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.
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.
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]
22 * Copyright 2010 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
27 * IPsec Security Policy Database.
29 * This module maintains the SPD and provides routines used by ip and ip6
30 * to apply IPsec policy to inbound and outbound datagrams.
33 #include <sys/types.h>
34 #include <sys/stream.h>
35 #include <sys/stropts.h>
36 #include <sys/sysmacros.h>
37 #include <sys/strsubr.h>
38 #include <sys/strsun.h>
39 #include <sys/strlog.h>
40 #include <sys/strsun.h>
41 #include <sys/cmn_err.h>
44 #include <sys/systm.h>
45 #include <sys/param.h>
49 #include <sys/crypto/api.h>
51 #include <inet/common.h>
54 #include <netinet/ip6.h>
55 #include <netinet/icmp6.h>
56 #include <netinet/udp.h>
61 #include <net/pfkeyv2.h>
62 #include <net/pfpolicy.h>
63 #include <inet/sadb.h>
64 #include <inet/ipsec_impl.h>
66 #include <inet/ip_impl.h> /* For IP_MOD_ID */
68 #include <inet/ipsecah.h>
69 #include <inet/ipsecesp.h>
70 #include <inet/ipdrop.h>
71 #include <inet/ipclassifier.h>
72 #include <inet/iptun.h>
73 #include <inet/iptun/iptun_impl.h>
75 static void ipsec_update_present_flags(ipsec_stack_t
*);
76 static ipsec_act_t
*ipsec_act_wildcard_expand(ipsec_act_t
*, uint_t
*,
78 static mblk_t
*ipsec_check_ipsecin_policy(mblk_t
*, ipsec_policy_t
*,
79 ipha_t
*, ip6_t
*, uint64_t, ip_recv_attr_t
*, netstack_t
*);
80 static void ipsec_action_free_table(ipsec_action_t
*);
81 static void ipsec_action_reclaim(void *);
82 static void ipsec_action_reclaim_stack(ipsec_stack_t
*);
83 static void ipsid_init(netstack_t
*);
84 static void ipsid_fini(netstack_t
*);
86 /* sel_flags values for ipsec_init_inbound_sel(). */
87 #define SEL_NONE 0x0000
88 #define SEL_PORT_POLICY 0x0001
89 #define SEL_IS_ICMP 0x0002
90 #define SEL_TUNNEL_MODE 0x0004
91 #define SEL_POST_FRAG 0x0008
93 /* Return values for ipsec_init_inbound_sel(). */
94 typedef enum { SELRET_NOMEM
, SELRET_BADPKT
, SELRET_SUCCESS
, SELRET_TUNFRAG
}
97 static selret_t
ipsec_init_inbound_sel(ipsec_selector_t
*, mblk_t
*,
98 ipha_t
*, ip6_t
*, uint8_t);
100 static boolean_t
ipsec_check_ipsecin_action(ip_recv_attr_t
*, mblk_t
*,
101 struct ipsec_action_s
*, ipha_t
*ipha
, ip6_t
*ip6h
, const char **,
102 kstat_named_t
**, netstack_t
*);
103 static void ipsec_unregister_prov_update(void);
104 static void ipsec_prov_update_callback_stack(uint32_t, void *, netstack_t
*);
105 static boolean_t
ipsec_compare_action(ipsec_policy_t
*, ipsec_policy_t
*);
106 static uint32_t selector_hash(ipsec_selector_t
*, ipsec_policy_root_t
*);
107 static boolean_t
ipsec_kstat_init(ipsec_stack_t
*);
108 static void ipsec_kstat_destroy(ipsec_stack_t
*);
109 static int ipsec_free_tables(ipsec_stack_t
*);
110 static int tunnel_compare(const void *, const void *);
111 static void ipsec_freemsg_chain(mblk_t
*);
112 static void ip_drop_packet_chain(mblk_t
*, boolean_t
, ill_t
*,
113 struct kstat_named
*, ipdropper_t
*);
114 static boolean_t
ipsec_kstat_init(ipsec_stack_t
*);
115 static void ipsec_kstat_destroy(ipsec_stack_t
*);
116 static int ipsec_free_tables(ipsec_stack_t
*);
117 static int tunnel_compare(const void *, const void *);
118 static void ipsec_freemsg_chain(mblk_t
*);
121 * Selector hash table is statically sized at module load time.
122 * we default to 251 buckets, which is the largest prime number under 255
125 #define IPSEC_SPDHASH_DEFAULT 251
127 /* SPD hash-size tunable per tunnel. */
128 #define TUN_SPDHASH_DEFAULT 5
130 uint32_t ipsec_spd_hashsize
;
131 uint32_t tun_spd_hashsize
;
133 #define IPSEC_SEL_NOHASH ((uint32_t)(~0))
136 * Handle global across all stack instances
138 static crypto_notify_handle_t prov_update_handle
= NULL
;
140 static kmem_cache_t
*ipsec_action_cache
;
141 static kmem_cache_t
*ipsec_sel_cache
;
142 static kmem_cache_t
*ipsec_pol_cache
;
144 /* Frag cache prototypes */
145 static void ipsec_fragcache_clean(ipsec_fragcache_t
*, ipsec_stack_t
*);
146 static ipsec_fragcache_entry_t
*fragcache_delentry(int,
147 ipsec_fragcache_entry_t
*, ipsec_fragcache_t
*, ipsec_stack_t
*);
148 boolean_t
ipsec_fragcache_init(ipsec_fragcache_t
*);
149 void ipsec_fragcache_uninit(ipsec_fragcache_t
*, ipsec_stack_t
*ipss
);
150 mblk_t
*ipsec_fragcache_add(ipsec_fragcache_t
*, mblk_t
*, mblk_t
*,
151 int, ipsec_stack_t
*);
153 int ipsec_hdr_pullup_needed
= 0;
154 int ipsec_weird_null_inbound_policy
= 0;
156 #define ALGBITS_ROUND_DOWN(x, align) (((x)/(align))*(align))
157 #define ALGBITS_ROUND_UP(x, align) ALGBITS_ROUND_DOWN((x)+(align)-1, align)
160 * Inbound traffic should have matching identities for both SA's.
163 #define SA_IDS_MATCH(sa1, sa2) \
164 (((sa1) == NULL) || ((sa2) == NULL) || \
165 (((sa1)->ipsa_src_cid == (sa2)->ipsa_src_cid) && \
166 (((sa1)->ipsa_dst_cid == (sa2)->ipsa_dst_cid))))
171 #define IS_V6_FRAGMENT(ipp) (ipp.ipp_fields & IPPF_FRAGHDR)
174 * Policy failure messages.
176 static char *ipsec_policy_failure_msgs
[] = {
178 /* IPSEC_POLICY_NOT_NEEDED */
179 "%s: Dropping the datagram because the incoming packet "
180 "is %s, but the recipient expects clear; Source %s, "
183 /* IPSEC_POLICY_MISMATCH */
184 "%s: Policy Failure for the incoming packet (%s); Source %s, "
187 /* IPSEC_POLICY_AUTH_NOT_NEEDED */
188 "%s: Authentication present while not expected in the "
189 "incoming %s packet; Source %s, Destination %s.\n",
191 /* IPSEC_POLICY_ENCR_NOT_NEEDED */
192 "%s: Encryption present while not expected in the "
193 "incoming %s packet; Source %s, Destination %s.\n",
195 /* IPSEC_POLICY_SE_NOT_NEEDED */
196 "%s: Self-Encapsulation present while not expected in the "
197 "incoming %s packet; Source %s, Destination %s.\n",
205 * All of the system policy structures are protected by a single
206 * rwlock. These structures are threaded in a
207 * fairly complex fashion and are not expected to change on a
208 * regular basis, so this should not cause scaling/contention
209 * problems. As a result, policy checks should (hopefully) be MT-hot.
213 * We use custom kmem cache types for the various
214 * bits & pieces of the policy data structures. All allocations
215 * use KM_NOSLEEP instead of KM_SLEEP for policy allocation. The
216 * policy table is of potentially unbounded size, so we don't
217 * want to provide a way to hog all system memory with policy
221 /* Convenient functions for freeing or dropping a b_next linked mblk chain */
223 /* Free all messages in an mblk chain */
225 ipsec_freemsg_chain(mblk_t
*mp
)
229 ASSERT(mp
->b_prev
== NULL
);
238 * ip_drop all messages in an mblk chain
239 * Can handle a b_next chain of ip_recv_attr_t mblks, or just a b_next chain
243 ip_drop_packet_chain(mblk_t
*mp
, boolean_t inbound
, ill_t
*ill
,
244 struct kstat_named
*counter
, ipdropper_t
*who_called
)
248 ASSERT(mp
->b_prev
== NULL
);
251 if (ip_recv_attr_is_mblk(mp
))
252 mp
= ip_recv_attr_free_mblk(mp
);
253 ip_drop_packet(mp
, inbound
, ill
, counter
, who_called
);
259 * AVL tree comparison function.
260 * the in-kernel avl assumes unique keys for all objects.
261 * Since sometimes policy will duplicate rules, we may insert
262 * multiple rules with the same rule id, so we need a tie-breaker.
265 ipsec_policy_cmpbyid(const void *a
, const void *b
)
267 const ipsec_policy_t
*ipa
, *ipb
;
270 ipa
= (const ipsec_policy_t
*)a
;
271 ipb
= (const ipsec_policy_t
*)b
;
272 idxa
= ipa
->ipsp_index
;
273 idxb
= ipb
->ipsp_index
;
280 * Tie-breaker #1: All installed policy rules have a non-NULL
281 * ipsl_sel (selector set), so an entry with a NULL ipsp_sel is not
282 * actually in-tree but rather a template node being used in
283 * an avl_find query; see ipsec_policy_delete(). This gives us
284 * a placeholder in the ordering just before the first entry with
285 * a key >= the one we're looking for, so we can walk forward from
286 * that point to get the remaining entries with the same id.
288 if ((ipa
->ipsp_sel
== NULL
) && (ipb
->ipsp_sel
!= NULL
))
290 if ((ipb
->ipsp_sel
== NULL
) && (ipa
->ipsp_sel
!= NULL
))
293 * At most one of the arguments to the comparison should have a
294 * NULL selector pointer; if not, the tree is broken.
296 ASSERT(ipa
->ipsp_sel
!= NULL
);
297 ASSERT(ipb
->ipsp_sel
!= NULL
);
299 * Tie-breaker #2: use the virtual address of the policy node
300 * to arbitrarily break ties. Since we use the new tree node in
301 * the avl_find() in ipsec_insert_always, the new node will be
302 * inserted into the tree in the right place in the sequence.
312 * Free what ipsec_alloc_table allocated.
315 ipsec_polhead_free_table(ipsec_policy_head_t
*iph
)
320 for (dir
= 0; dir
< IPSEC_NTYPES
; dir
++) {
321 ipsec_policy_root_t
*ipr
= &iph
->iph_root
[dir
];
323 if (ipr
->ipr_hash
== NULL
)
326 for (i
= 0; i
< ipr
->ipr_nchains
; i
++) {
327 ASSERT(ipr
->ipr_hash
[i
].hash_head
== NULL
);
329 kmem_free(ipr
->ipr_hash
, ipr
->ipr_nchains
*
330 sizeof (ipsec_policy_hash_t
));
331 ipr
->ipr_hash
= NULL
;
336 ipsec_polhead_destroy(ipsec_policy_head_t
*iph
)
340 avl_destroy(&iph
->iph_rulebyid
);
341 rw_destroy(&iph
->iph_lock
);
343 for (dir
= 0; dir
< IPSEC_NTYPES
; dir
++) {
344 ipsec_policy_root_t
*ipr
= &iph
->iph_root
[dir
];
347 for (chain
= 0; chain
< ipr
->ipr_nchains
; chain
++)
348 mutex_destroy(&(ipr
->ipr_hash
[chain
].hash_lock
));
351 ipsec_polhead_free_table(iph
);
355 * Free the IPsec stack instance.
359 ipsec_stack_fini(netstackid_t stackid
, void *arg
)
361 ipsec_stack_t
*ipss
= (ipsec_stack_t
*)arg
;
363 ipsec_tun_pol_t
*node
;
364 netstack_t
*ns
= ipss
->ipsec_netstack
;
366 ipsec_algtype_t algtype
;
368 ipsec_loader_destroy(ipss
);
370 rw_enter(&ipss
->ipsec_tunnel_policy_lock
, RW_WRITER
);
372 * It's possible we can just ASSERT() the tree is empty. After all,
373 * we aren't called until IP is ready to unload (and presumably all
374 * tunnels have been unplumbed). But we'll play it safe for now, the
375 * loop will just exit immediately if it's empty.
378 while ((node
= (ipsec_tun_pol_t
*)
379 avl_destroy_nodes(&ipss
->ipsec_tunnel_policies
,
381 ITP_REFRELE(node
, ns
);
383 avl_destroy(&ipss
->ipsec_tunnel_policies
);
384 rw_exit(&ipss
->ipsec_tunnel_policy_lock
);
385 rw_destroy(&ipss
->ipsec_tunnel_policy_lock
);
387 ipsec_config_flush(ns
);
389 ipsec_kstat_destroy(ipss
);
391 ip_drop_unregister(&ipss
->ipsec_dropper
);
393 ip_drop_unregister(&ipss
->ipsec_spd_dropper
);
394 ip_drop_destroy(ipss
);
396 * Globals start with ref == 1 to prevent IPPH_REFRELE() from
397 * attempting to free them, hence they should have 1 now.
399 ipsec_polhead_destroy(&ipss
->ipsec_system_policy
);
400 ASSERT(ipss
->ipsec_system_policy
.iph_refs
== 1);
401 ipsec_polhead_destroy(&ipss
->ipsec_inactive_policy
);
402 ASSERT(ipss
->ipsec_inactive_policy
.iph_refs
== 1);
404 for (i
= 0; i
< IPSEC_ACTION_HASH_SIZE
; i
++) {
405 ipsec_action_free_table(ipss
->ipsec_action_hash
[i
].hash_head
);
406 ipss
->ipsec_action_hash
[i
].hash_head
= NULL
;
407 mutex_destroy(&(ipss
->ipsec_action_hash
[i
].hash_lock
));
410 for (i
= 0; i
< ipss
->ipsec_spd_hashsize
; i
++) {
411 ASSERT(ipss
->ipsec_sel_hash
[i
].hash_head
== NULL
);
412 mutex_destroy(&(ipss
->ipsec_sel_hash
[i
].hash_lock
));
415 mutex_enter(&ipss
->ipsec_alg_lock
);
416 for (algtype
= 0; algtype
< IPSEC_NALGTYPES
; algtype
++) {
417 int nalgs
= ipss
->ipsec_nalgs
[algtype
];
419 for (i
= 0; i
< nalgs
; i
++) {
420 if (ipss
->ipsec_alglists
[algtype
][i
] != NULL
)
421 ipsec_alg_unreg(algtype
, i
, ns
);
424 mutex_exit(&ipss
->ipsec_alg_lock
);
425 mutex_destroy(&ipss
->ipsec_alg_lock
);
430 (void) ipsec_free_tables(ipss
);
431 kmem_free(ipss
, sizeof (*ipss
));
435 ipsec_policy_g_destroy(void)
437 kmem_cache_destroy(ipsec_action_cache
);
438 kmem_cache_destroy(ipsec_sel_cache
);
439 kmem_cache_destroy(ipsec_pol_cache
);
441 ipsec_unregister_prov_update();
443 netstack_unregister(NS_IPSEC
);
448 * Free what ipsec_alloc_tables allocated.
449 * Called when table allocation fails to free the table.
452 ipsec_free_tables(ipsec_stack_t
*ipss
)
456 if (ipss
->ipsec_sel_hash
!= NULL
) {
457 for (i
= 0; i
< ipss
->ipsec_spd_hashsize
; i
++) {
458 ASSERT(ipss
->ipsec_sel_hash
[i
].hash_head
== NULL
);
460 kmem_free(ipss
->ipsec_sel_hash
, ipss
->ipsec_spd_hashsize
*
461 sizeof (*ipss
->ipsec_sel_hash
));
462 ipss
->ipsec_sel_hash
= NULL
;
463 ipss
->ipsec_spd_hashsize
= 0;
465 ipsec_polhead_free_table(&ipss
->ipsec_system_policy
);
466 ipsec_polhead_free_table(&ipss
->ipsec_inactive_policy
);
472 * Attempt to allocate the tables in a single policy head.
473 * Return nonzero on failure after cleaning up any work in progress.
476 ipsec_alloc_table(ipsec_policy_head_t
*iph
, int nchains
, int kmflag
,
477 boolean_t global_cleanup
, netstack_t
*ns
)
481 for (dir
= 0; dir
< IPSEC_NTYPES
; dir
++) {
482 ipsec_policy_root_t
*ipr
= &iph
->iph_root
[dir
];
484 ipr
->ipr_nchains
= nchains
;
485 ipr
->ipr_hash
= kmem_zalloc(nchains
*
486 sizeof (ipsec_policy_hash_t
), kmflag
);
487 if (ipr
->ipr_hash
== NULL
)
488 return (global_cleanup
?
489 ipsec_free_tables(ns
->netstack_ipsec
) :
496 * Attempt to allocate the various tables. Return nonzero on failure
497 * after cleaning up any work in progress.
500 ipsec_alloc_tables(int kmflag
, netstack_t
*ns
)
503 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
505 error
= ipsec_alloc_table(&ipss
->ipsec_system_policy
,
506 ipss
->ipsec_spd_hashsize
, kmflag
, B_TRUE
, ns
);
510 error
= ipsec_alloc_table(&ipss
->ipsec_inactive_policy
,
511 ipss
->ipsec_spd_hashsize
, kmflag
, B_TRUE
, ns
);
515 ipss
->ipsec_sel_hash
= kmem_zalloc(ipss
->ipsec_spd_hashsize
*
516 sizeof (*ipss
->ipsec_sel_hash
), kmflag
);
518 if (ipss
->ipsec_sel_hash
== NULL
)
519 return (ipsec_free_tables(ipss
));
525 * After table allocation, initialize a policy head.
528 ipsec_polhead_init(ipsec_policy_head_t
*iph
, int nchains
)
532 rw_init(&iph
->iph_lock
, NULL
, RW_DEFAULT
, NULL
);
533 avl_create(&iph
->iph_rulebyid
, ipsec_policy_cmpbyid
,
534 sizeof (ipsec_policy_t
), offsetof(ipsec_policy_t
, ipsp_byid
));
536 for (dir
= 0; dir
< IPSEC_NTYPES
; dir
++) {
537 ipsec_policy_root_t
*ipr
= &iph
->iph_root
[dir
];
538 ipr
->ipr_nchains
= nchains
;
540 for (chain
= 0; chain
< nchains
; chain
++) {
541 mutex_init(&(ipr
->ipr_hash
[chain
].hash_lock
),
542 NULL
, MUTEX_DEFAULT
, NULL
);
548 ipsec_kstat_init(ipsec_stack_t
*ipss
)
550 ipss
->ipsec_ksp
= kstat_create_netstack("ip", 0, "ipsec_stat", "net",
551 KSTAT_TYPE_NAMED
, sizeof (ipsec_kstats_t
) / sizeof (kstat_named_t
),
552 KSTAT_FLAG_PERSISTENT
, ipss
->ipsec_netstack
->netstack_stackid
);
554 if (ipss
->ipsec_ksp
== NULL
|| ipss
->ipsec_ksp
->ks_data
== NULL
)
557 ipss
->ipsec_kstats
= ipss
->ipsec_ksp
->ks_data
;
559 #define KI(x) kstat_named_init(&ipss->ipsec_kstats->x, #x, KSTAT_DATA_UINT64)
560 KI(esp_stat_in_requests
);
561 KI(esp_stat_in_discards
);
562 KI(esp_stat_lookup_failure
);
563 KI(ah_stat_in_requests
);
564 KI(ah_stat_in_discards
);
565 KI(ah_stat_lookup_failure
);
566 KI(sadb_acquire_maxpackets
);
567 KI(sadb_acquire_qhiwater
);
570 kstat_install(ipss
->ipsec_ksp
);
575 ipsec_kstat_destroy(ipsec_stack_t
*ipss
)
577 kstat_delete_netstack(ipss
->ipsec_ksp
,
578 ipss
->ipsec_netstack
->netstack_stackid
);
579 ipss
->ipsec_kstats
= NULL
;
584 * Initialize the IPsec stack instance.
588 ipsec_stack_init(netstackid_t stackid
, netstack_t
*ns
)
593 ipss
= (ipsec_stack_t
*)kmem_zalloc(sizeof (*ipss
), KM_SLEEP
);
594 ipss
->ipsec_netstack
= ns
;
597 * FIXME: netstack_ipsec is used by some of the routines we call
598 * below, but it isn't set until this routine returns.
599 * Either we introduce optional xxx_stack_alloc() functions
600 * that will be called by the netstack framework before xxx_stack_init,
601 * or we switch spd.c and sadb.c to operate on ipsec_stack_t
602 * (latter has some include file order issues for sadb.h, but makes
603 * sense if we merge some of the ipsec related stack_t's together.
605 ns
->netstack_ipsec
= ipss
;
608 * Make two attempts to allocate policy hash tables; try it at
609 * the "preferred" size (may be set in /etc/system) first,
610 * then fall back to the default size.
612 ipss
->ipsec_spd_hashsize
= (ipsec_spd_hashsize
== 0) ?
613 IPSEC_SPDHASH_DEFAULT
: ipsec_spd_hashsize
;
615 if (ipsec_alloc_tables(KM_NOSLEEP
, ns
) != 0) {
617 "Unable to allocate %d entry IPsec policy hash table",
618 ipss
->ipsec_spd_hashsize
);
619 ipss
->ipsec_spd_hashsize
= IPSEC_SPDHASH_DEFAULT
;
620 cmn_err(CE_WARN
, "Falling back to %d entries",
621 ipss
->ipsec_spd_hashsize
);
622 (void) ipsec_alloc_tables(KM_SLEEP
, ns
);
625 /* Just set a default for tunnels. */
626 ipss
->ipsec_tun_spd_hashsize
= (tun_spd_hashsize
== 0) ?
627 TUN_SPDHASH_DEFAULT
: tun_spd_hashsize
;
631 * Globals need ref == 1 to prevent IPPH_REFRELE() from attempting
634 ipss
->ipsec_system_policy
.iph_refs
= 1;
635 ipss
->ipsec_inactive_policy
.iph_refs
= 1;
636 ipsec_polhead_init(&ipss
->ipsec_system_policy
,
637 ipss
->ipsec_spd_hashsize
);
638 ipsec_polhead_init(&ipss
->ipsec_inactive_policy
,
639 ipss
->ipsec_spd_hashsize
);
640 rw_init(&ipss
->ipsec_tunnel_policy_lock
, NULL
, RW_DEFAULT
, NULL
);
641 avl_create(&ipss
->ipsec_tunnel_policies
, tunnel_compare
,
642 sizeof (ipsec_tun_pol_t
), 0);
644 ipss
->ipsec_next_policy_index
= 1;
646 rw_init(&ipss
->ipsec_system_policy
.iph_lock
, NULL
, RW_DEFAULT
, NULL
);
647 rw_init(&ipss
->ipsec_inactive_policy
.iph_lock
, NULL
, RW_DEFAULT
, NULL
);
649 for (i
= 0; i
< IPSEC_ACTION_HASH_SIZE
; i
++)
650 mutex_init(&(ipss
->ipsec_action_hash
[i
].hash_lock
),
651 NULL
, MUTEX_DEFAULT
, NULL
);
653 for (i
= 0; i
< ipss
->ipsec_spd_hashsize
; i
++)
654 mutex_init(&(ipss
->ipsec_sel_hash
[i
].hash_lock
),
655 NULL
, MUTEX_DEFAULT
, NULL
);
657 mutex_init(&ipss
->ipsec_alg_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
658 for (i
= 0; i
< IPSEC_NALGTYPES
; i
++) {
659 ipss
->ipsec_nalgs
[i
] = 0;
663 ip_drop_register(&ipss
->ipsec_spd_dropper
, "IPsec SPD");
665 /* IP's IPsec code calls the packet dropper */
666 ip_drop_register(&ipss
->ipsec_dropper
, "IP IPsec processing");
668 (void) ipsec_kstat_init(ipss
);
670 ipsec_loader_init(ipss
);
671 ipsec_loader_start(ipss
);
676 /* Global across all stack instances */
678 ipsec_policy_g_init(void)
680 ipsec_action_cache
= kmem_cache_create("ipsec_actions",
681 sizeof (ipsec_action_t
), _POINTER_ALIGNMENT
, NULL
, NULL
,
682 ipsec_action_reclaim
, NULL
, NULL
, 0);
683 ipsec_sel_cache
= kmem_cache_create("ipsec_selectors",
684 sizeof (ipsec_sel_t
), _POINTER_ALIGNMENT
, NULL
, NULL
,
685 NULL
, NULL
, NULL
, 0);
686 ipsec_pol_cache
= kmem_cache_create("ipsec_policy",
687 sizeof (ipsec_policy_t
), _POINTER_ALIGNMENT
, NULL
, NULL
,
688 NULL
, NULL
, NULL
, 0);
691 * We want to be informed each time a stack is created or
692 * destroyed in the kernel, so we can maintain the
693 * set of ipsec_stack_t's.
695 netstack_register(NS_IPSEC
, ipsec_stack_init
, NULL
, ipsec_stack_fini
);
699 * Sort algorithm lists.
701 * I may need to split this based on
702 * authentication/encryption, and I may wish to have an administrator
703 * configure this list. Hold on to some NDD variables...
705 * XXX For now, sort on minimum key size (GAG!). While minimum key size is
706 * not the ideal metric, it's the only quantifiable measure available.
707 * We need a better metric for sorting algorithms by preference.
710 alg_insert_sortlist(enum ipsec_algtype at
, uint8_t algid
, netstack_t
*ns
)
712 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
713 ipsec_alginfo_t
*ai
= ipss
->ipsec_alglists
[at
][algid
];
714 uint8_t holder
, swap
;
716 uint_t count
= ipss
->ipsec_nalgs
[at
];
718 ASSERT(algid
== ai
->alg_id
);
720 ASSERT(MUTEX_HELD(&ipss
->ipsec_alg_lock
));
724 for (i
= 0; i
< count
- 1; i
++) {
725 ipsec_alginfo_t
*alt
;
727 alt
= ipss
->ipsec_alglists
[at
][ipss
->ipsec_sortlist
[at
][i
]];
729 * If you want to give precedence to newly added algs,
730 * add the = in the > comparison.
732 if ((holder
!= algid
) || (ai
->alg_minbits
> alt
->alg_minbits
)) {
733 /* Swap sortlist[i] and holder. */
734 swap
= ipss
->ipsec_sortlist
[at
][i
];
735 ipss
->ipsec_sortlist
[at
][i
] = holder
;
738 } /* Else just continue. */
741 /* Store holder in last slot. */
742 ipss
->ipsec_sortlist
[at
][i
] = holder
;
746 * Remove an algorithm from a sorted algorithm list.
747 * This should be considerably easier, even with complex sorting.
750 alg_remove_sortlist(enum ipsec_algtype at
, uint8_t algid
, netstack_t
*ns
)
752 boolean_t copyback
= B_FALSE
;
754 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
755 int newcount
= ipss
->ipsec_nalgs
[at
];
757 ASSERT(MUTEX_HELD(&ipss
->ipsec_alg_lock
));
759 for (i
= 0; i
<= newcount
; i
++) {
761 ipss
->ipsec_sortlist
[at
][i
-1] =
762 ipss
->ipsec_sortlist
[at
][i
];
763 } else if (ipss
->ipsec_sortlist
[at
][i
] == algid
) {
770 * Add the specified algorithm to the algorithm tables.
771 * Must be called while holding the algorithm table writer lock.
774 ipsec_alg_reg(ipsec_algtype_t algtype
, ipsec_alginfo_t
*alg
, netstack_t
*ns
)
776 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
778 ASSERT(MUTEX_HELD(&ipss
->ipsec_alg_lock
));
780 ASSERT(ipss
->ipsec_alglists
[algtype
][alg
->alg_id
] == NULL
);
781 ipsec_alg_fix_min_max(alg
, algtype
, ns
);
782 ipss
->ipsec_alglists
[algtype
][alg
->alg_id
] = alg
;
784 ipss
->ipsec_nalgs
[algtype
]++;
785 alg_insert_sortlist(algtype
, alg
->alg_id
, ns
);
789 * Remove the specified algorithm from the algorithm tables.
790 * Must be called while holding the algorithm table writer lock.
793 ipsec_alg_unreg(ipsec_algtype_t algtype
, uint8_t algid
, netstack_t
*ns
)
795 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
797 ASSERT(MUTEX_HELD(&ipss
->ipsec_alg_lock
));
799 ASSERT(ipss
->ipsec_alglists
[algtype
][algid
] != NULL
);
800 ipsec_alg_free(ipss
->ipsec_alglists
[algtype
][algid
]);
801 ipss
->ipsec_alglists
[algtype
][algid
] = NULL
;
803 ipss
->ipsec_nalgs
[algtype
]--;
804 alg_remove_sortlist(algtype
, algid
, ns
);
808 * Hooks for spdsock to get a grip on system policy.
811 ipsec_policy_head_t
*
812 ipsec_system_policy(netstack_t
*ns
)
814 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
815 ipsec_policy_head_t
*h
= &ipss
->ipsec_system_policy
;
821 ipsec_policy_head_t
*
822 ipsec_inactive_policy(netstack_t
*ns
)
824 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
825 ipsec_policy_head_t
*h
= &ipss
->ipsec_inactive_policy
;
832 * Lock inactive policy, then active policy, then exchange policy root
836 ipsec_swap_policy(ipsec_policy_head_t
*active
, ipsec_policy_head_t
*inactive
,
842 rw_enter(&inactive
->iph_lock
, RW_WRITER
);
843 rw_enter(&active
->iph_lock
, RW_WRITER
);
845 r1
= active
->iph_rulebyid
;
846 r2
= inactive
->iph_rulebyid
;
847 active
->iph_rulebyid
= r2
;
848 inactive
->iph_rulebyid
= r1
;
850 for (dir
= 0; dir
< IPSEC_NTYPES
; dir
++) {
851 ipsec_policy_hash_t
*h1
, *h2
;
853 h1
= active
->iph_root
[dir
].ipr_hash
;
854 h2
= inactive
->iph_root
[dir
].ipr_hash
;
855 active
->iph_root
[dir
].ipr_hash
= h2
;
856 inactive
->iph_root
[dir
].ipr_hash
= h1
;
858 for (af
= 0; af
< IPSEC_NAF
; af
++) {
859 ipsec_policy_t
*t1
, *t2
;
861 t1
= active
->iph_root
[dir
].ipr_nonhash
[af
];
862 t2
= inactive
->iph_root
[dir
].ipr_nonhash
[af
];
863 active
->iph_root
[dir
].ipr_nonhash
[af
] = t2
;
864 inactive
->iph_root
[dir
].ipr_nonhash
[af
] = t1
;
866 t1
->ipsp_hash
.hash_pp
=
867 &(inactive
->iph_root
[dir
].ipr_nonhash
[af
]);
870 t2
->ipsp_hash
.hash_pp
=
871 &(active
->iph_root
[dir
].ipr_nonhash
[af
]);
878 ipsec_update_present_flags(ns
->netstack_ipsec
);
879 rw_exit(&active
->iph_lock
);
880 rw_exit(&inactive
->iph_lock
);
884 * Swap global policy primary/secondary.
887 ipsec_swap_global_policy(netstack_t
*ns
)
889 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
891 ipsec_swap_policy(&ipss
->ipsec_system_policy
,
892 &ipss
->ipsec_inactive_policy
, ns
);
896 * Clone one policy rule..
898 static ipsec_policy_t
*
899 ipsec_copy_policy(const ipsec_policy_t
*src
)
901 ipsec_policy_t
*dst
= kmem_cache_alloc(ipsec_pol_cache
, KM_NOSLEEP
);
907 * Adjust refcounts of cloned state.
