xref: /xnu-12377.41.6/bsd/net/if_bond.c (revision bbb1b6f9e71b8cdde6e5cd6f4841f207dee3d828)

/*
 * Copyright (c) 2004-2024 Apple Inc. All rights reserved.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_START@
 *
 * This file contains Original Code and/or Modifications of Original Code
 * as defined in and that are subject to the Apple Public Source License
 * Version 2.0 (the 'License'). You may not use this file except in
 * compliance with the License. The rights granted to you under the License
 * may not be used to create, or enable the creation or redistribution of,
 * unlawful or unlicensed copies of an Apple operating system, or to
 * circumvent, violate, or enable the circumvention or violation of, any
 * terms of an Apple operating system software license agreement.
 *
 * Please obtain a copy of the License at
 * http://www.opensource.apple.com/apsl/ and read it before using this file.
 *
 * The Original Code and all software distributed under the License are
 * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER
 * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES,
 * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY,
 * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT.
 * Please see the License for the specific language governing rights and
 * limitations under the License.
 *
 * @APPLE_OSREFERENCE_LICENSE_HEADER_END@
 */

/*
 * if_bond.c
 * - bond/failover interface
 * - implements IEEE 802.3ad Link Aggregation
 */

/*
 * Modification History:
 *
 * April 29, 2004	Dieter Siegmund ([email protected])
 * - created
 */

#include <sys/param.h>
#include <sys/kernel.h>
#include <sys/malloc.h>
#include <sys/mbuf.h>
#include <sys/queue.h>
#include <sys/socket.h>
#include <sys/sockio.h>
#include <sys/sysctl.h>
#include <sys/systm.h>
#include <sys/kern_event.h>
#include <net/bpf.h>
#include <net/ethernet.h>
#include <net/if.h>
#include <net/kpi_interface.h>
#include <net/if_arp.h>
#include <net/if_dl.h>
#include <net/if_ether.h>
#include <net/if_types.h>
#include <net/if_bond_var.h>
#include <net/ieee8023ad.h>
#include <net/lacp.h>
#include <net/dlil.h>
#include <sys/time.h>
#include <net/devtimer.h>
#include <net/if_vlan_var.h>
#include <net/kpi_protocol.h>
#include <sys/protosw.h>
#include <kern/locks.h>
#include <kern/uipc_domain.h>
#include <kern/zalloc.h>
#include <os/refcnt.h>

#include <netinet/in.h>
#include <netinet/if_ether.h>
#include <netinet/in_systm.h>
#include <netinet/ip.h>
#include <netinet/ip6.h>

#include <net/if_media.h>
#include <net/multicast_list.h>

#include <net/sockaddr_utils.h>
#include <net/mblist.h>

static struct ether_addr slow_proto_multicast = {
	.octet = IEEE8023AD_SLOW_PROTO_MULTICAST
};

typedef struct ifbond_s ifbond, *__single ifbond_ref;
typedef struct bondport_s bondport, *__single bondport_ref;

#define BOND_MAXUNIT            128
#define BOND_ZONE_MAX_ELEM      MIN(IFNETS_MAX, BOND_MAXUNIT)
#define BONDNAME                "bond"

#define EA_FORMAT       "%x:%x:%x:%x:%x:%x"
#define EA_CH(e, i)     ((u_char)((u_char *)(e))[(i)])
#define EA_LIST(ea)     EA_CH(ea,0),EA_CH(ea,1),EA_CH(ea,2),EA_CH(ea,3),EA_CH(ea,4),EA_CH(ea,5)

/*
 * if_bond_debug, BD_DBGF_*
 * - 'if_bond_debug' is a bitmask of BD_DBGF_* flags that can be set
 *   to enable additional logs for the corresponding bond function
 * - "sysctl net.link.bond.debug" controls the value of
 *   'if_bond_debug'
 */
static uint32_t if_bond_debug = 0;
#define BD_DBGF_LIFECYCLE       0x0001
#define BD_DBGF_INPUT           0x0002
#define BD_DBGF_OUTPUT          0x0004
#define BD_DBGF_LACP            0x0008

/*
 * if_bond_log_level
 * - 'if_bond_log_level' ensures that by default important logs are
 *   logged regardless of if_bond_debug by comparing the log level
 *   in BOND_LOG to if_bond_log_level
 * - use "sysctl net.link.bond.log_level" controls the value of
 *   'if_bond_log_level'
 * - the default value of 'if_bond_log_level' is LOG_NOTICE; important
 *   logs must use LOG_NOTICE to ensure they appear by default
 */
static int if_bond_log_level = LOG_NOTICE;

#define BOND_DBGF_ENABLED(__flag)     ((if_bond_debug & __flag) != 0)

/*
 * BOND_LOG, BOND_LOG_SIMPLE
 * - macros to generate the specified log conditionally based on
 *   the specified log level and debug flags
 * - BOND_LOG_SIMPLE does not include the function name in the log
 */
#define BOND_LOG(__level, __dbgf, __string, ...)              \
	do {                                                            \
	        if (__level <= if_bond_log_level ||                   \
	            BOND_DBGF_ENABLED(__dbgf)) {                      \
	                os_log(OS_LOG_DEFAULT, "%s: " __string, \
	                       __func__, ## __VA_ARGS__);       \
	        }                                                       \
	} while (0)
#define BOND_LOG_SIMPLE(__level, __dbgf, __string, ...)               \
	do {                                                    \
	        if (__level <= if_bond_log_level ||           \
	            BOND_DBGF_ENABLED(__dbgf)) {                      \
	                os_log(OS_LOG_DEFAULT, __string, ## __VA_ARGS__); \
	        }                                                               \
	} while (0)

SYSCTL_DECL(_net_link);
SYSCTL_NODE(_net_link, OID_AUTO, bond, CTLFLAG_RW | CTLFLAG_LOCKED, 0,
    "Bond interface");

SYSCTL_INT(_net_link_bond, OID_AUTO, debug, CTLFLAG_RW | CTLFLAG_LOCKED,
    &if_bond_debug, 0, "Bond interface debug flags");

SYSCTL_INT(_net_link_bond, OID_AUTO, log_level, CTLFLAG_RW | CTLFLAG_LOCKED,
    &if_bond_log_level, 0, "Bond interface log level");

/**
** bond locks
**/

static LCK_GRP_DECLARE(bond_lck_grp, "if_bond");
static LCK_MTX_DECLARE(bond_lck_mtx, &bond_lck_grp);

static __inline__ void
bond_assert_lock_held(void)
{
	LCK_MTX_ASSERT(&bond_lck_mtx, LCK_MTX_ASSERT_OWNED);
}

static __inline__ void
bond_assert_lock_not_held(void)
{
	LCK_MTX_ASSERT(&bond_lck_mtx, LCK_MTX_ASSERT_NOTOWNED);
}

static __inline__ void
bond_lock(void)
{
	lck_mtx_lock(&bond_lck_mtx);
}

static __inline__ void
bond_unlock(void)
{
	lck_mtx_unlock(&bond_lck_mtx);
}

/**
** bond structures, types
**/

struct LAG_info_s {
	lacp_system                 li_system;
	lacp_system_priority        li_system_priority;
	lacp_key                    li_key;
};
typedef struct LAG_info_s LAG_info, * LAG_info_ref;

struct bondport_s;
TAILQ_HEAD(port_list, bondport_s);
struct ifbond_s;
TAILQ_HEAD(ifbond_list, ifbond_s);
struct LAG_s;
TAILQ_HEAD(lag_list, LAG_s);

struct LAG_s {
	TAILQ_ENTRY(LAG_s)          lag_list;
	struct port_list            lag_port_list;
	short                       lag_port_count;
	short                       lag_selected_port_count;
	int                         lag_active_media;
	LAG_info                    lag_info;
};
typedef struct LAG_s LAG, * LAG_ref;

typedef struct partner_state_s {
	LAG_info                    ps_lag_info;
	lacp_port                   ps_port;
	lacp_port_priority          ps_port_priority;
	lacp_actor_partner_state    ps_state;
} partner_state, * partner_state_ref;

struct ifbond_s {
	TAILQ_ENTRY(ifbond_s)       ifb_bond_list;
	int                         ifb_flags;
	struct os_refcnt            ifb_retain_count;
	char                        ifb_name[IFNAMSIZ];
	struct ifnet *              ifb_ifp;
	int                         ifb_altmtu;
	struct port_list            ifb_port_list;
	short                       ifb_port_count;
	struct lag_list             ifb_lag_list;
	lacp_key                    ifb_key;
	short                       ifb_max_active;/* 0 == unlimited */
	LAG_ref                     ifb_active_lag;
	struct ifmultiaddr *        ifb_ifma_slow_proto;
	bondport_ref *__counted_by(ifb_distributing_max) ifb_distributing_array;
	int                         ifb_distributing_count;
	int                         ifb_distributing_max;
	int                         ifb_last_link_event;
	int                         ifb_mode;/* LACP, STATIC */
};

struct media_info {
	int         mi_active;
	int         mi_status;
};

enum {
	ReceiveState_none = 0,
	ReceiveState_INITIALIZE = 1,
	ReceiveState_PORT_DISABLED = 2,
	ReceiveState_EXPIRED = 3,
	ReceiveState_LACP_DISABLED = 4,
	ReceiveState_DEFAULTED = 5,
	ReceiveState_CURRENT = 6,
};

typedef u_char ReceiveState;

enum {
	SelectedState_UNSELECTED = IF_BOND_STATUS_SELECTED_STATE_UNSELECTED,
	SelectedState_SELECTED = IF_BOND_STATUS_SELECTED_STATE_SELECTED,
	SelectedState_STANDBY = IF_BOND_STATUS_SELECTED_STATE_STANDBY
};
typedef u_char SelectedState;

static __inline__ const char *
SelectedStateString(SelectedState s)
{
	static const char * names[] = { "UNSELECTED", "SELECTED", "STANDBY" };

	if (s <= SelectedState_STANDBY) {
		return names[s];
	}
	return "<unknown>";
}

enum {
	MuxState_none = 0,
	MuxState_DETACHED = 1,
	MuxState_WAITING = 2,
	MuxState_ATTACHED = 3,
	MuxState_COLLECTING_DISTRIBUTING = 4,
};

typedef u_char MuxState;

#define PORT_CONTROL_FLAGS_IN_LIST               0x01
#define PORT_CONTROL_FLAGS_PROTO_ATTACHED        0x02
#define PORT_CONTROL_FLAGS_LLADDR_SET            0x04
#define PORT_CONTROL_FLAGS_MTU_SET               0x08
#define PORT_CONTROL_FLAGS_PROMISCUOUS_SET       0x10
#define PORT_CONTROL_FLAGS_BOND_PROMISCUOUS_SET  0x20


static inline bool
uint32_bit_is_set(uint32_t flags, uint32_t flags_to_test)
{
	return (flags & flags_to_test) != 0;
}

static inline void
uint32_bit_set(uint32_t * flags_p, uint32_t flags_to_set)
{
	*flags_p |= flags_to_set;
}

static inline void
uint32_bit_clear(uint32_t * flags_p, uint32_t flags_to_clear)
{
	*flags_p &= ~flags_to_clear;
}

struct bondport_s {
	TAILQ_ENTRY(bondport_s)     po_port_list;
	ifbond_ref                  po_bond;
	struct multicast_list       po_multicast;
	struct ifnet *              po_ifp;
	struct ether_addr           po_saved_addr;
	int                         po_enabled;
	char                        po_name[IFNAMSIZ];
	struct ifdevmtu             po_devmtu;
	uint32_t                    po_control_flags;

	/* LACP */
	TAILQ_ENTRY(bondport_s)     po_lag_port_list;
	devtimer_ref                po_current_while_timer;
	devtimer_ref                po_periodic_timer;
	devtimer_ref                po_wait_while_timer;
	devtimer_ref                po_transmit_timer;
	partner_state               po_partner_state;
	lacp_port_priority          po_priority;
	lacp_actor_partner_state    po_actor_state;
	u_char                      po_flags;
	u_char                      po_periodic_interval;
	u_char                      po_n_transmit;
	ReceiveState                po_receive_state;
	MuxState                    po_mux_state;
	SelectedState               po_selected;
	int32_t                     po_last_transmit_secs;
	struct media_info           po_media_info;
	uint64_t                    po_force_link_event_time;
	LAG_ref                     po_lag;
};

#define IFBF_PROMISC            0x1     /* promiscuous mode */
#define IFBF_IF_DETACHING       0x2     /* interface is detaching */
#define IFBF_LLADDR             0x4     /* specific link address requested */
#define IFBF_CHANGE_IN_PROGRESS 0x8     /* interface add/remove in progress */

static int bond_get_status(ifbond_ref ifb, struct if_bond_req * ibr_p,
    user_addr_t datap);

static __inline__ bool
ifbond_flags_if_detaching(ifbond_ref ifb)
{
	return (ifb->ifb_flags & IFBF_IF_DETACHING) != 0;
}

static __inline__ void
ifbond_flags_set_if_detaching(ifbond_ref ifb)
{
	ifb->ifb_flags |= IFBF_IF_DETACHING;
	return;
}

static __inline__ bool
ifbond_flags_lladdr(ifbond_ref ifb)
{
	return (ifb->ifb_flags & IFBF_LLADDR) != 0;
}

static __inline__ bool
ifbond_flags_change_in_progress(ifbond_ref ifb)
{
	return (ifb->ifb_flags & IFBF_CHANGE_IN_PROGRESS) != 0;
}

static __inline__ void
ifbond_flags_set_change_in_progress(ifbond_ref ifb)
{
	ifb->ifb_flags |= IFBF_CHANGE_IN_PROGRESS;
	return;
}

static __inline__ void
ifbond_flags_clear_change_in_progress(ifbond_ref ifb)
{
	ifb->ifb_flags &= ~IFBF_CHANGE_IN_PROGRESS;
	return;
}

static __inline__ bool
ifbond_flags_promisc(ifbond_ref ifb)
{
	return (ifb->ifb_flags & IFBF_PROMISC) != 0;
}

static __inline__ void
ifbond_flags_set_promisc(ifbond_ref ifb)
{
	ifb->ifb_flags |= IFBF_PROMISC;
	return;
}

static __inline__ void
ifbond_flags_clear_promisc(ifbond_ref ifb)
{
	ifb->ifb_flags &= ~IFBF_PROMISC;
	return;
}

/*
 * bondport_ref->po_flags bits
 */
#define BONDPORT_FLAGS_NTT              0x01
#define BONDPORT_FLAGS_READY            0x02
#define BONDPORT_FLAGS_SELECTED_CHANGED 0x04
#define BONDPORT_FLAGS_MUX_ATTACHED     0x08
#define BONDPORT_FLAGS_DISTRIBUTING     0x10
#define BONDPORT_FLAGS_UNUSED2          0x20
#define BONDPORT_FLAGS_UNUSED3          0x40
#define BONDPORT_FLAGS_UNUSED4          0x80

static __inline__ void
bondport_flags_set_ntt(bondport_ref p)
{
	p->po_flags |= BONDPORT_FLAGS_NTT;
	return;
}

static __inline__ void
bondport_flags_clear_ntt(bondport_ref p)
{
	p->po_flags &= ~BONDPORT_FLAGS_NTT;
	return;
}

static __inline__ int
bondport_flags_ntt(bondport_ref p)
{
	return (p->po_flags & BONDPORT_FLAGS_NTT) != 0;
}

static __inline__ void
bondport_flags_set_ready(bondport_ref p)
{
	p->po_flags |= BONDPORT_FLAGS_READY;
	return;
}

static __inline__ void
bondport_flags_clear_ready(bondport_ref p)
{
	p->po_flags &= ~BONDPORT_FLAGS_READY;
	return;
}

static __inline__ int
bondport_flags_ready(bondport_ref p)
{
	return (p->po_flags & BONDPORT_FLAGS_READY) != 0;
}

static __inline__ void
bondport_flags_set_selected_changed(bondport_ref p)
{
	p->po_flags |= BONDPORT_FLAGS_SELECTED_CHANGED;
	return;
}

static __inline__ void
bondport_flags_clear_selected_changed(bondport_ref p)
{
	p->po_flags &= ~BONDPORT_FLAGS_SELECTED_CHANGED;
	return;
}

static __inline__ int
bondport_flags_selected_changed(bondport_ref p)
{
	return (p->po_flags & BONDPORT_FLAGS_SELECTED_CHANGED) != 0;
}

static __inline__ void
bondport_flags_set_mux_attached(bondport_ref p)
{
	p->po_flags |= BONDPORT_FLAGS_MUX_ATTACHED;
	return;
}

static __inline__ void
bondport_flags_clear_mux_attached(bondport_ref p)
{
	p->po_flags &= ~BONDPORT_FLAGS_MUX_ATTACHED;
	return;
}

static __inline__ int
bondport_flags_mux_attached(bondport_ref p)
{
	return (p->po_flags & BONDPORT_FLAGS_MUX_ATTACHED) != 0;
}

static __inline__ void
bondport_flags_set_distributing(bondport_ref p)
{
	p->po_flags |= BONDPORT_FLAGS_DISTRIBUTING;
	return;
}

static __inline__ void
bondport_flags_clear_distributing(bondport_ref p)
{
	p->po_flags &= ~BONDPORT_FLAGS_DISTRIBUTING;
	return;
}

static __inline__ int
bondport_flags_distributing(bondport_ref p)
{
	return (p->po_flags & BONDPORT_FLAGS_DISTRIBUTING) != 0;
}

typedef struct bond_globals_s {
	struct ifbond_list          ifbond_list;
	lacp_system                 system;
	lacp_system_priority        system_priority;
} * bond_globals_ref;

static bond_globals_ref g_bond;

/**
** packet_buffer routines
** - thin wrapper for mbuf
**/

typedef struct mbuf * packet_buffer_ref;

static packet_buffer_ref
packet_buffer_allocate(int length)
{
	packet_buffer_ref   m;
	int                 size;

	/* leave room for ethernet header */
	size = length + sizeof(struct ether_header);
	if (size > (int)MHLEN) {
		if (size > (int)MCLBYTES) {
			BOND_LOG(LOG_NOTICE, 0,
			    "packet_buffer_allocate size %d > max %u",
			    size, MCLBYTES);
			return NULL;
		}
		m = m_getcl(M_WAITOK, MT_DATA, M_PKTHDR);
	} else {
		m = m_gethdr(M_WAITOK, MT_DATA);
	}
	if (m == NULL) {
		return NULL;
	}
	m->m_len = size;
	m->m_pkthdr.len = size;
	return m;
}

static void *__indexable
packet_buffer_byteptr(packet_buffer_ref buf)
{
	return m_mtod_current(buf) + sizeof(struct ether_header);
}

typedef enum {
	LAEventStart,
	LAEventTimeout,
	LAEventPacket,
	LAEventMediaChange,
	LAEventSelectedChange,
	LAEventPortMoved,
	LAEventReady
} LAEvent;

/**
** Receive machine
**/
static void
bondport_receive_machine(bondport_ref p, LAEvent event,
    void * event_data);
/**
** Periodic Transmission machine
**/
static void
bondport_periodic_transmit_machine(bondport_ref p, LAEvent event,
    void * event_data);

/**
** Transmit machine
**/
static void *TRANSMIT_MACHINE_TX_IMMEDIATE = __unsafe_forge_single(void *, 1);

static void
bondport_transmit_machine(bondport_ref p, LAEvent event,
    void * event_data);

/**
** Mux machine
**/
static void
bondport_mux_machine(bondport_ref p, LAEvent event,
    void * event_data);

/**
** bond, LAG
**/
static void
ifbond_activate_LAG(ifbond_ref bond, LAG_ref lag, int active_media);

static void
ifbond_deactivate_LAG(ifbond_ref bond, LAG_ref lag);

static int
ifbond_all_ports_ready(ifbond_ref bond);

static LAG_ref
ifbond_find_best_LAG(ifbond_ref bond, int * active_media);

static int
LAG_get_aggregatable_port_count(LAG_ref lag, int * active_media);

static int
ifbond_selection(ifbond_ref bond);

static void
bond_handle_event(struct ifnet * port_ifp, int event_code);

