Add administrative options to LACP trunk implementation.

The trunk driver now has a new ioctl (SIOCxTRUNKOPTS), which for now only
has options for LACP:
* Mode - Active or Passive (default Active)
* Timeout - Fast or Slow (default Slow)
* System Priority - 1(high) to 65535(low) (default 32768/0x8000)
* Port Priority - 1(high) to 65535(low) (default 32768/0x8000)
* IFQ Priority - 0 to NUM_QUEUES (default 6)

At the moment, ifconfig only has options for lacpmode and lacptimeout
plumbed as those are the immediate need.

The approach taken for the options was to make them on a "trunk" vs a
"port" as what's typically seen on various NOSes (JunOS, NXOS, etc...)
as it's uncommon for a host to have one link "Passive" and the other
"Active" in a given trunk.

Just like on a NOS, when applying lacpmode or lacptimeout, the settings
are immediately applied to all existing ports in the trunk and to all
future ports brought into the trunk.

Tested by many on a plethora of NIC drivers and switches.

Ok remi@

4 files changed, 190 insertions, 42 deletions

diff --git a/sys/net/if_trunk.c b/sys/net/if_trunk.c
index 49d2dbecbbe..6887002a863 100644
--- a/sys/net/if_trunk.c
+++ b/sys/net/if_trunk.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_trunk.c,v 1.136 2018/02/19 08:59:52 mpi Exp $	*/
+/*	$OpenBSD: if_trunk.c,v 1.137 2018/08/12 23:50:31 ccardenas Exp $	*/
 
 /*
  * Copyright (c) 2005, 2006, 2007 Reyk Floeter <reyk@openbsd.org>
@@ -604,8 +604,11 @@ trunk_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
 	struct trunk_softc *tr = (struct trunk_softc *)ifp->if_softc;
 	struct trunk_reqall *ra = (struct trunk_reqall *)data;
 	struct trunk_reqport *rp = (struct trunk_reqport *)data, rpbuf;
+	struct trunk_opts *tro = (struct trunk_opts *)data;
 	struct ifreq *ifr = (struct ifreq *)data;
+	struct lacp_softc *lsc;
 	struct trunk_port *tp;
+	struct lacp_port *lp;
 	struct ifnet *tpif;
 	int i, error = 0;
 
