xref: /illumos-gate/usr/src/uts/common/inet/ip/keysock.c (revision 60405de4d8688d96dd05157c28db3ade5c9bc234)
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 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 #pragma ident	"%Z%%M%	%I%	%E% SMI"
27 
28 
29 #include <sys/param.h>
30 #include <sys/types.h>
31 #include <sys/stream.h>
32 #include <sys/strsubr.h>
33 #include <sys/strsun.h>
34 #include <sys/stropts.h>
35 #include <sys/vnode.h>
36 #include <sys/strlog.h>
37 #include <sys/sysmacros.h>
38 #define	_SUN_TPI_VERSION 2
39 #include <sys/tihdr.h>
40 #include <sys/timod.h>
41 #include <sys/tiuser.h>
42 #include <sys/ddi.h>
43 #include <sys/sunddi.h>
44 #include <sys/sunldi.h>
45 #include <sys/file.h>
46 #include <sys/modctl.h>
47 #include <sys/debug.h>
48 #include <sys/kmem.h>
49 #include <sys/cmn_err.h>
50 #include <sys/proc.h>
51 #include <sys/suntpi.h>
52 #include <sys/atomic.h>
53 #include <sys/mkdev.h>
54 #include <sys/policy.h>
55 
56 #include <sys/socket.h>
57 #include <netinet/in.h>
58 #include <net/pfkeyv2.h>
59 
60 #include <inet/common.h>
61 #include <netinet/ip6.h>
62 #include <inet/ip.h>
63 #include <inet/mi.h>
64 #include <inet/nd.h>
65 #include <inet/optcom.h>
66 #include <inet/ipsec_info.h>
67 #include <inet/ipsec_impl.h>
68 #include <inet/keysock.h>
69 
70 #include <sys/isa_defs.h>
71 
72 /*
73  * This is a transport provider for the PF_KEY key mangement socket.
74  * (See RFC 2367 for details.)
75  * Downstream messages are wrapped in a keysock consumer interface KEYSOCK_IN
76  * messages (see ipsec_info.h), and passed to the appropriate consumer.
77  * Upstream messages are generated for all open PF_KEY sockets, when
78  * appropriate, as well as the sender (as long as SO_USELOOPBACK is enabled)
79  * in reply to downstream messages.
80  *
81  * Upstream messages must be created asynchronously for the following
82  * situations:
83  *
84  *	1.) A keysock consumer requires an SA, and there is currently none.
85  *	2.) An SA expires, either hard or soft lifetime.
86  *	3.) Other events a consumer deems fit.
87  *
88  * The MT model of this is PERMOD, with shared put procedures.  Two types of
89  * messages, SADB_FLUSH and SADB_DUMP, need to lock down the perimeter to send
90  * down the *multiple* messages they create.
91  */
92 
93 /* List of open PF_KEY sockets, protected by keysock_list_lock. */
94 static kmutex_t keysock_list_lock;
95 static keysock_t *keysock_list;
96 
97 static vmem_t *keysock_vmem;		/* for minor numbers. */
98 
99 /* Consumers table.  If an entry is NULL, keysock maintains the table. */
100 static kmutex_t keysock_consumers_lock;
101 
102 #define	KEYSOCK_MAX_CONSUMERS 256
103 static keysock_consumer_t *keysock_consumers[KEYSOCK_MAX_CONSUMERS];
104 
105 /* Default structure copied into T_INFO_ACK messages (from rts.c...) */
106 static struct T_info_ack keysock_g_t_info_ack = {
107 	T_INFO_ACK,
108 	T_INFINITE,	/* TSDU_size. Maximum size messages. */
109 	T_INVALID,	/* ETSDU_size. No expedited data. */
110 	T_INVALID,	/* CDATA_size. No connect data. */
111 	T_INVALID,	/* DDATA_size. No disconnect data. */
112 	0,		/* ADDR_size. */
113 	0,		/* OPT_size. No user-settable options */
114 	64 * 1024,	/* TIDU_size. keysock allows maximum size messages. */
115 	T_COTS,		/* SERV_type. keysock supports connection oriented. */
116 	TS_UNBND,	/* CURRENT_state. This is set from keysock_state. */
117 	(XPG4_1)	/* Provider flags */
118 };
119 
120 /* Named Dispatch Parameter Management Structure */
121 typedef struct keysockpparam_s {
122 	uint_t	keysock_param_min;
123 	uint_t	keysock_param_max;
124 	uint_t	keysock_param_value;
125 	char	*keysock_param_name;
126 } keysockparam_t;
127 
128 /*
129  * Table of NDD variables supported by keysock. These are loaded into
130  * keysock_g_nd in keysock_init_nd.
131  * All of these are alterable, within the min/max values given, at run time.
132  */
133 static	keysockparam_t	keysock_param_arr[] = {
134 	/* min	max	value	name */
135 	{ 4096, 65536,	8192,	"keysock_xmit_hiwat"},
136 	{ 0,	65536,	1024,	"keysock_xmit_lowat"},
137 	{ 4096, 65536,	8192,	"keysock_recv_hiwat"},
138 	{ 65536, 1024*1024*1024, 256*1024,	"keysock_max_buf"},
139 	{ 0,	3,	0,	"keysock_debug"},
140 };
141 #define	keysock_xmit_hiwat	keysock_param_arr[0].keysock_param_value
142 #define	keysock_xmit_lowat	keysock_param_arr[1].keysock_param_value
143 #define	keysock_recv_hiwat	keysock_param_arr[2].keysock_param_value
144 #define	keysock_max_buf		keysock_param_arr[3].keysock_param_value
145 #define	keysock_debug		keysock_param_arr[4].keysock_param_value
146 
147 kmutex_t keysock_param_lock;	/* Protects the NDD variables. */
148 
149 #define	ks0dbg(a)	printf a
150 /* NOTE:  != 0 instead of > 0 so lint doesn't complain. */
151 #define	ks1dbg(a)	if (keysock_debug != 0) printf a
152 #define	ks2dbg(a)	if (keysock_debug > 1) printf a
153 #define	ks3dbg(a)	if (keysock_debug > 2) printf a
154 
155 static IDP keysock_g_nd;
156 
157 /*
158  * State for flush/dump.  This would normally be a boolean_t, but
159  * cas32() works best for a known 32-bit quantity.
160  */
161 static uint32_t keysock_flushdump;
162 static int keysock_flushdump_errno;
163 
164 static int keysock_close(queue_t *);
165 static int keysock_open(queue_t *, dev_t *, int, int, cred_t *);
166 static void keysock_wput(queue_t *, mblk_t *);
167 static void keysock_rput(queue_t *, mblk_t *);
168 static void keysock_rsrv(queue_t *);
169 static void keysock_passup(mblk_t *, sadb_msg_t *, minor_t,
170     keysock_consumer_t *, boolean_t);
171 
172 static struct module_info info = {
173 	5138, "keysock", 1, INFPSZ, 512, 128
174 };
175 
176 static struct qinit rinit = {
177 	(pfi_t)keysock_rput, (pfi_t)keysock_rsrv, keysock_open, keysock_close,
178 	NULL, &info
179 };
180 
181 static struct qinit winit = {
182 	(pfi_t)keysock_wput, NULL, NULL, NULL, NULL, &info
183 };
184 
185 struct streamtab keysockinfo = {
186 	&rinit, &winit
187 };
188 
189 extern struct modlinkage *keysock_modlp;
190 
191 /*
192  * Plumb IPsec.
193  *
194  * NOTE:  New "default" modules will need to be loaded here if needed before
195  *	  boot time.
196  */
197 
198 /* Keep these in global space to keep the lint from complaining. */
199 static char *IPSECESP = "ipsecesp";
200 static char *IPSECESPDEV = "/devices/pseudo/ipsecesp@0:ipsecesp";
201 static char *IPSECAH = "ipsecah";
202 static char *IPSECAHDEV = "/devices/pseudo/ipsecah@0:ipsecah";
203 static char *IP6DEV = "/devices/pseudo/ip6@0:ip6";
204 static char *KEYSOCK = "keysock";
205 static char *STRMOD = "strmod";
206 
207 /*
208  * keysock_plumbed: zero if plumb not attempted, positive if it succeeded,
209  * negative if it failed.
210  */
211 static int keysock_plumbed = 0;
212 
213 /*
214  * This integer counts the number of extended REGISTERed sockets.  This
215  * determines if we should send extended REGISTERs.
216  */
217 static uint32_t keysock_num_extended = 0;
218 
219 /*
220  * Global sequence space for SADB_ACQUIRE messages of any sort.
221  */
222 static uint32_t keysock_acquire_seq = 0xffffffff;
223 
224 /*
225  * Load the other ipsec modules and plumb them together.
226  */
227 int
228 keysock_plumb_ipsec(void)
229 {
230 	ldi_handle_t	lh, ip6_lh = NULL;
231 	ldi_ident_t	li = NULL;
232 	int		err = 0;
233 	int		muxid, rval;
234 	boolean_t	esp_present = B_TRUE;
235 
236 
237 	keysock_plumbed = 0;	/* we're trying again.. */
238 
239 	/*
240 	 * Load up the drivers (AH/ESP).
241 	 *
242 	 * I do this separately from the actual plumbing in case this function
243 	 * ever gets called from a diskless boot before the root filesystem is
244 	 * up.  I don't have to worry about "keysock" because, well, if I'm
245 	 * here, keysock must've loaded successfully.
246 	 */
247 	if (i_ddi_attach_pseudo_node(IPSECAH) == NULL) {
248 		ks0dbg(("IPsec:  AH failed to attach.\n"));
249 		goto bail;
250 	}
251 	if (i_ddi_attach_pseudo_node(IPSECESP) == NULL) {
252 		ks0dbg(("IPsec:  ESP failed to attach.\n"));
253 		esp_present = B_FALSE;
254 	}
255 
256 	/*
257 	 * Set up the IP streams for AH and ESP, as well as tacking keysock
258 	 * on top of them.  Assume keysock has set the autopushes up already.
