xref: /titanic_51/usr/src/uts/common/inet/ip/ipsecah.c (revision 82629e3015252bf18319ba3815c773df23e21436)
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  */
25 
26 #include <sys/types.h>
27 #include <sys/stream.h>
28 #include <sys/stropts.h>
29 #include <sys/errno.h>
30 #include <sys/strlog.h>
31 #include <sys/tihdr.h>
32 #include <sys/socket.h>
33 #include <sys/ddi.h>
34 #include <sys/sunddi.h>
35 #include <sys/mkdev.h>
36 #include <sys/kmem.h>
37 #include <sys/zone.h>
38 #include <sys/sysmacros.h>
39 #include <sys/cmn_err.h>
40 #include <sys/vtrace.h>
41 #include <sys/debug.h>
42 #include <sys/atomic.h>
43 #include <sys/strsun.h>
44 #include <sys/random.h>
45 #include <netinet/in.h>
46 #include <net/if.h>
47 #include <netinet/ip6.h>
48 #include <netinet/icmp6.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/ip.h>
55 #include <inet/ip6.h>
56 #include <inet/nd.h>
57 #include <inet/ip_if.h>
58 #include <inet/ip_ndp.h>
59 #include <inet/ipsec_info.h>
60 #include <inet/ipsec_impl.h>
61 #include <inet/sadb.h>
62 #include <inet/ipsecah.h>
63 #include <inet/ipsec_impl.h>
64 #include <inet/ipdrop.h>
65 #include <sys/taskq.h>
66 #include <sys/policy.h>
67 #include <sys/strsun.h>
68 
69 #include <sys/crypto/common.h>
70 #include <sys/crypto/api.h>
71 #include <sys/kstat.h>
72 #include <sys/strsubr.h>
73 
74 #include <sys/tsol/tnet.h>
75 
76 /*
77  * Table of ND variables supported by ipsecah. These are loaded into
78  * ipsecah_g_nd in ipsecah_init_nd.
79  * All of these are alterable, within the min/max values given, at run time.
80  */
81 static	ipsecahparam_t	lcl_param_arr[] = {
82 	/* min	max			value	name */
83 	{ 0,	3,			0,	"ipsecah_debug"},
84 	{ 125,	32000, SADB_AGE_INTERVAL_DEFAULT,	"ipsecah_age_interval"},
85 	{ 1,	10,			1,	"ipsecah_reap_delay"},
86 	{ 1,	SADB_MAX_REPLAY,	64,	"ipsecah_replay_size"},
87 	{ 1,	300,			15,	"ipsecah_acquire_timeout"},
88 	{ 1,	1800,			90,	"ipsecah_larval_timeout"},
89 	/* Default lifetime values for ACQUIRE messages. */
90 	{ 0,	0xffffffffU,		0,	"ipsecah_default_soft_bytes"},
91 	{ 0,	0xffffffffU,		0,	"ipsecah_default_hard_bytes"},
92 	{ 0,	0xffffffffU,		24000,	"ipsecah_default_soft_addtime"},
93 	{ 0,	0xffffffffU,		28800,	"ipsecah_default_hard_addtime"},
94 	{ 0,	0xffffffffU,		0,	"ipsecah_default_soft_usetime"},
95 	{ 0,	0xffffffffU,		0,	"ipsecah_default_hard_usetime"},
96 	{ 0,	1,			0,	"ipsecah_log_unknown_spi"},
97 };
98 #define	ipsecah_debug			ipsecah_params[0].ipsecah_param_value
99 #define	ipsecah_age_interval		ipsecah_params[1].ipsecah_param_value
100 #define	ipsecah_age_int_max		ipsecah_params[1].ipsecah_param_max
101 #define	ipsecah_reap_delay		ipsecah_params[2].ipsecah_param_value
102 #define	ipsecah_replay_size		ipsecah_params[3].ipsecah_param_value
103 #define	ipsecah_acquire_timeout		ipsecah_params[4].ipsecah_param_value
104 #define	ipsecah_larval_timeout		ipsecah_params[5].ipsecah_param_value
105 #define	ipsecah_default_soft_bytes	ipsecah_params[6].ipsecah_param_value
106 #define	ipsecah_default_hard_bytes	ipsecah_params[7].ipsecah_param_value
107 #define	ipsecah_default_soft_addtime	ipsecah_params[8].ipsecah_param_value
108 #define	ipsecah_default_hard_addtime	ipsecah_params[9].ipsecah_param_value
109 #define	ipsecah_default_soft_usetime	ipsecah_params[10].ipsecah_param_value
110 #define	ipsecah_default_hard_usetime	ipsecah_params[11].ipsecah_param_value
111 #define	ipsecah_log_unknown_spi		ipsecah_params[12].ipsecah_param_value
112 
113 #define	ah0dbg(a)	printf a
114 /* NOTE:  != 0 instead of > 0 so lint doesn't complain. */
115 #define	ah1dbg(ahstack, a)	if (ahstack->ipsecah_debug != 0) printf a
116 #define	ah2dbg(ahstack, a)	if (ahstack->ipsecah_debug > 1) printf a
117 #define	ah3dbg(ahstack, a)	if (ahstack->ipsecah_debug > 2) printf a
118 
119 /*
120  * XXX This is broken. Padding should be determined dynamically
121  * depending on the ICV size and IP version number so that the
122  * total AH header size is a multiple of 32 bits or 64 bits
123  * for V4 and V6 respectively. For 96bit ICVs we have no problems.
124  * Anything different from that, we need to fix our code.
125  */
126 #define	IPV4_PADDING_ALIGN	0x04	/* Multiple of 32 bits */
127 #define	IPV6_PADDING_ALIGN	0x04	/* Multiple of 32 bits */
128 
129 /*
130  * Helper macro. Avoids a call to msgdsize if there is only one
131  * mblk in the chain.
132  */
133 #define	AH_MSGSIZE(mp) ((mp)->b_cont != NULL ? msgdsize(mp) : MBLKL(mp))
134 
135 
136 static mblk_t *ah_auth_out_done(mblk_t *, ip_xmit_attr_t *, ipsec_crypto_t *);
137 static mblk_t *ah_auth_in_done(mblk_t *, ip_recv_attr_t *, ipsec_crypto_t *);
138 static mblk_t *ah_process_ip_options_v4(mblk_t *, ipsa_t *, int *, uint_t,
139     boolean_t, ipsecah_stack_t *);
140 static mblk_t *ah_process_ip_options_v6(mblk_t *, ipsa_t *, int *, uint_t,
141     boolean_t, ipsecah_stack_t *);
142 static void ah_getspi(mblk_t *, keysock_in_t *, ipsecah_stack_t *);
143 static void ah_inbound_restart(mblk_t *, ip_recv_attr_t *);
144 
145 static mblk_t *ah_outbound(mblk_t *, ip_xmit_attr_t *);
146 static void ah_outbound_finish(mblk_t *, ip_xmit_attr_t *);
147 
148 static int ipsecah_open(queue_t *, dev_t *, int, int, cred_t *);
149 static int ipsecah_close(queue_t *);
150 static void ipsecah_wput(queue_t *, mblk_t *);
151 static void ah_send_acquire(ipsacq_t *, mblk_t *, netstack_t *);
152 static boolean_t ah_register_out(uint32_t, uint32_t, uint_t, ipsecah_stack_t *,
153     cred_t *);
154 static void	*ipsecah_stack_init(netstackid_t stackid, netstack_t *ns);
155 static void	ipsecah_stack_fini(netstackid_t stackid, void *arg);
156 
157 /* Setable in /etc/system */
158 uint32_t ah_hash_size = IPSEC_DEFAULT_HASH_SIZE;
159 
160 static taskq_t *ah_taskq;
161 
162 static struct module_info info = {
163 	5136, "ipsecah", 0, INFPSZ, 65536, 1024
164 };
165 
166 static struct qinit rinit = {
167 	(pfi_t)putnext, NULL, ipsecah_open, ipsecah_close, NULL, &info,
168 	NULL
169 };
170 
171 static struct qinit winit = {
172 	(pfi_t)ipsecah_wput, NULL, ipsecah_open, ipsecah_close, NULL, &info,
173 	NULL
174 };
175 
176 struct streamtab ipsecahinfo = {
177 	&rinit, &winit, NULL, NULL
178 };
179 
180 static int ah_kstat_update(kstat_t *, int);
181 
182 uint64_t ipsacq_maxpackets = IPSACQ_MAXPACKETS;
183 
184 static boolean_t
185 ah_kstat_init(ipsecah_stack_t *ahstack, netstackid_t stackid)
186 {
187 	ipsec_stack_t	*ipss = ahstack->ipsecah_netstack->netstack_ipsec;
188 
189 	ahstack->ah_ksp = kstat_create_netstack("ipsecah", 0, "ah_stat", "net",
190 	    KSTAT_TYPE_NAMED, sizeof (ah_kstats_t) / sizeof (kstat_named_t),
191 	    KSTAT_FLAG_PERSISTENT, stackid);
192 
193 	if (ahstack->ah_ksp == NULL || ahstack->ah_ksp->ks_data == NULL)
194 		return (B_FALSE);
195 
196 	ahstack->ah_kstats = ahstack->ah_ksp->ks_data;
197 
198 	ahstack->ah_ksp->ks_update = ah_kstat_update;
199 	ahstack->ah_ksp->ks_private = (void *)(uintptr_t)stackid;
200 
201 #define	K64 KSTAT_DATA_UINT64
202 #define	KI(x) kstat_named_init(&(ahstack->ah_kstats->ah_stat_##x), #x, K64)
203 
204 	KI(num_aalgs);
205 	KI(good_auth);
206 	KI(bad_auth);
207 	KI(replay_failures);
208 	KI(replay_early_failures);
209 	KI(keysock_in);
210 	KI(out_requests);
211 	KI(acquire_requests);
212 	KI(bytes_expired);
213 	KI(out_discards);
214 	KI(crypto_sync);
215 	KI(crypto_async);
216 	KI(crypto_failures);
217 
218 #undef KI
219 #undef K64
220 
221 	kstat_install(ahstack->ah_ksp);
222 	IP_ACQUIRE_STAT(ipss, maxpackets, ipsacq_maxpackets);
223 	return (B_TRUE);
224 }
225 
226 static int
227 ah_kstat_update(kstat_t *kp, int rw)
228 {
229 	ah_kstats_t	*ekp;
230 	netstackid_t	stackid = (netstackid_t)(uintptr_t)kp->ks_private;
231 	netstack_t	*ns;
232 	ipsec_stack_t	*ipss;
233 
234 	if ((kp == NULL) || (kp->ks_data == NULL))
235 		return (EIO);
236 
237 	if (rw == KSTAT_WRITE)
238 		return (EACCES);
239 
240 	ns = netstack_find_by_stackid(stackid);
241 	if (ns == NULL)
242 		return (-1);
243 	ipss = ns->netstack_ipsec;
244 	if (ipss == NULL) {
245 		netstack_rele(ns);
246 		return (-1);
247 	}
248 	ekp = (ah_kstats_t *)kp->ks_data;
249 
250 	mutex_enter(&ipss->ipsec_alg_lock);
251 	ekp->ah_stat_num_aalgs.value.ui64 = ipss->ipsec_nalgs[IPSEC_ALG_AUTH];
252 	mutex_exit(&ipss->ipsec_alg_lock);
253 
254 	netstack_rele(ns);
255 	return (0);
256 }
257 
258 /*
259  * Don't have to lock ipsec_age_interval, as only one thread will access it at
260  * a time, because I control the one function that does a qtimeout() on
261  * ah_pfkey_q.
262  */
263 static void
264 ah_ager(void *arg)
265 {
266 	ipsecah_stack_t *ahstack = (ipsecah_stack_t *)arg;
267 	netstack_t	*ns = ahstack->ipsecah_netstack;
268 	hrtime_t begin = gethrtime();
269 
270 	sadb_ager(&ahstack->ah_sadb.s_v4, ahstack->ah_pfkey_q,
271 	    ahstack->ipsecah_reap_delay, ns);
272 	sadb_ager(&ahstack->ah_sadb.s_v6, ahstack->ah_pfkey_q,
273 	    ahstack->ipsecah_reap_delay, ns);
274 
275 	ahstack->ah_event = sadb_retimeout(begin, ahstack->ah_pfkey_q,
276 	    ah_ager, ahstack,
277 	    &ahstack->ipsecah_age_interval, ahstack->ipsecah_age_int_max,
278 	    info.mi_idnum);
279 }
280 
281 /*
282  * Get an AH NDD parameter.
283  */
284 /* ARGSUSED */
285 static int
286 ipsecah_param_get(q, mp, cp, cr)
287 	queue_t	*q;
288 	mblk_t	*mp;
289 	caddr_t	cp;
290 	cred_t *cr;
291 {
292 	ipsecahparam_t	*ipsecahpa = (ipsecahparam_t *)cp;
293 	uint_t value;
294 	ipsecah_stack_t	*ahstack = (ipsecah_stack_t *)q->q_ptr;
295 
296 	mutex_enter(&ahstack->ipsecah_param_lock);
297 	value = ipsecahpa->ipsecah_param_value;
298 	mutex_exit(&ahstack->ipsecah_param_lock);
299 
300 	(void) mi_mpprintf(mp, "%u", value);
301 	return (0);
302 }
303 
304 /*
305  * This routine sets an NDD variable in a ipsecahparam_t structure.
306  */
307 /* ARGSUSED */
308 static int
309 ipsecah_param_set(q, mp, value, cp, cr)
310 	queue_t	*q;
311 	mblk_t	*mp;
312 	char	*value;
313 	caddr_t	cp;
314 	cred_t *cr;
315 {
316 	ulong_t	new_value;
317 	ipsecahparam_t	*ipsecahpa = (ipsecahparam_t *)cp;
318 	ipsecah_stack_t	*ahstack = (ipsecah_stack_t *)q->q_ptr;
319 
320 	/*
321 	 * Fail the request if the new value does not lie within the
322 	 * required bounds.
323 	 */
324 	if (ddi_strtoul(value, NULL, 10, &new_value) != 0 ||
325 	    new_value < ipsecahpa->ipsecah_param_min ||
326 	    new_value > ipsecahpa->ipsecah_param_max) {
327 		    return (EINVAL);
328 	}
329 
330 	/* Set the new value */
331 	mutex_enter(&ahstack->ipsecah_param_lock);
332 	ipsecahpa->ipsecah_param_value = new_value;
333 	mutex_exit(&ahstack->ipsecah_param_lock);
334 	return (0);
335 }
336 
337 /*
338  * Using lifetime NDD variables, fill in an extended combination's
339  * lifetime information.
340  */
341 void
342 ipsecah_fill_defs(sadb_x_ecomb_t *ecomb, netstack_t *ns)
343 {
344 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
345 
346 	ecomb->sadb_x_ecomb_soft_bytes = ahstack->ipsecah_default_soft_bytes;
347 	ecomb->sadb_x_ecomb_hard_bytes = ahstack->ipsecah_default_hard_bytes;
348 	ecomb->sadb_x_ecomb_soft_addtime =
349 	    ahstack->ipsecah_default_soft_addtime;
350 	ecomb->sadb_x_ecomb_hard_addtime =
351 	    ahstack->ipsecah_default_hard_addtime;
352 	ecomb->sadb_x_ecomb_soft_usetime =
353 	    ahstack->ipsecah_default_soft_usetime;
354 	ecomb->sadb_x_ecomb_hard_usetime =
355 	    ahstack->ipsecah_default_hard_usetime;
356 }
357 
358 /*
359  * Initialize things for AH at module load time.
360  */
361 boolean_t
362 ipsecah_ddi_init(void)
363 {
364 	ah_taskq = taskq_create("ah_taskq", 1, minclsyspri,
365 	    IPSEC_TASKQ_MIN, IPSEC_TASKQ_MAX, 0);
366 
367 	/*
368 	 * We want to be informed each time a stack is created or
369 	 * destroyed in the kernel, so we can maintain the
370 	 * set of ipsecah_stack_t's.
371 	 */
372 	netstack_register(NS_IPSECAH, ipsecah_stack_init, NULL,
373 	    ipsecah_stack_fini);
374 
375 	return (B_TRUE);
376 }
377 
378 /*
379  * Walk through the param array specified registering each element with the
380  * named dispatch handler.
381  */
382 static boolean_t
383 ipsecah_param_register(IDP *ndp, ipsecahparam_t *ahp, int cnt)
384 {
385 	for (; cnt-- > 0; ahp++) {
386 		if (ahp->ipsecah_param_name != NULL &&
387 		    ahp->ipsecah_param_name[0]) {
388 			if (!nd_load(ndp,
389 			    ahp->ipsecah_param_name,
390 			    ipsecah_param_get, ipsecah_param_set,
391 			    (caddr_t)ahp)) {
392 				nd_free(ndp);
393 				return (B_FALSE);
394 			}
395 		}
396 	}
397 	return (B_TRUE);
398 }
399 
400 /*
401  * Initialize things for AH for each stack instance
402  */
403 static void *
404 ipsecah_stack_init(netstackid_t stackid, netstack_t *ns)
405 {
406 	ipsecah_stack_t	*ahstack;
407 	ipsecahparam_t	*ahp;
408 
409 	ahstack = (ipsecah_stack_t *)kmem_zalloc(sizeof (*ahstack), KM_SLEEP);
410 	ahstack->ipsecah_netstack = ns;
411 
412 	ahp = (ipsecahparam_t *)kmem_alloc(sizeof (lcl_param_arr), KM_SLEEP);
413 	ahstack->ipsecah_params = ahp;
414 	bcopy(lcl_param_arr, ahp, sizeof (lcl_param_arr));
415 
416 	(void) ipsecah_param_register(&ahstack->ipsecah_g_nd, ahp,
417 	    A_CNT(lcl_param_arr));
418 
419 	(void) ah_kstat_init(ahstack, stackid);
420 
421 	ahstack->ah_sadb.s_acquire_timeout = &ahstack->ipsecah_acquire_timeout;
422 	ahstack->ah_sadb.s_acqfn = ah_send_acquire;
423 	sadbp_init("AH", &ahstack->ah_sadb, SADB_SATYPE_AH, ah_hash_size,
424 	    ahstack->ipsecah_netstack);
425 
426 	mutex_init(&ahstack->ipsecah_param_lock, NULL, MUTEX_DEFAULT, 0);
427 
428 	ip_drop_register(&ahstack->ah_dropper, "IPsec AH");
429 	return (ahstack);
430 }
431 
432 /*
433  * Destroy things for AH at module unload time.
434  */
435 void
436 ipsecah_ddi_destroy(void)
437 {
438 	netstack_unregister(NS_IPSECAH);
439 	taskq_destroy(ah_taskq);
440 }
441 
442 /*
443  * Destroy things for AH for one stack... Never called?
444  */
445 static void
446 ipsecah_stack_fini(netstackid_t stackid, void *arg)
447 {
448 	ipsecah_stack_t *ahstack = (ipsecah_stack_t *)arg;
449 
450 	if (ahstack->ah_pfkey_q != NULL) {
451 		(void) quntimeout(ahstack->ah_pfkey_q, ahstack->ah_event);
452 	}
453 	ahstack->ah_sadb.s_acqfn = NULL;
454 	ahstack->ah_sadb.s_acquire_timeout = NULL;
455 	sadbp_destroy(&ahstack->ah_sadb, ahstack->ipsecah_netstack);
456 	ip_drop_unregister(&ahstack->ah_dropper);
457 	mutex_destroy(&ahstack->ipsecah_param_lock);
458 	nd_free(&ahstack->ipsecah_g_nd);
459 
460 	kmem_free(ahstack->ipsecah_params, sizeof (lcl_param_arr));
461 	ahstack->ipsecah_params = NULL;
462 	kstat_delete_netstack(ahstack->ah_ksp, stackid);
463 	ahstack->ah_ksp = NULL;
464 	ahstack->ah_kstats = NULL;
465 
466 	kmem_free(ahstack, sizeof (*ahstack));
467 }
468 
469 /*
470  * AH module open routine, which is here for keysock plumbing.
471  * Keysock is pushed over {AH,ESP} which is an artifact from the Bad Old
472  * Days of export control, and fears that ESP would not be allowed
473  * to be shipped at all by default.  Eventually, keysock should
474  * either access AH and ESP via modstubs or krtld dependencies, or
475  * perhaps be folded in with AH and ESP into a single IPsec/netsec
476  * module ("netsec" if PF_KEY provides more than AH/ESP keying tables).
477  */
478 /* ARGSUSED */
479 static int
480 ipsecah_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
481 {
482 	netstack_t	*ns;
483 	ipsecah_stack_t	*ahstack;
484 
485 	if (secpolicy_ip_config(credp, B_FALSE) != 0)
486 		return (EPERM);
487 
488 	if (q->q_ptr != NULL)
489 		return (0);  /* Re-open of an already open instance. */
490 
491 	if (sflag != MODOPEN)
492 		return (EINVAL);
493 
494 	ns = netstack_find_by_cred(credp);
495 	ASSERT(ns != NULL);
496 	ahstack = ns->netstack_ipsecah;
497 	ASSERT(ahstack != NULL);
498 
499 	q->q_ptr = ahstack;
500 	WR(q)->q_ptr = q->q_ptr;
501 
502 	qprocson(q);
503 	return (0);
504 }
505 
506 /*
507  * AH module close routine.
508  */
509 static int
510 ipsecah_close(queue_t *q)
511 {
512 	ipsecah_stack_t	*ahstack = (ipsecah_stack_t *)q->q_ptr;
513 
514 	/*
515 	 * Clean up q_ptr, if needed.
