xref: /onnv/onnv-gate/usr/src/uts/common/inet/ip/keysock.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 2008 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#pragma ident	"%Z%%M%	%I%	%E% SMI"


#include <sys/param.h>
#include <sys/types.h>
#include <sys/stream.h>
#include <sys/strsubr.h>
#include <sys/strsun.h>
#include <sys/stropts.h>
#include <sys/vnode.h>
#include <sys/zone.h>
#include <sys/strlog.h>
#include <sys/sysmacros.h>
#define	_SUN_TPI_VERSION 2
#include <sys/tihdr.h>
#include <sys/timod.h>
#include <sys/tiuser.h>
#include <sys/ddi.h>
#include <sys/sunddi.h>
#include <sys/sunldi.h>
#include <sys/file.h>
#include <sys/modctl.h>
#include <sys/debug.h>
#include <sys/kmem.h>
#include <sys/cmn_err.h>
#include <sys/proc.h>
#include <sys/suntpi.h>
#include <sys/atomic.h>
#include <sys/mkdev.h>
#include <sys/policy.h>
#include <sys/disp.h>

#include <sys/socket.h>
#include <netinet/in.h>
#include <net/pfkeyv2.h>

#include <inet/common.h>
#include <netinet/ip6.h>
#include <inet/ip.h>
#include <inet/mi.h>
#include <inet/nd.h>
#include <inet/optcom.h>
#include <inet/ipsec_info.h>
#include <inet/ipsec_impl.h>
#include <inet/keysock.h>

#include <sys/isa_defs.h>

/*
 * This is a transport provider for the PF_KEY key mangement socket.
 * (See RFC 2367 for details.)
 * Downstream messages are wrapped in a keysock consumer interface KEYSOCK_IN
 * messages (see ipsec_info.h), and passed to the appropriate consumer.
 * Upstream messages are generated for all open PF_KEY sockets, when
 * appropriate, as well as the sender (as long as SO_USELOOPBACK is enabled)
 * in reply to downstream messages.
 *
 * Upstream messages must be created asynchronously for the following
 * situations:
 *
 *	1.) A keysock consumer requires an SA, and there is currently none.
 *	2.) An SA expires, either hard or soft lifetime.
 *	3.) Other events a consumer deems fit.
 *
 * The MT model of this is PERMOD, with shared put procedures.  Two types of
 * messages, SADB_FLUSH and SADB_DUMP, need to lock down the perimeter to send
 * down the *multiple* messages they create.
 */

static vmem_t *keysock_vmem;		/* for minor numbers. */

#define	KEYSOCK_MAX_CONSUMERS 256

/* Default structure copied into T_INFO_ACK messages (from rts.c...) */
static struct T_info_ack keysock_g_t_info_ack = {
	T_INFO_ACK,
	T_INFINITE,	/* TSDU_size. Maximum size messages. */
	T_INVALID,	/* ETSDU_size. No expedited data. */
	T_INVALID,	/* CDATA_size. No connect data. */
	T_INVALID,	/* DDATA_size. No disconnect data. */
	0,		/* ADDR_size. */
	0,		/* OPT_size. No user-settable options */
	64 * 1024,	/* TIDU_size. keysock allows maximum size messages. */
	T_COTS,		/* SERV_type. keysock supports connection oriented. */
	TS_UNBND,	/* CURRENT_state. This is set from keysock_state. */
	(XPG4_1)	/* Provider flags */
};

/* Named Dispatch Parameter Management Structure */
typedef struct keysockparam_s {
	uint_t	keysock_param_min;
	uint_t	keysock_param_max;
	uint_t	keysock_param_value;
	char	*keysock_param_name;
} keysockparam_t;

/*
 * Table of NDD variables supported by keysock. These are loaded into
 * keysock_g_nd in keysock_init_nd.
 * All of these are alterable, within the min/max values given, at run time.
 */
static	keysockparam_t	lcl_param_arr[] = {
	/* min	max	value	name */
	{ 4096, 65536,	8192,	"keysock_xmit_hiwat"},
	{ 0,	65536,	1024,	"keysock_xmit_lowat"},
	{ 4096, 65536,	8192,	"keysock_recv_hiwat"},
	{ 65536, 1024*1024*1024, 256*1024,	"keysock_max_buf"},
	{ 0,	3,	0,	"keysock_debug"},
};
#define	keystack_xmit_hiwat	keystack_params[0].keysock_param_value
#define	keystack_xmit_lowat	keystack_params[1].keysock_param_value
#define	keystack_recv_hiwat	keystack_params[2].keysock_param_value
#define	keystack_max_buf	keystack_params[3].keysock_param_value
#define	keystack_debug	keystack_params[4].keysock_param_value

