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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 */
25
26 #include <sys/param.h>
27 #include <sys/types.h>
28 #include <sys/stream.h>
29 #include <sys/strsubr.h>
30 #include <sys/strsun.h>
31 #include <sys/stropts.h>
32 #include <sys/vnode.h>
33 #include <sys/zone.h>
34 #include <sys/strlog.h>
35 #include <sys/sysmacros.h>
36 #define _SUN_TPI_VERSION 2
37 #include <sys/tihdr.h>
38 #include <sys/timod.h>
39 #include <sys/tiuser.h>
40 #include <sys/ddi.h>
41 #include <sys/sunddi.h>
42 #include <sys/sunldi.h>
43 #include <sys/file.h>
44 #include <sys/modctl.h>
45 #include <sys/debug.h>
46 #include <sys/kmem.h>
47 #include <sys/cmn_err.h>
48 #include <sys/proc.h>
49 #include <sys/suntpi.h>
50 #include <sys/atomic.h>
51 #include <sys/mkdev.h>
52 #include <sys/policy.h>
53 #include <sys/disp.h>
54
55 #include <sys/socket.h>
56 #include <netinet/in.h>
57 #include <net/pfkeyv2.h>
58
59 #include <inet/common.h>
60 #include <netinet/ip6.h>
61 #include <inet/ip.h>
62 #include <inet/proto_set.h>
63 #include <inet/nd.h>
64 #include <inet/optcom.h>
65 #include <inet/ipsec_info.h>
66 #include <inet/ipsec_impl.h>
67 #include <inet/keysock.h>
68
69 #include <sys/isa_defs.h>
70
71 /*
72 * This is a transport provider for the PF_KEY key mangement socket.
73 * (See RFC 2367 for details.)
74 * Downstream messages are wrapped in a keysock consumer interface KEYSOCK_IN
75 * messages (see ipsec_info.h), and passed to the appropriate consumer.
76 * Upstream messages are generated for all open PF_KEY sockets, when
77 * appropriate, as well as the sender (as long as SO_USELOOPBACK is enabled)
78 * in reply to downstream messages.
79 *
80 * Upstream messages must be created asynchronously for the following
81 * situations:
82 *
83 * 1.) A keysock consumer requires an SA, and there is currently none.
84 * 2.) An SA expires, either hard or soft lifetime.
85 * 3.) Other events a consumer deems fit.
86 *
87 * The MT model of this is PERMOD, with shared put procedures. Two types of
88 * messages, SADB_FLUSH and SADB_DUMP, need to lock down the perimeter to send
89 * down the *multiple* messages they create.
90 */
91
92 static vmem_t *keysock_vmem; /* for minor numbers. */
93
94 #define KEYSOCK_MAX_CONSUMERS 256
95
96 /* Default structure copied into T_INFO_ACK messages (from rts.c...) */
97 static struct T_info_ack keysock_g_t_info_ack = {
98 T_INFO_ACK,
99 T_INFINITE, /* TSDU_size. Maximum size messages. */
100 T_INVALID, /* ETSDU_size. No expedited data. */
101 T_INVALID, /* CDATA_size. No connect data. */
102 T_INVALID, /* DDATA_size. No disconnect data. */
103 0, /* ADDR_size. */
104 0, /* OPT_size. No user-settable options */
105 64 * 1024, /* TIDU_size. keysock allows maximum size messages. */
106 T_COTS, /* SERV_type. keysock supports connection oriented. */
107 TS_UNBND, /* CURRENT_state. This is set from keysock_state. */
108 (XPG4_1) /* Provider flags */
109 };
110
111 /* Named Dispatch Parameter Management Structure */
112 typedef struct keysockparam_s {
113 uint_t keysock_param_min;
114 uint_t keysock_param_max;
115 uint_t keysock_param_value;
116 char *keysock_param_name;
117 } keysockparam_t;
118
119 /*
120 * Table of NDD variables supported by keysock. These are loaded into
121 * keysock_g_nd in keysock_init_nd.
122 * All of these are alterable, within the min/max values given, at run time.
123 */
124 static keysockparam_t lcl_param_arr[] = {
125 /* min max value name */
126 { 4096, 65536, 8192, "keysock_xmit_hiwat"},
127 { 0, 65536, 1024, "keysock_xmit_lowat"},
128 { 4096, 65536, 8192, "keysock_recv_hiwat"},
129 { 65536, 1024*1024*1024, 256*1024, "keysock_max_buf"},
130 { 0, 3, 0, "keysock_debug"},
131 };
132 #define keystack_xmit_hiwat keystack_params[0].keysock_param_value
133 #define keystack_xmit_lowat keystack_params[1].keysock_param_value
134 #define keystack_recv_hiwat keystack_params[2].keysock_param_value
135 #define keystack_max_buf keystack_params[3].keysock_param_value
136 #define keystack_debug keystack_params[4].keysock_param_value
137
138 #define ks0dbg(a) printf a
139 /* NOTE: != 0 instead of > 0 so lint doesn't complain. */
140 #define ks1dbg(keystack, a) if (keystack->keystack_debug != 0) printf a
141 #define ks2dbg(keystack, a) if (keystack->keystack_debug > 1) printf a
142 #define ks3dbg(keystack, a) if (keystack->keystack_debug > 2) printf a
143
144 static int keysock_close(queue_t *);
145 static int keysock_open(queue_t *, dev_t *, int, int, cred_t *);
146 static void keysock_wput(queue_t *, mblk_t *);
147 static void keysock_rput(queue_t *, mblk_t *);
148 static void keysock_rsrv(queue_t *);
149 static void keysock_passup(mblk_t *, sadb_msg_t *, minor_t,
150 keysock_consumer_t *, boolean_t, keysock_stack_t *);
151 static void *keysock_stack_init(netstackid_t stackid, netstack_t *ns);
152 static void keysock_stack_fini(netstackid_t stackid, void *arg);
153
154 static struct module_info info = {
155 5138, "keysock", 1, INFPSZ, 512, 128
156 };
157
158 static struct qinit rinit = {
159 (pfi_t)keysock_rput, (pfi_t)keysock_rsrv, keysock_open, keysock_close,
160 NULL, &info
161 };
162
163 static struct qinit winit = {
164 (pfi_t)keysock_wput, NULL, NULL, NULL, NULL, &info
165 };
166
167 struct streamtab keysockinfo = {
168 &rinit, &winit
169 };
170
171 extern struct modlinkage *keysock_modlp;
172
173 /*
174 * Plumb IPsec.
175 *
176 * NOTE: New "default" modules will need to be loaded here if needed before
177 * boot time.
178 */
179
180 /* Keep these in global space to keep the lint from complaining. */
181 static char *IPSECESP = "ipsecesp";
182 static char *IPSECESPDEV = "/devices/pseudo/ipsecesp@0:ipsecesp";
183 static char *IPSECAH = "ipsecah";
184 static char *IPSECAHDEV = "/devices/pseudo/ipsecah@0:ipsecah";
185 static char *IP6DEV = "/devices/pseudo/ip6@0:ip6";
186 static char *KEYSOCK = "keysock";
187 static char *STRMOD = "strmod";
188
189 /*
190 * Load the other ipsec modules and plumb them together.
191 */
192 int
193 keysock_plumb_ipsec(netstack_t *ns)
194 {
195 ldi_handle_t lh, ip6_lh = NULL;
196 ldi_ident_t li = NULL;
197 int err = 0;
198 int muxid, rval;
199 boolean_t esp_present = B_TRUE;
200 cred_t *cr;
201 keysock_stack_t *keystack = ns->netstack_keysock;
202
203 #ifdef NS_DEBUG
204 (void) printf("keysock_plumb_ipsec(%d)\n",
205 ns->netstack_stackid);
206 #endif
207
208 keystack->keystack_plumbed = 0; /* we're trying again.. */
209
210 cr = zone_get_kcred(netstackid_to_zoneid(
211 keystack->keystack_netstack->netstack_stackid));
212 ASSERT(cr != NULL);
213 /*
214 * Load up the drivers (AH/ESP).
215 *
216 * I do this separately from the actual plumbing in case this function
217 * ever gets called from a diskless boot before the root filesystem is
218 * up. I don't have to worry about "keysock" because, well, if I'm
219 * here, keysock must've loaded successfully.
220 */
221 if (i_ddi_attach_pseudo_node(IPSECAH) == NULL) {
222 ks0dbg(("IPsec: AH failed to attach.\n"));
223 goto bail;
224 }
225 if (i_ddi_attach_pseudo_node(IPSECESP) == NULL) {
226 ks0dbg(("IPsec: ESP failed to attach.\n"));
227 esp_present = B_FALSE;
228 }
229
230 /*
231 * Set up the IP streams for AH and ESP, as well as tacking keysock
232 * on top of them. Assume keysock has set the autopushes up already.
233 */
234
235 /* Open IP. */
236 err = ldi_ident_from_mod(keysock_modlp, &li);
237 if (err) {
238 ks0dbg(("IPsec: lid_ident_from_mod failed (err %d).\n",
239 err));
240 goto bail;
241 }
242
243 err = ldi_open_by_name(IP6DEV, FREAD|FWRITE, cr, &ip6_lh, li);
244 if (err) {
245 ks0dbg(("IPsec: Open of IP6 failed (err %d).\n", err));
246 goto bail;
247 }
248
249 /* PLINK KEYSOCK/AH */
250 err = ldi_open_by_name(IPSECAHDEV, FREAD|FWRITE, cr, &lh, li);
251 if (err) {
252 ks0dbg(("IPsec: Open of AH failed (err %d).\n", err));
253 goto bail;
254 }
255 err = ldi_ioctl(lh,
256 I_PUSH, (intptr_t)KEYSOCK, FKIOCTL, cr, &rval);
257 if (err) {
258 ks0dbg(("IPsec: Push of KEYSOCK onto AH failed (err %d).\n",
259 err));
260 (void) ldi_close(lh, FREAD|FWRITE, cr);
261 goto bail;
262 }
263 err = ldi_ioctl(ip6_lh, I_PLINK, (intptr_t)lh,
264 FREAD+FWRITE+FNOCTTY+FKIOCTL, cr, &muxid);
265 if (err) {
266 ks0dbg(("IPsec: PLINK of KEYSOCK/AH failed (err %d).\n", err));
267 (void) ldi_close(lh, FREAD|FWRITE, cr);
268 goto bail;
269 }
270 (void) ldi_close(lh, FREAD|FWRITE, cr);
271
272 /* PLINK KEYSOCK/ESP */
273 if (esp_present) {
274 err = ldi_open_by_name(IPSECESPDEV,
275 FREAD|FWRITE, cr, &lh, li);
276 if (err) {
277 ks0dbg(("IPsec: Open of ESP failed (err %d).\n", err));
278 goto bail;
279 }
280 err = ldi_ioctl(lh,
281 I_PUSH, (intptr_t)KEYSOCK, FKIOCTL, cr, &rval);
282 if (err) {
283 ks0dbg(("IPsec: "
284 "Push of KEYSOCK onto ESP failed (err %d).\n",
285 err));
286 (void) ldi_close(lh, FREAD|FWRITE, cr);
287 goto bail;
288 }
289 err = ldi_ioctl(ip6_lh, I_PLINK, (intptr_t)lh,
290 FREAD+FWRITE+FNOCTTY+FKIOCTL, cr, &muxid);
291 if (err) {
292 ks0dbg(("IPsec: "
293 "PLINK of KEYSOCK/ESP failed (err %d).\n", err));
294 (void) ldi_close(lh, FREAD|FWRITE, cr);
295 goto bail;
296 }
297 (void) ldi_close(lh, FREAD|FWRITE, cr);
298 }
299
300 bail:
301 keystack->keystack_plumbed = (err == 0) ? 1 : -1;
302 if (ip6_lh != NULL) {
303 (void) ldi_close(ip6_lh, FREAD|FWRITE, cr);
304 }
305 if (li != NULL)
306 ldi_ident_release(li);
307 #ifdef NS_DEBUG
308 (void) printf("keysock_plumb_ipsec -> %d\n",
309 keystack->keystack_plumbed);
310 #endif
311 crfree(cr);
312 return (err);
313 }
314
315 /* ARGSUSED */
316 static int
317 keysock_param_get(q, mp, cp, cr)
318 queue_t *q;
319 mblk_t *mp;
320 caddr_t cp;
321 cred_t *cr;
322 {
323 keysockparam_t *keysockpa = (keysockparam_t *)cp;
324 uint_t value;
325 keysock_t *ks = (keysock_t *)q->q_ptr;
326 keysock_stack_t *keystack = ks->keysock_keystack;
327
328 mutex_enter(&keystack->keystack_param_lock);
329 value = keysockpa->keysock_param_value;
330 mutex_exit(&keystack->keystack_param_lock);
331
332 (void) mi_mpprintf(mp, "%u", value);
333 return (0);
334 }
335
336 /* This routine sets an NDD variable in a keysockparam_t structure. */
337 /* ARGSUSED */
338 static int
339 keysock_param_set(q, mp, value, cp, cr)
340 queue_t *q;
341 mblk_t *mp;
342 char *value;
343 caddr_t cp;
344 cred_t *cr;
345 {
346 ulong_t new_value;
347 keysockparam_t *keysockpa = (keysockparam_t *)cp;
348 keysock_t *ks = (keysock_t *)q->q_ptr;
349 keysock_stack_t *keystack = ks->keysock_keystack;
350
351 /* Convert the value from a string into a long integer. */
352 if (ddi_strtoul(value, NULL, 10, &new_value) != 0)
353 return (EINVAL);
354
355 mutex_enter(&keystack->keystack_param_lock);
356 /*
357 * Fail the request if the new value does not lie within the
358 * required bounds.