516 	 */
517 	qprocsoff(q);
518 
519 	/* Keysock queue check is safe, because of OCEXCL perimeter. */
520 
521 	if (q == ahstack->ah_pfkey_q) {
522 		ah1dbg(ahstack,
523 		    ("ipsecah_close:  Ummm... keysock is closing AH.\n"));
524 		ahstack->ah_pfkey_q = NULL;
525 		/* Detach qtimeouts. */
526 		(void) quntimeout(q, ahstack->ah_event);
527 	}
528 
529 	netstack_rele(ahstack->ipsecah_netstack);
530 	return (0);
531 }
532 
533 /*
534  * Construct an SADB_REGISTER message with the current algorithms.
535  */
536 static boolean_t
537 ah_register_out(uint32_t sequence, uint32_t pid, uint_t serial,
538     ipsecah_stack_t *ahstack, cred_t *cr)
539 {
540 	mblk_t *mp;
541 	boolean_t rc = B_TRUE;
542 	sadb_msg_t *samsg;
543 	sadb_supported_t *sasupp;
544 	sadb_alg_t *saalg;
545 	uint_t allocsize = sizeof (*samsg);
546 	uint_t i, numalgs_snap;
547 	ipsec_alginfo_t **authalgs;
548 	uint_t num_aalgs;
549 	ipsec_stack_t	*ipss = ahstack->ipsecah_netstack->netstack_ipsec;
550 	sadb_sens_t *sens;
551 	size_t sens_len = 0;
552 	sadb_ext_t *nextext;
553 	ts_label_t *sens_tsl = NULL;
554 
555 	/* Allocate the KEYSOCK_OUT. */
556 	mp = sadb_keysock_out(serial);
557 	if (mp == NULL) {
558 		ah0dbg(("ah_register_out: couldn't allocate mblk.\n"));
559 		return (B_FALSE);
560 	}
561 
562 	if (is_system_labeled() && (cr != NULL)) {
563 		sens_tsl = crgetlabel(cr);
564 		if (sens_tsl != NULL) {
565 			sens_len = sadb_sens_len_from_label(sens_tsl);
566 			allocsize += sens_len;
567 		}
568 	}
569 
570 	/*
571 	 * Allocate the PF_KEY message that follows KEYSOCK_OUT.
572 	 * The alg reader lock needs to be held while allocating
573 	 * the variable part (i.e. the algorithms) of the message.
574 	 */
575 
576 	mutex_enter(&ipss->ipsec_alg_lock);
577 
578 	/*
579 	 * Return only valid algorithms, so the number of algorithms
580 	 * to send up may be less than the number of algorithm entries
581 	 * in the table.
582 	 */
583 	authalgs = ipss->ipsec_alglists[IPSEC_ALG_AUTH];
584 	for (num_aalgs = 0, i = 0; i < IPSEC_MAX_ALGS; i++)
585 		if (authalgs[i] != NULL && ALG_VALID(authalgs[i]))
586 			num_aalgs++;
587 
588 	/*
589 	 * Fill SADB_REGISTER message's algorithm descriptors.  Hold
590 	 * down the lock while filling it.
591 	 */
592 	if (num_aalgs != 0) {
593 		allocsize += (num_aalgs * sizeof (*saalg));
594 		allocsize += sizeof (*sasupp);
595 	}
596 	mp->b_cont = allocb(allocsize, BPRI_HI);
597 	if (mp->b_cont == NULL) {
598 		mutex_exit(&ipss->ipsec_alg_lock);
599 		freemsg(mp);
600 		return (B_FALSE);
601 	}
602 
603 	mp->b_cont->b_wptr += allocsize;
604 	nextext = (sadb_ext_t *)(mp->b_cont->b_rptr + sizeof (*samsg));
605 
606 	if (num_aalgs != 0) {
607 
608 		saalg = (sadb_alg_t *)(((uint8_t *)nextext) + sizeof (*sasupp));
609 		ASSERT(((ulong_t)saalg & 0x7) == 0);
610 
611 		numalgs_snap = 0;
612 		for (i = 0;
613 		    ((i < IPSEC_MAX_ALGS) && (numalgs_snap < num_aalgs));
614 		    i++) {
615 			if (authalgs[i] == NULL || !ALG_VALID(authalgs[i]))
616 				continue;
617 
618 			saalg->sadb_alg_id = authalgs[i]->alg_id;
619 			saalg->sadb_alg_ivlen = 0;
620 			saalg->sadb_alg_minbits = authalgs[i]->alg_ef_minbits;
621 			saalg->sadb_alg_maxbits = authalgs[i]->alg_ef_maxbits;
622 			saalg->sadb_x_alg_increment =
623 			    authalgs[i]->alg_increment;
624 			/* For now, salt is meaningless in AH. */
625 			ASSERT(authalgs[i]->alg_saltlen == 0);
626 			saalg->sadb_x_alg_saltbits =
627 			    SADB_8TO1(authalgs[i]->alg_saltlen);
628 			numalgs_snap++;
629 			saalg++;
630 		}
631 		ASSERT(numalgs_snap == num_aalgs);
632 #ifdef DEBUG
633 		/*
634 		 * Reality check to make sure I snagged all of the
635 		 * algorithms.
636 		 */
637 		for (; i < IPSEC_MAX_ALGS; i++)
638 			if (authalgs[i] != NULL && ALG_VALID(authalgs[i]))
639 				cmn_err(CE_PANIC,
640 				    "ah_register_out()!  Missed #%d.\n", i);
641 #endif /* DEBUG */
642 		nextext = (sadb_ext_t *)saalg;
643 	}
644 
645 	mutex_exit(&ipss->ipsec_alg_lock);
646 
647 	if (sens_tsl != NULL) {
648 		sens = (sadb_sens_t *)nextext;
649 		sadb_sens_from_label(sens, SADB_EXT_SENSITIVITY,
650 		    sens_tsl, sens_len);
651 
652 		nextext = (sadb_ext_t *)(((uint8_t *)sens) + sens_len);
653 	}
654 
655 	/* Now fill the restof the SADB_REGISTER message. */
656 
657 	samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
658 	samsg->sadb_msg_version = PF_KEY_V2;
659 	samsg->sadb_msg_type = SADB_REGISTER;
660 	samsg->sadb_msg_errno = 0;
661 	samsg->sadb_msg_satype = SADB_SATYPE_AH;
662 	samsg->sadb_msg_len = SADB_8TO64(allocsize);
663 	samsg->sadb_msg_reserved = 0;
664 	/*
665 	 * Assume caller has sufficient sequence/pid number info.  If it's one
666 	 * from me over a new alg., I could give two hoots about sequence.
667 	 */
668 	samsg->sadb_msg_seq = sequence;
669 	samsg->sadb_msg_pid = pid;
670 
671 	if (num_aalgs != 0) {
672 		sasupp = (sadb_supported_t *)(samsg + 1);
673 		sasupp->sadb_supported_len = SADB_8TO64(
674 		    sizeof (*sasupp) + sizeof (*saalg) * num_aalgs);
675 		sasupp->sadb_supported_exttype = SADB_EXT_SUPPORTED_AUTH;
676 		sasupp->sadb_supported_reserved = 0;
677 	}
678 
679 	if (ahstack->ah_pfkey_q != NULL)
680 		putnext(ahstack->ah_pfkey_q, mp);
681 	else {
682 		rc = B_FALSE;
683 		freemsg(mp);
684 	}
685 
686 	return (rc);
687 }
688 
689 /*
690  * Invoked when the algorithm table changes. Causes SADB_REGISTER
691  * messages continaining the current list of algorithms to be
692  * sent up to the AH listeners.
693  */
694 void
695 ipsecah_algs_changed(netstack_t *ns)
696 {
697 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
698 
699 	/*
700 	 * Time to send a PF_KEY SADB_REGISTER message to AH listeners
701 	 * everywhere.  (The function itself checks for NULL ah_pfkey_q.)
702 	 */
703 	(void) ah_register_out(0, 0, 0, ahstack, NULL);
704 }
705 
706 /*
707  * Stub function that taskq_dispatch() invokes to take the mblk (in arg)
708  * and send it into AH and IP again.
709  */
710 static void
711 inbound_task(void *arg)
712 {
713 	mblk_t		*mp = (mblk_t *)arg;
714 	mblk_t		*async_mp;
715 	ip_recv_attr_t	iras;
716 
717 	async_mp = mp;
718 	mp = async_mp->b_cont;
719 	async_mp->b_cont = NULL;
720 	if (!ip_recv_attr_from_mblk(async_mp, &iras)) {
721 		/* The ill or ip_stack_t disappeared on us */
722 		ip_drop_input("ip_recv_attr_from_mblk", mp, NULL);
723 		freemsg(mp);
724 		goto done;
725 	}
726 
727 	ah_inbound_restart(mp, &iras);
728 done:
729 	ira_cleanup(&iras, B_TRUE);
730 }
731 
732 /*
733  * Restart ESP after the SA has been added.
734  */
735 static void
736 ah_inbound_restart(mblk_t *mp, ip_recv_attr_t *ira)
737 {
738 	ah_t		*ah;
739 	netstack_t	*ns;
740 	ipsecah_stack_t	*ahstack;
741 
742 	ns = ira->ira_ill->ill_ipst->ips_netstack;
743 	ahstack = ns->netstack_ipsecah;
744 
745 	ASSERT(ahstack != NULL);
746 	mp = ipsec_inbound_ah_sa(mp, ira, &ah);
747 	if (mp == NULL)
748 		return;
749 
750 	ASSERT(ah != NULL);
751 	ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
752 	ASSERT(ira->ira_ipsec_ah_sa != NULL);
753 
754 	mp = ira->ira_ipsec_ah_sa->ipsa_input_func(mp, ah, ira);
755 	if (mp == NULL) {
756 		/*
757 		 * Either it failed or is pending. In the former case
758 		 * ipIfStatsInDiscards was increased.
759 		 */
760 		return;
761 	}
762 	ip_input_post_ipsec(mp, ira);
763 }
764 
765 /*
766  * Now that weak-key passed, actually ADD the security association, and
767  * send back a reply ADD message.
768  */
769 static int
770 ah_add_sa_finish(mblk_t *mp, sadb_msg_t *samsg, keysock_in_t *ksi,
771     int *diagnostic, ipsecah_stack_t *ahstack)
772 {
773 	isaf_t *primary = NULL, *secondary;
774 	boolean_t clone = B_FALSE, is_inbound = B_FALSE;
775 	sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
776 	ipsa_t *larval;
777 	ipsacq_t *acqrec;
778 	iacqf_t *acq_bucket;
779 	mblk_t *acq_msgs = NULL;
780 	mblk_t *lpkt;
781 	int rc;
782 	ipsa_query_t sq;
783 	int error;
784 	netstack_t	*ns = ahstack->ipsecah_netstack;
785 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
786 
787 	/*
788 	 * Locate the appropriate table(s).
789 	 */
790 
791 	sq.spp = &ahstack->ah_sadb;
792 	error = sadb_form_query(ksi, IPSA_Q_SA|IPSA_Q_DST,
793 	    IPSA_Q_SA|IPSA_Q_DST|IPSA_Q_INBOUND|IPSA_Q_OUTBOUND,
794 	    &sq, diagnostic);
795 	if (error)
796 		return (error);
797 
798 	/*
799 	 * Use the direction flags provided by the KMD to determine
800 	 * if the inbound or outbound table should be the primary
801 	 * for this SA. If these flags were absent then make this
802 	 * decision based on the addresses.
803 	 */
804 	if (assoc->sadb_sa_flags & IPSA_F_INBOUND) {
805 		primary = sq.inbound;
806 		secondary = sq.outbound;
807 		is_inbound = B_TRUE;
808 		if (assoc->sadb_sa_flags & IPSA_F_OUTBOUND)
809 			clone = B_TRUE;
810 	} else {
811 		if (assoc->sadb_sa_flags & IPSA_F_OUTBOUND) {
812 			primary = sq.outbound;
813 			secondary = sq.inbound;
814 		}
815 	}
816 	if (primary == NULL) {
817 		/*
818 		 * The KMD did not set a direction flag, determine which
819 		 * table to insert the SA into based on addresses.
820 		 */
821 		switch (ksi->ks_in_dsttype) {
822 		case KS_IN_ADDR_MBCAST:
823 			clone = B_TRUE;	/* All mcast SAs can be bidirectional */
824 			assoc->sadb_sa_flags |= IPSA_F_OUTBOUND;
825 			/* FALLTHRU */
826 		/*
827 		 * If the source address is either one of mine, or unspecified
828 		 * (which is best summed up by saying "not 'not mine'"),
829 		 * then the association is potentially bi-directional,
830 		 * in that it can be used for inbound traffic and outbound
831 		 * traffic.  The best example of such and SA is a multicast
832 		 * SA (which allows me to receive the outbound traffic).
833 		 */
834 		case KS_IN_ADDR_ME:
835 			assoc->sadb_sa_flags |= IPSA_F_INBOUND;
836 			primary = sq.inbound;
837 			secondary = sq.outbound;
838 			if (ksi->ks_in_srctype != KS_IN_ADDR_NOTME)
839 				clone = B_TRUE;
840 			is_inbound = B_TRUE;
841 			break;
842 
843 		/*
844 		 * If the source address literally not mine (either
845 		 * unspecified or not mine), then this SA may have an
846 		 * address that WILL be mine after some configuration.
847 		 * We pay the price for this by making it a bi-directional
848 		 * SA.
849 		 */
850 		case KS_IN_ADDR_NOTME:
851 			assoc->sadb_sa_flags |= IPSA_F_OUTBOUND;
852 			primary = sq.outbound;
853 			secondary = sq.inbound;
854 			if (ksi->ks_in_srctype != KS_IN_ADDR_ME) {
855 				assoc->sadb_sa_flags |= IPSA_F_INBOUND;
856 				clone = B_TRUE;
857 			}
858 			break;
859 		default:
860 			*diagnostic = SADB_X_DIAGNOSTIC_BAD_DST;
861 			return (EINVAL);
862 		}
863 	}
864 
865 	/*
866 	 * Find a ACQUIRE list entry if possible.  If we've added an SA that
867 	 * suits the needs of an ACQUIRE list entry, we can eliminate the
868 	 * ACQUIRE list entry and transmit the enqueued packets.  Use the
869 	 * high-bit of the sequence number to queue it.  Key off destination
870 	 * addr, and change acqrec's state.
871 	 */
872 
873 	if (samsg->sadb_msg_seq & IACQF_LOWEST_SEQ) {
874 		acq_bucket = &(sq.sp->sdb_acq[sq.outhash]);
875 		mutex_enter(&acq_bucket->iacqf_lock);
876 		for (acqrec = acq_bucket->iacqf_ipsacq; acqrec != NULL;
877 		    acqrec = acqrec->ipsacq_next) {
878 			mutex_enter(&acqrec->ipsacq_lock);
879 			/*
880 			 * Q:  I only check sequence.  Should I check dst?
881 			 * A: Yes, check dest because those are the packets
882 			 *    that are queued up.
883 			 */
884 			if (acqrec->ipsacq_seq == samsg->sadb_msg_seq &&
885 			    IPSA_ARE_ADDR_EQUAL(sq.dstaddr,
886 			    acqrec->ipsacq_dstaddr, acqrec->ipsacq_addrfam))
887 				break;
888 			mutex_exit(&acqrec->ipsacq_lock);
889 		}
890 		if (acqrec != NULL) {
891 			/*
892 			 * AHA!  I found an ACQUIRE record for this SA.
893 			 * Grab the msg list, and free the acquire record.
894 			 * I already am holding the lock for this record,
895 			 * so all I have to do is free it.
896 			 */
897 			acq_msgs = acqrec->ipsacq_mp;
898 			acqrec->ipsacq_mp = NULL;
899 			mutex_exit(&acqrec->ipsacq_lock);
900 			sadb_destroy_acquire(acqrec, ns);
901 		}
902 		mutex_exit(&acq_bucket->iacqf_lock);
903 	}
904 
905 	/*
906 	 * Find PF_KEY message, and see if I'm an update.  If so, find entry
907 	 * in larval list (if there).
908 	 */
909 
910 	larval = NULL;
911 
912 	if (samsg->sadb_msg_type == SADB_UPDATE) {
913 		mutex_enter(&sq.inbound->isaf_lock);
914 		larval = ipsec_getassocbyspi(sq.inbound, sq.assoc->sadb_sa_spi,
915 		    ALL_ZEROES_PTR, sq.dstaddr, sq.dst->sin_family);
916 		mutex_exit(&sq.inbound->isaf_lock);
917 
918 		if ((larval == NULL) ||
919 		    (larval->ipsa_state != IPSA_STATE_LARVAL)) {
920 			*diagnostic = SADB_X_DIAGNOSTIC_SA_NOTFOUND;
921 			if (larval != NULL) {
922 				IPSA_REFRELE(larval);
923 			}
924 			ah0dbg(("Larval update, but larval disappeared.\n"));
925 			return (ESRCH);
926 		} /* Else sadb_common_add unlinks it for me! */
927 	}
928 
929 	if (larval != NULL) {
930 		/*
931 		 * Hold again, because sadb_common_add() consumes a reference,
932 		 * and we don't want to clear_lpkt() without a reference.
933 		 */
934 		IPSA_REFHOLD(larval);
935 	}
936 
937 	rc = sadb_common_add(ahstack->ah_pfkey_q, mp,
938 	    samsg, ksi, primary, secondary, larval, clone, is_inbound,
939 	    diagnostic, ns, &ahstack->ah_sadb);
940 
941 	if (larval != NULL) {
942 		if (rc == 0) {
943 			lpkt = sadb_clear_lpkt(larval);
944 			if (lpkt != NULL) {
945 				rc = !taskq_dispatch(ah_taskq, inbound_task,
946 				    lpkt, TQ_NOSLEEP);
947 			}
948 		}
949 		IPSA_REFRELE(larval);
950 	}
951 
952 	/*
953 	 * How much more stack will I create with all of these
954 	 * ah_outbound_*() calls?
955 	 */
956 
957 	/* Handle the packets queued waiting for the SA */
958 	while (acq_msgs != NULL) {
959 		mblk_t		*asyncmp;
960 		mblk_t		*data_mp;
961 		ip_xmit_attr_t	ixas;
962 		ill_t		*ill;
963 
964 		asyncmp = acq_msgs;
965 		acq_msgs = acq_msgs->b_next;
966 		asyncmp->b_next = NULL;
967 
968 		/*
969 		 * Extract the ip_xmit_attr_t from the first mblk.
970 		 * Verifies that the netstack and ill is still around; could
971 		 * have vanished while iked was doing its work.
972 		 * On succesful return we have a nce_t and the ill/ipst can't
973 		 * disappear until we do the nce_refrele in ixa_cleanup.
974 		 */
975 		data_mp = asyncmp->b_cont;
976 		asyncmp->b_cont = NULL;
977 		if (!ip_xmit_attr_from_mblk(asyncmp, &ixas)) {
978 			AH_BUMP_STAT(ahstack, out_discards);
979 			ip_drop_packet(data_mp, B_FALSE, NULL,
980 			    DROPPER(ipss, ipds_sadb_acquire_timeout),
981 			    &ahstack->ah_dropper);
982 		} else if (rc != 0) {
983 			ill = ixas.ixa_nce->nce_ill;
984 			AH_BUMP_STAT(ahstack, out_discards);
985 			ip_drop_packet(data_mp, B_FALSE, ill,
986 			    DROPPER(ipss, ipds_sadb_acquire_timeout),
987 			    &ahstack->ah_dropper);
988 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
989 		} else {
990 			ah_outbound_finish(data_mp, &ixas);
991 		}
992 		ixa_cleanup(&ixas);
993 	}
994 
995 	return (rc);
996 }
997 
998 
999 /*
1000  * Process one of the queued messages (from ipsacq_mp) once the SA
1001  * has been added.
1002  */
1003 static void
1004 ah_outbound_finish(mblk_t *data_mp, ip_xmit_attr_t *ixa)
1005 {
1006 	netstack_t	*ns = ixa->ixa_ipst->ips_netstack;
1007 	ipsecah_stack_t *ahstack = ns->netstack_ipsecah;
1008 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
1009 	ill_t		*ill = ixa->ixa_nce->nce_ill;
1010 
1011 	if (!ipsec_outbound_sa(data_mp, ixa, IPPROTO_AH)) {
1012 		AH_BUMP_STAT(ahstack, out_discards);
1013 		ip_drop_packet(data_mp, B_FALSE, ill,
1014 		    DROPPER(ipss, ipds_sadb_acquire_timeout),
1015 		    &ahstack->ah_dropper);
1016 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1017 		return;
1018 	}
1019 
1020 	data_mp = ah_outbound(data_mp, ixa);
1021 	if (data_mp == NULL)
1022 		return;
1023 
1024 	(void) ip_output_post_ipsec(data_mp, ixa);
1025 }
1026 
1027 /*
1028  * Add new AH security association.  This may become a generic AH/ESP
1029  * routine eventually.