#define	ks0dbg(a)	printf a
/* NOTE:  != 0 instead of > 0 so lint doesn't complain. */
#define	ks1dbg(keystack, a)	if (keystack->keystack_debug != 0) printf a
#define	ks2dbg(keystack, a)	if (keystack->keystack_debug > 1) printf a
#define	ks3dbg(keystack, a)	if (keystack->keystack_debug > 2) printf a

static int keysock_close(queue_t *);
static int keysock_open(queue_t *, dev_t *, int, int, cred_t *);
static void keysock_wput(queue_t *, mblk_t *);
static void keysock_rput(queue_t *, mblk_t *);
static void keysock_rsrv(queue_t *);
static void keysock_passup(mblk_t *, sadb_msg_t *, minor_t,
    keysock_consumer_t *, boolean_t, keysock_stack_t *);
static void *keysock_stack_init(netstackid_t stackid, netstack_t *ns);
static void keysock_stack_fini(netstackid_t stackid, void *arg);

static struct module_info info = {
	5138, "keysock", 1, INFPSZ, 512, 128
};

static struct qinit rinit = {
	(pfi_t)keysock_rput, (pfi_t)keysock_rsrv, keysock_open, keysock_close,
	NULL, &info
};

static struct qinit winit = {
	(pfi_t)keysock_wput, NULL, NULL, NULL, NULL, &info
};

struct streamtab keysockinfo = {
	&rinit, &winit
};

extern struct modlinkage *keysock_modlp;

/*
 * Plumb IPsec.
 *
 * NOTE:  New "default" modules will need to be loaded here if needed before
 *	  boot time.
 */

/* Keep these in global space to keep the lint from complaining. */
static char *IPSECESP = "ipsecesp";
static char *IPSECESPDEV = "/devices/pseudo/ipsecesp@0:ipsecesp";
static char *IPSECAH = "ipsecah";
static char *IPSECAHDEV = "/devices/pseudo/ipsecah@0:ipsecah";
static char *IP6DEV = "/devices/pseudo/ip6@0:ip6";
static char *KEYSOCK = "keysock";
static char *STRMOD = "strmod";

/*
 * Load the other ipsec modules and plumb them together.
 */
int
keysock_plumb_ipsec(netstack_t *ns)
{
	ldi_handle_t	lh, ip6_lh = NULL;
	ldi_ident_t	li = NULL;
	int		err = 0;
	int		muxid, rval;
	boolean_t	esp_present = B_TRUE;
	cred_t		*cr;
	keysock_stack_t *keystack = ns->netstack_keysock;

#ifdef NS_DEBUG
	(void) printf("keysock_plumb_ipsec(%d)\n",
	    ns->netstack_stackid);
#endif

	keystack->keystack_plumbed = 0;	/* we're trying again.. */

	cr = zone_get_kcred(netstackid_to_zoneid(
	    keystack->keystack_netstack->netstack_stackid));
	ASSERT(cr != NULL);
	/*
	 * Load up the drivers (AH/ESP).
	 *
	 * I do this separately from the actual plumbing in case this function
	 * ever gets called from a diskless boot before the root filesystem is
	 * up.  I don't have to worry about "keysock" because, well, if I'm
	 * here, keysock must've loaded successfully.
	 */
	if (i_ddi_attach_pseudo_node(IPSECAH) == NULL) {
		ks0dbg(("IPsec:  AH failed to attach.\n"));
		goto bail;
	}
	if (i_ddi_attach_pseudo_node(IPSECESP) == NULL) {
		ks0dbg(("IPsec:  ESP failed to attach.\n"));
		esp_present = B_FALSE;
	}

	/*
	 * Set up the IP streams for AH and ESP, as well as tacking keysock
	 * on top of them.  Assume keysock has set the autopushes up already.
	 */

	/* Open IP. */
	err = ldi_ident_from_mod(keysock_modlp, &li);
	if (err) {
		ks0dbg(("IPsec:  lid_ident_from_mod failed (err %d).\n",
		    err));
		goto bail;
	}

	err = ldi_open_by_name(IP6DEV, FREAD|FWRITE, cr, &ip6_lh, li);
	if (err) {
		ks0dbg(("IPsec:  Open of IP6 failed (err %d).\n", err));
		goto bail;
	}

