xref: /illumos-gate/usr/src/uts/common/os/streamio.c (revision 86c48bbfeb72d5a6ee171e713059939bab658b77)
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 2009 Sun Microsystems, Inc.  All rights reserved.
27  * Use is subject to license terms.
28  */
29 
30 #include <sys/types.h>
31 #include <sys/sysmacros.h>
32 #include <sys/param.h>
33 #include <sys/errno.h>
34 #include <sys/signal.h>
35 #include <sys/stat.h>
36 #include <sys/proc.h>
37 #include <sys/cred.h>
38 #include <sys/user.h>
39 #include <sys/vnode.h>
40 #include <sys/file.h>
41 #include <sys/stream.h>
42 #include <sys/strsubr.h>
43 #include <sys/stropts.h>
44 #include <sys/tihdr.h>
45 #include <sys/var.h>
46 #include <sys/poll.h>
47 #include <sys/termio.h>
48 #include <sys/ttold.h>
49 #include <sys/systm.h>
50 #include <sys/uio.h>
51 #include <sys/cmn_err.h>
52 #include <sys/sad.h>
53 #include <sys/netstack.h>
54 #include <sys/priocntl.h>
55 #include <sys/jioctl.h>
56 #include <sys/procset.h>
57 #include <sys/session.h>
58 #include <sys/kmem.h>
59 #include <sys/filio.h>
60 #include <sys/vtrace.h>
61 #include <sys/debug.h>
62 #include <sys/strredir.h>
63 #include <sys/fs/fifonode.h>
64 #include <sys/fs/snode.h>
65 #include <sys/strlog.h>
66 #include <sys/strsun.h>
67 #include <sys/project.h>
68 #include <sys/kbio.h>
69 #include <sys/msio.h>
70 #include <sys/tty.h>
71 #include <sys/ptyvar.h>
72 #include <sys/vuid_event.h>
73 #include <sys/modctl.h>
74 #include <sys/sunddi.h>
75 #include <sys/sunldi_impl.h>
76 #include <sys/autoconf.h>
77 #include <sys/policy.h>
78 #include <sys/dld.h>
79 #include <sys/zone.h>
80 #include <sys/sodirect.h>
81 
82 /*
83  * This define helps improve the readability of streams code while
84  * still maintaining a very old streams performance enhancement.  The
85  * performance enhancement basically involved having all callers
86  * of straccess() perform the first check that straccess() will do
87  * locally before actually calling straccess().  (There by reducing
88  * the number of unnecessary calls to straccess().)
89  */
90 #define	i_straccess(x, y)	((stp->sd_sidp == NULL) ? 0 : \
91 				    (stp->sd_vnode->v_type == VFIFO) ? 0 : \
92 				    straccess((x), (y)))
93 
94 /*
95  * what is mblk_pull_len?
96  *
97  * If a streams message consists of many short messages,
98  * a performance degradation occurs from copyout overhead.
99  * To decrease the per mblk overhead, messages that are
100  * likely to consist of many small mblks are pulled up into
101  * one continuous chunk of memory.
102  *
103  * To avoid the processing overhead of examining every
104  * mblk, a quick heuristic is used. If the first mblk in
105  * the message is shorter than mblk_pull_len, it is likely
106  * that the rest of the mblk will be short.
107  *
108  * This heuristic was decided upon after performance tests
109  * indicated that anything more complex slowed down the main
110  * code path.
111  */
112 #define	MBLK_PULL_LEN 64
113 uint32_t mblk_pull_len = MBLK_PULL_LEN;
114 
115 /*
116  * The sgttyb_handling flag controls the handling of the old BSD
117  * TIOCGETP, TIOCSETP, and TIOCSETN ioctls as follows:
118  *
119  * 0 - Emit no warnings at all and retain old, broken behavior.
120  * 1 - Emit no warnings and silently handle new semantics.
121  * 2 - Send cmn_err(CE_NOTE) when either TIOCSETP or TIOCSETN is used
122  *     (once per system invocation).  Handle with new semantics.
123  * 3 - Send SIGSYS when any TIOCGETP, TIOCSETP, or TIOCSETN call is
124  *     made (so that offenders drop core and are easy to debug).
125  *
126  * The "new semantics" are that TIOCGETP returns B38400 for
127  * sg_[io]speed if the corresponding value is over B38400, and that
128  * TIOCSET[PN] accept B38400 in these cases to mean "retain current
129  * bit rate."
130  */
131 int sgttyb_handling = 1;
132 static boolean_t sgttyb_complaint;
133 
134 /* don't push drcompat module by default on Style-2 streams */
135 static int push_drcompat = 0;
136 
137 /*
138  * id value used to distinguish between different ioctl messages
139  */
140 static uint32_t ioc_id;
141 
142 static void putback(struct stdata *, queue_t *, mblk_t *, int);
143 static void strcleanall(struct vnode *);
144 static int strwsrv(queue_t *);
145 static int strdocmd(struct stdata *, struct strcmd *, cred_t *);
146 static void struioainit(queue_t *, sodirect_t *, uio_t *);
147 
148 /*
149  * qinit and module_info structures for stream head read and write queues
150  */
151 struct module_info strm_info = { 0, "strrhead", 0, INFPSZ, STRHIGH, STRLOW };
152 struct module_info stwm_info = { 0, "strwhead", 0, 0, 0, 0 };
153 struct qinit strdata = { strrput, NULL, NULL, NULL, NULL, &strm_info };
154 struct qinit stwdata = { NULL, strwsrv, NULL, NULL, NULL, &stwm_info };
155 struct module_info fiform_info = { 0, "fifostrrhead", 0, PIPE_BUF, FIFOHIWAT,
156     FIFOLOWAT };
157 struct module_info fifowm_info = { 0, "fifostrwhead", 0, 0, 0, 0 };
158 struct qinit fifo_strdata = { strrput, NULL, NULL, NULL, NULL, &fiform_info };
159 struct qinit fifo_stwdata = { NULL, strwsrv, NULL, NULL, NULL, &fifowm_info };
160 
161 extern kmutex_t	strresources;	/* protects global resources */
162 extern kmutex_t muxifier;	/* single-threads multiplexor creation */
163 
164 static boolean_t msghasdata(mblk_t *bp);
165 #define	msgnodata(bp) (!msghasdata(bp))
166 
167 /*
168  * Stream head locking notes:
169  *	There are four monitors associated with the stream head:
170  *	1. v_stream monitor: in stropen() and strclose() v_lock
171  *		is held while the association of vnode and stream
172  *		head is established or tested for.
173  *	2. open/close/push/pop monitor: sd_lock is held while each
174  *		thread bids for exclusive access to this monitor
175  *		for opening or closing a stream.  In addition, this
176  *		monitor is entered during pushes and pops.  This
177  *		guarantees that during plumbing operations there
178  *		is only one thread trying to change the plumbing.
179  *		Any other threads present in the stream are only
180  *		using the plumbing.
181  *	3. read/write monitor: in the case of read, a thread holds
182  *		sd_lock while trying to get data from the stream
183  *		head queue.  if there is none to fulfill a read
184  *		request, it sets RSLEEP and calls cv_wait_sig() down
185  *		in strwaitq() to await the arrival of new data.
186  *		when new data arrives in strrput(), sd_lock is acquired
187  *		before testing for RSLEEP and calling cv_broadcast().
188  *		the behavior of strwrite(), strwsrv(), and WSLEEP
189  *		mirror this.
190  *	4. ioctl monitor: sd_lock is gotten to ensure that only one
191  *		thread is doing an ioctl at a time.
192  *
193  * Note, for sodirect case 3. is extended to (*sodirect_t.sod_enqueue)()
194  * call-back from below, further the sodirect support is for code paths
195  * called via kstgetmsg(), all other code paths ASSERT() that sodirect
196  * uioa generated mblk_t's (i.e. DBLK_UIOA) aren't processed.
197  */
198 
199 static int
200 push_mod(queue_t *qp, dev_t *devp, struct stdata *stp, const char *name,
201     int anchor, cred_t *crp, uint_t anchor_zoneid)
202 {
203 	int error;
204 	fmodsw_impl_t *fp;
205 
206 	if (stp->sd_flag & (STRHUP|STRDERR|STWRERR)) {
207 		error = (stp->sd_flag & STRHUP) ? ENXIO : EIO;
208 		return (error);
209 	}
210 	if (stp->sd_pushcnt >= nstrpush) {
211 		return (EINVAL);
212 	}
213 
214 	if ((fp = fmodsw_find(name, FMODSW_HOLD | FMODSW_LOAD)) == NULL) {
215 		stp->sd_flag |= STREOPENFAIL;
216 		return (EINVAL);
217 	}
218 
219 	/*
220 	 * push new module and call its open routine via qattach
221 	 */
222 	if ((error = qattach(qp, devp, 0, crp, fp, B_FALSE)) != 0)
223 		return (error);
224 
225 	/*
226 	 * Check to see if caller wants a STREAMS anchor
227 	 * put at this place in the stream, and add if so.
228 	 */
229 	mutex_enter(&stp->sd_lock);
230 	if (anchor == stp->sd_pushcnt) {
231 		stp->sd_anchor = stp->sd_pushcnt;
232 		stp->sd_anchorzone = anchor_zoneid;
233 	}
234 	mutex_exit(&stp->sd_lock);
235 
236 	return (0);
237 }
238 
239 /*
240  * Open a stream device.
241  */
242 int
243 stropen(vnode_t *vp, dev_t *devp, int flag, cred_t *crp)
244 {
245 	struct stdata *stp;
246 	queue_t *qp;
247 	int s;
248 	dev_t dummydev, savedev;
249 	struct autopush *ap;
250 	struct dlautopush dlap;
251 	int error = 0;
252 	ssize_t	rmin, rmax;
253 	int cloneopen;
254 	queue_t *brq;
255 	major_t major;
256 	str_stack_t *ss;
257 	zoneid_t zoneid;
258 	uint_t anchor;
259 
260 	if (audit_active)
261 		audit_stropen(vp, devp, flag, crp);
262 
263 	/*
264 	 * If the stream already exists, wait for any open in progress
265 	 * to complete, then call the open function of each module and
266 	 * driver in the stream.  Otherwise create the stream.
267 	 */
268 	TRACE_1(TR_FAC_STREAMS_FR, TR_STROPEN, "stropen:%p", vp);
269 retry:
270 	mutex_enter(&vp->v_lock);
271 	if ((stp = vp->v_stream) != NULL) {
272 
273 		/*
274 		 * Waiting for stream to be created to device
275 		 * due to another open.
276 		 */
277 		mutex_exit(&vp->v_lock);
278 
279 		if (STRMATED(stp)) {
280 			struct stdata *strmatep = stp->sd_mate;
281 
282 			STRLOCKMATES(stp);
283 			if (strmatep->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
284 				if (flag & (FNDELAY|FNONBLOCK)) {
285 					error = EAGAIN;
286 					mutex_exit(&strmatep->sd_lock);
287 					goto ckreturn;
288 				}
289 				mutex_exit(&stp->sd_lock);
290 				if (!cv_wait_sig(&strmatep->sd_monitor,
291 				    &strmatep->sd_lock)) {
292 					error = EINTR;
293 					mutex_exit(&strmatep->sd_lock);
294 					mutex_enter(&stp->sd_lock);
295 					goto ckreturn;
296 				}
297 				mutex_exit(&strmatep->sd_lock);
298 				goto retry;
299 			}
300 			if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
301 				if (flag & (FNDELAY|FNONBLOCK)) {
302 					error = EAGAIN;
303 					mutex_exit(&strmatep->sd_lock);
304 					goto ckreturn;
305 				}
306 				mutex_exit(&strmatep->sd_lock);
307 				if (!cv_wait_sig(&stp->sd_monitor,
308 				    &stp->sd_lock)) {
309 					error = EINTR;
310 					goto ckreturn;
311 				}
312 				mutex_exit(&stp->sd_lock);
313 				goto retry;
314 			}
315 
316 			if (stp->sd_flag & (STRDERR|STWRERR)) {
317 				error = EIO;
318 				mutex_exit(&strmatep->sd_lock);
319 				goto ckreturn;
320 			}
321 
322 			stp->sd_flag |= STWOPEN;
323 			STRUNLOCKMATES(stp);
324 		} else {
325 			mutex_enter(&stp->sd_lock);
326 			if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
327 				if (flag & (FNDELAY|FNONBLOCK)) {
328 					error = EAGAIN;
329 					goto ckreturn;
330 				}
331 				if (!cv_wait_sig(&stp->sd_monitor,
332 				    &stp->sd_lock)) {
333 					error = EINTR;
334 					goto ckreturn;
335 				}
336 				mutex_exit(&stp->sd_lock);
337 				goto retry;  /* could be clone! */
338 			}
339 
340 			if (stp->sd_flag & (STRDERR|STWRERR)) {
341 				error = EIO;
342 				goto ckreturn;
343 			}
344 
345 			stp->sd_flag |= STWOPEN;
346 			mutex_exit(&stp->sd_lock);
347 		}
348 
349 		/*
350 		 * Open all modules and devices down stream to notify
351 		 * that another user is streaming.  For modules, set the
352 		 * last argument to MODOPEN and do not pass any open flags.
353 		 * Ignore dummydev since this is not the first open.
354 		 */
355 		claimstr(stp->sd_wrq);
356 		qp = stp->sd_wrq;
357 		while (_SAMESTR(qp)) {
358 			qp = qp->q_next;
359 			if ((error = qreopen(_RD(qp), devp, flag, crp)) != 0)
360 				break;
361 		}
362 		releasestr(stp->sd_wrq);
363 		mutex_enter(&stp->sd_lock);
364 		stp->sd_flag &= ~(STRHUP|STWOPEN|STRDERR|STWRERR);
365 		stp->sd_rerror = 0;
366 		stp->sd_werror = 0;
367 ckreturn:
368 		cv_broadcast(&stp->sd_monitor);
369 		mutex_exit(&stp->sd_lock);
370 		return (error);
371 	}
372 
373 	/*
374 	 * This vnode isn't streaming.  SPECFS already
375 	 * checked for multiple vnodes pointing to the
376 	 * same stream, so create a stream to the driver.
377 	 */
378 	qp = allocq();
379 	stp = shalloc(qp);
380 
381 	/*
382 	 * Initialize stream head.  shalloc() has given us
383 	 * exclusive access, and we have the vnode locked;
384 	 * we can do whatever we want with stp.
385 	 */
386 	stp->sd_flag = STWOPEN;
387 	stp->sd_siglist = NULL;
388 	stp->sd_pollist.ph_list = NULL;
389 	stp->sd_sigflags = 0;
390 	stp->sd_mark = NULL;
391 	stp->sd_closetime = STRTIMOUT;
392 	stp->sd_sidp = NULL;
393 	stp->sd_pgidp = NULL;
394 	stp->sd_vnode = vp;
395 	stp->sd_rerror = 0;
396 	stp->sd_werror = 0;
397 	stp->sd_wroff = 0;
398 	stp->sd_tail = 0;
399 	stp->sd_iocblk = NULL;
400 	stp->sd_cmdblk = NULL;
401 	stp->sd_pushcnt = 0;
402 	stp->sd_qn_minpsz = 0;
403 	stp->sd_qn_maxpsz = INFPSZ - 1;	/* used to check for initialization */
404 	stp->sd_maxblk = INFPSZ;
405 	stp->sd_sodirect = NULL;
406 	qp->q_ptr = _WR(qp)->q_ptr = stp;
407 	STREAM(qp) = STREAM(_WR(qp)) = stp;
408 	vp->v_stream = stp;
409 	mutex_exit(&vp->v_lock);
410 	if (vp->v_type == VFIFO) {
411 		stp->sd_flag |= OLDNDELAY;
412 		/*
413 		 * This means, both for pipes and fifos
414 		 * strwrite will send SIGPIPE if the other
415 		 * end is closed. For putmsg it depends
416 		 * on whether it is a XPG4_2 application
417 		 * or not
418 		 */
419 		stp->sd_wput_opt = SW_SIGPIPE;
420 
421 		/* setq might sleep in kmem_alloc - avoid holding locks. */
422 		setq(qp, &fifo_strdata, &fifo_stwdata, NULL, QMTSAFE,
423 		    SQ_CI|SQ_CO, B_FALSE);
424 
425 		set_qend(qp);
426 		stp->sd_strtab = fifo_getinfo();
427 		_WR(qp)->q_nfsrv = _WR(qp);
428 		qp->q_nfsrv = qp;
429 		/*
430 		 * Wake up others that are waiting for stream to be created.
431 		 */
432 		mutex_enter(&stp->sd_lock);
433 		/*
434 		 * nothing is be pushed on stream yet, so
435 		 * optimized stream head packetsizes are just that
436 		 * of the read queue
437 		 */
438 		stp->sd_qn_minpsz = qp->q_minpsz;
439 		stp->sd_qn_maxpsz = qp->q_maxpsz;
440 		stp->sd_flag &= ~STWOPEN;
441 		goto fifo_opendone;
442 	}
443 	/* setq might sleep in kmem_alloc - avoid holding locks. */
444 	setq(qp, &strdata, &stwdata, NULL, QMTSAFE, SQ_CI|SQ_CO, B_FALSE);
445 
446 	set_qend(qp);
447 
448 	/*
449 	 * Open driver and create stream to it (via qattach).
450 	 */
451 	savedev = *devp;
452 	cloneopen = (getmajor(*devp) == clone_major);
453 	if ((error = qattach(qp, devp, flag, crp, NULL, B_FALSE)) != 0) {
454 		mutex_enter(&vp->v_lock);
455 		vp->v_stream = NULL;
456 		mutex_exit(&vp->v_lock);
457 		mutex_enter(&stp->sd_lock);
458 		cv_broadcast(&stp->sd_monitor);
459 		mutex_exit(&stp->sd_lock);
460 		freeq(_RD(qp));
461 		shfree(stp);
462 		return (error);
463 	}
464 	/*
465 	 * Set sd_strtab after open in order to handle clonable drivers
466 	 */
467 	stp->sd_strtab = STREAMSTAB(getmajor(*devp));
468 
469 	/*
470 	 * Historical note: dummydev used to be be prior to the initial
471 	 * open (via qattach above), which made the value seen
472 	 * inconsistent between an I_PUSH and an autopush of a module.
473 	 */
474 	dummydev = *devp;
475 
476 	/*
477 	 * For clone open of old style (Q not associated) network driver,
478 	 * push DRMODNAME module to handle DL_ATTACH/DL_DETACH
479 	 */
480 	brq = _RD(_WR(qp)->q_next);
481 	major = getmajor(*devp);
482 	if (push_drcompat && cloneopen && NETWORK_DRV(major) &&
483 	    ((brq->q_flag & _QASSOCIATED) == 0)) {
484 		if (push_mod(qp, &dummydev, stp, DRMODNAME, 0, crp, 0) != 0)
485 			cmn_err(CE_WARN, "cannot push " DRMODNAME
486 			    " streams module");
487 	}
488 
489 	if (!NETWORK_DRV(major)) {
490 		savedev = *devp;
491 	} else {
492 		/*
493 		 * For network devices, process differently based on the
494 		 * return value from dld_autopush():
495 		 *
496 		 *   0: the passed-in device points to a GLDv3 datalink with
497 		 *   per-link autopush configuration; use that configuration
498 		 *   and ignore any per-driver autopush configuration.
499 		 *
500 		 *   1: the passed-in device points to a physical GLDv3
501 		 *   datalink without per-link autopush configuration.  The
502 		 *   passed in device was changed to refer to the actual
503 		 *   physical device (if it's not already); we use that new
504 		 *   device to look up any per-driver autopush configuration.
505 		 *
506 		 *   -1: neither of the above cases applied; use the initial
507 		 *   device to look up any per-driver autopush configuration.
508 		 */
509 		switch (dld_autopush(&savedev, &dlap)) {
510 		case 0:
511 			zoneid = crgetzoneid(crp);
512 			for (s = 0; s < dlap.dap_npush; s++) {
513 				error = push_mod(qp, &dummydev, stp,
514 				    dlap.dap_aplist[s], dlap.dap_anchor, crp,
515 				    zoneid);
516 				if (error != 0)
517 					break;
518 			}
519 			goto opendone;
520 		case 1:
521 			break;
522 		case -1:
523 			savedev = *devp;
524 			break;
525 		}
526 	}
527 	/*
528 	 * Find the autopush configuration based on "savedev". Start with the
529 	 * global zone. If not found check in the local zone.
530 	 */
531 	zoneid = GLOBAL_ZONEID;
532 retryap:
533 	ss = netstack_find_by_stackid(zoneid_to_netstackid(zoneid))->
534 	    netstack_str;
535 	if ((ap = sad_ap_find_by_dev(savedev, ss)) == NULL) {
536 		netstack_rele(ss->ss_netstack);
537 		if (zoneid == GLOBAL_ZONEID) {
538 			/*
539 			 * None found. Also look in the zone's autopush table.
540 			 */
541 			zoneid = crgetzoneid(crp);
542 			if (zoneid != GLOBAL_ZONEID)
543 				goto retryap;
544 		}
545 		goto opendone;
546 	}
547 	anchor = ap->ap_anchor;
548 	zoneid = crgetzoneid(crp);
549 	for (s = 0; s < ap->ap_npush; s++) {
550 		error = push_mod(qp, &dummydev, stp, ap->ap_list[s],
551 		    anchor, crp, zoneid);
552 		if (error != 0)
553 			break;
554 	}
555 	sad_ap_rele(ap, ss);
556 	netstack_rele(ss->ss_netstack);
557 
558 opendone:
559 
560 	/*
561 	 * let specfs know that open failed part way through
562 	 */
563 	if (error) {
564 		mutex_enter(&stp->sd_lock);
565 		stp->sd_flag |= STREOPENFAIL;
566 		mutex_exit(&stp->sd_lock);
567 	}
568 
569 	/*
570 	 * Wake up others that are waiting for stream to be created.
571 	 */
572 	mutex_enter(&stp->sd_lock);
573 	stp->sd_flag &= ~STWOPEN;
574 
575 	/*
576 	 * As a performance concern we are caching the values of
577 	 * q_minpsz and q_maxpsz of the module below the stream
578 	 * head in the stream head.
579 	 */
580 	mutex_enter(QLOCK(stp->sd_wrq->q_next));
581 	rmin = stp->sd_wrq->q_next->q_minpsz;
582 	rmax = stp->sd_wrq->q_next->q_maxpsz;
583 	mutex_exit(QLOCK(stp->sd_wrq->q_next));
584 
585 	/* do this processing here as a performance concern */
586 	if (strmsgsz != 0) {
587 		if (rmax == INFPSZ)
588 			rmax = strmsgsz;
589 		else
590 			rmax = MIN(strmsgsz, rmax);
591 	}
592 
593 	mutex_enter(QLOCK(stp->sd_wrq));
594 	stp->sd_qn_minpsz = rmin;
595 	stp->sd_qn_maxpsz = rmax;
596 	mutex_exit(QLOCK(stp->sd_wrq));
597 
598 fifo_opendone:
599 	cv_broadcast(&stp->sd_monitor);
600 	mutex_exit(&stp->sd_lock);
601 	return (error);
602 }
603 
604 static int strsink(queue_t *, mblk_t *);
605 static struct qinit deadrend = {
606 	strsink, NULL, NULL, NULL, NULL, &strm_info, NULL
607 };
608 static struct qinit deadwend = {
609 	NULL, NULL, NULL, NULL, NULL, &stwm_info, NULL
610 };
611 
612 /*
613  * Close a stream.
614  * This is called from closef() on the last close of an open stream.
615  * Strclean() will already have removed the siglist and pollist
616  * information, so all that remains is to remove all multiplexor links
617  * for the stream, pop all the modules (and the driver), and free the
618  * stream structure.
619  */
620 
621 int
622 strclose(struct vnode *vp, int flag, cred_t *crp)
623 {
624 	struct stdata *stp;
625 	queue_t *qp;
626 	int rval;
627 	int freestp = 1;
628 	queue_t *rmq;
629 
630 	if (audit_active)
631 		audit_strclose(vp, flag, crp);
632 
633 	TRACE_1(TR_FAC_STREAMS_FR,
634 	    TR_STRCLOSE, "strclose:%p", vp);
635 	ASSERT(vp->v_stream);
636 
637 	stp = vp->v_stream;
638 	ASSERT(!(stp->sd_flag & STPLEX));
639 	qp = stp->sd_wrq;
640 
641 	/*
642 	 * Needed so that strpoll will return non-zero for this fd.
643 	 * Note that with POLLNOERR STRHUP does still cause POLLHUP.
644 	 */
645 	mutex_enter(&stp->sd_lock);
646 	stp->sd_flag |= STRHUP;
647 	mutex_exit(&stp->sd_lock);
648 
649 	/*
650 	 * If the registered process or process group did not have an
651 	 * open instance of this stream then strclean would not be
652 	 * called. Thus at the time of closing all remaining siglist entries
653 	 * are removed.
654 	 */
655 	if (stp->sd_siglist != NULL)
656 		strcleanall(vp);
657 
658 	ASSERT(stp->sd_siglist == NULL);
659 	ASSERT(stp->sd_sigflags == 0);
660 
661 	if (STRMATED(stp)) {
662 		struct stdata *strmatep = stp->sd_mate;
663 		int waited = 1;
664 
665 		STRLOCKMATES(stp);
666 		while (waited) {
667 			waited = 0;
668 			while (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
669 				mutex_exit(&strmatep->sd_lock);
670 				cv_wait(&stp->sd_monitor, &stp->sd_lock);
671 				mutex_exit(&stp->sd_lock);
672 				STRLOCKMATES(stp);
673 				waited = 1;
674 			}
675 			while (strmatep->sd_flag &
676 			    (STWOPEN|STRCLOSE|STRPLUMB)) {
677 				mutex_exit(&stp->sd_lock);
678 				cv_wait(&strmatep->sd_monitor,
679 				    &strmatep->sd_lock);
680 				mutex_exit(&strmatep->sd_lock);
681 				STRLOCKMATES(stp);
682 				waited = 1;
683 			}
684 		}
685 		stp->sd_flag |= STRCLOSE;
686 		STRUNLOCKMATES(stp);
687 	} else {
688 		mutex_enter(&stp->sd_lock);
689 		stp->sd_flag |= STRCLOSE;
690 		mutex_exit(&stp->sd_lock);
691 	}
692 
693 	ASSERT(qp->q_first == NULL);	/* No more delayed write */
694 
695 	/* Check if an I_LINK was ever done on this stream */
696 	if (stp->sd_flag & STRHASLINKS) {
697 		netstack_t *ns;
698 		str_stack_t *ss;
699 
700 		ns = netstack_find_by_cred(crp);
701 		ASSERT(ns != NULL);
702 		ss = ns->netstack_str;
703 		ASSERT(ss != NULL);
704 
705 		(void) munlinkall(stp, LINKCLOSE|LINKNORMAL, crp, &rval, ss);
706 		netstack_rele(ss->ss_netstack);
707 	}
708 
709 	while (_SAMESTR(qp)) {
710 		/*
711 		 * Holding sd_lock prevents q_next from changing in
712 		 * this stream.
713 		 */
714 		mutex_enter(&stp->sd_lock);
715 		if (!(flag & (FNDELAY|FNONBLOCK)) && (stp->sd_closetime > 0)) {
716 
717 			/*
718 			 * sleep until awakened by strwsrv() or timeout
719 			 */
720 			for (;;) {
721 				mutex_enter(QLOCK(qp->q_next));
722 				if (!(qp->q_next->q_mblkcnt)) {
723 					mutex_exit(QLOCK(qp->q_next));
724 					break;
725 				}
726 				stp->sd_flag |= WSLEEP;
727 
728 				/* ensure strwsrv gets enabled */
729 				qp->q_next->q_flag |= QWANTW;
730 				mutex_exit(QLOCK(qp->q_next));
731 				/* get out if we timed out or recv'd a signal */
732 				if (str_cv_wait(&qp->q_wait, &stp->sd_lock,
733 				    stp->sd_closetime, 0) <= 0) {
734 					break;
735 				}
736 			}
737 			stp->sd_flag &= ~WSLEEP;
738 		}
739 		mutex_exit(&stp->sd_lock);
740 
741 		rmq = qp->q_next;
742 		if (rmq->q_flag & QISDRV) {
743 			ASSERT(!_SAMESTR(rmq));
744 			wait_sq_svc(_RD(qp)->q_syncq);
745 		}
746 
747 		qdetach(_RD(rmq), 1, flag, crp, B_FALSE);
748 	}
749 
750 	/*
751 	 * Since we call pollwakeup in close() now, the poll list should
752 	 * be empty in most cases. The only exception is the layered devices
753 	 * (e.g. the console drivers with redirection modules pushed on top
754 	 * of it).  We have to do this after calling qdetach() because
755 	 * the redirection module won't have torn down the console
756 	 * redirection until after qdetach() has been invoked.
757 	 */
758 	if (stp->sd_pollist.ph_list != NULL) {
759 		pollwakeup(&stp->sd_pollist, POLLERR);
760 		pollhead_clean(&stp->sd_pollist);
761 	}
762 	ASSERT(stp->sd_pollist.ph_list == NULL);
763 	ASSERT(stp->sd_sidp == NULL);
764 	ASSERT(stp->sd_pgidp == NULL);
765 
766 	/* Prevent qenable from re-enabling the stream head queue */
767 	disable_svc(_RD(qp));
768 
769 	/*
770 	 * Wait until service procedure of each queue is
771 	 * run, if QINSERVICE is set.
772 	 */
773 	wait_svc(_RD(qp));
774 
775 	/*
776 	 * Now, flush both queues.
777 	 */
778 	flushq(_RD(qp), FLUSHALL);
779 	flushq(qp, FLUSHALL);
780 
781 	/*
782 	 * If the write queue of the stream head is pointing to a
783 	 * read queue, we have a twisted stream.  If the read queue
784 	 * is alive, convert the stream head queues into a dead end.
785 	 * If the read queue is dead, free the dead pair.
786 	 */
787 	if (qp->q_next && !_SAMESTR(qp)) {
788 		if (qp->q_next->q_qinfo == &deadrend) {	/* half-closed pipe */
789 			flushq(qp->q_next, FLUSHALL); /* ensure no message */
790 			shfree(qp->q_next->q_stream);
791 			freeq(qp->q_next);
792 			freeq(_RD(qp));
793 		} else if (qp->q_next == _RD(qp)) {	/* fifo */
794 			freeq(_RD(qp));
795 		} else {				/* pipe */
796 			freestp = 0;
797 			/*
798 			 * The q_info pointers are never accessed when
799 			 * SQLOCK is held.
800 			 */
801 			ASSERT(qp->q_syncq == _RD(qp)->q_syncq);
802 			mutex_enter(SQLOCK(qp->q_syncq));
803 			qp->q_qinfo = &deadwend;
804 			_RD(qp)->q_qinfo = &deadrend;
805 			mutex_exit(SQLOCK(qp->q_syncq));
806 		}
807 	} else {
808 		freeq(_RD(qp)); /* free stream head queue pair */
809 	}
810 
811 	mutex_enter(&vp->v_lock);
812 	if (stp->sd_iocblk) {
813 		if (stp->sd_iocblk != (mblk_t *)-1) {
814 			freemsg(stp->sd_iocblk);
815 		}
816 		stp->sd_iocblk = NULL;
817 	}
818 	stp->sd_vnode = NULL;
819 	vp->v_stream = NULL;
820 	mutex_exit(&vp->v_lock);
821 	mutex_enter(&stp->sd_lock);
822 	freemsg(stp->sd_cmdblk);
823 	stp->sd_cmdblk = NULL;
824 	stp->sd_flag &= ~STRCLOSE;
825 	cv_broadcast(&stp->sd_monitor);
826 	mutex_exit(&stp->sd_lock);
827 
828 	if (freestp)
829 		shfree(stp);
830 	return (0);
831 }
832 
833 static int
834 strsink(queue_t *q, mblk_t *bp)
835 {
836 	struct copyresp *resp;
837 
838 	switch (bp->b_datap->db_type) {
839 	case M_FLUSH:
840 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
841 			*bp->b_rptr &= ~FLUSHR;
842 			bp->b_flag |= MSGNOLOOP;
843 			/*
844 			 * Protect against the driver passing up
845 			 * messages after it has done a qprocsoff.
846 			 */
847 			if (_OTHERQ(q)->q_next == NULL)
848 				freemsg(bp);
849 			else
850 				qreply(q, bp);
851 		} else {
852 			freemsg(bp);
853 		}
854 		break;
855 
856 	case M_COPYIN:
857 	case M_COPYOUT:
858 		if (bp->b_cont) {
859 			freemsg(bp->b_cont);
860 			bp->b_cont = NULL;
861 		}
862 		bp->b_datap->db_type = M_IOCDATA;
863 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
864 		resp = (struct copyresp *)bp->b_rptr;
865 		resp->cp_rval = (caddr_t)1;	/* failure */
866 		/*
867 		 * Protect against the driver passing up
868 		 * messages after it has done a qprocsoff.
869 		 */
870 		if (_OTHERQ(q)->q_next == NULL)
871 			freemsg(bp);
872 		else
873 			qreply(q, bp);
874 		break;
875 
876 	case M_IOCTL:
877 		if (bp->b_cont) {
878 			freemsg(bp->b_cont);
879 			bp->b_cont = NULL;
880 		}
881 		bp->b_datap->db_type = M_IOCNAK;
882 		/*
883 		 * Protect against the driver passing up
884 		 * messages after it has done a qprocsoff.
885 		 */
886 		if (_OTHERQ(q)->q_next == NULL)
887 			freemsg(bp);
888 		else
889 			qreply(q, bp);
890 		break;
891 
892 	default:
893 		freemsg(bp);
894 		break;
895 	}
896 
897 	return (0);
898 }
899 
900 /*
901  * Clean up after a process when it closes a stream.  This is called
902  * from closef for all closes, whereas strclose is called only for the
903  * last close on a stream.  The siglist is scanned for entries for the
904  * current process, and these are removed.
905  */
906 void
907 strclean(struct vnode *vp)
908 {
909 	strsig_t *ssp, *pssp, *tssp;
910 	stdata_t *stp;
911 	int update = 0;
912 
913 	TRACE_1(TR_FAC_STREAMS_FR,
914 	    TR_STRCLEAN, "strclean:%p", vp);
915 	stp = vp->v_stream;
916 	pssp = NULL;
917 	mutex_enter(&stp->sd_lock);
918 	ssp = stp->sd_siglist;
919 	while (ssp) {
920 		if (ssp->ss_pidp == curproc->p_pidp) {
921 			tssp = ssp->ss_next;
922 			if (pssp)
923 				pssp->ss_next = tssp;
924 			else
925 				stp->sd_siglist = tssp;
926 			mutex_enter(&pidlock);
927 			PID_RELE(ssp->ss_pidp);
928 			mutex_exit(&pidlock);
929 			kmem_free(ssp, sizeof (strsig_t));
930 			update = 1;
931 			ssp = tssp;
932 		} else {
933 			pssp = ssp;
934 			ssp = ssp->ss_next;
935 		}
936 	}
937 	if (update) {
938 		stp->sd_sigflags = 0;
939 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
940 			stp->sd_sigflags |= ssp->ss_events;
941 	}
942 	mutex_exit(&stp->sd_lock);
943 }
944 
945 /*
946  * Used on the last close to remove any remaining items on the siglist.
947  * These could be present on the siglist due to I_ESETSIG calls that
948  * use process groups or processed that do not have an open file descriptor
949  * for this stream (Such entries would not be removed by strclean).
950  */
951 static void
952 strcleanall(struct vnode *vp)
953 {
954 	strsig_t *ssp, *nssp;
955 	stdata_t *stp;
956 
957 	stp = vp->v_stream;
958 	mutex_enter(&stp->sd_lock);
959 	ssp = stp->sd_siglist;
960 	stp->sd_siglist = NULL;
961 	while (ssp) {
962 		nssp = ssp->ss_next;
963 		mutex_enter(&pidlock);
964 		PID_RELE(ssp->ss_pidp);
965 		mutex_exit(&pidlock);
966 		kmem_free(ssp, sizeof (strsig_t));
967 		ssp = nssp;
968 	}
969 	stp->sd_sigflags = 0;
970 	mutex_exit(&stp->sd_lock);
971 }
972 
973 /*
974  * Retrieve the next message from the logical stream head read queue
975  * using either rwnext (if sync stream) or getq_noenab.
976  * It is the callers responsibility to call qbackenable after
977  * it is finished with the message. The caller should not call
978  * qbackenable until after any putback calls to avoid spurious backenabling.
979  *
980  * Also, handle uioa initialization and process any DBLK_UIOA flaged messages.
981  */
982 mblk_t *
983 strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
984     int *errorp)
985 {
986 	sodirect_t *sodp = stp->sd_sodirect;
987 	mblk_t *bp;
988 	int error;
989 	ssize_t rbytes = 0;
990 
991 	/* Holding sd_lock prevents the read queue from changing  */
992 	ASSERT(MUTEX_HELD(&stp->sd_lock));
993 
994 	if (uiop != NULL && stp->sd_struiordq != NULL &&
995 	    q->q_first == NULL &&
996 	    (!first || (stp->sd_wakeq & RSLEEP))) {
997 		/*
998 		 * Stream supports rwnext() for the read side.
999 		 * If this is the first time we're called by e.g. strread
1000 		 * only do the downcall if there is a deferred wakeup
1001 		 * (registered in sd_wakeq).
1002 		 */
1003 		struiod_t uiod;
1004 
1005 		if (first)
1006 			stp->sd_wakeq &= ~RSLEEP;
1007 
1008 		(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
1009 		    sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
1010 		uiod.d_mp = 0;
1011 		/*
1012 		 * Mark that a thread is in rwnext on the read side
1013 		 * to prevent strrput from nacking ioctls immediately.
1014 		 * When the last concurrent rwnext returns
1015 		 * the ioctls are nack'ed.
1016 		 */
1017 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1018 		stp->sd_struiodnak++;
1019 		/*
1020 		 * Note: rwnext will drop sd_lock.
1021 		 */
1022 		error = rwnext(q, &uiod);
1023 		ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
1024 		mutex_enter(&stp->sd_lock);
1025 		stp->sd_struiodnak--;
1026 		while (stp->sd_struiodnak == 0 &&
1027 		    ((bp = stp->sd_struionak) != NULL)) {
1028 			stp->sd_struionak = bp->b_next;
1029 			bp->b_next = NULL;
1030 			bp->b_datap->db_type = M_IOCNAK;
1031 			/*
1032 			 * Protect against the driver passing up
1033 			 * messages after it has done a qprocsoff.
1034 			 */
1035 			if (_OTHERQ(q)->q_next == NULL)
1036 				freemsg(bp);
1037 			else {
1038 				mutex_exit(&stp->sd_lock);
1039 				qreply(q, bp);
1040 				mutex_enter(&stp->sd_lock);
1041 			}
1042 		}
1043 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1044 		if (error == 0 || error == EWOULDBLOCK) {
1045 			if ((bp = uiod.d_mp) != NULL) {
1046 				*errorp = 0;
1047 				ASSERT(MUTEX_HELD(&stp->sd_lock));
1048 				return (bp);
1049 			}
1050 			error = 0;
1051 		} else if (error == EINVAL) {
1052 			/*
1053 			 * The stream plumbing must have
1054 			 * changed while we were away, so
1055 			 * just turn off rwnext()s.
1056 			 */
1057 			error = 0;
1058 		} else if (error == EBUSY) {
1059 			/*
1060 			 * The module might have data in transit using putnext
1061 			 * Fall back on waiting + getq.
1062 			 */
1063 			error = 0;
1064 		} else {
1065 			*errorp = error;
1066 			ASSERT(MUTEX_HELD(&stp->sd_lock));
1067 			return (NULL);
1068 		}
1069 		/*
1070 		 * Try a getq in case a rwnext() generated mblk
1071 		 * has bubbled up via strrput().
1072 		 */
1073 	}
1074 	*errorp = 0;
1075 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1076 
1077 	if (sodp != NULL && sodp->sod_state & SOD_ENABLED) {
1078 		if (sodp->sod_uioa.uioa_state & UIOA_INIT) {
1079 			/*
1080 			 * First kstrgetmsg() call for an uioa_t so if any
1081 			 * queued mblk_t's need to consume them before uioa
1082 			 * from below can occur.
1083 			 */
1084 			sodp->sod_uioa.uioa_state &= UIOA_CLR;
1085 			sodp->sod_uioa.uioa_state |= UIOA_ENABLED;
1086 			if (q->q_first != NULL) {
1087 				struioainit(q, sodp, uiop);
1088 			}
1089 		} else if (sodp->sod_uioa.uioa_state &
1090 		    (UIOA_ENABLED|UIOA_FINI)) {
1091 			ASSERT(uiop == (uio_t *)&sodp->sod_uioa);
1092 			rbytes = 0;
1093 		} else {
1094 			rbytes = uiop->uio_resid;
1095 		}
1096 	} else {
1097 		/*
1098 		 * If we have a valid uio, try and use this as a guide for how
1099 		 * many bytes to retrieve from the queue via getq_noenab().
1100 		 * Doing this can avoid unneccesary counting of overlong
1101 		 * messages in putback(). We currently only do this for sockets
1102 		 * and only if there is no sd_rputdatafunc hook.
1103 		 *
1104 		 * The sd_rputdatafunc hook transforms the entire message
1105 		 * before any bytes in it can be given to a client. So, rbytes
1106 		 * must be 0 if there is a hook.
