path: root/usr.sbin/bgpd/rde_update.c

/*	$OpenBSD: rde_update.c,v 1.32 2004/11/11 13:06:45 claudio Exp $ */

/*
 * Copyright (c) 2004 Claudio Jeker <claudio@openbsd.org>
 *
 * Permission to use, copy, modify, and distribute this software for any
 * purpose with or without fee is hereby granted, provided that the above
 * copyright notice and this permission notice appear in all copies.
 *
 * THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES
 * WITH REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF
 * MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR
 * ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES
 * WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN
 * ACTION OF CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF
 * OR IN CONNECTION WITH THE USE OR PERFORMANCE OF THIS SOFTWARE.
 */
#include <sys/types.h>
#include <sys/queue.h>

#include <stdlib.h>
#include <string.h>

#include "bgpd.h"
#include "rde.h"

int	up_generate_attr(struct rde_peer *, struct update_attr *,
	    struct rde_aspath *);

/* update stuff. */
struct update_prefix {
	struct bgpd_addr		 prefix;
	int				 prefixlen;
	struct uplist_prefix		*prefix_h;
	TAILQ_ENTRY(update_prefix)	 prefix_l;
	RB_ENTRY(update_prefix)		 entry;
};

struct update_attr {
	u_int32_t			 attr_hash;
	u_char				*attr;
	u_int16_t			 attr_len;
	struct uplist_prefix		 prefix_h;
	TAILQ_ENTRY(update_attr)	 attr_l;
	RB_ENTRY(update_attr)		 entry;
};

void	up_clear(struct uplist_attr *, struct uplist_prefix *);
int	up_prefix_cmp(struct update_prefix *, struct update_prefix *);
int	up_attr_cmp(struct update_attr *, struct update_attr *);
int	up_add(struct rde_peer *, struct update_prefix *, struct update_attr *);

RB_PROTOTYPE(uptree_prefix, update_prefix, entry, up_prefix_cmp)
RB_GENERATE(uptree_prefix, update_prefix, entry, up_prefix_cmp)

RB_PROTOTYPE(uptree_attr, update_attr, entry, up_attr_cmp)
RB_GENERATE(uptree_attr, update_attr, entry, up_attr_cmp)

void
up_init(struct rde_peer *peer)
{
	TAILQ_INIT(&peer->updates);
	TAILQ_INIT(&peer->withdraws);
	TAILQ_INIT(&peer->updates6);
	TAILQ_INIT(&peer->withdraws6);
	RB_INIT(&peer->up_prefix);
	RB_INIT(&peer->up_attrs);
	peer->up_pcnt = 0;
	peer->up_acnt = 0;
	peer->up_nlricnt = 0;
	peer->up_wcnt = 0;
}

void
up_clear(struct uplist_attr *updates, struct uplist_prefix *withdraws)
{
	struct update_attr	*ua;
	struct update_prefix	*up;

	while ((ua = TAILQ_FIRST(updates)) != NULL) {
		TAILQ_REMOVE(updates, ua, attr_l);
		while ((up = TAILQ_FIRST(&ua->prefix_h)) != NULL) {
			TAILQ_REMOVE(&ua->prefix_h, up, prefix_l);
			free(up);
		}
		free(ua->attr);
		free(ua);
	}

	while ((up = TAILQ_FIRST(withdraws)) != NULL) {
		TAILQ_REMOVE(withdraws, up, prefix_l);
		free(up);
	}
}

void
up_down(struct rde_peer *peer)
{
	up_clear(&peer->updates, &peer->withdraws);
	up_clear(&peer->updates6, &peer->withdraws6);

	RB_INIT(&peer->up_prefix);
	RB_INIT(&peer->up_attrs);

	peer->up_pcnt = 0;
	peer->up_acnt = 0;
	peer->up_nlricnt = 0;
	peer->up_wcnt = 0;
}

int
up_prefix_cmp(struct update_prefix *a, struct update_prefix *b)
{
	int	i;

	if (a->prefix.af < b->prefix.af)
		return (-1);
	if (a->prefix.af > b->prefix.af)
		return (1);

	switch (a->prefix.af) {
	case AF_INET:
		if (ntohl(a->prefix.v4.s_addr) < ntohl(b->prefix.v4.s_addr))
			return (-1);
		if (ntohl(a->prefix.v4.s_addr) > ntohl(b->prefix.v4.s_addr))
			return (1);
		break;
	case AF_INET6:
		i = memcmp(&a->prefix.v6, &b->prefix.v6,
		    sizeof(struct in6_addr));
		if (i > 0)
			return (1);
		if (i < 0)
			return (-1);
		break;
	default:
		fatalx("pt_prefix_cmp: unknown af");
	}
	if (a->prefixlen < b->prefixlen)
		return (-1);
	if (a->prefixlen > b->prefixlen)
		return (1);
	return (0);
}

