1 /*
2 * CDDL HEADER START
3 *
4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
7 *
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
12 *
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 *
19 * CDDL HEADER END
20 */
21 /*
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
24 *
25 * Copyright 2015 Garrett D'Amore <garrett@damore.org>
26 */
27
28 #include <sys/types.h>
29 #include <sys/stream.h>
30 #include <sys/stropts.h>
31 #include <sys/errno.h>
32 #include <sys/strlog.h>
33 #include <sys/tihdr.h>
34 #include <sys/socket.h>
35 #include <sys/ddi.h>
36 #include <sys/sunddi.h>
37 #include <sys/kmem.h>
38 #include <sys/zone.h>
39 #include <sys/sysmacros.h>
40 #include <sys/cmn_err.h>
41 #include <sys/vtrace.h>
42 #include <sys/debug.h>
43 #include <sys/atomic.h>
44 #include <sys/strsun.h>
45 #include <sys/random.h>
46 #include <netinet/in.h>
47 #include <net/if.h>
48 #include <netinet/ip6.h>
49 #include <net/pfkeyv2.h>
50 #include <net/pfpolicy.h>
51
52 #include <inet/common.h>
53 #include <inet/mi.h>
54 #include <inet/nd.h>
55 #include <inet/ip.h>
56 #include <inet/ip_impl.h>
57 #include <inet/ip6.h>
58 #include <inet/ip_if.h>
59 #include <inet/ip_ndp.h>
60 #include <inet/sadb.h>
61 #include <inet/ipsec_info.h>
62 #include <inet/ipsec_impl.h>
63 #include <inet/ipsecesp.h>
64 #include <inet/ipdrop.h>
65 #include <inet/tcp.h>
66 #include <sys/kstat.h>
67 #include <sys/policy.h>
68 #include <sys/strsun.h>
69 #include <sys/strsubr.h>
70 #include <inet/udp_impl.h>
71 #include <sys/taskq.h>
72 #include <sys/note.h>
73
74 #include <sys/tsol/tnet.h>
75
76 /*
77 * Table of ND variables supported by ipsecesp. These are loaded into
78 * ipsecesp_g_nd in ipsecesp_init_nd.
79 * All of these are alterable, within the min/max values given, at run time.
80 */
81 static ipsecespparam_t lcl_param_arr[] = {
82 /* min max value name */
83 { 0, 3, 0, "ipsecesp_debug"},
84 { 125, 32000, SADB_AGE_INTERVAL_DEFAULT, "ipsecesp_age_interval"},
85 { 1, 10, 1, "ipsecesp_reap_delay"},
86 { 1, SADB_MAX_REPLAY, 64, "ipsecesp_replay_size"},
87 { 1, 300, 15, "ipsecesp_acquire_timeout"},
88 { 1, 1800, 90, "ipsecesp_larval_timeout"},
89 /* Default lifetime values for ACQUIRE messages. */
90 { 0, 0xffffffffU, 0, "ipsecesp_default_soft_bytes"},
91 { 0, 0xffffffffU, 0, "ipsecesp_default_hard_bytes"},
92 { 0, 0xffffffffU, 24000, "ipsecesp_default_soft_addtime"},
93 { 0, 0xffffffffU, 28800, "ipsecesp_default_hard_addtime"},
94 { 0, 0xffffffffU, 0, "ipsecesp_default_soft_usetime"},
95 { 0, 0xffffffffU, 0, "ipsecesp_default_hard_usetime"},
96 { 0, 1, 0, "ipsecesp_log_unknown_spi"},
97 { 0, 2, 1, "ipsecesp_padding_check"},
98 { 0, 600, 20, "ipsecesp_nat_keepalive_interval"},
99 };
100 #define ipsecesp_debug ipsecesp_params[0].ipsecesp_param_value
101 #define ipsecesp_age_interval ipsecesp_params[1].ipsecesp_param_value
102 #define ipsecesp_age_int_max ipsecesp_params[1].ipsecesp_param_max
103 #define ipsecesp_reap_delay ipsecesp_params[2].ipsecesp_param_value
104 #define ipsecesp_replay_size ipsecesp_params[3].ipsecesp_param_value
105 #define ipsecesp_acquire_timeout \
106 ipsecesp_params[4].ipsecesp_param_value
107 #define ipsecesp_larval_timeout \
108 ipsecesp_params[5].ipsecesp_param_value
109 #define ipsecesp_default_soft_bytes \
110 ipsecesp_params[6].ipsecesp_param_value
111 #define ipsecesp_default_hard_bytes \
112 ipsecesp_params[7].ipsecesp_param_value
113 #define ipsecesp_default_soft_addtime \
114 ipsecesp_params[8].ipsecesp_param_value
115 #define ipsecesp_default_hard_addtime \
116 ipsecesp_params[9].ipsecesp_param_value
117 #define ipsecesp_default_soft_usetime \
118 ipsecesp_params[10].ipsecesp_param_value
119 #define ipsecesp_default_hard_usetime \
120 ipsecesp_params[11].ipsecesp_param_value
121 #define ipsecesp_log_unknown_spi \
122 ipsecesp_params[12].ipsecesp_param_value
123 #define ipsecesp_padding_check \
124 ipsecesp_params[13].ipsecesp_param_value
125 /* For ipsecesp_nat_keepalive_interval, see ipsecesp.h. */
126
127 #define esp0dbg(a) printf a
128 /* NOTE: != 0 instead of > 0 so lint doesn't complain. */
129 #define esp1dbg(espstack, a) if (espstack->ipsecesp_debug != 0) printf a
130 #define esp2dbg(espstack, a) if (espstack->ipsecesp_debug > 1) printf a
131 #define esp3dbg(espstack, a) if (espstack->ipsecesp_debug > 2) printf a
132
133 static int ipsecesp_open(queue_t *, dev_t *, int, int, cred_t *);
134 static int ipsecesp_close(queue_t *);
135 static void ipsecesp_wput(queue_t *, mblk_t *);
136 static void *ipsecesp_stack_init(netstackid_t stackid, netstack_t *ns);
137 static void ipsecesp_stack_fini(netstackid_t stackid, void *arg);
138 static void esp_send_acquire(ipsacq_t *, mblk_t *, netstack_t *);
139
140 static void esp_prepare_udp(netstack_t *, mblk_t *, ipha_t *);
141 static void esp_outbound_finish(mblk_t *, ip_xmit_attr_t *);
142 static void esp_inbound_restart(mblk_t *, ip_recv_attr_t *);
143
144 static boolean_t esp_register_out(uint32_t, uint32_t, uint_t,
145 ipsecesp_stack_t *, cred_t *);
146 static boolean_t esp_strip_header(mblk_t *, boolean_t, uint32_t,
147 kstat_named_t **, ipsecesp_stack_t *);
148 static mblk_t *esp_submit_req_inbound(mblk_t *, ip_recv_attr_t *,
149 ipsa_t *, uint_t);
150 static mblk_t *esp_submit_req_outbound(mblk_t *, ip_xmit_attr_t *,
151 ipsa_t *, uchar_t *, uint_t);
152
153 /* Setable in /etc/system */
154 uint32_t esp_hash_size = IPSEC_DEFAULT_HASH_SIZE;
155
156 static struct module_info info = {
157 5137, "ipsecesp", 0, INFPSZ, 65536, 1024
158 };
159
160 static struct qinit rinit = {
161 (pfi_t)putnext, NULL, ipsecesp_open, ipsecesp_close, NULL, &info,
162 NULL
163 };
164
165 static struct qinit winit = {
166 (pfi_t)ipsecesp_wput, NULL, ipsecesp_open, ipsecesp_close, NULL, &info,
167 NULL
168 };
169
170 struct streamtab ipsecespinfo = {
171 &rinit, &winit, NULL, NULL
172 };
173
174 static taskq_t *esp_taskq;
175
176 /*
177 * OTOH, this one is set at open/close, and I'm D_MTQPAIR for now.
178 *
179 * Question: Do I need this, given that all instance's esps->esps_wq point
180 * to IP?
181 *
182 * Answer: Yes, because I need to know which queue is BOUND to
183 * IPPROTO_ESP
184 */
185
186 /*
187 * Stats. This may eventually become a full-blown SNMP MIB once that spec
188 * stabilizes.
189 */
190
191 typedef struct esp_kstats_s {
192 kstat_named_t esp_stat_num_aalgs;
193 kstat_named_t esp_stat_good_auth;
194 kstat_named_t esp_stat_bad_auth;
195 kstat_named_t esp_stat_bad_padding;
196 kstat_named_t esp_stat_replay_failures;
197 kstat_named_t esp_stat_replay_early_failures;
198 kstat_named_t esp_stat_keysock_in;
199 kstat_named_t esp_stat_out_requests;
200 kstat_named_t esp_stat_acquire_requests;
201 kstat_named_t esp_stat_bytes_expired;
202 kstat_named_t esp_stat_out_discards;
203 kstat_named_t esp_stat_crypto_sync;
204 kstat_named_t esp_stat_crypto_async;
205 kstat_named_t esp_stat_crypto_failures;
206 kstat_named_t esp_stat_num_ealgs;
207 kstat_named_t esp_stat_bad_decrypt;
208 kstat_named_t esp_stat_sa_port_renumbers;
209 } esp_kstats_t;
210
211 /*
212 * espstack->esp_kstats is equal to espstack->esp_ksp->ks_data if
213 * kstat_create_netstack for espstack->esp_ksp succeeds, but when it
214 * fails, it will be NULL. Note this is done for all stack instances,
215 * so it *could* fail. hence a non-NULL checking is done for
216 * ESP_BUMP_STAT and ESP_DEBUMP_STAT
217 */
218 #define ESP_BUMP_STAT(espstack, x) \
219 do { \
220 if (espstack->esp_kstats != NULL) \
221 (espstack->esp_kstats->esp_stat_ ## x).value.ui64++; \
222 _NOTE(CONSTCOND) \
223 } while (0)
224
225 #define ESP_DEBUMP_STAT(espstack, x) \
226 do { \
227 if (espstack->esp_kstats != NULL) \
228 (espstack->esp_kstats->esp_stat_ ## x).value.ui64--; \
229 _NOTE(CONSTCOND) \
230 } while (0)
231
232 static int esp_kstat_update(kstat_t *, int);
233
234 static boolean_t
235 esp_kstat_init(ipsecesp_stack_t *espstack, netstackid_t stackid)
236 {
237 espstack->esp_ksp = kstat_create_netstack("ipsecesp", 0, "esp_stat",
238 "net", KSTAT_TYPE_NAMED,
239 sizeof (esp_kstats_t) / sizeof (kstat_named_t),
240 KSTAT_FLAG_PERSISTENT, stackid);
241
242 if (espstack->esp_ksp == NULL || espstack->esp_ksp->ks_data == NULL)
243 return (B_FALSE);
244
245 espstack->esp_kstats = espstack->esp_ksp->ks_data;
246
247 espstack->esp_ksp->ks_update = esp_kstat_update;
248 espstack->esp_ksp->ks_private = (void *)(uintptr_t)stackid;
249
250 #define K64 KSTAT_DATA_UINT64
251 #define KI(x) kstat_named_init(&(espstack->esp_kstats->esp_stat_##x), #x, K64)
252
253 KI(num_aalgs);
254 KI(num_ealgs);
255 KI(good_auth);
256 KI(bad_auth);
257 KI(bad_padding);
258 KI(replay_failures);
259 KI(replay_early_failures);
260 KI(keysock_in);
261 KI(out_requests);
262 KI(acquire_requests);
263 KI(bytes_expired);
264 KI(out_discards);
265 KI(crypto_sync);
266 KI(crypto_async);
267 KI(crypto_failures);
268 KI(bad_decrypt);
269 KI(sa_port_renumbers);
270
271 #undef KI
272 #undef K64
273
274 kstat_install(espstack->esp_ksp);
275
276 return (B_TRUE);
277 }
278
279 static int
280 esp_kstat_update(kstat_t *kp, int rw)
281 {
282 esp_kstats_t *ekp;
283 netstackid_t stackid = (zoneid_t)(uintptr_t)kp->ks_private;
284 netstack_t *ns;
285 ipsec_stack_t *ipss;
286
287 if ((kp == NULL) || (kp->ks_data == NULL))
288 return (EIO);
289
290 if (rw == KSTAT_WRITE)
291 return (EACCES);
292
293 ns = netstack_find_by_stackid(stackid);
294 if (ns == NULL)
295 return (-1);
296 ipss = ns->netstack_ipsec;
297 if (ipss == NULL) {
298 netstack_rele(ns);
299 return (-1);
300 }
301 ekp = (esp_kstats_t *)kp->ks_data;
302
303 mutex_enter(&ipss->ipsec_alg_lock);
304 ekp->esp_stat_num_aalgs.value.ui64 =
305 ipss->ipsec_nalgs[IPSEC_ALG_AUTH];
306 ekp->esp_stat_num_ealgs.value.ui64 =
307 ipss->ipsec_nalgs[IPSEC_ALG_ENCR];
308 mutex_exit(&ipss->ipsec_alg_lock);
309
310 netstack_rele(ns);
311 return (0);
312 }
313
314 #ifdef DEBUG
315 /*
316 * Debug routine, useful to see pre-encryption data.
317 */
318 static char *
319 dump_msg(mblk_t *mp)
320 {
321 char tmp_str[3], tmp_line[256];
322
323 while (mp != NULL) {
324 unsigned char *ptr;
325
326 printf("mblk address 0x%p, length %ld, db_ref %d "
327 "type %d, base 0x%p, lim 0x%p\n",
328 (void *) mp, (long)(mp->b_wptr - mp->b_rptr),
329 mp->b_datap->db_ref, mp->b_datap->db_type,
330 (void *)mp->b_datap->db_base, (void *)mp->b_datap->db_lim);
331 ptr = mp->b_rptr;
332
333 tmp_line[0] = '\0';
334 while (ptr < mp->b_wptr) {
335 uint_t diff;
336
337 diff = (ptr - mp->b_rptr);
338 if (!(diff & 0x1f)) {
339 if (strlen(tmp_line) > 0) {
340 printf("bytes: %s\n", tmp_line);
341 tmp_line[0] = '\0';
342 }
343 }
344 if (!(diff & 0x3))
345 (void) strcat(tmp_line, " ");
346 (void) sprintf(tmp_str, "%02x", *ptr);
347 (void) strcat(tmp_line, tmp_str);
348 ptr++;
349 }
350 if (strlen(tmp_line) > 0)
351 printf("bytes: %s\n", tmp_line);
352
353 mp = mp->b_cont;
354 }
355
356 return ("\n");
357 }
358
359 #else /* DEBUG */
360 static char *
361 dump_msg(mblk_t *mp)
362 {
363 printf("Find value of mp %p.\n", mp);
364 return ("\n");
365 }
366 #endif /* DEBUG */
367
368 /*
369 * Don't have to lock age_interval, as only one thread will access it at
370 * a time, because I control the one function that does with timeout().
371 */
372 static void
373 esp_ager(void *arg)
374 {
375 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)arg;
376 netstack_t *ns = espstack->ipsecesp_netstack;
377 hrtime_t begin = gethrtime();
378
379 sadb_ager(&espstack->esp_sadb.s_v4, espstack->esp_pfkey_q,
380 espstack->ipsecesp_reap_delay, ns);
381 sadb_ager(&espstack->esp_sadb.s_v6, espstack->esp_pfkey_q,
382 espstack->ipsecesp_reap_delay, ns);
383
384 espstack->esp_event = sadb_retimeout(begin, espstack->esp_pfkey_q,
385 esp_ager, espstack,
386 &espstack->ipsecesp_age_interval, espstack->ipsecesp_age_int_max,
387 info.mi_idnum);
388 }
389
390 /*
391 * Get an ESP NDD parameter.
392 */
393 /* ARGSUSED */
394 static int
395 ipsecesp_param_get(q, mp, cp, cr)
396 queue_t *q;
397 mblk_t *mp;
398 caddr_t cp;
399 cred_t *cr;
400 {
401 ipsecespparam_t *ipsecesppa = (ipsecespparam_t *)cp;
402 uint_t value;
403 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)q->q_ptr;
404
405 mutex_enter(&espstack->ipsecesp_param_lock);
406 value = ipsecesppa->ipsecesp_param_value;
407 mutex_exit(&espstack->ipsecesp_param_lock);
408
409 (void) mi_mpprintf(mp, "%u", value);
410 return (0);
411 }
412
413 /*
414 * This routine sets an NDD variable in a ipsecespparam_t structure.
415 */
416 /* ARGSUSED */
417 static int
418 ipsecesp_param_set(q, mp, value, cp, cr)
419 queue_t *q;
420 mblk_t *mp;
421 char *value;
422 caddr_t cp;
423 cred_t *cr;
424 {
425 ulong_t new_value;
426 ipsecespparam_t *ipsecesppa = (ipsecespparam_t *)cp;
427 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)q->q_ptr;
428
429 /*
430 * Fail the request if the new value does not lie within the
431 * required bounds.
432 */
433 if (ddi_strtoul(value, NULL, 10, &new_value) != 0 ||
434 new_value < ipsecesppa->ipsecesp_param_min ||
435 new_value > ipsecesppa->ipsecesp_param_max) {
436 return (EINVAL);
437 }
438
439 /* Set the new value */
440 mutex_enter(&espstack->ipsecesp_param_lock);
441 ipsecesppa->ipsecesp_param_value = new_value;
442 mutex_exit(&espstack->ipsecesp_param_lock);
443 return (0);
444 }
445
446 /*
447 * Using lifetime NDD variables, fill in an extended combination's
448 * lifetime information.
449 */
450 void
451 ipsecesp_fill_defs(sadb_x_ecomb_t *ecomb, netstack_t *ns)
452 {
453 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
454
455 ecomb->sadb_x_ecomb_soft_bytes = espstack->ipsecesp_default_soft_bytes;
456 ecomb->sadb_x_ecomb_hard_bytes = espstack->ipsecesp_default_hard_bytes;
457 ecomb->sadb_x_ecomb_soft_addtime =
458 espstack->ipsecesp_default_soft_addtime;
459 ecomb->sadb_x_ecomb_hard_addtime =
460 espstack->ipsecesp_default_hard_addtime;
461 ecomb->sadb_x_ecomb_soft_usetime =
462 espstack->ipsecesp_default_soft_usetime;
463 ecomb->sadb_x_ecomb_hard_usetime =
464 espstack->ipsecesp_default_hard_usetime;
465 }
466
467 /*
468 * Initialize things for ESP at module load time.
469 */
470 boolean_t
471 ipsecesp_ddi_init(void)
472 {
473 esp_taskq = taskq_create("esp_taskq", 1, minclsyspri,
474 IPSEC_TASKQ_MIN, IPSEC_TASKQ_MAX, 0);
475
476 /*
477 * We want to be informed each time a stack is created or
478 * destroyed in the kernel, so we can maintain the
479 * set of ipsecesp_stack_t's.
480 */
481 netstack_register(NS_IPSECESP, ipsecesp_stack_init, NULL,
482 ipsecesp_stack_fini);
483
484 return (B_TRUE);
485 }
486
487 /*
488 * Walk through the param array specified registering each element with the
489 * named dispatch handler.
490 */
491 static boolean_t
492 ipsecesp_param_register(IDP *ndp, ipsecespparam_t *espp, int cnt)
493 {
494 for (; cnt-- > 0; espp++) {
495 if (espp->ipsecesp_param_name != NULL &&
496 espp->ipsecesp_param_name[0]) {
497 if (!nd_load(ndp,
498 espp->ipsecesp_param_name,
499 ipsecesp_param_get, ipsecesp_param_set,
500 (caddr_t)espp)) {
501 nd_free(ndp);
502 return (B_FALSE);
503 }
504 }
505 }
506 return (B_TRUE);
507 }
508 /*
509 * Initialize things for ESP for each stack instance
510 */
511 static void *
512 ipsecesp_stack_init(netstackid_t stackid, netstack_t *ns)
513 {
514 ipsecesp_stack_t *espstack;
515 ipsecespparam_t *espp;
516
517 espstack = (ipsecesp_stack_t *)kmem_zalloc(sizeof (*espstack),
518 KM_SLEEP);
519 espstack->ipsecesp_netstack = ns;
520
521 espp = (ipsecespparam_t *)kmem_alloc(sizeof (lcl_param_arr), KM_SLEEP);
522 espstack->ipsecesp_params = espp;
523 bcopy(lcl_param_arr, espp, sizeof (lcl_param_arr));
524
525 (void) ipsecesp_param_register(&espstack->ipsecesp_g_nd, espp,
526 A_CNT(lcl_param_arr));
527
528 (void) esp_kstat_init(espstack, stackid);
529
530 espstack->esp_sadb.s_acquire_timeout =
531 &espstack->ipsecesp_acquire_timeout;
532 espstack->esp_sadb.s_acqfn = esp_send_acquire;
533 sadbp_init("ESP", &espstack->esp_sadb, SADB_SATYPE_ESP, esp_hash_size,
534 espstack->ipsecesp_netstack);
535
536 mutex_init(&espstack->ipsecesp_param_lock, NULL, MUTEX_DEFAULT, 0);
537
538 ip_drop_register(&espstack->esp_dropper, "IPsec ESP");
539 return (espstack);
540 }
541
542 /*
543 * Destroy things for ESP at module unload time.
