dls - zygaena/hammerhead

root / base / usr / src / uts / common / io / dls
dls Plain Text 4243 lines 107.2 KB
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 * Copyright 2012, Nexenta Systems, Inc. All rights reserved.
 */

/*
 * Copyright 2019 Joyent, Inc.
 * Copyright 2025 Oxide Computer Company
 */

/*
 * Data-Link Services Module
 */

#include	<sys/stdbit.h>
#include	<sys/strsun.h>
#include	<sys/vlan.h>
#include	<sys/dld_impl.h>
#include	<sys/mac_client_priv.h>

int
dls_open(dls_link_t *dlp, dls_dl_handle_t ddh, dld_str_t *dsp)
{
	zoneid_t	zid = getzoneid();
	boolean_t	local;
	int		err;

	/*
	 * Check whether this client belongs to the zone of this dlp. Note that
	 * a global zone client is allowed to open a local zone dlp.
	 */
	if (zid != GLOBAL_ZONEID && dlp->dl_zid != zid)
		return (ENOENT);

	/*
	 * mac_start() is required for non-legacy MACs to show accurate
	 * kstats even before the interface is brought up. For legacy
	 * drivers, this is not needed. Further, calling mac_start() for
	 * legacy drivers would make the shared-lower-stream to stay in
	 * the DL_IDLE state, which in turn causes performance regression.
	 */
	if (!mac_capab_get(dlp->dl_mh, MAC_CAPAB_LEGACY, NULL) &&
	    ((err = mac_start(dlp->dl_mh)) != 0)) {
		return (err);
	}

	local = (zid == dlp->dl_zid);
	dlp->dl_zone_ref += (local ? 1 : 0);

	/*
	 * Cache a copy of the MAC interface handle, a pointer to the
	 * immutable MAC info.
	 */
	dsp->ds_dlp = dlp;
	dsp->ds_mh = dlp->dl_mh;
	dsp->ds_mch = dlp->dl_mch;
	dsp->ds_mip = dlp->dl_mip;
	dsp->ds_ddh = ddh;
	dsp->ds_local = local;

	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));
	return (0);
}

void
dls_close(dld_str_t *dsp)
{
	dls_link_t		*dlp = dsp->ds_dlp;
	dls_multicst_addr_t	*p;
	dls_multicst_addr_t	*nextp;

	ASSERT(dsp->ds_datathr_cnt == 0);
	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));

	if (dsp->ds_local)
		dlp->dl_zone_ref--;
	dsp->ds_local = B_FALSE;

	/*
	 * Walk the list of multicast addresses, disabling each at the MAC.
	 * Note that we must remove multicast address before
	 * mac_unicast_remove() (called by dls_active_clear()) because
	 * mac_multicast_remove() relies on the unicast flows on the mac
	 * client.
	 */
	for (p = dsp->ds_dmap; p != NULL; p = nextp) {
		(void) mac_multicast_remove(dsp->ds_mch, p->dma_addr);
		nextp = p->dma_nextp;
		kmem_free(p, sizeof (dls_multicst_addr_t));
	}
	dsp->ds_dmap = NULL;

	dls_active_clear(dsp, B_TRUE);

	/*
	 * If the dld_str_t is bound then unbind it.
	 */
	if (dsp->ds_dlstate == DL_IDLE) {
		dls_unbind(dsp);
		dsp->ds_dlstate = DL_UNBOUND;
	}

	/*
	 * If the MAC has been set in promiscuous mode then disable it.
	 * This needs to be done before resetting ds_rx.
	 */
	(void) dls_promisc(dsp, 0);

	/*
	 * At this point we have cutoff inbound packet flow from the mac
	 * for this 'dsp'. The dls_link_remove above cut off packets meant
	 * for us and waited for upcalls to finish. Similarly the dls_promisc
	 * reset above waited for promisc callbacks to finish. Now we can
	 * safely reset ds_rx to NULL
	 */
	dsp->ds_rx = NULL;
	dsp->ds_rx_arg = NULL;

	dsp->ds_dlp = NULL;

	if (!mac_capab_get(dsp->ds_mh, MAC_CAPAB_LEGACY, NULL))
		mac_stop(dsp->ds_mh);

	/*
	 * Release our reference to the dls_link_t allowing that to be
	 * destroyed if there are no more dls_impl_t.
	 */
	dls_link_rele(dlp);
}

int
dls_bind(dld_str_t *dsp, uint32_t sap)
{
	uint32_t	dls_sap;

	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));

	/*
	 * Check to see the value is legal for the media type.
	 */
	if (!mac_sap_verify(dsp->ds_mh, sap, &dls_sap))
		return (EINVAL);

	if (dsp->ds_promisc & DLS_PROMISC_SAP)
		dls_sap = DLS_SAP_PROMISC;

	/*
	 * Set up the dld_str_t to mark it as able to receive packets.
	 */
	dsp->ds_sap = sap;

	/*
	 * The MAC layer does the VLAN demultiplexing and will only pass up
	 * untagged packets to non-promiscuous primary MAC clients. In order to
	 * support binding to the VLAN SAP, which is required by DLPI, DLS
	 * needs to get a copy of all tagged packets when the client binds to
	 * the VLAN SAP. We do this by registering a separate promiscuous
	 * callback for each DLS client binding to that SAP.
	 *
	 * Note: even though there are two promiscuous handles in dld_str_t,
	 * ds_mph is for the regular promiscuous mode, ds_vlan_mph is the handle
	 * to receive VLAN traffic when promiscuous mode is not on. Only one of
	 * them can be non-NULL at the same time, to avoid receiving duplicate
	 * copies of packets.
	 */
	if (sap == ETHERTYPE_VLAN && dsp->ds_promisc == 0) {
		int err;

		if (dsp->ds_vlan_mph != NULL)
			return (EINVAL);
		err = mac_promisc_add(dsp->ds_mch,
		    MAC_CLIENT_PROMISC_ALL, dls_rx_vlan_promisc, dsp,
		    &dsp->ds_vlan_mph, MAC_PROMISC_FLAGS_NO_PHYS);

		if (err == 0 && dsp->ds_nonip &&
		    dsp->ds_dlp->dl_nonip_cnt++ == 0)
			mac_rx_bypass_disable(dsp->ds_mch);

		return (err);
	}

	/*
	 * Now bind the dld_str_t by adding it into the hash table in the
	 * dls_link_t.
	 */
	dls_link_add(dsp->ds_dlp, dls_sap, dsp);
	if (dsp->ds_nonip && dsp->ds_dlp->dl_nonip_cnt++ == 0)
		mac_rx_bypass_disable(dsp->ds_mch);

	return (0);
}

void
dls_unbind(dld_str_t *dsp)
{
	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));

	if (dsp->ds_nonip && --dsp->ds_dlp->dl_nonip_cnt == 0)
		mac_rx_bypass_enable(dsp->ds_mch);

	/*
	 * A VLAN SAP does not actually add itself to the STREAM head today.
	 * While we initially set up a VLAN handle below, it's possible that
	 * something else will have come in and clobbered it.
	 */
	if (dsp->ds_sap == ETHERTYPE_VLAN) {
		if (dsp->ds_vlan_mph != NULL) {
			mac_promisc_remove(dsp->ds_vlan_mph);
			dsp->ds_vlan_mph = NULL;
		}
		return;
	}

	/*
	 * Unbind the dld_str_t by removing it from the hash table in the
	 * dls_link_t.
	 */
	dls_link_remove(dsp->ds_dlp, dsp);
	dsp->ds_sap = 0;
}

/*
 * In order to prevent promiscuous-mode processing with dsp->ds_promisc
 * set to inaccurate values, this function sets dsp->ds_promisc with new
 * flags.  For enabling (mac_promisc_add), the flags are set prior to the
 * actual enabling.  For disabling (mac_promisc_remove), the flags are set
 * after the actual disabling.
 */
int
dls_promisc(dld_str_t *dsp, uint32_t new_flags)
{
	int err = 0;
	uint32_t old_flags = dsp->ds_promisc;
	const uint32_t option_flags = DLS_PROMISC_RX_ONLY |
	    DLS_PROMISC_INCOMING | DLS_PROMISC_OUTGOING;
	uint32_t old_type = old_flags & ~option_flags;
	uint32_t new_type = new_flags & ~option_flags;
	mac_client_promisc_type_t mptype = MAC_CLIENT_PROMISC_ALL;
	uint16_t mac_flags = 0;

	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));
	ASSERT(!(new_flags & ~(DLS_PROMISC_SAP | DLS_PROMISC_MULTI |
	    DLS_PROMISC_PHYS | option_flags)));

	/*
	 * If the user has only requested DLS_PROMISC_MULTI then we need to make
	 * sure that they don't see all packets.
	 */
	if (new_type == DLS_PROMISC_MULTI)
		mptype = MAC_CLIENT_PROMISC_MULTI;

	/* Can choose only one of Rx-only or incoming/outoing. */
	if (stdc_count_ones_ui(new_flags & option_flags) > 1)
		return (EINVAL);

	/*
	 * Look at new flags and figure out the correct mac promisc flags.
	 * If we've only requested DLS_PROMISC_SAP and not _MULTI or _PHYS,
	 * don't turn on physical promisc mode.
	 *
	 * Note that there are two MAC flags, MAC_PROMISC_FLAGS_NO_TX_LOOP and
	 * MAC_PROMISC_FLAGS_RX_ONLY, which are similar but subtly different.
	 * NO_TX_LOOP stops outgoing packets transmitted by a MAC client from
	 * being delivered to a promsic handler on the same client. Whereas
	 * RX_ONLY suppresses outgoing packets across all MAC clients. For
	 * backwards compatibility, we map DLS_PROMISC_RX_ONLY to
	 * MAC_PROMISC_FLAGS_NO_TX_LOOP and DLS_PROMISC_INCOMING/OUTGOING to
	 * MAC_PROMISC_FLAGS_RX_ONLY/MAC_PROMISC_FLAGS_TX_ONLY.
	 */
	if (new_flags & DLS_PROMISC_RX_ONLY)
		mac_flags |= MAC_PROMISC_FLAGS_NO_TX_LOOP;
	if (new_flags & DLS_PROMISC_INCOMING)
		mac_flags |= MAC_PROMISC_FLAGS_RX_ONLY;
	if (new_flags & DLS_PROMISC_OUTGOING)
		mac_flags |= MAC_PROMISC_FLAGS_TX_ONLY;
	if (new_type == DLS_PROMISC_SAP)
		mac_flags |= MAC_PROMISC_FLAGS_NO_PHYS;

	/*
	 * There are three cases we care about here with respect to MAC. Going
	 * from nothing to something, something to nothing, something to
	 * something where we need to change how we're getting stuff from mac.
	 * In the last case, as long as they're not equal, we need to assume
	 * something has changed and do something about it.
	 */
	if (old_type == 0 && new_type == 0) {
		/*
		 * If there are only option flags in the old and new flags
		 * then there is no need to talk to MAC. Just update the flags.
		 */
		dsp->ds_promisc = new_flags;
	} else if (old_type == 0 && new_type != 0) {
		dsp->ds_promisc = new_flags;
		err = mac_promisc_add(dsp->ds_mch, mptype,
		    dls_rx_promisc, dsp, &dsp->ds_mph, mac_flags);
		if (err != 0) {
			dsp->ds_promisc = old_flags;
			return (err);
		}

		/* Remove vlan promisc handle to avoid sending dup copy up */
		if (dsp->ds_vlan_mph != NULL) {
			mac_promisc_remove(dsp->ds_vlan_mph);
			dsp->ds_vlan_mph = NULL;
		}
	} else if (old_type != 0 && new_type == 0) {
		ASSERT(dsp->ds_mph != NULL);

		mac_promisc_remove(dsp->ds_mph);
		dsp->ds_promisc = new_flags;
		dsp->ds_mph = NULL;

		if (dsp->ds_sap == ETHERTYPE_VLAN &&
		    dsp->ds_dlstate != DL_UNBOUND) {
			if (dsp->ds_vlan_mph != NULL)
				return (EINVAL);
			err = mac_promisc_add(dsp->ds_mch,
			    MAC_CLIENT_PROMISC_ALL, dls_rx_vlan_promisc, dsp,
			    &dsp->ds_vlan_mph, MAC_PROMISC_FLAGS_NO_PHYS);
		}
	} else if (new_flags != old_flags) {
		ASSERT(dsp->ds_mph != NULL);
		mac_promisc_remove(dsp->ds_mph);
		/* Honors both after-remove and before-add semantics! */
		dsp->ds_promisc = new_flags;
		err = mac_promisc_add(dsp->ds_mch, mptype,
		    dls_rx_promisc, dsp, &dsp->ds_mph, mac_flags);
		if (err != 0)
			dsp->ds_promisc = old_flags;
	}

	return (err);
}

int
dls_multicst_add(dld_str_t *dsp, const uint8_t *addr)
{
	int			err;
	dls_multicst_addr_t	**pp;
	dls_multicst_addr_t	*p;
	uint_t			addr_length;

	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));

	/*
	 * Check whether the address is in the list of enabled addresses for
	 * this dld_str_t.
	 */
	addr_length = dsp->ds_mip->mi_addr_length;

	/*
	 * Protect against concurrent access of ds_dmap by data threads using
	 * ds_rw_lock. The mac perimeter serializes the dls_multicst_add and
	 * remove operations. Dropping the ds_rw_lock across mac calls is thus
	 * ok and is also required by the locking protocol.
	 */
	rw_enter(&dsp->ds_rw_lock, RW_WRITER);
	for (pp = &(dsp->ds_dmap); (p = *pp) != NULL; pp = &(p->dma_nextp)) {
		if (bcmp(addr, p->dma_addr, addr_length) == 0) {
			/*
			 * It is there so there's nothing to do.
			 */
			err = 0;
			goto done;
		}
	}

	/*
	 * Allocate a new list item and add it to the list.
	 */
	p = kmem_zalloc(sizeof (dls_multicst_addr_t), KM_SLEEP);
	bcopy(addr, p->dma_addr, addr_length);
	*pp = p;
	rw_exit(&dsp->ds_rw_lock);

	/*
	 * Enable the address at the MAC.
	 */
	err = mac_multicast_add(dsp->ds_mch, addr);
	if (err == 0)
		return (0);

	/* Undo the operation as it has failed */
	rw_enter(&dsp->ds_rw_lock, RW_WRITER);
	ASSERT(*pp == p && p->dma_nextp == NULL);
	*pp = NULL;
	kmem_free(p, sizeof (dls_multicst_addr_t));
done:
	rw_exit(&dsp->ds_rw_lock);
	return (err);
}

int
dls_multicst_remove(dld_str_t *dsp, const uint8_t *addr)
{
	dls_multicst_addr_t	**pp;
	dls_multicst_addr_t	*p;
	uint_t			addr_length;

	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));

	/*
	 * Find the address in the list of enabled addresses for this
	 * dld_str_t.
	 */
	addr_length = dsp->ds_mip->mi_addr_length;

	/*
	 * Protect against concurrent access to ds_dmap by data threads using
	 * ds_rw_lock. The mac perimeter serializes the dls_multicst_add and
	 * remove operations. Dropping the ds_rw_lock across mac calls is thus
	 * ok and is also required by the locking protocol.
	 */
	rw_enter(&dsp->ds_rw_lock, RW_WRITER);
	for (pp = &(dsp->ds_dmap); (p = *pp) != NULL; pp = &(p->dma_nextp)) {
		if (bcmp(addr, p->dma_addr, addr_length) == 0)
			break;
	}

	/*
	 * If we walked to the end of the list then the given address is
	 * not currently enabled for this dld_str_t.
	 */
	if (p == NULL) {
		rw_exit(&dsp->ds_rw_lock);
		return (ENOENT);
	}

	/*
	 * Remove the address from the list.
	 */
	*pp = p->dma_nextp;
	rw_exit(&dsp->ds_rw_lock);

	/*
	 * Disable the address at the MAC.
	 */
	mac_multicast_remove(dsp->ds_mch, addr);
	kmem_free(p, sizeof (dls_multicst_addr_t));
	return (0);
}

mblk_t *
dls_header(dld_str_t *dsp, const uint8_t *addr, uint16_t sap, uint_t pri,
    mblk_t **payloadp)
{
	uint16_t vid;
	size_t extra_len;
	uint16_t mac_sap;
	mblk_t *mp, *payload;
	boolean_t is_ethernet = (dsp->ds_mip->mi_media == DL_ETHER);
	struct ether_vlan_header *evhp;

	vid = mac_client_vid(dsp->ds_mch);
	payload = (payloadp == NULL) ? NULL : (*payloadp);

	/*
	 * In the case of Ethernet, we need to tell mac_header() if we need
	 * extra room beyond the Ethernet header for a VLAN header.  We'll
	 * need to add a VLAN header if this isn't an ETHERTYPE_VLAN listener
	 * (because such streams will be handling VLAN headers on their own)
	 * and one of the following conditions is satisfied:
	 *
	 * - This is a VLAN stream
	 * - This is a physical stream, the priority is not 0, and user
	 *   priority tagging is allowed.
	 */
	if (is_ethernet && sap != ETHERTYPE_VLAN &&
	    (vid != VLAN_ID_NONE ||
	    (pri != 0 && dsp->ds_dlp->dl_tagmode != LINK_TAGMODE_VLANONLY))) {
		extra_len = sizeof (struct ether_vlan_header) -
		    sizeof (struct ether_header);
		mac_sap = ETHERTYPE_VLAN;
	} else {
		extra_len = 0;
		mac_sap = sap;
	}

	mp = mac_header(dsp->ds_mh, addr, mac_sap, payload, extra_len);
	if (mp == NULL)
		return (NULL);

	if ((vid == VLAN_ID_NONE && (pri == 0 ||
	    dsp->ds_dlp->dl_tagmode == LINK_TAGMODE_VLANONLY)) || !is_ethernet)
		return (mp);

