp("Entering RECV_PDU\n");
    395 
    396 	memset(&msg, 0, sizeof(msg));
    397 	msg.msg_control = control_un.control;
    398 	msg.msg_controllen = sizeof(control_un.control);
    399 	msg.msg_flags = 0;
    400 	msg.msg_name = &fromsa;
    401 	msg.msg_namelen = sizeof(fromsa);
    402 	iov[0].iov_base = recvspace;
    403 	iov[0].iov_len = sizeof(recvspace);
    404 	msg.msg_iov = iov;
    405 	msg.msg_iovlen = 1;
    406 
    407 	c = recvmsg(sock, &msg, MSG_WAITALL);
    408 	debugp("Incoming PDU size: %d\n", c);
    409 
    410 	debugp("PDU from: %s\n", inet_ntoa(fromsa.sin_addr));
    411 
    412 	/* Check to see if this is larger than MIN_PDU_SIZE */
    413 	if (c < MIN_PDU_SIZE)
    414 		return;
    415 
    416 	/* Read the PDU */
    417 	i = get_pdu(recvspace, &rpdu);
    418 
    419 	/* We currently understand Version 1 */
    420 	if (rpdu.version != LDP_VERSION) {
    421 		fatalp("recv_pdu: Version mismatch\n");
    422 		return;
    423 	}
    424 
    425 	/* Maybe it's our hello */
    426 	inet_aton(LDP_ID, &my_ldp_addr);
    427 	if (rpdu.ldp_id.s_addr == my_ldp_addr.s_addr) {
    428 		fatalp("Received our PDU..\n");	/* it should be not looped */
    429 		return;
    430 	}
    431 
    432 	if (msg.msg_controllen < (socklen_t)sizeof(struct cmsghdr) ||
    433 	    (msg.msg_flags & MSG_CTRUNC))
    434 		local_addr.s_addr = my_ldp_addr.s_addr;
    435 	else {
    436 		for (cmptr = CMSG_FIRSTHDR(&msg); cmptr != NULL;
    437 		    cmptr = CMSG_NXTHDR(&msg, cmptr))
    438 			if (cmptr->cmsg_level == IPPROTO_IP &&
    439 			    cmptr->cmsg_type == IP_RECVIF) {
    440 				sdl = (struct sockaddr_dl *) CMSG_DATA(cmptr);
    441 				break;
    442 			}
    443 		if (get_local_addr(sdl, &local_addr) != 0)
    444 			local_addr.s_addr = my_ldp_addr.s_addr;
    445 	}
    446 
    447 
    448 	debugp("Read %d bytes from address %s Length: %.4d Version: %d\n",
    449 	       c, inet_ntoa(rpdu.ldp_id), rpdu.length, rpdu.version);
    450 
    451 	/* Fill the TLV messages */
    452 	t = get_hello_tlv(recvspace + i, c - i);
    453 	run_ldp_hello(&rpdu, t, &fromsa.sin_addr, &local_addr, sock);
    454 }
    455 
    456 void
    457 send_hello_alarm(int unused)
    458 {
    459 	struct ldp_peer *p;
    460 	struct hello_info *hi;
    461 	time_t          t = time(NULL);
    462 	int             olderrno = errno;
    463 
    464 	/* Send hellos */
    465 	if (!(t % ldp_hello_time))
    466 		send_hello();
    467 
    468 	/* Timeout -- */
    469 	SLIST_FOREACH(p, &ldp_peer_head, peers)
    470 		p->timeout--;
    471 
    472 	/* Check for timeout */
    473 check_peer:
    474 	SLIST_FOREACH(p, &ldp_peer_head, peers)
    475 		if (p->timeout < 1)
    476 			switch (p->state) {
    477 			case LDP_PEER_HOLDDOWN:
    478 				debugp("LDP holddown expired for peer %s\n",
    479 				       inet_ntoa(p->ldp_id));
    480 				ldp_peer_delete(p);
    481 				goto check_peer;
    482 			case LDP_PEER_ESTABLISHED:
    483 			case LDP_PEER_CONNECTED:
