xref: /titanic_41/usr/src/uts/common/os/streamio.c (revision 4a6ec905b96eb96a398c346f59e034a90ce8ad37)
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 2008 Sun Microsystems, Inc.  All rights reserved.
27  * Use is subject to license terms.
28  */
29 
30 #pragma ident	"%Z%%M%	%I%	%E% SMI"
31 
32 #include <sys/types.h>
33 #include <sys/sysmacros.h>
34 #include <sys/param.h>
35 #include <sys/errno.h>
36 #include <sys/signal.h>
37 #include <sys/stat.h>
38 #include <sys/proc.h>
39 #include <sys/cred.h>
40 #include <sys/user.h>
41 #include <sys/vnode.h>
42 #include <sys/file.h>
43 #include <sys/stream.h>
44 #include <sys/strsubr.h>
45 #include <sys/stropts.h>
46 #include <sys/tihdr.h>
47 #include <sys/var.h>
48 #include <sys/poll.h>
49 #include <sys/termio.h>
50 #include <sys/ttold.h>
51 #include <sys/systm.h>
52 #include <sys/uio.h>
53 #include <sys/cmn_err.h>
54 #include <sys/sad.h>
55 #include <sys/netstack.h>
56 #include <sys/priocntl.h>
57 #include <sys/jioctl.h>
58 #include <sys/procset.h>
59 #include <sys/session.h>
60 #include <sys/kmem.h>
61 #include <sys/filio.h>
62 #include <sys/vtrace.h>
63 #include <sys/debug.h>
64 #include <sys/strredir.h>
65 #include <sys/fs/fifonode.h>
66 #include <sys/fs/snode.h>
67 #include <sys/strlog.h>
68 #include <sys/strsun.h>
69 #include <sys/project.h>
70 #include <sys/kbio.h>
71 #include <sys/msio.h>
72 #include <sys/tty.h>
73 #include <sys/ptyvar.h>
74 #include <sys/vuid_event.h>
75 #include <sys/modctl.h>
76 #include <sys/sunddi.h>
77 #include <sys/sunldi_impl.h>
78 #include <sys/autoconf.h>
79 #include <sys/policy.h>
80 #include <sys/dld.h>
81 #include <sys/zone.h>
82 
83 /*
84  * This define helps improve the readability of streams code while
85  * still maintaining a very old streams performance enhancement.  The
86  * performance enhancement basically involved having all callers
87  * of straccess() perform the first check that straccess() will do
88  * locally before actually calling straccess().  (There by reducing
89  * the number of unnecessary calls to straccess().)
90  */
91 #define	i_straccess(x, y)	((stp->sd_sidp == NULL) ? 0 : \
92 				    (stp->sd_vnode->v_type == VFIFO) ? 0 : \
93 				    straccess((x), (y)))
94 
95 /*
96  * what is mblk_pull_len?
97  *
98  * If a streams message consists of many short messages,
99  * a performance degradation occurs from copyout overhead.
100  * To decrease the per mblk overhead, messages that are
101  * likely to consist of many small mblks are pulled up into
102  * one continuous chunk of memory.
103  *
104  * To avoid the processing overhead of examining every
105  * mblk, a quick heuristic is used. If the first mblk in
106  * the message is shorter than mblk_pull_len, it is likely
107  * that the rest of the mblk will be short.
108  *
109  * This heuristic was decided upon after performance tests
110  * indicated that anything more complex slowed down the main
111  * code path.
112  */
113 #define	MBLK_PULL_LEN 64
114 uint32_t mblk_pull_len = MBLK_PULL_LEN;
115 
116 /*
117  * The sgttyb_handling flag controls the handling of the old BSD
118  * TIOCGETP, TIOCSETP, and TIOCSETN ioctls as follows:
119  *
120  * 0 - Emit no warnings at all and retain old, broken behavior.
121  * 1 - Emit no warnings and silently handle new semantics.
122  * 2 - Send cmn_err(CE_NOTE) when either TIOCSETP or TIOCSETN is used
123  *     (once per system invocation).  Handle with new semantics.
124  * 3 - Send SIGSYS when any TIOCGETP, TIOCSETP, or TIOCSETN call is
125  *     made (so that offenders drop core and are easy to debug).
126  *
127  * The "new semantics" are that TIOCGETP returns B38400 for
128  * sg_[io]speed if the corresponding value is over B38400, and that
129  * TIOCSET[PN] accept B38400 in these cases to mean "retain current
130  * bit rate."
131  */
132 int sgttyb_handling = 1;
133 static boolean_t sgttyb_complaint;
134 
135 /* don't push drcompat module by default on Style-2 streams */
136 static int push_drcompat = 0;
137 
138 /*
139  * id value used to distinguish between different ioctl messages
140  */
141 static uint32_t ioc_id;
142 
143 static void putback(struct stdata *, queue_t *, mblk_t *, int);
144 static void strcleanall(struct vnode *);
145 static int strwsrv(queue_t *);
146 static int strdocmd(struct stdata *, struct strcmd *, cred_t *);
147 static void struioainit(queue_t *, sodirect_t *, uio_t *);
148 
149 /*
150  * qinit and module_info structures for stream head read and write queues
151  */
152 struct module_info strm_info = { 0, "strrhead", 0, INFPSZ, STRHIGH, STRLOW };
153 struct module_info stwm_info = { 0, "strwhead", 0, 0, 0, 0 };
154 struct qinit strdata = { strrput, NULL, NULL, NULL, NULL, &strm_info };
155 struct qinit stwdata = { NULL, strwsrv, NULL, NULL, NULL, &stwm_info };
156 struct module_info fiform_info = { 0, "fifostrrhead", 0, PIPE_BUF, FIFOHIWAT,
157     FIFOLOWAT };
158 struct module_info fifowm_info = { 0, "fifostrwhead", 0, 0, 0, 0 };
159 struct qinit fifo_strdata = { strrput, NULL, NULL, NULL, NULL, &fiform_info };
160 struct qinit fifo_stwdata = { NULL, strwsrv, NULL, NULL, NULL, &fifowm_info };
161 
162 extern kmutex_t	strresources;	/* protects global resources */
163 extern kmutex_t muxifier;	/* single-threads multiplexor creation */
164 
165 static boolean_t msghasdata(mblk_t *bp);
166 #define	msgnodata(bp) (!msghasdata(bp))
167 
168 /*
169  * Stream head locking notes:
170  *	There are four monitors associated with the stream head:
171  *	1. v_stream monitor: in stropen() and strclose() v_lock
172  *		is held while the association of vnode and stream
173  *		head is established or tested for.
174  *	2. open/close/push/pop monitor: sd_lock is held while each
175  *		thread bids for exclusive access to this monitor
176  *		for opening or closing a stream.  In addition, this
177  *		monitor is entered during pushes and pops.  This
178  *		guarantees that during plumbing operations there
179  *		is only one thread trying to change the plumbing.
180  *		Any other threads present in the stream are only
181  *		using the plumbing.
182  *	3. read/write monitor: in the case of read, a thread holds
183  *		sd_lock while trying to get data from the stream
184  *		head queue.  if there is none to fulfill a read
185  *		request, it sets RSLEEP and calls cv_wait_sig() down
186  *		in strwaitq() to await the arrival of new data.
187  *		when new data arrives in strrput(), sd_lock is acquired
188  *		before testing for RSLEEP and calling cv_broadcast().
189  *		the behavior of strwrite(), strwsrv(), and WSLEEP
190  *		mirror this.
191  *	4. ioctl monitor: sd_lock is gotten to ensure that only one
192  *		thread is doing an ioctl at a time.
193  *
194  * Note, for sodirect case 3. is extended to (*sodirect_t.sod_enqueue)()
195  * call-back from below, further the sodirect support is for code paths
196  * called via kstgetmsg(), all other code paths ASSERT() that sodirect
197  * uioa generated mblk_t's (i.e. DBLK_UIOA) aren't processed.
198  */
199 
200 static int
201 push_mod(queue_t *qp, dev_t *devp, struct stdata *stp, const char *name,
202     int anchor, cred_t *crp, uint_t anchor_zoneid)
203 {
204 	int error;
205 	fmodsw_impl_t *fp;
206 
207 	if (stp->sd_flag & (STRHUP|STRDERR|STWRERR)) {
208 		error = (stp->sd_flag & STRHUP) ? ENXIO : EIO;
209 		return (error);
210 	}
211 	if (stp->sd_pushcnt >= nstrpush) {
212 		return (EINVAL);
213 	}
214 
215 	if ((fp = fmodsw_find(name, FMODSW_HOLD | FMODSW_LOAD)) == NULL) {
216 		stp->sd_flag |= STREOPENFAIL;
217 		return (EINVAL);
218 	}
219 
220 	/*
221 	 * push new module and call its open routine via qattach
222 	 */
223 	if ((error = qattach(qp, devp, 0, crp, fp, B_FALSE)) != 0)
224 		return (error);
225 
226 	/*
227 	 * Check to see if caller wants a STREAMS anchor
228 	 * put at this place in the stream, and add if so.
229 	 */
230 	mutex_enter(&stp->sd_lock);
231 	if (anchor == stp->sd_pushcnt) {
232 		stp->sd_anchor = stp->sd_pushcnt;
233 		stp->sd_anchorzone = anchor_zoneid;
234 	}
235 	mutex_exit(&stp->sd_lock);
236 
237 	return (0);
238 }
239 
240 /*
241  * Open a stream device.
242  */
243 int
244 stropen(vnode_t *vp, dev_t *devp, int flag, cred_t *crp)
245 {
246 	struct stdata *stp;
247 	queue_t *qp;
248 	int s;
249 	dev_t dummydev, savedev;
250 	struct autopush *ap;
251 	struct dlautopush dlap;
252 	int error = 0;
253 	ssize_t	rmin, rmax;
254 	int cloneopen;
255 	queue_t *brq;
256 	major_t major;
257 	str_stack_t *ss;
258 	zoneid_t zoneid;
259 	uint_t anchor;
260 
261 	if (audit_active)
262 		audit_stropen(vp, devp, flag, crp);
263 
264 	/*
265 	 * If the stream already exists, wait for any open in progress
266 	 * to complete, then call the open function of each module and
267 	 * driver in the stream.  Otherwise create the stream.
268 	 */
269 	TRACE_1(TR_FAC_STREAMS_FR, TR_STROPEN, "stropen:%p", vp);
270 retry:
271 	mutex_enter(&vp->v_lock);
272 	if ((stp = vp->v_stream) != NULL) {
273 
274 		/*
275 		 * Waiting for stream to be created to device
276 		 * due to another open.
277 		 */
278 		mutex_exit(&vp->v_lock);
279 
280 		if (STRMATED(stp)) {
281 			struct stdata *strmatep = stp->sd_mate;
282 
283 			STRLOCKMATES(stp);
284 			if (strmatep->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
285 				if (flag & (FNDELAY|FNONBLOCK)) {
286 					error = EAGAIN;
287 					mutex_exit(&strmatep->sd_lock);
288 					goto ckreturn;
289 				}
290 				mutex_exit(&stp->sd_lock);
291 				if (!cv_wait_sig(&strmatep->sd_monitor,
292 				    &strmatep->sd_lock)) {
293 					error = EINTR;
294 					mutex_exit(&strmatep->sd_lock);
295 					mutex_enter(&stp->sd_lock);
296 					goto ckreturn;
297 				}
298 				mutex_exit(&strmatep->sd_lock);
299 				goto retry;
300 			}
301 			if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
302 				if (flag & (FNDELAY|FNONBLOCK)) {
303 					error = EAGAIN;
304 					mutex_exit(&strmatep->sd_lock);
305 					goto ckreturn;
306 				}
307 				mutex_exit(&strmatep->sd_lock);
308 				if (!cv_wait_sig(&stp->sd_monitor,
309 				    &stp->sd_lock)) {
310 					error = EINTR;
311 					goto ckreturn;
312 				}
313 				mutex_exit(&stp->sd_lock);
314 				goto retry;
315 			}
316 
317 			if (stp->sd_flag & (STRDERR|STWRERR)) {
318 				error = EIO;
319 				mutex_exit(&strmatep->sd_lock);
320 				goto ckreturn;
321 			}
322 
323 			stp->sd_flag |= STWOPEN;
324 			STRUNLOCKMATES(stp);
325 		} else {
326 			mutex_enter(&stp->sd_lock);
327 			if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
328 				if (flag & (FNDELAY|FNONBLOCK)) {
329 					error = EAGAIN;
330 					goto ckreturn;
331 				}
332 				if (!cv_wait_sig(&stp->sd_monitor,
333 				    &stp->sd_lock)) {
334 					error = EINTR;
335 					goto ckreturn;
336 				}
337 				mutex_exit(&stp->sd_lock);
338 				goto retry;  /* could be clone! */
339 			}
340 
341 			if (stp->sd_flag & (STRDERR|STWRERR)) {
342 				error = EIO;
343 				goto ckreturn;
344 			}
345 
346 			stp->sd_flag |= STWOPEN;
347 			mutex_exit(&stp->sd_lock);
348 		}
349 
350 		/*
351 		 * Open all modules and devices down stream to notify
352 		 * that another user is streaming.  For modules, set the
353 		 * last argument to MODOPEN and do not pass any open flags.
354 		 * Ignore dummydev since this is not the first open.
355 		 */
356 		claimstr(stp->sd_wrq);
357 		qp = stp->sd_wrq;
358 		while (_SAMESTR(qp)) {
359 			qp = qp->q_next;
360 			if ((error = qreopen(_RD(qp), devp, flag, crp)) != 0)
361 				break;
362 		}
363 		releasestr(stp->sd_wrq);
364 		mutex_enter(&stp->sd_lock);
365 		stp->sd_flag &= ~(STRHUP|STWOPEN|STRDERR|STWRERR);
366 		stp->sd_rerror = 0;
367 		stp->sd_werror = 0;
368 ckreturn:
369 		cv_broadcast(&stp->sd_monitor);
370 		mutex_exit(&stp->sd_lock);
371 		return (error);
372 	}
373 
374 	/*
375 	 * This vnode isn't streaming.  SPECFS already
376 	 * checked for multiple vnodes pointing to the
377 	 * same stream, so create a stream to the driver.
378 	 */
379 	qp = allocq();
380 	stp = shalloc(qp);
381 
382 	/*
383 	 * Initialize stream head.  shalloc() has given us
384 	 * exclusive access, and we have the vnode locked;
385 	 * we can do whatever we want with stp.
386 	 */
387 	stp->sd_flag = STWOPEN;
388 	stp->sd_siglist = NULL;
389 	stp->sd_pollist.ph_list = NULL;
390 	stp->sd_sigflags = 0;
391 	stp->sd_mark = NULL;
392 	stp->sd_closetime = STRTIMOUT;
393 	stp->sd_sidp = NULL;
394 	stp->sd_pgidp = NULL;
395 	stp->sd_vnode = vp;
396 	stp->sd_rerror = 0;
397 	stp->sd_werror = 0;
398 	stp->sd_wroff = 0;
399 	stp->sd_tail = 0;
400 	stp->sd_iocblk = NULL;
401 	stp->sd_cmdblk = NULL;
402 	stp->sd_pushcnt = 0;
403 	stp->sd_qn_minpsz = 0;
404 	stp->sd_qn_maxpsz = INFPSZ - 1;	/* used to check for initialization */
405 	stp->sd_maxblk = INFPSZ;
406 	stp->sd_sodirect = NULL;
407 	qp->q_ptr = _WR(qp)->q_ptr = stp;
408 	STREAM(qp) = STREAM(_WR(qp)) = stp;
409 	vp->v_stream = stp;
410 	mutex_exit(&vp->v_lock);
411 	if (vp->v_type == VFIFO) {
412 		stp->sd_flag |= OLDNDELAY;
413 		/*
414 		 * This means, both for pipes and fifos
415 		 * strwrite will send SIGPIPE if the other
416 		 * end is closed. For putmsg it depends
417 		 * on whether it is a XPG4_2 application
418 		 * or not
419 		 */
420 		stp->sd_wput_opt = SW_SIGPIPE;
421 
422 		/* setq might sleep in kmem_alloc - avoid holding locks. */
423 		setq(qp, &fifo_strdata, &fifo_stwdata, NULL, QMTSAFE,
424 		    SQ_CI|SQ_CO, B_FALSE);
425 
426 		set_qend(qp);
427 		stp->sd_strtab = fifo_getinfo();
428 		_WR(qp)->q_nfsrv = _WR(qp);
429 		qp->q_nfsrv = qp;
430 		/*
431 		 * Wake up others that are waiting for stream to be created.
432 		 */
433 		mutex_enter(&stp->sd_lock);
434 		/*
435 		 * nothing is be pushed on stream yet, so
436 		 * optimized stream head packetsizes are just that
437 		 * of the read queue
438 		 */
439 		stp->sd_qn_minpsz = qp->q_minpsz;
440 		stp->sd_qn_maxpsz = qp->q_maxpsz;
441 		stp->sd_flag &= ~STWOPEN;
442 		goto fifo_opendone;
443 	}
444 	/* setq might sleep in kmem_alloc - avoid holding locks. */
445 	setq(qp, &strdata, &stwdata, NULL, QMTSAFE, SQ_CI|SQ_CO, B_FALSE);
446 
447 	set_qend(qp);
448 
449 	/*
450 	 * Open driver and create stream to it (via qattach).
451 	 */
452 	savedev = *devp;
453 	cloneopen = (getmajor(*devp) == clone_major);
454 	if ((error = qattach(qp, devp, flag, crp, NULL, B_FALSE)) != 0) {
455 		mutex_enter(&vp->v_lock);
456 		vp->v_stream = NULL;
457 		mutex_exit(&vp->v_lock);
458 		mutex_enter(&stp->sd_lock);
459 		cv_broadcast(&stp->sd_monitor);
460 		mutex_exit(&stp->sd_lock);
461 		freeq(_RD(qp));
462 		shfree(stp);
463 		return (error);
464 	}
465 	/*
466 	 * Set sd_strtab after open in order to handle clonable drivers
467 	 */
468 	stp->sd_strtab = STREAMSTAB(getmajor(*devp));
469 
470 	/*
471 	 * Historical note: dummydev used to be be prior to the initial
472 	 * open (via qattach above), which made the value seen
473 	 * inconsistent between an I_PUSH and an autopush of a module.
474 	 */
475 	dummydev = *devp;
476 
477 	/*
478 	 * For clone open of old style (Q not associated) network driver,
479 	 * push DRMODNAME module to handle DL_ATTACH/DL_DETACH
480 	 */
481 	brq = _RD(_WR(qp)->q_next);
482 	major = getmajor(*devp);
483 	if (push_drcompat && cloneopen && NETWORK_DRV(major) &&
484 	    ((brq->q_flag & _QASSOCIATED) == 0)) {
485 		if (push_mod(qp, &dummydev, stp, DRMODNAME, 0, crp, 0) != 0)
486 			cmn_err(CE_WARN, "cannot push " DRMODNAME
487 			    " streams module");
488 	}
489 
490 	if (!NETWORK_DRV(major)) {
491 		savedev = *devp;
492 	} else {
493 		/*
494 		 * For network devices, process differently based on the
495 		 * return value from dld_autopush():
496 		 *
497 		 *   0: the passed-in device points to a GLDv3 datalink with
498 		 *   per-link autopush configuration; use that configuration
499 		 *   and ignore any per-driver autopush configuration.
500 		 *
501 		 *   1: the passed-in device points to a physical GLDv3
502 		 *   datalink without per-link autopush configuration.  The
503 		 *   passed in device was changed to refer to the actual
504 		 *   physical device (if it's not already); we use that new
505 		 *   device to look up any per-driver autopush configuration.
506 		 *
507 		 *   -1: neither of the above cases applied; use the initial
508 		 *   device to look up any per-driver autopush configuration.
509 		 */
510 		switch (dld_autopush(&savedev, &dlap)) {
511 		case 0:
512 			zoneid = crgetzoneid(crp);
513 			for (s = 0; s < dlap.dap_npush; s++) {
514 				error = push_mod(qp, &dummydev, stp,
515 				    dlap.dap_aplist[s], dlap.dap_anchor, crp,
516 				    zoneid);
517 				if (error != 0)
518 					break;
519 			}
520 			goto opendone;
521 		case 1:
522 			break;
523 		case -1:
524 			savedev = *devp;
525 			break;
526 		}
527 	}
528 	/*
529 	 * Find the autopush configuration based on "savedev". Start with the
530 	 * global zone. If not found check in the local zone.
531 	 */
532 	zoneid = GLOBAL_ZONEID;
533 retryap:
534 	ss = netstack_find_by_stackid(zoneid_to_netstackid(zoneid))->
535 	    netstack_str;
536 	if ((ap = sad_ap_find_by_dev(savedev, ss)) == NULL) {
537 		netstack_rele(ss->ss_netstack);
538 		if (zoneid == GLOBAL_ZONEID) {
539 			/*
540 			 * None found. Also look in the zone's autopush table.
541 			 */
542 			zoneid = crgetzoneid(crp);
543 			if (zoneid != GLOBAL_ZONEID)
544 				goto retryap;
545 		}
546 		goto opendone;
547 	}
548 	anchor = ap->ap_anchor;
549 	zoneid = crgetzoneid(crp);
550 	for (s = 0; s < ap->ap_npush; s++) {
551 		error = push_mod(qp, &dummydev, stp, ap->ap_list[s],
552 		    anchor, crp, zoneid);
553 		if (error != 0)
554 			break;
555 	}
556 	sad_ap_rele(ap, ss);
557 	netstack_rele(ss->ss_netstack);
558 
559 opendone:
560 
561 	/*
562 	 * let specfs know that open failed part way through
563 	 */
564 	if (error) {
565 		mutex_enter(&stp->sd_lock);
566 		stp->sd_flag |= STREOPENFAIL;
567 		mutex_exit(&stp->sd_lock);
568 	}
569 
570 	/*
571 	 * Wake up others that are waiting for stream to be created.
572 	 */
573 	mutex_enter(&stp->sd_lock);
574 	stp->sd_flag &= ~STWOPEN;
575 
576 	/*
577 	 * As a performance concern we are caching the values of
578 	 * q_minpsz and q_maxpsz of the module below the stream
579 	 * head in the stream head.
580 	 */
581 	mutex_enter(QLOCK(stp->sd_wrq->q_next));
582 	rmin = stp->sd_wrq->q_next->q_minpsz;
583 	rmax = stp->sd_wrq->q_next->q_maxpsz;
584 	mutex_exit(QLOCK(stp->sd_wrq->q_next));
585 
586 	/* do this processing here as a performance concern */
587 	if (strmsgsz != 0) {
588 		if (rmax == INFPSZ)
589 			rmax = strmsgsz;
590 		else
591 			rmax = MIN(strmsgsz, rmax);
592 	}
593 
594 	mutex_enter(QLOCK(stp->sd_wrq));
595 	stp->sd_qn_minpsz = rmin;
596 	stp->sd_qn_maxpsz = rmax;
597 	mutex_exit(QLOCK(stp->sd_wrq));
598 
599 fifo_opendone:
600 	cv_broadcast(&stp->sd_monitor);
601 	mutex_exit(&stp->sd_lock);
602 	return (error);
603 }
604 
605 static int strsink(queue_t *, mblk_t *);
606 static struct qinit deadrend = {
607 	strsink, NULL, NULL, NULL, NULL, &strm_info, NULL
608 };
609 static struct qinit deadwend = {
610 	NULL, NULL, NULL, NULL, NULL, &stwm_info, NULL
611 };
612 
613 /*
614  * Close a stream.
615  * This is called from closef() on the last close of an open stream.
616  * Strclean() will already have removed the siglist and pollist
617  * information, so all that remains is to remove all multiplexor links
618  * for the stream, pop all the modules (and the driver), and free the
619  * stream structure.
620  */
621 
622 int
623 strclose(struct vnode *vp, int flag, cred_t *crp)
624 {
625 	struct stdata *stp;
626 	queue_t *qp;
627 	int rval;
628 	int freestp = 1;
629 	queue_t *rmq;
630 
631 	if (audit_active)
632 		audit_strclose(vp, flag, crp);
633 
634 	TRACE_1(TR_FAC_STREAMS_FR,
635 	    TR_STRCLOSE, "strclose:%p", vp);
636 	ASSERT(vp->v_stream);
637 
638 	stp = vp->v_stream;
639 	ASSERT(!(stp->sd_flag & STPLEX));
640 	qp = stp->sd_wrq;
641 
642 	/*
643 	 * Needed so that strpoll will return non-zero for this fd.
644 	 * Note that with POLLNOERR STRHUP does still cause POLLHUP.
645 	 */
646 	mutex_enter(&stp->sd_lock);
647 	stp->sd_flag |= STRHUP;
648 	mutex_exit(&stp->sd_lock);
649 
650 	/*
651 	 * If the registered process or process group did not have an
652 	 * open instance of this stream then strclean would not be
653 	 * called. Thus at the time of closing all remaining siglist entries
654 	 * are removed.
