xref: /illumos-gate/usr/src/uts/common/inet/sctp/sctp.c (revision e77c795bcbe51aebd7579fe13cbf2a6d56eca47f)
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 /*
23  * Copyright (c) 2004, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/stream.h>
28 #include <sys/strsubr.h>
29 #include <sys/stropts.h>
30 #include <sys/strsun.h>
31 #define	_SUN_TPI_VERSION 2
32 #include <sys/tihdr.h>
33 #include <sys/ddi.h>
34 #include <sys/sunddi.h>
35 #include <sys/xti_inet.h>
36 #include <sys/cmn_err.h>
37 #include <sys/debug.h>
38 #include <sys/vtrace.h>
39 #include <sys/kmem.h>
40 #include <sys/cpuvar.h>
41 #include <sys/random.h>
42 #include <sys/priv.h>
43 #include <sys/sunldi.h>
44 
45 #include <sys/errno.h>
46 #include <sys/signal.h>
47 #include <sys/socket.h>
48 #include <sys/isa_defs.h>
49 #include <netinet/in.h>
50 #include <netinet/tcp.h>
51 #include <netinet/ip6.h>
52 #include <netinet/icmp6.h>
53 #include <netinet/sctp.h>
54 #include <net/if.h>
55 
56 #include <inet/common.h>
57 #include <inet/ip.h>
58 #include <inet/ip_if.h>
59 #include <inet/ip_ire.h>
60 #include <inet/ip6.h>
61 #include <inet/mi.h>
62 #include <inet/mib2.h>
63 #include <inet/kstatcom.h>
64 #include <inet/optcom.h>
65 #include <inet/ipclassifier.h>
66 #include <inet/ipsec_impl.h>
67 #include <inet/sctp_ip.h>
68 #include <inet/sctp_crc32.h>
69 
70 #include "sctp_impl.h"
71 #include "sctp_addr.h"
72 #include "sctp_asconf.h"
73 
74 int sctpdebug;
75 sin6_t	sctp_sin6_null;	/* Zero address for quick clears */
76 
77 static void	sctp_closei_local(sctp_t *sctp);
78 static int	sctp_init_values(sctp_t *, sctp_t *, int);
79 static void	sctp_icmp_error_ipv6(sctp_t *sctp, mblk_t *mp);
80 static void	sctp_process_recvq(void *);
81 static void	sctp_rq_tq_init(sctp_stack_t *);
82 static void	sctp_rq_tq_fini(sctp_stack_t *);
83 static void	sctp_conn_cache_init();
84 static void	sctp_conn_cache_fini();
85 static int	sctp_conn_cache_constructor();
86 static void	sctp_conn_cache_destructor();
87 static void	sctp_conn_clear(conn_t *);
88 static void	sctp_notify(void *, ip_xmit_attr_t *, ixa_notify_type_t,
89     ixa_notify_arg_t);
90 
91 static void	*sctp_stack_init(netstackid_t stackid, netstack_t *ns);
92 static void	sctp_stack_fini(netstackid_t stackid, void *arg);
93 
94 /*
95  * SCTP receive queue taskq
96  *
97  * At SCTP initialization time, a default taskq is created for
98  * servicing packets received when the interrupt thread cannot
99  * get a hold on the sctp_t.  The number of taskq can be increased in
100  * sctp_find_next_tq() when an existing taskq cannot be dispatched.
101  * The taskqs are never removed.  But the max number of taskq which
102  * can be created is controlled by sctp_recvq_tq_list_max_sz.  Note
103  * that SCTP recvq taskq is not tied to any specific CPU or ill.
104  *
105  * Those taskqs are stored in an array recvq_tq_list.  And they are
106  * used in a round robin fashion.  The current taskq being used is
107  * determined by recvq_tq_list_cur.
108  */
109 
110 /* /etc/system variables */
111 /* The minimum number of threads for each taskq. */
112 int sctp_recvq_tq_thr_min = 4;
113 /* The maximum number of threads for each taskq. */
114 int sctp_recvq_tq_thr_max = 48;
115 /* The mnimum number of tasks for each taskq. */
116 int sctp_recvq_tq_task_min = 8;
117 /* Default value of sctp_recvq_tq_list_max_sz. */
118 int sctp_recvq_tq_list_max = 16;
119 
120 /*
121  * SCTP tunables related declarations. Definitions are in sctp_tunables.c
122  */
123 extern mod_prop_info_t sctp_propinfo_tbl[];
124 extern int sctp_propinfo_count;
125 
126 /*  sctp_t/conn_t kmem cache */
127 struct kmem_cache	*sctp_conn_cache;
128 
129 #define	SCTP_CONDEMNED(sctp)				\
130 	mutex_enter(&(sctp)->sctp_reflock);		\
131 	((sctp)->sctp_condemned = B_TRUE);		\
132 	mutex_exit(&(sctp)->sctp_reflock);
133 
134 /* Link/unlink a sctp_t to/from the global list. */
135 #define	SCTP_LINK(sctp, sctps)				\
136 	mutex_enter(&(sctps)->sctps_g_lock);		\
137 	list_insert_tail(&sctps->sctps_g_list, (sctp));	\
138 	mutex_exit(&(sctps)->sctps_g_lock);
139 
140 #define	SCTP_UNLINK(sctp, sctps)				\
141 	mutex_enter(&(sctps)->sctps_g_lock);		\
142 	ASSERT((sctp)->sctp_condemned);			\
143 	list_remove(&(sctps)->sctps_g_list, (sctp));	\
144 	mutex_exit(&(sctps)->sctps_g_lock);
145 
146 /*
147  * Hooks for Sun Cluster. On non-clustered nodes these will remain NULL.
148  * PSARC/2005/602.
149  */
150 void (*cl_sctp_listen)(sa_family_t, uchar_t *, uint_t, in_port_t) = NULL;
151 void (*cl_sctp_unlisten)(sa_family_t, uchar_t *, uint_t, in_port_t) = NULL;
152 void (*cl_sctp_connect)(sa_family_t, uchar_t *, uint_t, in_port_t,
153     uchar_t *, uint_t, in_port_t, boolean_t, cl_sctp_handle_t) = NULL;
154 void (*cl_sctp_disconnect)(sa_family_t, cl_sctp_handle_t) = NULL;
155 void (*cl_sctp_assoc_change)(sa_family_t, uchar_t *, size_t, uint_t,
156     uchar_t *, size_t, uint_t, int, cl_sctp_handle_t) = NULL;
157 void (*cl_sctp_check_addrs)(sa_family_t, in_port_t, uchar_t **, size_t,
158     uint_t *, boolean_t) = NULL;
159 /*
160  * Return the version number of the SCTP kernel interface.
161  */
162 int
163 sctp_itf_ver(int cl_ver)
164 {
165 	if (cl_ver != SCTP_ITF_VER)
166 		return (-1);
167 	return (SCTP_ITF_VER);
168 }
169 
170 /*
171  * Called when we need a new sctp instantiation but don't really have a
172  * new q to hang it off of. Copy the priv flag from the passed in structure.
173  */
174 sctp_t *
175 sctp_create_eager(sctp_t *psctp)
176 {
177 	sctp_t	*sctp;
178 	mblk_t	*ack_mp, *hb_mp;
179 	conn_t	*connp;
180 	cred_t	*credp;
181 	sctp_stack_t	*sctps = psctp->sctp_sctps;
182 
183 	if ((connp = ipcl_conn_create(IPCL_SCTPCONN, KM_NOSLEEP,
184 	    sctps->sctps_netstack)) == NULL) {
185 		return (NULL);
186 	}
187 
188 	sctp = CONN2SCTP(connp);
189 	sctp->sctp_sctps = sctps;
190 
191 	if ((ack_mp = sctp_timer_alloc(sctp, sctp_ack_timer,
192 	    KM_NOSLEEP)) == NULL ||
193 	    (hb_mp = sctp_timer_alloc(sctp, sctp_heartbeat_timer,
194 	    KM_NOSLEEP)) == NULL) {
195 		if (ack_mp != NULL)
196 			freeb(ack_mp);
197 		sctp_conn_clear(connp);
198 		sctp->sctp_sctps = NULL;
199 		kmem_cache_free(sctp_conn_cache, connp);
200 		return (NULL);
201 	}
202 
203 	sctp->sctp_ack_mp = ack_mp;
204 	sctp->sctp_heartbeat_mp = hb_mp;
205 
206 	if (sctp_init_values(sctp, psctp, KM_NOSLEEP) != 0) {
207 		freeb(ack_mp);
208 		freeb(hb_mp);
209 		sctp_conn_clear(connp);
210 		sctp->sctp_sctps = NULL;
211 		kmem_cache_free(sctp_conn_cache, connp);
212 		return (NULL);
213 	}
214 
215 	if ((credp = psctp->sctp_connp->conn_cred) != NULL) {
216 		connp->conn_cred = credp;
217 		crhold(credp);
218 	}
219 
220 	sctp->sctp_mss = psctp->sctp_mss;
221 	sctp->sctp_detached = B_TRUE;
222 	/*
223 	 * Link to the global as soon as possible so that this sctp_t
224 	 * can be found.
225 	 */
226 	SCTP_LINK(sctp, sctps);
227 
228 	/* If the listener has a limit, inherit the counter info. */
229 	sctp->sctp_listen_cnt = psctp->sctp_listen_cnt;
230 
231 	return (sctp);
232 }
233 
234 /*
235  * We are dying for some reason.  Try to do it gracefully.
236  */
237 void
238 sctp_clean_death(sctp_t *sctp, int err)
239 {
240 	ASSERT(sctp != NULL);
241 
242 	dprint(3, ("sctp_clean_death %p, state %d\n", (void *)sctp,
243 	    sctp->sctp_state));
244 
245 	sctp->sctp_client_errno = err;
246 	/*
247 	 * Check to see if we need to notify upper layer.
248 	 */
249 	if ((sctp->sctp_state >= SCTPS_COOKIE_WAIT) &&
250 	    !SCTP_IS_DETACHED(sctp)) {
251 		if (sctp->sctp_xmit_head || sctp->sctp_xmit_unsent) {
252 			sctp_regift_xmitlist(sctp);
253 		}
254 		if (sctp->sctp_ulp_disconnected(sctp->sctp_ulpd, 0, err)) {
255 			/*
256 			 * Socket is gone, detach.
257 			 */
258 			sctp->sctp_detached = B_TRUE;
259 			sctp->sctp_ulpd = NULL;
260 			sctp->sctp_upcalls = NULL;
261 		}
262 	}
263 
264 	/* Remove this sctp from all hashes. */
265 	sctp_closei_local(sctp);
266 
267 	/*
268 	 * If the sctp_t is detached, we need to finish freeing up
269 	 * the resources.  At this point, ip_fanout_sctp() should have
270 	 * a hold on this sctp_t.  Some thread doing snmp stuff can
271 	 * have a hold.  And a taskq can also have a hold waiting to
272 	 * work.  sctp_unlink() the sctp_t from the global list so
273 	 * that no new thread can find it.  Then do a SCTP_REFRELE().
274 	 * The sctp_t will be freed after all those threads are done.
275 	 */
276 	if (SCTP_IS_DETACHED(sctp)) {
277 		SCTP_CONDEMNED(sctp);
278 		SCTP_REFRELE(sctp);
279 	}
280 }
281 
282 /*
283  * Called by upper layer when it wants to close this association.
284  * Depending on the state of this assoication, we need to do
285  * different things.
286  *
287  * If the state is below COOKIE_ECHOED or it is COOKIE_ECHOED but with
288  * no sent data, just remove this sctp from all the hashes.  This
289  * makes sure that all packets from the other end will go to the default
290  * sctp handling.  The upper layer will then do a sctp_close() to clean
291  * up.
292  *
293  * Otherwise, check and see if SO_LINGER is set.  If it is set, check
294  * the value.  If the value is 0, consider this an abortive close.  Send
295  * an ABORT message and kill the associatiion.
