Big overhaul of the mrt code.

Dumping of incomming bgp messages is now possible and dumping the (not yet)
filtered updates works too. Per neighbor dumps are still missing.
OK henning@

9 files changed, 759 insertions, 471 deletions

diff --git a/usr.sbin/bgpd/bgpd.c b/usr.sbin/bgpd/bgpd.c
index a60a7a8c3d8..632c1cd8c63 100644
--- a/usr.sbin/bgpd/bgpd.c
+++ b/usr.sbin/bgpd/bgpd.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: bgpd.c,v 1.56 2004/01/05 19:10:24 henning Exp $ */
+/*	$OpenBSD: bgpd.c,v 1.57 2004/01/05 22:57:59 claudio Exp $ */
 
 /*
  * Copyright (c) 2003 Henning Brauer <henning@openbsd.org>
@@ -39,11 +39,10 @@ void	sighdlr(int);
 void	usage(void);
 int	main(int, char *[]);
 int	check_child(pid_t, const char *);
-int	reconfigure(char *, struct bgpd_config *, struct mrt_config *,
+int	reconfigure(char *, struct bgpd_config *, struct mrt_head *,
 	    struct peer *);
-int	dispatch_imsg(struct imsgbuf *, int, struct mrt_config *);
+int	dispatch_imsg(struct imsgbuf *, int, struct mrt_head *);
 
-int			mrtfd = -1;
 int			rfd = -1;
 volatile sig_atomic_t	mrtdump = 0;
 volatile sig_atomic_t	quit = 0;
@@ -67,10 +66,8 @@ sighdlr(int sig)
 		reconfig = 1;
 		break;
 	case SIGALRM:
-		mrtdump = 1;
-		break;
 	case SIGUSR1:
-		mrtdump = 2;
+		mrtdump = 1;
 		break;
 	}
 }
@@ -96,13 +93,13 @@ main(int argc, char *argv[])
 {
 	struct bgpd_config	 conf;
 	struct peer		*peer_l, *p, *next;
-	struct mrt_config	 mrtconf;
-	struct mrtdump_config	*mconf, *(mrt[POLL_MAX]);
+	struct mrt_head		 mrtconf;
+	struct mrt		*(mrt[POLL_MAX]);
 	struct pollfd		 pfd[POLL_MAX];
 	pid_t			 io_pid = 0, rde_pid = 0, pid;
 	char			*conffile;
 	int			 debug = 0;
-	int			 ch, i, j, n, nfds, csock;
+	int			 ch, csock, i, j, n, nfds, timeout;
 	int			 pipe_m2s[2];
 	int			 pipe_m2r[2];
 	int			 pipe_s2r[2];
@@ -113,7 +110,6 @@ main(int argc, char *argv[])
 	log_init(1);		/* log to stderr until daemonized */
 
 	bzero(&conf, sizeof(conf));
-	bzero(&mrtconf, sizeof(mrtconf));
 	LIST_INIT(&mrtconf);
 	peer_l = NULL;
 
@@ -203,6 +199,7 @@ main(int argc, char *argv[])
 
 	imsg_init(&ibuf_se, pipe_m2s[0]);
 	imsg_init(&ibuf_rde, pipe_m2r[0]);
+	mrt_init(&ibuf_rde, &ibuf_se);
 	if ((rfd = kroute_init(!(conf.flags & BGPD_FLAG_NO_FIB_UPDATE))) == -1)
 		quit = 1;
 
@@ -223,12 +220,7 @@ main(int argc, char *argv[])
 		pfd[PFD_SOCK_ROUTE].fd = rfd;
 		pfd[PFD_SOCK_ROUTE].events = POLLIN;
 		i = PFD_MRT_START;
-		LIST_FOREACH(mconf, &mrtconf, list)
-			if (mconf->msgbuf.queued > 0) {
-				pfd[i].fd = mconf->msgbuf.sock;
-				pfd[i].events |= POLLOUT;
-				mrt[i++] = mconf;
-			}
+		i = mrt_select(&mrtconf, pfd, mrt, i, POLL_MAX, &timeout);
 
 		if ((nfds = poll(pfd, i, INFTIM)) == -1)
 			if (errno != EINTR) {
@@ -270,9 +262,8 @@ main(int argc, char *argv[])
 
 		for (j = PFD_MRT_START; j < i && nfds > 0 ; j++) {
 			if (pfd[j].revents & POLLOUT) {
-				if ((n = msgbuf_write(&mrt[i]->msgbuf)) < 0) {
-					log_err("pipe write error (MRT)");
-					quit = 1;
+				if ((n = mrt_write(mrt[i])) < 0) {
+					log_err("mrt write error");
 				}
 			}
 		}
@@ -280,8 +271,6 @@ main(int argc, char *argv[])
 		if (reconfig) {
 			logit(LOG_CRIT, "rereading config");
 			reconfigure(conffile, &conf, &mrtconf, peer_l);
-			LIST_FOREACH(mconf, &mrtconf, list)
-				mrt_state(mconf, IMSG_NONE, &ibuf_rde);
 			reconfig = 0;
 		}
 
@@ -294,10 +283,7 @@ main(int argc, char *argv[])
 		}
 
 		if (mrtdump == 1) {
-			mrt_alrm(&mrtconf, &ibuf_rde);
-			mrtdump = 0;
-		} else if (mrtdump == 2) {
-			mrt_usr1(&mrtconf, &ibuf_rde);
+			mrt_handler(&mrtconf);
 			mrtdump = 0;
 		}
 	}
@@ -342,7 +328,7 @@ check_child(pid_t pid, const char *pname)
 }
 
 int
-reconfigure(char *conffile, struct bgpd_config *conf, struct mrt_config *mrtc,
+reconfigure(char *conffile, struct bgpd_config *conf, struct mrt_head *mrtc,
     struct peer *peer_l)
 {
 	struct peer		*p, *next;
@@ -376,18 +362,12 @@ reconfigure(char *conffile, struct bgpd_config *conf, struct mrt_config *mrtc,
 	return (0);
 }
 
-/*
- * XXX currently messages are only buffered for mrt files.
- */
 int
-dispatch_imsg(struct imsgbuf *ibuf, int idx, struct mrt_config *conf)
+dispatch_imsg(struct imsgbuf *ibuf, int idx, struct mrt_head *mrtc)
 {
-	struct imsg		 imsg;
-	struct buf		*wbuf;
-	struct mrtdump_config	*m;
-	ssize_t			 len;
-	int			 n;
-	in_addr_t		 ina;
+	struct imsg	 imsg;
+	int		 n;
+	in_addr_t	 ina;
 
 	if ((n = imsg_read(ibuf)) == -1)
 		return (-1);
@@ -407,27 +387,8 @@ dispatch_imsg(struct imsgbuf *ibuf, int idx, struct mrt_config *conf)
 		switch (imsg.hdr.type) {
 		case IMSG_MRT_MSG:
 		case IMSG_MRT_END:
-			LIST_FOREACH(m, conf, list) {
-				if (m->id != imsg.hdr.peerid)
-					continue;
-				if (mrt_state(m, imsg.hdr.type, ibuf) == 0)
-					break;
-				if (m->msgbuf.sock == -1)
-					break;
-				len = imsg.hdr.len - IMSG_HEADER_SIZE;
-				wbuf = buf_open(len);
-				if (wbuf == NULL)
-					return (-1);
-				if (buf_add(wbuf, imsg.data, len) == -1) {
-					buf_free(wbuf);
-					return (-1);
-				}
-				if ((n = buf_close(&m->msgbuf, wbuf)) < 0) {
-					buf_free(wbuf);
-					return (-1);
-				}
-				break;
-			}
+			if (mrt_queue(mrtc, &imsg) == -1)
+				logit(LOG_CRIT, "mrt_queue failed.");
 			break;
 		case IMSG_KROUTE_CHANGE:
 			if (idx != PFD_PIPE_ROUTE)
diff --git a/usr.sbin/bgpd/bgpd.h b/usr.sbin/bgpd/bgpd.h
index 56583a8a33f..e72dbbd79c3 100644
--- a/usr.sbin/bgpd/bgpd.h
+++ b/usr.sbin/bgpd/bgpd.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: bgpd.h,v 1.53 2004/01/05 16:21:14 henning Exp $ */
+/*	$OpenBSD: bgpd.h,v 1.54 2004/01/05 22:57:59 claudio Exp $ */
 
 /*
  * Copyright (c) 2003 Henning Brauer <henning@openbsd.org>
@@ -115,41 +115,6 @@ struct peer_config {
 	enum reconf_action	 reconf_action;
 };
 
-#define	MRT_FILE_LEN	512
-enum mrtdump_type {
-	MRT_NONE,
-	MRT_TABLE_DUMP
-/*
- *	MRT_UPDATE_START,
- *	MRT_SESSION_START,
- *	MRT_UPDATE_STOP,
- *	MRT_SESSION_STOP,
- */
-};
-
-enum mrtdump_state {
-	MRT_STATE_OPEN,
-	MRT_STATE_RUNNING,
-	MRT_STATE_DONE,
-	MRT_STATE_CLOSE,
-	MRT_STATE_REOPEN
-};
-
-LIST_HEAD(mrt_config, mrtdump_config);
-
-struct mrtdump_config {
-	enum mrtdump_type	 type;
-	u_int32_t		 id;
-	struct msgbuf		 msgbuf;
-	char			 name[MRT_FILE_LEN];	/* base file name */
-	char			 file[MRT_FILE_LEN];	/* actual file name */
-	time_t			 ReopenTimer;
-	time_t			 ReopenTimerInterval;
-	enum mrtdump_state	 state;
-	LIST_ENTRY(mrtdump_config)
-				 list;
-};
-
 /* ipc messages */
 
