xref: /illumos-gate/usr/src/uts/common/inet/ip/ipclassifier.c (revision 6a604193b70017bd933cd973200b3f13803674b2)
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 2008 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  * IP PACKET CLASSIFIER
28  *
29  * The IP packet classifier provides mapping between IP packets and persistent
30  * connection state for connection-oriented protocols. It also provides
31  * interface for managing connection states.
32  *
33  * The connection state is kept in conn_t data structure and contains, among
34  * other things:
35  *
36  *	o local/remote address and ports
37  *	o Transport protocol
38  *	o squeue for the connection (for TCP only)
39  *	o reference counter
40  *	o Connection state
41  *	o hash table linkage
42  *	o interface/ire information
43  *	o credentials
44  *	o ipsec policy
45  *	o send and receive functions.
46  *	o mutex lock.
47  *
48  * Connections use a reference counting scheme. They are freed when the
49  * reference counter drops to zero. A reference is incremented when connection
50  * is placed in a list or table, when incoming packet for the connection arrives
51  * and when connection is processed via squeue (squeue processing may be
52  * asynchronous and the reference protects the connection from being destroyed
53  * before its processing is finished).
54  *
55  * send and receive functions are currently used for TCP only. The send function
56  * determines the IP entry point for the packet once it leaves TCP to be sent to
57  * the destination address. The receive function is used by IP when the packet
58  * should be passed for TCP processing. When a new connection is created these
59  * are set to ip_output() and tcp_input() respectively. During the lifetime of
60  * the connection the send and receive functions may change depending on the
61  * changes in the connection state. For example, Once the connection is bound to
62  * an addresse, the receive function for this connection is set to
63  * tcp_conn_request().  This allows incoming SYNs to go directly into the
64  * listener SYN processing function without going to tcp_input() first.
65  *
66  * Classifier uses several hash tables:
67  *
68  * 	ipcl_conn_fanout:	contains all TCP connections in CONNECTED state
69  *	ipcl_bind_fanout:	contains all connections in BOUND state
70  *	ipcl_proto_fanout:	IPv4 protocol fanout
71  *	ipcl_proto_fanout_v6:	IPv6 protocol fanout
72  *	ipcl_udp_fanout:	contains all UDP connections
73  *	ipcl_globalhash_fanout:	contains all connections
74  *
75  * The ipcl_globalhash_fanout is used for any walkers (like snmp and Clustering)
76  * which need to view all existing connections.
77  *
78  * All tables are protected by per-bucket locks. When both per-bucket lock and
79  * connection lock need to be held, the per-bucket lock should be acquired
80  * first, followed by the connection lock.
81  *
82  * All functions doing search in one of these tables increment a reference
83  * counter on the connection found (if any). This reference should be dropped
84  * when the caller has finished processing the connection.
85  *
86  *
87  * INTERFACES:
88  * ===========
89  *
90  * Connection Lookup:
91  * ------------------
92  *
93  * conn_t *ipcl_classify_v4(mp, protocol, hdr_len, zoneid, ip_stack)
94  * conn_t *ipcl_classify_v6(mp, protocol, hdr_len, zoneid, ip_stack)
95  *
96  * Finds connection for an incoming IPv4 or IPv6 packet. Returns NULL if
97  * it can't find any associated connection. If the connection is found, its
98  * reference counter is incremented.
99  *
100  *	mp:	mblock, containing packet header. The full header should fit
101  *		into a single mblock. It should also contain at least full IP
102  *		and TCP or UDP header.
103  *
104  *	protocol: Either IPPROTO_TCP or IPPROTO_UDP.
105  *
106  *	hdr_len: The size of IP header. It is used to find TCP or UDP header in
107  *		 the packet.
108  *
109  * 	zoneid: The zone in which the returned connection must be; the zoneid
110  *		corresponding to the ire_zoneid on the IRE located for the
111  *		packet's destination address.
112  *
113  *	For TCP connections, the lookup order is as follows:
114  *		5-tuple {src, dst, protocol, local port, remote port}
115  *			lookup in ipcl_conn_fanout table.
116  *		3-tuple {dst, remote port, protocol} lookup in
117  *			ipcl_bind_fanout table.
118  *
119  *	For UDP connections, a 5-tuple {src, dst, protocol, local port,
120  *	remote port} lookup is done on ipcl_udp_fanout. Note that,
121  *	these interfaces do not handle cases where a packets belongs
122  *	to multiple UDP clients, which is handled in IP itself.
123  *
124  * If the destination IRE is ALL_ZONES (indicated by zoneid), then we must
125  * determine which actual zone gets the segment.  This is used only in a
126  * labeled environment.  The matching rules are:
127  *
128  *	- If it's not a multilevel port, then the label on the packet selects
129  *	  the zone.  Unlabeled packets are delivered to the global zone.
130  *
131  *	- If it's a multilevel port, then only the zone registered to receive
132  *	  packets on that port matches.
133  *
134  * Also, in a labeled environment, packet labels need to be checked.  For fully
135  * bound TCP connections, we can assume that the packet label was checked
136  * during connection establishment, and doesn't need to be checked on each
137  * packet.  For others, though, we need to check for strict equality or, for
138  * multilevel ports, membership in the range or set.  This part currently does
139  * a tnrh lookup on each packet, but could be optimized to use cached results
140  * if that were necessary.  (SCTP doesn't come through here, but if it did,
141  * we would apply the same rules as TCP.)
142  *
143  * An implication of the above is that fully-bound TCP sockets must always use
144  * distinct 4-tuples; they can't be discriminated by label alone.
145  *
146  * Note that we cannot trust labels on packets sent to fully-bound UDP sockets,
147  * as there's no connection set-up handshake and no shared state.
148  *
149  * Labels on looped-back packets within a single zone do not need to be
150  * checked, as all processes in the same zone have the same label.
151  *
152  * Finally, for unlabeled packets received by a labeled system, special rules
153  * apply.  We consider only the MLP if there is one.  Otherwise, we prefer a
154  * socket in the zone whose label matches the default label of the sender, if
155  * any.  In any event, the receiving socket must have SO_MAC_EXEMPT set and the
156  * receiver's label must dominate the sender's default label.
157  *
158  * conn_t *ipcl_tcp_lookup_reversed_ipv4(ipha_t *, tcph_t *, int, ip_stack);
159  * conn_t *ipcl_tcp_lookup_reversed_ipv6(ip6_t *, tcpha_t *, int, uint_t,
160  *					 ip_stack);
161  *
162  *	Lookup routine to find a exact match for {src, dst, local port,
163  *	remote port) for TCP connections in ipcl_conn_fanout. The address and
164  *	ports are read from the IP and TCP header respectively.
165  *
166  * conn_t	*ipcl_lookup_listener_v4(lport, laddr, protocol,
167  *					 zoneid, ip_stack);
168  * conn_t	*ipcl_lookup_listener_v6(lport, laddr, protocol, ifindex,
169  *					 zoneid, ip_stack);
170  *
171  * 	Lookup routine to find a listener with the tuple {lport, laddr,
172  * 	protocol} in the ipcl_bind_fanout table. For IPv6, an additional
173  * 	parameter interface index is also compared.
174  *
175  * void ipcl_walk(func, arg, ip_stack)
176  *
177  * 	Apply 'func' to every connection available. The 'func' is called as
178  *	(*func)(connp, arg). The walk is non-atomic so connections may be
179  *	created and destroyed during the walk. The CONN_CONDEMNED and
180  *	CONN_INCIPIENT flags ensure that connections which are newly created
181  *	or being destroyed are not selected by the walker.
182  *
183  * Table Updates
184  * -------------
185  *
186  * int ipcl_conn_insert(connp, protocol, src, dst, ports)
187  * int ipcl_conn_insert_v6(connp, protocol, src, dst, ports, ifindex)
188  *
189  *	Insert 'connp' in the ipcl_conn_fanout.
190  *	Arguements :
191  *		connp		conn_t to be inserted
192  *		protocol	connection protocol
193  *		src		source address
194  *		dst		destination address
195  *		ports		local and remote port
196  *		ifindex		interface index for IPv6 connections
197  *
198  *	Return value :
199  *		0		if connp was inserted
200  *		EADDRINUSE	if the connection with the same tuple
201  *				already exists.
202  *
203  * int ipcl_bind_insert(connp, protocol, src, lport);
204  * int ipcl_bind_insert_v6(connp, protocol, src, lport);
205  *
206  * 	Insert 'connp' in ipcl_bind_fanout.
207  * 	Arguements :
208  * 		connp		conn_t to be inserted
209  * 		protocol	connection protocol
210  * 		src		source address connection wants
211  * 				to bind to
212  * 		lport		local port connection wants to
213  * 				bind to
214  *
215  *
216  * void ipcl_hash_remove(connp);
217  *
218  * 	Removes the 'connp' from the connection fanout table.
219  *
220  * Connection Creation/Destruction
221  * -------------------------------
222  *
223  * conn_t *ipcl_conn_create(type, sleep, netstack_t *)
224  *
225  * 	Creates a new conn based on the type flag, inserts it into
226  * 	globalhash table.
227  *
228  *	type:	This flag determines the type of conn_t which needs to be
229  *		created i.e., which kmem_cache it comes from.
230  *		IPCL_TCPCONN	indicates a TCP connection
231  *		IPCL_SCTPCONN	indicates a SCTP connection
232  *		IPCL_UDPCONN	indicates a UDP conn_t.
233  *		IPCL_RAWIPCONN	indicates a RAWIP/ICMP conn_t.
234  *		IPCL_RTSCONN	indicates a RTS conn_t.
235  *		IPCL_IPCCONN	indicates all other connections.
236  *
237  * void ipcl_conn_destroy(connp)
238  *
239  * 	Destroys the connection state, removes it from the global
240  * 	connection hash table and frees its memory.
241  */
242 
243 #include <sys/types.h>
244 #include <sys/stream.h>
245 #include <sys/stropts.h>
246 #include <sys/sysmacros.h>
247 #include <sys/strsubr.h>
248 #include <sys/strsun.h>
249 #define	_SUN_TPI_VERSION 2
250 #include <sys/ddi.h>
251 #include <sys/cmn_err.h>
252 #include <sys/debug.h>
253 
254 #include <sys/systm.h>
255 #include <sys/param.h>
256 #include <sys/kmem.h>
257 #include <sys/isa_defs.h>
258 #include <inet/common.h>
259 #include <netinet/ip6.h>
260 #include <netinet/icmp6.h>
261 
262 #include <inet/ip.h>
263 #include <inet/ip6.h>
264 #include <inet/tcp.h>
265 #include <inet/ip_ndp.h>
266 #include <inet/udp_impl.h>
267 #include <inet/sctp_ip.h>
268 #include <inet/sctp/sctp_impl.h>
269 #include <inet/rawip_impl.h>
270 #include <inet/rts_impl.h>
271 
272 #include <sys/cpuvar.h>
273 
274 #include <inet/ipclassifier.h>
275 #include <inet/ipsec_impl.h>
276 
277 #include <sys/tsol/tnet.h>
278 
279 #ifdef DEBUG
280 #define	IPCL_DEBUG
281 #else
282 #undef	IPCL_DEBUG
283 #endif
284 
285 #ifdef	IPCL_DEBUG
286 int	ipcl_debug_level = 0;
287 #define	IPCL_DEBUG_LVL(level, args)	\
288 	if (ipcl_debug_level  & level) { printf args; }
289 #else
290 #define	IPCL_DEBUG_LVL(level, args) {; }
291 #endif
292 /* Old value for compatibility. Setable in /etc/system */
293 uint_t tcp_conn_hash_size = 0;
294 
295 /* New value. Zero means choose automatically.  Setable in /etc/system */
296 uint_t ipcl_conn_hash_size = 0;
297 uint_t ipcl_conn_hash_memfactor = 8192;
298 uint_t ipcl_conn_hash_maxsize = 82500;
299 
300 /* bind/udp fanout table size */
301 uint_t ipcl_bind_fanout_size = 512;
302 uint_t ipcl_udp_fanout_size = 16384;
303 
304 /* Raw socket fanout size.  Must be a power of 2. */
305 uint_t ipcl_raw_fanout_size = 256;
306 
307 /*
308  * Power of 2^N Primes useful for hashing for N of 0-28,
309  * these primes are the nearest prime <= 2^N - 2^(N-2).
310  */
311 
312 #define	P2Ps() {0, 0, 0, 5, 11, 23, 47, 89, 191, 383, 761, 1531, 3067,	\
313 		6143, 12281, 24571, 49139, 98299, 196597, 393209,	\
314 		786431, 1572853, 3145721, 6291449, 12582893, 25165813,	\
315 		50331599, 100663291, 201326557, 0}
316 
317 /*
318  * wrapper structure to ensure that conn and what follows it (tcp_t, etc)
319  * are aligned on cache lines.
320  */
321 typedef union itc_s {
322 	conn_t	itc_conn;
323 	char	itcu_filler[CACHE_ALIGN(conn_s)];
324 } itc_t;
325 
326 struct kmem_cache  *tcp_conn_cache;
327 struct kmem_cache  *ip_conn_cache;
328 extern struct kmem_cache  *sctp_conn_cache;
329 extern struct kmem_cache  *tcp_sack_info_cache;
330 extern struct kmem_cache  *tcp_iphc_cache;
331 struct kmem_cache  *udp_conn_cache;
332 struct kmem_cache  *rawip_conn_cache;
333 struct kmem_cache  *rts_conn_cache;
334 
335 extern void	tcp_timermp_free(tcp_t *);
336 extern mblk_t	*tcp_timermp_alloc(int);
337 
338 static int	ip_conn_constructor(void *, void *, int);
339 static void	ip_conn_destructor(void *, void *);
340 
341 static int	tcp_conn_constructor(void *, void *, int);
342 static void	tcp_conn_destructor(void *, void *);
343 
344 static int	udp_conn_constructor(void *, void *, int);
345 static void	udp_conn_destructor(void *, void *);
346 
347 static int	rawip_conn_constructor(void *, void *, int);
348 static void	rawip_conn_destructor(void *, void *);
349 
350 static int	rts_conn_constructor(void *, void *, int);
351 static void	rts_conn_destructor(void *, void *);
352 
353 #ifdef	IPCL_DEBUG
354 #define	INET_NTOA_BUFSIZE	18
355 
356 static char *
357 inet_ntoa_r(uint32_t in, char *b)
358 {
359 	unsigned char	*p;
360 
361 	p = (unsigned char *)&in;
362 	(void) sprintf(b, "%d.%d.%d.%d", p[0], p[1], p[2], p[3]);
363 	return (b);
364 }
365 #endif
366 
367 /*
368  * Global (for all stack instances) init routine
369  */
370 void
371 ipcl_g_init(void)
372 {
373 	ip_conn_cache = kmem_cache_create("ip_conn_cache",
374 	    sizeof (conn_t), CACHE_ALIGN_SIZE,
375 	    ip_conn_constructor, ip_conn_destructor,
376 	    NULL, NULL, NULL, 0);
377 
378 	tcp_conn_cache = kmem_cache_create("tcp_conn_cache",
379 	    sizeof (itc_t) + sizeof (tcp_t), CACHE_ALIGN_SIZE,
380 	    tcp_conn_constructor, tcp_conn_destructor,
381 	    NULL, NULL, NULL, 0);
382 
383 	udp_conn_cache = kmem_cache_create("udp_conn_cache",
384 	    sizeof (itc_t) + sizeof (udp_t), CACHE_ALIGN_SIZE,
385 	    udp_conn_constructor, udp_conn_destructor,
386 	    NULL, NULL, NULL, 0);
387 
388 	rawip_conn_cache = kmem_cache_create("rawip_conn_cache",
389 	    sizeof (itc_t) + sizeof (icmp_t), CACHE_ALIGN_SIZE,
390 	    rawip_conn_constructor, rawip_conn_destructor,
391 	    NULL, NULL, NULL, 0);
392 
393 	rts_conn_cache = kmem_cache_create("rts_conn_cache",
394 	    sizeof (itc_t) + sizeof (rts_t), CACHE_ALIGN_SIZE,
395 	    rts_conn_constructor, rts_conn_destructor,
396 	    NULL, NULL, NULL, 0);
397 }
398 
399 /*
400  * ipclassifier intialization routine, sets up hash tables.