296  *
297  */
298 int
299 sctp_disconnect(sctp_t *sctp)
300 {
301 	int		error = 0;
302 	conn_t		*connp = sctp->sctp_connp;
303 
304 	dprint(3, ("sctp_disconnect %p, state %d\n", (void *)sctp,
305 	    sctp->sctp_state));
306 
307 	RUN_SCTP(sctp);
308 
309 	switch (sctp->sctp_state) {
310 	case SCTPS_IDLE:
311 	case SCTPS_BOUND:
312 	case SCTPS_LISTEN:
313 		break;
314 	case SCTPS_COOKIE_WAIT:
315 	case SCTPS_COOKIE_ECHOED:
316 		/*
317 		 * Close during the connect 3-way handshake
318 		 * but here there may or may not be pending data
319 		 * already on queue. Process almost same as in
320 		 * the ESTABLISHED state.
321 		 */
322 		if (sctp->sctp_xmit_head == NULL &&
323 		    sctp->sctp_xmit_unsent == NULL) {
324 			break;
325 		}
326 		/* FALLTHRU */
327 	default:
328 		/*
329 		 * If SO_LINGER has set a zero linger time, terminate the
330 		 * association and send an ABORT.
331 		 */
332 		if (connp->conn_linger && connp->conn_lingertime == 0) {
333 			sctp_user_abort(sctp, NULL);
334 			WAKE_SCTP(sctp);
335 			return (error);
336 		}
337 
338 		/*
339 		 * If there is unread data, send an ABORT and terminate the
340 		 * association.
341 		 */
342 		if (sctp->sctp_rxqueued > 0 || sctp->sctp_ulp_rxqueued > 0) {
343 			sctp_user_abort(sctp, NULL);
344 			WAKE_SCTP(sctp);
345 			return (error);
346 		}
347 		/*
348 		 * Transmit the shutdown before detaching the sctp_t.
349 		 * After sctp_detach returns this queue/perimeter
350 		 * no longer owns the sctp_t thus others can modify it.
351 		 */
352 		sctp_send_shutdown(sctp, 0);
353 
354 		/* Pass gathered wisdom to IP for keeping */
355 		sctp_update_dce(sctp);
356 
357 		/*
358 		 * If lingering on close then wait until the shutdown
359 		 * is complete, or the SO_LINGER time passes, or an
360 		 * ABORT is sent/received.  Note that sctp_disconnect()
361 		 * can be called more than once.  Make sure that only
362 		 * one thread waits.
363 		 */
364 		if (connp->conn_linger && connp->conn_lingertime > 0 &&
365 		    sctp->sctp_state >= SCTPS_ESTABLISHED &&
366 		    !sctp->sctp_lingering) {
367 			clock_t stoptime;	/* in ticks */
368 			clock_t ret;
369 
370 			sctp->sctp_lingering = 1;
371 			sctp->sctp_client_errno = 0;
372 			stoptime = ddi_get_lbolt() +
373 			    connp->conn_lingertime * hz;
374 
375 			mutex_enter(&sctp->sctp_lock);
376 			sctp->sctp_running = B_FALSE;
377 			while (sctp->sctp_state >= SCTPS_ESTABLISHED &&
378 			    sctp->sctp_client_errno == 0) {
379 				cv_signal(&sctp->sctp_cv);
380 				ret = cv_timedwait_sig(&sctp->sctp_cv,
381 				    &sctp->sctp_lock, stoptime);
382 				if (ret < 0) {
383 					/* Stoptime has reached. */
384 					sctp->sctp_client_errno = EWOULDBLOCK;
385 					break;
386 				} else if (ret == 0) {
387 					/* Got a signal. */
388 					break;
389 				}
390 			}
391 			error = sctp->sctp_client_errno;
392 			sctp->sctp_client_errno = 0;
393 			mutex_exit(&sctp->sctp_lock);
394 		}
395 
396 		WAKE_SCTP(sctp);
397 		return (error);
398 	}
399 
400 
401 	/* Remove this sctp from all hashes so nobody can find it. */
402 	sctp_closei_local(sctp);
403 	WAKE_SCTP(sctp);
404 	return (error);
405 }
406 
407 void
408 sctp_close(sctp_t *sctp)
409 {
410 	dprint(3, ("sctp_close %p, state %d\n", (void *)sctp,
411 	    sctp->sctp_state));
412 
413 	RUN_SCTP(sctp);
414 	sctp->sctp_detached = 1;
415 	sctp->sctp_ulpd = NULL;
416 	sctp->sctp_upcalls = NULL;
417 	bzero(&sctp->sctp_events, sizeof (sctp->sctp_events));
418 
419 	/* If the graceful shutdown has not been completed, just return. */
420 	if (sctp->sctp_state != SCTPS_IDLE) {
421 		WAKE_SCTP(sctp);
422 		return;
423 	}
424 
425 	/*
426 	 * Since sctp_t is in SCTPS_IDLE state, so the only thread which
427 	 * can have a hold on the sctp_t is doing snmp stuff.  Just do
428 	 * a SCTP_REFRELE() here after the SCTP_UNLINK().  It will
429 	 * be freed when the other thread is done.
430 	 */
431 	SCTP_CONDEMNED(sctp);
432 	WAKE_SCTP(sctp);
433 	SCTP_REFRELE(sctp);
434 }
435 
436 /*
437  * Unlink from global list and do the eager close.
438  * Remove the refhold implicit in being on the global list.
439  */
440 void
441 sctp_close_eager(sctp_t *sctp)
442 {
443 	SCTP_CONDEMNED(sctp);
444 	sctp_closei_local(sctp);
445 	SCTP_REFRELE(sctp);
446 }
447 
448 /*
449  * The sctp_t is going away. Remove it from all lists and set it
450  * to SCTPS_IDLE. The caller has to remove it from the
451  * global list. The freeing up of memory is deferred until
452  * sctp_free(). This is needed since a thread in sctp_input() might have
453  * done a SCTP_REFHOLD on this structure before it was removed from the
454  * hashes.
455  */
456 static void
457 sctp_closei_local(sctp_t *sctp)
458 {
459 	mblk_t	*mp;
460 	conn_t	*connp = sctp->sctp_connp;
461 
462 	/* The counter is incremented only for established associations. */
463 	if (sctp->sctp_state >= SCTPS_ESTABLISHED)
464 		SCTPS_ASSOC_DEC(sctp->sctp_sctps);
465 
466 	if (sctp->sctp_listen_cnt != NULL)
467 		SCTP_DECR_LISTEN_CNT(sctp);
468 
469 	/* Sanity check, don't do the same thing twice.  */
470 	if (connp->conn_state_flags & CONN_CLOSING) {
471 		ASSERT(sctp->sctp_state == SCTPS_IDLE);
472 		return;
473 	}
474 
475 	/* Stop and free the timers */
476 	sctp_free_faddr_timers(sctp);
477 	if ((mp = sctp->sctp_heartbeat_mp) != NULL) {
478 		sctp_timer_free(mp);
479 		sctp->sctp_heartbeat_mp = NULL;
480 	}
481 	if ((mp = sctp->sctp_ack_mp) != NULL) {
482 		sctp_timer_free(mp);
483 		sctp->sctp_ack_mp = NULL;
484 	}
485 
486 	/* Set the CONN_CLOSING flag so that IP will not cache IRE again. */
487 	mutex_enter(&connp->conn_lock);
488 	connp->conn_state_flags |= CONN_CLOSING;
489 	mutex_exit(&connp->conn_lock);
490 
491 	/* Remove from all hashes. */
492 	sctp_bind_hash_remove(sctp);
493 	sctp_conn_hash_remove(sctp);
494 	sctp_listen_hash_remove(sctp);
495 	sctp->sctp_state = SCTPS_IDLE;
496 
497 	/*
498 	 * Clean up the recvq as much as possible.  All those packets
499 	 * will be silently dropped as this sctp_t is now in idle state.
500 	 */
501 	mutex_enter(&sctp->sctp_recvq_lock);
502 	while ((mp = sctp->sctp_recvq) != NULL) {
503 		sctp->sctp_recvq = mp->b_next;
504 		mp->b_next = NULL;
505 
506 		if (ip_recv_attr_is_mblk(mp))
507 			mp = ip_recv_attr_free_mblk(mp);
508 
509 		freemsg(mp);
510 	}
511 	mutex_exit(&sctp->sctp_recvq_lock);
512 }
513 
514 /*
515  * Free memory associated with the sctp/ip header template.
516  */
517 static void
518 sctp_headers_free(sctp_t *sctp)
519 {
520 	if (sctp->sctp_iphc != NULL) {
521 		kmem_free(sctp->sctp_iphc, sctp->sctp_iphc_len);
522 		sctp->sctp_iphc = NULL;
523 		sctp->sctp_ipha = NULL;
524 		sctp->sctp_hdr_len = 0;
525 		sctp->sctp_ip_hdr_len = 0;
526 		sctp->sctp_iphc_len = 0;
527 		sctp->sctp_sctph = NULL;
528 		sctp->sctp_hdr_len = 0;
529 	}
530 	if (sctp->sctp_iphc6 != NULL) {
531 		kmem_free(sctp->sctp_iphc6, sctp->sctp_iphc6_len);
532 		sctp->sctp_iphc6 = NULL;
533 		sctp->sctp_ip6h = NULL;
534 		sctp->sctp_hdr6_len = 0;
535 		sctp->sctp_ip_hdr6_len = 0;
536 		sctp->sctp_iphc6_len = 0;
537 		sctp->sctp_sctph6 = NULL;
538 		sctp->sctp_hdr6_len = 0;
539 	}
540 }
541 
542 static void
543 sctp_free_xmit_data(sctp_t *sctp)
544 {
545 	mblk_t	*ump = NULL;
546 	mblk_t	*nump;
547 	mblk_t	*mp;
548 	mblk_t	*nmp;
549 
550 	sctp->sctp_xmit_unacked = NULL;
551 	ump = sctp->sctp_xmit_head;
552 	sctp->sctp_xmit_tail = sctp->sctp_xmit_head = NULL;
553 free_unsent:
554 	for (; ump != NULL; ump = nump) {
555 		for (mp = ump->b_cont; mp != NULL; mp = nmp) {
556 			nmp = mp->b_next;
557 			mp->b_next = NULL;
558 			mp->b_prev = NULL;
559 			freemsg(mp);
560 		}
561 		ASSERT(DB_REF(ump) == 1);
562 		nump = ump->b_next;
563 		ump->b_next = NULL;
564 		ump->b_prev = NULL;
565 		ump->b_cont = NULL;
566 		freeb(ump);
567 	}
568 	if ((ump = sctp->sctp_xmit_unsent) == NULL) {
569 		ASSERT(sctp->sctp_xmit_unsent_tail == NULL);
570 		return;
571 	}
572 	sctp->sctp_xmit_unsent = sctp->sctp_xmit_unsent_tail = NULL;
573 	goto free_unsent;
574 }
575 
576 /*
577  * Cleanup all the messages in the stream queue and the reassembly lists.
578  * If 'free' is true, then delete the streams as well.
579  */
580 void
581 sctp_instream_cleanup(sctp_t *sctp, boolean_t free)
582 {
583 	int	i;
584 	mblk_t	*mp;
585 	mblk_t	*mp1;
586 
587 	if (sctp->sctp_instr != NULL) {
588 		/* walk thru and flush out anything remaining in the Q */
589 		for (i = 0; i < sctp->sctp_num_istr; i++) {
590 			mp = sctp->sctp_instr[i].istr_msgs;
591 			while (mp != NULL) {
592 				mp1 = mp->b_next;
593 				mp->b_next = mp->b_prev = NULL;
594 				freemsg(mp);
595 				mp = mp1;
596 			}
597 			sctp->sctp_instr[i].istr_msgs = NULL;
598 			sctp->sctp_instr[i].istr_nmsgs = 0;
599 			sctp_free_reass((sctp->sctp_instr) + i);
600 			sctp->sctp_instr[i].nextseq = 0;
601 		}
602 		if (free) {
603 			kmem_free(sctp->sctp_instr,
604 			    sizeof (*sctp->sctp_instr) * sctp->sctp_num_istr);
605 			sctp->sctp_instr = NULL;
606 			sctp->sctp_num_istr = 0;
607 		}
608 	}
609 	/* un-ordered fragments */
610 	if (sctp->sctp_uo_frags != NULL) {
611 		for (mp = sctp->sctp_uo_frags; mp != NULL; mp = mp1) {
612 			mp1 = mp->b_next;
613 			mp->b_next = mp->b_prev = NULL;
614 			freemsg(mp);
615 		}
616 		sctp->sctp_uo_frags = NULL;
617 	}
618 }
619 
620 /*
621  * Last reference to the sctp_t is gone. Free all memory associated with it.