 #define	IMSG_HEADER_SIZE	sizeof(struct imsg_hdr)
@@ -196,6 +161,9 @@ struct imsg {
 	void		*data;
 };
 
+/* needed for session.h parse prototype */
+LIST_HEAD(mrt_head, mrt);
+
 /* error subcode for UPDATE; needed in SE and RDE */
 enum suberr_update {
 	ERR_UPD_ATTRLIST = 1,
@@ -237,6 +205,8 @@ void		 buf_free(struct buf *);
 void		 msgbuf_init(struct msgbuf *);
 void		 msgbuf_clear(struct msgbuf *);
 int		 msgbuf_write(struct msgbuf *);
+int		 msgbuf_writebound(struct msgbuf *);
+int		 msgbuf_unbounded(struct msgbuf *msgbuf);
 
 /* log.c */
 void		 log_init(int);
@@ -258,9 +228,6 @@ int	 imsg_get(struct imsgbuf *, struct imsg *);
 int	 imsg_compose(struct imsgbuf *, int, u_int32_t, void *, u_int16_t);
 void	 imsg_free(struct imsg *);
 
-/* mrt.c */
-int	 mrt_mergeconfig(struct mrt_config *, struct mrt_config *);
-
 /* kroute.c */
 int	kroute_init(int);
 int	kroute_change(struct kroute *);
diff --git a/usr.sbin/bgpd/buffer.c b/usr.sbin/bgpd/buffer.c
index bc0223b7f50..55d898a4e4d 100644
--- a/usr.sbin/bgpd/buffer.c
+++ b/usr.sbin/bgpd/buffer.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: buffer.c,v 1.10 2003/12/26 17:13:52 henning Exp $ */
+/*	$OpenBSD: buffer.c,v 1.11 2004/01/05 22:57:59 claudio Exp $ */
 
 /*
  * Copyright (c) 2003 Henning Brauer <henning@openbsd.org>
@@ -98,7 +98,7 @@ buf_write(int sock, struct buf *buf)
 	ssize_t	n;
 
 	if ((n = write(sock, buf->buf + buf->rpos,
-	    buf->size-buf->rpos)) == -1) {
+	    buf->size - buf->rpos)) == -1) {
 		if (errno == EAGAIN)	/* cannot write immediately */
 			return (0);
 		else
@@ -170,6 +170,45 @@ msgbuf_write(struct msgbuf *msgbuf)
 	return (0);
 }
 
+int
+msgbuf_writebound(struct msgbuf *msgbuf)
+{
+	/*
+	 * possible race here
+	 * when we cannot write out data completely from a buffer,
+	 * we MUST return and NOT try to write out stuff from later buffers -
+	 * the socket might have become writeable again
+	 */
+	struct buf	*buf;
+	int		 n;
+
+	if (!msgbuf_unbounded(msgbuf))
+		return (1);
+
+	buf = TAILQ_FIRST(&msgbuf->bufs);
+	if ((n = buf_write(msgbuf->sock, buf)) < 0)
+		return (n);
+
+	if (n == 1) {	/* everything written out */
+		buf_dequeue(msgbuf, buf);
+		return (1);
+	} else
+		return (0);
+}
+
+int
+msgbuf_unbounded(struct msgbuf *msgbuf)
+{
+	struct buf	*buf;
+
+	/* return 1 if last buffer was not completely written. */
+	buf = TAILQ_FIRST(&msgbuf->bufs);
+	if (buf != NULL && buf->rpos != 0)
+		return (1);
+	else
+		return (0);
+}
+
 void
 buf_enqueue(struct msgbuf *msgbuf, struct buf *buf)
 {
diff --git a/usr.sbin/bgpd/mrt.c b/usr.sbin/bgpd/mrt.c
index 7da202df50f..480912f6f6a 100644
--- a/usr.sbin/bgpd/mrt.c
+++ b/usr.sbin/bgpd/mrt.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: mrt.c,v 1.14 2004/01/01 21:18:13 henning Exp $ */
+/*	$OpenBSD: mrt.c,v 1.15 2004/01/05 22:57:58 claudio Exp $ */
 
 /*
  * Copyright (c) 2003 Claudio Jeker <claudio@openbsd.org>
@@ -38,43 +38,51 @@
  * XXX imsg_create(), imsg_add(), imsg_close() ...
  */
 
-static int	mrt_dump_entry(struct mrt *, struct prefix *, u_int16_t,
+static int	mrt_dump_entry(struct mrt_config *, struct prefix *, u_int16_t,
 		    struct peer_config *);
-static void	mrt_dump_header(struct buf *, u_int16_t, u_int16_t, u_int32_t);
-static int	mrt_open(struct mrtdump_config *);
+static int	mrt_dump_header(struct buf *, u_int16_t, u_int16_t, u_int32_t);
+static int	mrt_open(struct mrt *);
 
 #define DUMP_BYTE(x, b)							\
 	do {								\
 		u_char		t = (b);				\
-		if (buf_add((x), &t, sizeof(t)) == -1)			\
-			fatalx("buf_add error");			\
+		if (buf_add((x), &t, sizeof(t)) == -1) {		\
+			logit(LOG_ERR, "mrt_dump1: buf_add error");	\
+			return (-1);					\
+		}							\
 	} while (0)
 
 #define DUMP_SHORT(x, s)						\
 	do {								\
 		u_int16_t	t;					\
 		t = htons((s));						\
-		if (buf_add((x), &t, sizeof(t)) == -1)			\
-			fatalx("buf_add error");			\
+		if (buf_add((x), &t, sizeof(t)) == -1) {		\
+			logit(LOG_ERR, "mrt_dump2: buf_add error");	\
+			return (-1);					\
+		}							\
 	} while (0)
 
 #define DUMP_LONG(x, l)							\
 	do {								\
 		u_int32_t	t;					\
 		t = htonl((l));						\
-		if (buf_add((x), &t, sizeof(t)) == -1)			\
-			fatalx("buf_add error");			\
+		if (buf_add((x), &t, sizeof(t)) == -1) {		\
+			logit(LOG_ERR, "mrt_dump3: buf_add error");	\
+			return (-1);					\
+		}							\
 	} while (0)
 
 #define DUMP_NLONG(x, l)						\
 	do {								\
 		u_int32_t	t = (l);				\
-		if (buf_add((x), &t, sizeof(t)) == -1)			\
-			fatalx("buf_add error");			\
+		if (buf_add((x), &t, sizeof(t)) == -1) {		\
+			logit(LOG_ERR, "mrt_dump4: buf_add error");	\
+			return (-1);					\
+		}							\
 	} while (0)
 
 int
-mrt_dump_bgp_msg(struct mrt *mrt, void *pkg, u_int16_t pkglen, int type,
+mrt_dump_bgp_msg(struct mrt_config *mrt, void *pkg, u_int16_t pkglen, int type,
     struct peer_config *peer, struct bgpd_config *bgp)
 {
 	struct buf	*buf;
@@ -82,18 +90,26 @@ mrt_dump_bgp_msg(struct mrt *mrt, void *pkg, u_int16_t pkglen, int type,
 	int		 i, n;
 	u_int16_t	 len;
 
-	len = pkglen + MRT_BGP4MP_HEADER_SIZE + type > 0 ? MSGSIZE_HEADER : 0;
+	len = pkglen + MRT_BGP4MP_HEADER_SIZE + (type > 0 ? MSGSIZE_HEADER : 0);
 
 	hdr.len = len + IMSG_HEADER_SIZE + MRT_HEADER_SIZE;
 	hdr.type = IMSG_MRT_MSG;
 	hdr.peerid = mrt->id;
 	buf = buf_open(hdr.len);
-	if (buf == NULL)
-		fatal("mrt_dump_bgp_msg");
-	if (buf_add(buf, &hdr, sizeof(hdr)) == -1)
-		fatalx("buf_add error");
+	if (buf == NULL) {
+		logit(LOG_ERR, "mrt_dump_bgp_msg: buf_open error");
+		return (-1);
+	}
+	if (buf_add(buf, &hdr, sizeof(hdr)) == -1) {
+		logit(LOG_ERR, "mrt_dump_bgp_msg: buf_add error");
+		return (-1);
+	}
 
-	mrt_dump_header(buf, MSG_PROTOCOL_BGP4MP, BGP4MP_MESSAGE, len);
+	if (mrt_dump_header(buf, MSG_PROTOCOL_BGP4MP, BGP4MP_MESSAGE, len) ==
+	    -1) {
+		logit(LOG_ERR, "mrt_dump_bgp_msg: buf_add error");
+		return (-1);
+	}
 
 	DUMP_SHORT(buf, bgp->as);
 	DUMP_SHORT(buf, peer->remote_as);
@@ -110,17 +126,69 @@ mrt_dump_bgp_msg(struct mrt *mrt, void *pkg, u_int16_t pkglen, int type,
 		DUMP_BYTE(buf, type);
 	}
 