1107 		 */
1108 		if ((uiop != NULL) && (stp->sd_vnode->v_type == VSOCK) &&
1109 		    (stp->sd_rputdatafunc == NULL))
1110 			rbytes = uiop->uio_resid;
1111 	}
1112 
1113 	bp = getq_noenab(q, rbytes);
1114 	sod_uioa_mblk_done(sodp, bp);
1115 
1116 	return (bp);
1117 }
1118 
1119 /*
1120  * Copy out the message pointed to by `bp' into the uio pointed to by `uiop'.
1121  * If the message does not fit in the uio the remainder of it is returned;
1122  * otherwise NULL is returned.  Any embedded zero-length mblk_t's are
1123  * consumed, even if uio_resid reaches zero.  On error, `*errorp' is set to
1124  * the error code, the message is consumed, and NULL is returned.
1125  */
1126 static mblk_t *
1127 struiocopyout(mblk_t *bp, struct uio *uiop, int *errorp)
1128 {
1129 	int error;
1130 	ptrdiff_t n;
1131 	mblk_t *nbp;
1132 
1133 	ASSERT(bp->b_wptr >= bp->b_rptr);
1134 
1135 	do {
1136 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
1137 
1138 		if ((n = MIN(uiop->uio_resid, MBLKL(bp))) != 0) {
1139 			ASSERT(n > 0);
1140 
1141 			error = uiomove(bp->b_rptr, n, UIO_READ, uiop);
1142 			if (error != 0) {
1143 				freemsg(bp);
1144 				*errorp = error;
1145 				return (NULL);
1146 			}
1147 		}
1148 
1149 		bp->b_rptr += n;
1150 		while (bp != NULL && (bp->b_rptr >= bp->b_wptr)) {
1151 			nbp = bp;
1152 			bp = bp->b_cont;
1153 			freeb(nbp);
1154 		}
1155 	} while (bp != NULL && uiop->uio_resid > 0);
1156 
1157 	*errorp = 0;
1158 	return (bp);
1159 }
1160 
1161 /*
1162  * Read a stream according to the mode flags in sd_flag:
1163  *
1164  * (default mode)		- Byte stream, msg boundaries are ignored
1165  * RD_MSGDIS (msg discard)	- Read on msg boundaries and throw away
1166  *				any data remaining in msg
1167  * RD_MSGNODIS (msg non-discard) - Read on msg boundaries and put back
1168  *				any remaining data on head of read queue
1169  *
1170  * Consume readable messages on the front of the queue until
1171  * ttolwp(curthread)->lwp_count
1172  * is satisfied, the readable messages are exhausted, or a message
1173  * boundary is reached in a message mode.  If no data was read and
1174  * the stream was not opened with the NDELAY flag, block until data arrives.
1175  * Otherwise return the data read and update the count.
1176  *
1177  * In default mode a 0 length message signifies end-of-file and terminates
1178  * a read in progress.  The 0 length message is removed from the queue
1179  * only if it is the only message read (no data is read).
1180  *
1181  * An attempt to read an M_PROTO or M_PCPROTO message results in an
1182  * EBADMSG error return, unless either RD_PROTDAT or RD_PROTDIS are set.
1183  * If RD_PROTDAT is set, M_PROTO and M_PCPROTO messages are read as data.
1184  * If RD_PROTDIS is set, the M_PROTO and M_PCPROTO parts of the message
1185  * are unlinked from and M_DATA blocks in the message, the protos are
1186  * thrown away, and the data is read.
1187  */
1188 /* ARGSUSED */
1189 int
1190 strread(struct vnode *vp, struct uio *uiop, cred_t *crp)
1191 {
1192 	struct stdata *stp;
1193 	mblk_t *bp, *nbp;
1194 	queue_t *q;
1195 	int error = 0;
1196 	uint_t old_sd_flag;
1197 	int first;
1198 	char rflg;
1199 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
1200 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
1201 	short delim;
1202 	unsigned char pri = 0;
1203 	char waitflag;
1204 	unsigned char type;
1205 
1206 	TRACE_1(TR_FAC_STREAMS_FR,
1207 	    TR_STRREAD_ENTER, "strread:%p", vp);
1208 	ASSERT(vp->v_stream);
1209 	stp = vp->v_stream;
1210 
1211 	mutex_enter(&stp->sd_lock);
1212 
1213 	if ((error = i_straccess(stp, JCREAD)) != 0) {
1214 		mutex_exit(&stp->sd_lock);
1215 		return (error);
1216 	}
1217 
1218 	if (stp->sd_flag & (STRDERR|STPLEX)) {
1219 		error = strgeterr(stp, STRDERR|STPLEX, 0);
1220 		if (error != 0) {
1221 			mutex_exit(&stp->sd_lock);
1222 			return (error);
1223 		}
1224 	}
1225 
1226 	/*
1227 	 * Loop terminates when uiop->uio_resid == 0.
1228 	 */
1229 	rflg = 0;
1230 	waitflag = READWAIT;
1231 	q = _RD(stp->sd_wrq);
1232 	for (;;) {
1233 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1234 		old_sd_flag = stp->sd_flag;
1235 		mark = 0;
1236 		delim = 0;
1237 		first = 1;
1238 		while ((bp = strget(stp, q, uiop, first, &error)) == NULL) {
1239 			int done = 0;
1240 
1241 			ASSERT(MUTEX_HELD(&stp->sd_lock));
1242 
1243 			if (error != 0)
1244 				goto oops;
1245 
1246 			if (stp->sd_flag & (STRHUP|STREOF)) {
1247 				goto oops;
1248 			}
1249 			if (rflg && !(stp->sd_flag & STRDELIM)) {
1250 				goto oops;
1251 			}
1252 			/*
1253 			 * If a read(fd,buf,0) has been done, there is no
1254 			 * need to sleep. We always have zero bytes to
1255 			 * return.
1256 			 */
1257 			if (uiop->uio_resid == 0) {
1258 				goto oops;
1259 			}
1260 
1261 			qbackenable(q, 0);
1262 
1263 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_WAIT,
1264 			    "strread calls strwaitq:%p, %p, %p",
1265 			    vp, uiop, crp);
1266 			if ((error = strwaitq(stp, waitflag, uiop->uio_resid,
1267 			    uiop->uio_fmode, -1, &done)) != 0 || done) {
1268 				TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_DONE,
1269 				    "strread error or done:%p, %p, %p",
1270 				    vp, uiop, crp);
1271 				if ((uiop->uio_fmode & FNDELAY) &&
1272 				    (stp->sd_flag & OLDNDELAY) &&
1273 				    (error == EAGAIN))
1274 					error = 0;
1275 				goto oops;
1276 			}
1277 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_AWAKE,
1278 			    "strread awakes:%p, %p, %p", vp, uiop, crp);
1279 			if ((error = i_straccess(stp, JCREAD)) != 0) {
1280 				goto oops;
1281 			}
1282 			first = 0;
1283 		}
1284 
1285 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1286 		ASSERT(bp);
1287 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
1288 		pri = bp->b_band;
1289 		/*
1290 		 * Extract any mark information. If the message is not
1291 		 * completely consumed this information will be put in the mblk
1292 		 * that is putback.
1293 		 * If MSGMARKNEXT is set and the message is completely consumed
1294 		 * the STRATMARK flag will be set below. Likewise, if
1295 		 * MSGNOTMARKNEXT is set and the message is
1296 		 * completely consumed STRNOTATMARK will be set.
1297 		 *
1298 		 * For some unknown reason strread only breaks the read at the
1299 		 * last mark.
1300 		 */
1301 		mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
1302 		ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
1303 		    (MSGMARKNEXT|MSGNOTMARKNEXT));
1304 		if (mark != 0 && bp == stp->sd_mark) {
1305 			if (rflg) {
1306 				putback(stp, q, bp, pri);
1307 				goto oops;
1308 			}
1309 			mark |= _LASTMARK;
1310 			stp->sd_mark = NULL;
1311 		}
1312 		if ((stp->sd_flag & STRDELIM) && (bp->b_flag & MSGDELIM))
1313 			delim = 1;
1314 		mutex_exit(&stp->sd_lock);
1315 
1316 		if (STREAM_NEEDSERVICE(stp))
1317 			stream_runservice(stp);
1318 
1319 		type = bp->b_datap->db_type;
1320 
1321 		switch (type) {
1322 
1323 		case M_DATA:
1324 ismdata:
1325 			if (msgnodata(bp)) {
1326 				if (mark || delim) {
1327 					freemsg(bp);
1328 				} else if (rflg) {
1329 
1330 					/*
1331 					 * If already read data put zero
1332 					 * length message back on queue else
1333 					 * free msg and return 0.
1334 					 */
1335 					bp->b_band = pri;
1336 					mutex_enter(&stp->sd_lock);
1337 					putback(stp, q, bp, pri);
1338 					mutex_exit(&stp->sd_lock);
1339 				} else {
1340 					freemsg(bp);
1341 				}
1342 				error =  0;
1343 				goto oops1;
1344 			}
1345 
1346 			rflg = 1;
1347 			waitflag |= NOINTR;
1348 			bp = struiocopyout(bp, uiop, &error);
1349 			if (error != 0)
1350 				goto oops1;
1351 
1352 			mutex_enter(&stp->sd_lock);
1353 			if (bp) {
1354 				/*
1355 				 * Have remaining data in message.
1356 				 * Free msg if in discard mode.
1357 				 */
1358 				if (stp->sd_read_opt & RD_MSGDIS) {
1359 					freemsg(bp);
1360 				} else {
1361 					bp->b_band = pri;
1362 					if ((mark & _LASTMARK) &&
1363 					    (stp->sd_mark == NULL))
1364 						stp->sd_mark = bp;
1365 					bp->b_flag |= mark & ~_LASTMARK;
1366 					if (delim)
1367 						bp->b_flag |= MSGDELIM;
1368 					if (msgnodata(bp))
1369 						freemsg(bp);
1370 					else
1371 						putback(stp, q, bp, pri);
1372 				}
1373 			} else {
1374 				/*
1375 				 * Consumed the complete message.
1376 				 * Move the MSG*MARKNEXT information
1377 				 * to the stream head just in case
1378 				 * the read queue becomes empty.
1379 				 *
1380 				 * If the stream head was at the mark
1381 				 * (STRATMARK) before we dropped sd_lock above
1382 				 * and some data was consumed then we have
1383 				 * moved past the mark thus STRATMARK is
1384 				 * cleared. However, if a message arrived in
1385 				 * strrput during the copyout above causing
1386 				 * STRATMARK to be set we can not clear that
1387 				 * flag.
1388 				 */
1389 				if (mark &
1390 				    (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
1391 					if (mark & MSGMARKNEXT) {
1392 						stp->sd_flag &= ~STRNOTATMARK;
1393 						stp->sd_flag |= STRATMARK;
1394 					} else if (mark & MSGNOTMARKNEXT) {
1395 						stp->sd_flag &= ~STRATMARK;
1396 						stp->sd_flag |= STRNOTATMARK;
1397 					} else {
1398 						stp->sd_flag &=
1399 						    ~(STRATMARK|STRNOTATMARK);
1400 					}
1401 				} else if (rflg && (old_sd_flag & STRATMARK)) {
1402 					stp->sd_flag &= ~STRATMARK;
1403 				}
1404 			}
1405 
1406 			/*
1407 			 * Check for signal messages at the front of the read
1408 			 * queue and generate the signal(s) if appropriate.
1409 			 * The only signal that can be on queue is M_SIG at
1410 			 * this point.
1411 			 */
1412 			while ((((bp = q->q_first)) != NULL) &&
1413 			    (bp->b_datap->db_type == M_SIG)) {
1414 				bp = getq_noenab(q, 0);
1415 				/*
1416 				 * sd_lock is held so the content of the
1417 				 * read queue can not change.
1418 				 */
1419 				ASSERT(bp != NULL && DB_TYPE(bp) == M_SIG);
1420 				strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
1421 				mutex_exit(&stp->sd_lock);
1422 				freemsg(bp);
1423 				if (STREAM_NEEDSERVICE(stp))
1424 					stream_runservice(stp);
1425 				mutex_enter(&stp->sd_lock);
1426 			}
1427 
1428 			if ((uiop->uio_resid == 0) || (mark & _LASTMARK) ||
1429 			    delim ||
1430 			    (stp->sd_read_opt & (RD_MSGDIS|RD_MSGNODIS))) {
1431 				goto oops;
1432 			}
1433 			continue;
1434 
1435 		case M_SIG:
1436 			strsignal(stp, *bp->b_rptr, (int32_t)bp->b_band);
1437 			freemsg(bp);
1438 			mutex_enter(&stp->sd_lock);
1439 			continue;
1440 
1441 		case M_PROTO:
1442 		case M_PCPROTO:
1443 			/*
1444 			 * Only data messages are readable.
1445 			 * Any others generate an error, unless
1446 			 * RD_PROTDIS or RD_PROTDAT is set.
1447 			 */
1448 			if (stp->sd_read_opt & RD_PROTDAT) {
1449 				for (nbp = bp; nbp; nbp = nbp->b_next) {
1450 					if ((nbp->b_datap->db_type ==
1451 					    M_PROTO) ||
1452 					    (nbp->b_datap->db_type ==
1453 					    M_PCPROTO)) {
1454 						nbp->b_datap->db_type = M_DATA;
1455 					} else {
1456 						break;
1457 					}
1458 				}
1459 				/*
1460 				 * clear stream head hi pri flag based on
1461 				 * first message
1462 				 */
1463 				if (type == M_PCPROTO) {
1464 					mutex_enter(&stp->sd_lock);
1465 					stp->sd_flag &= ~STRPRI;
1466 					mutex_exit(&stp->sd_lock);
1467 				}
1468 				goto ismdata;
1469 			} else if (stp->sd_read_opt & RD_PROTDIS) {
1470 				/*
1471 				 * discard non-data messages
1472 				 */
1473 				while (bp &&
1474 				    ((bp->b_datap->db_type == M_PROTO) ||
1475 				    (bp->b_datap->db_type == M_PCPROTO))) {
1476 					nbp = unlinkb(bp);
1477 					freeb(bp);
1478 					bp = nbp;
1479 				}
1480 				/*
1481 				 * clear stream head hi pri flag based on
1482 				 * first message
1483 				 */
1484 				if (type == M_PCPROTO) {
1485 					mutex_enter(&stp->sd_lock);
1486 					stp->sd_flag &= ~STRPRI;
1487 					mutex_exit(&stp->sd_lock);
1488 				}
1489 				if (bp) {
1490 					bp->b_band = pri;
1491 					goto ismdata;
1492 				} else {
1493 					break;
1494 				}
1495 			}
1496 			/* FALLTHRU */
1497 		case M_PASSFP:
1498 			if ((bp->b_datap->db_type == M_PASSFP) &&
1499 			    (stp->sd_read_opt & RD_PROTDIS)) {
1500 				freemsg(bp);
1501 				break;
1502 			}
1503 			mutex_enter(&stp->sd_lock);
1504 			putback(stp, q, bp, pri);
1505 			mutex_exit(&stp->sd_lock);
1506 			if (rflg == 0)
1507 				error = EBADMSG;
1508 			goto oops1;
1509 
1510 		default:
1511 			/*
1512 			 * Garbage on stream head read queue.
1513 			 */
1514 			cmn_err(CE_WARN, "bad %x found at stream head\n",
1515 			    bp->b_datap->db_type);
1516 			freemsg(bp);
1517 			goto oops1;
1518 		}
1519 		mutex_enter(&stp->sd_lock);
1520 	}
1521 oops:
1522 	mutex_exit(&stp->sd_lock);
1523 oops1:
1524 	qbackenable(q, pri);
1525 	return (error);
1526 #undef	_LASTMARK
1527 }
1528 
1529 /*
1530  * Default processing of M_PROTO/M_PCPROTO messages.
1531  * Determine which wakeups and signals are needed.
1532  * This can be replaced by a user-specified procedure for kernel users
1533  * of STREAMS.
1534  */
1535 /* ARGSUSED */
1536 mblk_t *
1537 strrput_proto(vnode_t *vp, mblk_t *mp,
1538     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1539     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1540 {
1541 	*wakeups = RSLEEP;
1542 	*allmsgsigs = 0;
1543 
1544 	switch (mp->b_datap->db_type) {
1545 	case M_PROTO:
1546 		if (mp->b_band == 0) {
1547 			*firstmsgsigs = S_INPUT | S_RDNORM;
1548 			*pollwakeups = POLLIN | POLLRDNORM;
1549 		} else {
1550 			*firstmsgsigs = S_INPUT | S_RDBAND;
1551 			*pollwakeups = POLLIN | POLLRDBAND;
1552 		}
1553 		break;
1554 	case M_PCPROTO:
1555 		*firstmsgsigs = S_HIPRI;
1556 		*pollwakeups = POLLPRI;
1557 		break;
1558 	}
1559 	return (mp);
1560 }
1561 
1562 /*
1563  * Default processing of everything but M_DATA, M_PROTO, M_PCPROTO and
1564  * M_PASSFP messages.
1565  * Determine which wakeups and signals are needed.
1566  * This can be replaced by a user-specified procedure for kernel users
1567  * of STREAMS.
1568  */
1569 /* ARGSUSED */
1570 mblk_t *
1571 strrput_misc(vnode_t *vp, mblk_t *mp,
1572     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1573     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1574 {
1575 	*wakeups = 0;
1576 	*firstmsgsigs = 0;
1577 	*allmsgsigs = 0;
1578 	*pollwakeups = 0;
1579 	return (mp);
1580 }
1581 
1582 /*
1583  * Stream read put procedure.  Called from downstream driver/module
1584  * with messages for the stream head.  Data, protocol, and in-stream
1585  * signal messages are placed on the queue, others are handled directly.
1586  */
1587 int
1588 strrput(queue_t *q, mblk_t *bp)
1589 {
1590 	struct stdata	*stp;
1591 	ulong_t		rput_opt;
1592 	strwakeup_t	wakeups;
1593 	strsigset_t	firstmsgsigs;	/* Signals if first message on queue */
1594 	strsigset_t	allmsgsigs;	/* Signals for all messages */
1595 	strsigset_t	signals;	/* Signals events to generate */
1596 	strpollset_t	pollwakeups;
1597 	mblk_t		*nextbp;
1598 	uchar_t		band = 0;
1599 	int		hipri_sig;
1600 
1601 	stp = (struct stdata *)q->q_ptr;
1602 	/*
1603 	 * Use rput_opt for optimized access to the SR_ flags except
1604 	 * SR_POLLIN. That flag has to be checked under sd_lock since it
1605 	 * is modified by strpoll().
1606 	 */
1607 	rput_opt = stp->sd_rput_opt;
1608 
1609 	ASSERT(qclaimed(q));
1610 	TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_ENTER,
1611 	    "strrput called with message type:q %p bp %p", q, bp);
1612 
1613 	/*
1614 	 * Perform initial processing and pass to the parameterized functions.
1615 	 */
1616 	ASSERT(bp->b_next == NULL);
1617 
1618 	switch (bp->b_datap->db_type) {
1619 	case M_DATA:
1620 		/*
1621 		 * sockfs is the only consumer of STREOF and when it is set,
1622 		 * it implies that the receiver is not interested in receiving
1623 		 * any more data, hence the mblk is freed to prevent unnecessary
1624 		 * message queueing at the stream head.
1625 		 */
1626 		if (stp->sd_flag == STREOF) {
1627 			freemsg(bp);
1628 			return (0);
1629 		}
1630 		if ((rput_opt & SR_IGN_ZEROLEN) &&
1631 		    bp->b_rptr == bp->b_wptr && msgnodata(bp)) {
1632 			/*
1633 			 * Ignore zero-length M_DATA messages. These might be
1634 			 * generated by some transports.
1635 			 * The zero-length M_DATA messages, even if they
1636 			 * are ignored, should effect the atmark tracking and
1637 			 * should wake up a thread sleeping in strwaitmark.
1638 			 */
1639 			mutex_enter(&stp->sd_lock);
1640 			if (bp->b_flag & MSGMARKNEXT) {
1641 				/*
1642 				 * Record the position of the mark either
1643 				 * in q_last or in STRATMARK.
1644 				 */
1645 				if (q->q_last != NULL) {
1646 					q->q_last->b_flag &= ~MSGNOTMARKNEXT;
1647 					q->q_last->b_flag |= MSGMARKNEXT;
1648 				} else {
1649 					stp->sd_flag &= ~STRNOTATMARK;
1650 					stp->sd_flag |= STRATMARK;
1651 				}
1652 			} else if (bp->b_flag & MSGNOTMARKNEXT) {
1653 				/*
1654 				 * Record that this is not the position of
1655 				 * the mark either in q_last or in
1656 				 * STRNOTATMARK.
1657 				 */
1658 				if (q->q_last != NULL) {
1659 					q->q_last->b_flag &= ~MSGMARKNEXT;
1660 					q->q_last->b_flag |= MSGNOTMARKNEXT;
1661 				} else {
1662 					stp->sd_flag &= ~STRATMARK;
1663 					stp->sd_flag |= STRNOTATMARK;
1664 				}
1665 			}
1666 			if (stp->sd_flag & RSLEEP) {
1667 				stp->sd_flag &= ~RSLEEP;
1668 				cv_broadcast(&q->q_wait);
1669 			}
1670 			mutex_exit(&stp->sd_lock);
1671 			freemsg(bp);
1672 			return (0);
1673 		}
1674 		wakeups = RSLEEP;
1675 		if (bp->b_band == 0) {
1676 			firstmsgsigs = S_INPUT | S_RDNORM;
1677 			pollwakeups = POLLIN | POLLRDNORM;
1678 		} else {
1679 			firstmsgsigs = S_INPUT | S_RDBAND;
1680 			pollwakeups = POLLIN | POLLRDBAND;
1681 		}
1682 		if (rput_opt & SR_SIGALLDATA)
1683 			allmsgsigs = firstmsgsigs;
1684 		else
1685 			allmsgsigs = 0;
1686 
1687 		mutex_enter(&stp->sd_lock);
1688 		if ((rput_opt & SR_CONSOL_DATA) &&
1689 		    (q->q_last != NULL) &&
1690 		    (bp->b_flag & (MSGMARK|MSGDELIM)) == 0) {
1691 			/*
1692 			 * Consolidate an M_DATA message onto an M_DATA,
1693 			 * M_PROTO, or M_PCPROTO by merging it with q_last.
1694 			 * The consolidation does not take place if
1695 			 * the old message is marked with either of the
1696 			 * marks or the delim flag or if the new
1697 			 * message is marked with MSGMARK. The MSGMARK
1698 			 * check is needed to handle the odd semantics of
1699 			 * MSGMARK where essentially the whole message
1700 			 * is to be treated as marked.
1701 			 * Carry any MSGMARKNEXT  and MSGNOTMARKNEXT from the
1702 			 * new message to the front of the b_cont chain.
1703 			 */
1704 			mblk_t *lbp = q->q_last;
1705 			unsigned char db_type = lbp->b_datap->db_type;
1706 
1707 			if ((db_type == M_DATA || db_type == M_PROTO ||
1708 			    db_type == M_PCPROTO) &&
1709 			    !(lbp->b_flag & (MSGDELIM|MSGMARK|MSGMARKNEXT))) {
1710 				rmvq_noenab(q, lbp);
1711 				/*
1712 				 * The first message in the b_cont list
1713 				 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
1714 				 * We need to handle the case where we
1715 				 * are appending:
1716 				 *
1717 				 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
1718 				 * 2) a MSGMARKNEXT to a plain message.
1719 				 * 3) a MSGNOTMARKNEXT to a plain message
1720 				 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
1721 				 *    message.
1722 				 *
1723 				 * Thus we never append a MSGMARKNEXT or
1724 				 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
1725 				 */
1726 				if (bp->b_flag & MSGMARKNEXT) {
1727 					lbp->b_flag |= MSGMARKNEXT;
1728 					lbp->b_flag &= ~MSGNOTMARKNEXT;
1729 					bp->b_flag &= ~MSGMARKNEXT;
1730 				} else if (bp->b_flag & MSGNOTMARKNEXT) {
1731 					lbp->b_flag |= MSGNOTMARKNEXT;
1732 					bp->b_flag &= ~MSGNOTMARKNEXT;
1733 				}
1734 
1735 				linkb(lbp, bp);
1736 				bp = lbp;
1737 				/*
1738 				 * The new message logically isn't the first
1739 				 * even though the q_first check below thinks
1740 				 * it is. Clear the firstmsgsigs to make it
1741 				 * not appear to be first.
1742 				 */
1743 				firstmsgsigs = 0;
1744 			}
1745 		}
1746 		break;
1747 
1748 	case M_PASSFP:
1749 		wakeups = RSLEEP;
1750 		allmsgsigs = 0;
1751 		if (bp->b_band == 0) {
1752 			firstmsgsigs = S_INPUT | S_RDNORM;
1753 			pollwakeups = POLLIN | POLLRDNORM;
1754 		} else {
1755 			firstmsgsigs = S_INPUT | S_RDBAND;
1756 			pollwakeups = POLLIN | POLLRDBAND;
1757 		}
1758 		mutex_enter(&stp->sd_lock);
1759 		break;
1760 
1761 	case M_PROTO:
1762 	case M_PCPROTO:
1763 		ASSERT(stp->sd_rprotofunc != NULL);
1764 		bp = (stp->sd_rprotofunc)(stp->sd_vnode, bp,
1765 		    &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1766 #define	ALLSIG	(S_INPUT|S_HIPRI|S_OUTPUT|S_MSG|S_ERROR|S_HANGUP|S_RDNORM|\
1767 		S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)
1768 #define	ALLPOLL	(POLLIN|POLLPRI|POLLOUT|POLLRDNORM|POLLWRNORM|POLLRDBAND|\
1769 		POLLWRBAND)
1770 
1771 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1772 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1773 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1774 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1775 
1776 		mutex_enter(&stp->sd_lock);
1777 		break;
1778 
1779 	default:
1780 		ASSERT(stp->sd_rmiscfunc != NULL);
1781 		bp = (stp->sd_rmiscfunc)(stp->sd_vnode, bp,
1782 		    &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1783 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1784 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1785 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1786 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1787 #undef	ALLSIG
1788 #undef	ALLPOLL
1789 		mutex_enter(&stp->sd_lock);
1790 		break;
1791 	}
1792 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1793 
1794 	/* By default generate superset of signals */
1795 	signals = (firstmsgsigs | allmsgsigs);
1796 
1797 	/*
1798 	 * The  proto and misc functions can return multiple messages
1799 	 * as a b_next chain. Such messages are processed separately.
1800 	 */
1801 one_more:
1802 	hipri_sig = 0;
1803 	if (bp == NULL) {
1804 		nextbp = NULL;
1805 	} else {
1806 		nextbp = bp->b_next;
1807 		bp->b_next = NULL;
1808 
1809 		switch (bp->b_datap->db_type) {
1810 		case M_PCPROTO:
1811 			/*
1812 			 * Only one priority protocol message is allowed at the
1813 			 * stream head at a time.
1814 			 */
1815 			if (stp->sd_flag & STRPRI) {
1816 				TRACE_0(TR_FAC_STREAMS_FR, TR_STRRPUT_PROTERR,
1817 				    "M_PCPROTO already at head");
1818 				freemsg(bp);
1819 				mutex_exit(&stp->sd_lock);
1820 				goto done;
1821 			}
1822 			stp->sd_flag |= STRPRI;
1823 			hipri_sig = 1;
1824 			/* FALLTHRU */
1825 		case M_DATA:
1826 		case M_PROTO:
1827 		case M_PASSFP:
1828 			band = bp->b_band;
1829 			/*
1830 			 * Marking doesn't work well when messages
1831 			 * are marked in more than one band.  We only
1832 			 * remember the last message received, even if
1833 			 * it is placed on the queue ahead of other
1834 			 * marked messages.
1835 			 */
1836 			if (bp->b_flag & MSGMARK)
1837 				stp->sd_mark = bp;
1838 			(void) putq(q, bp);
1839 
1840 			/*
1841 			 * If message is a PCPROTO message, always use
1842 			 * firstmsgsigs to determine if a signal should be
1843 			 * sent as strrput is the only place to send
1844 			 * signals for PCPROTO. Other messages are based on
1845 			 * the STRGETINPROG flag. The flag determines if
1846 			 * strrput or (k)strgetmsg will be responsible for
1847 			 * sending the signals, in the firstmsgsigs case.
1848 			 */
1849 			if ((hipri_sig == 1) ||
1850 			    (((stp->sd_flag & STRGETINPROG) == 0) &&
1851 			    (q->q_first == bp)))
1852 				signals = (firstmsgsigs | allmsgsigs);
1853 			else
1854 				signals = allmsgsigs;
1855 			break;
1856 
1857 		default:
1858 			mutex_exit(&stp->sd_lock);
1859 			(void) strrput_nondata(q, bp);
1860 			mutex_enter(&stp->sd_lock);
1861 			break;
1862 		}
1863 	}
1864 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1865 	/*
1866 	 * Wake sleeping read/getmsg and cancel deferred wakeup
1867 	 */
1868 	if (wakeups & RSLEEP)
1869 		stp->sd_wakeq &= ~RSLEEP;
1870 
1871 	wakeups &= stp->sd_flag;
1872 	if (wakeups & RSLEEP) {
1873 		stp->sd_flag &= ~RSLEEP;
1874 		cv_broadcast(&q->q_wait);
1875 	}
1876 	if (wakeups & WSLEEP) {
1877 		stp->sd_flag &= ~WSLEEP;
1878 		cv_broadcast(&_WR(q)->q_wait);
1879 	}
1880 
1881 	if (pollwakeups != 0) {
1882 		if (pollwakeups == (POLLIN | POLLRDNORM)) {
1883 			/*
1884 			 * Can't use rput_opt since it was not
1885 			 * read when sd_lock was held and SR_POLLIN is changed
1886 			 * by strpoll() under sd_lock.
1887 			 */
1888 			if (!(stp->sd_rput_opt & SR_POLLIN))
1889 				goto no_pollwake;
1890 			stp->sd_rput_opt &= ~SR_POLLIN;
1891 		}
1892 		mutex_exit(&stp->sd_lock);
1893 		pollwakeup(&stp->sd_pollist, pollwakeups);
1894 		mutex_enter(&stp->sd_lock);
1895 	}
1896 no_pollwake:
1897 
1898 	/*
1899 	 * strsendsig can handle multiple signals with a
1900 	 * single call.
1901 	 */
1902 	if (stp->sd_sigflags & signals)
1903 		strsendsig(stp->sd_siglist, signals, band, 0);
1904 	mutex_exit(&stp->sd_lock);
1905 
1906 
1907 done:
1908 	if (nextbp == NULL)
1909 		return (0);
1910 
1911 	/*
1912 	 * Any signals were handled the first time.
1913 	 * Wakeups and pollwakeups are redone to avoid any race
1914 	 * conditions - all the messages are not queued until the
1915 	 * last message has been processed by strrput.
1916 	 */
1917 	bp = nextbp;
1918 	signals = firstmsgsigs = allmsgsigs = 0;
1919 	mutex_enter(&stp->sd_lock);
1920 	goto one_more;
1921 }
1922 
1923 static void
1924 log_dupioc(queue_t *rq, mblk_t *bp)
1925 {
1926 	queue_t *wq, *qp;
1927 	char *modnames, *mnp, *dname;
1928 	size_t maxmodstr;
1929 	boolean_t islast;
1930 
1931 	/*
1932 	 * Allocate a buffer large enough to hold the names of nstrpush modules
1933 	 * and one driver, with spaces between and NUL terminator.  If we can't
1934 	 * get memory, then we'll just log the driver name.
1935 	 */
1936 	maxmodstr = nstrpush * (FMNAMESZ + 1);
1937 	mnp = modnames = kmem_alloc(maxmodstr, KM_NOSLEEP);
1938 
1939 	/* march down write side to print log message down to the driver */
1940 	wq = WR(rq);
1941 
1942 	/* make sure q_next doesn't shift around while we're grabbing data */
1943 	claimstr(wq);
1944 	qp = wq->q_next;
1945 	do {
1946 		dname = Q2NAME(qp);
1947 		islast = !SAMESTR(qp) || qp->q_next == NULL;
1948 		if (modnames == NULL) {
1949 			/*
1950 			 * If we don't have memory, then get the driver name in
1951 			 * the log where we can see it.  Note that memory
1952 			 * pressure is a possible cause of these sorts of bugs.
1953 			 */
1954 			if (islast) {
1955 				modnames = dname;
1956 				maxmodstr = 0;
1957 			}
1958 		} else {
1959 			mnp += snprintf(mnp, FMNAMESZ + 1, "%s", dname);
1960 			if (!islast)
1961 				*mnp++ = ' ';
1962 		}
1963 		qp = qp->q_next;
1964 	} while (!islast);
1965 	releasestr(wq);
1966 	/* Cannot happen unless stream head is corrupt. */
1967 	ASSERT(modnames != NULL);
1968 	(void) strlog(rq->q_qinfo->qi_minfo->mi_idnum, 0, 1,
1969 	    SL_CONSOLE|SL_TRACE|SL_ERROR,
1970 	    "Warning: stream %p received duplicate %X M_IOC%s; module list: %s",
1971 	    rq->q_ptr, ((struct iocblk *)bp->b_rptr)->ioc_cmd,
1972 	    (DB_TYPE(bp) == M_IOCACK ? "ACK" : "NAK"), modnames);
1973 	if (maxmodstr != 0)
1974 		kmem_free(modnames, maxmodstr);
1975 }
1976 
1977 int
1978 strrput_nondata(queue_t *q, mblk_t *bp)
1979 {
1980 	struct stdata *stp;
1981 	struct iocblk *iocbp;
1982 	struct stroptions *sop;
1983 	struct copyreq *reqp;
1984 	struct copyresp *resp;
1985 	unsigned char bpri;
1986 	unsigned char  flushed_already = 0;
1987 
1988 	stp = (struct stdata *)q->q_ptr;
1989 
1990 	ASSERT(!(stp->sd_flag & STPLEX));
1991 	ASSERT(qclaimed(q));
1992 
1993 	switch (bp->b_datap->db_type) {
1994 	case M_ERROR:
1995 		/*
1996 		 * An error has occurred downstream, the errno is in the first
1997 		 * bytes of the message.
1998 		 */
1999 		if ((bp->b_wptr - bp->b_rptr) == 2) {	/* New flavor */
2000 			unsigned char rw = 0;
2001 
2002 			mutex_enter(&stp->sd_lock);
2003 			if (*bp->b_rptr != NOERROR) {	/* read error */
2004 				if (*bp->b_rptr != 0) {
2005 					if (stp->sd_flag & STRDERR)
2006 						flushed_already |= FLUSHR;
2007 					stp->sd_flag |= STRDERR;
2008 					rw |= FLUSHR;
2009 				} else {
2010 					stp->sd_flag &= ~STRDERR;
2011 				}
2012 				stp->sd_rerror = *bp->b_rptr;
2013 			}
2014 			bp->b_rptr++;
2015 			if (*bp->b_rptr != NOERROR) {	/* write error */
2016 				if (*bp->b_rptr != 0) {
2017 					if (stp->sd_flag & STWRERR)
2018 						flushed_already |= FLUSHW;
2019 					stp->sd_flag |= STWRERR;
2020 					rw |= FLUSHW;
2021 				} else {
2022 					stp->sd_flag &= ~STWRERR;
2023 				}
2024 				stp->sd_werror = *bp->b_rptr;
2025 			}
2026 			if (rw) {
2027 				TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_WAKE,
2028 				    "strrput cv_broadcast:q %p, bp %p",
2029 				    q, bp);
2030 				cv_broadcast(&q->q_wait); /* readers */
2031 				cv_broadcast(&_WR(q)->q_wait); /* writers */
2032 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
2033 
2034 				mutex_exit(&stp->sd_lock);
2035 				pollwakeup(&stp->sd_pollist, POLLERR);
2036 				mutex_enter(&stp->sd_lock);
2037 
2038 				if (stp->sd_sigflags & S_ERROR)
2039 					strsendsig(stp->sd_siglist, S_ERROR, 0,
2040 					    ((rw & FLUSHR) ? stp->sd_rerror :
2041 					    stp->sd_werror));
2042 				mutex_exit(&stp->sd_lock);
2043 				/*
2044 				 * Send the M_FLUSH only
2045 				 * for the first M_ERROR
2046 				 * message on the stream
2047 				 */
2048 				if (flushed_already == rw) {
2049 					freemsg(bp);
2050 					return (0);
2051 				}
2052 
2053 				bp->b_datap->db_type = M_FLUSH;
2054 				*bp->b_rptr = rw;
2055 				bp->b_wptr = bp->b_rptr + 1;
2056 				/*
2057 				 * Protect against the driver
2058 				 * passing up messages after
2059 				 * it has done a qprocsoff
2060 				 */
2061 				if (_OTHERQ(q)->q_next == NULL)
2062 					freemsg(bp);
2063 				else
2064 					qreply(q, bp);
2065 				return (0);
2066 			} else
2067 				mutex_exit(&stp->sd_lock);
2068 		} else if (*bp->b_rptr != 0) {		/* Old flavor */
2069 				if (stp->sd_flag & (STRDERR|STWRERR))
2070 					flushed_already = FLUSHRW;
2071 				mutex_enter(&stp->sd_lock);
2072 				stp->sd_flag |= (STRDERR|STWRERR);
2073 				stp->sd_rerror = *bp->b_rptr;
2074 				stp->sd_werror = *bp->b_rptr;
2075 				TRACE_2(TR_FAC_STREAMS_FR,
2076 				    TR_STRRPUT_WAKE2,
2077 				    "strrput wakeup #2:q %p, bp %p", q, bp);
2078 				cv_broadcast(&q->q_wait); /* the readers */
2079 				cv_broadcast(&_WR(q)->q_wait); /* the writers */
2080 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
2081 
2082 				mutex_exit(&stp->sd_lock);
2083 				pollwakeup(&stp->sd_pollist, POLLERR);
2084 				mutex_enter(&stp->sd_lock);
2085 
2086 				if (stp->sd_sigflags & S_ERROR)
2087 					strsendsig(stp->sd_siglist, S_ERROR, 0,
2088 					    (stp->sd_werror ? stp->sd_werror :
2089 					    stp->sd_rerror));
2090 				mutex_exit(&stp->sd_lock);
2091 
2092 				/*
2093 				 * Send the M_FLUSH only
2094 				 * for the first M_ERROR
2095 				 * message on the stream
2096 				 */
2097 				if (flushed_already != FLUSHRW) {
2098 					bp->b_datap->db_type = M_FLUSH;
2099 					*bp->b_rptr = FLUSHRW;
2100 					/*
2101 					 * Protect against the driver passing up
2102 					 * messages after it has done a
2103 					 * qprocsoff.
2104 					 */
2105 				if (_OTHERQ(q)->q_next == NULL)
2106 					freemsg(bp);
2107 				else
2108 					qreply(q, bp);
2109 				return (0);
2110 				}
2111 		}
2112 		freemsg(bp);
2113 		return (0);
2114 
2115 	case M_HANGUP:
2116 
2117 		freemsg(bp);
2118 		mutex_enter(&stp->sd_lock);
2119 		stp->sd_werror = ENXIO;
2120 		stp->sd_flag |= STRHUP;
2121 		stp->sd_flag &= ~(WSLEEP|RSLEEP);
2122 
2123 		/*
2124 		 * send signal if controlling tty
2125 		 */
2126 
2127 		if (stp->sd_sidp) {
2128 			prsignal(stp->sd_sidp, SIGHUP);
2129 			if (stp->sd_sidp != stp->sd_pgidp)
2130 				pgsignal(stp->sd_pgidp, SIGTSTP);
2131 		}
2132 
2133 		/*
2134 		 * wake up read, write, and exception pollers and
2135 		 * reset wakeup mechanism.
2136 		 */
2137 		cv_broadcast(&q->q_wait);	/* the readers */
2138 		cv_broadcast(&_WR(q)->q_wait);	/* the writers */
2139 		cv_broadcast(&stp->sd_monitor);	/* the ioctllers */
2140 		strhup(stp);
2141 		mutex_exit(&stp->sd_lock);
2142 		return (0);
2143 
2144 	case M_UNHANGUP:
2145 		freemsg(bp);
2146 		mutex_enter(&stp->sd_lock);
2147 		stp->sd_werror = 0;
2148 		stp->sd_flag &= ~STRHUP;
2149 		mutex_exit(&stp->sd_lock);
2150 		return (0);
2151 
2152 	case M_SIG:
2153 		/*
2154 		 * Someone downstream wants to post a signal.  The
2155 		 * signal to post is contained in the first byte of the
2156 		 * message.  If the message would go on the front of
2157 		 * the queue, send a signal to the process group
2158 		 * (if not SIGPOLL) or to the siglist processes
2159 		 * (SIGPOLL).  If something is already on the queue,
2160 		 * OR if we are delivering a delayed suspend (*sigh*
2161 		 * another "tty" hack) and there's no one sleeping already,
2162 		 * just enqueue the message.
2163 		 */
2164 		mutex_enter(&stp->sd_lock);
2165 		if (q->q_first || (*bp->b_rptr == SIGTSTP &&
2166 		    !(stp->sd_flag & RSLEEP))) {
2167 			(void) putq(q, bp);
2168 			mutex_exit(&stp->sd_lock);
2169 			return (0);
2170 		}
2171 		mutex_exit(&stp->sd_lock);
2172 		/* FALLTHRU */
2173 
2174 	case M_PCSIG:
2175 		/*
2176 		 * Don't enqueue, just post the signal.