259 	 */
260 
261 	/* Open IP. */
262 	err = ldi_ident_from_mod(keysock_modlp, &li);
263 	if (err) {
264 		ks0dbg(("IPsec:  lid_ident_from_mod failed (err %d).\n",
265 		    err));
266 		goto bail;
267 	}
268 
269 	err = ldi_open_by_name(IP6DEV, FREAD|FWRITE, CRED(), &ip6_lh, li);
270 	if (err) {
271 		ks0dbg(("IPsec:  Open of IP6 failed (err %d).\n", err));
272 		goto bail;
273 	}
274 
275 	/* PLINK KEYSOCK/AH */
276 	err = ldi_open_by_name(IPSECAHDEV, FREAD|FWRITE, CRED(), &lh, li);
277 	if (err) {
278 		ks0dbg(("IPsec:  Open of AH failed (err %d).\n", err));
279 		goto bail;
280 	}
281 	err = ldi_ioctl(lh,
282 	    I_PUSH, (intptr_t)KEYSOCK, FKIOCTL, CRED(), &rval);
283 	if (err) {
284 		ks0dbg(("IPsec:  Push of KEYSOCK onto AH failed (err %d).\n",
285 		    err));
286 		(void) ldi_close(lh, FREAD|FWRITE, CRED());
287 		goto bail;
288 	}
289 	err = ldi_ioctl(ip6_lh, I_PLINK, (intptr_t)lh,
290 			FREAD+FWRITE+FNOCTTY+FKIOCTL, kcred, &muxid);
291 	if (err) {
292 		ks0dbg(("IPsec:  PLINK of KEYSOCK/AH failed (err %d).\n", err));
293 		(void) ldi_close(lh, FREAD|FWRITE, CRED());
294 		goto bail;
295 	}
296 	(void) ldi_close(lh, FREAD|FWRITE, CRED());
297 
298 	/* PLINK KEYSOCK/ESP */
299 	if (esp_present) {
300 		err = ldi_open_by_name(IPSECESPDEV,
301 		    FREAD|FWRITE, CRED(), &lh, li);
302 		if (err) {
303 			ks0dbg(("IPsec:  Open of ESP failed (err %d).\n", err));
304 			goto bail;
305 		}
306 		err = ldi_ioctl(lh,
307 		    I_PUSH, (intptr_t)KEYSOCK, FKIOCTL, CRED(), &rval);
308 		if (err) {
309 			ks0dbg(("IPsec:  "
310 			    "Push of KEYSOCK onto ESP failed (err %d).\n",
311 			    err));
312 			(void) ldi_close(lh, FREAD|FWRITE, CRED());
313 			goto bail;
314 		}
315 		err = ldi_ioctl(ip6_lh, I_PLINK, (intptr_t)lh,
316 				FREAD+FWRITE+FNOCTTY+FKIOCTL, kcred, &muxid);
317 		if (err) {
318 			ks0dbg(("IPsec:  "
319 			    "PLINK of KEYSOCK/ESP failed (err %d).\n", err));
320 			(void) ldi_close(lh, FREAD|FWRITE, CRED());
321 			goto bail;
322 		}
323 		(void) ldi_close(lh, FREAD|FWRITE, CRED());
324 	}
325 
326 bail:
327 	keysock_plumbed = (err == 0) ? 1 : -1;
328 	if (ip6_lh != NULL) {
329 		(void) ldi_close(ip6_lh, FREAD|FWRITE, CRED());
330 	}
331 	if (li != NULL)
332 		ldi_ident_release(li);
333 	return (err);
334 }
335 
336 /* ARGSUSED */
337 static int
338 keysock_param_get(q, mp, cp, cr)
339 	queue_t	*q;
340 	mblk_t	*mp;
341 	caddr_t	cp;
342 	cred_t *cr;
343 {
344 	keysockparam_t	*keysockpa = (keysockparam_t *)cp;
345 	uint_t value;
346 
347 	mutex_enter(&keysock_param_lock);
348 	value = keysockpa->keysock_param_value;
349 	mutex_exit(&keysock_param_lock);
350 
351 	(void) mi_mpprintf(mp, "%u", value);
352 	return (0);
353 }
354 
355 /* This routine sets an NDD variable in a keysockparam_t structure. */
356 /* ARGSUSED */
357 static int
358 keysock_param_set(q, mp, value, cp, cr)
359 	queue_t	*q;
360 	mblk_t	*mp;
361 	char	*value;
362 	caddr_t	cp;
363 	cred_t *cr;
364 {
365 	ulong_t	new_value;
366 	keysockparam_t	*keysockpa = (keysockparam_t *)cp;
367 
368 	/* Convert the value from a string into a long integer. */
369 	if (ddi_strtoul(value, NULL, 10, &new_value) != 0)
370 		return (EINVAL);
371 
372 	mutex_enter(&keysock_param_lock);
373 	/*
374 	 * Fail the request if the new value does not lie within the
375 	 * required bounds.
376 	 */
377 	if (new_value < keysockpa->keysock_param_min ||
378 	    new_value > keysockpa->keysock_param_max) {
379 		mutex_exit(&keysock_param_lock);
380 		return (EINVAL);
381 	}
382 
383 	/* Set the new value */
384 	keysockpa->keysock_param_value = new_value;
385 	mutex_exit(&keysock_param_lock);
386 
387 	return (0);
388 }
389 
390 /*
391  * Initialize NDD variables, and other things, for keysock.
392  */
393 boolean_t
394 keysock_ddi_init(void)
395 {
396 	keysockparam_t *ksp = keysock_param_arr;
397 	int count = A_CNT(keysock_param_arr);
398 
399 	if (!keysock_g_nd) {
400 		for (; count-- > 0; ksp++) {
401 			if (ksp->keysock_param_name != NULL &&
402 			    ksp->keysock_param_name[0]) {
403 				if (!nd_load(&keysock_g_nd,
404 				    ksp->keysock_param_name,
405 				    keysock_param_get, keysock_param_set,
406 				    (caddr_t)ksp)) {
407 					nd_free(&keysock_g_nd);
408 					return (B_FALSE);
409 				}
410 			}
411 		}
412 	}
413 
414 	keysock_max_optsize = optcom_max_optsize(
415 	    keysock_opt_obj.odb_opt_des_arr, keysock_opt_obj.odb_opt_arr_cnt);
416 
417 	keysock_vmem = vmem_create("keysock", (void *)1, MAXMIN, 1,
418 	    NULL, NULL, NULL, 1, VM_SLEEP | VMC_IDENTIFIER);
419 
420 	mutex_init(&keysock_list_lock, NULL, MUTEX_DEFAULT, NULL);
421 	mutex_init(&keysock_consumers_lock, NULL, MUTEX_DEFAULT, NULL);
422 	mutex_init(&keysock_param_lock, NULL, MUTEX_DEFAULT, NULL);
423 
424 	return (B_TRUE);
425 }
426 
427 /*
428  * Free NDD variable space, and other destructors, for keysock.
429  */
430 void
431 keysock_ddi_destroy(void)
432 {
433 	/* XXX Free instances? */
434 	ks0dbg(("keysock_ddi_destroy being called.\n"));
435 
436 	vmem_destroy(keysock_vmem);
437 	mutex_destroy(&keysock_list_lock);
438 	mutex_destroy(&keysock_consumers_lock);
439 	mutex_destroy(&keysock_param_lock);
440 	nd_free(&keysock_g_nd);
441 }
442 
443 /*
444  * Close routine for keysock.
445  */
446 static int
447 keysock_close(queue_t *q)
448 {
449 	keysock_t *ks;
450 	keysock_consumer_t *kc;
451 	void *ptr = q->q_ptr;
452 	int size;
453 
454 	qprocsoff(q);
455 
456 	/* Safe assumption. */
457 	ASSERT(ptr != NULL);
458 
459 	if (WR(q)->q_next) {
460 		kc = (keysock_consumer_t *)ptr;
461 		ks0dbg(("Module close, removing a consumer (%d).\n",
462 		    kc->kc_sa_type));
463 		/*
464 		 * Because of PERMOD open/close exclusive perimeter, I
465 		 * can inspect KC_FLUSHING w/o locking down kc->kc_lock.
466 		 */
467 		if (kc->kc_flags & KC_FLUSHING) {
468 			/*
469 			 * If this decrement was the last one, send
470 			 * down the next pending one, if any.
471 			 *
472 			 * With a PERMOD perimeter, the mutexes ops aren't
473 			 * really necessary, but if we ever loosen up, we will
474 			 * have this bit covered already.
475 			 */
476 			keysock_flushdump--;
477 			if (keysock_flushdump == 0) {
478 				/*
479 				 * The flush/dump terminated by having a
480 				 * consumer go away.  I need to send up to the
481 				 * appropriate keysock all of the relevant
482 				 * information.  Unfortunately, I don't
483 				 * have that handy.
484 				 */
485 				ks0dbg(("Consumer went away while flushing or"
486 				    " dumping.\n"));
487 			}
488 		}
489 		size = sizeof (keysock_consumer_t);
490 		mutex_enter(&keysock_consumers_lock);
491 		keysock_consumers[kc->kc_sa_type] = NULL;
492 		mutex_exit(&keysock_consumers_lock);
493 		mutex_destroy(&kc->kc_lock);
494 	} else {
495 		ks3dbg(("Driver close, PF_KEY socket is going away.\n"));
496 		ks = (keysock_t *)ptr;
497 		if ((ks->keysock_flags & KEYSOCK_EXTENDED) != 0)
498 			atomic_add_32(&keysock_num_extended, -1);
499 		size = sizeof (keysock_t);
500 		mutex_enter(&keysock_list_lock);
501 		*(ks->keysock_ptpn) = ks->keysock_next;
502 		if (ks->keysock_next != NULL)
503 			ks->keysock_next->keysock_ptpn = ks->keysock_ptpn;
504 		mutex_exit(&keysock_list_lock);
505 		mutex_destroy(&ks->keysock_lock);
506 		vmem_free(keysock_vmem, (void *)(uintptr_t)ks->keysock_serial,
507 		    1);
508 	}
509 
510 	/* Now I'm free. */
511 	kmem_free(ptr, size);
512 	return (0);
513 }
514 /*
515  * Open routine for keysock.
516  */
517 /* ARGSUSED */
518 static int
519 keysock_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
520 {
521 	keysock_t *ks;
522 	keysock_consumer_t *kc;
523 	mblk_t *mp;
524 	ipsec_info_t *ii;
525 
526 	ks3dbg(("Entering keysock open.\n"));
527 
528 	if (secpolicy_net_config(credp, B_FALSE) != 0) {
529 		/* Privilege debugging will log the error */
530 		return (EPERM);
531 	}
532 
533 	if (q->q_ptr != NULL)
534 		return (0);  /* Re-open of an already open instance. */
535 
536 	if (keysock_plumbed < 1) {
537 		keysock_plumbed = 0;
538 		/*
539 		 * Don't worry about ipsec_failure being true here.
540 		 * (See ip.c).  An open of keysock should try and force
541 		 * the issue.  Maybe it was a transient failure.
542 		 */
543 		ipsec_loader_loadnow();
544 	}
545 
546 	if (sflag & MODOPEN) {
547 		/* Initialize keysock_consumer state here. */
548 		kc = kmem_zalloc(sizeof (keysock_consumer_t), KM_NOSLEEP);
549 		if (kc == NULL)
550 			return (ENOMEM);
551 		mutex_init(&kc->kc_lock, NULL, MUTEX_DEFAULT, 0);
552 		kc->kc_rq = q;
553 		kc->kc_wq = WR(q);
554 
555 		q->q_ptr = kc;
556 		WR(q)->q_ptr = kc;
557 
558 		qprocson(q);
559 
560 		/*
561 		 * Send down initial message to whatever I was pushed on top
562 		 * of asking for its consumer type.  The reply will set it.
