[mirror_frr.git] / bgpd / bgp_bmp.h

/* BMP support.
 * Copyright (C) 2018 Yasuhiro Ohara
 * Copyright (C) 2019 David Lamparter for NetDEF, Inc.
 *
 * This program is free software; you can redistribute it and/or modify it
 * under the terms of the GNU General Public License as published by the Free
 * Software Foundation; either version 2 of the License, or (at your option)
 * any later version.
 *
 * This program is distributed in the hope that it will be useful, but WITHOUT
 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
 * FITNESS FOR A PARTICULAR PURPOSE.  See the GNU General Public License for
 * more details.
 *
 * You should have received a copy of the GNU General Public License along
 * with this program; see the file COPYING; if not, write to the Free Software
 * Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
 */

#ifndef _BGP_BMP_H_
#define _BGP_BMP_H_

#include "zebra.h"
#include "typesafe.h"
#include "pullwr.h"
#include "qobj.h"
#include "resolver.h"

#define BMP_VERSION_3	3

#define BMP_LENGTH_POS  1

/* BMP message types */
#define BMP_TYPE_ROUTE_MONITORING       0
#define BMP_TYPE_STATISTICS_REPORT      1
#define BMP_TYPE_PEER_DOWN_NOTIFICATION 2
#define BMP_TYPE_PEER_UP_NOTIFICATION   3
#define BMP_TYPE_INITIATION             4
#define BMP_TYPE_TERMINATION            5
#define BMP_TYPE_ROUTE_MIRRORING        6

#define BMP_READ_BUFSIZ	1024

/* bmp->state */
#define BMP_None        0
#define BMP_PeerUp      2
#define BMP_Run         3

/* This one is for BMP Route Monitoring messages, i.e. delivering updates
 * in somewhat processed (as opposed to fully raw, see mirroring below) form.
 * RFC explicitly says that we can skip old updates if we haven't sent them out
 * yet and another newer update for the same prefix arrives.
 *
 * So, at most one of these can exist for each (bgp, afi, safi, prefix, peerid)
 * tuple; if some prefix is "re-added" to the queue, the existing entry is
 * instead moved to the end of the queue.  This ensures that the queue size is
 * bounded by the BGP table size.
 *
 * bmp_qlist is the queue itself while bmp_qhash is used to efficiently check
 * whether a tuple is already on the list.  The queue is maintained per
 * bmp_target.
 *
 * refcount = number of "struct bmp *" whose queue position is before this
 * entry, i.e. number of BMP sessions where we still want to send this out.
 * Decremented on send so we know when we're done with an entry (i.e. this
 * always happens from the front of the queue.)
 */

PREDECL_DLIST(bmp_qlist)
PREDECL_HASH(bmp_qhash)

struct bmp_queue_entry {
	struct bmp_qlist_item bli;
	struct bmp_qhash_item bhi;

	struct prefix p;
	uint64_t peerid;
	afi_t afi;
	safi_t safi;

	size_t refcount;

	/* initialized only for L2VPN/EVPN (S)AFIs */
	struct prefix_rd rd;
};

/* This is for BMP Route Mirroring, which feeds fully raw BGP PDUs out to BMP
 * receivers.  So, this goes directly off packet RX/TX handling instead of
 * grabbing bits from tables.
 *
 * There is *one* queue for each "struct bgp *" where we throw everything on,
 * with a size limit.  Refcount works the same as for monitoring above.
 */

PREDECL_LIST(bmp_mirrorq)

struct bmp_mirrorq {
	struct bmp_mirrorq_item bmi;

	size_t refcount;
	uint64_t peerid;
	struct timeval tv;

	size_t len;
	uint8_t data[0];
};

enum {
	BMP_AFI_INACTIVE = 0,
	BMP_AFI_NEEDSYNC,
	BMP_AFI_SYNC,
	BMP_AFI_LIVE,
};

PREDECL_LIST(bmp_session)

struct bmp_active;
struct bmp_targets;

/* an established BMP session to a peer */
struct bmp {
	struct bmp_session_item bsi;
	struct bmp_targets *targets;
	struct bmp_active *active;

	int socket;
	char remote[SU_ADDRSTRLEN + 6];
	struct thread *t_read;

	struct pullwr *pullwr;

	int state;

