xref: /illumos-gate/usr/src/uts/common/fs/sockfs/sockstr.c (revision d019449136cec9f203f106de418421095790e4e2)
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 2010 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  */
26 
27 #include <sys/types.h>
28 #include <sys/inttypes.h>
29 #include <sys/t_lock.h>
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/buf.h>
33 #include <sys/conf.h>
34 #include <sys/cred.h>
35 #include <sys/kmem.h>
36 #include <sys/sysmacros.h>
37 #include <sys/vfs.h>
38 #include <sys/vnode.h>
39 #include <sys/debug.h>
40 #include <sys/errno.h>
41 #include <sys/time.h>
42 #include <sys/file.h>
43 #include <sys/user.h>
44 #include <sys/stream.h>
45 #include <sys/strsubr.h>
46 #include <sys/esunddi.h>
47 #include <sys/flock.h>
48 #include <sys/modctl.h>
49 #include <sys/vtrace.h>
50 #include <sys/strsun.h>
51 #include <sys/cmn_err.h>
52 #include <sys/proc.h>
53 #include <sys/ddi.h>
54 
55 #include <sys/suntpi.h>
56 #include <sys/socket.h>
57 #include <sys/sockio.h>
58 #include <sys/socketvar.h>
59 #include <netinet/in.h>
60 #include <inet/common.h>
61 #include <inet/proto_set.h>
62 
63 #include <sys/tiuser.h>
64 #define	_SUN_TPI_VERSION	2
65 #include <sys/tihdr.h>
66 
67 #include <inet/kssl/ksslapi.h>
68 
69 #include <c2/audit.h>
70 
71 #include <fs/sockfs/socktpi.h>
72 #include <fs/sockfs/socktpi_impl.h>
73 
74 int so_default_version = SOV_SOCKSTREAM;
75 
76 #ifdef DEBUG
77 /* Set sockdebug to print debug messages when SO_DEBUG is set */
78 int sockdebug = 0;
79 
80 /* Set sockprinterr to print error messages when SO_DEBUG is set */
81 int sockprinterr = 0;
82 
83 /*
84  * Set so_default_options to SO_DEBUG is all sockets should be created
85  * with SO_DEBUG set. This is needed to get debug printouts from the
86  * socket() call itself.
87  */
88 int so_default_options = 0;
89 #endif /* DEBUG */
90 
91 #ifdef SOCK_TEST
92 /*
93  * Set to number of ticks to limit cv_waits for code coverage testing.
94  * Set to 1000 when SO_DEBUG is set to 2.
95  */
96 clock_t sock_test_timelimit = 0;
97 #endif /* SOCK_TEST */
98 
99 /*
100  * For concurrency testing of e.g. opening /dev/ip which does not
101  * handle T_INFO_REQ messages.
102  */
103 int so_no_tinfo = 0;
104 
105 /*
106  * Timeout for getting a T_CAPABILITY_ACK - it is possible for a provider
107  * to simply ignore the T_CAPABILITY_REQ.
108  */
109 clock_t	sock_capability_timeout	= 2;	/* seconds */
110 
111 static int	do_tcapability(struct sonode *so, t_uscalar_t cap_bits1);
112 static void	so_removehooks(struct sonode *so);
113 
114 static mblk_t *strsock_proto(vnode_t *vp, mblk_t *mp,
115 		strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
116 		strsigset_t *allmsgsigs, strpollset_t *pollwakeups);
117 static mblk_t *strsock_misc(vnode_t *vp, mblk_t *mp,
118 		strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
119 		strsigset_t *allmsgsigs, strpollset_t *pollwakeups);
120 
121 /*
122  * Convert a socket to a stream. Invoked when the illusory sockmod
123  * is popped from the stream.
124  * Change the stream head back to default operation without losing
125  * any messages (T_conn_ind's are moved to the stream head queue).
126  */
127 int
128 so_sock2stream(struct sonode *so)
129 {
130 	struct vnode		*vp = SOTOV(so);
131 	queue_t			*rq;
132 	mblk_t			*mp;
133 	int			error = 0;
134 	sotpi_info_t		*sti = SOTOTPI(so);
135 
136 	ASSERT(MUTEX_HELD(&sti->sti_plumb_lock));
137 
138 	mutex_enter(&so->so_lock);
139 	so_lock_single(so);
140 
141 	ASSERT(so->so_version != SOV_STREAM);
142 
143 	if (sti->sti_direct) {
144 		mblk_t **mpp;
145 		int rval;
146 
147 		/*
148 		 * Tell the transport below that sockmod is being popped
149 		 */
150 		mutex_exit(&so->so_lock);
151 		error = strioctl(vp, _SIOCSOCKFALLBACK, 0, 0, K_TO_K, CRED(),
152 		    &rval);
153 		mutex_enter(&so->so_lock);
154 		if (error != 0) {
155 			dprintso(so, 0, ("so_sock2stream(%p): "
156 			    "_SIOCSOCKFALLBACK failed\n", (void *)so));
157 			goto exit;
158 		}
159 		sti->sti_direct = 0;
160 
161 		for (mpp = &sti->sti_conn_ind_head; (mp = *mpp) != NULL;
162 		    mpp = &mp->b_next) {
163 			struct T_conn_ind	*conn_ind;
164 
165 			/*
166 			 * strsock_proto() has already verified the length of
167 			 * this message block.
168 			 */
169 			ASSERT(MBLKL(mp) >= sizeof (struct T_conn_ind));
170 
171 			conn_ind = (struct T_conn_ind *)mp->b_rptr;
172 			if (conn_ind->OPT_length == 0 &&
173 			    conn_ind->OPT_offset == 0)
174 				continue;
175 
176 			if (DB_REF(mp) > 1) {
177 				mblk_t	*newmp;
178 				size_t	length;
179 				cred_t	*cr;
180 				pid_t	cpid;
181 				int error;	/* Dummy - error not returned */
182 
183 				/*
184 				 * Copy the message block because it is used
185 				 * elsewhere, too.
186 				 * Can't use copyb since we want to wait
187 				 * yet allow for EINTR.
188 				 */
189 				/* Round up size for reuse */
190 				length = MAX(MBLKL(mp), 64);
191 				cr = msg_getcred(mp, &cpid);
192 				if (cr != NULL) {
193 					newmp = allocb_cred_wait(length, 0,
194 					    &error, cr, cpid);
195 				} else {
196 					newmp = allocb_wait(length, 0, 0,
197 					    &error);
198 				}
199 				if (newmp == NULL) {
200 					error = EINTR;
201 					goto exit;
202 				}
203 				bcopy(mp->b_rptr, newmp->b_wptr, length);
204 				newmp->b_wptr += length;
205 				newmp->b_next = mp->b_next;
206 
207 				/*
208 				 * Link the new message block into the queue
209 				 * and free the old one.
210 				 */
211 				*mpp = newmp;
212 				mp->b_next = NULL;
213 				freemsg(mp);
214 
215 				mp = newmp;
216 				conn_ind = (struct T_conn_ind *)mp->b_rptr;
217 			}
218 
219 			/*
220 			 * Remove options added by TCP for accept fast-path.
221 			 */
222 			conn_ind->OPT_length = 0;
223 			conn_ind->OPT_offset = 0;
224 		}
225 	}
226 
227 	so->so_version = SOV_STREAM;
228 	so->so_proto_handle = NULL;
229 
230 	/*
231 	 * Remove the hooks in the stream head to avoid queuing more
232 	 * packets in sockfs.
233 	 */
234 	mutex_exit(&so->so_lock);
235 	so_removehooks(so);
236 	mutex_enter(&so->so_lock);
237 
238 	/*
239 	 * Clear any state related to urgent data. Leave any T_EXDATA_IND
240 	 * on the queue - the behavior of urgent data after a switch is
241 	 * left undefined.
242 	 */
243 	so->so_error = sti->sti_delayed_error = 0;
244 	freemsg(so->so_oobmsg);
245 	so->so_oobmsg = NULL;
246 	sti->sti_oobsigcnt = sti->sti_oobcnt = 0;
247 
248 	so->so_state &= ~(SS_RCVATMARK|SS_OOBPEND|SS_HAVEOOBDATA|SS_HADOOBDATA|
249 	    SS_SAVEDEOR);
250 	ASSERT(so_verify_oobstate(so));
251 
252 	freemsg(sti->sti_ack_mp);
253 	sti->sti_ack_mp = NULL;
254 
255 	/*
256 	 * Flush the T_DISCON_IND on sti_discon_ind_mp.
257 	 */
258 	so_flush_discon_ind(so);
259 
260 	/*
261 	 * Move any queued T_CONN_IND messages to stream head queue.
262 	 */
263 	rq = RD(strvp2wq(vp));
264 	while ((mp = sti->sti_conn_ind_head) != NULL) {
265 		sti->sti_conn_ind_head = mp->b_next;
266 		mp->b_next = NULL;
267 		if (sti->sti_conn_ind_head == NULL) {
268 			ASSERT(sti->sti_conn_ind_tail == mp);
269 			sti->sti_conn_ind_tail = NULL;
270 		}
271 		dprintso(so, 0,
272 		    ("so_sock2stream(%p): moving T_CONN_IND\n", (void *)so));
273 
274 		/* Drop lock across put() */
275 		mutex_exit(&so->so_lock);
276 		put(rq, mp);
277 		mutex_enter(&so->so_lock);
278 	}
279 
280 exit:
281 	ASSERT(MUTEX_HELD(&so->so_lock));
282 	so_unlock_single(so, SOLOCKED);
283 	mutex_exit(&so->so_lock);
284 	return (error);
285 }
286 
287 /*
288  * Covert a stream back to a socket. This is invoked when the illusory
289  * sockmod is pushed on a stream (where the stream was "created" by
290  * popping the illusory sockmod).
291  * This routine can not recreate the socket state (certain aspects of
292  * it like urgent data state and the bound/connected addresses for AF_UNIX
293  * sockets can not be recreated by asking the transport for information).
294  * Thus this routine implicitly assumes that the socket is in an initial
295  * state (as if it was just created). It flushes any messages queued on the
296  * read queue to avoid dealing with e.g. TPI acks or T_exdata_ind messages.
297  */
298 void
299 so_stream2sock(struct sonode *so)
300 {
301 	struct vnode *vp = SOTOV(so);
302 	sotpi_info_t *sti = SOTOTPI(so);
303 
304 	ASSERT(MUTEX_HELD(&sti->sti_plumb_lock));
305 
306 	mutex_enter(&so->so_lock);
307 	so_lock_single(so);
308 	ASSERT(so->so_version == SOV_STREAM);
309 	so->so_version = SOV_SOCKSTREAM;
310 	sti->sti_pushcnt = 0;
311 	mutex_exit(&so->so_lock);
312 
313 	/*
314 	 * Set a permenent error to force any thread in sorecvmsg to
315 	 * return (and drop SOREADLOCKED). Clear the error once
316 	 * we have SOREADLOCKED.
317 	 * This makes a read sleeping during the I_PUSH of sockmod return
318 	 * EIO.
319 	 */
320 	strsetrerror(SOTOV(so), EIO, 1, NULL);
321 
322 	/*
323 	 * Get the read lock before flushing data to avoid
324 	 * problems with the T_EXDATA_IND MSG_PEEK code in sorecvmsg.
325 	 */
326 	mutex_enter(&so->so_lock);
327 	(void) so_lock_read(so, 0);	/* Set SOREADLOCKED */
328 	mutex_exit(&so->so_lock);
329 
330 	strsetrerror(SOTOV(so), 0, 0, NULL);
331 	so_installhooks(so);
332 
333 	/*
334 	 * Flush everything on the read queue.
335 	 * This ensures that no T_CONN_IND remain and that no T_EXDATA_IND
336 	 * remain; those types of messages would confuse sockfs.
337 	 */
338 	strflushrq(vp, FLUSHALL);
339 	mutex_enter(&so->so_lock);
340 
341 	/*
342 	 * Flush the T_DISCON_IND on sti_discon_ind_mp.
343 	 */
344 	so_flush_discon_ind(so);
345 	so_unlock_read(so);	/* Clear SOREADLOCKED */
346 
347 	so_unlock_single(so, SOLOCKED);
348 	mutex_exit(&so->so_lock);
349 }
350 
351 /*
352  * Install the hooks in the stream head.
353  */
354 void
355 so_installhooks(struct sonode *so)
356 {
357 	struct vnode *vp = SOTOV(so);
358 
359 	strsetrputhooks(vp, SH_SIGALLDATA | SH_IGN_ZEROLEN | SH_CONSOL_DATA,
360 	    strsock_proto, strsock_misc);
361 	strsetwputhooks(vp, SH_SIGPIPE | SH_RECHECK_ERR, 0);
362 }
363 
364 /*
365  * Remove the hooks in the stream head.
366  */
367 static void
368 so_removehooks(struct sonode *so)
369 {
370 	struct vnode *vp = SOTOV(so);
371 
372 	strsetrputhooks(vp, 0, NULL, NULL);
373 	strsetwputhooks(vp, 0, STRTIMOUT);
374 	/*
375 	 * Leave read behavior as it would have been for a normal
376 	 * stream i.e. a read of an M_PROTO will fail.
377 	 */
378 }
379 
380 void
381 so_basic_strinit(struct sonode *so)
382 {
383 	struct vnode *vp = SOTOV(so);
384 	struct stdata *stp;
385 	mblk_t *mp;
386 	sotpi_info_t *sti = SOTOTPI(so);
387 
388 	/* Preallocate an unbind_req message */
389 	mp = soallocproto(sizeof (struct T_unbind_req), _ALLOC_SLEEP, CRED());
390 	mutex_enter(&so->so_lock);
391 	sti->sti_unbind_mp = mp;
392 #ifdef DEBUG
393 	so->so_options = so_default_options;
394 #endif /* DEBUG */
395 	mutex_exit(&so->so_lock);
396 
397 	so_installhooks(so);
398 
399 	stp = vp->v_stream;
400 	/*
401 	 * Have to keep minpsz at zero in order to allow write/send of zero
402 	 * bytes.
403 	 */
404 	mutex_enter(&stp->sd_lock);
405 	if (stp->sd_qn_minpsz == 1)
406 		stp->sd_qn_minpsz = 0;
407 	mutex_exit(&stp->sd_lock);
408 }
409 
410 /*
411  * Initialize the streams side of a socket including
412  * T_info_req/ack processing. If tso is not NULL its values are used thereby
413  * avoiding the T_INFO_REQ.
414  */
415 int
416 so_strinit(struct sonode *so, struct sonode *tso)
417 {
418 	sotpi_info_t *sti = SOTOTPI(so);
419 	sotpi_info_t *tsti;
420 	int error;
421 
422 	so_basic_strinit(so);
423 
424 	/*
425 	 * The T_CAPABILITY_REQ should be the first message sent down because
426 	 * at least TCP has a fast-path for this which avoids timeouts while
427 	 * waiting for the T_CAPABILITY_ACK under high system load.