-	if (buf_add(buf, pkg, pkglen) == -1)
-		fatalx("buf_add error");
+	if (buf_add(buf, pkg, pkglen) == -1) {
+		logit(LOG_ERR, "mrt_dump_bgp_msg: buf_add error");
+		return (-1);
+	}
+
+	if ((n = buf_close(mrt->msgbuf, buf)) < 0) {
+		logit(LOG_ERR, "mrt_dump_bgp_msg: buf_close error");
+		return (-1);
+	}
+
+	return (n);
+}
+
+int
+mrt_dump_state(struct mrt_config *mrt, u_int16_t old_state, u_int16_t new_state,
+    struct peer_config *peer, struct bgpd_config *bgp)
+{
+	struct buf	*buf;
+	struct imsg_hdr	 hdr;
+	int		 n;
+	u_int16_t	 len;
+
+	len = 4 + MRT_BGP4MP_HEADER_SIZE;
+	hdr.len = len + IMSG_HEADER_SIZE + MRT_HEADER_SIZE;
+	hdr.type = IMSG_MRT_MSG;
+	hdr.peerid = mrt->id;
+	buf = buf_open(hdr.len);
+	if (buf == NULL) {
+		logit(LOG_ERR, "mrt_dump_bgp_msg: buf_open error");
+		return (-1);
+	}
+	if (buf_add(buf, &hdr, sizeof(hdr)) == -1) {
+		logit(LOG_ERR, "mrt_dump_bgp_msg: buf_add error");
+		return (-1);
+	}
+
+	if (mrt_dump_header(buf, MSG_PROTOCOL_BGP4MP, BGP4MP_STATE_CHANGE,
+	    len) == -1) {
+		logit(LOG_ERR, "mrt_dump_bgp_msg: buf_add error");
+		return (-1);
+	}
+
+	DUMP_SHORT(buf, bgp->as);
+	DUMP_SHORT(buf, peer->remote_as);
+	DUMP_SHORT(buf, /* ifindex */ 0);
+	DUMP_SHORT(buf, 4);
+	DUMP_NLONG(buf, peer->local_addr.sin_addr.s_addr);
+	DUMP_NLONG(buf, peer->remote_addr.sin_addr.s_addr);
+
+	DUMP_SHORT(buf, old_state);
+	DUMP_SHORT(buf, new_state);
 
-	if ((n = buf_close(mrt->msgbuf, buf)) < 0)
-		fatalx("buf_close error");
+	if ((n = buf_close(mrt->msgbuf, buf)) < 0) {
+		logit(LOG_ERR, "mrt_dump_bgp_msg: buf_close error");
+		return (-1);
+	}
 
 	return (n);
+
 }
 
 static int
-mrt_dump_entry(struct mrt *mrt, struct prefix *p, u_int16_t snum,
+mrt_dump_entry(struct mrt_config *mrt, struct prefix *p, u_int16_t snum,
     struct peer_config *peer)
 {
 	struct buf	*buf;
@@ -136,31 +204,44 @@ mrt_dump_entry(struct mrt *mrt, struct prefix *p, u_int16_t snum,
 	hdr.type = IMSG_MRT_MSG;
 	hdr.peerid = mrt->id;
 	buf = buf_open(hdr.len);
-	if (buf == NULL)
-		fatal("mrt_dump_entry");
-	if (buf_add(buf, &hdr, sizeof(hdr)) == -1)
-		fatalx("buf_add error");
+	if (buf == NULL) {
+		logit(LOG_ERR, "mrt_dump_entry: buf_open error");
+		return (-1);
+	}
+	if (buf_add(buf, &hdr, sizeof(hdr)) == -1) {
+		logit(LOG_ERR, "mrt_dump_entry: buf_add error");
+		return (-1);
+	}
 
-	mrt_dump_header(buf, MSG_TABLE_DUMP, AFI_IPv4, len);
+	if (mrt_dump_header(buf, MSG_TABLE_DUMP, AFI_IPv4, len) == -1) {
+		logit(LOG_ERR, "mrt_dump_bgp_msg: buf_add error");
+		return (-1);
+	}
 
 	DUMP_SHORT(buf, 0);
 	DUMP_SHORT(buf, snum);
 	DUMP_NLONG(buf, p->prefix->prefix.s_addr);
 	DUMP_BYTE(buf, p->prefix->prefixlen);
-	DUMP_BYTE(buf, 1);
+	DUMP_BYTE(buf, 1);		/* state */
 	DUMP_LONG(buf, p->lastchange);	/* originated */
 	DUMP_NLONG(buf, peer->remote_addr.sin_addr.s_addr);
 	DUMP_SHORT(buf, peer->remote_as);
 	DUMP_SHORT(buf, attr_len);
 
-	if ((bptr = buf_reserve(buf, attr_len)) == NULL)
-		fatalx("buf_reserve error");
+	if ((bptr = buf_reserve(buf, attr_len)) == NULL) {
+		logit(LOG_ERR, "mrt_dump_entry: buf_reserve error");
+		return (-1);
+	}
 
-	if (attr_dump(bptr, attr_len, &p->aspath->flags) == -1)
-		fatalx("attr_dump error");
+	if (attr_dump(bptr, attr_len, &p->aspath->flags) == -1) {
+		logit(LOG_ERR, "mrt_dump_entry: attr_dump error");
+		return (-1);
+	}
 
-	if ((n = buf_close(mrt->msgbuf, buf)) < 0)
-		fatalx("buf_close error");
+	if ((n = buf_close(mrt->msgbuf, buf)) < 0) {
+		logit(LOG_ERR, "mrt_dump_entry: buf_close error");
+		return (-1);
+	}
 
 	return (n);
 }
@@ -176,8 +257,8 @@ mrt_clear_seq(void)
 void
 mrt_dump_upcall(struct pt_entry *pt, void *ptr)
 {
-	struct mrt	*mrtbuf = ptr;
-	struct prefix	*p;
+	struct mrt_config	*mrtbuf = ptr;
+	struct prefix		*p;
 
 	/*
 	 * dump all prefixes even the inactive ones. That is the way zebra
@@ -188,226 +269,357 @@ mrt_dump_upcall(struct pt_entry *pt, void *ptr)
 		mrt_dump_entry(mrtbuf, p, sequencenum++, &p->peer->conf);
 }
 
-static void
+static int
 mrt_dump_header(struct buf *buf, u_int16_t type, u_int16_t subtype,
     u_int32_t len)
 {
-	struct mrt_header	mrt;
 	time_t			now;
 
 	now = time(NULL);
 
-	mrt.timestamp = htonl(now);
-	mrt.type = ntohs(type);
-	mrt.subtype = ntohs(subtype);
-	mrt.length = htonl(len);
+	DUMP_LONG(buf, now);
+	DUMP_SHORT(buf, type);
+	DUMP_SHORT(buf, subtype);
+	DUMP_LONG(buf, len);
+
+	return (0);
+}
+
+static struct imsgbuf	*mrt_imsgbuf[2];
 
-	if (buf_add(buf, &mrt, sizeof(mrt)) == -1)
-		fatalx("buf_add error");
+void
+mrt_init(struct imsgbuf *rde, struct imsgbuf *se)
+{
+	mrt_imsgbuf[0] = rde;
+	mrt_imsgbuf[1] = se;
 }
 
 static int
-mrt_open(struct mrtdump_config *mconf)
+mrt_open(struct mrt *mrt)
 {
 	time_t	now;
 
 	now = time(NULL);
-	if (strftime(mconf->file, sizeof(mconf->file), mconf->name,
+	if (strftime(mrt->file, sizeof(mrt->file), mrt->name,
 		    localtime(&now)) == 0) {
-		logit(LOG_CRIT, "mrt_open strftime failed");
-		mconf->msgbuf.sock = -1;
-		return -1;
+		logit(LOG_ERR, "mrt_open: strftime conversion failed");
+		mrt->msgbuf.sock = -1;
+		return (0);
 	}
 
-	mconf->msgbuf.sock = open(mconf->file,
+	mrt->msgbuf.sock = open(mrt->file,
 	    O_WRONLY|O_NONBLOCK|O_CREAT|O_TRUNC, 0644);
-	if (mconf->msgbuf.sock == -1)
-		logit(LOG_CRIT, "mrt_open %s: %s",
-		    mconf->file, strerror(errno));
+	if (mrt->msgbuf.sock == -1) {
+		logit(LOG_ERR, "mrt_open %s: %s",
+		    mrt->file, strerror(errno));
+		return (0);
+	}
+	return (1);
+}
 