401  */
402 void
403 ipcl_init(ip_stack_t *ipst)
404 {
405 	int i;
406 	int sizes[] = P2Ps();
407 
408 	/*
409 	 * Calculate size of conn fanout table from /etc/system settings
410 	 */
411 	if (ipcl_conn_hash_size != 0) {
412 		ipst->ips_ipcl_conn_fanout_size = ipcl_conn_hash_size;
413 	} else if (tcp_conn_hash_size != 0) {
414 		ipst->ips_ipcl_conn_fanout_size = tcp_conn_hash_size;
415 	} else {
416 		extern pgcnt_t freemem;
417 
418 		ipst->ips_ipcl_conn_fanout_size =
419 		    (freemem * PAGESIZE) / ipcl_conn_hash_memfactor;
420 
421 		if (ipst->ips_ipcl_conn_fanout_size > ipcl_conn_hash_maxsize) {
422 			ipst->ips_ipcl_conn_fanout_size =
423 			    ipcl_conn_hash_maxsize;
424 		}
425 	}
426 
427 	for (i = 9; i < sizeof (sizes) / sizeof (*sizes) - 1; i++) {
428 		if (sizes[i] >= ipst->ips_ipcl_conn_fanout_size) {
429 			break;
430 		}
431 	}
432 	if ((ipst->ips_ipcl_conn_fanout_size = sizes[i]) == 0) {
433 		/* Out of range, use the 2^16 value */
434 		ipst->ips_ipcl_conn_fanout_size = sizes[16];
435 	}
436 
437 	/* Take values from /etc/system */
438 	ipst->ips_ipcl_bind_fanout_size = ipcl_bind_fanout_size;
439 	ipst->ips_ipcl_udp_fanout_size = ipcl_udp_fanout_size;
440 	ipst->ips_ipcl_raw_fanout_size = ipcl_raw_fanout_size;
441 
442 	ASSERT(ipst->ips_ipcl_conn_fanout == NULL);
443 
444 	ipst->ips_ipcl_conn_fanout = kmem_zalloc(
445 	    ipst->ips_ipcl_conn_fanout_size * sizeof (connf_t), KM_SLEEP);
446 
447 	for (i = 0; i < ipst->ips_ipcl_conn_fanout_size; i++) {
448 		mutex_init(&ipst->ips_ipcl_conn_fanout[i].connf_lock, NULL,
449 		    MUTEX_DEFAULT, NULL);
450 	}
451 
452 	ipst->ips_ipcl_bind_fanout = kmem_zalloc(
453 	    ipst->ips_ipcl_bind_fanout_size * sizeof (connf_t), KM_SLEEP);
454 
455 	for (i = 0; i < ipst->ips_ipcl_bind_fanout_size; i++) {
456 		mutex_init(&ipst->ips_ipcl_bind_fanout[i].connf_lock, NULL,
457 		    MUTEX_DEFAULT, NULL);
458 	}
459 
460 	ipst->ips_ipcl_proto_fanout = kmem_zalloc(IPPROTO_MAX *
461 	    sizeof (connf_t), KM_SLEEP);
462 	for (i = 0; i < IPPROTO_MAX; i++) {
463 		mutex_init(&ipst->ips_ipcl_proto_fanout[i].connf_lock, NULL,
464 		    MUTEX_DEFAULT, NULL);
465 	}
466 
467 	ipst->ips_ipcl_proto_fanout_v6 = kmem_zalloc(IPPROTO_MAX *
468 	    sizeof (connf_t), KM_SLEEP);
469 	for (i = 0; i < IPPROTO_MAX; i++) {
470 		mutex_init(&ipst->ips_ipcl_proto_fanout_v6[i].connf_lock, NULL,
471 		    MUTEX_DEFAULT, NULL);
472 	}
473 
474 	ipst->ips_rts_clients = kmem_zalloc(sizeof (connf_t), KM_SLEEP);
475 	mutex_init(&ipst->ips_rts_clients->connf_lock,
476 	    NULL, MUTEX_DEFAULT, NULL);
477 
478 	ipst->ips_ipcl_udp_fanout = kmem_zalloc(
479 	    ipst->ips_ipcl_udp_fanout_size * sizeof (connf_t), KM_SLEEP);
480 	for (i = 0; i < ipst->ips_ipcl_udp_fanout_size; i++) {
481 		mutex_init(&ipst->ips_ipcl_udp_fanout[i].connf_lock, NULL,
482 		    MUTEX_DEFAULT, NULL);
483 	}
484 
485 	ipst->ips_ipcl_raw_fanout = kmem_zalloc(
486 	    ipst->ips_ipcl_raw_fanout_size * sizeof (connf_t), KM_SLEEP);
487 	for (i = 0; i < ipst->ips_ipcl_raw_fanout_size; i++) {
488 		mutex_init(&ipst->ips_ipcl_raw_fanout[i].connf_lock, NULL,
489 		    MUTEX_DEFAULT, NULL);
490 	}
491 
492 	ipst->ips_ipcl_globalhash_fanout = kmem_zalloc(
493 	    sizeof (connf_t) * CONN_G_HASH_SIZE, KM_SLEEP);
494 	for (i = 0; i < CONN_G_HASH_SIZE; i++) {
495 		mutex_init(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock,
496 		    NULL, MUTEX_DEFAULT, NULL);
497 	}
498 }
499 
500 void
501 ipcl_g_destroy(void)
502 {
503 	kmem_cache_destroy(ip_conn_cache);
504 	kmem_cache_destroy(tcp_conn_cache);
505 	kmem_cache_destroy(udp_conn_cache);
506 	kmem_cache_destroy(rawip_conn_cache);
507 	kmem_cache_destroy(rts_conn_cache);
508 }
509 
510 /*
511  * All user-level and kernel use of the stack must be gone
512  * by now.
513  */
514 void
515 ipcl_destroy(ip_stack_t *ipst)
516 {
517 	int i;
518 
519 	for (i = 0; i < ipst->ips_ipcl_conn_fanout_size; i++) {
520 		ASSERT(ipst->ips_ipcl_conn_fanout[i].connf_head == NULL);
521 		mutex_destroy(&ipst->ips_ipcl_conn_fanout[i].connf_lock);
522 	}
523 	kmem_free(ipst->ips_ipcl_conn_fanout, ipst->ips_ipcl_conn_fanout_size *
524 	    sizeof (connf_t));
525 	ipst->ips_ipcl_conn_fanout = NULL;
526 
527 	for (i = 0; i < ipst->ips_ipcl_bind_fanout_size; i++) {
528 		ASSERT(ipst->ips_ipcl_bind_fanout[i].connf_head == NULL);
529 		mutex_destroy(&ipst->ips_ipcl_bind_fanout[i].connf_lock);
530 	}
531 	kmem_free(ipst->ips_ipcl_bind_fanout, ipst->ips_ipcl_bind_fanout_size *
532 	    sizeof (connf_t));
533 	ipst->ips_ipcl_bind_fanout = NULL;
534 
535 	for (i = 0; i < IPPROTO_MAX; i++) {
536 		ASSERT(ipst->ips_ipcl_proto_fanout[i].connf_head == NULL);
537 		mutex_destroy(&ipst->ips_ipcl_proto_fanout[i].connf_lock);
538 	}
539 	kmem_free(ipst->ips_ipcl_proto_fanout, IPPROTO_MAX * sizeof (connf_t));
540 	ipst->ips_ipcl_proto_fanout = NULL;
541 
542 	for (i = 0; i < IPPROTO_MAX; i++) {
543 		ASSERT(ipst->ips_ipcl_proto_fanout_v6[i].connf_head == NULL);
544 		mutex_destroy(&ipst->ips_ipcl_proto_fanout_v6[i].connf_lock);
545 	}
546 	kmem_free(ipst->ips_ipcl_proto_fanout_v6,
547 	    IPPROTO_MAX * sizeof (connf_t));
548 	ipst->ips_ipcl_proto_fanout_v6 = NULL;
549 
550 	for (i = 0; i < ipst->ips_ipcl_udp_fanout_size; i++) {
551 		ASSERT(ipst->ips_ipcl_udp_fanout[i].connf_head == NULL);
552 		mutex_destroy(&ipst->ips_ipcl_udp_fanout[i].connf_lock);
553 	}
554 	kmem_free(ipst->ips_ipcl_udp_fanout, ipst->ips_ipcl_udp_fanout_size *
555 	    sizeof (connf_t));
556 	ipst->ips_ipcl_udp_fanout = NULL;
557 
558 	for (i = 0; i < ipst->ips_ipcl_raw_fanout_size; i++) {
559 		ASSERT(ipst->ips_ipcl_raw_fanout[i].connf_head == NULL);
560 		mutex_destroy(&ipst->ips_ipcl_raw_fanout[i].connf_lock);
561 	}
562 	kmem_free(ipst->ips_ipcl_raw_fanout, ipst->ips_ipcl_raw_fanout_size *
563 	    sizeof (connf_t));
564 	ipst->ips_ipcl_raw_fanout = NULL;
565 
566 	for (i = 0; i < CONN_G_HASH_SIZE; i++) {
567 		ASSERT(ipst->ips_ipcl_globalhash_fanout[i].connf_head == NULL);
568 		mutex_destroy(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock);
569 	}
570 	kmem_free(ipst->ips_ipcl_globalhash_fanout,
571 	    sizeof (connf_t) * CONN_G_HASH_SIZE);
572 	ipst->ips_ipcl_globalhash_fanout = NULL;
573 
574 	ASSERT(ipst->ips_rts_clients->connf_head == NULL);
575 	mutex_destroy(&ipst->ips_rts_clients->connf_lock);
576 	kmem_free(ipst->ips_rts_clients, sizeof (connf_t));
577 	ipst->ips_rts_clients = NULL;
578 }
579 
580 /*
581  * conn creation routine. initialize the conn, sets the reference
582  * and inserts it in the global hash table.
583  */
584 conn_t *
585 ipcl_conn_create(uint32_t type, int sleep, netstack_t *ns)
586 {
587 	conn_t	*connp;
588 	sctp_stack_t *sctps;
589 	struct kmem_cache *conn_cache;
590 
591 	switch (type) {
592 	case IPCL_SCTPCONN:
593 		if ((connp = kmem_cache_alloc(sctp_conn_cache, sleep)) == NULL)
594 			return (NULL);
595 		sctp_conn_init(connp);
596 		sctps = ns->netstack_sctp;
597 		SCTP_G_Q_REFHOLD(sctps);
598 		netstack_hold(ns);
599 		connp->conn_netstack = ns;
600 		return (connp);
601 
602 	case IPCL_TCPCONN:
603 		conn_cache = tcp_conn_cache;
604 		break;
605 
606 	case IPCL_UDPCONN:
607 		conn_cache = udp_conn_cache;
608 		break;
609 
610 	case IPCL_RAWIPCONN:
611 		conn_cache = rawip_conn_cache;
612 		break;
613 
614 	case IPCL_RTSCONN:
615 		conn_cache = rts_conn_cache;
616 		break;
617 
618 	case IPCL_IPCCONN:
619 		conn_cache = ip_conn_cache;
620 		break;
621 
622 	default:
623 		connp = NULL;
624 		ASSERT(0);
625 	}
626 
627 	if ((connp = kmem_cache_alloc(conn_cache, sleep)) == NULL)
628 		return (NULL);
629 
630 	connp->conn_ref = 1;
631 	netstack_hold(ns);
632 	connp->conn_netstack = ns;
633 	ipcl_globalhash_insert(connp);
634 	return (connp);
635 }
636 
637 void
638 ipcl_conn_destroy(conn_t *connp)
639 {
640 	mblk_t	*mp;
641 	netstack_t	*ns = connp->conn_netstack;
642 
643 	ASSERT(!MUTEX_HELD(&connp->conn_lock));
644 	ASSERT(connp->conn_ref == 0);
645 	ASSERT(connp->conn_ire_cache == NULL);
646 
647 	DTRACE_PROBE1(conn__destroy, conn_t *, connp);
648 
649 	if (connp->conn_peercred != NULL &&
650 	    connp->conn_peercred != connp->conn_cred)
651 		crfree(connp->conn_peercred);
652 	connp->conn_peercred = NULL;
653 
654 	if (connp->conn_cred != NULL) {
655 		crfree(connp->conn_cred);
656 		connp->conn_cred = NULL;
657 	}
658 
659 	ipcl_globalhash_remove(connp);
660 
661 	/* FIXME: add separate tcp_conn_free()? */
662 	if (connp->conn_flags & IPCL_TCPCONN) {
663 		tcp_t	*tcp = connp->conn_tcp;
664 		tcp_stack_t *tcps;
665 
666 		ASSERT(tcp != NULL);
667 		tcps = tcp->tcp_tcps;
668 		if (tcps != NULL) {
669 			if (connp->conn_latch != NULL) {
670 				IPLATCH_REFRELE(connp->conn_latch, ns);
671 				connp->conn_latch = NULL;
672 			}
673 			if (connp->conn_policy != NULL) {
674 				IPPH_REFRELE(connp->conn_policy, ns);
675 				connp->conn_policy = NULL;
676 			}
677 			tcp->tcp_tcps = NULL;
678 			TCPS_REFRELE(tcps);
679 		}
680 
681 		tcp_free(tcp);
682 		mp = tcp->tcp_timercache;
683 		tcp->tcp_cred = NULL;
684 
685 		if (tcp->tcp_sack_info != NULL) {
686 			bzero(tcp->tcp_sack_info, sizeof (tcp_sack_info_t));
687 			kmem_cache_free(tcp_sack_info_cache,
688 			    tcp->tcp_sack_info);
689 		}
690 		if (tcp->tcp_iphc != NULL) {
691 			if (tcp->tcp_hdr_grown) {
692 				kmem_free(tcp->tcp_iphc, tcp->tcp_iphc_len);
693 			} else {
694 				bzero(tcp->tcp_iphc, tcp->tcp_iphc_len);
695 				kmem_cache_free(tcp_iphc_cache, tcp->tcp_iphc);
696 			}
697 			tcp->tcp_iphc_len = 0;
698 		}
699 		ASSERT(tcp->tcp_iphc_len == 0);
700 
701 		/*
702 		 * tcp_rsrv_mp can be NULL if tcp_get_conn() fails to allocate
703 		 * the mblk.