563 		 */
564 
565 		/* Allocate it. */
566 		mp = allocb(sizeof (ipsec_info_t), BPRI_HI);
567 		if (mp == NULL) {
568 			ks1dbg((
569 			    "keysock_open:  Cannot allocate KEYSOCK_HELLO.\n"));
570 			/* Do I need to set these to null? */
571 			q->q_ptr = NULL;
572 			WR(q)->q_ptr = NULL;
573 			mutex_destroy(&kc->kc_lock);
574 			kmem_free(kc, sizeof (*kc));
575 			return (ENOMEM);
576 		}
577 
578 		/* If I allocated okay, putnext to what I was pushed atop. */
579 		mp->b_wptr += sizeof (ipsec_info_t);
580 		mp->b_datap->db_type = M_CTL;
581 		ii = (ipsec_info_t *)mp->b_rptr;
582 		ii->ipsec_info_type = KEYSOCK_HELLO;
583 		/* Length only of type/len. */
584 		ii->ipsec_info_len = sizeof (ii->ipsec_allu);
585 		ks2dbg(("Ready to putnext KEYSOCK_HELLO.\n"));
586 		putnext(kc->kc_wq, mp);
587 	} else {
588 		minor_t ksminor;
589 
590 		/* Initialize keysock state. */
591 
592 		ks2dbg(("Made it into PF_KEY socket open.\n"));
593 
594 		ksminor = (minor_t)(uintptr_t)
595 		    vmem_alloc(keysock_vmem, 1, VM_NOSLEEP);
596 		if (ksminor == 0)
597 			return (ENOMEM);
598 
599 		ks = kmem_zalloc(sizeof (keysock_t), KM_NOSLEEP);
600 		if (ks == NULL) {
601 			vmem_free(keysock_vmem, (void *)(uintptr_t)ksminor, 1);
602 			return (ENOMEM);
603 		}
604 
605 		mutex_init(&ks->keysock_lock, NULL, MUTEX_DEFAULT, 0);
606 		ks->keysock_rq = q;
607 		ks->keysock_wq = WR(q);
608 		ks->keysock_state = TS_UNBND;
609 		ks->keysock_serial = ksminor;
610 
611 		q->q_ptr = ks;
612 		WR(q)->q_ptr = ks;
613 
614 		/*
615 		 * The receive hiwat is only looked at on the stream head
616 		 * queue.  Store in q_hiwat in order to return on SO_RCVBUF
617 		 * getsockopts.
618 		 */
619 
620 		q->q_hiwat = keysock_recv_hiwat;
621 
622 		/*
623 		 * The transmit hiwat/lowat is only looked at on IP's queue.
624 		 * Store in q_hiwat/q_lowat in order to return on
625 		 * SO_SNDBUF/SO_SNDLOWAT getsockopts.
626 		 */
627 
628 		WR(q)->q_hiwat = keysock_xmit_hiwat;
629 		WR(q)->q_lowat = keysock_xmit_lowat;
630 
631 		*devp = makedevice(getmajor(*devp), ksminor);
632 
633 		/*
634 		 * Thread keysock into the global keysock list.
635 		 */
636 		mutex_enter(&keysock_list_lock);
637 		ks->keysock_next = keysock_list;
638 		ks->keysock_ptpn = &keysock_list;
639 		if (keysock_list != NULL)
640 			keysock_list->keysock_ptpn = &ks->keysock_next;
641 		keysock_list = ks;
642 		mutex_exit(&keysock_list_lock);
643 
644 		qprocson(q);
645 		(void) mi_set_sth_hiwat(q, keysock_recv_hiwat);
646 		/*
647 		 * Wait outside the keysock module perimeter for IPsec
648 		 * plumbing to be completed.  If it fails, keysock_close()
649 		 * undoes everything we just did.
650 		 */
651 		if (!ipsec_loader_wait(q)) {
652 			(void) keysock_close(q);
653 			return (EPFNOSUPPORT);
654 		}
655 	}
656 
657 	return (0);
658 }
659 
660 /* BELOW THIS LINE ARE ROUTINES INCLUDING AND RELATED TO keysock_wput(). */
661 
662 /*
663  * Copy relevant state bits.
664  */
665 static void
666 keysock_copy_info(struct T_info_ack *tap, keysock_t *ks)
667 {
668 	*tap = keysock_g_t_info_ack;
669 	tap->CURRENT_state = ks->keysock_state;
670 	tap->OPT_size = keysock_max_optsize;
671 }
672 
673 /*
674  * This routine responds to T_CAPABILITY_REQ messages.  It is called by
675  * keysock_wput.  Much of the T_CAPABILITY_ACK information is copied from
676  * keysock_g_t_info_ack.  The current state of the stream is copied from
677  * keysock_state.
678  */
679 static void
680 keysock_capability_req(queue_t *q, mblk_t *mp)
681 {
682 	keysock_t *ks = (keysock_t *)q->q_ptr;
683 	t_uscalar_t cap_bits1;
684 	struct T_capability_ack	*tcap;
685 
686 	cap_bits1 = ((struct T_capability_req *)mp->b_rptr)->CAP_bits1;
687 
688 	mp = tpi_ack_alloc(mp, sizeof (struct T_capability_ack),
689 		mp->b_datap->db_type, T_CAPABILITY_ACK);
690 	if (mp == NULL)
691 		return;
692 
693 	tcap = (struct T_capability_ack *)mp->b_rptr;
694 	tcap->CAP_bits1 = 0;
695 
696 	if (cap_bits1 & TC1_INFO) {
697 		keysock_copy_info(&tcap->INFO_ack, ks);
698 		tcap->CAP_bits1 |= TC1_INFO;
699 	}
700 
701 	qreply(q, mp);
702 }
703 
704 /*
705  * This routine responds to T_INFO_REQ messages. It is called by
706  * keysock_wput_other.
707  * Most of the T_INFO_ACK information is copied from keysock_g_t_info_ack.
708  * The current state of the stream is copied from keysock_state.
709  */
710 static void
711 keysock_info_req(q, mp)
712 	queue_t	*q;
713 	mblk_t	*mp;
714 {
715 	mp = tpi_ack_alloc(mp, sizeof (struct T_info_ack), M_PCPROTO,
716 	    T_INFO_ACK);
717 	if (mp == NULL)
718 		return;
719 	keysock_copy_info((struct T_info_ack *)mp->b_rptr,
720 	    (keysock_t *)q->q_ptr);
721 	qreply(q, mp);
722 }
723 
724 /*
725  * keysock_err_ack. This routine creates a
726  * T_ERROR_ACK message and passes it
727  * upstream.
728  */
729 static void
730 keysock_err_ack(q, mp, t_error, sys_error)
731 	queue_t	*q;
732 	mblk_t	*mp;
733 	int	t_error;
734 	int	sys_error;
735 {
736 	if ((mp = mi_tpi_err_ack_alloc(mp, t_error, sys_error)) != NULL)
737 		qreply(q, mp);
738 }
739 
740 /*
741  * This routine retrieves the current status of socket options.
742  * It returns the size of the option retrieved.
743  */
744 /* ARGSUSED */
745 int
746 keysock_opt_get(queue_t *q, int level, int name, uchar_t *ptr)
747 {
748 	int *i1 = (int *)ptr;
749 	keysock_t *ks = (keysock_t *)q->q_ptr;
750 
751 	switch (level) {
752 	case SOL_SOCKET:
753 		mutex_enter(&ks->keysock_lock);
754 		switch (name) {
755 		case SO_TYPE:
756 			*i1 = SOCK_RAW;
757 			break;
758 		case SO_USELOOPBACK:
759 			*i1 = (int)(!((ks->keysock_flags & KEYSOCK_NOLOOP) ==
760 			    KEYSOCK_NOLOOP));
761 			break;
762 		/*
763 		 * The following two items can be manipulated,
764 		 * but changing them should do nothing.
765 		 */
766 		case SO_SNDBUF:
767 			*i1 = (int)q->q_hiwat;
768 			break;
769 		case SO_RCVBUF:
770 			*i1 = (int)(RD(q)->q_hiwat);
771 			break;
772 		}
773 		mutex_exit(&ks->keysock_lock);
774 		break;
775 	default:
776 		return (0);
777 	}
778 	return (sizeof (int));
779 }
780 
781 /*
782  * This routine sets socket options.
783  */
784 /* ARGSUSED */
785 int
786 keysock_opt_set(queue_t *q, uint_t mgmt_flags, int level,
787     int name, uint_t inlen, uchar_t *invalp, uint_t *outlenp,
788     uchar_t *outvalp, void *thisdg_attrs, cred_t *cr, mblk_t *mblk)
789 {
790 	int *i1 = (int *)invalp;
791 	keysock_t *ks = (keysock_t *)q->q_ptr;
792 
793 	switch (level) {
794 	case SOL_SOCKET:
795 		mutex_enter(&ks->keysock_lock);
796 		switch (name) {
797 		case SO_USELOOPBACK:
798 			if (!(*i1))
799 				ks->keysock_flags |= KEYSOCK_NOLOOP;
800 			else ks->keysock_flags &= ~KEYSOCK_NOLOOP;
801 			break;
802 		case SO_SNDBUF:
803 			if (*i1 > keysock_max_buf)
804 				return (ENOBUFS);
805 			q->q_hiwat = *i1;
806 			break;
807 		case SO_RCVBUF:
808 			if (*i1 > keysock_max_buf)
809 				return (ENOBUFS);
810 			RD(q)->q_hiwat = *i1;
811 			(void) mi_set_sth_hiwat(RD(q), *i1);
812 			break;
813 		}
814 		mutex_exit(&ks->keysock_lock);
815 		break;
816 	}
817 	return (0);
818 }
819 
820 /*
821  * Handle STREAMS messages.
822  */
823 static void
824 keysock_wput_other(queue_t *q, mblk_t *mp)
825 {
826 	struct iocblk *iocp;
827 	int error;
828 
829 	switch (mp->b_datap->db_type) {
830 	case M_PROTO:
831 	case M_PCPROTO:
832 		if ((mp->b_wptr - mp->b_rptr) < sizeof (long)) {
833 			ks3dbg((
834 			    "keysock_wput_other: Not big enough M_PROTO\n"));
835 			freemsg(mp);
836 			return;
837 		}
838 		switch (((union T_primitives *)mp->b_rptr)->type) {
839 		case T_CAPABILITY_REQ:
840 			keysock_capability_req(q, mp);
841 			return;
842 		case T_INFO_REQ:
843 			keysock_info_req(q, mp);
844 			return;
845 		case T_SVR4_OPTMGMT_REQ:
846 			(void) svr4_optcom_req(q, mp, DB_CREDDEF(mp, kcred),
847 			    &keysock_opt_obj);
848 			return;
849 		case T_OPTMGMT_REQ:
850 			(void) tpi_optcom_req(q, mp, DB_CREDDEF(mp, kcred),
851 			    &keysock_opt_obj);
852 			return;
853 		case T_DATA_REQ:
854 		case T_EXDATA_REQ:
855 		case T_ORDREL_REQ:
856 			/* Illegal for keysock. */
857 			freemsg(mp);
858 			(void) putnextctl1(RD(q), M_ERROR, EPROTO);
859 			return;
860 		default:
861 			/* Not supported by keysock. */
862 			keysock_err_ack(q, mp, TNOTSUPPORT, 0);
863 			return;
864 		}
865 	case M_IOCTL:
866 		iocp = (struct iocblk *)mp->b_rptr;
867 		error = EINVAL;
868 
869 		switch (iocp->ioc_cmd) {
870 		case ND_SET:
871 		case ND_GET:
872 			if (nd_getset(q, keysock_g_nd, mp)) {
873 				qreply(q, mp);
874 				return;
875 			} else
876 				error = ENOENT;
877 			/* FALLTHRU */
878 		default:
879 			miocnak(q, mp, 0, error);
880 			return;
881 		}
882 	case M_FLUSH:
883 		if (*mp->b_rptr & FLUSHW) {
884 			flushq(q, FLUSHALL);
885 			*mp->b_rptr &= ~FLUSHW;
886 		}
887 		if (*mp->b_rptr & FLUSHR) {
888 			qreply(q, mp);
889 			return;
890 		}
891 		/* Else FALLTHRU */
892 	}
893 
894 	/* If fell through, just black-hole the message. */
895 	freemsg(mp);
896 }
897 
898 /*
899  * Transmit a PF_KEY error message to the instance either pointed to
900  * by ks, the instance with serial number serial, or more, depending.