544 */
545 void
546 ipsecesp_ddi_destroy(void)
547 {
548 netstack_unregister(NS_IPSECESP);
549 taskq_destroy(esp_taskq);
550 }
551
552 /*
553 * Destroy things for ESP for one stack instance
554 */
555 static void
556 ipsecesp_stack_fini(netstackid_t stackid, void *arg)
557 {
558 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)arg;
559
560 if (espstack->esp_pfkey_q != NULL) {
561 (void) quntimeout(espstack->esp_pfkey_q, espstack->esp_event);
562 }
563 espstack->esp_sadb.s_acqfn = NULL;
564 espstack->esp_sadb.s_acquire_timeout = NULL;
565 sadbp_destroy(&espstack->esp_sadb, espstack->ipsecesp_netstack);
566 ip_drop_unregister(&espstack->esp_dropper);
567 mutex_destroy(&espstack->ipsecesp_param_lock);
568 nd_free(&espstack->ipsecesp_g_nd);
569
570 kmem_free(espstack->ipsecesp_params, sizeof (lcl_param_arr));
571 espstack->ipsecesp_params = NULL;
572 kstat_delete_netstack(espstack->esp_ksp, stackid);
573 espstack->esp_ksp = NULL;
574 espstack->esp_kstats = NULL;
575 kmem_free(espstack, sizeof (*espstack));
576 }
577
578 /*
579 * ESP module open routine, which is here for keysock plumbing.
580 * Keysock is pushed over {AH,ESP} which is an artifact from the Bad Old
581 * Days of export control, and fears that ESP would not be allowed
582 * to be shipped at all by default. Eventually, keysock should
583 * either access AH and ESP via modstubs or krtld dependencies, or
584 * perhaps be folded in with AH and ESP into a single IPsec/netsec
585 * module ("netsec" if PF_KEY provides more than AH/ESP keying tables).
586 */
587 /* ARGSUSED */
588 static int
589 ipsecesp_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
590 {
591 netstack_t *ns;
592 ipsecesp_stack_t *espstack;
593
594 if (secpolicy_ip_config(credp, B_FALSE) != 0)
595 return (EPERM);
596
597 if (q->q_ptr != NULL)
598 return (0); /* Re-open of an already open instance. */
599
600 if (sflag != MODOPEN)
601 return (EINVAL);
602
603 ns = netstack_find_by_cred(credp);
604 ASSERT(ns != NULL);
605 espstack = ns->netstack_ipsecesp;
606 ASSERT(espstack != NULL);
607
608 q->q_ptr = espstack;
609 WR(q)->q_ptr = q->q_ptr;
610
611 qprocson(q);
612 return (0);
613 }
614
615 /*
616 * ESP module close routine.
617 */
618 static int
619 ipsecesp_close(queue_t *q)
620 {
621 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)q->q_ptr;
622
623 /*
624 * Clean up q_ptr, if needed.
625 */
626 qprocsoff(q);
627
628 /* Keysock queue check is safe, because of OCEXCL perimeter. */
629
630 if (q == espstack->esp_pfkey_q) {
631 esp1dbg(espstack,
632 ("ipsecesp_close: Ummm... keysock is closing ESP.\n"));
633 espstack->esp_pfkey_q = NULL;
634 /* Detach qtimeouts. */
635 (void) quntimeout(q, espstack->esp_event);
636 }
637
638 netstack_rele(espstack->ipsecesp_netstack);
639 return (0);
640 }
641
642 /*
643 * Add a number of bytes to what the SA has protected so far. Return
644 * B_TRUE if the SA can still protect that many bytes.
645 *
646 * Caller must REFRELE the passed-in assoc. This function must REFRELE
647 * any obtained peer SA.
648 */
649 static boolean_t
650 esp_age_bytes(ipsa_t *assoc, uint64_t bytes, boolean_t inbound)
651 {
652 ipsa_t *inassoc, *outassoc;
653 isaf_t *bucket;
654 boolean_t inrc, outrc, isv6;
655 sadb_t *sp;
656 int outhash;
657 netstack_t *ns = assoc->ipsa_netstack;
658 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
659
660 /* No peer? No problem! */
661 if (!assoc->ipsa_haspeer) {
662 return (sadb_age_bytes(espstack->esp_pfkey_q, assoc, bytes,
663 B_TRUE));
664 }
665
666 /*
667 * Otherwise, we want to grab both the original assoc and its peer.
668 * There might be a race for this, but if it's a real race, two
669 * expire messages may occur. We limit this by only sending the
670 * expire message on one of the peers, we'll pick the inbound
671 * arbitrarily.
672 *
673 * If we need tight synchronization on the peer SA, then we need to
674 * reconsider.
675 */
676
677 /* Use address length to select IPv6/IPv4 */
678 isv6 = (assoc->ipsa_addrfam == AF_INET6);
679 sp = isv6 ? &espstack->esp_sadb.s_v6 : &espstack->esp_sadb.s_v4;
680
681 if (inbound) {
682 inassoc = assoc;
683 if (isv6) {
684 outhash = OUTBOUND_HASH_V6(sp, *((in6_addr_t *)
685 &inassoc->ipsa_dstaddr));
686 } else {
687 outhash = OUTBOUND_HASH_V4(sp, *((ipaddr_t *)
688 &inassoc->ipsa_dstaddr));
689 }
690 bucket = &sp->sdb_of[outhash];
691 mutex_enter(&bucket->isaf_lock);
692 outassoc = ipsec_getassocbyspi(bucket, inassoc->ipsa_spi,
693 inassoc->ipsa_srcaddr, inassoc->ipsa_dstaddr,
694 inassoc->ipsa_addrfam);
695 mutex_exit(&bucket->isaf_lock);
696 if (outassoc == NULL) {
697 /* Q: Do we wish to set haspeer == B_FALSE? */
698 esp0dbg(("esp_age_bytes: "
699 "can't find peer for inbound.\n"));
700 return (sadb_age_bytes(espstack->esp_pfkey_q, inassoc,
701 bytes, B_TRUE));
702 }
703 } else {
704 outassoc = assoc;
705 bucket = INBOUND_BUCKET(sp, outassoc->ipsa_spi);
706 mutex_enter(&bucket->isaf_lock);
707 inassoc = ipsec_getassocbyspi(bucket, outassoc->ipsa_spi,
708 outassoc->ipsa_srcaddr, outassoc->ipsa_dstaddr,
709 outassoc->ipsa_addrfam);
710 mutex_exit(&bucket->isaf_lock);
711 if (inassoc == NULL) {
712 /* Q: Do we wish to set haspeer == B_FALSE? */
713 esp0dbg(("esp_age_bytes: "
714 "can't find peer for outbound.\n"));
715 return (sadb_age_bytes(espstack->esp_pfkey_q, outassoc,
716 bytes, B_TRUE));
717 }
718 }
719
720 inrc = sadb_age_bytes(espstack->esp_pfkey_q, inassoc, bytes, B_TRUE);
721 outrc = sadb_age_bytes(espstack->esp_pfkey_q, outassoc, bytes, B_FALSE);
722
723 /*
724 * REFRELE any peer SA.
725 *
726 * Because of the multi-line macro nature of IPSA_REFRELE, keep
727 * them in { }.
728 */
729 if (inbound) {
730 IPSA_REFRELE(outassoc);
731 } else {
732 IPSA_REFRELE(inassoc);
733 }
734
735 return (inrc && outrc);
736 }
737
738 /*
739 * Do incoming NAT-T manipulations for packet.
740 * Returns NULL if the mblk chain is consumed.
741 */
742 static mblk_t *
743 esp_fix_natt_checksums(mblk_t *data_mp, ipsa_t *assoc)
744 {
745 ipha_t *ipha = (ipha_t *)data_mp->b_rptr;
746 tcpha_t *tcpha;
747 udpha_t *udpha;
748 /* Initialize to our inbound cksum adjustment... */
749 uint32_t sum = assoc->ipsa_inbound_cksum;
750
751 switch (ipha->ipha_protocol) {
752 case IPPROTO_TCP:
753 tcpha = (tcpha_t *)(data_mp->b_rptr +
754 IPH_HDR_LENGTH(ipha));
755
756 #define DOWN_SUM(x) (x) = ((x) & 0xFFFF) + ((x) >> 16)
757 sum += ~ntohs(tcpha->tha_sum) & 0xFFFF;
758 DOWN_SUM(sum);
759 DOWN_SUM(sum);
760 tcpha->tha_sum = ~htons(sum);
761 break;
762 case IPPROTO_UDP:
763 udpha = (udpha_t *)(data_mp->b_rptr + IPH_HDR_LENGTH(ipha));
764
765 if (udpha->uha_checksum != 0) {
766 /* Adujst if the inbound one was not zero. */
767 sum += ~ntohs(udpha->uha_checksum) & 0xFFFF;
768 DOWN_SUM(sum);
769 DOWN_SUM(sum);
770 udpha->uha_checksum = ~htons(sum);
771 if (udpha->uha_checksum == 0)
772 udpha->uha_checksum = 0xFFFF;
773 }
774 #undef DOWN_SUM
775 break;
776 case IPPROTO_IP:
777 /*
778 * This case is only an issue for self-encapsulated
779 * packets. So for now, fall through.
780 */
781 break;
782 }
783 return (data_mp);
784 }
785
786
787 /*
788 * Strip ESP header, check padding, and fix IP header.
789 * Returns B_TRUE on success, B_FALSE if an error occured.
790 */
791 static boolean_t
792 esp_strip_header(mblk_t *data_mp, boolean_t isv4, uint32_t ivlen,
793 kstat_named_t **counter, ipsecesp_stack_t *espstack)
794 {
795 ipha_t *ipha;
796 ip6_t *ip6h;
797 uint_t divpoint;
798 mblk_t *scratch;
799 uint8_t nexthdr, padlen;
800 uint8_t lastpad;
801 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec;
802 uint8_t *lastbyte;
803
804 /*
805 * Strip ESP data and fix IP header.
806 *
807 * XXX In case the beginning of esp_inbound() changes to not do a
808 * pullup, this part of the code can remain unchanged.
809 */
810 if (isv4) {
811 ASSERT((data_mp->b_wptr - data_mp->b_rptr) >= sizeof (ipha_t));
812 ipha = (ipha_t *)data_mp->b_rptr;
813 ASSERT((data_mp->b_wptr - data_mp->b_rptr) >= sizeof (esph_t) +
814 IPH_HDR_LENGTH(ipha));
815 divpoint = IPH_HDR_LENGTH(ipha);
816 } else {
817 ASSERT((data_mp->b_wptr - data_mp->b_rptr) >= sizeof (ip6_t));
818 ip6h = (ip6_t *)data_mp->b_rptr;
819 divpoint = ip_hdr_length_v6(data_mp, ip6h);
820 }
821
822 scratch = data_mp;
823 while (scratch->b_cont != NULL)
824 scratch = scratch->b_cont;
825
826 ASSERT((scratch->b_wptr - scratch->b_rptr) >= 3);
827
828 /*
829 * "Next header" and padding length are the last two bytes in the
830 * ESP-protected datagram, thus the explicit - 1 and - 2.
831 * lastpad is the last byte of the padding, which can be used for
832 * a quick check to see if the padding is correct.
833 */
834 lastbyte = scratch->b_wptr - 1;
835 nexthdr = *lastbyte--;
836 padlen = *lastbyte--;
837
838 if (isv4) {
839 /* Fix part of the IP header. */
840 ipha->ipha_protocol = nexthdr;
841 /*
842 * Reality check the padlen. The explicit - 2 is for the
843 * padding length and the next-header bytes.
844 */
845 if (padlen >= ntohs(ipha->ipha_length) - sizeof (ipha_t) - 2 -
846 sizeof (esph_t) - ivlen) {
847 ESP_BUMP_STAT(espstack, bad_decrypt);
848 ipsec_rl_strlog(espstack->ipsecesp_netstack,
849 info.mi_idnum, 0, 0,
850 SL_ERROR | SL_WARN,
851 "Corrupt ESP packet (padlen too big).\n");
852 esp1dbg(espstack, ("padlen (%d) is greater than:\n",
853 padlen));
854 esp1dbg(espstack, ("pkt len(%d) - ip hdr - esp "
855 "hdr - ivlen(%d) = %d.\n",
856 ntohs(ipha->ipha_length), ivlen,
857 (int)(ntohs(ipha->ipha_length) - sizeof (ipha_t) -
858 2 - sizeof (esph_t) - ivlen)));
859 *counter = DROPPER(ipss, ipds_esp_bad_padlen);
860 return (B_FALSE);
861 }
862
863 /*
864 * Fix the rest of the header. The explicit - 2 is for the
865 * padding length and the next-header bytes.
866 */
867 ipha->ipha_length = htons(ntohs(ipha->ipha_length) - padlen -
868 2 - sizeof (esph_t) - ivlen);
869 ipha->ipha_hdr_checksum = 0;
870 ipha->ipha_hdr_checksum = (uint16_t)ip_csum_hdr(ipha);
871 } else {
872 if (ip6h->ip6_nxt == IPPROTO_ESP) {
873 ip6h->ip6_nxt = nexthdr;
874 } else {
875 ip_pkt_t ipp;
876
877 bzero(&ipp, sizeof (ipp));
878 (void) ip_find_hdr_v6(data_mp, ip6h, B_FALSE, &ipp,
879 NULL);
880 if (ipp.ipp_dstopts != NULL) {
881 ipp.ipp_dstopts->ip6d_nxt = nexthdr;
882 } else if (ipp.ipp_rthdr != NULL) {
883 ipp.ipp_rthdr->ip6r_nxt = nexthdr;
884 } else if (ipp.ipp_hopopts != NULL) {
885 ipp.ipp_hopopts->ip6h_nxt = nexthdr;
886 } else {
887 /* Panic a DEBUG kernel. */
888 ASSERT(ipp.ipp_hopopts != NULL);
889 /* Otherwise, pretend it's IP + ESP. */
890 cmn_err(CE_WARN, "ESP IPv6 headers wrong.\n");
891 ip6h->ip6_nxt = nexthdr;
892 }
893 }
894
895 if (padlen >= ntohs(ip6h->ip6_plen) - 2 - sizeof (esph_t) -
896 ivlen) {
897 ESP_BUMP_STAT(espstack, bad_decrypt);
898 ipsec_rl_strlog(espstack->ipsecesp_netstack,
899 info.mi_idnum, 0, 0,
900 SL_ERROR | SL_WARN,
901 "Corrupt ESP packet (v6 padlen too big).\n");
902 esp1dbg(espstack, ("padlen (%d) is greater than:\n",
903 padlen));
904 esp1dbg(espstack,
905 ("pkt len(%u) - ip hdr - esp hdr - ivlen(%d) = "
906 "%u.\n", (unsigned)(ntohs(ip6h->ip6_plen)
907 + sizeof (ip6_t)), ivlen,
908 (unsigned)(ntohs(ip6h->ip6_plen) - 2 -
909 sizeof (esph_t) - ivlen)));
910 *counter = DROPPER(ipss, ipds_esp_bad_padlen);
911 return (B_FALSE);
912 }
913
914
915 /*
916 * Fix the rest of the header. The explicit - 2 is for the
917 * padding length and the next-header bytes. IPv6 is nice,
918 * because there's no hdr checksum!
919 */
920 ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) - padlen -
921 2 - sizeof (esph_t) - ivlen);
922 }
923
924 if (espstack->ipsecesp_padding_check > 0 && padlen > 0) {
925 /*
926 * Weak padding check: compare last-byte to length, they
927 * should be equal.
928 */
929 lastpad = *lastbyte--;
930
931 if (padlen != lastpad) {
932 ipsec_rl_strlog(espstack->ipsecesp_netstack,
933 info.mi_idnum, 0, 0, SL_ERROR | SL_WARN,
934 "Corrupt ESP packet (lastpad != padlen).\n");
935 esp1dbg(espstack,
936 ("lastpad (%d) not equal to padlen (%d):\n",
937 lastpad, padlen));
938 ESP_BUMP_STAT(espstack, bad_padding);
939 *counter = DROPPER(ipss, ipds_esp_bad_padding);
940 return (B_FALSE);
941 }
942
943 /*
944 * Strong padding check: Check all pad bytes to see that
945 * they're ascending. Go backwards using a descending counter
946 * to verify. padlen == 1 is checked by previous block, so
947 * only bother if we've more than 1 byte of padding.
948 * Consequently, start the check one byte before the location
949 * of "lastpad".
950 */
951 if (espstack->ipsecesp_padding_check > 1) {
952 /*
953 * This assert may have to become an if and a pullup
954 * if we start accepting multi-dblk mblks. For now,
955 * though, any packet here will have been pulled up in
956 * esp_inbound.
957 */
958 ASSERT(MBLKL(scratch) >= lastpad + 3);
959
960 /*
961 * Use "--lastpad" because we already checked the very
962 * last pad byte previously.
963 */
964 while (--lastpad != 0) {
965 if (lastpad != *lastbyte) {
966 ipsec_rl_strlog(
967 espstack->ipsecesp_netstack,
968 info.mi_idnum, 0, 0,
969 SL_ERROR | SL_WARN, "Corrupt ESP "
970 "packet (bad padding).\n");
971 esp1dbg(espstack,
972 ("padding not in correct"
973 " format:\n"));
974 ESP_BUMP_STAT(espstack, bad_padding);
975 *counter = DROPPER(ipss,
976 ipds_esp_bad_padding);
977 return (B_FALSE);
978 }
979 lastbyte--;
980 }
981 }
982 }
983
984 /* Trim off the padding. */
985 ASSERT(data_mp->b_cont == NULL);
986 data_mp->b_wptr -= (padlen + 2);
987
988 /*
989 * Remove the ESP header.
990 *
991 * The above assertions about data_mp's size will make this work.
992 *
993 * XXX Question: If I send up and get back a contiguous mblk,
994 * would it be quicker to bcopy over, or keep doing the dupb stuff?
995 * I go with copying for now.
996 */
997
998 if (IS_P2ALIGNED(data_mp->b_rptr, sizeof (uint32_t)) &&
999 IS_P2ALIGNED(ivlen, sizeof (uint32_t))) {
1000 uint8_t *start = data_mp->b_rptr;
1001 uint32_t *src, *dst;
1002
1003 src = (uint32_t *)(start + divpoint);
1004 dst = (uint32_t *)(start + divpoint + sizeof (esph_t) + ivlen);
1005
1006 ASSERT(IS_P2ALIGNED(dst, sizeof (uint32_t)) &&
1007 IS_P2ALIGNED(src, sizeof (uint32_t)));
1008
1009 do {
1010 src--;
1011 dst--;
1012 *dst = *src;
1013 } while (src != (uint32_t *)start);
1014
1015 data_mp->b_rptr = (uchar_t *)dst;
1016 } else {
1017 uint8_t *start = data_mp->b_rptr;
1018 uint8_t *src, *dst;
1019
1020 src = start + divpoint;
1021 dst = src + sizeof (esph_t) + ivlen;
1022
1023 do {
1024 src--;
1025 dst--;
1026 *dst = *src;
1027 } while (src != start);
1028
1029 data_mp->b_rptr = dst;
1030 }
1031
1032 esp2dbg(espstack, ("data_mp after inbound ESP adjustment:\n"));
1033 esp2dbg(espstack, (dump_msg(data_mp)));
1034
1035 return (B_TRUE);
1036 }
1037
1038 /*
1039 * Updating use times can be tricky business if the ipsa_haspeer flag is
1040 * set. This function is called once in an SA's lifetime.
1041 *
1042 * Caller has to REFRELE "assoc" which is passed in. This function has
1043 * to REFRELE any peer SA that is obtained.
1044 */
1045 static void
1046 esp_set_usetime(ipsa_t *assoc, boolean_t inbound)
1047 {
1048 ipsa_t *inassoc, *outassoc;
1049 isaf_t *bucket;
1050 sadb_t *sp;
1051 int outhash;
1052 boolean_t isv6;
1053 netstack_t *ns = assoc->ipsa_netstack;
1054 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1055
1056 /* No peer? No problem! */
1057 if (!assoc->ipsa_haspeer) {
1058 sadb_set_usetime(assoc);
1059 return;
1060 }
1061
1062 /*
1063 * Otherwise, we want to grab both the original assoc and its peer.
1064 * There might be a race for this, but if it's a real race, the times
1065 * will be out-of-synch by at most a second, and since our time
1066 * granularity is a second, this won't be a problem.
1067 *
1068 * If we need tight synchronization on the peer SA, then we need to
1069 * reconsider.
1070 */
1071
1072 /* Use address length to select IPv6/IPv4 */
1073 isv6 = (assoc->ipsa_addrfam == AF_INET6);
1074 sp = isv6 ? &espstack->esp_sadb.s_v6 : &espstack->esp_sadb.s_v4;
1075
1076 if (inbound) {
1077 inassoc = assoc;
1078 if (isv6) {
1079 outhash = OUTBOUND_HASH_V6(sp, *((in6_addr_t *)
1080 &inassoc->ipsa_dstaddr));
1081 } else {
1082 outhash = OUTBOUND_HASH_V4(sp, *((ipaddr_t *)
1083 &inassoc->ipsa_dstaddr));
1084 }
1085 bucket = &sp->sdb_of[outhash];
1086 mutex_enter(&bucket->isaf_lock);
1087 outassoc = ipsec_getassocbyspi(bucket, inassoc->ipsa_spi,
1088 inassoc->ipsa_srcaddr, inassoc->ipsa_dstaddr,
1089 inassoc->ipsa_addrfam);
1090 mutex_exit(&bucket->isaf_lock);
1091 if (outassoc == NULL) {
1092 /* Q: Do we wish to set haspeer == B_FALSE? */
1093 esp0dbg(("esp_set_usetime: "
1094 "can't find peer for inbound.\n"));
1095 sadb_set_usetime(inassoc);
1096 return;
1097 }
1098 } else {
1099 outassoc = assoc;
1100 bucket = INBOUND_BUCKET(sp, outassoc->ipsa_spi);
1101 mutex_enter(&bucket->isaf_lock);
1102 inassoc = ipsec_getassocbyspi(bucket, outassoc->ipsa_spi,
1103 outassoc->ipsa_srcaddr, outassoc->ipsa_dstaddr,
1104 outassoc->ipsa_addrfam);
1105 mutex_exit(&bucket->isaf_lock);
1106 if (inassoc == NULL) {
1107 /* Q: Do we wish to set haspeer == B_FALSE? */
1108 esp0dbg(("esp_set_usetime: "
1109 "can't find peer for outbound.\n"));
1110 sadb_set_usetime(outassoc);
1111 return;
1112 }
1113 }
1114
1115 /* Update usetime on both. */
1116 sadb_set_usetime(inassoc);
1117 sadb_set_usetime(outassoc);
1118
1119 /*
1120 * REFRELE any peer SA.