	/* queue positions must remain synced with refcounts in the items.
	 * Whenever appending a queue item, we need to know the correct number
	 * of "struct bmp *" that want it, and when moving these positions
	 * ahead we need to make sure that refcount is decremented.  Also, on
	 * disconnects we need to walk the queue and drop our reference.
	 */
	struct bmp_queue_entry *queuepos;
	struct bmp_mirrorq *mirrorpos;
	bool mirror_lost;

	/* enum BMP_AFI_* */
	uint8_t afistate[AFI_MAX][SAFI_MAX];

	/* counters for the various BMP packet types */
	uint64_t cnt_update, cnt_mirror;
	/* number of times this peer wasn't fast enough in consuming the
	 * mirror queue
	 */
	uint64_t cnt_mirror_overruns;
	struct timeval t_up;

	/* synchronization / startup works by repeatedly finding the next
	 * table entry, the sync* fields note down what we sent last
	 */
	struct prefix syncpos;
	struct bgp_dest *syncrdpos;
	uint64_t syncpeerid;
	afi_t syncafi;
	safi_t syncsafi;
};

/* config & state for an active outbound connection.  When the connection
 * succeeds, "bmp" is set up.
 */

PREDECL_SORTLIST_UNIQ(bmp_actives)

#define BMP_DFLT_MINRETRY	30000
#define BMP_DFLT_MAXRETRY	720000

struct bmp_active {
	struct bmp_actives_item bai;
	struct bmp_targets *targets;
	struct bmp *bmp;

	char *hostname;
	int port;
	unsigned minretry, maxretry;

	struct resolver_query resq;

	unsigned curretry;
	unsigned addrpos, addrtotal;
	union sockunion addrs[8];
	int socket;
	const char *last_err;
	struct thread *t_timer, *t_read, *t_write;
};

/* config & state for passive / listening sockets */
PREDECL_SORTLIST_UNIQ(bmp_listeners)

struct bmp_listener {
	struct bmp_listeners_item bli;

	struct bmp_targets *targets;

	union sockunion addr;
	int port;

	struct thread *t_accept;
	int sock;
};

/* bmp_targets - plural since it may contain multiple bmp_listener &
 * bmp_active items.  If they have the same config, BMP session should be
 * put in the same targets since that's a bit more effective.
 */
PREDECL_SORTLIST_UNIQ(bmp_targets)

struct bmp_targets {
	struct bmp_targets_item bti;

	struct bmp_bgp *bmpbgp;
	struct bgp *bgp;
	char *name;

	struct bmp_listeners_head listeners;

	char *acl_name;
	char *acl6_name;
#define BMP_STAT_DEFAULT_TIMER	60000
	int stat_msec;

	/* only supporting:
	 * - IPv4 / unicast & multicast
	 * - IPv6 / unicast & multicast
	 * - L2VPN / EVPN
	 */
#define BMP_MON_PREPOLICY	(1 << 0)
#define BMP_MON_POSTPOLICY	(1 << 1)
	uint8_t afimon[AFI_MAX][SAFI_MAX];
	bool mirror;

	struct bmp_actives_head actives;

	struct thread *t_stats;
	struct bmp_session_head sessions;

	struct bmp_qhash_head updhash;
	struct bmp_qlist_head updlist;

	uint64_t cnt_accept, cnt_aclrefused;

	QOBJ_FIELDS
};
DECLARE_QOBJ_TYPE(bmp_targets)

/* per struct peer * data.  Lookup by peer->qobj_node.nid, created on demand,
 * deleted in peer_backward hook. */
PREDECL_HASH(bmp_peerh)

struct bmp_bgp_peer {
	struct bmp_peerh_item bpi;

	uint64_t peerid;
	/* struct peer *peer; */

	uint8_t *open_rx;
	size_t open_rx_len;

	uint8_t *open_tx;
	size_t open_tx_len;
};

/* per struct bgp * data */
PREDECL_HASH(bmp_bgph)

struct bmp_bgp {
	struct bmp_bgph_item bbi;

	struct bgp *bgp;
	struct bmp_targets_head targets;

	struct bmp_mirrorq_head mirrorq;
	size_t mirror_qsize, mirror_qsizemax;

	size_t mirror_qsizelimit;
};

enum {
	BMP_PEERDOWN_LOCAL_NOTIFY       = 1,
	BMP_PEERDOWN_LOCAL_FSM          = 2,
	BMP_PEERDOWN_REMOTE_NOTIFY      = 3,
	BMP_PEERDOWN_REMOTE_CLOSE       = 4,
	BMP_PEERDOWN_ENDMONITOR         = 5,
};