655 	 */
656 	if (stp->sd_siglist != NULL)
657 		strcleanall(vp);
658 
659 	ASSERT(stp->sd_siglist == NULL);
660 	ASSERT(stp->sd_sigflags == 0);
661 
662 	if (STRMATED(stp)) {
663 		struct stdata *strmatep = stp->sd_mate;
664 		int waited = 1;
665 
666 		STRLOCKMATES(stp);
667 		while (waited) {
668 			waited = 0;
669 			while (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
670 				mutex_exit(&strmatep->sd_lock);
671 				cv_wait(&stp->sd_monitor, &stp->sd_lock);
672 				mutex_exit(&stp->sd_lock);
673 				STRLOCKMATES(stp);
674 				waited = 1;
675 			}
676 			while (strmatep->sd_flag &
677 			    (STWOPEN|STRCLOSE|STRPLUMB)) {
678 				mutex_exit(&stp->sd_lock);
679 				cv_wait(&strmatep->sd_monitor,
680 				    &strmatep->sd_lock);
681 				mutex_exit(&strmatep->sd_lock);
682 				STRLOCKMATES(stp);
683 				waited = 1;
684 			}
685 		}
686 		stp->sd_flag |= STRCLOSE;
687 		STRUNLOCKMATES(stp);
688 	} else {
689 		mutex_enter(&stp->sd_lock);
690 		stp->sd_flag |= STRCLOSE;
691 		mutex_exit(&stp->sd_lock);
692 	}
693 
694 	ASSERT(qp->q_first == NULL);	/* No more delayed write */
695 
696 	/* Check if an I_LINK was ever done on this stream */
697 	if (stp->sd_flag & STRHASLINKS) {
698 		netstack_t *ns;
699 		str_stack_t *ss;
700 
701 		ns = netstack_find_by_cred(crp);
702 		ASSERT(ns != NULL);
703 		ss = ns->netstack_str;
704 		ASSERT(ss != NULL);
705 
706 		(void) munlinkall(stp, LINKCLOSE|LINKNORMAL, crp, &rval, ss);
707 		netstack_rele(ss->ss_netstack);
708 	}
709 
710 	while (_SAMESTR(qp)) {
711 		/*
712 		 * Holding sd_lock prevents q_next from changing in
713 		 * this stream.
714 		 */
715 		mutex_enter(&stp->sd_lock);
716 		if (!(flag & (FNDELAY|FNONBLOCK)) && (stp->sd_closetime > 0)) {
717 
718 			/*
719 			 * sleep until awakened by strwsrv() or timeout
720 			 */
721 			for (;;) {
722 				mutex_enter(QLOCK(qp->q_next));
723 				if (!(qp->q_next->q_mblkcnt)) {
724 					mutex_exit(QLOCK(qp->q_next));
725 					break;
726 				}
727 				stp->sd_flag |= WSLEEP;
728 
729 				/* ensure strwsrv gets enabled */
730 				qp->q_next->q_flag |= QWANTW;
731 				mutex_exit(QLOCK(qp->q_next));
732 				/* get out if we timed out or recv'd a signal */
733 				if (str_cv_wait(&qp->q_wait, &stp->sd_lock,
734 				    stp->sd_closetime, 0) <= 0) {
735 					break;
736 				}
737 			}
738 			stp->sd_flag &= ~WSLEEP;
739 		}
740 		mutex_exit(&stp->sd_lock);
741 
742 		rmq = qp->q_next;
743 		if (rmq->q_flag & QISDRV) {
744 			ASSERT(!_SAMESTR(rmq));
745 			wait_sq_svc(_RD(qp)->q_syncq);
746 		}
747 
748 		qdetach(_RD(rmq), 1, flag, crp, B_FALSE);
749 	}
750 
751 	/*
752 	 * Since we call pollwakeup in close() now, the poll list should
753 	 * be empty in most cases. The only exception is the layered devices
754 	 * (e.g. the console drivers with redirection modules pushed on top
755 	 * of it).  We have to do this after calling qdetach() because
756 	 * the redirection module won't have torn down the console
757 	 * redirection until after qdetach() has been invoked.
758 	 */
759 	if (stp->sd_pollist.ph_list != NULL) {
760 		pollwakeup(&stp->sd_pollist, POLLERR);
761 		pollhead_clean(&stp->sd_pollist);
762 	}
763 	ASSERT(stp->sd_pollist.ph_list == NULL);
764 	ASSERT(stp->sd_sidp == NULL);
765 	ASSERT(stp->sd_pgidp == NULL);
766 
767 	/* Prevent qenable from re-enabling the stream head queue */
768 	disable_svc(_RD(qp));
769 
770 	/*
771 	 * Wait until service procedure of each queue is
772 	 * run, if QINSERVICE is set.
773 	 */
774 	wait_svc(_RD(qp));
775 
776 	/*
777 	 * Now, flush both queues.
778 	 */
779 	flushq(_RD(qp), FLUSHALL);
780 	flushq(qp, FLUSHALL);
781 
782 	/*
783 	 * If the write queue of the stream head is pointing to a
784 	 * read queue, we have a twisted stream.  If the read queue
785 	 * is alive, convert the stream head queues into a dead end.
786 	 * If the read queue is dead, free the dead pair.
787 	 */
788 	if (qp->q_next && !_SAMESTR(qp)) {
789 		if (qp->q_next->q_qinfo == &deadrend) {	/* half-closed pipe */
790 			flushq(qp->q_next, FLUSHALL); /* ensure no message */
791 			shfree(qp->q_next->q_stream);
792 			freeq(qp->q_next);
793 			freeq(_RD(qp));
794 		} else if (qp->q_next == _RD(qp)) {	/* fifo */
795 			freeq(_RD(qp));
796 		} else {				/* pipe */
797 			freestp = 0;
798 			/*
799 			 * The q_info pointers are never accessed when
800 			 * SQLOCK is held.
801 			 */
802 			ASSERT(qp->q_syncq == _RD(qp)->q_syncq);
803 			mutex_enter(SQLOCK(qp->q_syncq));
804 			qp->q_qinfo = &deadwend;
805 			_RD(qp)->q_qinfo = &deadrend;
806 			mutex_exit(SQLOCK(qp->q_syncq));
807 		}
808 	} else {
809 		freeq(_RD(qp)); /* free stream head queue pair */
810 	}
811 
812 	mutex_enter(&vp->v_lock);
813 	if (stp->sd_iocblk) {
814 		if (stp->sd_iocblk != (mblk_t *)-1) {
815 			freemsg(stp->sd_iocblk);
816 		}
817 		stp->sd_iocblk = NULL;
818 	}
819 	stp->sd_vnode = NULL;
820 	vp->v_stream = NULL;
821 	mutex_exit(&vp->v_lock);
822 	mutex_enter(&stp->sd_lock);
823 	freemsg(stp->sd_cmdblk);
824 	stp->sd_cmdblk = NULL;
825 	stp->sd_flag &= ~STRCLOSE;
826 	cv_broadcast(&stp->sd_monitor);
827 	mutex_exit(&stp->sd_lock);
828 
829 	if (freestp)
830 		shfree(stp);
831 	return (0);
832 }
833 
834 static int
835 strsink(queue_t *q, mblk_t *bp)
836 {
837 	struct copyresp *resp;
838 
839 	switch (bp->b_datap->db_type) {
840 	case M_FLUSH:
841 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
842 			*bp->b_rptr &= ~FLUSHR;
843 			bp->b_flag |= MSGNOLOOP;
844 			/*
845 			 * Protect against the driver passing up
846 			 * messages after it has done a qprocsoff.
847 			 */
848 			if (_OTHERQ(q)->q_next == NULL)
849 				freemsg(bp);
850 			else
851 				qreply(q, bp);
852 		} else {
853 			freemsg(bp);
854 		}
855 		break;
856 
857 	case M_COPYIN:
858 	case M_COPYOUT:
859 		if (bp->b_cont) {
860 			freemsg(bp->b_cont);
861 			bp->b_cont = NULL;
862 		}
863 		bp->b_datap->db_type = M_IOCDATA;
864 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
865 		resp = (struct copyresp *)bp->b_rptr;
866 		resp->cp_rval = (caddr_t)1;	/* failure */
867 		/*
868 		 * Protect against the driver passing up
869 		 * messages after it has done a qprocsoff.
870 		 */
871 		if (_OTHERQ(q)->q_next == NULL)
872 			freemsg(bp);
873 		else
874 			qreply(q, bp);
875 		break;
876 
877 	case M_IOCTL:
878 		if (bp->b_cont) {
879 			freemsg(bp->b_cont);
880 			bp->b_cont = NULL;
881 		}
882 		bp->b_datap->db_type = M_IOCNAK;
883 		/*
884 		 * Protect against the driver passing up
885 		 * messages after it has done a qprocsoff.
886 		 */
887 		if (_OTHERQ(q)->q_next == NULL)
888 			freemsg(bp);
889 		else
890 			qreply(q, bp);
891 		break;
892 
893 	default:
894 		freemsg(bp);
895 		break;
896 	}
897 
898 	return (0);
899 }
900 
901 /*
902  * Clean up after a process when it closes a stream.  This is called
903  * from closef for all closes, whereas strclose is called only for the
904  * last close on a stream.  The siglist is scanned for entries for the
905  * current process, and these are removed.
906  */
907 void
908 strclean(struct vnode *vp)
909 {
910 	strsig_t *ssp, *pssp, *tssp;
911 	stdata_t *stp;
912 	int update = 0;
913 
914 	TRACE_1(TR_FAC_STREAMS_FR,
915 	    TR_STRCLEAN, "strclean:%p", vp);
916 	stp = vp->v_stream;
917 	pssp = NULL;
918 	mutex_enter(&stp->sd_lock);
919 	ssp = stp->sd_siglist;
920 	while (ssp) {
921 		if (ssp->ss_pidp == curproc->p_pidp) {
922 			tssp = ssp->ss_next;
923 			if (pssp)
924 				pssp->ss_next = tssp;
925 			else
926 				stp->sd_siglist = tssp;
927 			mutex_enter(&pidlock);
928 			PID_RELE(ssp->ss_pidp);
929 			mutex_exit(&pidlock);
930 			kmem_free(ssp, sizeof (strsig_t));
931 			update = 1;
932 			ssp = tssp;
933 		} else {
934 			pssp = ssp;
935 			ssp = ssp->ss_next;
936 		}
937 	}
938 	if (update) {
939 		stp->sd_sigflags = 0;
940 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
941 			stp->sd_sigflags |= ssp->ss_events;
942 	}
943 	mutex_exit(&stp->sd_lock);
944 }
945 
946 /*
947  * Used on the last close to remove any remaining items on the siglist.
948  * These could be present on the siglist due to I_ESETSIG calls that
949  * use process groups or processed that do not have an open file descriptor
950  * for this stream (Such entries would not be removed by strclean).
951  */
952 static void
953 strcleanall(struct vnode *vp)
954 {
955 	strsig_t *ssp, *nssp;
956 	stdata_t *stp;
957 
958 	stp = vp->v_stream;
959 	mutex_enter(&stp->sd_lock);
960 	ssp = stp->sd_siglist;
961 	stp->sd_siglist = NULL;
962 	while (ssp) {
963 		nssp = ssp->ss_next;
964 		mutex_enter(&pidlock);
965 		PID_RELE(ssp->ss_pidp);
966 		mutex_exit(&pidlock);
967 		kmem_free(ssp, sizeof (strsig_t));
968 		ssp = nssp;
969 	}
970 	stp->sd_sigflags = 0;
971 	mutex_exit(&stp->sd_lock);
972 }
973 
974 /*
975  * Retrieve the next message from the logical stream head read queue
976  * using either rwnext (if sync stream) or getq_noenab.
977  * It is the callers responsibility to call qbackenable after
978  * it is finished with the message. The caller should not call
979  * qbackenable until after any putback calls to avoid spurious backenabling.
980  *
981  * Also, handle uioa initialization and process any DBLK_UIOA flaged messages.
982  */
983 mblk_t *
984 strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
985     int *errorp)
986 {
987 	sodirect_t *sodp = stp->sd_sodirect;
988 	mblk_t *bp;
989 	int error;
990 
991 	ASSERT(MUTEX_HELD(&stp->sd_lock));
992 	/* Holding sd_lock prevents the read queue from changing  */
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 	if (sodp != NULL && (sodp->sod_state & SOD_ENABLED) &&
1077 	    (sodp->sod_uioa.uioa_state & UIOA_INIT)) {
1078 		/*
1079 		 * First kstrgetmsg() call for an uioa_t so if any
1080 		 * queued mblk_t's need to consume them before uioa
1081 		 * from below can occur.
1082 		 */
1083 		sodp->sod_uioa.uioa_state &= UIOA_CLR;
1084 		sodp->sod_uioa.uioa_state |= UIOA_ENABLED;
1085 		if (q->q_first != NULL) {
1086 			struioainit(q, sodp, uiop);
1087 		}
1088 	}
1089 
1090 	bp = getq_noenab(q);
1091 
1092 	if (bp != NULL && (bp->b_datap->db_flags & DBLK_UIOA)) {
1093 		/*
1094 		 * A uioa flaged mblk_t chain, already uio processed,
1095 		 * add it to the sodirect uioa pending free list.
1096 		 *
1097 		 * Note, a b_cont chain headed by a DBLK_UIOA enable
1098 		 * mblk_t must have all mblk_t(s) DBLK_UIOA enabled.
1099 		 */
1100 		mblk_t	*bpt = sodp->sod_uioaft;
1101 
1102 		ASSERT(sodp != NULL);
1103 
1104 		/*
1105 		 * Add first mblk_t of "bp" chain to current sodirect uioa
1106 		 * free list tail mblk_t, if any, else empty list so new head.
1107 		 */
1108 		if (bpt == NULL)
1109 			sodp->sod_uioafh = bp;
1110 		else
1111 			bpt->b_cont = bp;
1112 
1113 		/*
1114 		 * Walk mblk_t "bp" chain to find tail and adjust rptr of
1115 		 * each to reflect that uioamove() has consumed all data.
1116 		 */
1117 		bpt = bp;
1118 		for (;;) {
1119 			bpt->b_rptr = bpt->b_wptr;
1120 			if (bpt->b_cont == NULL)
1121 				break;
1122 			bpt = bpt->b_cont;
1123 
1124 			ASSERT(bpt->b_datap->db_flags & DBLK_UIOA);
1125 		}
1126 		/* New sodirect uioa free list tail */
1127 		sodp->sod_uioaft = bpt;
1128 
1129 		/* Only 1 strget() with data returned per uioa_t */
1130 		if (sodp->sod_uioa.uioa_state & UIOA_ENABLED) {
1131 			sodp->sod_uioa.uioa_state &= UIOA_CLR;
1132 			sodp->sod_uioa.uioa_state |= UIOA_FINI;
1133 		}
1134 	}
1135 
1136 	return (bp);
1137 }
1138 
1139 /*
1140  * Copy out the message pointed to by `bp' into the uio pointed to by `uiop'.
1141  * If the message does not fit in the uio the remainder of it is returned;
1142  * otherwise NULL is returned.  Any embedded zero-length mblk_t's are
1143  * consumed, even if uio_resid reaches zero.  On error, `*errorp' is set to
1144  * the error code, the message is consumed, and NULL is returned.
1145  */
1146 static mblk_t *
1147 struiocopyout(mblk_t *bp, struct uio *uiop, int *errorp)
1148 {
1149 	int error;
1150 	ptrdiff_t n;
1151 	mblk_t *nbp;
1152 
1153 	ASSERT(bp->b_wptr >= bp->b_rptr);
1154 
1155 	do {
1156 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
1157 
1158 		if ((n = MIN(uiop->uio_resid, MBLKL(bp))) != 0) {
1159 			ASSERT(n > 0);
1160 
1161 			error = uiomove(bp->b_rptr, n, UIO_READ, uiop);
1162 			if (error != 0) {
1163 				freemsg(bp);
1164 				*errorp = error;
1165 				return (NULL);
1166 			}
1167 		}
1168 
1169 		bp->b_rptr += n;
1170 		while (bp != NULL && (bp->b_rptr >= bp->b_wptr)) {
1171 			nbp = bp;
1172 			bp = bp->b_cont;
1173 			freeb(nbp);
1174 		}
1175 	} while (bp != NULL && uiop->uio_resid > 0);
1176 
1177 	*errorp = 0;
1178 	return (bp);
1179 }
1180 
1181 /*
1182  * Read a stream according to the mode flags in sd_flag:
1183  *
1184  * (default mode)		- Byte stream, msg boundaries are ignored
1185  * RD_MSGDIS (msg discard)	- Read on msg boundaries and throw away
1186  *				any data remaining in msg
1187  * RD_MSGNODIS (msg non-discard) - Read on msg boundaries and put back
1188  *				any remaining data on head of read queue
1189  *
1190  * Consume readable messages on the front of the queue until
1191  * ttolwp(curthread)->lwp_count
1192  * is satisfied, the readable messages are exhausted, or a message
1193  * boundary is reached in a message mode.  If no data was read and
1194  * the stream was not opened with the NDELAY flag, block until data arrives.
1195  * Otherwise return the data read and update the count.
1196  *
1197  * In default mode a 0 length message signifies end-of-file and terminates
1198  * a read in progress.  The 0 length message is removed from the queue
1199  * only if it is the only message read (no data is read).
1200  *
1201  * An attempt to read an M_PROTO or M_PCPROTO message results in an
1202  * EBADMSG error return, unless either RD_PROTDAT or RD_PROTDIS are set.
1203  * If RD_PROTDAT is set, M_PROTO and M_PCPROTO messages are read as data.
1204  * If RD_PROTDIS is set, the M_PROTO and M_PCPROTO parts of the message
1205  * are unlinked from and M_DATA blocks in the message, the protos are
1206  * thrown away, and the data is read.
1207  */
1208 /* ARGSUSED */
1209 int
1210 strread(struct vnode *vp, struct uio *uiop, cred_t *crp)
1211 {
1212 	struct stdata *stp;
1213 	mblk_t *bp, *nbp;
1214 	queue_t *q;
1215 	int error = 0;
1216 	uint_t old_sd_flag;
1217 	int first;
1218 	char rflg;
1219 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
1220 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
1221 	short delim;
1222 	unsigned char pri = 0;
1223 	char waitflag;
1224 	unsigned char type;
1225 
1226 	TRACE_1(TR_FAC_STREAMS_FR,
1227 	    TR_STRREAD_ENTER, "strread:%p", vp);
1228 	ASSERT(vp->v_stream);
1229 	stp = vp->v_stream;
1230 
1231 	mutex_enter(&stp->sd_lock);
1232 
1233 	if ((error = i_straccess(stp, JCREAD)) != 0) {
1234 		mutex_exit(&stp->sd_lock);
1235 		return (error);
1236 	}
1237 
1238 	if (stp->sd_flag & (STRDERR|STPLEX)) {
1239 		error = strgeterr(stp, STRDERR|STPLEX, 0);
1240 		if (error != 0) {
1241 			mutex_exit(&stp->sd_lock);
1242 			return (error);
1243 		}
1244 	}
1245 
1246 	/*
1247 	 * Loop terminates when uiop->uio_resid == 0.
1248 	 */
1249 	rflg = 0;
1250 	waitflag = READWAIT;
1251 	q = _RD(stp->sd_wrq);
1252 	for (;;) {
1253 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1254 		old_sd_flag = stp->sd_flag;
1255 		mark = 0;
1256 		delim = 0;
1257 		first = 1;
1258 		while ((bp = strget(stp, q, uiop, first, &error)) == NULL) {
1259 			int done = 0;
1260 
1261 			ASSERT(MUTEX_HELD(&stp->sd_lock));
1262 
1263 			if (error != 0)
1264 				goto oops;
1265 
1266 			if (stp->sd_flag & (STRHUP|STREOF)) {
1267 				goto oops;
1268 			}
1269 			if (rflg && !(stp->sd_flag & STRDELIM)) {
1270 				goto oops;
1271 			}
1272 			/*
1273 			 * If a read(fd,buf,0) has been done, there is no
1274 			 * need to sleep. We always have zero bytes to
1275 			 * return.
1276 			 */
1277 			if (uiop->uio_resid == 0) {
1278 				goto oops;
1279 			}
1280 
1281 			qbackenable(q, 0);
1282 
1283 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_WAIT,
1284 			    "strread calls strwaitq:%p, %p, %p",
1285 			    vp, uiop, crp);
1286 			if ((error = strwaitq(stp, waitflag, uiop->uio_resid,
1287 			    uiop->uio_fmode, -1, &done)) != 0 || done) {
1288 				TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_DONE,
1289 				    "strread error or done:%p, %p, %p",
1290 				    vp, uiop, crp);
1291 				if ((uiop->uio_fmode & FNDELAY) &&
1292 				    (stp->sd_flag & OLDNDELAY) &&
1293 				    (error == EAGAIN))
1294 					error = 0;
1295 				goto oops;
1296 			}
1297 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_AWAKE,
1298 			    "strread awakes:%p, %p, %p", vp, uiop, crp);
1299 			if ((error = i_straccess(stp, JCREAD)) != 0) {
1300 				goto oops;
1301 			}
1302 			first = 0;
1303 		}
1304 
1305 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1306 		ASSERT(bp);
1307 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
1308 		pri = bp->b_band;
1309 		/*
1310 		 * Extract any mark information. If the message is not
1311 		 * completely consumed this information will be put in the mblk
1312 		 * that is putback.
1313 		 * If MSGMARKNEXT is set and the message is completely consumed
1314 		 * the STRATMARK flag will be set below. Likewise, if
1315 		 * MSGNOTMARKNEXT is set and the message is
1316 		 * completely consumed STRNOTATMARK will be set.
1317 		 *
1318 		 * For some unknown reason strread only breaks the read at the
1319 		 * last mark.
1320 		 */
1321 		mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
1322 		ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
1323 		    (MSGMARKNEXT|MSGNOTMARKNEXT));
1324 		if (mark != 0 && bp == stp->sd_mark) {
1325 			if (rflg) {
1326 				putback(stp, q, bp, pri);
1327 				goto oops;
1328 			}
1329 			mark |= _LASTMARK;
1330 			stp->sd_mark = NULL;
1331 		}
1332 		if ((stp->sd_flag & STRDELIM) && (bp->b_flag & MSGDELIM))
1333 			delim = 1;
1334 		mutex_exit(&stp->sd_lock);
1335 
1336 		if (STREAM_NEEDSERVICE(stp))
1337 			stream_runservice(stp);
1338 
1339 		type = bp->b_datap->db_type;
1340 
1341 		switch (type) {
1342 
1343 		case M_DATA:
1344 ismdata:
1345 			if (msgnodata(bp)) {
1346 				if (mark || delim) {
1347 					freemsg(bp);
1348 				} else if (rflg) {
1349 
1350 					/*
1351 					 * If already read data put zero
1352 					 * length message back on queue else
1353 					 * free msg and return 0.
1354 					 */
1355 					bp->b_band = pri;
1356 					mutex_enter(&stp->sd_lock);
1357 					putback(stp, q, bp, pri);
1358 					mutex_exit(&stp->sd_lock);
1359 				} else {
1360 					freemsg(bp);
1361 				}
1362 				error =  0;
1363 				goto oops1;
1364 			}
1365 
1366 			rflg = 1;
1367 			waitflag |= NOINTR;
1368 			bp = struiocopyout(bp, uiop, &error);
1369 			if (error != 0)
1370 				goto oops1;
1371 
1372 			mutex_enter(&stp->sd_lock);
1373 			if (bp) {
1374 				/*
1375 				 * Have remaining data in message.
1376 				 * Free msg if in discard mode.
1377 				 */
1378 				if (stp->sd_read_opt & RD_MSGDIS) {
1379 					freemsg(bp);
1380 				} else {
1381 					bp->b_band = pri;
1382 					if ((mark & _LASTMARK) &&
1383 					    (stp->sd_mark == NULL))
1384 						stp->sd_mark = bp;
1385 					bp->b_flag |= mark & ~_LASTMARK;
1386 					if (delim)
1387 						bp->b_flag |= MSGDELIM;
1388 					if (msgnodata(bp))
1389 						freemsg(bp);
1390 					else
1391 						putback(stp, q, bp, pri);
1392 				}
1393 			} else {
1394 				/*
1395 				 * Consumed the complete message.
1396 				 * Move the MSG*MARKNEXT information
1397 				 * to the stream head just in case
1398 				 * the read queue becomes empty.
1399 				 *
1400 				 * If the stream head was at the mark
1401 				 * (STRATMARK) before we dropped sd_lock above
1402 				 * and some data was consumed then we have
1403 				 * moved past the mark thus STRATMARK is
1404 				 * cleared. However, if a message arrived in
1405 				 * strrput during the copyout above causing
1406 				 * STRATMARK to be set we can not clear that
1407 				 * flag.
1408 				 */
1409 				if (mark &
1410 				    (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
1411 					if (mark & MSGMARKNEXT) {
1412 						stp->sd_flag &= ~STRNOTATMARK;
1413 						stp->sd_flag |= STRATMARK;
1414 					} else if (mark & MSGNOTMARKNEXT) {
1415 						stp->sd_flag &= ~STRATMARK;
1416 						stp->sd_flag |= STRNOTATMARK;
1417 					} else {
1418 						stp->sd_flag &=
1419 						    ~(STRATMARK|STRNOTATMARK);
1420 					}
1421 				} else if (rflg && (old_sd_flag & STRATMARK)) {
1422 					stp->sd_flag &= ~STRATMARK;
1423 				}
1424 			}
1425 
1426 			/*
1427 			 * Check for signal messages at the front of the read
1428 			 * queue and generate the signal(s) if appropriate.
1429 			 * The only signal that can be on queue is M_SIG at
1430 			 * this point.
1431 			 */
1432 			while ((((bp = q->q_first)) != NULL) &&
1433 			    (bp->b_datap->db_type == M_SIG)) {
1434 				bp = getq_noenab(q);
1435 				/*
1436 				 * sd_lock is held so the content of the
1437 				 * read queue can not change.