428 	 */
429 	if (tso == NULL) {
430 		error = do_tcapability(so, TC1_ACCEPTOR_ID | TC1_INFO);
431 		if (error)
432 			return (error);
433 	} else {
434 		tsti = SOTOTPI(tso);
435 
436 		mutex_enter(&so->so_lock);
437 		sti->sti_tsdu_size = tsti->sti_tsdu_size;
438 		sti->sti_etsdu_size = tsti->sti_etsdu_size;
439 		sti->sti_addr_size = tsti->sti_addr_size;
440 		sti->sti_opt_size = tsti->sti_opt_size;
441 		sti->sti_tidu_size = tsti->sti_tidu_size;
442 		sti->sti_serv_type = tsti->sti_serv_type;
443 		so->so_mode = tso->so_mode & ~SM_ACCEPTOR_ID;
444 		mutex_exit(&so->so_lock);
445 
446 		/* the following do_tcapability may update so->so_mode */
447 		if ((tsti->sti_serv_type != T_CLTS) &&
448 		    (sti->sti_direct == 0)) {
449 			error = do_tcapability(so, TC1_ACCEPTOR_ID);
450 			if (error)
451 				return (error);
452 		}
453 	}
454 	/*
455 	 * If the addr_size is 0 we treat it as already bound
456 	 * and connected. This is used by the routing socket.
457 	 * We set the addr_size to something to allocate a the address
458 	 * structures.
459 	 */
460 	if (sti->sti_addr_size == 0) {
461 		so->so_state |= SS_ISBOUND | SS_ISCONNECTED;
462 		/* Address size can vary with address families. */
463 		if (so->so_family == AF_INET6)
464 			sti->sti_addr_size =
465 			    (t_scalar_t)sizeof (struct sockaddr_in6);
466 		else
467 			sti->sti_addr_size =
468 			    (t_scalar_t)sizeof (struct sockaddr_in);
469 		ASSERT(sti->sti_unbind_mp);
470 	}
471 
472 	so_alloc_addr(so, sti->sti_addr_size);
473 
474 	return (0);
475 }
476 
477 static void
478 copy_tinfo(struct sonode *so, struct T_info_ack *tia)
479 {
480 	sotpi_info_t *sti = SOTOTPI(so);
481 
482 	sti->sti_tsdu_size = tia->TSDU_size;
483 	sti->sti_etsdu_size = tia->ETSDU_size;
484 	sti->sti_addr_size = tia->ADDR_size;
485 	sti->sti_opt_size = tia->OPT_size;
486 	sti->sti_tidu_size = tia->TIDU_size;
487 	sti->sti_serv_type = tia->SERV_type;
488 	switch (tia->CURRENT_state) {
489 	case TS_UNBND:
490 		break;
491 	case TS_IDLE:
492 		so->so_state |= SS_ISBOUND;
493 		sti->sti_laddr_len = 0;
494 		sti->sti_laddr_valid = 0;
495 		break;
496 	case TS_DATA_XFER:
497 		so->so_state |= SS_ISBOUND|SS_ISCONNECTED;
498 		sti->sti_laddr_len = 0;
499 		sti->sti_faddr_len = 0;
500 		sti->sti_laddr_valid = 0;
501 		sti->sti_faddr_valid = 0;
502 		break;
503 	}
504 
505 	/*
506 	 * Heuristics for determining the socket mode flags
507 	 * (SM_ATOMIC, SM_CONNREQUIRED, SM_ADDR, SM_FDPASSING,
508 	 * and SM_EXDATA, SM_OPTDATA, and SM_BYTESTREAM)
509 	 * from the info ack.
510 	 */
511 	if (sti->sti_serv_type == T_CLTS) {
512 		so->so_mode |= SM_ATOMIC | SM_ADDR;
513 	} else {
514 		so->so_mode |= SM_CONNREQUIRED;
515 		if (sti->sti_etsdu_size != 0 && sti->sti_etsdu_size != -2)
516 			so->so_mode |= SM_EXDATA;
517 	}
518 	if (so->so_type == SOCK_SEQPACKET || so->so_type == SOCK_RAW) {
519 		/* Semantics are to discard tail end of messages */
520 		so->so_mode |= SM_ATOMIC;
521 	}
522 	if (so->so_family == AF_UNIX) {
523 		so->so_mode |= SM_FDPASSING | SM_OPTDATA;
524 		if (sti->sti_addr_size == -1) {
525 			/* MAXPATHLEN + soun_family + nul termination */
526 			sti->sti_addr_size = (t_scalar_t)(MAXPATHLEN +
527 			    sizeof (short) + 1);
528 		}
529 		if (so->so_type == SOCK_STREAM) {
530 			/*
531 			 * Make it into a byte-stream transport.
532 			 * SOCK_SEQPACKET sockets are unchanged.
533 			 */
534 			sti->sti_tsdu_size = 0;
535 		}
536 	} else if (sti->sti_addr_size == -1) {
537 		/*
538 		 * Logic extracted from sockmod - have to pick some max address
539 		 * length in order to preallocate the addresses.
540 		 */
541 		sti->sti_addr_size = SOA_DEFSIZE;
542 	}
543 	if (sti->sti_tsdu_size == 0)
544 		so->so_mode |= SM_BYTESTREAM;
545 }
546 
547 static int
548 check_tinfo(struct sonode *so)
549 {
550 	sotpi_info_t *sti = SOTOTPI(so);
551 
552 	/* Consistency checks */
553 	if (so->so_type == SOCK_DGRAM && sti->sti_serv_type != T_CLTS) {
554 		eprintso(so, ("service type and socket type mismatch\n"));
555 		eprintsoline(so, EPROTO);
556 		return (EPROTO);
557 	}
558 	if (so->so_type == SOCK_STREAM && sti->sti_serv_type == T_CLTS) {
559 		eprintso(so, ("service type and socket type mismatch\n"));
560 		eprintsoline(so, EPROTO);
561 		return (EPROTO);
562 	}
563 	if (so->so_type == SOCK_SEQPACKET && sti->sti_serv_type == T_CLTS) {
564 		eprintso(so, ("service type and socket type mismatch\n"));
565 		eprintsoline(so, EPROTO);
566 		return (EPROTO);
567 	}
568 	if (so->so_family == AF_INET &&
569 	    sti->sti_addr_size != (t_scalar_t)sizeof (struct sockaddr_in)) {
570 		eprintso(so,
571 		    ("AF_INET must have sockaddr_in address length. Got %d\n",
572 		    sti->sti_addr_size));
573 		eprintsoline(so, EMSGSIZE);
574 		return (EMSGSIZE);
575 	}
576 	if (so->so_family == AF_INET6 &&
577 	    sti->sti_addr_size != (t_scalar_t)sizeof (struct sockaddr_in6)) {
578 		eprintso(so,
579 		    ("AF_INET6 must have sockaddr_in6 address length. Got %d\n",
580 		    sti->sti_addr_size));
581 		eprintsoline(so, EMSGSIZE);
582 		return (EMSGSIZE);
583 	}
584 
585 	dprintso(so, 1, (
586 	    "tinfo: serv %d tsdu %d, etsdu %d, addr %d, opt %d, tidu %d\n",
587 	    sti->sti_serv_type, sti->sti_tsdu_size, sti->sti_etsdu_size,
588 	    sti->sti_addr_size, sti->sti_opt_size,
589 	    sti->sti_tidu_size));
590 	dprintso(so, 1, ("tinfo: so_state %s\n",
591 	    pr_state(so->so_state, so->so_mode)));
592 	return (0);
593 }
594 
595 /*
596  * Send down T_info_req and wait for the ack.
597  * Record interesting T_info_ack values in the sonode.
598  */
599 static int
600 do_tinfo(struct sonode *so)
601 {
602 	struct T_info_req tir;
603 	mblk_t *mp;
604 	int error;
605 
606 	ASSERT(MUTEX_NOT_HELD(&so->so_lock));
607 
608 	if (so_no_tinfo) {
609 		SOTOTPI(so)->sti_addr_size = 0;
610 		return (0);
611 	}
612 
613 	dprintso(so, 1, ("do_tinfo(%p)\n", (void *)so));
614 
615 	/* Send T_INFO_REQ */
616 	tir.PRIM_type = T_INFO_REQ;
617 	mp = soallocproto1(&tir, sizeof (tir),
618 	    sizeof (struct T_info_req) + sizeof (struct T_info_ack),
619 	    _ALLOC_INTR, CRED());
620 	if (mp == NULL) {
621 		eprintsoline(so, ENOBUFS);
622 		return (ENOBUFS);
623 	}
624 	/* T_INFO_REQ has to be M_PCPROTO */
625 	DB_TYPE(mp) = M_PCPROTO;
626 
627 	error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0,
628 	    MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0);
629 	if (error) {
630 		eprintsoline(so, error);
631 		return (error);
632 	}
633 	mutex_enter(&so->so_lock);
634 	/* Wait for T_INFO_ACK */
635 	if ((error = sowaitprim(so, T_INFO_REQ, T_INFO_ACK,
636 	    (t_uscalar_t)sizeof (struct T_info_ack), &mp, 0))) {
637 		mutex_exit(&so->so_lock);
638 		eprintsoline(so, error);
639 		return (error);
640 	}
641 
642 	ASSERT(mp);
643 	copy_tinfo(so, (struct T_info_ack *)mp->b_rptr);
644 	mutex_exit(&so->so_lock);
645 	freemsg(mp);
646 	return (check_tinfo(so));
647 }
648 
649 /*
650  * Send down T_capability_req and wait for the ack.
651  * Record interesting T_capability_ack values in the sonode.
652  */
653 static int
654 do_tcapability(struct sonode *so, t_uscalar_t cap_bits1)
655 {
656 	struct T_capability_req tcr;
657 	struct T_capability_ack *tca;
658 	mblk_t *mp;
659 	int error;
660 	sotpi_info_t *sti = SOTOTPI(so);
661 
662 	ASSERT(cap_bits1 != 0);
663 	ASSERT((cap_bits1 & ~(TC1_ACCEPTOR_ID | TC1_INFO)) == 0);
664 	ASSERT(MUTEX_NOT_HELD(&so->so_lock));
665 
666 	if (sti->sti_provinfo->tpi_capability == PI_NO)
667 		return (do_tinfo(so));
668 
669 	if (so_no_tinfo) {
670 		sti->sti_addr_size = 0;
671 		if ((cap_bits1 &= ~TC1_INFO) == 0)
672 			return (0);
673 	}
674 
675 	dprintso(so, 1, ("do_tcapability(%p)\n", (void *)so));
676 
677 	/* Send T_CAPABILITY_REQ */
678 	tcr.PRIM_type = T_CAPABILITY_REQ;
679 	tcr.CAP_bits1 = cap_bits1;
680 	mp = soallocproto1(&tcr, sizeof (tcr),
681 	    sizeof (struct T_capability_req) + sizeof (struct T_capability_ack),
682 	    _ALLOC_INTR, CRED());
683 	if (mp == NULL) {
684 		eprintsoline(so, ENOBUFS);
685 		return (ENOBUFS);
686 	}
687 	/* T_CAPABILITY_REQ should be M_PCPROTO here */
688 	DB_TYPE(mp) = M_PCPROTO;
689 
690 	error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0,
691 	    MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0);
692 	if (error) {
693 		eprintsoline(so, error);
694 		return (error);
695 	}
696 	mutex_enter(&so->so_lock);
697 	/* Wait for T_CAPABILITY_ACK */
698 	if ((error = sowaitprim(so, T_CAPABILITY_REQ, T_CAPABILITY_ACK,
699 	    (t_uscalar_t)sizeof (*tca), &mp, sock_capability_timeout * hz))) {
700 		mutex_exit(&so->so_lock);
701 		PI_PROVLOCK(sti->sti_provinfo);
702 		if (sti->sti_provinfo->tpi_capability == PI_DONTKNOW)
703 			sti->sti_provinfo->tpi_capability = PI_NO;
704 		PI_PROVUNLOCK(sti->sti_provinfo);
705 		ASSERT((so->so_mode & SM_ACCEPTOR_ID) == 0);
706 		if (cap_bits1 & TC1_INFO) {
707 			/*
708 			 * If the T_CAPABILITY_REQ timed out and then a
709 			 * T_INFO_REQ gets a protocol error, most likely
710 			 * the capability was slow (vs. unsupported). Return
711 			 * ENOSR for this case as a best guess.
712 			 */
713 			if (error == ETIME) {
714 				return ((error = do_tinfo(so)) == EPROTO ?
715 				    ENOSR : error);
716 			}
717 			return (do_tinfo(so));
718 		}
719 		return (0);
720 	}
721 
722 	ASSERT(mp);
723 	tca = (struct T_capability_ack *)mp->b_rptr;
724 
725 	ASSERT((cap_bits1 & TC1_INFO) == (tca->CAP_bits1 & TC1_INFO));
726 	so_proc_tcapability_ack(so, tca);
727 
728 	cap_bits1 = tca->CAP_bits1;
729 
730 	mutex_exit(&so->so_lock);
731 	freemsg(mp);
732 
733 	if (cap_bits1 & TC1_INFO)
734 		return (check_tinfo(so));
735 
736 	return (0);
737 }
738 
739 /*
740  * Process a T_CAPABILITY_ACK
741  */
742 void
743 so_proc_tcapability_ack(struct sonode *so, struct T_capability_ack *tca)
744 {
745 	sotpi_info_t *sti = SOTOTPI(so);
746 
747 	if (sti->sti_provinfo->tpi_capability == PI_DONTKNOW) {
748 		PI_PROVLOCK(sti->sti_provinfo);
749 		sti->sti_provinfo->tpi_capability = PI_YES;
750 		PI_PROVUNLOCK(sti->sti_provinfo);
751 	}
752 
753 	if (tca->CAP_bits1 & TC1_ACCEPTOR_ID) {
754 		sti->sti_acceptor_id = tca->ACCEPTOR_id;
755 		so->so_mode |= SM_ACCEPTOR_ID;
756 	}
757 
758 	if (tca->CAP_bits1 & TC1_INFO)
759 		copy_tinfo(so, &tca->INFO_ack);
760 }
761 
762 /*
763  * Retrieve socket error, clear error if not peek.
764  */
765 int
766 sogeterr(struct sonode *so, boolean_t clear_err)
767 {
768 	int error;
769 
770 	ASSERT(MUTEX_HELD(&so->so_lock));
771 
772 	error = so->so_error;
773 	if (clear_err)
774 		so->so_error = 0;
775 
776 	return (error);
777 }
778 
779 /*
780  * This routine is registered with the stream head to retrieve read
781  * side errors.
782  * It does not clear the socket error for a peeking read side operation.
783  * It the error is to be cleared it sets *clearerr.
784  */
785 int
786 sogetrderr(vnode_t *vp, int ispeek, int *clearerr)
787 {
788 	struct sonode *so = VTOSO(vp);
789 	int error;
790 
791 	mutex_enter(&so->so_lock);
792 	if (ispeek) {
793 		error = so->so_error;
794 		*clearerr = 0;
795 	} else {
796 		error = so->so_error;
797 		so->so_error = 0;
798 		*clearerr = 1;
799 	}
800 	mutex_exit(&so->so_lock);
801 	return (error);
802 }
803 
804 /*
805  * This routine is registered with the stream head to retrieve write
806  * side errors.
807  * It does not clear the socket error for a peeking read side operation.
808  * It the error is to be cleared it sets *clearerr.
809  */
810 int
811 sogetwrerr(vnode_t *vp, int ispeek, int *clearerr)
812 {
813 	struct sonode *so = VTOSO(vp);
814 	int error;
815 
816 	mutex_enter(&so->so_lock);
817 	if (so->so_state & SS_CANTSENDMORE) {
818 		error = EPIPE;
819 		*clearerr = 0;
820 	} else {
821 		error = so->so_error;
822 		if (ispeek) {
823 			*clearerr = 0;
824 		} else {
825 			so->so_error = 0;
826 			*clearerr = 1;
827 		}
828 	}
829 	mutex_exit(&so->so_lock);
830 	return (error);
831 }
832 
833 /*
834  * Set a nonpersistent read and write error on the socket.
