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, IPPROTO_ESP);
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 uint16_t local = NEXT_16(ports);
1572 ipsa_t *outbound_peer;
1573 isaf_t *bucket;
1574 ipsecesp_stack_t *espstack = assoc->ipsa_netstack->netstack_ipsecesp;
1575
1576 /* We found a conn_t, therefore local != 0. */
1577 ASSERT(local != 0);
1578 /* Assume an IPv4 SA. */
1579 ASSERT(assoc->ipsa_addrfam == AF_INET);
1580
1581 /*
1582 * On-the-wire rport == 0 means something's very wrong.
1583 * An unpaired SA is also useless to us.
1584 * If we are behind the NAT, don't bother.
1585 * A zero local NAT port defaults to 4500, so check that too.
1586 * And, of course, if the ports already match, we don't need to
1587 * bother.
1588 */
1589 if (remote == 0 || assoc->ipsa_otherspi == 0 ||
1590 (assoc->ipsa_flags & IPSA_F_BEHIND_NAT) ||
1591 (assoc->ipsa_remote_nat_port == 0 &&
1592 remote == htons(IPPORT_IKE_NATT)) ||
1593 remote == assoc->ipsa_remote_nat_port)
1594 return;
1595
1596 /* Try and snag the peer. NOTE: Assume IPv4 for now. */
1597 bucket = OUTBOUND_BUCKET_V4(&(espstack->esp_sadb.s_v4),
1598 assoc->ipsa_srcaddr[0]);
1599 mutex_enter(&bucket->isaf_lock);
1600 outbound_peer = ipsec_getassocbyspi(bucket, assoc->ipsa_otherspi,
1601 assoc->ipsa_dstaddr, assoc->ipsa_srcaddr, AF_INET);
1602 mutex_exit(&bucket->isaf_lock);
1603
1604 /* We probably lost a race to a deleting or expiring thread. */
1605 if (outbound_peer == NULL)
1606 return;
1607
1608 /*
1609 * Hold the mutexes for both SAs so we don't race another inbound
1610 * thread. A lock-entry order shouldn't matter, since all other
1611 * per-ipsa locks are individually held-then-released.
1612 *
1613 * Luckily, this has nothing to do with the remote-NAT address,
1614 * so we don't have to re-scribble the cached-checksum differential.
1615 */
1616 mutex_enter(&outbound_peer->ipsa_lock);
1617 mutex_enter(&assoc->ipsa_lock);
1618 outbound_peer->ipsa_remote_nat_port = assoc->ipsa_remote_nat_port =
1619 remote;
1620 mutex_exit(&assoc->ipsa_lock);
1621 mutex_exit(&outbound_peer->ipsa_lock);
1622 IPSA_REFRELE(outbound_peer);
1623 ESP_BUMP_STAT(espstack, sa_port_renumbers);
1624 }
1625 /*
1626 * Finish processing of an inbound ESP packet after processing by the
1627 * crypto framework.
1628 * - Remove the ESP header.
1629 * - Send packet back to IP.
1630 * If authentication was performed on the packet, this function is called
1631 * only if the authentication succeeded.
1632 * On success returns B_TRUE, on failure returns B_FALSE and frees the
1633 * mblk chain data_mp.
1634 */
1635 static mblk_t *
1636 esp_in_done(mblk_t *data_mp, ip_recv_attr_t *ira, ipsec_crypto_t *ic)
1637 {
1638 ipsa_t *assoc;
1639 uint_t espstart;
1640 uint32_t ivlen = 0;
1641 uint_t processed_len;
1642 esph_t *esph;
1643 kstat_named_t *counter;
1644 boolean_t is_natt;
1645 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
1646 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1647 ipsec_stack_t *ipss = ns->netstack_ipsec;
1648
1649 assoc = ira->ira_ipsec_esp_sa;
1650 ASSERT(assoc != NULL);
1651
1652 is_natt = ((assoc->ipsa_flags & IPSA_F_NATT) != 0);
1653
1654 /* get the pointer to the ESP header */
1655 if (assoc->ipsa_encr_alg == SADB_EALG_NULL) {
1656 /* authentication-only ESP */
1657 espstart = ic->ic_crypto_data.cd_offset;
1658 processed_len = ic->ic_crypto_data.cd_length;
1659 } else {
1660 /* encryption present */
1661 ivlen = assoc->ipsa_iv_len;
1662 if (assoc->ipsa_auth_alg == SADB_AALG_NONE) {
1663 /* encryption-only ESP */
1664 espstart = ic->ic_crypto_data.cd_offset -
1665 sizeof (esph_t) - assoc->ipsa_iv_len;
1666 processed_len = ic->ic_crypto_data.cd_length +
1667 ivlen;
1668 } else {
1669 /* encryption with authentication */
1670 espstart = ic->ic_crypto_dual_data.dd_offset1;
1671 processed_len = ic->ic_crypto_dual_data.dd_len2 +
1672 ivlen;
1673 }
1674 }
1675
1676 esph = (esph_t *)(data_mp->b_rptr + espstart);
1677
1678 if (assoc->ipsa_auth_alg != IPSA_AALG_NONE ||
1679 (assoc->ipsa_flags & IPSA_F_COMBINED)) {
1680 /*
1681 * Authentication passed if we reach this point.
1682 * Packets with authentication will have the ICV
1683 * after the crypto data. Adjust b_wptr before
1684 * making padlen checks.
1685 */
1686 ESP_BUMP_STAT(espstack, good_auth);
1687 data_mp->b_wptr -= assoc->ipsa_mac_len;
1688
1689 /*
1690 * Check replay window here!
1691 * For right now, assume keysock will set the replay window
1692 * size to zero for SAs that have an unspecified sender.
1693 * This may change...
1694 */
1695
1696 if (!sadb_replay_check(assoc, esph->esph_replay)) {
1697 /*
1698 * Log the event. As of now we print out an event.
1699 * Do not print the replay failure number, or else
1700 * syslog cannot collate the error messages. Printing
1701 * the replay number that failed opens a denial-of-
1702 * service attack.
1703 */
1704 ipsec_assocfailure(info.mi_idnum, 0, 0,
1705 SL_ERROR | SL_WARN,
1706 "Replay failed for ESP spi 0x%x, dst %s.\n",
1707 assoc->ipsa_spi, assoc->ipsa_dstaddr,
1708 assoc->ipsa_addrfam, espstack->ipsecesp_netstack);
1709 ESP_BUMP_STAT(espstack, replay_failures);
1710 counter = DROPPER(ipss, ipds_esp_replay);
1711 goto drop_and_bail;
1712 }
1713
1714 if (is_natt) {
1715 ASSERT(ira->ira_flags & IRAF_ESP_UDP_PORTS);
1716 ASSERT(ira->ira_esp_udp_ports != 0);
1717 esp_port_freshness(ira->ira_esp_udp_ports, assoc);
1718 }
1719 }
1720
1721 esp_set_usetime(assoc, B_TRUE);
1722
1723 if (!esp_age_bytes(assoc, processed_len, B_TRUE)) {
1724 /* The ipsa has hit hard expiration, LOG and AUDIT. */
1725 ipsec_assocfailure(info.mi_idnum, 0, 0,
1726 SL_ERROR | SL_WARN,
1727 "ESP association 0x%x, dst %s had bytes expire.\n",
1728 assoc->ipsa_spi, assoc->ipsa_dstaddr, assoc->ipsa_addrfam,
1729 espstack->ipsecesp_netstack);
1730 ESP_BUMP_STAT(espstack, bytes_expired);
1731 counter = DROPPER(ipss, ipds_esp_bytes_expire);
1732 goto drop_and_bail;
1733 }
1734
1735 /*
1736 * Remove ESP header and padding from packet. I hope the compiler
1737 * spews "branch, predict taken" code for this.
1738 */
1739
1740 if (esp_strip_header(data_mp, (ira->ira_flags & IRAF_IS_IPV4),
1741 ivlen, &counter, espstack)) {
1742
1743 if (is_system_labeled() && assoc->ipsa_tsl != NULL) {
1744 if (!ip_recv_attr_replace_label(ira, assoc->ipsa_tsl)) {
1745 ip_drop_packet(data_mp, B_TRUE, ira->ira_ill,
1746 DROPPER(ipss, ipds_ah_nomem),
1747 &espstack->esp_dropper);
1748 BUMP_MIB(ira->ira_ill->ill_ip_mib,
1749 ipIfStatsInDiscards);
1750 return (NULL);
1751 }
1752 }
1753 if (is_natt)
1754 return (esp_fix_natt_checksums(data_mp, assoc));
1755
1756 if (assoc->ipsa_state == IPSA_STATE_IDLE) {
1757 /*
1758 * Cluster buffering case. Tell caller that we're
1759 * handling the packet.
1760 */
1761 sadb_buf_pkt(assoc, data_mp, ira);
1762 return (NULL);
1763 }
1764
1765 return (data_mp);
1766 }
1767
1768 esp1dbg(espstack, ("esp_in_done: esp_strip_header() failed\n"));
1769 drop_and_bail:
1770 IP_ESP_BUMP_STAT(ipss, in_discards);
1771 ip_drop_packet(data_mp, B_TRUE, ira->ira_ill, counter,
1772 &espstack->esp_dropper);
1773 BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
1774 return (NULL);
1775 }
1776
1777 /*
1778 * Called upon failing the inbound ICV check. The message passed as
1779 * argument is freed.
1780 */
1781 static void
1782 esp_log_bad_auth(mblk_t *mp, ip_recv_attr_t *ira)
1783 {
1784 ipsa_t *assoc = ira->ira_ipsec_esp_sa;
1785 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
1786 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
1787 ipsec_stack_t *ipss = ns->netstack_ipsec;
1788
1789 /*
1790 * Log the event. Don't print to the console, block
1791 * potential denial-of-service attack.
1792 */
1793 ESP_BUMP_STAT(espstack, bad_auth);
1794
1795 ipsec_assocfailure(info.mi_idnum, 0, 0, SL_ERROR | SL_WARN,
1796 "ESP Authentication failed for spi 0x%x, dst %s.\n",
1797 assoc->ipsa_spi, assoc->ipsa_dstaddr, assoc->ipsa_addrfam,
1798 espstack->ipsecesp_netstack);
1799
1800 IP_ESP_BUMP_STAT(ipss, in_discards);
1801 ip_drop_packet(mp, B_TRUE, ira->ira_ill,
1802 DROPPER(ipss, ipds_esp_bad_auth),
1803 &espstack->esp_dropper);
1804 }
1805
1806
1807 /*
1808 * Invoked for outbound packets after ESP processing. If the packet
1809 * also requires AH, performs the AH SA selection and AH processing.
1810 * Returns B_TRUE if the AH processing was not needed or if it was
1811 * performed successfully. Returns B_FALSE and consumes the passed mblk
1812 * if AH processing was required but could not be performed.
1813 *
1814 * Returns data_mp unless data_mp was consumed/queued.
1815 */
1816 static mblk_t *
1817 esp_do_outbound_ah(mblk_t *data_mp, ip_xmit_attr_t *ixa)
1818 {
1819 ipsec_action_t *ap;
1820
1821 ap = ixa->ixa_ipsec_action;
1822 if (ap == NULL) {
1823 ipsec_policy_t *pp = ixa->ixa_ipsec_policy;
1824 ap = pp->ipsp_act;
1825 }
1826
1827 if (!ap->ipa_want_ah)
1828 return (data_mp);
1829
1830 /*
1831 * Normally the AH SA would have already been put in place
1832 * but it could have been flushed so we need to look for it.
1833 */
1834 if (ixa->ixa_ipsec_ah_sa == NULL) {
1835 if (!ipsec_outbound_sa(data_mp, ixa, IPPROTO_AH)) {
1836 sadb_acquire(data_mp, ixa, B_TRUE, B_FALSE);
1837 return (NULL);
1838 }
1839 }
1840 ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
1841
1842 data_mp = ixa->ixa_ipsec_ah_sa->ipsa_output_func(data_mp, ixa);
1843 return (data_mp);
1844 }
1845
1846
1847 /*
1848 * Kernel crypto framework callback invoked after completion of async
1849 * crypto requests for outbound packets.
1850 */
1851 static void
1852 esp_kcf_callback_outbound(void *arg, int status)
1853 {
1854 mblk_t *mp = (mblk_t *)arg;
1855 mblk_t *async_mp;
1856 netstack_t *ns;
1857 ipsec_stack_t *ipss;
1858 ipsecesp_stack_t *espstack;
1859 mblk_t *data_mp;
1860 ip_xmit_attr_t ixas;
1861 ipsec_crypto_t *ic;
1862 ill_t *ill;
1863
1864 /*
1865 * First remove the ipsec_crypto_t mblk
1866 * Note that we need to ipsec_free_crypto_data(mp) once done with ic.
1867 */
1868 async_mp = ipsec_remove_crypto_data(mp, &ic);
1869 ASSERT(async_mp != NULL);
1870
1871 /*
1872 * Extract the ip_xmit_attr_t from the first mblk.
1873 * Verifies that the netstack and ill is still around; could
1874 * have vanished while kEf was doing its work.
1875 * On succesful return we have a nce_t and the ill/ipst can't
1876 * disappear until we do the nce_refrele in ixa_cleanup.
1877 */
1878 data_mp = async_mp->b_cont;
1879 async_mp->b_cont = NULL;
1880 if (!ip_xmit_attr_from_mblk(async_mp, &ixas)) {
1881 /* Disappeared on us - no ill/ipst for MIB */
1882 /* We have nowhere to do stats since ixa_ipst could be NULL */
1883 if (ixas.ixa_nce != NULL) {
1884 ill = ixas.ixa_nce->nce_ill;
1885 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1886 ip_drop_output("ipIfStatsOutDiscards", data_mp, ill);
1887 }
1888 freemsg(data_mp);
1889 goto done;
1890 }
1891 ns = ixas.ixa_ipst->ips_netstack;
1892 espstack = ns->netstack_ipsecesp;
1893 ipss = ns->netstack_ipsec;
1894 ill = ixas.ixa_nce->nce_ill;
1895
1896 if (status == CRYPTO_SUCCESS) {
1897 /*
1898 * If a ICV was computed, it was stored by the
1899 * crypto framework at the end of the packet.