-	return mconf->msgbuf.sock;
+static int
+mrt_close(struct mrt *mrt)
+{
+	/*
+	 * close the mrt filedescriptor but first ensure that the last
+	 * mrt message was written correctly. If not mrt_write needs to do
+	 * that the next time called.
+	 * To ensure this we need to fiddle around with internal msgbuf stuff.
+	 */
+	if (msgbuf_unbounded(&mrt->msgbuf))
+		return (0);
+
+	if (mrt->msgbuf.sock == -1) {
+		close(mrt->msgbuf.sock);
+		mrt->msgbuf.sock = -1;
+	}
+
+	return (1);
+}
+
+void
+mrt_abort(struct mrt *mrt)
+{
+	/*
+	 * something failed horribly. Stop all dumping and go back to start
+	 * position. Retry after MRT_MIN_RETRY or ReopenTimerInterval. Which-
+	 * ever is bigger.
+	 */
+	msgbuf_clear(&mrt->msgbuf);
+	mrt_close(mrt);
+	mrt->state = MRT_STATE_STOPPED;
+
+	if (MRT_MIN_RETRY > mrt->ReopenTimerInterval)
+		mrt->ReopenTimer = MRT_MIN_RETRY + time(NULL);
+	else
+		mrt->ReopenTimer = mrt->ReopenTimerInterval + time(NULL);
 }
 
 int
-mrt_state(struct mrtdump_config *m, enum imsg_type type,
-    struct imsgbuf *buf)
+mrt_queue(struct mrt_head *mrtc, struct imsg *imsg)
 {
-	switch (m->state) {
-	case MRT_STATE_DONE:
-		/* no dump expected */
-		return (0);
-	case MRT_STATE_CLOSE:
-		switch (type) {
-		case IMSG_NONE:
-			if (m->type == MRT_TABLE_DUMP)
-				imsg_compose(buf, IMSG_MRT_END, m->id, NULL, 0);
+	struct buf	*wbuf;
+	struct mrt	*m;
+	ssize_t		 len;
+	int		 n;
+
+	if (imsg->hdr.type != IMSG_MRT_MSG && imsg->hdr.type != IMSG_MRT_END)
+		return (-1);
+
+	LIST_FOREACH(m, mrtc, list) {
+		if (m->conf.id != imsg->hdr.peerid)
+			continue;
+		if (m->state != MRT_STATE_RUNNING &&
+		    m->state != MRT_STATE_REOPEN)
 			return (0);
-		case IMSG_MRT_END:
-			/* dump no longer valid */
-			close(m->msgbuf.sock);
-			LIST_REMOVE(m, list);
-			free(m);
+
+		if (imsg->hdr.type == IMSG_MRT_END) {
+			if (mrt_close(m) == 0) {
+				m->state = MRT_STATE_CLOSE;
+			} else {
+				msgbuf_clear(&m->msgbuf);
+				m->state = MRT_STATE_STOPPED;
+			}
 			return (0);
-		default:
-			break;
 		}
-		break;
-	case MRT_STATE_REOPEN:
-		switch (type) {
-		case IMSG_NONE:
-			if (m->type == MRT_TABLE_DUMP)
-				imsg_compose(buf, IMSG_MRT_END, m->id, NULL, 0);
-			return (0);
-		case IMSG_MRT_END:
-			if (m->msgbuf.sock != -1)
-				close(m->msgbuf.sock);
-			m->state = MRT_STATE_OPEN;
-			if (m->type == MRT_TABLE_DUMP)
-				imsg_compose(buf, IMSG_MRT_REQ, m->id, NULL, 0);
-			return (0);
-		default:
-			break;
+
+		len = imsg->hdr.len - IMSG_HEADER_SIZE;
+		wbuf = buf_open(len);
+		if (wbuf == NULL)
+			return (-1);
+		if (buf_add(wbuf, imsg->data, len) == -1) {
+			buf_free(wbuf);
+			return (-1);
 		}
-		break;
-	case MRT_STATE_OPEN:
-		switch (type) {
-		case IMSG_NONE:
-			if (m->type == MRT_TABLE_DUMP)
-				imsg_compose(buf, IMSG_MRT_REQ, m->id, NULL, 0);
-			return (0);
-		case IMSG_MRT_MSG:
-			mrt_open(m);
-			m->state = MRT_STATE_RUNNING;
-			break;
-		default:
-			return (0);
+		if ((n = buf_close(&m->msgbuf, wbuf)) < 0) {
+			buf_free(wbuf);
+			return (-1);
 		}
+		return (n);
+	}
+	return (0);
+}
+
+int
+mrt_write(struct mrt *mrt)
+{
+	int	r;
+
+	if (mrt->state == MRT_STATE_REOPEN ||
+	    mrt->state == MRT_STATE_CLOSE ||
+	    mrt->state == MRT_STATE_REMOVE)
+		r = msgbuf_writebound(&mrt->msgbuf);
+	else
+		r = msgbuf_write(&mrt->msgbuf);
+
+	switch (r) {
+	case 1:
+		/* only msgbuf_writebound returns 1 */
 		break;
-	case MRT_STATE_RUNNING:
-		if (type == IMSG_MRT_END) {
-			if (m->msgbuf.sock != -1)
-				close(m->msgbuf.sock);
-			m->msgbuf.sock = -1;
-			m->state = MRT_STATE_DONE;
+	case 0:
+		return (0);
+	case -1:
+		logit(LOG_ERR, "mrt_write: msgbuf_write: %s",
+		    strerror(errno));
+		mrt_abort(mrt);
+		return (0);
+	case -2:
+		logit(LOG_ERR, "mrt_write: msgbuf_write: %s",
+		    "connection closed");
+		mrt_abort(mrt);
+		return (0);
+	default:
+		fatalx("mrt_write: unexpected retval from msgbuf_write");
+	}
+
+	if (mrt_close(mrt) != 1) {
+		logit(LOG_ERR, "mrt_write: mrt_close failed");
+		return (0);
+	}
+
+	switch (mrt->state) {
+	case MRT_STATE_REMOVE:
+		/*
+		 * Remove request: free all left buffers and
+		 * remove the descriptor.
+		 */
+		msgbuf_clear(&mrt->msgbuf);
+		LIST_REMOVE(mrt, list);
+		free(mrt);
+		return (0);
+	case MRT_STATE_CLOSE:
+		/* Close request: free all left buffers */
+		msgbuf_clear(&mrt->msgbuf);
+		mrt->state = MRT_STATE_STOPPED;
+		return (0);
+	case MRT_STATE_REOPEN:
+		if (mrt_open(mrt) == 0) {
+			mrt_abort(mrt);
 			return (0);
+		} else {
+			if (mrt->ReopenTimerInterval != 0)
+				mrt->ReopenTimer = time(NULL) +
+				    mrt->ReopenTimerInterval;
+			mrt->state = MRT_STATE_RUNNING;
 		}
 		break;
+	default:
+		break;
 	}
 	return (1);
 }
 
 int
-mrt_usr1(struct mrt_config *mconf, struct imsgbuf *buf)
+mrt_select(struct mrt_head *mc, struct pollfd *pfd, struct mrt **mrt,
+    int start, int size, int *timeout)
 {
-	struct mrtdump_config	*m;
-	time_t			 now;
-	int			 interval = INT_MAX;
+	struct mrt	*m, *xm;
+	time_t		 now;
+	int		 t;
 
 	now = time(NULL);
-	LIST_FOREACH(m, mconf, list) {
-		if (m->type == MRT_TABLE_DUMP) {
-			switch (m->state) {
-			case MRT_STATE_REOPEN:
-			case MRT_STATE_CLOSE:
+	for (m = LIST_FIRST(mc); m != LIST_END(mc); m = xm) {
+		xm = LIST_NEXT(m, list);
+		if (m->state == MRT_STATE_TOREMOVE) {
+			imsg_compose(m->ibuf, IMSG_MRT_END, 0,
+					&m->conf, sizeof(m->conf));
+			if (mrt_close(m) == 0) {
+				m->state = MRT_STATE_REMOVE;
+				m->ReopenTimer = 0;
+			} else {
+				msgbuf_clear(&m->msgbuf);
+				LIST_REMOVE(m, list);
+				free(m);
+				continue;
+			}
+		}
+		if (m->state == MRT_STATE_OPEN) {
+			switch (m->conf.type) {
+			case MRT_TABLE_DUMP:
+			case MRT_FILTERED_IN:
+				m->ibuf = mrt_imsgbuf[0];
 				break;
-			case MRT_STATE_DONE:
-				m->state = MRT_STATE_OPEN;
-				imsg_compose(buf, IMSG_MRT_REQ, m->id, NULL, 0);
+			case MRT_ALL_IN:
+				m->ibuf = mrt_imsgbuf[1];
 				break;
 			default:
-				m->state = MRT_STATE_REOPEN;
-				imsg_compose(buf, IMSG_MRT_END, m->id, NULL, 0);
-				break;
+				continue;
+			}
+			if (mrt_open(m) == 0) {
+				mrt_abort(m);
+				t = m->ReopenTimer - now;
+				if (*timeout > t)
+					*timeout = t;
+			}
+			m->state = MRT_STATE_RUNNING;
+			imsg_compose(m->ibuf, IMSG_MRT_REQ, 0,
+					&m->conf, sizeof(m->conf));
+		}
+		if (m->ReopenTimer != 0) {
+			t = m->ReopenTimer - now;
+			if (t <= 0 && (m->state == MRT_STATE_RUNNING ||
+			    m->state == MRT_STATE_STOPPED)) {
+				if (m->state == MRT_STATE_RUNNING) {
+					/* reopen file */
+					if (mrt_close(m) == 0) {
+						m->state = MRT_STATE_REOPEN;
+						continue;
+					}
+				}
+				if (mrt_open(m) == 0) {
+					mrt_abort(m);
+					t = m->ReopenTimer - now;
+					if (*timeout > t)
+						*timeout = t;
+				}
+				m->state = MRT_STATE_RUNNING;
+				if (m->ReopenTimerInterval != 0) {
+					m->ReopenTimer = now +
+					    m->ReopenTimerInterval;
+					if (*timeout > m->ReopenTimerInterval)
+						*timeout = t;
+				}
+			}
+		}
+		if (m->msgbuf.queued > 0) {
+			if (m->msgbuf.sock == -1 ||
+			    m->state == MRT_STATE_STOPPED) {
+				logit(LOG_ERR, "mrt_select: orphaned buffer");
+				mrt_abort(m);
+				continue;
+			}
+			if (start > size) {
+				pfd[start].fd = m->msgbuf.sock;
+				pfd[start].events |= POLLOUT;
+				mrt[start++] = m;
 			}
-
-			m->ReopenTimer = now + m->ReopenTimerInterval;
 		}
-		if (m->ReopenTimer - now < interval)
-			interval = m->ReopenTimer - now;
 	}
-
-	if (interval != INT_MAX)
-		alarm(interval);
-
-	return (0);
+	return (start);
 }
 