359 */
360 if (new_value < keysockpa->keysock_param_min ||
361 new_value > keysockpa->keysock_param_max) {
362 mutex_exit(&keystack->keystack_param_lock);
363 return (EINVAL);
364 }
365
366 /* Set the new value */
367 keysockpa->keysock_param_value = new_value;
368 mutex_exit(&keystack->keystack_param_lock);
369
370 return (0);
371 }
372
373 /*
374 * Initialize keysock at module load time
375 */
376 boolean_t
377 keysock_ddi_init(void)
378 {
379 keysock_max_optsize = optcom_max_optsize(
380 keysock_opt_obj.odb_opt_des_arr, keysock_opt_obj.odb_opt_arr_cnt);
381
382 keysock_vmem = vmem_create("keysock", (void *)1, MAXMIN, 1,
383 NULL, NULL, NULL, 1, VM_SLEEP | VMC_IDENTIFIER);
384
385 /*
386 * We want to be informed each time a stack is created or
387 * destroyed in the kernel, so we can maintain the
388 * set of keysock_stack_t's.
389 */
390 netstack_register(NS_KEYSOCK, keysock_stack_init, NULL,
391 keysock_stack_fini);
392
393 return (B_TRUE);
394 }
395
396 /*
397 * Walk through the param array specified registering each element with the
398 * named dispatch handler.
399 */
400 static boolean_t
401 keysock_param_register(IDP *ndp, keysockparam_t *ksp, int cnt)
402 {
403 for (; cnt-- > 0; ksp++) {
404 if (ksp->keysock_param_name != NULL &&
405 ksp->keysock_param_name[0]) {
406 if (!nd_load(ndp,
407 ksp->keysock_param_name,
408 keysock_param_get, keysock_param_set,
409 (caddr_t)ksp)) {
410 nd_free(ndp);
411 return (B_FALSE);
412 }
413 }
414 }
415 return (B_TRUE);
416 }
417
418 /*
419 * Initialize keysock for one stack instance
420 */
421 /* ARGSUSED */
422 static void *
423 keysock_stack_init(netstackid_t stackid, netstack_t *ns)
424 {
425 keysock_stack_t *keystack;
426 keysockparam_t *ksp;
427
428 keystack = (keysock_stack_t *)kmem_zalloc(sizeof (*keystack), KM_SLEEP);
429 keystack->keystack_netstack = ns;
430
431 keystack->keystack_acquire_seq = 0xffffffff;
432
433 ksp = (keysockparam_t *)kmem_alloc(sizeof (lcl_param_arr), KM_SLEEP);
434 keystack->keystack_params = ksp;
435 bcopy(lcl_param_arr, ksp, sizeof (lcl_param_arr));
436
437 (void) keysock_param_register(&keystack->keystack_g_nd, ksp,
438 A_CNT(lcl_param_arr));
439
440 mutex_init(&keystack->keystack_list_lock, NULL, MUTEX_DEFAULT, NULL);
441 mutex_init(&keystack->keystack_consumers_lock,
442 NULL, MUTEX_DEFAULT, NULL);
443 mutex_init(&keystack->keystack_param_lock, NULL, MUTEX_DEFAULT, NULL);
444 return (keystack);
445 }
446
447 /*
448 * Free NDD variable space, and other destructors, for keysock.
449 */
450 void
451 keysock_ddi_destroy(void)
452 {
453 netstack_unregister(NS_KEYSOCK);
454 vmem_destroy(keysock_vmem);
455 }
456
457 /*
458 * Remove one stack instance from keysock
459 */
460 /* ARGSUSED */
461 static void
462 keysock_stack_fini(netstackid_t stackid, void *arg)
463 {
464 keysock_stack_t *keystack = (keysock_stack_t *)arg;
465
466 nd_free(&keystack->keystack_g_nd);
467 kmem_free(keystack->keystack_params, sizeof (lcl_param_arr));
468 keystack->keystack_params = NULL;
469
470 mutex_destroy(&keystack->keystack_list_lock);
471 mutex_destroy(&keystack->keystack_consumers_lock);
472 mutex_destroy(&keystack->keystack_param_lock);
473
474 kmem_free(keystack, sizeof (*keystack));
475 }
476
477 /*
478 * Close routine for keysock.
479 */
480 static int
481 keysock_close(queue_t *q)
482 {
483 keysock_t *ks;
484 keysock_consumer_t *kc;
485 void *ptr = q->q_ptr;
486 int size;
487 keysock_stack_t *keystack;
488
489
490 qprocsoff(q);
491
492 /* Safe assumption. */
493 ASSERT(ptr != NULL);
494
495 if (WR(q)->q_next) {
496 kc = (keysock_consumer_t *)ptr;
497 keystack = kc->kc_keystack;
498
499 ks1dbg(keystack, ("Module close, removing a consumer (%d).\n",
500 kc->kc_sa_type));
501 /*
502 * Because of PERMOD open/close exclusive perimeter, I
503 * can inspect KC_FLUSHING w/o locking down kc->kc_lock.
504 */
505 if (kc->kc_flags & KC_FLUSHING) {
506 /*
507 * If this decrement was the last one, send
508 * down the next pending one, if any.
509 *
510 * With a PERMOD perimeter, the mutexes ops aren't
511 * really necessary, but if we ever loosen up, we will
512 * have this bit covered already.
513 */
514 keystack->keystack_flushdump--;
515 if (keystack->keystack_flushdump == 0) {
516 /*
517 * The flush/dump terminated by having a
518 * consumer go away. I need to send up to the
519 * appropriate keysock all of the relevant
520 * information. Unfortunately, I don't
521 * have that handy.
522 */
523 ks0dbg(("Consumer went away while flushing or"
524 " dumping.\n"));
525 }
526 }
527 size = sizeof (keysock_consumer_t);
528 mutex_enter(&keystack->keystack_consumers_lock);
529 keystack->keystack_consumers[kc->kc_sa_type] = NULL;
530 mutex_exit(&keystack->keystack_consumers_lock);
531 mutex_destroy(&kc->kc_lock);
532 netstack_rele(kc->kc_keystack->keystack_netstack);
533 } else {
534 ks = (keysock_t *)ptr;
535 keystack = ks->keysock_keystack;
536
537 ks3dbg(keystack,
538 ("Driver close, PF_KEY socket is going away.\n"));
539 if ((ks->keysock_flags & KEYSOCK_EXTENDED) != 0)
540 atomic_dec_32(&keystack->keystack_num_extended);
541 size = sizeof (keysock_t);
542 mutex_enter(&keystack->keystack_list_lock);
543 *(ks->keysock_ptpn) = ks->keysock_next;
544 if (ks->keysock_next != NULL)
545 ks->keysock_next->keysock_ptpn = ks->keysock_ptpn;
546 mutex_exit(&keystack->keystack_list_lock);
547 mutex_destroy(&ks->keysock_lock);
548 vmem_free(keysock_vmem, (void *)(uintptr_t)ks->keysock_serial,
549 1);
550 netstack_rele(ks->keysock_keystack->keystack_netstack);
551 }
552
553 /* Now I'm free. */
554 kmem_free(ptr, size);
555 return (0);
556 }
557 /*
558 * Open routine for keysock.
559 */
560 /* ARGSUSED */
561 static int
562 keysock_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
563 {
564 keysock_t *ks;
565 keysock_consumer_t *kc;
566 mblk_t *mp;
567 ipsec_info_t *ii;
568 netstack_t *ns;
569 keysock_stack_t *keystack;
570
571 if (secpolicy_ip_config(credp, B_FALSE) != 0) {
572 /* Privilege debugging will log the error */
573 return (EPERM);
574 }
575
576 if (q->q_ptr != NULL)
577 return (0); /* Re-open of an already open instance. */
578
579 ns = netstack_find_by_cred(credp);
580 ASSERT(ns != NULL);
581 keystack = ns->netstack_keysock;
582 ASSERT(keystack != NULL);
583
584 ks3dbg(keystack, ("Entering keysock open.\n"));
585
586 if (keystack->keystack_plumbed < 1) {
587 netstack_t *ns = keystack->keystack_netstack;
588
589 keystack->keystack_plumbed = 0;
590 #ifdef NS_DEBUG
591 printf("keysock_open(%d) - plumb\n",
592 keystack->keystack_netstack->netstack_stackid);
593 #endif
594 /*
595 * Don't worry about ipsec_failure being true here.
596 * (See ip.c). An open of keysock should try and force
597 * the issue. Maybe it was a transient failure.
598 */
599 ipsec_loader_loadnow(ns->netstack_ipsec);
600 }
601
602 if (sflag & MODOPEN) {
603 /* Initialize keysock_consumer state here. */
604 kc = kmem_zalloc(sizeof (keysock_consumer_t), KM_NOSLEEP);
605 if (kc == NULL) {
606 netstack_rele(keystack->keystack_netstack);
607 return (ENOMEM);
608 }
609 mutex_init(&kc->kc_lock, NULL, MUTEX_DEFAULT, 0);
610 kc->kc_rq = q;
611 kc->kc_wq = WR(q);
612
613 q->q_ptr = kc;
614 WR(q)->q_ptr = kc;
615
616 kc->kc_keystack = keystack;
617 qprocson(q);
618
619 /*
620 * Send down initial message to whatever I was pushed on top
621 * of asking for its consumer type. The reply will set it.