1121 *
1122 * Because of the multi-line macro nature of IPSA_REFRELE, keep
1123 * them in { }.
1124 */
1125 if (inbound) {
1126 IPSA_REFRELE(outassoc);
1127 } else {
1128 IPSA_REFRELE(inassoc);
1129 }
1130 }
1131
1132 /*
1133 * Handle ESP inbound data for IPv4 and IPv6.
1134 * On success returns B_TRUE, on failure returns B_FALSE and frees the
1135 * mblk chain data_mp.
1136 */
1137 mblk_t *
1138 esp_inbound(mblk_t *data_mp, void *arg, ip_recv_attr_t *ira)
1139 {
1140 esph_t *esph = (esph_t *)arg;
1141 ipsa_t *ipsa = ira->ira_ipsec_esp_sa;
1142 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
1143 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1144 ipsec_stack_t *ipss = ns->netstack_ipsec;
1145
1146 /*
1147 * We may wish to check replay in-range-only here as an optimization.
1148 * Include the reality check of ipsa->ipsa_replay >
1149 * ipsa->ipsa_replay_wsize for times when it's the first N packets,
1150 * where N == ipsa->ipsa_replay_wsize.
1151 *
1152 * Another check that may come here later is the "collision" check.
1153 * If legitimate packets flow quickly enough, this won't be a problem,
1154 * but collisions may cause authentication algorithm crunching to
1155 * take place when it doesn't need to.
1156 */
1157 if (!sadb_replay_peek(ipsa, esph->esph_replay)) {
1158 ESP_BUMP_STAT(espstack, replay_early_failures);
1159 IP_ESP_BUMP_STAT(ipss, in_discards);
1160 ip_drop_packet(data_mp, B_TRUE, ira->ira_ill,
1161 DROPPER(ipss, ipds_esp_early_replay),
1162 &espstack->esp_dropper);
1163 BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
1164 return (NULL);
1165 }
1166
1167 /*
1168 * Adjust the IP header's payload length to reflect the removal
1169 * of the ICV.
1170 */
1171 if (!(ira->ira_flags & IRAF_IS_IPV4)) {
1172 ip6_t *ip6h = (ip6_t *)data_mp->b_rptr;
1173 ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) -
1174 ipsa->ipsa_mac_len);
1175 } else {
1176 ipha_t *ipha = (ipha_t *)data_mp->b_rptr;
1177 ipha->ipha_length = htons(ntohs(ipha->ipha_length) -
1178 ipsa->ipsa_mac_len);
1179 }
1180
1181 /* submit the request to the crypto framework */
1182 return (esp_submit_req_inbound(data_mp, ira, ipsa,
1183 (uint8_t *)esph - data_mp->b_rptr));
1184 }
1185
1186 /*
1187 * Perform the really difficult work of inserting the proposed situation.
1188 * Called while holding the algorithm lock.
1189 */
1190 static void
1191 esp_insert_prop(sadb_prop_t *prop, ipsacq_t *acqrec, uint_t combs,
1192 netstack_t *ns)
1193 {
1194 sadb_comb_t *comb = (sadb_comb_t *)(prop + 1);
1195 ipsec_action_t *ap;
1196 ipsec_prot_t *prot;
1197 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1198 ipsec_stack_t *ipss = ns->netstack_ipsec;
1199
1200 ASSERT(MUTEX_HELD(&ipss->ipsec_alg_lock));
1201
1202 prop->sadb_prop_exttype = SADB_EXT_PROPOSAL;
1203 prop->sadb_prop_len = SADB_8TO64(sizeof (sadb_prop_t));
1204 *(uint32_t *)(&prop->sadb_prop_replay) = 0; /* Quick zero-out! */
1205
1206 prop->sadb_prop_replay = espstack->ipsecesp_replay_size;
1207
1208 /*
1209 * Based upon algorithm properties, and what-not, prioritize a
1210 * proposal, based on the ordering of the ESP algorithms in the
1211 * alternatives in the policy rule or socket that was placed
1212 * in the acquire record.
1213 *
1214 * For each action in policy list
1215 * Add combination. If I've hit limit, return.
1216 */
1217
1218 for (ap = acqrec->ipsacq_act; ap != NULL;
1219 ap = ap->ipa_next) {
1220 ipsec_alginfo_t *ealg = NULL;
1221 ipsec_alginfo_t *aalg = NULL;
1222
1223 if (ap->ipa_act.ipa_type != IPSEC_POLICY_APPLY)
1224 continue;
1225
1226 prot = &ap->ipa_act.ipa_apply;
1227
1228 if (!(prot->ipp_use_esp))
1229 continue;
1230
1231 if (prot->ipp_esp_auth_alg != 0) {
1232 aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
1233 [prot->ipp_esp_auth_alg];
1234 if (aalg == NULL || !ALG_VALID(aalg))
1235 continue;
1236 }
1237
1238 ASSERT(prot->ipp_encr_alg > 0);
1239 ealg = ipss->ipsec_alglists[IPSEC_ALG_ENCR]
1240 [prot->ipp_encr_alg];
1241 if (ealg == NULL || !ALG_VALID(ealg))
1242 continue;
1243
1244 comb->sadb_comb_flags = 0;
1245 comb->sadb_comb_reserved = 0;
1246 comb->sadb_comb_encrypt = ealg->alg_id;
1247 comb->sadb_comb_encrypt_minbits =
1248 MAX(prot->ipp_espe_minbits, ealg->alg_ef_minbits);
1249 comb->sadb_comb_encrypt_maxbits =
1250 MIN(prot->ipp_espe_maxbits, ealg->alg_ef_maxbits);
1251
1252 if (aalg == NULL) {
1253 comb->sadb_comb_auth = 0;
1254 comb->sadb_comb_auth_minbits = 0;
1255 comb->sadb_comb_auth_maxbits = 0;
1256 } else {
1257 comb->sadb_comb_auth = aalg->alg_id;
1258 comb->sadb_comb_auth_minbits =
1259 MAX(prot->ipp_espa_minbits, aalg->alg_ef_minbits);
1260 comb->sadb_comb_auth_maxbits =
1261 MIN(prot->ipp_espa_maxbits, aalg->alg_ef_maxbits);
1262 }
1263
1264 /*
1265 * The following may be based on algorithm
1266 * properties, but in the meantime, we just pick
1267 * some good, sensible numbers. Key mgmt. can
1268 * (and perhaps should) be the place to finalize
1269 * such decisions.
1270 */
1271
1272 /*
1273 * No limits on allocations, since we really don't
1274 * support that concept currently.
1275 */
1276 comb->sadb_comb_soft_allocations = 0;
1277 comb->sadb_comb_hard_allocations = 0;
1278
1279 /*
1280 * These may want to come from policy rule..
1281 */
1282 comb->sadb_comb_soft_bytes =
1283 espstack->ipsecesp_default_soft_bytes;
1284 comb->sadb_comb_hard_bytes =
1285 espstack->ipsecesp_default_hard_bytes;
1286 comb->sadb_comb_soft_addtime =
1287 espstack->ipsecesp_default_soft_addtime;
1288 comb->sadb_comb_hard_addtime =
1289 espstack->ipsecesp_default_hard_addtime;
1290 comb->sadb_comb_soft_usetime =
1291 espstack->ipsecesp_default_soft_usetime;
1292 comb->sadb_comb_hard_usetime =
1293 espstack->ipsecesp_default_hard_usetime;
1294
1295 prop->sadb_prop_len += SADB_8TO64(sizeof (*comb));
1296 if (--combs == 0)
1297 break; /* out of space.. */
1298 comb++;
1299 }
1300 }
1301
1302 /*
1303 * Prepare and actually send the SADB_ACQUIRE message to PF_KEY.
1304 */
1305 static void
1306 esp_send_acquire(ipsacq_t *acqrec, mblk_t *extended, netstack_t *ns)
1307 {
1308 uint_t combs;
1309 sadb_msg_t *samsg;
1310 sadb_prop_t *prop;
1311 mblk_t *pfkeymp, *msgmp;
1312 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1313 ipsec_stack_t *ipss = ns->netstack_ipsec;
1314
1315 ESP_BUMP_STAT(espstack, acquire_requests);
1316
1317 if (espstack->esp_pfkey_q == NULL) {
1318 mutex_exit(&acqrec->ipsacq_lock);
1319 return;
1320 }
1321
1322 /* Set up ACQUIRE. */
1323 pfkeymp = sadb_setup_acquire(acqrec, SADB_SATYPE_ESP,
1324 ns->netstack_ipsec);
1325 if (pfkeymp == NULL) {
1326 esp0dbg(("sadb_setup_acquire failed.\n"));
1327 mutex_exit(&acqrec->ipsacq_lock);
1328 return;
1329 }
1330 ASSERT(MUTEX_HELD(&ipss->ipsec_alg_lock));
1331 combs = ipss->ipsec_nalgs[IPSEC_ALG_AUTH] *
1332 ipss->ipsec_nalgs[IPSEC_ALG_ENCR];
1333 msgmp = pfkeymp->b_cont;
1334 samsg = (sadb_msg_t *)(msgmp->b_rptr);
1335
1336 /* Insert proposal here. */
1337
1338 prop = (sadb_prop_t *)(((uint64_t *)samsg) + samsg->sadb_msg_len);
1339 esp_insert_prop(prop, acqrec, combs, ns);
1340 samsg->sadb_msg_len += prop->sadb_prop_len;
1341 msgmp->b_wptr += SADB_64TO8(samsg->sadb_msg_len);
1342
1343 mutex_exit(&ipss->ipsec_alg_lock);
1344
1345 /*
1346 * Must mutex_exit() before sending PF_KEY message up, in
1347 * order to avoid recursive mutex_enter() if there are no registered
1348 * listeners.
1349 *
1350 * Once I've sent the message, I'm cool anyway.
1351 */
1352 mutex_exit(&acqrec->ipsacq_lock);
1353 if (extended != NULL) {
1354 putnext(espstack->esp_pfkey_q, extended);
1355 }
1356 putnext(espstack->esp_pfkey_q, pfkeymp);
1357 }
1358
1359 /* XXX refactor me */
1360 /*
1361 * Handle the SADB_GETSPI message. Create a larval SA.
1362 */
1363 static void
1364 esp_getspi(mblk_t *mp, keysock_in_t *ksi, ipsecesp_stack_t *espstack)
1365 {
1366 ipsa_t *newbie, *target;
1367 isaf_t *outbound, *inbound;
1368 int rc, diagnostic;
1369 sadb_sa_t *assoc;
1370 keysock_out_t *kso;
1371 uint32_t newspi;
1372
1373 /*
1374 * Randomly generate a proposed SPI value
1375 */
1376 (void) random_get_pseudo_bytes((uint8_t *)&newspi, sizeof (uint32_t));
1377 newbie = sadb_getspi(ksi, newspi, &diagnostic,
1378 espstack->ipsecesp_netstack);
1379
1380 if (newbie == NULL) {
1381 sadb_pfkey_error(espstack->esp_pfkey_q, mp, ENOMEM, diagnostic,
1382 ksi->ks_in_serial);
1383 return;
1384 } else if (newbie == (ipsa_t *)-1) {
1385 sadb_pfkey_error(espstack->esp_pfkey_q, mp, EINVAL, diagnostic,
1386 ksi->ks_in_serial);
1387 return;
1388 }
1389
1390 /*
1391 * XXX - We may randomly collide. We really should recover from this.
1392 * Unfortunately, that could require spending way-too-much-time
1393 * in here. For now, let the user retry.
1394 */
1395
1396 if (newbie->ipsa_addrfam == AF_INET6) {
1397 outbound = OUTBOUND_BUCKET_V6(&espstack->esp_sadb.s_v6,
1398 *(uint32_t *)(newbie->ipsa_dstaddr));
1399 inbound = INBOUND_BUCKET(&espstack->esp_sadb.s_v6,
1400 newbie->ipsa_spi);
1401 } else {
1402 ASSERT(newbie->ipsa_addrfam == AF_INET);
1403 outbound = OUTBOUND_BUCKET_V4(&espstack->esp_sadb.s_v4,
1404 *(uint32_t *)(newbie->ipsa_dstaddr));
1405 inbound = INBOUND_BUCKET(&espstack->esp_sadb.s_v4,
1406 newbie->ipsa_spi);
1407 }
1408
1409 mutex_enter(&outbound->isaf_lock);
1410 mutex_enter(&inbound->isaf_lock);
1411
1412 /*
1413 * Check for collisions (i.e. did sadb_getspi() return with something
1414 * that already exists?).
1415 *
1416 * Try outbound first. Even though SADB_GETSPI is traditionally
1417 * for inbound SAs, you never know what a user might do.
1418 */
1419 target = ipsec_getassocbyspi(outbound, newbie->ipsa_spi,
1420 newbie->ipsa_srcaddr, newbie->ipsa_dstaddr, newbie->ipsa_addrfam);
1421 if (target == NULL) {
1422 target = ipsec_getassocbyspi(inbound, newbie->ipsa_spi,
1423 newbie->ipsa_srcaddr, newbie->ipsa_dstaddr,
1424 newbie->ipsa_addrfam);
1425 }
1426
1427 /*
1428 * I don't have collisions elsewhere!
1429 * (Nor will I because I'm still holding inbound/outbound locks.)
1430 */
1431
1432 if (target != NULL) {
1433 rc = EEXIST;
1434 IPSA_REFRELE(target);
1435 } else {
1436 /*
1437 * sadb_insertassoc() also checks for collisions, so
1438 * if there's a colliding entry, rc will be set
1439 * to EEXIST.
1440 */
1441 rc = sadb_insertassoc(newbie, inbound);
1442 newbie->ipsa_hardexpiretime = gethrestime_sec();
1443 newbie->ipsa_hardexpiretime +=
1444 espstack->ipsecesp_larval_timeout;
1445 }
1446
1447 /*
1448 * Can exit outbound mutex. Hold inbound until we're done
1449 * with newbie.
1450 */
1451 mutex_exit(&outbound->isaf_lock);
1452
1453 if (rc != 0) {
1454 mutex_exit(&inbound->isaf_lock);
1455 IPSA_REFRELE(newbie);
1456 sadb_pfkey_error(espstack->esp_pfkey_q, mp, rc,
1457 SADB_X_DIAGNOSTIC_NONE, ksi->ks_in_serial);
1458 return;
1459 }
1460
1461
1462 /* Can write here because I'm still holding the bucket lock. */
1463 newbie->ipsa_type = SADB_SATYPE_ESP;
1464
1465 /*
1466 * Construct successful return message. We have one thing going
1467 * for us in PF_KEY v2. That's the fact that
1468 * sizeof (sadb_spirange_t) == sizeof (sadb_sa_t)
1469 */
1470 assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SPIRANGE];
1471 assoc->sadb_sa_exttype = SADB_EXT_SA;
1472 assoc->sadb_sa_spi = newbie->ipsa_spi;
1473 *((uint64_t *)(&assoc->sadb_sa_replay)) = 0;
1474 mutex_exit(&inbound->isaf_lock);
1475
1476 /* Convert KEYSOCK_IN to KEYSOCK_OUT. */
1477 kso = (keysock_out_t *)ksi;
1478 kso->ks_out_len = sizeof (*kso);
1479 kso->ks_out_serial = ksi->ks_in_serial;
1480 kso->ks_out_type = KEYSOCK_OUT;
1481
1482 /*
1483 * Can safely putnext() to esp_pfkey_q, because this is a turnaround
1484 * from the esp_pfkey_q.
1485 */
1486 putnext(espstack->esp_pfkey_q, mp);
1487 }
1488
1489 /*
1490 * Insert the ESP header into a packet. Duplicate an mblk, and insert a newly
1491 * allocated mblk with the ESP header in between the two.
1492 */
1493 static boolean_t
1494 esp_insert_esp(mblk_t *mp, mblk_t *esp_mp, uint_t divpoint,
1495 ipsecesp_stack_t *espstack)
1496 {
1497 mblk_t *split_mp = mp;
1498 uint_t wheretodiv = divpoint;
1499
1500 while ((split_mp->b_wptr - split_mp->b_rptr) < wheretodiv) {
1501 wheretodiv -= (split_mp->b_wptr - split_mp->b_rptr);
1502 split_mp = split_mp->b_cont;
1503 ASSERT(split_mp != NULL);
1504 }
1505
1506 if (split_mp->b_wptr - split_mp->b_rptr != wheretodiv) {
1507 mblk_t *scratch;
1508
1509 /* "scratch" is the 2nd half, split_mp is the first. */
1510 scratch = dupb(split_mp);
1511 if (scratch == NULL) {
1512 esp1dbg(espstack,
1513 ("esp_insert_esp: can't allocate scratch.\n"));
1514 return (B_FALSE);
1515 }
1516 /* NOTE: dupb() doesn't set b_cont appropriately. */
1517 scratch->b_cont = split_mp->b_cont;
1518 scratch->b_rptr += wheretodiv;
1519 split_mp->b_wptr = split_mp->b_rptr + wheretodiv;
1520 split_mp->b_cont = scratch;
1521 }
1522 /*
1523 * At this point, split_mp is exactly "wheretodiv" bytes long, and
1524 * holds the end of the pre-ESP part of the datagram.
1525 */
1526 esp_mp->b_cont = split_mp->b_cont;
1527 split_mp->b_cont = esp_mp;
1528
1529 return (B_TRUE);
1530 }
1531
1532 /*
1533 * Section 7 of RFC 3947 says:
1534 *
1535 * 7. Recovering from the Expiring NAT Mappings
1536 *
1537 * There are cases where NAT box decides to remove mappings that are still
1538 * alive (for example, when the keepalive interval is too long, or when the
1539 * NAT box is rebooted). To recover from this, ends that are NOT behind
1540 * NAT SHOULD use the last valid UDP encapsulated IKE or IPsec packet from
1541 * the other end to determine which IP and port addresses should be used.
1542 * The host behind dynamic NAT MUST NOT do this, as otherwise it opens a
1543 * DoS attack possibility because the IP address or port of the other host
1544 * will not change (it is not behind NAT).
1545 *
1546 * Keepalives cannot be used for these purposes, as they are not
1547 * authenticated, but any IKE authenticated IKE packet or ESP packet can be
1548 * used to detect whether the IP address or the port has changed.
1549 *
1550 * The following function will check an SA and its explicitly-set pair to see
1551 * if the NAT-T remote port matches the received packet (which must have
1552 * passed ESP authentication, see esp_in_done() for the caller context). If
1553 * there is a mismatch, the SAs are updated. It is not important if we race
1554 * with a transmitting thread, as if there is a transmitting thread, it will
1555 * merely emit a packet that will most-likely be dropped.
1556 *
1557 * "ports" are ordered src,dst, and assoc is an inbound SA, where src should
1558 * match ipsa_remote_nat_port and dst should match ipsa_local_nat_port.
1559 */
1560 #ifdef _LITTLE_ENDIAN
1561 #define FIRST_16(x) ((x) & 0xFFFF)
1562 #define NEXT_16(x) (((x) >> 16) & 0xFFFF)
1563 #else
1564 #define FIRST_16(x) (((x) >> 16) & 0xFFFF)
1565 #define NEXT_16(x) ((x) & 0xFFFF)
1566 #endif
1567 static void
1568 esp_port_freshness(uint32_t ports, ipsa_t *assoc)
1569 {
1570 uint16_t remote = FIRST_16(ports);
1571 ipsa_t *outbound_peer;
1572 isaf_t *bucket;
1573 ipsecesp_stack_t *espstack = assoc->ipsa_netstack->netstack_ipsecesp;
1574
1575 /* We found a conn_t, therefore local != 0. */
1576 ASSERT(NEXT_16(ports) != 0);
1577 /* Assume an IPv4 SA. */
1578 ASSERT(assoc->ipsa_addrfam == AF_INET);
1579
1580 /*
1581 * On-the-wire rport == 0 means something's very wrong.
1582 * An unpaired SA is also useless to us.
1583 * If we are behind the NAT, don't bother.
1584 * A zero local NAT port defaults to 4500, so check that too.
1585 * And, of course, if the ports already match, we don't need to
1586 * bother.
