xref: /illumos-gate/usr/src/uts/common/inet/ipsec_impl.h (revision dec98d2aa2912b1fbd0abd2574307a695b566692)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2009 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  * Copyright (c) 2012 Nexenta Systems, Inc. All rights reserved.
25  */
26 
27 #ifndef _INET_IPSEC_IMPL_H
28 #define	_INET_IPSEC_IMPL_H
29 
30 #include <inet/ip.h>
31 #include <inet/ipdrop.h>
32 
33 #ifdef	__cplusplus
34 extern "C" {
35 #endif
36 
37 #define	IPSEC_CONF_SRC_ADDRESS	0	/* Source Address */
38 #define	IPSEC_CONF_SRC_PORT		1	/* Source Port */
39 #define	IPSEC_CONF_DST_ADDRESS	2	/* Dest Address */
40 #define	IPSEC_CONF_DST_PORT		3	/* Dest Port */
41 #define	IPSEC_CONF_SRC_MASK		4	/* Source Address Mask */
42 #define	IPSEC_CONF_DST_MASK		5	/* Destination Address Mask */
43 #define	IPSEC_CONF_ULP			6	/* Upper layer Port */
44 #define	IPSEC_CONF_IPSEC_PROT	7	/* AH or ESP or AH_ESP */
45 #define	IPSEC_CONF_IPSEC_AALGS	8	/* Auth Algorithms - MD5 etc. */
46 #define	IPSEC_CONF_IPSEC_EALGS	9	/* Encr Algorithms - DES etc. */
47 #define	IPSEC_CONF_IPSEC_EAALGS	10	/* Encr Algorithms - MD5 etc. */
48 #define	IPSEC_CONF_IPSEC_SA		11	/* Shared or unique SA */
49 #define	IPSEC_CONF_IPSEC_DIR 		12	/* Direction of traffic */
50 #define	IPSEC_CONF_ICMP_TYPE 		13	/* ICMP type */
51 #define	IPSEC_CONF_ICMP_CODE 		14	/* ICMP code */
52 #define	IPSEC_CONF_NEGOTIATE		15	/* Negotiation */
53 #define	IPSEC_CONF_TUNNEL		16	/* Tunnel */
54 
55 /* Type of an entry */
56 
57 #define	IPSEC_NTYPES			0x02
58 #define	IPSEC_TYPE_OUTBOUND		0x00
59 #define	IPSEC_TYPE_INBOUND		0x01
60 
61 /* Policy */
62 #define	IPSEC_POLICY_APPLY	0x01
63 #define	IPSEC_POLICY_DISCARD	0x02
64 #define	IPSEC_POLICY_BYPASS	0x03
65 
66 /* Shared or unique SA */
67 #define	IPSEC_SHARED_SA		0x01
68 #define	IPSEC_UNIQUE_SA		0x02
69 
70 /* IPsec protocols and combinations */
71 #define	IPSEC_AH_ONLY		0x01
72 #define	IPSEC_ESP_ONLY		0x02
73 #define	IPSEC_AH_ESP		0x03
74 
75 /*
76  * Internally defined "any" algorithm.
77  * Move to PF_KEY v3 when that RFC is released.
78  */
79 #define	SADB_AALG_ANY 255
80 
81 #ifdef _KERNEL
82 
83 #include <inet/common.h>
84 #include <netinet/ip6.h>
85 #include <netinet/icmp6.h>
86 #include <net/pfkeyv2.h>
87 #include <inet/ip.h>
88 #include <inet/sadb.h>
89 #include <inet/ipsecah.h>
90 #include <inet/ipsecesp.h>
91 #include <sys/crypto/common.h>
92 #include <sys/crypto/api.h>
93 #include <sys/avl.h>
94 
95 /*
96  * Maximum number of authentication algorithms (can be indexed by one byte
97  * per PF_KEY and the IKE IPsec DOI.
98  */
99 #define	MAX_AALGS 256
100 
101 /*
102  * IPsec task queue constants.
103  */
104 #define	IPSEC_TASKQ_MIN 10
105 #define	IPSEC_TASKQ_MAX 20
106 
107 /*
108  * So we can access IPsec global variables that live in keysock.c.
109  */
110 extern boolean_t keysock_extended_reg(netstack_t *);
111 extern uint32_t keysock_next_seq(netstack_t *);
112 
113 /*
114  * Locking for ipsec policy rules:
115  *
116  * policy heads: system policy is static; per-conn polheads are dynamic,
117  * and refcounted (and inherited); use atomic refcounts and "don't let
118  * go with both hands".
119  *
120  * policy: refcounted; references from polhead, ipsec_out
121  *
122  * actions: refcounted; referenced from: action hash table, policy, ipsec_out
123  * selectors: refcounted; referenced from: selector hash table, policy.
124  */
125 
126 /*
127  * the following are inspired by, but not directly based on,
128  * some of the sys/queue.h type-safe pseudo-polymorphic macros
129  * found in BSD.
130  *
131  * XXX If we use these more generally, we'll have to make the names
132  * less generic (HASH_* will probably clobber other namespaces).
