xref: /titanic_41/usr/src/uts/common/os/streamio.c (revision 74e20cfe817b82802b16fac8690dadcda76f54f5)
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 2006 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/priocntl.h>
56 #include <sys/jioctl.h>
57 #include <sys/procset.h>
58 #include <sys/session.h>
59 #include <sys/kmem.h>
60 #include <sys/filio.h>
61 #include <sys/vtrace.h>
62 #include <sys/debug.h>
63 #include <sys/strredir.h>
64 #include <sys/fs/fifonode.h>
65 #include <sys/fs/snode.h>
66 #include <sys/strlog.h>
67 #include <sys/strsun.h>
68 #include <sys/project.h>
69 #include <sys/kbio.h>
70 #include <sys/msio.h>
71 #include <sys/tty.h>
72 #include <sys/ptyvar.h>
73 #include <sys/vuid_event.h>
74 #include <sys/modctl.h>
75 #include <sys/sunddi.h>
76 #include <sys/sunldi_impl.h>
77 #include <sys/autoconf.h>
78 #include <sys/policy.h>
79 
80 
81 /*
82  * This define helps improve the readability of streams code while
83  * still maintaining a very old streams performance enhancement.  The
84  * performance enhancement basically involved having all callers
85  * of straccess() perform the first check that straccess() will do
86  * locally before actually calling straccess().  (There by reducing
87  * the number of unnecessary calls to straccess().)
88  */
89 #define	i_straccess(x, y)	((stp->sd_sidp == NULL) ? 0 : \
90 				    (stp->sd_vnode->v_type == VFIFO) ? 0 : \
91 				    straccess((x), (y)))
92 
93 /*
94  * what is mblk_pull_len?
95  *
96  * If a streams message consists of many short messages,
97  * a performance degradation occurs from copyout overhead.
98  * To decrease the per mblk overhead, messages that are
99  * likely to consist of many small mblks are pulled up into
100  * one continuous chunk of memory.
101  *
102  * To avoid the processing overhead of examining every
103  * mblk, a quick heuristic is used. If the first mblk in
104  * the message is shorter than mblk_pull_len, it is likely
105  * that the rest of the mblk will be short.
106  *
107  * This heuristic was decided upon after performance tests
108  * indicated that anything more complex slowed down the main
109  * code path.
110  */
111 #define	MBLK_PULL_LEN 64
112 uint32_t mblk_pull_len = MBLK_PULL_LEN;
113 
114 /*
115  * The sgttyb_handling flag controls the handling of the old BSD
116  * TIOCGETP, TIOCSETP, and TIOCSETN ioctls as follows:
117  *
118  * 0 - Emit no warnings at all and retain old, broken behavior.
119  * 1 - Emit no warnings and silently handle new semantics.
120  * 2 - Send cmn_err(CE_NOTE) when either TIOCSETP or TIOCSETN is used
121  *     (once per system invocation).  Handle with new semantics.
122  * 3 - Send SIGSYS when any TIOCGETP, TIOCSETP, or TIOCSETN call is
123  *     made (so that offenders drop core and are easy to debug).
124  *
125  * The "new semantics" are that TIOCGETP returns B38400 for
126  * sg_[io]speed if the corresponding value is over B38400, and that
127  * TIOCSET[PN] accept B38400 in these cases to mean "retain current
128  * bit rate."
129  */
130 int sgttyb_handling = 1;
131 static boolean_t sgttyb_complaint;
132 
133 /* don't push drcompat module by default on Style-2 streams */
134 static int push_drcompat = 0;
135 
136 /*
137  * id value used to distinguish between different ioctl messages
138  */
139 static uint32_t ioc_id;
140 
141 static void putback(struct stdata *, queue_t *, mblk_t *, int);
142 static void strcleanall(struct vnode *);
143 static int strwsrv(queue_t *);
144 
145 /*
146  * qinit and module_info structures for stream head read and write queues
147  */
148 struct module_info strm_info = { 0, "strrhead", 0, INFPSZ, STRHIGH, STRLOW };
149 struct module_info stwm_info = { 0, "strwhead", 0, 0, 0, 0 };
150 struct qinit strdata = { strrput, NULL, NULL, NULL, NULL, &strm_info };
151 struct qinit stwdata = { NULL, strwsrv, NULL, NULL, NULL, &stwm_info };
152 struct module_info fiform_info = { 0, "fifostrrhead", 0, PIPE_BUF, FIFOHIWAT,
153     FIFOLOWAT };
154 struct module_info fifowm_info = { 0, "fifostrwhead", 0, 0, 0, 0 };
155 struct qinit fifo_strdata = { strrput, NULL, NULL, NULL, NULL, &fiform_info };
156 struct qinit fifo_stwdata = { NULL, strwsrv, NULL, NULL, NULL, &fifowm_info };
157 
158 extern kmutex_t	strresources;	/* protects global resources */
159 extern kmutex_t muxifier;	/* single-threads multiplexor creation */
160 
161 static boolean_t msghasdata(mblk_t *bp);
162 #define	msgnodata(bp) (!msghasdata(bp))
163 
164 /*
165  * Stream head locking notes:
166  *	There are four monitors associated with the stream head:
167  *	1. v_stream monitor: in stropen() and strclose() v_lock
168  *		is held while the association of vnode and stream
169  *		head is established or tested for.
170  *	2. open/close/push/pop monitor: sd_lock is held while each
171  *		thread bids for exclusive access to this monitor
172  *		for opening or closing a stream.  In addition, this
173  *		monitor is entered during pushes and pops.  This
174  *		guarantees that during plumbing operations there
175  *		is only one thread trying to change the plumbing.
176  *		Any other threads present in the stream are only
177  *		using the plumbing.
178  *	3. read/write monitor: in the case of read, a thread holds
179  *		sd_lock while trying to get data from the stream
180  *		head queue.  if there is none to fulfill a read
181  *		request, it sets RSLEEP and calls cv_wait_sig() down
182  *		in strwaitq() to await the arrival of new data.
183  *		when new data arrives in strrput(), sd_lock is acquired
184  *		before testing for RSLEEP and calling cv_broadcast().
185  *		the behavior of strwrite(), strwsrv(), and WSLEEP
186  *		mirror this.
187  *	4. ioctl monitor: sd_lock is gotten to ensure that only one
188  *		thread is doing an ioctl at a time.
189  */
190 
191 static int
192 push_mod(queue_t *qp, dev_t *devp, struct stdata *stp, const char *name,
193     int anchor, cred_t *crp)
194 {
195 	int error;
196 	fmodsw_impl_t *fp;
197 
198 	if (stp->sd_flag & (STRHUP|STRDERR|STWRERR)) {
199 		error = (stp->sd_flag & STRHUP) ? ENXIO : EIO;
200 		return (error);
201 	}
202 	if (stp->sd_pushcnt >= nstrpush) {
203 		return (EINVAL);
204 	}
205 
206 	if ((fp = fmodsw_find(name, FMODSW_HOLD | FMODSW_LOAD)) == NULL) {
207 		stp->sd_flag |= STREOPENFAIL;
208 		return (EINVAL);
209 	}
210 
211 	/*
212 	 * push new module and call its open routine via qattach
213 	 */
214 	if ((error = qattach(qp, devp, 0, crp, fp, B_FALSE)) != 0)
215 		return (error);
216 
217 	/*
218 	 * Check to see if caller wants a STREAMS anchor
219 	 * put at this place in the stream, and add if so.
220 	 */
221 	mutex_enter(&stp->sd_lock);
222 	if (anchor == stp->sd_pushcnt)
223 		stp->sd_anchor = stp->sd_pushcnt;
224 	mutex_exit(&stp->sd_lock);
225 
226 	return (0);
227 }
228 
229 /*
230  * Open a stream device.
231  */
232 int
233 stropen(vnode_t *vp, dev_t *devp, int flag, cred_t *crp)
234 {
235 	struct stdata *stp;
236 	queue_t *qp;
237 	int s;
238 	dev_t dummydev;
239 	struct autopush *ap;
240 	int error = 0;
241 	ssize_t	rmin, rmax;
242 	int cloneopen;
243 	queue_t *brq;
244 	major_t major;
245 
246 #ifdef C2_AUDIT
247 	if (audit_active)
248 		audit_stropen(vp, devp, flag, crp);
249 #endif
250 
251 	/*
252 	 * If the stream already exists, wait for any open in progress
253 	 * to complete, then call the open function of each module and
254 	 * driver in the stream.  Otherwise create the stream.
255 	 */
256 	TRACE_1(TR_FAC_STREAMS_FR, TR_STROPEN, "stropen:%p", vp);
257 retry:
258 	mutex_enter(&vp->v_lock);
259 	if ((stp = vp->v_stream) != NULL) {
260 
261 		/*
262 		 * Waiting for stream to be created to device
263 		 * due to another open.
264 		 */
265 	    mutex_exit(&vp->v_lock);
266 
267 	    if (STRMATED(stp)) {
268 		struct stdata *strmatep = stp->sd_mate;
269 
270 		STRLOCKMATES(stp);
271 		if (strmatep->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
272 			if (flag & (FNDELAY|FNONBLOCK)) {
273 				error = EAGAIN;
274 				mutex_exit(&strmatep->sd_lock);
275 				goto ckreturn;
276 			}
277 			mutex_exit(&stp->sd_lock);
278 			if (!cv_wait_sig(&strmatep->sd_monitor,
279 			    &strmatep->sd_lock)) {
280 				error = EINTR;
281 				mutex_exit(&strmatep->sd_lock);
282 				mutex_enter(&stp->sd_lock);
283 				goto ckreturn;
284 			}
285 			mutex_exit(&strmatep->sd_lock);
286 			goto retry;
287 		}
288 		if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
289 			if (flag & (FNDELAY|FNONBLOCK)) {
290 				error = EAGAIN;
291 				mutex_exit(&strmatep->sd_lock);
292 				goto ckreturn;
293 			}
294 			mutex_exit(&strmatep->sd_lock);
295 			if (!cv_wait_sig(&stp->sd_monitor, &stp->sd_lock)) {
296 				error = EINTR;
297 				goto ckreturn;
298 			}
299 			mutex_exit(&stp->sd_lock);
300 			goto retry;
301 		}
302 
303 		if (stp->sd_flag & (STRDERR|STWRERR)) {
304 			error = EIO;
305 			mutex_exit(&strmatep->sd_lock);
306 			goto ckreturn;
307 		}
308 
309 		stp->sd_flag |= STWOPEN;
310 		STRUNLOCKMATES(stp);
311 	    } else {
312 		mutex_enter(&stp->sd_lock);
313 		if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
314 			if (flag & (FNDELAY|FNONBLOCK)) {
315 				error = EAGAIN;
316 				goto ckreturn;
317 			}
318 			if (!cv_wait_sig(&stp->sd_monitor, &stp->sd_lock)) {
319 				error = EINTR;
320 				goto ckreturn;
321 			}
322 			mutex_exit(&stp->sd_lock);
323 			goto retry;  /* could be clone! */
324 		}
325 
326 		if (stp->sd_flag & (STRDERR|STWRERR)) {
327 			error = EIO;
328 			goto ckreturn;
329 		}
330 
331 		stp->sd_flag |= STWOPEN;
332 		mutex_exit(&stp->sd_lock);
333 	    }
334 
335 		/*
336 		 * Open all modules and devices down stream to notify
337 		 * that another user is streaming.  For modules, set the
338 		 * last argument to MODOPEN and do not pass any open flags.
339 		 * Ignore dummydev since this is not the first open.
340 		 */
341 	    claimstr(stp->sd_wrq);
342 	    qp = stp->sd_wrq;
343 	    while (_SAMESTR(qp)) {
344 		qp = qp->q_next;
345 		if ((error = qreopen(_RD(qp), devp, flag, crp)) != 0)
346 			break;
347 	    }
348 	    releasestr(stp->sd_wrq);
349 	    mutex_enter(&stp->sd_lock);
350 	    stp->sd_flag &= ~(STRHUP|STWOPEN|STRDERR|STWRERR);
351 	    stp->sd_rerror = 0;
352 	    stp->sd_werror = 0;
353 ckreturn:
354 	    cv_broadcast(&stp->sd_monitor);
355 	    mutex_exit(&stp->sd_lock);
356 	    return (error);
357 	}
358 
359 	/*
360 	 * This vnode isn't streaming.  SPECFS already
361 	 * checked for multiple vnodes pointing to the
362 	 * same stream, so create a stream to the driver.
363 	 */
364 	qp = allocq();
365 	stp = shalloc(qp);
366 
367 	/*
368 	 * Initialize stream head.  shalloc() has given us
369 	 * exclusive access, and we have the vnode locked;
370 	 * we can do whatever we want with stp.
371 	 */
372 	stp->sd_flag = STWOPEN;
373 	stp->sd_siglist = NULL;
374 	stp->sd_pollist.ph_list = NULL;
375 	stp->sd_sigflags = 0;
376 	stp->sd_mark = NULL;
377 	stp->sd_closetime = STRTIMOUT;
378 	stp->sd_sidp = NULL;
379 	stp->sd_pgidp = NULL;
380 	stp->sd_vnode = vp;
381 	stp->sd_rerror = 0;
382 	stp->sd_werror = 0;
383 	stp->sd_wroff = 0;
384 	stp->sd_tail = 0;
385 	stp->sd_iocblk = NULL;
386 	stp->sd_pushcnt = 0;
387 	stp->sd_qn_minpsz = 0;
388 	stp->sd_qn_maxpsz = INFPSZ - 1;	/* used to check for initialization */
389 	stp->sd_maxblk = INFPSZ;
390 	qp->q_ptr = _WR(qp)->q_ptr = stp;
391 	STREAM(qp) = STREAM(_WR(qp)) = stp;
392 	vp->v_stream = stp;
393 	mutex_exit(&vp->v_lock);
394 	if (vp->v_type == VFIFO) {
395 		stp->sd_flag |= OLDNDELAY;
396 		/*
397 		 * This means, both for pipes and fifos
398 		 * strwrite will send SIGPIPE if the other
399 		 * end is closed. For putmsg it depends
400 		 * on whether it is a XPG4_2 application
401 		 * or not
402 		 */
403 		stp->sd_wput_opt = SW_SIGPIPE;
404 
405 		/* setq might sleep in kmem_alloc - avoid holding locks. */
406 		setq(qp, &fifo_strdata, &fifo_stwdata, NULL, QMTSAFE,
407 		    SQ_CI|SQ_CO, B_FALSE);
408 
409 		set_qend(qp);
410 		stp->sd_strtab = fifo_getinfo();
411 		_WR(qp)->q_nfsrv = _WR(qp);
412 		qp->q_nfsrv = qp;
413 		/*
414 		 * Wake up others that are waiting for stream to be created.
415 		 */
416 		mutex_enter(&stp->sd_lock);
417 		/*
418 		 * nothing is be pushed on stream yet, so
419 		 * optimized stream head packetsizes are just that
420 		 * of the read queue
421 		 */
422 		stp->sd_qn_minpsz = qp->q_minpsz;
423 		stp->sd_qn_maxpsz = qp->q_maxpsz;
424 		stp->sd_flag &= ~STWOPEN;
425 		goto fifo_opendone;
426 	}
427 	/* setq might sleep in kmem_alloc - avoid holding locks. */
428 	setq(qp, &strdata, &stwdata, NULL, QMTSAFE, SQ_CI|SQ_CO, B_FALSE);
429 
430 	set_qend(qp);
431 
432 	/*
433 	 * Open driver and create stream to it (via qattach).
434 	 */
435 	cloneopen = (getmajor(*devp) == clone_major);
436 	if ((error = qattach(qp, devp, flag, crp, NULL, B_FALSE)) != 0) {
437 		mutex_enter(&vp->v_lock);
438 		vp->v_stream = NULL;
439 		mutex_exit(&vp->v_lock);
440 		mutex_enter(&stp->sd_lock);
441 		cv_broadcast(&stp->sd_monitor);
442 		mutex_exit(&stp->sd_lock);
443 		freeq(_RD(qp));
444 		shfree(stp);
445 		return (error);
446 	}
447 	/*
448 	 * Set sd_strtab after open in order to handle clonable drivers
449 	 */
450 	stp->sd_strtab = STREAMSTAB(getmajor(*devp));
451 
452 	/*
453 	 * Historical note: dummydev used to be be prior to the initial
454 	 * open (via qattach above), which made the value seen
455 	 * inconsistent between an I_PUSH and an autopush of a module.
456 	 */
457 	dummydev = *devp;
458 
459 	/*
460 	 * For clone open of old style (Q not associated) network driver,
461 	 * push DRMODNAME module to handle DL_ATTACH/DL_DETACH
462 	 */
463 	brq = _RD(_WR(qp)->q_next);
464 	major = getmajor(*devp);
465 	if (push_drcompat && cloneopen && NETWORK_DRV(major) &&
466 	    ((brq->q_flag & _QASSOCIATED) == 0)) {
467 		if (push_mod(qp, &dummydev, stp, DRMODNAME, 0, crp) != 0)
468 			cmn_err(CE_WARN, "cannot push " DRMODNAME
469 			    " streams module");
470 	}
471 
472 	/*
473 	 * check for modules that need to be autopushed
474 	 */
475 	if ((ap = sad_ap_find_by_dev(*devp)) == NULL)
476 		goto opendone;
477 	for (s = 0; s < ap->ap_npush; s++) {
478 		error = push_mod(qp, &dummydev, stp, ap->ap_list[s],
479 		    ap->ap_anchor, crp);
480 		if (error != 0)
481 			break;
482 	}
483 	sad_ap_rele(ap);
484 
485 	/*
486 	 * let specfs know that open failed part way through
487 	 */
488 	if (error) {
489 		mutex_enter(&stp->sd_lock);
490 		stp->sd_flag |= STREOPENFAIL;
491 		mutex_exit(&stp->sd_lock);
492 	}
493 
494 opendone:
495 
496 	/*
497 	 * Wake up others that are waiting for stream to be created.
498 	 */
499 	mutex_enter(&stp->sd_lock);
500 	stp->sd_flag &= ~STWOPEN;
501 
502 	/*
503 	 * As a performance concern we are caching the values of
504 	 * q_minpsz and q_maxpsz of the module below the stream
505 	 * head in the stream head.
506 	 */
507 	mutex_enter(QLOCK(stp->sd_wrq->q_next));
508 	rmin = stp->sd_wrq->q_next->q_minpsz;
509 	rmax = stp->sd_wrq->q_next->q_maxpsz;
510 	mutex_exit(QLOCK(stp->sd_wrq->q_next));
511 
512 	/* do this processing here as a performance concern */
513 	if (strmsgsz != 0) {
514 		if (rmax == INFPSZ)
515 			rmax = strmsgsz;
516 		else
517 			rmax = MIN(strmsgsz, rmax);
518 	}
519 
520 	mutex_enter(QLOCK(stp->sd_wrq));
521 	stp->sd_qn_minpsz = rmin;
522 	stp->sd_qn_maxpsz = rmax;
523 	mutex_exit(QLOCK(stp->sd_wrq));
524 
525 fifo_opendone:
526 	cv_broadcast(&stp->sd_monitor);
527 	mutex_exit(&stp->sd_lock);
528 	return (error);
529 }
530 
531 static int strsink(queue_t *, mblk_t *);
532 static struct qinit deadrend = {
533 	strsink, NULL, NULL, NULL, NULL, &strm_info, NULL
534 };
535 static struct qinit deadwend = {
536 	NULL, NULL, NULL, NULL, NULL, &stwm_info, NULL
537 };
538 
539 /*
540  * Close a stream.
541  * This is called from closef() on the last close of an open stream.
542  * Strclean() will already have removed the siglist and pollist
543  * information, so all that remains is to remove all multiplexor links
544  * for the stream, pop all the modules (and the driver), and free the
545  * stream structure.
546  */
547 
548 int
549 strclose(struct vnode *vp, int flag, cred_t *crp)
550 {
551 	struct stdata *stp;
552 	queue_t *qp;
553 	int rval;
554 	int freestp = 1;
555 	queue_t *rmq;
556 
557 #ifdef C2_AUDIT
558 	if (audit_active)
559 		audit_strclose(vp, flag, crp);
560 #endif
561 
562 	TRACE_1(TR_FAC_STREAMS_FR,
563 		TR_STRCLOSE, "strclose:%p", vp);
564 	ASSERT(vp->v_stream);
565 
566 	stp = vp->v_stream;
567 	ASSERT(!(stp->sd_flag & STPLEX));
568 	qp = stp->sd_wrq;
569 
570 	/*
571 	 * Needed so that strpoll will return non-zero for this fd.
572 	 * Note that with POLLNOERR STRHUP does still cause POLLHUP.
573 	 */
574 	mutex_enter(&stp->sd_lock);
575 	stp->sd_flag |= STRHUP;
576 	mutex_exit(&stp->sd_lock);
577 
578 	/*
579 	 * If the registered process or process group did not have an
580 	 * open instance of this stream then strclean would not be
581 	 * called. Thus at the time of closing all remaining siglist entries
582 	 * are removed.
583 	 */
584 	if (stp->sd_siglist != NULL)
585 		strcleanall(vp);
586 
587 	ASSERT(stp->sd_siglist == NULL);
588 	ASSERT(stp->sd_sigflags == 0);
589 
590 	if (STRMATED(stp)) {
591 		struct stdata *strmatep = stp->sd_mate;
592 		int waited = 1;
593 
594 		STRLOCKMATES(stp);
595 		while (waited) {
596 			waited = 0;
597 			while (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
598 				mutex_exit(&strmatep->sd_lock);
599 				cv_wait(&stp->sd_monitor, &stp->sd_lock);
600 				mutex_exit(&stp->sd_lock);
601 				STRLOCKMATES(stp);
602 				waited = 1;
603 			}
604 			while (strmatep->sd_flag &
605 			    (STWOPEN|STRCLOSE|STRPLUMB)) {
606 				mutex_exit(&stp->sd_lock);
607 				cv_wait(&strmatep->sd_monitor,
608 				    &strmatep->sd_lock);
609 				mutex_exit(&strmatep->sd_lock);
610 				STRLOCKMATES(stp);
611 				waited = 1;
612 			}
613 		}
614 		stp->sd_flag |= STRCLOSE;
615 		STRUNLOCKMATES(stp);
616 	} else {
617 		mutex_enter(&stp->sd_lock);
618 		stp->sd_flag |= STRCLOSE;
619 		mutex_exit(&stp->sd_lock);
620 	}
621 
622 	ASSERT(qp->q_first == NULL);	/* No more delayed write */
623 
624 	/* Check if an I_LINK was ever done on this stream */
625 	if (stp->sd_flag & STRHASLINKS) {
626 		(void) munlinkall(stp, LINKCLOSE|LINKNORMAL, crp, &rval);
627 	}
628 
629 	while (_SAMESTR(qp)) {
630 		/*
631 		 * Holding sd_lock prevents q_next from changing in
632 		 * this stream.
633 		 */
634 		mutex_enter(&stp->sd_lock);
635 		if (!(flag & (FNDELAY|FNONBLOCK)) && (stp->sd_closetime > 0)) {
636 
637 			/*
638 			 * sleep until awakened by strwsrv() or timeout
639 			 */
640 			for (;;) {
641 				mutex_enter(QLOCK(qp->q_next));
642 				if (!(qp->q_next->q_mblkcnt)) {
643 					mutex_exit(QLOCK(qp->q_next));
644 					break;
645 				}
646 				stp->sd_flag |= WSLEEP;
647 
648 				/* ensure strwsrv gets enabled */
649 				qp->q_next->q_flag |= QWANTW;
650 				mutex_exit(QLOCK(qp->q_next));
651 				/* get out if we timed out or recv'd a signal */
652 				if (str_cv_wait(&qp->q_wait, &stp->sd_lock,
653 				    stp->sd_closetime, 0) <= 0) {
654 					break;
655 				}
656 			}
657 			stp->sd_flag &= ~WSLEEP;
658 		}
659 		mutex_exit(&stp->sd_lock);
660 
661 		rmq = qp->q_next;
662 		if (rmq->q_flag & QISDRV) {
663 			ASSERT(!_SAMESTR(rmq));
664 			wait_sq_svc(_RD(qp)->q_syncq);
665 		}
666 
667 		qdetach(_RD(rmq), 1, flag, crp, B_FALSE);
668 	}
669 
670 	/*
671 	 * Since we call pollwakeup in close() now, the poll list should
672 	 * be empty in most cases. The only exception is the layered devices
673 	 * (e.g. the console drivers with redirection modules pushed on top
674 	 * of it).  We have to do this after calling qdetach() because
675 	 * the redirection module won't have torn down the console
676 	 * redirection until after qdetach() has been invoked.
677 	 */
678 	if (stp->sd_pollist.ph_list != NULL) {
679 		pollwakeup(&stp->sd_pollist, POLLERR);
680 		pollhead_clean(&stp->sd_pollist);
681 	}
682 	ASSERT(stp->sd_pollist.ph_list == NULL);
683 	ASSERT(stp->sd_sidp == NULL);
684 	ASSERT(stp->sd_pgidp == NULL);
685 
686 	/* Prevent qenable from re-enabling the stream head queue */
687 	disable_svc(_RD(qp));
688 
689 	/*
690 	 * Wait until service procedure of each queue is
691 	 * run, if QINSERVICE is set.
692 	 */
693 	wait_svc(_RD(qp));
694 
695 	/*
696 	 * Now, flush both queues.
697 	 */
698 	flushq(_RD(qp), FLUSHALL);
699 	flushq(qp, FLUSHALL);
700 
701 	/*
702 	 * If the write queue of the stream head is pointing to a
703 	 * read queue, we have a twisted stream.  If the read queue
704 	 * is alive, convert the stream head queues into a dead end.
705 	 * If the read queue is dead, free the dead pair.
706 	 */
707 	if (qp->q_next && !_SAMESTR(qp)) {
708 		if (qp->q_next->q_qinfo == &deadrend) {	/* half-closed pipe */
709 			flushq(qp->q_next, FLUSHALL); /* ensure no message */
710 			shfree(qp->q_next->q_stream);
711 			freeq(qp->q_next);
712 			freeq(_RD(qp));
713 		} else if (qp->q_next == _RD(qp)) {	/* fifo */
714 			freeq(_RD(qp));
715 		} else {				/* pipe */
716 			freestp = 0;
717 			/*
718 			 * The q_info pointers are never accessed when
719 			 * SQLOCK is held.
720 			 */
721 			ASSERT(qp->q_syncq == _RD(qp)->q_syncq);
722 			mutex_enter(SQLOCK(qp->q_syncq));
723 			qp->q_qinfo = &deadwend;
724 			_RD(qp)->q_qinfo = &deadrend;
725 			mutex_exit(SQLOCK(qp->q_syncq));
726 		}
727 	} else {
728 		freeq(_RD(qp)); /* free stream head queue pair */
729 	}
730 
731 	mutex_enter(&vp->v_lock);
732 	if (stp->sd_iocblk) {
733 		if (stp->sd_iocblk != (mblk_t *)-1) {
734 			freemsg(stp->sd_iocblk);
735 		}
736 		stp->sd_iocblk = NULL;
737 	}
738 	stp->sd_vnode = NULL;
739 	vp->v_stream = NULL;
740 	mutex_exit(&vp->v_lock);
741 	mutex_enter(&stp->sd_lock);
742 	stp->sd_flag &= ~STRCLOSE;
743 	cv_broadcast(&stp->sd_monitor);
744 	mutex_exit(&stp->sd_lock);
745 
746 	if (freestp)
747 		shfree(stp);
748 	return (0);
749 }
750 
751 static int
752 strsink(queue_t *q, mblk_t *bp)
753 {
754 	struct copyresp *resp;
755 
756 	switch (bp->b_datap->db_type) {
757 	case M_FLUSH:
758 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
759 			*bp->b_rptr &= ~FLUSHR;
760 			bp->b_flag |= MSGNOLOOP;
761 			/*
762 			 * Protect against the driver passing up
763 			 * messages after it has done a qprocsoff.
764 			 */
765 			if (_OTHERQ(q)->q_next == NULL)
766 				freemsg(bp);
767 			else
768 				qreply(q, bp);
769 		} else {
770 			freemsg(bp);
771 		}
772 		break;
773 
774 	case M_COPYIN:
775 	case M_COPYOUT:
776 		if (bp->b_cont) {
777 			freemsg(bp->b_cont);
778 			bp->b_cont = NULL;
779 		}
780 		bp->b_datap->db_type = M_IOCDATA;
781 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
782 		resp = (struct copyresp *)bp->b_rptr;
783 		resp->cp_rval = (caddr_t)1;	/* failure */
784 		/*
785 		 * Protect against the driver passing up
786 		 * messages after it has done a qprocsoff.
787 		 */
788 		if (_OTHERQ(q)->q_next == NULL)
789 			freemsg(bp);
790 		else
791 			qreply(q, bp);
792 		break;
793 
794 	case M_IOCTL:
795 		if (bp->b_cont) {
796 			freemsg(bp->b_cont);
797 			bp->b_cont = NULL;
798 		}
799 		bp->b_datap->db_type = M_IOCNAK;
800 		/*
801 		 * Protect against the driver passing up
802 		 * messages after it has done a qprocsoff.
803 		 */
804 		if (_OTHERQ(q)->q_next == NULL)
805 			freemsg(bp);
806 		else
807 			qreply(q, bp);
808 		break;
809 
810 	default:
811 		freemsg(bp);
812 		break;
813 	}
814 
815 	return (0);
816 }
817 
818 /*
819  * Clean up after a process when it closes a stream.  This is called
820  * from closef for all closes, whereas strclose is called only for the
821  * last close on a stream.  The siglist is scanned for entries for the
822  * current process, and these are removed.
823  */
824 void
825 strclean(struct vnode *vp)
826 {
827 	strsig_t *ssp, *pssp, *tssp;
828 	stdata_t *stp;
829 	int update = 0;
830 
831 	TRACE_1(TR_FAC_STREAMS_FR,
832 		TR_STRCLEAN, "strclean:%p", vp);
833 	stp = vp->v_stream;
834 	pssp = NULL;
835 	mutex_enter(&stp->sd_lock);
836 	ssp = stp->sd_siglist;
837 	while (ssp) {
838 		if (ssp->ss_pidp == curproc->p_pidp) {
839 			tssp = ssp->ss_next;
840 			if (pssp)
841 				pssp->ss_next = tssp;
842 			else
843 				stp->sd_siglist = tssp;
844 			mutex_enter(&pidlock);
845 			PID_RELE(ssp->ss_pidp);
846 			mutex_exit(&pidlock);
847 			kmem_free(ssp, sizeof (strsig_t));
848 			update = 1;
849 			ssp = tssp;
850 		} else {
851 			pssp = ssp;
852 			ssp = ssp->ss_next;
853 		}
854 	}
855 	if (update) {
856 		stp->sd_sigflags = 0;
857 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
858 			stp->sd_sigflags |= ssp->ss_events;
859 	}
860 	mutex_exit(&stp->sd_lock);
861 }
862 
863 /*
864  * Used on the last close to remove any remaining items on the siglist.
865  * These could be present on the siglist due to I_ESETSIG calls that
866  * use process groups or processed that do not have an open file descriptor
867  * for this stream (Such entries would not be removed by strclean).
868  */
869 static void
870 strcleanall(struct vnode *vp)
871 {
872 	strsig_t *ssp, *nssp;
873 	stdata_t *stp;
874 
875 	stp = vp->v_stream;
876 	mutex_enter(&stp->sd_lock);
877 	ssp = stp->sd_siglist;
878 	stp->sd_siglist = NULL;
879 	while (ssp) {
880 		nssp = ssp->ss_next;
881 		mutex_enter(&pidlock);
882 		PID_RELE(ssp->ss_pidp);
883 		mutex_exit(&pidlock);
884 		kmem_free(ssp, sizeof (strsig_t));
885 		ssp = nssp;
886 	}
887 	stp->sd_sigflags = 0;
888 	mutex_exit(&stp->sd_lock);
889 }
890 
891 /*
892  * Retrieve the next message from the logical stream head read queue
893  * using either rwnext (if sync stream) or getq_noenab.
894  * It is the callers responsibility to call qbackenable after
895  * it is finished with the message. The caller should not call
896  * qbackenable until after any putback calls to avoid spurious backenabling.
897  */
898 mblk_t *
899 strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
900     int *errorp)
901 {
902 	mblk_t *bp;
903 	int error;
904 
905 	ASSERT(MUTEX_HELD(&stp->sd_lock));
906 	/* Holding sd_lock prevents the read queue from changing  */
907 
908 	if (uiop != NULL && stp->sd_struiordq != NULL &&
909 	    q->q_first == NULL &&
910 	    (!first || (stp->sd_wakeq & RSLEEP))) {
911 		/*
912 		 * Stream supports rwnext() for the read side.
913 		 * If this is the first time we're called by e.g. strread
914 		 * only do the downcall if there is a deferred wakeup
915 		 * (registered in sd_wakeq).
916 		 */
917 		struiod_t uiod;
918 
919 		if (first)
920 			stp->sd_wakeq &= ~RSLEEP;
921 
922 		(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
923 			sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
924 		uiod.d_mp = 0;
925 		/*
926 		 * Mark that a thread is in rwnext on the read side
927 		 * to prevent strrput from nacking ioctls immediately.
928 		 * When the last concurrent rwnext returns
929 		 * the ioctls are nack'ed.
930 		 */
931 		ASSERT(MUTEX_HELD(&stp->sd_lock));
932 		stp->sd_struiodnak++;
933 		/*
934 		 * Note: rwnext will drop sd_lock.
935 		 */
936 		error = rwnext(q, &uiod);
937 		ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
938 		mutex_enter(&stp->sd_lock);
939 		stp->sd_struiodnak--;
940 		while (stp->sd_struiodnak == 0 &&
941 		    ((bp = stp->sd_struionak) != NULL)) {
942 			stp->sd_struionak = bp->b_next;
943 			bp->b_next = NULL;
944 			bp->b_datap->db_type = M_IOCNAK;
945 			/*
946 			 * Protect against the driver passing up
947 			 * messages after it has done a qprocsoff.
948 			 */
949 			if (_OTHERQ(q)->q_next == NULL)
950 				freemsg(bp);
951 			else {
952 				mutex_exit(&stp->sd_lock);
953 				qreply(q, bp);
954 				mutex_enter(&stp->sd_lock);
955 			}
956 		}
957 		ASSERT(MUTEX_HELD(&stp->sd_lock));
958 		if (error == 0 || error == EWOULDBLOCK) {
959 			if ((bp = uiod.d_mp) != NULL) {
960 				*errorp = 0;
961 				ASSERT(MUTEX_HELD(&stp->sd_lock));
962 				return (bp);
963 			}
964 			error = 0;
965 		} else if (error == EINVAL) {
966 			/*
967 			 * The stream plumbing must have
968 			 * changed while we were away, so
969 			 * just turn off rwnext()s.
970 			 */
971 			error = 0;
972 		} else if (error == EBUSY) {
973 			/*
974 			 * The module might have data in transit using putnext
975 			 * Fall back on waiting + getq.
976 			 */
977 			error = 0;
978 		} else {
979 			*errorp = error;
980 			ASSERT(MUTEX_HELD(&stp->sd_lock));
981 			return (NULL);
982 		}
983 		/*
984 		 * Try a getq in case a rwnext() generated mblk
985 		 * has bubbled up via strrput().
986 		 */
987 	}
988 	*errorp = 0;
989 	ASSERT(MUTEX_HELD(&stp->sd_lock));
990 	return (getq_noenab(q));
991 }
992 
993 /*
994  * Copy out the message pointed to by `bp' into the uio pointed to by `uiop'.
995  * If the message does not fit in the uio the remainder of it is returned;
996  * otherwise NULL is returned.  Any embedded zero-length mblk_t's are
997  * consumed, even if uio_resid reaches zero.  On error, `*errorp' is set to
998  * the error code, the message is consumed, and NULL is returned.
999  */
1000 static mblk_t *
1001 struiocopyout(mblk_t *bp, struct uio *uiop, int *errorp)
1002 {
1003 	int error;
1004 	ptrdiff_t n;
1005 	mblk_t *nbp;
1006 
1007 	ASSERT(bp->b_wptr >= bp->b_rptr);
1008 
1009 	do {
1010 		if ((n = MIN(uiop->uio_resid, MBLKL(bp))) != 0) {
1011 			ASSERT(n > 0);
1012 
1013 			error = uiomove(bp->b_rptr, n, UIO_READ, uiop);
1014 			if (error != 0) {
1015 				freemsg(bp);
1016 				*errorp = error;
1017 				return (NULL);
1018 			}
1019 		}
1020 
1021 		bp->b_rptr += n;
1022 		while (bp != NULL && (bp->b_rptr >= bp->b_wptr)) {
1023 			nbp = bp;
1024 			bp = bp->b_cont;
1025 			freeb(nbp);
1026 		}
1027 	} while (bp != NULL && uiop->uio_resid > 0);
1028 
1029 	*errorp = 0;
1030 	return (bp);
1031 }
1032 
1033 /*
1034  * Read a stream according to the mode flags in sd_flag:
1035  *
1036  * (default mode)		- Byte stream, msg boundaries are ignored
1037  * RD_MSGDIS (msg discard)	- Read on msg boundaries and throw away
1038  *				any data remaining in msg
1039  * RD_MSGNODIS (msg non-discard) - Read on msg boundaries and put back
1040  *				any remaining data on head of read queue
1041  *
1042  * Consume readable messages on the front of the queue until
1043  * ttolwp(curthread)->lwp_count
1044  * is satisfied, the readable messages are exhausted, or a message
1045  * boundary is reached in a message mode.  If no data was read and
1046  * the stream was not opened with the NDELAY flag, block until data arrives.
1047  * Otherwise return the data read and update the count.
1048  *
1049  * In default mode a 0 length message signifies end-of-file and terminates
1050  * a read in progress.  The 0 length message is removed from the queue
1051  * only if it is the only message read (no data is read).
1052  *
1053  * An attempt to read an M_PROTO or M_PCPROTO message results in an
1054  * EBADMSG error return, unless either RD_PROTDAT or RD_PROTDIS are set.
1055  * If RD_PROTDAT is set, M_PROTO and M_PCPROTO messages are read as data.
1056  * If RD_PROTDIS is set, the M_PROTO and M_PCPROTO parts of the message
1057  * are unlinked from and M_DATA blocks in the message, the protos are
1058  * thrown away, and the data is read.