622  * Called from SCTP_REFRELE. Called inline in sctp_close()
623  */
624 void
625 sctp_free(conn_t *connp)
626 {
627 	sctp_t *sctp = CONN2SCTP(connp);
628 	int		cnt;
629 	sctp_stack_t	*sctps = sctp->sctp_sctps;
630 
631 	ASSERT(sctps != NULL);
632 	/* Unlink it from the global list */
633 	SCTP_UNLINK(sctp, sctps);
634 
635 	ASSERT(connp->conn_ref == 0);
636 	ASSERT(connp->conn_proto == IPPROTO_SCTP);
637 	ASSERT(!MUTEX_HELD(&sctp->sctp_reflock));
638 	ASSERT(sctp->sctp_refcnt == 0);
639 
640 	ASSERT(sctp->sctp_ptpbhn == NULL && sctp->sctp_bind_hash == NULL);
641 	ASSERT(sctp->sctp_conn_hash_next == NULL &&
642 	    sctp->sctp_conn_hash_prev == NULL);
643 
644 
645 	/* Free up all the resources. */
646 
647 	/* blow away sctp stream management */
648 	if (sctp->sctp_ostrcntrs != NULL) {
649 		kmem_free(sctp->sctp_ostrcntrs,
650 		    sizeof (uint16_t) * sctp->sctp_num_ostr);
651 		sctp->sctp_ostrcntrs = NULL;
652 	}
653 	sctp_instream_cleanup(sctp, B_TRUE);
654 
655 	/* Remove all data transfer resources. */
656 	sctp->sctp_istr_nmsgs = 0;
657 	sctp->sctp_rxqueued = 0;
658 	sctp_free_xmit_data(sctp);
659 	sctp->sctp_unacked = 0;
660 	sctp->sctp_unsent = 0;
661 	if (sctp->sctp_cxmit_list != NULL)
662 		sctp_asconf_free_cxmit(sctp, NULL);
663 
664 	sctp->sctp_lastdata = NULL;
665 
666 	/* Clear out default xmit settings */
667 	sctp->sctp_def_stream = 0;
668 	sctp->sctp_def_flags = 0;
669 	sctp->sctp_def_ppid = 0;
670 	sctp->sctp_def_context = 0;
671 	sctp->sctp_def_timetolive = 0;
672 
673 	if (sctp->sctp_sack_info != NULL) {
674 		sctp_free_set(sctp->sctp_sack_info);
675 		sctp->sctp_sack_info = NULL;
676 	}
677 	sctp->sctp_sack_gaps = 0;
678 
679 	if (sctp->sctp_cookie_mp != NULL) {
680 		freemsg(sctp->sctp_cookie_mp);
681 		sctp->sctp_cookie_mp = NULL;
682 	}
683 
684 	/* Remove all the address resources. */
685 	sctp_zap_addrs(sctp);
686 	for (cnt = 0; cnt < SCTP_IPIF_HASH; cnt++) {
687 		ASSERT(sctp->sctp_saddrs[cnt].ipif_count == 0);
688 		list_destroy(&sctp->sctp_saddrs[cnt].sctp_ipif_list);
689 	}
690 
691 	if (sctp->sctp_hopopts != NULL) {
692 		mi_free(sctp->sctp_hopopts);
693 		sctp->sctp_hopopts = NULL;
694 		sctp->sctp_hopoptslen = 0;
695 	}
696 	ASSERT(sctp->sctp_hopoptslen == 0);
697 	if (sctp->sctp_dstopts != NULL) {
698 		mi_free(sctp->sctp_dstopts);
699 		sctp->sctp_dstopts = NULL;
700 		sctp->sctp_dstoptslen = 0;
701 	}
702 	ASSERT(sctp->sctp_dstoptslen == 0);
703 	if (sctp->sctp_rthdrdstopts != NULL) {
704 		mi_free(sctp->sctp_rthdrdstopts);
705 		sctp->sctp_rthdrdstopts = NULL;
706 		sctp->sctp_rthdrdstoptslen = 0;
707 	}
708 	ASSERT(sctp->sctp_rthdrdstoptslen == 0);
709 	if (sctp->sctp_rthdr != NULL) {
710 		mi_free(sctp->sctp_rthdr);
711 		sctp->sctp_rthdr = NULL;
712 		sctp->sctp_rthdrlen = 0;
713 	}
714 	ASSERT(sctp->sctp_rthdrlen == 0);
715 	sctp_headers_free(sctp);
716 
717 	sctp->sctp_shutdown_faddr = NULL;
718 
719 	if (sctp->sctp_err_chunks != NULL) {
720 		freemsg(sctp->sctp_err_chunks);
721 		sctp->sctp_err_chunks = NULL;
722 		sctp->sctp_err_len = 0;
723 	}
724 
725 	/* Clear all the bitfields. */
726 	bzero(&sctp->sctp_bits, sizeof (sctp->sctp_bits));
727 
728 	/* It is time to update the global statistics. */
729 	SCTPS_UPDATE_MIB(sctps, sctpOutSCTPPkts, sctp->sctp_opkts);
730 	SCTPS_UPDATE_MIB(sctps, sctpOutCtrlChunks, sctp->sctp_obchunks);
731 	SCTPS_UPDATE_MIB(sctps, sctpOutOrderChunks, sctp->sctp_odchunks);
732 	SCTPS_UPDATE_MIB(sctps, sctpOutUnorderChunks, sctp->sctp_oudchunks);
733 	SCTPS_UPDATE_MIB(sctps, sctpRetransChunks, sctp->sctp_rxtchunks);
734 	SCTPS_UPDATE_MIB(sctps, sctpInSCTPPkts, sctp->sctp_ipkts);
735 	SCTPS_UPDATE_MIB(sctps, sctpInCtrlChunks, sctp->sctp_ibchunks);
736 	SCTPS_UPDATE_MIB(sctps, sctpInOrderChunks, sctp->sctp_idchunks);
737 	SCTPS_UPDATE_MIB(sctps, sctpInUnorderChunks, sctp->sctp_iudchunks);
738 	SCTPS_UPDATE_MIB(sctps, sctpFragUsrMsgs, sctp->sctp_fragdmsgs);
739 	SCTPS_UPDATE_MIB(sctps, sctpReasmUsrMsgs, sctp->sctp_reassmsgs);
740 	sctp->sctp_opkts = 0;
741 	sctp->sctp_obchunks = 0;
742 	sctp->sctp_odchunks = 0;
743 	sctp->sctp_oudchunks = 0;
744 	sctp->sctp_rxtchunks = 0;
745 	sctp->sctp_ipkts = 0;
746 	sctp->sctp_ibchunks = 0;
747 	sctp->sctp_idchunks = 0;
748 	sctp->sctp_iudchunks = 0;
749 	sctp->sctp_fragdmsgs = 0;
750 	sctp->sctp_reassmsgs = 0;
751 	sctp->sctp_outseqtsns = 0;
752 	sctp->sctp_osacks = 0;
753 	sctp->sctp_isacks = 0;
754 	sctp->sctp_idupchunks = 0;
755 	sctp->sctp_gapcnt = 0;
756 	sctp->sctp_cum_obchunks = 0;
757 	sctp->sctp_cum_odchunks = 0;
758 	sctp->sctp_cum_oudchunks = 0;
759 	sctp->sctp_cum_rxtchunks = 0;
760 	sctp->sctp_cum_ibchunks = 0;
761 	sctp->sctp_cum_idchunks = 0;
762 	sctp->sctp_cum_iudchunks = 0;
763 
764 	sctp->sctp_autoclose = 0;
765 	sctp->sctp_tx_adaptation_code = 0;
766 
767 	sctp->sctp_v6label_len = 0;
768 	sctp->sctp_v4label_len = 0;
769 
770 	sctp->sctp_sctps = NULL;
771 
772 	sctp_conn_clear(connp);
773 	kmem_cache_free(sctp_conn_cache, connp);
774 }
775 
776 /*
777  * Initialize protocol control block. If a parent exists, inherit
778  * all values set through setsockopt().
779  */
780 static int
781 sctp_init_values(sctp_t *sctp, sctp_t *psctp, int sleep)
782 {
783 	int	err;
784 	int	cnt;
785 	sctp_stack_t	*sctps = sctp->sctp_sctps;
786 	conn_t 	*connp;
787 
788 	connp = sctp->sctp_connp;
789 
790 	sctp->sctp_nsaddrs = 0;
791 	for (cnt = 0; cnt < SCTP_IPIF_HASH; cnt++) {
792 		sctp->sctp_saddrs[cnt].ipif_count = 0;
793 		list_create(&sctp->sctp_saddrs[cnt].sctp_ipif_list,
794 		    sizeof (sctp_saddr_ipif_t), offsetof(sctp_saddr_ipif_t,
795 		    saddr_ipif));
796 	}
797 	connp->conn_ports = 0;
798 	sctp->sctp_running = B_FALSE;
799 	sctp->sctp_state = SCTPS_IDLE;
800 
801 	sctp->sctp_refcnt = 1;
802 
803 	sctp->sctp_strikes = 0;
804 
805 	sctp->sctp_last_mtu_probe = ddi_get_lbolt64();
806 	sctp->sctp_mtu_probe_intvl = sctps->sctps_mtu_probe_interval;
807 
808 	sctp->sctp_sack_gaps = 0;
809 	/* So we will not delay sending the first SACK. */
810 	sctp->sctp_sack_toggle = sctps->sctps_deferred_acks_max;
811 
812 	/* Only need to do the allocation if there is no "cached" one. */
813 	if (sctp->sctp_pad_mp == NULL) {
814 		if (sleep == KM_SLEEP) {
815 			sctp->sctp_pad_mp = allocb_wait(SCTP_ALIGN, BPRI_MED,
816 			    STR_NOSIG, NULL);
817 		} else {
818 			sctp->sctp_pad_mp = allocb(SCTP_ALIGN, BPRI_MED);
819 			if (sctp->sctp_pad_mp == NULL)
820 				return (ENOMEM);
821 		}
822 		bzero(sctp->sctp_pad_mp->b_rptr, SCTP_ALIGN);
823 	}
824 
825 	if (psctp != NULL) {
826 		/*
827 		 * Inherit from parent
828 		 *
829 		 * Start by inheriting from the conn_t, including conn_ixa and
830 		 * conn_xmit_ipp.