835  * Used when there is a T_uderror_ind for a connected socket.
836  * The caller also needs to call strsetrerror and strsetwerror
837  * after dropping the lock.
838  */
839 void
840 soseterror(struct sonode *so, int error)
841 {
842 	ASSERT(error != 0);
843 
844 	ASSERT(MUTEX_HELD(&so->so_lock));
845 	so->so_error = (ushort_t)error;
846 }
847 
848 void
849 soisconnecting(struct sonode *so)
850 {
851 	ASSERT(MUTEX_HELD(&so->so_lock));
852 	so->so_state &= ~(SS_ISCONNECTED|SS_ISDISCONNECTING);
853 	so->so_state |= SS_ISCONNECTING;
854 	cv_broadcast(&so->so_state_cv);
855 }
856 
857 void
858 soisconnected(struct sonode *so)
859 {
860 	ASSERT(MUTEX_HELD(&so->so_lock));
861 	so->so_state &= ~(SS_ISCONNECTING|SS_ISDISCONNECTING);
862 	so->so_state |= SS_ISCONNECTED;
863 	cv_broadcast(&so->so_state_cv);
864 }
865 
866 /*
867  * The caller also needs to call strsetrerror, strsetwerror and strseteof.
868  */
869 void
870 soisdisconnected(struct sonode *so, int error)
871 {
872 	ASSERT(MUTEX_HELD(&so->so_lock));
873 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
874 	so->so_state |= (SS_CANTRCVMORE|SS_CANTSENDMORE);
875 	so->so_error = (ushort_t)error;
876 	if (so->so_peercred != NULL) {
877 		crfree(so->so_peercred);
878 		so->so_peercred = NULL;
879 	}
880 	cv_broadcast(&so->so_state_cv);
881 }
882 
883 /*
884  * For connected AF_UNIX SOCK_DGRAM sockets when the peer closes.
885  * Does not affect write side.
886  * The caller also has to call strsetrerror.
887  */
888 static void
889 sobreakconn(struct sonode *so, int error)
890 {
891 	ASSERT(MUTEX_HELD(&so->so_lock));
892 	so->so_state &= ~(SS_ISCONNECTING|SS_ISCONNECTED|SS_ISDISCONNECTING);
893 	so->so_error = (ushort_t)error;
894 	cv_broadcast(&so->so_state_cv);
895 }
896 
897 /*
898  * Can no longer send.
899  * Caller must also call strsetwerror.
900  *
901  * We mark the peer address as no longer valid for getpeername, but
902  * leave it around for so_unix_close to notify the peer (that
903  * transport has no addressing held at that layer).
904  */
905 void
906 socantsendmore(struct sonode *so)
907 {
908 	ASSERT(MUTEX_HELD(&so->so_lock));
909 	so->so_state |= SS_CANTSENDMORE;
910 	cv_broadcast(&so->so_state_cv);
911 }
912 
913 /*
914  * The caller must call strseteof(,1) as well as this routine
915  * to change the socket state.
916  */
917 void
918 socantrcvmore(struct sonode *so)
919 {
920 	ASSERT(MUTEX_HELD(&so->so_lock));
921 	so->so_state |= SS_CANTRCVMORE;
922 	cv_broadcast(&so->so_state_cv);
923 }
924 
925 /*
926  * The caller has sent down a "request_prim" primitive and wants to wait for
927  * an ack ("ack_prim") or an T_ERROR_ACK for it.
928  * The specified "ack_prim" can be a T_OK_ACK.
929  *
930  * Assumes that all the TPI acks are M_PCPROTO messages.
931  *
932  * Note that the socket is single-threaded (using so_lock_single)
933  * for all operations that generate TPI ack messages. Since
934  * only TPI ack messages are M_PCPROTO we should never receive
935  * anything except either the ack we are expecting or a T_ERROR_ACK
936  * for the same primitive.
937  */
938 int
939 sowaitprim(struct sonode *so, t_scalar_t request_prim, t_scalar_t ack_prim,
940 	    t_uscalar_t min_size, mblk_t **mpp, clock_t wait)
941 {
942 	mblk_t *mp;
943 	union T_primitives *tpr;
944 	int error;
945 
946 	dprintso(so, 1, ("sowaitprim(%p, %d, %d, %d, %p, %lu)\n",
947 	    (void *)so, request_prim, ack_prim, min_size, (void *)mpp, wait));
948 
949 	ASSERT(MUTEX_HELD(&so->so_lock));
950 
951 	error = sowaitack(so, &mp, wait);
952 	if (error)
953 		return (error);
954 
955 	dprintso(so, 1, ("got msg %p\n", (void *)mp));
956 	if (DB_TYPE(mp) != M_PCPROTO ||
957 	    MBLKL(mp) < sizeof (tpr->type)) {
958 		freemsg(mp);
959 		eprintsoline(so, EPROTO);
960 		return (EPROTO);
961 	}
962 	tpr = (union T_primitives *)mp->b_rptr;
963 	/*
964 	 * Did we get the primitive that we were asking for?
965 	 * For T_OK_ACK we also check that it matches the request primitive.
966 	 */
967 	if (tpr->type == ack_prim &&
968 	    (ack_prim != T_OK_ACK ||
969 	    tpr->ok_ack.CORRECT_prim == request_prim)) {
970 		if (MBLKL(mp) >= (ssize_t)min_size) {
971 			/* Found what we are looking for */
972 			*mpp = mp;
973 			return (0);
974 		}
975 		/* Too short */
976 		freemsg(mp);
977 		eprintsoline(so, EPROTO);
978 		return (EPROTO);
979 	}
980 
981 	if (tpr->type == T_ERROR_ACK &&
982 	    tpr->error_ack.ERROR_prim == request_prim) {
983 		/* Error to the primitive we were looking for */
984 		if (tpr->error_ack.TLI_error == TSYSERR) {
985 			error = tpr->error_ack.UNIX_error;
986 		} else {
987 			error = proto_tlitosyserr(tpr->error_ack.TLI_error);
988 		}
989 		dprintso(so, 0, ("error_ack for %d: %d/%d ->%d\n",
990 		    tpr->error_ack.ERROR_prim, tpr->error_ack.TLI_error,
991 		    tpr->error_ack.UNIX_error, error));
992 		freemsg(mp);
993 		return (error);
994 	}
995 	/*
996 	 * Wrong primitive or T_ERROR_ACK for the wrong primitive
997 	 */
998 #ifdef DEBUG
999 	if (tpr->type == T_ERROR_ACK) {
1000 		dprintso(so, 0, ("error_ack for %d: %d/%d\n",
1001 		    tpr->error_ack.ERROR_prim, tpr->error_ack.TLI_error,
1002 		    tpr->error_ack.UNIX_error));
1003 	} else if (tpr->type == T_OK_ACK) {
1004 		dprintso(so, 0, ("ok_ack for %d, expected %d for %d\n",
1005 		    tpr->ok_ack.CORRECT_prim, ack_prim, request_prim));
1006 	} else {
1007 		dprintso(so, 0,
1008 		    ("unexpected primitive %d, expected %d for %d\n",
1009 		    tpr->type, ack_prim, request_prim));
1010 	}
1011 #endif /* DEBUG */
1012 
1013 	freemsg(mp);
1014 	eprintsoline(so, EPROTO);
1015 	return (EPROTO);
1016 }
1017 
1018 /*
1019  * Wait for a T_OK_ACK for the specified primitive.
1020  */
1021 int
1022 sowaitokack(struct sonode *so, t_scalar_t request_prim)
1023 {
1024 	mblk_t *mp;
1025 	int error;
1026 
1027 	error = sowaitprim(so, request_prim, T_OK_ACK,
1028 	    (t_uscalar_t)sizeof (struct T_ok_ack), &mp, 0);
1029 	if (error)
1030 		return (error);
1031 	freemsg(mp);
1032 	return (0);
1033 }
1034 
1035 /*
1036  * Queue a received TPI ack message on sti_ack_mp.
1037  */
1038 void
1039 soqueueack(struct sonode *so, mblk_t *mp)
1040 {
1041 	sotpi_info_t *sti = SOTOTPI(so);
1042 
1043 	if (DB_TYPE(mp) != M_PCPROTO) {
1044 		zcmn_err(getzoneid(), CE_WARN,
1045 		    "sockfs: received unexpected M_PROTO TPI ack. Prim %d\n",
1046 		    *(t_scalar_t *)mp->b_rptr);
1047 		freemsg(mp);
1048 		return;
1049 	}
1050 
1051 	mutex_enter(&so->so_lock);
1052 	if (sti->sti_ack_mp != NULL) {
1053 		dprintso(so, 1, ("sti_ack_mp already set\n"));
1054 		freemsg(sti->sti_ack_mp);
1055 		sti->sti_ack_mp = NULL;
1056 	}
1057 	sti->sti_ack_mp = mp;
1058 	cv_broadcast(&sti->sti_ack_cv);
1059 	mutex_exit(&so->so_lock);
1060 }
1061 
1062 /*
1063  * Wait for a TPI ack ignoring signals and errors.
1064  */
1065 int
1066 sowaitack(struct sonode *so, mblk_t **mpp, clock_t wait)
1067 {
1068 	sotpi_info_t *sti = SOTOTPI(so);
1069 
1070 	ASSERT(MUTEX_HELD(&so->so_lock));
1071 
1072 	while (sti->sti_ack_mp == NULL) {
1073 #ifdef SOCK_TEST
1074 		if (wait == 0 && sock_test_timelimit != 0)
1075 			wait = sock_test_timelimit;
1076 #endif
1077 		if (wait != 0) {
1078 			/*
1079 			 * Only wait for the time limit.
1080 			 */
1081 			if (cv_reltimedwait(&sti->sti_ack_cv, &so->so_lock,
1082 			    wait, TR_CLOCK_TICK) == -1) {
1083 				eprintsoline(so, ETIME);
1084 				return (ETIME);
1085 			}
1086 		}
1087 		else
1088 			cv_wait(&sti->sti_ack_cv, &so->so_lock);
1089 	}
1090 	*mpp = sti->sti_ack_mp;
1091 #ifdef DEBUG
1092 	{
1093 		union T_primitives *tpr;
1094 		mblk_t *mp = *mpp;
1095 
1096 		tpr = (union T_primitives *)mp->b_rptr;
1097 		ASSERT(DB_TYPE(mp) == M_PCPROTO);
1098 		ASSERT(tpr->type == T_OK_ACK ||
1099 		    tpr->type == T_ERROR_ACK ||
1100 		    tpr->type == T_BIND_ACK ||
1101 		    tpr->type == T_CAPABILITY_ACK ||
1102 		    tpr->type == T_INFO_ACK ||
1103 		    tpr->type == T_OPTMGMT_ACK);
1104 	}
1105 #endif /* DEBUG */
1106 	sti->sti_ack_mp = NULL;
1107 	return (0);
1108 }
1109 
1110 /*
1111  * Queue a received T_CONN_IND message on sti_conn_ind_head/tail.
1112  */
1113 void
1114 soqueueconnind(struct sonode *so, mblk_t *mp)
1115 {
1116 	sotpi_info_t *sti = SOTOTPI(so);
1117 
1118 	if (DB_TYPE(mp) != M_PROTO) {
1119 		zcmn_err(getzoneid(), CE_WARN,
1120 		    "sockfs: received unexpected M_PCPROTO T_CONN_IND\n");
1121 		freemsg(mp);
1122 		return;
1123 	}
1124 
1125 	mutex_enter(&so->so_lock);
1126 	ASSERT(mp->b_next == NULL);
1127 	if (sti->sti_conn_ind_head == NULL) {
1128 		sti->sti_conn_ind_head = mp;
1129 	} else {
1130 		ASSERT(sti->sti_conn_ind_tail->b_next == NULL);
1131 		sti->sti_conn_ind_tail->b_next = mp;
1132 	}
1133 	sti->sti_conn_ind_tail = mp;
1134 	/* Wakeup a single consumer of the T_CONN_IND */
1135 	cv_signal(&so->so_acceptq_cv);
1136 	mutex_exit(&so->so_lock);
1137 }
1138 
1139 /*
1140  * Wait for a T_CONN_IND.
1141  * Don't wait if nonblocking.
1142  * Accept signals and socket errors.
1143  */
1144 int
1145 sowaitconnind(struct sonode *so, int fmode, mblk_t **mpp)
1146 {
1147 	mblk_t *mp;
1148 	sotpi_info_t *sti = SOTOTPI(so);
1149 	int error = 0;
1150 
1151 	ASSERT(MUTEX_NOT_HELD(&so->so_lock));
1152 	mutex_enter(&so->so_lock);
1153 check_error:
1154 	if (so->so_error) {
1155 		error = sogeterr(so, B_TRUE);
1156 		if (error) {
1157 			mutex_exit(&so->so_lock);
1158 			return (error);
1159 		}
1160 	}
1161 
1162 	if (sti->sti_conn_ind_head == NULL) {
1163 		if (fmode & (FNDELAY|FNONBLOCK)) {
1164 			error = EWOULDBLOCK;
1165 			goto done;
1166 		}
1167 
1168 		if (so->so_state & SS_CLOSING) {
1169 			error = EINTR;
1170 			goto done;
1171 		}
1172 
1173 		if (!cv_wait_sig_swap(&so->so_acceptq_cv, &so->so_lock)) {
1174 			error = EINTR;
1175 			goto done;
1176 		}
1177 		goto check_error;
1178 	}
1179 	mp = sti->sti_conn_ind_head;
1180 	sti->sti_conn_ind_head = mp->b_next;
1181 	mp->b_next = NULL;
1182 	if (sti->sti_conn_ind_head == NULL) {
1183 		ASSERT(sti->sti_conn_ind_tail == mp);
1184 		sti->sti_conn_ind_tail = NULL;
1185 	}
1186 	*mpp = mp;
1187 done:
1188 	mutex_exit(&so->so_lock);
1189 	return (error);
1190 }
1191 
1192 /*
1193  * Flush a T_CONN_IND matching the sequence number from the list.
1194  * Return zero if found; non-zero otherwise.
1195  * This is called very infrequently thus it is ok to do a linear search.
1196  */
1197 int
1198 soflushconnind(struct sonode *so, t_scalar_t seqno)
1199 {
1200 	mblk_t *prevmp, *mp;
1201 	struct T_conn_ind *tci;
1202 	sotpi_info_t *sti = SOTOTPI(so);
1203 
1204 	mutex_enter(&so->so_lock);
1205 	for (prevmp = NULL, mp = sti->sti_conn_ind_head; mp != NULL;
1206 	    prevmp = mp, mp = mp->b_next) {
1207 		tci = (struct T_conn_ind *)mp->b_rptr;
1208 		if (tci->SEQ_number == seqno) {
1209 			dprintso(so, 1,
1210 			    ("t_discon_ind: found T_CONN_IND %d\n", seqno));
1211 			/* Deleting last? */
1212 			if (sti->sti_conn_ind_tail == mp) {
1213 				sti->sti_conn_ind_tail = prevmp;
1214 			}
1215 			if (prevmp == NULL) {
1216 				/* Deleting first */
1217 				sti->sti_conn_ind_head = mp->b_next;
1218 			} else {
1219 				prevmp->b_next = mp->b_next;
1220 			}
1221 			mp->b_next = NULL;
1222 
1223 			ASSERT((sti->sti_conn_ind_head == NULL &&
1224 			    sti->sti_conn_ind_tail == NULL) ||
1225 			    (sti->sti_conn_ind_head != NULL &&
1226 			    sti->sti_conn_ind_tail != NULL));
1227 
1228 			so->so_error = ECONNABORTED;
1229 			mutex_exit(&so->so_lock);
1230 
1231 			/*
1232 			 * T_KSSL_PROXY_CONN_IND may carry a handle for
1233 			 * an SSL context, and needs to be released.