133  */
134 
135 #define	HASH_LOCK(table, hash) \
136 	mutex_enter(&(table)[hash].hash_lock)
137 #define	HASH_UNLOCK(table, hash) \
138 	mutex_exit(&(table)[hash].hash_lock)
139 
140 #define	HASH_LOCKED(table, hash) \
141 	MUTEX_HELD(&(table)[hash].hash_lock)
142 
143 #define	HASH_ITERATE(var, field, table, hash) 		\
144 	var = table[hash].hash_head; var != NULL; var = var->field.hash_next
145 
146 #define	HASH_NEXT(var, field) 		\
147 	(var)->field.hash_next
148 
149 #define	HASH_INSERT(var, field, table, hash)			\
150 {								\
151 	ASSERT(HASH_LOCKED(table, hash));			\
152 	(var)->field.hash_next = (table)[hash].hash_head;	\
153 	(var)->field.hash_pp = &(table)[hash].hash_head;	\
154 	(table)[hash].hash_head = var;				\
155 	if ((var)->field.hash_next != NULL)			\
156 		(var)->field.hash_next->field.hash_pp = 	\
157 			&((var)->field.hash_next); 		\
158 }
159 
160 
161 #define	HASH_UNCHAIN(var, field, table, hash)			\
162 {								\
163 	ASSERT(MUTEX_HELD(&(table)[hash].hash_lock));		\
164 	HASHLIST_UNCHAIN(var, field);				\
165 }
166 
167 #define	HASHLIST_INSERT(var, field, head)			\
168 {								\
169 	(var)->field.hash_next = head;				\
170 	(var)->field.hash_pp = &(head);				\
171 	head = var;						\
172 	if ((var)->field.hash_next != NULL)			\
173 		(var)->field.hash_next->field.hash_pp = 	\
174 			&((var)->field.hash_next); 		\
175 }
176 
177 #define	HASHLIST_UNCHAIN(var, field) 				\
178 {								\
179 	*var->field.hash_pp = var->field.hash_next;		\
180 	if (var->field.hash_next)				\
181 		var->field.hash_next->field.hash_pp = 		\
182 			var->field.hash_pp;			\
183 	HASH_NULL(var, field);					\
184 }
185 
186 
187 #define	HASH_NULL(var, field) 					\
188 {								\
189 	var->field.hash_next = NULL;				\
190 	var->field.hash_pp = NULL;				\
191 }
192 
193 #define	HASH_LINK(fieldname, type)				\
194 	struct {						\
195 		type *hash_next;				\
196 		type **hash_pp;					\
197 	} fieldname
198 
199 
200 #define	HASH_HEAD(tag)						\
201 	struct {						\
202 		struct tag *hash_head;				\
203 		kmutex_t hash_lock;				\
204 	}
205 
206 
207 typedef struct ipsec_policy_s ipsec_policy_t;
208 
209 typedef HASH_HEAD(ipsec_policy_s) ipsec_policy_hash_t;
210 
211 /*
212  * When adding new fields to ipsec_prot_t, make sure to update
213  * ipsec_in_to_out_action() as well as other code in spd.c
214  */
215 
216 typedef struct ipsec_prot
217 {
218 	unsigned int
219 		ipp_use_ah : 1,
220 		ipp_use_esp : 1,
221 		ipp_use_se : 1,
222 		ipp_use_unique : 1,
223 		ipp_use_espa : 1,
224 		ipp_pad : 27;
225 	uint8_t		ipp_auth_alg;		 /* DOI number */
226 	uint8_t		ipp_encr_alg;		 /* DOI number */
227 	uint8_t		ipp_esp_auth_alg;	 /* DOI number */
228 	uint16_t 	ipp_ah_minbits;		 /* AH: min keylen */
229 	uint16_t 	ipp_ah_maxbits;		 /* AH: max keylen */
230 	uint16_t	ipp_espe_minbits;	 /* ESP encr: min keylen */
231 	uint16_t	ipp_espe_maxbits;	 /* ESP encr: max keylen */
232 	uint16_t	ipp_espa_minbits;	 /* ESP auth: min keylen */
233 	uint16_t	ipp_espa_maxbits;	 /* ESP auth: max keylen */
234 	uint32_t	ipp_km_proto;		 /* key mgmt protocol */
235 	uint32_t	ipp_km_cookie;		 /* key mgmt cookie */
236 	uint32_t	ipp_replay_depth;	 /* replay window */
237 	/* XXX add lifetimes */
238 } ipsec_prot_t;
239 
240 #define	IPSEC_MAX_KEYBITS (0xffff)
241 
242 /*
243  * An individual policy action, possibly a member of a chain.
244  *
245  * Action chains may be shared between multiple policy rules.
246  *
247  * With one exception (IPSEC_POLICY_LOG), a chain consists of an
248  * ordered list of alternative ways to handle a packet.
249  *
250  * All actions are also "interned" into a hash table (to allow
251  * multiple rules with the same action chain to share one copy in
252  * memory).
253  */
254 
255 typedef struct ipsec_act
256 {
257 	uint8_t		ipa_type;
258 	uint8_t		ipa_log;
259 	union
260 	{
261 		ipsec_prot_t	ipau_apply;
262 		uint8_t		ipau_reject_type;
263 		uint32_t	ipau_resolve_id; /* magic cookie */
264 		uint8_t		ipau_log_type;
265 	} ipa_u;
266 #define	ipa_apply ipa_u.ipau_apply
267 #define	ipa_reject_type ipa_u.ipau_reject_type
268 #define	ipa_log_type ipa_u.ipau_log_type
269 #define	ipa_resolve_type ipa_u.ipau_resolve_type
270 } ipsec_act_t;
271 
272 #define	IPSEC_ACT_APPLY		0x01 /* match IPSEC_POLICY_APPLY */
273 #define	IPSEC_ACT_DISCARD	0x02 /* match IPSEC_POLICY_DISCARD */
274 #define	IPSEC_ACT_BYPASS	0x03 /* match IPSEC_POLICY_BYPASS */
275 #define	IPSEC_ACT_REJECT	0x04
276 #define	IPSEC_ACT_CLEAR		0x05
277 
278 typedef struct ipsec_action_s
279 {
280 	HASH_LINK(ipa_hash, struct ipsec_action_s);
281 	struct ipsec_action_s	*ipa_next;	/* next alternative */
282 	uint32_t		ipa_refs;		/* refcount */
283 	ipsec_act_t		ipa_act;
284 	/*
285 	 * The following bits are equivalent to an OR of bits included in the
286 	 * ipau_apply fields of this and subsequent actions in an
287 	 * action chain; this is an optimization for the sake of
288 	 * ipsec_out_process() in ip.c and a few other places.