int
up_attr_cmp(struct update_attr *a, struct update_attr *b)
{
	if (a->attr_hash < b->attr_hash)
		return (-1);
	if (a->attr_hash > b->attr_hash)
		return (1);
	if (a->attr_len < b->attr_len)
		return (-1);
	if (a->attr_len > b->attr_len)
		return (1);
	return (memcmp(a->attr, b->attr, a->attr_len));
}

int
up_add(struct rde_peer *peer, struct update_prefix *p, struct update_attr *a)
{
	struct update_attr	*na = NULL;
	struct update_prefix	*np;
	struct uplist_attr	*upl = NULL;
	struct uplist_prefix	*wdl = NULL;

	switch (p->prefix.af) {
	case AF_INET:
		upl = &peer->updates;
		wdl = &peer->withdraws;
		break;
	case AF_INET6:
		upl = &peer->updates6;
		wdl = &peer->withdraws6;
		break;
	default:
		fatalx("up_add: unknown AF");
	}

	/* 1. search for attr */
	if (a != NULL && (na = RB_FIND(uptree_attr, &peer->up_attrs, a)) ==
	    NULL) {
		/* 1.1 if not found -> add */
		TAILQ_INIT(&a->prefix_h);
		if (RB_INSERT(uptree_attr, &peer->up_attrs, a) != NULL) {
			log_warnx("uptree_attr insert failed");
			/* cleanup */
			free(a->attr);
			free(a);
			free(p);
			return (-1);
		}
		TAILQ_INSERT_TAIL(upl, a, attr_l);
		peer->up_acnt++;
	} else {
		/* 1.2 if found -> use that, free a */
		if (a != NULL) {
			free(a->attr);
			free(a);
			a = na;
			/* move to end of update queue */
			TAILQ_REMOVE(upl, a, attr_l);
			TAILQ_INSERT_TAIL(upl, a, attr_l);
		}
	}

	/* 2. search for prefix */
	if ((np = RB_FIND(uptree_prefix, &peer->up_prefix, p)) == NULL) {
		/* 2.1 if not found -> add */
		if (RB_INSERT(uptree_prefix, &peer->up_prefix, p) != NULL) {
			log_warnx("uptree_prefix insert failed");
			/*
			 * cleanup. But do not free a because it is already
			 * linked or NULL. up_dump_attrnlri() will remove and
			 * free the empty attribute later.
			 */
			free(p);
			return (-1);
		}
		peer->up_pcnt++;
	} else {
		/* 2.2 if found -> use that and free p */
		TAILQ_REMOVE(np->prefix_h, np, prefix_l);
		free(p);
		p = np;
		if (p->prefix_h == wdl)
			peer->up_wcnt--;
		else
			peer->up_nlricnt--;
	}
	/* 3. link prefix to attr */
	if (a == NULL) {
		TAILQ_INSERT_TAIL(wdl, p, prefix_l);
		p->prefix_h = wdl;
		peer->up_wcnt++;
	} else {
		TAILQ_INSERT_TAIL(&a->prefix_h, p, prefix_l);
		p->prefix_h = &a->prefix_h;
		peer->up_nlricnt++;
	}
	return (0);
}

void
up_generate_updates(struct rde_peer *peer,
    struct prefix *new, struct prefix *old)
{
	struct update_attr		*a;
	struct update_prefix		*p;
	struct attr			*attr;
	struct rde_aspath		*fasp;
	struct bgpd_addr		 addr;

	if (peer->state != PEER_UP)
		return;

	if (new == NULL || (new->aspath->nexthop != NULL &&
	    new->aspath->nexthop->state != NEXTHOP_REACH)) {
		if (old == NULL)
			/* new prefix got filtered and no old prefix avail */
			return;

		if (peer == old->peer)
			/* Do not send routes back to sender */
			return;

		if (peer->conf.ebgp &&
		    !aspath_loopfree(old->aspath->aspath,
		    peer->conf.remote_as))
			/*
			 * Do not send routes back to sender which would
			 * cause a aspath loop.
			 */
			return;

		if (old->peer->conf.ebgp == 0 && peer->conf.ebgp == 0) {
			/*
			 * route reflector redistribution rules:
			 * 1. if announce is set		-> announce
			 * 2. old non-client, new non-client	-> no 
			 * 3. old client, new non-client	-> yes
			 * 4. old non-client, new client	-> yes
			 * 5. old client, new client		-> yes
			 */
			if (old->peer->conf.reflector_client == 0 &&
			    peer->conf.reflector_client == 0 &&
			    (old->aspath->flags & F_PREFIX_ANNOUNCED) == 0)
				/* Do not redistribute updates to ibgp peers */
				return;
		}