1059  */
1060 /* ARGSUSED */
1061 int
1062 strread(struct vnode *vp, struct uio *uiop, cred_t *crp)
1063 {
1064 	struct stdata *stp;
1065 	mblk_t *bp, *nbp;
1066 	queue_t *q;
1067 	int error = 0;
1068 	uint_t old_sd_flag;
1069 	int first;
1070 	char rflg;
1071 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
1072 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
1073 	short delim;
1074 	unsigned char pri = 0;
1075 	char waitflag;
1076 	unsigned char type;
1077 
1078 	TRACE_1(TR_FAC_STREAMS_FR,
1079 		TR_STRREAD_ENTER, "strread:%p", vp);
1080 	ASSERT(vp->v_stream);
1081 	stp = vp->v_stream;
1082 
1083 	mutex_enter(&stp->sd_lock);
1084 
1085 	if ((error = i_straccess(stp, JCREAD)) != 0) {
1086 		mutex_exit(&stp->sd_lock);
1087 		return (error);
1088 	}
1089 
1090 	if (stp->sd_flag & (STRDERR|STPLEX)) {
1091 		error = strgeterr(stp, STRDERR|STPLEX, 0);
1092 		if (error != 0) {
1093 			mutex_exit(&stp->sd_lock);
1094 			return (error);
1095 		}
1096 	}
1097 
1098 	/*
1099 	 * Loop terminates when uiop->uio_resid == 0.
1100 	 */
1101 	rflg = 0;
1102 	waitflag = READWAIT;
1103 	q = _RD(stp->sd_wrq);
1104 	for (;;) {
1105 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1106 		old_sd_flag = stp->sd_flag;
1107 		mark = 0;
1108 		delim = 0;
1109 		first = 1;
1110 		while ((bp = strget(stp, q, uiop, first, &error)) == NULL) {
1111 			int done = 0;
1112 
1113 			ASSERT(MUTEX_HELD(&stp->sd_lock));
1114 
1115 			if (error != 0)
1116 				goto oops;
1117 
1118 			if (stp->sd_flag & (STRHUP|STREOF)) {
1119 				goto oops;
1120 			}
1121 			if (rflg && !(stp->sd_flag & STRDELIM)) {
1122 				goto oops;
1123 			}
1124 			/*
1125 			 * If a read(fd,buf,0) has been done, there is no
1126 			 * need to sleep. We always have zero bytes to
1127 			 * return.
1128 			 */
1129 			if (uiop->uio_resid == 0) {
1130 				goto oops;
1131 			}
1132 
1133 			qbackenable(q, 0);
1134 
1135 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_WAIT,
1136 				"strread calls strwaitq:%p, %p, %p",
1137 				vp, uiop, crp);
1138 			if ((error = strwaitq(stp, waitflag, uiop->uio_resid,
1139 			    uiop->uio_fmode, -1, &done)) != 0 || done) {
1140 				TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_DONE,
1141 					"strread error or done:%p, %p, %p",
1142 					vp, uiop, crp);
1143 				if ((uiop->uio_fmode & FNDELAY) &&
1144 				    (stp->sd_flag & OLDNDELAY) &&
1145 				    (error == EAGAIN))
1146 					error = 0;
1147 				goto oops;
1148 			}
1149 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_AWAKE,
1150 				"strread awakes:%p, %p, %p", vp, uiop, crp);
1151 			if ((error = i_straccess(stp, JCREAD)) != 0) {
1152 				goto oops;
1153 			}
1154 			first = 0;
1155 		}
1156 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1157 		ASSERT(bp);
1158 		pri = bp->b_band;
1159 		/*
1160 		 * Extract any mark information. If the message is not
1161 		 * completely consumed this information will be put in the mblk
1162 		 * that is putback.
1163 		 * If MSGMARKNEXT is set and the message is completely consumed
1164 		 * the STRATMARK flag will be set below. Likewise, if
1165 		 * MSGNOTMARKNEXT is set and the message is
1166 		 * completely consumed STRNOTATMARK will be set.
1167 		 *
1168 		 * For some unknown reason strread only breaks the read at the
1169 		 * last mark.
1170 		 */
1171 		mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
1172 		ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
1173 			(MSGMARKNEXT|MSGNOTMARKNEXT));
1174 		if (mark != 0 && bp == stp->sd_mark) {
1175 			if (rflg) {
1176 				putback(stp, q, bp, pri);
1177 				goto oops;
1178 			}
1179 			mark |= _LASTMARK;
1180 			stp->sd_mark = NULL;
1181 		}
1182 		if ((stp->sd_flag & STRDELIM) && (bp->b_flag & MSGDELIM))
1183 			delim = 1;
1184 		mutex_exit(&stp->sd_lock);
1185 
1186 		if (STREAM_NEEDSERVICE(stp))
1187 			stream_runservice(stp);
1188 
1189 		type = bp->b_datap->db_type;
1190 
1191 		switch (type) {
1192 
1193 		case M_DATA:
1194 ismdata:
1195 			if (msgnodata(bp)) {
1196 				if (mark || delim) {
1197 					freemsg(bp);
1198 				} else if (rflg) {
1199 
1200 					/*
1201 					 * If already read data put zero
1202 					 * length message back on queue else
1203 					 * free msg and return 0.
1204 					 */
1205 					bp->b_band = pri;
1206 					mutex_enter(&stp->sd_lock);
1207 					putback(stp, q, bp, pri);
1208 					mutex_exit(&stp->sd_lock);
1209 				} else {
1210 					freemsg(bp);
1211 				}
1212 				error =  0;
1213 				goto oops1;
1214 			}
1215 
1216 			rflg = 1;
1217 			waitflag |= NOINTR;
1218 			bp = struiocopyout(bp, uiop, &error);
1219 			if (error != 0)
1220 				goto oops1;
1221 
1222 			mutex_enter(&stp->sd_lock);
1223 			if (bp) {
1224 				/*
1225 				 * Have remaining data in message.
1226 				 * Free msg if in discard mode.
1227 				 */
1228 				if (stp->sd_read_opt & RD_MSGDIS) {
1229 					freemsg(bp);
1230 				} else {
1231 					bp->b_band = pri;
1232 					if ((mark & _LASTMARK) &&
1233 					    (stp->sd_mark == NULL))
1234 						stp->sd_mark = bp;
1235 					bp->b_flag |= mark & ~_LASTMARK;
1236 					if (delim)
1237 						bp->b_flag |= MSGDELIM;
1238 					if (msgnodata(bp))
1239 						freemsg(bp);
1240 					else
1241 						putback(stp, q, bp, pri);
1242 				}
1243 			} else {
1244 				/*
1245 				 * Consumed the complete message.
1246 				 * Move the MSG*MARKNEXT information
1247 				 * to the stream head just in case
1248 				 * the read queue becomes empty.
1249 				 *
1250 				 * If the stream head was at the mark
1251 				 * (STRATMARK) before we dropped sd_lock above
1252 				 * and some data was consumed then we have
1253 				 * moved past the mark thus STRATMARK is
1254 				 * cleared. However, if a message arrived in
1255 				 * strrput during the copyout above causing
1256 				 * STRATMARK to be set we can not clear that
1257 				 * flag.
1258 				 */
1259 				if (mark &
1260 				    (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
1261 					if (mark & MSGMARKNEXT) {
1262 						stp->sd_flag &= ~STRNOTATMARK;
1263 						stp->sd_flag |= STRATMARK;
1264 					} else if (mark & MSGNOTMARKNEXT) {
1265 						stp->sd_flag &= ~STRATMARK;
1266 						stp->sd_flag |= STRNOTATMARK;
1267 					} else {
1268 						stp->sd_flag &=
1269 						    ~(STRATMARK|STRNOTATMARK);
1270 					}
1271 				} else if (rflg && (old_sd_flag & STRATMARK)) {
1272 					stp->sd_flag &= ~STRATMARK;
1273 				}
1274 			}
1275 
1276 			/*
1277 			 * Check for signal messages at the front of the read
1278 			 * queue and generate the signal(s) if appropriate.
1279 			 * The only signal that can be on queue is M_SIG at
1280 			 * this point.
1281 			 */
1282 			while ((((bp = q->q_first)) != NULL) &&
1283 				(bp->b_datap->db_type == M_SIG)) {
1284 				bp = getq_noenab(q);
1285 				/*
1286 				 * sd_lock is held so the content of the
1287 				 * read queue can not change.
1288 				 */
1289 				ASSERT(bp != NULL &&
1290 					bp->b_datap->db_type == M_SIG);
1291 				strsignal_nolock(stp, *bp->b_rptr,
1292 					(int32_t)bp->b_band);
1293 				mutex_exit(&stp->sd_lock);
1294 				freemsg(bp);
1295 				if (STREAM_NEEDSERVICE(stp))
1296 					stream_runservice(stp);
1297 				mutex_enter(&stp->sd_lock);
1298 			}
1299 
1300 			if ((uiop->uio_resid == 0) || (mark & _LASTMARK) ||
1301 			    delim ||
1302 			    (stp->sd_read_opt & (RD_MSGDIS|RD_MSGNODIS))) {
1303 				goto oops;
1304 			}
1305 			continue;
1306 
1307 		case M_SIG:
1308 			strsignal(stp, *bp->b_rptr, (int32_t)bp->b_band);
1309 			freemsg(bp);
1310 			mutex_enter(&stp->sd_lock);
1311 			continue;
1312 
1313 		case M_PROTO:
1314 		case M_PCPROTO:
1315 			/*
1316 			 * Only data messages are readable.
1317 			 * Any others generate an error, unless
1318 			 * RD_PROTDIS or RD_PROTDAT is set.
1319 			 */
1320 			if (stp->sd_read_opt & RD_PROTDAT) {
1321 				for (nbp = bp; nbp; nbp = nbp->b_next) {
1322 				    if ((nbp->b_datap->db_type == M_PROTO) ||
1323 					(nbp->b_datap->db_type == M_PCPROTO))
1324 					nbp->b_datap->db_type = M_DATA;
1325 				    else
1326 					break;
1327 				}
1328 				/*
1329 				 * clear stream head hi pri flag based on
1330 				 * first message
1331 				 */
1332 				if (type == M_PCPROTO) {
1333 					mutex_enter(&stp->sd_lock);
1334 					stp->sd_flag &= ~STRPRI;
1335 					mutex_exit(&stp->sd_lock);
1336 				}
1337 				goto ismdata;
1338 			} else if (stp->sd_read_opt & RD_PROTDIS) {
1339 				/*
1340 				 * discard non-data messages
1341 				 */
1342 				while (bp &&
1343 				    ((bp->b_datap->db_type == M_PROTO) ||
1344 				    (bp->b_datap->db_type == M_PCPROTO))) {
1345 					nbp = unlinkb(bp);
1346 					freeb(bp);
1347 					bp = nbp;
1348 				}
1349 				/*
1350 				 * clear stream head hi pri flag based on
1351 				 * first message
1352 				 */
1353 				if (type == M_PCPROTO) {
1354 					mutex_enter(&stp->sd_lock);
1355 					stp->sd_flag &= ~STRPRI;
1356 					mutex_exit(&stp->sd_lock);
1357 				}
1358 				if (bp) {
1359 					bp->b_band = pri;
1360 					goto ismdata;
1361 				} else {
1362 					break;
1363 				}
1364 			}
1365 			/* FALLTHRU */
1366 		case M_PASSFP:
1367 			if ((bp->b_datap->db_type == M_PASSFP) &&
1368 			    (stp->sd_read_opt & RD_PROTDIS)) {
1369 				freemsg(bp);
1370 				break;
1371 			}
1372 			mutex_enter(&stp->sd_lock);
1373 			putback(stp, q, bp, pri);
1374 			mutex_exit(&stp->sd_lock);
1375 			if (rflg == 0)
1376 				error = EBADMSG;
1377 			goto oops1;
1378 
1379 		default:
1380 			/*
1381 			 * Garbage on stream head read queue.
1382 			 */
1383 			cmn_err(CE_WARN, "bad %x found at stream head\n",
1384 				bp->b_datap->db_type);
1385 			freemsg(bp);
1386 			goto oops1;
1387 		}
1388 		mutex_enter(&stp->sd_lock);
1389 	}
1390 oops:
1391 	mutex_exit(&stp->sd_lock);
1392 oops1:
1393 	qbackenable(q, pri);
1394 	return (error);
1395 #undef	_LASTMARK
1396 }
1397 
1398 /*
1399  * Default processing of M_PROTO/M_PCPROTO messages.
1400  * Determine which wakeups and signals are needed.
1401  * This can be replaced by a user-specified procedure for kernel users
1402  * of STREAMS.
1403  */
1404 /* ARGSUSED */
1405 mblk_t *
1406 strrput_proto(vnode_t *vp, mblk_t *mp,
1407     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1408     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1409 {
1410 	*wakeups = RSLEEP;
1411 	*allmsgsigs = 0;
1412 
1413 	switch (mp->b_datap->db_type) {
1414 	case M_PROTO:
1415 		if (mp->b_band == 0) {
1416 			*firstmsgsigs = S_INPUT | S_RDNORM;
1417 			*pollwakeups = POLLIN | POLLRDNORM;
1418 		} else {
1419 			*firstmsgsigs = S_INPUT | S_RDBAND;
1420 			*pollwakeups = POLLIN | POLLRDBAND;
1421 		}
1422 		break;
1423 	case M_PCPROTO:
1424 		*firstmsgsigs = S_HIPRI;
1425 		*pollwakeups = POLLPRI;
1426 		break;
1427 	}
1428 	return (mp);
1429 }
1430 
1431 /*
1432  * Default processing of everything but M_DATA, M_PROTO, M_PCPROTO and
1433  * M_PASSFP 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_misc(vnode_t *vp, mblk_t *mp,
1441     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1442     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1443 {
1444 	*wakeups = 0;
1445 	*firstmsgsigs = 0;
1446 	*allmsgsigs = 0;
1447 	*pollwakeups = 0;
1448 	return (mp);
1449 }
1450 
1451 /*
1452  * Stream read put procedure.  Called from downstream driver/module
1453  * with messages for the stream head.  Data, protocol, and in-stream
1454  * signal messages are placed on the queue, others are handled directly.
1455  */
1456 int
1457 strrput(queue_t *q, mblk_t *bp)
1458 {
1459 	struct stdata	*stp;
1460 	ulong_t		rput_opt;
1461 	strwakeup_t	wakeups;
1462 	strsigset_t	firstmsgsigs;	/* Signals if first message on queue */
1463 	strsigset_t	allmsgsigs;	/* Signals for all messages */
1464 	strsigset_t	signals;	/* Signals events to generate */
1465 	strpollset_t	pollwakeups;
1466 	mblk_t		*nextbp;
1467 	uchar_t		band = 0;
1468 	int		hipri_sig;
1469 
1470 	stp = (struct stdata *)q->q_ptr;
1471 	/*
1472 	 * Use rput_opt for optimized access to the SR_ flags except
1473 	 * SR_POLLIN. That flag has to be checked under sd_lock since it
1474 	 * is modified by strpoll().
1475 	 */
1476 	rput_opt = stp->sd_rput_opt;
1477 
1478 	ASSERT(qclaimed(q));
1479 	TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_ENTER,
1480 		"strrput called with message type:q %p bp %p", q, bp);
1481 
1482 	/*
1483 	 * Perform initial processing and pass to the parameterized functions.
1484 	 */
1485 	ASSERT(bp->b_next == NULL);
1486 
1487 	switch (bp->b_datap->db_type) {
1488 	case M_DATA:
1489 		/*
1490 		 * sockfs is the only consumer of STREOF and when it is set,
1491 		 * it implies that the receiver is not interested in receiving
1492 		 * any more data, hence the mblk is freed to prevent unnecessary
1493 		 * message queueing at the stream head.
1494 		 */
1495 		if (stp->sd_flag == STREOF) {
1496 			freemsg(bp);
1497 			return (0);
1498 		}
1499 		if ((rput_opt & SR_IGN_ZEROLEN) &&
1500 		    bp->b_rptr == bp->b_wptr && msgnodata(bp)) {
1501 			/*
1502 			 * Ignore zero-length M_DATA messages. These might be
1503 			 * generated by some transports.
1504 			 * The zero-length M_DATA messages, even if they
1505 			 * are ignored, should effect the atmark tracking and
1506 			 * should wake up a thread sleeping in strwaitmark.
1507 			 */
1508 			mutex_enter(&stp->sd_lock);
1509 			if (bp->b_flag & MSGMARKNEXT) {
1510 				/*
1511 				 * Record the position of the mark either
1512 				 * in q_last or in STRATMARK.
1513 				 */
1514 				if (q->q_last != NULL) {
1515 					q->q_last->b_flag &= ~MSGNOTMARKNEXT;
1516 					q->q_last->b_flag |= MSGMARKNEXT;
1517 				} else {
1518 					stp->sd_flag &= ~STRNOTATMARK;
1519 					stp->sd_flag |= STRATMARK;
1520 				}
1521 			} else if (bp->b_flag & MSGNOTMARKNEXT) {
1522 				/*
1523 				 * Record that this is not the position of
1524 				 * the mark either in q_last or in
1525 				 * STRNOTATMARK.
1526 				 */
1527 				if (q->q_last != NULL) {
1528 					q->q_last->b_flag &= ~MSGMARKNEXT;
1529 					q->q_last->b_flag |= MSGNOTMARKNEXT;
1530 				} else {
1531 					stp->sd_flag &= ~STRATMARK;
1532 					stp->sd_flag |= STRNOTATMARK;
1533 				}
1534 			}
1535 			if (stp->sd_flag & RSLEEP) {
1536 				stp->sd_flag &= ~RSLEEP;
1537 				cv_broadcast(&q->q_wait);
1538 			}
1539 			mutex_exit(&stp->sd_lock);
1540 			freemsg(bp);
1541 			return (0);
1542 		}
1543 		wakeups = RSLEEP;
1544 		if (bp->b_band == 0) {
1545 			firstmsgsigs = S_INPUT | S_RDNORM;
1546 			pollwakeups = POLLIN | POLLRDNORM;
1547 		} else {
1548 			firstmsgsigs = S_INPUT | S_RDBAND;
1549 			pollwakeups = POLLIN | POLLRDBAND;
1550 		}
1551 		if (rput_opt & SR_SIGALLDATA)
1552 			allmsgsigs = firstmsgsigs;
1553 		else
1554 			allmsgsigs = 0;
1555 
1556 		mutex_enter(&stp->sd_lock);
1557 		if ((rput_opt & SR_CONSOL_DATA) &&
1558 		    (bp->b_flag & (MSGMARK|MSGDELIM)) == 0) {
1559 			/*
1560 			 * Consolidate on M_DATA message onto an M_DATA,
1561 			 * M_PROTO, or M_PCPROTO by merging it with q_last.
1562 			 * The consolidation does not take place if
1563 			 * the old message is marked with either of the
1564 			 * marks or the delim flag or if the new
1565 			 * message is marked with MSGMARK. The MSGMARK
1566 			 * check is needed to handle the odd semantics of
1567 			 * MSGMARK where essentially the whole message
1568 			 * is to be treated as marked.
1569 			 * Carry any MSGMARKNEXT  and MSGNOTMARKNEXT from the
1570 			 * new message to the front of the b_cont chain.
1571 			 */
1572 			mblk_t *lbp;
1573 
1574 			lbp = q->q_last;
1575 			if (lbp != NULL &&
1576 			    (lbp->b_datap->db_type == M_DATA ||
1577 			    lbp->b_datap->db_type == M_PROTO ||
1578 			    lbp->b_datap->db_type == M_PCPROTO) &&
1579 			    !(lbp->b_flag & (MSGDELIM|MSGMARK|
1580 			    MSGMARKNEXT))) {
1581 				rmvq_noenab(q, lbp);
1582 				/*
1583 				 * The first message in the b_cont list
1584 				 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
1585 				 * We need to handle the case where we
1586 				 * are appending
1587 				 *
1588 				 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
1589 				 * 2) a MSGMARKNEXT to a plain message.
1590 				 * 3) a MSGNOTMARKNEXT to a plain message
1591 				 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
1592 				 *    message.
1593 				 *
1594 				 * Thus we never append a MSGMARKNEXT or
1595 				 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
1596 				 */
1597 				if (bp->b_flag & MSGMARKNEXT) {
1598 					lbp->b_flag |= MSGMARKNEXT;
1599 					lbp->b_flag &= ~MSGNOTMARKNEXT;
1600 					bp->b_flag &= ~MSGMARKNEXT;
1601 				} else if (bp->b_flag & MSGNOTMARKNEXT) {
1602 					lbp->b_flag |= MSGNOTMARKNEXT;
1603 					bp->b_flag &= ~MSGNOTMARKNEXT;
1604 				}
1605 
1606 				linkb(lbp, bp);
1607 				bp = lbp;
1608 				/*
1609 				 * The new message logically isn't the first
1610 				 * even though the q_first check below thinks
1611 				 * it is. Clear the firstmsgsigs to make it
1612 				 * not appear to be first.
1613 				 */
1614 				firstmsgsigs = 0;
1615 			}
1616 		}
1617 		break;
1618 
1619 	case M_PASSFP:
1620 		wakeups = RSLEEP;
1621 		allmsgsigs = 0;
1622 		if (bp->b_band == 0) {
1623 			firstmsgsigs = S_INPUT | S_RDNORM;
1624 			pollwakeups = POLLIN | POLLRDNORM;
1625 		} else {
1626 			firstmsgsigs = S_INPUT | S_RDBAND;
1627 			pollwakeups = POLLIN | POLLRDBAND;
1628 		}
1629 		mutex_enter(&stp->sd_lock);
1630 		break;
1631 
1632 	case M_PROTO:
1633 	case M_PCPROTO:
1634 		ASSERT(stp->sd_rprotofunc != NULL);
1635 		bp = (stp->sd_rprotofunc)(stp->sd_vnode, bp,
1636 			&wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1637 #define	ALLSIG	(S_INPUT|S_HIPRI|S_OUTPUT|S_MSG|S_ERROR|S_HANGUP|S_RDNORM|\
1638 		S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)
1639 #define	ALLPOLL	(POLLIN|POLLPRI|POLLOUT|POLLRDNORM|POLLWRNORM|POLLRDBAND|\
1640 		POLLWRBAND)
1641 
1642 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1643 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1644 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1645 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1646 
1647 		mutex_enter(&stp->sd_lock);
1648 		break;
1649 
1650 	default:
1651 		ASSERT(stp->sd_rmiscfunc != NULL);
1652 		bp = (stp->sd_rmiscfunc)(stp->sd_vnode, bp,
1653 			&wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1654 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1655 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1656 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1657 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1658 #undef	ALLSIG
1659 #undef	ALLPOLL
1660 		mutex_enter(&stp->sd_lock);
1661 		break;
1662 	}
1663 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1664 
1665 	/* By default generate superset of signals */
1666 	signals = (firstmsgsigs | allmsgsigs);
1667 
1668 	/*
1669 	 * The  proto and misc functions can return multiple messages
1670 	 * as a b_next chain. Such messages are processed separately.
1671 	 */
1672 one_more:
1673 	hipri_sig = 0;
1674 	if (bp == NULL) {
1675 		nextbp = NULL;
1676 	} else {
1677 		nextbp = bp->b_next;
1678 		bp->b_next = NULL;
1679 
1680 		switch (bp->b_datap->db_type) {
1681 		case M_PCPROTO:
1682 			/*
1683 			 * Only one priority protocol message is allowed at the
1684 			 * stream head at a time.
1685 			 */
1686 			if (stp->sd_flag & STRPRI) {
1687 				TRACE_0(TR_FAC_STREAMS_FR, TR_STRRPUT_PROTERR,
1688 				    "M_PCPROTO already at head");
1689 				freemsg(bp);
1690 				mutex_exit(&stp->sd_lock);
1691 				goto done;
1692 			}
1693 			stp->sd_flag |= STRPRI;
1694 			hipri_sig = 1;
1695 			/* FALLTHRU */
1696 		case M_DATA:
1697 		case M_PROTO:
1698 		case M_PASSFP:
1699 			band = bp->b_band;
1700 			/*
1701 			 * Marking doesn't work well when messages
1702 			 * are marked in more than one band.  We only
1703 			 * remember the last message received, even if
1704 			 * it is placed on the queue ahead of other
1705 			 * marked messages.
1706 			 */
1707 			if (bp->b_flag & MSGMARK)
1708 				stp->sd_mark = bp;
1709 			(void) putq(q, bp);
1710 
1711 			/*
1712 			 * If message is a PCPROTO message, always use
1713 			 * firstmsgsigs to determine if a signal should be
1714 			 * sent as strrput is the only place to send
1715 			 * signals for PCPROTO. Other messages are based on
1716 			 * the STRGETINPROG flag. The flag determines if
1717 			 * strrput or (k)strgetmsg will be responsible for
1718 			 * sending the signals, in the firstmsgsigs case.
1719 			 */
1720 			if ((hipri_sig == 1) ||
1721 			    (((stp->sd_flag & STRGETINPROG) == 0) &&
1722 			    (q->q_first == bp)))
1723 				signals = (firstmsgsigs | allmsgsigs);
1724 			else
1725 				signals = allmsgsigs;
1726 			break;
1727 
1728 		default:
1729 			mutex_exit(&stp->sd_lock);
1730 			(void) strrput_nondata(q, bp);
1731 			mutex_enter(&stp->sd_lock);
1732 			break;
1733 		}
1734 	}
1735 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1736 	/*
1737 	 * Wake sleeping read/getmsg and cancel deferred wakeup
1738 	 */
1739 	if (wakeups & RSLEEP)
1740 		stp->sd_wakeq &= ~RSLEEP;
1741 
1742 	wakeups &= stp->sd_flag;
1743 	if (wakeups & RSLEEP) {
1744 		stp->sd_flag &= ~RSLEEP;
1745 		cv_broadcast(&q->q_wait);
1746 	}
1747 	if (wakeups & WSLEEP) {
1748 		stp->sd_flag &= ~WSLEEP;
1749 		cv_broadcast(&_WR(q)->q_wait);
1750 	}
1751 
1752 	if (pollwakeups != 0) {
1753 		if (pollwakeups == (POLLIN | POLLRDNORM)) {
1754 			/*
1755 			 * Can't use rput_opt since it was not
1756 			 * read when sd_lock was held and SR_POLLIN is changed
1757 			 * by strpoll() under sd_lock.
1758 			 */
1759 			if (!(stp->sd_rput_opt & SR_POLLIN))
1760 				goto no_pollwake;
1761 			stp->sd_rput_opt &= ~SR_POLLIN;
1762 		}
1763 		mutex_exit(&stp->sd_lock);
1764 		pollwakeup(&stp->sd_pollist, pollwakeups);
1765 		mutex_enter(&stp->sd_lock);
1766 	}
1767 no_pollwake:
1768 
1769 	/*
1770 	 * strsendsig can handle multiple signals with a
1771 	 * single call.
1772 	 */
1773 	if (stp->sd_sigflags & signals)
1774 		strsendsig(stp->sd_siglist, signals, band, 0);
1775 	mutex_exit(&stp->sd_lock);
1776 
1777 
1778 done:
1779 	if (nextbp == NULL)
1780 		return (0);
1781 
1782 	/*
1783 	 * Any signals were handled the first time.
1784 	 * Wakeups and pollwakeups are redone to avoid any race
1785 	 * conditions - all the messages are not queued until the
1786 	 * last message has been processed by strrput.
1787 	 */
1788 	bp = nextbp;
1789 	signals = firstmsgsigs = allmsgsigs = 0;
1790 	mutex_enter(&stp->sd_lock);
1791 	goto one_more;
1792 }
1793 
1794 static void
1795 log_dupioc(queue_t *rq, mblk_t *bp)
1796 {
1797 	queue_t *wq, *qp;
1798 	char *modnames, *mnp, *dname;
1799 	size_t maxmodstr;
1800 	boolean_t islast;
1801 
1802 	/*
1803 	 * Allocate a buffer large enough to hold the names of nstrpush modules
1804 	 * and one driver, with spaces between and NUL terminator.  If we can't
1805 	 * get memory, then we'll just log the driver name.
1806 	 */
1807 	maxmodstr = nstrpush * (FMNAMESZ + 1);
1808 	mnp = modnames = kmem_alloc(maxmodstr, KM_NOSLEEP);
1809 
1810 	/* march down write side to print log message down to the driver */
1811 	wq = WR(rq);
1812 
1813 	/* make sure q_next doesn't shift around while we're grabbing data */
1814 	claimstr(wq);
1815 	qp = wq->q_next;
1816 	do {
1817 		if ((dname = qp->q_qinfo->qi_minfo->mi_idname) == NULL)
1818 			dname = "?";
1819 		islast = !SAMESTR(qp) || qp->q_next == NULL;
1820 		if (modnames == NULL) {
1821 			/*
1822 			 * If we don't have memory, then get the driver name in
1823 			 * the log where we can see it.  Note that memory
1824 			 * pressure is a possible cause of these sorts of bugs.
1825 			 */
1826 			if (islast) {
1827 				modnames = dname;
1828 				maxmodstr = 0;
1829 			}
1830 		} else {
1831 			mnp += snprintf(mnp, FMNAMESZ + 1, "%s", dname);
1832 			if (!islast)
1833 				*mnp++ = ' ';
1834 		}
1835 		qp = qp->q_next;
1836 	} while (!islast);
1837 	releasestr(wq);
1838 	/* Cannot happen unless stream head is corrupt. */
1839 	ASSERT(modnames != NULL);
1840 	(void) strlog(rq->q_qinfo->qi_minfo->mi_idnum, 0, 1,
1841 	    SL_CONSOLE|SL_TRACE|SL_ERROR,
1842 	    "Warning: stream %p received duplicate %X M_IOC%s; module list: %s",
1843 	    rq->q_ptr, ((struct iocblk *)bp->b_rptr)->ioc_cmd,
1844 	    (DB_TYPE(bp) == M_IOCACK ? "ACK" : "NAK"), modnames);
1845 	if (maxmodstr != 0)
1846 		kmem_free(modnames, maxmodstr);
1847 }
1848 
1849 int
1850 strrput_nondata(queue_t *q, mblk_t *bp)
1851 {
1852 	struct stdata *stp;
1853 	struct iocblk *iocbp;
1854 	struct stroptions *sop;
1855 	struct copyreq *reqp;
1856 	struct copyresp *resp;
1857 	unsigned char bpri;
1858 	unsigned char  flushed_already = 0;
1859 
1860 	stp = (struct stdata *)q->q_ptr;
1861 
1862 	ASSERT(!(stp->sd_flag & STPLEX));
1863 	ASSERT(qclaimed(q));
1864 
1865 	switch (bp->b_datap->db_type) {
1866 	case M_ERROR:
1867 		/*
1868 		 * An error has occurred downstream, the errno is in the first
1869 		 * bytes of the message.
1870 		 */
1871 		if ((bp->b_wptr - bp->b_rptr) == 2) {	/* New flavor */
1872 			unsigned char rw = 0;
1873 
1874 			mutex_enter(&stp->sd_lock);
1875 			if (*bp->b_rptr != NOERROR) {	/* read error */
1876 				if (*bp->b_rptr != 0) {
1877 					if (stp->sd_flag & STRDERR)
1878 						flushed_already |= FLUSHR;
1879 					stp->sd_flag |= STRDERR;
1880 					rw |= FLUSHR;
1881 				} else {
1882 					stp->sd_flag &= ~STRDERR;
1883 				}
1884 				stp->sd_rerror = *bp->b_rptr;
1885 			}
1886 			bp->b_rptr++;
1887 			if (*bp->b_rptr != NOERROR) {	/* write error */
1888 				if (*bp->b_rptr != 0) {
1889 					if (stp->sd_flag & STWRERR)
1890 						flushed_already |= FLUSHW;
1891 					stp->sd_flag |= STWRERR;
1892 					rw |= FLUSHW;
1893 				} else {
1894 					stp->sd_flag &= ~STWRERR;
1895 				}
1896 				stp->sd_werror = *bp->b_rptr;
1897 			}
1898 			if (rw) {
1899 				TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_WAKE,
1900 					"strrput cv_broadcast:q %p, bp %p",
1901 					q, bp);
1902 				cv_broadcast(&q->q_wait); /* readers */
1903 				cv_broadcast(&_WR(q)->q_wait); /* writers */
1904 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
1905 
1906 				mutex_exit(&stp->sd_lock);
1907 				pollwakeup(&stp->sd_pollist, POLLERR);
1908 				mutex_enter(&stp->sd_lock);
1909 
1910 				if (stp->sd_sigflags & S_ERROR)
1911 					strsendsig(stp->sd_siglist, S_ERROR, 0,
1912 					    ((rw & FLUSHR) ? stp->sd_rerror :
1913 					    stp->sd_werror));
1914 				mutex_exit(&stp->sd_lock);
1915 				/*
1916 				 * Send the M_FLUSH only
1917 				 * for the first M_ERROR
1918 				 * message on the stream
1919 				 */
1920 				if (flushed_already == rw) {
1921 					freemsg(bp);
1922 					return (0);
1923 				}
1924 
1925 				bp->b_datap->db_type = M_FLUSH;
1926 				*bp->b_rptr = rw;
1927 				bp->b_wptr = bp->b_rptr + 1;
1928 				/*
1929 				 * Protect against the driver
1930 				 * passing up messages after
1931 				 * it has done a qprocsoff
1932 				 */
1933 				if (_OTHERQ(q)->q_next == NULL)
1934 					freemsg(bp);
1935 				else
1936 					qreply(q, bp);
1937 				return (0);
1938 			} else
1939 				mutex_exit(&stp->sd_lock);
1940 		} else if (*bp->b_rptr != 0) {		/* Old flavor */
1941 				if (stp->sd_flag & (STRDERR|STWRERR))
1942 					flushed_already = FLUSHRW;
1943 				mutex_enter(&stp->sd_lock);
1944 				stp->sd_flag |= (STRDERR|STWRERR);
1945 				stp->sd_rerror = *bp->b_rptr;
1946 				stp->sd_werror = *bp->b_rptr;
1947 				TRACE_2(TR_FAC_STREAMS_FR,
1948 					TR_STRRPUT_WAKE2,
1949 					"strrput wakeup #2:q %p, bp %p", q, bp);
1950 				cv_broadcast(&q->q_wait); /* the readers */
1951 				cv_broadcast(&_WR(q)->q_wait); /* the writers */
1952 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
1953 
1954 				mutex_exit(&stp->sd_lock);
1955 				pollwakeup(&stp->sd_pollist, POLLERR);
1956 				mutex_enter(&stp->sd_lock);
1957 
1958 				if (stp->sd_sigflags & S_ERROR)
1959 					strsendsig(stp->sd_siglist, S_ERROR, 0,
1960 					    (stp->sd_werror ? stp->sd_werror :
1961 					    stp->sd_rerror));
1962 				mutex_exit(&stp->sd_lock);
1963 
1964 				/*
1965 				 * Send the M_FLUSH only
1966 				 * for the first M_ERROR
1967 				 * message on the stream
1968 				 */
1969 				if (flushed_already != FLUSHRW) {
1970 					bp->b_datap->db_type = M_FLUSH;
1971 					*bp->b_rptr = FLUSHRW;
1972 					/*
1973 					 * Protect against the driver passing up
1974 					 * messages after it has done a
1975 					 * qprocsoff.
1976 					 */
1977 				if (_OTHERQ(q)->q_next == NULL)
1978 					freemsg(bp);
1979 				else
1980 					qreply(q, bp);
1981 				return (0);
1982 				}
1983 		}
1984 		freemsg(bp);
1985 		return (0);
1986 
1987 	case M_HANGUP:
1988 
1989 		freemsg(bp);
1990 		mutex_enter(&stp->sd_lock);
1991 		stp->sd_werror = ENXIO;
1992 		stp->sd_flag |= STRHUP;
1993 		stp->sd_flag &= ~(WSLEEP|RSLEEP);
1994 
1995 		/*
1996 		 * send signal if controlling tty
1997 		 */
1998 
1999 		if (stp->sd_sidp) {
2000 			prsignal(stp->sd_sidp, SIGHUP);
2001 			if (stp->sd_sidp != stp->sd_pgidp)
2002 				pgsignal(stp->sd_pgidp, SIGTSTP);
2003 		}
2004 
2005 		/*
2006 		 * wake up read, write, and exception pollers and
2007 		 * reset wakeup mechanism.
2008 		 */
2009 		cv_broadcast(&q->q_wait);	/* the readers */
2010 		cv_broadcast(&_WR(q)->q_wait);	/* the writers */
2011 		cv_broadcast(&stp->sd_monitor);	/* the ioctllers */
2012 		strhup(stp);
2013 		mutex_exit(&stp->sd_lock);
2014 		return (0);
2015 
2016 	case M_UNHANGUP:
2017 		freemsg(bp);
2018 		mutex_enter(&stp->sd_lock);
2019 		stp->sd_werror = 0;
2020 		stp->sd_flag &= ~STRHUP;
2021 		mutex_exit(&stp->sd_lock);
2022 		return (0);
2023 
2024 	case M_SIG:
2025 		/*
2026 		 * Someone downstream wants to post a signal.  The
2027 		 * signal to post is contained in the first byte of the
2028 		 * message.  If the message would go on the front of
2029 		 * the queue, send a signal to the process group
2030 		 * (if not SIGPOLL) or to the siglist processes
2031 		 * (SIGPOLL).  If something is already on the queue,
2032 		 * OR if we are delivering a delayed suspend (*sigh*
2033 		 * another "tty" hack) and there's no one sleeping already,
2034 		 * just enqueue the message.
2035 		 */
2036 		mutex_enter(&stp->sd_lock);
2037 		if (q->q_first || (*bp->b_rptr == SIGTSTP &&
2038 		    !(stp->sd_flag & RSLEEP))) {
2039 			(void) putq(q, bp);
2040 			mutex_exit(&stp->sd_lock);
2041 			return (0);
2042 		}
2043 		mutex_exit(&stp->sd_lock);
2044 		/* FALLTHRU */
2045 
2046 	case M_PCSIG:
2047 		/*
2048 		 * Don't enqueue, just post the signal.
2049 		 */
2050 		strsignal(stp, *bp->b_rptr, 0L);
2051 		freemsg(bp);
2052 		return (0);
2053 
2054 	case M_FLUSH:
2055 		/*
2056 		 * Flush queues.  The indication of which queues to flush
2057 		 * is in the first byte of the message.  If the read queue
2058 		 * is specified, then flush it.  If FLUSHBAND is set, just
2059 		 * flush the band specified by the second byte of the message.
2060 		 *
2061 		 * If a module has issued a M_SETOPT to not flush hi
2062 		 * priority messages off of the stream head, then pass this
2063 		 * flag into the flushq code to preserve such messages.
