usr/src/uts/common/inet/ip/sadb.c

/*
 * 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.
 */

#include <sys/types.h>
#include <sys/stream.h>
#include <sys/stropts.h>
#include <sys/strsubr.h>
#include <sys/errno.h>
#include <sys/ddi.h>
#include <sys/debug.h>
#include <sys/cmn_err.h>
#include <sys/stream.h>
#include <sys/strlog.h>
#include <sys/kmem.h>
#include <sys/sunddi.h>
#include <sys/tihdr.h>
#include <sys/atomic.h>
#include <sys/socket.h>
#include <sys/sysmacros.h>
#include <sys/crypto/common.h>
#include <sys/crypto/api.h>
#include <sys/zone.h>
#include <netinet/in.h>
#include <net/if.h>
#include <net/pfkeyv2.h>
#include <inet/common.h>
#include <netinet/ip6.h>
#include <inet/ip.h>
#include <inet/ip_ire.h>
#include <inet/ip6.h>
#include <inet/ipsec_info.h>
#include <inet/tcp.h>
#include <inet/sadb.h>
#include <inet/ipsec_impl.h>
#include <inet/ipsecah.h>
#include <inet/ipsecesp.h>
#include <sys/random.h>
#include <sys/dlpi.h>
#include <sys/iphada.h>
#include <inet/ip_if.h>
#include <inet/ipdrop.h>
#include <inet/ipclassifier.h>
#include <inet/sctp_ip.h>
#include <inet/tun.h>

/*
 * This source file contains Security Association Database (SADB) common
 * routines.  They are linked in with the AH module.  Since AH has no chance
 * of falling under export control, it was safe to link it in there.
 */

static mblk_t *sadb_extended_acquire(ipsec_selector_t *, ipsec_policy_t *,
    ipsec_action_t *, boolean_t, uint32_t, uint32_t, netstack_t *);
static void sadb_ill_df(ill_t *, mblk_t *, isaf_t *, int, boolean_t);
static ipsa_t *sadb_torch_assoc(isaf_t *, ipsa_t *, boolean_t, mblk_t **);
static void sadb_drain_torchq(queue_t *, mblk_t *);
static void sadb_destroy_acqlist(iacqf_t **, uint_t, boolean_t,
			    netstack_t *);
static void sadb_destroy(sadb_t *, netstack_t *);
static mblk_t *sadb_sa2msg(ipsa_t *, sadb_msg_t *);

static time_t sadb_add_time(time_t, uint64_t);
static void lifetime_fuzz(ipsa_t *);
static void age_pair_peer_list(templist_t *, sadb_t *, boolean_t);
static void ipsa_set_replay(ipsa_t *ipsa, uint32_t offset);

extern void (*cl_inet_getspi)(netstackid_t stack_id, uint8_t protocol,
    uint8_t *ptr, size_t len, void *args);
extern int (*cl_inet_checkspi)(netstackid_t stack_id, uint8_t protocol,
    uint32_t spi, void *args);
extern void (*cl_inet_deletespi)(netstackid_t stack_id, uint8_t protocol,
    uint32_t spi, void *args);

/*
 * ipsacq_maxpackets is defined here to make it tunable
 * from /etc/system.
 */
extern uint64_t ipsacq_maxpackets;

#define	SET_EXPIRE(sa, delta, exp) {				\
	if (((sa)->ipsa_ ## delta) != 0) {				\
		(sa)->ipsa_ ## exp = sadb_add_time((sa)->ipsa_addtime,	\
			(sa)->ipsa_ ## delta);				\
	}								\
}

#define	UPDATE_EXPIRE(sa, delta, exp) {					\
	if (((sa)->ipsa_ ## delta) != 0) {				\
		time_t tmp = sadb_add_time((sa)->ipsa_usetime,		\
			(sa)->ipsa_ ## delta);				\
		if (((sa)->ipsa_ ## exp) == 0)				\
			(sa)->ipsa_ ## exp = tmp;			\
		else							\
			(sa)->ipsa_ ## exp = 				\
			    MIN((sa)->ipsa_ ## exp, tmp); 		\
	}								\
}


/* wrap the macro so we can pass it as a function pointer */
void
sadb_sa_refrele(void *target)
{
	IPSA_REFRELE(((ipsa_t *)target));
}

/*
 * We presume that sizeof (long) == sizeof (time_t) and that time_t is
 * a signed type.
 */
#define	TIME_MAX LONG_MAX

/*
 * PF_KEY gives us lifetimes in uint64_t seconds.  We presume that
 * time_t is defined to be a signed type with the same range as
 * "long".  On ILP32 systems, we thus run the risk of wrapping around
 * at end of time, as well as "overwrapping" the clock back around
 * into a seemingly valid but incorrect future date earlier than the
 * desired expiration.
 *
 * In order to avoid odd behavior (either negative lifetimes or loss
 * of high order bits) when someone asks for bizarrely long SA
 * lifetimes, we do a saturating add for expire times.
 *
 * We presume that ILP32 systems will be past end of support life when
 * the 32-bit time_t overflows (a dangerous assumption, mind you..).
 *
 * On LP64, 2^64 seconds are about 5.8e11 years, at which point we
 * will hopefully have figured out clever ways to avoid the use of
 * fixed-sized integers in computation.
 */
static time_t
sadb_add_time(time_t base, uint64_t delta)
{
	time_t sum;

	/*
	 * Clip delta to the maximum possible time_t value to
	 * prevent "overwrapping" back into a shorter-than-desired
	 * future time.
	 */
	if (delta > TIME_MAX)
		delta = TIME_MAX;
	/*
	 * This sum may still overflow.
	 */
	sum = base + delta;

	/*
	 * .. so if the result is less than the base, we overflowed.
	 */
	if (sum < base)
		sum = TIME_MAX;

	return (sum);
}

/*
 * Callers of this function have already created a working security
 * association, and have found the appropriate table & hash chain.  All this
 * function does is check duplicates, and insert the SA.  The caller needs to
 * hold the hash bucket lock and increment the refcnt before insertion.
 *
 * Return 0 if success, EEXIST if collision.
 */
#define	SA_UNIQUE_MATCH(sa1, sa2) \
	(((sa1)->ipsa_unique_id & (sa1)->ipsa_unique_mask) == \
	((sa2)->ipsa_unique_id & (sa2)->ipsa_unique_mask))

int
sadb_insertassoc(ipsa_t *ipsa, isaf_t *bucket)
{
	ipsa_t **ptpn = NULL;
	ipsa_t *walker;
	boolean_t unspecsrc;

	ASSERT(MUTEX_HELD(&bucket->isaf_lock));

	unspecsrc = IPSA_IS_ADDR_UNSPEC(ipsa->ipsa_srcaddr, ipsa->ipsa_addrfam);

	walker = bucket->isaf_ipsa;
	ASSERT(walker == NULL || ipsa->ipsa_addrfam == walker->ipsa_addrfam);

	/*
	 * Find insertion point (pointed to with **ptpn).  Insert at the head
	 * of the list unless there's an unspecified source address, then
	 * insert it after the last SA with a specified source address.
	 *
	 * BTW, you'll have to walk the whole chain, matching on {DST, SPI}
	 * checking for collisions.
	 */

	while (walker != NULL) {
		if (IPSA_ARE_ADDR_EQUAL(walker->ipsa_dstaddr,
		    ipsa->ipsa_dstaddr, ipsa->ipsa_addrfam)) {
			if (walker->ipsa_spi == ipsa->ipsa_spi)
				return (EEXIST);

			mutex_enter(&walker->ipsa_lock);
			if (ipsa->ipsa_state == IPSA_STATE_MATURE &&
			    (walker->ipsa_flags & IPSA_F_USED) &&
			    SA_UNIQUE_MATCH(walker, ipsa)) {
				walker->ipsa_flags |= IPSA_F_CINVALID;
			}
			mutex_exit(&walker->ipsa_lock);
		}

		if (ptpn == NULL && unspecsrc) {
			if (IPSA_IS_ADDR_UNSPEC(walker->ipsa_srcaddr,
			    walker->ipsa_addrfam))
				ptpn = walker->ipsa_ptpn;
			else if (walker->ipsa_next == NULL)
				ptpn = &walker->ipsa_next;
		}

		walker = walker->ipsa_next;
	}

	if (ptpn == NULL)
		ptpn = &bucket->isaf_ipsa;
	ipsa->ipsa_next = *ptpn;
	ipsa->ipsa_ptpn = ptpn;
	if (ipsa->ipsa_next != NULL)
		ipsa->ipsa_next->ipsa_ptpn = &ipsa->ipsa_next;
	*ptpn = ipsa;
	ipsa->ipsa_linklock = &bucket->isaf_lock;

	return (0);
}
#undef SA_UNIQUE_MATCH

/*
 * Free a security association.  Its reference count is 0, which means
 * I must free it.  The SA must be unlocked and must not be linked into
 * any fanout list.
 */
static void
sadb_freeassoc(ipsa_t *ipsa)
{
	ipsec_stack_t	*ipss = ipsa->ipsa_netstack->netstack_ipsec;

	ASSERT(ipss != NULL);
	ASSERT(MUTEX_NOT_HELD(&ipsa->ipsa_lock));
	ASSERT(ipsa->ipsa_refcnt == 0);
	ASSERT(ipsa->ipsa_next == NULL);
	ASSERT(ipsa->ipsa_ptpn == NULL);

	mutex_enter(&ipsa->ipsa_lock);
	/* Don't call sadb_clear_lpkt() since we hold the ipsa_lock anyway. */
	ip_drop_packet(ipsa->ipsa_lpkt, B_TRUE, NULL, NULL,
	    DROPPER(ipss, ipds_sadb_inlarval_timeout),
	    &ipss->ipsec_sadb_dropper);
	ipsec_destroy_ctx_tmpl(ipsa, IPSEC_ALG_AUTH);
	ipsec_destroy_ctx_tmpl(ipsa, IPSEC_ALG_ENCR);
	mutex_exit(&ipsa->ipsa_lock);

	/* bzero() these fields for paranoia's sake. */
	if (ipsa->ipsa_authkey != NULL) {
		bzero(ipsa->ipsa_authkey, ipsa->ipsa_authkeylen);
		kmem_free(ipsa->ipsa_authkey, ipsa->ipsa_authkeylen);
	}
	if (ipsa->ipsa_encrkey != NULL) {
		bzero(ipsa->ipsa_encrkey, ipsa->ipsa_encrkeylen);
		kmem_free(ipsa->ipsa_encrkey, ipsa->ipsa_encrkeylen);
	}
	if (ipsa->ipsa_src_cid != NULL) {
		IPSID_REFRELE(ipsa->ipsa_src_cid);
	}
	if (ipsa->ipsa_dst_cid != NULL) {
		IPSID_REFRELE(ipsa->ipsa_dst_cid);
	}
	if (ipsa->ipsa_integ != NULL)
		kmem_free(ipsa->ipsa_integ, ipsa->ipsa_integlen);
	if (ipsa->ipsa_sens != NULL)
		kmem_free(ipsa->ipsa_sens, ipsa->ipsa_senslen);

	mutex_destroy(&ipsa->ipsa_lock);
	kmem_free(ipsa, sizeof (*ipsa));
}

/*
 * Unlink a security association from a hash bucket.  Assume the hash bucket
 * lock is held, but the association's lock is not.
 *
 * Note that we do not bump the bucket's generation number here because
 * we might not be making a visible change to the set of visible SA's.
 * All callers MUST bump the bucket's generation number before they unlock
 * the bucket if they use sadb_unlinkassoc to permanetly remove an SA which
 * was present in the bucket at the time it was locked.
 */
void
sadb_unlinkassoc(ipsa_t *ipsa)
{
	ASSERT(ipsa->ipsa_linklock != NULL);
	ASSERT(MUTEX_HELD(ipsa->ipsa_linklock));

	/* These fields are protected by the link lock. */
	*(ipsa->ipsa_ptpn) = ipsa->ipsa_next;
	if (ipsa->ipsa_next != NULL) {
		ipsa->ipsa_next->ipsa_ptpn = ipsa->ipsa_ptpn;
		ipsa->ipsa_next = NULL;
	}

	ipsa->ipsa_ptpn = NULL;

	/* This may destroy the SA. */
	IPSA_REFRELE(ipsa);
}

void
sadb_delete_cluster(ipsa_t *assoc)
{
	uint8_t protocol;

	if (cl_inet_deletespi &&
	    ((assoc->ipsa_state == IPSA_STATE_LARVAL) ||
	    (assoc->ipsa_state == IPSA_STATE_MATURE))) {
		protocol = (assoc->ipsa_type == SADB_SATYPE_AH) ?
		    IPPROTO_AH : IPPROTO_ESP;
		cl_inet_deletespi(assoc->ipsa_netstack->netstack_stackid,
		    protocol, assoc->ipsa_spi, NULL);
	}
}

/*
 * Create a larval security association with the specified SPI.	 All other
 * fields are zeroed.
 */
static ipsa_t *
sadb_makelarvalassoc(uint32_t spi, uint32_t *src, uint32_t *dst, int addrfam,
    netstack_t *ns)
{
	ipsa_t *newbie;