2177 		 */
2178 		strsignal(stp, *bp->b_rptr, 0L);
2179 		freemsg(bp);
2180 		return (0);
2181 
2182 	case M_CMD:
2183 		if (MBLKL(bp) != sizeof (cmdblk_t)) {
2184 			freemsg(bp);
2185 			return (0);
2186 		}
2187 
2188 		mutex_enter(&stp->sd_lock);
2189 		if (stp->sd_flag & STRCMDWAIT) {
2190 			ASSERT(stp->sd_cmdblk == NULL);
2191 			stp->sd_cmdblk = bp;
2192 			cv_broadcast(&stp->sd_monitor);
2193 			mutex_exit(&stp->sd_lock);
2194 		} else {
2195 			mutex_exit(&stp->sd_lock);
2196 			freemsg(bp);
2197 		}
2198 		return (0);
2199 
2200 	case M_FLUSH:
2201 		/*
2202 		 * Flush queues.  The indication of which queues to flush
2203 		 * is in the first byte of the message.  If the read queue
2204 		 * is specified, then flush it.  If FLUSHBAND is set, just
2205 		 * flush the band specified by the second byte of the message.
2206 		 *
2207 		 * If a module has issued a M_SETOPT to not flush hi
2208 		 * priority messages off of the stream head, then pass this
2209 		 * flag into the flushq code to preserve such messages.
2210 		 */
2211 
2212 		if (*bp->b_rptr & FLUSHR) {
2213 			mutex_enter(&stp->sd_lock);
2214 			if (*bp->b_rptr & FLUSHBAND) {
2215 				ASSERT((bp->b_wptr - bp->b_rptr) >= 2);
2216 				flushband(q, *(bp->b_rptr + 1), FLUSHALL);
2217 			} else
2218 				flushq_common(q, FLUSHALL,
2219 				    stp->sd_read_opt & RFLUSHPCPROT);
2220 			if ((q->q_first == NULL) ||
2221 			    (q->q_first->b_datap->db_type < QPCTL))
2222 				stp->sd_flag &= ~STRPRI;
2223 			else {
2224 				ASSERT(stp->sd_flag & STRPRI);
2225 			}
2226 			mutex_exit(&stp->sd_lock);
2227 		}
2228 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
2229 			*bp->b_rptr &= ~FLUSHR;
2230 			bp->b_flag |= MSGNOLOOP;
2231 			/*
2232 			 * Protect against the driver passing up
2233 			 * messages after it has done a qprocsoff.
2234 			 */
2235 			if (_OTHERQ(q)->q_next == NULL)
2236 				freemsg(bp);
2237 			else
2238 				qreply(q, bp);
2239 			return (0);
2240 		}
2241 		freemsg(bp);
2242 		return (0);
2243 
2244 	case M_IOCACK:
2245 	case M_IOCNAK:
2246 		iocbp = (struct iocblk *)bp->b_rptr;
2247 		/*
2248 		 * If not waiting for ACK or NAK then just free msg.
2249 		 * If incorrect id sequence number then just free msg.
2250 		 * If already have ACK or NAK for user then this is a
2251 		 *    duplicate, display a warning and free the msg.
2252 		 */
2253 		mutex_enter(&stp->sd_lock);
2254 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2255 		    (stp->sd_iocid != iocbp->ioc_id)) {
2256 			/*
2257 			 * If the ACK/NAK is a dup, display a message
2258 			 * Dup is when sd_iocid == ioc_id, and
2259 			 * sd_iocblk == <valid ptr> or -1 (the former
2260 			 * is when an ioctl has been put on the stream
2261 			 * head, but has not yet been consumed, the
2262 			 * later is when it has been consumed).
2263 			 */
2264 			if ((stp->sd_iocid == iocbp->ioc_id) &&
2265 			    (stp->sd_iocblk != NULL)) {
2266 				log_dupioc(q, bp);
2267 			}
2268 			freemsg(bp);
2269 			mutex_exit(&stp->sd_lock);
2270 			return (0);
2271 		}
2272 
2273 		/*
2274 		 * Assign ACK or NAK to user and wake up.
2275 		 */
2276 		stp->sd_iocblk = bp;
2277 		cv_broadcast(&stp->sd_monitor);
2278 		mutex_exit(&stp->sd_lock);
2279 		return (0);
2280 
2281 	case M_COPYIN:
2282 	case M_COPYOUT:
2283 		reqp = (struct copyreq *)bp->b_rptr;
2284 
2285 		/*
2286 		 * If not waiting for ACK or NAK then just fail request.
2287 		 * If already have ACK, NAK, or copy request, then just
2288 		 * fail request.
2289 		 * If incorrect id sequence number then just fail request.
2290 		 */
2291 		mutex_enter(&stp->sd_lock);
2292 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2293 		    (stp->sd_iocid != reqp->cq_id)) {
2294 			if (bp->b_cont) {
2295 				freemsg(bp->b_cont);
2296 				bp->b_cont = NULL;
2297 			}
2298 			bp->b_datap->db_type = M_IOCDATA;
2299 			bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
2300 			resp = (struct copyresp *)bp->b_rptr;
2301 			resp->cp_rval = (caddr_t)1;	/* failure */
2302 			mutex_exit(&stp->sd_lock);
2303 			putnext(stp->sd_wrq, bp);
2304 			return (0);
2305 		}
2306 
2307 		/*
2308 		 * Assign copy request to user and wake up.
2309 		 */
2310 		stp->sd_iocblk = bp;
2311 		cv_broadcast(&stp->sd_monitor);
2312 		mutex_exit(&stp->sd_lock);
2313 		return (0);
2314 
2315 	case M_SETOPTS:
2316 		/*
2317 		 * Set stream head options (read option, write offset,
2318 		 * min/max packet size, and/or high/low water marks for
2319 		 * the read side only).
2320 		 */
2321 
2322 		bpri = 0;
2323 		sop = (struct stroptions *)bp->b_rptr;
2324 		mutex_enter(&stp->sd_lock);
2325 		if (sop->so_flags & SO_READOPT) {
2326 			switch (sop->so_readopt & RMODEMASK) {
2327 			case RNORM:
2328 				stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
2329 				break;
2330 
2331 			case RMSGD:
2332 				stp->sd_read_opt =
2333 				    ((stp->sd_read_opt & ~RD_MSGNODIS) |
2334 				    RD_MSGDIS);
2335 				break;
2336 
2337 			case RMSGN:
2338 				stp->sd_read_opt =
2339 				    ((stp->sd_read_opt & ~RD_MSGDIS) |
2340 				    RD_MSGNODIS);
2341 				break;
2342 			}
2343 			switch (sop->so_readopt & RPROTMASK) {
2344 			case RPROTNORM:
2345 				stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
2346 				break;
2347 
2348 			case RPROTDAT:
2349 				stp->sd_read_opt =
2350 				    ((stp->sd_read_opt & ~RD_PROTDIS) |
2351 				    RD_PROTDAT);
2352 				break;
2353 
2354 			case RPROTDIS:
2355 				stp->sd_read_opt =
2356 				    ((stp->sd_read_opt & ~RD_PROTDAT) |
2357 				    RD_PROTDIS);
2358 				break;
2359 			}
2360 			switch (sop->so_readopt & RFLUSHMASK) {
2361 			case RFLUSHPCPROT:
2362 				/*
2363 				 * This sets the stream head to NOT flush
2364 				 * M_PCPROTO messages.
2365 				 */
2366 				stp->sd_read_opt |= RFLUSHPCPROT;
2367 				break;
2368 			}
2369 		}
2370 		if (sop->so_flags & SO_ERROPT) {
2371 			switch (sop->so_erropt & RERRMASK) {
2372 			case RERRNORM:
2373 				stp->sd_flag &= ~STRDERRNONPERSIST;
2374 				break;
2375 			case RERRNONPERSIST:
2376 				stp->sd_flag |= STRDERRNONPERSIST;
2377 				break;
2378 			}
2379 			switch (sop->so_erropt & WERRMASK) {
2380 			case WERRNORM:
2381 				stp->sd_flag &= ~STWRERRNONPERSIST;
2382 				break;
2383 			case WERRNONPERSIST:
2384 				stp->sd_flag |= STWRERRNONPERSIST;
2385 				break;
2386 			}
2387 		}
2388 		if (sop->so_flags & SO_COPYOPT) {
2389 			if (sop->so_copyopt & ZCVMSAFE) {
2390 				stp->sd_copyflag |= STZCVMSAFE;
2391 				stp->sd_copyflag &= ~STZCVMUNSAFE;
2392 			} else if (sop->so_copyopt & ZCVMUNSAFE) {
2393 				stp->sd_copyflag |= STZCVMUNSAFE;
2394 				stp->sd_copyflag &= ~STZCVMSAFE;
2395 			}
2396 
2397 			if (sop->so_copyopt & COPYCACHED) {
2398 				stp->sd_copyflag |= STRCOPYCACHED;
2399 			}
2400 		}
2401 		if (sop->so_flags & SO_WROFF)
2402 			stp->sd_wroff = sop->so_wroff;
2403 		if (sop->so_flags & SO_TAIL)
2404 			stp->sd_tail = sop->so_tail;
2405 		if (sop->so_flags & SO_MINPSZ)
2406 			q->q_minpsz = sop->so_minpsz;
2407 		if (sop->so_flags & SO_MAXPSZ)
2408 			q->q_maxpsz = sop->so_maxpsz;
2409 		if (sop->so_flags & SO_MAXBLK)
2410 			stp->sd_maxblk = sop->so_maxblk;
2411 		if (sop->so_flags & SO_HIWAT) {
2412 			if (sop->so_flags & SO_BAND) {
2413 				if (strqset(q, QHIWAT,
2414 				    sop->so_band, sop->so_hiwat)) {
2415 					cmn_err(CE_WARN, "strrput: could not "
2416 					    "allocate qband\n");
2417 				} else {
2418 					bpri = sop->so_band;
2419 				}
2420 			} else {
2421 				q->q_hiwat = sop->so_hiwat;
2422 			}
2423 		}
2424 		if (sop->so_flags & SO_LOWAT) {
2425 			if (sop->so_flags & SO_BAND) {
2426 				if (strqset(q, QLOWAT,
2427 				    sop->so_band, sop->so_lowat)) {
2428 					cmn_err(CE_WARN, "strrput: could not "
2429 					    "allocate qband\n");
2430 				} else {
2431 					bpri = sop->so_band;
2432 				}
2433 			} else {
2434 				q->q_lowat = sop->so_lowat;
2435 			}
2436 		}
2437 		if (sop->so_flags & SO_MREADON)
2438 			stp->sd_flag |= SNDMREAD;
2439 		if (sop->so_flags & SO_MREADOFF)
2440 			stp->sd_flag &= ~SNDMREAD;
2441 		if (sop->so_flags & SO_NDELON)
2442 			stp->sd_flag |= OLDNDELAY;
2443 		if (sop->so_flags & SO_NDELOFF)
2444 			stp->sd_flag &= ~OLDNDELAY;
2445 		if (sop->so_flags & SO_ISTTY)
2446 			stp->sd_flag |= STRISTTY;
2447 		if (sop->so_flags & SO_ISNTTY)
2448 			stp->sd_flag &= ~STRISTTY;
2449 		if (sop->so_flags & SO_TOSTOP)
2450 			stp->sd_flag |= STRTOSTOP;
2451 		if (sop->so_flags & SO_TONSTOP)
2452 			stp->sd_flag &= ~STRTOSTOP;
2453 		if (sop->so_flags & SO_DELIM)
2454 			stp->sd_flag |= STRDELIM;
2455 		if (sop->so_flags & SO_NODELIM)
2456 			stp->sd_flag &= ~STRDELIM;
2457 
2458 		mutex_exit(&stp->sd_lock);
2459 		freemsg(bp);
2460 
2461 		/* Check backenable in case the water marks changed */
2462 		qbackenable(q, bpri);
2463 		return (0);
2464 
2465 	/*
2466 	 * The following set of cases deal with situations where two stream
2467 	 * heads are connected to each other (twisted streams).  These messages
2468 	 * have no meaning at the stream head.
2469 	 */
2470 	case M_BREAK:
2471 	case M_CTL:
2472 	case M_DELAY:
2473 	case M_START:
2474 	case M_STOP:
2475 	case M_IOCDATA:
2476 	case M_STARTI:
2477 	case M_STOPI:
2478 		freemsg(bp);
2479 		return (0);
2480 
2481 	case M_IOCTL:
2482 		/*
2483 		 * Always NAK this condition
2484 		 * (makes no sense)
2485 		 * If there is one or more threads in the read side
2486 		 * rwnext we have to defer the nacking until that thread
2487 		 * returns (in strget).
2488 		 */
2489 		mutex_enter(&stp->sd_lock);
2490 		if (stp->sd_struiodnak != 0) {
2491 			/*
2492 			 * Defer NAK to the streamhead. Queue at the end
2493 			 * the list.
2494 			 */
2495 			mblk_t *mp = stp->sd_struionak;
2496 
2497 			while (mp && mp->b_next)
2498 				mp = mp->b_next;
2499 			if (mp)
2500 				mp->b_next = bp;
2501 			else
2502 				stp->sd_struionak = bp;
2503 			bp->b_next = NULL;
2504 			mutex_exit(&stp->sd_lock);
2505 			return (0);
2506 		}
2507 		mutex_exit(&stp->sd_lock);
2508 
2509 		bp->b_datap->db_type = M_IOCNAK;
2510 		/*
2511 		 * Protect against the driver passing up
2512 		 * messages after it has done a qprocsoff.
2513 		 */
2514 		if (_OTHERQ(q)->q_next == NULL)
2515 			freemsg(bp);
2516 		else
2517 			qreply(q, bp);
2518 		return (0);
2519 
2520 	default:
2521 #ifdef DEBUG
2522 		cmn_err(CE_WARN,
2523 		    "bad message type %x received at stream head\n",
2524 		    bp->b_datap->db_type);
2525 #endif
2526 		freemsg(bp);
2527 		return (0);
2528 	}
2529 
2530 	/* NOTREACHED */
2531 }
2532 
2533 /*
2534  * Check if the stream pointed to by `stp' can be written to, and return an
2535  * error code if not.  If `eiohup' is set, then return EIO if STRHUP is set.
2536  * If `sigpipeok' is set and the SW_SIGPIPE option is enabled on the stream,
2537  * then always return EPIPE and send a SIGPIPE to the invoking thread.
2538  */
2539 static int
2540 strwriteable(struct stdata *stp, boolean_t eiohup, boolean_t sigpipeok)
2541 {
2542 	int error;
2543 
2544 	ASSERT(MUTEX_HELD(&stp->sd_lock));
2545 
2546 	/*
2547 	 * For modem support, POSIX states that on writes, EIO should
2548 	 * be returned if the stream has been hung up.
2549 	 */
2550 	if (eiohup && (stp->sd_flag & (STPLEX|STRHUP)) == STRHUP)
2551 		error = EIO;
2552 	else
2553 		error = strgeterr(stp, STRHUP|STPLEX|STWRERR, 0);
2554 
2555 	if (error != 0) {
2556 		if (!(stp->sd_flag & STPLEX) &&
2557 		    (stp->sd_wput_opt & SW_SIGPIPE) && sigpipeok) {
2558 			tsignal(curthread, SIGPIPE);
2559 			error = EPIPE;
2560 		}
2561 	}
2562 
2563 	return (error);
2564 }
2565 
2566 /*
2567  * Copyin and send data down a stream.
2568  * The caller will allocate and copyin any control part that precedes the
2569  * message and pass that in as mctl.
2570  *
2571  * Caller should *not* hold sd_lock.
2572  * When EWOULDBLOCK is returned the caller has to redo the canputnext
2573  * under sd_lock in order to avoid missing a backenabling wakeup.
2574  *
2575  * Use iosize = -1 to not send any M_DATA. iosize = 0 sends zero-length M_DATA.
2576  *
2577  * Set MSG_IGNFLOW in flags to ignore flow control for hipri messages.
2578  * For sync streams we can only ignore flow control by reverting to using
2579  * putnext.
2580  *
2581  * If sd_maxblk is less than *iosize this routine might return without
2582  * transferring all of *iosize. In all cases, on return *iosize will contain
2583  * the amount of data that was transferred.
2584  */
2585 static int
2586 strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
2587     int b_flag, int pri, int flags)
2588 {
2589 	struiod_t uiod;
2590 	mblk_t *mp;
2591 	queue_t *wqp = stp->sd_wrq;
2592 	int error = 0;
2593 	ssize_t count = *iosize;
2594 
2595 	ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
2596 
2597 	if (uiop != NULL && count >= 0)
2598 		flags |= stp->sd_struiowrq ? STRUIO_POSTPONE : 0;
2599 
2600 	if (!(flags & STRUIO_POSTPONE)) {
2601 		/*
2602 		 * Use regular canputnext, strmakedata, putnext sequence.
2603 		 */
2604 		if (pri == 0) {
2605 			if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2606 				freemsg(mctl);
2607 				return (EWOULDBLOCK);
2608 			}
2609 		} else {
2610 			if (!(flags & MSG_IGNFLOW) && !bcanputnext(wqp, pri)) {
2611 				freemsg(mctl);
2612 				return (EWOULDBLOCK);
2613 			}
2614 		}
2615 
2616 		if ((error = strmakedata(iosize, uiop, stp, flags,
2617 		    &mp)) != 0) {
2618 			freemsg(mctl);
2619 			/*
2620 			 * need to change return code to ENOMEM
2621 			 * so that this is not confused with
2622 			 * flow control, EAGAIN.
2623 			 */
2624 
2625 			if (error == EAGAIN)
2626 				return (ENOMEM);
2627 			else
2628 				return (error);
2629 		}
2630 		if (mctl != NULL) {
2631 			if (mctl->b_cont == NULL)
2632 				mctl->b_cont = mp;
2633 			else if (mp != NULL)
2634 				linkb(mctl, mp);
2635 			mp = mctl;
2636 		} else if (mp == NULL)
2637 			return (0);
2638 
2639 		mp->b_flag |= b_flag;
2640 		mp->b_band = (uchar_t)pri;
2641 
2642 		if (flags & MSG_IGNFLOW) {
2643 			/*
2644 			 * XXX Hack: Don't get stuck running service
2645 			 * procedures. This is needed for sockfs when
2646 			 * sending the unbind message out of the rput
2647 			 * procedure - we don't want a put procedure
2648 			 * to run service procedures.
2649 			 */
2650 			putnext(wqp, mp);
2651 		} else {
2652 			stream_willservice(stp);
2653 			putnext(wqp, mp);
2654 			stream_runservice(stp);
2655 		}
2656 		return (0);
2657 	}
2658 	/*
2659 	 * Stream supports rwnext() for the write side.
2660 	 */
2661 	if ((error = strmakedata(iosize, uiop, stp, flags, &mp)) != 0) {
2662 		freemsg(mctl);
2663 		/*
2664 		 * map EAGAIN to ENOMEM since EAGAIN means "flow controlled".
2665 		 */
2666 		return (error == EAGAIN ? ENOMEM : error);
2667 	}
2668 	if (mctl != NULL) {
2669 		if (mctl->b_cont == NULL)
2670 			mctl->b_cont = mp;
2671 		else if (mp != NULL)
2672 			linkb(mctl, mp);
2673 		mp = mctl;
2674 	} else if (mp == NULL) {
2675 		return (0);
2676 	}
2677 
2678 	mp->b_flag |= b_flag;
2679 	mp->b_band = (uchar_t)pri;
2680 
2681 	(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
2682 	    sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
2683 	uiod.d_uio.uio_offset = 0;
2684 	uiod.d_mp = mp;
2685 	error = rwnext(wqp, &uiod);
2686 	if (! uiod.d_mp) {
2687 		uioskip(uiop, *iosize);
2688 		return (error);
2689 	}
2690 	ASSERT(mp == uiod.d_mp);
2691 	if (error == EINVAL) {
2692 		/*
2693 		 * The stream plumbing must have changed while
2694 		 * we were away, so just turn off rwnext()s.
2695 		 */
2696 		error = 0;
2697 	} else if (error == EBUSY || error == EWOULDBLOCK) {
2698 		/*
2699 		 * Couldn't enter a perimeter or took a page fault,
2700 		 * so fall-back to putnext().
2701 		 */
2702 		error = 0;
2703 	} else {
2704 		freemsg(mp);
2705 		return (error);
2706 	}
2707 	/* Have to check canput before consuming data from the uio */
2708 	if (pri == 0) {
2709 		if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2710 			freemsg(mp);
2711 			return (EWOULDBLOCK);
2712 		}
2713 	} else {
2714 		if (!bcanputnext(wqp, pri) && !(flags & MSG_IGNFLOW)) {
2715 			freemsg(mp);
2716 			return (EWOULDBLOCK);
2717 		}
2718 	}
2719 	ASSERT(mp == uiod.d_mp);
2720 	/* Copyin data from the uio */
2721 	if ((error = struioget(wqp, mp, &uiod, 0)) != 0) {
2722 		freemsg(mp);
2723 		return (error);
2724 	}
2725 	uioskip(uiop, *iosize);
2726 	if (flags & MSG_IGNFLOW) {
2727 		/*
2728 		 * XXX Hack: Don't get stuck running service procedures.
2729 		 * This is needed for sockfs when sending the unbind message
2730 		 * out of the rput procedure - we don't want a put procedure
2731 		 * to run service procedures.
2732 		 */
2733 		putnext(wqp, mp);
2734 	} else {
2735 		stream_willservice(stp);
2736 		putnext(wqp, mp);
2737 		stream_runservice(stp);
2738 	}
2739 	return (0);
2740 }
2741 
2742 /*
2743  * Write attempts to break the write request into messages conforming
2744  * with the minimum and maximum packet sizes set downstream.
2745  *
2746  * Write will not block if downstream queue is full and
2747  * O_NDELAY is set, otherwise it will block waiting for the queue to get room.
2748  *
2749  * A write of zero bytes gets packaged into a zero length message and sent
2750  * downstream like any other message.
2751  *
2752  * If buffers of the requested sizes are not available, the write will
2753  * sleep until the buffers become available.
2754  *
2755  * Write (if specified) will supply a write offset in a message if it
2756  * makes sense. This can be specified by downstream modules as part of
2757  * a M_SETOPTS message.  Write will not supply the write offset if it
2758  * cannot supply any data in a buffer.  In other words, write will never
2759  * send down an empty packet due to a write offset.
2760  */
2761 /* ARGSUSED2 */
2762 int
2763 strwrite(struct vnode *vp, struct uio *uiop, cred_t *crp)
2764 {
2765 	return (strwrite_common(vp, uiop, crp, 0));
2766 }
2767 
2768 /* ARGSUSED2 */
2769 int
2770 strwrite_common(struct vnode *vp, struct uio *uiop, cred_t *crp, int wflag)
2771 {
2772 	struct stdata *stp;
2773 	struct queue *wqp;
2774 	ssize_t rmin, rmax;
2775 	ssize_t iosize;
2776 	int waitflag;
2777 	int tempmode;
2778 	int error = 0;
2779 	int b_flag;
2780 
2781 	ASSERT(vp->v_stream);
2782 	stp = vp->v_stream;
2783 
2784 	mutex_enter(&stp->sd_lock);
2785 
2786 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
2787 		mutex_exit(&stp->sd_lock);
2788 		return (error);
2789 	}
2790 
2791 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
2792 		error = strwriteable(stp, B_TRUE, B_TRUE);
2793 		if (error != 0) {
2794 			mutex_exit(&stp->sd_lock);
2795 			return (error);
2796 		}
2797 	}
2798 
2799 	mutex_exit(&stp->sd_lock);
2800 
2801 	wqp = stp->sd_wrq;
2802 
2803 	/* get these values from them cached in the stream head */
2804 	rmin = stp->sd_qn_minpsz;
2805 	rmax = stp->sd_qn_maxpsz;
2806 
2807 	/*
2808 	 * Check the min/max packet size constraints.  If min packet size
2809 	 * is non-zero, the write cannot be split into multiple messages
2810 	 * and still guarantee the size constraints.
2811 	 */
2812 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_IN, "strwrite in:q %p", wqp);
2813 
2814 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
2815 	if (rmax == 0) {
2816 		return (0);
2817 	}
2818 	if (rmin > 0) {
2819 		if (uiop->uio_resid < rmin) {
2820 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2821 			    "strwrite out:q %p out %d error %d",
2822 			    wqp, 0, ERANGE);
2823 			return (ERANGE);
2824 		}
2825 		if ((rmax != INFPSZ) && (uiop->uio_resid > rmax)) {
2826 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2827 			    "strwrite out:q %p out %d error %d",
2828 			    wqp, 1, ERANGE);
2829 			return (ERANGE);
2830 		}
2831 	}
2832 
2833 	/*
2834 	 * Do until count satisfied or error.
2835 	 */
2836 	waitflag = WRITEWAIT | wflag;
2837 	if (stp->sd_flag & OLDNDELAY)
2838 		tempmode = uiop->uio_fmode & ~FNDELAY;
2839 	else
2840 		tempmode = uiop->uio_fmode;
2841 
2842 	if (rmax == INFPSZ)
2843 		rmax = uiop->uio_resid;
2844 
2845 	/*
2846 	 * Note that tempmode does not get used in strput/strmakedata
2847 	 * but only in strwaitq. The other routines use uio_fmode
2848 	 * unmodified.
2849 	 */
2850 
2851 	/* LINTED: constant in conditional context */
2852 	while (1) {	/* breaks when uio_resid reaches zero */
2853 		/*
2854 		 * Determine the size of the next message to be
2855 		 * packaged.  May have to break write into several
2856 		 * messages based on max packet size.
2857 		 */
2858 		iosize = MIN(uiop->uio_resid, rmax);
2859 
2860 		/*
2861 		 * Put block downstream when flow control allows it.
2862 		 */
2863 		if ((stp->sd_flag & STRDELIM) && (uiop->uio_resid == iosize))
2864 			b_flag = MSGDELIM;
2865 		else
2866 			b_flag = 0;
2867 
2868 		for (;;) {
2869 			int done = 0;
2870 
2871 			error = strput(stp, NULL, uiop, &iosize, b_flag, 0, 0);
2872 			if (error == 0)
2873 				break;
2874 			if (error != EWOULDBLOCK)
2875 				goto out;
2876 
2877 			mutex_enter(&stp->sd_lock);
2878 			/*
2879 			 * Check for a missed wakeup.
2880 			 * Needed since strput did not hold sd_lock across
2881 			 * the canputnext.
2882 			 */
2883 			if (canputnext(wqp)) {
2884 				/* Try again */
2885 				mutex_exit(&stp->sd_lock);
2886 				continue;
2887 			}
2888 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAIT,
2889 			    "strwrite wait:q %p wait", wqp);
2890 			if ((error = strwaitq(stp, waitflag, (ssize_t)0,
2891 			    tempmode, -1, &done)) != 0 || done) {
2892 				mutex_exit(&stp->sd_lock);
2893 				if ((vp->v_type == VFIFO) &&
2894 				    (uiop->uio_fmode & FNDELAY) &&
2895 				    (error == EAGAIN))
2896 					error = 0;
2897 				goto out;
2898 			}
2899 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAKE,
2900 			    "strwrite wake:q %p awakes", wqp);
2901 			if ((error = i_straccess(stp, JCWRITE)) != 0) {
2902 				mutex_exit(&stp->sd_lock);
2903 				goto out;
2904 			}
2905 			mutex_exit(&stp->sd_lock);
2906 		}
2907 		waitflag |= NOINTR;
2908 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRWRITE_RESID,
2909 		    "strwrite resid:q %p uiop %p", wqp, uiop);
2910 		if (uiop->uio_resid) {
2911 			/* Recheck for errors - needed for sockets */
2912 			if ((stp->sd_wput_opt & SW_RECHECK_ERR) &&
2913 			    (stp->sd_flag & (STWRERR|STRHUP|STPLEX))) {
2914 				mutex_enter(&stp->sd_lock);
2915 				error = strwriteable(stp, B_FALSE, B_TRUE);
2916 				mutex_exit(&stp->sd_lock);
2917 				if (error != 0)
2918 					return (error);
2919 			}
2920 			continue;
2921 		}
2922 		break;
2923 	}
2924 out:
2925 	/*
2926 	 * For historical reasons, applications expect EAGAIN when a data
2927 	 * mblk_t cannot be allocated, so change ENOMEM back to EAGAIN.
2928 	 */
2929 	if (error == ENOMEM)
2930 		error = EAGAIN;
2931 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2932 	    "strwrite out:q %p out %d error %d", wqp, 2, error);
2933 	return (error);
2934 }
2935 
2936 /*
2937  * Stream head write service routine.
2938  * Its job is to wake up any sleeping writers when a queue
2939  * downstream needs data (part of the flow control in putq and getq).
2940  * It also must wake anyone sleeping on a poll().
2941  * For stream head right below mux module, it must also invoke put procedure
2942  * of next downstream module.
2943  */
2944 int
2945 strwsrv(queue_t *q)
2946 {
2947 	struct stdata *stp;
2948 	queue_t *tq;
2949 	qband_t *qbp;
2950 	int i;
2951 	qband_t *myqbp;
2952 	int isevent;
2953 	unsigned char	qbf[NBAND];	/* band flushing backenable flags */
2954 
2955 	TRACE_1(TR_FAC_STREAMS_FR,
2956 	    TR_STRWSRV, "strwsrv:q %p", q);
2957 	stp = (struct stdata *)q->q_ptr;
2958 	ASSERT(qclaimed(q));
2959 	mutex_enter(&stp->sd_lock);
2960 	ASSERT(!(stp->sd_flag & STPLEX));
2961 
2962 	if (stp->sd_flag & WSLEEP) {
2963 		stp->sd_flag &= ~WSLEEP;
2964 		cv_broadcast(&q->q_wait);
2965 	}
2966 	mutex_exit(&stp->sd_lock);
2967 
2968 	/* The other end of a stream pipe went away. */
2969 	if ((tq = q->q_next) == NULL) {
2970 		return (0);
2971 	}
2972 
2973 	/* Find the next module forward that has a service procedure */
2974 	claimstr(q);
2975 	tq = q->q_nfsrv;
2976 	ASSERT(tq != NULL);
2977 
2978 	if ((q->q_flag & QBACK)) {
2979 		if ((tq->q_flag & QFULL)) {
2980 			mutex_enter(QLOCK(tq));
2981 			if (!(tq->q_flag & QFULL)) {
2982 				mutex_exit(QLOCK(tq));
2983 				goto wakeup;
2984 			}
2985 			/*
2986 			 * The queue must have become full again. Set QWANTW
2987 			 * again so strwsrv will be back enabled when
2988 			 * the queue becomes non-full next time.
2989 			 */
2990 			tq->q_flag |= QWANTW;
2991 			mutex_exit(QLOCK(tq));
2992 		} else {
2993 		wakeup:
2994 			pollwakeup(&stp->sd_pollist, POLLWRNORM);
2995 			mutex_enter(&stp->sd_lock);
2996 			if (stp->sd_sigflags & S_WRNORM)
2997 				strsendsig(stp->sd_siglist, S_WRNORM, 0, 0);
2998 			mutex_exit(&stp->sd_lock);
2999 		}
3000 	}
3001 
3002 	isevent = 0;
3003 	i = 1;
3004 	bzero((caddr_t)qbf, NBAND);
3005 	mutex_enter(QLOCK(tq));
3006 	if ((myqbp = q->q_bandp) != NULL)
3007 		for (qbp = tq->q_bandp; qbp && myqbp; qbp = qbp->qb_next) {
3008 			ASSERT(myqbp);
3009 			if ((myqbp->qb_flag & QB_BACK)) {
3010 				if (qbp->qb_flag & QB_FULL) {
3011 					/*
3012 					 * The band must have become full again.
3013 					 * Set QB_WANTW again so strwsrv will
3014 					 * be back enabled when the band becomes
3015 					 * non-full next time.
3016 					 */
3017 					qbp->qb_flag |= QB_WANTW;
3018 				} else {
3019 					isevent = 1;
3020 					qbf[i] = 1;
3021 				}
3022 			}
3023 			myqbp = myqbp->qb_next;
3024 			i++;
3025 		}
3026 	mutex_exit(QLOCK(tq));
3027 
3028 	if (isevent) {
3029 		for (i = tq->q_nband; i; i--) {
3030 			if (qbf[i]) {
3031 				pollwakeup(&stp->sd_pollist, POLLWRBAND);
3032 				mutex_enter(&stp->sd_lock);
3033 				if (stp->sd_sigflags & S_WRBAND)
3034 					strsendsig(stp->sd_siglist, S_WRBAND,
3035 					    (uchar_t)i, 0);
3036 				mutex_exit(&stp->sd_lock);
3037 			}
3038 		}
3039 	}
3040 
3041 	releasestr(q);
3042 	return (0);
3043 }
3044 
3045 /*
3046  * Special case of strcopyin/strcopyout for copying
3047  * struct strioctl that can deal with both data
3048  * models.
3049  */
3050 
3051 #ifdef	_LP64
3052 
3053 static int
3054 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3055 {
3056 	struct	strioctl32 strioc32;
3057 	struct	strioctl *striocp;
3058 
3059 	if (copyflag & U_TO_K) {
3060 		ASSERT((copyflag & K_TO_K) == 0);
3061 
3062 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
3063 			if (copyin(from, &strioc32, sizeof (strioc32)))
3064 				return (EFAULT);
3065 
3066 			striocp = (struct strioctl *)to;
3067 			striocp->ic_cmd	= strioc32.ic_cmd;
3068 			striocp->ic_timout = strioc32.ic_timout;
3069 			striocp->ic_len	= strioc32.ic_len;
3070 			striocp->ic_dp	= (char *)(uintptr_t)strioc32.ic_dp;
3071 
3072 		} else { /* NATIVE data model */
3073 			if (copyin(from, to, sizeof (struct strioctl))) {
3074 				return (EFAULT);
3075 			} else {
3076 				return (0);
3077 			}
3078 		}
3079 	} else {
3080 		ASSERT(copyflag & K_TO_K);
3081 		bcopy(from, to, sizeof (struct strioctl));
3082 	}
3083 	return (0);
3084 }
3085 
3086 static int
3087 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3088 {
3089 	struct	strioctl32 strioc32;
3090 	struct	strioctl *striocp;
3091 
3092 	if (copyflag & U_TO_K) {
3093 		ASSERT((copyflag & K_TO_K) == 0);
3094 
3095 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
3096 			striocp = (struct strioctl *)from;
3097 			strioc32.ic_cmd	= striocp->ic_cmd;
3098 			strioc32.ic_timout = striocp->ic_timout;
3099 			strioc32.ic_len	= striocp->ic_len;
3100 			strioc32.ic_dp	= (caddr32_t)(uintptr_t)striocp->ic_dp;
3101 			ASSERT((char *)(uintptr_t)strioc32.ic_dp ==
3102 			    striocp->ic_dp);
3103 
3104 			if (copyout(&strioc32, to, sizeof (strioc32)))
3105 				return (EFAULT);
3106 
3107 		} else { /* NATIVE data model */
3108 			if (copyout(from, to, sizeof (struct strioctl))) {
3109 				return (EFAULT);
3110 			} else {
3111 				return (0);
3112 			}
3113 		}
3114 	} else {
3115 		ASSERT(copyflag & K_TO_K);
3116 		bcopy(from, to, sizeof (struct strioctl));
3117 	}
3118 	return (0);
3119 }
3120 
3121 #else	/* ! _LP64 */
3122 
3123 /* ARGSUSED2 */
3124 static int
3125 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3126 {
3127 	return (strcopyin(from, to, sizeof (struct strioctl), copyflag));
3128 }
3129 
3130 /* ARGSUSED2 */
3131 static int
3132 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3133 {
3134 	return (strcopyout(from, to, sizeof (struct strioctl), copyflag));
3135 }
3136 
3137 #endif	/* _LP64 */
3138 
3139 /*
3140  * Determine type of job control semantics expected by user.  The
3141  * possibilities are:
3142  *	JCREAD	- Behaves like read() on fd; send SIGTTIN
3143  *	JCWRITE	- Behaves like write() on fd; send SIGTTOU if TOSTOP set
3144  *	JCSETP	- Sets a value in the stream; send SIGTTOU, ignore TOSTOP
3145  *	JCGETP	- Gets a value in the stream; no signals.
3146  * See straccess in strsubr.c for usage of these values.
3147  *
3148  * This routine also returns -1 for I_STR as a special case; the
3149  * caller must call again with the real ioctl number for
3150  * classification.
3151  */
3152 static int
3153 job_control_type(int cmd)
3154 {
3155 	switch (cmd) {
3156 	case I_STR:
3157 		return (-1);
3158 
3159 	case I_RECVFD:
3160 	case I_E_RECVFD:
3161 		return (JCREAD);
3162 
3163 	case I_FDINSERT:
3164 	case I_SENDFD:
3165 		return (JCWRITE);
3166 
3167 	case TCSETA:
3168 	case TCSETAW:
3169 	case TCSETAF:
3170 	case TCSBRK:
3171 	case TCXONC:
3172 	case TCFLSH:
3173 	case TCDSET:	/* Obsolete */
3174 	case TIOCSWINSZ:
3175 	case TCSETS:
3176 	case TCSETSW:
3177 	case TCSETSF:
3178 	case TIOCSETD:
3179 	case TIOCHPCL:
3180 	case TIOCSETP:
3181 	case TIOCSETN:
3182 	case TIOCEXCL:
3183 	case TIOCNXCL:
3184 	case TIOCFLUSH:
3185 	case TIOCSETC:
3186 	case TIOCLBIS:
3187 	case TIOCLBIC:
3188 	case TIOCLSET:
3189 	case TIOCSBRK:
3190 	case TIOCCBRK:
3191 	case TIOCSDTR:
3192 	case TIOCCDTR:
3193 	case TIOCSLTC:
3194 	case TIOCSTOP:
3195 	case TIOCSTART:
3196 	case TIOCSTI:
3197 	case TIOCSPGRP:
3198 	case TIOCMSET:
3199 	case TIOCMBIS:
3200 	case TIOCMBIC:
3201 	case TIOCREMOTE:
3202 	case TIOCSIGNAL:
3203 	case LDSETT:
3204 	case LDSMAP:	/* Obsolete */
3205 	case DIOCSETP:
3206 	case I_FLUSH:
3207 	case I_SRDOPT:
3208 	case I_SETSIG:
3209 	case I_SWROPT:
3210 	case I_FLUSHBAND:
3211 	case I_SETCLTIME:
3212 	case I_SERROPT:
3213 	case I_ESETSIG:
3214 	case FIONBIO:
3215 	case FIOASYNC:
3216 	case FIOSETOWN:
3217 	case JBOOT:	/* Obsolete */
3218 	case JTERM:	/* Obsolete */
3219 	case JTIMOM:	/* Obsolete */
3220 	case JZOMBOOT:	/* Obsolete */
3221 	case JAGENT:	/* Obsolete */
3222 	case JTRUN:	/* Obsolete */
3223 	case JXTPROTO:	/* Obsolete */
3224 	case TIOCSETLD:
3225 		return (JCSETP);
3226 	}
3227 
3228 	return (JCGETP);
3229 }
3230 
3231 /*
3232  * ioctl for streams
3233  */
3234 int
3235 strioctl(struct vnode *vp, int cmd, intptr_t arg, int flag, int copyflag,
3236     cred_t *crp, int *rvalp)
3237 {
3238 	struct stdata *stp;
3239 	struct strcmd *scp;
3240 	struct strioctl strioc;
3241 	struct uio uio;
3242 	struct iovec iov;
3243 	int access;
3244 	mblk_t *mp;
3245 	int error = 0;
3246 	int done = 0;
3247 	ssize_t	rmin, rmax;
3248 	queue_t *wrq;
3249 	queue_t *rdq;
3250 	boolean_t kioctl = B_FALSE;
3251 
3252 	if (flag & FKIOCTL) {
3253 		copyflag = K_TO_K;
3254 		kioctl = B_TRUE;
3255 	}
3256 	ASSERT(vp->v_stream);
3257 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
3258 	stp = vp->v_stream;
3259 
3260 	TRACE_3(TR_FAC_STREAMS_FR, TR_IOCTL_ENTER,
3261 	    "strioctl:stp %p cmd %X arg %lX", stp, cmd, arg);
3262 
3263 	if (audit_active)
3264 		audit_strioctl(vp, cmd, arg, flag, copyflag, crp, rvalp);
3265 
3266 	/*
3267 	 * If the copy is kernel to kernel, make sure that the FNATIVE
3268 	 * flag is set.  After this it would be a serious error to have
3269 	 * no model flag.
3270 	 */
3271 	if (copyflag == K_TO_K)
3272 		flag = (flag & ~FMODELS) | FNATIVE;
3273 
3274 	ASSERT((flag & FMODELS) != 0);
3275 
3276 	wrq = stp->sd_wrq;
3277 	rdq = _RD(wrq);
3278 
3279 	access = job_control_type(cmd);
3280 
3281 	/* We should never see these here, should be handled by iwscn */
3282 	if (cmd == SRIOCSREDIR || cmd == SRIOCISREDIR)
3283 		return (EINVAL);
3284 
3285 	mutex_enter(&stp->sd_lock);
3286 	if ((access != -1) && ((error = i_straccess(stp, access)) != 0)) {
3287 		mutex_exit(&stp->sd_lock);
3288 		return (error);
3289 	}
3290 	mutex_exit(&stp->sd_lock);
3291 
3292 	/*
3293 	 * Check for sgttyb-related ioctls first, and complain as
3294 	 * necessary.
