xref: /titanic_52/usr/src/uts/common/fs/sockfs/socktpi.c (revision 3e95bd4ab92abca814bd28e854607d1975c7dc88)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 
22 /*
23  * Copyright (c) 1995, 2010, Oracle and/or its affiliates. All rights reserved.
24  */
25 
26 #include <sys/types.h>
27 #include <sys/t_lock.h>
28 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/buf.h>
31 #include <sys/conf.h>
32 #include <sys/cred.h>
33 #include <sys/kmem.h>
34 #include <sys/kmem_impl.h>
35 #include <sys/sysmacros.h>
36 #include <sys/vfs.h>
37 #include <sys/vnode.h>
38 #include <sys/debug.h>
39 #include <sys/errno.h>
40 #include <sys/time.h>
41 #include <sys/file.h>
42 #include <sys/open.h>
43 #include <sys/user.h>
44 #include <sys/termios.h>
45 #include <sys/stream.h>
46 #include <sys/strsubr.h>
47 #include <sys/strsun.h>
48 #include <sys/suntpi.h>
49 #include <sys/ddi.h>
50 #include <sys/esunddi.h>
51 #include <sys/flock.h>
52 #include <sys/modctl.h>
53 #include <sys/vtrace.h>
54 #include <sys/cmn_err.h>
55 #include <sys/pathname.h>
56 
57 #include <sys/socket.h>
58 #include <sys/socketvar.h>
59 #include <sys/sockio.h>
60 #include <netinet/in.h>
61 #include <sys/un.h>
62 #include <sys/strsun.h>
63 
64 #include <sys/tiuser.h>
65 #define	_SUN_TPI_VERSION	2
66 #include <sys/tihdr.h>
67 #include <sys/timod.h>		/* TI_GETMYNAME, TI_GETPEERNAME */
68 
69 #include <c2/audit.h>
70 
71 #include <inet/common.h>
72 #include <inet/ip.h>
73 #include <inet/ip6.h>
74 #include <inet/tcp.h>
75 #include <inet/udp_impl.h>
76 
77 #include <sys/zone.h>
78 
79 #include <fs/sockfs/nl7c.h>
80 #include <fs/sockfs/nl7curi.h>
81 
82 #include <inet/kssl/ksslapi.h>
83 
84 #include <fs/sockfs/sockcommon.h>
85 #include <fs/sockfs/socktpi.h>
86 #include <fs/sockfs/socktpi_impl.h>
87 
88 /*
89  * Possible failures when memory can't be allocated. The documented behavior:
90  *
91  * 		5.5:			4.X:		XNET:
92  * accept:	ENOMEM/ENOSR/EINTR	- (EINTR)	ENOMEM/ENOBUFS/ENOSR/
93  *							EINTR
94  *	(4.X does not document EINTR but returns it)
95  * bind:	ENOSR			-		ENOBUFS/ENOSR
96  * connect: 	EINTR			EINTR		ENOBUFS/ENOSR/EINTR
97  * getpeername:	ENOMEM/ENOSR		ENOBUFS (-)	ENOBUFS/ENOSR
98  * getsockname:	ENOMEM/ENOSR		ENOBUFS (-)	ENOBUFS/ENOSR
99  *	(4.X getpeername and getsockname do not fail in practice)
100  * getsockopt:	ENOMEM/ENOSR		-		ENOBUFS/ENOSR
101  * listen:	-			-		ENOBUFS
102  * recv:	ENOMEM/ENOSR/EINTR	EINTR		ENOBUFS/ENOMEM/ENOSR/
103  *							EINTR
104  * send:	ENOMEM/ENOSR/EINTR	ENOBUFS/EINTR	ENOBUFS/ENOMEM/ENOSR/
105  *							EINTR
106  * setsockopt:	ENOMEM/ENOSR		-		ENOBUFS/ENOMEM/ENOSR
107  * shutdown:	ENOMEM/ENOSR		-		ENOBUFS/ENOSR
108  * socket:	ENOMEM/ENOSR		ENOBUFS		ENOBUFS/ENOMEM/ENOSR
109  * socketpair:	ENOMEM/ENOSR		-		ENOBUFS/ENOMEM/ENOSR
110  *
111  * Resolution. When allocation fails:
112  *	recv: return EINTR
113  *	send: return EINTR
114  *	connect, accept: EINTR
115  *	bind, listen, shutdown (unbind, unix_close, disconnect): sleep
116  *	socket, socketpair: ENOBUFS
117  *	getpeername, getsockname: sleep
118  *	getsockopt, setsockopt: sleep
119  */
120 
121 #ifdef SOCK_TEST
122 /*
123  * Variables that make sockfs do something other than the standard TPI
124  * for the AF_INET transports.
125  *
126  * solisten_tpi_tcp:
127  *	TCP can handle a O_T_BIND_REQ with an increased backlog even though
128  *	the transport is already bound. This is needed to avoid loosing the
129  *	port number should listen() do a T_UNBIND_REQ followed by a
130  *	O_T_BIND_REQ.
131  *
132  * soconnect_tpi_udp:
133  *	UDP and ICMP can handle a T_CONN_REQ.
134  *	This is needed to make the sequence of connect(), getsockname()
135  *	return the local IP address used to send packets to the connected to
136  *	destination.
137  *
138  * soconnect_tpi_tcp:
139  *	TCP can handle a T_CONN_REQ without seeing a O_T_BIND_REQ.
140  *	Set this to non-zero to send TPI conformant messages to TCP in this
141  *	respect. This is a performance optimization.
142  *
143  * soaccept_tpi_tcp:
144  *	TCP can handle a T_CONN_REQ without the acceptor being bound.
145  *	This is a performance optimization that has been picked up in XTI.
146  *
147  * soaccept_tpi_multioptions:
148  *	When inheriting SOL_SOCKET options from the listener to the accepting
149  *	socket send them as a single message for AF_INET{,6}.
150  */
151 int solisten_tpi_tcp = 0;
152 int soconnect_tpi_udp = 0;
153 int soconnect_tpi_tcp = 0;
154 int soaccept_tpi_tcp = 0;
155 int soaccept_tpi_multioptions = 1;
156 #else /* SOCK_TEST */
157 #define	soconnect_tpi_tcp	0
158 #define	soconnect_tpi_udp	0
159 #define	solisten_tpi_tcp	0
160 #define	soaccept_tpi_tcp	0
161 #define	soaccept_tpi_multioptions	1
162 #endif /* SOCK_TEST */
163 
164 #ifdef SOCK_TEST
165 extern int do_useracc;
166 extern clock_t sock_test_timelimit;
167 #endif /* SOCK_TEST */
168 
169 /*
170  * Some X/Open added checks might have to be backed out to keep SunOS 4.X
171  * applications working. Turn on this flag to disable these checks.
172  */
173 int xnet_skip_checks = 0;
174 int xnet_check_print = 0;
175 int xnet_truncate_print = 0;
176 
177 static void sotpi_destroy(struct sonode *);
178 static struct sonode *sotpi_create(struct sockparams *, int, int, int, int,
179     int, int *, cred_t *cr);
180 
181 static boolean_t	sotpi_info_create(struct sonode *, int);
182 static void		sotpi_info_init(struct sonode *);
183 static void 		sotpi_info_fini(struct sonode *);
184 static void 		sotpi_info_destroy(struct sonode *);
185 
186 /*
187  * Do direct function call to the transport layer below; this would
188  * also allow the transport to utilize read-side synchronous stream
189  * interface if necessary.  This is a /etc/system tunable that must
190  * not be modified on a running system.  By default this is enabled
191  * for performance reasons and may be disabled for debugging purposes.
192  */
193 boolean_t socktpi_direct = B_TRUE;
194 
195 static struct kmem_cache *socktpi_cache, *socktpi_unix_cache;
196 
197 extern	void sigintr(k_sigset_t *, int);
198 extern	void sigunintr(k_sigset_t *);
199 
200 /* Sockets acting as an in-kernel SSL proxy */
201 extern mblk_t	*strsock_kssl_input(vnode_t *, mblk_t *, strwakeup_t *,
202 		    strsigset_t *, strsigset_t *, strpollset_t *);
203 extern mblk_t	*strsock_kssl_output(vnode_t *, mblk_t *, strwakeup_t *,
204 		    strsigset_t *, strsigset_t *, strpollset_t *);
205 
206 static int	sotpi_unbind(struct sonode *, int);
207 
208 /* TPI sockfs sonode operations */
209 int 		sotpi_init(struct sonode *, struct sonode *, struct cred *,
210 		    int);
211 static int	sotpi_accept(struct sonode *, int, struct cred *,
212 		    struct sonode **);
213 static int	sotpi_bind(struct sonode *, struct sockaddr *, socklen_t,
214 		    int, struct cred *);
215 static int	sotpi_listen(struct sonode *, int, struct cred *);
216 static int	sotpi_connect(struct sonode *, struct sockaddr *,
217 		    socklen_t, int, int, struct cred *);
218 extern int	sotpi_recvmsg(struct sonode *, struct nmsghdr *,
219 		    struct uio *, struct cred *);
220 static int	sotpi_sendmsg(struct sonode *, struct nmsghdr *,
221 		    struct uio *, struct cred *);
222 static int	sotpi_sendmblk(struct sonode *, struct nmsghdr *, int,
223 		    struct cred *, mblk_t **);
224 static int	sosend_dgramcmsg(struct sonode *, struct sockaddr *, socklen_t,
225 		    struct uio *, void *, t_uscalar_t, int);
226 static int	sodgram_direct(struct sonode *, struct sockaddr *,
227 		    socklen_t, struct uio *, int);
228 extern int	sotpi_getpeername(struct sonode *, struct sockaddr *,
229 		    socklen_t *, boolean_t, struct cred *);
230 static int	sotpi_getsockname(struct sonode *, struct sockaddr *,
231 		    socklen_t *, struct cred *);
232 static int	sotpi_shutdown(struct sonode *, int, struct cred *);
233 extern int	sotpi_getsockopt(struct sonode *, int, int, void *,
234 		    socklen_t *, int, struct cred *);
235 extern int	sotpi_setsockopt(struct sonode *, int, int, const void *,
236 		    socklen_t, struct cred *);
237 static int 	sotpi_ioctl(struct sonode *, int, intptr_t, int, struct cred *,
238 		    int32_t *);
239 static int 	socktpi_plumbioctl(struct vnode *, int, intptr_t, int,
240 		    struct cred *, int32_t *);
241 static int 	sotpi_poll(struct sonode *, short, int, short *,
242 		    struct pollhead **);
243 static int 	sotpi_close(struct sonode *, int, struct cred *);
244 
245 static int	i_sotpi_info_constructor(sotpi_info_t *);
246 static void 	i_sotpi_info_destructor(sotpi_info_t *);
247 
248 sonodeops_t sotpi_sonodeops = {
249 	sotpi_init,		/* sop_init		*/
250 	sotpi_accept,		/* sop_accept		*/
251 	sotpi_bind,		/* sop_bind		*/
252 	sotpi_listen,		/* sop_listen		*/
253 	sotpi_connect,		/* sop_connect		*/
254 	sotpi_recvmsg,		/* sop_recvmsg		*/
255 	sotpi_sendmsg,		/* sop_sendmsg		*/
256 	sotpi_sendmblk,		/* sop_sendmblk		*/
257 	sotpi_getpeername,	/* sop_getpeername	*/
258 	sotpi_getsockname,	/* sop_getsockname	*/
259 	sotpi_shutdown,		/* sop_shutdown		*/
260 	sotpi_getsockopt,	/* sop_getsockopt	*/
261 	sotpi_setsockopt,	/* sop_setsockopt	*/
262 	sotpi_ioctl,		/* sop_ioctl		*/
263 	sotpi_poll,		/* sop_poll		*/
264 	sotpi_close,		/* sop_close		*/
265 };
266 
267 /*
268  * Return a TPI socket vnode.
269  *
270  * Note that sockets assume that the driver will clone (either itself
271  * or by using the clone driver) i.e. a socket() call will always
272  * result in a new vnode being created.
273  */
274 
275 /*
276  * Common create code for socket and accept. If tso is set the values
277  * from that node is used instead of issuing a T_INFO_REQ.
278  */
279 
280 /* ARGSUSED */
281 static struct sonode *
282 sotpi_create(struct sockparams *sp, int family, int type, int protocol,
283     int version, int sflags, int *errorp, cred_t *cr)
284 {
285 	struct sonode	*so;
286 	kmem_cache_t 	*cp;
287 	int		sfamily = family;
288 
289 	ASSERT(sp->sp_sdev_info.sd_vnode != NULL);
290 
291 	if (family == AF_NCA) {
292 		/*
293 		 * The request is for an NCA socket so for NL7C use the
294 		 * INET domain instead and mark NL7C_AF_NCA below.
295 		 */
296 		family = AF_INET;
297 		/*
298 		 * NL7C is not supported in the non-global zone,
299 		 * we enforce this restriction here.
300 		 */
301 		if (getzoneid() != GLOBAL_ZONEID) {
302 			*errorp = ENOTSUP;
303 			return (NULL);
304 		}
305 	}
306 
307 	/*
308 	 * to be compatible with old tpi socket implementation ignore
309 	 * sleep flag (sflags) passed in
310 	 */
311 	cp = (family == AF_UNIX) ? socktpi_unix_cache : socktpi_cache;
312 	so = kmem_cache_alloc(cp, KM_SLEEP);
313 	if (so == NULL) {
314 		*errorp = ENOMEM;
315 		return (NULL);
316 	}
317 
318 	sonode_init(so, sp, family, type, protocol, &sotpi_sonodeops);
319 	sotpi_info_init(so);
320 
321 	if (sfamily == AF_NCA) {
322 		SOTOTPI(so)->sti_nl7c_flags = NL7C_AF_NCA;
323 	}
324 
325 	if (version == SOV_DEFAULT)
326 		version = so_default_version;
327 
328 	so->so_version = (short)version;
329 	*errorp = 0;
330 
331 	return (so);
332 }
333 
334 static void
335 sotpi_destroy(struct sonode *so)
336 {
337 	kmem_cache_t *cp;
338 	struct sockparams *origsp;
339 
340 	/*
341 	 * If there is a new dealloc function (ie. smod_destroy_func),
342 	 * then it should check the correctness of the ops.
343 	 */
344 
345 	ASSERT(so->so_ops == &sotpi_sonodeops);
346 
347 	origsp = SOTOTPI(so)->sti_orig_sp;
348 
349 	sotpi_info_fini(so);
350 
351 	if (so->so_state & SS_FALLBACK_COMP) {
352 		/*
353 		 * A fallback happend, which means that a sotpi_info_t struct
354 		 * was allocated (as opposed to being allocated from the TPI
355 		 * sonode cache. Therefore we explicitly free the struct
356 		 * here.
357 		 */
358 		sotpi_info_destroy(so);
359 		ASSERT(origsp != NULL);
360 
361 		origsp->sp_smod_info->smod_sock_destroy_func(so);
362 		SOCKPARAMS_DEC_REF(origsp);
363 	} else {
364 		sonode_fini(so);
365 		cp = (so->so_family == AF_UNIX) ? socktpi_unix_cache :
366 		    socktpi_cache;
367 		kmem_cache_free(cp, so);
368 	}
369 }
370 
371 /* ARGSUSED1 */
372 int
373 sotpi_init(struct sonode *so, struct sonode *tso, struct cred *cr, int flags)
374 {
375 	major_t maj;
376 	dev_t newdev;
377 	struct vnode *vp;
378 	int error = 0;
379 	struct stdata *stp;
380 
381 	sotpi_info_t *sti = SOTOTPI(so);
382 
383 	dprint(1, ("sotpi_init()\n"));
384 
385 	/*
386 	 * over write the sleep flag passed in but that is ok
387 	 * as tpi socket does not honor sleep flag.
388 	 */
389 	flags |= FREAD|FWRITE;
390 
391 	/*
392 	 * Record in so_flag that it is a clone.
393 	 */
394 	if (getmajor(sti->sti_dev) == clone_major)
395 		so->so_flag |= SOCLONE;
396 
397 	if ((so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM) &&
398 	    (so->so_family == AF_INET || so->so_family == AF_INET6) &&
399 	    (so->so_protocol == IPPROTO_TCP || so->so_protocol == IPPROTO_UDP ||
400 	    so->so_protocol == IPPROTO_IP)) {
401 		/* Tell tcp or udp that it's talking to sockets */
402 		flags |= SO_SOCKSTR;
403 
404 		/*
405 		 * Here we indicate to socktpi_open() our attempt to
406 		 * make direct calls between sockfs and transport.
407 		 * The final decision is left to socktpi_open().
408 		 */
409 		sti->sti_direct = 1;
410 
411 		ASSERT(so->so_type != SOCK_DGRAM || tso == NULL);
412 		if (so->so_type == SOCK_STREAM && tso != NULL) {
413 			if (SOTOTPI(tso)->sti_direct) {
414 				/*
415 				 * Inherit sti_direct from listener and pass
416 				 * SO_ACCEPTOR open flag to tcp, indicating
417 				 * that this is an accept fast-path instance.
418 				 */
419 				flags |= SO_ACCEPTOR;
420 			} else {
421 				/*
422 				 * sti_direct is not set on listener, meaning
423 				 * that the listener has been converted from
424 				 * a socket to a stream.  Ensure that the
425 				 * acceptor inherits these settings.
426 				 */
427 				sti->sti_direct = 0;
428 				flags &= ~SO_SOCKSTR;
429 			}
430 		}
431 	}
432 
433 	/*
434 	 * Tell local transport that it is talking to sockets.
435 	 */
436 	if (so->so_family == AF_UNIX) {
437 		flags |= SO_SOCKSTR;
438 	}
439 
440 	vp = SOTOV(so);
441 	newdev = vp->v_rdev;
442 	maj = getmajor(newdev);
443 	ASSERT(STREAMSTAB(maj));
444 
445 	error = stropen(vp, &newdev, flags, cr);
446 
447 	stp = vp->v_stream;
448 	if (error == 0) {
449 		if (so->so_flag & SOCLONE)
450 			ASSERT(newdev != vp->v_rdev);
451 		mutex_enter(&so->so_lock);
452 		sti->sti_dev = newdev;
453 		vp->v_rdev = newdev;
454 		mutex_exit(&so->so_lock);
455 
456 		if (stp->sd_flag & STRISTTY) {
457 			/*
458 			 * this is a post SVR4 tty driver - a socket can not
459 			 * be a controlling terminal. Fail the open.
460 			 */
461 			(void) sotpi_close(so, flags, cr);
462 			return (ENOTTY);	/* XXX */
463 		}
464 
465 		ASSERT(stp->sd_wrq != NULL);
466 		sti->sti_provinfo = tpi_findprov(stp->sd_wrq);
467 
468 		/*
469 		 * If caller is interested in doing direct function call
470 		 * interface to/from transport module, probe the module
471 		 * directly beneath the streamhead to see if it qualifies.
472 		 *
473 		 * We turn off the direct interface when qualifications fail.
474 		 * In the acceptor case, we simply turn off the sti_direct
475 		 * flag on the socket. We do the fallback after the accept
476 		 * has completed, before the new socket is returned to the
477 		 * application.
478 		 */
479 		if (sti->sti_direct) {
480 			queue_t *tq = stp->sd_wrq->q_next;
481 
482 			/*
483 			 * sti_direct is currently supported and tested
484 			 * only for tcp/udp; this is the main reason to
485 			 * have the following assertions.
486 			 */
487 			ASSERT(so->so_family == AF_INET ||
488 			    so->so_family == AF_INET6);
489 			ASSERT(so->so_protocol == IPPROTO_UDP ||
490 			    so->so_protocol == IPPROTO_TCP ||
491 			    so->so_protocol == IPPROTO_IP);
492 			ASSERT(so->so_type == SOCK_DGRAM ||
493 			    so->so_type == SOCK_STREAM);
494 
495 			/*
496 			 * Abort direct call interface if the module directly
497 			 * underneath the stream head is not defined with the
498 			 * _D_DIRECT flag.  This could happen in the tcp or
499 			 * udp case, when some other module is autopushed
500 			 * above it, or for some reasons the expected module
501 			 * isn't purely D_MP (which is the main requirement).
502 			 */
503 			if (!socktpi_direct || !(tq->q_flag & _QDIRECT) ||
504 			    !(_OTHERQ(tq)->q_flag & _QDIRECT)) {
505 				int rval;
506 
507 				/* Continue on without direct calls */
508 				sti->sti_direct = 0;
509 
510 				/*
511 				 * Cannot issue ioctl on fallback socket since
512 				 * there is no conn associated with the queue.
513 				 * The fallback downcall will notify the proto
514 				 * of the change.
515 				 */
516 				if (!(flags & SO_ACCEPTOR) &&
517 				    !(flags & SO_FALLBACK)) {
518 					if ((error = strioctl(vp,
519 					    _SIOCSOCKFALLBACK, 0, 0, K_TO_K,
520 					    cr, &rval)) != 0) {
521 						(void) sotpi_close(so, flags,
522 						    cr);
523 						return (error);
524 					}
525 				}
526 			}
527 		}
528 
529 		if (flags & SO_FALLBACK) {
530 			/*
531 			 * The stream created does not have a conn.
532 			 * do stream set up after conn has been assigned
533 			 */
534 			return (error);
535 		}
536 		if (error = so_strinit(so, tso)) {
537 			(void) sotpi_close(so, flags, cr);
538 			return (error);
539 		}
540 
541 		/* Wildcard */
542 		if (so->so_protocol != so->so_sockparams->sp_protocol) {
543 			int protocol = so->so_protocol;
544 			/*
545 			 * Issue SO_PROTOTYPE setsockopt.
546 			 */
547 			error = sotpi_setsockopt(so, SOL_SOCKET, SO_PROTOTYPE,
548 			    &protocol, (t_uscalar_t)sizeof (protocol), cr);
549 			if (error != 0) {
550 				(void) sotpi_close(so, flags, cr);
551 				/*
552 				 * Setsockopt often fails with ENOPROTOOPT but
553 				 * socket() should fail with
554 				 * EPROTONOSUPPORT/EPROTOTYPE.
555 				 */
556 				return (EPROTONOSUPPORT);
557 			}
558 		}
559 
560 	} else {
561 		/*
562 		 * While the same socket can not be reopened (unlike specfs)
563 		 * the stream head sets STREOPENFAIL when the autopush fails.
564 		 */
565 		if ((stp != NULL) &&
566 		    (stp->sd_flag & STREOPENFAIL)) {
567 			/*
568 			 * Open failed part way through.
569 			 */
570 			mutex_enter(&stp->sd_lock);
571 			stp->sd_flag &= ~STREOPENFAIL;
572 			mutex_exit(&stp->sd_lock);
573 			(void) sotpi_close(so, flags, cr);
574 			return (error);
575 			/*NOTREACHED*/
576 		}
577 		ASSERT(stp == NULL);
578 	}
579 	TRACE_4(TR_FAC_SOCKFS, TR_SOCKFS_OPEN,
580 	    "sockfs open:maj %d vp %p so %p error %d",
581 	    maj, vp, so, error);
582 	return (error);
583 }
584 
585 /*
586  * Bind the socket to an unspecified address in sockfs only.
587  * Used for TCP/UDP transports where we know that the O_T_BIND_REQ isn't
588  * required in all cases.
589  */
590 static void
591 so_automatic_bind(struct sonode *so)
592 {
593 	sotpi_info_t *sti = SOTOTPI(so);
594 	ASSERT(so->so_family == AF_INET || so->so_family == AF_INET6);
595 
596 	ASSERT(MUTEX_HELD(&so->so_lock));
597 	ASSERT(!(so->so_state & SS_ISBOUND));
598 	ASSERT(sti->sti_unbind_mp);
599 
600 	ASSERT(sti->sti_laddr_len <= sti->sti_laddr_maxlen);
601 	bzero(sti->sti_laddr_sa, sti->sti_laddr_len);
602 	sti->sti_laddr_sa->sa_family = so->so_family;
603 	so->so_state |= SS_ISBOUND;
604 }
605 
606 
607 /*
608  * bind the socket.
609  *
610  * If the socket is already bound and none of _SOBIND_SOCKBSD or _SOBIND_XPG4_2
611  * are passed in we allow rebinding. Note that for backwards compatibility
612  * even "svr4" sockets pass in _SOBIND_SOCKBSD/SOV_SOCKBSD to sobind/bind.
613  * Thus the rebinding code is currently not executed.
614  *
615  * The constraints for rebinding are:
616  * - it is a SOCK_DGRAM, or
617  * - it is a SOCK_STREAM/SOCK_SEQPACKET that has not been connected
618  *   and no listen() has been done.
619  * This rebinding code was added based on some language in the XNET book
620  * about not returning EINVAL it the protocol allows rebinding. However,
621  * this language is not present in the Posix socket draft. Thus maybe the
622  * rebinding logic should be deleted from the source.
623  *
624  * A null "name" can be used to unbind the socket if:
625  * - it is a SOCK_DGRAM, or
626  * - it is a SOCK_STREAM/SOCK_SEQPACKET that has not been connected
627  *   and no listen() has been done.
628  */
629 /* ARGSUSED */
630 static int
631 sotpi_bindlisten(struct sonode *so, struct sockaddr *name,
632     socklen_t namelen, int backlog, int flags, struct cred *cr)
633 {
634 	struct T_bind_req	bind_req;
635 	struct T_bind_ack	*bind_ack;
636 	int			error = 0;
637 	mblk_t			*mp;
638 	void			*addr;
639 	t_uscalar_t		addrlen;
640 	int			unbind_on_err = 1;
641 	boolean_t		clear_acceptconn_on_err = B_FALSE;
642 	boolean_t		restore_backlog_on_err = B_FALSE;
643 	int			save_so_backlog;
644 	t_scalar_t		PRIM_type = O_T_BIND_REQ;
645 	boolean_t		tcp_udp_xport;
646 	void			*nl7c = NULL;
647 	sotpi_info_t		*sti = SOTOTPI(so);
648 
649 	dprintso(so, 1, ("sotpi_bindlisten(%p, %p, %d, %d, 0x%x) %s\n",
650 	    (void *)so, (void *)name, namelen, backlog, flags,
651 	    pr_state(so->so_state, so->so_mode)));
652 
653 	tcp_udp_xport = so->so_type == SOCK_STREAM || so->so_type == SOCK_DGRAM;
654 
655 	if (!(flags & _SOBIND_LOCK_HELD)) {
656 		mutex_enter(&so->so_lock);
657 		so_lock_single(so);	/* Set SOLOCKED */
658 	} else {
659 		ASSERT(MUTEX_HELD(&so->so_lock));
660 		ASSERT(so->so_flag & SOLOCKED);
661 	}
662 
663 	/*
664 	 * Make sure that there is a preallocated unbind_req message
665 	 * before binding. This message allocated when the socket is
666 	 * created  but it might be have been consumed.
667 	 */
668 	if (sti->sti_unbind_mp == NULL) {
669 		dprintso(so, 1, ("sobind: allocating unbind_req\n"));
670 		/* NOTE: holding so_lock while sleeping */
671 		sti->sti_unbind_mp =
672 		    soallocproto(sizeof (struct T_unbind_req), _ALLOC_SLEEP,
673 		    cr);
674 	}
675 
676 	if (flags & _SOBIND_REBIND) {
677 		/*
678 		 * Called from solisten after doing an sotpi_unbind() or
679 		 * potentially without the unbind (latter for AF_INET{,6}).
680 		 */
681 		ASSERT(name == NULL && namelen == 0);
682 
683 		if (so->so_family == AF_UNIX) {
684 			ASSERT(sti->sti_ux_bound_vp);
685 			addr = &sti->sti_ux_laddr;
686 			addrlen = (t_uscalar_t)sizeof (sti->sti_ux_laddr);
687 			dprintso(so, 1, ("sobind rebind UNIX: addrlen %d, "
688 			    "addr 0x%p, vp %p\n",
689 			    addrlen,
690 			    (void *)((struct so_ux_addr *)addr)->soua_vp,
691 			    (void *)sti->sti_ux_bound_vp));
692 		} else {
693 			addr = sti->sti_laddr_sa;
694 			addrlen = (t_uscalar_t)sti->sti_laddr_len;
695 		}
696 	} else if (flags & _SOBIND_UNSPEC) {
697 		ASSERT(name == NULL && namelen == 0);
698 
699 		/*
700 		 * The caller checked SS_ISBOUND but not necessarily
701 		 * under so_lock
702 		 */
703 		if (so->so_state & SS_ISBOUND) {
704 			/* No error */
705 			goto done;
706 		}
707 
708 		/* Set an initial local address */
709 		switch (so->so_family) {
710 		case AF_UNIX:
711 			/*
712 			 * Use an address with same size as struct sockaddr
713 			 * just like BSD.
714 			 */
715 			sti->sti_laddr_len =
716 			    (socklen_t)sizeof (struct sockaddr);
717 			ASSERT(sti->sti_laddr_len <= sti->sti_laddr_maxlen);
718 			bzero(sti->sti_laddr_sa, sti->sti_laddr_len);
719 			sti->sti_laddr_sa->sa_family = so->so_family;
720 
721 			/*
722 			 * Pass down an address with the implicit bind
723 			 * magic number and the rest all zeros.
724 			 * The transport will return a unique address.
725 			 */
726 			sti->sti_ux_laddr.soua_vp = NULL;
727 			sti->sti_ux_laddr.soua_magic = SOU_MAGIC_IMPLICIT;
728 			addr = &sti->sti_ux_laddr;
729 			addrlen = (t_uscalar_t)sizeof (sti->sti_ux_laddr);
730 			break;
731 
732 		case AF_INET:
733 		case AF_INET6:
734 			/*
735 			 * An unspecified bind in TPI has a NULL address.
736 			 * Set the address in sockfs to have the sa_family.
737 			 */
738 			sti->sti_laddr_len = (so->so_family == AF_INET) ?
739 			    (socklen_t)sizeof (sin_t) :
740 			    (socklen_t)sizeof (sin6_t);
741 			ASSERT(sti->sti_laddr_len <= sti->sti_laddr_maxlen);
742 			bzero(sti->sti_laddr_sa, sti->sti_laddr_len);
743 			sti->sti_laddr_sa->sa_family = so->so_family;
744 			addr = NULL;
745 			addrlen = 0;
746 			break;
747 
748 		default:
749 			/*
750 			 * An unspecified bind in TPI has a NULL address.
751 			 * Set the address in sockfs to be zero length.
752 			 *
753 			 * Can not assume there is a sa_family for all
754 			 * protocol families. For example, AF_X25 does not
755 			 * have a family field.
756 			 */
757 			bzero(sti->sti_laddr_sa, sti->sti_laddr_len);
758 			sti->sti_laddr_len = 0;	/* XXX correct? */
759 			addr = NULL;
760 			addrlen = 0;
761 			break;
762 		}
763 
764 	} else {
765 		if (so->so_state & SS_ISBOUND) {
766 			/*
767 			 * If it is ok to rebind the socket, first unbind
768 			 * with the transport. A rebind to the NULL address
769 			 * is interpreted as an unbind.
770 			 * Note that a bind to NULL in BSD does unbind the
771 			 * socket but it fails with EINVAL.
772 			 * Note that regular sockets set SOV_SOCKBSD i.e.
773 			 * _SOBIND_SOCKBSD gets set here hence no type of
774 			 * socket does currently allow rebinding.
775 			 *
776 			 * If the name is NULL just do an unbind.
777 			 */
778 			if (flags & (_SOBIND_SOCKBSD|_SOBIND_XPG4_2) &&
779 			    name != NULL) {
780 				error = EINVAL;
781 				unbind_on_err = 0;
782 				eprintsoline(so, error);
783 				goto done;
784 			}
785 			if ((so->so_mode & SM_CONNREQUIRED) &&
786 			    (so->so_state & SS_CANTREBIND)) {
787 				error = EINVAL;
788 				unbind_on_err = 0;
789 				eprintsoline(so, error);
790 				goto done;
791 			}
792 			error = sotpi_unbind(so, 0);
793 			if (error) {
794 				eprintsoline(so, error);
795 				goto done;
796 			}
797 			ASSERT(!(so->so_state & SS_ISBOUND));
798 			if (name == NULL) {
799 				so->so_state &=
800 				    ~(SS_ISCONNECTED|SS_ISCONNECTING);
801 				goto done;
802 			}
803 		}
804 
805 		/* X/Open requires this check */
806 		if ((so->so_state & SS_CANTSENDMORE) && !xnet_skip_checks) {
807 			if (xnet_check_print) {
808 				printf("sockfs: X/Open bind state check "
809 				    "caused EINVAL\n");
810 			}
811 			error = EINVAL;
812 			goto done;
813 		}
814 
815 		switch (so->so_family) {
816 		case AF_UNIX:
817 			/*
818 			 * All AF_UNIX addresses are nul terminated
819 			 * when copied (copyin_name) in so the minimum
820 			 * length is 3 bytes.
821 			 */
822 			if (name == NULL ||
823 			    (ssize_t)namelen <= sizeof (short) + 1) {
824 				error = EISDIR;
825 				eprintsoline(so, error);
826 				goto done;
827 			}
828 			/*
829 			 * Verify so_family matches the bound family.
830 			 * BSD does not check this for AF_UNIX resulting
831 			 * in funny mknods.
832 			 */
833 			if (name->sa_family != so->so_family) {
834 				error = EAFNOSUPPORT;
835 				goto done;
836 			}
837 			break;
838 		case AF_INET:
839 			if (name == NULL) {
840 				error = EINVAL;
841 				eprintsoline(so, error);
842 				goto done;
843 			}
844 			if ((size_t)namelen != sizeof (sin_t)) {
845 				error = name->sa_family != so->so_family ?
846 				    EAFNOSUPPORT : EINVAL;
847 				eprintsoline(so, error);
848 				goto done;
849 			}
850 			if ((flags & _SOBIND_XPG4_2) &&
851 			    (name->sa_family != so->so_family)) {
852 				/*
853 				 * This check has to be made for X/Open
854 				 * sockets however application failures have
855 				 * been observed when it is applied to
856 				 * all sockets.
857 				 */
858 				error = EAFNOSUPPORT;
859 				eprintsoline(so, error);
860 				goto done;
861 			}
862 			/*
863 			 * Force a zero sa_family to match so_family.
864 			 *
865 			 * Some programs like inetd(1M) don't set the
866 			 * family field. Other programs leave
867 			 * sin_family set to garbage - SunOS 4.X does
868 			 * not check the family field on a bind.
869 			 * We use the family field that
870 			 * was passed in to the socket() call.
871 			 */
872 			name->sa_family = so->so_family;
873 			break;
874 
875 		case AF_INET6: {
876 #ifdef DEBUG
877 			sin6_t *sin6 = (sin6_t *)name;
878 #endif /* DEBUG */
879 
880 			if (name == NULL) {
881 				error = EINVAL;
882 				eprintsoline(so, error);
883 				goto done;
884 			}
885 			if ((size_t)namelen != sizeof (sin6_t)) {
886 				error = name->sa_family != so->so_family ?
887 				    EAFNOSUPPORT : EINVAL;
888 				eprintsoline(so, error);
889 				goto done;
890 			}
891 			if (name->sa_family != so->so_family) {
892 				/*
893 				 * With IPv6 we require the family to match
894 				 * unlike in IPv4.