/**
** bondport
**/

static void
bondport_receive_lacpdu(bondport_ref p, lacpdu_ref in_lacpdu_p);

static void
bondport_slow_proto_transmit(bondport_ref p, packet_buffer_ref buf);

static bondport_ref
bondport_create(struct ifnet * port_ifp, lacp_port_priority priority,
    int active, int short_timeout, int * error);
static void
bondport_start(bondport_ref p);

static void
bondport_free(bondport_ref p);

static int
bondport_aggregatable(bondport_ref p);

static int
bondport_remove_from_LAG(bondport_ref p);

static void
bondport_set_selected(bondport_ref p, SelectedState s);

static int
bondport_matches_LAG(bondport_ref p, LAG_ref lag);

static void
bondport_link_status_changed(bondport_ref p);

static void
bondport_enable_distributing(bondport_ref p);

static void
bondport_disable_distributing(bondport_ref p);

static __inline__ int
bondport_collecting(bondport_ref p)
{
	if (p->po_bond->ifb_mode == IF_BOND_MODE_LACP) {
		return lacp_actor_partner_state_collecting(p->po_actor_state);
	}
	return TRUE;
}

/**
** bond interface/dlil specific routines
**/
static int bond_clone_create(struct if_clone *, u_int32_t, void *);
static int bond_clone_destroy(struct ifnet *);
static int bond_output(ifnet_t ifp, mbuf_t list);
static int bond_ioctl(struct ifnet *ifp, u_long cmd, void * addr);
static int bond_attach_protocol(struct ifnet *ifp);
static int bond_detach_protocol(struct ifnet *ifp);
static int bond_setmulti(struct ifnet *ifp);
static int bond_add_interface(struct ifnet * ifp, struct ifnet * port_ifp);
static int bond_remove_interface(ifbond_ref ifb, struct ifnet * port_ifp);
static void bond_if_free(struct ifnet * ifp);
static void interface_link_event(struct ifnet * ifp, u_int32_t event_code);

static struct if_clone bond_cloner = IF_CLONE_INITIALIZER(BONDNAME,
    bond_clone_create,
    bond_clone_destroy,
    0,
    BOND_MAXUNIT);

static int
siocsifmtu(struct ifnet * ifp, int mtu)
{
	struct ifreq        ifr;

	bzero(&ifr, sizeof(ifr));
	ifr.ifr_mtu = mtu;
	return ifnet_ioctl(ifp, 0, SIOCSIFMTU, &ifr);
}

static int
siocgifdevmtu(struct ifnet * ifp, struct ifdevmtu * ifdm_p)
{
	struct ifreq        ifr;
	int                 error;

	bzero(&ifr, sizeof(ifr));
	error = ifnet_ioctl(ifp, 0, SIOCGIFDEVMTU, &ifr);
	if (error == 0) {
		*ifdm_p = ifr.ifr_devmtu;
	}
	return error;
}

static __inline__ void
ether_addr_copy(uint8_t *__sized_by(ETHER_ADDR_LEN) dest,
    const uint8_t *__sized_by(ETHER_ADDR_LEN) source)
{
	bcopy(source, dest, ETHER_ADDR_LEN);
	return;
}

static __inline__ void
ifbond_retain(ifbond_ref ifb)
{
	os_ref_retain(&ifb->ifb_retain_count);
}

static __inline__ void
ifbond_release(ifbond_ref ifb)
{
	if (os_ref_release(&ifb->ifb_retain_count) != 0) {
		return;
	}
	BOND_LOG(LOG_DEBUG, BD_DBGF_LIFECYCLE, "%s", ifb->ifb_name);
	if (ifb->ifb_ifma_slow_proto != NULL) {
		BOND_LOG(LOG_DEBUG, BD_DBGF_LIFECYCLE,
		    "%s: removing multicast", ifb->ifb_name);
		(void) if_delmulti_anon(ifb->ifb_ifma_slow_proto->ifma_ifp,
		    ifb->ifb_ifma_slow_proto->ifma_addr);
		IFMA_REMREF(ifb->ifb_ifma_slow_proto);
	}
	kfree_type_counted_by(bondport_ref, ifb->ifb_distributing_max,
	    ifb->ifb_distributing_array);
	kfree_type(struct ifbond_s, ifb);
}

/*
 * Function: ifbond_wait
 * Purpose:
 *   Allows a single thread to gain exclusive access to the ifbond
 *   data structure.  Some operations take a long time to complete,
 *   and some have side-effects that we can't predict.  Holding the
 *   bond_lock() across such operations is not possible.
 *
 *   For example:
 *   1) The SIOCSIFLLADDR ioctl takes a long time (several seconds) to
 *      complete.  Simply holding the bond_lock() would freeze all other
 *      data structure accesses during that time.
 *   2) When we attach our protocol to the interface, a dlil event is
 *      generated and invokes our bond_event() function.  bond_event()
 *      needs to take the bond_lock(), but we're already holding it, so
 *      we're deadlocked against ourselves.
 * Notes:
 *   Before calling, you must be holding the bond_lock and have taken
 *   a reference on the ifbond_ref.
 */
static void
ifbond_wait(ifbond_ref ifb, const char * msg)
{
	int         waited = 0;

	/* other add/remove in progress */
	while (ifbond_flags_change_in_progress(ifb)) {
		BOND_LOG(LOG_DEBUG, BD_DBGF_LIFECYCLE, "%s: %s msleep",
		    ifb->ifb_name, msg);
		waited = 1;
		(void)msleep(ifb, &bond_lck_mtx, PZERO, msg, 0);
	}
	/* prevent other bond list remove/add from taking place */
	ifbond_flags_set_change_in_progress(ifb);
	if (waited) {
		BOND_LOG(LOG_DEBUG, BD_DBGF_LIFECYCLE,
		    "%s: %s woke up", ifb->ifb_name, msg);
	}
	return;
}

/*
 * Function: ifbond_signal
 * Purpose:
 *   Allows the thread that previously invoked ifbond_wait() to
 *   give up exclusive access to the ifbond data structure, and wake up
 *   any other threads waiting to access
 * Notes:
 *   Before calling, you must be holding the bond_lock and have taken
 *   a reference on the ifbond_ref.
 */
static void
ifbond_signal(ifbond_ref ifb, const char * msg)
{
	ifbond_flags_clear_change_in_progress(ifb);
	wakeup((caddr_t)ifb);
	BOND_LOG(LOG_DEBUG, BD_DBGF_LIFECYCLE, "%s: %s wakeup",
	    ifb->ifb_name, msg);
	return;
}

/**
** Media information
**/

static int
link_speed(int active)
{
	switch (IFM_SUBTYPE(active)) {
	case IFM_AUTO:
	case IFM_MANUAL:
	case IFM_NONE:
		return 0;
	case IFM_10_T:
	case IFM_10_2:
	case IFM_10_5:
	case IFM_10_STP:
	case IFM_10_FL:
		return 10;
	case IFM_100_TX:
	case IFM_100_FX:
	case IFM_100_T4:
	case IFM_100_VG:
	case IFM_100_T2:
		return 100;
	case IFM_1000_SX:
	case IFM_1000_LX:
	case IFM_1000_CX:
	case IFM_1000_TX:
	case IFM_1000_CX_SGMII:
	case IFM_1000_KX:
		return 1000;
	case IFM_HPNA_1:
		return 1;
	default:
	/* assume that new defined types are going to be at least 10GigE */
	case IFM_10G_T:
	case IFM_10G_SR:
	case IFM_10G_LR:
	case IFM_10G_KX4:
	case IFM_10G_KR:
	case IFM_10G_CR1:
	case IFM_10G_ER:
		return 10000;
	case IFM_2500_T:
		return 2500;
	case IFM_5000_T:
		return 5000;
	case IFM_20G_KR2:
		return 20000;
	case IFM_25G_CR:
	case IFM_25G_KR:
	case IFM_25G_SR:
	case IFM_25G_LR:
		return 25000;
	case IFM_40G_CR4:
	case IFM_40G_SR4:
	case IFM_40G_LR4:
	case IFM_40G_KR4:
		return 40000;
	case IFM_50G_CR2:
	case IFM_50G_KR2:
	case IFM_50G_SR2:
	case IFM_50G_LR2:
		return 50000;
	case IFM_56G_R4:
		return 56000;
	case IFM_100G_CR4:
	case IFM_100G_SR4:
	case IFM_100G_KR4:
	case IFM_100G_LR4:
		return 100000;
	}
}

static __inline__ int
media_active(const struct media_info * mi)
{
	if ((mi->mi_status & IFM_AVALID) == 0) {
		return 1;
	}
	return (mi->mi_status & IFM_ACTIVE) != 0;
}

static __inline__ int
media_full_duplex(const struct media_info * mi)
{
	return (mi->mi_active & IFM_FDX) != 0;
}

static __inline__ int
media_type_unknown(const struct media_info * mi)
{
	int unknown;

	switch (IFM_SUBTYPE(mi->mi_active)) {
	case IFM_AUTO:
	case IFM_MANUAL:
	case IFM_NONE:
		unknown = 1;
		break;
	default:
		unknown = 0;
		break;
	}
	return unknown;
}

static __inline__ int
media_ok(const struct media_info * mi)
{
	return media_full_duplex(mi) || media_type_unknown(mi);
}

static __inline__ int
media_speed(const struct media_info * mi)
{
	return link_speed(mi->mi_active);
}

static struct media_info
interface_media_info(struct ifnet * ifp)
{
	struct ifmediareq   ifmr;
	struct media_info   mi;

	bzero(&mi, sizeof(mi));
	bzero(&ifmr, sizeof(ifmr));
	if (ifnet_ioctl(ifp, 0, SIOCGIFMEDIA, &ifmr) == 0) {
		if (ifmr.ifm_count != 0) {
			mi.mi_status = ifmr.ifm_status;
			mi.mi_active = ifmr.ifm_active;
		}
	}
	return mi;
}

static int
if_siflladdr(struct ifnet * ifp, const struct ether_addr * ea_p)
{
	struct ifreq        ifr;

	/*
	 * XXX setting the sa_len to ETHER_ADDR_LEN is wrong, but the driver
	 * currently expects it that way
	 */
	ifr.ifr_addr.sa_family = AF_UNSPEC;
	ifr.ifr_addr.sa_len = ETHER_ADDR_LEN;
	ether_addr_copy((uint8_t *__indexable)ifr.ifr_addr.sa_data, ea_p->octet);
	return ifnet_ioctl(ifp, 0, SIOCSIFLLADDR, &ifr);
}

/**
** bond_globals
**/
static bond_globals_ref
bond_globals_create(lacp_system_priority sys_pri,
    lacp_system_ref sys)
{
	bond_globals_ref    b;

	b = kalloc_type(struct bond_globals_s, Z_WAITOK | Z_ZERO | Z_NOFAIL);
	TAILQ_INIT(&b->ifbond_list);
	b->system = *sys;
	b->system_priority = sys_pri;
	return b;
}

static int
bond_globals_init(void)
{
	bond_globals_ref    b;
	int                 i;
	struct ifnet *      ifp;

	bond_assert_lock_not_held();

	if (g_bond != NULL) {
		return 0;
	}

	/*
	 * use en0's ethernet address as the system identifier, and if it's not
	 * there, use en1 .. en3
	 */
	ifp = NULL;
	for (i = 0; i < 4; i++) {
		char            ifname[IFNAMSIZ + 1];
		snprintf(ifname, sizeof(ifname), "en%d", i);
		ifp = ifunit(__unsafe_null_terminated_from_indexable(ifname));
		if (ifp != NULL) {
			break;
		}
	}
	b = NULL;
	if (ifp != NULL) {
		b = bond_globals_create(0x8000, (lacp_system_ref)IF_LLADDR(ifp));
	}
	bond_lock();
	if (g_bond != NULL) {
		bond_unlock();
		kfree_type(struct bond_globals_s, b);
		return 0;
	}
	g_bond = b;
	bond_unlock();
	if (ifp == NULL) {
		return ENXIO;
	}
	if (b == NULL) {
		return ENOMEM;
	}
	return 0;
}

/*
 * bpf tap
 */
static inline void *__indexable
get_bpf_header(mbuf_t m, struct ether_header * eh_p,
    struct ether_vlan_header * evl_p, size_t * header_len)
{
	void *header;

	/* no VLAN tag, just use the ethernet header */
	if ((m->m_pkthdr.csum_flags & CSUM_VLAN_TAG_VALID) == 0) {
		header = (struct ether_header *__bidi_indexable)eh_p;
		*header_len = sizeof(*eh_p);
		goto done;
	}

	/* has VLAN tag, populate the ether VLAN header */
	bcopy(eh_p, evl_p,
	    offsetof(struct ether_header, ether_type));   /* dst+src ether */
	evl_p->evl_encap_proto = htons(ETHERTYPE_VLAN);   /* VLAN encap */
	evl_p->evl_tag = htons(m->m_pkthdr.vlan_tag);     /* tag */
	evl_p->evl_proto = eh_p->ether_type;              /* proto */
	*header_len = sizeof(*evl_p);
	header = (struct ether_vlan_header *__bidi_indexable)evl_p;

done:
	return header;
}

typedef void (*_tap_func)(ifnet_t interface, u_int32_t dlt, mbuf_t packet,
    void *__sized_by(header_len) header, size_t header_len);

static void
bond_bpf_tap_common(ifnet_t ifp, mbuf_t m, struct ether_header * eh_p,
    _tap_func func)
{
	struct ether_vlan_header        evl;
	size_t                          header_len;
	void *                          header;

	header = get_bpf_header(m, eh_p, &evl, &header_len);
	(*func)(ifp, DLT_EN10MB, m, header, header_len);
}

static inline void
bond_bpf_tap_in(ifnet_t ifp, mbuf_t m, struct ether_header * eh_p)
{
	bond_bpf_tap_common(ifp, m, eh_p, bpf_tap_in);
}

static inline void
bond_bpf_tap_out(ifnet_t ifp, mbuf_t m, struct ether_header * eh_p)
{
	bond_bpf_tap_common(ifp, m, eh_p, bpf_tap_out);
}

/*
 * Function: bond_setmulti
 * Purpose:
 *   Enable multicast reception on "our" interface by enabling multicasts on
 *   each of the member ports.
 */
static int
bond_setmulti(struct ifnet * ifp)
{
	ifbond_ref          ifb;
	int                 error;
	int                 result = 0;
	bondport_ref        p;

	bond_lock();
	ifb = ifnet_softc(ifp);
	if (ifb == NULL || ifbond_flags_if_detaching(ifb)
	    || TAILQ_EMPTY(&ifb->ifb_port_list)) {
		bond_unlock();
		return 0;
	}
	ifbond_retain(ifb);
	ifbond_wait(ifb, "bond_setmulti");

	if (ifbond_flags_if_detaching(ifb)) {
		/* someone destroyed the bond while we were waiting */
		result = EBUSY;
		goto signal_done;
	}
	bond_unlock();

	/* ifbond_wait() let's us safely walk the list without holding the lock */
	TAILQ_FOREACH(p, &ifb->ifb_port_list, po_port_list) {
		struct ifnet *  port_ifp = p->po_ifp;

		error = multicast_list_program(&p->po_multicast,
		    ifp, port_ifp);
		if (error != 0) {
			BOND_LOG(LOG_NOTICE, 0,
			    "%s: multicast_list_program(%s) failed, %d",
			    ifb->ifb_name, port_ifp->if_xname, error);
			result = error;
		}
	}
	bond_lock();
signal_done:
	ifbond_signal(ifb, __func__);
	bond_unlock();
	ifbond_release(ifb);
	return result;
}

static int
bond_clone_attach(void)
{
	int error;

	if ((error = if_clone_attach(&bond_cloner)) != 0) {
		return error;
	}
	return 0;
}

static int
ifbond_add_slow_proto_multicast(ifbond_ref ifb)
{
	int                         error;
	struct ifmultiaddr *__single ifma = NULL;
	struct sockaddr_dl          sdl;

	bond_assert_lock_not_held();

	SOCKADDR_ZERO(&sdl, sizeof(sdl));
	sdl.sdl_len = sizeof(sdl);
	sdl.sdl_family = AF_LINK;
	sdl.sdl_type = IFT_ETHER;
	sdl.sdl_nlen = 0;
	sdl.sdl_alen = sizeof(slow_proto_multicast);
	bcopy(&slow_proto_multicast, sdl.sdl_data, sizeof(slow_proto_multicast));
	error = if_addmulti_anon(ifb->ifb_ifp, SA(&sdl), &ifma);
	if (error == 0) {
		ifb->ifb_ifma_slow_proto = ifma;
	}
	return error;
}

static int
bond_clone_create(struct if_clone * ifc, u_int32_t unit, __unused void *params)
{
	int                                             error;
	ifbond_ref                                      ifb;
	ifnet_ref_t                                     ifp;
	struct ifnet_init_eparams       bond_init;

	error = bond_globals_init();
	if (error != 0) {
		return error;
	}

	ifb = kalloc_type(struct ifbond_s, Z_WAITOK_ZERO_NOFAIL);
	os_ref_init(&ifb->ifb_retain_count, NULL);
	TAILQ_INIT(&ifb->ifb_port_list);
	TAILQ_INIT(&ifb->ifb_lag_list);
	ifb->ifb_key = unit + 1;

	/* use the interface name as the unique id for ifp recycle */
	if ((u_int32_t)snprintf(ifb->ifb_name, sizeof(ifb->ifb_name), "%s%d",
	    ifc->ifc_name, unit) >= sizeof(ifb->ifb_name)) {
		ifbond_release(ifb);
		return EINVAL;
	}

	bzero(&bond_init, sizeof(bond_init));
	bond_init.ver = IFNET_INIT_CURRENT_VERSION;
	bond_init.len = sizeof(bond_init);
	bond_init.flags = IFNET_INIT_LEGACY;
	bond_init.uniqueid_len = strbuflen(ifb->ifb_name);
	bond_init.uniqueid = ifb->ifb_name;
	bond_init.name = __unsafe_null_terminated_from_indexable(ifc->ifc_name);
	bond_init.unit = unit;
	bond_init.family = IFNET_FAMILY_BOND;
	bond_init.type = IFT_IEEE8023ADLAG;
	bond_init.output = bond_output;
	bond_init.demux = ether_demux;
	bond_init.add_proto = ether_add_proto;
	bond_init.del_proto = ether_del_proto;
	bond_init.check_multi = ether_check_multi;
	bond_init.framer_extended = ether_frameout_extended;
	bond_init.ioctl = bond_ioctl;
	bond_init.detach = bond_if_free;
	bond_init.broadcast_addr = etherbroadcastaddr;
	bond_init.broadcast_len = ETHER_ADDR_LEN;
	bond_init.softc = ifb;
	error = ifnet_allocate_extended(&bond_init, &ifp);

	if (error) {
		ifbond_release(ifb);
		return error;
	}

	ifb->ifb_ifp = ifp;
	ifnet_set_offload(ifp, 0);
	ifnet_set_addrlen(ifp, ETHER_ADDR_LEN); /* XXX ethernet specific */
	ifnet_set_flags(ifp, IFF_BROADCAST | IFF_MULTICAST | IFF_SIMPLEX, 0xffff);
	ifnet_set_mtu(ifp, ETHERMTU);

	error = ifnet_attach(ifp, NULL);
	if (error != 0) {
		ifnet_release(ifp);
		ifbond_release(ifb);
		return error;
	}
	error = ifbond_add_slow_proto_multicast(ifb);
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, 0,
		    "%s: failed to add slow_proto multicast, %d",
		    ifb->ifb_name, error);
	}

	/* attach as ethernet */
	bpfattach(ifp, DLT_EN10MB, sizeof(struct ether_header));

	bond_lock();
	TAILQ_INSERT_HEAD(&g_bond->ifbond_list, ifb, ifb_bond_list);
	bond_unlock();

	return 0;
}

static void
bond_remove_all_interfaces(ifbond_ref ifb)
{
	bondport_ref        p;

	bond_assert_lock_held();