3295 	 */
3296 	switch (cmd) {
3297 	case TIOCGETP:
3298 	case TIOCSETP:
3299 	case TIOCSETN:
3300 		if (sgttyb_handling >= 2 && !sgttyb_complaint) {
3301 			sgttyb_complaint = B_TRUE;
3302 			cmn_err(CE_NOTE,
3303 			    "application used obsolete TIOC[GS]ET");
3304 		}
3305 		if (sgttyb_handling >= 3) {
3306 			tsignal(curthread, SIGSYS);
3307 			return (EIO);
3308 		}
3309 		break;
3310 	}
3311 
3312 	mutex_enter(&stp->sd_lock);
3313 
3314 	switch (cmd) {
3315 	case I_RECVFD:
3316 	case I_E_RECVFD:
3317 	case I_PEEK:
3318 	case I_NREAD:
3319 	case FIONREAD:
3320 	case FIORDCHK:
3321 	case I_ATMARK:
3322 	case FIONBIO:
3323 	case FIOASYNC:
3324 		if (stp->sd_flag & (STRDERR|STPLEX)) {
3325 			error = strgeterr(stp, STRDERR|STPLEX, 0);
3326 			if (error != 0) {
3327 				mutex_exit(&stp->sd_lock);
3328 				return (error);
3329 			}
3330 		}
3331 		break;
3332 
3333 	default:
3334 		if (stp->sd_flag & (STRDERR|STWRERR|STPLEX)) {
3335 			error = strgeterr(stp, STRDERR|STWRERR|STPLEX, 0);
3336 			if (error != 0) {
3337 				mutex_exit(&stp->sd_lock);
3338 				return (error);
3339 			}
3340 		}
3341 	}
3342 
3343 	mutex_exit(&stp->sd_lock);
3344 
3345 	switch (cmd) {
3346 	default:
3347 		/*
3348 		 * The stream head has hardcoded knowledge of a
3349 		 * miscellaneous collection of terminal-, keyboard- and
3350 		 * mouse-related ioctls, enumerated below.  This hardcoded
3351 		 * knowledge allows the stream head to automatically
3352 		 * convert transparent ioctl requests made by userland
3353 		 * programs into I_STR ioctls which many old STREAMS
3354 		 * modules and drivers require.
3355 		 *
3356 		 * No new ioctls should ever be added to this list.
3357 		 * Instead, the STREAMS module or driver should be written
3358 		 * to either handle transparent ioctls or require any
3359 		 * userland programs to use I_STR ioctls (by returning
3360 		 * EINVAL to any transparent ioctl requests).
3361 		 *
3362 		 * More importantly, removing ioctls from this list should
3363 		 * be done with the utmost care, since our STREAMS modules
3364 		 * and drivers *count* on the stream head performing this
3365 		 * conversion, and thus may panic while processing
3366 		 * transparent ioctl request for one of these ioctls (keep
3367 		 * in mind that third party modules and drivers may have
3368 		 * similar problems).
3369 		 */
3370 		if (((cmd & IOCTYPE) == LDIOC) ||
3371 		    ((cmd & IOCTYPE) == tIOC) ||
3372 		    ((cmd & IOCTYPE) == TIOC) ||
3373 		    ((cmd & IOCTYPE) == KIOC) ||
3374 		    ((cmd & IOCTYPE) == MSIOC) ||
3375 		    ((cmd & IOCTYPE) == VUIOC)) {
3376 			/*
3377 			 * The ioctl is a tty ioctl - set up strioc buffer
3378 			 * and call strdoioctl() to do the work.
3379 			 */
3380 			if (stp->sd_flag & STRHUP)
3381 				return (ENXIO);
3382 			strioc.ic_cmd = cmd;
3383 			strioc.ic_timout = INFTIM;
3384 
3385 			switch (cmd) {
3386 
3387 			case TCXONC:
3388 			case TCSBRK:
3389 			case TCFLSH:
3390 			case TCDSET:
3391 				{
3392 				int native_arg = (int)arg;
3393 				strioc.ic_len = sizeof (int);
3394 				strioc.ic_dp = (char *)&native_arg;
3395 				return (strdoioctl(stp, &strioc, flag,
3396 				    K_TO_K, crp, rvalp));
3397 				}
3398 
3399 			case TCSETA:
3400 			case TCSETAW:
3401 			case TCSETAF:
3402 				strioc.ic_len = sizeof (struct termio);
3403 				strioc.ic_dp = (char *)arg;
3404 				return (strdoioctl(stp, &strioc, flag,
3405 				    copyflag, crp, rvalp));
3406 
3407 			case TCSETS:
3408 			case TCSETSW:
3409 			case TCSETSF:
3410 				strioc.ic_len = sizeof (struct termios);
3411 				strioc.ic_dp = (char *)arg;
3412 				return (strdoioctl(stp, &strioc, flag,
3413 				    copyflag, crp, rvalp));
3414 
3415 			case LDSETT:
3416 				strioc.ic_len = sizeof (struct termcb);
3417 				strioc.ic_dp = (char *)arg;
3418 				return (strdoioctl(stp, &strioc, flag,
3419 				    copyflag, crp, rvalp));
3420 
3421 			case TIOCSETP:
3422 				strioc.ic_len = sizeof (struct sgttyb);
3423 				strioc.ic_dp = (char *)arg;
3424 				return (strdoioctl(stp, &strioc, flag,
3425 				    copyflag, crp, rvalp));
3426 
3427 			case TIOCSTI:
3428 				if ((flag & FREAD) == 0 &&
3429 				    secpolicy_sti(crp) != 0) {
3430 					return (EPERM);
3431 				}
3432 				mutex_enter(&stp->sd_lock);
3433 				mutex_enter(&curproc->p_splock);
3434 				if (stp->sd_sidp != curproc->p_sessp->s_sidp &&
3435 				    secpolicy_sti(crp) != 0) {
3436 					mutex_exit(&curproc->p_splock);
3437 					mutex_exit(&stp->sd_lock);
3438 					return (EACCES);
3439 				}
3440 				mutex_exit(&curproc->p_splock);
3441 				mutex_exit(&stp->sd_lock);
3442 
3443 				strioc.ic_len = sizeof (char);
3444 				strioc.ic_dp = (char *)arg;
3445 				return (strdoioctl(stp, &strioc, flag,
3446 				    copyflag, crp, rvalp));
3447 
3448 			case TIOCSWINSZ:
3449 				strioc.ic_len = sizeof (struct winsize);
3450 				strioc.ic_dp = (char *)arg;
3451 				return (strdoioctl(stp, &strioc, flag,
3452 				    copyflag, crp, rvalp));
3453 
3454 			case TIOCSSIZE:
3455 				strioc.ic_len = sizeof (struct ttysize);
3456 				strioc.ic_dp = (char *)arg;
3457 				return (strdoioctl(stp, &strioc, flag,
3458 				    copyflag, crp, rvalp));
3459 
3460 			case TIOCSSOFTCAR:
3461 			case KIOCTRANS:
3462 			case KIOCTRANSABLE:
3463 			case KIOCCMD:
3464 			case KIOCSDIRECT:
3465 			case KIOCSCOMPAT:
3466 			case KIOCSKABORTEN:
3467 			case KIOCSRPTDELAY:
3468 			case KIOCSRPTRATE:
3469 			case VUIDSFORMAT:
3470 			case TIOCSPPS:
3471 				strioc.ic_len = sizeof (int);
3472 				strioc.ic_dp = (char *)arg;
3473 				return (strdoioctl(stp, &strioc, flag,
3474 				    copyflag, crp, rvalp));
3475 
3476 			case KIOCSETKEY:
3477 			case KIOCGETKEY:
3478 				strioc.ic_len = sizeof (struct kiockey);
3479 				strioc.ic_dp = (char *)arg;
3480 				return (strdoioctl(stp, &strioc, flag,
3481 				    copyflag, crp, rvalp));
3482 
3483 			case KIOCSKEY:
3484 			case KIOCGKEY:
3485 				strioc.ic_len = sizeof (struct kiockeymap);
3486 				strioc.ic_dp = (char *)arg;
3487 				return (strdoioctl(stp, &strioc, flag,
3488 				    copyflag, crp, rvalp));
3489 
3490 			case KIOCSLED:
3491 				/* arg is a pointer to char */
3492 				strioc.ic_len = sizeof (char);
3493 				strioc.ic_dp = (char *)arg;
3494 				return (strdoioctl(stp, &strioc, flag,
3495 				    copyflag, crp, rvalp));
3496 
3497 			case MSIOSETPARMS:
3498 				strioc.ic_len = sizeof (Ms_parms);
3499 				strioc.ic_dp = (char *)arg;
3500 				return (strdoioctl(stp, &strioc, flag,
3501 				    copyflag, crp, rvalp));
3502 
3503 			case VUIDSADDR:
3504 			case VUIDGADDR:
3505 				strioc.ic_len = sizeof (struct vuid_addr_probe);
3506 				strioc.ic_dp = (char *)arg;
3507 				return (strdoioctl(stp, &strioc, flag,
3508 				    copyflag, crp, rvalp));
3509 
3510 			/*
3511 			 * These M_IOCTL's don't require any data to be sent
3512 			 * downstream, and the driver will allocate and link
3513 			 * on its own mblk_t upon M_IOCACK -- thus we set
3514 			 * ic_len to zero and set ic_dp to arg so we know
3515 			 * where to copyout to later.
3516 			 */
3517 			case TIOCGSOFTCAR:
3518 			case TIOCGWINSZ:
3519 			case TIOCGSIZE:
3520 			case KIOCGTRANS:
3521 			case KIOCGTRANSABLE:
3522 			case KIOCTYPE:
3523 			case KIOCGDIRECT:
3524 			case KIOCGCOMPAT:
3525 			case KIOCLAYOUT:
3526 			case KIOCGLED:
3527 			case MSIOGETPARMS:
3528 			case MSIOBUTTONS:
3529 			case VUIDGFORMAT:
3530 			case TIOCGPPS:
3531 			case TIOCGPPSEV:
3532 			case TCGETA:
3533 			case TCGETS:
3534 			case LDGETT:
3535 			case TIOCGETP:
3536 			case KIOCGRPTDELAY:
3537 			case KIOCGRPTRATE:
3538 				strioc.ic_len = 0;
3539 				strioc.ic_dp = (char *)arg;
3540 				return (strdoioctl(stp, &strioc, flag,
3541 				    copyflag, crp, rvalp));
3542 			}
3543 		}
3544 
3545 		/*
3546 		 * Unknown cmd - send it down as a transparent ioctl.
3547 		 */
3548 		strioc.ic_cmd = cmd;
3549 		strioc.ic_timout = INFTIM;
3550 		strioc.ic_len = TRANSPARENT;
3551 		strioc.ic_dp = (char *)&arg;
3552 
3553 		return (strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp));
3554 
3555 	case I_STR:
3556 		/*
3557 		 * Stream ioctl.  Read in an strioctl buffer from the user
3558 		 * along with any data specified and send it downstream.
3559 		 * Strdoioctl will wait allow only one ioctl message at
3560 		 * a time, and waits for the acknowledgement.
3561 		 */
3562 
3563 		if (stp->sd_flag & STRHUP)
3564 			return (ENXIO);
3565 
3566 		error = strcopyin_strioctl((void *)arg, &strioc, flag,
3567 		    copyflag);
3568 		if (error != 0)
3569 			return (error);
3570 
3571 		if ((strioc.ic_len < 0) || (strioc.ic_timout < -1))
3572 			return (EINVAL);
3573 
3574 		access = job_control_type(strioc.ic_cmd);
3575 		mutex_enter(&stp->sd_lock);
3576 		if ((access != -1) &&
3577 		    ((error = i_straccess(stp, access)) != 0)) {
3578 			mutex_exit(&stp->sd_lock);
3579 			return (error);
3580 		}
3581 		mutex_exit(&stp->sd_lock);
3582 
3583 		/*
3584 		 * The I_STR facility provides a trap door for malicious
3585 		 * code to send down bogus streamio(7I) ioctl commands to
3586 		 * unsuspecting STREAMS modules and drivers which expect to
3587 		 * only get these messages from the stream head.
3588 		 * Explicitly prohibit any streamio ioctls which can be
3589 		 * passed downstream by the stream head.  Note that we do
3590 		 * not block all streamio ioctls because the ioctl
3591 		 * numberspace is not well managed and thus it's possible
3592 		 * that a module or driver's ioctl numbers may accidentally
3593 		 * collide with them.
3594 		 */
3595 		switch (strioc.ic_cmd) {
3596 		case I_LINK:
3597 		case I_PLINK:
3598 		case I_UNLINK:
3599 		case I_PUNLINK:
3600 		case _I_GETPEERCRED:
3601 		case _I_PLINK_LH:
3602 			return (EINVAL);
3603 		}
3604 
3605 		error = strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp);
3606 		if (error == 0) {
3607 			error = strcopyout_strioctl(&strioc, (void *)arg,
3608 			    flag, copyflag);
3609 		}
3610 		return (error);
3611 
3612 	case _I_CMD:
3613 		/*
3614 		 * Like I_STR, but without using M_IOC* messages and without
3615 		 * copyins/copyouts beyond the passed-in argument.
3616 		 */
3617 		if (stp->sd_flag & STRHUP)
3618 			return (ENXIO);
3619 
3620 		if ((scp = kmem_alloc(sizeof (strcmd_t), KM_NOSLEEP)) == NULL)
3621 			return (ENOMEM);
3622 
3623 		if (copyin((void *)arg, scp, sizeof (strcmd_t))) {
3624 			kmem_free(scp, sizeof (strcmd_t));
3625 			return (EFAULT);
3626 		}
3627 
3628 		access = job_control_type(scp->sc_cmd);
3629 		mutex_enter(&stp->sd_lock);
3630 		if (access != -1 && (error = i_straccess(stp, access)) != 0) {
3631 			mutex_exit(&stp->sd_lock);
3632 			kmem_free(scp, sizeof (strcmd_t));
3633 			return (error);
3634 		}
3635 		mutex_exit(&stp->sd_lock);
3636 
3637 		*rvalp = 0;
3638 		if ((error = strdocmd(stp, scp, crp)) == 0) {
3639 			if (copyout(scp, (void *)arg, sizeof (strcmd_t)))
3640 				error = EFAULT;
3641 		}
3642 		kmem_free(scp, sizeof (strcmd_t));
3643 		return (error);
3644 
3645 	case I_NREAD:
3646 		/*
3647 		 * Return number of bytes of data in first message
3648 		 * in queue in "arg" and return the number of messages
3649 		 * in queue in return value.
3650 		 */
3651 	{
3652 		size_t	size;
3653 		int	retval;
3654 		int	count = 0;
3655 
3656 		mutex_enter(QLOCK(rdq));
3657 
3658 		size = msgdsize(rdq->q_first);
3659 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3660 			count++;
3661 
3662 		mutex_exit(QLOCK(rdq));
3663 		if (stp->sd_struiordq) {
3664 			infod_t infod;
3665 
3666 			infod.d_cmd = INFOD_COUNT;
3667 			infod.d_count = 0;
3668 			if (count == 0) {
3669 				infod.d_cmd |= INFOD_FIRSTBYTES;
3670 				infod.d_bytes = 0;
3671 			}
3672 			infod.d_res = 0;
3673 			(void) infonext(rdq, &infod);
3674 			count += infod.d_count;
3675 			if (infod.d_res & INFOD_FIRSTBYTES)
3676 				size = infod.d_bytes;
3677 		}
3678 
3679 		/*
3680 		 * Drop down from size_t to the "int" required by the
3681 		 * interface.  Cap at INT_MAX.
3682 		 */
3683 		retval = MIN(size, INT_MAX);
3684 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3685 		    copyflag);
3686 		if (!error)
3687 			*rvalp = count;
3688 		return (error);
3689 	}
3690 
3691 	case FIONREAD:
3692 		/*
3693 		 * Return number of bytes of data in all data messages
3694 		 * in queue in "arg".
3695 		 */
3696 	{
3697 		size_t	size = 0;
3698 		int	retval;
3699 
3700 		mutex_enter(QLOCK(rdq));
3701 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3702 			size += msgdsize(mp);
3703 		mutex_exit(QLOCK(rdq));
3704 
3705 		if (stp->sd_struiordq) {
3706 			infod_t infod;
3707 
3708 			infod.d_cmd = INFOD_BYTES;
3709 			infod.d_res = 0;
3710 			infod.d_bytes = 0;
3711 			(void) infonext(rdq, &infod);
3712 			size += infod.d_bytes;
3713 		}
3714 
3715 		/*
3716 		 * Drop down from size_t to the "int" required by the
3717 		 * interface.  Cap at INT_MAX.
3718 		 */
3719 		retval = MIN(size, INT_MAX);
3720 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3721 		    copyflag);
3722 
3723 		*rvalp = 0;
3724 		return (error);
3725 	}
3726 	case FIORDCHK:
3727 		/*
3728 		 * FIORDCHK does not use arg value (like FIONREAD),
3729 		 * instead a count is returned. I_NREAD value may
3730 		 * not be accurate but safe. The real thing to do is
3731 		 * to add the msgdsizes of all data  messages until
3732 		 * a non-data message.
3733 		 */
3734 	{
3735 		size_t size = 0;
3736 
3737 		mutex_enter(QLOCK(rdq));
3738 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3739 			size += msgdsize(mp);
3740 		mutex_exit(QLOCK(rdq));
3741 
3742 		if (stp->sd_struiordq) {
3743 			infod_t infod;
3744 
3745 			infod.d_cmd = INFOD_BYTES;
3746 			infod.d_res = 0;
3747 			infod.d_bytes = 0;
3748 			(void) infonext(rdq, &infod);
3749 			size += infod.d_bytes;
3750 		}
3751 
3752 		/*
3753 		 * Since ioctl returns an int, and memory sizes under
3754 		 * LP64 may not fit, we return INT_MAX if the count was
3755 		 * actually greater.
3756 		 */
3757 		*rvalp = MIN(size, INT_MAX);
3758 		return (0);
3759 	}
3760 
3761 	case I_FIND:
3762 		/*
3763 		 * Get module name.
3764 		 */
3765 	{
3766 		char mname[FMNAMESZ + 1];
3767 		queue_t *q;
3768 
3769 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3770 		    mname, FMNAMESZ + 1, NULL);
3771 		if (error)
3772 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3773 
3774 		/*
3775 		 * Return EINVAL if we're handed a bogus module name.
3776 		 */
3777 		if (fmodsw_find(mname, FMODSW_LOAD) == NULL) {
3778 			TRACE_0(TR_FAC_STREAMS_FR,
3779 			    TR_I_CANT_FIND, "couldn't I_FIND");
3780 			return (EINVAL);
3781 		}
3782 
3783 		*rvalp = 0;
3784 
3785 		/* Look downstream to see if module is there. */
3786 		claimstr(stp->sd_wrq);
3787 		for (q = stp->sd_wrq->q_next; q; q = q->q_next) {
3788 			if (q->q_flag & QREADR) {
3789 				q = NULL;
3790 				break;
3791 			}
3792 			if (strcmp(mname, Q2NAME(q)) == 0)
3793 				break;
3794 		}
3795 		releasestr(stp->sd_wrq);
3796 
3797 		*rvalp = (q ? 1 : 0);
3798 		return (error);
3799 	}
3800 
3801 	case I_PUSH:
3802 	case __I_PUSH_NOCTTY:
3803 		/*
3804 		 * Push a module.
3805 		 * For the case __I_PUSH_NOCTTY push a module but
3806 		 * do not allocate controlling tty. See bugid 4025044
3807 		 */
3808 
3809 	{
3810 		char mname[FMNAMESZ + 1];
3811 		fmodsw_impl_t *fp;
3812 		dev_t dummydev;
3813 
3814 		if (stp->sd_flag & STRHUP)
3815 			return (ENXIO);
3816 
3817 		/*
3818 		 * Get module name and look up in fmodsw.
3819 		 */
3820 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3821 		    mname, FMNAMESZ + 1, NULL);
3822 		if (error)
3823 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3824 
3825 		if ((fp = fmodsw_find(mname, FMODSW_HOLD | FMODSW_LOAD)) ==
3826 		    NULL)
3827 			return (EINVAL);
3828 
3829 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_PUSH,
3830 		    "I_PUSH:fp %p stp %p", fp, stp);
3831 
3832 		if (error = strstartplumb(stp, flag, cmd)) {
3833 			fmodsw_rele(fp);
3834 			return (error);
3835 		}
3836 
3837 		/*
3838 		 * See if any more modules can be pushed on this stream.
3839 		 * Note that this check must be done after strstartplumb()
3840 		 * since otherwise multiple threads issuing I_PUSHes on
3841 		 * the same stream will be able to exceed nstrpush.
3842 		 */
3843 		mutex_enter(&stp->sd_lock);
3844 		if (stp->sd_pushcnt >= nstrpush) {
3845 			fmodsw_rele(fp);
3846 			strendplumb(stp);
3847 			mutex_exit(&stp->sd_lock);
3848 			return (EINVAL);
3849 		}
3850 		mutex_exit(&stp->sd_lock);
3851 
3852 		/*
3853 		 * Push new module and call its open routine
3854 		 * via qattach().  Modules don't change device
3855 		 * numbers, so just ignore dummydev here.
3856 		 */
3857 		dummydev = vp->v_rdev;
3858 		if ((error = qattach(rdq, &dummydev, 0, crp, fp,
3859 		    B_FALSE)) == 0) {
3860 			if (vp->v_type == VCHR && /* sorry, no pipes allowed */
3861 			    (cmd == I_PUSH) && (stp->sd_flag & STRISTTY)) {
3862 				/*
3863 				 * try to allocate it as a controlling terminal
3864 				 */
3865 				(void) strctty(stp);
3866 			}
3867 		}
3868 
3869 		mutex_enter(&stp->sd_lock);
3870 
3871 		/*
3872 		 * As a performance concern we are caching the values of
3873 		 * q_minpsz and q_maxpsz of the module below the stream
3874 		 * head in the stream head.
3875 		 */
3876 		mutex_enter(QLOCK(stp->sd_wrq->q_next));
3877 		rmin = stp->sd_wrq->q_next->q_minpsz;
3878 		rmax = stp->sd_wrq->q_next->q_maxpsz;
3879 		mutex_exit(QLOCK(stp->sd_wrq->q_next));
3880 
3881 		/* Do this processing here as a performance concern */
3882 		if (strmsgsz != 0) {
3883 			if (rmax == INFPSZ)
3884 				rmax = strmsgsz;
3885 			else  {
3886 				if (vp->v_type == VFIFO)
3887 					rmax = MIN(PIPE_BUF, rmax);
3888 				else	rmax = MIN(strmsgsz, rmax);
3889 			}
3890 		}
3891 
3892 		mutex_enter(QLOCK(wrq));
3893 		stp->sd_qn_minpsz = rmin;
3894 		stp->sd_qn_maxpsz = rmax;
3895 		mutex_exit(QLOCK(wrq));
3896 
3897 		strendplumb(stp);
3898 		mutex_exit(&stp->sd_lock);
3899 		return (error);
3900 	}
3901 
3902 	case I_POP:
3903 	{
3904 		queue_t	*q;
3905 
3906 		if (stp->sd_flag & STRHUP)
3907 			return (ENXIO);
3908 		if (!wrq->q_next)	/* for broken pipes */
3909 			return (EINVAL);
3910 
3911 		if (error = strstartplumb(stp, flag, cmd))
3912 			return (error);
3913 
3914 		/*
3915 		 * If there is an anchor on this stream and popping
3916 		 * the current module would attempt to pop through the
3917 		 * anchor, then disallow the pop unless we have sufficient
3918 		 * privileges; take the cheapest (non-locking) check
3919 		 * first.
3920 		 */
3921 		if (secpolicy_ip_config(crp, B_TRUE) != 0 ||
3922 		    (stp->sd_anchorzone != crgetzoneid(crp))) {
3923 			mutex_enter(&stp->sd_lock);
3924 			/*
3925 			 * Anchors only apply if there's at least one
3926 			 * module on the stream (sd_pushcnt > 0).
3927 			 */
3928 			if (stp->sd_pushcnt > 0 &&
3929 			    stp->sd_pushcnt == stp->sd_anchor &&
3930 			    stp->sd_vnode->v_type != VFIFO) {
3931 				strendplumb(stp);
3932 				mutex_exit(&stp->sd_lock);
3933 				if (stp->sd_anchorzone != crgetzoneid(crp))
3934 					return (EINVAL);
3935 				/* Audit and report error */
3936 				return (secpolicy_ip_config(crp, B_FALSE));
3937 			}
3938 			mutex_exit(&stp->sd_lock);
3939 		}
3940 
3941 		q = wrq->q_next;
3942 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_POP,
3943 		    "I_POP:%p from %p", q, stp);
3944 		if (q->q_next == NULL || (q->q_flag & (QREADR|QISDRV))) {
3945 			error = EINVAL;
3946 		} else {
3947 			qdetach(_RD(q), 1, flag, crp, B_FALSE);
3948 			error = 0;
3949 		}
3950 		mutex_enter(&stp->sd_lock);
3951 
3952 		/*
3953 		 * As a performance concern we are caching the values of
3954 		 * q_minpsz and q_maxpsz of the module below the stream
3955 		 * head in the stream head.
3956 		 */
3957 		mutex_enter(QLOCK(wrq->q_next));
3958 		rmin = wrq->q_next->q_minpsz;
3959 		rmax = wrq->q_next->q_maxpsz;
3960 		mutex_exit(QLOCK(wrq->q_next));
3961 
3962 		/* Do this processing here as a performance concern */
3963 		if (strmsgsz != 0) {
3964 			if (rmax == INFPSZ)
3965 				rmax = strmsgsz;
3966 			else  {
3967 				if (vp->v_type == VFIFO)
3968 					rmax = MIN(PIPE_BUF, rmax);
3969 				else	rmax = MIN(strmsgsz, rmax);
3970 			}
3971 		}
3972 
3973 		mutex_enter(QLOCK(wrq));
3974 		stp->sd_qn_minpsz = rmin;
3975 		stp->sd_qn_maxpsz = rmax;
3976 		mutex_exit(QLOCK(wrq));
3977 
3978 		/* If we popped through the anchor, then reset the anchor. */
3979 		if (stp->sd_pushcnt < stp->sd_anchor) {
3980 			stp->sd_anchor = 0;
3981 			stp->sd_anchorzone = 0;
3982 		}
3983 		strendplumb(stp);
3984 		mutex_exit(&stp->sd_lock);
3985 		return (error);
3986 	}
3987 
3988 	case _I_MUXID2FD:
3989 	{
3990 		/*
3991 		 * Create a fd for a I_PLINK'ed lower stream with a given
3992 		 * muxid.  With the fd, application can send down ioctls,
3993 		 * like I_LIST, to the previously I_PLINK'ed stream.  Note
3994 		 * that after getting the fd, the application has to do an
3995 		 * I_PUNLINK on the muxid before it can do any operation
3996 		 * on the lower stream.  This is required by spec1170.
3997 		 *
3998 		 * The fd used to do this ioctl should point to the same
3999 		 * controlling device used to do the I_PLINK.  If it uses
4000 		 * a different stream or an invalid muxid, I_MUXID2FD will
4001 		 * fail.  The error code is set to EINVAL.
4002 		 *
4003 		 * The intended use of this interface is the following.
4004 		 * An application I_PLINK'ed a stream and exits.  The fd
4005 		 * to the lower stream is gone.  Another application
4006 		 * wants to get a fd to the lower stream, it uses I_MUXID2FD.
4007 		 */
4008 		int muxid = (int)arg;
4009 		int fd;
4010 		linkinfo_t *linkp;
4011 		struct file *fp;
4012 		netstack_t *ns;
4013 		str_stack_t *ss;
4014 
4015 		/*
4016 		 * Do not allow the wildcard muxid.  This ioctl is not
4017 		 * intended to find arbitrary link.
4018 		 */
4019 		if (muxid == 0) {
4020 			return (EINVAL);
4021 		}
4022 
4023 		ns = netstack_find_by_cred(crp);
4024 		ASSERT(ns != NULL);
4025 		ss = ns->netstack_str;
4026 		ASSERT(ss != NULL);
4027 
4028 		mutex_enter(&muxifier);
4029 		linkp = findlinks(vp->v_stream, muxid, LINKPERSIST, ss);
4030 		if (linkp == NULL) {
4031 			mutex_exit(&muxifier);
4032 			netstack_rele(ss->ss_netstack);
4033 			return (EINVAL);
4034 		}
4035 
4036 		if ((fd = ufalloc(0)) == -1) {
4037 			mutex_exit(&muxifier);
4038 			netstack_rele(ss->ss_netstack);
4039 			return (EMFILE);
4040 		}
4041 		fp = linkp->li_fpdown;
4042 		mutex_enter(&fp->f_tlock);
4043 		fp->f_count++;
4044 		mutex_exit(&fp->f_tlock);
4045 		mutex_exit(&muxifier);
4046 		setf(fd, fp);
4047 		*rvalp = fd;
4048 		netstack_rele(ss->ss_netstack);
4049 		return (0);
4050 	}
4051 
4052 	case _I_INSERT:
4053 	{
4054 		/*
4055 		 * To insert a module to a given position in a stream.
4056 		 * In the first release, only allow privileged user
4057 		 * to use this ioctl. Furthermore, the insert is only allowed
4058 		 * below an anchor if the zoneid is the same as the zoneid
4059 		 * which created the anchor.
4060 		 *
4061 		 * Note that we do not plan to support this ioctl
4062 		 * on pipes in the first release.  We want to learn more
4063 		 * about the implications of these ioctls before extending
4064 		 * their support.  And we do not think these features are
4065 		 * valuable for pipes.
4066 		 */
4067 		STRUCT_DECL(strmodconf, strmodinsert);
4068 		char mod_name[FMNAMESZ + 1];
4069 		fmodsw_impl_t *fp;
4070 		dev_t dummydev;
4071 		queue_t *tmp_wrq;
4072 		int pos;
4073 		boolean_t is_insert;
4074 
4075 		STRUCT_INIT(strmodinsert, flag);
4076 		if (stp->sd_flag & STRHUP)
4077 			return (ENXIO);
4078 		if (STRMATED(stp))
4079 			return (EINVAL);
4080 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4081 			return (error);
4082 		if (stp->sd_anchor != 0 &&
4083 		    stp->sd_anchorzone != crgetzoneid(crp))
4084 			return (EINVAL);
4085 
4086 		error = strcopyin((void *)arg, STRUCT_BUF(strmodinsert),
4087 		    STRUCT_SIZE(strmodinsert), copyflag);
4088 		if (error)
4089 			return (error);
4090 
4091 		/*
4092 		 * Get module name and look up in fmodsw.
4093 		 */
4094 		error = (copyflag & U_TO_K ? copyinstr :
4095 		    copystr)(STRUCT_FGETP(strmodinsert, mod_name),
4096 		    mod_name, FMNAMESZ + 1, NULL);
4097 		if (error)
4098 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4099 
4100 		if ((fp = fmodsw_find(mod_name, FMODSW_HOLD | FMODSW_LOAD)) ==
4101 		    NULL)
4102 			return (EINVAL);
4103 
4104 		if (error = strstartplumb(stp, flag, cmd)) {
4105 			fmodsw_rele(fp);
4106 			return (error);
4107 		}
4108 
4109 		/*
4110 		 * Is this _I_INSERT just like an I_PUSH?  We need to know
4111 		 * this because we do some optimizations if this is a
4112 		 * module being pushed.
4113 		 */
4114 		pos = STRUCT_FGET(strmodinsert, pos);
4115 		is_insert = (pos != 0);
4116 
4117 		/*
4118 		 * Make sure pos is valid.  Even though it is not an I_PUSH,
4119 		 * we impose the same limit on the number of modules in a
4120 		 * stream.
4121 		 */
4122 		mutex_enter(&stp->sd_lock);
4123 		if (stp->sd_pushcnt >= nstrpush || pos < 0 ||
4124 		    pos > stp->sd_pushcnt) {
4125 			fmodsw_rele(fp);
4126 			strendplumb(stp);
4127 			mutex_exit(&stp->sd_lock);
4128 			return (EINVAL);
4129 		}
4130 		if (stp->sd_anchor != 0) {
4131 			/*
4132 			 * Is this insert below the anchor?
4133 			 * Pushcnt hasn't been increased yet hence
4134 			 * we test for greater than here, and greater or
4135 			 * equal after qattach.
4136 			 */
4137 			if (pos > (stp->sd_pushcnt - stp->sd_anchor) &&
4138 			    stp->sd_anchorzone != crgetzoneid(crp)) {
4139 				fmodsw_rele(fp);
4140 				strendplumb(stp);
4141 				mutex_exit(&stp->sd_lock);
4142 				return (EPERM);
4143 			}
4144 		}
4145 
4146 		mutex_exit(&stp->sd_lock);
4147 
4148 		/*
4149 		 * First find the correct position this module to
4150 		 * be inserted.  We don't need to call claimstr()
4151 		 * as the stream should not be changing at this point.
4152 		 *
4153 		 * Insert new module and call its open routine
4154 		 * via qattach().  Modules don't change device
4155 		 * numbers, so just ignore dummydev here.
4156 		 */
4157 		for (tmp_wrq = stp->sd_wrq; pos > 0;
4158 		    tmp_wrq = tmp_wrq->q_next, pos--) {
4159 			ASSERT(SAMESTR(tmp_wrq));
4160 		}
4161 		dummydev = vp->v_rdev;
4162 		if ((error = qattach(_RD(tmp_wrq), &dummydev, 0, crp,
4163 		    fp, is_insert)) != 0) {
4164 			mutex_enter(&stp->sd_lock);
4165 			strendplumb(stp);
4166 			mutex_exit(&stp->sd_lock);
4167 			return (error);
4168 		}
4169 
4170 		mutex_enter(&stp->sd_lock);
4171 
4172 		/*
4173 		 * As a performance concern we are caching the values of
4174 		 * q_minpsz and q_maxpsz of the module below the stream
4175 		 * head in the stream head.
4176 		 */
4177 		if (!is_insert) {
4178 			mutex_enter(QLOCK(stp->sd_wrq->q_next));
4179 			rmin = stp->sd_wrq->q_next->q_minpsz;
4180 			rmax = stp->sd_wrq->q_next->q_maxpsz;
4181 			mutex_exit(QLOCK(stp->sd_wrq->q_next));
4182 
4183 			/* Do this processing here as a performance concern */
4184 			if (strmsgsz != 0) {
4185 				if (rmax == INFPSZ) {
4186 					rmax = strmsgsz;
4187 				} else  {
4188 					rmax = MIN(strmsgsz, rmax);
4189 				}
4190 			}
4191 
4192 			mutex_enter(QLOCK(wrq));
4193 			stp->sd_qn_minpsz = rmin;
4194 			stp->sd_qn_maxpsz = rmax;
4195 			mutex_exit(QLOCK(wrq));
4196 		}
4197 
4198 		/*
4199 		 * Need to update the anchor value if this module is
4200 		 * inserted below the anchor point.
4201 		 */
4202 		if (stp->sd_anchor != 0) {
4203 			pos = STRUCT_FGET(strmodinsert, pos);
4204 			if (pos >= (stp->sd_pushcnt - stp->sd_anchor))
4205 				stp->sd_anchor++;
4206 		}
4207 
4208 		strendplumb(stp);
4209 		mutex_exit(&stp->sd_lock);
4210 		return (0);
4211 	}
4212 
4213 	case _I_REMOVE:
4214 	{
4215 		/*
4216 		 * To remove a module with a given name in a stream.  The
4217 		 * caller of this ioctl needs to provide both the name and
4218 		 * the position of the module to be removed.  This eliminates
4219 		 * the ambiguity of removal if a module is inserted/pushed
4220 		 * multiple times in a stream.  In the first release, only
4221 		 * allow privileged user to use this ioctl.
4222 		 * Furthermore, the remove is only allowed
4223 		 * below an anchor if the zoneid is the same as the zoneid
4224 		 * which created the anchor.
4225 		 *
4226 		 * Note that we do not plan to support this ioctl
4227 		 * on pipes in the first release.  We want to learn more
4228 		 * about the implications of these ioctls before extending
4229 		 * their support.  And we do not think these features are
4230 		 * valuable for pipes.
4231 		 *
4232 		 * Also note that _I_REMOVE cannot be used to remove a
4233 		 * driver or the stream head.
4234 		 */
4235 		STRUCT_DECL(strmodconf, strmodremove);
4236 		queue_t	*q;
4237 		int pos;
4238 		char mod_name[FMNAMESZ + 1];
4239 		boolean_t is_remove;
4240 
4241 		STRUCT_INIT(strmodremove, flag);
4242 		if (stp->sd_flag & STRHUP)
4243 			return (ENXIO);
4244 		if (STRMATED(stp))
4245 			return (EINVAL);
4246 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4247 			return (error);
4248 		if (stp->sd_anchor != 0 &&
4249 		    stp->sd_anchorzone != crgetzoneid(crp))
4250 			return (EINVAL);
4251 
4252 		error = strcopyin((void *)arg, STRUCT_BUF(strmodremove),
4253 		    STRUCT_SIZE(strmodremove), copyflag);
4254 		if (error)
4255 			return (error);
4256 
4257 		error = (copyflag & U_TO_K ? copyinstr :
4258 		    copystr)(STRUCT_FGETP(strmodremove, mod_name),
4259 		    mod_name, FMNAMESZ + 1, NULL);
4260 		if (error)
4261 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4262 
4263 		if ((error = strstartplumb(stp, flag, cmd)) != 0)
4264 			return (error);
4265 
4266 		/*
4267 		 * Match the name of given module to the name of module at
4268 		 * the given position.
4269 		 */
4270 		pos = STRUCT_FGET(strmodremove, pos);
4271 
4272 		is_remove = (pos != 0);
4273 		for (q = stp->sd_wrq->q_next; SAMESTR(q) && pos > 0;
4274 		    q = q->q_next, pos--)
4275 			;
4276 		if (pos > 0 || !SAMESTR(q) ||
4277 		    strcmp(Q2NAME(q), mod_name) != 0) {
4278 			mutex_enter(&stp->sd_lock);
4279 			strendplumb(stp);
4280 			mutex_exit(&stp->sd_lock);
4281 			return (EINVAL);
4282 		}
4283 
4284 		/*
4285 		 * If the position is at or below an anchor, then the zoneid
4286 		 * must match the zoneid that created the anchor.
4287 		 */
4288 		if (stp->sd_anchor != 0) {
4289 			pos = STRUCT_FGET(strmodremove, pos);
4290 			if (pos >= (stp->sd_pushcnt - stp->sd_anchor) &&
4291 			    stp->sd_anchorzone != crgetzoneid(crp)) {
4292 				mutex_enter(&stp->sd_lock);
4293 				strendplumb(stp);
4294 				mutex_exit(&stp->sd_lock);
4295 				return (EPERM);
4296 			}
4297 		}
4298 
4299 
4300 		ASSERT(!(q->q_flag & QREADR));
4301 		qdetach(_RD(q), 1, flag, crp, is_remove);
4302 
4303 		mutex_enter(&stp->sd_lock);
4304 
4305 		/*
4306 		 * As a performance concern we are caching the values of
4307 		 * q_minpsz and q_maxpsz of the module below the stream
4308 		 * head in the stream head.
4309 		 */
4310 		if (!is_remove) {
4311 			mutex_enter(QLOCK(wrq->q_next));
4312 			rmin = wrq->q_next->q_minpsz;
4313 			rmax = wrq->q_next->q_maxpsz;
4314 			mutex_exit(QLOCK(wrq->q_next));
4315 
4316 			/* Do this processing here as a performance concern */
4317 			if (strmsgsz != 0) {
4318 				if (rmax == INFPSZ)
4319 					rmax = strmsgsz;
4320 				else  {
4321 					if (vp->v_type == VFIFO)
4322 						rmax = MIN(PIPE_BUF, rmax);
4323 					else	rmax = MIN(strmsgsz, rmax);
4324 				}
4325 			}
4326 
4327 			mutex_enter(QLOCK(wrq));
4328 			stp->sd_qn_minpsz = rmin;
4329 			stp->sd_qn_maxpsz = rmax;
4330 			mutex_exit(QLOCK(wrq));
4331 		}
4332 
4333 		/*
4334 		 * Need to update the anchor value if this module is removed
4335 		 * at or below the anchor point.  If the removed module is at
4336 		 * the anchor point, remove the anchor for this stream if
4337 		 * there is no module above the anchor point.  Otherwise, if
4338 		 * the removed module is below the anchor point, decrement the
4339 		 * anchor point by 1.
4340 		 */
4341 		if (stp->sd_anchor != 0) {
4342 			pos = STRUCT_FGET(strmodremove, pos);
4343 			if (pos == stp->sd_pushcnt - stp->sd_anchor + 1)
4344 				stp->sd_anchor = 0;
4345 			else if (pos > (stp->sd_pushcnt - stp->sd_anchor + 1))
4346 				stp->sd_anchor--;
4347 		}
4348 
4349 		strendplumb(stp);
4350 		mutex_exit(&stp->sd_lock);
4351 		return (0);
4352 	}
4353 
4354 	case I_ANCHOR:
4355 		/*
4356 		 * Set the anchor position on the stream to reside at
4357 		 * the top module (in other words, the top module
4358 		 * cannot be popped).  Anchors with a FIFO make no
4359 		 * obvious sense, so they're not allowed.
