xref: /illumos-gate/usr/src/uts/common/inet/ip/ipsecesp.c (revision 069e6b7e31ba5dcbc5441b98af272714d9a5455c)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2010 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  * Copyright (c) 2012 Nexenta Systems, Inc. All rights reserved.
25  * Copyright (c) 2017 Joyent, Inc.
26  */
27 
28 #include <sys/types.h>
29 #include <sys/stream.h>
30 #include <sys/stropts.h>
31 #include <sys/errno.h>
32 #include <sys/strlog.h>
33 #include <sys/tihdr.h>
34 #include <sys/socket.h>
35 #include <sys/ddi.h>
36 #include <sys/sunddi.h>
37 #include <sys/kmem.h>
38 #include <sys/zone.h>
39 #include <sys/sysmacros.h>
40 #include <sys/cmn_err.h>
41 #include <sys/vtrace.h>
42 #include <sys/debug.h>
43 #include <sys/atomic.h>
44 #include <sys/strsun.h>
45 #include <sys/random.h>
46 #include <netinet/in.h>
47 #include <net/if.h>
48 #include <netinet/ip6.h>
49 #include <net/pfkeyv2.h>
50 #include <net/pfpolicy.h>
51 
52 #include <inet/common.h>
53 #include <inet/mi.h>
54 #include <inet/nd.h>
55 #include <inet/ip.h>
56 #include <inet/ip_impl.h>
57 #include <inet/ip6.h>
58 #include <inet/ip_if.h>
59 #include <inet/ip_ndp.h>
60 #include <inet/sadb.h>
61 #include <inet/ipsec_info.h>
62 #include <inet/ipsec_impl.h>
63 #include <inet/ipsecesp.h>
64 #include <inet/ipdrop.h>
65 #include <inet/tcp.h>
66 #include <sys/kstat.h>
67 #include <sys/policy.h>
68 #include <sys/strsun.h>
69 #include <sys/strsubr.h>
70 #include <inet/udp_impl.h>
71 #include <sys/taskq.h>
72 #include <sys/note.h>
73 
74 #include <sys/tsol/tnet.h>
75 
76 /*
77  * Table of ND variables supported by ipsecesp. These are loaded into
78  * ipsecesp_g_nd in ipsecesp_init_nd.
79  * All of these are alterable, within the min/max values given, at run time.
80  */
81 static	ipsecespparam_t	lcl_param_arr[] = {
82 	/* min	max			value	name */
83 	{ 0,	3,			0,	"ipsecesp_debug"},
84 	{ 125,	32000, SADB_AGE_INTERVAL_DEFAULT, "ipsecesp_age_interval"},
85 	{ 1,	10,			1,	"ipsecesp_reap_delay"},
86 	{ 1,	SADB_MAX_REPLAY,	64,	"ipsecesp_replay_size"},
87 	{ 1,	300,			15,	"ipsecesp_acquire_timeout"},
88 	{ 1,	1800,			90,	"ipsecesp_larval_timeout"},
89 	/* Default lifetime values for ACQUIRE messages. */
90 	{ 0,	0xffffffffU,	0,	"ipsecesp_default_soft_bytes"},
91 	{ 0,	0xffffffffU,	0,	"ipsecesp_default_hard_bytes"},
92 	{ 0,	0xffffffffU,	24000,	"ipsecesp_default_soft_addtime"},
93 	{ 0,	0xffffffffU,	28800,	"ipsecesp_default_hard_addtime"},
94 	{ 0,	0xffffffffU,	0,	"ipsecesp_default_soft_usetime"},
95 	{ 0,	0xffffffffU,	0,	"ipsecesp_default_hard_usetime"},
96 	{ 0,	1,		0,	"ipsecesp_log_unknown_spi"},
97 	{ 0,	2,		1,	"ipsecesp_padding_check"},
98 	{ 0,	600,		20,	"ipsecesp_nat_keepalive_interval"},
99 };
100 /* For ipsecesp_nat_keepalive_interval, see ipsecesp.h. */
101 
102 #define	esp0dbg(a)	printf a
103 /* NOTE:  != 0 instead of > 0 so lint doesn't complain. */
104 #define	esp1dbg(espstack, a)	if (espstack->ipsecesp_debug != 0) printf a
105 #define	esp2dbg(espstack, a)	if (espstack->ipsecesp_debug > 1) printf a
106 #define	esp3dbg(espstack, a)	if (espstack->ipsecesp_debug > 2) printf a
107 
108 static int ipsecesp_open(queue_t *, dev_t *, int, int, cred_t *);
109 static int ipsecesp_close(queue_t *, int, cred_t *);
110 static int ipsecesp_rput(queue_t *, mblk_t *);
111 static int ipsecesp_wput(queue_t *, mblk_t *);
112 static void	*ipsecesp_stack_init(netstackid_t stackid, netstack_t *ns);
113 static void	ipsecesp_stack_fini(netstackid_t stackid, void *arg);
114 
115 static void esp_prepare_udp(netstack_t *, mblk_t *, ipha_t *);
116 static void esp_outbound_finish(mblk_t *, ip_xmit_attr_t *);
117 static void esp_inbound_restart(mblk_t *, ip_recv_attr_t *);
118 
119 static boolean_t esp_register_out(uint32_t, uint32_t, uint_t,
120     ipsecesp_stack_t *, cred_t *);
121 static boolean_t esp_strip_header(mblk_t *, boolean_t, uint32_t,
122     kstat_named_t **, ipsecesp_stack_t *);
123 static mblk_t *esp_submit_req_inbound(mblk_t *, ip_recv_attr_t *,
124     ipsa_t *, uint_t);
125 static mblk_t *esp_submit_req_outbound(mblk_t *, ip_xmit_attr_t *,
126     ipsa_t *, uchar_t *, uint_t);
127 
128 /* Setable in /etc/system */
129 uint32_t esp_hash_size = IPSEC_DEFAULT_HASH_SIZE;
130 
131 static struct module_info info = {
132 	5137, "ipsecesp", 0, INFPSZ, 65536, 1024
133 };
134 
135 static struct qinit rinit = {
136 	ipsecesp_rput, NULL, ipsecesp_open, ipsecesp_close, NULL, &info,
137 	NULL
138 };
139 
140 static struct qinit winit = {
141 	ipsecesp_wput, NULL, ipsecesp_open, ipsecesp_close, NULL, &info,
142 	NULL
143 };
144 
145 struct streamtab ipsecespinfo = {
146 	&rinit, &winit, NULL, NULL
147 };
148 
149 static taskq_t *esp_taskq;
150 
151 /*
152  * OTOH, this one is set at open/close, and I'm D_MTQPAIR for now.
153  *
154  * Question:	Do I need this, given that all instance's esps->esps_wq point
155  *		to IP?
156  *
157  * Answer:	Yes, because I need to know which queue is BOUND to
158  *		IPPROTO_ESP
159  */
160 
161 static int	esp_kstat_update(kstat_t *, int);
162 
163 static boolean_t
164 esp_kstat_init(ipsecesp_stack_t *espstack, netstackid_t stackid)
165 {
166 	espstack->esp_ksp = kstat_create_netstack("ipsecesp", 0, "esp_stat",
167 	    "net", KSTAT_TYPE_NAMED,
168 	    sizeof (esp_kstats_t) / sizeof (kstat_named_t), 0, stackid);
169 
170 	if (espstack->esp_ksp == NULL || espstack->esp_ksp->ks_data == NULL)
171 		return (B_FALSE);
172 
173 	espstack->esp_kstats = espstack->esp_ksp->ks_data;
174 
175 	espstack->esp_ksp->ks_update = esp_kstat_update;
176 	espstack->esp_ksp->ks_private = (void *)(uintptr_t)stackid;
177 
178 #define	K64 KSTAT_DATA_UINT64
179 #define	KI(x) kstat_named_init(&(espstack->esp_kstats->esp_stat_##x), #x, K64)
180 
181 	KI(num_aalgs);
182 	KI(num_ealgs);
183 	KI(good_auth);
184 	KI(bad_auth);
185 	KI(bad_padding);
186 	KI(replay_failures);
187 	KI(replay_early_failures);
188 	KI(keysock_in);
189 	KI(out_requests);
190 	KI(acquire_requests);
191 	KI(bytes_expired);
192 	KI(out_discards);
193 	KI(crypto_sync);
194 	KI(crypto_async);
195 	KI(crypto_failures);
196 	KI(bad_decrypt);
197 	KI(sa_port_renumbers);
198 
199 #undef KI
200 #undef K64
201 
202 	kstat_install(espstack->esp_ksp);
203 
204 	return (B_TRUE);
205 }
206 
207 static int
208 esp_kstat_update(kstat_t *kp, int rw)
209 {
210 	esp_kstats_t *ekp;
211 	netstackid_t	stackid;
212 	netstack_t	*ns;
213 	ipsec_stack_t	*ipss;
214 
215 	if ((kp == NULL) || (kp->ks_data == NULL))
216 		return (EIO);
217 
218 	if (rw == KSTAT_WRITE)
219 		return (EACCES);
220 
221 	stackid = (zoneid_t)(uintptr_t)kp->ks_private;
222 	ns = netstack_find_by_stackid(stackid);
223 	if (ns == NULL)
224 		return (-1);
225 	ipss = ns->netstack_ipsec;
226 	if (ipss == NULL) {
227 		netstack_rele(ns);
228 		return (-1);
229 	}
230 	ekp = (esp_kstats_t *)kp->ks_data;
231 
232 	rw_enter(&ipss->ipsec_alg_lock, RW_READER);
233 	ekp->esp_stat_num_aalgs.value.ui64 =
234 	    ipss->ipsec_nalgs[IPSEC_ALG_AUTH];
235 	ekp->esp_stat_num_ealgs.value.ui64 =
236 	    ipss->ipsec_nalgs[IPSEC_ALG_ENCR];
237 	rw_exit(&ipss->ipsec_alg_lock);
238 
239 	netstack_rele(ns);
240 	return (0);
241 }
242 
243 #ifdef DEBUG
244 /*
245  * Debug routine, useful to see pre-encryption data.
246  */
247 static char *
248 dump_msg(mblk_t *mp)
249 {
250 	char tmp_str[3], tmp_line[256];
251 
252 	while (mp != NULL) {
253 		unsigned char *ptr;
254 
255 		printf("mblk address 0x%p, length %ld, db_ref %d "
256 		    "type %d, base 0x%p, lim 0x%p\n",
257 		    (void *) mp, (long)(mp->b_wptr - mp->b_rptr),
258 		    mp->b_datap->db_ref, mp->b_datap->db_type,
259 		    (void *)mp->b_datap->db_base, (void *)mp->b_datap->db_lim);
260 		ptr = mp->b_rptr;
261 
262 		tmp_line[0] = '\0';
263 		while (ptr < mp->b_wptr) {
264 			uint_t diff;
265 
266 			diff = (ptr - mp->b_rptr);
267 			if (!(diff & 0x1f)) {
268 				if (strlen(tmp_line) > 0) {
269 					printf("bytes: %s\n", tmp_line);
270 					tmp_line[0] = '\0';
271 				}
272 			}
273 			if (!(diff & 0x3))
274 				(void) strcat(tmp_line, " ");
275 			(void) sprintf(tmp_str, "%02x", *ptr);
276 			(void) strcat(tmp_line, tmp_str);
277 			ptr++;
278 		}
279 		if (strlen(tmp_line) > 0)
280 			printf("bytes: %s\n", tmp_line);
281 
282 		mp = mp->b_cont;
283 	}
284 
285 	return ("\n");
286 }
287 
288 #else /* DEBUG */
289 static char *
290 dump_msg(mblk_t *mp)
291 {
292 	printf("Find value of mp %p.\n", mp);
293 	return ("\n");
294 }
295 #endif /* DEBUG */
296 
297 /*
298  * Don't have to lock age_interval, as only one thread will access it at
299  * a time, because I control the one function that does with timeout().
300  */
301 static void
302 esp_ager(void *arg)
303 {
304 	ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)arg;
305 	netstack_t	*ns = espstack->ipsecesp_netstack;
306 	hrtime_t begin = gethrtime();
307 
308 	sadb_ager(&espstack->esp_sadb.s_v4, espstack->esp_pfkey_q,
309 	    espstack->ipsecesp_reap_delay, ns);
310 	sadb_ager(&espstack->esp_sadb.s_v6, espstack->esp_pfkey_q,
311 	    espstack->ipsecesp_reap_delay, ns);
312 
313 	espstack->esp_event = sadb_retimeout(begin, espstack->esp_pfkey_q,
314 	    esp_ager, espstack,
315 	    &espstack->ipsecesp_age_interval, espstack->ipsecesp_age_int_max,
316 	    info.mi_idnum);
317 }
318 
319 /*
320  * Get an ESP NDD parameter.
321  */
322 /* ARGSUSED */
323 static int
324 ipsecesp_param_get(
325     queue_t	*q,
326     mblk_t	*mp,
327     caddr_t	cp,
328     cred_t *cr)
329 {
330 	ipsecespparam_t	*ipsecesppa = (ipsecespparam_t *)cp;
331 	uint_t value;
332 	ipsecesp_stack_t	*espstack = (ipsecesp_stack_t *)q->q_ptr;
333 
334 	mutex_enter(&espstack->ipsecesp_param_lock);
335 	value = ipsecesppa->ipsecesp_param_value;
336 	mutex_exit(&espstack->ipsecesp_param_lock);
337 
338 	(void) mi_mpprintf(mp, "%u", value);
339 	return (0);
340 }
341 
342 /*
343  * This routine sets an NDD variable in a ipsecespparam_t structure.
344  */
345 /* ARGSUSED */
346 static int
347 ipsecesp_param_set(
348     queue_t	*q,
349     mblk_t	*mp,
350     char	*value,
351     caddr_t	cp,
352     cred_t *cr)
353 {
354 	ulong_t	new_value;
355 	ipsecespparam_t	*ipsecesppa = (ipsecespparam_t *)cp;
356 	ipsecesp_stack_t	*espstack = (ipsecesp_stack_t *)q->q_ptr;
357 
358 	/*
359 	 * Fail the request if the new value does not lie within the
360 	 * required bounds.
361 	 */
362 	if (ddi_strtoul(value, NULL, 10, &new_value) != 0 ||
363 	    new_value < ipsecesppa->ipsecesp_param_min ||
364 	    new_value > ipsecesppa->ipsecesp_param_max) {
365 		return (EINVAL);
366 	}
367 
368 	/* Set the new value */
369 	mutex_enter(&espstack->ipsecesp_param_lock);
370 	ipsecesppa->ipsecesp_param_value = new_value;
371 	mutex_exit(&espstack->ipsecesp_param_lock);
372 	return (0);
373 }
374 
375 /*
376  * Using lifetime NDD variables, fill in an extended combination's
377  * lifetime information.
378  */
379 void
380 ipsecesp_fill_defs(sadb_x_ecomb_t *ecomb, netstack_t *ns)
381 {
382 	ipsecesp_stack_t	*espstack = ns->netstack_ipsecesp;
383 
384 	ecomb->sadb_x_ecomb_soft_bytes = espstack->ipsecesp_default_soft_bytes;
385 	ecomb->sadb_x_ecomb_hard_bytes = espstack->ipsecesp_default_hard_bytes;
386 	ecomb->sadb_x_ecomb_soft_addtime =
387 	    espstack->ipsecesp_default_soft_addtime;
388 	ecomb->sadb_x_ecomb_hard_addtime =
389 	    espstack->ipsecesp_default_hard_addtime;
390 	ecomb->sadb_x_ecomb_soft_usetime =
391 	    espstack->ipsecesp_default_soft_usetime;
392 	ecomb->sadb_x_ecomb_hard_usetime =
393 	    espstack->ipsecesp_default_hard_usetime;
394 }
395 
396 /*
397  * Initialize things for ESP at module load time.
398  */
399 boolean_t
400 ipsecesp_ddi_init(void)
401 {
402 	esp_taskq = taskq_create("esp_taskq", 1, minclsyspri,
403 	    IPSEC_TASKQ_MIN, IPSEC_TASKQ_MAX, 0);
404 
405 	/*
406 	 * We want to be informed each time a stack is created or
407 	 * destroyed in the kernel, so we can maintain the
408 	 * set of ipsecesp_stack_t's.
409 	 */
410 	netstack_register(NS_IPSECESP, ipsecesp_stack_init, NULL,
411 	    ipsecesp_stack_fini);
412 
413 	return (B_TRUE);
414 }
415 
416 /*
417  * Walk through the param array specified registering each element with the
418  * named dispatch handler.
419  */
420 static boolean_t
421 ipsecesp_param_register(IDP *ndp, ipsecespparam_t *espp, int cnt)
422 {
423 	for (; cnt-- > 0; espp++) {
424 		if (espp->ipsecesp_param_name != NULL &&
425 		    espp->ipsecesp_param_name[0]) {
426 			if (!nd_load(ndp,
427 			    espp->ipsecesp_param_name,
428 			    ipsecesp_param_get, ipsecesp_param_set,
429 			    (caddr_t)espp)) {
430 				nd_free(ndp);
431 				return (B_FALSE);
432 			}
433 		}
434 	}
435 	return (B_TRUE);
436 }
437 
438 /*
439  * Initialize things for ESP for each stack instance
440  */
441 static void *
442 ipsecesp_stack_init(netstackid_t stackid, netstack_t *ns)
443 {
444 	ipsecesp_stack_t	*espstack;
445 	ipsecespparam_t		*espp;
446 
447 	espstack = (ipsecesp_stack_t *)kmem_zalloc(sizeof (*espstack),
448 	    KM_SLEEP);
449 	espstack->ipsecesp_netstack = ns;
450 
451 	espp = (ipsecespparam_t *)kmem_alloc(sizeof (lcl_param_arr), KM_SLEEP);
452 	espstack->ipsecesp_params = espp;
453 	bcopy(lcl_param_arr, espp, sizeof (lcl_param_arr));
454 
455 	(void) ipsecesp_param_register(&espstack->ipsecesp_g_nd, espp,
456 	    A_CNT(lcl_param_arr));
457 
458 	(void) esp_kstat_init(espstack, stackid);
459 
460 	espstack->esp_sadb.s_acquire_timeout =
461 	    &espstack->ipsecesp_acquire_timeout;
462 	sadbp_init("ESP", &espstack->esp_sadb, SADB_SATYPE_ESP, esp_hash_size,
463 	    espstack->ipsecesp_netstack);
464 
465 	mutex_init(&espstack->ipsecesp_param_lock, NULL, MUTEX_DEFAULT, 0);
466 
467 	ip_drop_register(&espstack->esp_dropper, "IPsec ESP");
468 	return (espstack);
469 }
470 
471 /*
472  * Destroy things for ESP at module unload time.
473  */
474 void
475 ipsecesp_ddi_destroy(void)
476 {
477 	netstack_unregister(NS_IPSECESP);
478 	taskq_destroy(esp_taskq);
479 }
480 
481 /*
482  * Destroy things for ESP for one stack instance
483  */
484 static void
485 ipsecesp_stack_fini(netstackid_t stackid, void *arg)
486 {
487 	ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)arg;
488 
489 	if (espstack->esp_pfkey_q != NULL) {
490 		(void) quntimeout(espstack->esp_pfkey_q, espstack->esp_event);
491 	}
492 	espstack->esp_sadb.s_acquire_timeout = NULL;
493 	sadbp_destroy(&espstack->esp_sadb, espstack->ipsecesp_netstack);
494 	ip_drop_unregister(&espstack->esp_dropper);
495 	mutex_destroy(&espstack->ipsecesp_param_lock);
496 	nd_free(&espstack->ipsecesp_g_nd);
497 
498 	kmem_free(espstack->ipsecesp_params, sizeof (lcl_param_arr));
499 	espstack->ipsecesp_params = NULL;
500 	kstat_delete_netstack(espstack->esp_ksp, stackid);
501 	espstack->esp_ksp = NULL;
502 	espstack->esp_kstats = NULL;
503 	kmem_free(espstack, sizeof (*espstack));
504 }
505 
506 /*
507  * ESP module open routine, which is here for keysock plumbing.
508  * Keysock is pushed over {AH,ESP} which is an artifact from the Bad Old
509  * Days of export control, and fears that ESP would not be allowed
510  * to be shipped at all by default.  Eventually, keysock should
511  * either access AH and ESP via modstubs or krtld dependencies, or
512  * perhaps be folded in with AH and ESP into a single IPsec/netsec
513  * module ("netsec" if PF_KEY provides more than AH/ESP keying tables).
514  */
515 /* ARGSUSED */
516 static int
517 ipsecesp_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
518 {
519 	netstack_t		*ns;
520 	ipsecesp_stack_t	*espstack;
521 
522 	if (secpolicy_ip_config(credp, B_FALSE) != 0)
523 		return (EPERM);
524 
525 	if (q->q_ptr != NULL)
526 		return (0);  /* Re-open of an already open instance. */
527 
528 	if (sflag != MODOPEN)
529 		return (EINVAL);
530 
531 	ns = netstack_find_by_cred(credp);
532 	ASSERT(ns != NULL);
533 	espstack = ns->netstack_ipsecesp;
534 	ASSERT(espstack != NULL);
535 
536 	q->q_ptr = espstack;
537 	WR(q)->q_ptr = q->q_ptr;
538 
539 	qprocson(q);
540 	return (0);
541 }
542 
543 /*
544  * ESP module close routine.
545  */
546 /* ARGSUSED */
547 static int
548 ipsecesp_close(queue_t *q, int flags __unused, cred_t *credp __unused)
549 {
550 	ipsecesp_stack_t	*espstack = (ipsecesp_stack_t *)q->q_ptr;
551 
552 	/*
553 	 * Clean up q_ptr, if needed.