1030  */
1031 static int
1032 ah_add_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic, netstack_t *ns)
1033 {
1034 	sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
1035 	sadb_address_t *srcext =
1036 	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
1037 	sadb_address_t *dstext =
1038 	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
1039 	sadb_address_t *isrcext =
1040 	    (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_SRC];
1041 	sadb_address_t *idstext =
1042 	    (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_DST];
1043 	sadb_key_t *key = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_AUTH];
1044 	struct sockaddr_in *src, *dst;
1045 	/* We don't need sockaddr_in6 for now. */
1046 	sadb_lifetime_t *soft =
1047 	    (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_SOFT];
1048 	sadb_lifetime_t *hard =
1049 	    (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_HARD];
1050 	sadb_lifetime_t *idle =
1051 	    (sadb_lifetime_t *)ksi->ks_in_extv[SADB_X_EXT_LIFETIME_IDLE];
1052 	ipsec_alginfo_t *aalg;
1053 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
1054 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
1055 
1056 	/* I need certain extensions present for an ADD message. */
1057 	if (srcext == NULL) {
1058 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_SRC;
1059 		return (EINVAL);
1060 	}
1061 	if (dstext == NULL) {
1062 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
1063 		return (EINVAL);
1064 	}
1065 	if (isrcext == NULL && idstext != NULL) {
1066 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_SRC;
1067 		return (EINVAL);
1068 	}
1069 	if (isrcext != NULL && idstext == NULL) {
1070 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_DST;
1071 		return (EINVAL);
1072 	}
1073 	if (assoc == NULL) {
1074 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
1075 		return (EINVAL);
1076 	}
1077 	if (key == NULL) {
1078 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_AKEY;
1079 		return (EINVAL);
1080 	}
1081 
1082 	src = (struct sockaddr_in *)(srcext + 1);
1083 	dst = (struct sockaddr_in *)(dstext + 1);
1084 
1085 	/* Sundry ADD-specific reality checks. */
1086 	/* XXX STATS : Logging/stats here? */
1087 
1088 	if ((assoc->sadb_sa_state != SADB_SASTATE_MATURE) &&
1089 	    (assoc->sadb_sa_state != SADB_X_SASTATE_ACTIVE_ELSEWHERE)) {
1090 		*diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
1091 		return (EINVAL);
1092 	}
1093 	if (assoc->sadb_sa_encrypt != SADB_EALG_NONE) {
1094 		*diagnostic = SADB_X_DIAGNOSTIC_ENCR_NOTSUPP;
1095 		return (EINVAL);
1096 	}
1097 	if (assoc->sadb_sa_flags & ~ahstack->ah_sadb.s_addflags) {
1098 		*diagnostic = SADB_X_DIAGNOSTIC_BAD_SAFLAGS;
1099 		return (EINVAL);
1100 	}
1101 	if ((*diagnostic = sadb_hardsoftchk(hard, soft, idle)) != 0)
1102 		return (EINVAL);
1103 
1104 	ASSERT(src->sin_family == dst->sin_family);
1105 
1106 	/* Stuff I don't support, for now.  XXX Diagnostic? */
1107 	if (ksi->ks_in_extv[SADB_EXT_LIFETIME_CURRENT] != NULL)
1108 		return (EOPNOTSUPP);
1109 
1110 	if (ksi->ks_in_extv[SADB_EXT_SENSITIVITY] != NULL) {
1111 		if (!is_system_labeled())
1112 			return (EOPNOTSUPP);
1113 	}
1114 
1115 	if (ksi->ks_in_extv[SADB_X_EXT_OUTER_SENS] != NULL) {
1116 		if (!is_system_labeled())
1117 			return (EOPNOTSUPP);
1118 	}
1119 	/*
1120 	 * XXX Policy : I'm not checking identities at this time, but
1121 	 * if I did, I'd do them here, before I sent the weak key
1122 	 * check up to the algorithm.
1123 	 */
1124 
1125 	/* verify that there is a mapping for the specified algorithm */
1126 	mutex_enter(&ipss->ipsec_alg_lock);
1127 	aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH][assoc->sadb_sa_auth];
1128 	if (aalg == NULL || !ALG_VALID(aalg)) {
1129 		mutex_exit(&ipss->ipsec_alg_lock);
1130 		ah1dbg(ahstack, ("Couldn't find auth alg #%d.\n",
1131 		    assoc->sadb_sa_auth));
1132 		*diagnostic = SADB_X_DIAGNOSTIC_BAD_AALG;
1133 		return (EINVAL);
1134 	}
1135 	ASSERT(aalg->alg_mech_type != CRYPTO_MECHANISM_INVALID);
1136 
1137 	/* sanity check key sizes */
1138 	if (!ipsec_valid_key_size(key->sadb_key_bits, aalg)) {
1139 		mutex_exit(&ipss->ipsec_alg_lock);
1140 		*diagnostic = SADB_X_DIAGNOSTIC_BAD_AKEYBITS;
1141 		return (EINVAL);
1142 	}
1143 
1144 	/* check key and fix parity if needed */
1145 	if (ipsec_check_key(aalg->alg_mech_type, key, B_TRUE,
1146 	    diagnostic) != 0) {
1147 		mutex_exit(&ipss->ipsec_alg_lock);
1148 		return (EINVAL);
1149 	}
1150 
1151 	mutex_exit(&ipss->ipsec_alg_lock);
1152 
1153 	return (ah_add_sa_finish(mp, (sadb_msg_t *)mp->b_cont->b_rptr, ksi,
1154 	    diagnostic, ahstack));
1155 }
1156 
1157 /* Refactor me */
1158 /*
1159  * Update a security association.  Updates come in two varieties.  The first
1160  * is an update of lifetimes on a non-larval SA.  The second is an update of
1161  * a larval SA, which ends up looking a lot more like an add.
1162  */
1163 static int
1164 ah_update_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic,
1165     ipsecah_stack_t *ahstack, uint8_t sadb_msg_type)
1166 {
1167 	sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
1168 	sadb_address_t *dstext =
1169 	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
1170 	mblk_t	*buf_pkt;
1171 	int rcode;
1172 
1173 	if (dstext == NULL) {
1174 		*diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
1175 		return (EINVAL);
1176 	}
1177 
1178 	rcode = sadb_update_sa(mp, ksi, &buf_pkt, &ahstack->ah_sadb,
1179 	    diagnostic, ahstack->ah_pfkey_q, ah_add_sa,
1180 	    ahstack->ipsecah_netstack, sadb_msg_type);
1181 
1182 	if ((assoc->sadb_sa_state != SADB_X_SASTATE_ACTIVE) ||
1183 	    (rcode != 0)) {
1184 		return (rcode);
1185 	}
1186 
1187 	HANDLE_BUF_PKT(ah_taskq, ahstack->ipsecah_netstack->netstack_ipsec,
1188 	    ahstack->ah_dropper, buf_pkt);
1189 
1190 	return (rcode);
1191 }
1192 
1193 /* Refactor me */
1194 /*
1195  * Delete a security association.  This is REALLY likely to be code common to
1196  * both AH and ESP.  Find the association, then unlink it.
1197  */
1198 static int
1199 ah_del_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic,
1200     ipsecah_stack_t *ahstack, uint8_t sadb_msg_type)
1201 {
1202 	sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
1203 	sadb_address_t *dstext =
1204 	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
1205 	sadb_address_t *srcext =
1206 	    (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
1207 	struct sockaddr_in *sin;
1208 
1209 	if (assoc == NULL) {
1210 		if (dstext != NULL)
1211 			sin = (struct sockaddr_in *)(dstext + 1);
1212 		else if (srcext != NULL)
1213 			sin = (struct sockaddr_in *)(srcext + 1);
1214 		else {
1215 			*diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
1216 			return (EINVAL);
1217 		}
1218 		return (sadb_purge_sa(mp, ksi,
1219 		    (sin->sin_family == AF_INET6) ? &ahstack->ah_sadb.s_v6 :
1220 		    &ahstack->ah_sadb.s_v4, diagnostic, ahstack->ah_pfkey_q));
1221 	}
1222 
1223 	return (sadb_delget_sa(mp, ksi, &ahstack->ah_sadb, diagnostic,
1224 	    ahstack->ah_pfkey_q, sadb_msg_type));
1225 }
1226 
1227 /* Refactor me */
1228 /*
1229  * Convert the entire contents of all of AH's SA tables into PF_KEY SADB_DUMP
1230  * messages.
1231  */
1232 static void
1233 ah_dump(mblk_t *mp, keysock_in_t *ksi, ipsecah_stack_t *ahstack)
1234 {
1235 	int error;
1236 	sadb_msg_t *samsg;
1237 
1238 	/*
1239 	 * Dump each fanout, bailing if error is non-zero.
1240 	 */
1241 
1242 	error = sadb_dump(ahstack->ah_pfkey_q, mp, ksi, &ahstack->ah_sadb.s_v4);
1243 	if (error != 0)
1244 		goto bail;
1245 
1246 	error = sadb_dump(ahstack->ah_pfkey_q, mp, ksi, &ahstack->ah_sadb.s_v6);
1247 bail:
1248 	ASSERT(mp->b_cont != NULL);
1249 	samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
1250 	samsg->sadb_msg_errno = (uint8_t)error;
1251 	sadb_pfkey_echo(ahstack->ah_pfkey_q, mp,
1252 	    (sadb_msg_t *)mp->b_cont->b_rptr, ksi, NULL);
1253 }
1254 
1255 /*
1256  * First-cut reality check for an inbound PF_KEY message.
1257  */
1258 static boolean_t
1259 ah_pfkey_reality_failures(mblk_t *mp, keysock_in_t *ksi,
1260     ipsecah_stack_t *ahstack)
1261 {
1262 	int diagnostic;
1263 
1264 	if (mp->b_cont == NULL) {
1265 		freemsg(mp);
1266 		return (B_TRUE);
1267 	}
1268 
1269 	if (ksi->ks_in_extv[SADB_EXT_KEY_ENCRYPT] != NULL) {
1270 		diagnostic = SADB_X_DIAGNOSTIC_EKEY_PRESENT;
1271 		goto badmsg;
1272 	}
1273 	if (ksi->ks_in_extv[SADB_EXT_PROPOSAL] != NULL) {
1274 		diagnostic = SADB_X_DIAGNOSTIC_PROP_PRESENT;
1275 		goto badmsg;
1276 	}
1277 	if (ksi->ks_in_extv[SADB_EXT_SUPPORTED_AUTH] != NULL ||
1278 	    ksi->ks_in_extv[SADB_EXT_SUPPORTED_ENCRYPT] != NULL) {
1279 		diagnostic = SADB_X_DIAGNOSTIC_SUPP_PRESENT;
1280 		goto badmsg;
1281 	}
1282 	return (B_FALSE);	/* False ==> no failures */
1283 
1284 badmsg:
1285 	sadb_pfkey_error(ahstack->ah_pfkey_q, mp, EINVAL,
1286 	    diagnostic, ksi->ks_in_serial);
1287 	return (B_TRUE);	/* True ==> failures */
1288 }
1289 
1290 /*
1291  * AH parsing of PF_KEY messages.  Keysock did most of the really silly
1292  * error cases.  What I receive is a fully-formed, syntactically legal
1293  * PF_KEY message.  I then need to check semantics...
1294  *
1295  * This code may become common to AH and ESP.  Stay tuned.
1296  *
1297  * I also make the assumption that db_ref's are cool.  If this assumption
1298  * is wrong, this means that someone other than keysock or me has been
1299  * mucking with PF_KEY messages.
1300  */
1301 static void
1302 ah_parse_pfkey(mblk_t *mp, ipsecah_stack_t *ahstack)
1303 {
1304 	mblk_t *msg = mp->b_cont;
1305 	sadb_msg_t *samsg;
1306 	keysock_in_t *ksi;
1307 	int error;
1308 	int diagnostic = SADB_X_DIAGNOSTIC_NONE;
1309 
1310 	ASSERT(msg != NULL);
1311 
1312 	samsg = (sadb_msg_t *)msg->b_rptr;
1313 	ksi = (keysock_in_t *)mp->b_rptr;
1314 
1315 	/*
1316 	 * If applicable, convert unspecified AF_INET6 to unspecified
1317 	 * AF_INET.
1318 	 */
1319 	if (!sadb_addrfix(ksi, ahstack->ah_pfkey_q, mp,
1320 	    ahstack->ipsecah_netstack) ||
1321 	    ah_pfkey_reality_failures(mp, ksi, ahstack)) {
1322 		return;
1323 	}
1324 
1325 	switch (samsg->sadb_msg_type) {
1326 	case SADB_ADD:
1327 		error = ah_add_sa(mp, ksi, &diagnostic,
1328 		    ahstack->ipsecah_netstack);
1329 		if (error != 0) {
1330 			sadb_pfkey_error(ahstack->ah_pfkey_q, mp, error,
1331 			    diagnostic, ksi->ks_in_serial);
1332 		}
1333 		/* else ah_add_sa() took care of things. */
1334 		break;
1335 	case SADB_DELETE:
1336 	case SADB_X_DELPAIR:
1337 	case SADB_X_DELPAIR_STATE:
1338 		error = ah_del_sa(mp, ksi, &diagnostic, ahstack,
1339 		    samsg->sadb_msg_type);
1340 		if (error != 0) {
1341 			sadb_pfkey_error(ahstack->ah_pfkey_q, mp, error,
1342 			    diagnostic, ksi->ks_in_serial);
1343 		}
1344 		/* Else ah_del_sa() took care of things. */
1345 		break;
1346 	case SADB_GET:
1347 		error = sadb_delget_sa(mp, ksi, &ahstack->ah_sadb, &diagnostic,
1348 		    ahstack->ah_pfkey_q, samsg->sadb_msg_type);
1349 		if (error != 0) {
1350 			sadb_pfkey_error(ahstack->ah_pfkey_q, mp, error,
1351 			    diagnostic, ksi->ks_in_serial);
1352 		}
1353 		/* Else sadb_get_sa() took care of things. */
1354 		break;
1355 	case SADB_FLUSH:
1356 		sadbp_flush(&ahstack->ah_sadb, ahstack->ipsecah_netstack);
1357 		sadb_pfkey_echo(ahstack->ah_pfkey_q, mp, samsg, ksi, NULL);
1358 		break;
1359 	case SADB_REGISTER:
1360 		/*
1361 		 * Hmmm, let's do it!  Check for extensions (there should
1362 		 * be none), extract the fields, call ah_register_out(),
1363 		 * then either free or report an error.
1364 		 *
1365 		 * Keysock takes care of the PF_KEY bookkeeping for this.
1366 		 */
1367 		if (ah_register_out(samsg->sadb_msg_seq, samsg->sadb_msg_pid,
1368 		    ksi->ks_in_serial, ahstack, msg_getcred(mp, NULL))) {
1369 			freemsg(mp);
1370 		} else {
1371 			/*
1372 			 * Only way this path hits is if there is a memory
1373 			 * failure.  It will not return B_FALSE because of
1374 			 * lack of ah_pfkey_q if I am in wput().
1375 			 */
1376 			sadb_pfkey_error(ahstack->ah_pfkey_q, mp, ENOMEM,
1377 			    diagnostic, ksi->ks_in_serial);
1378 		}
1379 		break;
1380 	case SADB_UPDATE:
1381 	case SADB_X_UPDATEPAIR:
1382 		/*
1383 		 * Find a larval, if not there, find a full one and get
1384 		 * strict.
1385 		 */
1386 		error = ah_update_sa(mp, ksi, &diagnostic, ahstack,
1387 		    samsg->sadb_msg_type);
1388 		if (error != 0) {
1389 			sadb_pfkey_error(ahstack->ah_pfkey_q, mp, error,
1390 			    diagnostic, ksi->ks_in_serial);
1391 		}
1392 		/* else ah_update_sa() took care of things. */
1393 		break;
1394 	case SADB_GETSPI:
1395 		/*
1396 		 * Reserve a new larval entry.
1397 		 */
1398 		ah_getspi(mp, ksi, ahstack);
1399 		break;
1400 	case SADB_ACQUIRE:
1401 		/*
1402 		 * Find larval and/or ACQUIRE record and kill it (them), I'm
1403 		 * most likely an error.  Inbound ACQUIRE messages should only
1404 		 * have the base header.
1405 		 */
1406 		sadb_in_acquire(samsg, &ahstack->ah_sadb, ahstack->ah_pfkey_q,
1407 		    ahstack->ipsecah_netstack);
1408 		freemsg(mp);
1409 		break;
1410 	case SADB_DUMP:
1411 		/*
1412 		 * Dump all entries.
1413 		 */
1414 		ah_dump(mp, ksi, ahstack);
1415 		/* ah_dump will take care of the return message, etc. */
1416 		break;
1417 	case SADB_EXPIRE:
1418 		/* Should never reach me. */
1419 		sadb_pfkey_error(ahstack->ah_pfkey_q, mp, EOPNOTSUPP,
1420 		    diagnostic, ksi->ks_in_serial);
1421 		break;
1422 	default:
1423 		sadb_pfkey_error(ahstack->ah_pfkey_q, mp, EINVAL,
1424 		    SADB_X_DIAGNOSTIC_UNKNOWN_MSG, ksi->ks_in_serial);
1425 		break;
1426 	}
1427 }
1428 
1429 /*
1430  * Handle case where PF_KEY says it can't find a keysock for one of my
1431  * ACQUIRE messages.
1432  */
1433 static void
1434 ah_keysock_no_socket(mblk_t *mp, ipsecah_stack_t *ahstack)
1435 {
1436 	sadb_msg_t *samsg;
1437 	keysock_out_err_t *kse = (keysock_out_err_t *)mp->b_rptr;
1438 
1439 	if (mp->b_cont == NULL) {
1440 		freemsg(mp);
1441 		return;
1442 	}
1443 	samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
1444 
1445 	/*
1446 	 * If keysock can't find any registered, delete the acquire record
1447 	 * immediately, and handle errors.
1448 	 */
1449 	if (samsg->sadb_msg_type == SADB_ACQUIRE) {
1450 		samsg->sadb_msg_errno = kse->ks_err_errno;
1451 		samsg->sadb_msg_len = SADB_8TO64(sizeof (*samsg));
1452 		/*
1453 		 * Use the write-side of the ah_pfkey_q
1454 		 */
1455 		sadb_in_acquire(samsg, &ahstack->ah_sadb,
1456 		    WR(ahstack->ah_pfkey_q), ahstack->ipsecah_netstack);
1457 	}
1458 
1459 	freemsg(mp);
1460 }
1461 
1462 /*
1463  * AH module write put routine.
1464  */
1465 static void
1466 ipsecah_wput(queue_t *q, mblk_t *mp)
1467 {
1468 	ipsec_info_t *ii;
1469 	struct iocblk *iocp;
1470 	ipsecah_stack_t	*ahstack = (ipsecah_stack_t *)q->q_ptr;
1471 
1472 	ah3dbg(ahstack, ("In ah_wput().\n"));
1473 
1474 	/* NOTE:  Each case must take care of freeing or passing mp. */
1475 	switch (mp->b_datap->db_type) {
1476 	case M_CTL:
1477 		if ((mp->b_wptr - mp->b_rptr) < sizeof (ipsec_info_t)) {
1478 			/* Not big enough message. */
1479 			freemsg(mp);
1480 			break;
1481 		}
1482 		ii = (ipsec_info_t *)mp->b_rptr;
1483 
1484 		switch (ii->ipsec_info_type) {
1485 		case KEYSOCK_OUT_ERR:
1486 			ah1dbg(ahstack, ("Got KEYSOCK_OUT_ERR message.\n"));
1487 			ah_keysock_no_socket(mp, ahstack);
1488 			break;
1489 		case KEYSOCK_IN:
1490 			AH_BUMP_STAT(ahstack, keysock_in);
1491 			ah3dbg(ahstack, ("Got KEYSOCK_IN message.\n"));
1492 
1493 			/* Parse the message. */
1494 			ah_parse_pfkey(mp, ahstack);
1495 			break;
1496 		case KEYSOCK_HELLO:
1497 			sadb_keysock_hello(&ahstack->ah_pfkey_q, q, mp,
1498 			    ah_ager, (void *)ahstack, &ahstack->ah_event,
1499 			    SADB_SATYPE_AH);
1500 			break;
1501 		default:
1502 			ah1dbg(ahstack, ("Got M_CTL from above of 0x%x.\n",
1503 			    ii->ipsec_info_type));
1504 			freemsg(mp);
1505 			break;
1506 		}
1507 		break;
1508 	case M_IOCTL:
1509 		iocp = (struct iocblk *)mp->b_rptr;
1510 		switch (iocp->ioc_cmd) {
1511 		case ND_SET:
1512 		case ND_GET:
1513 			if (nd_getset(q, ahstack->ipsecah_g_nd, mp)) {
1514 				qreply(q, mp);
1515 				return;
1516 			} else {
1517 				iocp->ioc_error = ENOENT;
1518 			}
1519 			/* FALLTHRU */
1520 		default:
1521 			/* We really don't support any other ioctls, do we? */
1522 
1523 			/* Return EINVAL */
1524 			if (iocp->ioc_error != ENOENT)
1525 				iocp->ioc_error = EINVAL;
1526 			iocp->ioc_count = 0;
1527 			mp->b_datap->db_type = M_IOCACK;
1528 			qreply(q, mp);
1529 			return;
1530 		}
1531 	default:
1532 		ah3dbg(ahstack,
1533 		    ("Got default message, type %d, passing to IP.\n",
1534 		    mp->b_datap->db_type));
1535 		putnext(q, mp);
1536 	}
1537 }
1538 
1539 /* Refactor me */
1540 /*
1541  * Updating use times can be tricky business if the ipsa_haspeer flag is
1542  * set.  This function is called once in an SA's lifetime.
1543  *
1544  * Caller has to REFRELE "assoc" which is passed in.  This function has
1545  * to REFRELE any peer SA that is obtained.
1546  */
1547 static void
1548 ah_set_usetime(ipsa_t *assoc, boolean_t inbound)
1549 {
1550 	ipsa_t *inassoc, *outassoc;
1551 	isaf_t *bucket;
1552 	sadb_t *sp;
1553 	int outhash;
1554 	boolean_t isv6;
1555 	netstack_t	*ns = assoc->ipsa_netstack;
1556 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
1557 
1558 	/* No peer?  No problem! */
1559 	if (!assoc->ipsa_haspeer) {
1560 		sadb_set_usetime(assoc);
1561 		return;
1562 	}
1563 
1564 	/*
1565 	 * Otherwise, we want to grab both the original assoc and its peer.
1566 	 * There might be a race for this, but if it's a real race, the times
1567 	 * will be out-of-synch by at most a second, and since our time
1568 	 * granularity is a second, this won't be a problem.
1569 	 *
1570 	 * If we need tight synchronization on the peer SA, then we need to
1571 	 * reconsider.
