xref: /titanic_41/usr/src/uts/common/os/streamio.c (revision 6f45ec7b0b964c3be967c4880e8867ac1e7763a5)
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 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
22 /*	  All Rights Reserved  	*/
23 
24 
25 /*
26  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
27  * Use is subject to license terms.
28  */
29 
30 #pragma ident	"%Z%%M%	%I%	%E% SMI"
31 
32 #include <sys/types.h>
33 #include <sys/sysmacros.h>
34 #include <sys/param.h>
35 #include <sys/errno.h>
36 #include <sys/signal.h>
37 #include <sys/stat.h>
38 #include <sys/proc.h>
39 #include <sys/cred.h>
40 #include <sys/user.h>
41 #include <sys/vnode.h>
42 #include <sys/file.h>
43 #include <sys/stream.h>
44 #include <sys/strsubr.h>
45 #include <sys/stropts.h>
46 #include <sys/tihdr.h>
47 #include <sys/var.h>
48 #include <sys/poll.h>
49 #include <sys/termio.h>
50 #include <sys/ttold.h>
51 #include <sys/systm.h>
52 #include <sys/uio.h>
53 #include <sys/cmn_err.h>
54 #include <sys/sad.h>
55 #include <sys/netstack.h>
56 #include <sys/priocntl.h>
57 #include <sys/jioctl.h>
58 #include <sys/procset.h>
59 #include <sys/session.h>
60 #include <sys/kmem.h>
61 #include <sys/filio.h>
62 #include <sys/vtrace.h>
63 #include <sys/debug.h>
64 #include <sys/strredir.h>
65 #include <sys/fs/fifonode.h>
66 #include <sys/fs/snode.h>
67 #include <sys/strlog.h>
68 #include <sys/strsun.h>
69 #include <sys/project.h>
70 #include <sys/kbio.h>
71 #include <sys/msio.h>
72 #include <sys/tty.h>
73 #include <sys/ptyvar.h>
74 #include <sys/vuid_event.h>
75 #include <sys/modctl.h>
76 #include <sys/sunddi.h>
77 #include <sys/sunldi_impl.h>
78 #include <sys/autoconf.h>
79 #include <sys/policy.h>
80 #include <sys/zone.h>
81 
82 
83 /*
84  * This define helps improve the readability of streams code while
85  * still maintaining a very old streams performance enhancement.  The
86  * performance enhancement basically involved having all callers
87  * of straccess() perform the first check that straccess() will do
88  * locally before actually calling straccess().  (There by reducing
89  * the number of unnecessary calls to straccess().)
90  */
91 #define	i_straccess(x, y)	((stp->sd_sidp == NULL) ? 0 : \
92 				    (stp->sd_vnode->v_type == VFIFO) ? 0 : \
93 				    straccess((x), (y)))
94 
95 /*
96  * what is mblk_pull_len?
97  *
98  * If a streams message consists of many short messages,
99  * a performance degradation occurs from copyout overhead.
100  * To decrease the per mblk overhead, messages that are
101  * likely to consist of many small mblks are pulled up into
102  * one continuous chunk of memory.
103  *
104  * To avoid the processing overhead of examining every
105  * mblk, a quick heuristic is used. If the first mblk in
106  * the message is shorter than mblk_pull_len, it is likely
107  * that the rest of the mblk will be short.
108  *
109  * This heuristic was decided upon after performance tests
110  * indicated that anything more complex slowed down the main
111  * code path.
112  */
113 #define	MBLK_PULL_LEN 64
114 uint32_t mblk_pull_len = MBLK_PULL_LEN;
115 
116 /*
117  * The sgttyb_handling flag controls the handling of the old BSD
118  * TIOCGETP, TIOCSETP, and TIOCSETN ioctls as follows:
119  *
120  * 0 - Emit no warnings at all and retain old, broken behavior.
121  * 1 - Emit no warnings and silently handle new semantics.
122  * 2 - Send cmn_err(CE_NOTE) when either TIOCSETP or TIOCSETN is used
123  *     (once per system invocation).  Handle with new semantics.
124  * 3 - Send SIGSYS when any TIOCGETP, TIOCSETP, or TIOCSETN call is
125  *     made (so that offenders drop core and are easy to debug).
126  *
127  * The "new semantics" are that TIOCGETP returns B38400 for
128  * sg_[io]speed if the corresponding value is over B38400, and that
129  * TIOCSET[PN] accept B38400 in these cases to mean "retain current
130  * bit rate."
131  */
132 int sgttyb_handling = 1;
133 static boolean_t sgttyb_complaint;
134 
135 /* don't push drcompat module by default on Style-2 streams */
136 static int push_drcompat = 0;
137 
138 /*
139  * id value used to distinguish between different ioctl messages
140  */
141 static uint32_t ioc_id;
142 
143 static void putback(struct stdata *, queue_t *, mblk_t *, int);
144 static void strcleanall(struct vnode *);
145 static int strwsrv(queue_t *);
146 
147 /*
148  * qinit and module_info structures for stream head read and write queues
149  */
150 struct module_info strm_info = { 0, "strrhead", 0, INFPSZ, STRHIGH, STRLOW };
151 struct module_info stwm_info = { 0, "strwhead", 0, 0, 0, 0 };
152 struct qinit strdata = { strrput, NULL, NULL, NULL, NULL, &strm_info };
153 struct qinit stwdata = { NULL, strwsrv, NULL, NULL, NULL, &stwm_info };
154 struct module_info fiform_info = { 0, "fifostrrhead", 0, PIPE_BUF, FIFOHIWAT,
155     FIFOLOWAT };
156 struct module_info fifowm_info = { 0, "fifostrwhead", 0, 0, 0, 0 };
157 struct qinit fifo_strdata = { strrput, NULL, NULL, NULL, NULL, &fiform_info };
158 struct qinit fifo_stwdata = { NULL, strwsrv, NULL, NULL, NULL, &fifowm_info };
159 
160 extern kmutex_t	strresources;	/* protects global resources */
161 extern kmutex_t muxifier;	/* single-threads multiplexor creation */
162 
163 static boolean_t msghasdata(mblk_t *bp);
164 #define	msgnodata(bp) (!msghasdata(bp))
165 
166 /*
167  * Stream head locking notes:
168  *	There are four monitors associated with the stream head:
169  *	1. v_stream monitor: in stropen() and strclose() v_lock
170  *		is held while the association of vnode and stream
171  *		head is established or tested for.
172  *	2. open/close/push/pop monitor: sd_lock is held while each
173  *		thread bids for exclusive access to this monitor
174  *		for opening or closing a stream.  In addition, this
175  *		monitor is entered during pushes and pops.  This
176  *		guarantees that during plumbing operations there
177  *		is only one thread trying to change the plumbing.
178  *		Any other threads present in the stream are only
179  *		using the plumbing.
180  *	3. read/write monitor: in the case of read, a thread holds
181  *		sd_lock while trying to get data from the stream
182  *		head queue.  if there is none to fulfill a read
183  *		request, it sets RSLEEP and calls cv_wait_sig() down
184  *		in strwaitq() to await the arrival of new data.
185  *		when new data arrives in strrput(), sd_lock is acquired
186  *		before testing for RSLEEP and calling cv_broadcast().
187  *		the behavior of strwrite(), strwsrv(), and WSLEEP
188  *		mirror this.
189  *	4. ioctl monitor: sd_lock is gotten to ensure that only one
190  *		thread is doing an ioctl at a time.
191  */
192 
193 static int
194 push_mod(queue_t *qp, dev_t *devp, struct stdata *stp, const char *name,
195     int anchor, cred_t *crp, uint_t anchor_zoneid)
196 {
197 	int error;
198 	fmodsw_impl_t *fp;
199 
200 	if (stp->sd_flag & (STRHUP|STRDERR|STWRERR)) {
201 		error = (stp->sd_flag & STRHUP) ? ENXIO : EIO;
202 		return (error);
203 	}
204 	if (stp->sd_pushcnt >= nstrpush) {
205 		return (EINVAL);
206 	}
207 
208 	if ((fp = fmodsw_find(name, FMODSW_HOLD | FMODSW_LOAD)) == NULL) {
209 		stp->sd_flag |= STREOPENFAIL;
210 		return (EINVAL);
211 	}
212 
213 	/*
214 	 * push new module and call its open routine via qattach
215 	 */
216 	if ((error = qattach(qp, devp, 0, crp, fp, B_FALSE)) != 0)
217 		return (error);
218 
219 	/*
220 	 * Check to see if caller wants a STREAMS anchor
221 	 * put at this place in the stream, and add if so.
222 	 */
223 	mutex_enter(&stp->sd_lock);
224 	if (anchor == stp->sd_pushcnt) {
225 		stp->sd_anchor = stp->sd_pushcnt;
226 		stp->sd_anchorzone = anchor_zoneid;
227 	}
228 	mutex_exit(&stp->sd_lock);
229 
230 	return (0);
231 }
232 
233 /*
234  * Open a stream device.
235  */
236 int
237 stropen(vnode_t *vp, dev_t *devp, int flag, cred_t *crp)
238 {
239 	struct stdata *stp;
240 	queue_t *qp;
241 	int s;
242 	dev_t dummydev;
243 	struct autopush *ap;
244 	int error = 0;
245 	ssize_t	rmin, rmax;
246 	int cloneopen;
247 	queue_t *brq;
248 	major_t major;
249 	str_stack_t *ss;
250 	zoneid_t zoneid;
251 	uint_t anchor;
252 
253 #ifdef C2_AUDIT
254 	if (audit_active)
255 		audit_stropen(vp, devp, flag, crp);
256 #endif
257 
258 	/*
259 	 * If the stream already exists, wait for any open in progress
260 	 * to complete, then call the open function of each module and
261 	 * driver in the stream.  Otherwise create the stream.
262 	 */
263 	TRACE_1(TR_FAC_STREAMS_FR, TR_STROPEN, "stropen:%p", vp);
264 retry:
265 	mutex_enter(&vp->v_lock);
266 	if ((stp = vp->v_stream) != NULL) {
267 
268 		/*
269 		 * Waiting for stream to be created to device
270 		 * due to another open.
271 		 */
272 	    mutex_exit(&vp->v_lock);
273 
274 	    if (STRMATED(stp)) {
275 		struct stdata *strmatep = stp->sd_mate;
276 
277 		STRLOCKMATES(stp);
278 		if (strmatep->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
279 			if (flag & (FNDELAY|FNONBLOCK)) {
280 				error = EAGAIN;
281 				mutex_exit(&strmatep->sd_lock);
282 				goto ckreturn;
283 			}
284 			mutex_exit(&stp->sd_lock);
285 			if (!cv_wait_sig(&strmatep->sd_monitor,
286 			    &strmatep->sd_lock)) {
287 				error = EINTR;
288 				mutex_exit(&strmatep->sd_lock);
289 				mutex_enter(&stp->sd_lock);
290 				goto ckreturn;
291 			}
292 			mutex_exit(&strmatep->sd_lock);
293 			goto retry;
294 		}
295 		if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
296 			if (flag & (FNDELAY|FNONBLOCK)) {
297 				error = EAGAIN;
298 				mutex_exit(&strmatep->sd_lock);
299 				goto ckreturn;
300 			}
301 			mutex_exit(&strmatep->sd_lock);
302 			if (!cv_wait_sig(&stp->sd_monitor, &stp->sd_lock)) {
303 				error = EINTR;
304 				goto ckreturn;
305 			}
306 			mutex_exit(&stp->sd_lock);
307 			goto retry;
308 		}
309 
310 		if (stp->sd_flag & (STRDERR|STWRERR)) {
311 			error = EIO;
312 			mutex_exit(&strmatep->sd_lock);
313 			goto ckreturn;
314 		}
315 
316 		stp->sd_flag |= STWOPEN;
317 		STRUNLOCKMATES(stp);
318 	    } else {
319 		mutex_enter(&stp->sd_lock);
320 		if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
321 			if (flag & (FNDELAY|FNONBLOCK)) {
322 				error = EAGAIN;
323 				goto ckreturn;
324 			}
325 			if (!cv_wait_sig(&stp->sd_monitor, &stp->sd_lock)) {
326 				error = EINTR;
327 				goto ckreturn;
328 			}
329 			mutex_exit(&stp->sd_lock);
330 			goto retry;  /* could be clone! */
331 		}
332 
333 		if (stp->sd_flag & (STRDERR|STWRERR)) {
334 			error = EIO;
335 			goto ckreturn;
336 		}
337 
338 		stp->sd_flag |= STWOPEN;
339 		mutex_exit(&stp->sd_lock);
340 	    }
341 
342 		/*
343 		 * Open all modules and devices down stream to notify
344 		 * that another user is streaming.  For modules, set the
345 		 * last argument to MODOPEN and do not pass any open flags.
346 		 * Ignore dummydev since this is not the first open.
347 		 */
348 	    claimstr(stp->sd_wrq);
349 	    qp = stp->sd_wrq;
350 	    while (_SAMESTR(qp)) {
351 		qp = qp->q_next;
352 		if ((error = qreopen(_RD(qp), devp, flag, crp)) != 0)
353 			break;
354 	    }
355 	    releasestr(stp->sd_wrq);
356 	    mutex_enter(&stp->sd_lock);
357 	    stp->sd_flag &= ~(STRHUP|STWOPEN|STRDERR|STWRERR);
358 	    stp->sd_rerror = 0;
359 	    stp->sd_werror = 0;
360 ckreturn:
361 	    cv_broadcast(&stp->sd_monitor);
362 	    mutex_exit(&stp->sd_lock);
363 	    return (error);
364 	}
365 
366 	/*
367 	 * This vnode isn't streaming.  SPECFS already
368 	 * checked for multiple vnodes pointing to the
369 	 * same stream, so create a stream to the driver.
370 	 */
371 	qp = allocq();
372 	stp = shalloc(qp);
373 
374 	/*
375 	 * Initialize stream head.  shalloc() has given us
376 	 * exclusive access, and we have the vnode locked;
377 	 * we can do whatever we want with stp.
378 	 */
379 	stp->sd_flag = STWOPEN;
380 	stp->sd_siglist = NULL;
381 	stp->sd_pollist.ph_list = NULL;
382 	stp->sd_sigflags = 0;
383 	stp->sd_mark = NULL;
384 	stp->sd_closetime = STRTIMOUT;
385 	stp->sd_sidp = NULL;
386 	stp->sd_pgidp = NULL;
387 	stp->sd_vnode = vp;
388 	stp->sd_rerror = 0;
389 	stp->sd_werror = 0;
390 	stp->sd_wroff = 0;
391 	stp->sd_tail = 0;
392 	stp->sd_iocblk = NULL;
393 	stp->sd_pushcnt = 0;
394 	stp->sd_qn_minpsz = 0;
395 	stp->sd_qn_maxpsz = INFPSZ - 1;	/* used to check for initialization */
396 	stp->sd_maxblk = INFPSZ;
397 	qp->q_ptr = _WR(qp)->q_ptr = stp;
398 	STREAM(qp) = STREAM(_WR(qp)) = stp;
399 	vp->v_stream = stp;
400 	mutex_exit(&vp->v_lock);
401 	if (vp->v_type == VFIFO) {
402 		stp->sd_flag |= OLDNDELAY;
403 		/*
404 		 * This means, both for pipes and fifos
405 		 * strwrite will send SIGPIPE if the other
406 		 * end is closed. For putmsg it depends
407 		 * on whether it is a XPG4_2 application
408 		 * or not
409 		 */
410 		stp->sd_wput_opt = SW_SIGPIPE;
411 
412 		/* setq might sleep in kmem_alloc - avoid holding locks. */
413 		setq(qp, &fifo_strdata, &fifo_stwdata, NULL, QMTSAFE,
414 		    SQ_CI|SQ_CO, B_FALSE);
415 
416 		set_qend(qp);
417 		stp->sd_strtab = fifo_getinfo();
418 		_WR(qp)->q_nfsrv = _WR(qp);
419 		qp->q_nfsrv = qp;
420 		/*
421 		 * Wake up others that are waiting for stream to be created.
422 		 */
423 		mutex_enter(&stp->sd_lock);
424 		/*
425 		 * nothing is be pushed on stream yet, so
426 		 * optimized stream head packetsizes are just that
427 		 * of the read queue
428 		 */
429 		stp->sd_qn_minpsz = qp->q_minpsz;
430 		stp->sd_qn_maxpsz = qp->q_maxpsz;
431 		stp->sd_flag &= ~STWOPEN;
432 		goto fifo_opendone;
433 	}
434 	/* setq might sleep in kmem_alloc - avoid holding locks. */
435 	setq(qp, &strdata, &stwdata, NULL, QMTSAFE, SQ_CI|SQ_CO, B_FALSE);
436 
437 	set_qend(qp);
438 
439 	/*
440 	 * Open driver and create stream to it (via qattach).
441 	 */
442 	cloneopen = (getmajor(*devp) == clone_major);
443 	if ((error = qattach(qp, devp, flag, crp, NULL, B_FALSE)) != 0) {
444 		mutex_enter(&vp->v_lock);
445 		vp->v_stream = NULL;
446 		mutex_exit(&vp->v_lock);
447 		mutex_enter(&stp->sd_lock);
448 		cv_broadcast(&stp->sd_monitor);
449 		mutex_exit(&stp->sd_lock);
450 		freeq(_RD(qp));
451 		shfree(stp);
452 		return (error);
453 	}
454 	/*
455 	 * Set sd_strtab after open in order to handle clonable drivers
456 	 */
457 	stp->sd_strtab = STREAMSTAB(getmajor(*devp));
458 
459 	/*
460 	 * Historical note: dummydev used to be be prior to the initial
461 	 * open (via qattach above), which made the value seen
462 	 * inconsistent between an I_PUSH and an autopush of a module.
463 	 */
464 	dummydev = *devp;
465 
466 	/*
467 	 * For clone open of old style (Q not associated) network driver,
468 	 * push DRMODNAME module to handle DL_ATTACH/DL_DETACH
469 	 */
470 	brq = _RD(_WR(qp)->q_next);
471 	major = getmajor(*devp);
472 	if (push_drcompat && cloneopen && NETWORK_DRV(major) &&
473 	    ((brq->q_flag & _QASSOCIATED) == 0)) {
474 		if (push_mod(qp, &dummydev, stp, DRMODNAME, 0, crp, 0) != 0)
475 			cmn_err(CE_WARN, "cannot push " DRMODNAME
476 			    " streams module");
477 	}
478 
479 	/*
480 	 * Check for autopush. Start with the global zone. If not found
481 	 * check in the local zone.
482 	 */
483 	zoneid = GLOBAL_ZONEID;
484 retryap:
485 	ss = netstack_find_by_stackid(zoneid_to_netstackid(zoneid))->
486 	    netstack_str;
487 	if ((ap = sad_ap_find_by_dev(*devp, ss)) == NULL) {
488 		netstack_rele(ss->ss_netstack);
489 		if (zoneid == GLOBAL_ZONEID) {
490 			/*
491 			 * None found. Also look in the zone's autopush table.
492 			 */
493 			zoneid = crgetzoneid(crp);
494 			if (zoneid != GLOBAL_ZONEID)
495 				goto retryap;
496 		}
497 		goto opendone;
498 	}
499 	anchor = ap->ap_anchor;
500 	zoneid = crgetzoneid(crp);
501 	for (s = 0; s < ap->ap_npush; s++) {
502 		error = push_mod(qp, &dummydev, stp, ap->ap_list[s],
503 		    anchor, crp, zoneid);
504 		if (error != 0)
505 			break;
506 	}
507 	sad_ap_rele(ap, ss);
508 	netstack_rele(ss->ss_netstack);
509 
510 	/*
511 	 * let specfs know that open failed part way through
512 	 */
513 	if (error) {
514 		mutex_enter(&stp->sd_lock);
515 		stp->sd_flag |= STREOPENFAIL;
516 		mutex_exit(&stp->sd_lock);
517 	}
518 
519 opendone:
520 
521 	/*
522 	 * Wake up others that are waiting for stream to be created.
523 	 */
524 	mutex_enter(&stp->sd_lock);
525 	stp->sd_flag &= ~STWOPEN;
526 
527 	/*
528 	 * As a performance concern we are caching the values of
529 	 * q_minpsz and q_maxpsz of the module below the stream
530 	 * head in the stream head.
531 	 */
532 	mutex_enter(QLOCK(stp->sd_wrq->q_next));
533 	rmin = stp->sd_wrq->q_next->q_minpsz;
534 	rmax = stp->sd_wrq->q_next->q_maxpsz;
535 	mutex_exit(QLOCK(stp->sd_wrq->q_next));
536 
537 	/* do this processing here as a performance concern */
538 	if (strmsgsz != 0) {
539 		if (rmax == INFPSZ)
540 			rmax = strmsgsz;
541 		else
542 			rmax = MIN(strmsgsz, rmax);
543 	}
544 
545 	mutex_enter(QLOCK(stp->sd_wrq));
546 	stp->sd_qn_minpsz = rmin;
547 	stp->sd_qn_maxpsz = rmax;
548 	mutex_exit(QLOCK(stp->sd_wrq));
549 
550 fifo_opendone:
551 	cv_broadcast(&stp->sd_monitor);
552 	mutex_exit(&stp->sd_lock);
553 	return (error);
554 }
555 
556 static int strsink(queue_t *, mblk_t *);
557 static struct qinit deadrend = {
558 	strsink, NULL, NULL, NULL, NULL, &strm_info, NULL
559 };
560 static struct qinit deadwend = {
561 	NULL, NULL, NULL, NULL, NULL, &stwm_info, NULL
562 };
563 
564 /*
565  * Close a stream.
566  * This is called from closef() on the last close of an open stream.
567  * Strclean() will already have removed the siglist and pollist
568  * information, so all that remains is to remove all multiplexor links
569  * for the stream, pop all the modules (and the driver), and free the
570  * stream structure.
571  */
572 
573 int
574 strclose(struct vnode *vp, int flag, cred_t *crp)
575 {
576 	struct stdata *stp;
577 	queue_t *qp;
578 	int rval;
579 	int freestp = 1;
580 	queue_t *rmq;
581 
582 #ifdef C2_AUDIT
583 	if (audit_active)
584 		audit_strclose(vp, flag, crp);
585 #endif
586 
587 	TRACE_1(TR_FAC_STREAMS_FR,
588 		TR_STRCLOSE, "strclose:%p", vp);
589 	ASSERT(vp->v_stream);
590 
591 	stp = vp->v_stream;
592 	ASSERT(!(stp->sd_flag & STPLEX));
593 	qp = stp->sd_wrq;
594 
595 	/*
596 	 * Needed so that strpoll will return non-zero for this fd.
597 	 * Note that with POLLNOERR STRHUP does still cause POLLHUP.
598 	 */
599 	mutex_enter(&stp->sd_lock);
600 	stp->sd_flag |= STRHUP;
601 	mutex_exit(&stp->sd_lock);
602 
603 	/*
604 	 * If the registered process or process group did not have an
605 	 * open instance of this stream then strclean would not be
606 	 * called. Thus at the time of closing all remaining siglist entries
607 	 * are removed.
608 	 */
609 	if (stp->sd_siglist != NULL)
610 		strcleanall(vp);
611 
612 	ASSERT(stp->sd_siglist == NULL);
613 	ASSERT(stp->sd_sigflags == 0);
614 
615 	if (STRMATED(stp)) {
616 		struct stdata *strmatep = stp->sd_mate;
617 		int waited = 1;
618 
619 		STRLOCKMATES(stp);
620 		while (waited) {
621 			waited = 0;
622 			while (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
623 				mutex_exit(&strmatep->sd_lock);
624 				cv_wait(&stp->sd_monitor, &stp->sd_lock);
625 				mutex_exit(&stp->sd_lock);
626 				STRLOCKMATES(stp);
627 				waited = 1;
628 			}
629 			while (strmatep->sd_flag &
630 			    (STWOPEN|STRCLOSE|STRPLUMB)) {
631 				mutex_exit(&stp->sd_lock);
632 				cv_wait(&strmatep->sd_monitor,
633 				    &strmatep->sd_lock);
634 				mutex_exit(&strmatep->sd_lock);
635 				STRLOCKMATES(stp);
636 				waited = 1;
637 			}
638 		}
639 		stp->sd_flag |= STRCLOSE;
640 		STRUNLOCKMATES(stp);
641 	} else {
642 		mutex_enter(&stp->sd_lock);
643 		stp->sd_flag |= STRCLOSE;
644 		mutex_exit(&stp->sd_lock);
645 	}
646 
647 	ASSERT(qp->q_first == NULL);	/* No more delayed write */
648 
649 	/* Check if an I_LINK was ever done on this stream */
650 	if (stp->sd_flag & STRHASLINKS) {
651 		netstack_t *ns;
652 		str_stack_t *ss;
653 
654 		ns = netstack_find_by_cred(crp);
655 		ASSERT(ns != NULL);
656 		ss = ns->netstack_str;
657 		ASSERT(ss != NULL);
658 
659 		(void) munlinkall(stp, LINKCLOSE|LINKNORMAL, crp, &rval, ss);
660 		netstack_rele(ss->ss_netstack);
661 	}
662 
663 	while (_SAMESTR(qp)) {
664 		/*
665 		 * Holding sd_lock prevents q_next from changing in
666 		 * this stream.
667 		 */
668 		mutex_enter(&stp->sd_lock);
669 		if (!(flag & (FNDELAY|FNONBLOCK)) && (stp->sd_closetime > 0)) {
670 
671 			/*
672 			 * sleep until awakened by strwsrv() or timeout
673 			 */
674 			for (;;) {
675 				mutex_enter(QLOCK(qp->q_next));
676 				if (!(qp->q_next->q_mblkcnt)) {
677 					mutex_exit(QLOCK(qp->q_next));
678 					break;
679 				}
680 				stp->sd_flag |= WSLEEP;
681 
682 				/* ensure strwsrv gets enabled */
683 				qp->q_next->q_flag |= QWANTW;
684 				mutex_exit(QLOCK(qp->q_next));
685 				/* get out if we timed out or recv'd a signal */
686 				if (str_cv_wait(&qp->q_wait, &stp->sd_lock,
687 				    stp->sd_closetime, 0) <= 0) {
688 					break;
689 				}
690 			}
691 			stp->sd_flag &= ~WSLEEP;
692 		}
693 		mutex_exit(&stp->sd_lock);
694 
695 		rmq = qp->q_next;
696 		if (rmq->q_flag & QISDRV) {
697 			ASSERT(!_SAMESTR(rmq));
698 			wait_sq_svc(_RD(qp)->q_syncq);
699 		}
700 
701 		qdetach(_RD(rmq), 1, flag, crp, B_FALSE);
702 	}
703 
704 	/*
705 	 * Since we call pollwakeup in close() now, the poll list should
706 	 * be empty in most cases. The only exception is the layered devices
707 	 * (e.g. the console drivers with redirection modules pushed on top
708 	 * of it).  We have to do this after calling qdetach() because
709 	 * the redirection module won't have torn down the console
710 	 * redirection until after qdetach() has been invoked.
711 	 */
712 	if (stp->sd_pollist.ph_list != NULL) {
713 		pollwakeup(&stp->sd_pollist, POLLERR);
714 		pollhead_clean(&stp->sd_pollist);
715 	}
716 	ASSERT(stp->sd_pollist.ph_list == NULL);
717 	ASSERT(stp->sd_sidp == NULL);
718 	ASSERT(stp->sd_pgidp == NULL);
719 
720 	/* Prevent qenable from re-enabling the stream head queue */
721 	disable_svc(_RD(qp));
722 
723 	/*
724 	 * Wait until service procedure of each queue is
725 	 * run, if QINSERVICE is set.
726 	 */
727 	wait_svc(_RD(qp));
728 
729 	/*
730 	 * Now, flush both queues.
731 	 */
732 	flushq(_RD(qp), FLUSHALL);
733 	flushq(qp, FLUSHALL);
734 
735 	/*
736 	 * If the write queue of the stream head is pointing to a
737 	 * read queue, we have a twisted stream.  If the read queue
738 	 * is alive, convert the stream head queues into a dead end.
739 	 * If the read queue is dead, free the dead pair.
740 	 */
741 	if (qp->q_next && !_SAMESTR(qp)) {
742 		if (qp->q_next->q_qinfo == &deadrend) {	/* half-closed pipe */
743 			flushq(qp->q_next, FLUSHALL); /* ensure no message */
744 			shfree(qp->q_next->q_stream);
745 			freeq(qp->q_next);
746 			freeq(_RD(qp));
747 		} else if (qp->q_next == _RD(qp)) {	/* fifo */
748 			freeq(_RD(qp));
749 		} else {				/* pipe */
750 			freestp = 0;
751 			/*
752 			 * The q_info pointers are never accessed when
753 			 * SQLOCK is held.
754 			 */
755 			ASSERT(qp->q_syncq == _RD(qp)->q_syncq);
756 			mutex_enter(SQLOCK(qp->q_syncq));
757 			qp->q_qinfo = &deadwend;
758 			_RD(qp)->q_qinfo = &deadrend;
759 			mutex_exit(SQLOCK(qp->q_syncq));
760 		}
761 	} else {
762 		freeq(_RD(qp)); /* free stream head queue pair */
763 	}
764 
765 	mutex_enter(&vp->v_lock);
766 	if (stp->sd_iocblk) {
767 		if (stp->sd_iocblk != (mblk_t *)-1) {
768 			freemsg(stp->sd_iocblk);
769 		}
770 		stp->sd_iocblk = NULL;
771 	}
772 	stp->sd_vnode = NULL;
773 	vp->v_stream = NULL;
774 	mutex_exit(&vp->v_lock);
775 	mutex_enter(&stp->sd_lock);
776 	stp->sd_flag &= ~STRCLOSE;
777 	cv_broadcast(&stp->sd_monitor);
778 	mutex_exit(&stp->sd_lock);
779 
780 	if (freestp)
781 		shfree(stp);
782 	return (0);
783 }
784 
785 static int
786 strsink(queue_t *q, mblk_t *bp)
787 {
788 	struct copyresp *resp;
789 
790 	switch (bp->b_datap->db_type) {
791 	case M_FLUSH:
792 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
793 			*bp->b_rptr &= ~FLUSHR;
794 			bp->b_flag |= MSGNOLOOP;
795 			/*
796 			 * Protect against the driver passing up
797 			 * messages after it has done a qprocsoff.
798 			 */
799 			if (_OTHERQ(q)->q_next == NULL)
800 				freemsg(bp);
801 			else
802 				qreply(q, bp);
803 		} else {
804 			freemsg(bp);
805 		}
806 		break;
807 
808 	case M_COPYIN:
809 	case M_COPYOUT:
810 		if (bp->b_cont) {
811 			freemsg(bp->b_cont);
812 			bp->b_cont = NULL;
813 		}
814 		bp->b_datap->db_type = M_IOCDATA;
815 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
816 		resp = (struct copyresp *)bp->b_rptr;
817 		resp->cp_rval = (caddr_t)1;	/* failure */
818 		/*
819 		 * Protect against the driver passing up
820 		 * messages after it has done a qprocsoff.
821 		 */
822 		if (_OTHERQ(q)->q_next == NULL)
823 			freemsg(bp);
824 		else
825 			qreply(q, bp);
826 		break;
827 
828 	case M_IOCTL:
829 		if (bp->b_cont) {
830 			freemsg(bp->b_cont);
831 			bp->b_cont = NULL;
832 		}
833 		bp->b_datap->db_type = M_IOCNAK;
834 		/*
835 		 * Protect against the driver passing up
836 		 * messages after it has done a qprocsoff.
837 		 */
838 		if (_OTHERQ(q)->q_next == NULL)
839 			freemsg(bp);
840 		else
841 			qreply(q, bp);
842 		break;
843 
844 	default:
845 		freemsg(bp);
846 		break;
847 	}
848 
849 	return (0);
850 }
851 
852 /*
853  * Clean up after a process when it closes a stream.  This is called
854  * from closef for all closes, whereas strclose is called only for the
855  * last close on a stream.  The siglist is scanned for entries for the
856  * current process, and these are removed.
857  */
858 void
859 strclean(struct vnode *vp)
860 {
861 	strsig_t *ssp, *pssp, *tssp;
862 	stdata_t *stp;
863 	int update = 0;
864 
865 	TRACE_1(TR_FAC_STREAMS_FR,
866 		TR_STRCLEAN, "strclean:%p", vp);
867 	stp = vp->v_stream;
868 	pssp = NULL;
869 	mutex_enter(&stp->sd_lock);
870 	ssp = stp->sd_siglist;
871 	while (ssp) {
872 		if (ssp->ss_pidp == curproc->p_pidp) {
873 			tssp = ssp->ss_next;
874 			if (pssp)
875 				pssp->ss_next = tssp;
876 			else
877 				stp->sd_siglist = tssp;
878 			mutex_enter(&pidlock);
879 			PID_RELE(ssp->ss_pidp);
880 			mutex_exit(&pidlock);
881 			kmem_free(ssp, sizeof (strsig_t));
882 			update = 1;
883 			ssp = tssp;
884 		} else {
885 			pssp = ssp;
886 			ssp = ssp->ss_next;
887 		}
888 	}
889 	if (update) {
890 		stp->sd_sigflags = 0;
891 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
892 			stp->sd_sigflags |= ssp->ss_events;
893 	}
894 	mutex_exit(&stp->sd_lock);
895 }
896 
897 /*
898  * Used on the last close to remove any remaining items on the siglist.
899  * These could be present on the siglist due to I_ESETSIG calls that
900  * use process groups or processed that do not have an open file descriptor
901  * for this stream (Such entries would not be removed by strclean).
902  */
903 static void
904 strcleanall(struct vnode *vp)
905 {
906 	strsig_t *ssp, *nssp;
907 	stdata_t *stp;
908 
909 	stp = vp->v_stream;
910 	mutex_enter(&stp->sd_lock);
911 	ssp = stp->sd_siglist;
912 	stp->sd_siglist = NULL;
913 	while (ssp) {
914 		nssp = ssp->ss_next;
915 		mutex_enter(&pidlock);
916 		PID_RELE(ssp->ss_pidp);
917 		mutex_exit(&pidlock);
918 		kmem_free(ssp, sizeof (strsig_t));
919 		ssp = nssp;
920 	}
921 	stp->sd_sigflags = 0;
922 	mutex_exit(&stp->sd_lock);
923 }
924 
925 /*
926  * Retrieve the next message from the logical stream head read queue
927  * using either rwnext (if sync stream) or getq_noenab.
928  * It is the callers responsibility to call qbackenable after
929  * it is finished with the message. The caller should not call
930  * qbackenable until after any putback calls to avoid spurious backenabling.
931  */
932 mblk_t *
933 strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
934     int *errorp)
935 {
936 	mblk_t *bp;
937 	int error;
938 
939 	ASSERT(MUTEX_HELD(&stp->sd_lock));
940 	/* Holding sd_lock prevents the read queue from changing  */
941 
942 	if (uiop != NULL && stp->sd_struiordq != NULL &&
943 	    q->q_first == NULL &&
944 	    (!first || (stp->sd_wakeq & RSLEEP))) {
945 		/*
946 		 * Stream supports rwnext() for the read side.
947 		 * If this is the first time we're called by e.g. strread
948 		 * only do the downcall if there is a deferred wakeup
949 		 * (registered in sd_wakeq).
950 		 */
951 		struiod_t uiod;
952 
953 		if (first)
954 			stp->sd_wakeq &= ~RSLEEP;
955 
956 		(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
957 			sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
958 		uiod.d_mp = 0;
959 		/*
960 		 * Mark that a thread is in rwnext on the read side
961 		 * to prevent strrput from nacking ioctls immediately.
962 		 * When the last concurrent rwnext returns
963 		 * the ioctls are nack'ed.
964 		 */
965 		ASSERT(MUTEX_HELD(&stp->sd_lock));
966 		stp->sd_struiodnak++;
967 		/*
968 		 * Note: rwnext will drop sd_lock.
969 		 */
970 		error = rwnext(q, &uiod);
971 		ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
972 		mutex_enter(&stp->sd_lock);
973 		stp->sd_struiodnak--;
974 		while (stp->sd_struiodnak == 0 &&
975 		    ((bp = stp->sd_struionak) != NULL)) {
976 			stp->sd_struionak = bp->b_next;
977 			bp->b_next = NULL;
978 			bp->b_datap->db_type = M_IOCNAK;
979 			/*
980 			 * Protect against the driver passing up
981 			 * messages after it has done a qprocsoff.
982 			 */
983 			if (_OTHERQ(q)->q_next == NULL)
984 				freemsg(bp);
985 			else {
986 				mutex_exit(&stp->sd_lock);
987 				qreply(q, bp);
988 				mutex_enter(&stp->sd_lock);
989 			}
990 		}
991 		ASSERT(MUTEX_HELD(&stp->sd_lock));
992 		if (error == 0 || error == EWOULDBLOCK) {
993 			if ((bp = uiod.d_mp) != NULL) {
994 				*errorp = 0;
995 				ASSERT(MUTEX_HELD(&stp->sd_lock));
996 				return (bp);
997 			}
998 			error = 0;
999 		} else if (error == EINVAL) {
1000 			/*
1001 			 * The stream plumbing must have
1002 			 * changed while we were away, so
1003 			 * just turn off rwnext()s.
1004 			 */
1005 			error = 0;
1006 		} else if (error == EBUSY) {
1007 			/*
1008 			 * The module might have data in transit using putnext
1009 			 * Fall back on waiting + getq.
1010 			 */
1011 			error = 0;
1012 		} else {
1013 			*errorp = error;
1014 			ASSERT(MUTEX_HELD(&stp->sd_lock));
1015 			return (NULL);
1016 		}
1017 		/*
1018 		 * Try a getq in case a rwnext() generated mblk
1019 		 * has bubbled up via strrput().
1020 		 */
1021 	}
1022 	*errorp = 0;
1023 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1024 	return (getq_noenab(q));
1025 }
1026 
1027 /*
1028  * Copy out the message pointed to by `bp' into the uio pointed to by `uiop'.
1029  * If the message does not fit in the uio the remainder of it is returned;
1030  * otherwise NULL is returned.  Any embedded zero-length mblk_t's are
1031  * consumed, even if uio_resid reaches zero.  On error, `*errorp' is set to
1032  * the error code, the message is consumed, and NULL is returned.
1033  */
1034 static mblk_t *
1035 struiocopyout(mblk_t *bp, struct uio *uiop, int *errorp)
1036 {
1037 	int error;
1038 	ptrdiff_t n;
1039 	mblk_t *nbp;
1040 
1041 	ASSERT(bp->b_wptr >= bp->b_rptr);
1042 
1043 	do {
1044 		if ((n = MIN(uiop->uio_resid, MBLKL(bp))) != 0) {
1045 			ASSERT(n > 0);
1046 
1047 			error = uiomove(bp->b_rptr, n, UIO_READ, uiop);
1048 			if (error != 0) {
1049 				freemsg(bp);
1050 				*errorp = error;
1051 				return (NULL);
1052 			}
1053 		}
1054 
1055 		bp->b_rptr += n;
1056 		while (bp != NULL && (bp->b_rptr >= bp->b_wptr)) {
1057 			nbp = bp;
1058 			bp = bp->b_cont;
1059 			freeb(nbp);
1060 		}
1061 	} while (bp != NULL && uiop->uio_resid > 0);
1062 
1063 	*errorp = 0;
1064 	return (bp);
1065 }
1066 
1067 /*
1068  * Read a stream according to the mode flags in sd_flag:
1069  *
1070  * (default mode)		- Byte stream, msg boundaries are ignored
1071  * RD_MSGDIS (msg discard)	- Read on msg boundaries and throw away
1072  *				any data remaining in msg
1073  * RD_MSGNODIS (msg non-discard) - Read on msg boundaries and put back
1074  *				any remaining data on head of read queue
1075  *
1076  * Consume readable messages on the front of the queue until
1077  * ttolwp(curthread)->lwp_count
1078  * is satisfied, the readable messages are exhausted, or a message
1079  * boundary is reached in a message mode.  If no data was read and
1080  * the stream was not opened with the NDELAY flag, block until data arrives.