	/*
	 * do this in reverse order to avoid re-programming the mac address
	 * as each head interface is removed
	 */
	while ((p = TAILQ_LAST(&ifb->ifb_port_list, port_list)) != NULL) {
		bond_remove_interface(ifb, p->po_ifp);
	}
	return;
}

static void
bond_remove(ifbond_ref ifb)
{
	bond_assert_lock_held();
	ifbond_flags_set_if_detaching(ifb);
	TAILQ_REMOVE(&g_bond->ifbond_list, ifb, ifb_bond_list);
	bond_remove_all_interfaces(ifb);
	return;
}

static void
bond_if_detach(struct ifnet * ifp)
{
	int         error;

	error = ifnet_detach(ifp);
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, 0, "%s: ifnet_detach failed, %d",
		    ifp->if_xname, error);
	}
	return;
}

static int
bond_clone_destroy(struct ifnet * ifp)
{
	ifbond_ref ifb;

	bond_lock();
	ifb = ifnet_softc(ifp);
	if (ifb == NULL || ifnet_type(ifp) != IFT_IEEE8023ADLAG) {
		bond_unlock();
		return 0;
	}
	if (ifbond_flags_if_detaching(ifb)) {
		bond_unlock();
		return 0;
	}
	bond_remove(ifb);
	bond_unlock();
	bond_if_detach(ifp);
	return 0;
}

static uint32_t
ether_header_hash(struct ether_header * eh_p)
{
	uint32_t    h;

	/* get 32-bits from destination ether and ether type */
	h = (*((uint16_t *)&eh_p->ether_dhost[4]) << 16)
	    | eh_p->ether_type;
	h ^= *((uint32_t *)&eh_p->ether_dhost[0]);
	return h;
}

#define BOND_HASH_L3_HEADER     0
#if BOND_HASH_L3_HEADER
static struct mbuf *
S_mbuf_skip_to_offset(struct mbuf * m, int32_t * offset)
{
	int                 len;

	len = m->m_len;
	while (*offset >= len) {
		*offset -= len;
		m = m->m_next;
		if (m == NULL) {
			break;
		}
		len = m->m_len;
	}
	return m;
}

#if BYTE_ORDER == BIG_ENDIAN
static __inline__ uint32_t
make_uint32(u_char c0, u_char c1, u_char c2, u_char c3)
{
	return ((uint32_t)c0 << 24) | ((uint32_t)c1 << 16)
	       | ((uint32_t)c2 << 8) | (uint32_t)c3;
}
#else /* BYTE_ORDER == LITTLE_ENDIAN */
static __inline__ uint32_t
make_uint32(u_char c0, u_char c1, u_char c2, u_char c3)
{
	return ((uint32_t)c3 << 24) | ((uint32_t)c2 << 16)
	       | ((uint32_t)c1 << 8) | (uint32_t)c0;
}
#endif /* BYTE_ORDER == LITTLE_ENDIAN */

static int
S_mbuf_copy_uint32(struct mbuf * m, int32_t offset, uint32_t * val)
{
	struct mbuf *       current;
	u_char *            current_data;
	struct mbuf *       next;
	u_char *            next_data;
	int                 space_current;

	current = S_mbuf_skip_to_offset(m, &offset);
	if (current == NULL) {
		return 1;
	}
	current_data = mtod(current, u_char *) + offset;
	space_current = current->m_len - offset;
	if (space_current >= (int)sizeof(uint32_t)) {
		*val = *((uint32_t *)current_data);
		return 0;
	}
	next = current->m_next;
	if (next == NULL || (next->m_len + space_current) < (int)sizeof(uint32_t)) {
		return 1;
	}
	next_data = mtod(next, u_char *);
	switch (space_current) {
	case 1:
		*val = make_uint32(current_data[0], next_data[0],
		    next_data[1], next_data[2]);
		break;
	case 2:
		*val = make_uint32(current_data[0], current_data[1],
		    next_data[0], next_data[1]);
		break;
	default:
		*val = make_uint32(current_data[0], current_data[1],
		    current_data[2], next_data[0]);
		break;
	}
	return 0;
}

#define IP_SRC_OFFSET (offsetof(struct ip, ip_src) - offsetof(struct ip, ip_p))
#define IP_DST_OFFSET (offsetof(struct ip, ip_dst) - offsetof(struct ip, ip_p))

static uint32_t
ip_header_hash(struct mbuf * m)
{
	u_char *            data;
	struct in_addr      ip_dst;
	struct in_addr      ip_src;
	u_char              ip_p;
	int32_t             offset;
	struct mbuf *       orig_m = m;

	/* find the IP protocol field relative to the start of the packet */
	offset = offsetof(struct ip, ip_p) + sizeof(struct ether_header);
	m = S_mbuf_skip_to_offset(m, &offset);
	if (m == NULL || m->m_len < 1) {
		goto bad_ip_packet;
	}
	data = mtod(m, u_char *) + offset;
	ip_p = *data;

	/* find the IP src relative to the IP protocol */
	if ((m->m_len - offset)
	    >= (int)(IP_SRC_OFFSET + sizeof(struct in_addr) * 2)) {
		/* this should be the normal case */
		ip_src = *(struct in_addr *)(data + IP_SRC_OFFSET);
		ip_dst = *(struct in_addr *)(data + IP_DST_OFFSET);
	} else {
		if (S_mbuf_copy_uint32(m, offset + IP_SRC_OFFSET,
		    (uint32_t *)&ip_src.s_addr)) {
			goto bad_ip_packet;
		}
		if (S_mbuf_copy_uint32(m, offset + IP_DST_OFFSET,
		    (uint32_t *)&ip_dst.s_addr)) {
			goto bad_ip_packet;
		}
	}
	return ntohl(ip_dst.s_addr) ^ ntohl(ip_src.s_addr) ^ ((uint32_t)ip_p);

bad_ip_packet:
	return ether_header_hash(mtod(orig_m, struct ether_header *));
}

#define IP6_ADDRS_LEN   (sizeof(struct in6_addr) * 2)
static uint32_t
ipv6_header_hash(struct mbuf * m)
{
	u_char *            data;
	int                 i;
	int32_t             offset;
	struct mbuf *       orig_m = m;
	uint32_t *          scan;
	uint32_t            val;

	/* find the IP protocol field relative to the start of the packet */
	offset = offsetof(struct ip6_hdr, ip6_src) + sizeof(struct ether_header);
	m = S_mbuf_skip_to_offset(m, &offset);
	if (m == NULL) {
		goto bad_ipv6_packet;
	}
	data = mtod(m, u_char *) + offset;
	val = 0;
	if ((m->m_len - offset) >= (int)IP6_ADDRS_LEN) {
		/* this should be the normal case */
		for (i = 0, scan = (uint32_t *)data;
		    i < (int)(IP6_ADDRS_LEN / sizeof(uint32_t));
		    i++, scan++) {
			val ^= *scan;
		}
	} else {
		for (i = 0; i < (int)(IP6_ADDRS_LEN / sizeof(uint32_t)); i++) {
			uint32_t    tmp;
			if (S_mbuf_copy_uint32(m, offset + i * sizeof(uint32_t),
			    (uint32_t *)&tmp)) {
				goto bad_ipv6_packet;
			}
			val ^= tmp;
		}
	}
	return ntohl(val);

bad_ipv6_packet:
	return ether_header_hash(mtod(orig_m, struct ether_header *));
}

#endif /* BOND_HASH_L3_HEADER */

static void
_mbuf_adjust_pkthdr_and_data(mbuf_t m, int len)
{
	mbuf_setdata(m, mtodo(m, len), mbuf_len(m) - len);
	mbuf_pkthdr_adjustlen(m, -len);
}

static uint32_t
get_packet_hash(mbuf_t m)
{
	uint32_t        flow_hash;

	if (m->m_pkthdr.pkt_flowid != 0) {
		flow_hash = m->m_pkthdr.pkt_flowid;
	} else {
		struct ether_header *   eh_p;

		eh_p = mtod(m, struct ether_header *);
#if BOND_HASH_L3_HEADER
		switch (ntohs(eh_p->ether_type)) {
		case ETHERTYPE_IP:
			flow_hash = ip_header_hash(m);
			break;
		case ETHERTYPE_IPV6:
			flow_hash = ipv6_header_hash(m);
			break;
		default:
			flow_hash = ether_header_hash(eh_p);
			break;
		}
#else /* BOND_HASH_L3_HEADER */
		flow_hash = ether_header_hash(eh_p);
#endif /* BOND_HASH_L3_HEADER */
	}
	return flow_hash;
}

static ifnet_t
bond_get_port_ifp(ifnet_t ifp, uint32_t hash)
{
	uint32_t        port_index;
	ifbond_ref      ifb;
	ifnet_t         port_ifp = NULL;

	bond_lock();
	ifb = ifnet_softc(ifp);
	if (ifb == NULL || ifbond_flags_if_detaching(ifb)
	    || ifb->ifb_distributing_count == 0) {
		goto done;
	}
	port_index = hash % ifb->ifb_distributing_count;
	port_ifp = ifb->ifb_distributing_array[port_index]->po_ifp;

done:
	bond_unlock();
	return port_ifp;
}

static int
bond_output(ifnet_t ifp, mbuf_t m)
{
	struct flowadv  adv = { .code = FADV_SUCCESS };
	int             error = 0;
	int             flags;
	uint32_t        hash;
	uint32_t        len;
	int             log_level;
	ifnet_t         port_ifp;

	VERIFY((m->m_flags & M_PKTHDR) != 0);
	hash = get_packet_hash(m);
	port_ifp = bond_get_port_ifp(ifp, hash);
	if (port_ifp == NULL) {
		m_freem(m);
		goto done;
	}
	if (ifp->if_bpf != NULL) {
		struct ether_header *   eh_p;

		eh_p = mtod(m, struct ether_header *);
		_mbuf_adjust_pkthdr_and_data(m, ETHER_HDR_LEN);
		bond_bpf_tap_out(ifp, m, eh_p);
		_mbuf_adjust_pkthdr_and_data(m, -ETHER_HDR_LEN);
	}
	len = m->m_pkthdr.len;
	if ((m->m_pkthdr.csum_flags & CSUM_VLAN_TAG_VALID) != 0) {
		len += ETHER_VLAN_ENCAP_LEN;
	}
	ifnet_stat_increment_out(ifp, 1, len, 0);
	flags = DLIL_OUTPUT_FLAGS_RAW;
	error = dlil_output(port_ifp, PF_BOND, m, NULL, NULL, flags, &adv);
	log_level = (error != 0) ? LOG_NOTICE : LOG_DEBUG;
	BOND_LOG(log_level, BD_DBGF_OUTPUT, "%s: %s bytes %d, error=%d",
	    ifp->if_xname, port_ifp->if_xname, len, error);
	if (error == 0) {
		if (adv.code == FADV_FLOW_CONTROLLED) {
			error = EQFULL;
		} else if (adv.code == FADV_SUSPENDED) {
			error = EQSUSPENDED;
		}
	}
done:
	return error;
}

static bondport_ref
ifbond_lookup_port(ifbond_ref ifb, struct ifnet * port_ifp)
{
	bondport_ref        p;
	TAILQ_FOREACH(p, &ifb->ifb_port_list, po_port_list) {
		if (p->po_ifp == port_ifp) {
			return p;
		}
	}
	return NULL;
}

static bondport_ref
bond_lookup_port(struct ifnet * port_ifp)
{
	ifbond_ref          ifb;
	bondport_ref        port;

	TAILQ_FOREACH(ifb, &g_bond->ifbond_list, ifb_bond_list) {
		port = ifbond_lookup_port(ifb, port_ifp);
		if (port != NULL) {
			return port;
		}
	}
	return NULL;
}

static void
bond_receive_lacpdu(struct mbuf * m, struct ifnet * port_ifp)
{
	struct ifnet *              bond_ifp = NULL;
	ifbond_ref                  ifb;
	int                         event_code = 0;
	bool                        need_link_update = false;
	bondport_ref                p;

	BOND_LOG(LOG_DEBUG, BD_DBGF_LACP, "%s", port_ifp->if_xname);

	bond_lock();
	if ((ifnet_eflags(port_ifp) & IFEF_BOND) == 0) {
		goto done;
	}
	p = bond_lookup_port(port_ifp);
	if (p == NULL) {
		goto done;
	}
	if (p->po_enabled == 0) {
		goto done;
	}
	ifb = p->po_bond;
	if (ifb->ifb_mode != IF_BOND_MODE_LACP) {
		goto done;
	}
	/*
	 * Work-around for rdar://problem/51372042
	 * Sometimes, the link comes up but the driver doesn't report the
	 * negotiated medium at that time. When we receive an LACPDU packet,
	 * and the medium is unknown, force a link status check. Don't force
	 * the link status check more often than _FORCE_LINK_EVENT_INTERVAL
	 * seconds.
	 */
#define _FORCE_LINK_EVENT_INTERVAL      1
	if (media_type_unknown(&p->po_media_info)) {
		uint64_t        now = net_uptime();

		if ((now - p->po_force_link_event_time) >=
		    _FORCE_LINK_EVENT_INTERVAL) {
			need_link_update = true;
			p->po_force_link_event_time = now;
		}
	}
	bondport_receive_lacpdu(p, (lacpdu_ref)m_mtod_current(m));
	if (ifbond_selection(ifb)) {
		event_code = (ifb->ifb_active_lag == NULL)
		    ? KEV_DL_LINK_OFF
		    : KEV_DL_LINK_ON;
		/* XXX need to take a reference on bond_ifp */
		bond_ifp = ifb->ifb_ifp;
		ifb->ifb_last_link_event = event_code;
	} else {
		event_code = (ifb->ifb_active_lag == NULL)
		    ? KEV_DL_LINK_OFF
		    : KEV_DL_LINK_ON;
		if (event_code != ifb->ifb_last_link_event) {
			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "%s: (receive) generating LINK event",
			    ifb->ifb_name);
			bond_ifp = ifb->ifb_ifp;
			ifb->ifb_last_link_event = event_code;
		}
	}

done:
	bond_unlock();
	if (bond_ifp != NULL) {
		interface_link_event(bond_ifp, event_code);
	}
	m_freem(m);
	if (need_link_update) {
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "simulating link status changed event");
		bond_handle_event(port_ifp, KEV_DL_LINK_ON);
	}
	return;
}

static void
bond_receive_la_marker_pdu(struct mbuf * m, struct ifnet * port_ifp)
{
	la_marker_pdu_ref           marker_p;
	bondport_ref                p;

	BOND_LOG(LOG_DEBUG, BD_DBGF_LACP, "%s", port_ifp->if_xname);

	marker_p = (la_marker_pdu_ref)(m_mtod_current(m) + ETHER_HDR_LEN);
	if (marker_p->lm_marker_tlv_type != LA_MARKER_TLV_TYPE_MARKER) {
		goto failed;
	}
	bond_lock();
	if ((ifnet_eflags(port_ifp) & IFEF_BOND) == 0) {
		bond_unlock();
		goto failed;
	}
	p = bond_lookup_port(port_ifp);
	if (p == NULL || p->po_enabled == 0
	    || p->po_bond->ifb_mode != IF_BOND_MODE_LACP) {
		bond_unlock();
		goto failed;
	}
	/* echo back the same packet as a marker response */
	marker_p->lm_marker_tlv_type = LA_MARKER_TLV_TYPE_MARKER_RESPONSE;
	bondport_slow_proto_transmit(p, (packet_buffer_ref)m);
	bond_unlock();
	return;

failed:
	m_freem(m);
	return;
}

static bool
is_slow_proto_multicast(struct ether_header * eh_p)
{
	return bcmp(eh_p->ether_dhost, &slow_proto_multicast,
	           sizeof(eh_p->ether_dhost)) == 0 &&
	       eh_p->ether_type == htons(IEEE8023AD_SLOW_PROTO_ETHERTYPE);
}

static void
bond_handle_slow_proto_multicast(ifnet_t port_ifp, mbuf_t m)
{
	u_char  subtype = *mtod(m, u_char *);

	if (subtype == IEEE8023AD_SLOW_PROTO_SUBTYPE_LACP) {
		if (m->m_pkthdr.len < LACPDU_MIN_SIZE) {
			BOND_LOG(LOG_DEBUG, BD_DBGF_INPUT,
			    "dropping short LACP frame %d < %d",
			    m->m_pkthdr.len, LACPDU_MIN_SIZE);
			goto discard;
		}
		/* send to lacp */
		if (m->m_len < LACPDU_MIN_SIZE) {
			m = m_pullup(m, LACPDU_MIN_SIZE);
			if (m == NULL) {
				BOND_LOG(LOG_NOTICE, BD_DBGF_INPUT,
				    "m_pullup LACPDU failed");
				return;
			}
		}
		bond_receive_lacpdu(m, port_ifp);
	} else if (subtype == IEEE8023AD_SLOW_PROTO_SUBTYPE_LA_MARKER_PROTOCOL) {
		int         min_size;

		/* restore the ethernet header pointer in the mbuf */
		m->m_pkthdr.len += ETHER_HDR_LEN;
		m->m_data -= ETHER_HDR_LEN;
		m->m_len += ETHER_HDR_LEN;
		min_size = ETHER_HDR_LEN + offsetof(la_marker_pdu, lm_reserved);
		if (m->m_pkthdr.len < min_size) {
			goto discard;
		}
		/* send to lacp */
		if (m->m_len < min_size) {
			m = m_pullup(m, min_size);
			if (m == NULL) {
				BOND_LOG(LOG_NOTICE, BD_DBGF_INPUT,
				    "m_pullup LA_MARKER failed");
				return;
			}
		}
		/* send to marker responder */
		bond_receive_la_marker_pdu(m, port_ifp);
	} else if (subtype == 0
	    || subtype > IEEE8023AD_SLOW_PROTO_SUBTYPE_RESERVED_END) {
		/* invalid subtype, discard the frame */
		goto discard;
	}
	return;

discard:
	m_freem(m);
	return;
}

static void
bond_input_packet_list(ifnet_t port_ifp, mbuf_t list)
{
	ifbond_ref                        ifb;
	struct ifnet *                    ifp;
	bondport_ref                      p;
	struct ifnet_stat_increment_param s;

	/* verify that we're ready to receive the packet list */
	bond_lock();
	if ((ifnet_eflags(port_ifp) & IFEF_BOND) == 0) {
		goto discard;
	}
	p = bond_lookup_port(port_ifp);
	if (p == NULL || bondport_collecting(p) == 0) {
		goto discard;
	}
	ifb = p->po_bond;
	ifp = ifb->ifb_ifp;
	bond_unlock();

	bzero(&s, sizeof(s));

	for (mbuf_t scan = list; scan != NULL; scan = scan->m_nextpkt) {
		struct ether_header *   eh_p;
		void * __single         frame_header;

		/* clear promisc so that the packet doesn't get dropped */
		mbuf_setflags_mask(scan, 0, MBUF_PROMISC);
		s.packets_in++;
		s.bytes_in += scan->m_pkthdr.len + ETHER_HDR_LEN;
		if ((scan->m_pkthdr.csum_flags & CSUM_VLAN_TAG_VALID) != 0) {
			s.bytes_in += ETHER_VLAN_ENCAP_LEN;
		}
		if (ifp->if_bpf != NULL) {
			frame_header = scan->m_pkthdr.pkt_hdr;
			eh_p = (struct ether_header *)frame_header;
			bond_bpf_tap_in(ifp, scan, eh_p);
		}
		scan->m_pkthdr.rcvif = ifp;
	}
	BOND_LOG(LOG_DEBUG, BD_DBGF_INPUT, "%s: %s packets %d bytes %d",
	    ifp->if_xname, port_ifp->if_xname, s.packets_in, s.bytes_in);

	dlil_input_packet_list(ifp, list);
	return;

discard:
	bond_unlock();
	m_freem_list(list);
	return;
}

static int
bond_input(ifnet_t port_ifp, __unused protocol_family_t protocol, mbuf_t m)
{
	struct ether_header *       eh_p;
	void * __single              frame_header;
	mblist                      list;
	mbuf_t                      next_packet = NULL;
	mbuf_t                      scan;

	mblist_init(&list);
	for (scan = m; scan != NULL; scan = next_packet) {
		next_packet = scan->m_nextpkt;
		scan->m_nextpkt = NULL;

		frame_header = scan->m_pkthdr.pkt_hdr;
		eh_p = (struct ether_header *)frame_header;
		if ((scan->m_flags & M_MCAST) != 0 &&
		    is_slow_proto_multicast(eh_p)) {
			/* send up what we have */
			if (list.head != NULL) {
				bond_input_packet_list(port_ifp, list.head);
				mblist_init(&list);
			}
			/* process this multicast */
			bond_handle_slow_proto_multicast(port_ifp, scan);
		} else {
			mblist_append(&list, scan);
		}
	}
	if (list.head != NULL) {
		bond_input_packet_list(port_ifp, list.head);
	}
	return 0;
}

static __inline__ const char *
bondport_get_name(bondport_ref p)
{
	return __unsafe_null_terminated_from_indexable(p->po_name);
}

static __inline__ int
bondport_get_index(bondport_ref p)
{
	return ifnet_index(p->po_ifp);
}

static void
bondport_slow_proto_transmit(bondport_ref p, packet_buffer_ref buf)
{
	struct ether_header *       eh_p;
	int                         error;