1587 */
1588 if (remote == 0 || assoc->ipsa_otherspi == 0 ||
1589 (assoc->ipsa_flags & IPSA_F_BEHIND_NAT) ||
1590 (assoc->ipsa_remote_nat_port == 0 &&
1591 remote == htons(IPPORT_IKE_NATT)) ||
1592 remote == assoc->ipsa_remote_nat_port)
1593 return;
1594
1595 /* Try and snag the peer. NOTE: Assume IPv4 for now. */
1596 bucket = OUTBOUND_BUCKET_V4(&(espstack->esp_sadb.s_v4),
1597 assoc->ipsa_srcaddr[0]);
1598 mutex_enter(&bucket->isaf_lock);
1599 outbound_peer = ipsec_getassocbyspi(bucket, assoc->ipsa_otherspi,
1600 assoc->ipsa_dstaddr, assoc->ipsa_srcaddr, AF_INET);
1601 mutex_exit(&bucket->isaf_lock);
1602
1603 /* We probably lost a race to a deleting or expiring thread. */
1604 if (outbound_peer == NULL)
1605 return;
1606
1607 /*
1608 * Hold the mutexes for both SAs so we don't race another inbound
1609 * thread. A lock-entry order shouldn't matter, since all other
1610 * per-ipsa locks are individually held-then-released.
1611 *
1612 * Luckily, this has nothing to do with the remote-NAT address,
1613 * so we don't have to re-scribble the cached-checksum differential.
1614 */
1615 mutex_enter(&outbound_peer->ipsa_lock);
1616 mutex_enter(&assoc->ipsa_lock);
1617 outbound_peer->ipsa_remote_nat_port = assoc->ipsa_remote_nat_port =
1618 remote;
1619 mutex_exit(&assoc->ipsa_lock);
1620 mutex_exit(&outbound_peer->ipsa_lock);
1621 IPSA_REFRELE(outbound_peer);
1622 ESP_BUMP_STAT(espstack, sa_port_renumbers);
1623 }
1624 /*
1625 * Finish processing of an inbound ESP packet after processing by the
1626 * crypto framework.
1627 * - Remove the ESP header.
1628 * - Send packet back to IP.
1629 * If authentication was performed on the packet, this function is called
1630 * only if the authentication succeeded.
1631 * On success returns B_TRUE, on failure returns B_FALSE and frees the
1632 * mblk chain data_mp.
1633 */
1634 static mblk_t *
1635 esp_in_done(mblk_t *data_mp, ip_recv_attr_t *ira, ipsec_crypto_t *ic)
1636 {
1637 ipsa_t *assoc;
1638 uint_t espstart;
1639 uint32_t ivlen = 0;
1640 uint_t processed_len;
1641 esph_t *esph;
1642 kstat_named_t *counter;
1643 boolean_t is_natt;
1644 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
1645 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1646 ipsec_stack_t *ipss = ns->netstack_ipsec;
1647
1648 assoc = ira->ira_ipsec_esp_sa;
1649 ASSERT(assoc != NULL);
1650
1651 is_natt = ((assoc->ipsa_flags & IPSA_F_NATT) != 0);
1652
1653 /* get the pointer to the ESP header */
1654 if (assoc->ipsa_encr_alg == SADB_EALG_NULL) {
1655 /* authentication-only ESP */
1656 espstart = ic->ic_crypto_data.cd_offset;
1657 processed_len = ic->ic_crypto_data.cd_length;
1658 } else {
1659 /* encryption present */
1660 ivlen = assoc->ipsa_iv_len;
1661 if (assoc->ipsa_auth_alg == SADB_AALG_NONE) {
1662 /* encryption-only ESP */
1663 espstart = ic->ic_crypto_data.cd_offset -
1664 sizeof (esph_t) - assoc->ipsa_iv_len;
1665 processed_len = ic->ic_crypto_data.cd_length +
1666 ivlen;
1667 } else {
1668 /* encryption with authentication */
1669 espstart = ic->ic_crypto_dual_data.dd_offset1;
1670 processed_len = ic->ic_crypto_dual_data.dd_len2 +
1671 ivlen;
1672 }
1673 }
1674
1675 esph = (esph_t *)(data_mp->b_rptr + espstart);
1676
1677 if (assoc->ipsa_auth_alg != IPSA_AALG_NONE ||
1678 (assoc->ipsa_flags & IPSA_F_COMBINED)) {
1679 /*
1680 * Authentication passed if we reach this point.
1681 * Packets with authentication will have the ICV
1682 * after the crypto data. Adjust b_wptr before
1683 * making padlen checks.
1684 */
1685 ESP_BUMP_STAT(espstack, good_auth);
1686 data_mp->b_wptr -= assoc->ipsa_mac_len;
1687
1688 /*
1689 * Check replay window here!
1690 * For right now, assume keysock will set the replay window
1691 * size to zero for SAs that have an unspecified sender.
1692 * This may change...
1693 */
1694
1695 if (!sadb_replay_check(assoc, esph->esph_replay)) {
1696 /*
1697 * Log the event. As of now we print out an event.
1698 * Do not print the replay failure number, or else
1699 * syslog cannot collate the error messages. Printing
1700 * the replay number that failed opens a denial-of-
1701 * service attack.
1702 */
1703 ipsec_assocfailure(info.mi_idnum, 0, 0,
1704 SL_ERROR | SL_WARN,
1705 "Replay failed for ESP spi 0x%x, dst %s.\n",
1706 assoc->ipsa_spi, assoc->ipsa_dstaddr,
1707 assoc->ipsa_addrfam, espstack->ipsecesp_netstack);
1708 ESP_BUMP_STAT(espstack, replay_failures);
1709 counter = DROPPER(ipss, ipds_esp_replay);
1710 goto drop_and_bail;
1711 }
1712
1713 if (is_natt) {
1714 ASSERT(ira->ira_flags & IRAF_ESP_UDP_PORTS);
1715 ASSERT(ira->ira_esp_udp_ports != 0);
1716 esp_port_freshness(ira->ira_esp_udp_ports, assoc);
1717 }
1718 }
1719
1720 esp_set_usetime(assoc, B_TRUE);
1721
1722 if (!esp_age_bytes(assoc, processed_len, B_TRUE)) {
1723 /* The ipsa has hit hard expiration, LOG and AUDIT. */
1724 ipsec_assocfailure(info.mi_idnum, 0, 0,
1725 SL_ERROR | SL_WARN,
1726 "ESP association 0x%x, dst %s had bytes expire.\n",
1727 assoc->ipsa_spi, assoc->ipsa_dstaddr, assoc->ipsa_addrfam,
1728 espstack->ipsecesp_netstack);
1729 ESP_BUMP_STAT(espstack, bytes_expired);
1730 counter = DROPPER(ipss, ipds_esp_bytes_expire);
1731 goto drop_and_bail;
1732 }
1733
1734 /*
1735 * Remove ESP header and padding from packet. I hope the compiler
1736 * spews "branch, predict taken" code for this.
1737 */
1738
1739 if (esp_strip_header(data_mp, (ira->ira_flags & IRAF_IS_IPV4),
1740 ivlen, &counter, espstack)) {
1741
1742 if (is_system_labeled() && assoc->ipsa_tsl != NULL) {
1743 if (!ip_recv_attr_replace_label(ira, assoc->ipsa_tsl)) {
1744 ip_drop_packet(data_mp, B_TRUE, ira->ira_ill,
1745 DROPPER(ipss, ipds_ah_nomem),
1746 &espstack->esp_dropper);
1747 BUMP_MIB(ira->ira_ill->ill_ip_mib,
1748 ipIfStatsInDiscards);
1749 return (NULL);
1750 }
1751 }
1752 if (is_natt)
1753 return (esp_fix_natt_checksums(data_mp, assoc));
1754
1755 return (data_mp);
1756 }
1757
1758 esp1dbg(espstack, ("esp_in_done: esp_strip_header() failed\n"));
1759 drop_and_bail:
1760 IP_ESP_BUMP_STAT(ipss, in_discards);
1761 ip_drop_packet(data_mp, B_TRUE, ira->ira_ill, counter,
1762 &espstack->esp_dropper);
1763 BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
1764 return (NULL);
1765 }
1766
1767 /*
1768 * Called upon failing the inbound ICV check. The message passed as
1769 * argument is freed.
1770 */
1771 static void
1772 esp_log_bad_auth(mblk_t *mp, ip_recv_attr_t *ira)
1773 {
1774 ipsa_t *assoc = ira->ira_ipsec_esp_sa;
1775 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
1776 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1777 ipsec_stack_t *ipss = ns->netstack_ipsec;
1778
1779 /*
1780 * Log the event. Don't print to the console, block
1781 * potential denial-of-service attack.
1782 */
1783 ESP_BUMP_STAT(espstack, bad_auth);
1784
1785 ipsec_assocfailure(info.mi_idnum, 0, 0, SL_ERROR | SL_WARN,
1786 "ESP Authentication failed for spi 0x%x, dst %s.\n",
1787 assoc->ipsa_spi, assoc->ipsa_dstaddr, assoc->ipsa_addrfam,
1788 espstack->ipsecesp_netstack);
1789
1790 IP_ESP_BUMP_STAT(ipss, in_discards);
1791 ip_drop_packet(mp, B_TRUE, ira->ira_ill,
1792 DROPPER(ipss, ipds_esp_bad_auth),
1793 &espstack->esp_dropper);
1794 }
1795
1796
1797 /*
1798 * Invoked for outbound packets after ESP processing. If the packet
1799 * also requires AH, performs the AH SA selection and AH processing.
1800 * Returns B_TRUE if the AH processing was not needed or if it was
1801 * performed successfully. Returns B_FALSE and consumes the passed mblk
1802 * if AH processing was required but could not be performed.
1803 *
1804 * Returns data_mp unless data_mp was consumed/queued.
1805 */
1806 static mblk_t *
1807 esp_do_outbound_ah(mblk_t *data_mp, ip_xmit_attr_t *ixa)
1808 {
1809 ipsec_action_t *ap;
1810
1811 ap = ixa->ixa_ipsec_action;
1812 if (ap == NULL) {
1813 ipsec_policy_t *pp = ixa->ixa_ipsec_policy;
1814 ap = pp->ipsp_act;
1815 }
1816
1817 if (!ap->ipa_want_ah)
1818 return (data_mp);
1819
1820 /*
1821 * Normally the AH SA would have already been put in place
1822 * but it could have been flushed so we need to look for it.
1823 */
1824 if (ixa->ixa_ipsec_ah_sa == NULL) {
1825 if (!ipsec_outbound_sa(data_mp, ixa, IPPROTO_AH)) {
1826 sadb_acquire(data_mp, ixa, B_TRUE, B_FALSE);
1827 return (NULL);
1828 }
1829 }
1830 ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
1831
1832 data_mp = ixa->ixa_ipsec_ah_sa->ipsa_output_func(data_mp, ixa);
1833 return (data_mp);
1834 }
1835
1836
1837 /*
1838 * Kernel crypto framework callback invoked after completion of async
1839 * crypto requests for outbound packets.
1840 */
1841 static void
1842 esp_kcf_callback_outbound(void *arg, int status)
1843 {
1844 mblk_t *mp = (mblk_t *)arg;
1845 mblk_t *async_mp;
1846 netstack_t *ns;
1847 ipsec_stack_t *ipss;
1848 ipsecesp_stack_t *espstack;
1849 mblk_t *data_mp;
1850 ip_xmit_attr_t ixas;
1851 ipsec_crypto_t *ic;
1852 ill_t *ill;
1853
1854 /*
1855 * First remove the ipsec_crypto_t mblk
1856 * Note that we need to ipsec_free_crypto_data(mp) once done with ic.
1857 */
1858 async_mp = ipsec_remove_crypto_data(mp, &ic);
1859 ASSERT(async_mp != NULL);
1860
1861 /*
1862 * Extract the ip_xmit_attr_t from the first mblk.
1863 * Verifies that the netstack and ill is still around; could
1864 * have vanished while kEf was doing its work.
1865 * On succesful return we have a nce_t and the ill/ipst can't
1866 * disappear until we do the nce_refrele in ixa_cleanup.
1867 */
1868 data_mp = async_mp->b_cont;
1869 async_mp->b_cont = NULL;
1870 if (!ip_xmit_attr_from_mblk(async_mp, &ixas)) {
1871 /* Disappeared on us - no ill/ipst for MIB */
1872 /* We have nowhere to do stats since ixa_ipst could be NULL */
1873 if (ixas.ixa_nce != NULL) {
1874 ill = ixas.ixa_nce->nce_ill;
1875 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1876 ip_drop_output("ipIfStatsOutDiscards", data_mp, ill);
1877 }
1878 freemsg(data_mp);
1879 goto done;
1880 }
1881 ns = ixas.ixa_ipst->ips_netstack;
1882 espstack = ns->netstack_ipsecesp;
1883 ipss = ns->netstack_ipsec;
1884 ill = ixas.ixa_nce->nce_ill;
1885
1886 if (status == CRYPTO_SUCCESS) {
1887 /*
1888 * If a ICV was computed, it was stored by the
1889 * crypto framework at the end of the packet.
1890 */
1891 ipha_t *ipha = (ipha_t *)data_mp->b_rptr;
1892
1893 esp_set_usetime(ixas.ixa_ipsec_esp_sa, B_FALSE);
1894 /* NAT-T packet. */
1895 if (IPH_HDR_VERSION(ipha) == IP_VERSION &&
1896 ipha->ipha_protocol == IPPROTO_UDP)
1897 esp_prepare_udp(ns, data_mp, ipha);
1898
1899 /* do AH processing if needed */
1900 data_mp = esp_do_outbound_ah(data_mp, &ixas);
1901 if (data_mp == NULL)
1902 goto done;
1903
1904 (void) ip_output_post_ipsec(data_mp, &ixas);
1905 } else {
1906 /* Outbound shouldn't see invalid MAC */
1907 ASSERT(status != CRYPTO_INVALID_MAC);
1908
1909 esp1dbg(espstack,
1910 ("esp_kcf_callback_outbound: crypto failed with 0x%x\n",
1911 status));
1912 ESP_BUMP_STAT(espstack, crypto_failures);
1913 ESP_BUMP_STAT(espstack, out_discards);
1914 ip_drop_packet(data_mp, B_FALSE, ill,
1915 DROPPER(ipss, ipds_esp_crypto_failed),
1916 &espstack->esp_dropper);
1917 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1918 }
1919 done:
1920 ixa_cleanup(&ixas);
1921 (void) ipsec_free_crypto_data(mp);
1922 }
1923
1924 /*
1925 * Kernel crypto framework callback invoked after completion of async
1926 * crypto requests for inbound packets.
1927 */
1928 static void
1929 esp_kcf_callback_inbound(void *arg, int status)
1930 {
1931 mblk_t *mp = (mblk_t *)arg;
1932 mblk_t *async_mp;
1933 netstack_t *ns;
1934 ipsecesp_stack_t *espstack;
1935 ipsec_stack_t *ipss;
1936 mblk_t *data_mp;
1937 ip_recv_attr_t iras;
1938 ipsec_crypto_t *ic;
1939
1940 /*
1941 * First remove the ipsec_crypto_t mblk
1942 * Note that we need to ipsec_free_crypto_data(mp) once done with ic.
1943 */
1944 async_mp = ipsec_remove_crypto_data(mp, &ic);
1945 ASSERT(async_mp != NULL);
1946
1947 /*
1948 * Extract the ip_recv_attr_t from the first mblk.
1949 * Verifies that the netstack and ill is still around; could
1950 * have vanished while kEf was doing its work.
1951 */
1952 data_mp = async_mp->b_cont;
1953 async_mp->b_cont = NULL;
1954 if (!ip_recv_attr_from_mblk(async_mp, &iras)) {
1955 /* The ill or ip_stack_t disappeared on us */
1956 ip_drop_input("ip_recv_attr_from_mblk", data_mp, NULL);
1957 freemsg(data_mp);
1958 goto done;
1959 }
1960
1961 ns = iras.ira_ill->ill_ipst->ips_netstack;
1962 espstack = ns->netstack_ipsecesp;
1963 ipss = ns->netstack_ipsec;
1964
1965 if (status == CRYPTO_SUCCESS) {
1966 data_mp = esp_in_done(data_mp, &iras, ic);
1967 if (data_mp == NULL)
1968 goto done;
1969
1970 /* finish IPsec processing */
1971 ip_input_post_ipsec(data_mp, &iras);
1972 } else if (status == CRYPTO_INVALID_MAC) {
1973 esp_log_bad_auth(data_mp, &iras);
1974 } else {
1975 esp1dbg(espstack,
1976 ("esp_kcf_callback: crypto failed with 0x%x\n",
1977 status));
1978 ESP_BUMP_STAT(espstack, crypto_failures);
1979 IP_ESP_BUMP_STAT(ipss, in_discards);
1980 ip_drop_packet(data_mp, B_TRUE, iras.ira_ill,
1981 DROPPER(ipss, ipds_esp_crypto_failed),
1982 &espstack->esp_dropper);
1983 BUMP_MIB(iras.ira_ill->ill_ip_mib, ipIfStatsInDiscards);
1984 }
1985 done:
1986 ira_cleanup(&iras, B_TRUE);
1987 (void) ipsec_free_crypto_data(mp);
1988 }
1989
1990 /*
1991 * Invoked on crypto framework failure during inbound and outbound processing.
1992 */
1993 static void
1994 esp_crypto_failed(mblk_t *data_mp, boolean_t is_inbound, int kef_rc,
1995 ill_t *ill, ipsecesp_stack_t *espstack)
1996 {
1997 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec;
1998
1999 esp1dbg(espstack, ("crypto failed for %s ESP with 0x%x\n",
2000 is_inbound ? "inbound" : "outbound", kef_rc));
2001 ip_drop_packet(data_mp, is_inbound, ill,
2002 DROPPER(ipss, ipds_esp_crypto_failed),
2003 &espstack->esp_dropper);
2004 ESP_BUMP_STAT(espstack, crypto_failures);
2005 if (is_inbound)
2006 IP_ESP_BUMP_STAT(ipss, in_discards);
2007 else
2008 ESP_BUMP_STAT(espstack, out_discards);
2009 }
2010
2011 /*
2012 * A statement-equivalent macro, _cr MUST point to a modifiable
2013 * crypto_call_req_t.
2014 */
2015 #define ESP_INIT_CALLREQ(_cr, _mp, _callback) \
2016 (_cr)->cr_flag = CRYPTO_SKIP_REQID|CRYPTO_ALWAYS_QUEUE; \
2017 (_cr)->cr_callback_arg = (_mp); \
2018 (_cr)->cr_callback_func = (_callback)
2019
2020 #define ESP_INIT_CRYPTO_MAC(mac, icvlen, icvbuf) { \
2021 (mac)->cd_format = CRYPTO_DATA_RAW; \
2022 (mac)->cd_offset = 0; \
2023 (mac)->cd_length = icvlen; \
2024 (mac)->cd_raw.iov_base = (char *)icvbuf; \
2025 (mac)->cd_raw.iov_len = icvlen; \
2026 }
2027
2028 #define ESP_INIT_CRYPTO_DATA(data, mp, off, len) { \
2029 if (MBLKL(mp) >= (len) + (off)) { \
2030 (data)->cd_format = CRYPTO_DATA_RAW; \
2031 (data)->cd_raw.iov_base = (char *)(mp)->b_rptr; \
2032 (data)->cd_raw.iov_len = MBLKL(mp); \
2033 (data)->cd_offset = off; \
2034 } else { \
2035 (data)->cd_format = CRYPTO_DATA_MBLK; \
2036 (data)->cd_mp = mp; \
2037 (data)->cd_offset = off; \
2038 } \
2039 (data)->cd_length = len; \
2040 }
2041
2042 #define ESP_INIT_CRYPTO_DUAL_DATA(data, mp, off1, len1, off2, len2) { \
2043 (data)->dd_format = CRYPTO_DATA_MBLK; \
2044 (data)->dd_mp = mp; \
2045 (data)->dd_len1 = len1; \
2046 (data)->dd_offset1 = off1; \
2047 (data)->dd_len2 = len2; \
2048 (data)->dd_offset2 = off2; \
2049 }
2050
2051 /*
2052 * Returns data_mp if successfully completed the request. Returns
2053 * NULL if it failed (and increments InDiscards) or if it is pending.
2054 */
2055 static mblk_t *
2056 esp_submit_req_inbound(mblk_t *esp_mp, ip_recv_attr_t *ira,
2057 ipsa_t *assoc, uint_t esph_offset)
2058 {
2059 uint_t auth_offset, msg_len, auth_len;
2060 crypto_call_req_t call_req, *callrp;
2061 mblk_t *mp;
2062 esph_t *esph_ptr;
2063 int kef_rc;
2064 uint_t icv_len = assoc->ipsa_mac_len;
2065 crypto_ctx_template_t auth_ctx_tmpl;
2066 boolean_t do_auth, do_encr, force;
2067 uint_t encr_offset, encr_len;
2068 uint_t iv_len = assoc->ipsa_iv_len;
2069 crypto_ctx_template_t encr_ctx_tmpl;
2070 ipsec_crypto_t *ic, icstack;
2071 uchar_t *iv_ptr;
2072 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
2073 ipsec_stack_t *ipss = ns->netstack_ipsec;
2074 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
2075
2076 do_auth = assoc->ipsa_auth_alg != SADB_AALG_NONE;
2077 do_encr = assoc->ipsa_encr_alg != SADB_EALG_NULL;
2078 force = (assoc->ipsa_flags & IPSA_F_ASYNC);
2079
2080 #ifdef IPSEC_LATENCY_TEST
2081 kef_rc = CRYPTO_SUCCESS;
2082 #else
2083 kef_rc = CRYPTO_FAILED;
2084 #endif
2085
2086 /*
2087 * An inbound packet is of the form:
2088 * [IP,options,ESP,IV,data,ICV,pad]
2089 */
2090 esph_ptr = (esph_t *)(esp_mp->b_rptr + esph_offset);
2091 iv_ptr = (uchar_t *)(esph_ptr + 1);
2092 /* Packet length starting at IP header ending after ESP ICV. */
2093 msg_len = MBLKL(esp_mp);
2094
2095 encr_offset = esph_offset + sizeof (esph_t) + iv_len;
2096 encr_len = msg_len - encr_offset;
2097
2098 /*
2099 * Counter mode algs need a nonce. This is setup in sadb_common_add().