1081  * Otherwise return the data read and update the count.
1082  *
1083  * In default mode a 0 length message signifies end-of-file and terminates
1084  * a read in progress.  The 0 length message is removed from the queue
1085  * only if it is the only message read (no data is read).
1086  *
1087  * An attempt to read an M_PROTO or M_PCPROTO message results in an
1088  * EBADMSG error return, unless either RD_PROTDAT or RD_PROTDIS are set.
1089  * If RD_PROTDAT is set, M_PROTO and M_PCPROTO messages are read as data.
1090  * If RD_PROTDIS is set, the M_PROTO and M_PCPROTO parts of the message
1091  * are unlinked from and M_DATA blocks in the message, the protos are
1092  * thrown away, and the data is read.
1093  */
1094 /* ARGSUSED */
1095 int
1096 strread(struct vnode *vp, struct uio *uiop, cred_t *crp)
1097 {
1098 	struct stdata *stp;
1099 	mblk_t *bp, *nbp;
1100 	queue_t *q;
1101 	int error = 0;
1102 	uint_t old_sd_flag;
1103 	int first;
1104 	char rflg;
1105 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
1106 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
1107 	short delim;
1108 	unsigned char pri = 0;
1109 	char waitflag;
1110 	unsigned char type;
1111 
1112 	TRACE_1(TR_FAC_STREAMS_FR,
1113 		TR_STRREAD_ENTER, "strread:%p", vp);
1114 	ASSERT(vp->v_stream);
1115 	stp = vp->v_stream;
1116 
1117 	mutex_enter(&stp->sd_lock);
1118 
1119 	if ((error = i_straccess(stp, JCREAD)) != 0) {
1120 		mutex_exit(&stp->sd_lock);
1121 		return (error);
1122 	}
1123 
1124 	if (stp->sd_flag & (STRDERR|STPLEX)) {
1125 		error = strgeterr(stp, STRDERR|STPLEX, 0);
1126 		if (error != 0) {
1127 			mutex_exit(&stp->sd_lock);
1128 			return (error);
1129 		}
1130 	}
1131 
1132 	/*
1133 	 * Loop terminates when uiop->uio_resid == 0.
1134 	 */
1135 	rflg = 0;
1136 	waitflag = READWAIT;
1137 	q = _RD(stp->sd_wrq);
1138 	for (;;) {
1139 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1140 		old_sd_flag = stp->sd_flag;
1141 		mark = 0;
1142 		delim = 0;
1143 		first = 1;
1144 		while ((bp = strget(stp, q, uiop, first, &error)) == NULL) {
1145 			int done = 0;
1146 
1147 			ASSERT(MUTEX_HELD(&stp->sd_lock));
1148 
1149 			if (error != 0)
1150 				goto oops;
1151 
1152 			if (stp->sd_flag & (STRHUP|STREOF)) {
1153 				goto oops;
1154 			}
1155 			if (rflg && !(stp->sd_flag & STRDELIM)) {
1156 				goto oops;
1157 			}
1158 			/*
1159 			 * If a read(fd,buf,0) has been done, there is no
1160 			 * need to sleep. We always have zero bytes to
1161 			 * return.
1162 			 */
1163 			if (uiop->uio_resid == 0) {
1164 				goto oops;
1165 			}
1166 
1167 			qbackenable(q, 0);
1168 
1169 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_WAIT,
1170 				"strread calls strwaitq:%p, %p, %p",
1171 				vp, uiop, crp);
1172 			if ((error = strwaitq(stp, waitflag, uiop->uio_resid,
1173 			    uiop->uio_fmode, -1, &done)) != 0 || done) {
1174 				TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_DONE,
1175 					"strread error or done:%p, %p, %p",
1176 					vp, uiop, crp);
1177 				if ((uiop->uio_fmode & FNDELAY) &&
1178 				    (stp->sd_flag & OLDNDELAY) &&
1179 				    (error == EAGAIN))
1180 					error = 0;
1181 				goto oops;
1182 			}
1183 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_AWAKE,
1184 				"strread awakes:%p, %p, %p", vp, uiop, crp);
1185 			if ((error = i_straccess(stp, JCREAD)) != 0) {
1186 				goto oops;
1187 			}
1188 			first = 0;
1189 		}
1190 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1191 		ASSERT(bp);
1192 		pri = bp->b_band;
1193 		/*
1194 		 * Extract any mark information. If the message is not
1195 		 * completely consumed this information will be put in the mblk
1196 		 * that is putback.
1197 		 * If MSGMARKNEXT is set and the message is completely consumed
1198 		 * the STRATMARK flag will be set below. Likewise, if
1199 		 * MSGNOTMARKNEXT is set and the message is
1200 		 * completely consumed STRNOTATMARK will be set.
1201 		 *
1202 		 * For some unknown reason strread only breaks the read at the
1203 		 * last mark.
1204 		 */
1205 		mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
1206 		ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
1207 			(MSGMARKNEXT|MSGNOTMARKNEXT));
1208 		if (mark != 0 && bp == stp->sd_mark) {
1209 			if (rflg) {
1210 				putback(stp, q, bp, pri);
1211 				goto oops;
1212 			}
1213 			mark |= _LASTMARK;
1214 			stp->sd_mark = NULL;
1215 		}
1216 		if ((stp->sd_flag & STRDELIM) && (bp->b_flag & MSGDELIM))
1217 			delim = 1;
1218 		mutex_exit(&stp->sd_lock);
1219 
1220 		if (STREAM_NEEDSERVICE(stp))
1221 			stream_runservice(stp);
1222 
1223 		type = bp->b_datap->db_type;
1224 
1225 		switch (type) {
1226 
1227 		case M_DATA:
1228 ismdata:
1229 			if (msgnodata(bp)) {
1230 				if (mark || delim) {
1231 					freemsg(bp);
1232 				} else if (rflg) {
1233 
1234 					/*
1235 					 * If already read data put zero
1236 					 * length message back on queue else
1237 					 * free msg and return 0.
1238 					 */
1239 					bp->b_band = pri;
1240 					mutex_enter(&stp->sd_lock);
1241 					putback(stp, q, bp, pri);
1242 					mutex_exit(&stp->sd_lock);
1243 				} else {
1244 					freemsg(bp);
1245 				}
1246 				error =  0;
1247 				goto oops1;
1248 			}
1249 
1250 			rflg = 1;
1251 			waitflag |= NOINTR;
1252 			bp = struiocopyout(bp, uiop, &error);
1253 			if (error != 0)
1254 				goto oops1;
1255 
1256 			mutex_enter(&stp->sd_lock);
1257 			if (bp) {
1258 				/*
1259 				 * Have remaining data in message.
1260 				 * Free msg if in discard mode.
1261 				 */
1262 				if (stp->sd_read_opt & RD_MSGDIS) {
1263 					freemsg(bp);
1264 				} else {
1265 					bp->b_band = pri;
1266 					if ((mark & _LASTMARK) &&
1267 					    (stp->sd_mark == NULL))
1268 						stp->sd_mark = bp;
1269 					bp->b_flag |= mark & ~_LASTMARK;
1270 					if (delim)
1271 						bp->b_flag |= MSGDELIM;
1272 					if (msgnodata(bp))
1273 						freemsg(bp);
1274 					else
1275 						putback(stp, q, bp, pri);
1276 				}
1277 			} else {
1278 				/*
1279 				 * Consumed the complete message.
1280 				 * Move the MSG*MARKNEXT information
1281 				 * to the stream head just in case
1282 				 * the read queue becomes empty.
1283 				 *
1284 				 * If the stream head was at the mark
1285 				 * (STRATMARK) before we dropped sd_lock above
1286 				 * and some data was consumed then we have
1287 				 * moved past the mark thus STRATMARK is
1288 				 * cleared. However, if a message arrived in
1289 				 * strrput during the copyout above causing
1290 				 * STRATMARK to be set we can not clear that
1291 				 * flag.
1292 				 */
1293 				if (mark &
1294 				    (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
1295 					if (mark & MSGMARKNEXT) {
1296 						stp->sd_flag &= ~STRNOTATMARK;
1297 						stp->sd_flag |= STRATMARK;
1298 					} else if (mark & MSGNOTMARKNEXT) {
1299 						stp->sd_flag &= ~STRATMARK;
1300 						stp->sd_flag |= STRNOTATMARK;
1301 					} else {
1302 						stp->sd_flag &=
1303 						    ~(STRATMARK|STRNOTATMARK);
1304 					}
1305 				} else if (rflg && (old_sd_flag & STRATMARK)) {
1306 					stp->sd_flag &= ~STRATMARK;
1307 				}
1308 			}
1309 
1310 			/*
1311 			 * Check for signal messages at the front of the read
1312 			 * queue and generate the signal(s) if appropriate.
1313 			 * The only signal that can be on queue is M_SIG at
1314 			 * this point.
1315 			 */
1316 			while ((((bp = q->q_first)) != NULL) &&
1317 				(bp->b_datap->db_type == M_SIG)) {
1318 				bp = getq_noenab(q);
1319 				/*
1320 				 * sd_lock is held so the content of the
1321 				 * read queue can not change.
1322 				 */
1323 				ASSERT(bp != NULL &&
1324 					bp->b_datap->db_type == M_SIG);
1325 				strsignal_nolock(stp, *bp->b_rptr,
1326 					(int32_t)bp->b_band);
1327 				mutex_exit(&stp->sd_lock);
1328 				freemsg(bp);
1329 				if (STREAM_NEEDSERVICE(stp))
1330 					stream_runservice(stp);
1331 				mutex_enter(&stp->sd_lock);
1332 			}
1333 
1334 			if ((uiop->uio_resid == 0) || (mark & _LASTMARK) ||
1335 			    delim ||
1336 			    (stp->sd_read_opt & (RD_MSGDIS|RD_MSGNODIS))) {
1337 				goto oops;
1338 			}
1339 			continue;
1340 
1341 		case M_SIG:
1342 			strsignal(stp, *bp->b_rptr, (int32_t)bp->b_band);
1343 			freemsg(bp);
1344 			mutex_enter(&stp->sd_lock);
1345 			continue;
1346 
1347 		case M_PROTO:
1348 		case M_PCPROTO:
1349 			/*
1350 			 * Only data messages are readable.
1351 			 * Any others generate an error, unless
1352 			 * RD_PROTDIS or RD_PROTDAT is set.
1353 			 */
1354 			if (stp->sd_read_opt & RD_PROTDAT) {
1355 				for (nbp = bp; nbp; nbp = nbp->b_next) {
1356 				    if ((nbp->b_datap->db_type == M_PROTO) ||
1357 					(nbp->b_datap->db_type == M_PCPROTO))
1358 					nbp->b_datap->db_type = M_DATA;
1359 				    else
1360 					break;
1361 				}
1362 				/*
1363 				 * clear stream head hi pri flag based on
1364 				 * first message
1365 				 */
1366 				if (type == M_PCPROTO) {
1367 					mutex_enter(&stp->sd_lock);
1368 					stp->sd_flag &= ~STRPRI;
1369 					mutex_exit(&stp->sd_lock);
1370 				}
1371 				goto ismdata;
1372 			} else if (stp->sd_read_opt & RD_PROTDIS) {
1373 				/*
1374 				 * discard non-data messages
1375 				 */
1376 				while (bp &&
1377 				    ((bp->b_datap->db_type == M_PROTO) ||
1378 				    (bp->b_datap->db_type == M_PCPROTO))) {
1379 					nbp = unlinkb(bp);
1380 					freeb(bp);
1381 					bp = nbp;
1382 				}
1383 				/*
1384 				 * clear stream head hi pri flag based on
1385 				 * first message
1386 				 */
1387 				if (type == M_PCPROTO) {
1388 					mutex_enter(&stp->sd_lock);
1389 					stp->sd_flag &= ~STRPRI;
1390 					mutex_exit(&stp->sd_lock);
1391 				}
1392 				if (bp) {
1393 					bp->b_band = pri;
1394 					goto ismdata;
1395 				} else {
1396 					break;
1397 				}
1398 			}
1399 			/* FALLTHRU */
1400 		case M_PASSFP:
1401 			if ((bp->b_datap->db_type == M_PASSFP) &&
1402 			    (stp->sd_read_opt & RD_PROTDIS)) {
1403 				freemsg(bp);
1404 				break;
1405 			}
1406 			mutex_enter(&stp->sd_lock);
1407 			putback(stp, q, bp, pri);
1408 			mutex_exit(&stp->sd_lock);
1409 			if (rflg == 0)
1410 				error = EBADMSG;
1411 			goto oops1;
1412 
1413 		default:
1414 			/*
1415 			 * Garbage on stream head read queue.
1416 			 */
1417 			cmn_err(CE_WARN, "bad %x found at stream head\n",
1418 				bp->b_datap->db_type);
1419 			freemsg(bp);
1420 			goto oops1;
1421 		}
1422 		mutex_enter(&stp->sd_lock);
1423 	}
1424 oops:
1425 	mutex_exit(&stp->sd_lock);
1426 oops1:
1427 	qbackenable(q, pri);
1428 	return (error);
1429 #undef	_LASTMARK
1430 }
1431 
1432 /*
1433  * Default processing of M_PROTO/M_PCPROTO messages.
1434  * Determine which wakeups and signals are needed.
1435  * This can be replaced by a user-specified procedure for kernel users
1436  * of STREAMS.
1437  */
1438 /* ARGSUSED */
1439 mblk_t *
1440 strrput_proto(vnode_t *vp, mblk_t *mp,
1441     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1442     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1443 {
1444 	*wakeups = RSLEEP;
1445 	*allmsgsigs = 0;
1446 
1447 	switch (mp->b_datap->db_type) {
1448 	case M_PROTO:
1449 		if (mp->b_band == 0) {
1450 			*firstmsgsigs = S_INPUT | S_RDNORM;
1451 			*pollwakeups = POLLIN | POLLRDNORM;
1452 		} else {
1453 			*firstmsgsigs = S_INPUT | S_RDBAND;
1454 			*pollwakeups = POLLIN | POLLRDBAND;
1455 		}
1456 		break;
1457 	case M_PCPROTO:
1458 		*firstmsgsigs = S_HIPRI;
1459 		*pollwakeups = POLLPRI;
1460 		break;
1461 	}
1462 	return (mp);
1463 }
1464 
1465 /*
1466  * Default processing of everything but M_DATA, M_PROTO, M_PCPROTO and
1467  * M_PASSFP messages.
1468  * Determine which wakeups and signals are needed.
1469  * This can be replaced by a user-specified procedure for kernel users
1470  * of STREAMS.
1471  */
1472 /* ARGSUSED */
1473 mblk_t *
1474 strrput_misc(vnode_t *vp, mblk_t *mp,
1475     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1476     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1477 {
1478 	*wakeups = 0;
1479 	*firstmsgsigs = 0;
1480 	*allmsgsigs = 0;
1481 	*pollwakeups = 0;
1482 	return (mp);
1483 }
1484 
1485 /*
1486  * Stream read put procedure.  Called from downstream driver/module
1487  * with messages for the stream head.  Data, protocol, and in-stream
1488  * signal messages are placed on the queue, others are handled directly.
1489  */
1490 int
1491 strrput(queue_t *q, mblk_t *bp)
1492 {
1493 	struct stdata	*stp;
1494 	ulong_t		rput_opt;
1495 	strwakeup_t	wakeups;
1496 	strsigset_t	firstmsgsigs;	/* Signals if first message on queue */
1497 	strsigset_t	allmsgsigs;	/* Signals for all messages */
1498 	strsigset_t	signals;	/* Signals events to generate */
1499 	strpollset_t	pollwakeups;
1500 	mblk_t		*nextbp;
1501 	uchar_t		band = 0;
1502 	int		hipri_sig;
1503 
1504 	stp = (struct stdata *)q->q_ptr;
1505 	/*
1506 	 * Use rput_opt for optimized access to the SR_ flags except
1507 	 * SR_POLLIN. That flag has to be checked under sd_lock since it
1508 	 * is modified by strpoll().
1509 	 */
1510 	rput_opt = stp->sd_rput_opt;
1511 
1512 	ASSERT(qclaimed(q));
1513 	TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_ENTER,
1514 		"strrput called with message type:q %p bp %p", q, bp);
1515 
1516 	/*
1517 	 * Perform initial processing and pass to the parameterized functions.
1518 	 */
1519 	ASSERT(bp->b_next == NULL);
1520 
1521 	switch (bp->b_datap->db_type) {
1522 	case M_DATA:
1523 		/*
1524 		 * sockfs is the only consumer of STREOF and when it is set,
1525 		 * it implies that the receiver is not interested in receiving
1526 		 * any more data, hence the mblk is freed to prevent unnecessary
1527 		 * message queueing at the stream head.
1528 		 */
1529 		if (stp->sd_flag == STREOF) {
1530 			freemsg(bp);
1531 			return (0);
1532 		}
1533 		if ((rput_opt & SR_IGN_ZEROLEN) &&
1534 		    bp->b_rptr == bp->b_wptr && msgnodata(bp)) {
1535 			/*
1536 			 * Ignore zero-length M_DATA messages. These might be
1537 			 * generated by some transports.
1538 			 * The zero-length M_DATA messages, even if they
1539 			 * are ignored, should effect the atmark tracking and
1540 			 * should wake up a thread sleeping in strwaitmark.
1541 			 */
1542 			mutex_enter(&stp->sd_lock);
1543 			if (bp->b_flag & MSGMARKNEXT) {
1544 				/*
1545 				 * Record the position of the mark either
1546 				 * in q_last or in STRATMARK.
1547 				 */
1548 				if (q->q_last != NULL) {
1549 					q->q_last->b_flag &= ~MSGNOTMARKNEXT;
1550 					q->q_last->b_flag |= MSGMARKNEXT;
1551 				} else {
1552 					stp->sd_flag &= ~STRNOTATMARK;
1553 					stp->sd_flag |= STRATMARK;
1554 				}
1555 			} else if (bp->b_flag & MSGNOTMARKNEXT) {
1556 				/*
1557 				 * Record that this is not the position of
1558 				 * the mark either in q_last or in
1559 				 * STRNOTATMARK.
1560 				 */
1561 				if (q->q_last != NULL) {
1562 					q->q_last->b_flag &= ~MSGMARKNEXT;
1563 					q->q_last->b_flag |= MSGNOTMARKNEXT;
1564 				} else {
1565 					stp->sd_flag &= ~STRATMARK;
1566 					stp->sd_flag |= STRNOTATMARK;
1567 				}
1568 			}
1569 			if (stp->sd_flag & RSLEEP) {
1570 				stp->sd_flag &= ~RSLEEP;
1571 				cv_broadcast(&q->q_wait);
1572 			}
1573 			mutex_exit(&stp->sd_lock);
1574 			freemsg(bp);
1575 			return (0);
1576 		}
1577 		wakeups = RSLEEP;
1578 		if (bp->b_band == 0) {
1579 			firstmsgsigs = S_INPUT | S_RDNORM;
1580 			pollwakeups = POLLIN | POLLRDNORM;
1581 		} else {
1582 			firstmsgsigs = S_INPUT | S_RDBAND;
1583 			pollwakeups = POLLIN | POLLRDBAND;
1584 		}
1585 		if (rput_opt & SR_SIGALLDATA)
1586 			allmsgsigs = firstmsgsigs;
1587 		else
1588 			allmsgsigs = 0;
1589 
1590 		mutex_enter(&stp->sd_lock);
1591 		if ((rput_opt & SR_CONSOL_DATA) &&
1592 		    (bp->b_flag & (MSGMARK|MSGDELIM)) == 0) {
1593 			/*
1594 			 * Consolidate on M_DATA message onto an M_DATA,
1595 			 * M_PROTO, or M_PCPROTO by merging it with q_last.
1596 			 * The consolidation does not take place if
1597 			 * the old message is marked with either of the
1598 			 * marks or the delim flag or if the new
1599 			 * message is marked with MSGMARK. The MSGMARK
1600 			 * check is needed to handle the odd semantics of
1601 			 * MSGMARK where essentially the whole message
1602 			 * is to be treated as marked.
1603 			 * Carry any MSGMARKNEXT  and MSGNOTMARKNEXT from the
1604 			 * new message to the front of the b_cont chain.
1605 			 */
1606 			mblk_t *lbp;
1607 
1608 			lbp = q->q_last;
1609 			if (lbp != NULL &&
1610 			    (lbp->b_datap->db_type == M_DATA ||
1611 			    lbp->b_datap->db_type == M_PROTO ||
1612 			    lbp->b_datap->db_type == M_PCPROTO) &&
1613 			    !(lbp->b_flag & (MSGDELIM|MSGMARK|
1614 			    MSGMARKNEXT))) {
1615 				rmvq_noenab(q, lbp);
1616 				/*
1617 				 * The first message in the b_cont list
1618 				 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
1619 				 * We need to handle the case where we
1620 				 * are appending
1621 				 *
1622 				 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
1623 				 * 2) a MSGMARKNEXT to a plain message.
1624 				 * 3) a MSGNOTMARKNEXT to a plain message
1625 				 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
1626 				 *    message.
1627 				 *
1628 				 * Thus we never append a MSGMARKNEXT or
1629 				 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
1630 				 */
1631 				if (bp->b_flag & MSGMARKNEXT) {
1632 					lbp->b_flag |= MSGMARKNEXT;
1633 					lbp->b_flag &= ~MSGNOTMARKNEXT;
1634 					bp->b_flag &= ~MSGMARKNEXT;
1635 				} else if (bp->b_flag & MSGNOTMARKNEXT) {
1636 					lbp->b_flag |= MSGNOTMARKNEXT;
1637 					bp->b_flag &= ~MSGNOTMARKNEXT;
1638 				}
1639 
1640 				linkb(lbp, bp);
1641 				bp = lbp;
1642 				/*
1643 				 * The new message logically isn't the first
1644 				 * even though the q_first check below thinks
1645 				 * it is. Clear the firstmsgsigs to make it
1646 				 * not appear to be first.
1647 				 */
1648 				firstmsgsigs = 0;
1649 			}
1650 		}
1651 		break;
1652 
1653 	case M_PASSFP:
1654 		wakeups = RSLEEP;
1655 		allmsgsigs = 0;
1656 		if (bp->b_band == 0) {
1657 			firstmsgsigs = S_INPUT | S_RDNORM;
1658 			pollwakeups = POLLIN | POLLRDNORM;
1659 		} else {
1660 			firstmsgsigs = S_INPUT | S_RDBAND;
1661 			pollwakeups = POLLIN | POLLRDBAND;
1662 		}
1663 		mutex_enter(&stp->sd_lock);
1664 		break;
1665 
1666 	case M_PROTO:
1667 	case M_PCPROTO:
1668 		ASSERT(stp->sd_rprotofunc != NULL);
1669 		bp = (stp->sd_rprotofunc)(stp->sd_vnode, bp,
1670 			&wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1671 #define	ALLSIG	(S_INPUT|S_HIPRI|S_OUTPUT|S_MSG|S_ERROR|S_HANGUP|S_RDNORM|\
1672 		S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)
1673 #define	ALLPOLL	(POLLIN|POLLPRI|POLLOUT|POLLRDNORM|POLLWRNORM|POLLRDBAND|\
1674 		POLLWRBAND)
1675 
1676 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1677 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1678 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1679 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1680 
1681 		mutex_enter(&stp->sd_lock);
1682 		break;
1683 
1684 	default:
1685 		ASSERT(stp->sd_rmiscfunc != NULL);
1686 		bp = (stp->sd_rmiscfunc)(stp->sd_vnode, bp,
1687 			&wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1688 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1689 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1690 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1691 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1692 #undef	ALLSIG
1693 #undef	ALLPOLL
1694 		mutex_enter(&stp->sd_lock);
1695 		break;
1696 	}
1697 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1698 
1699 	/* By default generate superset of signals */
1700 	signals = (firstmsgsigs | allmsgsigs);
1701 
1702 	/*
1703 	 * The  proto and misc functions can return multiple messages
1704 	 * as a b_next chain. Such messages are processed separately.
1705 	 */
1706 one_more:
1707 	hipri_sig = 0;
1708 	if (bp == NULL) {
1709 		nextbp = NULL;
1710 	} else {
1711 		nextbp = bp->b_next;
1712 		bp->b_next = NULL;
1713 
1714 		switch (bp->b_datap->db_type) {
1715 		case M_PCPROTO:
1716 			/*
1717 			 * Only one priority protocol message is allowed at the
1718 			 * stream head at a time.
1719 			 */
1720 			if (stp->sd_flag & STRPRI) {
1721 				TRACE_0(TR_FAC_STREAMS_FR, TR_STRRPUT_PROTERR,
1722 				    "M_PCPROTO already at head");
1723 				freemsg(bp);
1724 				mutex_exit(&stp->sd_lock);
1725 				goto done;
1726 			}
1727 			stp->sd_flag |= STRPRI;
1728 			hipri_sig = 1;
1729 			/* FALLTHRU */
1730 		case M_DATA:
1731 		case M_PROTO:
1732 		case M_PASSFP:
1733 			band = bp->b_band;
1734 			/*
1735 			 * Marking doesn't work well when messages
1736 			 * are marked in more than one band.  We only
1737 			 * remember the last message received, even if
1738 			 * it is placed on the queue ahead of other
1739 			 * marked messages.
1740 			 */
1741 			if (bp->b_flag & MSGMARK)
1742 				stp->sd_mark = bp;
1743 			(void) putq(q, bp);
1744 
1745 			/*
1746 			 * If message is a PCPROTO message, always use
1747 			 * firstmsgsigs to determine if a signal should be
1748 			 * sent as strrput is the only place to send
1749 			 * signals for PCPROTO. Other messages are based on
1750 			 * the STRGETINPROG flag. The flag determines if
1751 			 * strrput or (k)strgetmsg will be responsible for
1752 			 * sending the signals, in the firstmsgsigs case.
1753 			 */
1754 			if ((hipri_sig == 1) ||
1755 			    (((stp->sd_flag & STRGETINPROG) == 0) &&
1756 			    (q->q_first == bp)))
1757 				signals = (firstmsgsigs | allmsgsigs);
1758 			else
1759 				signals = allmsgsigs;
1760 			break;
1761 
1762 		default:
1763 			mutex_exit(&stp->sd_lock);
1764 			(void) strrput_nondata(q, bp);
1765 			mutex_enter(&stp->sd_lock);
1766 			break;
1767 		}
1768 	}
1769 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1770 	/*
1771 	 * Wake sleeping read/getmsg and cancel deferred wakeup
1772 	 */
1773 	if (wakeups & RSLEEP)
1774 		stp->sd_wakeq &= ~RSLEEP;
1775 
1776 	wakeups &= stp->sd_flag;
1777 	if (wakeups & RSLEEP) {
1778 		stp->sd_flag &= ~RSLEEP;
1779 		cv_broadcast(&q->q_wait);
1780 	}
1781 	if (wakeups & WSLEEP) {
1782 		stp->sd_flag &= ~WSLEEP;
1783 		cv_broadcast(&_WR(q)->q_wait);
1784 	}
1785 
1786 	if (pollwakeups != 0) {
1787 		if (pollwakeups == (POLLIN | POLLRDNORM)) {
1788 			/*
1789 			 * Can't use rput_opt since it was not
1790 			 * read when sd_lock was held and SR_POLLIN is changed
1791 			 * by strpoll() under sd_lock.
1792 			 */
1793 			if (!(stp->sd_rput_opt & SR_POLLIN))
1794 				goto no_pollwake;
1795 			stp->sd_rput_opt &= ~SR_POLLIN;
1796 		}
1797 		mutex_exit(&stp->sd_lock);
1798 		pollwakeup(&stp->sd_pollist, pollwakeups);
1799 		mutex_enter(&stp->sd_lock);
1800 	}
1801 no_pollwake:
1802 
1803 	/*
1804 	 * strsendsig can handle multiple signals with a
1805 	 * single call.
1806 	 */
1807 	if (stp->sd_sigflags & signals)
1808 		strsendsig(stp->sd_siglist, signals, band, 0);
1809 	mutex_exit(&stp->sd_lock);
1810 
1811 
1812 done:
1813 	if (nextbp == NULL)
1814 		return (0);
1815 
1816 	/*
1817 	 * Any signals were handled the first time.
1818 	 * Wakeups and pollwakeups are redone to avoid any race
1819 	 * conditions - all the messages are not queued until the
1820 	 * last message has been processed by strrput.
1821 	 */
1822 	bp = nextbp;
1823 	signals = firstmsgsigs = allmsgsigs = 0;
1824 	mutex_enter(&stp->sd_lock);
1825 	goto one_more;
1826 }
1827 
1828 static void
1829 log_dupioc(queue_t *rq, mblk_t *bp)
1830 {
1831 	queue_t *wq, *qp;
1832 	char *modnames, *mnp, *dname;
1833 	size_t maxmodstr;
1834 	boolean_t islast;
1835 
1836 	/*
1837 	 * Allocate a buffer large enough to hold the names of nstrpush modules
1838 	 * and one driver, with spaces between and NUL terminator.  If we can't
1839 	 * get memory, then we'll just log the driver name.
1840 	 */
1841 	maxmodstr = nstrpush * (FMNAMESZ + 1);
1842 	mnp = modnames = kmem_alloc(maxmodstr, KM_NOSLEEP);
1843 
1844 	/* march down write side to print log message down to the driver */
1845 	wq = WR(rq);
1846 
1847 	/* make sure q_next doesn't shift around while we're grabbing data */
1848 	claimstr(wq);
1849 	qp = wq->q_next;
1850 	do {
1851 		if ((dname = qp->q_qinfo->qi_minfo->mi_idname) == NULL)
1852 			dname = "?";
1853 		islast = !SAMESTR(qp) || qp->q_next == NULL;
1854 		if (modnames == NULL) {
1855 			/*
1856 			 * If we don't have memory, then get the driver name in
1857 			 * the log where we can see it.  Note that memory
1858 			 * pressure is a possible cause of these sorts of bugs.
1859 			 */
1860 			if (islast) {
1861 				modnames = dname;
1862 				maxmodstr = 0;
1863 			}
1864 		} else {
1865 			mnp += snprintf(mnp, FMNAMESZ + 1, "%s", dname);
1866 			if (!islast)
1867 				*mnp++ = ' ';
1868 		}
1869 		qp = qp->q_next;
1870 	} while (!islast);
1871 	releasestr(wq);
1872 	/* Cannot happen unless stream head is corrupt. */
1873 	ASSERT(modnames != NULL);
1874 	(void) strlog(rq->q_qinfo->qi_minfo->mi_idnum, 0, 1,
1875 	    SL_CONSOLE|SL_TRACE|SL_ERROR,
1876 	    "Warning: stream %p received duplicate %X M_IOC%s; module list: %s",
1877 	    rq->q_ptr, ((struct iocblk *)bp->b_rptr)->ioc_cmd,
1878 	    (DB_TYPE(bp) == M_IOCACK ? "ACK" : "NAK"), modnames);
1879 	if (maxmodstr != 0)
1880 		kmem_free(modnames, maxmodstr);
1881 }
1882 
1883 int
1884 strrput_nondata(queue_t *q, mblk_t *bp)
1885 {
1886 	struct stdata *stp;
1887 	struct iocblk *iocbp;
1888 	struct stroptions *sop;
1889 	struct copyreq *reqp;
1890 	struct copyresp *resp;
1891 	unsigned char bpri;
1892 	unsigned char  flushed_already = 0;
1893 
1894 	stp = (struct stdata *)q->q_ptr;
1895 
1896 	ASSERT(!(stp->sd_flag & STPLEX));
1897 	ASSERT(qclaimed(q));
1898 
1899 	switch (bp->b_datap->db_type) {
1900 	case M_ERROR:
1901 		/*
1902 		 * An error has occurred downstream, the errno is in the first
1903 		 * bytes of the message.
1904 		 */
1905 		if ((bp->b_wptr - bp->b_rptr) == 2) {	/* New flavor */
1906 			unsigned char rw = 0;
1907 
1908 			mutex_enter(&stp->sd_lock);
1909 			if (*bp->b_rptr != NOERROR) {	/* read error */
1910 				if (*bp->b_rptr != 0) {
1911 					if (stp->sd_flag & STRDERR)
1912 						flushed_already |= FLUSHR;
1913 					stp->sd_flag |= STRDERR;
1914 					rw |= FLUSHR;
1915 				} else {
1916 					stp->sd_flag &= ~STRDERR;
1917 				}
1918 				stp->sd_rerror = *bp->b_rptr;
1919 			}
1920 			bp->b_rptr++;
1921 			if (*bp->b_rptr != NOERROR) {	/* write error */
1922 				if (*bp->b_rptr != 0) {
1923 					if (stp->sd_flag & STWRERR)
1924 						flushed_already |= FLUSHW;
1925 					stp->sd_flag |= STWRERR;
1926 					rw |= FLUSHW;
1927 				} else {
1928 					stp->sd_flag &= ~STWRERR;
1929 				}
1930 				stp->sd_werror = *bp->b_rptr;
1931 			}
1932 			if (rw) {
1933 				TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_WAKE,
1934 					"strrput cv_broadcast:q %p, bp %p",
1935 					q, bp);
1936 				cv_broadcast(&q->q_wait); /* readers */
1937 				cv_broadcast(&_WR(q)->q_wait); /* writers */
1938 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
1939 
1940 				mutex_exit(&stp->sd_lock);
1941 				pollwakeup(&stp->sd_pollist, POLLERR);
1942 				mutex_enter(&stp->sd_lock);
1943 
1944 				if (stp->sd_sigflags & S_ERROR)
1945 					strsendsig(stp->sd_siglist, S_ERROR, 0,
1946 					    ((rw & FLUSHR) ? stp->sd_rerror :
1947 					    stp->sd_werror));
1948 				mutex_exit(&stp->sd_lock);
1949 				/*
1950 				 * Send the M_FLUSH only
1951 				 * for the first M_ERROR
1952 				 * message on the stream
1953 				 */
1954 				if (flushed_already == rw) {
1955 					freemsg(bp);
1956 					return (0);
1957 				}
1958 
1959 				bp->b_datap->db_type = M_FLUSH;
1960 				*bp->b_rptr = rw;
1961 				bp->b_wptr = bp->b_rptr + 1;
1962 				/*
1963 				 * Protect against the driver
1964 				 * passing up messages after
1965 				 * it has done a qprocsoff
1966 				 */
1967 				if (_OTHERQ(q)->q_next == NULL)
1968 					freemsg(bp);
1969 				else
1970 					qreply(q, bp);
1971 				return (0);
1972 			} else
1973 				mutex_exit(&stp->sd_lock);
1974 		} else if (*bp->b_rptr != 0) {		/* Old flavor */
1975 				if (stp->sd_flag & (STRDERR|STWRERR))
1976 					flushed_already = FLUSHRW;
1977 				mutex_enter(&stp->sd_lock);
1978 				stp->sd_flag |= (STRDERR|STWRERR);
1979 				stp->sd_rerror = *bp->b_rptr;
1980 				stp->sd_werror = *bp->b_rptr;
1981 				TRACE_2(TR_FAC_STREAMS_FR,
1982 					TR_STRRPUT_WAKE2,
1983 					"strrput wakeup #2:q %p, bp %p", q, bp);
1984 				cv_broadcast(&q->q_wait); /* the readers */
1985 				cv_broadcast(&_WR(q)->q_wait); /* the writers */
1986 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
1987 
1988 				mutex_exit(&stp->sd_lock);
1989 				pollwakeup(&stp->sd_pollist, POLLERR);
1990 				mutex_enter(&stp->sd_lock);
1991 
1992 				if (stp->sd_sigflags & S_ERROR)
1993 					strsendsig(stp->sd_siglist, S_ERROR, 0,
1994 					    (stp->sd_werror ? stp->sd_werror :
1995 					    stp->sd_rerror));
1996 				mutex_exit(&stp->sd_lock);
1997 
1998 				/*
1999 				 * Send the M_FLUSH only
2000 				 * for the first M_ERROR
2001 				 * message on the stream
2002 				 */
2003 				if (flushed_already != FLUSHRW) {
2004 					bp->b_datap->db_type = M_FLUSH;
2005 					*bp->b_rptr = FLUSHRW;
2006 					/*
2007 					 * Protect against the driver passing up
2008 					 * messages after it has done a
2009 					 * qprocsoff.
2010 					 */
2011 				if (_OTHERQ(q)->q_next == NULL)
2012 					freemsg(bp);
2013 				else
2014 					qreply(q, bp);
2015 				return (0);
2016 				}
2017 		}
2018 		freemsg(bp);
2019 		return (0);
2020 
2021 	case M_HANGUP:
2022 
2023 		freemsg(bp);
2024 		mutex_enter(&stp->sd_lock);
2025 		stp->sd_werror = ENXIO;
2026 		stp->sd_flag |= STRHUP;
2027 		stp->sd_flag &= ~(WSLEEP|RSLEEP);
2028 
2029 		/*
2030 		 * send signal if controlling tty
2031 		 */
2032 
2033 		if (stp->sd_sidp) {
2034 			prsignal(stp->sd_sidp, SIGHUP);
2035 			if (stp->sd_sidp != stp->sd_pgidp)
2036 				pgsignal(stp->sd_pgidp, SIGTSTP);
2037 		}
2038 
2039 		/*
2040 		 * wake up read, write, and exception pollers and
2041 		 * reset wakeup mechanism.
2042 		 */
2043 		cv_broadcast(&q->q_wait);	/* the readers */
2044 		cv_broadcast(&_WR(q)->q_wait);	/* the writers */
2045 		cv_broadcast(&stp->sd_monitor);	/* the ioctllers */
2046 		strhup(stp);
2047 		mutex_exit(&stp->sd_lock);
2048 		return (0);
2049 
2050 	case M_UNHANGUP:
2051 		freemsg(bp);
2052 		mutex_enter(&stp->sd_lock);
2053 		stp->sd_werror = 0;
2054 		stp->sd_flag &= ~STRHUP;
2055 		mutex_exit(&stp->sd_lock);
2056 		return (0);
2057 
2058 	case M_SIG:
2059 		/*
2060 		 * Someone downstream wants to post a signal.  The
2061 		 * signal to post is contained in the first byte of the
2062 		 * message.  If the message would go on the front of
2063 		 * the queue, send a signal to the process group
2064 		 * (if not SIGPOLL) or to the siglist processes
2065 		 * (SIGPOLL).  If something is already on the queue,
2066 		 * OR if we are delivering a delayed suspend (*sigh*
2067 		 * another "tty" hack) and there's no one sleeping already,
2068 		 * just enqueue the message.