909 IPACT_REFHOLD(src
->ipsp_act
);
910 src
->ipsp_sel
->ipsl_refs
++;
912 HASH_NULL(dst
, ipsp_hash
);
913 dst
->ipsp_netstack
= src
->ipsp_netstack
;
915 dst
->ipsp_sel
= src
->ipsp_sel
;
916 dst
->ipsp_act
= src
->ipsp_act
;
917 dst
->ipsp_prio
= src
->ipsp_prio
;
918 dst
->ipsp_index
= src
->ipsp_index
;
924 ipsec_insert_always(avl_tree_t
*tree
, void *new_node
)
929 node
= avl_find(tree
, new_node
, &where
);
930 ASSERT(node
== NULL
);
931 avl_insert(tree
, new_node
, where
);
936 ipsec_copy_chain(ipsec_policy_head_t
*dph
, ipsec_policy_t
*src
,
937 ipsec_policy_t
**dstp
)
939 for (; src
!= NULL
; src
= src
->ipsp_hash
.hash_next
) {
940 ipsec_policy_t
*dst
= ipsec_copy_policy(src
);
944 HASHLIST_INSERT(dst
, ipsp_hash
, *dstp
);
945 ipsec_insert_always(&dph
->iph_rulebyid
, dst
);
953 * Make one policy head look exactly like another.
955 * As with ipsec_swap_policy, we lock the destination policy head first, then
956 * the source policy head. Note that we only need to read-lock the source
957 * policy head as we are not changing it.
960 ipsec_copy_polhead(ipsec_policy_head_t
*sph
, ipsec_policy_head_t
*dph
,
963 int af
, dir
, chain
, nchains
;
965 rw_enter(&dph
->iph_lock
, RW_WRITER
);
967 ipsec_polhead_flush(dph
, ns
);
969 rw_enter(&sph
->iph_lock
, RW_READER
);
971 for (dir
= 0; dir
< IPSEC_NTYPES
; dir
++) {
972 ipsec_policy_root_t
*dpr
= &dph
->iph_root
[dir
];
973 ipsec_policy_root_t
*spr
= &sph
->iph_root
[dir
];
974 nchains
= dpr
->ipr_nchains
;
976 ASSERT(dpr
->ipr_nchains
== spr
->ipr_nchains
);
978 for (af
= 0; af
< IPSEC_NAF
; af
++) {
979 if (ipsec_copy_chain(dph
, spr
->ipr_nonhash
[af
],
980 &dpr
->ipr_nonhash
[af
]))
984 for (chain
= 0; chain
< nchains
; chain
++) {
985 if (ipsec_copy_chain(dph
,
986 spr
->ipr_hash
[chain
].hash_head
,
987 &dpr
->ipr_hash
[chain
].hash_head
))
994 rw_exit(&sph
->iph_lock
);
995 rw_exit(&dph
->iph_lock
);
999 ipsec_polhead_flush(dph
, ns
);
1000 rw_exit(&sph
->iph_lock
);
1001 rw_exit(&dph
->iph_lock
);
1006 * Clone currently active policy to the inactive policy list.
1009 ipsec_clone_system_policy(netstack_t
*ns
)
1011 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
1013 return (ipsec_copy_polhead(&ipss
->ipsec_system_policy
,
1014 &ipss
->ipsec_inactive_policy
, ns
));
1018 * Extract the string from ipsec_policy_failure_msgs[type] and
1023 ipsec_log_policy_failure(int type
, char *func_name
, ipha_t
*ipha
, ip6_t
*ip6h
,
1024 boolean_t secure
, netstack_t
*ns
)
1026 char sbuf
[INET6_ADDRSTRLEN
];
1027 char dbuf
[INET6_ADDRSTRLEN
];
1030 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
1032 ASSERT((ipha
== NULL
&& ip6h
!= NULL
) ||
1033 (ip6h
== NULL
&& ipha
!= NULL
));
1036 s
= inet_ntop(AF_INET
, &ipha
->ipha_src
, sbuf
, sizeof (sbuf
));
1037 d
= inet_ntop(AF_INET
, &ipha
->ipha_dst
, dbuf
, sizeof (dbuf
));
1039 s
= inet_ntop(AF_INET6
, &ip6h
->ip6_src
, sbuf
, sizeof (sbuf
));
1040 d
= inet_ntop(AF_INET6
, &ip6h
->ip6_dst
, dbuf
, sizeof (dbuf
));
1044 /* Always bump the policy failure counter. */
1045 ipss
->ipsec_policy_failure_count
[type
]++;
1047 ipsec_rl_strlog(ns
, IP_MOD_ID
, 0, 0, SL_ERROR
|SL_WARN
|SL_CONSOLE
,
1048 ipsec_policy_failure_msgs
[type
], func_name
,
1049 (secure
? "secure" : "not secure"), s
, d
);
1053 * Rate-limiting front-end to strlog() for AH and ESP. Uses the ndd variables
1054 * in /dev/ip and the same rate-limiting clock so that there's a single
1055 * knob to turn to throttle the rate of messages.
1058 ipsec_rl_strlog(netstack_t
*ns
, short mid
, short sid
, char level
, ushort_t sl
,
1062 hrtime_t current
= gethrtime();
1063 ip_stack_t
*ipst
= ns
->netstack_ip
;
1064 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
1068 * Throttle logging to stop syslog from being swamped. If variable
1069 * 'ipsec_policy_log_interval' is zero, don't log any messages at
1070 * all, otherwise log only one message every 'ipsec_policy_log_interval'
1071 * msec. Convert interval (in msec) to hrtime (in nsec).
1074 if (ipst
->ips_ipsec_policy_log_interval
) {
1075 if (ipss
->ipsec_policy_failure_last
+
1076 MSEC2NSEC(ipst
->ips_ipsec_policy_log_interval
) <= current
) {
1078 (void) vstrlog(mid
, sid
, level
, sl
, fmt
, adx
);
1080 ipss
->ipsec_policy_failure_last
= current
;
1086 ipsec_config_flush(netstack_t
*ns
)
1088 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
1090 rw_enter(&ipss
->ipsec_system_policy
.iph_lock
, RW_WRITER
);
1091 ipsec_polhead_flush(&ipss
->ipsec_system_policy
, ns
);
1092 ipss
->ipsec_next_policy_index
= 1;
1093 rw_exit(&ipss
->ipsec_system_policy
.iph_lock
);
1094 ipsec_action_reclaim_stack(ipss
);
1098 * Clip a policy's min/max keybits vs. the capabilities of the
1102 act_alg_adjust(uint_t algtype
, uint_t algid
,
1103 uint16_t *minbits
, uint16_t *maxbits
, netstack_t
*ns
)
1105 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
1106 ipsec_alginfo_t
*algp
= ipss
->ipsec_alglists
[algtype
][algid
];
1110 * If passed-in minbits is zero, we assume the caller trusts
1111 * us with setting the minimum key size. We pick the
1112 * algorithms DEFAULT key size for the minimum in this case.
1114 if (*minbits
== 0) {
1115 *minbits
= algp
->alg_default_bits
;
1116 ASSERT(*minbits
>= algp
->alg_minbits
);
1118 *minbits
= MAX(MIN(*minbits
, algp
->alg_maxbits
),
1122 *maxbits
= algp
->alg_maxbits
;
1124 *maxbits
= MIN(MAX(*maxbits
, algp
->alg_minbits
),
1126 ASSERT(*minbits
<= *maxbits
);
1134 * Check an action's requested algorithms against the algorithms currently
1135 * loaded in the system.
1138 ipsec_check_action(ipsec_act_t
*act
, int *diag
, netstack_t
*ns
)
1141 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
1143 ipp
= &act
->ipa_apply
;
1145 if (ipp
->ipp_use_ah
&&
1146 ipss
->ipsec_alglists
[IPSEC_ALG_AUTH
][ipp
->ipp_auth_alg
] == NULL
) {
1147 *diag
= SPD_DIAGNOSTIC_UNSUPP_AH_ALG
;
1150 if (ipp
->ipp_use_espa
&&
1151 ipss
->ipsec_alglists
[IPSEC_ALG_AUTH
][ipp
->ipp_esp_auth_alg
] ==
1153 *diag
= SPD_DIAGNOSTIC_UNSUPP_ESP_AUTH_ALG
;
1156 if (ipp
->ipp_use_esp
&&
1157 ipss
->ipsec_alglists
[IPSEC_ALG_ENCR
][ipp
->ipp_encr_alg
] == NULL
) {
1158 *diag
= SPD_DIAGNOSTIC_UNSUPP_ESP_ENCR_ALG
;
1162 act_alg_adjust(IPSEC_ALG_AUTH
, ipp
->ipp_auth_alg
,
1163 &ipp
->ipp_ah_minbits
, &ipp
->ipp_ah_maxbits
, ns
);
1164 act_alg_adjust(IPSEC_ALG_AUTH
, ipp
->ipp_esp_auth_alg
,
1165 &ipp
->ipp_espa_minbits
, &ipp
->ipp_espa_maxbits
, ns
);
1166 act_alg_adjust(IPSEC_ALG_ENCR
, ipp
->ipp_encr_alg
,
1167 &ipp
->ipp_espe_minbits
, &ipp
->ipp_espe_maxbits
, ns
);
1169 if (ipp
->ipp_ah_minbits
> ipp
->ipp_ah_maxbits
) {
1170 *diag
= SPD_DIAGNOSTIC_UNSUPP_AH_KEYSIZE
;
1173 if (ipp
->ipp_espa_minbits
> ipp
->ipp_espa_maxbits
) {
1174 *diag
= SPD_DIAGNOSTIC_UNSUPP_ESP_AUTH_KEYSIZE
;
1177 if (ipp
->ipp_espe_minbits
> ipp
->ipp_espe_maxbits
) {
1178 *diag
= SPD_DIAGNOSTIC_UNSUPP_ESP_ENCR_KEYSIZE
;
1181 /* TODO: sanity check lifetimes */
1186 * Set up a single action during wildcard expansion..
1189 ipsec_setup_act(ipsec_act_t
*outact
, ipsec_act_t
*act
,
1190 uint_t auth_alg
, uint_t encr_alg
, uint_t eauth_alg
, netstack_t
*ns
)
1195 ipp
= &outact
->ipa_apply
;
1196 ipp
->ipp_auth_alg
= (uint8_t)auth_alg
;
1197 ipp
->ipp_encr_alg
= (uint8_t)encr_alg
;
1198 ipp
->ipp_esp_auth_alg
= (uint8_t)eauth_alg
;
1200 act_alg_adjust(IPSEC_ALG_AUTH
, auth_alg
,
1201 &ipp
->ipp_ah_minbits
, &ipp
->ipp_ah_maxbits
, ns
);
1202 act_alg_adjust(IPSEC_ALG_AUTH
, eauth_alg
,
1203 &ipp
->ipp_espa_minbits
, &ipp
->ipp_espa_maxbits
, ns
);
1204 act_alg_adjust(IPSEC_ALG_ENCR
, encr_alg
,
1205 &ipp
->ipp_espe_minbits
, &ipp
->ipp_espe_maxbits
, ns
);
1209 * combinatoric expansion time: expand a wildcarded action into an
1210 * array of wildcarded actions; we return the exploded action list,
1211 * and return a count in *nact (output only).
1213 static ipsec_act_t
*
1214 ipsec_act_wildcard_expand(ipsec_act_t
*act
, uint_t
*nact
, netstack_t
*ns
)
1216 boolean_t use_ah
, use_esp
, use_espa
;
1217 boolean_t wild_auth
, wild_encr
, wild_eauth
;
1218 uint_t auth_alg
, auth_idx
, auth_min
, auth_max
;
1219 uint_t eauth_alg
, eauth_idx
, eauth_min
, eauth_max
;
1220 uint_t encr_alg
, encr_idx
, encr_min
, encr_max
;
1221 uint_t action_count
, ai
;
1222 ipsec_act_t
*outact
;
1223 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
1225 if (act
->ipa_type
!= IPSEC_ACT_APPLY
) {
1226 outact
= kmem_alloc(sizeof (*act
), KM_NOSLEEP
);
1229 bcopy(act
, outact
, sizeof (*act
));
1233 * compute the combinatoric explosion..
1235 * we assume a request for encr if esp_req is PREF_REQUIRED
1236 * we assume a request for ah auth if ah_req is PREF_REQUIRED.
1237 * we assume a request for esp auth if !ah and esp_req is PREF_REQUIRED
1240 use_ah
= act
->ipa_apply
.ipp_use_ah
;
1241 use_esp
= act
->ipa_apply
.ipp_use_esp
;
1242 use_espa
= act
->ipa_apply
.ipp_use_espa
;
1243 auth_alg
= act
->ipa_apply
.ipp_auth_alg
;
1244 eauth_alg
= act
->ipa_apply
.ipp_esp_auth_alg
;
1245 encr_alg
= act
->ipa_apply
.ipp_encr_alg
;
1247 wild_auth
= use_ah
&& (auth_alg
== 0);
1248 wild_eauth
= use_espa
&& (eauth_alg
== 0);
1249 wild_encr
= use_esp
&& (encr_alg
== 0);
1252 auth_min
= auth_max
= auth_alg
;
1253 eauth_min
= eauth_max
= eauth_alg
;
1254 encr_min
= encr_max
= encr_alg
;
1257 * set up for explosion.. for each dimension, expand output
1258 * size by the explosion factor.
1260 * Don't include the "any" algorithms, if defined, as no
1261 * kernel policies should be set for these algorithms.
1264 #define SET_EXP_MINMAX(type, wild, alg, min, max, ipss) \
1266 int nalgs = ipss->ipsec_nalgs[type]; \
1267 if (ipss->ipsec_alglists[type][alg] != NULL) \
1269 action_count *= nalgs; \
1271 max = ipss->ipsec_nalgs[type] - 1; \
1274 SET_EXP_MINMAX(IPSEC_ALG_AUTH
, wild_auth
, SADB_AALG_NONE
,
1275 auth_min
, auth_max
, ipss
);
1276 SET_EXP_MINMAX(IPSEC_ALG_AUTH
, wild_eauth
, SADB_AALG_NONE
,
1277 eauth_min
, eauth_max
, ipss
);
1278 SET_EXP_MINMAX(IPSEC_ALG_ENCR
, wild_encr
, SADB_EALG_NONE
,
1279 encr_min
, encr_max
, ipss
);
1281 #undef SET_EXP_MINMAX
1284 * ok, allocate the whole mess..
1287 outact
= kmem_alloc(sizeof (*outact
) * action_count
, KM_NOSLEEP
);
1292 * Now compute all combinations. Note that non-wildcarded
1293 * dimensions just get a single value from auth_min, while
1294 * wildcarded dimensions indirect through the sortlist.
1296 * We do encryption outermost since, at this time, there's
1297 * greater difference in security and performance between
1298 * encryption algorithms vs. authentication algorithms.
1303 #define WHICH_ALG(type, wild, idx, ipss) \
1304 ((wild)?(ipss->ipsec_sortlist[type][idx]):(idx))
1306 for (encr_idx
= encr_min
; encr_idx
<= encr_max
; encr_idx
++) {
1307 encr_alg
= WHICH_ALG(IPSEC_ALG_ENCR
, wild_encr
, encr_idx
, ipss
);
1308 if (wild_encr
&& encr_alg
== SADB_EALG_NONE
)
1310 for (auth_idx
= auth_min
; auth_idx
<= auth_max
; auth_idx
++) {
1311 auth_alg
= WHICH_ALG(IPSEC_ALG_AUTH
, wild_auth
,
1313 if (wild_auth
&& auth_alg
== SADB_AALG_NONE
)
1315 for (eauth_idx
= eauth_min
; eauth_idx
<= eauth_max
;
1317 eauth_alg
= WHICH_ALG(IPSEC_ALG_AUTH
,
1318 wild_eauth
, eauth_idx
, ipss
);
1319 if (wild_eauth
&& eauth_alg
== SADB_AALG_NONE
)
1322 ipsec_setup_act(&outact
[ai
], act
,
1323 auth_alg
, encr_alg
, eauth_alg
, ns
);
1331 ASSERT(ai
== action_count
);
1332 *nact
= action_count
;
1337 * Extract the parts of an ipsec_prot_t from an old-style ipsec_req_t.
1340 ipsec_prot_from_req(const ipsec_req_t
*req
, ipsec_prot_t
*ipp
)
1342 bzero(ipp
, sizeof (*ipp
));
1344 * ipp_use_* are bitfields. Look at "!!" in the following as a
1345 * "boolean canonicalization" operator.
1347 ipp
->ipp_use_ah
= !!(req
->ipsr_ah_req
& IPSEC_PREF_REQUIRED
);
1348 ipp
->ipp_use_esp
= !!(req
->ipsr_esp_req
& IPSEC_PREF_REQUIRED
);
1349 ipp
->ipp_use_espa
= !!(req
->ipsr_esp_auth_alg
);
1350 ipp
->ipp_use_se
= !!(req
->ipsr_self_encap_req
& IPSEC_PREF_REQUIRED
);
1351 ipp
->ipp_use_unique
= !!((req
->ipsr_ah_req
|req
->ipsr_esp_req
) &
1353 ipp
->ipp_encr_alg
= req
->ipsr_esp_alg
;
1355 * SADB_AALG_ANY is a placeholder to distinguish "any" from
1356 * "none" above. If auth is required, as determined above,
1357 * SADB_AALG_ANY becomes 0, which is the representation
1358 * of "any" and "none" in PF_KEY v2.
1360 ipp
->ipp_auth_alg
= (req
->ipsr_auth_alg
!= SADB_AALG_ANY
) ?
1361 req
->ipsr_auth_alg
: 0;
1362 ipp
->ipp_esp_auth_alg
= (req
->ipsr_esp_auth_alg
!= SADB_AALG_ANY
) ?
1363 req
->ipsr_esp_auth_alg
: 0;
1367 * Extract a new-style action from a request.
1370 ipsec_actvec_from_req(const ipsec_req_t
*req
, ipsec_act_t
**actp
, uint_t
*nactp
,
1373 struct ipsec_act act
;
1375 bzero(&act
, sizeof (act
));
1376 if ((req
->ipsr_ah_req
& IPSEC_PREF_NEVER
) &&
1377 (req
->ipsr_esp_req
& IPSEC_PREF_NEVER
)) {
1378 act
.ipa_type
= IPSEC_ACT_BYPASS
;
1380 act
.ipa_type
= IPSEC_ACT_APPLY
;
1381 ipsec_prot_from_req(req
, &act
.ipa_apply
);
1383 *actp
= ipsec_act_wildcard_expand(&act
, nactp
, ns
);
1387 * Convert a new-style "prot" back to an ipsec_req_t (more backwards compat).
1388 * We assume caller has already zero'ed *req for us.
1391 ipsec_req_from_prot(ipsec_prot_t
*ipp
, ipsec_req_t
*req
)
1393 req
->ipsr_esp_alg
= ipp
->ipp_encr_alg
;
1394 req
->ipsr_auth_alg
= ipp
->ipp_auth_alg
;
1395 req
->ipsr_esp_auth_alg
= ipp
->ipp_esp_auth_alg
;
1397 if (ipp
->ipp_use_unique
) {
1398 req
->ipsr_ah_req
|= IPSEC_PREF_UNIQUE
;
1399 req
->ipsr_esp_req
|= IPSEC_PREF_UNIQUE
;
1401 if (ipp
->ipp_use_se
)
1402 req
->ipsr_self_encap_req
|= IPSEC_PREF_REQUIRED
;
1403 if (ipp
->ipp_use_ah
)
1404 req
->ipsr_ah_req
|= IPSEC_PREF_REQUIRED
;
1405 if (ipp
->ipp_use_esp
)
1406 req
->ipsr_esp_req
|= IPSEC_PREF_REQUIRED
;
1407 return (sizeof (*req
));
1411 * Convert a new-style action back to an ipsec_req_t (more backwards compat).
1412 * We assume caller has already zero'ed *req for us.
1415 ipsec_req_from_act(ipsec_action_t
*ap
, ipsec_req_t
*req
)
1417 switch (ap
->ipa_act
.ipa_type
) {
1418 case IPSEC_ACT_BYPASS
:
1419 req
->ipsr_ah_req
= IPSEC_PREF_NEVER
;
1420 req
->ipsr_esp_req
= IPSEC_PREF_NEVER
;
1421 return (sizeof (*req
));
1422 case IPSEC_ACT_APPLY
:
1423 return (ipsec_req_from_prot(&ap
->ipa_act
.ipa_apply
, req
));
1425 return (sizeof (*req
));
1429 * Convert a new-style action back to an ipsec_req_t (more backwards compat).
1430 * We assume caller has already zero'ed *req for us.
1433 ipsec_req_from_head(ipsec_policy_head_t
*ph
, ipsec_req_t
*req
, int af
)
1438 * FULL-PERSOCK: consult hash table, too?
1440 for (p
= ph
->iph_root
[IPSEC_INBOUND
].ipr_nonhash
[af
];
1442 p
= p
->ipsp_hash
.hash_next
) {
1443 if ((p
->ipsp_sel
->ipsl_key
.ipsl_valid
& IPSL_WILDCARD
) == 0)
1444 return (ipsec_req_from_act(p
->ipsp_act
, req
));
1446 return (sizeof (*req
));
1450 * Based on per-socket or latched policy, convert to an appropriate
1451 * IP_SEC_OPT ipsec_req_t for the socket option; return size so we can
1452 * be tail-called from ip.
1455 ipsec_req_from_conn(conn_t
*connp
, ipsec_req_t
*req
, int af
)
1458 int rv
= sizeof (ipsec_req_t
);
1460 bzero(req
, sizeof (*req
));
1462 ASSERT(MUTEX_HELD(&connp
->conn_lock
));
1463 ipl
= connp
->conn_latch
;
1466 * Find appropriate policy. First choice is latched action;
1467 * failing that, see latched policy; failing that,
1468 * look at configured policy.
1471 if (connp
->conn_latch_in_action
!= NULL
) {
1472 rv
= ipsec_req_from_act(connp
->conn_latch_in_action
,
1476 if (connp
->conn_latch_in_policy
!= NULL
) {
1477 rv
= ipsec_req_from_act(
1478 connp
->conn_latch_in_policy
->ipsp_act
, req
);
1482 if (connp
->conn_policy
!= NULL
)
1483 rv
= ipsec_req_from_head(connp
->conn_policy
, req
, af
);
1489 ipsec_actvec_free(ipsec_act_t
*act
, uint_t nact
)
1491 kmem_free(act
, nact
* sizeof (*act
));
1495 * Consumes a reference to ipsp.
1498 ipsec_check_loopback_policy(mblk_t
*data_mp
, ip_recv_attr_t
*ira
,
1499 ipsec_policy_t
*ipsp
)
1501 if (!(ira
->ira_flags
& IRAF_IPSEC_SECURE
))
1504 ASSERT(ira
->ira_flags
& IRAF_LOOPBACK
);
1506 IPPOL_REFRELE(ipsp
);
1509 * We should do an actual policy check here. Revisit this
1510 * when we revisit the IPsec API. (And pass a conn_t in when we
1518 * Check that packet's inbound ports & proto match the selectors
1519 * expected by the SAs it traversed on the way in.
1522 ipsec_check_ipsecin_unique(ip_recv_attr_t
*ira
, const char **reason
,
1523 kstat_named_t
**counter
, uint64_t pkt_unique
, netstack_t
*ns
)
1525 uint64_t ah_mask
, esp_mask
;
1528 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
1530 ASSERT(ira
->ira_flags
& IRAF_IPSEC_SECURE
);
1531 ASSERT(!(ira
->ira_flags
& IRAF_LOOPBACK
));
1533 ah_assoc
= ira
->ira_ipsec_ah_sa
;
1534 esp_assoc
= ira
->ira_ipsec_esp_sa
;
1535 ASSERT((ah_assoc
!= NULL
) || (esp_assoc
!= NULL
));
1537 ah_mask
= (ah_assoc
!= NULL
) ? ah_assoc
->ipsa_unique_mask
: 0;
1538 esp_mask
= (esp_assoc
!= NULL
) ? esp_assoc
->ipsa_unique_mask
: 0;
1540 if ((ah_mask
== 0) && (esp_mask
== 0))
1544 * The pkt_unique check will also check for tunnel mode on the SA
1545 * vs. the tunneled_packet boolean. "Be liberal in what you receive"
1546 * should not apply in this case. ;)
1550 ah_assoc
->ipsa_unique_id
!= (pkt_unique
& ah_mask
)) {
1551 *reason
= "AH inner header mismatch";
1552 *counter
= DROPPER(ipss
, ipds_spd_ah_innermismatch
);
1555 if (esp_mask
!= 0 &&
1556 esp_assoc
->ipsa_unique_id
!= (pkt_unique
& esp_mask
)) {
1557 *reason
= "ESP inner header mismatch";
1558 *counter
= DROPPER(ipss
, ipds_spd_esp_innermismatch
);
1565 ipsec_check_ipsecin_action(ip_recv_attr_t
*ira
, mblk_t
*mp
, ipsec_action_t
*ap
,
1566 ipha_t
*ipha
, ip6_t
*ip6h
, const char **reason
, kstat_named_t
**counter
,
1569 boolean_t ret
= B_TRUE
;
1574 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
1576 ASSERT((ipha
== NULL
&& ip6h
!= NULL
) ||
1577 (ip6h
== NULL
&& ipha
!= NULL
));
1579 if (ira
->ira_flags
& IRAF_LOOPBACK
) {
1581 * Besides accepting pointer-equivalent actions, we also
1582 * accept any ICMP errors we generated for ourselves,
1583 * regardless of policy. If we do not wish to make this
1584 * assumption in the future, check here, and where
1585 * IXAF_TRUSTED_ICMP is initialized in ip.c and ip6.c.
1587 if (ap
== ira
->ira_ipsec_action
||
1588 (ira
->ira_flags
& IRAF_TRUSTED_ICMP
))
1591 /* Deep compare necessary here?? */
1592 *counter
= DROPPER(ipss
, ipds_spd_loopback_mismatch
);
1593 *reason
= "loopback policy mismatch";
1596 ASSERT(!(ira
->ira_flags
& IRAF_TRUSTED_ICMP
));
1597 ASSERT(ira
->ira_flags
& IRAF_IPSEC_SECURE
);
1599 ah_assoc
= ira
->ira_ipsec_ah_sa
;
1600 esp_assoc
= ira
->ira_ipsec_esp_sa
;
1602 decaps
= (ira
->ira_flags
& IRAF_IPSEC_DECAPS
);
1604 switch (ap
->ipa_act
.ipa_type
) {
1605 case IPSEC_ACT_DISCARD
:
1606 case IPSEC_ACT_REJECT
:
1607 /* Should "fail hard" */
1608 *counter
= DROPPER(ipss
, ipds_spd_explicit
);
1609 *reason
= "blocked by policy";
1612 case IPSEC_ACT_BYPASS
:
1613 case IPSEC_ACT_CLEAR
:
1614 *counter
= DROPPER(ipss
, ipds_spd_got_secure
);
1615 *reason
= "expected clear, got protected";
1618 case IPSEC_ACT_APPLY
:
1619 ipp
= &ap
->ipa_act
.ipa_apply
;
1621 * As of now we do the simple checks of whether
1622 * the datagram has gone through the required IPSEC
1623 * protocol constraints or not. We might have more
1624 * in the future like sensitive levels, key bits, etc.
1625 * If it fails the constraints, check whether we would
1626 * have accepted this if it had come in clear.
1628 if (ipp
->ipp_use_ah
) {
1629 if (ah_assoc
== NULL
) {
1630 ret
= ipsec_inbound_accept_clear(mp
, ipha
,
1632 *counter
= DROPPER(ipss
, ipds_spd_got_clear
);
1633 *reason
= "unprotected not accepted";
1636 ASSERT(ah_assoc
!= NULL
);
1637 ASSERT(ipp
->ipp_auth_alg
!= 0);
1639 if (ah_assoc
->ipsa_auth_alg
!=
1640 ipp
->ipp_auth_alg
) {
1641 *counter
= DROPPER(ipss
, ipds_spd_bad_ahalg
);
1642 *reason
= "unacceptable ah alg";
1646 } else if (ah_assoc
!= NULL
) {
1648 * Don't allow this. Check IPSEC NOTE above
1649 * ip_fanout_proto().
1651 *counter
= DROPPER(ipss
, ipds_spd_got_ah
);
1652 *reason
= "unexpected AH";
1656 if (ipp
->ipp_use_esp
) {
1657 if (esp_assoc
== NULL
) {
1658 ret
= ipsec_inbound_accept_clear(mp
, ipha
,
1660 *counter
= DROPPER(ipss
, ipds_spd_got_clear
);
1661 *reason
= "unprotected not accepted";
1664 ASSERT(esp_assoc
!= NULL
);
1665 ASSERT(ipp
->ipp_encr_alg
!= 0);
1667 if (esp_assoc
->ipsa_encr_alg
!=
1668 ipp
->ipp_encr_alg
) {
1669 *counter
= DROPPER(ipss
, ipds_spd_bad_espealg
);
1670 *reason
= "unacceptable esp alg";
1675 * If the client does not need authentication,
1676 * we don't verify the alogrithm.
1678 if (ipp
->ipp_use_espa
) {
1679 if (esp_assoc
->ipsa_auth_alg
!=
1680 ipp
->ipp_esp_auth_alg
) {
1681 *counter
= DROPPER(ipss
,
1682 ipds_spd_bad_espaalg
);
1683 *reason
= "unacceptable esp auth alg";
1688 } else if (esp_assoc
!= NULL
) {
1690 * Don't allow this. Check IPSEC NOTE above
1691 * ip_fanout_proto().
1693 *counter
= DROPPER(ipss
, ipds_spd_got_esp
);
1694 *reason
= "unexpected ESP";
1698 if (ipp
->ipp_use_se
) {
1700 ret
= ipsec_inbound_accept_clear(mp
, ipha
,
1704 *counter
= DROPPER(ipss
,
1705 ipds_spd_bad_selfencap
);
1706 *reason
= "self encap not found";
1710 } else if (decaps
) {
1712 * XXX If the packet comes in tunneled and the
1713 * recipient does not expect it to be tunneled, it
1714 * is okay. But we drop to be consistent with the
1717 *counter
= DROPPER(ipss
, ipds_spd_got_selfencap
);
1718 *reason
= "unexpected self encap";
1722 if (ira
->ira_ipsec_action
!= NULL
) {
1724 * This can happen if we do a double policy-check on
1726 * XXX XXX should fix this case!
1728 IPACT_REFRELE(ira
->ira_ipsec_action
);
1730 ASSERT(ira
->ira_flags
& IRAF_IPSEC_SECURE
);
1731 ASSERT(ira
->ira_ipsec_action
== NULL
);
1733 ira
->ira_ipsec_action
= ap
;
1734 break; /* from switch */
1740 spd_match_inbound_ids(ipsec_latch_t
*ipl
, ipsa_t
*sa
)
1742 ASSERT(ipl
->ipl_ids_latched
== B_TRUE
);
1743 return ipsid_equal(ipl
->ipl_remote_cid
, sa
->ipsa_src_cid
) &&
1744 ipsid_equal(ipl
->ipl_local_cid
, sa
->ipsa_dst_cid
);
1748 * Takes a latched conn and an inbound packet and returns a unique_id suitable
1749 * for SA comparisons. Most of the time we will copy from the conn_t, but
1750 * there are cases when the conn_t is latched but it has wildcard selectors,
1751 * and then we need to fallback to scooping them out of the packet.
1753 * Assume we'll never have 0 with a conn_t present, so use 0 as a failure. We
1754 * can get away with this because we only have non-zero ports/proto for
1757 * Ideal candidate for an "inline" keyword, as we're JUST convoluted enough
1758 * to not be a nice macro.