622 */
623
624 /* Allocate it. */
625 mp = allocb(sizeof (ipsec_info_t), BPRI_HI);
626 if (mp == NULL) {
627 ks1dbg(keystack, (
628 "keysock_open: Cannot allocate KEYSOCK_HELLO.\n"));
629 /* Do I need to set these to null? */
630 q->q_ptr = NULL;
631 WR(q)->q_ptr = NULL;
632 mutex_destroy(&kc->kc_lock);
633 kmem_free(kc, sizeof (*kc));
634 netstack_rele(keystack->keystack_netstack);
635 return (ENOMEM);
636 }
637
638 /* If I allocated okay, putnext to what I was pushed atop. */
639 mp->b_wptr += sizeof (ipsec_info_t);
640 mp->b_datap->db_type = M_CTL;
641 ii = (ipsec_info_t *)mp->b_rptr;
642 ii->ipsec_info_type = KEYSOCK_HELLO;
643 /* Length only of type/len. */
644 ii->ipsec_info_len = sizeof (ii->ipsec_allu);
645 ks2dbg(keystack, ("Ready to putnext KEYSOCK_HELLO.\n"));
646 putnext(kc->kc_wq, mp);
647 } else {
648 minor_t ksminor;
649
650 /* Initialize keysock state. */
651
652 ks2dbg(keystack, ("Made it into PF_KEY socket open.\n"));
653
654 ksminor = (minor_t)(uintptr_t)
655 vmem_alloc(keysock_vmem, 1, VM_NOSLEEP);
656 if (ksminor == 0) {
657 netstack_rele(keystack->keystack_netstack);
658 return (ENOMEM);
659 }
660 ks = kmem_zalloc(sizeof (keysock_t), KM_NOSLEEP);
661 if (ks == NULL) {
662 vmem_free(keysock_vmem, (void *)(uintptr_t)ksminor, 1);
663 netstack_rele(keystack->keystack_netstack);
664 return (ENOMEM);
665 }
666
667 mutex_init(&ks->keysock_lock, NULL, MUTEX_DEFAULT, 0);
668 ks->keysock_rq = q;
669 ks->keysock_wq = WR(q);
670 ks->keysock_state = TS_UNBND;
671 ks->keysock_serial = ksminor;
672
673 q->q_ptr = ks;
674 WR(q)->q_ptr = ks;
675 ks->keysock_keystack = keystack;
676
677 /*
678 * The receive hiwat is only looked at on the stream head
679 * queue. Store in q_hiwat in order to return on SO_RCVBUF
680 * getsockopts.
681 */
682
683 q->q_hiwat = keystack->keystack_recv_hiwat;
684
685 /*
686 * The transmit hiwat/lowat is only looked at on IP's queue.
687 * Store in q_hiwat/q_lowat in order to return on
688 * SO_SNDBUF/SO_SNDLOWAT getsockopts.
689 */
690
691 WR(q)->q_hiwat = keystack->keystack_xmit_hiwat;
692 WR(q)->q_lowat = keystack->keystack_xmit_lowat;
693
694 *devp = makedevice(getmajor(*devp), ksminor);
695
696 /*
697 * Thread keysock into the global keysock list.
698 */
699 mutex_enter(&keystack->keystack_list_lock);
700 ks->keysock_next = keystack->keystack_list;
701 ks->keysock_ptpn = &keystack->keystack_list;
702 if (keystack->keystack_list != NULL) {
703 keystack->keystack_list->keysock_ptpn =
704 &ks->keysock_next;
705 }
706 keystack->keystack_list = ks;
707 mutex_exit(&keystack->keystack_list_lock);
708
709 qprocson(q);
710 (void) proto_set_rx_hiwat(q, NULL,
711 keystack->keystack_recv_hiwat);
712 /*
713 * Wait outside the keysock module perimeter for IPsec
714 * plumbing to be completed. If it fails, keysock_close()
715 * undoes everything we just did.
716 */
717 if (!ipsec_loader_wait(q,
718 keystack->keystack_netstack->netstack_ipsec)) {
719 (void) keysock_close(q);
720 return (EPFNOSUPPORT);
721 }
722 }
723
724 return (0);
725 }
726
727 /* BELOW THIS LINE ARE ROUTINES INCLUDING AND RELATED TO keysock_wput(). */
728
729 /*
730 * Copy relevant state bits.
731 */
732 static void
733 keysock_copy_info(struct T_info_ack *tap, keysock_t *ks)
734 {
735 *tap = keysock_g_t_info_ack;
736 tap->CURRENT_state = ks->keysock_state;
737 tap->OPT_size = keysock_max_optsize;
738 }
739
740 /*
741 * This routine responds to T_CAPABILITY_REQ messages. It is called by
742 * keysock_wput. Much of the T_CAPABILITY_ACK information is copied from
743 * keysock_g_t_info_ack. The current state of the stream is copied from
744 * keysock_state.
745 */
746 static void
747 keysock_capability_req(queue_t *q, mblk_t *mp)
748 {
749 keysock_t *ks = (keysock_t *)q->q_ptr;
750 t_uscalar_t cap_bits1;
751 struct T_capability_ack *tcap;
752
753 cap_bits1 = ((struct T_capability_req *)mp->b_rptr)->CAP_bits1;
754
755 mp = tpi_ack_alloc(mp, sizeof (struct T_capability_ack),
756 mp->b_datap->db_type, T_CAPABILITY_ACK);
757 if (mp == NULL)
758 return;
759
760 tcap = (struct T_capability_ack *)mp->b_rptr;
761 tcap->CAP_bits1 = 0;
762
763 if (cap_bits1 & TC1_INFO) {
764 keysock_copy_info(&tcap->INFO_ack, ks);
765 tcap->CAP_bits1 |= TC1_INFO;
766 }
767
768 qreply(q, mp);
769 }
770
771 /*
772 * This routine responds to T_INFO_REQ messages. It is called by
773 * keysock_wput_other.
774 * Most of the T_INFO_ACK information is copied from keysock_g_t_info_ack.
775 * The current state of the stream is copied from keysock_state.
776 */
777 static void
778 keysock_info_req(q, mp)
779 queue_t *q;
780 mblk_t *mp;
781 {
782 mp = tpi_ack_alloc(mp, sizeof (struct T_info_ack), M_PCPROTO,
783 T_INFO_ACK);
784 if (mp == NULL)
785 return;
786 keysock_copy_info((struct T_info_ack *)mp->b_rptr,
787 (keysock_t *)q->q_ptr);
788 qreply(q, mp);
789 }
790
791 /*
792 * keysock_err_ack. This routine creates a
793 * T_ERROR_ACK message and passes it
794 * upstream.
795 */
796 static void
797 keysock_err_ack(q, mp, t_error, sys_error)
798 queue_t *q;
799 mblk_t *mp;
800 int t_error;
801 int sys_error;
802 {
803 if ((mp = mi_tpi_err_ack_alloc(mp, t_error, sys_error)) != NULL)
804 qreply(q, mp);
805 }
806
807 /*
808 * This routine retrieves the current status of socket options.
809 * It returns the size of the option retrieved.
810 */
811 /* ARGSUSED */
812 int
813 keysock_opt_get(queue_t *q, int level, int name, uchar_t *ptr)
814 {
815 int *i1 = (int *)ptr;
816 keysock_t *ks = (keysock_t *)q->q_ptr;
817
818 switch (level) {
819 case SOL_SOCKET:
820 mutex_enter(&ks->keysock_lock);
821 switch (name) {
822 case SO_TYPE:
823 *i1 = SOCK_RAW;
824 break;
825 case SO_USELOOPBACK:
826 *i1 = (int)(!((ks->keysock_flags & KEYSOCK_NOLOOP) ==
827 KEYSOCK_NOLOOP));
828 break;
829 /*
830 * The following two items can be manipulated,
831 * but changing them should do nothing.
832 */
833 case SO_SNDBUF:
834 *i1 = (int)q->q_hiwat;
835 break;
836 case SO_RCVBUF:
837 *i1 = (int)(RD(q)->q_hiwat);
838 break;
839 }
840 mutex_exit(&ks->keysock_lock);
841 break;
842 default:
843 return (0);
844 }
845 return (sizeof (int));
846 }
847
848 /*
849 * This routine sets socket options.
850 */
851 /* ARGSUSED */
852 int
853 keysock_opt_set(queue_t *q, uint_t mgmt_flags, int level,
854 int name, uint_t inlen, uchar_t *invalp, uint_t *outlenp,
855 uchar_t *outvalp, void *thisdg_attrs, cred_t *cr)
856 {
857 int *i1 = (int *)invalp, errno = 0;
858 keysock_t *ks = (keysock_t *)q->q_ptr;
859 keysock_stack_t *keystack = ks->keysock_keystack;
860
861 switch (level) {
862 case SOL_SOCKET:
863 mutex_enter(&ks->keysock_lock);
864 switch (name) {
865 case SO_USELOOPBACK:
866 if (!(*i1))
867 ks->keysock_flags |= KEYSOCK_NOLOOP;
868 else ks->keysock_flags &= ~KEYSOCK_NOLOOP;
869 break;
870 case SO_SNDBUF:
871 if (*i1 > keystack->keystack_max_buf)
872 errno = ENOBUFS;
873 else q->q_hiwat = *i1;
874 break;
875 case SO_RCVBUF:
876 if (*i1 > keystack->keystack_max_buf) {
877 errno = ENOBUFS;
878 } else {
879 RD(q)->q_hiwat = *i1;
880 (void) proto_set_rx_hiwat(RD(q), NULL, *i1);
881 }
882 break;
883 default:
884 errno = EINVAL;
885 }
886 mutex_exit(&ks->keysock_lock);
887 break;
888 default:
889 errno = EINVAL;
890 }
891 return (errno);
892 }
893
894 /*
895 * Handle STREAMS messages.
896 */
897 static void
898 keysock_wput_other(queue_t *q, mblk_t *mp)
899 {
900 struct iocblk *iocp;
901 int error;
902 keysock_t *ks = (keysock_t *)q->q_ptr;
903 keysock_stack_t *keystack = ks->keysock_keystack;
904 cred_t *cr;
905
906 switch (mp->b_datap->db_type) {
907 case M_PROTO:
908 case M_PCPROTO:
909 if ((mp->b_wptr - mp->b_rptr) < sizeof (long)) {
910 ks3dbg(keystack, (
911 "keysock_wput_other: Not big enough M_PROTO\n"));
912 freemsg(mp);
913 return;
914 }
915 switch (((union T_primitives *)mp->b_rptr)->type) {
916 case T_CAPABILITY_REQ:
917 keysock_capability_req(q, mp);
918 break;
919 case T_INFO_REQ:
920 keysock_info_req(q, mp);
921 break;
922 case T_SVR4_OPTMGMT_REQ:
923 case T_OPTMGMT_REQ:
924 /*
925 * All Solaris components should pass a db_credp
926 * for this TPI message, hence we ASSERT.
927 * But in case there is some other M_PROTO that looks
928 * like a TPI message sent by some other kernel
929 * component, we check and return an error.