 int
-mrt_alrm(struct mrt_config *mconf, struct imsgbuf *buf)
+mrt_handler(struct mrt_head *mrt)
 {
-	struct mrtdump_config	*m;
-	time_t			 now;
-	int			 interval = INT_MAX;
+	struct mrt	*m;
+	time_t		 now;
 
 	now = time(NULL);
-	LIST_FOREACH(m, mconf, list) {
-		if (m->ReopenTimerInterval == 0)
-			continue;
-		if (m->ReopenTimer <= now) {
-			switch (m->state) {
-			case MRT_STATE_REOPEN:
-			case MRT_STATE_CLOSE:
-				break;
-			case MRT_STATE_DONE:
-				m->state = MRT_STATE_OPEN;
-				if (m->type == MRT_TABLE_DUMP)
-					imsg_compose(buf, IMSG_MRT_REQ, m->id,
-					    NULL, 0);
-				break;
-			default:
-				m->state = MRT_STATE_REOPEN;
-				if (m->type == MRT_TABLE_DUMP)
-					imsg_compose(buf, IMSG_MRT_END, m->id,
-					    NULL, 0);
-				break;
+	LIST_FOREACH(m, mrt, list) {
+		if (m->conf.type == MRT_TABLE_DUMP) {
+			if (m->state == MRT_STATE_STOPPED) {
+				if (mrt_open(m) == 0) {
+					mrt_abort(m);
+					break;
+				}
+				imsg_compose(mrt_imsgbuf[0], IMSG_MRT_REQ, 0,
+						&m->conf, sizeof(m->conf));
+				m->state = MRT_STATE_RUNNING;
 			}
-
-			m->ReopenTimer = now + m->ReopenTimerInterval;
 		}
-		if (m->ReopenTimer - now < interval)
-			interval = m->ReopenTimer - now;
+		if (m->state == MRT_STATE_RUNNING)
+			m->state = MRT_STATE_REOPEN;
+		if (m->ReopenTimerInterval != 0)
+			m->ReopenTimer = now + m->ReopenTimerInterval;
 	}
-
-	if (interval != INT_MAX)
-		alarm(interval);
-
 	return (0);
 }
 
 static u_int32_t	 max_id = 1;
 
-static struct mrtdump_config *
-getconf(struct mrt_config *c, struct mrtdump_config *m)
+static struct mrt *
+getconf(struct mrt_head *c, struct mrt *m)
 {
-	struct mrtdump_config	*t;
-	LIST_FOREACH(t, c, list)
-		if (t->type == m->type)
+	struct mrt	*t;
+
+	LIST_FOREACH(t, c, list) {
+		if (t->conf.type != m->conf.type)
+			continue;
+		if (t->conf.type == MRT_TABLE_DUMP ||
+		    t->conf.type == MRT_ALL_IN ||
+		    t->conf.type == MRT_FILTERED_IN)
+			return t;
+		if (t->conf.peer_id == m->conf.peer_id)
 			return t;
+	}
 	return (NULL);
 }
 
 int
-mrt_mergeconfig(struct mrt_config *xconf, struct mrt_config *nconf)
+mrt_mergeconfig(struct mrt_head *xconf, struct mrt_head *nconf)
 {
-	struct mrtdump_config	*m, *xm;
-	time_t			 now;
-	int			 interval = INT_MAX;
+	struct mrt	*m, *xm;
 
-	now = time(NULL);
 	LIST_FOREACH(m, nconf, list)
 		if ((xm = getconf(xconf, m)) == NULL) {
 			/* NEW */
-			if ((xm = calloc(1, sizeof(struct mrtdump_config))) ==
+			if ((xm = calloc(1, sizeof(struct mrt))) ==
 			    NULL)
 				fatal("mrt_mergeconfig");
-			memcpy(xm, m, sizeof(struct mrtdump_config));
-			xm->id = max_id++;
-			if (xm->ReopenTimerInterval != 0) {
-				xm->ReopenTimer = now + xm->ReopenTimerInterval;
-				interval = xm->ReopenTimerInterval < interval ?
-				    xm->ReopenTimerInterval : interval;
-			}
-			xm->state=MRT_STATE_OPEN;
+			memcpy(xm, m, sizeof(struct mrt));
+			msgbuf_init(&xm->msgbuf);
+			xm->conf.id = max_id++;
+			xm->state = MRT_STATE_OPEN;
 			LIST_INSERT_HEAD(xconf, xm, list);
 		} else {
 			/* MERGE */
@@ -415,17 +627,13 @@ mrt_mergeconfig(struct mrt_config *xconf, struct mrt_config *nconf)
 			    sizeof(xm->name))
 				fatalx("mrt_mergeconfig: strlcpy");
 			xm->ReopenTimerInterval = m->ReopenTimerInterval;
-			if (xm->ReopenTimerInterval != 0) {
-				xm->ReopenTimer = now + xm->ReopenTimerInterval;
-				interval = xm->ReopenTimerInterval < interval ?
-				    xm->ReopenTimerInterval : interval;
-			}
-			xm->state=MRT_STATE_REOPEN;
+			xm->state = MRT_STATE_REOPEN;
 		}
+
 	LIST_FOREACH(xm, xconf, list)
 		if (getconf(nconf, xm) == NULL)
 			/* REMOVE */
-			xm->state = MRT_STATE_CLOSE;
+			xm->state = MRT_STATE_TOREMOVE;
 
 	/* free config */
 	for (m = LIST_FIRST(nconf); m != LIST_END(nconf); m = xm) {
@@ -433,9 +641,6 @@ mrt_mergeconfig(struct mrt_config *xconf, struct mrt_config *nconf)
 		free(m);
 	}
 
-	if (interval != INT_MAX)
-		alarm(interval);
-
 	return (0);
 }
 
diff --git a/usr.sbin/bgpd/mrt.h b/usr.sbin/bgpd/mrt.h
index a7e6381a03e..40f8e8ecace 100644
--- a/usr.sbin/bgpd/mrt.h
+++ b/usr.sbin/bgpd/mrt.h
@@ -1,7 +1,7 @@
-/*	$OpenBSD: mrt.h,v 1.6 2004/01/02 00:13:32 deraadt Exp $ */
+/*	$OpenBSD: mrt.h,v 1.7 2004/01/05 22:57:58 claudio Exp $ */
 
 /*
- * Copyright (c) 2003 Claudio Jeker <cjeker@diehard.n-r-g.com>
+ * Copyright (c) 2003 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
@@ -20,6 +20,9 @@
 
 #include "bgpd.h"
 
+/* In cases of failure wait at least MRT_MIN_RETRY. */
+#define MRT_MIN_RETRY	300
+
 /*
  * MRT binary packet format as used by zebra.
  * For more info see:
@@ -30,12 +33,17 @@
  * http://www.quagga.net/docs/docs-multi/Packet-Binary-Dump-Format.html
  */
 
-struct mrt_header {
-	u_int32_t	timestamp;
-	u_int16_t	type;
-	u_int16_t	subtype;
-	u_int32_t	length;		/* length of packet including header */
-};
+/*
+ * MRT header:
+ * +--------+--------+--------+--------+
+ * |             timestamp             |
+ * +--------+--------+--------+--------+
+ * |      type       |     subtype     |
+ * +--------+--------+--------+--------+
+ * |               length              | length of packet including header
+ * +--------+--------+--------+--------+
+ */
+#define MRT_HEADER_SIZE		12
 
 enum MRT_MSG_TYPES {
 	MSG_NULL,
@@ -54,47 +62,6 @@ enum MRT_MSG_TYPES {
 	MSG_PROTOCOL_BGP4MP=16	/* 16 zebras own packet format */
 };
 
