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 (c) 2003, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright 2013 Nexenta Systems, Inc. All rights reserved.
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 * conn_recv is used to pass up packets to the ULP.
56 * For TCP conn_recv changes. It is tcp_input_listener_unbound initially for
57 * a listener, and changes to tcp_input_listener as the listener has picked a
58 * good squeue. For other cases it is set to tcp_input_data.
59 *
60 * conn_recvicmp is used to pass up ICMP errors to the ULP.
61 *
62 * Classifier uses several hash tables:
63 *
64 * ipcl_conn_fanout: contains all TCP connections in CONNECTED state
65 * ipcl_bind_fanout: contains all connections in BOUND state
66 * ipcl_proto_fanout: IPv4 protocol fanout
67 * ipcl_proto_fanout_v6: IPv6 protocol fanout
68 * ipcl_udp_fanout: contains all UDP connections
69 * ipcl_iptun_fanout: contains all IP tunnel connections
70 * ipcl_globalhash_fanout: contains all connections
71 *
72 * The ipcl_globalhash_fanout is used for any walkers (like snmp and Clustering)
73 * which need to view all existing connections.
74 *
75 * All tables are protected by per-bucket locks. When both per-bucket lock and
76 * connection lock need to be held, the per-bucket lock should be acquired
77 * first, followed by the connection lock.
78 *
79 * All functions doing search in one of these tables increment a reference
80 * counter on the connection found (if any). This reference should be dropped
81 * when the caller has finished processing the connection.
82 *
83 *
84 * INTERFACES:
85 * ===========
86 *
87 * Connection Lookup:
88 * ------------------
89 *
90 * conn_t *ipcl_classify_v4(mp, protocol, hdr_len, ira, ip_stack)
91 * conn_t *ipcl_classify_v6(mp, protocol, hdr_len, ira, ip_stack)
92 *
93 * Finds connection for an incoming IPv4 or IPv6 packet. Returns NULL if
94 * it can't find any associated connection. If the connection is found, its
95 * reference counter is incremented.
96 *
97 * mp: mblock, containing packet header. The full header should fit
98 * into a single mblock. It should also contain at least full IP
99 * and TCP or UDP header.
100 *
101 * protocol: Either IPPROTO_TCP or IPPROTO_UDP.
102 *
103 * hdr_len: The size of IP header. It is used to find TCP or UDP header in
104 * the packet.
105 *
106 * ira->ira_zoneid: The zone in which the returned connection must be; the
107 * zoneid corresponding to the ire_zoneid on the IRE located for
108 * the packet's destination address.
109 *
110 * ira->ira_flags: Contains the IRAF_TX_MAC_EXEMPTABLE and
111 * IRAF_TX_SHARED_ADDR flags
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 *, tcpha_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);
187 * int ipcl_conn_insert_v4(connp);
188 * int ipcl_conn_insert_v6(connp);
189 *
190 * Insert 'connp' in the ipcl_conn_fanout.
191 * Arguements :
192 * connp conn_t to be inserted
193 *
194 * Return value :
195 * 0 if connp was inserted
196 * EADDRINUSE if the connection with the same tuple
197 * already exists.
198 *
199 * int ipcl_bind_insert(connp);
200 * int ipcl_bind_insert_v4(connp);
201 * int ipcl_bind_insert_v6(connp);
202 *
203 * Insert 'connp' in ipcl_bind_fanout.
204 * Arguements :
205 * connp conn_t to be inserted
206 *
207 *
208 * void ipcl_hash_remove(connp);
209 *
210 * Removes the 'connp' from the connection fanout table.
211 *
212 * Connection Creation/Destruction
213 * -------------------------------
214 *
215 * conn_t *ipcl_conn_create(type, sleep, netstack_t *)
216 *
217 * Creates a new conn based on the type flag, inserts it into
218 * globalhash table.
219 *
220 * type: This flag determines the type of conn_t which needs to be
221 * created i.e., which kmem_cache it comes from.
222 * IPCL_TCPCONN indicates a TCP connection
223 * IPCL_SCTPCONN indicates a SCTP connection
224 * IPCL_UDPCONN indicates a UDP conn_t.
225 * IPCL_RAWIPCONN indicates a RAWIP/ICMP conn_t.
226 * IPCL_RTSCONN indicates a RTS conn_t.
227 * IPCL_IPCCONN indicates all other connections.
228 *
229 * void ipcl_conn_destroy(connp)
230 *
231 * Destroys the connection state, removes it from the global
232 * connection hash table and frees its memory.
233 */
234
235 #include <sys/types.h>
236 #include <sys/stream.h>
237 #include <sys/stropts.h>
238 #include <sys/sysmacros.h>
239 #include <sys/strsubr.h>
240 #include <sys/strsun.h>
241 #define _SUN_TPI_VERSION 2
242 #include <sys/ddi.h>
243 #include <sys/cmn_err.h>
244 #include <sys/debug.h>
245
246 #include <sys/systm.h>
247 #include <sys/param.h>
248 #include <sys/kmem.h>
249 #include <sys/isa_defs.h>
250 #include <inet/common.h>
251 #include <netinet/ip6.h>
252 #include <netinet/icmp6.h>
253
254 #include <inet/ip.h>
255 #include <inet/ip_if.h>
256 #include <inet/ip_ire.h>
257 #include <inet/ip6.h>
258 #include <inet/ip_ndp.h>
259 #include <inet/ip_impl.h>
260 #include <inet/udp_impl.h>
261 #include <inet/sctp_ip.h>
262 #include <inet/sctp/sctp_impl.h>
263 #include <inet/rawip_impl.h>
264 #include <inet/rts_impl.h>
265 #include <inet/iptun/iptun_impl.h>
266
267 #include <sys/cpuvar.h>
268
269 #include <inet/ipclassifier.h>
270 #include <inet/tcp.h>
271 #include <inet/ipsec_impl.h>
272
273 #include <sys/tsol/tnet.h>
274 #include <sys/sockio.h>
275
276 /* Old value for compatibility. Setable in /etc/system */
277 uint_t tcp_conn_hash_size = 0;
278
279 /* New value. Zero means choose automatically. Setable in /etc/system */
280 volatile uint_t ipcl_conn_hash_size = 0;
281 uint_t ipcl_conn_hash_memfactor = 8192;
282 uint_t ipcl_conn_hash_maxsize = 82500;
283
284 /* bind/udp fanout table size */
285 uint_t ipcl_bind_fanout_size = 512;
286 uint_t ipcl_udp_fanout_size = 16384;
287
288 /* Raw socket fanout size. Must be a power of 2. */
289 uint_t ipcl_raw_fanout_size = 256;
290
291 /*
292 * The IPCL_IPTUN_HASH() function works best with a prime table size. We
293 * expect that most large deployments would have hundreds of tunnels, and
294 * thousands in the extreme case.
295 */
296 uint_t ipcl_iptun_fanout_size = 6143;
297
298 /*
299 * Power of 2^N Primes useful for hashing for N of 0-28,
300 * these primes are the nearest prime <= 2^N - 2^(N-2).
301 */
302
303 #define P2Ps() {0, 0, 0, 5, 11, 23, 47, 89, 191, 383, 761, 1531, 3067, \
304 6143, 12281, 24571, 49139, 98299, 196597, 393209, \
305 786431, 1572853, 3145721, 6291449, 12582893, 25165813, \
306 50331599, 100663291, 201326557, 0}
307
308 /*
309 * wrapper structure to ensure that conn and what follows it (tcp_t, etc)
310 * are aligned on cache lines.
311 */
312 typedef union itc_s {
313 conn_t itc_conn;
314 char itcu_filler[CACHE_ALIGN(conn_s)];
315 } itc_t;
316
317 struct kmem_cache *tcp_conn_cache;
318 struct kmem_cache *ip_conn_cache;
319 extern struct kmem_cache *sctp_conn_cache;
320 struct kmem_cache *udp_conn_cache;
321 struct kmem_cache *rawip_conn_cache;
322 struct kmem_cache *rts_conn_cache;
323
324 extern void tcp_timermp_free(tcp_t *);
325 extern mblk_t *tcp_timermp_alloc(int);
326
327 static int ip_conn_constructor(void *, void *, int);
328 static void ip_conn_destructor(void *, void *);
329
330 static int tcp_conn_constructor(void *, void *, int);
331 static void tcp_conn_destructor(void *, void *);
332
333 static int udp_conn_constructor(void *, void *, int);
334 static void udp_conn_destructor(void *, void *);
335
336 static int rawip_conn_constructor(void *, void *, int);
337 static void rawip_conn_destructor(void *, void *);
338
339 static int rts_conn_constructor(void *, void *, int);
340 static void rts_conn_destructor(void *, void *);
341
342 /*
343 * Global (for all stack instances) init routine
344 */
345 void
346 ipcl_g_init(void)
347 {
348 ip_conn_cache = kmem_cache_create("ip_conn_cache",
349 sizeof (conn_t), CACHE_ALIGN_SIZE,
350 ip_conn_constructor, ip_conn_destructor,
351 NULL, NULL, NULL, 0);
352
353 tcp_conn_cache = kmem_cache_create("tcp_conn_cache",
354 sizeof (itc_t) + sizeof (tcp_t), CACHE_ALIGN_SIZE,
355 tcp_conn_constructor, tcp_conn_destructor,
356 tcp_conn_reclaim, NULL, NULL, 0);
357
358 udp_conn_cache = kmem_cache_create("udp_conn_cache",
359 sizeof (itc_t) + sizeof (udp_t), CACHE_ALIGN_SIZE,
360 udp_conn_constructor, udp_conn_destructor,
361 NULL, NULL, NULL, 0);
362
363 rawip_conn_cache = kmem_cache_create("rawip_conn_cache",
364 sizeof (itc_t) + sizeof (icmp_t), CACHE_ALIGN_SIZE,
365 rawip_conn_constructor, rawip_conn_destructor,
366 NULL, NULL, NULL, 0);
367
368 rts_conn_cache = kmem_cache_create("rts_conn_cache",
369 sizeof (itc_t) + sizeof (rts_t), CACHE_ALIGN_SIZE,
370 rts_conn_constructor, rts_conn_destructor,
371 NULL, NULL, NULL, 0);
372 }
373
374 /*
375 * ipclassifier intialization routine, sets up hash tables.
376 */
377 void
378 ipcl_init(ip_stack_t *ipst)
379 {
380 int i;
381 int sizes[] = P2Ps();
382
383 /*
384 * Calculate size of conn fanout table from /etc/system settings
385 */
386 if (ipcl_conn_hash_size != 0) {
387 ipst->ips_ipcl_conn_fanout_size = ipcl_conn_hash_size;
388 } else if (tcp_conn_hash_size != 0) {
389 ipst->ips_ipcl_conn_fanout_size = tcp_conn_hash_size;
390 } else {
391 extern pgcnt_t freemem;
392
393 ipst->ips_ipcl_conn_fanout_size =
394 (freemem * PAGESIZE) / ipcl_conn_hash_memfactor;
395
396 if (ipst->ips_ipcl_conn_fanout_size > ipcl_conn_hash_maxsize) {
397 ipst->ips_ipcl_conn_fanout_size =
398 ipcl_conn_hash_maxsize;
399 }
400 }
401
402 for (i = 9; i < sizeof (sizes) / sizeof (*sizes) - 1; i++) {
403 if (sizes[i] >= ipst->ips_ipcl_conn_fanout_size) {
404 break;
405 }
406 }
407 if ((ipst->ips_ipcl_conn_fanout_size = sizes[i]) == 0) {
408 /* Out of range, use the 2^16 value */
409 ipst->ips_ipcl_conn_fanout_size = sizes[16];
410 }
411
412 /* Take values from /etc/system */
413 ipst->ips_ipcl_bind_fanout_size = ipcl_bind_fanout_size;
414 ipst->ips_ipcl_udp_fanout_size = ipcl_udp_fanout_size;
415 ipst->ips_ipcl_raw_fanout_size = ipcl_raw_fanout_size;
416 ipst->ips_ipcl_iptun_fanout_size = ipcl_iptun_fanout_size;
417
418 ASSERT(ipst->ips_ipcl_conn_fanout == NULL);
419
420 ipst->ips_ipcl_conn_fanout = kmem_zalloc(
421 ipst->ips_ipcl_conn_fanout_size * sizeof (connf_t), KM_SLEEP);
422
423 for (i = 0; i < ipst->ips_ipcl_conn_fanout_size; i++) {
424 mutex_init(&ipst->ips_ipcl_conn_fanout[i].connf_lock, NULL,
425 MUTEX_DEFAULT, NULL);
426 }
427
428 ipst->ips_ipcl_bind_fanout = kmem_zalloc(
429 ipst->ips_ipcl_bind_fanout_size * sizeof (connf_t), KM_SLEEP);
430
431 for (i = 0; i < ipst->ips_ipcl_bind_fanout_size; i++) {
432 mutex_init(&ipst->ips_ipcl_bind_fanout[i].connf_lock, NULL,
433 MUTEX_DEFAULT, NULL);
434 }
435
436 ipst->ips_ipcl_proto_fanout_v4 = kmem_zalloc(IPPROTO_MAX *
437 sizeof (connf_t), KM_SLEEP);
438 for (i = 0; i < IPPROTO_MAX; i++) {
439 mutex_init(&ipst->ips_ipcl_proto_fanout_v4[i].connf_lock, NULL,
440 MUTEX_DEFAULT, NULL);
441 }
442
443 ipst->ips_ipcl_proto_fanout_v6 = kmem_zalloc(IPPROTO_MAX *
444 sizeof (connf_t), KM_SLEEP);
445 for (i = 0; i < IPPROTO_MAX; i++) {
446 mutex_init(&ipst->ips_ipcl_proto_fanout_v6[i].connf_lock, NULL,
447 MUTEX_DEFAULT, NULL);
448 }
449
450 ipst->ips_rts_clients = kmem_zalloc(sizeof (connf_t), KM_SLEEP);
451 mutex_init(&ipst->ips_rts_clients->connf_lock,
452 NULL, MUTEX_DEFAULT, NULL);
453
454 ipst->ips_ipcl_udp_fanout = kmem_zalloc(
455 ipst->ips_ipcl_udp_fanout_size * sizeof (connf_t), KM_SLEEP);
456 for (i = 0; i < ipst->ips_ipcl_udp_fanout_size; i++) {
457 mutex_init(&ipst->ips_ipcl_udp_fanout[i].connf_lock, NULL,
458 MUTEX_DEFAULT, NULL);
459 }
460
461 ipst->ips_ipcl_iptun_fanout = kmem_zalloc(
462 ipst->ips_ipcl_iptun_fanout_size * sizeof (connf_t), KM_SLEEP);
463 for (i = 0; i < ipst->ips_ipcl_iptun_fanout_size; i++) {
464 mutex_init(&ipst->ips_ipcl_iptun_fanout[i].connf_lock, NULL,
465 MUTEX_DEFAULT, NULL);
466 }
467
468 ipst->ips_ipcl_raw_fanout = kmem_zalloc(
469 ipst->ips_ipcl_raw_fanout_size * sizeof (connf_t), KM_SLEEP);
470 for (i = 0; i < ipst->ips_ipcl_raw_fanout_size; i++) {
471 mutex_init(&ipst->ips_ipcl_raw_fanout[i].connf_lock, NULL,
472 MUTEX_DEFAULT, NULL);
473 }
474
475 ipst->ips_ipcl_globalhash_fanout = kmem_zalloc(
476 sizeof (connf_t) * CONN_G_HASH_SIZE, KM_SLEEP);
477 for (i = 0; i < CONN_G_HASH_SIZE; i++) {
478 mutex_init(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock,
479 NULL, MUTEX_DEFAULT, NULL);
480 }
481 }
482
483 void
484 ipcl_g_destroy(void)
485 {
486 kmem_cache_destroy(ip_conn_cache);
487 kmem_cache_destroy(tcp_conn_cache);
488 kmem_cache_destroy(udp_conn_cache);
489 kmem_cache_destroy(rawip_conn_cache);
490 kmem_cache_destroy(rts_conn_cache);
491 }
492
493 /*
494 * All user-level and kernel use of the stack must be gone
495 * by now.