1234 			 */
1235 			if ((tci->PRIM_type == T_SSL_PROXY_CONN_IND) &&
1236 			    (mp->b_cont != NULL)) {
1237 				kssl_ctx_t kssl_ctx;
1238 
1239 				ASSERT(MBLKL(mp->b_cont) ==
1240 				    sizeof (kssl_ctx_t));
1241 				kssl_ctx = *((kssl_ctx_t *)mp->b_cont->b_rptr);
1242 				kssl_release_ctx(kssl_ctx);
1243 			}
1244 			freemsg(mp);
1245 			return (0);
1246 		}
1247 	}
1248 	mutex_exit(&so->so_lock);
1249 	dprintso(so, 1,	("t_discon_ind: NOT found T_CONN_IND %d\n", seqno));
1250 	return (-1);
1251 }
1252 
1253 /*
1254  * Wait until the socket is connected or there is an error.
1255  * fmode should contain any nonblocking flags. nosig should be
1256  * set if the caller does not want the wait to be interrupted by a signal.
1257  */
1258 int
1259 sowaitconnected(struct sonode *so, int fmode, int nosig)
1260 {
1261 	int error;
1262 
1263 	ASSERT(MUTEX_HELD(&so->so_lock));
1264 
1265 	while ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) ==
1266 	    SS_ISCONNECTING && so->so_error == 0) {
1267 
1268 		dprintso(so, 1, ("waiting for SS_ISCONNECTED on %p\n",
1269 		    (void *)so));
1270 		if (fmode & (FNDELAY|FNONBLOCK))
1271 			return (EINPROGRESS);
1272 
1273 		if (so->so_state & SS_CLOSING)
1274 			return (EINTR);
1275 
1276 		if (nosig)
1277 			cv_wait(&so->so_state_cv, &so->so_lock);
1278 		else if (!cv_wait_sig_swap(&so->so_state_cv, &so->so_lock)) {
1279 			/*
1280 			 * Return EINTR and let the application use
1281 			 * nonblocking techniques for detecting when
1282 			 * the connection has been established.
1283 			 */
1284 			return (EINTR);
1285 		}
1286 		dprintso(so, 1, ("awoken on %p\n", (void *)so));
1287 	}
1288 
1289 	if (so->so_error != 0) {
1290 		error = sogeterr(so, B_TRUE);
1291 		ASSERT(error != 0);
1292 		dprintso(so, 1, ("sowaitconnected: error %d\n", error));
1293 		return (error);
1294 	}
1295 	if (!(so->so_state & SS_ISCONNECTED)) {
1296 		/*
1297 		 * Could have received a T_ORDREL_IND or a T_DISCON_IND with
1298 		 * zero errno. Or another thread could have consumed so_error
1299 		 * e.g. by calling read.
1300 		 */
1301 		error = ECONNREFUSED;
1302 		dprintso(so, 1, ("sowaitconnected: error %d\n", error));
1303 		return (error);
1304 	}
1305 	return (0);
1306 }
1307 
1308 
1309 /*
1310  * Handle the signal generation aspect of urgent data.
1311  */
1312 static void
1313 so_oob_sig(struct sonode *so, int extrasig,
1314     strsigset_t *signals, strpollset_t *pollwakeups)
1315 {
1316 	sotpi_info_t *sti = SOTOTPI(so);
1317 
1318 	ASSERT(MUTEX_HELD(&so->so_lock));
1319 
1320 	ASSERT(so_verify_oobstate(so));
1321 	ASSERT(sti->sti_oobsigcnt >= sti->sti_oobcnt);
1322 	if (sti->sti_oobsigcnt > sti->sti_oobcnt) {
1323 		/*
1324 		 * Signal has already been generated once for this
1325 		 * urgent "event". However, since TCP can receive updated
1326 		 * urgent pointers we still generate a signal.
1327 		 */
1328 		ASSERT(so->so_state & SS_OOBPEND);
1329 		if (extrasig) {
1330 			*signals |= S_RDBAND;
1331 			*pollwakeups |= POLLRDBAND;
1332 		}
1333 		return;
1334 	}
1335 
1336 	sti->sti_oobsigcnt++;
1337 	ASSERT(sti->sti_oobsigcnt > 0);	/* Wraparound */
1338 	ASSERT(sti->sti_oobsigcnt > sti->sti_oobcnt);
1339 
1340 	/*
1341 	 * Record (for select/poll) that urgent data is pending.
1342 	 */
1343 	so->so_state |= SS_OOBPEND;
1344 	/*
1345 	 * New urgent data on the way so forget about any old
1346 	 * urgent data.
1347 	 */
1348 	so->so_state &= ~(SS_HAVEOOBDATA|SS_HADOOBDATA);
1349 	if (so->so_oobmsg != NULL) {
1350 		dprintso(so, 1, ("sock: discarding old oob\n"));
1351 		freemsg(so->so_oobmsg);
1352 		so->so_oobmsg = NULL;
1353 	}
1354 	*signals |= S_RDBAND;
1355 	*pollwakeups |= POLLRDBAND;
1356 	ASSERT(so_verify_oobstate(so));
1357 }
1358 
1359 /*
1360  * Handle the processing of the T_EXDATA_IND with urgent data.
1361  * Returns the T_EXDATA_IND if it should be queued on the read queue.
1362  */
1363 /* ARGSUSED2 */
1364 static mblk_t *
1365 so_oob_exdata(struct sonode *so, mblk_t *mp,
1366 	strsigset_t *signals, strpollset_t *pollwakeups)
1367 {
1368 	sotpi_info_t *sti = SOTOTPI(so);
1369 
1370 	ASSERT(MUTEX_HELD(&so->so_lock));
1371 
1372 	ASSERT(so_verify_oobstate(so));
1373 
1374 	ASSERT(sti->sti_oobsigcnt > sti->sti_oobcnt);
1375 
1376 	sti->sti_oobcnt++;
1377 	ASSERT(sti->sti_oobcnt > 0);	/* wraparound? */
1378 	ASSERT(sti->sti_oobsigcnt >= sti->sti_oobcnt);
1379 
1380 	/*
1381 	 * Set MSGMARK for SIOCATMARK.
1382 	 */
1383 	mp->b_flag |= MSGMARK;
1384 
1385 	ASSERT(so_verify_oobstate(so));
1386 	return (mp);
1387 }
1388 
1389 /*
1390  * Handle the processing of the actual urgent data.
1391  * Returns the data mblk if it should be queued on the read queue.
1392  */
1393 static mblk_t *
1394 so_oob_data(struct sonode *so, mblk_t *mp,
1395 	strsigset_t *signals, strpollset_t *pollwakeups)
1396 {
1397 	sotpi_info_t *sti = SOTOTPI(so);
1398 
1399 	ASSERT(MUTEX_HELD(&so->so_lock));
1400 
1401 	ASSERT(so_verify_oobstate(so));
1402 
1403 	ASSERT(sti->sti_oobsigcnt >= sti->sti_oobcnt);
1404 	ASSERT(mp != NULL);
1405 	/*
1406 	 * For OOBINLINE we keep the data in the T_EXDATA_IND.
1407 	 * Otherwise we store it in so_oobmsg.
1408 	 */
1409 	ASSERT(so->so_oobmsg == NULL);
1410 	if (so->so_options & SO_OOBINLINE) {
1411 		*pollwakeups |= POLLIN | POLLRDNORM | POLLRDBAND;
1412 		*signals |= S_INPUT | S_RDNORM;
1413 	} else {
1414 		*pollwakeups |= POLLRDBAND;
1415 		so->so_state |= SS_HAVEOOBDATA;
1416 		so->so_oobmsg = mp;
1417 		mp = NULL;
1418 	}
1419 	ASSERT(so_verify_oobstate(so));
1420 	return (mp);
1421 }
1422 
1423 /*
1424  * Caller must hold the mutex.
1425  * For delayed processing, save the T_DISCON_IND received
1426  * from below on sti_discon_ind_mp.
1427  * When the message is processed the framework will call:
1428  *      (*func)(so, mp);
1429  */
1430 static void
1431 so_save_discon_ind(struct sonode *so,
1432 	mblk_t *mp,
1433 	void (*func)(struct sonode *so, mblk_t *))
1434 {
1435 	sotpi_info_t *sti = SOTOTPI(so);
1436 
1437 	ASSERT(MUTEX_HELD(&so->so_lock));
1438 
1439 	/*
1440 	 * Discard new T_DISCON_IND if we have already received another.
1441 	 * Currently the earlier message can either be on sti_discon_ind_mp
1442 	 * or being processed.
1443 	 */
1444 	if (sti->sti_discon_ind_mp != NULL || (so->so_flag & SOASYNC_UNBIND)) {
1445 		zcmn_err(getzoneid(), CE_WARN,
1446 		    "sockfs: received unexpected additional T_DISCON_IND\n");
1447 		freemsg(mp);
1448 		return;
1449 	}
1450 	mp->b_prev = (mblk_t *)func;
1451 	mp->b_next = NULL;
1452 	sti->sti_discon_ind_mp = mp;
1453 }
1454 
1455 /*
1456  * Caller must hold the mutex and make sure that either SOLOCKED
1457  * or SOASYNC_UNBIND is set. Called from so_unlock_single().
1458  * Perform delayed processing of T_DISCON_IND message on sti_discon_ind_mp.
1459  * Need to ensure that strsock_proto() will not end up sleeping for
1460  * SOASYNC_UNBIND, while executing this function.
1461  */
1462 void
1463 so_drain_discon_ind(struct sonode *so)
1464 {
1465 	mblk_t	*bp;
1466 	void (*func)(struct sonode *so, mblk_t *);
1467 	sotpi_info_t *sti = SOTOTPI(so);
1468 
1469 	ASSERT(MUTEX_HELD(&so->so_lock));
1470 	ASSERT(so->so_flag & (SOLOCKED|SOASYNC_UNBIND));
1471 
1472 	/* Process T_DISCON_IND on sti_discon_ind_mp */
1473 	if ((bp = sti->sti_discon_ind_mp) != NULL) {
1474 		sti->sti_discon_ind_mp = NULL;
1475 		func = (void (*)())bp->b_prev;
1476 		bp->b_prev = NULL;
1477 
1478 		/*
1479 		 * This (*func) is supposed to generate a message downstream
1480 		 * and we need to have a flag set until the corresponding
1481 		 * upstream message reaches stream head.
1482 		 * When processing T_DISCON_IND in strsock_discon_ind
1483 		 * we hold SOASYN_UNBIND when sending T_UNBIND_REQ down and
1484 		 * drop the flag after we get the ACK in strsock_proto.
1485 		 */
1486 		(void) (*func)(so, bp);
1487 	}
1488 }
1489 
1490 /*
1491  * Caller must hold the mutex.
1492  * Remove the T_DISCON_IND on sti_discon_ind_mp.
1493  */
1494 void
1495 so_flush_discon_ind(struct sonode *so)
1496 {
1497 	mblk_t	*bp;
1498 	sotpi_info_t *sti = SOTOTPI(so);
1499 
1500 	ASSERT(MUTEX_HELD(&so->so_lock));
1501 
1502 	/*
1503 	 * Remove T_DISCON_IND mblk at sti_discon_ind_mp.
1504 	 */
1505 	if ((bp = sti->sti_discon_ind_mp) != NULL) {
1506 		sti->sti_discon_ind_mp = NULL;
1507 		bp->b_prev = NULL;
1508 		freemsg(bp);
1509 	}
1510 }
1511 
1512 /*
1513  * Caller must hold the mutex.
1514  *
1515  * This function is used to process the T_DISCON_IND message. It does
1516  * immediate processing when called from strsock_proto and delayed
1517  * processing of discon_ind saved on sti_discon_ind_mp when called from
1518  * so_drain_discon_ind. When a T_DISCON_IND message is saved in
1519  * sti_discon_ind_mp for delayed processing, this function is registered
1520  * as the callback function to process the message.
1521  *
1522  * SOASYNC_UNBIND should be held in this function, during the non-blocking
1523  * unbind operation, and should be released only after we receive the ACK
1524  * in strsock_proto, for the T_UNBIND_REQ sent here. Since SOLOCKED is not set,
1525  * no TPI messages would be sent down at this time. This is to prevent M_FLUSH
1526  * sent from either this function or tcp_unbind(), flushing away any TPI
1527  * message that is being sent down and stays in a lower module's queue.
1528  *
1529  * This function drops so_lock and grabs it again.
1530  */
1531 static void
1532 strsock_discon_ind(struct sonode *so, mblk_t *discon_mp)
1533 {
1534 	struct vnode *vp;
1535 	struct stdata *stp;
1536 	union T_primitives *tpr;
1537 	struct T_unbind_req *ubr;
1538 	mblk_t *mp;
1539 	int error;
1540 	sotpi_info_t *sti = SOTOTPI(so);
1541 
1542 	ASSERT(MUTEX_HELD(&so->so_lock));
1543 	ASSERT(discon_mp);
1544 	ASSERT(discon_mp->b_rptr);
1545 
1546 	tpr = (union T_primitives *)discon_mp->b_rptr;
1547 	ASSERT(tpr->type == T_DISCON_IND);
1548 
1549 	vp = SOTOV(so);
1550 	stp = vp->v_stream;
1551 	ASSERT(stp);
1552 
1553 	/*
1554 	 * Not a listener
1555 	 */
1556 	ASSERT((so->so_state & SS_ACCEPTCONN) == 0);
1557 
1558 	/*
1559 	 * This assumes that the name space for DISCON_reason
1560 	 * is the errno name space.
1561 	 */
1562 	soisdisconnected(so, tpr->discon_ind.DISCON_reason);
1563 	sti->sti_laddr_valid = 0;
1564 	sti->sti_faddr_valid = 0;
1565 
1566 	/*
1567 	 * Unbind with the transport without blocking.
1568 	 * If we've already received a T_DISCON_IND do not unbind.
1569 	 *
1570 	 * If there is no preallocated unbind message, we have already
1571 	 * unbound with the transport
1572 	 *
1573 	 * If the socket is not bound, no need to unbind.
1574 	 */
1575 	mp = sti->sti_unbind_mp;
1576 	if (mp == NULL) {
1577 		ASSERT(!(so->so_state & SS_ISBOUND));
1578 		mutex_exit(&so->so_lock);
1579 	} else if (!(so->so_state & SS_ISBOUND))  {
1580 		mutex_exit(&so->so_lock);
1581 	} else {
1582 		sti->sti_unbind_mp = NULL;
1583 
1584 		/*
1585 		 * Is another T_DISCON_IND being processed.