831 		 */
832 		err = conn_inherit_parent(psctp->sctp_connp, connp);
833 		if (err != 0)
834 			goto failure;
835 
836 		sctp->sctp_upcalls = psctp->sctp_upcalls;
837 
838 		sctp->sctp_cookie_lifetime = psctp->sctp_cookie_lifetime;
839 
840 		sctp->sctp_cwnd_max = psctp->sctp_cwnd_max;
841 		sctp->sctp_rwnd = psctp->sctp_rwnd;
842 		sctp->sctp_arwnd = psctp->sctp_arwnd;
843 		sctp->sctp_pd_point = psctp->sctp_pd_point;
844 		sctp->sctp_rto_max = psctp->sctp_rto_max;
845 		sctp->sctp_rto_max_init = psctp->sctp_rto_max_init;
846 		sctp->sctp_rto_min = psctp->sctp_rto_min;
847 		sctp->sctp_rto_initial = psctp->sctp_rto_initial;
848 		sctp->sctp_pa_max_rxt = psctp->sctp_pa_max_rxt;
849 		sctp->sctp_pp_max_rxt = psctp->sctp_pp_max_rxt;
850 		sctp->sctp_max_init_rxt = psctp->sctp_max_init_rxt;
851 
852 		sctp->sctp_def_stream = psctp->sctp_def_stream;
853 		sctp->sctp_def_flags = psctp->sctp_def_flags;
854 		sctp->sctp_def_ppid = psctp->sctp_def_ppid;
855 		sctp->sctp_def_context = psctp->sctp_def_context;
856 		sctp->sctp_def_timetolive = psctp->sctp_def_timetolive;
857 
858 		sctp->sctp_num_istr = psctp->sctp_num_istr;
859 		sctp->sctp_num_ostr = psctp->sctp_num_ostr;
860 
861 		sctp->sctp_hb_interval = psctp->sctp_hb_interval;
862 		sctp->sctp_autoclose = psctp->sctp_autoclose;
863 		sctp->sctp_tx_adaptation_code = psctp->sctp_tx_adaptation_code;
864 
865 		/* xxx should be a better way to copy these flags xxx */
866 		sctp->sctp_bound_to_all = psctp->sctp_bound_to_all;
867 		sctp->sctp_cansleep = psctp->sctp_cansleep;
868 		sctp->sctp_send_adaptation = psctp->sctp_send_adaptation;
869 		sctp->sctp_ndelay = psctp->sctp_ndelay;
870 		sctp->sctp_events = psctp->sctp_events;
871 	} else {
872 		/*
873 		 * Set to system defaults
874 		 */
875 		sctp->sctp_cookie_lifetime =
876 		    MSEC_TO_TICK(sctps->sctps_cookie_life);
877 		connp->conn_sndlowat = sctps->sctps_xmit_lowat;
878 		connp->conn_sndbuf = sctps->sctps_xmit_hiwat;
879 		connp->conn_rcvbuf = sctps->sctps_recv_hiwat;
880 
881 		sctp->sctp_cwnd_max = sctps->sctps_cwnd_max_;
882 		sctp->sctp_rwnd = connp->conn_rcvbuf;
883 		sctp->sctp_arwnd = connp->conn_rcvbuf;
884 		sctp->sctp_pd_point = sctp->sctp_rwnd;
885 		sctp->sctp_rto_max = MSEC_TO_TICK(sctps->sctps_rto_maxg);
886 		sctp->sctp_rto_max_init = sctp->sctp_rto_max;
887 		sctp->sctp_rto_min = MSEC_TO_TICK(sctps->sctps_rto_ming);
888 		sctp->sctp_rto_initial = MSEC_TO_TICK(
889 		    sctps->sctps_rto_initialg);
890 		sctp->sctp_pa_max_rxt = sctps->sctps_pa_max_retr;
891 		sctp->sctp_pp_max_rxt = sctps->sctps_pp_max_retr;
892 		sctp->sctp_max_init_rxt = sctps->sctps_max_init_retr;
893 
894 		sctp->sctp_num_istr = sctps->sctps_max_in_streams;
895 		sctp->sctp_num_ostr = sctps->sctps_initial_out_streams;
896 
897 		sctp->sctp_hb_interval =
898 		    MSEC_TO_TICK(sctps->sctps_heartbeat_interval);
899 
900 		if (connp->conn_family == AF_INET)
901 			connp->conn_default_ttl = sctps->sctps_ipv4_ttl;
902 		else
903 			connp->conn_default_ttl = sctps->sctps_ipv6_hoplimit;
904 
905 		connp->conn_xmit_ipp.ipp_unicast_hops =
906 		    connp->conn_default_ttl;
907 
908 		/*
909 		 * Initialize the header template
910 		 */
911 		if ((err = sctp_build_hdrs(sctp, sleep)) != 0) {
912 			goto failure;
913 		}
914 	}
915 
916 	sctp->sctp_understands_asconf = B_TRUE;
917 	sctp->sctp_understands_addip = B_TRUE;
918 	sctp->sctp_prsctp_aware = B_FALSE;
919 
920 	sctp->sctp_connp->conn_ref = 1;
921 
922 	sctp->sctp_prsctpdrop = 0;
923 	sctp->sctp_msgcount = 0;
924 
925 	return (0);
926 
927 failure:
928 	sctp_headers_free(sctp);
929 	return (err);
930 }
931 
932 /*
933  * Extracts the init tag from an INIT chunk and checks if it matches
934  * the sctp's verification tag. Returns 0 if it doesn't match, 1 if
935  * it does.
936  */
937 static boolean_t
938 sctp_icmp_verf(sctp_t *sctp, sctp_hdr_t *sh, mblk_t *mp)
939 {
940 	sctp_chunk_hdr_t *sch;
941 	uint32_t verf, *vp;
942 
943 	sch = (sctp_chunk_hdr_t *)(sh + 1);
944 	vp = (uint32_t *)(sch + 1);
945 
946 	/* Need at least the data chunk hdr and the first 4 bytes of INIT */
947 	if ((unsigned char *)(vp + 1) > mp->b_wptr) {
948 		return (B_FALSE);
949 	}
950 
951 	bcopy(vp, &verf, sizeof (verf));
952 
953 	if (verf == sctp->sctp_lvtag) {
954 		return (B_TRUE);
955 	}
956 	return (B_FALSE);
957 }
958 
959 /*
960  * Update the SCTP state according to change of PMTU.
961  *
962  * Path MTU might have changed by either increase or decrease, so need to
963  * adjust the MSS based on the value of ixa_pmtu.
964  */
965 static void
966 sctp_update_pmtu(sctp_t *sctp, sctp_faddr_t *fp, boolean_t decrease_only)
967 {
968 	uint32_t	pmtu;
969 	int32_t		mss;
970 	ip_xmit_attr_t	*ixa = fp->sf_ixa;
971 
972 	if (sctp->sctp_state < SCTPS_ESTABLISHED)
973 		return;
974 
975 	/*
976 	 * Always call ip_get_pmtu() to make sure that IP has updated
977 	 * ixa_flags properly.
978 	 */
979 	pmtu = ip_get_pmtu(ixa);
980 
981 	/*
982 	 * Calculate the MSS by decreasing the PMTU by sctp_hdr_len and
983 	 * IPsec overhead if applied. Make sure to use the most recent
984 	 * IPsec information.
985 	 */
986 	mss = pmtu - conn_ipsec_length(sctp->sctp_connp);
987 	if (ixa->ixa_flags & IXAF_IS_IPV4)
988 		mss -= sctp->sctp_hdr_len;
989 	else
990 		mss -= sctp->sctp_hdr6_len;
991 
992 	/*
993 	 * Nothing to change, so just return.
994 	 */
995 	if (mss == fp->sf_pmss)
996 		return;
997 
998 	/*
999 	 * Currently, for ICMP errors, only PMTU decrease is handled.
1000 	 */
1001 	if (mss > fp->sf_pmss && decrease_only)
1002 		return;
1003 
1004 #ifdef DEBUG
1005 	(void) printf("sctp_update_pmtu mss from %d to %d\n",
1006 	    fp->sf_pmss, mss);
1007 #endif
1008 	DTRACE_PROBE2(sctp_update_pmtu, int32_t, fp->sf_pmss, uint32_t, mss);
1009 
1010 	/*
1011 	 * Update ixa_fragsize and ixa_pmtu.
1012 	 */
1013 	ixa->ixa_fragsize = ixa->ixa_pmtu = pmtu;
1014 
1015 	/*
1016 	 * Make sure that sfa_pmss is a multiple of
1017 	 * SCTP_ALIGN.
1018 	 */
1019 	fp->sf_pmss = mss & ~(SCTP_ALIGN - 1);
1020 	fp->sf_pmtu_discovered = 1;
1021 
1022 #ifdef notyet
1023 	if (mss < sctp->sctp_sctps->sctps_mss_min)
1024 		ixa->ixa_flags |= IXAF_PMTU_TOO_SMALL;
1025 #endif
1026 	if (ixa->ixa_flags & IXAF_PMTU_TOO_SMALL)
1027 		ixa->ixa_flags &= ~(IXAF_DONTFRAG | IXAF_PMTU_IPV4_DF);
1028 
1029 	/*
1030 	 * If below the min size then ip_get_pmtu cleared IXAF_PMTU_IPV4_DF.
1031 	 * Make sure to clear IXAF_DONTFRAG, which is used by IP to decide
1032 	 * whether to fragment the packet.
1033 	 */
1034 	if (ixa->ixa_flags & IXAF_IS_IPV4) {
1035 		if (!(ixa->ixa_flags & IXAF_PMTU_IPV4_DF)) {
1036 			fp->sf_df = B_FALSE;
1037 			if (fp == sctp->sctp_current) {
1038 				sctp->sctp_ipha->
1039 				    ipha_fragment_offset_and_flags = 0;
1040 			}
1041 		}
1042 	}
1043 }
1044 
1045 /*
1046  * Notify function registered with ip_xmit_attr_t. It's called in the context
1047  * of conn_ip_output so it's safe to update the SCTP state.
1048  * Currently only used for pmtu changes.
1049  */
1050 /* ARGSUSED1 */
1051 static void
1052 sctp_notify(void *arg, ip_xmit_attr_t *ixa, ixa_notify_type_t ntype,
1053     ixa_notify_arg_t narg)
1054 {
1055 	sctp_t		*sctp = (sctp_t *)arg;
1056 	sctp_faddr_t	*fp;
1057 
1058 	switch (ntype) {
1059 	case IXAN_PMTU:
1060 		/* Find the faddr based on the ip_xmit_attr_t pointer */
1061 		for (fp = sctp->sctp_faddrs; fp != NULL; fp = fp->sf_next) {
1062 			if (fp->sf_ixa == ixa)
1063 				break;
1064 		}
1065 		if (fp != NULL)
1066 			sctp_update_pmtu(sctp, fp, B_FALSE);
1067 		break;
1068 	default:
1069 		break;
1070 	}
1071 }
1072 
1073 /*
1074  * sctp_icmp_error is called by sctp_input() to process ICMP error messages
1075  * passed up by IP.  We need to find a sctp_t
1076  * that corresponds to the returned datagram.  Passes the message back in on
1077  * the correct queue once it has located the connection.
1078  * Assumes that IP has pulled up everything up to and including
1079  * the ICMP header.
1080  */
1081 void
1082 sctp_icmp_error(sctp_t *sctp, mblk_t *mp)
1083 {
1084 	icmph_t *icmph;
1085 	ipha_t	*ipha;
1086 	int	iph_hdr_length;
1087 	sctp_hdr_t *sctph;
1088 	in6_addr_t dst;
1089 	sctp_faddr_t *fp;
1090 	sctp_stack_t	*sctps = sctp->sctp_sctps;
1091 
1092 	dprint(1, ("sctp_icmp_error: sctp=%p, mp=%p\n", (void *)sctp,
1093 	    (void *)mp));
1094 
1095 	ipha = (ipha_t *)mp->b_rptr;
1096 	if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
1097 		ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
1098 		sctp_icmp_error_ipv6(sctp, mp);
1099 		return;
1100 	}
1101 
1102 	/* account for the ip hdr from the icmp message */
1103 	iph_hdr_length = IPH_HDR_LENGTH(ipha);
1104 	icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
1105 	/* now the ip hdr of message resulting in this icmp */
1106 	ipha = (ipha_t *)&icmph[1];
1107 	iph_hdr_length = IPH_HDR_LENGTH(ipha);
1108 	sctph = (sctp_hdr_t *)((char *)ipha + iph_hdr_length);
1109 	/* first_mp must expose the full sctp header. */
1110 	if ((uchar_t *)(sctph + 1) >= mp->b_wptr) {
1111 		/* not enough data for SCTP header */
1112 		freemsg(mp);
1113 		return;
1114 	}
1115 
1116 	switch (icmph->icmph_type) {
1117 	case ICMP_DEST_UNREACHABLE:
1118 		switch (icmph->icmph_code) {
1119 		case ICMP_FRAGMENTATION_NEEDED:
1120 			/*
1121 			 * Reduce the MSS based on the new MTU.  This will
1122 			 * eliminate any fragmentation locally.