289 	 */
290 	unsigned int
291 		ipa_hval: 8,
292 		ipa_allow_clear:1,		/* rule allows cleartext? */
293 		ipa_want_ah:1,			/* an action wants ah */
294 		ipa_want_esp:1,			/* an action wants esp */
295 		ipa_want_se:1,			/* an action wants se */
296 		ipa_want_unique:1,		/* want unique sa's */
297 		ipa_pad:19;
298 	uint32_t		ipa_ovhd;	/* per-packet encap ovhd */
299 } ipsec_action_t;
300 
301 #define	IPACT_REFHOLD(ipa) {			\
302 	atomic_inc_32(&(ipa)->ipa_refs);	\
303 	ASSERT((ipa)->ipa_refs != 0);	\
304 }
305 #define	IPACT_REFRELE(ipa) {					\
306 	ASSERT((ipa)->ipa_refs != 0);				\
307 	membar_exit();						\
308 	if (atomic_dec_32_nv(&(ipa)->ipa_refs) == 0)	\
309 		ipsec_action_free(ipa);				\
310 	(ipa) = 0;						\
311 }
312 
313 /*
314  * For now, use a trivially sized hash table for actions.
315  * In the future we can add the structure canonicalization necessary
316  * to get the hash function to behave correctly..
317  */
318 #define	IPSEC_ACTION_HASH_SIZE 1
319 
320 /*
321  * Merged address structure, for cheezy address-family independent
322  * matches in policy code.
323  */
324 
325 typedef union ipsec_addr
326 {
327 	in6_addr_t	ipsad_v6;
328 	in_addr_t	ipsad_v4;
329 } ipsec_addr_t;
330 
331 /*
332  * ipsec selector set, as used by the kernel policy structures.
333  * Note that that we specify "local" and "remote"
334  * rather than "source" and "destination", which allows the selectors
335  * for symmetric policy rules to be shared between inbound and
336  * outbound rules.
337  *
338  * "local" means "destination" on inbound, and "source" on outbound.
339  * "remote" means "source" on inbound, and "destination" on outbound.
340  * XXX if we add a fifth policy enforcement point for forwarded packets,
341  * what do we do?
342  *
343  * The ipsl_valid mask is not done as a bitfield; this is so we
344  * can use "ffs()" to find the "most interesting" valid tag.
345  *
346  * XXX should we have multiple types for space-conservation reasons?
347  * (v4 vs v6?  prefix vs. range)?
348  */
349 
350 typedef struct ipsec_selkey
351 {
352 	uint32_t	ipsl_valid;		/* bitmask of valid entries */
353 #define	IPSL_REMOTE_ADDR		0x00000001
354 #define	IPSL_LOCAL_ADDR			0x00000002
355 #define	IPSL_REMOTE_PORT		0x00000004
356 #define	IPSL_LOCAL_PORT			0x00000008
357 #define	IPSL_PROTOCOL			0x00000010
358 #define	IPSL_ICMP_TYPE			0x00000020
359 #define	IPSL_ICMP_CODE			0x00000040
360 #define	IPSL_IPV6			0x00000080
361 #define	IPSL_IPV4			0x00000100
362 
363 #define	IPSL_WILDCARD			0x0000007f
364 
365 	ipsec_addr_t	ipsl_local;
366 	ipsec_addr_t	ipsl_remote;
367 	uint16_t	ipsl_lport;
368 	uint16_t	ipsl_rport;
369 	/*
370 	 * ICMP type and code selectors. Both have an end value to
371 	 * specify ranges, or * and *_end are equal for a single
372 	 * value
373 	 */
374 	uint8_t		ipsl_icmp_type;
375 	uint8_t		ipsl_icmp_type_end;
376 	uint8_t		ipsl_icmp_code;
377 	uint8_t		ipsl_icmp_code_end;
378 
379 	uint8_t		ipsl_proto;		/* ip payload type */
380 	uint8_t		ipsl_local_pfxlen;	/* #bits of prefix */
381 	uint8_t		ipsl_remote_pfxlen;	/* #bits of prefix */
382 	uint8_t		ipsl_mbz;
383 
384 	/* Insert new elements above this line */
385 	uint32_t	ipsl_pol_hval;
386 	uint32_t	ipsl_sel_hval;
387 } ipsec_selkey_t;
388 
389 typedef struct ipsec_sel
390 {
391 	HASH_LINK(ipsl_hash, struct ipsec_sel);
392 	uint32_t	ipsl_refs;		/* # refs to this sel */
393 	ipsec_selkey_t	ipsl_key;		/* actual selector guts */
394 } ipsec_sel_t;
395 
396 /*
397  * One policy rule.  This will be linked into a single hash chain bucket in
398  * the parent rule structure.  If the selector is simple enough to
399  * allow hashing, it gets filed under ipsec_policy_root_t->ipr_hash.
400  * Otherwise it goes onto a linked list in ipsec_policy_root_t->ipr_nonhash[af]
401  *
402  * In addition, we file the rule into an avl tree keyed by the rule index.
403  * (Duplicate rules are permitted; the comparison function breaks ties).
404  */
405 struct ipsec_policy_s
406 {
407 	HASH_LINK(ipsp_hash, struct ipsec_policy_s);
408 	avl_node_t		ipsp_byid;
409 	uint64_t		ipsp_index;	/* unique id */
410 	uint32_t		ipsp_prio; 	/* rule priority */
411 	uint32_t		ipsp_refs;
412 	ipsec_sel_t		*ipsp_sel;	/* selector set (shared) */
413 	ipsec_action_t		*ipsp_act; 	/* action (may be shared) */
414 	netstack_t		*ipsp_netstack;	/* No netstack_hold */
415 };
416 
417 #define	IPPOL_REFHOLD(ipp) {			\
418 	atomic_inc_32(&(ipp)->ipsp_refs);	\
419 	ASSERT((ipp)->ipsp_refs != 0);		\
420 }
421 #define	IPPOL_REFRELE(ipp) {					\
422 	ASSERT((ipp)->ipsp_refs != 0);				\
423 	membar_exit();						\
424 	if (atomic_dec_32_nv(&(ipp)->ipsp_refs) == 0)	\
425 		ipsec_policy_free(ipp);				\
426 	(ipp) = 0;						\
427 }
428 
429 #define	IPPOL_UNCHAIN(php, ip)					\
430 	HASHLIST_UNCHAIN((ip), ipsp_hash);			\
431 	avl_remove(&(php)->iph_rulebyid, (ip));			\
432 	IPPOL_REFRELE(ip);
433 
434 /*
435  * Policy ruleset.  One per (protocol * direction) for system policy.