1572 	 */
1573 
1574 	/* Use address family to select IPv6/IPv4 */
1575 	isv6 = (assoc->ipsa_addrfam == AF_INET6);
1576 	if (isv6) {
1577 		sp = &ahstack->ah_sadb.s_v6;
1578 	} else {
1579 		sp = &ahstack->ah_sadb.s_v4;
1580 		ASSERT(assoc->ipsa_addrfam == AF_INET);
1581 	}
1582 	if (inbound) {
1583 		inassoc = assoc;
1584 		if (isv6)
1585 			outhash = OUTBOUND_HASH_V6(sp,
1586 			    *((in6_addr_t *)&inassoc->ipsa_dstaddr));
1587 		else
1588 			outhash = OUTBOUND_HASH_V4(sp,
1589 			    *((ipaddr_t *)&inassoc->ipsa_dstaddr));
1590 		bucket = &sp->sdb_of[outhash];
1591 
1592 		mutex_enter(&bucket->isaf_lock);
1593 		outassoc = ipsec_getassocbyspi(bucket, inassoc->ipsa_spi,
1594 		    inassoc->ipsa_srcaddr, inassoc->ipsa_dstaddr,
1595 		    inassoc->ipsa_addrfam);
1596 		mutex_exit(&bucket->isaf_lock);
1597 		if (outassoc == NULL) {
1598 			/* Q: Do we wish to set haspeer == B_FALSE? */
1599 			ah0dbg(("ah_set_usetime: "
1600 			    "can't find peer for inbound.\n"));
1601 			sadb_set_usetime(inassoc);
1602 			return;
1603 		}
1604 	} else {
1605 		outassoc = assoc;
1606 		bucket = INBOUND_BUCKET(sp, outassoc->ipsa_spi);
1607 		mutex_enter(&bucket->isaf_lock);
1608 		inassoc = ipsec_getassocbyspi(bucket, outassoc->ipsa_spi,
1609 		    outassoc->ipsa_srcaddr, outassoc->ipsa_dstaddr,
1610 		    outassoc->ipsa_addrfam);
1611 		mutex_exit(&bucket->isaf_lock);
1612 		if (inassoc == NULL) {
1613 			/* Q: Do we wish to set haspeer == B_FALSE? */
1614 			ah0dbg(("ah_set_usetime: "
1615 			    "can't find peer for outbound.\n"));
1616 			sadb_set_usetime(outassoc);
1617 			return;
1618 		}
1619 	}
1620 
1621 	/* Update usetime on both. */
1622 	sadb_set_usetime(inassoc);
1623 	sadb_set_usetime(outassoc);
1624 
1625 	/*
1626 	 * REFRELE any peer SA.
1627 	 *
1628 	 * Because of the multi-line macro nature of IPSA_REFRELE, keep
1629 	 * them in { }.
1630 	 */
1631 	if (inbound) {
1632 		IPSA_REFRELE(outassoc);
1633 	} else {
1634 		IPSA_REFRELE(inassoc);
1635 	}
1636 }
1637 
1638 /* Refactor me */
1639 /*
1640  * Add a number of bytes to what the SA has protected so far.  Return
1641  * B_TRUE if the SA can still protect that many bytes.
1642  *
1643  * Caller must REFRELE the passed-in assoc.  This function must REFRELE
1644  * any obtained peer SA.
1645  */
1646 static boolean_t
1647 ah_age_bytes(ipsa_t *assoc, uint64_t bytes, boolean_t inbound)
1648 {
1649 	ipsa_t *inassoc, *outassoc;
1650 	isaf_t *bucket;
1651 	boolean_t inrc, outrc, isv6;
1652 	sadb_t *sp;
1653 	int outhash;
1654 	netstack_t	*ns = assoc->ipsa_netstack;
1655 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
1656 
1657 	/* No peer?  No problem! */
1658 	if (!assoc->ipsa_haspeer) {
1659 		return (sadb_age_bytes(ahstack->ah_pfkey_q, assoc, bytes,
1660 		    B_TRUE));
1661 	}
1662 
1663 	/*
1664 	 * Otherwise, we want to grab both the original assoc and its peer.
1665 	 * There might be a race for this, but if it's a real race, two
1666 	 * expire messages may occur.  We limit this by only sending the
1667 	 * expire message on one of the peers, we'll pick the inbound
1668 	 * arbitrarily.
1669 	 *
1670 	 * If we need tight synchronization on the peer SA, then we need to
1671 	 * reconsider.
1672 	 */
1673 
1674 	/* Pick v4/v6 bucket based on addrfam. */
1675 	isv6 = (assoc->ipsa_addrfam == AF_INET6);
1676 	if (isv6) {
1677 		sp = &ahstack->ah_sadb.s_v6;
1678 	} else {
1679 		sp = &ahstack->ah_sadb.s_v4;
1680 		ASSERT(assoc->ipsa_addrfam == AF_INET);
1681 	}
1682 	if (inbound) {
1683 		inassoc = assoc;
1684 		if (isv6)
1685 			outhash = OUTBOUND_HASH_V6(sp,
1686 			    *((in6_addr_t *)&inassoc->ipsa_dstaddr));
1687 		else
1688 			outhash = OUTBOUND_HASH_V4(sp,
1689 			    *((ipaddr_t *)&inassoc->ipsa_dstaddr));
1690 		bucket = &sp->sdb_of[outhash];
1691 		mutex_enter(&bucket->isaf_lock);
1692 		outassoc = ipsec_getassocbyspi(bucket, inassoc->ipsa_spi,
1693 		    inassoc->ipsa_srcaddr, inassoc->ipsa_dstaddr,
1694 		    inassoc->ipsa_addrfam);
1695 		mutex_exit(&bucket->isaf_lock);
1696 		if (outassoc == NULL) {
1697 			/* Q: Do we wish to set haspeer == B_FALSE? */
1698 			ah0dbg(("ah_age_bytes: "
1699 			    "can't find peer for inbound.\n"));
1700 			return (sadb_age_bytes(ahstack->ah_pfkey_q, inassoc,
1701 			    bytes, B_TRUE));
1702 		}
1703 	} else {
1704 		outassoc = assoc;
1705 		bucket = INBOUND_BUCKET(sp, outassoc->ipsa_spi);
1706 		mutex_enter(&bucket->isaf_lock);
1707 		inassoc = ipsec_getassocbyspi(bucket, outassoc->ipsa_spi,
1708 		    outassoc->ipsa_srcaddr, outassoc->ipsa_dstaddr,
1709 		    outassoc->ipsa_addrfam);
1710 		mutex_exit(&bucket->isaf_lock);
1711 		if (inassoc == NULL) {
1712 			/* Q: Do we wish to set haspeer == B_FALSE? */
1713 			ah0dbg(("ah_age_bytes: "
1714 			    "can't find peer for outbound.\n"));
1715 			return (sadb_age_bytes(ahstack->ah_pfkey_q, outassoc,
1716 			    bytes, B_TRUE));
1717 		}
1718 	}
1719 
1720 	inrc = sadb_age_bytes(ahstack->ah_pfkey_q, inassoc, bytes, B_TRUE);
1721 	outrc = sadb_age_bytes(ahstack->ah_pfkey_q, outassoc, bytes, B_FALSE);
1722 
1723 	/*
1724 	 * REFRELE any peer SA.
1725 	 *
1726 	 * Because of the multi-line macro nature of IPSA_REFRELE, keep
1727 	 * them in { }.
1728 	 */
1729 	if (inbound) {
1730 		IPSA_REFRELE(outassoc);
1731 	} else {
1732 		IPSA_REFRELE(inassoc);
1733 	}
1734 
1735 	return (inrc && outrc);
1736 }
1737 
1738 /*
1739  * Perform the really difficult work of inserting the proposed situation.
1740  * Called while holding the algorithm lock.
1741  */
1742 static void
1743 ah_insert_prop(sadb_prop_t *prop, ipsacq_t *acqrec, uint_t combs,
1744     netstack_t *ns)
1745 {
1746 	sadb_comb_t *comb = (sadb_comb_t *)(prop + 1);
1747 	ipsec_action_t *ap;
1748 	ipsec_prot_t *prot;
1749 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
1750 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
1751 
1752 	ASSERT(MUTEX_HELD(&ipss->ipsec_alg_lock));
1753 
1754 	prop->sadb_prop_exttype = SADB_EXT_PROPOSAL;
1755 	prop->sadb_prop_len = SADB_8TO64(sizeof (sadb_prop_t));
1756 	*(uint32_t *)(&prop->sadb_prop_replay) = 0;	/* Quick zero-out! */
1757 
1758 	prop->sadb_prop_replay = ahstack->ipsecah_replay_size;
1759 
1760 	/*
1761 	 * Based upon algorithm properties, and what-not, prioritize a
1762 	 * proposal, based on the ordering of the AH algorithms in the
1763 	 * alternatives in the policy rule or socket that was placed
1764 	 * in the acquire record.
1765 	 */
1766 
1767 	for (ap = acqrec->ipsacq_act; ap != NULL;
1768 	    ap = ap->ipa_next) {
1769 		ipsec_alginfo_t *aalg;
1770 
1771 		if ((ap->ipa_act.ipa_type != IPSEC_POLICY_APPLY) ||
1772 		    (!ap->ipa_act.ipa_apply.ipp_use_ah))
1773 			continue;
1774 
1775 		prot = &ap->ipa_act.ipa_apply;
1776 
1777 		ASSERT(prot->ipp_auth_alg > 0);
1778 
1779 		aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
1780 		    [prot->ipp_auth_alg];
1781 		if (aalg == NULL || !ALG_VALID(aalg))
1782 			continue;
1783 
1784 		/* XXX check aalg for duplicates??.. */
1785 
1786 		comb->sadb_comb_flags = 0;
1787 		comb->sadb_comb_reserved = 0;
1788 		comb->sadb_comb_encrypt = 0;
1789 		comb->sadb_comb_encrypt_minbits = 0;
1790 		comb->sadb_comb_encrypt_maxbits = 0;
1791 
1792 		comb->sadb_comb_auth = aalg->alg_id;
1793 		comb->sadb_comb_auth_minbits =
1794 		    MAX(prot->ipp_ah_minbits, aalg->alg_ef_minbits);
1795 		comb->sadb_comb_auth_maxbits =
1796 		    MIN(prot->ipp_ah_maxbits, aalg->alg_ef_maxbits);
1797 
1798 		/*
1799 		 * The following may be based on algorithm
1800 		 * properties, but in the meantime, we just pick
1801 		 * some good, sensible numbers.  Key mgmt. can
1802 		 * (and perhaps should) be the place to finalize
1803 		 * such decisions.
1804 		 */
1805 
1806 		/*
1807 		 * No limits on allocations, since we really don't
1808 		 * support that concept currently.
1809 		 */
1810 		comb->sadb_comb_soft_allocations = 0;
1811 		comb->sadb_comb_hard_allocations = 0;
1812 
1813 		/*
1814 		 * These may want to come from policy rule..
1815 		 */
1816 		comb->sadb_comb_soft_bytes =
1817 		    ahstack->ipsecah_default_soft_bytes;
1818 		comb->sadb_comb_hard_bytes =
1819 		    ahstack->ipsecah_default_hard_bytes;
1820 		comb->sadb_comb_soft_addtime =
1821 		    ahstack->ipsecah_default_soft_addtime;
1822 		comb->sadb_comb_hard_addtime =
1823 		    ahstack->ipsecah_default_hard_addtime;
1824 		comb->sadb_comb_soft_usetime =
1825 		    ahstack->ipsecah_default_soft_usetime;
1826 		comb->sadb_comb_hard_usetime =
1827 		    ahstack->ipsecah_default_hard_usetime;
1828 
1829 		prop->sadb_prop_len += SADB_8TO64(sizeof (*comb));
1830 		if (--combs == 0)
1831 			return;	/* out of space.. */
1832 		comb++;
1833 	}
1834 }
1835 
1836 /*
1837  * Prepare and actually send the SADB_ACQUIRE message to PF_KEY.
1838  */
1839 static void
1840 ah_send_acquire(ipsacq_t *acqrec, mblk_t *extended, netstack_t *ns)
1841 {
1842 	uint_t combs;
1843 	sadb_msg_t *samsg;
1844 	sadb_prop_t *prop;
1845 	mblk_t *pfkeymp, *msgmp;
1846 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
1847 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
1848 
1849 	AH_BUMP_STAT(ahstack, acquire_requests);
1850 
1851 	if (ahstack->ah_pfkey_q == NULL) {
1852 		mutex_exit(&acqrec->ipsacq_lock);
1853 		return;
1854 	}
1855 
1856 	/* Set up ACQUIRE. */
1857 	pfkeymp = sadb_setup_acquire(acqrec, SADB_SATYPE_AH,
1858 	    ns->netstack_ipsec);
1859 	if (pfkeymp == NULL) {
1860 		ah0dbg(("sadb_setup_acquire failed.\n"));
1861 		mutex_exit(&acqrec->ipsacq_lock);
1862 		return;
1863 	}
1864 	ASSERT(MUTEX_HELD(&ipss->ipsec_alg_lock));
1865 	combs = ipss->ipsec_nalgs[IPSEC_ALG_AUTH];
1866 	msgmp = pfkeymp->b_cont;
1867 	samsg = (sadb_msg_t *)(msgmp->b_rptr);
1868 
1869 	/* Insert proposal here. */
1870 
1871 	prop = (sadb_prop_t *)(((uint64_t *)samsg) + samsg->sadb_msg_len);
1872 	ah_insert_prop(prop, acqrec, combs, ns);
1873 	samsg->sadb_msg_len += prop->sadb_prop_len;
1874 	msgmp->b_wptr += SADB_64TO8(samsg->sadb_msg_len);
1875 
1876 	mutex_exit(&ipss->ipsec_alg_lock);
1877 
1878 	/*
1879 	 * Must mutex_exit() before sending PF_KEY message up, in
1880 	 * order to avoid recursive mutex_enter() if there are no registered
1881 	 * listeners.
1882 	 *
1883 	 * Once I've sent the message, I'm cool anyway.
1884 	 */
1885 	mutex_exit(&acqrec->ipsacq_lock);
1886 	if (extended != NULL) {
1887 		putnext(ahstack->ah_pfkey_q, extended);
1888 	}
1889 	putnext(ahstack->ah_pfkey_q, pfkeymp);
1890 }
1891 
1892 /* Refactor me */
1893 /*
1894  * Handle the SADB_GETSPI message.  Create a larval SA.
1895  */
1896 static void
1897 ah_getspi(mblk_t *mp, keysock_in_t *ksi, ipsecah_stack_t *ahstack)
1898 {
1899 	ipsa_t *newbie, *target;
1900 	isaf_t *outbound, *inbound;
1901 	int rc, diagnostic;
1902 	sadb_sa_t *assoc;
1903 	keysock_out_t *kso;
1904 	uint32_t newspi;
1905 
1906 	/*
1907 	 * Randomly generate a proposed SPI value.
1908 	 */
1909 	if (cl_inet_getspi != NULL) {
1910 		cl_inet_getspi(ahstack->ipsecah_netstack->netstack_stackid,
1911 		    IPPROTO_AH, (uint8_t *)&newspi, sizeof (uint32_t), NULL);
1912 	} else {
1913 		(void) random_get_pseudo_bytes((uint8_t *)&newspi,
1914 		    sizeof (uint32_t));
1915 	}
1916 	newbie = sadb_getspi(ksi, newspi, &diagnostic,
1917 	    ahstack->ipsecah_netstack, IPPROTO_AH);
1918 
1919 	if (newbie == NULL) {
1920 		sadb_pfkey_error(ahstack->ah_pfkey_q, mp, ENOMEM, diagnostic,
1921 		    ksi->ks_in_serial);
1922 		return;
1923 	} else if (newbie == (ipsa_t *)-1) {
1924 		sadb_pfkey_error(ahstack->ah_pfkey_q, mp, EINVAL, diagnostic,
1925 		    ksi->ks_in_serial);
1926 		return;
1927 	}
1928 
1929 	/*
1930 	 * XXX - We may randomly collide.  We really should recover from this.
1931 	 *	 Unfortunately, that could require spending way-too-much-time
1932 	 *	 in here.  For now, let the user retry.
1933 	 */
1934 
1935 	if (newbie->ipsa_addrfam == AF_INET6) {
1936 		outbound = OUTBOUND_BUCKET_V6(&ahstack->ah_sadb.s_v6,
1937 		    *(uint32_t *)(newbie->ipsa_dstaddr));
1938 		inbound = INBOUND_BUCKET(&ahstack->ah_sadb.s_v6,
1939 		    newbie->ipsa_spi);
1940 	} else {
1941 		outbound = OUTBOUND_BUCKET_V4(&ahstack->ah_sadb.s_v4,
1942 		    *(uint32_t *)(newbie->ipsa_dstaddr));
1943 		inbound = INBOUND_BUCKET(&ahstack->ah_sadb.s_v4,
1944 		    newbie->ipsa_spi);
1945 	}
1946 
1947 	mutex_enter(&outbound->isaf_lock);
1948 	mutex_enter(&inbound->isaf_lock);
1949 
1950 	/*
1951 	 * Check for collisions (i.e. did sadb_getspi() return with something
1952 	 * that already exists?).
1953 	 *
1954 	 * Try outbound first.  Even though SADB_GETSPI is traditionally
1955 	 * for inbound SAs, you never know what a user might do.
1956 	 */
1957 	target = ipsec_getassocbyspi(outbound, newbie->ipsa_spi,
1958 	    newbie->ipsa_srcaddr, newbie->ipsa_dstaddr, newbie->ipsa_addrfam);
1959 	if (target == NULL) {
1960 		target = ipsec_getassocbyspi(inbound, newbie->ipsa_spi,
1961 		    newbie->ipsa_srcaddr, newbie->ipsa_dstaddr,
1962 		    newbie->ipsa_addrfam);
1963 	}
1964 
1965 	/*
1966 	 * I don't have collisions elsewhere!
1967 	 * (Nor will I because I'm still holding inbound/outbound locks.)
1968 	 */
1969 
1970 	if (target != NULL) {
1971 		rc = EEXIST;
1972 		IPSA_REFRELE(target);
1973 	} else {
1974 		/*
1975 		 * sadb_insertassoc() also checks for collisions, so
1976 		 * if there's a colliding larval entry, rc will be set
1977 		 * to EEXIST.
1978 		 */
1979 		rc = sadb_insertassoc(newbie, inbound);
1980 		newbie->ipsa_hardexpiretime = gethrestime_sec();
1981 		newbie->ipsa_hardexpiretime += ahstack->ipsecah_larval_timeout;
1982 	}
1983 
1984 	/*
1985 	 * Can exit outbound mutex.  Hold inbound until we're done with
1986 	 * newbie.
1987 	 */
1988 	mutex_exit(&outbound->isaf_lock);
1989 
1990 	if (rc != 0) {
1991 		mutex_exit(&inbound->isaf_lock);
1992 		IPSA_REFRELE(newbie);
1993 		sadb_pfkey_error(ahstack->ah_pfkey_q, mp, rc,
1994 		    SADB_X_DIAGNOSTIC_NONE, ksi->ks_in_serial);
1995 		return;
1996 	}
1997 
1998 	/* Can write here because I'm still holding the bucket lock. */
1999 	newbie->ipsa_type = SADB_SATYPE_AH;
2000 
2001 	/*
2002 	 * Construct successful return message.  We have one thing going
2003 	 * for us in PF_KEY v2.  That's the fact that
2004 	 *	sizeof (sadb_spirange_t) == sizeof (sadb_sa_t)
2005 	 */
2006 	assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SPIRANGE];
2007 	assoc->sadb_sa_exttype = SADB_EXT_SA;
2008 	assoc->sadb_sa_spi = newbie->ipsa_spi;
2009 	*((uint64_t *)(&assoc->sadb_sa_replay)) = 0;
2010 	mutex_exit(&inbound->isaf_lock);
2011 
2012 	/* Convert KEYSOCK_IN to KEYSOCK_OUT. */
2013 	kso = (keysock_out_t *)ksi;
2014 	kso->ks_out_len = sizeof (*kso);
2015 	kso->ks_out_serial = ksi->ks_in_serial;
2016 	kso->ks_out_type = KEYSOCK_OUT;
2017 
2018 	/*
2019 	 * Can safely putnext() to ah_pfkey_q, because this is a turnaround
2020 	 * from the ah_pfkey_q.
2021 	 */
2022 	putnext(ahstack->ah_pfkey_q, mp);
2023 }
2024 
2025 /*
2026  * IPv6 sends up the ICMP errors for validation and the removal of the AH
2027  * header.
2028  * If succesful, the mp has been modified to not include the AH header so
2029  * that the caller can fanout to the ULP's icmp error handler.
2030  */
2031 static mblk_t *
2032 ah_icmp_error_v6(mblk_t *mp, ip_recv_attr_t *ira, ipsecah_stack_t *ahstack)
2033 {
2034 	ip6_t *ip6h, *oip6h;
2035 	uint16_t hdr_length, ah_length;
2036 	uint8_t *nexthdrp;
2037 	ah_t *ah;
2038 	icmp6_t *icmp6;
2039 	isaf_t *isaf;
2040 	ipsa_t *assoc;
2041 	uint8_t *post_ah_ptr;
2042 	ipsec_stack_t	*ipss = ahstack->ipsecah_netstack->netstack_ipsec;
2043 
2044 	/*
2045 	 * Eat the cost of a pullupmsg() for now.  It makes the rest of this
2046 	 * code far less convoluted.