901  *
902  * The faulty message (or a reasonable facsimile thereof) is in mp.
903  * This function will free mp or recycle it for delivery, thereby causing
904  * the stream head to free it.
905  */
906 static void
907 keysock_error(keysock_t *ks, mblk_t *mp, int error, int diagnostic)
908 {
909 	sadb_msg_t *samsg = (sadb_msg_t *)mp->b_rptr;
910 
911 	ASSERT(mp->b_datap->db_type == M_DATA);
912 
913 	if (samsg->sadb_msg_type < SADB_GETSPI ||
914 	    samsg->sadb_msg_type > SADB_MAX)
915 		samsg->sadb_msg_type = SADB_RESERVED;
916 
917 	/*
918 	 * Strip out extension headers.
919 	 */
920 	ASSERT(mp->b_rptr + sizeof (*samsg) <= mp->b_datap->db_lim);
921 	mp->b_wptr = mp->b_rptr + sizeof (*samsg);
922 	samsg->sadb_msg_len = SADB_8TO64(sizeof (sadb_msg_t));
923 	samsg->sadb_msg_errno = (uint8_t)error;
924 	samsg->sadb_x_msg_diagnostic = (uint16_t)diagnostic;
925 
926 	keysock_passup(mp, samsg, ks->keysock_serial, NULL, B_FALSE);
927 }
928 
929 /*
930  * Pass down a message to a consumer.  Wrap it in KEYSOCK_IN, and copy
931  * in the extv if passed in.
932  */
933 static void
934 keysock_passdown(keysock_t *ks, mblk_t *mp, uint8_t satype, sadb_ext_t *extv[],
935     boolean_t flushmsg)
936 {
937 	keysock_consumer_t *kc;
938 	mblk_t *wrapper;
939 	keysock_in_t *ksi;
940 	int i;
941 
942 	wrapper = allocb(sizeof (ipsec_info_t), BPRI_HI);
943 	if (wrapper == NULL) {
944 		ks3dbg(("keysock_passdown: allocb failed.\n"));
945 		if (extv[SADB_EXT_KEY_ENCRYPT] != NULL)
946 			bzero(extv[SADB_EXT_KEY_ENCRYPT],
947 			    SADB_64TO8(
948 				extv[SADB_EXT_KEY_ENCRYPT]->sadb_ext_len));
949 		if (extv[SADB_EXT_KEY_AUTH] != NULL)
950 			bzero(extv[SADB_EXT_KEY_AUTH],
951 			    SADB_64TO8(
952 				extv[SADB_EXT_KEY_AUTH]->sadb_ext_len));
953 		if (flushmsg) {
954 			ks0dbg((
955 			    "keysock: Downwards flush/dump message failed!\n"));
956 			/* If this is true, I hold the perimeter. */
957 			keysock_flushdump--;
958 		}
959 		freemsg(mp);
960 		return;
961 	}
962 
963 	wrapper->b_datap->db_type = M_CTL;
964 	ksi = (keysock_in_t *)wrapper->b_rptr;
965 	ksi->ks_in_type = KEYSOCK_IN;
966 	ksi->ks_in_len = sizeof (keysock_in_t);
967 	if (extv[SADB_EXT_ADDRESS_SRC] != NULL)
968 		ksi->ks_in_srctype = KS_IN_ADDR_UNKNOWN;
969 	else ksi->ks_in_srctype = KS_IN_ADDR_NOTTHERE;
970 	if (extv[SADB_EXT_ADDRESS_DST] != NULL)
971 		ksi->ks_in_dsttype = KS_IN_ADDR_UNKNOWN;
972 	else ksi->ks_in_dsttype = KS_IN_ADDR_NOTTHERE;
973 	if (extv[SADB_EXT_ADDRESS_PROXY] != NULL)
974 		ksi->ks_in_proxytype = KS_IN_ADDR_UNKNOWN;
975 	else ksi->ks_in_proxytype = KS_IN_ADDR_NOTTHERE;
976 	for (i = 0; i <= SADB_EXT_MAX; i++)
977 		ksi->ks_in_extv[i] = extv[i];
978 	ksi->ks_in_serial = ks->keysock_serial;
979 	wrapper->b_wptr += sizeof (ipsec_info_t);
980 	wrapper->b_cont = mp;
981 
982 	/*
983 	 * Find the appropriate consumer where the message is passed down.
984 	 */
985 	kc = keysock_consumers[satype];
986 	if (kc == NULL) {
987 		freeb(wrapper);
988 		keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_UNKNOWN_SATYPE);
989 		if (flushmsg) {
990 			ks0dbg((
991 			    "keysock: Downwards flush/dump message failed!\n"));
992 			/* If this is true, I hold the perimeter. */
993 			keysock_flushdump--;
994 		}
995 		return;
996 	}
997 
998 	/*
999 	 * NOTE: There used to be code in here to spin while a flush or
1000 	 *	 dump finished.  Keysock now assumes that consumers have enough
1001 	 *	 MT-savviness to deal with that.
1002 	 */
1003 
1004 	/*
1005 	 * Current consumers (AH and ESP) are guaranteed to return a
1006 	 * FLUSH or DUMP message back, so when we reach here, we don't
1007 	 * have to worry about keysock_flushdumps.
1008 	 */
1009 
1010 	putnext(kc->kc_wq, wrapper);
1011 }
1012 
1013 /*
1014  * High-level reality checking of extensions.
1015  */
1016 static boolean_t
1017 ext_check(sadb_ext_t *ext)
1018 {
1019 	int i;
1020 	uint64_t *lp;
1021 	sadb_ident_t *id;
1022 	char *idstr;
1023 
1024 	switch (ext->sadb_ext_type) {
1025 	case SADB_EXT_ADDRESS_SRC:
1026 	case SADB_EXT_ADDRESS_DST:
1027 	case SADB_EXT_ADDRESS_PROXY:
1028 		/* Check for at least enough addtl length for a sockaddr. */
1029 		if (ext->sadb_ext_len <= SADB_8TO64(sizeof (sadb_address_t)))
1030 			return (B_FALSE);
1031 		break;
1032 	case SADB_EXT_LIFETIME_HARD:
1033 	case SADB_EXT_LIFETIME_SOFT:
1034 	case SADB_EXT_LIFETIME_CURRENT:
1035 		if (ext->sadb_ext_len != SADB_8TO64(sizeof (sadb_lifetime_t)))
1036 			return (B_FALSE);
1037 		break;
1038 	case SADB_EXT_SPIRANGE:
1039 		/* See if the SPI range is legit. */
1040 		if (htonl(((sadb_spirange_t *)ext)->sadb_spirange_min) >
1041 		    htonl(((sadb_spirange_t *)ext)->sadb_spirange_max))
1042 			return (B_FALSE);
1043 		break;
1044 	case SADB_EXT_KEY_AUTH:
1045 	case SADB_EXT_KEY_ENCRYPT:
1046 		/* Key length check. */
1047 		if (((sadb_key_t *)ext)->sadb_key_bits == 0)
1048 			return (B_FALSE);
1049 		/*
1050 		 * Check to see if the key length (in bits) is less than the
1051 		 * extension length (in 8-bits words).
1052 		 */
1053 		if ((roundup(SADB_1TO8(((sadb_key_t *)ext)->sadb_key_bits), 8) +
1054 		    sizeof (sadb_key_t)) != SADB_64TO8(ext->sadb_ext_len)) {
1055 			ks1dbg((
1056 			    "ext_check:  Key bits/length inconsistent.\n"));
1057 			ks1dbg(("%d bits, len is %d bytes.\n",
1058 			    ((sadb_key_t *)ext)->sadb_key_bits,
1059 			    SADB_64TO8(ext->sadb_ext_len)));
1060 			return (B_FALSE);
1061 		}
1062 
1063 		/* All-zeroes key check. */
1064 		lp = (uint64_t *)(((char *)ext) + sizeof (sadb_key_t));
1065 		for (i = 0;
1066 		    i < (ext->sadb_ext_len - SADB_8TO64(sizeof (sadb_key_t)));
1067 		    i++)
1068 			if (lp[i] != 0)
1069 				break;	/* Out of for loop. */
1070 		/* If finished the loop naturally, it's an all zero key. */
1071 		if (lp[i] == 0)
1072 			return (B_FALSE);
1073 		break;
1074 	case SADB_EXT_IDENTITY_SRC:
1075 	case SADB_EXT_IDENTITY_DST:
1076 		/*
1077 		 * Make sure the strings in these identities are
1078 		 * null-terminated.  RFC 2367 underspecified how to handle
1079 		 * such a case.  I "proactively" null-terminate the string
1080 		 * at the last byte if it's not terminated sooner.
1081 		 */
1082 		id = (sadb_ident_t *)ext;
1083 		i = SADB_64TO8(id->sadb_ident_len);
1084 		i -= sizeof (sadb_ident_t);
1085 		idstr = (char *)(id + 1);
1086 		while (*idstr != '\0' && i > 0) {
1087 			i--;
1088 			idstr++;
1089 		}
1090 		if (i == 0) {
1091 			/*
1092 			 * I.e., if the bozo user didn't NULL-terminate the
1093 			 * string...
1094 			 */
1095 			idstr--;
1096 			*idstr = '\0';
1097 		}
1098 		break;
1099 	}
1100 	return (B_TRUE);	/* For now... */
1101 }
1102 
1103 /* Return values for keysock_get_ext(). */
1104 #define	KGE_OK	0
1105 #define	KGE_DUP	1
1106 #define	KGE_UNK	2
1107 #define	KGE_LEN	3
1108 #define	KGE_CHK	4
1109 
1110 /*
1111  * Parse basic extension headers and return in the passed-in pointer vector.
1112  * Return values include:
1113  *
1114  *	KGE_OK	Everything's nice and parsed out.
1115  *		If there are no extensions, place NULL in extv[0].
1116  *	KGE_DUP	There is a duplicate extension.
1117  *		First instance in appropriate bin.  First duplicate in
1118  *		extv[0].
1119  *	KGE_UNK	Unknown extension type encountered.  extv[0] contains
1120  *		unknown header.
1121  *	KGE_LEN	Extension length error.
1122  *	KGE_CHK	High-level reality check failed on specific extension.
1123  *
1124  * My apologies for some of the pointer arithmetic in here.  I'm thinking
1125  * like an assembly programmer, yet trying to make the compiler happy.