895 				 */
896 				error = EAFNOSUPPORT;
897 				eprintsoline(so, error);
898 				goto done;
899 			}
900 #ifdef DEBUG
901 			/*
902 			 * Verify that apps don't forget to clear
903 			 * sin6_scope_id etc
904 			 */
905 			if (sin6->sin6_scope_id != 0 &&
906 			    !IN6_IS_ADDR_LINKSCOPE(&sin6->sin6_addr)) {
907 				zcmn_err(getzoneid(), CE_WARN,
908 				    "bind with uninitialized sin6_scope_id "
909 				    "(%d) on socket. Pid = %d\n",
910 				    (int)sin6->sin6_scope_id,
911 				    (int)curproc->p_pid);
912 			}
913 			if (sin6->__sin6_src_id != 0) {
914 				zcmn_err(getzoneid(), CE_WARN,
915 				    "bind with uninitialized __sin6_src_id "
916 				    "(%d) on socket. Pid = %d\n",
917 				    (int)sin6->__sin6_src_id,
918 				    (int)curproc->p_pid);
919 			}
920 #endif /* DEBUG */
921 			break;
922 		}
923 		default:
924 			/*
925 			 * Don't do any length or sa_family check to allow
926 			 * non-sockaddr style addresses.
927 			 */
928 			if (name == NULL) {
929 				error = EINVAL;
930 				eprintsoline(so, error);
931 				goto done;
932 			}
933 			break;
934 		}
935 
936 		if (namelen > (t_uscalar_t)sti->sti_laddr_maxlen) {
937 			error = ENAMETOOLONG;
938 			eprintsoline(so, error);
939 			goto done;
940 		}
941 		/*
942 		 * Save local address.
943 		 */
944 		sti->sti_laddr_len = (socklen_t)namelen;
945 		ASSERT(sti->sti_laddr_len <= sti->sti_laddr_maxlen);
946 		bcopy(name, sti->sti_laddr_sa, namelen);
947 
948 		addr = sti->sti_laddr_sa;
949 		addrlen = (t_uscalar_t)sti->sti_laddr_len;
950 		switch (so->so_family) {
951 		case AF_INET6:
952 		case AF_INET:
953 			break;
954 		case AF_UNIX: {
955 			struct sockaddr_un *soun =
956 			    (struct sockaddr_un *)sti->sti_laddr_sa;
957 			struct vnode *vp, *rvp;
958 			struct vattr vattr;
959 
960 			ASSERT(sti->sti_ux_bound_vp == NULL);
961 			/*
962 			 * Create vnode for the specified path name.
963 			 * Keep vnode held with a reference in sti_ux_bound_vp.
964 			 * Use the vnode pointer as the address used in the
965 			 * bind with the transport.
966 			 *
967 			 * Use the same mode as in BSD. In particular this does
968 			 * not observe the umask.
969 			 */
970 			/* MAXPATHLEN + soun_family + nul termination */
971 			if (sti->sti_laddr_len >
972 			    (socklen_t)(MAXPATHLEN + sizeof (short) + 1)) {
973 				error = ENAMETOOLONG;
974 				eprintsoline(so, error);
975 				goto done;
976 			}
977 			vattr.va_type = VSOCK;
978 			vattr.va_mode = 0777 & ~PTOU(curproc)->u_cmask;
979 			vattr.va_mask = AT_TYPE|AT_MODE;
980 			/* NOTE: holding so_lock */
981 			error = vn_create(soun->sun_path, UIO_SYSSPACE, &vattr,
982 			    EXCL, 0, &vp, CRMKNOD, 0, 0);
983 			if (error) {
984 				if (error == EEXIST)
985 					error = EADDRINUSE;
986 				eprintsoline(so, error);
987 				goto done;
988 			}
989 			/*
990 			 * Establish pointer from the underlying filesystem
991 			 * vnode to the socket node.
992 			 * sti_ux_bound_vp and v_stream->sd_vnode form the
993 			 * cross-linkage between the underlying filesystem
994 			 * node and the socket node.
995 			 */
996 
997 			if ((VOP_REALVP(vp, &rvp, NULL) == 0) && (vp != rvp)) {
998 				VN_HOLD(rvp);
999 				VN_RELE(vp);
1000 				vp = rvp;
1001 			}
1002 
1003 			ASSERT(SOTOV(so)->v_stream);
1004 			mutex_enter(&vp->v_lock);
1005 			vp->v_stream = SOTOV(so)->v_stream;
1006 			sti->sti_ux_bound_vp = vp;
1007 			mutex_exit(&vp->v_lock);
1008 
1009 			/*
1010 			 * Use the vnode pointer value as a unique address
1011 			 * (together with the magic number to avoid conflicts
1012 			 * with implicit binds) in the transport provider.
1013 			 */
1014 			sti->sti_ux_laddr.soua_vp =
1015 			    (void *)sti->sti_ux_bound_vp;
1016 			sti->sti_ux_laddr.soua_magic = SOU_MAGIC_EXPLICIT;
1017 			addr = &sti->sti_ux_laddr;
1018 			addrlen = (t_uscalar_t)sizeof (sti->sti_ux_laddr);
1019 			dprintso(so, 1, ("sobind UNIX: addrlen %d, addr %p\n",
1020 			    addrlen,
1021 			    (void *)((struct so_ux_addr *)addr)->soua_vp));
1022 			break;
1023 		}
1024 		} /* end switch (so->so_family) */
1025 	}
1026 
1027 	/*
1028 	 * set SS_ACCEPTCONN before sending down O_T_BIND_REQ since
1029 	 * the transport can start passing up T_CONN_IND messages
1030 	 * as soon as it receives the bind req and strsock_proto()
1031 	 * insists that SS_ACCEPTCONN is set when processing T_CONN_INDs.
1032 	 */
1033 	if (flags & _SOBIND_LISTEN) {
1034 		if ((so->so_state & SS_ACCEPTCONN) == 0)
1035 			clear_acceptconn_on_err = B_TRUE;
1036 		save_so_backlog = so->so_backlog;
1037 		restore_backlog_on_err = B_TRUE;
1038 		so->so_state |= SS_ACCEPTCONN;
1039 		so->so_backlog = backlog;
1040 	}
1041 
1042 	/*
1043 	 * If NL7C addr(s) have been configured check for addr/port match,
1044 	 * or if an implicit NL7C socket via AF_NCA mark socket as NL7C.
1045 	 *
1046 	 * NL7C supports the TCP transport only so check AF_INET and AF_INET6
1047 	 * family sockets only. If match mark as such.
1048 	 */
1049 	if (nl7c_enabled && ((addr != NULL &&
1050 	    (so->so_family == AF_INET || so->so_family == AF_INET6) &&
1051 	    (nl7c = nl7c_lookup_addr(addr, addrlen))) ||
1052 	    sti->sti_nl7c_flags == NL7C_AF_NCA)) {
1053 		/*
1054 		 * NL7C is not supported in non-global zones,
1055 		 * we enforce this restriction here.
1056 		 */
1057 		if (so->so_zoneid == GLOBAL_ZONEID) {
1058 			/* An NL7C socket, mark it */
1059 			sti->sti_nl7c_flags |= NL7C_ENABLED;
1060 			if (nl7c == NULL) {
1061 				/*
1062 				 * Was an AF_NCA bind() so add it to the
1063 				 * addr list for reporting purposes.
1064 				 */
1065 				nl7c = nl7c_add_addr(addr, addrlen);
1066 			}
1067 		} else
1068 			nl7c = NULL;
1069 	}
1070 
1071 	/*
1072 	 * We send a T_BIND_REQ for TCP/UDP since we know it supports it,
1073 	 * for other transports we will send in a O_T_BIND_REQ.
1074 	 */
1075 	if (tcp_udp_xport &&
1076 	    (so->so_family == AF_INET || so->so_family == AF_INET6))
1077 		PRIM_type = T_BIND_REQ;
1078 
1079 	bind_req.PRIM_type = PRIM_type;
1080 	bind_req.ADDR_length = addrlen;
1081 	bind_req.ADDR_offset = (t_scalar_t)sizeof (bind_req);
1082 	bind_req.CONIND_number = backlog;
1083 	/* NOTE: holding so_lock while sleeping */
1084 	mp = soallocproto2(&bind_req, sizeof (bind_req),
1085 	    addr, addrlen, 0, _ALLOC_SLEEP, cr);
1086 	sti->sti_laddr_valid = 0;
1087 
1088 	/* Done using sti_laddr_sa - can drop the lock */
1089 	mutex_exit(&so->so_lock);
1090 
1091 	/*
1092 	 * Intercept the bind_req message here to check if this <address/port>
1093 	 * was configured as an SSL proxy server, or if another endpoint was
1094 	 * already configured to act as a proxy for us.
1095 	 *
1096 	 * Note, only if NL7C not enabled for this socket.
1097 	 */
1098 	if (nl7c == NULL &&
1099 	    (so->so_family == AF_INET || so->so_family == AF_INET6) &&
1100 	    so->so_type == SOCK_STREAM) {
1101 
1102 		if (sti->sti_kssl_ent != NULL) {
1103 			kssl_release_ent(sti->sti_kssl_ent, so,
1104 			    sti->sti_kssl_type);
1105 			sti->sti_kssl_ent = NULL;
1106 		}
1107 
1108 		sti->sti_kssl_type = kssl_check_proxy(mp, so,
1109 		    &sti->sti_kssl_ent);
1110 		switch (sti->sti_kssl_type) {
1111 		case KSSL_NO_PROXY:
1112 			break;
1113 
1114 		case KSSL_HAS_PROXY:
1115 			mutex_enter(&so->so_lock);
1116 			goto skip_transport;
1117 
1118 		case KSSL_IS_PROXY:
1119 			break;
1120 		}
1121 	}
1122 
1123 	error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0,
1124 	    MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0);
1125 	if (error) {
1126 		eprintsoline(so, error);
1127 		mutex_enter(&so->so_lock);
1128 		goto done;
1129 	}
1130 
1131 	mutex_enter(&so->so_lock);
1132 	error = sowaitprim(so, PRIM_type, T_BIND_ACK,
1133 	    (t_uscalar_t)sizeof (*bind_ack), &mp, 0);
1134 	if (error) {
1135 		eprintsoline(so, error);
1136 		goto done;
1137 	}
1138 skip_transport:
1139 	ASSERT(mp);
1140 	/*
1141 	 * Even if some TPI message (e.g. T_DISCON_IND) was received in
1142 	 * strsock_proto while the lock was dropped above, the bind
1143 	 * is allowed to complete.
1144 	 */
1145 
1146 	/* Mark as bound. This will be undone if we detect errors below. */
1147 	if (flags & _SOBIND_NOXLATE) {
1148 		ASSERT(so->so_family == AF_UNIX);
1149 		sti->sti_faddr_noxlate = 1;
1150 	}
1151 	ASSERT(!(so->so_state & SS_ISBOUND) || (flags & _SOBIND_REBIND));
1152 	so->so_state |= SS_ISBOUND;
1153 	ASSERT(sti->sti_unbind_mp);
1154 
1155 	/* note that we've already set SS_ACCEPTCONN above */
1156 
1157 	/*
1158 	 * Recompute addrlen - an unspecied bind sent down an
1159 	 * address of length zero but we expect the appropriate length
1160 	 * in return.
1161 	 */
1162 	addrlen = (t_uscalar_t)(so->so_family == AF_UNIX ?
1163 	    sizeof (sti->sti_ux_laddr) : sti->sti_laddr_len);
1164 
1165 	bind_ack = (struct T_bind_ack *)mp->b_rptr;
1166 	/*
1167 	 * The alignment restriction is really too strict but
1168 	 * we want enough alignment to inspect the fields of
1169 	 * a sockaddr_in.
1170 	 */
1171 	addr = sogetoff(mp, bind_ack->ADDR_offset,
1172 	    bind_ack->ADDR_length,
1173 	    __TPI_ALIGN_SIZE);
1174 	if (addr == NULL) {
1175 		freemsg(mp);
1176 		error = EPROTO;
1177 		eprintsoline(so, error);
1178 		goto done;
1179 	}
1180 	if (!(flags & _SOBIND_UNSPEC)) {
1181 		/*
1182 		 * Verify that the transport didn't return something we
1183 		 * did not want e.g. an address other than what we asked for.
1184 		 *
1185 		 * NOTE: These checks would go away if/when we switch to
1186 		 * using the new TPI (in which the transport would fail
1187 		 * the request instead of assigning a different address).
1188 		 *
1189 		 * NOTE2: For protocols that we don't know (i.e. any
1190 		 * other than AF_INET6, AF_INET and AF_UNIX), we
1191 		 * cannot know if the transport should be expected to
1192 		 * return the same address as that requested.
1193 		 *
1194 		 * NOTE3: For AF_INET and AF_INET6, TCP/UDP, we send
1195 		 * down a T_BIND_REQ. We use O_T_BIND_REQ for others.
1196 		 *
1197 		 * For example, in the case of netatalk it may be
1198 		 * inappropriate for the transport to return the
1199 		 * requested address (as it may have allocated a local
1200 		 * port number in behaviour similar to that of an
1201 		 * AF_INET bind request with a port number of zero).
1202 		 *
1203 		 * Given the definition of O_T_BIND_REQ, where the
1204 		 * transport may bind to an address other than the
1205 		 * requested address, it's not possible to determine
1206 		 * whether a returned address that differs from the
1207 		 * requested address is a reason to fail (because the
1208 		 * requested address was not available) or succeed
1209 		 * (because the transport allocated an appropriate
1210 		 * address and/or port).
1211 		 *
1212 		 * sockfs currently requires that the transport return
1213 		 * the requested address in the T_BIND_ACK, unless
1214 		 * there is code here to allow for any discrepancy.
1215 		 * Such code exists for AF_INET and AF_INET6.
1216 		 *
1217 		 * Netatalk chooses to return the requested address
1218 		 * rather than the (correct) allocated address.  This
1219 		 * means that netatalk violates the TPI specification
1220 		 * (and would not function correctly if used from a
1221 		 * TLI application), but it does mean that it works
1222 		 * with sockfs.
1223 		 *
1224 		 * As noted above, using the newer XTI bind primitive
1225 		 * (T_BIND_REQ) in preference to O_T_BIND_REQ would
1226 		 * allow sockfs to be more sure about whether or not
1227 		 * the bind request had succeeded (as transports are
1228 		 * not permitted to bind to a different address than
1229 		 * that requested - they must return failure).
1230 		 * Unfortunately, support for T_BIND_REQ may not be
1231 		 * present in all transport implementations (netatalk,
1232 		 * for example, doesn't have it), making the
1233 		 * transition difficult.
1234 		 */
1235 		if (bind_ack->ADDR_length != addrlen) {
1236 			/* Assumes that the requested address was in use */
1237 			freemsg(mp);
1238 			error = EADDRINUSE;
1239 			eprintsoline(so, error);
1240 			goto done;
1241 		}
1242 
1243 		switch (so->so_family) {
1244 		case AF_INET6:
1245 		case AF_INET: {
1246 			sin_t *rname, *aname;
1247 
1248 			rname = (sin_t *)addr;
1249 			aname = (sin_t *)sti->sti_laddr_sa;
1250 
1251 			/*
1252 			 * Take advantage of the alignment
1253 			 * of sin_port and sin6_port which fall
1254 			 * in the same place in their data structures.
1255 			 * Just use sin_port for either address family.
1256 			 *
1257 			 * This may become a problem if (heaven forbid)
1258 			 * there's a separate ipv6port_reserved... :-P
1259 			 *
1260 			 * Binding to port 0 has the semantics of letting
1261 			 * the transport bind to any port.
1262 			 *
1263 			 * If the transport is TCP or UDP since we had sent
1264 			 * a T_BIND_REQ we would not get a port other than
1265 			 * what we asked for.
1266 			 */
1267 			if (tcp_udp_xport) {
1268 				/*
1269 				 * Pick up the new port number if we bound to
1270 				 * port 0.
1271 				 */
1272 				if (aname->sin_port == 0)
1273 					aname->sin_port = rname->sin_port;
1274 				sti->sti_laddr_valid = 1;
1275 				break;
1276 			}
1277 			if (aname->sin_port != 0 &&
1278 			    aname->sin_port != rname->sin_port) {
1279 				freemsg(mp);
1280 				error = EADDRINUSE;
1281 				eprintsoline(so, error);
1282 				goto done;
1283 			}
1284 			/*
1285 			 * Pick up the new port number if we bound to port 0.
1286 			 */
1287 			aname->sin_port = rname->sin_port;
1288 
1289 			/*
1290 			 * Unfortunately, addresses aren't _quite_ the same.
1291 			 */
1292 			if (so->so_family == AF_INET) {
1293 				if (aname->sin_addr.s_addr !=
1294 				    rname->sin_addr.s_addr) {
1295 					freemsg(mp);
1296 					error = EADDRNOTAVAIL;
1297 					eprintsoline(so, error);
1298 					goto done;
1299 				}
1300 			} else {
1301 				sin6_t *rname6 = (sin6_t *)rname;
1302 				sin6_t *aname6 = (sin6_t *)aname;
1303 
1304 				if (!IN6_ARE_ADDR_EQUAL(&aname6->sin6_addr,
1305 				    &rname6->sin6_addr)) {
1306 					freemsg(mp);
1307 					error = EADDRNOTAVAIL;
1308 					eprintsoline(so, error);
1309 					goto done;
1310 				}
1311 			}
1312 			break;
1313 		}
1314 		case AF_UNIX:
1315 			if (bcmp(addr, &sti->sti_ux_laddr, addrlen) != 0) {
1316 				freemsg(mp);
1317 				error = EADDRINUSE;
1318 				eprintsoline(so, error);
1319 				eprintso(so,
1320 				    ("addrlen %d, addr 0x%x, vp %p\n",
1321 				    addrlen, *((int *)addr),
1322 				    (void *)sti->sti_ux_bound_vp));
1323 				goto done;
1324 			}
1325 			sti->sti_laddr_valid = 1;
1326 			break;
1327 		default:
1328 			/*
1329 			 * NOTE: This assumes that addresses can be
1330 			 * byte-compared for equivalence.
1331 			 */
1332 			if (bcmp(addr, sti->sti_laddr_sa, addrlen) != 0) {
1333 				freemsg(mp);
1334 				error = EADDRINUSE;
1335 				eprintsoline(so, error);
1336 				goto done;
1337 			}
1338 			/*
1339 			 * Don't mark sti_laddr_valid, as we cannot be
1340 			 * sure that the returned address is the real
1341 			 * bound address when talking to an unknown
1342 			 * transport.
1343 			 */
1344 			break;
1345 		}
1346 	} else {
1347 		/*
1348 		 * Save for returned address for getsockname.
1349 		 * Needed for unspecific bind unless transport supports
1350 		 * the TI_GETMYNAME ioctl.
1351 		 * Do this for AF_INET{,6} even though they do, as
1352 		 * caching info here is much better performance than
1353 		 * a TPI/STREAMS trip to the transport for getsockname.
1354 		 * Any which can't for some reason _must_ _not_ set
1355 		 * sti_laddr_valid here for the caching version of
1356 		 * getsockname to not break;
1357 		 */
1358 		switch (so->so_family) {
1359 		case AF_UNIX:
1360 			/*
1361 			 * Record the address bound with the transport
1362 			 * for use by socketpair.
1363 			 */
1364 			bcopy(addr, &sti->sti_ux_laddr, addrlen);
1365 			sti->sti_laddr_valid = 1;
1366 			break;
1367 		case AF_INET:
1368 		case AF_INET6:
1369 			ASSERT(sti->sti_laddr_len <= sti->sti_laddr_maxlen);
1370 			bcopy(addr, sti->sti_laddr_sa, sti->sti_laddr_len);
1371 			sti->sti_laddr_valid = 1;
1372 			break;
1373 		default:
1374 			/*
1375 			 * Don't mark sti_laddr_valid, as we cannot be
1376 			 * sure that the returned address is the real
1377 			 * bound address when talking to an unknown
1378 			 * transport.
1379 			 */
1380 			break;
1381 		}
1382 	}
1383 
1384 	if (nl7c != NULL) {
1385 		/* Register listen()er sonode pointer with NL7C */
1386 		nl7c_listener_addr(nl7c, so);
1387 	}
1388 
1389 	freemsg(mp);
1390 
1391 done:
1392 	if (error) {
1393 		/* reset state & backlog to values held on entry */
1394 		if (clear_acceptconn_on_err == B_TRUE)
1395 			so->so_state &= ~SS_ACCEPTCONN;
1396 		if (restore_backlog_on_err == B_TRUE)
1397 			so->so_backlog = save_so_backlog;
1398 
1399 		if (unbind_on_err && so->so_state & SS_ISBOUND) {
1400 			int err;
1401 
1402 			err = sotpi_unbind(so, 0);
1403 			/* LINTED - statement has no consequent: if */
1404 			if (err) {
1405 				eprintsoline(so, error);
1406 			} else {
1407 				ASSERT(!(so->so_state & SS_ISBOUND));
1408 			}
1409 		}
1410 	}
1411 	if (!(flags & _SOBIND_LOCK_HELD)) {
1412 		so_unlock_single(so, SOLOCKED);
1413 		mutex_exit(&so->so_lock);
1414 	} else {
1415 		ASSERT(MUTEX_HELD(&so->so_lock));
1416 		ASSERT(so->so_flag & SOLOCKED);
1417 	}
1418 	return (error);
1419 }
1420 
1421 /* bind the socket */
1422 static int
1423 sotpi_bind(struct sonode *so, struct sockaddr *name, socklen_t namelen,
1424     int flags, struct cred *cr)
1425 {
1426 	if ((flags & _SOBIND_SOCKETPAIR) == 0)
1427 		return (sotpi_bindlisten(so, name, namelen, 0, flags, cr));
1428 
1429 	flags &= ~_SOBIND_SOCKETPAIR;
1430 	return (sotpi_bindlisten(so, name, namelen, 1, flags, cr));
1431 }
1432 
1433 /*
1434  * Unbind a socket - used when bind() fails, when bind() specifies a NULL
1435  * address, or when listen needs to unbind and bind.
1436  * If the _SOUNBIND_REBIND flag is specified the addresses are retained
1437  * so that a sobind can pick them up.
1438  */
1439 static int
1440 sotpi_unbind(struct sonode *so, int flags)
1441 {
1442 	struct T_unbind_req	unbind_req;
1443 	int			error = 0;
1444 	mblk_t			*mp;
1445 	sotpi_info_t		*sti = SOTOTPI(so);
1446 
1447 	dprintso(so, 1, ("sotpi_unbind(%p, 0x%x) %s\n",
1448 	    (void *)so, flags, pr_state(so->so_state, so->so_mode)));
1449 
1450 	ASSERT(MUTEX_HELD(&so->so_lock));
1451 	ASSERT(so->so_flag & SOLOCKED);
1452 
1453 	if (!(so->so_state & SS_ISBOUND)) {
1454 		error = EINVAL;
1455 		eprintsoline(so, error);
1456 		goto done;
1457 	}
1458 
1459 	mutex_exit(&so->so_lock);
1460 
1461 	/*
1462 	 * Flush the read and write side (except stream head read queue)
1463 	 * and send down T_UNBIND_REQ.
1464 	 */
1465 	(void) putnextctl1(strvp2wq(SOTOV(so)), M_FLUSH, FLUSHRW);
1466 
1467 	unbind_req.PRIM_type = T_UNBIND_REQ;
1468 	mp = soallocproto1(&unbind_req, sizeof (unbind_req),
1469 	    0, _ALLOC_SLEEP, CRED());
1470 	error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0,
1471 	    MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0);
1472 	mutex_enter(&so->so_lock);
1473 	if (error) {
1474 		eprintsoline(so, error);
1475 		goto done;
1476 	}
1477 
1478 	error = sowaitokack(so, T_UNBIND_REQ);
1479 	if (error) {
1480 		eprintsoline(so, error);
1481 		goto done;
1482 	}
1483 
1484 	/*
1485 	 * Even if some TPI message (e.g. T_DISCON_IND) was received in
1486 	 * strsock_proto while the lock was dropped above, the unbind
1487 	 * is allowed to complete.
1488 	 */
1489 	if (!(flags & _SOUNBIND_REBIND)) {
1490 		/*
1491 		 * Clear out bound address.
1492 		 */
1493 		vnode_t *vp;
1494 
1495 		if ((vp = sti->sti_ux_bound_vp) != NULL) {
1496 
1497 			/* Undo any SSL proxy setup */
1498 			if ((so->so_family == AF_INET ||
1499 			    so->so_family == AF_INET6) &&
1500 			    (so->so_type == SOCK_STREAM) &&
1501 			    (sti->sti_kssl_ent != NULL)) {
1502 				kssl_release_ent(sti->sti_kssl_ent, so,
1503 				    sti->sti_kssl_type);
1504 				sti->sti_kssl_ent = NULL;
1505 				sti->sti_kssl_type = KSSL_NO_PROXY;
1506 			}
1507 			sti->sti_ux_bound_vp = NULL;
1508 			vn_rele_stream(vp);
1509 		}
1510 		/* Clear out address */
1511 		sti->sti_laddr_len = 0;
1512 	}
1513 	so->so_state &= ~(SS_ISBOUND|SS_ACCEPTCONN);
1514 	sti->sti_laddr_valid = 0;
1515 
1516 done:
1517 
1518 	/* If the caller held the lock don't release it here */
1519 	ASSERT(MUTEX_HELD(&so->so_lock));
1520 	ASSERT(so->so_flag & SOLOCKED);
1521 
1522 	return (error);
1523 }
1524 
1525 /*
1526  * listen on the socket.
1527  * For TPI conforming transports this has to first unbind with the transport
1528  * and then bind again using the new backlog.
1529  */
1530 /* ARGSUSED */
1531 int
1532 sotpi_listen(struct sonode *so, int backlog, struct cred *cr)
1533 {
1534 	int		error = 0;
1535 	sotpi_info_t	*sti = SOTOTPI(so);
1536 
1537 	dprintso(so, 1, ("sotpi_listen(%p, %d) %s\n",
1538 	    (void *)so, backlog, pr_state(so->so_state, so->so_mode)));
1539 
1540 	if (sti->sti_serv_type == T_CLTS)
1541 		return (EOPNOTSUPP);
1542 
1543 	/*
1544 	 * If the socket is ready to accept connections already, then
1545 	 * return without doing anything.  This avoids a problem where
1546 	 * a second listen() call fails if a connection is pending and
1547 	 * leaves the socket unbound. Only when we are not unbinding
1548 	 * with the transport can we safely increase the backlog.
1549 	 */
1550 	if (so->so_state & SS_ACCEPTCONN &&
1551 	    !((so->so_family == AF_INET || so->so_family == AF_INET6) &&
1552 	    /*CONSTCOND*/
1553 	    !solisten_tpi_tcp))
1554 		return (0);
1555 
1556 	if (so->so_state & SS_ISCONNECTED)
1557 		return (EINVAL);
1558 
1559 	mutex_enter(&so->so_lock);
1560 	so_lock_single(so);	/* Set SOLOCKED */
1561 
1562 	/*
1563 	 * If the listen doesn't change the backlog we do nothing.
1564 	 * This avoids an EPROTO error from the transport.
1565 	 */
1566 	if ((so->so_state & SS_ACCEPTCONN) &&
1567 	    so->so_backlog == backlog)
1568 		goto done;
1569 
1570 	if (!(so->so_state & SS_ISBOUND)) {
1571 		/*
1572 		 * Must have been explicitly bound in the UNIX domain.
1573 		 */
1574 		if (so->so_family == AF_UNIX) {
1575 			error = EINVAL;
1576 			goto done;
1577 		}
1578 		error = sotpi_bindlisten(so, NULL, 0, backlog,
1579 		    _SOBIND_UNSPEC|_SOBIND_LOCK_HELD|_SOBIND_LISTEN, cr);
1580 	} else if (backlog > 0) {
1581 		/*
1582 		 * AF_INET{,6} hack to avoid losing the port.
1583 		 * Assumes that all AF_INET{,6} transports can handle a
1584 		 * O_T_BIND_REQ with a non-zero CONIND_number when the TPI
1585 		 * has already bound thus it is possible to avoid the unbind.
1586 		 */
1587 		if (!((so->so_family == AF_INET || so->so_family == AF_INET6) &&
1588 		    /*CONSTCOND*/
1589 		    !solisten_tpi_tcp)) {
1590 			error = sotpi_unbind(so, _SOUNBIND_REBIND);
1591 			if (error)
1592 				goto done;
1593 		}
1594 		error = sotpi_bindlisten(so, NULL, 0, backlog,
1595 		    _SOBIND_REBIND|_SOBIND_LOCK_HELD|_SOBIND_LISTEN, cr);
1596 	} else {
1597 		so->so_state |= SS_ACCEPTCONN;
1598 		so->so_backlog = backlog;
1599 	}
1600 	if (error)
1601 		goto done;
1602 	ASSERT(so->so_state & SS_ACCEPTCONN);
1603 done:
1604 	so_unlock_single(so, SOLOCKED);
1605 	mutex_exit(&so->so_lock);
1606 	return (error);
1607 }
1608 
1609 /*
1610  * Disconnect either a specified seqno or all (-1).
1611  * The former is used on listening sockets only.
1612  *
1613  * When seqno == -1 sodisconnect could call sotpi_unbind. However,
1614  * the current use of sodisconnect(seqno == -1) is only for shutdown
1615  * so there is no point (and potentially incorrect) to unbind.
1616  */
1617 static int
1618 sodisconnect(struct sonode *so, t_scalar_t seqno, int flags)
1619 {
1620 	struct T_discon_req	discon_req;
1621 	int			error = 0;
1622 	mblk_t			*mp;
1623 
1624 	dprintso(so, 1, ("sodisconnect(%p, %d, 0x%x) %s\n",
1625 	    (void *)so, seqno, flags, pr_state(so->so_state, so->so_mode)));
1626 
1627 	if (!(flags & _SODISCONNECT_LOCK_HELD)) {
1628 		mutex_enter(&so->so_lock);
1629 		so_lock_single(so);	/* Set SOLOCKED */
1630 	} else {
1631 		ASSERT(MUTEX_HELD(&so->so_lock));
1632 		ASSERT(so->so_flag & SOLOCKED);
1633 	}
1634 
1635 	if (!(so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING|SS_ACCEPTCONN))) {
1636 		error = EINVAL;
1637 		eprintsoline(so, error);
1638 		goto done;
1639 	}
1640 
1641 	mutex_exit(&so->so_lock);
1642 	/*
1643 	 * Flush the write side (unless this is a listener)
1644 	 * and then send down a T_DISCON_REQ.
1645 	 * (Don't flush on listener since it could flush {O_}T_CONN_RES
1646 	 * and other messages.)
1647 	 */
1648 	if (!(so->so_state & SS_ACCEPTCONN))
1649 		(void) putnextctl1(strvp2wq(SOTOV(so)), M_FLUSH, FLUSHW);
1650 
1651 	discon_req.PRIM_type = T_DISCON_REQ;
1652 	discon_req.SEQ_number = seqno;
1653 	mp = soallocproto1(&discon_req, sizeof (discon_req),
1654 	    0, _ALLOC_SLEEP, CRED());
1655 	error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0,
1656 	    MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0);
1657 	mutex_enter(&so->so_lock);
1658 	if (error) {
1659 		eprintsoline(so, error);
1660 		goto done;
1661 	}
1662 
1663 	error = sowaitokack(so, T_DISCON_REQ);
1664 	if (error) {
1665 		eprintsoline(so, error);
1666 		goto done;
1667 	}
1668 	/*
1669 	 * Even if some TPI message (e.g. T_DISCON_IND) was received in
1670 	 * strsock_proto while the lock was dropped above, the disconnect
1671 	 * is allowed to complete. However, it is not possible to
1672 	 * assert that SS_ISCONNECTED|SS_ISCONNECTING are set.
1673 	 */
1674 	so->so_state &= ~(SS_ISCONNECTED|SS_ISCONNECTING);
1675 	SOTOTPI(so)->sti_laddr_valid = 0;
1676 	SOTOTPI(so)->sti_faddr_valid = 0;
1677 done:
1678 	if (!(flags & _SODISCONNECT_LOCK_HELD)) {
1679 		so_unlock_single(so, SOLOCKED);
1680 		mutex_exit(&so->so_lock);
1681 	} else {
1682 		/* If the caller held the lock don't release it here */
1683 		ASSERT(MUTEX_HELD(&so->so_lock));
1684 		ASSERT(so->so_flag & SOLOCKED);
1685 	}
1686 	return (error);
1687 }
1688 
1689 /* ARGSUSED */
1690 int
1691 sotpi_accept(struct sonode *so, int fflag, struct cred *cr,
1692     struct sonode **nsop)
1693 {
1694 	struct T_conn_ind	*conn_ind;
1695 	struct T_conn_res	*conn_res;
1696 	int			error = 0;
1697 	mblk_t			*mp, *ctxmp, *ack_mp;
1698 	struct sonode		*nso;
1699 	vnode_t			*nvp;
1700 	void			*src;
1701 	t_uscalar_t		srclen;
1702 	void			*opt;
1703 	t_uscalar_t		optlen;
1704 	t_scalar_t		PRIM_type;
1705 	t_scalar_t		SEQ_number;
1706 	size_t			sinlen;
1707 	sotpi_info_t		*sti = SOTOTPI(so);
1708 	sotpi_info_t		*nsti;
1709 
1710 	dprintso(so, 1, ("sotpi_accept(%p, 0x%x, %p) %s\n",
1711 	    (void *)so, fflag, (void *)nsop,
1712 	    pr_state(so->so_state, so->so_mode)));
1713 
1714 	/*
1715 	 * Defer single-threading the accepting socket until
1716 	 * the T_CONN_IND has been received and parsed and the
1717 	 * new sonode has been opened.
1718 	 */
1719 
1720 	/* Check that we are not already connected */
1721 	if ((so->so_state & SS_ACCEPTCONN) == 0)
1722 		goto conn_bad;
1723 again:
1724 	if ((error = sowaitconnind(so, fflag, &mp)) != 0)
1725 		goto e_bad;
1726 
1727 	ASSERT(mp != NULL);
1728 	conn_ind = (struct T_conn_ind *)mp->b_rptr;
1729 	ctxmp = mp->b_cont;
1730 
1731 	/*
1732 	 * Save SEQ_number for error paths.
1733 	 */
1734 	SEQ_number = conn_ind->SEQ_number;
1735 
1736 	srclen = conn_ind->SRC_length;
1737 	src = sogetoff(mp, conn_ind->SRC_offset, srclen, 1);
1738 	if (src == NULL) {
1739 		error = EPROTO;
1740 		freemsg(mp);
1741 		eprintsoline(so, error);
1742 		goto disconnect_unlocked;
1743 	}
1744 	optlen = conn_ind->OPT_length;
1745 	switch (so->so_family) {
1746 	case AF_INET:
1747 	case AF_INET6:
1748 		if ((optlen == sizeof (intptr_t)) && (sti->sti_direct != 0)) {
1749 			bcopy(mp->b_rptr + conn_ind->OPT_offset,
1750 			    &opt, conn_ind->OPT_length);
1751 		} else {
1752 			/*
1753 			 * The transport (in this case TCP) hasn't sent up
1754 			 * a pointer to an instance for the accept fast-path.
1755 			 * Disable fast-path completely because the call to
1756 			 * sotpi_create() below would otherwise create an
1757 			 * incomplete TCP instance, which would lead to
1758 			 * problems when sockfs sends a normal T_CONN_RES
1759 			 * message down the new stream.