4360 		 */
4361 		mutex_enter(&stp->sd_lock);
4362 
4363 		if (stp->sd_vnode->v_type == VFIFO) {
4364 			mutex_exit(&stp->sd_lock);
4365 			return (EINVAL);
4366 		}
4367 		/* Only allow the same zoneid to update the anchor */
4368 		if (stp->sd_anchor != 0 &&
4369 		    stp->sd_anchorzone != crgetzoneid(crp)) {
4370 			mutex_exit(&stp->sd_lock);
4371 			return (EINVAL);
4372 		}
4373 		stp->sd_anchor = stp->sd_pushcnt;
4374 		stp->sd_anchorzone = crgetzoneid(crp);
4375 		mutex_exit(&stp->sd_lock);
4376 		return (0);
4377 
4378 	case I_LOOK:
4379 		/*
4380 		 * Get name of first module downstream.
4381 		 * If no module, return an error.
4382 		 */
4383 		claimstr(wrq);
4384 		if (_SAMESTR(wrq) && wrq->q_next->q_next != NULL) {
4385 			char *name = Q2NAME(wrq->q_next);
4386 
4387 			error = strcopyout(name, (void *)arg, strlen(name) + 1,
4388 			    copyflag);
4389 			releasestr(wrq);
4390 			return (error);
4391 		}
4392 		releasestr(wrq);
4393 		return (EINVAL);
4394 
4395 	case I_LINK:
4396 	case I_PLINK:
4397 		/*
4398 		 * Link a multiplexor.
4399 		 */
4400 		return (mlink(vp, cmd, (int)arg, crp, rvalp, 0));
4401 
4402 	case _I_PLINK_LH:
4403 		/*
4404 		 * Link a multiplexor: Call must originate from kernel.
4405 		 */
4406 		if (kioctl)
4407 			return (ldi_mlink_lh(vp, cmd, arg, crp, rvalp));
4408 
4409 		return (EINVAL);
4410 	case I_UNLINK:
4411 	case I_PUNLINK:
4412 		/*
4413 		 * Unlink a multiplexor.
4414 		 * If arg is -1, unlink all links for which this is the
4415 		 * controlling stream.  Otherwise, arg is an index number
4416 		 * for a link to be removed.
4417 		 */
4418 	{
4419 		struct linkinfo *linkp;
4420 		int native_arg = (int)arg;
4421 		int type;
4422 		netstack_t *ns;
4423 		str_stack_t *ss;
4424 
4425 		TRACE_1(TR_FAC_STREAMS_FR,
4426 		    TR_I_UNLINK, "I_UNLINK/I_PUNLINK:%p", stp);
4427 		if (vp->v_type == VFIFO) {
4428 			return (EINVAL);
4429 		}
4430 		if (cmd == I_UNLINK)
4431 			type = LINKNORMAL;
4432 		else	/* I_PUNLINK */
4433 			type = LINKPERSIST;
4434 		if (native_arg == 0) {
4435 			return (EINVAL);
4436 		}
4437 		ns = netstack_find_by_cred(crp);
4438 		ASSERT(ns != NULL);
4439 		ss = ns->netstack_str;
4440 		ASSERT(ss != NULL);
4441 
4442 		if (native_arg == MUXID_ALL)
4443 			error = munlinkall(stp, type, crp, rvalp, ss);
4444 		else {
4445 			mutex_enter(&muxifier);
4446 			if (!(linkp = findlinks(stp, (int)arg, type, ss))) {
4447 				/* invalid user supplied index number */
4448 				mutex_exit(&muxifier);
4449 				netstack_rele(ss->ss_netstack);
4450 				return (EINVAL);
4451 			}
4452 			/* munlink drops the muxifier lock */
4453 			error = munlink(stp, linkp, type, crp, rvalp, ss);
4454 		}
4455 		netstack_rele(ss->ss_netstack);
4456 		return (error);
4457 	}
4458 
4459 	case I_FLUSH:
4460 		/*
4461 		 * send a flush message downstream
4462 		 * flush message can indicate
4463 		 * FLUSHR - flush read queue
4464 		 * FLUSHW - flush write queue
4465 		 * FLUSHRW - flush read/write queue
4466 		 */
4467 		if (stp->sd_flag & STRHUP)
4468 			return (ENXIO);
4469 		if (arg & ~FLUSHRW)
4470 			return (EINVAL);
4471 
4472 		for (;;) {
4473 			if (putnextctl1(stp->sd_wrq, M_FLUSH, (int)arg)) {
4474 				break;
4475 			}
4476 			if (error = strwaitbuf(1, BPRI_HI)) {
4477 				return (error);
4478 			}
4479 		}
4480 
4481 		/*
4482 		 * Send down an unsupported ioctl and wait for the nack
4483 		 * in order to allow the M_FLUSH to propagate back
4484 		 * up to the stream head.
4485 		 * Replaces if (qready()) runqueues();
4486 		 */
4487 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4488 		strioc.ic_timout = 0;
4489 		strioc.ic_len = 0;
4490 		strioc.ic_dp = NULL;
4491 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4492 		*rvalp = 0;
4493 		return (0);
4494 
4495 	case I_FLUSHBAND:
4496 	{
4497 		struct bandinfo binfo;
4498 
4499 		error = strcopyin((void *)arg, &binfo, sizeof (binfo),
4500 		    copyflag);
4501 		if (error)
4502 			return (error);
4503 		if (stp->sd_flag & STRHUP)
4504 			return (ENXIO);
4505 		if (binfo.bi_flag & ~FLUSHRW)
4506 			return (EINVAL);
4507 		while (!(mp = allocb(2, BPRI_HI))) {
4508 			if (error = strwaitbuf(2, BPRI_HI))
4509 				return (error);
4510 		}
4511 		mp->b_datap->db_type = M_FLUSH;
4512 		*mp->b_wptr++ = binfo.bi_flag | FLUSHBAND;
4513 		*mp->b_wptr++ = binfo.bi_pri;
4514 		putnext(stp->sd_wrq, mp);
4515 		/*
4516 		 * Send down an unsupported ioctl and wait for the nack
4517 		 * in order to allow the M_FLUSH to propagate back
4518 		 * up to the stream head.
4519 		 * Replaces if (qready()) runqueues();
4520 		 */
4521 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4522 		strioc.ic_timout = 0;
4523 		strioc.ic_len = 0;
4524 		strioc.ic_dp = NULL;
4525 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4526 		*rvalp = 0;
4527 		return (0);
4528 	}
4529 
4530 	case I_SRDOPT:
4531 		/*
4532 		 * Set read options
4533 		 *
4534 		 * RNORM - default stream mode
4535 		 * RMSGN - message no discard
4536 		 * RMSGD - message discard
4537 		 * RPROTNORM - fail read with EBADMSG for M_[PC]PROTOs
4538 		 * RPROTDAT - convert M_[PC]PROTOs to M_DATAs
4539 		 * RPROTDIS - discard M_[PC]PROTOs and retain M_DATAs
4540 		 */
4541 		if (arg & ~(RMODEMASK | RPROTMASK))
4542 			return (EINVAL);
4543 
4544 		if ((arg & (RMSGD|RMSGN)) == (RMSGD|RMSGN))
4545 			return (EINVAL);
4546 
4547 		mutex_enter(&stp->sd_lock);
4548 		switch (arg & RMODEMASK) {
4549 		case RNORM:
4550 			stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
4551 			break;
4552 		case RMSGD:
4553 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGNODIS) |
4554 			    RD_MSGDIS;
4555 			break;
4556 		case RMSGN:
4557 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGDIS) |
4558 			    RD_MSGNODIS;
4559 			break;
4560 		}
4561 
4562 		switch (arg & RPROTMASK) {
4563 		case RPROTNORM:
4564 			stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
4565 			break;
4566 
4567 		case RPROTDAT:
4568 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDIS) |
4569 			    RD_PROTDAT);
4570 			break;
4571 
4572 		case RPROTDIS:
4573 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDAT) |
4574 			    RD_PROTDIS);
4575 			break;
4576 		}
4577 		mutex_exit(&stp->sd_lock);
4578 		return (0);
4579 
4580 	case I_GRDOPT:
4581 		/*
4582 		 * Get read option and return the value
4583 		 * to spot pointed to by arg
4584 		 */
4585 	{
4586 		int rdopt;
4587 
4588 		rdopt = ((stp->sd_read_opt & RD_MSGDIS) ? RMSGD :
4589 		    ((stp->sd_read_opt & RD_MSGNODIS) ? RMSGN : RNORM));
4590 		rdopt |= ((stp->sd_read_opt & RD_PROTDAT) ? RPROTDAT :
4591 		    ((stp->sd_read_opt & RD_PROTDIS) ? RPROTDIS : RPROTNORM));
4592 
4593 		return (strcopyout(&rdopt, (void *)arg, sizeof (int),
4594 		    copyflag));
4595 	}
4596 
4597 	case I_SERROPT:
4598 		/*
4599 		 * Set error options
4600 		 *
4601 		 * RERRNORM - persistent read errors
4602 		 * RERRNONPERSIST - non-persistent read errors
4603 		 * WERRNORM - persistent write errors
4604 		 * WERRNONPERSIST - non-persistent write errors
4605 		 */
4606 		if (arg & ~(RERRMASK | WERRMASK))
4607 			return (EINVAL);
4608 
4609 		mutex_enter(&stp->sd_lock);
4610 		switch (arg & RERRMASK) {
4611 		case RERRNORM:
4612 			stp->sd_flag &= ~STRDERRNONPERSIST;
4613 			break;
4614 		case RERRNONPERSIST:
4615 			stp->sd_flag |= STRDERRNONPERSIST;
4616 			break;
4617 		}
4618 		switch (arg & WERRMASK) {
4619 		case WERRNORM:
4620 			stp->sd_flag &= ~STWRERRNONPERSIST;
4621 			break;
4622 		case WERRNONPERSIST:
4623 			stp->sd_flag |= STWRERRNONPERSIST;
4624 			break;
4625 		}
4626 		mutex_exit(&stp->sd_lock);
4627 		return (0);
4628 
4629 	case I_GERROPT:
4630 		/*
4631 		 * Get error option and return the value
4632 		 * to spot pointed to by arg
4633 		 */
4634 	{
4635 		int erropt = 0;
4636 
4637 		erropt |= (stp->sd_flag & STRDERRNONPERSIST) ? RERRNONPERSIST :
4638 		    RERRNORM;
4639 		erropt |= (stp->sd_flag & STWRERRNONPERSIST) ? WERRNONPERSIST :
4640 		    WERRNORM;
4641 		return (strcopyout(&erropt, (void *)arg, sizeof (int),
4642 		    copyflag));
4643 	}
4644 
4645 	case I_SETSIG:
4646 		/*
4647 		 * Register the calling proc to receive the SIGPOLL
4648 		 * signal based on the events given in arg.  If
4649 		 * arg is zero, remove the proc from register list.
4650 		 */
4651 	{
4652 		strsig_t *ssp, *pssp;
4653 		struct pid *pidp;
4654 
4655 		pssp = NULL;
4656 		pidp = curproc->p_pidp;
4657 		/*
4658 		 * Hold sd_lock to prevent traversal of sd_siglist while
4659 		 * it is modified.
4660 		 */
4661 		mutex_enter(&stp->sd_lock);
4662 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pidp != pidp);
4663 		    pssp = ssp, ssp = ssp->ss_next)
4664 			;
4665 
4666 		if (arg) {
4667 			if (arg & ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4668 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4669 				mutex_exit(&stp->sd_lock);
4670 				return (EINVAL);
4671 			}
4672 			if ((arg & S_BANDURG) && !(arg & S_RDBAND)) {
4673 				mutex_exit(&stp->sd_lock);
4674 				return (EINVAL);
4675 			}
4676 
4677 			/*
4678 			 * If proc not already registered, add it
4679 			 * to list.
4680 			 */
4681 			if (!ssp) {
4682 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4683 				ssp->ss_pidp = pidp;
4684 				ssp->ss_pid = pidp->pid_id;
4685 				ssp->ss_next = NULL;
4686 				if (pssp)
4687 					pssp->ss_next = ssp;
4688 				else
4689 					stp->sd_siglist = ssp;
4690 				mutex_enter(&pidlock);
4691 				PID_HOLD(pidp);
4692 				mutex_exit(&pidlock);
4693 			}
4694 
4695 			/*
4696 			 * Set events.
4697 			 */
4698 			ssp->ss_events = (int)arg;
4699 		} else {
4700 			/*
4701 			 * Remove proc from register list.
4702 			 */
4703 			if (ssp) {
4704 				mutex_enter(&pidlock);
4705 				PID_RELE(pidp);
4706 				mutex_exit(&pidlock);
4707 				if (pssp)
4708 					pssp->ss_next = ssp->ss_next;
4709 				else
4710 					stp->sd_siglist = ssp->ss_next;
4711 				kmem_free(ssp, sizeof (strsig_t));
4712 			} else {
4713 				mutex_exit(&stp->sd_lock);
4714 				return (EINVAL);
4715 			}
4716 		}
4717 
4718 		/*
4719 		 * Recalculate OR of sig events.
4720 		 */
4721 		stp->sd_sigflags = 0;
4722 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4723 			stp->sd_sigflags |= ssp->ss_events;
4724 		mutex_exit(&stp->sd_lock);
4725 		return (0);
4726 	}
4727 
4728 	case I_GETSIG:
4729 		/*
4730 		 * Return (in arg) the current registration of events
4731 		 * for which the calling proc is to be signaled.
4732 		 */
4733 	{
4734 		struct strsig *ssp;
4735 		struct pid  *pidp;
4736 
4737 		pidp = curproc->p_pidp;
4738 		mutex_enter(&stp->sd_lock);
4739 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4740 			if (ssp->ss_pidp == pidp) {
4741 				error = strcopyout(&ssp->ss_events, (void *)arg,
4742 				    sizeof (int), copyflag);
4743 				mutex_exit(&stp->sd_lock);
4744 				return (error);
4745 			}
4746 		mutex_exit(&stp->sd_lock);
4747 		return (EINVAL);
4748 	}
4749 
4750 	case I_ESETSIG:
4751 		/*
4752 		 * Register the ss_pid to receive the SIGPOLL
4753 		 * signal based on the events is ss_events arg.  If
4754 		 * ss_events is zero, remove the proc from register list.
4755 		 */
4756 	{
4757 		struct strsig *ssp, *pssp;
4758 		struct proc *proc;
4759 		struct pid  *pidp;
4760 		pid_t pid;
4761 		struct strsigset ss;
4762 
4763 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4764 		if (error)
4765 			return (error);
4766 
4767 		pid = ss.ss_pid;
4768 
4769 		if (ss.ss_events != 0) {
4770 			/*
4771 			 * Permissions check by sending signal 0.
4772 			 * Note that when kill fails it does a set_errno
4773 			 * causing the system call to fail.
4774 			 */
4775 			error = kill(pid, 0);
4776 			if (error) {
4777 				return (error);
4778 			}
4779 		}
4780 		mutex_enter(&pidlock);
4781 		if (pid == 0)
4782 			proc = curproc;
4783 		else if (pid < 0)
4784 			proc = pgfind(-pid);
4785 		else
4786 			proc = prfind(pid);
4787 		if (proc == NULL) {
4788 			mutex_exit(&pidlock);
4789 			return (ESRCH);
4790 		}
4791 		if (pid < 0)
4792 			pidp = proc->p_pgidp;
4793 		else
4794 			pidp = proc->p_pidp;
4795 		ASSERT(pidp);
4796 		/*
4797 		 * Get a hold on the pid structure while referencing it.
4798 		 * There is a separate PID_HOLD should it be inserted
4799 		 * in the list below.
4800 		 */
4801 		PID_HOLD(pidp);
4802 		mutex_exit(&pidlock);
4803 
4804 		pssp = NULL;
4805 		/*
4806 		 * Hold sd_lock to prevent traversal of sd_siglist while
4807 		 * it is modified.
4808 		 */
4809 		mutex_enter(&stp->sd_lock);
4810 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pid != pid);
4811 		    pssp = ssp, ssp = ssp->ss_next)
4812 			;
4813 
4814 		if (ss.ss_events) {
4815 			if (ss.ss_events &
4816 			    ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4817 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4818 				mutex_exit(&stp->sd_lock);
4819 				mutex_enter(&pidlock);
4820 				PID_RELE(pidp);
4821 				mutex_exit(&pidlock);
4822 				return (EINVAL);
4823 			}
4824 			if ((ss.ss_events & S_BANDURG) &&
4825 			    !(ss.ss_events & S_RDBAND)) {
4826 				mutex_exit(&stp->sd_lock);
4827 				mutex_enter(&pidlock);
4828 				PID_RELE(pidp);
4829 				mutex_exit(&pidlock);
4830 				return (EINVAL);
4831 			}
4832 
4833 			/*
4834 			 * If proc not already registered, add it
4835 			 * to list.
4836 			 */
4837 			if (!ssp) {
4838 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4839 				ssp->ss_pidp = pidp;
4840 				ssp->ss_pid = pid;
4841 				ssp->ss_next = NULL;
4842 				if (pssp)
4843 					pssp->ss_next = ssp;
4844 				else
4845 					stp->sd_siglist = ssp;
4846 				mutex_enter(&pidlock);
4847 				PID_HOLD(pidp);
4848 				mutex_exit(&pidlock);
4849 			}
4850 
4851 			/*
4852 			 * Set events.
4853 			 */
4854 			ssp->ss_events = ss.ss_events;
4855 		} else {
4856 			/*
4857 			 * Remove proc from register list.
4858 			 */
4859 			if (ssp) {
4860 				mutex_enter(&pidlock);
4861 				PID_RELE(pidp);
4862 				mutex_exit(&pidlock);
4863 				if (pssp)
4864 					pssp->ss_next = ssp->ss_next;
4865 				else
4866 					stp->sd_siglist = ssp->ss_next;
4867 				kmem_free(ssp, sizeof (strsig_t));
4868 			} else {
4869 				mutex_exit(&stp->sd_lock);
4870 				mutex_enter(&pidlock);
4871 				PID_RELE(pidp);
4872 				mutex_exit(&pidlock);
4873 				return (EINVAL);
4874 			}
4875 		}
4876 
4877 		/*
4878 		 * Recalculate OR of sig events.
4879 		 */
4880 		stp->sd_sigflags = 0;
4881 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4882 			stp->sd_sigflags |= ssp->ss_events;
4883 		mutex_exit(&stp->sd_lock);
4884 		mutex_enter(&pidlock);
4885 		PID_RELE(pidp);
4886 		mutex_exit(&pidlock);
4887 		return (0);
4888 	}
4889 
4890 	case I_EGETSIG:
4891 		/*
4892 		 * Return (in arg) the current registration of events
4893 		 * for which the calling proc is to be signaled.
4894 		 */
4895 	{
4896 		struct strsig *ssp;
4897 		struct proc *proc;
4898 		pid_t pid;
4899 		struct pid  *pidp;
4900 		struct strsigset ss;
4901 
4902 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4903 		if (error)
4904 			return (error);
4905 
4906 		pid = ss.ss_pid;
4907 		mutex_enter(&pidlock);
4908 		if (pid == 0)
4909 			proc = curproc;
4910 		else if (pid < 0)
4911 			proc = pgfind(-pid);
4912 		else
4913 			proc = prfind(pid);
4914 		if (proc == NULL) {
4915 			mutex_exit(&pidlock);
4916 			return (ESRCH);
4917 		}
4918 		if (pid < 0)
4919 			pidp = proc->p_pgidp;
4920 		else
4921 			pidp = proc->p_pidp;
4922 
4923 		/* Prevent the pidp from being reassigned */
4924 		PID_HOLD(pidp);
4925 		mutex_exit(&pidlock);
4926 
4927 		mutex_enter(&stp->sd_lock);
4928 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4929 			if (ssp->ss_pid == pid) {
4930 				ss.ss_pid = ssp->ss_pid;
4931 				ss.ss_events = ssp->ss_events;
4932 				error = strcopyout(&ss, (void *)arg,
4933 				    sizeof (struct strsigset), copyflag);
4934 				mutex_exit(&stp->sd_lock);
4935 				mutex_enter(&pidlock);
4936 				PID_RELE(pidp);
4937 				mutex_exit(&pidlock);
4938 				return (error);
4939 			}
4940 		mutex_exit(&stp->sd_lock);
4941 		mutex_enter(&pidlock);
4942 		PID_RELE(pidp);
4943 		mutex_exit(&pidlock);
4944 		return (EINVAL);
4945 	}
4946 
4947 	case I_PEEK:
4948 	{
4949 		STRUCT_DECL(strpeek, strpeek);
4950 		size_t n;
4951 		mblk_t *fmp, *tmp_mp = NULL;
4952 
4953 		STRUCT_INIT(strpeek, flag);
4954 
4955 		error = strcopyin((void *)arg, STRUCT_BUF(strpeek),
4956 		    STRUCT_SIZE(strpeek), copyflag);
4957 		if (error)
4958 			return (error);
4959 
4960 		mutex_enter(QLOCK(rdq));
4961 		/*
4962 		 * Skip the invalid messages
4963 		 */
4964 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
4965 			if (mp->b_datap->db_type != M_SIG)
4966 				break;
4967 
4968 		/*
4969 		 * If user has requested to peek at a high priority message
4970 		 * and first message is not, return 0
4971 		 */
4972 		if (mp != NULL) {
4973 			if ((STRUCT_FGET(strpeek, flags) & RS_HIPRI) &&
4974 			    queclass(mp) == QNORM) {
4975 				*rvalp = 0;
4976 				mutex_exit(QLOCK(rdq));
4977 				return (0);
4978 			}
4979 		} else if (stp->sd_struiordq == NULL ||
4980 		    (STRUCT_FGET(strpeek, flags) & RS_HIPRI)) {
4981 			/*
4982 			 * No mblks to look at at the streamhead and
4983 			 * 1). This isn't a synch stream or
4984 			 * 2). This is a synch stream but caller wants high
4985 			 *	priority messages which is not supported by
4986 			 *	the synch stream. (it only supports QNORM)
4987 			 */
4988 			*rvalp = 0;
4989 			mutex_exit(QLOCK(rdq));
4990 			return (0);
4991 		}
4992 
4993 		fmp = mp;
4994 
4995 		if (mp && mp->b_datap->db_type == M_PASSFP) {
4996 			mutex_exit(QLOCK(rdq));
4997 			return (EBADMSG);
4998 		}
4999 
5000 		ASSERT(mp == NULL || mp->b_datap->db_type == M_PCPROTO ||
5001 		    mp->b_datap->db_type == M_PROTO ||
5002 		    mp->b_datap->db_type == M_DATA);
5003 
5004 		if (mp && mp->b_datap->db_type == M_PCPROTO) {
5005 			STRUCT_FSET(strpeek, flags, RS_HIPRI);
5006 		} else {
5007 			STRUCT_FSET(strpeek, flags, 0);
5008 		}
5009 
5010 
5011 		if (mp && ((tmp_mp = dupmsg(mp)) == NULL)) {
5012 			mutex_exit(QLOCK(rdq));
5013 			return (ENOSR);
5014 		}
5015 		mutex_exit(QLOCK(rdq));
5016 
5017 		/*
5018 		 * set mp = tmp_mp, so that I_PEEK processing can continue.
5019 		 * tmp_mp is used to free the dup'd message.
5020 		 */
5021 		mp = tmp_mp;
5022 
5023 		uio.uio_fmode = 0;
5024 		uio.uio_extflg = UIO_COPY_CACHED;
5025 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5026 		    UIO_SYSSPACE;
5027 		uio.uio_limit = 0;
5028 		/*
5029 		 * First process PROTO blocks, if any.
5030 		 * If user doesn't want to get ctl info by setting maxlen <= 0,
5031 		 * then set len to -1/0 and skip control blocks part.
5032 		 */
5033 		if (STRUCT_FGET(strpeek, ctlbuf.maxlen) < 0)
5034 			STRUCT_FSET(strpeek, ctlbuf.len, -1);
5035 		else if (STRUCT_FGET(strpeek, ctlbuf.maxlen) == 0)
5036 			STRUCT_FSET(strpeek, ctlbuf.len, 0);
5037 		else {
5038 			int	ctl_part = 0;
5039 
5040 			iov.iov_base = STRUCT_FGETP(strpeek, ctlbuf.buf);
5041 			iov.iov_len = STRUCT_FGET(strpeek, ctlbuf.maxlen);
5042 			uio.uio_iov = &iov;
5043 			uio.uio_resid = iov.iov_len;
5044 			uio.uio_loffset = 0;
5045 			uio.uio_iovcnt = 1;
5046 			while (mp && mp->b_datap->db_type != M_DATA &&
5047 			    uio.uio_resid >= 0) {
5048 				ASSERT(STRUCT_FGET(strpeek, flags) == 0 ?
5049 				    mp->b_datap->db_type == M_PROTO :
5050 				    mp->b_datap->db_type == M_PCPROTO);
5051 
5052 				if ((n = MIN(uio.uio_resid,
5053 				    mp->b_wptr - mp->b_rptr)) != 0 &&
5054 				    (error = uiomove((char *)mp->b_rptr, n,
5055 				    UIO_READ, &uio)) != 0) {
5056 					freemsg(tmp_mp);
5057 					return (error);
5058 				}
5059 				ctl_part = 1;
5060 				mp = mp->b_cont;
5061 			}
5062 			/* No ctl message */
5063 			if (ctl_part == 0)
5064 				STRUCT_FSET(strpeek, ctlbuf.len, -1);
5065 			else
5066 				STRUCT_FSET(strpeek, ctlbuf.len,
5067 				    STRUCT_FGET(strpeek, ctlbuf.maxlen) -
5068 				    uio.uio_resid);
5069 		}
5070 
5071 		/*
5072 		 * Now process DATA blocks, if any.
5073 		 * If user doesn't want to get data info by setting maxlen <= 0,
5074 		 * then set len to -1/0 and skip data blocks part.
5075 		 */
5076 		if (STRUCT_FGET(strpeek, databuf.maxlen) < 0)
5077 			STRUCT_FSET(strpeek, databuf.len, -1);
5078 		else if (STRUCT_FGET(strpeek, databuf.maxlen) == 0)
5079 			STRUCT_FSET(strpeek, databuf.len, 0);
5080 		else {
5081 			int	data_part = 0;
5082 
5083 			iov.iov_base = STRUCT_FGETP(strpeek, databuf.buf);
5084 			iov.iov_len = STRUCT_FGET(strpeek, databuf.maxlen);
5085 			uio.uio_iov = &iov;
5086 			uio.uio_resid = iov.iov_len;
5087 			uio.uio_loffset = 0;
5088 			uio.uio_iovcnt = 1;
5089 			while (mp && uio.uio_resid) {
5090 				if (mp->b_datap->db_type == M_DATA) {
5091 					if ((n = MIN(uio.uio_resid,
5092 					    mp->b_wptr - mp->b_rptr)) != 0 &&
5093 					    (error = uiomove((char *)mp->b_rptr,
5094 					    n, UIO_READ, &uio)) != 0) {
5095 						freemsg(tmp_mp);
5096 						return (error);
5097 					}
5098 					data_part = 1;
5099 				}
5100 				ASSERT(data_part == 0 ||
5101 				    mp->b_datap->db_type == M_DATA);
5102 				mp = mp->b_cont;
5103 			}
5104 			/* No data message */
5105 			if (data_part == 0)
5106 				STRUCT_FSET(strpeek, databuf.len, -1);
5107 			else
5108 				STRUCT_FSET(strpeek, databuf.len,
5109 				    STRUCT_FGET(strpeek, databuf.maxlen) -
5110 				    uio.uio_resid);
5111 		}
5112 		freemsg(tmp_mp);
5113 
5114 		/*
5115 		 * It is a synch stream and user wants to get
5116 		 * data (maxlen > 0).
5117 		 * uio setup is done by the codes that process DATA
5118 		 * blocks above.
5119 		 */
5120 		if ((fmp == NULL) && STRUCT_FGET(strpeek, databuf.maxlen) > 0) {
5121 			infod_t infod;
5122 
5123 			infod.d_cmd = INFOD_COPYOUT;
5124 			infod.d_res = 0;
5125 			infod.d_uiop = &uio;
5126 			error = infonext(rdq, &infod);
5127 			if (error == EINVAL || error == EBUSY)
5128 				error = 0;
5129 			if (error)
5130 				return (error);
5131 			STRUCT_FSET(strpeek, databuf.len, STRUCT_FGET(strpeek,
5132 			    databuf.maxlen) - uio.uio_resid);
5133 			if (STRUCT_FGET(strpeek, databuf.len) == 0) {
5134 				/*
5135 				 * No data found by the infonext().
5136 				 */
5137 				STRUCT_FSET(strpeek, databuf.len, -1);
5138 			}
5139 		}
5140 		error = strcopyout(STRUCT_BUF(strpeek), (void *)arg,
5141 		    STRUCT_SIZE(strpeek), copyflag);
5142 		if (error) {
5143 			return (error);
5144 		}
5145 		/*
5146 		 * If there is no message retrieved, set return code to 0
5147 		 * otherwise, set it to 1.
5148 		 */
5149 		if (STRUCT_FGET(strpeek, ctlbuf.len) == -1 &&
5150 		    STRUCT_FGET(strpeek, databuf.len) == -1)
5151 			*rvalp = 0;
5152 		else
5153 			*rvalp = 1;
5154 		return (0);
5155 	}
5156 
5157 	case I_FDINSERT:
5158 	{
5159 		STRUCT_DECL(strfdinsert, strfdinsert);
5160 		struct file *resftp;
5161 		struct stdata *resstp;
5162 		t_uscalar_t	ival;
5163 		ssize_t msgsize;
5164 		struct strbuf mctl;
5165 
5166 		STRUCT_INIT(strfdinsert, flag);
5167 		if (stp->sd_flag & STRHUP)
5168 			return (ENXIO);
5169 		/*
5170 		 * STRDERR, STWRERR and STPLEX tested above.
5171 		 */
5172 		error = strcopyin((void *)arg, STRUCT_BUF(strfdinsert),
5173 		    STRUCT_SIZE(strfdinsert), copyflag);
5174 		if (error)
5175 			return (error);
5176 
5177 		if (STRUCT_FGET(strfdinsert, offset) < 0 ||
5178 		    (STRUCT_FGET(strfdinsert, offset) %
5179 		    sizeof (t_uscalar_t)) != 0)
5180 			return (EINVAL);
5181 		if ((resftp = getf(STRUCT_FGET(strfdinsert, fildes))) != NULL) {
5182 			if ((resstp = resftp->f_vnode->v_stream) == NULL) {
5183 				releasef(STRUCT_FGET(strfdinsert, fildes));
5184 				return (EINVAL);
5185 			}
5186 		} else
5187 			return (EINVAL);
5188 
5189 		mutex_enter(&resstp->sd_lock);
5190 		if (resstp->sd_flag & (STRDERR|STWRERR|STRHUP|STPLEX)) {
5191 			error = strgeterr(resstp,
5192 			    STRDERR|STWRERR|STRHUP|STPLEX, 0);
5193 			if (error != 0) {
5194 				mutex_exit(&resstp->sd_lock);
5195 				releasef(STRUCT_FGET(strfdinsert, fildes));
5196 				return (error);
5197 			}
5198 		}
5199 		mutex_exit(&resstp->sd_lock);
5200 
5201 #ifdef	_ILP32
5202 		{
5203 			queue_t	*q;
5204 			queue_t	*mate = NULL;
5205 
5206 			/* get read queue of stream terminus */
5207 			claimstr(resstp->sd_wrq);
5208 			for (q = resstp->sd_wrq->q_next; q->q_next != NULL;
5209 			    q = q->q_next)
5210 				if (!STRMATED(resstp) && STREAM(q) != resstp &&
5211 				    mate == NULL) {
5212 					ASSERT(q->q_qinfo->qi_srvp);
5213 					ASSERT(_OTHERQ(q)->q_qinfo->qi_srvp);
5214 					claimstr(q);
5215 					mate = q;
5216 				}
5217 			q = _RD(q);
5218 			if (mate)
5219 				releasestr(mate);
5220 			releasestr(resstp->sd_wrq);
5221 			ival = (t_uscalar_t)q;
5222 		}
5223 #else
5224 		ival = (t_uscalar_t)getminor(resftp->f_vnode->v_rdev);
5225 #endif	/* _ILP32 */
5226 
5227 		if (STRUCT_FGET(strfdinsert, ctlbuf.len) <
5228 		    STRUCT_FGET(strfdinsert, offset) + sizeof (t_uscalar_t)) {
5229 			releasef(STRUCT_FGET(strfdinsert, fildes));
5230 			return (EINVAL);
5231 		}
5232 
5233 		/*
5234 		 * Check for legal flag value.
5235 		 */
5236 		if (STRUCT_FGET(strfdinsert, flags) & ~RS_HIPRI) {
5237 			releasef(STRUCT_FGET(strfdinsert, fildes));
5238 			return (EINVAL);
5239 		}
5240 
5241 		/* get these values from those cached in the stream head */
5242 		mutex_enter(QLOCK(stp->sd_wrq));
5243 		rmin = stp->sd_qn_minpsz;
5244 		rmax = stp->sd_qn_maxpsz;
5245 		mutex_exit(QLOCK(stp->sd_wrq));
5246 
5247 		/*
5248 		 * Make sure ctl and data sizes together fall within
5249 		 * the limits of the max and min receive packet sizes
5250 		 * and do not exceed system limit.  A negative data
5251 		 * length means that no data part is to be sent.
5252 		 */
5253 		ASSERT((rmax >= 0) || (rmax == INFPSZ));
5254 		if (rmax == 0) {
5255 			releasef(STRUCT_FGET(strfdinsert, fildes));
5256 			return (ERANGE);
5257 		}
5258 		if ((msgsize = STRUCT_FGET(strfdinsert, databuf.len)) < 0)
5259 			msgsize = 0;
5260 		if ((msgsize < rmin) ||
5261 		    ((msgsize > rmax) && (rmax != INFPSZ)) ||
5262 		    (STRUCT_FGET(strfdinsert, ctlbuf.len) > strctlsz)) {
5263 			releasef(STRUCT_FGET(strfdinsert, fildes));
5264 			return (ERANGE);
5265 		}
5266 
5267 		mutex_enter(&stp->sd_lock);
5268 		while (!(STRUCT_FGET(strfdinsert, flags) & RS_HIPRI) &&
5269 		    !canputnext(stp->sd_wrq)) {
5270 			if ((error = strwaitq(stp, WRITEWAIT, (ssize_t)0,
5271 			    flag, -1, &done)) != 0 || done) {
5272 				mutex_exit(&stp->sd_lock);
5273 				releasef(STRUCT_FGET(strfdinsert, fildes));
5274 				return (error);
5275 			}
5276 			if ((error = i_straccess(stp, access)) != 0) {
5277 				mutex_exit(&stp->sd_lock);
5278 				releasef(
5279 				    STRUCT_FGET(strfdinsert, fildes));
5280 				return (error);
5281 			}
5282 		}
5283 		mutex_exit(&stp->sd_lock);
5284 
5285 		/*
5286 		 * Copy strfdinsert.ctlbuf into native form of
5287 		 * ctlbuf to pass down into strmakemsg().
5288 		 */
5289 		mctl.maxlen = STRUCT_FGET(strfdinsert, ctlbuf.maxlen);
5290 		mctl.len = STRUCT_FGET(strfdinsert, ctlbuf.len);
5291 		mctl.buf = STRUCT_FGETP(strfdinsert, ctlbuf.buf);
5292 
5293 		iov.iov_base = STRUCT_FGETP(strfdinsert, databuf.buf);
5294 		iov.iov_len = STRUCT_FGET(strfdinsert, databuf.len);
5295 		uio.uio_iov = &iov;
5296 		uio.uio_iovcnt = 1;
5297 		uio.uio_loffset = 0;
5298 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5299 		    UIO_SYSSPACE;
5300 		uio.uio_fmode = 0;
5301 		uio.uio_extflg = UIO_COPY_CACHED;
5302 		uio.uio_resid = iov.iov_len;
5303 		if ((error = strmakemsg(&mctl,
5304 		    &msgsize, &uio, stp,
5305 		    STRUCT_FGET(strfdinsert, flags), &mp)) != 0 || !mp) {
5306 			STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5307 			releasef(STRUCT_FGET(strfdinsert, fildes));
5308 			return (error);
5309 		}
5310 
5311 		STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5312 
5313 		/*
5314 		 * Place the possibly reencoded queue pointer 'offset' bytes
5315 		 * from the start of the control portion of the message.
5316 		 */
5317 		*((t_uscalar_t *)(mp->b_rptr +
5318 		    STRUCT_FGET(strfdinsert, offset))) = ival;
5319 
5320 		/*
5321 		 * Put message downstream.
5322 		 */
5323 		stream_willservice(stp);
5324 		putnext(stp->sd_wrq, mp);
5325 		stream_runservice(stp);
5326 		releasef(STRUCT_FGET(strfdinsert, fildes));
5327 		return (error);
5328 	}
5329 
5330 	case I_SENDFD:
5331 	{
5332 		struct file *fp;
5333 
5334 		if ((fp = getf((int)arg)) == NULL)
5335 			return (EBADF);
5336 		error = do_sendfp(stp, fp, crp);
5337 		if (audit_active) {
5338 			audit_fdsend((int)arg, fp, error);
5339 		}
5340 		releasef((int)arg);
5341 		return (error);
5342 	}
5343 
5344 	case I_RECVFD:
5345 	case I_E_RECVFD:
5346 	{
5347 		struct k_strrecvfd *srf;
5348 		int i, fd;
5349 
5350 		mutex_enter(&stp->sd_lock);
5351 		while (!(mp = getq(rdq))) {
5352 			if (stp->sd_flag & (STRHUP|STREOF)) {
5353 				mutex_exit(&stp->sd_lock);
5354 				return (ENXIO);
5355 			}
5356 			if ((error = strwaitq(stp, GETWAIT, (ssize_t)0,
5357 			    flag, -1, &done)) != 0 || done) {
5358 				mutex_exit(&stp->sd_lock);
5359 				return (error);
5360 			}
5361 			if ((error = i_straccess(stp, access)) != 0) {
5362 				mutex_exit(&stp->sd_lock);
5363 				return (error);
5364 			}
5365 		}
5366 		if (mp->b_datap->db_type != M_PASSFP) {
5367 			putback(stp, rdq, mp, mp->b_band);
5368 			mutex_exit(&stp->sd_lock);
5369 			return (EBADMSG);
5370 		}
5371 		mutex_exit(&stp->sd_lock);
5372 
5373 		srf = (struct k_strrecvfd *)mp->b_rptr;
5374 		if ((fd = ufalloc(0)) == -1) {
5375 			mutex_enter(&stp->sd_lock);
5376 			putback(stp, rdq, mp, mp->b_band);
5377 			mutex_exit(&stp->sd_lock);
5378 			return (EMFILE);
5379 		}
5380 		if (cmd == I_RECVFD) {
5381 			struct o_strrecvfd	ostrfd;
5382 
5383 			/* check to see if uid/gid values are too large. */
5384 
5385 			if (srf->uid > (o_uid_t)USHRT_MAX ||
5386 			    srf->gid > (o_gid_t)USHRT_MAX) {
5387 				mutex_enter(&stp->sd_lock);
5388 				putback(stp, rdq, mp, mp->b_band);
5389 				mutex_exit(&stp->sd_lock);
5390 				setf(fd, NULL);	/* release fd entry */
5391 				return (EOVERFLOW);
5392 			}
5393 
5394 			ostrfd.fd = fd;
5395 			ostrfd.uid = (o_uid_t)srf->uid;
5396 			ostrfd.gid = (o_gid_t)srf->gid;
5397 
5398 			/* Null the filler bits */
5399 			for (i = 0; i < 8; i++)
5400 				ostrfd.fill[i] = 0;
5401 
5402 			error = strcopyout(&ostrfd, (void *)arg,
5403 			    sizeof (struct o_strrecvfd), copyflag);
5404 		} else {		/* I_E_RECVFD */
5405 			struct strrecvfd	strfd;
5406 
5407 			strfd.fd = fd;
5408 			strfd.uid = srf->uid;
5409 			strfd.gid = srf->gid;
5410 
5411 			/* null the filler bits */
5412 			for (i = 0; i < 8; i++)
5413 				strfd.fill[i] = 0;
5414 
5415 			error = strcopyout(&strfd, (void *)arg,
5416 			    sizeof (struct strrecvfd), copyflag);
5417 		}
5418 
5419 		if (error) {
5420 			setf(fd, NULL);	/* release fd entry */
5421 			mutex_enter(&stp->sd_lock);
5422 			putback(stp, rdq, mp, mp->b_band);
5423 			mutex_exit(&stp->sd_lock);
5424 			return (error);
5425 		}
5426 		if (audit_active) {
5427 			audit_fdrecv(fd, srf->fp);
5428 		}
5429 
5430 		/*
5431 		 * Always increment f_count since the freemsg() below will
5432 		 * always call free_passfp() which performs a closef().
5433 		 */
5434 		mutex_enter(&srf->fp->f_tlock);
5435 		srf->fp->f_count++;
5436 		mutex_exit(&srf->fp->f_tlock);
5437 		setf(fd, srf->fp);
5438 		freemsg(mp);
5439 		return (0);
5440 	}
5441 
5442 	case I_SWROPT:
5443 		/*
5444 		 * Set/clear the write options. arg is a bit
5445 		 * mask with any of the following bits set...