1126  */
1127 static int
1128 keysock_get_ext(sadb_ext_t *extv[], sadb_msg_t *basehdr, uint_t msgsize)
1129 {
1130 	bzero(extv, sizeof (sadb_ext_t *) * (SADB_EXT_MAX + 1));
1131 
1132 	/* Use extv[0] as the "current working pointer". */
1133 
1134 	extv[0] = (sadb_ext_t *)(basehdr + 1);
1135 
1136 	while (extv[0] < (sadb_ext_t *)(((uint8_t *)basehdr) + msgsize)) {
1137 		/* Check for unknown headers. */
1138 		if (extv[0]->sadb_ext_type == 0 ||
1139 		    extv[0]->sadb_ext_type > SADB_EXT_MAX)
1140 			return (KGE_UNK);
1141 
1142 		/*
1143 		 * Check length.  Use uint64_t because extlen is in units
1144 		 * of 64-bit words.  If length goes beyond the msgsize,
1145 		 * return an error.  (Zero length also qualifies here.)
1146 		 */
1147 		if (extv[0]->sadb_ext_len == 0 ||
1148 		    (void *)((uint64_t *)extv[0] + extv[0]->sadb_ext_len) >
1149 		    (void *)((uint8_t *)basehdr + msgsize))
1150 			return (KGE_LEN);
1151 
1152 		/* Check for redundant headers. */
1153 		if (extv[extv[0]->sadb_ext_type] != NULL)
1154 			return (KGE_DUP);
1155 
1156 		/*
1157 		 * Reality check the extension if possible at the keysock
1158 		 * level.
1159 		 */
1160 		if (!ext_check(extv[0]))
1161 			return (KGE_CHK);
1162 
1163 		/* If I make it here, assign the appropriate bin. */
1164 		extv[extv[0]->sadb_ext_type] = extv[0];
1165 
1166 		/* Advance pointer (See above for uint64_t ptr reasoning.) */
1167 		extv[0] = (sadb_ext_t *)
1168 		    ((uint64_t *)extv[0] + extv[0]->sadb_ext_len);
1169 	}
1170 
1171 	/* Everything's cool. */
1172 
1173 	/*
1174 	 * If extv[0] == NULL, then there are no extension headers in this
1175 	 * message.  Ensure that this is the case.
1176 	 */
1177 	if (extv[0] == (sadb_ext_t *)(basehdr + 1))
1178 		extv[0] = NULL;
1179 
1180 	return (KGE_OK);
1181 }
1182 
1183 /*
1184  * qwriter() callback to handle flushes and dumps.  This routine will hold
1185  * the inner perimeter.
1186  */
1187 void
1188 keysock_do_flushdump(queue_t *q, mblk_t *mp)
1189 {
1190 	int i, start, finish;
1191 	mblk_t *mp1 = NULL;
1192 	keysock_t *ks = (keysock_t *)q->q_ptr;
1193 	sadb_ext_t *extv[SADB_EXT_MAX + 1];
1194 	sadb_msg_t *samsg = (sadb_msg_t *)mp->b_rptr;
1195 
1196 	/*
1197 	 * I am guaranteed this will work.  I did the work in keysock_parse()
1198 	 * already.
1199 	 */
1200 	(void) keysock_get_ext(extv, samsg, SADB_64TO8(samsg->sadb_msg_len));
1201 
1202 	/*
1203 	 * I hold the perimeter, therefore I don't need to use atomic ops.
1204 	 */
1205 	if (keysock_flushdump != 0) {
1206 		/* XXX Should I instead use EBUSY? */
1207 		/* XXX Or is there a way to queue these up? */
1208 		keysock_error(ks, mp, ENOMEM, SADB_X_DIAGNOSTIC_NONE);
1209 		return;
1210 	}
1211 
1212 	if (samsg->sadb_msg_satype == SADB_SATYPE_UNSPEC) {
1213 		start = 0;
1214 		finish = KEYSOCK_MAX_CONSUMERS - 1;
1215 	} else {
1216 		start = samsg->sadb_msg_satype;
1217 		finish = samsg->sadb_msg_satype;
1218 	}
1219 
1220 	/*
1221 	 * Fill up keysock_flushdump with the number of outstanding dumps
1222 	 * and/or flushes.
1223 	 */
1224 
1225 	keysock_flushdump_errno = 0;
1226 
1227 	/*
1228 	 * Okay, I hold the perimeter.  Eventually keysock_flushdump will
1229 	 * contain the number of consumers with outstanding flush operations.
1230 	 *
1231 	 * SO, here's the plan:
1232 	 *	* For each relevant consumer (Might be one, might be all)
1233 	 *		* Twiddle on the FLUSHING flag.
1234 	 *		* Pass down the FLUSH/DUMP message.
1235 	 *
1236 	 * When I see upbound FLUSH/DUMP messages, I will decrement the
1237 	 * keysock_flushdump.  When I decrement it to 0, I will pass the
1238 	 * FLUSH/DUMP message back up to the PF_KEY sockets.  Because I will
1239 	 * pass down the right SA type to the consumer (either its own, or
1240 	 * that of UNSPEC), the right one will be reflected from each consumer,
1241 	 * and accordingly back to the socket.
1242 	 */
1243 
1244 	mutex_enter(&keysock_consumers_lock);
1245 	for (i = start; i <= finish; i++) {
1246 		if (keysock_consumers[i] != NULL) {
1247 			mp1 = copymsg(mp);
1248 			if (mp1 == NULL) {
1249 				ks0dbg(("SADB_FLUSH copymsg() failed.\n"));
1250 				/*
1251 				 * Error?  And what about outstanding
1252 				 * flushes?  Oh, yeah, they get sucked up and
1253 				 * the counter is decremented.  Consumers
1254 				 * (see keysock_passdown()) are guaranteed
1255 				 * to deliver back a flush request, even if
1256 				 * it's an error.
1257 				 */
1258 				keysock_error(ks, mp, ENOMEM,
1259 				    SADB_X_DIAGNOSTIC_NONE);
1260 				return;
1261 			}
1262 			/*
1263 			 * Because my entry conditions are met above, the
1264 			 * following assertion should hold true.
1265 			 */
1266 			mutex_enter(&(keysock_consumers[i]->kc_lock));
1267 			ASSERT((keysock_consumers[i]->kc_flags & KC_FLUSHING)
1268 			    == 0);
1269 			keysock_consumers[i]->kc_flags |= KC_FLUSHING;
1270 			mutex_exit(&(keysock_consumers[i]->kc_lock));
1271 			/* Always increment the number of flushes... */
1272 			keysock_flushdump++;
1273 			/* Guaranteed to return a message. */
1274 			keysock_passdown(ks, mp1, i, extv, B_TRUE);
1275 		} else if (start == finish) {
1276 			/*
1277 			 * In case where start == finish, and there's no
1278 			 * consumer, should we force an error?  Yes.
1279 			 */
1280 			mutex_exit(&keysock_consumers_lock);
1281 			keysock_error(ks, mp, EINVAL,
1282 			    SADB_X_DIAGNOSTIC_UNKNOWN_SATYPE);
1283 			return;
1284 		}
1285 	}
1286 	mutex_exit(&keysock_consumers_lock);
1287 
1288 	if (keysock_flushdump == 0) {
1289 		/*
1290 		 * There were no consumers at all for this message.
1291 		 * XXX For now return ESRCH.
1292 		 */
1293 		keysock_error(ks, mp, ESRCH, SADB_X_DIAGNOSTIC_NO_SADBS);
1294 	} else {
1295 		/* Otherwise, free the original message. */
1296 		freemsg(mp);
1297 	}
1298 }
1299 
1300 /*
1301  * Get the right diagnostic for a duplicate.  Should probably use a static
1302  * table lookup.
1303  */
1304 int
1305 keysock_duplicate(int ext_type)
1306 {
1307 	int rc = 0;
1308 
1309 	switch (ext_type) {
1310 	case SADB_EXT_ADDRESS_SRC:
1311 		rc = SADB_X_DIAGNOSTIC_DUPLICATE_SRC;
1312 		break;
1313 	case SADB_EXT_ADDRESS_DST:
1314 		rc = SADB_X_DIAGNOSTIC_DUPLICATE_DST;
1315 		break;
1316 	case SADB_EXT_SA:
1317 		rc = SADB_X_DIAGNOSTIC_DUPLICATE_SA;
1318 		break;
1319 	case SADB_EXT_SPIRANGE:
1320 		rc = SADB_X_DIAGNOSTIC_DUPLICATE_RANGE;
1321 		break;
1322 	case SADB_EXT_KEY_AUTH:
1323 		rc = SADB_X_DIAGNOSTIC_DUPLICATE_AKEY;
1324 		break;
1325 	case SADB_EXT_KEY_ENCRYPT:
1326 		rc = SADB_X_DIAGNOSTIC_DUPLICATE_EKEY;
1327 		break;
1328 	}
1329 	return (rc);
1330 }
1331 
1332 /*
1333  * Get the right diagnostic for a reality check failure.  Should probably use
1334  * a static table lookup.
1335  */
1336 int
1337 keysock_malformed(int ext_type)
1338 {
1339 	int rc = 0;
1340 
1341 	switch (ext_type) {
1342 	case SADB_EXT_ADDRESS_SRC:
1343 		rc = SADB_X_DIAGNOSTIC_MALFORMED_SRC;
1344 		break;
1345 	case SADB_EXT_ADDRESS_DST:
1346 		rc = SADB_X_DIAGNOSTIC_MALFORMED_DST;
1347 		break;
1348 	case SADB_EXT_SA:
1349 		rc = SADB_X_DIAGNOSTIC_MALFORMED_SA;
1350 		break;
1351 	case SADB_EXT_SPIRANGE:
1352 		rc = SADB_X_DIAGNOSTIC_MALFORMED_RANGE;
1353 		break;
1354 	case SADB_EXT_KEY_AUTH:
1355 		rc = SADB_X_DIAGNOSTIC_MALFORMED_AKEY;
1356 		break;
1357 	case SADB_EXT_KEY_ENCRYPT:
1358 		rc = SADB_X_DIAGNOSTIC_MALFORMED_EKEY;
1359 		break;
1360 	}
1361 	return (rc);
1362 }
1363 
1364 /*
1365  * Keysock massaging of an inverse ACQUIRE.  Consult policy,
1366  * and construct an appropriate response.
1367  */
1368 static void
1369 keysock_inverse_acquire(mblk_t *mp, sadb_msg_t *samsg, sadb_ext_t *extv[],
1370     keysock_t *ks)
1371 {
1372 	mblk_t *reply_mp;
1373 
1374 	/*
1375 	 * Reality check things...
1376 	 */
1377 	if (extv[SADB_EXT_ADDRESS_SRC] == NULL) {
1378 		keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_MISSING_SRC);
1379 		return;
1380 	}
1381 	if (extv[SADB_EXT_ADDRESS_DST] == NULL) {
1382 		keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_MISSING_DST);
1383 	}
1384 
1385 	reply_mp = ipsec_construct_inverse_acquire(samsg, extv);
1386 
1387 	if (reply_mp != NULL) {
1388 		freemsg(mp);
1389 		keysock_passup(reply_mp, (sadb_msg_t *)reply_mp->b_rptr,
1390 		    ks->keysock_serial, NULL, B_FALSE);
1391 	} else {
1392 		keysock_error(ks, mp, samsg->sadb_msg_errno,
1393 		    samsg->sadb_x_msg_diagnostic);
1394 	}
1395 }
1396 
1397 /*
1398  * Spew an extended REGISTER down to the relevant consumers.