	/*
	 * Allocate...
	 */

	newbie = (ipsa_t *)kmem_zalloc(sizeof (ipsa_t), KM_NOSLEEP);
	if (newbie == NULL) {
		/* Can't make new larval SA. */
		return (NULL);
	}

	/* Assigned requested SPI, assume caller does SPI allocation magic. */
	newbie->ipsa_spi = spi;
	newbie->ipsa_netstack = ns;	/* No netstack_hold */

	/*
	 * Copy addresses...
	 */

	IPSA_COPY_ADDR(newbie->ipsa_srcaddr, src, addrfam);
	IPSA_COPY_ADDR(newbie->ipsa_dstaddr, dst, addrfam);

	newbie->ipsa_addrfam = addrfam;

	/*
	 * Set common initialization values, including refcnt.
	 */
	mutex_init(&newbie->ipsa_lock, NULL, MUTEX_DEFAULT, NULL);
	newbie->ipsa_state = IPSA_STATE_LARVAL;
	newbie->ipsa_refcnt = 1;
	newbie->ipsa_freefunc = sadb_freeassoc;

	/*
	 * There aren't a lot of other common initialization values, as
	 * they are copied in from the PF_KEY message.
	 */

	return (newbie);
}

/*
 * Call me to initialize a security association fanout.
 */
static int
sadb_init_fanout(isaf_t **tablep, uint_t size, int kmflag)
{
	isaf_t *table;
	int i;

	table = (isaf_t *)kmem_alloc(size * sizeof (*table), kmflag);
	*tablep = table;

	if (table == NULL)
		return (ENOMEM);

	for (i = 0; i < size; i++) {
		mutex_init(&(table[i].isaf_lock), NULL, MUTEX_DEFAULT, NULL);
		table[i].isaf_ipsa = NULL;
		table[i].isaf_gen = 0;
	}

	return (0);
}

/*
 * Call me to initialize an acquire fanout
 */
static int
sadb_init_acfanout(iacqf_t **tablep, uint_t size, int kmflag)
{
	iacqf_t *table;
	int i;

	table = (iacqf_t *)kmem_alloc(size * sizeof (*table), kmflag);
	*tablep = table;

	if (table == NULL)
		return (ENOMEM);

	for (i = 0; i < size; i++) {
		mutex_init(&(table[i].iacqf_lock), NULL, MUTEX_DEFAULT, NULL);
		table[i].iacqf_ipsacq = NULL;
	}

	return (0);
}

/*
 * Attempt to initialize an SADB instance.  On failure, return ENOMEM;
 * caller must clean up partial allocations.
 */
static int
sadb_init_trial(sadb_t *sp, uint_t size, int kmflag)
{
	ASSERT(sp->sdb_of == NULL);
	ASSERT(sp->sdb_if == NULL);
	ASSERT(sp->sdb_acq == NULL);

	sp->sdb_hashsize = size;
	if (sadb_init_fanout(&sp->sdb_of, size, kmflag) != 0)
		return (ENOMEM);
	if (sadb_init_fanout(&sp->sdb_if, size, kmflag) != 0)
		return (ENOMEM);
	if (sadb_init_acfanout(&sp->sdb_acq, size, kmflag) != 0)
		return (ENOMEM);

	return (0);
}

/*
 * Call me to initialize an SADB instance; fall back to default size on failure.
 */
static void
sadb_init(const char *name, sadb_t *sp, uint_t size, uint_t ver,
    netstack_t *ns)
{
	ASSERT(sp->sdb_of == NULL);
	ASSERT(sp->sdb_if == NULL);
	ASSERT(sp->sdb_acq == NULL);

	if (size < IPSEC_DEFAULT_HASH_SIZE)
		size = IPSEC_DEFAULT_HASH_SIZE;

	if (sadb_init_trial(sp, size, KM_NOSLEEP) != 0) {

		cmn_err(CE_WARN,
		    "Unable to allocate %u entry IPv%u %s SADB hash table",
		    size, ver, name);

		sadb_destroy(sp, ns);
		size = IPSEC_DEFAULT_HASH_SIZE;
		cmn_err(CE_WARN, "Falling back to %d entries", size);
		(void) sadb_init_trial(sp, size, KM_SLEEP);
	}
}


/*
 * Initialize an SADB-pair.
 */
void
sadbp_init(const char *name, sadbp_t *sp, int type, int size, netstack_t *ns)
{
	sadb_init(name, &sp->s_v4, size, 4, ns);
	sadb_init(name, &sp->s_v6, size, 6, ns);

	sp->s_satype = type;

	ASSERT((type == SADB_SATYPE_AH) || (type == SADB_SATYPE_ESP));
	if (type == SADB_SATYPE_AH) {
		ipsec_stack_t	*ipss = ns->netstack_ipsec;

		ip_drop_register(&ipss->ipsec_sadb_dropper, "IPsec SADB");
		sp->s_addflags = AH_ADD_SETTABLE_FLAGS;
		sp->s_updateflags = AH_UPDATE_SETTABLE_FLAGS;
	} else {
		sp->s_addflags = ESP_ADD_SETTABLE_FLAGS;
		sp->s_updateflags = ESP_UPDATE_SETTABLE_FLAGS;
	}
}

/*
 * Deliver a single SADB_DUMP message representing a single SA.  This is
 * called many times by sadb_dump().
 *
 * If the return value of this is ENOBUFS (not the same as ENOMEM), then
 * the caller should take that as a hint that dupb() on the "original answer"
 * failed, and that perhaps the caller should try again with a copyb()ed
 * "original answer".
 */
static int
sadb_dump_deliver(queue_t *pfkey_q, mblk_t *original_answer, ipsa_t *ipsa,
    sadb_msg_t *samsg)
{
	mblk_t *answer;

	answer = dupb(original_answer);
	if (answer == NULL)
		return (ENOBUFS);
	answer->b_cont = sadb_sa2msg(ipsa, samsg);
	if (answer->b_cont == NULL) {
		freeb(answer);
		return (ENOMEM);
	}

	/* Just do a putnext, and let keysock deal with flow control. */
	putnext(pfkey_q, answer);
	return (0);
}

/*
 * Common function to allocate and prepare a keysock_out_t M_CTL message.
 */
mblk_t *
sadb_keysock_out(minor_t serial)
{
	mblk_t *mp;
	keysock_out_t *kso;

	mp = allocb(sizeof (ipsec_info_t), BPRI_HI);
	if (mp != NULL) {
		mp->b_datap->db_type = M_CTL;
		mp->b_wptr += sizeof (ipsec_info_t);
		kso = (keysock_out_t *)mp->b_rptr;
		kso->ks_out_type = KEYSOCK_OUT;
		kso->ks_out_len = sizeof (*kso);
		kso->ks_out_serial = serial;
	}

	return (mp);
}

/*
 * Perform an SADB_DUMP, spewing out every SA in an array of SA fanouts
 * to keysock.
 */
static int
sadb_dump_fanout(queue_t *pfkey_q, mblk_t *mp, minor_t serial, isaf_t *fanout,
    int num_entries, boolean_t do_peers, time_t active_time)
{
	int i, error = 0;
	mblk_t *original_answer;
	ipsa_t *walker;
	sadb_msg_t *samsg;
	time_t	current;

	/*
	 * For each IPSA hash bucket do:
	 *	- Hold the mutex
	 *	- Walk each entry, doing an sadb_dump_deliver() on it.
	 */
	ASSERT(mp->b_cont != NULL);
	samsg = (sadb_msg_t *)mp->b_cont->b_rptr;

	original_answer = sadb_keysock_out(serial);
	if (original_answer == NULL)
		return (ENOMEM);

	current = gethrestime_sec();
	for (i = 0; i < num_entries; i++) {
		mutex_enter(&fanout[i].isaf_lock);
		for (walker = fanout[i].isaf_ipsa; walker != NULL;
		    walker = walker->ipsa_next) {
			if (!do_peers && walker->ipsa_haspeer)
				continue;
			if ((active_time != 0) &&
			    ((current - walker->ipsa_lastuse) > active_time))
				continue;
			error = sadb_dump_deliver(pfkey_q, original_answer,
			    walker, samsg);
			if (error == ENOBUFS) {
				mblk_t *new_original_answer;

				/* Ran out of dupb's.  Try a copyb. */
				new_original_answer = copyb(original_answer);
				if (new_original_answer == NULL) {
					error = ENOMEM;
				} else {
					freeb(original_answer);
					original_answer = new_original_answer;
					error = sadb_dump_deliver(pfkey_q,
					    original_answer, walker, samsg);
				}
			}
			if (error != 0)
				break;	/* out of for loop. */
		}
		mutex_exit(&fanout[i].isaf_lock);
		if (error != 0)
			break;	/* out of for loop. */
	}

	freeb(original_answer);
	return (error);
}

/*
 * Dump an entire SADB; outbound first, then inbound.
 */

int
sadb_dump(queue_t *pfkey_q, mblk_t *mp, keysock_in_t *ksi, sadb_t *sp)
{
	int error;
	time_t	active_time = 0;
	sadb_x_edump_t	*edump =
	    (sadb_x_edump_t *)ksi->ks_in_extv[SADB_X_EXT_EDUMP];

	if (edump != NULL) {
		active_time = edump->sadb_x_edump_timeout;
	}

	/* Dump outbound */
	error = sadb_dump_fanout(pfkey_q, mp, ksi->ks_in_serial, sp->sdb_of,
	    sp->sdb_hashsize, B_TRUE, active_time);
	if (error)
		return (error);

	/* Dump inbound */
	return sadb_dump_fanout(pfkey_q, mp, ksi->ks_in_serial, sp->sdb_if,
	    sp->sdb_hashsize, B_FALSE, active_time);
}

/*
 * Generic sadb table walker.
 *
 * Call "walkfn" for each SA in each bucket in "table"; pass the
 * bucket, the entry and "cookie" to the callback function.
 * Take care to ensure that walkfn can delete the SA without screwing
 * up our traverse.
 *
 * The bucket is locked for the duration of the callback, both so that the
 * callback can just call sadb_unlinkassoc() when it wants to delete something,
 * and so that no new entries are added while we're walking the list.
 */
static void
sadb_walker(isaf_t *table, uint_t numentries,
    void (*walkfn)(isaf_t *head, ipsa_t *entry, void *cookie),
    void *cookie)
{
	int i;
	for (i = 0; i < numentries; i++) {
		ipsa_t *entry, *next;

		mutex_enter(&table[i].isaf_lock);

		for (entry = table[i].isaf_ipsa; entry != NULL;
		    entry = next) {
			next = entry->ipsa_next;
			(*walkfn)(&table[i], entry, cookie);
		}
		mutex_exit(&table[i].isaf_lock);
	}
}

/*
 * From the given SA, construct a dl_ct_ipsec_key and
 * a dl_ct_ipsec structures to be sent to the adapter as part
 * of a DL_CONTROL_REQ.
 *
 * ct_sa must point to the storage allocated for the key
 * structure and must be followed by storage allocated
 * for the SA information that must be sent to the driver
 * as part of the DL_CONTROL_REQ request.
 *
 * The is_inbound boolean indicates whether the specified
 * SA is part of an inbound SA table.
 *
 * Returns B_TRUE if the corresponding SA must be passed to
 * a provider, B_FALSE otherwise; frees *mp if it returns B_FALSE.
 */
static boolean_t
sadb_req_from_sa(ipsa_t *sa, mblk_t *mp, boolean_t is_inbound)
{
	dl_ct_ipsec_key_t *keyp;
	dl_ct_ipsec_t *sap;
	void *ct_sa = mp->b_wptr;

	ASSERT(MUTEX_HELD(&sa->ipsa_lock));

	keyp = (dl_ct_ipsec_key_t *)(ct_sa);
	sap = (dl_ct_ipsec_t *)(keyp + 1);

	IPSECHW_DEBUG(IPSECHW_CAPAB, ("sadb_req_from_sa: "
	    "is_inbound = %d\n", is_inbound));

	/* initialize flag */
	sap->sadb_sa_flags = 0;
	if (is_inbound) {
		sap->sadb_sa_flags |= DL_CT_IPSEC_INBOUND;
		/*
		 * If an inbound SA has a peer, then mark it has being
		 * an outbound SA as well.
		 */
		if (sa->ipsa_haspeer)
			sap->sadb_sa_flags |= DL_CT_IPSEC_OUTBOUND;
	} else {
		/*
		 * If an outbound SA has a peer, then don't send it,
		 * since we will send the copy from the inbound table.
		 */
		if (sa->ipsa_haspeer) {
			freemsg(mp);
			return (B_FALSE);
		}
		sap->sadb_sa_flags |= DL_CT_IPSEC_OUTBOUND;
	}

	keyp->dl_key_spi = sa->ipsa_spi;
	bcopy(sa->ipsa_dstaddr, keyp->dl_key_dest_addr,
	    DL_CTL_IPSEC_ADDR_LEN);
	keyp->dl_key_addr_family = sa->ipsa_addrfam;

	sap->sadb_sa_auth = sa->ipsa_auth_alg;
	sap->sadb_sa_encrypt = sa->ipsa_encr_alg;

	sap->sadb_key_len_a = sa->ipsa_authkeylen;
	sap->sadb_key_bits_a = sa->ipsa_authkeybits;
	bcopy(sa->ipsa_authkey,
	    sap->sadb_key_data_a, sap->sadb_key_len_a);

	sap->sadb_key_len_e = sa->ipsa_encrkeylen;
	sap->sadb_key_bits_e = sa->ipsa_encrkeybits;
	bcopy(sa->ipsa_encrkey,
	    sap->sadb_key_data_e, sap->sadb_key_len_e);

	mp->b_wptr += sizeof (dl_ct_ipsec_t) + sizeof (dl_ct_ipsec_key_t);
	return (B_TRUE);
}

