[mirror_frr.git] / bgpd / bgp_bmp.h

// SPDX-License-Identifier: GPL-2.0-or-later
/* BMP support.
 * Copyright (C) 2018 Yasuhiro Ohara
 * Copyright (C) 2019 David Lamparter for NetDEF, Inc.
 */

#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;
	char *ifsrc;
	union sockunion addrsrc;

	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);

#define BMP_PEER_DOWN_NO_RELEVANT_EVENT_CODE 0x00

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
acddc0ed	1	// SPDX-License-Identifier: GPL-2.0-or-later
6c29258c YO	2	/* BMP support.
6c29258c YO	3	* Copyright (C) 2018 Yasuhiro Ohara
ed18356f	4	* Copyright (C) 2019 David Lamparter for NetDEF, Inc.
6c29258c YO	5	*/
	6
	7	#ifndef _BGP_BMP_H_
	8	#define _BGP_BMP_H_
	9
ed18356f DL	10	#include "zebra.h"
	11	#include "typesafe.h"
	12	#include "pullwr.h"
	13	#include "qobj.h"
	14	#include "resolver.h"
	15
6c29258c YO	16	#define BMP_VERSION_3 3
	17
	18	#define BMP_LENGTH_POS 1
	19
	20	/* BMP message types */
	21	#define BMP_TYPE_ROUTE_MONITORING 0
	22	#define BMP_TYPE_STATISTICS_REPORT 1
	23	#define BMP_TYPE_PEER_DOWN_NOTIFICATION 2
	24	#define BMP_TYPE_PEER_UP_NOTIFICATION 3
	25	#define BMP_TYPE_INITIATION 4
	26	#define BMP_TYPE_TERMINATION 5
	27	#define BMP_TYPE_ROUTE_MIRRORING 6
	28
	29	#define BMP_READ_BUFSIZ 1024
	30
	31	/* bmp->state */
	32	#define BMP_None 0
6c29258c	33	#define BMP_PeerUp 2
ed18356f DL	34	#define BMP_Run 3
	35
	36	/* This one is for BMP Route Monitoring messages, i.e. delivering updates
	37	* in somewhat processed (as opposed to fully raw, see mirroring below) form.
	38	* RFC explicitly says that we can skip old updates if we haven't sent them out
	39	* yet and another newer update for the same prefix arrives.
	40	*
	41	* So, at most one of these can exist for each (bgp, afi, safi, prefix, peerid)
	42	* tuple; if some prefix is "re-added" to the queue, the existing entry is
	43	* instead moved to the end of the queue. This ensures that the queue size is
	44	* bounded by the BGP table size.
	45	*
	46	* bmp_qlist is the queue itself while bmp_qhash is used to efficiently check
	47	* whether a tuple is already on the list. The queue is maintained per
	48	* bmp_target.
	49	*
	50	* refcount = number of "struct bmp *" whose queue position is before this
	51	* entry, i.e. number of BMP sessions where we still want to send this out.
	52	* Decremented on send so we know when we're done with an entry (i.e. this
	53	* always happens from the front of the queue.)
	54	*/
	55
960b9a53 DL	56	PREDECL_DLIST(bmp_qlist);
960b9a53 DL	57	PREDECL_HASH(bmp_qhash);
ed18356f DL	58
	59	struct bmp_queue_entry {
	60	struct bmp_qlist_item bli;
	61	struct bmp_qhash_item bhi;
	62
	63	struct prefix p;
	64	uint64_t peerid;
	65	afi_t afi;
	66	safi_t safi;
	67
	68	size_t refcount;
87102aa0	69
	70	/* initialized only for L2VPN/EVPN (S)AFIs */
	71	struct prefix_rd rd;
ed18356f DL	72	};
	73
	74	/* This is for BMP Route Mirroring, which feeds fully raw BGP PDUs out to BMP
	75	* receivers. So, this goes directly off packet RX/TX handling instead of
	76	* grabbing bits from tables.
	77	*