436  */
437 
438 #define	IPSEC_AF_V4	0
439 #define	IPSEC_AF_V6	1
440 #define	IPSEC_NAF	2
441 
442 typedef struct ipsec_policy_root_s
443 {
444 	ipsec_policy_t		*ipr_nonhash[IPSEC_NAF];
445 	int			ipr_nchains;
446 	ipsec_policy_hash_t 	*ipr_hash;
447 } ipsec_policy_root_t;
448 
449 /*
450  * Policy head.  One for system policy; there may also be one present
451  * on ill_t's with interface-specific policy, as well as one present
452  * for sockets with per-socket policy allocated.
453  */
454 
455 typedef struct ipsec_policy_head_s
456 {
457 	uint32_t	iph_refs;
458 	krwlock_t	iph_lock;
459 	uint64_t	iph_gen; /* generation number */
460 	ipsec_policy_root_t iph_root[IPSEC_NTYPES];
461 	avl_tree_t	iph_rulebyid;
462 } ipsec_policy_head_t;
463 
464 #define	IPPH_REFHOLD(iph) {			\
465 	atomic_inc_32(&(iph)->iph_refs);	\
466 	ASSERT((iph)->iph_refs != 0);		\
467 }
468 #define	IPPH_REFRELE(iph, ns) {					\
469 	ASSERT((iph)->iph_refs != 0);				\
470 	membar_exit();						\
471 	if (atomic_dec_32_nv(&(iph)->iph_refs) == 0)	\
472 		ipsec_polhead_free(iph, ns);			\
473 	(iph) = 0;						\
474 }
475 
476 /*
477  * IPsec fragment related structures
478  */
479 
480 typedef struct ipsec_fragcache_entry {
481 	struct ipsec_fragcache_entry *itpfe_next;	/* hash list chain */
482 	mblk_t *itpfe_fraglist;			/* list of fragments */
483 	time_t itpfe_exp;			/* time when entry is stale */
484 	int itpfe_depth;			/* # of fragments in list */
485 	ipsec_addr_t itpfe_frag_src;
486 	ipsec_addr_t itpfe_frag_dst;
487 #define	itpfe_src itpfe_frag_src.ipsad_v4
488 #define	itpfe_src6 itpfe_frag_src.ipsad_v6
489 #define	itpfe_dst itpfe_frag_dst.ipsad_v4
490 #define	itpfe_dst6 itpfe_frag_dst.ipsad_v6
491 	uint32_t itpfe_id;			/* IP datagram ID */
492 	uint8_t itpfe_proto;			/* IP Protocol */
493 	uint8_t itpfe_last;			/* Last packet */
494 } ipsec_fragcache_entry_t;
495 
496 typedef struct ipsec_fragcache {
497 	kmutex_t itpf_lock;
498 	struct ipsec_fragcache_entry **itpf_ptr;
499 	struct ipsec_fragcache_entry *itpf_freelist;
500 	time_t itpf_expire_hint;	/* time when oldest entry is stale */
501 } ipsec_fragcache_t;
502 
503 /*
504  * Tunnel policies.  We keep a minature of the transport-mode/global policy
505  * per each tunnel instance.
506  *
507  * People who need both an itp held down AND one of its polheads need to
508  * first lock the itp, THEN the polhead, otherwise deadlock WILL occur.
509  */
510 typedef struct ipsec_tun_pol_s {
511 	avl_node_t itp_node;
512 	kmutex_t itp_lock;
513 	uint64_t itp_next_policy_index;
514 	ipsec_policy_head_t *itp_policy;
515 	ipsec_policy_head_t *itp_inactive;
516 	uint32_t itp_flags;
517 	uint32_t itp_refcnt;
518 	char itp_name[LIFNAMSIZ];
519 	ipsec_fragcache_t itp_fragcache;
520 } ipsec_tun_pol_t;
521 /* NOTE - Callers (tun code) synchronize their own instances for these flags. */
522 #define	ITPF_P_ACTIVE 0x1	/* Are we using IPsec right now? */
523 #define	ITPF_P_TUNNEL 0x2	/* Negotiate tunnel-mode */
524 /* Optimization -> Do we have per-port security entries in this polhead? */
525 #define	ITPF_P_PER_PORT_SECURITY 0x4
526 #define	ITPF_PFLAGS 0x7
527 #define	ITPF_SHIFT 3
528 
529 #define	ITPF_I_ACTIVE 0x8	/* Is the inactive using IPsec right now? */
530 #define	ITPF_I_TUNNEL 0x10	/* Negotiate tunnel-mode (on inactive) */
531 /* Optimization -> Do we have per-port security entries in this polhead? */
532 #define	ITPF_I_PER_PORT_SECURITY 0x20
533 #define	ITPF_IFLAGS 0x38
534 
535 /* NOTE:  f cannot be an expression. */
536 #define	ITPF_CLONE(f) (f) = (((f) & ITPF_PFLAGS) | \
537 	    (((f) & ITPF_PFLAGS) << ITPF_SHIFT));
538 #define	ITPF_SWAP(f) (f) = ((((f) & ITPF_PFLAGS) << ITPF_SHIFT) | \
539 	    (((f) & ITPF_IFLAGS) >> ITPF_SHIFT))
540 
541 #define	ITP_P_ISACTIVE(itp, iph) ((itp)->itp_flags & \
542 	(((itp)->itp_policy == (iph)) ? ITPF_P_ACTIVE : ITPF_I_ACTIVE))
543 
544 #define	ITP_P_ISTUNNEL(itp, iph) ((itp)->itp_flags & \
545 	(((itp)->itp_policy == (iph)) ? ITPF_P_TUNNEL : ITPF_I_TUNNEL))
546 
547 #define	ITP_P_ISPERPORT(itp, iph) ((itp)->itp_flags & \
548 	(((itp)->itp_policy == (iph)) ? ITPF_P_PER_PORT_SECURITY : \
549 	ITPF_I_PER_PORT_SECURITY))
550 
551 #define	ITP_REFHOLD(itp) { \
552 	atomic_inc_32(&((itp)->itp_refcnt));	\
553 	ASSERT((itp)->itp_refcnt != 0); \
554 }
555 
556 #define	ITP_REFRELE(itp, ns) { \
557 	ASSERT((itp)->itp_refcnt != 0); \
558 	membar_exit(); \
559 	if (atomic_dec_32_nv(&((itp)->itp_refcnt)) == 0) \
560 		itp_free(itp, ns); \
561 }
562 
563 /*
564  * Certificate identity.