	/* PLINK KEYSOCK/AH */
	err = ldi_open_by_name(IPSECAHDEV, FREAD|FWRITE, cr, &lh, li);
	if (err) {
		ks0dbg(("IPsec:  Open of AH failed (err %d).\n", err));
		goto bail;
	}
	err = ldi_ioctl(lh,
	    I_PUSH, (intptr_t)KEYSOCK, FKIOCTL, cr, &rval);
	if (err) {
		ks0dbg(("IPsec:  Push of KEYSOCK onto AH failed (err %d).\n",
		    err));
		(void) ldi_close(lh, FREAD|FWRITE, cr);
		goto bail;
	}
	err = ldi_ioctl(ip6_lh, I_PLINK, (intptr_t)lh,
	    FREAD+FWRITE+FNOCTTY+FKIOCTL, cr, &muxid);
	if (err) {
		ks0dbg(("IPsec:  PLINK of KEYSOCK/AH failed (err %d).\n", err));
		(void) ldi_close(lh, FREAD|FWRITE, cr);
		goto bail;
	}
	(void) ldi_close(lh, FREAD|FWRITE, cr);

	/* PLINK KEYSOCK/ESP */
	if (esp_present) {
		err = ldi_open_by_name(IPSECESPDEV,
		    FREAD|FWRITE, cr, &lh, li);
		if (err) {
			ks0dbg(("IPsec:  Open of ESP failed (err %d).\n", err));
			goto bail;
		}
		err = ldi_ioctl(lh,
		    I_PUSH, (intptr_t)KEYSOCK, FKIOCTL, cr, &rval);
		if (err) {
			ks0dbg(("IPsec:  "
			    "Push of KEYSOCK onto ESP failed (err %d).\n",
			    err));
			(void) ldi_close(lh, FREAD|FWRITE, cr);
			goto bail;
		}
		err = ldi_ioctl(ip6_lh, I_PLINK, (intptr_t)lh,
		    FREAD+FWRITE+FNOCTTY+FKIOCTL, cr, &muxid);
		if (err) {
			ks0dbg(("IPsec:  "
			    "PLINK of KEYSOCK/ESP failed (err %d).\n", err));
			(void) ldi_close(lh, FREAD|FWRITE, cr);
			goto bail;
		}
		(void) ldi_close(lh, FREAD|FWRITE, cr);
	}

bail:
	keystack->keystack_plumbed = (err == 0) ? 1 : -1;
	if (ip6_lh != NULL) {
		(void) ldi_close(ip6_lh, FREAD|FWRITE, cr);
	}
	if (li != NULL)
		ldi_ident_release(li);
#ifdef NS_DEBUG
	(void) printf("keysock_plumb_ipsec -> %d\n",
	    keystack->keystack_plumbed);
#endif
	crfree(cr);
	return (err);
}

/* ARGSUSED */
static int
keysock_param_get(q, mp, cp, cr)
	queue_t	*q;
	mblk_t	*mp;
	caddr_t	cp;
	cred_t *cr;
{
	keysockparam_t	*keysockpa = (keysockparam_t *)cp;
	uint_t value;
	keysock_t *ks = (keysock_t *)q->q_ptr;
	keysock_stack_t	*keystack = ks->keysock_keystack;

	mutex_enter(&keystack->keystack_param_lock);
	value = keysockpa->keysock_param_value;
	mutex_exit(&keystack->keystack_param_lock);

	(void) mi_mpprintf(mp, "%u", value);
	return (0);
}

/* This routine sets an NDD variable in a keysockparam_t structure. */
/* ARGSUSED */
static int
keysock_param_set(q, mp, value, cp, cr)
	queue_t	*q;
	mblk_t	*mp;
	char	*value;
	caddr_t	cp;
	cred_t *cr;
{
	ulong_t	new_value;
	keysockparam_t	*keysockpa = (keysockparam_t *)cp;
	keysock_t *ks = (keysock_t *)q->q_ptr;
	keysock_stack_t	*keystack = ks->keysock_keystack;

	/* Convert the value from a string into a long integer. */
	if (ddi_strtoul(value, NULL, 10, &new_value) != 0)
		return (EINVAL);

	mutex_enter(&keystack->keystack_param_lock);
	/*
	 * Fail the request if the new value does not lie within the
	 * required bounds.
	 */
	if (new_value < keysockpa->keysock_param_min ||
	    new_value > keysockpa->keysock_param_max) {
		mutex_exit(&keystack->keystack_param_lock);
		return (EINVAL);
	}