    484 				send_notification(p, 0,
    485 				    NOTIF_KEEP_ALIVE_TIMER_EXPIRED);
    486 				warnp("Keepalive expired for %s\n",
    487 				    inet_ntoa(p->ldp_id));
    488 				ldp_peer_holddown(p);
    489 				break;
    490 			}	/* switch */
    491 
    492 	/* send keepalives */
    493 	if (!(t % LDP_KEEPALIVE_TIME)) {
    494 		SLIST_FOREACH(p, &ldp_peer_head, peers)
    495 		    if (p->state == LDP_PEER_ESTABLISHED) {
    496 			debugp("Sending KeepAlive to %s\n",
    497 			    inet_ntoa(p->ldp_id));
    498 			keep_alive(p);
    499 		    }
    500 	}
    501 
    502 	/* Decrement hello info keepalives */
    503 	SLIST_FOREACH(hi, &hello_info_head, infos)
    504 		hi->keepalive--;
    505 
    506 	/* Check hello keepalives */
    507 check_hello:
    508 	SLIST_FOREACH(hi, &hello_info_head, infos)
    509 		if (hi->keepalive < 1) {
    510 			SLIST_REMOVE(&hello_info_head, hi, hello_info, infos);
    511 			goto check_hello;
    512 		}
    513 
    514 	/* Set the alarm again and bail out */
    515 	alarm(1);
    516 	errno = olderrno;
    517 }
    518 
    519 void
    520 bail_out(int x)
    521 {
    522 	ldp_peer_holddown_all();
    523 	flush_mpls_routes();
    524 	exit(0);
    525 }
    526 
    527 /*
    528  * The big poll that catches every single event
    529  * on every socket.
    530  */
    531 void
    532 the_big_loop(void)
    533 {
    534 	int		sock_error;
    535 	uint32_t	i;
    536 	socklen_t       sock_error_size = sizeof(int);
    537 	struct ldp_peer *p;
    538 	struct com_sock	*cs;
    539 	struct pollfd	pfd[MAX_POLL_FDS];
    540 
    541 	SLIST_INIT(&hello_info_head);
    542 
    543 	signal(SIGALRM, send_hello_alarm);
    544 	signal(SIGPIPE, SIG_IGN);
    545 	signal(SIGTERM, bail_out);
    546 	send_hello_alarm(1);
    547 
    548 	route_socket = socket(PF_ROUTE, SOCK_RAW, AF_UNSPEC);
    549 
    550 	if (bind_current_routes() != LDP_E_OK)
    551 		fatalp("Cannot get current routes\n");
    552 
    553 	for (;;) {
    554 		nfds_t pollsum = 4;
    555 
    556 		pfd[0].fd = ls;
    557 		pfd[0].events = POLLRDNORM;
    558 		pfd[0].revents = 0;
    559 
    560 		pfd[1].fd = route_socket;
    561 		pfd[1].events = POLLRDNORM;
    562 		pfd[1].revents = 0;
    563 
    564 		pfd[2].fd = command_socket;
    565 		pfd[2].events = POLLRDNORM;
    566 		pfd[2].revents = 0;
    567 
    568 		/* Hello socket */
    569 		pfd[3].fd = hello_socket;
    570 		pfd[3].events = POLLIN;
    571 		pfd[3].revents = 0;
    572 
    573 		/* Command sockets */
    574 		for (i=0; i < MAX_COMMAND_SOCKETS; i++)
    575 			if (csockets[i].socket != -1) {
    576 				pfd[pollsum].fd = csockets[i].socket;
    577 				pfd[pollsum].events = POLLIN;
    578 				pfd[pollsum].revents = 0;
    579 				pollsum++;
    580 			}
    581 
    582 		/* LDP Peer sockets */
    583 		SLIST_FOREACH(p, &ldp_peer_head, peers) {
    584 			if (p->socket < 1)
    585 				continue;
    586 			switch (p->state) {
    587 			    case LDP_PEER_CONNECTED:
    588 			    case LDP_PEER_ESTABLISHED:
    589 				pfd[pollsum].fd = p->socket;
    590 				pfd[pollsum].events = POLLRDNORM;
    591 				pfd[pollsum].revents = 0;
    592 				pollsum++;
    593 				break;
    594 			    case LDP_PEER_CONNECTING:
    595 				pfd[pollsum].fd = p->socket;
    596 				pfd[pollsum].events = POLLWRNORM;
    597 				pfd[pollsum].revents = 0;
    598 				pollsum++;
    599 				break;
    600 			}
    601 		}
    602 
    603 		if (pollsum >= MAX_POLL_FDS) {
    604 			fatalp("Too many sockets. Increase MAX_POLL_FDS\n");
    605 			return;
    606 			}
    607 		if (poll(pfd, pollsum, INFTIM) < 0) {
    608 			if (errno != EINTR)
    609 				fatalp("poll: %s", strerror(errno));
    610 			continue;
    611 			}
    612 
    613 		for (i = 0; i < pollsum; i++) {
    614 			if ((pfd[i].revents & POLLRDNORM) ||
    615 			    (pfd[i].revents & POLLIN)) {
    616 				if(pfd[i].fd == ls) {
    617 					new_peer_connection();
    618 				} else if (pfd[i].fd == route_socket) {
    619 					struct rt_msg xbuf;
    620 					int l, to_read;
    621 					do {
    622 					    l = recv(route_socket, &xbuf,
    623 					      sizeof(struct rt_msg), MSG_PEEK);
    624 					} while ((l == -1) && (errno == EINTR));
    625 
    626 					if (l == -1)
    627 						break;
    628 
    629 					to_read = l;
    630 					l = 0;
    631 					do {
    632 					    l += recv(route_socket, &xbuf,
    633 						to_read - l, MSG_WAITALL);
    634 					} while (l != to_read);
    635 
    636 					check_route(&xbuf, to_read);
    637 
    638 				} else if (pfd[i].fd == hello_socket) {
    639 					/* Receiving hello socket */
    640 					recv_pdu(pfd[i].fd);
    641 				} else if (pfd[i].fd == command_socket) {
    642 					command_accept(command_socket);
    643 				} else if ((cs = is_command_socket(pfd[i].fd))
    644 						!= NULL) {
    645 					command_dispatch(cs);
    646 				} else {
    647 					/* ldp peer socket */
    648 					p = get_ldp_peer_by_socket(pfd[i].fd);
    649 					if (p)
    650 						recv_session_pdu(p);
    651 				}
    652 			} else if(pfd[i].revents & POLLWRNORM) {
    653 				p = get_ldp_peer_by_socket(pfd[i].fd);
    654 				if (!p)
    655 					continue;
    656 				if ((getsockopt(pfd[i].fd, SOL_SOCKET, SO_ERROR,
    657 					&sock_error, &sock_error_size) != 0) ||
    658 					    (sock_error)) {
    659 						ldp_peer_holddown(p);
    660 					} else {
    661 						p->state = LDP_PEER_CONNECTED;
    662 						send_initialize(p);
    663 				}
    664 			}
    665 		}
    666 
    667 		for (int ri = 0; ri < replay_index; ri++) {
    668 			debugp("Replaying: PID %d, SEQ %d\n",
    669 				replay_rt[ri].m_rtm.rtm_pid,
    670 				replay_rt[ri].m_rtm.rtm_seq);
    671 			check_route(&replay_rt[ri], sizeof(struct rt_msg));
    672                 }
    673 		replay_index = 0;
    674 	}	/* for (;;) */
    675 }
    676 
    677 void
    678 new_peer_connection()
    679 {
    680 	struct sockaddr_in sa, sin_me;
    681 	int             s;
    682 