	/*
	 * Fill in the tag information.
	 */
	ASSERT(MBLKL(mp) == sizeof (struct ether_header));
	if (extra_len != 0) {
		mp->b_wptr += extra_len;
		evhp = (struct ether_vlan_header *)mp->b_rptr;
		evhp->ether_tci = htons(VLAN_TCI(pri, ETHER_CFI, vid));
		evhp->ether_type = htons(sap);
	} else {
		/*
		 * The stream is ETHERTYPE_VLAN listener, so its VLAN tag is
		 * in the payload. Update the priority.
		 */
		struct ether_vlan_extinfo *extinfo;
		size_t len = sizeof (struct ether_vlan_extinfo);

		ASSERT(sap == ETHERTYPE_VLAN);
		ASSERT(payload != NULL);

		if ((DB_REF(payload) > 1) || (MBLKL(payload) < len)) {
			mblk_t *newmp;

			/*
			 * Because some DLS consumers only check the db_ref
			 * count of the first mblk, we pullup 'payload' into
			 * a single mblk.
			 */
			newmp = msgpullup(payload, -1);
			if ((newmp == NULL) || (MBLKL(newmp) < len)) {
				freemsg(newmp);
				freemsg(mp);
				return (NULL);
			} else {
				freemsg(payload);
				*payloadp = payload = newmp;
			}
		}

		extinfo = (struct ether_vlan_extinfo *)payload->b_rptr;
		extinfo->ether_tci = htons(VLAN_TCI(pri, ETHER_CFI,
		    VLAN_ID(ntohs(extinfo->ether_tci))));
	}
	return (mp);
}

void
dls_rx_set(dld_str_t *dsp, dls_rx_t rx, void *arg)
{
	mutex_enter(&dsp->ds_lock);
	dsp->ds_rx = rx;
	dsp->ds_rx_arg = arg;
	mutex_exit(&dsp->ds_lock);
}

static boolean_t
dls_accept_common(dld_str_t *dsp, mac_header_info_t *mhip, dls_rx_t *ds_rx,
    void **ds_rx_arg, boolean_t promisc, boolean_t promisc_loopback)
{
	dls_multicst_addr_t	*dmap;
	size_t			addr_length = dsp->ds_mip->mi_addr_length;

	/*
	 * We must not accept packets if the dld_str_t is not marked as bound
	 * or is being removed.
	 */
	if (dsp->ds_dlstate != DL_IDLE)
		goto refuse;

	if (dsp->ds_promisc != 0) {
		/*
		 * Filter out packets that arrived from the data path
		 * (i_dls_link_rx) when promisc mode is on. We need to correlate
		 * the ds_promisc flags with the mac header destination type. If
		 * only DLS_PROMISC_MULTI is enabled, we need to only reject
		 * multicast packets as those are the only ones which filter up
		 * the promiscuous path. If we have DLS_PROMISC_PHYS or
		 * DLS_PROMISC_SAP set, then we know that we'll be seeing
		 * everything, so we should drop it now.
		 */
		if (!promisc && !(dsp->ds_promisc == DLS_PROMISC_MULTI &&
		    mhip->mhi_dsttype != MAC_ADDRTYPE_MULTICAST))
			goto refuse;
		/*
		 * If the dls_impl_t is in 'all physical' mode then
		 * always accept.
		 */
		if (dsp->ds_promisc & DLS_PROMISC_PHYS)
			goto accept;

		/*
		 * Loopback packets i.e. packets sent out by DLS on a given
		 * mac end point, will be accepted back by DLS on loopback
		 * from the mac, only in the 'all physical' mode which has been
		 * covered by the previous check above
		 */
		if (promisc_loopback)
			goto refuse;
	}

	switch (mhip->mhi_dsttype) {
	case MAC_ADDRTYPE_UNICAST:
	case MAC_ADDRTYPE_BROADCAST:
		/*
		 * We can accept unicast and broadcast packets because
		 * filtering is already done by the mac layer.
		 */
		goto accept;
	case MAC_ADDRTYPE_MULTICAST:
		/*
		 * Additional filtering is needed for multicast addresses
		 * because different streams may be interested in different
		 * addresses.
		 */
		if (dsp->ds_promisc & DLS_PROMISC_MULTI)
			goto accept;

		rw_enter(&dsp->ds_rw_lock, RW_READER);
		for (dmap = dsp->ds_dmap; dmap != NULL;
		    dmap = dmap->dma_nextp) {
			if (memcmp(mhip->mhi_daddr, dmap->dma_addr,
			    addr_length) == 0) {
				rw_exit(&dsp->ds_rw_lock);
				goto accept;
			}
		}
		rw_exit(&dsp->ds_rw_lock);
		break;
	}

refuse:
	return (B_FALSE);

accept:
	/*
	 * the returned ds_rx and ds_rx_arg will always be in sync.
	 */
	mutex_enter(&dsp->ds_lock);
	*ds_rx = dsp->ds_rx;
	*ds_rx_arg = dsp->ds_rx_arg;
	mutex_exit(&dsp->ds_lock);

	return (B_TRUE);
}

/* ARGSUSED */
boolean_t
dls_accept(dld_str_t *dsp, mac_header_info_t *mhip, dls_rx_t *ds_rx,
    void **ds_rx_arg)
{
	return (dls_accept_common(dsp, mhip, ds_rx, ds_rx_arg, B_FALSE,
	    B_FALSE));
}

boolean_t
dls_accept_promisc(dld_str_t *dsp, mac_header_info_t *mhip, dls_rx_t *ds_rx,
    void **ds_rx_arg, boolean_t loopback)
{
	return (dls_accept_common(dsp, mhip, ds_rx, ds_rx_arg, B_TRUE,
	    loopback));
}

int
dls_mac_active_set(dls_link_t *dlp)
{
	int err = 0;

	/*
	 * First client; add the primary unicast address.
	 */
	if (dlp->dl_nactive == 0) {
		/*
		 * First client; add the primary unicast address.
		 */
		mac_diag_t diag;

		/* request the primary MAC address */
		if ((err = mac_unicast_add(dlp->dl_mch, NULL,
		    MAC_UNICAST_PRIMARY | MAC_UNICAST_TAG_DISABLE |
		    MAC_UNICAST_DISABLE_TX_VID_CHECK, &dlp->dl_mah,
		    VLAN_ID_NONE, &diag)) != 0) {
			return (err);
		}

		/*
		 * Set the function to start receiving packets.
		 */
		mac_rx_set(dlp->dl_mch, i_dls_link_rx, dlp);
	}
	dlp->dl_nactive++;
	return (0);
}

void
dls_mac_active_clear(dls_link_t *dlp)
{
	if (--dlp->dl_nactive == 0) {
		ASSERT(dlp->dl_mah != NULL);
		(void) mac_unicast_remove(dlp->dl_mch, dlp->dl_mah);
		dlp->dl_mah = NULL;
		mac_rx_clear(dlp->dl_mch);
	}
}

int
dls_active_set(dld_str_t *dsp)
{
	int err = 0;

	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));

	if (dsp->ds_passivestate == DLD_PASSIVE)
		return (0);

	/* If we're already active, then there's nothing more to do. */
	if ((dsp->ds_nactive == 0) &&
	    ((err = dls_mac_active_set(dsp->ds_dlp)) != 0)) {
		/* except for ENXIO all other errors are mapped to EBUSY */
		if (err != ENXIO)
			return (EBUSY);
		return (err);
	}

	dsp->ds_passivestate = DLD_ACTIVE;
	dsp->ds_nactive++;
	return (0);
}

/*
 * Note that dls_active_set() is called whenever an active operation
 * (DL_BIND_REQ, DL_ENABMULTI_REQ ...) is processed and
 * dls_active_clear(dsp, B_FALSE) is called whenever the active operation
 * is being undone (DL_UNBIND_REQ, DL_DISABMULTI_REQ ...). In some cases,
 * a stream is closed without every active operation being undone and we
 * need to clear all the "active" states by calling
 * dls_active_clear(dsp, B_TRUE).
 */
void
dls_active_clear(dld_str_t *dsp, boolean_t all)
{
	ASSERT(MAC_PERIM_HELD(dsp->ds_mh));

	if (dsp->ds_passivestate == DLD_PASSIVE)
		return;

	if (all && dsp->ds_nactive == 0)
		return;

	ASSERT(dsp->ds_nactive > 0);

	dsp->ds_nactive -= (all ? dsp->ds_nactive : 1);
	if (dsp->ds_nactive != 0)
		return;

	ASSERT(dsp->ds_passivestate == DLD_ACTIVE);
	dls_mac_active_clear(dsp->ds_dlp);
	dsp->ds_passivestate = DLD_UNINITIALIZED;
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 * Copyright 2019 Joyent, Inc.
 */

/*
 * Data-Link Services Module
 */

#include	<sys/sysmacros.h>
#include	<sys/strsubr.h>
#include	<sys/strsun.h>
#include	<sys/vlan.h>
#include	<sys/dld_impl.h>
#include	<sys/sdt.h>
#include	<sys/atomic.h>
#include	<sys/sysevent.h>
#include	<sys/sysevent/eventdefs.h>
#include	<sys/sysevent/datalink.h>

static kmem_cache_t	*i_dls_link_cachep;
mod_hash_t		*i_dls_link_hash;
static uint_t		i_dls_link_count;

#define		LINK_HASHSZ	67	/* prime */
#define		IMPL_HASHSZ	67	/* prime */

/*
 * Construct a hash key from the DLSAP value.
 */
#define	MAKE_KEY(_sap)						\
	((mod_hash_key_t)(uintptr_t)((_sap) << VLAN_ID_SIZE))

#define	DLS_STRIP_PADDING(pktsize, p) {			\
	if (pktsize != 0) {				\
		ssize_t delta = pktsize - msgdsize(p);	\
							\
		if (delta < 0)				\
			(void) adjmsg(p, delta);	\
	}						\
}

/*
 * Private functions.
 */

/*ARGSUSED*/
static int
i_dls_link_constructor(void *buf, void *arg, int kmflag)
{
	dls_link_t	*dlp = buf;
	char		name[MAXNAMELEN];

	bzero(buf, sizeof (dls_link_t));

	(void) snprintf(name, MAXNAMELEN, "dls_link_t_%p_hash", buf);
	dlp->dl_str_hash = mod_hash_create_idhash(name, IMPL_HASHSZ,
	    mod_hash_null_valdtor);

	return (0);
}

/*ARGSUSED*/
static void
i_dls_link_destructor(void *buf, void *arg)
{
	dls_link_t	*dlp = buf;

	ASSERT(dlp->dl_ref == 0);
	ASSERT(dlp->dl_mh == NULL);
	ASSERT(dlp->dl_mah == NULL);
	ASSERT(dlp->dl_unknowns == 0);

	mod_hash_destroy_idhash(dlp->dl_str_hash);
	dlp->dl_str_hash = NULL;

}

/*
 * - Parse the mac header information of the given packet.
 * - Strip the padding and skip over the header. Note that because some
 *   DLS consumers only check the db_ref count of the first mblk, we
 *   pullup the message into a single mblk. Because the original message
 *   is freed as the result of message pulling up, mac_vlan_header_info()
 *   is called again to update the mhi_saddr and mhi_daddr pointers in the
 *   mhip. Further, the mac_vlan_header_info() function ensures that the
 *   size of the pulled message is greater than the MAC header size,
 *   therefore we can directly advance b_rptr to point at the payload.
 *
 * We choose to use a macro for performance reasons.
 */
#define	DLS_PREPARE_PKT(mh, mp, mhip, err) {				\
	mblk_t *nextp = (mp)->b_next;					\
	if (((err) = mac_vlan_header_info((mh), (mp), (mhip))) == 0) {	\
		DLS_STRIP_PADDING((mhip)->mhi_pktsize, (mp));		\
		if (MBLKL((mp)) < (mhip)->mhi_hdrsize) {		\
			mblk_t *newmp;					\
			if ((newmp = msgpullup((mp), -1)) == NULL) {	\
				(err) = EINVAL;				\
			} else {					\
				(mp)->b_next = NULL;			\
				freemsg((mp));				\
				(mp) = newmp;				\
				VERIFY(mac_vlan_header_info((mh),	\
				    (mp), (mhip)) == 0);		\
				(mp)->b_next = nextp;			\
				(mp)->b_rptr += (mhip)->mhi_hdrsize;	\
			}						\
		} else {						\
			(mp)->b_rptr += (mhip)->mhi_hdrsize;		\
		}							\
	}								\
}

/*
 * Truncate the chain starting at mp such that all packets in the chain
 * have identical source and destination addresses, saps, and tag types
 * (see below).  It returns a pointer to the mblk following the chain,
 * NULL if there is no further packet following the processed chain.
 * The countp argument is set to the number of valid packets in the chain.
 * Note that the whole MAC header (including the VLAN tag if any) in each
 * packet will be stripped.
 */
static mblk_t *
i_dls_link_subchain(dls_link_t *dlp, mblk_t *mp, const mac_header_info_t *mhip,
    uint_t *countp)
{
	mblk_t		*prevp;
	uint_t		npacket = 1;
	size_t		addr_size = dlp->dl_mip->mi_addr_length;
	uint16_t	vid = VLAN_ID(mhip->mhi_tci);
	uint16_t	pri = VLAN_PRI(mhip->mhi_tci);

	/*
	 * Compare with subsequent headers until we find one that has
	 * differing header information. After checking each packet
	 * strip padding and skip over the header.
	 */
	for (prevp = mp; (mp = mp->b_next) != NULL; prevp = mp) {
		mac_header_info_t cmhi;
		uint16_t cvid, cpri;
		int err;

		DLS_PREPARE_PKT(dlp->dl_mh, mp, &cmhi, err);
		if (err != 0)
			break;

		prevp->b_next = mp;

		/*
		 * The source, destination, sap, vlan tag must all match in
		 * a given subchain.
		 */
		if (mhip->mhi_saddr == NULL || cmhi.mhi_saddr == NULL ||
		    memcmp(mhip->mhi_daddr, cmhi.mhi_daddr, addr_size) != 0 ||
		    memcmp(mhip->mhi_saddr, cmhi.mhi_saddr, addr_size) != 0 ||
		    mhip->mhi_bindsap != cmhi.mhi_bindsap) {
			/*
			 * Note that we don't need to restore the padding.
			 */
			mp->b_rptr -= cmhi.mhi_hdrsize;
			break;
		}

		cvid = VLAN_ID(cmhi.mhi_tci);
		cpri = VLAN_PRI(cmhi.mhi_tci);

		/*
		 * There are several types of packets. Packets don't match
		 * if they are classified to different type or if they are
		 * VLAN packets but belong to different VLANs:
		 *
		 * packet type		tagged		vid		pri
		 * ---------------------------------------------------------
		 * untagged		No		zero		zero
		 * VLAN packets		Yes		non-zero	-
		 * priority tagged	Yes		zero		non-zero
		 * 0 tagged		Yes		zero		zero
		 */
		if ((mhip->mhi_istagged != cmhi.mhi_istagged) ||
		    (vid != cvid) || ((vid == VLAN_ID_NONE) &&
		    (((pri == 0) && (cpri != 0)) ||
		    ((pri != 0) && (cpri == 0))))) {
			mp->b_rptr -= cmhi.mhi_hdrsize;
			break;
		}

		npacket++;
	}

	/*
	 * Break the chain at this point and return a pointer to the next
	 * sub-chain.
	 */
	prevp->b_next = NULL;
	*countp = npacket;
	return (mp);
}

/* ARGSUSED */
static int
i_dls_head_hold(mod_hash_key_t key, mod_hash_val_t val)
{
	dls_head_t *dhp = (dls_head_t *)val;

	/*
	 * The lock order is  mod_hash's internal lock -> dh_lock as in the
	 * call to i_dls_link_rx -> mod_hash_find_cb_rval -> i_dls_head_hold
	 */
	mutex_enter(&dhp->dh_lock);
	if (dhp->dh_removing) {
		mutex_exit(&dhp->dh_lock);
		return (-1);
	}
	dhp->dh_ref++;
	mutex_exit(&dhp->dh_lock);
	return (0);
}

void
i_dls_head_rele(dls_head_t *dhp)
{
	mutex_enter(&dhp->dh_lock);
	dhp->dh_ref--;
	if (dhp->dh_ref == 0 && dhp->dh_removing != 0)
		cv_broadcast(&dhp->dh_cv);
	mutex_exit(&dhp->dh_lock);
}

static dls_head_t *
i_dls_head_alloc(mod_hash_key_t key)
{
	dls_head_t	*dhp;

	dhp = kmem_zalloc(sizeof (dls_head_t), KM_SLEEP);
	dhp->dh_key = key;
	return (dhp);
}

static void
i_dls_head_free(dls_head_t *dhp)
{
	ASSERT(dhp->dh_ref == 0);
	kmem_free(dhp, sizeof (dls_head_t));
}