1760 			 */
1761 			if (sti->sti_direct) {
1762 				int rval;
1763 				/*
1764 				 * For consistency we inform tcp to disable
1765 				 * direct interface on the listener, though
1766 				 * we can certainly live without doing this
1767 				 * because no data will ever travel upstream
1768 				 * on the listening socket.
1769 				 */
1770 				sti->sti_direct = 0;
1771 				(void) strioctl(SOTOV(so), _SIOCSOCKFALLBACK,
1772 				    0, 0, K_TO_K, cr, &rval);
1773 			}
1774 			opt = NULL;
1775 			optlen = 0;
1776 		}
1777 		break;
1778 	case AF_UNIX:
1779 	default:
1780 		if (optlen != 0) {
1781 			opt = sogetoff(mp, conn_ind->OPT_offset, optlen,
1782 			    __TPI_ALIGN_SIZE);
1783 			if (opt == NULL) {
1784 				error = EPROTO;
1785 				freemsg(mp);
1786 				eprintsoline(so, error);
1787 				goto disconnect_unlocked;
1788 			}
1789 		}
1790 		if (so->so_family == AF_UNIX) {
1791 			if (!sti->sti_faddr_noxlate) {
1792 				src = NULL;
1793 				srclen = 0;
1794 			}
1795 			/* Extract src address from options */
1796 			if (optlen != 0)
1797 				so_getopt_srcaddr(opt, optlen, &src, &srclen);
1798 		}
1799 		break;
1800 	}
1801 
1802 	/*
1803 	 * Create the new socket.
1804 	 */
1805 	nso = socket_newconn(so, NULL, NULL, SOCKET_SLEEP, &error);
1806 	if (nso == NULL) {
1807 		ASSERT(error != 0);
1808 		/*
1809 		 * Accept can not fail with ENOBUFS. sotpi_create
1810 		 * sleeps waiting for memory until a signal is caught
1811 		 * so return EINTR.
1812 		 */
1813 		freemsg(mp);
1814 		if (error == ENOBUFS)
1815 			error = EINTR;
1816 		goto e_disc_unl;
1817 	}
1818 	nvp = SOTOV(nso);
1819 	nsti = SOTOTPI(nso);
1820 
1821 	/*
1822 	 * If the transport sent up an SSL connection context, then attach
1823 	 * it the new socket, and set the (sd_wputdatafunc)() and
1824 	 * (sd_rputdatafunc)() stream head hooks to intercept and process
1825 	 * SSL records.
1826 	 */
1827 	if (ctxmp != NULL) {
1828 		/*
1829 		 * This kssl_ctx_t is already held for us by the transport.
1830 		 * So, we don't need to do a kssl_hold_ctx() here.
1831 		 */
1832 		nsti->sti_kssl_ctx = *((kssl_ctx_t *)ctxmp->b_rptr);
1833 		freemsg(ctxmp);
1834 		mp->b_cont = NULL;
1835 		strsetrwputdatahooks(nvp, strsock_kssl_input,
1836 		    strsock_kssl_output);
1837 	}
1838 #ifdef DEBUG
1839 	/*
1840 	 * SO_DEBUG is used to trigger the dprint* and eprint* macros thus
1841 	 * it's inherited early to allow debugging of the accept code itself.
1842 	 */
1843 	nso->so_options |= so->so_options & SO_DEBUG;
1844 #endif /* DEBUG */
1845 
1846 	/*
1847 	 * Save the SRC address from the T_CONN_IND
1848 	 * for getpeername to work on AF_UNIX and on transports that do not
1849 	 * support TI_GETPEERNAME.
1850 	 *
1851 	 * NOTE: AF_UNIX NUL termination is ensured by the sender's
1852 	 * copyin_name().
1853 	 */
1854 	if (srclen > (t_uscalar_t)nsti->sti_faddr_maxlen) {
1855 		error = EINVAL;
1856 		freemsg(mp);
1857 		eprintsoline(so, error);
1858 		goto disconnect_vp_unlocked;
1859 	}
1860 	nsti->sti_faddr_len = (socklen_t)srclen;
1861 	ASSERT(sti->sti_faddr_len <= sti->sti_faddr_maxlen);
1862 	bcopy(src, nsti->sti_faddr_sa, srclen);
1863 	nsti->sti_faddr_valid = 1;
1864 
1865 	/*
1866 	 * Record so_peercred and so_cpid from a cred in the T_CONN_IND.
1867 	 */
1868 	if ((DB_REF(mp) > 1) || MBLKSIZE(mp) <
1869 	    (sizeof (struct T_conn_res) + sizeof (intptr_t))) {
1870 		cred_t	*cr;
1871 		pid_t	cpid;
1872 
1873 		cr = msg_getcred(mp, &cpid);
1874 		if (cr != NULL) {
1875 			crhold(cr);
1876 			nso->so_peercred = cr;
1877 			nso->so_cpid = cpid;
1878 		}
1879 		freemsg(mp);
1880 
1881 		mp = soallocproto1(NULL, sizeof (struct T_conn_res) +
1882 		    sizeof (intptr_t), 0, _ALLOC_INTR, cr);
1883 		if (mp == NULL) {
1884 			/*
1885 			 * Accept can not fail with ENOBUFS.
1886 			 * A signal was caught so return EINTR.
1887 			 */
1888 			error = EINTR;
1889 			eprintsoline(so, error);
1890 			goto disconnect_vp_unlocked;
1891 		}
1892 		conn_res = (struct T_conn_res *)mp->b_rptr;
1893 	} else {
1894 		/*
1895 		 * For efficency reasons we use msg_extractcred; no crhold
1896 		 * needed since db_credp is cleared (i.e., we move the cred
1897 		 * from the message to so_peercred.
1898 		 */
1899 		nso->so_peercred = msg_extractcred(mp, &nso->so_cpid);
1900 
1901 		mp->b_rptr = DB_BASE(mp);
1902 		conn_res = (struct T_conn_res *)mp->b_rptr;
1903 		mp->b_wptr = mp->b_rptr + sizeof (struct T_conn_res);
1904 
1905 		mblk_setcred(mp, cr, curproc->p_pid);
1906 	}
1907 
1908 	/*
1909 	 * New socket must be bound at least in sockfs and, except for AF_INET,
1910 	 * (or AF_INET6) it also has to be bound in the transport provider.
1911 	 * We set the local address in the sonode from the T_OK_ACK of the
1912 	 * T_CONN_RES. For this reason the address we bind to here isn't
1913 	 * important.
1914 	 */
1915 	if ((nso->so_family == AF_INET || nso->so_family == AF_INET6) &&
1916 	    /*CONSTCOND*/
1917 	    nso->so_type == SOCK_STREAM && !soaccept_tpi_tcp) {
1918 		/*
1919 		 * Optimization for AF_INET{,6} transports
1920 		 * that can handle a T_CONN_RES without being bound.
1921 		 */
1922 		mutex_enter(&nso->so_lock);
1923 		so_automatic_bind(nso);
1924 		mutex_exit(&nso->so_lock);
1925 	} else {
1926 		/* Perform NULL bind with the transport provider. */
1927 		if ((error = sotpi_bind(nso, NULL, 0, _SOBIND_UNSPEC,
1928 		    cr)) != 0) {
1929 			ASSERT(error != ENOBUFS);
1930 			freemsg(mp);
1931 			eprintsoline(nso, error);
1932 			goto disconnect_vp_unlocked;
1933 		}
1934 	}
1935 
1936 	/*
1937 	 * Inherit SIOCSPGRP, SS_ASYNC before we send the {O_}T_CONN_RES
1938 	 * so that any data arriving on the new socket will cause the
1939 	 * appropriate signals to be delivered for the new socket.
1940 	 *
1941 	 * No other thread (except strsock_proto and strsock_misc)
1942 	 * can access the new socket thus we relax the locking.
1943 	 */
1944 	nso->so_pgrp = so->so_pgrp;
1945 	nso->so_state |= so->so_state & SS_ASYNC;
1946 	nsti->sti_faddr_noxlate = sti->sti_faddr_noxlate;
1947 
1948 	if (nso->so_pgrp != 0) {
1949 		if ((error = so_set_events(nso, nvp, cr)) != 0) {
1950 			eprintsoline(nso, error);
1951 			error = 0;
1952 			nso->so_pgrp = 0;
1953 		}
1954 	}
1955 
1956 	/*
1957 	 * Make note of the socket level options. TCP and IP level options
1958 	 * are already inherited. We could do all this after accept is
1959 	 * successful but doing it here simplifies code and no harm done
1960 	 * for error case.
1961 	 */
1962 	nso->so_options = so->so_options & (SO_DEBUG|SO_REUSEADDR|SO_KEEPALIVE|
1963 	    SO_DONTROUTE|SO_BROADCAST|SO_USELOOPBACK|
1964 	    SO_OOBINLINE|SO_DGRAM_ERRIND|SO_LINGER);
1965 	nso->so_sndbuf = so->so_sndbuf;
1966 	nso->so_rcvbuf = so->so_rcvbuf;
1967 	if (nso->so_options & SO_LINGER)
1968 		nso->so_linger = so->so_linger;
1969 
1970 	/*
1971 	 * Note that the following sti_direct code path should be
1972 	 * removed once we are confident that the direct sockets
1973 	 * do not result in any degradation.
1974 	 */
1975 	if (sti->sti_direct) {
1976 
1977 		ASSERT(opt != NULL);
1978 
1979 		conn_res->OPT_length = optlen;
1980 		conn_res->OPT_offset = MBLKL(mp);
1981 		bcopy(&opt, mp->b_wptr, optlen);
1982 		mp->b_wptr += optlen;
1983 		conn_res->PRIM_type = T_CONN_RES;
1984 		conn_res->ACCEPTOR_id = 0;
1985 		PRIM_type = T_CONN_RES;
1986 
1987 		/* Send down the T_CONN_RES on acceptor STREAM */
1988 		error = kstrputmsg(SOTOV(nso), mp, NULL,
1989 		    0, 0, MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0);
1990 		if (error) {
1991 			mutex_enter(&so->so_lock);
1992 			so_lock_single(so);
1993 			eprintsoline(so, error);
1994 			goto disconnect_vp;
1995 		}
1996 		mutex_enter(&nso->so_lock);
1997 		error = sowaitprim(nso, T_CONN_RES, T_OK_ACK,
1998 		    (t_uscalar_t)sizeof (struct T_ok_ack), &ack_mp, 0);
1999 		if (error) {
2000 			mutex_exit(&nso->so_lock);
2001 			mutex_enter(&so->so_lock);
2002 			so_lock_single(so);
2003 			eprintsoline(so, error);
2004 			goto disconnect_vp;
2005 		}
2006 		if (nso->so_family == AF_INET) {
2007 			sin_t *sin;
2008 
2009 			sin = (sin_t *)(ack_mp->b_rptr +
2010 			    sizeof (struct T_ok_ack));
2011 			bcopy(sin, nsti->sti_laddr_sa, sizeof (sin_t));
2012 			nsti->sti_laddr_len = sizeof (sin_t);
2013 		} else {
2014 			sin6_t *sin6;
2015 
2016 			sin6 = (sin6_t *)(ack_mp->b_rptr +
2017 			    sizeof (struct T_ok_ack));
2018 			bcopy(sin6, nsti->sti_laddr_sa, sizeof (sin6_t));
2019 			nsti->sti_laddr_len = sizeof (sin6_t);
2020 		}
2021 		freemsg(ack_mp);
2022 
2023 		nso->so_state |= SS_ISCONNECTED;
2024 		nso->so_proto_handle = (sock_lower_handle_t)opt;
2025 		nsti->sti_laddr_valid = 1;
2026 
2027 		if (sti->sti_nl7c_flags & NL7C_ENABLED) {
2028 			/*
2029 			 * A NL7C marked listen()er so the new socket
2030 			 * inherits the listen()er's NL7C state, except
2031 			 * for NL7C_POLLIN.
2032 			 *
2033 			 * Only call NL7C to process the new socket if
2034 			 * the listen socket allows blocking i/o.
2035 			 */
2036 			nsti->sti_nl7c_flags =
2037 			    sti->sti_nl7c_flags & (~NL7C_POLLIN);
2038 			if (so->so_state & (SS_NONBLOCK|SS_NDELAY)) {
2039 				/*
2040 				 * Nonblocking accept() just make it
2041 				 * persist to defer processing to the
2042 				 * read-side syscall (e.g. read).
2043 				 */
2044 				nsti->sti_nl7c_flags |= NL7C_SOPERSIST;
2045 			} else if (nl7c_process(nso, B_FALSE)) {
2046 				/*
2047 				 * NL7C has completed processing on the
2048 				 * socket, close the socket and back to
2049 				 * the top to await the next T_CONN_IND.
2050 				 */
2051 				mutex_exit(&nso->so_lock);
2052 				(void) VOP_CLOSE(nvp, 0, 1, (offset_t)0,
2053 				    cr, NULL);
2054 				VN_RELE(nvp);
2055 				goto again;
2056 			}
2057 			/* Pass the new socket out */
2058 		}
2059 
2060 		mutex_exit(&nso->so_lock);
2061 
2062 		/*
2063 		 * It's possible, through the use of autopush for example,
2064 		 * that the acceptor stream may not support sti_direct
2065 		 * semantics. If the new socket does not support sti_direct
2066 		 * we issue a _SIOCSOCKFALLBACK to inform the transport
2067 		 * as we would in the I_PUSH case.
2068 		 */
2069 		if (nsti->sti_direct == 0) {
2070 			int	rval;
2071 
2072 			if ((error = strioctl(SOTOV(nso), _SIOCSOCKFALLBACK,
2073 			    0, 0, K_TO_K, cr, &rval)) != 0) {
2074 				mutex_enter(&so->so_lock);
2075 				so_lock_single(so);
2076 				eprintsoline(so, error);
2077 				goto disconnect_vp;
2078 			}
2079 		}
2080 
2081 		/*
2082 		 * Pass out new socket.
2083 		 */
2084 		if (nsop != NULL)
2085 			*nsop = nso;
2086 
2087 		return (0);
2088 	}
2089 
2090 	/*
2091 	 * This is the non-performance case for sockets (e.g. AF_UNIX sockets)
2092 	 * which don't support the FireEngine accept fast-path. It is also
2093 	 * used when the virtual "sockmod" has been I_POP'd and I_PUSH'd
2094 	 * again. Neither sockfs nor TCP attempt to find out if some other
2095 	 * random module has been inserted in between (in which case we
2096 	 * should follow TLI accept behaviour). We blindly assume the worst
2097 	 * case and revert back to old behaviour i.e. TCP will not send us
2098 	 * any option (eager) and the accept should happen on the listener
2099 	 * queue. Any queued T_conn_ind have already got their options removed
2100 	 * by so_sock2_stream() when "sockmod" was I_POP'd.
2101 	 */
2102 	/*
2103 	 * Fill in the {O_}T_CONN_RES before getting SOLOCKED.
2104 	 */
2105 	if ((nso->so_mode & SM_ACCEPTOR_ID) == 0) {
2106 #ifdef	_ILP32
2107 		queue_t	*q;
2108 
2109 		/*
2110 		 * Find read queue in driver
2111 		 * Can safely do this since we "own" nso/nvp.
2112 		 */
2113 		q = strvp2wq(nvp)->q_next;
2114 		while (SAMESTR(q))
2115 			q = q->q_next;
2116 		q = RD(q);
2117 		conn_res->ACCEPTOR_id = (t_uscalar_t)q;
2118 #else
2119 		conn_res->ACCEPTOR_id = (t_uscalar_t)getminor(nvp->v_rdev);
2120 #endif	/* _ILP32 */
2121 		conn_res->PRIM_type = O_T_CONN_RES;
2122 		PRIM_type = O_T_CONN_RES;
2123 	} else {
2124 		conn_res->ACCEPTOR_id = nsti->sti_acceptor_id;
2125 		conn_res->PRIM_type = T_CONN_RES;
2126 		PRIM_type = T_CONN_RES;
2127 	}
2128 	conn_res->SEQ_number = SEQ_number;
2129 	conn_res->OPT_length = 0;
2130 	conn_res->OPT_offset = 0;
2131 
2132 	mutex_enter(&so->so_lock);
2133 	so_lock_single(so);	/* Set SOLOCKED */
2134 	mutex_exit(&so->so_lock);
2135 
2136 	error = kstrputmsg(SOTOV(so), mp, NULL,
2137 	    0, 0, MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0);
2138 	mutex_enter(&so->so_lock);
2139 	if (error) {
2140 		eprintsoline(so, error);
2141 		goto disconnect_vp;
2142 	}
2143 	error = sowaitprim(so, PRIM_type, T_OK_ACK,
2144 	    (t_uscalar_t)sizeof (struct T_ok_ack), &ack_mp, 0);
2145 	if (error) {
2146 		eprintsoline(so, error);
2147 		goto disconnect_vp;
2148 	}
2149 	/*
2150 	 * If there is a sin/sin6 appended onto the T_OK_ACK use
2151 	 * that to set the local address. If this is not present
2152 	 * then we zero out the address and don't set the
2153 	 * sti_laddr_valid bit. For AF_UNIX endpoints we copy over
2154 	 * the pathname from the listening socket.
2155 	 */
2156 	sinlen = (nso->so_family == AF_INET) ? sizeof (sin_t) : sizeof (sin6_t);
2157 	if ((nso->so_family == AF_INET) || (nso->so_family == AF_INET6) &&
2158 	    MBLKL(ack_mp) == (sizeof (struct T_ok_ack) + sinlen)) {
2159 		ack_mp->b_rptr += sizeof (struct T_ok_ack);
2160 		bcopy(ack_mp->b_rptr, nsti->sti_laddr_sa, sinlen);
2161 		nsti->sti_laddr_len = sinlen;
2162 		nsti->sti_laddr_valid = 1;
2163 	} else if (nso->so_family == AF_UNIX) {
2164 		ASSERT(so->so_family == AF_UNIX);
2165 		nsti->sti_laddr_len = sti->sti_laddr_len;
2166 		ASSERT(nsti->sti_laddr_len <= nsti->sti_laddr_maxlen);
2167 		bcopy(sti->sti_laddr_sa, nsti->sti_laddr_sa,
2168 		    nsti->sti_laddr_len);
2169 		nsti->sti_laddr_valid = 1;
2170 	} else {
2171 		nsti->sti_laddr_len = sti->sti_laddr_len;
2172 		ASSERT(nsti->sti_laddr_len <= nsti->sti_laddr_maxlen);
2173 		bzero(nsti->sti_laddr_sa, nsti->sti_addr_size);
2174 		nsti->sti_laddr_sa->sa_family = nso->so_family;
2175 	}
2176 	freemsg(ack_mp);
2177 
2178 	so_unlock_single(so, SOLOCKED);
2179 	mutex_exit(&so->so_lock);
2180 
2181 	nso->so_state |= SS_ISCONNECTED;
2182 
2183 	/*
2184 	 * Pass out new socket.
2185 	 */
2186 	if (nsop != NULL)
2187 		*nsop = nso;
2188 
2189 	return (0);
2190 
2191 
2192 eproto_disc_unl:
2193 	error = EPROTO;
2194 e_disc_unl:
2195 	eprintsoline(so, error);
2196 	goto disconnect_unlocked;
2197 
2198 pr_disc_vp_unl:
2199 	eprintsoline(so, error);
2200 disconnect_vp_unlocked:
2201 	(void) VOP_CLOSE(nvp, 0, 1, 0, cr, NULL);
2202 	VN_RELE(nvp);
2203 disconnect_unlocked:
2204 	(void) sodisconnect(so, SEQ_number, 0);
2205 	return (error);
2206 
2207 pr_disc_vp:
2208 	eprintsoline(so, error);
2209 disconnect_vp:
2210 	(void) sodisconnect(so, SEQ_number, _SODISCONNECT_LOCK_HELD);
2211 	so_unlock_single(so, SOLOCKED);
2212 	mutex_exit(&so->so_lock);
2213 	(void) VOP_CLOSE(nvp, 0, 1, 0, cr, NULL);
2214 	VN_RELE(nvp);
2215 	return (error);
2216 
2217 conn_bad:	/* Note: SunOS 4/BSD unconditionally returns EINVAL here */
2218 	error = (so->so_type == SOCK_DGRAM || so->so_type == SOCK_RAW)
2219 	    ? EOPNOTSUPP : EINVAL;
2220 e_bad:
2221 	eprintsoline(so, error);
2222 	return (error);
2223 }
2224 
2225 /*
2226  * connect a socket.
2227  *
2228  * Allow SOCK_DGRAM sockets to reconnect (by specifying a new address) and to
2229  * unconnect (by specifying a null address).
2230  */
2231 int
2232 sotpi_connect(struct sonode *so,
2233 	struct sockaddr *name,
2234 	socklen_t namelen,
2235 	int fflag,
2236 	int flags,
2237 	struct cred *cr)
2238 {
2239 	struct T_conn_req	conn_req;
2240 	int			error = 0;
2241 	mblk_t			*mp;
2242 	void			*src;
2243 	socklen_t		srclen;
2244 	void			*addr;
2245 	socklen_t		addrlen;
2246 	boolean_t		need_unlock;
2247 	sotpi_info_t		*sti = SOTOTPI(so);
2248 
2249 	dprintso(so, 1, ("sotpi_connect(%p, %p, %d, 0x%x, 0x%x) %s\n",
2250 	    (void *)so, (void *)name, namelen, fflag, flags,
2251 	    pr_state(so->so_state, so->so_mode)));
2252 
2253 	/*
2254 	 * Preallocate the T_CONN_REQ mblk before grabbing SOLOCKED to
2255 	 * avoid sleeping for memory with SOLOCKED held.
2256 	 * We know that the T_CONN_REQ can't be larger than 2 * sti_faddr_maxlen
2257 	 * + sizeof (struct T_opthdr).
2258 	 * (the AF_UNIX so_ux_addr_xlate() does not make the address
2259 	 * exceed sti_faddr_maxlen).
2260 	 */
2261 	mp = soallocproto(sizeof (struct T_conn_req) +
2262 	    2 * sti->sti_faddr_maxlen + sizeof (struct T_opthdr), _ALLOC_INTR,
2263 	    cr);
2264 	if (mp == NULL) {
2265 		/*
2266 		 * Connect can not fail with ENOBUFS. A signal was
2267 		 * caught so return EINTR.
2268 		 */
2269 		error = EINTR;
2270 		eprintsoline(so, error);
2271 		return (error);
2272 	}
2273 
2274 	mutex_enter(&so->so_lock);
2275 	/*
2276 	 * Make sure there is a preallocated T_unbind_req message
2277 	 * before any binding. This message is allocated when the
2278 	 * socket is created. Since another thread can consume
2279 	 * so_unbind_mp by the time we return from so_lock_single(),
2280 	 * we should check the availability of so_unbind_mp after
2281 	 * we return from so_lock_single().
2282 	 */
2283 
2284 	so_lock_single(so);	/* Set SOLOCKED */
2285 	need_unlock = B_TRUE;
2286 
2287 	if (sti->sti_unbind_mp == NULL) {
2288 		dprintso(so, 1, ("sotpi_connect: allocating unbind_req\n"));
2289 		/* NOTE: holding so_lock while sleeping */
2290 		sti->sti_unbind_mp =
2291 		    soallocproto(sizeof (struct T_unbind_req), _ALLOC_INTR, cr);
2292 		if (sti->sti_unbind_mp == NULL) {
2293 			error = EINTR;
2294 			goto done;
2295 		}
2296 	}
2297 
2298 	/*
2299 	 * Can't have done a listen before connecting.
2300 	 */
2301 	if (so->so_state & SS_ACCEPTCONN) {
2302 		error = EOPNOTSUPP;
2303 		goto done;
2304 	}
2305 
2306 	/*
2307 	 * Must be bound with the transport
2308 	 */
2309 	if (!(so->so_state & SS_ISBOUND)) {
2310 		if ((so->so_family == AF_INET || so->so_family == AF_INET6) &&
2311 		    /*CONSTCOND*/
2312 		    so->so_type == SOCK_STREAM && !soconnect_tpi_tcp) {
2313 			/*
2314 			 * Optimization for AF_INET{,6} transports
2315 			 * that can handle a T_CONN_REQ without being bound.
2316 			 */
2317 			so_automatic_bind(so);
2318 		} else {
2319 			error = sotpi_bind(so, NULL, 0,
2320 			    _SOBIND_UNSPEC|_SOBIND_LOCK_HELD, cr);
2321 			if (error)
2322 				goto done;
2323 		}
2324 		ASSERT(so->so_state & SS_ISBOUND);
2325 		flags |= _SOCONNECT_DID_BIND;
2326 	}
2327 
2328 	/*
2329 	 * Handle a connect to a name parameter of type AF_UNSPEC like a
2330 	 * connect to a null address. This is the portable method to
2331 	 * unconnect a socket.
2332 	 */
2333 	if ((namelen >= sizeof (sa_family_t)) &&
2334 	    (name->sa_family == AF_UNSPEC)) {
2335 		name = NULL;
2336 		namelen = 0;
2337 	}
2338 
2339 	/*
2340 	 * Check that we are not already connected.
2341 	 * A connection-oriented socket cannot be reconnected.
2342 	 * A connected connection-less socket can be
2343 	 * - connected to a different address by a subsequent connect
2344 	 * - "unconnected" by a connect to the NULL address
2345 	 */
2346 	if (so->so_state & (SS_ISCONNECTED|SS_ISCONNECTING)) {
2347 		ASSERT(!(flags & _SOCONNECT_DID_BIND));
2348 		if (so->so_mode & SM_CONNREQUIRED) {
2349 			/* Connection-oriented socket */
2350 			error = so->so_state & SS_ISCONNECTED ?
2351 			    EISCONN : EALREADY;
2352 			goto done;
2353 		}
2354 		/* Connection-less socket */
2355 		if (name == NULL) {
2356 			/*
2357 			 * Remove the connected state and clear SO_DGRAM_ERRIND
2358 			 * since it was set when the socket was connected.
2359 			 * If this is UDP also send down a T_DISCON_REQ.
2360 			 */
2361 			int val;
2362 
2363 			if ((so->so_family == AF_INET ||
2364 			    so->so_family == AF_INET6) &&
2365 			    (so->so_type == SOCK_DGRAM ||
2366 			    so->so_type == SOCK_RAW) &&
2367 			    /*CONSTCOND*/
2368 			    !soconnect_tpi_udp) {
2369 				/* XXX What about implicitly unbinding here? */
2370 				error = sodisconnect(so, -1,
2371 				    _SODISCONNECT_LOCK_HELD);
2372 			} else {
2373 				so->so_state &=
2374 				    ~(SS_ISCONNECTED | SS_ISCONNECTING);
2375 				sti->sti_faddr_valid = 0;
2376 				sti->sti_faddr_len = 0;
2377 			}
2378 
2379 			/* Remove SOLOCKED since setsockopt will grab it */
2380 			so_unlock_single(so, SOLOCKED);
2381 			mutex_exit(&so->so_lock);
2382 
2383 			val = 0;
2384 			(void) sotpi_setsockopt(so, SOL_SOCKET,
2385 			    SO_DGRAM_ERRIND, &val, (t_uscalar_t)sizeof (val),
2386 			    cr);
2387 
2388 			mutex_enter(&so->so_lock);
2389 			so_lock_single(so);	/* Set SOLOCKED */
2390 			goto done;
2391 		}
2392 	}
2393 	ASSERT(so->so_state & SS_ISBOUND);
2394 
2395 	if (name == NULL || namelen == 0) {
2396 		error = EINVAL;
2397 		goto done;
2398 	}
2399 	/*
2400 	 * Mark the socket if sti_faddr_sa represents the transport level
2401 	 * address.
2402 	 */
2403 	if (flags & _SOCONNECT_NOXLATE) {
2404 		struct sockaddr_ux	*soaddr_ux;
2405 
2406 		ASSERT(so->so_family == AF_UNIX);
2407 		if (namelen != sizeof (struct sockaddr_ux)) {
2408 			error = EINVAL;
2409 			goto done;
2410 		}
2411 		soaddr_ux = (struct sockaddr_ux *)name;
2412 		name = (struct sockaddr *)&soaddr_ux->sou_addr;
2413 		namelen = sizeof (soaddr_ux->sou_addr);
2414 		sti->sti_faddr_noxlate = 1;
2415 	}
2416 
2417 	/*
2418 	 * Length and family checks.
2419 	 */
2420 	error = so_addr_verify(so, name, namelen);
2421 	if (error)
2422 		goto bad;
2423 
2424 	/*
2425 	 * Save foreign address. Needed for AF_UNIX as well as
2426 	 * transport providers that do not support TI_GETPEERNAME.
2427 	 * Also used for cached foreign address for TCP and UDP.
2428 	 */
2429 	if (namelen > (t_uscalar_t)sti->sti_faddr_maxlen) {
2430 		error = EINVAL;
2431 		goto done;
2432 	}
2433 	sti->sti_faddr_len = (socklen_t)namelen;
2434 	ASSERT(sti->sti_faddr_len <= sti->sti_faddr_maxlen);
2435 	bcopy(name, sti->sti_faddr_sa, namelen);
2436 	sti->sti_faddr_valid = 1;
2437 
2438 	if (so->so_family == AF_UNIX) {
2439 		if (sti->sti_faddr_noxlate) {
2440 			/*
2441 			 * Already have a transport internal address. Do not
2442 			 * pass any (transport internal) source address.
2443 			 */
2444 			addr = sti->sti_faddr_sa;
2445 			addrlen = (t_uscalar_t)sti->sti_faddr_len;
2446 			src = NULL;
2447 			srclen = 0;
2448 		} else {
2449 			/*
2450 			 * Pass the sockaddr_un source address as an option
2451 			 * and translate the remote address.
2452 			 * Holding so_lock thus sti_laddr_sa can not change.
2453 			 */
2454 			src = sti->sti_laddr_sa;
2455 			srclen = (t_uscalar_t)sti->sti_laddr_len;
2456 			dprintso(so, 1,
2457 			    ("sotpi_connect UNIX: srclen %d, src %p\n",
2458 			    srclen, src));
2459 			error = so_ux_addr_xlate(so,
2460 			    sti->sti_faddr_sa, (socklen_t)sti->sti_faddr_len,
2461 			    (flags & _SOCONNECT_XPG4_2),
2462 			    &addr, &addrlen);
2463 			if (error)
2464 				goto bad;
2465 		}
2466 	} else {
2467 		addr = sti->sti_faddr_sa;
2468 		addrlen = (t_uscalar_t)sti->sti_faddr_len;
2469 		src = NULL;
2470 		srclen = 0;
2471 	}
2472 	/*
2473 	 * When connecting a datagram socket we issue the SO_DGRAM_ERRIND
2474 	 * option which asks the transport provider to send T_UDERR_IND
2475 	 * messages. These T_UDERR_IND messages are used to return connected
2476 	 * style errors (e.g. ECONNRESET) for connected datagram sockets.
2477 	 *
2478 	 * In addition, for UDP (and SOCK_RAW AF_INET{,6} sockets)
2479 	 * we send down a T_CONN_REQ. This is needed to let the
2480 	 * transport assign a local address that is consistent with
2481 	 * the remote address. Applications depend on a getsockname()
2482 	 * after a connect() to retrieve the "source" IP address for
2483 	 * the connected socket.  Invalidate the cached local address
2484 	 * to force getsockname() to enquire of the transport.
2485 	 */
2486 	if (!(so->so_mode & SM_CONNREQUIRED)) {
2487 		/*
2488 		 * Datagram socket.
2489 		 */
2490 		int32_t val;
2491 
2492 		so_unlock_single(so, SOLOCKED);
2493 		mutex_exit(&so->so_lock);
2494 
2495 		val = 1;
2496 		(void) sotpi_setsockopt(so, SOL_SOCKET, SO_DGRAM_ERRIND,
2497 		    &val, (t_uscalar_t)sizeof (val), cr);
2498 
2499 		mutex_enter(&so->so_lock);
2500 		so_lock_single(so);	/* Set SOLOCKED */
2501 		if ((so->so_family != AF_INET && so->so_family != AF_INET6) ||
2502 		    (so->so_type != SOCK_DGRAM && so->so_type != SOCK_RAW) ||
2503 		    soconnect_tpi_udp) {
2504 			soisconnected(so);
2505 			goto done;
2506 		}
2507 		/*
2508 		 * Send down T_CONN_REQ etc.
2509 		 * Clear fflag to avoid returning EWOULDBLOCK.
2510 		 */
2511 		fflag = 0;
2512 		ASSERT(so->so_family != AF_UNIX);
2513 		sti->sti_laddr_valid = 0;
2514 	} else if (sti->sti_laddr_len != 0) {
2515 		/*
2516 		 * If the local address or port was "any" then it may be
2517 		 * changed by the transport as a result of the
2518 		 * connect.  Invalidate the cached version if we have one.
2519 		 */
2520 		switch (so->so_family) {
2521 		case AF_INET:
2522 			ASSERT(sti->sti_laddr_len == (socklen_t)sizeof (sin_t));
2523 			if (((sin_t *)sti->sti_laddr_sa)->sin_addr.s_addr ==
2524 			    INADDR_ANY ||
2525 			    ((sin_t *)sti->sti_laddr_sa)->sin_port == 0)
2526 				sti->sti_laddr_valid = 0;
2527 			break;
2528 
2529 		case AF_INET6:
2530 			ASSERT(sti->sti_laddr_len ==
2531 			    (socklen_t)sizeof (sin6_t));
2532 			if (IN6_IS_ADDR_UNSPECIFIED(
2533 			    &((sin6_t *)sti->sti_laddr_sa) ->sin6_addr) ||
2534 			    IN6_IS_ADDR_V4MAPPED_ANY(
2535 			    &((sin6_t *)sti->sti_laddr_sa)->sin6_addr) ||
2536 			    ((sin6_t *)sti->sti_laddr_sa)->sin6_port == 0)
2537 				sti->sti_laddr_valid = 0;
2538 			break;
2539 
2540 		default:
2541 			break;
2542 		}
2543 	}
2544 
2545 	/*
2546 	 * Check for failure of an earlier call
2547 	 */
2548 	if (so->so_error != 0)
2549 		goto so_bad;
2550 
2551 	/*
2552 	 * Send down T_CONN_REQ. Message was allocated above.
2553 	 */
2554 	conn_req.PRIM_type = T_CONN_REQ;
2555 	conn_req.DEST_length = addrlen;
2556 	conn_req.DEST_offset = (t_scalar_t)sizeof (conn_req);
2557 	if (srclen == 0) {
2558 		conn_req.OPT_length = 0;
2559 		conn_req.OPT_offset = 0;
2560 		soappendmsg(mp, &conn_req, sizeof (conn_req));
2561 		soappendmsg(mp, addr, addrlen);
2562 	} else {
2563 		/*
2564 		 * There is a AF_UNIX sockaddr_un to include as a source
2565 		 * address option.