	/* packet_buffer_allocate leaves room for ethernet header */
	eh_p = mtod(buf, struct ether_header *);
	bcopy(&slow_proto_multicast, &eh_p->ether_dhost, sizeof(eh_p->ether_dhost));
	bcopy(&p->po_saved_addr, eh_p->ether_shost, sizeof(eh_p->ether_shost));
	eh_p->ether_type = htons(IEEE8023AD_SLOW_PROTO_ETHERTYPE);
	error = ifnet_output_raw(p->po_ifp, PF_BOND, buf);
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LACP,
		    "(%s) failed %d", bondport_get_name(p), error);
	}
	return;
}

static void
bondport_timer_process_func(devtimer_ref timer,
    devtimer_process_func_event event)
{
	bondport_ref        p;

	switch (event) {
	case devtimer_process_func_event_lock:
		bond_lock();
		devtimer_retain(timer);
		break;
	case devtimer_process_func_event_unlock:
		if (devtimer_valid(timer)) {
			/* as long as the devtimer is valid, we can look at arg0 */
			int                 event_code = 0;
			struct ifnet *      bond_ifp = NULL;

			p = (bondport_ref)devtimer_arg0(timer);
			if (ifbond_selection(p->po_bond)) {
				event_code = (p->po_bond->ifb_active_lag == NULL)
				    ? KEV_DL_LINK_OFF
				    : KEV_DL_LINK_ON;
				/* XXX need to take a reference on bond_ifp */
				bond_ifp = p->po_bond->ifb_ifp;
				p->po_bond->ifb_last_link_event = event_code;
			} else {
				event_code = (p->po_bond->ifb_active_lag == NULL)
				    ? KEV_DL_LINK_OFF
				    : KEV_DL_LINK_ON;
				if (event_code !=
				    p->po_bond->ifb_last_link_event) {
					BOND_LOG(LOG_NOTICE, BD_DBGF_LACP,
					    "%s: (timer) generating LINK event",
					    p->po_bond->ifb_name);
					bond_ifp = p->po_bond->ifb_ifp;
					p->po_bond->ifb_last_link_event = event_code;
				}
			}
			devtimer_release(timer);
			bond_unlock();
			if (bond_ifp != NULL) {
				interface_link_event(bond_ifp, event_code);
			}
		} else {
			/* timer is going away */
			devtimer_release(timer);
			bond_unlock();
		}
		break;
	default:
		break;
	}
}

static bondport_ref
bondport_create(struct ifnet * port_ifp, lacp_port_priority priority,
    int active, int short_timeout, int * ret_error)
{
	int                         error = 0;
	bondport_ref                p = NULL;
	lacp_actor_partner_state    s;

	*ret_error = 0;
	p = kalloc_type(struct bondport_s, Z_WAITOK | Z_ZERO | Z_NOFAIL);
	multicast_list_init(&p->po_multicast);
	if ((u_int32_t)snprintf(p->po_name, sizeof(p->po_name), "%s%d",
	    ifnet_name(port_ifp), ifnet_unit(port_ifp))
	    >= sizeof(p->po_name)) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "name too large");
		*ret_error = EINVAL;
		goto failed;
	}
	error = siocgifdevmtu(port_ifp, &p->po_devmtu);
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "SIOCGIFDEVMTU %s failed, %d",
		    bondport_get_name(p), error);
		goto failed;
	}
	/* remember the current interface MTU so it can be restored */
	p->po_devmtu.ifdm_current = ifnet_mtu(port_ifp);
	p->po_ifp = port_ifp;
	p->po_media_info = interface_media_info(port_ifp);
	p->po_current_while_timer = devtimer_create(bondport_timer_process_func, p);
	if (p->po_current_while_timer == NULL) {
		*ret_error = ENOMEM;
		goto failed;
	}
	p->po_periodic_timer = devtimer_create(bondport_timer_process_func, p);
	if (p->po_periodic_timer == NULL) {
		*ret_error = ENOMEM;
		goto failed;
	}
	p->po_wait_while_timer = devtimer_create(bondport_timer_process_func, p);
	if (p->po_wait_while_timer == NULL) {
		*ret_error = ENOMEM;
		goto failed;
	}
	p->po_transmit_timer = devtimer_create(bondport_timer_process_func, p);
	if (p->po_transmit_timer == NULL) {
		*ret_error = ENOMEM;
		goto failed;
	}
	p->po_receive_state = ReceiveState_none;
	p->po_mux_state = MuxState_none;
	p->po_priority = priority;
	s = 0;
	s = lacp_actor_partner_state_set_aggregatable(s);
	if (short_timeout) {
		s = lacp_actor_partner_state_set_short_timeout(s);
	}
	if (active) {
		s = lacp_actor_partner_state_set_active_lacp(s);
	}
	p->po_actor_state = s;
	return p;

failed:
	bondport_free(p);
	return NULL;
}

static void
bondport_start(bondport_ref p)
{
	bondport_receive_machine(p, LAEventStart, NULL);
	bondport_mux_machine(p, LAEventStart, NULL);
	bondport_periodic_transmit_machine(p, LAEventStart, NULL);
	bondport_transmit_machine(p, LAEventStart, NULL);
	return;
}

/*
 * Function: bondport_invalidate_timers
 * Purpose:
 *   Invalidate all of the timers for the bondport.
 */
static void
bondport_invalidate_timers(bondport_ref p)
{
	devtimer_invalidate(p->po_current_while_timer);
	devtimer_invalidate(p->po_periodic_timer);
	devtimer_invalidate(p->po_wait_while_timer);
	devtimer_invalidate(p->po_transmit_timer);
}

/*
 * Function: bondport_cancel_timers
 * Purpose:
 *   Cancel all of the timers for the bondport.
 */
static void
bondport_cancel_timers(bondport_ref p)
{
	devtimer_cancel(p->po_current_while_timer);
	devtimer_cancel(p->po_periodic_timer);
	devtimer_cancel(p->po_wait_while_timer);
	devtimer_cancel(p->po_transmit_timer);
}

static void
bondport_free(bondport_ref p)
{
	multicast_list_remove(&p->po_multicast);
	devtimer_release(p->po_current_while_timer);
	devtimer_release(p->po_periodic_timer);
	devtimer_release(p->po_wait_while_timer);
	devtimer_release(p->po_transmit_timer);
	kfree_type(struct bondport_s, p);
	return;
}

static __inline__ int
bond_device_mtu(struct ifnet * ifp, ifbond_ref ifb)
{
	return ((int)ifnet_mtu(ifp) > ifb->ifb_altmtu)
	       ? (int)ifnet_mtu(ifp) : ifb->ifb_altmtu;
}

static int
bond_add_interface(struct ifnet * ifp, struct ifnet * port_ifp)
{
	u_int32_t                   eflags;
	uint32_t                    control_flags = 0;
	int                         devmtu;
	int                         error = 0;
	int                         event_code = 0;
	int                         first = FALSE;
	ifbond_ref                  ifb;
	bondport_ref *              new_array = NULL;
	bondport_ref *              old_array = NULL;
	bondport_ref                p;
	int                         old_max = 0;
	int                         new_max = 0;

	if (IFNET_IS_INTCOPROC(port_ifp) || IFNET_IS_MANAGEMENT(port_ifp)) {
		return EINVAL;
	}

	/* pre-allocate space for new port */
	p = bondport_create(port_ifp, 0x8000, 1, 0, &error);
	if (p == NULL) {
		return error;
	}
	bond_lock();
	ifb = (ifbond_ref)ifnet_softc(ifp);
	if (ifb == NULL || ifbond_flags_if_detaching(ifb)) {
		bond_unlock();
		bondport_free(p);
		return ifb == NULL ? EOPNOTSUPP : EBUSY;
	}

	/* make sure this interface can handle our current MTU */
	devmtu = bond_device_mtu(ifp, ifb);
	if (devmtu != 0
	    && (devmtu > p->po_devmtu.ifdm_max || devmtu < p->po_devmtu.ifdm_min)) {
		bond_unlock();
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "interface %s doesn't support mtu %d",
		    bondport_get_name(p), devmtu);
		bondport_free(p);
		return EINVAL;
	}

	/* make sure ifb doesn't get de-allocated while we wait */
	ifbond_retain(ifb);

	/* wait for other add or remove to complete */
	ifbond_wait(ifb, __func__);

	if (ifbond_flags_if_detaching(ifb)) {
		/* someone destroyed the bond while we were waiting */
		error = EBUSY;
		goto signal_done;
	}
	if (bond_lookup_port(port_ifp) != NULL) {
		/* port is already part of a bond */
		error = EBUSY;
		goto signal_done;
	}
	if ((ifnet_eflags(port_ifp) & (IFEF_VLAN | IFEF_BOND)) != 0) {
		/* interface already has VLAN's, or is part of bond */
		error = EBUSY;
		goto signal_done;
	}

	/* mark the interface busy */
	eflags = if_set_eflags(port_ifp, IFEF_BOND);
	if ((eflags & IFEF_VLAN) != 0) {
		/* vlan got in ahead of us */
		if_clear_eflags(port_ifp, IFEF_BOND);
		error = EBUSY;
		goto signal_done;
	}

	if (TAILQ_EMPTY(&ifb->ifb_port_list)) {
		ifnet_set_offload(ifp, ifnet_offload(port_ifp));
		ifnet_set_flags(ifp, IFF_RUNNING, IFF_RUNNING);
		if (ifbond_flags_lladdr(ifb) == FALSE) {
			first = TRUE;
		}
	} else {
		ifnet_offload_t         ifp_offload;
		ifnet_offload_t         port_ifp_offload;

		ifp_offload = ifnet_offload(ifp);
		port_ifp_offload = ifnet_offload(port_ifp);
		if (ifp_offload != port_ifp_offload) {
			ifnet_offload_t     offload;

			offload = ifp_offload & port_ifp_offload;
			BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
			    "(%s, %s) hwassist values don't match 0x%x != 0x%x,"
			    " using 0x%x instead",
			    ifb->ifb_name, bondport_get_name(p),
			    ifp_offload, port_ifp_offload, offload);
			/*
			 * XXX
			 * if the bond has VLAN's, we can't simply change the hwassist
			 * field behind its back: this needs work
			 */
			ifnet_set_offload(ifp, offload);
		}
	}
	p->po_bond = ifb;

	/* remember the port's ethernet address so it can be restored */
	ether_addr_copy(p->po_saved_addr.octet,
	    (uint8_t *__indexable)IF_LLADDR(port_ifp));

	/* add it to the list of ports */
	TAILQ_INSERT_TAIL(&ifb->ifb_port_list, p, po_port_list);
	ifb->ifb_port_count++;

	bond_unlock();


	/* first port added to bond determines bond's ethernet address */
	if (first) {
		ifnet_set_lladdr_and_type(ifp, IF_LLADDR(port_ifp), ETHER_ADDR_LEN,
		    IFT_ETHER);
	}
	uint32_bit_set(&control_flags, PORT_CONTROL_FLAGS_IN_LIST);

	/* allocate a larger distributing array */
	new_max = ifb->ifb_port_count;
	new_array = kalloc_type(bondport_ref, new_max, Z_WAITOK);
	if (new_array == NULL) {
		error = ENOMEM;
		goto failed;
	}

	/* attach our BOND "protocol" to the interface */
	error = bond_attach_protocol(port_ifp);
	if (error) {
		goto failed;
	}
	uint32_bit_set(&control_flags, PORT_CONTROL_FLAGS_PROTO_ATTACHED);

	/* set the interface MTU */
	devmtu = bond_device_mtu(ifp, ifb);
	error = siocsifmtu(port_ifp, devmtu);
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "(%s, %s): SIOCSIFMTU %d failed %d",
		    ifb->ifb_name, bondport_get_name(p), devmtu, error);
		goto failed;
	}
	uint32_bit_set(&control_flags, PORT_CONTROL_FLAGS_MTU_SET);

	/* program the port with our multicast addresses */
	error = multicast_list_program(&p->po_multicast, ifp, port_ifp);
	if (error) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "(%s, %s): multicast_list_program failed %d",
		    ifb->ifb_name, bondport_get_name(p), error);
		goto failed;
	}

	/* mark the interface up */
	ifnet_set_flags(port_ifp, IFF_UP, IFF_UP);

	error = ifnet_ioctl(port_ifp, 0, SIOCSIFFLAGS, NULL);
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "(%s, %s): SIOCSIFFLAGS failed %d",
		    ifb->ifb_name, bondport_get_name(p), error);
		goto failed;
	}

	/* re-program the port's ethernet address */
	error = if_siflladdr(port_ifp,
	    (const struct ether_addr *)IF_LLADDR(ifp));
	if (error == 0) {
		if (memcmp(IF_LLADDR(ifp), IF_LLADDR(port_ifp), ETHER_ADDR_LEN)
		    != 0) {
			/* it lied, it really doesn't support setting lladdr */
			error = EOPNOTSUPP;
		}
	}
	if (error != 0) {
		/* port doesn't support setting the link address */
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "(%s, %s): if_siflladdr failed %d",
		    ifb->ifb_name, bondport_get_name(p), error);
		error = ifnet_set_promiscuous(port_ifp, 1);
		if (error != 0) {
			/* port doesn't support setting promiscuous mode */
			BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
			    "(%s, %s): set promiscuous failed %d",
			    ifb->ifb_name, bondport_get_name(p), error);
			goto failed;
		}
		uint32_bit_set(&control_flags,
		    PORT_CONTROL_FLAGS_PROMISCUOUS_SET);
	} else {
		uint32_bit_set(&control_flags,
		    PORT_CONTROL_FLAGS_LLADDR_SET);
	}

	/* if we're in promiscuous mode, enable that as well */
	if (ifbond_flags_promisc(ifb)) {
		error = ifnet_set_promiscuous(port_ifp, 1);
		if (error != 0) {
			/* port doesn't support setting promiscuous mode */
			BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
			    "(%s, %s): set promiscuous failed %d",
			    ifb->ifb_name, bondport_get_name(p), error);
			goto failed;
		}
		uint32_bit_set(&control_flags,
		    PORT_CONTROL_FLAGS_BOND_PROMISCUOUS_SET);
	}

	bond_lock();

	/* no failures past this point */
	p->po_enabled = 1;
	p->po_control_flags = control_flags;

	/* copy the contents of the existing distributing array */
	if (ifb->ifb_distributing_count) {
		bcopy(ifb->ifb_distributing_array, new_array,
		    sizeof(*new_array) * ifb->ifb_distributing_count);
	}
	old_array = ifb->ifb_distributing_array;
	old_max = ifb->ifb_distributing_max;
	ifb->ifb_distributing_array = new_array;
	ifb->ifb_distributing_max = new_max;

	if (ifb->ifb_mode == IF_BOND_MODE_LACP) {
		bondport_start(p);

		/* check if we need to generate a link status event */
		if (ifbond_selection(ifb)) {
			event_code = (ifb->ifb_active_lag == NULL)
			    ? KEV_DL_LINK_OFF
			    : KEV_DL_LINK_ON;
			ifb->ifb_last_link_event = event_code;
		}
	} else {
		/* are we adding the first distributing interface? */
		if (media_active(&p->po_media_info)) {
			if (ifb->ifb_distributing_count == 0) {
				ifb->ifb_last_link_event = event_code = KEV_DL_LINK_ON;
			}
			bondport_enable_distributing(p);
		} else {
			bondport_disable_distributing(p);
		}
	}

	/* clear the busy state, and wakeup anyone waiting */
	ifbond_signal(ifb, __func__);
	bond_unlock();
	if (event_code != 0) {
		interface_link_event(ifp, event_code);
	}
	kfree_type(bondport_ref, old_max, old_array);
	return 0;

failed:
	bond_assert_lock_not_held();

	/* if this was the first port to be added, clear our address */
	if (first) {
		ifnet_set_lladdr_and_type(ifp, NULL, 0, IFT_IEEE8023ADLAG);
	}

	kfree_type(bondport_ref, new_max, new_array);
	if (uint32_bit_is_set(control_flags,
	    PORT_CONTROL_FLAGS_LLADDR_SET)) {
		int     error1;

		error1 = if_siflladdr(port_ifp, &p->po_saved_addr);
		if (error1 != 0) {
			BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
			    "(%s, %s): if_siflladdr restore failed %d",
			    ifb->ifb_name, bondport_get_name(p), error1);
		}
	}
	if (uint32_bit_is_set(control_flags,
	    PORT_CONTROL_FLAGS_PROMISCUOUS_SET)) {
		int     error1;

		error1 = ifnet_set_promiscuous(port_ifp, 0);
		if (error1 != 0) {
			BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
			    "(%s, %s): promiscous mode disable failed %d",
			    ifb->ifb_name, bondport_get_name(p), error1);
		}
	}
	if (uint32_bit_is_set(control_flags,
	    PORT_CONTROL_FLAGS_PROTO_ATTACHED)) {
		(void)bond_detach_protocol(port_ifp);
	}
	if (uint32_bit_is_set(control_flags,
	    PORT_CONTROL_FLAGS_MTU_SET)) {
		int error1;

		error1 = siocsifmtu(port_ifp, p->po_devmtu.ifdm_current);
		if (error1 != 0) {
			BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
			    "(%s, %s): SIOCSIFMTU %d failed %d",
			    ifb->ifb_name, bondport_get_name(p),
			    p->po_devmtu.ifdm_current, error1);
		}
	}
	bond_lock();
	if (uint32_bit_is_set(control_flags,
	    PORT_CONTROL_FLAGS_IN_LIST)) {
		TAILQ_REMOVE(&ifb->ifb_port_list, p, po_port_list);
		ifb->ifb_port_count--;
	}
	if_clear_eflags(ifp, IFEF_BOND);
	if (TAILQ_EMPTY(&ifb->ifb_port_list)) {
		ifb->ifb_altmtu = 0;
		ifnet_set_mtu(ifp, ETHERMTU);
		ifnet_set_offload(ifp, 0);
	}

signal_done:
	ifbond_signal(ifb, __func__);
	bond_unlock();
	ifbond_release(ifb);
	bondport_free(p);
	return error;
}

static int
bond_remove_interface(ifbond_ref ifb, struct ifnet * port_ifp)
{
	int                         active_lag = 0;
	int                         error = 0;
	int                         event_code = 0;
	bondport_ref                head_port;
	struct ifnet *              ifp;
	int                         last = FALSE;
	int                         new_link_address = FALSE;
	bondport_ref                p;
	lacp_actor_partner_state    s;
	int                         was_distributing;

	bond_assert_lock_held();

	ifbond_retain(ifb);
	ifbond_wait(ifb, "bond_remove_interface");

	p = ifbond_lookup_port(ifb, port_ifp);
	if (p == NULL) {
		error = ENXIO;
		/* it got removed by another thread */
		goto signal_done;
	}

	/* de-select it and remove it from the lists */
	was_distributing = bondport_flags_distributing(p);
	bondport_disable_distributing(p);
	if (ifb->ifb_mode == IF_BOND_MODE_LACP) {
		bondport_set_selected(p, SelectedState_UNSELECTED);
		active_lag = bondport_remove_from_LAG(p);
		/* invalidate timers here while holding the bond_lock */
		bondport_invalidate_timers(p);

		/* announce that we're Individual now */
		s = p->po_actor_state;
		s = lacp_actor_partner_state_set_individual(s);
		s = lacp_actor_partner_state_set_not_collecting(s);
		s = lacp_actor_partner_state_set_not_distributing(s);
		s = lacp_actor_partner_state_set_out_of_sync(s);
		p->po_actor_state = s;
		bondport_flags_set_ntt(p);
	}