/*
 * Try to send mp up to the streams of the given sap. Return the
 * number of streams which accepted this message, or 0 if no streams
 * accepted the message.
 *
 * Note that this function copies the message chain and the original
 * mp remains valid after this function returns.
 */
static uint_t
i_dls_link_rx_func(dls_link_t *dlp, mac_resource_handle_t mrh,
    mac_header_info_t *mhip, mblk_t *mp, uint32_t sap,
    boolean_t (*acceptfunc)())
{
	mod_hash_t	*hash = dlp->dl_str_hash;
	mod_hash_key_t	key;
	dls_head_t	*dhp;
	dld_str_t	*dsp;
	mblk_t		*nmp;
	dls_rx_t	ds_rx;
	void		*ds_rx_arg;
	uint_t		naccepted = 0;
	int		rval;

	/*
	 * Construct a hash key from the DLSAP.
	 */
	key = MAKE_KEY(sap);

	/*
	 * Search the hash table for a dld_str_t eligible to receive a
	 * packet chain for this DLSAP. The mod hash's internal lock
	 * serializes find/insert/remove from the mod hash list.
	 * Incrementing the dh_ref (while holding the mod hash lock)
	 * ensures dls_link_remove will wait for the upcall to finish.
	 */
	if (mod_hash_find_cb_rval(hash, key, (mod_hash_val_t *)&dhp,
	    i_dls_head_hold, &rval) != 0 || (rval != 0)) {
		return (0);
	}

	/*
	 * Find all dld_str_t that will accept the sub-chain.
	 */
	for (dsp = dhp->dh_list; dsp != NULL; dsp = dsp->ds_next) {
		if (!acceptfunc(dsp, mhip, &ds_rx, &ds_rx_arg))
			continue;

		/*
		 * We have at least one acceptor.
		 */
		naccepted++;

		/*
		 * There will normally be at least one more dld_str_t
		 * (since we've yet to check for non-promiscuous
		 * dld_str_t) so dup the sub-chain.
		 */
		if ((nmp = copymsgchain(mp)) != NULL)
			ds_rx(ds_rx_arg, mrh, nmp, mhip);
	}

	/*
	 * Release the hold on the dld_str_t chain now that we have
	 * finished walking it.
	 */
	i_dls_head_rele(dhp);
	return (naccepted);
}

/* ARGSUSED */
void
i_dls_link_rx(void *arg, mac_resource_handle_t mrh, mblk_t *mp,
    boolean_t loopback)
{
	dls_link_t			*dlp = arg;
	mod_hash_t			*hash = dlp->dl_str_hash;
	mblk_t				*nextp;
	mac_header_info_t		mhi;
	dls_head_t			*dhp;
	dld_str_t			*dsp;
	dld_str_t			*ndsp;
	mblk_t				*nmp;
	mod_hash_key_t			key;
	uint_t				npacket;
	boolean_t			accepted;
	dls_rx_t			ds_rx, nds_rx;
	void				*ds_rx_arg, *nds_rx_arg;
	uint16_t			vid;
	int				err, rval;

	/*
	 * Walk the packet chain.
	 */
	for (; mp != NULL; mp = nextp) {
		/*
		 * Wipe the accepted state.
		 */
		accepted = B_FALSE;

		DLS_PREPARE_PKT(dlp->dl_mh, mp, &mhi, err);
		if (err != 0) {
			atomic_inc_32(&(dlp->dl_unknowns));
			nextp = mp->b_next;
			mp->b_next = NULL;
			freemsg(mp);
			continue;
		}

		/*
		 * Grab the longest sub-chain we can process as a single
		 * unit.
		 */
		nextp = i_dls_link_subchain(dlp, mp, &mhi, &npacket);
		ASSERT(npacket != 0);

		vid = VLAN_ID(mhi.mhi_tci);

		/*
		 * This condition is true only when a sun4v vsw client
		 * is on the scene; as it is the only type of client
		 * that multiplexes VLANs on a single client instance.
		 * All other types of clients have one VLAN per client
		 * instance. In that case, MAC strips the VLAN tag
		 * before delivering it to DLS (see mac_rx_deliver()).
		 */
		if (mhi.mhi_istagged) {

			/*
			 * If it is tagged traffic, send it upstream to
			 * all dld_str_t which are attached to the physical
			 * link and bound to SAP 0x8100.
			 */
			if (i_dls_link_rx_func(dlp, mrh, &mhi, mp,
			    ETHERTYPE_VLAN, dls_accept) > 0) {
				accepted = B_TRUE;
			}

			/*
			 * Don't pass the packets up if they are tagged
			 * packets and:
			 *  - their VID and priority are both zero and the
			 *    original packet isn't using the PVID (invalid
			 *    packets).
			 *  - their sap is ETHERTYPE_VLAN and their VID is
			 *    zero as they have already been sent upstreams.
			 */
			if ((vid == VLAN_ID_NONE && !mhi.mhi_ispvid &&
			    VLAN_PRI(mhi.mhi_tci) == 0) ||
			    (mhi.mhi_bindsap == ETHERTYPE_VLAN &&
			    vid == VLAN_ID_NONE)) {
				freemsgchain(mp);
				goto loop;
			}
		}

		/*
		 * Construct a hash key from the DLSAP.
		 */
		key = MAKE_KEY(mhi.mhi_bindsap);

		/*
		 * Search the hash table for dld_str_t eligible to receive
		 * a packet chain for this DLSAP.
		 */
		if (mod_hash_find_cb_rval(hash, key, (mod_hash_val_t *)&dhp,
		    i_dls_head_hold, &rval) != 0 || (rval != 0)) {
			freemsgchain(mp);
			goto loop;
		}

		/*
		 * Find the first dld_str_t that will accept the sub-chain.
		 */
		for (dsp = dhp->dh_list; dsp != NULL; dsp = dsp->ds_next)
			if (dls_accept(dsp, &mhi, &ds_rx, &ds_rx_arg))
				break;

		/*
		 * If we did not find any dld_str_t willing to accept the
		 * sub-chain then throw it away.
		 */
		if (dsp == NULL) {
			i_dls_head_rele(dhp);
			freemsgchain(mp);
			goto loop;
		}

		/*
		 * We have at least one acceptor.
		 */
		accepted = B_TRUE;
		for (;;) {
			/*
			 * Find the next dld_str_t that will accept the
			 * sub-chain.
			 */
			for (ndsp = dsp->ds_next; ndsp != NULL;
			    ndsp = ndsp->ds_next)
				if (dls_accept(ndsp, &mhi, &nds_rx,
				    &nds_rx_arg))
					break;

			/*
			 * If there are no more dld_str_t that are willing
			 * to accept the sub-chain then we don't need to dup
			 * it before handing it to the current one.
			 */
			if (ndsp == NULL) {
				ds_rx(ds_rx_arg, mrh, mp, &mhi);

				/*
				 * Since there are no more dld_str_t, we're
				 * done.
				 */
				break;
			}

			/*
			 * There are more dld_str_t so dup the sub-chain.
			 */
			if ((nmp = copymsgchain(mp)) != NULL)
				ds_rx(ds_rx_arg, mrh, nmp, &mhi);

			dsp = ndsp;
			ds_rx = nds_rx;
			ds_rx_arg = nds_rx_arg;
		}

		/*
		 * Release the hold on the dld_str_t chain now that we have
		 * finished walking it.
		 */
		i_dls_head_rele(dhp);

loop:
		/*
		 * If there were no acceptors then add the packet count to the
		 * 'unknown' count.
		 */
		if (!accepted)
			atomic_add_32(&(dlp->dl_unknowns), npacket);
	}
}

/* ARGSUSED */
void
dls_rx_vlan_promisc(void *arg, mac_resource_handle_t mrh, mblk_t *mp,
    boolean_t loopback)
{
	dld_str_t			*dsp = arg;
	dls_link_t			*dlp = dsp->ds_dlp;
	mac_header_info_t		mhi;
	dls_rx_t			ds_rx;
	void				*ds_rx_arg;
	int				err;

	DLS_PREPARE_PKT(dlp->dl_mh, mp, &mhi, err);
	if (err != 0)
		goto drop;

	/*
	 * If there is promiscuous handle for vlan, we filter out the untagged
	 * pkts and pkts that are not for the primary unicast address.
	 */
	if (dsp->ds_vlan_mph != NULL) {
		uint8_t prim_addr[MAXMACADDRLEN];
		size_t	addr_length = dsp->ds_mip->mi_addr_length;

		if (!(mhi.mhi_istagged))
			goto drop;
		ASSERT(dsp->ds_mh != NULL);
		mac_unicast_primary_get(dsp->ds_mh, (uint8_t *)prim_addr);
		if (memcmp(mhi.mhi_daddr, prim_addr, addr_length) != 0)
			goto drop;

		if (!dls_accept(dsp, &mhi, &ds_rx, &ds_rx_arg))
			goto drop;

		ds_rx(ds_rx_arg, NULL, mp, &mhi);
		return;
	}

drop:
	atomic_inc_32(&dlp->dl_unknowns);
	freemsg(mp);
}

/* ARGSUSED */
void
dls_rx_promisc(void *arg, mac_resource_handle_t mrh, mblk_t *mp,
    boolean_t loopback)
{
	dld_str_t			*dsp = arg;
	dls_link_t			*dlp = dsp->ds_dlp;
	mac_header_info_t		mhi;
	dls_rx_t			ds_rx;
	void				*ds_rx_arg;
	int				err;
	dls_head_t			*dhp;
	mod_hash_key_t			key;

	/*
	 * We expect to deal with only a single packet.
	 */
	ASSERT3P(mp->b_next, ==, NULL);

	DLS_PREPARE_PKT(dlp->dl_mh, mp, &mhi, err);

	if (err != 0)
		goto drop;

	/*
	 * In order to filter out sap pkt that no dls channel listens, search
	 * the hash table trying to find a dld_str_t eligible to receive the pkt
	 */
	if ((dsp->ds_promisc & DLS_PROMISC_SAP) == 0) {
		key = MAKE_KEY(mhi.mhi_bindsap);
		if (mod_hash_find(dsp->ds_dlp->dl_str_hash, key,
		    (mod_hash_val_t *)&dhp) != 0)
			goto drop;
	}

	if (!dls_accept_promisc(dsp, &mhi, &ds_rx, &ds_rx_arg, loopback))
		goto drop;

	ds_rx(ds_rx_arg, NULL, mp, &mhi);
	return;

drop:
	atomic_inc_32(&dlp->dl_unknowns);
	freemsg(mp);
}

/*
 * We'd like to notify via sysevents that a link state change has occurred.
 * There are a couple of challenges associated with this. The first is that if
 * the link is flapping a lot, we may not see an accurate state when we launch
 * the notification, we're told it changed, not what it changed to.
 *
 * The next problem is that all of the information that a user has associated
 * with this device is the exact opposite of what we have on the dls_link_t. We
 * have the name of the mac device, which has no bearing on what users see.
 * Likewise, we don't have the datalink id either. So we're going to have to get
 * this from dls.
 *
 * This is all further complicated by the fact that this could be going on in
 * another thread at the same time as someone is tearing down the dls_link_t
 * that we're associated with. We need to be careful not to grab the mac
 * perimeter, otherwise we stand a good chance of deadlock.
 */
static void
dls_link_notify(void *arg, mac_notify_type_t type)
{
	dls_link_t	*dlp = arg;
	dls_dl_handle_t	dhp;
	nvlist_t	*nvp;
	sysevent_t	*event;
	sysevent_id_t	eid;

	if (type != MAC_NOTE_LINK && type != MAC_NOTE_LOWLINK)
		return;

	/*
	 * If we can't find a devnet handle for this link, then there is no user
	 * knowable device for this at the moment and there's nothing we can
	 * really share with them that will make sense.
	 */
	if (dls_devnet_hold_tmp_by_link(dlp, &dhp) != 0)
		return;

	/*
	 * Because we're attaching this nvlist_t to the sysevent, it'll get
	 * cleaned up when we call sysevent_free.
	 */
	VERIFY(nvlist_alloc(&nvp, NV_UNIQUE_NAME, KM_SLEEP) == 0);
	VERIFY(nvlist_add_int32(nvp, DATALINK_EV_LINK_ID,
	    dls_devnet_linkid(dhp)) == 0);
	VERIFY(nvlist_add_string(nvp, DATALINK_EV_LINK_NAME,
	    dls_devnet_link(dhp)) == 0);
	VERIFY(nvlist_add_int32(nvp, DATALINK_EV_ZONE_ID,
	    dls_devnet_getzid(dhp)) == 0);

	dls_devnet_rele_tmp(dhp);

	event = sysevent_alloc(EC_DATALINK, ESC_DATALINK_LINK_STATE,
	    ILLUMOS_KERN_PUB"dls", SE_SLEEP);
	VERIFY(event != NULL);
	(void) sysevent_attach_attributes(event, (sysevent_attr_list_t *)nvp);

	(void) log_sysevent(event, SE_SLEEP, &eid);
	sysevent_free(event);

}

static void
i_dls_link_destroy(dls_link_t *dlp)
{
	ASSERT(dlp->dl_nactive == 0);
	ASSERT(dlp->dl_impl_count == 0);
	ASSERT(dlp->dl_zone_ref == 0);

	/*
	 * Free the structure back to the cache.
	 */
	if (dlp->dl_mnh != NULL)
		mac_notify_remove(dlp->dl_mnh, B_TRUE);

	if (dlp->dl_mch != NULL)
		mac_client_close(dlp->dl_mch, 0);

	if (dlp->dl_mh != NULL) {
		ASSERT(MAC_PERIM_HELD(dlp->dl_mh));
		mac_close(dlp->dl_mh);
	}

	dlp->dl_mh = NULL;
	dlp->dl_mch = NULL;
	dlp->dl_mip = NULL;
	dlp->dl_mnh = NULL;
	dlp->dl_unknowns = 0;
	dlp->dl_nonip_cnt = 0;
	kmem_cache_free(i_dls_link_cachep, dlp);
}

static int
i_dls_link_create(const char *name, dls_link_t **dlpp)
{
	dls_link_t		*dlp;
	int			err;

	/*
	 * Allocate a new dls_link_t structure.
	 */
	dlp = kmem_cache_alloc(i_dls_link_cachep, KM_SLEEP);

	/*
	 * Name the dls_link_t after the MAC interface it represents.
	 */
	(void) strlcpy(dlp->dl_name, name, sizeof (dlp->dl_name));

	/*
	 * First reference; hold open the MAC interface.
	 */
	ASSERT(dlp->dl_mh == NULL);
	err = mac_open(dlp->dl_name, &dlp->dl_mh);
	if (err != 0)
		goto bail;

	ASSERT(MAC_PERIM_HELD(dlp->dl_mh));
	dlp->dl_mip = mac_info(dlp->dl_mh);

	/* DLS is the "primary" MAC client */
	ASSERT(dlp->dl_mch == NULL);

	err = mac_client_open(dlp->dl_mh, &dlp->dl_mch, NULL,
	    MAC_OPEN_FLAGS_USE_DATALINK_NAME);
	if (err != 0)
		goto bail;

	dlp->dl_mnh = mac_notify_add(dlp->dl_mh, dls_link_notify, dlp);

	DTRACE_PROBE2(dls__primary__client, char *, dlp->dl_name, void *,
	    dlp->dl_mch);

	*dlpp = dlp;
	return (0);

bail:
	i_dls_link_destroy(dlp);
	return (err);
}

/*
 * Module initialization functions.
 */

void
dls_link_init(void)
{
	/*
	 * Create a kmem_cache of dls_link_t structures.
	 */
	i_dls_link_cachep = kmem_cache_create("dls_link_cache",
	    sizeof (dls_link_t), 0, i_dls_link_constructor,
	    i_dls_link_destructor, NULL, NULL, NULL, 0);
	ASSERT(i_dls_link_cachep != NULL);

	/*
	 * Create a dls_link_t hash table and associated lock.
	 */
	i_dls_link_hash = mod_hash_create_extended("dls_link_hash",
	    IMPL_HASHSZ, mod_hash_null_keydtor, mod_hash_null_valdtor,
	    mod_hash_bystr, NULL, mod_hash_strkey_cmp, KM_SLEEP);
	i_dls_link_count = 0;
}

int
dls_link_fini(void)
{
	if (i_dls_link_count > 0)
		return (EBUSY);

	/*
	 * Destroy the kmem_cache.
	 */
	kmem_cache_destroy(i_dls_link_cachep);

	/*
	 * Destroy the hash table and associated lock.
	 */
	mod_hash_destroy_hash(i_dls_link_hash);
	return (0);
}

/*
 * Exported functions.
 */

static int
dls_link_hold_common(const char *name, dls_link_t **dlpp, boolean_t create)
{
	dls_link_t		*dlp;
	int			err;

	/*
	 * Look up a dls_link_t corresponding to the given macname in the
	 * global hash table. The i_dls_link_hash itself is protected by the
	 * mod_hash package's internal lock which synchronizes
	 * find/insert/remove into the global mod_hash list. Assumes that
	 * inserts and removes are single threaded on a per mac end point
	 * by the mac perimeter.
	 */
	if ((err = mod_hash_find(i_dls_link_hash, (mod_hash_key_t)name,
	    (mod_hash_val_t *)&dlp)) == 0)
		goto done;