1438 				 */
1439 				ASSERT(bp != NULL &&
1440 				    bp->b_datap->db_type == M_SIG);
1441 				strsignal_nolock(stp, *bp->b_rptr,
1442 				    (int32_t)bp->b_band);
1443 				mutex_exit(&stp->sd_lock);
1444 				freemsg(bp);
1445 				if (STREAM_NEEDSERVICE(stp))
1446 					stream_runservice(stp);
1447 				mutex_enter(&stp->sd_lock);
1448 			}
1449 
1450 			if ((uiop->uio_resid == 0) || (mark & _LASTMARK) ||
1451 			    delim ||
1452 			    (stp->sd_read_opt & (RD_MSGDIS|RD_MSGNODIS))) {
1453 				goto oops;
1454 			}
1455 			continue;
1456 
1457 		case M_SIG:
1458 			strsignal(stp, *bp->b_rptr, (int32_t)bp->b_band);
1459 			freemsg(bp);
1460 			mutex_enter(&stp->sd_lock);
1461 			continue;
1462 
1463 		case M_PROTO:
1464 		case M_PCPROTO:
1465 			/*
1466 			 * Only data messages are readable.
1467 			 * Any others generate an error, unless
1468 			 * RD_PROTDIS or RD_PROTDAT is set.
1469 			 */
1470 			if (stp->sd_read_opt & RD_PROTDAT) {
1471 				for (nbp = bp; nbp; nbp = nbp->b_next) {
1472 					if ((nbp->b_datap->db_type ==
1473 					    M_PROTO) ||
1474 					    (nbp->b_datap->db_type ==
1475 					    M_PCPROTO)) {
1476 						nbp->b_datap->db_type = M_DATA;
1477 					} else {
1478 						break;
1479 					}
1480 				}
1481 				/*
1482 				 * clear stream head hi pri flag based on
1483 				 * first message
1484 				 */
1485 				if (type == M_PCPROTO) {
1486 					mutex_enter(&stp->sd_lock);
1487 					stp->sd_flag &= ~STRPRI;
1488 					mutex_exit(&stp->sd_lock);
1489 				}
1490 				goto ismdata;
1491 			} else if (stp->sd_read_opt & RD_PROTDIS) {
1492 				/*
1493 				 * discard non-data messages
1494 				 */
1495 				while (bp &&
1496 				    ((bp->b_datap->db_type == M_PROTO) ||
1497 				    (bp->b_datap->db_type == M_PCPROTO))) {
1498 					nbp = unlinkb(bp);
1499 					freeb(bp);
1500 					bp = nbp;
1501 				}
1502 				/*
1503 				 * clear stream head hi pri flag based on
1504 				 * first message
1505 				 */
1506 				if (type == M_PCPROTO) {
1507 					mutex_enter(&stp->sd_lock);
1508 					stp->sd_flag &= ~STRPRI;
1509 					mutex_exit(&stp->sd_lock);
1510 				}
1511 				if (bp) {
1512 					bp->b_band = pri;
1513 					goto ismdata;
1514 				} else {
1515 					break;
1516 				}
1517 			}
1518 			/* FALLTHRU */
1519 		case M_PASSFP:
1520 			if ((bp->b_datap->db_type == M_PASSFP) &&
1521 			    (stp->sd_read_opt & RD_PROTDIS)) {
1522 				freemsg(bp);
1523 				break;
1524 			}
1525 			mutex_enter(&stp->sd_lock);
1526 			putback(stp, q, bp, pri);
1527 			mutex_exit(&stp->sd_lock);
1528 			if (rflg == 0)
1529 				error = EBADMSG;
1530 			goto oops1;
1531 
1532 		default:
1533 			/*
1534 			 * Garbage on stream head read queue.
1535 			 */
1536 			cmn_err(CE_WARN, "bad %x found at stream head\n",
1537 			    bp->b_datap->db_type);
1538 			freemsg(bp);
1539 			goto oops1;
1540 		}
1541 		mutex_enter(&stp->sd_lock);
1542 	}
1543 oops:
1544 	mutex_exit(&stp->sd_lock);
1545 oops1:
1546 	qbackenable(q, pri);
1547 	return (error);
1548 #undef	_LASTMARK
1549 }
1550 
1551 /*
1552  * Default processing of M_PROTO/M_PCPROTO messages.
1553  * Determine which wakeups and signals are needed.
1554  * This can be replaced by a user-specified procedure for kernel users
1555  * of STREAMS.
1556  */
1557 /* ARGSUSED */
1558 mblk_t *
1559 strrput_proto(vnode_t *vp, mblk_t *mp,
1560     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1561     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1562 {
1563 	*wakeups = RSLEEP;
1564 	*allmsgsigs = 0;
1565 
1566 	switch (mp->b_datap->db_type) {
1567 	case M_PROTO:
1568 		if (mp->b_band == 0) {
1569 			*firstmsgsigs = S_INPUT | S_RDNORM;
1570 			*pollwakeups = POLLIN | POLLRDNORM;
1571 		} else {
1572 			*firstmsgsigs = S_INPUT | S_RDBAND;
1573 			*pollwakeups = POLLIN | POLLRDBAND;
1574 		}
1575 		break;
1576 	case M_PCPROTO:
1577 		*firstmsgsigs = S_HIPRI;
1578 		*pollwakeups = POLLPRI;
1579 		break;
1580 	}
1581 	return (mp);
1582 }
1583 
1584 /*
1585  * Default processing of everything but M_DATA, M_PROTO, M_PCPROTO and
1586  * M_PASSFP messages.
1587  * Determine which wakeups and signals are needed.
1588  * This can be replaced by a user-specified procedure for kernel users
1589  * of STREAMS.
1590  */
1591 /* ARGSUSED */
1592 mblk_t *
1593 strrput_misc(vnode_t *vp, mblk_t *mp,
1594     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1595     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1596 {
1597 	*wakeups = 0;
1598 	*firstmsgsigs = 0;
1599 	*allmsgsigs = 0;
1600 	*pollwakeups = 0;
1601 	return (mp);
1602 }
1603 
1604 /*
1605  * Stream read put procedure.  Called from downstream driver/module
1606  * with messages for the stream head.  Data, protocol, and in-stream
1607  * signal messages are placed on the queue, others are handled directly.
1608  */
1609 int
1610 strrput(queue_t *q, mblk_t *bp)
1611 {
1612 	struct stdata	*stp;
1613 	ulong_t		rput_opt;
1614 	strwakeup_t	wakeups;
1615 	strsigset_t	firstmsgsigs;	/* Signals if first message on queue */
1616 	strsigset_t	allmsgsigs;	/* Signals for all messages */
1617 	strsigset_t	signals;	/* Signals events to generate */
1618 	strpollset_t	pollwakeups;
1619 	mblk_t		*nextbp;
1620 	uchar_t		band = 0;
1621 	int		hipri_sig;
1622 
1623 	stp = (struct stdata *)q->q_ptr;
1624 	/*
1625 	 * Use rput_opt for optimized access to the SR_ flags except
1626 	 * SR_POLLIN. That flag has to be checked under sd_lock since it
1627 	 * is modified by strpoll().
1628 	 */
1629 	rput_opt = stp->sd_rput_opt;
1630 
1631 	ASSERT(qclaimed(q));
1632 	TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_ENTER,
1633 	    "strrput called with message type:q %p bp %p", q, bp);
1634 
1635 	/*
1636 	 * Perform initial processing and pass to the parameterized functions.
1637 	 */
1638 	ASSERT(bp->b_next == NULL);
1639 
1640 	switch (bp->b_datap->db_type) {
1641 	case M_DATA:
1642 		/*
1643 		 * sockfs is the only consumer of STREOF and when it is set,
1644 		 * it implies that the receiver is not interested in receiving
1645 		 * any more data, hence the mblk is freed to prevent unnecessary
1646 		 * message queueing at the stream head.
1647 		 */
1648 		if (stp->sd_flag == STREOF) {
1649 			freemsg(bp);
1650 			return (0);
1651 		}
1652 		if ((rput_opt & SR_IGN_ZEROLEN) &&
1653 		    bp->b_rptr == bp->b_wptr && msgnodata(bp)) {
1654 			/*
1655 			 * Ignore zero-length M_DATA messages. These might be
1656 			 * generated by some transports.
1657 			 * The zero-length M_DATA messages, even if they
1658 			 * are ignored, should effect the atmark tracking and
1659 			 * should wake up a thread sleeping in strwaitmark.
1660 			 */
1661 			mutex_enter(&stp->sd_lock);
1662 			if (bp->b_flag & MSGMARKNEXT) {
1663 				/*
1664 				 * Record the position of the mark either
1665 				 * in q_last or in STRATMARK.
1666 				 */
1667 				if (q->q_last != NULL) {
1668 					q->q_last->b_flag &= ~MSGNOTMARKNEXT;
1669 					q->q_last->b_flag |= MSGMARKNEXT;
1670 				} else {
1671 					stp->sd_flag &= ~STRNOTATMARK;
1672 					stp->sd_flag |= STRATMARK;
1673 				}
1674 			} else if (bp->b_flag & MSGNOTMARKNEXT) {
1675 				/*
1676 				 * Record that this is not the position of
1677 				 * the mark either in q_last or in
1678 				 * STRNOTATMARK.
1679 				 */
1680 				if (q->q_last != NULL) {
1681 					q->q_last->b_flag &= ~MSGMARKNEXT;
1682 					q->q_last->b_flag |= MSGNOTMARKNEXT;
1683 				} else {
1684 					stp->sd_flag &= ~STRATMARK;
1685 					stp->sd_flag |= STRNOTATMARK;
1686 				}
1687 			}
1688 			if (stp->sd_flag & RSLEEP) {
1689 				stp->sd_flag &= ~RSLEEP;
1690 				cv_broadcast(&q->q_wait);
1691 			}
1692 			mutex_exit(&stp->sd_lock);
1693 			freemsg(bp);
1694 			return (0);
1695 		}
1696 		wakeups = RSLEEP;
1697 		if (bp->b_band == 0) {
1698 			firstmsgsigs = S_INPUT | S_RDNORM;
1699 			pollwakeups = POLLIN | POLLRDNORM;
1700 		} else {
1701 			firstmsgsigs = S_INPUT | S_RDBAND;
1702 			pollwakeups = POLLIN | POLLRDBAND;
1703 		}
1704 		if (rput_opt & SR_SIGALLDATA)
1705 			allmsgsigs = firstmsgsigs;
1706 		else
1707 			allmsgsigs = 0;
1708 
1709 		mutex_enter(&stp->sd_lock);
1710 		if ((rput_opt & SR_CONSOL_DATA) &&
1711 		    (q->q_last != NULL) &&
1712 		    (bp->b_flag & (MSGMARK|MSGDELIM)) == 0) {
1713 			/*
1714 			 * Consolidate an M_DATA message onto an M_DATA,
1715 			 * M_PROTO, or M_PCPROTO by merging it with q_last.
1716 			 * The consolidation does not take place if
1717 			 * the old message is marked with either of the
1718 			 * marks or the delim flag or if the new
1719 			 * message is marked with MSGMARK. The MSGMARK
1720 			 * check is needed to handle the odd semantics of
1721 			 * MSGMARK where essentially the whole message
1722 			 * is to be treated as marked.
1723 			 * Carry any MSGMARKNEXT  and MSGNOTMARKNEXT from the
1724 			 * new message to the front of the b_cont chain.
1725 			 */
1726 			mblk_t *lbp = q->q_last;
1727 			unsigned char db_type = lbp->b_datap->db_type;
1728 
1729 			if ((db_type == M_DATA || db_type == M_PROTO ||
1730 			    db_type == M_PCPROTO) &&
1731 			    !(lbp->b_flag & (MSGDELIM|MSGMARK|MSGMARKNEXT))) {
1732 				rmvq_noenab(q, lbp);
1733 				/*
1734 				 * The first message in the b_cont list
1735 				 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
1736 				 * We need to handle the case where we
1737 				 * are appending:
1738 				 *
1739 				 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
1740 				 * 2) a MSGMARKNEXT to a plain message.
1741 				 * 3) a MSGNOTMARKNEXT to a plain message
1742 				 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
1743 				 *    message.
1744 				 *
1745 				 * Thus we never append a MSGMARKNEXT or
1746 				 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
1747 				 */
1748 				if (bp->b_flag & MSGMARKNEXT) {
1749 					lbp->b_flag |= MSGMARKNEXT;
1750 					lbp->b_flag &= ~MSGNOTMARKNEXT;
1751 					bp->b_flag &= ~MSGMARKNEXT;
1752 				} else if (bp->b_flag & MSGNOTMARKNEXT) {
1753 					lbp->b_flag |= MSGNOTMARKNEXT;
1754 					bp->b_flag &= ~MSGNOTMARKNEXT;
1755 				}
1756 
1757 				linkb(lbp, bp);
1758 				bp = lbp;
1759 				/*
1760 				 * The new message logically isn't the first
1761 				 * even though the q_first check below thinks
1762 				 * it is. Clear the firstmsgsigs to make it
1763 				 * not appear to be first.
1764 				 */
1765 				firstmsgsigs = 0;
1766 			}
1767 		}
1768 		break;
1769 
1770 	case M_PASSFP:
1771 		wakeups = RSLEEP;
1772 		allmsgsigs = 0;
1773 		if (bp->b_band == 0) {
1774 			firstmsgsigs = S_INPUT | S_RDNORM;
1775 			pollwakeups = POLLIN | POLLRDNORM;
1776 		} else {
1777 			firstmsgsigs = S_INPUT | S_RDBAND;
1778 			pollwakeups = POLLIN | POLLRDBAND;
1779 		}
1780 		mutex_enter(&stp->sd_lock);
1781 		break;
1782 
1783 	case M_PROTO:
1784 	case M_PCPROTO:
1785 		ASSERT(stp->sd_rprotofunc != NULL);
1786 		bp = (stp->sd_rprotofunc)(stp->sd_vnode, bp,
1787 		    &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1788 #define	ALLSIG	(S_INPUT|S_HIPRI|S_OUTPUT|S_MSG|S_ERROR|S_HANGUP|S_RDNORM|\
1789 		S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)
1790 #define	ALLPOLL	(POLLIN|POLLPRI|POLLOUT|POLLRDNORM|POLLWRNORM|POLLRDBAND|\
1791 		POLLWRBAND)
1792 
1793 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1794 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1795 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1796 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1797 
1798 		mutex_enter(&stp->sd_lock);
1799 		break;
1800 
1801 	default:
1802 		ASSERT(stp->sd_rmiscfunc != NULL);
1803 		bp = (stp->sd_rmiscfunc)(stp->sd_vnode, bp,
1804 		    &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1805 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1806 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1807 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1808 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1809 #undef	ALLSIG
1810 #undef	ALLPOLL
1811 		mutex_enter(&stp->sd_lock);
1812 		break;
1813 	}
1814 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1815 
1816 	/* By default generate superset of signals */
1817 	signals = (firstmsgsigs | allmsgsigs);
1818 
1819 	/*
1820 	 * The  proto and misc functions can return multiple messages
1821 	 * as a b_next chain. Such messages are processed separately.
1822 	 */
1823 one_more:
1824 	hipri_sig = 0;
1825 	if (bp == NULL) {
1826 		nextbp = NULL;
1827 	} else {
1828 		nextbp = bp->b_next;
1829 		bp->b_next = NULL;
1830 
1831 		switch (bp->b_datap->db_type) {
1832 		case M_PCPROTO:
1833 			/*
1834 			 * Only one priority protocol message is allowed at the
1835 			 * stream head at a time.
1836 			 */
1837 			if (stp->sd_flag & STRPRI) {
1838 				TRACE_0(TR_FAC_STREAMS_FR, TR_STRRPUT_PROTERR,
1839 				    "M_PCPROTO already at head");
1840 				freemsg(bp);
1841 				mutex_exit(&stp->sd_lock);
1842 				goto done;
1843 			}
1844 			stp->sd_flag |= STRPRI;
1845 			hipri_sig = 1;
1846 			/* FALLTHRU */
1847 		case M_DATA:
1848 		case M_PROTO:
1849 		case M_PASSFP:
1850 			band = bp->b_band;
1851 			/*
1852 			 * Marking doesn't work well when messages
1853 			 * are marked in more than one band.  We only
1854 			 * remember the last message received, even if
1855 			 * it is placed on the queue ahead of other
1856 			 * marked messages.
1857 			 */
1858 			if (bp->b_flag & MSGMARK)
1859 				stp->sd_mark = bp;
1860 			(void) putq(q, bp);
1861 
1862 			/*
1863 			 * If message is a PCPROTO message, always use
1864 			 * firstmsgsigs to determine if a signal should be
1865 			 * sent as strrput is the only place to send
1866 			 * signals for PCPROTO. Other messages are based on
1867 			 * the STRGETINPROG flag. The flag determines if
1868 			 * strrput or (k)strgetmsg will be responsible for
1869 			 * sending the signals, in the firstmsgsigs case.
1870 			 */
1871 			if ((hipri_sig == 1) ||
1872 			    (((stp->sd_flag & STRGETINPROG) == 0) &&
1873 			    (q->q_first == bp)))
1874 				signals = (firstmsgsigs | allmsgsigs);
1875 			else
1876 				signals = allmsgsigs;
1877 			break;
1878 
1879 		default:
1880 			mutex_exit(&stp->sd_lock);
1881 			(void) strrput_nondata(q, bp);
1882 			mutex_enter(&stp->sd_lock);
1883 			break;
1884 		}
1885 	}
1886 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1887 	/*
1888 	 * Wake sleeping read/getmsg and cancel deferred wakeup
1889 	 */
1890 	if (wakeups & RSLEEP)
1891 		stp->sd_wakeq &= ~RSLEEP;
1892 
1893 	wakeups &= stp->sd_flag;
1894 	if (wakeups & RSLEEP) {
1895 		stp->sd_flag &= ~RSLEEP;
1896 		cv_broadcast(&q->q_wait);
1897 	}
1898 	if (wakeups & WSLEEP) {
1899 		stp->sd_flag &= ~WSLEEP;
1900 		cv_broadcast(&_WR(q)->q_wait);
1901 	}
1902 
1903 	if (pollwakeups != 0) {
1904 		if (pollwakeups == (POLLIN | POLLRDNORM)) {
1905 			/*
1906 			 * Can't use rput_opt since it was not
1907 			 * read when sd_lock was held and SR_POLLIN is changed
1908 			 * by strpoll() under sd_lock.
1909 			 */
1910 			if (!(stp->sd_rput_opt & SR_POLLIN))
1911 				goto no_pollwake;
1912 			stp->sd_rput_opt &= ~SR_POLLIN;
1913 		}
1914 		mutex_exit(&stp->sd_lock);
1915 		pollwakeup(&stp->sd_pollist, pollwakeups);
1916 		mutex_enter(&stp->sd_lock);
1917 	}
1918 no_pollwake:
1919 
1920 	/*
1921 	 * strsendsig can handle multiple signals with a
1922 	 * single call.
1923 	 */
1924 	if (stp->sd_sigflags & signals)
1925 		strsendsig(stp->sd_siglist, signals, band, 0);
1926 	mutex_exit(&stp->sd_lock);
1927 
1928 
1929 done:
1930 	if (nextbp == NULL)
1931 		return (0);
1932 
1933 	/*
1934 	 * Any signals were handled the first time.
1935 	 * Wakeups and pollwakeups are redone to avoid any race
1936 	 * conditions - all the messages are not queued until the
1937 	 * last message has been processed by strrput.
1938 	 */
1939 	bp = nextbp;
1940 	signals = firstmsgsigs = allmsgsigs = 0;
1941 	mutex_enter(&stp->sd_lock);
1942 	goto one_more;
1943 }
1944 
1945 static void
1946 log_dupioc(queue_t *rq, mblk_t *bp)
1947 {
1948 	queue_t *wq, *qp;
1949 	char *modnames, *mnp, *dname;
1950 	size_t maxmodstr;
1951 	boolean_t islast;
1952 
1953 	/*
1954 	 * Allocate a buffer large enough to hold the names of nstrpush modules
1955 	 * and one driver, with spaces between and NUL terminator.  If we can't
1956 	 * get memory, then we'll just log the driver name.
1957 	 */
1958 	maxmodstr = nstrpush * (FMNAMESZ + 1);
1959 	mnp = modnames = kmem_alloc(maxmodstr, KM_NOSLEEP);
1960 
1961 	/* march down write side to print log message down to the driver */
1962 	wq = WR(rq);
1963 
1964 	/* make sure q_next doesn't shift around while we're grabbing data */
1965 	claimstr(wq);
1966 	qp = wq->q_next;
1967 	do {
1968 		if ((dname = qp->q_qinfo->qi_minfo->mi_idname) == NULL)
1969 			dname = "?";
1970 		islast = !SAMESTR(qp) || qp->q_next == NULL;
1971 		if (modnames == NULL) {
1972 			/*
1973 			 * If we don't have memory, then get the driver name in
1974 			 * the log where we can see it.  Note that memory
1975 			 * pressure is a possible cause of these sorts of bugs.
1976 			 */
1977 			if (islast) {
1978 				modnames = dname;
1979 				maxmodstr = 0;
1980 			}
1981 		} else {
1982 			mnp += snprintf(mnp, FMNAMESZ + 1, "%s", dname);
1983 			if (!islast)
1984 				*mnp++ = ' ';
1985 		}
1986 		qp = qp->q_next;
1987 	} while (!islast);
1988 	releasestr(wq);
1989 	/* Cannot happen unless stream head is corrupt. */
1990 	ASSERT(modnames != NULL);
1991 	(void) strlog(rq->q_qinfo->qi_minfo->mi_idnum, 0, 1,
1992 	    SL_CONSOLE|SL_TRACE|SL_ERROR,
1993 	    "Warning: stream %p received duplicate %X M_IOC%s; module list: %s",
1994 	    rq->q_ptr, ((struct iocblk *)bp->b_rptr)->ioc_cmd,
1995 	    (DB_TYPE(bp) == M_IOCACK ? "ACK" : "NAK"), modnames);
1996 	if (maxmodstr != 0)
1997 		kmem_free(modnames, maxmodstr);
1998 }
1999 
2000 int
2001 strrput_nondata(queue_t *q, mblk_t *bp)
2002 {
2003 	struct stdata *stp;
2004 	struct iocblk *iocbp;
2005 	struct stroptions *sop;
2006 	struct copyreq *reqp;
2007 	struct copyresp *resp;
2008 	unsigned char bpri;
2009 	unsigned char  flushed_already = 0;
2010 
2011 	stp = (struct stdata *)q->q_ptr;
2012 
2013 	ASSERT(!(stp->sd_flag & STPLEX));
2014 	ASSERT(qclaimed(q));
2015 
2016 	switch (bp->b_datap->db_type) {
2017 	case M_ERROR:
2018 		/*
2019 		 * An error has occurred downstream, the errno is in the first
2020 		 * bytes of the message.
2021 		 */
2022 		if ((bp->b_wptr - bp->b_rptr) == 2) {	/* New flavor */
2023 			unsigned char rw = 0;
2024 
2025 			mutex_enter(&stp->sd_lock);
2026 			if (*bp->b_rptr != NOERROR) {	/* read error */
2027 				if (*bp->b_rptr != 0) {
2028 					if (stp->sd_flag & STRDERR)
2029 						flushed_already |= FLUSHR;
2030 					stp->sd_flag |= STRDERR;
2031 					rw |= FLUSHR;
2032 				} else {
2033 					stp->sd_flag &= ~STRDERR;
2034 				}
2035 				stp->sd_rerror = *bp->b_rptr;
2036 			}
2037 			bp->b_rptr++;
2038 			if (*bp->b_rptr != NOERROR) {	/* write error */
2039 				if (*bp->b_rptr != 0) {
2040 					if (stp->sd_flag & STWRERR)
2041 						flushed_already |= FLUSHW;
2042 					stp->sd_flag |= STWRERR;
2043 					rw |= FLUSHW;
2044 				} else {
2045 					stp->sd_flag &= ~STWRERR;
2046 				}
2047 				stp->sd_werror = *bp->b_rptr;
2048 			}
2049 			if (rw) {
2050 				TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_WAKE,
2051 				    "strrput cv_broadcast:q %p, bp %p",
2052 				    q, bp);
2053 				cv_broadcast(&q->q_wait); /* readers */
2054 				cv_broadcast(&_WR(q)->q_wait); /* writers */
2055 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
2056 
2057 				mutex_exit(&stp->sd_lock);
2058 				pollwakeup(&stp->sd_pollist, POLLERR);
2059 				mutex_enter(&stp->sd_lock);
2060 
2061 				if (stp->sd_sigflags & S_ERROR)
2062 					strsendsig(stp->sd_siglist, S_ERROR, 0,
2063 					    ((rw & FLUSHR) ? stp->sd_rerror :
2064 					    stp->sd_werror));
2065 				mutex_exit(&stp->sd_lock);
2066 				/*
2067 				 * Send the M_FLUSH only
2068 				 * for the first M_ERROR
2069 				 * message on the stream
2070 				 */
2071 				if (flushed_already == rw) {
2072 					freemsg(bp);
2073 					return (0);
2074 				}
2075 
2076 				bp->b_datap->db_type = M_FLUSH;
2077 				*bp->b_rptr = rw;
2078 				bp->b_wptr = bp->b_rptr + 1;
2079 				/*
2080 				 * Protect against the driver
2081 				 * passing up messages after
2082 				 * it has done a qprocsoff
2083 				 */
2084 				if (_OTHERQ(q)->q_next == NULL)
2085 					freemsg(bp);
2086 				else
2087 					qreply(q, bp);
2088 				return (0);
2089 			} else
2090 				mutex_exit(&stp->sd_lock);
2091 		} else if (*bp->b_rptr != 0) {		/* Old flavor */
2092 				if (stp->sd_flag & (STRDERR|STWRERR))
2093 					flushed_already = FLUSHRW;
2094 				mutex_enter(&stp->sd_lock);
2095 				stp->sd_flag |= (STRDERR|STWRERR);
2096 				stp->sd_rerror = *bp->b_rptr;
2097 				stp->sd_werror = *bp->b_rptr;
2098 				TRACE_2(TR_FAC_STREAMS_FR,
2099 				    TR_STRRPUT_WAKE2,
2100 				    "strrput wakeup #2:q %p, bp %p", q, bp);
2101 				cv_broadcast(&q->q_wait); /* the readers */
2102 				cv_broadcast(&_WR(q)->q_wait); /* the writers */
2103 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
2104 
2105 				mutex_exit(&stp->sd_lock);
2106 				pollwakeup(&stp->sd_pollist, POLLERR);
2107 				mutex_enter(&stp->sd_lock);
2108 
2109 				if (stp->sd_sigflags & S_ERROR)
2110 					strsendsig(stp->sd_siglist, S_ERROR, 0,
2111 					    (stp->sd_werror ? stp->sd_werror :
2112 					    stp->sd_rerror));
2113 				mutex_exit(&stp->sd_lock);
2114 
2115 				/*
2116 				 * Send the M_FLUSH only
2117 				 * for the first M_ERROR
2118 				 * message on the stream
2119 				 */
2120 				if (flushed_already != FLUSHRW) {
2121 					bp->b_datap->db_type = M_FLUSH;
2122 					*bp->b_rptr = FLUSHRW;
2123 					/*
2124 					 * Protect against the driver passing up
2125 					 * messages after it has done a
2126 					 * qprocsoff.