	TAILQ_REMOVE(&ifb->ifb_port_list, p, po_port_list);
	ifb->ifb_port_count--;

	ifp = ifb->ifb_ifp;
	head_port = TAILQ_FIRST(&ifb->ifb_port_list);
	if (head_port == NULL) {
		ifnet_set_flags(ifp, 0, IFF_RUNNING);
		if (ifbond_flags_lladdr(ifb) == FALSE) {
			last = TRUE;
		}
		ifnet_set_offload(ifp, 0);
		ifnet_set_mtu(ifp, ETHERMTU);
		ifb->ifb_altmtu = 0;
	} else if (ifbond_flags_lladdr(ifb) == FALSE
	    && bcmp(&p->po_saved_addr, IF_LLADDR(ifp),
	    ETHER_ADDR_LEN) == 0) {
		new_link_address = TRUE;
	}
	/* check if we need to generate a link status event */
	if (ifb->ifb_mode == IF_BOND_MODE_LACP) {
		if (ifbond_selection(ifb) || active_lag) {
			event_code = (ifb->ifb_active_lag == NULL)
			    ? KEV_DL_LINK_OFF
			    : KEV_DL_LINK_ON;
			ifb->ifb_last_link_event = event_code;
		}
		bondport_transmit_machine(p, LAEventStart,
		    TRANSMIT_MACHINE_TX_IMMEDIATE);
	} else {
		/* are we removing the last distributing interface? */
		if (was_distributing && ifb->ifb_distributing_count == 0) {
			ifb->ifb_last_link_event = event_code = KEV_DL_LINK_OFF;
		}
	}
	bond_unlock();

	if (last) {
		ifnet_set_lladdr_and_type(ifp, NULL, 0, IFT_IEEE8023ADLAG);
	} else if (new_link_address) {
		struct ifnet *  scan_ifp;
		bondport_ref    scan_port;

		/* ifbond_wait() allows port list traversal without holding the lock */

		/* this port gave the bond its ethernet address, switch to new one */
		ifnet_set_lladdr_and_type(ifp,
		    &head_port->po_saved_addr, ETHER_ADDR_LEN,
		    IFT_ETHER);

		/* re-program each port with the new link address */
		TAILQ_FOREACH(scan_port, &ifb->ifb_port_list, po_port_list) {
			scan_ifp = scan_port->po_ifp;

			if (!uint32_bit_is_set(scan_port->po_control_flags,
			    PORT_CONTROL_FLAGS_LLADDR_SET)) {
				/* port doesn't support setting lladdr */
				continue;
			}
			error = if_siflladdr(scan_ifp,
			    (const struct ether_addr *) IF_LLADDR(ifp));
			if (error != 0) {
				BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
				    "(%s, %s): if_siflladdr (%s) failed %d",
				    ifb->ifb_name, bondport_get_name(p),
				    bondport_get_name(scan_port), error);
			}
		}
	}

	/* restore the port's ethernet address */
	if (uint32_bit_is_set(p->po_control_flags,
	    PORT_CONTROL_FLAGS_LLADDR_SET)) {
		error = if_siflladdr(port_ifp, &p->po_saved_addr);
		if (error != 0) {
			BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
			    "(%s, %s): if_siflladdr failed %d",
			    ifb->ifb_name, bondport_get_name(p), error);
		}
	}

	/* disable promiscous mode (if we enabled it) */
	if (uint32_bit_is_set(p->po_control_flags,
	    PORT_CONTROL_FLAGS_PROMISCUOUS_SET)) {
		error = ifnet_set_promiscuous(port_ifp, 0);
		if (error != 0) {
			BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
			    "(%s, %s): disable promiscuous failed %d",
			    ifb->ifb_name, bondport_get_name(p), error);
		}
	}

	/* disable promiscous mode from bond (if we enabled it) */
	if (uint32_bit_is_set(p->po_control_flags,
	    PORT_CONTROL_FLAGS_BOND_PROMISCUOUS_SET)) {
		error = ifnet_set_promiscuous(port_ifp, 0);
		if (error != 0) {
			BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
			    "(%s, %s): disable promiscuous failed %d",
			    ifb->ifb_name, bondport_get_name(p), error);
		}
	}

	/* restore the port's MTU */
	error = siocsifmtu(port_ifp, p->po_devmtu.ifdm_current);
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "(%s, %s): SIOCSIFMTU %d failed %d",
		    ifb->ifb_name, bondport_get_name(p),
		    p->po_devmtu.ifdm_current, error);
	}

	/* remove the bond "protocol" */
	bond_detach_protocol(port_ifp);

	/* generate link event */
	if (event_code != 0) {
		interface_link_event(ifp, event_code);
	}

	bond_lock();
	bondport_free(p);
	if_clear_eflags(port_ifp, IFEF_BOND);
	/* release this bondport's reference to the ifbond */
	ifbond_release(ifb);

signal_done:
	ifbond_signal(ifb, __func__);
	ifbond_release(ifb);
	return error;
}

static void
bond_set_lacp_mode(ifbond_ref ifb)
{
	bondport_ref                p;

	TAILQ_FOREACH(p, &ifb->ifb_port_list, po_port_list) {
		bondport_disable_distributing(p);
		bondport_start(p);
	}
	return;
}

static void
bond_set_static_mode(ifbond_ref ifb)
{
	bondport_ref                p;
	lacp_actor_partner_state    s;

	TAILQ_FOREACH(p, &ifb->ifb_port_list, po_port_list) {
		bondport_disable_distributing(p);
		bondport_set_selected(p, SelectedState_UNSELECTED);
		(void)bondport_remove_from_LAG(p);
		bondport_cancel_timers(p);

		/* announce that we're Individual now */
		s = p->po_actor_state;
		s = lacp_actor_partner_state_set_individual(s);
		s = lacp_actor_partner_state_set_not_collecting(s);
		s = lacp_actor_partner_state_set_not_distributing(s);
		s = lacp_actor_partner_state_set_out_of_sync(s);
		p->po_actor_state = s;
		bondport_flags_set_ntt(p);
		bondport_transmit_machine(p, LAEventStart,
		    TRANSMIT_MACHINE_TX_IMMEDIATE);
		/* clear state */
		p->po_actor_state = 0;
		bzero(&p->po_partner_state, sizeof(p->po_partner_state));

		if (media_active(&p->po_media_info)) {
			bondport_enable_distributing(p);
		} else {
			bondport_disable_distributing(p);
		}
	}
	return;
}

static int
bond_set_mode(struct ifnet * ifp, int mode)
{
	int                         error = 0;
	int                         event_code = 0;
	ifbond_ref                  ifb;

	bond_lock();
	ifb = (ifbond_ref)ifnet_softc(ifp);
	if (ifb == NULL || ifbond_flags_if_detaching(ifb)) {
		bond_unlock();
		return (ifb == NULL) ? EOPNOTSUPP : EBUSY;
	}
	if (ifb->ifb_mode == mode) {
		bond_unlock();
		return 0;
	}

	ifbond_retain(ifb);
	ifbond_wait(ifb, "bond_set_mode");

	/* verify (again) that the mode is actually different */
	if (ifb->ifb_mode == mode) {
		/* nothing to do */
		goto signal_done;
	}

	ifb->ifb_mode = mode;
	if (mode == IF_BOND_MODE_LACP) {
		bond_set_lacp_mode(ifb);

		/* check if we need to generate a link status event */
		if (ifbond_selection(ifb)) {
			event_code = (ifb->ifb_active_lag == NULL)
			    ? KEV_DL_LINK_OFF
			    : KEV_DL_LINK_ON;
		}
	} else {
		bond_set_static_mode(ifb);
		event_code = (ifb->ifb_distributing_count == 0)
		    ? KEV_DL_LINK_OFF
		    : KEV_DL_LINK_ON;
	}
	ifb->ifb_last_link_event = event_code;

signal_done:
	ifbond_signal(ifb, __func__);
	bond_unlock();
	ifbond_release(ifb);

	if (event_code != 0) {
		interface_link_event(ifp, event_code);
	}
	return error;
}

static int
bond_get_status(ifbond_ref ifb, struct if_bond_req * ibr_p, user_addr_t datap)
{
	int                         count;
	user_addr_t                 dst;
	int                         error = 0;
	struct if_bond_status_req * ibsr;
	struct if_bond_status       ibs;
	bondport_ref                port;

	ibsr = &(ibr_p->ibr_ibru.ibru_status);
	if (ibsr->ibsr_version != IF_BOND_STATUS_REQ_VERSION) {
		return EINVAL;
	}
	ibsr->ibsr_key = ifb->ifb_key;
	ibsr->ibsr_mode = ifb->ifb_mode;
	ibsr->ibsr_total = ifb->ifb_port_count;
	dst = proc_is64bit(current_proc())
	    ? ibsr->ibsr_ibsru.ibsru_buffer64
	    : CAST_USER_ADDR_T(ibsr->ibsr_ibsru.ibsru_buffer);
	if (dst == USER_ADDR_NULL) {
		/* just want to know how many there are */
		goto done;
	}
	if (ibsr->ibsr_count < 0) {
		return EINVAL;
	}
	count = (ifb->ifb_port_count < ibsr->ibsr_count)
	    ? ifb->ifb_port_count : ibsr->ibsr_count;
	TAILQ_FOREACH(port, &ifb->ifb_port_list, po_port_list) {
		struct if_bond_partner_state *  ibps_p;
		partner_state_ref               ps;

		if (count == 0) {
			break;
		}
		bzero(&ibs, sizeof(ibs));
		strbufcpy(ibs.ibs_if_name, port->po_name);
		ibs.ibs_port_priority = port->po_priority;
		if (ifb->ifb_mode == IF_BOND_MODE_LACP) {
			ibs.ibs_state = port->po_actor_state;
			ibs.ibs_selected_state = port->po_selected;
			ps = &port->po_partner_state;
			ibps_p = &ibs.ibs_partner_state;
			ibps_p->ibps_system = ps->ps_lag_info.li_system;
			ibps_p->ibps_system_priority = ps->ps_lag_info.li_system_priority;
			ibps_p->ibps_key = ps->ps_lag_info.li_key;
			ibps_p->ibps_port = ps->ps_port;
			ibps_p->ibps_port_priority = ps->ps_port_priority;
			ibps_p->ibps_state = ps->ps_state;
		} else {
			/* fake the selected information */
			ibs.ibs_selected_state = bondport_flags_distributing(port)
			    ? SelectedState_SELECTED : SelectedState_UNSELECTED;
		}
		error = copyout(&ibs, dst, sizeof(ibs));
		if (error != 0) {
			break;
		}
		dst += sizeof(ibs);
		count--;
	}

done:
	if (error == 0) {
		error = copyout(ibr_p, datap, sizeof(*ibr_p));
	} else {
		(void)copyout(ibr_p, datap, sizeof(*ibr_p));
	}
	return error;
}

static int
bond_set_promisc(struct ifnet * ifp)
{
	int                 error = 0;
	ifbond_ref          ifb;
	bool                is_promisc;
	bondport_ref        p;
	int                 val;

	is_promisc = (ifnet_flags(ifp) & IFF_PROMISC) != 0;

	/* determine whether promiscuous state needs to be changed */
	bond_lock();
	ifb = (ifbond_ref)ifnet_softc(ifp);
	if (ifb == NULL) {
		bond_unlock();
		error = EBUSY;
		goto done;
	}
	if (is_promisc == ifbond_flags_promisc(ifb)) {
		/* already in the right state */
		bond_unlock();
		goto done;
	}
	ifbond_retain(ifb);
	ifbond_wait(ifb, __func__);
	if (ifbond_flags_if_detaching(ifb)) {
		/* someone destroyed the bond while we were waiting */
		error = EBUSY;
		goto signal_done;
	}
	bond_unlock();

	/* update the promiscuous state of each memeber */
	val = is_promisc ? 1 : 0;
	TAILQ_FOREACH(p, &ifb->ifb_port_list, po_port_list) {
		struct ifnet *  port_ifp = p->po_ifp;
		bool            port_is_promisc;

		port_is_promisc = uint32_bit_is_set(p->po_control_flags,
		    PORT_CONTROL_FLAGS_BOND_PROMISCUOUS_SET);
		if (port_is_promisc == is_promisc) {
			/* already in the right state */
			continue;
		}
		error = ifnet_set_promiscuous(port_ifp, val);
		if (error != 0) {
			BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
			    "%s: ifnet_set_promiscuous(%s, %d): failed %d",
			    ifb->ifb_name, port_ifp->if_xname, val, error);
			continue;
		}
		BOND_LOG(LOG_DEBUG, BD_DBGF_LIFECYCLE,
		    "%s: ifnet_set_promiscuous(%s, %d): succeeded",
		    ifb->ifb_name, port_ifp->if_xname, val);
		if (is_promisc) {
			/* remember that we set it */
			uint32_bit_set(&p->po_control_flags,
			    PORT_CONTROL_FLAGS_BOND_PROMISCUOUS_SET);
		} else {
			uint32_bit_clear(&p->po_control_flags,
			    PORT_CONTROL_FLAGS_BOND_PROMISCUOUS_SET);
		}
	}

	/* assume that updating promiscuous state succeeded */
	error = 0;
	bond_lock();

	/* update our internal state */
	if (is_promisc) {
		ifbond_flags_set_promisc(ifb);
	} else {
		ifbond_flags_clear_promisc(ifb);
	}

signal_done:
	ifbond_signal(ifb, __func__);
	bond_unlock();
	ifbond_release(ifb);

done:
	return error;
}

static void
bond_get_mtu_values(ifbond_ref ifb, int * ret_min, int * ret_max)
{
	int                         mtu_min = 0;
	int                         mtu_max = 0;
	bondport_ref                p;

	if (TAILQ_FIRST(&ifb->ifb_port_list) != NULL) {
		mtu_min = IF_MINMTU;
	}
	TAILQ_FOREACH(p, &ifb->ifb_port_list, po_port_list) {
		struct ifdevmtu *       devmtu_p = &p->po_devmtu;

		if (devmtu_p->ifdm_min > mtu_min) {
			mtu_min = devmtu_p->ifdm_min;
		}
		if (mtu_max == 0 || devmtu_p->ifdm_max < mtu_max) {
			mtu_max = devmtu_p->ifdm_max;
		}
	}
	*ret_min = mtu_min;
	*ret_max = mtu_max;
	return;
}

static int
bond_set_mtu_on_ports(ifbond_ref ifb, int mtu)
{
	int                         error = 0;
	bondport_ref                p;

	TAILQ_FOREACH(p, &ifb->ifb_port_list, po_port_list) {
		error = siocsifmtu(p->po_ifp, mtu);
		if (error != 0) {
			BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
			    "%s: SIOCSIFMTU %s failed, %d",
			    ifb->ifb_name, bondport_get_name(p), error);
			break;
		}
	}
	return error;
}

static int
bond_set_mtu(struct ifnet * ifp, int mtu, int isdevmtu)
{
	int                 error = 0;
	ifbond_ref          ifb;
	int                 mtu_min;
	int                 mtu_max;
	int                 new_max;
	int                 old_max;

	bond_lock();
	ifb = (ifbond_ref)ifnet_softc(ifp);
	if (ifb == NULL || ifbond_flags_if_detaching(ifb)) {
		error = (ifb == NULL) ? EOPNOTSUPP : EBUSY;
		goto done;
	}
	ifbond_retain(ifb);
	ifbond_wait(ifb, "bond_set_mtu");

	/* check again */
	if (ifnet_softc(ifp) == NULL || ifbond_flags_if_detaching(ifb)) {
		error = EBUSY;
		goto signal_done;
	}
	bond_get_mtu_values(ifb, &mtu_min, &mtu_max);
	if (mtu > mtu_max) {
		error = EINVAL;
		goto signal_done;
	}
	if (mtu < mtu_min && (isdevmtu == 0 || mtu != 0)) {
		/* allow SIOCSIFALTMTU to set the mtu to 0 */
		error = EINVAL;
		goto signal_done;
	}
	if (isdevmtu) {
		new_max = (mtu > (int)ifnet_mtu(ifp)) ? mtu : (int)ifnet_mtu(ifp);
	} else {
		new_max = (mtu > ifb->ifb_altmtu) ? mtu : ifb->ifb_altmtu;
	}
	old_max = ((int)ifnet_mtu(ifp) > ifb->ifb_altmtu)
	    ? (int)ifnet_mtu(ifp) : ifb->ifb_altmtu;
	if (new_max != old_max) {
		/* we can safely walk the list of port without the lock held */
		bond_unlock();
		error = bond_set_mtu_on_ports(ifb, new_max);
		if (error != 0) {
			/* try our best to back out of it */
			(void)bond_set_mtu_on_ports(ifb, old_max);
		}
		bond_lock();
	}
	if (error == 0) {
		if (isdevmtu) {
			ifb->ifb_altmtu = mtu;
		} else {
			ifnet_set_mtu(ifp, mtu);
		}
	}

signal_done:
	ifbond_signal(ifb, __func__);
	ifbond_release(ifb);

done:
	bond_unlock();
	return error;
}

static int
bond_ioctl(struct ifnet *ifp, u_long cmd, void * data)
{
	int                 error = 0;
	struct if_bond_req  ibr;
	struct ifaddr *     ifa;
	ifbond_ref          ifb;
	struct ifreq *      ifr;
	struct ifmediareq32 * ifmr;
	struct ifnet *      port_ifp = NULL;
	user_addr_t         user_addr;

	if (ifnet_type(ifp) != IFT_IEEE8023ADLAG) {
		return EOPNOTSUPP;
	}
	ifr = (struct ifreq *)data;
	ifa = (struct ifaddr *)data;

	switch (cmd) {
	case SIOCSIFADDR:
		ifnet_set_flags(ifp, IFF_UP, IFF_UP);
		break;

	case SIOCGIFMEDIA32:
	case SIOCGIFMEDIA64:
		bond_lock();
		ifb = (ifbond_ref)ifnet_softc(ifp);
		if (ifb == NULL || ifbond_flags_if_detaching(ifb)) {
			bond_unlock();
			return ifb == NULL ? EOPNOTSUPP : EBUSY;
		}
		ifmr = (struct ifmediareq32 *)data;
		ifmr->ifm_current = IFM_ETHER;
		ifmr->ifm_mask = 0;
		ifmr->ifm_status = IFM_AVALID;
		ifmr->ifm_active = IFM_ETHER;
		ifmr->ifm_count = 1;
		if (ifb->ifb_mode == IF_BOND_MODE_LACP) {
			if (ifb->ifb_active_lag != NULL) {
				ifmr->ifm_active = ifb->ifb_active_lag->lag_active_media;
				ifmr->ifm_status |= IFM_ACTIVE;
			}
		} else if (ifb->ifb_distributing_count > 0) {
			ifmr->ifm_active
			        = ifb->ifb_distributing_array[0]->po_media_info.mi_active;
			ifmr->ifm_status |= IFM_ACTIVE;
		}
		bond_unlock();
		user_addr = (cmd == SIOCGIFMEDIA64) ?
		    ((struct ifmediareq64 *)data)->ifmu_ulist :
		    CAST_USER_ADDR_T(((struct ifmediareq32 *)data)->ifmu_ulist);
		if (user_addr != USER_ADDR_NULL) {
			error = copyout(&ifmr->ifm_current,
			    user_addr,
			    sizeof(int));
		}
		break;

	case SIOCSIFMEDIA:
		/* XXX send the SIFMEDIA to all children?  Or force autoselect? */
		error = EINVAL;
		break;

	case SIOCGIFDEVMTU:
		bond_lock();
		ifb = (ifbond_ref)ifnet_softc(ifp);
		if (ifb == NULL || ifbond_flags_if_detaching(ifb)) {
			bond_unlock();
			error = (ifb == NULL) ? EOPNOTSUPP : EBUSY;
			break;
		}
		ifr->ifr_devmtu.ifdm_current = bond_device_mtu(ifp, ifb);
		bond_get_mtu_values(ifb, &ifr->ifr_devmtu.ifdm_min,
		    &ifr->ifr_devmtu.ifdm_max);
		bond_unlock();
		break;