enum {
	BMP_STATS_PFX_REJECTED          = 0,
	BMP_STATS_PFX_DUP_ADV           = 1,
	BMP_STATS_PFX_DUP_WITHDRAW      = 2,
	BMP_STATS_UPD_LOOP_CLUSTER      = 3,
	BMP_STATS_UPD_LOOP_ASPATH       = 4,
	BMP_STATS_UPD_LOOP_ORIGINATOR   = 5,
	BMP_STATS_UPD_LOOP_CONFED       = 6,
	BMP_STATS_SIZE_ADJ_RIB_IN       = 7,
	BMP_STATS_SIZE_LOC_RIB          = 8,
	BMP_STATS_SIZE_ADJ_RIB_IN_SAFI  = 9,
	BMP_STATS_SIZE_LOC_RIB_IN_SAFI  = 10,
	BMP_STATS_UPD_7606_WITHDRAW     = 11,
	BMP_STATS_PFX_7606_WITHDRAW     = 12,
	BMP_STATS_UPD_DUP               = 13,
	BMP_STATS_FRR_NH_INVALID        = 65531,
};

DECLARE_MGROUP(BMP)

#endif /*_BGP_BMP_H_*/
Commit	Line	Data
6c29258c YO	1	/* BMP support.
6c29258c YO	2	* Copyright (C) 2018 Yasuhiro Ohara
ed18356f	3	* Copyright (C) 2019 David Lamparter for NetDEF, Inc.
6c29258c	4	*
ed18356f DL	5	* This program is free software; you can redistribute it and/or modify it
	6	* under the terms of the GNU General Public License as published by the Free
	7	* Software Foundation; either version 2 of the License, or (at your option)
	8	* any later version.
6c29258c	9	*
ed18356f DL	10	* This program is distributed in the hope that it will be useful, but WITHOUT
	11	* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
	12	* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
	13	* more details.
6c29258c YO	14	*
	15	* You should have received a copy of the GNU General Public License along
	16	* with this program; see the file COPYING; if not, write to the Free Software
	17	* Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
	18	*/
	19
	20	#ifndef _BGP_BMP_H_
	21	#define _BGP_BMP_H_
	22
ed18356f DL	23	#include "zebra.h"
	24	#include "typesafe.h"
	25	#include "pullwr.h"
	26	#include "qobj.h"
	27	#include "resolver.h"
	28
6c29258c YO	29	#define BMP_VERSION_3 3
	30
	31	#define BMP_LENGTH_POS 1
	32
	33	/* BMP message types */
	34	#define BMP_TYPE_ROUTE_MONITORING 0
	35	#define BMP_TYPE_STATISTICS_REPORT 1
	36	#define BMP_TYPE_PEER_DOWN_NOTIFICATION 2
	37	#define BMP_TYPE_PEER_UP_NOTIFICATION 3
	38	#define BMP_TYPE_INITIATION 4
	39	#define BMP_TYPE_TERMINATION 5
	40	#define BMP_TYPE_ROUTE_MIRRORING 6
	41
	42	#define BMP_READ_BUFSIZ 1024
	43
	44	/* bmp->state */
	45	#define BMP_None 0
6c29258c	46	#define BMP_PeerUp 2
ed18356f DL	47	#define BMP_Run 3
	48
	49	/* This one is for BMP Route Monitoring messages, i.e. delivering updates
	50	* in somewhat processed (as opposed to fully raw, see mirroring below) form.
	51	* RFC explicitly says that we can skip old updates if we haven't sent them out
	52	* yet and another newer update for the same prefix arrives.
	53	*
	54	* So, at most one of these can exist for each (bgp, afi, safi, prefix, peerid)
	55	* tuple; if some prefix is "re-added" to the queue, the existing entry is
	56	* instead moved to the end of the queue. This ensures that the queue size is
	57	* bounded by the BGP table size.
	58	*
	59	* bmp_qlist is the queue itself while bmp_qhash is used to efficiently check
	60	* whether a tuple is already on the list. The queue is maintained per
	61	* bmp_target.
	62	*
	63	* refcount = number of "struct bmp *" whose queue position is before this
	64	* entry, i.e. number of BMP sessions where we still want to send this out.