@@ -671,6 +674,94 @@ trunk_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
 		}
 		error = EPROTONOSUPPORT;
 		break;
+	case SIOCGTRUNKOPTS:
+		/* Only LACP trunks have options atm */
+		if (tro->to_proto != TRUNK_PROTO_LACP) {
+			error = EPROTONOSUPPORT;
+			break;
+		}
+		lsc = LACP_SOFTC(tr);
+		tro->to_lacpopts.lacp_mode = lsc->lsc_mode;
+		tro->to_lacpopts.lacp_timeout = lsc->lsc_timeout;
+		tro->to_lacpopts.lacp_prio = lsc->lsc_sys_prio;
+		tro->to_lacpopts.lacp_portprio = lsc->lsc_port_prio;
+		tro->to_lacpopts.lacp_ifqprio = lsc->lsc_ifq_prio;
+		break;
+	case SIOCSTRUNKOPTS:
+		if ((error = suser(curproc)) != 0) {
+			error = EPERM;
+			break;
+		}
+		/* Only LACP trunks have options atm */
+		if (tro->to_proto != TRUNK_PROTO_LACP) {
+			error = EPROTONOSUPPORT;
+			break;
+		}
+		lsc = LACP_SOFTC(tr);
+		switch(tro->to_opts) {
+			case TRUNK_OPT_LACP_MODE:
+				/*
+				 * Ensure mode changes occur immediately
+				 * on all ports
+				 */
+				lsc->lsc_mode = tro->to_lacpopts.lacp_mode;
+				if (lsc->lsc_mode == 0) {
+					LIST_FOREACH(lp, &lsc->lsc_ports,
+					    lp_next)
+						lp->lp_state &=
+						    ~LACP_STATE_ACTIVITY;
+				} else {
+					LIST_FOREACH(lp, &lsc->lsc_ports,
+					    lp_next)
+						lp->lp_state |=
+						    LACP_STATE_ACTIVITY;
+				}
+				break;
+			case TRUNK_OPT_LACP_TIMEOUT:
+				/*
+				 * Ensure timeout changes occur immediately
+				 * on all ports
+				 */
+				lsc->lsc_timeout =
+				    tro->to_lacpopts.lacp_timeout;
+				if (lsc->lsc_timeout == 0) {
+					LIST_FOREACH(lp, &lsc->lsc_ports,
+					    lp_next)
+						lp->lp_state &=
+						    ~LACP_STATE_TIMEOUT;
+				} else {
+					LIST_FOREACH(lp, &lsc->lsc_ports,
+					    lp_next)
+						lp->lp_state |=
+						    LACP_STATE_TIMEOUT;
+				}
+				break;
+			case TRUNK_OPT_LACP_SYS_PRIO:
+				if (tro->to_lacpopts.lacp_prio == 0) {
+					error = EINVAL;	
+					break;
+				}
+				lsc->lsc_sys_prio = tro->to_lacpopts.lacp_prio;
+				break;
+			case TRUNK_OPT_LACP_PORT_PRIO:
+				if (tro->to_lacpopts.lacp_portprio == 0) {
+					error = EINVAL;	
+					break;
+				}
+				lsc->lsc_port_prio =
+				    tro->to_lacpopts.lacp_portprio;
+				break;
+			case TRUNK_OPT_LACP_IFQ_PRIO:
+				if (tro->to_lacpopts.lacp_ifqprio >
+				    IFQ_MAXPRIO) {
+					error = EINVAL;	
+					break;
+				}
+				lsc->lsc_ifq_prio =
+				    tro->to_lacpopts.lacp_ifqprio;
+				break;
+		}
+		break;
 	case SIOCGTRUNKPORT:
 		if (rp->rp_portname[0] == '\0' ||
 		    (tpif = ifunit(rp->rp_portname)) == NULL) {
diff --git a/sys/net/if_trunk.h b/sys/net/if_trunk.h
index 05d731a3307..c4408dc5e5d 100644
--- a/sys/net/if_trunk.h
+++ b/sys/net/if_trunk.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: if_trunk.h,v 1.25 2015/09/23 12:40:12 mikeb Exp $	*/
+/*	$OpenBSD: if_trunk.h,v 1.26 2018/08/12 23:50:31 ccardenas Exp $	*/
 
 /*
  * Copyright (c) 2005, 2006, 2007 Reyk Floeter <reyk@openbsd.org>
@@ -121,6 +121,36 @@ struct trunk_reqall {
 #define SIOCGTRUNK		_IOWR('i', 143, struct trunk_reqall)
 #define SIOCSTRUNK		 _IOW('i', 144, struct trunk_reqall)
 
+/* LACP administrative options */
+struct lacp_adminopts {
+	u_int8_t		lacp_mode;		/* active or passive */
+	u_int8_t		lacp_timeout;		/* fast or slow */
+	u_int16_t		lacp_prio;		/* system priority */
+	u_int16_t		lacp_portprio;		/* port priority */
+	u_int8_t		lacp_ifqprio;		/* ifq priority */
+};
+
+/* Trunk administrative options */
+struct trunk_opts {
+	char			to_ifname[IFNAMSIZ];	/* name of the trunk */
+	u_int			to_proto;		/* trunk protocol */
+	int			to_opts;		/* option bitmap */
+#define TRUNK_OPT_NONE			0x00
+#define TRUNK_OPT_LACP_MODE		0x01		/* set active bit */
+#define TRUNK_OPT_LACP_TIMEOUT		0x02		/* set timeout bit */
+#define TRUNK_OPT_LACP_SYS_PRIO		0x04		/* set sys_prio bit */
+#define TRUNK_OPT_LACP_PORT_PRIO	0x08		/* set port_prio bit */
+#define TRUNK_OPT_LACP_IFQ_PRIO		0x10		/* set ifq_prio bit */
+
+	union {
+		struct lacp_adminopts rpsc_lacp;
+	} to_psc;
+#define to_lacpopts	to_psc.rpsc_lacp
+};
+
+#define SIOCGTRUNKOPTS		_IOWR('i', 145, struct trunk_opts)
+#define SIOCSTRUNKOPTS		 _IOW('i', 146, struct trunk_opts)
+
 #ifdef _KERNEL
 /*
  * Internal kernel part
diff --git a/sys/net/trunklacp.c b/sys/net/trunklacp.c
index 52ca84f6514..8d5b7bcadbf 100644
--- a/sys/net/trunklacp.c
+++ b/sys/net/trunklacp.c
@@ -1,4 +1,4 @@
-/*	$OpenBSD: trunklacp.c,v 1.29 2017/01/24 10:08:30 krw Exp $ */
+/*	$OpenBSD: trunklacp.c,v 1.30 2018/08/12 23:50:31 ccardenas Exp $ */
 /*	$NetBSD: ieee8023ad_lacp.c,v 1.3 2005/12/11 12:24:54 christos Exp $ */
 /*	$FreeBSD:ieee8023ad_lacp.c,v 1.15 2008/03/16 19:25:30 thompsa Exp $ */
 