2100 * If for some reason we are using a SA which does not have a nonce
2101 * then we must fail here.
2102 */
2103 if ((assoc->ipsa_flags & IPSA_F_COUNTERMODE) &&
2104 (assoc->ipsa_nonce == NULL)) {
2105 ip_drop_packet(esp_mp, B_TRUE, ira->ira_ill,
2106 DROPPER(ipss, ipds_esp_nomem), &espstack->esp_dropper);
2107 return (NULL);
2108 }
2109
2110 if (force) {
2111 /* We are doing asynch; allocate mblks to hold state */
2112 if ((mp = ip_recv_attr_to_mblk(ira)) == NULL ||
2113 (mp = ipsec_add_crypto_data(mp, &ic)) == NULL) {
2114 BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
2115 ip_drop_input("ipIfStatsInDiscards", esp_mp,
2116 ira->ira_ill);
2117 return (NULL);
2118 }
2119 linkb(mp, esp_mp);
2120 callrp = &call_req;
2121 ESP_INIT_CALLREQ(callrp, mp, esp_kcf_callback_inbound);
2122 } else {
2123 /*
2124 * If we know we are going to do sync then ipsec_crypto_t
2125 * should be on the stack.
2126 */
2127 ic = &icstack;
2128 bzero(ic, sizeof (*ic));
2129 callrp = NULL;
2130 }
2131
2132 if (do_auth) {
2133 /* authentication context template */
2134 IPSEC_CTX_TMPL(assoc, ipsa_authtmpl, IPSEC_ALG_AUTH,
2135 auth_ctx_tmpl);
2136
2137 /* ICV to be verified */
2138 ESP_INIT_CRYPTO_MAC(&ic->ic_crypto_mac,
2139 icv_len, esp_mp->b_wptr - icv_len);
2140
2141 /* authentication starts at the ESP header */
2142 auth_offset = esph_offset;
2143 auth_len = msg_len - auth_offset - icv_len;
2144 if (!do_encr) {
2145 /* authentication only */
2146 /* initialize input data argument */
2147 ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
2148 esp_mp, auth_offset, auth_len);
2149
2150 /* call the crypto framework */
2151 kef_rc = crypto_mac_verify(&assoc->ipsa_amech,
2152 &ic->ic_crypto_data,
2153 &assoc->ipsa_kcfauthkey, auth_ctx_tmpl,
2154 &ic->ic_crypto_mac, callrp);
2155 }
2156 }
2157
2158 if (do_encr) {
2159 /* encryption template */
2160 IPSEC_CTX_TMPL(assoc, ipsa_encrtmpl, IPSEC_ALG_ENCR,
2161 encr_ctx_tmpl);
2162
2163 /* Call the nonce update function. Also passes in IV */
2164 (assoc->ipsa_noncefunc)(assoc, (uchar_t *)esph_ptr, encr_len,
2165 iv_ptr, &ic->ic_cmm, &ic->ic_crypto_data);
2166
2167 if (!do_auth) {
2168 /* decryption only */
2169 /* initialize input data argument */
2170 ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
2171 esp_mp, encr_offset, encr_len);
2172
2173 /* call the crypto framework */
2174 kef_rc = crypto_decrypt((crypto_mechanism_t *)
2175 &ic->ic_cmm, &ic->ic_crypto_data,
2176 &assoc->ipsa_kcfencrkey, encr_ctx_tmpl,
2177 NULL, callrp);
2178 }
2179 }
2180
2181 if (do_auth && do_encr) {
2182 /* dual operation */
2183 /* initialize input data argument */
2184 ESP_INIT_CRYPTO_DUAL_DATA(&ic->ic_crypto_dual_data,
2185 esp_mp, auth_offset, auth_len,
2186 encr_offset, encr_len - icv_len);
2187
2188 /* specify IV */
2189 ic->ic_crypto_dual_data.dd_miscdata = (char *)iv_ptr;
2190
2191 /* call the framework */
2192 kef_rc = crypto_mac_verify_decrypt(&assoc->ipsa_amech,
2193 &assoc->ipsa_emech, &ic->ic_crypto_dual_data,
2194 &assoc->ipsa_kcfauthkey, &assoc->ipsa_kcfencrkey,
2195 auth_ctx_tmpl, encr_ctx_tmpl, &ic->ic_crypto_mac,
2196 NULL, callrp);
2197 }
2198
2199 switch (kef_rc) {
2200 case CRYPTO_SUCCESS:
2201 ESP_BUMP_STAT(espstack, crypto_sync);
2202 esp_mp = esp_in_done(esp_mp, ira, ic);
2203 if (force) {
2204 /* Free mp after we are done with ic */
2205 mp = ipsec_free_crypto_data(mp);
2206 (void) ip_recv_attr_free_mblk(mp);
2207 }
2208 return (esp_mp);
2209 case CRYPTO_QUEUED:
2210 /* esp_kcf_callback_inbound() will be invoked on completion */
2211 ESP_BUMP_STAT(espstack, crypto_async);
2212 return (NULL);
2213 case CRYPTO_INVALID_MAC:
2214 if (force) {
2215 mp = ipsec_free_crypto_data(mp);
2216 esp_mp = ip_recv_attr_free_mblk(mp);
2217 }
2218 ESP_BUMP_STAT(espstack, crypto_sync);
2219 BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
2220 esp_log_bad_auth(esp_mp, ira);
2221 /* esp_mp was passed to ip_drop_packet */
2222 return (NULL);
2223 }
2224
2225 if (force) {
2226 mp = ipsec_free_crypto_data(mp);
2227 esp_mp = ip_recv_attr_free_mblk(mp);
2228 }
2229 BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
2230 esp_crypto_failed(esp_mp, B_TRUE, kef_rc, ira->ira_ill, espstack);
2231 /* esp_mp was passed to ip_drop_packet */
2232 return (NULL);
2233 }
2234
2235 /*
2236 * Compute the IP and UDP checksums -- common code for both keepalives and
2237 * actual ESP-in-UDP packets. Be flexible with multiple mblks because ESP
2238 * uses mblk-insertion to insert the UDP header.
2239 * TODO - If there is an easy way to prep a packet for HW checksums, make
2240 * it happen here.
2241 * Note that this is used before both before calling ip_output_simple and
2242 * in the esp datapath. The former could use IXAF_SET_ULP_CKSUM but not the
2243 * latter.
2244 */
2245 static void
2246 esp_prepare_udp(netstack_t *ns, mblk_t *mp, ipha_t *ipha)
2247 {
2248 int offset;
2249 uint32_t cksum;
2250 uint16_t *arr;
2251 mblk_t *udpmp = mp;
2252 uint_t hlen = IPH_HDR_LENGTH(ipha);
2253
2254 ASSERT(MBLKL(mp) >= sizeof (ipha_t));
2255
2256 ipha->ipha_hdr_checksum = 0;
2257 ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
2258
2259 if (ns->netstack_udp->us_do_checksum) {
2260 ASSERT(MBLKL(udpmp) >= sizeof (udpha_t));
2261 /* arr points to the IP header. */
2262 arr = (uint16_t *)ipha;
2263 IP_STAT(ns->netstack_ip, ip_out_sw_cksum);
2264 IP_STAT_UPDATE(ns->netstack_ip, ip_out_sw_cksum_bytes,
2265 ntohs(htons(ipha->ipha_length) - hlen));
2266 /* arr[6-9] are the IP addresses. */
2267 cksum = IP_UDP_CSUM_COMP + arr[6] + arr[7] + arr[8] + arr[9] +
2268 ntohs(htons(ipha->ipha_length) - hlen);
2269 cksum = IP_CSUM(mp, hlen, cksum);
2270 offset = hlen + UDP_CHECKSUM_OFFSET;
2271 while (offset >= MBLKL(udpmp)) {
2272 offset -= MBLKL(udpmp);
2273 udpmp = udpmp->b_cont;
2274 }
2275 /* arr points to the UDP header's checksum field. */
2276 arr = (uint16_t *)(udpmp->b_rptr + offset);
2277 *arr = cksum;
2278 }
2279 }
2280
2281 /*
2282 * taskq handler so we can send the NAT-T keepalive on a separate thread.
2283 */
2284 static void
2285 actually_send_keepalive(void *arg)
2286 {
2287 mblk_t *mp = (mblk_t *)arg;
2288 ip_xmit_attr_t ixas;
2289 netstack_t *ns;
2290 netstackid_t stackid;
2291
2292 stackid = (netstackid_t)(uintptr_t)mp->b_prev;
2293 mp->b_prev = NULL;
2294 ns = netstack_find_by_stackid(stackid);
2295 if (ns == NULL) {
2296 /* Disappeared */
2297 ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
2298 freemsg(mp);
2299 return;
2300 }
2301
2302 bzero(&ixas, sizeof (ixas));
2303 ixas.ixa_zoneid = ALL_ZONES;
2304 ixas.ixa_cred = kcred;
2305 ixas.ixa_cpid = NOPID;
2306 ixas.ixa_tsl = NULL;
2307 ixas.ixa_ipst = ns->netstack_ip;
2308 /* No ULP checksum; done by esp_prepare_udp */
2309 ixas.ixa_flags = (IXAF_IS_IPV4 | IXAF_NO_IPSEC | IXAF_VERIFY_SOURCE);
2310
2311 (void) ip_output_simple(mp, &ixas);
2312 ixa_cleanup(&ixas);
2313 netstack_rele(ns);
2314 }
2315
2316 /*
2317 * Send a one-byte UDP NAT-T keepalive.
2318 */
2319 void
2320 ipsecesp_send_keepalive(ipsa_t *assoc)
2321 {
2322 mblk_t *mp;
2323 ipha_t *ipha;
2324 udpha_t *udpha;
2325 netstack_t *ns = assoc->ipsa_netstack;
2326
2327 ASSERT(MUTEX_NOT_HELD(&assoc->ipsa_lock));
2328
2329 mp = allocb(sizeof (ipha_t) + sizeof (udpha_t) + 1, BPRI_HI);
2330 if (mp == NULL)
2331 return;
2332 ipha = (ipha_t *)mp->b_rptr;
2333 ipha->ipha_version_and_hdr_length = IP_SIMPLE_HDR_VERSION;
2334 ipha->ipha_type_of_service = 0;
2335 ipha->ipha_length = htons(sizeof (ipha_t) + sizeof (udpha_t) + 1);
2336 /* Use the low-16 of the SPI so we have some clue where it came from. */
2337 ipha->ipha_ident = *(((uint16_t *)(&assoc->ipsa_spi)) + 1);
2338 ipha->ipha_fragment_offset_and_flags = 0; /* Too small to fragment! */
2339 ipha->ipha_ttl = 0xFF;
2340 ipha->ipha_protocol = IPPROTO_UDP;
2341 ipha->ipha_hdr_checksum = 0;
2342 ipha->ipha_src = assoc->ipsa_srcaddr[0];
2343 ipha->ipha_dst = assoc->ipsa_dstaddr[0];
2344 udpha = (udpha_t *)(ipha + 1);
2345 udpha->uha_src_port = (assoc->ipsa_local_nat_port != 0) ?
2346 assoc->ipsa_local_nat_port : htons(IPPORT_IKE_NATT);
2347 udpha->uha_dst_port = (assoc->ipsa_remote_nat_port != 0) ?
2348 assoc->ipsa_remote_nat_port : htons(IPPORT_IKE_NATT);
2349 udpha->uha_length = htons(sizeof (udpha_t) + 1);
2350 udpha->uha_checksum = 0;
2351 mp->b_wptr = (uint8_t *)(udpha + 1);
2352 *(mp->b_wptr++) = 0xFF;
2353
2354 esp_prepare_udp(ns, mp, ipha);
2355
2356 /*
2357 * We're holding an isaf_t bucket lock, so pawn off the actual
2358 * packet transmission to another thread. Just in case syncq
2359 * processing causes a same-bucket packet to be processed.
2360 */
2361 mp->b_prev = (mblk_t *)(uintptr_t)ns->netstack_stackid;
2362
2363 if (taskq_dispatch(esp_taskq, actually_send_keepalive, mp,
2364 TQ_NOSLEEP) == 0) {
2365 /* Assume no memory if taskq_dispatch() fails. */
2366 mp->b_prev = NULL;
2367 ip_drop_packet(mp, B_FALSE, NULL,
2368 DROPPER(ns->netstack_ipsec, ipds_esp_nomem),
2369 &ns->netstack_ipsecesp->esp_dropper);
2370 }
2371 }
2372
2373 /*
2374 * Returns mp if successfully completed the request. Returns
2375 * NULL if it failed (and increments InDiscards) or if it is pending.
2376 */
2377 static mblk_t *
2378 esp_submit_req_outbound(mblk_t *data_mp, ip_xmit_attr_t *ixa, ipsa_t *assoc,
2379 uchar_t *icv_buf, uint_t payload_len)
2380 {
2381 uint_t auth_len;
2382 crypto_call_req_t call_req, *callrp;
2383 mblk_t *esp_mp;
2384 esph_t *esph_ptr;
2385 mblk_t *mp;
2386 int kef_rc = CRYPTO_FAILED;
2387 uint_t icv_len = assoc->ipsa_mac_len;
2388 crypto_ctx_template_t auth_ctx_tmpl;
2389 boolean_t do_auth, do_encr, force;
2390 uint_t iv_len = assoc->ipsa_iv_len;
2391 crypto_ctx_template_t encr_ctx_tmpl;
2392 boolean_t is_natt = ((assoc->ipsa_flags & IPSA_F_NATT) != 0);
2393 size_t esph_offset = (is_natt ? UDPH_SIZE : 0);
2394 netstack_t *ns = ixa->ixa_ipst->ips_netstack;
2395 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
2396 ipsec_crypto_t *ic, icstack;
2397 uchar_t *iv_ptr;
2398 crypto_data_t *cd_ptr = NULL;
2399 ill_t *ill = ixa->ixa_nce->nce_ill;
2400 ipsec_stack_t *ipss = ns->netstack_ipsec;
2401
2402 esp3dbg(espstack, ("esp_submit_req_outbound:%s",
2403 is_natt ? "natt" : "not natt"));
2404
2405 do_encr = assoc->ipsa_encr_alg != SADB_EALG_NULL;
2406 do_auth = assoc->ipsa_auth_alg != SADB_AALG_NONE;
2407 force = (assoc->ipsa_flags & IPSA_F_ASYNC);
2408
2409 #ifdef IPSEC_LATENCY_TEST
2410 kef_rc = CRYPTO_SUCCESS;
2411 #else
2412 kef_rc = CRYPTO_FAILED;
2413 #endif
2414
2415 /*
2416 * Outbound IPsec packets are of the form:
2417 * [IP,options] -> [ESP,IV] -> [data] -> [pad,ICV]
2418 * unless it's NATT, then it's
2419 * [IP,options] -> [udp][ESP,IV] -> [data] -> [pad,ICV]
2420 * Get a pointer to the mblk containing the ESP header.
2421 */
2422 ASSERT(data_mp->b_cont != NULL);
2423 esp_mp = data_mp->b_cont;
2424 esph_ptr = (esph_t *)(esp_mp->b_rptr + esph_offset);
2425 iv_ptr = (uchar_t *)(esph_ptr + 1);
2426
2427 /*
2428 * Combined mode algs need a nonce. This is setup in sadb_common_add().
2429 * If for some reason we are using a SA which does not have a nonce
2430 * then we must fail here.
2431 */
2432 if ((assoc->ipsa_flags & IPSA_F_COUNTERMODE) &&
2433 (assoc->ipsa_nonce == NULL)) {
2434 ip_drop_packet(data_mp, B_FALSE, NULL,
2435 DROPPER(ipss, ipds_esp_nomem), &espstack->esp_dropper);
2436 return (NULL);
2437 }
2438
2439 if (force) {
2440 /* We are doing asynch; allocate mblks to hold state */
2441 if ((mp = ip_xmit_attr_to_mblk(ixa)) == NULL ||
2442 (mp = ipsec_add_crypto_data(mp, &ic)) == NULL) {
2443 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2444 ip_drop_output("ipIfStatsOutDiscards", data_mp, ill);
2445 freemsg(data_mp);
2446 return (NULL);
2447 }
2448
2449 linkb(mp, data_mp);
2450 callrp = &call_req;
2451 ESP_INIT_CALLREQ(callrp, mp, esp_kcf_callback_outbound);
2452 } else {
2453 /*
2454 * If we know we are going to do sync then ipsec_crypto_t
2455 * should be on the stack.
2456 */
2457 ic = &icstack;
2458 bzero(ic, sizeof (*ic));
2459 callrp = NULL;
2460 }
2461
2462
2463 if (do_auth) {
2464 /* authentication context template */
2465 IPSEC_CTX_TMPL(assoc, ipsa_authtmpl, IPSEC_ALG_AUTH,
2466 auth_ctx_tmpl);
2467
2468 /* where to store the computed mac */
2469 ESP_INIT_CRYPTO_MAC(&ic->ic_crypto_mac,
2470 icv_len, icv_buf);
2471
2472 /* authentication starts at the ESP header */
2473 auth_len = payload_len + iv_len + sizeof (esph_t);
2474 if (!do_encr) {
2475 /* authentication only */
2476 /* initialize input data argument */
2477 ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
2478 esp_mp, esph_offset, auth_len);
2479
2480 /* call the crypto framework */
2481 kef_rc = crypto_mac(&assoc->ipsa_amech,
2482 &ic->ic_crypto_data,
2483 &assoc->ipsa_kcfauthkey, auth_ctx_tmpl,
2484 &ic->ic_crypto_mac, callrp);
2485 }
2486 }
2487
2488 if (do_encr) {
2489 /* encryption context template */
2490 IPSEC_CTX_TMPL(assoc, ipsa_encrtmpl, IPSEC_ALG_ENCR,
2491 encr_ctx_tmpl);
2492 /* Call the nonce update function. */
2493 (assoc->ipsa_noncefunc)(assoc, (uchar_t *)esph_ptr, payload_len,
2494 iv_ptr, &ic->ic_cmm, &ic->ic_crypto_data);
2495
2496 if (!do_auth) {
2497 /* encryption only, skip mblk that contains ESP hdr */
2498 /* initialize input data argument */
2499 ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
2500 esp_mp->b_cont, 0, payload_len);
2501
2502 /*
2503 * For combined mode ciphers, the ciphertext is the same
2504 * size as the clear text, the ICV should follow the
2505 * ciphertext. To convince the kcf to allow in-line
2506 * encryption, with an ICV, use ipsec_out_crypto_mac
2507 * to point to the same buffer as the data. The calling
2508 * function need to ensure the buffer is large enough to
2509 * include the ICV.
2510 *
2511 * The IV is already written to the packet buffer, the
2512 * nonce setup function copied it to the params struct
2513 * for the cipher to use.
2514 */
2515 if (assoc->ipsa_flags & IPSA_F_COMBINED) {
2516 bcopy(&ic->ic_crypto_data,
2517 &ic->ic_crypto_mac,
2518 sizeof (crypto_data_t));
2519 ic->ic_crypto_mac.cd_length =
2520 payload_len + icv_len;
2521 cd_ptr = &ic->ic_crypto_mac;
2522 }
2523
2524 /* call the crypto framework */
2525 kef_rc = crypto_encrypt((crypto_mechanism_t *)
2526 &ic->ic_cmm, &ic->ic_crypto_data,
2527 &assoc->ipsa_kcfencrkey, encr_ctx_tmpl,
2528 cd_ptr, callrp);
2529
2530 }
2531 }
2532
2533 if (do_auth && do_encr) {
2534 /*
2535 * Encryption and authentication:
2536 * Pass the pointer to the mblk chain starting at the ESP
2537 * header to the framework. Skip the ESP header mblk
2538 * for encryption, which is reflected by an encryption
2539 * offset equal to the length of that mblk. Start
2540 * the authentication at the ESP header, i.e. use an
2541 * authentication offset of zero.
2542 */
2543 ESP_INIT_CRYPTO_DUAL_DATA(&ic->ic_crypto_dual_data,
2544 esp_mp, MBLKL(esp_mp), payload_len, esph_offset, auth_len);
2545
2546 /* specify IV */
2547 ic->ic_crypto_dual_data.dd_miscdata = (char *)iv_ptr;
2548
2549 /* call the framework */
2550 kef_rc = crypto_encrypt_mac(&assoc->ipsa_emech,
2551 &assoc->ipsa_amech, NULL,
2552 &assoc->ipsa_kcfencrkey, &assoc->ipsa_kcfauthkey,
2553 encr_ctx_tmpl, auth_ctx_tmpl,
2554 &ic->ic_crypto_dual_data,
2555 &ic->ic_crypto_mac, callrp);
2556 }
2557
2558 switch (kef_rc) {
2559 case CRYPTO_SUCCESS:
2560 ESP_BUMP_STAT(espstack, crypto_sync);
2561 esp_set_usetime(assoc, B_FALSE);
2562 if (force) {
2563 mp = ipsec_free_crypto_data(mp);
2564 data_mp = ip_xmit_attr_free_mblk(mp);
2565 }
2566 if (is_natt)
2567 esp_prepare_udp(ns, data_mp, (ipha_t *)data_mp->b_rptr);
2568 return (data_mp);
2569 case CRYPTO_QUEUED:
2570 /* esp_kcf_callback_outbound() will be invoked on completion */
2571 ESP_BUMP_STAT(espstack, crypto_async);
2572 return (NULL);
2573 }
2574
2575 if (force) {
2576 mp = ipsec_free_crypto_data(mp);
2577 data_mp = ip_xmit_attr_free_mblk(mp);
2578 }
2579 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2580 esp_crypto_failed(data_mp, B_FALSE, kef_rc, NULL, espstack);
2581 /* data_mp was passed to ip_drop_packet */
2582 return (NULL);
2583 }
2584
2585 /*
2586 * Handle outbound IPsec processing for IPv4 and IPv6
2587 *
2588 * Returns data_mp if successfully completed the request. Returns
2589 * NULL if it failed (and increments InDiscards) or if it is pending.