2047 	 */
2048 	if (!pullupmsg(mp, -1) ||
2049 	    !ip_hdr_length_nexthdr_v6(mp, (ip6_t *)mp->b_rptr, &hdr_length,
2050 	    &nexthdrp) ||
2051 	    mp->b_rptr + hdr_length + sizeof (icmp6_t) + sizeof (ip6_t) +
2052 	    sizeof (ah_t) > mp->b_wptr) {
2053 		IP_AH_BUMP_STAT(ipss, in_discards);
2054 		ip_drop_packet(mp, B_TRUE, ira->ira_ill,
2055 		    DROPPER(ipss, ipds_ah_nomem),
2056 		    &ahstack->ah_dropper);
2057 		return (NULL);
2058 	}
2059 
2060 	oip6h = (ip6_t *)mp->b_rptr;
2061 	icmp6 = (icmp6_t *)((uint8_t *)oip6h + hdr_length);
2062 	ip6h = (ip6_t *)(icmp6 + 1);
2063 	if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &hdr_length, &nexthdrp)) {
2064 		IP_AH_BUMP_STAT(ipss, in_discards);
2065 		ip_drop_packet(mp, B_TRUE, ira->ira_ill,
2066 		    DROPPER(ipss, ipds_ah_bad_v6_hdrs),
2067 		    &ahstack->ah_dropper);
2068 		return (NULL);
2069 	}
2070 	ah = (ah_t *)((uint8_t *)ip6h + hdr_length);
2071 
2072 	isaf = OUTBOUND_BUCKET_V6(&ahstack->ah_sadb.s_v6, ip6h->ip6_dst);
2073 	mutex_enter(&isaf->isaf_lock);
2074 	assoc = ipsec_getassocbyspi(isaf, ah->ah_spi,
2075 	    (uint32_t *)&ip6h->ip6_src, (uint32_t *)&ip6h->ip6_dst, AF_INET6);
2076 	mutex_exit(&isaf->isaf_lock);
2077 
2078 	if (assoc == NULL) {
2079 		IP_AH_BUMP_STAT(ipss, lookup_failure);
2080 		IP_AH_BUMP_STAT(ipss, in_discards);
2081 		if (ahstack->ipsecah_log_unknown_spi) {
2082 			ipsec_assocfailure(info.mi_idnum, 0, 0,
2083 			    SL_CONSOLE | SL_WARN | SL_ERROR,
2084 			    "Bad ICMP message - No association for the "
2085 			    "attached AH header whose spi is 0x%x, "
2086 			    "sender is 0x%x\n",
2087 			    ah->ah_spi, &oip6h->ip6_src, AF_INET6,
2088 			    ahstack->ipsecah_netstack);
2089 		}
2090 		ip_drop_packet(mp, B_TRUE, ira->ira_ill,
2091 		    DROPPER(ipss, ipds_ah_no_sa),
2092 		    &ahstack->ah_dropper);
2093 		return (NULL);
2094 	}
2095 
2096 	IPSA_REFRELE(assoc);
2097 
2098 	/*
2099 	 * There seems to be a valid association. If there is enough of AH
2100 	 * header remove it, otherwise bail.  One could check whether it has
2101 	 * complete AH header plus 8 bytes but it does not make sense if an
2102 	 * icmp error is returned for ICMP messages e.g ICMP time exceeded,
2103 	 * that are being sent up. Let the caller figure out.
2104 	 *
2105 	 * NOTE: ah_length is the number of 32 bit words minus 2.
2106 	 */
2107 	ah_length = (ah->ah_length << 2) + 8;
2108 	post_ah_ptr = (uint8_t *)ah + ah_length;
2109 
2110 	if (post_ah_ptr > mp->b_wptr) {
2111 		IP_AH_BUMP_STAT(ipss, in_discards);
2112 		ip_drop_packet(mp, B_TRUE, ira->ira_ill,
2113 		    DROPPER(ipss, ipds_ah_bad_length),
2114 		    &ahstack->ah_dropper);
2115 		return (NULL);
2116 	}
2117 
2118 	ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) - ah_length);
2119 	*nexthdrp = ah->ah_nexthdr;
2120 	ovbcopy(post_ah_ptr, ah,
2121 	    (size_t)((uintptr_t)mp->b_wptr - (uintptr_t)post_ah_ptr));
2122 	mp->b_wptr -= ah_length;
2123 
2124 	return (mp);
2125 }
2126 
2127 /*
2128  * IP sends up the ICMP errors for validation and the removal of
2129  * the AH header.
2130  * If succesful, the mp has been modified to not include the AH header so
2131  * that the caller can fanout to the ULP's icmp error handler.
2132  */
2133 static mblk_t *
2134 ah_icmp_error_v4(mblk_t *mp, ip_recv_attr_t *ira, ipsecah_stack_t *ahstack)
2135 {
2136 	mblk_t *mp1;
2137 	icmph_t *icmph;
2138 	int iph_hdr_length;
2139 	int hdr_length;
2140 	isaf_t *hptr;
2141 	ipsa_t *assoc;
2142 	int ah_length;
2143 	ipha_t *ipha;
2144 	ipha_t *oipha;
2145 	ah_t *ah;
2146 	uint32_t length;
2147 	int alloc_size;
2148 	uint8_t nexthdr;
2149 	ipsec_stack_t	*ipss = ahstack->ipsecah_netstack->netstack_ipsec;
2150 
2151 	oipha = ipha = (ipha_t *)mp->b_rptr;
2152 	iph_hdr_length = IPH_HDR_LENGTH(ipha);
2153 	icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2154 
2155 	ipha = (ipha_t *)&icmph[1];
2156 	hdr_length = IPH_HDR_LENGTH(ipha);
2157 
2158 	/*
2159 	 * See if we have enough to locate the SPI
2160 	 */
2161 	if ((uchar_t *)ipha + hdr_length + 8 > mp->b_wptr) {
2162 		if (!pullupmsg(mp, (uchar_t *)ipha + hdr_length + 8 -
2163 		    mp->b_rptr)) {
2164 			ipsec_rl_strlog(ahstack->ipsecah_netstack,
2165 			    info.mi_idnum, 0, 0,
2166 			    SL_WARN | SL_ERROR,
2167 			    "ICMP error: Small AH header\n");
2168 			IP_AH_BUMP_STAT(ipss, in_discards);
2169 			ip_drop_packet(mp, B_TRUE, ira->ira_ill,
2170 			    DROPPER(ipss, ipds_ah_bad_length),
2171 			    &ahstack->ah_dropper);
2172 			return (NULL);
2173 		}
2174 		icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2175 		ipha = (ipha_t *)&icmph[1];
2176 	}
2177 
2178 	ah = (ah_t *)((uint8_t *)ipha + hdr_length);
2179 	nexthdr = ah->ah_nexthdr;
2180 
2181 	hptr = OUTBOUND_BUCKET_V4(&ahstack->ah_sadb.s_v4, ipha->ipha_dst);
2182 	mutex_enter(&hptr->isaf_lock);
2183 	assoc = ipsec_getassocbyspi(hptr, ah->ah_spi,
2184 	    (uint32_t *)&ipha->ipha_src, (uint32_t *)&ipha->ipha_dst, AF_INET);
2185 	mutex_exit(&hptr->isaf_lock);
2186 
2187 	if (assoc == NULL) {
2188 		IP_AH_BUMP_STAT(ipss, lookup_failure);
2189 		IP_AH_BUMP_STAT(ipss, in_discards);
2190 		if (ahstack->ipsecah_log_unknown_spi) {
2191 			ipsec_assocfailure(info.mi_idnum, 0, 0,
2192 			    SL_CONSOLE | SL_WARN | SL_ERROR,
2193 			    "Bad ICMP message - No association for the "
2194 			    "attached AH header whose spi is 0x%x, "
2195 			    "sender is 0x%x\n",
2196 			    ah->ah_spi, &oipha->ipha_src, AF_INET,
2197 			    ahstack->ipsecah_netstack);
2198 		}
2199 		ip_drop_packet(mp, B_TRUE, ira->ira_ill,
2200 		    DROPPER(ipss, ipds_ah_no_sa),
2201 		    &ahstack->ah_dropper);
2202 		return (NULL);
2203 	}
2204 
2205 	IPSA_REFRELE(assoc);
2206 	/*
2207 	 * There seems to be a valid association. If there
2208 	 * is enough of AH header remove it, otherwise remove
2209 	 * as much as possible and send it back. One could check
2210 	 * whether it has complete AH header plus 8 bytes but it
2211 	 * does not make sense if an icmp error is returned for
2212 	 * ICMP messages e.g ICMP time exceeded, that are being
2213 	 * sent up. Let the caller figure out.
2214 	 *
2215 	 * NOTE: ah_length is the number of 32 bit words minus 2.
2216 	 */
2217 	ah_length = (ah->ah_length << 2) + 8;
2218 
2219 	if ((uchar_t *)ipha + hdr_length + ah_length > mp->b_wptr) {
2220 		if (mp->b_cont == NULL) {
2221 			/*
2222 			 * There is nothing to pullup. Just remove as
2223 			 * much as possible. This is a common case for
2224 			 * IPV4.
2225 			 */
2226 			ah_length = (mp->b_wptr - ((uchar_t *)ipha +
2227 			    hdr_length));
2228 			goto done;
2229 		}
2230 		/* Pullup the full ah header */
2231 		if (!pullupmsg(mp, (uchar_t *)ah + ah_length - mp->b_rptr)) {
2232 			/*
2233 			 * pullupmsg could have failed if there was not
2234 			 * enough to pullup or memory allocation failed.
2235 			 * We tried hard, give up now.
2236 			 */
2237 			IP_AH_BUMP_STAT(ipss, in_discards);
2238 			ip_drop_packet(mp, B_TRUE, ira->ira_ill,
2239 			    DROPPER(ipss, ipds_ah_nomem),
2240 			    &ahstack->ah_dropper);
2241 			return (NULL);
2242 		}
2243 		icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2244 		ipha = (ipha_t *)&icmph[1];
2245 	}
2246 done:
2247 	/*
2248 	 * Remove the AH header and change the protocol.
2249 	 * Don't update the spi fields in the ip_recv_attr_t
2250 	 * as we are called just to validate the
2251 	 * message attached to the ICMP message.
2252 	 *
2253 	 * If we never pulled up since all of the message
2254 	 * is in one single mblk, we can't remove the AH header
2255 	 * by just setting the b_wptr to the beginning of the
2256 	 * AH header. We need to allocate a mblk that can hold
2257 	 * up until the inner IP header and copy them.
2258 	 */
2259 	alloc_size = iph_hdr_length + sizeof (icmph_t) + hdr_length;
2260 
2261 	if ((mp1 = allocb(alloc_size, BPRI_LO)) == NULL) {
2262 		IP_AH_BUMP_STAT(ipss, in_discards);
2263 		ip_drop_packet(mp, B_TRUE, ira->ira_ill,
2264 		    DROPPER(ipss, ipds_ah_nomem),
2265 		    &ahstack->ah_dropper);
2266 		return (NULL);
2267 	}
2268 	bcopy(mp->b_rptr, mp1->b_rptr, alloc_size);
2269 	mp1->b_wptr += alloc_size;
2270 
2271 	/*
2272 	 * Skip whatever we have copied and as much of AH header
2273 	 * possible. If we still have something left in the original
2274 	 * message, tag on.
2275 	 */
2276 	mp->b_rptr = (uchar_t *)ipha + hdr_length + ah_length;
2277 
2278 	if (mp->b_rptr != mp->b_wptr) {
2279 		mp1->b_cont = mp;
2280 	} else {
2281 		if (mp->b_cont != NULL)
2282 			mp1->b_cont = mp->b_cont;
2283 		freeb(mp);
2284 	}
2285 
2286 	ipha = (ipha_t *)(mp1->b_rptr + iph_hdr_length + sizeof (icmph_t));
2287 	ipha->ipha_protocol = nexthdr;
2288 	length = ntohs(ipha->ipha_length);
2289 	length -= ah_length;
2290 	ipha->ipha_length = htons((uint16_t)length);
2291 	ipha->ipha_hdr_checksum = 0;
2292 	ipha->ipha_hdr_checksum = (uint16_t)ip_csum_hdr(ipha);
2293 
2294 	return (mp1);
2295 }
2296 
2297 /*
2298  * IP calls this to validate the ICMP errors that
2299  * we got from the network.
2300  */
2301 mblk_t *
2302 ipsecah_icmp_error(mblk_t *data_mp, ip_recv_attr_t *ira)
2303 {
2304 	netstack_t	*ns = ira->ira_ill->ill_ipst->ips_netstack;
2305 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
2306 
2307 	if (ira->ira_flags & IRAF_IS_IPV4)
2308 		return (ah_icmp_error_v4(data_mp, ira, ahstack));
2309 	else
2310 		return (ah_icmp_error_v6(data_mp, ira, ahstack));
2311 }
2312 
2313 static int
2314 ah_fix_tlv_options_v6(uint8_t *oi_opt, uint8_t *pi_opt, uint_t ehdrlen,
2315     uint8_t hdr_type, boolean_t copy_always)
2316 {
2317 	uint8_t opt_type;
2318 	uint_t optlen;
2319 
2320 	ASSERT(hdr_type == IPPROTO_DSTOPTS || hdr_type == IPPROTO_HOPOPTS);
2321 
2322 	/*
2323 	 * Copy the next header and hdr ext. len of the HOP-by-HOP
2324 	 * and Destination option.
2325 	 */
2326 	*pi_opt++ = *oi_opt++;
2327 	*pi_opt++ = *oi_opt++;
2328 	ehdrlen -= 2;
2329 
2330 	/*
2331 	 * Now handle all the TLV encoded options.
2332 	 */
2333 	while (ehdrlen != 0) {
2334 		opt_type = *oi_opt;
2335 
2336 		if (opt_type == IP6OPT_PAD1) {
2337 			optlen = 1;
2338 		} else {
2339 			if (ehdrlen < 2)
2340 				goto bad_opt;
2341 			optlen = 2 + oi_opt[1];
2342 			if (optlen > ehdrlen)
2343 				goto bad_opt;
2344 		}
2345 		if (copy_always || !(opt_type & IP6OPT_MUTABLE)) {
2346 			bcopy(oi_opt, pi_opt, optlen);
2347 		} else {
2348 			if (optlen == 1) {
2349 				*pi_opt = 0;
2350 			} else {
2351 				/*
2352 				 * Copy the type and data length fields.
2353 				 * Zero the option data by skipping
2354 				 * option type and option data len
2355 				 * fields.
2356 				 */
2357 				*pi_opt = *oi_opt;
2358 				*(pi_opt + 1) = *(oi_opt + 1);
2359 				bzero(pi_opt + 2, optlen - 2);
2360 			}
2361 		}
2362 		ehdrlen -= optlen;
2363 		oi_opt += optlen;
2364 		pi_opt += optlen;
2365 	}
2366 	return (0);
2367 bad_opt:
2368 	return (-1);
2369 }
2370 
2371 /*
2372  * Construct a pseudo header for AH, processing all the options.
2373  *
2374  * oip6h is the IPv6 header of the incoming or outgoing packet.
2375  * ip6h is the pointer to the pseudo headers IPV6 header. All
2376  * the space needed for the options have been allocated including
2377  * the AH header.
2378  *
2379  * If copy_always is set, all the options that appear before AH are copied
2380  * blindly without checking for IP6OPT_MUTABLE. This is used by
2381  * ah_auth_out_done().  Please refer to that function for details.
2382  *
2383  * NOTE :
2384  *
2385  * *  AH header is never copied in this function even if copy_always
2386  *    is set. It just returns the ah_offset - offset of the AH header
2387  *    and the caller needs to do the copying. This is done so that we
2388  *    don't have pass extra arguments e.g. SA etc. and also,
2389  *    it is not needed when ah_auth_out_done is calling this function.
2390  */
2391 static uint_t
2392 ah_fix_phdr_v6(ip6_t *ip6h, ip6_t *oip6h, boolean_t outbound,
2393     boolean_t copy_always)
2394 {
2395 	uint8_t	*oi_opt;
2396 	uint8_t	*pi_opt;
2397 	uint8_t nexthdr;
2398 	uint8_t *prev_nexthdr;
2399 	ip6_hbh_t *hbhhdr;
2400 	ip6_dest_t *dsthdr = NULL;
2401 	ip6_rthdr0_t *rthdr;
2402 	int ehdrlen;
2403 	ah_t *ah;
2404 	int ret;
2405 
2406 	/*
2407 	 * In the outbound case for source route, ULP has already moved
2408 	 * the first hop, which is now in ip6_dst. We need to re-arrange
2409 	 * the header to make it look like how it would appear in the
2410 	 * receiver i.e
2411 	 *
2412 	 * Because of ip_massage_options_v6 the header looks like
2413 	 * this :
2414 	 *
2415 	 * ip6_src = S, ip6_dst = I1. followed by I2,I3,D.
2416 	 *
2417 	 * When it reaches the receiver, it would look like
2418 	 *
2419 	 * ip6_src = S, ip6_dst = D. followed by I1,I2,I3.
2420 	 *
2421 	 * NOTE : We assume that there are no problems with the options
2422 	 * as IP should have already checked this.
2423 	 */
2424 
2425 	oi_opt = (uchar_t *)&oip6h[1];
2426 	pi_opt = (uchar_t *)&ip6h[1];
2427 
2428 	/*
2429 	 * We set the prev_nexthdr properly in the pseudo header.
2430 	 * After we finish authentication and come back from the
2431 	 * algorithm module, pseudo header will become the real
2432 	 * IP header.
2433 	 */
2434 	prev_nexthdr = (uint8_t *)&ip6h->ip6_nxt;
2435 	nexthdr = oip6h->ip6_nxt;
2436 	/* Assume IP has already stripped it */
2437 	ASSERT(nexthdr != IPPROTO_FRAGMENT);
2438 	ah = NULL;
2439 	dsthdr = NULL;
2440 	for (;;) {
2441 		switch (nexthdr) {
2442 		case IPPROTO_HOPOPTS:
2443 			hbhhdr = (ip6_hbh_t *)oi_opt;
2444 			nexthdr = hbhhdr->ip6h_nxt;
2445 			ehdrlen = 8 * (hbhhdr->ip6h_len + 1);
2446 			ret = ah_fix_tlv_options_v6(oi_opt, pi_opt, ehdrlen,
2447 			    IPPROTO_HOPOPTS, copy_always);
2448 			/*
2449 			 * Return a zero offset indicating error if there
2450 			 * was error.
2451 			 */
2452 			if (ret == -1)
2453 				return (0);
2454 			hbhhdr = (ip6_hbh_t *)pi_opt;
2455 			prev_nexthdr = (uint8_t *)&hbhhdr->ip6h_nxt;
2456 			break;
2457 		case IPPROTO_ROUTING:
2458 			rthdr = (ip6_rthdr0_t *)oi_opt;
2459 			nexthdr = rthdr->ip6r0_nxt;
2460 			ehdrlen = 8 * (rthdr->ip6r0_len + 1);
2461 			if (!copy_always && outbound) {
2462 				int i, left;
2463 				ip6_rthdr0_t *prthdr;
2464 				in6_addr_t *ap, *pap;
2465 
2466 				left = rthdr->ip6r0_segleft;
2467 				prthdr = (ip6_rthdr0_t *)pi_opt;
2468 				pap = (in6_addr_t *)(prthdr + 1);
2469 				ap = (in6_addr_t *)(rthdr + 1);
2470 				/*
2471 				 * First eight bytes except seg_left
2472 				 * does not change en route.
2473 				 */
2474 				bcopy(oi_opt, pi_opt, 8);
2475 				prthdr->ip6r0_segleft = 0;
2476 				/*
2477 				 * First address has been moved to
2478 				 * the destination address of the
2479 				 * ip header by ip_massage_options_v6.
2480 				 * And the real destination address is
2481 				 * in the last address part of the
2482 				 * option.
2483 				 */
2484 				*pap = oip6h->ip6_dst;
2485 				for (i = 1; i < left - 1; i++)
2486 					pap[i] = ap[i - 1];
2487 				ip6h->ip6_dst = *(ap + left - 1);
2488 			} else {
2489 				bcopy(oi_opt, pi_opt, ehdrlen);
2490 			}
2491 			rthdr = (ip6_rthdr0_t *)pi_opt;
2492 			prev_nexthdr = (uint8_t *)&rthdr->ip6r0_nxt;
2493 			break;
2494 		case IPPROTO_DSTOPTS:
2495 			/*
2496 			 * Destination options are tricky.  If there is
2497 			 * a terminal (e.g. non-IPv6-extension) header
2498 			 * following the destination options, don't
2499 			 * reset prev_nexthdr or advance the AH insertion
2500 			 * point and just treat this as a terminal header.
2501 			 *
2502 			 * If this is an inbound packet, just deal with
2503 			 * it as is.
2504 			 */
2505 			dsthdr = (ip6_dest_t *)oi_opt;
2506 			/*
2507 			 * XXX I hope common-subexpression elimination
2508 			 * saves us the double-evaluate.
2509 			 */
2510 			if (outbound && dsthdr->ip6d_nxt != IPPROTO_ROUTING &&
2511 			    dsthdr->ip6d_nxt != IPPROTO_HOPOPTS)
2512 				goto terminal_hdr;
2513 			nexthdr = dsthdr->ip6d_nxt;
2514 			ehdrlen = 8 * (dsthdr->ip6d_len + 1);
2515 			ret = ah_fix_tlv_options_v6(oi_opt, pi_opt, ehdrlen,
2516 			    IPPROTO_DSTOPTS, copy_always);
2517 			/*
2518 			 * Return a zero offset indicating error if there
2519 			 * was error.
2520 			 */
2521 			if (ret == -1)
2522 				return (0);
2523 			break;
2524 		case IPPROTO_AH:
2525 			/*
2526 			 * Be conservative in what you send.  We shouldn't
2527 			 * see two same-scoped AH's in one packet.
2528 			 * (Inner-IP-scoped AH will be hit by terminal
2529 			 * header of IP or IPv6.)