554 	 */
555 	qprocsoff(q);
556 
557 	/* Keysock queue check is safe, because of OCEXCL perimeter. */
558 
559 	if (q == espstack->esp_pfkey_q) {
560 		esp1dbg(espstack,
561 		    ("ipsecesp_close:  Ummm... keysock is closing ESP.\n"));
562 		espstack->esp_pfkey_q = NULL;
563 		/* Detach qtimeouts. */
564 		(void) quntimeout(q, espstack->esp_event);
565 	}
566 
567 	netstack_rele(espstack->ipsecesp_netstack);
568 	return (0);
569 }
570 
571 /*
572  * Add a number of bytes to what the SA has protected so far.  Return
573  * B_TRUE if the SA can still protect that many bytes.
574  *
575  * Caller must REFRELE the passed-in assoc.  This function must REFRELE
576  * any obtained peer SA.
577  */
578 static boolean_t
579 esp_age_bytes(ipsa_t *assoc, uint64_t bytes, boolean_t inbound)
580 {
581 	ipsa_t *inassoc, *outassoc;
582 	isaf_t *bucket;
583 	boolean_t inrc, outrc, isv6;
584 	sadb_t *sp;
585 	int outhash;
586 	netstack_t		*ns = assoc->ipsa_netstack;
587 	ipsecesp_stack_t	*espstack = ns->netstack_ipsecesp;
588 
589 	/* No peer?  No problem! */
590 	if (!assoc->ipsa_haspeer) {
591 		return (sadb_age_bytes(espstack->esp_pfkey_q, assoc, bytes,
592 		    B_TRUE));
593 	}
594 
595 	/*
596 	 * Otherwise, we want to grab both the original assoc and its peer.
597 	 * There might be a race for this, but if it's a real race, two
598 	 * expire messages may occur.  We limit this by only sending the
599 	 * expire message on one of the peers, we'll pick the inbound
600 	 * arbitrarily.
601 	 *
602 	 * If we need tight synchronization on the peer SA, then we need to
603 	 * reconsider.
604 	 */
605 
606 	/* Use address length to select IPv6/IPv4 */
607 	isv6 = (assoc->ipsa_addrfam == AF_INET6);
608 	sp = isv6 ? &espstack->esp_sadb.s_v6 : &espstack->esp_sadb.s_v4;
609 
610 	if (inbound) {
611 		inassoc = assoc;
612 		if (isv6) {
613 			outhash = OUTBOUND_HASH_V6(sp, *((in6_addr_t *)
614 			    &inassoc->ipsa_dstaddr));
615 		} else {
616 			outhash = OUTBOUND_HASH_V4(sp, *((ipaddr_t *)
617 			    &inassoc->ipsa_dstaddr));
618 		}
619 		bucket = &sp->sdb_of[outhash];
620 		mutex_enter(&bucket->isaf_lock);
621 		outassoc = ipsec_getassocbyspi(bucket, inassoc->ipsa_spi,
622 		    inassoc->ipsa_srcaddr, inassoc->ipsa_dstaddr,
623 		    inassoc->ipsa_addrfam);
624 		mutex_exit(&bucket->isaf_lock);
625 		if (outassoc == NULL) {
626 			/* Q: Do we wish to set haspeer == B_FALSE? */
627 			esp0dbg(("esp_age_bytes: "
628 			    "can't find peer for inbound.\n"));
629 			return (sadb_age_bytes(espstack->esp_pfkey_q, inassoc,
630 			    bytes, B_TRUE));
631 		}
632 	} else {
633 		outassoc = assoc;
634 		bucket = INBOUND_BUCKET(sp, outassoc->ipsa_spi);
635 		mutex_enter(&bucket->isaf_lock);
636 		inassoc = ipsec_getassocbyspi(bucket, outassoc->ipsa_spi,
637 		    outassoc->ipsa_srcaddr, outassoc->ipsa_dstaddr,
638 		    outassoc->ipsa_addrfam);
639 		mutex_exit(&bucket->isaf_lock);
640 		if (inassoc == NULL) {
641 			/* Q: Do we wish to set haspeer == B_FALSE? */
642 			esp0dbg(("esp_age_bytes: "
643 			    "can't find peer for outbound.\n"));
644 			return (sadb_age_bytes(espstack->esp_pfkey_q, outassoc,
645 			    bytes, B_TRUE));
646 		}
647 	}
648 
649 	inrc = sadb_age_bytes(espstack->esp_pfkey_q, inassoc, bytes, B_TRUE);
650 	outrc = sadb_age_bytes(espstack->esp_pfkey_q, outassoc, bytes, B_FALSE);
651 
652 	/*
653 	 * REFRELE any peer SA.
654 	 *
655 	 * Because of the multi-line macro nature of IPSA_REFRELE, keep
656 	 * them in { }.
657 	 */
658 	if (inbound) {
659 		IPSA_REFRELE(outassoc);
660 	} else {
661 		IPSA_REFRELE(inassoc);
662 	}
663 
664 	return (inrc && outrc);
665 }
666 
667 /*
668  * Do incoming NAT-T manipulations for packet.
669  * Returns NULL if the mblk chain is consumed.
670  */
671 static mblk_t *
672 esp_fix_natt_checksums(mblk_t *data_mp, ipsa_t *assoc)
673 {
674 	ipha_t *ipha = (ipha_t *)data_mp->b_rptr;
675 	tcpha_t *tcpha;
676 	udpha_t *udpha;
677 	/* Initialize to our inbound cksum adjustment... */
678 	uint32_t sum = assoc->ipsa_inbound_cksum;
679 
680 	switch (ipha->ipha_protocol) {
681 	case IPPROTO_TCP:
682 		tcpha = (tcpha_t *)(data_mp->b_rptr +
683 		    IPH_HDR_LENGTH(ipha));
684 
685 #define	DOWN_SUM(x) (x) = ((x) & 0xFFFF) +	 ((x) >> 16)
686 		sum += ~ntohs(tcpha->tha_sum) & 0xFFFF;
687 		DOWN_SUM(sum);
688 		DOWN_SUM(sum);
689 		tcpha->tha_sum = ~htons(sum);
690 		break;
691 	case IPPROTO_UDP:
692 		udpha = (udpha_t *)(data_mp->b_rptr + IPH_HDR_LENGTH(ipha));
693 
694 		if (udpha->uha_checksum != 0) {
695 			/* Adujst if the inbound one was not zero. */
696 			sum += ~ntohs(udpha->uha_checksum) & 0xFFFF;
697 			DOWN_SUM(sum);
698 			DOWN_SUM(sum);
699 			udpha->uha_checksum = ~htons(sum);
700 			if (udpha->uha_checksum == 0)
701 				udpha->uha_checksum = 0xFFFF;
702 		}
703 #undef DOWN_SUM
704 		break;
705 	case IPPROTO_IP:
706 		/*
707 		 * This case is only an issue for self-encapsulated
708 		 * packets.  So for now, fall through.
709 		 */
710 		break;
711 	}
712 	return (data_mp);
713 }
714 
715 
716 /*
717  * Strip ESP header, check padding, and fix IP header.
718  * Returns B_TRUE on success, B_FALSE if an error occured.
719  */
720 static boolean_t
721 esp_strip_header(mblk_t *data_mp, boolean_t isv4, uint32_t ivlen,
722     kstat_named_t **counter, ipsecesp_stack_t *espstack)
723 {
724 	ipha_t *ipha;
725 	ip6_t *ip6h;
726 	uint_t divpoint;
727 	mblk_t *scratch;
728 	uint8_t nexthdr, padlen;
729 	uint8_t lastpad;
730 	ipsec_stack_t	*ipss = espstack->ipsecesp_netstack->netstack_ipsec;
731 	uint8_t *lastbyte;
732 
733 	/*
734 	 * Strip ESP data and fix IP header.
735 	 *
736 	 * XXX In case the beginning of esp_inbound() changes to not do a
737 	 * pullup, this part of the code can remain unchanged.
738 	 */
739 	if (isv4) {
740 		ASSERT((data_mp->b_wptr - data_mp->b_rptr) >= sizeof (ipha_t));
741 		ipha = (ipha_t *)data_mp->b_rptr;
742 		ASSERT((data_mp->b_wptr - data_mp->b_rptr) >= sizeof (esph_t) +
743 		    IPH_HDR_LENGTH(ipha));
744 		divpoint = IPH_HDR_LENGTH(ipha);
745 	} else {
746 		ASSERT((data_mp->b_wptr - data_mp->b_rptr) >= sizeof (ip6_t));
747 		ip6h = (ip6_t *)data_mp->b_rptr;
748 		divpoint = ip_hdr_length_v6(data_mp, ip6h);
749 	}
750 
751 	scratch = data_mp;
752 	while (scratch->b_cont != NULL)
753 		scratch = scratch->b_cont;
754 
755 	ASSERT((scratch->b_wptr - scratch->b_rptr) >= 3);
756 
757 	/*
758 	 * "Next header" and padding length are the last two bytes in the
759 	 * ESP-protected datagram, thus the explicit - 1 and - 2.
760 	 * lastpad is the last byte of the padding, which can be used for
761 	 * a quick check to see if the padding is correct.
762 	 */
763 	lastbyte = scratch->b_wptr - 1;
764 	nexthdr = *lastbyte--;
765 	padlen = *lastbyte--;
766 
767 	if (isv4) {
768 		/* Fix part of the IP header. */
769 		ipha->ipha_protocol = nexthdr;
770 		/*
771 		 * Reality check the padlen.  The explicit - 2 is for the
772 		 * padding length and the next-header bytes.
773 		 */
774 		if (padlen >= ntohs(ipha->ipha_length) - sizeof (ipha_t) - 2 -
775 		    sizeof (esph_t) - ivlen) {
776 			ESP_BUMP_STAT(espstack, bad_decrypt);
777 			ipsec_rl_strlog(espstack->ipsecesp_netstack,
778 			    info.mi_idnum, 0, 0,
779 			    SL_ERROR | SL_WARN,
780 			    "Corrupt ESP packet (padlen too big).\n");
781 			esp1dbg(espstack, ("padlen (%d) is greater than:\n",
782 			    padlen));
783 			esp1dbg(espstack, ("pkt len(%d) - ip hdr - esp "
784 			    "hdr - ivlen(%d) = %d.\n",
785 			    ntohs(ipha->ipha_length), ivlen,
786 			    (int)(ntohs(ipha->ipha_length) - sizeof (ipha_t) -
787 			    2 - sizeof (esph_t) - ivlen)));
788 			*counter = DROPPER(ipss, ipds_esp_bad_padlen);
789 			return (B_FALSE);
790 		}
791 
792 		/*
793 		 * Fix the rest of the header.  The explicit - 2 is for the
794 		 * padding length and the next-header bytes.
795 		 */
796 		ipha->ipha_length = htons(ntohs(ipha->ipha_length) - padlen -
797 		    2 - sizeof (esph_t) - ivlen);
798 		ipha->ipha_hdr_checksum = 0;
799 		ipha->ipha_hdr_checksum = (uint16_t)ip_csum_hdr(ipha);
800 	} else {
801 		if (ip6h->ip6_nxt == IPPROTO_ESP) {
802 			ip6h->ip6_nxt = nexthdr;
803 		} else {
804 			ip_pkt_t ipp;
805 
806 			bzero(&ipp, sizeof (ipp));
807 			(void) ip_find_hdr_v6(data_mp, ip6h, B_FALSE, &ipp,
808 			    NULL);
809 			if (ipp.ipp_dstopts != NULL) {
810 				ipp.ipp_dstopts->ip6d_nxt = nexthdr;
811 			} else if (ipp.ipp_rthdr != NULL) {
812 				ipp.ipp_rthdr->ip6r_nxt = nexthdr;
813 			} else if (ipp.ipp_hopopts != NULL) {
814 				ipp.ipp_hopopts->ip6h_nxt = nexthdr;
815 			} else {
816 				/* Panic a DEBUG kernel. */
817 				ASSERT(ipp.ipp_hopopts != NULL);
818 				/* Otherwise, pretend it's IP + ESP. */
819 				cmn_err(CE_WARN, "ESP IPv6 headers wrong.\n");
820 				ip6h->ip6_nxt = nexthdr;
821 			}
822 		}
823 
824 		if (padlen >= ntohs(ip6h->ip6_plen) - 2 - sizeof (esph_t) -
825 		    ivlen) {
826 			ESP_BUMP_STAT(espstack, bad_decrypt);
827 			ipsec_rl_strlog(espstack->ipsecesp_netstack,
828 			    info.mi_idnum, 0, 0,
829 			    SL_ERROR | SL_WARN,
830 			    "Corrupt ESP packet (v6 padlen too big).\n");
831 			esp1dbg(espstack, ("padlen (%d) is greater than:\n",
832 			    padlen));
833 			esp1dbg(espstack,
834 			    ("pkt len(%u) - ip hdr - esp hdr - ivlen(%d) = "
835 			    "%u.\n", (unsigned)(ntohs(ip6h->ip6_plen)
836 			    + sizeof (ip6_t)), ivlen,
837 			    (unsigned)(ntohs(ip6h->ip6_plen) - 2 -
838 			    sizeof (esph_t) - ivlen)));
839 			*counter = DROPPER(ipss, ipds_esp_bad_padlen);
840 			return (B_FALSE);
841 		}
842 
843 
844 		/*
845 		 * Fix the rest of the header.  The explicit - 2 is for the
846 		 * padding length and the next-header bytes.  IPv6 is nice,
847 		 * because there's no hdr checksum!
848 		 */
849 		ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) - padlen -
850 		    2 - sizeof (esph_t) - ivlen);
851 	}
852 
853 	if (espstack->ipsecesp_padding_check > 0 && padlen > 0) {
854 		/*
855 		 * Weak padding check: compare last-byte to length, they
856 		 * should be equal.
857 		 */
858 		lastpad = *lastbyte--;
859 
860 		if (padlen != lastpad) {
861 			ipsec_rl_strlog(espstack->ipsecesp_netstack,
862 			    info.mi_idnum, 0, 0, SL_ERROR | SL_WARN,
863 			    "Corrupt ESP packet (lastpad != padlen).\n");
864 			esp1dbg(espstack,
865 			    ("lastpad (%d) not equal to padlen (%d):\n",
866 			    lastpad, padlen));
867 			ESP_BUMP_STAT(espstack, bad_padding);
868 			*counter = DROPPER(ipss, ipds_esp_bad_padding);
869 			return (B_FALSE);
870 		}
871 
872 		/*
873 		 * Strong padding check: Check all pad bytes to see that
874 		 * they're ascending.  Go backwards using a descending counter
875 		 * to verify.  padlen == 1 is checked by previous block, so
876 		 * only bother if we've more than 1 byte of padding.
877 		 * Consequently, start the check one byte before the location
878 		 * of "lastpad".
879 		 */
880 		if (espstack->ipsecesp_padding_check > 1) {
881 			/*
882 			 * This assert may have to become an if and a pullup
883 			 * if we start accepting multi-dblk mblks. For now,
884 			 * though, any packet here will have been pulled up in
885 			 * esp_inbound.
886 			 */
887 			ASSERT(MBLKL(scratch) >= lastpad + 3);
888 
889 			/*
890 			 * Use "--lastpad" because we already checked the very
891 			 * last pad byte previously.
892 			 */
893 			while (--lastpad != 0) {
894 				if (lastpad != *lastbyte) {
895 					ipsec_rl_strlog(
896 					    espstack->ipsecesp_netstack,
897 					    info.mi_idnum, 0, 0,
898 					    SL_ERROR | SL_WARN, "Corrupt ESP "
899 					    "packet (bad padding).\n");
900 					esp1dbg(espstack,
901 					    ("padding not in correct"
902 					    " format:\n"));
903 					ESP_BUMP_STAT(espstack, bad_padding);
904 					*counter = DROPPER(ipss,
905 					    ipds_esp_bad_padding);
906 					return (B_FALSE);
907 				}
908 				lastbyte--;
909 			}
910 		}
911 	}
912 
913 	/* Trim off the padding. */
914 	ASSERT(data_mp->b_cont == NULL);
915 	data_mp->b_wptr -= (padlen + 2);
916 
917 	/*
918 	 * Remove the ESP header.
919 	 *
920 	 * The above assertions about data_mp's size will make this work.
921 	 *
922 	 * XXX  Question:  If I send up and get back a contiguous mblk,
923 	 * would it be quicker to bcopy over, or keep doing the dupb stuff?
924 	 * I go with copying for now.
925 	 */
926 
927 	if (IS_P2ALIGNED(data_mp->b_rptr, sizeof (uint32_t)) &&
928 	    IS_P2ALIGNED(ivlen, sizeof (uint32_t))) {
929 		uint8_t *start = data_mp->b_rptr;
930 		uint32_t *src, *dst;
931 
932 		src = (uint32_t *)(start + divpoint);
933 		dst = (uint32_t *)(start + divpoint + sizeof (esph_t) + ivlen);
934 
935 		ASSERT(IS_P2ALIGNED(dst, sizeof (uint32_t)) &&
936 		    IS_P2ALIGNED(src, sizeof (uint32_t)));
937 
938 		do {
939 			src--;
940 			dst--;
941 			*dst = *src;
942 		} while (src != (uint32_t *)start);
943 
944 		data_mp->b_rptr = (uchar_t *)dst;
945 	} else {
946 		uint8_t *start = data_mp->b_rptr;
947 		uint8_t *src, *dst;
948 
949 		src = start + divpoint;
950 		dst = src + sizeof (esph_t) + ivlen;
951 
952 		do {
953 			src--;
954 			dst--;
955 			*dst = *src;
956 		} while (src != start);
957 
958 		data_mp->b_rptr = dst;
959 	}
960 
961 	esp2dbg(espstack, ("data_mp after inbound ESP adjustment:\n"));
962 	esp2dbg(espstack, (dump_msg(data_mp)));
963 
964 	return (B_TRUE);
965 }
966 
967 /*
968  * Updating use times can be tricky business if the ipsa_haspeer flag is
969  * set.  This function is called once in an SA's lifetime.
970  *
971  * Caller has to REFRELE "assoc" which is passed in.  This function has
972  * to REFRELE any peer SA that is obtained.
973  */
974 static void
975 esp_set_usetime(ipsa_t *assoc, boolean_t inbound)
976 {
977 	ipsa_t *inassoc, *outassoc;
978 	isaf_t *bucket;
979 	sadb_t *sp;
980 	int outhash;
981 	boolean_t isv6;
982 	netstack_t		*ns = assoc->ipsa_netstack;
983 	ipsecesp_stack_t	*espstack = ns->netstack_ipsecesp;
984 
985 	/* No peer?  No problem! */
986 	if (!assoc->ipsa_haspeer) {
987 		sadb_set_usetime(assoc);
988 		return;
989 	}
990 
991 	/*
992 	 * Otherwise, we want to grab both the original assoc and its peer.
993 	 * There might be a race for this, but if it's a real race, the times
994 	 * will be out-of-synch by at most a second, and since our time
995 	 * granularity is a second, this won't be a problem.
996 	 *
997 	 * If we need tight synchronization on the peer SA, then we need to
998 	 * reconsider.
999 	 */
1000 
1001 	/* Use address length to select IPv6/IPv4 */
1002 	isv6 = (assoc->ipsa_addrfam == AF_INET6);
1003 	sp = isv6 ? &espstack->esp_sadb.s_v6 : &espstack->esp_sadb.s_v4;
1004 
1005 	if (inbound) {
1006 		inassoc = assoc;
1007 		if (isv6) {
1008 			outhash = OUTBOUND_HASH_V6(sp, *((in6_addr_t *)
1009 			    &inassoc->ipsa_dstaddr));
1010 		} else {
1011 			outhash = OUTBOUND_HASH_V4(sp, *((ipaddr_t *)
1012 			    &inassoc->ipsa_dstaddr));
1013 		}
1014 		bucket = &sp->sdb_of[outhash];
1015 		mutex_enter(&bucket->isaf_lock);
1016 		outassoc = ipsec_getassocbyspi(bucket, inassoc->ipsa_spi,
1017 		    inassoc->ipsa_srcaddr, inassoc->ipsa_dstaddr,
1018 		    inassoc->ipsa_addrfam);
1019 		mutex_exit(&bucket->isaf_lock);
1020 		if (outassoc == NULL) {
1021 			/* Q: Do we wish to set haspeer == B_FALSE? */
1022 			esp0dbg(("esp_set_usetime: "
1023 			    "can't find peer for inbound.\n"));
1024 			sadb_set_usetime(inassoc);
1025 			return;
1026 		}
1027 	} else {
1028 		outassoc = assoc;
1029 		bucket = INBOUND_BUCKET(sp, outassoc->ipsa_spi);
1030 		mutex_enter(&bucket->isaf_lock);
1031 		inassoc = ipsec_getassocbyspi(bucket, outassoc->ipsa_spi,
1032 		    outassoc->ipsa_srcaddr, outassoc->ipsa_dstaddr,
1033 		    outassoc->ipsa_addrfam);
1034 		mutex_exit(&bucket->isaf_lock);
1035 		if (inassoc == NULL) {
1036 			/* Q: Do we wish to set haspeer == B_FALSE? */
1037 			esp0dbg(("esp_set_usetime: "
1038 			    "can't find peer for outbound.\n"));
1039 			sadb_set_usetime(outassoc);
1040 			return;
1041 		}
1042 	}
1043 
1044 	/* Update usetime on both. */
1045 	sadb_set_usetime(inassoc);
1046 	sadb_set_usetime(outassoc);
1047 
1048 	/*
1049 	 * REFRELE any peer SA.
1050 	 *
1051 	 * Because of the multi-line macro nature of IPSA_REFRELE, keep
1052 	 * them in { }.
1053 	 */
1054 	if (inbound) {
1055 		IPSA_REFRELE(outassoc);
1056 	} else {
1057 		IPSA_REFRELE(inassoc);
1058 	}
1059 }
1060 
1061 /*
1062  * Handle ESP inbound data for IPv4 and IPv6.