	case SIOCGIFALTMTU:
		bond_lock();
		ifb = (ifbond_ref)ifnet_softc(ifp);
		if (ifb == NULL || ifbond_flags_if_detaching(ifb)) {
			bond_unlock();
			error = (ifb == NULL) ? EOPNOTSUPP : EBUSY;
			break;
		}
		ifr->ifr_mtu = ifb->ifb_altmtu;
		bond_unlock();
		break;

	case SIOCSIFALTMTU:
		error = bond_set_mtu(ifp, ifr->ifr_mtu, 1);
		break;

	case SIOCSIFMTU:
		error = bond_set_mtu(ifp, ifr->ifr_mtu, 0);
		break;

	case SIOCSIFBOND:
		user_addr = proc_is64bit(current_proc())
		    ? ifr->ifr_data64 : CAST_USER_ADDR_T(ifr->ifr_data);
		error = copyin(user_addr, &ibr, sizeof(ibr));
		if (error) {
			break;
		}
		switch (ibr.ibr_op) {
		case IF_BOND_OP_ADD_INTERFACE:
		case IF_BOND_OP_REMOVE_INTERFACE:
			port_ifp = ifunit(__unsafe_null_terminated_from_indexable(ibr.ibr_ibru.ibru_if_name));
			if (port_ifp == NULL) {
				error = ENXIO;
				break;
			}
			if (ifnet_type(port_ifp) != IFT_ETHER) {
				error = EPROTONOSUPPORT;
				break;
			}
			break;
		case IF_BOND_OP_SET_MODE:
			break;
		default:
			error = EOPNOTSUPP;
			break;
		}
		if (error != 0) {
			break;
		}
		switch (ibr.ibr_op) {
		case IF_BOND_OP_ADD_INTERFACE:
			error = bond_add_interface(ifp, port_ifp);
			break;
		case IF_BOND_OP_REMOVE_INTERFACE:
			bond_lock();
			ifb = (ifbond_ref)ifnet_softc(ifp);
			if (ifb == NULL || ifbond_flags_if_detaching(ifb)) {
				bond_unlock();
				return ifb == NULL ? EOPNOTSUPP : EBUSY;
			}
			error = bond_remove_interface(ifb, port_ifp);
			bond_unlock();
			break;
		case IF_BOND_OP_SET_MODE:
			switch (ibr.ibr_ibru.ibru_int_val) {
			case IF_BOND_MODE_LACP:
			case IF_BOND_MODE_STATIC:
				break;
			default:
				error = EINVAL;
				break;
			}
			if (error != 0) {
				break;
			}
			error = bond_set_mode(ifp, ibr.ibr_ibru.ibru_int_val);
			break;
		}
		break; /* SIOCSIFBOND */

	case SIOCGIFBOND:
		user_addr = proc_is64bit(current_proc())
		    ? ifr->ifr_data64 : CAST_USER_ADDR_T(ifr->ifr_data);
		error = copyin(user_addr, &ibr, sizeof(ibr));
		if (error) {
			break;
		}
		switch (ibr.ibr_op) {
		case IF_BOND_OP_GET_STATUS:
			break;
		default:
			error = EOPNOTSUPP;
			break;
		}
		if (error != 0) {
			break;
		}
		bond_lock();
		ifb = (ifbond_ref)ifnet_softc(ifp);
		if (ifb == NULL || ifbond_flags_if_detaching(ifb)) {
			bond_unlock();
			return ifb == NULL ? EOPNOTSUPP : EBUSY;
		}
		switch (ibr.ibr_op) {
		case IF_BOND_OP_GET_STATUS:
			error = bond_get_status(ifb, &ibr, user_addr);
			break;
		}
		bond_unlock();
		break; /* SIOCGIFBOND */

	case SIOCSIFLLADDR:
		error = EOPNOTSUPP;
		break;

	case SIOCSIFFLAGS:
		/* enable promiscuous mode on members */
		error = bond_set_promisc(ifp);
		break;

	case SIOCADDMULTI:
	case SIOCDELMULTI:
		error = bond_setmulti(ifp);
		break;
	default:
		error = EOPNOTSUPP;
	}
	return error;
}

static void
bond_if_free(struct ifnet * ifp)
{
	ifbond_ref  ifb;

	if (ifp == NULL) {
		return;
	}
	bond_lock();
	ifb = (ifbond_ref)ifnet_softc(ifp);
	if (ifb == NULL) {
		bond_unlock();
		return;
	}
	ifbond_release(ifb);
	bond_unlock();
	ifnet_release(ifp);
	return;
}

static void
bond_handle_event(struct ifnet * port_ifp, int event_code)
{
	struct ifnet *      bond_ifp = NULL;
	ifbond_ref          ifb;
	int                 old_distributing_count;
	bondport_ref        p;
	struct media_info   media_info = { .mi_active = 0, .mi_status = 0 };

	switch (event_code) {
	case KEV_DL_IF_DETACHED:
	case KEV_DL_IF_DETACHING:
		break;
	case KEV_DL_LINK_OFF:
	case KEV_DL_LINK_ON:
		media_info = interface_media_info(port_ifp);
		break;
	default:
		return;
	}
	bond_lock();
	p = bond_lookup_port(port_ifp);
	if (p == NULL) {
		bond_unlock();
		return;
	}
	ifb = p->po_bond;
	old_distributing_count = ifb->ifb_distributing_count;
	switch (event_code) {
	case KEV_DL_IF_DETACHED:
	case KEV_DL_IF_DETACHING:
		bond_remove_interface(ifb, p->po_ifp);
		break;
	case KEV_DL_LINK_OFF:
	case KEV_DL_LINK_ON:
		p->po_media_info = media_info;
		if (p->po_enabled) {
			bondport_link_status_changed(p);
		}
		break;
	}
	/* generate a link-event */
	if (ifb->ifb_mode == IF_BOND_MODE_LACP) {
		if (ifbond_selection(ifb)) {
			event_code = (ifb->ifb_active_lag == NULL)
			    ? KEV_DL_LINK_OFF
			    : KEV_DL_LINK_ON;
			/* XXX need to take a reference on bond_ifp */
			bond_ifp = ifb->ifb_ifp;
			ifb->ifb_last_link_event = event_code;
		} else {
			event_code = (ifb->ifb_active_lag == NULL)
			    ? KEV_DL_LINK_OFF
			    : KEV_DL_LINK_ON;
			if (event_code != ifb->ifb_last_link_event) {
				BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
				    "%s: (event) generating LINK event",
				    ifb->ifb_name);
				bond_ifp = ifb->ifb_ifp;
				ifb->ifb_last_link_event = event_code;
			}
		}
	} else {
		/*
		 * if the distributing array membership changed from 0 <-> !0
		 * generate a link event
		 */
		if (old_distributing_count == 0
		    && ifb->ifb_distributing_count != 0) {
			event_code = KEV_DL_LINK_ON;
		} else if (old_distributing_count != 0
		    && ifb->ifb_distributing_count == 0) {
			event_code = KEV_DL_LINK_OFF;
		}
		if (event_code != 0 && event_code != ifb->ifb_last_link_event) {
			bond_ifp = ifb->ifb_ifp;
			ifb->ifb_last_link_event = event_code;
		}
	}

	bond_unlock();
	if (bond_ifp != NULL) {
		interface_link_event(bond_ifp, event_code);
	}
	return;
}

static void
bond_event(struct ifnet * port_ifp, __unused protocol_family_t protocol,
    const struct kev_msg * event)
{
	int         event_code;

	if (event->vendor_code != KEV_VENDOR_APPLE
	    || event->kev_class != KEV_NETWORK_CLASS
	    || event->kev_subclass != KEV_DL_SUBCLASS) {
		return;
	}
	event_code = event->event_code;
	switch (event_code) {
	case KEV_DL_LINK_OFF:
	case KEV_DL_LINK_ON:
	case KEV_DL_IF_DETACHING:
	case KEV_DL_IF_DETACHED:
		bond_handle_event(port_ifp, event_code);
		break;
	default:
		break;
	}
	return;
}

static errno_t
bond_detached(ifnet_t port_ifp, __unused protocol_family_t protocol)
{
	bond_handle_event(port_ifp, KEV_DL_IF_DETACHED);
	return 0;
}

static void
interface_link_event(struct ifnet * ifp, u_int32_t event_code)
{
	struct event {
		u_int32_t ifnet_family;
		u_int32_t unit;
		char if_name[IFNAMSIZ];
	};
	_Alignas(struct kern_event_msg) char message[sizeof(struct kern_event_msg) + sizeof(struct event)] = { 0 };
	struct kern_event_msg *header = (struct kern_event_msg*)message;
	struct event *data = (struct event *)(message + KEV_MSG_HEADER_SIZE);

	header->total_size   = sizeof(message);
	header->vendor_code  = KEV_VENDOR_APPLE;
	header->kev_class    = KEV_NETWORK_CLASS;
	header->kev_subclass = KEV_DL_SUBCLASS;
	header->event_code   = event_code;
	data->ifnet_family   = ifnet_family(ifp);
	data->unit           = (u_int32_t)ifnet_unit(ifp);
	strlcpy(data->if_name, ifnet_name(ifp), sizeof(data->if_name));
	ifnet_event(ifp, header);
}


/*
 * Function: bond_attach_protocol
 * Purpose:
 *   Attach a DLIL protocol to the interface.
 *
 *   The ethernet demux special cases to always return PF_BOND if the
 *   interface is bonded.  That means we receive all traffic from that
 *   interface without passing any of the traffic to any other attached
 *   protocol.
 */
static int
bond_attach_protocol(struct ifnet *ifp)
{
	int                                 error;
	struct ifnet_attach_proto_param_v2  reg;

	bzero(&reg, sizeof(reg));
	reg.input = bond_input;
	reg.event = bond_event;
	reg.detached = bond_detached;

	error = ifnet_attach_protocol_v2(ifp, PF_BOND, &reg);
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "%s: ifnet_attach_protocol failed, %d",
		    ifp->if_xname, error);
	}
	return error;
}

/*
 * Function: bond_detach_protocol
 * Purpose:
 *   Detach our DLIL protocol from an interface
 */
static int
bond_detach_protocol(struct ifnet *ifp)
{
	int         error;

	error = ifnet_detach_protocol(ifp, PF_BOND);
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "%s: ifnet_detach_protocol failed, %d",
		    ifp->if_xname, error);
	}
	return error;
}


/*
 * DLIL interface family functions
 */
extern int ether_attach_inet(ifnet_t ifp, protocol_family_t protocol_family);
extern void ether_detach_inet(ifnet_t ifp, protocol_family_t protocol_family);
extern int ether_attach_inet6(ifnet_t ifp, protocol_family_t protocol_family);
extern void ether_detach_inet6(ifnet_t ifp, protocol_family_t protocol_family);
extern int ether_attach_at(ifnet_t ifp, protocol_family_t protocol_family);
extern void ether_detach_at(ifnet_t ifp, protocol_family_t protocol_family);

__private_extern__ int
bond_family_init(void)
{
	int error = 0;

	error = proto_register_plumber(PF_INET, APPLE_IF_FAM_BOND,
	    ether_attach_inet,
	    ether_detach_inet);
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "proto_register_plumber failed for AF_INET error %d",
		    error);
		goto done;
	}
	error = proto_register_plumber(PF_INET6, APPLE_IF_FAM_BOND,
	    ether_attach_inet6,
	    ether_detach_inet6);
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "proto_register_plumber failed for AF_INET6 error %d",
		    error);
		goto done;
	}
	error = bond_clone_attach();
	if (error != 0) {
		BOND_LOG(LOG_NOTICE, BD_DBGF_LIFECYCLE,
		    "bond_clone_attach error %d",
		    error);
		goto done;
	}

done:
	return error;
}
/**
**
** LACP routines:
**
**/

/**
** LACP ifbond_list routines
**/
static bondport_ref
ifbond_list_find_moved_port(bondport_ref rx_port,
    const lacp_actor_partner_tlv_ref atlv)
{
	ifbond_ref          bond;
	bondport_ref        p;
	partner_state_ref   ps;
	LAG_info_ref        ps_li;

	TAILQ_FOREACH(bond, &g_bond->ifbond_list, ifb_bond_list) {
		TAILQ_FOREACH(p, &bond->ifb_port_list, po_port_list) {
			if (rx_port == p) {
				/* no point in comparing against ourselves */
				continue;
			}
			if (p->po_receive_state != ReceiveState_PORT_DISABLED) {
				/* it's not clear that we should be checking this */
				continue;
			}
			ps = &p->po_partner_state;
			if (lacp_actor_partner_state_defaulted(ps->ps_state)) {
				continue;
			}
			ps_li = &ps->ps_lag_info;
			if (ps->ps_port == lacp_actor_partner_tlv_get_port(atlv)
			    && bcmp(&ps_li->li_system, atlv->lap_system,
			    sizeof(ps_li->li_system)) == 0) {
				BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
				    "System " EA_FORMAT
				    " Port 0x%x moved from %s to %s",
				    EA_LIST(&ps_li->li_system), ps->ps_port,
				    bondport_get_name(p),
				    bondport_get_name(rx_port));
				return p;
			}
		}
	}
	return NULL;
}

/**
** LACP ifbond, LAG routines
**/

static int
ifbond_selection(ifbond_ref bond)
{
	int                 all_ports_ready = 0;
	int                 active_media = 0;
	LAG_ref             lag = NULL;
	int                 lag_changed = 0;
	bondport_ref        p;
	int                 port_speed = 0;

	lag = ifbond_find_best_LAG(bond, &active_media);
	if (lag != bond->ifb_active_lag) {
		if (bond->ifb_active_lag != NULL) {
			ifbond_deactivate_LAG(bond, bond->ifb_active_lag);
			bond->ifb_active_lag = NULL;
		}
		bond->ifb_active_lag = lag;
		if (lag != NULL) {
			ifbond_activate_LAG(bond, lag, active_media);
		}
		lag_changed = 1;
	} else if (lag != NULL) {
		if (lag->lag_active_media != active_media) {
			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "LAG PORT SPEED CHANGED from %d to %d",
			    link_speed(lag->lag_active_media),
			    link_speed(active_media));
			ifbond_deactivate_LAG(bond, lag);
			ifbond_activate_LAG(bond, lag, active_media);
			lag_changed = 1;
		}
	}
	if (lag != NULL) {
		port_speed = link_speed(active_media);
		all_ports_ready = ifbond_all_ports_ready(bond);
	}
	TAILQ_FOREACH(p, &bond->ifb_port_list, po_port_list) {
		if (lag != NULL && p->po_lag == lag
		    && media_speed(&p->po_media_info) == port_speed
		    && (p->po_mux_state == MuxState_DETACHED
		    || p->po_selected == SelectedState_SELECTED
		    || p->po_selected == SelectedState_STANDBY)
		    && bondport_aggregatable(p)) {
			if (bond->ifb_max_active > 0) {
				if (lag->lag_selected_port_count < bond->ifb_max_active) {
					if (p->po_selected == SelectedState_STANDBY
					    || p->po_selected == SelectedState_UNSELECTED) {
						bondport_set_selected(p, SelectedState_SELECTED);
					}
				} else if (p->po_selected == SelectedState_UNSELECTED) {
					bondport_set_selected(p, SelectedState_STANDBY);
				}
			} else {
				bondport_set_selected(p, SelectedState_SELECTED);
			}
		}
		if (bondport_flags_selected_changed(p)) {
			bondport_flags_clear_selected_changed(p);
			bondport_mux_machine(p, LAEventSelectedChange, NULL);
		}
		if (all_ports_ready
		    && bondport_flags_ready(p)
		    && p->po_mux_state == MuxState_WAITING) {
			bondport_mux_machine(p, LAEventReady, NULL);
		}
		bondport_transmit_machine(p, LAEventStart, NULL);
	}
	return lag_changed;
}

static LAG_ref
ifbond_find_best_LAG(ifbond_ref bond, int * active_media)
{
	int                 best_active = 0;
	LAG_ref             best_lag = NULL;
	int                 best_count = 0;
	int                 best_speed = 0;
	LAG_ref             lag;

	if (bond->ifb_active_lag != NULL) {
		best_lag = bond->ifb_active_lag;
		best_count = LAG_get_aggregatable_port_count(best_lag, &best_active);
		if (bond->ifb_max_active > 0
		    && best_count > bond->ifb_max_active) {
			best_count = bond->ifb_max_active;
		}
		best_speed = link_speed(best_active);
	}
	TAILQ_FOREACH(lag, &bond->ifb_lag_list, lag_list) {
		int     active;
		int     count;
		int     speed;

		if (lag == bond->ifb_active_lag) {
			/* we've already computed it */
			continue;
		}
		count = LAG_get_aggregatable_port_count(lag, &active);
		if (count == 0) {
			continue;
		}
		if (bond->ifb_max_active > 0
		    && count > bond->ifb_max_active) {
			/* if there's a limit, don't count extra links */
			count = bond->ifb_max_active;
		}
		speed = link_speed(active);
		if ((count * speed) > (best_count * best_speed)) {
			best_count = count;
			best_speed = speed;
			best_active = active;
			best_lag = lag;
		}
	}
	if (best_count == 0) {
		return NULL;
	}
	*active_media = best_active;
	return best_lag;
}

static void
ifbond_deactivate_LAG(__unused ifbond_ref bond, LAG_ref lag)
{
	bondport_ref        p;

	TAILQ_FOREACH(p, &lag->lag_port_list, po_lag_port_list) {
		bondport_set_selected(p, SelectedState_UNSELECTED);
	}
	return;
}

static void
ifbond_activate_LAG(ifbond_ref bond, LAG_ref lag, int active_media)
{
	int                 need = 0;
	bondport_ref        p;

	if (bond->ifb_max_active > 0) {
		need = bond->ifb_max_active;
	}
	lag->lag_active_media = active_media;
	TAILQ_FOREACH(p, &lag->lag_port_list, po_lag_port_list) {
		if (bondport_aggregatable(p) == 0) {
			bondport_set_selected(p, SelectedState_UNSELECTED);
		} else if (media_speed(&p->po_media_info) != link_speed(active_media)) {
			bondport_set_selected(p, SelectedState_UNSELECTED);
		} else if (p->po_mux_state == MuxState_DETACHED) {
			if (bond->ifb_max_active > 0) {
				if (need > 0) {
					bondport_set_selected(p, SelectedState_SELECTED);
					need--;
				} else {
					bondport_set_selected(p, SelectedState_STANDBY);
				}
			} else {
				bondport_set_selected(p, SelectedState_SELECTED);
			}
		} else {
			bondport_set_selected(p, SelectedState_UNSELECTED);
		}
	}
	return;
}

#if 0
static void
ifbond_set_max_active(ifbond_ref bond, int max_active)
{
	LAG_ref     lag = bond->ifb_active_lag;

	bond->ifb_max_active = max_active;
	if (bond->ifb_max_active <= 0 || lag == NULL) {
		return;
	}
	if (lag->lag_selected_port_count > bond->ifb_max_active) {
		bondport_ref    p;
		int                     remove_count;

		remove_count = lag->lag_selected_port_count - bond->ifb_max_active;
		TAILQ_FOREACH(p, &lag->lag_port_list, po_lag_port_list) {
			if (p->po_selected == SelectedState_SELECTED) {
				bondport_set_selected(p, SelectedState_UNSELECTED);
				remove_count--;
				if (remove_count == 0) {
					break;
				}
			}
		}
	}
	return;
}
#endif

static int
ifbond_all_ports_ready(ifbond_ref bond)
{
	int                 ready = 0;
	bondport_ref        p;

	if (bond->ifb_active_lag == NULL) {
		return 0;
	}
	TAILQ_FOREACH(p, &bond->ifb_active_lag->lag_port_list, po_lag_port_list) {
		if (p->po_mux_state == MuxState_WAITING
		    && p->po_selected == SelectedState_SELECTED) {
			if (bondport_flags_ready(p) == 0) {
				return 0;
			}
		}
		/* note that there was at least one ready port */
		ready = 1;
	}
	return ready;
}

static int
ifbond_all_ports_attached(ifbond_ref bond, bondport_ref this_port)
{
	bondport_ref        p;