1761 conn_to_unique(conn_t
*connp
, mblk_t
*data_mp
, ipha_t
*ipha
, ip6_t
*ip6h
)
1763 ipsec_selector_t sel
;
1764 uint8_t ulp
= connp
->conn_proto
;
1766 ASSERT(connp
->conn_latch_in_policy
!= NULL
);
1768 if ((ulp
== IPPROTO_TCP
|| ulp
== IPPROTO_UDP
|| ulp
== IPPROTO_SCTP
) &&
1769 (connp
->conn_fport
== 0 || connp
->conn_lport
== 0)) {
1770 /* Slow path - we gotta grab from the packet. */
1771 if (ipsec_init_inbound_sel(&sel
, data_mp
, ipha
, ip6h
,
1772 SEL_NONE
) != SELRET_SUCCESS
) {
1773 /* Failure -> have caller free packet with ENOMEM. */
1776 return (SA_UNIQUE_ID(sel
.ips_remote_port
, sel
.ips_local_port
,
1777 sel
.ips_protocol
, 0));
1780 #ifdef DEBUG_NOT_UNTIL_6478464
1781 if (ipsec_init_inbound_sel(&sel
, data_mp
, ipha
, ip6h
, SEL_NONE
) ==
1783 ASSERT(sel
.ips_local_port
== connp
->conn_lport
);
1784 ASSERT(sel
.ips_remote_port
== connp
->conn_fport
);
1785 ASSERT(sel
.ips_protocol
== connp
->conn_proto
);
1787 ASSERT(connp
->conn_proto
!= 0);
1790 return (SA_UNIQUE_ID(connp
->conn_fport
, connp
->conn_lport
, ulp
, 0));
1794 * Called to check policy on a latched connection.
1795 * Note that we don't dereference conn_latch or conn_ihere since the conn might
1796 * be closing. The caller passes a held ipsec_latch_t instead.
1799 ipsec_check_ipsecin_latch(ip_recv_attr_t
*ira
, mblk_t
*mp
, ipsec_latch_t
*ipl
,
1800 ipsec_action_t
*ap
, ipha_t
*ipha
, ip6_t
*ip6h
, const char **reason
,
1801 kstat_named_t
**counter
, conn_t
*connp
, netstack_t
*ns
)
1803 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
1805 ASSERT(ipl
->ipl_ids_latched
== B_TRUE
);
1806 ASSERT(ira
->ira_flags
& IRAF_IPSEC_SECURE
);
1808 if (!(ira
->ira_flags
& IRAF_LOOPBACK
)) {
1810 * Over loopback, there aren't real security associations,
1811 * so there are neither identities nor "unique" values
1812 * for us to check the packet against.
1814 if (ira
->ira_ipsec_ah_sa
!= NULL
) {
1815 if (!spd_match_inbound_ids(ipl
,
1816 ira
->ira_ipsec_ah_sa
)) {
1817 *counter
= DROPPER(ipss
, ipds_spd_ah_badid
);
1818 *reason
= "AH identity mismatch";
1823 if (ira
->ira_ipsec_esp_sa
!= NULL
) {
1824 if (!spd_match_inbound_ids(ipl
,
1825 ira
->ira_ipsec_esp_sa
)) {
1826 *counter
= DROPPER(ipss
, ipds_spd_esp_badid
);
1827 *reason
= "ESP identity mismatch";
1833 * Can fudge pkt_unique from connp because we're latched.
1834 * In DEBUG kernels (see conn_to_unique()'s implementation),
1835 * verify this even if it REALLY slows things down.
1837 if (!ipsec_check_ipsecin_unique(ira
, reason
, counter
,
1838 conn_to_unique(connp
, mp
, ipha
, ip6h
), ns
)) {
1842 return (ipsec_check_ipsecin_action(ira
, mp
, ap
, ipha
, ip6h
, reason
,
1847 * Check to see whether this secured datagram meets the policy
1848 * constraints specified in ipsp.
1850 * Called from ipsec_check_global_policy, and ipsec_check_inbound_policy.
1852 * Consumes a reference to ipsp.
1853 * Returns the mblk if ok.
1856 ipsec_check_ipsecin_policy(mblk_t
*data_mp
, ipsec_policy_t
*ipsp
,
1857 ipha_t
*ipha
, ip6_t
*ip6h
, uint64_t pkt_unique
, ip_recv_attr_t
*ira
,
1861 const char *reason
= "no policy actions found";
1862 ip_stack_t
*ipst
= ns
->netstack_ip
;
1863 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
1864 kstat_named_t
*counter
;
1866 counter
= DROPPER(ipss
, ipds_spd_got_secure
);
1868 ASSERT(ipsp
!= NULL
);
1870 ASSERT((ipha
== NULL
&& ip6h
!= NULL
) ||
1871 (ip6h
== NULL
&& ipha
!= NULL
));
1873 if (ira
->ira_flags
& IRAF_LOOPBACK
)
1874 return (ipsec_check_loopback_policy(data_mp
, ira
, ipsp
));
1876 ASSERT(ira
->ira_flags
& IRAF_IPSEC_SECURE
);
1878 if (ira
->ira_ipsec_action
!= NULL
) {
1880 * this can happen if we do a double policy-check on a packet
1881 * Would be nice to be able to delete this test..
1883 IPACT_REFRELE(ira
->ira_ipsec_action
);
1885 ASSERT(ira
->ira_ipsec_action
== NULL
);
1887 if (!SA_IDS_MATCH(ira
->ira_ipsec_ah_sa
, ira
->ira_ipsec_esp_sa
)) {
1888 reason
= "inbound AH and ESP identities differ";
1889 counter
= DROPPER(ipss
, ipds_spd_ahesp_diffid
);
1893 if (!ipsec_check_ipsecin_unique(ira
, &reason
, &counter
, pkt_unique
,
1898 * Ok, now loop through the possible actions and see if any
1899 * of them work for us.
1902 for (ap
= ipsp
->ipsp_act
; ap
!= NULL
; ap
= ap
->ipa_next
) {
1903 if (ipsec_check_ipsecin_action(ira
, data_mp
, ap
,
1904 ipha
, ip6h
, &reason
, &counter
, ns
)) {
1905 BUMP_MIB(&ipst
->ips_ip_mib
, ipsecInSucceeded
);
1906 IPPOL_REFRELE(ipsp
);
1911 ipsec_rl_strlog(ns
, IP_MOD_ID
, 0, 0, SL_ERROR
|SL_WARN
|SL_CONSOLE
,
1912 "ipsec inbound policy mismatch: %s, packet dropped\n",
1914 IPPOL_REFRELE(ipsp
);
1915 ASSERT(ira
->ira_ipsec_action
== NULL
);
1916 BUMP_MIB(&ipst
->ips_ip_mib
, ipsecInFailed
);
1917 ip_drop_packet(data_mp
, B_TRUE
, NULL
, counter
,
1918 &ipss
->ipsec_spd_dropper
);
1923 * sleazy prefix-length-based compare.
1924 * another inlining candidate..
1927 ip_addr_match(uint8_t *addr1
, int pfxlen
, in6_addr_t
*addr2p
)
1929 int offset
= pfxlen
>>3;
1930 int bitsleft
= pfxlen
& 7;
1931 uint8_t *addr2
= (uint8_t *)addr2p
;
1934 * and there was much evil..
1935 * XXX should inline-expand the bcmp here and do this 32 bits
1936 * or 64 bits at a time..
1938 return ((bcmp(addr1
, addr2
, offset
) == 0) &&
1940 (((addr1
[offset
] ^ addr2
[offset
]) & (0xff<<(8-bitsleft
))) == 0)));
1943 static ipsec_policy_t
*
1944 ipsec_find_policy_chain(ipsec_policy_t
*best
, ipsec_policy_t
*chain
,
1945 ipsec_selector_t
*sel
, boolean_t is_icmp_inv_acq
)
1947 ipsec_selkey_t
*isel
;
1949 int bpri
= best
? best
->ipsp_prio
: 0;
1951 for (p
= chain
; p
!= NULL
; p
= p
->ipsp_hash
.hash_next
) {
1954 if (p
->ipsp_prio
<= bpri
)
1956 isel
= &p
->ipsp_sel
->ipsl_key
;
1957 valid
= isel
->ipsl_valid
;
1959 if ((valid
& IPSL_PROTOCOL
) &&
1960 (isel
->ipsl_proto
!= sel
->ips_protocol
))
1963 if ((valid
& IPSL_REMOTE_ADDR
) &&
1964 !ip_addr_match((uint8_t *)&isel
->ipsl_remote
,
1965 isel
->ipsl_remote_pfxlen
, &sel
->ips_remote_addr_v6
))
1968 if ((valid
& IPSL_LOCAL_ADDR
) &&
1969 !ip_addr_match((uint8_t *)&isel
->ipsl_local
,
1970 isel
->ipsl_local_pfxlen
, &sel
->ips_local_addr_v6
))
1973 if ((valid
& IPSL_REMOTE_PORT
) &&
1974 isel
->ipsl_rport
!= sel
->ips_remote_port
)
1977 if ((valid
& IPSL_LOCAL_PORT
) &&
1978 isel
->ipsl_lport
!= sel
->ips_local_port
)
1981 if (!is_icmp_inv_acq
) {
1982 if ((valid
& IPSL_ICMP_TYPE
) &&
1983 (isel
->ipsl_icmp_type
> sel
->ips_icmp_type
||
1984 isel
->ipsl_icmp_type_end
< sel
->ips_icmp_type
)) {
1988 if ((valid
& IPSL_ICMP_CODE
) &&
1989 (isel
->ipsl_icmp_code
> sel
->ips_icmp_code
||
1990 isel
->ipsl_icmp_code_end
<
1991 sel
->ips_icmp_code
)) {
1996 * special case for icmp inverse acquire
1997 * we only want policies that aren't drop/pass
1999 if (p
->ipsp_act
->ipa_act
.ipa_type
!= IPSEC_ACT_APPLY
)
2003 /* we matched all the packet-port-field selectors! */
2005 bpri
= p
->ipsp_prio
;
2012 * Try to find and return the best policy entry under a given policy
2013 * root for a given set of selectors; the first parameter "best" is
2014 * the current best policy so far. If "best" is non-null, we have a
2015 * reference to it. We return a reference to a policy; if that policy
2016 * is not the original "best", we need to release that reference
2020 ipsec_find_policy_head(ipsec_policy_t
*best
, ipsec_policy_head_t
*head
,
2021 int direction
, ipsec_selector_t
*sel
)
2023 ipsec_policy_t
*curbest
;
2024 ipsec_policy_root_t
*root
;
2025 uint8_t is_icmp_inv_acq
= sel
->ips_is_icmp_inv_acq
;
2026 int af
= sel
->ips_isv4
? IPSEC_AF_V4
: IPSEC_AF_V6
;
2029 root
= &head
->iph_root
[direction
];
2032 if (is_icmp_inv_acq
) {
2033 if (sel
->ips_isv4
) {
2034 if (sel
->ips_protocol
!= IPPROTO_ICMP
) {
2035 cmn_err(CE_WARN
, "ipsec_find_policy_head:"
2036 " expecting icmp, got %d",
2040 if (sel
->ips_protocol
!= IPPROTO_ICMPV6
) {
2041 cmn_err(CE_WARN
, "ipsec_find_policy_head:"
2042 " expecting icmpv6, got %d",
2049 rw_enter(&head
->iph_lock
, RW_READER
);
2051 if (root
->ipr_nchains
> 0) {
2052 curbest
= ipsec_find_policy_chain(curbest
,
2053 root
->ipr_hash
[selector_hash(sel
, root
)].hash_head
, sel
,
2056 curbest
= ipsec_find_policy_chain(curbest
, root
->ipr_nonhash
[af
], sel
,
2060 * Adjust reference counts if we found anything new.
2062 if (curbest
!= best
) {
2063 ASSERT(curbest
!= NULL
);
2064 IPPOL_REFHOLD(curbest
);
2067 IPPOL_REFRELE(best
);
2071 rw_exit(&head
->iph_lock
);
2077 * Find the best system policy (either global or per-interface) which
2078 * applies to the given selector; look in all the relevant policy roots
2079 * to figure out which policy wins.
2081 * Returns a reference to a policy; caller must release this
2082 * reference when done.
2085 ipsec_find_policy(int direction
, const conn_t
*connp
, ipsec_selector_t
*sel
,
2089 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
2091 p
= ipsec_find_policy_head(NULL
, &ipss
->ipsec_system_policy
,
2093 if ((connp
!= NULL
) && (connp
->conn_policy
!= NULL
)) {
2094 p
= ipsec_find_policy_head(p
, connp
->conn_policy
,
2102 * Check with global policy and see whether this inbound
2103 * packet meets the policy constraints.
2105 * Locate appropriate policy from global policy, supplemented by the
2106 * conn's configured and/or cached policy if the conn is supplied.
2108 * Dispatch to ipsec_check_ipsecin_policy if we have policy and an
2109 * encrypted packet to see if they match.
2111 * Otherwise, see if the policy allows cleartext; if not, drop it on the
2115 ipsec_check_global_policy(mblk_t
*data_mp
, conn_t
*connp
,
2116 ipha_t
*ipha
, ip6_t
*ip6h
, ip_recv_attr_t
*ira
, netstack_t
*ns
)
2119 ipsec_selector_t sel
;
2120 boolean_t policy_present
;
2121 kstat_named_t
*counter
;
2122 uint64_t pkt_unique
;
2123 ip_stack_t
*ipst
= ns
->netstack_ip
;
2124 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
2126 sel
.ips_is_icmp_inv_acq
= 0;
2128 ASSERT((ipha
== NULL
&& ip6h
!= NULL
) ||
2129 (ip6h
== NULL
&& ipha
!= NULL
));
2132 policy_present
= ipss
->ipsec_inbound_v4_policy_present
;
2134 policy_present
= ipss
->ipsec_inbound_v6_policy_present
;
2136 if (!policy_present
&& connp
== NULL
) {
2138 * No global policy and no per-socket policy;
2139 * just pass it back (but we shouldn't get here in that case)
2145 * If we have cached policy, use it.
2146 * Otherwise consult system policy.
2148 if ((connp
!= NULL
) && (connp
->conn_latch
!= NULL
)) {
2149 p
= connp
->conn_latch_in_policy
;
2154 * Fudge sel for UNIQUE_ID setting below.
2156 pkt_unique
= conn_to_unique(connp
, data_mp
, ipha
, ip6h
);
2158 /* Initialize the ports in the selector */
2159 if (ipsec_init_inbound_sel(&sel
, data_mp
, ipha
, ip6h
,
2160 SEL_NONE
) == SELRET_NOMEM
) {
2162 * Technically not a policy mismatch, but it is
2163 * an internal failure.
2165 ipsec_log_policy_failure(IPSEC_POLICY_MISMATCH
,
2166 "ipsec_init_inbound_sel", ipha
, ip6h
, B_TRUE
, ns
);
2167 counter
= DROPPER(ipss
, ipds_spd_nomem
);
2172 * Find the policy which best applies.
2174 * If we find global policy, we should look at both
2175 * local policy and global policy and see which is
2176 * stronger and match accordingly.
2178 * If we don't find a global policy, check with
2179 * local policy alone.
2182 p
= ipsec_find_policy(IPSEC_TYPE_INBOUND
, connp
, &sel
, ns
);
2183 pkt_unique
= SA_UNIQUE_ID(sel
.ips_remote_port
,
2184 sel
.ips_local_port
, sel
.ips_protocol
, 0);
2188 if (!(ira
->ira_flags
& IRAF_IPSEC_SECURE
)) {
2190 * We have no policy; default to succeeding.
2191 * XXX paranoid system design doesn't do this.
2193 BUMP_MIB(&ipst
->ips_ip_mib
, ipsecInSucceeded
);
2196 counter
= DROPPER(ipss
, ipds_spd_got_secure
);
2197 ipsec_log_policy_failure(IPSEC_POLICY_NOT_NEEDED
,
2198 "ipsec_check_global_policy", ipha
, ip6h
, B_TRUE
,
2203 if (ira
->ira_flags
& IRAF_IPSEC_SECURE
) {
2204 return (ipsec_check_ipsecin_policy(data_mp
, p
, ipha
, ip6h
,
2205 pkt_unique
, ira
, ns
));
2207 if (p
->ipsp_act
->ipa_allow_clear
) {
2208 BUMP_MIB(&ipst
->ips_ip_mib
, ipsecInSucceeded
);
2214 * If we reach here, we will drop the packet because it failed the
2215 * global policy check because the packet was cleartext, and it
2216 * should not have been.
2218 ipsec_log_policy_failure(IPSEC_POLICY_MISMATCH
,
2219 "ipsec_check_global_policy", ipha
, ip6h
, B_FALSE
, ns
);
2220 counter
= DROPPER(ipss
, ipds_spd_got_clear
);
2223 ip_drop_packet(data_mp
, B_TRUE
, NULL
, counter
,
2224 &ipss
->ipsec_spd_dropper
);
2225 BUMP_MIB(&ipst
->ips_ip_mib
, ipsecInFailed
);
2230 * We check whether an inbound datagram is a valid one
2231 * to accept in clear. If it is secure, it is the job
2232 * of IPSEC to log information appropriately if it
2233 * suspects that it may not be the real one.
2235 * It is called only while fanning out to the ULP
2236 * where ULP accepts only secure data and the incoming
2237 * is clear. Usually we never accept clear datagrams in
2238 * such cases. ICMP is the only exception.
2240 * NOTE : We don't call this function if the client (ULP)
2241 * is willing to accept things in clear.
2244 ipsec_inbound_accept_clear(mblk_t
*mp
, ipha_t
*ipha
, ip6_t
*ip6h
)
2246 ushort_t iph_hdr_length
;
2251 ASSERT((ipha
!= NULL
&& ip6h
== NULL
) ||
2252 (ipha
== NULL
&& ip6h
!= NULL
));
2255 iph_hdr_length
= ip_hdr_length_v6(mp
, ip6h
);
2256 if (!ip_hdr_length_nexthdr_v6(mp
, ip6h
, &iph_hdr_length
,
2260 if (*nexthdrp
!= IPPROTO_ICMPV6
)
2262 icmp6
= (icmp6_t
*)(&mp
->b_rptr
[iph_hdr_length
]);
2263 /* Match IPv6 ICMP policy as closely as IPv4 as possible. */
2264 switch (icmp6
->icmp6_type
) {
2265 case ICMP6_PARAM_PROB
:
2266 /* Corresponds to port/proto unreach in IPv4. */
2267 case ICMP6_ECHO_REQUEST
:
2268 /* Just like IPv4. */
2271 case MLD_LISTENER_QUERY
:
2272 case MLD_LISTENER_REPORT
:
2273 case MLD_LISTENER_REDUCTION
:
2275 * XXX Seperate NDD in IPv4 what about here?
2276 * Plus, mcast is important to ND.
2278 case ICMP6_DST_UNREACH
:
2279 /* Corresponds to HOST/NET unreachable in IPv4. */
2280 case ICMP6_PACKET_TOO_BIG
:
2281 case ICMP6_ECHO_REPLY
:
2282 /* These are trusted in IPv4. */
2283 case ND_ROUTER_SOLICIT
:
2284 case ND_ROUTER_ADVERT
:
2285 case ND_NEIGHBOR_SOLICIT
:
2286 case ND_NEIGHBOR_ADVERT
:
2288 /* Trust ND messages for now. */
2289 case ICMP6_TIME_EXCEEDED
:
2295 * If it is not ICMP, fail this request.
2297 if (ipha
->ipha_protocol
!= IPPROTO_ICMP
) {
2298 #ifdef FRAGCACHE_DEBUG
2299 cmn_err(CE_WARN
, "Dropping - ipha_proto = %d\n",
2300 ipha
->ipha_protocol
);
2304 iph_hdr_length
= IPH_HDR_LENGTH(ipha
);
2305 icmph
= (icmph_t
*)&mp
->b_rptr
[iph_hdr_length
];
2307 * It is an insecure icmp message. Check to see whether we are
2308 * willing to accept this one.
2311 switch (icmph
->icmph_type
) {
2312 case ICMP_ECHO_REPLY
:
2313 case ICMP_TIME_STAMP_REPLY
:
2314 case ICMP_INFO_REPLY
:
2315 case ICMP_ROUTER_ADVERTISEMENT
:
2317 * We should not encourage clear replies if this
2318 * client expects secure. If somebody is replying
2319 * in clear some mailicious user watching both the
2320 * request and reply, can do chosen-plain-text attacks.
2321 * With global policy we might be just expecting secure
2322 * but sending out clear. We don't know what the right
2323 * thing is. We can't do much here as we can't control
2324 * the sender here. Till we are sure of what to do,
2328 case ICMP_ECHO_REQUEST
:
2329 case ICMP_TIME_STAMP_REQUEST
:
2330 case ICMP_INFO_REQUEST
:
2331 case ICMP_ADDRESS_MASK_REQUEST
:
2332 case ICMP_ROUTER_SOLICITATION
:
2333 case ICMP_ADDRESS_MASK_REPLY
:
2335 * Don't accept this as somebody could be sending
2336 * us plain text to get encrypted data. If we reply,
2337 * it will lead to chosen plain text attack.
2340 case ICMP_DEST_UNREACHABLE
:
2341 switch (icmph
->icmph_code
) {
2342 case ICMP_FRAGMENTATION_NEEDED
:
2344 * Be in sync with icmp_inbound, where we have
2345 * already set dce_pmtu
2347 #ifdef FRAGCACHE_DEBUG
2348 cmn_err(CE_WARN
, "ICMP frag needed\n");
2351 case ICMP_HOST_UNREACHABLE
:
2352 case ICMP_NET_UNREACHABLE
:
2354 * By accepting, we could reset a connection.
2355 * How do we solve the problem of some
2356 * intermediate router sending in-secure ICMP
2360 case ICMP_PORT_UNREACHABLE
:
2361 case ICMP_PROTOCOL_UNREACHABLE
:
2365 case ICMP_SOURCE_QUENCH
:
2367 * If this is an attack, TCP will slow start
2368 * because of this. Is it very harmful ?
2371 case ICMP_PARAM_PROBLEM
:
2373 case ICMP_TIME_EXCEEDED
:
2384 ipsec_latch_ids(ipsec_latch_t
*ipl
, ipsid_t
*local
, ipsid_t
*remote
)
2386 mutex_enter(&ipl
->ipl_lock
);
2388 if (ipl
->ipl_ids_latched
) {
2389 /* I lost, someone else got here before me */
2390 mutex_exit(&ipl
->ipl_lock
);
2395 IPSID_REFHOLD(local
);
2397 IPSID_REFHOLD(remote
);
2399 ipl
->ipl_local_cid
= local
;
2400 ipl
->ipl_remote_cid
= remote
;
2401 ipl
->ipl_ids_latched
= B_TRUE
;
2402 mutex_exit(&ipl
->ipl_lock
);
2406 ipsec_latch_inbound(conn_t
*connp
, ip_recv_attr_t
*ira
)
2409 ipsec_latch_t
*ipl
= connp
->conn_latch
;
2411 if (!ipl
->ipl_ids_latched
) {
2412 ipsid_t
*local
= NULL
;
2413 ipsid_t
*remote
= NULL
;
2415 if (!(ira
->ira_flags
& IRAF_LOOPBACK
)) {
2416 ASSERT(ira
->ira_flags
& IRAF_IPSEC_SECURE
);
2417 if (ira
->ira_ipsec_esp_sa
!= NULL
)
2418 sa
= ira
->ira_ipsec_esp_sa
;
2420 sa
= ira
->ira_ipsec_ah_sa
;
2422 local
= sa
->ipsa_dst_cid
;
2423 remote
= sa
->ipsa_src_cid
;
2425 ipsec_latch_ids(ipl
, local
, remote
);
2427 if (ira
->ira_flags
& IRAF_IPSEC_SECURE
) {
2428 if (connp
->conn_latch_in_action
!= NULL
) {
2430 * Previously cached action. This is probably
2431 * harmless, but in DEBUG kernels, check for
2434 * Preserve the existing action to preserve latch
2437 ASSERT(connp
->conn_latch_in_action
==
2438 ira
->ira_ipsec_action
);
2441 connp
->conn_latch_in_action
= ira
->ira_ipsec_action
;
2442 IPACT_REFHOLD(connp
->conn_latch_in_action
);
2447 * Check whether the policy constraints are met either for an
2448 * inbound datagram; called from IP in numerous places.
2450 * Note that this is not a chokepoint for inbound policy checks;
2451 * see also ipsec_check_ipsecin_latch() and ipsec_check_global_policy()
2454 ipsec_check_inbound_policy(mblk_t
*mp
, conn_t
*connp
,
2455 ipha_t
*ipha
, ip6_t
*ip6h
, ip_recv_attr_t
*ira
)
2461 ipsec_stack_t
*ipss
;
2464 ipsec_policy_head_t
*policy_head
;
2465 ipsec_policy_t
*p
= NULL
;
2467 ASSERT(connp
!= NULL
);
2468 ns
= connp
->conn_netstack
;
2469 ipss
= ns
->netstack_ipsec
;
2470 ipst
= ns
->netstack_ip
;
2472 if (!(ira
->ira_flags
& IRAF_IPSEC_SECURE
)) {
2474 * This is the case where the incoming datagram is
2475 * cleartext and we need to see whether this client
2476 * would like to receive such untrustworthy things from
2481 mutex_enter(&connp
->conn_lock
);
2482 if (connp
->conn_state_flags
& CONN_CONDEMNED
) {
2483 mutex_exit(&connp
->conn_lock
);
2484 ip_drop_packet(mp
, B_TRUE
, NULL
,
2485 DROPPER(ipss
, ipds_spd_got_clear
),
2486 &ipss
->ipsec_spd_dropper
);
2487 BUMP_MIB(&ipst
->ips_ip_mib
, ipsecInFailed
);
2490 if (connp
->conn_latch
!= NULL
) {
2491 /* Hold a reference in case the conn is closing */
2492 p
= connp
->conn_latch_in_policy
;
2495 mutex_exit(&connp
->conn_lock
);
2497 * Policy is cached in the conn.
2499 if (p
!= NULL
&& !p
->ipsp_act
->ipa_allow_clear
) {
2500 ret
= ipsec_inbound_accept_clear(mp
,
2503 BUMP_MIB(&ipst
->ips_ip_mib
,
2508 ipsec_log_policy_failure(
2509 IPSEC_POLICY_MISMATCH
,
2510 "ipsec_check_inbound_policy", ipha
,
2512 ip_drop_packet(mp
, B_TRUE
, NULL
,
2513 DROPPER(ipss
, ipds_spd_got_clear
),
2514 &ipss
->ipsec_spd_dropper
);
2515 BUMP_MIB(&ipst
->ips_ip_mib
,
2521 BUMP_MIB(&ipst
->ips_ip_mib
, ipsecInSucceeded
);
2527 policy_head
= connp
->conn_policy
;
2529 /* Hold a reference in case the conn is closing */
2530 if (policy_head
!= NULL
)
2531 IPPH_REFHOLD(policy_head
);
2532 mutex_exit(&connp
->conn_lock
);
2534 * As this is a non-hardbound connection we need
2535 * to look at both per-socket policy and global
2538 mp
= ipsec_check_global_policy(mp
, connp
,
2539 ipha
, ip6h
, ira
, ns
);
2540 if (policy_head
!= NULL
)
2541 IPPH_REFRELE(policy_head
, ns
);
2546 mutex_enter(&connp
->conn_lock
);
2547 /* Connection is closing */
2548 if (connp
->conn_state_flags
& CONN_CONDEMNED
) {
2549 mutex_exit(&connp
->conn_lock
);
2550 ip_drop_packet(mp
, B_TRUE
, NULL
,
2551 DROPPER(ipss
, ipds_spd_got_clear
),
2552 &ipss
->ipsec_spd_dropper
);
2553 BUMP_MIB(&ipst
->ips_ip_mib
, ipsecInFailed
);
2558 * Once a connection is latched it remains so for life, the conn_latch
2559 * pointer on the conn has not changed, simply initializing ipl here
2560 * as the earlier initialization was done only in the cleartext case.
2562 if ((ipl
= connp
->conn_latch
) == NULL
) {
2564 policy_head
= connp
->conn_policy
;
2566 /* Hold a reference in case the conn is closing */
2567 if (policy_head
!= NULL
)
2568 IPPH_REFHOLD(policy_head
);
2569 mutex_exit(&connp
->conn_lock
);
2571 * We don't have policies cached in the conn
2572 * for this stream. So, look at the global
2573 * policy. It will check against conn or global
2574 * depending on whichever is stronger.
2576 retmp
= ipsec_check_global_policy(mp
, connp
,
2577 ipha
, ip6h
, ira
, ns
);
2578 if (policy_head
!= NULL
)
2579 IPPH_REFRELE(policy_head
, ns
);
2583 IPLATCH_REFHOLD(ipl
);
2584 /* Hold reference on conn_latch_in_action in case conn is closing */
2585 ap
= connp
->conn_latch_in_action
;
2588 mutex_exit(&connp
->conn_lock
);
2591 /* Policy is cached & latched; fast(er) path */
2593 kstat_named_t
*counter
;
2595 if (ipsec_check_ipsecin_latch(ira
, mp
, ipl
, ap
,
2596 ipha
, ip6h
, &reason
, &counter
, connp
, ns
)) {
2597 BUMP_MIB(&ipst
->ips_ip_mib
, ipsecInSucceeded
);
2598 IPLATCH_REFRELE(ipl
);
2602 ipsec_rl_strlog(ns
, IP_MOD_ID
, 0, 0,
2603 SL_ERROR
|SL_WARN
|SL_CONSOLE
,
2604 "ipsec inbound policy mismatch: %s, packet dropped\n",
2606 ip_drop_packet(mp
, B_TRUE
, NULL
, counter
,
2607 &ipss
->ipsec_spd_dropper
);
2608 BUMP_MIB(&ipst
->ips_ip_mib
, ipsecInFailed
);
2609 IPLATCH_REFRELE(ipl
);
2613 if ((p
= connp
->conn_latch_in_policy
) == NULL
) {
2614 ipsec_weird_null_inbound_policy
++;
2615 IPLATCH_REFRELE(ipl
);
2619 unique_id
= conn_to_unique(connp
, mp
, ipha
, ip6h
);
2621 mp
= ipsec_check_ipsecin_policy(mp
, p
, ipha
, ip6h
, unique_id
, ira
, ns
);
2623 * NOTE: ipsecIn{Failed,Succeeeded} bumped by
2624 * ipsec_check_ipsecin_policy().
2627 ipsec_latch_inbound(connp
, ira
);
2628 IPLATCH_REFRELE(ipl
);
2633 * Handle all sorts of cases like tunnel-mode and ICMP.
2636 prepended_length(mblk_t
*mp
, uintptr_t hptr
)
2640 while (mp
!= NULL
) {
2641 if (hptr
>= (uintptr_t)mp
->b_rptr
&& hptr
<
2642 (uintptr_t)mp
->b_wptr
) {
2643 rc
+= (int)(hptr
- (uintptr_t)mp
->b_rptr
);
2644 break; /* out of while loop */
2646 rc
+= (int)MBLKL(mp
);
2652 * IF (big IF) we make it here by naturally exiting the loop,
2653 * then ip6h isn't in the mblk chain "mp" at all.
2655 * The only case where this happens is with a reversed IP
2656 * header that gets passed up by inbound ICMP processing.
2657 * This unfortunately triggers longstanding bug 6478464. For
2658 * now, just pass up 0 for the answer.
2660 #ifdef DEBUG_NOT_UNTIL_6478464
2672 * SELRET_NOMEM --> msgpullup() needed to gather things failed.
2673 * SELRET_BADPKT --> If we're being called after tunnel-mode fragment
2674 * gathering, the initial fragment is too short for
2675 * useful data. Only returned if SEL_TUNNEL_FIRSTFRAG is
2677 * SELRET_SUCCESS --> "sel" now has initialized IPsec selector data.
2678 * SELRET_TUNFRAG --> This is a fragment in a tunnel-mode packet. Caller
2679 * should put this packet in a fragment-gathering queue.
2680 * Only returned if SEL_TUNNEL_MODE and SEL_PORT_POLICY
2683 * Note that ipha/ip6h can be in a different mblk (mp->b_cont) in the case
2684 * of tunneled packets.
2685 * Also, mp->b_rptr can be an ICMP error where ipha/ip6h is the packet in
2686 * error past the ICMP error.