1900 */
1901 ipha_t *ipha = (ipha_t *)data_mp->b_rptr;
1902
1903 esp_set_usetime(ixas.ixa_ipsec_esp_sa, B_FALSE);
1904 /* NAT-T packet. */
1905 if (IPH_HDR_VERSION(ipha) == IP_VERSION &&
1906 ipha->ipha_protocol == IPPROTO_UDP)
1907 esp_prepare_udp(ns, data_mp, ipha);
1908
1909 /* do AH processing if needed */
1910 data_mp = esp_do_outbound_ah(data_mp, &ixas);
1911 if (data_mp == NULL)
1912 goto done;
1913
1914 (void) ip_output_post_ipsec(data_mp, &ixas);
1915 } else {
1916 /* Outbound shouldn't see invalid MAC */
1917 ASSERT(status != CRYPTO_INVALID_MAC);
1918
1919 esp1dbg(espstack,
1920 ("esp_kcf_callback_outbound: crypto failed with 0x%x\n",
1921 status));
1922 ESP_BUMP_STAT(espstack, crypto_failures);
1923 ESP_BUMP_STAT(espstack, out_discards);
1924 ip_drop_packet(data_mp, B_FALSE, ill,
1925 DROPPER(ipss, ipds_esp_crypto_failed),
1926 &espstack->esp_dropper);
1927 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
1928 }
1929 done:
1930 ixa_cleanup(&ixas);
1931 (void) ipsec_free_crypto_data(mp);
1932 }
1933
1934 /*
1935 * Kernel crypto framework callback invoked after completion of async
1936 * crypto requests for inbound packets.
1937 */
1938 static void
1939 esp_kcf_callback_inbound(void *arg, int status)
1940 {
1941 mblk_t *mp = (mblk_t *)arg;
1942 mblk_t *async_mp;
1943 netstack_t *ns;
1944 ipsecesp_stack_t *espstack;
1945 ipsec_stack_t *ipss;
1946 mblk_t *data_mp;
1947 ip_recv_attr_t iras;
1948 ipsec_crypto_t *ic;
1949
1950 /*
1951 * First remove the ipsec_crypto_t mblk
1952 * Note that we need to ipsec_free_crypto_data(mp) once done with ic.
1953 */
1954 async_mp = ipsec_remove_crypto_data(mp, &ic);
1955 ASSERT(async_mp != NULL);
1956
1957 /*
1958 * Extract the ip_recv_attr_t from the first mblk.
1959 * Verifies that the netstack and ill is still around; could
1960 * have vanished while kEf was doing its work.
1961 */
1962 data_mp = async_mp->b_cont;
1963 async_mp->b_cont = NULL;
1964 if (!ip_recv_attr_from_mblk(async_mp, &iras)) {
1965 /* The ill or ip_stack_t disappeared on us */
1966 ip_drop_input("ip_recv_attr_from_mblk", data_mp, NULL);
1967 freemsg(data_mp);
1968 goto done;
1969 }
1970
1971 ns = iras.ira_ill->ill_ipst->ips_netstack;
1972 espstack = ns->netstack_ipsecesp;
1973 ipss = ns->netstack_ipsec;
1974
1975 if (status == CRYPTO_SUCCESS) {
1976 data_mp = esp_in_done(data_mp, &iras, ic);
1977 if (data_mp == NULL)
1978 goto done;
1979
1980 /* finish IPsec processing */
1981 ip_input_post_ipsec(data_mp, &iras);
1982 } else if (status == CRYPTO_INVALID_MAC) {
1983 esp_log_bad_auth(data_mp, &iras);
1984 } else {
1985 esp1dbg(espstack,
1986 ("esp_kcf_callback: crypto failed with 0x%x\n",
1987 status));
1988 ESP_BUMP_STAT(espstack, crypto_failures);
1989 IP_ESP_BUMP_STAT(ipss, in_discards);
1990 ip_drop_packet(data_mp, B_TRUE, iras.ira_ill,
1991 DROPPER(ipss, ipds_esp_crypto_failed),
1992 &espstack->esp_dropper);
1993 BUMP_MIB(iras.ira_ill->ill_ip_mib, ipIfStatsInDiscards);
1994 }
1995 done:
1996 ira_cleanup(&iras, B_TRUE);
1997 (void) ipsec_free_crypto_data(mp);
1998 }
1999
2000 /*
2001 * Invoked on crypto framework failure during inbound and outbound processing.
2002 */
2003 static void
2004 esp_crypto_failed(mblk_t *data_mp, boolean_t is_inbound, int kef_rc,
2005 ill_t *ill, ipsecesp_stack_t *espstack)
2006 {
2007 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec;
2008
2009 esp1dbg(espstack, ("crypto failed for %s ESP with 0x%x\n",
2010 is_inbound ? "inbound" : "outbound", kef_rc));
2011 ip_drop_packet(data_mp, is_inbound, ill,
2012 DROPPER(ipss, ipds_esp_crypto_failed),
2013 &espstack->esp_dropper);
2014 ESP_BUMP_STAT(espstack, crypto_failures);
2015 if (is_inbound)
2016 IP_ESP_BUMP_STAT(ipss, in_discards);
2017 else
2018 ESP_BUMP_STAT(espstack, out_discards);
2019 }
2020
2021 /*
2022 * A statement-equivalent macro, _cr MUST point to a modifiable
2023 * crypto_call_req_t.
2024 */
2025 #define ESP_INIT_CALLREQ(_cr, _mp, _callback) \
2026 (_cr)->cr_flag = CRYPTO_SKIP_REQID|CRYPTO_ALWAYS_QUEUE; \
2027 (_cr)->cr_callback_arg = (_mp); \
2028 (_cr)->cr_callback_func = (_callback)
2029
2030 #define ESP_INIT_CRYPTO_MAC(mac, icvlen, icvbuf) { \
2031 (mac)->cd_format = CRYPTO_DATA_RAW; \
2032 (mac)->cd_offset = 0; \
2033 (mac)->cd_length = icvlen; \
2034 (mac)->cd_raw.iov_base = (char *)icvbuf; \
2035 (mac)->cd_raw.iov_len = icvlen; \
2036 }
2037
2038 #define ESP_INIT_CRYPTO_DATA(data, mp, off, len) { \
2039 if (MBLKL(mp) >= (len) + (off)) { \
2040 (data)->cd_format = CRYPTO_DATA_RAW; \
2041 (data)->cd_raw.iov_base = (char *)(mp)->b_rptr; \
2042 (data)->cd_raw.iov_len = MBLKL(mp); \
2043 (data)->cd_offset = off; \
2044 } else { \
2045 (data)->cd_format = CRYPTO_DATA_MBLK; \
2046 (data)->cd_mp = mp; \
2047 (data)->cd_offset = off; \
2048 } \
2049 (data)->cd_length = len; \
2050 }
2051
2052 #define ESP_INIT_CRYPTO_DUAL_DATA(data, mp, off1, len1, off2, len2) { \
2053 (data)->dd_format = CRYPTO_DATA_MBLK; \
2054 (data)->dd_mp = mp; \
2055 (data)->dd_len1 = len1; \
2056 (data)->dd_offset1 = off1; \
2057 (data)->dd_len2 = len2; \
2058 (data)->dd_offset2 = off2; \
2059 }
2060
2061 /*
2062 * Returns data_mp if successfully completed the request. Returns
2063 * NULL if it failed (and increments InDiscards) or if it is pending.
2064 */
2065 static mblk_t *
2066 esp_submit_req_inbound(mblk_t *esp_mp, ip_recv_attr_t *ira,
2067 ipsa_t *assoc, uint_t esph_offset)
2068 {
2069 uint_t auth_offset, msg_len, auth_len;
2070 crypto_call_req_t call_req, *callrp;
2071 mblk_t *mp;
2072 esph_t *esph_ptr;
2073 int kef_rc;
2074 uint_t icv_len = assoc->ipsa_mac_len;
2075 crypto_ctx_template_t auth_ctx_tmpl;
2076 boolean_t do_auth, do_encr, force;
2077 uint_t encr_offset, encr_len;
2078 uint_t iv_len = assoc->ipsa_iv_len;
2079 crypto_ctx_template_t encr_ctx_tmpl;
2080 ipsec_crypto_t *ic, icstack;
2081 uchar_t *iv_ptr;
2082 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
2083 ipsec_stack_t *ipss = ns->netstack_ipsec;
2084 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
2085
2086 do_auth = assoc->ipsa_auth_alg != SADB_AALG_NONE;
2087 do_encr = assoc->ipsa_encr_alg != SADB_EALG_NULL;
2088 force = (assoc->ipsa_flags & IPSA_F_ASYNC);
2089
2090 #ifdef IPSEC_LATENCY_TEST
2091 kef_rc = CRYPTO_SUCCESS;
2092 #else
2093 kef_rc = CRYPTO_FAILED;
2094 #endif
2095
2096 /*
2097 * An inbound packet is of the form:
2098 * [IP,options,ESP,IV,data,ICV,pad]
2099 */
2100 esph_ptr = (esph_t *)(esp_mp->b_rptr + esph_offset);
2101 iv_ptr = (uchar_t *)(esph_ptr + 1);
2102 /* Packet length starting at IP header ending after ESP ICV. */
2103 msg_len = MBLKL(esp_mp);
2104
2105 encr_offset = esph_offset + sizeof (esph_t) + iv_len;
2106 encr_len = msg_len - encr_offset;
2107
2108 /*
2109 * Counter mode algs need a nonce. This is setup in sadb_common_add().
2110 * If for some reason we are using a SA which does not have a nonce
2111 * then we must fail here.
2112 */
2113 if ((assoc->ipsa_flags & IPSA_F_COUNTERMODE) &&
2114 (assoc->ipsa_nonce == NULL)) {
2115 ip_drop_packet(esp_mp, B_TRUE, ira->ira_ill,
2116 DROPPER(ipss, ipds_esp_nomem), &espstack->esp_dropper);
2117 return (NULL);
2118 }
2119
2120 if (force) {
2121 /* We are doing asynch; allocate mblks to hold state */
2122 if ((mp = ip_recv_attr_to_mblk(ira)) == NULL ||
2123 (mp = ipsec_add_crypto_data(mp, &ic)) == NULL) {
2124 BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
2125 ip_drop_input("ipIfStatsInDiscards", esp_mp,
2126 ira->ira_ill);
2127 return (NULL);
2128 }
2129 linkb(mp, esp_mp);
2130 callrp = &call_req;
2131 ESP_INIT_CALLREQ(callrp, mp, esp_kcf_callback_inbound);
2132 } else {
2133 /*
2134 * If we know we are going to do sync then ipsec_crypto_t
2135 * should be on the stack.
2136 */
2137 ic = &icstack;
2138 bzero(ic, sizeof (*ic));
2139 callrp = NULL;
2140 }
2141
2142 if (do_auth) {
2143 /* authentication context template */
2144 IPSEC_CTX_TMPL(assoc, ipsa_authtmpl, IPSEC_ALG_AUTH,
2145 auth_ctx_tmpl);
2146
2147 /* ICV to be verified */
2148 ESP_INIT_CRYPTO_MAC(&ic->ic_crypto_mac,
2149 icv_len, esp_mp->b_wptr - icv_len);
2150
2151 /* authentication starts at the ESP header */
2152 auth_offset = esph_offset;
2153 auth_len = msg_len - auth_offset - icv_len;
2154 if (!do_encr) {
2155 /* authentication only */
2156 /* initialize input data argument */
2157 ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
2158 esp_mp, auth_offset, auth_len);
2159
2160 /* call the crypto framework */
2161 kef_rc = crypto_mac_verify(&assoc->ipsa_amech,
2162 &ic->ic_crypto_data,
2163 &assoc->ipsa_kcfauthkey, auth_ctx_tmpl,
2164 &ic->ic_crypto_mac, callrp);
2165 }
2166 }
2167
2168 if (do_encr) {
2169 /* encryption template */
2170 IPSEC_CTX_TMPL(assoc, ipsa_encrtmpl, IPSEC_ALG_ENCR,
2171 encr_ctx_tmpl);
2172
2173 /* Call the nonce update function. Also passes in IV */
2174 (assoc->ipsa_noncefunc)(assoc, (uchar_t *)esph_ptr, encr_len,
2175 iv_ptr, &ic->ic_cmm, &ic->ic_crypto_data);
2176
2177 if (!do_auth) {
2178 /* decryption only */
2179 /* initialize input data argument */
2180 ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
2181 esp_mp, encr_offset, encr_len);
2182
2183 /* call the crypto framework */
2184 kef_rc = crypto_decrypt((crypto_mechanism_t *)
2185 &ic->ic_cmm, &ic->ic_crypto_data,
2186 &assoc->ipsa_kcfencrkey, encr_ctx_tmpl,
2187 NULL, callrp);
2188 }
2189 }
2190
2191 if (do_auth && do_encr) {
2192 /* dual operation */
2193 /* initialize input data argument */
2194 ESP_INIT_CRYPTO_DUAL_DATA(&ic->ic_crypto_dual_data,
2195 esp_mp, auth_offset, auth_len,
2196 encr_offset, encr_len - icv_len);
2197
2198 /* specify IV */
2199 ic->ic_crypto_dual_data.dd_miscdata = (char *)iv_ptr;
2200
2201 /* call the framework */
2202 kef_rc = crypto_mac_verify_decrypt(&assoc->ipsa_amech,
2203 &assoc->ipsa_emech, &ic->ic_crypto_dual_data,
2204 &assoc->ipsa_kcfauthkey, &assoc->ipsa_kcfencrkey,
2205 auth_ctx_tmpl, encr_ctx_tmpl, &ic->ic_crypto_mac,
2206 NULL, callrp);
2207 }
2208
2209 switch (kef_rc) {
2210 case CRYPTO_SUCCESS:
2211 ESP_BUMP_STAT(espstack, crypto_sync);
2212 esp_mp = esp_in_done(esp_mp, ira, ic);
2213 if (force) {
2214 /* Free mp after we are done with ic */
2215 mp = ipsec_free_crypto_data(mp);
2216 (void) ip_recv_attr_free_mblk(mp);
2217 }
2218 return (esp_mp);
2219 case CRYPTO_QUEUED:
2220 /* esp_kcf_callback_inbound() will be invoked on completion */
2221 ESP_BUMP_STAT(espstack, crypto_async);
2222 return (NULL);
2223 case CRYPTO_INVALID_MAC:
2224 if (force) {
2225 mp = ipsec_free_crypto_data(mp);
2226 esp_mp = ip_recv_attr_free_mblk(mp);
2227 }
2228 ESP_BUMP_STAT(espstack, crypto_sync);
2229 BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
2230 esp_log_bad_auth(esp_mp, ira);
2231 /* esp_mp was passed to ip_drop_packet */
2232 return (NULL);
2233 }
2234
2235 if (force) {
2236 mp = ipsec_free_crypto_data(mp);
2237 esp_mp = ip_recv_attr_free_mblk(mp);
2238 }
2239 BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
2240 esp_crypto_failed(esp_mp, B_TRUE, kef_rc, ira->ira_ill, espstack);
2241 /* esp_mp was passed to ip_drop_packet */
2242 return (NULL);
2243 }
2244
2245 /*
2246 * Compute the IP and UDP checksums -- common code for both keepalives and
2247 * actual ESP-in-UDP packets. Be flexible with multiple mblks because ESP
2248 * uses mblk-insertion to insert the UDP header.