930 */
931 cr = msg_getcred(mp, NULL);
932 ASSERT(cr != NULL);
933 if (cr == NULL) {
934 keysock_err_ack(q, mp, TSYSERR, EINVAL);
935 return;
936 }
937 if (((union T_primitives *)mp->b_rptr)->type ==
938 T_SVR4_OPTMGMT_REQ) {
939 svr4_optcom_req(q, mp, cr, &keysock_opt_obj);
940 } else {
941 tpi_optcom_req(q, mp, cr, &keysock_opt_obj);
942 }
943 break;
944 case T_DATA_REQ:
945 case T_EXDATA_REQ:
946 case T_ORDREL_REQ:
947 /* Illegal for keysock. */
948 freemsg(mp);
949 (void) putnextctl1(RD(q), M_ERROR, EPROTO);
950 break;
951 default:
952 /* Not supported by keysock. */
953 keysock_err_ack(q, mp, TNOTSUPPORT, 0);
954 break;
955 }
956 return;
957 case M_IOCTL:
958 iocp = (struct iocblk *)mp->b_rptr;
959 error = EINVAL;
960
961 switch (iocp->ioc_cmd) {
962 case ND_SET:
963 case ND_GET:
964 if (nd_getset(q, keystack->keystack_g_nd, mp)) {
965 qreply(q, mp);
966 return;
967 } else
968 error = ENOENT;
969 /* FALLTHRU */
970 default:
971 miocnak(q, mp, 0, error);
972 return;
973 }
974 case M_FLUSH:
975 if (*mp->b_rptr & FLUSHW) {
976 flushq(q, FLUSHALL);
977 *mp->b_rptr &= ~FLUSHW;
978 }
979 if (*mp->b_rptr & FLUSHR) {
980 qreply(q, mp);
981 return;
982 }
983 /* Else FALLTHRU */
984 }
985
986 /* If fell through, just black-hole the message. */
987 freemsg(mp);
988 }
989
990 /*
991 * Transmit a PF_KEY error message to the instance either pointed to
992 * by ks, the instance with serial number serial, or more, depending.
993 *
994 * The faulty message (or a reasonable facsimile thereof) is in mp.
995 * This function will free mp or recycle it for delivery, thereby causing
996 * the stream head to free it.
997 */
998 static void
999 keysock_error(keysock_t *ks, mblk_t *mp, int error, int diagnostic)
1000 {
1001 sadb_msg_t *samsg = (sadb_msg_t *)mp->b_rptr;
1002 keysock_stack_t *keystack = ks->keysock_keystack;
1003
1004 ASSERT(mp->b_datap->db_type == M_DATA);
1005
1006 if (samsg->sadb_msg_type < SADB_GETSPI ||
1007 samsg->sadb_msg_type > SADB_MAX)
1008 samsg->sadb_msg_type = SADB_RESERVED;
1009
1010 /*
1011 * Strip out extension headers.
1012 */
1013 ASSERT(mp->b_rptr + sizeof (*samsg) <= mp->b_datap->db_lim);
1014 mp->b_wptr = mp->b_rptr + sizeof (*samsg);
1015 samsg->sadb_msg_len = SADB_8TO64(sizeof (sadb_msg_t));
1016 samsg->sadb_msg_errno = (uint8_t)error;
1017 samsg->sadb_x_msg_diagnostic = (uint16_t)diagnostic;
1018
1019 keysock_passup(mp, samsg, ks->keysock_serial, NULL, B_FALSE, keystack);
1020 }
1021
1022 /*
1023 * Pass down a message to a consumer. Wrap it in KEYSOCK_IN, and copy
1024 * in the extv if passed in.
1025 */
1026 static void
1027 keysock_passdown(keysock_t *ks, mblk_t *mp, uint8_t satype, sadb_ext_t *extv[],
1028 boolean_t flushmsg)
1029 {
1030 keysock_consumer_t *kc;
1031 mblk_t *wrapper;
1032 keysock_in_t *ksi;
1033 int i;
1034 keysock_stack_t *keystack = ks->keysock_keystack;
1035
1036 wrapper = allocb(sizeof (ipsec_info_t), BPRI_HI);
1037 if (wrapper == NULL) {
1038 ks3dbg(keystack, ("keysock_passdown: allocb failed.\n"));
1039 if (extv[SADB_EXT_KEY_ENCRYPT] != NULL)
1040 bzero(extv[SADB_EXT_KEY_ENCRYPT],
1041 SADB_64TO8(
1042 extv[SADB_EXT_KEY_ENCRYPT]->sadb_ext_len));
1043 if (extv[SADB_EXT_KEY_AUTH] != NULL)
1044 bzero(extv[SADB_EXT_KEY_AUTH],
1045 SADB_64TO8(
1046 extv[SADB_EXT_KEY_AUTH]->sadb_ext_len));
1047 if (flushmsg) {
1048 ks0dbg((
1049 "keysock: Downwards flush/dump message failed!\n"));
1050 /* If this is true, I hold the perimeter. */
1051 keystack->keystack_flushdump--;
1052 }
1053 freemsg(mp);
1054 return;
1055 }
1056
1057 wrapper->b_datap->db_type = M_CTL;
1058 ksi = (keysock_in_t *)wrapper->b_rptr;
1059 ksi->ks_in_type = KEYSOCK_IN;
1060 ksi->ks_in_len = sizeof (keysock_in_t);
1061 if (extv[SADB_EXT_ADDRESS_SRC] != NULL)
1062 ksi->ks_in_srctype = KS_IN_ADDR_UNKNOWN;
1063 else ksi->ks_in_srctype = KS_IN_ADDR_NOTTHERE;
1064 if (extv[SADB_EXT_ADDRESS_DST] != NULL)
1065 ksi->ks_in_dsttype = KS_IN_ADDR_UNKNOWN;
1066 else ksi->ks_in_dsttype = KS_IN_ADDR_NOTTHERE;
1067 for (i = 0; i <= SADB_EXT_MAX; i++)
1068 ksi->ks_in_extv[i] = extv[i];
1069 ksi->ks_in_serial = ks->keysock_serial;
1070 wrapper->b_wptr += sizeof (ipsec_info_t);
1071 wrapper->b_cont = mp;
1072
1073 /*
1074 * Find the appropriate consumer where the message is passed down.
1075 */
1076 kc = keystack->keystack_consumers[satype];
1077 if (kc == NULL) {
1078 freeb(wrapper);
1079 keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_UNKNOWN_SATYPE);
1080 if (flushmsg) {
1081 ks0dbg((
1082 "keysock: Downwards flush/dump message failed!\n"));
1083 /* If this is true, I hold the perimeter. */
1084 keystack->keystack_flushdump--;
1085 }
1086 return;
1087 }
1088
1089 /*
1090 * NOTE: There used to be code in here to spin while a flush or
1091 * dump finished. Keysock now assumes that consumers have enough
1092 * MT-savviness to deal with that.
1093 */
1094
1095 /*
1096 * Current consumers (AH and ESP) are guaranteed to return a
1097 * FLUSH or DUMP message back, so when we reach here, we don't
1098 * have to worry about keysock_flushdumps.
1099 */
1100
1101 putnext(kc->kc_wq, wrapper);
1102 }
1103
1104 /*
1105 * High-level reality checking of extensions.
1106 */
1107 static boolean_t
1108 ext_check(sadb_ext_t *ext, keysock_stack_t *keystack)
1109 {
1110 int i;
1111 uint64_t *lp;
1112 sadb_ident_t *id;
1113 char *idstr;
1114
1115 switch (ext->sadb_ext_type) {
1116 case SADB_EXT_ADDRESS_SRC:
1117 case SADB_EXT_ADDRESS_DST:
1118 case SADB_X_EXT_ADDRESS_INNER_SRC:
1119 case SADB_X_EXT_ADDRESS_INNER_DST:
1120 /* Check for at least enough addtl length for a sockaddr. */
1121 if (ext->sadb_ext_len <= SADB_8TO64(sizeof (sadb_address_t)))
1122 return (B_FALSE);
1123 break;
1124 case SADB_EXT_LIFETIME_HARD:
1125 case SADB_EXT_LIFETIME_SOFT:
1126 case SADB_EXT_LIFETIME_CURRENT:
1127 if (ext->sadb_ext_len != SADB_8TO64(sizeof (sadb_lifetime_t)))
1128 return (B_FALSE);
1129 break;
1130 case SADB_EXT_SPIRANGE:
1131 /* See if the SPI range is legit. */
1132 if (htonl(((sadb_spirange_t *)ext)->sadb_spirange_min) >
1133 htonl(((sadb_spirange_t *)ext)->sadb_spirange_max))
1134 return (B_FALSE);
1135 break;
1136 case SADB_EXT_KEY_AUTH:
1137 case SADB_EXT_KEY_ENCRYPT:
1138 /* Key length check. */
1139 if (((sadb_key_t *)ext)->sadb_key_bits == 0)
1140 return (B_FALSE);
1141 /*
1142 * Check to see if the key length (in bits) is less than the
1143 * extension length (in 8-bits words).
1144 */
1145 if ((roundup(SADB_1TO8(((sadb_key_t *)ext)->sadb_key_bits), 8) +
1146 sizeof (sadb_key_t)) != SADB_64TO8(ext->sadb_ext_len)) {
1147 ks1dbg(keystack, (
1148 "ext_check: Key bits/length inconsistent.\n"));
1149 ks1dbg(keystack, ("%d bits, len is %d bytes.\n",
1150 ((sadb_key_t *)ext)->sadb_key_bits,
1151 SADB_64TO8(ext->sadb_ext_len)));
1152 return (B_FALSE);
1153 }
1154
1155 /* All-zeroes key check. */
1156 lp = (uint64_t *)(((char *)ext) + sizeof (sadb_key_t));
1157 for (i = 0;
1158 i < (ext->sadb_ext_len - SADB_8TO64(sizeof (sadb_key_t)));
1159 i++)
1160 if (lp[i] != 0)
1161 break; /* Out of for loop. */
1162 /* If finished the loop naturally, it's an all zero key. */
1163 if (lp[i] == 0)
1164 return (B_FALSE);
1165 break;
1166 case SADB_EXT_IDENTITY_SRC:
1167 case SADB_EXT_IDENTITY_DST:
1168 /*
1169 * Make sure the strings in these identities are
1170 * null-terminated. RFC 2367 underspecified how to handle
1171 * such a case. I "proactively" null-terminate the string
1172 * at the last byte if it's not terminated sooner.
1173 */
1174 id = (sadb_ident_t *)ext;
1175 i = SADB_64TO8(id->sadb_ident_len);
1176 i -= sizeof (sadb_ident_t);
1177 idstr = (char *)(id + 1);
1178 while (*idstr != '\0' && i > 0) {
1179 i--;
1180 idstr++;
1181 }
1182 if (i == 0) {
1183 /*
1184 * I.e., if the bozo user didn't NULL-terminate the
1185 * string...
1186 */
1187 idstr--;
1188 *idstr = '\0';
1189 }
1190 break;
1191 }
1192 return (B_TRUE); /* For now... */
1193 }
1194
1195 /* Return values for keysock_get_ext(). */
1196 #define KGE_OK 0
1197 #define KGE_DUP 1
1198 #define KGE_UNK 2
1199 #define KGE_LEN 3
1200 #define KGE_CHK 4
1201
1202 /*
1203 * Parse basic extension headers and return in the passed-in pointer vector.
1204 * Return values include:
1205 *
1206 * KGE_OK Everything's nice and parsed out.
1207 * If there are no extensions, place NULL in extv[0].
1208 * KGE_DUP There is a duplicate extension.
1209 * First instance in appropriate bin. First duplicate in
1210 * extv[0].
1211 * KGE_UNK Unknown extension type encountered. extv[0] contains
1212 * unknown header.
1213 * KGE_LEN Extension length error.
1214 * KGE_CHK High-level reality check failed on specific extension.
1215 *
1216 * My apologies for some of the pointer arithmetic in here. I'm thinking
1217 * like an assembly programmer, yet trying to make the compiler happy.