2530 			 */
2531 			ASSERT(!outbound);
2532 			return ((uint_t)(pi_opt - (uint8_t *)ip6h));
2533 		default:
2534 			ASSERT(outbound);
2535 terminal_hdr:
2536 			*prev_nexthdr = IPPROTO_AH;
2537 			ah = (ah_t *)pi_opt;
2538 			ah->ah_nexthdr = nexthdr;
2539 			return ((uint_t)(pi_opt - (uint8_t *)ip6h));
2540 		}
2541 		pi_opt += ehdrlen;
2542 		oi_opt += ehdrlen;
2543 	}
2544 	/* NOTREACHED */
2545 }
2546 
2547 static boolean_t
2548 ah_finish_up(ah_t *phdr_ah, ah_t *inbound_ah, ipsa_t *assoc,
2549     int ah_data_sz, int ah_align_sz, ipsecah_stack_t *ahstack)
2550 {
2551 	int i;
2552 
2553 	/*
2554 	 * Padding :
2555 	 *
2556 	 * 1) Authentication data may have to be padded
2557 	 * before ICV calculation if ICV is not a multiple
2558 	 * of 64 bits. This padding is arbitrary and transmitted
2559 	 * with the packet at the end of the authentication data.
2560 	 * Payload length should include the padding bytes.
2561 	 *
2562 	 * 2) Explicit padding of the whole datagram may be
2563 	 * required by the algorithm which need not be
2564 	 * transmitted. It is assumed that this will be taken
2565 	 * care by the algorithm module.
2566 	 */
2567 	bzero(phdr_ah + 1, ah_data_sz);	/* Zero out ICV for pseudo-hdr. */
2568 
2569 	if (inbound_ah == NULL) {
2570 		/* Outbound AH datagram. */
2571 
2572 		phdr_ah->ah_length = (ah_align_sz >> 2) + 1;
2573 		phdr_ah->ah_reserved = 0;
2574 		phdr_ah->ah_spi = assoc->ipsa_spi;
2575 
2576 		phdr_ah->ah_replay =
2577 		    htonl(atomic_add_32_nv(&assoc->ipsa_replay, 1));
2578 		if (phdr_ah->ah_replay == 0 && assoc->ipsa_replay_wsize != 0) {
2579 			/*
2580 			 * XXX We have replay counter wrapping.  We probably
2581 			 * 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 AH SA (0x%x), dst %s has wrapped "
2586 			    "sequence.\n", phdr_ah->ah_spi,
2587 			    assoc->ipsa_dstaddr, assoc->ipsa_addrfam,
2588 			    ahstack->ipsecah_netstack);
2589 
2590 			sadb_replay_delete(assoc);
2591 			/* Caller will free phdr_mp and return NULL. */
2592 			return (B_FALSE);
2593 		}
2594 
2595 		if (ah_data_sz != ah_align_sz) {
2596 			uchar_t *pad = ((uchar_t *)phdr_ah + sizeof (ah_t) +
2597 			    ah_data_sz);
2598 
2599 			for (i = 0; i < (ah_align_sz - ah_data_sz); i++) {
2600 				pad[i] = (uchar_t)i;	/* Fill the padding */
2601 			}
2602 		}
2603 	} else {
2604 		/* Inbound AH datagram. */
2605 		phdr_ah->ah_nexthdr = inbound_ah->ah_nexthdr;
2606 		phdr_ah->ah_length = inbound_ah->ah_length;
2607 		phdr_ah->ah_reserved = 0;
2608 		ASSERT(inbound_ah->ah_spi == assoc->ipsa_spi);
2609 		phdr_ah->ah_spi = inbound_ah->ah_spi;
2610 		phdr_ah->ah_replay = inbound_ah->ah_replay;
2611 
2612 		if (ah_data_sz != ah_align_sz) {
2613 			uchar_t *opad = ((uchar_t *)inbound_ah +
2614 			    sizeof (ah_t) + ah_data_sz);
2615 			uchar_t *pad = ((uchar_t *)phdr_ah + sizeof (ah_t) +
2616 			    ah_data_sz);
2617 
2618 			for (i = 0; i < (ah_align_sz - ah_data_sz); i++) {
2619 				pad[i] = opad[i];	/* Copy the padding */
2620 			}
2621 		}
2622 	}
2623 
2624 	return (B_TRUE);
2625 }
2626 
2627 /*
2628  * Called upon failing the inbound ICV check. The message passed as
2629  * argument is freed.
2630  */
2631 static void
2632 ah_log_bad_auth(mblk_t *mp, ip_recv_attr_t *ira, ipsec_crypto_t *ic)
2633 {
2634 	boolean_t	isv4 = (ira->ira_flags & IRAF_IS_IPV4);
2635 	ipsa_t		*assoc = ira->ira_ipsec_ah_sa;
2636 	int		af;
2637 	void		*addr;
2638 	netstack_t	*ns = ira->ira_ill->ill_ipst->ips_netstack;
2639 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
2640 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
2641 
2642 	ASSERT(mp->b_datap->db_type == M_DATA);
2643 
2644 	mp->b_rptr -= ic->ic_skip_len;
2645 
2646 	if (isv4) {
2647 		ipha_t *ipha = (ipha_t *)mp->b_rptr;
2648 		addr = &ipha->ipha_dst;
2649 		af = AF_INET;
2650 	} else {
2651 		ip6_t *ip6h = (ip6_t *)mp->b_rptr;
2652 		addr = &ip6h->ip6_dst;
2653 		af = AF_INET6;
2654 	}
2655 
2656 	/*
2657 	 * Log the event. Don't print to the console, block
2658 	 * potential denial-of-service attack.
2659 	 */
2660 	AH_BUMP_STAT(ahstack, bad_auth);
2661 
2662 	ipsec_assocfailure(info.mi_idnum, 0, 0, SL_ERROR | SL_WARN,
2663 	    "AH Authentication failed spi %x, dst_addr %s",
2664 	    assoc->ipsa_spi, addr, af, ahstack->ipsecah_netstack);
2665 
2666 	IP_AH_BUMP_STAT(ipss, in_discards);
2667 	ip_drop_packet(mp, B_TRUE, ira->ira_ill,
2668 	    DROPPER(ipss, ipds_ah_bad_auth),
2669 	    &ahstack->ah_dropper);
2670 }
2671 
2672 /*
2673  * Kernel crypto framework callback invoked after completion of async
2674  * crypto requests for outbound packets.
2675  */
2676 static void
2677 ah_kcf_callback_outbound(void *arg, int status)
2678 {
2679 	mblk_t		*mp = (mblk_t *)arg;
2680 	mblk_t		*async_mp;
2681 	netstack_t	*ns;
2682 	ipsec_stack_t	*ipss;
2683 	ipsecah_stack_t	*ahstack;
2684 	mblk_t		*data_mp;
2685 	ip_xmit_attr_t	ixas;
2686 	ipsec_crypto_t	*ic;
2687 	ill_t		*ill;
2688 
2689 	/*
2690 	 * First remove the ipsec_crypto_t mblk
2691 	 * Note that we need to ipsec_free_crypto_data(mp) once done with ic.
2692 	 */
2693 	async_mp = ipsec_remove_crypto_data(mp, &ic);
2694 	ASSERT(async_mp != NULL);
2695 
2696 	/*
2697 	 * Extract the ip_xmit_attr_t from the first mblk.
2698 	 * Verifies that the netstack and ill is still around; could
2699 	 * have vanished while kEf was doing its work.
2700 	 * On succesful return we have a nce_t and the ill/ipst can't
2701 	 * disappear until we do the nce_refrele in ixa_cleanup.
2702 	 */
2703 	data_mp = async_mp->b_cont;
2704 	async_mp->b_cont = NULL;
2705 	if (!ip_xmit_attr_from_mblk(async_mp, &ixas)) {
2706 		/* Disappeared on us - no ill/ipst for MIB */
2707 		if (ixas.ixa_nce != NULL) {
2708 			ill = ixas.ixa_nce->nce_ill;
2709 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2710 			ip_drop_output("ipIfStatsOutDiscards", data_mp, ill);
2711 		}
2712 		freemsg(data_mp);
2713 		goto done;
2714 	}
2715 	ns = ixas.ixa_ipst->ips_netstack;
2716 	ahstack = ns->netstack_ipsecah;
2717 	ipss = ns->netstack_ipsec;
2718 	ill = ixas.ixa_nce->nce_ill;
2719 
2720 	if (status == CRYPTO_SUCCESS) {
2721 		data_mp = ah_auth_out_done(data_mp, &ixas, ic);
2722 		if (data_mp == NULL)
2723 			goto done;
2724 
2725 		(void) ip_output_post_ipsec(data_mp, &ixas);
2726 	} else {
2727 		/* Outbound shouldn't see invalid MAC */
2728 		ASSERT(status != CRYPTO_INVALID_MAC);
2729 
2730 		ah1dbg(ahstack,
2731 		    ("ah_kcf_callback_outbound: crypto failed with 0x%x\n",
2732 		    status));
2733 		AH_BUMP_STAT(ahstack, crypto_failures);
2734 		AH_BUMP_STAT(ahstack, out_discards);
2735 
2736 		ip_drop_packet(data_mp, B_FALSE, ill,
2737 		    DROPPER(ipss, ipds_ah_crypto_failed),
2738 		    &ahstack->ah_dropper);
2739 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2740 	}
2741 done:
2742 	ixa_cleanup(&ixas);
2743 	(void) ipsec_free_crypto_data(mp);
2744 }
2745 
2746 /*
2747  * Kernel crypto framework callback invoked after completion of async
2748  * crypto requests for inbound packets.
2749  */
2750 static void
2751 ah_kcf_callback_inbound(void *arg, int status)
2752 {
2753 	mblk_t		*mp = (mblk_t *)arg;
2754 	mblk_t		*async_mp;
2755 	netstack_t	*ns;
2756 	ipsec_stack_t	*ipss;
2757 	ipsecah_stack_t	*ahstack;
2758 	mblk_t		*data_mp;
2759 	ip_recv_attr_t	iras;
2760 	ipsec_crypto_t	*ic;
2761 
2762 	/*
2763 	 * First remove the ipsec_crypto_t mblk
2764 	 * Note that we need to ipsec_free_crypto_data(mp) once done with ic.
2765 	 */
2766 	async_mp = ipsec_remove_crypto_data(mp, &ic);
2767 	ASSERT(async_mp != NULL);
2768 
2769 	/*
2770 	 * Extract the ip_xmit_attr_t from the first mblk.
2771 	 * Verifies that the netstack and ill is still around; could
2772 	 * have vanished while kEf was doing its work.
2773 	 */
2774 	data_mp = async_mp->b_cont;
2775 	async_mp->b_cont = NULL;
2776 	if (!ip_recv_attr_from_mblk(async_mp, &iras)) {
2777 		/* The ill or ip_stack_t disappeared on us */
2778 		ip_drop_input("ip_recv_attr_from_mblk", data_mp, NULL);
2779 		freemsg(data_mp);
2780 		goto done;
2781 	}
2782 	ns = iras.ira_ill->ill_ipst->ips_netstack;
2783 	ahstack = ns->netstack_ipsecah;
2784 	ipss = ns->netstack_ipsec;
2785 
2786 	if (status == CRYPTO_SUCCESS) {
2787 		data_mp = ah_auth_in_done(data_mp, &iras, ic);
2788 		if (data_mp == NULL)
2789 			goto done;
2790 
2791 		/* finish IPsec processing */
2792 		ip_input_post_ipsec(data_mp, &iras);
2793 
2794 	} else if (status == CRYPTO_INVALID_MAC) {
2795 		ah_log_bad_auth(data_mp, &iras, ic);
2796 	} else {
2797 		ah1dbg(ahstack,
2798 		    ("ah_kcf_callback_inbound: crypto failed with 0x%x\n",
2799 		    status));
2800 		AH_BUMP_STAT(ahstack, crypto_failures);
2801 		IP_AH_BUMP_STAT(ipss, in_discards);
2802 		ip_drop_packet(data_mp, B_TRUE, iras.ira_ill,
2803 		    DROPPER(ipss, ipds_ah_crypto_failed),
2804 		    &ahstack->ah_dropper);
2805 		BUMP_MIB(iras.ira_ill->ill_ip_mib, ipIfStatsInDiscards);
2806 	}
2807 done:
2808 	ira_cleanup(&iras, B_TRUE);
2809 	(void) ipsec_free_crypto_data(mp);
2810 }
2811 
2812 /*
2813  * Invoked on kernel crypto failure during inbound and outbound processing.
2814  */
2815 static void
2816 ah_crypto_failed(mblk_t *data_mp, boolean_t is_inbound, int kef_rc,
2817     ill_t *ill, ipsecah_stack_t *ahstack)
2818 {
2819 	ipsec_stack_t	*ipss = ahstack->ipsecah_netstack->netstack_ipsec;
2820 
2821 	ah1dbg(ahstack, ("crypto failed for %s AH with 0x%x\n",
2822 	    is_inbound ? "inbound" : "outbound", kef_rc));
2823 	ip_drop_packet(data_mp, is_inbound, ill,
2824 	    DROPPER(ipss, ipds_ah_crypto_failed),
2825 	    &ahstack->ah_dropper);
2826 	AH_BUMP_STAT(ahstack, crypto_failures);
2827 	if (is_inbound)
2828 		IP_AH_BUMP_STAT(ipss, in_discards);
2829 	else
2830 		AH_BUMP_STAT(ahstack, out_discards);
2831 }
2832 
2833 /*
2834  * Helper macros for the ah_submit_req_{inbound,outbound}() functions.
2835  */
2836 
2837 /*
2838  * A statement-equivalent macro, _cr MUST point to a modifiable
2839  * crypto_call_req_t.
2840  */
2841 #define	AH_INIT_CALLREQ(_cr, _mp, _callback)		\
2842 	(_cr)->cr_flag = CRYPTO_SKIP_REQID|CRYPTO_ALWAYS_QUEUE;	\
2843 	(_cr)->cr_callback_arg = (_mp);				\
2844 	(_cr)->cr_callback_func = (_callback)
2845 
2846 #define	AH_INIT_CRYPTO_DATA(data, msglen, mblk) {			\
2847 	(data)->cd_format = CRYPTO_DATA_MBLK;				\
2848 	(data)->cd_mp = mblk;						\
2849 	(data)->cd_offset = 0;						\
2850 	(data)->cd_length = msglen;					\
2851 }
2852 
2853 #define	AH_INIT_CRYPTO_MAC(mac, icvlen, icvbuf) {			\
2854 	(mac)->cd_format = CRYPTO_DATA_RAW;				\
2855 	(mac)->cd_offset = 0;						\
2856 	(mac)->cd_length = icvlen;					\
2857 	(mac)->cd_raw.iov_base = icvbuf;				\
2858 	(mac)->cd_raw.iov_len = icvlen;					\
2859 }
2860 
2861 /*
2862  * Submit an inbound packet for processing by the crypto framework.
2863  */
2864 static mblk_t *
2865 ah_submit_req_inbound(mblk_t *phdr_mp, ip_recv_attr_t *ira,
2866     size_t skip_len, uint32_t ah_offset, ipsa_t *assoc)
2867 {
2868 	int kef_rc;
2869 	mblk_t *mp;
2870 	crypto_call_req_t call_req, *callrp;
2871 	uint_t icv_len = assoc->ipsa_mac_len;
2872 	crypto_ctx_template_t ctx_tmpl;
2873 	ipsecah_stack_t	*ahstack;
2874 	ipsec_crypto_t	*ic, icstack;
2875 	boolean_t force = (assoc->ipsa_flags & IPSA_F_ASYNC);
2876 
2877 	ahstack = ira->ira_ill->ill_ipst->ips_netstack->netstack_ipsecah;
2878 
2879 	ASSERT(phdr_mp != NULL);
2880 	ASSERT(phdr_mp->b_datap->db_type == M_DATA);
2881 
2882 	if (force) {
2883 		/* We are doing asynch; allocate mblks to hold state */
2884 		if ((mp = ip_recv_attr_to_mblk(ira)) == NULL ||
2885 		    (mp = ipsec_add_crypto_data(mp, &ic)) == NULL) {
2886 			BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
2887 			ip_drop_input("ipIfStatsInDiscards", phdr_mp,
2888 			    ira->ira_ill);
2889 			freemsg(phdr_mp);
2890 			return (NULL);
2891 		}
2892 
2893 		linkb(mp, phdr_mp);
2894 		callrp = &call_req;
2895 		AH_INIT_CALLREQ(callrp, mp, ah_kcf_callback_inbound);
2896 	} else {
2897 		/*
2898 		 * If we know we are going to do sync then ipsec_crypto_t
2899 		 * should be on the stack.
2900 		 */
2901 		ic = &icstack;
2902 		bzero(ic, sizeof (*ic));
2903 		callrp = NULL;
2904 	}
2905 
2906 	/* init arguments for the crypto framework */
2907 	AH_INIT_CRYPTO_DATA(&ic->ic_crypto_data, AH_MSGSIZE(phdr_mp),
2908 	    phdr_mp);
2909 
2910 	AH_INIT_CRYPTO_MAC(&ic->ic_crypto_mac, icv_len,
2911 	    (char *)phdr_mp->b_cont->b_rptr - skip_len + ah_offset +
2912 	    sizeof (ah_t));
2913 
2914 	ic->ic_skip_len = skip_len;
2915 
2916 	IPSEC_CTX_TMPL(assoc, ipsa_authtmpl, IPSEC_ALG_AUTH, ctx_tmpl);
2917 
2918 	/* call KEF to do the MAC operation */
2919 	kef_rc = crypto_mac_verify(&assoc->ipsa_amech,
2920 	    &ic->ic_crypto_data, &assoc->ipsa_kcfauthkey, ctx_tmpl,
2921 	    &ic->ic_crypto_mac, callrp);
2922 
2923 	switch (kef_rc) {
2924 	case CRYPTO_SUCCESS:
2925 		AH_BUMP_STAT(ahstack, crypto_sync);
2926 		phdr_mp = ah_auth_in_done(phdr_mp, ira, ic);
2927 		if (force) {
2928 			/* Free mp after we are done with ic */
2929 			mp = ipsec_free_crypto_data(mp);
2930 			(void) ip_recv_attr_free_mblk(mp);
2931 		}
2932 		return (phdr_mp);
2933 	case CRYPTO_QUEUED:
2934 		/* ah_kcf_callback_inbound() will be invoked on completion */
2935 		AH_BUMP_STAT(ahstack, crypto_async);
2936 		return (NULL);
2937 	case CRYPTO_INVALID_MAC:
2938 		/* Free mp after we are done with ic */
2939 		AH_BUMP_STAT(ahstack, crypto_sync);
2940 		BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
2941 		ah_log_bad_auth(phdr_mp, ira, ic);
2942 		/* phdr_mp was passed to ip_drop_packet */
2943 		if (force) {
2944 			mp = ipsec_free_crypto_data(mp);
2945 			(void) ip_recv_attr_free_mblk(mp);
2946 		}
2947 		return (NULL);
2948 	}
2949 
2950 	if (force) {
2951 		mp = ipsec_free_crypto_data(mp);
2952 		phdr_mp = ip_recv_attr_free_mblk(mp);
2953 	}
2954 	BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
2955 	ah_crypto_failed(phdr_mp, B_TRUE, kef_rc, ira->ira_ill, ahstack);
2956 	/* phdr_mp was passed to ip_drop_packet */
2957 	return (NULL);
2958 }
2959 
2960 /*
2961  * Submit an outbound packet for processing by the crypto framework.
2962  */
2963 static mblk_t *
2964 ah_submit_req_outbound(mblk_t *phdr_mp, ip_xmit_attr_t *ixa,
2965     size_t skip_len, ipsa_t *assoc)
2966 {
2967 	int kef_rc;
2968 	mblk_t *mp;
2969 	crypto_call_req_t call_req, *callrp;
2970 	uint_t icv_len = assoc->ipsa_mac_len;
2971 	ipsecah_stack_t	*ahstack;
2972 	ipsec_crypto_t	*ic, icstack;
2973 	ill_t		*ill = ixa->ixa_nce->nce_ill;
2974 	boolean_t force = (assoc->ipsa_flags & IPSA_F_ASYNC);
2975 
2976 	ahstack = ill->ill_ipst->ips_netstack->netstack_ipsecah;
2977 
2978 	ASSERT(phdr_mp != NULL);
2979 	ASSERT(phdr_mp->b_datap->db_type == M_DATA);
2980 
2981 	if (force) {
2982 		/* We are doing asynch; allocate mblks to hold state */
2983 		if ((mp = ip_xmit_attr_to_mblk(ixa)) == NULL ||
2984 		    (mp = ipsec_add_crypto_data(mp, &ic)) == NULL) {
2985 			BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2986 			ip_drop_output("ipIfStatsOutDiscards", phdr_mp, ill);
2987 			freemsg(phdr_mp);
2988 			return (NULL);
2989 		}
2990 		linkb(mp, phdr_mp);
2991 		callrp = &call_req;
2992 		AH_INIT_CALLREQ(callrp, mp, ah_kcf_callback_outbound);
2993 	} else {
2994 		/*
2995 		 * If we know we are going to do sync then ipsec_crypto_t
2996 		 * should be on the stack.
2997 		 */
2998 		ic = &icstack;
2999 		bzero(ic, sizeof (*ic));
3000 		callrp = NULL;
3001 	}
3002 
3003 	/* init arguments for the crypto framework */
3004 	AH_INIT_CRYPTO_DATA(&ic->ic_crypto_data, AH_MSGSIZE(phdr_mp),
3005 	    phdr_mp);
3006 
3007 	AH_INIT_CRYPTO_MAC(&ic->ic_crypto_mac, icv_len,
3008 	    (char *)phdr_mp->b_wptr);
3009 
3010 	ic->ic_skip_len = skip_len;
3011 
3012 	ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
3013 
3014 	/* call KEF to do the MAC operation */
3015 	kef_rc = crypto_mac(&assoc->ipsa_amech, &ic->ic_crypto_data,
3016 	    &assoc->ipsa_kcfauthkey, assoc->ipsa_authtmpl,
3017 	    &ic->ic_crypto_mac, callrp);
3018 
3019 	switch (kef_rc) {
3020 	case CRYPTO_SUCCESS:
3021 		AH_BUMP_STAT(ahstack, crypto_sync);
3022 		phdr_mp = ah_auth_out_done(phdr_mp, ixa, ic);
3023 		if (force) {
3024 			/* Free mp after we are done with ic */
3025 			mp = ipsec_free_crypto_data(mp);
3026 			(void) ip_xmit_attr_free_mblk(mp);
3027 		}
3028 		return (phdr_mp);
3029 	case CRYPTO_QUEUED:
3030 		/* ah_kcf_callback_outbound() will be invoked on completion */
3031 		AH_BUMP_STAT(ahstack, crypto_async);
3032 		return (NULL);
3033 	}
3034 
3035 	if (force) {
3036 		mp = ipsec_free_crypto_data(mp);
3037 		phdr_mp = ip_xmit_attr_free_mblk(mp);
3038 	}
3039 	BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
3040 	ah_crypto_failed(phdr_mp, B_FALSE, kef_rc, NULL, ahstack);
3041 	/* phdr_mp was passed to ip_drop_packet */
3042 	return (NULL);
3043 }
3044 
3045 /*
3046  * This function constructs a pseudo header by looking at the IP header
3047  * and options if any. This is called for both outbound and inbound,
3048  * before computing the ICV.