2249 * TODO - If there is an easy way to prep a packet for HW checksums, make
2250 * it happen here.
2251 * Note that this is used before both before calling ip_output_simple and
2252 * in the esp datapath. The former could use IXAF_SET_ULP_CKSUM but not the
2253 * latter.
2254 */
2255 static void
2256 esp_prepare_udp(netstack_t *ns, mblk_t *mp, ipha_t *ipha)
2257 {
2258 int offset;
2259 uint32_t cksum;
2260 uint16_t *arr;
2261 mblk_t *udpmp = mp;
2262 uint_t hlen = IPH_HDR_LENGTH(ipha);
2263
2264 ASSERT(MBLKL(mp) >= sizeof (ipha_t));
2265
2266 ipha->ipha_hdr_checksum = 0;
2267 ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
2268
2269 if (ns->netstack_udp->us_do_checksum) {
2270 ASSERT(MBLKL(udpmp) >= sizeof (udpha_t));
2271 /* arr points to the IP header. */
2272 arr = (uint16_t *)ipha;
2273 IP_STAT(ns->netstack_ip, ip_out_sw_cksum);
2274 IP_STAT_UPDATE(ns->netstack_ip, ip_out_sw_cksum_bytes,
2275 ntohs(htons(ipha->ipha_length) - hlen));
2276 /* arr[6-9] are the IP addresses. */
2277 cksum = IP_UDP_CSUM_COMP + arr[6] + arr[7] + arr[8] + arr[9] +
2278 ntohs(htons(ipha->ipha_length) - hlen);
2279 cksum = IP_CSUM(mp, hlen, cksum);
2280 offset = hlen + UDP_CHECKSUM_OFFSET;
2281 while (offset >= MBLKL(udpmp)) {
2282 offset -= MBLKL(udpmp);
2283 udpmp = udpmp->b_cont;
2284 }
2285 /* arr points to the UDP header's checksum field. */
2286 arr = (uint16_t *)(udpmp->b_rptr + offset);
2287 *arr = cksum;
2288 }
2289 }
2290
2291 /*
2292 * taskq handler so we can send the NAT-T keepalive on a separate thread.
2293 */
2294 static void
2295 actually_send_keepalive(void *arg)
2296 {
2297 mblk_t *mp = (mblk_t *)arg;
2298 ip_xmit_attr_t ixas;
2299 netstack_t *ns;
2300 netstackid_t stackid;
2301
2302 stackid = (netstackid_t)(uintptr_t)mp->b_prev;
2303 mp->b_prev = NULL;
2304 ns = netstack_find_by_stackid(stackid);
2305 if (ns == NULL) {
2306 /* Disappeared */
2307 ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
2308 freemsg(mp);
2309 return;
2310 }
2311
2312 bzero(&ixas, sizeof (ixas));
2313 ixas.ixa_zoneid = ALL_ZONES;
2314 ixas.ixa_cred = kcred;
2315 ixas.ixa_cpid = NOPID;
2316 ixas.ixa_tsl = NULL;
2317 ixas.ixa_ipst = ns->netstack_ip;
2318 /* No ULP checksum; done by esp_prepare_udp */
2319 ixas.ixa_flags = (IXAF_IS_IPV4 | IXAF_NO_IPSEC | IXAF_VERIFY_SOURCE);
2320
2321 (void) ip_output_simple(mp, &ixas);
2322 ixa_cleanup(&ixas);
2323 netstack_rele(ns);
2324 }
2325
2326 /*
2327 * Send a one-byte UDP NAT-T keepalive.
2328 */
2329 void
2330 ipsecesp_send_keepalive(ipsa_t *assoc)
2331 {
2332 mblk_t *mp;
2333 ipha_t *ipha;
2334 udpha_t *udpha;
2335 netstack_t *ns = assoc->ipsa_netstack;
2336
2337 ASSERT(MUTEX_NOT_HELD(&assoc->ipsa_lock));
2338
2339 mp = allocb(sizeof (ipha_t) + sizeof (udpha_t) + 1, BPRI_HI);
2340 if (mp == NULL)
2341 return;
2342 ipha = (ipha_t *)mp->b_rptr;
2343 ipha->ipha_version_and_hdr_length = IP_SIMPLE_HDR_VERSION;
2344 ipha->ipha_type_of_service = 0;
2345 ipha->ipha_length = htons(sizeof (ipha_t) + sizeof (udpha_t) + 1);
2346 /* Use the low-16 of the SPI so we have some clue where it came from. */
2347 ipha->ipha_ident = *(((uint16_t *)(&assoc->ipsa_spi)) + 1);
2348 ipha->ipha_fragment_offset_and_flags = 0; /* Too small to fragment! */
2349 ipha->ipha_ttl = 0xFF;
2350 ipha->ipha_protocol = IPPROTO_UDP;
2351 ipha->ipha_hdr_checksum = 0;
2352 ipha->ipha_src = assoc->ipsa_srcaddr[0];
2353 ipha->ipha_dst = assoc->ipsa_dstaddr[0];
2354 udpha = (udpha_t *)(ipha + 1);
2355 udpha->uha_src_port = (assoc->ipsa_local_nat_port != 0) ?
2356 assoc->ipsa_local_nat_port : htons(IPPORT_IKE_NATT);
2357 udpha->uha_dst_port = (assoc->ipsa_remote_nat_port != 0) ?
2358 assoc->ipsa_remote_nat_port : htons(IPPORT_IKE_NATT);
2359 udpha->uha_length = htons(sizeof (udpha_t) + 1);
2360 udpha->uha_checksum = 0;
2361 mp->b_wptr = (uint8_t *)(udpha + 1);
2362 *(mp->b_wptr++) = 0xFF;
2363
2364 esp_prepare_udp(ns, mp, ipha);
2365
2366 /*
2367 * We're holding an isaf_t bucket lock, so pawn off the actual
2368 * packet transmission to another thread. Just in case syncq
2369 * processing causes a same-bucket packet to be processed.
2370 */
2371 mp->b_prev = (mblk_t *)(uintptr_t)ns->netstack_stackid;
2372
2373 if (taskq_dispatch(esp_taskq, actually_send_keepalive, mp,
2374 TQ_NOSLEEP) == 0) {
2375 /* Assume no memory if taskq_dispatch() fails. */
2376 mp->b_prev = NULL;
2377 ip_drop_packet(mp, B_FALSE, NULL,
2378 DROPPER(ns->netstack_ipsec, ipds_esp_nomem),
2379 &ns->netstack_ipsecesp->esp_dropper);
2380 }
2381 }
2382
2383 /*
2384 * Returns mp if successfully completed the request. Returns
2385 * NULL if it failed (and increments InDiscards) or if it is pending.
2386 */
2387 static mblk_t *
2388 esp_submit_req_outbound(mblk_t *data_mp, ip_xmit_attr_t *ixa, ipsa_t *assoc,
2389 uchar_t *icv_buf, uint_t payload_len)
2390 {
2391 uint_t auth_len;
2392 crypto_call_req_t call_req, *callrp;
2393 mblk_t *esp_mp;
2394 esph_t *esph_ptr;
2395 mblk_t *mp;
2396 int kef_rc = CRYPTO_FAILED;
2397 uint_t icv_len = assoc->ipsa_mac_len;
2398 crypto_ctx_template_t auth_ctx_tmpl;
2399 boolean_t do_auth, do_encr, force;
2400 uint_t iv_len = assoc->ipsa_iv_len;
2401 crypto_ctx_template_t encr_ctx_tmpl;
2402 boolean_t is_natt = ((assoc->ipsa_flags & IPSA_F_NATT) != 0);
2403 size_t esph_offset = (is_natt ? UDPH_SIZE : 0);
2404 netstack_t *ns = ixa->ixa_ipst->ips_netstack;
2405 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
2406 ipsec_crypto_t *ic, icstack;
2407 uchar_t *iv_ptr;
2408 crypto_data_t *cd_ptr = NULL;
2409 ill_t *ill = ixa->ixa_nce->nce_ill;
2410 ipsec_stack_t *ipss = ns->netstack_ipsec;
2411
2412 esp3dbg(espstack, ("esp_submit_req_outbound:%s",
2413 is_natt ? "natt" : "not natt"));
2414
2415 do_encr = assoc->ipsa_encr_alg != SADB_EALG_NULL;
2416 do_auth = assoc->ipsa_auth_alg != SADB_AALG_NONE;
2417 force = (assoc->ipsa_flags & IPSA_F_ASYNC);
2418
2419 #ifdef IPSEC_LATENCY_TEST
2420 kef_rc = CRYPTO_SUCCESS;
2421 #else
2422 kef_rc = CRYPTO_FAILED;
2423 #endif
2424
2425 /*
2426 * Outbound IPsec packets are of the form:
2427 * [IP,options] -> [ESP,IV] -> [data] -> [pad,ICV]
2428 * unless it's NATT, then it's
2429 * [IP,options] -> [udp][ESP,IV] -> [data] -> [pad,ICV]
2430 * Get a pointer to the mblk containing the ESP header.
2431 */
2432 ASSERT(data_mp->b_cont != NULL);
2433 esp_mp = data_mp->b_cont;
2434 esph_ptr = (esph_t *)(esp_mp->b_rptr + esph_offset);
2435 iv_ptr = (uchar_t *)(esph_ptr + 1);
2436
2437 /*
2438 * Combined mode algs need a nonce. This is setup in sadb_common_add().
2439 * If for some reason we are using a SA which does not have a nonce
2440 * then we must fail here.
2441 */
2442 if ((assoc->ipsa_flags & IPSA_F_COUNTERMODE) &&
2443 (assoc->ipsa_nonce == NULL)) {
2444 ip_drop_packet(data_mp, B_FALSE, NULL,
2445 DROPPER(ipss, ipds_esp_nomem), &espstack->esp_dropper);
2446 return (NULL);
2447 }
2448
2449 if (force) {
2450 /* We are doing asynch; allocate mblks to hold state */
2451 if ((mp = ip_xmit_attr_to_mblk(ixa)) == NULL ||
2452 (mp = ipsec_add_crypto_data(mp, &ic)) == NULL) {
2453 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2454 ip_drop_output("ipIfStatsOutDiscards", data_mp, ill);
2455 freemsg(data_mp);
2456 return (NULL);
2457 }
2458
2459 linkb(mp, data_mp);
2460 callrp = &call_req;
2461 ESP_INIT_CALLREQ(callrp, mp, esp_kcf_callback_outbound);
2462 } else {
2463 /*
2464 * If we know we are going to do sync then ipsec_crypto_t
2465 * should be on the stack.
2466 */
2467 ic = &icstack;
2468 bzero(ic, sizeof (*ic));
2469 callrp = NULL;
2470 }
2471
2472
2473 if (do_auth) {
2474 /* authentication context template */
2475 IPSEC_CTX_TMPL(assoc, ipsa_authtmpl, IPSEC_ALG_AUTH,
2476 auth_ctx_tmpl);
2477
2478 /* where to store the computed mac */
2479 ESP_INIT_CRYPTO_MAC(&ic->ic_crypto_mac,
2480 icv_len, icv_buf);
2481
2482 /* authentication starts at the ESP header */
2483 auth_len = payload_len + iv_len + sizeof (esph_t);
2484 if (!do_encr) {
2485 /* authentication only */
2486 /* initialize input data argument */
2487 ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
2488 esp_mp, esph_offset, auth_len);
2489
2490 /* call the crypto framework */
2491 kef_rc = crypto_mac(&assoc->ipsa_amech,
2492 &ic->ic_crypto_data,
2493 &assoc->ipsa_kcfauthkey, auth_ctx_tmpl,
2494 &ic->ic_crypto_mac, callrp);
2495 }
2496 }
2497
2498 if (do_encr) {
2499 /* encryption context template */
2500 IPSEC_CTX_TMPL(assoc, ipsa_encrtmpl, IPSEC_ALG_ENCR,
2501 encr_ctx_tmpl);
2502 /* Call the nonce update function. */
2503 (assoc->ipsa_noncefunc)(assoc, (uchar_t *)esph_ptr, payload_len,
2504 iv_ptr, &ic->ic_cmm, &ic->ic_crypto_data);
2505
2506 if (!do_auth) {
2507 /* encryption only, skip mblk that contains ESP hdr */
2508 /* initialize input data argument */
2509 ESP_INIT_CRYPTO_DATA(&ic->ic_crypto_data,
2510 esp_mp->b_cont, 0, payload_len);
2511
2512 /*
2513 * For combined mode ciphers, the ciphertext is the same
2514 * size as the clear text, the ICV should follow the
2515 * ciphertext. To convince the kcf to allow in-line
2516 * encryption, with an ICV, use ipsec_out_crypto_mac
2517 * to point to the same buffer as the data. The calling
2518 * function need to ensure the buffer is large enough to
2519 * include the ICV.