1586 		 */
1587 		ASSERT((so->so_flag & SOASYNC_UNBIND) == 0);
1588 
1589 		/*
1590 		 * Make strsock_proto ignore T_OK_ACK and T_ERROR_ACK for
1591 		 * this unbind. Set SOASYNC_UNBIND. This should be cleared
1592 		 * only after we receive the ACK in strsock_proto.
1593 		 */
1594 		so->so_flag |= SOASYNC_UNBIND;
1595 		ASSERT(!(so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)));
1596 		so->so_state &= ~(SS_ISBOUND|SS_ACCEPTCONN);
1597 		sti->sti_laddr_valid = 0;
1598 		mutex_exit(&so->so_lock);
1599 
1600 		/*
1601 		 * Send down T_UNBIND_REQ ignoring flow control.
1602 		 * XXX Assumes that MSG_IGNFLOW implies that this thread
1603 		 * does not run service procedures.
1604 		 */
1605 		ASSERT(DB_TYPE(mp) == M_PROTO);
1606 		ubr = (struct T_unbind_req *)mp->b_rptr;
1607 		mp->b_wptr += sizeof (*ubr);
1608 		ubr->PRIM_type = T_UNBIND_REQ;
1609 
1610 		/*
1611 		 * Flush the read and write side (except stream head read queue)
1612 		 * and send down T_UNBIND_REQ.
1613 		 */
1614 		(void) putnextctl1(strvp2wq(SOTOV(so)), M_FLUSH, FLUSHRW);
1615 		error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0,
1616 		    MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR|MSG_IGNFLOW, 0);
1617 		/* LINTED - warning: statement has no consequent: if */
1618 		if (error) {
1619 			eprintsoline(so, error);
1620 		}
1621 	}
1622 
1623 	if (tpr->discon_ind.DISCON_reason != 0)
1624 		strsetrerror(SOTOV(so), 0, 0, sogetrderr);
1625 	strsetwerror(SOTOV(so), 0, 0, sogetwrerr);
1626 	strseteof(SOTOV(so), 1);
1627 	/*
1628 	 * strseteof takes care of read side wakeups,
1629 	 * pollwakeups, and signals.
1630 	 */
1631 	dprintso(so, 1, ("T_DISCON_IND: error %d\n", so->so_error));
1632 	freemsg(discon_mp);
1633 
1634 
1635 	pollwakeup(&stp->sd_pollist, POLLOUT);
1636 	mutex_enter(&stp->sd_lock);
1637 
1638 	/*
1639 	 * Wake sleeping write
1640 	 */
1641 	if (stp->sd_flag & WSLEEP) {
1642 		stp->sd_flag &= ~WSLEEP;
1643 		cv_broadcast(&stp->sd_wrq->q_wait);
1644 	}
1645 
1646 	/*
1647 	 * strsendsig can handle multiple signals with a
1648 	 * single call.  Send SIGPOLL for S_OUTPUT event.
1649 	 */
1650 	if (stp->sd_sigflags & S_OUTPUT)
1651 		strsendsig(stp->sd_siglist, S_OUTPUT, 0, 0);
1652 
1653 	mutex_exit(&stp->sd_lock);
1654 	mutex_enter(&so->so_lock);
1655 }
1656 
1657 /*
1658  * This routine is registered with the stream head to receive M_PROTO
1659  * and M_PCPROTO messages.
1660  *
1661  * Returns NULL if the message was consumed.
1662  * Returns an mblk to make that mblk be processed (and queued) by the stream
1663  * head.
1664  *
1665  * Sets the return parameters (*wakeups, *firstmsgsigs, *allmsgsigs, and
1666  * *pollwakeups) for the stream head to take action on. Note that since
1667  * sockets always deliver SIGIO for every new piece of data this routine
1668  * never sets *firstmsgsigs; any signals are returned in *allmsgsigs.
1669  *
1670  * This routine handles all data related TPI messages independent of
1671  * the type of the socket i.e. it doesn't care if T_UNITDATA_IND message
1672  * arrive on a SOCK_STREAM.
1673  */
1674 static mblk_t *
1675 strsock_proto(vnode_t *vp, mblk_t *mp,
1676 		strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1677 		strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1678 {
1679 	union T_primitives *tpr;
1680 	struct sonode *so;
1681 	sotpi_info_t *sti;
1682 	uint32_t auditing = AU_AUDITING();
1683 
1684 	so = VTOSO(vp);
1685 	sti = SOTOTPI(so);
1686 
1687 	dprintso(so, 1, ("strsock_proto(%p, %p)\n", (void *)vp, (void *)mp));
1688 
1689 	/* Set default return values */
1690 	*firstmsgsigs = *wakeups = *allmsgsigs = *pollwakeups = 0;
1691 
1692 	ASSERT(DB_TYPE(mp) == M_PROTO ||
1693 	    DB_TYPE(mp) == M_PCPROTO);
1694 
1695 	if (MBLKL(mp) < sizeof (tpr->type)) {
1696 		/* The message is too short to even contain the primitive */
1697 		zcmn_err(getzoneid(), CE_WARN,
1698 		    "sockfs: Too short TPI message received. Len = %ld\n",
1699 		    (ptrdiff_t)(MBLKL(mp)));
1700 		freemsg(mp);
1701 		return (NULL);
1702 	}
1703 	if (!__TPI_PRIM_ISALIGNED(mp->b_rptr)) {
1704 		/* The read pointer is not aligned correctly for TPI */
1705 		zcmn_err(getzoneid(), CE_WARN,
1706 		    "sockfs: Unaligned TPI message received. rptr = %p\n",
1707 		    (void *)mp->b_rptr);
1708 		freemsg(mp);
1709 		return (NULL);
1710 	}
1711 	tpr = (union T_primitives *)mp->b_rptr;
1712 	dprintso(so, 1, ("strsock_proto: primitive %d\n", tpr->type));
1713 
1714 	switch (tpr->type) {
1715 
1716 	case T_DATA_IND:
1717 		if (MBLKL(mp) < sizeof (struct T_data_ind)) {
1718 			zcmn_err(getzoneid(), CE_WARN,
1719 			    "sockfs: Too short T_DATA_IND. Len = %ld\n",
1720 			    (ptrdiff_t)(MBLKL(mp)));
1721 			freemsg(mp);
1722 			return (NULL);
1723 		}
1724 		/*
1725 		 * Ignore zero-length T_DATA_IND messages. These might be
1726 		 * generated by some transports.
1727 		 * This is needed to prevent read (which skips the M_PROTO
1728 		 * part) to unexpectedly return 0 (or return EWOULDBLOCK
1729 		 * on a non-blocking socket after select/poll has indicated
1730 		 * that data is available).
1731 		 */
1732 		if (msgdsize(mp->b_cont) == 0) {
1733 			dprintso(so, 0,
1734 			    ("strsock_proto: zero length T_DATA_IND\n"));
1735 			freemsg(mp);
1736 			return (NULL);
1737 		}
1738 		*allmsgsigs = S_INPUT | S_RDNORM;
1739 		*pollwakeups = POLLIN | POLLRDNORM;
1740 		*wakeups = RSLEEP;
1741 		return (mp);
1742 
1743 	case T_UNITDATA_IND: {
1744 		struct T_unitdata_ind	*tudi = &tpr->unitdata_ind;
1745 		void			*addr;
1746 		t_uscalar_t		addrlen;
1747 
1748 		if (MBLKL(mp) < sizeof (struct T_unitdata_ind)) {
1749 			zcmn_err(getzoneid(), CE_WARN,
1750 			    "sockfs: Too short T_UNITDATA_IND. Len = %ld\n",
1751 			    (ptrdiff_t)(MBLKL(mp)));
1752 			freemsg(mp);
1753 			return (NULL);
1754 		}
1755 
1756 		/* Is this is not a connected datagram socket? */
1757 		if ((so->so_mode & SM_CONNREQUIRED) ||
1758 		    !(so->so_state & SS_ISCONNECTED)) {
1759 			/*
1760 			 * Not a connected datagram socket. Look for
1761 			 * the SO_UNIX_CLOSE option. If such an option is found
1762 			 * discard the message (since it has no meaning
1763 			 * unless connected).
1764 			 */
1765 			if (so->so_family == AF_UNIX && msgdsize(mp) == 0 &&
1766 			    tudi->OPT_length != 0) {
1767 				void *opt;
1768 				t_uscalar_t optlen = tudi->OPT_length;
1769 
1770 				opt = sogetoff(mp, tudi->OPT_offset,
1771 				    optlen, __TPI_ALIGN_SIZE);
1772 				if (opt == NULL) {
1773 					/* The len/off falls outside mp */
1774 					freemsg(mp);
1775 					mutex_enter(&so->so_lock);
1776 					soseterror(so, EPROTO);
1777 					mutex_exit(&so->so_lock);
1778 					zcmn_err(getzoneid(), CE_WARN,
1779 					    "sockfs: T_unidata_ind with "
1780 					    "invalid optlen/offset %u/%d\n",
1781 					    optlen, tudi->OPT_offset);
1782 					return (NULL);
1783 				}
1784 				if (so_getopt_unix_close(opt, optlen)) {
1785 					freemsg(mp);
1786 					return (NULL);
1787 				}
1788 			}
1789 			*allmsgsigs = S_INPUT | S_RDNORM;
1790 			*pollwakeups = POLLIN | POLLRDNORM;
1791 			*wakeups = RSLEEP;
1792 			if (auditing)
1793 				audit_sock(T_UNITDATA_IND, strvp2wq(vp),
1794 				    mp, 0);
1795 			return (mp);
1796 		}
1797 
1798 		/*
1799 		 * A connect datagram socket. For AF_INET{,6} we verify that
1800 		 * the source address matches the "connected to" address.
1801 		 * The semantics of AF_UNIX sockets is to not verify
1802 		 * the source address.
1803 		 * Note that this source address verification is transport
1804 		 * specific. Thus the real fix would be to extent TPI
1805 		 * to allow T_CONN_REQ messages to be send to connectionless
1806 		 * transport providers and always let the transport provider
1807 		 * do whatever filtering is needed.
1808 		 *
1809 		 * The verification/filtering semantics for transports
1810 		 * other than AF_INET and AF_UNIX are unknown. The choice
1811 		 * would be to either filter using bcmp or let all messages
1812 		 * get through. This code does not filter other address
1813 		 * families since this at least allows the application to
1814 		 * work around any missing filtering.
1815 		 *
1816 		 * XXX Should we move filtering to UDP/ICMP???
1817 		 * That would require passing e.g. a T_DISCON_REQ to UDP
1818 		 * when the socket becomes unconnected.
1819 		 */
1820 		addrlen = tudi->SRC_length;
1821 		/*
1822 		 * The alignment restriction is really to strict but
1823 		 * we want enough alignment to inspect the fields of
1824 		 * a sockaddr_in.
1825 		 */
1826 		addr = sogetoff(mp, tudi->SRC_offset, addrlen,
1827 		    __TPI_ALIGN_SIZE);
1828 		if (addr == NULL) {
1829 			freemsg(mp);
1830 			mutex_enter(&so->so_lock);
1831 			soseterror(so, EPROTO);
1832 			mutex_exit(&so->so_lock);
1833 			zcmn_err(getzoneid(), CE_WARN,
1834 			    "sockfs: T_unidata_ind with invalid "
1835 			    "addrlen/offset %u/%d\n",
1836 			    addrlen, tudi->SRC_offset);
1837 			return (NULL);
1838 		}
1839 
1840 		if (so->so_family == AF_INET) {
1841 			/*
1842 			 * For AF_INET we allow wildcarding both sin_addr
1843 			 * and sin_port.
1844 			 */
1845 			struct sockaddr_in *faddr, *sin;
1846 
1847 			/* Prevent sti_faddr_sa from changing while accessed */
1848 			mutex_enter(&so->so_lock);
1849 			ASSERT(sti->sti_faddr_len ==
1850 			    (socklen_t)sizeof (struct sockaddr_in));
1851 			faddr = (struct sockaddr_in *)sti->sti_faddr_sa;
1852 			sin = (struct sockaddr_in *)addr;
1853 			if (addrlen !=
1854 			    (t_uscalar_t)sizeof (struct sockaddr_in) ||
1855 			    (sin->sin_addr.s_addr != faddr->sin_addr.s_addr &&
1856 			    faddr->sin_addr.s_addr != INADDR_ANY) ||
1857 			    (so->so_type != SOCK_RAW &&
1858 			    sin->sin_port != faddr->sin_port &&
1859 			    faddr->sin_port != 0)) {
1860 #ifdef DEBUG
1861 				dprintso(so, 0,
1862 				    ("sockfs: T_UNITDATA_IND mismatch: %s",
1863 				    pr_addr(so->so_family,
1864 				    (struct sockaddr *)addr, addrlen)));
1865 				dprintso(so, 0, (" - %s\n",
1866 				    pr_addr(so->so_family, sti->sti_faddr_sa,
1867 				    (t_uscalar_t)sti->sti_faddr_len)));
1868 #endif /* DEBUG */
1869 				mutex_exit(&so->so_lock);
1870 				freemsg(mp);
1871 				return (NULL);
1872 			}
1873 			mutex_exit(&so->so_lock);
1874 		} else if (so->so_family == AF_INET6) {
1875 			/*
1876 			 * For AF_INET6 we allow wildcarding both sin6_addr
1877 			 * and sin6_port.
1878 			 */
1879 			struct sockaddr_in6 *faddr6, *sin6;
1880 			static struct in6_addr zeroes; /* inits to all zeros */
1881 
1882 			/* Prevent sti_faddr_sa from changing while accessed */
1883 			mutex_enter(&so->so_lock);
1884 			ASSERT(sti->sti_faddr_len ==
1885 			    (socklen_t)sizeof (struct sockaddr_in6));
1886 			faddr6 = (struct sockaddr_in6 *)sti->sti_faddr_sa;
1887 			sin6 = (struct sockaddr_in6 *)addr;
1888 			/* XXX could we get a mapped address ::ffff:0.0.0.0 ? */
1889 			if (addrlen !=
1890 			    (t_uscalar_t)sizeof (struct sockaddr_in6) ||
1891 			    (!IN6_ARE_ADDR_EQUAL(&sin6->sin6_addr,
1892 			    &faddr6->sin6_addr) &&
1893 			    !IN6_ARE_ADDR_EQUAL(&faddr6->sin6_addr, &zeroes)) ||
1894 			    (so->so_type != SOCK_RAW &&
1895 			    sin6->sin6_port != faddr6->sin6_port &&
1896 			    faddr6->sin6_port != 0)) {
1897 #ifdef DEBUG
1898 				dprintso(so, 0,
1899 				    ("sockfs: T_UNITDATA_IND mismatch: %s",
1900 				    pr_addr(so->so_family,
1901 				    (struct sockaddr *)addr, addrlen)));
1902 				dprintso(so, 0, (" - %s\n",
1903 				    pr_addr(so->so_family, sti->sti_faddr_sa,
1904 				    (t_uscalar_t)sti->sti_faddr_len)));
1905 #endif /* DEBUG */
1906 				mutex_exit(&so->so_lock);
1907 				freemsg(mp);
1908 				return (NULL);
1909 			}
1910 			mutex_exit(&so->so_lock);
1911 		} else if (so->so_family == AF_UNIX &&
1912 		    msgdsize(mp->b_cont) == 0 &&
1913 		    tudi->OPT_length != 0) {
1914 			/*
1915 			 * Attempt to extract AF_UNIX
1916 			 * SO_UNIX_CLOSE indication from options.