5446 		 * 	SNDZERO - send zero length message
5447 		 *	SNDPIPE - send sigpipe to process if
5448 		 *		sd_werror is set and process is
5449 		 *		doing a write or putmsg.
5450 		 * The new stream head write options should reflect
5451 		 * what is in arg.
5452 		 */
5453 		if (arg & ~(SNDZERO|SNDPIPE))
5454 			return (EINVAL);
5455 
5456 		mutex_enter(&stp->sd_lock);
5457 		stp->sd_wput_opt &= ~(SW_SIGPIPE|SW_SNDZERO);
5458 		if (arg & SNDZERO)
5459 			stp->sd_wput_opt |= SW_SNDZERO;
5460 		if (arg & SNDPIPE)
5461 			stp->sd_wput_opt |= SW_SIGPIPE;
5462 		mutex_exit(&stp->sd_lock);
5463 		return (0);
5464 
5465 	case I_GWROPT:
5466 	{
5467 		int wropt = 0;
5468 
5469 		if (stp->sd_wput_opt & SW_SNDZERO)
5470 			wropt |= SNDZERO;
5471 		if (stp->sd_wput_opt & SW_SIGPIPE)
5472 			wropt |= SNDPIPE;
5473 		return (strcopyout(&wropt, (void *)arg, sizeof (wropt),
5474 		    copyflag));
5475 	}
5476 
5477 	case I_LIST:
5478 		/*
5479 		 * Returns all the modules found on this stream,
5480 		 * upto the driver. If argument is NULL, return the
5481 		 * number of modules (including driver). If argument
5482 		 * is not NULL, copy the names into the structure
5483 		 * provided.
5484 		 */
5485 
5486 	{
5487 		queue_t *q;
5488 		char *qname;
5489 		int i, nmods;
5490 		struct str_mlist *mlist;
5491 		STRUCT_DECL(str_list, strlist);
5492 
5493 		if (arg == NULL) { /* Return number of modules plus driver */
5494 			if (stp->sd_vnode->v_type == VFIFO)
5495 				*rvalp = stp->sd_pushcnt;
5496 			else
5497 				*rvalp = stp->sd_pushcnt + 1;
5498 			return (0);
5499 		}
5500 
5501 		STRUCT_INIT(strlist, flag);
5502 
5503 		error = strcopyin((void *)arg, STRUCT_BUF(strlist),
5504 		    STRUCT_SIZE(strlist), copyflag);
5505 		if (error != 0)
5506 			return (error);
5507 
5508 		mlist = STRUCT_FGETP(strlist, sl_modlist);
5509 		nmods = STRUCT_FGET(strlist, sl_nmods);
5510 		if (nmods <= 0)
5511 			return (EINVAL);
5512 
5513 		claimstr(stp->sd_wrq);
5514 		q = stp->sd_wrq;
5515 		for (i = 0; i < nmods && _SAMESTR(q); i++, q = q->q_next) {
5516 			qname = Q2NAME(q->q_next);
5517 			error = strcopyout(qname, &mlist[i], strlen(qname) + 1,
5518 			    copyflag);
5519 			if (error != 0) {
5520 				releasestr(stp->sd_wrq);
5521 				return (error);
5522 			}
5523 		}
5524 		releasestr(stp->sd_wrq);
5525 		return (strcopyout(&i, (void *)arg, sizeof (int), copyflag));
5526 	}
5527 
5528 	case I_CKBAND:
5529 	{
5530 		queue_t *q;
5531 		qband_t *qbp;
5532 
5533 		if ((arg < 0) || (arg >= NBAND))
5534 			return (EINVAL);
5535 		q = _RD(stp->sd_wrq);
5536 		mutex_enter(QLOCK(q));
5537 		if (arg > (int)q->q_nband) {
5538 			*rvalp = 0;
5539 		} else {
5540 			if (arg == 0) {
5541 				if (q->q_first)
5542 					*rvalp = 1;
5543 				else
5544 					*rvalp = 0;
5545 			} else {
5546 				qbp = q->q_bandp;
5547 				while (--arg > 0)
5548 					qbp = qbp->qb_next;
5549 				if (qbp->qb_first)
5550 					*rvalp = 1;
5551 				else
5552 					*rvalp = 0;
5553 			}
5554 		}
5555 		mutex_exit(QLOCK(q));
5556 		return (0);
5557 	}
5558 
5559 	case I_GETBAND:
5560 	{
5561 		int intpri;
5562 		queue_t *q;
5563 
5564 		q = _RD(stp->sd_wrq);
5565 		mutex_enter(QLOCK(q));
5566 		mp = q->q_first;
5567 		if (!mp) {
5568 			mutex_exit(QLOCK(q));
5569 			return (ENODATA);
5570 		}
5571 		intpri = (int)mp->b_band;
5572 		error = strcopyout(&intpri, (void *)arg, sizeof (int),
5573 		    copyflag);
5574 		mutex_exit(QLOCK(q));
5575 		return (error);
5576 	}
5577 
5578 	case I_ATMARK:
5579 	{
5580 		queue_t *q;
5581 
5582 		if (arg & ~(ANYMARK|LASTMARK))
5583 			return (EINVAL);
5584 		q = _RD(stp->sd_wrq);
5585 		mutex_enter(&stp->sd_lock);
5586 		if ((stp->sd_flag & STRATMARK) && (arg == ANYMARK)) {
5587 			*rvalp = 1;
5588 		} else {
5589 			mutex_enter(QLOCK(q));
5590 			mp = q->q_first;
5591 
5592 			if (mp == NULL)
5593 				*rvalp = 0;
5594 			else if ((arg == ANYMARK) && (mp->b_flag & MSGMARK))
5595 				*rvalp = 1;
5596 			else if ((arg == LASTMARK) && (mp == stp->sd_mark))
5597 				*rvalp = 1;
5598 			else
5599 				*rvalp = 0;
5600 			mutex_exit(QLOCK(q));
5601 		}
5602 		mutex_exit(&stp->sd_lock);
5603 		return (0);
5604 	}
5605 
5606 	case I_CANPUT:
5607 	{
5608 		char band;
5609 
5610 		if ((arg < 0) || (arg >= NBAND))
5611 			return (EINVAL);
5612 		band = (char)arg;
5613 		*rvalp = bcanputnext(stp->sd_wrq, band);
5614 		return (0);
5615 	}
5616 
5617 	case I_SETCLTIME:
5618 	{
5619 		int closetime;
5620 
5621 		error = strcopyin((void *)arg, &closetime, sizeof (int),
5622 		    copyflag);
5623 		if (error)
5624 			return (error);
5625 		if (closetime < 0)
5626 			return (EINVAL);
5627 
5628 		stp->sd_closetime = closetime;
5629 		return (0);
5630 	}
5631 
5632 	case I_GETCLTIME:
5633 	{
5634 		int closetime;
5635 
5636 		closetime = stp->sd_closetime;
5637 		return (strcopyout(&closetime, (void *)arg, sizeof (int),
5638 		    copyflag));
5639 	}
5640 
5641 	case TIOCGSID:
5642 	{
5643 		pid_t sid;
5644 
5645 		mutex_enter(&stp->sd_lock);
5646 		if (stp->sd_sidp == NULL) {
5647 			mutex_exit(&stp->sd_lock);
5648 			return (ENOTTY);
5649 		}
5650 		sid = stp->sd_sidp->pid_id;
5651 		mutex_exit(&stp->sd_lock);
5652 		return (strcopyout(&sid, (void *)arg, sizeof (pid_t),
5653 		    copyflag));
5654 	}
5655 
5656 	case TIOCSPGRP:
5657 	{
5658 		pid_t pgrp;
5659 		proc_t *q;
5660 		pid_t	sid, fg_pgid, bg_pgid;
5661 
5662 		if (error = strcopyin((void *)arg, &pgrp, sizeof (pid_t),
5663 		    copyflag))
5664 			return (error);
5665 		mutex_enter(&stp->sd_lock);
5666 		mutex_enter(&pidlock);
5667 		if (stp->sd_sidp != ttoproc(curthread)->p_sessp->s_sidp) {
5668 			mutex_exit(&pidlock);
5669 			mutex_exit(&stp->sd_lock);
5670 			return (ENOTTY);
5671 		}
5672 		if (pgrp == stp->sd_pgidp->pid_id) {
5673 			mutex_exit(&pidlock);
5674 			mutex_exit(&stp->sd_lock);
5675 			return (0);
5676 		}
5677 		if (pgrp <= 0 || pgrp >= maxpid) {
5678 			mutex_exit(&pidlock);
5679 			mutex_exit(&stp->sd_lock);
5680 			return (EINVAL);
5681 		}
5682 		if ((q = pgfind(pgrp)) == NULL ||
5683 		    q->p_sessp != ttoproc(curthread)->p_sessp) {
5684 			mutex_exit(&pidlock);
5685 			mutex_exit(&stp->sd_lock);
5686 			return (EPERM);
5687 		}
5688 		sid = stp->sd_sidp->pid_id;
5689 		fg_pgid = q->p_pgrp;
5690 		bg_pgid = stp->sd_pgidp->pid_id;
5691 		CL_SET_PROCESS_GROUP(curthread, sid, bg_pgid, fg_pgid);
5692 		PID_RELE(stp->sd_pgidp);
5693 		ctty_clear_sighuped();
5694 		stp->sd_pgidp = q->p_pgidp;
5695 		PID_HOLD(stp->sd_pgidp);
5696 		mutex_exit(&pidlock);
5697 		mutex_exit(&stp->sd_lock);
5698 		return (0);
5699 	}
5700 
5701 	case TIOCGPGRP:
5702 	{
5703 		pid_t pgrp;
5704 
5705 		mutex_enter(&stp->sd_lock);
5706 		if (stp->sd_sidp == NULL) {
5707 			mutex_exit(&stp->sd_lock);
5708 			return (ENOTTY);
5709 		}
5710 		pgrp = stp->sd_pgidp->pid_id;
5711 		mutex_exit(&stp->sd_lock);
5712 		return (strcopyout(&pgrp, (void *)arg, sizeof (pid_t),
5713 		    copyflag));
5714 	}
5715 
5716 	case TIOCSCTTY:
5717 	{
5718 		return (strctty(stp));
5719 	}
5720 
5721 	case TIOCNOTTY:
5722 	{
5723 		/* freectty() always assumes curproc. */
5724 		if (freectty(B_FALSE) != 0)
5725 			return (0);
5726 		return (ENOTTY);
5727 	}
5728 
5729 	case FIONBIO:
5730 	case FIOASYNC:
5731 		return (0);	/* handled by the upper layer */
5732 	}
5733 }
5734 
5735 /*
5736  * Custom free routine used for M_PASSFP messages.
5737  */
5738 static void
5739 free_passfp(struct k_strrecvfd *srf)
5740 {
5741 	(void) closef(srf->fp);
5742 	kmem_free(srf, sizeof (struct k_strrecvfd) + sizeof (frtn_t));
5743 }
5744 
5745 /* ARGSUSED */
5746 int
5747 do_sendfp(struct stdata *stp, struct file *fp, struct cred *cr)
5748 {
5749 	queue_t *qp, *nextqp;
5750 	struct k_strrecvfd *srf;
5751 	mblk_t *mp;
5752 	frtn_t *frtnp;
5753 	size_t bufsize;
5754 	queue_t	*mate = NULL;
5755 	syncq_t	*sq = NULL;
5756 	int retval = 0;
5757 
5758 	if (stp->sd_flag & STRHUP)
5759 		return (ENXIO);
5760 
5761 	claimstr(stp->sd_wrq);
5762 
5763 	/* Fastpath, we have a pipe, and we are already mated, use it. */
5764 	if (STRMATED(stp)) {
5765 		qp = _RD(stp->sd_mate->sd_wrq);
5766 		claimstr(qp);
5767 		mate = qp;
5768 	} else { /* Not already mated. */
5769 
5770 		/*
5771 		 * Walk the stream to the end of this one.
5772 		 * assumes that the claimstr() will prevent
5773 		 * plumbing between the stream head and the
5774 		 * driver from changing
5775 		 */
5776 		qp = stp->sd_wrq;
5777 
5778 		/*
5779 		 * Loop until we reach the end of this stream.
5780 		 * On completion, qp points to the write queue
5781 		 * at the end of the stream, or the read queue
5782 		 * at the stream head if this is a fifo.
5783 		 */
5784 		while (((qp = qp->q_next) != NULL) && _SAMESTR(qp))
5785 			;
5786 
5787 		/*
5788 		 * Just in case we get a q_next which is NULL, but
5789 		 * not at the end of the stream.  This is actually
5790 		 * broken, so we set an assert to catch it in
5791 		 * debug, and set an error and return if not debug.
5792 		 */
5793 		ASSERT(qp);
5794 		if (qp == NULL) {
5795 			releasestr(stp->sd_wrq);
5796 			return (EINVAL);
5797 		}
5798 
5799 		/*
5800 		 * Enter the syncq for the driver, so (hopefully)
5801 		 * the queue values will not change on us.
5802 		 * XXXX - This will only prevent the race IFF only
5803 		 *   the write side modifies the q_next member, and
5804 		 *   the put procedure is protected by at least
5805 		 *   MT_PERQ.
5806 		 */
5807 		if ((sq = qp->q_syncq) != NULL)
5808 			entersq(sq, SQ_PUT);
5809 
5810 		/* Now get the q_next value from this qp. */
5811 		nextqp = qp->q_next;
5812 
5813 		/*
5814 		 * If nextqp exists and the other stream is different
5815 		 * from this one claim the stream, set the mate, and
5816 		 * get the read queue at the stream head of the other
5817 		 * stream.  Assumes that nextqp was at least valid when
5818 		 * we got it.  Hopefully the entersq of the driver
5819 		 * will prevent it from changing on us.
5820 		 */
5821 		if ((nextqp != NULL) && (STREAM(nextqp) != stp)) {
5822 			ASSERT(qp->q_qinfo->qi_srvp);
5823 			ASSERT(_OTHERQ(qp)->q_qinfo->qi_srvp);
5824 			ASSERT(_OTHERQ(qp->q_next)->q_qinfo->qi_srvp);
5825 			claimstr(nextqp);
5826 
5827 			/* Make sure we still have a q_next */
5828 			if (nextqp != qp->q_next) {
5829 				releasestr(stp->sd_wrq);
5830 				releasestr(nextqp);
5831 				return (EINVAL);
5832 			}
5833 
5834 			qp = _RD(STREAM(nextqp)->sd_wrq);
5835 			mate = qp;
5836 		}
5837 		/* If we entered the synq above, leave it. */
5838 		if (sq != NULL)
5839 			leavesq(sq, SQ_PUT);
5840 	} /*  STRMATED(STP)  */
5841 
5842 	/* XXX prevents substitution of the ops vector */
5843 	if (qp->q_qinfo != &strdata && qp->q_qinfo != &fifo_strdata) {
5844 		retval = EINVAL;
5845 		goto out;
5846 	}
5847 
5848 	if (qp->q_flag & QFULL) {
5849 		retval = EAGAIN;
5850 		goto out;
5851 	}
5852 
5853 	/*
5854 	 * Since M_PASSFP messages include a file descriptor, we use
5855 	 * esballoc() and specify a custom free routine (free_passfp()) that
5856 	 * will close the descriptor as part of freeing the message.  For
5857 	 * convenience, we stash the frtn_t right after the data block.
5858 	 */
5859 	bufsize = sizeof (struct k_strrecvfd) + sizeof (frtn_t);
5860 	srf = kmem_alloc(bufsize, KM_NOSLEEP);
5861 	if (srf == NULL) {
5862 		retval = EAGAIN;
5863 		goto out;
5864 	}
5865 
5866 	frtnp = (frtn_t *)(srf + 1);
5867 	frtnp->free_arg = (caddr_t)srf;
5868 	frtnp->free_func = free_passfp;
5869 
5870 	mp = esballoc((uchar_t *)srf, bufsize, BPRI_MED, frtnp);
5871 	if (mp == NULL) {
5872 		kmem_free(srf, bufsize);
5873 		retval = EAGAIN;
5874 		goto out;
5875 	}
5876 	mp->b_wptr += sizeof (struct k_strrecvfd);
5877 	mp->b_datap->db_type = M_PASSFP;
5878 
5879 	srf->fp = fp;
5880 	srf->uid = crgetuid(curthread->t_cred);
5881 	srf->gid = crgetgid(curthread->t_cred);
5882 	mutex_enter(&fp->f_tlock);
5883 	fp->f_count++;
5884 	mutex_exit(&fp->f_tlock);
5885 
5886 	put(qp, mp);
5887 out:
5888 	releasestr(stp->sd_wrq);
5889 	if (mate)
5890 		releasestr(mate);
5891 	return (retval);
5892 }
5893 
5894 /*
5895  * Send an ioctl message downstream and wait for acknowledgement.
5896  * flags may be set to either U_TO_K or K_TO_K and a combination
5897  * of STR_NOERROR or STR_NOSIG
5898  * STR_NOSIG: Signals are essentially ignored or held and have
5899  *	no effect for the duration of the call.
5900  * STR_NOERROR: Ignores stream head read, write and hup errors.
5901  *	Additionally, if an existing ioctl times out, it is assumed
5902  *	lost and and this ioctl will continue as if the previous ioctl had
5903  *	finished.  ETIME may be returned if this ioctl times out (i.e.
5904  *	ic_timout is not INFTIM).  Non-stream head errors may be returned if
5905  *	the ioc_error indicates that the driver/module had problems,
5906  *	an EFAULT was found when accessing user data, a lack of
5907  * 	resources, etc.
5908  */
5909 int
5910 strdoioctl(
5911 	struct stdata *stp,
5912 	struct strioctl *strioc,
5913 	int fflags,		/* file flags with model info */
5914 	int flag,
5915 	cred_t *crp,
5916 	int *rvalp)
5917 {
5918 	mblk_t *bp;
5919 	struct iocblk *iocbp;
5920 	struct copyreq *reqp;
5921 	struct copyresp *resp;
5922 	int id;
5923 	int transparent = 0;
5924 	int error = 0;
5925 	int len = 0;
5926 	caddr_t taddr;
5927 	int copyflag = (flag & (U_TO_K | K_TO_K));
5928 	int sigflag = (flag & STR_NOSIG);
5929 	int errs;
5930 	uint_t waitflags;
5931 
5932 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
5933 	ASSERT((fflags & FMODELS) != 0);
5934 
5935 	TRACE_2(TR_FAC_STREAMS_FR,
5936 	    TR_STRDOIOCTL,
5937 	    "strdoioctl:stp %p strioc %p", stp, strioc);
5938 	if (strioc->ic_len == TRANSPARENT) {	/* send arg in M_DATA block */
5939 		transparent = 1;
5940 		strioc->ic_len = sizeof (intptr_t);
5941 	}
5942 
5943 	if (strioc->ic_len < 0 || (strmsgsz > 0 && strioc->ic_len > strmsgsz))
5944 		return (EINVAL);
5945 
5946 	if ((bp = allocb_cred_wait(sizeof (union ioctypes), sigflag, &error,
5947 	    crp, curproc->p_pid)) == NULL)
5948 			return (error);
5949 
5950 	bzero(bp->b_wptr, sizeof (union ioctypes));
5951 
5952 	iocbp = (struct iocblk *)bp->b_wptr;
5953 	iocbp->ioc_count = strioc->ic_len;
5954 	iocbp->ioc_cmd = strioc->ic_cmd;
5955 	iocbp->ioc_flag = (fflags & FMODELS);
5956 
5957 	crhold(crp);
5958 	iocbp->ioc_cr = crp;
5959 	DB_TYPE(bp) = M_IOCTL;
5960 	bp->b_wptr += sizeof (struct iocblk);
5961 
5962 	if (flag & STR_NOERROR)
5963 		errs = STPLEX;
5964 	else
5965 		errs = STRHUP|STRDERR|STWRERR|STPLEX;
5966 
5967 	/*
5968 	 * If there is data to copy into ioctl block, do so.
5969 	 */
5970 	if (iocbp->ioc_count > 0) {
5971 		if (transparent)
5972 			/*
5973 			 * Note: STR_NOERROR does not have an effect
5974 			 * in putiocd()
5975 			 */
5976 			id = K_TO_K | sigflag;
5977 		else
5978 			id = flag;
5979 		if ((error = putiocd(bp, strioc->ic_dp, id, crp)) != 0) {
5980 			freemsg(bp);
5981 			crfree(crp);
5982 			return (error);
5983 		}
5984 
5985 		/*
5986 		 * We could have slept copying in user pages.
5987 		 * Recheck the stream head state (the other end
5988 		 * of a pipe could have gone away).
5989 		 */
5990 		if (stp->sd_flag & errs) {
5991 			mutex_enter(&stp->sd_lock);
5992 			error = strgeterr(stp, errs, 0);
5993 			mutex_exit(&stp->sd_lock);
5994 			if (error != 0) {
5995 				freemsg(bp);
5996 				crfree(crp);
5997 				return (error);
5998 			}
5999 		}
6000 	}
6001 	if (transparent)
6002 		iocbp->ioc_count = TRANSPARENT;
6003 
6004 	/*
6005 	 * Block for up to STRTIMOUT milliseconds if there is an outstanding
6006 	 * ioctl for this stream already running.  All processes
6007 	 * sleeping here will be awakened as a result of an ACK
6008 	 * or NAK being received for the outstanding ioctl, or
6009 	 * as a result of the timer expiring on the outstanding
6010 	 * ioctl (a failure), or as a result of any waiting
6011 	 * process's timer expiring (also a failure).
6012 	 */
6013 
6014 	error = 0;
6015 	mutex_enter(&stp->sd_lock);
6016 	while (stp->sd_flag & (IOCWAIT | IOCWAITNE)) {
6017 		clock_t cv_rval;
6018 
6019 		TRACE_0(TR_FAC_STREAMS_FR,
6020 		    TR_STRDOIOCTL_WAIT,
6021 		    "strdoioctl sleeps - IOCWAIT");
6022 		cv_rval = str_cv_wait(&stp->sd_iocmonitor, &stp->sd_lock,
6023 		    STRTIMOUT, sigflag);
6024 		if (cv_rval <= 0) {
6025 			if (cv_rval == 0) {
6026 				error = EINTR;
6027 			} else {
6028 				if (flag & STR_NOERROR) {
6029 					/*
6030 					 * Terminating current ioctl in
6031 					 * progress -- assume it got lost and
6032 					 * wake up the other thread so that the
6033 					 * operation completes.
6034 					 */
6035 					if (!(stp->sd_flag & IOCWAITNE)) {
6036 						stp->sd_flag |= IOCWAITNE;
6037 						cv_broadcast(&stp->sd_monitor);
6038 					}
6039 					/*
6040 					 * Otherwise, there's a running
6041 					 * STR_NOERROR -- we have no choice
6042 					 * here but to wait forever (or until
6043 					 * interrupted).
6044 					 */
6045 				} else {
6046 					/*
6047 					 * pending ioctl has caused
6048 					 * us to time out
6049 					 */
6050 					error = ETIME;
6051 				}
6052 			}
6053 		} else if ((stp->sd_flag & errs)) {
6054 			error = strgeterr(stp, errs, 0);
6055 		}
6056 		if (error) {
6057 			mutex_exit(&stp->sd_lock);
6058 			freemsg(bp);
6059 			crfree(crp);
6060 			return (error);
6061 		}
6062 	}
6063 
6064 	/*
6065 	 * Have control of ioctl mechanism.
6066 	 * Send down ioctl packet and wait for response.
6067 	 */
6068 	if (stp->sd_iocblk != (mblk_t *)-1) {
6069 		freemsg(stp->sd_iocblk);
6070 	}
6071 	stp->sd_iocblk = NULL;
6072 
6073 	/*
6074 	 * If this is marked with 'noerror' (internal; mostly
6075 	 * I_{P,}{UN,}LINK), then make sure nobody else is able to get
6076 	 * in here by setting IOCWAITNE.
6077 	 */
6078 	waitflags = IOCWAIT;
6079 	if (flag & STR_NOERROR)
6080 		waitflags |= IOCWAITNE;
6081 
6082 	stp->sd_flag |= waitflags;
6083 
6084 	/*
6085 	 * Assign sequence number.
6086 	 */
6087 	iocbp->ioc_id = stp->sd_iocid = getiocseqno();
6088 
6089 	mutex_exit(&stp->sd_lock);
6090 
6091 	TRACE_1(TR_FAC_STREAMS_FR,
6092 	    TR_STRDOIOCTL_PUT, "strdoioctl put: stp %p", stp);
6093 	stream_willservice(stp);
6094 	putnext(stp->sd_wrq, bp);
6095 	stream_runservice(stp);
6096 
6097 	/*
6098 	 * Timed wait for acknowledgment.  The wait time is limited by the
6099 	 * timeout value, which must be a positive integer (number of
6100 	 * milliseconds) to wait, or 0 (use default value of STRTIMOUT
6101 	 * milliseconds), or -1 (wait forever).  This will be awakened
6102 	 * either by an ACK/NAK message arriving, the timer expiring, or
6103 	 * the timer expiring on another ioctl waiting for control of the
6104 	 * mechanism.
6105 	 */
6106 waitioc:
6107 	mutex_enter(&stp->sd_lock);
6108 
6109 
6110 	/*
6111 	 * If the reply has already arrived, don't sleep.  If awakened from
6112 	 * the sleep, fail only if the reply has not arrived by then.
6113 	 * Otherwise, process the reply.
6114 	 */
6115 	while (!stp->sd_iocblk) {
6116 		clock_t cv_rval;
6117 
6118 		if (stp->sd_flag & errs) {
6119 			error = strgeterr(stp, errs, 0);
6120 			if (error != 0) {
6121 				stp->sd_flag &= ~waitflags;
6122 				cv_broadcast(&stp->sd_iocmonitor);
6123 				mutex_exit(&stp->sd_lock);
6124 				crfree(crp);
6125 				return (error);
6126 			}
6127 		}
6128 
6129 		TRACE_0(TR_FAC_STREAMS_FR,
6130 		    TR_STRDOIOCTL_WAIT2,
6131 		    "strdoioctl sleeps awaiting reply");
6132 		ASSERT(error == 0);
6133 
6134 		cv_rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock,
6135 		    (strioc->ic_timout ?
6136 		    strioc->ic_timout * 1000 : STRTIMOUT), sigflag);
6137 
6138 		/*
6139 		 * There are four possible cases here: interrupt, timeout,
6140 		 * wakeup by IOCWAITNE (above), or wakeup by strrput_nondata (a
6141 		 * valid M_IOCTL reply).
6142 		 *
6143 		 * If we've been awakened by a STR_NOERROR ioctl on some other
6144 		 * thread, then sd_iocblk will still be NULL, and IOCWAITNE
6145 		 * will be set.  Pretend as if we just timed out.  Note that
6146 		 * this other thread waited at least STRTIMOUT before trying to
6147 		 * awaken our thread, so this is indistinguishable (even for
6148 		 * INFTIM) from the case where we failed with ETIME waiting on
6149 		 * IOCWAIT in the prior loop.
6150 		 */
6151 		if (cv_rval > 0 && !(flag & STR_NOERROR) &&
6152 		    stp->sd_iocblk == NULL && (stp->sd_flag & IOCWAITNE)) {
6153 			cv_rval = -1;
6154 		}
6155 
6156 		/*
6157 		 * note: STR_NOERROR does not protect
6158 		 * us here.. use ic_timout < 0
6159 		 */
6160 		if (cv_rval <= 0) {
6161 			if (cv_rval == 0) {
6162 				error = EINTR;
6163 			} else {
6164 				error =  ETIME;
6165 			}
6166 			/*
6167 			 * A message could have come in after we were scheduled
6168 			 * but before we were actually run.
6169 			 */
6170 			bp = stp->sd_iocblk;
6171 			stp->sd_iocblk = NULL;
6172 			if (bp != NULL) {
6173 				if ((bp->b_datap->db_type == M_COPYIN) ||
6174 				    (bp->b_datap->db_type == M_COPYOUT)) {
6175 					mutex_exit(&stp->sd_lock);
6176 					if (bp->b_cont) {
6177 						freemsg(bp->b_cont);
6178 						bp->b_cont = NULL;
6179 					}
6180 					bp->b_datap->db_type = M_IOCDATA;
6181 					bp->b_wptr = bp->b_rptr +
6182 					    sizeof (struct copyresp);
6183 					resp = (struct copyresp *)bp->b_rptr;
6184 					resp->cp_rval =
6185 					    (caddr_t)1; /* failure */
6186 					stream_willservice(stp);
6187 					putnext(stp->sd_wrq, bp);
6188 					stream_runservice(stp);
6189 					mutex_enter(&stp->sd_lock);
6190 				} else {
6191 					freemsg(bp);
6192 				}
6193 			}
6194 			stp->sd_flag &= ~waitflags;
6195 			cv_broadcast(&stp->sd_iocmonitor);
6196 			mutex_exit(&stp->sd_lock);
6197 			crfree(crp);
6198 			return (error);
6199 		}
6200 	}
6201 	bp = stp->sd_iocblk;
6202 	/*
6203 	 * Note: it is strictly impossible to get here with sd_iocblk set to
6204 	 * -1.  This is because the initial loop above doesn't allow any new
6205 	 * ioctls into the fray until all others have passed this point.
6206 	 */
6207 	ASSERT(bp != NULL && bp != (mblk_t *)-1);
6208 	TRACE_1(TR_FAC_STREAMS_FR,
6209 	    TR_STRDOIOCTL_ACK, "strdoioctl got reply: bp %p", bp);
6210 	if ((bp->b_datap->db_type == M_IOCACK) ||
6211 	    (bp->b_datap->db_type == M_IOCNAK)) {
6212 		/* for detection of duplicate ioctl replies */
6213 		stp->sd_iocblk = (mblk_t *)-1;
6214 		stp->sd_flag &= ~waitflags;
6215 		cv_broadcast(&stp->sd_iocmonitor);
6216 		mutex_exit(&stp->sd_lock);
6217 	} else {
6218 		/*
6219 		 * flags not cleared here because we're still doing
6220 		 * copy in/out for ioctl.
6221 		 */
6222 		stp->sd_iocblk = NULL;
6223 		mutex_exit(&stp->sd_lock);
6224 	}
6225 
6226 
6227 	/*
6228 	 * Have received acknowledgment.
6229 	 */
6230 
6231 	switch (bp->b_datap->db_type) {
6232 	case M_IOCACK:
6233 		/*
6234 		 * Positive ack.
6235 		 */
6236 		iocbp = (struct iocblk *)bp->b_rptr;
6237 
6238 		/*
6239 		 * Set error if indicated.
6240 		 */
6241 		if (iocbp->ioc_error) {
6242 			error = iocbp->ioc_error;
6243 			break;
6244 		}
6245 
6246 		/*
6247 		 * Set return value.
6248 		 */
6249 		*rvalp = iocbp->ioc_rval;
6250 
6251 		/*
6252 		 * Data may have been returned in ACK message (ioc_count > 0).
6253 		 * If so, copy it out to the user's buffer.
6254 		 */
6255 		if (iocbp->ioc_count && !transparent) {
6256 			if (error = getiocd(bp, strioc->ic_dp, copyflag))
6257 				break;
6258 		}
6259 		if (!transparent) {
6260 			if (len)	/* an M_COPYOUT was used with I_STR */
6261 				strioc->ic_len = len;
6262 			else
6263 				strioc->ic_len = (int)iocbp->ioc_count;
6264 		}
6265 		break;
6266 
6267 	case M_IOCNAK:
6268 		/*
6269 		 * Negative ack.
6270 		 *
6271 		 * The only thing to do is set error as specified
6272 		 * in neg ack packet.
6273 		 */
6274 		iocbp = (struct iocblk *)bp->b_rptr;
6275 
6276 		error = (iocbp->ioc_error ? iocbp->ioc_error : EINVAL);
6277 		break;
6278 
6279 	case M_COPYIN:
6280 		/*
6281 		 * Driver or module has requested user ioctl data.
6282 		 */
6283 		reqp = (struct copyreq *)bp->b_rptr;
6284 
6285 		/*
6286 		 * M_COPYIN should *never* have a message attached, though
6287 		 * it's harmless if it does -- thus, panic on a DEBUG
6288 		 * kernel and just free it on a non-DEBUG build.
6289 		 */
6290 		ASSERT(bp->b_cont == NULL);
6291 		if (bp->b_cont != NULL) {
6292 			freemsg(bp->b_cont);
6293 			bp->b_cont = NULL;
6294 		}
6295 
6296 		error = putiocd(bp, reqp->cq_addr, flag, crp);
6297 		if (error && bp->b_cont) {
6298 			freemsg(bp->b_cont);
6299 			bp->b_cont = NULL;
6300 		}
6301 
6302 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6303 		bp->b_datap->db_type = M_IOCDATA;
6304 
6305 		mblk_setcred(bp, crp, curproc->p_pid);
6306 		resp = (struct copyresp *)bp->b_rptr;
6307 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6308 		resp->cp_flag = (fflags & FMODELS);
6309 
6310 		stream_willservice(stp);
6311 		putnext(stp->sd_wrq, bp);
6312 		stream_runservice(stp);
6313 
6314 		if (error) {
6315 			mutex_enter(&stp->sd_lock);
6316 			stp->sd_flag &= ~waitflags;
6317 			cv_broadcast(&stp->sd_iocmonitor);
6318 			mutex_exit(&stp->sd_lock);
6319 			crfree(crp);
6320 			return (error);
6321 		}
6322 
6323 		goto waitioc;
6324 
6325 	case M_COPYOUT:
6326 		/*
6327 		 * Driver or module has ioctl data for a user.
6328 		 */
6329 		reqp = (struct copyreq *)bp->b_rptr;
6330 		ASSERT(bp->b_cont != NULL);
6331 
6332 		/*
6333 		 * Always (transparent or non-transparent )
6334 		 * use the address specified in the request
6335 		 */
6336 		taddr = reqp->cq_addr;
6337 		if (!transparent)
6338 			len = (int)reqp->cq_size;
6339 
6340 		/* copyout data to the provided address */
6341 		error = getiocd(bp, taddr, copyflag);
6342 
6343 		freemsg(bp->b_cont);
6344 		bp->b_cont = NULL;
6345 
6346 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6347 		bp->b_datap->db_type = M_IOCDATA;
6348 
6349 		mblk_setcred(bp, crp, curproc->p_pid);
6350 		resp = (struct copyresp *)bp->b_rptr;
6351 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6352 		resp->cp_flag = (fflags & FMODELS);
6353 
6354 		stream_willservice(stp);
6355 		putnext(stp->sd_wrq, bp);
6356 		stream_runservice(stp);
6357 
6358 		if (error) {
6359 			mutex_enter(&stp->sd_lock);
6360 			stp->sd_flag &= ~waitflags;
6361 			cv_broadcast(&stp->sd_iocmonitor);
6362 			mutex_exit(&stp->sd_lock);
6363 			crfree(crp);
6364 			return (error);
6365 		}
6366 		goto waitioc;
6367 
6368 	default:
6369 		ASSERT(0);
6370 		mutex_enter(&stp->sd_lock);
6371 		stp->sd_flag &= ~waitflags;
6372 		cv_broadcast(&stp->sd_iocmonitor);
6373 		mutex_exit(&stp->sd_lock);
6374 		break;
6375 	}
6376 
6377 	freemsg(bp);
6378 	crfree(crp);
6379 	return (error);
6380 }
6381 
6382 /*
6383  * Send an M_CMD message downstream and wait for a reply.  This is a ptools
6384  * special used to retrieve information from modules/drivers a stream without
6385  * being subjected to flow control or interfering with pending messages on the
6386  * stream (e.g. an ioctl in flight).
6387  */
6388 int
6389 strdocmd(struct stdata *stp, struct strcmd *scp, cred_t *crp)
6390 {
6391 	mblk_t *mp;
6392 	struct cmdblk *cmdp;
6393 	int error = 0;
6394 	int errs = STRHUP|STRDERR|STWRERR|STPLEX;
6395 	clock_t rval, timeout = STRTIMOUT;
6396 
6397 	if (scp->sc_len < 0 || scp->sc_len > sizeof (scp->sc_buf) ||
6398 	    scp->sc_timeout < -1)
6399 		return (EINVAL);
6400 
6401 	if (scp->sc_timeout > 0)
6402 		timeout = scp->sc_timeout * MILLISEC;
6403 
6404 	if ((mp = allocb_cred(sizeof (struct cmdblk), crp,
6405 	    curproc->p_pid)) == NULL)
6406 		return (ENOMEM);
6407 
6408 	crhold(crp);
6409 
6410 	cmdp = (struct cmdblk *)mp->b_wptr;
6411 	cmdp->cb_cr = crp;
6412 	cmdp->cb_cmd = scp->sc_cmd;
6413 	cmdp->cb_len = scp->sc_len;
6414 	cmdp->cb_error = 0;
6415 	mp->b_wptr += sizeof (struct cmdblk);
6416 
6417 	DB_TYPE(mp) = M_CMD;
6418 	DB_CPID(mp) = curproc->p_pid;
6419 
6420 	/*
6421 	 * Copy in the payload.
6422 	 */
6423 	if (cmdp->cb_len > 0) {
6424 		mp->b_cont = allocb_cred(sizeof (scp->sc_buf), crp,
6425 		    curproc->p_pid);
6426 		if (mp->b_cont == NULL) {
6427 			error = ENOMEM;
6428 			goto out;
6429 		}
6430 
6431 		/* cb_len comes from sc_len, which has already been checked */
6432 		ASSERT(cmdp->cb_len <= sizeof (scp->sc_buf));
6433 		(void) bcopy(scp->sc_buf, mp->b_cont->b_wptr, cmdp->cb_len);
6434 		mp->b_cont->b_wptr += cmdp->cb_len;
6435 		DB_CPID(mp->b_cont) = curproc->p_pid;
6436 	}
6437 
6438 	/*
6439 	 * Since this mechanism is strictly for ptools, and since only one
6440 	 * process can be grabbed at a time, we simply fail if there's
6441 	 * currently an operation pending.
6442 	 */
6443 	mutex_enter(&stp->sd_lock);
6444 	if (stp->sd_flag & STRCMDWAIT) {
6445 		mutex_exit(&stp->sd_lock);
6446 		error = EBUSY;
6447 		goto out;
6448 	}
6449 	stp->sd_flag |= STRCMDWAIT;
6450 	ASSERT(stp->sd_cmdblk == NULL);
6451 	mutex_exit(&stp->sd_lock);
6452 
6453 	putnext(stp->sd_wrq, mp);
6454 	mp = NULL;
6455 
6456 	/*
6457 	 * Timed wait for acknowledgment.  If the reply has already arrived,
6458 	 * don't sleep.  If awakened from the sleep, fail only if the reply
6459 	 * has not arrived by then.  Otherwise, process the reply.
6460 	 */
6461 	mutex_enter(&stp->sd_lock);
6462 	while (stp->sd_cmdblk == NULL) {
6463 		if (stp->sd_flag & errs) {
6464 			if ((error = strgeterr(stp, errs, 0)) != 0)
6465 				goto waitout;
6466 		}
6467 
6468 		rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock, timeout, 0);
6469 		if (stp->sd_cmdblk != NULL)
6470 			break;
6471 
6472 		if (rval <= 0) {
6473 			error = (rval == 0) ? EINTR : ETIME;
6474 			goto waitout;
6475 		}
6476 	}
6477 
6478 	/*
6479 	 * We received a reply.
6480 	 */
6481 	mp = stp->sd_cmdblk;
6482 	stp->sd_cmdblk = NULL;
6483 	ASSERT(mp != NULL && DB_TYPE(mp) == M_CMD);
6484 	ASSERT(stp->sd_flag & STRCMDWAIT);
6485 	stp->sd_flag &= ~STRCMDWAIT;
6486 	mutex_exit(&stp->sd_lock);
6487 
6488 	cmdp = (struct cmdblk *)mp->b_rptr;
6489 	if ((error = cmdp->cb_error) != 0)
6490 		goto out;
6491 
6492 	/*
6493 	 * Data may have been returned in the reply (cb_len > 0).
6494 	 * If so, copy it out to the user's buffer.
6495 	 */
6496 	if (cmdp->cb_len > 0) {
6497 		if (mp->b_cont == NULL || MBLKL(mp->b_cont) < cmdp->cb_len) {
6498 			error = EPROTO;
6499 			goto out;
6500 		}
6501 
6502 		cmdp->cb_len = MIN(cmdp->cb_len, sizeof (scp->sc_buf));
6503 		(void) bcopy(mp->b_cont->b_rptr, scp->sc_buf, cmdp->cb_len);
6504 	}
6505 	scp->sc_len = cmdp->cb_len;
6506 out:
6507 	freemsg(mp);
6508 	crfree(crp);
6509 	return (error);
6510 waitout:
6511 	ASSERT(stp->sd_cmdblk == NULL);
6512 	stp->sd_flag &= ~STRCMDWAIT;
6513 	mutex_exit(&stp->sd_lock);
6514 	crfree(crp);
6515 	return (error);
6516 }
6517 
6518 /*
6519  * For the SunOS keyboard driver.
6520  * Return the next available "ioctl" sequence number.
6521  * Exported, so that streams modules can send "ioctl" messages
6522  * downstream from their open routine.