2520 *
2521 * The IV is already written to the packet buffer, the
2522 * nonce setup function copied it to the params struct
2523 * for the cipher to use.
2524 */
2525 if (assoc->ipsa_flags & IPSA_F_COMBINED) {
2526 bcopy(&ic->ic_crypto_data,
2527 &ic->ic_crypto_mac,
2528 sizeof (crypto_data_t));
2529 ic->ic_crypto_mac.cd_length =
2530 payload_len + icv_len;
2531 cd_ptr = &ic->ic_crypto_mac;
2532 }
2533
2534 /* call the crypto framework */
2535 kef_rc = crypto_encrypt((crypto_mechanism_t *)
2536 &ic->ic_cmm, &ic->ic_crypto_data,
2537 &assoc->ipsa_kcfencrkey, encr_ctx_tmpl,
2538 cd_ptr, callrp);
2539
2540 }
2541 }
2542
2543 if (do_auth && do_encr) {
2544 /*
2545 * Encryption and authentication:
2546 * Pass the pointer to the mblk chain starting at the ESP
2547 * header to the framework. Skip the ESP header mblk
2548 * for encryption, which is reflected by an encryption
2549 * offset equal to the length of that mblk. Start
2550 * the authentication at the ESP header, i.e. use an
2551 * authentication offset of zero.
2552 */
2553 ESP_INIT_CRYPTO_DUAL_DATA(&ic->ic_crypto_dual_data,
2554 esp_mp, MBLKL(esp_mp), payload_len, esph_offset, auth_len);
2555
2556 /* specify IV */
2557 ic->ic_crypto_dual_data.dd_miscdata = (char *)iv_ptr;
2558
2559 /* call the framework */
2560 kef_rc = crypto_encrypt_mac(&assoc->ipsa_emech,
2561 &assoc->ipsa_amech, NULL,
2562 &assoc->ipsa_kcfencrkey, &assoc->ipsa_kcfauthkey,
2563 encr_ctx_tmpl, auth_ctx_tmpl,
2564 &ic->ic_crypto_dual_data,
2565 &ic->ic_crypto_mac, callrp);
2566 }
2567
2568 switch (kef_rc) {
2569 case CRYPTO_SUCCESS:
2570 ESP_BUMP_STAT(espstack, crypto_sync);
2571 esp_set_usetime(assoc, B_FALSE);
2572 if (force) {
2573 mp = ipsec_free_crypto_data(mp);
2574 data_mp = ip_xmit_attr_free_mblk(mp);
2575 }
2576 if (is_natt)
2577 esp_prepare_udp(ns, data_mp, (ipha_t *)data_mp->b_rptr);
2578 return (data_mp);
2579 case CRYPTO_QUEUED:
2580 /* esp_kcf_callback_outbound() will be invoked on completion */
2581 ESP_BUMP_STAT(espstack, crypto_async);
2582 return (NULL);
2583 }
2584
2585 if (force) {
2586 mp = ipsec_free_crypto_data(mp);
2587 data_mp = ip_xmit_attr_free_mblk(mp);
2588 }
2589 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2590 esp_crypto_failed(data_mp, B_FALSE, kef_rc, NULL, espstack);
2591 /* data_mp was passed to ip_drop_packet */
2592 return (NULL);
2593 }
2594
2595 /*
2596 * Handle outbound IPsec processing for IPv4 and IPv6
2597 *
2598 * Returns data_mp if successfully completed the request. Returns
2599 * NULL if it failed (and increments InDiscards) or if it is pending.
2600 */
2601 static mblk_t *
2602 esp_outbound(mblk_t *data_mp, ip_xmit_attr_t *ixa)
2603 {
2604 mblk_t *espmp, *tailmp;
2605 ipha_t *ipha;
2606 ip6_t *ip6h;
2607 esph_t *esph_ptr, *iv_ptr;
2608 uint_t af;
2609 uint8_t *nhp;
2610 uintptr_t divpoint, datalen, adj, padlen, i, alloclen;
2611 uintptr_t esplen = sizeof (esph_t);
2612 uint8_t protocol;
2613 ipsa_t *assoc;
2614 uint_t iv_len, block_size, mac_len = 0;
2615 uchar_t *icv_buf;
2616 udpha_t *udpha;
2617 boolean_t is_natt = B_FALSE;
2618 netstack_t *ns = ixa->ixa_ipst->ips_netstack;
2619 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
2620 ipsec_stack_t *ipss = ns->netstack_ipsec;
2621 ill_t *ill = ixa->ixa_nce->nce_ill;
2622 boolean_t need_refrele = B_FALSE;
2623
2624 ESP_BUMP_STAT(espstack, out_requests);
2625
2626 /*
2627 * <sigh> We have to copy the message here, because TCP (for example)
2628 * keeps a dupb() of the message lying around for retransmission.
2629 * Since ESP changes the whole of the datagram, we have to create our
2630 * own copy lest we clobber TCP's data. Since we have to copy anyway,
2631 * we might as well make use of msgpullup() and get the mblk into one
2632 * contiguous piece!
2633 */
2634 tailmp = msgpullup(data_mp, -1);
2635 if (tailmp == NULL) {
2636 esp0dbg(("esp_outbound: msgpullup() failed, "
2637 "dropping packet.\n"));
2638 ip_drop_packet(data_mp, B_FALSE, ill,
2639 DROPPER(ipss, ipds_esp_nomem),
2640 &espstack->esp_dropper);
2641 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2642 return (NULL);
2643 }
2644 freemsg(data_mp);
2645 data_mp = tailmp;
2646
2647 assoc = ixa->ixa_ipsec_esp_sa;
2648 ASSERT(assoc != NULL);
2649
2650 /*
2651 * Get the outer IP header in shape to escape this system..
2652 */
2653 if (is_system_labeled() && (assoc->ipsa_otsl != NULL)) {
2654 /*
2655 * Need to update packet with any CIPSO option and update
2656 * ixa_tsl to capture the new label.
2657 * We allocate a separate ixa for that purpose.
2658 */
2659 ixa = ip_xmit_attr_duplicate(ixa);
2660 if (ixa == NULL) {
2661 ip_drop_packet(data_mp, B_FALSE, ill,
2662 DROPPER(ipss, ipds_esp_nomem),
2663 &espstack->esp_dropper);
2664 return (NULL);
2665 }
2666 need_refrele = B_TRUE;
2667
2668 label_hold(assoc->ipsa_otsl);
2669 ip_xmit_attr_replace_tsl(ixa, assoc->ipsa_otsl);
2670
2671 data_mp = sadb_whack_label(data_mp, assoc, ixa,
2672 DROPPER(ipss, ipds_esp_nomem), &espstack->esp_dropper);
2673 if (data_mp == NULL) {
2674 /* Packet dropped by sadb_whack_label */
2675 ixa_refrele(ixa);
2676 return (NULL);
2677 }
2678 }
2679
2680 /*
2681 * Reality check....
2682 */
2683 ipha = (ipha_t *)data_mp->b_rptr; /* So we can call esp_acquire(). */
2684
2685 if (ixa->ixa_flags & IXAF_IS_IPV4) {
2686 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
2687
2688 af = AF_INET;
2689 divpoint = IPH_HDR_LENGTH(ipha);
2690 datalen = ntohs(ipha->ipha_length) - divpoint;
2691 nhp = (uint8_t *)&ipha->ipha_protocol;
2692 } else {
2693 ip_pkt_t ipp;
2694
2695 ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
2696
2697 af = AF_INET6;
2698 ip6h = (ip6_t *)ipha;
2699 bzero(&ipp, sizeof (ipp));
2700 divpoint = ip_find_hdr_v6(data_mp, ip6h, B_FALSE, &ipp, NULL);
2701 if (ipp.ipp_dstopts != NULL &&
2702 ipp.ipp_dstopts->ip6d_nxt != IPPROTO_ROUTING) {
2703 /*
2704 * Destination options are tricky. If we get in here,
2705 * then we have a terminal header following the
2706 * destination options. We need to adjust backwards
2707 * so we insert ESP BEFORE the destination options
2708 * bag. (So that the dstopts get encrypted!)
2709 *
2710 * Since this is for outbound packets only, we know
2711 * that non-terminal destination options only precede
2712 * routing headers.
2713 */
2714 divpoint -= ipp.ipp_dstoptslen;
2715 }
2716 datalen = ntohs(ip6h->ip6_plen) + sizeof (ip6_t) - divpoint;
2717
2718 if (ipp.ipp_rthdr != NULL) {
2719 nhp = &ipp.ipp_rthdr->ip6r_nxt;
2720 } else if (ipp.ipp_hopopts != NULL) {
2721 nhp = &ipp.ipp_hopopts->ip6h_nxt;
2722 } else {
2723 ASSERT(divpoint == sizeof (ip6_t));
2724 /* It's probably IP + ESP. */
2725 nhp = &ip6h->ip6_nxt;
2726 }
2727 }
2728
2729 mac_len = assoc->ipsa_mac_len;
2730
2731 if (assoc->ipsa_flags & IPSA_F_NATT) {
2732 /* wedge in UDP header */
2733 is_natt = B_TRUE;
2734 esplen += UDPH_SIZE;
2735 }
2736
2737 /*
2738 * Set up ESP header and encryption padding for ENCR PI request.
2739 */
2740
2741 /* Determine the padding length. Pad to 4-bytes for no-encryption. */
2742 if (assoc->ipsa_encr_alg != SADB_EALG_NULL) {
2743 iv_len = assoc->ipsa_iv_len;
2744 block_size = assoc->ipsa_datalen;
2745
2746 /*
2747 * Pad the data to the length of the cipher block size.
2748 * Include the two additional bytes (hence the - 2) for the
2749 * padding length and the next header. Take this into account
2750 * when calculating the actual length of the padding.
2751 */
2752 ASSERT(ISP2(iv_len));
2753 padlen = ((unsigned)(block_size - datalen - 2)) &
2754 (block_size - 1);
2755 } else {
2756 iv_len = 0;
2757 padlen = ((unsigned)(sizeof (uint32_t) - datalen - 2)) &
2758 (sizeof (uint32_t) - 1);
2759 }
2760
2761 /* Allocate ESP header and IV. */
2762 esplen += iv_len;
2763
2764 /*
2765 * Update association byte-count lifetimes. Don't forget to take
2766 * into account the padding length and next-header (hence the + 2).
2767 *
2768 * Use the amount of data fed into the "encryption algorithm". This
2769 * is the IV, the data length, the padding length, and the final two
2770 * bytes (padlen, and next-header).
2771 *
2772 */
2773
2774 if (!esp_age_bytes(assoc, datalen + padlen + iv_len + 2, B_FALSE)) {
2775 ip_drop_packet(data_mp, B_FALSE, ill,
2776 DROPPER(ipss, ipds_esp_bytes_expire),
2777 &espstack->esp_dropper);
2778 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2779 if (need_refrele)
2780 ixa_refrele(ixa);
2781 return (NULL);
2782 }
2783
2784 espmp = allocb(esplen, BPRI_HI);
2785 if (espmp == NULL) {
2786 ESP_BUMP_STAT(espstack, out_discards);
2787 esp1dbg(espstack, ("esp_outbound: can't allocate espmp.\n"));
2788 ip_drop_packet(data_mp, B_FALSE, ill,
2789 DROPPER(ipss, ipds_esp_nomem),
2790 &espstack->esp_dropper);
2791 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2792 if (need_refrele)
2793 ixa_refrele(ixa);
2794 return (NULL);
2795 }
2796 espmp->b_wptr += esplen;
2797 esph_ptr = (esph_t *)espmp->b_rptr;
2798
2799 if (is_natt) {
2800 esp3dbg(espstack, ("esp_outbound: NATT"));
2801
2802 udpha = (udpha_t *)espmp->b_rptr;
2803 udpha->uha_src_port = (assoc->ipsa_local_nat_port != 0) ?
2804 assoc->ipsa_local_nat_port : htons(IPPORT_IKE_NATT);
2805 udpha->uha_dst_port = (assoc->ipsa_remote_nat_port != 0) ?
2806 assoc->ipsa_remote_nat_port : htons(IPPORT_IKE_NATT);
2807 /*
2808 * Set the checksum to 0, so that the esp_prepare_udp() call
2809 * can do the right thing.
2810 */
2811 udpha->uha_checksum = 0;
2812 esph_ptr = (esph_t *)(udpha + 1);
2813 }
2814
2815 esph_ptr->esph_spi = assoc->ipsa_spi;
2816
2817 esph_ptr->esph_replay = htonl(atomic_inc_32_nv(&assoc->ipsa_replay));
2818 if (esph_ptr->esph_replay == 0 && assoc->ipsa_replay_wsize != 0) {
2819 /*
2820 * XXX We have replay counter wrapping.
2821 * We probably want to nuke this SA (and its peer).
2822 */
2823 ipsec_assocfailure(info.mi_idnum, 0, 0,
2824 SL_ERROR | SL_CONSOLE | SL_WARN,
2825 "Outbound ESP SA (0x%x, %s) has wrapped sequence.\n",
2826 esph_ptr->esph_spi, assoc->ipsa_dstaddr, af,
2827 espstack->ipsecesp_netstack);
2828
2829 ESP_BUMP_STAT(espstack, out_discards);
2830 sadb_replay_delete(assoc);
2831 ip_drop_packet(data_mp, B_FALSE, ill,
2832 DROPPER(ipss, ipds_esp_replay),
2833 &espstack->esp_dropper);
2834 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2835 if (need_refrele)
2836 ixa_refrele(ixa);
2837 return (NULL);
2838 }
2839
2840 iv_ptr = (esph_ptr + 1);
2841 /*
2842 * iv_ptr points to the mblk which will contain the IV once we have
2843 * written it there. This mblk will be part of a mblk chain that
2844 * will make up the packet.
2845 *
2846 * For counter mode algorithms, the IV is a 64 bit quantity, it
2847 * must NEVER repeat in the lifetime of the SA, otherwise an
2848 * attacker who had recorded enough packets might be able to
2849 * determine some clear text.