	78	* There is one queue for each "struct bgp *" where we throw everything on,
	79	* with a size limit. Refcount works the same as for monitoring above.
	80	*/
	81
960b9a53	82	PREDECL_LIST(bmp_mirrorq);
ed18356f DL	83
	84	struct bmp_mirrorq {
	85	struct bmp_mirrorq_item bmi;
	86
	87	size_t refcount;
	88	uint64_t peerid;
	89	struct timeval tv;
	90
	91	size_t len;
	92	uint8_t data[0];
	93	};
	94
	95	enum {
	96	BMP_AFI_INACTIVE = 0,
	97	BMP_AFI_NEEDSYNC,
	98	BMP_AFI_SYNC,
	99	BMP_AFI_LIVE,
	100	};
	101
960b9a53	102	PREDECL_LIST(bmp_session);
ed18356f DL	103
	104	struct bmp_active;
	105	struct bmp_targets;
	106
	107	/* an established BMP session to a peer */
	108	struct bmp {
	109	struct bmp_session_item bsi;
	110	struct bmp_targets *targets;
	111	struct bmp_active *active;
6c29258c	112
6c29258c	113	int socket;
ed18356f	114	char remote[SU_ADDRSTRLEN + 6];
6c29258c	115	struct thread *t_read;
ed18356f DL	116
ed18356f DL	117	struct pullwr *pullwr;
6c29258c YO	118
6c29258c YO	119	int state;
ed18356f DL	120
	121	/* queue positions must remain synced with refcounts in the items.
	122	* Whenever appending a queue item, we need to know the correct number
	123	* of "struct bmp *" that want it, and when moving these positions
	124	* ahead we need to make sure that refcount is decremented. Also, on
	125	* disconnects we need to walk the queue and drop our reference.
	126	*/
	127	struct bmp_queue_entry *queuepos;
	128	struct bmp_mirrorq *mirrorpos;
	129	bool mirror_lost;
	130
	131	/* enum BMP_AFI_* */
	132	uint8_t afistate[AFI_MAX][SAFI_MAX];
	133
	134	/* counters for the various BMP packet types */
	135	uint64_t cnt_update, cnt_mirror;
	136	/* number of times this peer wasn't fast enough in consuming the
	137	* mirror queue
	138	*/
	139	uint64_t cnt_mirror_overruns;
	140	struct timeval t_up;
	141
	142	/* synchronization / startup works by repeatedly finding the next
	143	* table entry, the sync* fields note down what we sent last
	144	*/
	145	struct prefix syncpos;
09319b4e	146	struct bgp_dest *syncrdpos;
ed18356f DL	147	uint64_t syncpeerid;
	148	afi_t syncafi;
	149	safi_t syncsafi;
	150	};
	151
	152	/* config & state for an active outbound connection. When the connection
	153	* succeeds, "bmp" is set up.
	154	*/
	155
960b9a53	156	PREDECL_SORTLIST_UNIQ(bmp_actives);
ed18356f DL	157
	158	#define BMP_DFLT_MINRETRY 30000
	159	#define BMP_DFLT_MAXRETRY 720000
	160
	161	struct bmp_active {
	162	struct bmp_actives_item bai;
	163	struct bmp_targets *targets;
	164	struct bmp *bmp;
	165
	166	char *hostname;
	167	int port;
	168	unsigned minretry, maxretry;
09f267ec FD	169	char *ifsrc;
09f267ec FD	170	union sockunion addrsrc;
ed18356f DL	171
	172	struct resolver_query resq;
	173
	174	unsigned curretry;
	175	unsigned addrpos, addrtotal;
	176	union sockunion addrs[8];
	177	int socket;
3286ca07	178	const char *last_err;
ed18356f DL	179	struct thread t_timer, t_read, *t_write;
	180	};
	181
	182	/* config & state for passive / listening sockets */
960b9a53	183	PREDECL_SORTLIST_UNIQ(bmp_listeners);
ed18356f DL	184
	185	struct bmp_listener {
	186	struct bmp_listeners_item bli;