	/* Set the new value */
	keysockpa->keysock_param_value = new_value;
	mutex_exit(&keystack->keystack_param_lock);

	return (0);
}

/*
 * Initialize keysock at module load time
 */
boolean_t
keysock_ddi_init(void)
{
	keysock_max_optsize = optcom_max_optsize(
	    keysock_opt_obj.odb_opt_des_arr, keysock_opt_obj.odb_opt_arr_cnt);

	keysock_vmem = vmem_create("keysock", (void *)1, MAXMIN, 1,
	    NULL, NULL, NULL, 1, VM_SLEEP | VMC_IDENTIFIER);

	/*
	 * We want to be informed each time a stack is created or
	 * destroyed in the kernel, so we can maintain the
	 * set of keysock_stack_t's.
	 */
	netstack_register(NS_KEYSOCK, keysock_stack_init, NULL,
	    keysock_stack_fini);

	return (B_TRUE);
}

/*
 * Walk through the param array specified registering each element with the
 * named dispatch handler.
 */
static boolean_t
keysock_param_register(IDP *ndp, keysockparam_t *ksp, int cnt)
{
	for (; cnt-- > 0; ksp++) {
		if (ksp->keysock_param_name != NULL &&
		    ksp->keysock_param_name[0]) {
			if (!nd_load(ndp,
			    ksp->keysock_param_name,
			    keysock_param_get, keysock_param_set,
			    (caddr_t)ksp)) {
				nd_free(ndp);
				return (B_FALSE);
			}
		}
	}
	return (B_TRUE);
}

/*
 * Initialize keysock for one stack instance
 */
/* ARGSUSED */
static void *
keysock_stack_init(netstackid_t stackid, netstack_t *ns)
{
	keysock_stack_t	*keystack;
	keysockparam_t *ksp;

	keystack = (keysock_stack_t *)kmem_zalloc(sizeof (*keystack), KM_SLEEP);
	keystack->keystack_netstack = ns;

	keystack->keystack_acquire_seq = 0xffffffff;

	ksp = (keysockparam_t *)kmem_alloc(sizeof (lcl_param_arr), KM_SLEEP);
	keystack->keystack_params = ksp;
	bcopy(lcl_param_arr, ksp, sizeof (lcl_param_arr));

	(void) keysock_param_register(&keystack->keystack_g_nd, ksp,
	    A_CNT(lcl_param_arr));

	mutex_init(&keystack->keystack_list_lock, NULL, MUTEX_DEFAULT, NULL);
	mutex_init(&keystack->keystack_consumers_lock,
	    NULL, MUTEX_DEFAULT, NULL);
	mutex_init(&keystack->keystack_param_lock, NULL, MUTEX_DEFAULT, NULL);
	return (keystack);
}

/*
 * Free NDD variable space, and other destructors, for keysock.
 */
void
keysock_ddi_destroy(void)
{
	netstack_unregister(NS_KEYSOCK);
	vmem_destroy(keysock_vmem);
}

/*
 * Remove one stack instance from keysock
 */
/* ARGSUSED */
static void
keysock_stack_fini(netstackid_t stackid, void *arg)
{
	keysock_stack_t *keystack = (keysock_stack_t *)arg;

	nd_free(&keystack->keystack_g_nd);
	kmem_free(keystack->keystack_params, sizeof (lcl_param_arr));
	keystack->keystack_params = NULL;

	mutex_destroy(&keystack->keystack_list_lock);
	mutex_destroy(&keystack->keystack_consumers_lock);
	mutex_destroy(&keystack->keystack_param_lock);

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

/*
 * Close routine for keysock.
 */
static int
keysock_close(queue_t *q)
{
	keysock_t *ks;
	keysock_consumer_t *kc;
	void *ptr = q->q_ptr;
	int size;
	keysock_stack_t	*keystack;


	qprocsoff(q);

	/* Safe assumption. */
	ASSERT(ptr != NULL);

	if (WR(q)->q_next) {
		kc = (keysock_consumer_t *)ptr;
		keystack = kc->kc_keystack;

		ks1dbg(keystack, ("Module close, removing a consumer (%d).\n",
		    kc->kc_sa_type));
		/*
		 * Because of PERMOD open/close exclusive perimeter, I
		 * can inspect KC_FLUSHING w/o locking down kc->kc_lock.
		 */
		if (kc->kc_flags & KC_FLUSHING) {
			/*
			 * If this decrement was the last one, send
			 * down the next pending one, if any.
			 *
			 * With a PERMOD perimeter, the mutexes ops aren't
			 * really necessary, but if we ever loosen up, we will
			 * have this bit covered already.
			 */
			keystack->keystack_flushdump--;
			if (keystack->keystack_flushdump == 0) {
				/*
				 * The flush/dump terminated by having a
				 * consumer go away.  I need to send up to the
				 * appropriate keysock all of the relevant
				 * information.  Unfortunately, I don't
				 * have that handy.
				 */
				ks0dbg(("Consumer went away while flushing or"
				    " dumping.\n"));
			}
		}
		size = sizeof (keysock_consumer_t);
		mutex_enter(&keystack->keystack_consumers_lock);
		keystack->keystack_consumers[kc->kc_sa_type] = NULL;
		mutex_exit(&keystack->keystack_consumers_lock);
		mutex_destroy(&kc->kc_lock);
		netstack_rele(kc->kc_keystack->keystack_netstack);
	} else {
		ks = (keysock_t *)ptr;
		keystack = ks->keysock_keystack;