1063  * On success returns B_TRUE, on failure returns B_FALSE and frees the
1064  * mblk chain data_mp.
1065  */
1066 mblk_t *
1067 esp_inbound(mblk_t *data_mp, void *arg, ip_recv_attr_t *ira)
1068 {
1069 	esph_t *esph = (esph_t *)arg;
1070 	ipsa_t *ipsa = ira->ira_ipsec_esp_sa;
1071 	netstack_t	*ns = ira->ira_ill->ill_ipst->ips_netstack;
1072 	ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1073 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
1074 
1075 	/*
1076 	 * We may wish to check replay in-range-only here as an optimization.
1077 	 * Include the reality check of ipsa->ipsa_replay >
1078 	 * ipsa->ipsa_replay_wsize for times when it's the first N packets,
1079 	 * where N == ipsa->ipsa_replay_wsize.
1080 	 *
1081 	 * Another check that may come here later is the "collision" check.
1082 	 * If legitimate packets flow quickly enough, this won't be a problem,
1083 	 * but collisions may cause authentication algorithm crunching to
1084 	 * take place when it doesn't need to.
1085 	 */
1086 	if (!sadb_replay_peek(ipsa, esph->esph_replay)) {
1087 		ESP_BUMP_STAT(espstack, replay_early_failures);
1088 		IP_ESP_BUMP_STAT(ipss, in_discards);
1089 		ip_drop_packet(data_mp, B_TRUE, ira->ira_ill,
1090 		    DROPPER(ipss, ipds_esp_early_replay),
1091 		    &espstack->esp_dropper);
1092 		BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
1093 		return (NULL);
1094 	}
1095 
1096 	/*
1097 	 * Adjust the IP header's payload length to reflect the removal
1098 	 * of the ICV.
1099 	 */
1100 	if (!(ira->ira_flags & IRAF_IS_IPV4)) {
1101 		ip6_t *ip6h = (ip6_t *)data_mp->b_rptr;
1102 		ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) -
1103 		    ipsa->ipsa_mac_len);
1104 	} else {
1105 		ipha_t *ipha = (ipha_t *)data_mp->b_rptr;
1106 		ipha->ipha_length = htons(ntohs(ipha->ipha_length) -
1107 		    ipsa->ipsa_mac_len);
1108 	}
1109 
1110 	/* submit the request to the crypto framework */
1111 	return (esp_submit_req_inbound(data_mp, ira, ipsa,
1112 	    (uint8_t *)esph - data_mp->b_rptr));
1113 }
1114 
1115 /* XXX refactor me */
1116 /*
1117  * Handle the SADB_GETSPI message.  Create a larval SA.
1118  */
1119 static void
1120 esp_getspi(mblk_t *mp, keysock_in_t *ksi, ipsecesp_stack_t *espstack)
1121 {
1122 	ipsa_t *newbie, *target;
1123 	isaf_t *outbound, *inbound;
1124 	int rc, diagnostic;
1125 	sadb_sa_t *assoc;
1126 	keysock_out_t *kso;
1127 	uint32_t newspi;
1128 
1129 	/*
1130 	 * Randomly generate a proposed SPI value
1131 	 */
1132 	if (cl_inet_getspi != NULL) {
1133 		cl_inet_getspi(espstack->ipsecesp_netstack->netstack_stackid,
1134 		    IPPROTO_ESP, (uint8_t *)&newspi, sizeof (uint32_t), NULL);
1135 	} else {
1136 		(void) random_get_pseudo_bytes((uint8_t *)&newspi,
1137 		    sizeof (uint32_t));
1138 	}
1139 	newbie = sadb_getspi(ksi, newspi, &diagnostic,
1140 	    espstack->ipsecesp_netstack, IPPROTO_ESP);
1141 
1142 	if (newbie == NULL) {
1143 		sadb_pfkey_error(espstack->esp_pfkey_q, mp, ENOMEM, diagnostic,
1144 		    ksi->ks_in_serial);
1145 		return;
1146 	} else if (newbie == (ipsa_t *)-1) {
1147 		sadb_pfkey_error(espstack->esp_pfkey_q, mp, EINVAL, diagnostic,
1148 		    ksi->ks_in_serial);
1149 		return;
1150 	}
1151 
1152 	/*
1153 	 * XXX - We may randomly collide.  We really should recover from this.
1154 	 *	 Unfortunately, that could require spending way-too-much-time
1155 	 *	 in here.  For now, let the user retry.
1156 	 */
1157 
1158 	if (newbie->ipsa_addrfam == AF_INET6) {
1159 		outbound = OUTBOUND_BUCKET_V6(&espstack->esp_sadb.s_v6,
1160 		    *(uint32_t *)(newbie->ipsa_dstaddr));
1161 		inbound = INBOUND_BUCKET(&espstack->esp_sadb.s_v6,
1162 		    newbie->ipsa_spi);
1163 	} else {
1164 		ASSERT(newbie->ipsa_addrfam == AF_INET);
1165 		outbound = OUTBOUND_BUCKET_V4(&espstack->esp_sadb.s_v4,
1166 		    *(uint32_t *)(newbie->ipsa_dstaddr));
1167 		inbound = INBOUND_BUCKET(&espstack->esp_sadb.s_v4,
1168 		    newbie->ipsa_spi);
1169 	}
1170 
1171 	mutex_enter(&outbound->isaf_lock);
1172 	mutex_enter(&inbound->isaf_lock);
1173 
1174 	/*
1175 	 * Check for collisions (i.e. did sadb_getspi() return with something
1176 	 * that already exists?).
1177 	 *
1178 	 * Try outbound first.  Even though SADB_GETSPI is traditionally
1179 	 * for inbound SAs, you never know what a user might do.
1180 	 */
1181 	target = ipsec_getassocbyspi(outbound, newbie->ipsa_spi,
1182 	    newbie->ipsa_srcaddr, newbie->ipsa_dstaddr, newbie->ipsa_addrfam);
1183 	if (target == NULL) {
1184 		target = ipsec_getassocbyspi(inbound, newbie->ipsa_spi,
1185 		    newbie->ipsa_srcaddr, newbie->ipsa_dstaddr,
1186 		    newbie->ipsa_addrfam);
1187 	}
1188 
1189 	/*
1190 	 * I don't have collisions elsewhere!
1191 	 * (Nor will I because I'm still holding inbound/outbound locks.)
1192 	 */
1193 
1194 	if (target != NULL) {
1195 		rc = EEXIST;
1196 		IPSA_REFRELE(target);
1197 	} else {
1198 		/*
1199 		 * sadb_insertassoc() also checks for collisions, so
1200 		 * if there's a colliding entry, rc will be set
1201 		 * to EEXIST.
1202 		 */
1203 		rc = sadb_insertassoc(newbie, inbound);
1204 		newbie->ipsa_hardexpiretime = gethrestime_sec();
1205 		newbie->ipsa_hardexpiretime +=
1206 		    espstack->ipsecesp_larval_timeout;
1207 	}
1208 
1209 	/*
1210 	 * Can exit outbound mutex.  Hold inbound until we're done
1211 	 * with newbie.
1212 	 */
1213 	mutex_exit(&outbound->isaf_lock);
1214 
1215 	if (rc != 0) {
1216 		mutex_exit(&inbound->isaf_lock);
1217 		IPSA_REFRELE(newbie);
1218 		sadb_pfkey_error(espstack->esp_pfkey_q, mp, rc,
1219 		    SADB_X_DIAGNOSTIC_NONE, ksi->ks_in_serial);
1220 		return;
1221 	}
1222 
1223 
1224 	/* Can write here because I'm still holding the bucket lock. */
1225 	newbie->ipsa_type = SADB_SATYPE_ESP;
1226 
1227 	/*
1228 	 * Construct successful return message. We have one thing going
1229 	 * for us in PF_KEY v2.  That's the fact that
1230 	 *	sizeof (sadb_spirange_t) == sizeof (sadb_sa_t)
1231 	 */
1232 	assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SPIRANGE];
1233 	assoc->sadb_sa_exttype = SADB_EXT_SA;
1234 	assoc->sadb_sa_spi = newbie->ipsa_spi;
1235 	*((uint64_t *)(&assoc->sadb_sa_replay)) = 0;
1236 	mutex_exit(&inbound->isaf_lock);
1237 
1238 	/* Convert KEYSOCK_IN to KEYSOCK_OUT. */
1239 	kso = (keysock_out_t *)ksi;
1240 	kso->ks_out_len = sizeof (*kso);
1241 	kso->ks_out_serial = ksi->ks_in_serial;
1242 	kso->ks_out_type = KEYSOCK_OUT;
1243 
1244 	/*
1245 	 * Can safely putnext() to esp_pfkey_q, because this is a turnaround
1246 	 * from the esp_pfkey_q.
1247 	 */
1248 	putnext(espstack->esp_pfkey_q, mp);
1249 }
1250 
1251 /*
1252  * Insert the ESP header into a packet.  Duplicate an mblk, and insert a newly
1253  * allocated mblk with the ESP header in between the two.
1254  */
1255 static boolean_t
1256 esp_insert_esp(mblk_t *mp, mblk_t *esp_mp, uint_t divpoint,
1257     ipsecesp_stack_t *espstack)
1258 {
1259 	mblk_t *split_mp = mp;
1260 	uint_t wheretodiv = divpoint;
1261 
1262 	while ((split_mp->b_wptr - split_mp->b_rptr) < wheretodiv) {
1263 		wheretodiv -= (split_mp->b_wptr - split_mp->b_rptr);
1264 		split_mp = split_mp->b_cont;
1265 		ASSERT(split_mp != NULL);
1266 	}
1267 
1268 	if (split_mp->b_wptr - split_mp->b_rptr != wheretodiv) {
1269 		mblk_t *scratch;
1270 
1271 		/* "scratch" is the 2nd half, split_mp is the first. */
1272 		scratch = dupb(split_mp);
1273 		if (scratch == NULL) {
1274 			esp1dbg(espstack,
1275 			    ("esp_insert_esp: can't allocate scratch.\n"));
1276 			return (B_FALSE);
1277 		}
1278 		/* NOTE:  dupb() doesn't set b_cont appropriately. */
1279 		scratch->b_cont = split_mp->b_cont;
1280 		scratch->b_rptr += wheretodiv;
1281 		split_mp->b_wptr = split_mp->b_rptr + wheretodiv;
1282 		split_mp->b_cont = scratch;
1283 	}
1284 	/*
1285 	 * At this point, split_mp is exactly "wheretodiv" bytes long, and
1286 	 * holds the end of the pre-ESP part of the datagram.
1287 	 */
1288 	esp_mp->b_cont = split_mp->b_cont;
1289 	split_mp->b_cont = esp_mp;
1290 
1291 	return (B_TRUE);
1292 }
1293 
1294 /*
1295  * Section 7 of RFC 3947 says:
1296  *
1297  * 7.  Recovering from the Expiring NAT Mappings
1298  *
1299  *    There are cases where NAT box decides to remove mappings that are still
1300  *    alive (for example, when the keepalive interval is too long, or when the
1301  *    NAT box is rebooted).  To recover from this, ends that are NOT behind
1302  *    NAT SHOULD use the last valid UDP encapsulated IKE or IPsec packet from
1303  *    the other end to determine which IP and port addresses should be used.
1304  *    The host behind dynamic NAT MUST NOT do this, as otherwise it opens a
1305  *    DoS attack possibility because the IP address or port of the other host
1306  *    will not change (it is not behind NAT).
1307  *
1308  *    Keepalives cannot be used for these purposes, as they are not
1309  *    authenticated, but any IKE authenticated IKE packet or ESP packet can be
1310  *    used to detect whether the IP address or the port has changed.
1311  *
1312  * The following function will check an SA and its explicitly-set pair to see
1313  * if the NAT-T remote port matches the received packet (which must have
1314  * passed ESP authentication, see esp_in_done() for the caller context).  If
1315  * there is a mismatch, the SAs are updated.  It is not important if we race
1316  * with a transmitting thread, as if there is a transmitting thread, it will
1317  * merely emit a packet that will most-likely be dropped.
1318  *
1319  * "ports" are ordered src,dst, and assoc is an inbound SA, where src should
1320  * match ipsa_remote_nat_port and dst should match ipsa_local_nat_port.
1321  */
1322 #ifdef _LITTLE_ENDIAN
1323 #define	FIRST_16(x) ((x) & 0xFFFF)
1324 #define	NEXT_16(x) (((x) >> 16) & 0xFFFF)
1325 #else
1326 #define	FIRST_16(x) (((x) >> 16) & 0xFFFF)
1327 #define	NEXT_16(x) ((x) & 0xFFFF)
1328 #endif
1329 static void
1330 esp_port_freshness(uint32_t ports, ipsa_t *assoc)
1331 {
1332 	uint16_t remote = FIRST_16(ports);
1333 	uint16_t local = NEXT_16(ports);
1334 	ipsa_t *outbound_peer;
1335 	isaf_t *bucket;
1336 	ipsecesp_stack_t *espstack = assoc->ipsa_netstack->netstack_ipsecesp;
1337 
1338 	/* We found a conn_t, therefore local != 0. */
1339 	ASSERT(local != 0);
1340 	/* Assume an IPv4 SA. */
1341 	ASSERT(assoc->ipsa_addrfam == AF_INET);
1342 
1343 	/*
1344 	 * On-the-wire rport == 0 means something's very wrong.
1345 	 * An unpaired SA is also useless to us.
1346 	 * If we are behind the NAT, don't bother.
1347 	 * A zero local NAT port defaults to 4500, so check that too.
1348 	 * And, of course, if the ports already match, we don't need to
1349 	 * bother.
1350 	 */
1351 	if (remote == 0 || assoc->ipsa_otherspi == 0 ||
1352 	    (assoc->ipsa_flags & IPSA_F_BEHIND_NAT) ||
1353 	    (assoc->ipsa_remote_nat_port == 0 &&
1354 	    remote == htons(IPPORT_IKE_NATT)) ||
1355 	    remote == assoc->ipsa_remote_nat_port)
1356 		return;
1357 
1358 	/* Try and snag the peer.   NOTE:  Assume IPv4 for now. */
1359 	bucket = OUTBOUND_BUCKET_V4(&(espstack->esp_sadb.s_v4),
1360 	    assoc->ipsa_srcaddr[0]);
1361 	mutex_enter(&bucket->isaf_lock);
1362 	outbound_peer = ipsec_getassocbyspi(bucket, assoc->ipsa_otherspi,
1363 	    assoc->ipsa_dstaddr, assoc->ipsa_srcaddr, AF_INET);
1364 	mutex_exit(&bucket->isaf_lock);
1365 
1366 	/* We probably lost a race to a deleting or expiring thread. */
1367 	if (outbound_peer == NULL)
1368 		return;
1369 
1370 	/*
1371 	 * Hold the mutexes for both SAs so we don't race another inbound
1372 	 * thread.  A lock-entry order shouldn't matter, since all other
1373 	 * per-ipsa locks are individually held-then-released.
1374 	 *
1375 	 * Luckily, this has nothing to do with the remote-NAT address,
1376 	 * so we don't have to re-scribble the cached-checksum differential.
1377 	 */
1378 	mutex_enter(&outbound_peer->ipsa_lock);
1379 	mutex_enter(&assoc->ipsa_lock);
1380 	outbound_peer->ipsa_remote_nat_port = assoc->ipsa_remote_nat_port =
1381 	    remote;
1382 	mutex_exit(&assoc->ipsa_lock);
1383 	mutex_exit(&outbound_peer->ipsa_lock);
1384 	IPSA_REFRELE(outbound_peer);
1385 	ESP_BUMP_STAT(espstack, sa_port_renumbers);
1386 }
1387 /*
1388  * Finish processing of an inbound ESP packet after processing by the
1389  * crypto framework.
1390  * - Remove the ESP header.
1391  * - Send packet back to IP.
1392  * If authentication was performed on the packet, this function is called
1393  * only if the authentication succeeded.
1394  * On success returns B_TRUE, on failure returns B_FALSE and frees the
1395  * mblk chain data_mp.
1396  */
1397 static mblk_t *
1398 esp_in_done(mblk_t *data_mp, ip_recv_attr_t *ira, ipsec_crypto_t *ic)
1399 {
1400 	ipsa_t *assoc;
1401 	uint_t espstart;
1402 	uint32_t ivlen = 0;
1403 	uint_t processed_len;
1404 	esph_t *esph;
1405 	kstat_named_t *counter;
1406 	boolean_t is_natt;
1407 	netstack_t	*ns = ira->ira_ill->ill_ipst->ips_netstack;
1408 	ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1409 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
1410 
1411 	assoc = ira->ira_ipsec_esp_sa;
1412 	ASSERT(assoc != NULL);
1413 
1414 	is_natt = ((assoc->ipsa_flags & IPSA_F_NATT) != 0);
1415 
1416 	/* get the pointer to the ESP header */
1417 	if (assoc->ipsa_encr_alg == SADB_EALG_NULL) {
1418 		/* authentication-only ESP */
1419 		espstart = ic->ic_crypto_data.cd_offset;
1420 		processed_len = ic->ic_crypto_data.cd_length;
1421 	} else {
1422 		/* encryption present */
1423 		ivlen = assoc->ipsa_iv_len;
1424 		if (assoc->ipsa_auth_alg == SADB_AALG_NONE) {
1425 			/* encryption-only ESP */
1426 			espstart = ic->ic_crypto_data.cd_offset -
1427 			    sizeof (esph_t) - assoc->ipsa_iv_len;
1428 			processed_len = ic->ic_crypto_data.cd_length +
1429 			    ivlen;
1430 		} else {
1431 			/* encryption with authentication */
1432 			espstart = ic->ic_crypto_dual_data.dd_offset1;
1433 			processed_len = ic->ic_crypto_dual_data.dd_len2 +
1434 			    ivlen;
1435 		}
1436 	}
1437 
1438 	esph = (esph_t *)(data_mp->b_rptr + espstart);
1439 
1440 	if (assoc->ipsa_auth_alg != IPSA_AALG_NONE ||
1441 	    (assoc->ipsa_flags & IPSA_F_COMBINED)) {
1442 		/*
1443 		 * Authentication passed if we reach this point.
1444 		 * Packets with authentication will have the ICV
1445 		 * after the crypto data. Adjust b_wptr before
1446 		 * making padlen checks.
1447 		 */
1448 		ESP_BUMP_STAT(espstack, good_auth);
1449 		data_mp->b_wptr -= assoc->ipsa_mac_len;
1450 
1451 		/*
1452 		 * Check replay window here!
1453 		 * For right now, assume keysock will set the replay window
1454 		 * size to zero for SAs that have an unspecified sender.
1455 		 * This may change...
1456 		 */
1457 
1458 		if (!sadb_replay_check(assoc, esph->esph_replay)) {
1459 			/*
1460 			 * Log the event. As of now we print out an event.
1461 			 * Do not print the replay failure number, or else
1462 			 * syslog cannot collate the error messages.  Printing
1463 			 * the replay number that failed opens a denial-of-
1464 			 * service attack.
1465 			 */
1466 			ipsec_assocfailure(info.mi_idnum, 0, 0,
1467 			    SL_ERROR | SL_WARN,
1468 			    "Replay failed for ESP spi 0x%x, dst %s.\n",
1469 			    assoc->ipsa_spi, assoc->ipsa_dstaddr,
1470 			    assoc->ipsa_addrfam, espstack->ipsecesp_netstack);
1471 			ESP_BUMP_STAT(espstack, replay_failures);
1472 			counter = DROPPER(ipss, ipds_esp_replay);
1473 			goto drop_and_bail;
1474 		}
1475 
1476 		if (is_natt) {
1477 			ASSERT(ira->ira_flags & IRAF_ESP_UDP_PORTS);
1478 			ASSERT(ira->ira_esp_udp_ports != 0);
1479 			esp_port_freshness(ira->ira_esp_udp_ports, assoc);
1480 		}
1481 	}
1482 
1483 	esp_set_usetime(assoc, B_TRUE);
1484 
1485 	if (!esp_age_bytes(assoc, processed_len, B_TRUE)) {
1486 		/* The ipsa has hit hard expiration, LOG and AUDIT. */
1487 		ipsec_assocfailure(info.mi_idnum, 0, 0,
1488 		    SL_ERROR | SL_WARN,
1489 		    "ESP association 0x%x, dst %s had bytes expire.\n",
1490 		    assoc->ipsa_spi, assoc->ipsa_dstaddr, assoc->ipsa_addrfam,
1491 		    espstack->ipsecesp_netstack);
1492 		ESP_BUMP_STAT(espstack, bytes_expired);
1493 		counter = DROPPER(ipss, ipds_esp_bytes_expire);
1494 		goto drop_and_bail;
1495 	}
1496 
1497 	/*
1498 	 * Remove ESP header and padding from packet.  I hope the compiler
1499 	 * spews "branch, predict taken" code for this.
1500 	 */
1501 
1502 	if (esp_strip_header(data_mp, (ira->ira_flags & IRAF_IS_IPV4),
1503 	    ivlen, &counter, espstack)) {
1504 
1505 		if (is_system_labeled() && assoc->ipsa_tsl != NULL) {
1506 			if (!ip_recv_attr_replace_label(ira, assoc->ipsa_tsl)) {
1507 				ip_drop_packet(data_mp, B_TRUE, ira->ira_ill,
1508 				    DROPPER(ipss, ipds_ah_nomem),
1509 				    &espstack->esp_dropper);
1510 				BUMP_MIB(ira->ira_ill->ill_ip_mib,
1511 				    ipIfStatsInDiscards);
1512 				return (NULL);
1513 			}
1514 		}
1515 		if (is_natt)
1516 			return (esp_fix_natt_checksums(data_mp, assoc));
1517 
1518 		if (assoc->ipsa_state == IPSA_STATE_IDLE) {
1519 			/*
1520 			 * Cluster buffering case.  Tell caller that we're
1521 			 * handling the packet.