	TAILQ_FOREACH(p, &bond->ifb_port_list, po_port_list) {
		if (this_port == p) {
			continue;
		}
		if (bondport_flags_mux_attached(p) == 0) {
			return 0;
		}
	}
	return 1;
}

static LAG_ref
ifbond_get_LAG_matching_port(ifbond_ref bond, bondport_ref p)
{
	LAG_ref     lag;

	TAILQ_FOREACH(lag, &bond->ifb_lag_list, lag_list) {
		if (bcmp(&lag->lag_info, &p->po_partner_state.ps_lag_info,
		    sizeof(lag->lag_info)) == 0) {
			return lag;
		}
	}
	return NULL;
}

static int
LAG_get_aggregatable_port_count(LAG_ref lag, int * active_media)
{
	int                 active;
	int                 count;
	bondport_ref        p;
	int                 speed;

	active = 0;
	count = 0;
	speed = 0;
	TAILQ_FOREACH(p, &lag->lag_port_list, po_lag_port_list) {
		if (bondport_aggregatable(p)) {
			int this_speed;

			this_speed = media_speed(&p->po_media_info);
			if (this_speed == 0) {
				continue;
			}
			if (this_speed > speed) {
				active = p->po_media_info.mi_active;
				speed = this_speed;
				count = 1;
			} else if (this_speed == speed) {
				count++;
			}
		}
	}
	*active_media = active;
	return count;
}


/**
** LACP bondport routines
**/
static void
bondport_link_status_changed(bondport_ref p)
{
	ifbond_ref  bond = p->po_bond;

	if (if_bond_debug) {
		if (media_active(&p->po_media_info)) {
			const char * duplex_string;

			if (media_full_duplex(&p->po_media_info)) {
				duplex_string = "full";
			} else if (media_type_unknown(&p->po_media_info)) {
				duplex_string = "unknown";
			} else {
				duplex_string = "half";
			}
			BOND_LOG(LOG_NOTICE, BD_DBGF_LACP,
			    "[%s] Link UP %d Mbit/s %s duplex",
			    bondport_get_name(p),
			    media_speed(&p->po_media_info),
			    duplex_string);
		} else {
			BOND_LOG(LOG_NOTICE, BD_DBGF_LACP,
			    "[%s] Link DOWN", bondport_get_name(p));
		}
	}
	if (bond->ifb_mode == IF_BOND_MODE_LACP) {
		if (media_active(&p->po_media_info)
		    && bond->ifb_active_lag != NULL
		    && p->po_lag == bond->ifb_active_lag
		    && p->po_selected != SelectedState_UNSELECTED) {
			if (media_speed(&p->po_media_info) != p->po_lag->lag_active_media) {
				BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
				    "[%s] Port speed %d differs from LAG %d",
				    bondport_get_name(p),
				    media_speed(&p->po_media_info),
				    link_speed(p->po_lag->lag_active_media));
				bondport_set_selected(p, SelectedState_UNSELECTED);
			}
		}
		bondport_receive_machine(p, LAEventMediaChange, NULL);
		bondport_mux_machine(p, LAEventMediaChange, NULL);
		bondport_periodic_transmit_machine(p, LAEventMediaChange, NULL);
	} else {
		if (media_active(&p->po_media_info)) {
			bondport_enable_distributing(p);
		} else {
			bondport_disable_distributing(p);
		}
	}
	return;
}

static int
bondport_aggregatable(bondport_ref p)
{
	partner_state_ref   ps = &p->po_partner_state;

	if (lacp_actor_partner_state_aggregatable(p->po_actor_state) == 0
	    || lacp_actor_partner_state_aggregatable(ps->ps_state) == 0) {
		/* we and/or our partner are individual */
		return 0;
	}
	if (p->po_lag == NULL) {
		return 0;
	}
	switch (p->po_receive_state) {
	default:
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Port is not selectable",
		    bondport_get_name(p));
		return 0;
	case ReceiveState_CURRENT:
	case ReceiveState_EXPIRED:
		break;
	}
	return 1;
}

static int
bondport_matches_LAG(bondport_ref p, LAG_ref lag)
{
	LAG_info_ref        lag_li;
	partner_state_ref   ps;
	LAG_info_ref        ps_li;

	ps = &p->po_partner_state;
	ps_li = &ps->ps_lag_info;
	lag_li = &lag->lag_info;
	if (ps_li->li_system_priority == lag_li->li_system_priority
	    && ps_li->li_key == lag_li->li_key
	    && (bcmp(&ps_li->li_system, &lag_li->li_system,
	    sizeof(lag_li->li_system))
	    == 0)) {
		return 1;
	}
	return 0;
}

static int
bondport_remove_from_LAG(bondport_ref p)
{
	int         active_lag = 0;
	ifbond_ref  bond = p->po_bond;
	LAG_ref     lag = p->po_lag;

	if (lag == NULL) {
		return 0;
	}
	TAILQ_REMOVE(&lag->lag_port_list, p, po_lag_port_list);
	BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
	    "[%s] Removed from LAG (0x%04x," EA_FORMAT ",0x%04x)",
	    bondport_get_name(p), lag->lag_info.li_system_priority,
	    EA_LIST(&lag->lag_info.li_system), lag->lag_info.li_key);
	p->po_lag = NULL;
	lag->lag_port_count--;
	if (lag->lag_port_count > 0) {
		return bond->ifb_active_lag == lag;
	}
	BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
	    "Key 0x%04x: LAG Released (%04x," EA_FORMAT ",0x%04x)",
	    bond->ifb_key,
	    lag->lag_info.li_system_priority,
	    EA_LIST(&lag->lag_info.li_system),
	    lag->lag_info.li_key);
	TAILQ_REMOVE(&bond->ifb_lag_list, lag, lag_list);
	if (bond->ifb_active_lag == lag) {
		bond->ifb_active_lag = NULL;
		active_lag = 1;
	}
	kfree_type(struct LAG_s, lag);
	return active_lag;
}

static void
bondport_add_to_LAG(bondport_ref p, LAG_ref lag)
{
	TAILQ_INSERT_TAIL(&lag->lag_port_list, p, po_lag_port_list);
	p->po_lag = lag;
	lag->lag_port_count++;
	BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
	    "[%s] Added to LAG (0x%04x," EA_FORMAT "0x%04x)",
	    bondport_get_name(p),
	    lag->lag_info.li_system_priority,
	    EA_LIST(&lag->lag_info.li_system),
	    lag->lag_info.li_key);
	return;
}

static void
bondport_assign_to_LAG(bondport_ref p)
{
	ifbond_ref  bond = p->po_bond;
	LAG_ref     lag;

	if (lacp_actor_partner_state_defaulted(p->po_actor_state)) {
		bondport_remove_from_LAG(p);
		return;
	}
	lag = p->po_lag;
	if (lag != NULL) {
		if (bondport_matches_LAG(p, lag)) {
			/* still OK */
			return;
		}
		bondport_remove_from_LAG(p);
	}
	lag = ifbond_get_LAG_matching_port(bond, p);
	if (lag != NULL) {
		bondport_add_to_LAG(p, lag);
		return;
	}
	lag = kalloc_type(struct LAG_s, Z_WAITOK);
	TAILQ_INIT(&lag->lag_port_list);
	lag->lag_port_count = 0;
	lag->lag_selected_port_count = 0;
	lag->lag_info = p->po_partner_state.ps_lag_info;
	TAILQ_INSERT_TAIL(&bond->ifb_lag_list, lag, lag_list);
	BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
	    "Key 0x%04x: LAG Created (0x%04x," EA_FORMAT ",0x%04x)",
	    bond->ifb_key, lag->lag_info.li_system_priority,
	    EA_LIST(&lag->lag_info.li_system), lag->lag_info.li_key);
	bondport_add_to_LAG(p, lag);
	return;
}

static void
bondport_receive_lacpdu(bondport_ref p, lacpdu_ref in_lacpdu_p)
{
	bondport_ref                moved_port;

	moved_port
	        = ifbond_list_find_moved_port(p, (const lacp_actor_partner_tlv_ref)
	    &in_lacpdu_p->la_actor_tlv);
	if (moved_port != NULL) {
		bondport_receive_machine(moved_port, LAEventPortMoved, NULL);
	}
	bondport_receive_machine(p, LAEventPacket, in_lacpdu_p);
	bondport_mux_machine(p, LAEventPacket, in_lacpdu_p);
	bondport_periodic_transmit_machine(p, LAEventPacket, in_lacpdu_p);
	return;
}

static void
bondport_set_selected(bondport_ref p, SelectedState s)
{
	if (s != p->po_selected) {
		ifbond_ref      bond = p->po_bond;
		LAG_ref         lag = p->po_lag;

		bondport_flags_set_selected_changed(p);
		if (lag != NULL && bond->ifb_active_lag == lag) {
			if (p->po_selected == SelectedState_SELECTED) {
				lag->lag_selected_port_count--;
			} else if (s == SelectedState_SELECTED) {
				lag->lag_selected_port_count++;
			}
			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "[%s] SetSelected: %s (was %s)",
			    bondport_get_name(p),
			    SelectedStateString(s),
			    SelectedStateString(p->po_selected));
		}
	}
	p->po_selected = s;
	return;
}

/**
** Receive machine
**/

static void
bondport_UpdateDefaultSelected(bondport_ref p)
{
	bondport_set_selected(p, SelectedState_UNSELECTED);
	return;
}

static void
bondport_RecordDefault(bondport_ref p)
{
	bzero(&p->po_partner_state, sizeof(p->po_partner_state));
	p->po_actor_state
	        = lacp_actor_partner_state_set_defaulted(p->po_actor_state);
	bondport_assign_to_LAG(p);
	return;
}

static void
bondport_UpdateSelected(bondport_ref p, lacpdu_ref lacpdu_p)
{
	lacp_actor_partner_tlv_ref  actor;
	partner_state_ref           ps;
	LAG_info_ref                ps_li;

	/* compare the PDU's Actor information to our Partner state */
	actor = (lacp_actor_partner_tlv_ref)lacpdu_p->la_actor_tlv;
	ps = &p->po_partner_state;
	ps_li = &ps->ps_lag_info;
	if (lacp_actor_partner_tlv_get_port(actor) != ps->ps_port
	    || (lacp_actor_partner_tlv_get_port_priority(actor)
	    != ps->ps_port_priority)
	    || bcmp(actor->lap_system, &ps_li->li_system, sizeof(ps_li->li_system))
	    || (lacp_actor_partner_tlv_get_system_priority(actor)
	    != ps_li->li_system_priority)
	    || (lacp_actor_partner_tlv_get_key(actor) != ps_li->li_key)
	    || (lacp_actor_partner_state_aggregatable(actor->lap_state)
	    != lacp_actor_partner_state_aggregatable(ps->ps_state))) {
		bondport_set_selected(p, SelectedState_UNSELECTED);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] updateSelected UNSELECTED",
		    bondport_get_name(p));
	}
	return;
}

static void
bondport_RecordPDU(bondport_ref p, lacpdu_ref lacpdu_p)
{
	lacp_actor_partner_tlv_ref  actor;
	ifbond_ref                  bond = p->po_bond;
	int                         lacp_maintain = 0;
	partner_state_ref           ps;
	lacp_actor_partner_tlv_ref  partner;
	LAG_info_ref                ps_li;

	/* copy the PDU's Actor information into our Partner state */
	actor = (lacp_actor_partner_tlv_ref)lacpdu_p->la_actor_tlv;
	ps = &p->po_partner_state;
	ps_li = &ps->ps_lag_info;
	ps->ps_port = lacp_actor_partner_tlv_get_port(actor);
	ps->ps_port_priority = lacp_actor_partner_tlv_get_port_priority(actor);
	ps_li->li_system = *((lacp_system_ref)actor->lap_system);
	ps_li->li_system_priority
	        = lacp_actor_partner_tlv_get_system_priority(actor);
	ps_li->li_key = lacp_actor_partner_tlv_get_key(actor);
	ps->ps_state = lacp_actor_partner_state_set_out_of_sync(actor->lap_state);
	p->po_actor_state
	        = lacp_actor_partner_state_set_not_defaulted(p->po_actor_state);

	/* compare the PDU's Partner information to our own information */
	partner = (lacp_actor_partner_tlv_ref)lacpdu_p->la_partner_tlv;

	if (lacp_actor_partner_state_active_lacp(ps->ps_state)
	    || (lacp_actor_partner_state_active_lacp(p->po_actor_state)
	    && lacp_actor_partner_state_active_lacp(partner->lap_state))) {
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] recordPDU: LACP will maintain", bondport_get_name(p));
		lacp_maintain = 1;
	}
	if ((lacp_actor_partner_tlv_get_port(partner)
	    == bondport_get_index(p))
	    && lacp_actor_partner_tlv_get_port_priority(partner) == p->po_priority
	    && bcmp(partner->lap_system, &g_bond->system,
	    sizeof(g_bond->system)) == 0
	    && (lacp_actor_partner_tlv_get_system_priority(partner)
	    == g_bond->system_priority)
	    && lacp_actor_partner_tlv_get_key(partner) == bond->ifb_key
	    && (lacp_actor_partner_state_aggregatable(partner->lap_state)
	    == lacp_actor_partner_state_aggregatable(p->po_actor_state))
	    && lacp_actor_partner_state_in_sync(actor->lap_state)
	    && lacp_maintain) {
		ps->ps_state = lacp_actor_partner_state_set_in_sync(ps->ps_state);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] recordPDU: LACP partner in sync",
		    bondport_get_name(p));
	} else if (lacp_actor_partner_state_aggregatable(actor->lap_state) == 0
	    && lacp_actor_partner_state_in_sync(actor->lap_state)
	    && lacp_maintain) {
		ps->ps_state = lacp_actor_partner_state_set_in_sync(ps->ps_state);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] recordPDU: LACP partner in sync (ind)",
		    bondport_get_name(p));
	}
	bondport_assign_to_LAG(p);
	return;
}

static __inline__ lacp_actor_partner_state
updateNTTBits(lacp_actor_partner_state s)
{
	return s & (LACP_ACTOR_PARTNER_STATE_LACP_ACTIVITY
	       | LACP_ACTOR_PARTNER_STATE_LACP_TIMEOUT
	       | LACP_ACTOR_PARTNER_STATE_AGGREGATION
	       | LACP_ACTOR_PARTNER_STATE_SYNCHRONIZATION);
}

static void
bondport_UpdateNTT(bondport_ref p, lacpdu_ref lacpdu_p)
{
	ifbond_ref                  bond = p->po_bond;
	lacp_actor_partner_tlv_ref  partner;

	/* compare the PDU's Actor information to our Partner state */
	partner = (lacp_actor_partner_tlv_ref)lacpdu_p->la_partner_tlv;
	if ((lacp_actor_partner_tlv_get_port(partner) != bondport_get_index(p))
	    || lacp_actor_partner_tlv_get_port_priority(partner) != p->po_priority
	    || bcmp(partner->lap_system, &g_bond->system, sizeof(g_bond->system))
	    || (lacp_actor_partner_tlv_get_system_priority(partner)
	    != g_bond->system_priority)
	    || lacp_actor_partner_tlv_get_key(partner) != bond->ifb_key
	    || (updateNTTBits(partner->lap_state)
	    != updateNTTBits(p->po_actor_state))) {
		bondport_flags_set_ntt(p);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] updateNTT: Need To Transmit", bondport_get_name(p));
	}
	return;
}

static void
bondport_AttachMuxToAggregator(bondport_ref p)
{
	if (bondport_flags_mux_attached(p) == 0) {
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Attached Mux To Aggregator", bondport_get_name(p));
		bondport_flags_set_mux_attached(p);
	}
	return;
}

static void
bondport_DetachMuxFromAggregator(bondport_ref p)
{
	if (bondport_flags_mux_attached(p)) {
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Detached Mux From Aggregator", bondport_get_name(p));
		bondport_flags_clear_mux_attached(p);
	}
	return;
}

static void
bondport_enable_distributing(bondport_ref p)
{
	if (bondport_flags_distributing(p) == 0) {
		ifbond_ref      bond = p->po_bond;

		bond->ifb_distributing_array[bond->ifb_distributing_count++] = p;
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Distribution Enabled", bondport_get_name(p));
		bondport_flags_set_distributing(p);
	}
	return;
}

static void
bondport_disable_distributing(bondport_ref p)
{
	if (bondport_flags_distributing(p)) {
		bondport_ref *  array;
		ifbond_ref      bond;
		int             count;
		int             i;

		bond = p->po_bond;
		array = bond->ifb_distributing_array;
		count = bond->ifb_distributing_count;
		for (i = 0; i < count; i++) {
			if (array[i] == p) {
				int     j;

				for (j = i; j < (count - 1); j++) {
					array[j] = array[j + 1];
				}
				break;
			}
		}
		bond->ifb_distributing_count--;
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Distribution Disabled", bondport_get_name(p));
		bondport_flags_clear_distributing(p);
	}
	return;
}

/**
** Receive machine functions
**/
static void
bondport_receive_machine_initialize(bondport_ref p, LAEvent event,
    void * event_data);
static void
bondport_receive_machine_port_disabled(bondport_ref p, LAEvent event,
    void * event_data);
static void
bondport_receive_machine_expired(bondport_ref p, LAEvent event,
    void * event_data);
static void
bondport_receive_machine_lacp_disabled(bondport_ref p, LAEvent event,
    void * event_data);
static void
bondport_receive_machine_defaulted(bondport_ref p, LAEvent event,
    void * event_data);
static void
bondport_receive_machine_current(bondport_ref p, LAEvent event,
    void * event_data);

static void
bondport_receive_machine_event(bondport_ref p, LAEvent event,
    void * event_data)
{
	switch (p->po_receive_state) {
	case ReceiveState_none:
		bondport_receive_machine_initialize(p, LAEventStart, NULL);
		break;
	case ReceiveState_INITIALIZE:
		bondport_receive_machine_initialize(p, event, event_data);
		break;
	case ReceiveState_PORT_DISABLED:
		bondport_receive_machine_port_disabled(p, event, event_data);
		break;
	case ReceiveState_EXPIRED:
		bondport_receive_machine_expired(p, event, event_data);
		break;
	case ReceiveState_LACP_DISABLED:
		bondport_receive_machine_lacp_disabled(p, event, event_data);
		break;
	case ReceiveState_DEFAULTED:
		bondport_receive_machine_defaulted(p, event, event_data);
		break;
	case ReceiveState_CURRENT:
		bondport_receive_machine_current(p, event, event_data);
		break;
	default:
		break;
	}
	return;
}

static void
bondport_receive_machine(bondport_ref p, LAEvent event,
    void * event_data)
{
	switch (event) {
	case LAEventPacket:
		if (p->po_receive_state != ReceiveState_LACP_DISABLED) {
			bondport_receive_machine_current(p, event, event_data);
		}
		break;
	case LAEventMediaChange:
		if (media_active(&p->po_media_info)) {
			switch (p->po_receive_state) {
			case ReceiveState_PORT_DISABLED:
			case ReceiveState_LACP_DISABLED:
				bondport_receive_machine_port_disabled(p, LAEventMediaChange, NULL);
				break;
			default:
				break;
			}
		} else {
			bondport_receive_machine_port_disabled(p, LAEventStart, NULL);
		}
		break;
	default:
		bondport_receive_machine_event(p, event, event_data);
		break;
	}
	return;
}

static void
bondport_receive_machine_initialize(bondport_ref p, LAEvent event,
    __unused void * event_data)
{
	switch (event) {
	case LAEventStart:
		devtimer_cancel(p->po_current_while_timer);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Receive INITIALIZE", bondport_get_name(p));
		p->po_receive_state = ReceiveState_INITIALIZE;
		bondport_set_selected(p, SelectedState_UNSELECTED);
		bondport_RecordDefault(p);
		p->po_actor_state
		        = lacp_actor_partner_state_set_not_expired(p->po_actor_state);
		bondport_receive_machine_port_disabled(p, LAEventStart, NULL);
		break;
	default:
		break;
	}
	return;
}

static void
bondport_receive_machine_port_disabled(bondport_ref p, LAEvent event,
    __unused void * event_data)
{
	partner_state_ref   ps;

	switch (event) {
	case LAEventStart:
		devtimer_cancel(p->po_current_while_timer);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Receive PORT_DISABLED", bondport_get_name(p));
		p->po_receive_state = ReceiveState_PORT_DISABLED;
		ps = &p->po_partner_state;
		ps->ps_state = lacp_actor_partner_state_set_out_of_sync(ps->ps_state);
		OS_FALLTHROUGH;
	case LAEventMediaChange:
		if (media_active(&p->po_media_info)) {
			if (media_ok(&p->po_media_info)) {
				bondport_receive_machine_expired(p, LAEventStart, NULL);
			} else {
				bondport_receive_machine_lacp_disabled(p, LAEventStart, NULL);
			}
		} else if (p->po_selected == SelectedState_SELECTED) {
			struct timeval      tv;