1218 */
1219 static int
1220 keysock_get_ext(sadb_ext_t *extv[], sadb_msg_t *basehdr, uint_t msgsize,
1221 keysock_stack_t *keystack)
1222 {
1223 bzero(extv, sizeof (sadb_ext_t *) * (SADB_EXT_MAX + 1));
1224
1225 /* Use extv[0] as the "current working pointer". */
1226
1227 extv[0] = (sadb_ext_t *)(basehdr + 1);
1228
1229 while (extv[0] < (sadb_ext_t *)(((uint8_t *)basehdr) + msgsize)) {
1230 /* Check for unknown headers. */
1231 if (extv[0]->sadb_ext_type == 0 ||
1232 extv[0]->sadb_ext_type > SADB_EXT_MAX)
1233 return (KGE_UNK);
1234
1235 /*
1236 * Check length. Use uint64_t because extlen is in units
1237 * of 64-bit words. If length goes beyond the msgsize,
1238 * return an error. (Zero length also qualifies here.)
1239 */
1240 if (extv[0]->sadb_ext_len == 0 ||
1241 (void *)((uint64_t *)extv[0] + extv[0]->sadb_ext_len) >
1242 (void *)((uint8_t *)basehdr + msgsize))
1243 return (KGE_LEN);
1244
1245 /* Check for redundant headers. */
1246 if (extv[extv[0]->sadb_ext_type] != NULL)
1247 return (KGE_DUP);
1248
1249 /*
1250 * Reality check the extension if possible at the keysock
1251 * level.
1252 */
1253 if (!ext_check(extv[0], keystack))
1254 return (KGE_CHK);
1255
1256 /* If I make it here, assign the appropriate bin. */
1257 extv[extv[0]->sadb_ext_type] = extv[0];
1258
1259 /* Advance pointer (See above for uint64_t ptr reasoning.) */
1260 extv[0] = (sadb_ext_t *)
1261 ((uint64_t *)extv[0] + extv[0]->sadb_ext_len);
1262 }
1263
1264 /* Everything's cool. */
1265
1266 /*
1267 * If extv[0] == NULL, then there are no extension headers in this
1268 * message. Ensure that this is the case.
1269 */
1270 if (extv[0] == (sadb_ext_t *)(basehdr + 1))
1271 extv[0] = NULL;
1272
1273 return (KGE_OK);
1274 }
1275
1276 /*
1277 * qwriter() callback to handle flushes and dumps. This routine will hold
1278 * the inner perimeter.
1279 */
1280 void
1281 keysock_do_flushdump(queue_t *q, mblk_t *mp)
1282 {
1283 int i, start, finish;
1284 mblk_t *mp1 = NULL;
1285 keysock_t *ks = (keysock_t *)q->q_ptr;
1286 sadb_ext_t *extv[SADB_EXT_MAX + 1];
1287 sadb_msg_t *samsg = (sadb_msg_t *)mp->b_rptr;
1288 keysock_stack_t *keystack = ks->keysock_keystack;
1289
1290 /*
1291 * I am guaranteed this will work. I did the work in keysock_parse()
1292 * already.
1293 */
1294 (void) keysock_get_ext(extv, samsg, SADB_64TO8(samsg->sadb_msg_len),
1295 keystack);
1296
1297 /*
1298 * I hold the perimeter, therefore I don't need to use atomic ops.
1299 */
1300 if (keystack->keystack_flushdump != 0) {
1301 /* XXX Should I instead use EBUSY? */
1302 /* XXX Or is there a way to queue these up? */
1303 keysock_error(ks, mp, ENOMEM, SADB_X_DIAGNOSTIC_NONE);
1304 return;
1305 }
1306
1307 if (samsg->sadb_msg_satype == SADB_SATYPE_UNSPEC) {
1308 start = 0;
1309 finish = KEYSOCK_MAX_CONSUMERS - 1;
1310 } else {
1311 start = samsg->sadb_msg_satype;
1312 finish = samsg->sadb_msg_satype;
1313 }
1314
1315 /*
1316 * Fill up keysock_flushdump with the number of outstanding dumps
1317 * and/or flushes.
1318 */
1319
1320 keystack->keystack_flushdump_errno = 0;
1321
1322 /*
1323 * Okay, I hold the perimeter. Eventually keysock_flushdump will
1324 * contain the number of consumers with outstanding flush operations.
1325 *
1326 * SO, here's the plan:
1327 * * For each relevant consumer (Might be one, might be all)
1328 * * Twiddle on the FLUSHING flag.
1329 * * Pass down the FLUSH/DUMP message.
1330 *
1331 * When I see upbound FLUSH/DUMP messages, I will decrement the
1332 * keysock_flushdump. When I decrement it to 0, I will pass the
1333 * FLUSH/DUMP message back up to the PF_KEY sockets. Because I will
1334 * pass down the right SA type to the consumer (either its own, or
1335 * that of UNSPEC), the right one will be reflected from each consumer,
1336 * and accordingly back to the socket.
1337 */
1338
1339 mutex_enter(&keystack->keystack_consumers_lock);
1340 for (i = start; i <= finish; i++) {
1341 if (keystack->keystack_consumers[i] != NULL) {
1342 mp1 = copymsg(mp);
1343 if (mp1 == NULL) {
1344 ks0dbg(("SADB_FLUSH copymsg() failed.\n"));
1345 /*
1346 * Error? And what about outstanding
1347 * flushes? Oh, yeah, they get sucked up and
1348 * the counter is decremented. Consumers
1349 * (see keysock_passdown()) are guaranteed
1350 * to deliver back a flush request, even if
1351 * it's an error.
1352 */
1353 keysock_error(ks, mp, ENOMEM,
1354 SADB_X_DIAGNOSTIC_NONE);
1355 return;
1356 }
1357 /*
1358 * Because my entry conditions are met above, the
1359 * following assertion should hold true.
1360 */
1361 mutex_enter(&keystack->keystack_consumers[i]->kc_lock);
1362 ASSERT((keystack->keystack_consumers[i]->kc_flags &
1363 KC_FLUSHING) == 0);
1364 keystack->keystack_consumers[i]->kc_flags |=
1365 KC_FLUSHING;
1366 mutex_exit(&(keystack->keystack_consumers[i]->kc_lock));
1367 /* Always increment the number of flushes... */
1368 keystack->keystack_flushdump++;
1369 /* Guaranteed to return a message. */
1370 keysock_passdown(ks, mp1, i, extv, B_TRUE);
1371 } else if (start == finish) {
1372 /*
1373 * In case where start == finish, and there's no
1374 * consumer, should we force an error? Yes.
1375 */
1376 mutex_exit(&keystack->keystack_consumers_lock);
1377 keysock_error(ks, mp, EINVAL,
1378 SADB_X_DIAGNOSTIC_UNKNOWN_SATYPE);
1379 return;
1380 }
1381 }
1382 mutex_exit(&keystack->keystack_consumers_lock);
1383
1384 if (keystack->keystack_flushdump == 0) {
1385 /*
1386 * There were no consumers at all for this message.
1387 * XXX For now return ESRCH.
1388 */
1389 keysock_error(ks, mp, ESRCH, SADB_X_DIAGNOSTIC_NO_SADBS);
1390 } else {
1391 /* Otherwise, free the original message. */
1392 freemsg(mp);
1393 }
1394 }
1395
1396 /*
1397 * Get the right diagnostic for a duplicate. Should probably use a static
1398 * table lookup.
1399 */
1400 int
1401 keysock_duplicate(int ext_type)
1402 {
1403 int rc = 0;
1404
1405 switch (ext_type) {
1406 case SADB_EXT_ADDRESS_SRC:
1407 rc = SADB_X_DIAGNOSTIC_DUPLICATE_SRC;
1408 break;
1409 case SADB_EXT_ADDRESS_DST:
1410 rc = SADB_X_DIAGNOSTIC_DUPLICATE_DST;
1411 break;
1412 case SADB_X_EXT_ADDRESS_INNER_SRC:
1413 rc = SADB_X_DIAGNOSTIC_DUPLICATE_INNER_SRC;
1414 break;
1415 case SADB_X_EXT_ADDRESS_INNER_DST:
1416 rc = SADB_X_DIAGNOSTIC_DUPLICATE_INNER_DST;
1417 break;
1418 case SADB_EXT_SA:
1419 rc = SADB_X_DIAGNOSTIC_DUPLICATE_SA;
1420 break;
1421 case SADB_EXT_SPIRANGE:
1422 rc = SADB_X_DIAGNOSTIC_DUPLICATE_RANGE;
1423 break;
1424 case SADB_EXT_KEY_AUTH:
1425 rc = SADB_X_DIAGNOSTIC_DUPLICATE_AKEY;
1426 break;
1427 case SADB_EXT_KEY_ENCRYPT:
1428 rc = SADB_X_DIAGNOSTIC_DUPLICATE_EKEY;
1429 break;
1430 }
1431 return (rc);
1432 }
1433
1434 /*
1435 * Get the right diagnostic for a reality check failure. Should probably use
1436 * a static table lookup.
1437 */
1438 int
1439 keysock_malformed(int ext_type)
1440 {
1441 int rc = 0;
1442
1443 switch (ext_type) {
1444 case SADB_EXT_ADDRESS_SRC:
1445 rc = SADB_X_DIAGNOSTIC_MALFORMED_SRC;
1446 break;
1447 case SADB_EXT_ADDRESS_DST:
1448 rc = SADB_X_DIAGNOSTIC_MALFORMED_DST;
1449 break;
1450 case SADB_X_EXT_ADDRESS_INNER_SRC:
1451 rc = SADB_X_DIAGNOSTIC_MALFORMED_INNER_SRC;
1452 break;
1453 case SADB_X_EXT_ADDRESS_INNER_DST:
1454 rc = SADB_X_DIAGNOSTIC_MALFORMED_INNER_DST;
1455 break;
1456 case SADB_EXT_SA:
1457 rc = SADB_X_DIAGNOSTIC_MALFORMED_SA;
1458 break;
1459 case SADB_EXT_SPIRANGE:
1460 rc = SADB_X_DIAGNOSTIC_MALFORMED_RANGE;
1461 break;
1462 case SADB_EXT_KEY_AUTH:
1463 rc = SADB_X_DIAGNOSTIC_MALFORMED_AKEY;
1464 break;
1465 case SADB_EXT_KEY_ENCRYPT:
1466 rc = SADB_X_DIAGNOSTIC_MALFORMED_EKEY;
1467 break;
1468 }
1469 return (rc);
1470 }
1471
1472 /*
1473 * Keysock massaging of an inverse ACQUIRE. Consult policy,
1474 * and construct an appropriate response.
1475 */
1476 static void
1477 keysock_inverse_acquire(mblk_t *mp, sadb_msg_t *samsg, sadb_ext_t *extv[],
1478 keysock_t *ks)
1479 {
1480 mblk_t *reply_mp;
1481 keysock_stack_t *keystack = ks->keysock_keystack;
1482
1483 /*
1484 * Reality check things...
1485 */
1486 if (extv[SADB_EXT_ADDRESS_SRC] == NULL) {
1487 keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_MISSING_SRC);
1488 return;
1489 }
1490 if (extv[SADB_EXT_ADDRESS_DST] == NULL) {
1491 keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_MISSING_DST);
1492 return;
1493 }
1494
1495 if (extv[SADB_X_EXT_ADDRESS_INNER_SRC] != NULL &&
1496 extv[SADB_X_EXT_ADDRESS_INNER_DST] == NULL) {
1497 keysock_error(ks, mp, EINVAL,
1498 SADB_X_DIAGNOSTIC_MISSING_INNER_DST);
1499 return;
1500 }
1501
1502 if (extv[SADB_X_EXT_ADDRESS_INNER_SRC] == NULL &&
1503 extv[SADB_X_EXT_ADDRESS_INNER_DST] != NULL) {
1504 keysock_error(ks, mp, EINVAL,
1505 SADB_X_DIAGNOSTIC_MISSING_INNER_SRC);
1506 return;
1507 }
1508
1509 reply_mp = ipsec_construct_inverse_acquire(samsg, extv,
1510 keystack->keystack_netstack);
1511
1512 if (reply_mp != NULL) {
1513 freemsg(mp);
1514 keysock_passup(reply_mp, (sadb_msg_t *)reply_mp->b_rptr,
1515 ks->keysock_serial, NULL, B_FALSE, keystack);
1516 } else {
1517 keysock_error(ks, mp, samsg->sadb_msg_errno,
1518 samsg->sadb_x_msg_diagnostic);
1519 }
1520 }
1521
1522 /*
1523 * Spew an extended REGISTER down to the relevant consumers.