2850 *
2851 * To ensure this does not happen, the IV is stored in the SA and
2852 * incremented for each packet, the IV is then copied into the
2853 * "packet" for transmission to the receiving system. The IV will
2854 * also be copied into the nonce, when the packet is encrypted.
2855 *
2856 * CBC mode algorithms use a random IV for each packet. We do not
2857 * require the highest quality random bits, but for best security
2858 * with CBC mode ciphers, the value must be unlikely to repeat and
2859 * must not be known in advance to an adversary capable of influencing
2860 * the clear text.
2861 */
2862 if (!update_iv((uint8_t *)iv_ptr, espstack->esp_pfkey_q, assoc,
2863 espstack)) {
2864 ip_drop_packet(data_mp, B_FALSE, ill,
2865 DROPPER(ipss, ipds_esp_iv_wrap), &espstack->esp_dropper);
2866 if (need_refrele)
2867 ixa_refrele(ixa);
2868 return (NULL);
2869 }
2870
2871 /* Fix the IP header. */
2872 alloclen = padlen + 2 + mac_len;
2873 adj = alloclen + (espmp->b_wptr - espmp->b_rptr);
2874
2875 protocol = *nhp;
2876
2877 if (ixa->ixa_flags & IXAF_IS_IPV4) {
2878 ipha->ipha_length = htons(ntohs(ipha->ipha_length) + adj);
2879 if (is_natt) {
2880 *nhp = IPPROTO_UDP;
2881 udpha->uha_length = htons(ntohs(ipha->ipha_length) -
2882 IPH_HDR_LENGTH(ipha));
2883 } else {
2884 *nhp = IPPROTO_ESP;
2885 }
2886 ipha->ipha_hdr_checksum = 0;
2887 ipha->ipha_hdr_checksum = (uint16_t)ip_csum_hdr(ipha);
2888 } else {
2889 ip6h->ip6_plen = htons(ntohs(ip6h->ip6_plen) + adj);
2890 *nhp = IPPROTO_ESP;
2891 }
2892
2893 /* I've got the two ESP mblks, now insert them. */
2894
2895 esp2dbg(espstack, ("data_mp before outbound ESP adjustment:\n"));
2896 esp2dbg(espstack, (dump_msg(data_mp)));
2897
2898 if (!esp_insert_esp(data_mp, espmp, divpoint, espstack)) {
2899 ESP_BUMP_STAT(espstack, out_discards);
2900 /* NOTE: esp_insert_esp() only fails if there's no memory. */
2901 ip_drop_packet(data_mp, B_FALSE, ill,
2902 DROPPER(ipss, ipds_esp_nomem),
2903 &espstack->esp_dropper);
2904 freeb(espmp);
2905 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2906 if (need_refrele)
2907 ixa_refrele(ixa);
2908 return (NULL);
2909 }
2910
2911 /* Append padding (and leave room for ICV). */
2912 for (tailmp = data_mp; tailmp->b_cont != NULL; tailmp = tailmp->b_cont)
2913 ;
2914 if (tailmp->b_wptr + alloclen > tailmp->b_datap->db_lim) {
2915 tailmp->b_cont = allocb(alloclen, BPRI_HI);
2916 if (tailmp->b_cont == NULL) {
2917 ESP_BUMP_STAT(espstack, out_discards);
2918 esp0dbg(("esp_outbound: Can't allocate tailmp.\n"));
2919 ip_drop_packet(data_mp, B_FALSE, ill,
2920 DROPPER(ipss, ipds_esp_nomem),
2921 &espstack->esp_dropper);
2922 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
2923 if (need_refrele)
2924 ixa_refrele(ixa);
2925 return (NULL);
2926 }
2927 tailmp = tailmp->b_cont;
2928 }
2929
2930 /*
2931 * If there's padding, N bytes of padding must be of the form 0x1,
2932 * 0x2, 0x3... 0xN.
2933 */
2934 for (i = 0; i < padlen; ) {
2935 i++;
2936 *tailmp->b_wptr++ = i;
2937 }
2938 *tailmp->b_wptr++ = i;
2939 *tailmp->b_wptr++ = protocol;
2940
2941 esp2dbg(espstack, ("data_Mp before encryption:\n"));
2942 esp2dbg(espstack, (dump_msg(data_mp)));
2943
2944 /*
2945 * Okay. I've set up the pre-encryption ESP. Let's do it!
2946 */
2947
2948 if (mac_len > 0) {
2949 ASSERT(tailmp->b_wptr + mac_len <= tailmp->b_datap->db_lim);
2950 icv_buf = tailmp->b_wptr;
2951 tailmp->b_wptr += mac_len;
2952 } else {
2953 icv_buf = NULL;
2954 }
2955
2956 data_mp = esp_submit_req_outbound(data_mp, ixa, assoc, icv_buf,
2957 datalen + padlen + 2);
2958 if (need_refrele)
2959 ixa_refrele(ixa);
2960 return (data_mp);
2961 }
2962
2963 /*
2964 * IP calls this to validate the ICMP errors that
2965 * we got from the network.
2966 */
2967 mblk_t *
2968 ipsecesp_icmp_error(mblk_t *data_mp, ip_recv_attr_t *ira)
2969 {
2970 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
2971 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
2972 ipsec_stack_t *ipss = ns->netstack_ipsec;
2973
2974 /*
2975 * Unless we get an entire packet back, this function is useless.
2976 * Why?
2977 *
2978 * 1.) Partial packets are useless, because the "next header"
2979 * is at the end of the decrypted ESP packet. Without the
2980 * whole packet, this is useless.
2981 *
2982 * 2.) If we every use a stateful cipher, such as a stream or a
2983 * one-time pad, we can't do anything.
2984 *
2985 * Since the chances of us getting an entire packet back are very
2986 * very small, we discard here.
2987 */
2988 IP_ESP_BUMP_STAT(ipss, in_discards);
2989 ip_drop_packet(data_mp, B_TRUE, ira->ira_ill,
2990 DROPPER(ipss, ipds_esp_icmp),
2991 &espstack->esp_dropper);
2992 return (NULL);
2993 }
2994
2995 /*
2996 * Construct an SADB_REGISTER message with the current algorithms.
2997 * This function gets called when 'ipsecalgs -s' is run or when
2998 * in.iked (or other KMD) starts.
2999 */
3000 static boolean_t
3001 esp_register_out(uint32_t sequence, uint32_t pid, uint_t serial,
3002 ipsecesp_stack_t *espstack, cred_t *cr)
3003 {
3004 mblk_t *pfkey_msg_mp, *keysock_out_mp;
3005 sadb_msg_t *samsg;
3006 sadb_supported_t *sasupp_auth = NULL;
3007 sadb_supported_t *sasupp_encr = NULL;
3008 sadb_alg_t *saalg;
3009 uint_t allocsize = sizeof (*samsg);
3010 uint_t i, numalgs_snap;
3011 int current_aalgs;
3012 ipsec_alginfo_t **authalgs;
3013 uint_t num_aalgs;
3014 int current_ealgs;
3015 ipsec_alginfo_t **encralgs;
3016 uint_t num_ealgs;
3017 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec;
3018 sadb_sens_t *sens;
3019 size_t sens_len = 0;
3020 sadb_ext_t *nextext;
3021 ts_label_t *sens_tsl = NULL;
3022
3023 /* Allocate the KEYSOCK_OUT. */
3024 keysock_out_mp = sadb_keysock_out(serial);
3025 if (keysock_out_mp == NULL) {
3026 esp0dbg(("esp_register_out: couldn't allocate mblk.\n"));
3027 return (B_FALSE);
3028 }
3029
3030 if (is_system_labeled() && (cr != NULL)) {
3031 sens_tsl = crgetlabel(cr);
3032 if (sens_tsl != NULL) {
3033 sens_len = sadb_sens_len_from_label(sens_tsl);
3034 allocsize += sens_len;
3035 }
3036 }
3037
3038 /*
3039 * Allocate the PF_KEY message that follows KEYSOCK_OUT.
3040 */
3041
3042 mutex_enter(&ipss->ipsec_alg_lock);
3043 /*
3044 * Fill SADB_REGISTER message's algorithm descriptors. Hold
3045 * down the lock while filling it.
3046 *
3047 * Return only valid algorithms, so the number of algorithms
3048 * to send up may be less than the number of algorithm entries
3049 * in the table.
3050 */
3051 authalgs = ipss->ipsec_alglists[IPSEC_ALG_AUTH];
3052 for (num_aalgs = 0, i = 0; i < IPSEC_MAX_ALGS; i++)
3053 if (authalgs[i] != NULL && ALG_VALID(authalgs[i]))
3054 num_aalgs++;
3055
3056 if (num_aalgs != 0) {
3057 allocsize += (num_aalgs * sizeof (*saalg));
3058 allocsize += sizeof (*sasupp_auth);
3059 }
3060 encralgs = ipss->ipsec_alglists[IPSEC_ALG_ENCR];
3061 for (num_ealgs = 0, i = 0; i < IPSEC_MAX_ALGS; i++)
3062 if (encralgs[i] != NULL && ALG_VALID(encralgs[i]))
3063 num_ealgs++;
3064
3065 if (num_ealgs != 0) {
3066 allocsize += (num_ealgs * sizeof (*saalg));
3067 allocsize += sizeof (*sasupp_encr);
3068 }
3069 keysock_out_mp->b_cont = allocb(allocsize, BPRI_HI);
3070 if (keysock_out_mp->b_cont == NULL) {
3071 mutex_exit(&ipss->ipsec_alg_lock);
3072 freemsg(keysock_out_mp);
3073 return (B_FALSE);
3074 }
3075 pfkey_msg_mp = keysock_out_mp->b_cont;
3076 pfkey_msg_mp->b_wptr += allocsize;
3077
3078 nextext = (sadb_ext_t *)(pfkey_msg_mp->b_rptr + sizeof (*samsg));
3079
3080 if (num_aalgs != 0) {
3081 sasupp_auth = (sadb_supported_t *)nextext;
3082 saalg = (sadb_alg_t *)(sasupp_auth + 1);
3083
3084 ASSERT(((ulong_t)saalg & 0x7) == 0);
3085
3086 numalgs_snap = 0;
3087 for (i = 0;
3088 ((i < IPSEC_MAX_ALGS) && (numalgs_snap < num_aalgs));
3089 i++) {
3090 if (authalgs[i] == NULL || !ALG_VALID(authalgs[i]))
3091 continue;
3092
3093 saalg->sadb_alg_id = authalgs[i]->alg_id;
3094 saalg->sadb_alg_ivlen = 0;
3095 saalg->sadb_alg_minbits = authalgs[i]->alg_ef_minbits;
3096 saalg->sadb_alg_maxbits = authalgs[i]->alg_ef_maxbits;
3097 saalg->sadb_x_alg_increment =
3098 authalgs[i]->alg_increment;
3099 saalg->sadb_x_alg_saltbits = SADB_8TO1(
3100 authalgs[i]->alg_saltlen);
3101 numalgs_snap++;
3102 saalg++;
3103 }
3104 ASSERT(numalgs_snap == num_aalgs);
3105 #ifdef DEBUG
3106 /*
3107 * Reality check to make sure I snagged all of the
3108 * algorithms.
3109 */
3110 for (; i < IPSEC_MAX_ALGS; i++) {
3111 if (authalgs[i] != NULL && ALG_VALID(authalgs[i])) {
3112 cmn_err(CE_PANIC, "esp_register_out()! "
3113 "Missed aalg #%d.\n", i);
3114 }
3115 }
3116 #endif /* DEBUG */
3117 nextext = (sadb_ext_t *)saalg;
3118 }
3119
3120 if (num_ealgs != 0) {
3121 sasupp_encr = (sadb_supported_t *)nextext;
3122 saalg = (sadb_alg_t *)(sasupp_encr + 1);
3123
3124 numalgs_snap = 0;
3125 for (i = 0;
3126 ((i < IPSEC_MAX_ALGS) && (numalgs_snap < num_ealgs)); i++) {
3127 if (encralgs[i] == NULL || !ALG_VALID(encralgs[i]))
3128 continue;
3129 saalg->sadb_alg_id = encralgs[i]->alg_id;
3130 saalg->sadb_alg_ivlen = encralgs[i]->alg_ivlen;
3131 saalg->sadb_alg_minbits = encralgs[i]->alg_ef_minbits;
3132 saalg->sadb_alg_maxbits = encralgs[i]->alg_ef_maxbits;
3133 /*
3134 * We could advertise the ICV length, except there
3135 * is not a value in sadb_x_algb to do this.
3136 * saalg->sadb_alg_maclen = encralgs[i]->alg_maclen;
3137 */
3138 saalg->sadb_x_alg_increment =
3139 encralgs[i]->alg_increment;
3140 saalg->sadb_x_alg_saltbits =
3141 SADB_8TO1(encralgs[i]->alg_saltlen);
3142
3143 numalgs_snap++;
3144 saalg++;
3145 }
3146 ASSERT(numalgs_snap == num_ealgs);
3147 #ifdef DEBUG
3148 /*
3149 * Reality check to make sure I snagged all of the
3150 * algorithms.
3151 */
3152 for (; i < IPSEC_MAX_ALGS; i++) {
3153 if (encralgs[i] != NULL && ALG_VALID(encralgs[i])) {
3154 cmn_err(CE_PANIC, "esp_register_out()! "
3155 "Missed ealg #%d.\n", i);
3156 }
3157 }
3158 #endif /* DEBUG */
3159 nextext = (sadb_ext_t *)saalg;
3160 }
3161
3162 current_aalgs = num_aalgs;
3163 current_ealgs = num_ealgs;
3164
3165 mutex_exit(&ipss->ipsec_alg_lock);
3166
3167 if (sens_tsl != NULL) {
3168 sens = (sadb_sens_t *)nextext;
3169 sadb_sens_from_label(sens, SADB_EXT_SENSITIVITY,
3170 sens_tsl, sens_len);
3171
3172 nextext = (sadb_ext_t *)(((uint8_t *)sens) + sens_len);
3173 }
3174
3175 /* Now fill the rest of the SADB_REGISTER message. */
3176
3177 samsg = (sadb_msg_t *)pfkey_msg_mp->b_rptr;
3178 samsg->sadb_msg_version = PF_KEY_V2;
3179 samsg->sadb_msg_type = SADB_REGISTER;
3180 samsg->sadb_msg_errno = 0;
3181 samsg->sadb_msg_satype = SADB_SATYPE_ESP;
3182 samsg->sadb_msg_len = SADB_8TO64(allocsize);
3183 samsg->sadb_msg_reserved = 0;
3184 /*
3185 * Assume caller has sufficient sequence/pid number info. If it's one
3186 * from me over a new alg., I could give two hoots about sequence.