496 */
497 void
498 ipcl_destroy(ip_stack_t *ipst)
499 {
500 int i;
501
502 for (i = 0; i < ipst->ips_ipcl_conn_fanout_size; i++) {
503 ASSERT(ipst->ips_ipcl_conn_fanout[i].connf_head == NULL);
504 mutex_destroy(&ipst->ips_ipcl_conn_fanout[i].connf_lock);
505 }
506 kmem_free(ipst->ips_ipcl_conn_fanout, ipst->ips_ipcl_conn_fanout_size *
507 sizeof (connf_t));
508 ipst->ips_ipcl_conn_fanout = NULL;
509
510 for (i = 0; i < ipst->ips_ipcl_bind_fanout_size; i++) {
511 ASSERT(ipst->ips_ipcl_bind_fanout[i].connf_head == NULL);
512 mutex_destroy(&ipst->ips_ipcl_bind_fanout[i].connf_lock);
513 }
514 kmem_free(ipst->ips_ipcl_bind_fanout, ipst->ips_ipcl_bind_fanout_size *
515 sizeof (connf_t));
516 ipst->ips_ipcl_bind_fanout = NULL;
517
518 for (i = 0; i < IPPROTO_MAX; i++) {
519 ASSERT(ipst->ips_ipcl_proto_fanout_v4[i].connf_head == NULL);
520 mutex_destroy(&ipst->ips_ipcl_proto_fanout_v4[i].connf_lock);
521 }
522 kmem_free(ipst->ips_ipcl_proto_fanout_v4,
523 IPPROTO_MAX * sizeof (connf_t));
524 ipst->ips_ipcl_proto_fanout_v4 = NULL;
525
526 for (i = 0; i < IPPROTO_MAX; i++) {
527 ASSERT(ipst->ips_ipcl_proto_fanout_v6[i].connf_head == NULL);
528 mutex_destroy(&ipst->ips_ipcl_proto_fanout_v6[i].connf_lock);
529 }
530 kmem_free(ipst->ips_ipcl_proto_fanout_v6,
531 IPPROTO_MAX * sizeof (connf_t));
532 ipst->ips_ipcl_proto_fanout_v6 = NULL;
533
534 for (i = 0; i < ipst->ips_ipcl_udp_fanout_size; i++) {
535 ASSERT(ipst->ips_ipcl_udp_fanout[i].connf_head == NULL);
536 mutex_destroy(&ipst->ips_ipcl_udp_fanout[i].connf_lock);
537 }
538 kmem_free(ipst->ips_ipcl_udp_fanout, ipst->ips_ipcl_udp_fanout_size *
539 sizeof (connf_t));
540 ipst->ips_ipcl_udp_fanout = NULL;
541
542 for (i = 0; i < ipst->ips_ipcl_iptun_fanout_size; i++) {
543 ASSERT(ipst->ips_ipcl_iptun_fanout[i].connf_head == NULL);
544 mutex_destroy(&ipst->ips_ipcl_iptun_fanout[i].connf_lock);
545 }
546 kmem_free(ipst->ips_ipcl_iptun_fanout,
547 ipst->ips_ipcl_iptun_fanout_size * sizeof (connf_t));
548 ipst->ips_ipcl_iptun_fanout = NULL;
549
550 for (i = 0; i < ipst->ips_ipcl_raw_fanout_size; i++) {
551 ASSERT(ipst->ips_ipcl_raw_fanout[i].connf_head == NULL);
552 mutex_destroy(&ipst->ips_ipcl_raw_fanout[i].connf_lock);
553 }
554 kmem_free(ipst->ips_ipcl_raw_fanout, ipst->ips_ipcl_raw_fanout_size *
555 sizeof (connf_t));
556 ipst->ips_ipcl_raw_fanout = NULL;
557
558 for (i = 0; i < CONN_G_HASH_SIZE; i++) {
559 ASSERT(ipst->ips_ipcl_globalhash_fanout[i].connf_head == NULL);
560 mutex_destroy(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock);
561 }
562 kmem_free(ipst->ips_ipcl_globalhash_fanout,
563 sizeof (connf_t) * CONN_G_HASH_SIZE);
564 ipst->ips_ipcl_globalhash_fanout = NULL;
565
566 ASSERT(ipst->ips_rts_clients->connf_head == NULL);
567 mutex_destroy(&ipst->ips_rts_clients->connf_lock);
568 kmem_free(ipst->ips_rts_clients, sizeof (connf_t));
569 ipst->ips_rts_clients = NULL;
570 }
571
572 /*
573 * conn creation routine. initialize the conn, sets the reference
574 * and inserts it in the global hash table.
575 */
576 conn_t *
577 ipcl_conn_create(uint32_t type, int sleep, netstack_t *ns)
578 {
579 conn_t *connp;
580 struct kmem_cache *conn_cache;
581
582 switch (type) {
583 case IPCL_SCTPCONN:
584 if ((connp = kmem_cache_alloc(sctp_conn_cache, sleep)) == NULL)
585 return (NULL);
586 sctp_conn_init(connp);
587 netstack_hold(ns);
588 connp->conn_netstack = ns;
589 connp->conn_ixa->ixa_ipst = ns->netstack_ip;
590 connp->conn_ixa->ixa_conn_id = (long)connp;
591 ipcl_globalhash_insert(connp);
592 return (connp);
593
594 case IPCL_TCPCONN:
595 conn_cache = tcp_conn_cache;
596 break;
597
598 case IPCL_UDPCONN:
599 conn_cache = udp_conn_cache;
600 break;
601
602 case IPCL_RAWIPCONN:
603 conn_cache = rawip_conn_cache;
604 break;
605
606 case IPCL_RTSCONN:
607 conn_cache = rts_conn_cache;
608 break;
609
610 case IPCL_IPCCONN:
611 conn_cache = ip_conn_cache;
612 break;
613
614 default:
615 connp = NULL;
616 ASSERT(0);
617 }
618
619 if ((connp = kmem_cache_alloc(conn_cache, sleep)) == NULL)
620 return (NULL);
621
622 connp->conn_ref = 1;
623 netstack_hold(ns);
624 connp->conn_netstack = ns;
625 connp->conn_ixa->ixa_ipst = ns->netstack_ip;
626 connp->conn_ixa->ixa_conn_id = (long)connp;
627 ipcl_globalhash_insert(connp);
628 return (connp);
629 }
630
631 void
632 ipcl_conn_destroy(conn_t *connp)
633 {
634 mblk_t *mp;
635 netstack_t *ns = connp->conn_netstack;
636
637 ASSERT(!MUTEX_HELD(&connp->conn_lock));
638 ASSERT(connp->conn_ref == 0);
639 ASSERT(connp->conn_ioctlref == 0);
640
641 DTRACE_PROBE1(conn__destroy, conn_t *, connp);
642
643 if (connp->conn_cred != NULL) {
644 crfree(connp->conn_cred);
645 connp->conn_cred = NULL;
646 /* ixa_cred done in ipcl_conn_cleanup below */
647 }
648
649 if (connp->conn_ht_iphc != NULL) {
650 kmem_free(connp->conn_ht_iphc, connp->conn_ht_iphc_allocated);
651 connp->conn_ht_iphc = NULL;
652 connp->conn_ht_iphc_allocated = 0;
653 connp->conn_ht_iphc_len = 0;
654 connp->conn_ht_ulp = NULL;
655 connp->conn_ht_ulp_len = 0;
656 }
657 ip_pkt_free(&connp->conn_xmit_ipp);
658
659 ipcl_globalhash_remove(connp);
660
661 if (connp->conn_latch != NULL) {
662 IPLATCH_REFRELE(connp->conn_latch);
663 connp->conn_latch = NULL;
664 }
665 if (connp->conn_latch_in_policy != NULL) {
666 IPPOL_REFRELE(connp->conn_latch_in_policy);
667 connp->conn_latch_in_policy = NULL;
668 }
669 if (connp->conn_latch_in_action != NULL) {
670 IPACT_REFRELE(connp->conn_latch_in_action);
671 connp->conn_latch_in_action = NULL;
672 }
673 if (connp->conn_policy != NULL) {
674 IPPH_REFRELE(connp->conn_policy, ns);
675 connp->conn_policy = NULL;
676 }
677
678 if (connp->conn_ipsec_opt_mp != NULL) {
679 freemsg(connp->conn_ipsec_opt_mp);
680 connp->conn_ipsec_opt_mp = NULL;
681 }
682
683 if (connp->conn_flags & IPCL_TCPCONN) {
684 tcp_t *tcp = connp->conn_tcp;
685
686 tcp_free(tcp);
687 mp = tcp->tcp_timercache;
688
689 tcp->tcp_tcps = NULL;
690
691 /*
692 * tcp_rsrv_mp can be NULL if tcp_get_conn() fails to allocate
693 * the mblk.
694 */
695 if (tcp->tcp_rsrv_mp != NULL) {
696 freeb(tcp->tcp_rsrv_mp);
697 tcp->tcp_rsrv_mp = NULL;
698 mutex_destroy(&tcp->tcp_rsrv_mp_lock);
699 }
700
701 ipcl_conn_cleanup(connp);
702 connp->conn_flags = IPCL_TCPCONN;
703 if (ns != NULL) {
704 ASSERT(tcp->tcp_tcps == NULL);
705 connp->conn_netstack = NULL;
706 connp->conn_ixa->ixa_ipst = NULL;
707 netstack_rele(ns);
708 }
709
710 bzero(tcp, sizeof (tcp_t));
711
712 tcp->tcp_timercache = mp;
713 tcp->tcp_connp = connp;
714 kmem_cache_free(tcp_conn_cache, connp);
715 return;
716 }
717
718 if (connp->conn_flags & IPCL_SCTPCONN) {
719 ASSERT(ns != NULL);
720 sctp_free(connp);
721 return;
722 }
723
724 ipcl_conn_cleanup(connp);
725 if (ns != NULL) {
726 connp->conn_netstack = NULL;
727 connp->conn_ixa->ixa_ipst = NULL;
728 netstack_rele(ns);
729 }
730
731 /* leave conn_priv aka conn_udp, conn_icmp, etc in place. */
732 if (connp->conn_flags & IPCL_UDPCONN) {
733 connp->conn_flags = IPCL_UDPCONN;
734 kmem_cache_free(udp_conn_cache, connp);
735 } else if (connp->conn_flags & IPCL_RAWIPCONN) {
736 connp->conn_flags = IPCL_RAWIPCONN;
737 connp->conn_proto = IPPROTO_ICMP;
738 connp->conn_ixa->ixa_protocol = connp->conn_proto;
739 kmem_cache_free(rawip_conn_cache, connp);
740 } else if (connp->conn_flags & IPCL_RTSCONN) {
741 connp->conn_flags = IPCL_RTSCONN;
742 kmem_cache_free(rts_conn_cache, connp);
743 } else {
744 connp->conn_flags = IPCL_IPCCONN;
745 ASSERT(connp->conn_flags & IPCL_IPCCONN);
746 ASSERT(connp->conn_priv == NULL);
747 kmem_cache_free(ip_conn_cache, connp);
748 }
749 }
750
751 /*
752 * Running in cluster mode - deregister listener information
753 */
754 static void
755 ipcl_conn_unlisten(conn_t *connp)
756 {
757 ASSERT((connp->conn_flags & IPCL_CL_LISTENER) != 0);
758 ASSERT(connp->conn_lport != 0);
759
760 if (cl_inet_unlisten != NULL) {
761 sa_family_t addr_family;
762 uint8_t *laddrp;
763
764 if (connp->conn_ipversion == IPV6_VERSION) {
765 addr_family = AF_INET6;
766 laddrp = (uint8_t *)&connp->conn_bound_addr_v6;
767 } else {
768 addr_family = AF_INET;
769 laddrp = (uint8_t *)&connp->conn_bound_addr_v4;
770 }
771 (*cl_inet_unlisten)(connp->conn_netstack->netstack_stackid,
772 IPPROTO_TCP, addr_family, laddrp, connp->conn_lport, NULL);
773 }
774 connp->conn_flags &= ~IPCL_CL_LISTENER;
775 }
776
777 /*
778 * We set the IPCL_REMOVED flag (instead of clearing the flag indicating
779 * which table the conn belonged to). So for debugging we can see which hash
780 * table this connection was in.