6523  */
6524 int
6525 getiocseqno(void)
6526 {
6527 	int	i;
6528 
6529 	mutex_enter(&strresources);
6530 	i = ++ioc_id;
6531 	mutex_exit(&strresources);
6532 	return (i);
6533 }
6534 
6535 /*
6536  * Get the next message from the read queue.  If the message is
6537  * priority, STRPRI will have been set by strrput().  This flag
6538  * should be reset only when the entire message at the front of the
6539  * queue as been consumed.
6540  *
6541  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
6542  */
6543 int
6544 strgetmsg(
6545 	struct vnode *vp,
6546 	struct strbuf *mctl,
6547 	struct strbuf *mdata,
6548 	unsigned char *prip,
6549 	int *flagsp,
6550 	int fmode,
6551 	rval_t *rvp)
6552 {
6553 	struct stdata *stp;
6554 	mblk_t *bp, *nbp;
6555 	mblk_t *savemp = NULL;
6556 	mblk_t *savemptail = NULL;
6557 	uint_t old_sd_flag;
6558 	int flg;
6559 	int more = 0;
6560 	int error = 0;
6561 	char first = 1;
6562 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
6563 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
6564 	unsigned char pri = 0;
6565 	queue_t *q;
6566 	int	pr = 0;			/* Partial read successful */
6567 	struct uio uios;
6568 	struct uio *uiop = &uios;
6569 	struct iovec iovs;
6570 	unsigned char type;
6571 
6572 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRGETMSG_ENTER,
6573 	    "strgetmsg:%p", vp);
6574 
6575 	ASSERT(vp->v_stream);
6576 	stp = vp->v_stream;
6577 	rvp->r_val1 = 0;
6578 
6579 	mutex_enter(&stp->sd_lock);
6580 
6581 	if ((error = i_straccess(stp, JCREAD)) != 0) {
6582 		mutex_exit(&stp->sd_lock);
6583 		return (error);
6584 	}
6585 
6586 	if (stp->sd_flag & (STRDERR|STPLEX)) {
6587 		error = strgeterr(stp, STRDERR|STPLEX, 0);
6588 		if (error != 0) {
6589 			mutex_exit(&stp->sd_lock);
6590 			return (error);
6591 		}
6592 	}
6593 	mutex_exit(&stp->sd_lock);
6594 
6595 	switch (*flagsp) {
6596 	case MSG_HIPRI:
6597 		if (*prip != 0)
6598 			return (EINVAL);
6599 		break;
6600 
6601 	case MSG_ANY:
6602 	case MSG_BAND:
6603 		break;
6604 
6605 	default:
6606 		return (EINVAL);
6607 	}
6608 	/*
6609 	 * Setup uio and iov for data part
6610 	 */
6611 	iovs.iov_base = mdata->buf;
6612 	iovs.iov_len = mdata->maxlen;
6613 	uios.uio_iov = &iovs;
6614 	uios.uio_iovcnt = 1;
6615 	uios.uio_loffset = 0;
6616 	uios.uio_segflg = UIO_USERSPACE;
6617 	uios.uio_fmode = 0;
6618 	uios.uio_extflg = UIO_COPY_CACHED;
6619 	uios.uio_resid = mdata->maxlen;
6620 	uios.uio_offset = 0;
6621 
6622 	q = _RD(stp->sd_wrq);
6623 	mutex_enter(&stp->sd_lock);
6624 	old_sd_flag = stp->sd_flag;
6625 	mark = 0;
6626 	for (;;) {
6627 		int done = 0;
6628 		mblk_t *q_first = q->q_first;
6629 
6630 		/*
6631 		 * Get the next message of appropriate priority
6632 		 * from the stream head.  If the caller is interested
6633 		 * in band or hipri messages, then they should already
6634 		 * be enqueued at the stream head.  On the other hand
6635 		 * if the caller wants normal (band 0) messages, they
6636 		 * might be deferred in a synchronous stream and they
6637 		 * will need to be pulled up.
6638 		 *
6639 		 * After we have dequeued a message, we might find that
6640 		 * it was a deferred M_SIG that was enqueued at the
6641 		 * stream head.  It must now be posted as part of the
6642 		 * read by calling strsignal_nolock().
6643 		 *
6644 		 * Also note that strrput does not enqueue an M_PCSIG,
6645 		 * and there cannot be more than one hipri message,
6646 		 * so there was no need to have the M_PCSIG case.
6647 		 *
6648 		 * At some time it might be nice to try and wrap the
6649 		 * functionality of kstrgetmsg() and strgetmsg() into
6650 		 * a common routine so to reduce the amount of replicated
6651 		 * code (since they are extremely similar).
6652 		 */
6653 		if (!(*flagsp & (MSG_HIPRI|MSG_BAND))) {
6654 			/* Asking for normal, band0 data */
6655 			bp = strget(stp, q, uiop, first, &error);
6656 			ASSERT(MUTEX_HELD(&stp->sd_lock));
6657 			if (bp != NULL) {
6658 				ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
6659 				if (DB_TYPE(bp) == M_SIG) {
6660 					strsignal_nolock(stp, *bp->b_rptr,
6661 					    bp->b_band);
6662 					freemsg(bp);
6663 					continue;
6664 				} else {
6665 					break;
6666 				}
6667 			}
6668 			if (error != 0)
6669 				goto getmout;
6670 
6671 		/*
6672 		 * We can't depend on the value of STRPRI here because
6673 		 * the stream head may be in transit. Therefore, we
6674 		 * must look at the type of the first message to
6675 		 * determine if a high priority messages is waiting
6676 		 */
6677 		} else if ((*flagsp & MSG_HIPRI) && q_first != NULL &&
6678 		    DB_TYPE(q_first) >= QPCTL &&
6679 		    (bp = getq_noenab(q, 0)) != NULL) {
6680 			/* Asked for HIPRI and got one */
6681 			ASSERT(DB_TYPE(bp) >= QPCTL);
6682 			break;
6683 		} else if ((*flagsp & MSG_BAND) && q_first != NULL &&
6684 		    ((q_first->b_band >= *prip) || DB_TYPE(q_first) >= QPCTL) &&
6685 		    (bp = getq_noenab(q, 0)) != NULL) {
6686 			/*
6687 			 * Asked for at least band "prip" and got either at
6688 			 * least that band or a hipri message.
6689 			 */
6690 			ASSERT(bp->b_band >= *prip || DB_TYPE(bp) >= QPCTL);
6691 			if (DB_TYPE(bp) == M_SIG) {
6692 				strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
6693 				freemsg(bp);
6694 				continue;
6695 			} else {
6696 				break;
6697 			}
6698 		}
6699 
6700 		/* No data. Time to sleep? */
6701 		qbackenable(q, 0);
6702 
6703 		/*
6704 		 * If STRHUP or STREOF, return 0 length control and data.
6705 		 * If resid is 0, then a read(fd,buf,0) was done. Do not
6706 		 * sleep to satisfy this request because by default we have
6707 		 * zero bytes to return.
6708 		 */
6709 		if ((stp->sd_flag & (STRHUP|STREOF)) || (mctl->maxlen == 0 &&
6710 		    mdata->maxlen == 0)) {
6711 			mctl->len = mdata->len = 0;
6712 			*flagsp = 0;
6713 			mutex_exit(&stp->sd_lock);
6714 			return (0);
6715 		}
6716 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_WAIT,
6717 		    "strgetmsg calls strwaitq:%p, %p",
6718 		    vp, uiop);
6719 		if (((error = strwaitq(stp, GETWAIT, (ssize_t)0, fmode, -1,
6720 		    &done)) != 0) || done) {
6721 			TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_DONE,
6722 			    "strgetmsg error or done:%p, %p",
6723 			    vp, uiop);
6724 			mutex_exit(&stp->sd_lock);
6725 			return (error);
6726 		}
6727 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_AWAKE,
6728 		    "strgetmsg awakes:%p, %p", vp, uiop);
6729 		if ((error = i_straccess(stp, JCREAD)) != 0) {
6730 			mutex_exit(&stp->sd_lock);
6731 			return (error);
6732 		}
6733 		first = 0;
6734 	}
6735 	ASSERT(bp != NULL);
6736 	/*
6737 	 * Extract any mark information. If the message is not completely
6738 	 * consumed this information will be put in the mblk
6739 	 * that is putback.
6740 	 * If MSGMARKNEXT is set and the message is completely consumed
6741 	 * the STRATMARK flag will be set below. Likewise, if
6742 	 * MSGNOTMARKNEXT is set and the message is
6743 	 * completely consumed STRNOTATMARK will be set.
6744 	 */
6745 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
6746 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
6747 	    (MSGMARKNEXT|MSGNOTMARKNEXT));
6748 	if (mark != 0 && bp == stp->sd_mark) {
6749 		mark |= _LASTMARK;
6750 		stp->sd_mark = NULL;
6751 	}
6752 	/*
6753 	 * keep track of the original message type and priority
6754 	 */
6755 	pri = bp->b_band;
6756 	type = bp->b_datap->db_type;
6757 	if (type == M_PASSFP) {
6758 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
6759 			stp->sd_mark = bp;
6760 		bp->b_flag |= mark & ~_LASTMARK;
6761 		putback(stp, q, bp, pri);
6762 		qbackenable(q, pri);
6763 		mutex_exit(&stp->sd_lock);
6764 		return (EBADMSG);
6765 	}
6766 	ASSERT(type != M_SIG);
6767 
6768 	/*
6769 	 * Set this flag so strrput will not generate signals. Need to
6770 	 * make sure this flag is cleared before leaving this routine
6771 	 * else signals will stop being sent.
6772 	 */
6773 	stp->sd_flag |= STRGETINPROG;
6774 	mutex_exit(&stp->sd_lock);
6775 
6776 	if (STREAM_NEEDSERVICE(stp))
6777 		stream_runservice(stp);
6778 
6779 	/*
6780 	 * Set HIPRI flag if message is priority.
6781 	 */
6782 	if (type >= QPCTL)
6783 		flg = MSG_HIPRI;
6784 	else
6785 		flg = MSG_BAND;
6786 
6787 	/*
6788 	 * First process PROTO or PCPROTO blocks, if any.
6789 	 */
6790 	if (mctl->maxlen >= 0 && type != M_DATA) {
6791 		size_t	n, bcnt;
6792 		char	*ubuf;
6793 
6794 		bcnt = mctl->maxlen;
6795 		ubuf = mctl->buf;
6796 		while (bp != NULL && bp->b_datap->db_type != M_DATA) {
6797 			if ((n = MIN(bcnt, bp->b_wptr - bp->b_rptr)) != 0 &&
6798 			    copyout(bp->b_rptr, ubuf, n)) {
6799 				error = EFAULT;
6800 				mutex_enter(&stp->sd_lock);
6801 				/*
6802 				 * clear stream head pri flag based on
6803 				 * first message type
6804 				 */
6805 				if (type >= QPCTL) {
6806 					ASSERT(type == M_PCPROTO);
6807 					stp->sd_flag &= ~STRPRI;
6808 				}
6809 				more = 0;
6810 				freemsg(bp);
6811 				goto getmout;
6812 			}
6813 			ubuf += n;
6814 			bp->b_rptr += n;
6815 			if (bp->b_rptr >= bp->b_wptr) {
6816 				nbp = bp;
6817 				bp = bp->b_cont;
6818 				freeb(nbp);
6819 			}
6820 			ASSERT(n <= bcnt);
6821 			bcnt -= n;
6822 			if (bcnt == 0)
6823 				break;
6824 		}
6825 		mctl->len = mctl->maxlen - bcnt;
6826 	} else
6827 		mctl->len = -1;
6828 
6829 	if (bp && bp->b_datap->db_type != M_DATA) {
6830 		/*
6831 		 * More PROTO blocks in msg.
6832 		 */
6833 		more |= MORECTL;
6834 		savemp = bp;
6835 		while (bp && bp->b_datap->db_type != M_DATA) {
6836 			savemptail = bp;
6837 			bp = bp->b_cont;
6838 		}
6839 		savemptail->b_cont = NULL;
6840 	}
6841 
6842 	/*
6843 	 * Now process DATA blocks, if any.
6844 	 */
6845 	if (mdata->maxlen >= 0 && bp) {
6846 		/*
6847 		 * struiocopyout will consume a potential zero-length
6848 		 * M_DATA even if uio_resid is zero.
6849 		 */
6850 		size_t oldresid = uiop->uio_resid;
6851 
6852 		bp = struiocopyout(bp, uiop, &error);
6853 		if (error != 0) {
6854 			mutex_enter(&stp->sd_lock);
6855 			/*
6856 			 * clear stream head hi pri flag based on
6857 			 * first message
6858 			 */
6859 			if (type >= QPCTL) {
6860 				ASSERT(type == M_PCPROTO);
6861 				stp->sd_flag &= ~STRPRI;
6862 			}
6863 			more = 0;
6864 			freemsg(savemp);
6865 			goto getmout;
6866 		}
6867 		/*
6868 		 * (pr == 1) indicates a partial read.
6869 		 */
6870 		if (oldresid > uiop->uio_resid)
6871 			pr = 1;
6872 		mdata->len = mdata->maxlen - uiop->uio_resid;
6873 	} else
6874 		mdata->len = -1;
6875 
6876 	if (bp) {			/* more data blocks in msg */
6877 		more |= MOREDATA;
6878 		if (savemp)
6879 			savemptail->b_cont = bp;
6880 		else
6881 			savemp = bp;
6882 	}
6883 
6884 	mutex_enter(&stp->sd_lock);
6885 	if (savemp) {
6886 		if (pr && (savemp->b_datap->db_type == M_DATA) &&
6887 		    msgnodata(savemp)) {
6888 			/*
6889 			 * Avoid queuing a zero-length tail part of
6890 			 * a message. pr=1 indicates that we read some of
6891 			 * the message.
6892 			 */
6893 			freemsg(savemp);
6894 			more &= ~MOREDATA;
6895 			/*
6896 			 * clear stream head hi pri flag based on
6897 			 * first message
6898 			 */
6899 			if (type >= QPCTL) {
6900 				ASSERT(type == M_PCPROTO);
6901 				stp->sd_flag &= ~STRPRI;
6902 			}
6903 		} else {
6904 			savemp->b_band = pri;
6905 			/*
6906 			 * If the first message was HIPRI and the one we're
6907 			 * putting back isn't, then clear STRPRI, otherwise
6908 			 * set STRPRI again.  Note that we must set STRPRI
6909 			 * again since the flush logic in strrput_nondata()
6910 			 * may have cleared it while we had sd_lock dropped.
6911 			 */
6912 			if (type >= QPCTL) {
6913 				ASSERT(type == M_PCPROTO);
6914 				if (queclass(savemp) < QPCTL)
6915 					stp->sd_flag &= ~STRPRI;
6916 				else
6917 					stp->sd_flag |= STRPRI;
6918 			} else if (queclass(savemp) >= QPCTL) {
6919 				/*
6920 				 * The first message was not a HIPRI message,
6921 				 * but the one we are about to putback is.
6922 				 * For simplicitly, we do not allow for HIPRI
6923 				 * messages to be embedded in the message
6924 				 * body, so just force it to same type as
6925 				 * first message.
6926 				 */
6927 				ASSERT(type == M_DATA || type == M_PROTO);
6928 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
6929 				savemp->b_datap->db_type = type;
6930 			}
6931 			if (mark != 0) {
6932 				savemp->b_flag |= mark & ~_LASTMARK;
6933 				if ((mark & _LASTMARK) &&
6934 				    (stp->sd_mark == NULL)) {
6935 					/*
6936 					 * If another marked message arrived
6937 					 * while sd_lock was not held sd_mark
6938 					 * would be non-NULL.
6939 					 */
6940 					stp->sd_mark = savemp;
6941 				}
6942 			}
6943 			putback(stp, q, savemp, pri);
6944 		}
6945 	} else {
6946 		/*
6947 		 * The complete message was consumed.
6948 		 *
6949 		 * If another M_PCPROTO arrived while sd_lock was not held
6950 		 * it would have been discarded since STRPRI was still set.
6951 		 *
6952 		 * Move the MSG*MARKNEXT information
6953 		 * to the stream head just in case
6954 		 * the read queue becomes empty.
6955 		 * clear stream head hi pri flag based on
6956 		 * first message
6957 		 *
6958 		 * If the stream head was at the mark
6959 		 * (STRATMARK) before we dropped sd_lock above
6960 		 * and some data was consumed then we have
6961 		 * moved past the mark thus STRATMARK is
6962 		 * cleared. However, if a message arrived in
6963 		 * strrput during the copyout above causing
6964 		 * STRATMARK to be set we can not clear that
6965 		 * flag.
6966 		 */
6967 		if (type >= QPCTL) {
6968 			ASSERT(type == M_PCPROTO);
6969 			stp->sd_flag &= ~STRPRI;
6970 		}
6971 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
6972 			if (mark & MSGMARKNEXT) {
6973 				stp->sd_flag &= ~STRNOTATMARK;
6974 				stp->sd_flag |= STRATMARK;
6975 			} else if (mark & MSGNOTMARKNEXT) {
6976 				stp->sd_flag &= ~STRATMARK;
6977 				stp->sd_flag |= STRNOTATMARK;
6978 			} else {
6979 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
6980 			}
6981 		} else if (pr && (old_sd_flag & STRATMARK)) {
6982 			stp->sd_flag &= ~STRATMARK;
6983 		}
6984 	}
6985 
6986 	*flagsp = flg;
6987 	*prip = pri;
6988 
6989 	/*
6990 	 * Getmsg cleanup processing - if the state of the queue has changed
6991 	 * some signals may need to be sent and/or poll awakened.
6992 	 */
6993 getmout:
6994 	qbackenable(q, pri);
6995 
6996 	/*
6997 	 * We dropped the stream head lock above. Send all M_SIG messages
6998 	 * before processing stream head for SIGPOLL messages.
6999 	 */
7000 	ASSERT(MUTEX_HELD(&stp->sd_lock));
7001 	while ((bp = q->q_first) != NULL &&
7002 	    (bp->b_datap->db_type == M_SIG)) {
7003 		/*
7004 		 * sd_lock is held so the content of the read queue can not
7005 		 * change.
7006 		 */
7007 		bp = getq(q);
7008 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7009 
7010 		strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7011 		mutex_exit(&stp->sd_lock);
7012 		freemsg(bp);
7013 		if (STREAM_NEEDSERVICE(stp))
7014 			stream_runservice(stp);
7015 		mutex_enter(&stp->sd_lock);
7016 	}
7017 
7018 	/*
7019 	 * stream head cannot change while we make the determination
7020 	 * whether or not to send a signal. Drop the flag to allow strrput
7021 	 * to send firstmsgsigs again.
7022 	 */
7023 	stp->sd_flag &= ~STRGETINPROG;
7024 
7025 	/*
7026 	 * If the type of message at the front of the queue changed
7027 	 * due to the receive the appropriate signals and pollwakeup events
7028 	 * are generated. The type of changes are:
7029 	 *	Processed a hipri message, q_first is not hipri.
7030 	 *	Processed a band X message, and q_first is band Y.
7031 	 * The generated signals and pollwakeups are identical to what
7032 	 * strrput() generates should the message that is now on q_first
7033 	 * arrive to an empty read queue.
7034 	 *
7035 	 * Note: only strrput will send a signal for a hipri message.
7036 	 */
7037 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7038 		strsigset_t signals = 0;
7039 		strpollset_t pollwakeups = 0;
7040 
7041 		if (flg & MSG_HIPRI) {
7042 			/*
7043 			 * Removed a hipri message. Regular data at
7044 			 * the front of  the queue.
7045 			 */
7046 			if (bp->b_band == 0) {
7047 				signals = S_INPUT | S_RDNORM;
7048 				pollwakeups = POLLIN | POLLRDNORM;
7049 			} else {
7050 				signals = S_INPUT | S_RDBAND;
7051 				pollwakeups = POLLIN | POLLRDBAND;
7052 			}
7053 		} else if (pri != bp->b_band) {
7054 			/*
7055 			 * The band is different for the new q_first.
7056 			 */
7057 			if (bp->b_band == 0) {
7058 				signals = S_RDNORM;
7059 				pollwakeups = POLLIN | POLLRDNORM;
7060 			} else {
7061 				signals = S_RDBAND;
7062 				pollwakeups = POLLIN | POLLRDBAND;
7063 			}
7064 		}
7065 
7066 		if (pollwakeups != 0) {
7067 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
7068 				if (!(stp->sd_rput_opt & SR_POLLIN))
7069 					goto no_pollwake;
7070 				stp->sd_rput_opt &= ~SR_POLLIN;
7071 			}
7072 			mutex_exit(&stp->sd_lock);
7073 			pollwakeup(&stp->sd_pollist, pollwakeups);
7074 			mutex_enter(&stp->sd_lock);
7075 		}
7076 no_pollwake:
7077 
7078 		if (stp->sd_sigflags & signals)
7079 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7080 	}
7081 	mutex_exit(&stp->sd_lock);
7082 
7083 	rvp->r_val1 = more;
7084 	return (error);
7085 #undef	_LASTMARK
7086 }
7087 
7088 /*
7089  * Get the next message from the read queue.  If the message is
7090  * priority, STRPRI will have been set by strrput().  This flag
7091  * should be reset only when the entire message at the front of the
7092  * queue as been consumed.
7093  *
7094  * If uiop is NULL all data is returned in mctlp.
7095  * Note that a NULL uiop implies that FNDELAY and FNONBLOCK are assumed
7096  * not enabled.
7097  * The timeout parameter is in milliseconds; -1 for infinity.
7098  * This routine handles the consolidation private flags:
7099  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
7100  *	MSG_DELAYERROR	Defer the error check until the queue is empty.
7101  *	MSG_HOLDSIG	Hold signals while waiting for data.
7102  *	MSG_IPEEK	Only peek at messages.
7103  *	MSG_DISCARDTAIL	Discard the tail M_DATA part of the message
7104  *			that doesn't fit.
7105  *	MSG_NOMARK	If the message is marked leave it on the queue.
7106  *
7107  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
7108  */
7109 int
7110 kstrgetmsg(
7111 	struct vnode *vp,
7112 	mblk_t **mctlp,
7113 	struct uio *uiop,
7114 	unsigned char *prip,
7115 	int *flagsp,
7116 	clock_t timout,
7117 	rval_t *rvp)
7118 {
7119 	struct stdata *stp;
7120 	mblk_t *bp, *nbp;
7121 	mblk_t *savemp = NULL;
7122 	mblk_t *savemptail = NULL;
7123 	int flags;
7124 	uint_t old_sd_flag;
7125 	int flg;
7126 	int more = 0;
7127 	int error = 0;
7128 	char first = 1;
7129 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
7130 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
7131 	unsigned char pri = 0;
7132 	queue_t *q;
7133 	int	pr = 0;			/* Partial read successful */
7134 	unsigned char type;
7135 
7136 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_ENTER,
7137 	    "kstrgetmsg:%p", vp);
7138 
7139 	ASSERT(vp->v_stream);
7140 	stp = vp->v_stream;
7141 	rvp->r_val1 = 0;
7142 
7143 	mutex_enter(&stp->sd_lock);
7144 
7145 	if ((error = i_straccess(stp, JCREAD)) != 0) {
7146 		mutex_exit(&stp->sd_lock);
7147 		return (error);
7148 	}
7149 
7150 	flags = *flagsp;
7151 	if (stp->sd_flag & (STRDERR|STPLEX)) {
7152 		if ((stp->sd_flag & STPLEX) ||
7153 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == 0) {
7154 			error = strgeterr(stp, STRDERR|STPLEX,
7155 			    (flags & MSG_IPEEK));
7156 			if (error != 0) {
7157 				mutex_exit(&stp->sd_lock);
7158 				return (error);
7159 			}
7160 		}
7161 	}
7162 	mutex_exit(&stp->sd_lock);
7163 
7164 	switch (flags & (MSG_HIPRI|MSG_ANY|MSG_BAND)) {
7165 	case MSG_HIPRI:
7166 		if (*prip != 0)
7167 			return (EINVAL);
7168 		break;
7169 
7170 	case MSG_ANY:
7171 	case MSG_BAND:
7172 		break;
7173 
7174 	default:
7175 		return (EINVAL);
7176 	}
7177 
7178 retry:
7179 	q = _RD(stp->sd_wrq);
7180 	mutex_enter(&stp->sd_lock);
7181 	old_sd_flag = stp->sd_flag;
7182 	mark = 0;
7183 	for (;;) {
7184 		int done = 0;
7185 		int waitflag;
7186 		int fmode;
7187 		mblk_t *q_first = q->q_first;
7188 
7189 		/*
7190 		 * This section of the code operates just like the code
7191 		 * in strgetmsg().  There is a comment there about what
7192 		 * is going on here.
7193 		 */
7194 		if (!(flags & (MSG_HIPRI|MSG_BAND))) {
7195 			/* Asking for normal, band0 data */
7196 			bp = strget(stp, q, uiop, first, &error);
7197 			ASSERT(MUTEX_HELD(&stp->sd_lock));
7198 			if (bp != NULL) {
7199 				if (DB_TYPE(bp) == M_SIG) {
7200 					strsignal_nolock(stp, *bp->b_rptr,
7201 					    bp->b_band);
7202 					freemsg(bp);
7203 					continue;
7204 				} else {
7205 					break;
7206 				}
7207 			}
7208 			if (error != 0) {
7209 				goto getmout;
7210 			}
7211 		/*
7212 		 * We can't depend on the value of STRPRI here because
7213 		 * the stream head may be in transit. Therefore, we
7214 		 * must look at the type of the first message to
7215 		 * determine if a high priority messages is waiting
7216 		 */
7217 		} else if ((flags & MSG_HIPRI) && q_first != NULL &&
7218 		    DB_TYPE(q_first) >= QPCTL &&
7219 		    (bp = getq_noenab(q, 0)) != NULL) {
7220 			ASSERT(DB_TYPE(bp) >= QPCTL);
7221 			break;
7222 		} else if ((flags & MSG_BAND) && q_first != NULL &&
7223 		    ((q_first->b_band >= *prip) || DB_TYPE(q_first) >= QPCTL) &&
7224 		    (bp = getq_noenab(q, 0)) != NULL) {
7225 			/*
7226 			 * Asked for at least band "prip" and got either at
7227 			 * least that band or a hipri message.
7228 			 */
7229 			ASSERT(bp->b_band >= *prip || DB_TYPE(bp) >= QPCTL);
7230 			if (DB_TYPE(bp) == M_SIG) {
7231 				strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7232 				freemsg(bp);
7233 				continue;
7234 			} else {
7235 				break;
7236 			}
7237 		}
7238 
7239 		/* No data. Time to sleep? */
7240 		qbackenable(q, 0);
7241 
7242 		/*
7243 		 * Delayed error notification?
7244 		 */
7245 		if ((stp->sd_flag & (STRDERR|STPLEX)) &&
7246 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == MSG_DELAYERROR) {
7247 			error = strgeterr(stp, STRDERR|STPLEX,
7248 			    (flags & MSG_IPEEK));
7249 			if (error != 0) {
7250 				mutex_exit(&stp->sd_lock);
7251 				return (error);
7252 			}
7253 		}
7254 
7255 		/*
7256 		 * If STRHUP or STREOF, return 0 length control and data.
7257 		 * If a read(fd,buf,0) has been done, do not sleep, just
7258 		 * return.
7259 		 *
7260 		 * If mctlp == NULL and uiop == NULL, then the code will
7261 		 * do the strwaitq. This is an understood way of saying
7262 		 * sleep "polling" until a message is received.
7263 		 */
7264 		if ((stp->sd_flag & (STRHUP|STREOF)) ||
7265 		    (uiop != NULL && uiop->uio_resid == 0)) {
7266 			if (mctlp != NULL)
7267 				*mctlp = NULL;
7268 			*flagsp = 0;
7269 			mutex_exit(&stp->sd_lock);
7270 			return (0);
7271 		}
7272 
7273 		waitflag = GETWAIT;
7274 		if (flags &
7275 		    (MSG_HOLDSIG|MSG_IGNERROR|MSG_IPEEK|MSG_DELAYERROR)) {
7276 			if (flags & MSG_HOLDSIG)
7277 				waitflag |= STR_NOSIG;
7278 			if (flags & MSG_IGNERROR)
7279 				waitflag |= STR_NOERROR;
7280 			if (flags & MSG_IPEEK)
7281 				waitflag |= STR_PEEK;
7282 			if (flags & MSG_DELAYERROR)
7283 				waitflag |= STR_DELAYERR;
7284 		}
7285 		if (uiop != NULL)
7286 			fmode = uiop->uio_fmode;
7287 		else
7288 			fmode = 0;
7289 
7290 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_WAIT,
7291 		    "kstrgetmsg calls strwaitq:%p, %p",
7292 		    vp, uiop);
7293 		if (((error = strwaitq(stp, waitflag, (ssize_t)0,
7294 		    fmode, timout, &done))) != 0 || done) {
7295 			TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_DONE,
7296 			    "kstrgetmsg error or done:%p, %p",
7297 			    vp, uiop);
7298 			mutex_exit(&stp->sd_lock);
7299 			return (error);
7300 		}
7301 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_AWAKE,
7302 		    "kstrgetmsg awakes:%p, %p", vp, uiop);
7303 		if ((error = i_straccess(stp, JCREAD)) != 0) {
7304 			mutex_exit(&stp->sd_lock);
7305 			return (error);
7306 		}
7307 		first = 0;
7308 	}
7309 	ASSERT(bp != NULL);
7310 	/*
7311 	 * Extract any mark information. If the message is not completely
7312 	 * consumed this information will be put in the mblk
7313 	 * that is putback.
7314 	 * If MSGMARKNEXT is set and the message is completely consumed
7315 	 * the STRATMARK flag will be set below. Likewise, if
7316 	 * MSGNOTMARKNEXT is set and the message is
7317 	 * completely consumed STRNOTATMARK will be set.
7318 	 */
7319 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
7320 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
7321 	    (MSGMARKNEXT|MSGNOTMARKNEXT));
7322 	pri = bp->b_band;
7323 	if (mark != 0) {
7324 		/*
7325 		 * If the caller doesn't want the mark return.
7326 		 * Used to implement MSG_WAITALL in sockets.
7327 		 */
7328 		if (flags & MSG_NOMARK) {
7329 			putback(stp, q, bp, pri);
7330 			qbackenable(q, pri);
7331 			mutex_exit(&stp->sd_lock);
7332 			return (EWOULDBLOCK);
7333 		}
7334 		if (bp == stp->sd_mark) {
7335 			mark |= _LASTMARK;
7336 			stp->sd_mark = NULL;
7337 		}
7338 	}
7339 
7340 	/*
7341 	 * keep track of the first message type
7342 	 */
7343 	type = bp->b_datap->db_type;
7344 
7345 	if (bp->b_datap->db_type == M_PASSFP) {
7346 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7347 			stp->sd_mark = bp;
7348 		bp->b_flag |= mark & ~_LASTMARK;
7349 		putback(stp, q, bp, pri);
7350 		qbackenable(q, pri);
7351 		mutex_exit(&stp->sd_lock);
7352 		return (EBADMSG);
7353 	}
7354 	ASSERT(type != M_SIG);
7355 
7356 	if (flags & MSG_IPEEK) {
7357 		/*
7358 		 * Clear any struioflag - we do the uiomove over again
7359 		 * when peeking since it simplifies the code.
7360 		 *
7361 		 * Dup the message and put the original back on the queue.
7362 		 * If dupmsg() fails, try again with copymsg() to see if
7363 		 * there is indeed a shortage of memory.  dupmsg() may fail
7364 		 * if db_ref in any of the messages reaches its limit.
7365 		 */
7366 
7367 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
7368 		if ((nbp = dupmsg(bp)) == NULL && (nbp = copymsg(bp)) == NULL) {
7369 			/*
7370 			 * Restore the state of the stream head since we
7371 			 * need to drop sd_lock (strwaitbuf is sleeping).
7372 			 */
7373 			size_t size = msgdsize(bp);
7374 
7375 			if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7376 				stp->sd_mark = bp;
7377 			bp->b_flag |= mark & ~_LASTMARK;
7378 			putback(stp, q, bp, pri);
7379 			mutex_exit(&stp->sd_lock);
7380 			error = strwaitbuf(size, BPRI_HI);
7381 			if (error) {
7382 				/*
7383 				 * There is no net change to the queue thus
7384 				 * no need to qbackenable.
7385 				 */
7386 				return (error);
7387 			}
7388 			goto retry;
7389 		}
7390 
7391 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7392 			stp->sd_mark = bp;
7393 		bp->b_flag |= mark & ~_LASTMARK;
7394 		putback(stp, q, bp, pri);
7395 		bp = nbp;
7396 	}
7397 
7398 	/*
7399 	 * Set this flag so strrput will not generate signals. Need to
7400 	 * make sure this flag is cleared before leaving this routine
7401 	 * else signals will stop being sent.
7402 	 */
7403 	stp->sd_flag |= STRGETINPROG;
7404 	mutex_exit(&stp->sd_lock);
7405 
7406 	if ((stp->sd_rputdatafunc != NULL) && (DB_TYPE(bp) == M_DATA)) {
7407 		mblk_t *tmp, *prevmp;
7408 
7409 		/*
7410 		 * Put first non-data mblk back to stream head and
7411 		 * cut the mblk chain so sd_rputdatafunc only sees
7412 		 * M_DATA mblks. We can skip the first mblk since it
7413 		 * is M_DATA according to the condition above.
7414 		 */
7415 		for (prevmp = bp, tmp = bp->b_cont; tmp != NULL;
7416 		    prevmp = tmp, tmp = tmp->b_cont) {
7417 			if (DB_TYPE(tmp) != M_DATA) {
7418 				prevmp->b_cont = NULL;
7419 				mutex_enter(&stp->sd_lock);
7420 				putback(stp, q, tmp, tmp->b_band);
7421 				mutex_exit(&stp->sd_lock);
7422 				break;
7423 			}
7424 		}
7425 
7426 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
7427 		bp = (stp->sd_rputdatafunc)(stp->sd_vnode, bp,
7428 		    NULL, NULL, NULL, NULL);
7429 
7430 		if (bp == NULL)
7431 			goto retry;
7432 	}
7433 
7434 	if (STREAM_NEEDSERVICE(stp))
7435 		stream_runservice(stp);
7436 
7437 	/*
7438 	 * Set HIPRI flag if message is priority.
7439 	 */
7440 	if (type >= QPCTL)
7441 		flg = MSG_HIPRI;
7442 	else
7443 		flg = MSG_BAND;
7444 
7445 	/*
7446 	 * First process PROTO or PCPROTO blocks, if any.
7447 	 */
7448 	if (mctlp != NULL && type != M_DATA) {
7449 		mblk_t *nbp;
7450 
7451 		*mctlp = bp;
7452 		while (bp->b_cont && bp->b_cont->b_datap->db_type != M_DATA)
7453 			bp = bp->b_cont;
7454 		nbp = bp->b_cont;
7455 		bp->b_cont = NULL;
7456 		bp = nbp;
7457 	}
7458 
7459 	if (bp && bp->b_datap->db_type != M_DATA) {
7460 		/*
7461 		 * More PROTO blocks in msg. Will only happen if mctlp is NULL.
7462 		 */
7463 		more |= MORECTL;
7464 		savemp = bp;
7465 		while (bp && bp->b_datap->db_type != M_DATA) {
7466 			savemptail = bp;
7467 			bp = bp->b_cont;
7468 		}
7469 		savemptail->b_cont = NULL;
7470 	}
7471 
7472 	/*
7473 	 * Now process DATA blocks, if any.
7474 	 */
7475 	if (uiop == NULL) {
7476 		/* Append data to tail of mctlp */
7477 
7478 		ASSERT(bp == NULL || !(bp->b_datap->db_flags & DBLK_UIOA));
7479 		if (mctlp != NULL) {
7480 			mblk_t **mpp = mctlp;
7481 
7482 			while (*mpp != NULL)
7483 				mpp = &((*mpp)->b_cont);
7484 			*mpp = bp;
7485 			bp = NULL;
7486 		}
7487 	} else if (bp && (bp->b_datap->db_flags & DBLK_UIOA)) {
7488 		/*
7489 		 * A uioa mblk_t chain, as uio processing has already
7490 		 * been done we simple skip over processing.
7491 		 */
7492 		bp = NULL;
7493 		pr = 0;
7494 
7495 	} else if (uiop->uio_resid >= 0 && bp) {
7496 		size_t oldresid = uiop->uio_resid;
7497 
7498 		/*
7499 		 * If a streams message is likely to consist
7500 		 * of many small mblks, it is pulled up into
7501 		 * one continuous chunk of memory.
7502 		 * see longer comment at top of page
7503 		 * by mblk_pull_len declaration.
7504 		 */
7505 
7506 		if (MBLKL(bp) < mblk_pull_len) {
7507 			(void) pullupmsg(bp, -1);
7508 		}
7509 
7510 		bp = struiocopyout(bp, uiop, &error);
7511 		if (error != 0) {
7512 			if (mctlp != NULL) {
7513 				freemsg(*mctlp);
7514 				*mctlp = NULL;
7515 			} else
7516 				freemsg(savemp);
7517 			mutex_enter(&stp->sd_lock);
7518 			/*
7519 			 * clear stream head hi pri flag based on
7520 			 * first message
7521 			 */
7522 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7523 				ASSERT(type == M_PCPROTO);
7524 				stp->sd_flag &= ~STRPRI;
7525 			}
7526 			more = 0;
7527 			goto getmout;
7528 		}
7529 		/*
7530 		 * (pr == 1) indicates a partial read.
7531 		 */
7532 		if (oldresid > uiop->uio_resid)
7533 			pr = 1;
7534 	}
7535 
7536 	if (bp) {			/* more data blocks in msg */
7537 		more |= MOREDATA;
7538 		if (savemp)
7539 			savemptail->b_cont = bp;
7540 		else
7541 			savemp = bp;
7542 	}
7543 
7544 	mutex_enter(&stp->sd_lock);
7545 	if (savemp) {
7546 		if (flags & (MSG_IPEEK|MSG_DISCARDTAIL)) {
7547 			/*
7548 			 * When MSG_DISCARDTAIL is set or
7549 			 * when peeking discard any tail. When peeking this
7550 			 * is the tail of the dup that was copied out - the
7551 			 * message has already been putback on the queue.
7552 			 * Return MOREDATA to the caller even though the data
7553 			 * is discarded. This is used by sockets (to
7554 			 * set MSG_TRUNC).
7555 			 */
7556 			freemsg(savemp);
7557 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7558 				ASSERT(type == M_PCPROTO);
7559 				stp->sd_flag &= ~STRPRI;
7560 			}
7561 		} else if (pr && (savemp->b_datap->db_type == M_DATA) &&
7562 		    msgnodata(savemp)) {
7563 			/*
7564 			 * Avoid queuing a zero-length tail part of
7565 			 * a message. pr=1 indicates that we read some of
7566 			 * the message.
7567 			 */
7568 			freemsg(savemp);
7569 			more &= ~MOREDATA;
7570 			if (type >= QPCTL) {
7571 				ASSERT(type == M_PCPROTO);
7572 				stp->sd_flag &= ~STRPRI;
7573 			}
7574 		} else {
7575 			savemp->b_band = pri;
7576 			/*
7577 			 * If the first message was HIPRI and the one we're
7578 			 * putting back isn't, then clear STRPRI, otherwise
7579 			 * set STRPRI again.  Note that we must set STRPRI
7580 			 * again since the flush logic in strrput_nondata()
7581 			 * may have cleared it while we had sd_lock dropped.
7582 			 */
7583 
7584 			ASSERT(!(savemp->b_datap->db_flags & DBLK_UIOA));
7585 			if (type >= QPCTL) {
7586 				ASSERT(type == M_PCPROTO);
7587 				if (queclass(savemp) < QPCTL)
7588 					stp->sd_flag &= ~STRPRI;
7589 				else
7590 					stp->sd_flag |= STRPRI;
7591 			} else if (queclass(savemp) >= QPCTL) {
7592 				/*
7593 				 * The first message was not a HIPRI message,
7594 				 * but the one we are about to putback is.
7595 				 * For simplicitly, we do not allow for HIPRI
7596 				 * messages to be embedded in the message
7597 				 * body, so just force it to same type as
7598 				 * first message.
7599 				 */
7600 				ASSERT(type == M_DATA || type == M_PROTO);
7601 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
7602 				savemp->b_datap->db_type = type;
7603 			}
7604 			if (mark != 0) {
7605 				if ((mark & _LASTMARK) &&
7606 				    (stp->sd_mark == NULL)) {
7607 					/*
7608 					 * If another marked message arrived
7609 					 * while sd_lock was not held sd_mark
7610 					 * would be non-NULL.
7611 					 */
7612 					stp->sd_mark = savemp;
7613 				}
7614 				savemp->b_flag |= mark & ~_LASTMARK;
7615 			}
7616 			putback(stp, q, savemp, pri);
7617 		}
7618 	} else if (!(flags & MSG_IPEEK)) {
7619 		/*
7620 		 * The complete message was consumed.
7621 		 *
7622 		 * If another M_PCPROTO arrived while sd_lock was not held
7623 		 * it would have been discarded since STRPRI was still set.
7624 		 *
7625 		 * Move the MSG*MARKNEXT information
7626 		 * to the stream head just in case
7627 		 * the read queue becomes empty.
7628 		 * clear stream head hi pri flag based on
7629 		 * first message
7630 		 *
7631 		 * If the stream head was at the mark
7632 		 * (STRATMARK) before we dropped sd_lock above
7633 		 * and some data was consumed then we have
7634 		 * moved past the mark thus STRATMARK is
7635 		 * cleared. However, if a message arrived in
7636 		 * strrput during the copyout above causing
7637 		 * STRATMARK to be set we can not clear that
7638 		 * flag.