1522 			 */
1523 			sadb_buf_pkt(assoc, data_mp, ira);
1524 			return (NULL);
1525 		}
1526 
1527 		return (data_mp);
1528 	}
1529 
1530 	esp1dbg(espstack, ("esp_in_done: esp_strip_header() failed\n"));
1531 drop_and_bail:
1532 	IP_ESP_BUMP_STAT(ipss, in_discards);
1533 	ip_drop_packet(data_mp, B_TRUE, ira->ira_ill, counter,
1534 	    &espstack->esp_dropper);
1535 	BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
1536 	return (NULL);
1537 }
1538 
1539 /*
1540  * Called upon failing the inbound ICV check. The message passed as
1541  * argument is freed.
1542  */
1543 static void
1544 esp_log_bad_auth(mblk_t *mp, ip_recv_attr_t *ira)
1545 {
1546 	ipsa_t		*assoc = ira->ira_ipsec_esp_sa;
1547 	netstack_t	*ns = ira->ira_ill->ill_ipst->ips_netstack;
1548 	ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1549 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
1550 
1551 	/*
1552 	 * Log the event. Don't print to the console, block
1553 	 * potential denial-of-service attack.
1554 	 */
1555 	ESP_BUMP_STAT(espstack, bad_auth);
1556 
1557 	ipsec_assocfailure(info.mi_idnum, 0, 0, SL_ERROR | SL_WARN,
1558 	    "ESP Authentication failed for spi 0x%x, dst %s.\n",
1559 	    assoc->ipsa_spi, assoc->ipsa_dstaddr, assoc->ipsa_addrfam,
1560 	    espstack->ipsecesp_netstack);
1561 
1562 	IP_ESP_BUMP_STAT(ipss, in_discards);
1563 	ip_drop_packet(mp, B_TRUE, ira->ira_ill,
1564 	    DROPPER(ipss, ipds_esp_bad_auth),
1565 	    &espstack->esp_dropper);
1566 }
1567 
1568 
1569 /*
1570  * Invoked for outbound packets after ESP processing. If the packet
1571  * also requires AH, performs the AH SA selection and AH processing.
1572  *
1573  * Returns data_mp (possibly with AH added) unless data_mp was consumed
1574  * due to an error, or queued due to async. crypto or an ACQUIRE trigger.
1575  */
1576 static mblk_t *
1577 esp_do_outbound_ah(mblk_t *data_mp, ip_xmit_attr_t *ixa)
1578 {
1579 	ipsec_action_t *ap;
1580 
1581 	ap = ixa->ixa_ipsec_action;
1582 	if (ap == NULL) {
1583 		ipsec_policy_t *pp = ixa->ixa_ipsec_policy;
1584 		ap = pp->ipsp_act;
1585 	}
1586 
1587 	if (!ap->ipa_want_ah)
1588 		return (data_mp);
1589 
1590 	/*
1591 	 * Normally the AH SA would have already been put in place
1592 	 * but it could have been flushed so we need to look for it.
1593 	 */
1594 	if (ixa->ixa_ipsec_ah_sa == NULL) {
1595 		if (!ipsec_outbound_sa(data_mp, ixa, IPPROTO_AH)) {
1596 			sadb_acquire(data_mp, ixa, B_TRUE, B_FALSE);
1597 			return (NULL);
1598 		}
1599 	}
1600 	ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
1601 
1602 	data_mp = ixa->ixa_ipsec_ah_sa->ipsa_output_func(data_mp, ixa);
1603 	return (data_mp);
1604 }
1605 
1606 
1607 /*
1608  * Kernel crypto framework callback invoked after completion of async
1609  * crypto requests for outbound packets.
1610  */
1611 static void
1612 esp_kcf_callback_outbound(void *arg, int status)
1613 {
1614 	mblk_t		*mp = (mblk_t *)arg;
1615 	mblk_t		*async_mp;
1616 	netstack_t	*ns;
1617 	ipsec_stack_t	*ipss;
1618 	ipsecesp_stack_t *espstack;
1619 	mblk_t		*data_mp;
1620 	ip_xmit_attr_t	ixas;
1621 	ipsec_crypto_t	*ic;
1622 	ill_t		*ill;
1623 
1624 	/*
1625 	 * First remove the ipsec_crypto_t mblk
1626 	 * Note that we need to ipsec_free_crypto_data(mp) once done with ic.
1627 	 */
1628 	async_mp = ipsec_remove_crypto_data(mp, &ic);
1629 	ASSERT(async_mp != NULL);
1630 
1631 	/*
1632 	 * Extract the ip_xmit_attr_t from the first mblk.
1633 	 * Verifies that the netstack and ill is still around; could
1634 	 * have vanished while kEf was doing its work.
1635 	 * On succesful return we have a nce_t and the ill/ipst can't
1636 	 * disappear until we do the nce_refrele in ixa_cleanup.
1637 	 */
1638 	data_mp = async_mp->b_cont;
1639 	async_mp->b_cont = NULL;
1640 	if (!ip_xmit_attr_from_mblk(async_mp, &ixas)) {
1641 		/* Disappeared on us - no ill/ipst for MIB */
1642 		/* We have nowhere to do stats since ixa_ipst could be NULL */
1643 		if (ixas.ixa_nce != NULL) {
1644 			ill = ixas.ixa_nce->nce_ill;
1645 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1646 			ip_drop_output("ipIfStatsOutDiscards", data_mp, ill);
1647 		}
1648 		freemsg(data_mp);
1649 		goto done;
1650 	}
1651 	ns = ixas.ixa_ipst->ips_netstack;
1652 	espstack = ns->netstack_ipsecesp;
1653 	ipss = ns->netstack_ipsec;
1654 	ill = ixas.ixa_nce->nce_ill;
1655 
1656 	if (status == CRYPTO_SUCCESS) {
1657 		/*
1658 		 * If a ICV was computed, it was stored by the
1659 		 * crypto framework at the end of the packet.
1660 		 */
1661 		ipha_t *ipha = (ipha_t *)data_mp->b_rptr;
1662 
1663 		esp_set_usetime(ixas.ixa_ipsec_esp_sa, B_FALSE);
1664 		/* NAT-T packet. */
1665 		if (IPH_HDR_VERSION(ipha) == IP_VERSION &&
1666 		    ipha->ipha_protocol == IPPROTO_UDP)
1667 			esp_prepare_udp(ns, data_mp, ipha);
1668 
1669 		/* do AH processing if needed */
1670 		data_mp = esp_do_outbound_ah(data_mp, &ixas);
1671 		if (data_mp == NULL)
1672 			goto done;
1673 
1674 		(void) ip_output_post_ipsec(data_mp, &ixas);
1675 	} else {
1676 		/* Outbound shouldn't see invalid MAC */
1677 		ASSERT(status != CRYPTO_INVALID_MAC);
1678 
1679 		esp1dbg(espstack,
1680 		    ("esp_kcf_callback_outbound: crypto failed with 0x%x\n",
1681 		    status));
1682 		ESP_BUMP_STAT(espstack, crypto_failures);
1683 		ESP_BUMP_STAT(espstack, out_discards);
1684 		ip_drop_packet(data_mp, B_FALSE, ill,
1685 		    DROPPER(ipss, ipds_esp_crypto_failed),
1686 		    &espstack->esp_dropper);
1687 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1688 	}
1689 done:
1690 	ixa_cleanup(&ixas);
1691 	(void) ipsec_free_crypto_data(mp);
1692 }
1693 
1694 /*
1695  * Kernel crypto framework callback invoked after completion of async
1696  * crypto requests for inbound packets.
1697  */
1698 static void
1699 esp_kcf_callback_inbound(void *arg, int status)
1700 {
1701 	mblk_t		*mp = (mblk_t *)arg;
1702 	mblk_t		*async_mp;
1703 	netstack_t	*ns;
1704 	ipsecesp_stack_t *espstack;
1705 	ipsec_stack_t	*ipss;
1706 	mblk_t		*data_mp;
1707 	ip_recv_attr_t	iras;
1708 	ipsec_crypto_t	*ic;
1709 
1710 	/*
1711 	 * First remove the ipsec_crypto_t mblk
1712 	 * Note that we need to ipsec_free_crypto_data(mp) once done with ic.
1713 	 */
1714 	async_mp = ipsec_remove_crypto_data(mp, &ic);
1715 	ASSERT(async_mp != NULL);
1716 
1717 	/*
1718 	 * Extract the ip_recv_attr_t from the first mblk.
1719 	 * Verifies that the netstack and ill is still around; could
1720 	 * have vanished while kEf was doing its work.
1721 	 */
1722 	data_mp = async_mp->b_cont;
1723 	async_mp->b_cont = NULL;
1724 	if (!ip_recv_attr_from_mblk(async_mp, &iras)) {
1725 		/* The ill or ip_stack_t disappeared on us */
1726 		ip_drop_input("ip_recv_attr_from_mblk", data_mp, NULL);
1727 		freemsg(data_mp);
1728 		goto done;
1729 	}
1730 
1731 	ns = iras.ira_ill->ill_ipst->ips_netstack;
1732 	espstack = ns->netstack_ipsecesp;
1733 	ipss = ns->netstack_ipsec;
1734 
1735 	if (status == CRYPTO_SUCCESS) {
1736 		data_mp = esp_in_done(data_mp, &iras, ic);
1737 		if (data_mp == NULL)
1738 			goto done;
1739 
1740 		/* finish IPsec processing */
1741 		ip_input_post_ipsec(data_mp, &iras);
1742 	} else if (status == CRYPTO_INVALID_MAC) {
1743 		esp_log_bad_auth(data_mp, &iras);
1744 	} else {
1745 		esp1dbg(espstack,
1746 		    ("esp_kcf_callback: crypto failed with 0x%x\n",
1747 		    status));
1748 		ESP_BUMP_STAT(espstack, crypto_failures);
1749 		IP_ESP_BUMP_STAT(ipss, in_discards);
1750 		ip_drop_packet(data_mp, B_TRUE, iras.ira_ill,
1751 		    DROPPER(ipss, ipds_esp_crypto_failed),
1752 		    &espstack->esp_dropper);
1753 		BUMP_MIB(iras.ira_ill->ill_ip_mib, ipIfStatsInDiscards);
1754 	}
1755 done:
1756 	ira_cleanup(&iras, B_TRUE);
1757 	(void) ipsec_free_crypto_data(mp);
1758 }
1759 
1760 /*
1761  * Invoked on crypto framework failure during inbound and outbound processing.
1762  */
1763 static void
1764 esp_crypto_failed(mblk_t *data_mp, boolean_t is_inbound, int kef_rc,
1765     ill_t *ill, ipsecesp_stack_t *espstack)
1766 {
1767 	ipsec_stack_t	*ipss = espstack->ipsecesp_netstack->netstack_ipsec;
1768 
1769 	esp1dbg(espstack, ("crypto failed for %s ESP with 0x%x\n",
1770 	    is_inbound ? "inbound" : "outbound", kef_rc));
1771 	ip_drop_packet(data_mp, is_inbound, ill,
1772 	    DROPPER(ipss, ipds_esp_crypto_failed),
1773 	    &espstack->esp_dropper);
1774 	ESP_BUMP_STAT(espstack, crypto_failures);
1775 	if (is_inbound)
1776 		IP_ESP_BUMP_STAT(ipss, in_discards);
1777 	else
1778 		ESP_BUMP_STAT(espstack, out_discards);
1779 }
1780 
1781 /*
1782  * A statement-equivalent macro, _cr MUST point to a modifiable
1783  * crypto_call_req_t.
1784  */
1785 #define	ESP_INIT_CALLREQ(_cr, _mp, _callback)				\
1786 	(_cr)->cr_flag = CRYPTO_SKIP_REQID|CRYPTO_ALWAYS_QUEUE;	\
1787 	(_cr)->cr_callback_arg = (_mp);				\
1788 	(_cr)->cr_callback_func = (_callback)
1789 
1790 #define	ESP_INIT_CRYPTO_MAC(mac, icvlen, icvbuf) {			\
1791 	(mac)->cd_format = CRYPTO_DATA_RAW;				\
1792 	(mac)->cd_offset = 0;						\
1793 	(mac)->cd_length = icvlen;					\
1794 	(mac)->cd_raw.iov_base = (char *)icvbuf;			\
1795 	(mac)->cd_raw.iov_len = icvlen;					\
1796 }
1797 
1798 #define	ESP_INIT_CRYPTO_DATA(data, mp, off, len) {			\
1799 	if (MBLKL(mp) >= (len) + (off)) {				\
1800 		(data)->cd_format = CRYPTO_DATA_RAW;			\
1801 		(data)->cd_raw.iov_base = (char *)(mp)->b_rptr;		\
1802 		(data)->cd_raw.iov_len = MBLKL(mp);			\
1803 		(data)->cd_offset = off;				\
1804 	} else {							\
1805 		(data)->cd_format = CRYPTO_DATA_MBLK;			\
1806 		(data)->cd_mp = mp;					\
1807 		(data)->cd_offset = off;				\
1808 	}								\
1809 	(data)->cd_length = len;					\
1810 }
1811 
1812 #define	ESP_INIT_CRYPTO_DUAL_DATA(data, mp, off1, len1, off2, len2) {	\
1813 	(data)->dd_format = CRYPTO_DATA_MBLK;				\
1814 	(data)->dd_mp = mp;						\
1815 	(data)->dd_len1 = len1;						\
1816 	(data)->dd_offset1 = off1;					\
1817 	(data)->dd_len2 = len2;						\
1818 	(data)->dd_offset2 = off2;					\
1819 }
1820 
1821 /*
1822  * Returns data_mp if successfully completed the request. Returns
1823  * NULL if it failed (and increments InDiscards) or if it is pending.
1824  */
1825 static mblk_t *
1826 esp_submit_req_inbound(mblk_t *esp_mp, ip_recv_attr_t *ira,
1827     ipsa_t *assoc, uint_t esph_offset)
1828 {
1829 	uint_t auth_offset, msg_len, auth_len;
1830 	crypto_call_req_t call_req, *callrp;
1831 	mblk_t *mp;
1832 	esph_t *esph_ptr;
1833 	int kef_rc;
1834 	uint_t icv_len = assoc->ipsa_mac_len;
1835 	crypto_ctx_template_t auth_ctx_tmpl;
1836 	boolean_t do_auth, do_encr, force;
1837 	uint_t encr_offset, encr_len;
1838 	uint_t iv_len = assoc->ipsa_iv_len;
1839 	crypto_ctx_template_t encr_ctx_tmpl;
1840 	ipsec_crypto_t	*ic, icstack;
1841 	uchar_t *iv_ptr;
1842 	netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
1843 	ipsec_stack_t *ipss = ns->netstack_ipsec;
1844 	ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1845 
1846 	do_auth = assoc->ipsa_auth_alg != SADB_AALG_NONE;
1847 	do_encr = assoc->ipsa_encr_alg != SADB_EALG_NULL;
1848 	force = (assoc->ipsa_flags & IPSA_F_ASYNC);
1849 
1850 #ifdef IPSEC_LATENCY_TEST
1851 	kef_rc = CRYPTO_SUCCESS;
1852 #else
1853 	kef_rc = CRYPTO_FAILED;
1854 #endif
1855 
1856 	/*
1857 	 * An inbound packet is of the form:
1858 	 * [IP,options,ESP,IV,data,ICV,pad]
1859 	 */
1860 	esph_ptr = (esph_t *)(esp_mp->b_rptr + esph_offset);
1861 	iv_ptr = (uchar_t *)(esph_ptr + 1);
1862 	/* Packet length starting at IP header ending after ESP ICV. */
1863 	msg_len = MBLKL(esp_mp);
1864 
1865 	encr_offset = esph_offset + sizeof (esph_t) + iv_len;
1866 	encr_len = msg_len - encr_offset;
1867 
1868 	/*
1869 	 * Counter mode algs need a nonce. This is setup in sadb_common_add().
1870 	 * If for some reason we are using a SA which does not have a nonce
1871 	 * then we must fail here.
1872 	 */
1873 	if ((assoc->ipsa_flags & IPSA_F_COUNTERMODE) &&
1874 	    (assoc->ipsa_nonce == NULL)) {
1875 		ip_drop_packet(esp_mp, B_TRUE, ira->ira_ill,
1876 		    DROPPER(ipss, ipds_esp_nomem), &espstack->esp_dropper);
1877 		return (NULL);
1878 	}
1879 
1880 	if (force) {
1881 		/* We are doing asynch; allocate mblks to hold state */
1882 		if ((mp = ip_recv_attr_to_mblk(ira)) == NULL ||
1883 		    (mp = ipsec_add_crypto_data(mp, &ic)) == NULL) {
1884 			BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
1885 			ip_drop_input("ipIfStatsInDiscards", esp_mp,
1886 			    ira->ira_ill);
1887 			return (NULL);
1888 		}
1889 		linkb(mp, esp_mp);
1890 		callrp = &call_req;
1891 		ESP_INIT_CALLREQ(callrp, mp, esp_kcf_callback_inbound);
1892 	} else {
1893 		/*
1894 		 * If we know we are going to do sync then ipsec_crypto_t
1895 		 * should be on the stack.
1896 		 */
1897 		ic = &icstack;
1898 		bzero(ic, sizeof (*ic));
1899 		callrp = NULL;
1900 	}
1901 
1902 	if (do_auth) {
1903 		/* authentication context template */
1904 		IPSEC_CTX_TMPL(assoc, ipsa_authtmpl, IPSEC_ALG_AUTH,
1905 		    auth_ctx_tmpl);
1906 
1907 		/* ICV to be verified */
1908 		ESP_INIT_CRYPTO_MAC(&ic->ic_crypto_mac,
1909 		    icv_len, esp_mp->b_wptr - icv_len);
1910 
1911 		/* authentication starts at the ESP header */
1912 		auth_offset = esph_offset;
1913 		auth_len = msg_len - auth_offset - icv_len;
1914 		if (!do_encr) {
1915 			/* authentication only */
1916 			/* initialize input data argument */
1917 			ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
1918 			    esp_mp, auth_offset, auth_len);
1919 
1920 			/* call the crypto framework */
1921 			kef_rc = crypto_mac_verify(&assoc->ipsa_amech,
1922 			    &ic->ic_crypto_data,
1923 			    &assoc->ipsa_kcfauthkey, auth_ctx_tmpl,
1924 			    &ic->ic_crypto_mac, callrp);
1925 		}
1926 	}
1927 
1928 	if (do_encr) {
1929 		/* encryption template */
1930 		IPSEC_CTX_TMPL(assoc, ipsa_encrtmpl, IPSEC_ALG_ENCR,
1931 		    encr_ctx_tmpl);
1932 
1933 		/* Call the nonce update function. Also passes in IV */
1934 		(assoc->ipsa_noncefunc)(assoc, (uchar_t *)esph_ptr, encr_len,
1935 		    iv_ptr, &ic->ic_cmm, &ic->ic_crypto_data);
1936 
1937 		if (!do_auth) {
1938 			/* decryption only */
1939 			/* initialize input data argument */
1940 			ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
1941 			    esp_mp, encr_offset, encr_len);
1942 
1943 			/* call the crypto framework */
1944 			kef_rc = crypto_decrypt((crypto_mechanism_t *)
1945 			    &ic->ic_cmm, &ic->ic_crypto_data,
1946 			    &assoc->ipsa_kcfencrkey, encr_ctx_tmpl,
1947 			    NULL, callrp);
1948 		}
1949 	}
1950 
1951 	if (do_auth && do_encr) {
1952 		/* dual operation */
1953 		/* initialize input data argument */
1954 		ESP_INIT_CRYPTO_DUAL_DATA(&ic->ic_crypto_dual_data,
1955 		    esp_mp, auth_offset, auth_len,
1956 		    encr_offset, encr_len - icv_len);
1957 
1958 		/* specify IV */
1959 		ic->ic_crypto_dual_data.dd_miscdata = (char *)iv_ptr;
1960 
1961 		/* call the framework */
1962 		kef_rc = crypto_mac_verify_decrypt(&assoc->ipsa_amech,
1963 		    &assoc->ipsa_emech, &ic->ic_crypto_dual_data,
1964 		    &assoc->ipsa_kcfauthkey, &assoc->ipsa_kcfencrkey,
1965 		    auth_ctx_tmpl, encr_ctx_tmpl, &ic->ic_crypto_mac,
1966 		    NULL, callrp);
1967 	}
1968 
1969 	switch (kef_rc) {
1970 	case CRYPTO_SUCCESS:
1971 		ESP_BUMP_STAT(espstack, crypto_sync);
1972 		esp_mp = esp_in_done(esp_mp, ira, ic);
1973 		if (force) {
1974 			/* Free mp after we are done with ic */
1975 			mp = ipsec_free_crypto_data(mp);
1976 			(void) ip_recv_attr_free_mblk(mp);
1977 		}
1978 		return (esp_mp);
1979 	case CRYPTO_QUEUED:
1980 		/* esp_kcf_callback_inbound() will be invoked on completion */
1981 		ESP_BUMP_STAT(espstack, crypto_async);
1982 		return (NULL);
1983 	case CRYPTO_INVALID_MAC:
1984 		if (force) {
1985 			mp = ipsec_free_crypto_data(mp);
1986 			esp_mp = ip_recv_attr_free_mblk(mp);
1987 		}
1988 		ESP_BUMP_STAT(espstack, crypto_sync);
1989 		BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
1990 		esp_log_bad_auth(esp_mp, ira);
1991 		/* esp_mp was passed to ip_drop_packet */
1992 		return (NULL);
1993 	}
1994 
1995 	if (force) {
1996 		mp = ipsec_free_crypto_data(mp);
1997 		esp_mp = ip_recv_attr_free_mblk(mp);
1998 	}
1999 	BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
2000 	esp_crypto_failed(esp_mp, B_TRUE, kef_rc, ira->ira_ill, espstack);
2001 	/* esp_mp was passed to ip_drop_packet */
2002 	return (NULL);
2003 }
2004 
2005 /*
2006  * Compute the IP and UDP checksums -- common code for both keepalives and
2007  * actual ESP-in-UDP packets.  Be flexible with multiple mblks because ESP
2008  * uses mblk-insertion to insert the UDP header.