781 */
782 #define IPCL_HASH_REMOVE(connp) { \
783 connf_t *connfp = (connp)->conn_fanout; \
784 ASSERT(!MUTEX_HELD(&((connp)->conn_lock))); \
785 if (connfp != NULL) { \
786 mutex_enter(&connfp->connf_lock); \
787 if ((connp)->conn_next != NULL) \
788 (connp)->conn_next->conn_prev = \
789 (connp)->conn_prev; \
790 if ((connp)->conn_prev != NULL) \
791 (connp)->conn_prev->conn_next = \
792 (connp)->conn_next; \
793 else \
794 connfp->connf_head = (connp)->conn_next; \
795 (connp)->conn_fanout = NULL; \
796 (connp)->conn_next = NULL; \
797 (connp)->conn_prev = NULL; \
798 (connp)->conn_flags |= IPCL_REMOVED; \
799 if (((connp)->conn_flags & IPCL_CL_LISTENER) != 0) \
800 ipcl_conn_unlisten((connp)); \
801 CONN_DEC_REF((connp)); \
802 mutex_exit(&connfp->connf_lock); \
803 } \
804 }
805
806 void
807 ipcl_hash_remove(conn_t *connp)
808 {
809 uint8_t protocol = connp->conn_proto;
810
811 IPCL_HASH_REMOVE(connp);
812 if (protocol == IPPROTO_RSVP)
813 ill_set_inputfn_all(connp->conn_netstack->netstack_ip);
814 }
815
816 /*
817 * The whole purpose of this function is allow removal of
818 * a conn_t from the connected hash for timewait reclaim.
819 * This is essentially a TW reclaim fastpath where timewait
820 * collector checks under fanout lock (so no one else can
821 * get access to the conn_t) that refcnt is 2 i.e. one for
822 * TCP and one for the classifier hash list. If ref count
823 * is indeed 2, we can just remove the conn under lock and
824 * avoid cleaning up the conn under squeue. This gives us
825 * improved performance.
826 */
827 void
828 ipcl_hash_remove_locked(conn_t *connp, connf_t *connfp)
829 {
830 ASSERT(MUTEX_HELD(&connfp->connf_lock));
831 ASSERT(MUTEX_HELD(&connp->conn_lock));
832 ASSERT((connp->conn_flags & IPCL_CL_LISTENER) == 0);
833
834 if ((connp)->conn_next != NULL) {
835 (connp)->conn_next->conn_prev = (connp)->conn_prev;
836 }
837 if ((connp)->conn_prev != NULL) {
838 (connp)->conn_prev->conn_next = (connp)->conn_next;
839 } else {
840 connfp->connf_head = (connp)->conn_next;
841 }
842 (connp)->conn_fanout = NULL;
843 (connp)->conn_next = NULL;
844 (connp)->conn_prev = NULL;
845 (connp)->conn_flags |= IPCL_REMOVED;
846 ASSERT((connp)->conn_ref == 2);
847 (connp)->conn_ref--;
848 }
849
850 #define IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp) { \
851 ASSERT((connp)->conn_fanout == NULL); \
852 ASSERT((connp)->conn_next == NULL); \
853 ASSERT((connp)->conn_prev == NULL); \
854 if ((connfp)->connf_head != NULL) { \
855 (connfp)->connf_head->conn_prev = (connp); \
856 (connp)->conn_next = (connfp)->connf_head; \
857 } \
858 (connp)->conn_fanout = (connfp); \
859 (connfp)->connf_head = (connp); \
860 (connp)->conn_flags = ((connp)->conn_flags & ~IPCL_REMOVED) | \
861 IPCL_CONNECTED; \
862 CONN_INC_REF(connp); \
863 }
864
865 #define IPCL_HASH_INSERT_CONNECTED(connfp, connp) { \
866 IPCL_HASH_REMOVE((connp)); \
867 mutex_enter(&(connfp)->connf_lock); \
868 IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp); \
869 mutex_exit(&(connfp)->connf_lock); \
870 }
871
872 #define IPCL_HASH_INSERT_BOUND(connfp, connp) { \
873 conn_t *pconnp = NULL, *nconnp; \
874 IPCL_HASH_REMOVE((connp)); \
875 mutex_enter(&(connfp)->connf_lock); \
876 nconnp = (connfp)->connf_head; \
877 while (nconnp != NULL && \
878 !_IPCL_V4_MATCH_ANY(nconnp->conn_laddr_v6)) { \
879 pconnp = nconnp; \
880 nconnp = nconnp->conn_next; \
881 } \
882 if (pconnp != NULL) { \
883 pconnp->conn_next = (connp); \
884 (connp)->conn_prev = pconnp; \
885 } else { \
886 (connfp)->connf_head = (connp); \
887 } \
888 if (nconnp != NULL) { \
889 (connp)->conn_next = nconnp; \
890 nconnp->conn_prev = (connp); \
891 } \
892 (connp)->conn_fanout = (connfp); \
893 (connp)->conn_flags = ((connp)->conn_flags & ~IPCL_REMOVED) | \
894 IPCL_BOUND; \
895 CONN_INC_REF(connp); \
896 mutex_exit(&(connfp)->connf_lock); \
897 }
898
899 #define IPCL_HASH_INSERT_WILDCARD(connfp, connp) { \
900 conn_t **list, *prev, *next; \
901 boolean_t isv4mapped = \
902 IN6_IS_ADDR_V4MAPPED(&(connp)->conn_laddr_v6); \
903 IPCL_HASH_REMOVE((connp)); \
904 mutex_enter(&(connfp)->connf_lock); \
905 list = &(connfp)->connf_head; \
906 prev = NULL; \
907 while ((next = *list) != NULL) { \
908 if (isv4mapped && \
909 IN6_IS_ADDR_UNSPECIFIED(&next->conn_laddr_v6) && \
910 connp->conn_zoneid == next->conn_zoneid) { \
911 (connp)->conn_next = next; \
912 if (prev != NULL) \
913 prev = next->conn_prev; \
914 next->conn_prev = (connp); \
915 break; \
916 } \
917 list = &next->conn_next; \
918 prev = next; \
919 } \
920 (connp)->conn_prev = prev; \
921 *list = (connp); \
922 (connp)->conn_fanout = (connfp); \
923 (connp)->conn_flags = ((connp)->conn_flags & ~IPCL_REMOVED) | \
924 IPCL_BOUND; \
925 CONN_INC_REF((connp)); \
926 mutex_exit(&(connfp)->connf_lock); \
927 }
928
929 void
930 ipcl_hash_insert_wildcard(connf_t *connfp, conn_t *connp)
931 {
932 IPCL_HASH_INSERT_WILDCARD(connfp, connp);
933 }
934
935 /*
936 * Because the classifier is used to classify inbound packets, the destination
937 * address is meant to be our local tunnel address (tunnel source), and the
938 * source the remote tunnel address (tunnel destination).
939 *
940 * Note that conn_proto can't be used for fanout since the upper protocol
941 * can be both 41 and 4 when IPv6 and IPv4 are over the same tunnel.
942 */
943 conn_t *
944 ipcl_iptun_classify_v4(ipaddr_t *src, ipaddr_t *dst, ip_stack_t *ipst)
945 {
946 connf_t *connfp;
947 conn_t *connp;
948
949 /* first look for IPv4 tunnel links */
950 connfp = &ipst->ips_ipcl_iptun_fanout[IPCL_IPTUN_HASH(*dst, *src)];
951 mutex_enter(&connfp->connf_lock);
952 for (connp = connfp->connf_head; connp != NULL;
953 connp = connp->conn_next) {
954 if (IPCL_IPTUN_MATCH(connp, *dst, *src))
955 break;
956 }
957 if (connp != NULL)
958 goto done;
959
960 mutex_exit(&connfp->connf_lock);
961
962 /* We didn't find an IPv4 tunnel, try a 6to4 tunnel */
963 connfp = &ipst->ips_ipcl_iptun_fanout[IPCL_IPTUN_HASH(*dst,
964 INADDR_ANY)];
965 mutex_enter(&connfp->connf_lock);
966 for (connp = connfp->connf_head; connp != NULL;
967 connp = connp->conn_next) {
968 if (IPCL_IPTUN_MATCH(connp, *dst, INADDR_ANY))
969 break;
970 }
971 done:
972 if (connp != NULL)
973 CONN_INC_REF(connp);
974 mutex_exit(&connfp->connf_lock);
975 return (connp);
976 }
977
978 conn_t *
979 ipcl_iptun_classify_v6(in6_addr_t *src, in6_addr_t *dst, ip_stack_t *ipst)
980 {
981 connf_t *connfp;
982 conn_t *connp;
983
984 /* Look for an IPv6 tunnel link */
985 connfp = &ipst->ips_ipcl_iptun_fanout[IPCL_IPTUN_HASH_V6(dst, src)];
986 mutex_enter(&connfp->connf_lock);
987 for (connp = connfp->connf_head; connp != NULL;
988 connp = connp->conn_next) {
989 if (IPCL_IPTUN_MATCH_V6(connp, dst, src)) {
990 CONN_INC_REF(connp);
991 break;
992 }
993 }
994 mutex_exit(&connfp->connf_lock);
995 return (connp);
996 }
997
998 /*
999 * This function is used only for inserting SCTP raw socket now.
1000 * This may change later.
1001 *
1002 * Note that only one raw socket can be bound to a port. The param
1003 * lport is in network byte order.
1004 */
1005 static int
1006 ipcl_sctp_hash_insert(conn_t *connp, in_port_t lport)
1007 {
1008 connf_t *connfp;
1009 conn_t *oconnp;
1010 ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
1011
1012 connfp = &ipst->ips_ipcl_raw_fanout[IPCL_RAW_HASH(ntohs(lport), ipst)];
1013
1014 /* Check for existing raw socket already bound to the port. */
1015 mutex_enter(&connfp->connf_lock);
1016 for (oconnp = connfp->connf_head; oconnp != NULL;
1017 oconnp = oconnp->conn_next) {
1018 if (oconnp->conn_lport == lport &&
1019 oconnp->conn_zoneid == connp->conn_zoneid &&
1020 oconnp->conn_family == connp->conn_family &&
1021 ((IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6) ||
1022 IN6_IS_ADDR_UNSPECIFIED(&oconnp->conn_laddr_v6) ||
1023 IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_laddr_v6) ||
1024 IN6_IS_ADDR_V4MAPPED_ANY(&oconnp->conn_laddr_v6)) ||
1025 IN6_ARE_ADDR_EQUAL(&oconnp->conn_laddr_v6,
1026 &connp->conn_laddr_v6))) {
1027 break;
1028 }
1029 }
1030 mutex_exit(&connfp->connf_lock);
1031 if (oconnp != NULL)
1032 return (EADDRNOTAVAIL);
1033
1034 if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6) ||
1035 IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_faddr_v6)) {
1036 if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6) ||
1037 IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_laddr_v6)) {
1038 IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1039 } else {
1040 IPCL_HASH_INSERT_BOUND(connfp, connp);
1041 }
1042 } else {
1043 IPCL_HASH_INSERT_CONNECTED(connfp, connp);
1044 }
1045 return (0);
1046 }
1047
1048 static int
1049 ipcl_iptun_hash_insert(conn_t *connp, ip_stack_t *ipst)
1050 {
1051 connf_t *connfp;
1052 conn_t *tconnp;
1053 ipaddr_t laddr = connp->conn_laddr_v4;
1054 ipaddr_t faddr = connp->conn_faddr_v4;
1055
1056 connfp = &ipst->ips_ipcl_iptun_fanout[IPCL_IPTUN_HASH(laddr, faddr)];
1057 mutex_enter(&connfp->connf_lock);
1058 for (tconnp = connfp->connf_head; tconnp != NULL;
1059 tconnp = tconnp->conn_next) {
1060 if (IPCL_IPTUN_MATCH(tconnp, laddr, faddr)) {
1061 /* A tunnel is already bound to these addresses. */
1062 mutex_exit(&connfp->connf_lock);
1063 return (EADDRINUSE);
1064 }
1065 }
1066 IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp);
1067 mutex_exit(&connfp->connf_lock);
1068 return (0);
1069 }
1070
1071 static int
1072 ipcl_iptun_hash_insert_v6(conn_t *connp, ip_stack_t *ipst)
1073 {
1074 connf_t *connfp;
1075 conn_t *tconnp;
1076 in6_addr_t *laddr = &connp->conn_laddr_v6;
1077 in6_addr_t *faddr = &connp->conn_faddr_v6;
1078
1079 connfp = &ipst->ips_ipcl_iptun_fanout[IPCL_IPTUN_HASH_V6(laddr, faddr)];
1080 mutex_enter(&connfp->connf_lock);
1081 for (tconnp = connfp->connf_head; tconnp != NULL;
1082 tconnp = tconnp->conn_next) {
1083 if (IPCL_IPTUN_MATCH_V6(tconnp, laddr, faddr)) {
1084 /* A tunnel is already bound to these addresses. */
1085 mutex_exit(&connfp->connf_lock);
1086 return (EADDRINUSE);
1087 }
1088 }
1089 IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp);
1090 mutex_exit(&connfp->connf_lock);
1091 return (0);
1092 }
1093
1094 /*
1095 * Check for a MAC exemption conflict on a labeled system. Note that for
1096 * protocols that use port numbers (UDP, TCP, SCTP), we do this check up in the
1097 * transport layer. This check is for binding all other protocols.