1399  */
1400 static void
1401 keysock_extended_register(keysock_t *ks, mblk_t *mp, sadb_ext_t *extv[])
1402 {
1403 	sadb_x_ereg_t *ereg = (sadb_x_ereg_t *)extv[SADB_X_EXT_EREG];
1404 	uint8_t *satypes, *fencepost;
1405 	mblk_t *downmp;
1406 	sadb_ext_t *downextv[SADB_EXT_MAX + 1];
1407 
1408 	if (ks->keysock_registered[0] != 0 || ks->keysock_registered[1] != 0 ||
1409 	    ks->keysock_registered[2] != 0 || ks->keysock_registered[3] != 0) {
1410 		keysock_error(ks, mp, EBUSY, 0);
1411 	}
1412 
1413 	ks->keysock_flags |= KEYSOCK_EXTENDED;
1414 	if (ereg == NULL) {
1415 		keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_SATYPE_NEEDED);
1416 	} else {
1417 		ASSERT(mp->b_rptr + msgdsize(mp) == mp->b_wptr);
1418 		fencepost = (uint8_t *)mp->b_wptr;
1419 		satypes = ereg->sadb_x_ereg_satypes;
1420 		while (*satypes != SADB_SATYPE_UNSPEC && satypes != fencepost) {
1421 			downmp = copymsg(mp);
1422 			if (downmp == NULL) {
1423 				keysock_error(ks, mp, ENOMEM, 0);
1424 				return;
1425 			}
1426 			/*
1427 			 * Since we've made it here, keysock_get_ext will work!
1428 			 */
1429 			(void) keysock_get_ext(downextv,
1430 			    (sadb_msg_t *)downmp->b_rptr, msgdsize(downmp));
1431 			keysock_passdown(ks, downmp, *satypes, downextv,
1432 			    B_FALSE);
1433 			++satypes;
1434 		}
1435 		freemsg(mp);
1436 	}
1437 
1438 	/*
1439 	 * Set global to indicate we prefer an extended ACQUIRE.
1440 	 */
1441 	atomic_add_32(&keysock_num_extended, 1);
1442 }
1443 
1444 /*
1445  * Handle PF_KEY messages.
1446  */
1447 static void
1448 keysock_parse(queue_t *q, mblk_t *mp)
1449 {
1450 	sadb_msg_t *samsg;
1451 	sadb_ext_t *extv[SADB_EXT_MAX + 1];
1452 	keysock_t *ks = (keysock_t *)q->q_ptr;
1453 	uint_t msgsize;
1454 	uint8_t satype;
1455 
1456 	/* Make sure I'm a PF_KEY socket.  (i.e. nothing's below me) */
1457 	ASSERT(WR(q)->q_next == NULL);
1458 
1459 	samsg = (sadb_msg_t *)mp->b_rptr;
1460 	ks2dbg(("Received possible PF_KEY message, type %d.\n",
1461 	    samsg->sadb_msg_type));
1462 
1463 	msgsize = SADB_64TO8(samsg->sadb_msg_len);
1464 
1465 	if (msgdsize(mp) != msgsize) {
1466 		/*
1467 		 * Message len incorrect w.r.t. actual size.  Send an error
1468 		 * (EMSGSIZE).	It may be necessary to massage things a
1469 		 * bit.	 For example, if the sadb_msg_type is hosed,
1470 		 * I need to set it to SADB_RESERVED to get delivery to
1471 		 * do the right thing.	Then again, maybe just letting
1472 		 * the error delivery do the right thing.
1473 		 */
1474 		ks2dbg(("mblk (%lu) and base (%d) message sizes don't jibe.\n",
1475 		    msgdsize(mp), msgsize));
1476 		keysock_error(ks, mp, EMSGSIZE, SADB_X_DIAGNOSTIC_NONE);
1477 		return;
1478 	}
1479 
1480 	if (msgsize > (uint_t)(mp->b_wptr - mp->b_rptr)) {
1481 		/* Get all message into one mblk. */
1482 		if (pullupmsg(mp, -1) == 0) {
1483 			/*
1484 			 * Something screwy happened.
1485 			 */
1486 			ks3dbg(("keysock_parse: pullupmsg() failed.\n"));
1487 			return;
1488 		} else {
1489 			samsg = (sadb_msg_t *)mp->b_rptr;
1490 		}
1491 	}
1492 
1493 	switch (keysock_get_ext(extv, samsg, msgsize)) {
1494 	case KGE_DUP:
1495 		/* Handle duplicate extension. */
1496 		ks1dbg(("Got duplicate extension of type %d.\n",
1497 		    extv[0]->sadb_ext_type));
1498 		keysock_error(ks, mp, EINVAL,
1499 		    keysock_duplicate(extv[0]->sadb_ext_type));
1500 		return;
1501 	case KGE_UNK:
1502 		/* Handle unknown extension. */
1503 		ks1dbg(("Got unknown extension of type %d.\n",
1504 		    extv[0]->sadb_ext_type));
1505 		keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_UNKNOWN_EXT);
1506 		return;
1507 	case KGE_LEN:
1508 		/* Length error. */
1509 		ks1dbg(("Length %d on extension type %d overrun or 0.\n",
1510 		    extv[0]->sadb_ext_len, extv[0]->sadb_ext_type));
1511 		keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_BAD_EXTLEN);
1512 		return;
1513 	case KGE_CHK:
1514 		/* Reality check failed. */
1515 		ks1dbg(("Reality check failed on extension type %d.\n",
1516 		    extv[0]->sadb_ext_type));
1517 		keysock_error(ks, mp, EINVAL,
1518 		    keysock_malformed(extv[0]->sadb_ext_type));
1519 		return;
1520 	default:
1521 		/* Default case is no errors. */
1522 		break;
1523 	}
1524 
1525 	switch (samsg->sadb_msg_type) {
1526 	case SADB_REGISTER:
1527 		/*
1528 		 * There's a semantic weirdness in that a message OTHER than
1529 		 * the return REGISTER message may be passed up if I set the
1530 		 * registered bit BEFORE I pass it down.
1531 		 *
1532 		 * SOOOO, I'll not twiddle any registered bits until I see
1533 		 * the upbound REGISTER (with a serial number in it).
1534 		 */
1535 		if (samsg->sadb_msg_satype == SADB_SATYPE_UNSPEC) {
1536 			/* Handle extended register here. */
1537 			keysock_extended_register(ks, mp, extv);
1538 			return;
1539 		} else if (ks->keysock_flags & KEYSOCK_EXTENDED) {
1540 			keysock_error(ks, mp, EBUSY, 0);
1541 			return;
1542 		}
1543 		/* FALLTHRU */
1544 	case SADB_GETSPI:
1545 	case SADB_ADD:
1546 	case SADB_UPDATE:
1547 	case SADB_DELETE:
1548 	case SADB_GET:
1549 		/*
1550 		 * Pass down to appropriate consumer.
1551 		 */
1552 		if (samsg->sadb_msg_satype != SADB_SATYPE_UNSPEC)
1553 			keysock_passdown(ks, mp, samsg->sadb_msg_satype, extv,
1554 			    B_FALSE);
1555 		else keysock_error(ks, mp, EINVAL,
1556 		    SADB_X_DIAGNOSTIC_SATYPE_NEEDED);
1557 		return;
1558 	case SADB_ACQUIRE:
1559 		/*
1560 		 * If I _receive_ an acquire, this means I should spread it
1561 		 * out to registered sockets.  Unless there's an errno...
1562 		 *
1563 		 * Need ADDRESS, may have ID, SENS, and PROP, unless errno,
1564 		 * in which case there should be NO extensions.
1565 		 *
1566 		 * Return to registered.
1567 		 */
1568 		if (samsg->sadb_msg_errno != 0) {
1569 			satype = samsg->sadb_msg_satype;
1570 			if (satype == SADB_SATYPE_UNSPEC) {
1571 				if (!(ks->keysock_flags & KEYSOCK_EXTENDED)) {
1572 					keysock_error(ks, mp, EINVAL,
1573 					    SADB_X_DIAGNOSTIC_SATYPE_NEEDED);
1574 					return;
1575 				}
1576 				/*
1577 				 * Reassign satype based on the first
1578 				 * flags that KEYSOCK_SETREG says.
1579 				 */
1580 				while (satype <= SADB_SATYPE_MAX) {
1581 					if (KEYSOCK_ISREG(ks, satype))
1582 						break;
1583 					satype++;
1584 				}
1585 				if (satype > SADB_SATYPE_MAX) {
1586 					keysock_error(ks, mp, EBUSY, 0);
1587 					return;
1588 				}
1589 			}
1590 			keysock_passdown(ks, mp, satype, extv, B_FALSE);
1591 		} else {
1592 			if (samsg->sadb_msg_satype == SADB_SATYPE_UNSPEC)
1593 				keysock_error(ks, mp, EINVAL,
1594 				    SADB_X_DIAGNOSTIC_SATYPE_NEEDED);
1595 			else
1596 				keysock_passup(mp, samsg, 0, NULL, B_FALSE);
1597 		}
1598 		return;
1599 	case SADB_EXPIRE:
1600 		/*
1601 		 * If someone sends this in, then send out to all senders.
1602 		 * (Save maybe ESP or AH, I have to be careful here.)
1603 		 *
1604 		 * Need ADDRESS, may have ID and SENS.
1605 		 *
1606 		 * XXX for now this is unsupported.
1607 		 */
1608 		break;
1609 	case SADB_FLUSH:
1610 	case SADB_DUMP:	 /* not used by normal applications */
1611 		/*
1612 		 * Nuke all SAs, or dump out the whole SA table to sender only.
1613 		 *
1614 		 * No extensions at all.  Return to all listeners.
1615 		 *
1616 		 * Question:	Should I hold a lock here to prevent
1617 		 *		additions/deletions while flushing?
1618 		 * Answer:	No.  (See keysock_passdown() for details.)
1619 		 */
1620 		if (extv[0] != NULL) {
1621 			/*
1622 			 * FLUSH or DUMP messages shouldn't have extensions.
1623 			 * Return EINVAL.
1624 			 */
1625 			ks2dbg(("FLUSH message with extension.\n"));
1626 			keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_NO_EXT);
1627 			return;
1628 		}
1629 
1630 		/* Passing down of DUMP/FLUSH messages are special. */
1631 		qwriter(q, mp, keysock_do_flushdump, PERIM_INNER);
1632 		return;
1633 	case SADB_X_PROMISC:
1634 		/*
1635 		 * Promiscuous processing message.