2064 		 */
2065 
2066 		if (*bp->b_rptr & FLUSHR) {
2067 			mutex_enter(&stp->sd_lock);
2068 			if (*bp->b_rptr & FLUSHBAND) {
2069 				ASSERT((bp->b_wptr - bp->b_rptr) >= 2);
2070 				flushband(q, *(bp->b_rptr + 1), FLUSHALL);
2071 			} else
2072 				flushq_common(q, FLUSHALL,
2073 				    stp->sd_read_opt & RFLUSHPCPROT);
2074 			if ((q->q_first == NULL) ||
2075 			    (q->q_first->b_datap->db_type < QPCTL))
2076 				stp->sd_flag &= ~STRPRI;
2077 			else {
2078 				ASSERT(stp->sd_flag & STRPRI);
2079 			}
2080 			mutex_exit(&stp->sd_lock);
2081 		}
2082 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
2083 			*bp->b_rptr &= ~FLUSHR;
2084 			bp->b_flag |= MSGNOLOOP;
2085 			/*
2086 			 * Protect against the driver passing up
2087 			 * messages after it has done a qprocsoff.
2088 			 */
2089 			if (_OTHERQ(q)->q_next == NULL)
2090 				freemsg(bp);
2091 			else
2092 				qreply(q, bp);
2093 			return (0);
2094 		}
2095 		freemsg(bp);
2096 		return (0);
2097 
2098 	case M_IOCACK:
2099 	case M_IOCNAK:
2100 		iocbp = (struct iocblk *)bp->b_rptr;
2101 		/*
2102 		 * If not waiting for ACK or NAK then just free msg.
2103 		 * If incorrect id sequence number then just free msg.
2104 		 * If already have ACK or NAK for user then this is a
2105 		 *    duplicate, display a warning and free the msg.
2106 		 */
2107 		mutex_enter(&stp->sd_lock);
2108 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2109 		    (stp->sd_iocid != iocbp->ioc_id)) {
2110 			/*
2111 			 * If the ACK/NAK is a dup, display a message
2112 			 * Dup is when sd_iocid == ioc_id, and
2113 			 * sd_iocblk == <valid ptr> or -1 (the former
2114 			 * is when an ioctl has been put on the stream
2115 			 * head, but has not yet been consumed, the
2116 			 * later is when it has been consumed).
2117 			 */
2118 			if ((stp->sd_iocid == iocbp->ioc_id) &&
2119 			    (stp->sd_iocblk != NULL)) {
2120 				log_dupioc(q, bp);
2121 			}
2122 			freemsg(bp);
2123 			mutex_exit(&stp->sd_lock);
2124 			return (0);
2125 		}
2126 
2127 		/*
2128 		 * Assign ACK or NAK to user and wake up.
2129 		 */
2130 		stp->sd_iocblk = bp;
2131 		cv_broadcast(&stp->sd_monitor);
2132 		mutex_exit(&stp->sd_lock);
2133 		return (0);
2134 
2135 	case M_COPYIN:
2136 	case M_COPYOUT:
2137 		reqp = (struct copyreq *)bp->b_rptr;
2138 
2139 		/*
2140 		 * If not waiting for ACK or NAK then just fail request.
2141 		 * If already have ACK, NAK, or copy request, then just
2142 		 * fail request.
2143 		 * If incorrect id sequence number then just fail request.
2144 		 */
2145 		mutex_enter(&stp->sd_lock);
2146 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2147 		    (stp->sd_iocid != reqp->cq_id)) {
2148 			if (bp->b_cont) {
2149 				freemsg(bp->b_cont);
2150 				bp->b_cont = NULL;
2151 			}
2152 			bp->b_datap->db_type = M_IOCDATA;
2153 			bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
2154 			resp = (struct copyresp *)bp->b_rptr;
2155 			resp->cp_rval = (caddr_t)1;	/* failure */
2156 			mutex_exit(&stp->sd_lock);
2157 			putnext(stp->sd_wrq, bp);
2158 			return (0);
2159 		}
2160 
2161 		/*
2162 		 * Assign copy request 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_SETOPTS:
2170 		/*
2171 		 * Set stream head options (read option, write offset,
2172 		 * min/max packet size, and/or high/low water marks for
2173 		 * the read side only).
2174 		 */
2175 
2176 		bpri = 0;
2177 		sop = (struct stroptions *)bp->b_rptr;
2178 		mutex_enter(&stp->sd_lock);
2179 		if (sop->so_flags & SO_READOPT) {
2180 			switch (sop->so_readopt & RMODEMASK) {
2181 			case RNORM:
2182 				stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
2183 				break;
2184 
2185 			case RMSGD:
2186 				stp->sd_read_opt =
2187 				    ((stp->sd_read_opt & ~RD_MSGNODIS) |
2188 				    RD_MSGDIS);
2189 				break;
2190 
2191 			case RMSGN:
2192 				stp->sd_read_opt =
2193 				    ((stp->sd_read_opt & ~RD_MSGDIS) |
2194 				    RD_MSGNODIS);
2195 				break;
2196 			}
2197 			switch (sop->so_readopt & RPROTMASK) {
2198 			case RPROTNORM:
2199 				stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
2200 				break;
2201 
2202 			case RPROTDAT:
2203 				stp->sd_read_opt =
2204 				    ((stp->sd_read_opt & ~RD_PROTDIS) |
2205 				    RD_PROTDAT);
2206 				break;
2207 
2208 			case RPROTDIS:
2209 				stp->sd_read_opt =
2210 				    ((stp->sd_read_opt & ~RD_PROTDAT) |
2211 				    RD_PROTDIS);
2212 				break;
2213 			}
2214 			switch (sop->so_readopt & RFLUSHMASK) {
2215 			case RFLUSHPCPROT:
2216 				/*
2217 				 * This sets the stream head to NOT flush
2218 				 * M_PCPROTO messages.
2219 				 */
2220 				stp->sd_read_opt |= RFLUSHPCPROT;
2221 				break;
2222 			}
2223 		}
2224 		if (sop->so_flags & SO_ERROPT) {
2225 			switch (sop->so_erropt & RERRMASK) {
2226 			case RERRNORM:
2227 				stp->sd_flag &= ~STRDERRNONPERSIST;
2228 				break;
2229 			case RERRNONPERSIST:
2230 				stp->sd_flag |= STRDERRNONPERSIST;
2231 				break;
2232 			}
2233 			switch (sop->so_erropt & WERRMASK) {
2234 			case WERRNORM:
2235 				stp->sd_flag &= ~STWRERRNONPERSIST;
2236 				break;
2237 			case WERRNONPERSIST:
2238 				stp->sd_flag |= STWRERRNONPERSIST;
2239 				break;
2240 			}
2241 		}
2242 		if (sop->so_flags & SO_COPYOPT) {
2243 			if (sop->so_copyopt & ZCVMSAFE) {
2244 				stp->sd_copyflag |= STZCVMSAFE;
2245 				stp->sd_copyflag &= ~STZCVMUNSAFE;
2246 			} else if (sop->so_copyopt & ZCVMUNSAFE) {
2247 				stp->sd_copyflag |= STZCVMUNSAFE;
2248 				stp->sd_copyflag &= ~STZCVMSAFE;
2249 			}
2250 
2251 			if (sop->so_copyopt & COPYCACHED) {
2252 				stp->sd_copyflag |= STRCOPYCACHED;
2253 			}
2254 		}
2255 		if (sop->so_flags & SO_WROFF)
2256 			stp->sd_wroff = sop->so_wroff;
2257 		if (sop->so_flags & SO_TAIL)
2258 			stp->sd_tail = sop->so_tail;
2259 		if (sop->so_flags & SO_MINPSZ)
2260 			q->q_minpsz = sop->so_minpsz;
2261 		if (sop->so_flags & SO_MAXPSZ)
2262 			q->q_maxpsz = sop->so_maxpsz;
2263 		if (sop->so_flags & SO_MAXBLK)
2264 			stp->sd_maxblk = sop->so_maxblk;
2265 		if (sop->so_flags & SO_HIWAT) {
2266 		    if (sop->so_flags & SO_BAND) {
2267 			if (strqset(q, QHIWAT, sop->so_band, sop->so_hiwat))
2268 				cmn_err(CE_WARN,
2269 				    "strrput: could not allocate qband\n");
2270 			else
2271 				bpri = sop->so_band;
2272 		    } else {
2273 			q->q_hiwat = sop->so_hiwat;
2274 		    }
2275 		}
2276 		if (sop->so_flags & SO_LOWAT) {
2277 		    if (sop->so_flags & SO_BAND) {
2278 			if (strqset(q, QLOWAT, sop->so_band, sop->so_lowat))
2279 				cmn_err(CE_WARN,
2280 				    "strrput: could not allocate qband\n");
2281 			else
2282 				bpri = sop->so_band;
2283 		    } else {
2284 			q->q_lowat = sop->so_lowat;
2285 		    }
2286 		}
2287 		if (sop->so_flags & SO_MREADON)
2288 			stp->sd_flag |= SNDMREAD;
2289 		if (sop->so_flags & SO_MREADOFF)
2290 			stp->sd_flag &= ~SNDMREAD;
2291 		if (sop->so_flags & SO_NDELON)
2292 			stp->sd_flag |= OLDNDELAY;
2293 		if (sop->so_flags & SO_NDELOFF)
2294 			stp->sd_flag &= ~OLDNDELAY;
2295 		if (sop->so_flags & SO_ISTTY)
2296 			stp->sd_flag |= STRISTTY;
2297 		if (sop->so_flags & SO_ISNTTY)
2298 			stp->sd_flag &= ~STRISTTY;
2299 		if (sop->so_flags & SO_TOSTOP)
2300 			stp->sd_flag |= STRTOSTOP;
2301 		if (sop->so_flags & SO_TONSTOP)
2302 			stp->sd_flag &= ~STRTOSTOP;
2303 		if (sop->so_flags & SO_DELIM)
2304 			stp->sd_flag |= STRDELIM;
2305 		if (sop->so_flags & SO_NODELIM)
2306 			stp->sd_flag &= ~STRDELIM;
2307 
2308 		mutex_exit(&stp->sd_lock);
2309 		freemsg(bp);
2310 
2311 		/* Check backenable in case the water marks changed */
2312 		qbackenable(q, bpri);
2313 		return (0);
2314 
2315 	/*
2316 	 * The following set of cases deal with situations where two stream
2317 	 * heads are connected to each other (twisted streams).  These messages
2318 	 * have no meaning at the stream head.
2319 	 */
2320 	case M_BREAK:
2321 	case M_CTL:
2322 	case M_DELAY:
2323 	case M_START:
2324 	case M_STOP:
2325 	case M_IOCDATA:
2326 	case M_STARTI:
2327 	case M_STOPI:
2328 		freemsg(bp);
2329 		return (0);
2330 
2331 	case M_IOCTL:
2332 		/*
2333 		 * Always NAK this condition
2334 		 * (makes no sense)
2335 		 * If there is one or more threads in the read side
2336 		 * rwnext we have to defer the nacking until that thread
2337 		 * returns (in strget).
2338 		 */
2339 		mutex_enter(&stp->sd_lock);
2340 		if (stp->sd_struiodnak != 0) {
2341 			/*
2342 			 * Defer NAK to the streamhead. Queue at the end
2343 			 * the list.
2344 			 */
2345 			mblk_t *mp = stp->sd_struionak;
2346 
2347 			while (mp && mp->b_next)
2348 				mp = mp->b_next;
2349 			if (mp)
2350 				mp->b_next = bp;
2351 			else
2352 				stp->sd_struionak = bp;
2353 			bp->b_next = NULL;
2354 			mutex_exit(&stp->sd_lock);
2355 			return (0);
2356 		}
2357 		mutex_exit(&stp->sd_lock);
2358 
2359 		bp->b_datap->db_type = M_IOCNAK;
2360 		/*
2361 		 * Protect against the driver passing up
2362 		 * messages after it has done a qprocsoff.
2363 		 */
2364 		if (_OTHERQ(q)->q_next == NULL)
2365 			freemsg(bp);
2366 		else
2367 			qreply(q, bp);
2368 		return (0);
2369 
2370 	default:
2371 #ifdef DEBUG
2372 		cmn_err(CE_WARN,
2373 			"bad message type %x received at stream head\n",
2374 			bp->b_datap->db_type);
2375 #endif
2376 		freemsg(bp);
2377 		return (0);
2378 	}
2379 
2380 	/* NOTREACHED */
2381 }
2382 
2383 /*
2384  * Check if the stream pointed to by `stp' can be written to, and return an
2385  * error code if not.  If `eiohup' is set, then return EIO if STRHUP is set.
2386  * If `sigpipeok' is set and the SW_SIGPIPE option is enabled on the stream,
2387  * then always return EPIPE and send a SIGPIPE to the invoking thread.
2388  */
2389 static int
2390 strwriteable(struct stdata *stp, boolean_t eiohup, boolean_t sigpipeok)
2391 {
2392 	int error;
2393 
2394 	ASSERT(MUTEX_HELD(&stp->sd_lock));
2395 
2396 	/*
2397 	 * For modem support, POSIX states that on writes, EIO should
2398 	 * be returned if the stream has been hung up.
2399 	 */
2400 	if (eiohup && (stp->sd_flag & (STPLEX|STRHUP)) == STRHUP)
2401 		error = EIO;
2402 	else
2403 		error = strgeterr(stp, STRHUP|STPLEX|STWRERR, 0);
2404 
2405 	if (error != 0) {
2406 		if (!(stp->sd_flag & STPLEX) &&
2407 		    (stp->sd_wput_opt & SW_SIGPIPE) && sigpipeok) {
2408 			tsignal(curthread, SIGPIPE);
2409 			error = EPIPE;
2410 		}
2411 	}
2412 
2413 	return (error);
2414 }
2415 
2416 /*
2417  * Copyin and send data down a stream.
2418  * The caller will allocate and copyin any control part that precedes the
2419  * message and pass than in as mctl.
2420  *
2421  * Caller should *not* hold sd_lock.
2422  * When EWOULDBLOCK is returned the caller has to redo the canputnext
2423  * under sd_lock in order to avoid missing a backenabling wakeup.
2424  *
2425  * Use iosize = -1 to not send any M_DATA. iosize = 0 sends zero-length M_DATA.
2426  *
2427  * Set MSG_IGNFLOW in flags to ignore flow control for hipri messages.
2428  * For sync streams we can only ignore flow control by reverting to using
2429  * putnext.
2430  *
2431  * If sd_maxblk is less than *iosize this routine might return without
2432  * transferring all of *iosize. In all cases, on return *iosize will contain
2433  * the amount of data that was transferred.
2434  */
2435 static int
2436 strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
2437     int b_flag, int pri, int flags)
2438 {
2439 	struiod_t uiod;
2440 	mblk_t *mp;
2441 	queue_t *wqp = stp->sd_wrq;
2442 	int error = 0;
2443 	ssize_t count = *iosize;
2444 	cred_t *cr;
2445 
2446 	ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
2447 
2448 	if (uiop != NULL && count >= 0)
2449 		flags |= stp->sd_struiowrq ? STRUIO_POSTPONE : 0;
2450 
2451 	if (!(flags & STRUIO_POSTPONE)) {
2452 		/*
2453 		 * Use regular canputnext, strmakedata, putnext sequence.
2454 		 */
2455 		if (pri == 0) {
2456 			if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2457 				freemsg(mctl);
2458 				return (EWOULDBLOCK);
2459 			}
2460 		} else {
2461 			if (!(flags & MSG_IGNFLOW) && !bcanputnext(wqp, pri)) {
2462 				freemsg(mctl);
2463 				return (EWOULDBLOCK);
2464 			}
2465 		}
2466 
2467 		if ((error = strmakedata(iosize, uiop, stp, flags,
2468 					&mp)) != 0) {
2469 			freemsg(mctl);
2470 			/*
2471 			 * need to change return code to ENOMEM
2472 			 * so that this is not confused with
2473 			 * flow control, EAGAIN.
2474 			 */
2475 
2476 			if (error == EAGAIN)
2477 				return (ENOMEM);
2478 			else
2479 				return (error);
2480 		}
2481 		if (mctl != NULL) {
2482 			if (mctl->b_cont == NULL)
2483 				mctl->b_cont = mp;
2484 			else if (mp != NULL)
2485 				linkb(mctl, mp);
2486 			mp = mctl;
2487 			/*
2488 			 * Note that for interrupt thread, the CRED() is
2489 			 * NULL. Don't bother with the pid either.
2490 			 */
2491 			if ((cr = CRED()) != NULL) {
2492 				mblk_setcred(mp, cr);
2493 				DB_CPID(mp) = curproc->p_pid;
2494 			}
2495 		} else if (mp == NULL)
2496 			return (0);
2497 
2498 		mp->b_flag |= b_flag;
2499 		mp->b_band = (uchar_t)pri;
2500 
2501 		if (flags & MSG_IGNFLOW) {
2502 			/*
2503 			 * XXX Hack: Don't get stuck running service
2504 			 * procedures. This is needed for sockfs when
2505 			 * sending the unbind message out of the rput
2506 			 * procedure - we don't want a put procedure
2507 			 * to run service procedures.
2508 			 */
2509 			putnext(wqp, mp);
2510 		} else {
2511 			stream_willservice(stp);
2512 			putnext(wqp, mp);
2513 			stream_runservice(stp);
2514 		}
2515 		return (0);
2516 	}
2517 	/*
2518 	 * Stream supports rwnext() for the write side.
2519 	 */
2520 	if ((error = strmakedata(iosize, uiop, stp, flags, &mp)) != 0) {
2521 		freemsg(mctl);
2522 		/*
2523 		 * map EAGAIN to ENOMEM since EAGAIN means "flow controlled".
2524 		 */
2525 		return (error == EAGAIN ? ENOMEM : error);
2526 	}
2527 	if (mctl != NULL) {
2528 		if (mctl->b_cont == NULL)
2529 			mctl->b_cont = mp;
2530 		else if (mp != NULL)
2531 			linkb(mctl, mp);
2532 		mp = mctl;
2533 		/*
2534 		 * Note that for interrupt thread, the CRED() is
2535 		 * NULL.  Don't bother with the pid either.
2536 		 */
2537 		if ((cr = CRED()) != NULL) {
2538 			mblk_setcred(mp, cr);
2539 			DB_CPID(mp) = curproc->p_pid;
2540 		}
2541 	} else if (mp == NULL) {
2542 		return (0);
2543 	}
2544 
2545 	mp->b_flag |= b_flag;
2546 	mp->b_band = (uchar_t)pri;
2547 
2548 	(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
2549 		sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
2550 	uiod.d_uio.uio_offset = 0;
2551 	uiod.d_mp = mp;
2552 	error = rwnext(wqp, &uiod);
2553 	if (! uiod.d_mp) {
2554 		uioskip(uiop, *iosize);
2555 		return (error);
2556 	}
2557 	ASSERT(mp == uiod.d_mp);
2558 	if (error == EINVAL) {
2559 		/*
2560 		 * The stream plumbing must have changed while
2561 		 * we were away, so just turn off rwnext()s.
2562 		 */
2563 		error = 0;
2564 	} else if (error == EBUSY || error == EWOULDBLOCK) {
2565 		/*
2566 		 * Couldn't enter a perimeter or took a page fault,
2567 		 * so fall-back to putnext().
2568 		 */
2569 		error = 0;
2570 	} else {
2571 		freemsg(mp);
2572 		return (error);
2573 	}
2574 	/* Have to check canput before consuming data from the uio */
2575 	if (pri == 0) {
2576 		if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2577 			freemsg(mp);
2578 			return (EWOULDBLOCK);
2579 		}
2580 	} else {
2581 		if (!bcanputnext(wqp, pri) && !(flags & MSG_IGNFLOW)) {
2582 			freemsg(mp);
2583 			return (EWOULDBLOCK);
2584 		}
2585 	}
2586 	ASSERT(mp == uiod.d_mp);
2587 	/* Copyin data from the uio */
2588 	if ((error = struioget(wqp, mp, &uiod, 0)) != 0) {
2589 		freemsg(mp);
2590 		return (error);
2591 	}
2592 	uioskip(uiop, *iosize);
2593 	if (flags & MSG_IGNFLOW) {
2594 		/*
2595 		 * XXX Hack: Don't get stuck running service procedures.
2596 		 * This is needed for sockfs when sending the unbind message
2597 		 * out of the rput procedure - we don't want a put procedure
2598 		 * to run service procedures.
2599 		 */
2600 		putnext(wqp, mp);
2601 	} else {
2602 		stream_willservice(stp);
2603 		putnext(wqp, mp);
2604 		stream_runservice(stp);
2605 	}
2606 	return (0);
2607 }
2608 
2609 /*
2610  * Write attempts to break the write request into messages conforming
2611  * with the minimum and maximum packet sizes set downstream.
2612  *
2613  * Write will not block if downstream queue is full and
2614  * O_NDELAY is set, otherwise it will block waiting for the queue to get room.
2615  *
2616  * A write of zero bytes gets packaged into a zero length message and sent
2617  * downstream like any other message.
2618  *
2619  * If buffers of the requested sizes are not available, the write will
2620  * sleep until the buffers become available.
2621  *
2622  * Write (if specified) will supply a write offset in a message if it
2623  * makes sense. This can be specified by downstream modules as part of
2624  * a M_SETOPTS message.  Write will not supply the write offset if it
2625  * cannot supply any data in a buffer.  In other words, write will never
2626  * send down an empty packet due to a write offset.
2627  */
2628 /* ARGSUSED2 */
2629 int
2630 strwrite(struct vnode *vp, struct uio *uiop, cred_t *crp)
2631 {
2632 	return (strwrite_common(vp, uiop, crp, 0));
2633 }
2634 
2635 /* ARGSUSED2 */
2636 int
2637 strwrite_common(struct vnode *vp, struct uio *uiop, cred_t *crp, int wflag)
2638 {
2639 	struct stdata *stp;
2640 	struct queue *wqp;
2641 	ssize_t rmin, rmax;
2642 	ssize_t iosize;
2643 	int waitflag;
2644 	int tempmode;
2645 	int error = 0;
2646 	int b_flag;
2647 
2648 	ASSERT(vp->v_stream);
2649 	stp = vp->v_stream;
2650 
2651 	mutex_enter(&stp->sd_lock);
2652 
2653 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
2654 		mutex_exit(&stp->sd_lock);
2655 		return (error);
2656 	}
2657 
2658 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
2659 		error = strwriteable(stp, B_TRUE, B_TRUE);
2660 		if (error != 0) {
2661 			mutex_exit(&stp->sd_lock);
2662 			return (error);
2663 		}
2664 	}
2665 
2666 	mutex_exit(&stp->sd_lock);
2667 
2668 	wqp = stp->sd_wrq;
2669 
2670 	/* get these values from them cached in the stream head */
2671 	rmin = stp->sd_qn_minpsz;
2672 	rmax = stp->sd_qn_maxpsz;
2673 
2674 	/*
2675 	 * Check the min/max packet size constraints.  If min packet size
2676 	 * is non-zero, the write cannot be split into multiple messages
2677 	 * and still guarantee the size constraints.
2678 	 */
2679 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_IN, "strwrite in:q %p", wqp);
2680 
2681 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
2682 	if (rmax == 0) {
2683 		return (0);
2684 	}
2685 	if (rmin > 0) {
2686 		if (uiop->uio_resid < rmin) {
2687 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2688 				"strwrite out:q %p out %d error %d",
2689 				wqp, 0, ERANGE);
2690 			return (ERANGE);
2691 		}
2692 		if ((rmax != INFPSZ) && (uiop->uio_resid > rmax)) {
2693 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2694 				"strwrite out:q %p out %d error %d",
2695 				wqp, 1, ERANGE);
2696 			return (ERANGE);
2697 		}
2698 	}
2699 
2700 	/*
2701 	 * Do until count satisfied or error.
2702 	 */
2703 	waitflag = WRITEWAIT | wflag;
2704 	if (stp->sd_flag & OLDNDELAY)
2705 		tempmode = uiop->uio_fmode & ~FNDELAY;
2706 	else
2707 		tempmode = uiop->uio_fmode;
2708 
2709 	if (rmax == INFPSZ)
2710 		rmax = uiop->uio_resid;
2711 
2712 	/*
2713 	 * Note that tempmode does not get used in strput/strmakedata
2714 	 * but only in strwaitq. The other routines use uio_fmode
2715 	 * unmodified.
2716 	 */
2717 
2718 	/* LINTED: constant in conditional context */
2719 	while (1) {	/* breaks when uio_resid reaches zero */
2720 		/*
2721 		 * Determine the size of the next message to be
2722 		 * packaged.  May have to break write into several
2723 		 * messages based on max packet size.
2724 		 */
2725 		iosize = MIN(uiop->uio_resid, rmax);
2726 
2727 		/*
2728 		 * Put block downstream when flow control allows it.
2729 		 */
2730 		if ((stp->sd_flag & STRDELIM) && (uiop->uio_resid == iosize))
2731 			b_flag = MSGDELIM;
2732 		else
2733 			b_flag = 0;
2734 
2735 		for (;;) {
2736 			int done = 0;
2737 
2738 			error = strput(stp, NULL, uiop, &iosize, b_flag,
2739 				0, 0);
2740 			if (error == 0)
2741 				break;
2742 			if (error != EWOULDBLOCK)
2743 				goto out;
2744 
2745 			mutex_enter(&stp->sd_lock);
2746 			/*
2747 			 * Check for a missed wakeup.
2748 			 * Needed since strput did not hold sd_lock across
2749 			 * the canputnext.
2750 			 */
2751 			if (canputnext(wqp)) {
2752 				/* Try again */
2753 				mutex_exit(&stp->sd_lock);
2754 				continue;
2755 			}
2756 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAIT,
2757 				"strwrite wait:q %p wait", wqp);
2758 			if ((error = strwaitq(stp, waitflag, (ssize_t)0,
2759 			    tempmode, -1, &done)) != 0 || done) {
2760 				mutex_exit(&stp->sd_lock);
2761 				if ((vp->v_type == VFIFO) &&
2762 				    (uiop->uio_fmode & FNDELAY) &&
2763 				    (error == EAGAIN))
2764 					error = 0;
2765 				goto out;
2766 			}
2767 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAKE,
2768 				"strwrite wake:q %p awakes", wqp);
2769 			if ((error = i_straccess(stp, JCWRITE)) != 0) {
2770 				mutex_exit(&stp->sd_lock);
2771 				goto out;
2772 			}
2773 			mutex_exit(&stp->sd_lock);
2774 		}
2775 		waitflag |= NOINTR;
2776 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRWRITE_RESID,
2777 			"strwrite resid:q %p uiop %p", wqp, uiop);
2778 		if (uiop->uio_resid) {
2779 			/* Recheck for errors - needed for sockets */
2780 			if ((stp->sd_wput_opt & SW_RECHECK_ERR) &&
2781 			    (stp->sd_flag & (STWRERR|STRHUP|STPLEX))) {
2782 				mutex_enter(&stp->sd_lock);
2783 				error = strwriteable(stp, B_FALSE, B_TRUE);
2784 				mutex_exit(&stp->sd_lock);
2785 				if (error != 0)
2786 					return (error);
2787 			}
2788 			continue;
2789 		}
2790 		break;
2791 	}
2792 out:
2793 	/*
2794 	 * For historical reasons, applications expect EAGAIN when a data
2795 	 * mblk_t cannot be allocated, so change ENOMEM back to EAGAIN.
2796 	 */
2797 	if (error == ENOMEM)
2798 		error = EAGAIN;
2799 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2800 		"strwrite out:q %p out %d error %d", wqp, 2, error);
2801 	return (error);
2802 }
2803 
2804 /*
2805  * Stream head write service routine.
2806  * Its job is to wake up any sleeping writers when a queue
2807  * downstream needs data (part of the flow control in putq and getq).
2808  * It also must wake anyone sleeping on a poll().
2809  * For stream head right below mux module, it must also invoke put procedure
2810  * of next downstream module.
2811  */
2812 int
2813 strwsrv(queue_t *q)
2814 {
2815 	struct stdata *stp;
2816 	queue_t *tq;
2817 	qband_t *qbp;
2818 	int i;
2819 	qband_t *myqbp;
2820 	int isevent;
2821 	unsigned char	qbf[NBAND];	/* band flushing backenable flags */
2822 
2823 	TRACE_1(TR_FAC_STREAMS_FR,
2824 		TR_STRWSRV, "strwsrv:q %p", q);
2825 	stp = (struct stdata *)q->q_ptr;
2826 	ASSERT(qclaimed(q));
2827 	mutex_enter(&stp->sd_lock);
2828 	ASSERT(!(stp->sd_flag & STPLEX));
2829 
2830 	if (stp->sd_flag & WSLEEP) {
2831 		stp->sd_flag &= ~WSLEEP;
2832 		cv_broadcast(&q->q_wait);
2833 	}
2834 	mutex_exit(&stp->sd_lock);
2835 
2836 	/* The other end of a stream pipe went away. */
2837 	if ((tq = q->q_next) == NULL) {
2838 		return (0);
2839 	}
2840 
2841 	/* Find the next module forward that has a service procedure */
2842 	claimstr(q);
2843 	tq = q->q_nfsrv;
2844 	ASSERT(tq != NULL);
2845 
2846 	if ((q->q_flag & QBACK)) {
2847 		if ((tq->q_flag & QFULL)) {
2848 			mutex_enter(QLOCK(tq));
2849 			if (!(tq->q_flag & QFULL)) {
2850 				mutex_exit(QLOCK(tq));
2851 				goto wakeup;
2852 			}
2853 			/*
2854 			 * The queue must have become full again. Set QWANTW
2855 			 * again so strwsrv will be back enabled when
2856 			 * the queue becomes non-full next time.
2857 			 */
2858 			tq->q_flag |= QWANTW;
2859 			mutex_exit(QLOCK(tq));
2860 		} else {
2861 		wakeup:
2862 			pollwakeup(&stp->sd_pollist, POLLWRNORM);
2863 			mutex_enter(&stp->sd_lock);
2864 			if (stp->sd_sigflags & S_WRNORM)
2865 				strsendsig(stp->sd_siglist, S_WRNORM, 0, 0);
2866 			mutex_exit(&stp->sd_lock);
2867 		}
2868 	}
2869 
2870 	isevent = 0;
2871 	i = 1;
2872 	bzero((caddr_t)qbf, NBAND);
2873 	mutex_enter(QLOCK(tq));
2874 	if ((myqbp = q->q_bandp) != NULL)
2875 		for (qbp = tq->q_bandp; qbp && myqbp; qbp = qbp->qb_next) {
2876 			ASSERT(myqbp);
2877 			if ((myqbp->qb_flag & QB_BACK)) {
2878 				if (qbp->qb_flag & QB_FULL) {
2879 					/*
2880 					 * The band must have become full again.
2881 					 * Set QB_WANTW again so strwsrv will
2882 					 * be back enabled when the band becomes
2883 					 * non-full next time.
2884 					 */
2885 					qbp->qb_flag |= QB_WANTW;
2886 				} else {
2887 					isevent = 1;
2888 					qbf[i] = 1;
2889 				}
2890 			}
2891 			myqbp = myqbp->qb_next;
2892 			i++;
2893 		}
2894 	mutex_exit(QLOCK(tq));
2895 
2896 	if (isevent) {
2897 	    for (i = tq->q_nband; i; i--) {
2898 		if (qbf[i]) {
2899 			pollwakeup(&stp->sd_pollist, POLLWRBAND);
2900 			mutex_enter(&stp->sd_lock);
2901 			if (stp->sd_sigflags & S_WRBAND)
2902 				strsendsig(stp->sd_siglist, S_WRBAND,
2903 					(uchar_t)i, 0);
2904 			mutex_exit(&stp->sd_lock);
2905 		}
2906 	    }
2907 	}
2908 
2909 	releasestr(q);
2910 	return (0);
2911 }
2912 
2913 /*
2914  * Special case of strcopyin/strcopyout for copying
2915  * struct strioctl that can deal with both data
2916  * models.
2917  */
2918 
2919 #ifdef	_LP64
2920 
2921 static int
2922 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
2923 {
2924 	struct	strioctl32 strioc32;
2925 	struct	strioctl *striocp;
2926 
2927 	if (copyflag & U_TO_K) {
2928 		ASSERT((copyflag & K_TO_K) == 0);
2929 
2930 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
2931 			if (copyin(from, &strioc32, sizeof (strioc32)))
2932 				return (EFAULT);
2933 
2934 			striocp = (struct strioctl *)to;
2935 			striocp->ic_cmd	= strioc32.ic_cmd;
2936 			striocp->ic_timout = strioc32.ic_timout;
2937 			striocp->ic_len	= strioc32.ic_len;
2938 			striocp->ic_dp	= (char *)(uintptr_t)strioc32.ic_dp;
2939 
2940 		} else { /* NATIVE data model */
2941 			if (copyin(from, to, sizeof (struct strioctl))) {
2942 				return (EFAULT);
2943 			} else {
2944 				return (0);
2945 			}
2946 		}
2947 	} else {
2948 		ASSERT(copyflag & K_TO_K);
2949 		bcopy(from, to, sizeof (struct strioctl));
2950 	}
2951 	return (0);
2952 }
2953 
2954 static int
2955 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
2956 {
2957 	struct	strioctl32 strioc32;
2958 	struct	strioctl *striocp;
2959 
2960 	if (copyflag & U_TO_K) {
2961 		ASSERT((copyflag & K_TO_K) == 0);
2962 
2963 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
2964 			striocp = (struct strioctl *)from;
2965 			strioc32.ic_cmd	= striocp->ic_cmd;
2966 			strioc32.ic_timout = striocp->ic_timout;
2967 			strioc32.ic_len	= striocp->ic_len;
2968 			strioc32.ic_dp	= (caddr32_t)(uintptr_t)striocp->ic_dp;
2969 			ASSERT((char *)(uintptr_t)strioc32.ic_dp ==
2970 			    striocp->ic_dp);
2971 
2972 			if (copyout(&strioc32, to, sizeof (strioc32)))
2973 				return (EFAULT);
2974 
2975 		} else { /* NATIVE data model */
2976 			if (copyout(from, to, sizeof (struct strioctl))) {
2977 				return (EFAULT);
2978 			} else {
2979 				return (0);
2980 			}
2981 		}
2982 	} else {
2983 		ASSERT(copyflag & K_TO_K);
2984 		bcopy(from, to, sizeof (struct strioctl));
2985 	}
2986 	return (0);
2987 }
2988 
2989 #else	/* ! _LP64 */
2990 
2991 /* ARGSUSED2 */
2992 static int
2993 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
2994 {
2995 	return (strcopyin(from, to, sizeof (struct strioctl), copyflag));
2996 }
2997 
2998 /* ARGSUSED2 */
2999 static int
3000 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3001 {
3002 	return (strcopyout(from, to, sizeof (struct strioctl), copyflag));
3003 }
3004 
3005 #endif	/* _LP64 */
3006 
3007 /*
3008  * Determine type of job control semantics expected by user.  The
3009  * possibilities are:
3010  *	JCREAD	- Behaves like read() on fd; send SIGTTIN
3011  *	JCWRITE	- Behaves like write() on fd; send SIGTTOU if TOSTOP set
3012  *	JCSETP	- Sets a value in the stream; send SIGTTOU, ignore TOSTOP
3013  *	JCGETP	- Gets a value in the stream; no signals.
3014  * See straccess in strsubr.c for usage of these values.
3015  *
3016  * This routine also returns -1 for I_STR as a special case; the
3017  * caller must call again with the real ioctl number for
3018  * classification.
3019  */
3020 static int
3021 job_control_type(int cmd)
3022 {
3023 	switch (cmd) {
3024 	case I_STR:
3025 		return (-1);
3026 
3027 	case I_RECVFD:
3028 	case I_E_RECVFD:
3029 		return (JCREAD);
3030 
3031 	case I_FDINSERT:
3032 	case I_SENDFD:
3033 		return (JCWRITE);
3034 
3035 	case TCSETA:
3036 	case TCSETAW:
3037 	case TCSETAF:
3038 	case TCSBRK:
3039 	case TCXONC:
3040 	case TCFLSH:
3041 	case TCDSET:	/* Obsolete */
3042 	case TIOCSWINSZ:
3043 	case TCSETS:
3044 	case TCSETSW:
3045 	case TCSETSF:
3046 	case TIOCSETD:
3047 	case TIOCHPCL:
3048 	case TIOCSETP:
3049 	case TIOCSETN:
3050 	case TIOCEXCL:
3051 	case TIOCNXCL:
3052 	case TIOCFLUSH:
3053 	case TIOCSETC:
3054 	case TIOCLBIS:
3055 	case TIOCLBIC:
3056 	case TIOCLSET:
3057 	case TIOCSBRK:
3058 	case TIOCCBRK:
3059 	case TIOCSDTR:
3060 	case TIOCCDTR:
3061 	case TIOCSLTC:
3062 	case TIOCSTOP:
3063 	case TIOCSTART:
3064 	case TIOCSTI:
3065 	case TIOCSPGRP:
3066 	case TIOCMSET:
3067 	case TIOCMBIS:
3068 	case TIOCMBIC:
3069 	case TIOCREMOTE:
3070 	case TIOCSIGNAL:
3071 	case LDSETT:
3072 	case LDSMAP:	/* Obsolete */
3073 	case DIOCSETP:
3074 	case I_FLUSH:
3075 	case I_SRDOPT:
3076 	case I_SETSIG:
3077 	case I_SWROPT:
3078 	case I_FLUSHBAND:
3079 	case I_SETCLTIME:
3080 	case I_SERROPT:
3081 	case I_ESETSIG:
3082 	case FIONBIO:
3083 	case FIOASYNC:
3084 	case FIOSETOWN:
3085 	case JBOOT:	/* Obsolete */
3086 	case JTERM:	/* Obsolete */
3087 	case JTIMOM:	/* Obsolete */
3088 	case JZOMBOOT:	/* Obsolete */
3089 	case JAGENT:	/* Obsolete */
3090 	case JTRUN:	/* Obsolete */
3091 	case JXTPROTO:	/* Obsolete */
3092 	case TIOCSETLD:
3093 		return (JCSETP);
3094 	}
3095 
3096 	return (JCGETP);
3097 }
3098 
3099 /*
3100  * ioctl for streams
3101  */
3102 int
3103 strioctl(struct vnode *vp, int cmd, intptr_t arg, int flag, int copyflag,
3104     cred_t *crp, int *rvalp)
3105 {
3106 	struct stdata *stp;
3107 	struct strioctl strioc;
3108 	struct uio uio;
3109 	struct iovec iov;
3110 	int access;
3111 	mblk_t *mp;
3112 	int error = 0;
3113 	int done = 0;
3114 	ssize_t	rmin, rmax;
3115 	queue_t *wrq;
3116 	queue_t *rdq;
3117 	boolean_t kioctl = B_FALSE;
3118 
3119 	if (flag & FKIOCTL) {
3120 		copyflag = K_TO_K;
3121 		kioctl = B_TRUE;
3122 	}
3123 	ASSERT(vp->v_stream);
3124 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
3125 	stp = vp->v_stream;
3126 
3127 	TRACE_3(TR_FAC_STREAMS_FR, TR_IOCTL_ENTER,
3128 		"strioctl:stp %p cmd %X arg %lX", stp, cmd, arg);
3129 
3130 #ifdef C2_AUDIT
3131 	if (audit_active)
3132 		audit_strioctl(vp, cmd, arg, flag, copyflag, crp, rvalp);
3133 #endif
3134 
3135 	/*
3136 	 * If the copy is kernel to kernel, make sure that the FNATIVE
3137 	 * flag is set.  After this it would be a serious error to have
3138 	 * no model flag.