3187 */
3188 samsg->sadb_msg_seq = sequence;
3189 samsg->sadb_msg_pid = pid;
3190
3191 if (sasupp_auth != NULL) {
3192 sasupp_auth->sadb_supported_len = SADB_8TO64(
3193 sizeof (*sasupp_auth) + sizeof (*saalg) * current_aalgs);
3194 sasupp_auth->sadb_supported_exttype = SADB_EXT_SUPPORTED_AUTH;
3195 sasupp_auth->sadb_supported_reserved = 0;
3196 }
3197
3198 if (sasupp_encr != NULL) {
3199 sasupp_encr->sadb_supported_len = SADB_8TO64(
3200 sizeof (*sasupp_encr) + sizeof (*saalg) * current_ealgs);
3201 sasupp_encr->sadb_supported_exttype =
3202 SADB_EXT_SUPPORTED_ENCRYPT;
3203 sasupp_encr->sadb_supported_reserved = 0;
3204 }
3205
3206 if (espstack->esp_pfkey_q != NULL)
3207 putnext(espstack->esp_pfkey_q, keysock_out_mp);
3208 else {
3209 freemsg(keysock_out_mp);
3210 return (B_FALSE);
3211 }
3212
3213 return (B_TRUE);
3214 }
3215
3216 /*
3217 * Invoked when the algorithm table changes. Causes SADB_REGISTER
3218 * messages continaining the current list of algorithms to be
3219 * sent up to the ESP listeners.
3220 */
3221 void
3222 ipsecesp_algs_changed(netstack_t *ns)
3223 {
3224 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3225
3226 /*
3227 * Time to send a PF_KEY SADB_REGISTER message to ESP listeners
3228 * everywhere. (The function itself checks for NULL esp_pfkey_q.)
3229 */
3230 (void) esp_register_out(0, 0, 0, espstack, NULL);
3231 }
3232
3233 /*
3234 * Stub function that taskq_dispatch() invokes to take the mblk (in arg)
3235 * and send() it into ESP and IP again.
3236 */
3237 static void
3238 inbound_task(void *arg)
3239 {
3240 mblk_t *mp = (mblk_t *)arg;
3241 mblk_t *async_mp;
3242 ip_recv_attr_t iras;
3243
3244 async_mp = mp;
3245 mp = async_mp->b_cont;
3246 async_mp->b_cont = NULL;
3247 if (!ip_recv_attr_from_mblk(async_mp, &iras)) {
3248 /* The ill or ip_stack_t disappeared on us */
3249 ip_drop_input("ip_recv_attr_from_mblk", mp, NULL);
3250 freemsg(mp);
3251 goto done;
3252 }
3253
3254 esp_inbound_restart(mp, &iras);
3255 done:
3256 ira_cleanup(&iras, B_TRUE);
3257 }
3258
3259 /*
3260 * Restart ESP after the SA has been added.
3261 */
3262 static void
3263 esp_inbound_restart(mblk_t *mp, ip_recv_attr_t *ira)
3264 {
3265 esph_t *esph;
3266 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
3267 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3268
3269 esp2dbg(espstack, ("in ESP inbound_task"));
3270 ASSERT(espstack != NULL);
3271
3272 mp = ipsec_inbound_esp_sa(mp, ira, &esph);
3273 if (mp == NULL)
3274 return;
3275
3276 ASSERT(esph != NULL);
3277 ASSERT(ira->ira_flags & IRAF_IPSEC_SECURE);
3278 ASSERT(ira->ira_ipsec_esp_sa != NULL);
3279
3280 mp = ira->ira_ipsec_esp_sa->ipsa_input_func(mp, esph, ira);
3281 if (mp == NULL) {
3282 /*
3283 * Either it failed or is pending. In the former case
3284 * ipIfStatsInDiscards was increased.
3285 */
3286 return;
3287 }
3288
3289 ip_input_post_ipsec(mp, ira);
3290 }
3291
3292 /*
3293 * Now that weak-key passed, actually ADD the security association, and
3294 * send back a reply ADD message.
3295 */
3296 static int
3297 esp_add_sa_finish(mblk_t *mp, sadb_msg_t *samsg, keysock_in_t *ksi,
3298 int *diagnostic, ipsecesp_stack_t *espstack)
3299 {
3300 isaf_t *primary = NULL, *secondary;
3301 boolean_t clone = B_FALSE, is_inbound = B_FALSE;
3302 ipsa_t *larval = NULL;
3303 ipsacq_t *acqrec;
3304 iacqf_t *acq_bucket;
3305 mblk_t *acq_msgs = NULL;
3306 int rc;
3307 mblk_t *lpkt;
3308 int error;
3309 ipsa_query_t sq;
3310 ipsec_stack_t *ipss = espstack->ipsecesp_netstack->netstack_ipsec;
3311
3312 /*
3313 * Locate the appropriate table(s).
3314 */
3315 sq.spp = &espstack->esp_sadb; /* XXX */
3316 error = sadb_form_query(ksi, IPSA_Q_SA|IPSA_Q_DST,
3317 IPSA_Q_SA|IPSA_Q_DST|IPSA_Q_INBOUND|IPSA_Q_OUTBOUND,
3318 &sq, diagnostic);
3319 if (error)
3320 return (error);
3321
3322 /*
3323 * Use the direction flags provided by the KMD to determine
3324 * if the inbound or outbound table should be the primary
3325 * for this SA. If these flags were absent then make this
3326 * decision based on the addresses.
3327 */
3328 if (sq.assoc->sadb_sa_flags & IPSA_F_INBOUND) {
3329 primary = sq.inbound;
3330 secondary = sq.outbound;
3331 is_inbound = B_TRUE;
3332 if (sq.assoc->sadb_sa_flags & IPSA_F_OUTBOUND)
3333 clone = B_TRUE;
3334 } else if (sq.assoc->sadb_sa_flags & IPSA_F_OUTBOUND) {
3335 primary = sq.outbound;
3336 secondary = sq.inbound;
3337 }
3338
3339 if (primary == NULL) {
3340 /*
3341 * The KMD did not set a direction flag, determine which
3342 * table to insert the SA into based on addresses.
3343 */
3344 switch (ksi->ks_in_dsttype) {
3345 case KS_IN_ADDR_MBCAST:
3346 clone = B_TRUE; /* All mcast SAs can be bidirectional */
3347 sq.assoc->sadb_sa_flags |= IPSA_F_OUTBOUND;
3348 /* FALLTHRU */
3349 /*
3350 * If the source address is either one of mine, or unspecified
3351 * (which is best summed up by saying "not 'not mine'"),
3352 * then the association is potentially bi-directional,
3353 * in that it can be used for inbound traffic and outbound
3354 * traffic. The best example of such an SA is a multicast
3355 * SA (which allows me to receive the outbound traffic).
3356 */
3357 case KS_IN_ADDR_ME:
3358 sq.assoc->sadb_sa_flags |= IPSA_F_INBOUND;
3359 primary = sq.inbound;
3360 secondary = sq.outbound;
3361 if (ksi->ks_in_srctype != KS_IN_ADDR_NOTME)
3362 clone = B_TRUE;
3363 is_inbound = B_TRUE;
3364 break;
3365 /*
3366 * If the source address literally not mine (either
3367 * unspecified or not mine), then this SA may have an
3368 * address that WILL be mine after some configuration.
3369 * We pay the price for this by making it a bi-directional
3370 * SA.
3371 */
3372 case KS_IN_ADDR_NOTME:
3373 sq.assoc->sadb_sa_flags |= IPSA_F_OUTBOUND;
3374 primary = sq.outbound;
3375 secondary = sq.inbound;
3376 if (ksi->ks_in_srctype != KS_IN_ADDR_ME) {
3377 sq.assoc->sadb_sa_flags |= IPSA_F_INBOUND;
3378 clone = B_TRUE;
3379 }
3380 break;
3381 default:
3382 *diagnostic = SADB_X_DIAGNOSTIC_BAD_DST;
3383 return (EINVAL);
3384 }
3385 }
3386
3387 /*
3388 * Find a ACQUIRE list entry if possible. If we've added an SA that
3389 * suits the needs of an ACQUIRE list entry, we can eliminate the
3390 * ACQUIRE list entry and transmit the enqueued packets. Use the
3391 * high-bit of the sequence number to queue it. Key off destination
3392 * addr, and change acqrec's state.
3393 */
3394
3395 if (samsg->sadb_msg_seq & IACQF_LOWEST_SEQ) {
3396 acq_bucket = &(sq.sp->sdb_acq[sq.outhash]);
3397 mutex_enter(&acq_bucket->iacqf_lock);
3398 for (acqrec = acq_bucket->iacqf_ipsacq; acqrec != NULL;
3399 acqrec = acqrec->ipsacq_next) {
3400 mutex_enter(&acqrec->ipsacq_lock);
3401 /*
3402 * Q: I only check sequence. Should I check dst?
3403 * A: Yes, check dest because those are the packets
3404 * that are queued up.
3405 */
3406 if (acqrec->ipsacq_seq == samsg->sadb_msg_seq &&
3407 IPSA_ARE_ADDR_EQUAL(sq.dstaddr,
3408 acqrec->ipsacq_dstaddr, acqrec->ipsacq_addrfam))
3409 break;
3410 mutex_exit(&acqrec->ipsacq_lock);
3411 }
3412 if (acqrec != NULL) {
3413 /*
3414 * AHA! I found an ACQUIRE record for this SA.
3415 * Grab the msg list, and free the acquire record.
3416 * I already am holding the lock for this record,
3417 * so all I have to do is free it.
3418 */
3419 acq_msgs = acqrec->ipsacq_mp;
3420 acqrec->ipsacq_mp = NULL;
3421 mutex_exit(&acqrec->ipsacq_lock);
3422 sadb_destroy_acquire(acqrec,
3423 espstack->ipsecesp_netstack);
3424 }
3425 mutex_exit(&acq_bucket->iacqf_lock);
3426 }
3427
3428 /*
3429 * Find PF_KEY message, and see if I'm an update. If so, find entry
3430 * in larval list (if there).
3431 */
3432 if (samsg->sadb_msg_type == SADB_UPDATE) {
3433 mutex_enter(&sq.inbound->isaf_lock);
3434 larval = ipsec_getassocbyspi(sq.inbound, sq.assoc->sadb_sa_spi,
3435 ALL_ZEROES_PTR, sq.dstaddr, sq.dst->sin_family);
3436 mutex_exit(&sq.inbound->isaf_lock);
3437
3438 if ((larval == NULL) ||
3439 (larval->ipsa_state != IPSA_STATE_LARVAL)) {
3440 *diagnostic = SADB_X_DIAGNOSTIC_SA_NOTFOUND;
3441 if (larval != NULL) {
3442 IPSA_REFRELE(larval);
3443 }
3444 esp0dbg(("Larval update, but larval disappeared.\n"));
3445 return (ESRCH);
3446 } /* Else sadb_common_add unlinks it for me! */
3447 }
3448
3449 if (larval != NULL) {
3450 /*
3451 * Hold again, because sadb_common_add() consumes a reference,
3452 * and we don't want to clear_lpkt() without a reference.
3453 */
3454 IPSA_REFHOLD(larval);
3455 }
3456
3457 rc = sadb_common_add(espstack->esp_pfkey_q,
3458 mp, samsg, ksi, primary, secondary, larval, clone, is_inbound,
3459 diagnostic, espstack->ipsecesp_netstack, &espstack->esp_sadb);
3460
3461 if (larval != NULL) {
3462 if (rc == 0) {
3463 lpkt = sadb_clear_lpkt(larval);
3464 if (lpkt != NULL) {
3465 rc = !taskq_dispatch(esp_taskq, inbound_task,
3466 lpkt, TQ_NOSLEEP);
3467 }
3468 }
3469 IPSA_REFRELE(larval);
3470 }
3471
3472 /*
3473 * How much more stack will I create with all of these
3474 * esp_outbound() calls?
3475 */
3476
3477 /* Handle the packets queued waiting for the SA */
3478 while (acq_msgs != NULL) {
3479 mblk_t *asyncmp;
3480 mblk_t *data_mp;
3481 ip_xmit_attr_t ixas;
3482 ill_t *ill;
3483
3484 asyncmp = acq_msgs;
3485 acq_msgs = acq_msgs->b_next;
3486 asyncmp->b_next = NULL;
3487
3488 /*
3489 * Extract the ip_xmit_attr_t from the first mblk.
3490 * Verifies that the netstack and ill is still around; could
3491 * have vanished while iked was doing its work.
3492 * On succesful return we have a nce_t and the ill/ipst can't
3493 * disappear until we do the nce_refrele in ixa_cleanup.
3494 */
3495 data_mp = asyncmp->b_cont;
3496 asyncmp->b_cont = NULL;
3497 if (!ip_xmit_attr_from_mblk(asyncmp, &ixas)) {
3498 ESP_BUMP_STAT(espstack, out_discards);
3499 ip_drop_packet(data_mp, B_FALSE, NULL,
3500 DROPPER(ipss, ipds_sadb_acquire_timeout),
3501 &espstack->esp_dropper);
3502 } else if (rc != 0) {
3503 ill = ixas.ixa_nce->nce_ill;
3504 ESP_BUMP_STAT(espstack, out_discards);
3505 ip_drop_packet(data_mp, B_FALSE, ill,
3506 DROPPER(ipss, ipds_sadb_acquire_timeout),
3507 &espstack->esp_dropper);
3508 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
3509 } else {
3510 esp_outbound_finish(data_mp, &ixas);
3511 }
3512 ixa_cleanup(&ixas);
3513 }
3514
3515 return (rc);
3516 }
3517
3518 /*
3519 * Process one of the queued messages (from ipsacq_mp) once the SA
3520 * has been added.