2009  * TODO - If there is an easy way to prep a packet for HW checksums, make
2010  * it happen here.
2011  * Note that this is used before both before calling ip_output_simple and
2012  * in the esp datapath. The former could use IXAF_SET_ULP_CKSUM but not the
2013  * latter.
2014  */
2015 static void
2016 esp_prepare_udp(netstack_t *ns, mblk_t *mp, ipha_t *ipha)
2017 {
2018 	int offset;
2019 	uint32_t cksum;
2020 	uint16_t *arr;
2021 	mblk_t *udpmp = mp;
2022 	uint_t hlen = IPH_HDR_LENGTH(ipha);
2023 
2024 	ASSERT(MBLKL(mp) >= sizeof (ipha_t));
2025 
2026 	ipha->ipha_hdr_checksum = 0;
2027 	ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
2028 
2029 	if (ns->netstack_udp->us_do_checksum) {
2030 		ASSERT(MBLKL(udpmp) >= sizeof (udpha_t));
2031 		/* arr points to the IP header. */
2032 		arr = (uint16_t *)ipha;
2033 		IP_STAT(ns->netstack_ip, ip_out_sw_cksum);
2034 		IP_STAT_UPDATE(ns->netstack_ip, ip_out_sw_cksum_bytes,
2035 		    ntohs(htons(ipha->ipha_length) - hlen));
2036 		/* arr[6-9] are the IP addresses. */
2037 		cksum = IP_UDP_CSUM_COMP + arr[6] + arr[7] + arr[8] + arr[9] +
2038 		    ntohs(htons(ipha->ipha_length) - hlen);
2039 		cksum = IP_CSUM(mp, hlen, cksum);
2040 		offset = hlen + UDP_CHECKSUM_OFFSET;
2041 		while (offset >= MBLKL(udpmp)) {
2042 			offset -= MBLKL(udpmp);
2043 			udpmp = udpmp->b_cont;
2044 		}
2045 		/* arr points to the UDP header's checksum field. */
2046 		arr = (uint16_t *)(udpmp->b_rptr + offset);
2047 		*arr = cksum;
2048 	}
2049 }
2050 
2051 /*
2052  * taskq handler so we can send the NAT-T keepalive on a separate thread.
2053  */
2054 static void
2055 actually_send_keepalive(void *arg)
2056 {
2057 	mblk_t *mp = (mblk_t *)arg;
2058 	ip_xmit_attr_t ixas;
2059 	netstack_t	*ns;
2060 	netstackid_t	stackid;
2061 
2062 	stackid = (netstackid_t)(uintptr_t)mp->b_prev;
2063 	mp->b_prev = NULL;
2064 	ns = netstack_find_by_stackid(stackid);
2065 	if (ns == NULL) {
2066 		/* Disappeared */
2067 		ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
2068 		freemsg(mp);
2069 		return;
2070 	}
2071 
2072 	bzero(&ixas, sizeof (ixas));
2073 	ixas.ixa_zoneid = ALL_ZONES;
2074 	ixas.ixa_cred = kcred;
2075 	ixas.ixa_cpid = NOPID;
2076 	ixas.ixa_tsl = NULL;
2077 	ixas.ixa_ipst = ns->netstack_ip;
2078 	/* No ULP checksum; done by esp_prepare_udp */
2079 	ixas.ixa_flags = (IXAF_IS_IPV4 | IXAF_NO_IPSEC | IXAF_VERIFY_SOURCE);
2080 
2081 	(void) ip_output_simple(mp, &ixas);
2082 	ixa_cleanup(&ixas);
2083 	netstack_rele(ns);
2084 }
2085 
2086 /*
2087  * Send a one-byte UDP NAT-T keepalive.
2088  */
2089 void
2090 ipsecesp_send_keepalive(ipsa_t *assoc)
2091 {
2092 	mblk_t		*mp;
2093 	ipha_t		*ipha;
2094 	udpha_t		*udpha;
2095 	netstack_t	*ns = assoc->ipsa_netstack;
2096 
2097 	ASSERT(MUTEX_NOT_HELD(&assoc->ipsa_lock));
2098 
2099 	mp = allocb(sizeof (ipha_t) + sizeof (udpha_t) + 1, BPRI_HI);
2100 	if (mp == NULL)
2101 		return;
2102 	ipha = (ipha_t *)mp->b_rptr;
2103 	ipha->ipha_version_and_hdr_length = IP_SIMPLE_HDR_VERSION;
2104 	ipha->ipha_type_of_service = 0;
2105 	ipha->ipha_length = htons(sizeof (ipha_t) + sizeof (udpha_t) + 1);
2106 	/* Use the low-16 of the SPI so we have some clue where it came from. */
2107 	ipha->ipha_ident = *(((uint16_t *)(&assoc->ipsa_spi)) + 1);
2108 	ipha->ipha_fragment_offset_and_flags = 0;  /* Too small to fragment! */
2109 	ipha->ipha_ttl = 0xFF;
2110 	ipha->ipha_protocol = IPPROTO_UDP;
2111 	ipha->ipha_hdr_checksum = 0;
2112 	ipha->ipha_src = assoc->ipsa_srcaddr[0];
2113 	ipha->ipha_dst = assoc->ipsa_dstaddr[0];
2114 	udpha = (udpha_t *)(ipha + 1);
2115 	udpha->uha_src_port = (assoc->ipsa_local_nat_port != 0) ?
2116 	    assoc->ipsa_local_nat_port : htons(IPPORT_IKE_NATT);
2117 	udpha->uha_dst_port = (assoc->ipsa_remote_nat_port != 0) ?
2118 	    assoc->ipsa_remote_nat_port : htons(IPPORT_IKE_NATT);
2119 	udpha->uha_length = htons(sizeof (udpha_t) + 1);
2120 	udpha->uha_checksum = 0;
2121 	mp->b_wptr = (uint8_t *)(udpha + 1);
2122 	*(mp->b_wptr++) = 0xFF;
2123 
2124 	esp_prepare_udp(ns, mp, ipha);
2125 
2126 	/*
2127 	 * We're holding an isaf_t bucket lock, so pawn off the actual
2128 	 * packet transmission to another thread.  Just in case syncq
2129 	 * processing causes a same-bucket packet to be processed.
2130 	 */
2131 	mp->b_prev = (mblk_t *)(uintptr_t)ns->netstack_stackid;
2132 
2133 	if (taskq_dispatch(esp_taskq, actually_send_keepalive, mp,
2134 	    TQ_NOSLEEP) == TASKQID_INVALID) {
2135 		/* Assume no memory if taskq_dispatch() fails. */
2136 		mp->b_prev = NULL;
2137 		ip_drop_packet(mp, B_FALSE, NULL,
2138 		    DROPPER(ns->netstack_ipsec, ipds_esp_nomem),
2139 		    &ns->netstack_ipsecesp->esp_dropper);
2140 	}
2141 }
2142 
2143 /*
2144  * Returns mp if successfully completed the request. Returns
2145  * NULL if it failed (and increments InDiscards) or if it is pending.
2146  */
2147 static mblk_t *
2148 esp_submit_req_outbound(mblk_t *data_mp, ip_xmit_attr_t *ixa, ipsa_t *assoc,
2149     uchar_t *icv_buf, uint_t payload_len)
2150 {
2151 	uint_t auth_len;
2152 	crypto_call_req_t call_req, *callrp;
2153 	mblk_t *esp_mp;
2154 	esph_t *esph_ptr;
2155 	mblk_t *mp;
2156 	int kef_rc = CRYPTO_FAILED;
2157 	uint_t icv_len = assoc->ipsa_mac_len;
2158 	crypto_ctx_template_t auth_ctx_tmpl;
2159 	boolean_t do_auth, do_encr, force;
2160 	uint_t iv_len = assoc->ipsa_iv_len;
2161 	crypto_ctx_template_t encr_ctx_tmpl;
2162 	boolean_t is_natt = ((assoc->ipsa_flags & IPSA_F_NATT) != 0);
2163 	size_t esph_offset = (is_natt ? UDPH_SIZE : 0);
2164 	netstack_t	*ns = ixa->ixa_ipst->ips_netstack;
2165 	ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
2166 	ipsec_crypto_t	*ic, icstack;
2167 	uchar_t		*iv_ptr;
2168 	crypto_data_t	*cd_ptr = NULL;
2169 	ill_t		*ill = ixa->ixa_nce->nce_ill;
2170 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
2171 
2172 	esp3dbg(espstack, ("esp_submit_req_outbound:%s",
2173 	    is_natt ? "natt" : "not natt"));
2174 
2175 	do_encr = assoc->ipsa_encr_alg != SADB_EALG_NULL;
2176 	do_auth = assoc->ipsa_auth_alg != SADB_AALG_NONE;
2177 	force = (assoc->ipsa_flags & IPSA_F_ASYNC);
2178 
2179 #ifdef IPSEC_LATENCY_TEST
2180 	kef_rc = CRYPTO_SUCCESS;
2181 #else
2182 	kef_rc = CRYPTO_FAILED;
2183 #endif
2184 
2185 	/*
2186 	 * Outbound IPsec packets are of the form:
2187 	 * [IP,options] -> [ESP,IV] -> [data] -> [pad,ICV]
2188 	 * unless it's NATT, then it's
2189 	 * [IP,options] -> [udp][ESP,IV] -> [data] -> [pad,ICV]
2190 	 * Get a pointer to the mblk containing the ESP header.
2191 	 */
2192 	ASSERT(data_mp->b_cont != NULL);
2193 	esp_mp = data_mp->b_cont;
2194 	esph_ptr = (esph_t *)(esp_mp->b_rptr + esph_offset);
2195 	iv_ptr = (uchar_t *)(esph_ptr + 1);
2196 
2197 	/*
2198 	 * Combined mode algs need a nonce. This is setup in sadb_common_add().
2199 	 * If for some reason we are using a SA which does not have a nonce
2200 	 * then we must fail here.
2201 	 */
2202 	if ((assoc->ipsa_flags & IPSA_F_COUNTERMODE) &&
2203 	    (assoc->ipsa_nonce == NULL)) {
2204 		ip_drop_packet(data_mp, B_FALSE, NULL,
2205 		    DROPPER(ipss, ipds_esp_nomem), &espstack->esp_dropper);
2206 		return (NULL);
2207 	}
2208 
2209 	if (force) {
2210 		/* We are doing asynch; allocate mblks to hold state */
2211 		if ((mp = ip_xmit_attr_to_mblk(ixa)) == NULL ||
2212 		    (mp = ipsec_add_crypto_data(mp, &ic)) == NULL) {
2213 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2214 			ip_drop_output("ipIfStatsOutDiscards", data_mp, ill);
2215 			freemsg(data_mp);
2216 			return (NULL);
2217 		}
2218 
2219 		linkb(mp, data_mp);
2220 		callrp = &call_req;
2221 		ESP_INIT_CALLREQ(callrp, mp, esp_kcf_callback_outbound);
2222 	} else {
2223 		/*
2224 		 * If we know we are going to do sync then ipsec_crypto_t
2225 		 * should be on the stack.
2226 		 */
2227 		ic = &icstack;
2228 		bzero(ic, sizeof (*ic));
2229 		callrp = NULL;
2230 	}
2231 
2232 
2233 	if (do_auth) {
2234 		/* authentication context template */
2235 		IPSEC_CTX_TMPL(assoc, ipsa_authtmpl, IPSEC_ALG_AUTH,
2236 		    auth_ctx_tmpl);
2237 
2238 		/* where to store the computed mac */
2239 		ESP_INIT_CRYPTO_MAC(&ic->ic_crypto_mac,
2240 		    icv_len, icv_buf);
2241 
2242 		/* authentication starts at the ESP header */
2243 		auth_len = payload_len + iv_len + sizeof (esph_t);
2244 		if (!do_encr) {
2245 			/* authentication only */
2246 			/* initialize input data argument */
2247 			ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
2248 			    esp_mp, esph_offset, auth_len);
2249 
2250 			/* call the crypto framework */
2251 			kef_rc = crypto_mac(&assoc->ipsa_amech,
2252 			    &ic->ic_crypto_data,
2253 			    &assoc->ipsa_kcfauthkey, auth_ctx_tmpl,
2254 			    &ic->ic_crypto_mac, callrp);
2255 		}
2256 	}
2257 
2258 	if (do_encr) {
2259 		/* encryption context template */
2260 		IPSEC_CTX_TMPL(assoc, ipsa_encrtmpl, IPSEC_ALG_ENCR,
2261 		    encr_ctx_tmpl);
2262 		/* Call the nonce update function. */
2263 		(assoc->ipsa_noncefunc)(assoc, (uchar_t *)esph_ptr, payload_len,
2264 		    iv_ptr, &ic->ic_cmm, &ic->ic_crypto_data);
2265 
2266 		if (!do_auth) {
2267 			/* encryption only, skip mblk that contains ESP hdr */
2268 			/* initialize input data argument */
2269 			ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
2270 			    esp_mp->b_cont, 0, payload_len);
2271 
2272 			/*
2273 			 * For combined mode ciphers, the ciphertext is the same
2274 			 * size as the clear text, the ICV should follow the
2275 			 * ciphertext. To convince the kcf to allow in-line
2276 			 * encryption, with an ICV, use ipsec_out_crypto_mac
2277 			 * to point to the same buffer as the data. The calling
2278 			 * function need to ensure the buffer is large enough to
2279 			 * include the ICV.
2280 			 *
2281 			 * The IV is already written to the packet buffer, the
2282 			 * nonce setup function copied it to the params struct
2283 			 * for the cipher to use.
2284 			 */
2285 			if (assoc->ipsa_flags & IPSA_F_COMBINED) {
2286 				bcopy(&ic->ic_crypto_data,
2287 				    &ic->ic_crypto_mac,
2288 				    sizeof (crypto_data_t));
2289 				ic->ic_crypto_mac.cd_length =
2290 				    payload_len + icv_len;
2291 				cd_ptr = &ic->ic_crypto_mac;
2292 			}
2293 
2294 			/* call the crypto framework */
2295 			kef_rc = crypto_encrypt((crypto_mechanism_t *)
2296 			    &ic->ic_cmm, &ic->ic_crypto_data,
2297 			    &assoc->ipsa_kcfencrkey, encr_ctx_tmpl,
2298 			    cd_ptr, callrp);
2299 
2300 		}
2301 	}
2302 
2303 	if (do_auth && do_encr) {
2304 		/*
2305 		 * Encryption and authentication:
2306 		 * Pass the pointer to the mblk chain starting at the ESP
2307 		 * header to the framework. Skip the ESP header mblk
2308 		 * for encryption, which is reflected by an encryption
2309 		 * offset equal to the length of that mblk. Start
2310 		 * the authentication at the ESP header, i.e. use an
2311 		 * authentication offset of zero.
2312 		 */
2313 		ESP_INIT_CRYPTO_DUAL_DATA(&ic->ic_crypto_dual_data,
2314 		    esp_mp, MBLKL(esp_mp), payload_len, esph_offset, auth_len);
2315 
2316 		/* specify IV */
2317 		ic->ic_crypto_dual_data.dd_miscdata = (char *)iv_ptr;
2318 
2319 		/* call the framework */
2320 		kef_rc = crypto_encrypt_mac(&assoc->ipsa_emech,
2321 		    &assoc->ipsa_amech, NULL,
2322 		    &assoc->ipsa_kcfencrkey, &assoc->ipsa_kcfauthkey,
2323 		    encr_ctx_tmpl, auth_ctx_tmpl,
2324 		    &ic->ic_crypto_dual_data,
2325 		    &ic->ic_crypto_mac, callrp);
2326 	}
2327 
2328 	switch (kef_rc) {
2329 	case CRYPTO_SUCCESS:
2330 		ESP_BUMP_STAT(espstack, crypto_sync);
2331 		esp_set_usetime(assoc, B_FALSE);
2332 		if (force) {
2333 			mp = ipsec_free_crypto_data(mp);
2334 			data_mp = ip_xmit_attr_free_mblk(mp);
2335 		}
2336 		if (is_natt)
2337 			esp_prepare_udp(ns, data_mp, (ipha_t *)data_mp->b_rptr);
2338 		return (data_mp);
2339 	case CRYPTO_QUEUED:
2340 		/* esp_kcf_callback_outbound() will be invoked on completion */
2341 		ESP_BUMP_STAT(espstack, crypto_async);
2342 		return (NULL);
2343 	}
2344 
2345 	if (force) {
2346 		mp = ipsec_free_crypto_data(mp);
2347 		data_mp = ip_xmit_attr_free_mblk(mp);
2348 	}
2349 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2350 	esp_crypto_failed(data_mp, B_FALSE, kef_rc, NULL, espstack);
2351 	/* data_mp was passed to ip_drop_packet */
2352 	return (NULL);
2353 }
2354 
2355 /*
2356  * Handle outbound IPsec processing for IPv4 and IPv6
2357  *
2358  * Returns data_mp if successfully completed the request. Returns
2359  * NULL if it failed (and increments InDiscards) or if it is pending.
2360  */
2361 static mblk_t *
2362 esp_outbound(mblk_t *data_mp, ip_xmit_attr_t *ixa)
2363 {
2364 	mblk_t *espmp, *tailmp;
2365 	ipha_t *ipha;
2366 	ip6_t *ip6h;
2367 	esph_t *esph_ptr, *iv_ptr;
2368 	uint_t af;
2369 	uint8_t *nhp;
2370 	uintptr_t divpoint, datalen, adj, padlen, i, alloclen;
2371 	uintptr_t esplen = sizeof (esph_t);
2372 	uint8_t protocol;
2373 	ipsa_t *assoc;
2374 	uint_t iv_len, block_size, mac_len = 0;
2375 	uchar_t *icv_buf;
2376 	udpha_t *udpha;
2377 	boolean_t is_natt = B_FALSE;
2378 	netstack_t	*ns = ixa->ixa_ipst->ips_netstack;
2379 	ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
2380 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
2381 	ill_t		*ill = ixa->ixa_nce->nce_ill;
2382 	boolean_t	need_refrele = B_FALSE;
2383 
2384 	ESP_BUMP_STAT(espstack, out_requests);
2385 
2386 	/*
2387 	 * <sigh> We have to copy the message here, because TCP (for example)
2388 	 * keeps a dupb() of the message lying around for retransmission.
2389 	 * Since ESP changes the whole of the datagram, we have to create our
2390 	 * own copy lest we clobber TCP's data.  Since we have to copy anyway,
2391 	 * we might as well make use of msgpullup() and get the mblk into one
2392 	 * contiguous piece!
2393 	 */
2394 	tailmp = msgpullup(data_mp, -1);
2395 	if (tailmp == NULL) {
2396 		esp0dbg(("esp_outbound: msgpullup() failed, "
2397 		    "dropping packet.\n"));
2398 		ip_drop_packet(data_mp, B_FALSE, ill,
2399 		    DROPPER(ipss, ipds_esp_nomem),
2400 		    &espstack->esp_dropper);
2401 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2402 		return (NULL);
2403 	}
2404 	freemsg(data_mp);
2405 	data_mp = tailmp;
2406 
2407 	assoc = ixa->ixa_ipsec_esp_sa;
2408 	ASSERT(assoc != NULL);
2409 
2410 	/*
2411 	 * Get the outer IP header in shape to escape this system..
2412 	 */
2413 	if (is_system_labeled() && (assoc->ipsa_otsl != NULL)) {
2414 		/*
2415 		 * Need to update packet with any CIPSO option and update
2416 		 * ixa_tsl to capture the new label.
2417 		 * We allocate a separate ixa for that purpose.
2418 		 */
2419 		ixa = ip_xmit_attr_duplicate(ixa);
2420 		if (ixa == NULL) {
2421 			ip_drop_packet(data_mp, B_FALSE, ill,
2422 			    DROPPER(ipss, ipds_esp_nomem),
2423 			    &espstack->esp_dropper);
2424 			return (NULL);
2425 		}
2426 		need_refrele = B_TRUE;
2427 
2428 		label_hold(assoc->ipsa_otsl);
2429 		ip_xmit_attr_replace_tsl(ixa, assoc->ipsa_otsl);
2430 
2431 		data_mp = sadb_whack_label(data_mp, assoc, ixa,
2432 		    DROPPER(ipss, ipds_esp_nomem), &espstack->esp_dropper);
2433 		if (data_mp == NULL) {
2434 			/* Packet dropped by sadb_whack_label */
2435 			ixa_refrele(ixa);
2436 			return (NULL);
2437 		}
2438 	}
2439 
2440 	/*
2441 	 * Reality check....
2442 	 */
2443 	ipha = (ipha_t *)data_mp->b_rptr;  /* So we can call esp_acquire(). */
2444 
2445 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
2446 		ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
2447 
2448 		af = AF_INET;
2449 		divpoint = IPH_HDR_LENGTH(ipha);
2450 		datalen = ntohs(ipha->ipha_length) - divpoint;
2451 		nhp = (uint8_t *)&ipha->ipha_protocol;
2452 	} else {
2453 		ip_pkt_t ipp;
2454 
2455 		ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
2456 
2457 		af = AF_INET6;
2458 		ip6h = (ip6_t *)ipha;
2459 		bzero(&ipp, sizeof (ipp));
2460 		divpoint = ip_find_hdr_v6(data_mp, ip6h, B_FALSE, &ipp, NULL);
2461 		if (ipp.ipp_dstopts != NULL &&
2462 		    ipp.ipp_dstopts->ip6d_nxt != IPPROTO_ROUTING) {
2463 			/*
2464 			 * Destination options are tricky.  If we get in here,
2465 			 * then we have a terminal header following the
2466 			 * destination options.  We need to adjust backwards
2467 			 * so we insert ESP BEFORE the destination options
2468 			 * bag.  (So that the dstopts get encrypted!)