565  */
566 
567 typedef struct ipsid_s
568 {
569 	struct ipsid_s *ipsid_next;
570 	struct ipsid_s **ipsid_ptpn;
571 	uint32_t	ipsid_refcnt;
572 	int		ipsid_type;	/* id type */
573 	char 		*ipsid_cid;	/* certificate id string */
574 } ipsid_t;
575 
576 /*
577  * ipsid_t reference hold/release macros, just like ipsa versions.
578  */
579 
580 #define	IPSID_REFHOLD(ipsid) {			\
581 	atomic_inc_32(&(ipsid)->ipsid_refcnt);	\
582 	ASSERT((ipsid)->ipsid_refcnt != 0);	\
583 }
584 
585 /*
586  * Decrement the reference count on the ID.  Someone else will clean up
587  * after us later.
588  */
589 
590 #define	IPSID_REFRELE(ipsid) {					\
591 	membar_exit();						\
592 	atomic_dec_32(&(ipsid)->ipsid_refcnt);		\
593 }
594 
595 /*
596  * Following are the estimates of what the maximum AH and ESP header size
597  * would be. This is used to tell the upper layer the right value of MSS
598  * it should use without consulting AH/ESP. If the size is something
599  * different from this, ULP will learn the right one through
600  * ICMP_FRAGMENTATION_NEEDED messages generated locally.
601  *
602  * AH : 12 bytes of constant header + 32 bytes of ICV checksum (SHA-512).
603  */
604 #define	IPSEC_MAX_AH_HDR_SIZE   (44)
605 
606 /*
607  * ESP : Is a bit more complex...
608  *
609  * A system of one inequality and one equation MUST be solved for proper ESP
610  * overhead.  The inequality is:
611  *
612  *    MTU - sizeof (IP header + options) >=
613  *		sizeof (esph_t) + sizeof (IV or ctr) + data-size + 2 + ICV
614  *
615  * IV or counter is almost always the cipher's block size.  The equation is:
616  *
617  *    data-size % block-size = (block-size - 2)
618  *
619  * so we can put as much data into the datagram as possible.  If we are
620  * pessimistic and include our largest overhead cipher (AES) and hash
621  * (SHA-512), and assume 1500-byte MTU minus IPv4 overhead of 20 bytes, we get:
622  *
623  *    1480 >= 8 + 16 + data-size + 2 + 32
624  *    1480 >= 58 + data-size
625  *    1422 >= data-size,      1422 % 16 = 14, so 58 is the overhead!
626  *
627  * But, let's re-run the numbers with the same algorithms, but with an IPv6
628  * header:
629  *
630  *    1460 >= 58 + data-size
631  *    1402 >= data-size,     1402 % 16 = 10, meaning shrink to 1390 to get 14,
632  *
633  * which means the overhead is now 70.
634  *
635  * Hmmm... IPv4 headers can never be anything other than multiples of 4-bytes,
636  * and IPv6 ones can never be anything other than multiples of 8-bytes.  We've
637  * seen overheads of 58 and 70.  58 % 16 == 10, and 70 % 16 == 6.  IPv4 could
638  * force us to have 62 ( % 16 == 14) or 66 ( % 16 == 2), or IPv6 could force us
639  * to have 78 ( % 16 = 14).  Let's compute IPv6 + 8-bytes of options:
640  *
641  *    1452 >= 58 + data-size
642  *    1394 >= data-size,     1394 % 16 = 2, meaning shrink to 1390 to get 14,
643  *
644  * Aha!  The "ESP overhead" shrinks to 62 (70 - 8).  This is good.  Let's try
645  * IPv4 + 8 bytes of IPv4 options:
646  *
647  *    1472 >= 58 + data-size
648  *    1414 >= data-size,      1414 % 16 = 6, meaning shrink to 1406,
649  *
650  * meaning 66 is the overhead.  Let's try 12 bytes:
651  *
652  *    1468 >= 58 + data-size
653  *    1410 >= data-size,      1410 % 16 = 2, meaning also shrink to 1406,
654  *
655  * meaning 62 is the overhead.  How about 16 bytes?
656  *
657  *    1464 >= 58 + data-size
658  *    1406 >= data-size,      1402 % 16 = 14, which is great!
659  *
660  * this means 58 is the overhead.  If I wrap and add 20 bytes, it looks just
661  * like IPv6's 70 bytes.  If I add 24, we go back to 66 bytes.
662  *
663  * So picking 70 is a sensible, conservative default.  Optimal calculations
664  * will depend on knowing pre-ESP header length (called "divpoint" in the ESP
665  * code), which could be cached in the conn_t for connected endpoints, or
666  * which must be computed on every datagram otherwise.
667  */
668 #define	IPSEC_MAX_ESP_HDR_SIZE  (70)
669 
670 /*
671  * Alternate, when we know the crypto block size via the SA.  Assume an ICV on
672  * the SA.  Use:
673  *
674  * sizeof (esph_t) + 2 * (sizeof (IV/counter)) - 2 + sizeof (ICV).  The "-2"
675  * discounts the overhead of the pad + padlen that gets swallowed up by the
676  * second (theoretically all-pad) cipher-block.  If you use our examples of
677  * AES and SHA512, you get:
678  *
679  *    8 + 32 - 2 + 32 == 70.
680  *
681  * Which is our pre-computed maximum above.