704 		 */
705 		if (tcp->tcp_rsrv_mp != NULL) {
706 			freeb(tcp->tcp_rsrv_mp);
707 			tcp->tcp_rsrv_mp = NULL;
708 			mutex_destroy(&tcp->tcp_rsrv_mp_lock);
709 		}
710 
711 		ASSERT(connp->conn_latch == NULL);
712 		ASSERT(connp->conn_policy == NULL);
713 
714 		if (ns != NULL) {
715 			ASSERT(tcp->tcp_tcps == NULL);
716 			connp->conn_netstack = NULL;
717 			netstack_rele(ns);
718 		}
719 
720 		ipcl_conn_cleanup(connp);
721 		connp->conn_flags = IPCL_TCPCONN;
722 		bzero(tcp, sizeof (tcp_t));
723 
724 		tcp->tcp_timercache = mp;
725 		tcp->tcp_connp = connp;
726 		kmem_cache_free(tcp_conn_cache, connp);
727 		return;
728 	}
729 	if (connp->conn_latch != NULL) {
730 		IPLATCH_REFRELE(connp->conn_latch, connp->conn_netstack);
731 		connp->conn_latch = NULL;
732 	}
733 	if (connp->conn_policy != NULL) {
734 		IPPH_REFRELE(connp->conn_policy, connp->conn_netstack);
735 		connp->conn_policy = NULL;
736 	}
737 	if (connp->conn_ipsec_opt_mp != NULL) {
738 		freemsg(connp->conn_ipsec_opt_mp);
739 		connp->conn_ipsec_opt_mp = NULL;
740 	}
741 
742 	if (connp->conn_flags & IPCL_SCTPCONN) {
743 		ASSERT(ns != NULL);
744 		sctp_free(connp);
745 		return;
746 	}
747 
748 	if (ns != NULL) {
749 		connp->conn_netstack = NULL;
750 		netstack_rele(ns);
751 	}
752 	ipcl_conn_cleanup(connp);
753 
754 	/* leave conn_priv aka conn_udp, conn_icmp, etc in place. */
755 	if (connp->conn_flags & IPCL_UDPCONN) {
756 		connp->conn_flags = IPCL_UDPCONN;
757 		kmem_cache_free(udp_conn_cache, connp);
758 	} else if (connp->conn_flags & IPCL_RAWIPCONN) {
759 		connp->conn_flags = IPCL_RAWIPCONN;
760 		connp->conn_ulp = IPPROTO_ICMP;
761 		kmem_cache_free(rawip_conn_cache, connp);
762 	} else if (connp->conn_flags & IPCL_RTSCONN) {
763 		connp->conn_flags = IPCL_RTSCONN;
764 		kmem_cache_free(rts_conn_cache, connp);
765 	} else {
766 		connp->conn_flags = IPCL_IPCCONN;
767 		ASSERT(connp->conn_flags & IPCL_IPCCONN);
768 		ASSERT(connp->conn_priv == NULL);
769 		kmem_cache_free(ip_conn_cache, connp);
770 	}
771 }
772 
773 /*
774  * Running in cluster mode - deregister listener information
775  */
776 
777 static void
778 ipcl_conn_unlisten(conn_t *connp)
779 {
780 	ASSERT((connp->conn_flags & IPCL_CL_LISTENER) != 0);
781 	ASSERT(connp->conn_lport != 0);
782 
783 	if (cl_inet_unlisten != NULL) {
784 		sa_family_t	addr_family;
785 		uint8_t		*laddrp;
786 
787 		if (connp->conn_pkt_isv6) {
788 			addr_family = AF_INET6;
789 			laddrp = (uint8_t *)&connp->conn_bound_source_v6;
790 		} else {
791 			addr_family = AF_INET;
792 			laddrp = (uint8_t *)&connp->conn_bound_source;
793 		}
794 		(*cl_inet_unlisten)(IPPROTO_TCP, addr_family, laddrp,
795 		    connp->conn_lport);
796 	}
797 	connp->conn_flags &= ~IPCL_CL_LISTENER;
798 }
799 
800 /*
801  * We set the IPCL_REMOVED flag (instead of clearing the flag indicating
802  * which table the conn belonged to). So for debugging we can see which hash
803  * table this connection was in.
804  */
805 #define	IPCL_HASH_REMOVE(connp)	{					\
806 	connf_t	*connfp = (connp)->conn_fanout;				\
807 	ASSERT(!MUTEX_HELD(&((connp)->conn_lock)));			\
808 	if (connfp != NULL) {						\
809 		IPCL_DEBUG_LVL(4, ("IPCL_HASH_REMOVE: connp %p",	\
810 		    (void *)(connp)));					\
811 		mutex_enter(&connfp->connf_lock);			\
812 		if ((connp)->conn_next != NULL)				\
813 			(connp)->conn_next->conn_prev =			\
814 			    (connp)->conn_prev;				\
815 		if ((connp)->conn_prev != NULL)				\
816 			(connp)->conn_prev->conn_next =			\
817 			    (connp)->conn_next;				\
818 		else							\
819 			connfp->connf_head = (connp)->conn_next;	\
820 		(connp)->conn_fanout = NULL;				\
821 		(connp)->conn_next = NULL;				\
822 		(connp)->conn_prev = NULL;				\
823 		(connp)->conn_flags |= IPCL_REMOVED;			\
824 		if (((connp)->conn_flags & IPCL_CL_LISTENER) != 0)	\
825 			ipcl_conn_unlisten((connp));			\
826 		CONN_DEC_REF((connp));					\
827 		mutex_exit(&connfp->connf_lock);			\
828 	}								\
829 }
830 
831 void
832 ipcl_hash_remove(conn_t *connp)
833 {
834 	IPCL_HASH_REMOVE(connp);
835 }
836 
837 /*
838  * The whole purpose of this function is allow removal of
839  * a conn_t from the connected hash for timewait reclaim.
840  * This is essentially a TW reclaim fastpath where timewait
841  * collector checks under fanout lock (so no one else can
842  * get access to the conn_t) that refcnt is 2 i.e. one for
843  * TCP and one for the classifier hash list. If ref count
844  * is indeed 2, we can just remove the conn under lock and
845  * avoid cleaning up the conn under squeue. This gives us
846  * improved performance.
847  */
848 void
849 ipcl_hash_remove_locked(conn_t *connp, connf_t	*connfp)
850 {
851 	ASSERT(MUTEX_HELD(&connfp->connf_lock));
852 	ASSERT(MUTEX_HELD(&connp->conn_lock));
853 	ASSERT((connp->conn_flags & IPCL_CL_LISTENER) == 0);
854 
855 	if ((connp)->conn_next != NULL) {
856 		(connp)->conn_next->conn_prev = (connp)->conn_prev;
857 	}
858 	if ((connp)->conn_prev != NULL) {
859 		(connp)->conn_prev->conn_next = (connp)->conn_next;
860 	} else {
861 		connfp->connf_head = (connp)->conn_next;
862 	}
863 	(connp)->conn_fanout = NULL;
864 	(connp)->conn_next = NULL;
865 	(connp)->conn_prev = NULL;
866 	(connp)->conn_flags |= IPCL_REMOVED;
867 	ASSERT((connp)->conn_ref == 2);
868 	(connp)->conn_ref--;
869 }
870 
871 #define	IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp) {		\
872 	ASSERT((connp)->conn_fanout == NULL);				\
873 	ASSERT((connp)->conn_next == NULL);				\
874 	ASSERT((connp)->conn_prev == NULL);				\
875 	if ((connfp)->connf_head != NULL) {				\
876 		(connfp)->connf_head->conn_prev = (connp);		\
877 		(connp)->conn_next = (connfp)->connf_head;		\
878 	}								\
879 	(connp)->conn_fanout = (connfp);				\
880 	(connfp)->connf_head = (connp);					\
881 	(connp)->conn_flags = ((connp)->conn_flags & ~IPCL_REMOVED) |	\
882 	    IPCL_CONNECTED;						\
883 	CONN_INC_REF(connp);						\
884 }
885 
886 #define	IPCL_HASH_INSERT_CONNECTED(connfp, connp) {			\
887 	IPCL_DEBUG_LVL(8, ("IPCL_HASH_INSERT_CONNECTED: connfp %p "	\
888 	    "connp %p", (void *)(connfp), (void *)(connp)));		\
889 	IPCL_HASH_REMOVE((connp));					\
890 	mutex_enter(&(connfp)->connf_lock);				\
891 	IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp);		\
892 	mutex_exit(&(connfp)->connf_lock);				\
893 }
894 
895 #define	IPCL_HASH_INSERT_BOUND(connfp, connp) {				\
896 	conn_t *pconnp = NULL, *nconnp;					\
897 	IPCL_DEBUG_LVL(32, ("IPCL_HASH_INSERT_BOUND: connfp %p "	\
898 	    "connp %p", (void *)connfp, (void *)(connp)));		\
899 	IPCL_HASH_REMOVE((connp));					\
900 	mutex_enter(&(connfp)->connf_lock);				\
901 	nconnp = (connfp)->connf_head;					\
902 	while (nconnp != NULL &&					\
903 	    !_IPCL_V4_MATCH_ANY(nconnp->conn_srcv6)) {			\
904 		pconnp = nconnp;					\
905 		nconnp = nconnp->conn_next;				\
906 	}								\
907 	if (pconnp != NULL) {						\
908 		pconnp->conn_next = (connp);				\
909 		(connp)->conn_prev = pconnp;				\
910 	} else {							\
911 		(connfp)->connf_head = (connp);				\
912 	}								\
913 	if (nconnp != NULL) {						\
914 		(connp)->conn_next = nconnp;				\
915 		nconnp->conn_prev = (connp);				\
916 	}								\
917 	(connp)->conn_fanout = (connfp);				\
918 	(connp)->conn_flags = ((connp)->conn_flags & ~IPCL_REMOVED) |	\
919 	    IPCL_BOUND;							\
920 	CONN_INC_REF(connp);						\
921 	mutex_exit(&(connfp)->connf_lock);				\
922 }
923 
924 #define	IPCL_HASH_INSERT_WILDCARD(connfp, connp) {			\
925 	conn_t **list, *prev, *next;					\
926 	boolean_t isv4mapped =						\
927 	    IN6_IS_ADDR_V4MAPPED(&(connp)->conn_srcv6);			\
928 	IPCL_DEBUG_LVL(32, ("IPCL_HASH_INSERT_WILDCARD: connfp %p "	\
929 	    "connp %p", (void *)(connfp), (void *)(connp)));		\
930 	IPCL_HASH_REMOVE((connp));					\
931 	mutex_enter(&(connfp)->connf_lock);				\
932 	list = &(connfp)->connf_head;					\
933 	prev = NULL;							\
934 	while ((next = *list) != NULL) {				\
935 		if (isv4mapped &&					\
936 		    IN6_IS_ADDR_UNSPECIFIED(&next->conn_srcv6) &&	\
937 		    connp->conn_zoneid == next->conn_zoneid) {		\
938 			(connp)->conn_next = next;			\
939 			if (prev != NULL)				\
940 				prev = next->conn_prev;			\
941 			next->conn_prev = (connp);			\
942 			break;						\
943 		}							\
944 		list = &next->conn_next;				\
945 		prev = next;						\
946 	}								\
947 	(connp)->conn_prev = prev;					\
948 	*list = (connp);						\
949 	(connp)->conn_fanout = (connfp);				\
950 	(connp)->conn_flags = ((connp)->conn_flags & ~IPCL_REMOVED) |	\
951 	    IPCL_BOUND;							\
952 	CONN_INC_REF((connp));						\
953 	mutex_exit(&(connfp)->connf_lock);				\
954 }
955 
956 void
957 ipcl_hash_insert_wildcard(connf_t *connfp, conn_t *connp)
958 {
959 	IPCL_HASH_INSERT_WILDCARD(connfp, connp);
960 }
961 
962 void
963 ipcl_proto_insert(conn_t *connp, uint8_t protocol)
964 {
965 	connf_t	*connfp;
966 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
967 
968 	ASSERT(connp != NULL);
969 	ASSERT(!connp->conn_mac_exempt || protocol == IPPROTO_AH ||
970 	    protocol == IPPROTO_ESP);
971 
972 	connp->conn_ulp = protocol;
973 
974 	/* Insert it in the protocol hash */
975 	connfp = &ipst->ips_ipcl_proto_fanout[protocol];
976 	IPCL_HASH_INSERT_WILDCARD(connfp, connp);
977 }
978 
979 void
980 ipcl_proto_insert_v6(conn_t *connp, uint8_t protocol)
981 {
982 	connf_t	*connfp;
983 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
984 
985 	ASSERT(connp != NULL);
986 	ASSERT(!connp->conn_mac_exempt || protocol == IPPROTO_AH ||
987 	    protocol == IPPROTO_ESP);
988 
989 	connp->conn_ulp = protocol;
990 
991 	/* Insert it in the Bind Hash */
992 	connfp = &ipst->ips_ipcl_proto_fanout_v6[protocol];
993 	IPCL_HASH_INSERT_WILDCARD(connfp, connp);
994 }
995 
996 /*
997  * This function is used only for inserting SCTP raw socket now.
998  * This may change later.
999  *
1000  * Note that only one raw socket can be bound to a port.  The param
1001  * lport is in network byte order.