1098 *
1099 * Returns true if there's a conflict.
1100 */
1101 static boolean_t
1102 check_exempt_conflict_v4(conn_t *connp, ip_stack_t *ipst)
1103 {
1104 connf_t *connfp;
1105 conn_t *tconn;
1106
1107 connfp = &ipst->ips_ipcl_proto_fanout_v4[connp->conn_proto];
1108 mutex_enter(&connfp->connf_lock);
1109 for (tconn = connfp->connf_head; tconn != NULL;
1110 tconn = tconn->conn_next) {
1111 /* We don't allow v4 fallback for v6 raw socket */
1112 if (connp->conn_family != tconn->conn_family)
1113 continue;
1114 /* If neither is exempt, then there's no conflict */
1115 if ((connp->conn_mac_mode == CONN_MAC_DEFAULT) &&
1116 (tconn->conn_mac_mode == CONN_MAC_DEFAULT))
1117 continue;
1118 /* We are only concerned about sockets for a different zone */
1119 if (connp->conn_zoneid == tconn->conn_zoneid)
1120 continue;
1121 /* If both are bound to different specific addrs, ok */
1122 if (connp->conn_laddr_v4 != INADDR_ANY &&
1123 tconn->conn_laddr_v4 != INADDR_ANY &&
1124 connp->conn_laddr_v4 != tconn->conn_laddr_v4)
1125 continue;
1126 /* These two conflict; fail */
1127 break;
1128 }
1129 mutex_exit(&connfp->connf_lock);
1130 return (tconn != NULL);
1131 }
1132
1133 static boolean_t
1134 check_exempt_conflict_v6(conn_t *connp, ip_stack_t *ipst)
1135 {
1136 connf_t *connfp;
1137 conn_t *tconn;
1138
1139 connfp = &ipst->ips_ipcl_proto_fanout_v6[connp->conn_proto];
1140 mutex_enter(&connfp->connf_lock);
1141 for (tconn = connfp->connf_head; tconn != NULL;
1142 tconn = tconn->conn_next) {
1143 /* We don't allow v4 fallback for v6 raw socket */
1144 if (connp->conn_family != tconn->conn_family)
1145 continue;
1146 /* If neither is exempt, then there's no conflict */
1147 if ((connp->conn_mac_mode == CONN_MAC_DEFAULT) &&
1148 (tconn->conn_mac_mode == CONN_MAC_DEFAULT))
1149 continue;
1150 /* We are only concerned about sockets for a different zone */
1151 if (connp->conn_zoneid == tconn->conn_zoneid)
1152 continue;
1153 /* If both are bound to different addrs, ok */
1154 if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6) &&
1155 !IN6_IS_ADDR_UNSPECIFIED(&tconn->conn_laddr_v6) &&
1156 !IN6_ARE_ADDR_EQUAL(&connp->conn_laddr_v6,
1157 &tconn->conn_laddr_v6))
1158 continue;
1159 /* These two conflict; fail */
1160 break;
1161 }
1162 mutex_exit(&connfp->connf_lock);
1163 return (tconn != NULL);
1164 }
1165
1166 /*
1167 * (v4, v6) bind hash insertion routines
1168 * The caller has already setup the conn (conn_proto, conn_laddr_v6, conn_lport)
1169 */
1170
1171 int
1172 ipcl_bind_insert(conn_t *connp)
1173 {
1174 if (connp->conn_ipversion == IPV6_VERSION)
1175 return (ipcl_bind_insert_v6(connp));
1176 else
1177 return (ipcl_bind_insert_v4(connp));
1178 }
1179
1180 int
1181 ipcl_bind_insert_v4(conn_t *connp)
1182 {
1183 connf_t *connfp;
1184 int ret = 0;
1185 ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
1186 uint16_t lport = connp->conn_lport;
1187 uint8_t protocol = connp->conn_proto;
1188
1189 if (IPCL_IS_IPTUN(connp))
1190 return (ipcl_iptun_hash_insert(connp, ipst));
1191
1192 switch (protocol) {
1193 default:
1194 if (is_system_labeled() &&
1195 check_exempt_conflict_v4(connp, ipst))
1196 return (EADDRINUSE);
1197 /* FALLTHROUGH */
1198 case IPPROTO_UDP:
1199 if (protocol == IPPROTO_UDP) {
1200 connfp = &ipst->ips_ipcl_udp_fanout[
1201 IPCL_UDP_HASH(lport, ipst)];
1202 } else {
1203 connfp = &ipst->ips_ipcl_proto_fanout_v4[protocol];
1204 }
1205
1206 if (connp->conn_faddr_v4 != INADDR_ANY) {
1207 IPCL_HASH_INSERT_CONNECTED(connfp, connp);
1208 } else if (connp->conn_laddr_v4 != INADDR_ANY) {
1209 IPCL_HASH_INSERT_BOUND(connfp, connp);
1210 } else {
1211 IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1212 }
1213 if (protocol == IPPROTO_RSVP)
1214 ill_set_inputfn_all(ipst);
1215 break;
1216
1217 case IPPROTO_TCP:
1218 /* Insert it in the Bind Hash */
1219 ASSERT(connp->conn_zoneid != ALL_ZONES);
1220 connfp = &ipst->ips_ipcl_bind_fanout[
1221 IPCL_BIND_HASH(lport, ipst)];
1222 if (connp->conn_laddr_v4 != INADDR_ANY) {
1223 IPCL_HASH_INSERT_BOUND(connfp, connp);
1224 } else {
1225 IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1226 }
1227 if (cl_inet_listen != NULL) {
1228 ASSERT(connp->conn_ipversion == IPV4_VERSION);
1229 connp->conn_flags |= IPCL_CL_LISTENER;
1230 (*cl_inet_listen)(
1231 connp->conn_netstack->netstack_stackid,
1232 IPPROTO_TCP, AF_INET,
1233 (uint8_t *)&connp->conn_bound_addr_v4, lport, NULL);
1234 }
1235 break;
1236
1237 case IPPROTO_SCTP:
1238 ret = ipcl_sctp_hash_insert(connp, lport);
1239 break;
1240 }
1241
1242 return (ret);
1243 }
1244
1245 int
1246 ipcl_bind_insert_v6(conn_t *connp)
1247 {
1248 connf_t *connfp;
1249 int ret = 0;
1250 ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
1251 uint16_t lport = connp->conn_lport;
1252 uint8_t protocol = connp->conn_proto;
1253
1254 if (IPCL_IS_IPTUN(connp)) {
1255 return (ipcl_iptun_hash_insert_v6(connp, ipst));
1256 }
1257
1258 switch (protocol) {
1259 default:
1260 if (is_system_labeled() &&
1261 check_exempt_conflict_v6(connp, ipst))
1262 return (EADDRINUSE);
1263 /* FALLTHROUGH */
1264 case IPPROTO_UDP:
1265 if (protocol == IPPROTO_UDP) {
1266 connfp = &ipst->ips_ipcl_udp_fanout[
1267 IPCL_UDP_HASH(lport, ipst)];
1268 } else {
1269 connfp = &ipst->ips_ipcl_proto_fanout_v6[protocol];
1270 }
1271
1272 if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6)) {
1273 IPCL_HASH_INSERT_CONNECTED(connfp, connp);
1274 } else if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6)) {
1275 IPCL_HASH_INSERT_BOUND(connfp, connp);
1276 } else {
1277 IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1278 }
1279 break;
1280
1281 case IPPROTO_TCP:
1282 /* Insert it in the Bind Hash */
1283 ASSERT(connp->conn_zoneid != ALL_ZONES);
1284 connfp = &ipst->ips_ipcl_bind_fanout[
1285 IPCL_BIND_HASH(lport, ipst)];
1286 if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6)) {
1287 IPCL_HASH_INSERT_BOUND(connfp, connp);
1288 } else {
1289 IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1290 }
1291 if (cl_inet_listen != NULL) {
1292 sa_family_t addr_family;
1293 uint8_t *laddrp;
1294
1295 if (connp->conn_ipversion == IPV6_VERSION) {
1296 addr_family = AF_INET6;
1297 laddrp =
1298 (uint8_t *)&connp->conn_bound_addr_v6;
1299 } else {
1300 addr_family = AF_INET;
1301 laddrp = (uint8_t *)&connp->conn_bound_addr_v4;
1302 }
1303 connp->conn_flags |= IPCL_CL_LISTENER;
1304 (*cl_inet_listen)(
1305 connp->conn_netstack->netstack_stackid,
1306 IPPROTO_TCP, addr_family, laddrp, lport, NULL);
1307 }
1308 break;
1309
1310 case IPPROTO_SCTP:
1311 ret = ipcl_sctp_hash_insert(connp, lport);
1312 break;
1313 }
1314
1315 return (ret);
1316 }
1317
1318 /*
1319 * ipcl_conn_hash insertion routines.
1320 * The caller has already set conn_proto and the addresses/ports in the conn_t.
1321 */
1322
1323 int
1324 ipcl_conn_insert(conn_t *connp)
1325 {
1326 if (connp->conn_ipversion == IPV6_VERSION)
1327 return (ipcl_conn_insert_v6(connp));
1328 else
1329 return (ipcl_conn_insert_v4(connp));
1330 }
1331
1332 int
1333 ipcl_conn_insert_v4(conn_t *connp)
1334 {
1335 connf_t *connfp;
1336 conn_t *tconnp;
1337 int ret = 0;
1338 ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
1339 uint16_t lport = connp->conn_lport;
1340 uint8_t protocol = connp->conn_proto;
1341
1342 if (IPCL_IS_IPTUN(connp))
1343 return (ipcl_iptun_hash_insert(connp, ipst));
1344
1345 switch (protocol) {
1346 case IPPROTO_TCP:
1347 /*
1348 * For TCP, we check whether the connection tuple already
1349 * exists before allowing the connection to proceed. We
1350 * also allow indexing on the zoneid. This is to allow
1351 * multiple shared stack zones to have the same tcp
1352 * connection tuple. In practice this only happens for
1353 * INADDR_LOOPBACK as it's the only local address which
1354 * doesn't have to be unique.
1355 */
1356 connfp = &ipst->ips_ipcl_conn_fanout[
1357 IPCL_CONN_HASH(connp->conn_faddr_v4,
1358 connp->conn_ports, ipst)];
1359 mutex_enter(&connfp->connf_lock);
1360 for (tconnp = connfp->connf_head; tconnp != NULL;
1361 tconnp = tconnp->conn_next) {
1362 if (IPCL_CONN_MATCH(tconnp, connp->conn_proto,
1363 connp->conn_faddr_v4, connp->conn_laddr_v4,
1364 connp->conn_ports) &&
1365 IPCL_ZONE_MATCH(tconnp, connp->conn_zoneid)) {
1366 /* Already have a conn. bail out */
1367 mutex_exit(&connfp->connf_lock);
1368 return (EADDRINUSE);
1369 }
1370 }
1371 if (connp->conn_fanout != NULL) {
1372 /*
1373 * Probably a XTI/TLI application trying to do a
1374 * rebind. Let it happen.
1375 */
1376 mutex_exit(&connfp->connf_lock);
1377 IPCL_HASH_REMOVE(connp);
1378 mutex_enter(&connfp->connf_lock);
1379 }
1380
1381 ASSERT(connp->conn_recv != NULL);
1382 ASSERT(connp->conn_recvicmp != NULL);
1383
1384 IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp);
1385 mutex_exit(&connfp->connf_lock);
1386 break;
1387
1388 case IPPROTO_SCTP:
1389 /*
1390 * The raw socket may have already been bound, remove it
1391 * from the hash first.
1392 */
1393 IPCL_HASH_REMOVE(connp);
1394 ret = ipcl_sctp_hash_insert(connp, lport);
1395 break;
1396
1397 default:
1398 /*
1399 * Check for conflicts among MAC exempt bindings. For
1400 * transports with port numbers, this is done by the upper
1401 * level per-transport binding logic. For all others, it's
1402 * done here.