682  */
683 #include <inet/ipsecesp.h>
684 #define	IPSEC_BASE_ESP_HDR_SIZE(sa) \
685 	(sizeof (esph_t) + ((sa)->ipsa_iv_len << 1) - 2 + (sa)->ipsa_mac_len)
686 
687 /*
688  * Identity hash table.
689  *
690  * Identities are refcounted and "interned" into the hash table.
691  * Only references coming from other objects (SA's, latching state)
692  * are counted in ipsid_refcnt.
693  *
694  * Locking: IPSID_REFHOLD is safe only when (a) the object's hash bucket
695  * is locked, (b) we know that the refcount must be > 0.
696  *
697  * The ipsid_next and ipsid_ptpn fields are only to be referenced or
698  * modified when the bucket lock is held; in particular, we only
699  * delete objects while holding the bucket lock, and we only increase
700  * the refcount from 0 to 1 while the bucket lock is held.
701  */
702 
703 #define	IPSID_HASHSIZE 64
704 
705 typedef struct ipsif_s
706 {
707 	ipsid_t *ipsif_head;
708 	kmutex_t ipsif_lock;
709 } ipsif_t;
710 
711 /*
712  * For call to the kernel crypto framework. State needed during
713  * the execution of a crypto request.
714  */
715 typedef struct ipsec_crypto_s {
716 	size_t		ic_skip_len;		/* len to skip for AH auth */
717 	crypto_data_t	ic_crypto_data;		/* single op crypto data */
718 	crypto_dual_data_t ic_crypto_dual_data; /* for dual ops */
719 	crypto_data_t	ic_crypto_mac;		/* to store the MAC */
720 	ipsa_cm_mech_t	ic_cmm;
721 } ipsec_crypto_t;
722 
723 /*
724  * IPsec stack instances
725  */
726 struct ipsec_stack {
727 	netstack_t		*ipsec_netstack;	/* Common netstack */
728 
729 	/* Packet dropper for IP IPsec processing failures */
730 	ipdropper_t		ipsec_dropper;
731 
732 /* From spd.c */
733 	/*
734 	 * Policy rule index generator.  We assume this won't wrap in the
735 	 * lifetime of a system.  If we make 2^20 policy changes per second,
736 	 * this will last 2^44 seconds, or roughly 500,000 years, so we don't
737 	 * have to worry about reusing policy index values.
738 	 */
739 	uint64_t		ipsec_next_policy_index;
740 
741 	HASH_HEAD(ipsec_action_s) ipsec_action_hash[IPSEC_ACTION_HASH_SIZE];
742 	HASH_HEAD(ipsec_sel)	  *ipsec_sel_hash;
743 	uint32_t		ipsec_spd_hashsize;
744 
745 	ipsif_t			ipsec_ipsid_buckets[IPSID_HASHSIZE];
746 
747 	/*
748 	 * Active & Inactive system policy roots
749 	 */
750 	ipsec_policy_head_t	ipsec_system_policy;
751 	ipsec_policy_head_t	ipsec_inactive_policy;
752 
753 	/* Packet dropper for generic SPD drops. */
754 	ipdropper_t		ipsec_spd_dropper;
755 
756 /* ipdrop.c */
757 	kstat_t			*ipsec_ip_drop_kstat;
758 	struct ip_dropstats	*ipsec_ip_drop_types;
759 
760 /* spd.c */
761 	/*
762 	 * Have a counter for every possible policy message in
763 	 * ipsec_policy_failure_msgs
764 	 */
765 	uint32_t		ipsec_policy_failure_count[IPSEC_POLICY_MAX];
766 	/* Time since last ipsec policy failure that printed a message. */
767 	hrtime_t		ipsec_policy_failure_last;
768 
769 /* ip_spd.c */
770 	/* stats */
771 	kstat_t			*ipsec_ksp;
772 	struct ipsec_kstats_s	*ipsec_kstats;
773 
774 /* sadb.c */
775 	/* Packet dropper for generic SADB drops. */
776 	ipdropper_t		ipsec_sadb_dropper;
777 
778 /* spd.c */
779 	boolean_t		ipsec_inbound_v4_policy_present;
780 	boolean_t		ipsec_outbound_v4_policy_present;
781 	boolean_t		ipsec_inbound_v6_policy_present;
782 	boolean_t		ipsec_outbound_v6_policy_present;
783 
784 /* spd.c */
785 	/*
786 	 * Because policy needs to know what algorithms are supported, keep the
787 	 * lists of algorithms here.
788 	 */
789 	krwlock_t 		ipsec_alg_lock;
790 
791 	uint8_t			ipsec_nalgs[IPSEC_NALGTYPES];
792 	ipsec_alginfo_t	*ipsec_alglists[IPSEC_NALGTYPES][IPSEC_MAX_ALGS];
793 
794 	uint8_t		ipsec_sortlist[IPSEC_NALGTYPES][IPSEC_MAX_ALGS];
795 
796 	int		ipsec_algs_exec_mode[IPSEC_NALGTYPES];
797 
798 	uint32_t 	ipsec_tun_spd_hashsize;
799 	/*
800 	 * Tunnel policies - AVL tree indexed by tunnel name.
801 	 */
802 	krwlock_t 	ipsec_tunnel_policy_lock;
803 	uint64_t	ipsec_tunnel_policy_gen;
804 	avl_tree_t	ipsec_tunnel_policies;
805 
806 /* ipsec_loader.c */
807 	kmutex_t	ipsec_loader_lock;
808 	int		ipsec_loader_state;
809 	int		ipsec_loader_sig;
810 	kt_did_t	ipsec_loader_tid;
811 	kcondvar_t	ipsec_loader_sig_cv;	/* For loader_sig conditions. */
812 
813 };
814 typedef struct ipsec_stack ipsec_stack_t;
815 
816 /* Handle the kstat_create in ip_drop_init() failing */
817 #define	DROPPER(_ipss, _dropper) \
818 	(((_ipss)->ipsec_ip_drop_types == NULL) ? NULL : \
819 	&((_ipss)->ipsec_ip_drop_types->_dropper))
820 
821 /*
822  * Loader states..