3521 */
3522 static void
3523 esp_outbound_finish(mblk_t *data_mp, ip_xmit_attr_t *ixa)
3524 {
3525 netstack_t *ns = ixa->ixa_ipst->ips_netstack;
3526 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3527 ipsec_stack_t *ipss = ns->netstack_ipsec;
3528 ill_t *ill = ixa->ixa_nce->nce_ill;
3529
3530 if (!ipsec_outbound_sa(data_mp, ixa, IPPROTO_ESP)) {
3531 ESP_BUMP_STAT(espstack, out_discards);
3532 ip_drop_packet(data_mp, B_FALSE, ill,
3533 DROPPER(ipss, ipds_sadb_acquire_timeout),
3534 &espstack->esp_dropper);
3535 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
3536 return;
3537 }
3538
3539 data_mp = esp_outbound(data_mp, ixa);
3540 if (data_mp == NULL)
3541 return;
3542
3543 /* do AH processing if needed */
3544 data_mp = esp_do_outbound_ah(data_mp, ixa);
3545 if (data_mp == NULL)
3546 return;
3547
3548 (void) ip_output_post_ipsec(data_mp, ixa);
3549 }
3550
3551 /*
3552 * Add new ESP security association. This may become a generic AH/ESP
3553 * routine eventually.
3554 */
3555 static int
3556 esp_add_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic, netstack_t *ns)
3557 {
3558 sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
3559 sadb_address_t *srcext =
3560 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
3561 sadb_address_t *dstext =
3562 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
3563 sadb_address_t *isrcext =
3564 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_SRC];
3565 sadb_address_t *idstext =
3566 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_INNER_DST];
3567 sadb_address_t *nttext_loc =
3568 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_LOC];
3569 sadb_address_t *nttext_rem =
3570 (sadb_address_t *)ksi->ks_in_extv[SADB_X_EXT_ADDRESS_NATT_REM];
3571 sadb_key_t *akey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_AUTH];
3572 sadb_key_t *ekey = (sadb_key_t *)ksi->ks_in_extv[SADB_EXT_KEY_ENCRYPT];
3573 struct sockaddr_in *src, *dst;
3574 struct sockaddr_in *natt_loc, *natt_rem;
3575 struct sockaddr_in6 *natt_loc6, *natt_rem6;
3576 sadb_lifetime_t *soft =
3577 (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_SOFT];
3578 sadb_lifetime_t *hard =
3579 (sadb_lifetime_t *)ksi->ks_in_extv[SADB_EXT_LIFETIME_HARD];
3580 sadb_lifetime_t *idle =
3581 (sadb_lifetime_t *)ksi->ks_in_extv[SADB_X_EXT_LIFETIME_IDLE];
3582 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
3583 ipsec_stack_t *ipss = ns->netstack_ipsec;
3584
3585
3586
3587 /* I need certain extensions present for an ADD message. */
3588 if (srcext == NULL) {
3589 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SRC;
3590 return (EINVAL);
3591 }
3592 if (dstext == NULL) {
3593 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
3594 return (EINVAL);
3595 }
3596 if (isrcext == NULL && idstext != NULL) {
3597 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_SRC;
3598 return (EINVAL);
3599 }
3600 if (isrcext != NULL && idstext == NULL) {
3601 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_INNER_DST;
3602 return (EINVAL);
3603 }
3604 if (assoc == NULL) {
3605 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
3606 return (EINVAL);
3607 }
3608 if (ekey == NULL && assoc->sadb_sa_encrypt != SADB_EALG_NULL) {
3609 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_EKEY;
3610 return (EINVAL);
3611 }
3612
3613 src = (struct sockaddr_in *)(srcext + 1);
3614 dst = (struct sockaddr_in *)(dstext + 1);
3615 natt_loc = (struct sockaddr_in *)(nttext_loc + 1);
3616 natt_loc6 = (struct sockaddr_in6 *)(nttext_loc + 1);
3617 natt_rem = (struct sockaddr_in *)(nttext_rem + 1);
3618 natt_rem6 = (struct sockaddr_in6 *)(nttext_rem + 1);
3619
3620 /* Sundry ADD-specific reality checks. */
3621 /* XXX STATS : Logging/stats here? */
3622
3623 if ((assoc->sadb_sa_state != SADB_SASTATE_MATURE) &&
3624 (assoc->sadb_sa_state != SADB_X_SASTATE_ACTIVE_ELSEWHERE)) {
3625 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SASTATE;
3626 return (EINVAL);
3627 }
3628 if (assoc->sadb_sa_encrypt == SADB_EALG_NONE) {
3629 *diagnostic = SADB_X_DIAGNOSTIC_BAD_EALG;
3630 return (EINVAL);
3631 }
3632
3633 #ifndef IPSEC_LATENCY_TEST
3634 if (assoc->sadb_sa_encrypt == SADB_EALG_NULL &&
3635 assoc->sadb_sa_auth == SADB_AALG_NONE) {
3636 *diagnostic = SADB_X_DIAGNOSTIC_BAD_AALG;
3637 return (EINVAL);
3638 }
3639 #endif
3640
3641 if (assoc->sadb_sa_flags & ~espstack->esp_sadb.s_addflags) {
3642 *diagnostic = SADB_X_DIAGNOSTIC_BAD_SAFLAGS;
3643 return (EINVAL);
3644 }
3645
3646 if ((*diagnostic = sadb_hardsoftchk(hard, soft, idle)) != 0) {
3647 return (EINVAL);
3648 }
3649 ASSERT(src->sin_family == dst->sin_family);
3650
3651 if (assoc->sadb_sa_flags & SADB_X_SAFLAGS_NATT_LOC) {
3652 if (nttext_loc == NULL) {
3653 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_LOC;
3654 return (EINVAL);
3655 }
3656
3657 if (natt_loc->sin_family == AF_INET6 &&
3658 !IN6_IS_ADDR_V4MAPPED(&natt_loc6->sin6_addr)) {
3659 *diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_NATT_LOC;
3660 return (EINVAL);
3661 }
3662 }
3663
3664 if (assoc->sadb_sa_flags & SADB_X_SAFLAGS_NATT_REM) {
3665 if (nttext_rem == NULL) {
3666 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_NATT_REM;
3667 return (EINVAL);
3668 }
3669 if (natt_rem->sin_family == AF_INET6 &&
3670 !IN6_IS_ADDR_V4MAPPED(&natt_rem6->sin6_addr)) {
3671 *diagnostic = SADB_X_DIAGNOSTIC_MALFORMED_NATT_REM;
3672 return (EINVAL);
3673 }
3674 }
3675
3676
3677 /* Stuff I don't support, for now. XXX Diagnostic? */
3678 if (ksi->ks_in_extv[SADB_EXT_LIFETIME_CURRENT] != NULL)
3679 return (EOPNOTSUPP);
3680
3681 if ((*diagnostic = sadb_labelchk(ksi)) != 0)
3682 return (EINVAL);
3683
3684 /*
3685 * XXX Policy : I'm not checking identities at this time,
3686 * but if I did, I'd do them here, before I sent
3687 * the weak key check up to the algorithm.
3688 */
3689
3690 mutex_enter(&ipss->ipsec_alg_lock);
3691
3692 /*
3693 * First locate the authentication algorithm.
3694 */
3695 #ifdef IPSEC_LATENCY_TEST
3696 if (akey != NULL && assoc->sadb_sa_auth != SADB_AALG_NONE) {
3697 #else
3698 if (akey != NULL) {
3699 #endif
3700 ipsec_alginfo_t *aalg;
3701
3702 aalg = ipss->ipsec_alglists[IPSEC_ALG_AUTH]
3703 [assoc->sadb_sa_auth];
3704 if (aalg == NULL || !ALG_VALID(aalg)) {
3705 mutex_exit(&ipss->ipsec_alg_lock);
3706 esp1dbg(espstack, ("Couldn't find auth alg #%d.\n",
3707 assoc->sadb_sa_auth));
3708 *diagnostic = SADB_X_DIAGNOSTIC_BAD_AALG;
3709 return (EINVAL);
3710 }
3711
3712 /*
3713 * Sanity check key sizes.
3714 * Note: It's not possible to use SADB_AALG_NONE because
3715 * this auth_alg is not defined with ALG_FLAG_VALID. If this
3716 * ever changes, the same check for SADB_AALG_NONE and
3717 * a auth_key != NULL should be made here ( see below).
3718 */
3719 if (!ipsec_valid_key_size(akey->sadb_key_bits, aalg)) {
3720 mutex_exit(&ipss->ipsec_alg_lock);
3721 *diagnostic = SADB_X_DIAGNOSTIC_BAD_AKEYBITS;
3722 return (EINVAL);
3723 }
3724 ASSERT(aalg->alg_mech_type != CRYPTO_MECHANISM_INVALID);
3725
3726 /* check key and fix parity if needed */
3727 if (ipsec_check_key(aalg->alg_mech_type, akey, B_TRUE,
3728 diagnostic) != 0) {
3729 mutex_exit(&ipss->ipsec_alg_lock);
3730 return (EINVAL);
3731 }
3732 }
3733
3734 /*
3735 * Then locate the encryption algorithm.
3736 */
3737 if (ekey != NULL) {
3738 uint_t keybits;
3739 ipsec_alginfo_t *ealg;
3740
3741 ealg = ipss->ipsec_alglists[IPSEC_ALG_ENCR]
3742 [assoc->sadb_sa_encrypt];
3743 if (ealg == NULL || !ALG_VALID(ealg)) {
3744 mutex_exit(&ipss->ipsec_alg_lock);
3745 esp1dbg(espstack, ("Couldn't find encr alg #%d.\n",
3746 assoc->sadb_sa_encrypt));
3747 *diagnostic = SADB_X_DIAGNOSTIC_BAD_EALG;
3748 return (EINVAL);
3749 }
3750
3751 /*
3752 * Sanity check key sizes. If the encryption algorithm is
3753 * SADB_EALG_NULL but the encryption key is NOT
3754 * NULL then complain.
3755 *
3756 * The keying material includes salt bits if required by
3757 * algorithm and optionally the Initial IV, check the
3758 * length of whats left.
3759 */
3760 keybits = ekey->sadb_key_bits;
3761 keybits -= ekey->sadb_key_reserved;
3762 keybits -= SADB_8TO1(ealg->alg_saltlen);
3763 if ((assoc->sadb_sa_encrypt == SADB_EALG_NULL) ||
3764 (!ipsec_valid_key_size(keybits, ealg))) {
3765 mutex_exit(&ipss->ipsec_alg_lock);
3766 *diagnostic = SADB_X_DIAGNOSTIC_BAD_EKEYBITS;
3767 return (EINVAL);
3768 }
3769 ASSERT(ealg->alg_mech_type != CRYPTO_MECHANISM_INVALID);
3770
3771 /* check key */
3772 if (ipsec_check_key(ealg->alg_mech_type, ekey, B_FALSE,
3773 diagnostic) != 0) {
3774 mutex_exit(&ipss->ipsec_alg_lock);
3775 return (EINVAL);
3776 }
3777 }
3778 mutex_exit(&ipss->ipsec_alg_lock);
3779
3780 return (esp_add_sa_finish(mp, (sadb_msg_t *)mp->b_cont->b_rptr, ksi,
3781 diagnostic, espstack));
3782 }
3783
3784 /*
3785 * Update a security association. Updates come in two varieties. The first
3786 * is an update of lifetimes on a non-larval SA. The second is an update of
3787 * a larval SA, which ends up looking a lot more like an add.
3788 */
3789 static int
3790 esp_update_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic,
3791 ipsecesp_stack_t *espstack, uint8_t sadb_msg_type)
3792 {
3793 sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
3794 mblk_t *buf_pkt;
3795 int rcode;
3796
3797 sadb_address_t *dstext =
3798 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
3799
3800 if (dstext == NULL) {
3801 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_DST;
3802 return (EINVAL);
3803 }
3804
3805 rcode = sadb_update_sa(mp, ksi, &buf_pkt, &espstack->esp_sadb,
3806 diagnostic, espstack->esp_pfkey_q, esp_add_sa,
3807 espstack->ipsecesp_netstack, sadb_msg_type);
3808
3809 if ((assoc->sadb_sa_state != SADB_X_SASTATE_ACTIVE) ||
3810 (rcode != 0)) {
3811 return (rcode);
3812 }
3813
3814 HANDLE_BUF_PKT(esp_taskq, espstack->ipsecesp_netstack->netstack_ipsec,
3815 espstack->esp_dropper, buf_pkt);
3816
3817 return (rcode);
3818 }
3819
3820 /* XXX refactor me */
3821 /*
3822 * Delete a security association. This is REALLY likely to be code common to
3823 * both AH and ESP. Find the association, then unlink it.
3824 */
3825 static int
3826 esp_del_sa(mblk_t *mp, keysock_in_t *ksi, int *diagnostic,
3827 ipsecesp_stack_t *espstack, uint8_t sadb_msg_type)
3828 {
3829 sadb_sa_t *assoc = (sadb_sa_t *)ksi->ks_in_extv[SADB_EXT_SA];
3830 sadb_address_t *dstext =
3831 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_DST];
3832 sadb_address_t *srcext =
3833 (sadb_address_t *)ksi->ks_in_extv[SADB_EXT_ADDRESS_SRC];
3834 struct sockaddr_in *sin;
3835
3836 if (assoc == NULL) {
3837 if (dstext != NULL) {
3838 sin = (struct sockaddr_in *)(dstext + 1);
3839 } else if (srcext != NULL) {
3840 sin = (struct sockaddr_in *)(srcext + 1);
3841 } else {
3842 *diagnostic = SADB_X_DIAGNOSTIC_MISSING_SA;
3843 return (EINVAL);
3844 }
3845 return (sadb_purge_sa(mp, ksi,
3846 (sin->sin_family == AF_INET6) ? &espstack->esp_sadb.s_v6 :
3847 &espstack->esp_sadb.s_v4, diagnostic,
3848 espstack->esp_pfkey_q));
3849 }
3850
3851 return (sadb_delget_sa(mp, ksi, &espstack->esp_sadb, diagnostic,
3852 espstack->esp_pfkey_q, sadb_msg_type));
3853 }
3854
3855 /* XXX refactor me */
3856 /*
3857 * Convert the entire contents of all of ESP's SA tables into PF_KEY SADB_DUMP
3858 * messages.