1403 */
1404 if (is_system_labeled() &&
1405 check_exempt_conflict_v4(connp, ipst))
1406 return (EADDRINUSE);
1407 /* FALLTHROUGH */
1408
1409 case IPPROTO_UDP:
1410 if (protocol == IPPROTO_UDP) {
1411 connfp = &ipst->ips_ipcl_udp_fanout[
1412 IPCL_UDP_HASH(lport, ipst)];
1413 } else {
1414 connfp = &ipst->ips_ipcl_proto_fanout_v4[protocol];
1415 }
1416
1417 if (connp->conn_faddr_v4 != INADDR_ANY) {
1418 IPCL_HASH_INSERT_CONNECTED(connfp, connp);
1419 } else if (connp->conn_laddr_v4 != INADDR_ANY) {
1420 IPCL_HASH_INSERT_BOUND(connfp, connp);
1421 } else {
1422 IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1423 }
1424 break;
1425 }
1426
1427 return (ret);
1428 }
1429
1430 int
1431 ipcl_conn_insert_v6(conn_t *connp)
1432 {
1433 connf_t *connfp;
1434 conn_t *tconnp;
1435 int ret = 0;
1436 ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
1437 uint16_t lport = connp->conn_lport;
1438 uint8_t protocol = connp->conn_proto;
1439 uint_t ifindex = connp->conn_bound_if;
1440
1441 if (IPCL_IS_IPTUN(connp))
1442 return (ipcl_iptun_hash_insert_v6(connp, ipst));
1443
1444 switch (protocol) {
1445 case IPPROTO_TCP:
1446
1447 /*
1448 * For tcp, we check whether the connection tuple already
1449 * exists before allowing the connection to proceed. We
1450 * also allow indexing on the zoneid. This is to allow
1451 * multiple shared stack zones to have the same tcp
1452 * connection tuple. In practice this only happens for
1453 * ipv6_loopback as it's the only local address which
1454 * doesn't have to be unique.
1455 */
1456 connfp = &ipst->ips_ipcl_conn_fanout[
1457 IPCL_CONN_HASH_V6(connp->conn_faddr_v6, connp->conn_ports,
1458 ipst)];
1459 mutex_enter(&connfp->connf_lock);
1460 for (tconnp = connfp->connf_head; tconnp != NULL;
1461 tconnp = tconnp->conn_next) {
1462 /* NOTE: need to match zoneid. Bug in onnv-gate */
1463 if (IPCL_CONN_MATCH_V6(tconnp, connp->conn_proto,
1464 connp->conn_faddr_v6, connp->conn_laddr_v6,
1465 connp->conn_ports) &&
1466 (tconnp->conn_bound_if == 0 ||
1467 tconnp->conn_bound_if == ifindex) &&
1468 IPCL_ZONE_MATCH(tconnp, connp->conn_zoneid)) {
1469 /* Already have a conn. bail out */
1470 mutex_exit(&connfp->connf_lock);
1471 return (EADDRINUSE);
1472 }
1473 }
1474 if (connp->conn_fanout != NULL) {
1475 /*
1476 * Probably a XTI/TLI application trying to do a
1477 * rebind. Let it happen.
1478 */
1479 mutex_exit(&connfp->connf_lock);
1480 IPCL_HASH_REMOVE(connp);
1481 mutex_enter(&connfp->connf_lock);
1482 }
1483 IPCL_HASH_INSERT_CONNECTED_LOCKED(connfp, connp);
1484 mutex_exit(&connfp->connf_lock);
1485 break;
1486
1487 case IPPROTO_SCTP:
1488 IPCL_HASH_REMOVE(connp);
1489 ret = ipcl_sctp_hash_insert(connp, lport);
1490 break;
1491
1492 default:
1493 if (is_system_labeled() &&
1494 check_exempt_conflict_v6(connp, ipst))
1495 return (EADDRINUSE);
1496 /* FALLTHROUGH */
1497 case IPPROTO_UDP:
1498 if (protocol == IPPROTO_UDP) {
1499 connfp = &ipst->ips_ipcl_udp_fanout[
1500 IPCL_UDP_HASH(lport, ipst)];
1501 } else {
1502 connfp = &ipst->ips_ipcl_proto_fanout_v6[protocol];
1503 }
1504
1505 if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6)) {
1506 IPCL_HASH_INSERT_CONNECTED(connfp, connp);
1507 } else if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_laddr_v6)) {
1508 IPCL_HASH_INSERT_BOUND(connfp, connp);
1509 } else {
1510 IPCL_HASH_INSERT_WILDCARD(connfp, connp);
1511 }
1512 break;
1513 }
1514
1515 return (ret);
1516 }
1517
1518 /*
1519 * v4 packet classifying function. looks up the fanout table to
1520 * find the conn, the packet belongs to. returns the conn with
1521 * the reference held, null otherwise.
1522 *
1523 * If zoneid is ALL_ZONES, then the search rules described in the "Connection
1524 * Lookup" comment block are applied. Labels are also checked as described
1525 * above. If the packet is from the inside (looped back), and is from the same
1526 * zone, then label checks are omitted.
1527 */
1528 conn_t *
1529 ipcl_classify_v4(mblk_t *mp, uint8_t protocol, uint_t hdr_len,
1530 ip_recv_attr_t *ira, ip_stack_t *ipst)
1531 {
1532 ipha_t *ipha;
1533 connf_t *connfp, *bind_connfp;
1534 uint16_t lport;
1535 uint16_t fport;
1536 uint32_t ports;
1537 conn_t *connp;
1538 uint16_t *up;
1539 zoneid_t zoneid = ira->ira_zoneid;
1540
1541 ipha = (ipha_t *)mp->b_rptr;
1542 up = (uint16_t *)((uchar_t *)ipha + hdr_len + TCP_PORTS_OFFSET);
1543
1544 switch (protocol) {
1545 case IPPROTO_TCP:
1546 ports = *(uint32_t *)up;
1547 connfp =
1548 &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH(ipha->ipha_src,
1549 ports, ipst)];
1550 mutex_enter(&connfp->connf_lock);
1551 for (connp = connfp->connf_head; connp != NULL;
1552 connp = connp->conn_next) {
1553 if (IPCL_CONN_MATCH(connp, protocol,
1554 ipha->ipha_src, ipha->ipha_dst, ports) &&
1555 (connp->conn_zoneid == zoneid ||
1556 connp->conn_allzones ||
1557 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) &&
1558 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) &&
1559 (ira->ira_flags & IRAF_TX_SHARED_ADDR))))
1560 break;
1561 }
1562
1563 if (connp != NULL) {
1564 /*
1565 * We have a fully-bound TCP connection.
1566 *
1567 * For labeled systems, there's no need to check the
1568 * label here. It's known to be good as we checked
1569 * before allowing the connection to become bound.
1570 */
1571 CONN_INC_REF(connp);
1572 mutex_exit(&connfp->connf_lock);
1573 return (connp);
1574 }
1575
1576 mutex_exit(&connfp->connf_lock);
1577 lport = up[1];
1578 bind_connfp =
1579 &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)];
1580 mutex_enter(&bind_connfp->connf_lock);
1581 for (connp = bind_connfp->connf_head; connp != NULL;
1582 connp = connp->conn_next) {
1583 if (IPCL_BIND_MATCH(connp, protocol, ipha->ipha_dst,
1584 lport) &&
1585 (connp->conn_zoneid == zoneid ||
1586 connp->conn_allzones ||
1587 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) &&
1588 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) &&
1589 (ira->ira_flags & IRAF_TX_SHARED_ADDR))))
1590 break;
1591 }
1592
1593 /*
1594 * If the matching connection is SLP on a private address, then
1595 * the label on the packet must match the local zone's label.
1596 * Otherwise, it must be in the label range defined by tnrh.
1597 * This is ensured by tsol_receive_local.
1598 *
1599 * Note that we don't check tsol_receive_local for
1600 * the connected case.
1601 */
1602 if (connp != NULL && (ira->ira_flags & IRAF_SYSTEM_LABELED) &&
1603 !tsol_receive_local(mp, &ipha->ipha_dst, IPV4_VERSION,
1604 ira, connp)) {
1605 DTRACE_PROBE3(tx__ip__log__info__classify__tcp,
1606 char *, "connp(1) could not receive mp(2)",
1607 conn_t *, connp, mblk_t *, mp);
1608 connp = NULL;
1609 }
1610
1611 if (connp != NULL) {
1612 /* Have a listener at least */
1613 CONN_INC_REF(connp);
1614 mutex_exit(&bind_connfp->connf_lock);
1615 return (connp);
1616 }
1617
1618 mutex_exit(&bind_connfp->connf_lock);
1619 break;
1620
1621 case IPPROTO_UDP:
1622 lport = up[1];
1623 fport = up[0];
1624 connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
1625 mutex_enter(&connfp->connf_lock);
1626 for (connp = connfp->connf_head; connp != NULL;
1627 connp = connp->conn_next) {
1628 if (IPCL_UDP_MATCH(connp, lport, ipha->ipha_dst,
1629 fport, ipha->ipha_src) &&
1630 (connp->conn_zoneid == zoneid ||
1631 connp->conn_allzones ||
1632 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) &&
1633 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE))))
1634 break;
1635 }
1636
1637 if (connp != NULL && (ira->ira_flags & IRAF_SYSTEM_LABELED) &&
1638 !tsol_receive_local(mp, &ipha->ipha_dst, IPV4_VERSION,
1639 ira, connp)) {
1640 DTRACE_PROBE3(tx__ip__log__info__classify__udp,
1641 char *, "connp(1) could not receive mp(2)",
1642 conn_t *, connp, mblk_t *, mp);
1643 connp = NULL;
1644 }
1645
1646 if (connp != NULL) {
1647 CONN_INC_REF(connp);
1648 mutex_exit(&connfp->connf_lock);
1649 return (connp);
1650 }
1651
1652 /*
1653 * We shouldn't come here for multicast/broadcast packets
1654 */
1655 mutex_exit(&connfp->connf_lock);
1656
1657 break;
1658
1659 case IPPROTO_ENCAP:
1660 case IPPROTO_IPV6:
1661 return (ipcl_iptun_classify_v4(&ipha->ipha_src,
1662 &ipha->ipha_dst, ipst));
1663 }
1664
1665 return (NULL);
1666 }
1667
1668 conn_t *
1669 ipcl_classify_v6(mblk_t *mp, uint8_t protocol, uint_t hdr_len,
1670 ip_recv_attr_t *ira, ip_stack_t *ipst)
1671 {
1672 ip6_t *ip6h;
1673 connf_t *connfp, *bind_connfp;
1674 uint16_t lport;
1675 uint16_t fport;
1676 tcpha_t *tcpha;
1677 uint32_t ports;
1678 conn_t *connp;
1679 uint16_t *up;
1680 zoneid_t zoneid = ira->ira_zoneid;
1681
1682 ip6h = (ip6_t *)mp->b_rptr;
1683
1684 switch (protocol) {
1685 case IPPROTO_TCP:
1686 tcpha = (tcpha_t *)&mp->b_rptr[hdr_len];
1687 up = &tcpha->tha_lport;
1688 ports = *(uint32_t *)up;
1689
1690 connfp =
1691 &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH_V6(ip6h->ip6_src,
1692 ports, ipst)];
1693 mutex_enter(&connfp->connf_lock);
1694 for (connp = connfp->connf_head; connp != NULL;
1695 connp = connp->conn_next) {
1696 if (IPCL_CONN_MATCH_V6(connp, protocol,
1697 ip6h->ip6_src, ip6h->ip6_dst, ports) &&
1698 (connp->conn_zoneid == zoneid ||
1699 connp->conn_allzones ||
1700 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) &&
1701 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) &&
1702 (ira->ira_flags & IRAF_TX_SHARED_ADDR))))
1703 break;
1704 }
1705
1706 if (connp != NULL) {
1707 /*
1708 * We have a fully-bound TCP connection.
1709 *
1710 * For labeled systems, there's no need to check the
1711 * label here. It's known to be good as we checked
1712 * before allowing the connection to become bound.
1713 */
1714 CONN_INC_REF(connp);
1715 mutex_exit(&connfp->connf_lock);
1716 return (connp);
1717 }
1718
1719 mutex_exit(&connfp->connf_lock);
1720
1721 lport = up[1];
1722 bind_connfp =
1723 &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)];
1724 mutex_enter(&bind_connfp->connf_lock);
1725 for (connp = bind_connfp->connf_head; connp != NULL;
1726 connp = connp->conn_next) {
1727 if (IPCL_BIND_MATCH_V6(connp, protocol,
1728 ip6h->ip6_dst, lport) &&
1729 (connp->conn_zoneid == zoneid ||
1730 connp->conn_allzones ||
1731 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) &&
1732 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) &&
1733 (ira->ira_flags & IRAF_TX_SHARED_ADDR))))
1734 break;
1735 }
1736
1737 if (connp != NULL && (ira->ira_flags & IRAF_SYSTEM_LABELED) &&
1738 !tsol_receive_local(mp, &ip6h->ip6_dst, IPV6_VERSION,
1739 ira, connp)) {
1740 DTRACE_PROBE3(tx__ip__log__info__classify__tcp6,
1741 char *, "connp(1) could not receive mp(2)",
1742 conn_t *, connp, mblk_t *, mp);
1743 connp = NULL;
1744 }
1745
1746 if (connp != NULL) {
1747 /* Have a listner at least */
1748 CONN_INC_REF(connp);
1749 mutex_exit(&bind_connfp->connf_lock);
1750 return (connp);
1751 }
1752
1753 mutex_exit(&bind_connfp->connf_lock);
1754 break;
1755
1756 case IPPROTO_UDP:
1757 up = (uint16_t *)&mp->b_rptr[hdr_len];
1758 lport = up[1];
1759 fport = up[0];
1760 connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
1761 mutex_enter(&connfp->connf_lock);
1762 for (connp = connfp->connf_head; connp != NULL;
1763 connp = connp->conn_next) {
1764 if (IPCL_UDP_MATCH_V6(connp, lport, ip6h->ip6_dst,
1765 fport, ip6h->ip6_src) &&
1766 (connp->conn_zoneid == zoneid ||
1767 connp->conn_allzones ||
1768 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) &&
1769 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) &&
1770 (ira->ira_flags & IRAF_TX_SHARED_ADDR))))
1771 break;
1772 }
1773
1774 if (connp != NULL && (ira->ira_flags & IRAF_SYSTEM_LABELED) &&
1775 !tsol_receive_local(mp, &ip6h->ip6_dst, IPV6_VERSION,
1776 ira, connp)) {
1777 DTRACE_PROBE3(tx__ip__log__info__classify__udp6,
1778 char *, "connp(1) could not receive mp(2)",
1779 conn_t *, connp, mblk_t *, mp);
1780 connp = NULL;
1781 }
1782
1783 if (connp != NULL) {
1784 CONN_INC_REF(connp);
1785 mutex_exit(&connfp->connf_lock);
1786 return (connp);
1787 }
1788
1789 /*
1790 * We shouldn't come here for multicast/broadcast packets
1791 */
1792 mutex_exit(&connfp->connf_lock);
1793 break;
1794 case IPPROTO_ENCAP:
1795 case IPPROTO_IPV6:
1796 return (ipcl_iptun_classify_v6(&ip6h->ip6_src,
1797 &ip6h->ip6_dst, ipst));
1798 }
1799
1800 return (NULL);
1801 }
1802
1803 /*
1804 * wrapper around ipcl_classify_(v4,v6) routines.