7639 		 * XXX A "perimeter" would help by single-threading strrput,
7640 		 * strread, strgetmsg and kstrgetmsg.
7641 		 */
7642 		if (type >= QPCTL) {
7643 			ASSERT(type == M_PCPROTO);
7644 			stp->sd_flag &= ~STRPRI;
7645 		}
7646 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
7647 			if (mark & MSGMARKNEXT) {
7648 				stp->sd_flag &= ~STRNOTATMARK;
7649 				stp->sd_flag |= STRATMARK;
7650 			} else if (mark & MSGNOTMARKNEXT) {
7651 				stp->sd_flag &= ~STRATMARK;
7652 				stp->sd_flag |= STRNOTATMARK;
7653 			} else {
7654 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
7655 			}
7656 		} else if (pr && (old_sd_flag & STRATMARK)) {
7657 			stp->sd_flag &= ~STRATMARK;
7658 		}
7659 	}
7660 
7661 	*flagsp = flg;
7662 	*prip = pri;
7663 
7664 	/*
7665 	 * Getmsg cleanup processing - if the state of the queue has changed
7666 	 * some signals may need to be sent and/or poll awakened.
7667 	 */
7668 getmout:
7669 	qbackenable(q, pri);
7670 
7671 	/*
7672 	 * We dropped the stream head lock above. Send all M_SIG messages
7673 	 * before processing stream head for SIGPOLL messages.
7674 	 */
7675 	ASSERT(MUTEX_HELD(&stp->sd_lock));
7676 	while ((bp = q->q_first) != NULL &&
7677 	    (bp->b_datap->db_type == M_SIG)) {
7678 		/*
7679 		 * sd_lock is held so the content of the read queue can not
7680 		 * change.
7681 		 */
7682 		bp = getq(q);
7683 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7684 
7685 		strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7686 		mutex_exit(&stp->sd_lock);
7687 		freemsg(bp);
7688 		if (STREAM_NEEDSERVICE(stp))
7689 			stream_runservice(stp);
7690 		mutex_enter(&stp->sd_lock);
7691 	}
7692 
7693 	/*
7694 	 * stream head cannot change while we make the determination
7695 	 * whether or not to send a signal. Drop the flag to allow strrput
7696 	 * to send firstmsgsigs again.
7697 	 */
7698 	stp->sd_flag &= ~STRGETINPROG;
7699 
7700 	/*
7701 	 * If the type of message at the front of the queue changed
7702 	 * due to the receive the appropriate signals and pollwakeup events
7703 	 * are generated. The type of changes are:
7704 	 *	Processed a hipri message, q_first is not hipri.
7705 	 *	Processed a band X message, and q_first is band Y.
7706 	 * The generated signals and pollwakeups are identical to what
7707 	 * strrput() generates should the message that is now on q_first
7708 	 * arrive to an empty read queue.
7709 	 *
7710 	 * Note: only strrput will send a signal for a hipri message.
7711 	 */
7712 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7713 		strsigset_t signals = 0;
7714 		strpollset_t pollwakeups = 0;
7715 
7716 		if (flg & MSG_HIPRI) {
7717 			/*
7718 			 * Removed a hipri message. Regular data at
7719 			 * the front of  the queue.
7720 			 */
7721 			if (bp->b_band == 0) {
7722 				signals = S_INPUT | S_RDNORM;
7723 				pollwakeups = POLLIN | POLLRDNORM;
7724 			} else {
7725 				signals = S_INPUT | S_RDBAND;
7726 				pollwakeups = POLLIN | POLLRDBAND;
7727 			}
7728 		} else if (pri != bp->b_band) {
7729 			/*
7730 			 * The band is different for the new q_first.
7731 			 */
7732 			if (bp->b_band == 0) {
7733 				signals = S_RDNORM;
7734 				pollwakeups = POLLIN | POLLRDNORM;
7735 			} else {
7736 				signals = S_RDBAND;
7737 				pollwakeups = POLLIN | POLLRDBAND;
7738 			}
7739 		}
7740 
7741 		if (pollwakeups != 0) {
7742 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
7743 				if (!(stp->sd_rput_opt & SR_POLLIN))
7744 					goto no_pollwake;
7745 				stp->sd_rput_opt &= ~SR_POLLIN;
7746 			}
7747 			mutex_exit(&stp->sd_lock);
7748 			pollwakeup(&stp->sd_pollist, pollwakeups);
7749 			mutex_enter(&stp->sd_lock);
7750 		}
7751 no_pollwake:
7752 
7753 		if (stp->sd_sigflags & signals)
7754 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7755 	}
7756 	mutex_exit(&stp->sd_lock);
7757 
7758 	rvp->r_val1 = more;
7759 	return (error);
7760 #undef	_LASTMARK
7761 }
7762 
7763 /*
7764  * Put a message downstream.
7765  *
7766  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7767  */
7768 int
7769 strputmsg(
7770 	struct vnode *vp,
7771 	struct strbuf *mctl,
7772 	struct strbuf *mdata,
7773 	unsigned char pri,
7774 	int flag,
7775 	int fmode)
7776 {
7777 	struct stdata *stp;
7778 	queue_t *wqp;
7779 	mblk_t *mp;
7780 	ssize_t msgsize;
7781 	ssize_t rmin, rmax;
7782 	int error;
7783 	struct uio uios;
7784 	struct uio *uiop = &uios;
7785 	struct iovec iovs;
7786 	int xpg4 = 0;
7787 
7788 	ASSERT(vp->v_stream);
7789 	stp = vp->v_stream;
7790 	wqp = stp->sd_wrq;
7791 
7792 	/*
7793 	 * If it is an XPG4 application, we need to send
7794 	 * SIGPIPE below
7795 	 */
7796 
7797 	xpg4 = (flag & MSG_XPG4) ? 1 : 0;
7798 	flag &= ~MSG_XPG4;
7799 
7800 	if (audit_active)
7801 		audit_strputmsg(vp, mctl, mdata, pri, flag, fmode);
7802 
7803 	mutex_enter(&stp->sd_lock);
7804 
7805 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
7806 		mutex_exit(&stp->sd_lock);
7807 		return (error);
7808 	}
7809 
7810 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7811 		error = strwriteable(stp, B_FALSE, xpg4);
7812 		if (error != 0) {
7813 			mutex_exit(&stp->sd_lock);
7814 			return (error);
7815 		}
7816 	}
7817 
7818 	mutex_exit(&stp->sd_lock);
7819 
7820 	/*
7821 	 * Check for legal flag value.
7822 	 */
7823 	switch (flag) {
7824 	case MSG_HIPRI:
7825 		if ((mctl->len < 0) || (pri != 0))
7826 			return (EINVAL);
7827 		break;
7828 	case MSG_BAND:
7829 		break;
7830 
7831 	default:
7832 		return (EINVAL);
7833 	}
7834 
7835 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_IN,
7836 	    "strputmsg in:stp %p", stp);
7837 
7838 	/* get these values from those cached in the stream head */
7839 	rmin = stp->sd_qn_minpsz;
7840 	rmax = stp->sd_qn_maxpsz;
7841 
7842 	/*
7843 	 * Make sure ctl and data sizes together fall within the
7844 	 * limits of the max and min receive packet sizes and do
7845 	 * not exceed system limit.
7846 	 */
7847 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
7848 	if (rmax == 0) {
7849 		return (ERANGE);
7850 	}
7851 	/*
7852 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
7853 	 * Needed to prevent partial failures in the strmakedata loop.
7854 	 */
7855 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
7856 		rmax = stp->sd_maxblk;
7857 
7858 	if ((msgsize = mdata->len) < 0) {
7859 		msgsize = 0;
7860 		rmin = 0;	/* no range check for NULL data part */
7861 	}
7862 	if ((msgsize < rmin) ||
7863 	    ((msgsize > rmax) && (rmax != INFPSZ)) ||
7864 	    (mctl->len > strctlsz)) {
7865 		return (ERANGE);
7866 	}
7867 
7868 	/*
7869 	 * Setup uio and iov for data part
7870 	 */
7871 	iovs.iov_base = mdata->buf;
7872 	iovs.iov_len = msgsize;
7873 	uios.uio_iov = &iovs;
7874 	uios.uio_iovcnt = 1;
7875 	uios.uio_loffset = 0;
7876 	uios.uio_segflg = UIO_USERSPACE;
7877 	uios.uio_fmode = fmode;
7878 	uios.uio_extflg = UIO_COPY_DEFAULT;
7879 	uios.uio_resid = msgsize;
7880 	uios.uio_offset = 0;
7881 
7882 	/* Ignore flow control in strput for HIPRI */
7883 	if (flag & MSG_HIPRI)
7884 		flag |= MSG_IGNFLOW;
7885 
7886 	for (;;) {
7887 		int done = 0;
7888 
7889 		/*
7890 		 * strput will always free the ctl mblk - even when strput
7891 		 * fails.
7892 		 */
7893 		if ((error = strmakectl(mctl, flag, fmode, &mp)) != 0) {
7894 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7895 			    "strputmsg out:stp %p out %d error %d",
7896 			    stp, 1, error);
7897 			return (error);
7898 		}
7899 		/*
7900 		 * Verify that the whole message can be transferred by
7901 		 * strput.
7902 		 */
7903 		ASSERT(stp->sd_maxblk == INFPSZ ||
7904 		    stp->sd_maxblk >= mdata->len);
7905 
7906 		msgsize = mdata->len;
7907 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
7908 		mdata->len = msgsize;
7909 
7910 		if (error == 0)
7911 			break;
7912 
7913 		if (error != EWOULDBLOCK)
7914 			goto out;
7915 
7916 		mutex_enter(&stp->sd_lock);
7917 		/*
7918 		 * Check for a missed wakeup.
7919 		 * Needed since strput did not hold sd_lock across
7920 		 * the canputnext.
7921 		 */
7922 		if (bcanputnext(wqp, pri)) {
7923 			/* Try again */
7924 			mutex_exit(&stp->sd_lock);
7925 			continue;
7926 		}
7927 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAIT,
7928 		    "strputmsg wait:stp %p waits pri %d", stp, pri);
7929 		if (((error = strwaitq(stp, WRITEWAIT, (ssize_t)0, fmode, -1,
7930 		    &done)) != 0) || done) {
7931 			mutex_exit(&stp->sd_lock);
7932 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7933 			    "strputmsg out:q %p out %d error %d",
7934 			    stp, 0, error);
7935 			return (error);
7936 		}
7937 		TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAKE,
7938 		    "strputmsg wake:stp %p wakes", stp);
7939 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
7940 			mutex_exit(&stp->sd_lock);
7941 			return (error);
7942 		}
7943 		mutex_exit(&stp->sd_lock);
7944 	}
7945 out:
7946 	/*
7947 	 * For historic reasons, applications expect EAGAIN
7948 	 * when data mblk could not be allocated. so change
7949 	 * ENOMEM back to EAGAIN
7950 	 */
7951 	if (error == ENOMEM)
7952 		error = EAGAIN;
7953 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7954 	    "strputmsg out:stp %p out %d error %d", stp, 2, error);
7955 	return (error);
7956 }
7957 
7958 /*
7959  * Put a message downstream.
7960  * Can send only an M_PROTO/M_PCPROTO by passing in a NULL uiop.
7961  * The fmode flag (NDELAY, NONBLOCK) is the or of the flags in the uio
7962  * and the fmode parameter.
7963  *
7964  * This routine handles the consolidation private flags:
7965  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
7966  *	MSG_HOLDSIG	Hold signals while waiting for data.
7967  *	MSG_IGNFLOW	Don't check streams flow control.
7968  *
7969  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7970  */
7971 int
7972 kstrputmsg(
7973 	struct vnode *vp,
7974 	mblk_t *mctl,
7975 	struct uio *uiop,
7976 	ssize_t msgsize,
7977 	unsigned char pri,
7978 	int flag,
7979 	int fmode)
7980 {
7981 	struct stdata *stp;
7982 	queue_t *wqp;
7983 	ssize_t rmin, rmax;
7984 	int error;
7985 
7986 	ASSERT(vp->v_stream);
7987 	stp = vp->v_stream;
7988 	wqp = stp->sd_wrq;
7989 	if (audit_active)
7990 		audit_strputmsg(vp, NULL, NULL, pri, flag, fmode);
7991 	if (mctl == NULL)
7992 		return (EINVAL);
7993 
7994 	mutex_enter(&stp->sd_lock);
7995 
7996 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
7997 		mutex_exit(&stp->sd_lock);
7998 		freemsg(mctl);
7999 		return (error);
8000 	}
8001 
8002 	if ((stp->sd_flag & STPLEX) || !(flag & MSG_IGNERROR)) {
8003 		if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
8004 			error = strwriteable(stp, B_FALSE, B_TRUE);
8005 			if (error != 0) {
8006 				mutex_exit(&stp->sd_lock);
8007 				freemsg(mctl);
8008 				return (error);
8009 			}
8010 		}
8011 	}
8012 
8013 	mutex_exit(&stp->sd_lock);
8014 
8015 	/*
8016 	 * Check for legal flag value.
8017 	 */
8018 	switch (flag & (MSG_HIPRI|MSG_BAND|MSG_ANY)) {
8019 	case MSG_HIPRI:
8020 		if (pri != 0) {
8021 			freemsg(mctl);
8022 			return (EINVAL);
8023 		}
8024 		break;
8025 	case MSG_BAND:
8026 		break;
8027 	default:
8028 		freemsg(mctl);
8029 		return (EINVAL);
8030 	}
8031 
8032 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_IN,
8033 	    "kstrputmsg in:stp %p", stp);
8034 
8035 	/* get these values from those cached in the stream head */
8036 	rmin = stp->sd_qn_minpsz;
8037 	rmax = stp->sd_qn_maxpsz;
8038 
8039 	/*
8040 	 * Make sure ctl and data sizes together fall within the
8041 	 * limits of the max and min receive packet sizes and do
8042 	 * not exceed system limit.
8043 	 */
8044 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
8045 	if (rmax == 0) {
8046 		freemsg(mctl);
8047 		return (ERANGE);
8048 	}
8049 	/*
8050 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
8051 	 * Needed to prevent partial failures in the strmakedata loop.
8052 	 */
8053 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
8054 		rmax = stp->sd_maxblk;
8055 
8056 	if (uiop == NULL) {
8057 		msgsize = -1;
8058 		rmin = -1;	/* no range check for NULL data part */
8059 	} else {
8060 		/* Use uio flags as well as the fmode parameter flags */
8061 		fmode |= uiop->uio_fmode;
8062 
8063 		if ((msgsize < rmin) ||
8064 		    ((msgsize > rmax) && (rmax != INFPSZ))) {
8065 			freemsg(mctl);
8066 			return (ERANGE);
8067 		}
8068 	}
8069 
8070 	/* Ignore flow control in strput for HIPRI */
8071 	if (flag & MSG_HIPRI)
8072 		flag |= MSG_IGNFLOW;
8073 
8074 	for (;;) {
8075 		int done = 0;
8076 		int waitflag;
8077 		mblk_t *mp;
8078 
8079 		/*
8080 		 * strput will always free the ctl mblk - even when strput
8081 		 * fails. If MSG_IGNFLOW is set then any error returned
8082 		 * will cause us to break the loop, so we don't need a copy
8083 		 * of the message. If MSG_IGNFLOW is not set, then we can
8084 		 * get hit by flow control and be forced to try again. In
8085 		 * this case we need to have a copy of the message. We
8086 		 * do this using copymsg since the message may get modified
8087 		 * by something below us.
8088 		 *
8089 		 * We've observed that many TPI providers do not check db_ref
8090 		 * on the control messages but blindly reuse them for the
8091 		 * T_OK_ACK/T_ERROR_ACK. Thus using copymsg is more
8092 		 * friendly to such providers than using dupmsg. Also, note
8093 		 * that sockfs uses MSG_IGNFLOW for all TPI control messages.
8094 		 * Only data messages are subject to flow control, hence
8095 		 * subject to this copymsg.
8096 		 */
8097 		if (flag & MSG_IGNFLOW) {
8098 			mp = mctl;
8099 			mctl = NULL;
8100 		} else {
8101 			do {
8102 				/*
8103 				 * If a message has a free pointer, the message
8104 				 * must be dupmsg to maintain this pointer.
8105 				 * Code using this facility must be sure
8106 				 * that modules below will not change the
8107 				 * contents of the dblk without checking db_ref
8108 				 * first. If db_ref is > 1, then the module
8109 				 * needs to do a copymsg first. Otherwise,
8110 				 * the contents of the dblk may become
8111 				 * inconsistent because the freesmg/freeb below
8112 				 * may end up calling atomic_add_32_nv.
8113 				 * The atomic_add_32_nv in freeb (accessing
8114 				 * all of db_ref, db_type, db_flags, and
8115 				 * db_struioflag) does not prevent other threads
8116 				 * from concurrently trying to modify e.g.
8117 				 * db_type.
8118 				 */
8119 				if (mctl->b_datap->db_frtnp != NULL)
8120 					mp = dupmsg(mctl);
8121 				else
8122 					mp = copymsg(mctl);
8123 
8124 				if (mp != NULL)
8125 					break;
8126 
8127 				error = strwaitbuf(msgdsize(mctl), BPRI_MED);
8128 				if (error) {
8129 					freemsg(mctl);
8130 					return (error);
8131 				}
8132 			} while (mp == NULL);
8133 		}
8134 		/*
8135 		 * Verify that all of msgsize can be transferred by
8136 		 * strput.
8137 		 */
8138 		ASSERT(stp->sd_maxblk == INFPSZ || stp->sd_maxblk >= msgsize);
8139 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
8140 		if (error == 0)
8141 			break;
8142 
8143 		if (error != EWOULDBLOCK)
8144 			goto out;
8145 
8146 		/*
8147 		 * IF MSG_IGNFLOW is set we should have broken out of loop
8148 		 * above.
8149 		 */
8150 		ASSERT(!(flag & MSG_IGNFLOW));
8151 		mutex_enter(&stp->sd_lock);
8152 		/*
8153 		 * Check for a missed wakeup.
8154 		 * Needed since strput did not hold sd_lock across
8155 		 * the canputnext.
8156 		 */
8157 		if (bcanputnext(wqp, pri)) {
8158 			/* Try again */
8159 			mutex_exit(&stp->sd_lock);
8160 			continue;
8161 		}
8162 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAIT,
8163 		    "kstrputmsg wait:stp %p waits pri %d", stp, pri);
8164 
8165 		waitflag = WRITEWAIT;
8166 		if (flag & (MSG_HOLDSIG|MSG_IGNERROR)) {
8167 			if (flag & MSG_HOLDSIG)
8168 				waitflag |= STR_NOSIG;
8169 			if (flag & MSG_IGNERROR)
8170 				waitflag |= STR_NOERROR;
8171 		}
8172 		if (((error = strwaitq(stp, waitflag,
8173 		    (ssize_t)0, fmode, -1, &done)) != 0) || done) {
8174 			mutex_exit(&stp->sd_lock);
8175 			TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8176 			    "kstrputmsg out:stp %p out %d error %d",
8177 			    stp, 0, error);
8178 			freemsg(mctl);
8179 			return (error);
8180 		}
8181 		TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAKE,
8182 		    "kstrputmsg wake:stp %p wakes", stp);
8183 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
8184 			mutex_exit(&stp->sd_lock);
8185 			freemsg(mctl);
8186 			return (error);
8187 		}
8188 		mutex_exit(&stp->sd_lock);
8189 	}
8190 out:
8191 	freemsg(mctl);
8192 	/*
8193 	 * For historic reasons, applications expect EAGAIN
8194 	 * when data mblk could not be allocated. so change
8195 	 * ENOMEM back to EAGAIN
8196 	 */
8197 	if (error == ENOMEM)
8198 		error = EAGAIN;
8199 	TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8200 	    "kstrputmsg out:stp %p out %d error %d", stp, 2, error);
8201 	return (error);
8202 }
8203 
8204 /*
8205  * Determines whether the necessary conditions are set on a stream
8206  * for it to be readable, writeable, or have exceptions.
8207  *
8208  * strpoll handles the consolidation private events:
8209  *	POLLNOERR	Do not return POLLERR even if there are stream
8210  *			head errors.
8211  *			Used by sockfs.
8212  *	POLLRDDATA	Do not return POLLIN unless at least one message on
8213  *			the queue contains one or more M_DATA mblks. Thus
8214  *			when this flag is set a queue with only
8215  *			M_PROTO/M_PCPROTO mblks does not return POLLIN.
8216  *			Used by sockfs to ignore T_EXDATA_IND messages.
8217  *
8218  * Note: POLLRDDATA assumes that synch streams only return messages with
8219  * an M_DATA attached (i.e. not messages consisting of only
8220  * an M_PROTO/M_PCPROTO part).
8221  */
8222 int
8223 strpoll(
8224 	struct stdata *stp,
8225 	short events_arg,
8226 	int anyyet,
8227 	short *reventsp,
8228 	struct pollhead **phpp)
8229 {
8230 	int events = (ushort_t)events_arg;
8231 	int retevents = 0;
8232 	mblk_t *mp;
8233 	qband_t *qbp;
8234 	long sd_flags = stp->sd_flag;
8235 	int headlocked = 0;
8236 
8237 	/*
8238 	 * For performance, a single 'if' tests for most possible edge
8239 	 * conditions in one shot
8240 	 */
8241 	if (sd_flags & (STPLEX | STRDERR | STWRERR)) {
8242 		if (sd_flags & STPLEX) {
8243 			*reventsp = POLLNVAL;
8244 			return (EINVAL);
8245 		}
8246 		if (((events & (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) &&
8247 		    (sd_flags & STRDERR)) ||
8248 		    ((events & (POLLOUT | POLLWRNORM | POLLWRBAND)) &&
8249 		    (sd_flags & STWRERR))) {
8250 			if (!(events & POLLNOERR)) {
8251 				*reventsp = POLLERR;
8252 				return (0);
8253 			}
8254 		}
8255 	}
8256 	if (sd_flags & STRHUP) {
8257 		retevents |= POLLHUP;
8258 	} else if (events & (POLLWRNORM | POLLWRBAND)) {
8259 		queue_t *tq;
8260 		queue_t	*qp = stp->sd_wrq;
8261 
8262 		claimstr(qp);
8263 		/* Find next module forward that has a service procedure */
8264 		tq = qp->q_next->q_nfsrv;
8265 		ASSERT(tq != NULL);
8266 
8267 		polllock(&stp->sd_pollist, QLOCK(tq));
8268 		if (events & POLLWRNORM) {
8269 			queue_t *sqp;
8270 
8271 			if (tq->q_flag & QFULL)
8272 				/* ensure backq svc procedure runs */
8273 				tq->q_flag |= QWANTW;
8274 			else if ((sqp = stp->sd_struiowrq) != NULL) {
8275 				/* Check sync stream barrier write q */
8276 				mutex_exit(QLOCK(tq));
8277 				polllock(&stp->sd_pollist, QLOCK(sqp));
8278 				if (sqp->q_flag & QFULL)
8279 					/* ensure pollwakeup() is done */
8280 					sqp->q_flag |= QWANTWSYNC;
8281 				else
8282 					retevents |= POLLOUT;
8283 				/* More write events to process ??? */
8284 				if (! (events & POLLWRBAND)) {
8285 					mutex_exit(QLOCK(sqp));
8286 					releasestr(qp);
8287 					goto chkrd;
8288 				}
8289 				mutex_exit(QLOCK(sqp));
8290 				polllock(&stp->sd_pollist, QLOCK(tq));
8291 			} else
8292 				retevents |= POLLOUT;
8293 		}
8294 		if (events & POLLWRBAND) {
8295 			qbp = tq->q_bandp;
8296 			if (qbp) {
8297 				while (qbp) {
8298 					if (qbp->qb_flag & QB_FULL)
8299 						qbp->qb_flag |= QB_WANTW;
8300 					else
8301 						retevents |= POLLWRBAND;
8302 					qbp = qbp->qb_next;
8303 				}
8304 			} else {
8305 				retevents |= POLLWRBAND;
8306 			}
8307 		}
8308 		mutex_exit(QLOCK(tq));
8309 		releasestr(qp);
8310 	}
8311 chkrd:
8312 	if (sd_flags & STRPRI) {
8313 		retevents |= (events & POLLPRI);
8314 	} else if (events & (POLLRDNORM | POLLRDBAND | POLLIN)) {
8315 		queue_t	*qp = _RD(stp->sd_wrq);
8316 		int normevents = (events & (POLLIN | POLLRDNORM));
8317 
8318 		/*
8319 		 * Note: Need to do polllock() here since ps_lock may be
8320 		 * held. See bug 4191544.
8321 		 */
8322 		polllock(&stp->sd_pollist, &stp->sd_lock);
8323 		headlocked = 1;
8324 		mp = qp->q_first;
8325 		while (mp) {
8326 			/*
8327 			 * For POLLRDDATA we scan b_cont and b_next until we
8328 			 * find an M_DATA.
8329 			 */
8330 			if ((events & POLLRDDATA) &&
8331 			    mp->b_datap->db_type != M_DATA) {
8332 				mblk_t *nmp = mp->b_cont;
8333 
8334 				while (nmp != NULL &&
8335 				    nmp->b_datap->db_type != M_DATA)
8336 					nmp = nmp->b_cont;
8337 				if (nmp == NULL) {
8338 					mp = mp->b_next;
8339 					continue;
8340 				}
8341 			}
8342 			if (mp->b_band == 0)
8343 				retevents |= normevents;
8344 			else
8345 				retevents |= (events & (POLLIN | POLLRDBAND));
8346 			break;
8347 		}
8348 		if (! (retevents & normevents) &&
8349 		    (stp->sd_wakeq & RSLEEP)) {
8350 			/*
8351 			 * Sync stream barrier read queue has data.
8352 			 */
8353 			retevents |= normevents;
8354 		}
8355 		/* Treat eof as normal data */
8356 		if (sd_flags & STREOF)
8357 			retevents |= normevents;
8358 	}
8359 
8360 	*reventsp = (short)retevents;
8361 	if (retevents) {
8362 		if (headlocked)
8363 			mutex_exit(&stp->sd_lock);
8364 		return (0);
8365 	}
8366 
8367 	/*
8368 	 * If poll() has not found any events yet, set up event cell
8369 	 * to wake up the poll if a requested event occurs on this
8370 	 * stream.  Check for collisions with outstanding poll requests.
8371 	 */
8372 	if (!anyyet) {
8373 		*phpp = &stp->sd_pollist;
8374 		if (headlocked == 0) {
8375 			polllock(&stp->sd_pollist, &stp->sd_lock);
8376 			headlocked = 1;
8377 		}
8378 		stp->sd_rput_opt |= SR_POLLIN;
8379 	}
8380 	if (headlocked)
8381 		mutex_exit(&stp->sd_lock);
8382 	return (0);
8383 }
8384 
8385 /*
8386  * The purpose of putback() is to assure sleeping polls/reads
8387  * are awakened when there are no new messages arriving at the,
8388  * stream head, and a message is placed back on the read queue.
8389  *
8390  * sd_lock must be held when messages are placed back on stream
8391  * head.  (getq() holds sd_lock when it removes messages from
8392  * the queue)
8393  */
8394 
8395 static void
8396 putback(struct stdata *stp, queue_t *q, mblk_t *bp, int band)
8397 {
8398 	mblk_t	*qfirst;
8399 	ASSERT(MUTEX_HELD(&stp->sd_lock));
8400 
8401 	/*
8402 	 * As a result of lock-step ordering around q_lock and sd_lock,
8403 	 * it's possible for function calls like putnext() and
8404 	 * canputnext() to get an inaccurate picture of how much
8405 	 * data is really being processed at the stream head.
8406 	 * We only consolidate with existing messages on the queue
8407 	 * if the length of the message we want to put back is smaller
8408 	 * than the queue hiwater mark.
8409 	 */
8410 	if ((stp->sd_rput_opt & SR_CONSOL_DATA) &&
8411 	    (DB_TYPE(bp) == M_DATA) && ((qfirst = q->q_first) != NULL) &&
8412 	    (DB_TYPE(qfirst) == M_DATA) &&
8413 	    ((qfirst->b_flag & (MSGMARK|MSGDELIM)) == 0) &&
8414 	    ((bp->b_flag & (MSGMARK|MSGDELIM|MSGMARKNEXT)) == 0) &&
8415 	    (mp_cont_len(bp, NULL) < q->q_hiwat)) {
8416 		/*
8417 		 * We use the same logic as defined in strrput()
8418 		 * but in reverse as we are putting back onto the
8419 		 * queue and want to retain byte ordering.
8420 		 * Consolidate M_DATA messages with M_DATA ONLY.
8421 		 * strrput() allows the consolidation of M_DATA onto
8422 		 * M_PROTO | M_PCPROTO but not the other way round.
8423 		 *
8424 		 * The consolidation does not take place if the message
8425 		 * we are returning to the queue is marked with either
8426 		 * of the marks or the delim flag or if q_first
8427 		 * is marked with MSGMARK. The MSGMARK check is needed to
8428 		 * handle the odd semantics of MSGMARK where essentially
8429 		 * the whole message is to be treated as marked.
8430 		 * Carry any MSGMARKNEXT and MSGNOTMARKNEXT from q_first
8431 		 * to the front of the b_cont chain.
8432 		 */
8433 		rmvq_noenab(q, qfirst);
8434 
8435 		/*
8436 		 * The first message in the b_cont list
8437 		 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
8438 		 * We need to handle the case where we
8439 		 * are appending:
8440 		 *
8441 		 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
8442 		 * 2) a MSGMARKNEXT to a plain message.
8443 		 * 3) a MSGNOTMARKNEXT to a plain message
8444 		 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
8445 		 *    message.
8446 		 *
8447 		 * Thus we never append a MSGMARKNEXT or
8448 		 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
8449 		 */
8450 		if (qfirst->b_flag & MSGMARKNEXT) {
8451 			bp->b_flag |= MSGMARKNEXT;
8452 			bp->b_flag &= ~MSGNOTMARKNEXT;
8453 			qfirst->b_flag &= ~MSGMARKNEXT;
8454 		} else if (qfirst->b_flag & MSGNOTMARKNEXT) {
8455 			bp->b_flag |= MSGNOTMARKNEXT;
8456 			qfirst->b_flag &= ~MSGNOTMARKNEXT;
8457 		}
8458 
8459 		linkb(bp, qfirst);
8460 	}
8461 	(void) putbq(q, bp);
8462 
8463 	/*
8464 	 * A message may have come in when the sd_lock was dropped in the
8465 	 * calling routine. If this is the case and STR*ATMARK info was
8466 	 * received, need to move that from the stream head to the q_last
8467 	 * so that SIOCATMARK can return the proper value.
8468 	 */
8469 	if (stp->sd_flag & (STRATMARK | STRNOTATMARK)) {
8470 		unsigned short *flagp = &q->q_last->b_flag;
8471 		uint_t b_flag = (uint_t)*flagp;
8472 
8473 		if (stp->sd_flag & STRATMARK) {
8474 			b_flag &= ~MSGNOTMARKNEXT;
8475 			b_flag |= MSGMARKNEXT;
8476 			stp->sd_flag &= ~STRATMARK;
8477 		} else {
8478 			b_flag &= ~MSGMARKNEXT;
8479 			b_flag |= MSGNOTMARKNEXT;
8480 			stp->sd_flag &= ~STRNOTATMARK;
8481 		}
8482 		*flagp = (unsigned short) b_flag;
8483 	}
8484 
8485 #ifdef	DEBUG
8486 	/*
8487 	 * Make sure that the flags are not messed up.
8488 	 */
8489 	{
8490 		mblk_t *mp;
8491 		mp = q->q_last;
8492 		while (mp != NULL) {
8493 			ASSERT((mp->b_flag & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
8494 			    (MSGMARKNEXT|MSGNOTMARKNEXT));
8495 			mp = mp->b_cont;
8496 		}
8497 	}
8498 #endif
8499 	if (q->q_first == bp) {
8500 		short pollevents;
8501 
8502 		if (stp->sd_flag & RSLEEP) {
8503 			stp->sd_flag &= ~RSLEEP;
8504 			cv_broadcast(&q->q_wait);
8505 		}
8506 		if (stp->sd_flag & STRPRI) {
8507 			pollevents = POLLPRI;
8508 		} else {
8509 			if (band == 0) {
8510 				if (!(stp->sd_rput_opt & SR_POLLIN))
8511 					return;
8512 				stp->sd_rput_opt &= ~SR_POLLIN;
8513 				pollevents = POLLIN | POLLRDNORM;
8514 			} else {
8515 				pollevents = POLLIN | POLLRDBAND;
8516 			}
8517 		}
8518 		mutex_exit(&stp->sd_lock);
8519 		pollwakeup(&stp->sd_pollist, pollevents);
8520 		mutex_enter(&stp->sd_lock);
8521 	}
8522 }
8523 
8524 /*
8525  * Return the held vnode attached to the stream head of a
8526  * given queue
8527  * It is the responsibility of the calling routine to ensure
8528  * that the queue does not go away (e.g. pop).
8529  */
8530 vnode_t *
8531 strq2vp(queue_t *qp)
8532 {
8533 	vnode_t *vp;
8534 	vp = STREAM(qp)->sd_vnode;
8535 	ASSERT(vp != NULL);
8536 	VN_HOLD(vp);
8537 	return (vp);
8538 }
8539 
8540 /*
8541  * return the stream head write queue for the given vp
8542  * It is the responsibility of the calling routine to ensure
8543  * that the stream or vnode do not close.
8544  */
8545 queue_t *
8546 strvp2wq(vnode_t *vp)
8547 {
8548 	ASSERT(vp->v_stream != NULL);
8549 	return (vp->v_stream->sd_wrq);
8550 }
8551 
8552 /*
8553  * pollwakeup stream head
8554  * It is the responsibility of the calling routine to ensure
8555  * that the stream or vnode do not close.
8556  */
8557 void
8558 strpollwakeup(vnode_t *vp, short event)
8559 {
8560 	ASSERT(vp->v_stream);
8561 	pollwakeup(&vp->v_stream->sd_pollist, event);
8562 }
8563 
8564 /*
8565  * Mate the stream heads of two vnodes together. If the two vnodes are the
8566  * same, we just make the write-side point at the read-side -- otherwise,
8567  * we do a full mate.  Only works on vnodes associated with streams that are
8568  * still being built and thus have only a stream head.
8569  */
8570 void
8571 strmate(vnode_t *vp1, vnode_t *vp2)
8572 {
8573 	queue_t *wrq1 = strvp2wq(vp1);
8574 	queue_t *wrq2 = strvp2wq(vp2);
8575 
8576 	/*
8577 	 * Verify that there are no modules on the stream yet.  We also
8578 	 * rely on the stream head always having a service procedure to
8579 	 * avoid tweaking q_nfsrv.
8580 	 */
8581 	ASSERT(wrq1->q_next == NULL && wrq2->q_next == NULL);
8582 	ASSERT(wrq1->q_qinfo->qi_srvp != NULL);
8583 	ASSERT(wrq2->q_qinfo->qi_srvp != NULL);
8584 
8585 	/*
8586 	 * If the queues are the same, just twist; otherwise do a full mate.
8587 	 */
8588 	if (wrq1 == wrq2) {
8589 		wrq1->q_next = _RD(wrq1);
8590 	} else {
8591 		wrq1->q_next = _RD(wrq2);
8592 		wrq2->q_next = _RD(wrq1);
8593 		STREAM(wrq1)->sd_mate = STREAM(wrq2);
8594 		STREAM(wrq1)->sd_flag |= STRMATE;
8595 		STREAM(wrq2)->sd_mate = STREAM(wrq1);
8596 		STREAM(wrq2)->sd_flag |= STRMATE;
8597 	}
8598 }
8599 
8600 /*
8601  * XXX will go away when console is correctly fixed.
8602  * Clean up the console PIDS, from previous I_SETSIG,
8603  * called only for cnopen which never calls strclean().
8604  */
8605 void
8606 str_cn_clean(struct vnode *vp)
8607 {
8608 	strsig_t *ssp, *pssp, *tssp;
8609 	struct stdata *stp;
8610 	struct pid  *pidp;
8611 	int update = 0;
8612 
8613 	ASSERT(vp->v_stream);
8614 	stp = vp->v_stream;
8615 	pssp = NULL;
8616 	mutex_enter(&stp->sd_lock);
8617 	ssp = stp->sd_siglist;
8618 	while (ssp) {
8619 		mutex_enter(&pidlock);
8620 		pidp = ssp->ss_pidp;
8621 		/*
8622 		 * Get rid of PID if the proc is gone.
8623 		 */
8624 		if (pidp->pid_prinactive) {
8625 			tssp = ssp->ss_next;
8626 			if (pssp)
8627 				pssp->ss_next = tssp;
8628 			else
8629 				stp->sd_siglist = tssp;
8630 			ASSERT(pidp->pid_ref <= 1);
8631 			PID_RELE(ssp->ss_pidp);
8632 			mutex_exit(&pidlock);
8633 			kmem_free(ssp, sizeof (strsig_t));
8634 			update = 1;
8635 			ssp = tssp;
8636 			continue;
8637 		} else
8638 			mutex_exit(&pidlock);
8639 		pssp = ssp;
8640 		ssp = ssp->ss_next;
8641 	}
8642 	if (update) {
8643 		stp->sd_sigflags = 0;
8644 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
8645 			stp->sd_sigflags |= ssp->ss_events;
8646 	}
8647 	mutex_exit(&stp->sd_lock);
8648 }
8649 
8650 /*
8651  * Return B_TRUE if there is data in the message, B_FALSE otherwise.
8652  */
8653 static boolean_t
8654 msghasdata(mblk_t *bp)
8655 {
8656 	for (; bp; bp = bp->b_cont)
8657 		if (bp->b_datap->db_type == M_DATA) {
8658 			ASSERT(bp->b_wptr >= bp->b_rptr);
8659 			if (bp->b_wptr > bp->b_rptr)
8660 				return (B_TRUE);
8661 		}
8662 	return (B_FALSE);
8663 }
8664 
8665 /*
8666  * Called on the first strget() of a sodirect/uioa enabled streamhead,
8667  * if any mblk_t(s) enqueued they must first be uioamove()d before uioa
8668  * can be enabled for the underlying transport's use.
8669  */
8670 void
8671 struioainit(queue_t *q, sodirect_t *sodp, uio_t *uiop)
8672 {
8673 	uioa_t	*uioap = (uioa_t *)uiop;
8674 	mblk_t	*bp;
8675 	mblk_t	*lbp = NULL;
8676 	mblk_t	*wbp;
8677 	int	len;
8678 	int	error;
8679 
8680 	ASSERT(MUTEX_HELD(sodp->sod_lockp));
8681 	ASSERT(&sodp->sod_uioa == uioap);
8682 
8683 	/*
8684 	 * Walk first b_cont chain in sod_q
8685 	 * and schedule any M_DATA mblk_t's for uio asynchronous move.
8686 	 */
8687 	mutex_enter(QLOCK(q));
8688 	if ((bp = q->q_first) == NULL) {
8689 		mutex_exit(QLOCK(q));
8690 		return;
8691 	}
8692 	/* Walk the chain */
8693 	wbp = bp;
8694 	do {
8695 		if (wbp->b_datap->db_type != M_DATA) {
8696 			/* Not M_DATA, no more uioa */
8697 			goto nouioa;
8698 		}
8699 		if ((len = wbp->b_wptr - wbp->b_rptr) > 0) {
8700 			/* Have a M_DATA mblk_t with data */
8701 			if (len > uioap->uio_resid) {
8702 				/* Not enough uio sapce */
8703 				goto nouioa;
8704 			}
8705 			ASSERT(!(wbp->b_datap->db_flags & DBLK_UIOA));
8706 			error = uioamove(wbp->b_rptr, len,
8707 			    UIO_READ, uioap);
8708 			if (!error) {
8709 				/* Scheduled, mark dblk_t as such */
8710 				wbp->b_datap->db_flags |= DBLK_UIOA;
8711 			} else {
8712 				/* Break the mblk chain */
8713 				goto nouioa;
8714 			}
8715 		}
8716 		/* Save last wbp processed */
8717 		lbp = wbp;
8718 	} while ((wbp = wbp->b_cont) != NULL);
8719 
8720 	mutex_exit(QLOCK(q));
8721 	return;
8722 
8723 nouioa:
8724 	/* No more uioa */
8725 	uioap->uioa_state &= UIOA_CLR;
8726 	uioap->uioa_state |= UIOA_FINI;
8727 
8728 	/*
8729 	 * If we processed 1 or more mblk_t(s) then we need to split the
8730 	 * current mblk_t chain in 2 so that all the uioamove()ed mblk_t(s)
8731 	 * are in the current chain and the rest are in the following new
8732 	 * chain.
8733 	 */
8734 	if (lbp != NULL) {
8735 		/* New end of current chain */
8736 		lbp->b_cont = NULL;
8737 
8738 		/* Insert new chain wbp after bp */
8739 		if ((wbp->b_next = bp->b_next) != NULL)
8740 			bp->b_next->b_prev = wbp;
8741 		else
8742 			q->q_last = wbp;
8743 		wbp->b_prev = bp;
8744 		bp->b_next = wbp;
8745 	}
8746 	mutex_exit(QLOCK(q));
8747 }
8748