1524 */
1525 static void
1526 keysock_extended_register(keysock_t *ks, mblk_t *mp, sadb_ext_t *extv[])
1527 {
1528 sadb_x_ereg_t *ereg = (sadb_x_ereg_t *)extv[SADB_X_EXT_EREG];
1529 uint8_t *satypes, *fencepost;
1530 mblk_t *downmp;
1531 sadb_ext_t *downextv[SADB_EXT_MAX + 1];
1532 keysock_stack_t *keystack = ks->keysock_keystack;
1533
1534 if (ks->keysock_registered[0] != 0 || ks->keysock_registered[1] != 0 ||
1535 ks->keysock_registered[2] != 0 || ks->keysock_registered[3] != 0) {
1536 keysock_error(ks, mp, EBUSY, 0);
1537 }
1538
1539 ks->keysock_flags |= KEYSOCK_EXTENDED;
1540 if (ereg == NULL) {
1541 keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_SATYPE_NEEDED);
1542 } else {
1543 ASSERT(mp->b_rptr + msgdsize(mp) == mp->b_wptr);
1544 fencepost = (uint8_t *)mp->b_wptr;
1545 satypes = ereg->sadb_x_ereg_satypes;
1546 while (*satypes != SADB_SATYPE_UNSPEC && satypes != fencepost) {
1547 downmp = copymsg(mp);
1548 if (downmp == NULL) {
1549 keysock_error(ks, mp, ENOMEM, 0);
1550 return;
1551 }
1552 /*
1553 * Since we've made it here, keysock_get_ext will work!
1554 */
1555 (void) keysock_get_ext(downextv,
1556 (sadb_msg_t *)downmp->b_rptr, msgdsize(downmp),
1557 keystack);
1558 keysock_passdown(ks, downmp, *satypes, downextv,
1559 B_FALSE);
1560 ++satypes;
1561 }
1562 freemsg(mp);
1563 }
1564
1565 /*
1566 * Set global to indicate we prefer an extended ACQUIRE.
1567 */
1568 atomic_inc_32(&keystack->keystack_num_extended);
1569 }
1570
1571 static void
1572 keysock_delpair_all(keysock_t *ks, mblk_t *mp, sadb_ext_t *extv[])
1573 {
1574 int i, start, finish;
1575 mblk_t *mp1 = NULL;
1576 keysock_stack_t *keystack = ks->keysock_keystack;
1577
1578 start = 0;
1579 finish = KEYSOCK_MAX_CONSUMERS - 1;
1580
1581 for (i = start; i <= finish; i++) {
1582 if (keystack->keystack_consumers[i] != NULL) {
1583 mp1 = copymsg(mp);
1584 if (mp1 == NULL) {
1585 keysock_error(ks, mp, ENOMEM,
1586 SADB_X_DIAGNOSTIC_NONE);
1587 return;
1588 }
1589 keysock_passdown(ks, mp1, i, extv, B_FALSE);
1590 }
1591 }
1592 }
1593
1594 /*
1595 * Handle PF_KEY messages.
1596 */
1597 static void
1598 keysock_parse(queue_t *q, mblk_t *mp)
1599 {
1600 sadb_msg_t *samsg;
1601 sadb_ext_t *extv[SADB_EXT_MAX + 1];
1602 keysock_t *ks = (keysock_t *)q->q_ptr;
1603 uint_t msgsize;
1604 uint8_t satype;
1605 keysock_stack_t *keystack = ks->keysock_keystack;
1606
1607 /* Make sure I'm a PF_KEY socket. (i.e. nothing's below me) */
1608 ASSERT(WR(q)->q_next == NULL);
1609
1610 samsg = (sadb_msg_t *)mp->b_rptr;
1611 ks2dbg(keystack, ("Received possible PF_KEY message, type %d.\n",
1612 samsg->sadb_msg_type));
1613
1614 msgsize = SADB_64TO8(samsg->sadb_msg_len);
1615
1616 if (msgdsize(mp) != msgsize) {
1617 /*
1618 * Message len incorrect w.r.t. actual size. Send an error
1619 * (EMSGSIZE). It may be necessary to massage things a
1620 * bit. For example, if the sadb_msg_type is hosed,
1621 * I need to set it to SADB_RESERVED to get delivery to
1622 * do the right thing. Then again, maybe just letting
1623 * the error delivery do the right thing.
1624 */
1625 ks2dbg(keystack,
1626 ("mblk (%lu) and base (%d) message sizes don't jibe.\n",
1627 msgdsize(mp), msgsize));
1628 keysock_error(ks, mp, EMSGSIZE, SADB_X_DIAGNOSTIC_NONE);
1629 return;
1630 }
1631
1632 if (msgsize > (uint_t)(mp->b_wptr - mp->b_rptr)) {
1633 /* Get all message into one mblk. */
1634 if (pullupmsg(mp, -1) == 0) {
1635 /*
1636 * Something screwy happened.
1637 */
1638 ks3dbg(keystack,
1639 ("keysock_parse: pullupmsg() failed.\n"));
1640 return;
1641 } else {
1642 samsg = (sadb_msg_t *)mp->b_rptr;
1643 }
1644 }
1645
1646 switch (keysock_get_ext(extv, samsg, msgsize, keystack)) {
1647 case KGE_DUP:
1648 /* Handle duplicate extension. */
1649 ks1dbg(keystack, ("Got duplicate extension of type %d.\n",
1650 extv[0]->sadb_ext_type));
1651 keysock_error(ks, mp, EINVAL,
1652 keysock_duplicate(extv[0]->sadb_ext_type));
1653 return;
1654 case KGE_UNK:
1655 /* Handle unknown extension. */
1656 ks1dbg(keystack, ("Got unknown extension of type %d.\n",
1657 extv[0]->sadb_ext_type));
1658 keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_UNKNOWN_EXT);
1659 return;
1660 case KGE_LEN:
1661 /* Length error. */
1662 ks1dbg(keystack,
1663 ("Length %d on extension type %d overrun or 0.\n",
1664 extv[0]->sadb_ext_len, extv[0]->sadb_ext_type));
1665 keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_BAD_EXTLEN);
1666 return;
1667 case KGE_CHK:
1668 /* Reality check failed. */
1669 ks1dbg(keystack,
1670 ("Reality check failed on extension type %d.\n",
1671 extv[0]->sadb_ext_type));
1672 keysock_error(ks, mp, EINVAL,
1673 keysock_malformed(extv[0]->sadb_ext_type));
1674 return;
1675 default:
1676 /* Default case is no errors. */
1677 break;
1678 }
1679
1680 switch (samsg->sadb_msg_type) {
1681 case SADB_REGISTER:
1682 /*
1683 * There's a semantic weirdness in that a message OTHER than
1684 * the return REGISTER message may be passed up if I set the
1685 * registered bit BEFORE I pass it down.
1686 *
1687 * SOOOO, I'll not twiddle any registered bits until I see
1688 * the upbound REGISTER (with a serial number in it).
1689 */
1690 if (samsg->sadb_msg_satype == SADB_SATYPE_UNSPEC) {
1691 /* Handle extended register here. */
1692 keysock_extended_register(ks, mp, extv);
1693 return;
1694 } else if (ks->keysock_flags & KEYSOCK_EXTENDED) {
1695 keysock_error(ks, mp, EBUSY, 0);
1696 return;
1697 }
1698 /* FALLTHRU */
1699 case SADB_GETSPI:
1700 case SADB_ADD:
1701 case SADB_UPDATE:
1702 case SADB_X_UPDATEPAIR:
1703 case SADB_DELETE:
1704 case SADB_X_DELPAIR:
1705 case SADB_GET:
1706 /*
1707 * Pass down to appropriate consumer.
1708 */
1709 if (samsg->sadb_msg_satype != SADB_SATYPE_UNSPEC)
1710 keysock_passdown(ks, mp, samsg->sadb_msg_satype, extv,
1711 B_FALSE);
1712 else keysock_error(ks, mp, EINVAL,
1713 SADB_X_DIAGNOSTIC_SATYPE_NEEDED);
1714 return;
1715 case SADB_X_DELPAIR_STATE:
1716 if (samsg->sadb_msg_satype == SADB_SATYPE_UNSPEC) {
1717 keysock_delpair_all(ks, mp, extv);
1718 } else {
1719 keysock_passdown(ks, mp, samsg->sadb_msg_satype, extv,
1720 B_FALSE);
1721 }
1722 return;
1723 case SADB_ACQUIRE:
1724 /*
1725 * If I _receive_ an acquire, this means I should spread it
1726 * out to registered sockets. Unless there's an errno...
1727 *
1728 * Need ADDRESS, may have ID, SENS, and PROP, unless errno,
1729 * in which case there should be NO extensions.
1730 *
1731 * Return to registered.
1732 */
1733 if (samsg->sadb_msg_errno != 0) {
1734 satype = samsg->sadb_msg_satype;
1735 if (satype == SADB_SATYPE_UNSPEC) {
1736 if (!(ks->keysock_flags & KEYSOCK_EXTENDED)) {
1737 keysock_error(ks, mp, EINVAL,
1738 SADB_X_DIAGNOSTIC_SATYPE_NEEDED);
1739 return;
1740 }
1741 /*
1742 * Reassign satype based on the first
1743 * flags that KEYSOCK_SETREG says.
1744 */
1745 while (satype <= SADB_SATYPE_MAX) {
1746 if (KEYSOCK_ISREG(ks, satype))
1747 break;
1748 satype++;
1749 }
1750 if (satype > SADB_SATYPE_MAX) {
1751 keysock_error(ks, mp, EBUSY, 0);
1752 return;
1753 }
1754 }
1755 keysock_passdown(ks, mp, satype, extv, B_FALSE);
1756 } else {
1757 if (samsg->sadb_msg_satype == SADB_SATYPE_UNSPEC) {
1758 keysock_error(ks, mp, EINVAL,
1759 SADB_X_DIAGNOSTIC_SATYPE_NEEDED);
1760 } else {
1761 keysock_passup(mp, samsg, 0, NULL, B_FALSE,
1762 keystack);
1763 }
1764 }
1765 return;
1766 case SADB_EXPIRE:
1767 /*
1768 * If someone sends this in, then send out to all senders.
1769 * (Save maybe ESP or AH, I have to be careful here.)
1770 *
1771 * Need ADDRESS, may have ID and SENS.
1772 *
1773 * XXX for now this is unsupported.
1774 */
1775 break;
1776 case SADB_FLUSH:
1777 /*
1778 * Nuke all SAs.
1779 *
1780 * No extensions at all. Return to all listeners.
1781 *
1782 * Question: Should I hold a lock here to prevent
1783 * additions/deletions while flushing?
1784 * Answer: No. (See keysock_passdown() for details.)
1785 */
1786 if (extv[0] != NULL) {
1787 /*
1788 * FLUSH messages shouldn't have extensions.
1789 * Return EINVAL.