		/* announce type handling */
		switch (peer->conf.announce_type) {
		case ANNOUNCE_UNDEF:
		case ANNOUNCE_NONE:
		case ANNOUNCE_DEFAULT_ROUTE:
			return;
		case ANNOUNCE_ALL:
			break;
		case ANNOUNCE_SELF:
			/*
			 * pass only prefix that have a aspath count
			 * of zero this is equal to the ^$ regex.
			 */
			if (old->aspath->aspath->ascnt != 0)
				return;
			break;
		}

		/* well known communities */
		if (rde_filter_community(old->aspath,
		    COMMUNITY_WELLKNOWN, COMMUNITY_NO_ADVERTISE))
			return;
		if (peer->conf.ebgp && rde_filter_community(old->aspath,
		    COMMUNITY_WELLKNOWN, COMMUNITY_NO_EXPORT))
			return;
		if (peer->conf.ebgp && rde_filter_community(old->aspath,
		    COMMUNITY_WELLKNOWN, COMMUNITY_NO_EXPSUBCONFED))
			return;

		/*
		 * Don't send messages back to originator 
		 * this is not specified in the RFC but seems logical.
		 */
		if ((attr = attr_optget(old->aspath,
		    ATTR_ORIGINATOR_ID)) != NULL) {
			if (memcmp(attr->data, &peer->remote_bgpid,
			    sizeof(peer->remote_bgpid)) == 0)
				/* would cause loop don't send */
				return;
		}

		/* copy attributes for output filter */
		fasp = path_copy(old->aspath);

		/* default override not needed here as this is a withdraw */

		pt_getaddr(old->prefix, &addr);
		if (rde_filter(peer, fasp, &addr,
		    old->prefix->prefixlen, DIR_OUT) == ACTION_DENY) {
			path_put(fasp);
			return;
		}
		path_put(fasp);

		/* withdraw prefix */
		p = calloc(1, sizeof(struct update_prefix));
		if (p == NULL)
			fatal("up_generate_updates");

		p->prefix = addr;
		p->prefixlen = old->prefix->prefixlen;
		if (up_add(peer, p, NULL) == -1)
			log_warnx("queuing withdraw failed.");
	} else {
		if (peer == new->peer) {
			/* Do not send routes back to sender */
			up_generate_updates(peer, NULL, old);
			return;
		}

		if (peer->conf.ebgp &&
		    !aspath_loopfree(new->aspath->aspath,
		    peer->conf.remote_as)) {
			/*
			 * Do not send routes back to sender which would
			 * cause a aspath loop.
			 */
			up_generate_updates(peer, NULL, old);
			return;
		}

		if (new->peer->conf.ebgp == 0 && peer->conf.ebgp == 0) {
			/*
			 * route reflector redistribution rules:
			 * 1. if announce is set		-> announce
			 * 2. old non-client, new non-client	-> no 
			 * 3. old client, new non-client	-> yes
			 * 4. old non-client, new client	-> yes
			 * 5. old client, new client		-> yes
			 */
			if (new->peer->conf.reflector_client == 0 &&
			    peer->conf.reflector_client == 0 &&
			    (new->aspath->flags & F_PREFIX_ANNOUNCED) == 0) {
				/* Do not redistribute updates to ibgp peers */
				up_generate_updates(peer, NULL, old);
				return;
			}
		}

		/* announce type handling */
		switch (peer->conf.announce_type) {
		case ANNOUNCE_UNDEF:
		case ANNOUNCE_NONE:
		case ANNOUNCE_DEFAULT_ROUTE:
			/*
			 * no need to withdraw old prefix as this will be
			 * filtered out to.
			 */
			return;
		case ANNOUNCE_ALL:
			break;
		case ANNOUNCE_SELF:
			/*
			 * pass only prefix that have a aspath count
			 * of zero this is equal to the ^$ regex.
			 */
			if (new->aspath->aspath->ascnt != 0) {
				up_generate_updates(peer, NULL, old);
				return;
			}
			break;
		}