3139 	 */
3140 	if (copyflag == K_TO_K)
3141 		flag = (flag & ~FMODELS) | FNATIVE;
3142 
3143 	ASSERT((flag & FMODELS) != 0);
3144 
3145 	wrq = stp->sd_wrq;
3146 	rdq = _RD(wrq);
3147 
3148 	access = job_control_type(cmd);
3149 
3150 	/* We should never see these here, should be handled by iwscn */
3151 	if (cmd == SRIOCSREDIR || cmd == SRIOCISREDIR)
3152 		return (EINVAL);
3153 
3154 	mutex_enter(&stp->sd_lock);
3155 	if ((access != -1) && ((error = i_straccess(stp, access)) != 0)) {
3156 		mutex_exit(&stp->sd_lock);
3157 		return (error);
3158 	}
3159 	mutex_exit(&stp->sd_lock);
3160 
3161 	/*
3162 	 * Check for sgttyb-related ioctls first, and complain as
3163 	 * necessary.
3164 	 */
3165 	switch (cmd) {
3166 	case TIOCGETP:
3167 	case TIOCSETP:
3168 	case TIOCSETN:
3169 		if (sgttyb_handling >= 2 && !sgttyb_complaint) {
3170 			sgttyb_complaint = B_TRUE;
3171 			cmn_err(CE_NOTE,
3172 			    "application used obsolete TIOC[GS]ET");
3173 		}
3174 		if (sgttyb_handling >= 3) {
3175 			tsignal(curthread, SIGSYS);
3176 			return (EIO);
3177 		}
3178 		break;
3179 	}
3180 
3181 	mutex_enter(&stp->sd_lock);
3182 
3183 	switch (cmd) {
3184 	case I_RECVFD:
3185 	case I_E_RECVFD:
3186 	case I_PEEK:
3187 	case I_NREAD:
3188 	case FIONREAD:
3189 	case FIORDCHK:
3190 	case I_ATMARK:
3191 	case FIONBIO:
3192 	case FIOASYNC:
3193 		if (stp->sd_flag & (STRDERR|STPLEX)) {
3194 			error = strgeterr(stp, STRDERR|STPLEX, 0);
3195 			if (error != 0) {
3196 				mutex_exit(&stp->sd_lock);
3197 				return (error);
3198 			}
3199 		}
3200 		break;
3201 
3202 	default:
3203 		if (stp->sd_flag & (STRDERR|STWRERR|STPLEX)) {
3204 			error = strgeterr(stp, STRDERR|STWRERR|STPLEX, 0);
3205 			if (error != 0) {
3206 				mutex_exit(&stp->sd_lock);
3207 				return (error);
3208 			}
3209 		}
3210 	}
3211 
3212 	mutex_exit(&stp->sd_lock);
3213 
3214 	switch (cmd) {
3215 	default:
3216 		/*
3217 		 * The stream head has hardcoded knowledge of a
3218 		 * miscellaneous collection of terminal-, keyboard- and
3219 		 * mouse-related ioctls, enumerated below.  This hardcoded
3220 		 * knowledge allows the stream head to automatically
3221 		 * convert transparent ioctl requests made by userland
3222 		 * programs into I_STR ioctls which many old STREAMS
3223 		 * modules and drivers require.
3224 		 *
3225 		 * No new ioctls should ever be added to this list.
3226 		 * Instead, the STREAMS module or driver should be written
3227 		 * to either handle transparent ioctls or require any
3228 		 * userland programs to use I_STR ioctls (by returning
3229 		 * EINVAL to any transparent ioctl requests).
3230 		 *
3231 		 * More importantly, removing ioctls from this list should
3232 		 * be done with the utmost care, since our STREAMS modules
3233 		 * and drivers *count* on the stream head performing this
3234 		 * conversion, and thus may panic while processing
3235 		 * transparent ioctl request for one of these ioctls (keep
3236 		 * in mind that third party modules and drivers may have
3237 		 * similar problems).
3238 		 */
3239 		if (((cmd & IOCTYPE) == LDIOC) ||
3240 		    ((cmd & IOCTYPE) == tIOC) ||
3241 		    ((cmd & IOCTYPE) == TIOC) ||
3242 		    ((cmd & IOCTYPE) == KIOC) ||
3243 		    ((cmd & IOCTYPE) == MSIOC) ||
3244 		    ((cmd & IOCTYPE) == VUIOC)) {
3245 			/*
3246 			 * The ioctl is a tty ioctl - set up strioc buffer
3247 			 * and call strdoioctl() to do the work.
3248 			 */
3249 			if (stp->sd_flag & STRHUP)
3250 				return (ENXIO);
3251 			strioc.ic_cmd = cmd;
3252 			strioc.ic_timout = INFTIM;
3253 
3254 			switch (cmd) {
3255 
3256 			case TCXONC:
3257 			case TCSBRK:
3258 			case TCFLSH:
3259 			case TCDSET:
3260 				{
3261 				int native_arg = (int)arg;
3262 				strioc.ic_len = sizeof (int);
3263 				strioc.ic_dp = (char *)&native_arg;
3264 				return (strdoioctl(stp, &strioc, flag,
3265 				    K_TO_K, crp, rvalp));
3266 				}
3267 
3268 			case TCSETA:
3269 			case TCSETAW:
3270 			case TCSETAF:
3271 				strioc.ic_len = sizeof (struct termio);
3272 				strioc.ic_dp = (char *)arg;
3273 				return (strdoioctl(stp, &strioc, flag,
3274 					copyflag, crp, rvalp));
3275 
3276 			case TCSETS:
3277 			case TCSETSW:
3278 			case TCSETSF:
3279 				strioc.ic_len = sizeof (struct termios);
3280 				strioc.ic_dp = (char *)arg;
3281 				return (strdoioctl(stp, &strioc, flag,
3282 					copyflag, crp, rvalp));
3283 
3284 			case LDSETT:
3285 				strioc.ic_len = sizeof (struct termcb);
3286 				strioc.ic_dp = (char *)arg;
3287 				return (strdoioctl(stp, &strioc, flag,
3288 					copyflag, crp, rvalp));
3289 
3290 			case TIOCSETP:
3291 				strioc.ic_len = sizeof (struct sgttyb);
3292 				strioc.ic_dp = (char *)arg;
3293 				return (strdoioctl(stp, &strioc, flag,
3294 					copyflag, crp, rvalp));
3295 
3296 			case TIOCSTI:
3297 				if ((flag & FREAD) == 0 &&
3298 				    secpolicy_sti(crp) != 0) {
3299 					return (EPERM);
3300 				}
3301 				mutex_enter(&stp->sd_lock);
3302 				mutex_enter(&curproc->p_splock);
3303 				if (stp->sd_sidp != curproc->p_sessp->s_sidp &&
3304 				    secpolicy_sti(crp) != 0) {
3305 					mutex_exit(&curproc->p_splock);
3306 					mutex_exit(&stp->sd_lock);
3307 					return (EACCES);
3308 				}
3309 				mutex_exit(&curproc->p_splock);
3310 				mutex_exit(&stp->sd_lock);
3311 
3312 				strioc.ic_len = sizeof (char);
3313 				strioc.ic_dp = (char *)arg;
3314 				return (strdoioctl(stp, &strioc, flag,
3315 					copyflag, crp, rvalp));
3316 
3317 			case TIOCSWINSZ:
3318 				strioc.ic_len = sizeof (struct winsize);
3319 				strioc.ic_dp = (char *)arg;
3320 				return (strdoioctl(stp, &strioc, flag,
3321 					copyflag, crp, rvalp));
3322 
3323 			case TIOCSSIZE:
3324 				strioc.ic_len = sizeof (struct ttysize);
3325 				strioc.ic_dp = (char *)arg;
3326 				return (strdoioctl(stp, &strioc, flag,
3327 					copyflag, crp, rvalp));
3328 
3329 			case TIOCSSOFTCAR:
3330 			case KIOCTRANS:
3331 			case KIOCTRANSABLE:
3332 			case KIOCCMD:
3333 			case KIOCSDIRECT:
3334 			case KIOCSCOMPAT:
3335 			case KIOCSKABORTEN:
3336 			case KIOCSRPTDELAY:
3337 			case KIOCSRPTRATE:
3338 			case VUIDSFORMAT:
3339 			case TIOCSPPS:
3340 				strioc.ic_len = sizeof (int);
3341 				strioc.ic_dp = (char *)arg;
3342 				return (strdoioctl(stp, &strioc, flag,
3343 					copyflag, crp, rvalp));
3344 
3345 			case KIOCSETKEY:
3346 			case KIOCGETKEY:
3347 				strioc.ic_len = sizeof (struct kiockey);
3348 				strioc.ic_dp = (char *)arg;
3349 				return (strdoioctl(stp, &strioc, flag,
3350 					copyflag, crp, rvalp));
3351 
3352 			case KIOCSKEY:
3353 			case KIOCGKEY:
3354 				strioc.ic_len = sizeof (struct kiockeymap);
3355 				strioc.ic_dp = (char *)arg;
3356 				return (strdoioctl(stp, &strioc, flag,
3357 					copyflag, crp, rvalp));
3358 
3359 			case KIOCSLED:
3360 				/* arg is a pointer to char */
3361 				strioc.ic_len = sizeof (char);
3362 				strioc.ic_dp = (char *)arg;
3363 				return (strdoioctl(stp, &strioc, flag,
3364 					copyflag, crp, rvalp));
3365 
3366 			case MSIOSETPARMS:
3367 				strioc.ic_len = sizeof (Ms_parms);
3368 				strioc.ic_dp = (char *)arg;
3369 				return (strdoioctl(stp, &strioc, flag,
3370 					copyflag, crp, rvalp));
3371 
3372 			case VUIDSADDR:
3373 			case VUIDGADDR:
3374 				strioc.ic_len = sizeof (struct vuid_addr_probe);
3375 				strioc.ic_dp = (char *)arg;
3376 				return (strdoioctl(stp, &strioc, flag,
3377 					copyflag, crp, rvalp));
3378 
3379 			/*
3380 			 * These M_IOCTL's don't require any data to be sent
3381 			 * downstream, and the driver will allocate and link
3382 			 * on its own mblk_t upon M_IOCACK -- thus we set
3383 			 * ic_len to zero and set ic_dp to arg so we know
3384 			 * where to copyout to later.
3385 			 */
3386 			case TIOCGSOFTCAR:
3387 			case TIOCGWINSZ:
3388 			case TIOCGSIZE:
3389 			case KIOCGTRANS:
3390 			case KIOCGTRANSABLE:
3391 			case KIOCTYPE:
3392 			case KIOCGDIRECT:
3393 			case KIOCGCOMPAT:
3394 			case KIOCLAYOUT:
3395 			case KIOCGLED:
3396 			case MSIOGETPARMS:
3397 			case MSIOBUTTONS:
3398 			case VUIDGFORMAT:
3399 			case TIOCGPPS:
3400 			case TIOCGPPSEV:
3401 			case TCGETA:
3402 			case TCGETS:
3403 			case LDGETT:
3404 			case TIOCGETP:
3405 			case KIOCGRPTDELAY:
3406 			case KIOCGRPTRATE:
3407 				strioc.ic_len = 0;
3408 				strioc.ic_dp = (char *)arg;
3409 				return (strdoioctl(stp, &strioc, flag,
3410 					copyflag, crp, rvalp));
3411 			}
3412 		}
3413 
3414 		/*
3415 		 * Unknown cmd - send it down as a transparent ioctl.
3416 		 */
3417 		strioc.ic_cmd = cmd;
3418 		strioc.ic_timout = INFTIM;
3419 		strioc.ic_len = TRANSPARENT;
3420 		strioc.ic_dp = (char *)&arg;
3421 
3422 		return (strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp));
3423 
3424 	case I_STR:
3425 		/*
3426 		 * Stream ioctl.  Read in an strioctl buffer from the user
3427 		 * along with any data specified and send it downstream.
3428 		 * Strdoioctl will wait allow only one ioctl message at
3429 		 * a time, and waits for the acknowledgement.
3430 		 */
3431 
3432 		if (stp->sd_flag & STRHUP)
3433 			return (ENXIO);
3434 
3435 		error = strcopyin_strioctl((void *)arg, &strioc, flag,
3436 		    copyflag);
3437 		if (error != 0)
3438 			return (error);
3439 
3440 		if ((strioc.ic_len < 0) || (strioc.ic_timout < -1))
3441 			return (EINVAL);
3442 
3443 		access = job_control_type(strioc.ic_cmd);
3444 		mutex_enter(&stp->sd_lock);
3445 		if ((access != -1) &&
3446 		    ((error = i_straccess(stp, access)) != 0)) {
3447 			mutex_exit(&stp->sd_lock);
3448 			return (error);
3449 		}
3450 		mutex_exit(&stp->sd_lock);
3451 
3452 		/*
3453 		 * The I_STR facility provides a trap door for malicious
3454 		 * code to send down bogus streamio(7I) ioctl commands to
3455 		 * unsuspecting STREAMS modules and drivers which expect to
3456 		 * only get these messages from the stream head.
3457 		 * Explicitly prohibit any streamio ioctls which can be
3458 		 * passed downstream by the stream head.  Note that we do
3459 		 * not block all streamio ioctls because the ioctl
3460 		 * numberspace is not well managed and thus it's possible
3461 		 * that a module or driver's ioctl numbers may accidentally
3462 		 * collide with them.
3463 		 */
3464 		switch (strioc.ic_cmd) {
3465 		case I_LINK:
3466 		case I_PLINK:
3467 		case I_UNLINK:
3468 		case I_PUNLINK:
3469 		case _I_GETPEERCRED:
3470 		case _I_PLINK_LH:
3471 			return (EINVAL);
3472 		}
3473 
3474 		error = strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp);
3475 		if (error == 0) {
3476 			error = strcopyout_strioctl(&strioc, (void *)arg,
3477 			    flag, copyflag);
3478 		}
3479 		return (error);
3480 
3481 	case I_NREAD:
3482 		/*
3483 		 * Return number of bytes of data in first message
3484 		 * in queue in "arg" and return the number of messages
3485 		 * in queue in return value.
3486 		 */
3487 	    {
3488 		size_t	size;
3489 		int	retval;
3490 		int	count = 0;
3491 
3492 		mutex_enter(QLOCK(rdq));
3493 
3494 		size = msgdsize(rdq->q_first);
3495 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3496 			count++;
3497 
3498 		mutex_exit(QLOCK(rdq));
3499 		if (stp->sd_struiordq) {
3500 			infod_t infod;
3501 
3502 			infod.d_cmd = INFOD_COUNT;
3503 			infod.d_count = 0;
3504 			if (count == 0) {
3505 				infod.d_cmd |= INFOD_FIRSTBYTES;
3506 				infod.d_bytes = 0;
3507 			}
3508 			infod.d_res = 0;
3509 			(void) infonext(rdq, &infod);
3510 			count += infod.d_count;
3511 			if (infod.d_res & INFOD_FIRSTBYTES)
3512 				size = infod.d_bytes;
3513 		}
3514 
3515 		/*
3516 		 * Drop down from size_t to the "int" required by the
3517 		 * interface.  Cap at INT_MAX.
3518 		 */
3519 		retval = MIN(size, INT_MAX);
3520 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3521 		    copyflag);
3522 		if (!error)
3523 			*rvalp = count;
3524 		return (error);
3525 	    }
3526 
3527 	case FIONREAD:
3528 		/*
3529 		 * Return number of bytes of data in all data messages
3530 		 * in queue in "arg".
3531 		 */
3532 	    {
3533 		size_t	size = 0;
3534 		int	retval;
3535 
3536 		mutex_enter(QLOCK(rdq));
3537 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3538 			size += msgdsize(mp);
3539 		mutex_exit(QLOCK(rdq));
3540 
3541 		if (stp->sd_struiordq) {
3542 			infod_t infod;
3543 
3544 			infod.d_cmd = INFOD_BYTES;
3545 			infod.d_res = 0;
3546 			infod.d_bytes = 0;
3547 			(void) infonext(rdq, &infod);
3548 			size += infod.d_bytes;
3549 		}
3550 
3551 		/*
3552 		 * Drop down from size_t to the "int" required by the
3553 		 * interface.  Cap at INT_MAX.
3554 		 */
3555 		retval = MIN(size, INT_MAX);
3556 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3557 		    copyflag);
3558 
3559 		*rvalp = 0;
3560 		return (error);
3561 	    }
3562 	case FIORDCHK:
3563 		/*
3564 		 * FIORDCHK does not use arg value (like FIONREAD),
3565 		 * instead a count is returned. I_NREAD value may
3566 		 * not be accurate but safe. The real thing to do is
3567 		 * to add the msgdsizes of all data  messages until
3568 		 * a non-data message.
3569 		 */
3570 	    {
3571 		size_t size = 0;
3572 
3573 		mutex_enter(QLOCK(rdq));
3574 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3575 			size += msgdsize(mp);
3576 		mutex_exit(QLOCK(rdq));
3577 
3578 		if (stp->sd_struiordq) {
3579 			infod_t infod;
3580 
3581 			infod.d_cmd = INFOD_BYTES;
3582 			infod.d_res = 0;
3583 			infod.d_bytes = 0;
3584 			(void) infonext(rdq, &infod);
3585 			size += infod.d_bytes;
3586 		}
3587 
3588 		/*
3589 		 * Since ioctl returns an int, and memory sizes under
3590 		 * LP64 may not fit, we return INT_MAX if the count was
3591 		 * actually greater.
3592 		 */
3593 		*rvalp = MIN(size, INT_MAX);
3594 		return (0);
3595 	    }
3596 
3597 	case I_FIND:
3598 		/*
3599 		 * Get module name.
3600 		 */
3601 	    {
3602 		char mname[FMNAMESZ + 1];
3603 		queue_t *q;
3604 
3605 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3606 		    mname, FMNAMESZ + 1, NULL);
3607 		if (error)
3608 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3609 
3610 		/*
3611 		 * Return EINVAL if we're handed a bogus module name.
3612 		 */
3613 		if (fmodsw_find(mname, FMODSW_LOAD) == NULL) {
3614 			TRACE_0(TR_FAC_STREAMS_FR,
3615 				TR_I_CANT_FIND, "couldn't I_FIND");
3616 			return (EINVAL);
3617 		}
3618 
3619 		*rvalp = 0;
3620 
3621 		/* Look downstream to see if module is there. */
3622 		claimstr(stp->sd_wrq);
3623 		for (q = stp->sd_wrq->q_next; q; q = q->q_next) {
3624 			if (q->q_flag&QREADR) {
3625 				q = NULL;
3626 				break;
3627 			}
3628 			if (strcmp(mname, q->q_qinfo->qi_minfo->mi_idname) == 0)
3629 				break;
3630 		}
3631 		releasestr(stp->sd_wrq);
3632 
3633 		*rvalp = (q ? 1 : 0);
3634 		return (error);
3635 	    }
3636 
3637 	case I_PUSH:
3638 	case __I_PUSH_NOCTTY:
3639 		/*
3640 		 * Push a module.
3641 		 * For the case __I_PUSH_NOCTTY push a module but
3642 		 * do not allocate controlling tty. See bugid 4025044
3643 		 */
3644 
3645 	    {
3646 		char mname[FMNAMESZ + 1];
3647 		fmodsw_impl_t *fp;
3648 		dev_t dummydev;
3649 
3650 		if (stp->sd_flag & STRHUP)
3651 			return (ENXIO);
3652 
3653 		/*
3654 		 * Get module name and look up in fmodsw.
3655 		 */
3656 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3657 		    mname, FMNAMESZ + 1, NULL);
3658 		if (error)
3659 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3660 
3661 		if ((fp = fmodsw_find(mname, FMODSW_HOLD | FMODSW_LOAD)) ==
3662 		    NULL)
3663 			return (EINVAL);
3664 
3665 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_PUSH,
3666 		    "I_PUSH:fp %p stp %p", fp, stp);
3667 
3668 		if (error = strstartplumb(stp, flag, cmd)) {
3669 			fmodsw_rele(fp);
3670 			return (error);
3671 		}
3672 
3673 		/*
3674 		 * See if any more modules can be pushed on this stream.
3675 		 * Note that this check must be done after strstartplumb()
3676 		 * since otherwise multiple threads issuing I_PUSHes on
3677 		 * the same stream will be able to exceed nstrpush.
3678 		 */
3679 		mutex_enter(&stp->sd_lock);
3680 		if (stp->sd_pushcnt >= nstrpush) {
3681 			fmodsw_rele(fp);
3682 			strendplumb(stp);
3683 			mutex_exit(&stp->sd_lock);
3684 			return (EINVAL);
3685 		}
3686 		mutex_exit(&stp->sd_lock);
3687 
3688 		/*
3689 		 * Push new module and call its open routine
3690 		 * via qattach().  Modules don't change device
3691 		 * numbers, so just ignore dummydev here.
3692 		 */
3693 		dummydev = vp->v_rdev;
3694 		if ((error = qattach(rdq, &dummydev, 0, crp, fp,
3695 		    B_FALSE)) == 0) {
3696 			if (vp->v_type == VCHR && /* sorry, no pipes allowed */
3697 			    (cmd == I_PUSH) && (stp->sd_flag & STRISTTY)) {
3698 				/*
3699 				 * try to allocate it as a controlling terminal
3700 				 */
3701 				(void) strctty(stp);
3702 			}
3703 		}
3704 
3705 		mutex_enter(&stp->sd_lock);
3706 
3707 		/*
3708 		 * As a performance concern we are caching the values of
3709 		 * q_minpsz and q_maxpsz of the module below the stream
3710 		 * head in the stream head.
3711 		 */
3712 		mutex_enter(QLOCK(stp->sd_wrq->q_next));
3713 		rmin = stp->sd_wrq->q_next->q_minpsz;
3714 		rmax = stp->sd_wrq->q_next->q_maxpsz;
3715 		mutex_exit(QLOCK(stp->sd_wrq->q_next));
3716 
3717 		/* Do this processing here as a performance concern */
3718 		if (strmsgsz != 0) {
3719 			if (rmax == INFPSZ)
3720 				rmax = strmsgsz;
3721 			else  {
3722 				if (vp->v_type == VFIFO)
3723 					rmax = MIN(PIPE_BUF, rmax);
3724 				else	rmax = MIN(strmsgsz, rmax);
3725 			}
3726 		}
3727 
3728 		mutex_enter(QLOCK(wrq));
3729 		stp->sd_qn_minpsz = rmin;
3730 		stp->sd_qn_maxpsz = rmax;
3731 		mutex_exit(QLOCK(wrq));
3732 
3733 		strendplumb(stp);
3734 		mutex_exit(&stp->sd_lock);
3735 		return (error);
3736 	    }
3737 
3738 	case I_POP:
3739 	    {
3740 		queue_t	*q;
3741 
3742 		if (stp->sd_flag & STRHUP)
3743 			return (ENXIO);
3744 		if (!wrq->q_next)	/* for broken pipes */
3745 			return (EINVAL);
3746 
3747 		if (error = strstartplumb(stp, flag, cmd))
3748 			return (error);
3749 
3750 		/*
3751 		 * If there is an anchor on this stream and popping
3752 		 * the current module would attempt to pop through the
3753 		 * anchor, then disallow the pop unless we have sufficient
3754 		 * privileges; take the cheapest (non-locking) check
3755 		 * first.
3756 		 */
3757 		if (secpolicy_net_config(crp, B_TRUE) != 0) {
3758 			mutex_enter(&stp->sd_lock);
3759 			/*
3760 			 * Anchors only apply if there's at least one
3761 			 * module on the stream (sd_pushcnt > 0).
3762 			 */
3763 			if (stp->sd_pushcnt > 0 &&
3764 			    stp->sd_pushcnt == stp->sd_anchor &&
3765 			    stp->sd_vnode->v_type != VFIFO) {
3766 				strendplumb(stp);
3767 				mutex_exit(&stp->sd_lock);
3768 				/* Audit and report error */
3769 				return (secpolicy_net_config(crp, B_FALSE));
3770 			}
3771 			mutex_exit(&stp->sd_lock);
3772 		}
3773 
3774 		q = wrq->q_next;
3775 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_POP,
3776 			"I_POP:%p from %p", q, stp);
3777 		if (q->q_next == NULL || (q->q_flag & (QREADR|QISDRV))) {
3778 			error = EINVAL;
3779 		} else {
3780 			qdetach(_RD(q), 1, flag, crp, B_FALSE);
3781 			error = 0;
3782 		}
3783 		mutex_enter(&stp->sd_lock);
3784 
3785 		/*
3786 		 * As a performance concern we are caching the values of
3787 		 * q_minpsz and q_maxpsz of the module below the stream
3788 		 * head in the stream head.
3789 		 */
3790 		mutex_enter(QLOCK(wrq->q_next));
3791 		rmin = wrq->q_next->q_minpsz;
3792 		rmax = wrq->q_next->q_maxpsz;
3793 		mutex_exit(QLOCK(wrq->q_next));
3794 
3795 		/* Do this processing here as a performance concern */
3796 		if (strmsgsz != 0) {
3797 			if (rmax == INFPSZ)
3798 				rmax = strmsgsz;
3799 			else  {
3800 				if (vp->v_type == VFIFO)
3801 					rmax = MIN(PIPE_BUF, rmax);
3802 				else	rmax = MIN(strmsgsz, rmax);
3803 			}
3804 		}
3805 
3806 		mutex_enter(QLOCK(wrq));
3807 		stp->sd_qn_minpsz = rmin;
3808 		stp->sd_qn_maxpsz = rmax;
3809 		mutex_exit(QLOCK(wrq));
3810 
3811 		/* If we popped through the anchor, then reset the anchor. */
3812 		if (stp->sd_pushcnt < stp->sd_anchor)
3813 			stp->sd_anchor = 0;
3814 
3815 		strendplumb(stp);
3816 		mutex_exit(&stp->sd_lock);
3817 		return (error);
3818 	    }
3819 
3820 	case _I_MUXID2FD:
3821 	{
3822 		/*
3823 		 * Create a fd for a I_PLINK'ed lower stream with a given
3824 		 * muxid.  With the fd, application can send down ioctls,
3825 		 * like I_LIST, to the previously I_PLINK'ed stream.  Note
3826 		 * that after getting the fd, the application has to do an
3827 		 * I_PUNLINK on the muxid before it can do any operation
3828 		 * on the lower stream.  This is required by spec1170.
3829 		 *
3830 		 * The fd used to do this ioctl should point to the same
3831 		 * controlling device used to do the I_PLINK.  If it uses
3832 		 * a different stream or an invalid muxid, I_MUXID2FD will
3833 		 * fail.  The error code is set to EINVAL.
3834 		 *
3835 		 * The intended use of this interface is the following.
3836 		 * An application I_PLINK'ed a stream and exits.  The fd
3837 		 * to the lower stream is gone.  Another application
3838 		 * wants to get a fd to the lower stream, it uses I_MUXID2FD.
3839 		 */
3840 		int muxid = (int)arg;
3841 		int fd;
3842 		linkinfo_t *linkp;
3843 		struct file *fp;
3844 
3845 		/*
3846 		 * Do not allow the wildcard muxid.  This ioctl is not
3847 		 * intended to find arbitrary link.
3848 		 */
3849 		if (muxid == 0) {
3850 			return (EINVAL);
3851 		}
3852 
3853 		mutex_enter(&muxifier);
3854 		linkp = findlinks(vp->v_stream, muxid, LINKPERSIST);
3855 		if (linkp == NULL) {
3856 			mutex_exit(&muxifier);
3857 			return (EINVAL);
3858 		}
3859 
3860 		if ((fd = ufalloc(0)) == -1) {
3861 			mutex_exit(&muxifier);
3862 			return (EMFILE);
3863 		}
3864 		fp = linkp->li_fpdown;
3865 		mutex_enter(&fp->f_tlock);
3866 		fp->f_count++;
3867 		mutex_exit(&fp->f_tlock);
3868 		mutex_exit(&muxifier);
3869 		setf(fd, fp);
3870 		*rvalp = fd;
3871 		return (0);
3872 	}
3873 
3874 	case _I_INSERT:
3875 	{
3876 		/*
3877 		 * To insert a module to a given position in a stream.
3878 		 * In the first release, only allow privileged user
3879 		 * to use this ioctl.
3880 		 *
3881 		 * Note that we do not plan to support this ioctl
3882 		 * on pipes in the first release.  We want to learn more
3883 		 * about the implications of these ioctls before extending
3884 		 * their support.  And we do not think these features are
3885 		 * valuable for pipes.
3886 		 *
3887 		 * Neither do we support O/C hot stream.  Note that only
3888 		 * the upper streams of TCP/IP stack are O/C hot streams.
3889 		 * The lower IP stream is not.
3890 		 * When there is a O/C cold barrier, we only allow inserts
3891 		 * above the barrier.
3892 		 */
3893 		STRUCT_DECL(strmodconf, strmodinsert);
3894 		char mod_name[FMNAMESZ + 1];
3895 		fmodsw_impl_t *fp;
3896 		dev_t dummydev;
3897 		queue_t *tmp_wrq;
3898 		int pos;
3899 		boolean_t is_insert;
3900 
3901 		STRUCT_INIT(strmodinsert, flag);
3902 		if (stp->sd_flag & STRHUP)
3903 			return (ENXIO);
3904 		if (STRMATED(stp))
3905 			return (EINVAL);
3906 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
3907 			return (error);
3908 
3909 		error = strcopyin((void *)arg, STRUCT_BUF(strmodinsert),
3910 		    STRUCT_SIZE(strmodinsert), copyflag);
3911 		if (error)
3912 			return (error);
3913 
3914 		/*
3915 		 * Get module name and look up in fmodsw.
3916 		 */
3917 		error = (copyflag & U_TO_K ? copyinstr :
3918 		    copystr)(STRUCT_FGETP(strmodinsert, mod_name),
3919 		    mod_name, FMNAMESZ + 1, NULL);
3920 		if (error)
3921 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3922 
3923 		if ((fp = fmodsw_find(mod_name, FMODSW_HOLD | FMODSW_LOAD)) ==
3924 		    NULL)
3925 			return (EINVAL);
3926 
3927 		if (error = strstartplumb(stp, flag, cmd)) {
3928 			fmodsw_rele(fp);
3929 			return (error);
3930 		}
3931 
3932 		/*
3933 		 * Is this _I_INSERT just like an I_PUSH?  We need to know
3934 		 * this because we do some optimizations if this is a
3935 		 * module being pushed.
3936 		 */
3937 		pos = STRUCT_FGET(strmodinsert, pos);
3938 		is_insert = (pos != 0);
3939 
3940 		/*
3941 		 * Make sure pos is valid.  Even though it is not an I_PUSH,
3942 		 * we impose the same limit on the number of modules in a
3943 		 * stream.
3944 		 */
3945 		mutex_enter(&stp->sd_lock);
3946 		if (stp->sd_pushcnt >= nstrpush || pos < 0 ||
3947 		    pos > stp->sd_pushcnt) {
3948 			fmodsw_rele(fp);
3949 			strendplumb(stp);
3950 			mutex_exit(&stp->sd_lock);
3951 			return (EINVAL);
3952 		}
3953 		mutex_exit(&stp->sd_lock);
3954 
3955 		/*
3956 		 * First find the correct position this module to
3957 		 * be inserted.  We don't need to call claimstr()
3958 		 * as the stream should not be changing at this point.
3959 		 *
3960 		 * Insert new module and call its open routine
3961 		 * via qattach().  Modules don't change device
3962 		 * numbers, so just ignore dummydev here.
3963 		 */
3964 		for (tmp_wrq = stp->sd_wrq; pos > 0;
3965 		    tmp_wrq = tmp_wrq->q_next, pos--) {
3966 			ASSERT(SAMESTR(tmp_wrq));
3967 		}
3968 		dummydev = vp->v_rdev;
3969 		if ((error = qattach(_RD(tmp_wrq), &dummydev, 0, crp,
3970 		    fp, is_insert)) != 0) {
3971 			mutex_enter(&stp->sd_lock);
3972 			strendplumb(stp);
3973 			mutex_exit(&stp->sd_lock);
3974 			return (error);
3975 		}
3976 
3977 		mutex_enter(&stp->sd_lock);
3978 
3979 		/*
3980 		 * As a performance concern we are caching the values of
3981 		 * q_minpsz and q_maxpsz of the module below the stream
3982 		 * head in the stream head.
3983 		 */
3984 		if (!is_insert) {
3985 			mutex_enter(QLOCK(stp->sd_wrq->q_next));
3986 			rmin = stp->sd_wrq->q_next->q_minpsz;
3987 			rmax = stp->sd_wrq->q_next->q_maxpsz;
3988 			mutex_exit(QLOCK(stp->sd_wrq->q_next));
3989 
3990 			/* Do this processing here as a performance concern */
3991 			if (strmsgsz != 0) {
3992 				if (rmax == INFPSZ) {
3993 					rmax = strmsgsz;
3994 				} else  {
3995 					rmax = MIN(strmsgsz, rmax);
3996 				}
3997 			}
3998 
3999 			mutex_enter(QLOCK(wrq));
4000 			stp->sd_qn_minpsz = rmin;
4001 			stp->sd_qn_maxpsz = rmax;
4002 			mutex_exit(QLOCK(wrq));
4003 		}
4004 
4005 		/*
4006 		 * Need to update the anchor value if this module is
4007 		 * inserted below the anchor point.
4008 		 */
4009 		if (stp->sd_anchor != 0) {
4010 			pos = STRUCT_FGET(strmodinsert, pos);
4011 			if (pos >= (stp->sd_pushcnt - stp->sd_anchor))
4012 				stp->sd_anchor++;
4013 		}
4014 
4015 		strendplumb(stp);
4016 		mutex_exit(&stp->sd_lock);
4017 		return (0);
4018 	}
4019 
4020 	case _I_REMOVE:
4021 	{
4022 		/*
4023 		 * To remove a module with a given name in a stream.  The
4024 		 * caller of this ioctl needs to provide both the name and
4025 		 * the position of the module to be removed.  This eliminates
4026 		 * the ambiguity of removal if a module is inserted/pushed
4027 		 * multiple times in a stream.  In the first release, only
4028 		 * allow privileged user to use this ioctl.
4029 		 *
4030 		 * Note that we do not plan to support this ioctl
4031 		 * on pipes in the first release.  We want to learn more
4032 		 * about the implications of these ioctls before extending
4033 		 * their support.  And we do not think these features are
4034 		 * valuable for pipes.
4035 		 *
4036 		 * Neither do we support O/C hot stream.  Note that only
4037 		 * the upper streams of TCP/IP stack are O/C hot streams.
4038 		 * The lower IP stream is not.
4039 		 * When there is a O/C cold barrier we do not allow removal
4040 		 * below the barrier.
4041 		 *
4042 		 * Also note that _I_REMOVE cannot be used to remove a
4043 		 * driver or the stream head.
4044 		 */
4045 		STRUCT_DECL(strmodconf, strmodremove);
4046 		queue_t	*q;
4047 		int pos;
4048 		char mod_name[FMNAMESZ + 1];
4049 		boolean_t is_remove;
4050 
4051 		STRUCT_INIT(strmodremove, flag);
4052 		if (stp->sd_flag & STRHUP)
4053 			return (ENXIO);
4054 		if (STRMATED(stp))
4055 			return (EINVAL);
4056 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4057 			return (error);
4058 
4059 		error = strcopyin((void *)arg, STRUCT_BUF(strmodremove),
4060 		    STRUCT_SIZE(strmodremove), copyflag);
4061 		if (error)
4062 			return (error);
4063 
4064 		error = (copyflag & U_TO_K ? copyinstr :
4065 		    copystr)(STRUCT_FGETP(strmodremove, mod_name),
4066 		    mod_name, FMNAMESZ + 1, NULL);
4067 		if (error)
4068 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4069 
4070 		if ((error = strstartplumb(stp, flag, cmd)) != 0)
4071 			return (error);
4072 
4073 		/*
4074 		 * Match the name of given module to the name of module at
4075 		 * the given position.
4076 		 */
4077 		pos = STRUCT_FGET(strmodremove, pos);
4078 
4079 		is_remove = (pos != 0);
4080 		for (q = stp->sd_wrq->q_next; SAMESTR(q) && pos > 0;
4081 		    q = q->q_next, pos--)
4082 			;
4083 		if (pos > 0 || ! SAMESTR(q) ||
4084 		    strncmp(q->q_qinfo->qi_minfo->mi_idname, mod_name,
4085 		    strlen(q->q_qinfo->qi_minfo->mi_idname)) != 0) {
4086 			mutex_enter(&stp->sd_lock);
4087 			strendplumb(stp);
4088 			mutex_exit(&stp->sd_lock);
4089 			return (EINVAL);
4090 		}
4091 
4092 		ASSERT(!(q->q_flag & QREADR));
4093 		qdetach(_RD(q), 1, flag, crp, is_remove);
4094 
4095 		mutex_enter(&stp->sd_lock);
4096 
4097 		/*
4098 		 * As a performance concern we are caching the values of
4099 		 * q_minpsz and q_maxpsz of the module below the stream
4100 		 * head in the stream head.
4101 		 */
4102 		if (!is_remove) {
4103 			mutex_enter(QLOCK(wrq->q_next));
4104 			rmin = wrq->q_next->q_minpsz;
4105 			rmax = wrq->q_next->q_maxpsz;
4106 			mutex_exit(QLOCK(wrq->q_next));
4107 
4108 			/* Do this processing here as a performance concern */
4109 			if (strmsgsz != 0) {
4110 				if (rmax == INFPSZ)
4111 					rmax = strmsgsz;
4112 				else  {
4113 					if (vp->v_type == VFIFO)
4114 						rmax = MIN(PIPE_BUF, rmax);
4115 					else	rmax = MIN(strmsgsz, rmax);
4116 				}
4117 			}
4118 
4119 			mutex_enter(QLOCK(wrq));
4120 			stp->sd_qn_minpsz = rmin;
4121 			stp->sd_qn_maxpsz = rmax;
4122 			mutex_exit(QLOCK(wrq));
4123 		}
4124 
4125 		/*
4126 		 * Need to update the anchor value if this module is removed
4127 		 * at or below the anchor point.  If the removed module is at
4128 		 * the anchor point, remove the anchor for this stream if
4129 		 * there is no module above the anchor point.  Otherwise, if
4130 		 * the removed module is below the anchor point, decrement the
4131 		 * anchor point by 1.