2689 ipsec_init_inbound_sel(ipsec_selector_t
*sel
, mblk_t
*mp
, ipha_t
*ipha
,
2690 ip6_t
*ip6h
, uint8_t sel_flags
)
2693 int outer_hdr_len
= 0; /* For ICMP or tunnel-mode cases... */
2695 mblk_t
*spare_mp
= NULL
;
2696 uint8_t *nexthdrp
, *transportp
;
2700 boolean_t port_policy_present
= (sel_flags
& SEL_PORT_POLICY
);
2701 boolean_t is_icmp
= (sel_flags
& SEL_IS_ICMP
);
2702 boolean_t tunnel_mode
= (sel_flags
& SEL_TUNNEL_MODE
);
2703 boolean_t post_frag
= (sel_flags
& SEL_POST_FRAG
);
2705 ASSERT((ipha
== NULL
&& ip6h
!= NULL
) ||
2706 (ipha
!= NULL
&& ip6h
== NULL
));
2709 outer_hdr_len
= prepended_length(mp
, (uintptr_t)ip6h
);
2710 nexthdr
= ip6h
->ip6_nxt
;
2711 icmp_proto
= IPPROTO_ICMPV6
;
2712 sel
->ips_isv4
= B_FALSE
;
2713 sel
->ips_local_addr_v6
= ip6h
->ip6_dst
;
2714 sel
->ips_remote_addr_v6
= ip6h
->ip6_src
;
2716 bzero(&ipp
, sizeof (ipp
));
2719 case IPPROTO_HOPOPTS
:
2720 case IPPROTO_ROUTING
:
2721 case IPPROTO_DSTOPTS
:
2722 case IPPROTO_FRAGMENT
:
2724 * Use ip_hdr_length_nexthdr_v6(). And have a spare
2725 * mblk that's contiguous to feed it
2727 if ((spare_mp
= msgpullup(mp
, -1)) == NULL
)
2728 return (SELRET_NOMEM
);
2729 if (!ip_hdr_length_nexthdr_v6(spare_mp
,
2730 (ip6_t
*)(spare_mp
->b_rptr
+ outer_hdr_len
),
2731 &hdr_len
, &nexthdrp
)) {
2732 /* Malformed packet - caller frees. */
2733 ipsec_freemsg_chain(spare_mp
);
2734 return (SELRET_BADPKT
);
2736 /* Repopulate now that we have the whole packet */
2737 ip6h
= (ip6_t
*)(spare_mp
->b_rptr
+ outer_hdr_len
);
2738 (void) ip_find_hdr_v6(spare_mp
, ip6h
, B_FALSE
, &ipp
,
2740 nexthdr
= *nexthdrp
;
2741 /* We can just extract based on hdr_len now. */
2744 (void) ip_find_hdr_v6(mp
, ip6h
, B_FALSE
, &ipp
, NULL
);
2745 hdr_len
= IPV6_HDR_LEN
;
2748 if (port_policy_present
&& IS_V6_FRAGMENT(ipp
) && !is_icmp
) {
2750 ipsec_freemsg_chain(spare_mp
);
2751 return (SELRET_TUNFRAG
);
2753 transportp
= (uint8_t *)ip6h
+ hdr_len
;
2755 outer_hdr_len
= prepended_length(mp
, (uintptr_t)ipha
);
2756 icmp_proto
= IPPROTO_ICMP
;
2757 sel
->ips_isv4
= B_TRUE
;
2758 sel
->ips_local_addr_v4
= ipha
->ipha_dst
;
2759 sel
->ips_remote_addr_v4
= ipha
->ipha_src
;
2760 nexthdr
= ipha
->ipha_protocol
;
2761 hdr_len
= IPH_HDR_LENGTH(ipha
);
2763 if (port_policy_present
&&
2764 IS_V4_FRAGMENT(ipha
->ipha_fragment_offset_and_flags
) &&
2767 ipsec_freemsg_chain(spare_mp
);
2768 return (SELRET_TUNFRAG
);
2770 transportp
= (uint8_t *)ipha
+ hdr_len
;
2772 sel
->ips_protocol
= nexthdr
;
2774 if ((nexthdr
!= IPPROTO_TCP
&& nexthdr
!= IPPROTO_UDP
&&
2775 nexthdr
!= IPPROTO_SCTP
&& nexthdr
!= icmp_proto
) ||
2776 (!port_policy_present
&& !post_frag
&& tunnel_mode
)) {
2777 sel
->ips_remote_port
= sel
->ips_local_port
= 0;
2778 ipsec_freemsg_chain(spare_mp
);
2779 return (SELRET_SUCCESS
);
2782 if (transportp
+ 4 > mp
->b_wptr
) {
2783 /* If we didn't pullup a copy already, do so now. */
2785 * XXX performance, will upper-layers frequently split TCP/UDP
2786 * apart from IP or options? If so, perhaps we should revisit
2787 * the spare_mp strategy.
2789 ipsec_hdr_pullup_needed
++;
2790 if (spare_mp
== NULL
&&
2791 (spare_mp
= msgpullup(mp
, -1)) == NULL
) {
2792 return (SELRET_NOMEM
);
2794 transportp
= &spare_mp
->b_rptr
[hdr_len
+ outer_hdr_len
];
2797 if (nexthdr
== icmp_proto
) {
2798 sel
->ips_icmp_type
= *transportp
++;
2799 sel
->ips_icmp_code
= *transportp
;
2800 sel
->ips_remote_port
= sel
->ips_local_port
= 0;
2802 ports
= (uint16_t *)transportp
;
2803 sel
->ips_remote_port
= *ports
++;
2804 sel
->ips_local_port
= *ports
;
2806 ipsec_freemsg_chain(spare_mp
);
2807 return (SELRET_SUCCESS
);
2811 * This is called with a b_next chain of messages from the fragcache code,
2812 * hence it needs to discard a chain on error.
2815 ipsec_init_outbound_ports(ipsec_selector_t
*sel
, mblk_t
*mp
, ipha_t
*ipha
,
2816 ip6_t
*ip6h
, int outer_hdr_len
, ipsec_stack_t
*ipss
)
2819 * XXX cut&paste shared with ipsec_init_inbound_sel
2823 mblk_t
*spare_mp
= NULL
;
2827 uint8_t check_proto
;
2829 ASSERT((ipha
== NULL
&& ip6h
!= NULL
) ||
2830 (ipha
!= NULL
&& ip6h
== NULL
));
2833 check_proto
= IPPROTO_ICMPV6
;
2834 nexthdr
= ip6h
->ip6_nxt
;
2836 case IPPROTO_HOPOPTS
:
2837 case IPPROTO_ROUTING
:
2838 case IPPROTO_DSTOPTS
:
2839 case IPPROTO_FRAGMENT
:
2841 * Use ip_hdr_length_nexthdr_v6(). And have a spare
2842 * mblk that's contiguous to feed it
2844 spare_mp
= msgpullup(mp
, -1);
2845 if (spare_mp
== NULL
||
2846 !ip_hdr_length_nexthdr_v6(spare_mp
,
2847 (ip6_t
*)(spare_mp
->b_rptr
+ outer_hdr_len
),
2848 &hdr_len
, &nexthdrp
)) {
2849 /* Always works, even if NULL. */
2850 ipsec_freemsg_chain(spare_mp
);
2851 ip_drop_packet_chain(mp
, B_FALSE
, NULL
,
2852 DROPPER(ipss
, ipds_spd_nomem
),
2853 &ipss
->ipsec_spd_dropper
);
2856 nexthdr
= *nexthdrp
;
2857 /* We can just extract based on hdr_len now. */
2861 hdr_len
= IPV6_HDR_LEN
;
2865 check_proto
= IPPROTO_ICMP
;
2866 hdr_len
= IPH_HDR_LENGTH(ipha
);
2867 nexthdr
= ipha
->ipha_protocol
;
2870 sel
->ips_protocol
= nexthdr
;
2871 if (nexthdr
!= IPPROTO_TCP
&& nexthdr
!= IPPROTO_UDP
&&
2872 nexthdr
!= IPPROTO_SCTP
&& nexthdr
!= check_proto
) {
2873 sel
->ips_local_port
= sel
->ips_remote_port
= 0;
2874 ipsec_freemsg_chain(spare_mp
); /* Always works, even if NULL */
2878 if (&mp
->b_rptr
[hdr_len
] + 4 + outer_hdr_len
> mp
->b_wptr
) {
2879 /* If we didn't pullup a copy already, do so now. */
2881 * XXX performance, will upper-layers frequently split TCP/UDP
2882 * apart from IP or options? If so, perhaps we should revisit
2883 * the spare_mp strategy.
2885 * XXX should this be msgpullup(mp, hdr_len+4) ???
2887 if (spare_mp
== NULL
&&
2888 (spare_mp
= msgpullup(mp
, -1)) == NULL
) {
2889 ip_drop_packet_chain(mp
, B_FALSE
, NULL
,
2890 DROPPER(ipss
, ipds_spd_nomem
),
2891 &ipss
->ipsec_spd_dropper
);
2894 ports
= (uint16_t *)&spare_mp
->b_rptr
[hdr_len
+ outer_hdr_len
];
2896 ports
= (uint16_t *)&mp
->b_rptr
[hdr_len
+ outer_hdr_len
];
2899 if (nexthdr
== check_proto
) {
2900 typecode
= (uint8_t *)ports
;
2901 sel
->ips_icmp_type
= *typecode
++;
2902 sel
->ips_icmp_code
= *typecode
;
2903 sel
->ips_remote_port
= sel
->ips_local_port
= 0;
2905 sel
->ips_local_port
= *ports
++;
2906 sel
->ips_remote_port
= *ports
;
2908 ipsec_freemsg_chain(spare_mp
); /* Always works, even if NULL */
2913 * Prepend an mblk with a ipsec_crypto_t to the message chain.
2914 * Frees the argument and returns NULL should the allocation fail.
2915 * Returns the pointer to the crypto data part.
2918 ipsec_add_crypto_data(mblk_t
*data_mp
, ipsec_crypto_t
**icp
)
2922 mp
= allocb(sizeof (ipsec_crypto_t
), BPRI_MED
);
2927 bzero(mp
->b_rptr
, sizeof (ipsec_crypto_t
));
2928 mp
->b_wptr
+= sizeof (ipsec_crypto_t
);
2929 mp
->b_cont
= data_mp
;
2930 mp
->b_datap
->db_type
= M_EVENT
; /* For ASSERT */
2931 *icp
= (ipsec_crypto_t
*)mp
->b_rptr
;
2936 * Remove what was prepended above. Return b_cont and a pointer to the
2938 * The caller must call ipsec_free_crypto_data for mblk once it is done
2939 * with the crypto data.
2942 ipsec_remove_crypto_data(mblk_t
*crypto_mp
, ipsec_crypto_t
**icp
)
2944 ASSERT(crypto_mp
->b_datap
->db_type
== M_EVENT
);
2945 ASSERT(MBLKL(crypto_mp
) == sizeof (ipsec_crypto_t
));
2947 *icp
= (ipsec_crypto_t
*)crypto_mp
->b_rptr
;
2948 return (crypto_mp
->b_cont
);
2952 * Free what was prepended above. Return b_cont.
2955 ipsec_free_crypto_data(mblk_t
*crypto_mp
)
2959 ASSERT(crypto_mp
->b_datap
->db_type
== M_EVENT
);
2960 ASSERT(MBLKL(crypto_mp
) == sizeof (ipsec_crypto_t
));
2962 mp
= crypto_mp
->b_cont
;
2968 * Create an ipsec_action_t based on the way an inbound packet was protected.
2969 * Used to reflect traffic back to a sender.
2971 * We don't bother interning the action into the hash table.
2974 ipsec_in_to_out_action(ip_recv_attr_t
*ira
)
2976 ipsa_t
*ah_assoc
, *esp_assoc
;
2977 uint_t auth_alg
= 0, encr_alg
= 0, espa_alg
= 0;
2981 ap
= kmem_cache_alloc(ipsec_action_cache
, KM_NOSLEEP
);
2986 bzero(ap
, sizeof (*ap
));
2987 HASH_NULL(ap
, ipa_hash
);
2988 ap
->ipa_next
= NULL
;
2992 * Get the algorithms that were used for this packet.
2994 ap
->ipa_act
.ipa_type
= IPSEC_ACT_APPLY
;
2995 ap
->ipa_act
.ipa_log
= 0;
2996 ASSERT(ira
->ira_flags
& IRAF_IPSEC_SECURE
);
2998 ah_assoc
= ira
->ira_ipsec_ah_sa
;
2999 ap
->ipa_act
.ipa_apply
.ipp_use_ah
= (ah_assoc
!= NULL
);
3001 esp_assoc
= ira
->ira_ipsec_esp_sa
;
3002 ap
->ipa_act
.ipa_apply
.ipp_use_esp
= (esp_assoc
!= NULL
);
3004 if (esp_assoc
!= NULL
) {
3005 encr_alg
= esp_assoc
->ipsa_encr_alg
;
3006 espa_alg
= esp_assoc
->ipsa_auth_alg
;
3007 ap
->ipa_act
.ipa_apply
.ipp_use_espa
= (espa_alg
!= 0);
3009 if (ah_assoc
!= NULL
)
3010 auth_alg
= ah_assoc
->ipsa_auth_alg
;
3012 ap
->ipa_act
.ipa_apply
.ipp_encr_alg
= (uint8_t)encr_alg
;
3013 ap
->ipa_act
.ipa_apply
.ipp_auth_alg
= (uint8_t)auth_alg
;
3014 ap
->ipa_act
.ipa_apply
.ipp_esp_auth_alg
= (uint8_t)espa_alg
;
3015 ap
->ipa_act
.ipa_apply
.ipp_use_se
=
3016 !!(ira
->ira_flags
& IRAF_IPSEC_DECAPS
);
3019 if (esp_assoc
!= NULL
) {
3020 ap
->ipa_act
.ipa_apply
.ipp_espa_minbits
=
3021 esp_assoc
->ipsa_authkeybits
;
3022 ap
->ipa_act
.ipa_apply
.ipp_espa_maxbits
=
3023 esp_assoc
->ipsa_authkeybits
;
3024 ap
->ipa_act
.ipa_apply
.ipp_espe_minbits
=
3025 esp_assoc
->ipsa_encrkeybits
;
3026 ap
->ipa_act
.ipa_apply
.ipp_espe_maxbits
=
3027 esp_assoc
->ipsa_encrkeybits
;
3028 ap
->ipa_act
.ipa_apply
.ipp_km_proto
= esp_assoc
->ipsa_kmp
;
3029 ap
->ipa_act
.ipa_apply
.ipp_km_cookie
= esp_assoc
->ipsa_kmc
;
3030 if (esp_assoc
->ipsa_flags
& IPSA_F_UNIQUE
)
3033 if (ah_assoc
!= NULL
) {
3034 ap
->ipa_act
.ipa_apply
.ipp_ah_minbits
=
3035 ah_assoc
->ipsa_authkeybits
;
3036 ap
->ipa_act
.ipa_apply
.ipp_ah_maxbits
=
3037 ah_assoc
->ipsa_authkeybits
;
3038 ap
->ipa_act
.ipa_apply
.ipp_km_proto
= ah_assoc
->ipsa_kmp
;
3039 ap
->ipa_act
.ipa_apply
.ipp_km_cookie
= ah_assoc
->ipsa_kmc
;
3040 if (ah_assoc
->ipsa_flags
& IPSA_F_UNIQUE
)
3043 ap
->ipa_act
.ipa_apply
.ipp_use_unique
= unique
;
3044 ap
->ipa_want_unique
= unique
;
3045 ap
->ipa_allow_clear
= B_FALSE
;
3046 ap
->ipa_want_se
= !!(ira
->ira_flags
& IRAF_IPSEC_DECAPS
);
3047 ap
->ipa_want_ah
= (ah_assoc
!= NULL
);
3048 ap
->ipa_want_esp
= (esp_assoc
!= NULL
);
3050 ap
->ipa_ovhd
= ipsec_act_ovhd(&ap
->ipa_act
);
3052 ap
->ipa_act
.ipa_apply
.ipp_replay_depth
= 0; /* don't care */
3059 * Compute the worst-case amount of extra space required by an action.
3060 * Note that, because of the ESP considerations listed below, this is
3061 * actually not the same as the best-case reduction in the MTU; in the
3062 * future, we should pass additional information to this function to
3063 * allow the actual MTU impact to be computed.
3065 * AH: Revisit this if we implement algorithms with
3066 * a verifier size of more than 12 bytes.
3068 * ESP: A more exact but more messy computation would take into
3069 * account the interaction between the cipher block size and the
3070 * effective MTU, yielding the inner payload size which reflects a
3071 * packet with *minimum* ESP padding..
3074 ipsec_act_ovhd(const ipsec_act_t
*act
)
3076 int32_t overhead
= 0;
3078 if (act
->ipa_type
== IPSEC_ACT_APPLY
) {
3079 const ipsec_prot_t
*ipp
= &act
->ipa_apply
;
3081 if (ipp
->ipp_use_ah
)
3082 overhead
+= IPSEC_MAX_AH_HDR_SIZE
;
3083 if (ipp
->ipp_use_esp
) {
3084 overhead
+= IPSEC_MAX_ESP_HDR_SIZE
;
3085 overhead
+= sizeof (struct udphdr
);
3087 if (ipp
->ipp_use_se
)
3088 overhead
+= IP_SIMPLE_HDR_LENGTH
;
3094 * This hash function is used only when creating policies and thus is not
3095 * performance-critical for packet flows.
3097 * Future work: canonicalize the structures hashed with this (i.e.,
3098 * zeroize padding) so the hash works correctly.
3102 policy_hash(int size
, const void *start
, const void *end
)
3109 * Hash function macros for each address type.
3111 * The IPV6 hash function assumes that the low order 32-bits of the
3112 * address (typically containing the low order 24 bits of the mac
3113 * address) are reasonably well-distributed. Revisit this if we run
3114 * into trouble from lots of collisions on ::1 addresses and the like
3117 #define IPSEC_IPV4_HASH(a, n) ((a) % (n))
3118 #define IPSEC_IPV6_HASH(a, n) (((a).s6_addr32[3]) % (n))
3121 * These two hash functions should produce coordinated values
3122 * but have slightly different roles.
3125 selkey_hash(const ipsec_selkey_t
*selkey
, netstack_t
*ns
)
3127 uint32_t valid
= selkey
->ipsl_valid
;
3128 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
3130 if (!(valid
& IPSL_REMOTE_ADDR
))
3131 return (IPSEC_SEL_NOHASH
);
3133 if (valid
& IPSL_IPV4
) {
3134 if (selkey
->ipsl_remote_pfxlen
== 32) {
3135 return (IPSEC_IPV4_HASH(selkey
->ipsl_remote
.ipsad_v4
,
3136 ipss
->ipsec_spd_hashsize
));
3139 if (valid
& IPSL_IPV6
) {
3140 if (selkey
->ipsl_remote_pfxlen
== 128) {
3141 return (IPSEC_IPV6_HASH(selkey
->ipsl_remote
.ipsad_v6
,
3142 ipss
->ipsec_spd_hashsize
));
3145 return (IPSEC_SEL_NOHASH
);
3149 selector_hash(ipsec_selector_t
*sel
, ipsec_policy_root_t
*root
)
3151 if (sel
->ips_isv4
) {
3152 return (IPSEC_IPV4_HASH(sel
->ips_remote_addr_v4
,
3153 root
->ipr_nchains
));
3155 return (IPSEC_IPV6_HASH(sel
->ips_remote_addr_v6
, root
->ipr_nchains
));
3159 * Intern actions into the action hash table.
3162 ipsec_act_find(const ipsec_act_t
*a
, int n
, netstack_t
*ns
)
3167 ipsec_action_t
*prev
= NULL
;
3168 int32_t overhead
, maxovhd
= 0;
3169 boolean_t allow_clear
= B_FALSE
;
3170 boolean_t want_ah
= B_FALSE
;
3171 boolean_t want_esp
= B_FALSE
;
3172 boolean_t want_se
= B_FALSE
;
3173 boolean_t want_unique
= B_FALSE
;
3174 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
3177 * TODO: should canonicalize a[] (i.e., zeroize any padding)
3178 * so we can use a non-trivial policy_hash function.
3180 for (i
= n
-1; i
>= 0; i
--) {
3181 hval
= policy_hash(IPSEC_ACTION_HASH_SIZE
, &a
[i
], &a
[n
]);
3183 HASH_LOCK(ipss
->ipsec_action_hash
, hval
);
3185 for (HASH_ITERATE(ap
, ipa_hash
,
3186 ipss
->ipsec_action_hash
, hval
)) {
3187 if (bcmp(&ap
->ipa_act
, &a
[i
], sizeof (*a
)) != 0)
3189 if (ap
->ipa_next
!= prev
)
3194 HASH_UNLOCK(ipss
->ipsec_action_hash
, hval
);
3199 * need to allocate a new one..
3201 ap
= kmem_cache_alloc(ipsec_action_cache
, KM_NOSLEEP
);
3203 HASH_UNLOCK(ipss
->ipsec_action_hash
, hval
);
3205 ipsec_action_free(prev
);
3208 HASH_INSERT(ap
, ipa_hash
, ipss
->ipsec_action_hash
, hval
);
3210 ap
->ipa_next
= prev
;
3213 overhead
= ipsec_act_ovhd(&a
[i
]);
3214 if (maxovhd
< overhead
)
3217 if ((a
[i
].ipa_type
== IPSEC_ACT_BYPASS
) ||
3218 (a
[i
].ipa_type
== IPSEC_ACT_CLEAR
))
3219 allow_clear
= B_TRUE
;
3220 if (a
[i
].ipa_type
== IPSEC_ACT_APPLY
) {
3221 const ipsec_prot_t
*ipp
= &a
[i
].ipa_apply
;
3223 ASSERT(ipp
->ipp_use_ah
|| ipp
->ipp_use_esp
);
3224 want_ah
|= ipp
->ipp_use_ah
;
3225 want_esp
|= ipp
->ipp_use_esp
;
3226 want_se
|= ipp
->ipp_use_se
;
3227 want_unique
|= ipp
->ipp_use_unique
;
3229 ap
->ipa_allow_clear
= allow_clear
;
3230 ap
->ipa_want_ah
= want_ah
;
3231 ap
->ipa_want_esp
= want_esp
;
3232 ap
->ipa_want_se
= want_se
;
3233 ap
->ipa_want_unique
= want_unique
;
3234 ap
->ipa_refs
= 1; /* from the hash table */
3235 ap
->ipa_ovhd
= maxovhd
;
3239 HASH_UNLOCK(ipss
->ipsec_action_hash
, hval
);
3242 ap
->ipa_refs
++; /* caller's reference */
3248 * Called when refcount goes to 0, indicating that all references to this
3251 * This does not unchain the action from the hash table.
3254 ipsec_action_free(ipsec_action_t
*ap
)
3257 ipsec_action_t
*np
= ap
->ipa_next
;
3258 ASSERT(ap
->ipa_refs
== 0);
3259 ASSERT(ap
->ipa_hash
.hash_pp
== NULL
);
3260 kmem_cache_free(ipsec_action_cache
, ap
);
3262 /* Inlined IPACT_REFRELE -- avoid recursion */
3266 if (atomic_dec_32_nv(&(ap
)->ipa_refs
) != 0)
3268 /* End inlined IPACT_REFRELE */
3273 * Called when the action hash table goes away.
3275 * The actions can be queued on an mblk with ipsec_in or
3276 * ipsec_out, hence the actions might still be around.
3277 * But we decrement ipa_refs here since we no longer have
3278 * a reference to the action from the hash table.
3281 ipsec_action_free_table(ipsec_action_t
*ap
)
3283 while (ap
!= NULL
) {
3284 ipsec_action_t
*np
= ap
->ipa_next
;
3286 /* FIXME: remove? */
3287 (void) printf("ipsec_action_free_table(%p) ref %d\n",
3288 (void *)ap
, ap
->ipa_refs
);
3289 ASSERT(ap
->ipa_refs
> 0);
3296 * Need to walk all stack instances since the reclaim function
3297 * is global for all instances
3301 ipsec_action_reclaim(void *arg
)
3303 netstack_handle_t nh
;
3305 ipsec_stack_t
*ipss
;
3307 netstack_next_init(&nh
);
3308 while ((ns
= netstack_next(&nh
)) != NULL
) {
3310 * netstack_next() can return a netstack_t with a NULL
3311 * netstack_ipsec at boot time.
3313 if ((ipss
= ns
->netstack_ipsec
) == NULL
) {
3317 ipsec_action_reclaim_stack(ipss
);
3320 netstack_next_fini(&nh
);
3324 * Periodically sweep action hash table for actions with refcount==1, and
3325 * nuke them. We cannot do this "on demand" (i.e., from IPACT_REFRELE)
3326 * because we can't close the race between another thread finding the action
3327 * in the hash table without holding the bucket lock during IPACT_REFRELE.
3328 * Instead, we run this function sporadically to clean up after ourselves;
3329 * we also set it as the "reclaim" function for the action kmem_cache.
3331 * Note that it may take several passes of ipsec_action_gc() to free all
3335 ipsec_action_reclaim_stack(ipsec_stack_t
*ipss
)
3339 for (i
= 0; i
< IPSEC_ACTION_HASH_SIZE
; i
++) {
3340 ipsec_action_t
*ap
, *np
;
3342 /* skip the lock if nobody home */
3343 if (ipss
->ipsec_action_hash
[i
].hash_head
== NULL
)
3346 HASH_LOCK(ipss
->ipsec_action_hash
, i
);
3347 for (ap
= ipss
->ipsec_action_hash
[i
].hash_head
;
3348 ap
!= NULL
; ap
= np
) {
3349 ASSERT(ap
->ipa_refs
> 0);
3350 np
= ap
->ipa_hash
.hash_next
;
3351 if (ap
->ipa_refs
> 1)
3353 HASH_UNCHAIN(ap
, ipa_hash
,
3354 ipss
->ipsec_action_hash
, i
);
3357 HASH_UNLOCK(ipss
->ipsec_action_hash
, i
);
3362 * Intern a selector set into the selector set hash table.
3363 * This is simpler than the actions case..
3365 static ipsec_sel_t
*
3366 ipsec_find_sel(ipsec_selkey_t
*selkey
, netstack_t
*ns
)
3369 uint32_t hval
, bucket
;
3370 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
3373 * Exactly one AF bit should be set in selkey.
3375 ASSERT(!(selkey
->ipsl_valid
& IPSL_IPV4
) ^
3376 !(selkey
->ipsl_valid
& IPSL_IPV6
));
3378 hval
= selkey_hash(selkey
, ns
);
3379 /* Set pol_hval to uninitialized until we put it in a polhead. */
3380 selkey
->ipsl_sel_hval
= hval
;
3382 bucket
= (hval
== IPSEC_SEL_NOHASH
) ? 0 : hval
;
3384 ASSERT(!HASH_LOCKED(ipss
->ipsec_sel_hash
, bucket
));
3385 HASH_LOCK(ipss
->ipsec_sel_hash
, bucket
);
3387 for (HASH_ITERATE(sp
, ipsl_hash
, ipss
->ipsec_sel_hash
, bucket
)) {
3388 if (bcmp(&sp
->ipsl_key
, selkey
,
3389 offsetof(ipsec_selkey_t
, ipsl_pol_hval
)) == 0)
3395 HASH_UNLOCK(ipss
->ipsec_sel_hash
, bucket
);
3399 sp
= kmem_cache_alloc(ipsec_sel_cache
, KM_NOSLEEP
);
3401 HASH_UNLOCK(ipss
->ipsec_sel_hash
, bucket
);
3405 HASH_INSERT(sp
, ipsl_hash
, ipss
->ipsec_sel_hash
, bucket
);
3406 sp
->ipsl_refs
= 2; /* one for hash table, one for caller */
3407 sp
->ipsl_key
= *selkey
;
3408 /* Set to uninitalized and have insertion into polhead fix things. */
3409 if (selkey
->ipsl_sel_hval
!= IPSEC_SEL_NOHASH
)
3410 sp
->ipsl_key
.ipsl_pol_hval
= 0;
3412 sp
->ipsl_key
.ipsl_pol_hval
= IPSEC_SEL_NOHASH
;
3414 HASH_UNLOCK(ipss
->ipsec_sel_hash
, bucket
);
3420 ipsec_sel_rel(ipsec_sel_t
**spp
, netstack_t
*ns
)
3422 ipsec_sel_t
*sp
= *spp
;
3423 int hval
= sp
->ipsl_key
.ipsl_sel_hval
;
3424 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
3428 if (hval
== IPSEC_SEL_NOHASH
)
3431 ASSERT(!HASH_LOCKED(ipss
->ipsec_sel_hash
, hval
));
3432 HASH_LOCK(ipss
->ipsec_sel_hash
, hval
);
3433 if (--sp
->ipsl_refs
== 1) {
3434 HASH_UNCHAIN(sp
, ipsl_hash
, ipss
->ipsec_sel_hash
, hval
);
3436 HASH_UNLOCK(ipss
->ipsec_sel_hash
, hval
);
3437 ASSERT(sp
->ipsl_refs
== 0);
3438 kmem_cache_free(ipsec_sel_cache
, sp
);
3439 /* Caller unlocks */
3443 HASH_UNLOCK(ipss
->ipsec_sel_hash
, hval
);
3447 * Free a policy rule which we know is no longer being referenced.
3450 ipsec_policy_free(ipsec_policy_t
*ipp
)
3452 ASSERT(ipp
->ipsp_refs
== 0);
3453 ASSERT(ipp
->ipsp_sel
!= NULL
);
3454 ASSERT(ipp
->ipsp_act
!= NULL
);
3455 ASSERT(ipp
->ipsp_netstack
!= NULL
);
3457 ipsec_sel_rel(&ipp
->ipsp_sel
, ipp
->ipsp_netstack
);
3458 IPACT_REFRELE(ipp
->ipsp_act
);
3459 kmem_cache_free(ipsec_pol_cache
, ipp
);
3463 * Construction of new policy rules; construct a policy, and add it to
3464 * the appropriate tables.
3467 ipsec_policy_create(ipsec_selkey_t
*keys
, const ipsec_act_t
*a
,
3468 int nacts
, int prio
, uint64_t *index_ptr
, netstack_t
*ns
)
3472 ipsec_policy_t
*ipp
;
3473 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
3475 if (index_ptr
== NULL
)
3476 index_ptr
= &ipss
->ipsec_next_policy_index
;
3478 ipp
= kmem_cache_alloc(ipsec_pol_cache
, KM_NOSLEEP
);
3479 ap
= ipsec_act_find(a
, nacts
, ns
);
3480 sp
= ipsec_find_sel(keys
, ns
);
3482 if ((ap
== NULL
) || (sp
== NULL
) || (ipp
== NULL
)) {
3487 ipsec_sel_rel(&sp
, ns
);
3489 kmem_cache_free(ipsec_pol_cache
, ipp
);
3493 HASH_NULL(ipp
, ipsp_hash
);
3495 ipp
->ipsp_netstack
= ns
; /* Needed for ipsec_policy_free */
3496 ipp
->ipsp_refs
= 1; /* caller's reference */
3499 ipp
->ipsp_prio
= prio
; /* rule priority */
3500 ipp
->ipsp_index
= *index_ptr
;
3507 ipsec_update_present_flags(ipsec_stack_t
*ipss
)
3511 hashpol
= (avl_numnodes(&ipss
->ipsec_system_policy
.iph_rulebyid
) > 0);
3514 ipss
->ipsec_outbound_v4_policy_present
= B_TRUE
;
3515 ipss
->ipsec_outbound_v6_policy_present
= B_TRUE
;
3516 ipss
->ipsec_inbound_v4_policy_present
= B_TRUE
;
3517 ipss
->ipsec_inbound_v6_policy_present
= B_TRUE
;
3521 ipss
->ipsec_outbound_v4_policy_present
= (NULL
!=
3522 ipss
->ipsec_system_policy
.iph_root
[IPSEC_TYPE_OUTBOUND
].
3523 ipr_nonhash
[IPSEC_AF_V4
]);
3524 ipss
->ipsec_outbound_v6_policy_present
= (NULL
!=
3525 ipss
->ipsec_system_policy
.iph_root
[IPSEC_TYPE_OUTBOUND
].