		ks3dbg(keystack,
		    ("Driver close, PF_KEY socket is going away.\n"));
		if ((ks->keysock_flags & KEYSOCK_EXTENDED) != 0)
			atomic_add_32(&keystack->keystack_num_extended, -1);
		size = sizeof (keysock_t);
		mutex_enter(&keystack->keystack_list_lock);
		*(ks->keysock_ptpn) = ks->keysock_next;
		if (ks->keysock_next != NULL)
			ks->keysock_next->keysock_ptpn = ks->keysock_ptpn;
		mutex_exit(&keystack->keystack_list_lock);
		mutex_destroy(&ks->keysock_lock);
		vmem_free(keysock_vmem, (void *)(uintptr_t)ks->keysock_serial,
		    1);
		netstack_rele(ks->keysock_keystack->keystack_netstack);
	}

	/* Now I'm free. */
	kmem_free(ptr, size);
	return (0);
}
/*
 * Open routine for keysock.
 */
/* ARGSUSED */
static int
keysock_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
{
	keysock_t *ks;
	keysock_consumer_t *kc;
	mblk_t *mp;
	ipsec_info_t *ii;
	netstack_t *ns;
	keysock_stack_t *keystack;

	if (secpolicy_ip_config(credp, B_FALSE) != 0) {
		/* Privilege debugging will log the error */
		return (EPERM);
	}

	if (q->q_ptr != NULL)
		return (0);  /* Re-open of an already open instance. */

	ns = netstack_find_by_cred(credp);
	ASSERT(ns != NULL);
	keystack = ns->netstack_keysock;
	ASSERT(keystack != NULL);

	ks3dbg(keystack, ("Entering keysock open.\n"));

	if (keystack->keystack_plumbed < 1) {
		netstack_t *ns = keystack->keystack_netstack;

		keystack->keystack_plumbed = 0;
#ifdef NS_DEBUG
		printf("keysock_open(%d) - plumb\n",
		    keystack->keystack_netstack->netstack_stackid);
#endif
		/*
		 * Don't worry about ipsec_failure being true here.
		 * (See ip.c).  An open of keysock should try and force
		 * the issue.  Maybe it was a transient failure.
		 */
		ipsec_loader_loadnow(ns->netstack_ipsec);
	}

	if (sflag & MODOPEN) {
		/* Initialize keysock_consumer state here. */
		kc = kmem_zalloc(sizeof (keysock_consumer_t), KM_NOSLEEP);
		if (kc == NULL) {
			netstack_rele(keystack->keystack_netstack);
			return (ENOMEM);
		}
		mutex_init(&kc->kc_lock, NULL, MUTEX_DEFAULT, 0);
		kc->kc_rq = q;
		kc->kc_wq = WR(q);

		q->q_ptr = kc;
		WR(q)->q_ptr = kc;

		kc->kc_keystack = keystack;
		qprocson(q);

		/*
		 * Send down initial message to whatever I was pushed on top
		 * of asking for its consumer type.  The reply will set it.
		 */

		/* Allocate it. */
		mp = allocb(sizeof (ipsec_info_t), BPRI_HI);
		if (mp == NULL) {
			ks1dbg(keystack, (
			    "keysock_open:  Cannot allocate KEYSOCK_HELLO.\n"));
			/* Do I need to set these to null? */
			q->q_ptr = NULL;
			WR(q)->q_ptr = NULL;
			mutex_destroy(&kc->kc_lock);
			kmem_free(kc, sizeof (*kc));
			netstack_rele(keystack->keystack_netstack);
			return (ENOMEM);
		}

		/* If I allocated okay, putnext to what I was pushed atop. */
		mp->b_wptr += sizeof (ipsec_info_t);
		mp->b_datap->db_type = M_CTL;
		ii = (ipsec_info_t *)mp->b_rptr;
		ii->ipsec_info_type = KEYSOCK_HELLO;
		/* Length only of type/len. */
		ii->ipsec_info_len = sizeof (ii->ipsec_allu);
		ks2dbg(keystack, ("Ready to putnext KEYSOCK_HELLO.\n"));
		putnext(kc->kc_wq, mp);
	} else {
		minor_t ksminor;