    683 	s = accept(ls, (struct sockaddr *) & sa,
    684 		& (socklen_t) { sizeof(struct sockaddr_in) } );
    685 	if (s < 0) {
    686 		fatalp("accept: %s", strerror(errno));
    687 		return;
    688 	}
    689 
    690 	if (get_ldp_peer(&sa.sin_addr) != NULL) {
    691 		close(s);
    692 		return;
    693 	}
    694 
    695 	warnp("Accepted a connection from %s\n", inet_ntoa(sa.sin_addr));
    696 
    697 	if (getsockname(s, (struct sockaddr *)&sin_me,
    698 	    & (socklen_t) { sizeof(struct sockaddr_in) } )) {
    699 		fatalp("new_peer_connection(): cannot getsockname\n");
    700 		close(s);
    701 		return;
    702 	}
    703 
    704 	if (ntohl(sa.sin_addr.s_addr) < ntohl(sin_me.sin_addr.s_addr)) {
    705 		fatalp("Peer %s: connect from lower ID\n",
    706 		    inet_ntoa(sa.sin_addr));
    707 		close(s);
    708 		return;
    709 	}
    710 	/* XXX: sa.sin_addr ain't peer LDP ID ... */
    711 	ldp_peer_new(&sa.sin_addr, &sa.sin_addr, NULL, LDP_HOLDTIME, s);
    712 
    713 }
    714 
    715 void
    716 send_initialize(struct ldp_peer * p)
    717 {
    718 	struct init_tlv ti;
    719 
    720 	ti.type = htons(LDP_INITIALIZE);
    721 	ti.length = htons(sizeof(struct init_tlv) - TLV_TYPE_LENGTH);
    722 	ti.messageid = htonl(get_message_id());
    723 	ti.cs_type = htons(TLV_COMMON_SESSION);
    724 	ti.cs_len = htons(CS_LEN);
    725 	ti.cs_version = htons(LDP_VERSION);
    726 	ti.cs_keepalive = htons(2 * LDP_KEEPALIVE_TIME);
    727 	ti.cs_adpvlim = 0;
    728 	ti.cs_maxpdulen = htons(MAX_PDU_SIZE);
    729 	ti.cs_peeraddress.s_addr = p->ldp_id.s_addr;
    730 	ti.cs_peeraddrspace = 0;
    731 
    732 	send_tlv(p, (struct tlv *) (void *) &ti);
    733 }
    734 
    735 void
    736 keep_alive(struct ldp_peer * p)
    737 {
    738 	struct ka_tlv   kt;
    739 
    740 	kt.type = htons(LDP_KEEPALIVE);
    741 	kt.length = htons(sizeof(kt.messageid));
    742 	kt.messageid = htonl(get_message_id());
    743 
    744 	send_tlv(p, (struct tlv *) (void *) &kt);
    745 
    746 }
    747 
    748 void
    749 recv_session_pdu(struct ldp_peer * p)
    750 {
    751 	struct ldp_pdu *rpdu;
    752 	struct address_tlv *atlv;
    753 	struct al_tlv  *altlv;
    754 	struct init_tlv	*itlv;
    755 	struct label_map_tlv *lmtlv;
    756 	struct fec_tlv *fectlv;
    757 	struct label_tlv *__packed labeltlv;
    758 	struct notification_tlv *nottlv;
    759 	struct hello_info *hi;
    760 
    761 	int             c;
    762 	int32_t         wo = 0;
    763 	struct tlv     *ttmp;
    764 	unsigned char   recvspace[MAX_PDU_SIZE];
    765 
    766 	memset(recvspace, 0, MAX_PDU_SIZE);
    767 
    768 	c = recv(p->socket, (void *) recvspace, MAX_PDU_SIZE, MSG_PEEK);
    769 
    770 	debugp("Ready to read %d bytes\n", c);
    771 
    772 	if (c < 1) {		/* Session closed */
    773 		warnp("Error in connection with %s\n", inet_ntoa(p->ldp_id));
    774 		ldp_peer_holddown(p);
    775 		return;
    776 	}
    777 	if (c > MAX_PDU_SIZE) {
    778 		debugp("Incoming PDU size exceeds MAX_PDU_SIZE !\n");