/*
 * Called from AH or ESP to format a message which will be used to inform
 * IPsec-acceleration-capable ills of a SADB change.
 * (It is not possible to send the message to IP directly from this function
 * since the SA, if any, is locked during the call).
 *
 * dl_operation: DL_CONTROL_REQ operation (add, delete, update, etc)
 * sa_type: identifies whether the operation applies to AH or ESP
 *	(must be one of SADB_SATYPE_AH or SADB_SATYPE_ESP)
 * sa: Pointer to an SA.  Must be non-NULL and locked
 *	for ADD, DELETE, GET, and UPDATE operations.
 * This function returns an mblk chain that must be passed to IP
 * for forwarding to the IPsec capable providers.
 */
mblk_t *
sadb_fmt_sa_req(uint_t dl_operation, uint_t sa_type, ipsa_t *sa,
    boolean_t is_inbound)
{
	mblk_t *mp;
	dl_control_req_t *ctrl;
	boolean_t need_key = B_FALSE;
	mblk_t *ctl_mp = NULL;
	ipsec_ctl_t *ctl;

	/*
	 * 1 allocate and initialize DL_CONTROL_REQ M_PROTO
	 * 2 if a key is needed for the operation
	 *    2.1 initialize key
	 *    2.2 if a full SA is needed for the operation
	 *	2.2.1 initialize full SA info
	 * 3 return message; caller will call ill_ipsec_capab_send_all()
	 * to send the resulting message to IPsec capable ills.
	 */

	ASSERT(sa_type == SADB_SATYPE_AH || sa_type == SADB_SATYPE_ESP);

	/*
	 * Allocate DL_CONTROL_REQ M_PROTO
	 * We allocate room for the SA even if it's not needed
	 * by some of the operations (for example flush)
	 */
	mp = allocb(sizeof (dl_control_req_t) +
	    sizeof (dl_ct_ipsec_key_t) + sizeof (dl_ct_ipsec_t), BPRI_HI);
	if (mp == NULL)
		return (NULL);
	mp->b_datap->db_type = M_PROTO;

	/* initialize dl_control_req_t */
	ctrl = (dl_control_req_t *)mp->b_wptr;
	ctrl->dl_primitive = DL_CONTROL_REQ;
	ctrl->dl_operation = dl_operation;
	ctrl->dl_type = sa_type == SADB_SATYPE_AH ? DL_CT_IPSEC_AH :
	    DL_CT_IPSEC_ESP;
	ctrl->dl_key_offset = sizeof (dl_control_req_t);
	ctrl->dl_key_length = sizeof (dl_ct_ipsec_key_t);
	ctrl->dl_data_offset = sizeof (dl_control_req_t) +
	    sizeof (dl_ct_ipsec_key_t);
	ctrl->dl_data_length = sizeof (dl_ct_ipsec_t);
	mp->b_wptr += sizeof (dl_control_req_t);

	if ((dl_operation == DL_CO_SET) || (dl_operation == DL_CO_DELETE)) {
		ASSERT(sa != NULL);
		ASSERT(MUTEX_HELD(&sa->ipsa_lock));

		need_key = B_TRUE;

		/*
		 * Initialize key and SA data. Note that for some
		 * operations the SA data is ignored by the provider
		 * (delete, etc.)
		 */
		if (!sadb_req_from_sa(sa, mp, is_inbound))
			return (NULL);
	}

	/* construct control message */
	ctl_mp = allocb(sizeof (ipsec_ctl_t), BPRI_HI);
	if (ctl_mp == NULL) {
		cmn_err(CE_WARN, "sadb_fmt_sa_req: allocb failed\n");
		freemsg(mp);
		return (NULL);
	}

	ctl_mp->b_datap->db_type = M_CTL;
	ctl_mp->b_wptr += sizeof (ipsec_ctl_t);
	ctl_mp->b_cont = mp;

	ctl = (ipsec_ctl_t *)ctl_mp->b_rptr;
	ctl->ipsec_ctl_type = IPSEC_CTL;
	ctl->ipsec_ctl_len  = sizeof (ipsec_ctl_t);
	ctl->ipsec_ctl_sa_type = sa_type;

	if (need_key) {
		/*
		 * Keep an additional reference on SA, since it will be
		 * needed by IP to send control messages corresponding
		 * to that SA from its perimeter. IP will do a
		 * IPSA_REFRELE when done with the request.
		 */
		ASSERT(MUTEX_HELD(&sa->ipsa_lock));
		IPSA_REFHOLD(sa);
		ctl->ipsec_ctl_sa = sa;
	} else
		ctl->ipsec_ctl_sa = NULL;

	return (ctl_mp);
}


/*
 * Called by sadb_ill_download() to dump the entries for a specific
 * fanout table.  For each SA entry in the table passed as argument,
 * use mp as a template and constructs a full DL_CONTROL message, and
 * call ill_dlpi_send(), provided by IP, to send the resulting
 * messages to the ill.
 */
static void
sadb_ill_df(ill_t *ill, mblk_t *mp, isaf_t *fanout, int num_entries,
    boolean_t is_inbound)
{
	ipsa_t *walker;
	mblk_t *nmp, *salist;
	int i, error = 0;
	ip_stack_t	*ipst = ill->ill_ipst;
	netstack_t	*ns = ipst->ips_netstack;

	IPSECHW_DEBUG(IPSECHW_SADB, ("sadb_ill_df: fanout at 0x%p ne=%d\n",
	    (void *)fanout, num_entries));
	/*
	 * For each IPSA hash bucket do:
	 *	- Hold the mutex
	 *	- Walk each entry, sending a corresponding request to IP
	 *	  for it.
	 */
	ASSERT(mp->b_datap->db_type == M_PROTO);

	for (i = 0; i < num_entries; i++) {
		mutex_enter(&fanout[i].isaf_lock);
		salist = NULL;

		for (walker = fanout[i].isaf_ipsa; walker != NULL;
		    walker = walker->ipsa_next) {
			IPSECHW_DEBUG(IPSECHW_SADB,
			    ("sadb_ill_df: sending SA to ill via IP \n"));
			/*
			 * Duplicate the template mp passed and
			 * complete DL_CONTROL_REQ data.
			 * To be more memory efficient, we could use
			 * dupb() for the M_CTL and copyb() for the M_PROTO
			 * as the M_CTL, since the M_CTL is the same for
			 * every SA entry passed down to IP for the same ill.
			 *
			 * Note that copymsg/copyb ensure that the new mblk
			 * is at least as large as the source mblk even if it's
			 * not using all its storage -- therefore, nmp
			 * has trailing space for sadb_req_from_sa to add
			 * the SA-specific bits.
			 */
			mutex_enter(&walker->ipsa_lock);
			if (ipsec_capab_match(ill,
			    ill->ill_phyint->phyint_ifindex, ill->ill_isv6,
			    walker, ns)) {
				nmp = copymsg(mp);
				if (nmp == NULL) {
					IPSECHW_DEBUG(IPSECHW_SADB,
					    ("sadb_ill_df: alloc error\n"));
					error = ENOMEM;
					mutex_exit(&walker->ipsa_lock);
					break;
				}
				if (sadb_req_from_sa(walker, nmp, is_inbound)) {
					nmp->b_next = salist;
					salist = nmp;
				}
			}
			mutex_exit(&walker->ipsa_lock);
		}
		mutex_exit(&fanout[i].isaf_lock);
		while (salist != NULL) {
			nmp = salist;
			salist = nmp->b_next;
			nmp->b_next = NULL;
			ill_dlpi_send(ill, nmp);
		}
		if (error != 0)
			break;	/* out of for loop. */
	}
}

/*
 * Called by ill_ipsec_capab_add(). Sends a copy of the SADB of
 * the type specified by sa_type to the specified ill.
 *
 * We call for each fanout table defined by the SADB (one per
 * protocol). sadb_ill_df() finally calls ill_dlpi_send() for
 * each SADB entry in order to send a corresponding DL_CONTROL_REQ
 * message to the ill.
 */
void
sadb_ill_download(ill_t *ill, uint_t sa_type)
{
	mblk_t *protomp;	/* prototype message */
	dl_control_req_t *ctrl;
	sadbp_t *spp;
	sadb_t *sp;
	int dlt;
	ip_stack_t	*ipst = ill->ill_ipst;
	netstack_t	*ns = ipst->ips_netstack;

	ASSERT(sa_type == SADB_SATYPE_AH || sa_type == SADB_SATYPE_ESP);

	/*
	 * Allocate and initialize prototype answer. A duplicate for
	 * each SA is sent down to the interface.
	 */

	/* DL_CONTROL_REQ M_PROTO mblk_t */
	protomp = allocb(sizeof (dl_control_req_t) +
	    sizeof (dl_ct_ipsec_key_t) + sizeof (dl_ct_ipsec_t), BPRI_HI);
	if (protomp == NULL)
		return;
	protomp->b_datap->db_type = M_PROTO;

	dlt = (sa_type == SADB_SATYPE_AH) ? DL_CT_IPSEC_AH : DL_CT_IPSEC_ESP;
	if (sa_type == SADB_SATYPE_ESP) {
		ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;

		spp = &espstack->esp_sadb;
	} else {
		ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;

		spp = &ahstack->ah_sadb;
	}

	ctrl = (dl_control_req_t *)protomp->b_wptr;
	ctrl->dl_primitive = DL_CONTROL_REQ;
	ctrl->dl_operation = DL_CO_SET;
	ctrl->dl_type = dlt;
	ctrl->dl_key_offset = sizeof (dl_control_req_t);
	ctrl->dl_key_length = sizeof (dl_ct_ipsec_key_t);
	ctrl->dl_data_offset = sizeof (dl_control_req_t) +
	    sizeof (dl_ct_ipsec_key_t);
	ctrl->dl_data_length = sizeof (dl_ct_ipsec_t);
	protomp->b_wptr += sizeof (dl_control_req_t);

	/*
	 * then for each SADB entry, we fill out the dl_ct_ipsec_key_t
	 * and dl_ct_ipsec_t
	 */
	sp = ill->ill_isv6 ? &(spp->s_v6) : &(spp->s_v4);
	sadb_ill_df(ill, protomp, sp->sdb_of, sp->sdb_hashsize, B_FALSE);
	sadb_ill_df(ill, protomp, sp->sdb_if, sp->sdb_hashsize, B_TRUE);
	freemsg(protomp);
}

/*
 * Call me to free up a security association fanout.  Use the forever
 * variable to indicate freeing up the SAs (forever == B_FALSE, e.g.
 * an SADB_FLUSH message), or destroying everything (forever == B_TRUE,
 * when a module is unloaded).
 */
static void
sadb_destroyer(isaf_t **tablep, uint_t numentries, boolean_t forever,
    boolean_t inbound)
{
	int i;
	isaf_t *table = *tablep;
	uint8_t protocol;
	ipsa_t *sa;
	netstackid_t sid;

	if (table == NULL)
		return;

	for (i = 0; i < numentries; i++) {
		mutex_enter(&table[i].isaf_lock);
		while ((sa = table[i].isaf_ipsa) != NULL) {
			if (inbound && cl_inet_deletespi &&
			    (sa->ipsa_state != IPSA_STATE_ACTIVE_ELSEWHERE) &&
			    (sa->ipsa_state != IPSA_STATE_IDLE)) {
				protocol = (sa->ipsa_type == SADB_SATYPE_AH) ?
				    IPPROTO_AH : IPPROTO_ESP;
				sid = sa->ipsa_netstack->netstack_stackid;
				cl_inet_deletespi(sid, protocol, sa->ipsa_spi,
				    NULL);
			}
			sadb_unlinkassoc(sa);
		}
		table[i].isaf_gen++;
		mutex_exit(&table[i].isaf_lock);
		if (forever)
			mutex_destroy(&(table[i].isaf_lock));
	}

	if (forever) {
		*tablep = NULL;
		kmem_free(table, numentries * sizeof (*table));
	}
}

/*
 * Entry points to sadb_destroyer().
 */
static void
sadb_flush(sadb_t *sp, netstack_t *ns)
{
	/*
	 * Flush out each bucket, one at a time.  Were it not for keysock's
	 * enforcement, there would be a subtlety where I could add on the
	 * heels of a flush.  With keysock's enforcement, however, this
	 * makes ESP's job easy.
	 */
	sadb_destroyer(&sp->sdb_of, sp->sdb_hashsize, B_FALSE, B_FALSE);
	sadb_destroyer(&sp->sdb_if, sp->sdb_hashsize, B_FALSE, B_TRUE);