		/* Initialize keysock state. */

		ks2dbg(keystack, ("Made it into PF_KEY socket open.\n"));

		ksminor = (minor_t)(uintptr_t)
		    vmem_alloc(keysock_vmem, 1, VM_NOSLEEP);
		if (ksminor == 0) {
			netstack_rele(keystack->keystack_netstack);
			return (ENOMEM);
		}
		ks = kmem_zalloc(sizeof (keysock_t), KM_NOSLEEP);
		if (ks == NULL) {
			vmem_free(keysock_vmem, (void *)(uintptr_t)ksminor, 1);
			netstack_rele(keystack->keystack_netstack);
			return (ENOMEM);
		}

		mutex_init(&ks->keysock_lock, NULL, MUTEX_DEFAULT, 0);
		ks->keysock_rq = q;
		ks->keysock_wq = WR(q);
		ks->keysock_state = TS_UNBND;
		ks->keysock_serial = ksminor;

		q->q_ptr = ks;
		WR(q)->q_ptr = ks;
		ks->keysock_keystack = keystack;

		/*
		 * The receive hiwat is only looked at on the stream head
		 * queue.  Store in q_hiwat in order to return on SO_RCVBUF
		 * getsockopts.
		 */

		q->q_hiwat = keystack->keystack_recv_hiwat;

		/*
		 * The transmit hiwat/lowat is only looked at on IP's queue.
		 * Store in q_hiwat/q_lowat in order to return on
		 * SO_SNDBUF/SO_SNDLOWAT getsockopts.
		 */

		WR(q)->q_hiwat = keystack->keystack_xmit_hiwat;
		WR(q)->q_lowat = keystack->keystack_xmit_lowat;

		*devp = makedevice(getmajor(*devp), ksminor);

		/*
		 * Thread keysock into the global keysock list.
		 */
		mutex_enter(&keystack->keystack_list_lock);
		ks->keysock_next = keystack->keystack_list;
		ks->keysock_ptpn = &keystack->keystack_list;
		if (keystack->keystack_list != NULL) {
			keystack->keystack_list->keysock_ptpn =
			    &ks->keysock_next;
		}
		keystack->keystack_list = ks;
		mutex_exit(&keystack->keystack_list_lock);

		qprocson(q);
		(void) mi_set_sth_hiwat(q, keystack->keystack_recv_hiwat);
		/*
		 * Wait outside the keysock module perimeter for IPsec
		 * plumbing to be completed.  If it fails, keysock_close()
		 * undoes everything we just did.
		 */
		if (!ipsec_loader_wait(q,
		    keystack->keystack_netstack->netstack_ipsec)) {
			(void) keysock_close(q);
			return (EPFNOSUPPORT);
		}
	}

	return (0);
}

/* BELOW THIS LINE ARE ROUTINES INCLUDING AND RELATED TO keysock_wput(). */

/*
 * Copy relevant state bits.
 */
static void
keysock_copy_info(struct T_info_ack *tap, keysock_t *ks)
{
	*tap = keysock_g_t_info_ack;
	tap->CURRENT_state = ks->keysock_state;
	tap->OPT_size = keysock_max_optsize;
}

/*
 * This routine responds to T_CAPABILITY_REQ messages.  It is called by
 * keysock_wput.  Much of the T_CAPABILITY_ACK information is copied from
 * keysock_g_t_info_ack.  The current state of the stream is copied from
 * keysock_state.
 */
static void
keysock_capability_req(queue_t *q, mblk_t *mp)
{
	keysock_t *ks = (keysock_t *)q->q_ptr;
	t_uscalar_t cap_bits1;
	struct T_capability_ack	*tcap;

	cap_bits1 = ((struct T_capability_req *)mp->b_rptr)->CAP_bits1;

	mp = tpi_ack_alloc(mp, sizeof (struct T_capability_ack),
	    mp->b_datap->db_type, T_CAPABILITY_ACK);
	if (mp == NULL)
		return;

	tcap = (struct T_capability_ack *)mp->b_rptr;
	tcap->CAP_bits1 = 0;

	if (cap_bits1 & TC1_INFO) {
		keysock_copy_info(&tcap->INFO_ack, ks);
		tcap->CAP_bits1 |= TC1_INFO;
	}

	qreply(q, mp);
}

/*
 * This routine responds to T_INFO_REQ messages. It is called by
 * keysock_wput_other.
 * Most of the T_INFO_ACK information is copied from keysock_g_t_info_ack.
 * The current state of the stream is copied from keysock_state.
 */
static void
keysock_info_req(q, mp)
	queue_t	*q;
	mblk_t	*mp;
{
	mp = tpi_ack_alloc(mp, sizeof (struct T_info_ack), M_PCPROTO,
	    T_INFO_ACK);
	if (mp == NULL)
		return;
	keysock_copy_info((struct T_info_ack *)mp->b_rptr,
	    (keysock_t *)q->q_ptr);
	qreply(q, mp);
}