    779 		return;
    780 	}
    781 	if (c < MIN_PDU_SIZE) {
    782 		debugp("PDU too small received from peer %s\n", inet_ntoa(p->ldp_id));
    783 		return;
    784 	}
    785 	rpdu = (struct ldp_pdu *) recvspace;
    786 	/* XXX: buggy messages may crash the whole thing */
    787 	c = recv(p->socket, (void *) recvspace,
    788 		ntohs(rpdu->length) + PDU_VER_LENGTH, MSG_WAITALL);
    789 	rpdu = (struct ldp_pdu *) recvspace;
    790 
    791 	/* Check if it's somehow OK... */
    792 	if (check_recv_pdu(p, rpdu, c) != 0)
    793 		return;
    794 
    795 	debugp("Read %d bytes, PDU size: %d bytes\n", c, ntohs(rpdu->length));
    796 	wo = sizeof(struct ldp_pdu);
    797 
    798 	while (wo + TLV_TYPE_LENGTH < (uint)c) {
    799 
    800 		ttmp = (struct tlv *) (&recvspace[wo]);
    801 
    802 		if ((ntohs(ttmp->type) != LDP_KEEPALIVE) &&
    803 		    (ntohs(ttmp->type) != LDP_LABEL_MAPPING)) {
    804 			debugp("Got Type: 0x%.4X (Length: %d) from %s\n",
    805 			    ntohs(ttmp->type), ntohs(ttmp->length),
    806 			    inet_ntoa(p->ldp_id));
    807 		} else
    808 			debugp("Got Type: 0x%.4X (Length: %d) from %s\n",
    809 			    ntohs(ttmp->type), ntohs(ttmp->length),
    810 			    inet_ntoa(p->ldp_id));
    811 
    812 		/* Should we get the message ? */
    813 		if (p->state != LDP_PEER_ESTABLISHED &&
    814 		    ntohs(ttmp->type) != LDP_INITIALIZE &&
    815 		    ntohs(ttmp->type) != LDP_KEEPALIVE)
    816 			break;
    817 		/* The big switch */
    818 		switch (ntohs(ttmp->type)) {
    819 		case LDP_INITIALIZE:
    820 			itlv = (struct init_tlv *)ttmp;
    821 			/* Check size */
    822 			if (ntohs(itlv->length) <
    823 			    sizeof(struct init_tlv) - TLV_TYPE_LENGTH) {
    824 				send_notification(p, 0,
    825 				    NOTIF_BAD_PDU_LEN | NOTIF_FATAL);
    826 				ldp_peer_holddown(p);
    827 				break;
    828 			}
    829 			/* Check version */
    830 			if (ntohs(itlv->cs_version) != LDP_VERSION) {
    831 				send_notification(p, ntohl(itlv->messageid),
    832 					NOTIF_BAD_LDP_VER | NOTIF_FATAL);
    833 				ldp_peer_holddown(p);
    834 				break;
    835 			}
    836 			/* Check if we got any hello from this one */
    837 			SLIST_FOREACH(hi, &hello_info_head, infos)
    838 				if (hi->ldp_id.s_addr == rpdu->ldp_id.s_addr)
    839 					break;
    840 			if (hi == NULL) {
    841 			    send_notification(p, ntohl(itlv->messageid),
    842 				NOTIF_SESSION_REJECTED_NO_HELLO | NOTIF_FATAL);
    843 			    ldp_peer_holddown(p);
    844 			    break;
    845 			}
    846 
    847 			if (!p->master) {
    848 				keep_alive(p);
    849 				send_initialize(p);
    850 			} else {
    851 				p->state = LDP_PEER_ESTABLISHED;
    852 				p->established_t = time(NULL);
    853 				keep_alive(p);
    854 
    855 				/*
    856 				 * Recheck here ldp id because we accepted
    857 				 * connection without knowing who is it for sure
    858 				 */
    859 				p->ldp_id.s_addr = rpdu->ldp_id.s_addr;
    860 
    861 				fatalp("LDP neighbour %s is UP\n",