	ASSERT(err == MH_ERR_NOTFOUND);
	if (!create)
		return (ENOENT);

	/*
	 * We didn't find anything so we need to create one.
	 */
	if ((err = i_dls_link_create(name, &dlp)) != 0)
		return (err);

	/*
	 * Insert the dls_link_t.
	 */
	err = mod_hash_insert(i_dls_link_hash, (mod_hash_key_t)dlp->dl_name,
	    (mod_hash_val_t)dlp);
	ASSERT(err == 0);

	atomic_inc_32(&i_dls_link_count);
	ASSERT(i_dls_link_count != 0);

done:
	ASSERT(MAC_PERIM_HELD(dlp->dl_mh));
	/*
	 * Bump the reference count and hand back the reference.
	 */
	dlp->dl_ref++;
	*dlpp = dlp;
	return (0);
}

int
dls_link_hold_create(const char *name, dls_link_t **dlpp)
{
	return (dls_link_hold_common(name, dlpp, B_TRUE));
}

int
dls_link_hold(const char *name, dls_link_t **dlpp)
{
	return (dls_link_hold_common(name, dlpp, B_FALSE));
}

dev_info_t *
dls_link_devinfo(dev_t dev)
{
	dls_link_t	*dlp;
	dev_info_t	*dip;
	char	macname[MAXNAMELEN];
	char	*drv;
	mac_perim_handle_t	mph;

	if ((drv = ddi_major_to_name(getmajor(dev))) == NULL)
		return (NULL);
	(void) snprintf(macname, MAXNAMELEN, "%s%d", drv,
	    DLS_MINOR2INST(getminor(dev)));

	/*
	 * The code below assumes that the name constructed above is the
	 * macname. This is not the case for legacy devices. Currently this
	 * is ok because this function is only called in the getinfo(9e) path,
	 * which for a legacy device would directly end up in the driver's
	 * getinfo, rather than here
	 */
	if (mac_perim_enter_by_macname(macname, &mph) != 0)
		return (NULL);

	if (dls_link_hold(macname, &dlp) != 0) {
		mac_perim_exit(mph);
		return (NULL);
	}

	dip = mac_devinfo_get(dlp->dl_mh);
	dls_link_rele(dlp);
	mac_perim_exit(mph);

	return (dip);
}

dev_t
dls_link_dev(dls_link_t *dlp)
{
	return (makedevice(ddi_driver_major(mac_devinfo_get(dlp->dl_mh)),
	    mac_minor(dlp->dl_mh)));
}

void
dls_link_rele(dls_link_t *dlp)
{
	mod_hash_val_t	val;

	ASSERT(MAC_PERIM_HELD(dlp->dl_mh));
	/*
	 * Check if there are any more references.
	 */
	if (--dlp->dl_ref == 0) {
		(void) mod_hash_remove(i_dls_link_hash,
		    (mod_hash_key_t)dlp->dl_name, &val);
		ASSERT(dlp == (dls_link_t *)val);

		/*
		 * Destroy the dls_link_t.
		 */
		i_dls_link_destroy(dlp);
		ASSERT(i_dls_link_count > 0);
		atomic_dec_32(&i_dls_link_count);
	}
}

int
dls_link_rele_by_name(const char *name)
{
	dls_link_t		*dlp;

	if (mod_hash_find(i_dls_link_hash, (mod_hash_key_t)name,
	    (mod_hash_val_t *)&dlp) != 0)
		return (ENOENT);

	ASSERT(MAC_PERIM_HELD(dlp->dl_mh));

	/*
	 * Must fail detach if mac client is busy.
	 */
	ASSERT(dlp->dl_ref > 0 && dlp->dl_mch != NULL);
	if (mac_link_has_flows(dlp->dl_mch))
		return (ENOTEMPTY);

	dls_link_rele(dlp);
	return (0);
}

int
dls_link_setzid(const char *name, zoneid_t zid)
{
	dls_link_t	*dlp;
	int		err = 0;
	zoneid_t	old_zid;

	if ((err = dls_link_hold_create(name, &dlp)) != 0)
		return (err);

	ASSERT(MAC_PERIM_HELD(dlp->dl_mh));

	if ((old_zid = dlp->dl_zid) == zid)
		goto done;

	/*
	 * Check whether this dlp is used by its own zone.  If yes, we cannot
	 * change its zoneid.
	 */
	if (dlp->dl_zone_ref != 0) {
		err = EBUSY;
		goto done;
	}

	dlp->dl_zid = zid;

	if (zid == GLOBAL_ZONEID) {
		/*
		 * The link is moving from a non-global zone to the global
		 * zone, so we need to release the reference that was held
		 * when the link was originally assigned to the non-global
		 * zone.
		 */
		dls_link_rele(dlp);
	}

done:
	/*
	 * We only keep the reference to this link open if the link has
	 * successfully moved from the global zone to a non-global zone.
	 */
	if (err != 0 || old_zid != GLOBAL_ZONEID)
		dls_link_rele(dlp);
	return (err);
}

int
dls_link_getzid(const char *name, zoneid_t *zidp)
{
	dls_link_t	*dlp;
	int		err = 0;

	if ((err = dls_link_hold(name, &dlp)) != 0)
		return (err);

	ASSERT(MAC_PERIM_HELD(dlp->dl_mh));

	*zidp = dlp->dl_zid;

	dls_link_rele(dlp);
	return (0);
}

void
dls_link_add(dls_link_t *dlp, uint32_t sap, dld_str_t *dsp)
{
	mod_hash_t	*hash = dlp->dl_str_hash;
	mod_hash_key_t	key;
	dls_head_t	*dhp;
	dld_str_t	*p;
	int		err;

	ASSERT(MAC_PERIM_HELD(dlp->dl_mh));

	/*
	 * Generate a hash key based on the sap.
	 */
	key = MAKE_KEY(sap);

	/*
	 * Search the table for a list head with this key.
	 */
	if ((err = mod_hash_find(hash, key, (mod_hash_val_t *)&dhp)) != 0) {
		ASSERT(err == MH_ERR_NOTFOUND);

		dhp = i_dls_head_alloc(key);
		err = mod_hash_insert(hash, key, (mod_hash_val_t)dhp);
		ASSERT(err == 0);
	}

	/*
	 * Add the dld_str_t to the head of the list. List walkers in
	 * i_dls_link_rx_* bump up dh_ref to ensure the list does not change
	 * while they walk the list. The membar below ensures that list walkers
	 * see exactly the old list or the new list.
	 */
	ASSERT(dsp->ds_next == NULL);
	p = dhp->dh_list;
	dsp->ds_next = p;

	membar_producer();

	dhp->dh_list = dsp;

	/*
	 * Save a pointer to the list head.
	 */
	dsp->ds_head = dhp;
	dlp->dl_impl_count++;
}

void
dls_link_remove(dls_link_t *dlp, dld_str_t *dsp)
{
	mod_hash_t	*hash = dlp->dl_str_hash;
	dld_str_t	**pp;
	dld_str_t	*p;
	dls_head_t	*dhp;

	ASSERT(MAC_PERIM_HELD(dlp->dl_mh));

	/*
	 * We set dh_removing here to tell the receive callbacks not to pass
	 * up packets anymore. Then wait till the current callbacks are done.
	 * This happens either in the close path or in processing the
	 * DL_UNBIND_REQ via a taskq thread, and it is ok to cv_wait in either.
	 * The dh_ref ensures there aren't and there won't be any upcalls
	 * walking or using the dh_list. The mod hash internal lock ensures
	 * that the insert/remove of the dls_head_t itself synchronizes with
	 * any i_dls_link_rx trying to locate it. The perimeter ensures that
	 * there isn't another simultaneous dls_link_add/remove.
	 */
	dhp = dsp->ds_head;

	mutex_enter(&dhp->dh_lock);
	dhp->dh_removing = B_TRUE;
	while (dhp->dh_ref != 0)
		cv_wait(&dhp->dh_cv, &dhp->dh_lock);
	mutex_exit(&dhp->dh_lock);

	/*
	 * Walk the list and remove the dld_str_t.
	 */
	for (pp = &dhp->dh_list; (p = *pp) != NULL; pp = &(p->ds_next)) {
		if (p == dsp)
			break;
	}
	ASSERT(p != NULL);
	*pp = p->ds_next;
	p->ds_next = NULL;
	p->ds_head = NULL;

	ASSERT(dlp->dl_impl_count != 0);
	dlp->dl_impl_count--;

	if (dhp->dh_list == NULL) {
		mod_hash_val_t	val = NULL;

		/*
		 * The list is empty so remove the hash table entry.
		 */
		(void) mod_hash_remove(hash, dhp->dh_key, &val);
		ASSERT(dhp == (dls_head_t *)val);
		i_dls_head_free(dhp);
	} else {
		mutex_enter(&dhp->dh_lock);
		dhp->dh_removing = B_FALSE;
		mutex_exit(&dhp->dh_lock);
	}
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 * Copyright (c) 2017 Joyent, Inc.
 * Copyright 2025 Oxide Computer Company.
 */
/*
 * Copyright (c) 2016 by Delphix. All rights reserved.
 */

/*
 * Datalink management routines.
 */

#include <sys/types.h>
#include <sys/door.h>
#include <sys/zone.h>
#include <sys/modctl.h>
#include <sys/file.h>
#include <sys/modhash.h>
#include <sys/kstat.h>
#include <sys/vnode.h>
#include <sys/cmn_err.h>
#include <sys/softmac.h>
#include <sys/dls.h>
#include <sys/dls_impl.h>
#include <sys/stropts.h>
#include <sys/netstack.h>
#include <inet/iptun/iptun_impl.h>

/*
 * This vanity name management module is treated as part of the GLD framework
 * and we don't hold any GLD framework lock across a call to any mac
 * function that needs to acquire the mac perimeter. The hierarchy is
 * mac perimeter -> framework locks
 */

typedef struct dls_stack {
	zoneid_t	dlss_zoneid;
} dls_stack_t;

static kmem_cache_t	*i_dls_devnet_cachep;

/* Upcall door handle and its lock. */
static kmutex_t		i_dls_mgmt_lock;
static door_handle_t	dls_mgmt_dh = NULL;

/*
 * Any association of <macname, linkid> (set, rename) can require an upcall to
 * the daemon for the link vanity name. We want set/rename/unset to be mutually
 * exclusive from start to finish, but it is unsafe to hold a write on
 * i_dls_devnet_hash_lock during an upcall. Enforce their exclusion using a
 * separate lock from the hash tables.
 *
 * i_dls_devnet_hash_lock protects the hash tables themselves. Taking a write on
 * it requires we first hold i_dls_devnet_lock. Thus, we can safely drop,
 * reacquire, and upgrade/downgrade it so long as all table updates occur in a
 * single write. If a write is intended i_dls_devnet_lock must be acquired
 * before i_dls_devnet_hash_lock, leaving the valid lock patterns:
 *  - i_dls_devnet_lock_enter -> i_dls_devnet_hashmap_write
 *  - i_dls_devnet_lock_enter -> i_dls_devnet_hashmap_read
 *  - i_dls_devnet_hashmap_read
 * i_dls_devnet_hashmap_write enforces the first invariant.
 */
static kmutex_t		i_dls_devnet_lock;
static kcondvar_t	i_dls_devnet_cv;
static kthread_t	*i_dls_devnet_own;

static krwlock_t	i_dls_devnet_hash_lock;
static mod_hash_t	*i_dls_devnet_id_hash;
static mod_hash_t	*i_dls_devnet_hash;

static void
i_dls_devnet_lock_enter(void)
{
	mutex_enter(&i_dls_devnet_lock);
	while (i_dls_devnet_own != NULL) {
		cv_wait(&i_dls_devnet_cv, &i_dls_devnet_lock);
	}
}

static void
i_dls_devnet_lock_exit(void)
{
	VERIFY3P(i_dls_devnet_own, ==, NULL);
	cv_broadcast(&i_dls_devnet_cv);
	mutex_exit(&i_dls_devnet_lock);
}

static void
i_dls_devnet_lock_upcall_start(void)
{
	VERIFY(MUTEX_HELD(&i_dls_devnet_lock));
	VERIFY3P(i_dls_devnet_own, ==, NULL);
	i_dls_devnet_own = curthread;
	mutex_exit(&i_dls_devnet_lock);
}

static void
i_dls_devnet_lock_upcall_end(void)
{
	mutex_enter(&i_dls_devnet_lock);
	VERIFY3P(i_dls_devnet_own, ==, curthread);
	i_dls_devnet_own = NULL;
}

static void
i_dls_devnet_hashmap_write(void)
{
	VERIFY(MUTEX_HELD(&i_dls_devnet_lock));
	rw_enter(&i_dls_devnet_hash_lock, RW_WRITER);
}

static void
i_dls_devnet_hashmap_read(void)
{
	rw_enter(&i_dls_devnet_hash_lock, RW_READER);
}

static void
i_dls_devnet_hashmap_exit(void)
{
	rw_exit(&i_dls_devnet_hash_lock);
}

boolean_t		devnet_need_rebuild;

#define	VLAN_HASHSZ	67	/* prime */

/*
 * The following macros take a link name without the trailing PPA as input.
 * Opening a /dev/net node with one of these names causes a tunnel link to be
 * implicitly created in dls_devnet_hold_by_name() for backward compatibility
 * with Solaris 10 and prior.
 */
#define	IS_IPV4_TUN(name)	(strcmp((name), "ip.tun") == 0)
#define	IS_IPV6_TUN(name)	(strcmp((name), "ip6.tun") == 0)
#define	IS_6TO4_TUN(name)	(strcmp((name), "ip.6to4tun") == 0)
#define	IS_IPTUN_LINK(name)	(					\
    IS_IPV4_TUN(name) || IS_IPV6_TUN(name) || IS_6TO4_TUN(name))

/* dls_devnet_t dd_flags */
#define	DD_CONDEMNED		0x1
#define	DD_IMPLICIT_IPTUN	0x2 /* Implicitly-created ip*.*tun* tunnel */
#define	DD_INITIALIZING		0x4

/*
 * If the link is marked as initializing or condemned then it should
 * not be visible outside of the DLS framework.
 */
#define	DD_NOT_VISIBLE(flags)	(					\
	(flags & (DD_CONDEMNED | DD_INITIALIZING)) != 0)

/*
 * This structure is used to keep the <linkid, macname> mapping.
 * This structure itself is not protected by the mac perimeter, but is
 * protected by the dd_mutex and i_dls_devnet_hash_lock. Thus most of the
 * functions manipulating this structure such as dls_devnet_set/unset etc.
 * may be called while not holding the mac perimeter.
 */
typedef struct dls_devnet_s {
	datalink_id_t	dd_linkid;
	char		dd_linkname[MAXLINKNAMELEN];
	char		dd_mac[MAXNAMELEN];
	kstat_t		*dd_ksp;	/* kstat in owner_zid */
	kstat_t		*dd_zone_ksp;	/* in dd_zid if != owner_zid */
	uint32_t	dd_ref;
	kmutex_t	dd_mutex;
	kcondvar_t	dd_cv;
	uint32_t	dd_tref;
	uint_t		dd_flags;
	zoneid_t	dd_owner_zid;	/* zone where node was created */
	zoneid_t	dd_zid;		/* current zone */
	boolean_t	dd_prop_loaded;
	taskqid_t	dd_prop_taskid;
	boolean_t	dd_transient;	/* link goes away when zone does */
} dls_devnet_t;

static int i_dls_devnet_create_iptun(const char *, const char *,
    datalink_id_t *);
static int i_dls_devnet_destroy_iptun(datalink_id_t);
static int i_dls_devnet_setzid(dls_devnet_t *, zoneid_t, boolean_t, boolean_t);
static int dls_devnet_unset(mac_handle_t, datalink_id_t *, boolean_t);

/*ARGSUSED*/
static int
i_dls_devnet_constructor(void *buf, void *arg, int kmflag)
{
	dls_devnet_t	*ddp = buf;

	bzero(buf, sizeof (dls_devnet_t));
	mutex_init(&ddp->dd_mutex, NULL, MUTEX_DEFAULT, NULL);
	cv_init(&ddp->dd_cv, NULL, CV_DEFAULT, NULL);
	return (0);
}

/*ARGSUSED*/
static void
i_dls_devnet_destructor(void *buf, void *arg)
{
	dls_devnet_t	*ddp = buf;

	VERIFY(ddp->dd_ksp == NULL);
	VERIFY(ddp->dd_ref == 0);
	VERIFY(ddp->dd_tref == 0);
	mutex_destroy(&ddp->dd_mutex);
	cv_destroy(&ddp->dd_cv);
}

/* ARGSUSED */
static int
dls_zone_remove(datalink_id_t linkid, void *arg)
{
	dls_devnet_t *ddp;

	if (dls_devnet_hold_tmp(linkid, &ddp) == 0) {
		/*
		 * Don't bother moving transient links back to the global zone
		 * since we will simply delete them in dls_devnet_unset.
		 */
		if (!ddp->dd_transient)
			(void) dls_devnet_setzid(ddp, GLOBAL_ZONEID);
		dls_devnet_rele_tmp(ddp);
	}
	return (0);
}