-#define MRT_HEADER_SIZE		sizeof(struct mrt_header)
-#define MRT_DUMP_HEADER_SIZE	22	/* sizeof(struct mrt_dump_v4_header) */
-#define MRT_BGP4MP_HEADER_SIZE	\
-	sizeof(struct mrt_bgp4mp_header) + \
-	sizeof(struct mrt_bgp4mp_IPv4)
-/*
- * format for routing table dumps in mrt format.
- * Type MSG_TABLE_DUMP and subtype is address family (IPv4)
- */
-struct mrt_dump_v4_header {
-	u_int16_t	view;	/* normally 0 */
-	u_int16_t	seqnum;	/* simple counter for this dump */
-	u_int32_t	prefix;
-	u_int8_t	prefixlen;
-	u_int8_t	status;	/* default seems to be 1 */
-	u_int32_t	originated; /* seems to be time of last update */
-	u_int32_t	peer_ip;
-	u_int16_t	peer_as;
-	u_int16_t	attr_len;
-	/* bgp attributes attr_len bytes long */
-};
-
-/*
- * format for routing table dumps in mrt format.
- * Type MSG_TABLE_DUMP and subtype is address family (IPv6)
- */
-struct mrt_dump_v6_header {
-	u_int16_t	view;	/* normally 0 */
-	u_int16_t	seqnum;	/* simple counter for this dump */
-	u_int32_t	prefix[4];
-	u_int8_t	prefixlen;
-	u_int8_t	status;	/* default seems to be 1 */
-	u_int32_t	originated; /* seems to be time of last update */
-	u_int32_t	peer_ip[4];
-	u_int16_t	peer_as;
-	u_int16_t	attr_len;
-	/* bgp attributes attr_len bytes long */
-};
-
-
-
 /*
  * Main zebra dump format is in MSG_PROTOCOL_BGP4MP exceptions are table dumps
  * that are normaly saved as MSG_TABLE_DUMP.
@@ -107,79 +74,106 @@ enum MRT_BGP4MP_TYPES {
 	BGP4MP_SNAPSHOT=3
 };
 
-/* if type PROTOCOL_BGP4MP and
- * subtype BGP4MP_STATE_CHANGE or BGP4MP_MESSAGE  */
-struct mrt_bgp4mp_header {
-	u_int16_t	source_as;
-	u_int16_t	dest_as;
-	u_int16_t	if_index;
-	u_int16_t	afi;
-	/*
-	 * Next comes either a struct mrt_bgp4mp_IPv4 or a
-	 * struct mrt_bgp4mp_IPv6 dependent on afi type.
-	 *
-	 * Last but not least the payload.
-	 */
-};
-
-/* if afi = 4 */
-struct mrt_bgp4mp_IPv4 {
-	u_int32_t	source_ip;
-	u_int32_t	dest_ip;
-};
+/* size of the BGP4MP headers without payload */
+#define MRT_BGP4MP_HEADER_SIZE	16
 
-/* if afi = 6 */
-struct mrt_bgp4mp_IPv6 {
-	u_int32_t	source_ip[4];
-	u_int32_t	dest_ip[4];
-};
-
-/*
- * payload of a BGP4MP_STATE_CHANGE packet
+/* If the type is PROTOCOL_BGP4MP and the subtype is either BGP4MP_STATE_CHANGE
+ * or BGP4MP_MESSAGE the message consists of a common header plus the payload.
+ * Header format:
+ * 
+ * +--------+--------+--------+--------+
+ * |    source_as    |     dest_as     |
+ * +--------+--------+--------+--------+
+ * |    if_index     |       afi       |
+ * +--------+--------+--------+--------+
+ * |             source_ip             |
+ * +--------+--------+--------+--------+
+ * |              dest_ip              |
+ * +--------+--------+--------+--------+
+ * |      message specific payload ...
+ * +--------+--------+--------+--------+
+ *
+ * The source_ip and dest_ip are dependant of the afi type. For IPv6 source_ip
+ * and dest_ip are both 16 bytes long.
+ *
+ * Payload of a BGP4MP_STATE_CHANGE packet:
+ *
+ * +--------+--------+--------+--------+
+ * |    old_state    |    new_state    |
+ * +--------+--------+--------+--------+
+ *
+ * The state values are the same as in RFC 1771.
+ *
+ * The payload of a BGP4MP_MESSAGE is the full bgp message with header.
  */
-struct mrt_bgp4mp_state_data {
-	u_int16_t	old_state;	/* as in RFC 1771 */
-	u_int16_t	new_state;
-};
 
 /*
- * The payload of a BGP4MP_MESSAGE is the full bgp message with header.
+ * The "new" table dump format consists of messages of type PROTOCOL_BGP4MP
+ * and subtype BGP4MP_ENTRY.
+ *
+ * +--------+--------+--------+--------+
+ * |      view       |     status      |
+ * +--------+--------+--------+--------+
+ * |            originated             |
+ * +--------+--------+--------+--------+
+ * |       afi       |  safi  | nhlen  |
+ * +--------+--------+--------+--------+
+ * |              nexthop              |
+ * +--------+--------+--------+--------+
+ * |  plen  |  prefix variable  ...    |
+ * +--------+--------+--------+--------+
+ * |    attr_len     | bgp attributes
+ * +--------+--------+--------+--------+
+ *  bgp attributes, attr_len bytes long
+ * +--------+--------+--------+--------+
+ *   ...                      |
+ * +--------+--------+--------+
+ *
+ * View is normaly 0 and originated the time of last change.
+ * The status seems to be 1 by default but probably something to indicate
+ * the status of a prefix would be more useful.
+ * The format of the nexthop address is defined via the afi value. For IPv6
+ * the nexthop field is 16 bytes long.
+ * The prefix field is as in the bgp update message variable length. It is
+ * plen bits long but rounded to the next octet.
  */
 
-/* if type PROTOCOL_BGP4MP, subtype BGP4MP_ENTRY */
-struct mrt_bgp4mp_entry_header {
-	u_int16_t	view;
-	u_int16_t	status;
-	u_int32_t	originated;	/* time last change */
-	u_int16_t	afi;
-	u_int8_t	safi;
-	u_int8_t	nexthop_len;
-	/*
-	 * u_int32_t nexthop    for IPv4 or
-	 * u_int32_t nexthop[4] for IPv6
-	 * u_int8_t  prefixlen
-	 * variable prefix (prefixlen bits long rounded to the next octet)
-	 * u_int16_t attrlen
-	 * variable length bgp attributes attrlen bytes long
-	 */
-};
-
 /*
- * What is this for?
- * if type MSG_PROTOCOL_BGP and subtype MSG_BGP_SYNC OR
- * if type MSG_PROTOCOL_BGP4MP and subtype BGP4MP_SNAPSHOT
+ * Format for routing table dumps in "old" mrt format.
+ * Type MSG_TABLE_DUMP and subtype is AFI_IPv4 (1) for IPv4 and AFI_IPv6 (2)
+ * for IPv6. In the IPv6 case prefix and peer_ip are both 16 bytes long.
+ *
+ * +--------+--------+--------+--------+
+ * |      view       |      seqnum     |
+ * +--------+--------+--------+--------+
+ * |               prefix              |
+ * +--------+--------+--------+--------+
+ * |  plen  | status |    originated
+ * +--------+--------+--------+--------+
+ *      originated   |     peer_ip
+ * +--------+--------+--------+--------+
+ *       peer_ip     |     peer_as     |
+ * +--------+--------+--------+--------+
+ * |    attr_len     | bgp attributes
+ * +--------+--------+--------+--------+
+ *  bgp attributes, attr_len bytes long
+ * +--------+--------+--------+--------+
+ *   ...                      |
+ * +--------+--------+--------+
+ *
+ *
+ * View is normaly 0 and seqnum just a simple counter for this dump.
+ * The status seems to be 1 by default but probably something to indicate
+ * the status of a prefix would be more useful.
  */
-struct mrt_bgp_sync_header {
-	u_int16_t	view;
-	char		filename[2]; /* variable */
-};
+
+/* size of the dump header until attr_len */
+#define MRT_DUMP_HEADER_SIZE	22
 
 /*
  * OLD MRT message headers. These structs are here for completion but
  * will not be used to generate dumps. It seems that nobody uses those.
- */
-
-/*
+ *
  * Only for bgp messages (type 5, 9 and 10)
  * Nota bene for bgp dumps MSG_PROTOCOL_BGP4MP should be used.
  */
@@ -192,47 +186,110 @@ enum MRT_BGP_TYPES {
 	MSG_BGP_SYNC
 };
 