1002  */
1003 static int
1004 ipcl_sctp_hash_insert(conn_t *connp, in_port_t lport)
1005 {
1006 	connf_t	*connfp;
1007 	conn_t	*oconnp;
1008 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
1009 
1010 	connfp = &ipst->ips_ipcl_raw_fanout[IPCL_RAW_HASH(ntohs(lport), ipst)];
1011 
1012 	/* Check for existing raw socket already bound to the port. */
1013 	mutex_enter(&connfp->connf_lock);
1014 	for (oconnp = connfp->connf_head; oconnp != NULL;
1015 	    oconnp = oconnp->conn_next) {
1016 		if (oconnp->conn_lport == lport &&
1017 		    oconnp->conn_zoneid == connp->conn_zoneid &&
1018 		    oconnp->conn_af_isv6 == connp->conn_af_isv6 &&
1019 		    ((IN6_IS_ADDR_UNSPECIFIED(&connp->conn_srcv6) ||
1020 		    IN6_IS_ADDR_UNSPECIFIED(&oconnp->conn_srcv6) ||
1021 		    IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_srcv6) ||
1022 		    IN6_IS_ADDR_V4MAPPED_ANY(&oconnp->conn_srcv6)) ||
1023 		    IN6_ARE_ADDR_EQUAL(&oconnp->conn_srcv6,
1024 		    &connp->conn_srcv6))) {
1025 			break;
1026 		}
1027 	}
1028 	mutex_exit(&connfp->connf_lock);
1029 	if (oconnp != NULL)
1030 		return (EADDRNOTAVAIL);
1031 
1032 	if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_remv6) ||
1033 	    IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_remv6)) {
1034 		if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_srcv6) ||
1035 		    IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_srcv6)) {
1036 			IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1037 		} else {
1038 			IPCL_HASH_INSERT_BOUND(connfp, connp);
1039 		}
1040 	} else {
1041 		IPCL_HASH_INSERT_CONNECTED(connfp, connp);
1042 	}
1043 	return (0);
1044 }
1045 
1046 /*
1047  * Check for a MAC exemption conflict on a labeled system.  Note that for
1048  * protocols that use port numbers (UDP, TCP, SCTP), we do this check up in the
1049  * transport layer.  This check is for binding all other protocols.
1050  *
1051  * Returns true if there's a conflict.
1052  */
1053 static boolean_t
1054 check_exempt_conflict_v4(conn_t *connp, ip_stack_t *ipst)
1055 {
1056 	connf_t	*connfp;
1057 	conn_t *tconn;
1058 
1059 	connfp = &ipst->ips_ipcl_proto_fanout[connp->conn_ulp];
1060 	mutex_enter(&connfp->connf_lock);
1061 	for (tconn = connfp->connf_head; tconn != NULL;
1062 	    tconn = tconn->conn_next) {
1063 		/* We don't allow v4 fallback for v6 raw socket */
1064 		if (connp->conn_af_isv6 != tconn->conn_af_isv6)
1065 			continue;
1066 		/* If neither is exempt, then there's no conflict */
1067 		if (!connp->conn_mac_exempt && !tconn->conn_mac_exempt)
1068 			continue;
1069 		/* If both are bound to different specific addrs, ok */
1070 		if (connp->conn_src != INADDR_ANY &&
1071 		    tconn->conn_src != INADDR_ANY &&
1072 		    connp->conn_src != tconn->conn_src)
1073 			continue;
1074 		/* These two conflict; fail */
1075 		break;
1076 	}
1077 	mutex_exit(&connfp->connf_lock);
1078 	return (tconn != NULL);
1079 }
1080 
1081 static boolean_t
1082 check_exempt_conflict_v6(conn_t *connp, ip_stack_t *ipst)
1083 {
1084 	connf_t	*connfp;
1085 	conn_t *tconn;
1086 
1087 	connfp = &ipst->ips_ipcl_proto_fanout[connp->conn_ulp];
1088 	mutex_enter(&connfp->connf_lock);
1089 	for (tconn = connfp->connf_head; tconn != NULL;
1090 	    tconn = tconn->conn_next) {
1091 		/* We don't allow v4 fallback for v6 raw socket */
1092 		if (connp->conn_af_isv6 != tconn->conn_af_isv6)
1093 			continue;
1094 		/* If neither is exempt, then there's no conflict */
1095 		if (!connp->conn_mac_exempt && !tconn->conn_mac_exempt)
1096 			continue;
1097 		/* If both are bound to different addrs, ok */
1098 		if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_srcv6) &&
1099 		    !IN6_IS_ADDR_UNSPECIFIED(&tconn->conn_srcv6) &&
1100 		    !IN6_ARE_ADDR_EQUAL(&connp->conn_srcv6, &tconn->conn_srcv6))
1101 			continue;
1102 		/* These two conflict; fail */
1103 		break;
1104 	}
1105 	mutex_exit(&connfp->connf_lock);
1106 	return (tconn != NULL);
1107 }
1108 
1109 /*
1110  * (v4, v6) bind hash insertion routines
1111  */
1112 int
1113 ipcl_bind_insert(conn_t *connp, uint8_t protocol, ipaddr_t src, uint16_t lport)
1114 {
1115 	connf_t	*connfp;
1116 #ifdef	IPCL_DEBUG
1117 	char	buf[INET_NTOA_BUFSIZE];
1118 #endif
1119 	int	ret = 0;
1120 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
1121 
1122 	ASSERT(connp);
1123 
1124 	IPCL_DEBUG_LVL(64, ("ipcl_bind_insert: connp %p, src = %s, "
1125 	    "port = %d\n", (void *)connp, inet_ntoa_r(src, buf), lport));
1126 
1127 	connp->conn_ulp = protocol;
1128 	IN6_IPADDR_TO_V4MAPPED(src, &connp->conn_srcv6);
1129 	connp->conn_lport = lport;
1130 
1131 	switch (protocol) {
1132 	default:
1133 		if (is_system_labeled() &&
1134 		    check_exempt_conflict_v4(connp, ipst))
1135 			return (EADDRINUSE);
1136 		/* FALLTHROUGH */
1137 	case IPPROTO_UDP:
1138 		if (protocol == IPPROTO_UDP) {
1139 			IPCL_DEBUG_LVL(64,
1140 			    ("ipcl_bind_insert: connp %p - udp\n",
1141 			    (void *)connp));
1142 			connfp = &ipst->ips_ipcl_udp_fanout[
1143 			    IPCL_UDP_HASH(lport, ipst)];
1144 		} else {
1145 			IPCL_DEBUG_LVL(64,
1146 			    ("ipcl_bind_insert: connp %p - protocol\n",
1147 			    (void *)connp));
1148 			connfp = &ipst->ips_ipcl_proto_fanout[protocol];
1149 		}
1150 
1151 		if (connp->conn_rem != INADDR_ANY) {
1152 			IPCL_HASH_INSERT_CONNECTED(connfp, connp);
1153 		} else if (connp->conn_src != INADDR_ANY) {
1154 			IPCL_HASH_INSERT_BOUND(connfp, connp);
1155 		} else {
1156 			IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1157 		}
1158 		break;
1159 
1160 	case IPPROTO_TCP:
1161 
1162 		/* Insert it in the Bind Hash */
1163 		ASSERT(connp->conn_zoneid != ALL_ZONES);
1164 		connfp = &ipst->ips_ipcl_bind_fanout[
1165 		    IPCL_BIND_HASH(lport, ipst)];
1166 		if (connp->conn_src != INADDR_ANY) {
1167 			IPCL_HASH_INSERT_BOUND(connfp, connp);
1168 		} else {
1169 			IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1170 		}
1171 		if (cl_inet_listen != NULL) {
1172 			ASSERT(!connp->conn_pkt_isv6);
1173 			connp->conn_flags |= IPCL_CL_LISTENER;
1174 			(*cl_inet_listen)(IPPROTO_TCP, AF_INET,
1175 			    (uint8_t *)&connp->conn_bound_source, lport);
1176 		}
1177 		break;
1178 
1179 	case IPPROTO_SCTP:
1180 		ret = ipcl_sctp_hash_insert(connp, lport);
1181 		break;
1182 	}
1183 
1184 	return (ret);
1185 }
1186 
1187 int
1188 ipcl_bind_insert_v6(conn_t *connp, uint8_t protocol, const in6_addr_t *src,
1189     uint16_t lport)
1190 {
1191 	connf_t	*connfp;
1192 	int	ret = 0;
1193 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
1194 
1195 	ASSERT(connp);
1196 
1197 	connp->conn_ulp = protocol;
1198 	connp->conn_srcv6 = *src;
1199 	connp->conn_lport = lport;
1200 
1201 	switch (protocol) {
1202 	default:
1203 		if (is_system_labeled() &&
1204 		    check_exempt_conflict_v6(connp, ipst))
1205 			return (EADDRINUSE);
1206 		/* FALLTHROUGH */
1207 	case IPPROTO_UDP:
1208 		if (protocol == IPPROTO_UDP) {
1209 			IPCL_DEBUG_LVL(128,
1210 			    ("ipcl_bind_insert_v6: connp %p - udp\n",
1211 			    (void *)connp));
1212 			connfp = &ipst->ips_ipcl_udp_fanout[
1213 			    IPCL_UDP_HASH(lport, ipst)];
1214 		} else {
1215 			IPCL_DEBUG_LVL(128,
1216 			    ("ipcl_bind_insert_v6: connp %p - protocol\n",
1217 			    (void *)connp));
1218 			connfp = &ipst->ips_ipcl_proto_fanout_v6[protocol];
1219 		}
1220 
1221 		if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_remv6)) {
1222 			IPCL_HASH_INSERT_CONNECTED(connfp, connp);
1223 		} else if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_srcv6)) {
1224 			IPCL_HASH_INSERT_BOUND(connfp, connp);
1225 		} else {
1226 			IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1227 		}
1228 		break;
1229 
1230 	case IPPROTO_TCP:
1231 		/* XXX - Need a separate table for IN6_IS_ADDR_UNSPECIFIED? */
1232 
1233 		/* Insert it in the Bind Hash */
1234 		ASSERT(connp->conn_zoneid != ALL_ZONES);
1235 		connfp = &ipst->ips_ipcl_bind_fanout[
1236 		    IPCL_BIND_HASH(lport, ipst)];
1237 		if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_srcv6)) {
1238 			IPCL_HASH_INSERT_BOUND(connfp, connp);
1239 		} else {
1240 			IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1241 		}
1242 		if (cl_inet_listen != NULL) {
1243 			sa_family_t	addr_family;
1244 			uint8_t		*laddrp;
1245 
1246 			if (connp->conn_pkt_isv6) {
1247 				addr_family = AF_INET6;
1248 				laddrp =
1249 				    (uint8_t *)&connp->conn_bound_source_v6;
1250 			} else {
1251 				addr_family = AF_INET;
1252 				laddrp = (uint8_t *)&connp->conn_bound_source;
1253 			}
1254 			connp->conn_flags |= IPCL_CL_LISTENER;
1255 			(*cl_inet_listen)(IPPROTO_TCP, addr_family, laddrp,
1256 			    lport);
1257 		}
1258 		break;
1259 
1260 	case IPPROTO_SCTP:
1261 		ret = ipcl_sctp_hash_insert(connp, lport);
1262 		break;
1263 	}
1264 
1265 	return (ret);
1266 }
1267 
1268 /*
1269  * ipcl_conn_hash insertion routines.
1270  */
1271 int
1272 ipcl_conn_insert(conn_t *connp, uint8_t protocol, ipaddr_t src,
1273     ipaddr_t rem, uint32_t ports)
1274 {
1275 	connf_t		*connfp;
1276 	uint16_t	*up;
1277 	conn_t		*tconnp;
1278 #ifdef	IPCL_DEBUG
1279 	char	sbuf[INET_NTOA_BUFSIZE], rbuf[INET_NTOA_BUFSIZE];
1280 #endif
1281 	in_port_t	lport;
1282 	int		ret = 0;
1283 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
1284 
1285 	IPCL_DEBUG_LVL(256, ("ipcl_conn_insert: connp %p, src = %s, "
1286 	    "dst = %s, ports = %x, protocol = %x", (void *)connp,
1287 	    inet_ntoa_r(src, sbuf), inet_ntoa_r(rem, rbuf),
1288 	    ports, protocol));
1289 
1290 	switch (protocol) {
1291 	case IPPROTO_TCP:
1292 		if (!(connp->conn_flags & IPCL_EAGER)) {
1293 			/*
1294 			 * for a eager connection, i.e connections which
1295 			 * have just been created, the initialization is
1296 			 * already done in ip at conn_creation time, so
1297 			 * we can skip the checks here.
1298 			 */
1299 			IPCL_CONN_INIT(connp, protocol, src, rem, ports);
1300 		}
1301 		connfp = &ipst->ips_ipcl_conn_fanout[
1302 		    IPCL_CONN_HASH(connp->conn_rem,
1303 		    connp->conn_ports, ipst)];
1304 		mutex_enter(&connfp->connf_lock);
1305 		for (tconnp = connfp->connf_head; tconnp != NULL;
1306 		    tconnp = tconnp->conn_next) {
1307 			if (IPCL_CONN_MATCH(tconnp, connp->conn_ulp,
1308 			    connp->conn_rem, connp->conn_src,
1309 			    connp->conn_ports)) {
1310 
1311 				/* Already have a conn. bail out */
1312 				mutex_exit(&connfp->connf_lock);
1313 				return (EADDRINUSE);
1314 			}
1315 		}
1316 		if (connp->conn_fanout != NULL) {
1317 			/*
1318 			 * Probably a XTI/TLI application trying to do a
1319 			 * rebind. Let it happen.
1320 			 */
1321 			mutex_exit(&connfp->connf_lock);
1322 			IPCL_HASH_REMOVE(connp);
1323 			mutex_enter(&connfp->connf_lock);
1324 		}
1325 
1326 		ASSERT(connp->conn_recv != NULL);
1327 
1328 		IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp);
1329 		mutex_exit(&connfp->connf_lock);
1330 		break;
1331 
1332 	case IPPROTO_SCTP:
1333 		/*
1334 		 * The raw socket may have already been bound, remove it
1335 		 * from the hash first.
1336 		 */
1337 		IPCL_HASH_REMOVE(connp);
1338 		lport = htons((uint16_t)(ntohl(ports) & 0xFFFF));
1339 		ret = ipcl_sctp_hash_insert(connp, lport);
1340 		break;
1341 
1342 	default:
1343 		/*
1344 		 * Check for conflicts among MAC exempt bindings.  For
1345 		 * transports with port numbers, this is done by the upper
1346 		 * level per-transport binding logic.  For all others, it's
1347 		 * done here.