	187
	188	struct bmp_targets *targets;
	189
	190	union sockunion addr;
	191	int port;
	192
	193	struct thread *t_accept;
	194	int sock;
	195	};
	196
	197	/* bmp_targets - plural since it may contain multiple bmp_listener &
	198	* bmp_active items. If they have the same config, BMP session should be
	199	* put in the same targets since that's a bit more effective.
	200	*/
960b9a53	201	PREDECL_SORTLIST_UNIQ(bmp_targets);
ed18356f DL	202
	203	struct bmp_targets {
	204	struct bmp_targets_item bti;
	205
	206	struct bmp_bgp *bmpbgp;
	207	struct bgp *bgp;
	208	char *name;
	209
	210	struct bmp_listeners_head listeners;
	211
	212	char *acl_name;
	213	char *acl6_name;
	214	#define BMP_STAT_DEFAULT_TIMER 60000
	215	int stat_msec;
	216
87102aa0	217	/* only supporting:
	218	* - IPv4 / unicast & multicast
	219	* - IPv6 / unicast & multicast
	220	* - L2VPN / EVPN
	221	*/
ed18356f DL	222	#define BMP_MON_PREPOLICY (1 << 0)
	223	#define BMP_MON_POSTPOLICY (1 << 1)
	224	uint8_t afimon[AFI_MAX][SAFI_MAX];
	225	bool mirror;
	226
	227	struct bmp_actives_head actives;
	228
	229	struct thread *t_stats;
	230	struct bmp_session_head sessions;
	231
	232	struct bmp_qhash_head updhash;
	233	struct bmp_qlist_head updlist;
	234
	235	uint64_t cnt_accept, cnt_aclrefused;
	236
96244aca	237	QOBJ_FIELDS;
ed18356f	238	};
96244aca	239	DECLARE_QOBJ_TYPE(bmp_targets);
ed18356f DL	240
	241	/* per struct peer * data. Lookup by peer->qobj_node.nid, created on demand,
	242	* deleted in peer_backward hook. */
960b9a53	243	PREDECL_HASH(bmp_peerh);
ed18356f DL	244
	245	struct bmp_bgp_peer {
	246	struct bmp_peerh_item bpi;
	247
	248	uint64_t peerid;
	249	/* struct peer peer; /
	250
	251	uint8_t *open_rx;
	252	size_t open_rx_len;
	253
	254	uint8_t *open_tx;
	255	size_t open_tx_len;
	256	};
	257
	258	/* per struct bgp * data */
960b9a53	259	PREDECL_HASH(bmp_bgph);
ed18356f	260
21e8caa2 QY	261	#define BMP_PEER_DOWN_NO_RELEVANT_EVENT_CODE 0x00
21e8caa2 QY	262
ed18356f DL	263	struct bmp_bgp {
	264	struct bmp_bgph_item bbi;
	265
	266	struct bgp *bgp;
	267	struct bmp_targets_head targets;
	268
	269	struct bmp_mirrorq_head mirrorq;
	270	size_t mirror_qsize, mirror_qsizemax;
	271
	272	size_t mirror_qsizelimit;
	273	};
	274
	275	enum {
	276	BMP_PEERDOWN_LOCAL_NOTIFY = 1,
	277	BMP_PEERDOWN_LOCAL_FSM = 2,
	278	BMP_PEERDOWN_REMOTE_NOTIFY = 3,
	279	BMP_PEERDOWN_REMOTE_CLOSE = 4,
	280	BMP_PEERDOWN_ENDMONITOR = 5,
	281	};
	282
	283	enum {
	284	BMP_STATS_PFX_REJECTED = 0,
	285	BMP_STATS_PFX_DUP_ADV = 1,
	286	BMP_STATS_PFX_DUP_WITHDRAW = 2,
	287	BMP_STATS_UPD_LOOP_CLUSTER = 3,
	288	BMP_STATS_UPD_LOOP_ASPATH = 4,
	289	BMP_STATS_UPD_LOOP_ORIGINATOR = 5,
	290	BMP_STATS_UPD_LOOP_CONFED = 6,
	291	BMP_STATS_SIZE_ADJ_RIB_IN = 7,
	292	BMP_STATS_SIZE_LOC_RIB = 8,
	293	BMP_STATS_SIZE_ADJ_RIB_IN_SAFI = 9,
	294	BMP_STATS_SIZE_LOC_RIB_IN_SAFI = 10,
	295	BMP_STATS_UPD_7606_WITHDRAW = 11,
	296	BMP_STATS_PFX_7606_WITHDRAW = 12,
	297	BMP_STATS_UPD_DUP = 13,
	298	BMP_STATS_FRR_NH_INVALID = 65531,
6c29258c YO	299	};
6c29258c YO	300
bf8d3d6a	301	DECLARE_MGROUP(BMP);
6c29258c	302
6c29258c	303	#endif /_BGP_BMP_H_/