2127 					 */
2128 				if (_OTHERQ(q)->q_next == NULL)
2129 					freemsg(bp);
2130 				else
2131 					qreply(q, bp);
2132 				return (0);
2133 				}
2134 		}
2135 		freemsg(bp);
2136 		return (0);
2137 
2138 	case M_HANGUP:
2139 
2140 		freemsg(bp);
2141 		mutex_enter(&stp->sd_lock);
2142 		stp->sd_werror = ENXIO;
2143 		stp->sd_flag |= STRHUP;
2144 		stp->sd_flag &= ~(WSLEEP|RSLEEP);
2145 
2146 		/*
2147 		 * send signal if controlling tty
2148 		 */
2149 
2150 		if (stp->sd_sidp) {
2151 			prsignal(stp->sd_sidp, SIGHUP);
2152 			if (stp->sd_sidp != stp->sd_pgidp)
2153 				pgsignal(stp->sd_pgidp, SIGTSTP);
2154 		}
2155 
2156 		/*
2157 		 * wake up read, write, and exception pollers and
2158 		 * reset wakeup mechanism.
2159 		 */
2160 		cv_broadcast(&q->q_wait);	/* the readers */
2161 		cv_broadcast(&_WR(q)->q_wait);	/* the writers */
2162 		cv_broadcast(&stp->sd_monitor);	/* the ioctllers */
2163 		strhup(stp);
2164 		mutex_exit(&stp->sd_lock);
2165 		return (0);
2166 
2167 	case M_UNHANGUP:
2168 		freemsg(bp);
2169 		mutex_enter(&stp->sd_lock);
2170 		stp->sd_werror = 0;
2171 		stp->sd_flag &= ~STRHUP;
2172 		mutex_exit(&stp->sd_lock);
2173 		return (0);
2174 
2175 	case M_SIG:
2176 		/*
2177 		 * Someone downstream wants to post a signal.  The
2178 		 * signal to post is contained in the first byte of the
2179 		 * message.  If the message would go on the front of
2180 		 * the queue, send a signal to the process group
2181 		 * (if not SIGPOLL) or to the siglist processes
2182 		 * (SIGPOLL).  If something is already on the queue,
2183 		 * OR if we are delivering a delayed suspend (*sigh*
2184 		 * another "tty" hack) and there's no one sleeping already,
2185 		 * just enqueue the message.
2186 		 */
2187 		mutex_enter(&stp->sd_lock);
2188 		if (q->q_first || (*bp->b_rptr == SIGTSTP &&
2189 		    !(stp->sd_flag & RSLEEP))) {
2190 			(void) putq(q, bp);
2191 			mutex_exit(&stp->sd_lock);
2192 			return (0);
2193 		}
2194 		mutex_exit(&stp->sd_lock);
2195 		/* FALLTHRU */
2196 
2197 	case M_PCSIG:
2198 		/*
2199 		 * Don't enqueue, just post the signal.
2200 		 */
2201 		strsignal(stp, *bp->b_rptr, 0L);
2202 		freemsg(bp);
2203 		return (0);
2204 
2205 	case M_CMD:
2206 		if (MBLKL(bp) != sizeof (cmdblk_t)) {
2207 			freemsg(bp);
2208 			return (0);
2209 		}
2210 
2211 		mutex_enter(&stp->sd_lock);
2212 		if (stp->sd_flag & STRCMDWAIT) {
2213 			ASSERT(stp->sd_cmdblk == NULL);
2214 			stp->sd_cmdblk = bp;
2215 			cv_broadcast(&stp->sd_monitor);
2216 			mutex_exit(&stp->sd_lock);
2217 		} else {
2218 			mutex_exit(&stp->sd_lock);
2219 			freemsg(bp);
2220 		}
2221 		return (0);
2222 
2223 	case M_FLUSH:
2224 		/*
2225 		 * Flush queues.  The indication of which queues to flush
2226 		 * is in the first byte of the message.  If the read queue
2227 		 * is specified, then flush it.  If FLUSHBAND is set, just
2228 		 * flush the band specified by the second byte of the message.
2229 		 *
2230 		 * If a module has issued a M_SETOPT to not flush hi
2231 		 * priority messages off of the stream head, then pass this
2232 		 * flag into the flushq code to preserve such messages.
2233 		 */
2234 
2235 		if (*bp->b_rptr & FLUSHR) {
2236 			mutex_enter(&stp->sd_lock);
2237 			if (*bp->b_rptr & FLUSHBAND) {
2238 				ASSERT((bp->b_wptr - bp->b_rptr) >= 2);
2239 				flushband(q, *(bp->b_rptr + 1), FLUSHALL);
2240 			} else
2241 				flushq_common(q, FLUSHALL,
2242 				    stp->sd_read_opt & RFLUSHPCPROT);
2243 			if ((q->q_first == NULL) ||
2244 			    (q->q_first->b_datap->db_type < QPCTL))
2245 				stp->sd_flag &= ~STRPRI;
2246 			else {
2247 				ASSERT(stp->sd_flag & STRPRI);
2248 			}
2249 			mutex_exit(&stp->sd_lock);
2250 		}
2251 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
2252 			*bp->b_rptr &= ~FLUSHR;
2253 			bp->b_flag |= MSGNOLOOP;
2254 			/*
2255 			 * Protect against the driver passing up
2256 			 * messages after it has done a qprocsoff.
2257 			 */
2258 			if (_OTHERQ(q)->q_next == NULL)
2259 				freemsg(bp);
2260 			else
2261 				qreply(q, bp);
2262 			return (0);
2263 		}
2264 		freemsg(bp);
2265 		return (0);
2266 
2267 	case M_IOCACK:
2268 	case M_IOCNAK:
2269 		iocbp = (struct iocblk *)bp->b_rptr;
2270 		/*
2271 		 * If not waiting for ACK or NAK then just free msg.
2272 		 * If incorrect id sequence number then just free msg.
2273 		 * If already have ACK or NAK for user then this is a
2274 		 *    duplicate, display a warning and free the msg.
2275 		 */
2276 		mutex_enter(&stp->sd_lock);
2277 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2278 		    (stp->sd_iocid != iocbp->ioc_id)) {
2279 			/*
2280 			 * If the ACK/NAK is a dup, display a message
2281 			 * Dup is when sd_iocid == ioc_id, and
2282 			 * sd_iocblk == <valid ptr> or -1 (the former
2283 			 * is when an ioctl has been put on the stream
2284 			 * head, but has not yet been consumed, the
2285 			 * later is when it has been consumed).
2286 			 */
2287 			if ((stp->sd_iocid == iocbp->ioc_id) &&
2288 			    (stp->sd_iocblk != NULL)) {
2289 				log_dupioc(q, bp);
2290 			}
2291 			freemsg(bp);
2292 			mutex_exit(&stp->sd_lock);
2293 			return (0);
2294 		}
2295 
2296 		/*
2297 		 * Assign ACK or NAK to user and wake up.
2298 		 */
2299 		stp->sd_iocblk = bp;
2300 		cv_broadcast(&stp->sd_monitor);
2301 		mutex_exit(&stp->sd_lock);
2302 		return (0);
2303 
2304 	case M_COPYIN:
2305 	case M_COPYOUT:
2306 		reqp = (struct copyreq *)bp->b_rptr;
2307 
2308 		/*
2309 		 * If not waiting for ACK or NAK then just fail request.
2310 		 * If already have ACK, NAK, or copy request, then just
2311 		 * fail request.
2312 		 * If incorrect id sequence number then just fail request.
2313 		 */
2314 		mutex_enter(&stp->sd_lock);
2315 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2316 		    (stp->sd_iocid != reqp->cq_id)) {
2317 			if (bp->b_cont) {
2318 				freemsg(bp->b_cont);
2319 				bp->b_cont = NULL;
2320 			}
2321 			bp->b_datap->db_type = M_IOCDATA;
2322 			bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
2323 			resp = (struct copyresp *)bp->b_rptr;
2324 			resp->cp_rval = (caddr_t)1;	/* failure */
2325 			mutex_exit(&stp->sd_lock);
2326 			putnext(stp->sd_wrq, bp);
2327 			return (0);
2328 		}
2329 
2330 		/*
2331 		 * Assign copy request to user and wake up.
2332 		 */
2333 		stp->sd_iocblk = bp;
2334 		cv_broadcast(&stp->sd_monitor);
2335 		mutex_exit(&stp->sd_lock);
2336 		return (0);
2337 
2338 	case M_SETOPTS:
2339 		/*
2340 		 * Set stream head options (read option, write offset,
2341 		 * min/max packet size, and/or high/low water marks for
2342 		 * the read side only).
2343 		 */
2344 
2345 		bpri = 0;
2346 		sop = (struct stroptions *)bp->b_rptr;
2347 		mutex_enter(&stp->sd_lock);
2348 		if (sop->so_flags & SO_READOPT) {
2349 			switch (sop->so_readopt & RMODEMASK) {
2350 			case RNORM:
2351 				stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
2352 				break;
2353 
2354 			case RMSGD:
2355 				stp->sd_read_opt =
2356 				    ((stp->sd_read_opt & ~RD_MSGNODIS) |
2357 				    RD_MSGDIS);
2358 				break;
2359 
2360 			case RMSGN:
2361 				stp->sd_read_opt =
2362 				    ((stp->sd_read_opt & ~RD_MSGDIS) |
2363 				    RD_MSGNODIS);
2364 				break;
2365 			}
2366 			switch (sop->so_readopt & RPROTMASK) {
2367 			case RPROTNORM:
2368 				stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
2369 				break;
2370 
2371 			case RPROTDAT:
2372 				stp->sd_read_opt =
2373 				    ((stp->sd_read_opt & ~RD_PROTDIS) |
2374 				    RD_PROTDAT);
2375 				break;
2376 
2377 			case RPROTDIS:
2378 				stp->sd_read_opt =
2379 				    ((stp->sd_read_opt & ~RD_PROTDAT) |
2380 				    RD_PROTDIS);
2381 				break;
2382 			}
2383 			switch (sop->so_readopt & RFLUSHMASK) {
2384 			case RFLUSHPCPROT:
2385 				/*
2386 				 * This sets the stream head to NOT flush
2387 				 * M_PCPROTO messages.
2388 				 */
2389 				stp->sd_read_opt |= RFLUSHPCPROT;
2390 				break;
2391 			}
2392 		}
2393 		if (sop->so_flags & SO_ERROPT) {
2394 			switch (sop->so_erropt & RERRMASK) {
2395 			case RERRNORM:
2396 				stp->sd_flag &= ~STRDERRNONPERSIST;
2397 				break;
2398 			case RERRNONPERSIST:
2399 				stp->sd_flag |= STRDERRNONPERSIST;
2400 				break;
2401 			}
2402 			switch (sop->so_erropt & WERRMASK) {
2403 			case WERRNORM:
2404 				stp->sd_flag &= ~STWRERRNONPERSIST;
2405 				break;
2406 			case WERRNONPERSIST:
2407 				stp->sd_flag |= STWRERRNONPERSIST;
2408 				break;
2409 			}
2410 		}
2411 		if (sop->so_flags & SO_COPYOPT) {
2412 			if (sop->so_copyopt & ZCVMSAFE) {
2413 				stp->sd_copyflag |= STZCVMSAFE;
2414 				stp->sd_copyflag &= ~STZCVMUNSAFE;
2415 			} else if (sop->so_copyopt & ZCVMUNSAFE) {
2416 				stp->sd_copyflag |= STZCVMUNSAFE;
2417 				stp->sd_copyflag &= ~STZCVMSAFE;
2418 			}
2419 
2420 			if (sop->so_copyopt & COPYCACHED) {
2421 				stp->sd_copyflag |= STRCOPYCACHED;
2422 			}
2423 		}
2424 		if (sop->so_flags & SO_WROFF)
2425 			stp->sd_wroff = sop->so_wroff;
2426 		if (sop->so_flags & SO_TAIL)
2427 			stp->sd_tail = sop->so_tail;
2428 		if (sop->so_flags & SO_MINPSZ)
2429 			q->q_minpsz = sop->so_minpsz;
2430 		if (sop->so_flags & SO_MAXPSZ)
2431 			q->q_maxpsz = sop->so_maxpsz;
2432 		if (sop->so_flags & SO_MAXBLK)
2433 			stp->sd_maxblk = sop->so_maxblk;
2434 		if (sop->so_flags & SO_HIWAT) {
2435 			if (sop->so_flags & SO_BAND) {
2436 				if (strqset(q, QHIWAT,
2437 				    sop->so_band, sop->so_hiwat)) {
2438 					cmn_err(CE_WARN, "strrput: could not "
2439 					    "allocate qband\n");
2440 				} else {
2441 					bpri = sop->so_band;
2442 				}
2443 			} else {
2444 				q->q_hiwat = sop->so_hiwat;
2445 			}
2446 		}
2447 		if (sop->so_flags & SO_LOWAT) {
2448 			if (sop->so_flags & SO_BAND) {
2449 				if (strqset(q, QLOWAT,
2450 				    sop->so_band, sop->so_lowat)) {
2451 					cmn_err(CE_WARN, "strrput: could not "
2452 					    "allocate qband\n");
2453 				} else {
2454 					bpri = sop->so_band;
2455 				}
2456 			} else {
2457 				q->q_lowat = sop->so_lowat;
2458 			}
2459 		}
2460 		if (sop->so_flags & SO_MREADON)
2461 			stp->sd_flag |= SNDMREAD;
2462 		if (sop->so_flags & SO_MREADOFF)
2463 			stp->sd_flag &= ~SNDMREAD;
2464 		if (sop->so_flags & SO_NDELON)
2465 			stp->sd_flag |= OLDNDELAY;
2466 		if (sop->so_flags & SO_NDELOFF)
2467 			stp->sd_flag &= ~OLDNDELAY;
2468 		if (sop->so_flags & SO_ISTTY)
2469 			stp->sd_flag |= STRISTTY;
2470 		if (sop->so_flags & SO_ISNTTY)
2471 			stp->sd_flag &= ~STRISTTY;
2472 		if (sop->so_flags & SO_TOSTOP)
2473 			stp->sd_flag |= STRTOSTOP;
2474 		if (sop->so_flags & SO_TONSTOP)
2475 			stp->sd_flag &= ~STRTOSTOP;
2476 		if (sop->so_flags & SO_DELIM)
2477 			stp->sd_flag |= STRDELIM;
2478 		if (sop->so_flags & SO_NODELIM)
2479 			stp->sd_flag &= ~STRDELIM;
2480 
2481 		mutex_exit(&stp->sd_lock);
2482 		freemsg(bp);
2483 
2484 		/* Check backenable in case the water marks changed */
2485 		qbackenable(q, bpri);
2486 		return (0);
2487 
2488 	/*
2489 	 * The following set of cases deal with situations where two stream
2490 	 * heads are connected to each other (twisted streams).  These messages
2491 	 * have no meaning at the stream head.
2492 	 */
2493 	case M_BREAK:
2494 	case M_CTL:
2495 	case M_DELAY:
2496 	case M_START:
2497 	case M_STOP:
2498 	case M_IOCDATA:
2499 	case M_STARTI:
2500 	case M_STOPI:
2501 		freemsg(bp);
2502 		return (0);
2503 
2504 	case M_IOCTL:
2505 		/*
2506 		 * Always NAK this condition
2507 		 * (makes no sense)
2508 		 * If there is one or more threads in the read side
2509 		 * rwnext we have to defer the nacking until that thread
2510 		 * returns (in strget).
2511 		 */
2512 		mutex_enter(&stp->sd_lock);
2513 		if (stp->sd_struiodnak != 0) {
2514 			/*
2515 			 * Defer NAK to the streamhead. Queue at the end
2516 			 * the list.
2517 			 */
2518 			mblk_t *mp = stp->sd_struionak;
2519 
2520 			while (mp && mp->b_next)
2521 				mp = mp->b_next;
2522 			if (mp)
2523 				mp->b_next = bp;
2524 			else
2525 				stp->sd_struionak = bp;
2526 			bp->b_next = NULL;
2527 			mutex_exit(&stp->sd_lock);
2528 			return (0);
2529 		}
2530 		mutex_exit(&stp->sd_lock);
2531 
2532 		bp->b_datap->db_type = M_IOCNAK;
2533 		/*
2534 		 * Protect against the driver passing up
2535 		 * messages after it has done a qprocsoff.
2536 		 */
2537 		if (_OTHERQ(q)->q_next == NULL)
2538 			freemsg(bp);
2539 		else
2540 			qreply(q, bp);
2541 		return (0);
2542 
2543 	default:
2544 #ifdef DEBUG
2545 		cmn_err(CE_WARN,
2546 		    "bad message type %x received at stream head\n",
2547 		    bp->b_datap->db_type);
2548 #endif
2549 		freemsg(bp);
2550 		return (0);
2551 	}
2552 
2553 	/* NOTREACHED */
2554 }
2555 
2556 /*
2557  * Check if the stream pointed to by `stp' can be written to, and return an
2558  * error code if not.  If `eiohup' is set, then return EIO if STRHUP is set.
2559  * If `sigpipeok' is set and the SW_SIGPIPE option is enabled on the stream,
2560  * then always return EPIPE and send a SIGPIPE to the invoking thread.
2561  */
2562 static int
2563 strwriteable(struct stdata *stp, boolean_t eiohup, boolean_t sigpipeok)
2564 {
2565 	int error;
2566 
2567 	ASSERT(MUTEX_HELD(&stp->sd_lock));
2568 
2569 	/*
2570 	 * For modem support, POSIX states that on writes, EIO should
2571 	 * be returned if the stream has been hung up.
2572 	 */
2573 	if (eiohup && (stp->sd_flag & (STPLEX|STRHUP)) == STRHUP)
2574 		error = EIO;
2575 	else
2576 		error = strgeterr(stp, STRHUP|STPLEX|STWRERR, 0);
2577 
2578 	if (error != 0) {
2579 		if (!(stp->sd_flag & STPLEX) &&
2580 		    (stp->sd_wput_opt & SW_SIGPIPE) && sigpipeok) {
2581 			tsignal(curthread, SIGPIPE);
2582 			error = EPIPE;
2583 		}
2584 	}
2585 
2586 	return (error);
2587 }
2588 
2589 /*
2590  * Copyin and send data down a stream.
2591  * The caller will allocate and copyin any control part that precedes the
2592  * message and pass than in as mctl.
2593  *
2594  * Caller should *not* hold sd_lock.
2595  * When EWOULDBLOCK is returned the caller has to redo the canputnext
2596  * under sd_lock in order to avoid missing a backenabling wakeup.
2597  *
2598  * Use iosize = -1 to not send any M_DATA. iosize = 0 sends zero-length M_DATA.
2599  *
2600  * Set MSG_IGNFLOW in flags to ignore flow control for hipri messages.
2601  * For sync streams we can only ignore flow control by reverting to using
2602  * putnext.
2603  *
2604  * If sd_maxblk is less than *iosize this routine might return without
2605  * transferring all of *iosize. In all cases, on return *iosize will contain
2606  * the amount of data that was transferred.
2607  */
2608 static int
2609 strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
2610     int b_flag, int pri, int flags)
2611 {
2612 	struiod_t uiod;
2613 	mblk_t *mp;
2614 	queue_t *wqp = stp->sd_wrq;
2615 	int error = 0;
2616 	ssize_t count = *iosize;
2617 	cred_t *cr;
2618 
2619 	ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
2620 
2621 	if (uiop != NULL && count >= 0)
2622 		flags |= stp->sd_struiowrq ? STRUIO_POSTPONE : 0;
2623 
2624 	if (!(flags & STRUIO_POSTPONE)) {
2625 		/*
2626 		 * Use regular canputnext, strmakedata, putnext sequence.
2627 		 */
2628 		if (pri == 0) {
2629 			if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2630 				freemsg(mctl);
2631 				return (EWOULDBLOCK);
2632 			}
2633 		} else {
2634 			if (!(flags & MSG_IGNFLOW) && !bcanputnext(wqp, pri)) {
2635 				freemsg(mctl);
2636 				return (EWOULDBLOCK);
2637 			}
2638 		}
2639 
2640 		if ((error = strmakedata(iosize, uiop, stp, flags,
2641 		    &mp)) != 0) {
2642 			freemsg(mctl);
2643 			/*
2644 			 * need to change return code to ENOMEM
2645 			 * so that this is not confused with
2646 			 * flow control, EAGAIN.
2647 			 */
2648 
2649 			if (error == EAGAIN)
2650 				return (ENOMEM);
2651 			else
2652 				return (error);
2653 		}
2654 		if (mctl != NULL) {
2655 			if (mctl->b_cont == NULL)
2656 				mctl->b_cont = mp;
2657 			else if (mp != NULL)
2658 				linkb(mctl, mp);
2659 			mp = mctl;
2660 			/*
2661 			 * Note that for interrupt thread, the CRED() is
2662 			 * NULL. Don't bother with the pid either.
2663 			 */
2664 			if ((cr = CRED()) != NULL) {
2665 				mblk_setcred(mp, cr);
2666 				DB_CPID(mp) = curproc->p_pid;
2667 			}
2668 		} else if (mp == NULL)
2669 			return (0);
2670 
2671 		mp->b_flag |= b_flag;
2672 		mp->b_band = (uchar_t)pri;
2673 
2674 		if (flags & MSG_IGNFLOW) {
2675 			/*
2676 			 * XXX Hack: Don't get stuck running service
2677 			 * procedures. This is needed for sockfs when
2678 			 * sending the unbind message out of the rput
2679 			 * procedure - we don't want a put procedure
2680 			 * to run service procedures.
2681 			 */
2682 			putnext(wqp, mp);
2683 		} else {
2684 			stream_willservice(stp);
2685 			putnext(wqp, mp);
2686 			stream_runservice(stp);
2687 		}
2688 		return (0);
2689 	}
2690 	/*
2691 	 * Stream supports rwnext() for the write side.
2692 	 */
2693 	if ((error = strmakedata(iosize, uiop, stp, flags, &mp)) != 0) {
2694 		freemsg(mctl);
2695 		/*
2696 		 * map EAGAIN to ENOMEM since EAGAIN means "flow controlled".
2697 		 */
2698 		return (error == EAGAIN ? ENOMEM : error);
2699 	}
2700 	if (mctl != NULL) {
2701 		if (mctl->b_cont == NULL)
2702 			mctl->b_cont = mp;
2703 		else if (mp != NULL)
2704 			linkb(mctl, mp);
2705 		mp = mctl;
2706 		/*
2707 		 * Note that for interrupt thread, the CRED() is
2708 		 * NULL.  Don't bother with the pid either.
2709 		 */
2710 		if ((cr = CRED()) != NULL) {
2711 			mblk_setcred(mp, cr);
2712 			DB_CPID(mp) = curproc->p_pid;
2713 		}
2714 	} else if (mp == NULL) {
2715 		return (0);
2716 	}
2717 
2718 	mp->b_flag |= b_flag;
2719 	mp->b_band = (uchar_t)pri;
2720 
2721 	(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
2722 	    sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
2723 	uiod.d_uio.uio_offset = 0;
2724 	uiod.d_mp = mp;
2725 	error = rwnext(wqp, &uiod);
2726 	if (! uiod.d_mp) {
2727 		uioskip(uiop, *iosize);
2728 		return (error);
2729 	}
2730 	ASSERT(mp == uiod.d_mp);
2731 	if (error == EINVAL) {
2732 		/*
2733 		 * The stream plumbing must have changed while
2734 		 * we were away, so just turn off rwnext()s.