1917 			 */
1918 			void *opt;
1919 			t_uscalar_t optlen = tudi->OPT_length;
1920 
1921 			opt = sogetoff(mp, tudi->OPT_offset,
1922 			    optlen, __TPI_ALIGN_SIZE);
1923 			if (opt == NULL) {
1924 				/* The len/off falls outside mp */
1925 				freemsg(mp);
1926 				mutex_enter(&so->so_lock);
1927 				soseterror(so, EPROTO);
1928 				mutex_exit(&so->so_lock);
1929 				zcmn_err(getzoneid(), CE_WARN,
1930 				    "sockfs: T_unidata_ind with invalid "
1931 				    "optlen/offset %u/%d\n",
1932 				    optlen, tudi->OPT_offset);
1933 				return (NULL);
1934 			}
1935 			/*
1936 			 * If we received a unix close indication mark the
1937 			 * socket and discard this message.
1938 			 */
1939 			if (so_getopt_unix_close(opt, optlen)) {
1940 				mutex_enter(&so->so_lock);
1941 				sobreakconn(so, ECONNRESET);
1942 				mutex_exit(&so->so_lock);
1943 				strsetrerror(SOTOV(so), 0, 0, sogetrderr);
1944 				freemsg(mp);
1945 				*pollwakeups = POLLIN | POLLRDNORM;
1946 				*allmsgsigs = S_INPUT | S_RDNORM;
1947 				*wakeups = RSLEEP;
1948 				return (NULL);
1949 			}
1950 		}
1951 		*allmsgsigs = S_INPUT | S_RDNORM;
1952 		*pollwakeups = POLLIN | POLLRDNORM;
1953 		*wakeups = RSLEEP;
1954 		return (mp);
1955 	}
1956 
1957 	case T_OPTDATA_IND: {
1958 		struct T_optdata_ind	*tdi = &tpr->optdata_ind;
1959 
1960 		if (MBLKL(mp) < sizeof (struct T_optdata_ind)) {
1961 			zcmn_err(getzoneid(), CE_WARN,
1962 			    "sockfs: Too short T_OPTDATA_IND. Len = %ld\n",
1963 			    (ptrdiff_t)(MBLKL(mp)));
1964 			freemsg(mp);
1965 			return (NULL);
1966 		}
1967 		/*
1968 		 * Allow zero-length messages carrying options.
1969 		 * This is used when carrying the SO_UNIX_CLOSE option.
1970 		 */
1971 		if (so->so_family == AF_UNIX && msgdsize(mp->b_cont) == 0 &&
1972 		    tdi->OPT_length != 0) {
1973 			/*
1974 			 * Attempt to extract AF_UNIX close indication
1975 			 * from the options. Ignore any other options -
1976 			 * those are handled once the message is removed
1977 			 * from the queue.
1978 			 * The close indication message should not carry data.
1979 			 */
1980 			void *opt;
1981 			t_uscalar_t optlen = tdi->OPT_length;
1982 
1983 			opt = sogetoff(mp, tdi->OPT_offset,
1984 			    optlen, __TPI_ALIGN_SIZE);
1985 			if (opt == NULL) {
1986 				/* The len/off falls outside mp */
1987 				freemsg(mp);
1988 				mutex_enter(&so->so_lock);
1989 				soseterror(so, EPROTO);
1990 				mutex_exit(&so->so_lock);
1991 				zcmn_err(getzoneid(), CE_WARN,
1992 				    "sockfs: T_optdata_ind with invalid "
1993 				    "optlen/offset %u/%d\n",
1994 				    optlen, tdi->OPT_offset);
1995 				return (NULL);
1996 			}
1997 			/*
1998 			 * If we received a close indication mark the
1999 			 * socket and discard this message.
2000 			 */
2001 			if (so_getopt_unix_close(opt, optlen)) {
2002 				mutex_enter(&so->so_lock);
2003 				socantsendmore(so);
2004 				sti->sti_faddr_valid = 0;
2005 				mutex_exit(&so->so_lock);
2006 				strsetwerror(SOTOV(so), 0, 0, sogetwrerr);
2007 				freemsg(mp);
2008 				return (NULL);
2009 			}
2010 		}
2011 		*allmsgsigs = S_INPUT | S_RDNORM;
2012 		*pollwakeups = POLLIN | POLLRDNORM;
2013 		*wakeups = RSLEEP;
2014 		return (mp);
2015 	}
2016 
2017 	case T_EXDATA_IND: {
2018 		mblk_t		*mctl, *mdata;
2019 		mblk_t *lbp;
2020 		union T_primitives *tprp;
2021 		struct stdata   *stp;
2022 		queue_t *qp;
2023 
2024 		if (MBLKL(mp) < sizeof (struct T_exdata_ind)) {
2025 			zcmn_err(getzoneid(), CE_WARN,
2026 			    "sockfs: Too short T_EXDATA_IND. Len = %ld\n",
2027 			    (ptrdiff_t)(MBLKL(mp)));
2028 			freemsg(mp);
2029 			return (NULL);
2030 		}
2031 		/*
2032 		 * Ignore zero-length T_EXDATA_IND messages. These might be
2033 		 * generated by some transports.
2034 		 *
2035 		 * This is needed to prevent read (which skips the M_PROTO
2036 		 * part) to unexpectedly return 0 (or return EWOULDBLOCK
2037 		 * on a non-blocking socket after select/poll has indicated
2038 		 * that data is available).
2039 		 */
2040 		dprintso(so, 1,
2041 		    ("T_EXDATA_IND(%p): counts %d/%d state %s\n",
2042 		    (void *)vp, sti->sti_oobsigcnt, sti->sti_oobcnt,
2043 		    pr_state(so->so_state, so->so_mode)));
2044 
2045 		if (msgdsize(mp->b_cont) == 0) {
2046 			dprintso(so, 0,
2047 			    ("strsock_proto: zero length T_EXDATA_IND\n"));
2048 			freemsg(mp);
2049 			return (NULL);
2050 		}
2051 
2052 		/*
2053 		 * Split into the T_EXDATA_IND and the M_DATA part.
2054 		 * We process these three pieces separately:
2055 		 *	signal generation
2056 		 *	handling T_EXDATA_IND
2057 		 *	handling M_DATA component
2058 		 */
2059 		mctl = mp;
2060 		mdata = mctl->b_cont;
2061 		mctl->b_cont = NULL;
2062 		mutex_enter(&so->so_lock);
2063 		so_oob_sig(so, 0, allmsgsigs, pollwakeups);
2064 		mctl = so_oob_exdata(so, mctl, allmsgsigs, pollwakeups);
2065 		mdata = so_oob_data(so, mdata, allmsgsigs, pollwakeups);
2066 
2067 		stp = vp->v_stream;
2068 		ASSERT(stp != NULL);
2069 		qp = _RD(stp->sd_wrq);
2070 
2071 		mutex_enter(QLOCK(qp));
2072 		lbp = qp->q_last;
2073 
2074 		/*
2075 		 * We want to avoid queueing up a string of T_EXDATA_IND
2076 		 * messages with no intervening data messages at the stream
2077 		 * head. These messages contribute to the total message
2078 		 * count. Eventually this can lead to STREAMS flow contol
2079 		 * and also cause TCP to advertise a zero window condition
2080 		 * to the peer. This can happen in the degenerate case where
2081 		 * the sender and receiver exchange only OOB data. The sender
2082 		 * only sends messages with MSG_OOB flag and the receiver
2083 		 * receives only MSG_OOB messages and does not use SO_OOBINLINE.
2084 		 * An example of this scenario has been reported in applications
2085 		 * that use OOB data to exchange heart beats. Flow control
2086 		 * relief will never happen if the application only reads OOB
2087 		 * data which is done directly by sorecvoob() and the
2088 		 * T_EXDATA_IND messages at the streamhead won't be consumed.
2089 		 * Note that there is no correctness issue in compressing the
2090 		 * string of T_EXDATA_IND messages into a single T_EXDATA_IND
2091 		 * message. A single read that does not specify MSG_OOB will
2092 		 * read across all the marks in a loop in sotpi_recvmsg().
2093 		 * Each mark is individually distinguishable only if the
2094 		 * T_EXDATA_IND messages are separated by data messages.
2095 		 */
2096 		if ((qp->q_first != NULL) && (DB_TYPE(lbp) == M_PROTO)) {
2097 			tprp = (union T_primitives *)lbp->b_rptr;
2098 			if ((tprp->type == T_EXDATA_IND) &&
2099 			    !(so->so_options & SO_OOBINLINE)) {
2100 
2101 				/*
2102 				 * free the new M_PROTO message
2103 				 */
2104 				freemsg(mctl);
2105 
2106 				/*
2107 				 * adjust the OOB count and OOB	signal count
2108 				 * just incremented for the new OOB data.
2109 				 */
2110 				sti->sti_oobcnt--;
2111 				sti->sti_oobsigcnt--;
2112 				mutex_exit(QLOCK(qp));
2113 				mutex_exit(&so->so_lock);
2114 				return (NULL);
2115 			}
2116 		}
2117 		mutex_exit(QLOCK(qp));
2118 
2119 		/*
2120 		 * Pass the T_EXDATA_IND and the M_DATA back separately
2121 		 * by using b_next linkage. (The stream head will queue any
2122 		 * b_next linked messages separately.) This is needed
2123 		 * since MSGMARK applies to the last by of the message
2124 		 * hence we can not have any M_DATA component attached
2125 		 * to the marked T_EXDATA_IND. Note that the stream head
2126 		 * will not consolidate M_DATA messages onto an MSGMARK'ed
2127 		 * message in order to preserve the constraint that
2128 		 * the T_EXDATA_IND always is a separate message.
2129 		 */
2130 		ASSERT(mctl != NULL);
2131 		mctl->b_next = mdata;
2132 		mp = mctl;
2133 #ifdef DEBUG
2134 		if (mdata == NULL) {
2135 			dprintso(so, 1,
2136 			    ("after outofline T_EXDATA_IND(%p): "
2137 			    "counts %d/%d  poll 0x%x sig 0x%x state %s\n",
2138 			    (void *)vp, sti->sti_oobsigcnt,
2139 			    sti->sti_oobcnt, *pollwakeups, *allmsgsigs,
2140 			    pr_state(so->so_state, so->so_mode)));
2141 		} else {
2142 			dprintso(so, 1,
2143 			    ("after inline T_EXDATA_IND(%p): "
2144 			    "counts %d/%d  poll 0x%x sig 0x%x state %s\n",
2145 			    (void *)vp, sti->sti_oobsigcnt,
2146 			    sti->sti_oobcnt, *pollwakeups, *allmsgsigs,
2147 			    pr_state(so->so_state, so->so_mode)));
2148 		}
2149 #endif /* DEBUG */
2150 		mutex_exit(&so->so_lock);
2151 		*wakeups = RSLEEP;
2152 		return (mp);
2153 	}
2154 
2155 	case T_CONN_CON: {
2156 		struct T_conn_con	*conn_con;
2157 		void			*addr;
2158 		t_uscalar_t		addrlen;
2159 
2160 		/*
2161 		 * Verify the state, update the state to ISCONNECTED,
2162 		 * record the potentially new address in the message,
2163 		 * and drop the message.
2164 		 */
2165 		if (MBLKL(mp) < sizeof (struct T_conn_con)) {
2166 			zcmn_err(getzoneid(), CE_WARN,
2167 			    "sockfs: Too short T_CONN_CON. Len = %ld\n",
2168 			    (ptrdiff_t)(MBLKL(mp)));
2169 			freemsg(mp);
2170 			return (NULL);
2171 		}
2172 
2173 		mutex_enter(&so->so_lock);
2174 		if ((so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) !=
2175 		    SS_ISCONNECTING) {
2176 			mutex_exit(&so->so_lock);
2177 			dprintso(so, 1,
2178 			    ("T_CONN_CON: state %x\n", so->so_state));
2179 			freemsg(mp);
2180 			return (NULL);
2181 		}
2182 
2183 		conn_con = &tpr->conn_con;
2184 		addrlen = conn_con->RES_length;
2185 		/*
2186 		 * Allow the address to be of different size than sent down
2187 		 * in the T_CONN_REQ as long as it doesn't exceed the maxlen.
2188 		 * For AF_UNIX require the identical length.
2189 		 */
2190 		if (so->so_family == AF_UNIX ?
2191 		    addrlen != (t_uscalar_t)sizeof (sti->sti_ux_laddr) :
2192 		    addrlen > (t_uscalar_t)sti->sti_faddr_maxlen) {
2193 			zcmn_err(getzoneid(), CE_WARN,
2194 			    "sockfs: T_conn_con with different "
2195 			    "length %u/%d\n",
2196 			    addrlen, conn_con->RES_length);
2197 			soisdisconnected(so, EPROTO);
2198 			sti->sti_laddr_valid = 0;
2199 			sti->sti_faddr_valid = 0;
2200 			mutex_exit(&so->so_lock);
2201 			strsetrerror(SOTOV(so), 0, 0, sogetrderr);
2202 			strsetwerror(SOTOV(so), 0, 0, sogetwrerr);
2203 			strseteof(SOTOV(so), 1);
2204 			freemsg(mp);
2205 			/*
2206 			 * strseteof takes care of read side wakeups,
2207 			 * pollwakeups, and signals.
2208 			 */
2209 			*wakeups = WSLEEP;
2210 			*allmsgsigs = S_OUTPUT;
2211 			*pollwakeups = POLLOUT;
2212 			return (NULL);
2213 		}
2214 		addr = sogetoff(mp, conn_con->RES_offset, addrlen, 1);
2215 		if (addr == NULL) {
2216 			zcmn_err(getzoneid(), CE_WARN,
2217 			    "sockfs: T_conn_con with invalid "
2218 			    "addrlen/offset %u/%d\n",
2219 			    addrlen, conn_con->RES_offset);
2220 			mutex_exit(&so->so_lock);
2221 			strsetrerror(SOTOV(so), 0, 0, sogetrderr);
2222 			strsetwerror(SOTOV(so), 0, 0, sogetwrerr);
2223 			strseteof(SOTOV(so), 1);
2224 			freemsg(mp);
2225 			/*
2226 			 * strseteof takes care of read side wakeups,
2227 			 * pollwakeups, and signals.
2228 			 */
2229 			*wakeups = WSLEEP;
2230 			*allmsgsigs = S_OUTPUT;
2231 			*pollwakeups = POLLOUT;
2232 			return (NULL);
2233 		}
2234 
2235 		/*
2236 		 * Save for getpeername.
2237 		 */
2238 		if (so->so_family != AF_UNIX) {
2239 			sti->sti_faddr_len = (socklen_t)addrlen;
2240 			ASSERT(sti->sti_faddr_len <= sti->sti_faddr_maxlen);
2241 			bcopy(addr, sti->sti_faddr_sa, addrlen);
2242 			sti->sti_faddr_valid = 1;
2243 		}
2244 
2245 		if (so->so_peercred != NULL)
2246 			crfree(so->so_peercred);
2247 		so->so_peercred = msg_getcred(mp, &so->so_cpid);
2248 		if (so->so_peercred != NULL)
2249 			crhold(so->so_peercred);
2250 
2251 		/* Wakeup anybody sleeping in sowaitconnected */
2252 		soisconnected(so);
2253 		mutex_exit(&so->so_lock);
2254 
2255 		/*
2256 		 * The socket is now available for sending data.