2069 		 */
2070 		mutex_enter(&stp->sd_lock);
2071 		if (q->q_first || (*bp->b_rptr == SIGTSTP &&
2072 		    !(stp->sd_flag & RSLEEP))) {
2073 			(void) putq(q, bp);
2074 			mutex_exit(&stp->sd_lock);
2075 			return (0);
2076 		}
2077 		mutex_exit(&stp->sd_lock);
2078 		/* FALLTHRU */
2079 
2080 	case M_PCSIG:
2081 		/*
2082 		 * Don't enqueue, just post the signal.
2083 		 */
2084 		strsignal(stp, *bp->b_rptr, 0L);
2085 		freemsg(bp);
2086 		return (0);
2087 
2088 	case M_FLUSH:
2089 		/*
2090 		 * Flush queues.  The indication of which queues to flush
2091 		 * is in the first byte of the message.  If the read queue
2092 		 * is specified, then flush it.  If FLUSHBAND is set, just
2093 		 * flush the band specified by the second byte of the message.
2094 		 *
2095 		 * If a module has issued a M_SETOPT to not flush hi
2096 		 * priority messages off of the stream head, then pass this
2097 		 * flag into the flushq code to preserve such messages.
2098 		 */
2099 
2100 		if (*bp->b_rptr & FLUSHR) {
2101 			mutex_enter(&stp->sd_lock);
2102 			if (*bp->b_rptr & FLUSHBAND) {
2103 				ASSERT((bp->b_wptr - bp->b_rptr) >= 2);
2104 				flushband(q, *(bp->b_rptr + 1), FLUSHALL);
2105 			} else
2106 				flushq_common(q, FLUSHALL,
2107 				    stp->sd_read_opt & RFLUSHPCPROT);
2108 			if ((q->q_first == NULL) ||
2109 			    (q->q_first->b_datap->db_type < QPCTL))
2110 				stp->sd_flag &= ~STRPRI;
2111 			else {
2112 				ASSERT(stp->sd_flag & STRPRI);
2113 			}
2114 			mutex_exit(&stp->sd_lock);
2115 		}
2116 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
2117 			*bp->b_rptr &= ~FLUSHR;
2118 			bp->b_flag |= MSGNOLOOP;
2119 			/*
2120 			 * Protect against the driver passing up
2121 			 * messages after it has done a qprocsoff.
2122 			 */
2123 			if (_OTHERQ(q)->q_next == NULL)
2124 				freemsg(bp);
2125 			else
2126 				qreply(q, bp);
2127 			return (0);
2128 		}
2129 		freemsg(bp);
2130 		return (0);
2131 
2132 	case M_IOCACK:
2133 	case M_IOCNAK:
2134 		iocbp = (struct iocblk *)bp->b_rptr;
2135 		/*
2136 		 * If not waiting for ACK or NAK then just free msg.
2137 		 * If incorrect id sequence number then just free msg.
2138 		 * If already have ACK or NAK for user then this is a
2139 		 *    duplicate, display a warning and free the msg.
2140 		 */
2141 		mutex_enter(&stp->sd_lock);
2142 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2143 		    (stp->sd_iocid != iocbp->ioc_id)) {
2144 			/*
2145 			 * If the ACK/NAK is a dup, display a message
2146 			 * Dup is when sd_iocid == ioc_id, and
2147 			 * sd_iocblk == <valid ptr> or -1 (the former
2148 			 * is when an ioctl has been put on the stream
2149 			 * head, but has not yet been consumed, the
2150 			 * later is when it has been consumed).
2151 			 */
2152 			if ((stp->sd_iocid == iocbp->ioc_id) &&
2153 			    (stp->sd_iocblk != NULL)) {
2154 				log_dupioc(q, bp);
2155 			}
2156 			freemsg(bp);
2157 			mutex_exit(&stp->sd_lock);
2158 			return (0);
2159 		}
2160 
2161 		/*
2162 		 * Assign ACK or NAK to user and wake up.
2163 		 */
2164 		stp->sd_iocblk = bp;
2165 		cv_broadcast(&stp->sd_monitor);
2166 		mutex_exit(&stp->sd_lock);
2167 		return (0);
2168 
2169 	case M_COPYIN:
2170 	case M_COPYOUT:
2171 		reqp = (struct copyreq *)bp->b_rptr;
2172 
2173 		/*
2174 		 * If not waiting for ACK or NAK then just fail request.
2175 		 * If already have ACK, NAK, or copy request, then just
2176 		 * fail request.
2177 		 * If incorrect id sequence number then just fail request.
2178 		 */
2179 		mutex_enter(&stp->sd_lock);
2180 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2181 		    (stp->sd_iocid != reqp->cq_id)) {
2182 			if (bp->b_cont) {
2183 				freemsg(bp->b_cont);
2184 				bp->b_cont = NULL;
2185 			}
2186 			bp->b_datap->db_type = M_IOCDATA;
2187 			bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
2188 			resp = (struct copyresp *)bp->b_rptr;
2189 			resp->cp_rval = (caddr_t)1;	/* failure */
2190 			mutex_exit(&stp->sd_lock);
2191 			putnext(stp->sd_wrq, bp);
2192 			return (0);
2193 		}
2194 
2195 		/*
2196 		 * Assign copy request to user and wake up.
2197 		 */
2198 		stp->sd_iocblk = bp;
2199 		cv_broadcast(&stp->sd_monitor);
2200 		mutex_exit(&stp->sd_lock);
2201 		return (0);
2202 
2203 	case M_SETOPTS:
2204 		/*
2205 		 * Set stream head options (read option, write offset,
2206 		 * min/max packet size, and/or high/low water marks for
2207 		 * the read side only).
2208 		 */
2209 
2210 		bpri = 0;
2211 		sop = (struct stroptions *)bp->b_rptr;
2212 		mutex_enter(&stp->sd_lock);
2213 		if (sop->so_flags & SO_READOPT) {
2214 			switch (sop->so_readopt & RMODEMASK) {
2215 			case RNORM:
2216 				stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
2217 				break;
2218 
2219 			case RMSGD:
2220 				stp->sd_read_opt =
2221 				    ((stp->sd_read_opt & ~RD_MSGNODIS) |
2222 				    RD_MSGDIS);
2223 				break;
2224 
2225 			case RMSGN:
2226 				stp->sd_read_opt =
2227 				    ((stp->sd_read_opt & ~RD_MSGDIS) |
2228 				    RD_MSGNODIS);
2229 				break;
2230 			}
2231 			switch (sop->so_readopt & RPROTMASK) {
2232 			case RPROTNORM:
2233 				stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
2234 				break;
2235 
2236 			case RPROTDAT:
2237 				stp->sd_read_opt =
2238 				    ((stp->sd_read_opt & ~RD_PROTDIS) |
2239 				    RD_PROTDAT);
2240 				break;
2241 
2242 			case RPROTDIS:
2243 				stp->sd_read_opt =
2244 				    ((stp->sd_read_opt & ~RD_PROTDAT) |
2245 				    RD_PROTDIS);
2246 				break;
2247 			}
2248 			switch (sop->so_readopt & RFLUSHMASK) {
2249 			case RFLUSHPCPROT:
2250 				/*
2251 				 * This sets the stream head to NOT flush
2252 				 * M_PCPROTO messages.
2253 				 */
2254 				stp->sd_read_opt |= RFLUSHPCPROT;
2255 				break;
2256 			}
2257 		}
2258 		if (sop->so_flags & SO_ERROPT) {
2259 			switch (sop->so_erropt & RERRMASK) {
2260 			case RERRNORM:
2261 				stp->sd_flag &= ~STRDERRNONPERSIST;
2262 				break;
2263 			case RERRNONPERSIST:
2264 				stp->sd_flag |= STRDERRNONPERSIST;
2265 				break;
2266 			}
2267 			switch (sop->so_erropt & WERRMASK) {
2268 			case WERRNORM:
2269 				stp->sd_flag &= ~STWRERRNONPERSIST;
2270 				break;
2271 			case WERRNONPERSIST:
2272 				stp->sd_flag |= STWRERRNONPERSIST;
2273 				break;
2274 			}
2275 		}
2276 		if (sop->so_flags & SO_COPYOPT) {
2277 			if (sop->so_copyopt & ZCVMSAFE) {
2278 				stp->sd_copyflag |= STZCVMSAFE;
2279 				stp->sd_copyflag &= ~STZCVMUNSAFE;
2280 			} else if (sop->so_copyopt & ZCVMUNSAFE) {
2281 				stp->sd_copyflag |= STZCVMUNSAFE;
2282 				stp->sd_copyflag &= ~STZCVMSAFE;
2283 			}
2284 
2285 			if (sop->so_copyopt & COPYCACHED) {
2286 				stp->sd_copyflag |= STRCOPYCACHED;
2287 			}
2288 		}
2289 		if (sop->so_flags & SO_WROFF)
2290 			stp->sd_wroff = sop->so_wroff;
2291 		if (sop->so_flags & SO_TAIL)
2292 			stp->sd_tail = sop->so_tail;
2293 		if (sop->so_flags & SO_MINPSZ)
2294 			q->q_minpsz = sop->so_minpsz;
2295 		if (sop->so_flags & SO_MAXPSZ)
2296 			q->q_maxpsz = sop->so_maxpsz;
2297 		if (sop->so_flags & SO_MAXBLK)
2298 			stp->sd_maxblk = sop->so_maxblk;
2299 		if (sop->so_flags & SO_HIWAT) {
2300 		    if (sop->so_flags & SO_BAND) {
2301 			if (strqset(q, QHIWAT, sop->so_band, sop->so_hiwat))
2302 				cmn_err(CE_WARN,
2303 				    "strrput: could not allocate qband\n");
2304 			else
2305 				bpri = sop->so_band;
2306 		    } else {
2307 			q->q_hiwat = sop->so_hiwat;
2308 		    }
2309 		}
2310 		if (sop->so_flags & SO_LOWAT) {
2311 		    if (sop->so_flags & SO_BAND) {
2312 			if (strqset(q, QLOWAT, sop->so_band, sop->so_lowat))
2313 				cmn_err(CE_WARN,
2314 				    "strrput: could not allocate qband\n");
2315 			else
2316 				bpri = sop->so_band;
2317 		    } else {
2318 			q->q_lowat = sop->so_lowat;
2319 		    }
2320 		}
2321 		if (sop->so_flags & SO_MREADON)
2322 			stp->sd_flag |= SNDMREAD;
2323 		if (sop->so_flags & SO_MREADOFF)
2324 			stp->sd_flag &= ~SNDMREAD;
2325 		if (sop->so_flags & SO_NDELON)
2326 			stp->sd_flag |= OLDNDELAY;
2327 		if (sop->so_flags & SO_NDELOFF)
2328 			stp->sd_flag &= ~OLDNDELAY;
2329 		if (sop->so_flags & SO_ISTTY)
2330 			stp->sd_flag |= STRISTTY;
2331 		if (sop->so_flags & SO_ISNTTY)
2332 			stp->sd_flag &= ~STRISTTY;
2333 		if (sop->so_flags & SO_TOSTOP)
2334 			stp->sd_flag |= STRTOSTOP;
2335 		if (sop->so_flags & SO_TONSTOP)
2336 			stp->sd_flag &= ~STRTOSTOP;
2337 		if (sop->so_flags & SO_DELIM)
2338 			stp->sd_flag |= STRDELIM;
2339 		if (sop->so_flags & SO_NODELIM)
2340 			stp->sd_flag &= ~STRDELIM;
2341 
2342 		mutex_exit(&stp->sd_lock);
2343 		freemsg(bp);
2344 
2345 		/* Check backenable in case the water marks changed */
2346 		qbackenable(q, bpri);
2347 		return (0);
2348 
2349 	/*
2350 	 * The following set of cases deal with situations where two stream
2351 	 * heads are connected to each other (twisted streams).  These messages
2352 	 * have no meaning at the stream head.
2353 	 */
2354 	case M_BREAK:
2355 	case M_CTL:
2356 	case M_DELAY:
2357 	case M_START:
2358 	case M_STOP:
2359 	case M_IOCDATA:
2360 	case M_STARTI:
2361 	case M_STOPI:
2362 		freemsg(bp);
2363 		return (0);
2364 
2365 	case M_IOCTL:
2366 		/*
2367 		 * Always NAK this condition
2368 		 * (makes no sense)
2369 		 * If there is one or more threads in the read side
2370 		 * rwnext we have to defer the nacking until that thread
2371 		 * returns (in strget).
2372 		 */
2373 		mutex_enter(&stp->sd_lock);
2374 		if (stp->sd_struiodnak != 0) {
2375 			/*
2376 			 * Defer NAK to the streamhead. Queue at the end
2377 			 * the list.
2378 			 */
2379 			mblk_t *mp = stp->sd_struionak;
2380 
2381 			while (mp && mp->b_next)
2382 				mp = mp->b_next;
2383 			if (mp)
2384 				mp->b_next = bp;
2385 			else
2386 				stp->sd_struionak = bp;
2387 			bp->b_next = NULL;
2388 			mutex_exit(&stp->sd_lock);
2389 			return (0);
2390 		}
2391 		mutex_exit(&stp->sd_lock);
2392 
2393 		bp->b_datap->db_type = M_IOCNAK;
2394 		/*
2395 		 * Protect against the driver passing up
2396 		 * messages after it has done a qprocsoff.
2397 		 */
2398 		if (_OTHERQ(q)->q_next == NULL)
2399 			freemsg(bp);
2400 		else
2401 			qreply(q, bp);
2402 		return (0);
2403 
2404 	default:
2405 #ifdef DEBUG
2406 		cmn_err(CE_WARN,
2407 			"bad message type %x received at stream head\n",
2408 			bp->b_datap->db_type);
2409 #endif
2410 		freemsg(bp);
2411 		return (0);
2412 	}
2413 
2414 	/* NOTREACHED */
2415 }
2416 
2417 /*
2418  * Check if the stream pointed to by `stp' can be written to, and return an
2419  * error code if not.  If `eiohup' is set, then return EIO if STRHUP is set.
2420  * If `sigpipeok' is set and the SW_SIGPIPE option is enabled on the stream,
2421  * then always return EPIPE and send a SIGPIPE to the invoking thread.
2422  */
2423 static int
2424 strwriteable(struct stdata *stp, boolean_t eiohup, boolean_t sigpipeok)
2425 {
2426 	int error;
2427 
2428 	ASSERT(MUTEX_HELD(&stp->sd_lock));
2429 
2430 	/*
2431 	 * For modem support, POSIX states that on writes, EIO should
2432 	 * be returned if the stream has been hung up.
2433 	 */
2434 	if (eiohup && (stp->sd_flag & (STPLEX|STRHUP)) == STRHUP)
2435 		error = EIO;
2436 	else
2437 		error = strgeterr(stp, STRHUP|STPLEX|STWRERR, 0);
2438 
2439 	if (error != 0) {
2440 		if (!(stp->sd_flag & STPLEX) &&
2441 		    (stp->sd_wput_opt & SW_SIGPIPE) && sigpipeok) {
2442 			tsignal(curthread, SIGPIPE);
2443 			error = EPIPE;
2444 		}
2445 	}
2446 
2447 	return (error);
2448 }
2449 
2450 /*
2451  * Copyin and send data down a stream.
2452  * The caller will allocate and copyin any control part that precedes the
2453  * message and pass than in as mctl.
2454  *
2455  * Caller should *not* hold sd_lock.
2456  * When EWOULDBLOCK is returned the caller has to redo the canputnext
2457  * under sd_lock in order to avoid missing a backenabling wakeup.
2458  *
2459  * Use iosize = -1 to not send any M_DATA. iosize = 0 sends zero-length M_DATA.
2460  *
2461  * Set MSG_IGNFLOW in flags to ignore flow control for hipri messages.
2462  * For sync streams we can only ignore flow control by reverting to using
2463  * putnext.
2464  *
2465  * If sd_maxblk is less than *iosize this routine might return without
2466  * transferring all of *iosize. In all cases, on return *iosize will contain
2467  * the amount of data that was transferred.
2468  */
2469 static int
2470 strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
2471     int b_flag, int pri, int flags)
2472 {
2473 	struiod_t uiod;
2474 	mblk_t *mp;
2475 	queue_t *wqp = stp->sd_wrq;
2476 	int error = 0;
2477 	ssize_t count = *iosize;
2478 	cred_t *cr;
2479 
2480 	ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
2481 
2482 	if (uiop != NULL && count >= 0)
2483 		flags |= stp->sd_struiowrq ? STRUIO_POSTPONE : 0;
2484 
2485 	if (!(flags & STRUIO_POSTPONE)) {
2486 		/*
2487 		 * Use regular canputnext, strmakedata, putnext sequence.
2488 		 */
2489 		if (pri == 0) {
2490 			if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2491 				freemsg(mctl);
2492 				return (EWOULDBLOCK);
2493 			}
2494 		} else {
2495 			if (!(flags & MSG_IGNFLOW) && !bcanputnext(wqp, pri)) {
2496 				freemsg(mctl);
2497 				return (EWOULDBLOCK);
2498 			}
2499 		}
2500 
2501 		if ((error = strmakedata(iosize, uiop, stp, flags,
2502 					&mp)) != 0) {
2503 			freemsg(mctl);
2504 			/*
2505 			 * need to change return code to ENOMEM
2506 			 * so that this is not confused with
2507 			 * flow control, EAGAIN.
2508 			 */
2509 
2510 			if (error == EAGAIN)
2511 				return (ENOMEM);
2512 			else
2513 				return (error);
2514 		}
2515 		if (mctl != NULL) {
2516 			if (mctl->b_cont == NULL)
2517 				mctl->b_cont = mp;
2518 			else if (mp != NULL)
2519 				linkb(mctl, mp);
2520 			mp = mctl;
2521 			/*
2522 			 * Note that for interrupt thread, the CRED() is
2523 			 * NULL. Don't bother with the pid either.
2524 			 */
2525 			if ((cr = CRED()) != NULL) {
2526 				mblk_setcred(mp, cr);
2527 				DB_CPID(mp) = curproc->p_pid;
2528 			}
2529 		} else if (mp == NULL)
2530 			return (0);
2531 
2532 		mp->b_flag |= b_flag;
2533 		mp->b_band = (uchar_t)pri;
2534 
2535 		if (flags & MSG_IGNFLOW) {
2536 			/*
2537 			 * XXX Hack: Don't get stuck running service
2538 			 * procedures. This is needed for sockfs when
2539 			 * sending the unbind message out of the rput
2540 			 * procedure - we don't want a put procedure
2541 			 * to run service procedures.
2542 			 */
2543 			putnext(wqp, mp);
2544 		} else {
2545 			stream_willservice(stp);
2546 			putnext(wqp, mp);
2547 			stream_runservice(stp);
2548 		}
2549 		return (0);
2550 	}
2551 	/*
2552 	 * Stream supports rwnext() for the write side.
2553 	 */
2554 	if ((error = strmakedata(iosize, uiop, stp, flags, &mp)) != 0) {
2555 		freemsg(mctl);
2556 		/*
2557 		 * map EAGAIN to ENOMEM since EAGAIN means "flow controlled".
2558 		 */
2559 		return (error == EAGAIN ? ENOMEM : error);
2560 	}
2561 	if (mctl != NULL) {
2562 		if (mctl->b_cont == NULL)
2563 			mctl->b_cont = mp;
2564 		else if (mp != NULL)
2565 			linkb(mctl, mp);
2566 		mp = mctl;
2567 		/*
2568 		 * Note that for interrupt thread, the CRED() is
2569 		 * NULL.  Don't bother with the pid either.
2570 		 */
2571 		if ((cr = CRED()) != NULL) {
2572 			mblk_setcred(mp, cr);
2573 			DB_CPID(mp) = curproc->p_pid;
2574 		}
2575 	} else if (mp == NULL) {
2576 		return (0);
2577 	}
2578 
2579 	mp->b_flag |= b_flag;
2580 	mp->b_band = (uchar_t)pri;
2581 
2582 	(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
2583 		sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
2584 	uiod.d_uio.uio_offset = 0;
2585 	uiod.d_mp = mp;
2586 	error = rwnext(wqp, &uiod);
2587 	if (! uiod.d_mp) {
2588 		uioskip(uiop, *iosize);
2589 		return (error);
2590 	}
2591 	ASSERT(mp == uiod.d_mp);
2592 	if (error == EINVAL) {
2593 		/*
2594 		 * The stream plumbing must have changed while
2595 		 * we were away, so just turn off rwnext()s.
2596 		 */
2597 		error = 0;
2598 	} else if (error == EBUSY || error == EWOULDBLOCK) {
2599 		/*
2600 		 * Couldn't enter a perimeter or took a page fault,
2601 		 * so fall-back to putnext().
2602 		 */
2603 		error = 0;
2604 	} else {
2605 		freemsg(mp);
2606 		return (error);
2607 	}
2608 	/* Have to check canput before consuming data from the uio */
2609 	if (pri == 0) {
2610 		if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2611 			freemsg(mp);
2612 			return (EWOULDBLOCK);
2613 		}
2614 	} else {
2615 		if (!bcanputnext(wqp, pri) && !(flags & MSG_IGNFLOW)) {
2616 			freemsg(mp);
2617 			return (EWOULDBLOCK);
2618 		}
2619 	}
2620 	ASSERT(mp == uiod.d_mp);
2621 	/* Copyin data from the uio */
2622 	if ((error = struioget(wqp, mp, &uiod, 0)) != 0) {
2623 		freemsg(mp);
2624 		return (error);
2625 	}
2626 	uioskip(uiop, *iosize);
2627 	if (flags & MSG_IGNFLOW) {
2628 		/*
2629 		 * XXX Hack: Don't get stuck running service procedures.
2630 		 * This is needed for sockfs when sending the unbind message
2631 		 * out of the rput procedure - we don't want a put procedure
2632 		 * to run service procedures.
2633 		 */
2634 		putnext(wqp, mp);
2635 	} else {
2636 		stream_willservice(stp);
2637 		putnext(wqp, mp);
2638 		stream_runservice(stp);
2639 	}
2640 	return (0);
2641 }
2642 
2643 /*
2644  * Write attempts to break the write request into messages conforming
2645  * with the minimum and maximum packet sizes set downstream.
2646  *
2647  * Write will not block if downstream queue is full and
2648  * O_NDELAY is set, otherwise it will block waiting for the queue to get room.
2649  *
2650  * A write of zero bytes gets packaged into a zero length message and sent
2651  * downstream like any other message.
2652  *
2653  * If buffers of the requested sizes are not available, the write will
2654  * sleep until the buffers become available.
2655  *
2656  * Write (if specified) will supply a write offset in a message if it
2657  * makes sense. This can be specified by downstream modules as part of
2658  * a M_SETOPTS message.  Write will not supply the write offset if it
2659  * cannot supply any data in a buffer.  In other words, write will never
2660  * send down an empty packet due to a write offset.
2661  */
2662 /* ARGSUSED2 */
2663 int
2664 strwrite(struct vnode *vp, struct uio *uiop, cred_t *crp)
2665 {
2666 	return (strwrite_common(vp, uiop, crp, 0));
2667 }
2668 
2669 /* ARGSUSED2 */
2670 int
2671 strwrite_common(struct vnode *vp, struct uio *uiop, cred_t *crp, int wflag)
2672 {
2673 	struct stdata *stp;
2674 	struct queue *wqp;
2675 	ssize_t rmin, rmax;
2676 	ssize_t iosize;
2677 	int waitflag;
2678 	int tempmode;
2679 	int error = 0;
2680 	int b_flag;
2681 
2682 	ASSERT(vp->v_stream);
2683 	stp = vp->v_stream;
2684 
2685 	mutex_enter(&stp->sd_lock);
2686 
2687 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
2688 		mutex_exit(&stp->sd_lock);
2689 		return (error);
2690 	}
2691 
2692 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
2693 		error = strwriteable(stp, B_TRUE, B_TRUE);
2694 		if (error != 0) {
2695 			mutex_exit(&stp->sd_lock);
2696 			return (error);
2697 		}
2698 	}
2699 
2700 	mutex_exit(&stp->sd_lock);
2701 
2702 	wqp = stp->sd_wrq;
2703 
2704 	/* get these values from them cached in the stream head */
2705 	rmin = stp->sd_qn_minpsz;
2706 	rmax = stp->sd_qn_maxpsz;
2707 
2708 	/*
2709 	 * Check the min/max packet size constraints.  If min packet size
2710 	 * is non-zero, the write cannot be split into multiple messages
2711 	 * and still guarantee the size constraints.
2712 	 */
2713 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_IN, "strwrite in:q %p", wqp);
2714 
2715 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
2716 	if (rmax == 0) {
2717 		return (0);
2718 	}
2719 	if (rmin > 0) {
2720 		if (uiop->uio_resid < rmin) {
2721 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2722 				"strwrite out:q %p out %d error %d",
2723 				wqp, 0, ERANGE);
2724 			return (ERANGE);
2725 		}
2726 		if ((rmax != INFPSZ) && (uiop->uio_resid > rmax)) {
2727 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2728 				"strwrite out:q %p out %d error %d",
2729 				wqp, 1, ERANGE);
2730 			return (ERANGE);
2731 		}
2732 	}
2733 
2734 	/*
2735 	 * Do until count satisfied or error.
2736 	 */
2737 	waitflag = WRITEWAIT | wflag;
2738 	if (stp->sd_flag & OLDNDELAY)
2739 		tempmode = uiop->uio_fmode & ~FNDELAY;
2740 	else
2741 		tempmode = uiop->uio_fmode;
2742 
2743 	if (rmax == INFPSZ)
2744 		rmax = uiop->uio_resid;
2745 
2746 	/*
2747 	 * Note that tempmode does not get used in strput/strmakedata
2748 	 * but only in strwaitq. The other routines use uio_fmode
2749 	 * unmodified.
2750 	 */
2751 
2752 	/* LINTED: constant in conditional context */
2753 	while (1) {	/* breaks when uio_resid reaches zero */
2754 		/*
2755 		 * Determine the size of the next message to be
2756 		 * packaged.  May have to break write into several
2757 		 * messages based on max packet size.
2758 		 */
2759 		iosize = MIN(uiop->uio_resid, rmax);
2760 
2761 		/*
2762 		 * Put block downstream when flow control allows it.
2763 		 */
2764 		if ((stp->sd_flag & STRDELIM) && (uiop->uio_resid == iosize))
2765 			b_flag = MSGDELIM;
2766 		else
2767 			b_flag = 0;
2768 
2769 		for (;;) {
2770 			int done = 0;
2771 
2772 			error = strput(stp, NULL, uiop, &iosize, b_flag,
2773 				0, 0);
2774 			if (error == 0)
2775 				break;
2776 			if (error != EWOULDBLOCK)
2777 				goto out;
2778 
2779 			mutex_enter(&stp->sd_lock);
2780 			/*
2781 			 * Check for a missed wakeup.
2782 			 * Needed since strput did not hold sd_lock across
2783 			 * the canputnext.
2784 			 */
2785 			if (canputnext(wqp)) {
2786 				/* Try again */
2787 				mutex_exit(&stp->sd_lock);
2788 				continue;
2789 			}
2790 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAIT,
2791 				"strwrite wait:q %p wait", wqp);
2792 			if ((error = strwaitq(stp, waitflag, (ssize_t)0,
2793 			    tempmode, -1, &done)) != 0 || done) {
2794 				mutex_exit(&stp->sd_lock);
2795 				if ((vp->v_type == VFIFO) &&
2796 				    (uiop->uio_fmode & FNDELAY) &&
2797 				    (error == EAGAIN))
2798 					error = 0;
2799 				goto out;
2800 			}
2801 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAKE,
2802 				"strwrite wake:q %p awakes", wqp);
2803 			if ((error = i_straccess(stp, JCWRITE)) != 0) {
2804 				mutex_exit(&stp->sd_lock);
2805 				goto out;
2806 			}
2807 			mutex_exit(&stp->sd_lock);
2808 		}
2809 		waitflag |= NOINTR;
2810 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRWRITE_RESID,
2811 			"strwrite resid:q %p uiop %p", wqp, uiop);
2812 		if (uiop->uio_resid) {
2813 			/* Recheck for errors - needed for sockets */
2814 			if ((stp->sd_wput_opt & SW_RECHECK_ERR) &&
2815 			    (stp->sd_flag & (STWRERR|STRHUP|STPLEX))) {
2816 				mutex_enter(&stp->sd_lock);
2817 				error = strwriteable(stp, B_FALSE, B_TRUE);
2818 				mutex_exit(&stp->sd_lock);
2819 				if (error != 0)
2820 					return (error);
2821 			}
2822 			continue;
2823 		}
2824 		break;
2825 	}
2826 out:
2827 	/*
2828 	 * For historical reasons, applications expect EAGAIN when a data
2829 	 * mblk_t cannot be allocated, so change ENOMEM back to EAGAIN.
2830 	 */
2831 	if (error == ENOMEM)
2832 		error = EAGAIN;
2833 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2834 		"strwrite out:q %p out %d error %d", wqp, 2, error);
2835 	return (error);
2836 }
2837 
2838 /*
2839  * Stream head write service routine.
2840  * Its job is to wake up any sleeping writers when a queue
2841  * downstream needs data (part of the flow control in putq and getq).
2842  * It also must wake anyone sleeping on a poll().
2843  * For stream head right below mux module, it must also invoke put procedure
2844  * of next downstream module.
2845  */
2846 int
2847 strwsrv(queue_t *q)
2848 {
2849 	struct stdata *stp;
2850 	queue_t *tq;
2851 	qband_t *qbp;
2852 	int i;
2853 	qband_t *myqbp;
2854 	int isevent;
2855 	unsigned char	qbf[NBAND];	/* band flushing backenable flags */
2856 
2857 	TRACE_1(TR_FAC_STREAMS_FR,
2858 		TR_STRWSRV, "strwsrv:q %p", q);
2859 	stp = (struct stdata *)q->q_ptr;
2860 	ASSERT(qclaimed(q));
2861 	mutex_enter(&stp->sd_lock);
2862 	ASSERT(!(stp->sd_flag & STPLEX));
2863 
2864 	if (stp->sd_flag & WSLEEP) {
2865 		stp->sd_flag &= ~WSLEEP;
2866 		cv_broadcast(&q->q_wait);
2867 	}
2868 	mutex_exit(&stp->sd_lock);
2869 
2870 	/* The other end of a stream pipe went away. */
2871 	if ((tq = q->q_next) == NULL) {
2872 		return (0);
2873 	}
2874 
2875 	/* Find the next module forward that has a service procedure */
2876 	claimstr(q);
2877 	tq = q->q_nfsrv;
2878 	ASSERT(tq != NULL);
2879 
2880 	if ((q->q_flag & QBACK)) {
2881 		if ((tq->q_flag & QFULL)) {
2882 			mutex_enter(QLOCK(tq));
2883 			if (!(tq->q_flag & QFULL)) {
2884 				mutex_exit(QLOCK(tq));
2885 				goto wakeup;
2886 			}
2887 			/*
2888 			 * The queue must have become full again. Set QWANTW
2889 			 * again so strwsrv will be back enabled when
2890 			 * the queue becomes non-full next time.
2891 			 */
2892 			tq->q_flag |= QWANTW;
2893 			mutex_exit(QLOCK(tq));
2894 		} else {
2895 		wakeup:
2896 			pollwakeup(&stp->sd_pollist, POLLWRNORM);
2897 			mutex_enter(&stp->sd_lock);
2898 			if (stp->sd_sigflags & S_WRNORM)
2899 				strsendsig(stp->sd_siglist, S_WRNORM, 0, 0);
2900 			mutex_exit(&stp->sd_lock);
2901 		}
2902 	}
2903 
2904 	isevent = 0;
2905 	i = 1;
2906 	bzero((caddr_t)qbf, NBAND);
2907 	mutex_enter(QLOCK(tq));
2908 	if ((myqbp = q->q_bandp) != NULL)
2909 		for (qbp = tq->q_bandp; qbp && myqbp; qbp = qbp->qb_next) {
2910 			ASSERT(myqbp);
2911 			if ((myqbp->qb_flag & QB_BACK)) {
2912 				if (qbp->qb_flag & QB_FULL) {
2913 					/*
2914 					 * The band must have become full again.
2915 					 * Set QB_WANTW again so strwsrv will
2916 					 * be back enabled when the band becomes
2917 					 * non-full next time.
2918 					 */
2919 					qbp->qb_flag |= QB_WANTW;
2920 				} else {
2921 					isevent = 1;
2922 					qbf[i] = 1;
2923 				}
2924 			}
2925 			myqbp = myqbp->qb_next;
2926 			i++;
2927 		}
2928 	mutex_exit(QLOCK(tq));
2929 
2930 	if (isevent) {
2931 	    for (i = tq->q_nband; i; i--) {
2932 		if (qbf[i]) {
2933 			pollwakeup(&stp->sd_pollist, POLLWRBAND);
2934 			mutex_enter(&stp->sd_lock);
2935 			if (stp->sd_sigflags & S_WRBAND)
2936 				strsendsig(stp->sd_siglist, S_WRBAND,
2937 					(uchar_t)i, 0);
2938 			mutex_exit(&stp->sd_lock);
2939 		}
2940 	    }
2941 	}
2942 
2943 	releasestr(q);
2944 	return (0);
2945 }
2946 
2947 /*
2948  * Special case of strcopyin/strcopyout for copying
2949  * struct strioctl that can deal with both data
2950  * models.
2951  */
2952 
2953 #ifdef	_LP64
2954 
2955 static int
2956 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
2957 {
2958 	struct	strioctl32 strioc32;
2959 	struct	strioctl *striocp;
2960 
2961 	if (copyflag & U_TO_K) {
2962 		ASSERT((copyflag & K_TO_K) == 0);
2963 
2964 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
2965 			if (copyin(from, &strioc32, sizeof (strioc32)))
2966 				return (EFAULT);
2967 
2968 			striocp = (struct strioctl *)to;
2969 			striocp->ic_cmd	= strioc32.ic_cmd;
2970 			striocp->ic_timout = strioc32.ic_timout;
2971 			striocp->ic_len	= strioc32.ic_len;
2972 			striocp->ic_dp	= (char *)(uintptr_t)strioc32.ic_dp;
2973 
2974 		} else { /* NATIVE data model */
2975 			if (copyin(from, to, sizeof (struct strioctl))) {
2976 				return (EFAULT);
2977 			} else {
2978 				return (0);
2979 			}
2980 		}
2981 	} else {
2982 		ASSERT(copyflag & K_TO_K);
2983 		bcopy(from, to, sizeof (struct strioctl));
2984 	}
2985 	return (0);
2986 }
2987 
2988 static int
2989 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
2990 {
2991 	struct	strioctl32 strioc32;
2992 	struct	strioctl *striocp;
2993 
2994 	if (copyflag & U_TO_K) {
2995 		ASSERT((copyflag & K_TO_K) == 0);
2996 
2997 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
2998 			striocp = (struct strioctl *)from;
2999 			strioc32.ic_cmd	= striocp->ic_cmd;
3000 			strioc32.ic_timout = striocp->ic_timout;
3001 			strioc32.ic_len	= striocp->ic_len;
3002 			strioc32.ic_dp	= (caddr32_t)(uintptr_t)striocp->ic_dp;
3003 			ASSERT((char *)(uintptr_t)strioc32.ic_dp ==
3004 			    striocp->ic_dp);
3005 
3006 			if (copyout(&strioc32, to, sizeof (strioc32)))
3007 				return (EFAULT);
3008 
3009 		} else { /* NATIVE data model */
3010 			if (copyout(from, to, sizeof (struct strioctl))) {
3011 				return (EFAULT);
3012 			} else {
3013 				return (0);
3014 			}
3015 		}
3016 	} else {
3017 		ASSERT(copyflag & K_TO_K);
3018 		bcopy(from, to, sizeof (struct strioctl));
3019 	}
3020 	return (0);
3021 }
3022 
3023 #else	/* ! _LP64 */
3024 
3025 /* ARGSUSED2 */
3026 static int
3027 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3028 {
3029 	return (strcopyin(from, to, sizeof (struct strioctl), copyflag));
3030 }
3031 
3032 /* ARGSUSED2 */
3033 static int
3034 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3035 {
3036 	return (strcopyout(from, to, sizeof (struct strioctl), copyflag));
3037 }
3038 
3039 #endif	/* _LP64 */
3040 
3041 /*
3042  * Determine type of job control semantics expected by user.  The
3043  * possibilities are:
3044  *	JCREAD	- Behaves like read() on fd; send SIGTTIN
3045  *	JCWRITE	- Behaves like write() on fd; send SIGTTOU if TOSTOP set
3046  *	JCSETP	- Sets a value in the stream; send SIGTTOU, ignore TOSTOP
3047  *	JCGETP	- Gets a value in the stream; no signals.
3048  * See straccess in strsubr.c for usage of these values.
3049  *
3050  * This routine also returns -1 for I_STR as a special case; the
3051  * caller must call again with the real ioctl number for
3052  * classification.
3053  */
3054 static int
3055 job_control_type(int cmd)
3056 {
3057 	switch (cmd) {
3058 	case I_STR:
3059 		return (-1);
3060 
3061 	case I_RECVFD:
3062 	case I_E_RECVFD:
3063 		return (JCREAD);
3064 
3065 	case I_FDINSERT:
3066 	case I_SENDFD:
3067 		return (JCWRITE);
3068 
3069 	case TCSETA:
3070 	case TCSETAW:
3071 	case TCSETAF:
3072 	case TCSBRK:
3073 	case TCXONC:
3074 	case TCFLSH:
3075 	case TCDSET:	/* Obsolete */
3076 	case TIOCSWINSZ:
3077 	case TCSETS:
3078 	case TCSETSW:
3079 	case TCSETSF:
3080 	case TIOCSETD:
3081 	case TIOCHPCL:
3082 	case TIOCSETP:
3083 	case TIOCSETN:
3084 	case TIOCEXCL:
3085 	case TIOCNXCL:
3086 	case TIOCFLUSH:
3087 	case TIOCSETC:
3088 	case TIOCLBIS:
3089 	case TIOCLBIC:
3090 	case TIOCLSET:
3091 	case TIOCSBRK:
3092 	case TIOCCBRK:
3093 	case TIOCSDTR:
3094 	case TIOCCDTR:
3095 	case TIOCSLTC:
3096 	case TIOCSTOP:
3097 	case TIOCSTART:
3098 	case TIOCSTI:
3099 	case TIOCSPGRP:
3100 	case TIOCMSET:
3101 	case TIOCMBIS:
3102 	case TIOCMBIC:
3103 	case TIOCREMOTE:
3104 	case TIOCSIGNAL:
3105 	case LDSETT:
3106 	case LDSMAP:	/* Obsolete */
3107 	case DIOCSETP:
3108 	case I_FLUSH:
3109 	case I_SRDOPT:
3110 	case I_SETSIG:
3111 	case I_SWROPT:
3112 	case I_FLUSHBAND:
3113 	case I_SETCLTIME:
3114 	case I_SERROPT:
3115 	case I_ESETSIG:
3116 	case FIONBIO:
3117 	case FIOASYNC:
3118 	case FIOSETOWN:
3119 	case JBOOT:	/* Obsolete */
3120 	case JTERM:	/* Obsolete */
3121 	case JTIMOM:	/* Obsolete */
3122 	case JZOMBOOT:	/* Obsolete */
3123 	case JAGENT:	/* Obsolete */
3124 	case JTRUN:	/* Obsolete */
3125 	case JXTPROTO:	/* Obsolete */
3126 	case TIOCSETLD:
3127 		return (JCSETP);
3128 	}
3129 
3130 	return (JCGETP);
3131 }
3132 
3133 /*
3134  * ioctl for streams
3135  */
3136 int
3137 strioctl(struct vnode *vp, int cmd, intptr_t arg, int flag, int copyflag,
3138     cred_t *crp, int *rvalp)
3139 {
3140 	struct stdata *stp;
3141 	struct strioctl strioc;
3142 	struct uio uio;
3143 	struct iovec iov;
3144 	int access;
3145 	mblk_t *mp;
3146 	int error = 0;
3147 	int done = 0;
3148 	ssize_t	rmin, rmax;
3149 	queue_t *wrq;
3150 	queue_t *rdq;
3151 	boolean_t kioctl = B_FALSE;
3152 
3153 	if (flag & FKIOCTL) {
3154 		copyflag = K_TO_K;
3155 		kioctl = B_TRUE;
3156 	}
3157 	ASSERT(vp->v_stream);
3158 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
3159 	stp = vp->v_stream;
3160 
3161 	TRACE_3(TR_FAC_STREAMS_FR, TR_IOCTL_ENTER,
3162 		"strioctl:stp %p cmd %X arg %lX", stp, cmd, arg);
3163 
3164 #ifdef C2_AUDIT
3165 	if (audit_active)
3166 		audit_strioctl(vp, cmd, arg, flag, copyflag, crp, rvalp);
3167 #endif
3168 
3169 	/*
3170 	 * If the copy is kernel to kernel, make sure that the FNATIVE
3171 	 * flag is set.  After this it would be a serious error to have
3172 	 * no model flag.