1123 			 * N.B.  There may well be some funny side-effects on
1124 			 * the local send policy and the remote receive policy.
1125 			 * Pending further research, we provide
1126 			 * sctp_ignore_path_mtu just in case this proves
1127 			 * disastrous somewhere.
1128 			 *
1129 			 * After updating the MSS, retransmit part of the
1130 			 * dropped segment using the new mss by calling
1131 			 * sctp_wput_slow().  Need to adjust all those
1132 			 * params to make sure sctp_wput_slow() work properly.
1133 			 */
1134 			if (sctps->sctps_ignore_path_mtu)
1135 				break;
1136 
1137 			/* find the offending faddr */
1138 			IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &dst);
1139 			fp = sctp_lookup_faddr(sctp, &dst);
1140 			if (fp == NULL) {
1141 				break;
1142 			}
1143 			sctp_update_pmtu(sctp, fp, B_TRUE);
1144 			/*
1145 			 * It is possible, even likely that a fast retransmit
1146 			 * attempt has been dropped by ip as a result of this
1147 			 * error, retransmission bundles as much as possible.
1148 			 * A retransmit here prevents significant delays waiting
1149 			 * on the timer. Analogous to behaviour of TCP after
1150 			 * ICMP too big.
1151 			 */
1152 			sctp_rexmit(sctp, fp);
1153 			break;
1154 		case ICMP_PORT_UNREACHABLE:
1155 		case ICMP_PROTOCOL_UNREACHABLE:
1156 			switch (sctp->sctp_state) {
1157 			case SCTPS_COOKIE_WAIT:
1158 			case SCTPS_COOKIE_ECHOED:
1159 				/* make sure the verification tag matches */
1160 				if (!sctp_icmp_verf(sctp, sctph, mp)) {
1161 					break;
1162 				}
1163 				SCTPS_BUMP_MIB(sctps, sctpAborted);
1164 				sctp_assoc_event(sctp, SCTP_CANT_STR_ASSOC, 0,
1165 				    NULL);
1166 				sctp_clean_death(sctp, ECONNREFUSED);
1167 				break;
1168 			}
1169 			break;
1170 		case ICMP_HOST_UNREACHABLE:
1171 		case ICMP_NET_UNREACHABLE:
1172 			/* Record the error in case we finally time out. */
1173 			sctp->sctp_client_errno = (icmph->icmph_code ==
1174 			    ICMP_HOST_UNREACHABLE) ? EHOSTUNREACH : ENETUNREACH;
1175 			break;
1176 		default:
1177 			break;
1178 		}
1179 		break;
1180 	case ICMP_SOURCE_QUENCH: {
1181 		/* Reduce the sending rate as if we got a retransmit timeout */
1182 		break;
1183 	}
1184 	}
1185 	freemsg(mp);
1186 }
1187 
1188 /*
1189  * sctp_icmp_error_ipv6() is called by sctp_icmp_error() to process ICMPv6
1190  * error messages passed up by IP.
1191  * Assumes that IP has pulled up all the extension headers as well
1192  * as the ICMPv6 header.
1193  */
1194 static void
1195 sctp_icmp_error_ipv6(sctp_t *sctp, mblk_t *mp)
1196 {
1197 	icmp6_t *icmp6;
1198 	ip6_t	*ip6h;
1199 	uint16_t	iph_hdr_length;
1200 	sctp_hdr_t *sctpha;
1201 	uint8_t	*nexthdrp;
1202 	sctp_faddr_t *fp;
1203 	sctp_stack_t	*sctps = sctp->sctp_sctps;
1204 
1205 	ip6h = (ip6_t *)mp->b_rptr;
1206 	iph_hdr_length = (ip6h->ip6_nxt != IPPROTO_SCTP) ?
1207 	    ip_hdr_length_v6(mp, ip6h) : IPV6_HDR_LEN;
1208 
1209 	icmp6 = (icmp6_t *)&mp->b_rptr[iph_hdr_length];
1210 	ip6h = (ip6_t *)&icmp6[1];
1211 	if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &iph_hdr_length, &nexthdrp)) {
1212 		freemsg(mp);
1213 		return;
1214 	}
1215 	ASSERT(*nexthdrp == IPPROTO_SCTP);
1216 
1217 	/* XXX need ifindex to find connection */
1218 	sctpha = (sctp_hdr_t *)((char *)ip6h + iph_hdr_length);
1219 	if ((uchar_t *)sctpha >= mp->b_wptr) {
1220 		/* not enough data for SCTP header */
1221 		freemsg(mp);
1222 		return;
1223 	}
1224 	switch (icmp6->icmp6_type) {
1225 	case ICMP6_PACKET_TOO_BIG:
1226 		/*
1227 		 * Reduce the MSS based on the new MTU.  This will
1228 		 * eliminate any fragmentation locally.
1229 		 * N.B.  There may well be some funny side-effects on
1230 		 * the local send policy and the remote receive policy.
1231 		 * Pending further research, we provide
1232 		 * sctp_ignore_path_mtu just in case this proves
1233 		 * disastrous somewhere.
1234 		 *
1235 		 * After updating the MSS, retransmit part of the
1236 		 * dropped segment using the new mss by calling
1237 		 * sctp_wput_slow().  Need to adjust all those
1238 		 * params to make sure sctp_wput_slow() work properly.
1239 		 */
1240 		if (sctps->sctps_ignore_path_mtu)
1241 			break;
1242 
1243 		/* find the offending faddr */
1244 		fp = sctp_lookup_faddr(sctp, &ip6h->ip6_dst);
1245 		if (fp == NULL) {
1246 			break;
1247 		}
1248 
1249 		sctp_update_pmtu(sctp, fp, B_TRUE);
1250 		/*
1251 		 * It is possible, even likely that a fast retransmit
1252 		 * attempt has been dropped by ip as a result of this
1253 		 * error, retransmission bundles as much as possible.
1254 		 * A retransmit here prevents significant delays waiting
1255 		 * on the timer. Analogous to behaviour of TCP after
1256 		 * ICMP too big.
1257 		 */
1258 		sctp_rexmit(sctp, fp);
1259 		break;
1260 
1261 	case ICMP6_DST_UNREACH:
1262 		switch (icmp6->icmp6_code) {
1263 		case ICMP6_DST_UNREACH_NOPORT:
1264 			/* make sure the verification tag matches */
1265 			if (!sctp_icmp_verf(sctp, sctpha, mp)) {
1266 				break;
1267 			}
1268 			if (sctp->sctp_state == SCTPS_COOKIE_WAIT ||
1269 			    sctp->sctp_state == SCTPS_COOKIE_ECHOED) {
1270 				SCTPS_BUMP_MIB(sctps, sctpAborted);
1271 				sctp_assoc_event(sctp, SCTP_CANT_STR_ASSOC, 0,
1272 				    NULL);
1273 				sctp_clean_death(sctp, ECONNREFUSED);
1274 			}
1275 			break;
1276 
1277 		case ICMP6_DST_UNREACH_ADMIN:
1278 		case ICMP6_DST_UNREACH_NOROUTE:
1279 		case ICMP6_DST_UNREACH_NOTNEIGHBOR:
1280 		case ICMP6_DST_UNREACH_ADDR:
1281 			/* Record the error in case we finally time out. */
1282 			sctp->sctp_client_errno = EHOSTUNREACH;
1283 			break;
1284 		default:
1285 			break;
1286 		}
1287 		break;
1288 
1289 	case ICMP6_PARAM_PROB:
1290 		/* If this corresponds to an ICMP_PROTOCOL_UNREACHABLE */
1291 		if (icmp6->icmp6_code == ICMP6_PARAMPROB_NEXTHEADER &&
1292 		    (uchar_t *)ip6h + icmp6->icmp6_pptr ==
1293 		    (uchar_t *)nexthdrp) {
1294 			/* make sure the verification tag matches */
1295 			if (!sctp_icmp_verf(sctp, sctpha, mp)) {
1296 				break;
1297 			}
1298 			if (sctp->sctp_state == SCTPS_COOKIE_WAIT) {
1299 				SCTPS_BUMP_MIB(sctps, sctpAborted);
1300 				sctp_assoc_event(sctp, SCTP_CANT_STR_ASSOC, 0,
1301 				    NULL);
1302 				sctp_clean_death(sctp, ECONNREFUSED);
1303 			}
1304 			break;
1305 		}
1306 		break;
1307 
1308 	case ICMP6_TIME_EXCEEDED:
1309 	default:
1310 		break;
1311 	}
1312 	freemsg(mp);
1313 }
1314 
1315 /*
1316  * Called by sockfs to create a new sctp instance.
1317  *
1318  * If parent pointer is passed in, inherit settings from it.
1319  */
1320 sctp_t *
1321 sctp_create(void *ulpd, sctp_t *parent, int family, int type, int flags,
1322     sock_upcalls_t *upcalls, sctp_sockbuf_limits_t *sbl,
1323     cred_t *credp)
1324 {
1325 	sctp_t		*sctp, *psctp;
1326 	conn_t		*connp;
1327 	mblk_t		*ack_mp, *hb_mp;
1328 	int		sleep = flags & SCTP_CAN_BLOCK ? KM_SLEEP : KM_NOSLEEP;
1329 	zoneid_t	zoneid;
1330 	sctp_stack_t	*sctps;
1331 
1332 	/* User must supply a credential. */
1333 	if (credp == NULL)
1334 		return (NULL);
1335 
1336 	psctp = (sctp_t *)parent;
1337 	if (psctp != NULL) {
1338 		sctps = psctp->sctp_sctps;
1339 		/* Increase here to have common decrease at end */
1340 		netstack_hold(sctps->sctps_netstack);
1341 		ASSERT(sctps->sctps_recvq_tq_list_cur_sz > 0);
1342 	} else {
1343 		netstack_t *ns;
1344 
1345 		ns = netstack_find_by_cred(credp);
1346 		sctps = ns->netstack_sctp;
1347 		/*
1348 		 * Check if the receive queue taskq for this sctp_stack_t has
1349 		 * been set up.
1350 		 */
1351 		if (sctps->sctps_recvq_tq_list_cur_sz == 0)
1352 			sctp_rq_tq_init(sctps);
1353 
1354 		/*
1355 		 * For exclusive stacks we set the zoneid to zero
1356 		 * to make SCTP operate as if in the global zone.
1357 		 */
1358 		if (sctps->sctps_netstack->netstack_stackid !=
1359 		    GLOBAL_NETSTACKID)
1360 			zoneid = GLOBAL_ZONEID;
1361 		else
1362 			zoneid = crgetzoneid(credp);
1363 	}
1364 	if ((connp = ipcl_conn_create(IPCL_SCTPCONN, sleep,
1365 	    sctps->sctps_netstack)) == NULL) {
1366 		netstack_rele(sctps->sctps_netstack);
1367 		SCTP_KSTAT(sctps, sctp_conn_create);
1368 		return (NULL);
1369 	}
1370 	/*
1371 	 * ipcl_conn_create did a netstack_hold. Undo the hold that was
1372 	 * done at top of sctp_create.
1373 	 */
1374 	netstack_rele(sctps->sctps_netstack);
1375 	sctp = CONN2SCTP(connp);
1376 	sctp->sctp_sctps = sctps;
1377 
1378 	if ((ack_mp = sctp_timer_alloc(sctp, sctp_ack_timer, sleep)) == NULL ||
1379 	    (hb_mp = sctp_timer_alloc(sctp, sctp_heartbeat_timer,
1380 	    sleep)) == NULL) {
1381 		if (ack_mp != NULL)
1382 			freeb(ack_mp);
1383 		sctp_conn_clear(connp);
1384 		sctp->sctp_sctps = NULL;
1385 		kmem_cache_free(sctp_conn_cache, connp);
1386 		return (NULL);
1387 	}
1388 
1389 	sctp->sctp_ack_mp = ack_mp;
1390 	sctp->sctp_heartbeat_mp = hb_mp;
1391 
1392 	/*
1393 	 * Have conn_ip_output drop packets should our outer source
1394 	 * go invalid, and tell us about mtu changes.