2590 */
2591 static mblk_t *
2592 esp_outbound(mblk_t *data_mp, ip_xmit_attr_t *ixa)
2593 {
2594 mblk_t *espmp, *tailmp;
2595 ipha_t *ipha;
2596 ip6_t *ip6h;
2597 esph_t *esph_ptr, *iv_ptr;
2598 uint_t af;
2599 uint8_t *nhp;
2600 uintptr_t divpoint, datalen, adj, padlen, i, alloclen;
2601 uintptr_t esplen = sizeof (esph_t);
2602 uint8_t protocol;
2603 ipsa_t *assoc;
2604 uint_t iv_len, block_size, mac_len = 0;
2605 uchar_t *icv_buf;
2606 udpha_t *udpha;
2607 boolean_t is_natt = B_FALSE;
2608 netstack_t *ns = ixa->ixa_ipst->ips_netstack;
2609 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
2610 ipsec_stack_t *ipss = ns->netstack_ipsec;
2611 ill_t *ill = ixa->ixa_nce->nce_ill;
2612 boolean_t need_refrele = B_FALSE;
2613
2614 ESP_BUMP_STAT(espstack, out_requests);
2615
2616 /*
2617 * <sigh> We have to copy the message here, because TCP (for example)
2618 * keeps a dupb() of the message lying around for retransmission.
2619 * Since ESP changes the whole of the datagram, we have to create our
2620 * own copy lest we clobber TCP's data. Since we have to copy anyway,
2621 * we might as well make use of msgpullup() and get the mblk into one
2622 * contiguous piece!
2623 */
2624 tailmp = msgpullup(data_mp, -1);
2625 if (tailmp == NULL) {
2626 esp0dbg(("esp_outbound: msgpullup() failed, "
2627 "dropping packet.\n"));
2628 ip_drop_packet(data_mp, B_FALSE, ill,
2629 DROPPER(ipss, ipds_esp_nomem),
2630 &espstack->esp_dropper);
2631 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2632 return (NULL);
2633 }
2634 freemsg(data_mp);
2635 data_mp = tailmp;
2636
2637 assoc = ixa->ixa_ipsec_esp_sa;
2638 ASSERT(assoc != NULL);
2639
2640 /*
2641 * Get the outer IP header in shape to escape this system..
2642 */
2643 if (is_system_labeled() && (assoc->ipsa_otsl != NULL)) {
2644 /*
2645 * Need to update packet with any CIPSO option and update
2646 * ixa_tsl to capture the new label.
2647 * We allocate a separate ixa for that purpose.
2648 */
2649 ixa = ip_xmit_attr_duplicate(ixa);
2650 if (ixa == NULL) {
2651 ip_drop_packet(data_mp, B_FALSE, ill,
2652 DROPPER(ipss, ipds_esp_nomem),
2653 &espstack->esp_dropper);
2654 return (NULL);
2655 }
2656 need_refrele = B_TRUE;
2657
2658 label_hold(assoc->ipsa_otsl);
2659 ip_xmit_attr_replace_tsl(ixa, assoc->ipsa_otsl);
2660
2661 data_mp = sadb_whack_label(data_mp, assoc, ixa,
2662 DROPPER(ipss, ipds_esp_nomem), &espstack->esp_dropper);
2663 if (data_mp == NULL) {
2664 /* Packet dropped by sadb_whack_label */
2665 ixa_refrele(ixa);
2666 return (NULL);
2667 }
2668 }
2669
2670 /*
2671 * Reality check....
2672 */
2673 ipha = (ipha_t *)data_mp->b_rptr; /* So we can call esp_acquire(). */
2674
2675 if (ixa->ixa_flags & IXAF_IS_IPV4) {
2676 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
2677
2678 af = AF_INET;
2679 divpoint = IPH_HDR_LENGTH(ipha);
2680 datalen = ntohs(ipha->ipha_length) - divpoint;
2681 nhp = (uint8_t *)&ipha->ipha_protocol;
2682 } else {
2683 ip_pkt_t ipp;
2684
2685 ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
2686
2687 af = AF_INET6;
2688 ip6h = (ip6_t *)ipha;
2689 bzero(&ipp, sizeof (ipp));
2690 divpoint = ip_find_hdr_v6(data_mp, ip6h, B_FALSE, &ipp, NULL);
2691 if (ipp.ipp_dstopts != NULL &&
2692 ipp.ipp_dstopts->ip6d_nxt != IPPROTO_ROUTING) {
2693 /*
2694 * Destination options are tricky. If we get in here,
2695 * then we have a terminal header following the
2696 * destination options. We need to adjust backwards
2697 * so we insert ESP BEFORE the destination options
2698 * bag. (So that the dstopts get encrypted!)
2699 *
2700 * Since this is for outbound packets only, we know
2701 * that non-terminal destination options only precede
2702 * routing headers.
2703 */
2704 divpoint -= ipp.ipp_dstoptslen;
2705 }
2706 datalen = ntohs(ip6h->ip6_plen) + sizeof (ip6_t) - divpoint;
2707
2708 if (ipp.ipp_rthdr != NULL) {
2709 nhp = &ipp.ipp_rthdr->ip6r_nxt;
2710 } else if (ipp.ipp_hopopts != NULL) {
2711 nhp = &ipp.ipp_hopopts->ip6h_nxt;
2712 } else {
2713 ASSERT(divpoint == sizeof (ip6_t));
2714 /* It's probably IP + ESP. */
2715 nhp = &ip6h->ip6_nxt;
2716 }
2717 }
2718
2719 mac_len = assoc->ipsa_mac_len;
2720
2721 if (assoc->ipsa_flags & IPSA_F_NATT) {
2722 /* wedge in UDP header */
2723 is_natt = B_TRUE;
2724 esplen += UDPH_SIZE;
2725 }
2726
2727 /*
2728 * Set up ESP header and encryption padding for ENCR PI request.
2729 */
2730
2731 /* Determine the padding length. Pad to 4-bytes for no-encryption. */
2732 if (assoc->ipsa_encr_alg != SADB_EALG_NULL) {
2733 iv_len = assoc->ipsa_iv_len;
2734 block_size = assoc->ipsa_datalen;
2735
2736 /*
2737 * Pad the data to the length of the cipher block size.
2738 * Include the two additional bytes (hence the - 2) for the
2739 * padding length and the next header. Take this into account
2740 * when calculating the actual length of the padding.
2741 */
2742 ASSERT(ISP2(iv_len));
2743 padlen = ((unsigned)(block_size - datalen - 2)) &
2744 (block_size - 1);
2745 } else {
2746 iv_len = 0;
2747 padlen = ((unsigned)(sizeof (uint32_t) - datalen - 2)) &
2748 (sizeof (uint32_t) - 1);
2749 }
2750
2751 /* Allocate ESP header and IV. */
2752 esplen += iv_len;
2753
2754 /*
2755 * Update association byte-count lifetimes. Don't forget to take
2756 * into account the padding length and next-header (hence the + 2).
2757 *
2758 * Use the amount of data fed into the "encryption algorithm". This
2759 * is the IV, the data length, the padding length, and the final two
2760 * bytes (padlen, and next-header).
2761 *
2762 */
2763
2764 if (!esp_age_bytes(assoc, datalen + padlen + iv_len + 2, B_FALSE)) {
2765 ip_drop_packet(data_mp, B_FALSE, ill,
2766 DROPPER(ipss, ipds_esp_bytes_expire),
2767 &espstack->esp_dropper);
2768 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2769 if (need_refrele)
2770 ixa_refrele(ixa);
2771 return (NULL);
2772 }
2773
2774 espmp = allocb(esplen, BPRI_HI);
2775 if (espmp == NULL) {
2776 ESP_BUMP_STAT(espstack, out_discards);
2777 esp1dbg(espstack, ("esp_outbound: can't allocate espmp.\n"));
2778 ip_drop_packet(data_mp, B_FALSE, ill,
2779 DROPPER(ipss, ipds_esp_nomem),
2780 &espstack->esp_dropper);
2781 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2782 if (need_refrele)
2783 ixa_refrele(ixa);
2784 return (NULL);
2785 }
2786 espmp->b_wptr += esplen;
2787 esph_ptr = (esph_t *)espmp->b_rptr;
2788
2789 if (is_natt) {
2790 esp3dbg(espstack, ("esp_outbound: NATT"));
2791
2792 udpha = (udpha_t *)espmp->b_rptr;
2793 udpha->uha_src_port = (assoc->ipsa_local_nat_port != 0) ?
2794 assoc->ipsa_local_nat_port : htons(IPPORT_IKE_NATT);
2795 udpha->uha_dst_port = (assoc->ipsa_remote_nat_port != 0) ?
2796 assoc->ipsa_remote_nat_port : htons(IPPORT_IKE_NATT);
2797 /*
2798 * Set the checksum to 0, so that the esp_prepare_udp() call
2799 * can do the right thing.
2800 */
2801 udpha->uha_checksum = 0;
2802 esph_ptr = (esph_t *)(udpha + 1);
2803 }
2804
2805 esph_ptr->esph_spi = assoc->ipsa_spi;
2806
2807 esph_ptr->esph_replay = htonl(atomic_inc_32_nv(&assoc->ipsa_replay));
2808 if (esph_ptr->esph_replay == 0 && assoc->ipsa_replay_wsize != 0) {
2809 /*
2810 * XXX We have replay counter wrapping.
2811 * We probably want to nuke this SA (and its peer).
2812 */
2813 ipsec_assocfailure(info.mi_idnum, 0, 0,
2814 SL_ERROR | SL_CONSOLE | SL_WARN,
2815 "Outbound ESP SA (0x%x, %s) has wrapped sequence.\n",
2816 esph_ptr->esph_spi, assoc->ipsa_dstaddr, af,
2817 espstack->ipsecesp_netstack);
2818
2819 ESP_BUMP_STAT(espstack, out_discards);
2820 sadb_replay_delete(assoc);
2821 ip_drop_packet(data_mp, B_FALSE, ill,
2822 DROPPER(ipss, ipds_esp_replay),
2823 &espstack->esp_dropper);
2824 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2825 if (need_refrele)
2826 ixa_refrele(ixa);
2827 return (NULL);
2828 }
2829
2830 iv_ptr = (esph_ptr + 1);
2831 /*
2832 * iv_ptr points to the mblk which will contain the IV once we have
2833 * written it there. This mblk will be part of a mblk chain that
2834 * will make up the packet.
2835 *
2836 * For counter mode algorithms, the IV is a 64 bit quantity, it
2837 * must NEVER repeat in the lifetime of the SA, otherwise an
2838 * attacker who had recorded enough packets might be able to
2839 * determine some clear text.
2840 *
2841 * To ensure this does not happen, the IV is stored in the SA and
2842 * incremented for each packet, the IV is then copied into the
2843 * "packet" for transmission to the receiving system. The IV will
2844 * also be copied into the nonce, when the packet is encrypted.
2845 *
2846 * CBC mode algorithms use a random IV for each packet. We do not
2847 * require the highest quality random bits, but for best security
2848 * with CBC mode ciphers, the value must be unlikely to repeat and
2849 * must not be known in advance to an adversary capable of influencing
2850 * the clear text.
2851 */
2852 if (!update_iv((uint8_t *)iv_ptr, espstack->esp_pfkey_q, assoc,
2853 espstack)) {
2854 ip_drop_packet(data_mp, B_FALSE, ill,
2855 DROPPER(ipss, ipds_esp_iv_wrap), &espstack->esp_dropper);
2856 if (need_refrele)
2857 ixa_refrele(ixa);
2858 return (NULL);
2859 }
2860
2861 /* Fix the IP header. */
2862 alloclen = padlen + 2 + mac_len;
2863 adj = alloclen + (espmp->b_wptr - espmp->b_rptr);
2864
2865 protocol = *nhp;
2866
2867 if (ixa->ixa_flags & IXAF_IS_IPV4) {
2868 ipha->ipha_length = htons(ntohs(ipha->ipha_length) + adj);
2869 if (is_natt) {
2870 *nhp = IPPROTO_UDP;
2871 udpha->uha_length = htons(ntohs(ipha->ipha_length) -
2872 IPH_HDR_LENGTH(ipha));
2873 } else {
2874 *nhp = IPPROTO_ESP;
2875 }
2876 ipha->ipha_hdr_checksum = 0;
2877 ipha->ipha_hdr_checksum = (uint16_t)ip_csum_hdr(ipha);
2878 } else {
2879 ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) + adj);
2880 *nhp = IPPROTO_ESP;
2881 }
2882
2883 /* I've got the two ESP mblks, now insert them. */
2884
2885 esp2dbg(espstack, ("data_mp before outbound ESP adjustment:\n"));
2886 esp2dbg(espstack, (dump_msg(data_mp)));
2887
2888 if (!esp_insert_esp(data_mp, espmp, divpoint, espstack)) {
2889 ESP_BUMP_STAT(espstack, out_discards);
2890 /* NOTE: esp_insert_esp() only fails if there's no memory. */
2891 ip_drop_packet(data_mp, B_FALSE, ill,
2892 DROPPER(ipss, ipds_esp_nomem),
2893 &espstack->esp_dropper);
2894 freeb(espmp);
2895 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2896 if (need_refrele)
2897 ixa_refrele(ixa);
2898 return (NULL);
2899 }
2900
2901 /* Append padding (and leave room for ICV). */
2902 for (tailmp = data_mp; tailmp->b_cont != NULL; tailmp = tailmp->b_cont)
2903 ;
2904 if (tailmp->b_wptr + alloclen > tailmp->b_datap->db_lim) {
2905 tailmp->b_cont = allocb(alloclen, BPRI_HI);
2906 if (tailmp->b_cont == NULL) {
2907 ESP_BUMP_STAT(espstack, out_discards);
2908 esp0dbg(("esp_outbound: Can't allocate tailmp.\n"));
2909 ip_drop_packet(data_mp, B_FALSE, ill,
2910 DROPPER(ipss, ipds_esp_nomem),
2911 &espstack->esp_dropper);
2912 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2913 if (need_refrele)
2914 ixa_refrele(ixa);
2915 return (NULL);
2916 }
2917 tailmp = tailmp->b_cont;
2918 }
2919
2920 /*
2921 * If there's padding, N bytes of padding must be of the form 0x1,
2922 * 0x2, 0x3... 0xN.
2923 */
2924 for (i = 0; i < padlen; ) {
2925 i++;
2926 *tailmp->b_wptr++ = i;
2927 }
2928 *tailmp->b_wptr++ = i;
2929 *tailmp->b_wptr++ = protocol;
2930
2931 esp2dbg(espstack, ("data_Mp before encryption:\n"));
2932 esp2dbg(espstack, (dump_msg(data_mp)));
2933
2934 /*
2935 * Okay. I've set up the pre-encryption ESP. Let's do it!
2936 */
2937
2938 if (mac_len > 0) {
2939 ASSERT(tailmp->b_wptr + mac_len <= tailmp->b_datap->db_lim);
2940 icv_buf = tailmp->b_wptr;
2941 tailmp->b_wptr += mac_len;
2942 } else {
2943 icv_buf = NULL;
2944 }
2945
2946 data_mp = esp_submit_req_outbound(data_mp, ixa, assoc, icv_buf,
2947 datalen + padlen + 2);
2948 if (need_refrele)
2949 ixa_refrele(ixa);
2950 return (data_mp);
2951 }
2952
2953 /*
2954 * IP calls this to validate the ICMP errors that
2955 * we got from the network.
2956 */
2957 mblk_t *
2958 ipsecesp_icmp_error(mblk_t *data_mp, ip_recv_attr_t *ira)
2959 {
2960 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
2961 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
2962 ipsec_stack_t *ipss = ns->netstack_ipsec;
2963
2964 /*
2965 * Unless we get an entire packet back, this function is useless.
2966 * Why?
2967 *
2968 * 1.) Partial packets are useless, because the "next header"
2969 * is at the end of the decrypted ESP packet. Without the
2970 * whole packet, this is useless.
2971 *
2972 * 2.) If we every use a stateful cipher, such as a stream or a
2973 * one-time pad, we can't do anything.
2974 *
2975 * Since the chances of us getting an entire packet back are very
2976 * very small, we discard here.
2977 */
2978 IP_ESP_BUMP_STAT(ipss, in_discards);
2979 ip_drop_packet(data_mp, B_TRUE, ira->ira_ill,
2980 DROPPER(ipss, ipds_esp_icmp),
2981 &espstack->esp_dropper);
2982 return (NULL);
2983 }
2984
2985 /*
2986 * Construct an SADB_REGISTER message with the current algorithms.
2987 * This function gets called when 'ipsecalgs -s' is run or when
2988 * in.iked (or other KMD) starts.
2989 */
2990 static boolean_t
2991 esp_register_out(uint32_t sequence, uint32_t pid, uint_t serial,
2992 ipsecesp_stack_t *espstack, cred_t *cr)
2993 {
2994 mblk_t *pfkey_msg_mp, *keysock_out_mp;
2995 sadb_msg_t *samsg;
2996 sadb_supported_t *sasupp_auth = NULL;
2997 sadb_supported_t *sasupp_encr = NULL;
2998 sadb_alg_t *saalg;
2999 uint_t allocsize = sizeof (*samsg);
3000 uint_t i, numalgs_snap;
3001 int current_aalgs;
3002 ipsec_alginfo_t **authalgs;
3003 uint_t num_aalgs;
3004 int current_ealgs;
3005 ipsec_alginfo_t **encralgs;
3006 uint_t num_ealgs;
3007 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec;
3008 sadb_sens_t *sens;
3009 size_t sens_len = 0;
3010 sadb_ext_t *nextext;
3011 ts_label_t *sens_tsl = NULL;
3012
3013 /* Allocate the KEYSOCK_OUT. */
3014 keysock_out_mp = sadb_keysock_out(serial);
3015 if (keysock_out_mp == NULL) {
3016 esp0dbg(("esp_register_out: couldn't allocate mblk.\n"));
3017 return (B_FALSE);
3018 }
3019
3020 if (is_system_labeled() && (cr != NULL)) {
3021 sens_tsl = crgetlabel(cr);
3022 if (sens_tsl != NULL) {
3023 sens_len = sadb_sens_len_from_label(sens_tsl);
3024 allocsize += sens_len;
3025 }
3026 }
3027
3028 /*
3029 * Allocate the PF_KEY message that follows KEYSOCK_OUT.
3030 */
3031
3032 mutex_enter(&ipss->ipsec_alg_lock);
3033 /*
3034 * Fill SADB_REGISTER message's algorithm descriptors. Hold
3035 * down the lock while filling it.
3036 *
3037 * Return only valid algorithms, so the number of algorithms
3038 * to send up may be less than the number of algorithm entries
3039 * in the table.
3040 */
3041 authalgs = ipss->ipsec_alglists[IPSEC_ALG_AUTH];
3042 for (num_aalgs = 0, i = 0; i < IPSEC_MAX_ALGS; i++)
3043 if (authalgs[i] != NULL && ALG_VALID(authalgs[i]))
3044 num_aalgs++;
3045
3046 if (num_aalgs != 0) {
3047 allocsize += (num_aalgs * sizeof (*saalg));
3048 allocsize += sizeof (*sasupp_auth);
3049 }
3050 encralgs = ipss->ipsec_alglists[IPSEC_ALG_ENCR];
3051 for (num_ealgs = 0, i = 0; i < IPSEC_MAX_ALGS; i++)
3052 if (encralgs[i] != NULL && ALG_VALID(encralgs[i]))
3053 num_ealgs++;
3054
3055 if (num_ealgs != 0) {
3056 allocsize += (num_ealgs * sizeof (*saalg));
3057 allocsize += sizeof (*sasupp_encr);
3058 }
3059 keysock_out_mp->b_cont = allocb(allocsize, BPRI_HI);
3060 if (keysock_out_mp->b_cont == NULL) {
3061 mutex_exit(&ipss->ipsec_alg_lock);
3062 freemsg(keysock_out_mp);
3063 return (B_FALSE);
3064 }
3065 pfkey_msg_mp = keysock_out_mp->b_cont;
3066 pfkey_msg_mp->b_wptr += allocsize;
3067
3068 nextext = (sadb_ext_t *)(pfkey_msg_mp->b_rptr + sizeof (*samsg));
3069
3070 if (num_aalgs != 0) {
3071 sasupp_auth = (sadb_supported_t *)nextext;
3072 saalg = (sadb_alg_t *)(sasupp_auth + 1);
3073
3074 ASSERT(((ulong_t)saalg & 0x7) == 0);
3075
3076 numalgs_snap = 0;
3077 for (i = 0;
3078 ((i < IPSEC_MAX_ALGS) && (numalgs_snap < num_aalgs));
3079 i++) {
3080 if (authalgs[i] == NULL || !ALG_VALID(authalgs[i]))
3081 continue;
3082
3083 saalg->sadb_alg_id = authalgs[i]->alg_id;
3084 saalg->sadb_alg_ivlen = 0;
3085 saalg->sadb_alg_minbits = authalgs[i]->alg_ef_minbits;
3086 saalg->sadb_alg_maxbits = authalgs[i]->alg_ef_maxbits;
3087 saalg->sadb_x_alg_increment =
3088 authalgs[i]->alg_increment;
3089 saalg->sadb_x_alg_saltbits = SADB_8TO1(
3090 authalgs[i]->alg_saltlen);
3091 numalgs_snap++;
3092 saalg++;
3093 }
3094 ASSERT(numalgs_snap == num_aalgs);
3095 #ifdef DEBUG
3096 /*
3097 * Reality check to make sure I snagged all of the
3098 * algorithms.