	/* For each acquire, destroy it; leave the bucket mutex alone. */
	sadb_destroy_acqlist(&sp->sdb_acq, sp->sdb_hashsize, B_FALSE, ns);
}

static void
sadb_destroy(sadb_t *sp, netstack_t *ns)
{
	sadb_destroyer(&sp->sdb_of, sp->sdb_hashsize, B_TRUE, B_FALSE);
	sadb_destroyer(&sp->sdb_if, sp->sdb_hashsize, B_TRUE, B_TRUE);

	/* For each acquire, destroy it, including the bucket mutex. */
	sadb_destroy_acqlist(&sp->sdb_acq, sp->sdb_hashsize, B_TRUE, ns);

	ASSERT(sp->sdb_of == NULL);
	ASSERT(sp->sdb_if == NULL);
	ASSERT(sp->sdb_acq == NULL);
}

static void
sadb_send_flush_req(sadbp_t *spp)
{
	mblk_t *ctl_mp;

	/*
	 * we've been unplumbed, or never were plumbed; don't go there.
	 */
	if (spp->s_ip_q == NULL)
		return;

	/* have IP send a flush msg to the IPsec accelerators */
	ctl_mp = sadb_fmt_sa_req(DL_CO_FLUSH, spp->s_satype, NULL, B_TRUE);
	if (ctl_mp != NULL)
		putnext(spp->s_ip_q, ctl_mp);
}

void
sadbp_flush(sadbp_t *spp, netstack_t *ns)
{
	sadb_flush(&spp->s_v4, ns);
	sadb_flush(&spp->s_v6, ns);

	sadb_send_flush_req(spp);
}

void
sadbp_destroy(sadbp_t *spp, netstack_t *ns)
{
	sadb_destroy(&spp->s_v4, ns);
	sadb_destroy(&spp->s_v6, ns);

	sadb_send_flush_req(spp);
	if (spp->s_satype == SADB_SATYPE_AH) {
		ipsec_stack_t	*ipss = ns->netstack_ipsec;

		ip_drop_unregister(&ipss->ipsec_sadb_dropper);
	}
}


/*
 * Check hard vs. soft lifetimes.  If there's a reality mismatch (e.g.
 * soft lifetimes > hard lifetimes) return an appropriate diagnostic for
 * EINVAL.
 */
int
sadb_hardsoftchk(sadb_lifetime_t *hard, sadb_lifetime_t *soft,
    sadb_lifetime_t *idle)
{
	if (hard == NULL || soft == NULL)
		return (0);

	if (hard->sadb_lifetime_allocations != 0 &&
	    soft->sadb_lifetime_allocations != 0 &&
	    hard->sadb_lifetime_allocations < soft->sadb_lifetime_allocations)
		return (SADB_X_DIAGNOSTIC_ALLOC_HSERR);

	if (hard->sadb_lifetime_bytes != 0 &&
	    soft->sadb_lifetime_bytes != 0 &&
	    hard->sadb_lifetime_bytes < soft->sadb_lifetime_bytes)
		return (SADB_X_DIAGNOSTIC_BYTES_HSERR);

	if (hard->sadb_lifetime_addtime != 0 &&
	    soft->sadb_lifetime_addtime != 0 &&
	    hard->sadb_lifetime_addtime < soft->sadb_lifetime_addtime)
		return (SADB_X_DIAGNOSTIC_ADDTIME_HSERR);

	if (hard->sadb_lifetime_usetime != 0 &&
	    soft->sadb_lifetime_usetime != 0 &&
	    hard->sadb_lifetime_usetime < soft->sadb_lifetime_usetime)
		return (SADB_X_DIAGNOSTIC_USETIME_HSERR);

	if (idle != NULL) {
		if (hard->sadb_lifetime_addtime != 0 &&
		    idle->sadb_lifetime_addtime != 0 &&
		    hard->sadb_lifetime_addtime < idle->sadb_lifetime_addtime)
			return (SADB_X_DIAGNOSTIC_ADDTIME_HSERR);

		if (soft->sadb_lifetime_addtime != 0 &&
		    idle->sadb_lifetime_addtime != 0 &&
		    soft->sadb_lifetime_addtime < idle->sadb_lifetime_addtime)
			return (SADB_X_DIAGNOSTIC_ADDTIME_HSERR);

		if (hard->sadb_lifetime_usetime != 0 &&
		    idle->sadb_lifetime_usetime != 0 &&
		    hard->sadb_lifetime_usetime < idle->sadb_lifetime_usetime)
			return (SADB_X_DIAGNOSTIC_USETIME_HSERR);

		if (soft->sadb_lifetime_usetime != 0 &&
		    idle->sadb_lifetime_usetime != 0 &&
		    soft->sadb_lifetime_usetime < idle->sadb_lifetime_usetime)
			return (SADB_X_DIAGNOSTIC_USETIME_HSERR);
	}

	return (0);
}

/*
 * Clone a security association for the purposes of inserting a single SA
 * into inbound and outbound tables respectively. This function should only
 * be called from sadb_common_add().
 */
static ipsa_t *
sadb_cloneassoc(ipsa_t *ipsa)
{
	ipsa_t *newbie;
	boolean_t error = B_FALSE;

	ASSERT(MUTEX_NOT_HELD(&(ipsa->ipsa_lock)));

	newbie = kmem_alloc(sizeof (ipsa_t), KM_NOSLEEP);
	if (newbie == NULL)
		return (NULL);

	/* Copy over what we can. */
	*newbie = *ipsa;

	/* bzero and initialize locks, in case *_init() allocates... */
	mutex_init(&newbie->ipsa_lock, NULL, MUTEX_DEFAULT, NULL);

	/*
	 * While somewhat dain-bramaged, the most graceful way to
	 * recover from errors is to keep plowing through the
	 * allocations, and getting what I can.  It's easier to call
	 * sadb_freeassoc() on the stillborn clone when all the
	 * pointers aren't pointing to the parent's data.
	 */

	if (ipsa->ipsa_authkey != NULL) {
		newbie->ipsa_authkey = kmem_alloc(newbie->ipsa_authkeylen,
		    KM_NOSLEEP);
		if (newbie->ipsa_authkey == NULL) {
			error = B_TRUE;
		} else {
			bcopy(ipsa->ipsa_authkey, newbie->ipsa_authkey,
			    newbie->ipsa_authkeylen);

			newbie->ipsa_kcfauthkey.ck_data =
			    newbie->ipsa_authkey;
		}

		if (newbie->ipsa_amech.cm_param != NULL) {
			newbie->ipsa_amech.cm_param =
			    (char *)&newbie->ipsa_mac_len;
		}
	}

	if (ipsa->ipsa_encrkey != NULL) {
		newbie->ipsa_encrkey = kmem_alloc(newbie->ipsa_encrkeylen,
		    KM_NOSLEEP);
		if (newbie->ipsa_encrkey == NULL) {
			error = B_TRUE;
		} else {
			bcopy(ipsa->ipsa_encrkey, newbie->ipsa_encrkey,
			    newbie->ipsa_encrkeylen);

			newbie->ipsa_kcfencrkey.ck_data =
			    newbie->ipsa_encrkey;
		}
	}

	newbie->ipsa_authtmpl = NULL;
	newbie->ipsa_encrtmpl = NULL;
	newbie->ipsa_haspeer = B_TRUE;

	if (ipsa->ipsa_integ != NULL) {
		newbie->ipsa_integ = kmem_alloc(newbie->ipsa_integlen,
		    KM_NOSLEEP);
		if (newbie->ipsa_integ == NULL) {
			error = B_TRUE;
		} else {
			bcopy(ipsa->ipsa_integ, newbie->ipsa_integ,
			    newbie->ipsa_integlen);
		}
	}

	if (ipsa->ipsa_sens != NULL) {
		newbie->ipsa_sens = kmem_alloc(newbie->ipsa_senslen,
		    KM_NOSLEEP);
		if (newbie->ipsa_sens == NULL) {
			error = B_TRUE;
		} else {
			bcopy(ipsa->ipsa_sens, newbie->ipsa_sens,
			    newbie->ipsa_senslen);
		}
	}

	if (ipsa->ipsa_src_cid != NULL) {
		newbie->ipsa_src_cid = ipsa->ipsa_src_cid;
		IPSID_REFHOLD(ipsa->ipsa_src_cid);
	}

	if (ipsa->ipsa_dst_cid != NULL) {
		newbie->ipsa_dst_cid = ipsa->ipsa_dst_cid;
		IPSID_REFHOLD(ipsa->ipsa_dst_cid);
	}

	if (error) {
		sadb_freeassoc(newbie);
		return (NULL);
	}

	return (newbie);
}

/*
 * Initialize a SADB address extension at the address specified by addrext.
 * Return a pointer to the end of the new address extension.
 */
static uint8_t *
sadb_make_addr_ext(uint8_t *start, uint8_t *end, uint16_t exttype,
    sa_family_t af, uint32_t *addr, uint16_t port, uint8_t proto, int prefix)
{
	struct sockaddr_in *sin;
	struct sockaddr_in6 *sin6;
	uint8_t *cur = start;
	int addrext_len;
	int sin_len;
	sadb_address_t *addrext	= (sadb_address_t *)cur;

	if (cur == NULL)
		return (NULL);

	cur += sizeof (*addrext);
	if (cur > end)
		return (NULL);

	addrext->sadb_address_proto = proto;
	addrext->sadb_address_prefixlen = prefix;
	addrext->sadb_address_reserved = 0;
	addrext->sadb_address_exttype = exttype;

	switch (af) {
	case AF_INET:
		sin = (struct sockaddr_in *)cur;
		sin_len = sizeof (*sin);
		cur += sin_len;
		if (cur > end)
			return (NULL);

		sin->sin_family = af;
		bzero(sin->sin_zero, sizeof (sin->sin_zero));
		sin->sin_port = port;
		IPSA_COPY_ADDR(&sin->sin_addr, addr, af);
		break;
	case AF_INET6:
		sin6 = (struct sockaddr_in6 *)cur;
		sin_len = sizeof (*sin6);
		cur += sin_len;
		if (cur > end)
			return (NULL);

		bzero(sin6, sizeof (*sin6));
		sin6->sin6_family = af;
		sin6->sin6_port = port;
		IPSA_COPY_ADDR(&sin6->sin6_addr, addr, af);
		break;
	}

	addrext_len = roundup(cur - start, sizeof (uint64_t));
	addrext->sadb_address_len = SADB_8TO64(addrext_len);

	cur = start + addrext_len;
	if (cur > end)
		cur = NULL;

	return (cur);
}

/*
 * Construct a key management cookie extension.
 */

static uint8_t *
sadb_make_kmc_ext(uint8_t *cur, uint8_t *end, uint32_t kmp, uint32_t kmc)
{
	sadb_x_kmc_t *kmcext = (sadb_x_kmc_t *)cur;

	if (cur == NULL)
		return (NULL);

	cur += sizeof (*kmcext);

	if (cur > end)
		return (NULL);

	kmcext->sadb_x_kmc_len = SADB_8TO64(sizeof (*kmcext));
	kmcext->sadb_x_kmc_exttype = SADB_X_EXT_KM_COOKIE;
	kmcext->sadb_x_kmc_proto = kmp;
	kmcext->sadb_x_kmc_cookie = kmc;
	kmcext->sadb_x_kmc_reserved = 0;

	return (cur);
}

/*
 * Given an original message header with sufficient space following it, and an
 * SA, construct a full PF_KEY message with all of the relevant extensions.
 * This is mostly used for SADB_GET, and SADB_DUMP.
 */
static mblk_t *
sadb_sa2msg(ipsa_t *ipsa, sadb_msg_t *samsg)
{
	int alloclen, addrsize, paddrsize, authsize, encrsize;
	int srcidsize, dstidsize;
	sa_family_t fam, pfam;	/* Address family for SADB_EXT_ADDRESS */
				/* src/dst and proxy sockaddrs. */
	/*
	 * The following are pointers into the PF_KEY message this PF_KEY
	 * message creates.
	 */
	sadb_msg_t *newsamsg;
	sadb_sa_t *assoc;
	sadb_lifetime_t *lt;
	sadb_key_t *key;
	sadb_ident_t *ident;
	sadb_sens_t *sens;
	sadb_ext_t *walker;	/* For when we need a generic ext. pointer. */
	sadb_x_replay_ctr_t *repl_ctr;
	sadb_x_pair_t *pair_ext;

	mblk_t *mp;
	uint64_t *bitmap;
	uint8_t *cur, *end;
	/* These indicate the presence of the above extension fields. */
	boolean_t soft, hard, isrc, idst, auth, encr, sensinteg, srcid, dstid;
	boolean_t idle;
	boolean_t paired;
	uint32_t otherspi;

	/* First off, figure out the allocation length for this message. */

	/*
	 * Constant stuff.  This includes base, SA, address (src, dst),
	 * and lifetime (current).
	 */
	alloclen = sizeof (sadb_msg_t) + sizeof (sadb_sa_t) +
	    sizeof (sadb_lifetime_t);

	fam = ipsa->ipsa_addrfam;
	switch (fam) {
	case AF_INET:
		addrsize = roundup(sizeof (struct sockaddr_in) +
		    sizeof (sadb_address_t), sizeof (uint64_t));
		break;
	case AF_INET6:
		addrsize = roundup(sizeof (struct sockaddr_in6) +
		    sizeof (sadb_address_t), sizeof (uint64_t));
		break;
	default:
		return (NULL);
	}
	/*
	 * Allocate TWO address extensions, for source and destination.
	 * (Thus, the * 2.)
	 */
	alloclen += addrsize * 2;
	if (ipsa->ipsa_flags & IPSA_F_NATT_REM)
		alloclen += addrsize;
	if (ipsa->ipsa_flags & IPSA_F_NATT_LOC)
		alloclen += addrsize;

	if (ipsa->ipsa_flags & IPSA_F_PAIRED) {
		paired = B_TRUE;
		alloclen += sizeof (sadb_x_pair_t);
		otherspi = ipsa->ipsa_otherspi;
	} else {
		paired = B_FALSE;
	}