4132 		 */
4133 		if (stp->sd_anchor != 0) {
4134 			pos = STRUCT_FGET(strmodremove, pos);
4135 			if (pos == 0)
4136 				stp->sd_anchor = 0;
4137 			else if (pos > (stp->sd_pushcnt - stp->sd_anchor + 1))
4138 				stp->sd_anchor--;
4139 		}
4140 
4141 		strendplumb(stp);
4142 		mutex_exit(&stp->sd_lock);
4143 		return (0);
4144 	}
4145 
4146 	case I_ANCHOR:
4147 		/*
4148 		 * Set the anchor position on the stream to reside at
4149 		 * the top module (in other words, the top module
4150 		 * cannot be popped).  Anchors with a FIFO make no
4151 		 * obvious sense, so they're not allowed.
4152 		 */
4153 		mutex_enter(&stp->sd_lock);
4154 
4155 		if (stp->sd_vnode->v_type == VFIFO) {
4156 			mutex_exit(&stp->sd_lock);
4157 			return (EINVAL);
4158 		}
4159 
4160 		stp->sd_anchor = stp->sd_pushcnt;
4161 
4162 		mutex_exit(&stp->sd_lock);
4163 		return (0);
4164 
4165 	case I_LOOK:
4166 		/*
4167 		 * Get name of first module downstream.
4168 		 * If no module, return an error.
4169 		 */
4170 	    {
4171 		claimstr(wrq);
4172 		if (_SAMESTR(wrq) && wrq->q_next->q_next) {
4173 			char *name = wrq->q_next->q_qinfo->qi_minfo->mi_idname;
4174 			error = strcopyout(name, (void *)arg, strlen(name) + 1,
4175 			    copyflag);
4176 			releasestr(wrq);
4177 			return (error);
4178 		}
4179 		releasestr(wrq);
4180 		return (EINVAL);
4181 	    }
4182 
4183 	case I_LINK:
4184 	case I_PLINK:
4185 		/*
4186 		 * Link a multiplexor.
4187 		 */
4188 		return (mlink(vp, cmd, (int)arg, crp, rvalp, 0));
4189 
4190 	case _I_PLINK_LH:
4191 		/*
4192 		 * Link a multiplexor: Call must originate from kernel.
4193 		 */
4194 		if (kioctl)
4195 			return (ldi_mlink_lh(vp, cmd, arg, crp, rvalp));
4196 
4197 		return (EINVAL);
4198 	case I_UNLINK:
4199 	case I_PUNLINK:
4200 		/*
4201 		 * Unlink a multiplexor.
4202 		 * If arg is -1, unlink all links for which this is the
4203 		 * controlling stream.  Otherwise, arg is an index number
4204 		 * for a link to be removed.
4205 		 */
4206 	    {
4207 		struct linkinfo *linkp;
4208 		int native_arg = (int)arg;
4209 		int type;
4210 
4211 		TRACE_1(TR_FAC_STREAMS_FR,
4212 			TR_I_UNLINK, "I_UNLINK/I_PUNLINK:%p", stp);
4213 		if (vp->v_type == VFIFO) {
4214 			return (EINVAL);
4215 		}
4216 		if (cmd == I_UNLINK)
4217 			type = LINKNORMAL;
4218 		else	/* I_PUNLINK */
4219 			type = LINKPERSIST;
4220 		if (native_arg == 0) {
4221 			return (EINVAL);
4222 		}
4223 		if (native_arg == MUXID_ALL)
4224 			error = munlinkall(stp, type, crp, rvalp);
4225 		else {
4226 			mutex_enter(&muxifier);
4227 			if (!(linkp = findlinks(stp, (int)arg, type))) {
4228 				/* invalid user supplied index number */
4229 				mutex_exit(&muxifier);
4230 				return (EINVAL);
4231 			}
4232 			/* munlink drops the muxifier lock */
4233 			error = munlink(stp, linkp, type, crp, rvalp);
4234 		}
4235 		return (error);
4236 	    }
4237 
4238 	case I_FLUSH:
4239 		/*
4240 		 * send a flush message downstream
4241 		 * flush message can indicate
4242 		 * FLUSHR - flush read queue
4243 		 * FLUSHW - flush write queue
4244 		 * FLUSHRW - flush read/write queue
4245 		 */
4246 		if (stp->sd_flag & STRHUP)
4247 			return (ENXIO);
4248 		if (arg & ~FLUSHRW)
4249 			return (EINVAL);
4250 
4251 		for (;;) {
4252 			if (putnextctl1(stp->sd_wrq, M_FLUSH, (int)arg)) {
4253 				break;
4254 			}
4255 			if (error = strwaitbuf(1, BPRI_HI)) {
4256 				return (error);
4257 			}
4258 		}
4259 
4260 		/*
4261 		 * Send down an unsupported ioctl and wait for the nack
4262 		 * in order to allow the M_FLUSH to propagate back
4263 		 * up to the stream head.
4264 		 * Replaces if (qready()) runqueues();
4265 		 */
4266 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4267 		strioc.ic_timout = 0;
4268 		strioc.ic_len = 0;
4269 		strioc.ic_dp = NULL;
4270 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4271 		*rvalp = 0;
4272 		return (0);
4273 
4274 	case I_FLUSHBAND:
4275 	    {
4276 		struct bandinfo binfo;
4277 
4278 		error = strcopyin((void *)arg, &binfo, sizeof (binfo),
4279 		    copyflag);
4280 		if (error)
4281 			return (error);
4282 		if (stp->sd_flag & STRHUP)
4283 			return (ENXIO);
4284 		if (binfo.bi_flag & ~FLUSHRW)
4285 			return (EINVAL);
4286 		while (!(mp = allocb(2, BPRI_HI))) {
4287 			if (error = strwaitbuf(2, BPRI_HI))
4288 				return (error);
4289 		}
4290 		mp->b_datap->db_type = M_FLUSH;
4291 		*mp->b_wptr++ = binfo.bi_flag | FLUSHBAND;
4292 		*mp->b_wptr++ = binfo.bi_pri;
4293 		putnext(stp->sd_wrq, mp);
4294 		/*
4295 		 * Send down an unsupported ioctl and wait for the nack
4296 		 * in order to allow the M_FLUSH to propagate back
4297 		 * up to the stream head.
4298 		 * Replaces if (qready()) runqueues();
4299 		 */
4300 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4301 		strioc.ic_timout = 0;
4302 		strioc.ic_len = 0;
4303 		strioc.ic_dp = NULL;
4304 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4305 		*rvalp = 0;
4306 		return (0);
4307 	    }
4308 
4309 	case I_SRDOPT:
4310 		/*
4311 		 * Set read options
4312 		 *
4313 		 * RNORM - default stream mode
4314 		 * RMSGN - message no discard
4315 		 * RMSGD - message discard
4316 		 * RPROTNORM - fail read with EBADMSG for M_[PC]PROTOs
4317 		 * RPROTDAT - convert M_[PC]PROTOs to M_DATAs
4318 		 * RPROTDIS - discard M_[PC]PROTOs and retain M_DATAs
4319 		 */
4320 		if (arg & ~(RMODEMASK | RPROTMASK))
4321 			return (EINVAL);
4322 
4323 		if ((arg & (RMSGD|RMSGN)) == (RMSGD|RMSGN))
4324 			return (EINVAL);
4325 
4326 		mutex_enter(&stp->sd_lock);
4327 		switch (arg & RMODEMASK) {
4328 		case RNORM:
4329 			stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
4330 			break;
4331 		case RMSGD:
4332 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGNODIS) |
4333 			    RD_MSGDIS;
4334 			break;
4335 		case RMSGN:
4336 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGDIS) |
4337 			    RD_MSGNODIS;
4338 			break;
4339 		}
4340 
4341 		switch (arg & RPROTMASK) {
4342 		case RPROTNORM:
4343 			stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
4344 			break;
4345 
4346 		case RPROTDAT:
4347 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDIS) |
4348 			    RD_PROTDAT);
4349 			break;
4350 
4351 		case RPROTDIS:
4352 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDAT) |
4353 			    RD_PROTDIS);
4354 			break;
4355 		}
4356 		mutex_exit(&stp->sd_lock);
4357 		return (0);
4358 
4359 	case I_GRDOPT:
4360 		/*
4361 		 * Get read option and return the value
4362 		 * to spot pointed to by arg
4363 		 */
4364 	    {
4365 		int rdopt;
4366 
4367 		rdopt = ((stp->sd_read_opt & RD_MSGDIS) ? RMSGD :
4368 		    ((stp->sd_read_opt & RD_MSGNODIS) ? RMSGN : RNORM));
4369 		rdopt |= ((stp->sd_read_opt & RD_PROTDAT) ? RPROTDAT :
4370 		    ((stp->sd_read_opt & RD_PROTDIS) ? RPROTDIS : RPROTNORM));
4371 
4372 		return (strcopyout(&rdopt, (void *)arg, sizeof (int),
4373 		    copyflag));
4374 	    }
4375 
4376 	case I_SERROPT:
4377 		/*
4378 		 * Set error options
4379 		 *
4380 		 * RERRNORM - persistent read errors
4381 		 * RERRNONPERSIST - non-persistent read errors
4382 		 * WERRNORM - persistent write errors
4383 		 * WERRNONPERSIST - non-persistent write errors
4384 		 */
4385 		if (arg & ~(RERRMASK | WERRMASK))
4386 			return (EINVAL);
4387 
4388 		mutex_enter(&stp->sd_lock);
4389 		switch (arg & RERRMASK) {
4390 		case RERRNORM:
4391 			stp->sd_flag &= ~STRDERRNONPERSIST;
4392 			break;
4393 		case RERRNONPERSIST:
4394 			stp->sd_flag |= STRDERRNONPERSIST;
4395 			break;
4396 		}
4397 		switch (arg & WERRMASK) {
4398 		case WERRNORM:
4399 			stp->sd_flag &= ~STWRERRNONPERSIST;
4400 			break;
4401 		case WERRNONPERSIST:
4402 			stp->sd_flag |= STWRERRNONPERSIST;
4403 			break;
4404 		}
4405 		mutex_exit(&stp->sd_lock);
4406 		return (0);
4407 
4408 	case I_GERROPT:
4409 		/*
4410 		 * Get error option and return the value
4411 		 * to spot pointed to by arg
4412 		 */
4413 	    {
4414 		int erropt = 0;
4415 
4416 		erropt |= (stp->sd_flag & STRDERRNONPERSIST) ? RERRNONPERSIST :
4417 			RERRNORM;
4418 		erropt |= (stp->sd_flag & STWRERRNONPERSIST) ? WERRNONPERSIST :
4419 			WERRNORM;
4420 		return (strcopyout(&erropt, (void *)arg, sizeof (int),
4421 		    copyflag));
4422 	    }
4423 
4424 	case I_SETSIG:
4425 		/*
4426 		 * Register the calling proc to receive the SIGPOLL
4427 		 * signal based on the events given in arg.  If
4428 		 * arg is zero, remove the proc from register list.
4429 		 */
4430 	    {
4431 		strsig_t *ssp, *pssp;
4432 		struct pid *pidp;
4433 
4434 		pssp = NULL;
4435 		pidp = curproc->p_pidp;
4436 		/*
4437 		 * Hold sd_lock to prevent traversal of sd_siglist while
4438 		 * it is modified.
4439 		 */
4440 		mutex_enter(&stp->sd_lock);
4441 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pidp != pidp);
4442 			pssp = ssp, ssp = ssp->ss_next)
4443 			;
4444 
4445 		if (arg) {
4446 			if (arg & ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4447 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4448 				mutex_exit(&stp->sd_lock);
4449 				return (EINVAL);
4450 			}
4451 			if ((arg & S_BANDURG) && !(arg & S_RDBAND)) {
4452 				mutex_exit(&stp->sd_lock);
4453 				return (EINVAL);
4454 			}
4455 
4456 			/*
4457 			 * If proc not already registered, add it
4458 			 * to list.
4459 			 */
4460 			if (!ssp) {
4461 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4462 				ssp->ss_pidp = pidp;
4463 				ssp->ss_pid = pidp->pid_id;
4464 				ssp->ss_next = NULL;
4465 				if (pssp)
4466 					pssp->ss_next = ssp;
4467 				else
4468 					stp->sd_siglist = ssp;
4469 				mutex_enter(&pidlock);
4470 				PID_HOLD(pidp);
4471 				mutex_exit(&pidlock);
4472 			}
4473 
4474 			/*
4475 			 * Set events.
4476 			 */
4477 			ssp->ss_events = (int)arg;
4478 		} else {
4479 			/*
4480 			 * Remove proc from register list.
4481 			 */
4482 			if (ssp) {
4483 				mutex_enter(&pidlock);
4484 				PID_RELE(pidp);
4485 				mutex_exit(&pidlock);
4486 				if (pssp)
4487 					pssp->ss_next = ssp->ss_next;
4488 				else
4489 					stp->sd_siglist = ssp->ss_next;
4490 				kmem_free(ssp, sizeof (strsig_t));
4491 			} else {
4492 				mutex_exit(&stp->sd_lock);
4493 				return (EINVAL);
4494 			}
4495 		}
4496 
4497 		/*
4498 		 * Recalculate OR of sig events.
4499 		 */
4500 		stp->sd_sigflags = 0;
4501 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4502 			stp->sd_sigflags |= ssp->ss_events;
4503 		mutex_exit(&stp->sd_lock);
4504 		return (0);
4505 	    }
4506 
4507 	case I_GETSIG:
4508 		/*
4509 		 * Return (in arg) the current registration of events
4510 		 * for which the calling proc is to be signaled.
4511 		 */
4512 	    {
4513 		struct strsig *ssp;
4514 		struct pid  *pidp;
4515 
4516 		pidp = curproc->p_pidp;
4517 		mutex_enter(&stp->sd_lock);
4518 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4519 			if (ssp->ss_pidp == pidp) {
4520 				error = strcopyout(&ssp->ss_events, (void *)arg,
4521 				    sizeof (int), copyflag);
4522 				mutex_exit(&stp->sd_lock);
4523 				return (error);
4524 			}
4525 		mutex_exit(&stp->sd_lock);
4526 		return (EINVAL);
4527 	    }
4528 
4529 	case I_ESETSIG:
4530 		/*
4531 		 * Register the ss_pid to receive the SIGPOLL
4532 		 * signal based on the events is ss_events arg.  If
4533 		 * ss_events is zero, remove the proc from register list.
4534 		 */
4535 	{
4536 		struct strsig *ssp, *pssp;
4537 		struct proc *proc;
4538 		struct pid  *pidp;
4539 		pid_t pid;
4540 		struct strsigset ss;
4541 
4542 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4543 		if (error)
4544 			return (error);
4545 
4546 		pid = ss.ss_pid;
4547 
4548 		if (ss.ss_events != 0) {
4549 			/*
4550 			 * Permissions check by sending signal 0.
4551 			 * Note that when kill fails it does a set_errno
4552 			 * causing the system call to fail.
4553 			 */
4554 			error = kill(pid, 0);
4555 			if (error) {
4556 				return (error);
4557 			}
4558 		}
4559 		mutex_enter(&pidlock);
4560 		if (pid == 0)
4561 			proc = curproc;
4562 		else if (pid < 0)
4563 			proc = pgfind(-pid);
4564 		else
4565 			proc = prfind(pid);
4566 		if (proc == NULL) {
4567 			mutex_exit(&pidlock);
4568 			return (ESRCH);
4569 		}
4570 		if (pid < 0)
4571 			pidp = proc->p_pgidp;
4572 		else
4573 			pidp = proc->p_pidp;
4574 		ASSERT(pidp);
4575 		/*
4576 		 * Get a hold on the pid structure while referencing it.
4577 		 * There is a separate PID_HOLD should it be inserted
4578 		 * in the list below.
4579 		 */
4580 		PID_HOLD(pidp);
4581 		mutex_exit(&pidlock);
4582 
4583 		pssp = NULL;
4584 		/*
4585 		 * Hold sd_lock to prevent traversal of sd_siglist while
4586 		 * it is modified.
4587 		 */
4588 		mutex_enter(&stp->sd_lock);
4589 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pid != pid);
4590 				pssp = ssp, ssp = ssp->ss_next)
4591 			;
4592 
4593 		if (ss.ss_events) {
4594 			if (ss.ss_events &
4595 			    ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4596 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4597 				mutex_exit(&stp->sd_lock);
4598 				mutex_enter(&pidlock);
4599 				PID_RELE(pidp);
4600 				mutex_exit(&pidlock);
4601 				return (EINVAL);
4602 			}
4603 			if ((ss.ss_events & S_BANDURG) &&
4604 			    !(ss.ss_events & S_RDBAND)) {
4605 				mutex_exit(&stp->sd_lock);
4606 				mutex_enter(&pidlock);
4607 				PID_RELE(pidp);
4608 				mutex_exit(&pidlock);
4609 				return (EINVAL);
4610 			}
4611 
4612 			/*
4613 			 * If proc not already registered, add it
4614 			 * to list.
4615 			 */
4616 			if (!ssp) {
4617 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4618 				ssp->ss_pidp = pidp;
4619 				ssp->ss_pid = pid;
4620 				ssp->ss_next = NULL;
4621 				if (pssp)
4622 					pssp->ss_next = ssp;
4623 				else
4624 					stp->sd_siglist = ssp;
4625 				mutex_enter(&pidlock);
4626 				PID_HOLD(pidp);
4627 				mutex_exit(&pidlock);
4628 			}
4629 
4630 			/*
4631 			 * Set events.
4632 			 */
4633 			ssp->ss_events = ss.ss_events;
4634 		} else {
4635 			/*
4636 			 * Remove proc from register list.
4637 			 */
4638 			if (ssp) {
4639 				mutex_enter(&pidlock);
4640 				PID_RELE(pidp);
4641 				mutex_exit(&pidlock);
4642 				if (pssp)
4643 					pssp->ss_next = ssp->ss_next;
4644 				else
4645 					stp->sd_siglist = ssp->ss_next;
4646 				kmem_free(ssp, sizeof (strsig_t));
4647 			} else {
4648 				mutex_exit(&stp->sd_lock);
4649 				mutex_enter(&pidlock);
4650 				PID_RELE(pidp);
4651 				mutex_exit(&pidlock);
4652 				return (EINVAL);
4653 			}
4654 		}
4655 
4656 		/*
4657 		 * Recalculate OR of sig events.
4658 		 */
4659 		stp->sd_sigflags = 0;
4660 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4661 			stp->sd_sigflags |= ssp->ss_events;
4662 		mutex_exit(&stp->sd_lock);
4663 		mutex_enter(&pidlock);
4664 		PID_RELE(pidp);
4665 		mutex_exit(&pidlock);
4666 		return (0);
4667 	    }
4668 
4669 	case I_EGETSIG:
4670 		/*
4671 		 * Return (in arg) the current registration of events
4672 		 * for which the calling proc is to be signaled.
4673 		 */
4674 	    {
4675 		struct strsig *ssp;
4676 		struct proc *proc;
4677 		pid_t pid;
4678 		struct pid  *pidp;
4679 		struct strsigset ss;
4680 
4681 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4682 		if (error)
4683 			return (error);
4684 
4685 		pid = ss.ss_pid;
4686 		mutex_enter(&pidlock);
4687 		if (pid == 0)
4688 			proc = curproc;
4689 		else if (pid < 0)
4690 			proc = pgfind(-pid);
4691 		else
4692 			proc = prfind(pid);
4693 		if (proc == NULL) {
4694 			mutex_exit(&pidlock);
4695 			return (ESRCH);
4696 		}
4697 		if (pid < 0)
4698 			pidp = proc->p_pgidp;
4699 		else
4700 			pidp = proc->p_pidp;
4701 
4702 		/* Prevent the pidp from being reassigned */
4703 		PID_HOLD(pidp);
4704 		mutex_exit(&pidlock);
4705 
4706 		mutex_enter(&stp->sd_lock);
4707 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4708 			if (ssp->ss_pid == pid) {
4709 				ss.ss_pid = ssp->ss_pid;
4710 				ss.ss_events = ssp->ss_events;
4711 				error = strcopyout(&ss, (void *)arg,
4712 				    sizeof (struct strsigset), copyflag);
4713 				mutex_exit(&stp->sd_lock);
4714 				mutex_enter(&pidlock);
4715 				PID_RELE(pidp);
4716 				mutex_exit(&pidlock);
4717 				return (error);
4718 			}
4719 		mutex_exit(&stp->sd_lock);
4720 		mutex_enter(&pidlock);
4721 		PID_RELE(pidp);
4722 		mutex_exit(&pidlock);
4723 		return (EINVAL);
4724 	    }
4725 
4726 	case I_PEEK:
4727 	    {
4728 		STRUCT_DECL(strpeek, strpeek);
4729 		size_t n;
4730 		mblk_t *fmp, *tmp_mp = NULL;
4731 
4732 		STRUCT_INIT(strpeek, flag);
4733 
4734 		error = strcopyin((void *)arg, STRUCT_BUF(strpeek),
4735 		    STRUCT_SIZE(strpeek), copyflag);
4736 		if (error)
4737 			return (error);
4738 
4739 		mutex_enter(QLOCK(rdq));
4740 		/*
4741 		 * Skip the invalid messages
4742 		 */
4743 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
4744 			if (mp->b_datap->db_type != M_SIG)
4745 				break;
4746 
4747 		/*
4748 		 * If user has requested to peek at a high priority message
4749 		 * and first message is not, return 0
4750 		 */
4751 		if (mp != NULL) {
4752 			if ((STRUCT_FGET(strpeek, flags) & RS_HIPRI) &&
4753 			    queclass(mp) == QNORM) {
4754 				*rvalp = 0;
4755 				mutex_exit(QLOCK(rdq));
4756 				return (0);
4757 			}
4758 		} else if (stp->sd_struiordq == NULL ||
4759 		    (STRUCT_FGET(strpeek, flags) & RS_HIPRI)) {
4760 			/*
4761 			 * No mblks to look at at the streamhead and
4762 			 * 1). This isn't a synch stream or
4763 			 * 2). This is a synch stream but caller wants high
4764 			 *	priority messages which is not supported by
4765 			 *	the synch stream. (it only supports QNORM)
4766 			 */
4767 			*rvalp = 0;
4768 			mutex_exit(QLOCK(rdq));
4769 			return (0);
4770 		}
4771 
4772 		fmp = mp;
4773 
4774 		if (mp && mp->b_datap->db_type == M_PASSFP) {
4775 			mutex_exit(QLOCK(rdq));
4776 			return (EBADMSG);
4777 		}
4778 
4779 		ASSERT(mp == NULL || mp->b_datap->db_type == M_PCPROTO ||
4780 		    mp->b_datap->db_type == M_PROTO ||
4781 		    mp->b_datap->db_type == M_DATA);
4782 
4783 		if (mp && mp->b_datap->db_type == M_PCPROTO) {
4784 			STRUCT_FSET(strpeek, flags, RS_HIPRI);
4785 		} else {
4786 			STRUCT_FSET(strpeek, flags, 0);
4787 		}
4788 
4789 
4790 		if (mp && ((tmp_mp = dupmsg(mp)) == NULL)) {
4791 			mutex_exit(QLOCK(rdq));
4792 			return (ENOSR);
4793 		}
4794 		mutex_exit(QLOCK(rdq));
4795 
4796 		/*
4797 		 * set mp = tmp_mp, so that I_PEEK processing can continue.
4798 		 * tmp_mp is used to free the dup'd message.
4799 		 */
4800 		mp = tmp_mp;
4801 
4802 		uio.uio_fmode = 0;
4803 		uio.uio_extflg = UIO_COPY_CACHED;
4804 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
4805 		    UIO_SYSSPACE;
4806 		uio.uio_limit = 0;
4807 		/*
4808 		 * First process PROTO blocks, if any.
4809 		 * If user doesn't want to get ctl info by setting maxlen <= 0,
4810 		 * then set len to -1/0 and skip control blocks part.
4811 		 */
4812 		if (STRUCT_FGET(strpeek, ctlbuf.maxlen) < 0)
4813 			STRUCT_FSET(strpeek, ctlbuf.len, -1);
4814 		else if (STRUCT_FGET(strpeek, ctlbuf.maxlen) == 0)
4815 			STRUCT_FSET(strpeek, ctlbuf.len, 0);
4816 		else {
4817 			int	ctl_part = 0;
4818 
4819 			iov.iov_base = STRUCT_FGETP(strpeek, ctlbuf.buf);
4820 			iov.iov_len = STRUCT_FGET(strpeek, ctlbuf.maxlen);
4821 			uio.uio_iov = &iov;
4822 			uio.uio_resid = iov.iov_len;
4823 			uio.uio_loffset = 0;
4824 			uio.uio_iovcnt = 1;
4825 			while (mp && mp->b_datap->db_type != M_DATA &&
4826 			    uio.uio_resid >= 0) {
4827 				ASSERT(STRUCT_FGET(strpeek, flags) == 0 ?
4828 				    mp->b_datap->db_type == M_PROTO :
4829 				    mp->b_datap->db_type == M_PCPROTO);
4830 
4831 				if ((n = MIN(uio.uio_resid,
4832 				    mp->b_wptr - mp->b_rptr)) != 0 &&
4833 				    (error = uiomove((char *)mp->b_rptr, n,
4834 				    UIO_READ, &uio)) != 0) {
4835 					freemsg(tmp_mp);
4836 					return (error);
4837 				}
4838 				ctl_part = 1;
4839 				mp = mp->b_cont;
4840 			}
4841 			/* No ctl message */
4842 			if (ctl_part == 0)
4843 				STRUCT_FSET(strpeek, ctlbuf.len, -1);
4844 			else
4845 				STRUCT_FSET(strpeek, ctlbuf.len,
4846 				    STRUCT_FGET(strpeek, ctlbuf.maxlen) -
4847 				    uio.uio_resid);
4848 		}
4849 
4850 		/*
4851 		 * Now process DATA blocks, if any.
4852 		 * If user doesn't want to get data info by setting maxlen <= 0,
4853 		 * then set len to -1/0 and skip data blocks part.
4854 		 */
4855 		if (STRUCT_FGET(strpeek, databuf.maxlen) < 0)
4856 			STRUCT_FSET(strpeek, databuf.len, -1);
4857 		else if (STRUCT_FGET(strpeek, databuf.maxlen) == 0)
4858 			STRUCT_FSET(strpeek, databuf.len, 0);
4859 		else {
4860 			int	data_part = 0;
4861 
4862 			iov.iov_base = STRUCT_FGETP(strpeek, databuf.buf);
4863 			iov.iov_len = STRUCT_FGET(strpeek, databuf.maxlen);
4864 			uio.uio_iov = &iov;
4865 			uio.uio_resid = iov.iov_len;
4866 			uio.uio_loffset = 0;
4867 			uio.uio_iovcnt = 1;
4868 			while (mp && uio.uio_resid) {
4869 				if (mp->b_datap->db_type == M_DATA) {
4870 					if ((n = MIN(uio.uio_resid,
4871 					    mp->b_wptr - mp->b_rptr)) != 0 &&
4872 					    (error = uiomove((char *)mp->b_rptr,
4873 						n, UIO_READ, &uio)) != 0) {
4874 						freemsg(tmp_mp);
4875 						return (error);
4876 					}
4877 					data_part = 1;
4878 				}
4879 				ASSERT(data_part == 0 ||
4880 				    mp->b_datap->db_type == M_DATA);
4881 				mp = mp->b_cont;
4882 			}
4883 			/* No data message */
4884 			if (data_part == 0)
4885 				STRUCT_FSET(strpeek, databuf.len, -1);
4886 			else
4887 				STRUCT_FSET(strpeek, databuf.len,
4888 				    STRUCT_FGET(strpeek, databuf.maxlen) -
4889 				    uio.uio_resid);
4890 		}
4891 		freemsg(tmp_mp);
4892 
4893 		/*
4894 		 * It is a synch stream and user wants to get
4895 		 * data (maxlen > 0).
4896 		 * uio setup is done by the codes that process DATA
4897 		 * blocks above.
4898 		 */
4899 		if ((fmp == NULL) && STRUCT_FGET(strpeek, databuf.maxlen) > 0) {
4900 			infod_t infod;
4901 
4902 			infod.d_cmd = INFOD_COPYOUT;
4903 			infod.d_res = 0;
4904 			infod.d_uiop = &uio;
4905 			error = infonext(rdq, &infod);
4906 			if (error == EINVAL || error == EBUSY)
4907 				error = 0;
4908 			if (error)
4909 				return (error);
4910 			STRUCT_FSET(strpeek, databuf.len, STRUCT_FGET(strpeek,
4911 			    databuf.maxlen) - uio.uio_resid);
4912 			if (STRUCT_FGET(strpeek, databuf.len) == 0) {
4913 				/*
4914 				 * No data found by the infonext().
4915 				 */
4916 				STRUCT_FSET(strpeek, databuf.len, -1);
4917 			}
4918 		}
4919 		error = strcopyout(STRUCT_BUF(strpeek), (void *)arg,
4920 		    STRUCT_SIZE(strpeek), copyflag);
4921 		if (error) {
4922 			return (error);
4923 		}
4924 		/*
4925 		 * If there is no message retrieved, set return code to 0
4926 		 * otherwise, set it to 1.
4927 		 */
4928 		if (STRUCT_FGET(strpeek, ctlbuf.len) == -1 &&
4929 		    STRUCT_FGET(strpeek, databuf.len) == -1)
4930 			*rvalp = 0;
4931 		else
4932 			*rvalp = 1;
4933 		return (0);
4934 	    }
4935 
4936 	case I_FDINSERT:
4937 	    {
4938 		STRUCT_DECL(strfdinsert, strfdinsert);
4939 		struct file *resftp;
4940 		struct stdata *resstp;
4941 		t_uscalar_t	ival;
4942 		ssize_t msgsize;
4943 		struct strbuf mctl;
4944 
4945 		STRUCT_INIT(strfdinsert, flag);
4946 		if (stp->sd_flag & STRHUP)
4947 			return (ENXIO);
4948 		/*
4949 		 * STRDERR, STWRERR and STPLEX tested above.
4950 		 */
4951 		error = strcopyin((void *)arg, STRUCT_BUF(strfdinsert),
4952 		    STRUCT_SIZE(strfdinsert), copyflag);
4953 		if (error)
4954 			return (error);
4955 
4956 		if (STRUCT_FGET(strfdinsert, offset) < 0 ||
4957 		    (STRUCT_FGET(strfdinsert, offset) %
4958 		    sizeof (t_uscalar_t)) != 0)
4959 			return (EINVAL);
4960 		if ((resftp = getf(STRUCT_FGET(strfdinsert, fildes))) != NULL) {
4961 			if ((resstp = resftp->f_vnode->v_stream) == NULL) {
4962 				releasef(STRUCT_FGET(strfdinsert, fildes));
4963 				return (EINVAL);
4964 			}
4965 		} else
4966 			return (EINVAL);
4967 
4968 		mutex_enter(&resstp->sd_lock);
4969 		if (resstp->sd_flag & (STRDERR|STWRERR|STRHUP|STPLEX)) {
4970 			error = strgeterr(resstp,
4971 					STRDERR|STWRERR|STRHUP|STPLEX, 0);
4972 			if (error != 0) {
4973 				mutex_exit(&resstp->sd_lock);
4974 				releasef(STRUCT_FGET(strfdinsert, fildes));
4975 				return (error);
4976 			}
4977 		}
4978 		mutex_exit(&resstp->sd_lock);
4979 
4980 #ifdef	_ILP32
4981 		{
4982 			queue_t	*q;
4983 			queue_t	*mate = NULL;
4984 
4985 			/* get read queue of stream terminus */
4986 			claimstr(resstp->sd_wrq);
4987 			for (q = resstp->sd_wrq->q_next; q->q_next != NULL;
4988 			    q = q->q_next)
4989 				if (!STRMATED(resstp) && STREAM(q) != resstp &&
4990 				    mate == NULL) {
4991 					ASSERT(q->q_qinfo->qi_srvp);
4992 					ASSERT(_OTHERQ(q)->q_qinfo->qi_srvp);
4993 					claimstr(q);
4994 					mate = q;
4995 				}
4996 			q = _RD(q);
4997 			if (mate)
4998 				releasestr(mate);
4999 			releasestr(resstp->sd_wrq);
5000 			ival = (t_uscalar_t)q;
5001 		}
5002 #else
5003 		ival = (t_uscalar_t)getminor(resftp->f_vnode->v_rdev);
5004 #endif	/* _ILP32 */
5005 
5006 		if (STRUCT_FGET(strfdinsert, ctlbuf.len) <
5007 		    STRUCT_FGET(strfdinsert, offset) + sizeof (t_uscalar_t)) {
5008 			releasef(STRUCT_FGET(strfdinsert, fildes));
5009 			return (EINVAL);
5010 		}
5011 
5012 		/*
5013 		 * Check for legal flag value.
5014 		 */
5015 		if (STRUCT_FGET(strfdinsert, flags) & ~RS_HIPRI) {
5016 			releasef(STRUCT_FGET(strfdinsert, fildes));
5017 			return (EINVAL);
5018 		}
5019 
5020 		/* get these values from those cached in the stream head */
5021 		mutex_enter(QLOCK(stp->sd_wrq));
5022 		rmin = stp->sd_qn_minpsz;
5023 		rmax = stp->sd_qn_maxpsz;
5024 		mutex_exit(QLOCK(stp->sd_wrq));
5025 
5026 		/*
5027 		 * Make sure ctl and data sizes together fall within
5028 		 * the limits of the max and min receive packet sizes
5029 		 * and do not exceed system limit.  A negative data
5030 		 * length means that no data part is to be sent.
5031 		 */
5032 		ASSERT((rmax >= 0) || (rmax == INFPSZ));
5033 		if (rmax == 0) {
5034 			releasef(STRUCT_FGET(strfdinsert, fildes));
5035 			return (ERANGE);
5036 		}
5037 		if ((msgsize = STRUCT_FGET(strfdinsert, databuf.len)) < 0)
5038 			msgsize = 0;
5039 		if ((msgsize < rmin) ||
5040 		    ((msgsize > rmax) && (rmax != INFPSZ)) ||
5041 		    (STRUCT_FGET(strfdinsert, ctlbuf.len) > strctlsz)) {
5042 			releasef(STRUCT_FGET(strfdinsert, fildes));
5043 			return (ERANGE);
5044 		}
5045 
5046 		mutex_enter(&stp->sd_lock);
5047 		while (!(STRUCT_FGET(strfdinsert, flags) & RS_HIPRI) &&
5048 		    !canputnext(stp->sd_wrq)) {
5049 			if ((error = strwaitq(stp, WRITEWAIT, (ssize_t)0,
5050 			    flag, -1, &done)) != 0 || done) {
5051 				mutex_exit(&stp->sd_lock);
5052 				releasef(STRUCT_FGET(strfdinsert, fildes));
5053 				return (error);
5054 			}
5055 			if ((error = i_straccess(stp, access)) != 0) {
5056 				mutex_exit(&stp->sd_lock);
5057 				releasef(
5058 				    STRUCT_FGET(strfdinsert, fildes));
5059 				return (error);
5060 			}
5061 		}
5062 		mutex_exit(&stp->sd_lock);
5063 
5064 		/*
5065 		 * Copy strfdinsert.ctlbuf into native form of
5066 		 * ctlbuf to pass down into strmakemsg().
5067 		 */
5068 		mctl.maxlen = STRUCT_FGET(strfdinsert, ctlbuf.maxlen);
5069 		mctl.len = STRUCT_FGET(strfdinsert, ctlbuf.len);
5070 		mctl.buf = STRUCT_FGETP(strfdinsert, ctlbuf.buf);
5071 
5072 		iov.iov_base = STRUCT_FGETP(strfdinsert, databuf.buf);
5073 		iov.iov_len = STRUCT_FGET(strfdinsert, databuf.len);
5074 		uio.uio_iov = &iov;
5075 		uio.uio_iovcnt = 1;
5076 		uio.uio_loffset = 0;
5077 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5078 		    UIO_SYSSPACE;
5079 		uio.uio_fmode = 0;
5080 		uio.uio_extflg = UIO_COPY_CACHED;
5081 		uio.uio_resid = iov.iov_len;
5082 		if ((error = strmakemsg(&mctl,
5083 		    &msgsize, &uio, stp,
5084 		    STRUCT_FGET(strfdinsert, flags), &mp)) != 0 || !mp) {
5085 			STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5086 			releasef(STRUCT_FGET(strfdinsert, fildes));
5087 			return (error);
5088 		}
5089 
5090 		STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5091 
5092 		/*
5093 		 * Place the possibly reencoded queue pointer 'offset' bytes
5094 		 * from the start of the control portion of the message.
5095 		 */
5096 		*((t_uscalar_t *)(mp->b_rptr +
5097 		    STRUCT_FGET(strfdinsert, offset))) = ival;
5098 
5099 		/*
5100 		 * Put message downstream.