1348 		 */
1349 		if (is_system_labeled() &&
1350 		    check_exempt_conflict_v4(connp, ipst))
1351 			return (EADDRINUSE);
1352 		/* FALLTHROUGH */
1353 
1354 	case IPPROTO_UDP:
1355 		up = (uint16_t *)&ports;
1356 		IPCL_CONN_INIT(connp, protocol, src, rem, ports);
1357 		if (protocol == IPPROTO_UDP) {
1358 			connfp = &ipst->ips_ipcl_udp_fanout[
1359 			    IPCL_UDP_HASH(up[1], ipst)];
1360 		} else {
1361 			connfp = &ipst->ips_ipcl_proto_fanout[protocol];
1362 		}
1363 
1364 		if (connp->conn_rem != INADDR_ANY) {
1365 			IPCL_HASH_INSERT_CONNECTED(connfp, connp);
1366 		} else if (connp->conn_src != INADDR_ANY) {
1367 			IPCL_HASH_INSERT_BOUND(connfp, connp);
1368 		} else {
1369 			IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1370 		}
1371 		break;
1372 	}
1373 
1374 	return (ret);
1375 }
1376 
1377 int
1378 ipcl_conn_insert_v6(conn_t *connp, uint8_t protocol, const in6_addr_t *src,
1379     const in6_addr_t *rem, uint32_t ports, uint_t ifindex)
1380 {
1381 	connf_t		*connfp;
1382 	uint16_t	*up;
1383 	conn_t		*tconnp;
1384 	in_port_t	lport;
1385 	int		ret = 0;
1386 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
1387 
1388 	switch (protocol) {
1389 	case IPPROTO_TCP:
1390 		/* Just need to insert a conn struct */
1391 		if (!(connp->conn_flags & IPCL_EAGER)) {
1392 			IPCL_CONN_INIT_V6(connp, protocol, *src, *rem, ports);
1393 		}
1394 		connfp = &ipst->ips_ipcl_conn_fanout[
1395 		    IPCL_CONN_HASH_V6(connp->conn_remv6, connp->conn_ports,
1396 		    ipst)];
1397 		mutex_enter(&connfp->connf_lock);
1398 		for (tconnp = connfp->connf_head; tconnp != NULL;
1399 		    tconnp = tconnp->conn_next) {
1400 			if (IPCL_CONN_MATCH_V6(tconnp, connp->conn_ulp,
1401 			    connp->conn_remv6, connp->conn_srcv6,
1402 			    connp->conn_ports) &&
1403 			    (tconnp->conn_tcp->tcp_bound_if == 0 ||
1404 			    tconnp->conn_tcp->tcp_bound_if == ifindex)) {
1405 				/* Already have a conn. bail out */
1406 				mutex_exit(&connfp->connf_lock);
1407 				return (EADDRINUSE);
1408 			}
1409 		}
1410 		if (connp->conn_fanout != NULL) {
1411 			/*
1412 			 * Probably a XTI/TLI application trying to do a
1413 			 * rebind. Let it happen.
1414 			 */
1415 			mutex_exit(&connfp->connf_lock);
1416 			IPCL_HASH_REMOVE(connp);
1417 			mutex_enter(&connfp->connf_lock);
1418 		}
1419 		IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp);
1420 		mutex_exit(&connfp->connf_lock);
1421 		break;
1422 
1423 	case IPPROTO_SCTP:
1424 		IPCL_HASH_REMOVE(connp);
1425 		lport = htons((uint16_t)(ntohl(ports) & 0xFFFF));
1426 		ret = ipcl_sctp_hash_insert(connp, lport);
1427 		break;
1428 
1429 	default:
1430 		if (is_system_labeled() &&
1431 		    check_exempt_conflict_v6(connp, ipst))
1432 			return (EADDRINUSE);
1433 		/* FALLTHROUGH */
1434 	case IPPROTO_UDP:
1435 		up = (uint16_t *)&ports;
1436 		IPCL_CONN_INIT_V6(connp, protocol, *src, *rem, ports);
1437 		if (protocol == IPPROTO_UDP) {
1438 			connfp = &ipst->ips_ipcl_udp_fanout[
1439 			    IPCL_UDP_HASH(up[1], ipst)];
1440 		} else {
1441 			connfp = &ipst->ips_ipcl_proto_fanout_v6[protocol];
1442 		}
1443 
1444 		if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_remv6)) {
1445 			IPCL_HASH_INSERT_CONNECTED(connfp, connp);
1446 		} else if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_srcv6)) {
1447 			IPCL_HASH_INSERT_BOUND(connfp, connp);
1448 		} else {
1449 			IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1450 		}
1451 		break;
1452 	}
1453 
1454 	return (ret);
1455 }
1456 
1457 /*
1458  * v4 packet classifying function. looks up the fanout table to
1459  * find the conn, the packet belongs to. returns the conn with
1460  * the reference held, null otherwise.
1461  *
1462  * If zoneid is ALL_ZONES, then the search rules described in the "Connection
1463  * Lookup" comment block are applied.  Labels are also checked as described
1464  * above.  If the packet is from the inside (looped back), and is from the same
1465  * zone, then label checks are omitted.
1466  */
1467 conn_t *
1468 ipcl_classify_v4(mblk_t *mp, uint8_t protocol, uint_t hdr_len, zoneid_t zoneid,
1469     ip_stack_t *ipst)
1470 {
1471 	ipha_t	*ipha;
1472 	connf_t	*connfp, *bind_connfp;
1473 	uint16_t lport;
1474 	uint16_t fport;
1475 	uint32_t ports;
1476 	conn_t	*connp;
1477 	uint16_t  *up;
1478 	boolean_t shared_addr;
1479 	boolean_t unlabeled;
1480 
1481 	ipha = (ipha_t *)mp->b_rptr;
1482 	up = (uint16_t *)((uchar_t *)ipha + hdr_len + TCP_PORTS_OFFSET);
1483 
1484 	switch (protocol) {
1485 	case IPPROTO_TCP:
1486 		ports = *(uint32_t *)up;
1487 		connfp =
1488 		    &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH(ipha->ipha_src,
1489 		    ports, ipst)];
1490 		mutex_enter(&connfp->connf_lock);
1491 		for (connp = connfp->connf_head; connp != NULL;
1492 		    connp = connp->conn_next) {
1493 			if (IPCL_CONN_MATCH(connp, protocol,
1494 			    ipha->ipha_src, ipha->ipha_dst, ports))
1495 				break;
1496 		}
1497 
1498 		if (connp != NULL) {
1499 			/*
1500 			 * We have a fully-bound TCP connection.
1501 			 *
1502 			 * For labeled systems, there's no need to check the
1503 			 * label here.  It's known to be good as we checked
1504 			 * before allowing the connection to become bound.
1505 			 */
1506 			CONN_INC_REF(connp);
1507 			mutex_exit(&connfp->connf_lock);
1508 			return (connp);
1509 		}
1510 
1511 		mutex_exit(&connfp->connf_lock);
1512 
1513 		lport = up[1];
1514 		unlabeled = B_FALSE;
1515 		/* Cred cannot be null on IPv4 */
1516 		if (is_system_labeled())
1517 			unlabeled = (crgetlabel(DB_CRED(mp))->tsl_flags &
1518 			    TSLF_UNLABELED) != 0;
1519 		shared_addr = (zoneid == ALL_ZONES);
1520 		if (shared_addr) {
1521 			/*
1522 			 * No need to handle exclusive-stack zones since
1523 			 * ALL_ZONES only applies to the shared stack.
1524 			 */
1525 			zoneid = tsol_mlp_findzone(protocol, lport);
1526 			/*
1527 			 * If no shared MLP is found, tsol_mlp_findzone returns
1528 			 * ALL_ZONES.  In that case, we assume it's SLP, and
1529 			 * search for the zone based on the packet label.
1530 			 *
1531 			 * If there is such a zone, we prefer to find a
1532 			 * connection in it.  Otherwise, we look for a
1533 			 * MAC-exempt connection in any zone whose label
1534 			 * dominates the default label on the packet.
1535 			 */
1536 			if (zoneid == ALL_ZONES)
1537 				zoneid = tsol_packet_to_zoneid(mp);
1538 			else
1539 				unlabeled = B_FALSE;
1540 		}
1541 
1542 		bind_connfp =
1543 		    &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)];
1544 		mutex_enter(&bind_connfp->connf_lock);
1545 		for (connp = bind_connfp->connf_head; connp != NULL;
1546 		    connp = connp->conn_next) {
1547 			if (IPCL_BIND_MATCH(connp, protocol, ipha->ipha_dst,
1548 			    lport) && (IPCL_ZONE_MATCH(connp, zoneid) ||
1549 			    (unlabeled && connp->conn_mac_exempt)))
1550 				break;
1551 		}
1552 
1553 		/*
1554 		 * If the matching connection is SLP on a private address, then
1555 		 * the label on the packet must match the local zone's label.
1556 		 * Otherwise, it must be in the label range defined by tnrh.
1557 		 * This is ensured by tsol_receive_label.
1558 		 */
1559 		if (connp != NULL && is_system_labeled() &&
1560 		    !tsol_receive_local(mp, &ipha->ipha_dst, IPV4_VERSION,
1561 		    shared_addr, connp)) {
1562 				DTRACE_PROBE3(
1563 				    tx__ip__log__info__classify__tcp,
1564 				    char *,
1565 				    "connp(1) could not receive mp(2)",
1566 				    conn_t *, connp, mblk_t *, mp);
1567 			connp = NULL;
1568 		}
1569 
1570 		if (connp != NULL) {
1571 			/* Have a listener at least */
1572 			CONN_INC_REF(connp);
1573 			mutex_exit(&bind_connfp->connf_lock);
1574 			return (connp);
1575 		}
1576 
1577 		mutex_exit(&bind_connfp->connf_lock);
1578 
1579 		IPCL_DEBUG_LVL(512,
1580 		    ("ipcl_classify: couldn't classify mp = %p\n",
1581 		    (void *)mp));
1582 		break;
1583 
1584 	case IPPROTO_UDP:
1585 		lport = up[1];
1586 		unlabeled = B_FALSE;
1587 		/* Cred cannot be null on IPv4 */
1588 		if (is_system_labeled())
1589 			unlabeled = (crgetlabel(DB_CRED(mp))->tsl_flags &
1590 			    TSLF_UNLABELED) != 0;
1591 		shared_addr = (zoneid == ALL_ZONES);
1592 		if (shared_addr) {
1593 			/*
1594 			 * No need to handle exclusive-stack zones since
1595 			 * ALL_ZONES only applies to the shared stack.
1596 			 */
1597 			zoneid = tsol_mlp_findzone(protocol, lport);
1598 			/*
1599 			 * If no shared MLP is found, tsol_mlp_findzone returns
1600 			 * ALL_ZONES.  In that case, we assume it's SLP, and
1601 			 * search for the zone based on the packet label.
1602 			 *
1603 			 * If there is such a zone, we prefer to find a
1604 			 * connection in it.  Otherwise, we look for a
1605 			 * MAC-exempt connection in any zone whose label
1606 			 * dominates the default label on the packet.
1607 			 */
1608 			if (zoneid == ALL_ZONES)
1609 				zoneid = tsol_packet_to_zoneid(mp);
1610 			else
1611 				unlabeled = B_FALSE;
1612 		}
1613 		fport = up[0];
1614 		IPCL_DEBUG_LVL(512, ("ipcl_udp_classify %x %x", lport, fport));
1615 		connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
1616 		mutex_enter(&connfp->connf_lock);
1617 		for (connp = connfp->connf_head; connp != NULL;
1618 		    connp = connp->conn_next) {
1619 			if (IPCL_UDP_MATCH(connp, lport, ipha->ipha_dst,
1620 			    fport, ipha->ipha_src) &&
1621 			    (IPCL_ZONE_MATCH(connp, zoneid) ||
1622 			    (unlabeled && connp->conn_mac_exempt)))
1623 				break;
1624 		}
1625 
1626 		if (connp != NULL && is_system_labeled() &&
1627 		    !tsol_receive_local(mp, &ipha->ipha_dst, IPV4_VERSION,
1628 		    shared_addr, connp)) {
1629 			DTRACE_PROBE3(tx__ip__log__info__classify__udp,
1630 			    char *, "connp(1) could not receive mp(2)",
1631 			    conn_t *, connp, mblk_t *, mp);
1632 			connp = NULL;
1633 		}
1634 
1635 		if (connp != NULL) {
1636 			CONN_INC_REF(connp);
1637 			mutex_exit(&connfp->connf_lock);
1638 			return (connp);
1639 		}
1640 
1641 		/*
1642 		 * We shouldn't come here for multicast/broadcast packets
1643 		 */
1644 		mutex_exit(&connfp->connf_lock);
1645 		IPCL_DEBUG_LVL(512,
1646 		    ("ipcl_classify: cant find udp conn_t for ports : %x %x",
1647 		    lport, fport));
1648 		break;
1649 	}
1650 
1651 	return (NULL);
1652 }
1653 
1654 conn_t *
1655 ipcl_classify_v6(mblk_t *mp, uint8_t protocol, uint_t hdr_len, zoneid_t zoneid,
1656     ip_stack_t *ipst)
1657 {
1658 	ip6_t		*ip6h;
1659 	connf_t		*connfp, *bind_connfp;
1660 	uint16_t	lport;
1661 	uint16_t	fport;
1662 	tcph_t		*tcph;
1663 	uint32_t	ports;
1664 	conn_t		*connp;
1665 	uint16_t	*up;
1666 	boolean_t	shared_addr;
1667 	boolean_t	unlabeled;
1668 
1669 	ip6h = (ip6_t *)mp->b_rptr;
1670 
1671 	switch (protocol) {
1672 	case IPPROTO_TCP:
1673 		tcph = (tcph_t *)&mp->b_rptr[hdr_len];
1674 		up = (uint16_t *)tcph->th_lport;
1675 		ports = *(uint32_t *)up;
1676 
1677 		connfp =
1678 		    &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH_V6(ip6h->ip6_src,
1679 		    ports, ipst)];
1680 		mutex_enter(&connfp->connf_lock);
1681 		for (connp = connfp->connf_head; connp != NULL;
1682 		    connp = connp->conn_next) {
1683 			if (IPCL_CONN_MATCH_V6(connp, protocol,
1684 			    ip6h->ip6_src, ip6h->ip6_dst, ports))
1685 				break;
1686 		}
1687 
1688 		if (connp != NULL) {
1689 			/*
1690 			 * We have a fully-bound TCP connection.
1691 			 *
1692 			 * For labeled systems, there's no need to check the
1693 			 * label here.  It's known to be good as we checked
1694 			 * before allowing the connection to become bound.
1695 			 */
1696 			CONN_INC_REF(connp);
1697 			mutex_exit(&connfp->connf_lock);
1698 			return (connp);
1699 		}
1700 
1701 		mutex_exit(&connfp->connf_lock);
1702 
1703 		lport = up[1];
1704 		unlabeled = B_FALSE;
1705 		/* Cred can be null on IPv6 */
1706 		if (is_system_labeled()) {
1707 			cred_t *cr = DB_CRED(mp);
1708 
1709 			unlabeled = (cr != NULL &&
1710 			    crgetlabel(cr)->tsl_flags & TSLF_UNLABELED) != 0;
1711 		}
1712 		shared_addr = (zoneid == ALL_ZONES);
1713 		if (shared_addr) {
1714 			/*
1715 			 * No need to handle exclusive-stack zones since
1716 			 * ALL_ZONES only applies to the shared stack.