1395 	 */
1396 	connp->conn_ixa->ixa_flags |= IXAF_SET_ULP_CKSUM | IXAF_VERIFY_SOURCE |
1397 	    IXAF_VERIFY_PMTU;
1398 	connp->conn_family = family;
1399 	connp->conn_so_type = type;
1400 
1401 	if (sctp_init_values(sctp, psctp, sleep) != 0) {
1402 		freeb(ack_mp);
1403 		freeb(hb_mp);
1404 		sctp_conn_clear(connp);
1405 		sctp->sctp_sctps = NULL;
1406 		kmem_cache_free(sctp_conn_cache, connp);
1407 		return (NULL);
1408 	}
1409 	sctp->sctp_cansleep = ((flags & SCTP_CAN_BLOCK) == SCTP_CAN_BLOCK);
1410 
1411 	sctp->sctp_mss = sctps->sctps_initial_mtu - ((family == AF_INET6) ?
1412 	    sctp->sctp_hdr6_len : sctp->sctp_hdr_len);
1413 
1414 	if (psctp != NULL) {
1415 		conn_t	*pconnp = psctp->sctp_connp;
1416 
1417 		RUN_SCTP(psctp);
1418 		/*
1419 		 * Inherit local address list, local port. Parent is either
1420 		 * in SCTPS_BOUND, or SCTPS_LISTEN state.
1421 		 */
1422 		ASSERT((psctp->sctp_state == SCTPS_BOUND) ||
1423 		    (psctp->sctp_state == SCTPS_LISTEN));
1424 		if (sctp_dup_saddrs(psctp, sctp, sleep)) {
1425 			WAKE_SCTP(psctp);
1426 			freeb(ack_mp);
1427 			freeb(hb_mp);
1428 			sctp_headers_free(sctp);
1429 			sctp_conn_clear(connp);
1430 			sctp->sctp_sctps = NULL;
1431 			kmem_cache_free(sctp_conn_cache, connp);
1432 			return (NULL);
1433 		}
1434 
1435 		/*
1436 		 * If the parent is specified, it'll be immediatelly
1437 		 * followed by sctp_connect(). So don't add this guy to
1438 		 * bind hash.
1439 		 */
1440 		connp->conn_lport = pconnp->conn_lport;
1441 		sctp->sctp_state = SCTPS_BOUND;
1442 		WAKE_SCTP(psctp);
1443 	} else {
1444 		ASSERT(connp->conn_cred == NULL);
1445 		connp->conn_zoneid = zoneid;
1446 		/*
1447 		 * conn_allzones can not be set this early, hence
1448 		 * no IPCL_ZONEID
1449 		 */
1450 		connp->conn_ixa->ixa_zoneid = zoneid;
1451 		connp->conn_open_time = ddi_get_lbolt64();
1452 		connp->conn_cred = credp;
1453 		crhold(credp);
1454 		connp->conn_cpid = curproc->p_pid;
1455 
1456 		/*
1457 		 * If the caller has the process-wide flag set, then default to
1458 		 * MAC exempt mode.  This allows read-down to unlabeled hosts.
1459 		 */
1460 		if (getpflags(NET_MAC_AWARE, credp) != 0)
1461 			connp->conn_mac_mode = CONN_MAC_AWARE;
1462 
1463 		connp->conn_zone_is_global =
1464 		    (crgetzoneid(credp) == GLOBAL_ZONEID);
1465 	}
1466 
1467 	/* Initialize SCTP instance values,  our verf tag must never be 0 */
1468 	(void) random_get_pseudo_bytes((uint8_t *)&sctp->sctp_lvtag,
1469 	    sizeof (sctp->sctp_lvtag));
1470 	if (sctp->sctp_lvtag == 0)
1471 		sctp->sctp_lvtag = (uint32_t)gethrtime();
1472 	ASSERT(sctp->sctp_lvtag != 0);
1473 
1474 	sctp->sctp_ltsn = sctp->sctp_lvtag + 1;
1475 	sctp->sctp_lcsn = sctp->sctp_ltsn;
1476 	sctp->sctp_recovery_tsn = sctp->sctp_lastack_rxd = sctp->sctp_ltsn - 1;
1477 	sctp->sctp_adv_pap = sctp->sctp_lastack_rxd;
1478 
1479 	/* Information required by upper layer */
1480 	ASSERT(ulpd != NULL);
1481 	sctp->sctp_ulpd = ulpd;
1482 
1483 	ASSERT(upcalls != NULL);
1484 	sctp->sctp_upcalls = upcalls;
1485 	ASSERT(sbl != NULL);
1486 	/* Fill in the socket buffer limits for sctpsockfs */
1487 	sbl->sbl_txlowat = connp->conn_sndlowat;
1488 	sbl->sbl_txbuf = connp->conn_sndbuf;
1489 	sbl->sbl_rxbuf = sctp->sctp_rwnd;
1490 	sbl->sbl_rxlowat = SCTP_RECV_LOWATER;
1491 
1492 	/* Insert this in the global list. */
1493 	SCTP_LINK(sctp, sctps);
1494 
1495 	return (sctp);
1496 }
1497 
1498 /* Run at module load time */
1499 void
1500 sctp_ddi_g_init(void)
1501 {
1502 	/* Create sctp_t/conn_t cache */
1503 	sctp_conn_cache_init();
1504 
1505 	/* Create the faddr cache */
1506 	sctp_faddr_init();
1507 
1508 	/* Create the sets cache */
1509 	sctp_sets_init();
1510 
1511 	/* Create the PR-SCTP sets cache */
1512 	sctp_ftsn_sets_init();
1513 
1514 	/* Initialize tables used for CRC calculation */
1515 	sctp_crc32_init();
1516 
1517 	/*
1518 	 * We want to be informed each time a stack is created or
1519 	 * destroyed in the kernel, so we can maintain the
1520 	 * set of sctp_stack_t's.
1521 	 */
1522 	netstack_register(NS_SCTP, sctp_stack_init, NULL, sctp_stack_fini);
1523 }
1524 
1525 static void *
1526 sctp_stack_init(netstackid_t stackid, netstack_t *ns)
1527 {
1528 	sctp_stack_t	*sctps;
1529 	size_t		arrsz;
1530 	int		i;
1531 
1532 	sctps = kmem_zalloc(sizeof (*sctps), KM_SLEEP);
1533 	sctps->sctps_netstack = ns;
1534 
1535 	/* Initialize locks */
1536 	mutex_init(&sctps->sctps_g_lock, NULL, MUTEX_DEFAULT, NULL);
1537 	mutex_init(&sctps->sctps_epriv_port_lock, NULL, MUTEX_DEFAULT, NULL);
1538 	sctps->sctps_g_num_epriv_ports = SCTP_NUM_EPRIV_PORTS;
1539 	sctps->sctps_g_epriv_ports[0] = ULP_DEF_EPRIV_PORT1;
1540 	sctps->sctps_g_epriv_ports[1] = ULP_DEF_EPRIV_PORT2;
1541 
1542 	/* Initialize SCTP hash arrays. */
1543 	sctp_hash_init(sctps);
1544 
1545 	arrsz = sctp_propinfo_count * sizeof (mod_prop_info_t);
1546 	sctps->sctps_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz,
1547 	    KM_SLEEP);
1548 	bcopy(sctp_propinfo_tbl, sctps->sctps_propinfo_tbl, arrsz);
1549 
1550 	/* saddr init */
1551 	sctp_saddr_init(sctps);
1552 
1553 	/* Global SCTP PCB list. */
1554 	list_create(&sctps->sctps_g_list, sizeof (sctp_t),
1555 	    offsetof(sctp_t, sctp_list));
1556 
1557 	/* Initialize SCTP kstats. */
1558 	sctps->sctps_mibkp = sctp_kstat_init(stackid);
1559 	sctps->sctps_kstat = sctp_kstat2_init(stackid);
1560 
1561 	mutex_init(&sctps->sctps_reclaim_lock, NULL, MUTEX_DEFAULT, NULL);
1562 	sctps->sctps_reclaim = B_FALSE;
1563 	sctps->sctps_reclaim_tid = 0;
1564 	sctps->sctps_reclaim_period = sctps->sctps_rto_maxg;
1565 
1566 	/* Allocate the per netstack stats */
1567 	mutex_enter(&cpu_lock);
1568 	sctps->sctps_sc_cnt = MAX(ncpus, boot_ncpus);
1569 	mutex_exit(&cpu_lock);
1570 	sctps->sctps_sc = kmem_zalloc(max_ncpus  * sizeof (sctp_stats_cpu_t *),
1571 	    KM_SLEEP);
1572 	for (i = 0; i < sctps->sctps_sc_cnt; i++) {
1573 		sctps->sctps_sc[i] = kmem_zalloc(sizeof (sctp_stats_cpu_t),
1574 		    KM_SLEEP);
1575 	}
1576 
1577 	mutex_init(&sctps->sctps_listener_conf_lock, NULL, MUTEX_DEFAULT, NULL);
1578 	list_create(&sctps->sctps_listener_conf, sizeof (sctp_listener_t),
1579 	    offsetof(sctp_listener_t, sl_link));
1580 
1581 	return (sctps);
1582 }
1583 
1584 /*
1585  * Called when the module is about to be unloaded.
1586  */
1587 void
1588 sctp_ddi_g_destroy(void)
1589 {
1590 	/* Destroy sctp_t/conn_t caches */
1591 	sctp_conn_cache_fini();
1592 
1593 	/* Destroy the faddr cache */
1594 	sctp_faddr_fini();
1595 
1596 	/* Destroy the sets cache */
1597 	sctp_sets_fini();
1598 
1599 	/* Destroy the PR-SCTP sets cache */
1600 	sctp_ftsn_sets_fini();
1601 
1602 	netstack_unregister(NS_SCTP);
1603 }
1604 
1605 /*
1606  * Free the SCTP stack instance.
1607  */
1608 static void
1609 sctp_stack_fini(netstackid_t stackid, void *arg)
1610 {
1611 	sctp_stack_t *sctps = (sctp_stack_t *)arg;
1612 	int i;
1613 
1614 	/*
1615 	 * Set sctps_reclaim to false tells sctp_reclaim_timer() not to restart
1616 	 * the timer.
1617 	 */
1618 	mutex_enter(&sctps->sctps_reclaim_lock);
1619 	sctps->sctps_reclaim = B_FALSE;
1620 	mutex_exit(&sctps->sctps_reclaim_lock);
1621 	if (sctps->sctps_reclaim_tid != 0)
1622 		(void) untimeout(sctps->sctps_reclaim_tid);
1623 	mutex_destroy(&sctps->sctps_reclaim_lock);
1624 
1625 	sctp_listener_conf_cleanup(sctps);
1626 
1627 	kmem_free(sctps->sctps_propinfo_tbl,
1628 	    sctp_propinfo_count * sizeof (mod_prop_info_t));
1629 	sctps->sctps_propinfo_tbl = NULL;
1630 
1631 	/* Destroy the recvq taskqs. */
1632 	sctp_rq_tq_fini(sctps);
1633 
1634 	/* Destroy saddr  */
1635 	sctp_saddr_fini(sctps);
1636 
1637 	/* Global SCTP PCB list. */
1638 	list_destroy(&sctps->sctps_g_list);
1639 
1640 	/* Destroy SCTP hash arrays. */
1641 	sctp_hash_destroy(sctps);
1642 
1643 	/* Destroy SCTP kernel stats. */
1644 	for (i = 0; i < sctps->sctps_sc_cnt; i++)
1645 		kmem_free(sctps->sctps_sc[i], sizeof (sctp_stats_cpu_t));
1646 	kmem_free(sctps->sctps_sc, max_ncpus * sizeof (sctp_stats_cpu_t *));
1647 
1648 	sctp_kstat_fini(stackid, sctps->sctps_mibkp);
1649 	sctps->sctps_mibkp = NULL;
1650 	sctp_kstat2_fini(stackid, sctps->sctps_kstat);
1651 	sctps->sctps_kstat = NULL;
1652 
1653 	mutex_destroy(&sctps->sctps_g_lock);
1654 	mutex_destroy(&sctps->sctps_epriv_port_lock);
1655 
1656 	kmem_free(sctps, sizeof (*sctps));
1657 }
1658 
1659 static void
1660 sctp_rq_tq_init(sctp_stack_t *sctps)
1661 {
1662 	char tq_name[TASKQ_NAMELEN];
1663 	int thrs;
1664 	int max_tasks;
1665 
1666 	mutex_enter(&sctps->sctps_g_lock);
1667 	/* Someone may have beaten us in creating the taskqs. */
1668 	if (sctps->sctps_recvq_tq_list_cur_sz > 0) {
1669 		mutex_exit(&sctps->sctps_g_lock);
1670 		return;
1671 	}
1672 
1673 	thrs = MIN(sctp_recvq_tq_thr_max, MAX(sctp_recvq_tq_thr_min,
1674 	    MAX(ncpus, boot_ncpus)));
1675 	/*
1676 	 * Make sure that the maximum number of tasks is at least thrice as
1677 	 * large as the number of threads.