@@ -58,14 +58,6 @@
 #include <net/bpf.h>
 #endif
 
-/*
- * actor system priority and port priority.
- * XXX should be configurable.
- */
-#define	LACP_SYSTEM_PRIO	0x8000
-#define	LACP_PORT_PRIO		0x8000
-#define	LACP_IFQ_PRIO		6
-
 const u_int8_t ethermulticastaddr_slowprotocols[ETHER_ADDR_LEN] =
     { 0x01, 0x80, 0xc2, 0x00, 0x00, 0x02 };
 
@@ -93,6 +85,8 @@ const struct tlv_template marker_response_tlv_template[] = {
 
 typedef void (*lacp_timer_func_t)(struct lacp_port *);
 
+void		lacp_default_partner(struct lacp_softc *,
+		    struct lacp_peerinfo *);
 void		lacp_fill_actorinfo(struct lacp_port *, struct lacp_peerinfo *);
 void		lacp_fill_markerinfo(struct lacp_port *,
 		    struct lacp_markerinfo *);
@@ -197,31 +191,22 @@ void		lacp_dprintf(const struct lacp_port *, const char *, ...)
 #define LACP_DPRINTF(a) /* nothing */
 #endif
 
-/*
- * partner administration variables.
- * XXX should be configurable.
- */
-
-const struct lacp_peerinfo lacp_partner_admin = {
-	{ 0xffff },	/* lip_systemid.lsi_prio */
-	0,		/* lip_key */
-	{ 0xffff },	/* lip_portid.lpi_prio */
-#if 1
-	/* optimistic lip_state */
-	LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
-	    LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING
-#else
-	/* pessimistic lip_state */
-	0
-#endif
-};
-
 const lacp_timer_func_t lacp_timer_funcs[LACP_NTIMER] = {
 	[LACP_TIMER_CURRENT_WHILE] = lacp_sm_rx_timer,
 	[LACP_TIMER_PERIODIC] = lacp_sm_ptx_timer,
 	[LACP_TIMER_WAIT_WHILE] = lacp_sm_mux_timer,
 };
 
+void
+lacp_default_partner(struct lacp_softc *lsc, struct lacp_peerinfo *peer)
+{
+	peer->lip_systemid.lsi_prio = lsc->lsc_sys_prio;
+	peer->lip_key = 0;
+	peer->lip_portid.lpi_prio = lsc->lsc_port_prio;
+	peer->lip_state = LACP_STATE_SYNC | LACP_STATE_AGGREGATION |
+	    LACP_STATE_COLLECTING | LACP_STATE_DISTRIBUTING;
+}
+
 int
 lacp_input(struct trunk_port *tp, struct mbuf *m)
 {
@@ -351,13 +336,14 @@ bad:
 void
 lacp_fill_actorinfo(struct lacp_port *lp, struct lacp_peerinfo *info)
 {
+	struct lacp_softc *lsc = lp->lp_lsc;
 	struct trunk_port *tp = lp->lp_trunk;
 	struct trunk_softc *sc = tp->tp_trunk;
 