			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "[%s] Receive PORT_DISABLED: link timer started",
			    bondport_get_name(p));
			tv.tv_sec = 1;
			tv.tv_usec = 0;
			devtimer_set_relative(p->po_current_while_timer, tv,
			    (devtimer_timeout_func)(void (*)(void))
			    bondport_receive_machine_port_disabled,
			    __unsafe_forge_single(void *, LAEventTimeout), NULL);
		} else if (p->po_selected == SelectedState_STANDBY) {
			bondport_set_selected(p, SelectedState_UNSELECTED);
		}
		break;
	case LAEventTimeout:
		if (p->po_selected == SelectedState_SELECTED) {
			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "[%s] Receive PORT_DISABLED: "
			    "link timer completed, marking UNSELECTED",
			    bondport_get_name(p));
			bondport_set_selected(p, SelectedState_UNSELECTED);
		}
		break;
	case LAEventPortMoved:
		bondport_receive_machine_initialize(p, LAEventStart, NULL);
		break;
	default:
		break;
	}
	return;
}

static void
bondport_receive_machine_expired(bondport_ref p, LAEvent event,
    __unused void * event_data)
{
	lacp_actor_partner_state    s;
	struct timeval              tv;

	switch (event) {
	case LAEventStart:
		devtimer_cancel(p->po_current_while_timer);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Receive EXPIRED", bondport_get_name(p));
		p->po_receive_state = ReceiveState_EXPIRED;
		s = p->po_partner_state.ps_state;
		s = lacp_actor_partner_state_set_out_of_sync(s);
		s = lacp_actor_partner_state_set_short_timeout(s);
		p->po_partner_state.ps_state = s;
		p->po_actor_state
		        = lacp_actor_partner_state_set_expired(p->po_actor_state);
		/* start current_while timer */
		tv.tv_sec = LACP_SHORT_TIMEOUT_TIME;
		tv.tv_usec = 0;
		devtimer_set_relative(p->po_current_while_timer, tv,
		    (devtimer_timeout_func)(void (*)(void))
		    bondport_receive_machine_expired,
		    __unsafe_forge_single(void *, LAEventTimeout), NULL);

		break;
	case LAEventTimeout:
		bondport_receive_machine_defaulted(p, LAEventStart, NULL);
		break;
	default:
		break;
	}
	return;
}

static void
bondport_receive_machine_lacp_disabled(bondport_ref p, LAEvent event,
    __unused void * event_data)
{
	partner_state_ref   ps;
	switch (event) {
	case LAEventStart:
		devtimer_cancel(p->po_current_while_timer);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Receive LACP_DISABLED", bondport_get_name(p));
		p->po_receive_state = ReceiveState_LACP_DISABLED;
		bondport_set_selected(p, SelectedState_UNSELECTED);
		bondport_RecordDefault(p);
		ps = &p->po_partner_state;
		ps->ps_state = lacp_actor_partner_state_set_individual(ps->ps_state);
		p->po_actor_state
		        = lacp_actor_partner_state_set_not_expired(p->po_actor_state);
		break;
	default:
		break;
	}
	return;
}

static void
bondport_receive_machine_defaulted(bondport_ref p, LAEvent event,
    __unused void * event_data)
{
	switch (event) {
	case LAEventStart:
		devtimer_cancel(p->po_current_while_timer);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Receive DEFAULTED", bondport_get_name(p));
		p->po_receive_state = ReceiveState_DEFAULTED;
		bondport_UpdateDefaultSelected(p);
		bondport_RecordDefault(p);
		p->po_actor_state
		        = lacp_actor_partner_state_set_not_expired(p->po_actor_state);
		break;
	default:
		break;
	}
	return;
}

static void
bondport_receive_machine_current(bondport_ref p, LAEvent event,
    void * event_data)
{
	partner_state_ref   ps;
	struct timeval      tv;

	switch (event) {
	case LAEventPacket:
		devtimer_cancel(p->po_current_while_timer);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Receive CURRENT", bondport_get_name(p));
		p->po_receive_state = ReceiveState_CURRENT;
		bondport_UpdateSelected(p, event_data);
		bondport_UpdateNTT(p, event_data);
		bondport_RecordPDU(p, event_data);
		p->po_actor_state
		        = lacp_actor_partner_state_set_not_expired(p->po_actor_state);
		bondport_assign_to_LAG(p);
		/* start current_while timer */
		ps = &p->po_partner_state;
		if (lacp_actor_partner_state_short_timeout(ps->ps_state)) {
			tv.tv_sec = LACP_SHORT_TIMEOUT_TIME;
		} else {
			tv.tv_sec = LACP_LONG_TIMEOUT_TIME;
		}
		tv.tv_usec = 0;
		devtimer_set_relative(p->po_current_while_timer, tv,
		    (devtimer_timeout_func)(void (*)(void))
		    bondport_receive_machine_current,
		    __unsafe_forge_single(void *, LAEventTimeout), NULL);
		break;
	case LAEventTimeout:
		bondport_receive_machine_expired(p, LAEventStart, NULL);
		break;
	default:
		break;
	}
	return;
}

/**
** Periodic Transmission machine
**/

static void
bondport_periodic_transmit_machine(bondport_ref p, LAEvent event,
    __unused void * event_data)
{
	int                 interval;
	partner_state_ref   ps;
	struct timeval      tv;

	switch (event) {
	case LAEventStart:
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] periodic_transmit Start", bondport_get_name(p));
		OS_FALLTHROUGH;
	case LAEventMediaChange:
		devtimer_cancel(p->po_periodic_timer);
		p->po_periodic_interval = 0;
		if (media_active(&p->po_media_info) == 0
		    || media_ok(&p->po_media_info) == 0) {
			break;
		}
		OS_FALLTHROUGH;
	case LAEventPacket:
		/* Neither Partner nor Actor are LACP Active, no periodic tx */
		ps = &p->po_partner_state;
		if (lacp_actor_partner_state_active_lacp(p->po_actor_state) == 0
		    && (lacp_actor_partner_state_active_lacp(ps->ps_state)
		    == 0)) {
			devtimer_cancel(p->po_periodic_timer);
			p->po_periodic_interval = 0;
			break;
		}
		if (lacp_actor_partner_state_short_timeout(ps->ps_state)) {
			interval = LACP_FAST_PERIODIC_TIME;
		} else {
			interval = LACP_SLOW_PERIODIC_TIME;
		}
		if (p->po_periodic_interval != interval) {
			if (interval == LACP_FAST_PERIODIC_TIME
			    && p->po_periodic_interval
			    == LACP_SLOW_PERIODIC_TIME) {
				BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
				    "[%s] periodic_transmit: Need To Transmit",
				    bondport_get_name(p));
				bondport_flags_set_ntt(p);
			}
			p->po_periodic_interval = interval;
			tv.tv_usec = 0;
			tv.tv_sec = interval;
			devtimer_set_relative(p->po_periodic_timer, tv,
			    (devtimer_timeout_func)(void (*)(void))
			    bondport_periodic_transmit_machine,
			    __unsafe_forge_single(void *, LAEventTimeout), NULL);
			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "[%s] Periodic Transmission Timer: %d secs",
			    bondport_get_name(p),
			    p->po_periodic_interval);
		}
		break;
	case LAEventTimeout:
		bondport_flags_set_ntt(p);
		tv.tv_sec = p->po_periodic_interval;
		tv.tv_usec = 0;
		devtimer_set_relative(p->po_periodic_timer, tv, (devtimer_timeout_func)(void (*)(void))
		    bondport_periodic_transmit_machine,
		    __unsafe_forge_single(void *, LAEventTimeout), NULL);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Periodic Transmission Timer: %d secs",
		    bondport_get_name(p), p->po_periodic_interval);
		break;
	default:
		break;
	}
	return;
}

/**
** Transmit machine
**/
static int
bondport_can_transmit(bondport_ref p, int32_t current_secs,
    __darwin_time_t * next_secs)
{
	if (p->po_last_transmit_secs != current_secs) {
		p->po_last_transmit_secs = current_secs;
		p->po_n_transmit = 0;
	}
	if (p->po_n_transmit < LACP_PACKET_RATE) {
		p->po_n_transmit++;
		return 1;
	}
	if (next_secs != NULL) {
		*next_secs = current_secs + 1;
	}
	return 0;
}

static void
bondport_transmit_machine(bondport_ref p, LAEvent event,
    void * event_data)
{
	lacp_actor_partner_tlv_ref  aptlv;
	lacp_collector_tlv_ref      ctlv;
	struct timeval              next_tick_time = {.tv_sec = 0, .tv_usec = 0};
	lacpdu_ref          out_lacpdu_p;
	packet_buffer_ref           pkt;
	partner_state_ref           ps;
	LAG_info_ref                ps_li;

	switch (event) {
	case LAEventTimeout:
	case LAEventStart:
		if (p->po_periodic_interval == 0 || bondport_flags_ntt(p) == 0) {
			break;
		}
		if (event_data == TRANSMIT_MACHINE_TX_IMMEDIATE) {
			/* we're going away, transmit the packet no matter what */
		} else if (bondport_can_transmit(p, devtimer_current_secs(),
		    &next_tick_time.tv_sec) == 0) {
			if (devtimer_enabled(p->po_transmit_timer)) {
				BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
				    "[%s] Transmit Timer Already Set",
				    bondport_get_name(p));
			} else {
				devtimer_set_absolute(p->po_transmit_timer, next_tick_time,
				    (devtimer_timeout_func)(void (*)(void))
				    bondport_transmit_machine,
				    __unsafe_forge_single(void *, LAEventTimeout), NULL);
				BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
				    "[%s] Transmit Timer Deadline %d secs",
				    bondport_get_name(p),
				    (int)next_tick_time.tv_sec);
			}
			break;
		}
		if (event == LAEventTimeout) {
			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "[%s] Transmit Timer Complete",
			    bondport_get_name(p));
		}
		pkt = packet_buffer_allocate(sizeof(*out_lacpdu_p));
		if (pkt == NULL) {
			printf("[%s] Transmit: failed to allocate packet buffer\n",
			    bondport_get_name(p));
			break;
		}
		out_lacpdu_p = (lacpdu_ref)packet_buffer_byteptr(pkt);
		bzero(out_lacpdu_p, sizeof(*out_lacpdu_p));
		out_lacpdu_p->la_subtype = IEEE8023AD_SLOW_PROTO_SUBTYPE_LACP;
		out_lacpdu_p->la_version = LACPDU_VERSION_1;

		/* Actor */
		aptlv = (lacp_actor_partner_tlv_ref)out_lacpdu_p->la_actor_tlv;
		aptlv->lap_tlv_type = LACPDU_TLV_TYPE_ACTOR;
		aptlv->lap_length = LACPDU_ACTOR_TLV_LENGTH;
		*((lacp_system_ref)aptlv->lap_system) = g_bond->system;
		lacp_actor_partner_tlv_set_system_priority(aptlv,
		    g_bond->system_priority);
		lacp_actor_partner_tlv_set_port_priority(aptlv, p->po_priority);
		lacp_actor_partner_tlv_set_port(aptlv, bondport_get_index(p));
		lacp_actor_partner_tlv_set_key(aptlv, p->po_bond->ifb_key);
		aptlv->lap_state = p->po_actor_state;

		/* Partner */
		aptlv = (lacp_actor_partner_tlv_ref)out_lacpdu_p->la_partner_tlv;
		aptlv->lap_tlv_type = LACPDU_TLV_TYPE_PARTNER;
		aptlv->lap_length = LACPDU_PARTNER_TLV_LENGTH;
		ps = &p->po_partner_state;
		ps_li = &ps->ps_lag_info;
		lacp_actor_partner_tlv_set_port(aptlv, ps->ps_port);
		lacp_actor_partner_tlv_set_port_priority(aptlv, ps->ps_port_priority);
		*((lacp_system_ref)aptlv->lap_system) = ps_li->li_system;
		lacp_actor_partner_tlv_set_system_priority(aptlv,
		    ps_li->li_system_priority);
		lacp_actor_partner_tlv_set_key(aptlv, ps_li->li_key);
		aptlv->lap_state = ps->ps_state;

		/* Collector */
		ctlv = (lacp_collector_tlv_ref)out_lacpdu_p->la_collector_tlv;
		ctlv->lac_tlv_type = LACPDU_TLV_TYPE_COLLECTOR;
		ctlv->lac_length = LACPDU_COLLECTOR_TLV_LENGTH;

		bondport_slow_proto_transmit(p, pkt);
		bondport_flags_clear_ntt(p);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP, "[%s] Transmit Packet %d",
		    bondport_get_name(p), p->po_n_transmit);
		break;
	default:
		break;
	}
	return;
}

/**
** Mux machine functions
**/

static void
bondport_mux_machine_detached(bondport_ref p, LAEvent event,
    void * event_data);
static void
bondport_mux_machine_waiting(bondport_ref p, LAEvent event,
    void * event_data);
static void
bondport_mux_machine_attached(bondport_ref p, LAEvent event,
    void * event_data);

static void
bondport_mux_machine_collecting_distributing(bondport_ref p, LAEvent event,
    void * event_data);

static void
bondport_mux_machine(bondport_ref p, LAEvent event, void * event_data)
{
	switch (p->po_mux_state) {
	case MuxState_none:
		bondport_mux_machine_detached(p, LAEventStart, NULL);
		break;
	case MuxState_DETACHED:
		bondport_mux_machine_detached(p, event, event_data);
		break;
	case MuxState_WAITING:
		bondport_mux_machine_waiting(p, event, event_data);
		break;
	case MuxState_ATTACHED:
		bondport_mux_machine_attached(p, event, event_data);
		break;
	case MuxState_COLLECTING_DISTRIBUTING:
		bondport_mux_machine_collecting_distributing(p, event, event_data);
		break;
	default:
		break;
	}
	return;
}

static void
bondport_mux_machine_detached(bondport_ref p, LAEvent event,
    __unused void * event_data)
{
	lacp_actor_partner_state    s;

	switch (event) {
	case LAEventStart:
		devtimer_cancel(p->po_wait_while_timer);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP, "[%s] Mux DETACHED",
		    bondport_get_name(p));
		p->po_mux_state = MuxState_DETACHED;
		bondport_flags_clear_ready(p);
		bondport_DetachMuxFromAggregator(p);
		bondport_disable_distributing(p);
		s = p->po_actor_state;
		s = lacp_actor_partner_state_set_out_of_sync(s);
		s = lacp_actor_partner_state_set_not_collecting(s);
		s = lacp_actor_partner_state_set_not_distributing(s);
		p->po_actor_state = s;
		bondport_flags_set_ntt(p);
		break;
	case LAEventSelectedChange:
	case LAEventPacket:
	case LAEventMediaChange:
		if (p->po_selected == SelectedState_SELECTED
		    || p->po_selected == SelectedState_STANDBY) {
			bondport_mux_machine_waiting(p, LAEventStart, NULL);
		}
		break;
	default:
		break;
	}
	return;
}

static void
bondport_mux_machine_waiting(bondport_ref p, LAEvent event,
    __unused void * event_data)
{
	struct timeval      tv;

	switch (event) {
	case LAEventStart:
		devtimer_cancel(p->po_wait_while_timer);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP, "[%s] Mux WAITING",
		    bondport_get_name(p));
		p->po_mux_state = MuxState_WAITING;
		OS_FALLTHROUGH;
	default:
	case LAEventSelectedChange:
		if (p->po_selected == SelectedState_UNSELECTED) {
			bondport_mux_machine_detached(p, LAEventStart, NULL);
			break;
		}
		if (p->po_selected == SelectedState_STANDBY) {
			devtimer_cancel(p->po_wait_while_timer);
			/* wait until state changes to SELECTED */
			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "[%s] Mux WAITING: Standby", bondport_get_name(p));
			break;
		}
		if (bondport_flags_ready(p)) {
			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "[%s] Mux WAITING: Port is already ready",
			    bondport_get_name(p));
			break;
		}
		if (devtimer_enabled(p->po_wait_while_timer)) {
			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "[%s] Mux WAITING: Timer already set",
			    bondport_get_name(p));
			break;
		}
		if (ifbond_all_ports_attached(p->po_bond, p)) {
			devtimer_cancel(p->po_wait_while_timer);
			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "[%s] Mux WAITING: No waiting",
			    bondport_get_name(p));
			bondport_flags_set_ready(p);
			goto no_waiting;
		}
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Mux WAITING: 2 seconds", bondport_get_name(p));
		tv.tv_sec = LACP_AGGREGATE_WAIT_TIME;
		tv.tv_usec = 0;
		devtimer_set_relative(p->po_wait_while_timer, tv,
		    (devtimer_timeout_func)(void (*)(void))
		    bondport_mux_machine_waiting,
		    __unsafe_forge_single(void *, LAEventTimeout), NULL);
		break;
	case LAEventTimeout:
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP, "[%s] Mux WAITING: Ready",
		    bondport_get_name(p));
		bondport_flags_set_ready(p);
		break;
	case LAEventReady:
no_waiting:
		if (bondport_flags_ready(p)) {
			BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
			    "[%s] Mux WAITING: All Ports Ready",
			    bondport_get_name(p));
			bondport_mux_machine_attached(p, LAEventStart, NULL);
			break;
		}
		break;
	}
	return;
}

static void
bondport_mux_machine_attached(bondport_ref p, LAEvent event,
    __unused void * event_data)
{
	lacp_actor_partner_state    s;

	switch (event) {
	case LAEventStart:
		devtimer_cancel(p->po_wait_while_timer);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP, "[%s] Mux ATTACHED",
		    bondport_get_name(p));
		p->po_mux_state = MuxState_ATTACHED;
		bondport_AttachMuxToAggregator(p);
		s = p->po_actor_state;
		s = lacp_actor_partner_state_set_in_sync(s);
		s = lacp_actor_partner_state_set_not_collecting(s);
		s = lacp_actor_partner_state_set_not_distributing(s);
		bondport_disable_distributing(p);
		p->po_actor_state = s;
		bondport_flags_set_ntt(p);
		OS_FALLTHROUGH;
	default:
		switch (p->po_selected) {
		case SelectedState_SELECTED:
			s = p->po_partner_state.ps_state;
			if (lacp_actor_partner_state_in_sync(s)) {
				bondport_mux_machine_collecting_distributing(p,
				    LAEventStart, NULL);
			}
			break;
		default:
			bondport_mux_machine_detached(p, LAEventStart, NULL);
			break;
		}
		break;
	}
	return;
}

static void
bondport_mux_machine_collecting_distributing(bondport_ref p,
    LAEvent event,
    __unused void * event_data)
{
	lacp_actor_partner_state    s;

	switch (event) {
	case LAEventStart:
		devtimer_cancel(p->po_wait_while_timer);
		BOND_LOG(LOG_DEBUG, BD_DBGF_LACP,
		    "[%s] Mux COLLECTING_DISTRIBUTING",
		    bondport_get_name(p));
		p->po_mux_state = MuxState_COLLECTING_DISTRIBUTING;
		bondport_enable_distributing(p);
		s = p->po_actor_state;
		s = lacp_actor_partner_state_set_collecting(s);
		s = lacp_actor_partner_state_set_distributing(s);
		p->po_actor_state = s;
		bondport_flags_set_ntt(p);
		OS_FALLTHROUGH;
	default:
		s = p->po_partner_state.ps_state;
		if (lacp_actor_partner_state_in_sync(s) == 0) {
			bondport_mux_machine_attached(p, LAEventStart, NULL);
			break;
		}
		switch (p->po_selected) {
		case SelectedState_UNSELECTED:
		case SelectedState_STANDBY:
			bondport_mux_machine_attached(p, LAEventStart, NULL);
			break;
		default:
			break;
		}
		break;
	}
	return;
}