1678 	 */
1679 	max_tasks = MAX(sctp_recvq_tq_task_min, thrs) * 3;
1680 
1681 	/*
1682 	 * This helps differentiate the default taskqs in different IP stacks.
1683 	 */
1684 	(void) snprintf(tq_name, sizeof (tq_name), "sctp_def_rq_taskq_%d",
1685 	    sctps->sctps_netstack->netstack_stackid);
1686 
1687 	sctps->sctps_recvq_tq_list_max_sz = sctp_recvq_tq_list_max;
1688 	sctps->sctps_recvq_tq_list_cur_sz = 1;
1689 
1690 	/*
1691 	 * Initialize the recvq_tq_list and create the first recvq taskq.
1692 	 * What to do if it fails?
1693 	 */
1694 	sctps->sctps_recvq_tq_list =
1695 	    kmem_zalloc(sctps->sctps_recvq_tq_list_max_sz * sizeof (taskq_t *),
1696 	    KM_SLEEP);
1697 	sctps->sctps_recvq_tq_list[0] = taskq_create(tq_name, thrs,
1698 	    minclsyspri, sctp_recvq_tq_task_min, max_tasks, TASKQ_PREPOPULATE);
1699 	mutex_init(&sctps->sctps_rq_tq_lock, NULL, MUTEX_DEFAULT, NULL);
1700 
1701 	mutex_exit(&sctps->sctps_g_lock);
1702 }
1703 
1704 static void
1705 sctp_rq_tq_fini(sctp_stack_t *sctps)
1706 {
1707 	int i;
1708 
1709 	if (sctps->sctps_recvq_tq_list_cur_sz == 0)
1710 		return;
1711 
1712 	for (i = 0; i < sctps->sctps_recvq_tq_list_cur_sz; i++) {
1713 		ASSERT(sctps->sctps_recvq_tq_list[i] != NULL);
1714 		taskq_destroy(sctps->sctps_recvq_tq_list[i]);
1715 	}
1716 	kmem_free(sctps->sctps_recvq_tq_list,
1717 	    sctps->sctps_recvq_tq_list_max_sz * sizeof (taskq_t *));
1718 	sctps->sctps_recvq_tq_list = NULL;
1719 }
1720 
1721 /* Add another taskq for a new ill. */
1722 void
1723 sctp_inc_taskq(sctp_stack_t *sctps)
1724 {
1725 	taskq_t *tq;
1726 	char tq_name[TASKQ_NAMELEN];
1727 	int thrs;
1728 	int max_tasks;
1729 
1730 	thrs = MIN(sctp_recvq_tq_thr_max, MAX(sctp_recvq_tq_thr_min,
1731 	    MAX(ncpus, boot_ncpus)));
1732 	/*
1733 	 * Make sure that the maximum number of tasks is at least thrice as
1734 	 * large as the number of threads.
1735 	 */
1736 	max_tasks = MAX(sctp_recvq_tq_task_min, thrs) * 3;
1737 
1738 	mutex_enter(&sctps->sctps_rq_tq_lock);
1739 	if (sctps->sctps_recvq_tq_list_cur_sz + 1 >
1740 	    sctps->sctps_recvq_tq_list_max_sz) {
1741 		mutex_exit(&sctps->sctps_rq_tq_lock);
1742 		cmn_err(CE_NOTE, "Cannot create more SCTP recvq taskq");
1743 		return;
1744 	}
1745 
1746 	(void) snprintf(tq_name, sizeof (tq_name), "sctp_rq_taskq_%d_%u",
1747 	    sctps->sctps_netstack->netstack_stackid,
1748 	    sctps->sctps_recvq_tq_list_cur_sz);
1749 	tq = taskq_create(tq_name, thrs, minclsyspri, sctp_recvq_tq_task_min,
1750 	    max_tasks, TASKQ_PREPOPULATE);
1751 	if (tq == NULL) {
1752 		mutex_exit(&sctps->sctps_rq_tq_lock);
1753 		cmn_err(CE_NOTE, "SCTP recvq taskq creation failed");
1754 		return;
1755 	}
1756 	ASSERT(sctps->sctps_recvq_tq_list[
1757 	    sctps->sctps_recvq_tq_list_cur_sz] == NULL);
1758 	sctps->sctps_recvq_tq_list[sctps->sctps_recvq_tq_list_cur_sz] = tq;
1759 	atomic_add_32(&sctps->sctps_recvq_tq_list_cur_sz, 1);
1760 	mutex_exit(&sctps->sctps_rq_tq_lock);
1761 }
1762 
1763 #ifdef DEBUG
1764 uint32_t recvq_loop_cnt = 0;
1765 uint32_t recvq_call = 0;
1766 #endif
1767 
1768 /*
1769  * Find the next recvq_tq to use.  This routine will go thru all the
1770  * taskqs until it can dispatch a job for the sctp.  If this fails,
1771  * it will create a new taskq and try it.
1772  */
1773 static boolean_t
1774 sctp_find_next_tq(sctp_t *sctp)
1775 {
1776 	int next_tq, try;
1777 	taskq_t *tq;
1778 	sctp_stack_t	*sctps = sctp->sctp_sctps;
1779 
1780 	/*
1781 	 * Note that since we don't hold a lock on sctp_rq_tq_lock for
1782 	 * performance reason, recvq_ta_list_cur_sz can be changed during
1783 	 * this loop.  The problem this will create is that the loop may
1784 	 * not have tried all the recvq_tq.  This should be OK.
1785 	 */
1786 	next_tq = atomic_add_32_nv(&sctps->sctps_recvq_tq_list_cur, 1) %
1787 	    sctps->sctps_recvq_tq_list_cur_sz;
1788 	for (try = 0; try < sctps->sctps_recvq_tq_list_cur_sz; try++) {
1789 		tq = sctps->sctps_recvq_tq_list[next_tq];
1790 		if (taskq_dispatch(tq, sctp_process_recvq, sctp,
1791 		    TQ_NOSLEEP) != NULL) {
1792 			sctp->sctp_recvq_tq = tq;
1793 			return (B_TRUE);
1794 		}
1795 		next_tq = (next_tq + 1) % sctps->sctps_recvq_tq_list_cur_sz;
1796 	}
1797 
1798 	/*
1799 	 * Create one more taskq and try it.  Note that sctp_inc_taskq()
1800 	 * may not have created another taskq if the number of recvq
1801 	 * taskqs is at the maximum.  We are probably in a pretty bad
1802 	 * shape if this actually happens...
1803 	 */
1804 	sctp_inc_taskq(sctps);
1805 	tq = sctps->sctps_recvq_tq_list[sctps->sctps_recvq_tq_list_cur_sz - 1];
1806 	if (taskq_dispatch(tq, sctp_process_recvq, sctp, TQ_NOSLEEP) != NULL) {
1807 		sctp->sctp_recvq_tq = tq;
1808 		return (B_TRUE);
1809 	}
1810 	SCTP_KSTAT(sctps, sctp_find_next_tq);
1811 	return (B_FALSE);
1812 }
1813 
1814 /*
1815  * To add a message to the recvq.  Note that the sctp_timer_fire()
1816  * routine also uses this function to add the timer message to the
1817  * receive queue for later processing.  And it should be the only
1818  * caller of sctp_add_recvq() which sets the try_harder argument
1819  * to B_TRUE.
1820  *
1821  * If the try_harder argument is B_TRUE, this routine sctp_find_next_tq()
1822  * will try very hard to dispatch the task.  Refer to the comment
1823  * for that routine on how it does that.
1824  *
1825  * On failure the message has been freed i.e., this routine always consumes the
1826  * message. It bumps ipIfStatsInDiscards and and uses ip_drop_input to drop.
1827  */
1828 void
1829 sctp_add_recvq(sctp_t *sctp, mblk_t *mp, boolean_t caller_hold_lock,
1830     ip_recv_attr_t *ira)
1831 {
1832 	mblk_t	*attrmp;
1833 	ip_stack_t	*ipst = sctp->sctp_sctps->sctps_netstack->netstack_ip;
1834 
1835 	ASSERT(ira->ira_ill == NULL);
1836 
1837 	if (!caller_hold_lock)
1838 		mutex_enter(&sctp->sctp_recvq_lock);
1839 
1840 	/* If the taskq dispatch has not been scheduled, do it now. */
1841 	if (sctp->sctp_recvq_tq == NULL) {
1842 		ASSERT(sctp->sctp_recvq == NULL);
1843 		if (!sctp_find_next_tq(sctp)) {
1844 			if (!caller_hold_lock)
1845 				mutex_exit(&sctp->sctp_recvq_lock);
1846 			BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInDiscards);
1847 			ip_drop_input("ipIfStatsInDiscards", mp, NULL);
1848 			freemsg(mp);
1849 			return;
1850 		}
1851 		/* Make sure the sctp_t will not go away. */
1852 		SCTP_REFHOLD(sctp);
1853 	}
1854 
1855 	attrmp = ip_recv_attr_to_mblk(ira);
1856 	if (attrmp == NULL) {
1857 		if (!caller_hold_lock)
1858 			mutex_exit(&sctp->sctp_recvq_lock);
1859 		BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInDiscards);
1860 		ip_drop_input("ipIfStatsInDiscards", mp, NULL);
1861 		freemsg(mp);
1862 		return;
1863 	}
1864 	ASSERT(attrmp->b_cont == NULL);
1865 	attrmp->b_cont = mp;
1866 	mp = attrmp;
1867 
1868 	if (sctp->sctp_recvq == NULL) {
1869 		sctp->sctp_recvq = mp;
1870 		sctp->sctp_recvq_tail = mp;
1871 	} else {
1872 		sctp->sctp_recvq_tail->b_next = mp;
1873 		sctp->sctp_recvq_tail = mp;
1874 	}
1875 
1876 	if (!caller_hold_lock)
1877 		mutex_exit(&sctp->sctp_recvq_lock);
1878 }
1879 
1880 static void
1881 sctp_process_recvq(void *arg)
1882 {
1883 	sctp_t		*sctp = (sctp_t *)arg;
1884 	mblk_t		*mp;
1885 #ifdef DEBUG
1886 	uint32_t	loop_cnt = 0;
1887 #endif
1888 	ip_recv_attr_t	iras;
1889 
1890 #ifdef	_BIG_ENDIAN
1891 #define	IPVER(ip6h)	((((uint32_t *)ip6h)[0] >> 28) & 0x7)
1892 #else
1893 #define	IPVER(ip6h)	((((uint32_t *)ip6h)[0] >> 4) & 0x7)
1894 #endif
1895 
1896 	RUN_SCTP(sctp);
1897 	mutex_enter(&sctp->sctp_recvq_lock);
1898 
1899 #ifdef DEBUG
1900 	recvq_call++;
1901 #endif
1902 	/*
1903 	 * Note that while we are in this loop, other thread can put
1904 	 * new packets in the receive queue.  We may be looping for
1905 	 * quite a while.