3859 */
3860 static void
3861 esp_dump(mblk_t *mp, keysock_in_t *ksi, ipsecesp_stack_t *espstack)
3862 {
3863 int error;
3864 sadb_msg_t *samsg;
3865
3866 /*
3867 * Dump each fanout, bailing if error is non-zero.
3868 */
3869
3870 error = sadb_dump(espstack->esp_pfkey_q, mp, ksi,
3871 &espstack->esp_sadb.s_v4);
3872 if (error != 0)
3873 goto bail;
3874
3875 error = sadb_dump(espstack->esp_pfkey_q, mp, ksi,
3876 &espstack->esp_sadb.s_v6);
3877 bail:
3878 ASSERT(mp->b_cont != NULL);
3879 samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
3880 samsg->sadb_msg_errno = (uint8_t)error;
3881 sadb_pfkey_echo(espstack->esp_pfkey_q, mp,
3882 (sadb_msg_t *)mp->b_cont->b_rptr, ksi, NULL);
3883 }
3884
3885 /*
3886 * First-cut reality check for an inbound PF_KEY message.
3887 */
3888 static boolean_t
3889 esp_pfkey_reality_failures(mblk_t *mp, keysock_in_t *ksi,
3890 ipsecesp_stack_t *espstack)
3891 {
3892 int diagnostic;
3893
3894 if (ksi->ks_in_extv[SADB_EXT_PROPOSAL] != NULL) {
3895 diagnostic = SADB_X_DIAGNOSTIC_PROP_PRESENT;
3896 goto badmsg;
3897 }
3898 if (ksi->ks_in_extv[SADB_EXT_SUPPORTED_AUTH] != NULL ||
3899 ksi->ks_in_extv[SADB_EXT_SUPPORTED_ENCRYPT] != NULL) {
3900 diagnostic = SADB_X_DIAGNOSTIC_SUPP_PRESENT;
3901 goto badmsg;
3902 }
3903 return (B_FALSE); /* False ==> no failures */
3904
3905 badmsg:
3906 sadb_pfkey_error(espstack->esp_pfkey_q, mp, EINVAL, diagnostic,
3907 ksi->ks_in_serial);
3908 return (B_TRUE); /* True ==> failures */
3909 }
3910
3911 /*
3912 * ESP parsing of PF_KEY messages. Keysock did most of the really silly
3913 * error cases. What I receive is a fully-formed, syntactically legal
3914 * PF_KEY message. I then need to check semantics...
3915 *
3916 * This code may become common to AH and ESP. Stay tuned.
3917 *
3918 * I also make the assumption that db_ref's are cool. If this assumption
3919 * is wrong, this means that someone other than keysock or me has been
3920 * mucking with PF_KEY messages.
3921 */
3922 static void
3923 esp_parse_pfkey(mblk_t *mp, ipsecesp_stack_t *espstack)
3924 {
3925 mblk_t *msg = mp->b_cont;
3926 sadb_msg_t *samsg;
3927 keysock_in_t *ksi;
3928 int error;
3929 int diagnostic = SADB_X_DIAGNOSTIC_NONE;
3930
3931 ASSERT(msg != NULL);
3932
3933 samsg = (sadb_msg_t *)msg->b_rptr;
3934 ksi = (keysock_in_t *)mp->b_rptr;
3935
3936 /*
3937 * If applicable, convert unspecified AF_INET6 to unspecified
3938 * AF_INET. And do other address reality checks.
3939 */
3940 if (!sadb_addrfix(ksi, espstack->esp_pfkey_q, mp,
3941 espstack->ipsecesp_netstack) ||
3942 esp_pfkey_reality_failures(mp, ksi, espstack)) {
3943 return;
3944 }
3945
3946 switch (samsg->sadb_msg_type) {
3947 case SADB_ADD:
3948 error = esp_add_sa(mp, ksi, &diagnostic,
3949 espstack->ipsecesp_netstack);
3950 if (error != 0) {
3951 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
3952 diagnostic, ksi->ks_in_serial);
3953 }
3954 /* else esp_add_sa() took care of things. */
3955 break;
3956 case SADB_DELETE:
3957 case SADB_X_DELPAIR:
3958 case SADB_X_DELPAIR_STATE:
3959 error = esp_del_sa(mp, ksi, &diagnostic, espstack,
3960 samsg->sadb_msg_type);
3961 if (error != 0) {
3962 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
3963 diagnostic, ksi->ks_in_serial);
3964 }
3965 /* Else esp_del_sa() took care of things. */
3966 break;
3967 case SADB_GET:
3968 error = sadb_delget_sa(mp, ksi, &espstack->esp_sadb,
3969 &diagnostic, espstack->esp_pfkey_q, samsg->sadb_msg_type);
3970 if (error != 0) {
3971 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
3972 diagnostic, ksi->ks_in_serial);
3973 }
3974 /* Else sadb_get_sa() took care of things. */
3975 break;
3976 case SADB_FLUSH:
3977 sadbp_flush(&espstack->esp_sadb, espstack->ipsecesp_netstack);
3978 sadb_pfkey_echo(espstack->esp_pfkey_q, mp, samsg, ksi, NULL);
3979 break;
3980 case SADB_REGISTER:
3981 /*
3982 * Hmmm, let's do it! Check for extensions (there should
3983 * be none), extract the fields, call esp_register_out(),
3984 * then either free or report an error.
3985 *
3986 * Keysock takes care of the PF_KEY bookkeeping for this.
3987 */
3988 if (esp_register_out(samsg->sadb_msg_seq, samsg->sadb_msg_pid,
3989 ksi->ks_in_serial, espstack, msg_getcred(mp, NULL))) {
3990 freemsg(mp);
3991 } else {
3992 /*
3993 * Only way this path hits is if there is a memory
3994 * failure. It will not return B_FALSE because of
3995 * lack of esp_pfkey_q if I am in wput().
3996 */
3997 sadb_pfkey_error(espstack->esp_pfkey_q, mp, ENOMEM,
3998 diagnostic, ksi->ks_in_serial);
3999 }
4000 break;
4001 case SADB_UPDATE:
4002 case SADB_X_UPDATEPAIR:
4003 /*
4004 * Find a larval, if not there, find a full one and get
4005 * strict.
4006 */
4007 error = esp_update_sa(mp, ksi, &diagnostic, espstack,
4008 samsg->sadb_msg_type);
4009 if (error != 0) {
4010 sadb_pfkey_error(espstack->esp_pfkey_q, mp, error,
4011 diagnostic, ksi->ks_in_serial);
4012 }
4013 /* else esp_update_sa() took care of things. */
4014 break;
4015 case SADB_GETSPI:
4016 /*
4017 * Reserve a new larval entry.
4018 */
4019 esp_getspi(mp, ksi, espstack);
4020 break;
4021 case SADB_ACQUIRE:
4022 /*
4023 * Find larval and/or ACQUIRE record and kill it (them), I'm
4024 * most likely an error. Inbound ACQUIRE messages should only
4025 * have the base header.
4026 */
4027 sadb_in_acquire(samsg, &espstack->esp_sadb,
4028 espstack->esp_pfkey_q, espstack->ipsecesp_netstack);
4029 freemsg(mp);
4030 break;
4031 case SADB_DUMP:
4032 /*
4033 * Dump all entries.
4034 */
4035 esp_dump(mp, ksi, espstack);
4036 /* esp_dump will take care of the return message, etc. */
4037 break;
4038 case SADB_EXPIRE:
4039 /* Should never reach me. */
4040 sadb_pfkey_error(espstack->esp_pfkey_q, mp, EOPNOTSUPP,
4041 diagnostic, ksi->ks_in_serial);
4042 break;
4043 default:
4044 sadb_pfkey_error(espstack->esp_pfkey_q, mp, EINVAL,
4045 SADB_X_DIAGNOSTIC_UNKNOWN_MSG, ksi->ks_in_serial);
4046 break;
4047 }
4048 }
4049
4050 /*
4051 * Handle case where PF_KEY says it can't find a keysock for one of my
4052 * ACQUIRE messages.
4053 */
4054 static void
4055 esp_keysock_no_socket(mblk_t *mp, ipsecesp_stack_t *espstack)
4056 {
4057 sadb_msg_t *samsg;
4058 keysock_out_err_t *kse = (keysock_out_err_t *)mp->b_rptr;
4059
4060 if (mp->b_cont == NULL) {
4061 freemsg(mp);
4062 return;
4063 }
4064 samsg = (sadb_msg_t *)mp->b_cont->b_rptr;
4065
4066 /*
4067 * If keysock can't find any registered, delete the acquire record
4068 * immediately, and handle errors.
4069 */
4070 if (samsg->sadb_msg_type == SADB_ACQUIRE) {
4071 samsg->sadb_msg_errno = kse->ks_err_errno;
4072 samsg->sadb_msg_len = SADB_8TO64(sizeof (*samsg));
4073 /*
4074 * Use the write-side of the esp_pfkey_q
4075 */
4076 sadb_in_acquire(samsg, &espstack->esp_sadb,
4077 WR(espstack->esp_pfkey_q), espstack->ipsecesp_netstack);
4078 }
4079
4080 freemsg(mp);
4081 }
4082
4083 /*
4084 * ESP module write put routine.
4085 */
4086 static void
4087 ipsecesp_wput(queue_t *q, mblk_t *mp)
4088 {
4089 ipsec_info_t *ii;
4090 struct iocblk *iocp;
4091 ipsecesp_stack_t *espstack = (ipsecesp_stack_t *)q->q_ptr;
4092
4093 esp3dbg(espstack, ("In esp_wput().\n"));
4094
4095 /* NOTE: Each case must take care of freeing or passing mp. */
4096 switch (mp->b_datap->db_type) {
4097 case M_CTL:
4098 if ((mp->b_wptr - mp->b_rptr) < sizeof (ipsec_info_t)) {
4099 /* Not big enough message. */
4100 freemsg(mp);
4101 break;
4102 }
4103 ii = (ipsec_info_t *)mp->b_rptr;
4104
4105 switch (ii->ipsec_info_type) {
4106 case KEYSOCK_OUT_ERR:
4107 esp1dbg(espstack, ("Got KEYSOCK_OUT_ERR message.\n"));
4108 esp_keysock_no_socket(mp, espstack);
4109 break;
4110 case KEYSOCK_IN:
4111 ESP_BUMP_STAT(espstack, keysock_in);
4112 esp3dbg(espstack, ("Got KEYSOCK_IN message.\n"));
4113
4114 /* Parse the message. */
4115 esp_parse_pfkey(mp, espstack);
4116 break;
4117 case KEYSOCK_HELLO:
4118 sadb_keysock_hello(&espstack->esp_pfkey_q, q, mp,
4119 esp_ager, (void *)espstack, &espstack->esp_event,
4120 SADB_SATYPE_ESP);
4121 break;
4122 default:
4123 esp2dbg(espstack, ("Got M_CTL from above of 0x%x.\n",
4124 ii->ipsec_info_type));
4125 freemsg(mp);
4126 break;
4127 }
4128 break;
4129 case M_IOCTL:
4130 iocp = (struct iocblk *)mp->b_rptr;
4131 switch (iocp->ioc_cmd) {
4132 case ND_SET:
4133 case ND_GET:
4134 if (nd_getset(q, espstack->ipsecesp_g_nd, mp)) {
4135 qreply(q, mp);
4136 return;
4137 } else {
4138 iocp->ioc_error = ENOENT;
4139 }
4140 /* FALLTHRU */
4141 default:
4142 /* We really don't support any other ioctls, do we? */
4143
4144 /* Return EINVAL */
4145 if (iocp->ioc_error != ENOENT)
4146 iocp->ioc_error = EINVAL;
4147 iocp->ioc_count = 0;
4148 mp->b_datap->db_type = M_IOCACK;
4149 qreply(q, mp);
4150 return;
4151 }
4152 default:
4153 esp3dbg(espstack,
4154 ("Got default message, type %d, passing to IP.\n",
4155 mp->b_datap->db_type));
4156 putnext(q, mp);
4157 }
4158 }
4159
4160 /*
4161 * Wrapper to allow IP to trigger an ESP association failure message
4162 * during inbound SA selection.
4163 */
4164 void
4165 ipsecesp_in_assocfailure(mblk_t *mp, char level, ushort_t sl, char *fmt,
4166 uint32_t spi, void *addr, int af, ip_recv_attr_t *ira)
4167 {
4168 netstack_t *ns = ira->ira_ill->ill_ipst->ips_netstack;
4169 ipsecesp_stack_t *espstack = ns->netstack_ipsecesp;
4170 ipsec_stack_t *ipss = ns->netstack_ipsec;
4171
4172 if (espstack->ipsecesp_log_unknown_spi) {
4173 ipsec_assocfailure(info.mi_idnum, 0, level, sl, fmt, spi,
4174 addr, af, espstack->ipsecesp_netstack);
4175 }
4176
4177 ip_drop_packet(mp, B_TRUE, ira->ira_ill,
4178 DROPPER(ipss, ipds_esp_no_sa),
4179 &espstack->esp_dropper);
4180 }
4181
4182 /*
4183 * Initialize the ESP input and output processing functions.
4184 */
4185 void
4186 ipsecesp_init_funcs(ipsa_t *sa)
4187 {
4188 if (sa->ipsa_output_func == NULL)
4189 sa->ipsa_output_func = esp_outbound;
4190 if (sa->ipsa_input_func == NULL)
4191 sa->ipsa_input_func = esp_inbound;
4192 }