3526 ipr_nonhash
[IPSEC_AF_V6
]);
3527 ipss
->ipsec_inbound_v4_policy_present
= (NULL
!=
3528 ipss
->ipsec_system_policy
.iph_root
[IPSEC_TYPE_INBOUND
].
3529 ipr_nonhash
[IPSEC_AF_V4
]);
3530 ipss
->ipsec_inbound_v6_policy_present
= (NULL
!=
3531 ipss
->ipsec_system_policy
.iph_root
[IPSEC_TYPE_INBOUND
].
3532 ipr_nonhash
[IPSEC_AF_V6
]);
3536 ipsec_policy_delete(ipsec_policy_head_t
*php
, ipsec_selkey_t
*keys
, int dir
,
3540 ipsec_policy_t
*ip
, *nip
, *head
;
3542 ipsec_policy_root_t
*pr
= &php
->iph_root
[dir
];
3544 sp
= ipsec_find_sel(keys
, ns
);
3549 af
= (sp
->ipsl_key
.ipsl_valid
& IPSL_IPV4
) ? IPSEC_AF_V4
: IPSEC_AF_V6
;
3551 rw_enter(&php
->iph_lock
, RW_WRITER
);
3553 if (sp
->ipsl_key
.ipsl_pol_hval
== IPSEC_SEL_NOHASH
) {
3554 head
= pr
->ipr_nonhash
[af
];
3556 head
= pr
->ipr_hash
[sp
->ipsl_key
.ipsl_pol_hval
].hash_head
;
3559 for (ip
= head
; ip
!= NULL
; ip
= nip
) {
3560 nip
= ip
->ipsp_hash
.hash_next
;
3561 if (ip
->ipsp_sel
!= sp
) {
3565 IPPOL_UNCHAIN(php
, ip
);
3568 ipsec_update_present_flags(ns
->netstack_ipsec
);
3570 rw_exit(&php
->iph_lock
);
3572 ipsec_sel_rel(&sp
, ns
);
3577 rw_exit(&php
->iph_lock
);
3578 ipsec_sel_rel(&sp
, ns
);
3583 ipsec_policy_delete_index(ipsec_policy_head_t
*php
, uint64_t policy_index
,
3586 boolean_t found
= B_FALSE
;
3587 ipsec_policy_t ipkey
;
3591 bzero(&ipkey
, sizeof (ipkey
));
3592 ipkey
.ipsp_index
= policy_index
;
3594 rw_enter(&php
->iph_lock
, RW_WRITER
);
3597 * We could be cleverer here about the walk.
3598 * but well, (k+1)*log(N) will do for now (k==number of matches,
3599 * N==number of table entries
3602 ip
= (ipsec_policy_t
*)avl_find(&php
->iph_rulebyid
,
3603 (void *)&ipkey
, &where
);
3606 ip
= avl_nearest(&php
->iph_rulebyid
, where
, AVL_AFTER
);
3611 if (ip
->ipsp_index
!= policy_index
) {
3612 ASSERT(ip
->ipsp_index
> policy_index
);
3616 IPPOL_UNCHAIN(php
, ip
);
3622 ipsec_update_present_flags(ns
->netstack_ipsec
);
3625 rw_exit(&php
->iph_lock
);
3627 return (found
? 0 : ENOENT
);
3631 * Given a constructed ipsec_policy_t policy rule, see if it can be entered
3632 * into the correct policy ruleset. As a side-effect, it sets the hash
3633 * entries on "ipp"'s ipsp_pol_hval.
3635 * Returns B_TRUE if it can be entered, B_FALSE if it can't be (because a
3636 * duplicate policy exists with exactly the same selectors), or an icmp
3637 * rule exists with a different encryption/authentication action.
3640 ipsec_check_policy(ipsec_policy_head_t
*php
, ipsec_policy_t
*ipp
, int direction
)
3642 ipsec_policy_root_t
*pr
= &php
->iph_root
[direction
];
3644 ipsec_policy_t
*p2
, *head
;
3645 uint8_t check_proto
;
3646 ipsec_selkey_t
*selkey
= &ipp
->ipsp_sel
->ipsl_key
;
3647 uint32_t valid
= selkey
->ipsl_valid
;
3649 if (valid
& IPSL_IPV6
) {
3650 ASSERT(!(valid
& IPSL_IPV4
));
3652 check_proto
= IPPROTO_ICMPV6
;
3654 ASSERT(valid
& IPSL_IPV4
);
3656 check_proto
= IPPROTO_ICMP
;
3659 ASSERT(RW_WRITE_HELD(&php
->iph_lock
));
3662 * Double-check that we don't have any duplicate selectors here.
3663 * Because selectors are interned below, we need only compare pointers
3666 if (selkey
->ipsl_sel_hval
== IPSEC_SEL_NOHASH
) {
3667 head
= pr
->ipr_nonhash
[af
];
3669 selkey
->ipsl_pol_hval
=
3670 (selkey
->ipsl_valid
& IPSL_IPV4
) ?
3671 IPSEC_IPV4_HASH(selkey
->ipsl_remote
.ipsad_v4
,
3673 IPSEC_IPV6_HASH(selkey
->ipsl_remote
.ipsad_v6
,
3676 head
= pr
->ipr_hash
[selkey
->ipsl_pol_hval
].hash_head
;
3679 for (p2
= head
; p2
!= NULL
; p2
= p2
->ipsp_hash
.hash_next
) {
3680 if (p2
->ipsp_sel
== ipp
->ipsp_sel
)
3685 * If it's ICMP and not a drop or pass rule, run through the ICMP
3686 * rules and make sure the action is either new or the same as any
3687 * other actions. We don't have to check the full chain because
3688 * discard and bypass will override all other actions
3691 if (valid
& IPSL_PROTOCOL
&&
3692 selkey
->ipsl_proto
== check_proto
&&
3693 (ipp
->ipsp_act
->ipa_act
.ipa_type
== IPSEC_ACT_APPLY
)) {
3695 for (p2
= head
; p2
!= NULL
; p2
= p2
->ipsp_hash
.hash_next
) {
3697 if (p2
->ipsp_sel
->ipsl_key
.ipsl_valid
& IPSL_PROTOCOL
&&
3698 p2
->ipsp_sel
->ipsl_key
.ipsl_proto
== check_proto
&&
3699 (p2
->ipsp_act
->ipa_act
.ipa_type
==
3701 return (ipsec_compare_action(p2
, ipp
));
3710 * compare the action chains of two policies for equality
3711 * B_TRUE -> effective equality
3715 ipsec_compare_action(ipsec_policy_t
*p1
, ipsec_policy_t
*p2
)
3718 ipsec_action_t
*act1
, *act2
;
3720 /* We have a valid rule. Let's compare the actions */
3721 if (p1
->ipsp_act
== p2
->ipsp_act
) {
3722 /* same action. We are good */
3726 /* we have to walk the chain */
3728 act1
= p1
->ipsp_act
;
3729 act2
= p2
->ipsp_act
;
3731 while (act1
!= NULL
&& act2
!= NULL
) {
3733 /* otherwise, Are we close enough? */
3734 if (act1
->ipa_allow_clear
!= act2
->ipa_allow_clear
||
3735 act1
->ipa_want_ah
!= act2
->ipa_want_ah
||
3736 act1
->ipa_want_esp
!= act2
->ipa_want_esp
||
3737 act1
->ipa_want_se
!= act2
->ipa_want_se
) {
3738 /* Nope, we aren't */
3742 if (act1
->ipa_want_ah
) {
3743 if (act1
->ipa_act
.ipa_apply
.ipp_auth_alg
!=
3744 act2
->ipa_act
.ipa_apply
.ipp_auth_alg
) {
3748 if (act1
->ipa_act
.ipa_apply
.ipp_ah_minbits
!=
3749 act2
->ipa_act
.ipa_apply
.ipp_ah_minbits
||
3750 act1
->ipa_act
.ipa_apply
.ipp_ah_maxbits
!=
3751 act2
->ipa_act
.ipa_apply
.ipp_ah_maxbits
) {
3756 if (act1
->ipa_want_esp
) {
3757 if (act1
->ipa_act
.ipa_apply
.ipp_use_esp
!=
3758 act2
->ipa_act
.ipa_apply
.ipp_use_esp
||
3759 act1
->ipa_act
.ipa_apply
.ipp_use_espa
!=
3760 act2
->ipa_act
.ipa_apply
.ipp_use_espa
) {
3764 if (act1
->ipa_act
.ipa_apply
.ipp_use_esp
) {
3765 if (act1
->ipa_act
.ipa_apply
.ipp_encr_alg
!=
3766 act2
->ipa_act
.ipa_apply
.ipp_encr_alg
) {
3770 if (act1
->ipa_act
.ipa_apply
.ipp_espe_minbits
!=
3771 act2
->ipa_act
.ipa_apply
.ipp_espe_minbits
||
3772 act1
->ipa_act
.ipa_apply
.ipp_espe_maxbits
!=
3773 act2
->ipa_act
.ipa_apply
.ipp_espe_maxbits
) {
3778 if (act1
->ipa_act
.ipa_apply
.ipp_use_espa
) {
3779 if (act1
->ipa_act
.ipa_apply
.ipp_esp_auth_alg
!=
3780 act2
->ipa_act
.ipa_apply
.ipp_esp_auth_alg
) {
3784 if (act1
->ipa_act
.ipa_apply
.ipp_espa_minbits
!=
3785 act2
->ipa_act
.ipa_apply
.ipp_espa_minbits
||
3786 act1
->ipa_act
.ipa_apply
.ipp_espa_maxbits
!=
3787 act2
->ipa_act
.ipa_apply
.ipp_espa_maxbits
) {
3794 act1
= act1
->ipa_next
;
3795 act2
= act2
->ipa_next
;
3798 if (act1
!= NULL
|| act2
!= NULL
) {
3807 * Given a constructed ipsec_policy_t policy rule, enter it into
3808 * the correct policy ruleset.
3810 * ipsec_check_policy() is assumed to have succeeded first (to check for
3814 ipsec_enter_policy(ipsec_policy_head_t
*php
, ipsec_policy_t
*ipp
, int direction
,
3817 ipsec_policy_root_t
*pr
= &php
->iph_root
[direction
];
3818 ipsec_selkey_t
*selkey
= &ipp
->ipsp_sel
->ipsl_key
;
3819 uint32_t valid
= selkey
->ipsl_valid
;
3820 uint32_t hval
= selkey
->ipsl_pol_hval
;
3823 ASSERT(RW_WRITE_HELD(&php
->iph_lock
));
3825 if (valid
& IPSL_IPV6
) {
3826 ASSERT(!(valid
& IPSL_IPV4
));
3829 ASSERT(valid
& IPSL_IPV4
);
3835 if (hval
== IPSEC_SEL_NOHASH
) {
3836 HASHLIST_INSERT(ipp
, ipsp_hash
, pr
->ipr_nonhash
[af
]);
3838 HASH_LOCK(pr
->ipr_hash
, hval
);
3839 HASH_INSERT(ipp
, ipsp_hash
, pr
->ipr_hash
, hval
);
3840 HASH_UNLOCK(pr
->ipr_hash
, hval
);
3843 ipsec_insert_always(&php
->iph_rulebyid
, ipp
);
3845 ipsec_update_present_flags(ns
->netstack_ipsec
);
3849 ipsec_ipr_flush(ipsec_policy_head_t
*php
, ipsec_policy_root_t
*ipr
)
3851 ipsec_policy_t
*ip
, *nip
;
3852 int af
, chain
, nchain
;
3854 for (af
= 0; af
< IPSEC_NAF
; af
++) {
3855 for (ip
= ipr
->ipr_nonhash
[af
]; ip
!= NULL
; ip
= nip
) {
3856 nip
= ip
->ipsp_hash
.hash_next
;
3857 IPPOL_UNCHAIN(php
, ip
);
3859 ipr
->ipr_nonhash
[af
] = NULL
;
3861 nchain
= ipr
->ipr_nchains
;
3863 for (chain
= 0; chain
< nchain
; chain
++) {
3864 for (ip
= ipr
->ipr_hash
[chain
].hash_head
; ip
!= NULL
;
3866 nip
= ip
->ipsp_hash
.hash_next
;
3867 IPPOL_UNCHAIN(php
, ip
);
3869 ipr
->ipr_hash
[chain
].hash_head
= NULL
;
3874 * Create and insert inbound or outbound policy associated with actp for the
3875 * address family fam into the policy head ph. Returns B_TRUE if policy was
3876 * inserted, and B_FALSE otherwise.
3879 ipsec_polhead_insert(ipsec_policy_head_t
*ph
, ipsec_act_t
*actp
, uint_t nact
,
3880 int fam
, int ptype
, netstack_t
*ns
)
3883 ipsec_policy_t
*pol
;
3884 ipsec_policy_root_t
*pr
;
3886 bzero(&sel
, sizeof (sel
));
3887 sel
.ipsl_valid
= (fam
== IPSEC_AF_V4
? IPSL_IPV4
: IPSL_IPV6
);
3888 if ((pol
= ipsec_policy_create(&sel
, actp
, nact
, IPSEC_PRIO_SOCKET
,
3889 NULL
, ns
)) != NULL
) {
3890 pr
= &ph
->iph_root
[ptype
];
3891 HASHLIST_INSERT(pol
, ipsp_hash
, pr
->ipr_nonhash
[fam
]);
3892 ipsec_insert_always(&ph
->iph_rulebyid
, pol
);
3894 return (pol
!= NULL
);
3898 ipsec_polhead_flush(ipsec_policy_head_t
*php
, netstack_t
*ns
)
3902 ASSERT(RW_WRITE_HELD(&php
->iph_lock
));
3904 for (dir
= 0; dir
< IPSEC_NTYPES
; dir
++)
3905 ipsec_ipr_flush(php
, &php
->iph_root
[dir
]);
3908 ipsec_update_present_flags(ns
->netstack_ipsec
);
3912 ipsec_polhead_free(ipsec_policy_head_t
*php
, netstack_t
*ns
)
3916 ASSERT(php
->iph_refs
== 0);
3918 rw_enter(&php
->iph_lock
, RW_WRITER
);
3919 ipsec_polhead_flush(php
, ns
);
3920 rw_exit(&php
->iph_lock
);
3921 rw_destroy(&php
->iph_lock
);
3922 for (dir
= 0; dir
< IPSEC_NTYPES
; dir
++) {
3923 ipsec_policy_root_t
*ipr
= &php
->iph_root
[dir
];
3926 for (chain
= 0; chain
< ipr
->ipr_nchains
; chain
++)
3927 mutex_destroy(&(ipr
->ipr_hash
[chain
].hash_lock
));
3930 ipsec_polhead_free_table(php
);
3931 kmem_free(php
, sizeof (*php
));
3935 ipsec_ipr_init(ipsec_policy_root_t
*ipr
)
3939 ipr
->ipr_nchains
= 0;
3940 ipr
->ipr_hash
= NULL
;
3942 for (af
= 0; af
< IPSEC_NAF
; af
++) {
3943 ipr
->ipr_nonhash
[af
] = NULL
;
3947 ipsec_policy_head_t
*
3948 ipsec_polhead_create(void)
3950 ipsec_policy_head_t
*php
;
3952 php
= kmem_alloc(sizeof (*php
), KM_NOSLEEP
);
3956 rw_init(&php
->iph_lock
, NULL
, RW_DEFAULT
, NULL
);
3960 ipsec_ipr_init(&php
->iph_root
[IPSEC_TYPE_INBOUND
]);
3961 ipsec_ipr_init(&php
->iph_root
[IPSEC_TYPE_OUTBOUND
]);
3963 avl_create(&php
->iph_rulebyid
, ipsec_policy_cmpbyid
,
3964 sizeof (ipsec_policy_t
), offsetof(ipsec_policy_t
, ipsp_byid
));
3970 * Clone the policy head into a new polhead; release one reference to the
3971 * old one and return the only reference to the new one.
3972 * If the old one had a refcount of 1, just return it.
3974 ipsec_policy_head_t
*
3975 ipsec_polhead_split(ipsec_policy_head_t
*php
, netstack_t
*ns
)
3977 ipsec_policy_head_t
*nphp
;
3980 return (ipsec_polhead_create());
3981 else if (php
->iph_refs
== 1)
3984 nphp
= ipsec_polhead_create();
3988 if (ipsec_copy_polhead(php
, nphp
, ns
) != 0) {
3989 ipsec_polhead_free(nphp
, ns
);
3992 IPPH_REFRELE(php
, ns
);
3997 * When sending a response to a ICMP request or generating a RST
3998 * in the TCP case, the outbound packets need to go at the same level
3999 * of protection as the incoming ones i.e we associate our outbound
4000 * policy with how the packet came in. We call this after we have
4001 * accepted the incoming packet which may or may not have been in
4002 * clear and hence we are sending the reply back with the policy
4003 * matching the incoming datagram's policy.
4005 * NOTE : This technology serves two purposes :
4007 * 1) If we have multiple outbound policies, we send out a reply
4008 * matching with how it came in rather than matching the outbound
4011 * 2) For assymetric policies, we want to make sure that incoming
4012 * and outgoing has the same level of protection. Assymetric
4013 * policies exist only with global policy where we may not have
4014 * both outbound and inbound at the same time.
4016 * NOTE2: This function is called by cleartext cases, so it needs to be
4019 * Note: the caller has moved other parts of ira into ixa already.
4022 ipsec_in_to_out(ip_recv_attr_t
*ira
, ip_xmit_attr_t
*ixa
, mblk_t
*data_mp
,
4023 ipha_t
*ipha
, ip6_t
*ip6h
)
4025 ipsec_selector_t sel
;
4026 ipsec_action_t
*reflect_action
= NULL
;
4027 netstack_t
*ns
= ixa
->ixa_ipst
->ips_netstack
;
4029 bzero((void*)&sel
, sizeof (sel
));
4031 if (ira
->ira_ipsec_action
!= NULL
) {
4032 /* transfer reference.. */
4033 reflect_action
= ira
->ira_ipsec_action
;
4034 ira
->ira_ipsec_action
= NULL
;
4035 } else if (!(ira
->ira_flags
& IRAF_LOOPBACK
))
4036 reflect_action
= ipsec_in_to_out_action(ira
);
4039 * The caller is going to send the datagram out which might
4040 * go on the wire or delivered locally through ire_send_local.
4042 * 1) If it goes out on the wire, new associations will be
4044 * 2) If it is delivered locally, ire_send_local will convert
4045 * this ip_xmit_attr_t back to a ip_recv_attr_t looking at the
4048 ixa
->ixa_ipsec_action
= reflect_action
;
4050 if (!ipsec_init_outbound_ports(&sel
, data_mp
, ipha
, ip6h
, 0,
4051 ns
->netstack_ipsec
)) {
4052 /* Note: data_mp already consumed and ip_drop_packet done */
4055 ixa
->ixa_ipsec_src_port
= sel
.ips_local_port
;
4056 ixa
->ixa_ipsec_dst_port
= sel
.ips_remote_port
;
4057 ixa
->ixa_ipsec_proto
= sel
.ips_protocol
;
4058 ixa
->ixa_ipsec_icmp_type
= sel
.ips_icmp_type
;
4059 ixa
->ixa_ipsec_icmp_code
= sel
.ips_icmp_code
;
4062 * Don't use global policy for this, as we want
4063 * to use the same protection that was applied to the inbound packet.
4064 * Thus we set IXAF_NO_IPSEC is it arrived in the clear to make
4065 * it be sent in the clear.
4067 if (ira
->ira_flags
& IRAF_IPSEC_SECURE
)
4068 ixa
->ixa_flags
|= IXAF_IPSEC_SECURE
;
4070 ixa
->ixa_flags
|= IXAF_NO_IPSEC
;
4076 ipsec_out_release_refs(ip_xmit_attr_t
*ixa
)
4078 if (!(ixa
->ixa_flags
& IXAF_IPSEC_SECURE
))
4081 if (ixa
->ixa_ipsec_ah_sa
!= NULL
) {
4082 IPSA_REFRELE(ixa
->ixa_ipsec_ah_sa
);
4083 ixa
->ixa_ipsec_ah_sa
= NULL
;
4085 if (ixa
->ixa_ipsec_esp_sa
!= NULL
) {
4086 IPSA_REFRELE(ixa
->ixa_ipsec_esp_sa
);
4087 ixa
->ixa_ipsec_esp_sa
= NULL
;
4089 if (ixa
->ixa_ipsec_policy
!= NULL
) {
4090 IPPOL_REFRELE(ixa
->ixa_ipsec_policy
);
4091 ixa
->ixa_ipsec_policy
= NULL
;
4093 if (ixa
->ixa_ipsec_action
!= NULL
) {
4094 IPACT_REFRELE(ixa
->ixa_ipsec_action
);
4095 ixa
->ixa_ipsec_action
= NULL
;
4097 if (ixa
->ixa_ipsec_latch
) {
4098 IPLATCH_REFRELE(ixa
->ixa_ipsec_latch
);
4099 ixa
->ixa_ipsec_latch
= NULL
;
4101 /* Clear the soft references to the SAs */
4102 ixa
->ixa_ipsec_ref
[0].ipsr_sa
= NULL
;
4103 ixa
->ixa_ipsec_ref
[0].ipsr_bucket
= NULL
;
4104 ixa
->ixa_ipsec_ref
[0].ipsr_gen
= 0;
4105 ixa
->ixa_ipsec_ref
[1].ipsr_sa
= NULL
;
4106 ixa
->ixa_ipsec_ref
[1].ipsr_bucket
= NULL
;
4107 ixa
->ixa_ipsec_ref
[1].ipsr_gen
= 0;
4108 ixa
->ixa_flags
&= ~IXAF_IPSEC_SECURE
;
4112 ipsec_in_release_refs(ip_recv_attr_t
*ira
)
4114 if (!(ira
->ira_flags
& IRAF_IPSEC_SECURE
))
4117 if (ira
->ira_ipsec_ah_sa
!= NULL
) {
4118 IPSA_REFRELE(ira
->ira_ipsec_ah_sa
);
4119 ira
->ira_ipsec_ah_sa
= NULL
;
4121 if (ira
->ira_ipsec_esp_sa
!= NULL
) {
4122 IPSA_REFRELE(ira
->ira_ipsec_esp_sa
);
4123 ira
->ira_ipsec_esp_sa
= NULL
;
4125 ira
->ira_flags
&= ~IRAF_IPSEC_SECURE
;
4129 * This is called from ire_send_local when a packet
4130 * is looped back. We setup the ip_recv_attr_t "borrowing" the references
4131 * held by the callers.
4132 * Note that we don't do any IPsec but we carry the actions and IPSEC flags
4133 * across so that the fanout policy checks see that IPsec was applied.
4135 * The caller should do ipsec_in_release_refs() on the ira by calling
4139 ipsec_out_to_in(ip_xmit_attr_t
*ixa
, ill_t
*ill
, ip_recv_attr_t
*ira
)
4141 ipsec_policy_t
*pol
;
4142 ipsec_action_t
*act
;
4144 /* Non-IPsec operations */
4145 ira
->ira_free_flags
= 0;
4146 ira
->ira_zoneid
= ixa
->ixa_zoneid
;
4147 ira
->ira_cred
= ixa
->ixa_cred
;
4148 ira
->ira_cpid
= ixa
->ixa_cpid
;
4149 ira
->ira_tsl
= ixa
->ixa_tsl
;
4150 ira
->ira_ill
= ira
->ira_rill
= ill
;
4151 ira
->ira_flags
= ixa
->ixa_flags
& IAF_MASK
;
4152 ira
->ira_no_loop_zoneid
= ixa
->ixa_no_loop_zoneid
;
4153 ira
->ira_pktlen
= ixa
->ixa_pktlen
;
4154 ira
->ira_ip_hdr_length
= ixa
->ixa_ip_hdr_length
;
4155 ira
->ira_protocol
= ixa
->ixa_protocol
;
4156 ira
->ira_mhip
= NULL
;
4158 ira
->ira_flags
|= IRAF_LOOPBACK
| IRAF_L2SRC_LOOPBACK
;
4160 ira
->ira_sqp
= ixa
->ixa_sqp
;
4161 ira
->ira_ring
= NULL
;
4163 ira
->ira_ruifindex
= ill
->ill_phyint
->phyint_ifindex
;
4164 ira
->ira_rifindex
= ira
->ira_ruifindex
;
4166 if (!(ixa
->ixa_flags
& IXAF_IPSEC_SECURE
))
4169 ira
->ira_flags
|= IRAF_IPSEC_SECURE
;
4171 ira
->ira_ipsec_ah_sa
= NULL
;
4172 ira
->ira_ipsec_esp_sa
= NULL
;
4174 act
= ixa
->ixa_ipsec_action
;
4176 pol
= ixa
->ixa_ipsec_policy
;
4178 act
= pol
->ipsp_act
;
4182 ixa
->ixa_ipsec_action
= NULL
;
4183 ira
->ira_ipsec_action
= act
;
4187 * Consults global policy and per-socket policy to see whether this datagram
4188 * should go out secure. If so it updates the ip_xmit_attr_t
4189 * Should not be used when connecting, since then we want to latch the policy.
4191 * If connp is NULL we just look at the global policy.
4193 * Returns NULL if the packet was dropped, in which case the MIB has
4194 * been incremented and ip_drop_packet done.
4197 ip_output_attach_policy(mblk_t
*mp
, ipha_t
*ipha
, ip6_t
*ip6h
,
4198 const conn_t
*connp
, ip_xmit_attr_t
*ixa
)
4200 ipsec_selector_t sel
;
4201 boolean_t policy_present
;
4202 ip_stack_t
*ipst
= ixa
->ixa_ipst
;
4203 netstack_t
*ns
= ipst
->ips_netstack
;
4204 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
4207 ixa
->ixa_ipsec_policy_gen
= ipss
->ipsec_system_policy
.iph_gen
;
4208 ASSERT((ipha
!= NULL
&& ip6h
== NULL
) ||
4209 (ip6h
!= NULL
&& ipha
== NULL
));
4212 policy_present
= ipss
->ipsec_outbound_v4_policy_present
;
4214 policy_present
= ipss
->ipsec_outbound_v6_policy_present
;
4216 if (!policy_present
&& (connp
== NULL
|| connp
->conn_policy
== NULL
))
4219 bzero((void*)&sel
, sizeof (sel
));
4222 sel
.ips_local_addr_v4
= ipha
->ipha_src
;
4223 sel
.ips_remote_addr_v4
= ip_get_dst(ipha
);
4224 sel
.ips_isv4
= B_TRUE
;
4226 sel
.ips_isv4
= B_FALSE
;
4227 sel
.ips_local_addr_v6
= ip6h
->ip6_src
;
4228 sel
.ips_remote_addr_v6
= ip_get_dst_v6(ip6h
, mp
, NULL
);
4230 sel
.ips_protocol
= ixa
->ixa_protocol
;
4232 if (!ipsec_init_outbound_ports(&sel
, mp
, ipha
, ip6h
, 0, ipss
)) {
4234 BUMP_MIB(&ipst
->ips_ip_mib
, ipIfStatsOutDiscards
);
4236 BUMP_MIB(&ipst
->ips_ip6_mib
, ipIfStatsOutDiscards
);
4238 /* Note: mp already consumed and ip_drop_packet done */
4242 ASSERT(ixa
->ixa_ipsec_policy
== NULL
);
4243 p
= ipsec_find_policy(IPSEC_TYPE_OUTBOUND
, connp
, &sel
, ns
);
4244 ixa
->ixa_ipsec_policy
= p
;
4246 ixa
->ixa_flags
|= IXAF_IPSEC_SECURE
;
4247 if (connp
== NULL
|| connp
->conn_policy
== NULL
)
4248 ixa
->ixa_flags
|= IXAF_IPSEC_GLOBAL_POLICY
;
4250 ixa
->ixa_flags
&= ~IXAF_IPSEC_SECURE
;
4254 * Copy the right port information.
4256 ixa
->ixa_ipsec_src_port
= sel
.ips_local_port
;
4257 ixa
->ixa_ipsec_dst_port
= sel
.ips_remote_port
;
4258 ixa
->ixa_ipsec_icmp_type
= sel
.ips_icmp_type
;
4259 ixa
->ixa_ipsec_icmp_code
= sel
.ips_icmp_code
;
4260 ixa
->ixa_ipsec_proto
= sel
.ips_protocol
;
4265 * When appropriate, this function caches inbound and outbound policy
4266 * for this connection. The outbound policy is stored in conn_ixa.
4267 * Note that it can not be used for SCTP since conn_faddr isn't set for SCTP.
4269 * XXX need to work out more details about per-interface policy and
4272 * XXX may want to split inbound and outbound caching for ill..
4275 ipsec_conn_cache_policy(conn_t
*connp
, boolean_t isv4
)
4277 boolean_t global_policy_present
;
4278 netstack_t
*ns
= connp
->conn_netstack
;
4279 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
4281 connp
->conn_ixa
->ixa_ipsec_policy_gen
=
4282 ipss
->ipsec_system_policy
.iph_gen
;
4284 * There is no policy latching for ICMP sockets because we can't
4285 * decide on which policy to use until we see the packet and get
4286 * type/code selectors.
4288 if (connp
->conn_proto
== IPPROTO_ICMP
||
4289 connp
->conn_proto
== IPPROTO_ICMPV6
) {
4290 connp
->conn_in_enforce_policy
=
4291 connp
->conn_out_enforce_policy
= B_TRUE
;
4292 if (connp
->conn_latch
!= NULL
) {
4293 IPLATCH_REFRELE(connp
->conn_latch
);
4294 connp
->conn_latch
= NULL
;
4296 if (connp
->conn_latch_in_policy
!= NULL
) {
4297 IPPOL_REFRELE(connp
->conn_latch_in_policy
);
4298 connp
->conn_latch_in_policy
= NULL
;
4300 if (connp
->conn_latch_in_action
!= NULL
) {
4301 IPACT_REFRELE(connp
->conn_latch_in_action
);
4302 connp
->conn_latch_in_action
= NULL
;
4304 if (connp
->conn_ixa
->ixa_ipsec_policy
!= NULL
) {
4305 IPPOL_REFRELE(connp
->conn_ixa
->ixa_ipsec_policy
);
4306 connp
->conn_ixa
->ixa_ipsec_policy
= NULL
;
4308 if (connp
->conn_ixa
->ixa_ipsec_action
!= NULL
) {
4309 IPACT_REFRELE(connp
->conn_ixa
->ixa_ipsec_action
);
4310 connp
->conn_ixa
->ixa_ipsec_action
= NULL
;
4312 connp
->conn_ixa
->ixa_flags
&= ~IXAF_IPSEC_SECURE
;
4316 global_policy_present
= isv4
?