823  */
824 #define	IPSEC_LOADER_WAIT	0
825 #define	IPSEC_LOADER_FAILED	-1
826 #define	IPSEC_LOADER_SUCCEEDED	1
827 
828 /*
829  * ipsec_loader entrypoints.
830  */
831 extern void ipsec_loader_init(ipsec_stack_t *);
832 extern void ipsec_loader_start(ipsec_stack_t *);
833 extern void ipsec_loader_destroy(ipsec_stack_t *);
834 extern void ipsec_loader_loadnow(ipsec_stack_t *);
835 extern boolean_t ipsec_loader_wait(queue_t *q, ipsec_stack_t *);
836 extern boolean_t ipsec_loaded(ipsec_stack_t *);
837 extern boolean_t ipsec_failed(ipsec_stack_t *);
838 
839 /*
840  * ipsec policy entrypoints (spd.c)
841  */
842 
843 extern void ipsec_policy_g_destroy(void);
844 extern void ipsec_policy_g_init(void);
845 
846 extern mblk_t	*ipsec_add_crypto_data(mblk_t *, ipsec_crypto_t **);
847 extern mblk_t	*ipsec_remove_crypto_data(mblk_t *, ipsec_crypto_t **);
848 extern mblk_t	*ipsec_free_crypto_data(mblk_t *);
849 extern int ipsec_alloc_table(ipsec_policy_head_t *, int, int, boolean_t,
850     netstack_t *);
851 extern void ipsec_polhead_init(ipsec_policy_head_t *, int);
852 extern void ipsec_polhead_destroy(ipsec_policy_head_t *);
853 extern void ipsec_polhead_free_table(ipsec_policy_head_t *);
854 extern mblk_t *ipsec_check_global_policy(mblk_t *, conn_t *, ipha_t *,
855     ip6_t *, ip_recv_attr_t *, netstack_t *ns);
856 extern mblk_t *ipsec_check_inbound_policy(mblk_t *, conn_t *, ipha_t *, ip6_t *,
857     ip_recv_attr_t *);
858 
859 extern boolean_t ipsec_in_to_out(ip_recv_attr_t *, ip_xmit_attr_t *,
860     mblk_t *, ipha_t *, ip6_t *);
861 extern void ipsec_in_release_refs(ip_recv_attr_t *);
862 extern void ipsec_out_release_refs(ip_xmit_attr_t *);
863 extern void ipsec_log_policy_failure(int, char *, ipha_t *, ip6_t *, boolean_t,
864     netstack_t *);
865 extern boolean_t ipsec_inbound_accept_clear(mblk_t *, ipha_t *, ip6_t *);
866 extern int ipsec_conn_cache_policy(conn_t *, boolean_t);
867 extern void ipsec_cache_outbound_policy(const conn_t *, const in6_addr_t *,
868     const in6_addr_t *, in_port_t, ip_xmit_attr_t *);
869 extern boolean_t ipsec_outbound_policy_current(ip_xmit_attr_t *);
870 extern ipsec_action_t *ipsec_in_to_out_action(ip_recv_attr_t *);
871 extern void ipsec_latch_inbound(conn_t *connp, ip_recv_attr_t *ira);
872 
873 extern void ipsec_policy_free(ipsec_policy_t *);
874 extern void ipsec_action_free(ipsec_action_t *);
875 extern void ipsec_polhead_free(ipsec_policy_head_t *, netstack_t *);
876 extern ipsec_policy_head_t *ipsec_polhead_split(ipsec_policy_head_t *,
877     netstack_t *);
878 extern ipsec_policy_head_t *ipsec_polhead_create(void);
879 extern ipsec_policy_head_t *ipsec_system_policy(netstack_t *);
880 extern ipsec_policy_head_t *ipsec_inactive_policy(netstack_t *);
881 extern void ipsec_swap_policy(ipsec_policy_head_t *, ipsec_policy_head_t *,
882     netstack_t *);
883 extern void ipsec_swap_global_policy(netstack_t *);
884 
885 extern int ipsec_clone_system_policy(netstack_t *);
886 extern ipsec_policy_t *ipsec_policy_create(ipsec_selkey_t *,
887     const ipsec_act_t *, int, int, uint64_t *, netstack_t *);
888 extern boolean_t ipsec_policy_delete(ipsec_policy_head_t *,
889     ipsec_selkey_t *, int, netstack_t *);
890 extern int ipsec_policy_delete_index(ipsec_policy_head_t *, uint64_t,
891     netstack_t *);
892 extern boolean_t ipsec_polhead_insert(ipsec_policy_head_t *, ipsec_act_t *,
893     uint_t, int, int, netstack_t *);
894 extern void ipsec_polhead_flush(ipsec_policy_head_t *, netstack_t *);
895 extern int ipsec_copy_polhead(ipsec_policy_head_t *, ipsec_policy_head_t *,
896     netstack_t *);
897 extern void ipsec_actvec_from_req(const ipsec_req_t *, ipsec_act_t **, uint_t *,
898     netstack_t *);
899 extern void ipsec_actvec_free(ipsec_act_t *, uint_t);
900 extern int ipsec_req_from_head(ipsec_policy_head_t *, ipsec_req_t *, int);
901 extern mblk_t *ipsec_construct_inverse_acquire(sadb_msg_t *, sadb_ext_t **,
902     netstack_t *);
903 extern ipsec_policy_t *ipsec_find_policy(int, const conn_t *,
904     ipsec_selector_t *, netstack_t *);
905 extern ipsid_t *ipsid_lookup(int, char *, netstack_t *);
906 extern boolean_t ipsid_equal(ipsid_t *, ipsid_t *);
907 extern void ipsid_gc(netstack_t *);
908 extern void ipsec_latch_ids(ipsec_latch_t *, ipsid_t *, ipsid_t *);
909 
910 extern void ipsec_config_flush(netstack_t *);
911 extern boolean_t ipsec_check_policy(ipsec_policy_head_t *, ipsec_policy_t *,
912     int);
913 extern void ipsec_enter_policy(ipsec_policy_head_t *, ipsec_policy_t *, int,
914     netstack_t *);
915 extern boolean_t ipsec_check_action(ipsec_act_t *, int *, netstack_t *);
916 
917 extern void iplatch_free(ipsec_latch_t *);
918 extern ipsec_latch_t *iplatch_create(void);
919 extern int ipsec_set_req(cred_t *, conn_t *, ipsec_req_t *);
920 
921 extern void ipsec_insert_always(avl_tree_t *tree, void *new_node);
922 
923 extern int32_t ipsec_act_ovhd(const ipsec_act_t *act);
924 extern mblk_t *sadb_whack_label(mblk_t *, ipsa_t *, ip_xmit_attr_t *,
925     kstat_named_t *, ipdropper_t *);
926 extern mblk_t *sadb_whack_label_v4(mblk_t *, ipsa_t *, kstat_named_t *,
927     ipdropper_t *);
928 extern mblk_t *sadb_whack_label_v6(mblk_t *, ipsa_t *, kstat_named_t *,
929     ipdropper_t *);
930 extern boolean_t update_iv(uint8_t *, queue_t *, ipsa_t *, ipsecesp_stack_t *);
931 
932 /*
933  * Tunnel-support SPD functions and variables.