1717 			 */
1718 			zoneid = tsol_mlp_findzone(protocol, lport);
1719 			/*
1720 			 * If no shared MLP is found, tsol_mlp_findzone returns
1721 			 * ALL_ZONES.  In that case, we assume it's SLP, and
1722 			 * search for the zone based on the packet label.
1723 			 *
1724 			 * If there is such a zone, we prefer to find a
1725 			 * connection in it.  Otherwise, we look for a
1726 			 * MAC-exempt connection in any zone whose label
1727 			 * dominates the default label on the packet.
1728 			 */
1729 			if (zoneid == ALL_ZONES)
1730 				zoneid = tsol_packet_to_zoneid(mp);
1731 			else
1732 				unlabeled = B_FALSE;
1733 		}
1734 
1735 		bind_connfp =
1736 		    &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)];
1737 		mutex_enter(&bind_connfp->connf_lock);
1738 		for (connp = bind_connfp->connf_head; connp != NULL;
1739 		    connp = connp->conn_next) {
1740 			if (IPCL_BIND_MATCH_V6(connp, protocol,
1741 			    ip6h->ip6_dst, lport) &&
1742 			    (IPCL_ZONE_MATCH(connp, zoneid) ||
1743 			    (unlabeled && connp->conn_mac_exempt)))
1744 				break;
1745 		}
1746 
1747 		if (connp != NULL && is_system_labeled() &&
1748 		    !tsol_receive_local(mp, &ip6h->ip6_dst, IPV6_VERSION,
1749 		    shared_addr, connp)) {
1750 			DTRACE_PROBE3(tx__ip__log__info__classify__tcp6,
1751 			    char *, "connp(1) could not receive mp(2)",
1752 			    conn_t *, connp, mblk_t *, mp);
1753 			connp = NULL;
1754 		}
1755 
1756 		if (connp != NULL) {
1757 			/* Have a listner at least */
1758 			CONN_INC_REF(connp);
1759 			mutex_exit(&bind_connfp->connf_lock);
1760 			IPCL_DEBUG_LVL(512,
1761 			    ("ipcl_classify_v6: found listner "
1762 			    "connp = %p\n", (void *)connp));
1763 
1764 			return (connp);
1765 		}
1766 
1767 		mutex_exit(&bind_connfp->connf_lock);
1768 
1769 		IPCL_DEBUG_LVL(512,
1770 		    ("ipcl_classify_v6: couldn't classify mp = %p\n",
1771 		    (void *)mp));
1772 		break;
1773 
1774 	case IPPROTO_UDP:
1775 		up = (uint16_t *)&mp->b_rptr[hdr_len];
1776 		lport = up[1];
1777 		unlabeled = B_FALSE;
1778 		/* Cred can be null on IPv6 */
1779 		if (is_system_labeled()) {
1780 			cred_t *cr = DB_CRED(mp);
1781 
1782 			unlabeled = (cr != NULL &&
1783 			    crgetlabel(cr)->tsl_flags & TSLF_UNLABELED) != 0;
1784 		}
1785 		shared_addr = (zoneid == ALL_ZONES);
1786 		if (shared_addr) {
1787 			/*
1788 			 * No need to handle exclusive-stack zones since
1789 			 * ALL_ZONES only applies to the shared stack.
1790 			 */
1791 			zoneid = tsol_mlp_findzone(protocol, lport);
1792 			/*
1793 			 * If no shared MLP is found, tsol_mlp_findzone returns
1794 			 * ALL_ZONES.  In that case, we assume it's SLP, and
1795 			 * search for the zone based on the packet label.
1796 			 *
1797 			 * If there is such a zone, we prefer to find a
1798 			 * connection in it.  Otherwise, we look for a
1799 			 * MAC-exempt connection in any zone whose label
1800 			 * dominates the default label on the packet.
1801 			 */
1802 			if (zoneid == ALL_ZONES)
1803 				zoneid = tsol_packet_to_zoneid(mp);
1804 			else
1805 				unlabeled = B_FALSE;
1806 		}
1807 
1808 		fport = up[0];
1809 		IPCL_DEBUG_LVL(512, ("ipcl_udp_classify_v6 %x %x", lport,
1810 		    fport));
1811 		connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
1812 		mutex_enter(&connfp->connf_lock);
1813 		for (connp = connfp->connf_head; connp != NULL;
1814 		    connp = connp->conn_next) {
1815 			if (IPCL_UDP_MATCH_V6(connp, lport, ip6h->ip6_dst,
1816 			    fport, ip6h->ip6_src) &&
1817 			    (IPCL_ZONE_MATCH(connp, zoneid) ||
1818 			    (unlabeled && connp->conn_mac_exempt)))
1819 				break;
1820 		}
1821 
1822 		if (connp != NULL && is_system_labeled() &&
1823 		    !tsol_receive_local(mp, &ip6h->ip6_dst, IPV6_VERSION,
1824 		    shared_addr, connp)) {
1825 			DTRACE_PROBE3(tx__ip__log__info__classify__udp6,
1826 			    char *, "connp(1) could not receive mp(2)",
1827 			    conn_t *, connp, mblk_t *, mp);
1828 			connp = NULL;
1829 		}
1830 
1831 		if (connp != NULL) {
1832 			CONN_INC_REF(connp);
1833 			mutex_exit(&connfp->connf_lock);
1834 			return (connp);
1835 		}
1836 
1837 		/*
1838 		 * We shouldn't come here for multicast/broadcast packets
1839 		 */
1840 		mutex_exit(&connfp->connf_lock);
1841 		IPCL_DEBUG_LVL(512,
1842 		    ("ipcl_classify_v6: cant find udp conn_t for ports : %x %x",
1843 		    lport, fport));
1844 		break;
1845 	}
1846 
1847 	return (NULL);
1848 }
1849 
1850 /*
1851  * wrapper around ipcl_classify_(v4,v6) routines.
1852  */
1853 conn_t *
1854 ipcl_classify(mblk_t *mp, zoneid_t zoneid, ip_stack_t *ipst)
1855 {
1856 	uint16_t	hdr_len;
1857 	ipha_t		*ipha;
1858 	uint8_t		*nexthdrp;
1859 
1860 	if (MBLKL(mp) < sizeof (ipha_t))
1861 		return (NULL);
1862 
1863 	switch (IPH_HDR_VERSION(mp->b_rptr)) {
1864 	case IPV4_VERSION:
1865 		ipha = (ipha_t *)mp->b_rptr;
1866 		hdr_len = IPH_HDR_LENGTH(ipha);
1867 		return (ipcl_classify_v4(mp, ipha->ipha_protocol, hdr_len,
1868 		    zoneid, ipst));
1869 	case IPV6_VERSION:
1870 		if (!ip_hdr_length_nexthdr_v6(mp, (ip6_t *)mp->b_rptr,
1871 		    &hdr_len, &nexthdrp))
1872 			return (NULL);
1873 
1874 		return (ipcl_classify_v6(mp, *nexthdrp, hdr_len, zoneid, ipst));
1875 	}
1876 
1877 	return (NULL);
1878 }
1879 
1880 conn_t *
1881 ipcl_classify_raw(mblk_t *mp, uint8_t protocol, zoneid_t zoneid,
1882     uint32_t ports, ipha_t *hdr, ip_stack_t *ipst)
1883 {
1884 	connf_t		*connfp;
1885 	conn_t		*connp;
1886 	in_port_t	lport;
1887 	int		af;
1888 	boolean_t	shared_addr;
1889 	boolean_t	unlabeled;
1890 	const void	*dst;
1891 
1892 	lport = ((uint16_t *)&ports)[1];
1893 
1894 	unlabeled = B_FALSE;
1895 	/* Cred can be null on IPv6 */
1896 	if (is_system_labeled()) {
1897 		cred_t *cr = DB_CRED(mp);
1898 
1899 		unlabeled = (cr != NULL &&
1900 		    crgetlabel(cr)->tsl_flags & TSLF_UNLABELED) != 0;
1901 	}
1902 	shared_addr = (zoneid == ALL_ZONES);
1903 	if (shared_addr) {
1904 		/*
1905 		 * No need to handle exclusive-stack zones since ALL_ZONES
1906 		 * only applies to the shared stack.
1907 		 */
1908 		zoneid = tsol_mlp_findzone(protocol, lport);
1909 		/*
1910 		 * If no shared MLP is found, tsol_mlp_findzone returns
1911 		 * ALL_ZONES.  In that case, we assume it's SLP, and search for
1912 		 * the zone based on the packet label.
1913 		 *
1914 		 * If there is such a zone, we prefer to find a connection in
1915 		 * it.  Otherwise, we look for a MAC-exempt connection in any
1916 		 * zone whose label dominates the default label on the packet.
1917 		 */
1918 		if (zoneid == ALL_ZONES)
1919 			zoneid = tsol_packet_to_zoneid(mp);
1920 		else
1921 			unlabeled = B_FALSE;
1922 	}
1923 
1924 	af = IPH_HDR_VERSION(hdr);
1925 	dst = af == IPV4_VERSION ? (const void *)&hdr->ipha_dst :
1926 	    (const void *)&((ip6_t *)hdr)->ip6_dst;
1927 	connfp = &ipst->ips_ipcl_raw_fanout[IPCL_RAW_HASH(ntohs(lport), ipst)];
1928 
1929 	mutex_enter(&connfp->connf_lock);
1930 	for (connp = connfp->connf_head; connp != NULL;
1931 	    connp = connp->conn_next) {
1932 		/* We don't allow v4 fallback for v6 raw socket. */
1933 		if (af == (connp->conn_af_isv6 ? IPV4_VERSION :
1934 		    IPV6_VERSION))
1935 			continue;
1936 		if (connp->conn_fully_bound) {
1937 			if (af == IPV4_VERSION) {
1938 				if (!IPCL_CONN_MATCH(connp, protocol,
1939 				    hdr->ipha_src, hdr->ipha_dst, ports))
1940 					continue;
1941 			} else {
1942 				if (!IPCL_CONN_MATCH_V6(connp, protocol,
1943 				    ((ip6_t *)hdr)->ip6_src,
1944 				    ((ip6_t *)hdr)->ip6_dst, ports))
1945 					continue;
1946 			}
1947 		} else {
1948 			if (af == IPV4_VERSION) {
1949 				if (!IPCL_BIND_MATCH(connp, protocol,
1950 				    hdr->ipha_dst, lport))
1951 					continue;
1952 			} else {
1953 				if (!IPCL_BIND_MATCH_V6(connp, protocol,
1954 				    ((ip6_t *)hdr)->ip6_dst, lport))
1955 					continue;
1956 			}
1957 		}
1958 
1959 		if (IPCL_ZONE_MATCH(connp, zoneid) ||
1960 		    (unlabeled && connp->conn_mac_exempt))
1961 			break;
1962 	}
1963 	/*
1964 	 * If the connection is fully-bound and connection-oriented (TCP or
1965 	 * SCTP), then we've already validated the remote system's label.
1966 	 * There's no need to do it again for every packet.