/* ARGSUSED */
static void *
dls_stack_init(netstackid_t stackid, netstack_t *ns)
{
	dls_stack_t *dlss;

	dlss = kmem_zalloc(sizeof (*dlss), KM_SLEEP);
	dlss->dlss_zoneid = netstackid_to_zoneid(stackid);
	return (dlss);
}

/* ARGSUSED */
static void
dls_stack_shutdown(netstackid_t stackid, void *arg)
{
	dls_stack_t	*dlss = (dls_stack_t *)arg;

	/* Move remaining datalinks in this zone back to the global zone. */
	(void) zone_datalink_walk(dlss->dlss_zoneid, dls_zone_remove, NULL);
}

/* ARGSUSED */
static void
dls_stack_fini(netstackid_t stackid, void *arg)
{
	dls_stack_t	*dlss = (dls_stack_t *)arg;

	kmem_free(dlss, sizeof (*dlss));
}

/*
 * Module initialization and finalization functions.
 */
void
dls_mgmt_init(void)
{
	mutex_init(&i_dls_mgmt_lock, NULL, MUTEX_DEFAULT, NULL);
	mutex_init(&i_dls_devnet_lock, NULL, MUTEX_DEFAULT, NULL);
	cv_init(&i_dls_devnet_cv, NULL, CV_DEFAULT, NULL);
	i_dls_devnet_own = NULL;
	rw_init(&i_dls_devnet_hash_lock, NULL, RW_DEFAULT, NULL);

	/*
	 * Create a kmem_cache of dls_devnet_t structures.
	 */
	i_dls_devnet_cachep = kmem_cache_create("dls_devnet_cache",
	    sizeof (dls_devnet_t), 0, i_dls_devnet_constructor,
	    i_dls_devnet_destructor, NULL, NULL, NULL, 0);
	ASSERT(i_dls_devnet_cachep != NULL);

	/*
	 * Create a hash table, keyed by dd_linkid, of dls_devnet_t.
	 */
	i_dls_devnet_id_hash = mod_hash_create_idhash("dls_devnet_id_hash",
	    VLAN_HASHSZ, mod_hash_null_valdtor);

	/*
	 * Create a hash table, keyed by dd_mac
	 */
	i_dls_devnet_hash = mod_hash_create_extended("dls_devnet_hash",
	    VLAN_HASHSZ, mod_hash_null_keydtor, mod_hash_null_valdtor,
	    mod_hash_bystr, NULL, mod_hash_strkey_cmp, KM_SLEEP);

	devnet_need_rebuild = B_FALSE;

	netstack_register(NS_DLS, dls_stack_init, dls_stack_shutdown,
	    dls_stack_fini);
}

void
dls_mgmt_fini(void)
{
	netstack_unregister(NS_DLS);
	mod_hash_destroy_hash(i_dls_devnet_hash);
	mod_hash_destroy_hash(i_dls_devnet_id_hash);
	kmem_cache_destroy(i_dls_devnet_cachep);
	rw_destroy(&i_dls_devnet_hash_lock);
	cv_destroy(&i_dls_devnet_cv);
	mutex_destroy(&i_dls_devnet_lock);
	mutex_destroy(&i_dls_mgmt_lock);
}

int
dls_mgmt_door_set(boolean_t start)
{
	int	err;

	/* handle daemon restart */
	mutex_enter(&i_dls_mgmt_lock);
	if (dls_mgmt_dh != NULL) {
		door_ki_rele(dls_mgmt_dh);
		dls_mgmt_dh = NULL;
	}

	if (start && ((err = door_ki_open(DLMGMT_DOOR, &dls_mgmt_dh)) != 0)) {
		mutex_exit(&i_dls_mgmt_lock);
		return (err);
	}

	mutex_exit(&i_dls_mgmt_lock);

	/*
	 * Create and associate <link name, linkid> mapping for network devices
	 * which are already attached before the daemon is started.
	 */
	if (start)
		softmac_recreate();
	return (0);
}

static boolean_t
i_dls_mgmt_door_revoked(door_handle_t dh)
{
	struct door_info info;
	extern int sys_shutdown;

	ASSERT(dh != NULL);

	if (sys_shutdown) {
		cmn_err(CE_NOTE, "dls_mgmt_door: shutdown observed\n");
		return (B_TRUE);
	}

	if (door_ki_info(dh, &info) != 0)
		return (B_TRUE);

	return ((info.di_attributes & DOOR_REVOKED) != 0);
}

/*
 * Upcall to the datalink management daemon (dlmgmtd).
 */
static int
i_dls_mgmt_upcall(void *arg, size_t asize, void *rbuf, size_t rsize)
{
	door_arg_t			darg, save_arg;
	door_handle_t			dh;
	int				err;
	int				retry = 0;

#define	MAXRETRYNUM	3

	ASSERT(arg);
	darg.data_ptr = arg;
	darg.data_size = asize;
	darg.desc_ptr = NULL;
	darg.desc_num = 0;
	darg.rbuf = rbuf;
	darg.rsize = rsize;
	save_arg = darg;

retry:
	mutex_enter(&i_dls_mgmt_lock);
	dh = dls_mgmt_dh;
	if ((dh == NULL) || i_dls_mgmt_door_revoked(dh)) {
		mutex_exit(&i_dls_mgmt_lock);
		return (EBADF);
	}
	door_ki_hold(dh);
	mutex_exit(&i_dls_mgmt_lock);

	for (;;) {
		retry++;
		if ((err = door_ki_upcall_limited(dh, &darg, zone_kcred(),
		    SIZE_MAX, 0)) == 0)
			break;

		/*
		 * handle door call errors
		 */
		darg = save_arg;
		switch (err) {
		case EINTR:
			/*
			 * If the operation which caused this door upcall gets
			 * interrupted, return directly.
			 */
			goto done;
		case EAGAIN:
			/*
			 * Repeat upcall if the maximum attempt limit has not
			 * been reached.
			 */
			if (retry < MAXRETRYNUM) {
				delay(2 * hz);
				break;
			}
			cmn_err(CE_WARN, "dls: dlmgmtd fatal error %d\n", err);
			goto done;
		default:
			/* A fatal door error */
			if (i_dls_mgmt_door_revoked(dh)) {
				cmn_err(CE_NOTE,
				    "dls: dlmgmtd door service revoked\n");

				if (retry < MAXRETRYNUM) {
					door_ki_rele(dh);
					goto retry;
				}
			}
			cmn_err(CE_WARN, "dls: dlmgmtd fatal error %d\n", err);
			goto done;
		}
	}

	if (darg.rbuf != rbuf) {
		/*
		 * The size of the input rbuf was not big enough, so the
		 * upcall allocated the rbuf itself.  If this happens, assume
		 * that this was an invalid door call request.
		 */
		kmem_free(darg.rbuf, darg.rsize);
		err = ENOSPC;
		goto done;
	}

	if (darg.rsize != rsize) {
		err = EINVAL;
		goto done;
	}

	err = ((dlmgmt_retval_t *)rbuf)->lr_err;

done:
	door_ki_rele(dh);
	return (err);
}

/*
 * Request the datalink management daemon to create a link with the attributes
 * below.  Upon success, zero is returned and linkidp contains the linkid for
 * the new link; otherwise, an errno is returned.
 *
 *     - dev		physical dev_t.  required for all physical links,
 *		        including GLDv3 links.  It will be used to force the
 *		        attachment of a physical device, hence the
 *		        registration of its mac
 *     - class		datalink class
 *     - media type	media type; DL_OTHER means unknown
 *     - persist	whether to persist the datalink
 */
int
dls_mgmt_create(const char *devname, dev_t dev, datalink_class_t class,
    uint32_t media, boolean_t persist, datalink_id_t *linkidp)
{
	dlmgmt_upcall_arg_create_t	create;
	dlmgmt_create_retval_t		retval;
	int				err;

	create.ld_cmd = DLMGMT_CMD_DLS_CREATE;
	create.ld_class = class;
	create.ld_media = media;
	create.ld_phymaj = getmajor(dev);
	create.ld_phyinst = getminor(dev);
	create.ld_persist = persist;
	if (strlcpy(create.ld_devname, devname, sizeof (create.ld_devname)) >=
	    sizeof (create.ld_devname))
		return (EINVAL);

	if ((err = i_dls_mgmt_upcall(&create, sizeof (create), &retval,
	    sizeof (retval))) == 0) {
		*linkidp = retval.lr_linkid;
	}
	return (err);
}

/*
 * Request the datalink management daemon to destroy the specified link.
 * Returns zero upon success, or an errno upon failure.
 */
int
dls_mgmt_destroy(datalink_id_t linkid, boolean_t persist)
{
	dlmgmt_upcall_arg_destroy_t	destroy;
	dlmgmt_destroy_retval_t		retval;

	destroy.ld_cmd = DLMGMT_CMD_DLS_DESTROY;
	destroy.ld_linkid = linkid;
	destroy.ld_persist = persist;

	return (i_dls_mgmt_upcall(&destroy, sizeof (destroy),
	    &retval, sizeof (retval)));
}

/*
 * Request the datalink management daemon to verify/update the information
 * for a physical link.  Upon success, get its linkid.
 *
 *     - media type	media type
 *     - novanity	whether this physical datalink supports vanity naming.
 *			physical links that do not use the GLDv3 MAC plugin
 *			cannot suport vanity naming
 *
 * This function could fail with ENOENT or EEXIST.  Two cases return EEXIST:
 *
 * 1. A link with devname already exists, but the media type does not match.
 *    In this case, mediap will bee set to the media type of the existing link.
 * 2. A link with devname already exists, but its link name does not match
 *    the device name, although this link does not support vanity naming.
 */
int
dls_mgmt_update(const char *devname, uint32_t media, boolean_t novanity,
    uint32_t *mediap, datalink_id_t *linkidp)
{
	dlmgmt_upcall_arg_update_t	update;
	dlmgmt_update_retval_t		retval;
	int				err;

	update.ld_cmd = DLMGMT_CMD_DLS_UPDATE;

	if (strlcpy(update.ld_devname, devname, sizeof (update.ld_devname)) >=
	    sizeof (update.ld_devname))
		return (EINVAL);

	update.ld_media = media;
	update.ld_novanity = novanity;

	if ((err = i_dls_mgmt_upcall(&update, sizeof (update), &retval,
	    sizeof (retval))) == EEXIST) {
		*linkidp = retval.lr_linkid;
		*mediap = retval.lr_media;
	} else if (err == 0) {
		*linkidp = retval.lr_linkid;
	}

	return (err);
}

/*
 * Request the datalink management daemon to get the information for a link.
 * Returns zero upon success, or an errno upon failure.
 *
 * Only fills in information for argument pointers that are non-NULL.
 * Note that the link argument is expected to be MAXLINKNAMELEN bytes.
 */
int
dls_mgmt_get_linkinfo(datalink_id_t linkid, char *link,
    datalink_class_t *classp, uint32_t *mediap, uint32_t *flagsp)
{
	dlmgmt_door_getname_t	getname;
	dlmgmt_getname_retval_t	retval;
	int			err, len;

	getname.ld_cmd = DLMGMT_CMD_GETNAME;
	getname.ld_linkid = linkid;

	if ((err = i_dls_mgmt_upcall(&getname, sizeof (getname), &retval,
	    sizeof (retval))) != 0) {
		return (err);
	}

	len = strlen(retval.lr_link);
	if (len <= 1 || len >= MAXLINKNAMELEN)
		return (EINVAL);

	if (link != NULL)
		(void) strlcpy(link, retval.lr_link, MAXLINKNAMELEN);
	if (classp != NULL)
		*classp = retval.lr_class;
	if (mediap != NULL)
		*mediap = retval.lr_media;
	if (flagsp != NULL)
		*flagsp = retval.lr_flags;
	return (0);
}

/*
 * Request the datalink management daemon to get the linkid for a link.
 * Returns a non-zero error code on failure.  The linkid argument is only
 * set on success (when zero is returned.)
 */
int
dls_mgmt_get_linkid(const char *link, datalink_id_t *linkid)
{
	dlmgmt_door_getlinkid_t		getlinkid;
	dlmgmt_getlinkid_retval_t	retval;
	int				err;

	getlinkid.ld_cmd = DLMGMT_CMD_GETLINKID;
	(void) strlcpy(getlinkid.ld_link, link, MAXLINKNAMELEN);

	if ((err = i_dls_mgmt_upcall(&getlinkid, sizeof (getlinkid), &retval,
	    sizeof (retval))) == 0) {
		*linkid = retval.lr_linkid;
	}
	return (err);
}

datalink_id_t
dls_mgmt_get_next(datalink_id_t linkid, datalink_class_t class,
    datalink_media_t dmedia, uint32_t flags)
{
	dlmgmt_door_getnext_t	getnext;
	dlmgmt_getnext_retval_t	retval;

	getnext.ld_cmd = DLMGMT_CMD_GETNEXT;
	getnext.ld_class = class;
	getnext.ld_dmedia = dmedia;
	getnext.ld_flags = flags;
	getnext.ld_linkid = linkid;

	if (i_dls_mgmt_upcall(&getnext, sizeof (getnext), &retval,
	    sizeof (retval)) != 0) {
		return (DATALINK_INVALID_LINKID);
	}

	return (retval.lr_linkid);
}

static int
i_dls_mgmt_get_linkattr(const datalink_id_t linkid, const char *attr,
    void *attrval, size_t *attrszp)
{
	dlmgmt_upcall_arg_getattr_t	getattr;
	dlmgmt_getattr_retval_t		retval;
	int				err;

	getattr.ld_cmd = DLMGMT_CMD_DLS_GETATTR;
	getattr.ld_linkid = linkid;
	(void) strlcpy(getattr.ld_attr, attr, MAXLINKATTRLEN);

	if ((err = i_dls_mgmt_upcall(&getattr, sizeof (getattr), &retval,
	    sizeof (retval))) == 0) {
		if (*attrszp < retval.lr_attrsz)
			return (EINVAL);
		*attrszp = retval.lr_attrsz;
		bcopy(retval.lr_attrval, attrval, retval.lr_attrsz);
	}

	return (err);
}

/*
 * Note that this function can only get devp successfully for non-VLAN link.
 */
int
dls_mgmt_get_phydev(datalink_id_t linkid, dev_t *devp)
{
	uint64_t	maj, inst;
	size_t		attrsz = sizeof (uint64_t);

	if (i_dls_mgmt_get_linkattr(linkid, FPHYMAJ, &maj, &attrsz) != 0 ||
	    attrsz != sizeof (uint64_t) ||
	    i_dls_mgmt_get_linkattr(linkid, FPHYINST, &inst, &attrsz) != 0 ||
	    attrsz != sizeof (uint64_t)) {
		return (EINVAL);
	}

	*devp = makedevice((major_t)maj, (minor_t)inst);
	return (0);
}

/*
 * Request the datalink management daemon to push in
 * all properties associated with the link.
 * Returns a non-zero error code on failure.
 */
int
dls_mgmt_linkprop_init(datalink_id_t linkid)
{
	dlmgmt_door_linkprop_init_t	li;
	dlmgmt_linkprop_init_retval_t	retval;
	int				err;

	li.ld_cmd = DLMGMT_CMD_LINKPROP_INIT;
	li.ld_linkid = linkid;

	err = i_dls_mgmt_upcall(&li, sizeof (li), &retval, sizeof (retval));
	return (err);
}

static void
dls_devnet_prop_task(void *arg)
{
	dls_devnet_t		*ddp = arg;

	(void) dls_mgmt_linkprop_init(ddp->dd_linkid);

	mutex_enter(&ddp->dd_mutex);
	ddp->dd_prop_loaded = B_TRUE;
	ddp->dd_prop_taskid = 0;
	cv_broadcast(&ddp->dd_cv);
	mutex_exit(&ddp->dd_mutex);
}

/*
 * Ensure property loading task is completed.
 */
void
dls_devnet_prop_task_wait(dls_dl_handle_t ddp)
{
	mutex_enter(&ddp->dd_mutex);
	while (ddp->dd_prop_taskid != 0)
		cv_wait(&ddp->dd_cv, &ddp->dd_mutex);
	mutex_exit(&ddp->dd_mutex);
}

void
dls_devnet_rele_tmp(dls_dl_handle_t dlh)
{
	dls_devnet_t		*ddp = dlh;

	mutex_enter(&ddp->dd_mutex);
	ASSERT(ddp->dd_tref != 0);
	if (--ddp->dd_tref == 0)
		cv_signal(&ddp->dd_cv);
	mutex_exit(&ddp->dd_mutex);
}

int
dls_devnet_hold_link(datalink_id_t linkid, dls_dl_handle_t *ddhp,
    dls_link_t **dlpp)
{
	dls_dl_handle_t	dlh;
	dls_link_t	*dlp;
	int		err;

	if ((err = dls_devnet_hold_tmp(linkid, &dlh)) != 0)
		return (err);

	if ((err = dls_link_hold(dls_devnet_mac(dlh), &dlp)) != 0) {
		dls_devnet_rele_tmp(dlh);
		return (err);
	}

	ASSERT(MAC_PERIM_HELD(dlp->dl_mh));