1790 */
1791 ks2dbg(keystack, ("FLUSH message with extension.\n"));
1792 keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_NO_EXT);
1793 return;
1794 }
1795
1796 /* Passing down of DUMP/FLUSH messages are special. */
1797 qwriter(q, mp, keysock_do_flushdump, PERIM_INNER);
1798 return;
1799 case SADB_DUMP: /* not used by normal applications */
1800 if ((extv[0] != NULL) &&
1801 ((msgsize >
1802 (sizeof (sadb_msg_t) + sizeof (sadb_x_edump_t))) ||
1803 (extv[SADB_X_EXT_EDUMP] == NULL))) {
1804 keysock_error(ks, mp, EINVAL,
1805 SADB_X_DIAGNOSTIC_NO_EXT);
1806 return;
1807 }
1808 qwriter(q, mp, keysock_do_flushdump, PERIM_INNER);
1809 return;
1810 case SADB_X_PROMISC:
1811 /*
1812 * Promiscuous processing message.
1813 */
1814 if (samsg->sadb_msg_satype == 0)
1815 ks->keysock_flags &= ~KEYSOCK_PROMISC;
1816 else
1817 ks->keysock_flags |= KEYSOCK_PROMISC;
1818 keysock_passup(mp, samsg, ks->keysock_serial, NULL, B_FALSE,
1819 keystack);
1820 return;
1821 case SADB_X_INVERSE_ACQUIRE:
1822 keysock_inverse_acquire(mp, samsg, extv, ks);
1823 return;
1824 default:
1825 ks2dbg(keystack, ("Got unknown message type %d.\n",
1826 samsg->sadb_msg_type));
1827 keysock_error(ks, mp, EINVAL, SADB_X_DIAGNOSTIC_UNKNOWN_MSG);
1828 return;
1829 }
1830
1831 /* As a placeholder... */
1832 ks0dbg(("keysock_parse(): Hit EOPNOTSUPP\n"));
1833 keysock_error(ks, mp, EOPNOTSUPP, SADB_X_DIAGNOSTIC_NONE);
1834 }
1835
1836 /*
1837 * wput routing for PF_KEY/keysock/whatever. Unlike the routing socket,
1838 * I don't convert to ioctl()'s for IP. I am the end-all driver as far
1839 * as PF_KEY sockets are concerned. I do some conversion, but not as much
1840 * as IP/rts does.
1841 */
1842 static void
1843 keysock_wput(queue_t *q, mblk_t *mp)
1844 {
1845 uchar_t *rptr = mp->b_rptr;
1846 mblk_t *mp1;
1847 keysock_t *ks;
1848 keysock_stack_t *keystack;
1849
1850 if (WR(q)->q_next) {
1851 keysock_consumer_t *kc = (keysock_consumer_t *)q->q_ptr;
1852 keystack = kc->kc_keystack;
1853
1854 ks3dbg(keystack, ("In keysock_wput\n"));
1855
1856 /*
1857 * We shouldn't get writes on a consumer instance.
1858 * But for now, just passthru.
1859 */
1860 ks1dbg(keystack, ("Huh? wput for an consumer instance (%d)?\n",
1861 kc->kc_sa_type));
1862 putnext(q, mp);
1863 return;
1864 }
1865 ks = (keysock_t *)q->q_ptr;
1866 keystack = ks->keysock_keystack;
1867
1868 ks3dbg(keystack, ("In keysock_wput\n"));
1869
1870 switch (mp->b_datap->db_type) {
1871 case M_DATA:
1872 /*
1873 * Silently discard.
1874 */
1875 ks2dbg(keystack, ("raw M_DATA in keysock.\n"));
1876 freemsg(mp);
1877 return;
1878 case M_PROTO:
1879 case M_PCPROTO:
1880 if ((mp->b_wptr - rptr) >= sizeof (struct T_data_req)) {
1881 if (((union T_primitives *)rptr)->type == T_DATA_REQ) {
1882 if ((mp1 = mp->b_cont) == NULL) {
1883 /* No data after T_DATA_REQ. */
1884 ks2dbg(keystack,
1885 ("No data after DATA_REQ.\n"));
1886 freemsg(mp);
1887 return;
1888 }
1889 freeb(mp);
1890 mp = mp1;
1891 ks2dbg(keystack, ("T_DATA_REQ\n"));
1892 break; /* Out of switch. */
1893 }
1894 }
1895 /* FALLTHRU */
1896 default:
1897 ks3dbg(keystack, ("In default wput case (%d %d).\n",
1898 mp->b_datap->db_type, ((union T_primitives *)rptr)->type));
1899 keysock_wput_other(q, mp);
1900 return;
1901 }
1902
1903 /* I now have a PF_KEY message in an M_DATA block, pointed to by mp. */
1904 keysock_parse(q, mp);
1905 }
1906
1907 /* BELOW THIS LINE ARE ROUTINES INCLUDING AND RELATED TO keysock_rput(). */
1908
1909 /*
1910 * Called upon receipt of a KEYSOCK_HELLO_ACK to set up the appropriate
1911 * state vectors.
1912 */
1913 static void
1914 keysock_link_consumer(uint8_t satype, keysock_consumer_t *kc)
1915 {
1916 keysock_t *ks;
1917 keysock_stack_t *keystack = kc->kc_keystack;
1918
1919 mutex_enter(&keystack->keystack_consumers_lock);
1920 mutex_enter(&kc->kc_lock);
1921 if (keystack->keystack_consumers[satype] != NULL) {
1922 ks0dbg((
1923 "Hmmmm, someone closed %d before the HELLO_ACK happened.\n",
1924 satype));
1925 /*
1926 * Perhaps updating the new below-me consumer with what I have
1927 * so far would work too?
1928 */
1929 mutex_exit(&kc->kc_lock);
1930 mutex_exit(&keystack->keystack_consumers_lock);
1931 } else {
1932 /* Add new below-me consumer. */
1933 keystack->keystack_consumers[satype] = kc;
1934
1935 kc->kc_flags = 0;
1936 kc->kc_sa_type = satype;
1937 mutex_exit(&kc->kc_lock);
1938 mutex_exit(&keystack->keystack_consumers_lock);
1939
1940 /* Scan the keysock list. */
1941 mutex_enter(&keystack->keystack_list_lock);
1942 for (ks = keystack->keystack_list; ks != NULL;
1943 ks = ks->keysock_next) {
1944 if (KEYSOCK_ISREG(ks, satype)) {
1945 /*
1946 * XXX Perhaps send an SADB_REGISTER down on
1947 * the socket's behalf.
1948 */
1949 ks1dbg(keystack,
1950 ("Socket %u registered already for "
1951 "new consumer.\n", ks->keysock_serial));
1952 }
1953 }
1954 mutex_exit(&keystack->keystack_list_lock);
1955 }
1956 }
1957
1958 /*
1959 * Generate a KEYSOCK_OUT_ERR message for my consumer.
1960 */
1961 static void
1962 keysock_out_err(keysock_consumer_t *kc, int ks_errno, mblk_t *mp)
1963 {
1964 keysock_out_err_t *kse;
1965 mblk_t *imp;
1966 keysock_stack_t *keystack = kc->kc_keystack;
1967
1968 imp = allocb(sizeof (ipsec_info_t), BPRI_HI);
1969 if (imp == NULL) {
1970 ks1dbg(keystack, ("keysock_out_err: Can't alloc message.\n"));
1971 return;
1972 }
1973
1974 imp->b_datap->db_type = M_CTL;
1975 imp->b_wptr += sizeof (ipsec_info_t);
1976
1977 kse = (keysock_out_err_t *)imp->b_rptr;
1978 imp->b_cont = mp;
1979 kse->ks_err_type = KEYSOCK_OUT_ERR;
1980 kse->ks_err_len = sizeof (*kse);
1981 /* Is serial necessary? */
1982 kse->ks_err_serial = 0;
1983 kse->ks_err_errno = ks_errno;
1984
1985 /*
1986 * XXX What else do I need to do here w.r.t. information
1987 * to tell the consumer what caused this error?
1988 *
1989 * I believe the answer is the PF_KEY ACQUIRE (or other) message
1990 * attached in mp, which is appended at the end. I believe the
1991 * db_ref won't matter here, because the PF_KEY message is only read
1992 * for KEYSOCK_OUT_ERR.
1993 */
1994
1995 putnext(kc->kc_wq, imp);
1996 }
1997
1998 /* XXX this is a hack errno. */
1999 #define EIPSECNOSA 255
2000
2001 /*
2002 * Route message (pointed by mp, header in samsg) toward appropriate
2003 * sockets. Assume the message's creator did its job correctly.
2004 *
2005 * This should be a function that is followed by a return in its caller.
2006 * The compiler _should_ be able to use tail-call optimizations to make the
2007 * large ## of parameters not a huge deal.
2008 */
2009 static void
2010 keysock_passup(mblk_t *mp, sadb_msg_t *samsg, minor_t serial,
2011 keysock_consumer_t *kc, boolean_t persistent, keysock_stack_t *keystack)
2012 {
2013 keysock_t *ks;
2014 uint8_t satype = samsg->sadb_msg_satype;
2015 boolean_t toall = B_FALSE, allreg = B_FALSE, allereg = B_FALSE,
2016 setalg = B_FALSE;
2017 mblk_t *mp1;
2018 int err = EIPSECNOSA;
2019
2020 /* Convert mp, which is M_DATA, into an M_PROTO of type T_DATA_IND */
2021 mp1 = allocb(sizeof (struct T_data_req), BPRI_HI);
2022 if (mp1 == NULL) {
2023 err = ENOMEM;
2024 goto error;
2025 }
2026 mp1->b_wptr += sizeof (struct T_data_req);
2027 ((struct T_data_ind *)mp1->b_rptr)->PRIM_type = T_DATA_IND;
2028 ((struct T_data_ind *)mp1->b_rptr)->MORE_flag = 0;
2029 mp1->b_datap->db_type = M_PROTO;
2030 mp1->b_cont = mp;
2031 mp = mp1;
2032
2033 switch (samsg->sadb_msg_type) {
2034 case SADB_FLUSH:
2035 case SADB_GETSPI:
2036 case SADB_UPDATE:
2037 case SADB_X_UPDATEPAIR:
2038 case SADB_ADD:
2039 case SADB_DELETE:
2040 case SADB_X_DELPAIR:
2041 case SADB_EXPIRE:
2042 /*
2043 * These are most likely replies. Don't worry about
2044 * KEYSOCK_OUT_ERR handling. Deliver to all sockets.
2045 */
2046 ks3dbg(keystack,
2047 ("Delivering normal message (%d) to all sockets.\n",
2048 samsg->sadb_msg_type));
2049 toall = B_TRUE;
2050 break;
2051 case SADB_REGISTER:
2052 /*
2053 * REGISTERs come up for one of three reasons:
2054 *
2055 * 1.) In response to a normal SADB_REGISTER
2056 * (samsg->sadb_msg_satype != SADB_SATYPE_UNSPEC &&
2057 * serial != 0)
2058 * Deliver to normal SADB_REGISTERed sockets.
2059 * 2.) In response to an extended REGISTER
2060 * (samsg->sadb_msg_satype == SADB_SATYPE_UNSPEC)
2061 * Deliver to extended REGISTERed socket.
2062 * 3.) Spontaneous algorithm changes
2063 * (samsg->sadb_msg_satype != SADB_SATYPE_UNSPEC &&
2064 * serial == 0)
2065 * Deliver to REGISTERed sockets of all sorts.
2066 */
2067 if (kc == NULL) {
2068 /* Here because of keysock_error() call. */
2069 ASSERT(samsg->sadb_msg_errno != 0);
2070 break; /* Out of switch. */
2071 }
2072 ks3dbg(keystack, ("Delivering REGISTER.\n"));
2073 if (satype == SADB_SATYPE_UNSPEC) {
2074 /* REGISTER Reason #2 */
2075 allereg = B_TRUE;
2076 /*
2077 * Rewhack SA type so PF_KEY socket holder knows what
2078 * consumer generated this algorithm list.