2566 		 */
2567 		struct T_opthdr toh;
2568 
2569 		toh.level = SOL_SOCKET;
2570 		toh.name = SO_SRCADDR;
2571 		toh.len = (t_uscalar_t)(srclen + sizeof (struct T_opthdr));
2572 		toh.status = 0;
2573 		conn_req.OPT_length =
2574 		    (t_scalar_t)(sizeof (toh) + _TPI_ALIGN_TOPT(srclen));
2575 		conn_req.OPT_offset = (t_scalar_t)(sizeof (conn_req) +
2576 		    _TPI_ALIGN_TOPT(addrlen));
2577 
2578 		soappendmsg(mp, &conn_req, sizeof (conn_req));
2579 		soappendmsg(mp, addr, addrlen);
2580 		mp->b_wptr += _TPI_ALIGN_TOPT(addrlen) - addrlen;
2581 		soappendmsg(mp, &toh, sizeof (toh));
2582 		soappendmsg(mp, src, srclen);
2583 		mp->b_wptr += _TPI_ALIGN_TOPT(srclen) - srclen;
2584 		ASSERT(mp->b_wptr <= mp->b_datap->db_lim);
2585 	}
2586 	/*
2587 	 * Set SS_ISCONNECTING before sending down the T_CONN_REQ
2588 	 * in order to have the right state when the T_CONN_CON shows up.
2589 	 */
2590 	soisconnecting(so);
2591 	mutex_exit(&so->so_lock);
2592 
2593 	if (AU_AUDITING())
2594 		audit_sock(T_CONN_REQ, strvp2wq(SOTOV(so)), mp, 0);
2595 
2596 	error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0,
2597 	    MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR, 0);
2598 	mp = NULL;
2599 	mutex_enter(&so->so_lock);
2600 	if (error != 0)
2601 		goto bad;
2602 
2603 	if ((error = sowaitokack(so, T_CONN_REQ)) != 0)
2604 		goto bad;
2605 
2606 	/* Allow other threads to access the socket */
2607 	so_unlock_single(so, SOLOCKED);
2608 	need_unlock = B_FALSE;
2609 
2610 	/*
2611 	 * Wait until we get a T_CONN_CON or an error
2612 	 */
2613 	if ((error = sowaitconnected(so, fflag, 0)) != 0) {
2614 		so_lock_single(so);	/* Set SOLOCKED */
2615 		need_unlock = B_TRUE;
2616 	}
2617 
2618 done:
2619 	freemsg(mp);
2620 	switch (error) {
2621 	case EINPROGRESS:
2622 	case EALREADY:
2623 	case EISCONN:
2624 	case EINTR:
2625 		/* Non-fatal errors */
2626 		sti->sti_laddr_valid = 0;
2627 		/* FALLTHRU */
2628 	case 0:
2629 		break;
2630 	default:
2631 		ASSERT(need_unlock);
2632 		/*
2633 		 * Fatal errors: clear SS_ISCONNECTING in case it was set,
2634 		 * and invalidate local-address cache
2635 		 */
2636 		so->so_state &= ~SS_ISCONNECTING;
2637 		sti->sti_laddr_valid = 0;
2638 		/* A discon_ind might have already unbound us */
2639 		if ((flags & _SOCONNECT_DID_BIND) &&
2640 		    (so->so_state & SS_ISBOUND)) {
2641 			int err;
2642 
2643 			err = sotpi_unbind(so, 0);
2644 			/* LINTED - statement has no conseq */
2645 			if (err) {
2646 				eprintsoline(so, err);
2647 			}
2648 		}
2649 		break;
2650 	}
2651 	if (need_unlock)
2652 		so_unlock_single(so, SOLOCKED);
2653 	mutex_exit(&so->so_lock);
2654 	return (error);
2655 
2656 so_bad:	error = sogeterr(so, B_TRUE);
2657 bad:	eprintsoline(so, error);
2658 	goto done;
2659 }
2660 
2661 /* ARGSUSED */
2662 int
2663 sotpi_shutdown(struct sonode *so, int how, struct cred *cr)
2664 {
2665 	struct T_ordrel_req	ordrel_req;
2666 	mblk_t			*mp;
2667 	uint_t			old_state, state_change;
2668 	int			error = 0;
2669 	sotpi_info_t		*sti = SOTOTPI(so);
2670 
2671 	dprintso(so, 1, ("sotpi_shutdown(%p, %d) %s\n",
2672 	    (void *)so, how, pr_state(so->so_state, so->so_mode)));
2673 
2674 	mutex_enter(&so->so_lock);
2675 	so_lock_single(so);	/* Set SOLOCKED */
2676 
2677 	/*
2678 	 * SunOS 4.X has no check for datagram sockets.
2679 	 * 5.X checks that it is connected (ENOTCONN)
2680 	 * X/Open requires that we check the connected state.
2681 	 */
2682 	if (!(so->so_state & SS_ISCONNECTED)) {
2683 		if (!xnet_skip_checks) {
2684 			error = ENOTCONN;
2685 			if (xnet_check_print) {
2686 				printf("sockfs: X/Open shutdown check "
2687 				    "caused ENOTCONN\n");
2688 			}
2689 		}
2690 		goto done;
2691 	}
2692 	/*
2693 	 * Record the current state and then perform any state changes.
2694 	 * Then use the difference between the old and new states to
2695 	 * determine which messages need to be sent.
2696 	 * This prevents e.g. duplicate T_ORDREL_REQ when there are
2697 	 * duplicate calls to shutdown().
2698 	 */
2699 	old_state = so->so_state;
2700 
2701 	switch (how) {
2702 	case 0:
2703 		socantrcvmore(so);
2704 		break;
2705 	case 1:
2706 		socantsendmore(so);
2707 		break;
2708 	case 2:
2709 		socantsendmore(so);
2710 		socantrcvmore(so);
2711 		break;
2712 	default:
2713 		error = EINVAL;
2714 		goto done;
2715 	}
2716 
2717 	/*
2718 	 * Assumes that the SS_CANT* flags are never cleared in the above code.
2719 	 */
2720 	state_change = (so->so_state & (SS_CANTRCVMORE|SS_CANTSENDMORE)) -
2721 	    (old_state & (SS_CANTRCVMORE|SS_CANTSENDMORE));
2722 	ASSERT((state_change & ~(SS_CANTRCVMORE|SS_CANTSENDMORE)) == 0);
2723 
2724 	switch (state_change) {
2725 	case 0:
2726 		dprintso(so, 1,
2727 		    ("sotpi_shutdown: nothing to send in state 0x%x\n",
2728 		    so->so_state));
2729 		goto done;
2730 
2731 	case SS_CANTRCVMORE:
2732 		mutex_exit(&so->so_lock);
2733 		strseteof(SOTOV(so), 1);
2734 		/*
2735 		 * strseteof takes care of read side wakeups,
2736 		 * pollwakeups, and signals.
2737 		 */
2738 		/*
2739 		 * Get the read lock before flushing data to avoid problems
2740 		 * with the T_EXDATA_IND MSG_PEEK code in sotpi_recvmsg.
2741 		 */
2742 		mutex_enter(&so->so_lock);
2743 		(void) so_lock_read(so, 0);	/* Set SOREADLOCKED */
2744 		mutex_exit(&so->so_lock);
2745 
2746 		/* Flush read side queue */
2747 		strflushrq(SOTOV(so), FLUSHALL);
2748 
2749 		mutex_enter(&so->so_lock);
2750 		so_unlock_read(so);		/* Clear SOREADLOCKED */
2751 		break;
2752 
2753 	case SS_CANTSENDMORE:
2754 		mutex_exit(&so->so_lock);
2755 		strsetwerror(SOTOV(so), 0, 0, sogetwrerr);
2756 		mutex_enter(&so->so_lock);
2757 		break;
2758 
2759 	case SS_CANTSENDMORE|SS_CANTRCVMORE:
2760 		mutex_exit(&so->so_lock);
2761 		strsetwerror(SOTOV(so), 0, 0, sogetwrerr);
2762 		strseteof(SOTOV(so), 1);
2763 		/*
2764 		 * strseteof takes care of read side wakeups,
2765 		 * pollwakeups, and signals.
2766 		 */
2767 		/*
2768 		 * Get the read lock before flushing data to avoid problems
2769 		 * with the T_EXDATA_IND MSG_PEEK code in sotpi_recvmsg.
2770 		 */
2771 		mutex_enter(&so->so_lock);
2772 		(void) so_lock_read(so, 0);	/* Set SOREADLOCKED */
2773 		mutex_exit(&so->so_lock);
2774 
2775 		/* Flush read side queue */
2776 		strflushrq(SOTOV(so), FLUSHALL);
2777 
2778 		mutex_enter(&so->so_lock);
2779 		so_unlock_read(so);		/* Clear SOREADLOCKED */
2780 		break;
2781 	}
2782 
2783 	ASSERT(MUTEX_HELD(&so->so_lock));
2784 
2785 	/*
2786 	 * If either SS_CANTSENDMORE or SS_CANTRCVMORE or both of them
2787 	 * was set due to this call and the new state has both of them set:
2788 	 *	Send the AF_UNIX close indication
2789 	 *	For T_COTS send a discon_ind
2790 	 *
2791 	 * If cantsend was set due to this call:
2792 	 *	For T_COTSORD send an ordrel_ind
2793 	 *
2794 	 * Note that for T_CLTS there is no message sent here.
2795 	 */
2796 	if ((so->so_state & (SS_CANTRCVMORE|SS_CANTSENDMORE)) ==
2797 	    (SS_CANTRCVMORE|SS_CANTSENDMORE)) {
2798 		/*
2799 		 * For SunOS 4.X compatibility we tell the other end
2800 		 * that we are unable to receive at this point.
2801 		 */
2802 		if (so->so_family == AF_UNIX && sti->sti_serv_type != T_CLTS)
2803 			so_unix_close(so);
2804 
2805 		if (sti->sti_serv_type == T_COTS)
2806 			error = sodisconnect(so, -1, _SODISCONNECT_LOCK_HELD);
2807 	}
2808 	if ((state_change & SS_CANTSENDMORE) &&
2809 	    (sti->sti_serv_type == T_COTS_ORD)) {
2810 		/* Send an orderly release */
2811 		ordrel_req.PRIM_type = T_ORDREL_REQ;
2812 
2813 		mutex_exit(&so->so_lock);
2814 		mp = soallocproto1(&ordrel_req, sizeof (ordrel_req),
2815 		    0, _ALLOC_SLEEP, cr);
2816 		/*
2817 		 * Send down the T_ORDREL_REQ even if there is flow control.
2818 		 * This prevents shutdown from blocking.
2819 		 * Note that there is no T_OK_ACK for ordrel_req.
2820 		 */
2821 		error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0,
2822 		    MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR|MSG_IGNFLOW, 0);
2823 		mutex_enter(&so->so_lock);
2824 		if (error) {
2825 			eprintsoline(so, error);
2826 			goto done;
2827 		}
2828 	}
2829 
2830 done:
2831 	so_unlock_single(so, SOLOCKED);
2832 	mutex_exit(&so->so_lock);
2833 	return (error);
2834 }
2835 
2836 /*
2837  * For any connected SOCK_STREAM/SOCK_SEQPACKET AF_UNIX socket we send
2838  * a zero-length T_OPTDATA_REQ with the SO_UNIX_CLOSE option to inform the peer
2839  * that we have closed.
2840  * Also, for connected AF_UNIX SOCK_DGRAM sockets we send a zero-length
2841  * T_UNITDATA_REQ containing the same option.
2842  *
2843  * For SOCK_DGRAM half-connections (somebody connected to this end
2844  * but this end is not connect) we don't know where to send any
2845  * SO_UNIX_CLOSE.
2846  *
2847  * We have to ignore stream head errors just in case there has been
2848  * a shutdown(output).
2849  * Ignore any flow control to try to get the message more quickly to the peer.
2850  * While locally ignoring flow control solves the problem when there
2851  * is only the loopback transport on the stream it would not provide
2852  * the correct AF_UNIX socket semantics when one or more modules have
2853  * been pushed.
2854  */
2855 void
2856 so_unix_close(struct sonode *so)
2857 {
2858 	int		error;
2859 	struct T_opthdr	toh;
2860 	mblk_t		*mp;
2861 	sotpi_info_t	*sti = SOTOTPI(so);
2862 
2863 	ASSERT(MUTEX_HELD(&so->so_lock));
2864 
2865 	ASSERT(so->so_family == AF_UNIX);
2866 
2867 	if ((so->so_state & (SS_ISCONNECTED|SS_ISBOUND)) !=
2868 	    (SS_ISCONNECTED|SS_ISBOUND))
2869 		return;
2870 
2871 	dprintso(so, 1, ("so_unix_close(%p) %s\n",
2872 	    (void *)so, pr_state(so->so_state, so->so_mode)));
2873 
2874 	toh.level = SOL_SOCKET;
2875 	toh.name = SO_UNIX_CLOSE;
2876 
2877 	/* zero length + header */
2878 	toh.len = (t_uscalar_t)sizeof (struct T_opthdr);
2879 	toh.status = 0;
2880 
2881 	if (so->so_type == SOCK_STREAM || so->so_type == SOCK_SEQPACKET) {
2882 		struct T_optdata_req tdr;
2883 
2884 		tdr.PRIM_type = T_OPTDATA_REQ;
2885 		tdr.DATA_flag = 0;
2886 
2887 		tdr.OPT_length = (t_scalar_t)sizeof (toh);
2888 		tdr.OPT_offset = (t_scalar_t)sizeof (tdr);
2889 
2890 		/* NOTE: holding so_lock while sleeping */
2891 		mp = soallocproto2(&tdr, sizeof (tdr),
2892 		    &toh, sizeof (toh), 0, _ALLOC_SLEEP, CRED());
2893 	} else {
2894 		struct T_unitdata_req	tudr;
2895 		void			*addr;
2896 		socklen_t		addrlen;
2897 		void			*src;
2898 		socklen_t		srclen;
2899 		struct T_opthdr		toh2;
2900 		t_scalar_t		size;
2901 
2902 		/* Connecteded DGRAM socket */
2903 
2904 		/*
2905 		 * For AF_UNIX the destination address is translated to
2906 		 * an internal name and the source address is passed as
2907 		 * an option.
2908 		 */
2909 		/*
2910 		 * Length and family checks.
2911 		 */
2912 		error = so_addr_verify(so, sti->sti_faddr_sa,
2913 		    (t_uscalar_t)sti->sti_faddr_len);
2914 		if (error) {
2915 			eprintsoline(so, error);
2916 			return;
2917 		}
2918 		if (sti->sti_faddr_noxlate) {
2919 			/*
2920 			 * Already have a transport internal address. Do not
2921 			 * pass any (transport internal) source address.
2922 			 */
2923 			addr = sti->sti_faddr_sa;
2924 			addrlen = (t_uscalar_t)sti->sti_faddr_len;
2925 			src = NULL;
2926 			srclen = 0;
2927 		} else {
2928 			/*
2929 			 * Pass the sockaddr_un source address as an option
2930 			 * and translate the remote address.
2931 			 * Holding so_lock thus sti_laddr_sa can not change.
2932 			 */
2933 			src = sti->sti_laddr_sa;
2934 			srclen = (socklen_t)sti->sti_laddr_len;
2935 			dprintso(so, 1,
2936 			    ("so_ux_close: srclen %d, src %p\n",
2937 			    srclen, src));
2938 			error = so_ux_addr_xlate(so,
2939 			    sti->sti_faddr_sa,
2940 			    (socklen_t)sti->sti_faddr_len, 0,
2941 			    &addr, &addrlen);
2942 			if (error) {
2943 				eprintsoline(so, error);
2944 				return;
2945 			}
2946 		}
2947 		tudr.PRIM_type = T_UNITDATA_REQ;
2948 		tudr.DEST_length = addrlen;
2949 		tudr.DEST_offset = (t_scalar_t)sizeof (tudr);
2950 		if (srclen == 0) {
2951 			tudr.OPT_length = (t_scalar_t)sizeof (toh);
2952 			tudr.OPT_offset = (t_scalar_t)(sizeof (tudr) +
2953 			    _TPI_ALIGN_TOPT(addrlen));
2954 
2955 			size = tudr.OPT_offset + tudr.OPT_length;
2956 			/* NOTE: holding so_lock while sleeping */
2957 			mp = soallocproto2(&tudr, sizeof (tudr),
2958 			    addr, addrlen, size, _ALLOC_SLEEP, CRED());
2959 			mp->b_wptr += (_TPI_ALIGN_TOPT(addrlen) - addrlen);
2960 			soappendmsg(mp, &toh, sizeof (toh));
2961 		} else {
2962 			/*
2963 			 * There is a AF_UNIX sockaddr_un to include as a
2964 			 * source address option.
2965 			 */
2966 			tudr.OPT_length = (t_scalar_t)(2 * sizeof (toh) +
2967 			    _TPI_ALIGN_TOPT(srclen));
2968 			tudr.OPT_offset = (t_scalar_t)(sizeof (tudr) +
2969 			    _TPI_ALIGN_TOPT(addrlen));
2970 
2971 			toh2.level = SOL_SOCKET;
2972 			toh2.name = SO_SRCADDR;
2973 			toh2.len = (t_uscalar_t)(srclen +
2974 			    sizeof (struct T_opthdr));
2975 			toh2.status = 0;
2976 
2977 			size = tudr.OPT_offset + tudr.OPT_length;
2978 
2979 			/* NOTE: holding so_lock while sleeping */
2980 			mp = soallocproto2(&tudr, sizeof (tudr),
2981 			    addr, addrlen, size, _ALLOC_SLEEP, CRED());
2982 			mp->b_wptr += _TPI_ALIGN_TOPT(addrlen) - addrlen;
2983 			soappendmsg(mp, &toh, sizeof (toh));
2984 			soappendmsg(mp, &toh2, sizeof (toh2));
2985 			soappendmsg(mp, src, srclen);
2986 			mp->b_wptr += _TPI_ALIGN_TOPT(srclen) - srclen;
2987 		}
2988 		ASSERT(mp->b_wptr <= mp->b_datap->db_lim);
2989 	}
2990 	mutex_exit(&so->so_lock);
2991 	error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0,
2992 	    MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR|MSG_IGNFLOW, 0);
2993 	mutex_enter(&so->so_lock);
2994 }
2995 
2996 /*
2997  * Called by sotpi_recvmsg when reading a non-zero amount of data.
2998  * In addition, the caller typically verifies that there is some
2999  * potential state to clear by checking
3000  *	if (so->so_state & (SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK))
3001  * before calling this routine.
3002  * Note that such a check can be made without holding so_lock since
3003  * sotpi_recvmsg is single-threaded (using SOREADLOCKED) and only sotpi_recvmsg
3004  * decrements sti_oobsigcnt.
3005  *
3006  * When data is read *after* the point that all pending
3007  * oob data has been consumed the oob indication is cleared.
3008  *
3009  * This logic keeps select/poll returning POLLRDBAND and
3010  * SIOCATMARK returning true until we have read past
3011  * the mark.
3012  */
3013 static void
3014 sorecv_update_oobstate(struct sonode *so)
3015 {
3016 	sotpi_info_t *sti = SOTOTPI(so);
3017 
3018 	mutex_enter(&so->so_lock);
3019 	ASSERT(so_verify_oobstate(so));
3020 	dprintso(so, 1,
3021 	    ("sorecv_update_oobstate: counts %d/%d state %s\n",
3022 	    sti->sti_oobsigcnt,
3023 	    sti->sti_oobcnt, pr_state(so->so_state, so->so_mode)));
3024 	if (sti->sti_oobsigcnt == 0) {
3025 		/* No more pending oob indications */
3026 		so->so_state &= ~(SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK);
3027 		freemsg(so->so_oobmsg);
3028 		so->so_oobmsg = NULL;
3029 	}
3030 	ASSERT(so_verify_oobstate(so));
3031 	mutex_exit(&so->so_lock);
3032 }
3033 
3034 /*
3035  * Handle recv* calls for an so which has NL7C saved recv mblk_t(s).
3036  */
3037 static int
3038 nl7c_sorecv(struct sonode *so, mblk_t **rmp, uio_t *uiop, rval_t *rp)
3039 {
3040 	sotpi_info_t *sti = SOTOTPI(so);
3041 	int	error = 0;
3042 	mblk_t *tmp = NULL;
3043 	mblk_t *pmp = NULL;
3044 	mblk_t *nmp = sti->sti_nl7c_rcv_mp;
3045 
3046 	ASSERT(nmp != NULL);
3047 
3048 	while (nmp != NULL && uiop->uio_resid > 0) {
3049 		ssize_t n;
3050 
3051 		if (DB_TYPE(nmp) == M_DATA) {
3052 			/*
3053 			 * We have some data, uiomove up to resid bytes.
3054 			 */
3055 			n = MIN(MBLKL(nmp), uiop->uio_resid);
3056 			if (n > 0)
3057 				error = uiomove(nmp->b_rptr, n, UIO_READ, uiop);
3058 			nmp->b_rptr += n;
3059 			if (nmp->b_rptr == nmp->b_wptr) {
3060 				pmp = nmp;
3061 				nmp = nmp->b_cont;
3062 			}
3063 			if (error)
3064 				break;
3065 		} else {
3066 			/*
3067 			 * We only handle data, save for caller to handle.
3068 			 */
3069 			if (pmp != NULL) {
3070 				pmp->b_cont = nmp->b_cont;
3071 			}
3072 			nmp->b_cont = NULL;
3073 			if (*rmp == NULL) {
3074 				*rmp = nmp;
3075 			} else {
3076 				tmp->b_cont = nmp;
3077 			}
3078 			nmp = nmp->b_cont;
3079 			tmp = nmp;
3080 		}
3081 	}
3082 	if (pmp != NULL) {
3083 		/* Free any mblk_t(s) which we have consumed */
3084 		pmp->b_cont = NULL;
3085 		freemsg(sti->sti_nl7c_rcv_mp);
3086 	}
3087 	if ((sti->sti_nl7c_rcv_mp = nmp) == NULL) {
3088 		/* Last mblk_t so return the saved kstrgetmsg() rval/error */
3089 		if (error == 0) {
3090 			rval_t	*p = (rval_t *)&sti->sti_nl7c_rcv_rval;
3091 
3092 			error = p->r_v.r_v2;
3093 			p->r_v.r_v2 = 0;
3094 		}
3095 		rp->r_vals = sti->sti_nl7c_rcv_rval;
3096 		sti->sti_nl7c_rcv_rval = 0;
3097 	} else {
3098 		/* More mblk_t(s) to process so no rval to return */
3099 		rp->r_vals = 0;
3100 	}
3101 	return (error);
3102 }
3103 /*
3104  * Receive the next message on the queue.
3105  * If msg_controllen is non-zero when called the caller is interested in
3106  * any received control info (options).
3107  * If msg_namelen is non-zero when called the caller is interested in
3108  * any received source address.
3109  * The routine returns with msg_control and msg_name pointing to
3110  * kmem_alloc'ed memory which the caller has to free.
3111  */
3112 /* ARGSUSED */
3113 int
3114 sotpi_recvmsg(struct sonode *so, struct nmsghdr *msg, struct uio *uiop,
3115     struct cred *cr)
3116 {
3117 	union T_primitives	*tpr;
3118 	mblk_t			*mp;
3119 	uchar_t			pri;
3120 	int			pflag, opflag;
3121 	void			*control;
3122 	t_uscalar_t		controllen;
3123 	t_uscalar_t		namelen;
3124 	int			so_state = so->so_state; /* Snapshot */
3125 	ssize_t			saved_resid;
3126 	rval_t			rval;
3127 	int			flags;
3128 	clock_t			timout;
3129 	int			error = 0;
3130 	sotpi_info_t		*sti = SOTOTPI(so);
3131 
3132 	flags = msg->msg_flags;
3133 	msg->msg_flags = 0;
3134 
3135 	dprintso(so, 1, ("sotpi_recvmsg(%p, %p, 0x%x) state %s err %d\n",
3136 	    (void *)so, (void *)msg, flags,
3137 	    pr_state(so->so_state, so->so_mode), so->so_error));
3138 
3139 	if (so->so_version == SOV_STREAM) {
3140 		so_update_attrs(so, SOACC);
3141 		/* The imaginary "sockmod" has been popped - act as a stream */
3142 		return (strread(SOTOV(so), uiop, cr));
3143 	}
3144 
3145 	/*
3146 	 * If we are not connected because we have never been connected
3147 	 * we return ENOTCONN. If we have been connected (but are no longer
3148 	 * connected) then SS_CANTRCVMORE is set and we let kstrgetmsg return
3149 	 * the EOF.
3150 	 *
3151 	 * An alternative would be to post an ENOTCONN error in stream head
3152 	 * (read+write) and clear it when we're connected. However, that error
3153 	 * would cause incorrect poll/select behavior!
3154 	 */
3155 	if ((so_state & (SS_ISCONNECTED|SS_CANTRCVMORE)) == 0 &&
3156 	    (so->so_mode & SM_CONNREQUIRED)) {
3157 		return (ENOTCONN);
3158 	}
3159 
3160 	/*
3161 	 * Note: SunOS 4.X checks uio_resid == 0 before going to sleep (but
3162 	 * after checking that the read queue is empty) and returns zero.
3163 	 * This implementation will sleep (in kstrgetmsg) even if uio_resid
3164 	 * is zero.
3165 	 */
3166 
3167 	if (flags & MSG_OOB) {
3168 		/* Check that the transport supports OOB */
3169 		if (!(so->so_mode & SM_EXDATA))
3170 			return (EOPNOTSUPP);
3171 		so_update_attrs(so, SOACC);
3172 		return (sorecvoob(so, msg, uiop, flags,
3173 		    (so->so_options & SO_OOBINLINE)));
3174 	}
3175 
3176 	so_update_attrs(so, SOACC);
3177 
3178 	/*
3179 	 * Set msg_controllen and msg_namelen to zero here to make it
3180 	 * simpler in the cases that no control or name is returned.
3181 	 */
3182 	controllen = msg->msg_controllen;
3183 	namelen = msg->msg_namelen;
3184 	msg->msg_controllen = 0;
3185 	msg->msg_namelen = 0;
3186 
3187 	dprintso(so, 1, ("sotpi_recvmsg: namelen %d controllen %d\n",
3188 	    namelen, controllen));
3189 
3190 	mutex_enter(&so->so_lock);
3191 	/*
3192 	 * If an NL7C enabled socket and not waiting for write data.
3193 	 */
3194 	if ((sti->sti_nl7c_flags & (NL7C_ENABLED | NL7C_WAITWRITE)) ==
3195 	    NL7C_ENABLED) {
3196 		if (sti->sti_nl7c_uri) {
3197 			/* Close uri processing for a previous request */
3198 			nl7c_close(so);
3199 		}
3200 		if ((so_state & SS_CANTRCVMORE) &&
3201 		    sti->sti_nl7c_rcv_mp == NULL) {
3202 			/* Nothing to process, EOF */
3203 			mutex_exit(&so->so_lock);
3204 			return (0);
3205 		} else if (sti->sti_nl7c_flags & NL7C_SOPERSIST) {
3206 			/* Persistent NL7C socket, try to process request */
3207 			boolean_t ret;
3208 
3209 			ret = nl7c_process(so,
3210 			    (so->so_state & (SS_NONBLOCK|SS_NDELAY)));
3211 			rval.r_vals = sti->sti_nl7c_rcv_rval;
3212 			error = rval.r_v.r_v2;
3213 			if (error) {
3214 				/* Error of some sort, return it */
3215 				mutex_exit(&so->so_lock);
3216 				return (error);
3217 			}
3218 			if (sti->sti_nl7c_flags &&
3219 			    ! (sti->sti_nl7c_flags & NL7C_WAITWRITE)) {
3220 				/*
3221 				 * Still an NL7C socket and no data
3222 				 * to pass up to the caller.
3223 				 */
3224 				mutex_exit(&so->so_lock);
3225 				if (ret) {
3226 					/* EOF */
3227 					return (0);
3228 				} else {
3229 					/* Need more data */
3230 					return (EAGAIN);
3231 				}
3232 			}
3233 		} else {
3234 			/*
3235 			 * Not persistent so no further NL7C processing.
3236 			 */
3237 			sti->sti_nl7c_flags = 0;
3238 		}
3239 	}
3240 	/*
3241 	 * Only one reader is allowed at any given time. This is needed
3242 	 * for T_EXDATA handling and, in the future, MSG_WAITALL.
3243 	 *
3244 	 * This is slightly different that BSD behavior in that it fails with
3245 	 * EWOULDBLOCK when using nonblocking io. In BSD the read queue access
3246 	 * is single-threaded using sblock(), which is dropped while waiting
3247 	 * for data to appear. The difference shows up e.g. if one
3248 	 * file descriptor does not have O_NONBLOCK but a dup'ed file descriptor
3249 	 * does use nonblocking io and different threads are reading each
3250 	 * file descriptor. In BSD there would never be an EWOULDBLOCK error
3251 	 * in this case as long as the read queue doesn't get empty.
3252 	 * In this implementation the thread using nonblocking io can
3253 	 * get an EWOULDBLOCK error due to the blocking thread executing
3254 	 * e.g. in the uiomove in kstrgetmsg.
3255 	 * This difference is not believed to be significant.
3256 	 */
3257 	/* Set SOREADLOCKED */
3258 	error = so_lock_read_intr(so,
3259 	    uiop->uio_fmode | ((flags & MSG_DONTWAIT) ? FNONBLOCK : 0));
3260 	mutex_exit(&so->so_lock);
3261 	if (error)
3262 		return (error);
3263 
3264 	/*
3265 	 * Tell kstrgetmsg to not inspect the stream head errors until all
3266 	 * queued data has been consumed.
3267 	 * Use a timeout=-1 to wait forever unless MSG_DONTWAIT is set.
3268 	 * Also, If uio_fmode indicates nonblocking kstrgetmsg will not block.
3269 	 *
3270 	 * MSG_WAITALL only applies to M_DATA and T_DATA_IND messages and
3271 	 * to T_OPTDATA_IND that do not contain any user-visible control msg.
3272 	 * Note that MSG_WAITALL set with MSG_PEEK is a noop.
3273 	 */
3274 	pflag = MSG_ANY | MSG_DELAYERROR;
3275 	if (flags & MSG_PEEK) {
3276 		pflag |= MSG_IPEEK;
3277 		flags &= ~MSG_WAITALL;
3278 	}
3279 	if (so->so_mode & SM_ATOMIC)
3280 		pflag |= MSG_DISCARDTAIL;
3281 
3282 	if (flags & MSG_DONTWAIT)
3283 		timout = 0;
3284 	else
3285 		timout = -1;
3286 	opflag = pflag;
3287 retry:
3288 	saved_resid = uiop->uio_resid;
3289 	pri = 0;
3290 	mp = NULL;
3291 	if (sti->sti_nl7c_rcv_mp != NULL) {
3292 		/* Already kstrgetmsg()ed saved mblk(s) from NL7C */
3293 		error = nl7c_sorecv(so, &mp, uiop, &rval);
3294 	} else {
3295 		error = kstrgetmsg(SOTOV(so), &mp, uiop, &pri, &pflag,
3296 		    timout, &rval);
3297 	}
3298 	if (error != 0) {
3299 		/* kstrgetmsg returns ETIME when timeout expires */
3300 		if (error == ETIME)
3301 			error = EWOULDBLOCK;
3302 		goto out;
3303 	}
3304 	/*
3305 	 * For datagrams the MOREDATA flag is used to set MSG_TRUNC.
3306 	 * For non-datagrams MOREDATA is used to set MSG_EOR.
3307 	 */
3308 	ASSERT(!(rval.r_val1 & MORECTL));
3309 	if ((rval.r_val1 & MOREDATA) && (so->so_mode & SM_ATOMIC))
3310 		msg->msg_flags |= MSG_TRUNC;
3311 
3312 	if (mp == NULL) {
3313 		dprintso(so, 1, ("sotpi_recvmsg: got M_DATA\n"));
3314 		/*
3315 		 * 4.3BSD and 4.4BSD clears the mark when peeking across it.
3316 		 * The draft Posix socket spec states that the mark should
3317 		 * not be cleared when peeking. We follow the latter.
3318 		 */
3319 		if ((so->so_state &
3320 		    (SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK)) &&
3321 		    (uiop->uio_resid != saved_resid) &&
3322 		    !(flags & MSG_PEEK)) {
3323 			sorecv_update_oobstate(so);
3324 		}
3325 
3326 		mutex_enter(&so->so_lock);
3327 		/* Set MSG_EOR based on MOREDATA */
3328 		if (!(rval.r_val1 & MOREDATA)) {
3329 			if (so->so_state & SS_SAVEDEOR) {
3330 				msg->msg_flags |= MSG_EOR;
3331 				so->so_state &= ~SS_SAVEDEOR;
3332 			}
3333 		}
3334 		/*
3335 		 * If some data was received (i.e. not EOF) and the
3336 		 * read/recv* has not been satisfied wait for some more.
3337 		 */
3338 		if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) &&
3339 		    uiop->uio_resid != saved_resid && uiop->uio_resid > 0) {
3340 			mutex_exit(&so->so_lock);
3341 			pflag = opflag | MSG_NOMARK;
3342 			goto retry;
3343 		}
3344 		goto out_locked;
3345 	}
3346 
3347 	/* strsock_proto has already verified length and alignment */
3348 	tpr = (union T_primitives *)mp->b_rptr;
3349 	dprintso(so, 1, ("sotpi_recvmsg: type %d\n", tpr->type));
3350 
3351 	switch (tpr->type) {
3352 	case T_DATA_IND: {
3353 		if ((so->so_state &
3354 		    (SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK)) &&
3355 		    (uiop->uio_resid != saved_resid) &&
3356 		    !(flags & MSG_PEEK)) {
3357 			sorecv_update_oobstate(so);
3358 		}
3359 
3360 		/*
3361 		 * Set msg_flags to MSG_EOR based on
3362 		 * MORE_flag and MOREDATA.
3363 		 */
3364 		mutex_enter(&so->so_lock);
3365 		so->so_state &= ~SS_SAVEDEOR;
3366 		if (!(tpr->data_ind.MORE_flag & 1)) {
3367 			if (!(rval.r_val1 & MOREDATA))
3368 				msg->msg_flags |= MSG_EOR;
3369 			else
3370 				so->so_state |= SS_SAVEDEOR;
3371 		}
3372 		freemsg(mp);
3373 		/*
3374 		 * If some data was received (i.e. not EOF) and the
3375 		 * read/recv* has not been satisfied wait for some more.
3376 		 */
3377 		if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) &&
3378 		    uiop->uio_resid != saved_resid && uiop->uio_resid > 0) {
3379 			mutex_exit(&so->so_lock);
3380 			pflag = opflag | MSG_NOMARK;
3381 			goto retry;
3382 		}
3383 		goto out_locked;
3384 	}
3385 	case T_UNITDATA_IND: {
3386 		void *addr;
3387 		t_uscalar_t addrlen;
3388 		void *abuf;
3389 		t_uscalar_t optlen;
3390 		void *opt;
3391 
3392 		if ((so->so_state &
3393 		    (SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK)) &&
3394 		    (uiop->uio_resid != saved_resid) &&
3395 		    !(flags & MSG_PEEK)) {
3396 			sorecv_update_oobstate(so);
3397 		}
3398 
3399 		if (namelen != 0) {
3400 			/* Caller wants source address */
3401 			addrlen = tpr->unitdata_ind.SRC_length;
3402 			addr = sogetoff(mp,
3403 			    tpr->unitdata_ind.SRC_offset,
3404 			    addrlen, 1);
3405 			if (addr == NULL) {
3406 				freemsg(mp);
3407 				error = EPROTO;
3408 				eprintsoline(so, error);
3409 				goto out;
3410 			}
3411 			if (so->so_family == AF_UNIX) {
3412 				/*
3413 				 * Can not use the transport level address.