2257 		 */
2258 		*wakeups = WSLEEP;
2259 		*allmsgsigs = S_OUTPUT;
2260 		*pollwakeups = POLLOUT;
2261 		freemsg(mp);
2262 		return (NULL);
2263 	}
2264 
2265 	/*
2266 	 * Extra processing in case of an SSL proxy, before queuing or
2267 	 * forwarding to the fallback endpoint
2268 	 */
2269 	case T_SSL_PROXY_CONN_IND:
2270 	case T_CONN_IND:
2271 		/*
2272 		 * Verify the min size and queue the message on
2273 		 * the sti_conn_ind_head/tail list.
2274 		 */
2275 		if (MBLKL(mp) < sizeof (struct T_conn_ind)) {
2276 			zcmn_err(getzoneid(), CE_WARN,
2277 			    "sockfs: Too short T_CONN_IND. Len = %ld\n",
2278 			    (ptrdiff_t)(MBLKL(mp)));
2279 			freemsg(mp);
2280 			return (NULL);
2281 		}
2282 
2283 		if (auditing)
2284 			audit_sock(T_CONN_IND, strvp2wq(vp), mp, 0);
2285 		if (!(so->so_state & SS_ACCEPTCONN)) {
2286 			zcmn_err(getzoneid(), CE_WARN,
2287 			    "sockfs: T_conn_ind on non-listening socket\n");
2288 			freemsg(mp);
2289 			return (NULL);
2290 		}
2291 
2292 		if (tpr->type == T_SSL_PROXY_CONN_IND && mp->b_cont == NULL) {
2293 			/* No context: need to fall back */
2294 			struct sonode *fbso;
2295 			stdata_t *fbstp;
2296 
2297 			tpr->type = T_CONN_IND;
2298 
2299 			fbso = kssl_find_fallback(sti->sti_kssl_ent);
2300 
2301 			/*
2302 			 * No fallback: the remote will timeout and
2303 			 * disconnect.
2304 			 */
2305 			if (fbso == NULL) {
2306 				freemsg(mp);
2307 				return (NULL);
2308 			}
2309 			fbstp = SOTOV(fbso)->v_stream;
2310 			qreply(fbstp->sd_wrq->q_next, mp);
2311 			return (NULL);
2312 		}
2313 		soqueueconnind(so, mp);
2314 		*allmsgsigs = S_INPUT | S_RDNORM;
2315 		*pollwakeups = POLLIN | POLLRDNORM;
2316 		*wakeups = RSLEEP;
2317 		return (NULL);
2318 
2319 	case T_ORDREL_IND:
2320 		if (MBLKL(mp) < sizeof (struct T_ordrel_ind)) {
2321 			zcmn_err(getzoneid(), CE_WARN,
2322 			    "sockfs: Too short T_ORDREL_IND. Len = %ld\n",
2323 			    (ptrdiff_t)(MBLKL(mp)));
2324 			freemsg(mp);
2325 			return (NULL);
2326 		}
2327 
2328 		/*
2329 		 * Some providers send this when not fully connected.
2330 		 * SunLink X.25 needs to retrieve disconnect reason after
2331 		 * disconnect for compatibility. It uses T_ORDREL_IND
2332 		 * instead of T_DISCON_IND so that it may use the
2333 		 * endpoint after a connect failure to retrieve the
2334 		 * reason using an ioctl. Thus we explicitly clear
2335 		 * SS_ISCONNECTING here for SunLink X.25.
2336 		 * This is a needed TPI violation.
2337 		 */
2338 		mutex_enter(&so->so_lock);
2339 		so->so_state &= ~SS_ISCONNECTING;
2340 		socantrcvmore(so);
2341 		mutex_exit(&so->so_lock);
2342 		strseteof(SOTOV(so), 1);
2343 		/*
2344 		 * strseteof takes care of read side wakeups,
2345 		 * pollwakeups, and signals.
2346 		 */
2347 		freemsg(mp);
2348 		return (NULL);
2349 
2350 	case T_DISCON_IND:
2351 		if (MBLKL(mp) < sizeof (struct T_discon_ind)) {
2352 			zcmn_err(getzoneid(), CE_WARN,
2353 			    "sockfs: Too short T_DISCON_IND. Len = %ld\n",
2354 			    (ptrdiff_t)(MBLKL(mp)));
2355 			freemsg(mp);
2356 			return (NULL);
2357 		}
2358 		if (so->so_state & SS_ACCEPTCONN) {
2359 			/*
2360 			 * This is a listener. Look for a queued T_CONN_IND
2361 			 * with a matching sequence number and remove it
2362 			 * from the list.
2363 			 * It is normal to not find the sequence number since
2364 			 * the soaccept might have already dequeued it
2365 			 * (in which case the T_CONN_RES will fail with
2366 			 * TBADSEQ).
2367 			 */
2368 			(void) soflushconnind(so, tpr->discon_ind.SEQ_number);
2369 			freemsg(mp);
2370 			return (0);
2371 		}
2372 
2373 		/*
2374 		 * Not a listener
2375 		 *
2376 		 * If SS_CANTRCVMORE for AF_UNIX ignore the discon_reason.
2377 		 * Such a discon_ind appears when the peer has first done
2378 		 * a shutdown() followed by a close() in which case we just
2379 		 * want to record socantsendmore.
2380 		 * In this case sockfs first receives a T_ORDREL_IND followed
2381 		 * by a T_DISCON_IND.
2382 		 * Note that for other transports (e.g. TCP) we need to handle
2383 		 * the discon_ind in this case since it signals an error.
2384 		 */
2385 		mutex_enter(&so->so_lock);
2386 		if ((so->so_state & SS_CANTRCVMORE) &&
2387 		    (so->so_family == AF_UNIX)) {
2388 			socantsendmore(so);
2389 			sti->sti_faddr_valid = 0;
2390 			mutex_exit(&so->so_lock);
2391 			strsetwerror(SOTOV(so), 0, 0, sogetwrerr);
2392 			dprintso(so, 1,
2393 			    ("T_DISCON_IND: error %d\n", so->so_error));
2394 			freemsg(mp);
2395 			/*
2396 			 * Set these variables for caller to process them.
2397 			 * For the else part where T_DISCON_IND is processed,
2398 			 * this will be done in the function being called
2399 			 * (strsock_discon_ind())
2400 			 */
2401 			*wakeups = WSLEEP;
2402 			*allmsgsigs = S_OUTPUT;
2403 			*pollwakeups = POLLOUT;
2404 		} else if (so->so_flag & (SOASYNC_UNBIND | SOLOCKED)) {
2405 			/*
2406 			 * Deferred processing of T_DISCON_IND
2407 			 */
2408 			so_save_discon_ind(so, mp, strsock_discon_ind);
2409 			mutex_exit(&so->so_lock);
2410 		} else {
2411 			/*
2412 			 * Process T_DISCON_IND now
2413 			 */
2414 			(void) strsock_discon_ind(so, mp);
2415 			mutex_exit(&so->so_lock);
2416 		}
2417 		return (NULL);
2418 
2419 	case T_UDERROR_IND: {
2420 		struct T_uderror_ind	*tudi = &tpr->uderror_ind;
2421 		void			*addr;
2422 		t_uscalar_t		addrlen;
2423 		int			error;
2424 
2425 		dprintso(so, 0,
2426 		    ("T_UDERROR_IND: error %d\n", tudi->ERROR_type));
2427 
2428 		if (MBLKL(mp) < sizeof (struct T_uderror_ind)) {
2429 			zcmn_err(getzoneid(), CE_WARN,
2430 			    "sockfs: Too short T_UDERROR_IND. Len = %ld\n",
2431 			    (ptrdiff_t)(MBLKL(mp)));
2432 			freemsg(mp);
2433 			return (NULL);
2434 		}
2435 		/* Ignore on connection-oriented transports */
2436 		if (so->so_mode & SM_CONNREQUIRED) {
2437 			freemsg(mp);
2438 			eprintsoline(so, 0);
2439 			zcmn_err(getzoneid(), CE_WARN,
2440 			    "sockfs: T_uderror_ind on connection-oriented "
2441 			    "transport\n");
2442 			return (NULL);
2443 		}
2444 		addrlen = tudi->DEST_length;
2445 		addr = sogetoff(mp, tudi->DEST_offset, addrlen, 1);
2446 		if (addr == NULL) {
2447 			zcmn_err(getzoneid(), CE_WARN,
2448 			    "sockfs: T_uderror_ind with invalid "
2449 			    "addrlen/offset %u/%d\n",
2450 			    addrlen, tudi->DEST_offset);
2451 			freemsg(mp);
2452 			return (NULL);
2453 		}
2454 
2455 		/* Verify source address for connected socket. */
2456 		mutex_enter(&so->so_lock);
2457 		if (so->so_state & SS_ISCONNECTED) {
2458 			void *faddr;
2459 			t_uscalar_t faddr_len;
2460 			boolean_t match = B_FALSE;
2461 
2462 			switch (so->so_family) {
2463 			case AF_INET: {
2464 				/* Compare just IP address and port */
2465 				struct sockaddr_in *sin1, *sin2;
2466 
2467 				sin1 = (struct sockaddr_in *)sti->sti_faddr_sa;
2468 				sin2 = (struct sockaddr_in *)addr;
2469 				if (addrlen == sizeof (struct sockaddr_in) &&
2470 				    sin1->sin_port == sin2->sin_port &&
2471 				    sin1->sin_addr.s_addr ==
2472 				    sin2->sin_addr.s_addr)
2473 					match = B_TRUE;
2474 				break;
2475 			}
2476 			case AF_INET6: {
2477 				/* Compare just IP address and port. Not flow */
2478 				struct sockaddr_in6 *sin1, *sin2;
2479 
2480 				sin1 = (struct sockaddr_in6 *)sti->sti_faddr_sa;
2481 				sin2 = (struct sockaddr_in6 *)addr;
2482 				if (addrlen == sizeof (struct sockaddr_in6) &&
2483 				    sin1->sin6_port == sin2->sin6_port &&
2484 				    IN6_ARE_ADDR_EQUAL(&sin1->sin6_addr,
2485 				    &sin2->sin6_addr))
2486 					match = B_TRUE;
2487 				break;
2488 			}
2489 			case AF_UNIX:
2490 				faddr = &sti->sti_ux_faddr;
2491 				faddr_len =
2492 				    (t_uscalar_t)sizeof (sti->sti_ux_faddr);
2493 				if (faddr_len == addrlen &&
2494 				    bcmp(addr, faddr, addrlen) == 0)
2495 					match = B_TRUE;
2496 				break;
2497 			default:
2498 				faddr = sti->sti_faddr_sa;
2499 				faddr_len = (t_uscalar_t)sti->sti_faddr_len;
2500 				if (faddr_len == addrlen &&
2501 				    bcmp(addr, faddr, addrlen) == 0)
2502 					match = B_TRUE;
2503 				break;
2504 			}
2505 
2506 			if (!match) {
2507 #ifdef DEBUG
2508 				dprintso(so, 0,
2509 				    ("sockfs: T_UDERR_IND mismatch: %s - ",
2510 				    pr_addr(so->so_family,
2511 				    (struct sockaddr *)addr, addrlen)));
2512 				dprintso(so, 0, ("%s\n",
2513 				    pr_addr(so->so_family, sti->sti_faddr_sa,
2514 				    sti->sti_faddr_len)));
2515 #endif /* DEBUG */
2516 				mutex_exit(&so->so_lock);
2517 				freemsg(mp);
2518 				return (NULL);
2519 			}
2520 			/*
2521 			 * Make the write error nonpersistent. If the error
2522 			 * is zero we use ECONNRESET.
2523 			 * This assumes that the name space for ERROR_type
2524 			 * is the errno name space.
2525 			 */
2526 			if (tudi->ERROR_type != 0)
2527 				error = tudi->ERROR_type;
2528 			else
2529 				error = ECONNRESET;
2530 
2531 			soseterror(so, error);
2532 			mutex_exit(&so->so_lock);
2533 			strsetrerror(SOTOV(so), 0, 0, sogetrderr);
2534 			strsetwerror(SOTOV(so), 0, 0, sogetwrerr);
2535 			*wakeups = RSLEEP | WSLEEP;
2536 			*allmsgsigs = S_INPUT | S_RDNORM | S_OUTPUT;
2537 			*pollwakeups = POLLIN | POLLRDNORM | POLLOUT;
2538 			freemsg(mp);
2539 			return (NULL);
2540 		}
2541 		/*
2542 		 * If the application asked for delayed errors
2543 		 * record the T_UDERROR_IND sti_eaddr_mp and the reason in
2544 		 * sti_delayed_error for delayed error posting. If the reason
2545 		 * is zero use ECONNRESET.
2546 		 * Note that delayed error indications do not make sense for
2547 		 * AF_UNIX sockets since sendto checks that the destination
2548 		 * address is valid at the time of the sendto.
2549 		 */
2550 		if (!(so->so_options & SO_DGRAM_ERRIND)) {
2551 			mutex_exit(&so->so_lock);
2552 			freemsg(mp);
2553 			return (NULL);
2554 		}
2555 		if (sti->sti_eaddr_mp != NULL)
2556 			freemsg(sti->sti_eaddr_mp);
2557 
2558 		sti->sti_eaddr_mp = mp;
2559 		if (tudi->ERROR_type != 0)
2560 			error = tudi->ERROR_type;
2561 		else
2562 			error = ECONNRESET;
2563 		sti->sti_delayed_error = (ushort_t)error;
2564 		mutex_exit(&so->so_lock);
2565 		return (NULL);
2566 	}
2567 
2568 	case T_ERROR_ACK:
2569 		dprintso(so, 0,
2570 		    ("strsock_proto: T_ERROR_ACK for %d, error %d/%d\n",
2571 		    tpr->error_ack.ERROR_prim,
2572 		    tpr->error_ack.TLI_error,
2573 		    tpr->error_ack.UNIX_error));
2574 
2575 		if (MBLKL(mp) < sizeof (struct T_error_ack)) {
2576 			zcmn_err(getzoneid(), CE_WARN,
2577 			    "sockfs: Too short T_ERROR_ACK. Len = %ld\n",
2578 			    (ptrdiff_t)(MBLKL(mp)));
2579 			freemsg(mp);
2580 			return (NULL);
2581 		}
2582 		/*
2583 		 * Check if we were waiting for the async message
2584 		 */
2585 		mutex_enter(&so->so_lock);
2586 		if ((so->so_flag & SOASYNC_UNBIND) &&
2587 		    tpr->error_ack.ERROR_prim == T_UNBIND_REQ) {
2588 			so_unlock_single(so, SOASYNC_UNBIND);
2589 			mutex_exit(&so->so_lock);
2590 			freemsg(mp);
2591 			return (NULL);
2592 		}
2593 		mutex_exit(&so->so_lock);
2594 		soqueueack(so, mp);
2595 		return (NULL);
2596 
2597 	case T_OK_ACK:
2598 		if (MBLKL(mp) < sizeof (struct T_ok_ack)) {
2599 			zcmn_err(getzoneid(), CE_WARN,
2600 			    "sockfs: Too short T_OK_ACK. Len = %ld\n",
2601 			    (ptrdiff_t)(MBLKL(mp)));
2602 			freemsg(mp);
2603 			return (NULL);
2604 		}
2605 		/*
2606 		 * Check if we were waiting for the async message
2607 		 */
2608 		mutex_enter(&so->so_lock);
2609 		if ((so->so_flag & SOASYNC_UNBIND) &&
2610 		    tpr->ok_ack.CORRECT_prim == T_UNBIND_REQ) {
2611 			dprintso(so, 1,
2612 			    ("strsock_proto: T_OK_ACK async unbind\n"));
2613 			so_unlock_single(so, SOASYNC_UNBIND);
2614 			mutex_exit(&so->so_lock);
2615 			freemsg(mp);
2616 			return (NULL);
2617 		}
2618 		mutex_exit(&so->so_lock);
2619 		soqueueack(so, mp);
2620 		return (NULL);
2621 
2622 	case T_INFO_ACK:
2623 		if (MBLKL(mp) < sizeof (struct T_info_ack)) {
2624 			zcmn_err(getzoneid(), CE_WARN,
2625 			    "sockfs: Too short T_INFO_ACK. Len = %ld\n",
2626 			    (ptrdiff_t)(MBLKL(mp)));
2627 			freemsg(mp);
2628 			return (NULL);
2629 		}
2630 		soqueueack(so, mp);
2631 		return (NULL);
2632 
2633 	case T_CAPABILITY_ACK:
2634 		/*
2635 		 * A T_capability_ack need only be large enough to hold
2636 		 * the PRIM_type and CAP_bits1 fields; checking for anything
2637 		 * larger might reject a correct response from an older
2638 		 * provider.