2079 */
2080 satype = kc->kc_sa_type;
2081 samsg->sadb_msg_satype = satype;
2082 setalg = B_TRUE;
2083 } else if (serial == 0) {
2084 /* REGISTER Reason #3 */
2085 allreg = B_TRUE;
2086 allereg = B_TRUE;
2087 } else {
2088 /* REGISTER Reason #1 */
2089 allreg = B_TRUE;
2090 setalg = B_TRUE;
2091 }
2092 break;
2093 case SADB_ACQUIRE:
2094 /*
2095 * ACQUIREs are either extended (sadb_msg_satype == 0) or
2096 * regular (sadb_msg_satype != 0). And we're guaranteed
2097 * that serial == 0 for an ACQUIRE.
2098 */
2099 ks3dbg(keystack, ("Delivering ACQUIRE.\n"));
2100 allereg = (satype == SADB_SATYPE_UNSPEC);
2101 allreg = !allereg;
2102 /*
2103 * Corner case - if we send a regular ACQUIRE and there's
2104 * extended ones registered, don't send an error down to
2105 * consumers if nobody's listening and prematurely destroy
2106 * their ACQUIRE record. This might be too hackish of a
2107 * solution.
2108 */
2109 if (allreg && keystack->keystack_num_extended > 0)
2110 err = 0;
2111 break;
2112 case SADB_X_PROMISC:
2113 case SADB_X_INVERSE_ACQUIRE:
2114 case SADB_DUMP:
2115 case SADB_GET:
2116 default:
2117 /*
2118 * Deliver to the sender and promiscuous only.
2119 */
2120 ks3dbg(keystack, ("Delivering sender/promisc only (%d).\n",
2121 samsg->sadb_msg_type));
2122 break;
2123 }
2124
2125 mutex_enter(&keystack->keystack_list_lock);
2126 for (ks = keystack->keystack_list; ks != NULL; ks = ks->keysock_next) {
2127 /* Delivery loop. */
2128
2129 /*
2130 * Check special keysock-setting cases (REGISTER replies)
2131 * here.
2132 */
2133 if (setalg && serial == ks->keysock_serial) {
2134 ASSERT(kc != NULL);
2135 ASSERT(kc->kc_sa_type == satype);
2136 KEYSOCK_SETREG(ks, satype);
2137 }
2138
2139 /*
2140 * NOLOOP takes precedence over PROMISC. So if you've set
2141 * !SO_USELOOPBACK, don't expect to see any data...
2142 */
2143 if (ks->keysock_flags & KEYSOCK_NOLOOP)
2144 continue;
2145
2146 /*
2147 * Messages to all, or promiscuous sockets just GET the
2148 * message. Perform rules-type checking iff it's not for all
2149 * listeners or the socket is in promiscuous mode.
2150 *
2151 * NOTE:Because of the (kc != NULL && ISREG()), make sure
2152 * extended ACQUIREs arrive off a consumer that is
2153 * part of the extended REGISTER set of consumers.
2154 */
2155 if (serial != ks->keysock_serial &&
2156 !toall &&
2157 !(ks->keysock_flags & KEYSOCK_PROMISC) &&
2158 !((ks->keysock_flags & KEYSOCK_EXTENDED) ?
2159 allereg : allreg && kc != NULL &&
2160 KEYSOCK_ISREG(ks, kc->kc_sa_type)))
2161 continue;
2162
2163 mp1 = dupmsg(mp);
2164 if (mp1 == NULL) {
2165 ks2dbg(keystack, (
2166 "keysock_passup(): dupmsg() failed.\n"));
2167 mp1 = mp;
2168 mp = NULL;
2169 err = ENOMEM;
2170 }
2171
2172 /*
2173 * At this point, we can deliver or attempt to deliver
2174 * this message. We're free of obligation to report
2175 * no listening PF_KEY sockets. So set err to 0.
2176 */
2177 err = 0;
2178
2179 /*
2180 * See if we canputnext(), as well as see if the message
2181 * needs to be queued if we can't.
2182 */
2183 if (!canputnext(ks->keysock_rq)) {
2184 if (persistent) {
2185 if (putq(ks->keysock_rq, mp1) == 0) {
2186 ks1dbg(keystack, (
2187 "keysock_passup: putq failed.\n"));
2188 } else {
2189 continue;
2190 }
2191 }
2192 freemsg(mp1);
2193 continue;
2194 }
2195
2196 ks3dbg(keystack,
2197 ("Putting to serial %d.\n", ks->keysock_serial));
2198 /*
2199 * Unlike the specific keysock instance case, this
2200 * will only hit for listeners, so we will only
2201 * putnext() if we can.
2202 */
2203 putnext(ks->keysock_rq, mp1);
2204 if (mp == NULL)
2205 break; /* out of for loop. */
2206 }
2207 mutex_exit(&keystack->keystack_list_lock);
2208
2209 error:
2210 if ((err != 0) && (kc != NULL)) {
2211 /*
2212 * Generate KEYSOCK_OUT_ERR for consumer.
2213 * Basically, I send this back if I have not been able to
2214 * transmit (for whatever reason)
2215 */
2216 ks1dbg(keystack,
2217 ("keysock_passup(): No registered of type %d.\n",
2218 satype));
2219 if (mp != NULL) {
2220 if (mp->b_datap->db_type == M_PROTO) {
2221 mp1 = mp;
2222 mp = mp->b_cont;
2223 freeb(mp1);
2224 }
2225 /*
2226 * Do a copymsg() because people who get
2227 * KEYSOCK_OUT_ERR may alter the message contents.
2228 */
2229 mp1 = copymsg(mp);
2230 if (mp1 == NULL) {
2231 ks2dbg(keystack,
2232 ("keysock_passup: copymsg() failed.\n"));
2233 mp1 = mp;
2234 mp = NULL;
2235 }
2236 keysock_out_err(kc, err, mp1);
2237 }
2238 }
2239
2240 /*
2241 * XXX Blank the message somehow. This is difficult because we don't
2242 * know at this point if the message has db_ref > 1, etc.
2243 *
2244 * Optimally, keysock messages containing actual keying material would
2245 * be allocated with esballoc(), with a zeroing free function.
2246 */
2247 if (mp != NULL)
2248 freemsg(mp);
2249 }
2250
2251 /*
2252 * Keysock's read service procedure is there only for PF_KEY reply
2253 * messages that really need to reach the top.
2254 */
2255 static void
2256 keysock_rsrv(queue_t *q)
2257 {
2258 mblk_t *mp;
2259
2260 while ((mp = getq(q)) != NULL) {
2261 if (canputnext(q)) {
2262 putnext(q, mp);
2263 } else {
2264 (void) putbq(q, mp);
2265 return;
2266 }
2267 }
2268 }
2269
2270 /*
2271 * The read procedure should only be invoked by a keysock consumer, like
2272 * ESP, AH, etc. I should only see KEYSOCK_OUT and KEYSOCK_HELLO_ACK
2273 * messages on my read queues.
2274 */
2275 static void
2276 keysock_rput(queue_t *q, mblk_t *mp)
2277 {
2278 keysock_consumer_t *kc = (keysock_consumer_t *)q->q_ptr;
2279 ipsec_info_t *ii;
2280 keysock_hello_ack_t *ksa;
2281 minor_t serial;
2282 mblk_t *mp1;
2283 sadb_msg_t *samsg;
2284 keysock_stack_t *keystack = kc->kc_keystack;
2285
2286 /* Make sure I'm a consumer instance. (i.e. something's below me) */
2287 ASSERT(WR(q)->q_next != NULL);
2288
2289 if (mp->b_datap->db_type != M_CTL) {
2290 /*
2291 * Keysock should only see keysock consumer interface
2292 * messages (see ipsec_info.h) on its read procedure.
2293 * To be robust, however, putnext() up so the STREAM head can
2294 * deal with it appropriately.
2295 */
2296 ks1dbg(keystack,
2297 ("Hmmm, a non M_CTL (%d, 0x%x) on keysock_rput.\n",
2298 mp->b_datap->db_type, mp->b_datap->db_type));
2299 putnext(q, mp);
2300 return;
2301 }
2302
2303 ii = (ipsec_info_t *)mp->b_rptr;
2304
2305 switch (ii->ipsec_info_type) {
2306 case KEYSOCK_OUT:
2307 /*
2308 * A consumer needs to pass a response message or an ACQUIRE
2309 * UP. I assume that the consumer has done the right
2310 * thing w.r.t. message creation, etc.
2311 */
2312 serial = ((keysock_out_t *)mp->b_rptr)->ks_out_serial;
2313 mp1 = mp->b_cont; /* Get M_DATA portion. */
2314 freeb(mp);
2315 samsg = (sadb_msg_t *)mp1->b_rptr;
2316 if (samsg->sadb_msg_type == SADB_FLUSH ||
2317 (samsg->sadb_msg_type == SADB_DUMP &&
2318 samsg->sadb_msg_len == SADB_8TO64(sizeof (*samsg)))) {
2319 /*
2320 * If I'm an end-of-FLUSH or an end-of-DUMP marker...
2321 */
2322 ASSERT(keystack->keystack_flushdump != 0);
2323 /* Am I flushing? */
2324
2325 mutex_enter(&kc->kc_lock);
2326 kc->kc_flags &= ~KC_FLUSHING;
2327 mutex_exit(&kc->kc_lock);
2328
2329 if (samsg->sadb_msg_errno != 0)
2330 keystack->keystack_flushdump_errno =
2331 samsg->sadb_msg_errno;
2332
2333 /*
2334 * Lower the atomic "flushing" count. If it's
2335 * the last one, send up the end-of-{FLUSH,DUMP} to
2336 * the appropriate PF_KEY socket.
2337 */
2338 if (atomic_dec_32_nv(&keystack->keystack_flushdump) !=
2339 0) {
2340 ks1dbg(keystack,
2341 ("One flush/dump message back from %d,"
2342 " more to go.\n", samsg->sadb_msg_satype));
2343 freemsg(mp1);
2344 return;
2345 }
2346
2347 samsg->sadb_msg_errno =
2348 (uint8_t)keystack->keystack_flushdump_errno;
2349 if (samsg->sadb_msg_type == SADB_DUMP) {
2350 samsg->sadb_msg_seq = 0;
2351 }
2352 }
2353 keysock_passup(mp1, samsg, serial, kc,
2354 (samsg->sadb_msg_type == SADB_DUMP), keystack);
2355 return;
2356 case KEYSOCK_HELLO_ACK:
2357 /* Aha, now we can link in the consumer! */
2358 ksa = (keysock_hello_ack_t *)ii;
2359 keysock_link_consumer(ksa->ks_hello_satype, kc);
2360 freemsg(mp);
2361 return;
2362 default:
2363 ks1dbg(keystack, ("Hmmm, an IPsec info I'm not used to, 0x%x\n",
2364 ii->ipsec_info_type));
2365 putnext(q, mp);
2366 }
2367 }
2368
2369 /*
2370 * So we can avoid external linking problems....
2371 */
2372 boolean_t
2373 keysock_extended_reg(netstack_t *ns)
2374 {
2375 keysock_stack_t *keystack = ns->netstack_keysock;
2376
2377 return (keystack->keystack_num_extended != 0);
2378 }
2379
2380 uint32_t
2381 keysock_next_seq(netstack_t *ns)
2382 {
2383 keysock_stack_t *keystack = ns->netstack_keysock;
2384
2385 return (atomic_dec_32_nv(&keystack->keystack_acquire_seq));
2386 }