	*ddhp = dlh;
	*dlpp = dlp;
	return (0);
}

void
dls_devnet_rele_link(dls_dl_handle_t dlh, dls_link_t *dlp)
{
	ASSERT(MAC_PERIM_HELD(dlp->dl_mh));

	dls_link_rele(dlp);
	dls_devnet_rele_tmp(dlh);
}

/*
 * "link" kstats related functions.
 */

/*
 * Query the "link" kstats.
 *
 * We may be called from the kstat subsystem in an arbitrary context.
 * If the caller is the stack, the context could be an upcall data
 * thread. Hence we can't acquire the mac perimeter in this function
 * for fear of deadlock.
 */
static int
dls_devnet_stat_update(kstat_t *ksp, int rw)
{
	datalink_id_t	linkid = (datalink_id_t)(uintptr_t)ksp->ks_private;
	dls_devnet_t	*ddp;
	dls_link_t	*dlp;
	int		err;

	if ((err = dls_devnet_hold_tmp(linkid, &ddp)) != 0) {
		return (err);
	}

	/*
	 * If a device detach happens at this time, it will block in
	 * dls_devnet_unset since the dd_tref has been bumped in
	 * dls_devnet_hold_tmp(). So the access to 'dlp' is safe even though
	 * we don't hold the mac perimeter.
	 */
	if (mod_hash_find(i_dls_link_hash, (mod_hash_key_t)ddp->dd_mac,
	    (mod_hash_val_t *)&dlp) != 0) {
		dls_devnet_rele_tmp(ddp);
		return (ENOENT);
	}

	err = dls_stat_update(ksp, dlp, rw);

	dls_devnet_rele_tmp(ddp);
	return (err);
}

/*
 * Create the "link" kstats.
 */
static void
dls_devnet_stat_create(dls_devnet_t *ddp, zoneid_t zoneid)
{
	kstat_t	*ksp;

	if (dls_stat_create("link", 0, ddp->dd_linkname, zoneid,
	    dls_devnet_stat_update, (void *)(uintptr_t)ddp->dd_linkid,
	    &ksp) == 0) {
		ASSERT(ksp != NULL);
		if (zoneid == ddp->dd_owner_zid) {
			ASSERT(ddp->dd_ksp == NULL);
			ddp->dd_ksp = ksp;
		} else {
			ASSERT(ddp->dd_zone_ksp == NULL);
			ddp->dd_zone_ksp = ksp;
		}
	}
}

/*
 * Destroy the "link" kstats.
 */
static void
dls_devnet_stat_destroy(dls_devnet_t *ddp, zoneid_t zoneid)
{
	if (zoneid == ddp->dd_owner_zid) {
		if (ddp->dd_ksp != NULL) {
			kstat_delete(ddp->dd_ksp);
			ddp->dd_ksp = NULL;
		}
	} else {
		if (ddp->dd_zone_ksp != NULL) {
			kstat_delete(ddp->dd_zone_ksp);
			ddp->dd_zone_ksp = NULL;
		}
	}
}

/*
 * The link has been renamed. Destroy the old non-legacy kstats ("link kstats")
 * and create the new set using the new name.
 */
static void
dls_devnet_stat_rename(dls_devnet_t *ddp)
{
	if (ddp->dd_ksp != NULL) {
		kstat_delete(ddp->dd_ksp);
		ddp->dd_ksp = NULL;
	}
	/* We can't rename a link while it's assigned to a non-global zone. */
	ASSERT(ddp->dd_zone_ksp == NULL);
	dls_devnet_stat_create(ddp, ddp->dd_owner_zid);
}

/*
 * Associate the linkid with the link identified by macname. If this
 * is called on behalf of a physical link then linkid may be
 * DATALINK_INVALID_LINKID. Otherwise, if called on behalf of a
 * virtual link, linkid must have a value.
 */
static int
dls_devnet_set(mac_handle_t mh, datalink_id_t linkid, zoneid_t zoneid,
    dls_devnet_t **ddpp)
{
	const char		*macname = mac_name(mh);
	dls_devnet_t		*ddp = NULL;
	datalink_class_t	class;
	int			err;
	boolean_t		stat_create = B_FALSE;
	char			linkname[MAXLINKNAMELEN];

	i_dls_devnet_lock_enter();

	/*
	 * Don't allow callers to set a link name with a linkid that already
	 * has a name association (that's what rename is for).
	 */
	if (linkid != DATALINK_INVALID_LINKID) {
		/*
		 * This temporary read access is valid, as no other set/rename
		 * operation can attempt an insert on the same linkid while
		 * i_dls_devnet_lock is held.
		 */
		i_dls_devnet_hashmap_read();
		if (mod_hash_find(i_dls_devnet_id_hash,
		    (mod_hash_key_t)(uintptr_t)linkid,
		    (mod_hash_val_t *)&ddp) == 0) {
			err = EEXIST;
			goto done;
		}
		i_dls_devnet_hashmap_exit();

		i_dls_devnet_lock_upcall_start();
		err = dls_mgmt_get_linkinfo(linkid, linkname, &class,
		    NULL, NULL);
		i_dls_devnet_lock_upcall_end();

		if (err != 0)
			goto done_rw_unlocked;
	}

	i_dls_devnet_hashmap_write();

	if ((err = mod_hash_find(i_dls_devnet_hash,
	    (mod_hash_key_t)macname, (mod_hash_val_t *)&ddp)) == 0) {
		if (ddp->dd_linkid != DATALINK_INVALID_LINKID) {
			err = EEXIST;
			goto done;
		}

		/*
		 * If we arrive here we know we are attempting to set
		 * the linkid on a physical link. A virtual link
		 * should never arrive here because it should never
		 * call this function without a linkid. Virtual links
		 * are created through dlgmtmd and thus we know
		 * dlmgmtd is alive to assign it a linkid (search for
		 * uses of dladm_create_datalink_id() to prove this to
		 * yourself); we don't have the same guarantee for a
		 * physical link which may perform an upcall for a
		 * linkid while dlmgmtd is down but will continue
		 * creating a devnet without the linkid (see
		 * softmac_create_datalink() to see how physical link
		 * creation works). That is why there is no entry in
		 * the id hash but there is one in the macname hash --
		 * softmac couldn't acquire a linkid the first time it
		 * called this function.
		 *
		 * Because of the check above, we also know that
		 * ddp->dd_linkid is not set. Following this, the link
		 * must still be in the DD_INITIALIZING state because
		 * that flag is removed IFF dd_linkid is set. This is
		 * why we can ASSERT the DD_INITIALIZING flag below if
		 * the call to i_dls_devnet_setzid() fails.
		 */
		if (linkid == DATALINK_INVALID_LINKID ||
		    class != DATALINK_CLASS_PHYS) {
			err = EINVAL;
			goto done;
		}

		ASSERT(ddp->dd_flags & DD_INITIALIZING);
	} else {
		ddp = kmem_cache_alloc(i_dls_devnet_cachep, KM_SLEEP);
		ddp->dd_flags = DD_INITIALIZING;
		ddp->dd_tref = 0;
		ddp->dd_ref++;
		ddp->dd_owner_zid = zoneid;
		/*
		 * If we are creating a new devnet which will be owned by a NGZ
		 * then mark it as transient. This link has never been in the
		 * GZ, the GZ will not have a hold on its reference, and we do
		 * not want to return it to the GZ when the zone halts.
		 */
		if (zoneid != GLOBAL_ZONEID)
			ddp->dd_transient = B_TRUE;
		(void) strlcpy(ddp->dd_mac, macname, sizeof (ddp->dd_mac));
		VERIFY(mod_hash_insert(i_dls_devnet_hash,
		    (mod_hash_key_t)ddp->dd_mac, (mod_hash_val_t)ddp) == 0);
	}

	if (linkid != DATALINK_INVALID_LINKID) {
		ddp->dd_linkid = linkid;
		(void) strlcpy(ddp->dd_linkname, linkname,
		    sizeof (ddp->dd_linkname));
		VERIFY(mod_hash_insert(i_dls_devnet_id_hash,
		    (mod_hash_key_t)(uintptr_t)linkid,
		    (mod_hash_val_t)ddp) == 0);
		devnet_need_rebuild = B_TRUE;
		stat_create = B_TRUE;
	}
	err = 0;
done:
	/*
	 * It is safe to drop the i_dls_devnet_hash_lock at this point. In the
	 * case of physical devices, the softmac framework will fail the device
	 * detach based on the smac_state or smac_hold_cnt. Other cases like
	 * vnic and aggr use their own scheme to serialize creates and deletes
	 * and ensure that *ddp is valid.
	 */
	i_dls_devnet_hashmap_exit();
done_rw_unlocked:
	i_dls_devnet_lock_exit();

	if (err == 0 && zoneid != GLOBAL_ZONEID) {
		/*
		 * If this link is being created directly within a non-global
		 * zone, then flag it as transient so that it will be cleaned
		 * up when the zone is shut down.
		 */
		err = i_dls_devnet_setzid(ddp, zoneid, B_FALSE, B_TRUE);
		if (err != 0) {
			/*
			 * At this point the link is marked as
			 * DD_INITIALIZING -- there can be no
			 * outstanding temp refs and therefore no need
			 * to wait for them.
			 */
			ASSERT(ddp->dd_flags & DD_INITIALIZING);
			(void) dls_devnet_unset(mh, &linkid, B_FALSE);
			return (err);
		}
	}

	if (err == 0) {
		/*
		 * The kstat subsystem holds its own locks (rather perimeter)
		 * before calling the ks_update (dls_devnet_stat_update) entry
		 * point which in turn grabs the i_dls_devnet_hash_lock. So the
		 * lock hierarchy is kstat locks -> i_dls_devnet_hash_lock.
		 */
		if (stat_create)
			dls_devnet_stat_create(ddp, zoneid);
		if (ddpp != NULL)
			*ddpp = ddp;

		mutex_enter(&ddp->dd_mutex);
		if (linkid != DATALINK_INVALID_LINKID &&
		    !ddp->dd_prop_loaded && ddp->dd_prop_taskid == 0) {
			ddp->dd_prop_taskid = taskq_dispatch(system_taskq,
			    dls_devnet_prop_task, ddp, TQ_SLEEP);
		}
		mutex_exit(&ddp->dd_mutex);

	}
	return (err);
}

/*
 * Disassociate the linkid from the link identified by macname. If
 * wait is B_TRUE, wait until all temporary refs are released and the
 * prop task is finished.
 *
 * If waiting then you SHOULD NOT call this from inside the MAC perim
 * as deadlock will ensue. Otherwise, this function is safe to call
 * from inside or outside the MAC perim.
 */
static int
dls_devnet_unset(mac_handle_t mh, datalink_id_t *id, boolean_t wait)
{
	const char	*macname = mac_name(mh);
	dls_devnet_t	*ddp;
	int		err;
	mod_hash_val_t	val;

	i_dls_devnet_lock_enter();
	i_dls_devnet_hashmap_write();

	if ((err = mod_hash_find(i_dls_devnet_hash,
	    (mod_hash_key_t)macname, (mod_hash_val_t *)&ddp)) != 0) {
		ASSERT(err == MH_ERR_NOTFOUND);
		i_dls_devnet_hashmap_exit();
		return (ENOENT);
	}

	mutex_enter(&ddp->dd_mutex);

	/*
	 * Make sure downcalls into softmac_create or softmac_destroy from
	 * devfs don't cv_wait on any devfs related condition for fear of
	 * deadlock. Return EBUSY if the asynchronous thread started for
	 * property loading as part of the post attach hasn't yet completed.
	 */
	VERIFY(ddp->dd_ref != 0);
	if ((ddp->dd_ref != 1) || (!wait &&
	    (ddp->dd_tref != 0 || ddp->dd_prop_taskid != 0))) {
		int zstatus = 0;

		/*
		 * There are a couple of alternatives that might be going on
		 * here; a) the zone is shutting down and it has a transient
		 * link assigned, in which case we want to clean it up instead
		 * of moving it back to the global zone, or b) its possible
		 * that we're trying to clean up an orphaned vnic that was
		 * delegated to a zone and which wasn't cleaned up properly
		 * when the zone went away.  Check for either of these cases
		 * before we simply return EBUSY.
		 *
		 * zstatus indicates which situation we are dealing with:
		 *	 0 - means return EBUSY
		 *	 1 - means case (a), cleanup transient link
		 *	-1 - means case (b), orphaned VNIC
		 */
		if (ddp->dd_ref > 1 && ddp->dd_zid != GLOBAL_ZONEID) {
			zone_t	*zp;

			if ((zp = zone_find_by_id(ddp->dd_zid)) == NULL) {
				zstatus = -1;
			} else {
				if (ddp->dd_transient) {
					zone_status_t s = zone_status_get(zp);

					if (s >= ZONE_IS_SHUTTING_DOWN)
						zstatus = 1;
				}
				zone_rele(zp);
			}
		}

		if (zstatus == 0) {
			mutex_exit(&ddp->dd_mutex);
			i_dls_devnet_hashmap_exit();
			i_dls_devnet_lock_exit();
			return (EBUSY);
		}

		/*
		 * We want to delete the link, reset ref to 1;
		 */
		if (zstatus == -1) {
			/* Log a warning, but continue in this case */
			cmn_err(CE_WARN, "clear orphaned datalink: %s\n",
			    ddp->dd_linkname);
		}
		ddp->dd_ref = 1;
	}

	ddp->dd_flags |= DD_CONDEMNED;
	ddp->dd_ref--;
	*id = ddp->dd_linkid;

	/*
	 * Remove this dls_devnet_t from the hash table.
	 */
	VERIFY(mod_hash_remove(i_dls_devnet_hash,
	    (mod_hash_key_t)ddp->dd_mac, &val) == 0);

	if (ddp->dd_linkid != DATALINK_INVALID_LINKID) {
		VERIFY(mod_hash_remove(i_dls_devnet_id_hash,
		    (mod_hash_key_t)(uintptr_t)ddp->dd_linkid, &val) == 0);

		devnet_need_rebuild = B_TRUE;
	}

	i_dls_devnet_hashmap_exit();
	i_dls_devnet_lock_exit();

	/*
	 * It is important to call i_dls_devnet_setzid() WITHOUT the
	 * i_dls_devnet_hash_lock held. The setzid call grabs the MAC
	 * perim; thus causing DLS -> MAC lock ordering if performed
	 * with the i_dls_devnet_hash_lock held. This forces consumers to
	 * grab the MAC perim before calling dls_devnet_unset() (the
	 * locking rules state MAC -> DLS order). By performing the
	 * setzid outside of the i_dls_devnet_hash_lock consumers can
	 * safely call dls_devnet_unset() outside the MAC perim.
	 */
	if (ddp->dd_zid != GLOBAL_ZONEID) {
		/*
		 * We need to release the dd_mutex before we try and destroy the
		 * stat. When we destroy it, we'll need to grab the lock for the
		 * kstat but if there's a concurrent reader of the kstat, we'll
		 * be blocked on it. This will lead to deadlock because these
		 * kstats employ a ks_update function (dls_devnet_stat_update)
		 * which needs the dd_mutex that we currently hold.
		 *
		 * Because we've already flagged the dls_devnet_t as
		 * DD_CONDEMNED and we still have a write lock on
		 * i_dls_devnet_hash_lock, we should be able to release the
		 * dd_mutex.
		 */
		mutex_exit(&ddp->dd_mutex);
		dls_devnet_stat_destroy(ddp, ddp->dd_zid);
		mutex_enter(&ddp->dd_mutex);
		(void) i_dls_devnet_setzid(ddp, GLOBAL_ZONEID, B_FALSE,
		    B_FALSE);
	}

	if (wait) {
		/*
		 * Wait until all temporary references are released.
		 * The holders of the tref need the MAC perim to
		 * perform their work and release the tref. To avoid
		 * deadlock, assert that the perim is never held here.
		 */
		ASSERT0(MAC_PERIM_HELD(mh));
		while ((ddp->dd_tref != 0) || (ddp->dd_prop_taskid != 0))
			cv_wait(&ddp->dd_cv, &ddp->dd_mutex);
	} else {
		VERIFY(ddp->dd_tref == 0);
		VERIFY(ddp->dd_prop_taskid == 0);
	}

	if (ddp->dd_linkid != DATALINK_INVALID_LINKID)
		dls_devnet_stat_destroy(ddp, ddp->dd_owner_zid);

	ddp->dd_prop_loaded = B_FALSE;
	ddp->dd_linkid = DATALINK_INVALID_LINKID;
	ddp->dd_flags = 0;
	mutex_exit(&ddp->dd_mutex);
	kmem_cache_free(i_dls_devnet_cachep, ddp);

	return (0);
}

/*
 * This is a private hold routine used when we already have the dls_link_t, thus
 * we know that it cannot go away.
 */
int
dls_devnet_hold_tmp_by_link(dls_link_t *dlp, dls_dl_handle_t *ddhp)
{
	int err;
	dls_devnet_t *ddp = NULL;

	i_dls_devnet_hashmap_read();
	if ((err = mod_hash_find(i_dls_devnet_hash,
	    (mod_hash_key_t)dlp->dl_name, (mod_hash_val_t *)&ddp)) != 0) {
		ASSERT(err == MH_ERR_NOTFOUND);
		i_dls_devnet_hashmap_exit();
		return (ENOENT);
	}

	mutex_enter(&ddp->dd_mutex);
	VERIFY(ddp->dd_ref > 0);
	if (DD_NOT_VISIBLE(ddp->dd_flags)) {
		mutex_exit(&ddp->dd_mutex);
		i_dls_devnet_hashmap_exit();
		return (ENOENT);
	}
	ddp->dd_tref++;
	mutex_exit(&ddp->dd_mutex);
	i_dls_devnet_hashmap_exit();