3173 	 */
3174 	if (copyflag == K_TO_K)
3175 		flag = (flag & ~FMODELS) | FNATIVE;
3176 
3177 	ASSERT((flag & FMODELS) != 0);
3178 
3179 	wrq = stp->sd_wrq;
3180 	rdq = _RD(wrq);
3181 
3182 	access = job_control_type(cmd);
3183 
3184 	/* We should never see these here, should be handled by iwscn */
3185 	if (cmd == SRIOCSREDIR || cmd == SRIOCISREDIR)
3186 		return (EINVAL);
3187 
3188 	mutex_enter(&stp->sd_lock);
3189 	if ((access != -1) && ((error = i_straccess(stp, access)) != 0)) {
3190 		mutex_exit(&stp->sd_lock);
3191 		return (error);
3192 	}
3193 	mutex_exit(&stp->sd_lock);
3194 
3195 	/*
3196 	 * Check for sgttyb-related ioctls first, and complain as
3197 	 * necessary.
3198 	 */
3199 	switch (cmd) {
3200 	case TIOCGETP:
3201 	case TIOCSETP:
3202 	case TIOCSETN:
3203 		if (sgttyb_handling >= 2 && !sgttyb_complaint) {
3204 			sgttyb_complaint = B_TRUE;
3205 			cmn_err(CE_NOTE,
3206 			    "application used obsolete TIOC[GS]ET");
3207 		}
3208 		if (sgttyb_handling >= 3) {
3209 			tsignal(curthread, SIGSYS);
3210 			return (EIO);
3211 		}
3212 		break;
3213 	}
3214 
3215 	mutex_enter(&stp->sd_lock);
3216 
3217 	switch (cmd) {
3218 	case I_RECVFD:
3219 	case I_E_RECVFD:
3220 	case I_PEEK:
3221 	case I_NREAD:
3222 	case FIONREAD:
3223 	case FIORDCHK:
3224 	case I_ATMARK:
3225 	case FIONBIO:
3226 	case FIOASYNC:
3227 		if (stp->sd_flag & (STRDERR|STPLEX)) {
3228 			error = strgeterr(stp, STRDERR|STPLEX, 0);
3229 			if (error != 0) {
3230 				mutex_exit(&stp->sd_lock);
3231 				return (error);
3232 			}
3233 		}
3234 		break;
3235 
3236 	default:
3237 		if (stp->sd_flag & (STRDERR|STWRERR|STPLEX)) {
3238 			error = strgeterr(stp, STRDERR|STWRERR|STPLEX, 0);
3239 			if (error != 0) {
3240 				mutex_exit(&stp->sd_lock);
3241 				return (error);
3242 			}
3243 		}
3244 	}
3245 
3246 	mutex_exit(&stp->sd_lock);
3247 
3248 	switch (cmd) {
3249 	default:
3250 		/*
3251 		 * The stream head has hardcoded knowledge of a
3252 		 * miscellaneous collection of terminal-, keyboard- and
3253 		 * mouse-related ioctls, enumerated below.  This hardcoded
3254 		 * knowledge allows the stream head to automatically
3255 		 * convert transparent ioctl requests made by userland
3256 		 * programs into I_STR ioctls which many old STREAMS
3257 		 * modules and drivers require.
3258 		 *
3259 		 * No new ioctls should ever be added to this list.
3260 		 * Instead, the STREAMS module or driver should be written
3261 		 * to either handle transparent ioctls or require any
3262 		 * userland programs to use I_STR ioctls (by returning
3263 		 * EINVAL to any transparent ioctl requests).
3264 		 *
3265 		 * More importantly, removing ioctls from this list should
3266 		 * be done with the utmost care, since our STREAMS modules
3267 		 * and drivers *count* on the stream head performing this
3268 		 * conversion, and thus may panic while processing
3269 		 * transparent ioctl request for one of these ioctls (keep
3270 		 * in mind that third party modules and drivers may have
3271 		 * similar problems).
3272 		 */
3273 		if (((cmd & IOCTYPE) == LDIOC) ||
3274 		    ((cmd & IOCTYPE) == tIOC) ||
3275 		    ((cmd & IOCTYPE) == TIOC) ||
3276 		    ((cmd & IOCTYPE) == KIOC) ||
3277 		    ((cmd & IOCTYPE) == MSIOC) ||
3278 		    ((cmd & IOCTYPE) == VUIOC)) {
3279 			/*
3280 			 * The ioctl is a tty ioctl - set up strioc buffer
3281 			 * and call strdoioctl() to do the work.
3282 			 */
3283 			if (stp->sd_flag & STRHUP)
3284 				return (ENXIO);
3285 			strioc.ic_cmd = cmd;
3286 			strioc.ic_timout = INFTIM;
3287 
3288 			switch (cmd) {
3289 
3290 			case TCXONC:
3291 			case TCSBRK:
3292 			case TCFLSH:
3293 			case TCDSET:
3294 				{
3295 				int native_arg = (int)arg;
3296 				strioc.ic_len = sizeof (int);
3297 				strioc.ic_dp = (char *)&native_arg;
3298 				return (strdoioctl(stp, &strioc, flag,
3299 				    K_TO_K, crp, rvalp));
3300 				}
3301 
3302 			case TCSETA:
3303 			case TCSETAW:
3304 			case TCSETAF:
3305 				strioc.ic_len = sizeof (struct termio);
3306 				strioc.ic_dp = (char *)arg;
3307 				return (strdoioctl(stp, &strioc, flag,
3308 					copyflag, crp, rvalp));
3309 
3310 			case TCSETS:
3311 			case TCSETSW:
3312 			case TCSETSF:
3313 				strioc.ic_len = sizeof (struct termios);
3314 				strioc.ic_dp = (char *)arg;
3315 				return (strdoioctl(stp, &strioc, flag,
3316 					copyflag, crp, rvalp));
3317 
3318 			case LDSETT:
3319 				strioc.ic_len = sizeof (struct termcb);
3320 				strioc.ic_dp = (char *)arg;
3321 				return (strdoioctl(stp, &strioc, flag,
3322 					copyflag, crp, rvalp));
3323 
3324 			case TIOCSETP:
3325 				strioc.ic_len = sizeof (struct sgttyb);
3326 				strioc.ic_dp = (char *)arg;
3327 				return (strdoioctl(stp, &strioc, flag,
3328 					copyflag, crp, rvalp));
3329 
3330 			case TIOCSTI:
3331 				if ((flag & FREAD) == 0 &&
3332 				    secpolicy_sti(crp) != 0) {
3333 					return (EPERM);
3334 				}
3335 				mutex_enter(&stp->sd_lock);
3336 				mutex_enter(&curproc->p_splock);
3337 				if (stp->sd_sidp != curproc->p_sessp->s_sidp &&
3338 				    secpolicy_sti(crp) != 0) {
3339 					mutex_exit(&curproc->p_splock);
3340 					mutex_exit(&stp->sd_lock);
3341 					return (EACCES);
3342 				}
3343 				mutex_exit(&curproc->p_splock);
3344 				mutex_exit(&stp->sd_lock);
3345 
3346 				strioc.ic_len = sizeof (char);
3347 				strioc.ic_dp = (char *)arg;
3348 				return (strdoioctl(stp, &strioc, flag,
3349 					copyflag, crp, rvalp));
3350 
3351 			case TIOCSWINSZ:
3352 				strioc.ic_len = sizeof (struct winsize);
3353 				strioc.ic_dp = (char *)arg;
3354 				return (strdoioctl(stp, &strioc, flag,
3355 					copyflag, crp, rvalp));
3356 
3357 			case TIOCSSIZE:
3358 				strioc.ic_len = sizeof (struct ttysize);
3359 				strioc.ic_dp = (char *)arg;
3360 				return (strdoioctl(stp, &strioc, flag,
3361 					copyflag, crp, rvalp));
3362 
3363 			case TIOCSSOFTCAR:
3364 			case KIOCTRANS:
3365 			case KIOCTRANSABLE:
3366 			case KIOCCMD:
3367 			case KIOCSDIRECT:
3368 			case KIOCSCOMPAT:
3369 			case KIOCSKABORTEN:
3370 			case KIOCSRPTDELAY:
3371 			case KIOCSRPTRATE:
3372 			case VUIDSFORMAT:
3373 			case TIOCSPPS:
3374 				strioc.ic_len = sizeof (int);
3375 				strioc.ic_dp = (char *)arg;
3376 				return (strdoioctl(stp, &strioc, flag,
3377 					copyflag, crp, rvalp));
3378 
3379 			case KIOCSETKEY:
3380 			case KIOCGETKEY:
3381 				strioc.ic_len = sizeof (struct kiockey);
3382 				strioc.ic_dp = (char *)arg;
3383 				return (strdoioctl(stp, &strioc, flag,
3384 					copyflag, crp, rvalp));
3385 
3386 			case KIOCSKEY:
3387 			case KIOCGKEY:
3388 				strioc.ic_len = sizeof (struct kiockeymap);
3389 				strioc.ic_dp = (char *)arg;
3390 				return (strdoioctl(stp, &strioc, flag,
3391 					copyflag, crp, rvalp));
3392 
3393 			case KIOCSLED:
3394 				/* arg is a pointer to char */
3395 				strioc.ic_len = sizeof (char);
3396 				strioc.ic_dp = (char *)arg;
3397 				return (strdoioctl(stp, &strioc, flag,
3398 					copyflag, crp, rvalp));
3399 
3400 			case MSIOSETPARMS:
3401 				strioc.ic_len = sizeof (Ms_parms);
3402 				strioc.ic_dp = (char *)arg;
3403 				return (strdoioctl(stp, &strioc, flag,
3404 					copyflag, crp, rvalp));
3405 
3406 			case VUIDSADDR:
3407 			case VUIDGADDR:
3408 				strioc.ic_len = sizeof (struct vuid_addr_probe);
3409 				strioc.ic_dp = (char *)arg;
3410 				return (strdoioctl(stp, &strioc, flag,
3411 					copyflag, crp, rvalp));
3412 
3413 			/*
3414 			 * These M_IOCTL's don't require any data to be sent
3415 			 * downstream, and the driver will allocate and link
3416 			 * on its own mblk_t upon M_IOCACK -- thus we set
3417 			 * ic_len to zero and set ic_dp to arg so we know
3418 			 * where to copyout to later.
3419 			 */
3420 			case TIOCGSOFTCAR:
3421 			case TIOCGWINSZ:
3422 			case TIOCGSIZE:
3423 			case KIOCGTRANS:
3424 			case KIOCGTRANSABLE:
3425 			case KIOCTYPE:
3426 			case KIOCGDIRECT:
3427 			case KIOCGCOMPAT:
3428 			case KIOCLAYOUT:
3429 			case KIOCGLED:
3430 			case MSIOGETPARMS:
3431 			case MSIOBUTTONS:
3432 			case VUIDGFORMAT:
3433 			case TIOCGPPS:
3434 			case TIOCGPPSEV:
3435 			case TCGETA:
3436 			case TCGETS:
3437 			case LDGETT:
3438 			case TIOCGETP:
3439 			case KIOCGRPTDELAY:
3440 			case KIOCGRPTRATE:
3441 				strioc.ic_len = 0;
3442 				strioc.ic_dp = (char *)arg;
3443 				return (strdoioctl(stp, &strioc, flag,
3444 					copyflag, crp, rvalp));
3445 			}
3446 		}
3447 
3448 		/*
3449 		 * Unknown cmd - send it down as a transparent ioctl.
3450 		 */
3451 		strioc.ic_cmd = cmd;
3452 		strioc.ic_timout = INFTIM;
3453 		strioc.ic_len = TRANSPARENT;
3454 		strioc.ic_dp = (char *)&arg;
3455 
3456 		return (strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp));
3457 
3458 	case I_STR:
3459 		/*
3460 		 * Stream ioctl.  Read in an strioctl buffer from the user
3461 		 * along with any data specified and send it downstream.
3462 		 * Strdoioctl will wait allow only one ioctl message at
3463 		 * a time, and waits for the acknowledgement.
3464 		 */
3465 
3466 		if (stp->sd_flag & STRHUP)
3467 			return (ENXIO);
3468 
3469 		error = strcopyin_strioctl((void *)arg, &strioc, flag,
3470 		    copyflag);
3471 		if (error != 0)
3472 			return (error);
3473 
3474 		if ((strioc.ic_len < 0) || (strioc.ic_timout < -1))
3475 			return (EINVAL);
3476 
3477 		access = job_control_type(strioc.ic_cmd);
3478 		mutex_enter(&stp->sd_lock);
3479 		if ((access != -1) &&
3480 		    ((error = i_straccess(stp, access)) != 0)) {
3481 			mutex_exit(&stp->sd_lock);
3482 			return (error);
3483 		}
3484 		mutex_exit(&stp->sd_lock);
3485 
3486 		/*
3487 		 * The I_STR facility provides a trap door for malicious
3488 		 * code to send down bogus streamio(7I) ioctl commands to
3489 		 * unsuspecting STREAMS modules and drivers which expect to
3490 		 * only get these messages from the stream head.
3491 		 * Explicitly prohibit any streamio ioctls which can be
3492 		 * passed downstream by the stream head.  Note that we do
3493 		 * not block all streamio ioctls because the ioctl
3494 		 * numberspace is not well managed and thus it's possible
3495 		 * that a module or driver's ioctl numbers may accidentally
3496 		 * collide with them.
3497 		 */
3498 		switch (strioc.ic_cmd) {
3499 		case I_LINK:
3500 		case I_PLINK:
3501 		case I_UNLINK:
3502 		case I_PUNLINK:
3503 		case _I_GETPEERCRED:
3504 		case _I_PLINK_LH:
3505 			return (EINVAL);
3506 		}
3507 
3508 		error = strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp);
3509 		if (error == 0) {
3510 			error = strcopyout_strioctl(&strioc, (void *)arg,
3511 			    flag, copyflag);
3512 		}
3513 		return (error);
3514 
3515 	case I_NREAD:
3516 		/*
3517 		 * Return number of bytes of data in first message
3518 		 * in queue in "arg" and return the number of messages
3519 		 * in queue in return value.
3520 		 */
3521 	    {
3522 		size_t	size;
3523 		int	retval;
3524 		int	count = 0;
3525 
3526 		mutex_enter(QLOCK(rdq));
3527 
3528 		size = msgdsize(rdq->q_first);
3529 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3530 			count++;
3531 
3532 		mutex_exit(QLOCK(rdq));
3533 		if (stp->sd_struiordq) {
3534 			infod_t infod;
3535 
3536 			infod.d_cmd = INFOD_COUNT;
3537 			infod.d_count = 0;
3538 			if (count == 0) {
3539 				infod.d_cmd |= INFOD_FIRSTBYTES;
3540 				infod.d_bytes = 0;
3541 			}
3542 			infod.d_res = 0;
3543 			(void) infonext(rdq, &infod);
3544 			count += infod.d_count;
3545 			if (infod.d_res & INFOD_FIRSTBYTES)
3546 				size = infod.d_bytes;
3547 		}
3548 
3549 		/*
3550 		 * Drop down from size_t to the "int" required by the
3551 		 * interface.  Cap at INT_MAX.
3552 		 */
3553 		retval = MIN(size, INT_MAX);
3554 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3555 		    copyflag);
3556 		if (!error)
3557 			*rvalp = count;
3558 		return (error);
3559 	    }
3560 
3561 	case FIONREAD:
3562 		/*
3563 		 * Return number of bytes of data in all data messages
3564 		 * in queue in "arg".
3565 		 */
3566 	    {
3567 		size_t	size = 0;
3568 		int	retval;
3569 
3570 		mutex_enter(QLOCK(rdq));
3571 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3572 			size += msgdsize(mp);
3573 		mutex_exit(QLOCK(rdq));
3574 
3575 		if (stp->sd_struiordq) {
3576 			infod_t infod;
3577 
3578 			infod.d_cmd = INFOD_BYTES;
3579 			infod.d_res = 0;
3580 			infod.d_bytes = 0;
3581 			(void) infonext(rdq, &infod);
3582 			size += infod.d_bytes;
3583 		}
3584 
3585 		/*
3586 		 * Drop down from size_t to the "int" required by the
3587 		 * interface.  Cap at INT_MAX.
3588 		 */
3589 		retval = MIN(size, INT_MAX);
3590 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3591 		    copyflag);
3592 
3593 		*rvalp = 0;
3594 		return (error);
3595 	    }
3596 	case FIORDCHK:
3597 		/*
3598 		 * FIORDCHK does not use arg value (like FIONREAD),
3599 		 * instead a count is returned. I_NREAD value may
3600 		 * not be accurate but safe. The real thing to do is
3601 		 * to add the msgdsizes of all data  messages until
3602 		 * a non-data message.
3603 		 */
3604 	    {
3605 		size_t size = 0;
3606 
3607 		mutex_enter(QLOCK(rdq));
3608 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3609 			size += msgdsize(mp);
3610 		mutex_exit(QLOCK(rdq));
3611 
3612 		if (stp->sd_struiordq) {
3613 			infod_t infod;
3614 
3615 			infod.d_cmd = INFOD_BYTES;
3616 			infod.d_res = 0;
3617 			infod.d_bytes = 0;
3618 			(void) infonext(rdq, &infod);
3619 			size += infod.d_bytes;
3620 		}
3621 
3622 		/*
3623 		 * Since ioctl returns an int, and memory sizes under
3624 		 * LP64 may not fit, we return INT_MAX if the count was
3625 		 * actually greater.
3626 		 */
3627 		*rvalp = MIN(size, INT_MAX);
3628 		return (0);
3629 	    }
3630 
3631 	case I_FIND:
3632 		/*
3633 		 * Get module name.
3634 		 */
3635 	    {
3636 		char mname[FMNAMESZ + 1];
3637 		queue_t *q;
3638 
3639 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3640 		    mname, FMNAMESZ + 1, NULL);
3641 		if (error)
3642 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3643 
3644 		/*
3645 		 * Return EINVAL if we're handed a bogus module name.
3646 		 */
3647 		if (fmodsw_find(mname, FMODSW_LOAD) == NULL) {
3648 			TRACE_0(TR_FAC_STREAMS_FR,
3649 				TR_I_CANT_FIND, "couldn't I_FIND");
3650 			return (EINVAL);
3651 		}
3652 
3653 		*rvalp = 0;
3654 
3655 		/* Look downstream to see if module is there. */
3656 		claimstr(stp->sd_wrq);
3657 		for (q = stp->sd_wrq->q_next; q; q = q->q_next) {
3658 			if (q->q_flag&QREADR) {
3659 				q = NULL;
3660 				break;
3661 			}
3662 			if (strcmp(mname, q->q_qinfo->qi_minfo->mi_idname) == 0)
3663 				break;
3664 		}
3665 		releasestr(stp->sd_wrq);
3666 
3667 		*rvalp = (q ? 1 : 0);
3668 		return (error);
3669 	    }
3670 
3671 	case I_PUSH:
3672 	case __I_PUSH_NOCTTY:
3673 		/*
3674 		 * Push a module.
3675 		 * For the case __I_PUSH_NOCTTY push a module but
3676 		 * do not allocate controlling tty. See bugid 4025044
3677 		 */
3678 
3679 	    {
3680 		char mname[FMNAMESZ + 1];
3681 		fmodsw_impl_t *fp;
3682 		dev_t dummydev;
3683 
3684 		if (stp->sd_flag & STRHUP)
3685 			return (ENXIO);
3686 
3687 		/*
3688 		 * Get module name and look up in fmodsw.
3689 		 */
3690 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3691 		    mname, FMNAMESZ + 1, NULL);
3692 		if (error)
3693 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3694 
3695 		if ((fp = fmodsw_find(mname, FMODSW_HOLD | FMODSW_LOAD)) ==
3696 		    NULL)
3697 			return (EINVAL);
3698 
3699 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_PUSH,
3700 		    "I_PUSH:fp %p stp %p", fp, stp);
3701 
3702 		if (error = strstartplumb(stp, flag, cmd)) {
3703 			fmodsw_rele(fp);
3704 			return (error);
3705 		}
3706 
3707 		/*
3708 		 * See if any more modules can be pushed on this stream.
3709 		 * Note that this check must be done after strstartplumb()
3710 		 * since otherwise multiple threads issuing I_PUSHes on
3711 		 * the same stream will be able to exceed nstrpush.
3712 		 */
3713 		mutex_enter(&stp->sd_lock);
3714 		if (stp->sd_pushcnt >= nstrpush) {
3715 			fmodsw_rele(fp);
3716 			strendplumb(stp);
3717 			mutex_exit(&stp->sd_lock);
3718 			return (EINVAL);
3719 		}
3720 		mutex_exit(&stp->sd_lock);
3721 
3722 		/*
3723 		 * Push new module and call its open routine
3724 		 * via qattach().  Modules don't change device
3725 		 * numbers, so just ignore dummydev here.
3726 		 */
3727 		dummydev = vp->v_rdev;
3728 		if ((error = qattach(rdq, &dummydev, 0, crp, fp,
3729 		    B_FALSE)) == 0) {
3730 			if (vp->v_type == VCHR && /* sorry, no pipes allowed */
3731 			    (cmd == I_PUSH) && (stp->sd_flag & STRISTTY)) {
3732 				/*
3733 				 * try to allocate it as a controlling terminal
3734 				 */
3735 				(void) strctty(stp);
3736 			}
3737 		}
3738 
3739 		mutex_enter(&stp->sd_lock);
3740 
3741 		/*
3742 		 * As a performance concern we are caching the values of
3743 		 * q_minpsz and q_maxpsz of the module below the stream
3744 		 * head in the stream head.
3745 		 */
3746 		mutex_enter(QLOCK(stp->sd_wrq->q_next));
3747 		rmin = stp->sd_wrq->q_next->q_minpsz;
3748 		rmax = stp->sd_wrq->q_next->q_maxpsz;
3749 		mutex_exit(QLOCK(stp->sd_wrq->q_next));
3750 
3751 		/* Do this processing here as a performance concern */
3752 		if (strmsgsz != 0) {
3753 			if (rmax == INFPSZ)
3754 				rmax = strmsgsz;
3755 			else  {
3756 				if (vp->v_type == VFIFO)
3757 					rmax = MIN(PIPE_BUF, rmax);
3758 				else	rmax = MIN(strmsgsz, rmax);
3759 			}
3760 		}
3761 
3762 		mutex_enter(QLOCK(wrq));
3763 		stp->sd_qn_minpsz = rmin;
3764 		stp->sd_qn_maxpsz = rmax;
3765 		mutex_exit(QLOCK(wrq));
3766 
3767 		strendplumb(stp);
3768 		mutex_exit(&stp->sd_lock);
3769 		return (error);
3770 	    }
3771 
3772 	case I_POP:
3773 	    {
3774 		queue_t	*q;
3775 
3776 		if (stp->sd_flag & STRHUP)
3777 			return (ENXIO);
3778 		if (!wrq->q_next)	/* for broken pipes */
3779 			return (EINVAL);
3780 
3781 		if (error = strstartplumb(stp, flag, cmd))
3782 			return (error);
3783 
3784 		/*
3785 		 * If there is an anchor on this stream and popping
3786 		 * the current module would attempt to pop through the
3787 		 * anchor, then disallow the pop unless we have sufficient
3788 		 * privileges; take the cheapest (non-locking) check
3789 		 * first.
3790 		 */
3791 		if (secpolicy_ip_config(crp, B_TRUE) != 0 ||
3792 		    (stp->sd_anchorzone != crgetzoneid(crp))) {
3793 			mutex_enter(&stp->sd_lock);
3794 			/*
3795 			 * Anchors only apply if there's at least one
3796 			 * module on the stream (sd_pushcnt > 0).
3797 			 */
3798 			if (stp->sd_pushcnt > 0 &&
3799 			    stp->sd_pushcnt == stp->sd_anchor &&
3800 			    stp->sd_vnode->v_type != VFIFO) {
3801 				strendplumb(stp);
3802 				mutex_exit(&stp->sd_lock);
3803 				if (stp->sd_anchorzone != crgetzoneid(crp))
3804 					return (EINVAL);
3805 				/* Audit and report error */
3806 				return (secpolicy_ip_config(crp, B_FALSE));
3807 			}
3808 			mutex_exit(&stp->sd_lock);
3809 		}
3810 
3811 		q = wrq->q_next;
3812 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_POP,
3813 			"I_POP:%p from %p", q, stp);
3814 		if (q->q_next == NULL || (q->q_flag & (QREADR|QISDRV))) {
3815 			error = EINVAL;
3816 		} else {
3817 			qdetach(_RD(q), 1, flag, crp, B_FALSE);
3818 			error = 0;
3819 		}
3820 		mutex_enter(&stp->sd_lock);
3821 
3822 		/*
3823 		 * As a performance concern we are caching the values of
3824 		 * q_minpsz and q_maxpsz of the module below the stream
3825 		 * head in the stream head.
3826 		 */
3827 		mutex_enter(QLOCK(wrq->q_next));
3828 		rmin = wrq->q_next->q_minpsz;
3829 		rmax = wrq->q_next->q_maxpsz;
3830 		mutex_exit(QLOCK(wrq->q_next));
3831 
3832 		/* Do this processing here as a performance concern */
3833 		if (strmsgsz != 0) {
3834 			if (rmax == INFPSZ)
3835 				rmax = strmsgsz;
3836 			else  {
3837 				if (vp->v_type == VFIFO)
3838 					rmax = MIN(PIPE_BUF, rmax);
3839 				else	rmax = MIN(strmsgsz, rmax);
3840 			}
3841 		}
3842 
3843 		mutex_enter(QLOCK(wrq));
3844 		stp->sd_qn_minpsz = rmin;
3845 		stp->sd_qn_maxpsz = rmax;
3846 		mutex_exit(QLOCK(wrq));
3847 
3848 		/* If we popped through the anchor, then reset the anchor. */
3849 		if (stp->sd_pushcnt < stp->sd_anchor) {
3850 			stp->sd_anchor = 0;
3851 			stp->sd_anchorzone = 0;
3852 		}
3853 		strendplumb(stp);
3854 		mutex_exit(&stp->sd_lock);
3855 		return (error);
3856 	    }
3857 
3858 	case _I_MUXID2FD:
3859 	{
3860 		/*
3861 		 * Create a fd for a I_PLINK'ed lower stream with a given
3862 		 * muxid.  With the fd, application can send down ioctls,
3863 		 * like I_LIST, to the previously I_PLINK'ed stream.  Note
3864 		 * that after getting the fd, the application has to do an
3865 		 * I_PUNLINK on the muxid before it can do any operation
3866 		 * on the lower stream.  This is required by spec1170.
3867 		 *
3868 		 * The fd used to do this ioctl should point to the same
3869 		 * controlling device used to do the I_PLINK.  If it uses
3870 		 * a different stream or an invalid muxid, I_MUXID2FD will
3871 		 * fail.  The error code is set to EINVAL.
3872 		 *
3873 		 * The intended use of this interface is the following.
3874 		 * An application I_PLINK'ed a stream and exits.  The fd
3875 		 * to the lower stream is gone.  Another application
3876 		 * wants to get a fd to the lower stream, it uses I_MUXID2FD.
3877 		 */
3878 		int muxid = (int)arg;
3879 		int fd;
3880 		linkinfo_t *linkp;
3881 		struct file *fp;
3882 		netstack_t *ns;
3883 		str_stack_t *ss;
3884 
3885 		/*
3886 		 * Do not allow the wildcard muxid.  This ioctl is not
3887 		 * intended to find arbitrary link.
3888 		 */
3889 		if (muxid == 0) {
3890 			return (EINVAL);
3891 		}
3892 
3893 		ns = netstack_find_by_cred(crp);
3894 		ASSERT(ns != NULL);
3895 		ss = ns->netstack_str;
3896 		ASSERT(ss != NULL);
3897 
3898 		mutex_enter(&muxifier);
3899 		linkp = findlinks(vp->v_stream, muxid, LINKPERSIST, ss);
3900 		if (linkp == NULL) {
3901 			mutex_exit(&muxifier);
3902 			netstack_rele(ss->ss_netstack);
3903 			return (EINVAL);
3904 		}
3905 
3906 		if ((fd = ufalloc(0)) == -1) {
3907 			mutex_exit(&muxifier);
3908 			netstack_rele(ss->ss_netstack);
3909 			return (EMFILE);
3910 		}
3911 		fp = linkp->li_fpdown;
3912 		mutex_enter(&fp->f_tlock);
3913 		fp->f_count++;
3914 		mutex_exit(&fp->f_tlock);
3915 		mutex_exit(&muxifier);
3916 		setf(fd, fp);
3917 		*rvalp = fd;
3918 		netstack_rele(ss->ss_netstack);
3919 		return (0);
3920 	}
3921 
3922 	case _I_INSERT:
3923 	{
3924 		/*
3925 		 * To insert a module to a given position in a stream.
3926 		 * In the first release, only allow privileged user
3927 		 * to use this ioctl. Furthermore, the insert is only allowed
3928 		 * below an anchor if the zoneid is the same as the zoneid
3929 		 * which created the anchor.
3930 		 *
3931 		 * Note that we do not plan to support this ioctl
3932 		 * on pipes in the first release.  We want to learn more
3933 		 * about the implications of these ioctls before extending
3934 		 * their support.  And we do not think these features are
3935 		 * valuable for pipes.
3936 		 *
3937 		 * Neither do we support O/C hot stream.  Note that only
3938 		 * the upper streams of TCP/IP stack are O/C hot streams.
3939 		 * The lower IP stream is not.
3940 		 * When there is a O/C cold barrier, we only allow inserts
3941 		 * above the barrier.
3942 		 */
3943 		STRUCT_DECL(strmodconf, strmodinsert);
3944 		char mod_name[FMNAMESZ + 1];
3945 		fmodsw_impl_t *fp;
3946 		dev_t dummydev;
3947 		queue_t *tmp_wrq;
3948 		int pos;
3949 		boolean_t is_insert;
3950 
3951 		STRUCT_INIT(strmodinsert, flag);
3952 		if (stp->sd_flag & STRHUP)
3953 			return (ENXIO);
3954 		if (STRMATED(stp))
3955 			return (EINVAL);
3956 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
3957 			return (error);
3958 		if (stp->sd_anchor != 0 &&
3959 		    stp->sd_anchorzone != crgetzoneid(crp))
3960 			return (EINVAL);
3961 
3962 		error = strcopyin((void *)arg, STRUCT_BUF(strmodinsert),
3963 		    STRUCT_SIZE(strmodinsert), copyflag);
3964 		if (error)
3965 			return (error);
3966 
3967 		/*
3968 		 * Get module name and look up in fmodsw.
3969 		 */
3970 		error = (copyflag & U_TO_K ? copyinstr :
3971 		    copystr)(STRUCT_FGETP(strmodinsert, mod_name),
3972 		    mod_name, FMNAMESZ + 1, NULL);
3973 		if (error)
3974 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3975 
3976 		if ((fp = fmodsw_find(mod_name, FMODSW_HOLD | FMODSW_LOAD)) ==
3977 		    NULL)
3978 			return (EINVAL);
3979 
3980 		if (error = strstartplumb(stp, flag, cmd)) {
3981 			fmodsw_rele(fp);
3982 			return (error);
3983 		}
3984 
3985 		/*
3986 		 * Is this _I_INSERT just like an I_PUSH?  We need to know
3987 		 * this because we do some optimizations if this is a
3988 		 * module being pushed.
3989 		 */
3990 		pos = STRUCT_FGET(strmodinsert, pos);
3991 		is_insert = (pos != 0);
3992 
3993 		/*
3994 		 * Make sure pos is valid.  Even though it is not an I_PUSH,
3995 		 * we impose the same limit on the number of modules in a
3996 		 * stream.
3997 		 */
3998 		mutex_enter(&stp->sd_lock);
3999 		if (stp->sd_pushcnt >= nstrpush || pos < 0 ||
4000 		    pos > stp->sd_pushcnt) {
4001 			fmodsw_rele(fp);
4002 			strendplumb(stp);
4003 			mutex_exit(&stp->sd_lock);
4004 			return (EINVAL);
4005 		}
4006 		if (stp->sd_anchor != 0) {
4007 			/*
4008 			 * Is this insert below the anchor?
4009 			 * Pushcnt hasn't been increased yet hence
4010 			 * we test for greater than here, and greater or
4011 			 * equal after qattach.
4012 			 */
4013 			if (pos > (stp->sd_pushcnt - stp->sd_anchor) &&
4014 			    stp->sd_anchorzone != crgetzoneid(crp)) {
4015 				fmodsw_rele(fp);
4016 				strendplumb(stp);
4017 				mutex_exit(&stp->sd_lock);
4018 				return (EPERM);
4019 			}
4020 		}
4021 
4022 		mutex_exit(&stp->sd_lock);
4023 
4024 		/*
4025 		 * First find the correct position this module to
4026 		 * be inserted.  We don't need to call claimstr()
4027 		 * as the stream should not be changing at this point.
4028 		 *
4029 		 * Insert new module and call its open routine
4030 		 * via qattach().  Modules don't change device
4031 		 * numbers, so just ignore dummydev here.
4032 		 */
4033 		for (tmp_wrq = stp->sd_wrq; pos > 0;
4034 		    tmp_wrq = tmp_wrq->q_next, pos--) {
4035 			ASSERT(SAMESTR(tmp_wrq));
4036 		}
4037 		dummydev = vp->v_rdev;
4038 		if ((error = qattach(_RD(tmp_wrq), &dummydev, 0, crp,
4039 		    fp, is_insert)) != 0) {
4040 			mutex_enter(&stp->sd_lock);
4041 			strendplumb(stp);
4042 			mutex_exit(&stp->sd_lock);
4043 			return (error);
4044 		}
4045 
4046 		mutex_enter(&stp->sd_lock);
4047 
4048 		/*
4049 		 * As a performance concern we are caching the values of
4050 		 * q_minpsz and q_maxpsz of the module below the stream
4051 		 * head in the stream head.
4052 		 */
4053 		if (!is_insert) {
4054 			mutex_enter(QLOCK(stp->sd_wrq->q_next));
4055 			rmin = stp->sd_wrq->q_next->q_minpsz;
4056 			rmax = stp->sd_wrq->q_next->q_maxpsz;
4057 			mutex_exit(QLOCK(stp->sd_wrq->q_next));
4058 
4059 			/* Do this processing here as a performance concern */
4060 			if (strmsgsz != 0) {
4061 				if (rmax == INFPSZ) {
4062 					rmax = strmsgsz;
4063 				} else  {
4064 					rmax = MIN(strmsgsz, rmax);
4065 				}
4066 			}
4067 
4068 			mutex_enter(QLOCK(wrq));
4069 			stp->sd_qn_minpsz = rmin;
4070 			stp->sd_qn_maxpsz = rmax;
4071 			mutex_exit(QLOCK(wrq));
4072 		}
4073 
4074 		/*
4075 		 * Need to update the anchor value if this module is
4076 		 * inserted below the anchor point.
4077 		 */
4078 		if (stp->sd_anchor != 0) {
4079 			pos = STRUCT_FGET(strmodinsert, pos);
4080 			if (pos >= (stp->sd_pushcnt - stp->sd_anchor))
4081 				stp->sd_anchor++;
4082 		}
4083 
4084 		strendplumb(stp);
4085 		mutex_exit(&stp->sd_lock);
4086 		return (0);
4087 	}
4088 
4089 	case _I_REMOVE:
4090 	{
4091 		/*
4092 		 * To remove a module with a given name in a stream.  The
4093 		 * caller of this ioctl needs to provide both the name and
4094 		 * the position of the module to be removed.  This eliminates
4095 		 * the ambiguity of removal if a module is inserted/pushed
4096 		 * multiple times in a stream.  In the first release, only
4097 		 * allow privileged user to use this ioctl.
4098 		 * Furthermore, the remove is only allowed
4099 		 * below an anchor if the zoneid is the same as the zoneid
4100 		 * which created the anchor.
4101 		 *
4102 		 * Note that we do not plan to support this ioctl
4103 		 * on pipes in the first release.  We want to learn more
4104 		 * about the implications of these ioctls before extending
4105 		 * their support.  And we do not think these features are
4106 		 * valuable for pipes.
4107 		 *
4108 		 * Neither do we support O/C hot stream.  Note that only
4109 		 * the upper streams of TCP/IP stack are O/C hot streams.
4110 		 * The lower IP stream is not.
4111 		 * When there is a O/C cold barrier we do not allow removal
4112 		 * below the barrier.
4113 		 *
4114 		 * Also note that _I_REMOVE cannot be used to remove a
4115 		 * driver or the stream head.
4116 		 */
4117 		STRUCT_DECL(strmodconf, strmodremove);
4118 		queue_t	*q;
4119 		int pos;
4120 		char mod_name[FMNAMESZ + 1];
4121 		boolean_t is_remove;
4122 
4123 		STRUCT_INIT(strmodremove, flag);
4124 		if (stp->sd_flag & STRHUP)
4125 			return (ENXIO);
4126 		if (STRMATED(stp))
4127 			return (EINVAL);
4128 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4129 			return (error);
4130 		if (stp->sd_anchor != 0 &&
4131 		    stp->sd_anchorzone != crgetzoneid(crp))
4132 			return (EINVAL);
4133 
4134 		error = strcopyin((void *)arg, STRUCT_BUF(strmodremove),
4135 		    STRUCT_SIZE(strmodremove), copyflag);
4136 		if (error)
4137 			return (error);
4138 
4139 		error = (copyflag & U_TO_K ? copyinstr :
4140 		    copystr)(STRUCT_FGETP(strmodremove, mod_name),
4141 		    mod_name, FMNAMESZ + 1, NULL);
4142 		if (error)
4143 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4144 
4145 		if ((error = strstartplumb(stp, flag, cmd)) != 0)
4146 			return (error);
4147 
4148 		/*
4149 		 * Match the name of given module to the name of module at
4150 		 * the given position.