1636 		 */
1637 		if (samsg->sadb_msg_satype == 0)
1638 			ks->keysock_flags &= ~KEYSOCK_PROMISC;
1639 		else
1640 			ks->keysock_flags |= KEYSOCK_PROMISC;
1641 		keysock_passup(mp, samsg, ks->keysock_serial, NULL, B_FALSE);
1642 		return;
1643 	case SADB_X_INVERSE_ACQUIRE:
1644 		keysock_inverse_acquire(mp, samsg, extv, ks);
1645 		return;
1646 	default:
1647 		ks2dbg(("Got unknown message type %d.\n",
1648 		    samsg->sadb_msg_type));
1649 		keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_UNKNOWN_MSG);
1650 		return;
1651 	}
1652 
1653 	/* As a placeholder... */
1654 	ks0dbg(("keysock_parse():  Hit EOPNOTSUPP\n"));
1655 	keysock_error(ks, mp, EOPNOTSUPP, SADB_X_DIAGNOSTIC_NONE);
1656 }
1657 
1658 /*
1659  * wput routing for PF_KEY/keysock/whatever.  Unlike the routing socket,
1660  * I don't convert to ioctl()'s for IP.  I am the end-all driver as far
1661  * as PF_KEY sockets are concerned.  I do some conversion, but not as much
1662  * as IP/rts does.
1663  */
1664 static void
1665 keysock_wput(queue_t *q, mblk_t *mp)
1666 {
1667 	uchar_t *rptr = mp->b_rptr;
1668 	mblk_t *mp1;
1669 
1670 	ks3dbg(("In keysock_wput\n"));
1671 
1672 	if (WR(q)->q_next) {
1673 		keysock_consumer_t *kc = (keysock_consumer_t *)q->q_ptr;
1674 
1675 		/*
1676 		 * We shouldn't get writes on a consumer instance.
1677 		 * But for now, just passthru.
1678 		 */
1679 		ks1dbg(("Huh?  wput for an consumer instance (%d)?\n",
1680 		    kc->kc_sa_type));
1681 		putnext(q, mp);
1682 		return;
1683 	}
1684 
1685 	switch (mp->b_datap->db_type) {
1686 	case M_DATA:
1687 		/*
1688 		 * Silently discard.
1689 		 */
1690 		ks2dbg(("raw M_DATA in keysock.\n"));
1691 		freemsg(mp);
1692 		return;
1693 	case M_PROTO:
1694 	case M_PCPROTO:
1695 		if ((mp->b_wptr - rptr) >= sizeof (struct T_data_req)) {
1696 			if (((union T_primitives *)rptr)->type == T_DATA_REQ) {
1697 				if ((mp1 = mp->b_cont) == NULL) {
1698 					/* No data after T_DATA_REQ. */
1699 					ks2dbg(("No data after DATA_REQ.\n"));
1700 					freemsg(mp);
1701 					return;
1702 				}
1703 				freeb(mp);
1704 				mp = mp1;
1705 				ks2dbg(("T_DATA_REQ\n"));
1706 				break;	/* Out of switch. */
1707 			}
1708 		}
1709 		/* FALLTHRU */
1710 	default:
1711 		ks3dbg(("In default wput case (%d %d).\n",
1712 		    mp->b_datap->db_type, ((union T_primitives *)rptr)->type));
1713 		keysock_wput_other(q, mp);
1714 		return;
1715 	}
1716 
1717 	/* I now have a PF_KEY message in an M_DATA block, pointed to by mp. */
1718 	keysock_parse(q, mp);
1719 }
1720 
1721 /* BELOW THIS LINE ARE ROUTINES INCLUDING AND RELATED TO keysock_rput(). */
1722 
1723 /*
1724  * Called upon receipt of a KEYSOCK_HELLO_ACK to set up the appropriate
1725  * state vectors.
1726  */
1727 static void
1728 keysock_link_consumer(uint8_t satype, keysock_consumer_t *kc)
1729 {
1730 	keysock_t *ks;
1731 
1732 	mutex_enter(&keysock_consumers_lock);
1733 	mutex_enter(&kc->kc_lock);
1734 	if (keysock_consumers[satype] != NULL) {
1735 		ks0dbg((
1736 		    "Hmmmm, someone closed %d before the HELLO_ACK happened.\n",
1737 		    satype));
1738 		/*
1739 		 * Perhaps updating the new below-me consumer with what I have
1740 		 * so far would work too?
1741 		 */
1742 		mutex_exit(&kc->kc_lock);
1743 		mutex_exit(&keysock_consumers_lock);
1744 	} else {
1745 		/* Add new below-me consumer. */
1746 		keysock_consumers[satype] = kc;
1747 
1748 		kc->kc_flags = 0;
1749 		kc->kc_sa_type = satype;
1750 		mutex_exit(&kc->kc_lock);
1751 		mutex_exit(&keysock_consumers_lock);
1752 
1753 		/* Scan the keysock list. */
1754 		mutex_enter(&keysock_list_lock);
1755 		for (ks = keysock_list; ks != NULL; ks = ks->keysock_next) {
1756 			if (KEYSOCK_ISREG(ks, satype)) {
1757 				/*
1758 				 * XXX Perhaps send an SADB_REGISTER down on
1759 				 * the socket's behalf.
1760 				 */
1761 				ks1dbg(("Socket %u registered already for "
1762 				    "new consumer.\n", ks->keysock_serial));
1763 			}
1764 		}
1765 		mutex_exit(&keysock_list_lock);
1766 	}
1767 }
1768 
1769 /*
1770  * Generate a KEYSOCK_OUT_ERR message for my consumer.
1771  */
1772 static void
1773 keysock_out_err(keysock_consumer_t *kc, int ks_errno, mblk_t *mp)
1774 {
1775 	keysock_out_err_t *kse;
1776 	mblk_t *imp;
1777 
1778 	imp = allocb(sizeof (ipsec_info_t), BPRI_HI);
1779 	if (imp == NULL) {
1780 		ks1dbg(("keysock_out_err:  Can't alloc message.\n"));
1781 		return;
1782 	}
1783 
1784 	imp->b_datap->db_type = M_CTL;
1785 	imp->b_wptr += sizeof (ipsec_info_t);
1786 
1787 	kse = (keysock_out_err_t *)imp->b_rptr;
1788 	imp->b_cont = mp;
1789 	kse->ks_err_type = KEYSOCK_OUT_ERR;
1790 	kse->ks_err_len = sizeof (*kse);
1791 	/* Is serial necessary? */
1792 	kse->ks_err_serial = 0;
1793 	kse->ks_err_errno = ks_errno;
1794 
1795 	/*
1796 	 * XXX What else do I need to do here w.r.t. information
1797 	 * to tell the consumer what caused this error?
1798 	 *
1799 	 * I believe the answer is the PF_KEY ACQUIRE (or other) message
1800 	 * attached in mp, which is appended at the end.  I believe the
1801 	 * db_ref won't matter here, because the PF_KEY message is only read
1802 	 * for KEYSOCK_OUT_ERR.
1803 	 */
1804 
1805 	putnext(kc->kc_wq, imp);
1806 }
1807 
1808 /* XXX this is a hack errno. */
1809 #define	EIPSECNOSA 255
1810 
1811 /*
1812  * Route message (pointed by mp, header in samsg) toward appropriate
1813  * sockets.  Assume the message's creator did its job correctly.
1814  *
1815  * This should be a function that is followed by a return in its caller.
1816  * The compiler _should_ be able to use tail-call optimizations to make the
1817  * large ## of parameters not a huge deal.
1818  */
1819 static void
1820 keysock_passup(mblk_t *mp, sadb_msg_t *samsg, minor_t serial,
1821     keysock_consumer_t *kc, boolean_t persistent)
1822 {
1823 	keysock_t *ks;
1824 	uint8_t satype = samsg->sadb_msg_satype;
1825 	boolean_t toall = B_FALSE, allreg = B_FALSE, allereg = B_FALSE,
1826 	    setalg = B_FALSE;
1827 	mblk_t *mp1;
1828 	int err = EIPSECNOSA;
1829 
1830 	/* Convert mp, which is M_DATA, into an M_PROTO of type T_DATA_IND */
1831 	mp1 = allocb(sizeof (struct T_data_req), BPRI_HI);
1832 	if (mp1 == NULL) {
1833 		err = ENOMEM;
1834 		goto error;
1835 	}
1836 	mp1->b_wptr += sizeof (struct T_data_req);
1837 	((struct T_data_ind *)mp1->b_rptr)->PRIM_type = T_DATA_IND;
1838 	((struct T_data_ind *)mp1->b_rptr)->MORE_flag = 0;
1839 	mp1->b_datap->db_type = M_PROTO;
1840 	mp1->b_cont = mp;
1841 	mp = mp1;
1842 
1843 	switch (samsg->sadb_msg_type) {
1844 	case SADB_FLUSH:
1845 	case SADB_GETSPI:
1846 	case SADB_UPDATE:
1847 	case SADB_ADD:
1848 	case SADB_DELETE:
1849 	case SADB_EXPIRE:
1850 		/*
1851 		 * These are most likely replies.  Don't worry about
1852 		 * KEYSOCK_OUT_ERR handling.  Deliver to all sockets.
1853 		 */
1854 		ks3dbg(("Delivering normal message (%d) to all sockets.\n",
1855 		    samsg->sadb_msg_type));
1856 		toall = B_TRUE;
1857 		break;
1858 	case SADB_REGISTER:
1859 		/*
1860 		 * REGISTERs come up for one of three reasons:
1861 		 *
1862 		 *	1.) In response to a normal SADB_REGISTER
1863 		 *		(samsg->sadb_msg_satype != SADB_SATYPE_UNSPEC &&
1864 		 *		    serial != 0)
1865 		 *		Deliver to normal SADB_REGISTERed sockets.
1866 		 *	2.) In response to an extended REGISTER
1867 		 *		(samsg->sadb_msg_satype == SADB_SATYPE_UNSPEC)
1868 		 *		Deliver to extended REGISTERed socket.
1869 		 *	3.) Spontaneous algorithm changes
1870 		 *		(samsg->sadb_msg_satype != SADB_SATYPE_UNSPEC &&
1871 		 *		    serial == 0)
1872 		 *		Deliver to REGISTERed sockets of all sorts.
1873 		 */
1874 		if (kc == NULL) {
1875 			/* Here because of keysock_error() call. */
1876 			ASSERT(samsg->sadb_msg_errno != 0);
1877 			break;	/* Out of switch. */
1878 		}
1879 		ks3dbg(("Delivering REGISTER.\n"));
1880 		if (satype == SADB_SATYPE_UNSPEC) {
1881 			/* REGISTER Reason #2 */
1882 			allereg = B_TRUE;
1883 			/*
1884 			 * Rewhack SA type so PF_KEY socket holder knows what
1885 			 * consumer generated this algorithm list.
1886 			 */
1887 			satype = kc->kc_sa_type;
1888 			samsg->sadb_msg_satype = satype;
1889 			setalg = B_TRUE;
1890 		} else if (serial == 0) {
1891 			/* REGISTER Reason #3 */
1892 			allreg = B_TRUE;
1893 			allereg = B_TRUE;
1894 		} else {
1895 			/* REGISTER Reason #1 */
1896 			allreg = B_TRUE;
1897 			setalg = B_TRUE;
1898 		}
1899 		break;
1900 	case SADB_ACQUIRE:
1901 		/*
1902 		 * ACQUIREs are either extended (sadb_msg_satype == 0) or
1903 		 * regular (sadb_msg_satype != 0).  And we're guaranteed
1904 		 * that serial == 0 for an ACQUIRE.