/*
 * keysock_err_ack. This routine creates a
 * T_ERROR_ACK message and passes it
 * upstream.
 */
static void
keysock_err_ack(q, mp, t_error, sys_error)
	queue_t	*q;
	mblk_t	*mp;
	int	t_error;
	int	sys_error;
{
	if ((mp = mi_tpi_err_ack_alloc(mp, t_error, sys_error)) != NULL)
		qreply(q, mp);
}

/*
 * This routine retrieves the current status of socket options.
 * It returns the size of the option retrieved.
 */
/* ARGSUSED */
int
keysock_opt_get(queue_t *q, int level, int name, uchar_t *ptr)
{
	int *i1 = (int *)ptr;
	keysock_t *ks = (keysock_t *)q->q_ptr;

	switch (level) {
	case SOL_SOCKET:
		mutex_enter(&ks->keysock_lock);
		switch (name) {
		case SO_TYPE:
			*i1 = SOCK_RAW;
			break;
		case SO_USELOOPBACK:
			*i1 = (int)(!((ks->keysock_flags & KEYSOCK_NOLOOP) ==
			    KEYSOCK_NOLOOP));
			break;
		/*
		 * The following two items can be manipulated,
		 * but changing them should do nothing.
		 */
		case SO_SNDBUF:
			*i1 = (int)q->q_hiwat;
			break;
		case SO_RCVBUF:
			*i1 = (int)(RD(q)->q_hiwat);
			break;
		}
		mutex_exit(&ks->keysock_lock);
		break;
	default:
		return (0);
	}
	return (sizeof (int));
}

/*
 * This routine sets socket options.
 */
/* ARGSUSED */
int
keysock_opt_set(queue_t *q, uint_t mgmt_flags, int level,
    int name, uint_t inlen, uchar_t *invalp, uint_t *outlenp,
    uchar_t *outvalp, void *thisdg_attrs, cred_t *cr, mblk_t *mblk)
{
	int *i1 = (int *)invalp;
	keysock_t *ks = (keysock_t *)q->q_ptr;
	keysock_stack_t	*keystack = ks->keysock_keystack;

	switch (level) {
	case SOL_SOCKET:
		mutex_enter(&ks->keysock_lock);
		switch (name) {
		case SO_USELOOPBACK:
			if (!(*i1))
				ks->keysock_flags |= KEYSOCK_NOLOOP;
			else ks->keysock_flags &= ~KEYSOCK_NOLOOP;
			break;
		case SO_SNDBUF:
			if (*i1 > keystack->keystack_max_buf)
				return (ENOBUFS);
			q->q_hiwat = *i1;
			break;
		case SO_RCVBUF:
			if (*i1 > keystack->keystack_max_buf)
				return (ENOBUFS);
			RD(q)->q_hiwat = *i1;
			(void) mi_set_sth_hiwat(RD(q), *i1);
			break;
		}
		mutex_exit(&ks->keysock_lock);
		break;
	}
	return (0);
}

/*
 * Handle STREAMS messages.
 */
static void
keysock_wput_other(queue_t *q, mblk_t *mp)
{
	struct iocblk *iocp;
	int error;
	keysock_t *ks = (keysock_t *)q->q_ptr;
	keysock_stack_t	*keystack = ks->keysock_keystack;
	cred_t		*cr;

	switch (mp->b_datap->db_type) {
	case M_PROTO:
	case M_PCPROTO:
		if ((mp->b_wptr - mp->b_rptr) < sizeof (long)) {
			ks3dbg(keystack, (
			    "keysock_wput_other: Not big enough M_PROTO\n"));
			freemsg(mp);
			return;
		}
		cr = zone_get_kcred(netstackid_to_zoneid(
		    keystack->keystack_netstack->netstack_stackid));
		ASSERT(cr != NULL);