4317 (ipss
->ipsec_outbound_v4_policy_present
||
4318 ipss
->ipsec_inbound_v4_policy_present
) :
4319 (ipss
->ipsec_outbound_v6_policy_present
||
4320 ipss
->ipsec_inbound_v6_policy_present
);
4322 if ((connp
->conn_policy
!= NULL
) || global_policy_present
) {
4323 ipsec_selector_t sel
;
4326 if (connp
->conn_latch
== NULL
&&
4327 (connp
->conn_latch
= iplatch_create()) == NULL
) {
4331 bzero((void*)&sel
, sizeof (sel
));
4333 sel
.ips_protocol
= connp
->conn_proto
;
4334 sel
.ips_local_port
= connp
->conn_lport
;
4335 sel
.ips_remote_port
= connp
->conn_fport
;
4336 sel
.ips_is_icmp_inv_acq
= 0;
4337 sel
.ips_isv4
= isv4
;
4339 sel
.ips_local_addr_v4
= connp
->conn_laddr_v4
;
4340 sel
.ips_remote_addr_v4
= connp
->conn_faddr_v4
;
4342 sel
.ips_local_addr_v6
= connp
->conn_laddr_v6
;
4343 sel
.ips_remote_addr_v6
= connp
->conn_faddr_v6
;
4346 p
= ipsec_find_policy(IPSEC_TYPE_INBOUND
, connp
, &sel
, ns
);
4347 if (connp
->conn_latch_in_policy
!= NULL
)
4348 IPPOL_REFRELE(connp
->conn_latch_in_policy
);
4349 connp
->conn_latch_in_policy
= p
;
4350 connp
->conn_in_enforce_policy
= (p
!= NULL
);
4352 p
= ipsec_find_policy(IPSEC_TYPE_OUTBOUND
, connp
, &sel
, ns
);
4353 if (connp
->conn_ixa
->ixa_ipsec_policy
!= NULL
)
4354 IPPOL_REFRELE(connp
->conn_ixa
->ixa_ipsec_policy
);
4355 connp
->conn_ixa
->ixa_ipsec_policy
= p
;
4356 connp
->conn_out_enforce_policy
= (p
!= NULL
);
4358 connp
->conn_ixa
->ixa_flags
|= IXAF_IPSEC_SECURE
;
4359 if (connp
->conn_policy
== NULL
) {
4360 connp
->conn_ixa
->ixa_flags
|=
4361 IXAF_IPSEC_GLOBAL_POLICY
;
4364 connp
->conn_ixa
->ixa_flags
&= ~IXAF_IPSEC_SECURE
;
4366 /* Clear the latched actions too, in case we're recaching. */
4367 if (connp
->conn_ixa
->ixa_ipsec_action
!= NULL
) {
4368 IPACT_REFRELE(connp
->conn_ixa
->ixa_ipsec_action
);
4369 connp
->conn_ixa
->ixa_ipsec_action
= NULL
;
4371 if (connp
->conn_latch_in_action
!= NULL
) {
4372 IPACT_REFRELE(connp
->conn_latch_in_action
);
4373 connp
->conn_latch_in_action
= NULL
;
4375 connp
->conn_ixa
->ixa_ipsec_src_port
= sel
.ips_local_port
;
4376 connp
->conn_ixa
->ixa_ipsec_dst_port
= sel
.ips_remote_port
;
4377 connp
->conn_ixa
->ixa_ipsec_icmp_type
= sel
.ips_icmp_type
;
4378 connp
->conn_ixa
->ixa_ipsec_icmp_code
= sel
.ips_icmp_code
;
4379 connp
->conn_ixa
->ixa_ipsec_proto
= sel
.ips_protocol
;
4381 connp
->conn_ixa
->ixa_flags
&= ~IXAF_IPSEC_SECURE
;
4385 * We may or may not have policy for this endpoint. We still set
4386 * conn_policy_cached so that inbound datagrams don't have to look
4387 * at global policy as policy is considered latched for these
4388 * endpoints. We should not set conn_policy_cached until the conn
4389 * reflects the actual policy. If we *set* this before inheriting
4390 * the policy there is a window where the check
4391 * CONN_INBOUND_POLICY_PRESENT, will neither check with the policy
4392 * on the conn (because we have not yet copied the policy on to
4393 * conn and hence not set conn_in_enforce_policy) nor with the
4394 * global policy (because conn_policy_cached is already set).
4396 connp
->conn_policy_cached
= B_TRUE
;
4401 * When appropriate, this function caches outbound policy for faddr/fport.
4402 * It is used when we are not connected i.e., when we can not latch the
4406 ipsec_cache_outbound_policy(const conn_t
*connp
, const in6_addr_t
*v6src
,
4407 const in6_addr_t
*v6dst
, in_port_t dstport
, ip_xmit_attr_t
*ixa
)
4409 boolean_t isv4
= (ixa
->ixa_flags
& IXAF_IS_IPV4
) != 0;
4410 boolean_t global_policy_present
;
4411 netstack_t
*ns
= connp
->conn_netstack
;
4412 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
4414 ixa
->ixa_ipsec_policy_gen
= ipss
->ipsec_system_policy
.iph_gen
;
4417 * There is no policy caching for ICMP sockets because we can't
4418 * decide on which policy to use until we see the packet and get
4419 * type/code selectors.
4421 if (connp
->conn_proto
== IPPROTO_ICMP
||
4422 connp
->conn_proto
== IPPROTO_ICMPV6
) {
4423 ixa
->ixa_flags
&= ~IXAF_IPSEC_SECURE
;
4424 if (ixa
->ixa_ipsec_policy
!= NULL
) {
4425 IPPOL_REFRELE(ixa
->ixa_ipsec_policy
);
4426 ixa
->ixa_ipsec_policy
= NULL
;
4428 if (ixa
->ixa_ipsec_action
!= NULL
) {
4429 IPACT_REFRELE(ixa
->ixa_ipsec_action
);
4430 ixa
->ixa_ipsec_action
= NULL
;
4435 global_policy_present
= isv4
?
4436 (ipss
->ipsec_outbound_v4_policy_present
||
4437 ipss
->ipsec_inbound_v4_policy_present
) :
4438 (ipss
->ipsec_outbound_v6_policy_present
||
4439 ipss
->ipsec_inbound_v6_policy_present
);
4441 if ((connp
->conn_policy
!= NULL
) || global_policy_present
) {
4442 ipsec_selector_t sel
;
4445 bzero((void*)&sel
, sizeof (sel
));
4447 sel
.ips_protocol
= connp
->conn_proto
;
4448 sel
.ips_local_port
= connp
->conn_lport
;
4449 sel
.ips_remote_port
= dstport
;
4450 sel
.ips_is_icmp_inv_acq
= 0;
4451 sel
.ips_isv4
= isv4
;
4453 IN6_V4MAPPED_TO_IPADDR(v6src
, sel
.ips_local_addr_v4
);
4454 IN6_V4MAPPED_TO_IPADDR(v6dst
, sel
.ips_remote_addr_v4
);
4456 sel
.ips_local_addr_v6
= *v6src
;
4457 sel
.ips_remote_addr_v6
= *v6dst
;
4460 p
= ipsec_find_policy(IPSEC_TYPE_OUTBOUND
, connp
, &sel
, ns
);
4461 if (ixa
->ixa_ipsec_policy
!= NULL
)
4462 IPPOL_REFRELE(ixa
->ixa_ipsec_policy
);
4463 ixa
->ixa_ipsec_policy
= p
;
4465 ixa
->ixa_flags
|= IXAF_IPSEC_SECURE
;
4466 if (connp
->conn_policy
== NULL
)
4467 ixa
->ixa_flags
|= IXAF_IPSEC_GLOBAL_POLICY
;
4469 ixa
->ixa_flags
&= ~IXAF_IPSEC_SECURE
;
4471 /* Clear the latched actions too, in case we're recaching. */
4472 if (ixa
->ixa_ipsec_action
!= NULL
) {
4473 IPACT_REFRELE(ixa
->ixa_ipsec_action
);
4474 ixa
->ixa_ipsec_action
= NULL
;
4477 ixa
->ixa_ipsec_src_port
= sel
.ips_local_port
;
4478 ixa
->ixa_ipsec_dst_port
= sel
.ips_remote_port
;
4479 ixa
->ixa_ipsec_icmp_type
= sel
.ips_icmp_type
;
4480 ixa
->ixa_ipsec_icmp_code
= sel
.ips_icmp_code
;
4481 ixa
->ixa_ipsec_proto
= sel
.ips_protocol
;
4483 ixa
->ixa_flags
&= ~IXAF_IPSEC_SECURE
;
4484 if (ixa
->ixa_ipsec_policy
!= NULL
) {
4485 IPPOL_REFRELE(ixa
->ixa_ipsec_policy
);
4486 ixa
->ixa_ipsec_policy
= NULL
;
4488 if (ixa
->ixa_ipsec_action
!= NULL
) {
4489 IPACT_REFRELE(ixa
->ixa_ipsec_action
);
4490 ixa
->ixa_ipsec_action
= NULL
;
4496 * Returns B_FALSE if the policy has gone stale.
4499 ipsec_outbound_policy_current(ip_xmit_attr_t
*ixa
)
4501 ipsec_stack_t
*ipss
= ixa
->ixa_ipst
->ips_netstack
->netstack_ipsec
;
4503 if (!(ixa
->ixa_flags
& IXAF_IPSEC_GLOBAL_POLICY
))
4506 return (ixa
->ixa_ipsec_policy_gen
== ipss
->ipsec_system_policy
.iph_gen
);
4510 iplatch_free(ipsec_latch_t
*ipl
)
4512 if (ipl
->ipl_local_cid
!= NULL
)
4513 IPSID_REFRELE(ipl
->ipl_local_cid
);
4514 if (ipl
->ipl_remote_cid
!= NULL
)
4515 IPSID_REFRELE(ipl
->ipl_remote_cid
);
4516 mutex_destroy(&ipl
->ipl_lock
);
4517 kmem_free(ipl
, sizeof (*ipl
));
4523 ipsec_latch_t
*ipl
= kmem_alloc(sizeof (*ipl
), KM_NOSLEEP
);
4526 bzero(ipl
, sizeof (*ipl
));
4527 mutex_init(&ipl
->ipl_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
4528 ipl
->ipl_refcnt
= 1;
4533 * Hash function for ID hash table.
4536 ipsid_hash(int idtype
, char *idstring
)
4538 uint32_t hval
= idtype
;
4541 while ((c
= *idstring
++) != 0) {
4542 hval
= (hval
<< 4) | (hval
>> 28);
4545 hval
= hval
^ (hval
>> 16);
4546 return (hval
& (IPSID_HASHSIZE
-1));
4550 * Look up identity string in hash table. Return identity object
4551 * corresponding to the name -- either preexisting, or newly allocated.
4553 * Return NULL if we need to allocate a new one and can't get memory.
4556 ipsid_lookup(int idtype
, char *idstring
, netstack_t
*ns
)
4560 int idlen
= strlen(idstring
) + 1;
4561 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
4564 bucket
= &ipss
->ipsec_ipsid_buckets
[ipsid_hash(idtype
, idstring
)];
4566 mutex_enter(&bucket
->ipsif_lock
);
4568 for (retval
= bucket
->ipsif_head
; retval
!= NULL
;
4569 retval
= retval
->ipsid_next
) {
4570 if (idtype
!= retval
->ipsid_type
)
4572 if (bcmp(idstring
, retval
->ipsid_cid
, idlen
) != 0)
4575 IPSID_REFHOLD(retval
);
4576 mutex_exit(&bucket
->ipsif_lock
);
4580 retval
= kmem_alloc(sizeof (*retval
), KM_NOSLEEP
);
4582 mutex_exit(&bucket
->ipsif_lock
);
4586 nstr
= kmem_alloc(idlen
, KM_NOSLEEP
);
4588 mutex_exit(&bucket
->ipsif_lock
);
4589 kmem_free(retval
, sizeof (*retval
));
4593 retval
->ipsid_refcnt
= 1;
4594 retval
->ipsid_next
= bucket
->ipsif_head
;
4595 if (retval
->ipsid_next
!= NULL
)
4596 retval
->ipsid_next
->ipsid_ptpn
= &retval
->ipsid_next
;
4597 retval
->ipsid_ptpn
= &bucket
->ipsif_head
;
4598 retval
->ipsid_type
= idtype
;
4599 retval
->ipsid_cid
= nstr
;
4600 bucket
->ipsif_head
= retval
;
4601 bcopy(idstring
, nstr
, idlen
);
4602 mutex_exit(&bucket
->ipsif_lock
);
4608 * Garbage collect the identity hash table.
4611 ipsid_gc(netstack_t
*ns
)
4616 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
4618 for (i
= 0; i
< IPSID_HASHSIZE
; i
++) {
4619 bucket
= &ipss
->ipsec_ipsid_buckets
[i
];
4620 mutex_enter(&bucket
->ipsif_lock
);
4621 for (id
= bucket
->ipsif_head
; id
!= NULL
; id
= nid
) {
4622 nid
= id
->ipsid_next
;
4623 if (id
->ipsid_refcnt
== 0) {
4624 *id
->ipsid_ptpn
= nid
;
4626 nid
->ipsid_ptpn
= id
->ipsid_ptpn
;
4627 len
= strlen(id
->ipsid_cid
) + 1;
4628 kmem_free(id
->ipsid_cid
, len
);
4629 kmem_free(id
, sizeof (*id
));
4632 mutex_exit(&bucket
->ipsif_lock
);
4637 * Return true if two identities are the same.
4640 ipsid_equal(ipsid_t
*id1
, ipsid_t
*id2
)
4645 if ((id1
== NULL
) || (id2
== NULL
))
4648 * test that we're interning id's correctly..
4650 ASSERT((strcmp(id1
->ipsid_cid
, id2
->ipsid_cid
) != 0) ||
4651 (id1
->ipsid_type
!= id2
->ipsid_type
));
4657 * Initialize identity table; called during module initialization.
4660 ipsid_init(netstack_t
*ns
)
4664 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
4666 for (i
= 0; i
< IPSID_HASHSIZE
; i
++) {
4667 bucket
= &ipss
->ipsec_ipsid_buckets
[i
];
4668 mutex_init(&bucket
->ipsif_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
4673 * Free identity table (preparatory to module unload)
4676 ipsid_fini(netstack_t
*ns
)
4680 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
4682 for (i
= 0; i
< IPSID_HASHSIZE
; i
++) {
4683 bucket
= &ipss
->ipsec_ipsid_buckets
[i
];
4684 ASSERT(bucket
->ipsif_head
== NULL
);
4685 mutex_destroy(&bucket
->ipsif_lock
);
4690 * Update the minimum and maximum supported key sizes for the
4691 * specified algorithm. Must be called while holding the algorithms lock.
4694 ipsec_alg_fix_min_max(ipsec_alginfo_t
*alg
, ipsec_algtype_t alg_type
,
4697 size_t crypto_min
= (size_t)-1, crypto_max
= 0;
4698 size_t cur_crypto_min
, cur_crypto_max
;
4700 crypto_mechanism_info_t
*mech_infos
;
4703 crypto_mech_usage_t mask
;
4704 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
4706 ASSERT(MUTEX_HELD(&ipss
->ipsec_alg_lock
));
4709 * Compute the min, max, and default key sizes (in number of
4710 * increments to the default key size in bits) as defined
4711 * by the algorithm mappings. This range of key sizes is used
4712 * for policy related operations. The effective key sizes
4713 * supported by the framework could be more limited than
4714 * those defined for an algorithm.
4716 alg
->alg_default_bits
= alg
->alg_key_sizes
[0];
4717 alg
->alg_default
= 0;
4718 if (alg
->alg_increment
!= 0) {
4719 /* key sizes are defined by range & increment */
4720 alg
->alg_minbits
= alg
->alg_key_sizes
[1];
4721 alg
->alg_maxbits
= alg
->alg_key_sizes
[2];
4722 } else if (alg
->alg_nkey_sizes
== 0) {
4723 /* no specified key size for algorithm */
4724 alg
->alg_minbits
= alg
->alg_maxbits
= 0;
4726 /* key sizes are defined by enumeration */
4727 alg
->alg_minbits
= (uint16_t)-1;
4728 alg
->alg_maxbits
= 0;
4730 for (i
= 0; i
< alg
->alg_nkey_sizes
; i
++) {
4731 if (alg
->alg_key_sizes
[i
] < alg
->alg_minbits
)
4732 alg
->alg_minbits
= alg
->alg_key_sizes
[i
];
4733 if (alg
->alg_key_sizes
[i
] > alg
->alg_maxbits
)
4734 alg
->alg_maxbits
= alg
->alg_key_sizes
[i
];
4738 if (!(alg
->alg_flags
& ALG_FLAG_VALID
))
4742 * Mechanisms do not apply to the NULL encryption
4743 * algorithm, so simply return for this case.
4745 if (alg
->alg_id
== SADB_EALG_NULL
)
4749 * Find the min and max key sizes supported by the cryptographic
4750 * framework providers.
4753 /* get the key sizes supported by the framework */
4754 crypto_rc
= crypto_get_all_mech_info(alg
->alg_mech_type
,
4755 &mech_infos
, &nmech_infos
, KM_SLEEP
);
4756 if (crypto_rc
!= CRYPTO_SUCCESS
|| nmech_infos
== 0) {
4757 alg
->alg_flags
&= ~ALG_FLAG_VALID
;
4761 /* min and max key sizes supported by framework */
4762 for (i
= 0, is_valid
= B_FALSE
; i
< nmech_infos
; i
++) {
4766 * Ignore entries that do not support the operations
4767 * needed for the algorithm type.
4769 if (alg_type
== IPSEC_ALG_AUTH
) {
4770 mask
= CRYPTO_MECH_USAGE_MAC
;
4772 mask
= CRYPTO_MECH_USAGE_ENCRYPT
|
4773 CRYPTO_MECH_USAGE_DECRYPT
;
4775 if ((mech_infos
[i
].mi_usage
& mask
) != mask
)
4778 unit_bits
= (mech_infos
[i
].mi_keysize_unit
==
4779 CRYPTO_KEYSIZE_UNIT_IN_BYTES
) ? 8 : 1;
4780 /* adjust min/max supported by framework */
4781 cur_crypto_min
= mech_infos
[i
].mi_min_key_size
* unit_bits
;
4782 cur_crypto_max
= mech_infos
[i
].mi_max_key_size
* unit_bits
;
4784 if (cur_crypto_min
< crypto_min
)
4785 crypto_min
= cur_crypto_min
;
4788 * CRYPTO_EFFECTIVELY_INFINITE is a special value of
4789 * the crypto framework which means "no upper limit".
4791 if (mech_infos
[i
].mi_max_key_size
==
4792 CRYPTO_EFFECTIVELY_INFINITE
) {
4793 crypto_max
= (size_t)-1;
4794 } else if (cur_crypto_max
> crypto_max
) {
4795 crypto_max
= cur_crypto_max
;
4801 kmem_free(mech_infos
, sizeof (crypto_mechanism_info_t
) *
4805 /* no key sizes supported by framework */
4806 alg
->alg_flags
&= ~ALG_FLAG_VALID
;
4811 * Determine min and max key sizes from alg_key_sizes[].
4812 * defined for the algorithm entry. Adjust key sizes based on
4813 * those supported by the framework.
4815 alg
->alg_ef_default_bits
= alg
->alg_key_sizes
[0];
4818 * For backwards compatability, assume that the IV length
4819 * is the same as the data length.
4821 alg
->alg_ivlen
= alg
->alg_datalen
;
4824 * Copy any algorithm parameters (if provided) into dedicated
4825 * elements in the ipsec_alginfo_t structure.
4826 * There may be a better place to put this code.
4828 for (i
= 0; i
< alg
->alg_nparams
; i
++) {
4831 /* Initialisation Vector length (bytes) */
4832 alg
->alg_ivlen
= alg
->alg_params
[0];
4835 /* Integrity Check Vector length (bytes) */
4836 alg
->alg_icvlen
= alg
->alg_params
[1];
4839 /* Salt length (bytes) */
4840 alg
->alg_saltlen
= (uint8_t)alg
->alg_params
[2];
4847 /* Default if the IV length is not specified. */
4848 if (alg_type
== IPSEC_ALG_ENCR
&& alg
->alg_ivlen
== 0)
4849 alg
->alg_ivlen
= alg
->alg_datalen
;
4851 alg_flag_check(alg
);
4853 if (alg
->alg_increment
!= 0) {
4854 /* supported key sizes are defined by range & increment */
4855 crypto_min
= ALGBITS_ROUND_UP(crypto_min
, alg
->alg_increment
);
4856 crypto_max
= ALGBITS_ROUND_DOWN(crypto_max
, alg
->alg_increment
);
4858 alg
->alg_ef_minbits
= MAX(alg
->alg_minbits
,
4859 (uint16_t)crypto_min
);
4860 alg
->alg_ef_maxbits
= MIN(alg
->alg_maxbits
,
4861 (uint16_t)crypto_max
);
4864 * If the sizes supported by the framework are outside
4865 * the range of sizes defined by the algorithm mappings,
4866 * the algorithm cannot be used. Check for this
4869 if (alg
->alg_ef_minbits
> alg
->alg_ef_maxbits
) {
4870 alg
->alg_flags
&= ~ALG_FLAG_VALID
;
4873 if (alg
->alg_ef_default_bits
< alg
->alg_ef_minbits
)
4874 alg
->alg_ef_default_bits
= alg
->alg_ef_minbits
;
4875 if (alg
->alg_ef_default_bits
> alg
->alg_ef_maxbits
)
4876 alg
->alg_ef_default_bits
= alg
->alg_ef_maxbits
;
4877 } else if (alg
->alg_nkey_sizes
== 0) {
4878 /* no specified key size for algorithm */
4879 alg
->alg_ef_minbits
= alg
->alg_ef_maxbits
= 0;
4881 /* supported key sizes are defined by enumeration */
4882 alg
->alg_ef_minbits
= (uint16_t)-1;
4883 alg
->alg_ef_maxbits
= 0;
4885 for (i
= 0, is_valid
= B_FALSE
; i
< alg
->alg_nkey_sizes
; i
++) {
4887 * Ignore the current key size if it is not in the
4888 * range of sizes supported by the framework.
4890 if (alg
->alg_key_sizes
[i
] < crypto_min
||
4891 alg
->alg_key_sizes
[i
] > crypto_max
)
4893 if (alg
->alg_key_sizes
[i
] < alg
->alg_ef_minbits
)
4894 alg
->alg_ef_minbits
= alg
->alg_key_sizes
[i
];
4895 if (alg
->alg_key_sizes
[i
] > alg
->alg_ef_maxbits
)
4896 alg
->alg_ef_maxbits
= alg
->alg_key_sizes
[i
];
4901 alg
->alg_flags
&= ~ALG_FLAG_VALID
;
4904 alg
->alg_ef_default
= 0;
4909 * Sanity check parameters provided by ipsecalgs(1m). Assume that
4910 * the algoritm is marked as valid, there is a check at the top
4911 * of this function. If any of the checks below fail, the algorithm
4915 alg_flag_check(ipsec_alginfo_t
*alg
)
4917 alg
->alg_flags
&= ~ALG_FLAG_VALID
;
4920 * Can't have the algorithm marked as CCM and GCM.
4921 * Check the ALG_FLAG_COMBINED and ALG_FLAG_COUNTERMODE
4922 * flags are set for CCM & GCM.
4924 if ((alg
->alg_flags
& (ALG_FLAG_CCM
|ALG_FLAG_GCM
)) ==
4925 (ALG_FLAG_CCM
|ALG_FLAG_GCM
))
4927 if (alg
->alg_flags
& (ALG_FLAG_CCM
|ALG_FLAG_GCM
)) {
4928 if (!(alg
->alg_flags
& ALG_FLAG_COUNTERMODE
))
4930 if (!(alg
->alg_flags
& ALG_FLAG_COMBINED
))
4935 * For ALG_FLAG_COUNTERMODE, check the parameters
4936 * fit in the ipsec_nonce_t structure.
4938 if (alg
->alg_flags
& ALG_FLAG_COUNTERMODE
) {
4939 if (alg
->alg_ivlen
!= sizeof (((ipsec_nonce_t
*)NULL
)->iv
))
4941 if (alg
->alg_saltlen
> sizeof (((ipsec_nonce_t
*)NULL
)->salt
))
4944 if ((alg
->alg_flags
& ALG_FLAG_COMBINED
) &&
4945 (alg
->alg_icvlen
== 0))
4949 alg
->alg_flags
|= ALG_FLAG_VALID
;
4953 * Free the memory used by the specified algorithm.
4956 ipsec_alg_free(ipsec_alginfo_t
*alg
)
4961 if (alg
->alg_key_sizes
!= NULL
) {
4962 kmem_free(alg
->alg_key_sizes
,
4963 (alg
->alg_nkey_sizes
+ 1) * sizeof (uint16_t));
4964 alg
->alg_key_sizes
= NULL
;
4966 if (alg
->alg_block_sizes
!= NULL
) {
4967 kmem_free(alg
->alg_block_sizes
,
4968 (alg
->alg_nblock_sizes
+ 1) * sizeof (uint16_t));
4969 alg
->alg_block_sizes
= NULL
;
4971 if (alg
->alg_params
!= NULL
) {
4972 kmem_free(alg
->alg_params
,
4973 (alg
->alg_nparams
+ 1) * sizeof (uint16_t));
4974 alg
->alg_params
= NULL
;
4976 kmem_free(alg
, sizeof (*alg
));
4980 * Check the validity of the specified key size for an algorithm.
4981 * Returns B_TRUE if key size is valid, B_FALSE otherwise.
4984 ipsec_valid_key_size(uint16_t key_size
, ipsec_alginfo_t
*alg
)
4986 if (key_size
< alg
->alg_ef_minbits
|| key_size
> alg
->alg_ef_maxbits
)
4989 if (alg
->alg_increment
== 0 && alg
->alg_nkey_sizes
!= 0) {
4991 * If the key sizes are defined by enumeration, the new
4992 * key size must be equal to one of the supported values.
4996 for (i
= 0; i
< alg
->alg_nkey_sizes
; i
++)
4997 if (key_size
== alg
->alg_key_sizes
[i
])
4999 if (i
== alg
->alg_nkey_sizes
)
5007 * Callback function invoked by the crypto framework when a provider
5008 * registers or unregisters. This callback updates the algorithms
5009 * tables when a crypto algorithm is no longer available or becomes
5010 * available, and triggers the freeing/creation of context templates
5011 * associated with existing SAs, if needed.
5013 * Need to walk all stack instances since the callback is global
5017 ipsec_prov_update_callback(uint32_t event
, void *event_arg
)
5019 netstack_handle_t nh
;
5022 netstack_next_init(&nh
);
5023 while ((ns
= netstack_next(&nh
)) != NULL
) {
5024 ipsec_prov_update_callback_stack(event
, event_arg
, ns
);
5027 netstack_next_fini(&nh
);
5031 ipsec_prov_update_callback_stack(uint32_t event
, void *event_arg
,
5034 crypto_notify_event_change_t
*prov_change
=
5035 (crypto_notify_event_change_t
*)event_arg
;
5036 uint_t algidx
, algid
, algtype
, mech_count
, mech_idx
;
5037 ipsec_alginfo_t
*alg
;
5038 ipsec_alginfo_t oalg
;
5039 crypto_mech_name_t
*mechs
;
5040 boolean_t alg_changed
= B_FALSE
;
5041 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
5043 /* ignore events for which we didn't register */
5044 if (event
!= CRYPTO_EVENT_MECHS_CHANGED
) {
5045 ip1dbg(("ipsec_prov_update_callback: unexpected event 0x%x "
5046 " received from crypto framework\n", event
));
5050 mechs
= crypto_get_mech_list(&mech_count
, KM_SLEEP
);
5055 * Walk the list of currently defined IPsec algorithm. Update
5056 * the algorithm valid flag and trigger an update of the
5057 * SAs that depend on that algorithm.
5059 mutex_enter(&ipss
->ipsec_alg_lock
);
5060 for (algtype
= 0; algtype
< IPSEC_NALGTYPES
; algtype
++) {
5061 for (algidx
= 0; algidx
< ipss
->ipsec_nalgs
[algtype
];
5064 algid
= ipss
->ipsec_sortlist
[algtype
][algidx
];
5065 alg
= ipss
->ipsec_alglists
[algtype
][algid
];
5066 ASSERT(alg
!= NULL
);
5069 * Skip the algorithms which do not map to the
5070 * crypto framework provider being added or removed.
5072 if (strncmp(alg
->alg_mech_name
,
5073 prov_change
->ec_mech_name
,
5074 CRYPTO_MAX_MECH_NAME
) != 0)
5078 * Determine if the mechanism is valid. If it
5079 * is not, mark the algorithm as being invalid. If
5080 * it is, mark the algorithm as being valid.
5082 for (mech_idx
= 0; mech_idx
< mech_count
; mech_idx
++)
5083 if (strncmp(alg
->alg_mech_name
,
5084 mechs
[mech_idx
], CRYPTO_MAX_MECH_NAME
) == 0)
5086 if (mech_idx
== mech_count
&&
5087 alg
->alg_flags
& ALG_FLAG_VALID
) {
5088 alg
->alg_flags
&= ~ALG_FLAG_VALID
;
5089 alg_changed
= B_TRUE
;
5090 } else if (mech_idx
< mech_count
&&
5091 !(alg
->alg_flags
& ALG_FLAG_VALID
)) {
5092 alg
->alg_flags
|= ALG_FLAG_VALID
;
5093 alg_changed
= B_TRUE
;
5097 * Update the supported key sizes, regardless
5098 * of whether a crypto provider was added or
5102 ipsec_alg_fix_min_max(alg
, algtype
, ns
);
5104 alg
->alg_ef_minbits
!= oalg
.alg_ef_minbits
||
5105 alg
->alg_ef_maxbits
!= oalg
.alg_ef_maxbits
||
5106 alg
->alg_ef_default
!= oalg
.alg_ef_default
||
5107 alg
->alg_ef_default_bits
!=
5108 oalg
.alg_ef_default_bits
)
5109 alg_changed
= B_TRUE
;
5112 * Update the affected SAs if a software provider is
5113 * being added or removed.
5115 if (prov_change
->ec_provider_type
==
5117 sadb_alg_update(algtype
, alg
->alg_id
,
5118 prov_change
->ec_change
==
5119 CRYPTO_MECH_ADDED
, ns
);
5122 mutex_exit(&ipss
->ipsec_alg_lock
);
5123 crypto_free_mech_list(mechs
, mech_count
);
5127 * An algorithm has changed, i.e. it became valid or
5128 * invalid, or its support key sizes have changed.
5129 * Notify ipsecah and ipsecesp of this change so
5130 * that they can send a SADB_REGISTER to their consumers.
5132 ipsecah_algs_changed(ns
);
5133 ipsecesp_algs_changed(ns
);
5138 * Registers with the crypto framework to be notified of crypto
5139 * providers changes. Used to update the algorithm tables and
5140 * to free or create context templates if needed. Invoked after IPsec
5141 * is loaded successfully.
5143 * This is called separately for each IP instance, so we ensure we only
5147 ipsec_register_prov_update(void)
5149 if (prov_update_handle
!= NULL
)
5152 prov_update_handle
= crypto_notify_events(
5153 ipsec_prov_update_callback
, CRYPTO_EVENT_MECHS_CHANGED
);
5157 * Unregisters from the framework to be notified of crypto providers
5158 * changes. Called from ipsec_policy_g_destroy().
5161 ipsec_unregister_prov_update(void)
5163 if (prov_update_handle
!= NULL
)
5164 crypto_unnotify_events(prov_update_handle
);
5168 * Tunnel-mode support routines.
5172 * Returns an mblk chain suitable for putnext() if policies match and IPsec
5173 * SAs are available. If there's no per-tunnel policy, or a match comes back
5174 * with no match, then still return the packet and have global policy take
5175 * a crack at it in IP.
5176 * This updates the ip_xmit_attr with the IPsec policy.
5178 * Remember -> we can be forwarding packets. Keep that in mind w.r.t.
5179 * inner-packet contents.
5182 ipsec_tun_outbound(mblk_t
*mp
, iptun_t
*iptun
, ipha_t
*inner_ipv4
,
5183 ip6_t
*inner_ipv6
, ipha_t
*outer_ipv4
, ip6_t
*outer_ipv6
, int outer_hdr_len
,
5184 ip_xmit_attr_t
*ixa
)
5186 ipsec_policy_head_t
*polhead
;
5187 ipsec_selector_t sel
;
5189 boolean_t is_fragment
;
5190 ipsec_policy_t
*pol
;
5191 ipsec_tun_pol_t
*itp
= iptun
->iptun_itp
;
5192 netstack_t
*ns
= iptun
->iptun_ns
;
5193 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
5195 ASSERT(outer_ipv6
!= NULL
&& outer_ipv4
== NULL
||
5196 outer_ipv4
!= NULL
&& outer_ipv6
== NULL
);
5197 /* We take care of inners in a bit. */
5199 /* Are the IPsec fields initialized at all? */
5200 if (!(ixa
->ixa_flags
& IXAF_IPSEC_SECURE
)) {
5201 ASSERT(ixa
->ixa_ipsec_policy
== NULL
);
5202 ASSERT(ixa
->ixa_ipsec_latch
== NULL
);
5203 ASSERT(ixa
->ixa_ipsec_action
== NULL
);
5204 ASSERT(ixa
->ixa_ipsec_ah_sa
== NULL
);
5205 ASSERT(ixa
->ixa_ipsec_esp_sa
== NULL
);
5208 ASSERT(itp
!= NULL
&& (itp
->itp_flags
& ITPF_P_ACTIVE
));
5209 polhead
= itp
->itp_policy
;
5211 bzero(&sel
, sizeof (sel
));
5212 if (inner_ipv4
!= NULL
) {
5213 ASSERT(inner_ipv6
== NULL
);
5214 sel
.ips_isv4
= B_TRUE
;
5215 sel
.ips_local_addr_v4
= inner_ipv4
->ipha_src
;
5216 sel
.ips_remote_addr_v4
= inner_ipv4
->ipha_dst
;
5217 sel
.ips_protocol
= (uint8_t)inner_ipv4
->ipha_protocol
;
5219 ASSERT(inner_ipv6
!= NULL
);
5220 sel
.ips_isv4
= B_FALSE
;
5221 sel
.ips_local_addr_v6
= inner_ipv6
->ip6_src
;
5223 * We don't care about routing-header dests in the
5224 * forwarding/tunnel path, so just grab ip6_dst.