	65	* Decremented on send so we know when we're done with an entry (i.e. this
	66	* always happens from the front of the queue.)
	67	*/
	68
	69	PREDECL_DLIST(bmp_qlist)
	70	PREDECL_HASH(bmp_qhash)
	71
	72	struct bmp_queue_entry {
	73	struct bmp_qlist_item bli;
	74	struct bmp_qhash_item bhi;
	75
	76	struct prefix p;
	77	uint64_t peerid;
	78	afi_t afi;
	79	safi_t safi;
	80
	81	size_t refcount;
87102aa0	82
	83	/* initialized only for L2VPN/EVPN (S)AFIs */
	84	struct prefix_rd rd;
ed18356f DL	85	};
	86
	87	/* This is for BMP Route Mirroring, which feeds fully raw BGP PDUs out to BMP
	88	* receivers. So, this goes directly off packet RX/TX handling instead of
	89	* grabbing bits from tables.
	90	*
	91	* There is one queue for each "struct bgp *" where we throw everything on,
	92	* with a size limit. Refcount works the same as for monitoring above.
	93	*/
	94
	95	PREDECL_LIST(bmp_mirrorq)
	96
	97	struct bmp_mirrorq {
	98	struct bmp_mirrorq_item bmi;
	99
	100	size_t refcount;
	101	uint64_t peerid;
	102	struct timeval tv;
	103
	104	size_t len;
	105	uint8_t data[0];
	106	};
	107
	108	enum {
	109	BMP_AFI_INACTIVE = 0,
	110	BMP_AFI_NEEDSYNC,
	111	BMP_AFI_SYNC,
	112	BMP_AFI_LIVE,
	113	};
	114
	115	PREDECL_LIST(bmp_session)
	116
	117	struct bmp_active;
	118	struct bmp_targets;
	119
	120	/* an established BMP session to a peer */
	121	struct bmp {
	122	struct bmp_session_item bsi;
	123	struct bmp_targets *targets;
	124	struct bmp_active *active;
6c29258c	125
6c29258c	126	int socket;
ed18356f	127	char remote[SU_ADDRSTRLEN + 6];
6c29258c	128	struct thread *t_read;
ed18356f DL	129
ed18356f DL	130	struct pullwr *pullwr;
6c29258c YO	131
6c29258c YO	132	int state;
ed18356f DL	133
	134	/* queue positions must remain synced with refcounts in the items.
	135	* Whenever appending a queue item, we need to know the correct number
	136	* of "struct bmp *" that want it, and when moving these positions
	137	* ahead we need to make sure that refcount is decremented. Also, on
	138	* disconnects we need to walk the queue and drop our reference.
	139	*/
	140	struct bmp_queue_entry *queuepos;
	141	struct bmp_mirrorq *mirrorpos;
	142	bool mirror_lost;
	143
	144	/* enum BMP_AFI_* */
	145	uint8_t afistate[AFI_MAX][SAFI_MAX];
	146
	147	/* counters for the various BMP packet types */
	148	uint64_t cnt_update, cnt_mirror;
	149	/* number of times this peer wasn't fast enough in consuming the
	150	* mirror queue
	151	*/
	152	uint64_t cnt_mirror_overruns;
	153	struct timeval t_up;
	154
	155	/* synchronization / startup works by repeatedly finding the next
	156	* table entry, the sync* fields note down what we sent last
	157	*/
	158	struct prefix syncpos;
09319b4e	159	struct bgp_dest *syncrdpos;
ed18356f DL	160	uint64_t syncpeerid;
	161	afi_t syncafi;
	162	safi_t syncsafi;
	163	};
	164
	165	/* config & state for an active outbound connection. When the connection
	166	* succeeds, "bmp" is set up.
	167	*/
	168
	169	PREDECL_SORTLIST_UNIQ(bmp_actives)
	170
	171	#define BMP_DFLT_MINRETRY 30000
	172	#define BMP_DFLT_MAXRETRY 720000
	173
	174	struct bmp_active {
	175	struct bmp_actives_item bai;
	176	struct bmp_targets *targets;
	177	struct bmp *bmp;
	178
	179	char *hostname;
	180	int port;
	181	unsigned minretry, maxretry;
	182
	183	struct resolver_query resq;
	184