2735 		 */
2736 		error = 0;
2737 	} else if (error == EBUSY || error == EWOULDBLOCK) {
2738 		/*
2739 		 * Couldn't enter a perimeter or took a page fault,
2740 		 * so fall-back to putnext().
2741 		 */
2742 		error = 0;
2743 	} else {
2744 		freemsg(mp);
2745 		return (error);
2746 	}
2747 	/* Have to check canput before consuming data from the uio */
2748 	if (pri == 0) {
2749 		if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2750 			freemsg(mp);
2751 			return (EWOULDBLOCK);
2752 		}
2753 	} else {
2754 		if (!bcanputnext(wqp, pri) && !(flags & MSG_IGNFLOW)) {
2755 			freemsg(mp);
2756 			return (EWOULDBLOCK);
2757 		}
2758 	}
2759 	ASSERT(mp == uiod.d_mp);
2760 	/* Copyin data from the uio */
2761 	if ((error = struioget(wqp, mp, &uiod, 0)) != 0) {
2762 		freemsg(mp);
2763 		return (error);
2764 	}
2765 	uioskip(uiop, *iosize);
2766 	if (flags & MSG_IGNFLOW) {
2767 		/*
2768 		 * XXX Hack: Don't get stuck running service procedures.
2769 		 * This is needed for sockfs when sending the unbind message
2770 		 * out of the rput procedure - we don't want a put procedure
2771 		 * to run service procedures.
2772 		 */
2773 		putnext(wqp, mp);
2774 	} else {
2775 		stream_willservice(stp);
2776 		putnext(wqp, mp);
2777 		stream_runservice(stp);
2778 	}
2779 	return (0);
2780 }
2781 
2782 /*
2783  * Write attempts to break the write request into messages conforming
2784  * with the minimum and maximum packet sizes set downstream.
2785  *
2786  * Write will not block if downstream queue is full and
2787  * O_NDELAY is set, otherwise it will block waiting for the queue to get room.
2788  *
2789  * A write of zero bytes gets packaged into a zero length message and sent
2790  * downstream like any other message.
2791  *
2792  * If buffers of the requested sizes are not available, the write will
2793  * sleep until the buffers become available.
2794  *
2795  * Write (if specified) will supply a write offset in a message if it
2796  * makes sense. This can be specified by downstream modules as part of
2797  * a M_SETOPTS message.  Write will not supply the write offset if it
2798  * cannot supply any data in a buffer.  In other words, write will never
2799  * send down an empty packet due to a write offset.
2800  */
2801 /* ARGSUSED2 */
2802 int
2803 strwrite(struct vnode *vp, struct uio *uiop, cred_t *crp)
2804 {
2805 	return (strwrite_common(vp, uiop, crp, 0));
2806 }
2807 
2808 /* ARGSUSED2 */
2809 int
2810 strwrite_common(struct vnode *vp, struct uio *uiop, cred_t *crp, int wflag)
2811 {
2812 	struct stdata *stp;
2813 	struct queue *wqp;
2814 	ssize_t rmin, rmax;
2815 	ssize_t iosize;
2816 	int waitflag;
2817 	int tempmode;
2818 	int error = 0;
2819 	int b_flag;
2820 
2821 	ASSERT(vp->v_stream);
2822 	stp = vp->v_stream;
2823 
2824 	mutex_enter(&stp->sd_lock);
2825 
2826 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
2827 		mutex_exit(&stp->sd_lock);
2828 		return (error);
2829 	}
2830 
2831 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
2832 		error = strwriteable(stp, B_TRUE, B_TRUE);
2833 		if (error != 0) {
2834 			mutex_exit(&stp->sd_lock);
2835 			return (error);
2836 		}
2837 	}
2838 
2839 	mutex_exit(&stp->sd_lock);
2840 
2841 	wqp = stp->sd_wrq;
2842 
2843 	/* get these values from them cached in the stream head */
2844 	rmin = stp->sd_qn_minpsz;
2845 	rmax = stp->sd_qn_maxpsz;
2846 
2847 	/*
2848 	 * Check the min/max packet size constraints.  If min packet size
2849 	 * is non-zero, the write cannot be split into multiple messages
2850 	 * and still guarantee the size constraints.
2851 	 */
2852 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_IN, "strwrite in:q %p", wqp);
2853 
2854 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
2855 	if (rmax == 0) {
2856 		return (0);
2857 	}
2858 	if (rmin > 0) {
2859 		if (uiop->uio_resid < rmin) {
2860 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2861 			    "strwrite out:q %p out %d error %d",
2862 			    wqp, 0, ERANGE);
2863 			return (ERANGE);
2864 		}
2865 		if ((rmax != INFPSZ) && (uiop->uio_resid > rmax)) {
2866 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2867 			    "strwrite out:q %p out %d error %d",
2868 			    wqp, 1, ERANGE);
2869 			return (ERANGE);
2870 		}
2871 	}
2872 
2873 	/*
2874 	 * Do until count satisfied or error.
2875 	 */
2876 	waitflag = WRITEWAIT | wflag;
2877 	if (stp->sd_flag & OLDNDELAY)
2878 		tempmode = uiop->uio_fmode & ~FNDELAY;
2879 	else
2880 		tempmode = uiop->uio_fmode;
2881 
2882 	if (rmax == INFPSZ)
2883 		rmax = uiop->uio_resid;
2884 
2885 	/*
2886 	 * Note that tempmode does not get used in strput/strmakedata
2887 	 * but only in strwaitq. The other routines use uio_fmode
2888 	 * unmodified.
2889 	 */
2890 
2891 	/* LINTED: constant in conditional context */
2892 	while (1) {	/* breaks when uio_resid reaches zero */
2893 		/*
2894 		 * Determine the size of the next message to be
2895 		 * packaged.  May have to break write into several
2896 		 * messages based on max packet size.
2897 		 */
2898 		iosize = MIN(uiop->uio_resid, rmax);
2899 
2900 		/*
2901 		 * Put block downstream when flow control allows it.
2902 		 */
2903 		if ((stp->sd_flag & STRDELIM) && (uiop->uio_resid == iosize))
2904 			b_flag = MSGDELIM;
2905 		else
2906 			b_flag = 0;
2907 
2908 		for (;;) {
2909 			int done = 0;
2910 
2911 			error = strput(stp, NULL, uiop, &iosize, b_flag, 0, 0);
2912 			if (error == 0)
2913 				break;
2914 			if (error != EWOULDBLOCK)
2915 				goto out;
2916 
2917 			mutex_enter(&stp->sd_lock);
2918 			/*
2919 			 * Check for a missed wakeup.
2920 			 * Needed since strput did not hold sd_lock across
2921 			 * the canputnext.
2922 			 */
2923 			if (canputnext(wqp)) {
2924 				/* Try again */
2925 				mutex_exit(&stp->sd_lock);
2926 				continue;
2927 			}
2928 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAIT,
2929 			    "strwrite wait:q %p wait", wqp);
2930 			if ((error = strwaitq(stp, waitflag, (ssize_t)0,
2931 			    tempmode, -1, &done)) != 0 || done) {
2932 				mutex_exit(&stp->sd_lock);
2933 				if ((vp->v_type == VFIFO) &&
2934 				    (uiop->uio_fmode & FNDELAY) &&
2935 				    (error == EAGAIN))
2936 					error = 0;
2937 				goto out;
2938 			}
2939 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAKE,
2940 			    "strwrite wake:q %p awakes", wqp);
2941 			if ((error = i_straccess(stp, JCWRITE)) != 0) {
2942 				mutex_exit(&stp->sd_lock);
2943 				goto out;
2944 			}
2945 			mutex_exit(&stp->sd_lock);
2946 		}
2947 		waitflag |= NOINTR;
2948 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRWRITE_RESID,
2949 		    "strwrite resid:q %p uiop %p", wqp, uiop);
2950 		if (uiop->uio_resid) {
2951 			/* Recheck for errors - needed for sockets */
2952 			if ((stp->sd_wput_opt & SW_RECHECK_ERR) &&
2953 			    (stp->sd_flag & (STWRERR|STRHUP|STPLEX))) {
2954 				mutex_enter(&stp->sd_lock);
2955 				error = strwriteable(stp, B_FALSE, B_TRUE);
2956 				mutex_exit(&stp->sd_lock);
2957 				if (error != 0)
2958 					return (error);
2959 			}
2960 			continue;
2961 		}
2962 		break;
2963 	}
2964 out:
2965 	/*
2966 	 * For historical reasons, applications expect EAGAIN when a data
2967 	 * mblk_t cannot be allocated, so change ENOMEM back to EAGAIN.
2968 	 */
2969 	if (error == ENOMEM)
2970 		error = EAGAIN;
2971 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2972 	    "strwrite out:q %p out %d error %d", wqp, 2, error);
2973 	return (error);
2974 }
2975 
2976 /*
2977  * Stream head write service routine.
2978  * Its job is to wake up any sleeping writers when a queue
2979  * downstream needs data (part of the flow control in putq and getq).
2980  * It also must wake anyone sleeping on a poll().
2981  * For stream head right below mux module, it must also invoke put procedure
2982  * of next downstream module.
2983  */
2984 int
2985 strwsrv(queue_t *q)
2986 {
2987 	struct stdata *stp;
2988 	queue_t *tq;
2989 	qband_t *qbp;
2990 	int i;
2991 	qband_t *myqbp;
2992 	int isevent;
2993 	unsigned char	qbf[NBAND];	/* band flushing backenable flags */
2994 
2995 	TRACE_1(TR_FAC_STREAMS_FR,
2996 	    TR_STRWSRV, "strwsrv:q %p", q);
2997 	stp = (struct stdata *)q->q_ptr;
2998 	ASSERT(qclaimed(q));
2999 	mutex_enter(&stp->sd_lock);
3000 	ASSERT(!(stp->sd_flag & STPLEX));
3001 
3002 	if (stp->sd_flag & WSLEEP) {
3003 		stp->sd_flag &= ~WSLEEP;
3004 		cv_broadcast(&q->q_wait);
3005 	}
3006 	mutex_exit(&stp->sd_lock);
3007 
3008 	/* The other end of a stream pipe went away. */
3009 	if ((tq = q->q_next) == NULL) {
3010 		return (0);
3011 	}
3012 
3013 	/* Find the next module forward that has a service procedure */
3014 	claimstr(q);
3015 	tq = q->q_nfsrv;
3016 	ASSERT(tq != NULL);
3017 
3018 	if ((q->q_flag & QBACK)) {
3019 		if ((tq->q_flag & QFULL)) {
3020 			mutex_enter(QLOCK(tq));
3021 			if (!(tq->q_flag & QFULL)) {
3022 				mutex_exit(QLOCK(tq));
3023 				goto wakeup;
3024 			}
3025 			/*
3026 			 * The queue must have become full again. Set QWANTW
3027 			 * again so strwsrv will be back enabled when
3028 			 * the queue becomes non-full next time.
3029 			 */
3030 			tq->q_flag |= QWANTW;
3031 			mutex_exit(QLOCK(tq));
3032 		} else {
3033 		wakeup:
3034 			pollwakeup(&stp->sd_pollist, POLLWRNORM);
3035 			mutex_enter(&stp->sd_lock);
3036 			if (stp->sd_sigflags & S_WRNORM)
3037 				strsendsig(stp->sd_siglist, S_WRNORM, 0, 0);
3038 			mutex_exit(&stp->sd_lock);
3039 		}
3040 	}
3041 
3042 	isevent = 0;
3043 	i = 1;
3044 	bzero((caddr_t)qbf, NBAND);
3045 	mutex_enter(QLOCK(tq));
3046 	if ((myqbp = q->q_bandp) != NULL)
3047 		for (qbp = tq->q_bandp; qbp && myqbp; qbp = qbp->qb_next) {
3048 			ASSERT(myqbp);
3049 			if ((myqbp->qb_flag & QB_BACK)) {
3050 				if (qbp->qb_flag & QB_FULL) {
3051 					/*
3052 					 * The band must have become full again.
3053 					 * Set QB_WANTW again so strwsrv will
3054 					 * be back enabled when the band becomes
3055 					 * non-full next time.
3056 					 */
3057 					qbp->qb_flag |= QB_WANTW;
3058 				} else {
3059 					isevent = 1;
3060 					qbf[i] = 1;
3061 				}
3062 			}
3063 			myqbp = myqbp->qb_next;
3064 			i++;
3065 		}
3066 	mutex_exit(QLOCK(tq));
3067 
3068 	if (isevent) {
3069 		for (i = tq->q_nband; i; i--) {
3070 			if (qbf[i]) {
3071 				pollwakeup(&stp->sd_pollist, POLLWRBAND);
3072 				mutex_enter(&stp->sd_lock);
3073 				if (stp->sd_sigflags & S_WRBAND)
3074 					strsendsig(stp->sd_siglist, S_WRBAND,
3075 					    (uchar_t)i, 0);
3076 				mutex_exit(&stp->sd_lock);
3077 			}
3078 		}
3079 	}
3080 
3081 	releasestr(q);
3082 	return (0);
3083 }
3084 
3085 /*
3086  * Special case of strcopyin/strcopyout for copying
3087  * struct strioctl that can deal with both data
3088  * models.
3089  */
3090 
3091 #ifdef	_LP64
3092 
3093 static int
3094 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3095 {
3096 	struct	strioctl32 strioc32;
3097 	struct	strioctl *striocp;
3098 
3099 	if (copyflag & U_TO_K) {
3100 		ASSERT((copyflag & K_TO_K) == 0);
3101 
3102 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
3103 			if (copyin(from, &strioc32, sizeof (strioc32)))
3104 				return (EFAULT);
3105 
3106 			striocp = (struct strioctl *)to;
3107 			striocp->ic_cmd	= strioc32.ic_cmd;
3108 			striocp->ic_timout = strioc32.ic_timout;
3109 			striocp->ic_len	= strioc32.ic_len;
3110 			striocp->ic_dp	= (char *)(uintptr_t)strioc32.ic_dp;
3111 
3112 		} else { /* NATIVE data model */
3113 			if (copyin(from, to, sizeof (struct strioctl))) {
3114 				return (EFAULT);
3115 			} else {
3116 				return (0);
3117 			}
3118 		}
3119 	} else {
3120 		ASSERT(copyflag & K_TO_K);
3121 		bcopy(from, to, sizeof (struct strioctl));
3122 	}
3123 	return (0);
3124 }
3125 
3126 static int
3127 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3128 {
3129 	struct	strioctl32 strioc32;
3130 	struct	strioctl *striocp;
3131 
3132 	if (copyflag & U_TO_K) {
3133 		ASSERT((copyflag & K_TO_K) == 0);
3134 
3135 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
3136 			striocp = (struct strioctl *)from;
3137 			strioc32.ic_cmd	= striocp->ic_cmd;
3138 			strioc32.ic_timout = striocp->ic_timout;
3139 			strioc32.ic_len	= striocp->ic_len;
3140 			strioc32.ic_dp	= (caddr32_t)(uintptr_t)striocp->ic_dp;
3141 			ASSERT((char *)(uintptr_t)strioc32.ic_dp ==
3142 			    striocp->ic_dp);
3143 
3144 			if (copyout(&strioc32, to, sizeof (strioc32)))
3145 				return (EFAULT);
3146 
3147 		} else { /* NATIVE data model */
3148 			if (copyout(from, to, sizeof (struct strioctl))) {
3149 				return (EFAULT);
3150 			} else {
3151 				return (0);
3152 			}
3153 		}
3154 	} else {
3155 		ASSERT(copyflag & K_TO_K);
3156 		bcopy(from, to, sizeof (struct strioctl));
3157 	}
3158 	return (0);
3159 }
3160 
3161 #else	/* ! _LP64 */
3162 
3163 /* ARGSUSED2 */
3164 static int
3165 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3166 {
3167 	return (strcopyin(from, to, sizeof (struct strioctl), copyflag));
3168 }
3169 
3170 /* ARGSUSED2 */
3171 static int
3172 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3173 {
3174 	return (strcopyout(from, to, sizeof (struct strioctl), copyflag));
3175 }
3176 
3177 #endif	/* _LP64 */
3178 
3179 /*
3180  * Determine type of job control semantics expected by user.  The
3181  * possibilities are:
3182  *	JCREAD	- Behaves like read() on fd; send SIGTTIN
3183  *	JCWRITE	- Behaves like write() on fd; send SIGTTOU if TOSTOP set
3184  *	JCSETP	- Sets a value in the stream; send SIGTTOU, ignore TOSTOP
3185  *	JCGETP	- Gets a value in the stream; no signals.
3186  * See straccess in strsubr.c for usage of these values.
3187  *
3188  * This routine also returns -1 for I_STR as a special case; the
3189  * caller must call again with the real ioctl number for
3190  * classification.
3191  */
3192 static int
3193 job_control_type(int cmd)
3194 {
3195 	switch (cmd) {
3196 	case I_STR:
3197 		return (-1);
3198 
3199 	case I_RECVFD:
3200 	case I_E_RECVFD:
3201 		return (JCREAD);
3202 
3203 	case I_FDINSERT:
3204 	case I_SENDFD:
3205 		return (JCWRITE);
3206 
3207 	case TCSETA:
3208 	case TCSETAW:
3209 	case TCSETAF:
3210 	case TCSBRK:
3211 	case TCXONC:
3212 	case TCFLSH:
3213 	case TCDSET:	/* Obsolete */
3214 	case TIOCSWINSZ:
3215 	case TCSETS:
3216 	case TCSETSW:
3217 	case TCSETSF:
3218 	case TIOCSETD:
3219 	case TIOCHPCL:
3220 	case TIOCSETP:
3221 	case TIOCSETN:
3222 	case TIOCEXCL:
3223 	case TIOCNXCL:
3224 	case TIOCFLUSH:
3225 	case TIOCSETC:
3226 	case TIOCLBIS:
3227 	case TIOCLBIC:
3228 	case TIOCLSET:
3229 	case TIOCSBRK:
3230 	case TIOCCBRK:
3231 	case TIOCSDTR:
3232 	case TIOCCDTR:
3233 	case TIOCSLTC:
3234 	case TIOCSTOP:
3235 	case TIOCSTART:
3236 	case TIOCSTI:
3237 	case TIOCSPGRP:
3238 	case TIOCMSET:
3239 	case TIOCMBIS:
3240 	case TIOCMBIC:
3241 	case TIOCREMOTE:
3242 	case TIOCSIGNAL:
3243 	case LDSETT:
3244 	case LDSMAP:	/* Obsolete */
3245 	case DIOCSETP:
3246 	case I_FLUSH:
3247 	case I_SRDOPT:
3248 	case I_SETSIG:
3249 	case I_SWROPT:
3250 	case I_FLUSHBAND:
3251 	case I_SETCLTIME:
3252 	case I_SERROPT:
3253 	case I_ESETSIG:
3254 	case FIONBIO:
3255 	case FIOASYNC:
3256 	case FIOSETOWN:
3257 	case JBOOT:	/* Obsolete */
3258 	case JTERM:	/* Obsolete */
3259 	case JTIMOM:	/* Obsolete */
3260 	case JZOMBOOT:	/* Obsolete */
3261 	case JAGENT:	/* Obsolete */
3262 	case JTRUN:	/* Obsolete */
3263 	case JXTPROTO:	/* Obsolete */
3264 	case TIOCSETLD:
3265 		return (JCSETP);
3266 	}
3267 
3268 	return (JCGETP);
3269 }
3270 
3271 /*
3272  * ioctl for streams
3273  */
3274 int
3275 strioctl(struct vnode *vp, int cmd, intptr_t arg, int flag, int copyflag,
3276     cred_t *crp, int *rvalp)
3277 {
3278 	struct stdata *stp;
3279 	struct strcmd *scp;
3280 	struct strioctl strioc;
3281 	struct uio uio;
3282 	struct iovec iov;
3283 	int access;
3284 	mblk_t *mp;
3285 	int error = 0;
3286 	int done = 0;
3287 	ssize_t	rmin, rmax;
3288 	queue_t *wrq;
3289 	queue_t *rdq;
3290 	boolean_t kioctl = B_FALSE;
3291 
3292 	if (flag & FKIOCTL) {
3293 		copyflag = K_TO_K;
3294 		kioctl = B_TRUE;
3295 	}
3296 	ASSERT(vp->v_stream);
3297 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
3298 	stp = vp->v_stream;
3299 
3300 	TRACE_3(TR_FAC_STREAMS_FR, TR_IOCTL_ENTER,
3301 	    "strioctl:stp %p cmd %X arg %lX", stp, cmd, arg);
3302 
3303 	if (audit_active)
3304 		audit_strioctl(vp, cmd, arg, flag, copyflag, crp, rvalp);
3305 
3306 	/*
3307 	 * If the copy is kernel to kernel, make sure that the FNATIVE
3308 	 * flag is set.  After this it would be a serious error to have
3309 	 * no model flag.
3310 	 */
3311 	if (copyflag == K_TO_K)
3312 		flag = (flag & ~FMODELS) | FNATIVE;
3313 
3314 	ASSERT((flag & FMODELS) != 0);
3315 
3316 	wrq = stp->sd_wrq;
3317 	rdq = _RD(wrq);
3318 
3319 	access = job_control_type(cmd);
3320 
3321 	/* We should never see these here, should be handled by iwscn */
3322 	if (cmd == SRIOCSREDIR || cmd == SRIOCISREDIR)
3323 		return (EINVAL);
3324 
3325 	mutex_enter(&stp->sd_lock);
3326 	if ((access != -1) && ((error = i_straccess(stp, access)) != 0)) {
3327 		mutex_exit(&stp->sd_lock);
3328 		return (error);
3329 	}
3330 	mutex_exit(&stp->sd_lock);
3331 
3332 	/*
3333 	 * Check for sgttyb-related ioctls first, and complain as
3334 	 * necessary.
3335 	 */
3336 	switch (cmd) {
3337 	case TIOCGETP:
3338 	case TIOCSETP:
3339 	case TIOCSETN:
3340 		if (sgttyb_handling >= 2 && !sgttyb_complaint) {
3341 			sgttyb_complaint = B_TRUE;
3342 			cmn_err(CE_NOTE,
3343 			    "application used obsolete TIOC[GS]ET");
3344 		}
3345 		if (sgttyb_handling >= 3) {
3346 			tsignal(curthread, SIGSYS);
3347 			return (EIO);
3348 		}
3349 		break;
3350 	}
3351 
3352 	mutex_enter(&stp->sd_lock);
3353 
3354 	switch (cmd) {
3355 	case I_RECVFD:
3356 	case I_E_RECVFD:
3357 	case I_PEEK:
3358 	case I_NREAD:
3359 	case FIONREAD:
3360 	case FIORDCHK:
3361 	case I_ATMARK:
3362 	case FIONBIO:
3363 	case FIOASYNC:
3364 		if (stp->sd_flag & (STRDERR|STPLEX)) {
3365 			error = strgeterr(stp, STRDERR|STPLEX, 0);
3366 			if (error != 0) {
3367 				mutex_exit(&stp->sd_lock);
3368 				return (error);
3369 			}
3370 		}
3371 		break;
3372 
3373 	default:
3374 		if (stp->sd_flag & (STRDERR|STWRERR|STPLEX)) {
3375 			error = strgeterr(stp, STRDERR|STWRERR|STPLEX, 0);
3376 			if (error != 0) {
3377 				mutex_exit(&stp->sd_lock);
3378 				return (error);
3379 			}
3380 		}
3381 	}
3382 
3383 	mutex_exit(&stp->sd_lock);
3384 
3385 	switch (cmd) {
3386 	default:
3387 		/*
3388 		 * The stream head has hardcoded knowledge of a
3389 		 * miscellaneous collection of terminal-, keyboard- and
3390 		 * mouse-related ioctls, enumerated below.  This hardcoded
3391 		 * knowledge allows the stream head to automatically
3392 		 * convert transparent ioctl requests made by userland
3393 		 * programs into I_STR ioctls which many old STREAMS
3394 		 * modules and drivers require.
3395 		 *
3396 		 * No new ioctls should ever be added to this list.
3397 		 * Instead, the STREAMS module or driver should be written
3398 		 * to either handle transparent ioctls or require any
3399 		 * userland programs to use I_STR ioctls (by returning
3400 		 * EINVAL to any transparent ioctl requests).
3401 		 *
3402 		 * More importantly, removing ioctls from this list should
3403 		 * be done with the utmost care, since our STREAMS modules
3404 		 * and drivers *count* on the stream head performing this
3405 		 * conversion, and thus may panic while processing
3406 		 * transparent ioctl request for one of these ioctls (keep
3407 		 * in mind that third party modules and drivers may have
3408 		 * similar problems).
3409 		 */
3410 		if (((cmd & IOCTYPE) == LDIOC) ||
3411 		    ((cmd & IOCTYPE) == tIOC) ||
3412 		    ((cmd & IOCTYPE) == TIOC) ||
3413 		    ((cmd & IOCTYPE) == KIOC) ||
3414 		    ((cmd & IOCTYPE) == MSIOC) ||
3415 		    ((cmd & IOCTYPE) == VUIOC)) {
3416 			/*
3417 			 * The ioctl is a tty ioctl - set up strioc buffer
3418 			 * and call strdoioctl() to do the work.