1805 */
1806 conn_t *
1807 ipcl_classify(mblk_t *mp, ip_recv_attr_t *ira, ip_stack_t *ipst)
1808 {
1809 if (ira->ira_flags & IRAF_IS_IPV4) {
1810 return (ipcl_classify_v4(mp, ira->ira_protocol,
1811 ira->ira_ip_hdr_length, ira, ipst));
1812 } else {
1813 return (ipcl_classify_v6(mp, ira->ira_protocol,
1814 ira->ira_ip_hdr_length, ira, ipst));
1815 }
1816 }
1817
1818 /*
1819 * Only used to classify SCTP RAW sockets
1820 */
1821 conn_t *
1822 ipcl_classify_raw(mblk_t *mp, uint8_t protocol, uint32_t ports,
1823 ipha_t *ipha, ip6_t *ip6h, ip_recv_attr_t *ira, ip_stack_t *ipst)
1824 {
1825 connf_t *connfp;
1826 conn_t *connp;
1827 in_port_t lport;
1828 int ipversion;
1829 const void *dst;
1830 zoneid_t zoneid = ira->ira_zoneid;
1831
1832 lport = ((uint16_t *)&ports)[1];
1833 if (ira->ira_flags & IRAF_IS_IPV4) {
1834 dst = (const void *)&ipha->ipha_dst;
1835 ipversion = IPV4_VERSION;
1836 } else {
1837 dst = (const void *)&ip6h->ip6_dst;
1838 ipversion = IPV6_VERSION;
1839 }
1840
1841 connfp = &ipst->ips_ipcl_raw_fanout[IPCL_RAW_HASH(ntohs(lport), ipst)];
1842 mutex_enter(&connfp->connf_lock);
1843 for (connp = connfp->connf_head; connp != NULL;
1844 connp = connp->conn_next) {
1845 /* We don't allow v4 fallback for v6 raw socket. */
1846 if (ipversion != connp->conn_ipversion)
1847 continue;
1848 if (!IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6) &&
1849 !IN6_IS_ADDR_V4MAPPED_ANY(&connp->conn_faddr_v6)) {
1850 if (ipversion == IPV4_VERSION) {
1851 if (!IPCL_CONN_MATCH(connp, protocol,
1852 ipha->ipha_src, ipha->ipha_dst, ports))
1853 continue;
1854 } else {
1855 if (!IPCL_CONN_MATCH_V6(connp, protocol,
1856 ip6h->ip6_src, ip6h->ip6_dst, ports))
1857 continue;
1858 }
1859 } else {
1860 if (ipversion == IPV4_VERSION) {
1861 if (!IPCL_BIND_MATCH(connp, protocol,
1862 ipha->ipha_dst, lport))
1863 continue;
1864 } else {
1865 if (!IPCL_BIND_MATCH_V6(connp, protocol,
1866 ip6h->ip6_dst, lport))
1867 continue;
1868 }
1869 }
1870
1871 if (connp->conn_zoneid == zoneid ||
1872 connp->conn_allzones ||
1873 ((connp->conn_mac_mode != CONN_MAC_DEFAULT) &&
1874 (ira->ira_flags & IRAF_TX_MAC_EXEMPTABLE) &&
1875 (ira->ira_flags & IRAF_TX_SHARED_ADDR)))
1876 break;
1877 }
1878
1879 if (connp != NULL && (ira->ira_flags & IRAF_SYSTEM_LABELED) &&
1880 !tsol_receive_local(mp, dst, ipversion, ira, connp)) {
1881 DTRACE_PROBE3(tx__ip__log__info__classify__rawip,
1882 char *, "connp(1) could not receive mp(2)",
1883 conn_t *, connp, mblk_t *, mp);
1884 connp = NULL;
1885 }
1886
1887 if (connp != NULL)
1888 goto found;
1889 mutex_exit(&connfp->connf_lock);
1890
1891 /* Try to look for a wildcard SCTP RAW socket match. */
1892 connfp = &ipst->ips_ipcl_raw_fanout[IPCL_RAW_HASH(0, ipst)];
1893 mutex_enter(&connfp->connf_lock);
1894 for (connp = connfp->connf_head; connp != NULL;
1895 connp = connp->conn_next) {
1896 /* We don't allow v4 fallback for v6 raw socket. */
1897 if (ipversion != connp->conn_ipversion)
1898 continue;
1899 if (!IPCL_ZONE_MATCH(connp, zoneid))
1900 continue;
1901
1902 if (ipversion == IPV4_VERSION) {
1903 if (IPCL_RAW_MATCH(connp, protocol, ipha->ipha_dst))
1904 break;
1905 } else {
1906 if (IPCL_RAW_MATCH_V6(connp, protocol, ip6h->ip6_dst)) {
1907 break;
1908 }
1909 }
1910 }
1911
1912 if (connp != NULL)
1913 goto found;
1914
1915 mutex_exit(&connfp->connf_lock);
1916 return (NULL);
1917
1918 found:
1919 ASSERT(connp != NULL);
1920 CONN_INC_REF(connp);
1921 mutex_exit(&connfp->connf_lock);
1922 return (connp);
1923 }
1924
1925 /* ARGSUSED */
1926 static int
1927 tcp_conn_constructor(void *buf, void *cdrarg, int kmflags)
1928 {
1929 itc_t *itc = (itc_t *)buf;
1930 conn_t *connp = &itc->itc_conn;
1931 tcp_t *tcp = (tcp_t *)&itc[1];
1932
1933 bzero(connp, sizeof (conn_t));
1934 bzero(tcp, sizeof (tcp_t));
1935
1936 mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL);
1937 cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL);
1938 cv_init(&connp->conn_sq_cv, NULL, CV_DEFAULT, NULL);
1939 tcp->tcp_timercache = tcp_timermp_alloc(kmflags);
1940 if (tcp->tcp_timercache == NULL)
1941 return (ENOMEM);
1942 connp->conn_tcp = tcp;
1943 connp->conn_flags = IPCL_TCPCONN;
1944 connp->conn_proto = IPPROTO_TCP;
1945 tcp->tcp_connp = connp;
1946 rw_init(&connp->conn_ilg_lock, NULL, RW_DEFAULT, NULL);
1947
1948 connp->conn_ixa = kmem_zalloc(sizeof (ip_xmit_attr_t), kmflags);
1949 if (connp->conn_ixa == NULL) {
1950 tcp_timermp_free(tcp);
1951 return (ENOMEM);
1952 }
1953 connp->conn_ixa->ixa_refcnt = 1;
1954 connp->conn_ixa->ixa_protocol = connp->conn_proto;
1955 connp->conn_ixa->ixa_xmit_hint = CONN_TO_XMIT_HINT(connp);
1956 return (0);
1957 }
1958
1959 /* ARGSUSED */
1960 static void
1961 tcp_conn_destructor(void *buf, void *cdrarg)
1962 {
1963 itc_t *itc = (itc_t *)buf;
1964 conn_t *connp = &itc->itc_conn;
1965 tcp_t *tcp = (tcp_t *)&itc[1];
1966
1967 ASSERT(connp->conn_flags & IPCL_TCPCONN);
1968 ASSERT(tcp->tcp_connp == connp);
1969 ASSERT(connp->conn_tcp == tcp);
1970 tcp_timermp_free(tcp);
1971 mutex_destroy(&connp->conn_lock);
1972 cv_destroy(&connp->conn_cv);
1973 cv_destroy(&connp->conn_sq_cv);
1974 rw_destroy(&connp->conn_ilg_lock);
1975
1976 /* Can be NULL if constructor failed */
1977 if (connp->conn_ixa != NULL) {
1978 ASSERT(connp->conn_ixa->ixa_refcnt == 1);
1979 ASSERT(connp->conn_ixa->ixa_ire == NULL);
1980 ASSERT(connp->conn_ixa->ixa_nce == NULL);
1981 ixa_refrele(connp->conn_ixa);
1982 }
1983 }
1984
1985 /* ARGSUSED */
1986 static int
1987 ip_conn_constructor(void *buf, void *cdrarg, int kmflags)
1988 {
1989 itc_t *itc = (itc_t *)buf;
1990 conn_t *connp = &itc->itc_conn;
1991
1992 bzero(connp, sizeof (conn_t));
1993 mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL);
1994 cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL);
1995 connp->conn_flags = IPCL_IPCCONN;
1996 rw_init(&connp->conn_ilg_lock, NULL, RW_DEFAULT, NULL);
1997
1998 connp->conn_ixa = kmem_zalloc(sizeof (ip_xmit_attr_t), kmflags);
1999 if (connp->conn_ixa == NULL)
2000 return (ENOMEM);
2001 connp->conn_ixa->ixa_refcnt = 1;
2002 connp->conn_ixa->ixa_xmit_hint = CONN_TO_XMIT_HINT(connp);
2003 return (0);
2004 }
2005
2006 /* ARGSUSED */
2007 static void
2008 ip_conn_destructor(void *buf, void *cdrarg)
2009 {
2010 itc_t *itc = (itc_t *)buf;
2011 conn_t *connp = &itc->itc_conn;
2012
2013 ASSERT(connp->conn_flags & IPCL_IPCCONN);
2014 ASSERT(connp->conn_priv == NULL);
2015 mutex_destroy(&connp->conn_lock);
2016 cv_destroy(&connp->conn_cv);
2017 rw_destroy(&connp->conn_ilg_lock);
2018
2019 /* Can be NULL if constructor failed */
2020 if (connp->conn_ixa != NULL) {
2021 ASSERT(connp->conn_ixa->ixa_refcnt == 1);
2022 ASSERT(connp->conn_ixa->ixa_ire == NULL);
2023 ASSERT(connp->conn_ixa->ixa_nce == NULL);
2024 ixa_refrele(connp->conn_ixa);
2025 }
2026 }
2027
2028 /* ARGSUSED */
2029 static int
2030 udp_conn_constructor(void *buf, void *cdrarg, int kmflags)
2031 {
2032 itc_t *itc = (itc_t *)buf;
2033 conn_t *connp = &itc->itc_conn;
2034 udp_t *udp = (udp_t *)&itc[1];
2035
2036 bzero(connp, sizeof (conn_t));
2037 bzero(udp, sizeof (udp_t));
2038
2039 mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL);
2040 cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL);
2041 connp->conn_udp = udp;
2042 connp->conn_flags = IPCL_UDPCONN;
2043 connp->conn_proto = IPPROTO_UDP;
2044 udp->udp_connp = connp;
2045 rw_init(&connp->conn_ilg_lock, NULL, RW_DEFAULT, NULL);
2046 connp->conn_ixa = kmem_zalloc(sizeof (ip_xmit_attr_t), kmflags);
2047 if (connp->conn_ixa == NULL)
2048 return (ENOMEM);
2049 connp->conn_ixa->ixa_refcnt = 1;
2050 connp->conn_ixa->ixa_protocol = connp->conn_proto;
2051 connp->conn_ixa->ixa_xmit_hint = CONN_TO_XMIT_HINT(connp);
2052 return (0);
2053 }
2054
2055 /* ARGSUSED */
2056 static void
2057 udp_conn_destructor(void *buf, void *cdrarg)
2058 {
2059 itc_t *itc = (itc_t *)buf;
2060 conn_t *connp = &itc->itc_conn;
2061 udp_t *udp = (udp_t *)&itc[1];
2062
2063 ASSERT(connp->conn_flags & IPCL_UDPCONN);
2064 ASSERT(udp->udp_connp == connp);
2065 ASSERT(connp->conn_udp == udp);
2066 mutex_destroy(&connp->conn_lock);
2067 cv_destroy(&connp->conn_cv);
2068 rw_destroy(&connp->conn_ilg_lock);
2069
2070 /* Can be NULL if constructor failed */
2071 if (connp->conn_ixa != NULL) {
2072 ASSERT(connp->conn_ixa->ixa_refcnt == 1);
2073 ASSERT(connp->conn_ixa->ixa_ire == NULL);
2074 ASSERT(connp->conn_ixa->ixa_nce == NULL);
2075 ixa_refrele(connp->conn_ixa);
2076 }
2077 }
2078
2079 /* ARGSUSED */
2080 static int
2081 rawip_conn_constructor(void *buf, void *cdrarg, int kmflags)
2082 {
2083 itc_t *itc = (itc_t *)buf;
2084 conn_t *connp = &itc->itc_conn;
2085 icmp_t *icmp = (icmp_t *)&itc[1];
2086
2087 bzero(connp, sizeof (conn_t));
2088 bzero(icmp, sizeof (icmp_t));
2089
2090 mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL);
2091 cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL);
2092 connp->conn_icmp = icmp;
2093 connp->conn_flags = IPCL_RAWIPCONN;
2094 connp->conn_proto = IPPROTO_ICMP;
2095 icmp->icmp_connp = connp;
2096 rw_init(&connp->conn_ilg_lock, NULL, RW_DEFAULT, NULL);
2097 connp->conn_ixa = kmem_zalloc(sizeof (ip_xmit_attr_t), kmflags);
2098 if (connp->conn_ixa == NULL)
2099 return (ENOMEM);
2100 connp->conn_ixa->ixa_refcnt = 1;
2101 connp->conn_ixa->ixa_protocol = connp->conn_proto;
2102 connp->conn_ixa->ixa_xmit_hint = CONN_TO_XMIT_HINT(connp);
2103 return (0);
2104 }
2105
2106 /* ARGSUSED */
2107 static void
2108 rawip_conn_destructor(void *buf, void *cdrarg)
2109 {
2110 itc_t *itc = (itc_t *)buf;
2111 conn_t *connp = &itc->itc_conn;
2112 icmp_t *icmp = (icmp_t *)&itc[1];
2113
2114 ASSERT(connp->conn_flags & IPCL_RAWIPCONN);
2115 ASSERT(icmp->icmp_connp == connp);
2116 ASSERT(connp->conn_icmp == icmp);
2117 mutex_destroy(&connp->conn_lock);
2118 cv_destroy(&connp->conn_cv);
2119 rw_destroy(&connp->conn_ilg_lock);