	/* How 'bout other lifetimes? */
	if (ipsa->ipsa_softaddlt != 0 || ipsa->ipsa_softuselt != 0 ||
	    ipsa->ipsa_softbyteslt != 0 || ipsa->ipsa_softalloc != 0) {
		alloclen += sizeof (sadb_lifetime_t);
		soft = B_TRUE;
	} else {
		soft = B_FALSE;
	}

	if (ipsa->ipsa_hardaddlt != 0 || ipsa->ipsa_harduselt != 0 ||
	    ipsa->ipsa_hardbyteslt != 0 || ipsa->ipsa_hardalloc != 0) {
		alloclen += sizeof (sadb_lifetime_t);
		hard = B_TRUE;
	} else {
		hard = B_FALSE;
	}

	if (ipsa->ipsa_idleaddlt != 0 || ipsa->ipsa_idleuselt != 0) {
		alloclen += sizeof (sadb_lifetime_t);
		idle = B_TRUE;
	} else {
		idle = B_FALSE;
	}

	/* Inner addresses. */
	if (ipsa->ipsa_innerfam == 0) {
		isrc = B_FALSE;
		idst = B_FALSE;
	} else {
		pfam = ipsa->ipsa_innerfam;
		switch (pfam) {
		case AF_INET6:
			paddrsize = roundup(sizeof (struct sockaddr_in6) +
			    sizeof (sadb_address_t), sizeof (uint64_t));
			break;
		case AF_INET:
			paddrsize = roundup(sizeof (struct sockaddr_in) +
			    sizeof (sadb_address_t), sizeof (uint64_t));
			break;
		default:
			cmn_err(CE_PANIC,
			    "IPsec SADB: Proxy length failure.\n");
			break;
		}
		isrc = B_TRUE;
		idst = B_TRUE;
		alloclen += 2 * paddrsize;
	}

	/* For the following fields, assume that length != 0 ==> stuff */
	if (ipsa->ipsa_authkeylen != 0) {
		authsize = roundup(sizeof (sadb_key_t) + ipsa->ipsa_authkeylen,
		    sizeof (uint64_t));
		alloclen += authsize;
		auth = B_TRUE;
	} else {
		auth = B_FALSE;
	}

	if (ipsa->ipsa_encrkeylen != 0) {
		encrsize = roundup(sizeof (sadb_key_t) + ipsa->ipsa_encrkeylen,
		    sizeof (uint64_t));
		alloclen += encrsize;
		encr = B_TRUE;
	} else {
		encr = B_FALSE;
	}

	/* No need for roundup on sens and integ. */
	if (ipsa->ipsa_integlen != 0 || ipsa->ipsa_senslen != 0) {
		alloclen += sizeof (sadb_key_t) + ipsa->ipsa_integlen +
		    ipsa->ipsa_senslen;
		sensinteg = B_TRUE;
	} else {
		sensinteg = B_FALSE;
	}

	/*
	 * Must use strlen() here for lengths.	Identities use NULL
	 * pointers to indicate their nonexistence.
	 */
	if (ipsa->ipsa_src_cid != NULL) {
		srcidsize = roundup(sizeof (sadb_ident_t) +
		    strlen(ipsa->ipsa_src_cid->ipsid_cid) + 1,
		    sizeof (uint64_t));
		alloclen += srcidsize;
		srcid = B_TRUE;
	} else {
		srcid = B_FALSE;
	}

	if (ipsa->ipsa_dst_cid != NULL) {
		dstidsize = roundup(sizeof (sadb_ident_t) +
		    strlen(ipsa->ipsa_dst_cid->ipsid_cid) + 1,
		    sizeof (uint64_t));
		alloclen += dstidsize;
		dstid = B_TRUE;
	} else {
		dstid = B_FALSE;
	}

	if ((ipsa->ipsa_kmp != 0) || (ipsa->ipsa_kmc != 0))
		alloclen += sizeof (sadb_x_kmc_t);

	if (ipsa->ipsa_replay != 0) {
		alloclen += sizeof (sadb_x_replay_ctr_t);
	}

	/* Make sure the allocation length is a multiple of 8 bytes. */
	ASSERT((alloclen & 0x7) == 0);

	/* XXX Possibly make it esballoc, with a bzero-ing free_ftn. */
	mp = allocb(alloclen, BPRI_HI);
	if (mp == NULL)
		return (NULL);

	mp->b_wptr += alloclen;
	end = mp->b_wptr;
	newsamsg = (sadb_msg_t *)mp->b_rptr;
	*newsamsg = *samsg;
	newsamsg->sadb_msg_len = (uint16_t)SADB_8TO64(alloclen);

	mutex_enter(&ipsa->ipsa_lock);	/* Since I'm grabbing SA fields... */

	newsamsg->sadb_msg_satype = ipsa->ipsa_type;

	assoc = (sadb_sa_t *)(newsamsg + 1);
	assoc->sadb_sa_len = SADB_8TO64(sizeof (*assoc));
	assoc->sadb_sa_exttype = SADB_EXT_SA;
	assoc->sadb_sa_spi = ipsa->ipsa_spi;
	assoc->sadb_sa_replay = ipsa->ipsa_replay_wsize;
	assoc->sadb_sa_state = ipsa->ipsa_state;
	assoc->sadb_sa_auth = ipsa->ipsa_auth_alg;
	assoc->sadb_sa_encrypt = ipsa->ipsa_encr_alg;
	assoc->sadb_sa_flags = ipsa->ipsa_flags;

	lt = (sadb_lifetime_t *)(assoc + 1);
	lt->sadb_lifetime_len = SADB_8TO64(sizeof (*lt));
	lt->sadb_lifetime_exttype = SADB_EXT_LIFETIME_CURRENT;
	/* We do not support the concept. */
	lt->sadb_lifetime_allocations = 0;
	lt->sadb_lifetime_bytes = ipsa->ipsa_bytes;
	lt->sadb_lifetime_addtime = ipsa->ipsa_addtime;
	lt->sadb_lifetime_usetime = ipsa->ipsa_usetime;

	if (hard) {
		lt++;
		lt->sadb_lifetime_len = SADB_8TO64(sizeof (*lt));
		lt->sadb_lifetime_exttype = SADB_EXT_LIFETIME_HARD;
		lt->sadb_lifetime_allocations = ipsa->ipsa_hardalloc;
		lt->sadb_lifetime_bytes = ipsa->ipsa_hardbyteslt;
		lt->sadb_lifetime_addtime = ipsa->ipsa_hardaddlt;
		lt->sadb_lifetime_usetime = ipsa->ipsa_harduselt;
	}

	if (soft) {
		lt++;
		lt->sadb_lifetime_len = SADB_8TO64(sizeof (*lt));
		lt->sadb_lifetime_exttype = SADB_EXT_LIFETIME_SOFT;
		lt->sadb_lifetime_allocations = ipsa->ipsa_softalloc;
		lt->sadb_lifetime_bytes = ipsa->ipsa_softbyteslt;
		lt->sadb_lifetime_addtime = ipsa->ipsa_softaddlt;
		lt->sadb_lifetime_usetime = ipsa->ipsa_softuselt;
	}

	if (idle) {
		lt++;
		lt->sadb_lifetime_len = SADB_8TO64(sizeof (*lt));
		lt->sadb_lifetime_exttype = SADB_X_EXT_LIFETIME_IDLE;
		lt->sadb_lifetime_addtime = ipsa->ipsa_idleaddlt;
		lt->sadb_lifetime_usetime = ipsa->ipsa_idleuselt;
	}

	cur = (uint8_t *)(lt + 1);

	/* NOTE:  Don't fill in ports here if we are a tunnel-mode SA. */
	cur = sadb_make_addr_ext(cur, end, SADB_EXT_ADDRESS_SRC, fam,
	    ipsa->ipsa_srcaddr, (!isrc && !idst) ? SA_SRCPORT(ipsa) : 0,
	    SA_PROTO(ipsa), 0);
	if (cur == NULL) {
		freemsg(mp);
		mp = NULL;
		goto bail;
	}

	cur = sadb_make_addr_ext(cur, end, SADB_EXT_ADDRESS_DST, fam,
	    ipsa->ipsa_dstaddr, (!isrc && !idst) ? SA_DSTPORT(ipsa) : 0,
	    SA_PROTO(ipsa), 0);
	if (cur == NULL) {
		freemsg(mp);
		mp = NULL;
		goto bail;
	}

	if (ipsa->ipsa_flags & IPSA_F_NATT_LOC) {
		cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_NATT_LOC,
		    fam, &ipsa->ipsa_natt_addr_loc, ipsa->ipsa_local_nat_port,
		    IPPROTO_UDP, 0);
		if (cur == NULL) {
			freemsg(mp);
			mp = NULL;
			goto bail;
		}
	}

	if (ipsa->ipsa_flags & IPSA_F_NATT_REM) {
		cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_NATT_REM,
		    fam, &ipsa->ipsa_natt_addr_rem, ipsa->ipsa_remote_nat_port,
		    IPPROTO_UDP, 0);
		if (cur == NULL) {
			freemsg(mp);
			mp = NULL;
			goto bail;
		}
	}

	/* If we are a tunnel-mode SA, fill in the inner-selectors. */
	if (isrc) {
		cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_INNER_SRC,
		    pfam, ipsa->ipsa_innersrc, SA_SRCPORT(ipsa),
		    SA_IPROTO(ipsa), ipsa->ipsa_innersrcpfx);
		if (cur == NULL) {
			freemsg(mp);
			mp = NULL;
			goto bail;
		}
	}

	if (idst) {
		cur = sadb_make_addr_ext(cur, end, SADB_X_EXT_ADDRESS_INNER_DST,
		    pfam, ipsa->ipsa_innerdst, SA_DSTPORT(ipsa),
		    SA_IPROTO(ipsa), ipsa->ipsa_innerdstpfx);
		if (cur == NULL) {
			freemsg(mp);
			mp = NULL;
			goto bail;
		}
	}

	if ((ipsa->ipsa_kmp != 0) || (ipsa->ipsa_kmc != 0)) {
		cur = sadb_make_kmc_ext(cur, end,
		    ipsa->ipsa_kmp, ipsa->ipsa_kmc);
		if (cur == NULL) {
			freemsg(mp);
			mp = NULL;
			goto bail;
		}
	}

	walker = (sadb_ext_t *)cur;
	if (auth) {
		key = (sadb_key_t *)walker;
		key->sadb_key_len = SADB_8TO64(authsize);
		key->sadb_key_exttype = SADB_EXT_KEY_AUTH;
		key->sadb_key_bits = ipsa->ipsa_authkeybits;
		key->sadb_key_reserved = 0;
		bcopy(ipsa->ipsa_authkey, key + 1, ipsa->ipsa_authkeylen);
		walker = (sadb_ext_t *)((uint64_t *)walker +
		    walker->sadb_ext_len);
	}

	if (encr) {
		key = (sadb_key_t *)walker;
		key->sadb_key_len = SADB_8TO64(encrsize);
		key->sadb_key_exttype = SADB_EXT_KEY_ENCRYPT;
		key->sadb_key_bits = ipsa->ipsa_encrkeybits;
		key->sadb_key_reserved = 0;
		bcopy(ipsa->ipsa_encrkey, key + 1, ipsa->ipsa_encrkeylen);
		walker = (sadb_ext_t *)((uint64_t *)walker +
		    walker->sadb_ext_len);
	}

	if (srcid) {
		ident = (sadb_ident_t *)walker;
		ident->sadb_ident_len = SADB_8TO64(srcidsize);
		ident->sadb_ident_exttype = SADB_EXT_IDENTITY_SRC;
		ident->sadb_ident_type = ipsa->ipsa_src_cid->ipsid_type;
		ident->sadb_ident_id = 0;
		ident->sadb_ident_reserved = 0;
		(void) strcpy((char *)(ident + 1),
		    ipsa->ipsa_src_cid->ipsid_cid);
		walker = (sadb_ext_t *)((uint64_t *)walker +
		    walker->sadb_ext_len);
	}

	if (dstid) {
		ident = (sadb_ident_t *)walker;
		ident->sadb_ident_len = SADB_8TO64(dstidsize);
		ident->sadb_ident_exttype = SADB_EXT_IDENTITY_DST;
		ident->sadb_ident_type = ipsa->ipsa_dst_cid->ipsid_type;
		ident->sadb_ident_id = 0;
		ident->sadb_ident_reserved = 0;
		(void) strcpy((char *)(ident + 1),
		    ipsa->ipsa_dst_cid->ipsid_cid);
		walker = (sadb_ext_t *)((uint64_t *)walker +
		    walker->sadb_ext_len);
	}