-	info->lip_systemid.lsi_prio = htons(LACP_SYSTEM_PRIO);
+	info->lip_systemid.lsi_prio = htons(lsc->lsc_sys_prio);
 	memcpy(&info->lip_systemid.lsi_mac,
 	    sc->tr_ac.ac_enaddr, ETHER_ADDR_LEN);
-	info->lip_portid.lpi_prio = htons(LACP_PORT_PRIO);
+	info->lip_portid.lpi_prio = htons(lsc->lsc_port_prio);
 	info->lip_portid.lpi_portno = htons(lp->lp_ifp->if_index);
 	info->lip_state = lp->lp_state;
 }
@@ -376,6 +362,7 @@ lacp_fill_markerinfo(struct lacp_port *lp, struct lacp_markerinfo *info)
 int
 lacp_xmit_lacpdu(struct lacp_port *lp)
 {
+	struct lacp_softc *lsc = lp->lp_lsc;
 	struct trunk_port *tp = lp->lp_trunk;
 	struct mbuf *m;
 	struct lacpdu *du;
@@ -385,7 +372,7 @@ lacp_xmit_lacpdu(struct lacp_port *lp)
 	if (m == NULL)
 		return (ENOMEM);
 	m->m_len = m->m_pkthdr.len = sizeof(*du);
-	m->m_pkthdr.pf.prio = LACP_IFQ_PRIO;
+	m->m_pkthdr.pf.prio = lsc->lsc_ifq_prio;
 
 	du = mtod(m, struct lacpdu *);
 	memset(du, 0, sizeof(*du));
@@ -427,6 +414,7 @@ lacp_xmit_lacpdu(struct lacp_port *lp)
 int
 lacp_xmit_marker(struct lacp_port *lp)
 {
+	struct lacp_softc *lsc = lp->lp_lsc;
 	struct trunk_port *tp = lp->lp_trunk;
 	struct mbuf *m;
 	struct markerdu *mdu;
@@ -436,7 +424,7 @@ lacp_xmit_marker(struct lacp_port *lp)
 	if (m == NULL)
 		return (ENOMEM);
 	m->m_len = m->m_pkthdr.len = sizeof(*mdu);
-	m->m_pkthdr.pf.prio = LACP_IFQ_PRIO;
+	m->m_pkthdr.pf.prio = lsc->lsc_ifq_prio;
 
 	mdu = mtod(m, struct markerdu *);
 	memset(mdu, 0, sizeof(*mdu));
@@ -522,9 +510,6 @@ lacp_port_create(struct trunk_port *tp)
 	struct ifreq ifr;
 	int error;
 
-	int active = 1; /* XXX should be configurable */
-	int fast = 0; /* XXX should be configurable */
-
 	bzero(&ifr, sizeof(ifr));
 	ifr.ifr_addr.sa_family = AF_UNSPEC;
 	ifr.ifr_addr.sa_len = ETHER_ADDR_LEN;
@@ -554,8 +539,8 @@ lacp_port_create(struct trunk_port *tp)
 	lacp_fill_actorinfo(lp, &lp->lp_actor);
 	lacp_fill_markerinfo(lp, &lp->lp_marker);
 	lp->lp_state =
-	    (active ? LACP_STATE_ACTIVITY : 0) |
-	    (fast ? LACP_STATE_TIMEOUT : 0);
+	    (lsc->lsc_mode ? LACP_STATE_ACTIVITY : 0) |
+	    (lsc->lsc_timeout ? LACP_STATE_TIMEOUT : 0);
 	lp->lp_aggregator = NULL;
 	lacp_sm_rx_set_expired(lp);
 
@@ -764,6 +749,13 @@ lacp_attach(struct trunk_softc *sc)
 	TAILQ_INIT(&lsc->lsc_aggregators);
 	LIST_INIT(&lsc->lsc_ports);
 
+	/* set default admin values */
+	lsc->lsc_mode = LACP_DEFAULT_MODE;
+	lsc->lsc_timeout = LACP_DEFAULT_TIMEOUT;
+	lsc->lsc_sys_prio = LACP_DEFAULT_SYSTEM_PRIO;
+	lsc->lsc_port_prio = LACP_DEFAULT_PORT_PRIO;
+	lsc->lsc_ifq_prio = LACP_DEFAULT_IFQ_PRIO;
+
 	timeout_set(&lsc->lsc_transit_callout, lacp_transit_expire, lsc);
 	timeout_set(&lsc->lsc_callout, lacp_tick, lsc);
 	task_set(&lsc->lsc_input, lacp_input_process, lsc);
@@ -1555,12 +1547,15 @@ lacp_sm_rx_update_ntt(struct lacp_port *lp, const struct lacpdu *du)
 void
 lacp_sm_rx_record_default(struct lacp_port *lp)
 {
+	struct lacp_softc *lsc;
 	u_int8_t oldpstate;
 