5101 		 */
5102 		stream_willservice(stp);
5103 		putnext(stp->sd_wrq, mp);
5104 		stream_runservice(stp);
5105 		releasef(STRUCT_FGET(strfdinsert, fildes));
5106 		return (error);
5107 	    }
5108 
5109 	case I_SENDFD:
5110 	    {
5111 		struct file *fp;
5112 
5113 		if ((fp = getf((int)arg)) == NULL)
5114 			return (EBADF);
5115 		error = do_sendfp(stp, fp, crp);
5116 #ifdef C2_AUDIT
5117 		if (audit_active) {
5118 			audit_fdsend((int)arg, fp, error);
5119 		}
5120 #endif
5121 		releasef((int)arg);
5122 		return (error);
5123 	    }
5124 
5125 	case I_RECVFD:
5126 	case I_E_RECVFD:
5127 	    {
5128 		struct k_strrecvfd *srf;
5129 		int i, fd;
5130 
5131 		mutex_enter(&stp->sd_lock);
5132 		while (!(mp = getq(rdq))) {
5133 			if (stp->sd_flag & (STRHUP|STREOF)) {
5134 				mutex_exit(&stp->sd_lock);
5135 				return (ENXIO);
5136 			}
5137 			if ((error = strwaitq(stp, GETWAIT, (ssize_t)0,
5138 			    flag, -1, &done)) != 0 || done) {
5139 				mutex_exit(&stp->sd_lock);
5140 				return (error);
5141 			}
5142 			if ((error = i_straccess(stp, access)) != 0) {
5143 				mutex_exit(&stp->sd_lock);
5144 				return (error);
5145 			}
5146 		}
5147 		if (mp->b_datap->db_type != M_PASSFP) {
5148 			putback(stp, rdq, mp, mp->b_band);
5149 			mutex_exit(&stp->sd_lock);
5150 			return (EBADMSG);
5151 		}
5152 		mutex_exit(&stp->sd_lock);
5153 
5154 		srf = (struct k_strrecvfd *)mp->b_rptr;
5155 		if ((fd = ufalloc(0)) == -1) {
5156 			mutex_enter(&stp->sd_lock);
5157 			putback(stp, rdq, mp, mp->b_band);
5158 			mutex_exit(&stp->sd_lock);
5159 			return (EMFILE);
5160 		}
5161 		if (cmd == I_RECVFD) {
5162 			struct o_strrecvfd	ostrfd;
5163 
5164 			/* check to see if uid/gid values are too large. */
5165 
5166 			if (srf->uid > (o_uid_t)USHRT_MAX ||
5167 			    srf->gid > (o_gid_t)USHRT_MAX) {
5168 				mutex_enter(&stp->sd_lock);
5169 				putback(stp, rdq, mp, mp->b_band);
5170 				mutex_exit(&stp->sd_lock);
5171 				setf(fd, NULL);	/* release fd entry */
5172 				return (EOVERFLOW);
5173 			}
5174 
5175 			ostrfd.fd = fd;
5176 			ostrfd.uid = (o_uid_t)srf->uid;
5177 			ostrfd.gid = (o_gid_t)srf->gid;
5178 
5179 			/* Null the filler bits */
5180 			for (i = 0; i < 8; i++)
5181 				ostrfd.fill[i] = 0;
5182 
5183 			error = strcopyout(&ostrfd, (void *)arg,
5184 			    sizeof (struct o_strrecvfd), copyflag);
5185 		} else {		/* I_E_RECVFD */
5186 			struct strrecvfd	strfd;
5187 
5188 			strfd.fd = fd;
5189 			strfd.uid = srf->uid;
5190 			strfd.gid = srf->gid;
5191 
5192 			/* null the filler bits */
5193 			for (i = 0; i < 8; i++)
5194 				strfd.fill[i] = 0;
5195 
5196 			error = strcopyout(&strfd, (void *)arg,
5197 			    sizeof (struct strrecvfd), copyflag);
5198 		}
5199 
5200 		if (error) {
5201 			setf(fd, NULL);	/* release fd entry */
5202 			mutex_enter(&stp->sd_lock);
5203 			putback(stp, rdq, mp, mp->b_band);
5204 			mutex_exit(&stp->sd_lock);
5205 			return (error);
5206 		}
5207 #ifdef C2_AUDIT
5208 		if (audit_active) {
5209 			audit_fdrecv(fd, srf->fp);
5210 		}
5211 #endif
5212 
5213 		/*
5214 		 * Always increment f_count since the freemsg() below will
5215 		 * always call free_passfp() which performs a closef().
5216 		 */
5217 		mutex_enter(&srf->fp->f_tlock);
5218 		srf->fp->f_count++;
5219 		mutex_exit(&srf->fp->f_tlock);
5220 		setf(fd, srf->fp);
5221 		freemsg(mp);
5222 		return (0);
5223 	    }
5224 
5225 	case I_SWROPT:
5226 		/*
5227 		 * Set/clear the write options. arg is a bit
5228 		 * mask with any of the following bits set...
5229 		 * 	SNDZERO - send zero length message
5230 		 *	SNDPIPE - send sigpipe to process if
5231 		 *		sd_werror is set and process is
5232 		 *		doing a write or putmsg.
5233 		 * The new stream head write options should reflect
5234 		 * what is in arg.
5235 		 */
5236 		if (arg & ~(SNDZERO|SNDPIPE))
5237 			return (EINVAL);
5238 
5239 		mutex_enter(&stp->sd_lock);
5240 		stp->sd_wput_opt &= ~(SW_SIGPIPE|SW_SNDZERO);
5241 		if (arg & SNDZERO)
5242 			stp->sd_wput_opt |= SW_SNDZERO;
5243 		if (arg & SNDPIPE)
5244 			stp->sd_wput_opt |= SW_SIGPIPE;
5245 		mutex_exit(&stp->sd_lock);
5246 		return (0);
5247 
5248 	case I_GWROPT:
5249 	    {
5250 		int wropt = 0;
5251 
5252 		if (stp->sd_wput_opt & SW_SNDZERO)
5253 			wropt |= SNDZERO;
5254 		if (stp->sd_wput_opt & SW_SIGPIPE)
5255 			wropt |= SNDPIPE;
5256 		return (strcopyout(&wropt, (void *)arg, sizeof (wropt),
5257 		    copyflag));
5258 	    }
5259 
5260 	case I_LIST:
5261 		/*
5262 		 * Returns all the modules found on this stream,
5263 		 * upto the driver. If argument is NULL, return the
5264 		 * number of modules (including driver). If argument
5265 		 * is not NULL, copy the names into the structure
5266 		 * provided.
5267 		 */
5268 
5269 	    {
5270 		queue_t *q;
5271 		int num_modules, space_allocated;
5272 		STRUCT_DECL(str_list, strlist);
5273 		struct str_mlist *mlist_ptr;
5274 
5275 		if (arg == NULL) { /* Return number of modules plus driver */
5276 			q = stp->sd_wrq;
5277 			if (stp->sd_vnode->v_type == VFIFO) {
5278 				*rvalp = stp->sd_pushcnt;
5279 			} else {
5280 				*rvalp = stp->sd_pushcnt + 1;
5281 			}
5282 		} else {
5283 			STRUCT_INIT(strlist, flag);
5284 
5285 			error = strcopyin((void *)arg, STRUCT_BUF(strlist),
5286 			    STRUCT_SIZE(strlist), copyflag);
5287 			if (error)
5288 				return (error);
5289 
5290 			space_allocated = STRUCT_FGET(strlist, sl_nmods);
5291 			if ((space_allocated) <= 0)
5292 				return (EINVAL);
5293 			claimstr(stp->sd_wrq);
5294 			q = stp->sd_wrq;
5295 			num_modules = 0;
5296 			while (_SAMESTR(q) && (space_allocated != 0)) {
5297 				char *name =
5298 				    q->q_next->q_qinfo->qi_minfo->mi_idname;
5299 
5300 				mlist_ptr = STRUCT_FGETP(strlist, sl_modlist);
5301 
5302 				error = strcopyout(name, mlist_ptr,
5303 				    strlen(name) + 1, copyflag);
5304 
5305 				if (error) {
5306 					releasestr(stp->sd_wrq);
5307 					return (error);
5308 				}
5309 				q = q->q_next;
5310 				space_allocated--;
5311 				num_modules++;
5312 				mlist_ptr =
5313 				    (struct str_mlist *)((uintptr_t)mlist_ptr +
5314 				    sizeof (struct str_mlist));
5315 				STRUCT_FSETP(strlist, sl_modlist, mlist_ptr);
5316 			}
5317 			releasestr(stp->sd_wrq);
5318 			error = strcopyout(&num_modules, (void *)arg,
5319 			    sizeof (int), copyflag);
5320 		}
5321 		return (error);
5322 	    }
5323 
5324 	case I_CKBAND:
5325 	    {
5326 		queue_t *q;
5327 		qband_t *qbp;
5328 
5329 		if ((arg < 0) || (arg >= NBAND))
5330 			return (EINVAL);
5331 		q = _RD(stp->sd_wrq);
5332 		mutex_enter(QLOCK(q));
5333 		if (arg > (int)q->q_nband) {
5334 			*rvalp = 0;
5335 		} else {
5336 			if (arg == 0) {
5337 				if (q->q_first)
5338 					*rvalp = 1;
5339 				else
5340 					*rvalp = 0;
5341 			} else {
5342 				qbp = q->q_bandp;
5343 				while (--arg > 0)
5344 					qbp = qbp->qb_next;
5345 				if (qbp->qb_first)
5346 					*rvalp = 1;
5347 				else
5348 					*rvalp = 0;
5349 			}
5350 		}
5351 		mutex_exit(QLOCK(q));
5352 		return (0);
5353 	    }
5354 
5355 	case I_GETBAND:
5356 	    {
5357 		int intpri;
5358 		queue_t *q;
5359 
5360 		q = _RD(stp->sd_wrq);
5361 		mutex_enter(QLOCK(q));
5362 		mp = q->q_first;
5363 		if (!mp) {
5364 			mutex_exit(QLOCK(q));
5365 			return (ENODATA);
5366 		}
5367 		intpri = (int)mp->b_band;
5368 		error = strcopyout(&intpri, (void *)arg, sizeof (int),
5369 		    copyflag);
5370 		mutex_exit(QLOCK(q));
5371 		return (error);
5372 	    }
5373 
5374 	case I_ATMARK:
5375 	    {
5376 		queue_t *q;
5377 
5378 		if (arg & ~(ANYMARK|LASTMARK))
5379 			return (EINVAL);
5380 		q = _RD(stp->sd_wrq);
5381 		mutex_enter(&stp->sd_lock);
5382 		if ((stp->sd_flag & STRATMARK) && (arg == ANYMARK)) {
5383 			*rvalp = 1;
5384 		} else {
5385 			mutex_enter(QLOCK(q));
5386 			mp = q->q_first;
5387 
5388 			if (mp == NULL)
5389 				*rvalp = 0;
5390 			else if ((arg == ANYMARK) && (mp->b_flag & MSGMARK))
5391 				*rvalp = 1;
5392 			else if ((arg == LASTMARK) && (mp == stp->sd_mark))
5393 				*rvalp = 1;
5394 			else
5395 				*rvalp = 0;
5396 			mutex_exit(QLOCK(q));
5397 		}
5398 		mutex_exit(&stp->sd_lock);
5399 		return (0);
5400 	    }
5401 
5402 	case I_CANPUT:
5403 	    {
5404 		char band;
5405 
5406 		if ((arg < 0) || (arg >= NBAND))
5407 			return (EINVAL);
5408 		band = (char)arg;
5409 		*rvalp = bcanputnext(stp->sd_wrq, band);
5410 		return (0);
5411 	    }
5412 
5413 	case I_SETCLTIME:
5414 	    {
5415 		int closetime;
5416 
5417 		error = strcopyin((void *)arg, &closetime, sizeof (int),
5418 		    copyflag);
5419 		if (error)
5420 			return (error);
5421 		if (closetime < 0)
5422 			return (EINVAL);
5423 
5424 		stp->sd_closetime = closetime;
5425 		return (0);
5426 	    }
5427 
5428 	case I_GETCLTIME:
5429 	    {
5430 		int closetime;
5431 
5432 		closetime = stp->sd_closetime;
5433 		return (strcopyout(&closetime, (void *)arg, sizeof (int),
5434 		    copyflag));
5435 	    }
5436 
5437 	case TIOCGSID:
5438 	{
5439 		pid_t sid;
5440 
5441 		mutex_enter(&stp->sd_lock);
5442 		if (stp->sd_sidp == NULL) {
5443 			mutex_exit(&stp->sd_lock);
5444 			return (ENOTTY);
5445 		}
5446 		sid = stp->sd_sidp->pid_id;
5447 		mutex_exit(&stp->sd_lock);
5448 		return (strcopyout(&sid, (void *)arg, sizeof (pid_t),
5449 		    copyflag));
5450 	}
5451 
5452 	case TIOCSPGRP:
5453 	{
5454 		pid_t pgrp;
5455 		proc_t *q;
5456 		pid_t	sid, fg_pgid, bg_pgid;
5457 
5458 		if (error = strcopyin((void *)arg, &pgrp, sizeof (pid_t),
5459 		    copyflag))
5460 			return (error);
5461 		mutex_enter(&stp->sd_lock);
5462 		mutex_enter(&pidlock);
5463 		if (stp->sd_sidp != ttoproc(curthread)->p_sessp->s_sidp) {
5464 			mutex_exit(&pidlock);
5465 			mutex_exit(&stp->sd_lock);
5466 			return (ENOTTY);
5467 		}
5468 		if (pgrp == stp->sd_pgidp->pid_id) {
5469 			mutex_exit(&pidlock);
5470 			mutex_exit(&stp->sd_lock);
5471 			return (0);
5472 		}
5473 		if (pgrp <= 0 || pgrp >= maxpid) {
5474 			mutex_exit(&pidlock);
5475 			mutex_exit(&stp->sd_lock);
5476 			return (EINVAL);
5477 		}
5478 		if ((q = pgfind(pgrp)) == NULL ||
5479 		    q->p_sessp != ttoproc(curthread)->p_sessp) {
5480 			mutex_exit(&pidlock);
5481 			mutex_exit(&stp->sd_lock);
5482 			return (EPERM);
5483 		}
5484 		sid = stp->sd_sidp->pid_id;
5485 		fg_pgid = q->p_pgrp;
5486 		bg_pgid = stp->sd_pgidp->pid_id;
5487 		CL_SET_PROCESS_GROUP(curthread, sid, bg_pgid, fg_pgid);
5488 		PID_RELE(stp->sd_pgidp);
5489 		ctty_clear_sighuped();
5490 		stp->sd_pgidp = q->p_pgidp;
5491 		PID_HOLD(stp->sd_pgidp);
5492 		mutex_exit(&pidlock);
5493 		mutex_exit(&stp->sd_lock);
5494 		return (0);
5495 	}
5496 
5497 	case TIOCGPGRP:
5498 	{
5499 		pid_t pgrp;
5500 
5501 		mutex_enter(&stp->sd_lock);
5502 		if (stp->sd_sidp == NULL) {
5503 			mutex_exit(&stp->sd_lock);
5504 			return (ENOTTY);
5505 		}
5506 		pgrp = stp->sd_pgidp->pid_id;
5507 		mutex_exit(&stp->sd_lock);
5508 		return (strcopyout(&pgrp, (void *)arg, sizeof (pid_t),
5509 		    copyflag));
5510 	}
5511 
5512 	case TIOCSCTTY:
5513 	{
5514 		return (strctty(stp));
5515 	}
5516 
5517 	case TIOCNOTTY:
5518 	{
5519 		/* freectty() always assumes curproc. */
5520 		if (freectty(B_FALSE) != 0)
5521 			return (0);
5522 		return (ENOTTY);
5523 	}
5524 
5525 	case FIONBIO:
5526 	case FIOASYNC:
5527 		return (0);	/* handled by the upper layer */
5528 	}
5529 }
5530 
5531 /*
5532  * Custom free routine used for M_PASSFP messages.
5533  */
5534 static void
5535 free_passfp(struct k_strrecvfd *srf)
5536 {
5537 	(void) closef(srf->fp);
5538 	kmem_free(srf, sizeof (struct k_strrecvfd) + sizeof (frtn_t));
5539 }
5540 
5541 /* ARGSUSED */
5542 int
5543 do_sendfp(struct stdata *stp, struct file *fp, struct cred *cr)
5544 {
5545 	queue_t *qp, *nextqp;
5546 	struct k_strrecvfd *srf;
5547 	mblk_t *mp;
5548 	frtn_t *frtnp;
5549 	size_t bufsize;
5550 	queue_t	*mate = NULL;
5551 	syncq_t	*sq = NULL;
5552 	int retval = 0;
5553 
5554 	if (stp->sd_flag & STRHUP)
5555 		return (ENXIO);
5556 
5557 	claimstr(stp->sd_wrq);
5558 
5559 	/* Fastpath, we have a pipe, and we are already mated, use it. */
5560 	if (STRMATED(stp)) {
5561 		qp = _RD(stp->sd_mate->sd_wrq);
5562 		claimstr(qp);
5563 		mate = qp;
5564 	} else { /* Not already mated. */
5565 
5566 		/*
5567 		 * Walk the stream to the end of this one.
5568 		 * assumes that the claimstr() will prevent
5569 		 * plumbing between the stream head and the
5570 		 * driver from changing
5571 		 */
5572 		qp = stp->sd_wrq;
5573 
5574 		/*
5575 		 * Loop until we reach the end of this stream.
5576 		 * On completion, qp points to the write queue
5577 		 * at the end of the stream, or the read queue
5578 		 * at the stream head if this is a fifo.
5579 		 */
5580 		while (((qp = qp->q_next) != NULL) && _SAMESTR(qp))
5581 			;
5582 
5583 		/*
5584 		 * Just in case we get a q_next which is NULL, but
5585 		 * not at the end of the stream.  This is actually
5586 		 * broken, so we set an assert to catch it in
5587 		 * debug, and set an error and return if not debug.
5588 		 */
5589 		ASSERT(qp);
5590 		if (qp == NULL) {
5591 			releasestr(stp->sd_wrq);
5592 			return (EINVAL);
5593 		}
5594 
5595 		/*
5596 		 * Enter the syncq for the driver, so (hopefully)
5597 		 * the queue values will not change on us.
5598 		 * XXXX - This will only prevent the race IFF only
5599 		 *   the write side modifies the q_next member, and
5600 		 *   the put procedure is protected by at least
5601 		 *   MT_PERQ.
5602 		 */
5603 		if ((sq = qp->q_syncq) != NULL)
5604 			entersq(sq, SQ_PUT);
5605 
5606 		/* Now get the q_next value from this qp. */
5607 		nextqp = qp->q_next;
5608 
5609 		/*
5610 		 * If nextqp exists and the other stream is different
5611 		 * from this one claim the stream, set the mate, and
5612 		 * get the read queue at the stream head of the other
5613 		 * stream.  Assumes that nextqp was at least valid when
5614 		 * we got it.  Hopefully the entersq of the driver
5615 		 * will prevent it from changing on us.
5616 		 */
5617 		if ((nextqp != NULL) && (STREAM(nextqp) != stp)) {
5618 			ASSERT(qp->q_qinfo->qi_srvp);
5619 			ASSERT(_OTHERQ(qp)->q_qinfo->qi_srvp);
5620 			ASSERT(_OTHERQ(qp->q_next)->q_qinfo->qi_srvp);
5621 			claimstr(nextqp);
5622 
5623 			/* Make sure we still have a q_next */
5624 			if (nextqp != qp->q_next) {
5625 				releasestr(stp->sd_wrq);
5626 				releasestr(nextqp);
5627 				return (EINVAL);
5628 			}
5629 
5630 			qp = _RD(STREAM(nextqp)->sd_wrq);
5631 			mate = qp;
5632 		}
5633 		/* If we entered the synq above, leave it. */
5634 		if (sq != NULL)
5635 			leavesq(sq, SQ_PUT);
5636 	} /*  STRMATED(STP)  */
5637 
5638 	/* XXX prevents substitution of the ops vector */
5639 	if (qp->q_qinfo != &strdata && qp->q_qinfo != &fifo_strdata) {
5640 		retval = EINVAL;
5641 		goto out;
5642 	}
5643 
5644 	if (qp->q_flag & QFULL) {
5645 		retval = EAGAIN;
5646 		goto out;
5647 	}
5648 
5649 	/*
5650 	 * Since M_PASSFP messages include a file descriptor, we use
5651 	 * esballoc() and specify a custom free routine (free_passfp()) that
5652 	 * will close the descriptor as part of freeing the message.  For
5653 	 * convenience, we stash the frtn_t right after the data block.
5654 	 */
5655 	bufsize = sizeof (struct k_strrecvfd) + sizeof (frtn_t);
5656 	srf = kmem_alloc(bufsize, KM_NOSLEEP);
5657 	if (srf == NULL) {
5658 		retval = EAGAIN;
5659 		goto out;
5660 	}
5661 
5662 	frtnp = (frtn_t *)(srf + 1);
5663 	frtnp->free_arg = (caddr_t)srf;
5664 	frtnp->free_func = free_passfp;
5665 
5666 	mp = esballoc((uchar_t *)srf, bufsize, BPRI_MED, frtnp);
5667 	if (mp == NULL) {
5668 		kmem_free(srf, bufsize);
5669 		retval = EAGAIN;
5670 		goto out;
5671 	}
5672 	mp->b_wptr += sizeof (struct k_strrecvfd);
5673 	mp->b_datap->db_type = M_PASSFP;
5674 
5675 	srf->fp = fp;
5676 	srf->uid = crgetuid(curthread->t_cred);
5677 	srf->gid = crgetgid(curthread->t_cred);
5678 	mutex_enter(&fp->f_tlock);
5679 	fp->f_count++;
5680 	mutex_exit(&fp->f_tlock);
5681 
5682 	put(qp, mp);
5683 out:
5684 	releasestr(stp->sd_wrq);
5685 	if (mate)
5686 		releasestr(mate);
5687 	return (retval);
5688 }
5689 
5690 /*
5691  * Send an ioctl message downstream and wait for acknowledgement.
5692  * flags may be set to either U_TO_K or K_TO_K and a combination
5693  * of STR_NOERROR or STR_NOSIG
5694  * STR_NOSIG: Signals are essentially ignored or held and have
5695  *	no effect for the duration of the call.
5696  * STR_NOERROR: Ignores stream head read, write and hup errors.
5697  *	Additionally, if an existing ioctl times out, it is assumed
5698  *	lost and and this ioctl will continue as if the previous ioctl had
5699  *	finished.  ETIME may be returned if this ioctl times out (i.e.
5700  *	ic_timout is not INFTIM).  Non-stream head errors may be returned if
5701  *	the ioc_error indicates that the driver/module had problems,
5702  *	an EFAULT was found when accessing user data, a lack of
5703  * 	resources, etc.
5704  */
5705 int
5706 strdoioctl(
5707 	struct stdata *stp,
5708 	struct strioctl *strioc,
5709 	int fflags,		/* file flags with model info */
5710 	int flag,
5711 	cred_t *crp,
5712 	int *rvalp)
5713 {
5714 	mblk_t *bp;
5715 	struct iocblk *iocbp;
5716 	struct copyreq *reqp;
5717 	struct copyresp *resp;
5718 	int id;
5719 	int transparent = 0;
5720 	int error = 0;
5721 	int len = 0;
5722 	caddr_t taddr;
5723 	int copyflag = (flag & (U_TO_K | K_TO_K));
5724 	int sigflag = (flag & STR_NOSIG);
5725 	int errs;
5726 	uint_t waitflags;
5727 
5728 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
5729 	ASSERT((fflags & FMODELS) != 0);
5730 
5731 	TRACE_2(TR_FAC_STREAMS_FR,
5732 		TR_STRDOIOCTL,
5733 		"strdoioctl:stp %p strioc %p", stp, strioc);
5734 	if (strioc->ic_len == TRANSPARENT) {	/* send arg in M_DATA block */
5735 		transparent = 1;
5736 		strioc->ic_len = sizeof (intptr_t);
5737 	}
5738 
5739 	if (strioc->ic_len < 0 || (strmsgsz > 0 && strioc->ic_len > strmsgsz))
5740 		return (EINVAL);
5741 
5742 	if ((bp = allocb_cred_wait(sizeof (union ioctypes), sigflag, &error,
5743 	    crp)) == NULL)
5744 			return (error);
5745 
5746 	bzero(bp->b_wptr, sizeof (union ioctypes));
5747 
5748 	iocbp = (struct iocblk *)bp->b_wptr;
5749 	iocbp->ioc_count = strioc->ic_len;
5750 	iocbp->ioc_cmd = strioc->ic_cmd;
5751 	iocbp->ioc_flag = (fflags & FMODELS);
5752 
5753 	crhold(crp);
5754 	iocbp->ioc_cr = crp;
5755 	DB_TYPE(bp) = M_IOCTL;
5756 	DB_CPID(bp) = curproc->p_pid;
5757 	bp->b_wptr += sizeof (struct iocblk);
5758 
5759 	if (flag & STR_NOERROR)
5760 		errs = STPLEX;
5761 	else
5762 		errs = STRHUP|STRDERR|STWRERR|STPLEX;
5763 
5764 	/*
5765 	 * If there is data to copy into ioctl block, do so.
5766 	 */
5767 	if (iocbp->ioc_count > 0) {
5768 		if (transparent)
5769 			/*
5770 			 * Note: STR_NOERROR does not have an effect
5771 			 * in putiocd()
5772 			 */
5773 			id = K_TO_K | sigflag;
5774 		else
5775 			id = flag;
5776 		if ((error = putiocd(bp, strioc->ic_dp, id, crp)) != 0) {
5777 			freemsg(bp);
5778 			crfree(crp);
5779 			return (error);
5780 		}
5781 
5782 		/*
5783 		 * We could have slept copying in user pages.
5784 		 * Recheck the stream head state (the other end
5785 		 * of a pipe could have gone away).
5786 		 */
5787 		if (stp->sd_flag & errs) {
5788 			mutex_enter(&stp->sd_lock);
5789 			error = strgeterr(stp, errs, 0);
5790 			mutex_exit(&stp->sd_lock);
5791 			if (error != 0) {
5792 				freemsg(bp);
5793 				crfree(crp);
5794 				return (error);
5795 			}
5796 		}
5797 	}
5798 	if (transparent)
5799 		iocbp->ioc_count = TRANSPARENT;
5800 
5801 	/*
5802 	 * Block for up to STRTIMOUT milliseconds if there is an outstanding
5803 	 * ioctl for this stream already running.  All processes
5804 	 * sleeping here will be awakened as a result of an ACK
5805 	 * or NAK being received for the outstanding ioctl, or
5806 	 * as a result of the timer expiring on the outstanding
5807 	 * ioctl (a failure), or as a result of any waiting
5808 	 * process's timer expiring (also a failure).
5809 	 */
5810 
5811 	error = 0;
5812 	mutex_enter(&stp->sd_lock);
5813 	while (stp->sd_flag & (IOCWAIT | IOCWAITNE)) {
5814 		clock_t cv_rval;
5815 
5816 		TRACE_0(TR_FAC_STREAMS_FR,
5817 			TR_STRDOIOCTL_WAIT,
5818 			"strdoioctl sleeps - IOCWAIT");
5819 		cv_rval = str_cv_wait(&stp->sd_iocmonitor, &stp->sd_lock,
5820 		    STRTIMOUT, sigflag);
5821 		if (cv_rval <= 0) {
5822 			if (cv_rval == 0) {
5823 				error = EINTR;
5824 			} else {
5825 				if (flag & STR_NOERROR) {
5826 					/*
5827 					 * Terminating current ioctl in
5828 					 * progress -- assume it got lost and
5829 					 * wake up the other thread so that the
5830 					 * operation completes.
5831 					 */
5832 					if (!(stp->sd_flag & IOCWAITNE)) {
5833 						stp->sd_flag |= IOCWAITNE;
5834 						cv_broadcast(&stp->sd_monitor);
5835 					}
5836 					/*
5837 					 * Otherwise, there's a running
5838 					 * STR_NOERROR -- we have no choice
5839 					 * here but to wait forever (or until
5840 					 * interrupted).
5841 					 */
5842 				} else {
5843 					/*
5844 					 * pending ioctl has caused
5845 					 * us to time out
5846 					 */
5847 					error = ETIME;
5848 				}
5849 			}
5850 		} else if ((stp->sd_flag & errs)) {
5851 			error = strgeterr(stp, errs, 0);
5852 		}
5853 		if (error) {
5854 			mutex_exit(&stp->sd_lock);
5855 			freemsg(bp);
5856 			crfree(crp);
5857 			return (error);
5858 		}
5859 	}
5860 
5861 	/*
5862 	 * Have control of ioctl mechanism.
5863 	 * Send down ioctl packet and wait for response.
5864 	 */
5865 	if (stp->sd_iocblk != (mblk_t *)-1) {
5866 		freemsg(stp->sd_iocblk);
5867 	}
5868 	stp->sd_iocblk = NULL;
5869 
5870 	/*
5871 	 * If this is marked with 'noerror' (internal; mostly
5872 	 * I_{P,}{UN,}LINK), then make sure nobody else is able to get
5873 	 * in here by setting IOCWAITNE.
5874 	 */
5875 	waitflags = IOCWAIT;
5876 	if (flag & STR_NOERROR)
5877 		waitflags |= IOCWAITNE;
5878 
5879 	stp->sd_flag |= waitflags;
5880 
5881 	/*
5882 	 * Assign sequence number.
5883 	 */
5884 	iocbp->ioc_id = stp->sd_iocid = getiocseqno();
5885 
5886 	mutex_exit(&stp->sd_lock);
5887 
5888 	TRACE_1(TR_FAC_STREAMS_FR,
5889 		TR_STRDOIOCTL_PUT, "strdoioctl put: stp %p", stp);
5890 	stream_willservice(stp);
5891 	putnext(stp->sd_wrq, bp);
5892 	stream_runservice(stp);
5893 
5894 	/*
5895 	 * Timed wait for acknowledgment.  The wait time is limited by the
5896 	 * timeout value, which must be a positive integer (number of
5897 	 * milliseconds) to wait, or 0 (use default value of STRTIMOUT
5898 	 * milliseconds), or -1 (wait forever).  This will be awakened
5899 	 * either by an ACK/NAK message arriving, the timer expiring, or
5900 	 * the timer expiring on another ioctl waiting for control of the
5901 	 * mechanism.
5902 	 */
5903 waitioc:
5904 	mutex_enter(&stp->sd_lock);
5905 
5906 
5907 	/*
5908 	 * If the reply has already arrived, don't sleep.  If awakened from
5909 	 * the sleep, fail only if the reply has not arrived by then.
5910 	 * Otherwise, process the reply.
5911 	 */
5912 	while (!stp->sd_iocblk) {
5913 		clock_t cv_rval;
5914 
5915 		if (stp->sd_flag & errs) {
5916 			error = strgeterr(stp, errs, 0);
5917 			if (error != 0) {
5918 				stp->sd_flag &= ~waitflags;
5919 				cv_broadcast(&stp->sd_iocmonitor);
5920 				mutex_exit(&stp->sd_lock);
5921 				crfree(crp);
5922 				return (error);
5923 			}
5924 		}
5925 
5926 		TRACE_0(TR_FAC_STREAMS_FR,
5927 			TR_STRDOIOCTL_WAIT2,
5928 			"strdoioctl sleeps awaiting reply");
5929 		ASSERT(error == 0);
5930 
5931 		cv_rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock,
5932 		    (strioc->ic_timout ?
5933 		    strioc->ic_timout * 1000 : STRTIMOUT), sigflag);
5934 
5935 		/*
5936 		 * There are four possible cases here: interrupt, timeout,
5937 		 * wakeup by IOCWAITNE (above), or wakeup by strrput_nondata (a
5938 		 * valid M_IOCTL reply).
5939 		 *
5940 		 * If we've been awakened by a STR_NOERROR ioctl on some other
5941 		 * thread, then sd_iocblk will still be NULL, and IOCWAITNE
5942 		 * will be set.  Pretend as if we just timed out.  Note that
5943 		 * this other thread waited at least STRTIMOUT before trying to
5944 		 * awaken our thread, so this is indistinguishable (even for
5945 		 * INFTIM) from the case where we failed with ETIME waiting on
5946 		 * IOCWAIT in the prior loop.
5947 		 */
5948 		if (cv_rval > 0 && !(flag & STR_NOERROR) &&
5949 		    stp->sd_iocblk == NULL && (stp->sd_flag & IOCWAITNE)) {
5950 			cv_rval = -1;
5951 		}
5952 
5953 		/*
5954 		 * note: STR_NOERROR does not protect
5955 		 * us here.. use ic_timout < 0
5956 		 */
5957 		if (cv_rval <= 0) {
5958 			if (cv_rval == 0) {
5959 				error = EINTR;
5960 			} else {
5961 				error =  ETIME;
5962 			}
5963 			/*
5964 			 * A message could have come in after we were scheduled
5965 			 * but before we were actually run.
5966 			 */
5967 			bp = stp->sd_iocblk;
5968 			stp->sd_iocblk = NULL;
5969 			if (bp != NULL) {
5970 				if ((bp->b_datap->db_type == M_COPYIN) ||
5971 				    (bp->b_datap->db_type == M_COPYOUT)) {
5972 					mutex_exit(&stp->sd_lock);
5973 					if (bp->b_cont) {
5974 						freemsg(bp->b_cont);
5975 						bp->b_cont = NULL;
5976 					}
5977 					bp->b_datap->db_type = M_IOCDATA;
5978 					bp->b_wptr = bp->b_rptr +
5979 						sizeof (struct copyresp);
5980 					resp = (struct copyresp *)bp->b_rptr;
5981 					resp->cp_rval =
5982 					    (caddr_t)1; /* failure */
5983 					stream_willservice(stp);
5984 					putnext(stp->sd_wrq, bp);
5985 					stream_runservice(stp);
5986 					mutex_enter(&stp->sd_lock);
5987 				} else {
5988 					freemsg(bp);
5989 				}
5990 			}
5991 			stp->sd_flag &= ~waitflags;
5992 			cv_broadcast(&stp->sd_iocmonitor);
5993 			mutex_exit(&stp->sd_lock);
5994 			crfree(crp);
5995 			return (error);
5996 		}
5997 	}
5998 	bp = stp->sd_iocblk;
5999 	/*
6000 	 * Note: it is strictly impossible to get here with sd_iocblk set to
6001 	 * -1.  This is because the initial loop above doesn't allow any new
6002 	 * ioctls into the fray until all others have passed this point.
6003 	 */
6004 	ASSERT(bp != NULL && bp != (mblk_t *)-1);
6005 	TRACE_1(TR_FAC_STREAMS_FR,
6006 		TR_STRDOIOCTL_ACK, "strdoioctl got reply: bp %p", bp);
6007 	if ((bp->b_datap->db_type == M_IOCACK) ||
6008 	    (bp->b_datap->db_type == M_IOCNAK)) {
6009 		/* for detection of duplicate ioctl replies */
6010 		stp->sd_iocblk = (mblk_t *)-1;
6011 		stp->sd_flag &= ~waitflags;
6012 		cv_broadcast(&stp->sd_iocmonitor);
6013 		mutex_exit(&stp->sd_lock);
6014 	} else {
6015 		/*
6016 		 * flags not cleared here because we're still doing
6017 		 * copy in/out for ioctl.
6018 		 */
6019 		stp->sd_iocblk = NULL;
6020 		mutex_exit(&stp->sd_lock);
6021 	}
6022 
6023 
6024 	/*
6025 	 * Have received acknowledgment.
6026 	 */
6027 
6028 	switch (bp->b_datap->db_type) {
6029 	case M_IOCACK:
6030 		/*
6031 		 * Positive ack.
6032 		 */
6033 		iocbp = (struct iocblk *)bp->b_rptr;
6034 
6035 		/*
6036 		 * Set error if indicated.
6037 		 */
6038 		if (iocbp->ioc_error) {
6039 			error = iocbp->ioc_error;
6040 			break;
6041 		}
6042 
6043 		/*
6044 		 * Set return value.
6045 		 */
6046 		*rvalp = iocbp->ioc_rval;
6047 
6048 		/*
6049 		 * Data may have been returned in ACK message (ioc_count > 0).
6050 		 * If so, copy it out to the user's buffer.
6051 		 */
6052 		if (iocbp->ioc_count && !transparent) {
6053 			if (error = getiocd(bp, strioc->ic_dp, copyflag))
6054 				break;
6055 		}
6056 		if (!transparent) {
6057 			if (len)	/* an M_COPYOUT was used with I_STR */
6058 				strioc->ic_len = len;
6059 			else
6060 				strioc->ic_len = (int)iocbp->ioc_count;
6061 		}
6062 		break;
6063 
6064 	case M_IOCNAK:
6065 		/*
6066 		 * Negative ack.
6067 		 *
6068 		 * The only thing to do is set error as specified
6069 		 * in neg ack packet.
6070 		 */
6071 		iocbp = (struct iocblk *)bp->b_rptr;
6072 
6073 		error = (iocbp->ioc_error ? iocbp->ioc_error : EINVAL);
6074 		break;
6075 
6076 	case M_COPYIN:
6077 		/*
6078 		 * Driver or module has requested user ioctl data.
6079 		 */
6080 		reqp = (struct copyreq *)bp->b_rptr;
6081 
6082 		/*
6083 		 * M_COPYIN should *never* have a message attached, though
6084 		 * it's harmless if it does -- thus, panic on a DEBUG
6085 		 * kernel and just free it on a non-DEBUG build.
6086 		 */
6087 		ASSERT(bp->b_cont == NULL);
6088 		if (bp->b_cont != NULL) {
6089 			freemsg(bp->b_cont);
6090 			bp->b_cont = NULL;
6091 		}
6092 
6093 		error = putiocd(bp, reqp->cq_addr, flag, crp);
6094 		if (error && bp->b_cont) {
6095 			freemsg(bp->b_cont);
6096 			bp->b_cont = NULL;
6097 		}
6098 
6099 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6100 		bp->b_datap->db_type = M_IOCDATA;
6101 
6102 		mblk_setcred(bp, crp);
6103 		DB_CPID(bp) = curproc->p_pid;
6104 		resp = (struct copyresp *)bp->b_rptr;
6105 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6106 		resp->cp_flag = (fflags & FMODELS);
6107 
6108 		stream_willservice(stp);
6109 		putnext(stp->sd_wrq, bp);
6110 		stream_runservice(stp);
6111 
6112 		if (error) {
6113 			mutex_enter(&stp->sd_lock);
6114 			stp->sd_flag &= ~waitflags;
6115 			cv_broadcast(&stp->sd_iocmonitor);
6116 			mutex_exit(&stp->sd_lock);
6117 			crfree(crp);
6118 			return (error);
6119 		}
6120 
6121 		goto waitioc;
6122 
6123 	case M_COPYOUT:
6124 		/*
6125 		 * Driver or module has ioctl data for a user.