3049  */
3050 static mblk_t *
3051 ah_process_ip_options_v6(mblk_t *mp, ipsa_t *assoc, int *length_to_skip,
3052     uint_t ah_data_sz, boolean_t outbound, ipsecah_stack_t *ahstack)
3053 {
3054 	ip6_t	*ip6h;
3055 	ip6_t	*oip6h;
3056 	mblk_t 	*phdr_mp;
3057 	int option_length;
3058 	uint_t	ah_align_sz;
3059 	uint_t ah_offset;
3060 	int hdr_size;
3061 
3062 	/*
3063 	 * Allocate space for the authentication data also. It is
3064 	 * useful both during the ICV calculation where we need to
3065 	 * feed in zeroes and while sending the datagram back to IP
3066 	 * where we will be using the same space.
3067 	 *
3068 	 * We need to allocate space for padding bytes if it is not
3069 	 * a multiple of IPV6_PADDING_ALIGN.
3070 	 *
3071 	 * In addition, we allocate space for the ICV computed by
3072 	 * the kernel crypto framework, saving us a separate kmem
3073 	 * allocation down the road.
3074 	 */
3075 
3076 	ah_align_sz = P2ALIGN(ah_data_sz + IPV6_PADDING_ALIGN - 1,
3077 	    IPV6_PADDING_ALIGN);
3078 
3079 	ASSERT(ah_align_sz >= ah_data_sz);
3080 
3081 	hdr_size = ipsec_ah_get_hdr_size_v6(mp, B_FALSE);
3082 	option_length = hdr_size - IPV6_HDR_LEN;
3083 
3084 	/* This was not included in ipsec_ah_get_hdr_size_v6() */
3085 	hdr_size += (sizeof (ah_t) + ah_align_sz);
3086 
3087 	if (!outbound && (MBLKL(mp) < hdr_size)) {
3088 		/*
3089 		 * We have post-AH header options in a separate mblk,
3090 		 * a pullup is required.
3091 		 */
3092 		if (!pullupmsg(mp, hdr_size))
3093 			return (NULL);
3094 	}
3095 
3096 	if ((phdr_mp = allocb_tmpl(hdr_size + ah_data_sz, mp)) == NULL) {
3097 		return (NULL);
3098 	}
3099 
3100 	oip6h = (ip6_t *)mp->b_rptr;
3101 
3102 	/*
3103 	 * Form the basic IP header first. Zero out the header
3104 	 * so that the mutable fields are zeroed out.
3105 	 */
3106 	ip6h = (ip6_t *)phdr_mp->b_rptr;
3107 	bzero(ip6h, sizeof (ip6_t));
3108 	ip6h->ip6_vcf = IPV6_DEFAULT_VERS_AND_FLOW;
3109 
3110 	if (outbound) {
3111 		/*
3112 		 * Include the size of AH and authentication data.
3113 		 * This is how our recipient would compute the
3114 		 * authentication data. Look at what we do in the
3115 		 * inbound case below.
3116 		 */
3117 		ip6h->ip6_plen = htons(ntohs(oip6h->ip6_plen) +
3118 		    sizeof (ah_t) + ah_align_sz);
3119 	} else {
3120 		ip6h->ip6_plen = oip6h->ip6_plen;
3121 	}
3122 
3123 	ip6h->ip6_src = oip6h->ip6_src;
3124 	ip6h->ip6_dst = oip6h->ip6_dst;
3125 
3126 	*length_to_skip = IPV6_HDR_LEN;
3127 	if (option_length == 0) {
3128 		/* Form the AH header */
3129 		ip6h->ip6_nxt = IPPROTO_AH;
3130 		((ah_t *)(ip6h + 1))->ah_nexthdr = oip6h->ip6_nxt;
3131 		ah_offset = *length_to_skip;
3132 	} else {
3133 		ip6h->ip6_nxt = oip6h->ip6_nxt;
3134 		/* option_length does not include the AH header's size */
3135 		*length_to_skip += option_length;
3136 
3137 		ah_offset = ah_fix_phdr_v6(ip6h, oip6h, outbound, B_FALSE);
3138 		if (ah_offset == 0) {
3139 			return (NULL);
3140 		}
3141 	}
3142 
3143 	if (!ah_finish_up(((ah_t *)((uint8_t *)ip6h + ah_offset)),
3144 	    (outbound ? NULL : ((ah_t *)((uint8_t *)oip6h + ah_offset))),
3145 	    assoc, ah_data_sz, ah_align_sz, ahstack)) {
3146 		freeb(phdr_mp);
3147 		/*
3148 		 * Returning NULL will tell the caller to
3149 		 * IPSA_REFELE(), free the memory, etc.
3150 		 */
3151 		return (NULL);
3152 	}
3153 
3154 	phdr_mp->b_wptr = ((uint8_t *)ip6h + ah_offset + sizeof (ah_t) +
3155 	    ah_align_sz);
3156 	if (!outbound)
3157 		*length_to_skip += sizeof (ah_t) + ah_align_sz;
3158 	return (phdr_mp);
3159 }
3160 
3161 /*
3162  * This function constructs a pseudo header by looking at the IP header
3163  * and options if any. This is called for both outbound and inbound,
3164  * before computing the ICV.
3165  */
3166 static mblk_t *
3167 ah_process_ip_options_v4(mblk_t *mp, ipsa_t *assoc, int *length_to_skip,
3168     uint_t ah_data_sz, boolean_t outbound, ipsecah_stack_t *ahstack)
3169 {
3170 	ipoptp_t opts;
3171 	uint32_t option_length;
3172 	ipha_t	*ipha;
3173 	ipha_t	*oipha;
3174 	mblk_t 	*phdr_mp;
3175 	int	 size;
3176 	uchar_t	*optptr;
3177 	uint8_t optval;
3178 	uint8_t optlen;
3179 	ipaddr_t dst;
3180 	uint32_t v_hlen_tos_len;
3181 	int ip_hdr_length;
3182 	uint_t	ah_align_sz;
3183 	uint32_t off;
3184 
3185 #ifdef	_BIG_ENDIAN
3186 #define	V_HLEN	(v_hlen_tos_len >> 24)
3187 #else
3188 #define	V_HLEN	(v_hlen_tos_len & 0xFF)
3189 #endif
3190 
3191 	oipha = (ipha_t *)mp->b_rptr;
3192 	v_hlen_tos_len = ((uint32_t *)oipha)[0];
3193 
3194 	/*
3195 	 * Allocate space for the authentication data also. It is
3196 	 * useful both during the ICV calculation where we need to
3197 	 * feed in zeroes and while sending the datagram back to IP
3198 	 * where we will be using the same space.
3199 	 *
3200 	 * We need to allocate space for padding bytes if it is not
3201 	 * a multiple of IPV4_PADDING_ALIGN.
3202 	 *
3203 	 * In addition, we allocate space for the ICV computed by
3204 	 * the kernel crypto framework, saving us a separate kmem
3205 	 * allocation down the road.
3206 	 */
3207 
3208 	ah_align_sz = P2ALIGN(ah_data_sz + IPV4_PADDING_ALIGN - 1,
3209 	    IPV4_PADDING_ALIGN);
3210 
3211 	ASSERT(ah_align_sz >= ah_data_sz);
3212 
3213 	size = IP_SIMPLE_HDR_LENGTH + sizeof (ah_t) + ah_align_sz +
3214 	    ah_data_sz;
3215 
3216 	if (V_HLEN != IP_SIMPLE_HDR_VERSION) {
3217 		option_length = oipha->ipha_version_and_hdr_length -
3218 		    (uint8_t)((IP_VERSION << 4) +
3219 		    IP_SIMPLE_HDR_LENGTH_IN_WORDS);
3220 		option_length <<= 2;
3221 		size += option_length;
3222 	}
3223 
3224 	if ((phdr_mp = allocb_tmpl(size, mp)) == NULL) {
3225 		return (NULL);
3226 	}
3227 
3228 	/*
3229 	 * Form the basic IP header first.
3230 	 */
3231 	ipha = (ipha_t *)phdr_mp->b_rptr;
3232 	ipha->ipha_version_and_hdr_length = oipha->ipha_version_and_hdr_length;
3233 	ipha->ipha_type_of_service = 0;
3234 
3235 	if (outbound) {
3236 		/*
3237 		 * Include the size of AH and authentication data.
3238 		 * This is how our recipient would compute the
3239 		 * authentication data. Look at what we do in the
3240 		 * inbound case below.
3241 		 */
3242 		ipha->ipha_length = ntohs(htons(oipha->ipha_length) +
3243 		    sizeof (ah_t) + ah_align_sz);
3244 	} else {
3245 		ipha->ipha_length = oipha->ipha_length;
3246 	}
3247 
3248 	ipha->ipha_ident = oipha->ipha_ident;
3249 	ipha->ipha_fragment_offset_and_flags = 0;
3250 	ipha->ipha_ttl = 0;
3251 	ipha->ipha_protocol = IPPROTO_AH;
3252 	ipha->ipha_hdr_checksum = 0;
3253 	ipha->ipha_src = oipha->ipha_src;
3254 	ipha->ipha_dst = dst = oipha->ipha_dst;
3255 
3256 	/*
3257 	 * If there is no option to process return now.
3258 	 */
3259 	ip_hdr_length = IP_SIMPLE_HDR_LENGTH;
3260 
3261 	if (V_HLEN == IP_SIMPLE_HDR_VERSION) {
3262 		/* Form the AH header */
3263 		goto ah_hdr;
3264 	}
3265 
3266 	ip_hdr_length += option_length;
3267 
3268 	/*
3269 	 * We have options. In the outbound case for source route,
3270 	 * ULP has already moved the first hop, which is now in
3271 	 * ipha_dst. We need the final destination for the calculation
3272 	 * of authentication data. And also make sure that mutable
3273 	 * and experimental fields are zeroed out in the IP options.
3274 	 */
3275 
3276 	bcopy(&oipha[1], &ipha[1], option_length);
3277 
3278 	for (optval = ipoptp_first(&opts, ipha);
3279 	    optval != IPOPT_EOL;
3280 	    optval = ipoptp_next(&opts)) {
3281 		optptr = opts.ipoptp_cur;
3282 		optlen = opts.ipoptp_len;
3283 		switch (optval) {
3284 		case IPOPT_EXTSEC:
3285 		case IPOPT_COMSEC:
3286 		case IPOPT_RA:
3287 		case IPOPT_SDMDD:
3288 		case IPOPT_SECURITY:
3289 			/*
3290 			 * These options are Immutable, leave them as-is.
3291 			 * Note that IPOPT_NOP is also Immutable, but it
3292 			 * was skipped by ipoptp_next() and thus remains
3293 			 * intact in the header.
3294 			 */
3295 			break;
3296 		case IPOPT_SSRR:
3297 		case IPOPT_LSRR:
3298 			if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0)
3299 				goto bad_ipv4opt;
3300 			/*
3301 			 * These two are mutable and will be zeroed, but
3302 			 * first get the final destination.
3303 			 */
3304 			off = optptr[IPOPT_OFFSET];
3305 			/*
3306 			 * If one of the conditions is true, it means
3307 			 * end of options and dst already has the right
3308 			 * value. So, just fall through.
3309 			 */
3310 			if (!(optlen < IP_ADDR_LEN || off > optlen - 3)) {
3311 				off = optlen - IP_ADDR_LEN;
3312 				bcopy(&optptr[off], &dst, IP_ADDR_LEN);
3313 			}
3314 			/* FALLTHRU */
3315 		case IPOPT_RR:
3316 		case IPOPT_TS:
3317 		case IPOPT_SATID:
3318 		default:
3319 			/*
3320 			 * optlen should include from the beginning of an
3321 			 * option.
3322 			 * NOTE : Stream Identifier Option (SID): RFC 791
3323 			 * shows the bit pattern of optlen as 2 and documents
3324 			 * the length as 4. We assume it to be 2 here.
3325 			 */
3326 			bzero(optptr, optlen);
3327 			break;
3328 		}
3329 	}
3330 
3331 	if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
3332 bad_ipv4opt:
3333 		ah1dbg(ahstack, ("AH : bad IPv4 option"));
3334 		freeb(phdr_mp);
3335 		return (NULL);
3336 	}
3337 
3338 	/*
3339 	 * Don't change ipha_dst for an inbound datagram as it points
3340 	 * to the right value. Only for the outbound with LSRR/SSRR,
3341 	 * because of ip_massage_options called by the ULP, ipha_dst
3342 	 * points to the first hop and we need to use the final
3343 	 * destination for computing the ICV.
3344 	 */
3345 
3346 	if (outbound)
3347 		ipha->ipha_dst = dst;
3348 ah_hdr:
3349 	((ah_t *)((uint8_t *)ipha + ip_hdr_length))->ah_nexthdr =
3350 	    oipha->ipha_protocol;
3351 	if (!ah_finish_up(((ah_t *)((uint8_t *)ipha + ip_hdr_length)),
3352 	    (outbound ? NULL : ((ah_t *)((uint8_t *)oipha + ip_hdr_length))),
3353 	    assoc, ah_data_sz, ah_align_sz, ahstack)) {
3354 		freeb(phdr_mp);
3355 		/*
3356 		 * Returning NULL will tell the caller to IPSA_REFELE(), free
3357 		 * the memory, etc.
3358 		 */
3359 		return (NULL);
3360 	}
3361 
3362 	phdr_mp->b_wptr = ((uchar_t *)ipha + ip_hdr_length +
3363 	    sizeof (ah_t) + ah_align_sz);
3364 
3365 	ASSERT(phdr_mp->b_wptr <= phdr_mp->b_datap->db_lim);
3366 	if (outbound)
3367 		*length_to_skip = ip_hdr_length;
3368 	else
3369 		*length_to_skip = ip_hdr_length + sizeof (ah_t) + ah_align_sz;
3370 	return (phdr_mp);
3371 }
3372 
3373 /*
3374  * Authenticate an outbound datagram. This function is called
3375  * whenever IP sends an outbound datagram that needs authentication.
3376  * Returns a modified packet if done. Returns NULL if error or queued.
3377  * If error return then ipIfStatsOutDiscards has been increased.
3378  */
3379 static mblk_t *
3380 ah_outbound(mblk_t *data_mp, ip_xmit_attr_t *ixa)
3381 {
3382 	mblk_t *phdr_mp;
3383 	ipsa_t *assoc;
3384 	int length_to_skip;
3385 	uint_t ah_align_sz;
3386 	uint_t age_bytes;
3387 	netstack_t	*ns = ixa->ixa_ipst->ips_netstack;
3388 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
3389 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
3390 	ill_t		*ill = ixa->ixa_nce->nce_ill;
3391 	boolean_t	need_refrele = B_FALSE;
3392 
3393 	/*
3394 	 * Construct the chain of mblks
3395 	 *
3396 	 * PSEUDO_HDR->DATA
3397 	 *
3398 	 * one by one.
3399 	 */
3400 
3401 	AH_BUMP_STAT(ahstack, out_requests);
3402 
3403 	ASSERT(data_mp->b_datap->db_type == M_DATA);
3404 
3405 	assoc = ixa->ixa_ipsec_ah_sa;
3406 	ASSERT(assoc != NULL);
3407 
3408 
3409 	/*
3410 	 * Get the outer IP header in shape to escape this system..
3411 	 */
3412 	if (is_system_labeled() && (assoc->ipsa_otsl != NULL)) {
3413 		/*
3414 		 * Need to update packet with any CIPSO option and update
3415 		 * ixa_tsl to capture the new label.
3416 		 * We allocate a separate ixa for that purpose.
3417 		 */
3418 		ixa = ip_xmit_attr_duplicate(ixa);
3419 		if (ixa == NULL) {
3420 			ip_drop_packet(data_mp, B_FALSE, ill,
3421 			    DROPPER(ipss, ipds_ah_nomem),
3422 			    &ahstack->ah_dropper);
3423 			return (NULL);
3424 		}
3425 		need_refrele = B_TRUE;
3426 
3427 		label_hold(assoc->ipsa_otsl);
3428 		ip_xmit_attr_replace_tsl(ixa, assoc->ipsa_otsl);
3429 
3430 		data_mp = sadb_whack_label(data_mp, assoc, ixa,
3431 		    DROPPER(ipss, ipds_ah_nomem), &ahstack->ah_dropper);
3432 		if (data_mp == NULL) {
3433 			/* Packet dropped by sadb_whack_label */
3434 			ixa_refrele(ixa);
3435 			return (NULL);
3436 		}
3437 	}
3438 
3439 	/*
3440 	 * Age SA according to number of bytes that will be sent after
3441 	 * adding the AH header, ICV, and padding to the packet.
3442 	 */
3443 
3444 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
3445 		ipha_t *ipha = (ipha_t *)data_mp->b_rptr;
3446 		ah_align_sz = P2ALIGN(assoc->ipsa_mac_len +
3447 		    IPV4_PADDING_ALIGN - 1, IPV4_PADDING_ALIGN);
3448 		age_bytes = ntohs(ipha->ipha_length) + sizeof (ah_t) +
3449 		    ah_align_sz;
3450 	} else {
3451 		ip6_t *ip6h = (ip6_t *)data_mp->b_rptr;
3452 		ah_align_sz = P2ALIGN(assoc->ipsa_mac_len +
3453 		    IPV6_PADDING_ALIGN - 1, IPV6_PADDING_ALIGN);
3454 		age_bytes = sizeof (ip6_t) + ntohs(ip6h->ip6_plen) +
3455 		    sizeof (ah_t) + ah_align_sz;
3456 	}
3457 
3458 	if (!ah_age_bytes(assoc, age_bytes, B_FALSE)) {
3459 		/* rig things as if ipsec_getassocbyconn() failed */
3460 		ipsec_assocfailure(info.mi_idnum, 0, 0, SL_ERROR | SL_WARN,
3461 		    "AH association 0x%x, dst %s had bytes expire.\n",
3462 		    ntohl(assoc->ipsa_spi), assoc->ipsa_dstaddr, AF_INET,
3463 		    ahstack->ipsecah_netstack);
3464 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
3465 		ip_drop_output("ipIfStatsOutDiscards", data_mp, ill);
3466 		freemsg(data_mp);
3467 		if (need_refrele)
3468 			ixa_refrele(ixa);
3469 		return (NULL);
3470 	}
3471 
3472 	/*
3473 	 * XXX We need to have fixed up the outer label before we get here.
3474 	 * (AH is computing the checksum over the outer label).
3475 	 */
3476 
3477 	/*
3478 	 * Insert pseudo header:
3479 	 * [IP, ULP] => [IP, AH, ICV] -> ULP
3480 	 */
3481 
3482 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
3483 		phdr_mp = ah_process_ip_options_v4(data_mp, assoc,
3484 		    &length_to_skip, assoc->ipsa_mac_len, B_TRUE, ahstack);
3485 	} else {
3486 		phdr_mp = ah_process_ip_options_v6(data_mp, assoc,
3487 		    &length_to_skip, assoc->ipsa_mac_len, B_TRUE, ahstack);
3488 	}
3489 
3490 	if (phdr_mp == NULL) {
3491 		AH_BUMP_STAT(ahstack, out_discards);
3492 		ip_drop_packet(data_mp, B_FALSE, ixa->ixa_nce->nce_ill,
3493 		    DROPPER(ipss, ipds_ah_bad_v4_opts),
3494 		    &ahstack->ah_dropper);
3495 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
3496 		if (need_refrele)
3497 			ixa_refrele(ixa);
3498 		return (NULL);
3499 	}
3500 
3501 	phdr_mp->b_cont = data_mp;
3502 	data_mp->b_rptr += length_to_skip;
3503 	data_mp = phdr_mp;
3504 
3505 	/*
3506 	 * At this point data_mp points to
3507 	 * an mblk containing the pseudo header (IP header,
3508 	 * AH header, and ICV with mutable fields zero'ed out).
3509 	 * mp points to the mblk containing the ULP data. The original
3510 	 * IP header is kept before the ULP data in data_mp.
3511 	 */
3512 
3513 	/* submit MAC request to KCF */
3514 	data_mp = ah_submit_req_outbound(data_mp, ixa, length_to_skip, assoc);
3515 	if (need_refrele)
3516 		ixa_refrele(ixa);
3517 	return (data_mp);
3518 }
3519 
3520 static mblk_t *
3521 ah_inbound(mblk_t *data_mp, void *arg, ip_recv_attr_t *ira)
3522 {
3523 	ah_t		*ah = (ah_t *)arg;
3524 	ipsa_t		*assoc = ira->ira_ipsec_ah_sa;
3525 	int		length_to_skip;
3526 	int		ah_length;
3527 	mblk_t		*phdr_mp;
3528 	uint32_t	ah_offset;
3529 	netstack_t	*ns = ira->ira_ill->ill_ipst->ips_netstack;
3530 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
3531 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
3532 
3533 	ASSERT(assoc != NULL);
3534 
3535 	/*
3536 	 * We may wish to check replay in-range-only here as an optimization.