	*ddhp = ddp;
	return (0);
}

static int
dls_devnet_hold_common(datalink_id_t linkid, dls_devnet_t **ddpp,
    boolean_t tmp_hold)
{
	dls_devnet_t		*ddp;
	int			err;

	i_dls_devnet_hashmap_read();
	if ((err = mod_hash_find(i_dls_devnet_id_hash,
	    (mod_hash_key_t)(uintptr_t)linkid, (mod_hash_val_t *)&ddp)) != 0) {
		ASSERT(err == MH_ERR_NOTFOUND);
		i_dls_devnet_hashmap_exit();
		return (ENOENT);
	}

	mutex_enter(&ddp->dd_mutex);
	VERIFY(ddp->dd_ref > 0);
	if (DD_NOT_VISIBLE(ddp->dd_flags)) {
		mutex_exit(&ddp->dd_mutex);
		i_dls_devnet_hashmap_exit();
		return (ENOENT);
	}
	if (tmp_hold)
		ddp->dd_tref++;
	else
		ddp->dd_ref++;
	mutex_exit(&ddp->dd_mutex);
	i_dls_devnet_hashmap_exit();

	*ddpp = ddp;
	return (0);
}

int
dls_devnet_hold(datalink_id_t linkid, dls_devnet_t **ddpp)
{
	return (dls_devnet_hold_common(linkid, ddpp, B_FALSE));
}

/*
 * Hold the vanity naming structure (dls_devnet_t) temporarily.  The request to
 * delete the dls_devnet_t will wait until the temporary reference is released.
 */
int
dls_devnet_hold_tmp(datalink_id_t linkid, dls_devnet_t **ddpp)
{
	return (dls_devnet_hold_common(linkid, ddpp, B_TRUE));
}

/*
 * This funtion is called when a DLS client tries to open a device node.
 * This dev_t could be a result of a /dev/net node access (returned by
 * devnet_create_rvp->dls_devnet_open()) or a direct /dev node access.
 * In both cases, this function bumps up the reference count of the
 * dls_devnet_t structure. The reference is held as long as the device node
 * is open. In the case of /dev/net while it is true that the initial reference
 * is held when the devnet_create_rvp->dls_devnet_open call happens, this
 * initial reference is released immediately in devnet_inactive_callback ->
 * dls_devnet_close(). (Note that devnet_inactive_callback() is called right
 * after dld_open completes, not when the /dev/net node is being closed).
 * To undo this function, call dls_devnet_rele()
 */
int
dls_devnet_hold_by_dev(dev_t dev, dls_dl_handle_t *ddhp)
{
	char			name[MAXNAMELEN];
	char			*drv;
	dls_devnet_t		*ddp;
	int			err;

	if ((drv = ddi_major_to_name(getmajor(dev))) == NULL)
		return (EINVAL);

	(void) snprintf(name, sizeof (name), "%s%d", drv,
	    DLS_MINOR2INST(getminor(dev)));

	i_dls_devnet_hashmap_read();
	if ((err = mod_hash_find(i_dls_devnet_hash,
	    (mod_hash_key_t)name, (mod_hash_val_t *)&ddp)) != 0) {
		ASSERT(err == MH_ERR_NOTFOUND);
		i_dls_devnet_hashmap_exit();
		return (ENOENT);
	}
	mutex_enter(&ddp->dd_mutex);
	VERIFY(ddp->dd_ref > 0);
	if (DD_NOT_VISIBLE(ddp->dd_flags)) {
		mutex_exit(&ddp->dd_mutex);
		i_dls_devnet_hashmap_exit();
		return (ENOENT);
	}
	ddp->dd_ref++;
	mutex_exit(&ddp->dd_mutex);
	i_dls_devnet_hashmap_exit();

	*ddhp = ddp;
	return (0);
}

void
dls_devnet_rele(dls_devnet_t *ddp)
{
	mutex_enter(&ddp->dd_mutex);
	VERIFY(ddp->dd_ref > 1);
	ddp->dd_ref--;
	if ((ddp->dd_flags & DD_IMPLICIT_IPTUN) && ddp->dd_ref == 1) {
		mutex_exit(&ddp->dd_mutex);
		if (i_dls_devnet_destroy_iptun(ddp->dd_linkid) != 0)
			ddp->dd_flags |= DD_IMPLICIT_IPTUN;
		return;
	}
	mutex_exit(&ddp->dd_mutex);
}

static int
dls_devnet_hold_by_name(const char *link, dls_devnet_t **ddpp)
{
	char			drv[MAXLINKNAMELEN];
	uint_t			ppa;
	major_t			major;
	dev_t			phy_dev, tmp_dev;
	datalink_id_t		linkid;
	dls_dev_handle_t	ddh;
	int			err;

	if ((err = dls_mgmt_get_linkid(link, &linkid)) == 0)
		return (dls_devnet_hold(linkid, ddpp));

	/*
	 * If we failed to get the link's linkid because the dlmgmtd daemon
	 * has not been started, return ENOENT so that the application can
	 * fallback to open the /dev node.
	 */
	if (err == EBADF)
		return (ENOENT);

	if (err != ENOENT)
		return (err);

	/*
	 * If we reach this point it means dlmgmtd is up but has no
	 * mapping for the link name.
	 */
	if (ddi_parse_dlen(link, drv, MAXLINKNAMELEN, &ppa) != DDI_SUCCESS)
		return (ENOENT);

	if (IS_IPTUN_LINK(drv)) {
		if ((err = i_dls_devnet_create_iptun(link, drv, &linkid)) != 0)
			return (err);
		/*
		 * At this point, an IP tunnel MAC has registered, which
		 * resulted in a link being created.
		 */
		err = dls_devnet_hold(linkid, ddpp);
		if (err != 0) {
			VERIFY(i_dls_devnet_destroy_iptun(linkid) == 0);
			return (err);
		}
		/*
		 * dls_devnet_rele() will know to destroy the implicit IP
		 * tunnel on last reference release if DD_IMPLICIT_IPTUN is
		 * set.
		 */
		(*ddpp)->dd_flags |= DD_IMPLICIT_IPTUN;
		return (0);
	}

	/*
	 * If this link:
	 * (a) is a physical device, (b) this is the first boot, (c) the MAC
	 * is not registered yet, and (d) we cannot find its linkid, then the
	 * linkname is the same as the devname.
	 *
	 * First filter out invalid names.
	 */
	if ((major = ddi_name_to_major(drv)) == (major_t)-1)
		return (ENOENT);

	phy_dev = makedevice(major, DLS_PPA2MINOR(ppa));
	if (softmac_hold_device(phy_dev, &ddh) != 0)
		return (ENOENT);

	/*
	 * At this time, the MAC should be registered, check its phy_dev using
	 * the given name.
	 */
	if ((err = dls_mgmt_get_linkid(link, &linkid)) != 0 ||
	    (err = dls_mgmt_get_phydev(linkid, &tmp_dev)) != 0) {
		softmac_rele_device(ddh);
		return (err);
	}
	if (tmp_dev != phy_dev) {
		softmac_rele_device(ddh);
		return (ENOENT);
	}

	err = dls_devnet_hold(linkid, ddpp);
	softmac_rele_device(ddh);
	return (err);
}

int
dls_devnet_macname2linkid(const char *macname, datalink_id_t *linkidp)
{
	dls_devnet_t	*ddp;

	i_dls_devnet_hashmap_read();
	if (mod_hash_find(i_dls_devnet_hash, (mod_hash_key_t)macname,
	    (mod_hash_val_t *)&ddp) != 0) {
		i_dls_devnet_hashmap_exit();
		return (ENOENT);
	}

	*linkidp = ddp->dd_linkid;
	i_dls_devnet_hashmap_exit();
	return (0);
}

/*
 * Get linkid for the given dev.
 */
int
dls_devnet_dev2linkid(dev_t dev, datalink_id_t *linkidp)
{
	char	macname[MAXNAMELEN];
	char	*drv;

	if ((drv = ddi_major_to_name(getmajor(dev))) == NULL)
		return (EINVAL);

	(void) snprintf(macname, sizeof (macname), "%s%d", drv,
	    DLS_MINOR2INST(getminor(dev)));
	return (dls_devnet_macname2linkid(macname, linkidp));
}

/*
 * Get the link's physical dev_t. It this is a VLAN, get the dev_t of the
 * link this VLAN is created on.
 */
int
dls_devnet_phydev(datalink_id_t vlanid, dev_t *devp)
{
	dls_devnet_t	*ddp;
	int		err;

	if ((err = dls_devnet_hold_tmp(vlanid, &ddp)) != 0)
		return (err);

	err = dls_mgmt_get_phydev(ddp->dd_linkid, devp);
	dls_devnet_rele_tmp(ddp);
	return (err);
}

/*
 * Handle the renaming requests.  There are two rename cases:
 *
 * 1. Request to rename a valid link (id1) to an non-existent link name
 *    (id2). In this case id2 is DATALINK_INVALID_LINKID.  Just check whether
 *    id1 is held by any applications.
 *
 *    In this case, the link's kstats need to be updated using the given name.
 *
 * 2. Request to rename a valid link (id1) to the name of a REMOVED
 *    physical link (id2). In this case, check that id1 and its associated
 *    mac is not held by any application, and update the link's linkid to id2.
 *
 *    This case does not change the <link name, linkid> mapping, so the link's
 *    kstats need to be updated with using name associated the given id2.
 */
int
dls_devnet_rename(datalink_id_t id1, datalink_id_t id2, const char *link)
{
	dls_dev_handle_t	ddh = NULL;
	int			err = 0;
	dev_t			phydev = 0;
	dls_devnet_t		*ddp;
	mac_perim_handle_t	mph = NULL;
	mac_handle_t		mh;
	mod_hash_val_t		val;

	/*
	 * In the second case, id2 must be a REMOVED physical link.
	 */
	if ((id2 != DATALINK_INVALID_LINKID) &&
	    (dls_mgmt_get_phydev(id2, &phydev) == 0) &&
	    softmac_hold_device(phydev, &ddh) == 0) {
		softmac_rele_device(ddh);
		return (EEXIST);
	}

	/*
	 * Hold id1 to prevent it from being detached (if a physical link).
	 */
	if (dls_mgmt_get_phydev(id1, &phydev) == 0)
		(void) softmac_hold_device(phydev, &ddh);

	/*
	 * The framework does not hold hold locks across calls to the
	 * mac perimeter, hence enter the perimeter first. This also waits
	 * for the property loading to finish.
	 */
	if ((err = mac_perim_enter_by_linkid(id1, &mph)) != 0) {
		softmac_rele_device(ddh);
		return (err);
	}

	i_dls_devnet_lock_enter();
	i_dls_devnet_hashmap_read();

	if ((err = mod_hash_find(i_dls_devnet_id_hash,
	    (mod_hash_key_t)(uintptr_t)id1, (mod_hash_val_t *)&ddp)) != 0) {
		ASSERT(err == MH_ERR_NOTFOUND);
		err = ENOENT;
		goto done;
	}

	mutex_enter(&ddp->dd_mutex);
	if (ddp->dd_ref > 1) {
		mutex_exit(&ddp->dd_mutex);
		err = EBUSY;
		goto done;
	}
	mutex_exit(&ddp->dd_mutex);

	if (id2 == DATALINK_INVALID_LINKID) {
		(void) strlcpy(ddp->dd_linkname, link,
		    sizeof (ddp->dd_linkname));

		/* rename mac client name and its flow if exists */
		if ((err = mac_open(ddp->dd_mac, &mh)) != 0)
			goto done;
		(void) mac_rename_primary(mh, link);
		mac_close(mh);
		goto done;
	}

	/*
	 * The second case, check whether the MAC is used by any MAC
	 * user.  This must be a physical link so ddh must not be NULL.
	 */
	if (ddh == NULL) {
		err = EINVAL;
		goto done;
	}

	if ((err = mac_open(ddp->dd_mac, &mh)) != 0)
		goto done;

	/*
	 * We release the reference of the MAC which mac_open() is
	 * holding. Note that this mac will not be unregistered
	 * because the physical device is held.
	 */
	mac_close(mh);

	/*
	 * Check if there is any other MAC clients, if not, hold this mac
	 * exclusively until we are done.
	 */
	if ((err = mac_mark_exclusive(mh)) != 0)
		goto done;

	/*
	 * Update the link's linkid.
	 */
	if ((err = mod_hash_find(i_dls_devnet_id_hash,
	    (mod_hash_key_t)(uintptr_t)id2, &val)) != MH_ERR_NOTFOUND) {
		mac_unmark_exclusive(mh);
		err = EEXIST;
		goto done;
	}

	/*
	 * Temporarily drop the hashmap lock for the upcall -- ddp will remain
	 * valid because we hold i_dls_devnet_lock. Taking this is a
	 * prerequisite for dls_devnet_unset to proceed, and it is the only
	 * pathway through which ddp can be freed.
	 */
	i_dls_devnet_hashmap_exit();
	i_dls_devnet_lock_upcall_start();
	err = dls_mgmt_get_linkinfo(id2, ddp->dd_linkname, NULL, NULL, NULL);
	i_dls_devnet_lock_upcall_end();

	if (err != 0) {
		mac_unmark_exclusive(mh);
		goto done_rw_unlocked;
	}

	i_dls_devnet_hashmap_write();

	(void) mod_hash_remove(i_dls_devnet_id_hash,
	    (mod_hash_key_t)(uintptr_t)id1, &val);

	ddp->dd_linkid = id2;
	(void) mod_hash_insert(i_dls_devnet_id_hash,
	    (mod_hash_key_t)(uintptr_t)ddp->dd_linkid, (mod_hash_val_t)ddp);

	mac_unmark_exclusive(mh);

	/* load properties for new id */
	mutex_enter(&ddp->dd_mutex);
	ddp->dd_prop_loaded = B_FALSE;
	ddp->dd_prop_taskid = taskq_dispatch(system_taskq,
	    dls_devnet_prop_task, ddp, TQ_SLEEP);
	mutex_exit(&ddp->dd_mutex);

done:
	i_dls_devnet_hashmap_exit();
done_rw_unlocked:
	i_dls_devnet_lock_exit();

	if (err == 0)
		dls_devnet_stat_rename(ddp);

	if (mph != NULL)
		mac_perim_exit(mph);
	softmac_rele_device(ddh);
	return (err);
}

static int
i_dls_devnet_setzid(dls_devnet_t *ddp, zoneid_t new_zoneid, boolean_t setprop,
    boolean_t transient)
{
	int			err;
	mac_perim_handle_t	mph;
	boolean_t		upcall_done = B_FALSE;
	datalink_id_t		linkid = ddp->dd_linkid;
	zoneid_t		old_zoneid = ddp->dd_zid;
	dlmgmt_door_setzoneid_t	setzid;
	dlmgmt_setzoneid_retval_t retval;

	if (old_zoneid == new_zoneid)
		return (0);

	if ((err = mac_perim_enter_by_macname(ddp->dd_mac, &mph)) != 0)
		return (err);

	/*
	 * When changing the zoneid of an existing link, we need to tell
	 * dlmgmtd about it.  dlmgmtd already knows the zoneid associated with
	 * newly created links.
	 */
	if (setprop) {
		setzid.ld_cmd = DLMGMT_CMD_SETZONEID;
		setzid.ld_linkid = linkid;
		setzid.ld_zoneid = new_zoneid;
		err = i_dls_mgmt_upcall(&setzid, sizeof (setzid), &retval,
		    sizeof (retval));
		if (err != 0)
			goto done;

		/*
		 * We set upcall_done only if the upcall is
		 * successful. This way, if dls_link_setzid() fails,
		 * we know another upcall must be done to reset the
		 * dlmgmtd state.
		 */
		upcall_done = B_TRUE;
	}
	if ((err = dls_link_setzid(ddp->dd_mac, new_zoneid)) == 0) {
		ddp->dd_zid = new_zoneid;
		ddp->dd_transient = transient;
		devnet_need_rebuild = B_TRUE;
	}

done:
	if (err != 0 && upcall_done) {
		setzid.ld_zoneid = old_zoneid;
		(void) i_dls_mgmt_upcall(&setzid, sizeof (setzid), &retval,
		    sizeof (retval));
	}
	mac_perim_exit(mph);
	return (err);
}

int
dls_devnet_setzid(dls_dl_handle_t ddh, zoneid_t new_zid)
{
	dls_devnet_t	*ddp;
	int		err;
	zoneid_t	old_zid;
	boolean_t	refheld = B_FALSE;

	old_zid = ddh->dd_zid;

	if (old_zid == new_zid)
		return (0);

	/*
	 * Acquire an additional reference to the link if it is being assigned
	 * to a non-global zone from the global zone.
	 */
	if (old_zid == GLOBAL_ZONEID && new_zid != GLOBAL_ZONEID) {
		if ((err = dls_devnet_hold(ddh->dd_linkid, &ddp)) != 0)
			return (err);
		refheld = B_TRUE;
	}

	if ((err = i_dls_devnet_setzid(ddh, new_zid, B_TRUE, B_FALSE)) != 0) {
		if (refheld)
			dls_devnet_rele(ddp);
		return (err);
	}

	/*
	 * Release the additional reference if the link is returning to the
	 * global zone from a non-global zone.
	 */
	if (old_zid != GLOBAL_ZONEID && new_zid == GLOBAL_ZONEID)
		dls_devnet_rele(ddh);