2639 		 */
2640 		if (MBLKL(mp) < 2 * sizeof (t_uscalar_t)) {
2641 			zcmn_err(getzoneid(), CE_WARN,
2642 			    "sockfs: Too short T_CAPABILITY_ACK. Len = %ld\n",
2643 			    (ptrdiff_t)(MBLKL(mp)));
2644 			freemsg(mp);
2645 			return (NULL);
2646 		}
2647 		soqueueack(so, mp);
2648 		return (NULL);
2649 
2650 	case T_BIND_ACK:
2651 		if (MBLKL(mp) < sizeof (struct T_bind_ack)) {
2652 			zcmn_err(getzoneid(), CE_WARN,
2653 			    "sockfs: Too short T_BIND_ACK. Len = %ld\n",
2654 			    (ptrdiff_t)(MBLKL(mp)));
2655 			freemsg(mp);
2656 			return (NULL);
2657 		}
2658 		soqueueack(so, mp);
2659 		return (NULL);
2660 
2661 	case T_OPTMGMT_ACK:
2662 		if (MBLKL(mp) < sizeof (struct T_optmgmt_ack)) {
2663 			zcmn_err(getzoneid(), CE_WARN,
2664 			    "sockfs: Too short T_OPTMGMT_ACK. Len = %ld\n",
2665 			    (ptrdiff_t)(MBLKL(mp)));
2666 			freemsg(mp);
2667 			return (NULL);
2668 		}
2669 		soqueueack(so, mp);
2670 		return (NULL);
2671 	default:
2672 #ifdef DEBUG
2673 		zcmn_err(getzoneid(), CE_WARN,
2674 		    "sockfs: unknown TPI primitive %d received\n",
2675 		    tpr->type);
2676 #endif /* DEBUG */
2677 		freemsg(mp);
2678 		return (NULL);
2679 	}
2680 }
2681 
2682 /*
2683  * This routine is registered with the stream head to receive other
2684  * (non-data, and non-proto) messages.
2685  *
2686  * Returns NULL if the message was consumed.
2687  * Returns an mblk to make that mblk be processed by the stream head.
2688  *
2689  * Sets the return parameters (*wakeups, *firstmsgsigs, *allmsgsigs, and
2690  * *pollwakeups) for the stream head to take action on.
2691  */
2692 static mblk_t *
2693 strsock_misc(vnode_t *vp, mblk_t *mp,
2694 		strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
2695 		strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
2696 {
2697 	struct sonode *so;
2698 	sotpi_info_t *sti;
2699 
2700 	so = VTOSO(vp);
2701 	sti = SOTOTPI(so);
2702 
2703 	dprintso(so, 1, ("strsock_misc(%p, %p, 0x%x)\n",
2704 	    (void *)vp, (void *)mp, DB_TYPE(mp)));
2705 
2706 	/* Set default return values */
2707 	*wakeups = *allmsgsigs = *firstmsgsigs = *pollwakeups = 0;
2708 
2709 	switch (DB_TYPE(mp)) {
2710 	case M_PCSIG:
2711 		/*
2712 		 * This assumes that an M_PCSIG for the urgent data arrives
2713 		 * before the corresponding T_EXDATA_IND.
2714 		 *
2715 		 * Note: Just like in SunOS 4.X and 4.4BSD a poll will be
2716 		 * awoken before the urgent data shows up.
2717 		 * For OOBINLINE this can result in select returning
2718 		 * only exceptions as opposed to except|read.
2719 		 */
2720 		if (*mp->b_rptr == SIGURG) {
2721 			mutex_enter(&so->so_lock);
2722 			dprintso(so, 1,
2723 			    ("SIGURG(%p): counts %d/%d state %s\n",
2724 			    (void *)vp, sti->sti_oobsigcnt, sti->sti_oobcnt,
2725 			    pr_state(so->so_state, so->so_mode)));
2726 			so_oob_sig(so, 1, allmsgsigs, pollwakeups);
2727 			dprintso(so, 1,
2728 			    ("after SIGURG(%p): counts %d/%d "
2729 			    " poll 0x%x sig 0x%x state %s\n",
2730 			    (void *)vp, sti->sti_oobsigcnt, sti->sti_oobcnt,
2731 			    *pollwakeups, *allmsgsigs,
2732 			    pr_state(so->so_state, so->so_mode)));
2733 			mutex_exit(&so->so_lock);
2734 		}
2735 		freemsg(mp);
2736 		return (NULL);
2737 
2738 	case M_SIG:
2739 	case M_HANGUP:
2740 	case M_UNHANGUP:
2741 	case M_ERROR:
2742 		/* M_ERRORs etc are ignored */
2743 		freemsg(mp);
2744 		return (NULL);
2745 
2746 	case M_FLUSH:
2747 		/*
2748 		 * Do not flush read queue. If the M_FLUSH
2749 		 * arrives because of an impending T_discon_ind
2750 		 * we still have to keep any queued data - this is part of
2751 		 * socket semantics.
2752 		 */
2753 		if (*mp->b_rptr & FLUSHW) {
2754 			*mp->b_rptr &= ~FLUSHR;
2755 			return (mp);
2756 		}
2757 		freemsg(mp);
2758 		return (NULL);
2759 
2760 	default:
2761 		return (mp);
2762 	}
2763 }
2764 
2765 
2766 /* Register to receive signals for certain events */
2767 int
2768 so_set_asyncsigs(vnode_t *vp, pid_t pgrp, int events, int mode, cred_t *cr)
2769 {
2770 	struct strsigset ss;
2771 	int32_t rval;
2772 
2773 	/*
2774 	 * Note that SOLOCKED will be set except for the call from soaccept().
2775 	 */
2776 	ASSERT(!mutex_owned(&VTOSO(vp)->so_lock));
2777 	ss.ss_pid = pgrp;
2778 	ss.ss_events = events;
2779 	return (strioctl(vp, I_ESETSIG, (intptr_t)&ss, mode, K_TO_K, cr,
2780 	    &rval));
2781 }
2782 
2783 
2784 /* Register for events matching the SS_ASYNC flag */
2785 int
2786 so_set_events(struct sonode *so, vnode_t *vp, cred_t *cr)
2787 {
2788 	int events = so->so_state & SS_ASYNC ?
2789 	    S_RDBAND | S_BANDURG | S_RDNORM | S_OUTPUT :
2790 	    S_RDBAND | S_BANDURG;
2791 
2792 	return (so_set_asyncsigs(vp, so->so_pgrp, events, 0, cr));
2793 }
2794 
2795 
2796 /* Change the SS_ASYNC flag, and update signal delivery if needed */
2797 int
2798 so_flip_async(struct sonode *so, vnode_t *vp, int mode, cred_t *cr)
2799 {
2800 	ASSERT(mutex_owned(&so->so_lock));
2801 	if (so->so_pgrp != 0) {
2802 		int error;
2803 		int events = so->so_state & SS_ASYNC ?		/* Old flag */
2804 		    S_RDBAND | S_BANDURG :			/* New sigs */
2805 		    S_RDBAND | S_BANDURG | S_RDNORM | S_OUTPUT;
2806 
2807 		so_lock_single(so);
2808 		mutex_exit(&so->so_lock);
2809 
2810 		error = so_set_asyncsigs(vp, so->so_pgrp, events, mode, cr);
2811 
2812 		mutex_enter(&so->so_lock);
2813 		so_unlock_single(so, SOLOCKED);
2814 		if (error)
2815 			return (error);
2816 	}
2817 	so->so_state ^= SS_ASYNC;
2818 	return (0);
2819 }
2820 
2821 /*
2822  * Set new pid/pgrp for SIGPOLL (or SIGIO for FIOASYNC mode), replacing
2823  * any existing one.  If passed zero, just clear the existing one.
2824  */
2825 int
2826 so_set_siggrp(struct sonode *so, vnode_t *vp, pid_t pgrp, int mode, cred_t *cr)
2827 {
2828 	int events = so->so_state & SS_ASYNC ?
2829 	    S_RDBAND | S_BANDURG | S_RDNORM | S_OUTPUT :
2830 	    S_RDBAND | S_BANDURG;
2831 	int error;
2832 
2833 	ASSERT(mutex_owned(&so->so_lock));
2834 
2835 	/*
2836 	 * Change socket process (group).
2837 	 *
2838 	 * strioctl (via so_set_asyncsigs) will perform permission check and
2839 	 * also keep a PID_HOLD to prevent the pid from being reused.
2840 	 */
2841 	so_lock_single(so);
2842 	mutex_exit(&so->so_lock);
2843 
2844 	if (pgrp != 0) {
2845 		dprintso(so, 1, ("setown: adding pgrp %d ev 0x%x\n",
2846 		    pgrp, events));
2847 		error = so_set_asyncsigs(vp, pgrp, events, mode, cr);
2848 		if (error != 0) {
2849 			eprintsoline(so, error);
2850 			goto bad;
2851 		}
2852 	}
2853 	/* Remove the previously registered process/group */
2854 	if (so->so_pgrp != 0) {
2855 		dprintso(so, 1, ("setown: removing pgrp %d\n", so->so_pgrp));
2856 		error = so_set_asyncsigs(vp, so->so_pgrp, 0, mode, cr);
2857 		if (error != 0) {
2858 			eprintsoline(so, error);
2859 			error = 0;
2860 		}
2861 	}
2862 	mutex_enter(&so->so_lock);
2863 	so_unlock_single(so, SOLOCKED);
2864 	so->so_pgrp = pgrp;
2865 	return (0);
2866 bad:
2867 	mutex_enter(&so->so_lock);
2868 	so_unlock_single(so, SOLOCKED);
2869 	return (error);
2870 }
2871 
2872 /*
2873  * Wrapper for getmsg. If the socket has been converted to a stream
2874  * pass the request to the stream head.
2875  */
2876 int
2877 sock_getmsg(
2878 	struct vnode *vp,
2879 	struct strbuf *mctl,
2880 	struct strbuf *mdata,
2881 	uchar_t *prip,
2882 	int *flagsp,
2883 	int fmode,
2884 	rval_t *rvp
2885 )
2886 {
2887 	struct sonode *so;
2888 
2889 	ASSERT(vp->v_type == VSOCK);
2890 	/*
2891 	 * Use the stream head to find the real socket vnode.
2892 	 * This is needed when namefs sits above sockfs.  Some
2893 	 * sockets (like SCTP) are not streams.
2894 	 */
2895 	if (!vp->v_stream) {
2896 		return (ENOSTR);
2897 	}
2898 	ASSERT(vp->v_stream->sd_vnode);
2899 	vp = vp->v_stream->sd_vnode;
2900 	ASSERT(vn_matchops(vp, socket_vnodeops));
2901 	so = VTOSO(vp);
2902 
2903 	dprintso(so, 1, ("sock_getmsg(%p) %s\n",
2904 	    (void *)so, pr_state(so->so_state, so->so_mode)));
2905 
2906 	if (so->so_version == SOV_STREAM) {
2907 		/* The imaginary "sockmod" has been popped - act as a stream */
2908 		return (strgetmsg(vp, mctl, mdata, prip, flagsp, fmode, rvp));
2909 	}
2910 	eprintsoline(so, ENOSTR);
2911 	return (ENOSTR);
2912 }
2913 
2914 /*
2915  * Wrapper for putmsg. If the socket has been converted to a stream
2916  * pass the request to the stream head.
2917  *
2918  * Note that a while a regular socket (SOV_SOCKSTREAM) does support the
2919  * streams ioctl set it does not support putmsg and getmsg.
2920  * Allowing putmsg would prevent sockfs from tracking the state of
2921  * the socket/transport and would also invalidate the locking in sockfs.
2922  */
2923 int
2924 sock_putmsg(
2925 	struct vnode *vp,
2926 	struct strbuf *mctl,
2927 	struct strbuf *mdata,
2928 	uchar_t pri,
2929 	int flag,
2930 	int fmode
2931 )
2932 {
2933 	struct sonode *so;
2934 
2935 	ASSERT(vp->v_type == VSOCK);
2936 	/*
2937 	 * Use the stream head to find the real socket vnode.
2938 	 * This is needed when namefs sits above sockfs.
2939 	 */
2940 	if (!vp->v_stream) {
2941 		return (ENOSTR);
2942 	}
2943 	ASSERT(vp->v_stream->sd_vnode);
2944 	vp = vp->v_stream->sd_vnode;
2945 	ASSERT(vn_matchops(vp, socket_vnodeops));
2946 	so = VTOSO(vp);
2947 
2948 	dprintso(so, 1, ("sock_putmsg(%p) %s\n",
2949 	    (void *)so, pr_state(so->so_state, so->so_mode)));
2950 
2951 	if (so->so_version == SOV_STREAM) {
2952 		/* The imaginary "sockmod" has been popped - act as a stream */
2953 		return (strputmsg(vp, mctl, mdata, pri, flag, fmode));
2954 	}
2955 	eprintsoline(so, ENOSTR);
2956 	return (ENOSTR);
2957 }
2958 
2959 /*
2960  * Special function called only from f_getfl().
2961  * Returns FASYNC if the SS_ASYNC flag is set on a socket, else 0.
2962  * No locks are acquired here, so it is safe to use while uf_lock is held.
2963  * This exists solely for BSD fcntl() FASYNC compatibility.
2964  */
2965 int
2966 sock_getfasync(vnode_t *vp)
2967 {
2968 	struct sonode *so;
2969 
2970 	ASSERT(vp->v_type == VSOCK);
2971 	/*
2972 	 * For stream model, v_stream is used; For non-stream, v_stream always
2973 	 * equals NULL
2974 	 */
2975 	if (vp->v_stream != NULL)
2976 		so = VTOSO(vp->v_stream->sd_vnode);
2977 	else
2978 		so = VTOSO(vp);
2979 
2980 	if (so->so_version == SOV_STREAM || !(so->so_state & SS_ASYNC))
2981 		return (0);
2982 
2983 	return (FASYNC);
2984 }
2985