3414 				 * If there is a SO_SRCADDR option carrying
3415 				 * the socket level address it will be
3416 				 * extracted below.
3417 				 */
3418 				addr = NULL;
3419 				addrlen = 0;
3420 			}
3421 		}
3422 		optlen = tpr->unitdata_ind.OPT_length;
3423 		if (optlen != 0) {
3424 			t_uscalar_t ncontrollen;
3425 
3426 			/*
3427 			 * Extract any source address option.
3428 			 * Determine how large cmsg buffer is needed.
3429 			 */
3430 			opt = sogetoff(mp,
3431 			    tpr->unitdata_ind.OPT_offset,
3432 			    optlen, __TPI_ALIGN_SIZE);
3433 
3434 			if (opt == NULL) {
3435 				freemsg(mp);
3436 				error = EPROTO;
3437 				eprintsoline(so, error);
3438 				goto out;
3439 			}
3440 			if (so->so_family == AF_UNIX)
3441 				so_getopt_srcaddr(opt, optlen, &addr, &addrlen);
3442 			ncontrollen = so_cmsglen(mp, opt, optlen,
3443 			    !(flags & MSG_XPG4_2));
3444 			if (controllen != 0)
3445 				controllen = ncontrollen;
3446 			else if (ncontrollen != 0)
3447 				msg->msg_flags |= MSG_CTRUNC;
3448 		} else {
3449 			controllen = 0;
3450 		}
3451 
3452 		if (namelen != 0) {
3453 			/*
3454 			 * Return address to caller.
3455 			 * Caller handles truncation if length
3456 			 * exceeds msg_namelen.
3457 			 * NOTE: AF_UNIX NUL termination is ensured by
3458 			 * the sender's copyin_name().
3459 			 */
3460 			abuf = kmem_alloc(addrlen, KM_SLEEP);
3461 
3462 			bcopy(addr, abuf, addrlen);
3463 			msg->msg_name = abuf;
3464 			msg->msg_namelen = addrlen;
3465 		}
3466 
3467 		if (controllen != 0) {
3468 			/*
3469 			 * Return control msg to caller.
3470 			 * Caller handles truncation if length
3471 			 * exceeds msg_controllen.
3472 			 */
3473 			control = kmem_zalloc(controllen, KM_SLEEP);
3474 
3475 			error = so_opt2cmsg(mp, opt, optlen,
3476 			    !(flags & MSG_XPG4_2),
3477 			    control, controllen);
3478 			if (error) {
3479 				freemsg(mp);
3480 				if (msg->msg_namelen != 0)
3481 					kmem_free(msg->msg_name,
3482 					    msg->msg_namelen);
3483 				kmem_free(control, controllen);
3484 				eprintsoline(so, error);
3485 				goto out;
3486 			}
3487 			msg->msg_control = control;
3488 			msg->msg_controllen = controllen;
3489 		}
3490 
3491 		freemsg(mp);
3492 		goto out;
3493 	}
3494 	case T_OPTDATA_IND: {
3495 		struct T_optdata_req *tdr;
3496 		void *opt;
3497 		t_uscalar_t optlen;
3498 
3499 		if ((so->so_state &
3500 		    (SS_OOBPEND|SS_HAVEOOBDATA|SS_RCVATMARK)) &&
3501 		    (uiop->uio_resid != saved_resid) &&
3502 		    !(flags & MSG_PEEK)) {
3503 			sorecv_update_oobstate(so);
3504 		}
3505 
3506 		tdr = (struct T_optdata_req *)mp->b_rptr;
3507 		optlen = tdr->OPT_length;
3508 		if (optlen != 0) {
3509 			t_uscalar_t ncontrollen;
3510 			/*
3511 			 * Determine how large cmsg buffer is needed.
3512 			 */
3513 			opt = sogetoff(mp,
3514 			    tpr->optdata_ind.OPT_offset,
3515 			    optlen, __TPI_ALIGN_SIZE);
3516 
3517 			if (opt == NULL) {
3518 				freemsg(mp);
3519 				error = EPROTO;
3520 				eprintsoline(so, error);
3521 				goto out;
3522 			}
3523 
3524 			ncontrollen = so_cmsglen(mp, opt, optlen,
3525 			    !(flags & MSG_XPG4_2));
3526 			if (controllen != 0)
3527 				controllen = ncontrollen;
3528 			else if (ncontrollen != 0)
3529 				msg->msg_flags |= MSG_CTRUNC;
3530 		} else {
3531 			controllen = 0;
3532 		}
3533 
3534 		if (controllen != 0) {
3535 			/*
3536 			 * Return control msg to caller.
3537 			 * Caller handles truncation if length
3538 			 * exceeds msg_controllen.
3539 			 */
3540 			control = kmem_zalloc(controllen, KM_SLEEP);
3541 
3542 			error = so_opt2cmsg(mp, opt, optlen,
3543 			    !(flags & MSG_XPG4_2),
3544 			    control, controllen);
3545 			if (error) {
3546 				freemsg(mp);
3547 				kmem_free(control, controllen);
3548 				eprintsoline(so, error);
3549 				goto out;
3550 			}
3551 			msg->msg_control = control;
3552 			msg->msg_controllen = controllen;
3553 		}
3554 
3555 		/*
3556 		 * Set msg_flags to MSG_EOR based on
3557 		 * DATA_flag and MOREDATA.
3558 		 */
3559 		mutex_enter(&so->so_lock);
3560 		so->so_state &= ~SS_SAVEDEOR;
3561 		if (!(tpr->data_ind.MORE_flag & 1)) {
3562 			if (!(rval.r_val1 & MOREDATA))
3563 				msg->msg_flags |= MSG_EOR;
3564 			else
3565 				so->so_state |= SS_SAVEDEOR;
3566 		}
3567 		freemsg(mp);
3568 		/*
3569 		 * If some data was received (i.e. not EOF) and the
3570 		 * read/recv* has not been satisfied wait for some more.
3571 		 * Not possible to wait if control info was received.
3572 		 */
3573 		if ((flags & MSG_WAITALL) && !(msg->msg_flags & MSG_EOR) &&
3574 		    controllen == 0 &&
3575 		    uiop->uio_resid != saved_resid && uiop->uio_resid > 0) {
3576 			mutex_exit(&so->so_lock);
3577 			pflag = opflag | MSG_NOMARK;
3578 			goto retry;
3579 		}
3580 		goto out_locked;
3581 	}
3582 	case T_EXDATA_IND: {
3583 		dprintso(so, 1,
3584 		    ("sotpi_recvmsg: EXDATA_IND counts %d/%d consumed %ld "
3585 		    "state %s\n",
3586 		    sti->sti_oobsigcnt, sti->sti_oobcnt,
3587 		    saved_resid - uiop->uio_resid,
3588 		    pr_state(so->so_state, so->so_mode)));
3589 		/*
3590 		 * kstrgetmsg handles MSGMARK so there is nothing to
3591 		 * inspect in the T_EXDATA_IND.
3592 		 * strsock_proto makes the stream head queue the T_EXDATA_IND
3593 		 * as a separate message with no M_DATA component. Furthermore,
3594 		 * the stream head does not consolidate M_DATA messages onto
3595 		 * an MSGMARK'ed message ensuring that the T_EXDATA_IND
3596 		 * remains a message by itself. This is needed since MSGMARK
3597 		 * marks both the whole message as well as the last byte
3598 		 * of the message.
3599 		 */
3600 		freemsg(mp);
3601 		ASSERT(uiop->uio_resid == saved_resid);	/* No data */
3602 		if (flags & MSG_PEEK) {
3603 			/*
3604 			 * Even though we are peeking we consume the
3605 			 * T_EXDATA_IND thereby moving the mark information
3606 			 * to SS_RCVATMARK. Then the oob code below will
3607 			 * retry the peeking kstrgetmsg.
3608 			 * Note that the stream head read queue is
3609 			 * never flushed without holding SOREADLOCKED
3610 			 * thus the T_EXDATA_IND can not disappear
3611 			 * underneath us.
3612 			 */
3613 			dprintso(so, 1,
3614 			    ("sotpi_recvmsg: consume EXDATA_IND "
3615 			    "counts %d/%d state %s\n",
3616 			    sti->sti_oobsigcnt,
3617 			    sti->sti_oobcnt,
3618 			    pr_state(so->so_state, so->so_mode)));
3619 
3620 			pflag = MSG_ANY | MSG_DELAYERROR;
3621 			if (so->so_mode & SM_ATOMIC)
3622 				pflag |= MSG_DISCARDTAIL;
3623 
3624 			pri = 0;
3625 			mp = NULL;
3626 
3627 			error = kstrgetmsg(SOTOV(so), &mp, uiop,
3628 			    &pri, &pflag, (clock_t)-1, &rval);
3629 			ASSERT(uiop->uio_resid == saved_resid);
3630 
3631 			if (error) {
3632 #ifdef SOCK_DEBUG
3633 				if (error != EWOULDBLOCK && error != EINTR) {
3634 					eprintsoline(so, error);
3635 				}
3636 #endif /* SOCK_DEBUG */
3637 				goto out;
3638 			}
3639 			ASSERT(mp);
3640 			tpr = (union T_primitives *)mp->b_rptr;
3641 			ASSERT(tpr->type == T_EXDATA_IND);
3642 			freemsg(mp);
3643 		} /* end "if (flags & MSG_PEEK)" */
3644 
3645 		/*
3646 		 * Decrement the number of queued and pending oob.
3647 		 *
3648 		 * SS_RCVATMARK is cleared when we read past a mark.
3649 		 * SS_HAVEOOBDATA is cleared when we've read past the
3650 		 * last mark.
3651 		 * SS_OOBPEND is cleared if we've read past the last
3652 		 * mark and no (new) SIGURG has been posted.
3653 		 */
3654 		mutex_enter(&so->so_lock);
3655 		ASSERT(so_verify_oobstate(so));
3656 		ASSERT(sti->sti_oobsigcnt >= sti->sti_oobcnt);
3657 		ASSERT(sti->sti_oobsigcnt > 0);
3658 		sti->sti_oobsigcnt--;
3659 		ASSERT(sti->sti_oobcnt > 0);
3660 		sti->sti_oobcnt--;
3661 		/*
3662 		 * Since the T_EXDATA_IND has been removed from the stream
3663 		 * head, but we have not read data past the mark,
3664 		 * sockfs needs to track that the socket is still at the mark.
3665 		 *
3666 		 * Since no data was received call kstrgetmsg again to wait
3667 		 * for data.
3668 		 */
3669 		so->so_state |= SS_RCVATMARK;
3670 		mutex_exit(&so->so_lock);
3671 		dprintso(so, 1,
3672 		    ("sotpi_recvmsg: retry EXDATA_IND counts %d/%d state %s\n",
3673 		    sti->sti_oobsigcnt, sti->sti_oobcnt,
3674 		    pr_state(so->so_state, so->so_mode)));
3675 		pflag = opflag;
3676 		goto retry;
3677 	}
3678 	default:
3679 		cmn_err(CE_CONT, "sotpi_recvmsg: so %p prim %d mp %p\n",
3680 		    (void *)so, tpr->type, (void *)mp);
3681 		ASSERT(0);
3682 		freemsg(mp);
3683 		error = EPROTO;
3684 		eprintsoline(so, error);
3685 		goto out;
3686 	}
3687 	/* NOTREACHED */
3688 out:
3689 	mutex_enter(&so->so_lock);
3690 out_locked:
3691 	so_unlock_read(so);	/* Clear SOREADLOCKED */
3692 	mutex_exit(&so->so_lock);
3693 	return (error);
3694 }
3695 
3696 /*
3697  * Sending data with options on a datagram socket.
3698  * Assumes caller has verified that SS_ISBOUND etc. are set.
3699  */
3700 static int
3701 sosend_dgramcmsg(struct sonode *so, struct sockaddr *name, socklen_t namelen,
3702     struct uio *uiop, void *control, t_uscalar_t controllen, int flags)
3703 {
3704 	struct T_unitdata_req	tudr;
3705 	mblk_t			*mp;
3706 	int			error;
3707 	void			*addr;
3708 	socklen_t		addrlen;
3709 	void			*src;
3710 	socklen_t		srclen;
3711 	ssize_t			len;
3712 	int			size;
3713 	struct T_opthdr		toh;
3714 	struct fdbuf		*fdbuf;
3715 	t_uscalar_t		optlen;
3716 	void			*fds;
3717 	int			fdlen;
3718 	sotpi_info_t		*sti = SOTOTPI(so);
3719 
3720 	ASSERT(name && namelen);
3721 	ASSERT(control && controllen);
3722 
3723 	len = uiop->uio_resid;
3724 	if (len > (ssize_t)sti->sti_tidu_size) {
3725 		return (EMSGSIZE);
3726 	}
3727 
3728 	/*
3729 	 * For AF_UNIX the destination address is translated to an internal
3730 	 * name and the source address is passed as an option.
3731 	 * Also, file descriptors are passed as file pointers in an
3732 	 * option.
3733 	 */
3734 
3735 	/*
3736 	 * Length and family checks.
3737 	 */
3738 	error = so_addr_verify(so, name, namelen);
3739 	if (error) {
3740 		eprintsoline(so, error);
3741 		return (error);
3742 	}
3743 	if (so->so_family == AF_UNIX) {
3744 		if (sti->sti_faddr_noxlate) {
3745 			/*
3746 			 * Already have a transport internal address. Do not
3747 			 * pass any (transport internal) source address.
3748 			 */
3749 			addr = name;
3750 			addrlen = namelen;
3751 			src = NULL;
3752 			srclen = 0;
3753 		} else {
3754 			/*
3755 			 * Pass the sockaddr_un source address as an option
3756 			 * and translate the remote address.
3757 			 *
3758 			 * Note that this code does not prevent sti_laddr_sa
3759 			 * from changing while it is being used. Thus
3760 			 * if an unbind+bind occurs concurrently with this
3761 			 * send the peer might see a partially new and a
3762 			 * partially old "from" address.
3763 			 */
3764 			src = sti->sti_laddr_sa;
3765 			srclen = (t_uscalar_t)sti->sti_laddr_len;
3766 			dprintso(so, 1,
3767 			    ("sosend_dgramcmsg UNIX: srclen %d, src %p\n",
3768 			    srclen, src));
3769 			error = so_ux_addr_xlate(so, name, namelen,
3770 			    (flags & MSG_XPG4_2),
3771 			    &addr, &addrlen);
3772 			if (error) {
3773 				eprintsoline(so, error);
3774 				return (error);
3775 			}
3776 		}
3777 	} else {
3778 		addr = name;
3779 		addrlen = namelen;
3780 		src = NULL;
3781 		srclen = 0;
3782 	}
3783 	optlen = so_optlen(control, controllen,
3784 	    !(flags & MSG_XPG4_2));
3785 	tudr.PRIM_type = T_UNITDATA_REQ;
3786 	tudr.DEST_length = addrlen;
3787 	tudr.DEST_offset = (t_scalar_t)sizeof (tudr);
3788 	if (srclen != 0)
3789 		tudr.OPT_length = (t_scalar_t)(optlen + sizeof (toh) +
3790 		    _TPI_ALIGN_TOPT(srclen));
3791 	else
3792 		tudr.OPT_length = optlen;
3793 	tudr.OPT_offset = (t_scalar_t)(sizeof (tudr) +
3794 	    _TPI_ALIGN_TOPT(addrlen));
3795 
3796 	size = tudr.OPT_offset + tudr.OPT_length;
3797 
3798 	/*
3799 	 * File descriptors only when SM_FDPASSING set.
3800 	 */
3801 	error = so_getfdopt(control, controllen,
3802 	    !(flags & MSG_XPG4_2), &fds, &fdlen);
3803 	if (error)
3804 		return (error);
3805 	if (fdlen != -1) {
3806 		if (!(so->so_mode & SM_FDPASSING))
3807 			return (EOPNOTSUPP);
3808 
3809 		error = fdbuf_create(fds, fdlen, &fdbuf);
3810 		if (error)
3811 			return (error);
3812 		mp = fdbuf_allocmsg(size, fdbuf);
3813 	} else {
3814 		mp = soallocproto(size, _ALLOC_INTR, CRED());
3815 		if (mp == NULL) {
3816 			/*
3817 			 * Caught a signal waiting for memory.
3818 			 * Let send* return EINTR.
3819 			 */
3820 			return (EINTR);
3821 		}
3822 	}
3823 	soappendmsg(mp, &tudr, sizeof (tudr));
3824 	soappendmsg(mp, addr, addrlen);
3825 	mp->b_wptr += _TPI_ALIGN_TOPT(addrlen) - addrlen;
3826 
3827 	if (fdlen != -1) {
3828 		ASSERT(fdbuf != NULL);
3829 		toh.level = SOL_SOCKET;
3830 		toh.name = SO_FILEP;
3831 		toh.len = fdbuf->fd_size +
3832 		    (t_uscalar_t)sizeof (struct T_opthdr);
3833 		toh.status = 0;
3834 		soappendmsg(mp, &toh, sizeof (toh));
3835 		soappendmsg(mp, fdbuf, fdbuf->fd_size);
3836 		ASSERT(__TPI_TOPT_ISALIGNED(mp->b_wptr));
3837 	}
3838 	if (srclen != 0) {
3839 		/*
3840 		 * There is a AF_UNIX sockaddr_un to include as a source
3841 		 * address option.
3842 		 */
3843 		toh.level = SOL_SOCKET;
3844 		toh.name = SO_SRCADDR;
3845 		toh.len = (t_uscalar_t)(srclen + sizeof (struct T_opthdr));
3846 		toh.status = 0;
3847 		soappendmsg(mp, &toh, sizeof (toh));
3848 		soappendmsg(mp, src, srclen);
3849 		mp->b_wptr += _TPI_ALIGN_TOPT(srclen) - srclen;
3850 		ASSERT(__TPI_TOPT_ISALIGNED(mp->b_wptr));
3851 	}
3852 	ASSERT(mp->b_wptr <= mp->b_datap->db_lim);
3853 	so_cmsg2opt(control, controllen, !(flags & MSG_XPG4_2), mp);
3854 	/* At most 3 bytes left in the message */
3855 	ASSERT(MBLKL(mp) > (ssize_t)(size - __TPI_ALIGN_SIZE));
3856 	ASSERT(MBLKL(mp) <= (ssize_t)size);
3857 
3858 	ASSERT(mp->b_wptr <= mp->b_datap->db_lim);
3859 	if (AU_AUDITING())
3860 		audit_sock(T_UNITDATA_REQ, strvp2wq(SOTOV(so)), mp, 0);
3861 
3862 	error = kstrputmsg(SOTOV(so), mp, uiop, len, 0, MSG_BAND, 0);
3863 #ifdef SOCK_DEBUG
3864 	if (error) {
3865 		eprintsoline(so, error);
3866 	}
3867 #endif /* SOCK_DEBUG */
3868 	return (error);
3869 }
3870 
3871 /*
3872  * Sending data with options on a connected stream socket.
3873  * Assumes caller has verified that SS_ISCONNECTED is set.
3874  */
3875 static int
3876 sosend_svccmsg(struct sonode *so, struct uio *uiop, int more, void *control,
3877     t_uscalar_t controllen, int flags)
3878 {
3879 	struct T_optdata_req	tdr;
3880 	mblk_t			*mp;
3881 	int			error;
3882 	ssize_t			iosize;
3883 	int			size;
3884 	struct fdbuf		*fdbuf;
3885 	t_uscalar_t		optlen;
3886 	void			*fds;
3887 	int			fdlen;
3888 	struct T_opthdr		toh;
3889 	sotpi_info_t		*sti = SOTOTPI(so);
3890 
3891 	dprintso(so, 1,
3892 	    ("sosend_svccmsg: resid %ld bytes\n", uiop->uio_resid));
3893 
3894 	/*
3895 	 * Has to be bound and connected. However, since no locks are
3896 	 * held the state could have changed after sotpi_sendmsg checked it
3897 	 * thus it is not possible to ASSERT on the state.
3898 	 */
3899 
3900 	/* Options on connection-oriented only when SM_OPTDATA set. */
3901 	if (!(so->so_mode & SM_OPTDATA))
3902 		return (EOPNOTSUPP);
3903 
3904 	do {
3905 		/*
3906 		 * Set the MORE flag if uio_resid does not fit in this
3907 		 * message or if the caller passed in "more".
3908 		 * Error for transports with zero tidu_size.
3909 		 */
3910 		tdr.PRIM_type = T_OPTDATA_REQ;
3911 		iosize = sti->sti_tidu_size;
3912 		if (iosize <= 0)
3913 			return (EMSGSIZE);
3914 		if (uiop->uio_resid > iosize) {
3915 			tdr.DATA_flag = 1;
3916 		} else {
3917 			if (more)
3918 				tdr.DATA_flag = 1;
3919 			else
3920 				tdr.DATA_flag = 0;
3921 			iosize = uiop->uio_resid;
3922 		}
3923 		dprintso(so, 1, ("sosend_svccmsg: sending %d, %ld bytes\n",
3924 		    tdr.DATA_flag, iosize));
3925 
3926 		optlen = so_optlen(control, controllen, !(flags & MSG_XPG4_2));
3927 		tdr.OPT_length = optlen;
3928 		tdr.OPT_offset = (t_scalar_t)sizeof (tdr);
3929 
3930 		size = (int)sizeof (tdr) + optlen;
3931 		/*
3932 		 * File descriptors only when SM_FDPASSING set.
3933 		 */
3934 		error = so_getfdopt(control, controllen,
3935 		    !(flags & MSG_XPG4_2), &fds, &fdlen);
3936 		if (error)
3937 			return (error);
3938 		if (fdlen != -1) {
3939 			if (!(so->so_mode & SM_FDPASSING))
3940 				return (EOPNOTSUPP);
3941 
3942 			error = fdbuf_create(fds, fdlen, &fdbuf);
3943 			if (error)
3944 				return (error);
3945 			mp = fdbuf_allocmsg(size, fdbuf);
3946 		} else {
3947 			mp = soallocproto(size, _ALLOC_INTR, CRED());
3948 			if (mp == NULL) {
3949 				/*
3950 				 * Caught a signal waiting for memory.
3951 				 * Let send* return EINTR.
3952 				 */
3953 				return (EINTR);
3954 			}
3955 		}
3956 		soappendmsg(mp, &tdr, sizeof (tdr));
3957 
3958 		if (fdlen != -1) {
3959 			ASSERT(fdbuf != NULL);
3960 			toh.level = SOL_SOCKET;
3961 			toh.name = SO_FILEP;
3962 			toh.len = fdbuf->fd_size +
3963 			    (t_uscalar_t)sizeof (struct T_opthdr);
3964 			toh.status = 0;
3965 			soappendmsg(mp, &toh, sizeof (toh));
3966 			soappendmsg(mp, fdbuf, fdbuf->fd_size);
3967 			ASSERT(__TPI_TOPT_ISALIGNED(mp->b_wptr));
3968 		}
3969 		so_cmsg2opt(control, controllen, !(flags & MSG_XPG4_2), mp);
3970 		/* At most 3 bytes left in the message */
3971 		ASSERT(MBLKL(mp) > (ssize_t)(size - __TPI_ALIGN_SIZE));
3972 		ASSERT(MBLKL(mp) <= (ssize_t)size);
3973 
3974 		ASSERT(mp->b_wptr <= mp->b_datap->db_lim);
3975 
3976 		error = kstrputmsg(SOTOV(so), mp, uiop, iosize,
3977 		    0, MSG_BAND, 0);
3978 		if (error) {
3979 			eprintsoline(so, error);
3980 			return (error);
3981 		}
3982 		control = NULL;
3983 		if (uiop->uio_resid > 0) {
3984 			/*
3985 			 * Recheck for fatal errors. Fail write even though
3986 			 * some data have been written. This is consistent
3987 			 * with strwrite semantics and BSD sockets semantics.
3988 			 */
3989 			if (so->so_state & SS_CANTSENDMORE) {
3990 				eprintsoline(so, error);
3991 				return (EPIPE);
3992 			}
3993 			if (so->so_error != 0) {
3994 				mutex_enter(&so->so_lock);
3995 				error = sogeterr(so, B_TRUE);
3996 				mutex_exit(&so->so_lock);
3997 				if (error != 0) {
3998 					eprintsoline(so, error);
3999 					return (error);
4000 				}
4001 			}
4002 		}
4003 	} while (uiop->uio_resid > 0);
4004 	return (0);
4005 }
4006 
4007 /*
4008  * Sending data on a datagram socket.
4009  * Assumes caller has verified that SS_ISBOUND etc. are set.
4010  *
4011  * For AF_UNIX the destination address is translated to an internal
4012  * name and the source address is passed as an option.
4013  */
4014 int
4015 sosend_dgram(struct sonode *so, struct sockaddr	*name, socklen_t namelen,
4016     struct uio *uiop, int flags)
4017 {
4018 	struct T_unitdata_req	tudr;
4019 	mblk_t			*mp;
4020 	int			error;
4021 	void			*addr;
4022 	socklen_t		addrlen;
4023 	void			*src;
4024 	socklen_t		srclen;
4025 	ssize_t			len;
4026 	sotpi_info_t		*sti = SOTOTPI(so);
4027 
4028 	ASSERT(name != NULL && namelen != 0);
4029 
4030 	len = uiop->uio_resid;
4031 	if (len > sti->sti_tidu_size) {
4032 		error = EMSGSIZE;
4033 		goto done;
4034 	}
4035 
4036 	/* Length and family checks */
4037 	error = so_addr_verify(so, name, namelen);
4038 	if (error != 0)
4039 		goto done;
4040 
4041 	if (sti->sti_direct)
4042 		return (sodgram_direct(so, name, namelen, uiop, flags));
4043 
4044 	if (so->so_family == AF_UNIX) {
4045 		if (sti->sti_faddr_noxlate) {
4046 			/*
4047 			 * Already have a transport internal address. Do not
4048 			 * pass any (transport internal) source address.
4049 			 */
4050 			addr = name;
4051 			addrlen = namelen;
4052 			src = NULL;
4053 			srclen = 0;
4054 		} else {
4055 			/*
4056 			 * Pass the sockaddr_un source address as an option
4057 			 * and translate the remote address.
4058 			 *
4059 			 * Note that this code does not prevent sti_laddr_sa
4060 			 * from changing while it is being used. Thus
4061 			 * if an unbind+bind occurs concurrently with this
4062 			 * send the peer might see a partially new and a
4063 			 * partially old "from" address.
4064 			 */
4065 			src = sti->sti_laddr_sa;
4066 			srclen = (socklen_t)sti->sti_laddr_len;
4067 			dprintso(so, 1,
4068 			    ("sosend_dgram UNIX: srclen %d, src %p\n",
4069 			    srclen, src));
4070 			error = so_ux_addr_xlate(so, name, namelen,
4071 			    (flags & MSG_XPG4_2),
4072 			    &addr, &addrlen);
4073 			if (error) {
4074 				eprintsoline(so, error);
4075 				goto done;
4076 			}
4077 		}
4078 	} else {
4079 		addr = name;
4080 		addrlen = namelen;
4081 		src = NULL;
4082 		srclen = 0;
4083 	}
4084 	tudr.PRIM_type = T_UNITDATA_REQ;
4085 	tudr.DEST_length = addrlen;
4086 	tudr.DEST_offset = (t_scalar_t)sizeof (tudr);
4087 	if (srclen == 0) {
4088 		tudr.OPT_length = 0;
4089 		tudr.OPT_offset = 0;
4090 
4091 		mp = soallocproto2(&tudr, sizeof (tudr),
4092 		    addr, addrlen, 0, _ALLOC_INTR, CRED());
4093 		if (mp == NULL) {
4094 			/*
4095 			 * Caught a signal waiting for memory.
4096 			 * Let send* return EINTR.
4097 			 */
4098 			error = EINTR;
4099 			goto done;
4100 		}
4101 	} else {
4102 		/*
4103 		 * There is a AF_UNIX sockaddr_un to include as a source
4104 		 * address option.
4105 		 */
4106 		struct T_opthdr toh;
4107 		ssize_t size;
4108 
4109 		tudr.OPT_length = (t_scalar_t)(sizeof (toh) +
4110 		    _TPI_ALIGN_TOPT(srclen));
4111 		tudr.OPT_offset = (t_scalar_t)(sizeof (tudr) +
4112 		    _TPI_ALIGN_TOPT(addrlen));
4113 
4114 		toh.level = SOL_SOCKET;
4115 		toh.name = SO_SRCADDR;
4116 		toh.len = (t_uscalar_t)(srclen + sizeof (struct T_opthdr));
4117 		toh.status = 0;
4118 
4119 		size = tudr.OPT_offset + tudr.OPT_length;
4120 		mp = soallocproto2(&tudr, sizeof (tudr),
4121 		    addr, addrlen, size, _ALLOC_INTR, CRED());
4122 		if (mp == NULL) {
4123 			/*
4124 			 * Caught a signal waiting for memory.
4125 			 * Let send* return EINTR.
4126 			 */
4127 			error = EINTR;
4128 			goto done;
4129 		}
4130 		mp->b_wptr += _TPI_ALIGN_TOPT(addrlen) - addrlen;
4131 		soappendmsg(mp, &toh, sizeof (toh));
4132 		soappendmsg(mp, src, srclen);
4133 		mp->b_wptr += _TPI_ALIGN_TOPT(srclen) - srclen;
4134 		ASSERT(mp->b_wptr <= mp->b_datap->db_lim);
4135 	}
4136 
4137 	if (AU_AUDITING())
4138 		audit_sock(T_UNITDATA_REQ, strvp2wq(SOTOV(so)), mp, 0);
4139 
4140 	error = kstrputmsg(SOTOV(so), mp, uiop, len, 0, MSG_BAND, 0);
4141 done:
4142 #ifdef SOCK_DEBUG
4143 	if (error) {
4144 		eprintsoline(so, error);
4145 	}
4146 #endif /* SOCK_DEBUG */
4147 	return (error);
4148 }
4149 
4150 /*
4151  * Sending data on a connected stream socket.
4152  * Assumes caller has verified that SS_ISCONNECTED is set.
4153  */
4154 int
4155 sosend_svc(struct sonode *so, struct uio *uiop, t_scalar_t prim, int more,
4156     int sflag)
4157 {
4158 	struct T_data_req	tdr;
4159 	mblk_t			*mp;
4160 	int			error;
4161 	ssize_t			iosize;
4162 	sotpi_info_t		*sti = SOTOTPI(so);
4163 
4164 	dprintso(so, 1,
4165 	    ("sosend_svc: %p, resid %ld bytes, prim %d, sflag 0x%x\n",
4166 	    (void *)so, uiop->uio_resid, prim, sflag));
4167 
4168 	/*
4169 	 * Has to be bound and connected. However, since no locks are
4170 	 * held the state could have changed after sotpi_sendmsg checked it
4171 	 * thus it is not possible to ASSERT on the state.
4172 	 */
4173 
4174 	do {
4175 		/*
4176 		 * Set the MORE flag if uio_resid does not fit in this
4177 		 * message or if the caller passed in "more".
4178 		 * Error for transports with zero tidu_size.
4179 		 */
4180 		tdr.PRIM_type = prim;
4181 		iosize = sti->sti_tidu_size;
4182 		if (iosize <= 0)
4183 			return (EMSGSIZE);
4184 		if (uiop->uio_resid > iosize) {
4185 			tdr.MORE_flag = 1;
4186 		} else {
4187 			if (more)
4188 				tdr.MORE_flag = 1;
4189 			else
4190 				tdr.MORE_flag = 0;
4191 			iosize = uiop->uio_resid;
4192 		}
4193 		dprintso(so, 1, ("sosend_svc: sending 0x%x %d, %ld bytes\n",
4194 		    prim, tdr.MORE_flag, iosize));
4195 		mp = soallocproto1(&tdr, sizeof (tdr), 0, _ALLOC_INTR, CRED());
4196 		if (mp == NULL) {
4197 			/*
4198 			 * Caught a signal waiting for memory.
4199 			 * Let send* return EINTR.
4200 			 */
4201 			return (EINTR);
4202 		}
4203 
4204 		error = kstrputmsg(SOTOV(so), mp, uiop, iosize,
4205 		    0, sflag | MSG_BAND, 0);
4206 		if (error) {
4207 			eprintsoline(so, error);
4208 			return (error);
4209 		}
4210 		if (uiop->uio_resid > 0) {
4211 			/*
4212 			 * Recheck for fatal errors. Fail write even though
4213 			 * some data have been written. This is consistent
4214 			 * with strwrite semantics and BSD sockets semantics.
4215 			 */
4216 			if (so->so_state & SS_CANTSENDMORE) {
4217 				eprintsoline(so, error);
4218 				return (EPIPE);
4219 			}
4220 			if (so->so_error != 0) {
4221 				mutex_enter(&so->so_lock);
4222 				error = sogeterr(so, B_TRUE);
4223 				mutex_exit(&so->so_lock);
4224 				if (error != 0) {
4225 					eprintsoline(so, error);
4226 					return (error);
4227 				}
4228 			}
4229 		}
4230 	} while (uiop->uio_resid > 0);
4231 	return (0);
4232 }
4233 
4234 /*
4235  * Check the state for errors and call the appropriate send function.
4236  *
4237  * If MSG_DONTROUTE is set (and SO_DONTROUTE isn't already set)
4238  * this function issues a setsockopt to toggle SO_DONTROUTE before and
4239  * after sending the message.
4240  */
4241 static int
4242 sotpi_sendmsg(struct sonode *so, struct nmsghdr *msg, struct uio *uiop,
4243     struct cred *cr)
4244 {
4245 	int		so_state;
4246 	int		so_mode;
4247 	int		error;
4248 	struct sockaddr *name;
4249 	t_uscalar_t	namelen;
4250 	int		dontroute;
4251 	int		flags;
4252 	sotpi_info_t	*sti = SOTOTPI(so);
4253 
4254 	dprintso(so, 1, ("sotpi_sendmsg(%p, %p, 0x%x) state %s, error %d\n",
4255 	    (void *)so, (void *)msg, msg->msg_flags,
4256 	    pr_state(so->so_state, so->so_mode), so->so_error));
4257 
4258 	if (so->so_version == SOV_STREAM) {
4259 		/* The imaginary "sockmod" has been popped - act as a stream */
4260 		so_update_attrs(so, SOMOD);
4261 		return (strwrite(SOTOV(so), uiop, cr));
4262 	}
4263 
4264 	mutex_enter(&so->so_lock);
4265 	so_state = so->so_state;
4266 
4267 	if (so_state & SS_CANTSENDMORE) {
4268 		mutex_exit(&so->so_lock);
4269 		return (EPIPE);
4270 	}
4271 
4272 	if (so->so_error != 0) {
4273 		error = sogeterr(so, B_TRUE);
4274 		if (error != 0) {
4275 			mutex_exit(&so->so_lock);
4276 			return (error);
4277 		}
4278 	}
4279 
4280 	name = (struct sockaddr *)msg->msg_name;
4281 	namelen = msg->msg_namelen;
4282 
4283 	so_mode = so->so_mode;
4284 
4285 	if (name == NULL) {
4286 		if (!(so_state & SS_ISCONNECTED)) {
4287 			mutex_exit(&so->so_lock);
4288 			if (so_mode & SM_CONNREQUIRED)
4289 				return (ENOTCONN);
4290 			else
4291 				return (EDESTADDRREQ);
4292 		}
4293 		if (so_mode & SM_CONNREQUIRED) {
4294 			name = NULL;
4295 			namelen = 0;
4296 		} else {
4297 			/*
4298 			 * Note that this code does not prevent sti_faddr_sa
4299 			 * from changing while it is being used. Thus
4300 			 * if an "unconnect"+connect occurs concurrently with
4301 			 * this send the datagram might be delivered to a
4302 			 * garbaled address.