1967 	 */
1968 	if (connp != NULL && is_system_labeled() && (!connp->conn_fully_bound ||
1969 	    !(connp->conn_flags & (IPCL_TCP|IPCL_SCTPCONN))) &&
1970 	    !tsol_receive_local(mp, dst, af, shared_addr, connp)) {
1971 		DTRACE_PROBE3(tx__ip__log__info__classify__rawip,
1972 		    char *, "connp(1) could not receive mp(2)",
1973 		    conn_t *, connp, mblk_t *, mp);
1974 		connp = NULL;
1975 	}
1976 
1977 	if (connp != NULL)
1978 		goto found;
1979 	mutex_exit(&connfp->connf_lock);
1980 
1981 	/* Try to look for a wildcard match. */
1982 	connfp = &ipst->ips_ipcl_raw_fanout[IPCL_RAW_HASH(0, ipst)];
1983 	mutex_enter(&connfp->connf_lock);
1984 	for (connp = connfp->connf_head; connp != NULL;
1985 	    connp = connp->conn_next) {
1986 		/* We don't allow v4 fallback for v6 raw socket. */
1987 		if ((af == (connp->conn_af_isv6 ? IPV4_VERSION :
1988 		    IPV6_VERSION)) || !IPCL_ZONE_MATCH(connp, zoneid)) {
1989 			continue;
1990 		}
1991 		if (af == IPV4_VERSION) {
1992 			if (IPCL_RAW_MATCH(connp, protocol, hdr->ipha_dst))
1993 				break;
1994 		} else {
1995 			if (IPCL_RAW_MATCH_V6(connp, protocol,
1996 			    ((ip6_t *)hdr)->ip6_dst)) {
1997 				break;
1998 			}
1999 		}
2000 	}
2001 
2002 	if (connp != NULL)
2003 		goto found;
2004 
2005 	mutex_exit(&connfp->connf_lock);
2006 	return (NULL);
2007 
2008 found:
2009 	ASSERT(connp != NULL);
2010 	CONN_INC_REF(connp);
2011 	mutex_exit(&connfp->connf_lock);
2012 	return (connp);
2013 }
2014 
2015 /* ARGSUSED */
2016 static int
2017 tcp_conn_constructor(void *buf, void *cdrarg, int kmflags)
2018 {
2019 	itc_t	*itc = (itc_t *)buf;
2020 	conn_t 	*connp = &itc->itc_conn;
2021 	tcp_t	*tcp = (tcp_t *)&itc[1];
2022 
2023 	bzero(connp, sizeof (conn_t));
2024 	bzero(tcp, sizeof (tcp_t));
2025 
2026 	mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL);
2027 	cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL);
2028 	tcp->tcp_timercache = tcp_timermp_alloc(KM_NOSLEEP);
2029 	connp->conn_tcp = tcp;
2030 	connp->conn_flags = IPCL_TCPCONN;
2031 	connp->conn_ulp = IPPROTO_TCP;
2032 	tcp->tcp_connp = connp;
2033 	return (0);
2034 }
2035 
2036 /* ARGSUSED */
2037 static void
2038 tcp_conn_destructor(void *buf, void *cdrarg)
2039 {
2040 	itc_t	*itc = (itc_t *)buf;
2041 	conn_t 	*connp = &itc->itc_conn;
2042 	tcp_t	*tcp = (tcp_t *)&itc[1];
2043 
2044 	ASSERT(connp->conn_flags & IPCL_TCPCONN);
2045 	ASSERT(tcp->tcp_connp == connp);
2046 	ASSERT(connp->conn_tcp == tcp);
2047 	tcp_timermp_free(tcp);
2048 	mutex_destroy(&connp->conn_lock);
2049 	cv_destroy(&connp->conn_cv);
2050 }
2051 
2052 /* ARGSUSED */
2053 static int
2054 ip_conn_constructor(void *buf, void *cdrarg, int kmflags)
2055 {
2056 	itc_t	*itc = (itc_t *)buf;
2057 	conn_t 	*connp = &itc->itc_conn;
2058 
2059 	bzero(connp, sizeof (conn_t));
2060 	mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL);
2061 	cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL);
2062 	connp->conn_flags = IPCL_IPCCONN;
2063 
2064 	return (0);
2065 }
2066 
2067 /* ARGSUSED */
2068 static void
2069 ip_conn_destructor(void *buf, void *cdrarg)
2070 {
2071 	itc_t	*itc = (itc_t *)buf;
2072 	conn_t 	*connp = &itc->itc_conn;
2073 
2074 	ASSERT(connp->conn_flags & IPCL_IPCCONN);
2075 	ASSERT(connp->conn_priv == NULL);
2076 	mutex_destroy(&connp->conn_lock);
2077 	cv_destroy(&connp->conn_cv);
2078 }
2079 
2080 /* ARGSUSED */
2081 static int
2082 udp_conn_constructor(void *buf, void *cdrarg, int kmflags)
2083 {
2084 	itc_t	*itc = (itc_t *)buf;
2085 	conn_t 	*connp = &itc->itc_conn;
2086 	udp_t	*udp = (udp_t *)&itc[1];
2087 
2088 	bzero(connp, sizeof (conn_t));
2089 	bzero(udp, sizeof (udp_t));
2090 
2091 	mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL);
2092 	cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL);
2093 	connp->conn_udp = udp;
2094 	connp->conn_flags = IPCL_UDPCONN;
2095 	connp->conn_ulp = IPPROTO_UDP;
2096 	udp->udp_connp = connp;
2097 	return (0);
2098 }
2099 
2100 /* ARGSUSED */
2101 static void
2102 udp_conn_destructor(void *buf, void *cdrarg)
2103 {
2104 	itc_t	*itc = (itc_t *)buf;
2105 	conn_t 	*connp = &itc->itc_conn;
2106 	udp_t	*udp = (udp_t *)&itc[1];
2107 
2108 	ASSERT(connp->conn_flags & IPCL_UDPCONN);
2109 	ASSERT(udp->udp_connp == connp);
2110 	ASSERT(connp->conn_udp == udp);
2111 	mutex_destroy(&connp->conn_lock);
2112 	cv_destroy(&connp->conn_cv);
2113 }
2114 
2115 /* ARGSUSED */
2116 static int
2117 rawip_conn_constructor(void *buf, void *cdrarg, int kmflags)
2118 {
2119 	itc_t	*itc = (itc_t *)buf;
2120 	conn_t 	*connp = &itc->itc_conn;
2121 	icmp_t	*icmp = (icmp_t *)&itc[1];
2122 
2123 	bzero(connp, sizeof (conn_t));
2124 	bzero(icmp, sizeof (icmp_t));
2125 
2126 	mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL);
2127 	cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL);
2128 	connp->conn_icmp = icmp;
2129 	connp->conn_flags = IPCL_RAWIPCONN;
2130 	connp->conn_ulp = IPPROTO_ICMP;
2131 	icmp->icmp_connp = connp;
2132 	return (0);
2133 }
2134 
2135 /* ARGSUSED */
2136 static void
2137 rawip_conn_destructor(void *buf, void *cdrarg)
2138 {
2139 	itc_t	*itc = (itc_t *)buf;
2140 	conn_t 	*connp = &itc->itc_conn;
2141 	icmp_t	*icmp = (icmp_t *)&itc[1];
2142 
2143 	ASSERT(connp->conn_flags & IPCL_RAWIPCONN);
2144 	ASSERT(icmp->icmp_connp == connp);
2145 	ASSERT(connp->conn_icmp == icmp);
2146 	mutex_destroy(&connp->conn_lock);
2147 	cv_destroy(&connp->conn_cv);
2148 }
2149 
2150 /* ARGSUSED */
2151 static int
2152 rts_conn_constructor(void *buf, void *cdrarg, int kmflags)
2153 {
2154 	itc_t	*itc = (itc_t *)buf;
2155 	conn_t 	*connp = &itc->itc_conn;
2156 	rts_t	*rts = (rts_t *)&itc[1];
2157 
2158 	bzero(connp, sizeof (conn_t));
2159 	bzero(rts, sizeof (rts_t));
2160 
2161 	mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL);
2162 	cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL);
2163 	connp->conn_rts = rts;
2164 	connp->conn_flags = IPCL_RTSCONN;
2165 	rts->rts_connp = connp;
2166 	return (0);
2167 }
2168 
2169 /* ARGSUSED */
2170 static void
2171 rts_conn_destructor(void *buf, void *cdrarg)
2172 {
2173 	itc_t	*itc = (itc_t *)buf;
2174 	conn_t 	*connp = &itc->itc_conn;
2175 	rts_t	*rts = (rts_t *)&itc[1];
2176 
2177 	ASSERT(connp->conn_flags & IPCL_RTSCONN);
2178 	ASSERT(rts->rts_connp == connp);
2179 	ASSERT(connp->conn_rts == rts);
2180 	mutex_destroy(&connp->conn_lock);
2181 	cv_destroy(&connp->conn_cv);
2182 }
2183 
2184 /*
2185  * Called as part of ipcl_conn_destroy to assert and clear any pointers
2186  * in the conn_t.
2187  *
2188  * Below we list all the pointers in the conn_t as a documentation aid.
2189  * The ones that we can not ASSERT to be NULL are #ifdef'ed out.
2190  * If you add any pointers to the conn_t please add an ASSERT here
2191  * and #ifdef it out if it can't be actually asserted to be NULL.
2192  * In any case, we bzero most of the conn_t at the end of the function.
2193  */
2194 void
2195 ipcl_conn_cleanup(conn_t *connp)
2196 {
2197 	ASSERT(connp->conn_ire_cache == NULL);
2198 	ASSERT(connp->conn_latch == NULL);
2199 #ifdef notdef
2200 	/* These are not cleared */
2201 	ASSERT(connp->conn_rq == NULL);
2202 	ASSERT(connp->conn_wq == NULL);
2203 #endif
2204 	ASSERT(connp->conn_cred == NULL);
2205 	ASSERT(connp->conn_g_fanout == NULL);
2206 	ASSERT(connp->conn_g_next == NULL);
2207 	ASSERT(connp->conn_g_prev == NULL);
2208 	ASSERT(connp->conn_policy == NULL);
2209 	ASSERT(connp->conn_fanout == NULL);
2210 	ASSERT(connp->conn_next == NULL);
2211 	ASSERT(connp->conn_prev == NULL);
2212 #ifdef notdef
2213 	/*
2214 	 * The ill and ipif pointers are not cleared before the conn_t
2215 	 * goes away since they do not hold a reference on the ill/ipif.
2216 	 * We should replace these pointers with ifindex/ipaddr_t to
2217 	 * make the code less complex.
2218 	 */
2219 	ASSERT(connp->conn_xmit_if_ill == NULL);
2220 	ASSERT(connp->conn_nofailover_ill == NULL);
2221 	ASSERT(connp->conn_outgoing_ill == NULL);
2222 	ASSERT(connp->conn_incoming_ill == NULL);
2223 	ASSERT(connp->conn_outgoing_pill == NULL);
2224 	ASSERT(connp->conn_multicast_ipif == NULL);
2225 	ASSERT(connp->conn_multicast_ill == NULL);
2226 #endif
2227 	ASSERT(connp->conn_oper_pending_ill == NULL);
2228 	ASSERT(connp->conn_ilg == NULL);
2229 	ASSERT(connp->conn_drain_next == NULL);
2230 	ASSERT(connp->conn_drain_prev == NULL);
2231 #ifdef notdef
2232 	/* conn_idl is not cleared when removed from idl list */
2233 	ASSERT(connp->conn_idl == NULL);
2234 #endif
2235 	ASSERT(connp->conn_ipsec_opt_mp == NULL);
2236 	ASSERT(connp->conn_peercred == NULL);
2237 	ASSERT(connp->conn_netstack == NULL);
2238 
2239 	/* Clear out the conn_t fields that are not preserved */
2240 	bzero(&connp->conn_start_clr,
2241 	    sizeof (conn_t) -
2242 	    ((uchar_t *)&connp->conn_start_clr - (uchar_t *)connp));
2243 
2244 }
2245 
2246 /*
2247  * All conns are inserted in a global multi-list for the benefit of
2248  * walkers. The walk is guaranteed to walk all open conns at the time
2249  * of the start of the walk exactly once. This property is needed to
2250  * achieve some cleanups during unplumb of interfaces. This is achieved
2251  * as follows.
2252  *
2253  * ipcl_conn_create and ipcl_conn_destroy are the only functions that
2254  * call the insert and delete functions below at creation and deletion
2255  * time respectively. The conn never moves or changes its position in this
2256  * multi-list during its lifetime. CONN_CONDEMNED ensures that the refcnt
2257  * won't increase due to walkers, once the conn deletion has started. Note
2258  * that we can't remove the conn from the global list and then wait for
2259  * the refcnt to drop to zero, since walkers would then see a truncated
2260  * list. CONN_INCIPIENT ensures that walkers don't start looking at
2261  * conns until ip_open is ready to make them globally visible.
2262  * The global round robin multi-list locks are held only to get the
2263  * next member/insertion/deletion and contention should be negligible
2264  * if the multi-list is much greater than the number of cpus.
2265  */
2266 void
2267 ipcl_globalhash_insert(conn_t *connp)
2268 {
2269 	int	index;
2270 	struct connf_s	*connfp;
2271 	ip_stack_t	*ipst = connp->conn_netstack->netstack_ip;
2272 
2273 	/*
2274 	 * No need for atomic here. Approximate even distribution
2275 	 * in the global lists is sufficient.
2276 	 */
2277 	ipst->ips_conn_g_index++;
2278 	index = ipst->ips_conn_g_index & (CONN_G_HASH_SIZE - 1);
2279 
2280 	connp->conn_g_prev = NULL;
2281 	/*
2282 	 * Mark as INCIPIENT, so that walkers will ignore this
2283 	 * for now, till ip_open is ready to make it visible globally.
2284 	 */
2285 	connp->conn_state_flags |= CONN_INCIPIENT;
2286 
2287 	connfp = &ipst->ips_ipcl_globalhash_fanout[index];
2288 	/* Insert at the head of the list */
2289 	mutex_enter(&connfp->connf_lock);
2290 	connp->conn_g_next = connfp->connf_head;
2291 	if (connp->conn_g_next != NULL)
2292 		connp->conn_g_next->conn_g_prev = connp;
2293 	connfp->connf_head = connp;
2294 
2295 	/* The fanout bucket this conn points to */
2296 	connp->conn_g_fanout = connfp;
2297 
2298 	mutex_exit(&connfp->connf_lock);
2299 }
2300 
2301 void
2302 ipcl_globalhash_remove(conn_t *connp)
2303 {
2304 	struct connf_s	*connfp;
2305 
2306 	/*
2307 	 * We were never inserted in the global multi list.
2308 	 * IPCL_NONE variety is never inserted in the global multilist
2309 	 * since it is presumed to not need any cleanup and is transient.
2310 	 */
2311 	if (connp->conn_g_fanout == NULL)
2312 		return;
2313 
2314 	connfp = connp->conn_g_fanout;
2315 	mutex_enter(&connfp->connf_lock);
2316 	if (connp->conn_g_prev != NULL)
2317 		connp->conn_g_prev->conn_g_next = connp->conn_g_next;
2318 	else
2319 		connfp->connf_head = connp->conn_g_next;
2320 	if (connp->conn_g_next != NULL)
2321 		connp->conn_g_next->conn_g_prev = connp->conn_g_prev;
2322 	mutex_exit(&connfp->connf_lock);
2323 
2324 	/* Better to stumble on a null pointer than to corrupt memory */
2325 	connp->conn_g_next = NULL;
2326 	connp->conn_g_prev = NULL;
2327 	connp->conn_g_fanout = NULL;
2328 }
2329 
2330 /*
2331  * Walk the list of all conn_t's in the system, calling the function provided
2332  * with the specified argument for each.
2333  * Applies to both IPv4 and IPv6.
2334  *
2335  * IPCs may hold pointers to ipif/ill. To guard against stale pointers
2336  * ipcl_walk() is called to cleanup the conn_t's, typically when an interface is
2337  * unplumbed or removed. New conn_t's that are created while we are walking
2338  * may be missed by this walk, because they are not necessarily inserted
2339  * at the tail of the list. They are new conn_t's and thus don't have any
2340  * stale pointers. The CONN_CLOSING flag ensures that no new reference
2341  * is created to the struct that is going away.
2342  */
2343 void
2344 ipcl_walk(pfv_t func, void *arg, ip_stack_t *ipst)
2345 {
2346 	int	i;
2347 	conn_t	*connp;
2348 	conn_t	*prev_connp;
2349 
2350 	for (i = 0; i < CONN_G_HASH_SIZE; i++) {
2351 		mutex_enter(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock);
2352 		prev_connp = NULL;
2353 		connp = ipst->ips_ipcl_globalhash_fanout[i].connf_head;
2354 		while (connp != NULL) {
2355 			mutex_enter(&connp->conn_lock);
2356 			if (connp->conn_state_flags &
2357 			    (CONN_CONDEMNED | CONN_INCIPIENT)) {
2358 				mutex_exit(&connp->conn_lock);
2359 				connp = connp->conn_g_next;
2360 				continue;
2361 			}
2362 			CONN_INC_REF_LOCKED(connp);
2363 			mutex_exit(&connp->conn_lock);
2364 			mutex_exit(
2365 			    &ipst->ips_ipcl_globalhash_fanout[i].connf_lock);
2366 			(*func)(connp, arg);
2367 			if (prev_connp != NULL)
2368 				CONN_DEC_REF(prev_connp);
2369 			mutex_enter(
2370 			    &ipst->ips_ipcl_globalhash_fanout[i].connf_lock);
2371 			prev_connp = connp;
2372 			connp = connp->conn_g_next;
2373 		}
2374 		mutex_exit(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock);
2375 		if (prev_connp != NULL)
2376 			CONN_DEC_REF(prev_connp);
2377 	}
2378 }
2379 
2380 /*
2381  * Search for a peer TCP/IPv4 loopback conn by doing a reverse lookup on
2382  * the {src, dst, lport, fport} quadruplet.  Returns with conn reference
2383  * held; caller must call CONN_DEC_REF.  Only checks for connected entries
2384  * (peer tcp in ESTABLISHED state).