+	lsc = lp->lp_lsc;
+
 	/* LACP_DPRINTF((lp, "%s\n", __func__)); */
 
 	oldpstate = lp->lp_partner.lip_state;
-	lp->lp_partner = lacp_partner_admin;
+	lacp_default_partner(lsc, &(lp->lp_partner));
 	lp->lp_state |= LACP_STATE_DEFAULTED;
 	lacp_sm_ptx_update_timeout(lp, oldpstate);
 }
@@ -1590,9 +1585,14 @@ lacp_sm_rx_update_selected(struct lacp_port *lp, const struct lacpdu *du)
 void
 lacp_sm_rx_update_default_selected(struct lacp_port *lp)
 {
+	struct lacp_softc *lsc;
+	struct lacp_peerinfo peer;
+
+	lsc = lp->lp_lsc;
+	lacp_default_partner(lsc, &peer);
 	/* LACP_DPRINTF((lp, "%s\n", __func__)); */
 
-	lacp_sm_rx_update_selected_from_peerinfo(lp, &lacp_partner_admin);
+	lacp_sm_rx_update_selected_from_peerinfo(lp, &peer);
 }
 
 /* transmit machine */
diff --git a/sys/net/trunklacp.h b/sys/net/trunklacp.h
index 869076aa4ea..1abc67f279e 100644
--- a/sys/net/trunklacp.h
+++ b/sys/net/trunklacp.h
@@ -1,4 +1,4 @@
-/*	$OpenBSD: trunklacp.h,v 1.13 2018/05/12 02:02:34 ccardenas Exp $	*/
+/*	$OpenBSD: trunklacp.h,v 1.14 2018/08/12 23:50:31 ccardenas Exp $	*/
 /*	$NetBSD: ieee8023ad_impl.h,v 1.2 2005/12/10 23:21:39 elad Exp $	*/
 
 /*
@@ -39,6 +39,19 @@
 #define	SLOWPROTOCOLS_SUBTYPE_LACP	1
 #define	SLOWPROTOCOLS_SUBTYPE_MARKER	2
 
+/*
+ * default administrative values
+ */
+#define	LACP_DEFAULT_MODE		1 /* Active Mode */
+#define	LACP_DEFAULT_TIMEOUT		0 /* Slow Timeout */
+#define	LACP_DEFAULT_SYSTEM_PRIO	0x8000 /* Medium Priority */
+#define	LACP_LOW_SYSTEM_PRIO		0xffff
+#define	LACP_HIGH_SYSTEM_PRIO		0x0001
+#define	LACP_DEFAULT_PORT_PRIO		0x8000 /* Medium Priority */
+#define	LACP_LOW_PORT_PRIO		0xffff
+#define	LACP_HIGH_PORT_PRIO		0x0001
+#define	LACP_DEFAULT_IFQ_PRIO		6
+
 struct slowprothdr {
 	u_int8_t		sph_subtype;
 	u_int8_t		sph_version;
@@ -221,6 +234,14 @@ struct lacp_aggregator {
 	int			la_pending; /* number of ports in wait_while */
 };
 
+struct lacp_admin_def {
+	u_int8_t		lad_mode; /* active or passive */
+	u_int8_t		lad_timeout; /* fast or slow */
+	u_int16_t		lad_prio; /* system priority */
+	u_int16_t		lad_portprio; /* port priority */
+	u_int8_t		lad_ifqprio; /* ifq priority */
+};
+
 struct lacp_softc {
 	struct trunk_softc	*lsc_softc;
 	struct lacp_aggregator	*lsc_active_aggregator;
@@ -233,6 +254,12 @@ struct lacp_softc {
 	volatile u_int		lsc_activemap;
 	SIPHASH_KEY		lsc_hashkey;
 	struct task		lsc_input;
+	struct lacp_admin_def	lsc_admin_defaults;
+#define lsc_mode		lsc_admin_defaults.lad_mode
+#define lsc_timeout		lsc_admin_defaults.lad_timeout
+#define lsc_sys_prio	lsc_admin_defaults.lad_prio
+#define lsc_port_prio	lsc_admin_defaults.lad_portprio
+#define lsc_ifq_prio	lsc_admin_defaults.lad_ifqprio
 };
 
 #define	LACP_TYPE_ACTORINFO	1