3099 */
3100 for (; i < IPSEC_MAX_ALGS; i++) {
3101 if (authalgs[i] != NULL && ALG_VALID(authalgs[i])) {
3102 cmn_err(CE_PANIC, "esp_register_out()! "
3103 "Missed aalg #%d.\n", i);
3104 }
3105 }
3106 #endif /* DEBUG */
3107 nextext = (sadb_ext_t *)saalg;
3108 }
3109
3110 if (num_ealgs != 0) {
3111 sasupp_encr = (sadb_supported_t *)nextext;
3112 saalg = (sadb_alg_t *)(sasupp_encr + 1);
3113
3114 numalgs_snap = 0;
3115 for (i = 0;
3116 ((i < IPSEC_MAX_ALGS) && (numalgs_snap < num_ealgs)); i++) {
3117 if (encralgs[i] == NULL || !ALG_VALID(encralgs[i]))
3118 continue;
3119 saalg->sadb_alg_id = encralgs[i]->alg_id;
3120 saalg->sadb_alg_ivlen = encralgs[i]->alg_ivlen;
3121 saalg->sadb_alg_minbits = encralgs[i]->alg_ef_minbits;
3122 saalg->sadb_alg_maxbits = encralgs[i]->alg_ef_maxbits;
3123 /*
3124 * We could advertise the ICV length, except there
3125 * is not a value in sadb_x_algb to do this.
3126 * saalg->sadb_alg_maclen = encralgs[i]->alg_maclen;
3127 */
3128 saalg->sadb_x_alg_increment =
3129 encralgs[i]->alg_increment;
3130 saalg->sadb_x_alg_saltbits =
3131 SADB_8TO1(encralgs[i]->alg_saltlen);
3132
3133 numalgs_snap++;
3134 saalg++;
3135 }
3136 ASSERT(numalgs_snap == num_ealgs);
3137 #ifdef DEBUG
3138 /*
3139 * Reality check to make sure I snagged all of the
3140 * algorithms.
3141 */
3142 for (; i < IPSEC_MAX_ALGS; i++) {
3143 if (encralgs[i] != NULL && ALG_VALID(encralgs[i])) {
3144 cmn_err(CE_PANIC, "esp_register_out()! "
3145 "Missed ealg #%d.\n", i);
3146 }
3147 }
3148 #endif /* DEBUG */
3149 nextext = (sadb_ext_t *)saalg;
3150 }
3151
3152 current_aalgs = num_aalgs;
3153 current_ealgs = num_ealgs;
3154
3155 mutex_exit(&ipss->ipsec_alg_lock);
3156
3157 if (sens_tsl != NULL) {
3158 sens = (sadb_sens_t *)nextext;
3159 sadb_sens_from_label(sens, SADB_EXT_SENSITIVITY,
3160 sens_tsl, sens_len);
3161
3162 nextext = (sadb_ext_t *)(((uint8_t *)sens) + sens_len);
3163 }
3164
3165 /* Now fill the rest of the SADB_REGISTER message. */
3166
3167 samsg = (sadb_msg_t *)pfkey_msg_mp->b_rptr;
3168 samsg->sadb_msg_version = PF_KEY_V2;
3169 samsg->sadb_msg_type = SADB_REGISTER;
3170 samsg->sadb_msg_errno = 0;
3171 samsg->sadb_msg_satype = SADB_SATYPE_ESP;
3172 samsg->sadb_msg_len = SADB_8TO64(allocsize);
3173 samsg->sadb_msg_reserved = 0;
3174 /*
3175 * Assume caller has sufficient sequence/pid number info. If it's one
3176 * from me over a new alg., I could give two hoots about sequence.
3177 */
3178 samsg->sadb_msg_seq = sequence;
3179 samsg->sadb_msg_pid = pid;
3180
3181 if (sasupp_auth != NULL) {
3182 sasupp_auth->sadb_supported_len = SADB_8TO64(
3183 sizeof (*sasupp_auth) + sizeof (*saalg) * current_aalgs);
3184 sasupp_auth->sadb_supported_exttype = SADB_EXT_SUPPORTED_AUTH;
3185 sasupp_auth->sadb_supported_reserved = 0;
3186 }
3187
3188 if (sasupp_encr != NULL) {
3189 sasupp_encr->sadb_supported_len = SADB_8TO64(
3190 sizeof (*sasupp_encr) + sizeof (*saalg) * current_ealgs);
3191 sasupp_encr->sadb_supported_exttype =
3192 SADB_EXT_SUPPORTED_ENCRYPT;
3193 sasupp_encr->sadb_supported_reserved = 0;
3194 }
3195
3196 if (espstack->esp_pfkey_q != NULL)
3197 putnext(espstack->esp_pfkey_q, keysock_out_mp);
3198 else {
3199 freemsg(keysock_out_mp);
3200 return (B_FALSE);
3201 }
3202
3203 return (B_TRUE);
3204 }
3205
3206 /*
3207 * Invoked when the algorithm table changes. Causes SADB_REGISTER
3208 * messages continaining the current list of algorithms to be
3209 * sent up to the ESP listeners.
3210 */
3211 void
3212 ipsecesp_algs_changed(netstack_t *ns)
3213 {
3214 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3215
3216 /*
3217 * Time to send a PF_KEY SADB_REGISTER message to ESP listeners
3218 * everywhere. (The function itself checks for NULL esp_pfkey_q.)
3219 */
3220 (void) esp_register_out(0, 0, 0, espstack, NULL);
3221 }
3222
3223 /*
3224 * Stub function that taskq_dispatch() invokes to take the mblk (in arg)
3225 * and send() it into ESP and IP again.
3226 */
3227 static void
3228 inbound_task(void *arg)
3229 {
3230 mblk_t *mp = (mblk_t *)arg;
3231 mblk_t *async_mp;
3232 ip_recv_attr_t iras;
3233
3234 async_mp = mp;
3235 mp = async_mp->b_cont;
3236 async_mp->b_cont = NULL;
3237 if (!ip_recv_attr_from_mblk(async_mp, &iras)) {
3238 /* The ill or ip_stack_t disappeared on us */
3239 ip_drop_input("ip_recv_attr_from_mblk", mp, NULL);
3240 freemsg(mp);
3241 goto done;
3242 }
3243
3244 esp_inbound_restart(mp, &iras);
3245 done:
3246 ira_cleanup(&iras, B_TRUE);
3247 }
3248
3249 /*
3250 * Restart ESP after the SA has been added.
3251 */
3252 static void
3253 esp_inbound_restart(mblk_t *mp, ip_recv_attr_t *ira)
3254 {
3255 esph_t *esph;
3256 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
3257 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3258
3259 esp2dbg(espstack, ("in ESP inbound_task"));
3260 ASSERT(espstack != NULL);
3261
3262 mp = ipsec_inbound_esp_sa(mp, ira, &esph);
3263 if (mp == NULL)
3264 return;
3265
3266 ASSERT(esph != NULL);
3267 ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
3268 ASSERT(ira->ira_ipsec_esp_sa != NULL);
3269
3270 mp = ira->ira_ipsec_esp_sa->ipsa_input_func(mp, esph, ira);
3271 if (mp == NULL) {
3272 /*
3273 * Either it failed or is pending. In the former case
3274 * ipIfStatsInDiscards was increased.
3275 */
3276 return;
3277 }
3278
3279 ip_input_post_ipsec(mp, ira);
3280 }
3281
3282 /*
3283 * Now that weak-key passed, actually ADD the security association, and
3284 * send back a reply ADD message.
3285 */
3286 static int
3287 esp_add_sa_finish(mblk_t *mp, sadb_msg_t *samsg, keysock_in_t *ksi,
3288 int *diagnostic, ipsecesp_stack_t *espstack)
3289 {
3290 isaf_t *primary = NULL, *secondary;
3291 boolean_t clone = B_FALSE, is_inbound = B_FALSE;
3292 ipsa_t *larval = NULL;
3293 ipsacq_t *acqrec;
3294 iacqf_t *acq_bucket;
3295 mblk_t *acq_msgs = NULL;
3296 int rc;
3297 mblk_t *lpkt;
3298 int error;
3299 ipsa_query_t sq;
3300 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec;
3301
3302 /*
3303 * Locate the appropriate table(s).
3304 */
3305 sq.spp = &espstack->esp_sadb; /* XXX */
3306 error = sadb_form_query(ksi, IPSA_Q_SA|IPSA_Q_DST,
3307 IPSA_Q_SA|IPSA_Q_DST|IPSA_Q_INBOUND|IPSA_Q_OUTBOUND,
3308 &sq, diagnostic);
3309 if (error)
3310 return (error);
3311
3312 /*
3313 * Use the direction flags provided by the KMD to determine
3314 * if the inbound or outbound table should be the primary
3315 * for this SA. If these flags were absent then make this
3316 * decision based on the addresses.
3317 */
3318 if (sq.assoc->sadb_sa_flags & IPSA_F_INBOUND) {
3319 primary = sq.inbound;
3320 secondary = sq.outbound;
3321 is_inbound = B_TRUE;
3322 if (sq.assoc->sadb_sa_flags & IPSA_F_OUTBOUND)
3323 clone = B_TRUE;
3324 } else if (sq.assoc->sadb_sa_flags & IPSA_F_OUTBOUND) {
3325 primary = sq.outbound;
3326 secondary = sq.inbound;
3327 }
3328
3329 if (primary == NULL) {
3330 /*
3331 * The KMD did not set a direction flag, determine which
3332 * table to insert the SA into based on addresses.
3333 */
3334 switch (ksi->ks_in_dsttype) {
3335 case KS_IN_ADDR_MBCAST:
3336 clone = B_TRUE; /* All mcast SAs can be bidirectional */
3337 sq.assoc->sadb_sa_flags |= IPSA_F_OUTBOUND;
3338 /* FALLTHRU */
3339 /*
3340 * If the source address is either one of mine, or unspecified
3341 * (which is best summed up by saying "not 'not mine'"),
3342 * then the association is potentially bi-directional,
3343 * in that it can be used for inbound traffic and outbound
3344 * traffic. The best example of such an SA is a multicast
3345 * SA (which allows me to receive the outbound traffic).
3346 */
3347 case KS_IN_ADDR_ME:
3348 sq.assoc->sadb_sa_flags |= IPSA_F_INBOUND;
3349 primary = sq.inbound;
3350 secondary = sq.outbound;
3351 if (ksi->ks_in_srctype != KS_IN_ADDR_NOTME)
3352 clone = B_TRUE;
3353 is_inbound = B_TRUE;
3354 break;
3355 /*
3356 * If the source address literally not mine (either
3357 * unspecified or not mine), then this SA may have an
3358 * address that WILL be mine after some configuration.
3359 * We pay the price for this by making it a bi-directional
3360 * SA.
3361 */
3362 case KS_IN_ADDR_NOTME:
3363 sq.assoc->sadb_sa_flags |= IPSA_F_OUTBOUND;
3364 primary = sq.outbound;
3365 secondary = sq.inbound;
3366 if (ksi->ks_in_srctype != KS_IN_ADDR_ME) {
3367 sq.assoc->sadb_sa_flags |= IPSA_F_INBOUND;
3368 clone = B_TRUE;
3369 }
3370 break;
3371 default:
3372 *diagnostic = SADB_X_DIAGNOSTIC_BAD_DST;
3373 return (EINVAL);
3374 }
3375 }
3376
3377 /*
3378 * Find a ACQUIRE list entry if possible. If we've added an SA that
3379 * suits the needs of an ACQUIRE list entry, we can eliminate the
3380 * ACQUIRE list entry and transmit the enqueued packets. Use the
3381 * high-bit of the sequence number to queue it. Key off destination
3382 * addr, and change acqrec's state.
3383 */
3384
3385 if (samsg->sadb_msg_seq & IACQF_LOWEST_SEQ) {
3386 acq_bucket = &(sq.sp->sdb_acq[sq.outhash]);
3387 mutex_enter(&acq_bucket->iacqf_lock);
3388 for (acqrec = acq_bucket->iacqf_ipsacq; acqrec != NULL;
3389 acqrec = acqrec->ipsacq_next) {
3390 mutex_enter(&acqrec->ipsacq_lock);
3391 /*
3392 * Q: I only check sequence. Should I check dst?
3393 * A: Yes, check dest because those are the packets
3394 * that are queued up.
3395 */
3396 if (acqrec->ipsacq_seq == samsg->sadb_msg_seq &&
3397 IPSA_ARE_ADDR_EQUAL(sq.dstaddr,
3398 acqrec->ipsacq_dstaddr, acqrec->ipsacq_addrfam))
3399 break;
3400 mutex_exit(&acqrec->ipsacq_lock);
3401 }
3402 if (acqrec != NULL) {
3403 /*
3404 * AHA! I found an ACQUIRE record for this SA.
3405 * Grab the msg list, and free the acquire record.
3406 * I already am holding the lock for this record,
3407 * so all I have to do is free it.
3408 */
3409 acq_msgs = acqrec->ipsacq_mp;
3410 acqrec->ipsacq_mp = NULL;
3411 mutex_exit(&acqrec->ipsacq_lock);
3412 sadb_destroy_acquire(acqrec,
3413 espstack->ipsecesp_netstack);
3414 }
3415 mutex_exit(&acq_bucket->iacqf_lock);
3416 }
3417
3418 /*
3419 * Find PF_KEY message, and see if I'm an update. If so, find entry
3420 * in larval list (if there).
3421 */
3422 if (samsg->sadb_msg_type == SADB_UPDATE) {
3423 mutex_enter(&sq.inbound->isaf_lock);
3424 larval = ipsec_getassocbyspi(sq.inbound, sq.assoc->sadb_sa_spi,
3425 ALL_ZEROES_PTR, sq.dstaddr, sq.dst->sin_family);
3426 mutex_exit(&sq.inbound->isaf_lock);
3427
3428 if ((larval == NULL) ||
3429 (larval->ipsa_state != IPSA_STATE_LARVAL)) {
3430 *diagnostic = SADB_X_DIAGNOSTIC_SA_NOTFOUND;
3431 if (larval != NULL) {
3432 IPSA_REFRELE(larval);
3433 }
3434 esp0dbg(("Larval update, but larval disappeared.\n"));
3435 return (ESRCH);
3436 } /* Else sadb_common_add unlinks it for me! */
3437 }
3438
3439 if (larval != NULL) {
3440 /*
3441 * Hold again, because sadb_common_add() consumes a reference,
3442 * and we don't want to clear_lpkt() without a reference.
3443 */
3444 IPSA_REFHOLD(larval);
3445 }
3446
3447 rc = sadb_common_add(espstack->esp_pfkey_q,
3448 mp, samsg, ksi, primary, secondary, larval, clone, is_inbound,
3449 diagnostic, espstack->ipsecesp_netstack, &espstack->esp_sadb);
3450
3451 if (larval != NULL) {
3452 if (rc == 0) {
3453 lpkt = sadb_clear_lpkt(larval);
3454 if (lpkt != NULL) {
3455 rc = !taskq_dispatch(esp_taskq, inbound_task,
3456 lpkt, TQ_NOSLEEP);
3457 }
3458 }
3459 IPSA_REFRELE(larval);
3460 }
3461
3462 /*
3463 * How much more stack will I create with all of these
3464 * esp_outbound() calls?
3465 */
3466
3467 /* Handle the packets queued waiting for the SA */
3468 while (acq_msgs != NULL) {
3469 mblk_t *asyncmp;
3470 mblk_t *data_mp;
3471 ip_xmit_attr_t ixas;
3472 ill_t *ill;
3473
3474 asyncmp = acq_msgs;
3475 acq_msgs = acq_msgs->b_next;
3476 asyncmp->b_next = NULL;
3477
3478 /*
3479 * Extract the ip_xmit_attr_t from the first mblk.
3480 * Verifies that the netstack and ill is still around; could
3481 * have vanished while iked was doing its work.
3482 * On succesful return we have a nce_t and the ill/ipst can't
3483 * disappear until we do the nce_refrele in ixa_cleanup.
3484 */
3485 data_mp = asyncmp->b_cont;
3486 asyncmp->b_cont = NULL;
3487 if (!ip_xmit_attr_from_mblk(asyncmp, &ixas)) {
3488 ESP_BUMP_STAT(espstack, out_discards);
3489 ip_drop_packet(data_mp, B_FALSE, NULL,
3490 DROPPER(ipss, ipds_sadb_acquire_timeout),
3491 &espstack->esp_dropper);
3492 } else if (rc != 0) {
3493 ill = ixas.ixa_nce->nce_ill;
3494 ESP_BUMP_STAT(espstack, out_discards);
3495 ip_drop_packet(data_mp, B_FALSE, ill,
3496 DROPPER(ipss, ipds_sadb_acquire_timeout),
3497 &espstack->esp_dropper);
3498 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
3499 } else {
3500 esp_outbound_finish(data_mp, &ixas);
3501 }
3502 ixa_cleanup(&ixas);
3503 }
3504
3505 return (rc);
3506 }
3507
3508 /*
3509 * Process one of the queued messages (from ipsacq_mp) once the SA
3510 * has been added.
3511 */
3512 static void
3513 esp_outbound_finish(mblk_t *data_mp, ip_xmit_attr_t *ixa)
3514 {
3515 netstack_t *ns = ixa->ixa_ipst->ips_netstack;
3516 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3517 ipsec_stack_t *ipss = ns->netstack_ipsec;
3518 ill_t *ill = ixa->ixa_nce->nce_ill;
3519
3520 if (!ipsec_outbound_sa(data_mp, ixa, IPPROTO_ESP)) {
3521 ESP_BUMP_STAT(espstack, out_discards);
3522 ip_drop_packet(data_mp, B_FALSE, ill,
3523 DROPPER(ipss, ipds_sadb_acquire_timeout),
3524 &espstack->esp_dropper);
3525 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
3526 return;
3527 }
3528
3529 data_mp = esp_outbound(data_mp, ixa);
3530 if (data_mp == NULL)
3531 return;
3532
3533 /* do AH processing if needed */
3534 data_mp = esp_do_outbound_ah(data_mp, ixa);
3535 if (data_mp == NULL)
3536 return;
3537
3538 (void) ip_output_post_ipsec(data_mp, ixa);
3539 }
3540
3541 /*
3542 * Add new ESP security association. This may become a generic AH/ESP
3543 * routine eventually.
3544 */
3545 static int
3546 esp_add_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic, netstack_t *ns)
3547 {
3548 sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
3549 sadb_address_t *srcext =
3550 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
3551 sadb_address_t *dstext =
3552 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
3553 sadb_address_t *isrcext =
3554 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_SRC];
3555 sadb_address_t *idstext =
3556 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_DST];
3557 sadb_address_t *nttext_loc =
3558 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_LOC];
3559 sadb_address_t *nttext_rem =
3560 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_REM];
3561 sadb_key_t *akey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_AUTH];
3562 sadb_key_t *ekey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_ENCRYPT];
3563 struct sockaddr_in *src, *dst;
3564 struct sockaddr_in *natt_loc, *natt_rem;
3565 struct sockaddr_in6 *natt_loc6, *natt_rem6;
3566 sadb_lifetime_t *soft =
3567 (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_SOFT];
3568 sadb_lifetime_t *hard =
3569 (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_HARD];
3570 sadb_lifetime_t *idle =
3571 (sadb_lifetime_t *)ksi->ks_in_extv[SADB_X_EXT_LIFETIME_IDLE];
3572 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3573 ipsec_stack_t *ipss = ns->netstack_ipsec;
3574
3575
3576
3577 /* I need certain extensions present for an ADD message. */
3578 if (srcext == NULL) {
3579 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SRC;
3580 return (EINVAL);
3581 }
3582 if (dstext == NULL) {
3583 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
3584 return (EINVAL);
3585 }
3586 if (isrcext == NULL && idstext != NULL) {
3587 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_SRC;
3588 return (EINVAL);
3589 }
3590 if (isrcext != NULL && idstext == NULL) {
3591 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_DST;
3592 return (EINVAL);
3593 }
3594 if (assoc == NULL) {
3595 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
3596 return (EINVAL);
3597 }
3598 if (ekey == NULL && assoc->sadb_sa_encrypt != SADB_EALG_NULL) {
3599 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_EKEY;
3600 return (EINVAL);
3601 }
3602
3603 src = (struct sockaddr_in *)(srcext + 1);
3604 dst = (struct sockaddr_in *)(dstext + 1);
3605 natt_loc = (struct sockaddr_in *)(nttext_loc + 1);
3606 natt_loc6 = (struct sockaddr_in6 *)(nttext_loc + 1);
3607 natt_rem = (struct sockaddr_in *)(nttext_rem + 1);
3608 natt_rem6 = (struct sockaddr_in6 *)(nttext_rem + 1);
3609
3610 /* Sundry ADD-specific reality checks. */
3611 /* XXX STATS : Logging/stats here? */
3612
3613 if (assoc->sadb_sa_state != SADB_SASTATE_MATURE) {
3614 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
3615 return (EINVAL);
3616 }
3617 if (assoc->sadb_sa_encrypt == SADB_EALG_NONE) {
3618 *diagnostic = SADB_X_DIAGNOSTIC_BAD_EALG;
3619 return (EINVAL);
3620 }
3621
3622 #ifndef IPSEC_LATENCY_TEST
3623 if (assoc->sadb_sa_encrypt == SADB_EALG_NULL &&
3624 assoc->sadb_sa_auth == SADB_AALG_NONE) {
3625 *diagnostic = SADB_X_DIAGNOSTIC_BAD_AALG;
3626 return (EINVAL);
3627 }
3628 #endif
3629
3630 if (assoc->sadb_sa_flags & ~espstack->esp_sadb.s_addflags) {
3631 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SAFLAGS;
3632 return (EINVAL);
3633 }
3634
3635 if ((*diagnostic = sadb_hardsoftchk(hard, soft, idle)) != 0) {
3636 return (EINVAL);
3637 }
3638 ASSERT(src->sin_family == dst->sin_family);
3639
3640 if (assoc->sadb_sa_flags & SADB_X_SAFLAGS_NATT_LOC) {
3641 if (nttext_loc == NULL) {
3642 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_LOC;
3643 return (EINVAL);
3644 }
3645
3646 if (natt_loc->sin_family == AF_INET6 &&
3647 !IN6_IS_ADDR_V4MAPPED(&natt_loc6->sin6_addr)) {
3648 *diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_NATT_LOC;
3649 return (EINVAL);
3650 }
3651 }
3652
3653 if (assoc->sadb_sa_flags & SADB_X_SAFLAGS_NATT_REM) {
3654 if (nttext_rem == NULL) {
3655 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_REM;
3656 return (EINVAL);
3657 }
3658 if (natt_rem->sin_family == AF_INET6 &&
3659 !IN6_IS_ADDR_V4MAPPED(&natt_rem6->sin6_addr)) {
3660 *diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_NATT_REM;
3661 return (EINVAL);
3662 }
3663 }
3664
3665
3666 /* Stuff I don't support, for now. XXX Diagnostic? */
3667 if (ksi->ks_in_extv[SADB_EXT_LIFETIME_CURRENT] != NULL)
3668 return (EOPNOTSUPP);
3669
3670 if ((*diagnostic = sadb_labelchk(ksi)) != 0)
3671 return (EINVAL);
3672
3673 /*
3674 * XXX Policy : I'm not checking identities at this time,
3675 * but if I did, I'd do them here, before I sent
3676 * the weak key check up to the algorithm.