4151 		 */
4152 		pos = STRUCT_FGET(strmodremove, pos);
4153 
4154 		is_remove = (pos != 0);
4155 		for (q = stp->sd_wrq->q_next; SAMESTR(q) && pos > 0;
4156 		    q = q->q_next, pos--)
4157 			;
4158 		if (pos > 0 || ! SAMESTR(q) ||
4159 		    strncmp(q->q_qinfo->qi_minfo->mi_idname, mod_name,
4160 		    strlen(q->q_qinfo->qi_minfo->mi_idname)) != 0) {
4161 			mutex_enter(&stp->sd_lock);
4162 			strendplumb(stp);
4163 			mutex_exit(&stp->sd_lock);
4164 			return (EINVAL);
4165 		}
4166 
4167 		/*
4168 		 * If the position is at or below an anchor, then the zoneid
4169 		 * must match the zoneid that created the anchor.
4170 		 */
4171 		if (stp->sd_anchor != 0) {
4172 			pos = STRUCT_FGET(strmodremove, pos);
4173 			if (pos >= (stp->sd_pushcnt - stp->sd_anchor) &&
4174 			    stp->sd_anchorzone != crgetzoneid(crp)) {
4175 				mutex_enter(&stp->sd_lock);
4176 				strendplumb(stp);
4177 				mutex_exit(&stp->sd_lock);
4178 				return (EPERM);
4179 			}
4180 		}
4181 
4182 
4183 		ASSERT(!(q->q_flag & QREADR));
4184 		qdetach(_RD(q), 1, flag, crp, is_remove);
4185 
4186 		mutex_enter(&stp->sd_lock);
4187 
4188 		/*
4189 		 * As a performance concern we are caching the values of
4190 		 * q_minpsz and q_maxpsz of the module below the stream
4191 		 * head in the stream head.
4192 		 */
4193 		if (!is_remove) {
4194 			mutex_enter(QLOCK(wrq->q_next));
4195 			rmin = wrq->q_next->q_minpsz;
4196 			rmax = wrq->q_next->q_maxpsz;
4197 			mutex_exit(QLOCK(wrq->q_next));
4198 
4199 			/* Do this processing here as a performance concern */
4200 			if (strmsgsz != 0) {
4201 				if (rmax == INFPSZ)
4202 					rmax = strmsgsz;
4203 				else  {
4204 					if (vp->v_type == VFIFO)
4205 						rmax = MIN(PIPE_BUF, rmax);
4206 					else	rmax = MIN(strmsgsz, rmax);
4207 				}
4208 			}
4209 
4210 			mutex_enter(QLOCK(wrq));
4211 			stp->sd_qn_minpsz = rmin;
4212 			stp->sd_qn_maxpsz = rmax;
4213 			mutex_exit(QLOCK(wrq));
4214 		}
4215 
4216 		/*
4217 		 * Need to update the anchor value if this module is removed
4218 		 * at or below the anchor point.  If the removed module is at
4219 		 * the anchor point, remove the anchor for this stream if
4220 		 * there is no module above the anchor point.  Otherwise, if
4221 		 * the removed module is below the anchor point, decrement the
4222 		 * anchor point by 1.
4223 		 */
4224 		if (stp->sd_anchor != 0) {
4225 			pos = STRUCT_FGET(strmodremove, pos);
4226 			if (pos == stp->sd_pushcnt - stp->sd_anchor + 1)
4227 				stp->sd_anchor = 0;
4228 			else if (pos > (stp->sd_pushcnt - stp->sd_anchor + 1))
4229 				stp->sd_anchor--;
4230 		}
4231 
4232 		strendplumb(stp);
4233 		mutex_exit(&stp->sd_lock);
4234 		return (0);
4235 	}
4236 
4237 	case I_ANCHOR:
4238 		/*
4239 		 * Set the anchor position on the stream to reside at
4240 		 * the top module (in other words, the top module
4241 		 * cannot be popped).  Anchors with a FIFO make no
4242 		 * obvious sense, so they're not allowed.
4243 		 */
4244 		mutex_enter(&stp->sd_lock);
4245 
4246 		if (stp->sd_vnode->v_type == VFIFO) {
4247 			mutex_exit(&stp->sd_lock);
4248 			return (EINVAL);
4249 		}
4250 		/* Only allow the same zoneid to update the anchor */
4251 		if (stp->sd_anchor != 0 &&
4252 		    stp->sd_anchorzone != crgetzoneid(crp)) {
4253 			mutex_exit(&stp->sd_lock);
4254 			return (EINVAL);
4255 		}
4256 		stp->sd_anchor = stp->sd_pushcnt;
4257 		stp->sd_anchorzone = crgetzoneid(crp);
4258 		mutex_exit(&stp->sd_lock);
4259 		return (0);
4260 
4261 	case I_LOOK:
4262 		/*
4263 		 * Get name of first module downstream.
4264 		 * If no module, return an error.
4265 		 */
4266 	    {
4267 		claimstr(wrq);
4268 		if (_SAMESTR(wrq) && wrq->q_next->q_next) {
4269 			char *name = wrq->q_next->q_qinfo->qi_minfo->mi_idname;
4270 			error = strcopyout(name, (void *)arg, strlen(name) + 1,
4271 			    copyflag);
4272 			releasestr(wrq);
4273 			return (error);
4274 		}
4275 		releasestr(wrq);
4276 		return (EINVAL);
4277 	    }
4278 
4279 	case I_LINK:
4280 	case I_PLINK:
4281 		/*
4282 		 * Link a multiplexor.
4283 		 */
4284 		error = mlink(vp, cmd, (int)arg, crp, rvalp, 0);
4285 		return (error);
4286 
4287 	case _I_PLINK_LH:
4288 		/*
4289 		 * Link a multiplexor: Call must originate from kernel.
4290 		 */
4291 		if (kioctl)
4292 			return (ldi_mlink_lh(vp, cmd, arg, crp, rvalp));
4293 
4294 		return (EINVAL);
4295 	case I_UNLINK:
4296 	case I_PUNLINK:
4297 		/*
4298 		 * Unlink a multiplexor.
4299 		 * If arg is -1, unlink all links for which this is the
4300 		 * controlling stream.  Otherwise, arg is an index number
4301 		 * for a link to be removed.
4302 		 */
4303 	    {
4304 		struct linkinfo *linkp;
4305 		int native_arg = (int)arg;
4306 		int type;
4307 		netstack_t *ns;
4308 		str_stack_t *ss;
4309 
4310 		TRACE_1(TR_FAC_STREAMS_FR,
4311 			TR_I_UNLINK, "I_UNLINK/I_PUNLINK:%p", stp);
4312 		if (vp->v_type == VFIFO) {
4313 			return (EINVAL);
4314 		}
4315 		if (cmd == I_UNLINK)
4316 			type = LINKNORMAL;
4317 		else	/* I_PUNLINK */
4318 			type = LINKPERSIST;
4319 		if (native_arg == 0) {
4320 			return (EINVAL);
4321 		}
4322 		ns = netstack_find_by_cred(crp);
4323 		ASSERT(ns != NULL);
4324 		ss = ns->netstack_str;
4325 		ASSERT(ss != NULL);
4326 
4327 		if (native_arg == MUXID_ALL)
4328 			error = munlinkall(stp, type, crp, rvalp, ss);
4329 		else {
4330 			mutex_enter(&muxifier);
4331 			if (!(linkp = findlinks(stp, (int)arg, type, ss))) {
4332 				/* invalid user supplied index number */
4333 				mutex_exit(&muxifier);
4334 				netstack_rele(ss->ss_netstack);
4335 				return (EINVAL);
4336 			}
4337 			/* munlink drops the muxifier lock */
4338 			error = munlink(stp, linkp, type, crp, rvalp, ss);
4339 		}
4340 		netstack_rele(ss->ss_netstack);
4341 		return (error);
4342 	    }
4343 
4344 	case I_FLUSH:
4345 		/*
4346 		 * send a flush message downstream
4347 		 * flush message can indicate
4348 		 * FLUSHR - flush read queue
4349 		 * FLUSHW - flush write queue
4350 		 * FLUSHRW - flush read/write queue
4351 		 */
4352 		if (stp->sd_flag & STRHUP)
4353 			return (ENXIO);
4354 		if (arg & ~FLUSHRW)
4355 			return (EINVAL);
4356 
4357 		for (;;) {
4358 			if (putnextctl1(stp->sd_wrq, M_FLUSH, (int)arg)) {
4359 				break;
4360 			}
4361 			if (error = strwaitbuf(1, BPRI_HI)) {
4362 				return (error);
4363 			}
4364 		}
4365 
4366 		/*
4367 		 * Send down an unsupported ioctl and wait for the nack
4368 		 * in order to allow the M_FLUSH to propagate back
4369 		 * up to the stream head.
4370 		 * Replaces if (qready()) runqueues();
4371 		 */
4372 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4373 		strioc.ic_timout = 0;
4374 		strioc.ic_len = 0;
4375 		strioc.ic_dp = NULL;
4376 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4377 		*rvalp = 0;
4378 		return (0);
4379 
4380 	case I_FLUSHBAND:
4381 	    {
4382 		struct bandinfo binfo;
4383 
4384 		error = strcopyin((void *)arg, &binfo, sizeof (binfo),
4385 		    copyflag);
4386 		if (error)
4387 			return (error);
4388 		if (stp->sd_flag & STRHUP)
4389 			return (ENXIO);
4390 		if (binfo.bi_flag & ~FLUSHRW)
4391 			return (EINVAL);
4392 		while (!(mp = allocb(2, BPRI_HI))) {
4393 			if (error = strwaitbuf(2, BPRI_HI))
4394 				return (error);
4395 		}
4396 		mp->b_datap->db_type = M_FLUSH;
4397 		*mp->b_wptr++ = binfo.bi_flag | FLUSHBAND;
4398 		*mp->b_wptr++ = binfo.bi_pri;
4399 		putnext(stp->sd_wrq, mp);
4400 		/*
4401 		 * Send down an unsupported ioctl and wait for the nack
4402 		 * in order to allow the M_FLUSH to propagate back
4403 		 * up to the stream head.
4404 		 * Replaces if (qready()) runqueues();
4405 		 */
4406 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4407 		strioc.ic_timout = 0;
4408 		strioc.ic_len = 0;
4409 		strioc.ic_dp = NULL;
4410 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4411 		*rvalp = 0;
4412 		return (0);
4413 	    }
4414 
4415 	case I_SRDOPT:
4416 		/*
4417 		 * Set read options
4418 		 *
4419 		 * RNORM - default stream mode
4420 		 * RMSGN - message no discard
4421 		 * RMSGD - message discard
4422 		 * RPROTNORM - fail read with EBADMSG for M_[PC]PROTOs
4423 		 * RPROTDAT - convert M_[PC]PROTOs to M_DATAs
4424 		 * RPROTDIS - discard M_[PC]PROTOs and retain M_DATAs
4425 		 */
4426 		if (arg & ~(RMODEMASK | RPROTMASK))
4427 			return (EINVAL);
4428 
4429 		if ((arg & (RMSGD|RMSGN)) == (RMSGD|RMSGN))
4430 			return (EINVAL);
4431 
4432 		mutex_enter(&stp->sd_lock);
4433 		switch (arg & RMODEMASK) {
4434 		case RNORM:
4435 			stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
4436 			break;
4437 		case RMSGD:
4438 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGNODIS) |
4439 			    RD_MSGDIS;
4440 			break;
4441 		case RMSGN:
4442 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGDIS) |
4443 			    RD_MSGNODIS;
4444 			break;
4445 		}
4446 
4447 		switch (arg & RPROTMASK) {
4448 		case RPROTNORM:
4449 			stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
4450 			break;
4451 
4452 		case RPROTDAT:
4453 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDIS) |
4454 			    RD_PROTDAT);
4455 			break;
4456 
4457 		case RPROTDIS:
4458 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDAT) |
4459 			    RD_PROTDIS);
4460 			break;
4461 		}
4462 		mutex_exit(&stp->sd_lock);
4463 		return (0);
4464 
4465 	case I_GRDOPT:
4466 		/*
4467 		 * Get read option and return the value
4468 		 * to spot pointed to by arg
4469 		 */
4470 	    {
4471 		int rdopt;
4472 
4473 		rdopt = ((stp->sd_read_opt & RD_MSGDIS) ? RMSGD :
4474 		    ((stp->sd_read_opt & RD_MSGNODIS) ? RMSGN : RNORM));
4475 		rdopt |= ((stp->sd_read_opt & RD_PROTDAT) ? RPROTDAT :
4476 		    ((stp->sd_read_opt & RD_PROTDIS) ? RPROTDIS : RPROTNORM));
4477 
4478 		return (strcopyout(&rdopt, (void *)arg, sizeof (int),
4479 		    copyflag));
4480 	    }
4481 
4482 	case I_SERROPT:
4483 		/*
4484 		 * Set error options
4485 		 *
4486 		 * RERRNORM - persistent read errors
4487 		 * RERRNONPERSIST - non-persistent read errors
4488 		 * WERRNORM - persistent write errors
4489 		 * WERRNONPERSIST - non-persistent write errors
4490 		 */
4491 		if (arg & ~(RERRMASK | WERRMASK))
4492 			return (EINVAL);
4493 
4494 		mutex_enter(&stp->sd_lock);
4495 		switch (arg & RERRMASK) {
4496 		case RERRNORM:
4497 			stp->sd_flag &= ~STRDERRNONPERSIST;
4498 			break;
4499 		case RERRNONPERSIST:
4500 			stp->sd_flag |= STRDERRNONPERSIST;
4501 			break;
4502 		}
4503 		switch (arg & WERRMASK) {
4504 		case WERRNORM:
4505 			stp->sd_flag &= ~STWRERRNONPERSIST;
4506 			break;
4507 		case WERRNONPERSIST:
4508 			stp->sd_flag |= STWRERRNONPERSIST;
4509 			break;
4510 		}
4511 		mutex_exit(&stp->sd_lock);
4512 		return (0);
4513 
4514 	case I_GERROPT:
4515 		/*
4516 		 * Get error option and return the value
4517 		 * to spot pointed to by arg
4518 		 */
4519 	    {
4520 		int erropt = 0;
4521 
4522 		erropt |= (stp->sd_flag & STRDERRNONPERSIST) ? RERRNONPERSIST :
4523 			RERRNORM;
4524 		erropt |= (stp->sd_flag & STWRERRNONPERSIST) ? WERRNONPERSIST :
4525 			WERRNORM;
4526 		return (strcopyout(&erropt, (void *)arg, sizeof (int),
4527 		    copyflag));
4528 	    }
4529 
4530 	case I_SETSIG:
4531 		/*
4532 		 * Register the calling proc to receive the SIGPOLL
4533 		 * signal based on the events given in arg.  If
4534 		 * arg is zero, remove the proc from register list.
4535 		 */
4536 	    {
4537 		strsig_t *ssp, *pssp;
4538 		struct pid *pidp;
4539 
4540 		pssp = NULL;
4541 		pidp = curproc->p_pidp;
4542 		/*
4543 		 * Hold sd_lock to prevent traversal of sd_siglist while
4544 		 * it is modified.
4545 		 */
4546 		mutex_enter(&stp->sd_lock);
4547 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pidp != pidp);
4548 			pssp = ssp, ssp = ssp->ss_next)
4549 			;
4550 
4551 		if (arg) {
4552 			if (arg & ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4553 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4554 				mutex_exit(&stp->sd_lock);
4555 				return (EINVAL);
4556 			}
4557 			if ((arg & S_BANDURG) && !(arg & S_RDBAND)) {
4558 				mutex_exit(&stp->sd_lock);
4559 				return (EINVAL);
4560 			}
4561 
4562 			/*
4563 			 * If proc not already registered, add it
4564 			 * to list.
4565 			 */
4566 			if (!ssp) {
4567 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4568 				ssp->ss_pidp = pidp;
4569 				ssp->ss_pid = pidp->pid_id;
4570 				ssp->ss_next = NULL;
4571 				if (pssp)
4572 					pssp->ss_next = ssp;
4573 				else
4574 					stp->sd_siglist = ssp;
4575 				mutex_enter(&pidlock);
4576 				PID_HOLD(pidp);
4577 				mutex_exit(&pidlock);
4578 			}
4579 
4580 			/*
4581 			 * Set events.
4582 			 */
4583 			ssp->ss_events = (int)arg;
4584 		} else {
4585 			/*
4586 			 * Remove proc from register list.
4587 			 */
4588 			if (ssp) {
4589 				mutex_enter(&pidlock);
4590 				PID_RELE(pidp);
4591 				mutex_exit(&pidlock);
4592 				if (pssp)
4593 					pssp->ss_next = ssp->ss_next;
4594 				else
4595 					stp->sd_siglist = ssp->ss_next;
4596 				kmem_free(ssp, sizeof (strsig_t));
4597 			} else {
4598 				mutex_exit(&stp->sd_lock);
4599 				return (EINVAL);
4600 			}
4601 		}
4602 
4603 		/*
4604 		 * Recalculate OR of sig events.
4605 		 */
4606 		stp->sd_sigflags = 0;
4607 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4608 			stp->sd_sigflags |= ssp->ss_events;
4609 		mutex_exit(&stp->sd_lock);
4610 		return (0);
4611 	    }
4612 
4613 	case I_GETSIG:
4614 		/*
4615 		 * Return (in arg) the current registration of events
4616 		 * for which the calling proc is to be signaled.
4617 		 */
4618 	    {
4619 		struct strsig *ssp;
4620 		struct pid  *pidp;
4621 
4622 		pidp = curproc->p_pidp;
4623 		mutex_enter(&stp->sd_lock);
4624 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4625 			if (ssp->ss_pidp == pidp) {
4626 				error = strcopyout(&ssp->ss_events, (void *)arg,
4627 				    sizeof (int), copyflag);
4628 				mutex_exit(&stp->sd_lock);
4629 				return (error);
4630 			}
4631 		mutex_exit(&stp->sd_lock);
4632 		return (EINVAL);
4633 	    }
4634 
4635 	case I_ESETSIG:
4636 		/*
4637 		 * Register the ss_pid to receive the SIGPOLL
4638 		 * signal based on the events is ss_events arg.  If
4639 		 * ss_events is zero, remove the proc from register list.
4640 		 */
4641 	{
4642 		struct strsig *ssp, *pssp;
4643 		struct proc *proc;
4644 		struct pid  *pidp;
4645 		pid_t pid;
4646 		struct strsigset ss;
4647 
4648 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4649 		if (error)
4650 			return (error);
4651 
4652 		pid = ss.ss_pid;
4653 
4654 		if (ss.ss_events != 0) {
4655 			/*
4656 			 * Permissions check by sending signal 0.
4657 			 * Note that when kill fails it does a set_errno
4658 			 * causing the system call to fail.
4659 			 */
4660 			error = kill(pid, 0);
4661 			if (error) {
4662 				return (error);
4663 			}
4664 		}
4665 		mutex_enter(&pidlock);
4666 		if (pid == 0)
4667 			proc = curproc;
4668 		else if (pid < 0)
4669 			proc = pgfind(-pid);
4670 		else
4671 			proc = prfind(pid);
4672 		if (proc == NULL) {
4673 			mutex_exit(&pidlock);
4674 			return (ESRCH);
4675 		}
4676 		if (pid < 0)
4677 			pidp = proc->p_pgidp;
4678 		else
4679 			pidp = proc->p_pidp;
4680 		ASSERT(pidp);
4681 		/*
4682 		 * Get a hold on the pid structure while referencing it.
4683 		 * There is a separate PID_HOLD should it be inserted
4684 		 * in the list below.
4685 		 */
4686 		PID_HOLD(pidp);
4687 		mutex_exit(&pidlock);
4688 
4689 		pssp = NULL;
4690 		/*
4691 		 * Hold sd_lock to prevent traversal of sd_siglist while
4692 		 * it is modified.
4693 		 */
4694 		mutex_enter(&stp->sd_lock);
4695 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pid != pid);
4696 				pssp = ssp, ssp = ssp->ss_next)
4697 			;
4698 
4699 		if (ss.ss_events) {
4700 			if (ss.ss_events &
4701 			    ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4702 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4703 				mutex_exit(&stp->sd_lock);
4704 				mutex_enter(&pidlock);
4705 				PID_RELE(pidp);
4706 				mutex_exit(&pidlock);
4707 				return (EINVAL);
4708 			}
4709 			if ((ss.ss_events & S_BANDURG) &&
4710 			    !(ss.ss_events & S_RDBAND)) {
4711 				mutex_exit(&stp->sd_lock);
4712 				mutex_enter(&pidlock);
4713 				PID_RELE(pidp);
4714 				mutex_exit(&pidlock);
4715 				return (EINVAL);
4716 			}
4717 
4718 			/*
4719 			 * If proc not already registered, add it
4720 			 * to list.
4721 			 */
4722 			if (!ssp) {
4723 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4724 				ssp->ss_pidp = pidp;
4725 				ssp->ss_pid = pid;
4726 				ssp->ss_next = NULL;
4727 				if (pssp)
4728 					pssp->ss_next = ssp;
4729 				else
4730 					stp->sd_siglist = ssp;
4731 				mutex_enter(&pidlock);
4732 				PID_HOLD(pidp);
4733 				mutex_exit(&pidlock);
4734 			}
4735 
4736 			/*
4737 			 * Set events.
4738 			 */
4739 			ssp->ss_events = ss.ss_events;
4740 		} else {
4741 			/*
4742 			 * Remove proc from register list.
4743 			 */
4744 			if (ssp) {
4745 				mutex_enter(&pidlock);
4746 				PID_RELE(pidp);
4747 				mutex_exit(&pidlock);
4748 				if (pssp)
4749 					pssp->ss_next = ssp->ss_next;
4750 				else
4751 					stp->sd_siglist = ssp->ss_next;
4752 				kmem_free(ssp, sizeof (strsig_t));
4753 			} else {
4754 				mutex_exit(&stp->sd_lock);
4755 				mutex_enter(&pidlock);
4756 				PID_RELE(pidp);
4757 				mutex_exit(&pidlock);
4758 				return (EINVAL);
4759 			}
4760 		}
4761 
4762 		/*
4763 		 * Recalculate OR of sig events.
4764 		 */
4765 		stp->sd_sigflags = 0;
4766 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4767 			stp->sd_sigflags |= ssp->ss_events;
4768 		mutex_exit(&stp->sd_lock);
4769 		mutex_enter(&pidlock);
4770 		PID_RELE(pidp);
4771 		mutex_exit(&pidlock);
4772 		return (0);
4773 	    }
4774 
4775 	case I_EGETSIG:
4776 		/*
4777 		 * Return (in arg) the current registration of events
4778 		 * for which the calling proc is to be signaled.
4779 		 */
4780 	    {
4781 		struct strsig *ssp;
4782 		struct proc *proc;
4783 		pid_t pid;
4784 		struct pid  *pidp;
4785 		struct strsigset ss;
4786 
4787 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4788 		if (error)
4789 			return (error);
4790 
4791 		pid = ss.ss_pid;
4792 		mutex_enter(&pidlock);
4793 		if (pid == 0)
4794 			proc = curproc;
4795 		else if (pid < 0)
4796 			proc = pgfind(-pid);
4797 		else
4798 			proc = prfind(pid);
4799 		if (proc == NULL) {
4800 			mutex_exit(&pidlock);
4801 			return (ESRCH);
4802 		}
4803 		if (pid < 0)
4804 			pidp = proc->p_pgidp;
4805 		else
4806 			pidp = proc->p_pidp;
4807 
4808 		/* Prevent the pidp from being reassigned */
4809 		PID_HOLD(pidp);
4810 		mutex_exit(&pidlock);
4811 
4812 		mutex_enter(&stp->sd_lock);
4813 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4814 			if (ssp->ss_pid == pid) {
4815 				ss.ss_pid = ssp->ss_pid;
4816 				ss.ss_events = ssp->ss_events;
4817 				error = strcopyout(&ss, (void *)arg,
4818 				    sizeof (struct strsigset), copyflag);
4819 				mutex_exit(&stp->sd_lock);
4820 				mutex_enter(&pidlock);
4821 				PID_RELE(pidp);
4822 				mutex_exit(&pidlock);
4823 				return (error);
4824 			}
4825 		mutex_exit(&stp->sd_lock);
4826 		mutex_enter(&pidlock);
4827 		PID_RELE(pidp);
4828 		mutex_exit(&pidlock);
4829 		return (EINVAL);
4830 	    }
4831 
4832 	case I_PEEK:
4833 	    {
4834 		STRUCT_DECL(strpeek, strpeek);
4835 		size_t n;
4836 		mblk_t *fmp, *tmp_mp = NULL;
4837 
4838 		STRUCT_INIT(strpeek, flag);
4839 
4840 		error = strcopyin((void *)arg, STRUCT_BUF(strpeek),
4841 		    STRUCT_SIZE(strpeek), copyflag);
4842 		if (error)
4843 			return (error);
4844 
4845 		mutex_enter(QLOCK(rdq));
4846 		/*
4847 		 * Skip the invalid messages
4848 		 */
4849 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
4850 			if (mp->b_datap->db_type != M_SIG)
4851 				break;
4852 
4853 		/*
4854 		 * If user has requested to peek at a high priority message
4855 		 * and first message is not, return 0
4856 		 */
4857 		if (mp != NULL) {
4858 			if ((STRUCT_FGET(strpeek, flags) & RS_HIPRI) &&
4859 			    queclass(mp) == QNORM) {
4860 				*rvalp = 0;
4861 				mutex_exit(QLOCK(rdq));
4862 				return (0);
4863 			}
4864 		} else if (stp->sd_struiordq == NULL ||
4865 		    (STRUCT_FGET(strpeek, flags) & RS_HIPRI)) {
4866 			/*
4867 			 * No mblks to look at at the streamhead and
4868 			 * 1). This isn't a synch stream or
4869 			 * 2). This is a synch stream but caller wants high
4870 			 *	priority messages which is not supported by
4871 			 *	the synch stream. (it only supports QNORM)
4872 			 */
4873 			*rvalp = 0;
4874 			mutex_exit(QLOCK(rdq));
4875 			return (0);
4876 		}
4877 
4878 		fmp = mp;
4879 
4880 		if (mp && mp->b_datap->db_type == M_PASSFP) {
4881 			mutex_exit(QLOCK(rdq));
4882 			return (EBADMSG);
4883 		}
4884 
4885 		ASSERT(mp == NULL || mp->b_datap->db_type == M_PCPROTO ||
4886 		    mp->b_datap->db_type == M_PROTO ||
4887 		    mp->b_datap->db_type == M_DATA);
4888 
4889 		if (mp && mp->b_datap->db_type == M_PCPROTO) {
4890 			STRUCT_FSET(strpeek, flags, RS_HIPRI);
4891 		} else {
4892 			STRUCT_FSET(strpeek, flags, 0);
4893 		}
4894 
4895 
4896 		if (mp && ((tmp_mp = dupmsg(mp)) == NULL)) {
4897 			mutex_exit(QLOCK(rdq));
4898 			return (ENOSR);
4899 		}
4900 		mutex_exit(QLOCK(rdq));
4901 
4902 		/*
4903 		 * set mp = tmp_mp, so that I_PEEK processing can continue.
4904 		 * tmp_mp is used to free the dup'd message.
4905 		 */
4906 		mp = tmp_mp;
4907 
4908 		uio.uio_fmode = 0;
4909 		uio.uio_extflg = UIO_COPY_CACHED;
4910 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
4911 		    UIO_SYSSPACE;
4912 		uio.uio_limit = 0;
4913 		/*
4914 		 * First process PROTO blocks, if any.
4915 		 * If user doesn't want to get ctl info by setting maxlen <= 0,
4916 		 * then set len to -1/0 and skip control blocks part.
4917 		 */
4918 		if (STRUCT_FGET(strpeek, ctlbuf.maxlen) < 0)
4919 			STRUCT_FSET(strpeek, ctlbuf.len, -1);
4920 		else if (STRUCT_FGET(strpeek, ctlbuf.maxlen) == 0)
4921 			STRUCT_FSET(strpeek, ctlbuf.len, 0);
4922 		else {
4923 			int	ctl_part = 0;
4924 
4925 			iov.iov_base = STRUCT_FGETP(strpeek, ctlbuf.buf);
4926 			iov.iov_len = STRUCT_FGET(strpeek, ctlbuf.maxlen);
4927 			uio.uio_iov = &iov;
4928 			uio.uio_resid = iov.iov_len;
4929 			uio.uio_loffset = 0;
4930 			uio.uio_iovcnt = 1;
4931 			while (mp && mp->b_datap->db_type != M_DATA &&
4932 			    uio.uio_resid >= 0) {
4933 				ASSERT(STRUCT_FGET(strpeek, flags) == 0 ?
4934 				    mp->b_datap->db_type == M_PROTO :
4935 				    mp->b_datap->db_type == M_PCPROTO);
4936 
4937 				if ((n = MIN(uio.uio_resid,
4938 				    mp->b_wptr - mp->b_rptr)) != 0 &&
4939 				    (error = uiomove((char *)mp->b_rptr, n,
4940 				    UIO_READ, &uio)) != 0) {
4941 					freemsg(tmp_mp);
4942 					return (error);
4943 				}
4944 				ctl_part = 1;
4945 				mp = mp->b_cont;
4946 			}
4947 			/* No ctl message */
4948 			if (ctl_part == 0)
4949 				STRUCT_FSET(strpeek, ctlbuf.len, -1);
4950 			else
4951 				STRUCT_FSET(strpeek, ctlbuf.len,
4952 				    STRUCT_FGET(strpeek, ctlbuf.maxlen) -
4953 				    uio.uio_resid);
4954 		}
4955 
4956 		/*
4957 		 * Now process DATA blocks, if any.
4958 		 * If user doesn't want to get data info by setting maxlen <= 0,
4959 		 * then set len to -1/0 and skip data blocks part.
4960 		 */
4961 		if (STRUCT_FGET(strpeek, databuf.maxlen) < 0)
4962 			STRUCT_FSET(strpeek, databuf.len, -1);
4963 		else if (STRUCT_FGET(strpeek, databuf.maxlen) == 0)
4964 			STRUCT_FSET(strpeek, databuf.len, 0);
4965 		else {
4966 			int	data_part = 0;
4967 
4968 			iov.iov_base = STRUCT_FGETP(strpeek, databuf.buf);
4969 			iov.iov_len = STRUCT_FGET(strpeek, databuf.maxlen);
4970 			uio.uio_iov = &iov;
4971 			uio.uio_resid = iov.iov_len;
4972 			uio.uio_loffset = 0;
4973 			uio.uio_iovcnt = 1;
4974 			while (mp && uio.uio_resid) {
4975 				if (mp->b_datap->db_type == M_DATA) {
4976 					if ((n = MIN(uio.uio_resid,
4977 					    mp->b_wptr - mp->b_rptr)) != 0 &&
4978 					    (error = uiomove((char *)mp->b_rptr,
4979 						n, UIO_READ, &uio)) != 0) {
4980 						freemsg(tmp_mp);
4981 						return (error);
4982 					}
4983 					data_part = 1;
4984 				}
4985 				ASSERT(data_part == 0 ||
4986 				    mp->b_datap->db_type == M_DATA);
4987 				mp = mp->b_cont;
4988 			}
4989 			/* No data message */
4990 			if (data_part == 0)
4991 				STRUCT_FSET(strpeek, databuf.len, -1);
4992 			else
4993 				STRUCT_FSET(strpeek, databuf.len,
4994 				    STRUCT_FGET(strpeek, databuf.maxlen) -
4995 				    uio.uio_resid);
4996 		}
4997 		freemsg(tmp_mp);
4998 
4999 		/*
5000 		 * It is a synch stream and user wants to get
5001 		 * data (maxlen > 0).
5002 		 * uio setup is done by the codes that process DATA
5003 		 * blocks above.
5004 		 */
5005 		if ((fmp == NULL) && STRUCT_FGET(strpeek, databuf.maxlen) > 0) {
5006 			infod_t infod;
5007 
5008 			infod.d_cmd = INFOD_COPYOUT;
5009 			infod.d_res = 0;
5010 			infod.d_uiop = &uio;
5011 			error = infonext(rdq, &infod);
5012 			if (error == EINVAL || error == EBUSY)
5013 				error = 0;
5014 			if (error)
5015 				return (error);
5016 			STRUCT_FSET(strpeek, databuf.len, STRUCT_FGET(strpeek,
5017 			    databuf.maxlen) - uio.uio_resid);
5018 			if (STRUCT_FGET(strpeek, databuf.len) == 0) {
5019 				/*
5020 				 * No data found by the infonext().
5021 				 */
5022 				STRUCT_FSET(strpeek, databuf.len, -1);
5023 			}
5024 		}
5025 		error = strcopyout(STRUCT_BUF(strpeek), (void *)arg,
5026 		    STRUCT_SIZE(strpeek), copyflag);
5027 		if (error) {
5028 			return (error);
5029 		}
5030 		/*
5031 		 * If there is no message retrieved, set return code to 0
5032 		 * otherwise, set it to 1.
5033 		 */
5034 		if (STRUCT_FGET(strpeek, ctlbuf.len) == -1 &&
5035 		    STRUCT_FGET(strpeek, databuf.len) == -1)
5036 			*rvalp = 0;
5037 		else
5038 			*rvalp = 1;
5039 		return (0);
5040 	    }
5041 
5042 	case I_FDINSERT:
5043 	    {
5044 		STRUCT_DECL(strfdinsert, strfdinsert);
5045 		struct file *resftp;
5046 		struct stdata *resstp;
5047 		t_uscalar_t	ival;
5048 		ssize_t msgsize;
5049 		struct strbuf mctl;
5050 
5051 		STRUCT_INIT(strfdinsert, flag);
5052 		if (stp->sd_flag & STRHUP)
5053 			return (ENXIO);
5054 		/*
5055 		 * STRDERR, STWRERR and STPLEX tested above.
5056 		 */
5057 		error = strcopyin((void *)arg, STRUCT_BUF(strfdinsert),
5058 		    STRUCT_SIZE(strfdinsert), copyflag);
5059 		if (error)
5060 			return (error);
5061 
5062 		if (STRUCT_FGET(strfdinsert, offset) < 0 ||
5063 		    (STRUCT_FGET(strfdinsert, offset) %
5064 		    sizeof (t_uscalar_t)) != 0)
5065 			return (EINVAL);
5066 		if ((resftp = getf(STRUCT_FGET(strfdinsert, fildes))) != NULL) {
5067 			if ((resstp = resftp->f_vnode->v_stream) == NULL) {
5068 				releasef(STRUCT_FGET(strfdinsert, fildes));
5069 				return (EINVAL);
5070 			}
5071 		} else
5072 			return (EINVAL);
5073 
5074 		mutex_enter(&resstp->sd_lock);
5075 		if (resstp->sd_flag & (STRDERR|STWRERR|STRHUP|STPLEX)) {
5076 			error = strgeterr(resstp,
5077 					STRDERR|STWRERR|STRHUP|STPLEX, 0);
5078 			if (error != 0) {
5079 				mutex_exit(&resstp->sd_lock);
5080 				releasef(STRUCT_FGET(strfdinsert, fildes));
5081 				return (error);
5082 			}
5083 		}
5084 		mutex_exit(&resstp->sd_lock);
5085 
5086 #ifdef	_ILP32
5087 		{
5088 			queue_t	*q;
5089 			queue_t	*mate = NULL;
5090 
5091 			/* get read queue of stream terminus */
5092 			claimstr(resstp->sd_wrq);
5093 			for (q = resstp->sd_wrq->q_next; q->q_next != NULL;
5094 			    q = q->q_next)
5095 				if (!STRMATED(resstp) && STREAM(q) != resstp &&
5096 				    mate == NULL) {
5097 					ASSERT(q->q_qinfo->qi_srvp);
5098 					ASSERT(_OTHERQ(q)->q_qinfo->qi_srvp);
5099 					claimstr(q);
5100 					mate = q;
5101 				}
5102 			q = _RD(q);
5103 			if (mate)
5104 				releasestr(mate);
5105 			releasestr(resstp->sd_wrq);
5106 			ival = (t_uscalar_t)q;
5107 		}
5108 #else
5109 		ival = (t_uscalar_t)getminor(resftp->f_vnode->v_rdev);
5110 #endif	/* _ILP32 */
5111 
5112 		if (STRUCT_FGET(strfdinsert, ctlbuf.len) <
5113 		    STRUCT_FGET(strfdinsert, offset) + sizeof (t_uscalar_t)) {
5114 			releasef(STRUCT_FGET(strfdinsert, fildes));
5115 			return (EINVAL);
5116 		}
5117 
5118 		/*
5119 		 * Check for legal flag value.
5120 		 */
5121 		if (STRUCT_FGET(strfdinsert, flags) & ~RS_HIPRI) {
5122 			releasef(STRUCT_FGET(strfdinsert, fildes));
5123 			return (EINVAL);
5124 		}
5125 
5126 		/* get these values from those cached in the stream head */
5127 		mutex_enter(QLOCK(stp->sd_wrq));
5128 		rmin = stp->sd_qn_minpsz;
5129 		rmax = stp->sd_qn_maxpsz;
5130 		mutex_exit(QLOCK(stp->sd_wrq));
5131 
5132 		/*
5133 		 * Make sure ctl and data sizes together fall within
5134 		 * the limits of the max and min receive packet sizes
5135 		 * and do not exceed system limit.  A negative data
5136 		 * length means that no data part is to be sent.
5137 		 */
5138 		ASSERT((rmax >= 0) || (rmax == INFPSZ));
5139 		if (rmax == 0) {
5140 			releasef(STRUCT_FGET(strfdinsert, fildes));
5141 			return (ERANGE);
5142 		}
5143 		if ((msgsize = STRUCT_FGET(strfdinsert, databuf.len)) < 0)
5144 			msgsize = 0;
5145 		if ((msgsize < rmin) ||
5146 		    ((msgsize > rmax) && (rmax != INFPSZ)) ||
5147 		    (STRUCT_FGET(strfdinsert, ctlbuf.len) > strctlsz)) {
5148 			releasef(STRUCT_FGET(strfdinsert, fildes));
5149 			return (ERANGE);
5150 		}
5151 
5152 		mutex_enter(&stp->sd_lock);
5153 		while (!(STRUCT_FGET(strfdinsert, flags) & RS_HIPRI) &&
5154 		    !canputnext(stp->sd_wrq)) {
5155 			if ((error = strwaitq(stp, WRITEWAIT, (ssize_t)0,
5156 			    flag, -1, &done)) != 0 || done) {
5157 				mutex_exit(&stp->sd_lock);
5158 				releasef(STRUCT_FGET(strfdinsert, fildes));
5159 				return (error);
5160 			}
5161 			if ((error = i_straccess(stp, access)) != 0) {
5162 				mutex_exit(&stp->sd_lock);
5163 				releasef(
5164 				    STRUCT_FGET(strfdinsert, fildes));
5165 				return (error);
5166 			}
5167 		}
5168 		mutex_exit(&stp->sd_lock);
5169 
5170 		/*
5171 		 * Copy strfdinsert.ctlbuf into native form of
5172 		 * ctlbuf to pass down into strmakemsg().