3537 	 * Include the reality check of ipsa->ipsa_replay >
3538 	 * ipsa->ipsa_replay_wsize for times when it's the first N packets,
3539 	 * where N == ipsa->ipsa_replay_wsize.
3540 	 *
3541 	 * Another check that may come here later is the "collision" check.
3542 	 * If legitimate packets flow quickly enough, this won't be a problem,
3543 	 * but collisions may cause authentication algorithm crunching to
3544 	 * take place when it doesn't need to.
3545 	 */
3546 	if (!sadb_replay_peek(assoc, ah->ah_replay)) {
3547 		AH_BUMP_STAT(ahstack, replay_early_failures);
3548 		IP_AH_BUMP_STAT(ipss, in_discards);
3549 		ip_drop_packet(data_mp, B_TRUE, ira->ira_ill,
3550 		    DROPPER(ipss, ipds_ah_early_replay),
3551 		    &ahstack->ah_dropper);
3552 		BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
3553 		return (NULL);
3554 	}
3555 
3556 	/*
3557 	 * The offset of the AH header can be computed from its pointer
3558 	 * within the data mblk, which was pulled up until the AH header
3559 	 * by ipsec_inbound_ah_sa() during SA selection.
3560 	 */
3561 	ah_offset = (uchar_t *)ah - data_mp->b_rptr;
3562 
3563 	/*
3564 	 * We need to pullup until the ICV before we call
3565 	 * ah_process_ip_options_v6.
3566 	 */
3567 	ah_length = (ah->ah_length << 2) + 8;
3568 
3569 	/*
3570 	 * NOTE : If we want to use any field of IP/AH header, you need
3571 	 * to re-assign following the pullup.
3572 	 */
3573 	if (((uchar_t *)ah + ah_length) > data_mp->b_wptr) {
3574 		if (!pullupmsg(data_mp, (uchar_t *)ah + ah_length -
3575 		    data_mp->b_rptr)) {
3576 			(void) ipsec_rl_strlog(ns, info.mi_idnum, 0, 0,
3577 			    SL_WARN | SL_ERROR,
3578 			    "ah_inbound: Small AH header\n");
3579 			IP_AH_BUMP_STAT(ipss, in_discards);
3580 			ip_drop_packet(data_mp, B_TRUE, ira->ira_ill,
3581 			    DROPPER(ipss, ipds_ah_nomem),
3582 			    &ahstack->ah_dropper);
3583 			BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
3584 			return (NULL);
3585 		}
3586 	}
3587 
3588 	/*
3589 	 * Insert pseudo header:
3590 	 * [IP, ULP] => [IP, AH, ICV] -> ULP
3591 	 */
3592 	if (ira->ira_flags & IRAF_IS_IPV4) {
3593 		phdr_mp = ah_process_ip_options_v4(data_mp, assoc,
3594 		    &length_to_skip, assoc->ipsa_mac_len, B_FALSE, ahstack);
3595 	} else {
3596 		phdr_mp = ah_process_ip_options_v6(data_mp, assoc,
3597 		    &length_to_skip, assoc->ipsa_mac_len, B_FALSE, ahstack);
3598 	}
3599 
3600 	if (phdr_mp == NULL) {
3601 		IP_AH_BUMP_STAT(ipss, in_discards);
3602 		ip_drop_packet(data_mp, B_TRUE, ira->ira_ill,
3603 		    ((ira->ira_flags & IRAF_IS_IPV4) ?
3604 		    DROPPER(ipss, ipds_ah_bad_v4_opts) :
3605 		    DROPPER(ipss, ipds_ah_bad_v6_hdrs)),
3606 		    &ahstack->ah_dropper);
3607 		BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
3608 		return (NULL);
3609 	}
3610 
3611 	phdr_mp->b_cont = data_mp;
3612 	data_mp->b_rptr += length_to_skip;
3613 	data_mp = phdr_mp;
3614 
3615 	/* submit request to KCF */
3616 	return (ah_submit_req_inbound(data_mp, ira, length_to_skip, ah_offset,
3617 	    assoc));
3618 }
3619 
3620 /*
3621  * Invoked after processing of an inbound packet by the
3622  * kernel crypto framework. Called by ah_submit_req() for a sync request,
3623  * or by the kcf callback for an async request.
3624  * Returns NULL if the mblk chain is consumed.
3625  */
3626 static mblk_t *
3627 ah_auth_in_done(mblk_t *phdr_mp, ip_recv_attr_t *ira, ipsec_crypto_t *ic)
3628 {
3629 	ipha_t *ipha;
3630 	uint_t ah_offset = 0;
3631 	mblk_t *mp;
3632 	int align_len, newpos;
3633 	ah_t *ah;
3634 	uint32_t length;
3635 	uint32_t *dest32;
3636 	uint8_t *dest;
3637 	boolean_t isv4;
3638 	ip6_t *ip6h;
3639 	uint_t icv_len;
3640 	ipsa_t *assoc;
3641 	kstat_named_t *counter;
3642 	netstack_t	*ns = ira->ira_ill->ill_ipst->ips_netstack;
3643 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
3644 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
3645 
3646 	isv4 = (ira->ira_flags & IRAF_IS_IPV4);
3647 	assoc = ira->ira_ipsec_ah_sa;
3648 	icv_len = (uint_t)ic->ic_crypto_mac.cd_raw.iov_len;
3649 
3650 	if (phdr_mp == NULL) {
3651 		ip_drop_packet(phdr_mp, B_TRUE, ira->ira_ill,
3652 		    DROPPER(ipss, ipds_ah_nomem),
3653 		    &ahstack->ah_dropper);
3654 		BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
3655 		return (NULL);
3656 	}
3657 
3658 	mp = phdr_mp->b_cont;
3659 	if (mp == NULL) {
3660 		ip_drop_packet(phdr_mp, B_TRUE, ira->ira_ill,
3661 		    DROPPER(ipss, ipds_ah_nomem),
3662 		    &ahstack->ah_dropper);
3663 		BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
3664 		return (NULL);
3665 	}
3666 	mp->b_rptr -= ic->ic_skip_len;
3667 
3668 	ah_set_usetime(assoc, B_TRUE);
3669 
3670 	if (isv4) {
3671 		ipha = (ipha_t *)mp->b_rptr;
3672 		ah_offset = ipha->ipha_version_and_hdr_length -
3673 		    (uint8_t)((IP_VERSION << 4));
3674 		ah_offset <<= 2;
3675 		align_len = P2ALIGN(icv_len + IPV4_PADDING_ALIGN - 1,
3676 		    IPV4_PADDING_ALIGN);
3677 	} else {
3678 		ip6h = (ip6_t *)mp->b_rptr;
3679 		ah_offset = ipsec_ah_get_hdr_size_v6(mp, B_TRUE);
3680 		ASSERT((mp->b_wptr - mp->b_rptr) >= ah_offset);
3681 		align_len = P2ALIGN(icv_len + IPV6_PADDING_ALIGN - 1,
3682 		    IPV6_PADDING_ALIGN);
3683 	}
3684 
3685 	ah = (ah_t *)(mp->b_rptr + ah_offset);
3686 	newpos = sizeof (ah_t) + align_len;
3687 
3688 	/*
3689 	 * We get here only when authentication passed.
3690 	 */
3691 
3692 	ah3dbg(ahstack, ("AH succeeded, checking replay\n"));
3693 	AH_BUMP_STAT(ahstack, good_auth);
3694 
3695 	if (!sadb_replay_check(assoc, ah->ah_replay)) {
3696 		int af;
3697 		void *addr;
3698 
3699 		if (isv4) {
3700 			addr = &ipha->ipha_dst;
3701 			af = AF_INET;
3702 		} else {
3703 			addr = &ip6h->ip6_dst;
3704 			af = AF_INET6;
3705 		}
3706 
3707 		/*
3708 		 * Log the event. As of now we print out an event.
3709 		 * Do not print the replay failure number, or else
3710 		 * syslog cannot collate the error messages.  Printing
3711 		 * the replay number that failed (or printing to the
3712 		 * console) opens a denial-of-service attack.
3713 		 */
3714 		AH_BUMP_STAT(ahstack, replay_failures);
3715 		ipsec_assocfailure(info.mi_idnum, 0, 0,
3716 		    SL_ERROR | SL_WARN,
3717 		    "Replay failed for AH spi %x, dst_addr %s",
3718 		    assoc->ipsa_spi, addr, af, ahstack->ipsecah_netstack);
3719 		counter = DROPPER(ipss, ipds_ah_replay);
3720 		goto ah_in_discard;
3721 	}
3722 
3723 	/*
3724 	 * We need to remove the AH header from the original
3725 	 * datagram. Best way to do this is to move the pre-AH headers
3726 	 * forward in the (relatively simple) IPv4 case.  In IPv6, it's
3727 	 * a bit more complicated because of IPv6's next-header chaining,
3728 	 * but it's doable.
3729 	 */
3730 	if (isv4) {
3731 		/*
3732 		 * Assign the right protocol, adjust the length as we
3733 		 * are removing the AH header and adjust the checksum to
3734 		 * account for the protocol and length.
3735 		 */
3736 		length = ntohs(ipha->ipha_length);
3737 		if (!ah_age_bytes(assoc, length, B_TRUE)) {
3738 			/* The ipsa has hit hard expiration, LOG and AUDIT. */
3739 			ipsec_assocfailure(info.mi_idnum, 0, 0,
3740 			    SL_ERROR | SL_WARN,
3741 			    "AH Association 0x%x, dst %s had bytes expire.\n",
3742 			    assoc->ipsa_spi, assoc->ipsa_dstaddr,
3743 			    AF_INET, ahstack->ipsecah_netstack);
3744 			AH_BUMP_STAT(ahstack, bytes_expired);
3745 			counter = DROPPER(ipss, ipds_ah_bytes_expire);
3746 			goto ah_in_discard;
3747 		}
3748 		ipha->ipha_protocol = ah->ah_nexthdr;
3749 		length -= newpos;
3750 
3751 		ipha->ipha_length = htons((uint16_t)length);
3752 		ipha->ipha_hdr_checksum = 0;
3753 		ipha->ipha_hdr_checksum = (uint16_t)ip_csum_hdr(ipha);
3754 	} else {
3755 		uchar_t *whereptr;
3756 		int hdrlen;
3757 		uint8_t *nexthdr;
3758 		ip6_hbh_t *hbhhdr;
3759 		ip6_dest_t *dsthdr;
3760 		ip6_rthdr0_t *rthdr;
3761 
3762 		/*
3763 		 * Make phdr_mp hold until the AH header and make
3764 		 * mp hold everything past AH header.
3765 		 */
3766 		length = ntohs(ip6h->ip6_plen);
3767 		if (!ah_age_bytes(assoc, length + sizeof (ip6_t), B_TRUE)) {
3768 			/* The ipsa has hit hard expiration, LOG and AUDIT. */
3769 			ipsec_assocfailure(info.mi_idnum, 0, 0,
3770 			    SL_ERROR | SL_WARN,
3771 			    "AH Association 0x%x, dst %s had bytes "
3772 			    "expire.\n", assoc->ipsa_spi, &ip6h->ip6_dst,
3773 			    AF_INET6, ahstack->ipsecah_netstack);
3774 			AH_BUMP_STAT(ahstack, bytes_expired);
3775 			counter = DROPPER(ipss, ipds_ah_bytes_expire);
3776 			goto ah_in_discard;
3777 		}
3778 
3779 		/*
3780 		 * Update the next header field of the header preceding
3781 		 * AH with the next header field of AH. Start with the
3782 		 * IPv6 header and proceed with the extension headers
3783 		 * until we find what we're looking for.
3784 		 */
3785 		nexthdr = &ip6h->ip6_nxt;
3786 		whereptr =  (uchar_t *)ip6h;
3787 		hdrlen = sizeof (ip6_t);
3788 
3789 		while (*nexthdr != IPPROTO_AH) {
3790 			whereptr += hdrlen;
3791 			/* Assume IP has already stripped it */
3792 			ASSERT(*nexthdr != IPPROTO_FRAGMENT);
3793 			switch (*nexthdr) {
3794 			case IPPROTO_HOPOPTS:
3795 				hbhhdr = (ip6_hbh_t *)whereptr;
3796 				nexthdr = &hbhhdr->ip6h_nxt;
3797 				hdrlen = 8 * (hbhhdr->ip6h_len + 1);
3798 				break;
3799 			case IPPROTO_DSTOPTS:
3800 				dsthdr = (ip6_dest_t *)whereptr;
3801 				nexthdr = &dsthdr->ip6d_nxt;
3802 				hdrlen = 8 * (dsthdr->ip6d_len + 1);
3803 				break;
3804 			case IPPROTO_ROUTING:
3805 				rthdr = (ip6_rthdr0_t *)whereptr;
3806 				nexthdr = &rthdr->ip6r0_nxt;
3807 				hdrlen = 8 * (rthdr->ip6r0_len + 1);
3808 				break;
3809 			}
3810 		}
3811 		*nexthdr = ah->ah_nexthdr;
3812 		length -= newpos;
3813 		ip6h->ip6_plen = htons((uint16_t)length);
3814 	}
3815 
3816 	/* Now that we've fixed the IP header, move it forward. */
3817 	mp->b_rptr += newpos;
3818 	if (IS_P2ALIGNED(mp->b_rptr, sizeof (uint32_t))) {
3819 		dest32 = (uint32_t *)(mp->b_rptr + ah_offset);
3820 		while (--dest32 >= (uint32_t *)mp->b_rptr)
3821 			*dest32 = *(dest32 - (newpos >> 2));
3822 	} else {
3823 		dest = mp->b_rptr + ah_offset;
3824 		while (--dest >= mp->b_rptr)
3825 			*dest = *(dest - newpos);
3826 	}
3827 	freeb(phdr_mp);
3828 
3829 	/*
3830 	 * If SA is labelled, use its label, else inherit the label
3831 	 */
3832 	if (is_system_labeled() && (assoc->ipsa_tsl != NULL)) {
3833 		if (!ip_recv_attr_replace_label(ira, assoc->ipsa_tsl)) {
3834 			ip_drop_packet(mp, B_TRUE, ira->ira_ill,
3835 			    DROPPER(ipss, ipds_ah_nomem), &ahstack->ah_dropper);
3836 			BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
3837 			return (NULL);
3838 		}
3839 	}
3840 
3841 	if (assoc->ipsa_state == IPSA_STATE_IDLE) {
3842 		/*
3843 		 * Cluster buffering case.  Tell caller that we're
3844 		 * handling the packet.
3845 		 */
3846 		sadb_buf_pkt(assoc, mp, ira);
3847 		return (NULL);
3848 	}
3849 
3850 	return (mp);
3851 
3852 ah_in_discard:
3853 	IP_AH_BUMP_STAT(ipss, in_discards);
3854 	ip_drop_packet(phdr_mp, B_TRUE, ira->ira_ill, counter,
3855 	    &ahstack->ah_dropper);
3856 	BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
3857 	return (NULL);
3858 }
3859 
3860 /*
3861  * Invoked after processing of an outbound packet by the
3862  * kernel crypto framework, either by ah_submit_req() for a request
3863  * executed syncrhonously, or by the KEF callback for a request
3864  * executed asynchronously.
3865  */
3866 static mblk_t *
3867 ah_auth_out_done(mblk_t *phdr_mp, ip_xmit_attr_t *ixa, ipsec_crypto_t *ic)
3868 {
3869 	mblk_t *mp;
3870 	int align_len;
3871 	uint32_t hdrs_length;
3872 	uchar_t *ptr;
3873 	uint32_t length;
3874 	boolean_t isv4;
3875 	size_t icv_len;
3876 	netstack_t	*ns = ixa->ixa_ipst->ips_netstack;
3877 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
3878 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
3879 	ill_t		*ill = ixa->ixa_nce->nce_ill;
3880 
3881 	isv4 = (ixa->ixa_flags & IXAF_IS_IPV4);
3882 	icv_len = ic->ic_crypto_mac.cd_raw.iov_len;
3883 
3884 	mp = phdr_mp->b_cont;
3885 	if (mp == NULL) {
3886 		ip_drop_packet(phdr_mp, B_FALSE, ill,
3887 		    DROPPER(ipss, ipds_ah_nomem),
3888 		    &ahstack->ah_dropper);
3889 		BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
3890 		return (NULL);
3891 	}
3892 	mp->b_rptr -= ic->ic_skip_len;
3893 
3894 	ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
3895 	ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
3896 	ah_set_usetime(ixa->ixa_ipsec_ah_sa, B_FALSE);
3897 
3898 	if (isv4) {
3899 		ipha_t *ipha;
3900 		ipha_t *nipha;
3901 
3902 		ipha = (ipha_t *)mp->b_rptr;
3903 		hdrs_length = ipha->ipha_version_and_hdr_length -
3904 		    (uint8_t)((IP_VERSION << 4));
3905 		hdrs_length <<= 2;
3906 		align_len = P2ALIGN(icv_len + IPV4_PADDING_ALIGN - 1,
3907 		    IPV4_PADDING_ALIGN);
3908 		/*
3909 		 * phdr_mp must have the right amount of space for the
3910 		 * combined IP and AH header. Copy the IP header and
3911 		 * the ack_data onto AH. Note that the AH header was
3912 		 * already formed before the ICV calculation and hence
3913 		 * you don't have to copy it here.
3914 		 */
3915 		bcopy(mp->b_rptr, phdr_mp->b_rptr, hdrs_length);
3916 
3917 		ptr = phdr_mp->b_rptr + hdrs_length + sizeof (ah_t);
3918 		bcopy(phdr_mp->b_wptr, ptr, icv_len);
3919 
3920 		/*
3921 		 * Compute the new header checksum as we are assigning
3922 		 * IPPROTO_AH and adjusting the length here.
3923 		 */
3924 		nipha = (ipha_t *)phdr_mp->b_rptr;
3925 
3926 		nipha->ipha_protocol = IPPROTO_AH;
3927 		length = ntohs(nipha->ipha_length);
3928 		length += (sizeof (ah_t) + align_len);
3929 		nipha->ipha_length = htons((uint16_t)length);
3930 		nipha->ipha_hdr_checksum = 0;
3931 		nipha->ipha_hdr_checksum = (uint16_t)ip_csum_hdr(nipha);
3932 	} else {
3933 		ip6_t *ip6h;
3934 		ip6_t *nip6h;
3935 		uint_t ah_offset;
3936 
3937 		ip6h = (ip6_t *)mp->b_rptr;
3938 		nip6h = (ip6_t *)phdr_mp->b_rptr;
3939 		align_len = P2ALIGN(icv_len + IPV6_PADDING_ALIGN - 1,
3940 		    IPV6_PADDING_ALIGN);
3941 		/*
3942 		 * phdr_mp must have the right amount of space for the
3943 		 * combined IP and AH header. Copy the IP header with
3944 		 * options into the pseudo header. When we constructed
3945 		 * a pseudo header, we did not copy some of the mutable
3946 		 * fields. We do it now by calling ah_fix_phdr_v6()
3947 		 * with the last argument B_TRUE. It returns the
3948 		 * ah_offset into the pseudo header.
3949 		 */
3950 
3951 		bcopy(ip6h, nip6h, IPV6_HDR_LEN);
3952 		ah_offset = ah_fix_phdr_v6(nip6h, ip6h, B_TRUE, B_TRUE);
3953 		ASSERT(ah_offset != 0);
3954 		/*
3955 		 * phdr_mp can hold exactly the whole IP header with options
3956 		 * plus the AH header also. Thus subtracting the AH header's
3957 		 * size should give exactly how much of the original header
3958 		 * should be skipped.
3959 		 */
3960 		hdrs_length = (phdr_mp->b_wptr - phdr_mp->b_rptr) -
3961 		    sizeof (ah_t) - icv_len;
3962 		bcopy(phdr_mp->b_wptr, ((uint8_t *)nip6h + ah_offset +
3963 		    sizeof (ah_t)), icv_len);
3964 		length = ntohs(nip6h->ip6_plen);
3965 		length += (sizeof (ah_t) + align_len);
3966 		nip6h->ip6_plen = htons((uint16_t)length);
3967 	}
3968 
3969 	/* Skip the original IP header */
3970 	mp->b_rptr += hdrs_length;
3971 	if (mp->b_rptr == mp->b_wptr) {
3972 		phdr_mp->b_cont = mp->b_cont;
3973 		freeb(mp);
3974 	}
3975 
3976 	return (phdr_mp);
3977 }
3978 
3979 /* Refactor me */
3980 /*
3981  * Wrapper to allow IP to trigger an AH association failure message
3982  * during SA inbound selection.
3983  */
3984 void
3985 ipsecah_in_assocfailure(mblk_t *mp, char level, ushort_t sl, char *fmt,
3986     uint32_t spi, void *addr, int af, ip_recv_attr_t *ira)
3987 {
3988 	netstack_t	*ns = ira->ira_ill->ill_ipst->ips_netstack;
3989 	ipsecah_stack_t	*ahstack = ns->netstack_ipsecah;
3990 	ipsec_stack_t	*ipss = ns->netstack_ipsec;
3991 
3992 	if (ahstack->ipsecah_log_unknown_spi) {
3993 		ipsec_assocfailure(info.mi_idnum, 0, level, sl, fmt, spi,
3994 		    addr, af, ahstack->ipsecah_netstack);
3995 	}
3996 
3997 	ip_drop_packet(mp, B_TRUE, ira->ira_ill,
3998 	    DROPPER(ipss, ipds_ah_no_sa),
3999 	    &ahstack->ah_dropper);
4000 }
4001 
4002 /*
4003  * Initialize the AH input and output processing functions.
4004  */
4005 void
4006 ipsecah_init_funcs(ipsa_t *sa)
4007 {
4008 	if (sa->ipsa_output_func == NULL)
4009 		sa->ipsa_output_func = ah_outbound;
4010 	if (sa->ipsa_input_func == NULL)
4011 		sa->ipsa_input_func = ah_inbound;
4012 }
4013