4303 			 */
4304 			ASSERT(sti->sti_faddr_sa);
4305 			name = sti->sti_faddr_sa;
4306 			namelen = (t_uscalar_t)sti->sti_faddr_len;
4307 		}
4308 	} else {
4309 		if (!(so_state & SS_ISCONNECTED) &&
4310 		    (so_mode & SM_CONNREQUIRED)) {
4311 			/* Required but not connected */
4312 			mutex_exit(&so->so_lock);
4313 			return (ENOTCONN);
4314 		}
4315 		/*
4316 		 * Ignore the address on connection-oriented sockets.
4317 		 * Just like BSD this code does not generate an error for
4318 		 * TCP (a CONNREQUIRED socket) when sending to an address
4319 		 * passed in with sendto/sendmsg. Instead the data is
4320 		 * delivered on the connection as if no address had been
4321 		 * supplied.
4322 		 */
4323 		if ((so_state & SS_ISCONNECTED) &&
4324 		    !(so_mode & SM_CONNREQUIRED)) {
4325 			mutex_exit(&so->so_lock);
4326 			return (EISCONN);
4327 		}
4328 		if (!(so_state & SS_ISBOUND)) {
4329 			so_lock_single(so);	/* Set SOLOCKED */
4330 			error = sotpi_bind(so, NULL, 0,
4331 			    _SOBIND_UNSPEC|_SOBIND_LOCK_HELD, cr);
4332 			so_unlock_single(so, SOLOCKED);
4333 			if (error) {
4334 				mutex_exit(&so->so_lock);
4335 				eprintsoline(so, error);
4336 				return (error);
4337 			}
4338 		}
4339 		/*
4340 		 * Handle delayed datagram errors. These are only queued
4341 		 * when the application sets SO_DGRAM_ERRIND.
4342 		 * Return the error if we are sending to the address
4343 		 * that was returned in the last T_UDERROR_IND.
4344 		 * If sending to some other address discard the delayed
4345 		 * error indication.
4346 		 */
4347 		if (sti->sti_delayed_error) {
4348 			struct T_uderror_ind	*tudi;
4349 			void			*addr;
4350 			t_uscalar_t		addrlen;
4351 			boolean_t		match = B_FALSE;
4352 
4353 			ASSERT(sti->sti_eaddr_mp);
4354 			error = sti->sti_delayed_error;
4355 			sti->sti_delayed_error = 0;
4356 			tudi =
4357 			    (struct T_uderror_ind *)sti->sti_eaddr_mp->b_rptr;
4358 			addrlen = tudi->DEST_length;
4359 			addr = sogetoff(sti->sti_eaddr_mp,
4360 			    tudi->DEST_offset, addrlen, 1);
4361 			ASSERT(addr);	/* Checked by strsock_proto */
4362 			switch (so->so_family) {
4363 			case AF_INET: {
4364 				/* Compare just IP address and port */
4365 				sin_t *sin1 = (sin_t *)name;
4366 				sin_t *sin2 = (sin_t *)addr;
4367 
4368 				if (addrlen == sizeof (sin_t) &&
4369 				    namelen == addrlen &&
4370 				    sin1->sin_port == sin2->sin_port &&
4371 				    sin1->sin_addr.s_addr ==
4372 				    sin2->sin_addr.s_addr)
4373 					match = B_TRUE;
4374 				break;
4375 			}
4376 			case AF_INET6: {
4377 				/* Compare just IP address and port. Not flow */
4378 				sin6_t *sin1 = (sin6_t *)name;
4379 				sin6_t *sin2 = (sin6_t *)addr;
4380 
4381 				if (addrlen == sizeof (sin6_t) &&
4382 				    namelen == addrlen &&
4383 				    sin1->sin6_port == sin2->sin6_port &&
4384 				    IN6_ARE_ADDR_EQUAL(&sin1->sin6_addr,
4385 				    &sin2->sin6_addr))
4386 					match = B_TRUE;
4387 				break;
4388 			}
4389 			case AF_UNIX:
4390 			default:
4391 				if (namelen == addrlen &&
4392 				    bcmp(name, addr, namelen) == 0)
4393 					match = B_TRUE;
4394 			}
4395 			if (match) {
4396 				freemsg(sti->sti_eaddr_mp);
4397 				sti->sti_eaddr_mp = NULL;
4398 				mutex_exit(&so->so_lock);
4399 #ifdef DEBUG
4400 				dprintso(so, 0,
4401 				    ("sockfs delayed error %d for %s\n",
4402 				    error,
4403 				    pr_addr(so->so_family, name, namelen)));
4404 #endif /* DEBUG */
4405 				return (error);
4406 			}
4407 			freemsg(sti->sti_eaddr_mp);
4408 			sti->sti_eaddr_mp = NULL;
4409 		}
4410 	}
4411 	mutex_exit(&so->so_lock);
4412 
4413 	flags = msg->msg_flags;
4414 	dontroute = 0;
4415 	if ((flags & MSG_DONTROUTE) && !(so->so_options & SO_DONTROUTE)) {
4416 		uint32_t	val;
4417 
4418 		val = 1;
4419 		error = sotpi_setsockopt(so, SOL_SOCKET, SO_DONTROUTE,
4420 		    &val, (t_uscalar_t)sizeof (val), cr);
4421 		if (error)
4422 			return (error);
4423 		dontroute = 1;
4424 	}
4425 
4426 	if ((flags & MSG_OOB) && !(so_mode & SM_EXDATA)) {
4427 		error = EOPNOTSUPP;
4428 		goto done;
4429 	}
4430 	if (msg->msg_controllen != 0) {
4431 		if (!(so_mode & SM_CONNREQUIRED)) {
4432 			so_update_attrs(so, SOMOD);
4433 			error = sosend_dgramcmsg(so, name, namelen, uiop,
4434 			    msg->msg_control, msg->msg_controllen, flags);
4435 		} else {
4436 			if (flags & MSG_OOB) {
4437 				/* Can't generate T_EXDATA_REQ with options */
4438 				error = EOPNOTSUPP;
4439 				goto done;
4440 			}
4441 			so_update_attrs(so, SOMOD);
4442 			error = sosend_svccmsg(so, uiop,
4443 			    !(flags & MSG_EOR),
4444 			    msg->msg_control, msg->msg_controllen,
4445 			    flags);
4446 		}
4447 		goto done;
4448 	}
4449 
4450 	so_update_attrs(so, SOMOD);
4451 	if (!(so_mode & SM_CONNREQUIRED)) {
4452 		/*
4453 		 * If there is no SO_DONTROUTE to turn off return immediately
4454 		 * from send_dgram. This can allow tail-call optimizations.
4455 		 */
4456 		if (!dontroute) {
4457 			return (sosend_dgram(so, name, namelen, uiop, flags));
4458 		}
4459 		error = sosend_dgram(so, name, namelen, uiop, flags);
4460 	} else {
4461 		t_scalar_t prim;
4462 		int sflag;
4463 
4464 		/* Ignore msg_name in the connected state */
4465 		if (flags & MSG_OOB) {
4466 			prim = T_EXDATA_REQ;
4467 			/*
4468 			 * Send down T_EXDATA_REQ even if there is flow
4469 			 * control for data.
4470 			 */
4471 			sflag = MSG_IGNFLOW;
4472 		} else {
4473 			if (so_mode & SM_BYTESTREAM) {
4474 				/* Byte stream transport - use write */
4475 				dprintso(so, 1, ("sotpi_sendmsg: write\n"));
4476 
4477 				/* Send M_DATA messages */
4478 				if ((sti->sti_nl7c_flags & NL7C_ENABLED) &&
4479 				    (error = nl7c_data(so, uiop)) >= 0) {
4480 					/* NL7C consumed the data */
4481 					return (error);
4482 				}
4483 				/*
4484 				 * If there is no SO_DONTROUTE to turn off,
4485 				 * sti_direct is on, and there is no flow
4486 				 * control, we can take the fast path.
4487 				 */
4488 				if (!dontroute && sti->sti_direct != 0 &&
4489 				    canputnext(SOTOV(so)->v_stream->sd_wrq)) {
4490 					return (sostream_direct(so, uiop,
4491 					    NULL, cr));
4492 				}
4493 				error = strwrite(SOTOV(so), uiop, cr);
4494 				goto done;
4495 			}
4496 			prim = T_DATA_REQ;
4497 			sflag = 0;
4498 		}
4499 		/*
4500 		 * If there is no SO_DONTROUTE to turn off return immediately
4501 		 * from sosend_svc. This can allow tail-call optimizations.
4502 		 */
4503 		if (!dontroute)
4504 			return (sosend_svc(so, uiop, prim,
4505 			    !(flags & MSG_EOR), sflag));
4506 		error = sosend_svc(so, uiop, prim,
4507 		    !(flags & MSG_EOR), sflag);
4508 	}
4509 	ASSERT(dontroute);
4510 done:
4511 	if (dontroute) {
4512 		uint32_t	val;
4513 
4514 		val = 0;
4515 		(void) sotpi_setsockopt(so, SOL_SOCKET, SO_DONTROUTE,
4516 		    &val, (t_uscalar_t)sizeof (val), cr);
4517 	}
4518 	return (error);
4519 }
4520 
4521 /*
4522  * kstrwritemp() has very similar semantics as that of strwrite().
4523  * The main difference is it obtains mblks from the caller and also
4524  * does not do any copy as done in strwrite() from user buffers to
4525  * kernel buffers.
4526  *
4527  * Currently, this routine is used by sendfile to send data allocated
4528  * within the kernel without any copying. This interface does not use the
4529  * synchronous stream interface as synch. stream interface implies
4530  * copying.
4531  */
4532 int
4533 kstrwritemp(struct vnode *vp, mblk_t *mp, ushort_t fmode)
4534 {
4535 	struct stdata *stp;
4536 	struct queue *wqp;
4537 	mblk_t *newmp;
4538 	char waitflag;
4539 	int tempmode;
4540 	int error = 0;
4541 	int done = 0;
4542 	struct sonode *so;
4543 	boolean_t direct;
4544 
4545 	ASSERT(vp->v_stream);
4546 	stp = vp->v_stream;
4547 
4548 	so = VTOSO(vp);
4549 	direct = _SOTOTPI(so)->sti_direct;
4550 
4551 	/*
4552 	 * This is the sockfs direct fast path. canputnext() need
4553 	 * not be accurate so we don't grab the sd_lock here. If
4554 	 * we get flow-controlled, we grab sd_lock just before the
4555 	 * do..while loop below to emulate what strwrite() does.
4556 	 */
4557 	wqp = stp->sd_wrq;
4558 	if (canputnext(wqp) && direct &&
4559 	    !(stp->sd_flag & (STWRERR|STRHUP|STPLEX))) {
4560 		return (sostream_direct(so, NULL, mp, CRED()));
4561 	} else if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
4562 		/* Fast check of flags before acquiring the lock */
4563 		mutex_enter(&stp->sd_lock);
4564 		error = strgeterr(stp, STWRERR|STRHUP|STPLEX, 0);
4565 		mutex_exit(&stp->sd_lock);
4566 		if (error != 0) {
4567 			if (!(stp->sd_flag & STPLEX) &&
4568 			    (stp->sd_wput_opt & SW_SIGPIPE)) {
4569 				error = EPIPE;
4570 			}
4571 			return (error);
4572 		}
4573 	}
4574 
4575 	waitflag = WRITEWAIT;
4576 	if (stp->sd_flag & OLDNDELAY)
4577 		tempmode = fmode & ~FNDELAY;
4578 	else
4579 		tempmode = fmode;
4580 
4581 	mutex_enter(&stp->sd_lock);
4582 	do {
4583 		if (canputnext(wqp)) {
4584 			mutex_exit(&stp->sd_lock);
4585 			if (stp->sd_wputdatafunc != NULL) {
4586 				newmp = (stp->sd_wputdatafunc)(vp, mp, NULL,
4587 				    NULL, NULL, NULL);
4588 				if (newmp == NULL) {
4589 					/* The caller will free mp */
4590 					return (ECOMM);
4591 				}
4592 				mp = newmp;
4593 			}
4594 			putnext(wqp, mp);
4595 			return (0);
4596 		}
4597 		error = strwaitq(stp, waitflag, (ssize_t)0, tempmode, -1,
4598 		    &done);
4599 	} while (error == 0 && !done);
4600 
4601 	mutex_exit(&stp->sd_lock);
4602 	/*
4603 	 * EAGAIN tells the application to try again. ENOMEM
4604 	 * is returned only if the memory allocation size
4605 	 * exceeds the physical limits of the system. ENOMEM
4606 	 * can't be true here.
4607 	 */
4608 	if (error == ENOMEM)
4609 		error = EAGAIN;
4610 	return (error);
4611 }
4612 
4613 /* ARGSUSED */
4614 static int
4615 sotpi_sendmblk(struct sonode *so, struct nmsghdr *msg, int fflag,
4616     struct cred *cr, mblk_t **mpp)
4617 {
4618 	int error;
4619 
4620 	if (so->so_family != AF_INET && so->so_family != AF_INET6)
4621 		return (EAFNOSUPPORT);
4622 
4623 	if (so->so_state & SS_CANTSENDMORE)
4624 		return (EPIPE);
4625 
4626 	if (so->so_type != SOCK_STREAM)
4627 		return (EOPNOTSUPP);
4628 
4629 	if ((so->so_state & SS_ISCONNECTED) == 0)
4630 		return (ENOTCONN);
4631 
4632 	error = kstrwritemp(so->so_vnode, *mpp, fflag);
4633 	if (error == 0)
4634 		*mpp = NULL;
4635 	return (error);
4636 }
4637 
4638 /*
4639  * Sending data on a datagram socket.
4640  * Assumes caller has verified that SS_ISBOUND etc. are set.
4641  */
4642 /* ARGSUSED */
4643 static int
4644 sodgram_direct(struct sonode *so, struct sockaddr *name,
4645     socklen_t namelen, struct uio *uiop, int flags)
4646 {
4647 	struct T_unitdata_req	tudr;
4648 	mblk_t			*mp = NULL;
4649 	int			error = 0;
4650 	void			*addr;
4651 	socklen_t		addrlen;
4652 	ssize_t			len;
4653 	struct stdata		*stp = SOTOV(so)->v_stream;
4654 	int			so_state;
4655 	queue_t			*udp_wq;
4656 	boolean_t		connected;
4657 	mblk_t			*mpdata = NULL;
4658 	sotpi_info_t		*sti = SOTOTPI(so);
4659 	uint32_t		auditing = AU_AUDITING();
4660 
4661 	ASSERT(name != NULL && namelen != 0);
4662 	ASSERT(!(so->so_mode & SM_CONNREQUIRED));
4663 	ASSERT(!(so->so_mode & SM_EXDATA));
4664 	ASSERT(so->so_family == AF_INET || so->so_family == AF_INET6);
4665 	ASSERT(SOTOV(so)->v_type == VSOCK);
4666 
4667 	/* Caller checked for proper length */
4668 	len = uiop->uio_resid;
4669 	ASSERT(len <= sti->sti_tidu_size);
4670 
4671 	/* Length and family checks have been done by caller */
4672 	ASSERT(name->sa_family == so->so_family);
4673 	ASSERT(so->so_family == AF_INET ||
4674 	    (namelen == (socklen_t)sizeof (struct sockaddr_in6)));
4675 	ASSERT(so->so_family == AF_INET6 ||
4676 	    (namelen == (socklen_t)sizeof (struct sockaddr_in)));
4677 
4678 	addr = name;
4679 	addrlen = namelen;
4680 
4681 	if (stp->sd_sidp != NULL &&
4682 	    (error = straccess(stp, JCWRITE)) != 0)
4683 		goto done;
4684 
4685 	so_state = so->so_state;
4686 
4687 	connected = so_state & SS_ISCONNECTED;
4688 	if (!connected) {
4689 		tudr.PRIM_type = T_UNITDATA_REQ;
4690 		tudr.DEST_length = addrlen;
4691 		tudr.DEST_offset = (t_scalar_t)sizeof (tudr);
4692 		tudr.OPT_length = 0;
4693 		tudr.OPT_offset = 0;
4694 
4695 		mp = soallocproto2(&tudr, sizeof (tudr), addr, addrlen, 0,
4696 		    _ALLOC_INTR, CRED());
4697 		if (mp == NULL) {
4698 			/*
4699 			 * Caught a signal waiting for memory.
4700 			 * Let send* return EINTR.
4701 			 */
4702 			error = EINTR;
4703 			goto done;
4704 		}
4705 	}
4706 
4707 	/*
4708 	 * For UDP we don't break up the copyin into smaller pieces
4709 	 * as in the TCP case.  That means if ENOMEM is returned by
4710 	 * mcopyinuio() then the uio vector has not been modified at
4711 	 * all and we fallback to either strwrite() or kstrputmsg()
4712 	 * below.  Note also that we never generate priority messages
4713 	 * from here.
4714 	 */
4715 	udp_wq = stp->sd_wrq->q_next;
4716 	if (canput(udp_wq) &&
4717 	    (mpdata = mcopyinuio(stp, uiop, -1, -1, &error)) != NULL) {
4718 		ASSERT(DB_TYPE(mpdata) == M_DATA);
4719 		ASSERT(uiop->uio_resid == 0);
4720 		if (!connected)
4721 			linkb(mp, mpdata);
4722 		else
4723 			mp = mpdata;
4724 		if (auditing)
4725 			audit_sock(T_UNITDATA_REQ, strvp2wq(SOTOV(so)), mp, 0);
4726 
4727 		udp_wput(udp_wq, mp);
4728 		return (0);
4729 	}
4730 
4731 	ASSERT(mpdata == NULL);
4732 	if (error != 0 && error != ENOMEM) {
4733 		freemsg(mp);
4734 		return (error);
4735 	}
4736 
4737 	/*
4738 	 * For connected, let strwrite() handle the blocking case.
4739 	 * Otherwise we fall thru and use kstrputmsg().
4740 	 */
4741 	if (connected)
4742 		return (strwrite(SOTOV(so), uiop, CRED()));
4743 
4744 	if (auditing)
4745 		audit_sock(T_UNITDATA_REQ, strvp2wq(SOTOV(so)), mp, 0);
4746 
4747 	error = kstrputmsg(SOTOV(so), mp, uiop, len, 0, MSG_BAND, 0);
4748 done:
4749 #ifdef SOCK_DEBUG
4750 	if (error != 0) {
4751 		eprintsoline(so, error);
4752 	}
4753 #endif /* SOCK_DEBUG */
4754 	return (error);
4755 }
4756 
4757 int
4758 sostream_direct(struct sonode *so, struct uio *uiop, mblk_t *mp, cred_t *cr)
4759 {
4760 	struct stdata *stp = SOTOV(so)->v_stream;
4761 	ssize_t iosize, rmax, maxblk;
4762 	queue_t *tcp_wq = stp->sd_wrq->q_next;
4763 	mblk_t *newmp;
4764 	int error = 0, wflag = 0;
4765 
4766 	ASSERT(so->so_mode & SM_BYTESTREAM);
4767 	ASSERT(SOTOV(so)->v_type == VSOCK);
4768 
4769 	if (stp->sd_sidp != NULL &&
4770 	    (error = straccess(stp, JCWRITE)) != 0)
4771 		return (error);
4772 
4773 	if (uiop == NULL) {
4774 		/*
4775 		 * kstrwritemp() should have checked sd_flag and
4776 		 * flow-control before coming here.  If we end up
4777 		 * here it means that we can simply pass down the
4778 		 * data to tcp.
4779 		 */
4780 		ASSERT(mp != NULL);
4781 		if (stp->sd_wputdatafunc != NULL) {
4782 			newmp = (stp->sd_wputdatafunc)(SOTOV(so), mp, NULL,
4783 			    NULL, NULL, NULL);
4784 			if (newmp == NULL) {
4785 				/* The caller will free mp */
4786 				return (ECOMM);
4787 			}
4788 			mp = newmp;
4789 		}
4790 		tcp_wput(tcp_wq, mp);
4791 		return (0);
4792 	}
4793 
4794 	/* Fallback to strwrite() to do proper error handling */
4795 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX|STRDELIM|OLDNDELAY))
4796 		return (strwrite(SOTOV(so), uiop, cr));
4797 
4798 	rmax = stp->sd_qn_maxpsz;
4799 	ASSERT(rmax >= 0 || rmax == INFPSZ);
4800 	if (rmax == 0 || uiop->uio_resid <= 0)
4801 		return (0);
4802 
4803 	if (rmax == INFPSZ)
4804 		rmax = uiop->uio_resid;
4805 
4806 	maxblk = stp->sd_maxblk;
4807 
4808 	for (;;) {
4809 		iosize = MIN(uiop->uio_resid, rmax);
4810 
4811 		mp = mcopyinuio(stp, uiop, iosize, maxblk, &error);
4812 		if (mp == NULL) {
4813 			/*
4814 			 * Fallback to strwrite() for ENOMEM; if this
4815 			 * is our first time in this routine and the uio
4816 			 * vector has not been modified, we will end up
4817 			 * calling strwrite() without any flag set.
4818 			 */
4819 			if (error == ENOMEM)
4820 				goto slow_send;
4821 			else
4822 				return (error);
4823 		}
4824 		ASSERT(uiop->uio_resid >= 0);
4825 		/*
4826 		 * If mp is non-NULL and ENOMEM is set, it means that
4827 		 * mcopyinuio() was able to break down some of the user
4828 		 * data into one or more mblks.  Send the partial data
4829 		 * to tcp and let the rest be handled in strwrite().
4830 		 */
4831 		ASSERT(error == 0 || error == ENOMEM);
4832 		if (stp->sd_wputdatafunc != NULL) {
4833 			newmp = (stp->sd_wputdatafunc)(SOTOV(so), mp, NULL,
4834 			    NULL, NULL, NULL);
4835 			if (newmp == NULL) {
4836 				/* The caller will free mp */
4837 				return (ECOMM);
4838 			}
4839 			mp = newmp;
4840 		}
4841 		tcp_wput(tcp_wq, mp);
4842 
4843 		wflag |= NOINTR;
4844 
4845 		if (uiop->uio_resid == 0) {	/* No more data; we're done */
4846 			ASSERT(error == 0);
4847 			break;
4848 		} else if (error == ENOMEM || !canput(tcp_wq) || (stp->sd_flag &
4849 		    (STWRERR|STRHUP|STPLEX|STRDELIM|OLDNDELAY))) {
4850 slow_send:
4851 			/*
4852 			 * We were able to send down partial data using
4853 			 * the direct call interface, but are now relying
4854 			 * on strwrite() to handle the non-fastpath cases.
4855 			 * If the socket is blocking we will sleep in
4856 			 * strwaitq() until write is permitted, otherwise,
4857 			 * we will need to return the amount of bytes
4858 			 * written so far back to the app.  This is the
4859 			 * reason why we pass NOINTR flag to strwrite()
4860 			 * for non-blocking socket, because we don't want
4861 			 * to return EAGAIN when portion of the user data
4862 			 * has actually been sent down.
4863 			 */
4864 			return (strwrite_common(SOTOV(so), uiop, cr, wflag));
4865 		}
4866 	}
4867 	return (0);
4868 }
4869 
4870 /*
4871  * Update sti_faddr by asking the transport (unless AF_UNIX).
4872  */
4873 /* ARGSUSED */
4874 int
4875 sotpi_getpeername(struct sonode *so, struct sockaddr *name, socklen_t *namelen,
4876     boolean_t accept, struct cred *cr)
4877 {
4878 	struct strbuf	strbuf;
4879 	int		error = 0, res;
4880 	void		*addr;
4881 	t_uscalar_t	addrlen;
4882 	k_sigset_t	smask;
4883 	sotpi_info_t	*sti = SOTOTPI(so);
4884 
4885 	dprintso(so, 1, ("sotpi_getpeername(%p) %s\n",
4886 	    (void *)so, pr_state(so->so_state, so->so_mode)));
4887 
4888 	ASSERT(*namelen > 0);
4889 	mutex_enter(&so->so_lock);
4890 	so_lock_single(so);	/* Set SOLOCKED */
4891 
4892 	if (accept) {
4893 		bcopy(sti->sti_faddr_sa, name,
4894 		    MIN(*namelen, sti->sti_faddr_len));
4895 		*namelen = sti->sti_faddr_noxlate ? 0: sti->sti_faddr_len;
4896 		goto done;
4897 	}
4898 
4899 	if (!(so->so_state & SS_ISCONNECTED)) {
4900 		error = ENOTCONN;
4901 		goto done;
4902 	}
4903 	/* Added this check for X/Open */
4904 	if ((so->so_state & SS_CANTSENDMORE) && !xnet_skip_checks) {
4905 		error = EINVAL;
4906 		if (xnet_check_print) {
4907 			printf("sockfs: X/Open getpeername check => EINVAL\n");
4908 		}
4909 		goto done;
4910 	}
4911 
4912 	if (sti->sti_faddr_valid) {
4913 		bcopy(sti->sti_faddr_sa, name,
4914 		    MIN(*namelen, sti->sti_faddr_len));
4915 		*namelen = sti->sti_faddr_noxlate ? 0: sti->sti_faddr_len;
4916 		goto done;
4917 	}
4918 
4919 #ifdef DEBUG
4920 	dprintso(so, 1, ("sotpi_getpeername (local): %s\n",
4921 	    pr_addr(so->so_family, sti->sti_faddr_sa,
4922 	    (t_uscalar_t)sti->sti_faddr_len)));
4923 #endif /* DEBUG */
4924 
4925 	if (so->so_family == AF_UNIX) {
4926 		/* Transport has different name space - return local info */
4927 		if (sti->sti_faddr_noxlate)
4928 			*namelen = 0;
4929 		error = 0;
4930 		goto done;
4931 	}
4932 
4933 	ASSERT(so->so_family != AF_UNIX && sti->sti_faddr_noxlate == 0);
4934 
4935 	ASSERT(sti->sti_faddr_sa);
4936 	/* Allocate local buffer to use with ioctl */
4937 	addrlen = (t_uscalar_t)sti->sti_faddr_maxlen;
4938 	mutex_exit(&so->so_lock);
4939 	addr = kmem_alloc(addrlen, KM_SLEEP);
4940 
4941 	/*
4942 	 * Issue TI_GETPEERNAME with signals masked.
4943 	 * Put the result in sti_faddr_sa so that getpeername works after
4944 	 * a shutdown(output).
4945 	 * If the ioctl fails (e.g. due to a ECONNRESET) the error is reposted
4946 	 * back to the socket.
4947 	 */
4948 	strbuf.buf = addr;
4949 	strbuf.maxlen = addrlen;
4950 	strbuf.len = 0;
4951 
4952 	sigintr(&smask, 0);
4953 	res = 0;
4954 	ASSERT(cr);
4955 	error = strioctl(SOTOV(so), TI_GETPEERNAME, (intptr_t)&strbuf,
4956 	    0, K_TO_K, cr, &res);
4957 	sigunintr(&smask);
4958 
4959 	mutex_enter(&so->so_lock);
4960 	/*
4961 	 * If there is an error record the error in so_error put don't fail
4962 	 * the getpeername. Instead fallback on the recorded
4963 	 * sti->sti_faddr_sa.
4964 	 */
4965 	if (error) {
4966 		/*
4967 		 * Various stream head errors can be returned to the ioctl.
4968 		 * However, it is impossible to determine which ones of
4969 		 * these are really socket level errors that were incorrectly
4970 		 * consumed by the ioctl. Thus this code silently ignores the
4971 		 * error - to code explicitly does not reinstate the error
4972 		 * using soseterror().
4973 		 * Experiments have shows that at least this set of
4974 		 * errors are reported and should not be reinstated on the
4975 		 * socket:
4976 		 *	EINVAL	E.g. if an I_LINK was in effect when
4977 		 *		getpeername was called.
4978 		 *	EPIPE	The ioctl error semantics prefer the write
4979 		 *		side error over the read side error.
4980 		 *	ENOTCONN The transport just got disconnected but
4981 		 *		sockfs had not yet seen the T_DISCON_IND
4982 		 *		when issuing the ioctl.
4983 		 */
4984 		error = 0;
4985 	} else if (res == 0 && strbuf.len > 0 &&
4986 	    (so->so_state & SS_ISCONNECTED)) {
4987 		ASSERT(strbuf.len <= (int)sti->sti_faddr_maxlen);
4988 		sti->sti_faddr_len = (socklen_t)strbuf.len;
4989 		bcopy(addr, sti->sti_faddr_sa, sti->sti_faddr_len);
4990 		sti->sti_faddr_valid = 1;
4991 
4992 		bcopy(addr, name, MIN(*namelen, sti->sti_faddr_len));
4993 		*namelen = sti->sti_faddr_len;
4994 	}
4995 	kmem_free(addr, addrlen);
4996 #ifdef DEBUG
4997 	dprintso(so, 1, ("sotpi_getpeername (tp): %s\n",
4998 	    pr_addr(so->so_family, sti->sti_faddr_sa,
4999 	    (t_uscalar_t)sti->sti_faddr_len)));
5000 #endif /* DEBUG */
5001 done:
5002 	so_unlock_single(so, SOLOCKED);
5003 	mutex_exit(&so->so_lock);
5004 	return (error);
5005 }
5006 
5007 /*
5008  * Update sti_laddr by asking the transport (unless AF_UNIX).
5009  */
5010 int
5011 sotpi_getsockname(struct sonode *so, struct sockaddr *name, socklen_t *namelen,
5012     struct cred *cr)
5013 {
5014 	struct strbuf	strbuf;
5015 	int		error = 0, res;
5016 	void		*addr;
5017 	t_uscalar_t	addrlen;
5018 	k_sigset_t	smask;
5019 	sotpi_info_t	*sti = SOTOTPI(so);
5020 
5021 	dprintso(so, 1, ("sotpi_getsockname(%p) %s\n",
5022 	    (void *)so, pr_state(so->so_state, so->so_mode)));
5023 
5024 	ASSERT(*namelen > 0);
5025 	mutex_enter(&so->so_lock);
5026 	so_lock_single(so);	/* Set SOLOCKED */
5027 
5028 #ifdef DEBUG
5029 
5030 	dprintso(so, 1, ("sotpi_getsockname (local): %s\n",
5031 	    pr_addr(so->so_family, sti->sti_laddr_sa,
5032 	    (t_uscalar_t)sti->sti_laddr_len)));
5033 #endif /* DEBUG */
5034 	if (sti->sti_laddr_valid) {
5035 		bcopy(sti->sti_laddr_sa, name,
5036 		    MIN(*namelen, sti->sti_laddr_len));
5037 		*namelen = sti->sti_laddr_len;
5038 		goto done;
5039 	}
5040 
5041 	if (so->so_family == AF_UNIX) {
5042 		/* Transport has different name space - return local info */
5043 		error = 0;
5044 		*namelen = 0;
5045 		goto done;
5046 	}
5047 	if (!(so->so_state & SS_ISBOUND)) {
5048 		/* If not bound, then nothing to return. */
5049 		error = 0;
5050 		goto done;
5051 	}
5052 
5053 	/* Allocate local buffer to use with ioctl */
5054 	addrlen = (t_uscalar_t)sti->sti_laddr_maxlen;
5055 	mutex_exit(&so->so_lock);
5056 	addr = kmem_alloc(addrlen, KM_SLEEP);
5057 
5058 	/*
5059 	 * Issue TI_GETMYNAME with signals masked.
5060 	 * Put the result in sti_laddr_sa so that getsockname works after
5061 	 * a shutdown(output).
5062 	 * If the ioctl fails (e.g. due to a ECONNRESET) the error is reposted
5063 	 * back to the socket.
5064 	 */
5065 	strbuf.buf = addr;
5066 	strbuf.maxlen = addrlen;
5067 	strbuf.len = 0;
5068 
5069 	sigintr(&smask, 0);
5070 	res = 0;
5071 	ASSERT(cr);
5072 	error = strioctl(SOTOV(so), TI_GETMYNAME, (intptr_t)&strbuf,
5073 	    0, K_TO_K, cr, &res);
5074 	sigunintr(&smask);
5075 
5076 	mutex_enter(&so->so_lock);
5077 	/*
5078 	 * If there is an error record the error in so_error put don't fail
5079 	 * the getsockname. Instead fallback on the recorded
5080 	 * sti->sti_laddr_sa.
5081 	 */
5082 	if (error) {
5083 		/*
5084 		 * Various stream head errors can be returned to the ioctl.
5085 		 * However, it is impossible to determine which ones of
5086 		 * these are really socket level errors that were incorrectly
5087 		 * consumed by the ioctl. Thus this code silently ignores the
5088 		 * error - to code explicitly does not reinstate the error
5089 		 * using soseterror().
5090 		 * Experiments have shows that at least this set of
5091 		 * errors are reported and should not be reinstated on the
5092 		 * socket:
5093 		 *	EINVAL	E.g. if an I_LINK was in effect when
5094 		 *		getsockname was called.
5095 		 *	EPIPE	The ioctl error semantics prefer the write
5096 		 *		side error over the read side error.
5097 		 */
5098 		error = 0;
5099 	} else if (res == 0 && strbuf.len > 0 &&
5100 	    (so->so_state & SS_ISBOUND)) {
5101 		ASSERT(strbuf.len <= (int)sti->sti_laddr_maxlen);
5102 		sti->sti_laddr_len = (socklen_t)strbuf.len;
5103 		bcopy(addr, sti->sti_laddr_sa, sti->sti_laddr_len);
5104 		sti->sti_laddr_valid = 1;
5105 
5106 		bcopy(addr, name, MIN(sti->sti_laddr_len, *namelen));
5107 		*namelen = sti->sti_laddr_len;
5108 	}
5109 	kmem_free(addr, addrlen);
5110 #ifdef DEBUG
5111 	dprintso(so, 1, ("sotpi_getsockname (tp): %s\n",
5112 	    pr_addr(so->so_family, sti->sti_laddr_sa,
5113 	    (t_uscalar_t)sti->sti_laddr_len)));
5114 #endif /* DEBUG */
5115 done:
5116 	so_unlock_single(so, SOLOCKED);
5117 	mutex_exit(&so->so_lock);
5118 	return (error);
5119 }
5120 
5121 /*
5122  * Get socket options. For SOL_SOCKET options some options are handled
5123  * by the sockfs while others use the value recorded in the sonode as a
5124  * fallback should the T_SVR4_OPTMGMT_REQ fail.
5125  *
5126  * On the return most *optlenp bytes are copied to optval.
5127  */
5128 /* ARGSUSED */
5129 int
5130 sotpi_getsockopt(struct sonode *so, int level, int option_name,
5131 		void *optval, socklen_t *optlenp, int flags, struct cred *cr)
5132 {
5133 	struct T_optmgmt_req	optmgmt_req;
5134 	struct T_optmgmt_ack	*optmgmt_ack;
5135 	struct opthdr		oh;
5136 	struct opthdr		*opt_res;
5137 	mblk_t			*mp = NULL;
5138 	int			error = 0;
5139 	void			*option = NULL;	/* Set if fallback value */
5140 	t_uscalar_t		maxlen = *optlenp;
5141 	t_uscalar_t		len;
5142 	uint32_t		value;
5143 	struct timeval		tmo_val; /* used for SO_RCVTIMEO, SO_SNDTIMEO */
5144 	struct timeval32	tmo_val32;
5145 	struct so_snd_bufinfo	snd_bufinfo;	/* used for zero copy */
5146 
5147 	dprintso(so, 1, ("sotpi_getsockopt(%p, 0x%x, 0x%x, %p, %p) %s\n",
5148 	    (void *)so, level, option_name, optval, (void *)optlenp,
5149 	    pr_state(so->so_state, so->so_mode)));
5150 
5151 	mutex_enter(&so->so_lock);
5152 	so_lock_single(so);	/* Set SOLOCKED */
5153 
5154 	/*
5155 	 * Check for SOL_SOCKET options.