2385  */
2386 conn_t *
2387 ipcl_conn_tcp_lookup_reversed_ipv4(conn_t *connp, ipha_t *ipha, tcph_t *tcph,
2388     ip_stack_t *ipst)
2389 {
2390 	uint32_t ports;
2391 	uint16_t *pports = (uint16_t *)&ports;
2392 	connf_t	*connfp;
2393 	conn_t	*tconnp;
2394 	boolean_t zone_chk;
2395 
2396 	/*
2397 	 * If either the source of destination address is loopback, then
2398 	 * both endpoints must be in the same Zone.  Otherwise, both of
2399 	 * the addresses are system-wide unique (tcp is in ESTABLISHED
2400 	 * state) and the endpoints may reside in different Zones.
2401 	 */
2402 	zone_chk = (ipha->ipha_src == htonl(INADDR_LOOPBACK) ||
2403 	    ipha->ipha_dst == htonl(INADDR_LOOPBACK));
2404 
2405 	bcopy(tcph->th_fport, &pports[0], sizeof (uint16_t));
2406 	bcopy(tcph->th_lport, &pports[1], sizeof (uint16_t));
2407 
2408 	connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH(ipha->ipha_dst,
2409 	    ports, ipst)];
2410 
2411 	mutex_enter(&connfp->connf_lock);
2412 	for (tconnp = connfp->connf_head; tconnp != NULL;
2413 	    tconnp = tconnp->conn_next) {
2414 
2415 		if (IPCL_CONN_MATCH(tconnp, IPPROTO_TCP,
2416 		    ipha->ipha_dst, ipha->ipha_src, ports) &&
2417 		    tconnp->conn_tcp->tcp_state == TCPS_ESTABLISHED &&
2418 		    (!zone_chk || tconnp->conn_zoneid == connp->conn_zoneid)) {
2419 
2420 			ASSERT(tconnp != connp);
2421 			CONN_INC_REF(tconnp);
2422 			mutex_exit(&connfp->connf_lock);
2423 			return (tconnp);
2424 		}
2425 	}
2426 	mutex_exit(&connfp->connf_lock);
2427 	return (NULL);
2428 }
2429 
2430 /*
2431  * Search for a peer TCP/IPv6 loopback conn by doing a reverse lookup on
2432  * the {src, dst, lport, fport} quadruplet.  Returns with conn reference
2433  * held; caller must call CONN_DEC_REF.  Only checks for connected entries
2434  * (peer tcp in ESTABLISHED state).
2435  */
2436 conn_t *
2437 ipcl_conn_tcp_lookup_reversed_ipv6(conn_t *connp, ip6_t *ip6h, tcph_t *tcph,
2438     ip_stack_t *ipst)
2439 {
2440 	uint32_t ports;
2441 	uint16_t *pports = (uint16_t *)&ports;
2442 	connf_t	*connfp;
2443 	conn_t	*tconnp;
2444 	boolean_t zone_chk;
2445 
2446 	/*
2447 	 * If either the source of destination address is loopback, then
2448 	 * both endpoints must be in the same Zone.  Otherwise, both of
2449 	 * the addresses are system-wide unique (tcp is in ESTABLISHED
2450 	 * state) and the endpoints may reside in different Zones.  We
2451 	 * don't do Zone check for link local address(es) because the
2452 	 * current Zone implementation treats each link local address as
2453 	 * being unique per system node, i.e. they belong to global Zone.
2454 	 */
2455 	zone_chk = (IN6_IS_ADDR_LOOPBACK(&ip6h->ip6_src) ||
2456 	    IN6_IS_ADDR_LOOPBACK(&ip6h->ip6_dst));
2457 
2458 	bcopy(tcph->th_fport, &pports[0], sizeof (uint16_t));
2459 	bcopy(tcph->th_lport, &pports[1], sizeof (uint16_t));
2460 
2461 	connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH_V6(ip6h->ip6_dst,
2462 	    ports, ipst)];
2463 
2464 	mutex_enter(&connfp->connf_lock);
2465 	for (tconnp = connfp->connf_head; tconnp != NULL;
2466 	    tconnp = tconnp->conn_next) {
2467 
2468 		/* We skip tcp_bound_if check here as this is loopback tcp */
2469 		if (IPCL_CONN_MATCH_V6(tconnp, IPPROTO_TCP,
2470 		    ip6h->ip6_dst, ip6h->ip6_src, ports) &&
2471 		    tconnp->conn_tcp->tcp_state == TCPS_ESTABLISHED &&
2472 		    (!zone_chk || tconnp->conn_zoneid == connp->conn_zoneid)) {
2473 
2474 			ASSERT(tconnp != connp);
2475 			CONN_INC_REF(tconnp);
2476 			mutex_exit(&connfp->connf_lock);
2477 			return (tconnp);
2478 		}
2479 	}
2480 	mutex_exit(&connfp->connf_lock);
2481 	return (NULL);
2482 }
2483 
2484 /*
2485  * Find an exact {src, dst, lport, fport} match for a bounced datagram.
2486  * Returns with conn reference held. Caller must call CONN_DEC_REF.
2487  * Only checks for connected entries i.e. no INADDR_ANY checks.
2488  */
2489 conn_t *
2490 ipcl_tcp_lookup_reversed_ipv4(ipha_t *ipha, tcph_t *tcph, int min_state,
2491     ip_stack_t *ipst)
2492 {
2493 	uint32_t ports;
2494 	uint16_t *pports;
2495 	connf_t	*connfp;
2496 	conn_t	*tconnp;
2497 
2498 	pports = (uint16_t *)&ports;
2499 	bcopy(tcph->th_fport, &pports[0], sizeof (uint16_t));
2500 	bcopy(tcph->th_lport, &pports[1], sizeof (uint16_t));
2501 
2502 	connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH(ipha->ipha_dst,
2503 	    ports, ipst)];
2504 
2505 	mutex_enter(&connfp->connf_lock);
2506 	for (tconnp = connfp->connf_head; tconnp != NULL;
2507 	    tconnp = tconnp->conn_next) {
2508 
2509 		if (IPCL_CONN_MATCH(tconnp, IPPROTO_TCP,
2510 		    ipha->ipha_dst, ipha->ipha_src, ports) &&
2511 		    tconnp->conn_tcp->tcp_state >= min_state) {
2512 
2513 			CONN_INC_REF(tconnp);
2514 			mutex_exit(&connfp->connf_lock);
2515 			return (tconnp);
2516 		}
2517 	}
2518 	mutex_exit(&connfp->connf_lock);
2519 	return (NULL);
2520 }
2521 
2522 /*
2523  * Find an exact {src, dst, lport, fport} match for a bounced datagram.
2524  * Returns with conn reference held. Caller must call CONN_DEC_REF.
2525  * Only checks for connected entries i.e. no INADDR_ANY checks.
2526  * Match on ifindex in addition to addresses.
2527  */
2528 conn_t *
2529 ipcl_tcp_lookup_reversed_ipv6(ip6_t *ip6h, tcpha_t *tcpha, int min_state,
2530     uint_t ifindex, ip_stack_t *ipst)
2531 {
2532 	tcp_t	*tcp;
2533 	uint32_t ports;
2534 	uint16_t *pports;
2535 	connf_t	*connfp;
2536 	conn_t	*tconnp;
2537 
2538 	pports = (uint16_t *)&ports;
2539 	pports[0] = tcpha->tha_fport;
2540 	pports[1] = tcpha->tha_lport;
2541 
2542 	connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH_V6(ip6h->ip6_dst,
2543 	    ports, ipst)];
2544 
2545 	mutex_enter(&connfp->connf_lock);
2546 	for (tconnp = connfp->connf_head; tconnp != NULL;
2547 	    tconnp = tconnp->conn_next) {
2548 
2549 		tcp = tconnp->conn_tcp;
2550 		if (IPCL_CONN_MATCH_V6(tconnp, IPPROTO_TCP,
2551 		    ip6h->ip6_dst, ip6h->ip6_src, ports) &&
2552 		    tcp->tcp_state >= min_state &&
2553 		    (tcp->tcp_bound_if == 0 ||
2554 		    tcp->tcp_bound_if == ifindex)) {
2555 
2556 			CONN_INC_REF(tconnp);
2557 			mutex_exit(&connfp->connf_lock);
2558 			return (tconnp);
2559 		}
2560 	}
2561 	mutex_exit(&connfp->connf_lock);
2562 	return (NULL);
2563 }
2564 
2565 /*
2566  * Finds a TCP/IPv4 listening connection; called by tcp_disconnect to locate
2567  * a listener when changing state.
2568  */
2569 conn_t *
2570 ipcl_lookup_listener_v4(uint16_t lport, ipaddr_t laddr, zoneid_t zoneid,
2571     ip_stack_t *ipst)
2572 {
2573 	connf_t		*bind_connfp;
2574 	conn_t		*connp;
2575 	tcp_t		*tcp;
2576 
2577 	/*
2578 	 * Avoid false matches for packets sent to an IP destination of
2579 	 * all zeros.
2580 	 */
2581 	if (laddr == 0)
2582 		return (NULL);
2583 
2584 	ASSERT(zoneid != ALL_ZONES);
2585 
2586 	bind_connfp = &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)];
2587 	mutex_enter(&bind_connfp->connf_lock);
2588 	for (connp = bind_connfp->connf_head; connp != NULL;
2589 	    connp = connp->conn_next) {
2590 		tcp = connp->conn_tcp;
2591 		if (IPCL_BIND_MATCH(connp, IPPROTO_TCP, laddr, lport) &&
2592 		    IPCL_ZONE_MATCH(connp, zoneid) &&
2593 		    (tcp->tcp_listener == NULL)) {
2594 			CONN_INC_REF(connp);
2595 			mutex_exit(&bind_connfp->connf_lock);
2596 			return (connp);
2597 		}
2598 	}
2599 	mutex_exit(&bind_connfp->connf_lock);
2600 	return (NULL);
2601 }
2602 
2603 /*
2604  * Finds a TCP/IPv6 listening connection; called by tcp_disconnect to locate
2605  * a listener when changing state.
2606  */
2607 conn_t *
2608 ipcl_lookup_listener_v6(uint16_t lport, in6_addr_t *laddr, uint_t ifindex,
2609     zoneid_t zoneid, ip_stack_t *ipst)
2610 {
2611 	connf_t		*bind_connfp;
2612 	conn_t		*connp = NULL;
2613 	tcp_t		*tcp;
2614 
2615 	/*
2616 	 * Avoid false matches for packets sent to an IP destination of
2617 	 * all zeros.
2618 	 */
2619 	if (IN6_IS_ADDR_UNSPECIFIED(laddr))
2620 		return (NULL);
2621 
2622 	ASSERT(zoneid != ALL_ZONES);
2623 
2624 	bind_connfp = &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)];
2625 	mutex_enter(&bind_connfp->connf_lock);
2626 	for (connp = bind_connfp->connf_head; connp != NULL;
2627 	    connp = connp->conn_next) {
2628 		tcp = connp->conn_tcp;
2629 		if (IPCL_BIND_MATCH_V6(connp, IPPROTO_TCP, *laddr, lport) &&
2630 		    IPCL_ZONE_MATCH(connp, zoneid) &&
2631 		    (tcp->tcp_bound_if == 0 ||
2632 		    tcp->tcp_bound_if == ifindex) &&
2633 		    tcp->tcp_listener == NULL) {
2634 			CONN_INC_REF(connp);
2635 			mutex_exit(&bind_connfp->connf_lock);
2636 			return (connp);
2637 		}
2638 	}
2639 	mutex_exit(&bind_connfp->connf_lock);
2640 	return (NULL);
2641 }
2642 
2643 /*
2644  * ipcl_get_next_conn
2645  *	get the next entry in the conn global list
2646  *	and put a reference on the next_conn.
2647  *	decrement the reference on the current conn.
2648  *
2649  * This is an iterator based walker function that also provides for
2650  * some selection by the caller. It walks through the conn_hash bucket
2651  * searching for the next valid connp in the list, and selects connections
2652  * that are neither closed nor condemned. It also REFHOLDS the conn
2653  * thus ensuring that the conn exists when the caller uses the conn.
2654  */
2655 conn_t *
2656 ipcl_get_next_conn(connf_t *connfp, conn_t *connp, uint32_t conn_flags)
2657 {
2658 	conn_t	*next_connp;
2659 
2660 	if (connfp == NULL)
2661 		return (NULL);
2662 
2663 	mutex_enter(&connfp->connf_lock);
2664 
2665 	next_connp = (connp == NULL) ?
2666 	    connfp->connf_head : connp->conn_g_next;
2667 
2668 	while (next_connp != NULL) {
2669 		mutex_enter(&next_connp->conn_lock);
2670 		if (!(next_connp->conn_flags & conn_flags) ||
2671 		    (next_connp->conn_state_flags &
2672 		    (CONN_CONDEMNED | CONN_INCIPIENT))) {
2673 			/*
2674 			 * This conn has been condemned or
2675 			 * is closing, or the flags don't match
2676 			 */
2677 			mutex_exit(&next_connp->conn_lock);
2678 			next_connp = next_connp->conn_g_next;
2679 			continue;
2680 		}
2681 		CONN_INC_REF_LOCKED(next_connp);
2682 		mutex_exit(&next_connp->conn_lock);
2683 		break;
2684 	}
2685 
2686 	mutex_exit(&connfp->connf_lock);
2687 
2688 	if (connp != NULL)
2689 		CONN_DEC_REF(connp);
2690 
2691 	return (next_connp);
2692 }
2693 
2694 #ifdef CONN_DEBUG
2695 /*
2696  * Trace of the last NBUF refhold/refrele
2697  */
2698 int
2699 conn_trace_ref(conn_t *connp)
2700 {
2701 	int	last;
2702 	conn_trace_t	*ctb;
2703 
2704 	ASSERT(MUTEX_HELD(&connp->conn_lock));
2705 	last = connp->conn_trace_last;
2706 	last++;
2707 	if (last == CONN_TRACE_MAX)
2708 		last = 0;
2709 
2710 	ctb = &connp->conn_trace_buf[last];
2711 	ctb->ctb_depth = getpcstack(ctb->ctb_stack, CONN_STACK_DEPTH);
2712 	connp->conn_trace_last = last;
2713 	return (1);
2714 }
2715 
2716 int
2717 conn_untrace_ref(conn_t *connp)
2718 {
2719 	int	last;
2720 	conn_trace_t	*ctb;
2721 
2722 	ASSERT(MUTEX_HELD(&connp->conn_lock));
2723 	last = connp->conn_trace_last;
2724 	last++;
2725 	if (last == CONN_TRACE_MAX)
2726 		last = 0;
2727 
2728 	ctb = &connp->conn_trace_buf[last];
2729 	ctb->ctb_depth = getpcstack(ctb->ctb_stack, CONN_STACK_DEPTH);
2730 	connp->conn_trace_last = last;
2731 	return (1);
2732 }
2733 #endif
2734