3677 */
3678
3679 mutex_enter(&ipss->ipsec_alg_lock);
3680
3681 /*
3682 * First locate the authentication algorithm.
3683 */
3684 #ifdef IPSEC_LATENCY_TEST
3685 if (akey != NULL && assoc->sadb_sa_auth != SADB_AALG_NONE) {
3686 #else
3687 if (akey != NULL) {
3688 #endif
3689 ipsec_alginfo_t *aalg;
3690
3691 aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
3692 [assoc->sadb_sa_auth];
3693 if (aalg == NULL || !ALG_VALID(aalg)) {
3694 mutex_exit(&ipss->ipsec_alg_lock);
3695 esp1dbg(espstack, ("Couldn't find auth alg #%d.\n",
3696 assoc->sadb_sa_auth));
3697 *diagnostic = SADB_X_DIAGNOSTIC_BAD_AALG;
3698 return (EINVAL);
3699 }
3700
3701 /*
3702 * Sanity check key sizes.
3703 * Note: It's not possible to use SADB_AALG_NONE because
3704 * this auth_alg is not defined with ALG_FLAG_VALID. If this
3705 * ever changes, the same check for SADB_AALG_NONE and
3706 * a auth_key != NULL should be made here ( see below).
3707 */
3708 if (!ipsec_valid_key_size(akey->sadb_key_bits, aalg)) {
3709 mutex_exit(&ipss->ipsec_alg_lock);
3710 *diagnostic = SADB_X_DIAGNOSTIC_BAD_AKEYBITS;
3711 return (EINVAL);
3712 }
3713 ASSERT(aalg->alg_mech_type != CRYPTO_MECHANISM_INVALID);
3714
3715 /* check key and fix parity if needed */
3716 if (ipsec_check_key(aalg->alg_mech_type, akey, B_TRUE,
3717 diagnostic) != 0) {
3718 mutex_exit(&ipss->ipsec_alg_lock);
3719 return (EINVAL);
3720 }
3721 }
3722
3723 /*
3724 * Then locate the encryption algorithm.
3725 */
3726 if (ekey != NULL) {
3727 uint_t keybits;
3728 ipsec_alginfo_t *ealg;
3729
3730 ealg = ipss->ipsec_alglists[IPSEC_ALG_ENCR]
3731 [assoc->sadb_sa_encrypt];
3732 if (ealg == NULL || !ALG_VALID(ealg)) {
3733 mutex_exit(&ipss->ipsec_alg_lock);
3734 esp1dbg(espstack, ("Couldn't find encr alg #%d.\n",
3735 assoc->sadb_sa_encrypt));
3736 *diagnostic = SADB_X_DIAGNOSTIC_BAD_EALG;
3737 return (EINVAL);
3738 }
3739
3740 /*
3741 * Sanity check key sizes. If the encryption algorithm is
3742 * SADB_EALG_NULL but the encryption key is NOT
3743 * NULL then complain.
3744 *
3745 * The keying material includes salt bits if required by
3746 * algorithm and optionally the Initial IV, check the
3747 * length of whats left.
3748 */
3749 keybits = ekey->sadb_key_bits;
3750 keybits -= ekey->sadb_key_reserved;
3751 keybits -= SADB_8TO1(ealg->alg_saltlen);
3752 if ((assoc->sadb_sa_encrypt == SADB_EALG_NULL) ||
3753 (!ipsec_valid_key_size(keybits, ealg))) {
3754 mutex_exit(&ipss->ipsec_alg_lock);
3755 *diagnostic = SADB_X_DIAGNOSTIC_BAD_EKEYBITS;
3756 return (EINVAL);
3757 }
3758 ASSERT(ealg->alg_mech_type != CRYPTO_MECHANISM_INVALID);
3759
3760 /* check key */
3761 if (ipsec_check_key(ealg->alg_mech_type, ekey, B_FALSE,
3762 diagnostic) != 0) {
3763 mutex_exit(&ipss->ipsec_alg_lock);
3764 return (EINVAL);
3765 }
3766 }
3767 mutex_exit(&ipss->ipsec_alg_lock);
3768
3769 return (esp_add_sa_finish(mp, (sadb_msg_t *)mp->b_cont->b_rptr, ksi,
3770 diagnostic, espstack));
3771 }
3772
3773 /*
3774 * Update a security association. Updates come in two varieties. The first
3775 * is an update of lifetimes on a non-larval SA. The second is an update of
3776 * a larval SA, which ends up looking a lot more like an add.
3777 */
3778 static int
3779 esp_update_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic,
3780 ipsecesp_stack_t *espstack, uint8_t sadb_msg_type)
3781 {
3782 sadb_address_t *dstext =
3783 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
3784
3785 if (dstext == NULL) {
3786 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
3787 return (EINVAL);
3788 }
3789
3790 return (sadb_update_sa(mp, ksi, &espstack->esp_sadb, diagnostic,
3791 espstack->esp_pfkey_q, esp_add_sa, espstack->ipsecesp_netstack,
3792 sadb_msg_type));
3793 }
3794
3795 /* XXX refactor me */
3796 /*
3797 * Delete a security association. This is REALLY likely to be code common to
3798 * both AH and ESP. Find the association, then unlink it.
3799 */
3800 static int
3801 esp_del_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic,
3802 ipsecesp_stack_t *espstack, uint8_t sadb_msg_type)
3803 {
3804 sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
3805 sadb_address_t *dstext =
3806 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
3807 sadb_address_t *srcext =
3808 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
3809 struct sockaddr_in *sin;
3810
3811 if (assoc == NULL) {
3812 if (dstext != NULL) {
3813 sin = (struct sockaddr_in *)(dstext + 1);
3814 } else if (srcext != NULL) {
3815 sin = (struct sockaddr_in *)(srcext + 1);
3816 } else {
3817 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
3818 return (EINVAL);
3819 }
3820 return (sadb_purge_sa(mp, ksi,
3821 (sin->sin_family == AF_INET6) ? &espstack->esp_sadb.s_v6 :
3822 &espstack->esp_sadb.s_v4, diagnostic,
3823 espstack->esp_pfkey_q));
3824 }
3825
3826 return (sadb_delget_sa(mp, ksi, &espstack->esp_sadb, diagnostic,
3827 espstack->esp_pfkey_q, sadb_msg_type));
3828 }
3829
3830 /* XXX refactor me */
3831 /*
3832 * Convert the entire contents of all of ESP's SA tables into PF_KEY SADB_DUMP
3833 * messages.
3834 */
3835 static void
3836 esp_dump(mblk_t *mp, keysock_in_t *ksi, ipsecesp_stack_t *espstack)
3837 {
3838 int error;
3839 sadb_msg_t *samsg;
3840
3841 /*
3842 * Dump each fanout, bailing if error is non-zero.
3843 */
3844
3845 error = sadb_dump(espstack->esp_pfkey_q, mp, ksi,
3846 &espstack->esp_sadb.s_v4);
3847 if (error != 0)
3848 goto bail;
3849
3850 error = sadb_dump(espstack->esp_pfkey_q, mp, ksi,
3851 &espstack->esp_sadb.s_v6);
3852 bail:
3853 ASSERT(mp->b_cont != NULL);
3854 samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
3855 samsg->sadb_msg_errno = (uint8_t)error;
3856 sadb_pfkey_echo(espstack->esp_pfkey_q, mp,
3857 (sadb_msg_t *)mp->b_cont->b_rptr, ksi, NULL);
3858 }
3859
3860 /*
3861 * First-cut reality check for an inbound PF_KEY message.
3862 */
3863 static boolean_t
3864 esp_pfkey_reality_failures(mblk_t *mp, keysock_in_t *ksi,
3865 ipsecesp_stack_t *espstack)
3866 {
3867 int diagnostic;
3868
3869 if (ksi->ks_in_extv[SADB_EXT_PROPOSAL] != NULL) {
3870 diagnostic = SADB_X_DIAGNOSTIC_PROP_PRESENT;
3871 goto badmsg;
3872 }
3873 if (ksi->ks_in_extv[SADB_EXT_SUPPORTED_AUTH] != NULL ||
3874 ksi->ks_in_extv[SADB_EXT_SUPPORTED_ENCRYPT] != NULL) {
3875 diagnostic = SADB_X_DIAGNOSTIC_SUPP_PRESENT;
3876 goto badmsg;
3877 }
3878 return (B_FALSE); /* False ==> no failures */
3879
3880 badmsg:
3881 sadb_pfkey_error(espstack->esp_pfkey_q, mp, EINVAL, diagnostic,
3882 ksi->ks_in_serial);
3883 return (B_TRUE); /* True ==> failures */
3884 }
3885
3886 /*
3887 * ESP parsing of PF_KEY messages. Keysock did most of the really silly
3888 * error cases. What I receive is a fully-formed, syntactically legal
3889 * PF_KEY message. I then need to check semantics...
3890 *
3891 * This code may become common to AH and ESP. Stay tuned.
3892 *
3893 * I also make the assumption that db_ref's are cool. If this assumption
3894 * is wrong, this means that someone other than keysock or me has been
3895 * mucking with PF_KEY messages.
3896 */
3897 static void
3898 esp_parse_pfkey(mblk_t *mp, ipsecesp_stack_t *espstack)
3899 {
3900 mblk_t *msg = mp->b_cont;
3901 sadb_msg_t *samsg;
3902 keysock_in_t *ksi;
3903 int error;
3904 int diagnostic = SADB_X_DIAGNOSTIC_NONE;
3905
3906 ASSERT(msg != NULL);
3907
3908 samsg = (sadb_msg_t *)msg->b_rptr;
3909 ksi = (keysock_in_t *)mp->b_rptr;
3910
3911 /*
3912 * If applicable, convert unspecified AF_INET6 to unspecified
3913 * AF_INET. And do other address reality checks.
3914 */
3915 if (!sadb_addrfix(ksi, espstack->esp_pfkey_q, mp,
3916 espstack->ipsecesp_netstack) ||
3917 esp_pfkey_reality_failures(mp, ksi, espstack)) {
3918 return;
3919 }
3920
3921 switch (samsg->sadb_msg_type) {
3922 case SADB_ADD:
3923 error = esp_add_sa(mp, ksi, &diagnostic,
3924 espstack->ipsecesp_netstack);
3925 if (error != 0) {
3926 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
3927 diagnostic, ksi->ks_in_serial);
3928 }
3929 /* else esp_add_sa() took care of things. */
3930 break;
3931 case SADB_DELETE:
3932 case SADB_X_DELPAIR:
3933 error = esp_del_sa(mp, ksi, &diagnostic, espstack,
3934 samsg->sadb_msg_type);
3935 if (error != 0) {
3936 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
3937 diagnostic, ksi->ks_in_serial);
3938 }
3939 /* Else esp_del_sa() took care of things. */
3940 break;
3941 case SADB_GET:
3942 error = sadb_delget_sa(mp, ksi, &espstack->esp_sadb,
3943 &diagnostic, espstack->esp_pfkey_q, samsg->sadb_msg_type);
3944 if (error != 0) {
3945 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
3946 diagnostic, ksi->ks_in_serial);
3947 }
3948 /* Else sadb_get_sa() took care of things. */
3949 break;
3950 case SADB_FLUSH:
3951 sadbp_flush(&espstack->esp_sadb, espstack->ipsecesp_netstack);
3952 sadb_pfkey_echo(espstack->esp_pfkey_q, mp, samsg, ksi, NULL);
3953 break;
3954 case SADB_REGISTER:
3955 /*
3956 * Hmmm, let's do it! Check for extensions (there should
3957 * be none), extract the fields, call esp_register_out(),
3958 * then either free or report an error.
3959 *
3960 * Keysock takes care of the PF_KEY bookkeeping for this.
3961 */
3962 if (esp_register_out(samsg->sadb_msg_seq, samsg->sadb_msg_pid,
3963 ksi->ks_in_serial, espstack, msg_getcred(mp, NULL))) {
3964 freemsg(mp);
3965 } else {
3966 /*
3967 * Only way this path hits is if there is a memory
3968 * failure. It will not return B_FALSE because of
3969 * lack of esp_pfkey_q if I am in wput().
3970 */
3971 sadb_pfkey_error(espstack->esp_pfkey_q, mp, ENOMEM,
3972 diagnostic, ksi->ks_in_serial);
3973 }
3974 break;
3975 case SADB_UPDATE:
3976 case SADB_X_UPDATEPAIR:
3977 /*
3978 * Find a larval, if not there, find a full one and get
3979 * strict.
3980 */
3981 error = esp_update_sa(mp, ksi, &diagnostic, espstack,
3982 samsg->sadb_msg_type);
3983 if (error != 0) {
3984 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
3985 diagnostic, ksi->ks_in_serial);
3986 }
3987 /* else esp_update_sa() took care of things. */
3988 break;
3989 case SADB_GETSPI:
3990 /*
3991 * Reserve a new larval entry.
3992 */
3993 esp_getspi(mp, ksi, espstack);
3994 break;
3995 case SADB_ACQUIRE:
3996 /*
3997 * Find larval and/or ACQUIRE record and kill it (them), I'm
3998 * most likely an error. Inbound ACQUIRE messages should only
3999 * have the base header.
4000 */
4001 sadb_in_acquire(samsg, &espstack->esp_sadb,
4002 espstack->esp_pfkey_q, espstack->ipsecesp_netstack);
4003 freemsg(mp);
4004 break;
4005 case SADB_DUMP:
4006 /*
4007 * Dump all entries.
4008 */
4009 esp_dump(mp, ksi, espstack);
4010 /* esp_dump will take care of the return message, etc. */
4011 break;
4012 case SADB_EXPIRE:
4013 /* Should never reach me. */
4014 sadb_pfkey_error(espstack->esp_pfkey_q, mp, EOPNOTSUPP,
4015 diagnostic, ksi->ks_in_serial);
4016 break;
4017 default:
4018 sadb_pfkey_error(espstack->esp_pfkey_q, mp, EINVAL,
4019 SADB_X_DIAGNOSTIC_UNKNOWN_MSG, ksi->ks_in_serial);
4020 break;
4021 }
4022 }
4023
4024 /*
4025 * Handle case where PF_KEY says it can't find a keysock for one of my
4026 * ACQUIRE messages.
4027 */
4028 static void
4029 esp_keysock_no_socket(mblk_t *mp, ipsecesp_stack_t *espstack)
4030 {
4031 sadb_msg_t *samsg;
4032 keysock_out_err_t *kse = (keysock_out_err_t *)mp->b_rptr;
4033
4034 if (mp->b_cont == NULL) {
4035 freemsg(mp);
4036 return;
4037 }
4038 samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
4039
4040 /*
4041 * If keysock can't find any registered, delete the acquire record
4042 * immediately, and handle errors.
4043 */
4044 if (samsg->sadb_msg_type == SADB_ACQUIRE) {
4045 samsg->sadb_msg_errno = kse->ks_err_errno;
4046 samsg->sadb_msg_len = SADB_8TO64(sizeof (*samsg));
4047 /*
4048 * Use the write-side of the esp_pfkey_q
4049 */
4050 sadb_in_acquire(samsg, &espstack->esp_sadb,
4051 WR(espstack->esp_pfkey_q), espstack->ipsecesp_netstack);
4052 }
4053
4054 freemsg(mp);
4055 }
4056
4057 /*
4058 * ESP module write put routine.
4059 */
4060 static void
4061 ipsecesp_wput(queue_t *q, mblk_t *mp)
4062 {
4063 ipsec_info_t *ii;
4064 struct iocblk *iocp;
4065 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)q->q_ptr;
4066
4067 esp3dbg(espstack, ("In esp_wput().\n"));
4068
4069 /* NOTE: Each case must take care of freeing or passing mp. */
4070 switch (mp->b_datap->db_type) {
4071 case M_CTL:
4072 if ((mp->b_wptr - mp->b_rptr) < sizeof (ipsec_info_t)) {
4073 /* Not big enough message. */
4074 freemsg(mp);
4075 break;
4076 }
4077 ii = (ipsec_info_t *)mp->b_rptr;
4078
4079 switch (ii->ipsec_info_type) {
4080 case KEYSOCK_OUT_ERR:
4081 esp1dbg(espstack, ("Got KEYSOCK_OUT_ERR message.\n"));
4082 esp_keysock_no_socket(mp, espstack);
4083 break;
4084 case KEYSOCK_IN:
4085 ESP_BUMP_STAT(espstack, keysock_in);
4086 esp3dbg(espstack, ("Got KEYSOCK_IN message.\n"));
4087
4088 /* Parse the message. */
4089 esp_parse_pfkey(mp, espstack);
4090 break;
4091 case KEYSOCK_HELLO:
4092 sadb_keysock_hello(&espstack->esp_pfkey_q, q, mp,
4093 esp_ager, (void *)espstack, &espstack->esp_event,
4094 SADB_SATYPE_ESP);
4095 break;
4096 default:
4097 esp2dbg(espstack, ("Got M_CTL from above of 0x%x.\n",
4098 ii->ipsec_info_type));
4099 freemsg(mp);
4100 break;
4101 }
4102 break;
4103 case M_IOCTL:
4104 iocp = (struct iocblk *)mp->b_rptr;
4105 switch (iocp->ioc_cmd) {
4106 case ND_SET:
4107 case ND_GET:
4108 if (nd_getset(q, espstack->ipsecesp_g_nd, mp)) {
4109 qreply(q, mp);
4110 return;
4111 } else {
4112 iocp->ioc_error = ENOENT;
4113 }
4114 /* FALLTHRU */
4115 default:
4116 /* We really don't support any other ioctls, do we? */
4117
4118 /* Return EINVAL */
4119 if (iocp->ioc_error != ENOENT)
4120 iocp->ioc_error = EINVAL;
4121 iocp->ioc_count = 0;
4122 mp->b_datap->db_type = M_IOCACK;
4123 qreply(q, mp);
4124 return;
4125 }
4126 default:
4127 esp3dbg(espstack,
4128 ("Got default message, type %d, passing to IP.\n",
4129 mp->b_datap->db_type));
4130 putnext(q, mp);
4131 }
4132 }
4133
4134 /*
4135 * Wrapper to allow IP to trigger an ESP association failure message
4136 * during inbound SA selection.
4137 */
4138 void
4139 ipsecesp_in_assocfailure(mblk_t *mp, char level, ushort_t sl, char *fmt,
4140 uint32_t spi, void *addr, int af, ip_recv_attr_t *ira)
4141 {
4142 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
4143 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
4144 ipsec_stack_t *ipss = ns->netstack_ipsec;
4145
4146 if (espstack->ipsecesp_log_unknown_spi) {
4147 ipsec_assocfailure(info.mi_idnum, 0, level, sl, fmt, spi,
4148 addr, af, espstack->ipsecesp_netstack);
4149 }
4150
4151 ip_drop_packet(mp, B_TRUE, ira->ira_ill,
4152 DROPPER(ipss, ipds_esp_no_sa),
4153 &espstack->esp_dropper);
4154 }
4155
4156 /*
4157 * Initialize the ESP input and output processing functions.
4158 */
4159 void
4160 ipsecesp_init_funcs(ipsa_t *sa)
4161 {
4162 if (sa->ipsa_output_func == NULL)
4163 sa->ipsa_output_func = esp_outbound;
4164 if (sa->ipsa_input_func == NULL)
4165 sa->ipsa_input_func = esp_inbound;
4166 }