	if (sensinteg) {
		sens = (sadb_sens_t *)walker;
		sens->sadb_sens_len = SADB_8TO64(sizeof (sadb_sens_t *) +
		    ipsa->ipsa_senslen + ipsa->ipsa_integlen);
		sens->sadb_sens_dpd = ipsa->ipsa_dpd;
		sens->sadb_sens_sens_level = ipsa->ipsa_senslevel;
		sens->sadb_sens_integ_level = ipsa->ipsa_integlevel;
		sens->sadb_sens_sens_len = SADB_8TO64(ipsa->ipsa_senslen);
		sens->sadb_sens_integ_len = SADB_8TO64(ipsa->ipsa_integlen);
		sens->sadb_sens_reserved = 0;
		bitmap = (uint64_t *)(sens + 1);
		if (ipsa->ipsa_sens != NULL) {
			bcopy(ipsa->ipsa_sens, bitmap, ipsa->ipsa_senslen);
			bitmap += sens->sadb_sens_sens_len;
		}
		if (ipsa->ipsa_integ != NULL)
			bcopy(ipsa->ipsa_integ, bitmap, ipsa->ipsa_integlen);
		walker = (sadb_ext_t *)((uint64_t *)walker +
		    walker->sadb_ext_len);
	}

	if (paired) {
		pair_ext = (sadb_x_pair_t *)walker;

		pair_ext->sadb_x_pair_len = SADB_8TO64(sizeof (sadb_x_pair_t));
		pair_ext->sadb_x_pair_exttype = SADB_X_EXT_PAIR;
		pair_ext->sadb_x_pair_spi = otherspi;

		walker = (sadb_ext_t *)((uint64_t *)walker +
		    walker->sadb_ext_len);
	}

	if (ipsa->ipsa_replay != 0) {
		repl_ctr = (sadb_x_replay_ctr_t *)walker;
		repl_ctr->sadb_x_rc_len = SADB_8TO64(sizeof (*repl_ctr));
		repl_ctr->sadb_x_rc_exttype = SADB_X_EXT_REPLAY_VALUE;
		repl_ctr->sadb_x_rc_replay32 = ipsa->ipsa_replay;
		repl_ctr->sadb_x_rc_replay64 = 0;
		walker = (sadb_ext_t *)(repl_ctr + 1);
	}

bail:
	/* Pardon any delays... */
	mutex_exit(&ipsa->ipsa_lock);

	return (mp);
}

/*
 * Strip out key headers or unmarked headers (SADB_EXT_KEY_*, SADB_EXT_UNKNOWN)
 * and adjust base message accordingly.
 *
 * Assume message is pulled up in one piece of contiguous memory.
 *
 * Say if we start off with:
 *
 * +------+----+-------------+-----------+---------------+---------------+
 * | base | SA | source addr | dest addr | rsrvd. or key | soft lifetime |
 * +------+----+-------------+-----------+---------------+---------------+
 *
 * we will end up with
 *
 * +------+----+-------------+-----------+---------------+
 * | base | SA | source addr | dest addr | soft lifetime |
 * +------+----+-------------+-----------+---------------+
 */
static void
sadb_strip(sadb_msg_t *samsg)
{
	sadb_ext_t *ext;
	uint8_t *target = NULL;
	uint8_t *msgend;
	int sofar = SADB_8TO64(sizeof (*samsg));
	int copylen;

	ext = (sadb_ext_t *)(samsg + 1);
	msgend = (uint8_t *)samsg;
	msgend += SADB_64TO8(samsg->sadb_msg_len);
	while ((uint8_t *)ext < msgend) {
		if (ext->sadb_ext_type == SADB_EXT_RESERVED ||
		    ext->sadb_ext_type == SADB_EXT_KEY_AUTH ||
		    ext->sadb_ext_type == SADB_X_EXT_EDUMP ||
		    ext->sadb_ext_type == SADB_EXT_KEY_ENCRYPT) {
			/*
			 * Aha!	 I found a header to be erased.
			 */

			if (target != NULL) {
				/*
				 * If I had a previous header to be erased,
				 * copy over it.  I can get away with just
				 * copying backwards because the target will
				 * always be 8 bytes behind the source.
				 */
				copylen = ((uint8_t *)ext) - (target +
				    SADB_64TO8(
				    ((sadb_ext_t *)target)->sadb_ext_len));
				ovbcopy(((uint8_t *)ext - copylen), target,
				    copylen);
				target += copylen;
				((sadb_ext_t *)target)->sadb_ext_len =
				    SADB_8TO64(((uint8_t *)ext) - target +
				    SADB_64TO8(ext->sadb_ext_len));
			} else {
				target = (uint8_t *)ext;
			}
		} else {
			sofar += ext->sadb_ext_len;
		}

		ext = (sadb_ext_t *)(((uint64_t *)ext) + ext->sadb_ext_len);
	}

	ASSERT((uint8_t *)ext == msgend);

	if (target != NULL) {
		copylen = ((uint8_t *)ext) - (target +
		    SADB_64TO8(((sadb_ext_t *)target)->sadb_ext_len));
		if (copylen != 0)
			ovbcopy(((uint8_t *)ext - copylen), target, copylen);
	}

	/* Adjust samsg. */
	samsg->sadb_msg_len = (uint16_t)sofar;

	/* Assume all of the rest is cleared by caller in sadb_pfkey_echo(). */
}

/*
 * AH needs to send an error to PF_KEY.	 Assume mp points to an M_CTL
 * followed by an M_DATA with a PF_KEY message in it.  The serial of
 * the sending keysock instance is included.
 */
void
sadb_pfkey_error(queue_t *pfkey_q, mblk_t *mp, int error, int diagnostic,
    uint_t serial)
{
	mblk_t *msg = mp->b_cont;
	sadb_msg_t *samsg;
	keysock_out_t *kso;

	/*
	 * Enough functions call this to merit a NULL queue check.
	 */
	if (pfkey_q == NULL) {
		freemsg(mp);
		return;
	}

	ASSERT(msg != NULL);
	ASSERT((mp->b_wptr - mp->b_rptr) == sizeof (ipsec_info_t));
	ASSERT((msg->b_wptr - msg->b_rptr) >= sizeof (sadb_msg_t));
	samsg = (sadb_msg_t *)msg->b_rptr;
	kso = (keysock_out_t *)mp->b_rptr;

	kso->ks_out_type = KEYSOCK_OUT;
	kso->ks_out_len = sizeof (*kso);
	kso->ks_out_serial = serial;

	/*
	 * Only send the base message up in the event of an error.
	 * Don't worry about bzero()-ing, because it was probably bogus
	 * anyway.
	 */
	msg->b_wptr = msg->b_rptr + sizeof (*samsg);
	samsg = (sadb_msg_t *)msg->b_rptr;
	samsg->sadb_msg_len = SADB_8TO64(sizeof (*samsg));
	samsg->sadb_msg_errno = (uint8_t)error;
	if (diagnostic != SADB_X_DIAGNOSTIC_PRESET)
		samsg->sadb_x_msg_diagnostic = (uint16_t)diagnostic;

	putnext(pfkey_q, mp);
}

/*
 * Send a successful return packet back to keysock via the queue in pfkey_q.
 *
 * Often, an SA is associated with the reply message, it's passed in if needed,
 * and NULL if not.  BTW, that ipsa will have its refcnt appropriately held,
 * and the caller will release said refcnt.
 */
void
sadb_pfkey_echo(queue_t *pfkey_q, mblk_t *mp, sadb_msg_t *samsg,
    keysock_in_t *ksi, ipsa_t *ipsa)
{
	keysock_out_t *kso;
	mblk_t *mp1;
	sadb_msg_t *newsamsg;
	uint8_t *oldend;

	ASSERT((mp->b_cont != NULL) &&
	    ((void *)samsg == (void *)mp->b_cont->b_rptr) &&
	    ((void *)mp->b_rptr == (void *)ksi));

	switch (samsg->sadb_msg_type) {
	case SADB_ADD:
	case SADB_UPDATE:
	case SADB_X_UPDATEPAIR:
	case SADB_X_DELPAIR_STATE:
	case SADB_FLUSH:
	case SADB_DUMP:
		/*
		 * I have all of the message already.  I just need to strip
		 * out the keying material and echo the message back.
		 *
		 * NOTE: for SADB_DUMP, the function sadb_dump() did the
		 * work.  When DUMP reaches here, it should only be a base
		 * message.
		 */
	justecho:
		if (ksi->ks_in_extv[SADB_EXT_KEY_AUTH] != NULL ||
		    ksi->ks_in_extv[SADB_EXT_KEY_ENCRYPT] != NULL ||
		    ksi->ks_in_extv[SADB_X_EXT_EDUMP] != NULL) {
			sadb_strip(samsg);
			/* Assume PF_KEY message is contiguous. */
			ASSERT(mp->b_cont->b_cont == NULL);
			oldend = mp->b_cont->b_wptr;
			mp->b_cont->b_wptr = mp->b_cont->b_rptr +
			    SADB_64TO8(samsg->sadb_msg_len);
			bzero(mp->b_cont->b_wptr, oldend - mp->b_cont->b_wptr);
		}
		break;
	case SADB_GET:
		/*
		 * Do a lot of work here, because of the ipsa I just found.
		 * First construct the new PF_KEY message, then abandon
		 * the old one.
		 */
		mp1 = sadb_sa2msg(ipsa, samsg);
		if (mp1 == NULL) {
			sadb_pfkey_error(pfkey_q, mp, ENOMEM,
			    SADB_X_DIAGNOSTIC_NONE, ksi->ks_in_serial);
			return;
		}
		freemsg(mp->b_cont);
		mp->b_cont = mp1;
		break;
	case SADB_DELETE:
	case SADB_X_DELPAIR:
		if (ipsa == NULL)
			goto justecho;
		/*
		 * Because listening KMds may require more info, treat
		 * DELETE like a special case of GET.
		 */
		mp1 = sadb_sa2msg(ipsa, samsg);
		if (mp1 == NULL) {
			sadb_pfkey_error(pfkey_q, mp, ENOMEM,
			    SADB_X_DIAGNOSTIC_NONE, ksi->ks_in_serial);
			return;
		}
		newsamsg = (sadb_msg_t *)mp1->b_rptr;
		sadb_strip(newsamsg);
		oldend = mp1->b_wptr;
		mp1->b_wptr = mp1->b_rptr + SADB_64TO8(newsamsg->sadb_msg_len);
		bzero(mp1->b_wptr, oldend - mp1->b_wptr);
		freemsg(mp->b_cont);
		mp->b_cont = mp1;
		break;
	default:
		if (mp != NULL)
			freemsg(mp);
		return;
	}

	/* ksi is now null and void. */
	kso = (keysock_out_t *)ksi;
	kso->ks_out_type = KEYSOCK_OUT;
	kso->ks_out_len = sizeof (*kso);
	kso->ks_out_serial = ksi->ks_in_serial;
	/* We're ready to send... */
	putnext(pfkey_q, mp);
}

/*
 * Set up a global pfkey_q instance for AH, ESP, or some other consumer.
 */
void
sadb_keysock_hello(queue_t **pfkey_qp, queue_t *q, mblk_t *mp,
    void (*ager)(void *), void *agerarg, timeout_id_t *top, int satype)
{
	keysock_hello_ack_t *kha;
	queue_t *oldq;

	ASSERT(OTHERQ(q) != NULL);

	/*
	 * First, check atomically that I'm the first and only keysock
	 * instance.
	 *
	 * Use OTHERQ(q), because qreply(q, mp) == putnext(OTHERQ(q), mp),
	 * and I want this module to say putnext(*_pfkey_q, mp) for PF_KEY
	 * messages.
	 */

	oldq = casptr((void **)pfkey_qp, NULL, OTHERQ(q));
	if (oldq != NULL) {
		ASSERT(oldq != q);
		cmn_err(CE_WARN, "Danger!  Multiple keysocks on top of %s.\n",
		    (satype == SADB_SATYPE_ESP)? "ESP" : "AH or other");
		freemsg(mp);
		return;
	}

	kha = (keysock_hello_ack_t *)mp->b_rptr;
	kha->ks_hello_len = sizeof (keysock_hello_ack_t);
	kha->ks_hello_type = KEYSOCK_HELLO_ACK;
	kha->ks_hello_satype = (uint8_t)satype;

	/*
	 * If we made it past the casptr, then we have "exclusive" access
	 * to the timeout handle.  Fire it off in 4 seconds, because it
	 * just seems like a good interval.
	 */
	*top = qtimeout(*pfkey_qp, ager, agerarg, drv_usectohz(4000000));

	putnext(*pfkey_qp, mp);
}

/*
 * Normalize IPv4-mapped IPv6 addresses (and prefixes) as appropriate.
 *
 * Check addresses themselves for wildcard or multicast.
 * Check ire table for local/non-local/broadcast.
 */
int
sadb_addrcheck(queue_t *pfkey_q, mblk_t *mp, sadb_ext_t *ext, uint_t serial,
    netstack_t *ns)
{
	sadb_address_t *addr = (sadb_address_t *)ext;
	struct sockaddr_in *sin;
	struct sockaddr_in6 *sin6;
	ire_t *ire;
	int diagnostic, type;
	boolean_t normalized = B_FALSE;