    862 				    inet_ntoa(p->ldp_id));
    863 				mpls_add_ldp_peer(p);
    864 				send_addresses(p);
    865 				send_all_bindings(p);
    866 			}
    867 			break;
    868 		case LDP_KEEPALIVE:
    869 			if ((p->state == LDP_PEER_CONNECTED) && (!p->master)) {
    870 				p->state = LDP_PEER_ESTABLISHED;
    871 				p->established_t = time(NULL);
    872 				fatalp("LDP neighbour %s is UP\n",
    873 				    inet_ntoa(p->ldp_id));
    874 				mpls_add_ldp_peer(p);
    875 				send_addresses(p);
    876 				send_all_bindings(p);
    877 			}
    878 			p->timeout = p->holdtime;
    879 			break;
    880 		case LDP_ADDRESS:
    881 			/* Add peer addresses */
    882 			atlv = (struct address_tlv *) ttmp;
    883 			altlv = (struct al_tlv *) (&atlv[1]);
    884 			add_ifaddresses(p, altlv);
    885 			print_bounded_addresses(p);
    886 			break;
    887 		case LDP_ADDRESS_WITHDRAW:
    888 			atlv = (struct address_tlv *) ttmp;
    889 			altlv = (struct al_tlv *) (&atlv[1]);
    890 			del_ifaddresses(p, altlv);
    891 			break;
    892 		case LDP_LABEL_MAPPING:
    893 			lmtlv = (struct label_map_tlv *) ttmp;
    894 			fectlv = (struct fec_tlv *) (&lmtlv[1]);
    895 			labeltlv = (struct label_tlv *)((unsigned char *)fectlv
    896 				+ ntohs(fectlv->length) + TLV_TYPE_LENGTH);
    897 			map_label(p, fectlv, labeltlv);
    898 			break;
    899 		case LDP_LABEL_REQUEST:
    900 			lmtlv = (struct label_map_tlv *) ttmp;
    901 			fectlv = (struct fec_tlv *) (&lmtlv[1]);
    902 			switch (request_respond(p, lmtlv, fectlv)) {
    903 			case LDP_E_BAD_FEC:
    904 				send_notification(p, ntohl(lmtlv->messageid),
    905 					NOTIF_UNKNOWN_TLV);
    906 				break;
    907 			case LDP_E_BAD_AF:
    908 				send_notification(p, ntohl(lmtlv->messageid),
    909 					NOTIF_UNSUPPORTED_AF);
    910 				break;
    911 			case LDP_E_NO_SUCH_ROUTE:
    912 				send_notification(p, ntohl(lmtlv->messageid),
    913 					NOTIF_NO_ROUTE);
    914 				break;
    915 			}
    916 			break;
    917 		case LDP_LABEL_WITHDRAW:
    918 			lmtlv = (struct label_map_tlv *) ttmp;
    919 			fectlv = (struct fec_tlv *) (&lmtlv[1]);
    920 			if (withdraw_label(p, fectlv) == LDP_E_OK) {
    921 				/* Send RELEASE */
    922 				prepare_release(ttmp);
    923 				send_tlv(p, ttmp);
    924 				}
    925 			break;
    926 		case LDP_LABEL_RELEASE:
    927 			/*
    928 			 * XXX: we need to make a timed queue...
    929 			 * For now I just assume peers are processing messages
    930 			 * correctly so I just ignore confirmations
    931 			 */
    932 			wo = -1;	/* Ignore rest of message */
    933 			break;
    934 		case LDP_LABEL_ABORT:
    935 		/* XXX: For now I pretend I can process everything
    936 		 * RFC 3036, Section 3.5.9.1
    937 		 * If an LSR receives a Label Abort Request Message after it
    938 		 * has responded to the Label Request in question with a Label
    939 		 * Mapping message or a Notification message, it ignores the
    940 		 * abort request.