	/* Re-create kstats in the appropriate zones. */
	if (old_zid != GLOBAL_ZONEID)
		dls_devnet_stat_destroy(ddh, old_zid);
	if (new_zid != GLOBAL_ZONEID)
		dls_devnet_stat_create(ddh, new_zid);

	return (0);
}

zoneid_t
dls_devnet_getzid(dls_dl_handle_t ddh)
{
	return (((dls_devnet_t *)ddh)->dd_zid);
}

zoneid_t
dls_devnet_getownerzid(dls_dl_handle_t ddh)
{
	return (((dls_devnet_t *)ddh)->dd_owner_zid);
}

/*
 * Is linkid visible from zoneid?  A link is visible if it was created in the
 * zone, or if it is currently assigned to the zone.
 */
boolean_t
dls_devnet_islinkvisible(datalink_id_t linkid, zoneid_t zoneid)
{
	dls_devnet_t	*ddp;
	boolean_t	result;

	if (dls_devnet_hold_tmp(linkid, &ddp) != 0)
		return (B_FALSE);
	result = (ddp->dd_owner_zid == zoneid || ddp->dd_zid == zoneid);
	dls_devnet_rele_tmp(ddp);
	return (result);
}

/*
 * Access a vanity naming node.
 */
int
dls_devnet_open(const char *link, dls_dl_handle_t *dhp, dev_t *devp)
{
	dls_devnet_t	*ddp;
	dls_link_t	*dlp;
	zoneid_t	zid = getzoneid();
	int		err;
	mac_perim_handle_t	mph;

	if ((err = dls_devnet_hold_by_name(link, &ddp)) != 0)
		return (err);

	dls_devnet_prop_task_wait(ddp);

	/*
	 * Opening a link that does not belong to the current non-global zone
	 * is not allowed.
	 */
	if (zid != GLOBAL_ZONEID && ddp->dd_zid != zid) {
		dls_devnet_rele(ddp);
		return (ENOENT);
	}

	err = mac_perim_enter_by_macname(ddp->dd_mac, &mph);
	if (err != 0) {
		dls_devnet_rele(ddp);
		return (err);
	}

	err = dls_link_hold_create(ddp->dd_mac, &dlp);
	mac_perim_exit(mph);

	if (err != 0) {
		dls_devnet_rele(ddp);
		return (err);
	}

	*dhp = ddp;
	*devp = dls_link_dev(dlp);
	return (0);
}

/*
 * Close access to a vanity naming node.
 */
void
dls_devnet_close(dls_dl_handle_t dlh)
{
	dls_devnet_t	*ddp = dlh;
	dls_link_t	*dlp;
	mac_perim_handle_t	mph;

	VERIFY(mac_perim_enter_by_macname(ddp->dd_mac, &mph) == 0);
	VERIFY(dls_link_hold(ddp->dd_mac, &dlp) == 0);

	/*
	 * One rele for the hold placed in dls_devnet_open, another for
	 * the hold done just above
	 */
	dls_link_rele(dlp);
	dls_link_rele(dlp);
	mac_perim_exit(mph);

	dls_devnet_rele(ddp);
}

/*
 * This is used by /dev/net to rebuild the nodes for readdir().  It is not
 * critical and no protection is needed.
 */
boolean_t
dls_devnet_rebuild()
{
	boolean_t updated = devnet_need_rebuild;

	devnet_need_rebuild = B_FALSE;
	return (updated);
}

int
dls_devnet_create(mac_handle_t mh, datalink_id_t linkid, zoneid_t zoneid)
{
	dls_link_t	*dlp;
	dls_devnet_t	*ddp;
	int		err;
	mac_perim_handle_t mph;

	/*
	 * Holding the mac perimeter ensures that the downcall from the
	 * dlmgmt daemon which does the property loading does not proceed
	 * until we relinquish the perimeter.
	 */
	mac_perim_enter_by_mh(mh, &mph);
	/*
	 * Make this association before we call dls_link_hold_create as
	 * we need to use the linkid to get the user name for the link
	 * when we create the MAC client.
	 */
	if ((err = dls_devnet_set(mh, linkid, zoneid, &ddp)) == 0) {
		if ((err = dls_link_hold_create(mac_name(mh), &dlp)) != 0) {
			mac_perim_exit(mph);
			(void) dls_devnet_unset(mh, &linkid, B_FALSE);
			return (err);
		}

		/*
		 * If dd_linkid is set then the link was successfully
		 * initialized. In this case we can remove the
		 * initializing flag and make the link visible to the
		 * rest of the system.
		 *
		 * If not set then we were called by softmac and it
		 * was unable to obtain a linkid for the physical link
		 * because dlmgmtd is down. In that case softmac will
		 * eventually obtain a linkid and call
		 * dls_devnet_recreate() to complete initialization.
		 */
		mutex_enter(&ddp->dd_mutex);
		if (ddp->dd_linkid != DATALINK_INVALID_LINKID)
			ddp->dd_flags &= ~DD_INITIALIZING;
		mutex_exit(&ddp->dd_mutex);

	}

	mac_perim_exit(mph);
	return (err);
}

/*
 * Set the linkid of the dls_devnet_t and add it into the i_dls_devnet_id_hash.
 * This is called in the case that the dlmgmtd daemon is started later than
 * the physical devices get attached, and the linkid is only known after the
 * daemon starts.
 */
int
dls_devnet_recreate(mac_handle_t mh, datalink_id_t linkid)
{
	dls_devnet_t	*ddp;
	int		err;

	VERIFY(linkid != DATALINK_INVALID_LINKID);
	if ((err = dls_devnet_set(mh, linkid, GLOBAL_ZONEID, &ddp)) == 0) {
		mutex_enter(&ddp->dd_mutex);
		if (ddp->dd_linkid != DATALINK_INVALID_LINKID)
			ddp->dd_flags &= ~DD_INITIALIZING;
		mutex_exit(&ddp->dd_mutex);
	}

	return (err);

}

int
dls_devnet_destroy(mac_handle_t mh, datalink_id_t *idp, boolean_t wait)
{
	int			err;
	mac_perim_handle_t	mph;

	*idp = DATALINK_INVALID_LINKID;
	err = dls_devnet_unset(mh, idp, wait);

	/*
	 * We continue on in the face of ENOENT because the devnet
	 * unset and DLS link release are not atomic and we may have a
	 * scenario where there is no entry in i_dls_devnet_hash for
	 * the MAC name but there is an entry in i_dls_link_hash. For
	 * example, if the following occurred:
	 *
	 * 1. dls_devnet_unset() returns success, and
	 *
	 * 2. dls_link_rele_by_name() fails with ENOTEMPTY because
	 *    flows still exist, and
	 *
	 * 3. dls_devnet_set() fails to set the zone id and calls
	 *    dls_devnet_unset() -- leaving an entry in
	 *    i_dls_link_hash but no corresponding entry in
	 *    i_dls_devnet_hash.
	 *
	 * Even if #3 wasn't true the dls_devnet_set() may fail for
	 * different reasons in the future; the point is that it _can_
	 * fail as part of its contract. We can't rely on it working
	 * so we must assume that these two pieces of state (devnet
	 * and link hashes), which should always be in sync, can get
	 * out of sync and thus even if we get ENOENT from the devnet
	 * hash we should still try to delete from the link hash just
	 * in case.
	 *
	 * We could prevent the ENOTEMPTY from dls_link_rele_by_name()
	 * by calling mac_disable() before calling
	 * dls_devnet_destroy() but that's not currently possible due
	 * to a long-standing bug. OpenSolaris 6791335: The semantics
	 * of mac_disable() were modified by Crossbow such that
	 * dls_devnet_destroy() needs to be called before
	 * mac_disable() can succeed. This is because of the implicit
	 * reference that dls has on the mac_impl_t.
	 */
	if (err != 0 && err != ENOENT)
		return (err);

	mac_perim_enter_by_mh(mh, &mph);
	err = dls_link_rele_by_name(mac_name(mh));
	mac_perim_exit(mph);

	if (err != 0) {
		dls_devnet_t	*ddp;

		/*
		 * XXX It is a general GLDv3 bug that dls_devnet_set() has to
		 * be called to re-set the link when destroy fails.  The
		 * zoneid below will be incorrect if this function is ever
		 * called from kernel context or from a zone other than that
		 * which initially created the link.
		 */
		(void) dls_devnet_set(mh, *idp, crgetzoneid(CRED()), &ddp);

		/*
		 * You might think dd_linkid should always be set
		 * here, but in the case where dls_devnet_unset()
		 * returns ENOENT it will be DATALINK_INVALID_LINKID.
		 * Stay consistent with the rest of DLS and only
		 * remove the initializing flag if linkid is set.
		 */
		mutex_enter(&ddp->dd_mutex);
		if (ddp->dd_linkid != DATALINK_INVALID_LINKID)
			ddp->dd_flags &= ~DD_INITIALIZING;
		mutex_exit(&ddp->dd_mutex);
	}
	return (err);
}

/*
 * Implicitly create an IP tunnel link.
 */
static int
i_dls_devnet_create_iptun(const char *linkname, const char *drvname,
    datalink_id_t *linkid)
{
	int		err;
	iptun_kparams_t	ik;
	uint32_t	media;
	netstack_t	*ns;
	major_t		iptun_major;
	dev_info_t	*iptun_dip;

	/* First ensure that the iptun device is attached. */
	if ((iptun_major = ddi_name_to_major(IPTUN_DRIVER_NAME)) == (major_t)-1)
		return (EINVAL);
	if ((iptun_dip = ddi_hold_devi_by_instance(iptun_major, 0, 0)) == NULL)
		return (EINVAL);

	if (IS_IPV4_TUN(drvname)) {
		ik.iptun_kparam_type = IPTUN_TYPE_IPV4;
		media = DL_IPV4;
	} else if (IS_6TO4_TUN(drvname)) {
		ik.iptun_kparam_type = IPTUN_TYPE_6TO4;
		media = DL_6TO4;
	} else if (IS_IPV6_TUN(drvname)) {
		ik.iptun_kparam_type = IPTUN_TYPE_IPV6;
		media = DL_IPV6;
	}
	ik.iptun_kparam_flags = (IPTUN_KPARAM_TYPE | IPTUN_KPARAM_IMPLICIT);

	/* Obtain a datalink id for this tunnel. */
	err = dls_mgmt_create((char *)linkname, 0, DATALINK_CLASS_IPTUN, media,
	    B_FALSE, &ik.iptun_kparam_linkid);
	if (err != 0) {
		ddi_release_devi(iptun_dip);
		return (err);
	}

	ns = netstack_get_current();
	err = iptun_create(&ik, CRED());
	netstack_rele(ns);

	if (err != 0)
		VERIFY(dls_mgmt_destroy(ik.iptun_kparam_linkid, B_FALSE) == 0);
	else
		*linkid = ik.iptun_kparam_linkid;

	ddi_release_devi(iptun_dip);
	return (err);
}

static int
i_dls_devnet_destroy_iptun(datalink_id_t linkid)
{
	int err;

	/*
	 * Note the use of zone_kcred() here as opposed to CRED().  This is
	 * because the process that does the last close of this /dev/net node
	 * may not have necessary privileges to delete this IP tunnel, but the
	 * tunnel must always be implicitly deleted on last close.
	 */
	if ((err = iptun_delete(linkid, zone_kcred())) == 0)
		(void) dls_mgmt_destroy(linkid, B_FALSE);
	return (err);
}

const char *
dls_devnet_link(dls_dl_handle_t ddh)
{
	return (ddh->dd_linkname);
}

const char *
dls_devnet_mac(dls_dl_handle_t ddh)
{
	return (ddh->dd_mac);
}

datalink_id_t
dls_devnet_linkid(dls_dl_handle_t ddh)
{
	return (ddh->dd_linkid);
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

/*
 * Data-Link Services Module
 */

#include	<sys/modctl.h>
#include	<sys/dld_impl.h>

static struct modlmisc		i_dls_modlmisc = {
	&mod_miscops,
	DLS_INFO
};

static struct modlinkage	i_dls_modlinkage = {
	MODREV_1,
	&i_dls_modlmisc,
	NULL
};

/*
 * Module initialization functions.
 */

static void
i_dls_mod_init(void)
{
	dls_link_init();
	dls_mgmt_init();
}

static int
i_dls_mod_fini(void)
{
	int	err;

	if ((err = dls_link_fini()) != 0)
		return (err);

	dls_mgmt_fini();
	return (0);
}

/*
 * modlinkage functions.
 */

int
_init(void)
{
	int	err;

	i_dls_mod_init();

	if ((err = mod_install(&i_dls_modlinkage)) != 0) {
		(void) i_dls_mod_fini();
		return (err);
	}

#ifdef	DEBUG
	cmn_err(CE_NOTE, "!%s loaded", DLS_INFO);
#endif	/* DEBUG */

	return (0);
}

int
_fini(void)
{
	int	err;

	if ((err = i_dls_mod_fini()) != 0)
		return (err);

	if ((err = mod_remove(&i_dls_modlinkage)) != 0)
		return (err);

#ifdef	DEBUG
	cmn_err(CE_NOTE, "!%s unloaded", DLS_INFO);
#endif	/* DEBUG */

	return (0);
}

int
_info(struct modinfo *modinfop)
{
	return (mod_info(&i_dls_modlinkage, modinfop));
}
/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

/*
 * Data-Link Services Module
 */

#include <sys/dld_impl.h>
#include <sys/mac_ether.h>

static mac_stat_info_t	i_dls_si[] = {
	{ MAC_STAT_IFSPEED, "ifspeed", KSTAT_DATA_UINT64, 0 },
	{ MAC_STAT_MULTIRCV, "multircv", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_BRDCSTRCV, "brdcstrcv", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_MULTIXMT, "multixmt", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_BRDCSTXMT, "brdcstxmt", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_NORCVBUF, "norcvbuf", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_IERRORS, "ierrors", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_NOXMTBUF, "noxmtbuf", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_OERRORS, "oerrors", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_COLLISIONS, "collisions", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_RBYTES, "rbytes", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_IPACKETS, "ipackets", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_OBYTES, "obytes", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_OPACKETS, "opackets", KSTAT_DATA_UINT32, 0 },
	{ MAC_STAT_RBYTES, "rbytes64", KSTAT_DATA_UINT64, 0 },
	{ MAC_STAT_IPACKETS, "ipackets64", KSTAT_DATA_UINT64, 0 },
	{ MAC_STAT_OBYTES, "obytes64", KSTAT_DATA_UINT64, 0 },
	{ MAC_STAT_OPACKETS, "opackets64", KSTAT_DATA_UINT64, 0 },
	{ MAC_STAT_LINK_STATE, "link_state", KSTAT_DATA_UINT32,
	    (uint64_t)LINK_STATE_UNKNOWN}
};

#define	STAT_INFO_COUNT	(sizeof (i_dls_si) / sizeof (i_dls_si[0]))

/*
 * Exported functions.
 */
int
dls_stat_update(kstat_t *ksp, dls_link_t *dlp, int rw)
{
	kstat_named_t	*knp;
	uint_t		i;
	uint64_t	val;

	if (rw != KSTAT_READ)
		return (EACCES);

	knp = (kstat_named_t *)ksp->ks_data;
	for (i = 0; i < STAT_INFO_COUNT; i++) {
		val = mac_stat_get(dlp->dl_mh, i_dls_si[i].msi_stat);

		switch (i_dls_si[i].msi_type) {
		case KSTAT_DATA_UINT64:
			knp->value.ui64 = val;
			break;
		case KSTAT_DATA_UINT32:
			knp->value.ui32 = (uint32_t)val;
			break;
		default:
			ASSERT(B_FALSE);
		}

		knp++;
	}

	/*
	 * Ethernet specific kstat "link_duplex"
	 */
	if (dlp->dl_mip->mi_nativemedia != DL_ETHER) {
		knp->value.ui32 = LINK_DUPLEX_UNKNOWN;
	} else {
		val = mac_stat_get(dlp->dl_mh, ETHER_STAT_LINK_DUPLEX);
		knp->value.ui32 = (uint32_t)val;
	}
	knp++;
	knp->value.ui32 = dlp->dl_unknowns;

	return (0);
}

int
dls_stat_create(const char *module, int instance, const char *name,
    zoneid_t zoneid, int (*update)(struct kstat *, int), void *private,
    kstat_t **kspp)
{
	kstat_t		*ksp;
	kstat_named_t	*knp;
	uint_t		i;

	if ((ksp = kstat_create_zone(module, instance, name, "net",
	    KSTAT_TYPE_NAMED, STAT_INFO_COUNT + 2, 0, zoneid)) == NULL) {
		return (EINVAL);
	}

	ksp->ks_update = update;
	ksp->ks_private = private;

	knp = (kstat_named_t *)ksp->ks_data;
	for (i = 0; i < STAT_INFO_COUNT; i++) {
		kstat_named_init(knp, i_dls_si[i].msi_name,
		    i_dls_si[i].msi_type);
		knp++;
	}

	kstat_named_init(knp++, "link_duplex", KSTAT_DATA_UINT32);
	kstat_named_init(knp, "unknowns", KSTAT_DATA_UINT32);
	kstat_install(ksp);
	*kspp = ksp;
	return (0);
}