5156 	 * Certain SOL_SOCKET options are returned directly whereas
5157 	 * others only provide a default (fallback) value should
5158 	 * the T_SVR4_OPTMGMT_REQ fail.
5159 	 */
5160 	if (level == SOL_SOCKET) {
5161 		/* Check parameters */
5162 		switch (option_name) {
5163 		case SO_TYPE:
5164 		case SO_ERROR:
5165 		case SO_DEBUG:
5166 		case SO_ACCEPTCONN:
5167 		case SO_REUSEADDR:
5168 		case SO_KEEPALIVE:
5169 		case SO_DONTROUTE:
5170 		case SO_BROADCAST:
5171 		case SO_USELOOPBACK:
5172 		case SO_OOBINLINE:
5173 		case SO_SNDBUF:
5174 		case SO_RCVBUF:
5175 #ifdef notyet
5176 		case SO_SNDLOWAT:
5177 		case SO_RCVLOWAT:
5178 #endif /* notyet */
5179 		case SO_DOMAIN:
5180 		case SO_DGRAM_ERRIND:
5181 			if (maxlen < (t_uscalar_t)sizeof (int32_t)) {
5182 				error = EINVAL;
5183 				eprintsoline(so, error);
5184 				goto done2;
5185 			}
5186 			break;
5187 		case SO_RCVTIMEO:
5188 		case SO_SNDTIMEO:
5189 			if (get_udatamodel() == DATAMODEL_NONE ||
5190 			    get_udatamodel() == DATAMODEL_NATIVE) {
5191 				if (maxlen < sizeof (struct timeval)) {
5192 					error = EINVAL;
5193 					eprintsoline(so, error);
5194 					goto done2;
5195 				}
5196 			} else {
5197 				if (maxlen < sizeof (struct timeval32)) {
5198 					error = EINVAL;
5199 					eprintsoline(so, error);
5200 					goto done2;
5201 				}
5202 
5203 			}
5204 			break;
5205 		case SO_LINGER:
5206 			if (maxlen < (t_uscalar_t)sizeof (struct linger)) {
5207 				error = EINVAL;
5208 				eprintsoline(so, error);
5209 				goto done2;
5210 			}
5211 			break;
5212 		case SO_SND_BUFINFO:
5213 			if (maxlen < (t_uscalar_t)
5214 			    sizeof (struct so_snd_bufinfo)) {
5215 				error = EINVAL;
5216 				eprintsoline(so, error);
5217 				goto done2;
5218 			}
5219 			break;
5220 		}
5221 
5222 		len = (t_uscalar_t)sizeof (uint32_t);	/* Default */
5223 
5224 		switch (option_name) {
5225 		case SO_TYPE:
5226 			value = so->so_type;
5227 			option = &value;
5228 			goto copyout; /* No need to issue T_SVR4_OPTMGMT_REQ */
5229 
5230 		case SO_ERROR:
5231 			value = sogeterr(so, B_TRUE);
5232 			option = &value;
5233 			goto copyout; /* No need to issue T_SVR4_OPTMGMT_REQ */
5234 
5235 		case SO_ACCEPTCONN:
5236 			if (so->so_state & SS_ACCEPTCONN)
5237 				value = SO_ACCEPTCONN;
5238 			else
5239 				value = 0;
5240 #ifdef DEBUG
5241 			if (value) {
5242 				dprintso(so, 1,
5243 				    ("sotpi_getsockopt: 0x%x is set\n",
5244 				    option_name));
5245 			} else {
5246 				dprintso(so, 1,
5247 				    ("sotpi_getsockopt: 0x%x not set\n",
5248 				    option_name));
5249 			}
5250 #endif /* DEBUG */
5251 			option = &value;
5252 			goto copyout; /* No need to issue T_SVR4_OPTMGMT_REQ */
5253 
5254 		case SO_DEBUG:
5255 		case SO_REUSEADDR:
5256 		case SO_KEEPALIVE:
5257 		case SO_DONTROUTE:
5258 		case SO_BROADCAST:
5259 		case SO_USELOOPBACK:
5260 		case SO_OOBINLINE:
5261 		case SO_DGRAM_ERRIND:
5262 			value = (so->so_options & option_name);
5263 #ifdef DEBUG
5264 			if (value) {
5265 				dprintso(so, 1,
5266 				    ("sotpi_getsockopt: 0x%x is set\n",
5267 				    option_name));
5268 			} else {
5269 				dprintso(so, 1,
5270 				    ("sotpi_getsockopt: 0x%x not set\n",
5271 				    option_name));
5272 			}
5273 #endif /* DEBUG */
5274 			option = &value;
5275 			goto copyout; /* No need to issue T_SVR4_OPTMGMT_REQ */
5276 
5277 		/*
5278 		 * The following options are only returned by sockfs when the
5279 		 * T_SVR4_OPTMGMT_REQ fails.
5280 		 */
5281 		case SO_LINGER:
5282 			option = &so->so_linger;
5283 			len = (t_uscalar_t)sizeof (struct linger);
5284 			break;
5285 		case SO_SNDBUF: {
5286 			ssize_t lvalue;
5287 
5288 			/*
5289 			 * If the option has not been set then get a default
5290 			 * value from the read queue. This value is
5291 			 * returned if the transport fails
5292 			 * the T_SVR4_OPTMGMT_REQ.
5293 			 */
5294 			lvalue = so->so_sndbuf;
5295 			if (lvalue == 0) {
5296 				mutex_exit(&so->so_lock);
5297 				(void) strqget(strvp2wq(SOTOV(so))->q_next,
5298 				    QHIWAT, 0, &lvalue);
5299 				mutex_enter(&so->so_lock);
5300 				dprintso(so, 1,
5301 				    ("got SO_SNDBUF %ld from q\n", lvalue));
5302 			}
5303 			value = (int)lvalue;
5304 			option = &value;
5305 			len = (t_uscalar_t)sizeof (so->so_sndbuf);
5306 			break;
5307 		}
5308 		case SO_RCVBUF: {
5309 			ssize_t lvalue;
5310 
5311 			/*
5312 			 * If the option has not been set then get a default
5313 			 * value from the read queue. This value is
5314 			 * returned if the transport fails
5315 			 * the T_SVR4_OPTMGMT_REQ.
5316 			 *
5317 			 * XXX If SO_RCVBUF has been set and this is an
5318 			 * XPG 4.2 application then do not ask the transport
5319 			 * since the transport might adjust the value and not
5320 			 * return exactly what was set by the application.
5321 			 * For non-XPG 4.2 application we return the value
5322 			 * that the transport is actually using.
5323 			 */
5324 			lvalue = so->so_rcvbuf;
5325 			if (lvalue == 0) {
5326 				mutex_exit(&so->so_lock);
5327 				(void) strqget(RD(strvp2wq(SOTOV(so))),
5328 				    QHIWAT, 0, &lvalue);
5329 				mutex_enter(&so->so_lock);
5330 				dprintso(so, 1,
5331 				    ("got SO_RCVBUF %ld from q\n", lvalue));
5332 			} else if (flags & _SOGETSOCKOPT_XPG4_2) {
5333 				value = (int)lvalue;
5334 				option = &value;
5335 				goto copyout;	/* skip asking transport */
5336 			}
5337 			value = (int)lvalue;
5338 			option = &value;
5339 			len = (t_uscalar_t)sizeof (so->so_rcvbuf);
5340 			break;
5341 		}
5342 		case SO_DOMAIN:
5343 			value = so->so_family;
5344 			option = &value;
5345 			goto copyout; /* No need to issue T_SVR4_OPTMGMT_REQ */
5346 
5347 #ifdef notyet
5348 		/*
5349 		 * We do not implement the semantics of these options
5350 		 * thus we shouldn't implement the options either.
5351 		 */
5352 		case SO_SNDLOWAT:
5353 			value = so->so_sndlowat;
5354 			option = &value;
5355 			break;
5356 		case SO_RCVLOWAT:
5357 			value = so->so_rcvlowat;
5358 			option = &value;
5359 			break;
5360 #endif /* notyet */
5361 		case SO_SNDTIMEO:
5362 		case SO_RCVTIMEO: {
5363 			clock_t val;
5364 
5365 			if (option_name == SO_RCVTIMEO)
5366 				val = drv_hztousec(so->so_rcvtimeo);
5367 			else
5368 				val = drv_hztousec(so->so_sndtimeo);
5369 			tmo_val.tv_sec = val / (1000 * 1000);
5370 			tmo_val.tv_usec = val % (1000 * 1000);
5371 			if (get_udatamodel() == DATAMODEL_NONE ||
5372 			    get_udatamodel() == DATAMODEL_NATIVE) {
5373 				option = &tmo_val;
5374 				len = sizeof (struct timeval);
5375 			} else {
5376 				TIMEVAL_TO_TIMEVAL32(&tmo_val32, &tmo_val);
5377 				option = &tmo_val32;
5378 				len = sizeof (struct timeval32);
5379 			}
5380 			break;
5381 		}
5382 		case SO_SND_BUFINFO: {
5383 			snd_bufinfo.sbi_wroff =
5384 			    (so->so_proto_props).sopp_wroff;
5385 			snd_bufinfo.sbi_maxblk =
5386 			    (so->so_proto_props).sopp_maxblk;
5387 			snd_bufinfo.sbi_maxpsz =
5388 			    (so->so_proto_props).sopp_maxpsz;
5389 			snd_bufinfo.sbi_tail =
5390 			    (so->so_proto_props).sopp_tail;
5391 			option = &snd_bufinfo;
5392 			len = (t_uscalar_t)sizeof (struct so_snd_bufinfo);
5393 			break;
5394 		}
5395 		}
5396 	}
5397 
5398 	mutex_exit(&so->so_lock);
5399 
5400 	/* Send request */
5401 	optmgmt_req.PRIM_type = T_SVR4_OPTMGMT_REQ;
5402 	optmgmt_req.MGMT_flags = T_CHECK;
5403 	optmgmt_req.OPT_length = (t_scalar_t)(sizeof (oh) + maxlen);
5404 	optmgmt_req.OPT_offset = (t_scalar_t)sizeof (optmgmt_req);
5405 
5406 	oh.level = level;
5407 	oh.name = option_name;
5408 	oh.len = maxlen;
5409 
5410 	mp = soallocproto3(&optmgmt_req, sizeof (optmgmt_req),
5411 	    &oh, sizeof (oh), NULL, maxlen, 0, _ALLOC_SLEEP, cr);
5412 	/* Let option management work in the presence of data flow control */
5413 	error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0,
5414 	    MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR|MSG_IGNFLOW, 0);
5415 	mp = NULL;
5416 	mutex_enter(&so->so_lock);
5417 	if (error) {
5418 		eprintsoline(so, error);
5419 		goto done2;
5420 	}
5421 	error = sowaitprim(so, T_SVR4_OPTMGMT_REQ, T_OPTMGMT_ACK,
5422 	    (t_uscalar_t)(sizeof (*optmgmt_ack) + sizeof (*opt_res)), &mp, 0);
5423 	if (error) {
5424 		if (option != NULL) {
5425 			/* We have a fallback value */
5426 			error = 0;
5427 			goto copyout;
5428 		}
5429 		eprintsoline(so, error);
5430 		goto done2;
5431 	}
5432 	ASSERT(mp);
5433 	optmgmt_ack = (struct T_optmgmt_ack *)mp->b_rptr;
5434 	opt_res = (struct opthdr *)sogetoff(mp, optmgmt_ack->OPT_offset,
5435 	    optmgmt_ack->OPT_length, __TPI_ALIGN_SIZE);
5436 	if (opt_res == NULL) {
5437 		if (option != NULL) {
5438 			/* We have a fallback value */
5439 			error = 0;
5440 			goto copyout;
5441 		}
5442 		error = EPROTO;
5443 		eprintsoline(so, error);
5444 		goto done;
5445 	}
5446 	option = &opt_res[1];
5447 
5448 	/* check to ensure that the option is within bounds */
5449 	if (((uintptr_t)option + opt_res->len < (uintptr_t)option) ||
5450 	    (uintptr_t)option + opt_res->len > (uintptr_t)mp->b_wptr) {
5451 		if (option != NULL) {
5452 			/* We have a fallback value */
5453 			error = 0;
5454 			goto copyout;
5455 		}
5456 		error = EPROTO;
5457 		eprintsoline(so, error);
5458 		goto done;
5459 	}
5460 
5461 	len = opt_res->len;
5462 
5463 copyout: {
5464 		t_uscalar_t size = MIN(len, maxlen);
5465 		bcopy(option, optval, size);
5466 		bcopy(&size, optlenp, sizeof (size));
5467 	}
5468 done:
5469 	freemsg(mp);
5470 done2:
5471 	so_unlock_single(so, SOLOCKED);
5472 	mutex_exit(&so->so_lock);
5473 
5474 	return (error);
5475 }
5476 
5477 /*
5478  * Set socket options. All options are passed down in a T_SVR4_OPTMGMT_REQ.
5479  * SOL_SOCKET options are also recorded in the sonode. A setsockopt for
5480  * SOL_SOCKET options will not fail just because the T_SVR4_OPTMGMT_REQ fails -
5481  * setsockopt has to work even if the transport does not support the option.
5482  */
5483 /* ARGSUSED */
5484 int
5485 sotpi_setsockopt(struct sonode *so, int level, int option_name,
5486 	const void *optval, t_uscalar_t optlen, struct cred *cr)
5487 {
5488 	struct T_optmgmt_req	optmgmt_req;
5489 	struct opthdr		oh;
5490 	mblk_t			*mp;
5491 	int			error = 0;
5492 	boolean_t		handled = B_FALSE;
5493 
5494 	dprintso(so, 1, ("sotpi_setsockopt(%p, 0x%x, 0x%x, %p, %d) %s\n",
5495 	    (void *)so, level, option_name, optval, optlen,
5496 	    pr_state(so->so_state, so->so_mode)));
5497 
5498 	/* X/Open requires this check */
5499 	if ((so->so_state & SS_CANTSENDMORE) && !xnet_skip_checks) {
5500 		if (xnet_check_print)
5501 			printf("sockfs: X/Open setsockopt check => EINVAL\n");
5502 		return (EINVAL);
5503 	}
5504 
5505 	mutex_enter(&so->so_lock);
5506 	so_lock_single(so);	/* Set SOLOCKED */
5507 	mutex_exit(&so->so_lock);
5508 
5509 	optmgmt_req.PRIM_type = T_SVR4_OPTMGMT_REQ;
5510 	optmgmt_req.MGMT_flags = T_NEGOTIATE;
5511 	optmgmt_req.OPT_length = (t_scalar_t)sizeof (oh) + optlen;
5512 	optmgmt_req.OPT_offset = (t_scalar_t)sizeof (optmgmt_req);
5513 
5514 	oh.level = level;
5515 	oh.name = option_name;
5516 	oh.len = optlen;
5517 
5518 	mp = soallocproto3(&optmgmt_req, sizeof (optmgmt_req),
5519 	    &oh, sizeof (oh), optval, optlen, 0, _ALLOC_SLEEP, cr);
5520 	/* Let option management work in the presence of data flow control */
5521 	error = kstrputmsg(SOTOV(so), mp, NULL, 0, 0,
5522 	    MSG_BAND|MSG_HOLDSIG|MSG_IGNERROR|MSG_IGNFLOW, 0);
5523 	mp = NULL;
5524 	mutex_enter(&so->so_lock);
5525 	if (error) {
5526 		eprintsoline(so, error);
5527 		goto done2;
5528 	}
5529 	error = sowaitprim(so, T_SVR4_OPTMGMT_REQ, T_OPTMGMT_ACK,
5530 	    (t_uscalar_t)sizeof (struct T_optmgmt_ack), &mp, 0);
5531 	if (error) {
5532 		eprintsoline(so, error);
5533 		goto done;
5534 	}
5535 	ASSERT(mp);
5536 	/* No need to verify T_optmgmt_ack */
5537 	freemsg(mp);
5538 done:
5539 	/*
5540 	 * Check for SOL_SOCKET options and record their values.
5541 	 * If we know about a SOL_SOCKET parameter and the transport
5542 	 * failed it with TBADOPT or TOUTSTATE (i.e. ENOPROTOOPT or
5543 	 * EPROTO) we let the setsockopt succeed.
5544 	 */
5545 	if (level == SOL_SOCKET) {
5546 		/* Check parameters */
5547 		switch (option_name) {
5548 		case SO_DEBUG:
5549 		case SO_REUSEADDR:
5550 		case SO_KEEPALIVE:
5551 		case SO_DONTROUTE:
5552 		case SO_BROADCAST:
5553 		case SO_USELOOPBACK:
5554 		case SO_OOBINLINE:
5555 		case SO_SNDBUF:
5556 		case SO_RCVBUF:
5557 #ifdef notyet
5558 		case SO_SNDLOWAT:
5559 		case SO_RCVLOWAT:
5560 #endif /* notyet */
5561 		case SO_DGRAM_ERRIND:
5562 			if (optlen != (t_uscalar_t)sizeof (int32_t)) {
5563 				error = EINVAL;
5564 				eprintsoline(so, error);
5565 				goto done2;
5566 			}
5567 			ASSERT(optval);
5568 			handled = B_TRUE;
5569 			break;
5570 		case SO_SNDTIMEO:
5571 		case SO_RCVTIMEO:
5572 			if (get_udatamodel() == DATAMODEL_NONE ||
5573 			    get_udatamodel() == DATAMODEL_NATIVE) {
5574 				if (optlen != sizeof (struct timeval)) {
5575 					error = EINVAL;
5576 					eprintsoline(so, error);
5577 					goto done2;
5578 				}
5579 			} else {
5580 				if (optlen != sizeof (struct timeval32)) {
5581 					error = EINVAL;
5582 					eprintsoline(so, error);
5583 					goto done2;
5584 				}
5585 			}
5586 			ASSERT(optval);
5587 			handled = B_TRUE;
5588 			break;
5589 		case SO_LINGER:
5590 			if (optlen != (t_uscalar_t)sizeof (struct linger)) {
5591 				error = EINVAL;
5592 				eprintsoline(so, error);
5593 				goto done2;
5594 			}
5595 			ASSERT(optval);
5596 			handled = B_TRUE;
5597 			break;
5598 		}
5599 
5600 #define	intvalue	(*(int32_t *)optval)
5601 
5602 		switch (option_name) {
5603 		case SO_TYPE:
5604 		case SO_ERROR:
5605 		case SO_ACCEPTCONN:
5606 			/* Can't be set */
5607 			error = ENOPROTOOPT;
5608 			goto done2;
5609 		case SO_LINGER: {
5610 			struct linger *l = (struct linger *)optval;
5611 
5612 			so->so_linger.l_linger = l->l_linger;
5613 			if (l->l_onoff) {
5614 				so->so_linger.l_onoff = SO_LINGER;
5615 				so->so_options |= SO_LINGER;
5616 			} else {
5617 				so->so_linger.l_onoff = 0;
5618 				so->so_options &= ~SO_LINGER;
5619 			}
5620 			break;
5621 		}
5622 
5623 		case SO_DEBUG:
5624 #ifdef SOCK_TEST
5625 			if (intvalue & 2)
5626 				sock_test_timelimit = 10 * hz;
5627 			else
5628 				sock_test_timelimit = 0;
5629 
5630 			if (intvalue & 4)
5631 				do_useracc = 0;
5632 			else
5633 				do_useracc = 1;
5634 #endif /* SOCK_TEST */
5635 			/* FALLTHRU */
5636 		case SO_REUSEADDR:
5637 		case SO_KEEPALIVE:
5638 		case SO_DONTROUTE:
5639 		case SO_BROADCAST:
5640 		case SO_USELOOPBACK:
5641 		case SO_OOBINLINE:
5642 		case SO_DGRAM_ERRIND:
5643 			if (intvalue != 0) {
5644 				dprintso(so, 1,
5645 				    ("socket_setsockopt: setting 0x%x\n",
5646 				    option_name));
5647 				so->so_options |= option_name;
5648 			} else {
5649 				dprintso(so, 1,
5650 				    ("socket_setsockopt: clearing 0x%x\n",
5651 				    option_name));
5652 				so->so_options &= ~option_name;
5653 			}
5654 			break;
5655 		/*
5656 		 * The following options are only returned by us when the
5657 		 * transport layer fails.
5658 		 * XXX XPG 4.2 applications retrieve SO_RCVBUF from sockfs
5659 		 * since the transport might adjust the value and not
5660 		 * return exactly what was set by the application.
5661 		 */
5662 		case SO_SNDBUF:
5663 			so->so_sndbuf = intvalue;
5664 			break;
5665 		case SO_RCVBUF:
5666 			so->so_rcvbuf = intvalue;
5667 			break;
5668 		case SO_RCVPSH:
5669 			so->so_rcv_timer_interval = intvalue;
5670 			break;
5671 #ifdef notyet
5672 		/*
5673 		 * We do not implement the semantics of these options
5674 		 * thus we shouldn't implement the options either.
5675 		 */
5676 		case SO_SNDLOWAT:
5677 			so->so_sndlowat = intvalue;
5678 			break;
5679 		case SO_RCVLOWAT:
5680 			so->so_rcvlowat = intvalue;
5681 			break;
5682 #endif /* notyet */
5683 		case SO_SNDTIMEO:
5684 		case SO_RCVTIMEO: {
5685 			struct timeval tl;
5686 			clock_t val;
5687 
5688 			if (get_udatamodel() == DATAMODEL_NONE ||
5689 			    get_udatamodel() == DATAMODEL_NATIVE)
5690 				bcopy(&tl, (struct timeval *)optval,
5691 				    sizeof (struct timeval));
5692 			else
5693 				TIMEVAL32_TO_TIMEVAL(&tl,
5694 				    (struct timeval32 *)optval);
5695 			val = tl.tv_sec * 1000 * 1000 + tl.tv_usec;
5696 			if (option_name == SO_RCVTIMEO)
5697 				so->so_rcvtimeo = drv_usectohz(val);
5698 			else
5699 				so->so_sndtimeo = drv_usectohz(val);
5700 			break;
5701 		}
5702 		}
5703 #undef	intvalue
5704 
5705 		if (error) {
5706 			if ((error == ENOPROTOOPT || error == EPROTO ||
5707 			    error == EINVAL) && handled) {
5708 				dprintso(so, 1,
5709 				    ("setsockopt: ignoring error %d for 0x%x\n",
5710 				    error, option_name));
5711 				error = 0;
5712 			}
5713 		}
5714 	}
5715 done2:
5716 	so_unlock_single(so, SOLOCKED);
5717 	mutex_exit(&so->so_lock);
5718 	return (error);
5719 }
5720 
5721 /*
5722  * sotpi_close() is called when the last open reference goes away.
5723  */
5724 /* ARGSUSED */
5725 int
5726 sotpi_close(struct sonode *so, int flag, struct cred *cr)
5727 {
5728 	struct vnode *vp = SOTOV(so);
5729 	dev_t dev;
5730 	int error = 0;
5731 	sotpi_info_t *sti = SOTOTPI(so);
5732 
5733 	dprintso(so, 1, ("sotpi_close(%p, %x) %s\n",
5734 	    (void *)vp, flag, pr_state(so->so_state, so->so_mode)));
5735 
5736 	dev = sti->sti_dev;
5737 
5738 	ASSERT(STREAMSTAB(getmajor(dev)));
5739 
5740 	mutex_enter(&so->so_lock);
5741 	so_lock_single(so);	/* Set SOLOCKED */
5742 
5743 	ASSERT(so_verify_oobstate(so));
5744 
5745 	if (sti->sti_nl7c_flags & NL7C_ENABLED) {
5746 		sti->sti_nl7c_flags = 0;
5747 		nl7c_close(so);
5748 	}
5749 
5750 	if (vp->v_stream != NULL) {
5751 		vnode_t *ux_vp;
5752 
5753 		if (so->so_family == AF_UNIX) {
5754 			/* Could avoid this when CANTSENDMORE for !dgram */
5755 			so_unix_close(so);
5756 		}
5757 
5758 		mutex_exit(&so->so_lock);
5759 		/*
5760 		 * Disassemble the linkage from the AF_UNIX underlying file
5761 		 * system vnode to this socket (by atomically clearing
5762 		 * v_stream in vn_rele_stream) before strclose clears sd_vnode
5763 		 * and frees the stream head.
5764 		 */
5765 		if ((ux_vp = sti->sti_ux_bound_vp) != NULL) {
5766 			ASSERT(ux_vp->v_stream);
5767 			sti->sti_ux_bound_vp = NULL;
5768 			vn_rele_stream(ux_vp);
5769 		}
5770 		if (so->so_family == AF_INET || so->so_family == AF_INET6) {
5771 			strsetrwputdatahooks(SOTOV(so), NULL, NULL);
5772 			if (sti->sti_kssl_ent != NULL) {
5773 				kssl_release_ent(sti->sti_kssl_ent, so,
5774 				    sti->sti_kssl_type);
5775 				sti->sti_kssl_ent = NULL;
5776 			}
5777 			if (sti->sti_kssl_ctx != NULL) {
5778 				kssl_release_ctx(sti->sti_kssl_ctx);
5779 				sti->sti_kssl_ctx = NULL;
5780 			}
5781 			sti->sti_kssl_type = KSSL_NO_PROXY;
5782 		}
5783 		error = strclose(vp, flag, cr);
5784 		vp->v_stream = NULL;
5785 		mutex_enter(&so->so_lock);
5786 	}
5787 
5788 	/*
5789 	 * Flush the T_DISCON_IND on sti_discon_ind_mp.
5790 	 */
5791 	so_flush_discon_ind(so);
5792 
5793 	so_unlock_single(so, SOLOCKED);
5794 	mutex_exit(&so->so_lock);
5795 
5796 	/*
5797 	 * Needed for STREAMs.
5798 	 * Decrement the device driver's reference count for streams
5799 	 * opened via the clone dip. The driver was held in clone_open().
5800 	 * The absence of clone_close() forces this asymmetry.
5801 	 */
5802 	if (so->so_flag & SOCLONE)
5803 		ddi_rele_driver(getmajor(dev));
5804 
5805 	return (error);
5806 }
5807 
5808 static int
5809 sotpi_ioctl(struct sonode *so, int cmd, intptr_t arg, int mode,
5810     struct cred *cr, int32_t *rvalp)
5811 {
5812 	struct vnode *vp = SOTOV(so);
5813 	sotpi_info_t *sti = SOTOTPI(so);
5814 	int error = 0;
5815 
5816 	dprintso(so, 0, ("sotpi_ioctl: cmd 0x%x, arg 0x%lx, state %s\n",
5817 	    cmd, arg, pr_state(so->so_state, so->so_mode)));
5818 
5819 	switch (cmd) {
5820 	case SIOCSQPTR:
5821 		/*
5822 		 * SIOCSQPTR is valid only when helper stream is created
5823 		 * by the protocol.
5824 		 */
5825 	case _I_INSERT:
5826 	case _I_REMOVE:
5827 		/*
5828 		 * Since there's no compelling reason to support these ioctls
5829 		 * on sockets, and doing so would increase the complexity
5830 		 * markedly, prevent it.
5831 		 */
5832 		return (EOPNOTSUPP);
5833 
5834 	case I_FIND:
5835 	case I_LIST:
5836 	case I_LOOK:
5837 	case I_POP:
5838 	case I_PUSH:
5839 		/*
5840 		 * To prevent races and inconsistencies between the actual
5841 		 * state of the stream and the state according to the sonode,
5842 		 * we serialize all operations which modify or operate on the
5843 		 * list of modules on the socket's stream.
5844 		 */
5845 		mutex_enter(&sti->sti_plumb_lock);
5846 		error = socktpi_plumbioctl(vp, cmd, arg, mode, cr, rvalp);
5847 		mutex_exit(&sti->sti_plumb_lock);
5848 		return (error);
5849 
5850 	default:
5851 		if (so->so_version != SOV_STREAM)
5852 			break;
5853 
5854 		/*
5855 		 * The imaginary "sockmod" has been popped; act as a stream.
5856 		 */
5857 		return (strioctl(vp, cmd, arg, mode, U_TO_K, cr, rvalp));
5858 	}
5859 
5860 	ASSERT(so->so_version != SOV_STREAM);
5861 
5862 	/*
5863 	 * Process socket-specific ioctls.
5864 	 */
5865 	switch (cmd) {
5866 	case FIONBIO: {
5867 		int32_t value;
5868 
5869 		if (so_copyin((void *)arg, &value, sizeof (int32_t),
5870 		    (mode & (int)FKIOCTL)))
5871 			return (EFAULT);
5872 
5873 		mutex_enter(&so->so_lock);
5874 		if (value) {
5875 			so->so_state |= SS_NDELAY;
5876 		} else {
5877 			so->so_state &= ~SS_NDELAY;
5878 		}
5879 		mutex_exit(&so->so_lock);
5880 		return (0);
5881 	}
5882 
5883 	case FIOASYNC: {
5884 		int32_t value;
5885 
5886 		if (so_copyin((void *)arg, &value, sizeof (int32_t),
5887 		    (mode & (int)FKIOCTL)))
5888 			return (EFAULT);
5889 
5890 		mutex_enter(&so->so_lock);
5891 		/*
5892 		 * SS_ASYNC flag not already set correctly?
5893 		 * (!value != !(so->so_state & SS_ASYNC))
5894 		 * but some engineers find that too hard to read.
5895 		 */
5896 		if (value == 0 && (so->so_state & SS_ASYNC) != 0 ||
5897 		    value != 0 && (so->so_state & SS_ASYNC) == 0)
5898 			error = so_flip_async(so, vp, mode, cr);
5899 		mutex_exit(&so->so_lock);
5900 		return (error);
5901 	}
5902 
5903 	case SIOCSPGRP:
5904 	case FIOSETOWN: {
5905 		pid_t pgrp;
5906 
5907 		if (so_copyin((void *)arg, &pgrp, sizeof (pid_t),
5908 		    (mode & (int)FKIOCTL)))
5909 			return (EFAULT);
5910 
5911 		mutex_enter(&so->so_lock);
5912 		dprintso(so, 1, ("setown: new %d old %d\n", pgrp, so->so_pgrp));
5913 		/* Any change? */
5914 		if (pgrp != so->so_pgrp)
5915 			error = so_set_siggrp(so, vp, pgrp, mode, cr);
5916 		mutex_exit(&so->so_lock);
5917 		return (error);
5918 	}
5919 	case SIOCGPGRP:
5920 	case FIOGETOWN:
5921 		if (so_copyout(&so->so_pgrp, (void *)arg,
5922 		    sizeof (pid_t), (mode & (int)FKIOCTL)))
5923 			return (EFAULT);
5924 		return (0);
5925 
5926 	case SIOCATMARK: {
5927 		int retval;
5928 		uint_t so_state;
5929 
5930 		/*
5931 		 * strwaitmark has a finite timeout after which it
5932 		 * returns -1 if the mark state is undetermined.
5933 		 * In order to avoid any race between the mark state
5934 		 * in sockfs and the mark state in the stream head this
5935 		 * routine loops until the mark state can be determined
5936 		 * (or the urgent data indication has been removed by some
5937 		 * other thread).
5938 		 */
5939 		do {
5940 			mutex_enter(&so->so_lock);
5941 			so_state = so->so_state;
5942 			mutex_exit(&so->so_lock);
5943 			if (so_state & SS_RCVATMARK) {
5944 				retval = 1;
5945 			} else if (!(so_state & SS_OOBPEND)) {
5946 				/*
5947 				 * No SIGURG has been generated -- there is no
5948 				 * pending or present urgent data. Thus can't
5949 				 * possibly be at the mark.
5950 				 */
5951 				retval = 0;
5952 			} else {
5953 				/*
5954 				 * Have the stream head wait until there is
5955 				 * either some messages on the read queue, or
5956 				 * STRATMARK or STRNOTATMARK gets set. The
5957 				 * STRNOTATMARK flag is used so that the
5958 				 * transport can send up a MSGNOTMARKNEXT
5959 				 * M_DATA to indicate that it is not
5960 				 * at the mark and additional data is not about
5961 				 * to be send upstream.
5962 				 *
5963 				 * If the mark state is undetermined this will
5964 				 * return -1 and we will loop rechecking the
5965 				 * socket state.
5966 				 */
5967 				retval = strwaitmark(vp);
5968 			}
5969 		} while (retval == -1);
5970 
5971 		if (so_copyout(&retval, (void *)arg, sizeof (int),
5972 		    (mode & (int)FKIOCTL)))
5973 			return (EFAULT);
5974 		return (0);
5975 	}
5976 
5977 	case I_FDINSERT:
5978 	case I_SENDFD:
5979 	case I_RECVFD:
5980 	case I_ATMARK:
5981 	case _SIOCSOCKFALLBACK:
5982 		/*
5983 		 * These ioctls do not apply to sockets. I_FDINSERT can be
5984 		 * used to send M_PROTO messages without modifying the socket
5985 		 * state. I_SENDFD/RECVFD should not be used for socket file
5986 		 * descriptor passing since they assume a twisted stream.
5987 		 * SIOCATMARK must be used instead of I_ATMARK.
5988 		 *
5989 		 * _SIOCSOCKFALLBACK from an application should never be
5990 		 * processed.  It is only generated by socktpi_open() or
5991 		 * in response to I_POP or I_PUSH.
5992 		 */
5993 #ifdef DEBUG
5994 		zcmn_err(getzoneid(), CE_WARN,
5995 		    "Unsupported STREAMS ioctl 0x%x on socket. "
5996 		    "Pid = %d\n", cmd, curproc->p_pid);
5997 #endif /* DEBUG */
5998 		return (EOPNOTSUPP);
5999 
6000 	case _I_GETPEERCRED:
6001 		if ((mode & FKIOCTL) == 0)
6002 			return (EINVAL);
6003 
6004 		mutex_enter(&so->so_lock);
6005 		if ((so->so_mode & SM_CONNREQUIRED) == 0) {
6006 			error = ENOTSUP;
6007 		} else if ((so->so_state & SS_ISCONNECTED) == 0) {
6008 			error = ENOTCONN;
6009 		} else if (so->so_peercred != NULL) {
6010 			k_peercred_t *kp = (k_peercred_t *)arg;
6011 			kp->pc_cr = so->so_peercred;
6012 			kp->pc_cpid = so->so_cpid;
6013 			crhold(so->so_peercred);
6014 		} else {
6015 			error = EINVAL;
6016 		}
6017 		mutex_exit(&so->so_lock);
6018 		return (error);
6019 
6020 	default:
6021 		/*
6022 		 * Do the higher-order bits of the ioctl cmd indicate
6023 		 * that it is an I_* streams ioctl?