1906 	 */
1907 	while ((mp = sctp->sctp_recvq) != NULL) {
1908 		mblk_t *data_mp;
1909 
1910 		sctp->sctp_recvq = mp->b_next;
1911 		mutex_exit(&sctp->sctp_recvq_lock);
1912 		mp->b_next = NULL;
1913 #ifdef DEBUG
1914 		loop_cnt++;
1915 #endif
1916 		mp->b_prev = NULL;
1917 
1918 		data_mp = mp->b_cont;
1919 		mp->b_cont = NULL;
1920 		if (!ip_recv_attr_from_mblk(mp, &iras)) {
1921 			ip_drop_input("ip_recv_attr_from_mblk", mp, NULL);
1922 			freemsg(mp);
1923 			ira_cleanup(&iras, B_TRUE);
1924 			continue;
1925 		}
1926 
1927 		if (iras.ira_flags & IRAF_ICMP_ERROR)
1928 			sctp_icmp_error(sctp, data_mp);
1929 		else
1930 			sctp_input_data(sctp, data_mp, &iras);
1931 
1932 		ira_cleanup(&iras, B_TRUE);
1933 		mutex_enter(&sctp->sctp_recvq_lock);
1934 	}
1935 
1936 	sctp->sctp_recvq_tail = NULL;
1937 	sctp->sctp_recvq_tq = NULL;
1938 
1939 	mutex_exit(&sctp->sctp_recvq_lock);
1940 
1941 	WAKE_SCTP(sctp);
1942 
1943 #ifdef DEBUG
1944 	if (loop_cnt > recvq_loop_cnt)
1945 		recvq_loop_cnt = loop_cnt;
1946 #endif
1947 	/* Now it can go away. */
1948 	SCTP_REFRELE(sctp);
1949 }
1950 
1951 /* ARGSUSED */
1952 static int
1953 sctp_conn_cache_constructor(void *buf, void *cdrarg, int kmflags)
1954 {
1955 	conn_t	*connp = (conn_t *)buf;
1956 	sctp_t	*sctp = (sctp_t *)&connp[1];
1957 	int	cnt;
1958 
1959 	bzero(connp, sizeof (conn_t));
1960 	bzero(buf, (char *)&sctp[1] - (char *)buf);
1961 
1962 	mutex_init(&sctp->sctp_reflock, NULL, MUTEX_DEFAULT, NULL);
1963 	mutex_init(&sctp->sctp_lock, NULL, MUTEX_DEFAULT, NULL);
1964 	mutex_init(&sctp->sctp_recvq_lock, NULL, MUTEX_DEFAULT, NULL);
1965 	cv_init(&sctp->sctp_cv, NULL, CV_DEFAULT, NULL);
1966 	for (cnt = 0; cnt < SCTP_IPIF_HASH; cnt++) {
1967 		rw_init(&sctp->sctp_saddrs[cnt].ipif_hash_lock, NULL,
1968 		    RW_DEFAULT, NULL);
1969 	}
1970 
1971 	mutex_init(&connp->conn_lock, NULL, MUTEX_DEFAULT, NULL);
1972 	cv_init(&connp->conn_cv, NULL, CV_DEFAULT, NULL);
1973 	connp->conn_flags = IPCL_SCTPCONN;
1974 	connp->conn_proto = IPPROTO_SCTP;
1975 	connp->conn_sctp = sctp;
1976 	sctp->sctp_connp = connp;
1977 	rw_init(&connp->conn_ilg_lock, NULL, RW_DEFAULT, NULL);
1978 
1979 	connp->conn_ixa = kmem_zalloc(sizeof (ip_xmit_attr_t), kmflags);
1980 	if (connp->conn_ixa == NULL) {
1981 		return (ENOMEM);
1982 	}
1983 	connp->conn_ixa->ixa_refcnt = 1;
1984 	connp->conn_ixa->ixa_protocol = connp->conn_proto;
1985 	connp->conn_ixa->ixa_xmit_hint = CONN_TO_XMIT_HINT(connp);
1986 	return (0);
1987 }
1988 
1989 /* ARGSUSED */
1990 static void
1991 sctp_conn_cache_destructor(void *buf, void *cdrarg)
1992 {
1993 	conn_t	*connp = (conn_t *)buf;
1994 	sctp_t	*sctp = (sctp_t *)&connp[1];
1995 	int	cnt;
1996 
1997 	ASSERT(sctp->sctp_connp == connp);
1998 	ASSERT(!MUTEX_HELD(&sctp->sctp_lock));
1999 	ASSERT(!MUTEX_HELD(&sctp->sctp_reflock));
2000 	ASSERT(!MUTEX_HELD(&sctp->sctp_recvq_lock));
2001 
2002 	ASSERT(sctp->sctp_conn_hash_next == NULL);
2003 	ASSERT(sctp->sctp_conn_hash_prev == NULL);
2004 	ASSERT(sctp->sctp_listen_hash_next == NULL);
2005 	ASSERT(sctp->sctp_listen_hash_prev == NULL);
2006 	ASSERT(sctp->sctp_listen_tfp == NULL);
2007 	ASSERT(sctp->sctp_conn_tfp == NULL);
2008 
2009 	ASSERT(sctp->sctp_faddrs == NULL);
2010 	ASSERT(sctp->sctp_nsaddrs == 0);
2011 
2012 	ASSERT(sctp->sctp_ulpd == NULL);
2013 
2014 	ASSERT(sctp->sctp_lastfaddr == NULL);
2015 	ASSERT(sctp->sctp_primary == NULL);
2016 	ASSERT(sctp->sctp_current == NULL);
2017 	ASSERT(sctp->sctp_lastdata == NULL);
2018 
2019 	ASSERT(sctp->sctp_xmit_head == NULL);
2020 	ASSERT(sctp->sctp_xmit_tail == NULL);
2021 	ASSERT(sctp->sctp_xmit_unsent == NULL);
2022 	ASSERT(sctp->sctp_xmit_unsent_tail == NULL);
2023 
2024 	ASSERT(sctp->sctp_ostrcntrs == NULL);
2025 
2026 	ASSERT(sctp->sctp_sack_info == NULL);
2027 	ASSERT(sctp->sctp_ack_mp == NULL);
2028 	ASSERT(sctp->sctp_instr == NULL);
2029 
2030 	ASSERT(sctp->sctp_iphc == NULL);
2031 	ASSERT(sctp->sctp_iphc6 == NULL);
2032 	ASSERT(sctp->sctp_ipha == NULL);
2033 	ASSERT(sctp->sctp_ip6h == NULL);
2034 	ASSERT(sctp->sctp_sctph == NULL);
2035 	ASSERT(sctp->sctp_sctph6 == NULL);
2036 
2037 	ASSERT(sctp->sctp_cookie_mp == NULL);
2038 
2039 	ASSERT(sctp->sctp_refcnt == 0);
2040 	ASSERT(sctp->sctp_timer_mp == NULL);
2041 	ASSERT(sctp->sctp_connp->conn_ref == 0);
2042 	ASSERT(sctp->sctp_heartbeat_mp == NULL);
2043 	ASSERT(sctp->sctp_ptpbhn == NULL && sctp->sctp_bind_hash == NULL);
2044 
2045 	ASSERT(sctp->sctp_shutdown_faddr == NULL);
2046 
2047 	ASSERT(sctp->sctp_cxmit_list == NULL);
2048 
2049 	ASSERT(sctp->sctp_recvq == NULL);
2050 	ASSERT(sctp->sctp_recvq_tail == NULL);
2051 	ASSERT(sctp->sctp_recvq_tq == NULL);
2052 
2053 	/*
2054 	 * sctp_pad_mp can be NULL if the memory allocation fails
2055 	 * in sctp_init_values() and the conn_t is freed.
2056 	 */
2057 	if (sctp->sctp_pad_mp != NULL) {
2058 		freeb(sctp->sctp_pad_mp);
2059 		sctp->sctp_pad_mp = NULL;
2060 	}
2061 
2062 	mutex_destroy(&sctp->sctp_reflock);
2063 	mutex_destroy(&sctp->sctp_lock);
2064 	mutex_destroy(&sctp->sctp_recvq_lock);
2065 	cv_destroy(&sctp->sctp_cv);
2066 	for (cnt = 0; cnt < SCTP_IPIF_HASH; cnt++) {
2067 		rw_destroy(&sctp->sctp_saddrs[cnt].ipif_hash_lock);
2068 	}
2069 
2070 	mutex_destroy(&connp->conn_lock);
2071 	cv_destroy(&connp->conn_cv);
2072 	rw_destroy(&connp->conn_ilg_lock);
2073 
2074 	/* Can be NULL if constructor failed */
2075 	if (connp->conn_ixa != NULL) {
2076 		ASSERT(connp->conn_ixa->ixa_refcnt == 1);
2077 		ASSERT(connp->conn_ixa->ixa_ire == NULL);
2078 		ASSERT(connp->conn_ixa->ixa_nce == NULL);
2079 		ixa_refrele(connp->conn_ixa);
2080 	}
2081 }
2082 
2083 static void
2084 sctp_conn_cache_init()
2085 {
2086 	sctp_conn_cache = kmem_cache_create("sctp_conn_cache",
2087 	    sizeof (sctp_t) + sizeof (conn_t), 0, sctp_conn_cache_constructor,
2088 	    sctp_conn_cache_destructor, sctp_conn_reclaim, NULL, NULL, 0);
2089 }
2090 
2091 static void
2092 sctp_conn_cache_fini()
2093 {
2094 	kmem_cache_destroy(sctp_conn_cache);
2095 }
2096 
2097 void
2098 sctp_conn_init(conn_t *connp)
2099 {
2100 	ASSERT(connp->conn_flags == IPCL_SCTPCONN);
2101 	connp->conn_rq = connp->conn_wq = NULL;
2102 	connp->conn_ixa->ixa_flags |= IXAF_SET_ULP_CKSUM | IXAF_VERIFY_SOURCE |
2103 	    IXAF_VERIFY_PMTU;
2104 
2105 	ASSERT(connp->conn_proto == IPPROTO_SCTP);
2106 	ASSERT(connp->conn_ixa->ixa_protocol == connp->conn_proto);
2107 	connp->conn_state_flags |= CONN_INCIPIENT;
2108 
2109 	ASSERT(connp->conn_sctp != NULL);
2110 
2111 	/*
2112 	 * Register sctp_notify to listen to capability changes detected by IP.
2113 	 * This upcall is made in the context of the call to conn_ip_output
2114 	 * thus it holds whatever locks sctp holds across conn_ip_output.
2115 	 */
2116 	connp->conn_ixa->ixa_notify = sctp_notify;
2117 	connp->conn_ixa->ixa_notify_cookie = connp->conn_sctp;
2118 }
2119 
2120 static void
2121 sctp_conn_clear(conn_t *connp)
2122 {
2123 	/* Clean up conn_t stuff */
2124 	if (connp->conn_latch != NULL) {
2125 		IPLATCH_REFRELE(connp->conn_latch);
2126 		connp->conn_latch = NULL;
2127 	}
2128 	if (connp->conn_latch_in_policy != NULL) {
2129 		IPPOL_REFRELE(connp->conn_latch_in_policy);
2130 		connp->conn_latch_in_policy = NULL;
2131 	}
2132 	if (connp->conn_latch_in_action != NULL) {
2133 		IPACT_REFRELE(connp->conn_latch_in_action);
2134 		connp->conn_latch_in_action = NULL;
2135 	}
2136 	if (connp->conn_policy != NULL) {
2137 		IPPH_REFRELE(connp->conn_policy, connp->conn_netstack);
2138 		connp->conn_policy = NULL;
2139 	}
2140 	if (connp->conn_ipsec_opt_mp != NULL) {
2141 		freemsg(connp->conn_ipsec_opt_mp);
2142 		connp->conn_ipsec_opt_mp = NULL;
2143 	}
2144 	netstack_rele(connp->conn_netstack);
2145 	connp->conn_netstack = NULL;
2146 
2147 	/* Leave conn_ixa and other constructed fields in place */
2148 	ipcl_conn_cleanup(connp);
2149 }
2150