3419 			 */
3420 			if (stp->sd_flag & STRHUP)
3421 				return (ENXIO);
3422 			strioc.ic_cmd = cmd;
3423 			strioc.ic_timout = INFTIM;
3424 
3425 			switch (cmd) {
3426 
3427 			case TCXONC:
3428 			case TCSBRK:
3429 			case TCFLSH:
3430 			case TCDSET:
3431 				{
3432 				int native_arg = (int)arg;
3433 				strioc.ic_len = sizeof (int);
3434 				strioc.ic_dp = (char *)&native_arg;
3435 				return (strdoioctl(stp, &strioc, flag,
3436 				    K_TO_K, crp, rvalp));
3437 				}
3438 
3439 			case TCSETA:
3440 			case TCSETAW:
3441 			case TCSETAF:
3442 				strioc.ic_len = sizeof (struct termio);
3443 				strioc.ic_dp = (char *)arg;
3444 				return (strdoioctl(stp, &strioc, flag,
3445 				    copyflag, crp, rvalp));
3446 
3447 			case TCSETS:
3448 			case TCSETSW:
3449 			case TCSETSF:
3450 				strioc.ic_len = sizeof (struct termios);
3451 				strioc.ic_dp = (char *)arg;
3452 				return (strdoioctl(stp, &strioc, flag,
3453 				    copyflag, crp, rvalp));
3454 
3455 			case LDSETT:
3456 				strioc.ic_len = sizeof (struct termcb);
3457 				strioc.ic_dp = (char *)arg;
3458 				return (strdoioctl(stp, &strioc, flag,
3459 				    copyflag, crp, rvalp));
3460 
3461 			case TIOCSETP:
3462 				strioc.ic_len = sizeof (struct sgttyb);
3463 				strioc.ic_dp = (char *)arg;
3464 				return (strdoioctl(stp, &strioc, flag,
3465 				    copyflag, crp, rvalp));
3466 
3467 			case TIOCSTI:
3468 				if ((flag & FREAD) == 0 &&
3469 				    secpolicy_sti(crp) != 0) {
3470 					return (EPERM);
3471 				}
3472 				mutex_enter(&stp->sd_lock);
3473 				mutex_enter(&curproc->p_splock);
3474 				if (stp->sd_sidp != curproc->p_sessp->s_sidp &&
3475 				    secpolicy_sti(crp) != 0) {
3476 					mutex_exit(&curproc->p_splock);
3477 					mutex_exit(&stp->sd_lock);
3478 					return (EACCES);
3479 				}
3480 				mutex_exit(&curproc->p_splock);
3481 				mutex_exit(&stp->sd_lock);
3482 
3483 				strioc.ic_len = sizeof (char);
3484 				strioc.ic_dp = (char *)arg;
3485 				return (strdoioctl(stp, &strioc, flag,
3486 				    copyflag, crp, rvalp));
3487 
3488 			case TIOCSWINSZ:
3489 				strioc.ic_len = sizeof (struct winsize);
3490 				strioc.ic_dp = (char *)arg;
3491 				return (strdoioctl(stp, &strioc, flag,
3492 				    copyflag, crp, rvalp));
3493 
3494 			case TIOCSSIZE:
3495 				strioc.ic_len = sizeof (struct ttysize);
3496 				strioc.ic_dp = (char *)arg;
3497 				return (strdoioctl(stp, &strioc, flag,
3498 				    copyflag, crp, rvalp));
3499 
3500 			case TIOCSSOFTCAR:
3501 			case KIOCTRANS:
3502 			case KIOCTRANSABLE:
3503 			case KIOCCMD:
3504 			case KIOCSDIRECT:
3505 			case KIOCSCOMPAT:
3506 			case KIOCSKABORTEN:
3507 			case KIOCSRPTDELAY:
3508 			case KIOCSRPTRATE:
3509 			case VUIDSFORMAT:
3510 			case TIOCSPPS:
3511 				strioc.ic_len = sizeof (int);
3512 				strioc.ic_dp = (char *)arg;
3513 				return (strdoioctl(stp, &strioc, flag,
3514 				    copyflag, crp, rvalp));
3515 
3516 			case KIOCSETKEY:
3517 			case KIOCGETKEY:
3518 				strioc.ic_len = sizeof (struct kiockey);
3519 				strioc.ic_dp = (char *)arg;
3520 				return (strdoioctl(stp, &strioc, flag,
3521 				    copyflag, crp, rvalp));
3522 
3523 			case KIOCSKEY:
3524 			case KIOCGKEY:
3525 				strioc.ic_len = sizeof (struct kiockeymap);
3526 				strioc.ic_dp = (char *)arg;
3527 				return (strdoioctl(stp, &strioc, flag,
3528 				    copyflag, crp, rvalp));
3529 
3530 			case KIOCSLED:
3531 				/* arg is a pointer to char */
3532 				strioc.ic_len = sizeof (char);
3533 				strioc.ic_dp = (char *)arg;
3534 				return (strdoioctl(stp, &strioc, flag,
3535 				    copyflag, crp, rvalp));
3536 
3537 			case MSIOSETPARMS:
3538 				strioc.ic_len = sizeof (Ms_parms);
3539 				strioc.ic_dp = (char *)arg;
3540 				return (strdoioctl(stp, &strioc, flag,
3541 				    copyflag, crp, rvalp));
3542 
3543 			case VUIDSADDR:
3544 			case VUIDGADDR:
3545 				strioc.ic_len = sizeof (struct vuid_addr_probe);
3546 				strioc.ic_dp = (char *)arg;
3547 				return (strdoioctl(stp, &strioc, flag,
3548 				    copyflag, crp, rvalp));
3549 
3550 			/*
3551 			 * These M_IOCTL's don't require any data to be sent
3552 			 * downstream, and the driver will allocate and link
3553 			 * on its own mblk_t upon M_IOCACK -- thus we set
3554 			 * ic_len to zero and set ic_dp to arg so we know
3555 			 * where to copyout to later.
3556 			 */
3557 			case TIOCGSOFTCAR:
3558 			case TIOCGWINSZ:
3559 			case TIOCGSIZE:
3560 			case KIOCGTRANS:
3561 			case KIOCGTRANSABLE:
3562 			case KIOCTYPE:
3563 			case KIOCGDIRECT:
3564 			case KIOCGCOMPAT:
3565 			case KIOCLAYOUT:
3566 			case KIOCGLED:
3567 			case MSIOGETPARMS:
3568 			case MSIOBUTTONS:
3569 			case VUIDGFORMAT:
3570 			case TIOCGPPS:
3571 			case TIOCGPPSEV:
3572 			case TCGETA:
3573 			case TCGETS:
3574 			case LDGETT:
3575 			case TIOCGETP:
3576 			case KIOCGRPTDELAY:
3577 			case KIOCGRPTRATE:
3578 				strioc.ic_len = 0;
3579 				strioc.ic_dp = (char *)arg;
3580 				return (strdoioctl(stp, &strioc, flag,
3581 				    copyflag, crp, rvalp));
3582 			}
3583 		}
3584 
3585 		/*
3586 		 * Unknown cmd - send it down as a transparent ioctl.
3587 		 */
3588 		strioc.ic_cmd = cmd;
3589 		strioc.ic_timout = INFTIM;
3590 		strioc.ic_len = TRANSPARENT;
3591 		strioc.ic_dp = (char *)&arg;
3592 
3593 		return (strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp));
3594 
3595 	case I_STR:
3596 		/*
3597 		 * Stream ioctl.  Read in an strioctl buffer from the user
3598 		 * along with any data specified and send it downstream.
3599 		 * Strdoioctl will wait allow only one ioctl message at
3600 		 * a time, and waits for the acknowledgement.
3601 		 */
3602 
3603 		if (stp->sd_flag & STRHUP)
3604 			return (ENXIO);
3605 
3606 		error = strcopyin_strioctl((void *)arg, &strioc, flag,
3607 		    copyflag);
3608 		if (error != 0)
3609 			return (error);
3610 
3611 		if ((strioc.ic_len < 0) || (strioc.ic_timout < -1))
3612 			return (EINVAL);
3613 
3614 		access = job_control_type(strioc.ic_cmd);
3615 		mutex_enter(&stp->sd_lock);
3616 		if ((access != -1) &&
3617 		    ((error = i_straccess(stp, access)) != 0)) {
3618 			mutex_exit(&stp->sd_lock);
3619 			return (error);
3620 		}
3621 		mutex_exit(&stp->sd_lock);
3622 
3623 		/*
3624 		 * The I_STR facility provides a trap door for malicious
3625 		 * code to send down bogus streamio(7I) ioctl commands to
3626 		 * unsuspecting STREAMS modules and drivers which expect to
3627 		 * only get these messages from the stream head.
3628 		 * Explicitly prohibit any streamio ioctls which can be
3629 		 * passed downstream by the stream head.  Note that we do
3630 		 * not block all streamio ioctls because the ioctl
3631 		 * numberspace is not well managed and thus it's possible
3632 		 * that a module or driver's ioctl numbers may accidentally
3633 		 * collide with them.
3634 		 */
3635 		switch (strioc.ic_cmd) {
3636 		case I_LINK:
3637 		case I_PLINK:
3638 		case I_UNLINK:
3639 		case I_PUNLINK:
3640 		case _I_GETPEERCRED:
3641 		case _I_PLINK_LH:
3642 			return (EINVAL);
3643 		}
3644 
3645 		error = strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp);
3646 		if (error == 0) {
3647 			error = strcopyout_strioctl(&strioc, (void *)arg,
3648 			    flag, copyflag);
3649 		}
3650 		return (error);
3651 
3652 	case _I_CMD:
3653 		/*
3654 		 * Like I_STR, but without using M_IOC* messages and without
3655 		 * copyins/copyouts beyond the passed-in argument.
3656 		 */
3657 		if (stp->sd_flag & STRHUP)
3658 			return (ENXIO);
3659 
3660 		if ((scp = kmem_alloc(sizeof (strcmd_t), KM_NOSLEEP)) == NULL)
3661 			return (ENOMEM);
3662 
3663 		if (copyin((void *)arg, scp, sizeof (strcmd_t))) {
3664 			kmem_free(scp, sizeof (strcmd_t));
3665 			return (EFAULT);
3666 		}
3667 
3668 		access = job_control_type(scp->sc_cmd);
3669 		mutex_enter(&stp->sd_lock);
3670 		if (access != -1 && (error = i_straccess(stp, access)) != 0) {
3671 			mutex_exit(&stp->sd_lock);
3672 			kmem_free(scp, sizeof (strcmd_t));
3673 			return (error);
3674 		}
3675 		mutex_exit(&stp->sd_lock);
3676 
3677 		*rvalp = 0;
3678 		if ((error = strdocmd(stp, scp, crp)) == 0) {
3679 			if (copyout(scp, (void *)arg, sizeof (strcmd_t)))
3680 				error = EFAULT;
3681 		}
3682 		kmem_free(scp, sizeof (strcmd_t));
3683 		return (error);
3684 
3685 	case I_NREAD:
3686 		/*
3687 		 * Return number of bytes of data in first message
3688 		 * in queue in "arg" and return the number of messages
3689 		 * in queue in return value.
3690 		 */
3691 	{
3692 		size_t	size;
3693 		int	retval;
3694 		int	count = 0;
3695 
3696 		mutex_enter(QLOCK(rdq));
3697 
3698 		size = msgdsize(rdq->q_first);
3699 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3700 			count++;
3701 
3702 		mutex_exit(QLOCK(rdq));
3703 		if (stp->sd_struiordq) {
3704 			infod_t infod;
3705 
3706 			infod.d_cmd = INFOD_COUNT;
3707 			infod.d_count = 0;
3708 			if (count == 0) {
3709 				infod.d_cmd |= INFOD_FIRSTBYTES;
3710 				infod.d_bytes = 0;
3711 			}
3712 			infod.d_res = 0;
3713 			(void) infonext(rdq, &infod);
3714 			count += infod.d_count;
3715 			if (infod.d_res & INFOD_FIRSTBYTES)
3716 				size = infod.d_bytes;
3717 		}
3718 
3719 		/*
3720 		 * Drop down from size_t to the "int" required by the
3721 		 * interface.  Cap at INT_MAX.
3722 		 */
3723 		retval = MIN(size, INT_MAX);
3724 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3725 		    copyflag);
3726 		if (!error)
3727 			*rvalp = count;
3728 		return (error);
3729 	}
3730 
3731 	case FIONREAD:
3732 		/*
3733 		 * Return number of bytes of data in all data messages
3734 		 * in queue in "arg".
3735 		 */
3736 	{
3737 		size_t	size = 0;
3738 		int	retval;
3739 
3740 		mutex_enter(QLOCK(rdq));
3741 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3742 			size += msgdsize(mp);
3743 		mutex_exit(QLOCK(rdq));
3744 
3745 		if (stp->sd_struiordq) {
3746 			infod_t infod;
3747 
3748 			infod.d_cmd = INFOD_BYTES;
3749 			infod.d_res = 0;
3750 			infod.d_bytes = 0;
3751 			(void) infonext(rdq, &infod);
3752 			size += infod.d_bytes;
3753 		}
3754 
3755 		/*
3756 		 * Drop down from size_t to the "int" required by the
3757 		 * interface.  Cap at INT_MAX.
3758 		 */
3759 		retval = MIN(size, INT_MAX);
3760 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3761 		    copyflag);
3762 
3763 		*rvalp = 0;
3764 		return (error);
3765 	}
3766 	case FIORDCHK:
3767 		/*
3768 		 * FIORDCHK does not use arg value (like FIONREAD),
3769 		 * instead a count is returned. I_NREAD value may
3770 		 * not be accurate but safe. The real thing to do is
3771 		 * to add the msgdsizes of all data  messages until
3772 		 * a non-data message.
3773 		 */
3774 	{
3775 		size_t size = 0;
3776 
3777 		mutex_enter(QLOCK(rdq));
3778 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3779 			size += msgdsize(mp);
3780 		mutex_exit(QLOCK(rdq));
3781 
3782 		if (stp->sd_struiordq) {
3783 			infod_t infod;
3784 
3785 			infod.d_cmd = INFOD_BYTES;
3786 			infod.d_res = 0;
3787 			infod.d_bytes = 0;
3788 			(void) infonext(rdq, &infod);
3789 			size += infod.d_bytes;
3790 		}
3791 
3792 		/*
3793 		 * Since ioctl returns an int, and memory sizes under
3794 		 * LP64 may not fit, we return INT_MAX if the count was
3795 		 * actually greater.
3796 		 */
3797 		*rvalp = MIN(size, INT_MAX);
3798 		return (0);
3799 	}
3800 
3801 	case I_FIND:
3802 		/*
3803 		 * Get module name.
3804 		 */
3805 	{
3806 		char mname[FMNAMESZ + 1];
3807 		queue_t *q;
3808 
3809 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3810 		    mname, FMNAMESZ + 1, NULL);
3811 		if (error)
3812 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3813 
3814 		/*
3815 		 * Return EINVAL if we're handed a bogus module name.
3816 		 */
3817 		if (fmodsw_find(mname, FMODSW_LOAD) == NULL) {
3818 			TRACE_0(TR_FAC_STREAMS_FR,
3819 			    TR_I_CANT_FIND, "couldn't I_FIND");
3820 			return (EINVAL);
3821 		}
3822 
3823 		*rvalp = 0;
3824 
3825 		/* Look downstream to see if module is there. */
3826 		claimstr(stp->sd_wrq);
3827 		for (q = stp->sd_wrq->q_next; q; q = q->q_next) {
3828 			if (q->q_flag&QREADR) {
3829 				q = NULL;
3830 				break;
3831 			}
3832 			if (strcmp(mname, q->q_qinfo->qi_minfo->mi_idname) == 0)
3833 				break;
3834 		}
3835 		releasestr(stp->sd_wrq);
3836 
3837 		*rvalp = (q ? 1 : 0);
3838 		return (error);
3839 	}
3840 
3841 	case I_PUSH:
3842 	case __I_PUSH_NOCTTY:
3843 		/*
3844 		 * Push a module.
3845 		 * For the case __I_PUSH_NOCTTY push a module but
3846 		 * do not allocate controlling tty. See bugid 4025044
3847 		 */
3848 
3849 	{
3850 		char mname[FMNAMESZ + 1];
3851 		fmodsw_impl_t *fp;
3852 		dev_t dummydev;
3853 
3854 		if (stp->sd_flag & STRHUP)
3855 			return (ENXIO);
3856 
3857 		/*
3858 		 * Get module name and look up in fmodsw.
3859 		 */
3860 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3861 		    mname, FMNAMESZ + 1, NULL);
3862 		if (error)
3863 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3864 
3865 		if ((fp = fmodsw_find(mname, FMODSW_HOLD | FMODSW_LOAD)) ==
3866 		    NULL)
3867 			return (EINVAL);
3868 
3869 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_PUSH,
3870 		    "I_PUSH:fp %p stp %p", fp, stp);
3871 
3872 		if (error = strstartplumb(stp, flag, cmd)) {
3873 			fmodsw_rele(fp);
3874 			return (error);
3875 		}
3876 
3877 		/*
3878 		 * See if any more modules can be pushed on this stream.
3879 		 * Note that this check must be done after strstartplumb()
3880 		 * since otherwise multiple threads issuing I_PUSHes on
3881 		 * the same stream will be able to exceed nstrpush.
3882 		 */
3883 		mutex_enter(&stp->sd_lock);
3884 		if (stp->sd_pushcnt >= nstrpush) {
3885 			fmodsw_rele(fp);
3886 			strendplumb(stp);
3887 			mutex_exit(&stp->sd_lock);
3888 			return (EINVAL);
3889 		}
3890 		mutex_exit(&stp->sd_lock);
3891 
3892 		/*
3893 		 * Push new module and call its open routine
3894 		 * via qattach().  Modules don't change device
3895 		 * numbers, so just ignore dummydev here.
3896 		 */
3897 		dummydev = vp->v_rdev;
3898 		if ((error = qattach(rdq, &dummydev, 0, crp, fp,
3899 		    B_FALSE)) == 0) {
3900 			if (vp->v_type == VCHR && /* sorry, no pipes allowed */
3901 			    (cmd == I_PUSH) && (stp->sd_flag & STRISTTY)) {
3902 				/*
3903 				 * try to allocate it as a controlling terminal
3904 				 */
3905 				(void) strctty(stp);
3906 			}
3907 		}
3908 
3909 		mutex_enter(&stp->sd_lock);
3910 
3911 		/*
3912 		 * As a performance concern we are caching the values of
3913 		 * q_minpsz and q_maxpsz of the module below the stream
3914 		 * head in the stream head.
3915 		 */
3916 		mutex_enter(QLOCK(stp->sd_wrq->q_next));
3917 		rmin = stp->sd_wrq->q_next->q_minpsz;
3918 		rmax = stp->sd_wrq->q_next->q_maxpsz;
3919 		mutex_exit(QLOCK(stp->sd_wrq->q_next));
3920 
3921 		/* Do this processing here as a performance concern */
3922 		if (strmsgsz != 0) {
3923 			if (rmax == INFPSZ)
3924 				rmax = strmsgsz;
3925 			else  {
3926 				if (vp->v_type == VFIFO)
3927 					rmax = MIN(PIPE_BUF, rmax);
3928 				else	rmax = MIN(strmsgsz, rmax);
3929 			}
3930 		}
3931 
3932 		mutex_enter(QLOCK(wrq));
3933 		stp->sd_qn_minpsz = rmin;
3934 		stp->sd_qn_maxpsz = rmax;
3935 		mutex_exit(QLOCK(wrq));
3936 
3937 		strendplumb(stp);
3938 		mutex_exit(&stp->sd_lock);
3939 		return (error);
3940 	}
3941 
3942 	case I_POP:
3943 	{
3944 		queue_t	*q;
3945 
3946 		if (stp->sd_flag & STRHUP)
3947 			return (ENXIO);
3948 		if (!wrq->q_next)	/* for broken pipes */
3949 			return (EINVAL);
3950 
3951 		if (error = strstartplumb(stp, flag, cmd))
3952 			return (error);
3953 
3954 		/*
3955 		 * If there is an anchor on this stream and popping
3956 		 * the current module would attempt to pop through the
3957 		 * anchor, then disallow the pop unless we have sufficient
3958 		 * privileges; take the cheapest (non-locking) check
3959 		 * first.
3960 		 */
3961 		if (secpolicy_ip_config(crp, B_TRUE) != 0 ||
3962 		    (stp->sd_anchorzone != crgetzoneid(crp))) {
3963 			mutex_enter(&stp->sd_lock);
3964 			/*
3965 			 * Anchors only apply if there's at least one
3966 			 * module on the stream (sd_pushcnt > 0).
3967 			 */
3968 			if (stp->sd_pushcnt > 0 &&
3969 			    stp->sd_pushcnt == stp->sd_anchor &&
3970 			    stp->sd_vnode->v_type != VFIFO) {
3971 				strendplumb(stp);
3972 				mutex_exit(&stp->sd_lock);
3973 				if (stp->sd_anchorzone != crgetzoneid(crp))
3974 					return (EINVAL);
3975 				/* Audit and report error */
3976 				return (secpolicy_ip_config(crp, B_FALSE));
3977 			}
3978 			mutex_exit(&stp->sd_lock);
3979 		}
3980 
3981 		q = wrq->q_next;
3982 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_POP,
3983 		    "I_POP:%p from %p", q, stp);
3984 		if (q->q_next == NULL || (q->q_flag & (QREADR|QISDRV))) {
3985 			error = EINVAL;
3986 		} else {
3987 			qdetach(_RD(q), 1, flag, crp, B_FALSE);
3988 			error = 0;
3989 		}
3990 		mutex_enter(&stp->sd_lock);
3991 
3992 		/*
3993 		 * As a performance concern we are caching the values of
3994 		 * q_minpsz and q_maxpsz of the module below the stream
3995 		 * head in the stream head.
3996 		 */
3997 		mutex_enter(QLOCK(wrq->q_next));
3998 		rmin = wrq->q_next->q_minpsz;
3999 		rmax = wrq->q_next->q_maxpsz;
4000 		mutex_exit(QLOCK(wrq->q_next));
4001 
4002 		/* Do this processing here as a performance concern */
4003 		if (strmsgsz != 0) {
4004 			if (rmax == INFPSZ)
4005 				rmax = strmsgsz;
4006 			else  {
4007 				if (vp->v_type == VFIFO)
4008 					rmax = MIN(PIPE_BUF, rmax);
4009 				else	rmax = MIN(strmsgsz, rmax);
4010 			}
4011 		}
4012 
4013 		mutex_enter(QLOCK(wrq));
4014 		stp->sd_qn_minpsz = rmin;
4015 		stp->sd_qn_maxpsz = rmax;
4016 		mutex_exit(QLOCK(wrq));
4017 
4018 		/* If we popped through the anchor, then reset the anchor. */
4019 		if (stp->sd_pushcnt < stp->sd_anchor) {
4020 			stp->sd_anchor = 0;
4021 			stp->sd_anchorzone = 0;
4022 		}
4023 		strendplumb(stp);
4024 		mutex_exit(&stp->sd_lock);
4025 		return (error);
4026 	}
4027 
4028 	case _I_MUXID2FD:
4029 	{
4030 		/*
4031 		 * Create a fd for a I_PLINK'ed lower stream with a given
4032 		 * muxid.  With the fd, application can send down ioctls,
4033 		 * like I_LIST, to the previously I_PLINK'ed stream.  Note
4034 		 * that after getting the fd, the application has to do an
4035 		 * I_PUNLINK on the muxid before it can do any operation
4036 		 * on the lower stream.  This is required by spec1170.
4037 		 *
4038 		 * The fd used to do this ioctl should point to the same
4039 		 * controlling device used to do the I_PLINK.  If it uses
4040 		 * a different stream or an invalid muxid, I_MUXID2FD will
4041 		 * fail.  The error code is set to EINVAL.
4042 		 *
4043 		 * The intended use of this interface is the following.
4044 		 * An application I_PLINK'ed a stream and exits.  The fd
4045 		 * to the lower stream is gone.  Another application
4046 		 * wants to get a fd to the lower stream, it uses I_MUXID2FD.
4047 		 */
4048 		int muxid = (int)arg;
4049 		int fd;
4050 		linkinfo_t *linkp;
4051 		struct file *fp;
4052 		netstack_t *ns;
4053 		str_stack_t *ss;
4054 
4055 		/*
4056 		 * Do not allow the wildcard muxid.  This ioctl is not
4057 		 * intended to find arbitrary link.
4058 		 */
4059 		if (muxid == 0) {
4060 			return (EINVAL);
4061 		}
4062 
4063 		ns = netstack_find_by_cred(crp);
4064 		ASSERT(ns != NULL);
4065 		ss = ns->netstack_str;
4066 		ASSERT(ss != NULL);
4067 
4068 		mutex_enter(&muxifier);
4069 		linkp = findlinks(vp->v_stream, muxid, LINKPERSIST, ss);
4070 		if (linkp == NULL) {
4071 			mutex_exit(&muxifier);
4072 			netstack_rele(ss->ss_netstack);
4073 			return (EINVAL);
4074 		}
4075 
4076 		if ((fd = ufalloc(0)) == -1) {
4077 			mutex_exit(&muxifier);
4078 			netstack_rele(ss->ss_netstack);
4079 			return (EMFILE);
4080 		}
4081 		fp = linkp->li_fpdown;
4082 		mutex_enter(&fp->f_tlock);
4083 		fp->f_count++;
4084 		mutex_exit(&fp->f_tlock);
4085 		mutex_exit(&muxifier);
4086 		setf(fd, fp);
4087 		*rvalp = fd;
4088 		netstack_rele(ss->ss_netstack);
4089 		return (0);
4090 	}
4091 
4092 	case _I_INSERT:
4093 	{
4094 		/*
4095 		 * To insert a module to a given position in a stream.