-/* if type MSG_PROTOCOL_BGP and subtype MSG_BGP_UPDATE */
-struct mrt_bgp_update_header {
-	u_int16_t	source_as;
-	u_int32_t	source_ip;
-	u_int16_t	dest_as;
-	u_int32_t	dest_ip;
-	/* bgp update packet */
+/* if type MSG_PROTOCOL_BGP and subtype MSG_BGP_UPDATE
+ *
+ * +--------+--------+--------+--------+
+ * |    source_as    |    source_ip
+ * +--------+--------+--------+--------+
+ *      source_ip    |    dest_as      |
+ * +--------+--------+--------+--------+
+ * |               dest_ip             |
+ * +--------+--------+--------+--------+
+ * | bgp update packet with header et
+ * +--------+--------+--------+--------+
+ *   al. (variable length) ...
+ * +--------+--------+--------+--------+
+ *
+ * For IPv6 the type is MSG_PROTOCOL_BGP4PLUS and the subtype remains
+ * MSG_BGP_UPDATE. The sourec_ip and dest_ip are again extended to 16 bytes.
+ */
+
+/*
+ * For subtype MSG_BGP_STATECHANGE (for all BGP types or just for the
+ * MSG_PROTOCOL_BGP4PLUS case? Unclear.)
+ *
+ * +--------+--------+--------+--------+
+ * |    source_as    |    source_ip
+ * +--------+--------+--------+--------+
+ *      source_ip    |    old_state    |
+ * +--------+--------+--------+--------+
+ * |    new_state    |
+ * +--------+--------+
+ *
+ * State are defined in RFC 1771.
+ */
+
+/*
+ * if type MSG_PROTOCOL_BGP and subtype MSG_BGP_SYNC OR
+ * if type MSG_PROTOCOL_BGP4MP and subtype BGP4MP_SNAPSHOT
+ * What is this for?
+ *
+ * +--------+--------+--------+--------+
+ * |      view       |    filename
+ * +--------+--------+--------+--------+
+ *    filename variable length zero
+ * +--------+--------+--------+--------+
+ *    terminated ... |   0    |
+ * +--------+--------+--------+
+ */
+
+#define	MRT_FILE_LEN	512
+enum mrt_type {
+	MRT_NONE,
+	MRT_TABLE_DUMP,
+	MRT_FILTERED_IN,
+	MRT_ALL_IN,
+	MRT_SESSION_IN,
+	MRT_SESSION_OUT
 };
 
-/* if type MSG_PROTOCOL_BGP4PLUS and subtype MSG_BGP_UPDATE */
-struct mrt_bgp_update_plus_header {
-	u_int16_t	source_as;
-	u_int32_t	source_ip[4];
-	u_int16_t	dest_as;
-	u_int32_t	dest_ip[4];
-	/* bgp update packet */
+enum mrt_state {
+	MRT_STATE_STOPPED,
+	MRT_STATE_RUNNING,
+	MRT_STATE_OPEN,
+	MRT_STATE_CLOSE,
+	MRT_STATE_REOPEN,
+	MRT_STATE_TOREMOVE,
+	MRT_STATE_REMOVE
 };
 
-/* for subtype MSG_BGP_STATECHANGE (for all BGP types ???) */
-struct mrt_bgp_state_header {
-	u_int16_t	source_as;
-	u_int32_t	source_ip[4];
-	u_int16_t	old_state;	/* as in RFC 1771 ??? */
-	u_int16_t	new_state;
+struct mrt_config {
+	enum mrt_type		 type;
+	u_int32_t		 id;
+	u_int32_t		 peer_id;
+	struct msgbuf		*msgbuf;
 };
 
-/* pseudo predeclarations */
-struct prefix;
-struct pt_entry;
 struct mrt {
-	struct msgbuf	*msgbuf;
-	u_int32_t	 id;
+	struct mrt_config	 conf;
+	time_t			 ReopenTimer;
+	time_t			 ReopenTimerInterval;
+	enum mrt_state		 state;
+	struct msgbuf		 msgbuf;
+	struct imsgbuf		*ibuf;
+	char			 name[MRT_FILE_LEN];	/* base file name */
+	char			 file[MRT_FILE_LEN];	/* actual file name */
+	LIST_ENTRY(mrt)		 list;
 };
 
+
+struct prefix;
+struct pt_entry;
+
 /* prototypes */
-int	mrt_dump_bgp_msg(struct mrt *, void *, u_int16_t, int,
-    struct peer_config *, struct bgpd_config *);
+int	mrt_dump_bgp_msg(struct mrt_config *, void *, u_int16_t, int,
+	    struct peer_config *, struct bgpd_config *);
+int	mrt_dump_state(struct mrt_config *, u_int16_t, u_int16_t,
+	    struct peer_config *, struct bgpd_config *);
 void	mrt_clear_seq(void);
 void	mrt_dump_upcall(struct pt_entry *, void *);
-int	mrt_state(struct mrtdump_config *, enum imsg_type, struct imsgbuf *);
-int	mrt_alrm(struct mrt_config *, struct imsgbuf *);
-int	mrt_usr1(struct mrt_config *, struct imsgbuf *);
+void	mrt_init(struct imsgbuf *, struct imsgbuf *);
+void	mrt_abort(struct mrt *);
+int	mrt_queue(struct mrt_head *, struct imsg *);
+int	mrt_write(struct mrt *);
+int	mrt_select(struct mrt_head *, struct pollfd *, struct mrt **,
+	    int, int, int *);
+int	mrt_handler(struct mrt_head *);
+int	mrt_mergeconfig(struct mrt_head *, struct mrt_head *);
 
 #endif
diff --git a/usr.sbin/bgpd/parse.y b/usr.sbin/bgpd/parse.y
index a0afb14a334..d09fe67d5f9 100644
--- a/usr.sbin/bgpd/parse.y
+++ b/usr.sbin/bgpd/parse.y
@@ -1,4 +1,4 @@
-/*	$OpenBSD: parse.y,v 1.29 2004/01/05 19:14:41 henning Exp $ */
+/*	$OpenBSD: parse.y,v 1.30 2004/01/05 22:57:58 claudio Exp $ */
 
 /*
  * Copyright (c) 2002, 2003 Henning Brauer <henning@openbsd.org>
@@ -33,10 +33,11 @@
 #include <string.h>
 
 #include "bgpd.h"
+#include "mrt.h"
 #include "session.h"
 
 static struct bgpd_config	*conf;
-static struct mrt_config	*mrtconf;
+static struct mrt_head		*mrtconf;
 static struct peer		*peer_l;
 static struct peer		*curpeer;
 static struct peer		*curgroup;
@@ -57,7 +58,7 @@ int	 yylex(void);
 
 struct peer	*new_peer(void);
 struct peer	*new_group(void);
-int		 add_mrtconfig(enum mrtdump_type, char *, time_t);
+int		 add_mrtconfig(enum mrt_type, char *, time_t);
 
 TAILQ_HEAD(symhead, sym)	 symhead = TAILQ_HEAD_INITIALIZER(symhead);
 struct sym {
@@ -88,10 +89,10 @@ typedef struct {
 %token	GROUP NEIGHBOR
 %token	REMOTEAS DESCR LOCALADDR MULTIHOP PASSIVE
 %token	ERROR
-%token	MRTDUMP
+%token	DUMP MSG IN TABLE
 %token	LOG UPDATES
 %token	<v.string>	STRING
-%type	<v.number>	number yesno
+%type	<v.number>	number optnumber yesno
 %type	<v.string>	string
 %type	<v.addr>	address
 %%
@@ -177,21 +178,26 @@ conf_main	: AS number		{
 		| LOG UPDATES		{
 			conf->log |= BGPD_LOG_UPDATES;
 		}
-		/*
-		 *  XXX this is bad.
-		 *  a) number should be optional
-		 *  b) there are multiple dump types
-		 */
-		| MRTDUMP STRING STRING number	{
-			if (strcmp($2, "table") == 0) {
-				if (add_mrtconfig(MRT_TABLE_DUMP, $3, $4) == -1)
-					YYERROR;
-			} else {
-				yyerror("unknown mrtdump type %s", $2);
+		| DUMP MSG STRING IN STRING optnumber	{
+			int action;
+
+			if (!strcmp($3, "all"))
+				action = MRT_ALL_IN;
+			else if (!strcmp($3, "filtered"))
+				action = MRT_FILTERED_IN;
+			else {
+				yyerror("unknown mrt msg dump type");
 				YYERROR;
 			}
+			if (add_mrtconfig(action, $5, $6) == -1)
+				YYERROR;
+		}
+		| DUMP TABLE STRING optnumber		{
+			if (add_mrtconfig(MRT_TABLE_DUMP, $3, $4) == -1)
+				YYERROR;
 		}
-		;
+
+;
 
 address		: STRING		{
 			int	n;
@@ -211,6 +217,10 @@ optnl		: '\n' optnl
 		|
 		;
 
+optnumber	: /* empty */		{ $$ = 0; }
+		| number
+		;
+
 neighbor	: NEIGHBOR address optnl '{' optnl {
 			curpeer = new_peer();
 			curpeer->conf.remote_addr.sin_len =
@@ -326,14 +336,16 @@ lookup(char *s)
 	static const struct keywords keywords[] = {
 		{ "AS",			AS},
 		{ "descr",		DESCR},
+		{ "dump",		DUMP},
 		{ "fib-update",		FIBUPDATE},
 		{ "group",		GROUP},
 		{ "holdtime",		HOLDTIME},
+		{ "in",			IN},
 		{ "listen",		LISTEN},
 		{ "local-address",	LOCALADDR},
 		{ "log",		LOG},
 		{ "min",		YMIN},
-		{ "mrtdump",		MRTDUMP},
+		{ "msg",		MSG},
 		{ "multihop",		MULTIHOP},
 		{ "neighbor",		NEIGHBOR},
 		{ "on",			ON},
@@ -341,6 +353,7 @@ lookup(char *s)
 		{ "remote-as",		REMOTEAS},
 		{ "router-id",		ROUTERID},
 		{ "set",		SET},
+		{ "table",		TABLE},
 		{ "updates",		UPDATES},
 	};
 	const struct keywords	*p;
@@ -549,13 +562,13 @@ top:
 
 int
 parse_config(char *filename, struct bgpd_config *xconf,
-    struct mrt_config *xmconf, struct peer **xpeers)
+    struct mrt_head *xmconf, struct peer **xpeers)
 {
 	struct sym	*sym, *next;
 
 	if ((conf = calloc(1, sizeof(struct bgpd_config))) == NULL)
 		fatal(NULL);
-	if ((mrtconf = calloc(1, sizeof(struct mrt_config))) == NULL)
+	if ((mrtconf = calloc(1, sizeof(struct mrt_head))) == NULL)
 		fatal(NULL);
 	LIST_INIT(mrtconf);
 