    941 		 */
    942 			wo = -1;
    943 			break;
    944 		case LDP_NOTIFICATION:
    945 			nottlv = (struct notification_tlv *) ttmp;
    946 			nottlv->st_code = ntohl(nottlv->st_code);
    947 			fatalp("Got notification 0x%X from peer %s\n",
    948 			    nottlv->st_code, inet_ntoa(p->ldp_id));
    949 			if (nottlv->st_code >> 31) {
    950 				fatalp("LDP peer %s signalized %s\n",
    951 				    inet_ntoa(p->ldp_id),
    952 				    NOTIF_STR[(nottlv->st_code << 1) >> 1]);
    953 				ldp_peer_holddown(p);
    954 				wo = -1;
    955 			}
    956 			break;
    957 		case LDP_HELLO:
    958 			/* No hellos should came on tcp session */
    959 			wo = -1;
    960 			break;
    961 		default:
    962 			warnp("Unknown TLV received from %s\n",
    963 			    inet_ntoa(p->ldp_id));
    964 			debug_tlv(ttmp);
    965 			wo = -1;/* discard the rest of the message */
    966 			break;
    967 		}
    968 		if (wo < 0) {
    969 			debugp("Discarding the rest of the message\n");
    970 			break;
    971 		} else {
    972 			wo += ntohs(ttmp->length) + TLV_TYPE_LENGTH;
    973 			debugp("WORKED ON %u bytes (Left %d)\n", wo, c - wo);
    974 		}
    975 	}			/* while */
    976 
    977 }
    978 
    979 /* Sends a pdu, tlv pair to a connected peer */
    980 int
    981 send_message(struct ldp_peer * p, struct ldp_pdu * pdu, struct tlv * t)
    982 {
    983 	unsigned char   sendspace[MAX_PDU_SIZE];
    984 
    985 	/* Check if peer is connected */
    986 	switch (p->state) {
    987 	case LDP_PEER_CONNECTED:
    988 	case LDP_PEER_ESTABLISHED:
    989 		break;
    990 	default:
    991 		return -1;
    992 	}
    993 
    994 	/* Check length validity first */
    995 	if (ntohs(pdu->length) !=
    996 	    ntohs(t->length) + TLV_TYPE_LENGTH + PDU_PAYLOAD_LENGTH) {
    997 		fatalp("LDP: TLV - PDU incompability. Message discarded\n");
    998 		fatalp("LDP: TLV len %d - PDU len %d\n", ntohs(t->length),
    999 		    ntohs(pdu->length));
   1000 		return -1;
   1001 	}
   1002 	if (ntohs(t->length) + PDU_VER_LENGTH > MAX_PDU_SIZE) {
   1003 		fatalp("Message to large discarded\n");
   1004 		return -1;
   1005 	}
   1006 	/* Arrange them in a buffer and send */
   1007 	memcpy(sendspace, pdu, sizeof(struct ldp_pdu));
   1008 	memcpy(sendspace + sizeof(struct ldp_pdu), t,
   1009 	    ntohs(t->length) + TLV_TYPE_LENGTH);
   1010 
   1011 	/* Report keepalives only for DEBUG */
   1012 	if ((ntohs(t->type) != 0x201) && (ntohs(t->type) != 0x400)) {
   1013 		debugp("Sending message type 0x%.4X to %s (size: %d)\n",
   1014 		    ntohs(t->type), inet_ntoa(p->ldp_id), ntohs(t->length));
   1015 	} else
   1016 	/* downgraded from warnp to debugp for now */
   1017 		debugp("Sending message type 0x%.4X to %s (size: %d)\n",
   1018 		    ntohs(t->type), inet_ntoa(p->ldp_id), ntohs(t->length));
   1019 
   1020 	/* Send it finally */
   1021 	return send(p->socket, sendspace,
   1022 		ntohs(pdu->length) + PDU_VER_LENGTH, 0);
   1023 }
   1024 
   1025 /*
   1026  * Encapsulates TLV into a PDU and sends it to a peer
   1027  */
   1028 int
   1029 send_tlv(struct ldp_peer * p, struct tlv * t)
   1030 {
   1031 	struct ldp_pdu  pdu;
   1032 
   1033 	pdu.version = htons(LDP_VERSION);
   1034 	inet_aton(LDP_ID, &pdu.ldp_id);
   1035 	pdu.label_space = 0;
   1036 	pdu.length = htons(ntohs(t->length) + TLV_TYPE_LENGTH +
   1037 		PDU_PAYLOAD_LENGTH);
   1038 
   1039 	return send_message(p, &pdu, t);
   1040 }
   1041 
   1042 
   1043 int
   1044 send_addresses(struct ldp_peer * p)
   1045 {
   1046 	struct address_list_tlv *t;
   1047 	int             ret;
   1048 
   1049 	t = build_address_list_tlv();
   1050 
   1051 	ret = send_tlv(p, (struct tlv *) t);
   1052 	free(t);
   1053 	return ret;
   1054 
   1055 }
   1056