6024 		 */
6025 		if ((cmd & 0xffffff00U) == STR &&
6026 		    so->so_version == SOV_SOCKBSD) {
6027 #ifdef DEBUG
6028 			zcmn_err(getzoneid(), CE_WARN,
6029 			    "Unsupported STREAMS ioctl 0x%x on socket. "
6030 			    "Pid = %d\n", cmd, 	curproc->p_pid);
6031 #endif /* DEBUG */
6032 			return (EOPNOTSUPP);
6033 		}
6034 		return (strioctl(vp, cmd, arg, mode, U_TO_K, cr, rvalp));
6035 	}
6036 }
6037 
6038 /*
6039  * Handle plumbing-related ioctls.
6040  */
6041 static int
6042 socktpi_plumbioctl(struct vnode *vp, int cmd, intptr_t arg, int mode,
6043     struct cred *cr, int32_t *rvalp)
6044 {
6045 	static const char sockmod_name[] = "sockmod";
6046 	struct sonode	*so = VTOSO(vp);
6047 	char		mname[FMNAMESZ + 1];
6048 	int		error;
6049 	sotpi_info_t	*sti = SOTOTPI(so);
6050 
6051 	ASSERT(MUTEX_HELD(&sti->sti_plumb_lock));
6052 
6053 	if (so->so_version == SOV_SOCKBSD)
6054 		return (EOPNOTSUPP);
6055 
6056 	if (so->so_version == SOV_STREAM) {
6057 		/*
6058 		 * The imaginary "sockmod" has been popped - act as a stream.
6059 		 * If this is a push of sockmod then change back to a socket.
6060 		 */
6061 		if (cmd == I_PUSH) {
6062 			error = ((mode & FKIOCTL) ? copystr : copyinstr)(
6063 			    (void *)arg, mname, sizeof (mname), NULL);
6064 
6065 			if (error == 0 && strcmp(mname, sockmod_name) == 0) {
6066 				dprintso(so, 0, ("socktpi_ioctl: going to "
6067 				    "socket version\n"));
6068 				so_stream2sock(so);
6069 				return (0);
6070 			}
6071 		}
6072 		return (strioctl(vp, cmd, arg, mode, U_TO_K, cr, rvalp));
6073 	}
6074 
6075 	switch (cmd) {
6076 	case I_PUSH:
6077 		if (sti->sti_direct) {
6078 			mutex_enter(&so->so_lock);
6079 			so_lock_single(so);
6080 			mutex_exit(&so->so_lock);
6081 
6082 			error = strioctl(vp, _SIOCSOCKFALLBACK, 0, 0, K_TO_K,
6083 			    cr, rvalp);
6084 
6085 			mutex_enter(&so->so_lock);
6086 			if (error == 0)
6087 				sti->sti_direct = 0;
6088 			so_unlock_single(so, SOLOCKED);
6089 			mutex_exit(&so->so_lock);
6090 
6091 			if (error != 0)
6092 				return (error);
6093 		}
6094 
6095 		error = strioctl(vp, cmd, arg, mode, U_TO_K, cr, rvalp);
6096 		if (error == 0)
6097 			sti->sti_pushcnt++;
6098 		return (error);
6099 
6100 	case I_POP:
6101 		if (sti->sti_pushcnt == 0) {
6102 			/* Emulate sockmod being popped */
6103 			dprintso(so, 0,
6104 			    ("socktpi_ioctl: going to STREAMS version\n"));
6105 			return (so_sock2stream(so));
6106 		}
6107 
6108 		error = strioctl(vp, cmd, arg, mode, U_TO_K, cr, rvalp);
6109 		if (error == 0)
6110 			sti->sti_pushcnt--;
6111 		return (error);
6112 
6113 	case I_LIST: {
6114 		struct str_mlist *kmlistp, *umlistp;
6115 		struct str_list	kstrlist;
6116 		ssize_t		kstrlistsize;
6117 		int		i, nmods;
6118 
6119 		STRUCT_DECL(str_list, ustrlist);
6120 		STRUCT_INIT(ustrlist, mode);
6121 
6122 		if (arg == NULL) {
6123 			error = strioctl(vp, cmd, arg, mode, U_TO_K, cr, rvalp);
6124 			if (error == 0)
6125 				(*rvalp)++;	/* Add one for sockmod */
6126 			return (error);
6127 		}
6128 
6129 		error = so_copyin((void *)arg, STRUCT_BUF(ustrlist),
6130 		    STRUCT_SIZE(ustrlist), mode & FKIOCTL);
6131 		if (error != 0)
6132 			return (error);
6133 
6134 		nmods = STRUCT_FGET(ustrlist, sl_nmods);
6135 		if (nmods <= 0)
6136 			return (EINVAL);
6137 		/*
6138 		 * Ceiling nmods at nstrpush to prevent someone from
6139 		 * maliciously consuming lots of kernel memory.
6140 		 */
6141 		nmods = MIN(nmods, nstrpush);
6142 
6143 		kstrlistsize = (nmods + 1) * sizeof (struct str_mlist);
6144 		kstrlist.sl_nmods = nmods;
6145 		kstrlist.sl_modlist = kmem_zalloc(kstrlistsize, KM_SLEEP);
6146 
6147 		error = strioctl(vp, cmd, (intptr_t)&kstrlist, mode, K_TO_K,
6148 		    cr, rvalp);
6149 		if (error != 0)
6150 			goto done;
6151 
6152 		/*
6153 		 * Considering the module list as a 0-based array of sl_nmods
6154 		 * modules, sockmod should conceptually exist at slot
6155 		 * sti_pushcnt.  Insert sockmod at this location by sliding all
6156 		 * of the module names after so_pushcnt over by one.  We know
6157 		 * that there will be room to do this since we allocated
6158 		 * sl_modlist with an additional slot.
6159 		 */
6160 		for (i = kstrlist.sl_nmods; i > sti->sti_pushcnt; i--)
6161 			kstrlist.sl_modlist[i] = kstrlist.sl_modlist[i - 1];
6162 
6163 		(void) strcpy(kstrlist.sl_modlist[i].l_name, sockmod_name);
6164 		kstrlist.sl_nmods++;
6165 
6166 		/*
6167 		 * Copy all of the entries out to ustrlist.
6168 		 */
6169 		kmlistp = kstrlist.sl_modlist;
6170 		umlistp = STRUCT_FGETP(ustrlist, sl_modlist);
6171 		for (i = 0; i < nmods && i < kstrlist.sl_nmods; i++) {
6172 			error = so_copyout(kmlistp++, umlistp++,
6173 			    sizeof (struct str_mlist), mode & FKIOCTL);
6174 			if (error != 0)
6175 				goto done;
6176 		}
6177 
6178 		error = so_copyout(&i, (void *)arg, sizeof (int32_t),
6179 		    mode & FKIOCTL);
6180 		if (error == 0)
6181 			*rvalp = 0;
6182 	done:
6183 		kmem_free(kstrlist.sl_modlist, kstrlistsize);
6184 		return (error);
6185 	}
6186 	case I_LOOK:
6187 		if (sti->sti_pushcnt == 0) {
6188 			return (so_copyout(sockmod_name, (void *)arg,
6189 			    sizeof (sockmod_name), mode & FKIOCTL));
6190 		}
6191 		return (strioctl(vp, cmd, arg, mode, U_TO_K, cr, rvalp));
6192 
6193 	case I_FIND:
6194 		error = strioctl(vp, cmd, arg, mode, U_TO_K, cr, rvalp);
6195 		if (error && error != EINVAL)
6196 			return (error);
6197 
6198 		/* if not found and string was sockmod return 1 */
6199 		if (*rvalp == 0 || error == EINVAL) {
6200 			error = ((mode & FKIOCTL) ? copystr : copyinstr)(
6201 			    (void *)arg, mname, sizeof (mname), NULL);
6202 			if (error == ENAMETOOLONG)
6203 				error = EINVAL;
6204 
6205 			if (error == 0 && strcmp(mname, sockmod_name) == 0)
6206 				*rvalp = 1;
6207 		}
6208 		return (error);
6209 
6210 	default:
6211 		panic("socktpi_plumbioctl: unknown ioctl %d", cmd);
6212 		break;
6213 	}
6214 
6215 	return (0);
6216 }
6217 
6218 /*
6219  * Wrapper around the streams poll routine that implements socket poll
6220  * semantics.
6221  * The sockfs never calls pollwakeup itself - the stream head take care
6222  * of all pollwakeups. Since sockfs never holds so_lock when calling the
6223  * stream head there can never be a deadlock due to holding so_lock across
6224  * pollwakeup and acquiring so_lock in this routine.
6225  *
6226  * However, since the performance of VOP_POLL is critical we avoid
6227  * acquiring so_lock here. This is based on two assumptions:
6228  *  - The poll implementation holds locks to serialize the VOP_POLL call
6229  *    and a pollwakeup for the same pollhead. This ensures that should
6230  *    e.g. so_state change during a socktpi_poll call the pollwakeup
6231  *    (which strsock_* and strrput conspire to issue) is issued after
6232  *    the state change. Thus the pollwakeup will block until VOP_POLL has
6233  *    returned and then wake up poll and have it call VOP_POLL again.
6234  *  - The reading of so_state without holding so_lock does not result in
6235  *    stale data that is older than the latest state change that has dropped
6236  *    so_lock. This is ensured by the mutex_exit issuing the appropriate
6237  *    memory barrier to force the data into the coherency domain.
6238  */
6239 static int
6240 sotpi_poll(
6241 	struct sonode	*so,
6242 	short		events,
6243 	int		anyyet,
6244 	short		*reventsp,
6245 	struct pollhead **phpp)
6246 {
6247 	short origevents = events;
6248 	struct vnode *vp = SOTOV(so);
6249 	int error;
6250 	int so_state = so->so_state;	/* snapshot */
6251 	sotpi_info_t *sti = SOTOTPI(so);
6252 
6253 	dprintso(so, 0, ("socktpi_poll(%p): state %s err %d\n",
6254 	    (void *)vp, pr_state(so_state, so->so_mode), so->so_error));
6255 
6256 	ASSERT(vp->v_type == VSOCK);
6257 	ASSERT(vp->v_stream != NULL);
6258 
6259 	if (so->so_version == SOV_STREAM) {
6260 		/* The imaginary "sockmod" has been popped - act as a stream */
6261 		return (strpoll(vp->v_stream, events, anyyet,
6262 		    reventsp, phpp));
6263 	}
6264 
6265 	if (!(so_state & SS_ISCONNECTED) &&
6266 	    (so->so_mode & SM_CONNREQUIRED)) {
6267 		/* Not connected yet - turn off write side events */
6268 		events &= ~(POLLOUT|POLLWRBAND);
6269 	}
6270 	/*
6271 	 * Check for errors without calling strpoll if the caller wants them.
6272 	 * In sockets the errors are represented as input/output events
6273 	 * and there is no need to ask the stream head for this information.
6274 	 */
6275 	if (so->so_error != 0 &&
6276 	    ((POLLIN|POLLRDNORM|POLLOUT) & origevents)  != 0) {
6277 		*reventsp = (POLLIN|POLLRDNORM|POLLOUT) & origevents;
6278 		return (0);
6279 	}
6280 	/*
6281 	 * Ignore M_PROTO only messages such as the T_EXDATA_IND messages.
6282 	 * These message with only an M_PROTO/M_PCPROTO part and no M_DATA
6283 	 * will not trigger a POLLIN event with POLLRDDATA set.
6284 	 * The handling of urgent data (causing POLLRDBAND) is done by
6285 	 * inspecting SS_OOBPEND below.
6286 	 */
6287 	events |= POLLRDDATA;
6288 
6289 	/*
6290 	 * After shutdown(output) a stream head write error is set.
6291 	 * However, we should not return output events.
6292 	 */
6293 	events |= POLLNOERR;
6294 	error = strpoll(vp->v_stream, events, anyyet,
6295 	    reventsp, phpp);
6296 	if (error)
6297 		return (error);
6298 
6299 	ASSERT(!(*reventsp & POLLERR));
6300 
6301 	/*
6302 	 * Notes on T_CONN_IND handling for sockets.
6303 	 *
6304 	 * If strpoll() returned without events, SR_POLLIN is guaranteed
6305 	 * to be set, ensuring any subsequent strrput() runs pollwakeup().
6306 	 *
6307 	 * Since the so_lock is not held, soqueueconnind() may have run
6308 	 * and a T_CONN_IND may be waiting. We now check for any queued
6309 	 * T_CONN_IND msgs on sti_conn_ind_head and set appropriate events
6310 	 * to ensure poll returns.
6311 	 *
6312 	 * However:
6313 	 * If the T_CONN_IND hasn't arrived by the time strpoll() returns,
6314 	 * when strrput() does run for an arriving M_PROTO with T_CONN_IND
6315 	 * the following actions will occur; taken together they ensure the
6316 	 * syscall will return.
6317 	 *
6318 	 * 1. If a socket, soqueueconnind() will queue the T_CONN_IND but if
6319 	 *    the accept() was run on a non-blocking socket sowaitconnind()
6320 	 *    may have already returned EWOULDBLOCK, so not be waiting to
6321 	 *    process the message. Additionally socktpi_poll() has probably
6322 	 *    proceeded past the sti_conn_ind_head check below.
6323 	 * 2. strrput() runs pollwakeup()->pollnotify()->cv_signal() to wake
6324 	 *    this thread,  however that could occur before poll_common()
6325 	 *    has entered cv_wait.
6326 	 * 3. pollnotify() sets T_POLLWAKE, while holding the pc_lock.
6327 	 *
6328 	 * Before proceeding to cv_wait() in poll_common() for an event,
6329 	 * poll_common() atomically checks for T_POLLWAKE under the pc_lock,
6330 	 * and if set, re-calls strpoll() to ensure the late arriving
6331 	 * T_CONN_IND is recognized, and pollsys() returns.
6332 	 */
6333 
6334 	if (sti->sti_conn_ind_head != NULL)
6335 		*reventsp |= (POLLIN|POLLRDNORM) & events;
6336 
6337 	if (so->so_state & SS_OOBPEND)
6338 		*reventsp |= POLLRDBAND & events;
6339 
6340 	if (sti->sti_nl7c_rcv_mp != NULL) {
6341 		*reventsp |= (POLLIN|POLLRDNORM) & events;
6342 	}
6343 	if ((sti->sti_nl7c_flags & NL7C_ENABLED) &&
6344 	    ((POLLIN|POLLRDNORM) & *reventsp)) {
6345 		sti->sti_nl7c_flags |= NL7C_POLLIN;
6346 	}
6347 
6348 	return (0);
6349 }
6350 
6351 /*ARGSUSED*/
6352 static int
6353 socktpi_constructor(void *buf, void *cdrarg, int kmflags)
6354 {
6355 	sotpi_sonode_t *st = (sotpi_sonode_t *)buf;
6356 	int error = 0;
6357 
6358 	error = sonode_constructor(buf, cdrarg, kmflags);
6359 	if (error != 0)
6360 		return (error);
6361 
6362 	error = i_sotpi_info_constructor(&st->st_info);
6363 	if (error != 0)
6364 		sonode_destructor(buf, cdrarg);
6365 
6366 	st->st_sonode.so_priv = &st->st_info;
6367 
6368 	return (error);
6369 }
6370 
6371 /*ARGSUSED1*/
6372 static void
6373 socktpi_destructor(void *buf, void *cdrarg)
6374 {
6375 	sotpi_sonode_t *st = (sotpi_sonode_t *)buf;
6376 
6377 	ASSERT(st->st_sonode.so_priv == &st->st_info);
6378 	st->st_sonode.so_priv = NULL;
6379 
6380 	i_sotpi_info_destructor(&st->st_info);
6381 	sonode_destructor(buf, cdrarg);
6382 }
6383 
6384 static int
6385 socktpi_unix_constructor(void *buf, void *cdrarg, int kmflags)
6386 {
6387 	int retval;
6388 
6389 	if ((retval = socktpi_constructor(buf, cdrarg, kmflags)) == 0) {
6390 		struct sonode *so = (struct sonode *)buf;
6391 		sotpi_info_t *sti = SOTOTPI(so);
6392 
6393 		mutex_enter(&socklist.sl_lock);
6394 
6395 		sti->sti_next_so = socklist.sl_list;
6396 		sti->sti_prev_so = NULL;
6397 		if (sti->sti_next_so != NULL)
6398 			SOTOTPI(sti->sti_next_so)->sti_prev_so = so;
6399 		socklist.sl_list = so;
6400 
6401 		mutex_exit(&socklist.sl_lock);
6402 
6403 	}
6404 	return (retval);
6405 }
6406 
6407 static void
6408 socktpi_unix_destructor(void *buf, void *cdrarg)
6409 {
6410 	struct sonode	*so = (struct sonode *)buf;
6411 	sotpi_info_t	*sti = SOTOTPI(so);
6412 
6413 	mutex_enter(&socklist.sl_lock);
6414 
6415 	if (sti->sti_next_so != NULL)
6416 		SOTOTPI(sti->sti_next_so)->sti_prev_so = sti->sti_prev_so;
6417 	if (sti->sti_prev_so != NULL)
6418 		SOTOTPI(sti->sti_prev_so)->sti_next_so = sti->sti_next_so;
6419 	else
6420 		socklist.sl_list = sti->sti_next_so;
6421 
6422 	mutex_exit(&socklist.sl_lock);
6423 
6424 	socktpi_destructor(buf, cdrarg);
6425 }
6426 
6427 int
6428 socktpi_init(void)
6429 {
6430 	/*
6431 	 * Create sonode caches.  We create a special one for AF_UNIX so
6432 	 * that we can track them for netstat(1m).
6433 	 */
6434 	socktpi_cache = kmem_cache_create("socktpi_cache",
6435 	    sizeof (struct sotpi_sonode), 0, socktpi_constructor,
6436 	    socktpi_destructor, NULL, NULL, NULL, 0);
6437 
6438 	socktpi_unix_cache = kmem_cache_create("socktpi_unix_cache",
6439 	    sizeof (struct sotpi_sonode), 0, socktpi_unix_constructor,
6440 	    socktpi_unix_destructor, NULL, NULL, NULL, 0);
6441 
6442 	return (0);
6443 }
6444 
6445 /*
6446  * Given a non-TPI sonode, allocate and prep it to be ready for TPI.
6447  *
6448  * Caller must still update state and mode using sotpi_update_state().
6449  */
6450 int
6451 sotpi_convert_sonode(struct sonode *so, struct sockparams *newsp,
6452     boolean_t *direct, queue_t **qp, struct cred *cr)
6453 {
6454 	sotpi_info_t *sti;
6455 	struct sockparams *origsp = so->so_sockparams;
6456 	sock_lower_handle_t handle = so->so_proto_handle;
6457 	struct stdata *stp;
6458 	struct vnode *vp;
6459 	queue_t *q;
6460 	int error = 0;
6461 
6462 	ASSERT((so->so_state & (SS_FALLBACK_PENDING|SS_FALLBACK_COMP)) ==
6463 	    SS_FALLBACK_PENDING);
6464 	ASSERT(SOCK_IS_NONSTR(so));
6465 
6466 	*qp = NULL;
6467 	*direct = B_FALSE;
6468 	so->so_sockparams = newsp;
6469 	/*
6470 	 * Allocate and initalize fields required by TPI.
6471 	 */
6472 	(void) sotpi_info_create(so, KM_SLEEP);
6473 	sotpi_info_init(so);
6474 
6475 	if ((error = sotpi_init(so, NULL, cr, SO_FALLBACK)) != 0) {
6476 		sotpi_info_fini(so);
6477 		sotpi_info_destroy(so);
6478 		return (error);
6479 	}
6480 	ASSERT(handle == so->so_proto_handle);
6481 	sti = SOTOTPI(so);
6482 	if (sti->sti_direct != 0)
6483 		*direct = B_TRUE;
6484 
6485 	/*
6486 	 * Keep the original sp around so we can properly dispose of the
6487 	 * sonode when the socket is being closed.
6488 	 */
6489 	sti->sti_orig_sp = origsp;
6490 
6491 	so_basic_strinit(so);	/* skips the T_CAPABILITY_REQ */
6492 	so_alloc_addr(so, so->so_max_addr_len);
6493 
6494 	/*
6495 	 * If the application has done a SIOCSPGRP, make sure the
6496 	 * STREAM head is aware. This needs to take place before
6497 	 * the protocol start sending up messages. Otherwise we
6498 	 * might miss to generate SIGPOLL.
6499 	 *
6500 	 * It is possible that the application will receive duplicate
6501 	 * signals if some were already generated for either data or
6502 	 * connection indications.
6503 	 */
6504 	if (so->so_pgrp != 0) {
6505 		if (so_set_events(so, so->so_vnode, cr) != 0)
6506 			so->so_pgrp = 0;
6507 	}
6508 
6509 	/*
6510 	 * Determine which queue to use.
6511 	 */
6512 	vp = SOTOV(so);
6513 	stp = vp->v_stream;
6514 	ASSERT(stp != NULL);
6515 	q = stp->sd_wrq->q_next;
6516 
6517 	/*
6518 	 * Skip any modules that may have been auto pushed when the device
6519 	 * was opened
6520 	 */
6521 	while (q->q_next != NULL)
6522 		q = q->q_next;
6523 	*qp = _RD(q);
6524 
6525 	/* This is now a STREAMS sockets */
6526 	so->so_not_str = B_FALSE;
6527 
6528 	return (error);
6529 }
6530 
6531 /*
6532  * Revert a TPI sonode. It is only allowed to revert the sonode during
6533  * the fallback process.
6534  */
6535 void
6536 sotpi_revert_sonode(struct sonode *so, struct cred *cr)
6537 {
6538 	vnode_t *vp = SOTOV(so);
6539 
6540 	ASSERT((so->so_state & (SS_FALLBACK_PENDING|SS_FALLBACK_COMP)) ==
6541 	    SS_FALLBACK_PENDING);
6542 	ASSERT(!SOCK_IS_NONSTR(so));
6543 	ASSERT(vp->v_stream != NULL);
6544 
6545 	strclean(vp);
6546 	(void) strclose(vp, FREAD|FWRITE|SO_FALLBACK, cr);
6547 
6548 	/*
6549 	 * Restore the original sockparams. The caller is responsible for
6550 	 * dropping the ref to the new sp.
6551 	 */
6552 	so->so_sockparams = SOTOTPI(so)->sti_orig_sp;
6553 
6554 	sotpi_info_fini(so);
6555 	sotpi_info_destroy(so);
6556 
6557 	/* This is no longer a STREAMS sockets */
6558 	so->so_not_str = B_TRUE;
6559 }
6560 
6561 void
6562 sotpi_update_state(struct sonode *so, struct T_capability_ack *tcap,
6563     struct sockaddr *laddr, socklen_t laddrlen, struct sockaddr *faddr,
6564     socklen_t faddrlen, short opts)
6565 {
6566 	sotpi_info_t *sti = SOTOTPI(so);
6567 
6568 	so_proc_tcapability_ack(so, tcap);
6569 
6570 	so->so_options |= opts;
6571 
6572 	/*
6573 	 * Determine whether the foreign and local address are valid
6574 	 */
6575 	if (laddrlen != 0) {
6576 		ASSERT(laddrlen <= sti->sti_laddr_maxlen);
6577 		sti->sti_laddr_len = laddrlen;
6578 		bcopy(laddr, sti->sti_laddr_sa, laddrlen);
6579 		sti->sti_laddr_valid = (so->so_state & SS_ISBOUND);
6580 	}
6581 
6582 	if (faddrlen != 0) {
6583 		ASSERT(faddrlen <= sti->sti_faddr_maxlen);
6584 		sti->sti_faddr_len = faddrlen;
6585 		bcopy(faddr, sti->sti_faddr_sa, faddrlen);
6586 		sti->sti_faddr_valid = (so->so_state & SS_ISCONNECTED);
6587 	}
6588 
6589 }
6590 
6591 /*
6592  * Allocate enough space to cache the local and foreign addresses.
6593  */
6594 void
6595 so_alloc_addr(struct sonode *so, t_uscalar_t maxlen)
6596 {
6597 	sotpi_info_t *sti = SOTOTPI(so);
6598 
6599 	ASSERT(sti->sti_laddr_sa == NULL && sti->sti_faddr_sa == NULL);
6600 	ASSERT(sti->sti_laddr_len == 0 && sti->sti_faddr_len == 0);
6601 	sti->sti_laddr_maxlen = sti->sti_faddr_maxlen =
6602 	    P2ROUNDUP(maxlen, KMEM_ALIGN);
6603 	so->so_max_addr_len = sti->sti_laddr_maxlen;
6604 	sti->sti_laddr_sa = kmem_alloc(sti->sti_laddr_maxlen * 2, KM_SLEEP);
6605 	sti->sti_faddr_sa = (struct sockaddr *)((caddr_t)sti->sti_laddr_sa
6606 	    + sti->sti_laddr_maxlen);
6607 
6608 	if (so->so_family == AF_UNIX) {
6609 		/*
6610 		 * Initialize AF_UNIX related fields.
6611 		 */
6612 		bzero(&sti->sti_ux_laddr, sizeof (sti->sti_ux_laddr));
6613 		bzero(&sti->sti_ux_faddr, sizeof (sti->sti_ux_faddr));
6614 	}
6615 }
6616 
6617 
6618 sotpi_info_t *
6619 sotpi_sototpi(struct sonode *so)
6620 {
6621 	sotpi_info_t *sti;
6622 
6623 	ASSERT(so != NULL);
6624 
6625 	sti = (sotpi_info_t *)so->so_priv;
6626 
6627 	ASSERT(sti != NULL);
6628 	ASSERT(sti->sti_magic == SOTPI_INFO_MAGIC);
6629 
6630 	return (sti);
6631 }
6632 
6633 static int
6634 i_sotpi_info_constructor(sotpi_info_t *sti)
6635 {
6636 	sti->sti_magic		= SOTPI_INFO_MAGIC;
6637 	sti->sti_ack_mp		= NULL;
6638 	sti->sti_discon_ind_mp	= NULL;
6639 	sti->sti_ux_bound_vp	= NULL;
6640 	sti->sti_unbind_mp	= NULL;
6641 
6642 	sti->sti_conn_ind_head	= NULL;
6643 	sti->sti_conn_ind_tail	= NULL;
6644 
6645 	sti->sti_laddr_sa	= NULL;
6646 	sti->sti_faddr_sa	= NULL;
6647 
6648 	sti->sti_nl7c_flags	= 0;
6649 	sti->sti_nl7c_uri	= NULL;
6650 	sti->sti_nl7c_rcv_mp	= NULL;
6651 
6652 	mutex_init(&sti->sti_plumb_lock, NULL, MUTEX_DEFAULT, NULL);
6653 	cv_init(&sti->sti_ack_cv, NULL, CV_DEFAULT, NULL);
6654 
6655 	return (0);
6656 }
6657 
6658 static void
6659 i_sotpi_info_destructor(sotpi_info_t *sti)
6660 {
6661 	ASSERT(sti->sti_magic == SOTPI_INFO_MAGIC);
6662 	ASSERT(sti->sti_ack_mp == NULL);
6663 	ASSERT(sti->sti_discon_ind_mp == NULL);
6664 	ASSERT(sti->sti_ux_bound_vp == NULL);
6665 	ASSERT(sti->sti_unbind_mp == NULL);
6666 
6667 	ASSERT(sti->sti_conn_ind_head == NULL);
6668 	ASSERT(sti->sti_conn_ind_tail == NULL);
6669 
6670 	ASSERT(sti->sti_laddr_sa == NULL);
6671 	ASSERT(sti->sti_faddr_sa == NULL);
6672 
6673 	ASSERT(sti->sti_nl7c_flags == 0);
6674 	ASSERT(sti->sti_nl7c_uri == NULL);
6675 	ASSERT(sti->sti_nl7c_rcv_mp == NULL);
6676 
6677 	mutex_destroy(&sti->sti_plumb_lock);
6678 	cv_destroy(&sti->sti_ack_cv);
6679 }
6680 
6681 /*
6682  * Creates and attaches TPI information to the given sonode
6683  */
6684 static boolean_t
6685 sotpi_info_create(struct sonode *so, int kmflags)
6686 {
6687 	sotpi_info_t *sti;
6688 
6689 	ASSERT(so->so_priv == NULL);
6690 
6691 	if ((sti = kmem_zalloc(sizeof (*sti), kmflags)) == NULL)
6692 		return (B_FALSE);
6693 
6694 	if (i_sotpi_info_constructor(sti) != 0) {
6695 		kmem_free(sti, sizeof (*sti));
6696 		return (B_FALSE);
6697 	}
6698 
6699 	so->so_priv = (void *)sti;
6700 	return (B_TRUE);
6701 }
6702 
6703 /*
6704  * Initializes the TPI information.
6705  */
6706 static void
6707 sotpi_info_init(struct sonode *so)
6708 {
6709 	struct vnode *vp = SOTOV(so);
6710 	sotpi_info_t *sti = SOTOTPI(so);
6711 	time_t now;
6712 
6713 	sti->sti_dev 	= so->so_sockparams->sp_sdev_info.sd_vnode->v_rdev;
6714 	vp->v_rdev	= sti->sti_dev;
6715 
6716 	sti->sti_orig_sp = NULL;
6717 
6718 	sti->sti_pushcnt = 0;
6719 
6720 	now = gethrestime_sec();
6721 	sti->sti_atime	= now;
6722 	sti->sti_mtime	= now;
6723 	sti->sti_ctime	= now;
6724 
6725 	sti->sti_eaddr_mp = NULL;
6726 	sti->sti_delayed_error = 0;
6727 
6728 	sti->sti_provinfo = NULL;
6729 
6730 	sti->sti_oobcnt = 0;
6731 	sti->sti_oobsigcnt = 0;
6732 
6733 	ASSERT(sti->sti_laddr_sa == NULL && sti->sti_faddr_sa == NULL);
6734 
6735 	sti->sti_laddr_sa	= 0;
6736 	sti->sti_faddr_sa	= 0;
6737 	sti->sti_laddr_maxlen = sti->sti_faddr_maxlen = 0;
6738 	sti->sti_laddr_len = sti->sti_faddr_len = 0;
6739 
6740 	sti->sti_laddr_valid = 0;
6741 	sti->sti_faddr_valid = 0;
6742 	sti->sti_faddr_noxlate = 0;
6743 
6744 	sti->sti_direct = 0;
6745 
6746 	ASSERT(sti->sti_ack_mp == NULL);
6747 	ASSERT(sti->sti_ux_bound_vp == NULL);
6748 	ASSERT(sti->sti_unbind_mp == NULL);
6749 
6750 	ASSERT(sti->sti_conn_ind_head == NULL);
6751 	ASSERT(sti->sti_conn_ind_tail == NULL);
6752 
6753 	/* Initialize the kernel SSL proxy fields */
6754 	sti->sti_kssl_type = KSSL_NO_PROXY;
6755 	sti->sti_kssl_ent = NULL;
6756 	sti->sti_kssl_ctx = NULL;
6757 }
6758 
6759 /*
6760  * Given a sonode, grab the TPI info and free any data.
6761  */
6762 static void
6763 sotpi_info_fini(struct sonode *so)
6764 {
6765 	sotpi_info_t *sti = SOTOTPI(so);
6766 	mblk_t *mp;
6767 
6768 	ASSERT(sti->sti_discon_ind_mp == NULL);
6769 
6770 	if ((mp = sti->sti_conn_ind_head) != NULL) {
6771 		mblk_t *mp1;
6772 
6773 		while (mp) {
6774 			mp1 = mp->b_next;
6775 			mp->b_next = NULL;
6776 			freemsg(mp);
6777 			mp = mp1;
6778 		}
6779 		sti->sti_conn_ind_head = sti->sti_conn_ind_tail = NULL;
6780 	}
6781 
6782 	/*
6783 	 * Protect so->so_[lf]addr_sa so that sockfs_snapshot() can safely
6784 	 * indirect them.  It also uses so_count as a validity test.
6785 	 */
6786 	mutex_enter(&so->so_lock);
6787 
6788 	if (sti->sti_laddr_sa) {
6789 		ASSERT((caddr_t)sti->sti_faddr_sa ==
6790 		    (caddr_t)sti->sti_laddr_sa + sti->sti_laddr_maxlen);
6791 		ASSERT(sti->sti_faddr_maxlen == sti->sti_laddr_maxlen);
6792 		sti->sti_laddr_valid = 0;
6793 		sti->sti_faddr_valid = 0;
6794 		kmem_free(sti->sti_laddr_sa, sti->sti_laddr_maxlen * 2);
6795 		sti->sti_laddr_sa = NULL;
6796 		sti->sti_laddr_len = sti->sti_laddr_maxlen = 0;
6797 		sti->sti_faddr_sa = NULL;
6798 		sti->sti_faddr_len = sti->sti_faddr_maxlen = 0;
6799 	}
6800 
6801 	mutex_exit(&so->so_lock);
6802 
6803 	if ((mp = sti->sti_eaddr_mp) != NULL) {
6804 		freemsg(mp);
6805 		sti->sti_eaddr_mp = NULL;
6806 		sti->sti_delayed_error = 0;
6807 	}
6808 
6809 	if ((mp = sti->sti_ack_mp) != NULL) {
6810 		freemsg(mp);
6811 		sti->sti_ack_mp = NULL;
6812 	}
6813 
6814 	if ((mp = sti->sti_nl7c_rcv_mp) != NULL) {
6815 		sti->sti_nl7c_rcv_mp = NULL;
6816 		freemsg(mp);
6817 	}
6818 	sti->sti_nl7c_rcv_rval = 0;
6819 	if (sti->sti_nl7c_uri != NULL) {
6820 		nl7c_urifree(so);
6821 		/* urifree() cleared nl7c_uri */
6822 	}
6823 	if (sti->sti_nl7c_flags) {
6824 		sti->sti_nl7c_flags = 0;
6825 	}
6826 
6827 	ASSERT(sti->sti_ux_bound_vp == NULL);
6828 	if ((mp = sti->sti_unbind_mp) != NULL) {
6829 		freemsg(mp);
6830 		sti->sti_unbind_mp = NULL;
6831 	}
6832 }
6833 
6834 /*
6835  * Destroys the TPI information attached to a sonode.
6836  */
6837 static void
6838 sotpi_info_destroy(struct sonode *so)
6839 {
6840 	sotpi_info_t *sti = SOTOTPI(so);
6841 
6842 	i_sotpi_info_destructor(sti);
6843 	kmem_free(sti, sizeof (*sti));
6844 
6845 	so->so_priv = NULL;
6846 }
6847 
6848 /*
6849  * Create the global sotpi socket module entry. It will never be freed.
6850  */
6851 smod_info_t *
6852 sotpi_smod_create(void)
6853 {
6854 	smod_info_t *smodp;
6855 
6856 	smodp = kmem_zalloc(sizeof (*smodp), KM_SLEEP);
6857 	smodp->smod_name = kmem_alloc(sizeof (SOTPI_SMOD_NAME), KM_SLEEP);
6858 	(void) strcpy(smodp->smod_name, SOTPI_SMOD_NAME);
6859 	/*
6860 	 * Initialize the smod_refcnt to 1 so it will never be freed.
6861 	 */
6862 	smodp->smod_refcnt = 1;
6863 	smodp->smod_uc_version = SOCK_UC_VERSION;
6864 	smodp->smod_dc_version = SOCK_DC_VERSION;
6865 	smodp->smod_sock_create_func = &sotpi_create;
6866 	smodp->smod_sock_destroy_func = &sotpi_destroy;
6867 	return (smodp);
6868 }
6869