	ASSERT(ext != NULL);
	ASSERT((ext->sadb_ext_type == SADB_EXT_ADDRESS_SRC) ||
	    (ext->sadb_ext_type == SADB_EXT_ADDRESS_DST) ||
	    (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_INNER_SRC) ||
	    (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_INNER_DST) ||
	    (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_NATT_LOC) ||
	    (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_NATT_REM));

	/* Assign both sockaddrs, the compiler will do the right thing. */
	sin = (struct sockaddr_in *)(addr + 1);
	sin6 = (struct sockaddr_in6 *)(addr + 1);

	if (sin6->sin6_family == AF_INET6) {
		if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
			/*
			 * Convert to an AF_INET sockaddr.  This means the
			 * return messages will have the extra space, but have
			 * AF_INET sockaddrs instead of AF_INET6.
			 *
			 * Yes, RFC 2367 isn't clear on what to do here w.r.t.
			 * mapped addresses, but since AF_INET6 ::ffff:<v4> is
			 * equal to AF_INET <v4>, it shouldnt be a huge
			 * problem.
			 */
			sin->sin_family = AF_INET;
			IN6_V4MAPPED_TO_INADDR(&sin6->sin6_addr,
			    &sin->sin_addr);
			bzero(&sin->sin_zero, sizeof (sin->sin_zero));
			normalized = B_TRUE;
		}
	} else if (sin->sin_family != AF_INET) {
		switch (ext->sadb_ext_type) {
		case SADB_EXT_ADDRESS_SRC:
			diagnostic = SADB_X_DIAGNOSTIC_BAD_SRC_AF;
			break;
		case SADB_EXT_ADDRESS_DST:
			diagnostic = SADB_X_DIAGNOSTIC_BAD_DST_AF;
			break;
		case SADB_X_EXT_ADDRESS_INNER_SRC:
			diagnostic = SADB_X_DIAGNOSTIC_BAD_PROXY_AF;
			break;
		case SADB_X_EXT_ADDRESS_INNER_DST:
			diagnostic = SADB_X_DIAGNOSTIC_BAD_INNER_DST_AF;
			break;
		case SADB_X_EXT_ADDRESS_NATT_LOC:
			diagnostic = SADB_X_DIAGNOSTIC_BAD_NATT_LOC_AF;
			break;
		case SADB_X_EXT_ADDRESS_NATT_REM:
			diagnostic = SADB_X_DIAGNOSTIC_BAD_NATT_REM_AF;
			break;
			/* There is no default, see above ASSERT. */
		}
bail:
		if (pfkey_q != NULL) {
			sadb_pfkey_error(pfkey_q, mp, EINVAL, diagnostic,
			    serial);
		} else {
			/*
			 * Scribble in sadb_msg that we got passed in.
			 * Overload "mp" to be an sadb_msg pointer.
			 */
			sadb_msg_t *samsg = (sadb_msg_t *)mp;

			samsg->sadb_msg_errno = EINVAL;
			samsg->sadb_x_msg_diagnostic = diagnostic;
		}
		return (KS_IN_ADDR_UNKNOWN);
	}

	if (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_INNER_SRC ||
	    ext->sadb_ext_type == SADB_X_EXT_ADDRESS_INNER_DST) {
		/*
		 * We need only check for prefix issues.
		 */

		/* Set diagnostic now, in case we need it later. */
		diagnostic =
		    (ext->sadb_ext_type == SADB_X_EXT_ADDRESS_INNER_SRC) ?
		    SADB_X_DIAGNOSTIC_PREFIX_INNER_SRC :
		    SADB_X_DIAGNOSTIC_PREFIX_INNER_DST;

		if (normalized)
			addr->sadb_address_prefixlen -= 96;

		/*
		 * Verify and mask out inner-addresses based on prefix length.
		 */
		if (sin->sin_family == AF_INET) {
			if (addr->sadb_address_prefixlen > 32)
				goto bail;
			sin->sin_addr.s_addr &=
			    ip_plen_to_mask(addr->sadb_address_prefixlen);
		} else {
			in6_addr_t mask;

			ASSERT(sin->sin_family == AF_INET6);
			/*
			 * ip_plen_to_mask_v6() returns NULL if the value in
			 * question is out of range.
			 */
			if (ip_plen_to_mask_v6(addr->sadb_address_prefixlen,
			    &mask) == NULL)
				goto bail;
			sin6->sin6_addr.s6_addr32[0] &= mask.s6_addr32[0];
			sin6->sin6_addr.s6_addr32[1] &= mask.s6_addr32[1];
			sin6->sin6_addr.s6_addr32[2] &= mask.s6_addr32[2];
			sin6->sin6_addr.s6_addr32[3] &= mask.s6_addr32[3];
		}

		/* We don't care in these cases. */
		return (KS_IN_ADDR_DONTCARE);
	}

	if (sin->sin_family == AF_INET6) {
		/* Check the easy ones now. */
		if (IN6_IS_ADDR_MULTICAST(&sin6->sin6_addr))
			return (KS_IN_ADDR_MBCAST);
		if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr))
			return (KS_IN_ADDR_UNSPEC);
		/*
		 * At this point, we're a unicast IPv6 address.
		 *
		 * A ctable lookup for local is sufficient here.  If we're
		 * local, return KS_IN_ADDR_ME, otherwise KS_IN_ADDR_NOTME.
		 *
		 * XXX Zones alert -> me/notme decision needs to be tempered
		 * by what zone we're in when we go to zone-aware IPsec.
		 */
		ire = ire_ctable_lookup_v6(&sin6->sin6_addr, NULL,
		    IRE_LOCAL, NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE,
		    ns->netstack_ip);
		if (ire != NULL) {
			/* Hey hey, it's local. */
			IRE_REFRELE(ire);
			return (KS_IN_ADDR_ME);
		}
	} else {
		ASSERT(sin->sin_family == AF_INET);
		if (sin->sin_addr.s_addr == INADDR_ANY)
			return (KS_IN_ADDR_UNSPEC);
		if (CLASSD(sin->sin_addr.s_addr))
			return (KS_IN_ADDR_MBCAST);
		/*
		 * At this point we're a unicast or broadcast IPv4 address.
		 *
		 * Lookup on the ctable for IRE_BROADCAST or IRE_LOCAL.
		 * A NULL return value is NOTME, otherwise, look at the
		 * returned ire for broadcast or not and return accordingly.
		 *
		 * XXX Zones alert -> me/notme decision needs to be tempered
		 * by what zone we're in when we go to zone-aware IPsec.
		 */
		ire = ire_ctable_lookup(sin->sin_addr.s_addr, 0,
		    IRE_LOCAL | IRE_BROADCAST, NULL, ALL_ZONES, NULL,
		    MATCH_IRE_TYPE, ns->netstack_ip);
		if (ire != NULL) {
			/* Check for local or broadcast */
			type = ire->ire_type;
			IRE_REFRELE(ire);
			ASSERT(type == IRE_LOCAL || type == IRE_BROADCAST);
			return ((type == IRE_LOCAL) ? KS_IN_ADDR_ME :
			    KS_IN_ADDR_MBCAST);
		}
	}

	return (KS_IN_ADDR_NOTME);
}

/*
 * Address normalizations and reality checks for inbound PF_KEY messages.
 *
 * For the case of src == unspecified AF_INET6, and dst == AF_INET, convert
 * the source to AF_INET.  Do the same for the inner sources.
 */
boolean_t
sadb_addrfix(keysock_in_t *ksi, queue_t *pfkey_q, mblk_t *mp, netstack_t *ns)
{
	struct sockaddr_in *src, *isrc;
	struct sockaddr_in6 *dst, *idst;
	sadb_address_t *srcext, *dstext;
	uint16_t sport;
	sadb_ext_t **extv = ksi->ks_in_extv;
	int rc;

	if (extv[SADB_EXT_ADDRESS_SRC] != NULL) {
		rc = sadb_addrcheck(pfkey_q, mp, extv[SADB_EXT_ADDRESS_SRC],
		    ksi->ks_in_serial, ns);
		if (rc == KS_IN_ADDR_UNKNOWN)
			return (B_FALSE);
		if (rc == KS_IN_ADDR_MBCAST) {
			sadb_pfkey_error(pfkey_q, mp, EINVAL,
			    SADB_X_DIAGNOSTIC_BAD_SRC, ksi->ks_in_serial);
			return (B_FALSE);
		}
		ksi->ks_in_srctype = rc;
	}

	if (extv[SADB_EXT_ADDRESS_DST] != NULL) {
		rc = sadb_addrcheck(pfkey_q, mp, extv[SADB_EXT_ADDRESS_DST],
		    ksi->ks_in_serial, ns);
		if (rc == KS_IN_ADDR_UNKNOWN)
			return (B_FALSE);
		if (rc == KS_IN_ADDR_UNSPEC) {
			sadb_pfkey_error(pfkey_q, mp, EINVAL,
			    SADB_X_DIAGNOSTIC_BAD_DST, ksi->ks_in_serial);
			return (B_FALSE);
		}
		ksi->ks_in_dsttype = rc;
	}

	/*
	 * NAT-Traversal addrs are simple enough to not require all of
	 * the checks in sadb_addrcheck().  Just normalize or reject if not
	 * AF_INET.
	 */
	if (extv[SADB_X_EXT_ADDRESS_NATT_LOC] != NULL) {
		rc = sadb_addrcheck(pfkey_q, mp,
		    extv[SADB_X_EXT_ADDRESS_NATT_LOC], ksi->ks_in_serial, ns);

		/*
		 * Local NAT-T addresses never use an IRE_LOCAL, so it should
		 * always be NOTME, or UNSPEC (to handle both tunnel mode
		 * AND local-port flexibility).
		 */
		if (rc != KS_IN_ADDR_NOTME && rc != KS_IN_ADDR_UNSPEC) {
			sadb_pfkey_error(pfkey_q, mp, EINVAL,
			    SADB_X_DIAGNOSTIC_MALFORMED_NATT_LOC,
			    ksi->ks_in_serial);
			return (B_FALSE);
		}
		src = (struct sockaddr_in *)
		    (((sadb_address_t *)extv[SADB_X_EXT_ADDRESS_NATT_LOC]) + 1);
		if (src->sin_family != AF_INET) {
			sadb_pfkey_error(pfkey_q, mp, EINVAL,
			    SADB_X_DIAGNOSTIC_BAD_NATT_LOC_AF,
			    ksi->ks_in_serial);
			return (B_FALSE);
		}
	}

	if (extv[SADB_X_EXT_ADDRESS_NATT_REM] != NULL) {
		rc = sadb_addrcheck(pfkey_q, mp,
		    extv[SADB_X_EXT_ADDRESS_NATT_REM], ksi->ks_in_serial, ns);

		/*
		 * Remote NAT-T addresses never use an IRE_LOCAL, so it should
		 * always be NOTME, or UNSPEC if it's a tunnel-mode SA.
		 */
		if (rc != KS_IN_ADDR_NOTME &&
		    !(extv[SADB_X_EXT_ADDRESS_INNER_SRC] != NULL &&
		    rc == KS_IN_ADDR_UNSPEC)) {
			sadb_pfkey_error(pfkey_q, mp, EINVAL,
			    SADB_X_DIAGNOSTIC_MALFORMED_NATT_REM,
			    ksi->ks_in_serial);
			return (B_FALSE);
		}
		src = (struct sockaddr_in *)
		    (((sadb_address_t *)extv[SADB_X_EXT_ADDRESS_NATT_REM]) + 1);
		if (src->sin_family != AF_INET) {
			sadb_pfkey_error(pfkey_q, mp, EINVAL,
			    SADB_X_DIAGNOSTIC_BAD_NATT_REM_AF,
			    ksi->ks_in_serial);
			return (B_FALSE);
		}
	}

	if (extv[SADB_X_EXT_ADDRESS_INNER_SRC] != NULL) {
		if (extv[SADB_X_EXT_ADDRESS_INNER_DST] == NULL) {
			sadb_pfkey_error(pfkey_q, mp, EINVAL,
			    SADB_X_DIAGNOSTIC_MISSING_INNER_DST,
			    ksi->ks_in_serial);
			return (B_FALSE);
		}

		if (sadb_addrcheck(pfkey_q, mp,
		    extv[SADB_X_EXT_ADDRESS_INNER_DST], ksi->ks_in_serial, ns)
		    == KS_IN_ADDR_UNKNOWN ||
		    sadb_addrcheck(pfkey_q, mp,
		    extv[SADB_X_EXT_ADDRESS_INNER_SRC], ksi->ks_in_serial, ns)
		    == KS_IN_ADDR_UNKNOWN)
			return (B_FALSE);