	185	unsigned curretry;
	186	unsigned addrpos, addrtotal;
	187	union sockunion addrs[8];
	188	int socket;
3286ca07	189	const char *last_err;
ed18356f DL	190	struct thread t_timer, t_read, *t_write;
	191	};
	192
	193	/* config & state for passive / listening sockets */
	194	PREDECL_SORTLIST_UNIQ(bmp_listeners)
	195
	196	struct bmp_listener {
	197	struct bmp_listeners_item bli;
	198
	199	struct bmp_targets *targets;
	200
	201	union sockunion addr;
	202	int port;
	203
	204	struct thread *t_accept;
	205	int sock;
	206	};
	207
	208	/* bmp_targets - plural since it may contain multiple bmp_listener &
	209	* bmp_active items. If they have the same config, BMP session should be
	210	* put in the same targets since that's a bit more effective.
	211	*/
	212	PREDECL_SORTLIST_UNIQ(bmp_targets)
	213
	214	struct bmp_targets {
	215	struct bmp_targets_item bti;
	216
	217	struct bmp_bgp *bmpbgp;
	218	struct bgp *bgp;
	219	char *name;
	220
	221	struct bmp_listeners_head listeners;
	222
	223	char *acl_name;
	224	char *acl6_name;
	225	#define BMP_STAT_DEFAULT_TIMER 60000
	226	int stat_msec;
	227
87102aa0	228	/* only supporting:
	229	* - IPv4 / unicast & multicast
	230	* - IPv6 / unicast & multicast
	231	* - L2VPN / EVPN
	232	*/
ed18356f DL	233	#define BMP_MON_PREPOLICY (1 << 0)
	234	#define BMP_MON_POSTPOLICY (1 << 1)
	235	uint8_t afimon[AFI_MAX][SAFI_MAX];
	236	bool mirror;
	237
	238	struct bmp_actives_head actives;
	239
	240	struct thread *t_stats;
	241	struct bmp_session_head sessions;
	242
	243	struct bmp_qhash_head updhash;
	244	struct bmp_qlist_head updlist;
	245
	246	uint64_t cnt_accept, cnt_aclrefused;
	247
	248	QOBJ_FIELDS
	249	};
	250	DECLARE_QOBJ_TYPE(bmp_targets)
	251
	252	/* per struct peer * data. Lookup by peer->qobj_node.nid, created on demand,
	253	* deleted in peer_backward hook. */
	254	PREDECL_HASH(bmp_peerh)
	255
	256	struct bmp_bgp_peer {
	257	struct bmp_peerh_item bpi;
	258
	259	uint64_t peerid;
	260	/* struct peer peer; /
	261
	262	uint8_t *open_rx;
	263	size_t open_rx_len;
	264
	265	uint8_t *open_tx;
	266	size_t open_tx_len;
	267	};
	268
	269	/* per struct bgp * data */
	270	PREDECL_HASH(bmp_bgph)
	271
	272	struct bmp_bgp {
	273	struct bmp_bgph_item bbi;
	274
	275	struct bgp *bgp;
	276	struct bmp_targets_head targets;
	277
	278	struct bmp_mirrorq_head mirrorq;
	279	size_t mirror_qsize, mirror_qsizemax;
	280
	281	size_t mirror_qsizelimit;
	282	};
	283
	284	enum {
	285	BMP_PEERDOWN_LOCAL_NOTIFY = 1,
	286	BMP_PEERDOWN_LOCAL_FSM = 2,
	287	BMP_PEERDOWN_REMOTE_NOTIFY = 3,
	288	BMP_PEERDOWN_REMOTE_CLOSE = 4,
	289	BMP_PEERDOWN_ENDMONITOR = 5,
	290	};
	291
	292	enum {
	293	BMP_STATS_PFX_REJECTED = 0,
	294	BMP_STATS_PFX_DUP_ADV = 1,
	295	BMP_STATS_PFX_DUP_WITHDRAW = 2,
	296	BMP_STATS_UPD_LOOP_CLUSTER = 3,
297	BMP_STATS_UPD_LOOP_ASPATH = 4,
298	BMP_STATS_UPD_LOOP_ORIGINATOR = 5,
299	BMP_STATS_UPD_LOOP_CONFED = 6,
300	BMP_STATS_SIZE_ADJ_RIB_IN = 7,
301	BMP_STATS_SIZE_LOC_RIB = 8,
302	BMP_STATS_SIZE_ADJ_RIB_IN_SAFI = 9,
303	BMP_STATS_SIZE_LOC_RIB_IN_SAFI = 10,
304	BMP_STATS_UPD_7606_WITHDRAW = 11,
305	BMP_STATS_PFX_7606_WITHDRAW = 12,
306	BMP_STATS_UPD_DUP = 13,
307	BMP_STATS_FRR_NH_INVALID = 65531,
6c29258c YO	308	};
6c29258c YO	309
ed18356f	310	DECLARE_MGROUP(BMP)
6c29258c	311
6c29258c	312	#endif /_BGP_BMP_H_/