5226 sel
.ips_remote_addr_v6
= inner_ipv6
->ip6_dst
;
5229 if (itp
->itp_flags
& ITPF_P_PER_PORT_SECURITY
) {
5231 * Caller can prepend the outer header, which means
5232 * inner_ipv[46] may be stuck in the middle. Pullup the whole
5233 * mess now if need-be, for easier processing later. Don't
5234 * forget to rewire the outer header too.
5236 if (mp
->b_cont
!= NULL
) {
5237 nmp
= msgpullup(mp
, -1);
5239 ip_drop_packet(mp
, B_FALSE
, NULL
,
5240 DROPPER(ipss
, ipds_spd_nomem
),
5241 &ipss
->ipsec_spd_dropper
);
5246 if (outer_ipv4
!= NULL
)
5247 outer_ipv4
= (ipha_t
*)mp
->b_rptr
;
5249 outer_ipv6
= (ip6_t
*)mp
->b_rptr
;
5250 if (inner_ipv4
!= NULL
) {
5252 (ipha_t
*)(mp
->b_rptr
+ outer_hdr_len
);
5255 (ip6_t
*)(mp
->b_rptr
+ outer_hdr_len
);
5258 if (inner_ipv4
!= NULL
) {
5259 is_fragment
= IS_V4_FRAGMENT(
5260 inner_ipv4
->ipha_fragment_offset_and_flags
);
5262 sel
.ips_remote_addr_v6
= ip_get_dst_v6(inner_ipv6
, mp
,
5271 uint16_t ip6_hdr_length
;
5273 uint8_t *v6_proto_p
;
5276 * We have a fragment we need to track!
5278 mp
= ipsec_fragcache_add(&itp
->itp_fragcache
, NULL
, mp
,
5279 outer_hdr_len
, ipss
);
5282 ASSERT(mp
->b_cont
== NULL
);
5285 * If we get here, we have a full fragment chain
5288 oiph
= (ipha_t
*)mp
->b_rptr
;
5289 if (IPH_HDR_VERSION(oiph
) == IPV4_VERSION
) {
5290 hdr_len
= ((outer_hdr_len
!= 0) ?
5291 IPH_HDR_LENGTH(oiph
) : 0);
5292 iph
= (ipha_t
*)(mp
->b_rptr
+ hdr_len
);
5294 ASSERT(IPH_HDR_VERSION(oiph
) == IPV6_VERSION
);
5295 ip6h
= (ip6_t
*)mp
->b_rptr
;
5296 if (!ip_hdr_length_nexthdr_v6(mp
, ip6h
,
5297 &ip6_hdr_length
, &v6_proto_p
)) {
5298 ip_drop_packet_chain(mp
, B_FALSE
, NULL
,
5300 ipds_spd_malformed_packet
),
5301 &ipss
->ipsec_spd_dropper
);
5304 hdr_len
= ip6_hdr_length
;
5306 outer_hdr_len
= hdr_len
;
5311 iph
= (ipha_t
*)(mp
->b_rptr
+ hdr_len
);
5314 sel
.ips_local_addr_v4
= inner_ipv4
->ipha_src
;
5315 sel
.ips_remote_addr_v4
= inner_ipv4
->ipha_dst
;
5317 (uint8_t)inner_ipv4
->ipha_protocol
;
5319 inner_ipv6
= (ip6_t
*)(mp
->b_rptr
+
5321 sel
.ips_local_addr_v6
= inner_ipv6
->ip6_src
;
5322 sel
.ips_remote_addr_v6
= inner_ipv6
->ip6_dst
;
5323 if (!ip_hdr_length_nexthdr_v6(mp
,
5324 inner_ipv6
, &ip6_hdr_length
, &v6_proto_p
)) {
5325 ip_drop_packet_chain(mp
, B_FALSE
, NULL
,
5327 ipds_spd_malformed_frag
),
5328 &ipss
->ipsec_spd_dropper
);
5331 v6_proto
= *v6_proto_p
;
5332 sel
.ips_protocol
= v6_proto
;
5333 #ifdef FRAGCACHE_DEBUG
5334 cmn_err(CE_WARN
, "v6_sel.ips_protocol = %d\n",
5338 /* Ports are extracted below */
5342 if (!ipsec_init_outbound_ports(&sel
, mp
,
5343 inner_ipv4
, inner_ipv6
, outer_hdr_len
, ipss
)) {
5344 /* callee did ip_drop_packet_chain() on mp. */
5347 #ifdef FRAGCACHE_DEBUG
5348 if (inner_ipv4
!= NULL
)
5350 "(v4) sel.ips_protocol = %d, "
5351 "sel.ips_local_port = %d, "
5352 "sel.ips_remote_port = %d\n",
5353 sel
.ips_protocol
, ntohs(sel
.ips_local_port
),
5354 ntohs(sel
.ips_remote_port
));
5355 if (inner_ipv6
!= NULL
)
5357 "(v6) sel.ips_protocol = %d, "
5358 "sel.ips_local_port = %d, "
5359 "sel.ips_remote_port = %d\n",
5360 sel
.ips_protocol
, ntohs(sel
.ips_local_port
),
5361 ntohs(sel
.ips_remote_port
));
5363 /* Success so far! */
5365 rw_enter(&polhead
->iph_lock
, RW_READER
);
5366 pol
= ipsec_find_policy_head(NULL
, polhead
, IPSEC_TYPE_OUTBOUND
, &sel
);
5367 rw_exit(&polhead
->iph_lock
);
5370 * No matching policy on this tunnel, drop the packet.
5372 * NOTE: Tunnel-mode tunnels are different from the
5373 * IP global transport mode policy head. For a tunnel-mode
5374 * tunnel, we drop the packet in lieu of passing it
5375 * along accepted the way a global-policy miss would.
5377 * NOTE2: "negotiate transport" tunnels should match ALL
5378 * inbound packets, but we do not uncomment the ASSERT()
5379 * below because if/when we open PF_POLICY, a user can
5380 * shoot him/her-self in the foot with a 0 priority.
5383 /* ASSERT(itp->itp_flags & ITPF_P_TUNNEL); */
5384 #ifdef FRAGCACHE_DEBUG
5385 cmn_err(CE_WARN
, "ipsec_tun_outbound(): No matching tunnel "
5386 "per-port policy\n");
5388 ip_drop_packet_chain(mp
, B_FALSE
, NULL
,
5389 DROPPER(ipss
, ipds_spd_explicit
),
5390 &ipss
->ipsec_spd_dropper
);
5394 #ifdef FRAGCACHE_DEBUG
5395 cmn_err(CE_WARN
, "Having matching tunnel per-port policy\n");
5399 * NOTE: ixa_cleanup() function will release pol references.
5401 ixa
->ixa_ipsec_policy
= pol
;
5403 * NOTE: There is a subtle difference between iptun_zoneid and
5404 * iptun_connp->conn_zoneid explained in iptun_conn_create(). When
5405 * interacting with the ip module, we must use conn_zoneid.
5407 ixa
->ixa_zoneid
= iptun
->iptun_connp
->conn_zoneid
;
5409 ASSERT((outer_ipv4
!= NULL
) ? (ixa
->ixa_flags
& IXAF_IS_IPV4
) :
5410 !(ixa
->ixa_flags
& IXAF_IS_IPV4
));
5411 ASSERT(ixa
->ixa_ipsec_policy
!= NULL
);
5412 ixa
->ixa_flags
|= IXAF_IPSEC_SECURE
;
5414 if (!(itp
->itp_flags
& ITPF_P_TUNNEL
)) {
5415 /* Set up transport mode for tunnelled packets. */
5416 ixa
->ixa_ipsec_proto
= (inner_ipv4
!= NULL
) ? IPPROTO_ENCAP
:
5421 /* Fill in tunnel-mode goodies here. */
5422 ixa
->ixa_flags
|= IXAF_IPSEC_TUNNEL
;
5423 /* XXX Do I need to fill in all of the goodies here? */
5425 ixa
->ixa_ipsec_inaf
= AF_INET
;
5426 ixa
->ixa_ipsec_insrc
[0] =
5427 pol
->ipsp_sel
->ipsl_key
.ipsl_local
.ipsad_v4
;
5428 ixa
->ixa_ipsec_indst
[0] =
5429 pol
->ipsp_sel
->ipsl_key
.ipsl_remote
.ipsad_v4
;
5431 ixa
->ixa_ipsec_inaf
= AF_INET6
;
5432 ixa
->ixa_ipsec_insrc
[0] =
5433 pol
->ipsp_sel
->ipsl_key
.ipsl_local
.ipsad_v6
.s6_addr32
[0];
5434 ixa
->ixa_ipsec_insrc
[1] =
5435 pol
->ipsp_sel
->ipsl_key
.ipsl_local
.ipsad_v6
.s6_addr32
[1];
5436 ixa
->ixa_ipsec_insrc
[2] =
5437 pol
->ipsp_sel
->ipsl_key
.ipsl_local
.ipsad_v6
.s6_addr32
[2];
5438 ixa
->ixa_ipsec_insrc
[3] =
5439 pol
->ipsp_sel
->ipsl_key
.ipsl_local
.ipsad_v6
.s6_addr32
[3];
5440 ixa
->ixa_ipsec_indst
[0] =
5441 pol
->ipsp_sel
->ipsl_key
.ipsl_remote
.ipsad_v6
.s6_addr32
[0];
5442 ixa
->ixa_ipsec_indst
[1] =
5443 pol
->ipsp_sel
->ipsl_key
.ipsl_remote
.ipsad_v6
.s6_addr32
[1];
5444 ixa
->ixa_ipsec_indst
[2] =
5445 pol
->ipsp_sel
->ipsl_key
.ipsl_remote
.ipsad_v6
.s6_addr32
[2];
5446 ixa
->ixa_ipsec_indst
[3] =
5447 pol
->ipsp_sel
->ipsl_key
.ipsl_remote
.ipsad_v6
.s6_addr32
[3];
5449 ixa
->ixa_ipsec_insrcpfx
= pol
->ipsp_sel
->ipsl_key
.ipsl_local_pfxlen
;
5450 ixa
->ixa_ipsec_indstpfx
= pol
->ipsp_sel
->ipsl_key
.ipsl_remote_pfxlen
;
5451 /* NOTE: These are used for transport mode too. */
5452 ixa
->ixa_ipsec_src_port
= pol
->ipsp_sel
->ipsl_key
.ipsl_lport
;
5453 ixa
->ixa_ipsec_dst_port
= pol
->ipsp_sel
->ipsl_key
.ipsl_rport
;
5454 ixa
->ixa_ipsec_proto
= pol
->ipsp_sel
->ipsl_key
.ipsl_proto
;
5460 * NOTE: The following releases pol's reference and
5461 * calls ip_drop_packet() for me on NULL returns.
5464 ipsec_check_ipsecin_policy_reasm(mblk_t
*attr_mp
, ipsec_policy_t
*pol
,
5465 ipha_t
*inner_ipv4
, ip6_t
*inner_ipv6
, uint64_t pkt_unique
, netstack_t
*ns
)
5467 /* Assume attr_mp is a chain of b_next-linked ip_recv_attr mblk. */
5468 mblk_t
*data_chain
= NULL
, *data_tail
= NULL
;
5471 ip_recv_attr_t iras
;
5473 while (attr_mp
!= NULL
) {
5474 ASSERT(ip_recv_attr_is_mblk(attr_mp
));
5475 next
= attr_mp
->b_next
;
5476 attr_mp
->b_next
= NULL
; /* No tripping asserts. */
5478 data_mp
= attr_mp
->b_cont
;
5479 attr_mp
->b_cont
= NULL
;
5480 if (!ip_recv_attr_from_mblk(attr_mp
, &iras
)) {
5481 /* The ill or ip_stack_t disappeared on us */
5482 freemsg(data_mp
); /* ip_drop_packet?? */
5483 ira_cleanup(&iras
, B_TRUE
);
5488 * Need IPPOL_REFHOLD(pol) for extras because
5489 * ipsecin_policy does the refrele.
5493 data_mp
= ipsec_check_ipsecin_policy(data_mp
, pol
, inner_ipv4
,
5494 inner_ipv6
, pkt_unique
, &iras
, ns
);
5495 ira_cleanup(&iras
, B_TRUE
);
5497 if (data_mp
== NULL
)
5500 if (data_tail
== NULL
) {
5502 data_chain
= data_tail
= data_mp
;
5504 data_tail
->b_next
= data_mp
;
5505 data_tail
= data_mp
;
5510 * One last release because either the loop bumped it up, or we never
5511 * called ipsec_check_ipsecin_policy().
5515 /* data_chain is ready for return to tun module. */
5516 return (data_chain
);
5520 * Need to get rid of any extra pol
5521 * references, and any remaining bits as well.
5524 ipsec_freemsg_chain(data_chain
);
5525 ipsec_freemsg_chain(next
); /* ipdrop stats? */
5530 * Return a message if the inbound packet passed an IPsec policy check. Returns
5531 * NULL if it failed or if it is a fragment needing its friends before a
5532 * policy check can be performed.
5534 * Expects a non-NULL data_mp, and a non-NULL polhead.
5535 * The returned mblk may be a b_next chain of packets if fragments
5536 * neeeded to be collected for a proper policy check.
5538 * This function calls ip_drop_packet() on data_mp if need be.
5540 * NOTE: outer_hdr_len is signed. If it's a negative value, the caller
5541 * is inspecting an ICMP packet.
5544 ipsec_tun_inbound(ip_recv_attr_t
*ira
, mblk_t
*data_mp
, ipsec_tun_pol_t
*itp
,
5545 ipha_t
*inner_ipv4
, ip6_t
*inner_ipv6
, ipha_t
*outer_ipv4
,
5546 ip6_t
*outer_ipv6
, int outer_hdr_len
, netstack_t
*ns
)
5548 ipsec_policy_head_t
*polhead
;
5549 ipsec_selector_t sel
;
5550 ipsec_policy_t
*pol
;
5553 boolean_t port_policy_present
, is_icmp
, global_present
;
5556 uint8_t flags
, *inner_hdr
;
5557 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
5559 sel
.ips_is_icmp_inv_acq
= 0;
5561 if (outer_ipv4
!= NULL
) {
5562 ASSERT(outer_ipv6
== NULL
);
5563 global_present
= ipss
->ipsec_inbound_v4_policy_present
;
5565 ASSERT(outer_ipv6
!= NULL
);
5566 global_present
= ipss
->ipsec_inbound_v6_policy_present
;
5569 ASSERT(inner_ipv4
!= NULL
&& inner_ipv6
== NULL
||
5570 inner_ipv4
== NULL
&& inner_ipv6
!= NULL
);
5572 if (outer_hdr_len
< 0) {
5573 outer_hdr_len
= (-outer_hdr_len
);
5579 if (itp
!= NULL
&& (itp
->itp_flags
& ITPF_P_ACTIVE
)) {
5580 mblk_t
*mp
= data_mp
;
5582 polhead
= itp
->itp_policy
;
5584 * We need to perform full Tunnel-Mode enforcement,
5585 * and we need to have inner-header data for such enforcement.
5587 * See ipsec_init_inbound_sel() for the 0x80000000 on inbound
5591 port_policy_present
= ((itp
->itp_flags
&
5592 ITPF_P_PER_PORT_SECURITY
) ? B_TRUE
: B_FALSE
);
5594 * NOTE: Even if our policy is transport mode, set the
5595 * SEL_TUNNEL_MODE flag so ipsec_init_inbound_sel() can
5596 * do the right thing w.r.t. outer headers.
5598 flags
= ((port_policy_present
? SEL_PORT_POLICY
: SEL_NONE
) |
5599 (is_icmp
? SEL_IS_ICMP
: SEL_NONE
) | SEL_TUNNEL_MODE
);
5601 rc
= ipsec_init_inbound_sel(&sel
, data_mp
, inner_ipv4
,
5606 ip_drop_packet(data_mp
, B_TRUE
, NULL
,
5607 DROPPER(ipss
, ipds_spd_nomem
),
5608 &ipss
->ipsec_spd_dropper
);
5610 case SELRET_TUNFRAG
:
5612 * At this point, if we're cleartext, we don't want
5615 if (!(ira
->ira_flags
& IRAF_IPSEC_SECURE
)) {
5616 ip_drop_packet(data_mp
, B_TRUE
, NULL
,
5617 DROPPER(ipss
, ipds_spd_got_clear
),
5618 &ipss
->ipsec_spd_dropper
);
5623 * Inner and outer headers may not be contiguous.
5624 * Pullup the data_mp now to satisfy assumptions of
5625 * ipsec_fragcache_add()
5627 if (data_mp
->b_cont
!= NULL
) {
5630 nmp
= msgpullup(data_mp
, -1);
5632 ip_drop_packet(data_mp
, B_TRUE
, NULL
,
5633 DROPPER(ipss
, ipds_spd_nomem
),
5634 &ipss
->ipsec_spd_dropper
);
5639 if (outer_ipv4
!= NULL
)
5641 (ipha_t
*)data_mp
->b_rptr
;
5644 (ip6_t
*)data_mp
->b_rptr
;
5645 if (inner_ipv4
!= NULL
) {
5647 (ipha_t
*)(data_mp
->b_rptr
+
5651 (ip6_t
*)(data_mp
->b_rptr
+
5657 * If we need to queue the packet. First we
5658 * get an mblk with the attributes. ipsec_fragcache_add
5659 * will prepend that to the queued data and return
5660 * a list of b_next messages each of which starts with
5661 * the attribute mblk.
5663 mp
= ip_recv_attr_to_mblk(ira
);
5665 ip_drop_packet(data_mp
, B_TRUE
, NULL
,
5666 DROPPER(ipss
, ipds_spd_nomem
),
5667 &ipss
->ipsec_spd_dropper
);
5671 mp
= ipsec_fragcache_add(&itp
->itp_fragcache
,
5672 mp
, data_mp
, outer_hdr_len
, ipss
);
5676 * Data is cached, fragment chain is not
5683 * If we get here, we have a full fragment chain.
5684 * Reacquire headers and selectors from first fragment.
5686 ASSERT(ip_recv_attr_is_mblk(mp
));
5687 data_mp
= mp
->b_cont
;
5688 inner_hdr
= data_mp
->b_rptr
;
5689 if (outer_ipv4
!= NULL
) {
5690 inner_hdr
+= IPH_HDR_LENGTH(
5691 (ipha_t
*)data_mp
->b_rptr
);
5693 inner_hdr
+= ip_hdr_length_v6(data_mp
,
5694 (ip6_t
*)data_mp
->b_rptr
);
5696 ASSERT(inner_hdr
<= data_mp
->b_wptr
);
5698 if (inner_ipv4
!= NULL
) {
5699 inner_ipv4
= (ipha_t
*)inner_hdr
;
5702 inner_ipv6
= (ip6_t
*)inner_hdr
;
5707 * Use SEL_TUNNEL_MODE to take into account the outer
5708 * header. Use SEL_POST_FRAG so we always get ports.
5710 rc
= ipsec_init_inbound_sel(&sel
, data_mp
,
5711 inner_ipv4
, inner_ipv6
,
5712 SEL_TUNNEL_MODE
| SEL_POST_FRAG
);
5714 case SELRET_SUCCESS
:
5716 * Get to same place as first caller's
5717 * SELRET_SUCCESS case.
5721 ip_drop_packet_chain(mp
, B_TRUE
, NULL
,
5722 DROPPER(ipss
, ipds_spd_nomem
),
5723 &ipss
->ipsec_spd_dropper
);
5726 ip_drop_packet_chain(mp
, B_TRUE
, NULL
,
5727 DROPPER(ipss
, ipds_spd_malformed_frag
),
5728 &ipss
->ipsec_spd_dropper
);
5730 case SELRET_TUNFRAG
:
5731 cmn_err(CE_WARN
, "(TUNFRAG on 2nd call...)");
5734 cmn_err(CE_WARN
, "ipsec_init_inbound_sel(mark2)"
5735 " returns bizarro 0x%x", rc
);
5736 /* Guaranteed panic! */
5737 ASSERT(rc
== SELRET_NOMEM
);
5741 case SELRET_SUCCESS
:
5744 * No per-port policy or a non-fragment. Keep going.
5749 * We may receive ICMP (with IPv6 inner) packets that
5750 * trigger this return value. Send 'em in for
5751 * enforcement checking.
5753 cmn_err(CE_NOTE
, "ipsec_tun_inbound(): "
5754 "sending 'bad packet' in for enforcement");
5758 "ipsec_init_inbound_sel() returns bizarro 0x%x",
5760 ASSERT(rc
== SELRET_NOMEM
); /* Guaranteed panic! */
5766 * Swap local/remote because this is an ICMP packet.
5768 tmpaddr
= sel
.ips_local_addr_v6
;
5769 sel
.ips_local_addr_v6
= sel
.ips_remote_addr_v6
;
5770 sel
.ips_remote_addr_v6
= tmpaddr
;
5771 tmpport
= sel
.ips_local_port
;
5772 sel
.ips_local_port
= sel
.ips_remote_port
;
5773 sel
.ips_remote_port
= tmpport
;
5776 /* find_policy_head() */
5777 rw_enter(&polhead
->iph_lock
, RW_READER
);
5778 pol
= ipsec_find_policy_head(NULL
, polhead
, IPSEC_TYPE_INBOUND
,
5780 rw_exit(&polhead
->iph_lock
);
5782 uint64_t pkt_unique
;
5784 if (!(ira
->ira_flags
& IRAF_IPSEC_SECURE
)) {
5785 if (!pol
->ipsp_act
->ipa_allow_clear
) {
5787 * XXX should never get here with
5788 * tunnel reassembled fragments?
5790 ASSERT(mp
== data_mp
);
5791 ip_drop_packet(data_mp
, B_TRUE
, NULL
,
5792 DROPPER(ipss
, ipds_spd_got_clear
),
5793 &ipss
->ipsec_spd_dropper
);
5801 pkt_unique
= SA_UNIQUE_ID(sel
.ips_remote_port
,
5803 (inner_ipv4
== NULL
) ? IPPROTO_IPV6
:
5804 IPPROTO_ENCAP
, sel
.ips_protocol
);
5807 * NOTE: The following releases pol's reference and
5808 * calls ip_drop_packet() for me on NULL returns.
5810 * "sel" is still good here, so let's use it!
5812 if (data_mp
== mp
) {
5813 /* A single packet without attributes */
5814 data_mp
= ipsec_check_ipsecin_policy(data_mp
,
5815 pol
, inner_ipv4
, inner_ipv6
, pkt_unique
,
5819 * We pass in the b_next chain of attr_mp's
5820 * and get back a b_next chain of data_mp's.
5822 data_mp
= ipsec_check_ipsecin_policy_reasm(mp
,
5823 pol
, inner_ipv4
, inner_ipv6
, pkt_unique
,
5830 * Else fallthru and check the global policy on the outer
5831 * header(s) if this tunnel is an old-style transport-mode
5832 * one. Drop the packet explicitly (no policy entry) for
5833 * a new-style tunnel-mode tunnel.
5835 if ((itp
->itp_flags
& ITPF_P_TUNNEL
) && !is_icmp
) {
5836 ip_drop_packet_chain(data_mp
, B_TRUE
, NULL
,
5837 DROPPER(ipss
, ipds_spd_explicit
),
5838 &ipss
->ipsec_spd_dropper
);
5844 * NOTE: If we reach here, we will not have packet chains from
5845 * fragcache_add(), because the only way I get chains is on a
5846 * tunnel-mode tunnel, which either returns with a pass, or gets
5847 * hit by the ip_drop_packet_chain() call right above here.
5849 ASSERT(data_mp
->b_next
== NULL
);
5851 /* If no per-tunnel security, check global policy now. */
5852 if ((ira
->ira_flags
& IRAF_IPSEC_SECURE
) && !global_present
) {
5853 if (ira
->ira_flags
& IRAF_TRUSTED_ICMP
) {
5855 * This is an ICMP message that was geenrated locally.
5856 * We should accept it.
5861 ip_drop_packet(data_mp
, B_TRUE
, NULL
,
5862 DROPPER(ipss
, ipds_spd_got_secure
),
5863 &ipss
->ipsec_spd_dropper
);
5869 * For ICMP packets, "outer_ipvN" is set to the outer header
5870 * that is *INSIDE* the ICMP payload. For global policy
5871 * checking, we need to reverse src/dst on the payload in
5872 * order to construct selectors appropriately. See "ripha"
5873 * constructions in ip.c. To avoid a bug like 6478464 (see
5874 * earlier in this file), we will actually exchange src/dst
5875 * in the packet, and reverse if after the call to
5876 * ipsec_check_global_policy().
5878 if (outer_ipv4
!= NULL
) {
5879 tmp4
= outer_ipv4
->ipha_src
;
5880 outer_ipv4
->ipha_src
= outer_ipv4
->ipha_dst
;
5881 outer_ipv4
->ipha_dst
= tmp4
;
5883 ASSERT(outer_ipv6
!= NULL
);
5884 tmpaddr
= outer_ipv6
->ip6_src
;
5885 outer_ipv6
->ip6_src
= outer_ipv6
->ip6_dst
;
5886 outer_ipv6
->ip6_dst
= tmpaddr
;
5890 data_mp
= ipsec_check_global_policy(data_mp
, NULL
, outer_ipv4
,
5891 outer_ipv6
, ira
, ns
);
5892 if (data_mp
== NULL
)
5896 /* Set things back to normal. */
5897 if (outer_ipv4
!= NULL
) {
5898 tmp4
= outer_ipv4
->ipha_src
;
5899 outer_ipv4
->ipha_src
= outer_ipv4
->ipha_dst
;
5900 outer_ipv4
->ipha_dst
= tmp4
;
5902 /* No need for ASSERT()s now. */
5903 tmpaddr
= outer_ipv6
->ip6_src
;
5904 outer_ipv6
->ip6_src
= outer_ipv6
->ip6_dst
;
5905 outer_ipv6
->ip6_dst
= tmpaddr
;
5910 * At this point, we pretend it's a cleartext accepted
5917 * AVL comparison routine for our list of tunnel polheads.
5920 tunnel_compare(const void *arg1
, const void *arg2
)
5922 ipsec_tun_pol_t
*left
, *right
;
5925 left
= (ipsec_tun_pol_t
*)arg1
;
5926 right
= (ipsec_tun_pol_t
*)arg2
;
5928 rc
= strncmp(left
->itp_name
, right
->itp_name
, LIFNAMSIZ
);
5929 return (rc
== 0 ? rc
: (rc
> 0 ? 1 : -1));
5933 * Free a tunnel policy node.
5936 itp_free(ipsec_tun_pol_t
*node
, netstack_t
*ns
)
5938 if (node
->itp_policy
!= NULL
) {
5939 IPPH_REFRELE(node
->itp_policy
, ns
);
5940 node
->itp_policy
= NULL
;
5942 if (node
->itp_inactive
!= NULL
) {
5943 IPPH_REFRELE(node
->itp_inactive
, ns
);
5944 node
->itp_inactive
= NULL
;
5946 mutex_destroy(&node
->itp_lock
);
5947 kmem_free(node
, sizeof (*node
));
5951 itp_unlink(ipsec_tun_pol_t
*node
, netstack_t
*ns
)
5953 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
5955 rw_enter(&ipss
->ipsec_tunnel_policy_lock
, RW_WRITER
);
5956 ipss
->ipsec_tunnel_policy_gen
++;
5957 ipsec_fragcache_uninit(&node
->itp_fragcache
, ipss
);
5958 avl_remove(&ipss
->ipsec_tunnel_policies
, node
);
5959 rw_exit(&ipss
->ipsec_tunnel_policy_lock
);
5960 ITP_REFRELE(node
, ns
);
5964 * Public interface to look up a tunnel security policy by name. Used by
5965 * spdsock mostly. Returns "node" with a bumped refcnt.
5968 get_tunnel_policy(char *name
, netstack_t
*ns
)
5970 ipsec_tun_pol_t
*node
, lookup
;
5971 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
5973 (void) strncpy(lookup
.itp_name
, name
, LIFNAMSIZ
);
5975 rw_enter(&ipss
->ipsec_tunnel_policy_lock
, RW_READER
);
5976 node
= (ipsec_tun_pol_t
*)avl_find(&ipss
->ipsec_tunnel_policies
,
5981 rw_exit(&ipss
->ipsec_tunnel_policy_lock
);
5987 * Public interface to walk all tunnel security polcies. Useful for spdsock
5988 * DUMP operations. iterator() will not consume a reference.
5991 itp_walk(void (*iterator
)(ipsec_tun_pol_t
*, void *, netstack_t
*),
5992 void *arg
, netstack_t
*ns
)
5994 ipsec_tun_pol_t
*node
;
5995 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
5997 rw_enter(&ipss
->ipsec_tunnel_policy_lock
, RW_READER
);
5998 for (node
= avl_first(&ipss
->ipsec_tunnel_policies
); node
!= NULL
;
5999 node
= AVL_NEXT(&ipss
->ipsec_tunnel_policies
, node
)) {
6000 iterator(node
, arg
, ns
);
6002 rw_exit(&ipss
->ipsec_tunnel_policy_lock
);
6006 * Initialize policy head. This can only fail if there's a memory problem.
6009 tunnel_polhead_init(ipsec_policy_head_t
*iph
, netstack_t
*ns
)
6011 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
6013 rw_init(&iph
->iph_lock
, NULL
, RW_DEFAULT
, NULL
);
6016 if (ipsec_alloc_table(iph
, ipss
->ipsec_tun_spd_hashsize
,
6017 KM_SLEEP
, B_FALSE
, ns
) != 0) {
6018 ipsec_polhead_free_table(iph
);
6021 ipsec_polhead_init(iph
, ipss
->ipsec_tun_spd_hashsize
);
6026 * Create a tunnel policy node with "name". Set errno with
6027 * ENOMEM if there's a memory problem, and EEXIST if there's an existing
6031 create_tunnel_policy(char *name
, int *errno
, uint64_t *gen
, netstack_t
*ns
)
6033 ipsec_tun_pol_t
*newbie
, *existing
;
6035 ipsec_stack_t
*ipss
= ns
->netstack_ipsec
;
6037 newbie
= kmem_zalloc(sizeof (*newbie
), KM_NOSLEEP
);
6038 if (newbie
== NULL
) {
6042 if (!ipsec_fragcache_init(&newbie
->itp_fragcache
)) {
6043 kmem_free(newbie
, sizeof (*newbie
));
6048 (void) strncpy(newbie
->itp_name
, name
, LIFNAMSIZ
);
6050 rw_enter(&ipss
->ipsec_tunnel_policy_lock
, RW_WRITER
);
6051 existing
= (ipsec_tun_pol_t
*)avl_find(&ipss
->ipsec_tunnel_policies
,
6053 if (existing
!= NULL
) {
6054 itp_free(newbie
, ns
);
6056 rw_exit(&ipss
->ipsec_tunnel_policy_lock
);
6059 ipss
->ipsec_tunnel_policy_gen
++;
6060 *gen
= ipss
->ipsec_tunnel_policy_gen
;
6061 newbie
->itp_refcnt
= 2; /* One for the caller, one for the tree. */
6062 newbie
->itp_next_policy_index
= 1;
6063 avl_insert(&ipss
->ipsec_tunnel_policies
, newbie
, where
);
6064 mutex_init(&newbie
->itp_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6065 newbie
->itp_policy
= kmem_zalloc(sizeof (ipsec_policy_head_t
),
6067 if (newbie
->itp_policy
== NULL
)
6069 newbie
->itp_inactive
= kmem_zalloc(sizeof (ipsec_policy_head_t
),
6071 if (newbie
->itp_inactive
== NULL
) {
6072 kmem_free(newbie
->itp_policy
, sizeof (ipsec_policy_head_t
));
6076 if (!tunnel_polhead_init(newbie
->itp_policy
, ns
)) {
6077 kmem_free(newbie
->itp_policy
, sizeof (ipsec_policy_head_t
));
6078 kmem_free(newbie
->itp_inactive
, sizeof (ipsec_policy_head_t
));
6080 } else if (!tunnel_polhead_init(newbie
->itp_inactive
, ns
)) {
6081 IPPH_REFRELE(newbie
->itp_policy
, ns
);
6082 kmem_free(newbie
->itp_inactive
, sizeof (ipsec_policy_head_t
));
6085 rw_exit(&ipss
->ipsec_tunnel_policy_lock
);
6090 kmem_free(newbie
, sizeof (*newbie
));
6095 * Given two addresses, find a tunnel instance's IPsec policy heads.