		switch (((union T_primitives *)mp->b_rptr)->type) {
		case T_CAPABILITY_REQ:
			keysock_capability_req(q, mp);
			break;
		case T_INFO_REQ:
			keysock_info_req(q, mp);
			break;
		case T_SVR4_OPTMGMT_REQ:
			(void) svr4_optcom_req(q, mp, DB_CREDDEF(mp, cr),
			    &keysock_opt_obj, B_FALSE);
			break;
		case T_OPTMGMT_REQ:
			(void) tpi_optcom_req(q, mp, DB_CREDDEF(mp, cr),
			    &keysock_opt_obj, B_FALSE);
			break;
		case T_DATA_REQ:
		case T_EXDATA_REQ:
		case T_ORDREL_REQ:
			/* Illegal for keysock. */
			freemsg(mp);
			(void) putnextctl1(RD(q), M_ERROR, EPROTO);
			break;
		default:
			/* Not supported by keysock. */
			keysock_err_ack(q, mp, TNOTSUPPORT, 0);
			break;
		}
		crfree(cr);
		return;
	case M_IOCTL:
		iocp = (struct iocblk *)mp->b_rptr;
		error = EINVAL;

		switch (iocp->ioc_cmd) {
		case ND_SET:
		case ND_GET:
			if (nd_getset(q, keystack->keystack_g_nd, mp)) {
				qreply(q, mp);
				return;
			} else
				error = ENOENT;
			/* FALLTHRU */
		default:
			miocnak(q, mp, 0, error);
			return;
		}
	case M_FLUSH:
		if (*mp->b_rptr & FLUSHW) {
			flushq(q, FLUSHALL);
			*mp->b_rptr &= ~FLUSHW;
		}
		if (*mp->b_rptr & FLUSHR) {
			qreply(q, mp);
			return;
		}
		/* Else FALLTHRU */
	}

	/* If fell through, just black-hole the message. */
	freemsg(mp);
}

/*
 * Transmit a PF_KEY error message to the instance either pointed to
 * by ks, the instance with serial number serial, or more, depending.
 *
 * The faulty message (or a reasonable facsimile thereof) is in mp.
 * This function will free mp or recycle it for delivery, thereby causing
 * the stream head to free it.
 */
static void
keysock_error(keysock_t *ks, mblk_t *mp, int error, int diagnostic)
{
	sadb_msg_t *samsg = (sadb_msg_t *)mp->b_rptr;
	keysock_stack_t	*keystack = ks->keysock_keystack;

	ASSERT(mp->b_datap->db_type == M_DATA);

	if (samsg->sadb_msg_type < SADB_GETSPI ||
	    samsg->sadb_msg_type > SADB_MAX)
		samsg->sadb_msg_type = SADB_RESERVED;

	/*
	 * Strip out extension headers.
	 */
	ASSERT(mp->b_rptr + sizeof (*samsg) <= mp->b_datap->db_lim);
	mp->b_wptr = mp->b_rptr + sizeof (*samsg);
	samsg->sadb_msg_len = SADB_8TO64(sizeof (sadb_msg_t));
	samsg->sadb_msg_errno = (uint8_t)error;
	samsg->sadb_x_msg_diagnostic = (uint16_t)diagnostic;

	keysock_passup(mp, samsg, ks->keysock_serial, NULL, B_FALSE, keystack);
}

/*
 * Pass down a message to a consumer.  Wrap it in KEYSOCK_IN, and copy
 * in the extv if passed in.
 */
static void
keysock_passdown(keysock_t *ks, mblk_t *mp, uint8_t satype, sadb_ext_t *extv[],
    boolean_t flushmsg)
{
	keysock_consumer_t *kc;
	mblk_t *wrapper;
	keysock_in_t *ksi;
	int i;
	keysock_stack_t	*keystack = ks->keysock_keystack;

	wrapper = allocb(sizeof (ipsec_info_t), BPRI_HI);
	if (wrapper == NULL) {
		ks3dbg(keystack, ("keysock_passdown: allocb failed.\n"));
		if (extv[SADB_EXT_KEY_ENCRYPT] != NULL)
			bzero(extv[SADB_EXT_KEY_ENCRYPT],
			    SADB_64TO8(
			    extv[SADB_EXT_KEY_ENCRYPT]->sadb_ext_len));
		if (extv[SADB_EXT_KEY_AUTH] != NULL)
			bzero(extv[SADB_EXT_KEY_AUTH],
			    SADB_64TO8(
			    extv[SADB_EXT_KEY_AUTH]->sadb_ext_len));
		if (flushmsg) {
			ks0dbg((
			    "keysock: Downwards flush/dump message failed!\n"));
			/* If this is true, I hold the perimeter. */