6126 		 */
6127 		reqp = (struct copyreq *)bp->b_rptr;
6128 		ASSERT(bp->b_cont != NULL);
6129 
6130 		/*
6131 		 * Always (transparent or non-transparent )
6132 		 * use the address specified in the request
6133 		 */
6134 		taddr = reqp->cq_addr;
6135 		if (!transparent)
6136 			len = (int)reqp->cq_size;
6137 
6138 		/* copyout data to the provided address */
6139 		error = getiocd(bp, taddr, copyflag);
6140 
6141 		freemsg(bp->b_cont);
6142 		bp->b_cont = NULL;
6143 
6144 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6145 		bp->b_datap->db_type = M_IOCDATA;
6146 
6147 		mblk_setcred(bp, crp);
6148 		DB_CPID(bp) = curproc->p_pid;
6149 		resp = (struct copyresp *)bp->b_rptr;
6150 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6151 		resp->cp_flag = (fflags & FMODELS);
6152 
6153 		stream_willservice(stp);
6154 		putnext(stp->sd_wrq, bp);
6155 		stream_runservice(stp);
6156 
6157 		if (error) {
6158 			mutex_enter(&stp->sd_lock);
6159 			stp->sd_flag &= ~waitflags;
6160 			cv_broadcast(&stp->sd_iocmonitor);
6161 			mutex_exit(&stp->sd_lock);
6162 			crfree(crp);
6163 			return (error);
6164 		}
6165 		goto waitioc;
6166 
6167 	default:
6168 		ASSERT(0);
6169 		mutex_enter(&stp->sd_lock);
6170 		stp->sd_flag &= ~waitflags;
6171 		cv_broadcast(&stp->sd_iocmonitor);
6172 		mutex_exit(&stp->sd_lock);
6173 		break;
6174 	}
6175 
6176 	freemsg(bp);
6177 	crfree(crp);
6178 	return (error);
6179 }
6180 
6181 /*
6182  * For the SunOS keyboard driver.
6183  * Return the next available "ioctl" sequence number.
6184  * Exported, so that streams modules can send "ioctl" messages
6185  * downstream from their open routine.
6186  */
6187 int
6188 getiocseqno(void)
6189 {
6190 	int	i;
6191 
6192 	mutex_enter(&strresources);
6193 	i = ++ioc_id;
6194 	mutex_exit(&strresources);
6195 	return (i);
6196 }
6197 
6198 /*
6199  * Get the next message from the read queue.  If the message is
6200  * priority, STRPRI will have been set by strrput().  This flag
6201  * should be reset only when the entire message at the front of the
6202  * queue as been consumed.
6203  *
6204  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
6205  */
6206 int
6207 strgetmsg(
6208 	struct vnode *vp,
6209 	struct strbuf *mctl,
6210 	struct strbuf *mdata,
6211 	unsigned char *prip,
6212 	int *flagsp,
6213 	int fmode,
6214 	rval_t *rvp)
6215 {
6216 	struct stdata *stp;
6217 	mblk_t *bp, *nbp;
6218 	mblk_t *savemp = NULL;
6219 	mblk_t *savemptail = NULL;
6220 	uint_t old_sd_flag;
6221 	int flg;
6222 	int more = 0;
6223 	int error = 0;
6224 	char first = 1;
6225 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
6226 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
6227 	unsigned char pri = 0;
6228 	queue_t *q;
6229 	int	pr = 0;			/* Partial read successful */
6230 	struct uio uios;
6231 	struct uio *uiop = &uios;
6232 	struct iovec iovs;
6233 	unsigned char type;
6234 
6235 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRGETMSG_ENTER,
6236 		"strgetmsg:%p", vp);
6237 
6238 	ASSERT(vp->v_stream);
6239 	stp = vp->v_stream;
6240 	rvp->r_val1 = 0;
6241 
6242 	mutex_enter(&stp->sd_lock);
6243 
6244 	if ((error = i_straccess(stp, JCREAD)) != 0) {
6245 		mutex_exit(&stp->sd_lock);
6246 		return (error);
6247 	}
6248 
6249 	if (stp->sd_flag & (STRDERR|STPLEX)) {
6250 		error = strgeterr(stp, STRDERR|STPLEX, 0);
6251 		if (error != 0) {
6252 			mutex_exit(&stp->sd_lock);
6253 			return (error);
6254 		}
6255 	}
6256 	mutex_exit(&stp->sd_lock);
6257 
6258 	switch (*flagsp) {
6259 	case MSG_HIPRI:
6260 		if (*prip != 0)
6261 			return (EINVAL);
6262 		break;
6263 
6264 	case MSG_ANY:
6265 	case MSG_BAND:
6266 		break;
6267 
6268 	default:
6269 		return (EINVAL);
6270 	}
6271 	/*
6272 	 * Setup uio and iov for data part
6273 	 */
6274 	iovs.iov_base = mdata->buf;
6275 	iovs.iov_len = mdata->maxlen;
6276 	uios.uio_iov = &iovs;
6277 	uios.uio_iovcnt = 1;
6278 	uios.uio_loffset = 0;
6279 	uios.uio_segflg = UIO_USERSPACE;
6280 	uios.uio_fmode = 0;
6281 	uios.uio_extflg = UIO_COPY_CACHED;
6282 	uios.uio_resid = mdata->maxlen;
6283 	uios.uio_offset = 0;
6284 
6285 	q = _RD(stp->sd_wrq);
6286 	mutex_enter(&stp->sd_lock);
6287 	old_sd_flag = stp->sd_flag;
6288 	mark = 0;
6289 	for (;;) {
6290 		int done = 0;
6291 		mblk_t *q_first = q->q_first;
6292 
6293 		/*
6294 		 * Get the next message of appropriate priority
6295 		 * from the stream head.  If the caller is interested
6296 		 * in band or hipri messages, then they should already
6297 		 * be enqueued at the stream head.  On the other hand
6298 		 * if the caller wants normal (band 0) messages, they
6299 		 * might be deferred in a synchronous stream and they
6300 		 * will need to be pulled up.
6301 		 *
6302 		 * After we have dequeued a message, we might find that
6303 		 * it was a deferred M_SIG that was enqueued at the
6304 		 * stream head.  It must now be posted as part of the
6305 		 * read by calling strsignal_nolock().
6306 		 *
6307 		 * Also note that strrput does not enqueue an M_PCSIG,
6308 		 * and there cannot be more than one hipri message,
6309 		 * so there was no need to have the M_PCSIG case.
6310 		 *
6311 		 * At some time it might be nice to try and wrap the
6312 		 * functionality of kstrgetmsg() and strgetmsg() into
6313 		 * a common routine so to reduce the amount of replicated
6314 		 * code (since they are extremely similar).
6315 		 */
6316 		if (!(*flagsp & (MSG_HIPRI|MSG_BAND))) {
6317 			/* Asking for normal, band0 data */
6318 			bp = strget(stp, q, uiop, first, &error);
6319 			ASSERT(MUTEX_HELD(&stp->sd_lock));
6320 			if (bp != NULL) {
6321 				if (bp->b_datap->db_type == M_SIG) {
6322 					strsignal_nolock(stp, *bp->b_rptr,
6323 					    (int32_t)bp->b_band);
6324 					continue;
6325 				} else {
6326 					break;
6327 				}
6328 			}
6329 			if (error != 0) {
6330 				goto getmout;
6331 			}
6332 
6333 		/*
6334 		 * We can't depend on the value of STRPRI here because
6335 		 * the stream head may be in transit. Therefore, we
6336 		 * must look at the type of the first message to
6337 		 * determine if a high priority messages is waiting
6338 		 */
6339 		} else if ((*flagsp & MSG_HIPRI) && q_first != NULL &&
6340 			    q_first->b_datap->db_type >= QPCTL &&
6341 			    (bp = getq_noenab(q)) != NULL) {
6342 			/* Asked for HIPRI and got one */
6343 			ASSERT(bp->b_datap->db_type >= QPCTL);
6344 			break;
6345 		} else if ((*flagsp & MSG_BAND) && q_first != NULL &&
6346 			    ((q_first->b_band >= *prip) ||
6347 			    q_first->b_datap->db_type >= QPCTL) &&
6348 			    (bp = getq_noenab(q)) != NULL) {
6349 			/*
6350 			 * Asked for at least band "prip" and got either at
6351 			 * least that band or a hipri message.
6352 			 */
6353 			ASSERT(bp->b_band >= *prip ||
6354 				bp->b_datap->db_type >= QPCTL);
6355 			if (bp->b_datap->db_type == M_SIG) {
6356 				strsignal_nolock(stp, *bp->b_rptr,
6357 				    (int32_t)bp->b_band);
6358 				continue;
6359 			} else {
6360 				break;
6361 			}
6362 		}
6363 
6364 		/* No data. Time to sleep? */
6365 		qbackenable(q, 0);
6366 
6367 		/*
6368 		 * If STRHUP or STREOF, return 0 length control and data.
6369 		 * If resid is 0, then a read(fd,buf,0) was done. Do not
6370 		 * sleep to satisfy this request because by default we have
6371 		 * zero bytes to return.
6372 		 */
6373 		if ((stp->sd_flag & (STRHUP|STREOF)) || (mctl->maxlen == 0 &&
6374 		    mdata->maxlen == 0)) {
6375 			mctl->len = mdata->len = 0;
6376 			*flagsp = 0;
6377 			mutex_exit(&stp->sd_lock);
6378 			return (0);
6379 		}
6380 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_WAIT,
6381 			"strgetmsg calls strwaitq:%p, %p",
6382 			vp, uiop);
6383 		if (((error = strwaitq(stp, GETWAIT, (ssize_t)0, fmode, -1,
6384 		    &done)) != 0) || done) {
6385 			TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_DONE,
6386 				"strgetmsg error or done:%p, %p",
6387 				vp, uiop);
6388 			mutex_exit(&stp->sd_lock);
6389 			return (error);
6390 		}
6391 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_AWAKE,
6392 			"strgetmsg awakes:%p, %p", vp, uiop);
6393 		if ((error = i_straccess(stp, JCREAD)) != 0) {
6394 			mutex_exit(&stp->sd_lock);
6395 			return (error);
6396 		}
6397 		first = 0;
6398 	}
6399 	ASSERT(bp != NULL);
6400 	/*
6401 	 * Extract any mark information. If the message is not completely
6402 	 * consumed this information will be put in the mblk
6403 	 * that is putback.
6404 	 * If MSGMARKNEXT is set and the message is completely consumed
6405 	 * the STRATMARK flag will be set below. Likewise, if
6406 	 * MSGNOTMARKNEXT is set and the message is
6407 	 * completely consumed STRNOTATMARK will be set.
6408 	 */
6409 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
6410 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
6411 		(MSGMARKNEXT|MSGNOTMARKNEXT));
6412 	if (mark != 0 && bp == stp->sd_mark) {
6413 		mark |= _LASTMARK;
6414 		stp->sd_mark = NULL;
6415 	}
6416 	/*
6417 	 * keep track of the original message type and priority
6418 	 */
6419 	pri = bp->b_band;
6420 	type = bp->b_datap->db_type;
6421 	if (type == M_PASSFP) {
6422 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
6423 			stp->sd_mark = bp;
6424 		bp->b_flag |= mark & ~_LASTMARK;
6425 		putback(stp, q, bp, pri);
6426 		qbackenable(q, pri);
6427 		mutex_exit(&stp->sd_lock);
6428 		return (EBADMSG);
6429 	}
6430 	ASSERT(type != M_SIG);
6431 
6432 	/*
6433 	 * Set this flag so strrput will not generate signals. Need to
6434 	 * make sure this flag is cleared before leaving this routine
6435 	 * else signals will stop being sent.
6436 	 */
6437 	stp->sd_flag |= STRGETINPROG;
6438 	mutex_exit(&stp->sd_lock);
6439 
6440 	if (STREAM_NEEDSERVICE(stp))
6441 		stream_runservice(stp);
6442 
6443 	/*
6444 	 * Set HIPRI flag if message is priority.
6445 	 */
6446 	if (type >= QPCTL)
6447 		flg = MSG_HIPRI;
6448 	else
6449 		flg = MSG_BAND;
6450 
6451 	/*
6452 	 * First process PROTO or PCPROTO blocks, if any.
6453 	 */
6454 	if (mctl->maxlen >= 0 && type != M_DATA) {
6455 		size_t	n, bcnt;
6456 		char	*ubuf;
6457 
6458 		bcnt = mctl->maxlen;
6459 		ubuf = mctl->buf;
6460 		while (bp != NULL && bp->b_datap->db_type != M_DATA) {
6461 			if ((n = MIN(bcnt, bp->b_wptr - bp->b_rptr)) != 0 &&
6462 			    copyout(bp->b_rptr, ubuf, n)) {
6463 				error = EFAULT;
6464 				mutex_enter(&stp->sd_lock);
6465 				/*
6466 				 * clear stream head pri flag based on
6467 				 * first message type
6468 				 */
6469 				if (type >= QPCTL) {
6470 					ASSERT(type == M_PCPROTO);
6471 					stp->sd_flag &= ~STRPRI;
6472 				}
6473 				more = 0;
6474 				freemsg(bp);
6475 				goto getmout;
6476 			}
6477 			ubuf += n;
6478 			bp->b_rptr += n;
6479 			if (bp->b_rptr >= bp->b_wptr) {
6480 				nbp = bp;
6481 				bp = bp->b_cont;
6482 				freeb(nbp);
6483 			}
6484 			ASSERT(n <= bcnt);
6485 			bcnt -= n;
6486 			if (bcnt == 0)
6487 				break;
6488 		}
6489 		mctl->len = mctl->maxlen - bcnt;
6490 	} else
6491 		mctl->len = -1;
6492 
6493 	if (bp && bp->b_datap->db_type != M_DATA) {
6494 		/*
6495 		 * More PROTO blocks in msg.
6496 		 */
6497 		more |= MORECTL;
6498 		savemp = bp;
6499 		while (bp && bp->b_datap->db_type != M_DATA) {
6500 			savemptail = bp;
6501 			bp = bp->b_cont;
6502 		}
6503 		savemptail->b_cont = NULL;
6504 	}
6505 
6506 	/*
6507 	 * Now process DATA blocks, if any.
6508 	 */
6509 	if (mdata->maxlen >= 0 && bp) {
6510 		/*
6511 		 * struiocopyout will consume a potential zero-length
6512 		 * M_DATA even if uio_resid is zero.
6513 		 */
6514 		size_t oldresid = uiop->uio_resid;
6515 
6516 		bp = struiocopyout(bp, uiop, &error);
6517 		if (error != 0) {
6518 			mutex_enter(&stp->sd_lock);
6519 			/*
6520 			 * clear stream head hi pri flag based on
6521 			 * first message
6522 			 */
6523 			if (type >= QPCTL) {
6524 				ASSERT(type == M_PCPROTO);
6525 				stp->sd_flag &= ~STRPRI;
6526 			}
6527 			more = 0;
6528 			freemsg(savemp);
6529 			goto getmout;
6530 		}
6531 		/*
6532 		 * (pr == 1) indicates a partial read.
6533 		 */
6534 		if (oldresid > uiop->uio_resid)
6535 			pr = 1;
6536 		mdata->len = mdata->maxlen - uiop->uio_resid;
6537 	} else
6538 		mdata->len = -1;
6539 
6540 	if (bp) {			/* more data blocks in msg */
6541 		more |= MOREDATA;
6542 		if (savemp)
6543 			savemptail->b_cont = bp;
6544 		else
6545 			savemp = bp;
6546 	}
6547 
6548 	mutex_enter(&stp->sd_lock);
6549 	if (savemp) {
6550 		if (pr && (savemp->b_datap->db_type == M_DATA) &&
6551 		    msgnodata(savemp)) {
6552 			/*
6553 			 * Avoid queuing a zero-length tail part of
6554 			 * a message. pr=1 indicates that we read some of
6555 			 * the message.
6556 			 */
6557 			freemsg(savemp);
6558 			more &= ~MOREDATA;
6559 			/*
6560 			 * clear stream head hi pri flag based on
6561 			 * first message
6562 			 */
6563 			if (type >= QPCTL) {
6564 				ASSERT(type == M_PCPROTO);
6565 				stp->sd_flag &= ~STRPRI;
6566 			}
6567 		} else {
6568 			savemp->b_band = pri;
6569 			/*
6570 			 * If the first message was HIPRI and the one we're
6571 			 * putting back isn't, then clear STRPRI, otherwise
6572 			 * set STRPRI again.  Note that we must set STRPRI
6573 			 * again since the flush logic in strrput_nondata()
6574 			 * may have cleared it while we had sd_lock dropped.
6575 			 */
6576 			if (type >= QPCTL) {
6577 				ASSERT(type == M_PCPROTO);
6578 				if (queclass(savemp) < QPCTL)
6579 					stp->sd_flag &= ~STRPRI;
6580 				else
6581 					stp->sd_flag |= STRPRI;
6582 			} else if (queclass(savemp) >= QPCTL) {
6583 				/*
6584 				 * The first message was not a HIPRI message,
6585 				 * but the one we are about to putback is.
6586 				 * For simplicitly, we do not allow for HIPRI
6587 				 * messages to be embedded in the message
6588 				 * body, so just force it to same type as
6589 				 * first message.
6590 				 */
6591 				ASSERT(type == M_DATA || type == M_PROTO);
6592 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
6593 				savemp->b_datap->db_type = type;
6594 			}
6595 			if (mark != 0) {
6596 				savemp->b_flag |= mark & ~_LASTMARK;
6597 				if ((mark & _LASTMARK) &&
6598 				    (stp->sd_mark == NULL)) {
6599 					/*
6600 					 * If another marked message arrived
6601 					 * while sd_lock was not held sd_mark
6602 					 * would be non-NULL.
6603 					 */
6604 					stp->sd_mark = savemp;
6605 				}
6606 			}
6607 			putback(stp, q, savemp, pri);
6608 		}
6609 	} else {
6610 		/*
6611 		 * The complete message was consumed.
6612 		 *
6613 		 * If another M_PCPROTO arrived while sd_lock was not held
6614 		 * it would have been discarded since STRPRI was still set.
6615 		 *
6616 		 * Move the MSG*MARKNEXT information
6617 		 * to the stream head just in case
6618 		 * the read queue becomes empty.
6619 		 * clear stream head hi pri flag based on
6620 		 * first message
6621 		 *
6622 		 * If the stream head was at the mark
6623 		 * (STRATMARK) before we dropped sd_lock above
6624 		 * and some data was consumed then we have
6625 		 * moved past the mark thus STRATMARK is
6626 		 * cleared. However, if a message arrived in
6627 		 * strrput during the copyout above causing
6628 		 * STRATMARK to be set we can not clear that
6629 		 * flag.
6630 		 */
6631 		if (type >= QPCTL) {
6632 			ASSERT(type == M_PCPROTO);
6633 			stp->sd_flag &= ~STRPRI;
6634 		}
6635 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
6636 			if (mark & MSGMARKNEXT) {
6637 				stp->sd_flag &= ~STRNOTATMARK;
6638 				stp->sd_flag |= STRATMARK;
6639 			} else if (mark & MSGNOTMARKNEXT) {
6640 				stp->sd_flag &= ~STRATMARK;
6641 				stp->sd_flag |= STRNOTATMARK;
6642 			} else {
6643 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
6644 			}
6645 		} else if (pr && (old_sd_flag & STRATMARK)) {
6646 			stp->sd_flag &= ~STRATMARK;
6647 		}
6648 	}
6649 
6650 	*flagsp = flg;
6651 	*prip = pri;
6652 
6653 	/*
6654 	 * Getmsg cleanup processing - if the state of the queue has changed
6655 	 * some signals may need to be sent and/or poll awakened.
6656 	 */
6657 getmout:
6658 	qbackenable(q, pri);
6659 
6660 	/*
6661 	 * We dropped the stream head lock above. Send all M_SIG messages
6662 	 * before processing stream head for SIGPOLL messages.
6663 	 */
6664 	ASSERT(MUTEX_HELD(&stp->sd_lock));
6665 	while ((bp = q->q_first) != NULL &&
6666 	    (bp->b_datap->db_type == M_SIG)) {
6667 		/*
6668 		 * sd_lock is held so the content of the read queue can not
6669 		 * change.
6670 		 */
6671 		bp = getq(q);
6672 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
6673 
6674 		strsignal_nolock(stp, *bp->b_rptr, (int32_t)bp->b_band);
6675 		mutex_exit(&stp->sd_lock);
6676 		freemsg(bp);
6677 		if (STREAM_NEEDSERVICE(stp))
6678 			stream_runservice(stp);
6679 		mutex_enter(&stp->sd_lock);
6680 	}
6681 
6682 	/*
6683 	 * stream head cannot change while we make the determination
6684 	 * whether or not to send a signal. Drop the flag to allow strrput
6685 	 * to send firstmsgsigs again.
6686 	 */
6687 	stp->sd_flag &= ~STRGETINPROG;
6688 
6689 	/*
6690 	 * If the type of message at the front of the queue changed
6691 	 * due to the receive the appropriate signals and pollwakeup events
6692 	 * are generated. The type of changes are:
6693 	 *	Processed a hipri message, q_first is not hipri.
6694 	 *	Processed a band X message, and q_first is band Y.
6695 	 * The generated signals and pollwakeups are identical to what
6696 	 * strrput() generates should the message that is now on q_first
6697 	 * arrive to an empty read queue.
6698 	 *
6699 	 * Note: only strrput will send a signal for a hipri message.
6700 	 */
6701 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
6702 		strsigset_t signals = 0;
6703 		strpollset_t pollwakeups = 0;
6704 
6705 		if (flg & MSG_HIPRI) {
6706 			/*
6707 			 * Removed a hipri message. Regular data at
6708 			 * the front of  the queue.
6709 			 */
6710 			if (bp->b_band == 0) {
6711 				signals = S_INPUT | S_RDNORM;
6712 				pollwakeups = POLLIN | POLLRDNORM;
6713 			} else {
6714 				signals = S_INPUT | S_RDBAND;
6715 				pollwakeups = POLLIN | POLLRDBAND;
6716 			}
6717 		} else if (pri != bp->b_band) {
6718 			/*
6719 			 * The band is different for the new q_first.
6720 			 */
6721 			if (bp->b_band == 0) {
6722 				signals = S_RDNORM;
6723 				pollwakeups = POLLIN | POLLRDNORM;
6724 			} else {
6725 				signals = S_RDBAND;
6726 				pollwakeups = POLLIN | POLLRDBAND;
6727 			}
6728 		}
6729 
6730 		if (pollwakeups != 0) {
6731 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
6732 				if (!(stp->sd_rput_opt & SR_POLLIN))
6733 					goto no_pollwake;
6734 				stp->sd_rput_opt &= ~SR_POLLIN;
6735 			}
6736 			mutex_exit(&stp->sd_lock);
6737 			pollwakeup(&stp->sd_pollist, pollwakeups);
6738 			mutex_enter(&stp->sd_lock);
6739 		}
6740 no_pollwake:
6741 
6742 		if (stp->sd_sigflags & signals)
6743 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
6744 	}
6745 	mutex_exit(&stp->sd_lock);
6746 
6747 	rvp->r_val1 = more;
6748 	return (error);
6749 #undef	_LASTMARK
6750 }
6751 
6752 /*
6753  * Get the next message from the read queue.  If the message is
6754  * priority, STRPRI will have been set by strrput().  This flag
6755  * should be reset only when the entire message at the front of the
6756  * queue as been consumed.
6757  *
6758  * If uiop is NULL all data is returned in mctlp.
6759  * Note that a NULL uiop implies that FNDELAY and FNONBLOCK are assumed
6760  * not enabled.
6761  * The timeout parameter is in milliseconds; -1 for infinity.
6762  * This routine handles the consolidation private flags:
6763  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
6764  *	MSG_DELAYERROR	Defer the error check until the queue is empty.
6765  *	MSG_HOLDSIG	Hold signals while waiting for data.
6766  *	MSG_IPEEK	Only peek at messages.
6767  *	MSG_DISCARDTAIL	Discard the tail M_DATA part of the message
6768  *			that doesn't fit.
6769  *	MSG_NOMARK	If the message is marked leave it on the queue.
6770  *
6771  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
6772  */
6773 int
6774 kstrgetmsg(
6775 	struct vnode *vp,
6776 	mblk_t **mctlp,
6777 	struct uio *uiop,
6778 	unsigned char *prip,
6779 	int *flagsp,
6780 	clock_t timout,
6781 	rval_t *rvp)
6782 {
6783 	struct stdata *stp;
6784 	mblk_t *bp, *nbp;
6785 	mblk_t *savemp = NULL;
6786 	mblk_t *savemptail = NULL;
6787 	int flags;
6788 	uint_t old_sd_flag;
6789 	int flg;
6790 	int more = 0;
6791 	int error = 0;
6792 	char first = 1;
6793 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
6794 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
6795 	unsigned char pri = 0;
6796 	queue_t *q;
6797 	int	pr = 0;			/* Partial read successful */
6798 	unsigned char type;
6799 
6800 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_ENTER,
6801 		"kstrgetmsg:%p", vp);
6802 
6803 	ASSERT(vp->v_stream);
6804 	stp = vp->v_stream;
6805 	rvp->r_val1 = 0;
6806 
6807 	mutex_enter(&stp->sd_lock);
6808 
6809 	if ((error = i_straccess(stp, JCREAD)) != 0) {
6810 		mutex_exit(&stp->sd_lock);
6811 		return (error);
6812 	}
6813 
6814 	flags = *flagsp;
6815 	if (stp->sd_flag & (STRDERR|STPLEX)) {
6816 		if ((stp->sd_flag & STPLEX) ||
6817 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == 0) {
6818 			error = strgeterr(stp, STRDERR|STPLEX,
6819 					(flags & MSG_IPEEK));
6820 			if (error != 0) {
6821 				mutex_exit(&stp->sd_lock);
6822 				return (error);
6823 			}
6824 		}
6825 	}
6826 	mutex_exit(&stp->sd_lock);
6827 
6828 	switch (flags & (MSG_HIPRI|MSG_ANY|MSG_BAND)) {
6829 	case MSG_HIPRI:
6830 		if (*prip != 0)
6831 			return (EINVAL);
6832 		break;
6833 
6834 	case MSG_ANY:
6835 	case MSG_BAND:
6836 		break;
6837 
6838 	default:
6839 		return (EINVAL);
6840 	}
6841 
6842 retry:
6843 	q = _RD(stp->sd_wrq);
6844 	mutex_enter(&stp->sd_lock);
6845 	old_sd_flag = stp->sd_flag;
6846 	mark = 0;
6847 	for (;;) {
6848 		int done = 0;
6849 		int waitflag;
6850 		int fmode;
6851 		mblk_t *q_first = q->q_first;
6852 
6853 		/*
6854 		 * This section of the code operates just like the code
6855 		 * in strgetmsg().  There is a comment there about what
6856 		 * is going on here.
6857 		 */
6858 		if (!(flags & (MSG_HIPRI|MSG_BAND))) {
6859 			/* Asking for normal, band0 data */
6860 			bp = strget(stp, q, uiop, first, &error);
6861 			ASSERT(MUTEX_HELD(&stp->sd_lock));
6862 			if (bp != NULL) {
6863 				if (bp->b_datap->db_type == M_SIG) {
6864 					strsignal_nolock(stp, *bp->b_rptr,
6865 					    (int32_t)bp->b_band);
6866 					continue;
6867 				} else {
6868 					break;
6869 				}
6870 			}
6871 			if (error != 0) {
6872 				goto getmout;
6873 			}
6874 		/*
6875 		 * We can't depend on the value of STRPRI here because
6876 		 * the stream head may be in transit. Therefore, we
6877 		 * must look at the type of the first message to
6878 		 * determine if a high priority messages is waiting
6879 		 */
6880 		} else if ((flags & MSG_HIPRI) && q_first != NULL &&
6881 			    q_first->b_datap->db_type >= QPCTL &&
6882 			    (bp = getq_noenab(q)) != NULL) {
6883 			ASSERT(bp->b_datap->db_type >= QPCTL);
6884 			break;
6885 		} else if ((flags & MSG_BAND) && q_first != NULL &&
6886 			    ((q_first->b_band >= *prip) ||
6887 			    q_first->b_datap->db_type >= QPCTL) &&
6888 			    (bp = getq_noenab(q)) != NULL) {
6889 			/*
6890 			 * Asked for at least band "prip" and got either at
6891 			 * least that band or a hipri message.
6892 			 */
6893 			ASSERT(bp->b_band >= *prip ||
6894 				bp->b_datap->db_type >= QPCTL);
6895 			if (bp->b_datap->db_type == M_SIG) {
6896 				strsignal_nolock(stp, *bp->b_rptr,
6897 				    (int32_t)bp->b_band);
6898 				continue;
6899 			} else {
6900 				break;
6901 			}
6902 		}
6903 
6904 		/* No data. Time to sleep? */
6905 		qbackenable(q, 0);
6906 
6907 		/*
6908 		 * Delayed error notification?
6909 		 */
6910 		if ((stp->sd_flag & (STRDERR|STPLEX)) &&
6911 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == MSG_DELAYERROR) {
6912 			error = strgeterr(stp, STRDERR|STPLEX,
6913 					(flags & MSG_IPEEK));
6914 			if (error != 0) {
6915 				mutex_exit(&stp->sd_lock);
6916 				return (error);
6917 			}
6918 		}
6919 
6920 		/*
6921 		 * If STRHUP or STREOF, return 0 length control and data.
6922 		 * If a read(fd,buf,0) has been done, do not sleep, just
6923 		 * return.
6924 		 *
6925 		 * If mctlp == NULL and uiop == NULL, then the code will
6926 		 * do the strwaitq. This is an understood way of saying
6927 		 * sleep "polling" until a message is received.
6928 		 */
6929 		if ((stp->sd_flag & (STRHUP|STREOF)) ||
6930 		    (uiop != NULL && uiop->uio_resid == 0)) {
6931 			if (mctlp != NULL)
6932 				*mctlp = NULL;
6933 			*flagsp = 0;
6934 			mutex_exit(&stp->sd_lock);
6935 			return (0);
6936 		}
6937 
6938 		waitflag = GETWAIT;
6939 		if (flags &
6940 		    (MSG_HOLDSIG|MSG_IGNERROR|MSG_IPEEK|MSG_DELAYERROR)) {
6941 			if (flags & MSG_HOLDSIG)
6942 				waitflag |= STR_NOSIG;
6943 			if (flags & MSG_IGNERROR)
6944 				waitflag |= STR_NOERROR;
6945 			if (flags & MSG_IPEEK)
6946 				waitflag |= STR_PEEK;
6947 			if (flags & MSG_DELAYERROR)
6948 				waitflag |= STR_DELAYERR;
6949 		}
6950 		if (uiop != NULL)
6951 			fmode = uiop->uio_fmode;
6952 		else
6953 			fmode = 0;
6954 
6955 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_WAIT,
6956 			"kstrgetmsg calls strwaitq:%p, %p",
6957 			vp, uiop);
6958 		if (((error = strwaitq(stp, waitflag, (ssize_t)0,
6959 		    fmode, timout, &done)) != 0) || done) {
6960 			TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_DONE,
6961 				"kstrgetmsg error or done:%p, %p",
6962 				vp, uiop);
6963 			mutex_exit(&stp->sd_lock);
6964 			return (error);
6965 		}
6966 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_AWAKE,
6967 			"kstrgetmsg awakes:%p, %p", vp, uiop);
6968 		if ((error = i_straccess(stp, JCREAD)) != 0) {
6969 			mutex_exit(&stp->sd_lock);
6970 			return (error);
6971 		}
6972 		first = 0;
6973 	}
6974 	ASSERT(bp != NULL);
6975 	/*
6976 	 * Extract any mark information. If the message is not completely
6977 	 * consumed this information will be put in the mblk
6978 	 * that is putback.
6979 	 * If MSGMARKNEXT is set and the message is completely consumed
6980 	 * the STRATMARK flag will be set below. Likewise, if
6981 	 * MSGNOTMARKNEXT is set and the message is
6982 	 * completely consumed STRNOTATMARK will be set.
6983 	 */
6984 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
6985 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
6986 		(MSGMARKNEXT|MSGNOTMARKNEXT));
6987 	pri = bp->b_band;
6988 	if (mark != 0) {
6989 		/*
6990 		 * If the caller doesn't want the mark return.
6991 		 * Used to implement MSG_WAITALL in sockets.
6992 		 */
6993 		if (flags & MSG_NOMARK) {
6994 			putback(stp, q, bp, pri);
6995 			qbackenable(q, pri);
6996 			mutex_exit(&stp->sd_lock);
6997 			return (EWOULDBLOCK);
6998 		}
6999 		if (bp == stp->sd_mark) {
7000 			mark |= _LASTMARK;
7001 			stp->sd_mark = NULL;
7002 		}
7003 	}
7004 
7005 	/*
7006 	 * keep track of the first message type
7007 	 */
7008 	type = bp->b_datap->db_type;
7009 
7010 	if (bp->b_datap->db_type == M_PASSFP) {
7011 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7012 			stp->sd_mark = bp;
7013 		bp->b_flag |= mark & ~_LASTMARK;
7014 		putback(stp, q, bp, pri);
7015 		qbackenable(q, pri);
7016 		mutex_exit(&stp->sd_lock);
7017 		return (EBADMSG);
7018 	}
7019 	ASSERT(type != M_SIG);
7020 
7021 	if (flags & MSG_IPEEK) {
7022 		/*
7023 		 * Clear any struioflag - we do the uiomove over again
7024 		 * when peeking since it simplifies the code.
7025 		 *
7026 		 * Dup the message and put the original back on the queue.
7027 		 * If dupmsg() fails, try again with copymsg() to see if
7028 		 * there is indeed a shortage of memory.  dupmsg() may fail
7029 		 * if db_ref in any of the messages reaches its limit.
7030 		 */
7031 		if ((nbp = dupmsg(bp)) == NULL && (nbp = copymsg(bp)) == NULL) {
7032 			/*
7033 			 * Restore the state of the stream head since we
7034 			 * need to drop sd_lock (strwaitbuf is sleeping).
7035 			 */
7036 			size_t size = msgdsize(bp);
7037 
7038 			if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7039 				stp->sd_mark = bp;
7040 			bp->b_flag |= mark & ~_LASTMARK;
7041 			putback(stp, q, bp, pri);
7042 			mutex_exit(&stp->sd_lock);
7043 			error = strwaitbuf(size, BPRI_HI);
7044 			if (error) {
7045 				/*
7046 				 * There is no net change to the queue thus
7047 				 * no need to qbackenable.
7048 				 */
7049 				return (error);
7050 			}
7051 			goto retry;
7052 		}
7053 
7054 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7055 			stp->sd_mark = bp;
7056 		bp->b_flag |= mark & ~_LASTMARK;
7057 		putback(stp, q, bp, pri);
7058 		bp = nbp;
7059 	}
7060 
7061 	/*
7062 	 * Set this flag so strrput will not generate signals. Need to
7063 	 * make sure this flag is cleared before leaving this routine
7064 	 * else signals will stop being sent.
7065 	 */
7066 	stp->sd_flag |= STRGETINPROG;
7067 	mutex_exit(&stp->sd_lock);
7068 
7069 	if ((stp->sd_rputdatafunc != NULL) && (DB_TYPE(bp) == M_DATA) &&
7070 	    (!(DB_FLAGS(bp) & DBLK_COOKED))) {
7071 
7072 		bp = (stp->sd_rputdatafunc)(
7073 		    stp->sd_vnode, bp, NULL,
7074 		    NULL, NULL, NULL);
7075 
7076 		if (bp == NULL)
7077 			goto retry;
7078 
7079 		DB_FLAGS(bp) |= DBLK_COOKED;
7080 	}
7081 
7082 	if (STREAM_NEEDSERVICE(stp))
7083 		stream_runservice(stp);
7084 
7085 	/*
7086 	 * Set HIPRI flag if message is priority.
7087 	 */
7088 	if (type >= QPCTL)
7089 		flg = MSG_HIPRI;
7090 	else
7091 		flg = MSG_BAND;
7092 
7093 	/*
7094 	 * First process PROTO or PCPROTO blocks, if any.
7095 	 */
7096 	if (mctlp != NULL && type != M_DATA) {
7097 		mblk_t *nbp;
7098 
7099 		*mctlp = bp;
7100 		while (bp->b_cont && bp->b_cont->b_datap->db_type != M_DATA)
7101 			bp = bp->b_cont;
7102 		nbp = bp->b_cont;
7103 		bp->b_cont = NULL;
7104 		bp = nbp;
7105 	}
7106 
7107 	if (bp && bp->b_datap->db_type != M_DATA) {
7108 		/*
7109 		 * More PROTO blocks in msg. Will only happen if mctlp is NULL.
7110 		 */
7111 		more |= MORECTL;
7112 		savemp = bp;
7113 		while (bp && bp->b_datap->db_type != M_DATA) {
7114 			savemptail = bp;
7115 			bp = bp->b_cont;
7116 		}
7117 		savemptail->b_cont = NULL;
7118 	}
7119 
7120 	/*
7121 	 * Now process DATA blocks, if any.
7122 	 */
7123 	if (uiop == NULL) {
7124 		/* Append data to tail of mctlp */
7125 		if (mctlp != NULL) {
7126 			mblk_t **mpp = mctlp;
7127 
7128 			while (*mpp != NULL)
7129 				mpp = &((*mpp)->b_cont);
7130 			*mpp = bp;
7131 			bp = NULL;
7132 		}
7133 	} else if (uiop->uio_resid >= 0 && bp) {
7134 		size_t oldresid = uiop->uio_resid;
7135 
7136 		/*
7137 		 * If a streams message is likely to consist
7138 		 * of many small mblks, it is pulled up into
7139 		 * one continuous chunk of memory.
7140 		 * see longer comment at top of page
7141 		 * by mblk_pull_len declaration.
7142 		 */
7143 
7144 		if (MBLKL(bp) < mblk_pull_len) {
7145 			(void) pullupmsg(bp, -1);
7146 		}
7147 
7148 		bp = struiocopyout(bp, uiop, &error);
7149 		if (error != 0) {
7150 			if (mctlp != NULL) {
7151 				freemsg(*mctlp);
7152 				*mctlp = NULL;
7153 			} else
7154 				freemsg(savemp);
7155 			mutex_enter(&stp->sd_lock);
7156 			/*
7157 			 * clear stream head hi pri flag based on
7158 			 * first message
7159 			 */
7160 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7161 				ASSERT(type == M_PCPROTO);
7162 				stp->sd_flag &= ~STRPRI;
7163 			}
7164 			more = 0;
7165 			goto getmout;
7166 		}
7167 		/*
7168 		 * (pr == 1) indicates a partial read.
7169 		 */
7170 		if (oldresid > uiop->uio_resid)
7171 			pr = 1;
7172 	}
7173 
7174 	if (bp) {			/* more data blocks in msg */
7175 		more |= MOREDATA;
7176 		if (savemp)
7177 			savemptail->b_cont = bp;
7178 		else
7179 			savemp = bp;
7180 	}
7181 
7182 	mutex_enter(&stp->sd_lock);
7183 	if (savemp) {
7184 		if (flags & (MSG_IPEEK|MSG_DISCARDTAIL)) {
7185 			/*
7186 			 * When MSG_DISCARDTAIL is set or
7187 			 * when peeking discard any tail. When peeking this
7188 			 * is the tail of the dup that was copied out - the
7189 			 * message has already been putback on the queue.