6096 * Returns NULL on failure.
6099 itp_get_byaddr(uint32_t *laddr
, uint32_t *faddr
, int af
, ip_stack_t
*ipst
)
6103 ipsec_tun_pol_t
*itp
= NULL
;
6105 /* Classifiers are used to "src" being foreign. */
6106 if (af
== AF_INET
) {
6107 connp
= ipcl_iptun_classify_v4((ipaddr_t
*)faddr
,
6108 (ipaddr_t
*)laddr
, ipst
);
6110 ASSERT(af
== AF_INET6
);
6111 ASSERT(!IN6_IS_ADDR_V4MAPPED((in6_addr_t
*)laddr
));
6112 ASSERT(!IN6_IS_ADDR_V4MAPPED((in6_addr_t
*)faddr
));
6113 connp
= ipcl_iptun_classify_v6((in6_addr_t
*)faddr
,
6114 (in6_addr_t
*)laddr
, ipst
);
6120 if (IPCL_IS_IPTUN(connp
)) {
6121 iptun
= connp
->conn_iptun
;
6122 if (iptun
!= NULL
) {
6123 itp
= iptun
->iptun_itp
;
6125 /* Braces due to the macro's nature... */
6128 } /* Else itp is already NULL. */
6131 CONN_DEC_REF(connp
);
6136 * Frag cache code, based on SunScreen 3.2 source
6137 * screen/kernel/common/screen_fragcache.c
6140 #define IPSEC_FRAG_TTL_MAX 5
6142 * Note that the following parameters create 256 hash buckets
6143 * with 1024 free entries to be distributed. Things are cleaned
6144 * periodically and are attempted to be cleaned when there is no
6145 * free space, but this system errs on the side of dropping packets
6146 * over creating memory exhaustion. We may decide to make hash
6147 * factor a tunable if this proves to be a bad decision.
6149 #define IPSEC_FRAG_HASH_SLOTS (1<<8)
6150 #define IPSEC_FRAG_HASH_FACTOR 4
6151 #define IPSEC_FRAG_HASH_SIZE (IPSEC_FRAG_HASH_SLOTS * IPSEC_FRAG_HASH_FACTOR)
6153 #define IPSEC_FRAG_HASH_MASK (IPSEC_FRAG_HASH_SLOTS - 1)
6154 #define IPSEC_FRAG_HASH_FUNC(id) (((id) & IPSEC_FRAG_HASH_MASK) ^ \
6156 (ushort_t)IPSEC_FRAG_HASH_SLOTS) & \
6157 IPSEC_FRAG_HASH_MASK))
6159 /* Maximum fragments per packet. 48 bytes payload x 1366 packets > 64KB */
6160 #define IPSEC_MAX_FRAGS 1366
6162 #define V4_FRAG_OFFSET(ipha) ((ntohs(ipha->ipha_fragment_offset_and_flags) & \
6164 #define V4_MORE_FRAGS(ipha) (ntohs(ipha->ipha_fragment_offset_and_flags) & \
6168 * Initialize an ipsec fragcache instance.
6169 * Returns B_FALSE if memory allocation fails.
6172 ipsec_fragcache_init(ipsec_fragcache_t
*frag
)
6174 ipsec_fragcache_entry_t
*ftemp
;
6177 mutex_init(&frag
->itpf_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
6178 frag
->itpf_ptr
= (ipsec_fragcache_entry_t
**)
6179 kmem_zalloc(sizeof (ipsec_fragcache_entry_t
*) *
6180 IPSEC_FRAG_HASH_SLOTS
, KM_NOSLEEP
);
6181 if (frag
->itpf_ptr
== NULL
)
6184 ftemp
= (ipsec_fragcache_entry_t
*)
6185 kmem_zalloc(sizeof (ipsec_fragcache_entry_t
) *
6186 IPSEC_FRAG_HASH_SIZE
, KM_NOSLEEP
);
6187 if (ftemp
== NULL
) {
6188 kmem_free(frag
->itpf_ptr
, sizeof (ipsec_fragcache_entry_t
*) *
6189 IPSEC_FRAG_HASH_SLOTS
);
6193 frag
->itpf_freelist
= NULL
;
6195 for (i
= 0; i
< IPSEC_FRAG_HASH_SIZE
; i
++) {
6196 ftemp
->itpfe_next
= frag
->itpf_freelist
;
6197 frag
->itpf_freelist
= ftemp
;
6201 frag
->itpf_expire_hint
= 0;
6207 ipsec_fragcache_uninit(ipsec_fragcache_t
*frag
, ipsec_stack_t
*ipss
)
6209 ipsec_fragcache_entry_t
*fep
;
6212 mutex_enter(&frag
->itpf_lock
);
6213 if (frag
->itpf_ptr
) {
6214 /* Delete any existing fragcache entry chains */
6215 for (i
= 0; i
< IPSEC_FRAG_HASH_SLOTS
; i
++) {
6216 fep
= (frag
->itpf_ptr
)[i
];
6217 while (fep
!= NULL
) {
6218 /* Returned fep is next in chain or NULL */
6219 fep
= fragcache_delentry(i
, fep
, frag
, ipss
);
6223 * Chase the pointers back to the beginning
6224 * of the memory allocation and then
6225 * get rid of the allocated freelist
6227 while (frag
->itpf_freelist
->itpfe_next
!= NULL
)
6228 frag
->itpf_freelist
= frag
->itpf_freelist
->itpfe_next
;
6230 * XXX - If we ever dynamically grow the freelist
6231 * then we'll have to free entries individually
6232 * or determine how many entries or chunks we have
6233 * grown since the initial allocation.
6235 kmem_free(frag
->itpf_freelist
,
6236 sizeof (ipsec_fragcache_entry_t
) *
6237 IPSEC_FRAG_HASH_SIZE
);
6238 /* Free the fragcache structure */
6239 kmem_free(frag
->itpf_ptr
,
6240 sizeof (ipsec_fragcache_entry_t
*) *
6241 IPSEC_FRAG_HASH_SLOTS
);
6243 mutex_exit(&frag
->itpf_lock
);
6244 mutex_destroy(&frag
->itpf_lock
);
6248 * Add a fragment to the fragment cache. Consumes mp if NULL is returned.
6249 * Returns mp if a whole fragment has been assembled, NULL otherwise
6250 * The returned mp could be a b_next chain of fragments.
6252 * The iramp argument is set on inbound; NULL if outbound.
6255 ipsec_fragcache_add(ipsec_fragcache_t
*frag
, mblk_t
*iramp
, mblk_t
*mp
,
6256 int outer_hdr_len
, ipsec_stack_t
*ipss
)
6264 uint8_t *v6_proto_p
;
6265 uint16_t ip6_hdr_length
;
6267 ip6_frag_t
*fraghdr
;
6268 ipsec_fragcache_entry_t
*fep
;
6270 mblk_t
*nmp
, *prevmp
;
6271 int firstbyte
, lastbyte
;
6274 boolean_t inbound
= (iramp
!= NULL
);
6276 #ifdef FRAGCACHE_DEBUG
6277 cmn_err(CE_WARN
, "Fragcache: %s\n", inbound
? "INBOUND" : "OUTBOUND");
6280 * You're on the slow path, so insure that every packet in the
6281 * cache is a single-mblk one.
6283 if (mp
->b_cont
!= NULL
) {
6284 nmp
= msgpullup(mp
, -1);
6286 ip_drop_packet(mp
, inbound
, NULL
,
6287 DROPPER(ipss
, ipds_spd_nomem
),
6288 &ipss
->ipsec_spd_dropper
);
6290 (void) ip_recv_attr_free_mblk(iramp
);
6297 mutex_enter(&frag
->itpf_lock
);
6299 oiph
= (ipha_t
*)mp
->b_rptr
;
6300 iph
= (ipha_t
*)(mp
->b_rptr
+ outer_hdr_len
);
6302 if (IPH_HDR_VERSION(iph
) == IPV4_VERSION
) {
6305 ASSERT(IPH_HDR_VERSION(iph
) == IPV6_VERSION
);
6306 ip6h
= (ip6_t
*)(mp
->b_rptr
+ outer_hdr_len
);
6308 if (!ip_hdr_length_nexthdr_v6(mp
, ip6h
, &ip6_hdr_length
,
6311 * Find upper layer protocol.
6312 * If it fails we have a malformed packet
6314 mutex_exit(&frag
->itpf_lock
);
6315 ip_drop_packet(mp
, inbound
, NULL
,
6316 DROPPER(ipss
, ipds_spd_malformed_packet
),
6317 &ipss
->ipsec_spd_dropper
);
6319 (void) ip_recv_attr_free_mblk(iramp
);
6322 v6_proto
= *v6_proto_p
;
6326 bzero(&ipp
, sizeof (ipp
));
6327 (void) ip_find_hdr_v6(mp
, ip6h
, B_FALSE
, &ipp
, NULL
);
6328 if (!(ipp
.ipp_fields
& IPPF_FRAGHDR
)) {
6330 * We think this is a fragment, but didn't find
6331 * a fragment header. Something is wrong.
6333 mutex_exit(&frag
->itpf_lock
);
6334 ip_drop_packet(mp
, inbound
, NULL
,
6335 DROPPER(ipss
, ipds_spd_malformed_frag
),
6336 &ipss
->ipsec_spd_dropper
);
6338 (void) ip_recv_attr_free_mblk(iramp
);
6341 fraghdr
= ipp
.ipp_fraghdr
;
6345 /* Anything to cleanup? */
6348 * This cleanup call could be put in a timer loop
6349 * but it may actually be just as reasonable a decision to
6350 * leave it here. The disadvantage is this only gets called when
6351 * frags are added. The advantage is that it is not
6352 * susceptible to race conditions like a time-based cleanup
6355 itpf_time
= gethrestime_sec();
6356 if (itpf_time
>= frag
->itpf_expire_hint
)
6357 ipsec_fragcache_clean(frag
, ipss
);
6359 /* Lookup to see if there is an existing entry */
6362 i
= IPSEC_FRAG_HASH_FUNC(iph
->ipha_ident
);
6364 i
= IPSEC_FRAG_HASH_FUNC(fraghdr
->ip6f_ident
);
6366 for (fep
= (frag
->itpf_ptr
)[i
]; fep
; fep
= fep
->itpfe_next
) {
6368 ASSERT(iph
!= NULL
);
6369 if ((fep
->itpfe_id
== iph
->ipha_ident
) &&
6370 (fep
->itpfe_src
== iph
->ipha_src
) &&
6371 (fep
->itpfe_dst
== iph
->ipha_dst
) &&
6372 (fep
->itpfe_proto
== iph
->ipha_protocol
))
6375 ASSERT(fraghdr
!= NULL
);
6376 ASSERT(fep
!= NULL
);
6377 if ((fep
->itpfe_id
== fraghdr
->ip6f_ident
) &&
6378 IN6_ARE_ADDR_EQUAL(&fep
->itpfe_src6
,
6380 IN6_ARE_ADDR_EQUAL(&fep
->itpfe_dst6
,
6381 &ip6h
->ip6_dst
) && (fep
->itpfe_proto
== v6_proto
))
6387 firstbyte
= V4_FRAG_OFFSET(iph
);
6388 lastbyte
= firstbyte
+ ntohs(iph
->ipha_length
) -
6389 IPH_HDR_LENGTH(iph
);
6390 last
= (V4_MORE_FRAGS(iph
) == 0);
6391 #ifdef FRAGCACHE_DEBUG
6392 cmn_err(CE_WARN
, "V4 fragcache: firstbyte = %d, lastbyte = %d, "
6393 "is_last_frag = %d, id = %d, mp = %p\n", firstbyte
,
6394 lastbyte
, last
, iph
->ipha_ident
, mp
);
6397 firstbyte
= ntohs(fraghdr
->ip6f_offlg
& IP6F_OFF_MASK
);
6398 lastbyte
= firstbyte
+ ntohs(ip6h
->ip6_plen
) +
6399 sizeof (ip6_t
) - ip6_hdr_length
;
6400 last
= (fraghdr
->ip6f_offlg
& IP6F_MORE_FRAG
) == 0;
6401 #ifdef FRAGCACHE_DEBUG
6402 cmn_err(CE_WARN
, "V6 fragcache: firstbyte = %d, lastbyte = %d, "
6403 "is_last_frag = %d, id = %d, fraghdr = %p, mp = %p\n",
6404 firstbyte
, lastbyte
, last
, fraghdr
->ip6f_ident
, fraghdr
,
6409 /* check for bogus fragments and delete the entry */
6410 if (firstbyte
> 0 && firstbyte
<= 8) {
6412 (void) fragcache_delentry(i
, fep
, frag
, ipss
);
6413 mutex_exit(&frag
->itpf_lock
);
6414 ip_drop_packet(mp
, inbound
, NULL
,
6415 DROPPER(ipss
, ipds_spd_malformed_frag
),
6416 &ipss
->ipsec_spd_dropper
);
6418 (void) ip_recv_attr_free_mblk(iramp
);
6422 /* Not found, allocate a new entry */
6424 if (frag
->itpf_freelist
== NULL
) {
6425 /* see if there is some space */
6426 ipsec_fragcache_clean(frag
, ipss
);
6427 if (frag
->itpf_freelist
== NULL
) {
6428 mutex_exit(&frag
->itpf_lock
);
6429 ip_drop_packet(mp
, inbound
, NULL
,
6430 DROPPER(ipss
, ipds_spd_nomem
),
6431 &ipss
->ipsec_spd_dropper
);
6433 (void) ip_recv_attr_free_mblk(iramp
);
6438 fep
= frag
->itpf_freelist
;
6439 frag
->itpf_freelist
= fep
->itpfe_next
;
6442 bcopy((caddr_t
)&iph
->ipha_src
, (caddr_t
)&fep
->itpfe_src
,
6443 sizeof (struct in_addr
));
6444 bcopy((caddr_t
)&iph
->ipha_dst
, (caddr_t
)&fep
->itpfe_dst
,
6445 sizeof (struct in_addr
));
6446 fep
->itpfe_id
= iph
->ipha_ident
;
6447 fep
->itpfe_proto
= iph
->ipha_protocol
;
6448 i
= IPSEC_FRAG_HASH_FUNC(fep
->itpfe_id
);
6450 bcopy((in6_addr_t
*)&ip6h
->ip6_src
,
6451 (in6_addr_t
*)&fep
->itpfe_src6
,
6452 sizeof (struct in6_addr
));
6453 bcopy((in6_addr_t
*)&ip6h
->ip6_dst
,
6454 (in6_addr_t
*)&fep
->itpfe_dst6
,
6455 sizeof (struct in6_addr
));
6456 fep
->itpfe_id
= fraghdr
->ip6f_ident
;
6457 fep
->itpfe_proto
= v6_proto
;
6458 i
= IPSEC_FRAG_HASH_FUNC(fep
->itpfe_id
);
6460 itpf_time
= gethrestime_sec();
6461 fep
->itpfe_exp
= itpf_time
+ IPSEC_FRAG_TTL_MAX
+ 1;
6462 fep
->itpfe_last
= 0;
6463 fep
->itpfe_fraglist
= NULL
;
6464 fep
->itpfe_depth
= 0;
6465 fep
->itpfe_next
= (frag
->itpf_ptr
)[i
];
6466 (frag
->itpf_ptr
)[i
] = fep
;
6468 if (frag
->itpf_expire_hint
> fep
->itpfe_exp
)
6469 frag
->itpf_expire_hint
= fep
->itpfe_exp
;
6473 /* Insert it in the frag list */
6474 /* List is in order by starting offset of fragments */
6477 for (nmp
= fep
->itpfe_fraglist
; nmp
; nmp
= nmp
->b_next
) {
6482 ip6_frag_t
*nfraghdr
;
6483 uint16_t nip6_hdr_length
;
6484 uint8_t *nv6_proto_p
;
6485 int nfirstbyte
, nlastbyte
;
6487 mblk_t
*ndata_mp
= (inbound
? nmp
->b_cont
: nmp
);
6490 oniph
= (ipha_t
*)mp
->b_rptr
;
6495 * Determine outer header type and length and set
6496 * pointers appropriately
6499 if (IPH_HDR_VERSION(oniph
) == IPV4_VERSION
) {
6500 hdr_len
= ((outer_hdr_len
!= 0) ?
6501 IPH_HDR_LENGTH(oiph
) : 0);
6502 niph
= (ipha_t
*)(ndata_mp
->b_rptr
+ hdr_len
);
6504 ASSERT(IPH_HDR_VERSION(oniph
) == IPV6_VERSION
);
6505 ASSERT(ndata_mp
->b_cont
== NULL
);
6506 nip6h
= (ip6_t
*)ndata_mp
->b_rptr
;
6507 (void) ip_hdr_length_nexthdr_v6(ndata_mp
, nip6h
,
6508 &nip6_hdr_length
, &v6_proto_p
);
6509 hdr_len
= ((outer_hdr_len
!= 0) ? nip6_hdr_length
: 0);
6513 * Determine inner header type and length and set
6514 * pointers appropriately
6520 niph
= (ipha_t
*)(ndata_mp
->b_rptr
+ hdr_len
);
6522 nfirstbyte
= V4_FRAG_OFFSET(niph
);
6523 nlastbyte
= nfirstbyte
+ ntohs(niph
->ipha_length
) -
6524 IPH_HDR_LENGTH(niph
);
6526 ASSERT(ndata_mp
->b_cont
== NULL
);
6527 nip6h
= (ip6_t
*)(ndata_mp
->b_rptr
+ hdr_len
);
6528 if (!ip_hdr_length_nexthdr_v6(ndata_mp
, nip6h
,
6529 &nip6_hdr_length
, &nv6_proto_p
)) {
6530 mutex_exit(&frag
->itpf_lock
);
6531 ip_drop_packet_chain(nmp
, inbound
, NULL
,
6532 DROPPER(ipss
, ipds_spd_malformed_frag
),
6533 &ipss
->ipsec_spd_dropper
);
6534 ipsec_freemsg_chain(ndata_mp
);
6536 (void) ip_recv_attr_free_mblk(iramp
);
6539 bzero(&nipp
, sizeof (nipp
));
6540 (void) ip_find_hdr_v6(ndata_mp
, nip6h
, B_FALSE
, &nipp
,
6542 nfraghdr
= nipp
.ipp_fraghdr
;
6543 nfirstbyte
= ntohs(nfraghdr
->ip6f_offlg
&
6545 nlastbyte
= nfirstbyte
+ ntohs(nip6h
->ip6_plen
) +
6546 sizeof (ip6_t
) - nip6_hdr_length
;
6549 /* Check for overlapping fragments */
6550 if (firstbyte
>= nfirstbyte
&& firstbyte
< nlastbyte
) {
6553 * ~~~~--------- # Check if the newly
6554 * ~ ndata_mp| # received fragment
6555 * ~~~~--------- # overlaps with the
6556 * ---------~~~~~~ # current fragment.
6561 data
= (char *)iph
+ IPH_HDR_LENGTH(iph
) +
6562 firstbyte
- nfirstbyte
;
6563 ndata
= (char *)niph
+ IPH_HDR_LENGTH(niph
);
6565 data
= (char *)ip6h
+
6566 nip6_hdr_length
+ firstbyte
-
6568 ndata
= (char *)nip6h
+ nip6_hdr_length
;
6570 if (bcmp(data
, ndata
, MIN(lastbyte
, nlastbyte
) -
6572 /* Overlapping data does not match */
6573 (void) fragcache_delentry(i
, fep
, frag
, ipss
);
6574 mutex_exit(&frag
->itpf_lock
);
6575 ip_drop_packet(mp
, inbound
, NULL
,
6576 DROPPER(ipss
, ipds_spd_overlap_frag
),
6577 &ipss
->ipsec_spd_dropper
);
6579 (void) ip_recv_attr_free_mblk(iramp
);
6582 /* Part of defense for jolt2.c fragmentation attack */
6583 if (firstbyte
>= nfirstbyte
&& lastbyte
<= nlastbyte
) {
6585 * Check for identical or subset fragments:
6586 * ---------- ~~~~--------~~~~~
6587 * | nmp | or ~ nmp ~
6588 * ---------- ~~~~--------~~~~~
6593 mutex_exit(&frag
->itpf_lock
);
6594 ip_drop_packet(mp
, inbound
, NULL
,
6595 DROPPER(ipss
, ipds_spd_evil_frag
),
6596 &ipss
->ipsec_spd_dropper
);
6598 (void) ip_recv_attr_free_mblk(iramp
);
6604 /* Correct location for this fragment? */
6605 if (firstbyte
<= nfirstbyte
) {
6607 * Check if the tail end of the new fragment overlaps
6608 * with the head of the current fragment.
6616 if (lastbyte
> nfirstbyte
) {
6617 /* Fragments overlap */
6618 data
= (char *)iph
+ IPH_HDR_LENGTH(iph
) +
6619 firstbyte
- nfirstbyte
;
6620 ndata
= (char *)niph
+ IPH_HDR_LENGTH(niph
);
6622 data
= (char *)iph
+
6623 IPH_HDR_LENGTH(iph
) + firstbyte
-
6625 ndata
= (char *)niph
+
6626 IPH_HDR_LENGTH(niph
);
6628 data
= (char *)ip6h
+
6629 nip6_hdr_length
+ firstbyte
-
6631 ndata
= (char *)nip6h
+ nip6_hdr_length
;
6633 if (bcmp(data
, ndata
, MIN(lastbyte
, nlastbyte
)
6635 /* Overlap mismatch */
6636 (void) fragcache_delentry(i
, fep
, frag
,
6638 mutex_exit(&frag
->itpf_lock
);
6639 ip_drop_packet(mp
, inbound
, NULL
,
6641 ipds_spd_overlap_frag
),
6642 &ipss
->ipsec_spd_dropper
);
6644 (void) ip_recv_attr_free_mblk(
6652 * Fragment does not illegally overlap and can now
6653 * be inserted into the chain
6660 /* Prepend the attributes before we link it in */
6661 if (iramp
!= NULL
) {
6662 ASSERT(iramp
->b_cont
== NULL
);
6669 if (prevmp
== NULL
) {
6670 fep
->itpfe_fraglist
= mp
;
6672 prevmp
->b_next
= mp
;
6675 fep
->itpfe_last
= 1;
6677 /* Part of defense for jolt2.c fragmentation attack */
6678 if (++(fep
->itpfe_depth
) > IPSEC_MAX_FRAGS
) {
6679 (void) fragcache_delentry(i
, fep
, frag
, ipss
);
6680 mutex_exit(&frag
->itpf_lock
);
6682 mp
= ip_recv_attr_free_mblk(mp
);
6684 ip_drop_packet(mp
, inbound
, NULL
,
6685 DROPPER(ipss
, ipds_spd_max_frags
),
6686 &ipss
->ipsec_spd_dropper
);
6690 /* Check for complete packet */
6692 if (!fep
->itpfe_last
) {
6693 mutex_exit(&frag
->itpf_lock
);
6694 #ifdef FRAGCACHE_DEBUG
6695 cmn_err(CE_WARN
, "Fragment cached, last not yet seen.\n");
6701 for (mp
= fep
->itpfe_fraglist
; mp
; mp
= mp
->b_next
) {
6702 mblk_t
*data_mp
= (inbound
? mp
->b_cont
: mp
);
6705 oiph
= (ipha_t
*)data_mp
->b_rptr
;
6709 if (IPH_HDR_VERSION(oiph
) == IPV4_VERSION
) {
6710 hdr_len
= ((outer_hdr_len
!= 0) ?
6711 IPH_HDR_LENGTH(oiph
) : 0);
6712 iph
= (ipha_t
*)(data_mp
->b_rptr
+ hdr_len
);
6714 ASSERT(IPH_HDR_VERSION(oiph
) == IPV6_VERSION
);
6715 ASSERT(data_mp
->b_cont
== NULL
);
6716 ip6h
= (ip6_t
*)data_mp
->b_rptr
;
6717 (void) ip_hdr_length_nexthdr_v6(data_mp
, ip6h
,
6718 &ip6_hdr_length
, &v6_proto_p
);
6719 hdr_len
= ((outer_hdr_len
!= 0) ? ip6_hdr_length
: 0);
6722 /* Calculate current fragment start/end */
6726 iph
= (ipha_t
*)(data_mp
->b_rptr
+ hdr_len
);
6728 firstbyte
= V4_FRAG_OFFSET(iph
);
6729 lastbyte
= firstbyte
+ ntohs(iph
->ipha_length
) -
6730 IPH_HDR_LENGTH(iph
);
6732 ASSERT(data_mp
->b_cont
== NULL
);
6733 ip6h
= (ip6_t
*)(data_mp
->b_rptr
+ hdr_len
);
6734 if (!ip_hdr_length_nexthdr_v6(data_mp
, ip6h
,
6735 &ip6_hdr_length
, &v6_proto_p
)) {
6736 mutex_exit(&frag
->itpf_lock
);
6737 ip_drop_packet_chain(mp
, inbound
, NULL
,
6738 DROPPER(ipss
, ipds_spd_malformed_frag
),
6739 &ipss
->ipsec_spd_dropper
);
6742 v6_proto
= *v6_proto_p
;
6743 bzero(&ipp
, sizeof (ipp
));
6744 (void) ip_find_hdr_v6(data_mp
, ip6h
, B_FALSE
, &ipp
,
6746 fraghdr
= ipp
.ipp_fraghdr
;
6747 firstbyte
= ntohs(fraghdr
->ip6f_offlg
&
6749 lastbyte
= firstbyte
+ ntohs(ip6h
->ip6_plen
) +
6750 sizeof (ip6_t
) - ip6_hdr_length
;
6754 * If this fragment is greater than current offset,
6755 * we have a missing fragment so return NULL
6757 if (firstbyte
> offset
) {
6758 mutex_exit(&frag
->itpf_lock
);
6759 #ifdef FRAGCACHE_DEBUG
6761 * Note, this can happen when the last frag
6762 * gets sent through because it is smaller
6763 * than the MTU. It is not necessarily an
6766 cmn_err(CE_WARN
, "Frag greater than offset! : "
6767 "missing fragment: firstbyte = %d, offset = %d, "
6768 "mp = %p\n", firstbyte
, offset
, mp
);
6772 #ifdef FRAGCACHE_DEBUG
6773 cmn_err(CE_WARN
, "Frag offsets : "
6774 "firstbyte = %d, offset = %d, mp = %p\n",
6775 firstbyte
, offset
, mp
);
6779 * If we are at the last fragment, we have the complete
6780 * packet, so rechain things and return it to caller
6784 if ((is_v4
&& !V4_MORE_FRAGS(iph
)) ||
6785 (!is_v4
&& !(fraghdr
->ip6f_offlg
& IP6F_MORE_FRAG
))) {
6786 mp
= fep
->itpfe_fraglist
;
6787 fep
->itpfe_fraglist
= NULL
;
6788 (void) fragcache_delentry(i
, fep
, frag
, ipss
);
6789 mutex_exit(&frag
->itpf_lock
);
6791 if ((is_v4
&& (firstbyte
+ ntohs(iph
->ipha_length
) >
6792 65535)) || (!is_v4
&& (firstbyte
+
6793 ntohs(ip6h
->ip6_plen
) > 65535))) {
6794 /* It is an invalid "ping-o-death" packet */
6796 ip_drop_packet_chain(mp
, inbound
, NULL
,
6797 DROPPER(ipss
, ipds_spd_evil_frag
),
6798 &ipss
->ipsec_spd_dropper
);
6801 #ifdef FRAGCACHE_DEBUG
6802 cmn_err(CE_WARN
, "Fragcache returning mp = %p, "
6803 "mp->b_next = %p", mp
, mp
->b_next
);
6806 * For inbound case, mp has attrmp b_next'd chain
6807 * For outbound case, it is just data mp chain
6813 * Update new ending offset if this
6814 * fragment extends the packet
6816 if (offset
< lastbyte
)
6820 mutex_exit(&frag
->itpf_lock
);
6822 /* Didn't find last fragment, so return NULL */
6827 ipsec_fragcache_clean(ipsec_fragcache_t
*frag
, ipsec_stack_t
*ipss
)
6829 ipsec_fragcache_entry_t
*fep
;
6831 ipsec_fragcache_entry_t
*earlyfep
= NULL
;
6836 ASSERT(MUTEX_HELD(&frag
->itpf_lock
));
6838 itpf_time
= gethrestime_sec();
6839 earlyexp
= itpf_time
+ 10000;
6841 for (i
= 0; i
< IPSEC_FRAG_HASH_SLOTS
; i
++) {
6842 fep
= (frag
->itpf_ptr
)[i
];
6844 if (fep
->itpfe_exp
< itpf_time
) {
6846 fep
= fragcache_delentry(i
, fep
, frag
, ipss
);
6848 if (fep
->itpfe_exp
< earlyexp
) {
6850 earlyexp
= fep
->itpfe_exp
;
6853 fep
= fep
->itpfe_next
;
6858 frag
->itpf_expire_hint
= earlyexp
;
6861 if (frag
->itpf_freelist
== NULL
)
6862 (void) fragcache_delentry(earlyi
, earlyfep
, frag
, ipss
);
6865 static ipsec_fragcache_entry_t
*
6866 fragcache_delentry(int slot
, ipsec_fragcache_entry_t
*fep
,
6867 ipsec_fragcache_t
*frag
, ipsec_stack_t
*ipss
)
6869 ipsec_fragcache_entry_t
*targp
;
6870 ipsec_fragcache_entry_t
*nextp
= fep
->itpfe_next
;
6872 ASSERT(MUTEX_HELD(&frag
->itpf_lock
));
6874 /* Free up any fragment list still in cache entry */
6875 if (fep
->itpfe_fraglist
!= NULL
) {
6876 ip_drop_packet_chain(fep
->itpfe_fraglist
,
6877 ip_recv_attr_is_mblk(fep
->itpfe_fraglist
), NULL
,
6878 DROPPER(ipss
, ipds_spd_expired_frags
),
6879 &ipss
->ipsec_spd_dropper
);
6881 fep
->itpfe_fraglist
= NULL
;
6883 targp
= (frag
->itpf_ptr
)[slot
];
6887 /* unlink from head of hash chain */
6888 (frag
->itpf_ptr
)[slot
] = nextp
;
6889 /* link into free list */
6890 fep
->itpfe_next
= frag
->itpf_freelist
;
6891 frag
->itpf_freelist
= fep
;
6895 /* maybe should use double linked list to make update faster */
6896 /* must be past front of chain */
6898 if (targp
->itpfe_next
== fep
) {
6899 /* unlink from hash chain */
6900 targp
->itpfe_next
= nextp
;
6901 /* link into free list */
6902 fep
->itpfe_next
= frag
->itpf_freelist
;
6903 frag
->itpf_freelist
= fep
;
6906 targp
= targp
->itpfe_next
;