4096 		 * In the first release, only allow privileged user
4097 		 * to use this ioctl. Furthermore, the insert is only allowed
4098 		 * below an anchor if the zoneid is the same as the zoneid
4099 		 * which created the anchor.
4100 		 *
4101 		 * Note that we do not plan to support this ioctl
4102 		 * on pipes in the first release.  We want to learn more
4103 		 * about the implications of these ioctls before extending
4104 		 * their support.  And we do not think these features are
4105 		 * valuable for pipes.
4106 		 *
4107 		 * Neither do we support O/C hot stream.  Note that only
4108 		 * the upper streams of TCP/IP stack are O/C hot streams.
4109 		 * The lower IP stream is not.
4110 		 * When there is a O/C cold barrier, we only allow inserts
4111 		 * above the barrier.
4112 		 */
4113 		STRUCT_DECL(strmodconf, strmodinsert);
4114 		char mod_name[FMNAMESZ + 1];
4115 		fmodsw_impl_t *fp;
4116 		dev_t dummydev;
4117 		queue_t *tmp_wrq;
4118 		int pos;
4119 		boolean_t is_insert;
4120 
4121 		STRUCT_INIT(strmodinsert, flag);
4122 		if (stp->sd_flag & STRHUP)
4123 			return (ENXIO);
4124 		if (STRMATED(stp))
4125 			return (EINVAL);
4126 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4127 			return (error);
4128 		if (stp->sd_anchor != 0 &&
4129 		    stp->sd_anchorzone != crgetzoneid(crp))
4130 			return (EINVAL);
4131 
4132 		error = strcopyin((void *)arg, STRUCT_BUF(strmodinsert),
4133 		    STRUCT_SIZE(strmodinsert), copyflag);
4134 		if (error)
4135 			return (error);
4136 
4137 		/*
4138 		 * Get module name and look up in fmodsw.
4139 		 */
4140 		error = (copyflag & U_TO_K ? copyinstr :
4141 		    copystr)(STRUCT_FGETP(strmodinsert, mod_name),
4142 		    mod_name, FMNAMESZ + 1, NULL);
4143 		if (error)
4144 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4145 
4146 		if ((fp = fmodsw_find(mod_name, FMODSW_HOLD | FMODSW_LOAD)) ==
4147 		    NULL)
4148 			return (EINVAL);
4149 
4150 		if (error = strstartplumb(stp, flag, cmd)) {
4151 			fmodsw_rele(fp);
4152 			return (error);
4153 		}
4154 
4155 		/*
4156 		 * Is this _I_INSERT just like an I_PUSH?  We need to know
4157 		 * this because we do some optimizations if this is a
4158 		 * module being pushed.
4159 		 */
4160 		pos = STRUCT_FGET(strmodinsert, pos);
4161 		is_insert = (pos != 0);
4162 
4163 		/*
4164 		 * Make sure pos is valid.  Even though it is not an I_PUSH,
4165 		 * we impose the same limit on the number of modules in a
4166 		 * stream.
4167 		 */
4168 		mutex_enter(&stp->sd_lock);
4169 		if (stp->sd_pushcnt >= nstrpush || pos < 0 ||
4170 		    pos > stp->sd_pushcnt) {
4171 			fmodsw_rele(fp);
4172 			strendplumb(stp);
4173 			mutex_exit(&stp->sd_lock);
4174 			return (EINVAL);
4175 		}
4176 		if (stp->sd_anchor != 0) {
4177 			/*
4178 			 * Is this insert below the anchor?
4179 			 * Pushcnt hasn't been increased yet hence
4180 			 * we test for greater than here, and greater or
4181 			 * equal after qattach.
4182 			 */
4183 			if (pos > (stp->sd_pushcnt - stp->sd_anchor) &&
4184 			    stp->sd_anchorzone != crgetzoneid(crp)) {
4185 				fmodsw_rele(fp);
4186 				strendplumb(stp);
4187 				mutex_exit(&stp->sd_lock);
4188 				return (EPERM);
4189 			}
4190 		}
4191 
4192 		mutex_exit(&stp->sd_lock);
4193 
4194 		/*
4195 		 * First find the correct position this module to
4196 		 * be inserted.  We don't need to call claimstr()
4197 		 * as the stream should not be changing at this point.
4198 		 *
4199 		 * Insert new module and call its open routine
4200 		 * via qattach().  Modules don't change device
4201 		 * numbers, so just ignore dummydev here.
4202 		 */
4203 		for (tmp_wrq = stp->sd_wrq; pos > 0;
4204 		    tmp_wrq = tmp_wrq->q_next, pos--) {
4205 			ASSERT(SAMESTR(tmp_wrq));
4206 		}
4207 		dummydev = vp->v_rdev;
4208 		if ((error = qattach(_RD(tmp_wrq), &dummydev, 0, crp,
4209 		    fp, is_insert)) != 0) {
4210 			mutex_enter(&stp->sd_lock);
4211 			strendplumb(stp);
4212 			mutex_exit(&stp->sd_lock);
4213 			return (error);
4214 		}
4215 
4216 		mutex_enter(&stp->sd_lock);
4217 
4218 		/*
4219 		 * As a performance concern we are caching the values of
4220 		 * q_minpsz and q_maxpsz of the module below the stream
4221 		 * head in the stream head.
4222 		 */
4223 		if (!is_insert) {
4224 			mutex_enter(QLOCK(stp->sd_wrq->q_next));
4225 			rmin = stp->sd_wrq->q_next->q_minpsz;
4226 			rmax = stp->sd_wrq->q_next->q_maxpsz;
4227 			mutex_exit(QLOCK(stp->sd_wrq->q_next));
4228 
4229 			/* Do this processing here as a performance concern */
4230 			if (strmsgsz != 0) {
4231 				if (rmax == INFPSZ) {
4232 					rmax = strmsgsz;
4233 				} else  {
4234 					rmax = MIN(strmsgsz, rmax);
4235 				}
4236 			}
4237 
4238 			mutex_enter(QLOCK(wrq));
4239 			stp->sd_qn_minpsz = rmin;
4240 			stp->sd_qn_maxpsz = rmax;
4241 			mutex_exit(QLOCK(wrq));
4242 		}
4243 
4244 		/*
4245 		 * Need to update the anchor value if this module is
4246 		 * inserted below the anchor point.
4247 		 */
4248 		if (stp->sd_anchor != 0) {
4249 			pos = STRUCT_FGET(strmodinsert, pos);
4250 			if (pos >= (stp->sd_pushcnt - stp->sd_anchor))
4251 				stp->sd_anchor++;
4252 		}
4253 
4254 		strendplumb(stp);
4255 		mutex_exit(&stp->sd_lock);
4256 		return (0);
4257 	}
4258 
4259 	case _I_REMOVE:
4260 	{
4261 		/*
4262 		 * To remove a module with a given name in a stream.  The
4263 		 * caller of this ioctl needs to provide both the name and
4264 		 * the position of the module to be removed.  This eliminates
4265 		 * the ambiguity of removal if a module is inserted/pushed
4266 		 * multiple times in a stream.  In the first release, only
4267 		 * allow privileged user to use this ioctl.
4268 		 * Furthermore, the remove is only allowed
4269 		 * below an anchor if the zoneid is the same as the zoneid
4270 		 * which created the anchor.
4271 		 *
4272 		 * Note that we do not plan to support this ioctl
4273 		 * on pipes in the first release.  We want to learn more
4274 		 * about the implications of these ioctls before extending
4275 		 * their support.  And we do not think these features are
4276 		 * valuable for pipes.
4277 		 *
4278 		 * Neither do we support O/C hot stream.  Note that only
4279 		 * the upper streams of TCP/IP stack are O/C hot streams.
4280 		 * The lower IP stream is not.
4281 		 * When there is a O/C cold barrier we do not allow removal
4282 		 * below the barrier.
4283 		 *
4284 		 * Also note that _I_REMOVE cannot be used to remove a
4285 		 * driver or the stream head.
4286 		 */
4287 		STRUCT_DECL(strmodconf, strmodremove);
4288 		queue_t	*q;
4289 		int pos;
4290 		char mod_name[FMNAMESZ + 1];
4291 		boolean_t is_remove;
4292 
4293 		STRUCT_INIT(strmodremove, flag);
4294 		if (stp->sd_flag & STRHUP)
4295 			return (ENXIO);
4296 		if (STRMATED(stp))
4297 			return (EINVAL);
4298 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4299 			return (error);
4300 		if (stp->sd_anchor != 0 &&
4301 		    stp->sd_anchorzone != crgetzoneid(crp))
4302 			return (EINVAL);
4303 
4304 		error = strcopyin((void *)arg, STRUCT_BUF(strmodremove),
4305 		    STRUCT_SIZE(strmodremove), copyflag);
4306 		if (error)
4307 			return (error);
4308 
4309 		error = (copyflag & U_TO_K ? copyinstr :
4310 		    copystr)(STRUCT_FGETP(strmodremove, mod_name),
4311 		    mod_name, FMNAMESZ + 1, NULL);
4312 		if (error)
4313 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4314 
4315 		if ((error = strstartplumb(stp, flag, cmd)) != 0)
4316 			return (error);
4317 
4318 		/*
4319 		 * Match the name of given module to the name of module at
4320 		 * the given position.
4321 		 */
4322 		pos = STRUCT_FGET(strmodremove, pos);
4323 
4324 		is_remove = (pos != 0);
4325 		for (q = stp->sd_wrq->q_next; SAMESTR(q) && pos > 0;
4326 		    q = q->q_next, pos--)
4327 			;
4328 		if (pos > 0 || ! SAMESTR(q) ||
4329 		    strncmp(q->q_qinfo->qi_minfo->mi_idname, mod_name,
4330 		    strlen(q->q_qinfo->qi_minfo->mi_idname)) != 0) {
4331 			mutex_enter(&stp->sd_lock);
4332 			strendplumb(stp);
4333 			mutex_exit(&stp->sd_lock);
4334 			return (EINVAL);
4335 		}
4336 
4337 		/*
4338 		 * If the position is at or below an anchor, then the zoneid
4339 		 * must match the zoneid that created the anchor.
4340 		 */
4341 		if (stp->sd_anchor != 0) {
4342 			pos = STRUCT_FGET(strmodremove, pos);
4343 			if (pos >= (stp->sd_pushcnt - stp->sd_anchor) &&
4344 			    stp->sd_anchorzone != crgetzoneid(crp)) {
4345 				mutex_enter(&stp->sd_lock);
4346 				strendplumb(stp);
4347 				mutex_exit(&stp->sd_lock);
4348 				return (EPERM);
4349 			}
4350 		}
4351 
4352 
4353 		ASSERT(!(q->q_flag & QREADR));
4354 		qdetach(_RD(q), 1, flag, crp, is_remove);
4355 
4356 		mutex_enter(&stp->sd_lock);
4357 
4358 		/*
4359 		 * As a performance concern we are caching the values of
4360 		 * q_minpsz and q_maxpsz of the module below the stream
4361 		 * head in the stream head.
4362 		 */
4363 		if (!is_remove) {
4364 			mutex_enter(QLOCK(wrq->q_next));
4365 			rmin = wrq->q_next->q_minpsz;
4366 			rmax = wrq->q_next->q_maxpsz;
4367 			mutex_exit(QLOCK(wrq->q_next));
4368 
4369 			/* Do this processing here as a performance concern */
4370 			if (strmsgsz != 0) {
4371 				if (rmax == INFPSZ)
4372 					rmax = strmsgsz;
4373 				else  {
4374 					if (vp->v_type == VFIFO)
4375 						rmax = MIN(PIPE_BUF, rmax);
4376 					else	rmax = MIN(strmsgsz, rmax);
4377 				}
4378 			}
4379 
4380 			mutex_enter(QLOCK(wrq));
4381 			stp->sd_qn_minpsz = rmin;
4382 			stp->sd_qn_maxpsz = rmax;
4383 			mutex_exit(QLOCK(wrq));
4384 		}
4385 
4386 		/*
4387 		 * Need to update the anchor value if this module is removed
4388 		 * at or below the anchor point.  If the removed module is at
4389 		 * the anchor point, remove the anchor for this stream if
4390 		 * there is no module above the anchor point.  Otherwise, if
4391 		 * the removed module is below the anchor point, decrement the
4392 		 * anchor point by 1.
4393 		 */
4394 		if (stp->sd_anchor != 0) {
4395 			pos = STRUCT_FGET(strmodremove, pos);
4396 			if (pos == stp->sd_pushcnt - stp->sd_anchor + 1)
4397 				stp->sd_anchor = 0;
4398 			else if (pos > (stp->sd_pushcnt - stp->sd_anchor + 1))
4399 				stp->sd_anchor--;
4400 		}
4401 
4402 		strendplumb(stp);
4403 		mutex_exit(&stp->sd_lock);
4404 		return (0);
4405 	}
4406 
4407 	case I_ANCHOR:
4408 		/*
4409 		 * Set the anchor position on the stream to reside at
4410 		 * the top module (in other words, the top module
4411 		 * cannot be popped).  Anchors with a FIFO make no
4412 		 * obvious sense, so they're not allowed.
4413 		 */
4414 		mutex_enter(&stp->sd_lock);
4415 
4416 		if (stp->sd_vnode->v_type == VFIFO) {
4417 			mutex_exit(&stp->sd_lock);
4418 			return (EINVAL);
4419 		}
4420 		/* Only allow the same zoneid to update the anchor */
4421 		if (stp->sd_anchor != 0 &&
4422 		    stp->sd_anchorzone != crgetzoneid(crp)) {
4423 			mutex_exit(&stp->sd_lock);
4424 			return (EINVAL);
4425 		}
4426 		stp->sd_anchor = stp->sd_pushcnt;
4427 		stp->sd_anchorzone = crgetzoneid(crp);
4428 		mutex_exit(&stp->sd_lock);
4429 		return (0);
4430 
4431 	case I_LOOK:
4432 		/*
4433 		 * Get name of first module downstream.
4434 		 * If no module, return an error.
4435 		 */
4436 	{
4437 		claimstr(wrq);
4438 		if (_SAMESTR(wrq) && wrq->q_next->q_next) {
4439 			char *name = wrq->q_next->q_qinfo->qi_minfo->mi_idname;
4440 			error = strcopyout(name, (void *)arg, strlen(name) + 1,
4441 			    copyflag);
4442 			releasestr(wrq);
4443 			return (error);
4444 		}
4445 		releasestr(wrq);
4446 		return (EINVAL);
4447 	}
4448 
4449 	case I_LINK:
4450 	case I_PLINK:
4451 		/*
4452 		 * Link a multiplexor.
4453 		 */
4454 		error = mlink(vp, cmd, (int)arg, crp, rvalp, 0);
4455 		return (error);
4456 
4457 	case _I_PLINK_LH:
4458 		/*
4459 		 * Link a multiplexor: Call must originate from kernel.
4460 		 */
4461 		if (kioctl)
4462 			return (ldi_mlink_lh(vp, cmd, arg, crp, rvalp));
4463 
4464 		return (EINVAL);
4465 	case I_UNLINK:
4466 	case I_PUNLINK:
4467 		/*
4468 		 * Unlink a multiplexor.
4469 		 * If arg is -1, unlink all links for which this is the
4470 		 * controlling stream.  Otherwise, arg is an index number
4471 		 * for a link to be removed.
4472 		 */
4473 	{
4474 		struct linkinfo *linkp;
4475 		int native_arg = (int)arg;
4476 		int type;
4477 		netstack_t *ns;
4478 		str_stack_t *ss;
4479 
4480 		TRACE_1(TR_FAC_STREAMS_FR,
4481 		    TR_I_UNLINK, "I_UNLINK/I_PUNLINK:%p", stp);
4482 		if (vp->v_type == VFIFO) {
4483 			return (EINVAL);
4484 		}
4485 		if (cmd == I_UNLINK)
4486 			type = LINKNORMAL;
4487 		else	/* I_PUNLINK */
4488 			type = LINKPERSIST;
4489 		if (native_arg == 0) {
4490 			return (EINVAL);
4491 		}
4492 		ns = netstack_find_by_cred(crp);
4493 		ASSERT(ns != NULL);
4494 		ss = ns->netstack_str;
4495 		ASSERT(ss != NULL);
4496 
4497 		if (native_arg == MUXID_ALL)
4498 			error = munlinkall(stp, type, crp, rvalp, ss);
4499 		else {
4500 			mutex_enter(&muxifier);
4501 			if (!(linkp = findlinks(stp, (int)arg, type, ss))) {
4502 				/* invalid user supplied index number */
4503 				mutex_exit(&muxifier);
4504 				netstack_rele(ss->ss_netstack);
4505 				return (EINVAL);
4506 			}
4507 			/* munlink drops the muxifier lock */
4508 			error = munlink(stp, linkp, type, crp, rvalp, ss);
4509 		}
4510 		netstack_rele(ss->ss_netstack);
4511 		return (error);
4512 	}
4513 
4514 	case I_FLUSH:
4515 		/*
4516 		 * send a flush message downstream
4517 		 * flush message can indicate
4518 		 * FLUSHR - flush read queue
4519 		 * FLUSHW - flush write queue
4520 		 * FLUSHRW - flush read/write queue
4521 		 */
4522 		if (stp->sd_flag & STRHUP)
4523 			return (ENXIO);
4524 		if (arg & ~FLUSHRW)
4525 			return (EINVAL);
4526 
4527 		for (;;) {
4528 			if (putnextctl1(stp->sd_wrq, M_FLUSH, (int)arg)) {
4529 				break;
4530 			}
4531 			if (error = strwaitbuf(1, BPRI_HI)) {
4532 				return (error);
4533 			}
4534 		}
4535 
4536 		/*
4537 		 * Send down an unsupported ioctl and wait for the nack
4538 		 * in order to allow the M_FLUSH to propagate back
4539 		 * up to the stream head.
4540 		 * Replaces if (qready()) runqueues();
4541 		 */
4542 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4543 		strioc.ic_timout = 0;
4544 		strioc.ic_len = 0;
4545 		strioc.ic_dp = NULL;
4546 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4547 		*rvalp = 0;
4548 		return (0);
4549 
4550 	case I_FLUSHBAND:
4551 	{
4552 		struct bandinfo binfo;
4553 
4554 		error = strcopyin((void *)arg, &binfo, sizeof (binfo),
4555 		    copyflag);
4556 		if (error)
4557 			return (error);
4558 		if (stp->sd_flag & STRHUP)
4559 			return (ENXIO);
4560 		if (binfo.bi_flag & ~FLUSHRW)
4561 			return (EINVAL);
4562 		while (!(mp = allocb(2, BPRI_HI))) {
4563 			if (error = strwaitbuf(2, BPRI_HI))
4564 				return (error);
4565 		}
4566 		mp->b_datap->db_type = M_FLUSH;
4567 		*mp->b_wptr++ = binfo.bi_flag | FLUSHBAND;
4568 		*mp->b_wptr++ = binfo.bi_pri;
4569 		putnext(stp->sd_wrq, mp);
4570 		/*
4571 		 * Send down an unsupported ioctl and wait for the nack
4572 		 * in order to allow the M_FLUSH to propagate back
4573 		 * up to the stream head.
4574 		 * Replaces if (qready()) runqueues();
4575 		 */
4576 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4577 		strioc.ic_timout = 0;
4578 		strioc.ic_len = 0;
4579 		strioc.ic_dp = NULL;
4580 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4581 		*rvalp = 0;
4582 		return (0);
4583 	}
4584 
4585 	case I_SRDOPT:
4586 		/*
4587 		 * Set read options
4588 		 *
4589 		 * RNORM - default stream mode
4590 		 * RMSGN - message no discard
4591 		 * RMSGD - message discard
4592 		 * RPROTNORM - fail read with EBADMSG for M_[PC]PROTOs
4593 		 * RPROTDAT - convert M_[PC]PROTOs to M_DATAs
4594 		 * RPROTDIS - discard M_[PC]PROTOs and retain M_DATAs
4595 		 */
4596 		if (arg & ~(RMODEMASK | RPROTMASK))
4597 			return (EINVAL);
4598 
4599 		if ((arg & (RMSGD|RMSGN)) == (RMSGD|RMSGN))
4600 			return (EINVAL);
4601 
4602 		mutex_enter(&stp->sd_lock);
4603 		switch (arg & RMODEMASK) {
4604 		case RNORM:
4605 			stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
4606 			break;
4607 		case RMSGD:
4608 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGNODIS) |
4609 			    RD_MSGDIS;
4610 			break;
4611 		case RMSGN:
4612 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGDIS) |
4613 			    RD_MSGNODIS;
4614 			break;
4615 		}
4616 
4617 		switch (arg & RPROTMASK) {
4618 		case RPROTNORM:
4619 			stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
4620 			break;
4621 
4622 		case RPROTDAT:
4623 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDIS) |
4624 			    RD_PROTDAT);
4625 			break;
4626 
4627 		case RPROTDIS:
4628 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDAT) |
4629 			    RD_PROTDIS);
4630 			break;
4631 		}
4632 		mutex_exit(&stp->sd_lock);
4633 		return (0);
4634 
4635 	case I_GRDOPT:
4636 		/*
4637 		 * Get read option and return the value
4638 		 * to spot pointed to by arg
4639 		 */
4640 	{
4641 		int rdopt;
4642 
4643 		rdopt = ((stp->sd_read_opt & RD_MSGDIS) ? RMSGD :
4644 		    ((stp->sd_read_opt & RD_MSGNODIS) ? RMSGN : RNORM));
4645 		rdopt |= ((stp->sd_read_opt & RD_PROTDAT) ? RPROTDAT :
4646 		    ((stp->sd_read_opt & RD_PROTDIS) ? RPROTDIS : RPROTNORM));
4647 
4648 		return (strcopyout(&rdopt, (void *)arg, sizeof (int),
4649 		    copyflag));
4650 	}
4651 
4652 	case I_SERROPT:
4653 		/*
4654 		 * Set error options
4655 		 *
4656 		 * RERRNORM - persistent read errors
4657 		 * RERRNONPERSIST - non-persistent read errors
4658 		 * WERRNORM - persistent write errors
4659 		 * WERRNONPERSIST - non-persistent write errors
4660 		 */
4661 		if (arg & ~(RERRMASK | WERRMASK))
4662 			return (EINVAL);
4663 
4664 		mutex_enter(&stp->sd_lock);
4665 		switch (arg & RERRMASK) {
4666 		case RERRNORM:
4667 			stp->sd_flag &= ~STRDERRNONPERSIST;
4668 			break;
4669 		case RERRNONPERSIST:
4670 			stp->sd_flag |= STRDERRNONPERSIST;
4671 			break;
4672 		}
4673 		switch (arg & WERRMASK) {
4674 		case WERRNORM:
4675 			stp->sd_flag &= ~STWRERRNONPERSIST;
4676 			break;
4677 		case WERRNONPERSIST:
4678 			stp->sd_flag |= STWRERRNONPERSIST;
4679 			break;
4680 		}
4681 		mutex_exit(&stp->sd_lock);
4682 		return (0);
4683 
4684 	case I_GERROPT:
4685 		/*
4686 		 * Get error option and return the value
4687 		 * to spot pointed to by arg
4688 		 */
4689 	{
4690 		int erropt = 0;
4691 
4692 		erropt |= (stp->sd_flag & STRDERRNONPERSIST) ? RERRNONPERSIST :
4693 		    RERRNORM;
4694 		erropt |= (stp->sd_flag & STWRERRNONPERSIST) ? WERRNONPERSIST :
4695 		    WERRNORM;
4696 		return (strcopyout(&erropt, (void *)arg, sizeof (int),
4697 		    copyflag));
4698 	}
4699 
4700 	case I_SETSIG:
4701 		/*
4702 		 * Register the calling proc to receive the SIGPOLL
4703 		 * signal based on the events given in arg.  If
4704 		 * arg is zero, remove the proc from register list.
4705 		 */
4706 	{
4707 		strsig_t *ssp, *pssp;
4708 		struct pid *pidp;
4709 
4710 		pssp = NULL;
4711 		pidp = curproc->p_pidp;
4712 		/*
4713 		 * Hold sd_lock to prevent traversal of sd_siglist while
4714 		 * it is modified.
4715 		 */
4716 		mutex_enter(&stp->sd_lock);
4717 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pidp != pidp);
4718 		    pssp = ssp, ssp = ssp->ss_next)
4719 			;
4720 
4721 		if (arg) {
4722 			if (arg & ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4723 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4724 				mutex_exit(&stp->sd_lock);
4725 				return (EINVAL);
4726 			}
4727 			if ((arg & S_BANDURG) && !(arg & S_RDBAND)) {
4728 				mutex_exit(&stp->sd_lock);
4729 				return (EINVAL);
4730 			}
4731 
4732 			/*
4733 			 * If proc not already registered, add it
4734 			 * to list.
4735 			 */
4736 			if (!ssp) {
4737 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4738 				ssp->ss_pidp = pidp;
4739 				ssp->ss_pid = pidp->pid_id;
4740 				ssp->ss_next = NULL;
4741 				if (pssp)
4742 					pssp->ss_next = ssp;
4743 				else
4744 					stp->sd_siglist = ssp;
4745 				mutex_enter(&pidlock);
4746 				PID_HOLD(pidp);
4747 				mutex_exit(&pidlock);
4748 			}
4749 
4750 			/*
4751 			 * Set events.