		isrc = (struct sockaddr_in *)
		    (((sadb_address_t *)extv[SADB_X_EXT_ADDRESS_INNER_SRC]) +
		    1);
		idst = (struct sockaddr_in6 *)
		    (((sadb_address_t *)extv[SADB_X_EXT_ADDRESS_INNER_DST]) +
		    1);
		if (isrc->sin_family != idst->sin6_family) {
			sadb_pfkey_error(pfkey_q, mp, EINVAL,
			    SADB_X_DIAGNOSTIC_INNER_AF_MISMATCH,
			    ksi->ks_in_serial);
			return (B_FALSE);
		}
	} else if (extv[SADB_X_EXT_ADDRESS_INNER_DST] != NULL) {
			sadb_pfkey_error(pfkey_q, mp, EINVAL,
			    SADB_X_DIAGNOSTIC_MISSING_INNER_SRC,
			    ksi->ks_in_serial);
			return (B_FALSE);
	} else {
		isrc = NULL;	/* For inner/outer port check below. */
	}

	dstext = (sadb_address_t *)extv[SADB_EXT_ADDRESS_DST];
	srcext = (sadb_address_t *)extv[SADB_EXT_ADDRESS_SRC];

	if (dstext == NULL || srcext == NULL)
		return (B_TRUE);

	dst = (struct sockaddr_in6 *)(dstext + 1);
	src = (struct sockaddr_in *)(srcext + 1);

	if (isrc != NULL &&
	    (isrc->sin_port != 0 || idst->sin6_port != 0) &&
	    (src->sin_port != 0 || dst->sin6_port != 0)) {
		/* Can't set inner and outer ports in one SA. */
		sadb_pfkey_error(pfkey_q, mp, EINVAL,
		    SADB_X_DIAGNOSTIC_DUAL_PORT_SETS,
		    ksi->ks_in_serial);
		return (B_FALSE);
	}

	if (dst->sin6_family == src->sin_family)
		return (B_TRUE);

	if (srcext->sadb_address_proto != dstext->sadb_address_proto) {
		if (srcext->sadb_address_proto == 0) {
			srcext->sadb_address_proto = dstext->sadb_address_proto;
		} else if (dstext->sadb_address_proto == 0) {
			dstext->sadb_address_proto = srcext->sadb_address_proto;
		} else {
			/* Inequal protocols, neither were 0.  Report error. */
			sadb_pfkey_error(pfkey_q, mp, EINVAL,
			    SADB_X_DIAGNOSTIC_PROTO_MISMATCH,
			    ksi->ks_in_serial);
			return (B_FALSE);
		}
	}

	/*
	 * With the exception of an unspec IPv6 source and an IPv4
	 * destination, address families MUST me matched.
	 */
	if (src->sin_family == AF_INET ||
	    ksi->ks_in_srctype != KS_IN_ADDR_UNSPEC) {
		sadb_pfkey_error(pfkey_q, mp, EINVAL,
		    SADB_X_DIAGNOSTIC_AF_MISMATCH, ksi->ks_in_serial);
		return (B_FALSE);
	}

	/*
	 * Convert "src" to AF_INET INADDR_ANY.  We rely on sin_port being
	 * in the same place for sockaddr_in and sockaddr_in6.
	 */
	sport = src->sin_port;
	bzero(src, sizeof (*src));
	src->sin_family = AF_INET;
	src->sin_port = sport;

	return (B_TRUE);
}

/*
 * Set the results in "addrtype", given an IRE as requested by
 * sadb_addrcheck().
 */
int
sadb_addrset(ire_t *ire)
{
	if ((ire->ire_type & IRE_BROADCAST) ||
	    (ire->ire_ipversion == IPV4_VERSION && CLASSD(ire->ire_addr)) ||
	    (ire->ire_ipversion == IPV6_VERSION &&
	    IN6_IS_ADDR_MULTICAST(&(ire->ire_addr_v6))))
		return (KS_IN_ADDR_MBCAST);
	if (ire->ire_type & (IRE_LOCAL | IRE_LOOPBACK))
		return (KS_IN_ADDR_ME);
	return (KS_IN_ADDR_NOTME);
}


/*
 * Walker callback function to delete sa's based on src/dst address.
 * Assumes that we're called with *head locked, no other locks held;
 * Conveniently, and not coincidentally, this is both what sadb_walker
 * gives us and also what sadb_unlinkassoc expects.
 */

struct sadb_purge_state
{
	uint32_t *src;
	uint32_t *dst;
	sa_family_t af;
	boolean_t inbnd;
	char *sidstr;
	char *didstr;
	uint16_t sidtype;
	uint16_t didtype;
	uint32_t kmproto;
	uint8_t sadb_sa_state;
	mblk_t *mq;
	sadb_t *sp;
};

static void
sadb_purge_cb(isaf_t *head, ipsa_t *entry, void *cookie)
{
	struct sadb_purge_state *ps = (struct sadb_purge_state *)cookie;

	ASSERT(MUTEX_HELD(&head->isaf_lock));

	mutex_enter(&entry->ipsa_lock);

	if ((entry->ipsa_state == IPSA_STATE_LARVAL) ||
	    (ps->src != NULL &&
	    !IPSA_ARE_ADDR_EQUAL(entry->ipsa_srcaddr, ps->src, ps->af)) ||
	    (ps->dst != NULL &&
	    !IPSA_ARE_ADDR_EQUAL(entry->ipsa_dstaddr, ps->dst, ps->af)) ||
	    (ps->didstr != NULL && (entry->ipsa_dst_cid != NULL) &&
	    !(ps->didtype == entry->ipsa_dst_cid->ipsid_type &&
	    strcmp(ps->didstr, entry->ipsa_dst_cid->ipsid_cid) == 0)) ||
	    (ps->sidstr != NULL && (entry->ipsa_src_cid != NULL) &&
	    !(ps->sidtype == entry->ipsa_src_cid->ipsid_type &&
	    strcmp(ps->sidstr, entry->ipsa_src_cid->ipsid_cid) == 0)) ||
	    (ps->kmproto <= SADB_X_KMP_MAX && ps->kmproto != entry->ipsa_kmp)) {
		mutex_exit(&entry->ipsa_lock);
		return;
	}

	if (ps->inbnd) {
		sadb_delete_cluster(entry);
	}
	entry->ipsa_state = IPSA_STATE_DEAD;
	(void) sadb_torch_assoc(head, entry, ps->inbnd, &ps->mq);
}

/*
 * Common code to purge an SA with a matching src or dst address.
 * Don't kill larval SA's in such a purge.
 */
int
sadb_purge_sa(mblk_t *mp, keysock_in_t *ksi, sadb_t *sp, queue_t *pfkey_q,
    queue_t *ip_q)
{
	sadb_address_t *dstext =
	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
	sadb_address_t *srcext =
	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
	sadb_ident_t *dstid =
	    (sadb_ident_t *)ksi->ks_in_extv[SADB_EXT_IDENTITY_DST];
	sadb_ident_t *srcid =
	    (sadb_ident_t *)ksi->ks_in_extv[SADB_EXT_IDENTITY_SRC];
	sadb_x_kmc_t *kmc =
	    (sadb_x_kmc_t *)ksi->ks_in_extv[SADB_X_EXT_KM_COOKIE];
	struct sockaddr_in *src, *dst;
	struct sockaddr_in6 *src6, *dst6;
	struct sadb_purge_state ps;

	/*
	 * Don't worry about IPv6 v4-mapped addresses, sadb_addrcheck()
	 * takes care of them.
	 */

	/* enforced by caller */
	ASSERT((dstext != NULL) || (srcext != NULL));

	ps.src = NULL;
	ps.dst = NULL;
#ifdef DEBUG
	ps.af = (sa_family_t)-1;
#endif
	ps.mq = NULL;
	ps.sidstr = NULL;
	ps.didstr = NULL;
	ps.kmproto = SADB_X_KMP_MAX + 1;

	if (dstext != NULL) {
		dst = (struct sockaddr_in *)(dstext + 1);
		ps.af = dst->sin_family;
		if (dst->sin_family == AF_INET6) {
			dst6 = (struct sockaddr_in6 *)dst;
			ps.dst = (uint32_t *)&dst6->sin6_addr;
		} else {
			ps.dst = (uint32_t *)&dst->sin_addr;
		}
	}

	if (srcext != NULL) {
		src = (struct sockaddr_in *)(srcext + 1);
		ps.af = src->sin_family;
		if (src->sin_family == AF_INET6) {
			src6 = (struct sockaddr_in6 *)(srcext + 1);
			ps.src = (uint32_t *)&src6->sin6_addr;
		} else {
			ps.src = (uint32_t *)&src->sin_addr;
		}
		ASSERT(dstext == NULL || src->sin_family == dst->sin_family);
	}

	ASSERT(ps.af != (sa_family_t)-1);

	if (dstid != NULL) {
		/*
		 * NOTE:  May need to copy string in the future
		 * if the inbound keysock message disappears for some strange
		 * reason.
		 */
		ps.didstr = (char *)(dstid + 1);
		ps.didtype = dstid->sadb_ident_type;
	}

	if (srcid != NULL) {
		/*
		 * NOTE:  May need to copy string in the future
		 * if the inbound keysock message disappears for some strange
		 * reason.
		 */
		ps.sidstr = (char *)(srcid + 1);
		ps.sidtype = srcid->sadb_ident_type;
	}

	if (kmc != NULL)
		ps.kmproto = kmc->sadb_x_kmc_proto;

	/*
	 * This is simple, crude, and effective.
	 * Unimplemented optimizations (TBD):
	 * - we can limit how many places we search based on where we
	 * think the SA is filed.
	 * - if we get a dst address, we can hash based on dst addr to find
	 * the correct bucket in the outbound table.
	 */
	ps.inbnd = B_TRUE;
	sadb_walker(sp->sdb_if, sp->sdb_hashsize, sadb_purge_cb, &ps);
	ps.inbnd = B_FALSE;
	sadb_walker(sp->sdb_of, sp->sdb_hashsize, sadb_purge_cb, &ps);

	if (ps.mq != NULL)
		sadb_drain_torchq(ip_q, ps.mq);

	ASSERT(mp->b_cont != NULL);
	sadb_pfkey_echo(pfkey_q, mp, (sadb_msg_t *)mp->b_cont->b_rptr, ksi,
	    NULL);
	return (0);
}

static void
sadb_delpair_state(isaf_t *head, ipsa_t *entry, void *cookie)
{
	struct sadb_purge_state *ps = (struct sadb_purge_state *)cookie;
	isaf_t  *inbound_bucket;
	ipsa_t *peer_assoc;

	ASSERT(MUTEX_HELD(&head->isaf_lock));

	mutex_enter(&entry->ipsa_lock);

	if ((entry->ipsa_state != ps->sadb_sa_state) ||
	    ((ps->src != NULL) &&
	    !IPSA_ARE_ADDR_EQUAL(entry->ipsa_srcaddr, ps->src, ps->af))) {
		mutex_exit(&entry->ipsa_lock);
		return;
	}

	/*
	 * The isaf_t *, which is passed in , is always an outbound bucket,
	 * and we are preserving the outbound-then-inbound hash-bucket lock
	 * ordering. The sadb_walker() which triggers this function is called
	 * only on the outbound fanout, and the corresponding inbound bucket
	 * lock is safe to acquire here.
	 */

	if (entry->ipsa_haspeer) {
		inbound_bucket = INBOUND_BUCKET(ps->sp, entry->ipsa_spi);
		mutex_enter(&inbound_bucket->isaf_lock);
		peer_assoc = ipsec_getassocbyspi(inbound_bucket,
		    entry->ipsa_spi, entry->ipsa_srcaddr,
		    entry->ipsa_dstaddr, entry->ipsa_addrfam);
	} else {
		inbound_bucket = INBOUND_BUCKET(ps->sp, entry->ipsa_otherspi);
		mutex_enter(&inbound_bucket->isaf_lock);
		peer_assoc = ipsec_getassocbyspi(inbound_bucket,
		    entry->ipsa_otherspi, entry->ipsa_dstaddr,
		    entry->ipsa_srcaddr, entry->ipsa_addrfam);
	}

	entry->ipsa_state = IPSA_STATE_DEAD;
	(void) sadb_torch_assoc(head, entry, B_FALSE, &ps->mq);
	if (peer_assoc != NULL) {
		mutex_enter(&peer_assoc->ipsa_lock);
		peer_assoc->ipsa_state = IPSA_STATE_DEAD;
		(void) sadb_torch_assoc(inbound_bucket, peer_assoc,
		    B_FALSE, &ps->mq);
	}
	mutex_exit(&inbound_bucket->isaf_lock);
}

/*
 * Common code to delete/get an SA.
 */
int
sadb_delget_sa(mblk_t *mp, keysock_in_t *ksi, sadbp_t *spp,
    int *diagnostic, queue_t *pfkey_q, uint8_t sadb_msg_type)
{
	sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
	sadb_address_t *srcext =
	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
	sadb_address_t *dstext =
	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
	ipsa_t *echo_target = NULL;
	ipsap_t *ipsapp;
	mblk_t *torchq = NULL;
	uint_t	error = 0;

	if (assoc == NULL) {
		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
		return (EINVAL);
	}

	if (sadb_msg_type == SADB_X_DELPAIR_STATE) {
		struct sockaddr_in *src;
		struct sockaddr_in6 *src6;
		struct sadb_purge_state ps;

		if (srcext == NULL) {
			*diagnostic = SADB_X_DIAGNOSTIC_MISSING_SRC;
			return (EINVAL);
		}
		ps.src = NULL;
		ps.mq = NULL;
		src = (struct sockaddr_in *)(srcext + 1);
		ps