2469 			 *
2470 			 * Since this is for outbound packets only, we know
2471 			 * that non-terminal destination options only precede
2472 			 * routing headers.
2473 			 */
2474 			divpoint -= ipp.ipp_dstoptslen;
2475 		}
2476 		datalen = ntohs(ip6h->ip6_plen) + sizeof (ip6_t) - divpoint;
2477 
2478 		if (ipp.ipp_rthdr != NULL) {
2479 			nhp = &ipp.ipp_rthdr->ip6r_nxt;
2480 		} else if (ipp.ipp_hopopts != NULL) {
2481 			nhp = &ipp.ipp_hopopts->ip6h_nxt;
2482 		} else {
2483 			ASSERT(divpoint == sizeof (ip6_t));
2484 			/* It's probably IP + ESP. */
2485 			nhp = &ip6h->ip6_nxt;
2486 		}
2487 	}
2488 
2489 	mac_len = assoc->ipsa_mac_len;
2490 
2491 	if (assoc->ipsa_flags & IPSA_F_NATT) {
2492 		/* wedge in UDP header */
2493 		is_natt = B_TRUE;
2494 		esplen += UDPH_SIZE;
2495 	}
2496 
2497 	/*
2498 	 * Set up ESP header and encryption padding for ENCR PI request.
2499 	 */
2500 
2501 	/* Determine the padding length.  Pad to 4-bytes for no-encryption. */
2502 	if (assoc->ipsa_encr_alg != SADB_EALG_NULL) {
2503 		iv_len = assoc->ipsa_iv_len;
2504 		block_size = assoc->ipsa_datalen;
2505 
2506 		/*
2507 		 * Pad the data to the length of the cipher block size.
2508 		 * Include the two additional bytes (hence the - 2) for the
2509 		 * padding length and the next header.  Take this into account
2510 		 * when calculating the actual length of the padding.
2511 		 */
2512 		ASSERT(ISP2(iv_len));
2513 		padlen = ((unsigned)(block_size - datalen - 2)) &
2514 		    (block_size - 1);
2515 	} else {
2516 		iv_len = 0;
2517 		padlen = ((unsigned)(sizeof (uint32_t) - datalen - 2)) &
2518 		    (sizeof (uint32_t) - 1);
2519 	}
2520 
2521 	/* Allocate ESP header and IV. */
2522 	esplen += iv_len;
2523 
2524 	/*
2525 	 * Update association byte-count lifetimes.  Don't forget to take
2526 	 * into account the padding length and next-header (hence the + 2).
2527 	 *
2528 	 * Use the amount of data fed into the "encryption algorithm".  This
2529 	 * is the IV, the data length, the padding length, and the final two
2530 	 * bytes (padlen, and next-header).
2531 	 *
2532 	 */
2533 
2534 	if (!esp_age_bytes(assoc, datalen + padlen + iv_len + 2, B_FALSE)) {
2535 		ip_drop_packet(data_mp, B_FALSE, ill,
2536 		    DROPPER(ipss, ipds_esp_bytes_expire),
2537 		    &espstack->esp_dropper);
2538 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2539 		if (need_refrele)
2540 			ixa_refrele(ixa);
2541 		return (NULL);
2542 	}
2543 
2544 	espmp = allocb(esplen, BPRI_HI);
2545 	if (espmp == NULL) {
2546 		ESP_BUMP_STAT(espstack, out_discards);
2547 		esp1dbg(espstack, ("esp_outbound: can't allocate espmp.\n"));
2548 		ip_drop_packet(data_mp, B_FALSE, ill,
2549 		    DROPPER(ipss, ipds_esp_nomem),
2550 		    &espstack->esp_dropper);
2551 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2552 		if (need_refrele)
2553 			ixa_refrele(ixa);
2554 		return (NULL);
2555 	}
2556 	espmp->b_wptr += esplen;
2557 	esph_ptr = (esph_t *)espmp->b_rptr;
2558 
2559 	if (is_natt) {
2560 		esp3dbg(espstack, ("esp_outbound: NATT"));
2561 
2562 		udpha = (udpha_t *)espmp->b_rptr;
2563 		udpha->uha_src_port = (assoc->ipsa_local_nat_port != 0) ?
2564 		    assoc->ipsa_local_nat_port : htons(IPPORT_IKE_NATT);
2565 		udpha->uha_dst_port = (assoc->ipsa_remote_nat_port != 0) ?
2566 		    assoc->ipsa_remote_nat_port : htons(IPPORT_IKE_NATT);
2567 		/*
2568 		 * Set the checksum to 0, so that the esp_prepare_udp() call
2569 		 * can do the right thing.
2570 		 */
2571 		udpha->uha_checksum = 0;
2572 		esph_ptr = (esph_t *)(udpha + 1);
2573 	}
2574 
2575 	esph_ptr->esph_spi = assoc->ipsa_spi;
2576 
2577 	esph_ptr->esph_replay = htonl(atomic_inc_32_nv(&assoc->ipsa_replay));
2578 	if (esph_ptr->esph_replay == 0 && assoc->ipsa_replay_wsize != 0) {
2579 		/*
2580 		 * XXX We have replay counter wrapping.
2581 		 * We probably want to nuke this SA (and its peer).
2582 		 */
2583 		ipsec_assocfailure(info.mi_idnum, 0, 0,
2584 		    SL_ERROR | SL_CONSOLE | SL_WARN,
2585 		    "Outbound ESP SA (0x%x, %s) has wrapped sequence.\n",
2586 		    esph_ptr->esph_spi, assoc->ipsa_dstaddr, af,
2587 		    espstack->ipsecesp_netstack);
2588 
2589 		ESP_BUMP_STAT(espstack, out_discards);
2590 		sadb_replay_delete(assoc);
2591 		ip_drop_packet(data_mp, B_FALSE, ill,
2592 		    DROPPER(ipss, ipds_esp_replay),
2593 		    &espstack->esp_dropper);
2594 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2595 		if (need_refrele)
2596 			ixa_refrele(ixa);
2597 		return (NULL);
2598 	}
2599 
2600 	iv_ptr = (esph_ptr + 1);
2601 	/*
2602 	 * iv_ptr points to the mblk which will contain the IV once we have
2603 	 * written it there. This mblk will be part of a mblk chain that
2604 	 * will make up the packet.
2605 	 *
2606 	 * For counter mode algorithms, the IV is a 64 bit quantity, it
2607 	 * must NEVER repeat in the lifetime of the SA, otherwise an
2608 	 * attacker who had recorded enough packets might be able to
2609 	 * determine some clear text.
2610 	 *
2611 	 * To ensure this does not happen, the IV is stored in the SA and
2612 	 * incremented for each packet, the IV is then copied into the
2613 	 * "packet" for transmission to the receiving system. The IV will
2614 	 * also be copied into the nonce, when the packet is encrypted.
2615 	 *
2616 	 * CBC mode algorithms use a random IV for each packet. We do not
2617 	 * require the highest quality random bits, but for best security
2618 	 * with CBC mode ciphers, the value must be unlikely to repeat and
2619 	 * must not be known in advance to an adversary capable of influencing
2620 	 * the clear text.
2621 	 */
2622 	if (!update_iv((uint8_t *)iv_ptr, espstack->esp_pfkey_q, assoc,
2623 	    espstack)) {
2624 		ip_drop_packet(data_mp, B_FALSE, ill,
2625 		    DROPPER(ipss, ipds_esp_iv_wrap), &espstack->esp_dropper);
2626 		if (need_refrele)
2627 			ixa_refrele(ixa);
2628 		return (NULL);
2629 	}
2630 
2631 	/* Fix the IP header. */
2632 	alloclen = padlen + 2 + mac_len;
2633 	adj = alloclen + (espmp->b_wptr - espmp->b_rptr);
2634 
2635 	protocol = *nhp;
2636 
2637 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
2638 		ipha->ipha_length = htons(ntohs(ipha->ipha_length) + adj);
2639 		if (is_natt) {
2640 			*nhp = IPPROTO_UDP;
2641 			udpha->uha_length = htons(ntohs(ipha->ipha_length) -
2642 			    IPH_HDR_LENGTH(ipha));
2643 		} else {
2644 			*nhp = IPPROTO_ESP;
2645 		}
2646 		ipha->ipha_hdr_checksum = 0;
2647 		ipha->ipha_hdr_checksum = (uint16_t)ip_csum_hdr(ipha);
2648 	} else {
2649 		ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) + adj);
2650 		*nhp = IPPROTO_ESP;
2651 	}
2652 
2653 	/* I've got the two ESP mblks, now insert them. */
2654 
2655 	esp2dbg(espstack, ("data_mp before outbound ESP adjustment:\n"));
2656 	esp2dbg(espstack, (dump_msg(data_mp)));
2657 
2658 	if (!esp_insert_esp(data_mp, espmp, divpoint, espstack)) {
2659 		ESP_BUMP_STAT(espstack, out_discards);
2660 		/* NOTE:  esp_insert_esp() only fails if there's no memory. */
2661 		ip_drop_packet(data_mp, B_FALSE, ill,
2662 		    DROPPER(ipss, ipds_esp_nomem),
2663 		    &espstack->esp_dropper);
2664 		freeb(espmp);
2665 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2666 		if (need_refrele)
2667 			ixa_refrele(ixa);
2668 		return (NULL);
2669 	}
2670 
2671 	/* Append padding (and leave room for ICV). */
2672 	for (tailmp = data_mp; tailmp->b_cont != NULL; tailmp = tailmp->b_cont)
2673 		;
2674 	if (tailmp->b_wptr + alloclen > tailmp->b_datap->db_lim) {
2675 		tailmp->b_cont = allocb(alloclen, BPRI_HI);
2676 		if (tailmp->b_cont == NULL) {
2677 			ESP_BUMP_STAT(espstack, out_discards);
2678 			esp0dbg(("esp_outbound:  Can't allocate tailmp.\n"));
2679 			ip_drop_packet(data_mp, B_FALSE, ill,
2680 			    DROPPER(ipss, ipds_esp_nomem),
2681 			    &espstack->esp_dropper);
2682 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2683 			if (need_refrele)
2684 				ixa_refrele(ixa);
2685 			return (NULL);
2686 		}
2687 		tailmp = tailmp->b_cont;
2688 	}
2689 
2690 	/*
2691 	 * If there's padding, N bytes of padding must be of the form 0x1,
2692 	 * 0x2, 0x3... 0xN.
2693 	 */
2694 	for (i = 0; i < padlen; ) {
2695 		i++;
2696 		*tailmp->b_wptr++ = i;
2697 	}
2698 	*tailmp->b_wptr++ = i;
2699 	*tailmp->b_wptr++ = protocol;
2700 
2701 	esp2dbg(espstack, ("data_Mp before encryption:\n"));
2702 	esp2dbg(espstack, (dump_msg(data_mp)));
2703 
2704 	/*
2705 	 * Okay.  I've set up the pre-encryption ESP.  Let's do it!
2706 	 */
2707 
2708 	if (mac_len > 0) {
2709 		ASSERT(tailmp->b_wptr + mac_len <= tailmp->b_datap->db_lim);
2710 		icv_buf = tailmp->b_wptr;
2711 		tailmp->b_wptr += mac_len;
2712 	} else {
2713 		icv_buf = NULL;
2714 	}
2715 
2716 	data_mp = esp_submit_req_outbound(data_mp, ixa, assoc, icv_buf,
2717 	    datalen + padlen + 2);
2718 	if (need_refrele)
2719 		ixa_refrele(ixa);
2720 	return (data_mp);
2721 }
2722 
2723 /*
2724  * IP calls this to validate the ICMP errors that
2725  * we got from the network.
2726  */
2727 mblk_t *
2728 ipsecesp_icmp_error(mblk_t *data_mp, ip_recv_attr_t *ira)
2729 {
2730 	netstack_t	*ns = ira->ira_ill->ill_ipst->ips_netstack;
2731 	ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
2732 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
2733 
2734 	/*
2735 	 * Unless we get an entire packet back, this function is useless.
2736 	 * Why?
2737 	 *
2738 	 * 1.)	Partial packets are useless, because the "next header"
2739 	 *	is at the end of the decrypted ESP packet.  Without the
2740 	 *	whole packet, this is useless.
2741 	 *
2742 	 * 2.)	If we every use a stateful cipher, such as a stream or a
2743 	 *	one-time pad, we can't do anything.
2744 	 *
2745 	 * Since the chances of us getting an entire packet back are very
2746 	 * very small, we discard here.
2747 	 */
2748 	IP_ESP_BUMP_STAT(ipss, in_discards);
2749 	ip_drop_packet(data_mp, B_TRUE, ira->ira_ill,
2750 	    DROPPER(ipss, ipds_esp_icmp),
2751 	    &espstack->esp_dropper);
2752 	return (NULL);
2753 }
2754 
2755 /*
2756  * Construct an SADB_REGISTER message with the current algorithms.
2757  * This function gets called when 'ipsecalgs -s' is run or when
2758  * in.iked (or other KMD) starts.
2759  */
2760 static boolean_t
2761 esp_register_out(uint32_t sequence, uint32_t pid, uint_t serial,
2762     ipsecesp_stack_t *espstack, cred_t *cr)
2763 {
2764 	mblk_t *pfkey_msg_mp, *keysock_out_mp;
2765 	sadb_msg_t *samsg;
2766 	sadb_supported_t *sasupp_auth = NULL;
2767 	sadb_supported_t *sasupp_encr = NULL;
2768 	sadb_alg_t *saalg;
2769 	uint_t allocsize = sizeof (*samsg);
2770 	uint_t i, numalgs_snap;
2771 	int current_aalgs;
2772 	ipsec_alginfo_t **authalgs;
2773 	uint_t num_aalgs;
2774 	int current_ealgs;
2775 	ipsec_alginfo_t **encralgs;
2776 	uint_t num_ealgs;
2777 	ipsec_stack_t	*ipss = espstack->ipsecesp_netstack->netstack_ipsec;
2778 	sadb_sens_t *sens;
2779 	size_t sens_len = 0;
2780 	sadb_ext_t *nextext;
2781 	ts_label_t *sens_tsl = NULL;
2782 
2783 	/* Allocate the KEYSOCK_OUT. */
2784 	keysock_out_mp = sadb_keysock_out(serial);
2785 	if (keysock_out_mp == NULL) {
2786 		esp0dbg(("esp_register_out: couldn't allocate mblk.\n"));
2787 		return (B_FALSE);
2788 	}
2789 
2790 	if (is_system_labeled() && (cr != NULL)) {
2791 		sens_tsl = crgetlabel(cr);
2792 		if (sens_tsl != NULL) {
2793 			sens_len = sadb_sens_len_from_label(sens_tsl);
2794 			allocsize += sens_len;
2795 		}
2796 	}
2797 
2798 	/*
2799 	 * Allocate the PF_KEY message that follows KEYSOCK_OUT.
2800 	 */
2801 
2802 	rw_enter(&ipss->ipsec_alg_lock, RW_READER);
2803 	/*
2804 	 * Fill SADB_REGISTER message's algorithm descriptors.  Hold
2805 	 * down the lock while filling it.
2806 	 *
2807 	 * Return only valid algorithms, so the number of algorithms
2808 	 * to send up may be less than the number of algorithm entries
2809 	 * in the table.
2810 	 */
2811 	authalgs = ipss->ipsec_alglists[IPSEC_ALG_AUTH];
2812 	for (num_aalgs = 0, i = 0; i < IPSEC_MAX_ALGS; i++)
2813 		if (authalgs[i] != NULL && ALG_VALID(authalgs[i]))
2814 			num_aalgs++;
2815 
2816 	if (num_aalgs != 0) {
2817 		allocsize += (num_aalgs * sizeof (*saalg));
2818 		allocsize += sizeof (*sasupp_auth);
2819 	}
2820 	encralgs = ipss->ipsec_alglists[IPSEC_ALG_ENCR];
2821 	for (num_ealgs = 0, i = 0; i < IPSEC_MAX_ALGS; i++)
2822 		if (encralgs[i] != NULL && ALG_VALID(encralgs[i]))
2823 			num_ealgs++;
2824 
2825 	if (num_ealgs != 0) {
2826 		allocsize += (num_ealgs * sizeof (*saalg));
2827 		allocsize += sizeof (*sasupp_encr);
2828 	}
2829 	keysock_out_mp->b_cont = allocb(allocsize, BPRI_HI);
2830 	if (keysock_out_mp->b_cont == NULL) {
2831 		rw_exit(&ipss->ipsec_alg_lock);
2832 		freemsg(keysock_out_mp);
2833 		return (B_FALSE);
2834 	}
2835 	pfkey_msg_mp = keysock_out_mp->b_cont;
2836 	pfkey_msg_mp->b_wptr += allocsize;
2837 
2838 	nextext = (sadb_ext_t *)(pfkey_msg_mp->b_rptr + sizeof (*samsg));
2839 
2840 	if (num_aalgs != 0) {
2841 		sasupp_auth = (sadb_supported_t *)nextext;
2842 		saalg = (sadb_alg_t *)(sasupp_auth + 1);
2843 
2844 		ASSERT(((ulong_t)saalg & 0x7) == 0);
2845 
2846 		numalgs_snap = 0;
2847 		for (i = 0;
2848 		    ((i < IPSEC_MAX_ALGS) && (numalgs_snap < num_aalgs));
2849 		    i++) {
2850 			if (authalgs[i] == NULL || !ALG_VALID(authalgs[i]))
2851 				continue;
2852 
2853 			saalg->sadb_alg_id = authalgs[i]->alg_id;
2854 			saalg->sadb_alg_ivlen = 0;
2855 			saalg->sadb_alg_minbits	= authalgs[i]->alg_ef_minbits;
2856 			saalg->sadb_alg_maxbits	= authalgs[i]->alg_ef_maxbits;
2857 			saalg->sadb_x_alg_increment =
2858 			    authalgs[i]->alg_increment;
2859 			saalg->sadb_x_alg_saltbits = SADB_8TO1(
2860 			    authalgs[i]->alg_saltlen);
2861 			numalgs_snap++;
2862 			saalg++;
2863 		}
2864 		ASSERT(numalgs_snap == num_aalgs);
2865 #ifdef DEBUG
2866 		/*
2867 		 * Reality check to make sure I snagged all of the
2868 		 * algorithms.
2869 		 */
2870 		for (; i < IPSEC_MAX_ALGS; i++) {
2871 			if (authalgs[i] != NULL && ALG_VALID(authalgs[i])) {
2872 				cmn_err(CE_PANIC, "esp_register_out()! "
2873 				    "Missed aalg #%d.\n", i);
2874 			}
2875 		}
2876 #endif /* DEBUG */
2877 		nextext = (sadb_ext_t *)saalg;
2878 	}
2879 
2880 	if (num_ealgs != 0) {
2881 		sasupp_encr = (sadb_supported_t *)nextext;
2882 		saalg = (sadb_alg_t *)(sasupp_encr + 1);
2883 
2884 		numalgs_snap = 0;
2885 		for (i = 0;
2886 		    ((i < IPSEC_MAX_ALGS) && (numalgs_snap < num_ealgs)); i++) {
2887 			if (encralgs[i] == NULL || !ALG_VALID(encralgs[i]))
2888 				continue;
2889 			saalg->sadb_alg_id = encralgs[i]->alg_id;
2890 			saalg->sadb_alg_ivlen = encralgs[i]->alg_ivlen;
2891 			saalg->sadb_alg_minbits	= encralgs[i]->alg_ef_minbits;
2892 			saalg->sadb_alg_maxbits	= encralgs[i]->alg_ef_maxbits;
2893 			/*
2894 			 * We could advertise the ICV length, except there
2895 			 * is not a value in sadb_x_algb to do this.
2896 			 * saalg->sadb_alg_maclen = encralgs[i]->alg_maclen;
2897 			 */
2898 			saalg->sadb_x_alg_increment =
2899 			    encralgs[i]->alg_increment;
2900 			saalg->sadb_x_alg_saltbits =
2901 			    SADB_8TO1(encralgs[i]->alg_saltlen);
2902 
2903 			numalgs_snap++;
2904 			saalg++;
2905 		}
2906 		ASSERT(numalgs_snap == num_ealgs);
2907 #ifdef DEBUG
2908 		/*
2909 		 * Reality check to make sure I snagged all of the
2910 		 * algorithms.
2911 		 */
2912 		for (; i < IPSEC_MAX_ALGS; i++) {
2913 			if (encralgs[i] != NULL && ALG_VALID(encralgs[i])) {
2914 				cmn_err(CE_PANIC, "esp_register_out()! "
2915 				    "Missed ealg #%d.\n", i);
2916 			}
2917 		}
2918 #endif /* DEBUG */
2919 		nextext = (sadb_ext_t *)saalg;
2920 	}
2921 
2922 	current_aalgs = num_aalgs;
2923 	current_ealgs = num_ealgs;
2924 
2925 	rw_exit(&ipss->ipsec_alg_lock);
2926 
2927 	if (sens_tsl != NULL) {
2928 		sens = (sadb_sens_t *)nextext;
2929 		sadb_sens_from_label(sens, SADB_EXT_SENSITIVITY,
2930 		    sens_tsl, sens_len);
2931 
2932 		nextext = (sadb_ext_t *)(((uint8_t *)sens) + sens_len);
2933 	}
2934 
2935 	/* Now fill the rest of the SADB_REGISTER message. */
2936 
2937 	samsg = (sadb_msg_t *)pfkey_msg_mp->b_rptr;
2938 	samsg->sadb_msg_version = PF_KEY_V2;
2939 	samsg->sadb_msg_type = SADB_REGISTER;
2940 	samsg->sadb_msg_errno = 0;
2941 	samsg->sadb_msg_satype = SADB_SATYPE_ESP;
2942 	samsg->sadb_msg_len = SADB_8TO64(allocsize);
2943 	samsg->sadb_msg_reserved = 0;
2944 	/*
2945 	 * Assume caller has sufficient sequence/pid number info.  If it's one
2946 	 * from me over a new alg., I could give two hoots about sequence.