934  */
935 struct iptun_s;	/* Defined in inet/iptun/iptun_impl.h. */
936 extern mblk_t *ipsec_tun_inbound(ip_recv_attr_t *, mblk_t *,  ipsec_tun_pol_t *,
937     ipha_t *, ip6_t *, ipha_t *, ip6_t *, int, netstack_t *);
938 extern mblk_t *ipsec_tun_outbound(mblk_t *, struct iptun_s *, ipha_t *,
939     ip6_t *, ipha_t *, ip6_t *, int, ip_xmit_attr_t *);
940 extern void itp_free(ipsec_tun_pol_t *, netstack_t *);
941 extern ipsec_tun_pol_t *create_tunnel_policy(char *, int *, uint64_t *,
942     netstack_t *);
943 extern ipsec_tun_pol_t *get_tunnel_policy(char *, netstack_t *);
944 extern void itp_unlink(ipsec_tun_pol_t *, netstack_t *);
945 extern void itp_walk(void (*)(ipsec_tun_pol_t *, void *, netstack_t *),
946     void *, netstack_t *);
947 
948 extern ipsec_tun_pol_t *itp_get_byaddr(uint32_t *, uint32_t *, int,
949     ip_stack_t *);
950 
951 /*
952  * IPsec AH/ESP functions called from IP or the common SADB code in AH.
953  */
954 
955 extern void ipsecah_in_assocfailure(mblk_t *, char, ushort_t, char *,
956     uint32_t, void *, int, ip_recv_attr_t *ira);
957 extern void ipsecesp_in_assocfailure(mblk_t *, char, ushort_t, char *,
958     uint32_t, void *, int, ip_recv_attr_t *ira);
959 extern void ipsecesp_send_keepalive(ipsa_t *);
960 
961 /*
962  * Algorithm management helper functions.
963  */
964 extern boolean_t ipsec_valid_key_size(uint16_t, ipsec_alginfo_t *);
965 
966 /*
967  * Per-socket policy, for now, takes precedence... this priority value
968  * insures it.
969  */
970 #define	IPSEC_PRIO_SOCKET		0x1000000
971 
972 /* DDI initialization functions. */
973 extern	boolean_t    ipsecesp_ddi_init(void);
974 extern	boolean_t    ipsecah_ddi_init(void);
975 extern	boolean_t    keysock_ddi_init(void);
976 extern	boolean_t    spdsock_ddi_init(void);
977 
978 extern	void    ipsecesp_ddi_destroy(void);
979 extern	void    ipsecah_ddi_destroy(void);
980 extern	void	keysock_ddi_destroy(void);
981 extern	void    spdsock_ddi_destroy(void);
982 
983 /*
984  * AH- and ESP-specific functions that are called directly by other modules.
985  */
986 extern void ipsecah_fill_defs(struct sadb_x_ecomb *, netstack_t *);
987 extern void ipsecesp_fill_defs(struct sadb_x_ecomb *, netstack_t *);
988 extern void ipsecah_algs_changed(netstack_t *);
989 extern void ipsecesp_algs_changed(netstack_t *);
990 extern void ipsecesp_init_funcs(ipsa_t *);
991 extern void ipsecah_init_funcs(ipsa_t *);
992 extern mblk_t *ipsecah_icmp_error(mblk_t *, ip_recv_attr_t *);
993 extern mblk_t *ipsecesp_icmp_error(mblk_t *, ip_recv_attr_t *);
994 
995 /*
996  * spdsock functions that are called directly by IP.
997  */
998 extern void spdsock_update_pending_algs(netstack_t *);
999 
1000 /*
1001  * IP functions that are called from AH and ESP.
1002  */
1003 extern boolean_t ipsec_outbound_sa(mblk_t *, ip_xmit_attr_t *, uint_t);
1004 extern mblk_t *ipsec_inbound_esp_sa(mblk_t *, ip_recv_attr_t *, esph_t **);
1005 extern mblk_t *ipsec_inbound_ah_sa(mblk_t *, ip_recv_attr_t *, ah_t **);
1006 extern ipsec_policy_t *ipsec_find_policy_head(ipsec_policy_t *,
1007     ipsec_policy_head_t *, int, ipsec_selector_t *);
1008 
1009 /*
1010  * IP dropper init/destroy.
1011  */
1012 void ip_drop_init(ipsec_stack_t *);
1013 void ip_drop_destroy(ipsec_stack_t *);
1014 
1015 /*
1016  * Common functions
1017  */
1018 extern boolean_t ip_addr_match(uint8_t *, int, in6_addr_t *);
1019 extern boolean_t ipsec_label_match(ts_label_t *, ts_label_t *);
1020 
1021 /*
1022  * AH and ESP counters types.
1023  */
1024 typedef uint32_t ah_counter;
1025 typedef uint32_t esp_counter;
1026 
1027 #endif /* _KERNEL */
1028 
1029 #ifdef	__cplusplus
1030 }
1031 #endif
1032 
1033 #endif	/* _INET_IPSEC_IMPL_H */
1034