2120
2121 /* Can be NULL if constructor failed */
2122 if (connp->conn_ixa != NULL) {
2123 ASSERT(connp->conn_ixa->ixa_refcnt == 1);
2124 ASSERT(connp->conn_ixa->ixa_ire == NULL);
2125 ASSERT(connp->conn_ixa->ixa_nce == NULL);
2126 ixa_refrele(connp->conn_ixa);
2127 }
2128 }
2129
2130 /* ARGSUSED */
2131 static int
2132 rts_conn_constructor(void *buf, void *cdrarg, int kmflags)
2133 {
2134 itc_t *itc = (itc_t *)buf;
2135 conn_t *connp = &itc->itc_conn;
2136 rts_t *rts = (rts_t *)&itc[1];
2137
2138 bzero(connp, sizeof (conn_t));
2139 bzero(rts, sizeof (rts_t));
2140
2141 mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL);
2142 cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL);
2143 connp->conn_rts = rts;
2144 connp->conn_flags = IPCL_RTSCONN;
2145 rts->rts_connp = connp;
2146 rw_init(&connp->conn_ilg_lock, NULL, RW_DEFAULT, NULL);
2147 connp->conn_ixa = kmem_zalloc(sizeof (ip_xmit_attr_t), kmflags);
2148 if (connp->conn_ixa == NULL)
2149 return (ENOMEM);
2150 connp->conn_ixa->ixa_refcnt = 1;
2151 connp->conn_ixa->ixa_xmit_hint = CONN_TO_XMIT_HINT(connp);
2152 return (0);
2153 }
2154
2155 /* ARGSUSED */
2156 static void
2157 rts_conn_destructor(void *buf, void *cdrarg)
2158 {
2159 itc_t *itc = (itc_t *)buf;
2160 conn_t *connp = &itc->itc_conn;
2161 rts_t *rts = (rts_t *)&itc[1];
2162
2163 ASSERT(connp->conn_flags & IPCL_RTSCONN);
2164 ASSERT(rts->rts_connp == connp);
2165 ASSERT(connp->conn_rts == rts);
2166 mutex_destroy(&connp->conn_lock);
2167 cv_destroy(&connp->conn_cv);
2168 rw_destroy(&connp->conn_ilg_lock);
2169
2170 /* Can be NULL if constructor failed */
2171 if (connp->conn_ixa != NULL) {
2172 ASSERT(connp->conn_ixa->ixa_refcnt == 1);
2173 ASSERT(connp->conn_ixa->ixa_ire == NULL);
2174 ASSERT(connp->conn_ixa->ixa_nce == NULL);
2175 ixa_refrele(connp->conn_ixa);
2176 }
2177 }
2178
2179 /*
2180 * Called as part of ipcl_conn_destroy to assert and clear any pointers
2181 * in the conn_t.
2182 *
2183 * Below we list all the pointers in the conn_t as a documentation aid.
2184 * The ones that we can not ASSERT to be NULL are #ifdef'ed out.
2185 * If you add any pointers to the conn_t please add an ASSERT here
2186 * and #ifdef it out if it can't be actually asserted to be NULL.
2187 * In any case, we bzero most of the conn_t at the end of the function.
2188 */
2189 void
2190 ipcl_conn_cleanup(conn_t *connp)
2191 {
2192 ip_xmit_attr_t *ixa;
2193
2194 ASSERT(connp->conn_latch == NULL);
2195 ASSERT(connp->conn_latch_in_policy == NULL);
2196 ASSERT(connp->conn_latch_in_action == NULL);
2197 #ifdef notdef
2198 ASSERT(connp->conn_rq == NULL);
2199 ASSERT(connp->conn_wq == NULL);
2200 #endif
2201 ASSERT(connp->conn_cred == NULL);
2202 ASSERT(connp->conn_g_fanout == NULL);
2203 ASSERT(connp->conn_g_next == NULL);
2204 ASSERT(connp->conn_g_prev == NULL);
2205 ASSERT(connp->conn_policy == NULL);
2206 ASSERT(connp->conn_fanout == NULL);
2207 ASSERT(connp->conn_next == NULL);
2208 ASSERT(connp->conn_prev == NULL);
2209 ASSERT(connp->conn_oper_pending_ill == NULL);
2210 ASSERT(connp->conn_ilg == NULL);
2211 ASSERT(connp->conn_drain_next == NULL);
2212 ASSERT(connp->conn_drain_prev == NULL);
2213 #ifdef notdef
2214 /* conn_idl is not cleared when removed from idl list */
2215 ASSERT(connp->conn_idl == NULL);
2216 #endif
2217 ASSERT(connp->conn_ipsec_opt_mp == NULL);
2218 #ifdef notdef
2219 /* conn_netstack is cleared by the caller; needed by ixa_cleanup */
2220 ASSERT(connp->conn_netstack == NULL);
2221 #endif
2222
2223 ASSERT(connp->conn_helper_info == NULL);
2224 ASSERT(connp->conn_ixa != NULL);
2225 ixa = connp->conn_ixa;
2226 ASSERT(ixa->ixa_refcnt == 1);
2227 /* Need to preserve ixa_protocol */
2228 ixa_cleanup(ixa);
2229 ixa->ixa_flags = 0;
2230
2231 /* Clear out the conn_t fields that are not preserved */
2232 bzero(&connp->conn_start_clr,
2233 sizeof (conn_t) -
2234 ((uchar_t *)&connp->conn_start_clr - (uchar_t *)connp));
2235 }
2236
2237 /*
2238 * All conns are inserted in a global multi-list for the benefit of
2239 * walkers. The walk is guaranteed to walk all open conns at the time
2240 * of the start of the walk exactly once. This property is needed to
2241 * achieve some cleanups during unplumb of interfaces. This is achieved
2242 * as follows.
2243 *
2244 * ipcl_conn_create and ipcl_conn_destroy are the only functions that
2245 * call the insert and delete functions below at creation and deletion
2246 * time respectively. The conn never moves or changes its position in this
2247 * multi-list during its lifetime. CONN_CONDEMNED ensures that the refcnt
2248 * won't increase due to walkers, once the conn deletion has started. Note
2249 * that we can't remove the conn from the global list and then wait for
2250 * the refcnt to drop to zero, since walkers would then see a truncated
2251 * list. CONN_INCIPIENT ensures that walkers don't start looking at
2252 * conns until ip_open is ready to make them globally visible.
2253 * The global round robin multi-list locks are held only to get the
2254 * next member/insertion/deletion and contention should be negligible
2255 * if the multi-list is much greater than the number of cpus.
2256 */
2257 void
2258 ipcl_globalhash_insert(conn_t *connp)
2259 {
2260 int index;
2261 struct connf_s *connfp;
2262 ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
2263
2264 /*
2265 * No need for atomic here. Approximate even distribution
2266 * in the global lists is sufficient.
2267 */
2268 ipst->ips_conn_g_index++;
2269 index = ipst->ips_conn_g_index & (CONN_G_HASH_SIZE - 1);
2270
2271 connp->conn_g_prev = NULL;
2272 /*
2273 * Mark as INCIPIENT, so that walkers will ignore this
2274 * for now, till ip_open is ready to make it visible globally.
2275 */
2276 connp->conn_state_flags |= CONN_INCIPIENT;
2277
2278 connfp = &ipst->ips_ipcl_globalhash_fanout[index];
2279 /* Insert at the head of the list */
2280 mutex_enter(&connfp->connf_lock);
2281 connp->conn_g_next = connfp->connf_head;
2282 if (connp->conn_g_next != NULL)
2283 connp->conn_g_next->conn_g_prev = connp;
2284 connfp->connf_head = connp;
2285
2286 /* The fanout bucket this conn points to */
2287 connp->conn_g_fanout = connfp;
2288
2289 mutex_exit(&connfp->connf_lock);
2290 }
2291
2292 void
2293 ipcl_globalhash_remove(conn_t *connp)
2294 {
2295 struct connf_s *connfp;
2296
2297 /*
2298 * We were never inserted in the global multi list.
2299 * IPCL_NONE variety is never inserted in the global multilist
2300 * since it is presumed to not need any cleanup and is transient.
2301 */
2302 if (connp->conn_g_fanout == NULL)
2303 return;
2304
2305 connfp = connp->conn_g_fanout;
2306 mutex_enter(&connfp->connf_lock);
2307 if (connp->conn_g_prev != NULL)
2308 connp->conn_g_prev->conn_g_next = connp->conn_g_next;
2309 else
2310 connfp->connf_head = connp->conn_g_next;
2311 if (connp->conn_g_next != NULL)
2312 connp->conn_g_next->conn_g_prev = connp->conn_g_prev;
2313 mutex_exit(&connfp->connf_lock);
2314
2315 /* Better to stumble on a null pointer than to corrupt memory */
2316 connp->conn_g_next = NULL;
2317 connp->conn_g_prev = NULL;
2318 connp->conn_g_fanout = NULL;
2319 }
2320
2321 /*
2322 * Walk the list of all conn_t's in the system, calling the function provided
2323 * With the specified argument for each.
2324 * Applies to both IPv4 and IPv6.
2325 *
2326 * CONNs may hold pointers to ills (conn_dhcpinit_ill and
2327 * conn_oper_pending_ill). To guard against stale pointers
2328 * ipcl_walk() is called to cleanup the conn_t's, typically when an interface is
2329 * unplumbed or removed. New conn_t's that are created while we are walking
2330 * may be missed by this walk, because they are not necessarily inserted
2331 * at the tail of the list. They are new conn_t's and thus don't have any
2332 * stale pointers. The CONN_CLOSING flag ensures that no new reference
2333 * is created to the struct that is going away.
2334 */
2335 void
2336 ipcl_walk(pfv_t func, void *arg, ip_stack_t *ipst)
2337 {
2338 int i;
2339 conn_t *connp;
2340 conn_t *prev_connp;
2341
2342 for (i = 0; i < CONN_G_HASH_SIZE; i++) {
2343 mutex_enter(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock);
2344 prev_connp = NULL;
2345 connp = ipst->ips_ipcl_globalhash_fanout[i].connf_head;
2346 while (connp != NULL) {
2347 mutex_enter(&connp->conn_lock);
2348 if (connp->conn_state_flags &
2349 (CONN_CONDEMNED | CONN_INCIPIENT)) {
2350 mutex_exit(&connp->conn_lock);
2351 connp = connp->conn_g_next;
2352 continue;
2353 }
2354 CONN_INC_REF_LOCKED(connp);
2355 mutex_exit(&connp->conn_lock);
2356 mutex_exit(
2357 &ipst->ips_ipcl_globalhash_fanout[i].connf_lock);
2358 (*func)(connp, arg);
2359 if (prev_connp != NULL)
2360 CONN_DEC_REF(prev_connp);
2361 mutex_enter(
2362 &ipst->ips_ipcl_globalhash_fanout[i].connf_lock);
2363 prev_connp = connp;
2364 connp = connp->conn_g_next;
2365 }
2366 mutex_exit(&ipst->ips_ipcl_globalhash_fanout[i].connf_lock);
2367 if (prev_connp != NULL)
2368 CONN_DEC_REF(prev_connp);
2369 }
2370 }
2371
2372 /*
2373 * Search for a peer TCP/IPv4 loopback conn by doing a reverse lookup on
2374 * the {src, dst, lport, fport} quadruplet. Returns with conn reference
2375 * held; caller must call CONN_DEC_REF. Only checks for connected entries
2376 * (peer tcp in ESTABLISHED state).
2377 */
2378 conn_t *
2379 ipcl_conn_tcp_lookup_reversed_ipv4(conn_t *connp, ipha_t *ipha, tcpha_t *tcpha,
2380 ip_stack_t *ipst)
2381 {
2382 uint32_t ports;
2383 uint16_t *pports = (uint16_t *)&ports;
2384 connf_t *connfp;
2385 conn_t *tconnp;
2386 boolean_t zone_chk;
2387
2388 /*
2389 * If either the source of destination address is loopback, then
2390 * both endpoints must be in the same Zone. Otherwise, both of
2391 * the addresses are system-wide unique (tcp is in ESTABLISHED
2392 * state) and the endpoints may reside in different Zones.