2947 	 */
2948 	samsg->sadb_msg_seq = sequence;
2949 	samsg->sadb_msg_pid = pid;
2950 
2951 	if (sasupp_auth != NULL) {
2952 		sasupp_auth->sadb_supported_len = SADB_8TO64(
2953 		    sizeof (*sasupp_auth) + sizeof (*saalg) * current_aalgs);
2954 		sasupp_auth->sadb_supported_exttype = SADB_EXT_SUPPORTED_AUTH;
2955 		sasupp_auth->sadb_supported_reserved = 0;
2956 	}
2957 
2958 	if (sasupp_encr != NULL) {
2959 		sasupp_encr->sadb_supported_len = SADB_8TO64(
2960 		    sizeof (*sasupp_encr) + sizeof (*saalg) * current_ealgs);
2961 		sasupp_encr->sadb_supported_exttype =
2962 		    SADB_EXT_SUPPORTED_ENCRYPT;
2963 		sasupp_encr->sadb_supported_reserved = 0;
2964 	}
2965 
2966 	if (espstack->esp_pfkey_q != NULL)
2967 		putnext(espstack->esp_pfkey_q, keysock_out_mp);
2968 	else {
2969 		freemsg(keysock_out_mp);
2970 		return (B_FALSE);
2971 	}
2972 
2973 	return (B_TRUE);
2974 }
2975 
2976 /*
2977  * Invoked when the algorithm table changes. Causes SADB_REGISTER
2978  * messages continaining the current list of algorithms to be
2979  * sent up to the ESP listeners.
2980  */
2981 void
2982 ipsecesp_algs_changed(netstack_t *ns)
2983 {
2984 	ipsecesp_stack_t	*espstack = ns->netstack_ipsecesp;
2985 
2986 	/*
2987 	 * Time to send a PF_KEY SADB_REGISTER message to ESP listeners
2988 	 * everywhere.  (The function itself checks for NULL esp_pfkey_q.)
2989 	 */
2990 	(void) esp_register_out(0, 0, 0, espstack, NULL);
2991 }
2992 
2993 /*
2994  * Stub function that taskq_dispatch() invokes to take the mblk (in arg)
2995  * and send() it into ESP and IP again.
2996  */
2997 static void
2998 inbound_task(void *arg)
2999 {
3000 	mblk_t		*mp = (mblk_t *)arg;
3001 	mblk_t		*async_mp;
3002 	ip_recv_attr_t	iras;
3003 
3004 	async_mp = mp;
3005 	mp = async_mp->b_cont;
3006 	async_mp->b_cont = NULL;
3007 	if (!ip_recv_attr_from_mblk(async_mp, &iras)) {
3008 		/* The ill or ip_stack_t disappeared on us */
3009 		ip_drop_input("ip_recv_attr_from_mblk", mp, NULL);
3010 		freemsg(mp);
3011 		goto done;
3012 	}
3013 
3014 	esp_inbound_restart(mp, &iras);
3015 done:
3016 	ira_cleanup(&iras, B_TRUE);
3017 }
3018 
3019 /*
3020  * Restart ESP after the SA has been added.
3021  */
3022 static void
3023 esp_inbound_restart(mblk_t *mp, ip_recv_attr_t *ira)
3024 {
3025 	esph_t		*esph;
3026 	netstack_t	*ns = ira->ira_ill->ill_ipst->ips_netstack;
3027 	ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3028 
3029 	esp2dbg(espstack, ("in ESP inbound_task"));
3030 	ASSERT(espstack != NULL);
3031 
3032 	mp = ipsec_inbound_esp_sa(mp, ira, &esph);
3033 	if (mp == NULL)
3034 		return;
3035 
3036 	ASSERT(esph != NULL);
3037 	ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
3038 	ASSERT(ira->ira_ipsec_esp_sa != NULL);
3039 
3040 	mp = ira->ira_ipsec_esp_sa->ipsa_input_func(mp, esph, ira);
3041 	if (mp == NULL) {
3042 		/*
3043 		 * Either it failed or is pending. In the former case
3044 		 * ipIfStatsInDiscards was increased.
3045 		 */
3046 		return;
3047 	}
3048 
3049 	ip_input_post_ipsec(mp, ira);
3050 }
3051 
3052 /*
3053  * Now that weak-key passed, actually ADD the security association, and
3054  * send back a reply ADD message.
3055  */
3056 static int
3057 esp_add_sa_finish(mblk_t *mp, sadb_msg_t *samsg, keysock_in_t *ksi,
3058     int *diagnostic, ipsecesp_stack_t *espstack)
3059 {
3060 	isaf_t *primary = NULL, *secondary;
3061 	boolean_t clone = B_FALSE, is_inbound = B_FALSE;
3062 	ipsa_t *larval = NULL;
3063 	ipsacq_t *acqrec;
3064 	iacqf_t *acq_bucket;
3065 	mblk_t *acq_msgs = NULL;
3066 	int rc;
3067 	mblk_t *lpkt;
3068 	int error;
3069 	ipsa_query_t sq;
3070 	ipsec_stack_t	*ipss = espstack->ipsecesp_netstack->netstack_ipsec;
3071 
3072 	/*
3073 	 * Locate the appropriate table(s).
3074 	 */
3075 	sq.spp = &espstack->esp_sadb;	/* XXX */
3076 	error = sadb_form_query(ksi, IPSA_Q_SA|IPSA_Q_DST,
3077 	    IPSA_Q_SA|IPSA_Q_DST|IPSA_Q_INBOUND|IPSA_Q_OUTBOUND,
3078 	    &sq, diagnostic);
3079 	if (error)
3080 		return (error);
3081 
3082 	/*
3083 	 * Use the direction flags provided by the KMD to determine
3084 	 * if the inbound or outbound table should be the primary
3085 	 * for this SA. If these flags were absent then make this
3086 	 * decision based on the addresses.
3087 	 */
3088 	if (sq.assoc->sadb_sa_flags & IPSA_F_INBOUND) {
3089 		primary = sq.inbound;
3090 		secondary = sq.outbound;
3091 		is_inbound = B_TRUE;
3092 		if (sq.assoc->sadb_sa_flags & IPSA_F_OUTBOUND)
3093 			clone = B_TRUE;
3094 	} else if (sq.assoc->sadb_sa_flags & IPSA_F_OUTBOUND) {
3095 		primary = sq.outbound;
3096 		secondary = sq.inbound;
3097 	}
3098 
3099 	if (primary == NULL) {
3100 		/*
3101 		 * The KMD did not set a direction flag, determine which
3102 		 * table to insert the SA into based on addresses.
3103 		 */
3104 		switch (ksi->ks_in_dsttype) {
3105 		case KS_IN_ADDR_MBCAST:
3106 			clone = B_TRUE;	/* All mcast SAs can be bidirectional */
3107 			sq.assoc->sadb_sa_flags |= IPSA_F_OUTBOUND;
3108 			/* FALLTHRU */
3109 		/*
3110 		 * If the source address is either one of mine, or unspecified
3111 		 * (which is best summed up by saying "not 'not mine'"),
3112 		 * then the association is potentially bi-directional,
3113 		 * in that it can be used for inbound traffic and outbound
3114 		 * traffic.  The best example of such an SA is a multicast
3115 		 * SA (which allows me to receive the outbound traffic).
3116 		 */
3117 		case KS_IN_ADDR_ME:
3118 			sq.assoc->sadb_sa_flags |= IPSA_F_INBOUND;
3119 			primary = sq.inbound;
3120 			secondary = sq.outbound;
3121 			if (ksi->ks_in_srctype != KS_IN_ADDR_NOTME)
3122 				clone = B_TRUE;
3123 			is_inbound = B_TRUE;
3124 			break;
3125 		/*
3126 		 * If the source address literally not mine (either
3127 		 * unspecified or not mine), then this SA may have an
3128 		 * address that WILL be mine after some configuration.
3129 		 * We pay the price for this by making it a bi-directional
3130 		 * SA.
3131 		 */
3132 		case KS_IN_ADDR_NOTME:
3133 			sq.assoc->sadb_sa_flags |= IPSA_F_OUTBOUND;
3134 			primary = sq.outbound;
3135 			secondary = sq.inbound;
3136 			if (ksi->ks_in_srctype != KS_IN_ADDR_ME) {
3137 				sq.assoc->sadb_sa_flags |= IPSA_F_INBOUND;
3138 				clone = B_TRUE;
3139 			}
3140 			break;
3141 		default:
3142 			*diagnostic = SADB_X_DIAGNOSTIC_BAD_DST;
3143 			return (EINVAL);
3144 		}
3145 	}
3146 
3147 	/*
3148 	 * Find a ACQUIRE list entry if possible.  If we've added an SA that
3149 	 * suits the needs of an ACQUIRE list entry, we can eliminate the
3150 	 * ACQUIRE list entry and transmit the enqueued packets.  Use the
3151 	 * high-bit of the sequence number to queue it.  Key off destination
3152 	 * addr, and change acqrec's state.
3153 	 */
3154 
3155 	if (samsg->sadb_msg_seq & IACQF_LOWEST_SEQ) {
3156 		acq_bucket = &(sq.sp->sdb_acq[sq.outhash]);
3157 		mutex_enter(&acq_bucket->iacqf_lock);
3158 		for (acqrec = acq_bucket->iacqf_ipsacq; acqrec != NULL;
3159 		    acqrec = acqrec->ipsacq_next) {
3160 			mutex_enter(&acqrec->ipsacq_lock);
3161 			/*
3162 			 * Q:  I only check sequence.  Should I check dst?
3163 			 * A: Yes, check dest because those are the packets
3164 			 *    that are queued up.
3165 			 */
3166 			if (acqrec->ipsacq_seq == samsg->sadb_msg_seq &&
3167 			    IPSA_ARE_ADDR_EQUAL(sq.dstaddr,
3168 			    acqrec->ipsacq_dstaddr, acqrec->ipsacq_addrfam))
3169 				break;
3170 			mutex_exit(&acqrec->ipsacq_lock);
3171 		}
3172 		if (acqrec != NULL) {
3173 			/*
3174 			 * AHA!  I found an ACQUIRE record for this SA.
3175 			 * Grab the msg list, and free the acquire record.
3176 			 * I already am holding the lock for this record,
3177 			 * so all I have to do is free it.
3178 			 */
3179 			acq_msgs = acqrec->ipsacq_mp;
3180 			acqrec->ipsacq_mp = NULL;
3181 			mutex_exit(&acqrec->ipsacq_lock);
3182 			sadb_destroy_acquire(acqrec,
3183 			    espstack->ipsecesp_netstack);
3184 		}
3185 		mutex_exit(&acq_bucket->iacqf_lock);
3186 	}
3187 
3188 	/*
3189 	 * Find PF_KEY message, and see if I'm an update.  If so, find entry
3190 	 * in larval list (if there).
3191 	 */
3192 	if (samsg->sadb_msg_type == SADB_UPDATE) {
3193 		mutex_enter(&sq.inbound->isaf_lock);
3194 		larval = ipsec_getassocbyspi(sq.inbound, sq.assoc->sadb_sa_spi,
3195 		    ALL_ZEROES_PTR, sq.dstaddr, sq.dst->sin_family);
3196 		mutex_exit(&sq.inbound->isaf_lock);
3197 
3198 		if ((larval == NULL) ||
3199 		    (larval->ipsa_state != IPSA_STATE_LARVAL)) {
3200 			*diagnostic = SADB_X_DIAGNOSTIC_SA_NOTFOUND;
3201 			if (larval != NULL) {
3202 				IPSA_REFRELE(larval);
3203 			}
3204 			esp0dbg(("Larval update, but larval disappeared.\n"));
3205 			return (ESRCH);
3206 		} /* Else sadb_common_add unlinks it for me! */
3207 	}
3208 
3209 	if (larval != NULL) {
3210 		/*
3211 		 * Hold again, because sadb_common_add() consumes a reference,
3212 		 * and we don't want to clear_lpkt() without a reference.
3213 		 */
3214 		IPSA_REFHOLD(larval);
3215 	}
3216 
3217 	rc = sadb_common_add(espstack->esp_pfkey_q,
3218 	    mp, samsg, ksi, primary, secondary, larval, clone, is_inbound,
3219 	    diagnostic, espstack->ipsecesp_netstack, &espstack->esp_sadb);
3220 
3221 	if (larval != NULL) {
3222 		if (rc == 0) {
3223 			lpkt = sadb_clear_lpkt(larval);
3224 			if (lpkt != NULL) {
3225 				rc = taskq_dispatch(esp_taskq, inbound_task,
3226 				    lpkt, TQ_NOSLEEP) == TASKQID_INVALID;
3227 			}
3228 		}
3229 		IPSA_REFRELE(larval);
3230 	}
3231 
3232 	/*
3233 	 * How much more stack will I create with all of these
3234 	 * esp_outbound() calls?
3235 	 */
3236 
3237 	/* Handle the packets queued waiting for the SA */
3238 	while (acq_msgs != NULL) {
3239 		mblk_t		*asyncmp;
3240 		mblk_t		*data_mp;
3241 		ip_xmit_attr_t	ixas;
3242 		ill_t		*ill;
3243 
3244 		asyncmp = acq_msgs;
3245 		acq_msgs = acq_msgs->b_next;
3246 		asyncmp->b_next = NULL;
3247 
3248 		/*
3249 		 * Extract the ip_xmit_attr_t from the first mblk.
3250 		 * Verifies that the netstack and ill is still around; could
3251 		 * have vanished while iked was doing its work.
3252 		 * On succesful return we have a nce_t and the ill/ipst can't
3253 		 * disappear until we do the nce_refrele in ixa_cleanup.
3254 		 */
3255 		data_mp = asyncmp->b_cont;
3256 		asyncmp->b_cont = NULL;
3257 		if (!ip_xmit_attr_from_mblk(asyncmp, &ixas)) {
3258 			ESP_BUMP_STAT(espstack, out_discards);
3259 			ip_drop_packet(data_mp, B_FALSE, NULL,
3260 			    DROPPER(ipss, ipds_sadb_acquire_timeout),
3261 			    &espstack->esp_dropper);
3262 		} else if (rc != 0) {
3263 			ill = ixas.ixa_nce->nce_ill;
3264 			ESP_BUMP_STAT(espstack, out_discards);
3265 			ip_drop_packet(data_mp, B_FALSE, ill,
3266 			    DROPPER(ipss, ipds_sadb_acquire_timeout),
3267 			    &espstack->esp_dropper);
3268 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
3269 		} else {
3270 			esp_outbound_finish(data_mp, &ixas);
3271 		}
3272 		ixa_cleanup(&ixas);
3273 	}
3274 
3275 	return (rc);
3276 }
3277 
3278 /*
3279  * Process one of the queued messages (from ipsacq_mp) once the SA
3280  * has been added.
3281  */
3282 static void
3283 esp_outbound_finish(mblk_t *data_mp, ip_xmit_attr_t *ixa)
3284 {
3285 	netstack_t	*ns = ixa->ixa_ipst->ips_netstack;
3286 	ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3287 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
3288 	ill_t		*ill = ixa->ixa_nce->nce_ill;
3289 
3290 	if (!ipsec_outbound_sa(data_mp, ixa, IPPROTO_ESP)) {
3291 		ESP_BUMP_STAT(espstack, out_discards);
3292 		ip_drop_packet(data_mp, B_FALSE, ill,
3293 		    DROPPER(ipss, ipds_sadb_acquire_timeout),
3294 		    &espstack->esp_dropper);
3295 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
3296 		return;
3297 	}
3298 
3299 	data_mp = esp_outbound(data_mp, ixa);
3300 	if (data_mp == NULL)
3301 		return;
3302 
3303 	/* do AH processing if needed */
3304 	data_mp = esp_do_outbound_ah(data_mp, ixa);
3305 	if (data_mp == NULL)
3306 		return;
3307 
3308 	(void) ip_output_post_ipsec(data_mp, ixa);
3309 }
3310 
3311 /*
3312  * Add new ESP security association.  This may become a generic AH/ESP
3313  * routine eventually.
3314  */
3315 static int
3316 esp_add_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic, netstack_t *ns)
3317 {
3318 	sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
3319 	sadb_address_t *srcext =
3320 	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
3321 	sadb_address_t *dstext =
3322 	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
3323 	sadb_address_t *isrcext =
3324 	    (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_SRC];
3325 	sadb_address_t *idstext =
3326 	    (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_DST];
3327 	sadb_address_t *nttext_loc =
3328 	    (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_LOC];
3329 	sadb_address_t *nttext_rem =
3330 	    (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_REM];
3331 	sadb_key_t *akey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_AUTH];
3332 	sadb_key_t *ekey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_ENCRYPT];
3333 	struct sockaddr_in *src, *dst;
3334 	struct sockaddr_in *natt_loc, *natt_rem;
3335 	struct sockaddr_in6 *natt_loc6, *natt_rem6;
3336 	sadb_lifetime_t *soft =
3337 	    (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_SOFT];
3338 	sadb_lifetime_t *hard =
3339 	    (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_HARD];
3340 	sadb_lifetime_t *idle =
3341 	    (sadb_lifetime_t *)ksi->ks_in_extv[SADB_X_EXT_LIFETIME_IDLE];
3342 	ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3343 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
3344 
3345 
3346 
3347 	/* I need certain extensions present for an ADD message. */
3348 	if (srcext == NULL) {
3349 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_SRC;
3350 		return (EINVAL);
3351 	}
3352 	if (dstext == NULL) {
3353 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
3354 		return (EINVAL);
3355 	}
3356 	if (isrcext == NULL && idstext != NULL) {
3357 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_SRC;
3358 		return (EINVAL);
3359 	}
3360 	if (isrcext != NULL && idstext == NULL) {
3361 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_DST;
3362 		return (EINVAL);
3363 	}
3364 	if (assoc == NULL) {
3365 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
3366 		return (EINVAL);
3367 	}
3368 	if (ekey == NULL && assoc->sadb_sa_encrypt != SADB_EALG_NULL) {
3369 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_EKEY;
3370 		return (EINVAL);
3371 	}
3372 
3373 	src = (struct sockaddr_in *)(srcext + 1);
3374 	dst = (struct sockaddr_in *)(dstext + 1);
3375 	natt_loc = (struct sockaddr_in *)(nttext_loc + 1);
3376 	natt_loc6 = (struct sockaddr_in6 *)(nttext_loc + 1);
3377 	natt_rem = (struct sockaddr_in *)(nttext_rem + 1);
3378 	natt_rem6 = (struct sockaddr_in6 *)(nttext_rem + 1);
3379 
3380 	/* Sundry ADD-specific reality checks. */
3381 	/* XXX STATS :  Logging/stats here? */
3382 
3383 	if ((assoc->sadb_sa_state != SADB_SASTATE_MATURE) &&
3384 	    (assoc->sadb_sa_state != SADB_X_SASTATE_ACTIVE_ELSEWHERE)) {
3385 		*diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
3386 		return (EINVAL);
3387 	}
3388 	if (assoc->sadb_sa_encrypt == SADB_EALG_NONE) {
3389 		*diagnostic = SADB_X_DIAGNOSTIC_BAD_EALG;
3390 		return (EINVAL);
3391 	}
3392 
3393 #ifndef IPSEC_LATENCY_TEST
3394 	if (assoc->sadb_sa_encrypt == SADB_EALG_NULL &&
3395 	    assoc->sadb_sa_auth == SADB_AALG_NONE) {
3396 		*diagnostic = SADB_X_DIAGNOSTIC_BAD_AALG;
3397 		return (EINVAL);
3398 	}
3399 #endif
3400 
3401 	if (assoc->sadb_sa_flags & ~espstack->esp_sadb.s_addflags) {
3402 		*diagnostic = SADB_X_DIAGNOSTIC_BAD_SAFLAGS;
3403 		return (EINVAL);
3404 	}
3405 
3406 	if ((*diagnostic = sadb_hardsoftchk(hard, soft, idle)) != 0) {
3407 		return (EINVAL);
3408 	}
3409 	ASSERT(src->sin_family == dst->sin_family);
3410 
3411 	if (assoc->sadb_sa_flags & SADB_X_SAFLAGS_NATT_LOC) {
3412 		if (nttext_loc == NULL) {
3413 			*diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_LOC;
3414 			return (EINVAL);
3415 		}
3416 
3417 		if (natt_loc->sin_family == AF_INET6 &&
3418 		    !IN6_IS_ADDR_V4MAPPED(&natt_loc6->sin6_addr)) {
3419 			*diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_NATT_LOC;
3420 			return (EINVAL);
3421 		}
3422 	}
3423 
3424 	if (assoc->sadb_sa_flags & SADB_X_SAFLAGS_NATT_REM) {
3425 		if (nttext_rem == NULL) {
3426 			*diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_REM;
3427 			return (EINVAL);
3428 		}
3429 		if (natt_rem->sin_family == AF_INET6 &&
3430 		    !IN6_IS_ADDR_V4MAPPED(&natt_rem6->sin6_addr)) {
3431 			*diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_NATT_REM;
3432 			return (EINVAL);
3433 		}
3434 	}
3435 
3436 
3437 	/* Stuff I don't support, for now.  XXX Diagnostic? */
3438 	if (ksi->ks_in_extv[SADB_EXT_LIFETIME_CURRENT] != NULL)
3439 		return (EOPNOTSUPP);
3440 
3441 	if ((*diagnostic = sadb_labelchk(ksi)) != 0)
3442 		return (EINVAL);
3443 
3444 	/*
3445 	 * XXX Policy :  I'm not checking identities at this time,
3446 	 * but if I did, I'd do them here, before I sent
3447 	 * the weak key check up to the algorithm.