7190 			 * Return MOREDATA to the caller even though the data
7191 			 * is discarded. This is used by sockets (to
7192 			 * set MSG_TRUNC).
7193 			 */
7194 			freemsg(savemp);
7195 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7196 				ASSERT(type == M_PCPROTO);
7197 				stp->sd_flag &= ~STRPRI;
7198 			}
7199 		} else if (pr && (savemp->b_datap->db_type == M_DATA) &&
7200 			    msgnodata(savemp)) {
7201 			/*
7202 			 * Avoid queuing a zero-length tail part of
7203 			 * a message. pr=1 indicates that we read some of
7204 			 * the message.
7205 			 */
7206 			freemsg(savemp);
7207 			more &= ~MOREDATA;
7208 			if (type >= QPCTL) {
7209 				ASSERT(type == M_PCPROTO);
7210 				stp->sd_flag &= ~STRPRI;
7211 			}
7212 		} else {
7213 			savemp->b_band = pri;
7214 			/*
7215 			 * If the first message was HIPRI and the one we're
7216 			 * putting back isn't, then clear STRPRI, otherwise
7217 			 * set STRPRI again.  Note that we must set STRPRI
7218 			 * again since the flush logic in strrput_nondata()
7219 			 * may have cleared it while we had sd_lock dropped.
7220 			 */
7221 			if (type >= QPCTL) {
7222 				ASSERT(type == M_PCPROTO);
7223 				if (queclass(savemp) < QPCTL)
7224 					stp->sd_flag &= ~STRPRI;
7225 				else
7226 					stp->sd_flag |= STRPRI;
7227 			} else if (queclass(savemp) >= QPCTL) {
7228 				/*
7229 				 * The first message was not a HIPRI message,
7230 				 * but the one we are about to putback is.
7231 				 * For simplicitly, we do not allow for HIPRI
7232 				 * messages to be embedded in the message
7233 				 * body, so just force it to same type as
7234 				 * first message.
7235 				 */
7236 				ASSERT(type == M_DATA || type == M_PROTO);
7237 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
7238 				savemp->b_datap->db_type = type;
7239 			}
7240 			if (mark != 0) {
7241 				if ((mark & _LASTMARK) &&
7242 				    (stp->sd_mark == NULL)) {
7243 					/*
7244 					 * If another marked message arrived
7245 					 * while sd_lock was not held sd_mark
7246 					 * would be non-NULL.
7247 					 */
7248 					stp->sd_mark = savemp;
7249 				}
7250 				savemp->b_flag |= mark & ~_LASTMARK;
7251 			}
7252 			putback(stp, q, savemp, pri);
7253 		}
7254 	} else if (!(flags & MSG_IPEEK)) {
7255 		/*
7256 		 * The complete message was consumed.
7257 		 *
7258 		 * If another M_PCPROTO arrived while sd_lock was not held
7259 		 * it would have been discarded since STRPRI was still set.
7260 		 *
7261 		 * Move the MSG*MARKNEXT information
7262 		 * to the stream head just in case
7263 		 * the read queue becomes empty.
7264 		 * clear stream head hi pri flag based on
7265 		 * first message
7266 		 *
7267 		 * If the stream head was at the mark
7268 		 * (STRATMARK) before we dropped sd_lock above
7269 		 * and some data was consumed then we have
7270 		 * moved past the mark thus STRATMARK is
7271 		 * cleared. However, if a message arrived in
7272 		 * strrput during the copyout above causing
7273 		 * STRATMARK to be set we can not clear that
7274 		 * flag.
7275 		 * XXX A "perimeter" would help by single-threading strrput,
7276 		 * strread, strgetmsg and kstrgetmsg.
7277 		 */
7278 		if (type >= QPCTL) {
7279 			ASSERT(type == M_PCPROTO);
7280 			stp->sd_flag &= ~STRPRI;
7281 		}
7282 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
7283 			if (mark & MSGMARKNEXT) {
7284 				stp->sd_flag &= ~STRNOTATMARK;
7285 				stp->sd_flag |= STRATMARK;
7286 			} else if (mark & MSGNOTMARKNEXT) {
7287 				stp->sd_flag &= ~STRATMARK;
7288 				stp->sd_flag |= STRNOTATMARK;
7289 			} else {
7290 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
7291 			}
7292 		} else if (pr && (old_sd_flag & STRATMARK)) {
7293 			stp->sd_flag &= ~STRATMARK;
7294 		}
7295 	}
7296 
7297 	*flagsp = flg;
7298 	*prip = pri;
7299 
7300 	/*
7301 	 * Getmsg cleanup processing - if the state of the queue has changed
7302 	 * some signals may need to be sent and/or poll awakened.
7303 	 */
7304 getmout:
7305 	qbackenable(q, pri);
7306 
7307 	/*
7308 	 * We dropped the stream head lock above. Send all M_SIG messages
7309 	 * before processing stream head for SIGPOLL messages.
7310 	 */
7311 	ASSERT(MUTEX_HELD(&stp->sd_lock));
7312 	while ((bp = q->q_first) != NULL &&
7313 	    (bp->b_datap->db_type == M_SIG)) {
7314 		/*
7315 		 * sd_lock is held so the content of the read queue can not
7316 		 * change.
7317 		 */
7318 		bp = getq(q);
7319 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7320 
7321 		strsignal_nolock(stp, *bp->b_rptr, (int32_t)bp->b_band);
7322 		mutex_exit(&stp->sd_lock);
7323 		freemsg(bp);
7324 		if (STREAM_NEEDSERVICE(stp))
7325 			stream_runservice(stp);
7326 		mutex_enter(&stp->sd_lock);
7327 	}
7328 
7329 	/*
7330 	 * stream head cannot change while we make the determination
7331 	 * whether or not to send a signal. Drop the flag to allow strrput
7332 	 * to send firstmsgsigs again.
7333 	 */
7334 	stp->sd_flag &= ~STRGETINPROG;
7335 
7336 	/*
7337 	 * If the type of message at the front of the queue changed
7338 	 * due to the receive the appropriate signals and pollwakeup events
7339 	 * are generated. The type of changes are:
7340 	 *	Processed a hipri message, q_first is not hipri.
7341 	 *	Processed a band X message, and q_first is band Y.
7342 	 * The generated signals and pollwakeups are identical to what
7343 	 * strrput() generates should the message that is now on q_first
7344 	 * arrive to an empty read queue.
7345 	 *
7346 	 * Note: only strrput will send a signal for a hipri message.
7347 	 */
7348 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7349 		strsigset_t signals = 0;
7350 		strpollset_t pollwakeups = 0;
7351 
7352 		if (flg & MSG_HIPRI) {
7353 			/*
7354 			 * Removed a hipri message. Regular data at
7355 			 * the front of  the queue.
7356 			 */
7357 			if (bp->b_band == 0) {
7358 				signals = S_INPUT | S_RDNORM;
7359 				pollwakeups = POLLIN | POLLRDNORM;
7360 			} else {
7361 				signals = S_INPUT | S_RDBAND;
7362 				pollwakeups = POLLIN | POLLRDBAND;
7363 			}
7364 		} else if (pri != bp->b_band) {
7365 			/*
7366 			 * The band is different for the new q_first.
7367 			 */
7368 			if (bp->b_band == 0) {
7369 				signals = S_RDNORM;
7370 				pollwakeups = POLLIN | POLLRDNORM;
7371 			} else {
7372 				signals = S_RDBAND;
7373 				pollwakeups = POLLIN | POLLRDBAND;
7374 			}
7375 		}
7376 
7377 		if (pollwakeups != 0) {
7378 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
7379 				if (!(stp->sd_rput_opt & SR_POLLIN))
7380 					goto no_pollwake;
7381 				stp->sd_rput_opt &= ~SR_POLLIN;
7382 			}
7383 			mutex_exit(&stp->sd_lock);
7384 			pollwakeup(&stp->sd_pollist, pollwakeups);
7385 			mutex_enter(&stp->sd_lock);
7386 		}
7387 no_pollwake:
7388 
7389 		if (stp->sd_sigflags & signals)
7390 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7391 	}
7392 	mutex_exit(&stp->sd_lock);
7393 
7394 	rvp->r_val1 = more;
7395 	return (error);
7396 #undef	_LASTMARK
7397 }
7398 
7399 /*
7400  * Put a message downstream.
7401  *
7402  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7403  */
7404 int
7405 strputmsg(
7406 	struct vnode *vp,
7407 	struct strbuf *mctl,
7408 	struct strbuf *mdata,
7409 	unsigned char pri,
7410 	int flag,
7411 	int fmode)
7412 {
7413 	struct stdata *stp;
7414 	queue_t *wqp;
7415 	mblk_t *mp;
7416 	ssize_t msgsize;
7417 	ssize_t rmin, rmax;
7418 	int error;
7419 	struct uio uios;
7420 	struct uio *uiop = &uios;
7421 	struct iovec iovs;
7422 	int xpg4 = 0;
7423 
7424 	ASSERT(vp->v_stream);
7425 	stp = vp->v_stream;
7426 	wqp = stp->sd_wrq;
7427 
7428 	/*
7429 	 * If it is an XPG4 application, we need to send
7430 	 * SIGPIPE below
7431 	 */
7432 
7433 	xpg4 = (flag & MSG_XPG4) ? 1 : 0;
7434 	flag &= ~MSG_XPG4;
7435 
7436 #ifdef C2_AUDIT
7437 	if (audit_active)
7438 		audit_strputmsg(vp, mctl, mdata, pri, flag, fmode);
7439 #endif
7440 
7441 	mutex_enter(&stp->sd_lock);
7442 
7443 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
7444 		mutex_exit(&stp->sd_lock);
7445 		return (error);
7446 	}
7447 
7448 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7449 		error = strwriteable(stp, B_FALSE, xpg4);
7450 		if (error != 0) {
7451 			mutex_exit(&stp->sd_lock);
7452 			return (error);
7453 		}
7454 	}
7455 
7456 	mutex_exit(&stp->sd_lock);
7457 
7458 	/*
7459 	 * Check for legal flag value.
7460 	 */
7461 	switch (flag) {
7462 	case MSG_HIPRI:
7463 		if ((mctl->len < 0) || (pri != 0))
7464 			return (EINVAL);
7465 		break;
7466 	case MSG_BAND:
7467 		break;
7468 
7469 	default:
7470 		return (EINVAL);
7471 	}
7472 
7473 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_IN,
7474 		"strputmsg in:stp %p", stp);
7475 
7476 	/* get these values from those cached in the stream head */
7477 	rmin = stp->sd_qn_minpsz;
7478 	rmax = stp->sd_qn_maxpsz;
7479 
7480 	/*
7481 	 * Make sure ctl and data sizes together fall within the
7482 	 * limits of the max and min receive packet sizes and do
7483 	 * not exceed system limit.
7484 	 */
7485 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
7486 	if (rmax == 0) {
7487 		return (ERANGE);
7488 	}
7489 	/*
7490 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
7491 	 * Needed to prevent partial failures in the strmakedata loop.
7492 	 */
7493 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
7494 		rmax = stp->sd_maxblk;
7495 
7496 	if ((msgsize = mdata->len) < 0) {
7497 		msgsize = 0;
7498 		rmin = 0;	/* no range check for NULL data part */
7499 	}
7500 	if ((msgsize < rmin) ||
7501 	    ((msgsize > rmax) && (rmax != INFPSZ)) ||
7502 	    (mctl->len > strctlsz)) {
7503 		return (ERANGE);
7504 	}
7505 
7506 	/*
7507 	 * Setup uio and iov for data part
7508 	 */
7509 	iovs.iov_base = mdata->buf;
7510 	iovs.iov_len = msgsize;
7511 	uios.uio_iov = &iovs;
7512 	uios.uio_iovcnt = 1;
7513 	uios.uio_loffset = 0;
7514 	uios.uio_segflg = UIO_USERSPACE;
7515 	uios.uio_fmode = fmode;
7516 	uios.uio_extflg = UIO_COPY_DEFAULT;
7517 	uios.uio_resid = msgsize;
7518 	uios.uio_offset = 0;
7519 
7520 	/* Ignore flow control in strput for HIPRI */
7521 	if (flag & MSG_HIPRI)
7522 		flag |= MSG_IGNFLOW;
7523 
7524 	for (;;) {
7525 		int done = 0;
7526 
7527 		/*
7528 		 * strput will always free the ctl mblk - even when strput
7529 		 * fails.
7530 		 */
7531 		if ((error = strmakectl(mctl, flag, fmode, &mp)) != 0) {
7532 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7533 				"strputmsg out:stp %p out %d error %d",
7534 				stp, 1, error);
7535 			return (error);
7536 		}
7537 		/*
7538 		 * Verify that the whole message can be transferred by
7539 		 * strput.
7540 		 */
7541 		ASSERT(stp->sd_maxblk == INFPSZ ||
7542 			stp->sd_maxblk >= mdata->len);
7543 
7544 		msgsize = mdata->len;
7545 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
7546 		mdata->len = msgsize;
7547 
7548 		if (error == 0)
7549 			break;
7550 
7551 		if (error != EWOULDBLOCK)
7552 			goto out;
7553 
7554 		mutex_enter(&stp->sd_lock);
7555 		/*
7556 		 * Check for a missed wakeup.
7557 		 * Needed since strput did not hold sd_lock across
7558 		 * the canputnext.
7559 		 */
7560 		if (bcanputnext(wqp, pri)) {
7561 			/* Try again */
7562 			mutex_exit(&stp->sd_lock);
7563 			continue;
7564 		}
7565 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAIT,
7566 			"strputmsg wait:stp %p waits pri %d", stp, pri);
7567 		if (((error = strwaitq(stp, WRITEWAIT, (ssize_t)0, fmode, -1,
7568 		    &done)) != 0) || done) {
7569 			mutex_exit(&stp->sd_lock);
7570 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7571 				"strputmsg out:q %p out %d error %d",
7572 				stp, 0, error);
7573 			return (error);
7574 		}
7575 		TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAKE,
7576 			"strputmsg wake:stp %p wakes", stp);
7577 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
7578 			mutex_exit(&stp->sd_lock);
7579 			return (error);
7580 		}
7581 		mutex_exit(&stp->sd_lock);
7582 	}
7583 out:
7584 	/*
7585 	 * For historic reasons, applications expect EAGAIN
7586 	 * when data mblk could not be allocated. so change
7587 	 * ENOMEM back to EAGAIN
7588 	 */
7589 	if (error == ENOMEM)
7590 		error = EAGAIN;
7591 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7592 		"strputmsg out:stp %p out %d error %d", stp, 2, error);
7593 	return (error);
7594 }
7595 
7596 /*
7597  * Put a message downstream.
7598  * Can send only an M_PROTO/M_PCPROTO by passing in a NULL uiop.
7599  * The fmode flag (NDELAY, NONBLOCK) is the or of the flags in the uio
7600  * and the fmode parameter.
7601  *
7602  * This routine handles the consolidation private flags:
7603  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
7604  *	MSG_HOLDSIG	Hold signals while waiting for data.
7605  *	MSG_IGNFLOW	Don't check streams flow control.
7606  *
7607  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7608  */
7609 int
7610 kstrputmsg(
7611 	struct vnode *vp,
7612 	mblk_t *mctl,
7613 	struct uio *uiop,
7614 	ssize_t msgsize,
7615 	unsigned char pri,
7616 	int flag,
7617 	int fmode)
7618 {
7619 	struct stdata *stp;
7620 	queue_t *wqp;
7621 	ssize_t rmin, rmax;
7622 	int error;
7623 
7624 	ASSERT(vp->v_stream);
7625 	stp = vp->v_stream;
7626 	wqp = stp->sd_wrq;
7627 #ifdef C2_AUDIT
7628 	if (audit_active)
7629 		audit_strputmsg(vp, NULL, NULL, pri, flag, fmode);
7630 #endif
7631 	if (mctl == NULL)
7632 		return (EINVAL);
7633 
7634 	mutex_enter(&stp->sd_lock);
7635 
7636 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
7637 		mutex_exit(&stp->sd_lock);
7638 		freemsg(mctl);
7639 		return (error);
7640 	}
7641 
7642 	if ((stp->sd_flag & STPLEX) || !(flag & MSG_IGNERROR)) {
7643 		if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7644 			error = strwriteable(stp, B_FALSE, B_TRUE);
7645 			if (error != 0) {
7646 				mutex_exit(&stp->sd_lock);
7647 				freemsg(mctl);
7648 				return (error);
7649 			}
7650 		}
7651 	}
7652 
7653 	mutex_exit(&stp->sd_lock);
7654 
7655 	/*
7656 	 * Check for legal flag value.
7657 	 */
7658 	switch (flag & (MSG_HIPRI|MSG_BAND|MSG_ANY)) {
7659 	case MSG_HIPRI:
7660 		if (pri != 0) {
7661 			freemsg(mctl);
7662 			return (EINVAL);
7663 		}
7664 		break;
7665 	case MSG_BAND:
7666 		break;
7667 	default:
7668 		freemsg(mctl);
7669 		return (EINVAL);
7670 	}
7671 
7672 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_IN,
7673 		"kstrputmsg in:stp %p", stp);
7674 
7675 	/* get these values from those cached in the stream head */
7676 	rmin = stp->sd_qn_minpsz;
7677 	rmax = stp->sd_qn_maxpsz;
7678 
7679 	/*
7680 	 * Make sure ctl and data sizes together fall within the
7681 	 * limits of the max and min receive packet sizes and do
7682 	 * not exceed system limit.
7683 	 */
7684 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
7685 	if (rmax == 0) {
7686 		freemsg(mctl);
7687 		return (ERANGE);
7688 	}
7689 	/*
7690 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
7691 	 * Needed to prevent partial failures in the strmakedata loop.
7692 	 */
7693 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
7694 		rmax = stp->sd_maxblk;
7695 
7696 	if (uiop == NULL) {
7697 		msgsize = -1;
7698 		rmin = -1;	/* no range check for NULL data part */
7699 	} else {
7700 		/* Use uio flags as well as the fmode parameter flags */
7701 		fmode |= uiop->uio_fmode;
7702 
7703 		if ((msgsize < rmin) ||
7704 		    ((msgsize > rmax) && (rmax != INFPSZ))) {
7705 			freemsg(mctl);
7706 			return (ERANGE);
7707 		}
7708 	}
7709 
7710 	/* Ignore flow control in strput for HIPRI */
7711 	if (flag & MSG_HIPRI)
7712 		flag |= MSG_IGNFLOW;
7713 
7714 	for (;;) {
7715 		int done = 0;
7716 		int waitflag;
7717 		mblk_t *mp;
7718 
7719 		/*
7720 		 * strput will always free the ctl mblk - even when strput
7721 		 * fails. If MSG_IGNFLOW is set then any error returned
7722 		 * will cause us to break the loop, so we don't need a copy
7723 		 * of the message. If MSG_IGNFLOW is not set, then we can
7724 		 * get hit by flow control and be forced to try again. In
7725 		 * this case we need to have a copy of the message. We
7726 		 * do this using copymsg since the message may get modified
7727 		 * by something below us.
7728 		 *
7729 		 * We've observed that many TPI providers do not check db_ref
7730 		 * on the control messages but blindly reuse them for the
7731 		 * T_OK_ACK/T_ERROR_ACK. Thus using copymsg is more
7732 		 * friendly to such providers than using dupmsg. Also, note
7733 		 * that sockfs uses MSG_IGNFLOW for all TPI control messages.
7734 		 * Only data messages are subject to flow control, hence
7735 		 * subject to this copymsg.
7736 		 */
7737 		if (flag & MSG_IGNFLOW) {
7738 			mp = mctl;
7739 			mctl = NULL;
7740 		} else {
7741 			do {
7742 				/*
7743 				 * If a message has a free pointer, the message
7744 				 * must be dupmsg to maintain this pointer.
7745 				 * Code using this facility must be sure
7746 				 * that modules below will not change the
7747 				 * contents of the dblk without checking db_ref
7748 				 * first. If db_ref is > 1, then the module
7749 				 * needs to do a copymsg first. Otherwise,
7750 				 * the contents of the dblk may become
7751 				 * inconsistent because the freesmg/freeb below
7752 				 * may end up calling atomic_add_32_nv.
7753 				 * The atomic_add_32_nv in freeb (accessing
7754 				 * all of db_ref, db_type, db_flags, and
7755 				 * db_struioflag) does not prevent other threads
7756 				 * from concurrently trying to modify e.g.
7757 				 * db_type.
7758 				 */
7759 				if (mctl->b_datap->db_frtnp != NULL)
7760 					mp = dupmsg(mctl);
7761 				else
7762 					mp = copymsg(mctl);
7763 
7764 				if (mp != NULL)
7765 					break;
7766 
7767 				error = strwaitbuf(msgdsize(mctl), BPRI_MED);
7768 				if (error) {
7769 					freemsg(mctl);
7770 					return (error);
7771 				}
7772 			} while (mp == NULL);
7773 		}
7774 		/*
7775 		 * Verify that all of msgsize can be transferred by
7776 		 * strput.
7777 		 */
7778 		ASSERT(stp->sd_maxblk == INFPSZ || stp->sd_maxblk >= msgsize);
7779 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
7780 		if (error == 0)
7781 			break;
7782 
7783 		if (error != EWOULDBLOCK)
7784 			goto out;
7785 
7786 		/*
7787 		 * IF MSG_IGNFLOW is set we should have broken out of loop
7788 		 * above.
7789 		 */
7790 		ASSERT(!(flag & MSG_IGNFLOW));
7791 		mutex_enter(&stp->sd_lock);
7792 		/*
7793 		 * Check for a missed wakeup.
7794 		 * Needed since strput did not hold sd_lock across
7795 		 * the canputnext.
7796 		 */
7797 		if (bcanputnext(wqp, pri)) {
7798 			/* Try again */
7799 			mutex_exit(&stp->sd_lock);
7800 			continue;
7801 		}
7802 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAIT,
7803 			"kstrputmsg wait:stp %p waits pri %d", stp, pri);
7804 
7805 		waitflag = WRITEWAIT;
7806 		if (flag & (MSG_HOLDSIG|MSG_IGNERROR)) {
7807 			if (flag & MSG_HOLDSIG)
7808 				waitflag |= STR_NOSIG;
7809 			if (flag & MSG_IGNERROR)
7810 				waitflag |= STR_NOERROR;
7811 		}
7812 		if (((error = strwaitq(stp, waitflag,
7813 		    (ssize_t)0, fmode, -1, &done)) != 0) || done) {
7814 			mutex_exit(&stp->sd_lock);
7815 			TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
7816 				"kstrputmsg out:stp %p out %d error %d",
7817 				stp, 0, error);
7818 			freemsg(mctl);
7819 			return (error);
7820 		}
7821 		TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAKE,
7822 			"kstrputmsg wake:stp %p wakes", stp);
7823 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
7824 			mutex_exit(&stp->sd_lock);
7825 			freemsg(mctl);
7826 			return (error);
7827 		}
7828 		mutex_exit(&stp->sd_lock);
7829 	}
7830 out:
7831 	freemsg(mctl);
7832 	/*
7833 	 * For historic reasons, applications expect EAGAIN
7834 	 * when data mblk could not be allocated. so change
7835 	 * ENOMEM back to EAGAIN
7836 	 */
7837 	if (error == ENOMEM)
7838 		error = EAGAIN;
7839 	TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
7840 		"kstrputmsg out:stp %p out %d error %d", stp, 2, error);
7841 	return (error);
7842 }
7843 
7844 /*
7845  * Determines whether the necessary conditions are set on a stream
7846  * for it to be readable, writeable, or have exceptions.
7847  *
7848  * strpoll handles the consolidation private events:
7849  *	POLLNOERR	Do not return POLLERR even if there are stream
7850  *			head errors.
7851  *			Used by sockfs.
7852  *	POLLRDDATA	Do not return POLLIN unless at least one message on
7853  *			the queue contains one or more M_DATA mblks. Thus
7854  *			when this flag is set a queue with only
7855  *			M_PROTO/M_PCPROTO mblks does not return POLLIN.
7856  *			Used by sockfs to ignore T_EXDATA_IND messages.
7857  *
7858  * Note: POLLRDDATA assumes that synch streams only return messages with
7859  * an M_DATA attached (i.e. not messages consisting of only
7860  * an M_PROTO/M_PCPROTO part).
7861  */
7862 int
7863 strpoll(
7864 	struct stdata *stp,
7865 	short events_arg,
7866 	int anyyet,
7867 	short *reventsp,
7868 	struct pollhead **phpp)
7869 {
7870 	int events = (ushort_t)events_arg;
7871 	int retevents = 0;
7872 	mblk_t *mp;
7873 	qband_t *qbp;
7874 	long sd_flags = stp->sd_flag;
7875 	int headlocked = 0;
7876 
7877 	/*
7878 	 * For performance, a single 'if' tests for most possible edge
7879 	 * conditions in one shot
7880 	 */
7881 	if (sd_flags & (STPLEX | STRDERR | STWRERR)) {
7882 		if (sd_flags & STPLEX) {
7883 			*reventsp = POLLNVAL;
7884 			return (EINVAL);
7885 		}
7886 		if (((events & (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) &&
7887 		    (sd_flags & STRDERR)) ||
7888 		    ((events & (POLLOUT | POLLWRNORM | POLLWRBAND)) &&
7889 		    (sd_flags & STWRERR))) {
7890 			if (!(events & POLLNOERR)) {
7891 				*reventsp = POLLERR;
7892 				return (0);
7893 			}
7894 		}
7895 	}
7896 	if (sd_flags & STRHUP) {
7897 		retevents |= POLLHUP;
7898 	} else if (events & (POLLWRNORM | POLLWRBAND)) {
7899 		queue_t *tq;
7900 		queue_t	*qp = stp->sd_wrq;
7901 
7902 		claimstr(qp);
7903 		/* Find next module forward that has a service procedure */
7904 		tq = qp->q_next->q_nfsrv;
7905 		ASSERT(tq != NULL);
7906 
7907 		polllock(&stp->sd_pollist, QLOCK(tq));
7908 		if (events & POLLWRNORM) {
7909 			queue_t *sqp;
7910 
7911 			if (tq->q_flag & QFULL)
7912 				/* ensure backq svc procedure runs */
7913 				tq->q_flag |= QWANTW;
7914 			else if ((sqp = stp->sd_struiowrq) != NULL) {
7915 				/* Check sync stream barrier write q */
7916 				mutex_exit(QLOCK(tq));
7917 				polllock(&stp->sd_pollist, QLOCK(sqp));
7918 				if (sqp->q_flag & QFULL)
7919 					/* ensure pollwakeup() is done */
7920 					sqp->q_flag |= QWANTWSYNC;
7921 				else
7922 					retevents |= POLLOUT;
7923 				/* More write events to process ??? */
7924 				if (! (events & POLLWRBAND)) {
7925 					mutex_exit(QLOCK(sqp));
7926 					releasestr(qp);
7927 					goto chkrd;
7928 				}
7929 				mutex_exit(QLOCK(sqp));
7930 				polllock(&stp->sd_pollist, QLOCK(tq));
7931 			} else
7932 				retevents |= POLLOUT;
7933 		}
7934 		if (events & POLLWRBAND) {
7935 			qbp = tq->q_bandp;
7936 			if (qbp) {
7937 				while (qbp) {
7938 					if (qbp->qb_flag & QB_FULL)
7939 						qbp->qb_flag |= QB_WANTW;
7940 					else
7941 						retevents |= POLLWRBAND;
7942 					qbp = qbp->qb_next;
7943 				}
7944 			} else {
7945 				retevents |= POLLWRBAND;
7946 			}
7947 		}
7948 		mutex_exit(QLOCK(tq));
7949 		releasestr(qp);
7950 	}
7951 chkrd:
7952 	if (sd_flags & STRPRI) {
7953 		retevents |= (events & POLLPRI);
7954 	} else if (events & (POLLRDNORM | POLLRDBAND | POLLIN)) {
7955 		queue_t	*qp = _RD(stp->sd_wrq);
7956 		int normevents = (events & (POLLIN | POLLRDNORM));
7957 
7958 		/*
7959 		 * Note: Need to do polllock() here since ps_lock may be
7960 		 * held. See bug 4191544.
7961 		 */
7962 		polllock(&stp->sd_pollist, &stp->sd_lock);
7963 		headlocked = 1;
7964 		mp = qp->q_first;
7965 		while (mp) {
7966 			/*
7967 			 * For POLLRDDATA we scan b_cont and b_next until we
7968 			 * find an M_DATA.
7969 			 */
7970 			if ((events & POLLRDDATA) &&
7971 			    mp->b_datap->db_type != M_DATA) {
7972 				mblk_t *nmp = mp->b_cont;
7973 
7974 				while (nmp != NULL &&
7975 				    nmp->b_datap->db_type != M_DATA)
7976 					nmp = nmp->b_cont;
7977 				if (nmp == NULL) {
7978 					mp = mp->b_next;
7979 					continue;
7980 				}
7981 			}
7982 			if (mp->b_band == 0)
7983 				retevents |= normevents;
7984 			else
7985 				retevents |= (events & (POLLIN | POLLRDBAND));
7986 			break;
7987 		}
7988 		if (! (retevents & normevents) &&
7989 		    (stp->sd_wakeq & RSLEEP)) {
7990 			/*
7991 			 * Sync stream barrier read queue has data.
7992 			 */
7993 			retevents |= normevents;
7994 		}
7995 		/* Treat eof as normal data */
7996 		if (sd_flags & STREOF)
7997 			retevents |= normevents;
7998 	}
7999 
8000 	*reventsp = (short)retevents;
8001 	if (retevents) {
8002 		if (headlocked)
8003 			mutex_exit(&stp->sd_lock);
8004 		return (0);
8005 	}
8006 
8007 	/*
8008 	 * If poll() has not found any events yet, set up event cell
8009 	 * to wake up the poll if a requested event occurs on this
8010 	 * stream.  Check for collisions with outstanding poll requests.
8011 	 */
8012 	if (!anyyet) {
8013 		*phpp = &stp->sd_pollist;
8014 		if (headlocked == 0) {
8015 			polllock(&stp->sd_pollist, &stp->sd_lock);
8016 			headlocked = 1;
8017 		}
8018 		stp->sd_rput_opt |= SR_POLLIN;
8019 	}
8020 	if (headlocked)
8021 		mutex_exit(&stp->sd_lock);
8022 	return (0);
8023 }
8024 
8025 /*
8026  * The purpose of putback() is to assure sleeping polls/reads
8027  * are awakened when there are no new messages arriving at the,
8028  * stream head, and a message is placed back on the read queue.
8029  *
8030  * sd_lock must be held when messages are placed back on stream
8031  * head.  (getq() holds sd_lock when it removes messages from
8032  * the queue)
8033  */
8034 
8035 static void
8036 putback(struct stdata *stp, queue_t *q, mblk_t *bp, int band)
8037 {
8038 	ASSERT(MUTEX_HELD(&stp->sd_lock));
8039 	(void) putbq(q, bp);
8040 	/*
8041 	 * A message may have come in when the sd_lock was dropped in the
8042 	 * calling routine. If this is the case and STR*ATMARK info was
8043 	 * received, need to move that from the stream head to the q_last
8044 	 * so that SIOCATMARK can return the proper value.
8045 	 */
8046 	if (stp->sd_flag & (STRATMARK | STRNOTATMARK)) {
8047 		unsigned short *flagp = &q->q_last->b_flag;
8048 		uint_t b_flag = (uint_t)*flagp;
8049 
8050 		if (stp->sd_flag & STRATMARK) {
8051 			b_flag &= ~MSGNOTMARKNEXT;
8052 			b_flag |= MSGMARKNEXT;
8053 			stp->sd_flag &= ~STRATMARK;
8054 		} else {
8055 			b_flag &= ~MSGMARKNEXT;
8056 			b_flag |= MSGNOTMARKNEXT;
8057 			stp->sd_flag &= ~STRNOTATMARK;
8058 		}
8059 		*flagp = (unsigned short) b_flag;
8060 	}
8061 
8062 #ifdef	DEBUG
8063 	/*
8064 	 * Make sure that the flags are not messed up.
8065 	 */
8066 	{
8067 		mblk_t *mp;
8068 		mp = q->q_last;
8069 		while (mp != NULL) {
8070 			ASSERT((mp->b_flag & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
8071 			    (MSGMARKNEXT|MSGNOTMARKNEXT));
8072 			mp = mp->b_cont;
8073 		}
8074 	}
8075 #endif
8076 	if (q->q_first == bp) {
8077 		short pollevents;
8078 
8079 		if (stp->sd_flag & RSLEEP) {
8080 			stp->sd_flag &= ~RSLEEP;
8081 			cv_broadcast(&q->q_wait);
8082 		}
8083 		if (stp->sd_flag & STRPRI) {
8084 			pollevents = POLLPRI;
8085 		} else {
8086 			if (band == 0) {
8087 				if (!(stp->sd_rput_opt & SR_POLLIN))
8088 					return;
8089 				stp->sd_rput_opt &= ~SR_POLLIN;
8090 				pollevents = POLLIN | POLLRDNORM;
8091 			} else {
8092 				pollevents = POLLIN | POLLRDBAND;
8093 			}
8094 		}
8095 		mutex_exit(&stp->sd_lock);
8096 		pollwakeup(&stp->sd_pollist, pollevents);
8097 		mutex_enter(&stp->sd_lock);
8098 	}
8099 }
8100 
8101 /*
8102  * Return the held vnode attached to the stream head of a
8103  * given queue
8104  * It is the responsibility of the calling routine to ensure
8105  * that the queue does not go away (e.g. pop).
8106  */
8107 vnode_t *
8108 strq2vp(queue_t *qp)
8109 {
8110 	vnode_t *vp;
8111 	vp = STREAM(qp)->sd_vnode;
8112 	ASSERT(vp != NULL);
8113 	VN_HOLD(vp);
8114 	return (vp);
8115 }
8116 
8117 /*
8118  * return the stream head write queue for the given vp
8119  * It is the responsibility of the calling routine to ensure
8120  * that the stream or vnode do not close.
8121  */
8122 queue_t *
8123 strvp2wq(vnode_t *vp)
8124 {
8125 	ASSERT(vp->v_stream != NULL);
8126 	return (vp->v_stream->sd_wrq);
8127 }
8128 
8129 /*
8130  * pollwakeup stream head
8131  * It is the responsibility of the calling routine to ensure
8132  * that the stream or vnode do not close.
8133  */
8134 void
8135 strpollwakeup(vnode_t *vp, short event)
8136 {
8137 	ASSERT(vp->v_stream);
8138 	pollwakeup(&vp->v_stream->sd_pollist, event);
8139 }
8140 
8141 /*
8142  * Mate the stream heads of two vnodes together. If the two vnodes are the
8143  * same, we just make the write-side point at the read-side -- otherwise,
8144  * we do a full mate.  Only works on vnodes associated with streams that are
8145  * still being built and thus have only a stream head.
8146  */
8147 void
8148 strmate(vnode_t *vp1, vnode_t *vp2)
8149 {
8150 	queue_t *wrq1 = strvp2wq(vp1);
8151 	queue_t *wrq2 = strvp2wq(vp2);
8152 
8153 	/*
8154 	 * Verify that there are no modules on the stream yet.  We also
8155 	 * rely on the stream head always having a service procedure to
8156 	 * avoid tweaking q_nfsrv.
8157 	 */
8158 	ASSERT(wrq1->q_next == NULL && wrq2->q_next == NULL);
8159 	ASSERT(wrq1->q_qinfo->qi_srvp != NULL);
8160 	ASSERT(wrq2->q_qinfo->qi_srvp != NULL);
8161 
8162 	/*
8163 	 * If the queues are the same, just twist; otherwise do a full mate.
8164 	 */
8165 	if (wrq1 == wrq2) {
8166 		wrq1->q_next = _RD(wrq1);
8167 	} else {
8168 		wrq1->q_next = _RD(wrq2);
8169 		wrq2->q_next = _RD(wrq1);
8170 		STREAM(wrq1)->sd_mate = STREAM(wrq2);
8171 		STREAM(wrq1)->sd_flag |= STRMATE;
8172 		STREAM(wrq2)->sd_mate = STREAM(wrq1);
8173 		STREAM(wrq2)->sd_flag |= STRMATE;
8174 	}
8175 }
8176 
8177 /*
8178  * XXX will go away when console is correctly fixed.
8179  * Clean up the console PIDS, from previous I_SETSIG,
8180  * called only for cnopen which never calls strclean().
8181  */
8182 void
8183 str_cn_clean(struct vnode *vp)
8184 {
8185 	strsig_t *ssp, *pssp, *tssp;
8186 	struct stdata *stp;
8187 	struct pid  *pidp;
8188 	int update = 0;
8189 
8190 	ASSERT(vp->v_stream);
8191 	stp = vp->v_stream;
8192 	pssp = NULL;
8193 	mutex_enter(&stp->sd_lock);
8194 	ssp = stp->sd_siglist;
8195 	while (ssp) {
8196 		mutex_enter(&pidlock);
8197 		pidp = ssp->ss_pidp;
8198 		/*
8199 		 * Get rid of PID if the proc is gone.
8200 		 */
8201 		if (pidp->pid_prinactive) {
8202 			tssp = ssp->ss_next;
8203 			if (pssp)
8204 				pssp->ss_next = tssp;
8205 			else
8206 				stp->sd_siglist = tssp;
8207 			ASSERT(pidp->pid_ref <= 1);
8208 			PID_RELE(ssp->ss_pidp);
8209 			mutex_exit(&pidlock);
8210 			kmem_free(ssp, sizeof (strsig_t));
8211 			update = 1;
8212 			ssp = tssp;
8213 			continue;
8214 		} else
8215 			mutex_exit(&pidlock);
8216 		pssp = ssp;
8217 		ssp = ssp->ss_next;
8218 	}
8219 	if (update) {
8220 		stp->sd_sigflags = 0;
8221 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
8222 			stp->sd_sigflags |= ssp->ss_events;
8223 	}
8224 	mutex_exit(&stp->sd_lock);
8225 }
8226 
8227 /*
8228  * Return B_TRUE if there is data in the message, B_FALSE otherwise.
8229  */
8230 static boolean_t
8231 msghasdata(mblk_t *bp)
8232 {
8233 	for (; bp; bp = bp->b_cont)
8234 		if (bp->b_datap->db_type == M_DATA) {
8235 			ASSERT(bp->b_wptr >= bp->b_rptr);
8236 			if (bp->b_wptr > bp->b_rptr)
8237 				return (B_TRUE);
8238 		}
8239 	return (B_FALSE);
8240 }
8241