		/* well known communities */
		if (rde_filter_community(new->aspath,
		    COMMUNITY_WELLKNOWN, COMMUNITY_NO_ADVERTISE)) {
			up_generate_updates(peer, NULL, old);
			return;
		}
		if (peer->conf.ebgp && rde_filter_community(new->aspath,
		    COMMUNITY_WELLKNOWN, COMMUNITY_NO_EXPORT)) {
			up_generate_updates(peer, NULL, old);
			return;
		}
		if (peer->conf.ebgp && rde_filter_community(new->aspath,
		    COMMUNITY_WELLKNOWN, COMMUNITY_NO_EXPSUBCONFED)) {
			up_generate_updates(peer, NULL, old);
			return;
		}

		/* copy attributes for output filter */
		fasp = path_copy(new->aspath);

		/*
		 * apply default outgoing overrides,
		 * acctually only prepend-self
		 */
		rde_apply_set(fasp, &peer->conf.attrset, new->prefix->af,
		    fasp->peer, DIR_DEFAULT_OUT);

		pt_getaddr(new->prefix, &addr);
		if (rde_filter(peer, fasp, &addr,
		    new->prefix->prefixlen, DIR_OUT) == ACTION_DENY) {
			path_put(fasp);
			up_generate_updates(peer, NULL, old);
			return;
		}

		/*
		 * Don't send messages back to originator 
		 * this is not specified in the RFC but seems logical.
		 */
		if ((attr = attr_optget(new->aspath,
		    ATTR_ORIGINATOR_ID)) != NULL) {
			if (memcmp(attr->data, &peer->remote_bgpid,
			    sizeof(peer->remote_bgpid)) == 0) {
				/* would cause loop don't send */
				path_put(fasp);
				return;
			}
		}

		/* generate update */
		p = calloc(1, sizeof(struct update_prefix));
		if (p == NULL)
			fatal("up_generate_updates");

		a = calloc(1, sizeof(struct update_attr));
		if (a == NULL)
			fatal("up_generate_updates");

		if (up_generate_attr(peer, a, fasp) == -1) {
			log_warnx("generation of bgp path attributes failed");
			free(a);
			free(p);
			return;
		}

		/*
		 * use aspath_hash as attr_hash, this may be unoptimal
		 * but currently I don't care.
		 */
		a->attr_hash = aspath_hash(fasp->aspath->data,
		    fasp->aspath->len);
		p->prefix = addr;
		p->prefixlen = new->prefix->prefixlen;

		/* no longer needed */
		path_put(fasp);

		if (up_add(peer, p, a) == -1)
			log_warnx("queuing update failed.");
	}
}

void
up_generate_default(struct rde_peer *peer, sa_family_t af)
{
	struct update_attr	*a;
	struct update_prefix	*p;
	struct rde_aspath	*asp;
	struct bgpd_addr	 addr;

	asp = path_get();
	asp->aspath = aspath_get(NULL, 0);
	asp->origin = ORIGIN_IGP;
	/* the other default values are OK, nexthop is once again NULL */

	/* XXX apply default overrides. Not yet possible */

	/* filter as usual */
	bzero(&addr, sizeof(addr));
	addr.af = af;
	if (rde_filter(peer, asp, &addr, 0, DIR_OUT) == ACTION_DENY) {
		path_put(asp);
		return;
	}

	/* generate update */
	p = calloc(1, sizeof(struct update_prefix));
	if (p == NULL)
		fatal("up_generate_default");

	a = calloc(1, sizeof(struct update_attr));
	if (a == NULL)
		fatal("up_generate_default");

	if (up_generate_attr(peer, a, asp) == -1) {
		log_warnx("generation of bgp path attributes failed");
		free(a);
		free(p);
		return;
	}

	/*
	 * use aspath_hash as attr_hash, this may be unoptimal
	 * but currently I don't care.
	 */
	a->attr_hash = aspath_hash(asp->aspath->data, asp->aspath->len);
	p->prefix = addr;
	p->prefixlen = 0; /* default route */

	/* no longer needed */
	path_put(asp);

	if (up_add(peer, p, a) == -1)
		log_warnx("queuing update failed.");