5173 		 */
5174 		mctl.maxlen = STRUCT_FGET(strfdinsert, ctlbuf.maxlen);
5175 		mctl.len = STRUCT_FGET(strfdinsert, ctlbuf.len);
5176 		mctl.buf = STRUCT_FGETP(strfdinsert, ctlbuf.buf);
5177 
5178 		iov.iov_base = STRUCT_FGETP(strfdinsert, databuf.buf);
5179 		iov.iov_len = STRUCT_FGET(strfdinsert, databuf.len);
5180 		uio.uio_iov = &iov;
5181 		uio.uio_iovcnt = 1;
5182 		uio.uio_loffset = 0;
5183 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5184 		    UIO_SYSSPACE;
5185 		uio.uio_fmode = 0;
5186 		uio.uio_extflg = UIO_COPY_CACHED;
5187 		uio.uio_resid = iov.iov_len;
5188 		if ((error = strmakemsg(&mctl,
5189 		    &msgsize, &uio, stp,
5190 		    STRUCT_FGET(strfdinsert, flags), &mp)) != 0 || !mp) {
5191 			STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5192 			releasef(STRUCT_FGET(strfdinsert, fildes));
5193 			return (error);
5194 		}
5195 
5196 		STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5197 
5198 		/*
5199 		 * Place the possibly reencoded queue pointer 'offset' bytes
5200 		 * from the start of the control portion of the message.
5201 		 */
5202 		*((t_uscalar_t *)(mp->b_rptr +
5203 		    STRUCT_FGET(strfdinsert, offset))) = ival;
5204 
5205 		/*
5206 		 * Put message downstream.
5207 		 */
5208 		stream_willservice(stp);
5209 		putnext(stp->sd_wrq, mp);
5210 		stream_runservice(stp);
5211 		releasef(STRUCT_FGET(strfdinsert, fildes));
5212 		return (error);
5213 	    }
5214 
5215 	case I_SENDFD:
5216 	    {
5217 		struct file *fp;
5218 
5219 		if ((fp = getf((int)arg)) == NULL)
5220 			return (EBADF);
5221 		error = do_sendfp(stp, fp, crp);
5222 #ifdef C2_AUDIT
5223 		if (audit_active) {
5224 			audit_fdsend((int)arg, fp, error);
5225 		}
5226 #endif
5227 		releasef((int)arg);
5228 		return (error);
5229 	    }
5230 
5231 	case I_RECVFD:
5232 	case I_E_RECVFD:
5233 	    {
5234 		struct k_strrecvfd *srf;
5235 		int i, fd;
5236 
5237 		mutex_enter(&stp->sd_lock);
5238 		while (!(mp = getq(rdq))) {
5239 			if (stp->sd_flag & (STRHUP|STREOF)) {
5240 				mutex_exit(&stp->sd_lock);
5241 				return (ENXIO);
5242 			}
5243 			if ((error = strwaitq(stp, GETWAIT, (ssize_t)0,
5244 			    flag, -1, &done)) != 0 || done) {
5245 				mutex_exit(&stp->sd_lock);
5246 				return (error);
5247 			}
5248 			if ((error = i_straccess(stp, access)) != 0) {
5249 				mutex_exit(&stp->sd_lock);
5250 				return (error);
5251 			}
5252 		}
5253 		if (mp->b_datap->db_type != M_PASSFP) {
5254 			putback(stp, rdq, mp, mp->b_band);
5255 			mutex_exit(&stp->sd_lock);
5256 			return (EBADMSG);
5257 		}
5258 		mutex_exit(&stp->sd_lock);
5259 
5260 		srf = (struct k_strrecvfd *)mp->b_rptr;
5261 		if ((fd = ufalloc(0)) == -1) {
5262 			mutex_enter(&stp->sd_lock);
5263 			putback(stp, rdq, mp, mp->b_band);
5264 			mutex_exit(&stp->sd_lock);
5265 			return (EMFILE);
5266 		}
5267 		if (cmd == I_RECVFD) {
5268 			struct o_strrecvfd	ostrfd;
5269 
5270 			/* check to see if uid/gid values are too large. */
5271 
5272 			if (srf->uid > (o_uid_t)USHRT_MAX ||
5273 			    srf->gid > (o_gid_t)USHRT_MAX) {
5274 				mutex_enter(&stp->sd_lock);
5275 				putback(stp, rdq, mp, mp->b_band);
5276 				mutex_exit(&stp->sd_lock);
5277 				setf(fd, NULL);	/* release fd entry */
5278 				return (EOVERFLOW);
5279 			}
5280 
5281 			ostrfd.fd = fd;
5282 			ostrfd.uid = (o_uid_t)srf->uid;
5283 			ostrfd.gid = (o_gid_t)srf->gid;
5284 
5285 			/* Null the filler bits */
5286 			for (i = 0; i < 8; i++)
5287 				ostrfd.fill[i] = 0;
5288 
5289 			error = strcopyout(&ostrfd, (void *)arg,
5290 			    sizeof (struct o_strrecvfd), copyflag);
5291 		} else {		/* I_E_RECVFD */
5292 			struct strrecvfd	strfd;
5293 
5294 			strfd.fd = fd;
5295 			strfd.uid = srf->uid;
5296 			strfd.gid = srf->gid;
5297 
5298 			/* null the filler bits */
5299 			for (i = 0; i < 8; i++)
5300 				strfd.fill[i] = 0;
5301 
5302 			error = strcopyout(&strfd, (void *)arg,
5303 			    sizeof (struct strrecvfd), copyflag);
5304 		}
5305 
5306 		if (error) {
5307 			setf(fd, NULL);	/* release fd entry */
5308 			mutex_enter(&stp->sd_lock);
5309 			putback(stp, rdq, mp, mp->b_band);
5310 			mutex_exit(&stp->sd_lock);
5311 			return (error);
5312 		}
5313 #ifdef C2_AUDIT
5314 		if (audit_active) {
5315 			audit_fdrecv(fd, srf->fp);
5316 		}
5317 #endif
5318 
5319 		/*
5320 		 * Always increment f_count since the freemsg() below will
5321 		 * always call free_passfp() which performs a closef().
5322 		 */
5323 		mutex_enter(&srf->fp->f_tlock);
5324 		srf->fp->f_count++;
5325 		mutex_exit(&srf->fp->f_tlock);
5326 		setf(fd, srf->fp);
5327 		freemsg(mp);
5328 		return (0);
5329 	    }
5330 
5331 	case I_SWROPT:
5332 		/*
5333 		 * Set/clear the write options. arg is a bit
5334 		 * mask with any of the following bits set...
5335 		 * 	SNDZERO - send zero length message
5336 		 *	SNDPIPE - send sigpipe to process if
5337 		 *		sd_werror is set and process is
5338 		 *		doing a write or putmsg.
5339 		 * The new stream head write options should reflect
5340 		 * what is in arg.
5341 		 */
5342 		if (arg & ~(SNDZERO|SNDPIPE))
5343 			return (EINVAL);
5344 
5345 		mutex_enter(&stp->sd_lock);
5346 		stp->sd_wput_opt &= ~(SW_SIGPIPE|SW_SNDZERO);
5347 		if (arg & SNDZERO)
5348 			stp->sd_wput_opt |= SW_SNDZERO;
5349 		if (arg & SNDPIPE)
5350 			stp->sd_wput_opt |= SW_SIGPIPE;
5351 		mutex_exit(&stp->sd_lock);
5352 		return (0);
5353 
5354 	case I_GWROPT:
5355 	    {
5356 		int wropt = 0;
5357 
5358 		if (stp->sd_wput_opt & SW_SNDZERO)
5359 			wropt |= SNDZERO;
5360 		if (stp->sd_wput_opt & SW_SIGPIPE)
5361 			wropt |= SNDPIPE;
5362 		return (strcopyout(&wropt, (void *)arg, sizeof (wropt),
5363 		    copyflag));
5364 	    }
5365 
5366 	case I_LIST:
5367 		/*
5368 		 * Returns all the modules found on this stream,
5369 		 * upto the driver. If argument is NULL, return the
5370 		 * number of modules (including driver). If argument
5371 		 * is not NULL, copy the names into the structure
5372 		 * provided.
5373 		 */
5374 
5375 	    {
5376 		queue_t *q;
5377 		int num_modules, space_allocated;
5378 		STRUCT_DECL(str_list, strlist);
5379 		struct str_mlist *mlist_ptr;
5380 
5381 		if (arg == NULL) { /* Return number of modules plus driver */
5382 			q = stp->sd_wrq;
5383 			if (stp->sd_vnode->v_type == VFIFO) {
5384 				*rvalp = stp->sd_pushcnt;
5385 			} else {
5386 				*rvalp = stp->sd_pushcnt + 1;
5387 			}
5388 		} else {
5389 			STRUCT_INIT(strlist, flag);
5390 
5391 			error = strcopyin((void *)arg, STRUCT_BUF(strlist),
5392 			    STRUCT_SIZE(strlist), copyflag);
5393 			if (error)
5394 				return (error);
5395 
5396 			space_allocated = STRUCT_FGET(strlist, sl_nmods);
5397 			if ((space_allocated) <= 0)
5398 				return (EINVAL);
5399 			claimstr(stp->sd_wrq);
5400 			q = stp->sd_wrq;
5401 			num_modules = 0;
5402 			while (_SAMESTR(q) && (space_allocated != 0)) {
5403 				char *name =
5404 				    q->q_next->q_qinfo->qi_minfo->mi_idname;
5405 
5406 				mlist_ptr = STRUCT_FGETP(strlist, sl_modlist);
5407 
5408 				error = strcopyout(name, mlist_ptr,
5409 				    strlen(name) + 1, copyflag);
5410 
5411 				if (error) {
5412 					releasestr(stp->sd_wrq);
5413 					return (error);
5414 				}
5415 				q = q->q_next;
5416 				space_allocated--;
5417 				num_modules++;
5418 				mlist_ptr =
5419 				    (struct str_mlist *)((uintptr_t)mlist_ptr +
5420 				    sizeof (struct str_mlist));
5421 				STRUCT_FSETP(strlist, sl_modlist, mlist_ptr);
5422 			}
5423 			releasestr(stp->sd_wrq);
5424 			error = strcopyout(&num_modules, (void *)arg,
5425 			    sizeof (int), copyflag);
5426 		}
5427 		return (error);
5428 	    }
5429 
5430 	case I_CKBAND:
5431 	    {
5432 		queue_t *q;
5433 		qband_t *qbp;
5434 
5435 		if ((arg < 0) || (arg >= NBAND))
5436 			return (EINVAL);
5437 		q = _RD(stp->sd_wrq);
5438 		mutex_enter(QLOCK(q));
5439 		if (arg > (int)q->q_nband) {
5440 			*rvalp = 0;
5441 		} else {
5442 			if (arg == 0) {
5443 				if (q->q_first)
5444 					*rvalp = 1;
5445 				else
5446 					*rvalp = 0;
5447 			} else {
5448 				qbp = q->q_bandp;
5449 				while (--arg > 0)
5450 					qbp = qbp->qb_next;
5451 				if (qbp->qb_first)
5452 					*rvalp = 1;
5453 				else
5454 					*rvalp = 0;
5455 			}
5456 		}
5457 		mutex_exit(QLOCK(q));
5458 		return (0);
5459 	    }
5460 
5461 	case I_GETBAND:
5462 	    {
5463 		int intpri;
5464 		queue_t *q;
5465 
5466 		q = _RD(stp->sd_wrq);
5467 		mutex_enter(QLOCK(q));
5468 		mp = q->q_first;
5469 		if (!mp) {
5470 			mutex_exit(QLOCK(q));
5471 			return (ENODATA);
5472 		}
5473 		intpri = (int)mp->b_band;
5474 		error = strcopyout(&intpri, (void *)arg, sizeof (int),
5475 		    copyflag);
5476 		mutex_exit(QLOCK(q));
5477 		return (error);
5478 	    }
5479 
5480 	case I_ATMARK:
5481 	    {
5482 		queue_t *q;
5483 
5484 		if (arg & ~(ANYMARK|LASTMARK))
5485 			return (EINVAL);
5486 		q = _RD(stp->sd_wrq);
5487 		mutex_enter(&stp->sd_lock);
5488 		if ((stp->sd_flag & STRATMARK) && (arg == ANYMARK)) {
5489 			*rvalp = 1;
5490 		} else {
5491 			mutex_enter(QLOCK(q));
5492 			mp = q->q_first;
5493 
5494 			if (mp == NULL)
5495 				*rvalp = 0;
5496 			else if ((arg == ANYMARK) && (mp->b_flag & MSGMARK))
5497 				*rvalp = 1;
5498 			else if ((arg == LASTMARK) && (mp == stp->sd_mark))
5499 				*rvalp = 1;
5500 			else
5501 				*rvalp = 0;
5502 			mutex_exit(QLOCK(q));
5503 		}
5504 		mutex_exit(&stp->sd_lock);
5505 		return (0);
5506 	    }
5507 
5508 	case I_CANPUT:
5509 	    {
5510 		char band;
5511 
5512 		if ((arg < 0) || (arg >= NBAND))
5513 			return (EINVAL);
5514 		band = (char)arg;
5515 		*rvalp = bcanputnext(stp->sd_wrq, band);
5516 		return (0);
5517 	    }
5518 
5519 	case I_SETCLTIME:
5520 	    {
5521 		int closetime;
5522 
5523 		error = strcopyin((void *)arg, &closetime, sizeof (int),
5524 		    copyflag);
5525 		if (error)
5526 			return (error);
5527 		if (closetime < 0)
5528 			return (EINVAL);
5529 
5530 		stp->sd_closetime = closetime;
5531 		return (0);
5532 	    }
5533 
5534 	case I_GETCLTIME:
5535 	    {
5536 		int closetime;
5537 
5538 		closetime = stp->sd_closetime;
5539 		return (strcopyout(&closetime, (void *)arg, sizeof (int),
5540 		    copyflag));
5541 	    }
5542 
5543 	case TIOCGSID:
5544 	{
5545 		pid_t sid;
5546 
5547 		mutex_enter(&stp->sd_lock);
5548 		if (stp->sd_sidp == NULL) {
5549 			mutex_exit(&stp->sd_lock);
5550 			return (ENOTTY);
5551 		}
5552 		sid = stp->sd_sidp->pid_id;
5553 		mutex_exit(&stp->sd_lock);
5554 		return (strcopyout(&sid, (void *)arg, sizeof (pid_t),
5555 		    copyflag));
5556 	}
5557 
5558 	case TIOCSPGRP:
5559 	{
5560 		pid_t pgrp;
5561 		proc_t *q;
5562 		pid_t	sid, fg_pgid, bg_pgid;
5563 
5564 		if (error = strcopyin((void *)arg, &pgrp, sizeof (pid_t),
5565 		    copyflag))
5566 			return (error);
5567 		mutex_enter(&stp->sd_lock);
5568 		mutex_enter(&pidlock);
5569 		if (stp->sd_sidp != ttoproc(curthread)->p_sessp->s_sidp) {
5570 			mutex_exit(&pidlock);
5571 			mutex_exit(&stp->sd_lock);
5572 			return (ENOTTY);
5573 		}
5574 		if (pgrp == stp->sd_pgidp->pid_id) {
5575 			mutex_exit(&pidlock);
5576 			mutex_exit(&stp->sd_lock);
5577 			return (0);
5578 		}
5579 		if (pgrp <= 0 || pgrp >= maxpid) {
5580 			mutex_exit(&pidlock);
5581 			mutex_exit(&stp->sd_lock);
5582 			return (EINVAL);
5583 		}
5584 		if ((q = pgfind(pgrp)) == NULL ||
5585 		    q->p_sessp != ttoproc(curthread)->p_sessp) {
5586 			mutex_exit(&pidlock);
5587 			mutex_exit(&stp->sd_lock);
5588 			return (EPERM);
5589 		}
5590 		sid = stp->sd_sidp->pid_id;
5591 		fg_pgid = q->p_pgrp;
5592 		bg_pgid = stp->sd_pgidp->pid_id;
5593 		CL_SET_PROCESS_GROUP(curthread, sid, bg_pgid, fg_pgid);
5594 		PID_RELE(stp->sd_pgidp);
5595 		ctty_clear_sighuped();
5596 		stp->sd_pgidp = q->p_pgidp;
5597 		PID_HOLD(stp->sd_pgidp);
5598 		mutex_exit(&pidlock);
5599 		mutex_exit(&stp->sd_lock);
5600 		return (0);
5601 	}
5602 
5603 	case TIOCGPGRP:
5604 	{
5605 		pid_t pgrp;
5606 
5607 		mutex_enter(&stp->sd_lock);
5608 		if (stp->sd_sidp == NULL) {
5609 			mutex_exit(&stp->sd_lock);
5610 			return (ENOTTY);
5611 		}
5612 		pgrp = stp->sd_pgidp->pid_id;
5613 		mutex_exit(&stp->sd_lock);
5614 		return (strcopyout(&pgrp, (void *)arg, sizeof (pid_t),
5615 		    copyflag));
5616 	}
5617 
5618 	case TIOCSCTTY:
5619 	{
5620 		return (strctty(stp));
5621 	}
5622 
5623 	case TIOCNOTTY:
5624 	{
5625 		/* freectty() always assumes curproc. */
5626 		if (freectty(B_FALSE) != 0)
5627 			return (0);
5628 		return (ENOTTY);
5629 	}
5630 
5631 	case FIONBIO:
5632 	case FIOASYNC:
5633 		return (0);	/* handled by the upper layer */
5634 	}
5635 }
5636 
5637 /*
5638  * Custom free routine used for M_PASSFP messages.
5639  */
5640 static void
5641 free_passfp(struct k_strrecvfd *srf)
5642 {
5643 	(void) closef(srf->fp);
5644 	kmem_free(srf, sizeof (struct k_strrecvfd) + sizeof (frtn_t));
5645 }
5646 
5647 /* ARGSUSED */
5648 int
5649 do_sendfp(struct stdata *stp, struct file *fp, struct cred *cr)
5650 {
5651 	queue_t *qp, *nextqp;
5652 	struct k_strrecvfd *srf;
5653 	mblk_t *mp;
5654 	frtn_t *frtnp;
5655 	size_t bufsize;
5656 	queue_t	*mate = NULL;
5657 	syncq_t	*sq = NULL;
5658 	int retval = 0;
5659 
5660 	if (stp->sd_flag & STRHUP)
5661 		return (ENXIO);
5662 
5663 	claimstr(stp->sd_wrq);
5664 
5665 	/* Fastpath, we have a pipe, and we are already mated, use it. */
5666 	if (STRMATED(stp)) {
5667 		qp = _RD(stp->sd_mate->sd_wrq);
5668 		claimstr(qp);
5669 		mate = qp;
5670 	} else { /* Not already mated. */
5671 
5672 		/*
5673 		 * Walk the stream to the end of this one.
5674 		 * assumes that the claimstr() will prevent
5675 		 * plumbing between the stream head and the
5676 		 * driver from changing
5677 		 */
5678 		qp = stp->sd_wrq;
5679 
5680 		/*
5681 		 * Loop until we reach the end of this stream.
5682 		 * On completion, qp points to the write queue
5683 		 * at the end of the stream, or the read queue
5684 		 * at the stream head if this is a fifo.
5685 		 */
5686 		while (((qp = qp->q_next) != NULL) && _SAMESTR(qp))
5687 			;
5688 
5689 		/*
5690 		 * Just in case we get a q_next which is NULL, but
5691 		 * not at the end of the stream.  This is actually
5692 		 * broken, so we set an assert to catch it in
5693 		 * debug, and set an error and return if not debug.
5694 		 */
5695 		ASSERT(qp);
5696 		if (qp == NULL) {
5697 			releasestr(stp->sd_wrq);
5698 			return (EINVAL);
5699 		}
5700 
5701 		/*
5702 		 * Enter the syncq for the driver, so (hopefully)
5703 		 * the queue values will not change on us.
5704 		 * XXXX - This will only prevent the race IFF only
5705 		 *   the write side modifies the q_next member, and
5706 		 *   the put procedure is protected by at least
5707 		 *   MT_PERQ.
5708 		 */
5709 		if ((sq = qp->q_syncq) != NULL)
5710 			entersq(sq, SQ_PUT);
5711 
5712 		/* Now get the q_next value from this qp. */
5713 		nextqp = qp->q_next;
5714 
5715 		/*
5716 		 * If nextqp exists and the other stream is different
5717 		 * from this one claim the stream, set the mate, and
5718 		 * get the read queue at the stream head of the other
5719 		 * stream.  Assumes that nextqp was at least valid when
5720 		 * we got it.  Hopefully the entersq of the driver
5721 		 * will prevent it from changing on us.
5722 		 */
5723 		if ((nextqp != NULL) && (STREAM(nextqp) != stp)) {
5724 			ASSERT(qp->q_qinfo->qi_srvp);
5725 			ASSERT(_OTHERQ(qp)->q_qinfo->qi_srvp);
5726 			ASSERT(_OTHERQ(qp->q_next)->q_qinfo->qi_srvp);
5727 			claimstr(nextqp);
5728 
5729 			/* Make sure we still have a q_next */
5730 			if (nextqp != qp->q_next) {
5731 				releasestr(stp->sd_wrq);
5732 				releasestr(nextqp);
5733 				return (EINVAL);
5734 			}
5735 
5736 			qp = _RD(STREAM(nextqp)->sd_wrq);
5737 			mate = qp;
5738 		}
5739 		/* If we entered the synq above, leave it. */
5740 		if (sq != NULL)
5741 			leavesq(sq, SQ_PUT);
5742 	} /*  STRMATED(STP)  */
5743 
5744 	/* XXX prevents substitution of the ops vector */
5745 	if (qp->q_qinfo != &strdata && qp->q_qinfo != &fifo_strdata) {
5746 		retval = EINVAL;
5747 		goto out;
5748 	}
5749 
5750 	if (qp->q_flag & QFULL) {
5751 		retval = EAGAIN;
5752 		goto out;
5753 	}
5754 
5755 	/*
5756 	 * Since M_PASSFP messages include a file descriptor, we use
5757 	 * esballoc() and specify a custom free routine (free_passfp()) that
5758 	 * will close the descriptor as part of freeing the message.  For
5759 	 * convenience, we stash the frtn_t right after the data block.
5760 	 */
5761 	bufsize = sizeof (struct k_strrecvfd) + sizeof (frtn_t);
5762 	srf = kmem_alloc(bufsize, KM_NOSLEEP);
5763 	if (srf == NULL) {
5764 		retval = EAGAIN;
5765 		goto out;
5766 	}
5767 
5768 	frtnp = (frtn_t *)(srf + 1);
5769 	frtnp->free_arg = (caddr_t)srf;
5770 	frtnp->free_func = free_passfp;
5771 
5772 	mp = esballoc((uchar_t *)srf, bufsize, BPRI_MED, frtnp);
5773 	if (mp == NULL) {
5774 		kmem_free(srf, bufsize);
5775 		retval = EAGAIN;
5776 		goto out;
5777 	}
5778 	mp->b_wptr += sizeof (struct k_strrecvfd);
5779 	mp->b_datap->db_type = M_PASSFP;
5780 
5781 	srf->fp = fp;
5782 	srf->uid = crgetuid(curthread->t_cred);
5783 	srf->gid = crgetgid(curthread->t_cred);
5784 	mutex_enter(&fp->f_tlock);
5785 	fp->f_count++;
5786 	mutex_exit(&fp->f_tlock);
5787 
5788 	put(qp, mp);
5789 out:
5790 	releasestr(stp->sd_wrq);
5791 	if (mate)
5792 		releasestr(mate);
5793 	return (retval);
5794 }
5795 
5796 /*
5797  * Send an ioctl message downstream and wait for acknowledgement.
5798  * flags may be set to either U_TO_K or K_TO_K and a combination
5799  * of STR_NOERROR or STR_NOSIG
5800  * STR_NOSIG: Signals are essentially ignored or held and have
5801  *	no effect for the duration of the call.
5802  * STR_NOERROR: Ignores stream head read, write and hup errors.
5803  *	Additionally, if an existing ioctl times out, it is assumed
5804  *	lost and and this ioctl will continue as if the previous ioctl had
5805  *	finished.  ETIME may be returned if this ioctl times out (i.e.
5806  *	ic_timout is not INFTIM).  Non-stream head errors may be returned if
5807  *	the ioc_error indicates that the driver/module had problems,
5808  *	an EFAULT was found when accessing user data, a lack of
5809  * 	resources, etc.
5810  */
5811 int
5812 strdoioctl(
5813 	struct stdata *stp,
5814 	struct strioctl *strioc,
5815 	int fflags,		/* file flags with model info */
5816 	int flag,
5817 	cred_t *crp,
5818 	int *rvalp)
5819 {
5820 	mblk_t *bp;
5821 	struct iocblk *iocbp;
5822 	struct copyreq *reqp;
5823 	struct copyresp *resp;
5824 	int id;
5825 	int transparent = 0;
5826 	int error = 0;
5827 	int len = 0;
5828 	caddr_t taddr;
5829 	int copyflag = (flag & (U_TO_K | K_TO_K));
5830 	int sigflag = (flag & STR_NOSIG);
5831 	int errs;
5832 	uint_t waitflags;
5833 
5834 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
5835 	ASSERT((fflags & FMODELS) != 0);
5836 
5837 	TRACE_2(TR_FAC_STREAMS_FR,
5838 		TR_STRDOIOCTL,
5839 		"strdoioctl:stp %p strioc %p", stp, strioc);
5840 	if (strioc->ic_len == TRANSPARENT) {	/* send arg in M_DATA block */
5841 		transparent = 1;
5842 		strioc->ic_len = sizeof (intptr_t);
5843 	}
5844 
5845 	if (strioc->ic_len < 0 || (strmsgsz > 0 && strioc->ic_len > strmsgsz))
5846 		return (EINVAL);
5847 
5848 	if ((bp = allocb_cred_wait(sizeof (union ioctypes), sigflag, &error,
5849 	    crp)) == NULL)
5850 			return (error);
5851 
5852 	bzero(bp->b_wptr, sizeof (union ioctypes));
5853 
5854 	iocbp = (struct iocblk *)bp->b_wptr;
5855 	iocbp->ioc_count = strioc->ic_len;
5856 	iocbp->ioc_cmd = strioc->ic_cmd;
5857 	iocbp->ioc_flag = (fflags & FMODELS);
5858 
5859 	crhold(crp);
5860 	iocbp->ioc_cr = crp;
5861 	DB_TYPE(bp) = M_IOCTL;
5862 	DB_CPID(bp) = curproc->p_pid;
5863 	bp->b_wptr += sizeof (struct iocblk);
5864 
5865 	if (flag & STR_NOERROR)
5866 		errs = STPLEX;
5867 	else
5868 		errs = STRHUP|STRDERR|STWRERR|STPLEX;
5869 
5870 	/*
5871 	 * If there is data to copy into ioctl block, do so.
5872 	 */
5873 	if (iocbp->ioc_count > 0) {
5874 		if (transparent)
5875 			/*
5876 			 * Note: STR_NOERROR does not have an effect
5877 			 * in putiocd()
5878 			 */
5879 			id = K_TO_K | sigflag;
5880 		else
5881 			id = flag;
5882 		if ((error = putiocd(bp, strioc->ic_dp, id, crp)) != 0) {
5883 			freemsg(bp);
5884 			crfree(crp);
5885 			return (error);
5886 		}
5887 
5888 		/*
5889 		 * We could have slept copying in user pages.
5890 		 * Recheck the stream head state (the other end
5891 		 * of a pipe could have gone away).
5892 		 */
5893 		if (stp->sd_flag & errs) {
5894 			mutex_enter(&stp->sd_lock);
5895 			error = strgeterr(stp, errs, 0);
5896 			mutex_exit(&stp->sd_lock);
5897 			if (error != 0) {
5898 				freemsg(bp);
5899 				crfree(crp);
5900 				return (error);
5901 			}
5902 		}
5903 	}
5904 	if (transparent)
5905 		iocbp->ioc_count = TRANSPARENT;
5906 
5907 	/*
5908 	 * Block for up to STRTIMOUT milliseconds if there is an outstanding
5909 	 * ioctl for this stream already running.  All processes
5910 	 * sleeping here will be awakened as a result of an ACK
5911 	 * or NAK being received for the outstanding ioctl, or
5912 	 * as a result of the timer expiring on the outstanding
5913 	 * ioctl (a failure), or as a result of any waiting
5914 	 * process's timer expiring (also a failure).
5915 	 */
5916 
5917 	error = 0;
5918 	mutex_enter(&stp->sd_lock);
5919 	while (stp->sd_flag & (IOCWAIT | IOCWAITNE)) {
5920 		clock_t cv_rval;
5921 
5922 		TRACE_0(TR_FAC_STREAMS_FR,
5923 			TR_STRDOIOCTL_WAIT,
5924 			"strdoioctl sleeps - IOCWAIT");
5925 		cv_rval = str_cv_wait(&stp->sd_iocmonitor, &stp->sd_lock,
5926 		    STRTIMOUT, sigflag);
5927 		if (cv_rval <= 0) {
5928 			if (cv_rval == 0) {
5929 				error = EINTR;
5930 			} else {
5931 				if (flag & STR_NOERROR) {
5932 					/*
5933 					 * Terminating current ioctl in
5934 					 * progress -- assume it got lost and
5935 					 * wake up the other thread so that the
5936 					 * operation completes.
5937 					 */
5938 					if (!(stp->sd_flag & IOCWAITNE)) {
5939 						stp->sd_flag |= IOCWAITNE;
5940 						cv_broadcast(&stp->sd_monitor);
5941 					}
5942 					/*
5943 					 * Otherwise, there's a running
5944 					 * STR_NOERROR -- we have no choice
5945 					 * here but to wait forever (or until
5946 					 * interrupted).
5947 					 */
5948 				} else {
5949 					/*
5950 					 * pending ioctl has caused
5951 					 * us to time out
5952 					 */
5953 					error = ETIME;
5954 				}
5955 			}
5956 		} else if ((stp->sd_flag & errs)) {
5957 			error = strgeterr(stp, errs, 0);
5958 		}
5959 		if (error) {
5960 			mutex_exit(&stp->sd_lock);
5961 			freemsg(bp);
5962 			crfree(crp);
5963 			return (error);
5964 		}
5965 	}
5966 
5967 	/*
5968 	 * Have control of ioctl mechanism.
5969 	 * Send down ioctl packet and wait for response.
5970 	 */
5971 	if (stp->sd_iocblk != (mblk_t *)-1) {
5972 		freemsg(stp->sd_iocblk);
5973 	}
5974 	stp->sd_iocblk = NULL;
5975 
5976 	/*
5977 	 * If this is marked with 'noerror' (internal; mostly
5978 	 * I_{P,}{UN,}LINK), then make sure nobody else is able to get
5979 	 * in here by setting IOCWAITNE.
5980 	 */
5981 	waitflags = IOCWAIT;
5982 	if (flag & STR_NOERROR)
5983 		waitflags |= IOCWAITNE;
5984 
5985 	stp->sd_flag |= waitflags;
5986 
5987 	/*
5988 	 * Assign sequence number.
5989 	 */
5990 	iocbp->ioc_id = stp->sd_iocid = getiocseqno();
5991 
5992 	mutex_exit(&stp->sd_lock);
5993 
5994 	TRACE_1(TR_FAC_STREAMS_FR,
5995 		TR_STRDOIOCTL_PUT, "strdoioctl put: stp %p", stp);
5996 	stream_willservice(stp);
5997 	putnext(stp->sd_wrq, bp);
5998 	stream_runservice(stp);
5999 
6000 	/*
6001 	 * Timed wait for acknowledgment.  The wait time is limited by the
6002 	 * timeout value, which must be a positive integer (number of
6003 	 * milliseconds) to wait, or 0 (use default value of STRTIMOUT
6004 	 * milliseconds), or -1 (wait forever).  This will be awakened
6005 	 * either by an ACK/NAK message arriving, the timer expiring, or
6006 	 * the timer expiring on another ioctl waiting for control of the
6007 	 * mechanism.
6008 	 */
6009 waitioc:
6010 	mutex_enter(&stp->sd_lock);
6011 
6012 
6013 	/*
6014 	 * If the reply has already arrived, don't sleep.  If awakened from
6015 	 * the sleep, fail only if the reply has not arrived by then.
6016 	 * Otherwise, process the reply.
6017 	 */
6018 	while (!stp->sd_iocblk) {
6019 		clock_t cv_rval;
6020 
6021 		if (stp->sd_flag & errs) {
6022 			error = strgeterr(stp, errs, 0);
6023 			if (error != 0) {
6024 				stp->sd_flag &= ~waitflags;
6025 				cv_broadcast(&stp->sd_iocmonitor);
6026 				mutex_exit(&stp->sd_lock);
6027 				crfree(crp);
6028 				return (error);
6029 			}
6030 		}
6031 
6032 		TRACE_0(TR_FAC_STREAMS_FR,
6033 			TR_STRDOIOCTL_WAIT2,
6034 			"strdoioctl sleeps awaiting reply");
6035 		ASSERT(error == 0);
6036 
6037 		cv_rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock,
6038 		    (strioc->ic_timout ?
6039 		    strioc->ic_timout * 1000 : STRTIMOUT), sigflag);
6040 
6041 		/*
6042 		 * There are four possible cases here: interrupt, timeout,
6043 		 * wakeup by IOCWAITNE (above), or wakeup by strrput_nondata (a
6044 		 * valid M_IOCTL reply).
6045 		 *
6046 		 * If we've been awakened by a STR_NOERROR ioctl on some other
6047 		 * thread, then sd_iocblk will still be NULL, and IOCWAITNE
6048 		 * will be set.  Pretend as if we just timed out.  Note that
6049 		 * this other thread waited at least STRTIMOUT before trying to
6050 		 * awaken our thread, so this is indistinguishable (even for
6051 		 * INFTIM) from the case where we failed with ETIME waiting on
6052 		 * IOCWAIT in the prior loop.
6053 		 */
6054 		if (cv_rval > 0 && !(flag & STR_NOERROR) &&
6055 		    stp->sd_iocblk == NULL && (stp->sd_flag & IOCWAITNE)) {
6056 			cv_rval = -1;
6057 		}
6058 
6059 		/*
6060 		 * note: STR_NOERROR does not protect
6061 		 * us here.. use ic_timout < 0
6062 		 */
6063 		if (cv_rval <= 0) {
6064 			if (cv_rval == 0) {
6065 				error = EINTR;
6066 			} else {
6067 				error =  ETIME;
6068 			}
6069 			/*
6070 			 * A message could have come in after we were scheduled
6071 			 * but before we were actually run.
6072 			 */
6073 			bp = stp->sd_iocblk;
6074 			stp->sd_iocblk = NULL;
6075 			if (bp != NULL) {
6076 				if ((bp->b_datap->db_type == M_COPYIN) ||
6077 				    (bp->b_datap->db_type == M_COPYOUT)) {
6078 					mutex_exit(&stp->sd_lock);
6079 					if (bp->b_cont) {
6080 						freemsg(bp->b_cont);
6081 						bp->b_cont = NULL;
6082 					}
6083 					bp->b_datap->db_type = M_IOCDATA;
6084 					bp->b_wptr = bp->b_rptr +
6085 						sizeof (struct copyresp);
6086 					resp = (struct copyresp *)bp->b_rptr;
6087 					resp->cp_rval =
6088 					    (caddr_t)1; /* failure */
6089 					stream_willservice(stp);
6090 					putnext(stp->sd_wrq, bp);
6091 					stream_runservice(stp);
6092 					mutex_enter(&stp->sd_lock);
6093 				} else {
6094 					freemsg(bp);
6095 				}
6096 			}
6097 			stp->sd_flag &= ~waitflags;
6098 			cv_broadcast(&stp->sd_iocmonitor);
6099 			mutex_exit(&stp->sd_lock);
6100 			crfree(crp);
6101 			return (error);
6102 		}
6103 	}
6104 	bp = stp->sd_iocblk;
6105 	/*
6106 	 * Note: it is strictly impossible to get here with sd_iocblk set to
6107 	 * -1.  This is because the initial loop above doesn't allow any new
6108 	 * ioctls into the fray until all others have passed this point.
6109 	 */
6110 	ASSERT(bp != NULL && bp != (mblk_t *)-1);
6111 	TRACE_1(TR_FAC_STREAMS_FR,
6112 		TR_STRDOIOCTL_ACK, "strdoioctl got reply: bp %p", bp);
6113 	if ((bp->b_datap->db_type == M_IOCACK) ||
6114 	    (bp->b_datap->db_type == M_IOCNAK)) {
6115 		/* for detection of duplicate ioctl replies */
6116 		stp->sd_iocblk = (mblk_t *)-1;
6117 		stp->sd_flag &= ~waitflags;
6118 		cv_broadcast(&stp->sd_iocmonitor);
6119 		mutex_exit(&stp->sd_lock);
6120 	} else {
6121 		/*
6122 		 * flags not cleared here because we're still doing
6123 		 * copy in/out for ioctl.
6124 		 */
6125 		stp->sd_iocblk = NULL;
6126 		mutex_exit(&stp->sd_lock);
6127 	}
6128 
6129 
6130 	/*
6131 	 * Have received acknowledgment.
6132 	 */
6133 
6134 	switch (bp->b_datap->db_type) {
6135 	case M_IOCACK:
6136 		/*
6137 		 * Positive ack.
6138 		 */
6139 		iocbp = (struct iocblk *)bp->b_rptr;
6140 
6141 		/*
6142 		 * Set error if indicated.
6143 		 */
6144 		if (iocbp->ioc_error) {
6145 			error = iocbp->ioc_error;
6146 			break;
6147 		}
6148 
6149 		/*
6150 		 * Set return value.
6151 		 */
6152 		*rvalp = iocbp->ioc_rval;
6153 
6154 		/*
6155 		 * Data may have been returned in ACK message (ioc_count > 0).
6156 		 * If so, copy it out to the user's buffer.
6157 		 */
6158 		if (iocbp->ioc_count && !transparent) {
6159 			if (error = getiocd(bp, strioc->ic_dp, copyflag))
6160 				break;
6161 		}
6162 		if (!transparent) {
6163 			if (len)	/* an M_COPYOUT was used with I_STR */
6164 				strioc->ic_len = len;
6165 			else
6166 				strioc->ic_len = (int)iocbp->ioc_count;
6167 		}
6168 		break;
6169 
6170 	case M_IOCNAK:
6171 		/*
6172 		 * Negative ack.
6173 		 *
6174 		 * The only thing to do is set error as specified
6175 		 * in neg ack packet.
6176 		 */
6177 		iocbp = (struct iocblk *)bp->b_rptr;
6178 
6179 		error = (iocbp->ioc_error ? iocbp->ioc_error : EINVAL);
6180 		break;
6181 
6182 	case M_COPYIN:
6183 		/*
6184 		 * Driver or module has requested user ioctl data.
6185 		 */
6186 		reqp = (struct copyreq *)bp->b_rptr;
6187 
6188 		/*
6189 		 * M_COPYIN should *never* have a message attached, though
6190 		 * it's harmless if it does -- thus, panic on a DEBUG
6191 		 * kernel and just free it on a non-DEBUG build.