1905 		 */
1906 		ks3dbg(("Delivering ACQUIRE.\n"));
1907 		allereg = (satype == SADB_SATYPE_UNSPEC);
1908 		allreg = !allereg;
1909 		/*
1910 		 * Corner case - if we send a regular ACQUIRE and there's
1911 		 * extended ones registered, don't send an error down to
1912 		 * consumers if nobody's listening and prematurely destroy
1913 		 * their ACQUIRE record.  This might be too hackish of a
1914 		 * solution.
1915 		 */
1916 		if (allreg && keysock_num_extended > 0)
1917 			err = 0;
1918 		break;
1919 	case SADB_X_PROMISC:
1920 	case SADB_X_INVERSE_ACQUIRE:
1921 	case SADB_DUMP:
1922 	case SADB_GET:
1923 	default:
1924 		/*
1925 		 * Deliver to the sender and promiscuous only.
1926 		 */
1927 		ks3dbg(("Delivering sender/promisc only (%d).\n",
1928 		    samsg->sadb_msg_type));
1929 		break;
1930 	}
1931 
1932 	mutex_enter(&keysock_list_lock);
1933 	for (ks = keysock_list; ks != NULL; ks = ks->keysock_next) {
1934 		/* Delivery loop. */
1935 
1936 		/*
1937 		 * Check special keysock-setting cases (REGISTER replies)
1938 		 * here.
1939 		 */
1940 		if (setalg && serial == ks->keysock_serial) {
1941 			ASSERT(kc != NULL);
1942 			ASSERT(kc->kc_sa_type == satype);
1943 			KEYSOCK_SETREG(ks, satype);
1944 		}
1945 
1946 		/*
1947 		 * NOLOOP takes precedence over PROMISC.  So if you've set
1948 		 * !SO_USELOOPBACK, don't expect to see any data...
1949 		 */
1950 		if (ks->keysock_flags & KEYSOCK_NOLOOP)
1951 			continue;
1952 
1953 		/*
1954 		 * Messages to all, or promiscuous sockets just GET the
1955 		 * message.  Perform rules-type checking iff it's not for all
1956 		 * listeners or the socket is in promiscuous mode.
1957 		 *
1958 		 * NOTE:Because of the (kc != NULL && ISREG()), make sure
1959 		 *	extended ACQUIREs arrive off a consumer that is
1960 		 *	part of the extended REGISTER set of consumers.
1961 		 */
1962 		if (serial != ks->keysock_serial &&
1963 		    !toall &&
1964 		    !(ks->keysock_flags & KEYSOCK_PROMISC) &&
1965 		    !((ks->keysock_flags & KEYSOCK_EXTENDED) ?
1966 			allereg : allreg && kc != NULL &&
1967 			KEYSOCK_ISREG(ks, kc->kc_sa_type)))
1968 			continue;
1969 
1970 		mp1 = dupmsg(mp);
1971 		if (mp1 == NULL) {
1972 			ks2dbg((
1973 			    "keysock_passup():  dupmsg() failed.\n"));
1974 			mp1 = mp;
1975 			mp = NULL;
1976 			err = ENOMEM;
1977 		}
1978 
1979 		/*
1980 		 * At this point, we can deliver or attempt to deliver
1981 		 * this message.  We're free of obligation to report
1982 		 * no listening PF_KEY sockets.  So set err to 0.
1983 		 */
1984 		err = 0;
1985 
1986 		/*
1987 		 * See if we canputnext(), as well as see if the message
1988 		 * needs to be queued if we can't.
1989 		 */
1990 		if (!canputnext(ks->keysock_rq)) {
1991 			if (persistent) {
1992 				if (putq(ks->keysock_rq, mp1) == 0) {
1993 					ks1dbg((
1994 					    "keysock_passup: putq failed.\n"));
1995 				} else {
1996 					continue;
1997 				}
1998 			}
1999 			freemsg(mp1);
2000 			continue;
2001 		}
2002 
2003 		ks3dbg(("Putting to serial %d.\n", ks->keysock_serial));
2004 		/*
2005 		 * Unlike the specific keysock instance case, this
2006 		 * will only hit for listeners, so we will only
2007 		 * putnext() if we can.
2008 		 */
2009 		putnext(ks->keysock_rq, mp1);
2010 		if (mp == NULL)
2011 			break;	/* out of for loop. */
2012 	}
2013 	mutex_exit(&keysock_list_lock);
2014 
2015 error:
2016 	if ((err != 0) && (kc != NULL)) {
2017 		/*
2018 		 * Generate KEYSOCK_OUT_ERR for consumer.
2019 		 * Basically, I send this back if I have not been able to
2020 		 * transmit (for whatever reason)
2021 		 */
2022 		ks1dbg(("keysock_passup():  No registered of type %d.\n",
2023 		    satype));
2024 		if (mp != NULL) {
2025 			if (mp->b_datap->db_type == M_PROTO) {
2026 				mp1 = mp;
2027 				mp = mp->b_cont;
2028 				freeb(mp1);
2029 			}
2030 			/*
2031 			 * Do a copymsg() because people who get
2032 			 * KEYSOCK_OUT_ERR may alter the message contents.
2033 			 */
2034 			mp1 = copymsg(mp);
2035 			if (mp1 == NULL) {
2036 				ks2dbg(("keysock_passup: copymsg() failed.\n"));
2037 				mp1 = mp;
2038 				mp = NULL;
2039 			}
2040 			keysock_out_err(kc, err, mp1);
2041 		}
2042 	}
2043 
2044 	/*
2045 	 * XXX Blank the message somehow.  This is difficult because we don't
2046 	 * know at this point if the message has db_ref > 1, etc.
2047 	 *
2048 	 * Optimally, keysock messages containing actual keying material would
2049 	 * be allocated with esballoc(), with a zeroing free function.
2050 	 */
2051 	if (mp != NULL)
2052 		freemsg(mp);
2053 }
2054 
2055 /*
2056  * Keysock's read service procedure is there only for PF_KEY reply
2057  * messages that really need to reach the top.
2058  */
2059 static void
2060 keysock_rsrv(queue_t *q)
2061 {
2062 	mblk_t *mp;
2063 
2064 	while ((mp = getq(q)) != NULL) {
2065 		if (canputnext(q)) {
2066 			putnext(q, mp);
2067 		} else {
2068 			(void) putbq(q, mp);
2069 			return;
2070 		}
2071 	}
2072 }
2073 
2074 /*
2075  * The read procedure should only be invoked by a keysock consumer, like
2076  * ESP, AH, etc.  I should only see KEYSOCK_OUT and KEYSOCK_HELLO_ACK
2077  * messages on my read queues.
2078  */
2079 static void
2080 keysock_rput(queue_t *q, mblk_t *mp)
2081 {
2082 	keysock_consumer_t *kc = (keysock_consumer_t *)q->q_ptr;
2083 	ipsec_info_t *ii;
2084 	keysock_hello_ack_t *ksa;
2085 	minor_t serial;
2086 	mblk_t *mp1;
2087 	sadb_msg_t *samsg;
2088 
2089 	/* Make sure I'm a consumer instance.  (i.e. something's below me) */
2090 	ASSERT(WR(q)->q_next != NULL);
2091 
2092 	if (mp->b_datap->db_type != M_CTL) {
2093 		/*
2094 		 * Keysock should only see keysock consumer interface
2095 		 * messages (see ipsec_info.h) on its read procedure.
2096 		 * To be robust, however, putnext() up so the STREAM head can
2097 		 * deal with it appropriately.
2098 		 */
2099 		ks1dbg(("Hmmm, a non M_CTL (%d, 0x%x) on keysock_rput.\n",
2100 		    mp->b_datap->db_type, mp->b_datap->db_type));
2101 		putnext(q, mp);
2102 		return;
2103 	}
2104 
2105 	ii = (ipsec_info_t *)mp->b_rptr;
2106 
2107 	switch (ii->ipsec_info_type) {
2108 	case KEYSOCK_OUT:
2109 		/*
2110 		 * A consumer needs to pass a response message or an ACQUIRE
2111 		 * UP.  I assume that the consumer has done the right
2112 		 * thing w.r.t. message creation, etc.
2113 		 */
2114 		serial = ((keysock_out_t *)mp->b_rptr)->ks_out_serial;
2115 		mp1 = mp->b_cont;	/* Get M_DATA portion. */
2116 		freeb(mp);
2117 		samsg = (sadb_msg_t *)mp1->b_rptr;
2118 		if (samsg->sadb_msg_type == SADB_FLUSH ||
2119 		    (samsg->sadb_msg_type == SADB_DUMP &&
2120 			samsg->sadb_msg_len == SADB_8TO64(sizeof (*samsg)))) {
2121 			/*
2122 			 * If I'm an end-of-FLUSH or an end-of-DUMP marker...
2123 			 */
2124 			ASSERT(keysock_flushdump != 0);  /* Am I flushing? */
2125 
2126 			mutex_enter(&kc->kc_lock);
2127 			kc->kc_flags &= ~KC_FLUSHING;
2128 			mutex_exit(&kc->kc_lock);
2129 
2130 			if (samsg->sadb_msg_errno != 0)
2131 				keysock_flushdump_errno = samsg->sadb_msg_errno;
2132 
2133 			/*
2134 			 * Lower the atomic "flushing" count.  If it's
2135 			 * the last one, send up the end-of-{FLUSH,DUMP} to
2136 			 * the appropriate PF_KEY socket.
2137 			 */
2138 			if (atomic_add_32_nv(&keysock_flushdump, -1) != 0) {
2139 				ks1dbg(("One flush/dump message back from %d,"
2140 				    " more to go.\n", samsg->sadb_msg_satype));
2141 				freemsg(mp1);
2142 				return;
2143 			}
2144 
2145 			samsg->sadb_msg_errno =
2146 			    (uint8_t)keysock_flushdump_errno;
2147 			if (samsg->sadb_msg_type == SADB_DUMP) {
2148 				samsg->sadb_msg_seq = 0;
2149 			}
2150 		}
2151 		keysock_passup(mp1, samsg, serial, kc,
2152 		    (samsg->sadb_msg_type == SADB_DUMP));
2153 		return;
2154 	case KEYSOCK_HELLO_ACK:
2155 		/* Aha, now we can link in the consumer! */
2156 		ksa = (keysock_hello_ack_t *)ii;
2157 		keysock_link_consumer(ksa->ks_hello_satype, kc);
2158 		freemsg(mp);
2159 		return;
2160 	default:
2161 		ks1dbg(("Hmmm, an IPsec info I'm not used to, 0x%x\n",
2162 		    ii->ipsec_info_type));
2163 		putnext(q, mp);
2164 	}
2165 }
2166 
2167 /*
2168  * So we can avoid external linking problems....
2169  */
2170 boolean_t
2171 keysock_extended_reg(void)
2172 {
2173 	return (keysock_num_extended != 0);
2174 }
2175 
2176 uint32_t
2177 keysock_next_seq(void)
2178 {
2179 	return (atomic_add_32_nv(&keysock_acquire_seq, -1));
2180 }
2181