}

void
up_dump_upcall(struct pt_entry *pt, void *ptr)
{
	struct rde_peer	*peer = ptr;

	if (pt->active == NULL)
		return;
	up_generate_updates(peer, pt->active, NULL);
}

u_char	up_attr_buf[4096];

int
up_generate_attr(struct rde_peer *peer, struct update_attr *upa,
    struct rde_aspath *a)
{
	struct aspath	*path;
	struct attr	*oa;
	u_int32_t	 tmp32;
	in_addr_t	 nexthop, mask;
	int		 r;
	u_int16_t	 len = sizeof(up_attr_buf), wlen = 0;

	/* origin */
	if ((r = attr_write(up_attr_buf + wlen, len, ATTR_WELL_KNOWN,
	    ATTR_ORIGIN, &a->origin, 1)) == -1)
		return (-1);
	wlen += r; len -= r;

	/* aspath */
	path = aspath_prepend(a->aspath, rde_local_as(), peer->conf.ebgp);
	if ((r = attr_write(up_attr_buf + wlen, len, ATTR_WELL_KNOWN,
	    ATTR_ASPATH, path->data, path->len)) == -1)
		return (-1);
	aspath_put(path);
	wlen += r; len -= r;

	/* nexthop, already network byte order */
	if (peer->conf.ebgp == 0) {
		/*
		 * If directly connected use peer->local_v4_addr
		 * this is only true for announced networks.
		 */
		if (a->nexthop == NULL)
			nexthop = peer->local_v4_addr.v4.s_addr;
		else if (a->nexthop->exit_nexthop.v4.s_addr ==
		    peer->remote_addr.v4.s_addr)
			/*
			 * per rfc: if remote peer address is equal to
			 * the nexthop set the nexthop to our local address.
			 * This reduces the risk of routing loops.
			 */
			nexthop = peer->local_v4_addr.v4.s_addr;
		else
			nexthop = a->nexthop->exit_nexthop.v4.s_addr;
	} else if (peer->conf.distance == 1) {
		/* ebgp directly connected */
		if (a->nexthop != NULL &&
		    a->nexthop->flags & NEXTHOP_CONNECTED) {
			mask = 0xffffffff << (32 - a->nexthop->nexthop_netlen);
			mask = htonl(mask);
			if ((peer->remote_addr.v4.s_addr & mask) ==
			    (a->nexthop->nexthop_net.v4.s_addr & mask))
				/* nexthop and peer are in the same net */
				nexthop = a->nexthop->exit_nexthop.v4.s_addr;
			else
				nexthop = peer->local_v4_addr.v4.s_addr;
		} else
			nexthop = peer->local_v4_addr.v4.s_addr;
	} else
		/* ebgp multihop */
		/*
		 * For ebgp multihop nh->flags should never have
		 * NEXTHOP_CONNECTED set so it should be possible to unify the
		 * two ebgp cases. But this is save and RFC compliant.
		 */
		nexthop = peer->local_v4_addr.v4.s_addr;

	if ((r = attr_write(up_attr_buf + wlen, len, ATTR_WELL_KNOWN,
	    ATTR_NEXTHOP, &nexthop, 4)) == -1)
		return (-1);
	wlen += r; len -= r;

	/*
	 * The old MED from other peers MUST not be announced to others
	 * unless the MED is originating from us or the peer is a IBGP one.
	 */
	if (a->flags & F_ATTR_MED && (peer->conf.ebgp == 0 ||
	    a->flags & F_ATTR_MED_ANNOUNCE)) {
		tmp32 = htonl(a->med);
		if ((r = attr_write(up_attr_buf + wlen, len, ATTR_OPTIONAL,
		    ATTR_MED, &tmp32, 4)) == -1)
			return (-1);
		wlen += r; len -= r;
	}

	if (peer->conf.ebgp == 0) {
		/* local preference, only valid for ibgp */
		tmp32 = htonl(a->lpref);
		if ((r = attr_write(up_attr_buf + wlen, len, ATTR_WELL_KNOWN,
		    ATTR_LOCALPREF, &tmp32, 4)) == -1)
			return (-1);
		wlen += r; len -= r;
	}