6192 		 */
6193 		ASSERT(bp->b_cont == NULL);
6194 		if (bp->b_cont != NULL) {
6195 			freemsg(bp->b_cont);
6196 			bp->b_cont = NULL;
6197 		}
6198 
6199 		error = putiocd(bp, reqp->cq_addr, flag, crp);
6200 		if (error && bp->b_cont) {
6201 			freemsg(bp->b_cont);
6202 			bp->b_cont = NULL;
6203 		}
6204 
6205 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6206 		bp->b_datap->db_type = M_IOCDATA;
6207 
6208 		mblk_setcred(bp, crp);
6209 		DB_CPID(bp) = curproc->p_pid;
6210 		resp = (struct copyresp *)bp->b_rptr;
6211 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6212 		resp->cp_flag = (fflags & FMODELS);
6213 
6214 		stream_willservice(stp);
6215 		putnext(stp->sd_wrq, bp);
6216 		stream_runservice(stp);
6217 
6218 		if (error) {
6219 			mutex_enter(&stp->sd_lock);
6220 			stp->sd_flag &= ~waitflags;
6221 			cv_broadcast(&stp->sd_iocmonitor);
6222 			mutex_exit(&stp->sd_lock);
6223 			crfree(crp);
6224 			return (error);
6225 		}
6226 
6227 		goto waitioc;
6228 
6229 	case M_COPYOUT:
6230 		/*
6231 		 * Driver or module has ioctl data for a user.
6232 		 */
6233 		reqp = (struct copyreq *)bp->b_rptr;
6234 		ASSERT(bp->b_cont != NULL);
6235 
6236 		/*
6237 		 * Always (transparent or non-transparent )
6238 		 * use the address specified in the request
6239 		 */
6240 		taddr = reqp->cq_addr;
6241 		if (!transparent)
6242 			len = (int)reqp->cq_size;
6243 
6244 		/* copyout data to the provided address */
6245 		error = getiocd(bp, taddr, copyflag);
6246 
6247 		freemsg(bp->b_cont);
6248 		bp->b_cont = NULL;
6249 
6250 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6251 		bp->b_datap->db_type = M_IOCDATA;
6252 
6253 		mblk_setcred(bp, crp);
6254 		DB_CPID(bp) = curproc->p_pid;
6255 		resp = (struct copyresp *)bp->b_rptr;
6256 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6257 		resp->cp_flag = (fflags & FMODELS);
6258 
6259 		stream_willservice(stp);
6260 		putnext(stp->sd_wrq, bp);
6261 		stream_runservice(stp);
6262 
6263 		if (error) {
6264 			mutex_enter(&stp->sd_lock);
6265 			stp->sd_flag &= ~waitflags;
6266 			cv_broadcast(&stp->sd_iocmonitor);
6267 			mutex_exit(&stp->sd_lock);
6268 			crfree(crp);
6269 			return (error);
6270 		}
6271 		goto waitioc;
6272 
6273 	default:
6274 		ASSERT(0);
6275 		mutex_enter(&stp->sd_lock);
6276 		stp->sd_flag &= ~waitflags;
6277 		cv_broadcast(&stp->sd_iocmonitor);
6278 		mutex_exit(&stp->sd_lock);
6279 		break;
6280 	}
6281 
6282 	freemsg(bp);
6283 	crfree(crp);
6284 	return (error);
6285 }
6286 
6287 /*
6288  * For the SunOS keyboard driver.
6289  * Return the next available "ioctl" sequence number.
6290  * Exported, so that streams modules can send "ioctl" messages
6291  * downstream from their open routine.
6292  */
6293 int
6294 getiocseqno(void)
6295 {
6296 	int	i;
6297 
6298 	mutex_enter(&strresources);
6299 	i = ++ioc_id;
6300 	mutex_exit(&strresources);
6301 	return (i);
6302 }
6303 
6304 /*
6305  * Get the next message from the read queue.  If the message is
6306  * priority, STRPRI will have been set by strrput().  This flag
6307  * should be reset only when the entire message at the front of the
6308  * queue as been consumed.
6309  *
6310  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
6311  */
6312 int
6313 strgetmsg(
6314 	struct vnode *vp,
6315 	struct strbuf *mctl,
6316 	struct strbuf *mdata,
6317 	unsigned char *prip,
6318 	int *flagsp,
6319 	int fmode,
6320 	rval_t *rvp)
6321 {
6322 	struct stdata *stp;
6323 	mblk_t *bp, *nbp;
6324 	mblk_t *savemp = NULL;
6325 	mblk_t *savemptail = NULL;
6326 	uint_t old_sd_flag;
6327 	int flg;
6328 	int more = 0;
6329 	int error = 0;
6330 	char first = 1;
6331 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
6332 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
6333 	unsigned char pri = 0;
6334 	queue_t *q;
6335 	int	pr = 0;			/* Partial read successful */
6336 	struct uio uios;
6337 	struct uio *uiop = &uios;
6338 	struct iovec iovs;
6339 	unsigned char type;
6340 
6341 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRGETMSG_ENTER,
6342 		"strgetmsg:%p", vp);
6343 
6344 	ASSERT(vp->v_stream);
6345 	stp = vp->v_stream;
6346 	rvp->r_val1 = 0;
6347 
6348 	mutex_enter(&stp->sd_lock);
6349 
6350 	if ((error = i_straccess(stp, JCREAD)) != 0) {
6351 		mutex_exit(&stp->sd_lock);
6352 		return (error);
6353 	}
6354 
6355 	if (stp->sd_flag & (STRDERR|STPLEX)) {
6356 		error = strgeterr(stp, STRDERR|STPLEX, 0);
6357 		if (error != 0) {
6358 			mutex_exit(&stp->sd_lock);
6359 			return (error);
6360 		}
6361 	}
6362 	mutex_exit(&stp->sd_lock);
6363 
6364 	switch (*flagsp) {
6365 	case MSG_HIPRI:
6366 		if (*prip != 0)
6367 			return (EINVAL);
6368 		break;
6369 
6370 	case MSG_ANY:
6371 	case MSG_BAND:
6372 		break;
6373 
6374 	default:
6375 		return (EINVAL);
6376 	}
6377 	/*
6378 	 * Setup uio and iov for data part
6379 	 */
6380 	iovs.iov_base = mdata->buf;
6381 	iovs.iov_len = mdata->maxlen;
6382 	uios.uio_iov = &iovs;
6383 	uios.uio_iovcnt = 1;
6384 	uios.uio_loffset = 0;
6385 	uios.uio_segflg = UIO_USERSPACE;
6386 	uios.uio_fmode = 0;
6387 	uios.uio_extflg = UIO_COPY_CACHED;
6388 	uios.uio_resid = mdata->maxlen;
6389 	uios.uio_offset = 0;
6390 
6391 	q = _RD(stp->sd_wrq);
6392 	mutex_enter(&stp->sd_lock);
6393 	old_sd_flag = stp->sd_flag;
6394 	mark = 0;
6395 	for (;;) {
6396 		int done = 0;
6397 		mblk_t *q_first = q->q_first;
6398 
6399 		/*
6400 		 * Get the next message of appropriate priority
6401 		 * from the stream head.  If the caller is interested
6402 		 * in band or hipri messages, then they should already
6403 		 * be enqueued at the stream head.  On the other hand
6404 		 * if the caller wants normal (band 0) messages, they
6405 		 * might be deferred in a synchronous stream and they
6406 		 * will need to be pulled up.
6407 		 *
6408 		 * After we have dequeued a message, we might find that
6409 		 * it was a deferred M_SIG that was enqueued at the
6410 		 * stream head.  It must now be posted as part of the
6411 		 * read by calling strsignal_nolock().
6412 		 *
6413 		 * Also note that strrput does not enqueue an M_PCSIG,
6414 		 * and there cannot be more than one hipri message,
6415 		 * so there was no need to have the M_PCSIG case.
6416 		 *
6417 		 * At some time it might be nice to try and wrap the
6418 		 * functionality of kstrgetmsg() and strgetmsg() into
6419 		 * a common routine so to reduce the amount of replicated
6420 		 * code (since they are extremely similar).
6421 		 */
6422 		if (!(*flagsp & (MSG_HIPRI|MSG_BAND))) {
6423 			/* Asking for normal, band0 data */
6424 			bp = strget(stp, q, uiop, first, &error);
6425 			ASSERT(MUTEX_HELD(&stp->sd_lock));
6426 			if (bp != NULL) {
6427 				if (bp->b_datap->db_type == M_SIG) {
6428 					strsignal_nolock(stp, *bp->b_rptr,
6429 					    (int32_t)bp->b_band);
6430 					continue;
6431 				} else {
6432 					break;
6433 				}
6434 			}
6435 			if (error != 0) {
6436 				goto getmout;
6437 			}
6438 
6439 		/*
6440 		 * We can't depend on the value of STRPRI here because
6441 		 * the stream head may be in transit. Therefore, we
6442 		 * must look at the type of the first message to
6443 		 * determine if a high priority messages is waiting
6444 		 */
6445 		} else if ((*flagsp & MSG_HIPRI) && q_first != NULL &&
6446 			    q_first->b_datap->db_type >= QPCTL &&
6447 			    (bp = getq_noenab(q)) != NULL) {
6448 			/* Asked for HIPRI and got one */
6449 			ASSERT(bp->b_datap->db_type >= QPCTL);
6450 			break;
6451 		} else if ((*flagsp & MSG_BAND) && q_first != NULL &&
6452 			    ((q_first->b_band >= *prip) ||
6453 			    q_first->b_datap->db_type >= QPCTL) &&
6454 			    (bp = getq_noenab(q)) != NULL) {
6455 			/*
6456 			 * Asked for at least band "prip" and got either at
6457 			 * least that band or a hipri message.
6458 			 */
6459 			ASSERT(bp->b_band >= *prip ||
6460 				bp->b_datap->db_type >= QPCTL);
6461 			if (bp->b_datap->db_type == M_SIG) {
6462 				strsignal_nolock(stp, *bp->b_rptr,
6463 				    (int32_t)bp->b_band);
6464 				continue;
6465 			} else {
6466 				break;
6467 			}
6468 		}
6469 
6470 		/* No data. Time to sleep? */
6471 		qbackenable(q, 0);
6472 
6473 		/*
6474 		 * If STRHUP or STREOF, return 0 length control and data.
6475 		 * If resid is 0, then a read(fd,buf,0) was done. Do not
6476 		 * sleep to satisfy this request because by default we have
6477 		 * zero bytes to return.
6478 		 */
6479 		if ((stp->sd_flag & (STRHUP|STREOF)) || (mctl->maxlen == 0 &&
6480 		    mdata->maxlen == 0)) {
6481 			mctl->len = mdata->len = 0;
6482 			*flagsp = 0;
6483 			mutex_exit(&stp->sd_lock);
6484 			return (0);
6485 		}
6486 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_WAIT,
6487 			"strgetmsg calls strwaitq:%p, %p",
6488 			vp, uiop);
6489 		if (((error = strwaitq(stp, GETWAIT, (ssize_t)0, fmode, -1,
6490 		    &done)) != 0) || done) {
6491 			TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_DONE,
6492 				"strgetmsg error or done:%p, %p",
6493 				vp, uiop);
6494 			mutex_exit(&stp->sd_lock);
6495 			return (error);
6496 		}
6497 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_AWAKE,
6498 			"strgetmsg awakes:%p, %p", vp, uiop);
6499 		if ((error = i_straccess(stp, JCREAD)) != 0) {
6500 			mutex_exit(&stp->sd_lock);
6501 			return (error);
6502 		}
6503 		first = 0;
6504 	}
6505 	ASSERT(bp != NULL);
6506 	/*
6507 	 * Extract any mark information. If the message is not completely
6508 	 * consumed this information will be put in the mblk
6509 	 * that is putback.
6510 	 * If MSGMARKNEXT is set and the message is completely consumed
6511 	 * the STRATMARK flag will be set below. Likewise, if
6512 	 * MSGNOTMARKNEXT is set and the message is
6513 	 * completely consumed STRNOTATMARK will be set.
6514 	 */
6515 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
6516 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
6517 		(MSGMARKNEXT|MSGNOTMARKNEXT));
6518 	if (mark != 0 && bp == stp->sd_mark) {
6519 		mark |= _LASTMARK;
6520 		stp->sd_mark = NULL;
6521 	}
6522 	/*
6523 	 * keep track of the original message type and priority
6524 	 */
6525 	pri = bp->b_band;
6526 	type = bp->b_datap->db_type;
6527 	if (type == M_PASSFP) {
6528 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
6529 			stp->sd_mark = bp;
6530 		bp->b_flag |= mark & ~_LASTMARK;
6531 		putback(stp, q, bp, pri);
6532 		qbackenable(q, pri);
6533 		mutex_exit(&stp->sd_lock);
6534 		return (EBADMSG);
6535 	}
6536 	ASSERT(type != M_SIG);
6537 
6538 	/*
6539 	 * Set this flag so strrput will not generate signals. Need to
6540 	 * make sure this flag is cleared before leaving this routine
6541 	 * else signals will stop being sent.
6542 	 */
6543 	stp->sd_flag |= STRGETINPROG;
6544 	mutex_exit(&stp->sd_lock);
6545 
6546 	if (STREAM_NEEDSERVICE(stp))
6547 		stream_runservice(stp);
6548 
6549 	/*
6550 	 * Set HIPRI flag if message is priority.
6551 	 */
6552 	if (type >= QPCTL)
6553 		flg = MSG_HIPRI;
6554 	else
6555 		flg = MSG_BAND;
6556 
6557 	/*
6558 	 * First process PROTO or PCPROTO blocks, if any.
6559 	 */
6560 	if (mctl->maxlen >= 0 && type != M_DATA) {
6561 		size_t	n, bcnt;
6562 		char	*ubuf;
6563 
6564 		bcnt = mctl->maxlen;
6565 		ubuf = mctl->buf;
6566 		while (bp != NULL && bp->b_datap->db_type != M_DATA) {
6567 			if ((n = MIN(bcnt, bp->b_wptr - bp->b_rptr)) != 0 &&
6568 			    copyout(bp->b_rptr, ubuf, n)) {
6569 				error = EFAULT;
6570 				mutex_enter(&stp->sd_lock);
6571 				/*
6572 				 * clear stream head pri flag based on
6573 				 * first message type
6574 				 */
6575 				if (type >= QPCTL) {
6576 					ASSERT(type == M_PCPROTO);
6577 					stp->sd_flag &= ~STRPRI;
6578 				}
6579 				more = 0;
6580 				freemsg(bp);
6581 				goto getmout;
6582 			}
6583 			ubuf += n;
6584 			bp->b_rptr += n;
6585 			if (bp->b_rptr >= bp->b_wptr) {
6586 				nbp = bp;
6587 				bp = bp->b_cont;
6588 				freeb(nbp);
6589 			}
6590 			ASSERT(n <= bcnt);
6591 			bcnt -= n;
6592 			if (bcnt == 0)
6593 				break;
6594 		}
6595 		mctl->len = mctl->maxlen - bcnt;
6596 	} else
6597 		mctl->len = -1;
6598 
6599 	if (bp && bp->b_datap->db_type != M_DATA) {
6600 		/*
6601 		 * More PROTO blocks in msg.
6602 		 */
6603 		more |= MORECTL;
6604 		savemp = bp;
6605 		while (bp && bp->b_datap->db_type != M_DATA) {
6606 			savemptail = bp;
6607 			bp = bp->b_cont;
6608 		}
6609 		savemptail->b_cont = NULL;
6610 	}
6611 
6612 	/*
6613 	 * Now process DATA blocks, if any.
6614 	 */
6615 	if (mdata->maxlen >= 0 && bp) {
6616 		/*
6617 		 * struiocopyout will consume a potential zero-length
6618 		 * M_DATA even if uio_resid is zero.
6619 		 */
6620 		size_t oldresid = uiop->uio_resid;
6621 
6622 		bp = struiocopyout(bp, uiop, &error);
6623 		if (error != 0) {
6624 			mutex_enter(&stp->sd_lock);
6625 			/*
6626 			 * clear stream head hi pri flag based on
6627 			 * first message
6628 			 */
6629 			if (type >= QPCTL) {
6630 				ASSERT(type == M_PCPROTO);
6631 				stp->sd_flag &= ~STRPRI;
6632 			}
6633 			more = 0;
6634 			freemsg(savemp);
6635 			goto getmout;
6636 		}
6637 		/*
6638 		 * (pr == 1) indicates a partial read.
6639 		 */
6640 		if (oldresid > uiop->uio_resid)
6641 			pr = 1;
6642 		mdata->len = mdata->maxlen - uiop->uio_resid;
6643 	} else
6644 		mdata->len = -1;
6645 
6646 	if (bp) {			/* more data blocks in msg */
6647 		more |= MOREDATA;
6648 		if (savemp)
6649 			savemptail->b_cont = bp;
6650 		else
6651 			savemp = bp;
6652 	}
6653 
6654 	mutex_enter(&stp->sd_lock);
6655 	if (savemp) {
6656 		if (pr && (savemp->b_datap->db_type == M_DATA) &&
6657 		    msgnodata(savemp)) {
6658 			/*
6659 			 * Avoid queuing a zero-length tail part of
6660 			 * a message. pr=1 indicates that we read some of
6661 			 * the message.
6662 			 */
6663 			freemsg(savemp);
6664 			more &= ~MOREDATA;
6665 			/*
6666 			 * clear stream head hi pri flag based on
6667 			 * first message
6668 			 */
6669 			if (type >= QPCTL) {
6670 				ASSERT(type == M_PCPROTO);
6671 				stp->sd_flag &= ~STRPRI;
6672 			}
6673 		} else {
6674 			savemp->b_band = pri;
6675 			/*
6676 			 * If the first message was HIPRI and the one we're
6677 			 * putting back isn't, then clear STRPRI, otherwise
6678 			 * set STRPRI again.  Note that we must set STRPRI
6679 			 * again since the flush logic in strrput_nondata()
6680 			 * may have cleared it while we had sd_lock dropped.
6681 			 */
6682 			if (type >= QPCTL) {
6683 				ASSERT(type == M_PCPROTO);
6684 				if (queclass(savemp) < QPCTL)
6685 					stp->sd_flag &= ~STRPRI;
6686 				else
6687 					stp->sd_flag |= STRPRI;
6688 			} else if (queclass(savemp) >= QPCTL) {
6689 				/*
6690 				 * The first message was not a HIPRI message,
6691 				 * but the one we are about to putback is.
6692 				 * For simplicitly, we do not allow for HIPRI
6693 				 * messages to be embedded in the message
6694 				 * body, so just force it to same type as
6695 				 * first message.
6696 				 */
6697 				ASSERT(type == M_DATA || type == M_PROTO);
6698 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
6699 				savemp->b_datap->db_type = type;
6700 			}
6701 			if (mark != 0) {
6702 				savemp->b_flag |= mark & ~_LASTMARK;
6703 				if ((mark & _LASTMARK) &&
6704 				    (stp->sd_mark == NULL)) {
6705 					/*
6706 					 * If another marked message arrived
6707 					 * while sd_lock was not held sd_mark
6708 					 * would be non-NULL.
6709 					 */
6710 					stp->sd_mark = savemp;
6711 				}
6712 			}
6713 			putback(stp, q, savemp, pri);
6714 		}
6715 	} else {
6716 		/*
6717 		 * The complete message was consumed.
6718 		 *
6719 		 * If another M_PCPROTO arrived while sd_lock was not held
6720 		 * it would have been discarded since STRPRI was still set.
6721 		 *
6722 		 * Move the MSG*MARKNEXT information
6723 		 * to the stream head just in case
6724 		 * the read queue becomes empty.
6725 		 * clear stream head hi pri flag based on
6726 		 * first message
6727 		 *
6728 		 * If the stream head was at the mark
6729 		 * (STRATMARK) before we dropped sd_lock above
6730 		 * and some data was consumed then we have
6731 		 * moved past the mark thus STRATMARK is
6732 		 * cleared. However, if a message arrived in
6733 		 * strrput during the copyout above causing
6734 		 * STRATMARK to be set we can not clear that
6735 		 * flag.
6736 		 */
6737 		if (type >= QPCTL) {
6738 			ASSERT(type == M_PCPROTO);
6739 			stp->sd_flag &= ~STRPRI;
6740 		}
6741 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
6742 			if (mark & MSGMARKNEXT) {
6743 				stp->sd_flag &= ~STRNOTATMARK;
6744 				stp->sd_flag |= STRATMARK;
6745 			} else if (mark & MSGNOTMARKNEXT) {
6746 				stp->sd_flag &= ~STRATMARK;
6747 				stp->sd_flag |= STRNOTATMARK;
6748 			} else {
6749 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
6750 			}
6751 		} else if (pr && (old_sd_flag & STRATMARK)) {
6752 			stp->sd_flag &= ~STRATMARK;
6753 		}
6754 	}
6755 
6756 	*flagsp = flg;
6757 	*prip = pri;
6758 
6759 	/*
6760 	 * Getmsg cleanup processing - if the state of the queue has changed
6761 	 * some signals may need to be sent and/or poll awakened.
6762 	 */
6763 getmout:
6764 	qbackenable(q, pri);
6765 
6766 	/*
6767 	 * We dropped the stream head lock above. Send all M_SIG messages
6768 	 * before processing stream head for SIGPOLL messages.
6769 	 */
6770 	ASSERT(MUTEX_HELD(&stp->sd_lock));
6771 	while ((bp = q->q_first) != NULL &&
6772 	    (bp->b_datap->db_type == M_SIG)) {
6773 		/*
6774 		 * sd_lock is held so the content of the read queue can not
6775 		 * change.
6776 		 */
6777 		bp = getq(q);
6778 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
6779 
6780 		strsignal_nolock(stp, *bp->b_rptr, (int32_t)bp->b_band);
6781 		mutex_exit(&stp->sd_lock);
6782 		freemsg(bp);
6783 		if (STREAM_NEEDSERVICE(stp))
6784 			stream_runservice(stp);
6785 		mutex_enter(&stp->sd_lock);
6786 	}
6787 
6788 	/*
6789 	 * stream head cannot change while we make the determination
6790 	 * whether or not to send a signal. Drop the flag to allow strrput
6791 	 * to send firstmsgsigs again.
6792 	 */
6793 	stp->sd_flag &= ~STRGETINPROG;
6794 
6795 	/*
6796 	 * If the type of message at the front of the queue changed
6797 	 * due to the receive the appropriate signals and pollwakeup events
6798 	 * are generated. The type of changes are:
6799 	 *	Processed a hipri message, q_first is not hipri.
6800 	 *	Processed a band X message, and q_first is band Y.
6801 	 * The generated signals and pollwakeups are identical to what
6802 	 * strrput() generates should the message that is now on q_first
6803 	 * arrive to an empty read queue.
6804 	 *
6805 	 * Note: only strrput will send a signal for a hipri message.
6806 	 */
6807 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
6808 		strsigset_t signals = 0;
6809 		strpollset_t pollwakeups = 0;
6810 
6811 		if (flg & MSG_HIPRI) {
6812 			/*
6813 			 * Removed a hipri message. Regular data at
6814 			 * the front of  the queue.
6815 			 */
6816 			if (bp->b_band == 0) {
6817 				signals = S_INPUT | S_RDNORM;
6818 				pollwakeups = POLLIN | POLLRDNORM;
6819 			} else {
6820 				signals = S_INPUT | S_RDBAND;
6821 				pollwakeups = POLLIN | POLLRDBAND;
6822 			}
6823 		} else if (pri != bp->b_band) {
6824 			/*
6825 			 * The band is different for the new q_first.
6826 			 */
6827 			if (bp->b_band == 0) {
6828 				signals = S_RDNORM;
6829 				pollwakeups = POLLIN | POLLRDNORM;
6830 			} else {
6831 				signals = S_RDBAND;
6832 				pollwakeups = POLLIN | POLLRDBAND;
6833 			}
6834 		}
6835 
6836 		if (pollwakeups != 0) {
6837 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
6838 				if (!(stp->sd_rput_opt & SR_POLLIN))
6839 					goto no_pollwake;
6840 				stp->sd_rput_opt &= ~SR_POLLIN;
6841 			}
6842 			mutex_exit(&stp->sd_lock);
6843 			pollwakeup(&stp->sd_pollist, pollwakeups);
6844 			mutex_enter(&stp->sd_lock);
6845 		}
6846 no_pollwake:
6847 
6848 		if (stp->sd_sigflags & signals)
6849 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
6850 	}
6851 	mutex_exit(&stp->sd_lock);
6852 
6853 	rvp->r_val1 = more;
6854 	return (error);
6855 #undef	_LASTMARK
6856 }
6857 
6858 /*
6859  * Get the next message from the read queue.  If the message is
6860  * priority, STRPRI will have been set by strrput().  This flag
6861  * should be reset only when the entire message at the front of the
6862  * queue as been consumed.
6863  *
6864  * If uiop is NULL all data is returned in mctlp.
6865  * Note that a NULL uiop implies that FNDELAY and FNONBLOCK are assumed
6866  * not enabled.
6867  * The timeout parameter is in milliseconds; -1 for infinity.
6868  * This routine handles the consolidation private flags:
6869  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
6870  *	MSG_DELAYERROR	Defer the error check until the queue is empty.
6871  *	MSG_HOLDSIG	Hold signals while waiting for data.
6872  *	MSG_IPEEK	Only peek at messages.
6873  *	MSG_DISCARDTAIL	Discard the tail M_DATA part of the message
6874  *			that doesn't fit.
6875  *	MSG_NOMARK	If the message is marked leave it on the queue.
6876  *
6877  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
6878  */
6879 int
6880 kstrgetmsg(
6881 	struct vnode *vp,
6882 	mblk_t **mctlp,
6883 	struct uio *uiop,
6884 	unsigned char *prip,
6885 	int *flagsp,
6886 	clock_t timout,
6887 	rval_t *rvp)
6888 {
6889 	struct stdata *stp;
6890 	mblk_t *bp, *nbp;
6891 	mblk_t *savemp = NULL;
6892 	mblk_t *savemptail = NULL;
6893 	int flags;
6894 	uint_t old_sd_flag;
6895 	int flg;
6896 	int more = 0;
6897 	int error = 0;
6898 	char first = 1;
6899 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
6900 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
6901 	unsigned char pri = 0;
6902 	queue_t *q;
6903 	int	pr = 0;			/* Partial read successful */
6904 	unsigned char type;
6905 
6906 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_ENTER,
6907 		"kstrgetmsg:%p", vp);
6908 
6909 	ASSERT(vp->v_stream);
6910 	stp = vp->v_stream;
6911 	rvp->r_val1 = 0;
6912 
6913 	mutex_enter(&stp->sd_lock);
6914 
6915 	if ((error = i_straccess(stp, JCREAD)) != 0) {
6916 		mutex_exit(&stp->sd_lock);
6917 		return (error);
6918 	}
6919 
6920 	flags = *flagsp;
6921 	if (stp->sd_flag & (STRDERR|STPLEX)) {
6922 		if ((stp->sd_flag & STPLEX) ||
6923 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == 0) {
6924 			error = strgeterr(stp, STRDERR|STPLEX,
6925 					(flags & MSG_IPEEK));
6926 			if (error != 0) {
6927 				mutex_exit(&stp->sd_lock);
6928 				return (error);
6929 			}
6930 		}
6931 	}
6932 	mutex_exit(&stp->sd_lock);
6933 
6934 	switch (flags & (MSG_HIPRI|MSG_ANY|MSG_BAND)) {
6935 	case MSG_HIPRI:
6936 		if (*prip != 0)
6937 			return (EINVAL);
6938 		break;
6939 
6940 	case MSG_ANY:
6941 	case MSG_BAND:
6942 		break;
6943 
6944 	default:
6945 		return (EINVAL);
6946 	}
6947 
6948 retry:
6949 	q = _RD(stp->sd_wrq);
6950 	mutex_enter(&stp->sd_lock);
6951 	old_sd_flag = stp->sd_flag;
6952 	mark = 0;
6953 	for (;;) {
6954 		int done = 0;
6955 		int waitflag;
6956 		int fmode;
6957 		mblk_t *q_first = q->q_first;
6958 
6959 		/*
6960 		 * This section of the code operates just like the code
6961 		 * in strgetmsg().  There is a comment there about what
6962 		 * is going on here.
6963 		 */
6964 		if (!(flags & (MSG_HIPRI|MSG_BAND))) {
6965 			/* Asking for normal, band0 data */
6966 			bp = strget(stp, q, uiop, first, &error);
6967 			ASSERT(MUTEX_HELD(&stp->sd_lock));
6968 			if (bp != NULL) {
6969 				if (bp->b_datap->db_type == M_SIG) {
6970 					strsignal_nolock(stp, *bp->b_rptr,
6971 					    (int32_t)bp->b_band);
6972 					continue;
6973 				} else {
6974 					break;
6975 				}
6976 			}
6977 			if (error != 0) {
6978 				goto getmout;
6979 			}
6980 		/*
6981 		 * We can't depend on the value of STRPRI here because
6982 		 * the stream head may be in transit. Therefore, we
6983 		 * must look at the type of the first message to
6984 		 * determine if a high priority messages is waiting
6985 		 */
6986 		} else if ((flags & MSG_HIPRI) && q_first != NULL &&
6987 			    q_first->b_datap->db_type >= QPCTL &&
6988 			    (bp = getq_noenab(q)) != NULL) {
6989 			ASSERT(bp->b_datap->db_type >= QPCTL);
6990 			break;
6991 		} else if ((flags & MSG_BAND) && q_first != NULL &&
6992 			    ((q_first->b_band >= *prip) ||
6993 			    q_first->b_datap->db_type >= QPCTL) &&
6994 			    (bp = getq_noenab(q)) != NULL) {
6995 			/*
6996 			 * Asked for at least band "prip" and got either at
6997 			 * least that band or a hipri message.
6998 			 */
6999 			ASSERT(bp->b_band >= *prip ||
7000 				bp->b_datap->db_type >= QPCTL);
7001 			if (bp->b_datap->db_type == M_SIG) {
7002 				strsignal_nolock(stp, *bp->b_rptr,
7003 				    (int32_t)bp->b_band);
7004 				continue;
7005 			} else {
7006 				break;
7007 			}
7008 		}
7009 
7010 		/* No data. Time to sleep? */
7011 		qbackenable(q, 0);
7012 
7013 		/*
7014 		 * Delayed error notification?
7015 		 */
7016 		if ((stp->sd_flag & (STRDERR|STPLEX)) &&
7017 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == MSG_DELAYERROR) {
7018 			error = strgeterr(stp, STRDERR|STPLEX,
7019 					(flags & MSG_IPEEK));
7020 			if (error != 0) {
7021 				mutex_exit(&stp->sd_lock);
7022 				return (error);
7023 			}
7024 		}
7025 
7026 		/*
7027 		 * If STRHUP or STREOF, return 0 length control and data.
7028 		 * If a read(fd,buf,0) has been done, do not sleep, just
7029 		 * return.
7030 		 *
7031 		 * If mctlp == NULL and uiop == NULL, then the code will
7032 		 * do the strwaitq. This is an understood way of saying
7033 		 * sleep "polling" until a message is received.
7034 		 */
7035 		if ((stp->sd_flag & (STRHUP|STREOF)) ||
7036 		    (uiop != NULL && uiop->uio_resid == 0)) {
7037 			if (mctlp != NULL)
7038 				*mctlp = NULL;
7039 			*flagsp = 0;
7040 			mutex_exit(&stp->sd_lock);
7041 			return (0);
7042 		}
7043 
7044 		waitflag = GETWAIT;
7045 		if (flags &
7046 		    (MSG_HOLDSIG|MSG_IGNERROR|MSG_IPEEK|MSG_DELAYERROR)) {
7047 			if (flags & MSG_HOLDSIG)
7048 				waitflag |= STR_NOSIG;
7049 			if (flags & MSG_IGNERROR)
7050 				waitflag |= STR_NOERROR;
7051 			if (flags & MSG_IPEEK)
7052 				waitflag |= STR_PEEK;
7053 			if (flags & MSG_DELAYERROR)
7054 				waitflag |= STR_DELAYERR;
7055 		}
7056 		if (uiop != NULL)
7057 			fmode = uiop->uio_fmode;
7058 		else
7059 			fmode = 0;
7060 
7061 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_WAIT,
7062 			"kstrgetmsg calls strwaitq:%p, %p",
7063 			vp, uiop);
7064 		if (((error = strwaitq(stp, waitflag, (ssize_t)0,
7065 		    fmode, timout, &done)) != 0) || done) {
7066 			TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_DONE,
7067 				"kstrgetmsg error or done:%p, %p",
7068 				vp, uiop);
7069 			mutex_exit(&stp->sd_lock);
7070 			return (error);
7071 		}
7072 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_AWAKE,
7073 			"kstrgetmsg awakes:%p, %p", vp, uiop);
7074 		if ((error = i_straccess(stp, JCREAD)) != 0) {
7075 			mutex_exit(&stp->sd_lock);
7076 			return (error);
7077 		}
7078 		first = 0;
7079 	}
7080 	ASSERT(bp != NULL);
7081 	/*
7082 	 * Extract any mark information. If the message is not completely
7083 	 * consumed this information will be put in the mblk
7084 	 * that is putback.
7085 	 * If MSGMARKNEXT is set and the message is completely consumed
7086 	 * the STRATMARK flag will be set below. Likewise, if
7087 	 * MSGNOTMARKNEXT is set and the message is
7088 	 * completely consumed STRNOTATMARK will be set.
7089 	 */
7090 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
7091 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
7092 		(MSGMARKNEXT|MSGNOTMARKNEXT));
7093 	pri = bp->b_band;
7094 	if (mark != 0) {
7095 		/*
7096 		 * If the caller doesn't want the mark return.
7097 		 * Used to implement MSG_WAITALL in sockets.
7098 		 */
7099 		if (flags & MSG_NOMARK) {
7100 			putback(stp, q, bp, pri);
7101 			qbackenable(q, pri);
7102 			mutex_exit(&stp->sd_lock);
7103 			return (EWOULDBLOCK);
7104 		}
7105 		if (bp == stp->sd_mark) {
7106 			mark |= _LASTMARK;
7107 			stp->sd_mark = NULL;
7108 		}
7109 	}
7110 
7111 	/*
7112 	 * keep track of the first message type
7113 	 */
7114 	type = bp->b_datap->db_type;
7115 
7116 	if (bp->b_datap->db_type == M_PASSFP) {
7117 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7118 			stp->sd_mark = bp;
7119 		bp->b_flag |= mark & ~_LASTMARK;
7120 		putback(stp, q, bp, pri);
7121 		qbackenable(q, pri);
7122 		mutex_exit(&stp->sd_lock);
7123 		return (EBADMSG);
7124 	}
7125 	ASSERT(type != M_SIG);
7126 
7127 	if (flags & MSG_IPEEK) {
7128 		/*
7129 		 * Clear any struioflag - we do the uiomove over again
7130 		 * when peeking since it simplifies the code.
7131 		 *
7132 		 * Dup the message and put the original back on the queue.
7133 		 * If dupmsg() fails, try again with copymsg() to see if
7134 		 * there is indeed a shortage of memory.  dupmsg() may fail
7135 		 * if db_ref in any of the messages reaches its limit.
7136 		 */
7137 		if ((nbp = dupmsg(bp)) == NULL && (nbp = copymsg(bp)) == NULL) {
7138 			/*
7139 			 * Restore the state of the stream head since we
7140 			 * need to drop sd_lock (strwaitbuf is sleeping).
7141 			 */
7142 			size_t size = msgdsize(bp);
7143 
7144 			if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7145 				stp->sd_mark = bp;
7146 			bp->b_flag |= mark & ~_LASTMARK;
7147 			putback(stp, q, bp, pri);
7148 			mutex_exit(&stp->sd_lock);
7149 			error = strwaitbuf(size, BPRI_HI);
7150 			if (error) {
7151 				/*
7152 				 * There is no net change to the queue thus
7153 				 * no need to qbackenable.
7154 				 */
7155 				return (error);
7156 			}
7157 			goto retry;
7158 		}
7159 
7160 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7161 			stp->sd_mark = bp;
7162 		bp->b_flag |= mark & ~_LASTMARK;
7163 		putback(stp, q, bp, pri);
7164 		bp = nbp;
7165 	}
7166 
7167 	/*
7168 	 * Set this flag so strrput will not generate signals. Need to
7169 	 * make sure this flag is cleared before leaving this routine
7170 	 * else signals will stop being sent.
7171 	 */
7172 	stp->sd_flag |= STRGETINPROG;
7173 	mutex_exit(&stp->sd_lock);
7174 
7175 	if ((stp->sd_rputdatafunc != NULL) && (DB_TYPE(bp) == M_DATA) &&
7176 	    (!(DB_FLAGS(bp) & DBLK_COOKED))) {
7177 
7178 		bp = (stp->sd_rputdatafunc)(
7179 		    stp->sd_vnode, bp, NULL,
7180 		    NULL, NULL, NULL);
7181 
7182 		if (bp == NULL)
7183 			goto retry;
7184 
7185 		DB_FLAGS(bp) |= DBLK_COOKED;
7186 	}
7187 
7188 	if (STREAM_NEEDSERVICE(stp))
7189 		stream_runservice(stp);
7190 
7191 	/*
7192 	 * Set HIPRI flag if message is priority.
7193 	 */
7194 	if (type >= QPCTL)
7195 		flg = MSG_HIPRI;
7196 	else
7197 		flg = MSG_BAND;
7198 
7199 	/*
7200 	 * First process PROTO or PCPROTO blocks, if any.
7201 	 */
7202 	if (mctlp != NULL && type != M_DATA) {
7203 		mblk_t *nbp;
7204 
7205 		*mctlp = bp;
7206 		while (bp->b_cont && bp->b_cont->b_datap->db_type != M_DATA)
7207 			bp = bp->b_cont;
7208 		nbp = bp->b_cont;
7209 		bp->b_cont = NULL;
7210 		bp = nbp;
7211 	}
7212 
7213 	if (bp && bp->b_datap->db_type != M_DATA) {
7214 		/*
7215 		 * More PROTO blocks in msg. Will only happen if mctlp is NULL.
7216 		 */
7217 		more |= MORECTL;
7218 		savemp = bp;
7219 		while (bp && bp->b_datap->db_type != M_DATA) {
7220 			savemptail = bp;
7221 			bp = bp->b_cont;
7222 		}
7223 		savemptail->b_cont = NULL;
7224 	}
7225 
7226 	/*
7227 	 * Now process DATA blocks, if any.
7228 	 */
7229 	if (uiop == NULL) {
7230 		/* Append data to tail of mctlp */
7231 		if (mctlp != NULL) {
7232 			mblk_t **mpp = mctlp;
7233 
7234 			while (*mpp != NULL)
7235 				mpp = &((*mpp)->b_cont);
7236 			*mpp = bp;
7237 			bp = NULL;
7238 		}
7239 	} else if (uiop->uio_resid >= 0 && bp) {
7240 		size_t oldresid = uiop->uio_resid;
7241 
7242 		/*
7243 		 * If a streams message is likely to consist
7244 		 * of many small mblks, it is pulled up into
7245 		 * one continuous chunk of memory.
7246 		 * see longer comment at top of page
7247 		 * by mblk_pull_len declaration.
7248 		 */
7249 
7250 		if (MBLKL(bp) < mblk_pull_len) {
7251 			(void) pullupmsg(bp, -1);
7252 		}
7253 
7254 		bp = struiocopyout(bp, uiop, &error);
7255 		if (error != 0) {
7256 			if (mctlp != NULL) {
7257 				freemsg(*mctlp);
7258 				*mctlp = NULL;
7259 			} else
7260 				freemsg(savemp);
7261 			mutex_enter(&stp->sd_lock);
7262 			/*
7263 			 * clear stream head hi pri flag based on
7264 			 * first message
7265 			 */
7266 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7267 				ASSERT(type == M_PCPROTO);
7268 				stp->sd_flag &= ~STRPRI;
7269 			}
7270 			more = 0;
7271 			goto getmout;
7272 		}
7273 		/*
7274 		 * (pr == 1) indicates a partial read.