@@ -734,21 +747,21 @@ new_group(void)
 }
 
 int
-add_mrtconfig(enum mrtdump_type type, char *name, time_t timeout)
+add_mrtconfig(enum mrt_type type, char *name, time_t timeout)
 {
-	struct mrtdump_config	*m, *n;
+	struct mrt	*m, *n;
 
 	LIST_FOREACH(m, mrtconf, list) {
-		if (m->type == type) {
+		if (m->conf.type == type) {
 			yyerror("only one mrtdump per type allowed.");
 			return (-1);
 		}
 	}
 
-	if ((n = calloc(1, sizeof(struct mrtdump_config))) == NULL)
+	if ((n = calloc(1, sizeof(struct mrt))) == NULL)
 		fatal("add_mrtconfig");
 
-	n->type = MRT_TABLE_DUMP;
+	n->conf.type = type;
 	n->msgbuf.sock = -1;
 	if (strlcpy(n->name, name, sizeof(n->name)) >= sizeof(n->name)) {
 		yyerror("filename \"%s\" too long: max %u",
diff --git a/usr.sbin/bgpd/rde.c b/usr.sbin/bgpd/rde.c
index ecff99ecce7..c4d6d99cdd0 100644
--- a/usr.sbin/bgpd/rde.c
+++ b/usr.sbin/bgpd/rde.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: rde.c,v 1.45 2004/01/04 20:47:34 henning Exp $ */
+/*	$OpenBSD: rde.c,v 1.46 2004/01/05 22:57:58 claudio Exp $ */
 
 /*
  * Copyright (c) 2003 Henning Brauer <henning@openbsd.org>
@@ -61,6 +61,9 @@ struct rde_peer_head	 peerlist;
 struct imsgbuf		 ibuf_se;
 struct imsgbuf		 ibuf_main;
 
+int			 mrt_flagfilter = 0;
+struct mrt_config	 mrt_filter;
+
 void
 rde_sighdlr(int sig)
 {
@@ -211,7 +214,7 @@ void
 rde_dispatch_imsg_parent(struct imsgbuf *ibuf)
 {
 	struct imsg		 imsg;
-	struct mrt		 mrtdump;
+	struct mrt_config	 mrt;
 	struct peer_config	*pconf;
 	struct rde_peer		*p, *np;
 	int			 n;
@@ -270,15 +273,26 @@ rde_dispatch_imsg_parent(struct imsgbuf *ibuf)
 			nexthop_update(imsg.data);
 			break;
 		case IMSG_MRT_REQ:
-			mrtdump.id = imsg.hdr.peerid;
-			mrtdump.msgbuf = &ibuf_main.w;
-			pt_dump(mrt_dump_upcall, &mrtdump);
-			/* FALLTHROUGH */
+			memcpy(&mrt, imsg.data, sizeof(mrt));
+			mrt.msgbuf = &ibuf_main.w;
+			if (mrt.type == MRT_TABLE_DUMP) {
+				mrt_clear_seq();
+				pt_dump(mrt_dump_upcall, &mrt);
+				if (imsg_compose(&ibuf_main, IMSG_MRT_END,
+				    mrt.id, NULL, 0) == -1)
+					fatalx("imsg_compose error");
+			} else if (mrt.type == MRT_FILTERED_IN) {
+				mrt_flagfilter = 1;
+				memcpy(&mrt_filter, &mrt, sizeof(mrt_filter));
+			}
+			break;
 		case IMSG_MRT_END:
+			memcpy(&mrt, imsg.data, sizeof(mrt));
 			/* ignore end message because a dump is atomic */
-			if (imsg_compose(&ibuf_main, IMSG_MRT_END,
-			    imsg.hdr.peerid, NULL, 0) == -1)
-				fatalx("imsg_compose error");
+			if (mrt.type == MRT_FILTERED_IN) {
+				mrt_flagfilter = 0;
+				bzero(&mrt_filter, sizeof(mrt_filter));
+			}
 			break;
 		default:
 			break;
@@ -308,6 +322,11 @@ rde_update_dispatch(struct imsg *imsg)
 	if (peer->state != PEER_UP)
 		return (-1);	/* peer is not yet up, cannot happen */
 
+	if (mrt_flagfilter == 1)
+		mrt_dump_bgp_msg(&mrt_filter, imsg->data,
+		    imsg->hdr.len - IMSG_HEADER_SIZE, UPDATE,
+		    &peer->conf, conf);
+
 	p = imsg->data;
 
 	memcpy(&len, p, 2);
diff --git a/usr.sbin/bgpd/session.c b/usr.sbin/bgpd/session.c
index 99c5a63da65..c08c4ca18b3 100644
--- a/usr.sbin/bgpd/session.c
+++ b/usr.sbin/bgpd/session.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: session.c,v 1.64 2004/01/04 23:08:09 henning Exp $ */
+/*	$OpenBSD: session.c,v 1.65 2004/01/05 22:57:58 claudio Exp $ */
 
 /*
  * Copyright (c) 2003 Henning Brauer <henning@openbsd.org>
@@ -41,6 +41,7 @@
 #include <unistd.h>
 
 #include "bgpd.h"
+#include "mrt.h"
 #include "session.h"
 
 #define	PFD_LISTEN	0
@@ -87,6 +88,9 @@ int			 csock = -1;
 struct imsgbuf		 ibuf_rde;
 struct imsgbuf		 ibuf_main;
 
+int			 mrt_flagall = 0;
+struct mrt_config	 mrt_allin;
+
 void
 session_sighdlr(int sig)
 {
@@ -656,6 +660,10 @@ change_state(struct peer *peer, enum session_state state,
 	}
 
 	log_statechange(peer, state, event);
+	if (mrt_flagall == 1)
+		mrt_dump_state(&mrt_allin, peer->state, state,
+		    &peer->conf, conf);
+	/* XXX mrt dump per peer */
 	peer->state = state;
 }
 
@@ -1143,6 +1151,9 @@ parse_header(struct peer *peer, u_char *data, u_int16_t *len, u_int8_t *type)
 		    type, 1);
 		return (-1);
 	}
+	if (mrt_flagall == 1)
+		mrt_dump_bgp_msg(&mrt_allin, data, *len, 0, &peer->conf, conf);
+	/* XXX mrt dump per peer */
 	return (0);
 }
 
@@ -1282,6 +1293,7 @@ void
 session_dispatch_imsg(struct imsgbuf *ibuf, int idx)
 {
 	struct imsg		 imsg;
+	struct mrt_config	 mrt;
 	struct peer_config	*pconf;
 	struct peer		*p, *next;
 	enum reconf_action	 reconf;
@@ -1383,6 +1395,21 @@ session_dispatch_imsg(struct imsgbuf *ibuf, int idx)
 			pending_reconf = 0;
 			logit(LOG_INFO, "SE reconfigured");
 			break;
+		case IMSG_MRT_REQ:
+			memcpy(&mrt, imsg.data, sizeof(mrt));
+			mrt.msgbuf = &ibuf_main.w;
+			if (mrt.type == MRT_ALL_IN) {
+				mrt_flagall = 1;
+				memcpy(&mrt_allin, &mrt, sizeof(mrt_allin));
+			}
+			break;
+		case IMSG_MRT_END:
+			memcpy(&mrt, imsg.data, sizeof(mrt));
+			if (mrt.type == MRT_ALL_IN) {
+				mrt_flagall = 0;
+				bzero(&mrt_allin, sizeof(mrt_allin));
+			}
+			break;
 		default:
 			break;
 		}
@@ -1425,4 +1452,4 @@ int
 imsg_compose_parent(int type, u_int32_t peerid, void *data, u_int16_t datalen)
 {
 	return (imsg_compose(&ibuf_main, type, peerid, data, datalen));
-}
\ No newline at end of file
+}
diff --git a/usr.sbin/bgpd/session.h b/usr.sbin/bgpd/session.h
index aea136d2830..7b8e99cddc2 100644
--- a/usr.sbin/bgpd/session.h
+++ b/usr.sbin/bgpd/session.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: session.h,v 1.10 2004/01/04 20:07:30 henning Exp $ */
+/*	$OpenBSD: session.h,v 1.11 2004/01/05 22:57:58 claudio Exp $ */
 
 /*
  * Copyright (c) 2003 Henning Brauer <henning@openbsd.org>
@@ -166,7 +166,7 @@ void		 log_notification(const struct peer *, u_int8_t, u_int8_t,
 void		 log_conn_attempt(const struct peer *, struct in_addr);
 
 /* parse.y */
-int	 parse_config(char *, struct bgpd_config *, struct mrt_config *,
+int	 parse_config(char *, struct bgpd_config *, struct mrt_head *,
 	    struct peer **);
 
 /* config.c */