	/*
	 * dump all other path attributes. Following rules apply:
	 *  1. well-known attrs: ATTR_ATOMIC_AGGREGATE and ATTR_AGGREGATOR
	 *     pass unmodified (enforce flags to correct values)
	 *  2. non-transitive attrs: don't re-announce to ebgp peers
	 *  3. transitive known attrs: announce unmodified
	 *  4. transitive unknown attrs: set partial bit and re-announce
	 */
	TAILQ_FOREACH(oa, &a->others, entry) {
		switch (oa->type) {
		case ATTR_ATOMIC_AGGREGATE:
			if ((r = attr_write(up_attr_buf + wlen, len,
			    ATTR_WELL_KNOWN, ATTR_ATOMIC_AGGREGATE,
			    NULL, 0)) == -1)
				return (-1);
			break;
		case ATTR_AGGREGATOR:
		case ATTR_COMMUNITIES:
		case ATTR_ORIGINATOR_ID:
		case ATTR_CLUSTER_LIST:
			if ((!(oa->flags & ATTR_TRANSITIVE)) &&
			    peer->conf.ebgp != 0) {
				r = 0;
				break;
			}
			if ((r = attr_write(up_attr_buf + wlen, len,
			    oa->flags, oa->type, oa->data, oa->len)) == -1)
				return (-1);
			break;
		default:
			/* unknown attribute */
			if (!(oa->flags & ATTR_TRANSITIVE)) {
				/*
				 * RFC 1771:
				 * Unrecognized non-transitive optional
				 * attributes must be quietly ignored and
				 * not passed along to other BGP peers.
				 */
				r = 0;
				break;
			}
			if ((r = attr_write(up_attr_buf + wlen, len,
			    oa->flags | ATTR_PARTIAL, oa->type,
			    oa->data, oa->len)) == -1)
				return (-1);
			break;
		}
		wlen += r; len -= r;
	}

	/* the bgp path attributes are now stored in the global buf */
	upa->attr = malloc(wlen);
	if (upa->attr == NULL)
		fatal("up_generate_attr");
	memcpy(upa->attr, up_attr_buf, wlen);
	upa->attr_len = wlen;
	return (wlen);
}

#define MIN_PREFIX_LEN	5	/* 1 byte prefix length + 4 bytes addr */
int
up_dump_prefix(u_char *buf, int len, struct uplist_prefix *prefix_head,
    struct rde_peer *peer)
{
	struct update_prefix	*upp;
	int			 r, wpos = 0;

	while ((upp = TAILQ_FIRST(prefix_head)) != NULL) {
		if ((r = prefix_write(buf + wpos, len - wpos,
		    &upp->prefix, upp->prefixlen)) == -1)
			break;
		wpos += r;
		if (RB_REMOVE(uptree_prefix, &peer->up_prefix, upp) == NULL)
			log_warnx("dequeuing update failed.");
		TAILQ_REMOVE(upp->prefix_h, upp, prefix_l);
		peer->up_pcnt--;
		if (upp->prefix_h == &peer->withdraws)
			peer->up_wcnt--;
		else
			peer->up_nlricnt--;
		free(upp);
	}
	return (wpos);
}

int
up_dump_attrnlri(u_char *buf, int len, struct rde_peer *peer)
{
	struct update_attr	*upa;
	int			 r, wpos;
	u_int16_t		 attr_len;

	/*
	 * It is possible that a queued path attribute has no nlri prefix.
	 * Ignore and remove those path attributes.
	 */
	while ((upa = TAILQ_FIRST(&peer->updates)) != NULL)
		if (TAILQ_EMPTY(&upa->prefix_h)) {
			if (RB_REMOVE(uptree_attr, &peer->up_attrs,
			    upa) == NULL)
				log_warnx("dequeuing update failed.");
			TAILQ_REMOVE(&peer->updates, upa, attr_l);
			free(upa->attr);
			free(upa);
			peer->up_acnt--;
		} else
			break;

	if (upa == NULL || upa->attr_len + MIN_PREFIX_LEN > len) {
		/*
		 * either no packet or not enough space.
		 * The length field needs to be set to zero else it would be
		 * an invalid bgp update.
		 */
		bzero(buf, 2);
		return (2);
	}

	/* first dump the 2-byte path attribute length */
	attr_len = htons(upa->attr_len);
	memcpy(buf, &attr_len, 2);
	wpos = 2;

	/* then the path attributes them self */
	memcpy(buf + wpos, upa->attr, upa->attr_len);
	wpos += upa->attr_len;

	/* last but not least dump the nlri */
	r = up_dump_prefix(buf + wpos, len - wpos, &upa->prefix_h, peer);
	wpos += r;

	/* now check if all prefixes where written */
	if (TAILQ_EMPTY(&upa->prefix_h)) {
		if (RB_REMOVE(uptree_attr, &peer->up_attrs, upa) == NULL)
			log_warnx("dequeuing update failed.");
		TAILQ_REMOVE(&peer->updates, upa, attr_l);
		free(upa->attr);
		free(upa);
		peer->up_acnt--;
	}

	return (wpos);
}