7275 		 */
7276 		if (oldresid > uiop->uio_resid)
7277 			pr = 1;
7278 	}
7279 
7280 	if (bp) {			/* more data blocks in msg */
7281 		more |= MOREDATA;
7282 		if (savemp)
7283 			savemptail->b_cont = bp;
7284 		else
7285 			savemp = bp;
7286 	}
7287 
7288 	mutex_enter(&stp->sd_lock);
7289 	if (savemp) {
7290 		if (flags & (MSG_IPEEK|MSG_DISCARDTAIL)) {
7291 			/*
7292 			 * When MSG_DISCARDTAIL is set or
7293 			 * when peeking discard any tail. When peeking this
7294 			 * is the tail of the dup that was copied out - the
7295 			 * message has already been putback on the queue.
7296 			 * Return MOREDATA to the caller even though the data
7297 			 * is discarded. This is used by sockets (to
7298 			 * set MSG_TRUNC).
7299 			 */
7300 			freemsg(savemp);
7301 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7302 				ASSERT(type == M_PCPROTO);
7303 				stp->sd_flag &= ~STRPRI;
7304 			}
7305 		} else if (pr && (savemp->b_datap->db_type == M_DATA) &&
7306 			    msgnodata(savemp)) {
7307 			/*
7308 			 * Avoid queuing a zero-length tail part of
7309 			 * a message. pr=1 indicates that we read some of
7310 			 * the message.
7311 			 */
7312 			freemsg(savemp);
7313 			more &= ~MOREDATA;
7314 			if (type >= QPCTL) {
7315 				ASSERT(type == M_PCPROTO);
7316 				stp->sd_flag &= ~STRPRI;
7317 			}
7318 		} else {
7319 			savemp->b_band = pri;
7320 			/*
7321 			 * If the first message was HIPRI and the one we're
7322 			 * putting back isn't, then clear STRPRI, otherwise
7323 			 * set STRPRI again.  Note that we must set STRPRI
7324 			 * again since the flush logic in strrput_nondata()
7325 			 * may have cleared it while we had sd_lock dropped.
7326 			 */
7327 			if (type >= QPCTL) {
7328 				ASSERT(type == M_PCPROTO);
7329 				if (queclass(savemp) < QPCTL)
7330 					stp->sd_flag &= ~STRPRI;
7331 				else
7332 					stp->sd_flag |= STRPRI;
7333 			} else if (queclass(savemp) >= QPCTL) {
7334 				/*
7335 				 * The first message was not a HIPRI message,
7336 				 * but the one we are about to putback is.
7337 				 * For simplicitly, we do not allow for HIPRI
7338 				 * messages to be embedded in the message
7339 				 * body, so just force it to same type as
7340 				 * first message.
7341 				 */
7342 				ASSERT(type == M_DATA || type == M_PROTO);
7343 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
7344 				savemp->b_datap->db_type = type;
7345 			}
7346 			if (mark != 0) {
7347 				if ((mark & _LASTMARK) &&
7348 				    (stp->sd_mark == NULL)) {
7349 					/*
7350 					 * If another marked message arrived
7351 					 * while sd_lock was not held sd_mark
7352 					 * would be non-NULL.
7353 					 */
7354 					stp->sd_mark = savemp;
7355 				}
7356 				savemp->b_flag |= mark & ~_LASTMARK;
7357 			}
7358 			putback(stp, q, savemp, pri);
7359 		}
7360 	} else if (!(flags & MSG_IPEEK)) {
7361 		/*
7362 		 * The complete message was consumed.
7363 		 *
7364 		 * If another M_PCPROTO arrived while sd_lock was not held
7365 		 * it would have been discarded since STRPRI was still set.
7366 		 *
7367 		 * Move the MSG*MARKNEXT information
7368 		 * to the stream head just in case
7369 		 * the read queue becomes empty.
7370 		 * clear stream head hi pri flag based on
7371 		 * first message
7372 		 *
7373 		 * If the stream head was at the mark
7374 		 * (STRATMARK) before we dropped sd_lock above
7375 		 * and some data was consumed then we have
7376 		 * moved past the mark thus STRATMARK is
7377 		 * cleared. However, if a message arrived in
7378 		 * strrput during the copyout above causing
7379 		 * STRATMARK to be set we can not clear that
7380 		 * flag.
7381 		 * XXX A "perimeter" would help by single-threading strrput,
7382 		 * strread, strgetmsg and kstrgetmsg.
7383 		 */
7384 		if (type >= QPCTL) {
7385 			ASSERT(type == M_PCPROTO);
7386 			stp->sd_flag &= ~STRPRI;
7387 		}
7388 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
7389 			if (mark & MSGMARKNEXT) {
7390 				stp->sd_flag &= ~STRNOTATMARK;
7391 				stp->sd_flag |= STRATMARK;
7392 			} else if (mark & MSGNOTMARKNEXT) {
7393 				stp->sd_flag &= ~STRATMARK;
7394 				stp->sd_flag |= STRNOTATMARK;
7395 			} else {
7396 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
7397 			}
7398 		} else if (pr && (old_sd_flag & STRATMARK)) {
7399 			stp->sd_flag &= ~STRATMARK;
7400 		}
7401 	}
7402 
7403 	*flagsp = flg;
7404 	*prip = pri;
7405 
7406 	/*
7407 	 * Getmsg cleanup processing - if the state of the queue has changed
7408 	 * some signals may need to be sent and/or poll awakened.
7409 	 */
7410 getmout:
7411 	qbackenable(q, pri);
7412 
7413 	/*
7414 	 * We dropped the stream head lock above. Send all M_SIG messages
7415 	 * before processing stream head for SIGPOLL messages.
7416 	 */
7417 	ASSERT(MUTEX_HELD(&stp->sd_lock));
7418 	while ((bp = q->q_first) != NULL &&
7419 	    (bp->b_datap->db_type == M_SIG)) {
7420 		/*
7421 		 * sd_lock is held so the content of the read queue can not
7422 		 * change.
7423 		 */
7424 		bp = getq(q);
7425 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7426 
7427 		strsignal_nolock(stp, *bp->b_rptr, (int32_t)bp->b_band);
7428 		mutex_exit(&stp->sd_lock);
7429 		freemsg(bp);
7430 		if (STREAM_NEEDSERVICE(stp))
7431 			stream_runservice(stp);
7432 		mutex_enter(&stp->sd_lock);
7433 	}
7434 
7435 	/*
7436 	 * stream head cannot change while we make the determination
7437 	 * whether or not to send a signal. Drop the flag to allow strrput
7438 	 * to send firstmsgsigs again.
7439 	 */
7440 	stp->sd_flag &= ~STRGETINPROG;
7441 
7442 	/*
7443 	 * If the type of message at the front of the queue changed
7444 	 * due to the receive the appropriate signals and pollwakeup events
7445 	 * are generated. The type of changes are:
7446 	 *	Processed a hipri message, q_first is not hipri.
7447 	 *	Processed a band X message, and q_first is band Y.
7448 	 * The generated signals and pollwakeups are identical to what
7449 	 * strrput() generates should the message that is now on q_first
7450 	 * arrive to an empty read queue.
7451 	 *
7452 	 * Note: only strrput will send a signal for a hipri message.
7453 	 */
7454 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7455 		strsigset_t signals = 0;
7456 		strpollset_t pollwakeups = 0;
7457 
7458 		if (flg & MSG_HIPRI) {
7459 			/*
7460 			 * Removed a hipri message. Regular data at
7461 			 * the front of  the queue.
7462 			 */
7463 			if (bp->b_band == 0) {
7464 				signals = S_INPUT | S_RDNORM;
7465 				pollwakeups = POLLIN | POLLRDNORM;
7466 			} else {
7467 				signals = S_INPUT | S_RDBAND;
7468 				pollwakeups = POLLIN | POLLRDBAND;
7469 			}
7470 		} else if (pri != bp->b_band) {
7471 			/*
7472 			 * The band is different for the new q_first.
7473 			 */
7474 			if (bp->b_band == 0) {
7475 				signals = S_RDNORM;
7476 				pollwakeups = POLLIN | POLLRDNORM;
7477 			} else {
7478 				signals = S_RDBAND;
7479 				pollwakeups = POLLIN | POLLRDBAND;
7480 			}
7481 		}
7482 
7483 		if (pollwakeups != 0) {
7484 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
7485 				if (!(stp->sd_rput_opt & SR_POLLIN))
7486 					goto no_pollwake;
7487 				stp->sd_rput_opt &= ~SR_POLLIN;
7488 			}
7489 			mutex_exit(&stp->sd_lock);
7490 			pollwakeup(&stp->sd_pollist, pollwakeups);
7491 			mutex_enter(&stp->sd_lock);
7492 		}
7493 no_pollwake:
7494 
7495 		if (stp->sd_sigflags & signals)
7496 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7497 	}
7498 	mutex_exit(&stp->sd_lock);
7499 
7500 	rvp->r_val1 = more;
7501 	return (error);
7502 #undef	_LASTMARK
7503 }
7504 
7505 /*
7506  * Put a message downstream.
7507  *
7508  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7509  */
7510 int
7511 strputmsg(
7512 	struct vnode *vp,
7513 	struct strbuf *mctl,
7514 	struct strbuf *mdata,
7515 	unsigned char pri,
7516 	int flag,
7517 	int fmode)
7518 {
7519 	struct stdata *stp;
7520 	queue_t *wqp;
7521 	mblk_t *mp;
7522 	ssize_t msgsize;
7523 	ssize_t rmin, rmax;
7524 	int error;
7525 	struct uio uios;
7526 	struct uio *uiop = &uios;
7527 	struct iovec iovs;
7528 	int xpg4 = 0;
7529 
7530 	ASSERT(vp->v_stream);
7531 	stp = vp->v_stream;
7532 	wqp = stp->sd_wrq;
7533 
7534 	/*
7535 	 * If it is an XPG4 application, we need to send
7536 	 * SIGPIPE below
7537 	 */
7538 
7539 	xpg4 = (flag & MSG_XPG4) ? 1 : 0;
7540 	flag &= ~MSG_XPG4;
7541 
7542 #ifdef C2_AUDIT
7543 	if (audit_active)
7544 		audit_strputmsg(vp, mctl, mdata, pri, flag, fmode);
7545 #endif
7546 
7547 	mutex_enter(&stp->sd_lock);
7548 
7549 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
7550 		mutex_exit(&stp->sd_lock);
7551 		return (error);
7552 	}
7553 
7554 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7555 		error = strwriteable(stp, B_FALSE, xpg4);
7556 		if (error != 0) {
7557 			mutex_exit(&stp->sd_lock);
7558 			return (error);
7559 		}
7560 	}
7561 
7562 	mutex_exit(&stp->sd_lock);
7563 
7564 	/*
7565 	 * Check for legal flag value.
7566 	 */
7567 	switch (flag) {
7568 	case MSG_HIPRI:
7569 		if ((mctl->len < 0) || (pri != 0))
7570 			return (EINVAL);
7571 		break;
7572 	case MSG_BAND:
7573 		break;
7574 
7575 	default:
7576 		return (EINVAL);
7577 	}
7578 
7579 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_IN,
7580 		"strputmsg in:stp %p", stp);
7581 
7582 	/* get these values from those cached in the stream head */
7583 	rmin = stp->sd_qn_minpsz;
7584 	rmax = stp->sd_qn_maxpsz;
7585 
7586 	/*
7587 	 * Make sure ctl and data sizes together fall within the
7588 	 * limits of the max and min receive packet sizes and do
7589 	 * not exceed system limit.
7590 	 */
7591 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
7592 	if (rmax == 0) {
7593 		return (ERANGE);
7594 	}
7595 	/*
7596 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
7597 	 * Needed to prevent partial failures in the strmakedata loop.
7598 	 */
7599 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
7600 		rmax = stp->sd_maxblk;
7601 
7602 	if ((msgsize = mdata->len) < 0) {
7603 		msgsize = 0;
7604 		rmin = 0;	/* no range check for NULL data part */
7605 	}
7606 	if ((msgsize < rmin) ||
7607 	    ((msgsize > rmax) && (rmax != INFPSZ)) ||
7608 	    (mctl->len > strctlsz)) {
7609 		return (ERANGE);
7610 	}
7611 
7612 	/*
7613 	 * Setup uio and iov for data part
7614 	 */
7615 	iovs.iov_base = mdata->buf;
7616 	iovs.iov_len = msgsize;
7617 	uios.uio_iov = &iovs;
7618 	uios.uio_iovcnt = 1;
7619 	uios.uio_loffset = 0;
7620 	uios.uio_segflg = UIO_USERSPACE;
7621 	uios.uio_fmode = fmode;
7622 	uios.uio_extflg = UIO_COPY_DEFAULT;
7623 	uios.uio_resid = msgsize;
7624 	uios.uio_offset = 0;
7625 
7626 	/* Ignore flow control in strput for HIPRI */
7627 	if (flag & MSG_HIPRI)
7628 		flag |= MSG_IGNFLOW;
7629 
7630 	for (;;) {
7631 		int done = 0;
7632 
7633 		/*
7634 		 * strput will always free the ctl mblk - even when strput
7635 		 * fails.
7636 		 */
7637 		if ((error = strmakectl(mctl, flag, fmode, &mp)) != 0) {
7638 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7639 				"strputmsg out:stp %p out %d error %d",
7640 				stp, 1, error);
7641 			return (error);
7642 		}
7643 		/*
7644 		 * Verify that the whole message can be transferred by
7645 		 * strput.
7646 		 */
7647 		ASSERT(stp->sd_maxblk == INFPSZ ||
7648 			stp->sd_maxblk >= mdata->len);
7649 
7650 		msgsize = mdata->len;
7651 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
7652 		mdata->len = msgsize;
7653 
7654 		if (error == 0)
7655 			break;
7656 
7657 		if (error != EWOULDBLOCK)
7658 			goto out;
7659 
7660 		mutex_enter(&stp->sd_lock);
7661 		/*
7662 		 * Check for a missed wakeup.
7663 		 * Needed since strput did not hold sd_lock across
7664 		 * the canputnext.
7665 		 */
7666 		if (bcanputnext(wqp, pri)) {
7667 			/* Try again */
7668 			mutex_exit(&stp->sd_lock);
7669 			continue;
7670 		}
7671 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAIT,
7672 			"strputmsg wait:stp %p waits pri %d", stp, pri);
7673 		if (((error = strwaitq(stp, WRITEWAIT, (ssize_t)0, fmode, -1,
7674 		    &done)) != 0) || done) {
7675 			mutex_exit(&stp->sd_lock);
7676 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7677 				"strputmsg out:q %p out %d error %d",
7678 				stp, 0, error);
7679 			return (error);
7680 		}
7681 		TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAKE,
7682 			"strputmsg wake:stp %p wakes", stp);
7683 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
7684 			mutex_exit(&stp->sd_lock);
7685 			return (error);
7686 		}
7687 		mutex_exit(&stp->sd_lock);
7688 	}
7689 out:
7690 	/*
7691 	 * For historic reasons, applications expect EAGAIN
7692 	 * when data mblk could not be allocated. so change
7693 	 * ENOMEM back to EAGAIN
7694 	 */
7695 	if (error == ENOMEM)
7696 		error = EAGAIN;
7697 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7698 		"strputmsg out:stp %p out %d error %d", stp, 2, error);
7699 	return (error);
7700 }
7701 
7702 /*
7703  * Put a message downstream.
7704  * Can send only an M_PROTO/M_PCPROTO by passing in a NULL uiop.
7705  * The fmode flag (NDELAY, NONBLOCK) is the or of the flags in the uio
7706  * and the fmode parameter.
7707  *
7708  * This routine handles the consolidation private flags:
7709  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
7710  *	MSG_HOLDSIG	Hold signals while waiting for data.
7711  *	MSG_IGNFLOW	Don't check streams flow control.
7712  *
7713  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7714  */
7715 int
7716 kstrputmsg(
7717 	struct vnode *vp,
7718 	mblk_t *mctl,
7719 	struct uio *uiop,
7720 	ssize_t msgsize,
7721 	unsigned char pri,
7722 	int flag,
7723 	int fmode)
7724 {
7725 	struct stdata *stp;
7726 	queue_t *wqp;
7727 	ssize_t rmin, rmax;
7728 	int error;
7729 
7730 	ASSERT(vp->v_stream);
7731 	stp = vp->v_stream;
7732 	wqp = stp->sd_wrq;
7733 #ifdef C2_AUDIT
7734 	if (audit_active)
7735 		audit_strputmsg(vp, NULL, NULL, pri, flag, fmode);
7736 #endif
7737 	if (mctl == NULL)
7738 		return (EINVAL);
7739 
7740 	mutex_enter(&stp->sd_lock);
7741 
7742 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
7743 		mutex_exit(&stp->sd_lock);
7744 		freemsg(mctl);
7745 		return (error);
7746 	}
7747 
7748 	if ((stp->sd_flag & STPLEX) || !(flag & MSG_IGNERROR)) {
7749 		if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7750 			error = strwriteable(stp, B_FALSE, B_TRUE);
7751 			if (error != 0) {
7752 				mutex_exit(&stp->sd_lock);
7753 				freemsg(mctl);
7754 				return (error);
7755 			}
7756 		}
7757 	}
7758 
7759 	mutex_exit(&stp->sd_lock);
7760 
7761 	/*
7762 	 * Check for legal flag value.
7763 	 */
7764 	switch (flag & (MSG_HIPRI|MSG_BAND|MSG_ANY)) {
7765 	case MSG_HIPRI:
7766 		if (pri != 0) {
7767 			freemsg(mctl);
7768 			return (EINVAL);
7769 		}
7770 		break;
7771 	case MSG_BAND:
7772 		break;
7773 	default:
7774 		freemsg(mctl);
7775 		return (EINVAL);
7776 	}
7777 
7778 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_IN,
7779 		"kstrputmsg in:stp %p", stp);
7780 
7781 	/* get these values from those cached in the stream head */
7782 	rmin = stp->sd_qn_minpsz;
7783 	rmax = stp->sd_qn_maxpsz;
7784 
7785 	/*
7786 	 * Make sure ctl and data sizes together fall within the
7787 	 * limits of the max and min receive packet sizes and do
7788 	 * not exceed system limit.
7789 	 */
7790 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
7791 	if (rmax == 0) {
7792 		freemsg(mctl);
7793 		return (ERANGE);
7794 	}
7795 	/*
7796 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
7797 	 * Needed to prevent partial failures in the strmakedata loop.
7798 	 */
7799 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
7800 		rmax = stp->sd_maxblk;
7801 
7802 	if (uiop == NULL) {
7803 		msgsize = -1;
7804 		rmin = -1;	/* no range check for NULL data part */
7805 	} else {
7806 		/* Use uio flags as well as the fmode parameter flags */
7807 		fmode |= uiop->uio_fmode;
7808 
7809 		if ((msgsize < rmin) ||
7810 		    ((msgsize > rmax) && (rmax != INFPSZ))) {
7811 			freemsg(mctl);
7812 			return (ERANGE);
7813 		}
7814 	}
7815 
7816 	/* Ignore flow control in strput for HIPRI */
7817 	if (flag & MSG_HIPRI)
7818 		flag |= MSG_IGNFLOW;
7819 
7820 	for (;;) {
7821 		int done = 0;
7822 		int waitflag;
7823 		mblk_t *mp;
7824 
7825 		/*
7826 		 * strput will always free the ctl mblk - even when strput
7827 		 * fails. If MSG_IGNFLOW is set then any error returned
7828 		 * will cause us to break the loop, so we don't need a copy
7829 		 * of the message. If MSG_IGNFLOW is not set, then we can
7830 		 * get hit by flow control and be forced to try again. In
7831 		 * this case we need to have a copy of the message. We
7832 		 * do this using copymsg since the message may get modified
7833 		 * by something below us.
7834 		 *
7835 		 * We've observed that many TPI providers do not check db_ref
7836 		 * on the control messages but blindly reuse them for the
7837 		 * T_OK_ACK/T_ERROR_ACK. Thus using copymsg is more
7838 		 * friendly to such providers than using dupmsg. Also, note
7839 		 * that sockfs uses MSG_IGNFLOW for all TPI control messages.
7840 		 * Only data messages are subject to flow control, hence
7841 		 * subject to this copymsg.
7842 		 */
7843 		if (flag & MSG_IGNFLOW) {
7844 			mp = mctl;
7845 			mctl = NULL;
7846 		} else {
7847 			do {
7848 				/*
7849 				 * If a message has a free pointer, the message
7850 				 * must be dupmsg to maintain this pointer.
7851 				 * Code using this facility must be sure
7852 				 * that modules below will not change the
7853 				 * contents of the dblk without checking db_ref
7854 				 * first. If db_ref is > 1, then the module
7855 				 * needs to do a copymsg first. Otherwise,
7856 				 * the contents of the dblk may become
7857 				 * inconsistent because the freesmg/freeb below
7858 				 * may end up calling atomic_add_32_nv.
7859 				 * The atomic_add_32_nv in freeb (accessing
7860 				 * all of db_ref, db_type, db_flags, and
7861 				 * db_struioflag) does not prevent other threads
7862 				 * from concurrently trying to modify e.g.
7863 				 * db_type.
7864 				 */
7865 				if (mctl->b_datap->db_frtnp != NULL)
7866 					mp = dupmsg(mctl);
7867 				else
7868 					mp = copymsg(mctl);
7869 
7870 				if (mp != NULL)
7871 					break;
7872 
7873 				error = strwaitbuf(msgdsize(mctl), BPRI_MED);
7874 				if (error) {
7875 					freemsg(mctl);
7876 					return (error);
7877 				}
7878 			} while (mp == NULL);
7879 		}
7880 		/*
7881 		 * Verify that all of msgsize can be transferred by
7882 		 * strput.
7883 		 */
7884 		ASSERT(stp->sd_maxblk == INFPSZ || stp->sd_maxblk >= msgsize);
7885 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
7886 		if (error == 0)
7887 			break;
7888 
7889 		if (error != EWOULDBLOCK)
7890 			goto out;
7891 
7892 		/*
7893 		 * IF MSG_IGNFLOW is set we should have broken out of loop
7894 		 * above.
7895 		 */
7896 		ASSERT(!(flag & MSG_IGNFLOW));
7897 		mutex_enter(&stp->sd_lock);
7898 		/*
7899 		 * Check for a missed wakeup.
7900 		 * Needed since strput did not hold sd_lock across
7901 		 * the canputnext.
7902 		 */
7903 		if (bcanputnext(wqp, pri)) {
7904 			/* Try again */
7905 			mutex_exit(&stp->sd_lock);
7906 			continue;
7907 		}
7908 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAIT,
7909 			"kstrputmsg wait:stp %p waits pri %d", stp, pri);
7910 
7911 		waitflag = WRITEWAIT;
7912 		if (flag & (MSG_HOLDSIG|MSG_IGNERROR)) {
7913 			if (flag & MSG_HOLDSIG)
7914 				waitflag |= STR_NOSIG;
7915 			if (flag & MSG_IGNERROR)
7916 				waitflag |= STR_NOERROR;
7917 		}
7918 		if (((error = strwaitq(stp, waitflag,
7919 		    (ssize_t)0, fmode, -1, &done)) != 0) || done) {
7920 			mutex_exit(&stp->sd_lock);
7921 			TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
7922 				"kstrputmsg out:stp %p out %d error %d",
7923 				stp, 0, error);
7924 			freemsg(mctl);
7925 			return (error);
7926 		}
7927 		TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAKE,
7928 			"kstrputmsg wake:stp %p wakes", stp);
7929 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
7930 			mutex_exit(&stp->sd_lock);
7931 			freemsg(mctl);
7932 			return (error);
7933 		}
7934 		mutex_exit(&stp->sd_lock);
7935 	}
7936 out:
7937 	freemsg(mctl);
7938 	/*
7939 	 * For historic reasons, applications expect EAGAIN
7940 	 * when data mblk could not be allocated. so change
7941 	 * ENOMEM back to EAGAIN
7942 	 */
7943 	if (error == ENOMEM)
7944 		error = EAGAIN;
7945 	TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
7946 		"kstrputmsg out:stp %p out %d error %d", stp, 2, error);
7947 	return (error);
7948 }
7949 
7950 /*
7951  * Determines whether the necessary conditions are set on a stream
7952  * for it to be readable, writeable, or have exceptions.
7953  *
7954  * strpoll handles the consolidation private events:
7955  *	POLLNOERR	Do not return POLLERR even if there are stream
7956  *			head errors.
7957  *			Used by sockfs.
7958  *	POLLRDDATA	Do not return POLLIN unless at least one message on
7959  *			the queue contains one or more M_DATA mblks. Thus
7960  *			when this flag is set a queue with only
7961  *			M_PROTO/M_PCPROTO mblks does not return POLLIN.
7962  *			Used by sockfs to ignore T_EXDATA_IND messages.
7963  *
7964  * Note: POLLRDDATA assumes that synch streams only return messages with
7965  * an M_DATA attached (i.e. not messages consisting of only
7966  * an M_PROTO/M_PCPROTO part).
7967  */
7968 int
7969 strpoll(
7970 	struct stdata *stp,
7971 	short events_arg,
7972 	int anyyet,
7973 	short *reventsp,
7974 	struct pollhead **phpp)
7975 {
7976 	int events = (ushort_t)events_arg;
7977 	int retevents = 0;
7978 	mblk_t *mp;
7979 	qband_t *qbp;
7980 	long sd_flags = stp->sd_flag;
7981 	int headlocked = 0;
7982 
7983 	/*
7984 	 * For performance, a single 'if' tests for most possible edge
7985 	 * conditions in one shot
7986 	 */
7987 	if (sd_flags & (STPLEX | STRDERR | STWRERR)) {
7988 		if (sd_flags & STPLEX) {
7989 			*reventsp = POLLNVAL;
7990 			return (EINVAL);
7991 		}
7992 		if (((events & (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) &&
7993 		    (sd_flags & STRDERR)) ||
7994 		    ((events & (POLLOUT | POLLWRNORM | POLLWRBAND)) &&
7995 		    (sd_flags & STWRERR))) {
7996 			if (!(events & POLLNOERR)) {
7997 				*reventsp = POLLERR;
7998 				return (0);
7999 			}
8000 		}
8001 	}
8002 	if (sd_flags & STRHUP) {
8003 		retevents |= POLLHUP;
8004 	} else if (events & (POLLWRNORM | POLLWRBAND)) {
8005 		queue_t *tq;
8006 		queue_t	*qp = stp->sd_wrq;
8007 
8008 		claimstr(qp);
8009 		/* Find next module forward that has a service procedure */
8010 		tq = qp->q_next->q_nfsrv;
8011 		ASSERT(tq != NULL);
8012 
8013 		polllock(&stp->sd_pollist, QLOCK(tq));
8014 		if (events & POLLWRNORM) {
8015 			queue_t *sqp;
8016 
8017 			if (tq->q_flag & QFULL)
8018 				/* ensure backq svc procedure runs */
8019 				tq->q_flag |= QWANTW;
8020 			else if ((sqp = stp->sd_struiowrq) != NULL) {
8021 				/* Check sync stream barrier write q */
8022 				mutex_exit(QLOCK(tq));
8023 				polllock(&stp->sd_pollist, QLOCK(sqp));
8024 				if (sqp->q_flag & QFULL)
8025 					/* ensure pollwakeup() is done */
8026 					sqp->q_flag |= QWANTWSYNC;
8027 				else
8028 					retevents |= POLLOUT;
8029 				/* More write events to process ??? */
8030 				if (! (events & POLLWRBAND)) {
8031 					mutex_exit(QLOCK(sqp));
8032 					releasestr(qp);
8033 					goto chkrd;
8034 				}
8035 				mutex_exit(QLOCK(sqp));
8036 				polllock(&stp->sd_pollist, QLOCK(tq));
8037 			} else
8038 				retevents |= POLLOUT;
8039 		}
8040 		if (events & POLLWRBAND) {
8041 			qbp = tq->q_bandp;
8042 			if (qbp) {
8043 				while (qbp) {
8044 					if (qbp->qb_flag & QB_FULL)
8045 						qbp->qb_flag |= QB_WANTW;
8046 					else
8047 						retevents |= POLLWRBAND;
8048 					qbp = qbp->qb_next;
8049 				}
8050 			} else {
8051 				retevents |= POLLWRBAND;
8052 			}
8053 		}
8054 		mutex_exit(QLOCK(tq));
8055 		releasestr(qp);
8056 	}
8057 chkrd:
8058 	if (sd_flags & STRPRI) {
8059 		retevents |= (events & POLLPRI);
8060 	} else if (events & (POLLRDNORM | POLLRDBAND | POLLIN)) {
8061 		queue_t	*qp = _RD(stp->sd_wrq);
8062 		int normevents = (events & (POLLIN | POLLRDNORM));
8063 
8064 		/*
8065 		 * Note: Need to do polllock() here since ps_lock may be
8066 		 * held. See bug 4191544.
8067 		 */
8068 		polllock(&stp->sd_pollist, &stp->sd_lock);
8069 		headlocked = 1;
8070 		mp = qp->q_first;
8071 		while (mp) {
8072 			/*
8073 			 * For POLLRDDATA we scan b_cont and b_next until we
8074 			 * find an M_DATA.
8075 			 */
8076 			if ((events & POLLRDDATA) &&
8077 			    mp->b_datap->db_type != M_DATA) {
8078 				mblk_t *nmp = mp->b_cont;
8079 
8080 				while (nmp != NULL &&
8081 				    nmp->b_datap->db_type != M_DATA)
8082 					nmp = nmp->b_cont;
8083 				if (nmp == NULL) {
8084 					mp = mp->b_next;
8085 					continue;
8086 				}
8087 			}
8088 			if (mp->b_band == 0)
8089 				retevents |= normevents;
8090 			else
8091 				retevents |= (events & (POLLIN | POLLRDBAND));
8092 			break;
8093 		}
8094 		if (! (retevents & normevents) &&
8095 		    (stp->sd_wakeq & RSLEEP)) {
8096 			/*
8097 			 * Sync stream barrier read queue has data.
8098 			 */
8099 			retevents |= normevents;
8100 		}
8101 		/* Treat eof as normal data */
8102 		if (sd_flags & STREOF)
8103 			retevents |= normevents;
8104 	}
8105 
8106 	*reventsp = (short)retevents;
8107 	if (retevents) {
8108 		if (headlocked)
8109 			mutex_exit(&stp->sd_lock);
8110 		return (0);
8111 	}
8112 
8113 	/*
8114 	 * If poll() has not found any events yet, set up event cell
8115 	 * to wake up the poll if a requested event occurs on this
8116 	 * stream.  Check for collisions with outstanding poll requests.
8117 	 */
8118 	if (!anyyet) {
8119 		*phpp = &stp->sd_pollist;
8120 		if (headlocked == 0) {
8121 			polllock(&stp->sd_pollist, &stp->sd_lock);
8122 			headlocked = 1;
8123 		}
8124 		stp->sd_rput_opt |= SR_POLLIN;
8125 	}
8126 	if (headlocked)
8127 		mutex_exit(&stp->sd_lock);
8128 	return (0);
8129 }
8130 
8131 /*
8132  * The purpose of putback() is to assure sleeping polls/reads
8133  * are awakened when there are no new messages arriving at the,
8134  * stream head, and a message is placed back on the read queue.
8135  *
8136  * sd_lock must be held when messages are placed back on stream
8137  * head.  (getq() holds sd_lock when it removes messages from
8138  * the queue)
8139  */
8140 
8141 static void
8142 putback(struct stdata *stp, queue_t *q, mblk_t *bp, int band)
8143 {
8144 	ASSERT(MUTEX_HELD(&stp->sd_lock));
8145 	(void) putbq(q, bp);
8146 	/*
8147 	 * A message may have come in when the sd_lock was dropped in the
8148 	 * calling routine. If this is the case and STR*ATMARK info was
8149 	 * received, need to move that from the stream head to the q_last
8150 	 * so that SIOCATMARK can return the proper value.
8151 	 */
8152 	if (stp->sd_flag & (STRATMARK | STRNOTATMARK)) {
8153 		unsigned short *flagp = &q->q_last->b_flag;
8154 		uint_t b_flag = (uint_t)*flagp;
8155 
8156 		if (stp->sd_flag & STRATMARK) {
8157 			b_flag &= ~MSGNOTMARKNEXT;
8158 			b_flag |= MSGMARKNEXT;
8159 			stp->sd_flag &= ~STRATMARK;
8160 		} else {
8161 			b_flag &= ~MSGMARKNEXT;
8162 			b_flag |= MSGNOTMARKNEXT;
8163 			stp->sd_flag &= ~STRNOTATMARK;
8164 		}
8165 		*flagp = (unsigned short) b_flag;
8166 	}
8167 
8168 #ifdef	DEBUG
8169 	/*
8170 	 * Make sure that the flags are not messed up.
8171 	 */
8172 	{
8173 		mblk_t *mp;
8174 		mp = q->q_last;
8175 		while (mp != NULL) {
8176 			ASSERT((mp->b_flag & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
8177 			    (MSGMARKNEXT|MSGNOTMARKNEXT));
8178 			mp = mp->b_cont;
8179 		}
8180 	}
8181 #endif
8182 	if (q->q_first == bp) {
8183 		short pollevents;
8184 
8185 		if (stp->sd_flag & RSLEEP) {
8186 			stp->sd_flag &= ~RSLEEP;
8187 			cv_broadcast(&q->q_wait);
8188 		}
8189 		if (stp->sd_flag & STRPRI) {
8190 			pollevents = POLLPRI;
8191 		} else {
8192 			if (band == 0) {
8193 				if (!(stp->sd_rput_opt & SR_POLLIN))
8194 					return;
8195 				stp->sd_rput_opt &= ~SR_POLLIN;
8196 				pollevents = POLLIN | POLLRDNORM;
8197 			} else {
8198 				pollevents = POLLIN | POLLRDBAND;
8199 			}
8200 		}
8201 		mutex_exit(&stp->sd_lock);
8202 		pollwakeup(&stp->sd_pollist, pollevents);
8203 		mutex_enter(&stp->sd_lock);
8204 	}
8205 }
8206 
8207 /*
8208  * Return the held vnode attached to the stream head of a
8209  * given queue
8210  * It is the responsibility of the calling routine to ensure
8211  * that the queue does not go away (e.g. pop).
8212  */
8213 vnode_t *
8214 strq2vp(queue_t *qp)
8215 {
8216 	vnode_t *vp;
8217 	vp = STREAM(qp)->sd_vnode;
8218 	ASSERT(vp != NULL);
8219 	VN_HOLD(vp);
8220 	return (vp);
8221 }
8222 
8223 /*
8224  * return the stream head write queue for the given vp
8225  * It is the responsibility of the calling routine to ensure
8226  * that the stream or vnode do not close.
8227  */
8228 queue_t *
8229 strvp2wq(vnode_t *vp)
8230 {
8231 	ASSERT(vp->v_stream != NULL);
8232 	return (vp->v_stream->sd_wrq);
8233 }
8234 
8235 /*
8236  * pollwakeup stream head
8237  * It is the responsibility of the calling routine to ensure
8238  * that the stream or vnode do not close.
8239  */
8240 void
8241 strpollwakeup(vnode_t *vp, short event)
8242 {
8243 	ASSERT(vp->v_stream);
8244 	pollwakeup(&vp->v_stream->sd_pollist, event);
8245 }
8246 
8247 /*
8248  * Mate the stream heads of two vnodes together. If the two vnodes are the
8249  * same, we just make the write-side point at the read-side -- otherwise,
8250  * we do a full mate.  Only works on vnodes associated with streams that are
8251  * still being built and thus have only a stream head.
8252  */
8253 void
8254 strmate(vnode_t *vp1, vnode_t *vp2)
8255 {
8256 	queue_t *wrq1 = strvp2wq(vp1);
8257 	queue_t *wrq2 = strvp2wq(vp2);
8258 
8259 	/*
8260 	 * Verify that there are no modules on the stream yet.  We also
8261 	 * rely on the stream head always having a service procedure to
8262 	 * avoid tweaking q_nfsrv.
8263 	 */
8264 	ASSERT(wrq1->q_next == NULL && wrq2->q_next == NULL);
8265 	ASSERT(wrq1->q_qinfo->qi_srvp != NULL);
8266 	ASSERT(wrq2->q_qinfo->qi_srvp != NULL);
8267 
8268 	/*
8269 	 * If the queues are the same, just twist; otherwise do a full mate.
8270 	 */
8271 	if (wrq1 == wrq2) {
8272 		wrq1->q_next = _RD(wrq1);
8273 	} else {
8274 		wrq1->q_next = _RD(wrq2);
8275 		wrq2->q_next = _RD(wrq1);
8276 		STREAM(wrq1)->sd_mate = STREAM(wrq2);
8277 		STREAM(wrq1)->sd_flag |= STRMATE;
8278 		STREAM(wrq2)->sd_mate = STREAM(wrq1);
8279 		STREAM(wrq2)->sd_flag |= STRMATE;
8280 	}
8281 }
8282 
8283 /*
8284  * XXX will go away when console is correctly fixed.
8285  * Clean up the console PIDS, from previous I_SETSIG,
8286  * called only for cnopen which never calls strclean().
8287  */
8288 void
8289 str_cn_clean(struct vnode *vp)
8290 {
8291 	strsig_t *ssp, *pssp, *tssp;
8292 	struct stdata *stp;
8293 	struct pid  *pidp;
8294 	int update = 0;
8295 
8296 	ASSERT(vp->v_stream);
8297 	stp = vp->v_stream;
8298 	pssp = NULL;
8299 	mutex_enter(&stp->sd_lock);
8300 	ssp = stp->sd_siglist;
8301 	while (ssp) {
8302 		mutex_enter(&pidlock);
8303 		pidp = ssp->ss_pidp;
8304 		/*
8305 		 * Get rid of PID if the proc is gone.
8306 		 */
8307 		if (pidp->pid_prinactive) {
8308 			tssp = ssp->ss_next;
8309 			if (pssp)
8310 				pssp->ss_next = tssp;
8311 			else
8312 				stp->sd_siglist = tssp;
8313 			ASSERT(pidp->pid_ref <= 1);
8314 			PID_RELE(ssp->ss_pidp);
8315 			mutex_exit(&pidlock);
8316 			kmem_free(ssp, sizeof (strsig_t));
8317 			update = 1;
8318 			ssp = tssp;
8319 			continue;
8320 		} else
8321 			mutex_exit(&pidlock);
8322 		pssp = ssp;
8323 		ssp = ssp->ss_next;
8324 	}
8325 	if (update) {
8326 		stp->sd_sigflags = 0;
8327 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
8328 			stp->sd_sigflags |= ssp->ss_events;
8329 	}
8330 	mutex_exit(&stp->sd_lock);
8331 }
8332 
8333 /*
8334  * Return B_TRUE if there is data in the message, B_FALSE otherwise.
8335  */
8336 static boolean_t
8337 msghasdata(mblk_t *bp)
8338 {
8339 	for (; bp; bp = bp->b_cont)
8340 		if (bp->b_datap->db_type == M_DATA) {
8341 			ASSERT(bp->b_wptr >= bp->b_rptr);
8342 			if (bp->b_wptr > bp->b_rptr)
8343 				return (B_TRUE);
8344 		}
8345 	return (B_FALSE);
8346 }
8347