3448 	 */
3449 
3450 	rw_enter(&ipss->ipsec_alg_lock, RW_READER);
3451 
3452 	/*
3453 	 * First locate the authentication algorithm.
3454 	 */
3455 #ifdef IPSEC_LATENCY_TEST
3456 	if (akey != NULL && assoc->sadb_sa_auth != SADB_AALG_NONE) {
3457 #else
3458 	if (akey != NULL) {
3459 #endif
3460 		ipsec_alginfo_t *aalg;
3461 
3462 		aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
3463 		    [assoc->sadb_sa_auth];
3464 		if (aalg == NULL || !ALG_VALID(aalg)) {
3465 			rw_exit(&ipss->ipsec_alg_lock);
3466 			esp1dbg(espstack, ("Couldn't find auth alg #%d.\n",
3467 			    assoc->sadb_sa_auth));
3468 			*diagnostic = SADB_X_DIAGNOSTIC_BAD_AALG;
3469 			return (EINVAL);
3470 		}
3471 
3472 		/*
3473 		 * Sanity check key sizes.
3474 		 * Note: It's not possible to use SADB_AALG_NONE because
3475 		 * this auth_alg is not defined with ALG_FLAG_VALID. If this
3476 		 * ever changes, the same check for SADB_AALG_NONE and
3477 		 * a auth_key != NULL should be made here ( see below).
3478 		 */
3479 		if (!ipsec_valid_key_size(akey->sadb_key_bits, aalg)) {
3480 			rw_exit(&ipss->ipsec_alg_lock);
3481 			*diagnostic = SADB_X_DIAGNOSTIC_BAD_AKEYBITS;
3482 			return (EINVAL);
3483 		}
3484 		ASSERT(aalg->alg_mech_type != CRYPTO_MECHANISM_INVALID);
3485 
3486 		/* check key and fix parity if needed */
3487 		if (ipsec_check_key(aalg->alg_mech_type, akey, B_TRUE,
3488 		    diagnostic) != 0) {
3489 			rw_exit(&ipss->ipsec_alg_lock);
3490 			return (EINVAL);
3491 		}
3492 	}
3493 
3494 	/*
3495 	 * Then locate the encryption algorithm.
3496 	 */
3497 	if (ekey != NULL) {
3498 		uint_t keybits;
3499 		ipsec_alginfo_t *ealg;
3500 
3501 		ealg = ipss->ipsec_alglists[IPSEC_ALG_ENCR]
3502 		    [assoc->sadb_sa_encrypt];
3503 		if (ealg == NULL || !ALG_VALID(ealg)) {
3504 			rw_exit(&ipss->ipsec_alg_lock);
3505 			esp1dbg(espstack, ("Couldn't find encr alg #%d.\n",
3506 			    assoc->sadb_sa_encrypt));
3507 			*diagnostic = SADB_X_DIAGNOSTIC_BAD_EALG;
3508 			return (EINVAL);
3509 		}
3510 
3511 		/*
3512 		 * Sanity check key sizes. If the encryption algorithm is
3513 		 * SADB_EALG_NULL but the encryption key is NOT
3514 		 * NULL then complain.
3515 		 *
3516 		 * The keying material includes salt bits if required by
3517 		 * algorithm and optionally the Initial IV, check the
3518 		 * length of whats left.
3519 		 */
3520 		keybits = ekey->sadb_key_bits;
3521 		keybits -= ekey->sadb_key_reserved;
3522 		keybits -= SADB_8TO1(ealg->alg_saltlen);
3523 		if ((assoc->sadb_sa_encrypt == SADB_EALG_NULL) ||
3524 		    (!ipsec_valid_key_size(keybits, ealg))) {
3525 			rw_exit(&ipss->ipsec_alg_lock);
3526 			*diagnostic = SADB_X_DIAGNOSTIC_BAD_EKEYBITS;
3527 			return (EINVAL);
3528 		}
3529 		ASSERT(ealg->alg_mech_type != CRYPTO_MECHANISM_INVALID);
3530 
3531 		/* check key */
3532 		if (ipsec_check_key(ealg->alg_mech_type, ekey, B_FALSE,
3533 		    diagnostic) != 0) {
3534 			rw_exit(&ipss->ipsec_alg_lock);
3535 			return (EINVAL);
3536 		}
3537 	}
3538 	rw_exit(&ipss->ipsec_alg_lock);
3539 
3540 	return (esp_add_sa_finish(mp, (sadb_msg_t *)mp->b_cont->b_rptr, ksi,
3541 	    diagnostic, espstack));
3542 }
3543 
3544 /*
3545  * Update a security association.  Updates come in two varieties.  The first
3546  * is an update of lifetimes on a non-larval SA.  The second is an update of
3547  * a larval SA, which ends up looking a lot more like an add.
3548  */
3549 static int
3550 esp_update_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic,
3551     ipsecesp_stack_t *espstack, uint8_t sadb_msg_type)
3552 {
3553 	sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
3554 	mblk_t    *buf_pkt;
3555 	int rcode;
3556 
3557 	sadb_address_t *dstext =
3558 	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
3559 
3560 	if (dstext == NULL) {
3561 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
3562 		return (EINVAL);
3563 	}
3564 
3565 	rcode = sadb_update_sa(mp, ksi, &buf_pkt, &espstack->esp_sadb,
3566 	    diagnostic, espstack->esp_pfkey_q, esp_add_sa,
3567 	    espstack->ipsecesp_netstack, sadb_msg_type);
3568 
3569 	if ((assoc->sadb_sa_state != SADB_X_SASTATE_ACTIVE) ||
3570 	    (rcode != 0)) {
3571 		return (rcode);
3572 	}
3573 
3574 	HANDLE_BUF_PKT(esp_taskq, espstack->ipsecesp_netstack->netstack_ipsec,
3575 	    espstack->esp_dropper, buf_pkt);
3576 
3577 	return (rcode);
3578 }
3579 
3580 /* XXX refactor me */
3581 /*
3582  * Delete a security association.  This is REALLY likely to be code common to
3583  * both AH and ESP.  Find the association, then unlink it.
3584  */
3585 static int
3586 esp_del_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic,
3587     ipsecesp_stack_t *espstack, uint8_t sadb_msg_type)
3588 {
3589 	sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
3590 	sadb_address_t *dstext =
3591 	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
3592 	sadb_address_t *srcext =
3593 	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
3594 	struct sockaddr_in *sin;
3595 
3596 	if (assoc == NULL) {
3597 		if (dstext != NULL) {
3598 			sin = (struct sockaddr_in *)(dstext + 1);
3599 		} else if (srcext != NULL) {
3600 			sin = (struct sockaddr_in *)(srcext + 1);
3601 		} else {
3602 			*diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
3603 			return (EINVAL);
3604 		}
3605 		return (sadb_purge_sa(mp, ksi,
3606 		    (sin->sin_family == AF_INET6) ? &espstack->esp_sadb.s_v6 :
3607 		    &espstack->esp_sadb.s_v4, diagnostic,
3608 		    espstack->esp_pfkey_q));
3609 	}
3610 
3611 	return (sadb_delget_sa(mp, ksi, &espstack->esp_sadb, diagnostic,
3612 	    espstack->esp_pfkey_q, sadb_msg_type));
3613 }
3614 
3615 /* XXX refactor me */
3616 /*
3617  * Convert the entire contents of all of ESP's SA tables into PF_KEY SADB_DUMP
3618  * messages.
3619  */
3620 static void
3621 esp_dump(mblk_t *mp, keysock_in_t *ksi, ipsecesp_stack_t *espstack)
3622 {
3623 	int error;
3624 	sadb_msg_t *samsg;
3625 
3626 	/*
3627 	 * Dump each fanout, bailing if error is non-zero.
3628 	 */
3629 
3630 	error = sadb_dump(espstack->esp_pfkey_q, mp, ksi,
3631 	    &espstack->esp_sadb.s_v4);
3632 	if (error != 0)
3633 		goto bail;
3634 
3635 	error = sadb_dump(espstack->esp_pfkey_q, mp, ksi,
3636 	    &espstack->esp_sadb.s_v6);
3637 bail:
3638 	ASSERT(mp->b_cont != NULL);
3639 	samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
3640 	samsg->sadb_msg_errno = (uint8_t)error;
3641 	sadb_pfkey_echo(espstack->esp_pfkey_q, mp,
3642 	    (sadb_msg_t *)mp->b_cont->b_rptr, ksi, NULL);
3643 }
3644 
3645 /*
3646  * First-cut reality check for an inbound PF_KEY message.
3647  */
3648 static boolean_t
3649 esp_pfkey_reality_failures(mblk_t *mp, keysock_in_t *ksi,
3650     ipsecesp_stack_t *espstack)
3651 {
3652 	int diagnostic;
3653 
3654 	if (ksi->ks_in_extv[SADB_EXT_PROPOSAL] != NULL) {
3655 		diagnostic = SADB_X_DIAGNOSTIC_PROP_PRESENT;
3656 		goto badmsg;
3657 	}
3658 	if (ksi->ks_in_extv[SADB_EXT_SUPPORTED_AUTH] != NULL ||
3659 	    ksi->ks_in_extv[SADB_EXT_SUPPORTED_ENCRYPT] != NULL) {
3660 		diagnostic = SADB_X_DIAGNOSTIC_SUPP_PRESENT;
3661 		goto badmsg;
3662 	}
3663 	return (B_FALSE);	/* False ==> no failures */
3664 
3665 badmsg:
3666 	sadb_pfkey_error(espstack->esp_pfkey_q, mp, EINVAL, diagnostic,
3667 	    ksi->ks_in_serial);
3668 	return (B_TRUE);	/* True ==> failures */
3669 }
3670 
3671 /*
3672  * ESP parsing of PF_KEY messages.  Keysock did most of the really silly
3673  * error cases.  What I receive is a fully-formed, syntactically legal
3674  * PF_KEY message.  I then need to check semantics...
3675  *
3676  * This code may become common to AH and ESP.  Stay tuned.
3677  *
3678  * I also make the assumption that db_ref's are cool.  If this assumption
3679  * is wrong, this means that someone other than keysock or me has been
3680  * mucking with PF_KEY messages.
3681  */
3682 static void
3683 esp_parse_pfkey(mblk_t *mp, ipsecesp_stack_t *espstack)
3684 {
3685 	mblk_t *msg = mp->b_cont;
3686 	sadb_msg_t *samsg;
3687 	keysock_in_t *ksi;
3688 	int error;
3689 	int diagnostic = SADB_X_DIAGNOSTIC_NONE;
3690 
3691 	ASSERT(msg != NULL);
3692 
3693 	samsg = (sadb_msg_t *)msg->b_rptr;
3694 	ksi = (keysock_in_t *)mp->b_rptr;
3695 
3696 	/*
3697 	 * If applicable, convert unspecified AF_INET6 to unspecified
3698 	 * AF_INET.  And do other address reality checks.
3699 	 */
3700 	if (!sadb_addrfix(ksi, espstack->esp_pfkey_q, mp,
3701 	    espstack->ipsecesp_netstack) ||
3702 	    esp_pfkey_reality_failures(mp, ksi, espstack)) {
3703 		return;
3704 	}
3705 
3706 	switch (samsg->sadb_msg_type) {
3707 	case SADB_ADD:
3708 		error = esp_add_sa(mp, ksi, &diagnostic,
3709 		    espstack->ipsecesp_netstack);
3710 		if (error != 0) {
3711 			sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
3712 			    diagnostic, ksi->ks_in_serial);
3713 		}
3714 		/* else esp_add_sa() took care of things. */
3715 		break;
3716 	case SADB_DELETE:
3717 	case SADB_X_DELPAIR:
3718 	case SADB_X_DELPAIR_STATE:
3719 		error = esp_del_sa(mp, ksi, &diagnostic, espstack,
3720 		    samsg->sadb_msg_type);
3721 		if (error != 0) {
3722 			sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
3723 			    diagnostic, ksi->ks_in_serial);
3724 		}
3725 		/* Else esp_del_sa() took care of things. */
3726 		break;
3727 	case SADB_GET:
3728 		error = sadb_delget_sa(mp, ksi, &espstack->esp_sadb,
3729 		    &diagnostic, espstack->esp_pfkey_q, samsg->sadb_msg_type);
3730 		if (error != 0) {
3731 			sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
3732 			    diagnostic, ksi->ks_in_serial);
3733 		}
3734 		/* Else sadb_get_sa() took care of things. */
3735 		break;
3736 	case SADB_FLUSH:
3737 		sadbp_flush(&espstack->esp_sadb, espstack->ipsecesp_netstack);
3738 		sadb_pfkey_echo(espstack->esp_pfkey_q, mp, samsg, ksi, NULL);
3739 		break;
3740 	case SADB_REGISTER:
3741 		/*
3742 		 * Hmmm, let's do it!  Check for extensions (there should
3743 		 * be none), extract the fields, call esp_register_out(),
3744 		 * then either free or report an error.
3745 		 *
3746 		 * Keysock takes care of the PF_KEY bookkeeping for this.
3747 		 */
3748 		if (esp_register_out(samsg->sadb_msg_seq, samsg->sadb_msg_pid,
3749 		    ksi->ks_in_serial, espstack, msg_getcred(mp, NULL))) {
3750 			freemsg(mp);
3751 		} else {
3752 			/*
3753 			 * Only way this path hits is if there is a memory
3754 			 * failure.  It will not return B_FALSE because of
3755 			 * lack of esp_pfkey_q if I am in wput().
3756 			 */
3757 			sadb_pfkey_error(espstack->esp_pfkey_q, mp, ENOMEM,
3758 			    diagnostic, ksi->ks_in_serial);
3759 		}
3760 		break;
3761 	case SADB_UPDATE:
3762 	case SADB_X_UPDATEPAIR:
3763 		/*
3764 		 * Find a larval, if not there, find a full one and get
3765 		 * strict.
3766 		 */
3767 		error = esp_update_sa(mp, ksi, &diagnostic, espstack,
3768 		    samsg->sadb_msg_type);
3769 		if (error != 0) {
3770 			sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
3771 			    diagnostic, ksi->ks_in_serial);
3772 		}
3773 		/* else esp_update_sa() took care of things. */
3774 		break;
3775 	case SADB_GETSPI:
3776 		/*
3777 		 * Reserve a new larval entry.
3778 		 */
3779 		esp_getspi(mp, ksi, espstack);
3780 		break;
3781 	case SADB_ACQUIRE:
3782 		/*
3783 		 * Find larval and/or ACQUIRE record and kill it (them), I'm
3784 		 * most likely an error.  Inbound ACQUIRE messages should only
3785 		 * have the base header.
3786 		 */
3787 		sadb_in_acquire(samsg, &espstack->esp_sadb,
3788 		    espstack->esp_pfkey_q, espstack->ipsecesp_netstack);
3789 		freemsg(mp);
3790 		break;
3791 	case SADB_DUMP:
3792 		/*
3793 		 * Dump all entries.
3794 		 */
3795 		esp_dump(mp, ksi, espstack);
3796 		/* esp_dump will take care of the return message, etc. */
3797 		break;
3798 	case SADB_EXPIRE:
3799 		/* Should never reach me. */
3800 		sadb_pfkey_error(espstack->esp_pfkey_q, mp, EOPNOTSUPP,
3801 		    diagnostic, ksi->ks_in_serial);
3802 		break;
3803 	default:
3804 		sadb_pfkey_error(espstack->esp_pfkey_q, mp, EINVAL,
3805 		    SADB_X_DIAGNOSTIC_UNKNOWN_MSG, ksi->ks_in_serial);
3806 		break;
3807 	}
3808 }
3809 
3810 /*
3811  * Handle case where PF_KEY says it can't find a keysock for one of my
3812  * ACQUIRE messages.
3813  */
3814 static void
3815 esp_keysock_no_socket(mblk_t *mp, ipsecesp_stack_t *espstack)
3816 {
3817 	sadb_msg_t *samsg;
3818 	keysock_out_err_t *kse = (keysock_out_err_t *)mp->b_rptr;
3819 
3820 	if (mp->b_cont == NULL) {
3821 		freemsg(mp);
3822 		return;
3823 	}
3824 	samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
3825 
3826 	/*
3827 	 * If keysock can't find any registered, delete the acquire record
3828 	 * immediately, and handle errors.
3829 	 */
3830 	if (samsg->sadb_msg_type == SADB_ACQUIRE) {
3831 		samsg->sadb_msg_errno = kse->ks_err_errno;
3832 		samsg->sadb_msg_len = SADB_8TO64(sizeof (*samsg));
3833 		/*
3834 		 * Use the write-side of the esp_pfkey_q
3835 		 */
3836 		sadb_in_acquire(samsg, &espstack->esp_sadb,
3837 		    WR(espstack->esp_pfkey_q), espstack->ipsecesp_netstack);
3838 	}
3839 
3840 	freemsg(mp);
3841 }
3842 
3843 /*
3844  * ESP module read put routine.
3845  */
3846 static int
3847 ipsecesp_rput(queue_t *q, mblk_t *mp)
3848 {
3849 	putnext(q, mp);
3850 	return (0);
3851 }
3852 
3853 /*
3854  * ESP module write put routine.
3855  */
3856 static int
3857 ipsecesp_wput(queue_t *q, mblk_t *mp)
3858 {
3859 	ipsec_info_t *ii;
3860 	struct iocblk *iocp;
3861 	ipsecesp_stack_t	*espstack = (ipsecesp_stack_t *)q->q_ptr;
3862 
3863 	esp3dbg(espstack, ("In esp_wput().\n"));
3864 
3865 	/* NOTE: Each case must take care of freeing or passing mp. */
3866 	switch (mp->b_datap->db_type) {
3867 	case M_CTL:
3868 		if ((mp->b_wptr - mp->b_rptr) < sizeof (ipsec_info_t)) {
3869 			/* Not big enough message. */
3870 			freemsg(mp);
3871 			break;
3872 		}
3873 		ii = (ipsec_info_t *)mp->b_rptr;
3874 
3875 		switch (ii->ipsec_info_type) {
3876 		case KEYSOCK_OUT_ERR:
3877 			esp1dbg(espstack, ("Got KEYSOCK_OUT_ERR message.\n"));
3878 			esp_keysock_no_socket(mp, espstack);
3879 			break;
3880 		case KEYSOCK_IN:
3881 			ESP_BUMP_STAT(espstack, keysock_in);
3882 			esp3dbg(espstack, ("Got KEYSOCK_IN message.\n"));
3883 
3884 			/* Parse the message. */
3885 			esp_parse_pfkey(mp, espstack);
3886 			break;
3887 		case KEYSOCK_HELLO:
3888 			sadb_keysock_hello(&espstack->esp_pfkey_q, q, mp,
3889 			    esp_ager, (void *)espstack, &espstack->esp_event,
3890 			    SADB_SATYPE_ESP);
3891 			break;
3892 		default:
3893 			esp2dbg(espstack, ("Got M_CTL from above of 0x%x.\n",
3894 			    ii->ipsec_info_type));
3895 			freemsg(mp);
3896 			break;
3897 		}
3898 		break;
3899 	case M_IOCTL:
3900 		iocp = (struct iocblk *)mp->b_rptr;
3901 		switch (iocp->ioc_cmd) {
3902 		case ND_SET:
3903 		case ND_GET:
3904 			if (nd_getset(q, espstack->ipsecesp_g_nd, mp)) {
3905 				qreply(q, mp);
3906 				return (0);
3907 			} else {
3908 				iocp->ioc_error = ENOENT;
3909 			}
3910 			/* FALLTHRU */
3911 		default:
3912 			/* We really don't support any other ioctls, do we? */
3913 
3914 			/* Return EINVAL */
3915 			if (iocp->ioc_error != ENOENT)
3916 				iocp->ioc_error = EINVAL;
3917 			iocp->ioc_count = 0;
3918 			mp->b_datap->db_type = M_IOCACK;
3919 			qreply(q, mp);
3920 			return (0);
3921 		}
3922 	default:
3923 		esp3dbg(espstack,
3924 		    ("Got default message, type %d, passing to IP.\n",
3925 		    mp->b_datap->db_type));
3926 		putnext(q, mp);
3927 	}
3928 	return (0);
3929 }
3930 
3931 /*
3932  * Wrapper to allow IP to trigger an ESP association failure message
3933  * during inbound SA selection.
3934  */
3935 void
3936 ipsecesp_in_assocfailure(mblk_t *mp, char level, ushort_t sl, char *fmt,
3937     uint32_t spi, void *addr, int af, ip_recv_attr_t *ira)
3938 {
3939 	netstack_t	*ns = ira->ira_ill->ill_ipst->ips_netstack;
3940 	ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3941 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
3942 
3943 	if (espstack->ipsecesp_log_unknown_spi) {
3944 		ipsec_assocfailure(info.mi_idnum, 0, level, sl, fmt, spi,
3945 		    addr, af, espstack->ipsecesp_netstack);
3946 	}
3947 
3948 	ip_drop_packet(mp, B_TRUE, ira->ira_ill,
3949 	    DROPPER(ipss, ipds_esp_no_sa),
3950 	    &espstack->esp_dropper);
3951 }
3952 
3953 /*
3954  * Initialize the ESP input and output processing functions.
3955  */
3956 void
3957 ipsecesp_init_funcs(ipsa_t *sa)
3958 {
3959 	if (sa->ipsa_output_func == NULL)
3960 		sa->ipsa_output_func = esp_outbound;
3961 	if (sa->ipsa_input_func == NULL)
3962 		sa->ipsa_input_func = esp_inbound;
3963 }
3964