2393 */
2394 zone_chk = (ipha->ipha_src == htonl(INADDR_LOOPBACK) ||
2395 ipha->ipha_dst == htonl(INADDR_LOOPBACK));
2396
2397 pports[0] = tcpha->tha_fport;
2398 pports[1] = tcpha->tha_lport;
2399
2400 connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH(ipha->ipha_dst,
2401 ports, ipst)];
2402
2403 mutex_enter(&connfp->connf_lock);
2404 for (tconnp = connfp->connf_head; tconnp != NULL;
2405 tconnp = tconnp->conn_next) {
2406
2407 if (IPCL_CONN_MATCH(tconnp, IPPROTO_TCP,
2408 ipha->ipha_dst, ipha->ipha_src, ports) &&
2409 tconnp->conn_tcp->tcp_state == TCPS_ESTABLISHED &&
2410 (!zone_chk || tconnp->conn_zoneid == connp->conn_zoneid)) {
2411
2412 ASSERT(tconnp != connp);
2413 CONN_INC_REF(tconnp);
2414 mutex_exit(&connfp->connf_lock);
2415 return (tconnp);
2416 }
2417 }
2418 mutex_exit(&connfp->connf_lock);
2419 return (NULL);
2420 }
2421
2422 /*
2423 * Search for a peer TCP/IPv6 loopback conn by doing a reverse lookup on
2424 * the {src, dst, lport, fport} quadruplet. Returns with conn reference
2425 * held; caller must call CONN_DEC_REF. Only checks for connected entries
2426 * (peer tcp in ESTABLISHED state).
2427 */
2428 conn_t *
2429 ipcl_conn_tcp_lookup_reversed_ipv6(conn_t *connp, ip6_t *ip6h, tcpha_t *tcpha,
2430 ip_stack_t *ipst)
2431 {
2432 uint32_t ports;
2433 uint16_t *pports = (uint16_t *)&ports;
2434 connf_t *connfp;
2435 conn_t *tconnp;
2436 boolean_t zone_chk;
2437
2438 /*
2439 * If either the source of destination address is loopback, then
2440 * both endpoints must be in the same Zone. Otherwise, both of
2441 * the addresses are system-wide unique (tcp is in ESTABLISHED
2442 * state) and the endpoints may reside in different Zones. We
2443 * don't do Zone check for link local address(es) because the
2444 * current Zone implementation treats each link local address as
2445 * being unique per system node, i.e. they belong to global Zone.
2446 */
2447 zone_chk = (IN6_IS_ADDR_LOOPBACK(&ip6h->ip6_src) ||
2448 IN6_IS_ADDR_LOOPBACK(&ip6h->ip6_dst));
2449
2450 pports[0] = tcpha->tha_fport;
2451 pports[1] = tcpha->tha_lport;
2452
2453 connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH_V6(ip6h->ip6_dst,
2454 ports, ipst)];
2455
2456 mutex_enter(&connfp->connf_lock);
2457 for (tconnp = connfp->connf_head; tconnp != NULL;
2458 tconnp = tconnp->conn_next) {
2459
2460 /* We skip conn_bound_if check here as this is loopback tcp */
2461 if (IPCL_CONN_MATCH_V6(tconnp, IPPROTO_TCP,
2462 ip6h->ip6_dst, ip6h->ip6_src, ports) &&
2463 tconnp->conn_tcp->tcp_state == TCPS_ESTABLISHED &&
2464 (!zone_chk || tconnp->conn_zoneid == connp->conn_zoneid)) {
2465
2466 ASSERT(tconnp != connp);
2467 CONN_INC_REF(tconnp);
2468 mutex_exit(&connfp->connf_lock);
2469 return (tconnp);
2470 }
2471 }
2472 mutex_exit(&connfp->connf_lock);
2473 return (NULL);
2474 }
2475
2476 /*
2477 * Find an exact {src, dst, lport, fport} match for a bounced datagram.
2478 * Returns with conn reference held. Caller must call CONN_DEC_REF.
2479 * Only checks for connected entries i.e. no INADDR_ANY checks.
2480 */
2481 conn_t *
2482 ipcl_tcp_lookup_reversed_ipv4(ipha_t *ipha, tcpha_t *tcpha, int min_state,
2483 ip_stack_t *ipst)
2484 {
2485 uint32_t ports;
2486 uint16_t *pports;
2487 connf_t *connfp;
2488 conn_t *tconnp;
2489
2490 pports = (uint16_t *)&ports;
2491 pports[0] = tcpha->tha_fport;
2492 pports[1] = tcpha->tha_lport;
2493
2494 connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH(ipha->ipha_dst,
2495 ports, ipst)];
2496
2497 mutex_enter(&connfp->connf_lock);
2498 for (tconnp = connfp->connf_head; tconnp != NULL;
2499 tconnp = tconnp->conn_next) {
2500
2501 if (IPCL_CONN_MATCH(tconnp, IPPROTO_TCP,
2502 ipha->ipha_dst, ipha->ipha_src, ports) &&
2503 tconnp->conn_tcp->tcp_state >= min_state) {
2504
2505 CONN_INC_REF(tconnp);
2506 mutex_exit(&connfp->connf_lock);
2507 return (tconnp);
2508 }
2509 }
2510 mutex_exit(&connfp->connf_lock);
2511 return (NULL);
2512 }
2513
2514 /*
2515 * Find an exact {src, dst, lport, fport} match for a bounced datagram.
2516 * Returns with conn reference held. Caller must call CONN_DEC_REF.
2517 * Only checks for connected entries i.e. no INADDR_ANY checks.
2518 * Match on ifindex in addition to addresses.
2519 */
2520 conn_t *
2521 ipcl_tcp_lookup_reversed_ipv6(ip6_t *ip6h, tcpha_t *tcpha, int min_state,
2522 uint_t ifindex, ip_stack_t *ipst)
2523 {
2524 tcp_t *tcp;
2525 uint32_t ports;
2526 uint16_t *pports;
2527 connf_t *connfp;
2528 conn_t *tconnp;
2529
2530 pports = (uint16_t *)&ports;
2531 pports[0] = tcpha->tha_fport;
2532 pports[1] = tcpha->tha_lport;
2533
2534 connfp = &ipst->ips_ipcl_conn_fanout[IPCL_CONN_HASH_V6(ip6h->ip6_dst,
2535 ports, ipst)];
2536
2537 mutex_enter(&connfp->connf_lock);
2538 for (tconnp = connfp->connf_head; tconnp != NULL;
2539 tconnp = tconnp->conn_next) {
2540
2541 tcp = tconnp->conn_tcp;
2542 if (IPCL_CONN_MATCH_V6(tconnp, IPPROTO_TCP,
2543 ip6h->ip6_dst, ip6h->ip6_src, ports) &&
2544 tcp->tcp_state >= min_state &&
2545 (tconnp->conn_bound_if == 0 ||
2546 tconnp->conn_bound_if == ifindex)) {
2547
2548 CONN_INC_REF(tconnp);
2549 mutex_exit(&connfp->connf_lock);
2550 return (tconnp);
2551 }
2552 }
2553 mutex_exit(&connfp->connf_lock);
2554 return (NULL);
2555 }
2556
2557 /*
2558 * Finds a TCP/IPv4 listening connection; called by tcp_disconnect to locate
2559 * a listener when changing state.
2560 */
2561 conn_t *
2562 ipcl_lookup_listener_v4(uint16_t lport, ipaddr_t laddr, zoneid_t zoneid,
2563 ip_stack_t *ipst)
2564 {
2565 connf_t *bind_connfp;
2566 conn_t *connp;
2567 tcp_t *tcp;
2568
2569 /*
2570 * Avoid false matches for packets sent to an IP destination of
2571 * all zeros.
2572 */
2573 if (laddr == 0)
2574 return (NULL);
2575
2576 ASSERT(zoneid != ALL_ZONES);
2577
2578 bind_connfp = &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)];
2579 mutex_enter(&bind_connfp->connf_lock);
2580 for (connp = bind_connfp->connf_head; connp != NULL;
2581 connp = connp->conn_next) {
2582 tcp = connp->conn_tcp;
2583 if (IPCL_BIND_MATCH(connp, IPPROTO_TCP, laddr, lport) &&
2584 IPCL_ZONE_MATCH(connp, zoneid) &&
2585 (tcp->tcp_listener == NULL)) {
2586 CONN_INC_REF(connp);
2587 mutex_exit(&bind_connfp->connf_lock);
2588 return (connp);
2589 }
2590 }
2591 mutex_exit(&bind_connfp->connf_lock);
2592 return (NULL);
2593 }
2594
2595 /*
2596 * Finds a TCP/IPv6 listening connection; called by tcp_disconnect to locate
2597 * a listener when changing state.
2598 */
2599 conn_t *
2600 ipcl_lookup_listener_v6(uint16_t lport, in6_addr_t *laddr, uint_t ifindex,
2601 zoneid_t zoneid, ip_stack_t *ipst)
2602 {
2603 connf_t *bind_connfp;
2604 conn_t *connp = NULL;
2605 tcp_t *tcp;
2606
2607 /*
2608 * Avoid false matches for packets sent to an IP destination of
2609 * all zeros.
2610 */
2611 if (IN6_IS_ADDR_UNSPECIFIED(laddr))
2612 return (NULL);
2613
2614 ASSERT(zoneid != ALL_ZONES);
2615
2616 bind_connfp = &ipst->ips_ipcl_bind_fanout[IPCL_BIND_HASH(lport, ipst)];
2617 mutex_enter(&bind_connfp->connf_lock);
2618 for (connp = bind_connfp->connf_head; connp != NULL;
2619 connp = connp->conn_next) {
2620 tcp = connp->conn_tcp;
2621 if (IPCL_BIND_MATCH_V6(connp, IPPROTO_TCP, *laddr, lport) &&
2622 IPCL_ZONE_MATCH(connp, zoneid) &&
2623 (connp->conn_bound_if == 0 ||
2624 connp->conn_bound_if == ifindex) &&
2625 tcp->tcp_listener == NULL) {
2626 CONN_INC_REF(connp);
2627 mutex_exit(&bind_connfp->connf_lock);
2628 return (connp);
2629 }
2630 }
2631 mutex_exit(&bind_connfp->connf_lock);
2632 return (NULL);
2633 }
2634
2635 /*
2636 * ipcl_get_next_conn
2637 * get the next entry in the conn global list
2638 * and put a reference on the next_conn.
2639 * decrement the reference on the current conn.
2640 *
2641 * This is an iterator based walker function that also provides for
2642 * some selection by the caller. It walks through the conn_hash bucket
2643 * searching for the next valid connp in the list, and selects connections
2644 * that are neither closed nor condemned. It also REFHOLDS the conn
2645 * thus ensuring that the conn exists when the caller uses the conn.
2646 */
2647 conn_t *
2648 ipcl_get_next_conn(connf_t *connfp, conn_t *connp, uint32_t conn_flags)
2649 {
2650 conn_t *next_connp;
2651
2652 if (connfp == NULL)
2653 return (NULL);
2654
2655 mutex_enter(&connfp->connf_lock);
2656
2657 next_connp = (connp == NULL) ?
2658 connfp->connf_head : connp->conn_g_next;
2659
2660 while (next_connp != NULL) {
2661 mutex_enter(&next_connp->conn_lock);
2662 if (!(next_connp->conn_flags & conn_flags) ||
2663 (next_connp->conn_state_flags &
2664 (CONN_CONDEMNED | CONN_INCIPIENT))) {
2665 /*
2666 * This conn has been condemned or
2667 * is closing, or the flags don't match
2668 */
2669 mutex_exit(&next_connp->conn_lock);
2670 next_connp = next_connp->conn_g_next;
2671 continue;
2672 }
2673 CONN_INC_REF_LOCKED(next_connp);
2674 mutex_exit(&next_connp->conn_lock);
2675 break;
2676 }
2677
2678 mutex_exit(&connfp->connf_lock);
2679
2680 if (connp != NULL)
2681 CONN_DEC_REF(connp);
2682
2683 return (next_connp);
2684 }
2685
2686 #ifdef CONN_DEBUG
2687 /*
2688 * Trace of the last NBUF refhold/refrele
2689 */
2690 int
2691 conn_trace_ref(conn_t *connp)
2692 {
2693 int last;
2694 conn_trace_t *ctb;
2695
2696 ASSERT(MUTEX_HELD(&connp->conn_lock));
2697 last = connp->conn_trace_last;
2698 last++;
2699 if (last == CONN_TRACE_MAX)
2700 last = 0;
2701
2702 ctb = &connp->conn_trace_buf[last];
2703 ctb->ctb_depth = getpcstack(ctb->ctb_stack, CONN_STACK_DEPTH);
2704 connp->conn_trace_last = last;
2705 return (1);
2706 }
2707
2708 int
2709 conn_untrace_ref(conn_t *connp)
2710 {
2711 int last;
2712 conn_trace_t *ctb;
2713
2714 ASSERT(MUTEX_HELD(&connp->conn_lock));
2715 last = connp->conn_trace_last;
2716 last++;
2717 if (last == CONN_TRACE_MAX)
2718 last = 0;
2719
2720 ctb = &connp->conn_trace_buf[last];
2721 ctb->ctb_depth = getpcstack(ctb->ctb_stack, CONN_STACK_DEPTH);
2722 connp->conn_trace_last = last;
2723 return (1);
2724 }
2725 #endif