4752 			 */
4753 			ssp->ss_events = (int)arg;
4754 		} else {
4755 			/*
4756 			 * Remove proc from register list.
4757 			 */
4758 			if (ssp) {
4759 				mutex_enter(&pidlock);
4760 				PID_RELE(pidp);
4761 				mutex_exit(&pidlock);
4762 				if (pssp)
4763 					pssp->ss_next = ssp->ss_next;
4764 				else
4765 					stp->sd_siglist = ssp->ss_next;
4766 				kmem_free(ssp, sizeof (strsig_t));
4767 			} else {
4768 				mutex_exit(&stp->sd_lock);
4769 				return (EINVAL);
4770 			}
4771 		}
4772 
4773 		/*
4774 		 * Recalculate OR of sig events.
4775 		 */
4776 		stp->sd_sigflags = 0;
4777 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4778 			stp->sd_sigflags |= ssp->ss_events;
4779 		mutex_exit(&stp->sd_lock);
4780 		return (0);
4781 	}
4782 
4783 	case I_GETSIG:
4784 		/*
4785 		 * Return (in arg) the current registration of events
4786 		 * for which the calling proc is to be signaled.
4787 		 */
4788 	{
4789 		struct strsig *ssp;
4790 		struct pid  *pidp;
4791 
4792 		pidp = curproc->p_pidp;
4793 		mutex_enter(&stp->sd_lock);
4794 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4795 			if (ssp->ss_pidp == pidp) {
4796 				error = strcopyout(&ssp->ss_events, (void *)arg,
4797 				    sizeof (int), copyflag);
4798 				mutex_exit(&stp->sd_lock);
4799 				return (error);
4800 			}
4801 		mutex_exit(&stp->sd_lock);
4802 		return (EINVAL);
4803 	}
4804 
4805 	case I_ESETSIG:
4806 		/*
4807 		 * Register the ss_pid to receive the SIGPOLL
4808 		 * signal based on the events is ss_events arg.  If
4809 		 * ss_events is zero, remove the proc from register list.
4810 		 */
4811 	{
4812 		struct strsig *ssp, *pssp;
4813 		struct proc *proc;
4814 		struct pid  *pidp;
4815 		pid_t pid;
4816 		struct strsigset ss;
4817 
4818 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4819 		if (error)
4820 			return (error);
4821 
4822 		pid = ss.ss_pid;
4823 
4824 		if (ss.ss_events != 0) {
4825 			/*
4826 			 * Permissions check by sending signal 0.
4827 			 * Note that when kill fails it does a set_errno
4828 			 * causing the system call to fail.
4829 			 */
4830 			error = kill(pid, 0);
4831 			if (error) {
4832 				return (error);
4833 			}
4834 		}
4835 		mutex_enter(&pidlock);
4836 		if (pid == 0)
4837 			proc = curproc;
4838 		else if (pid < 0)
4839 			proc = pgfind(-pid);
4840 		else
4841 			proc = prfind(pid);
4842 		if (proc == NULL) {
4843 			mutex_exit(&pidlock);
4844 			return (ESRCH);
4845 		}
4846 		if (pid < 0)
4847 			pidp = proc->p_pgidp;
4848 		else
4849 			pidp = proc->p_pidp;
4850 		ASSERT(pidp);
4851 		/*
4852 		 * Get a hold on the pid structure while referencing it.
4853 		 * There is a separate PID_HOLD should it be inserted
4854 		 * in the list below.
4855 		 */
4856 		PID_HOLD(pidp);
4857 		mutex_exit(&pidlock);
4858 
4859 		pssp = NULL;
4860 		/*
4861 		 * Hold sd_lock to prevent traversal of sd_siglist while
4862 		 * it is modified.
4863 		 */
4864 		mutex_enter(&stp->sd_lock);
4865 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pid != pid);
4866 		    pssp = ssp, ssp = ssp->ss_next)
4867 			;
4868 
4869 		if (ss.ss_events) {
4870 			if (ss.ss_events &
4871 			    ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4872 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4873 				mutex_exit(&stp->sd_lock);
4874 				mutex_enter(&pidlock);
4875 				PID_RELE(pidp);
4876 				mutex_exit(&pidlock);
4877 				return (EINVAL);
4878 			}
4879 			if ((ss.ss_events & S_BANDURG) &&
4880 			    !(ss.ss_events & S_RDBAND)) {
4881 				mutex_exit(&stp->sd_lock);
4882 				mutex_enter(&pidlock);
4883 				PID_RELE(pidp);
4884 				mutex_exit(&pidlock);
4885 				return (EINVAL);
4886 			}
4887 
4888 			/*
4889 			 * If proc not already registered, add it
4890 			 * to list.
4891 			 */
4892 			if (!ssp) {
4893 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4894 				ssp->ss_pidp = pidp;
4895 				ssp->ss_pid = pid;
4896 				ssp->ss_next = NULL;
4897 				if (pssp)
4898 					pssp->ss_next = ssp;
4899 				else
4900 					stp->sd_siglist = ssp;
4901 				mutex_enter(&pidlock);
4902 				PID_HOLD(pidp);
4903 				mutex_exit(&pidlock);
4904 			}
4905 
4906 			/*
4907 			 * Set events.
4908 			 */
4909 			ssp->ss_events = ss.ss_events;
4910 		} else {
4911 			/*
4912 			 * Remove proc from register list.
4913 			 */
4914 			if (ssp) {
4915 				mutex_enter(&pidlock);
4916 				PID_RELE(pidp);
4917 				mutex_exit(&pidlock);
4918 				if (pssp)
4919 					pssp->ss_next = ssp->ss_next;
4920 				else
4921 					stp->sd_siglist = ssp->ss_next;
4922 				kmem_free(ssp, sizeof (strsig_t));
4923 			} else {
4924 				mutex_exit(&stp->sd_lock);
4925 				mutex_enter(&pidlock);
4926 				PID_RELE(pidp);
4927 				mutex_exit(&pidlock);
4928 				return (EINVAL);
4929 			}
4930 		}
4931 
4932 		/*
4933 		 * Recalculate OR of sig events.
4934 		 */
4935 		stp->sd_sigflags = 0;
4936 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4937 			stp->sd_sigflags |= ssp->ss_events;
4938 		mutex_exit(&stp->sd_lock);
4939 		mutex_enter(&pidlock);
4940 		PID_RELE(pidp);
4941 		mutex_exit(&pidlock);
4942 		return (0);
4943 	}
4944 
4945 	case I_EGETSIG:
4946 		/*
4947 		 * Return (in arg) the current registration of events
4948 		 * for which the calling proc is to be signaled.
4949 		 */
4950 	{
4951 		struct strsig *ssp;
4952 		struct proc *proc;
4953 		pid_t pid;
4954 		struct pid  *pidp;
4955 		struct strsigset ss;
4956 
4957 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4958 		if (error)
4959 			return (error);
4960 
4961 		pid = ss.ss_pid;
4962 		mutex_enter(&pidlock);
4963 		if (pid == 0)
4964 			proc = curproc;
4965 		else if (pid < 0)
4966 			proc = pgfind(-pid);
4967 		else
4968 			proc = prfind(pid);
4969 		if (proc == NULL) {
4970 			mutex_exit(&pidlock);
4971 			return (ESRCH);
4972 		}
4973 		if (pid < 0)
4974 			pidp = proc->p_pgidp;
4975 		else
4976 			pidp = proc->p_pidp;
4977 
4978 		/* Prevent the pidp from being reassigned */
4979 		PID_HOLD(pidp);
4980 		mutex_exit(&pidlock);
4981 
4982 		mutex_enter(&stp->sd_lock);
4983 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4984 			if (ssp->ss_pid == pid) {
4985 				ss.ss_pid = ssp->ss_pid;
4986 				ss.ss_events = ssp->ss_events;
4987 				error = strcopyout(&ss, (void *)arg,
4988 				    sizeof (struct strsigset), copyflag);
4989 				mutex_exit(&stp->sd_lock);
4990 				mutex_enter(&pidlock);
4991 				PID_RELE(pidp);
4992 				mutex_exit(&pidlock);
4993 				return (error);
4994 			}
4995 		mutex_exit(&stp->sd_lock);
4996 		mutex_enter(&pidlock);
4997 		PID_RELE(pidp);
4998 		mutex_exit(&pidlock);
4999 		return (EINVAL);
5000 	}
5001 
5002 	case I_PEEK:
5003 	{
5004 		STRUCT_DECL(strpeek, strpeek);
5005 		size_t n;
5006 		mblk_t *fmp, *tmp_mp = NULL;
5007 
5008 		STRUCT_INIT(strpeek, flag);
5009 
5010 		error = strcopyin((void *)arg, STRUCT_BUF(strpeek),
5011 		    STRUCT_SIZE(strpeek), copyflag);
5012 		if (error)
5013 			return (error);
5014 
5015 		mutex_enter(QLOCK(rdq));
5016 		/*
5017 		 * Skip the invalid messages
5018 		 */
5019 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
5020 			if (mp->b_datap->db_type != M_SIG)
5021 				break;
5022 
5023 		/*
5024 		 * If user has requested to peek at a high priority message
5025 		 * and first message is not, return 0
5026 		 */
5027 		if (mp != NULL) {
5028 			if ((STRUCT_FGET(strpeek, flags) & RS_HIPRI) &&
5029 			    queclass(mp) == QNORM) {
5030 				*rvalp = 0;
5031 				mutex_exit(QLOCK(rdq));
5032 				return (0);
5033 			}
5034 		} else if (stp->sd_struiordq == NULL ||
5035 		    (STRUCT_FGET(strpeek, flags) & RS_HIPRI)) {
5036 			/*
5037 			 * No mblks to look at at the streamhead and
5038 			 * 1). This isn't a synch stream or
5039 			 * 2). This is a synch stream but caller wants high
5040 			 *	priority messages which is not supported by
5041 			 *	the synch stream. (it only supports QNORM)
5042 			 */
5043 			*rvalp = 0;
5044 			mutex_exit(QLOCK(rdq));
5045 			return (0);
5046 		}
5047 
5048 		fmp = mp;
5049 
5050 		if (mp && mp->b_datap->db_type == M_PASSFP) {
5051 			mutex_exit(QLOCK(rdq));
5052 			return (EBADMSG);
5053 		}
5054 
5055 		ASSERT(mp == NULL || mp->b_datap->db_type == M_PCPROTO ||
5056 		    mp->b_datap->db_type == M_PROTO ||
5057 		    mp->b_datap->db_type == M_DATA);
5058 
5059 		if (mp && mp->b_datap->db_type == M_PCPROTO) {
5060 			STRUCT_FSET(strpeek, flags, RS_HIPRI);
5061 		} else {
5062 			STRUCT_FSET(strpeek, flags, 0);
5063 		}
5064 
5065 
5066 		if (mp && ((tmp_mp = dupmsg(mp)) == NULL)) {
5067 			mutex_exit(QLOCK(rdq));
5068 			return (ENOSR);
5069 		}
5070 		mutex_exit(QLOCK(rdq));
5071 
5072 		/*
5073 		 * set mp = tmp_mp, so that I_PEEK processing can continue.
5074 		 * tmp_mp is used to free the dup'd message.
5075 		 */
5076 		mp = tmp_mp;
5077 
5078 		uio.uio_fmode = 0;
5079 		uio.uio_extflg = UIO_COPY_CACHED;
5080 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5081 		    UIO_SYSSPACE;
5082 		uio.uio_limit = 0;
5083 		/*
5084 		 * First process PROTO blocks, if any.
5085 		 * If user doesn't want to get ctl info by setting maxlen <= 0,
5086 		 * then set len to -1/0 and skip control blocks part.
5087 		 */
5088 		if (STRUCT_FGET(strpeek, ctlbuf.maxlen) < 0)
5089 			STRUCT_FSET(strpeek, ctlbuf.len, -1);
5090 		else if (STRUCT_FGET(strpeek, ctlbuf.maxlen) == 0)
5091 			STRUCT_FSET(strpeek, ctlbuf.len, 0);
5092 		else {
5093 			int	ctl_part = 0;
5094 
5095 			iov.iov_base = STRUCT_FGETP(strpeek, ctlbuf.buf);
5096 			iov.iov_len = STRUCT_FGET(strpeek, ctlbuf.maxlen);
5097 			uio.uio_iov = &iov;
5098 			uio.uio_resid = iov.iov_len;
5099 			uio.uio_loffset = 0;
5100 			uio.uio_iovcnt = 1;
5101 			while (mp && mp->b_datap->db_type != M_DATA &&
5102 			    uio.uio_resid >= 0) {
5103 				ASSERT(STRUCT_FGET(strpeek, flags) == 0 ?
5104 				    mp->b_datap->db_type == M_PROTO :
5105 				    mp->b_datap->db_type == M_PCPROTO);
5106 
5107 				if ((n = MIN(uio.uio_resid,
5108 				    mp->b_wptr - mp->b_rptr)) != 0 &&
5109 				    (error = uiomove((char *)mp->b_rptr, n,
5110 				    UIO_READ, &uio)) != 0) {
5111 					freemsg(tmp_mp);
5112 					return (error);
5113 				}
5114 				ctl_part = 1;
5115 				mp = mp->b_cont;
5116 			}
5117 			/* No ctl message */
5118 			if (ctl_part == 0)
5119 				STRUCT_FSET(strpeek, ctlbuf.len, -1);
5120 			else
5121 				STRUCT_FSET(strpeek, ctlbuf.len,
5122 				    STRUCT_FGET(strpeek, ctlbuf.maxlen) -
5123 				    uio.uio_resid);
5124 		}
5125 
5126 		/*
5127 		 * Now process DATA blocks, if any.
5128 		 * If user doesn't want to get data info by setting maxlen <= 0,
5129 		 * then set len to -1/0 and skip data blocks part.
5130 		 */
5131 		if (STRUCT_FGET(strpeek, databuf.maxlen) < 0)
5132 			STRUCT_FSET(strpeek, databuf.len, -1);
5133 		else if (STRUCT_FGET(strpeek, databuf.maxlen) == 0)
5134 			STRUCT_FSET(strpeek, databuf.len, 0);
5135 		else {
5136 			int	data_part = 0;
5137 
5138 			iov.iov_base = STRUCT_FGETP(strpeek, databuf.buf);
5139 			iov.iov_len = STRUCT_FGET(strpeek, databuf.maxlen);
5140 			uio.uio_iov = &iov;
5141 			uio.uio_resid = iov.iov_len;
5142 			uio.uio_loffset = 0;
5143 			uio.uio_iovcnt = 1;
5144 			while (mp && uio.uio_resid) {
5145 				if (mp->b_datap->db_type == M_DATA) {
5146 					if ((n = MIN(uio.uio_resid,
5147 					    mp->b_wptr - mp->b_rptr)) != 0 &&
5148 					    (error = uiomove((char *)mp->b_rptr,
5149 					    n, UIO_READ, &uio)) != 0) {
5150 						freemsg(tmp_mp);
5151 						return (error);
5152 					}
5153 					data_part = 1;
5154 				}
5155 				ASSERT(data_part == 0 ||
5156 				    mp->b_datap->db_type == M_DATA);
5157 				mp = mp->b_cont;
5158 			}
5159 			/* No data message */
5160 			if (data_part == 0)
5161 				STRUCT_FSET(strpeek, databuf.len, -1);
5162 			else
5163 				STRUCT_FSET(strpeek, databuf.len,
5164 				    STRUCT_FGET(strpeek, databuf.maxlen) -
5165 				    uio.uio_resid);
5166 		}
5167 		freemsg(tmp_mp);
5168 
5169 		/*
5170 		 * It is a synch stream and user wants to get
5171 		 * data (maxlen > 0).
5172 		 * uio setup is done by the codes that process DATA
5173 		 * blocks above.
5174 		 */
5175 		if ((fmp == NULL) && STRUCT_FGET(strpeek, databuf.maxlen) > 0) {
5176 			infod_t infod;
5177 
5178 			infod.d_cmd = INFOD_COPYOUT;
5179 			infod.d_res = 0;
5180 			infod.d_uiop = &uio;
5181 			error = infonext(rdq, &infod);
5182 			if (error == EINVAL || error == EBUSY)
5183 				error = 0;
5184 			if (error)
5185 				return (error);
5186 			STRUCT_FSET(strpeek, databuf.len, STRUCT_FGET(strpeek,
5187 			    databuf.maxlen) - uio.uio_resid);
5188 			if (STRUCT_FGET(strpeek, databuf.len) == 0) {
5189 				/*
5190 				 * No data found by the infonext().
5191 				 */
5192 				STRUCT_FSET(strpeek, databuf.len, -1);
5193 			}
5194 		}
5195 		error = strcopyout(STRUCT_BUF(strpeek), (void *)arg,
5196 		    STRUCT_SIZE(strpeek), copyflag);
5197 		if (error) {
5198 			return (error);
5199 		}
5200 		/*
5201 		 * If there is no message retrieved, set return code to 0
5202 		 * otherwise, set it to 1.
5203 		 */
5204 		if (STRUCT_FGET(strpeek, ctlbuf.len) == -1 &&
5205 		    STRUCT_FGET(strpeek, databuf.len) == -1)
5206 			*rvalp = 0;
5207 		else
5208 			*rvalp = 1;
5209 		return (0);
5210 	}
5211 
5212 	case I_FDINSERT:
5213 	{
5214 		STRUCT_DECL(strfdinsert, strfdinsert);
5215 		struct file *resftp;
5216 		struct stdata *resstp;
5217 		t_uscalar_t	ival;
5218 		ssize_t msgsize;
5219 		struct strbuf mctl;
5220 
5221 		STRUCT_INIT(strfdinsert, flag);
5222 		if (stp->sd_flag & STRHUP)
5223 			return (ENXIO);
5224 		/*
5225 		 * STRDERR, STWRERR and STPLEX tested above.
5226 		 */
5227 		error = strcopyin((void *)arg, STRUCT_BUF(strfdinsert),
5228 		    STRUCT_SIZE(strfdinsert), copyflag);
5229 		if (error)
5230 			return (error);
5231 
5232 		if (STRUCT_FGET(strfdinsert, offset) < 0 ||
5233 		    (STRUCT_FGET(strfdinsert, offset) %
5234 		    sizeof (t_uscalar_t)) != 0)
5235 			return (EINVAL);
5236 		if ((resftp = getf(STRUCT_FGET(strfdinsert, fildes))) != NULL) {
5237 			if ((resstp = resftp->f_vnode->v_stream) == NULL) {
5238 				releasef(STRUCT_FGET(strfdinsert, fildes));
5239 				return (EINVAL);
5240 			}
5241 		} else
5242 			return (EINVAL);
5243 
5244 		mutex_enter(&resstp->sd_lock);
5245 		if (resstp->sd_flag & (STRDERR|STWRERR|STRHUP|STPLEX)) {
5246 			error = strgeterr(resstp,
5247 			    STRDERR|STWRERR|STRHUP|STPLEX, 0);
5248 			if (error != 0) {
5249 				mutex_exit(&resstp->sd_lock);
5250 				releasef(STRUCT_FGET(strfdinsert, fildes));
5251 				return (error);
5252 			}
5253 		}
5254 		mutex_exit(&resstp->sd_lock);
5255 
5256 #ifdef	_ILP32
5257 		{
5258 			queue_t	*q;
5259 			queue_t	*mate = NULL;
5260 
5261 			/* get read queue of stream terminus */
5262 			claimstr(resstp->sd_wrq);
5263 			for (q = resstp->sd_wrq->q_next; q->q_next != NULL;
5264 			    q = q->q_next)
5265 				if (!STRMATED(resstp) && STREAM(q) != resstp &&
5266 				    mate == NULL) {
5267 					ASSERT(q->q_qinfo->qi_srvp);
5268 					ASSERT(_OTHERQ(q)->q_qinfo->qi_srvp);
5269 					claimstr(q);
5270 					mate = q;
5271 				}
5272 			q = _RD(q);
5273 			if (mate)
5274 				releasestr(mate);
5275 			releasestr(resstp->sd_wrq);
5276 			ival = (t_uscalar_t)q;
5277 		}
5278 #else
5279 		ival = (t_uscalar_t)getminor(resftp->f_vnode->v_rdev);
5280 #endif	/* _ILP32 */
5281 
5282 		if (STRUCT_FGET(strfdinsert, ctlbuf.len) <
5283 		    STRUCT_FGET(strfdinsert, offset) + sizeof (t_uscalar_t)) {
5284 			releasef(STRUCT_FGET(strfdinsert, fildes));
5285 			return (EINVAL);
5286 		}
5287 
5288 		/*
5289 		 * Check for legal flag value.
5290 		 */
5291 		if (STRUCT_FGET(strfdinsert, flags) & ~RS_HIPRI) {
5292 			releasef(STRUCT_FGET(strfdinsert, fildes));
5293 			return (EINVAL);
5294 		}
5295 
5296 		/* get these values from those cached in the stream head */
5297 		mutex_enter(QLOCK(stp->sd_wrq));
5298 		rmin = stp->sd_qn_minpsz;
5299 		rmax = stp->sd_qn_maxpsz;
5300 		mutex_exit(QLOCK(stp->sd_wrq));
5301 
5302 		/*
5303 		 * Make sure ctl and data sizes together fall within
5304 		 * the limits of the max and min receive packet sizes
5305 		 * and do not exceed system limit.  A negative data
5306 		 * length means that no data part is to be sent.
5307 		 */
5308 		ASSERT((rmax >= 0) || (rmax == INFPSZ));
5309 		if (rmax == 0) {
5310 			releasef(STRUCT_FGET(strfdinsert, fildes));
5311 			return (ERANGE);
5312 		}
5313 		if ((msgsize = STRUCT_FGET(strfdinsert, databuf.len)) < 0)
5314 			msgsize = 0;
5315 		if ((msgsize < rmin) ||
5316 		    ((msgsize > rmax) && (rmax != INFPSZ)) ||
5317 		    (STRUCT_FGET(strfdinsert, ctlbuf.len) > strctlsz)) {
5318 			releasef(STRUCT_FGET(strfdinsert, fildes));
5319 			return (ERANGE);
5320 		}
5321 
5322 		mutex_enter(&stp->sd_lock);
5323 		while (!(STRUCT_FGET(strfdinsert, flags) & RS_HIPRI) &&
5324 		    !canputnext(stp->sd_wrq)) {
5325 			if ((error = strwaitq(stp, WRITEWAIT, (ssize_t)0,
5326 			    flag, -1, &done)) != 0 || done) {
5327 				mutex_exit(&stp->sd_lock);
5328 				releasef(STRUCT_FGET(strfdinsert, fildes));
5329 				return (error);
5330 			}
5331 			if ((error = i_straccess(stp, access)) != 0) {
5332 				mutex_exit(&stp->sd_lock);
5333 				releasef(
5334 				    STRUCT_FGET(strfdinsert, fildes));
5335 				return (error);
5336 			}
5337 		}
5338 		mutex_exit(&stp->sd_lock);
5339 
5340 		/*
5341 		 * Copy strfdinsert.ctlbuf into native form of
5342 		 * ctlbuf to pass down into strmakemsg().
5343 		 */
5344 		mctl.maxlen = STRUCT_FGET(strfdinsert, ctlbuf.maxlen);
5345 		mctl.len = STRUCT_FGET(strfdinsert, ctlbuf.len);
5346 		mctl.buf = STRUCT_FGETP(strfdinsert, ctlbuf.buf);
5347 
5348 		iov.iov_base = STRUCT_FGETP(strfdinsert, databuf.buf);
5349 		iov.iov_len = STRUCT_FGET(strfdinsert, databuf.len);
5350 		uio.uio_iov = &iov;
5351 		uio.uio_iovcnt = 1;
5352 		uio.uio_loffset = 0;
5353 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5354 		    UIO_SYSSPACE;
5355 		uio.uio_fmode = 0;
5356 		uio.uio_extflg = UIO_COPY_CACHED;
5357 		uio.uio_resid = iov.iov_len;
5358 		if ((error = strmakemsg(&mctl,
5359 		    &msgsize, &uio, stp,
5360 		    STRUCT_FGET(strfdinsert, flags), &mp)) != 0 || !mp) {
5361 			STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5362 			releasef(STRUCT_FGET(strfdinsert, fildes));
5363 			return (error);
5364 		}
5365 
5366 		STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5367 
5368 		/*
5369 		 * Place the possibly reencoded queue pointer 'offset' bytes
5370 		 * from the start of the control portion of the message.
5371 		 */
5372 		*((t_uscalar_t *)(mp->b_rptr +
5373 		    STRUCT_FGET(strfdinsert, offset))) = ival;
5374 
5375 		/*
5376 		 * Put message downstream.
5377 		 */
5378 		stream_willservice(stp);
5379 		putnext(stp->sd_wrq, mp);
5380 		stream_runservice(stp);
5381 		releasef(STRUCT_FGET(strfdinsert, fildes));
5382 		return (error);
5383 	}
5384 
5385 	case I_SENDFD:
5386 	{
5387 		struct file *fp;
5388 
5389 		if ((fp = getf((int)arg)) == NULL)
5390 			return (EBADF);
5391 		error = do_sendfp(stp, fp, crp);
5392 		if (audit_active) {
5393 			audit_fdsend((int)arg, fp, error);
5394 		}
5395 		releasef((int)arg);
5396 		return (error);
5397 	}
5398 
5399 	case I_RECVFD:
5400 	case I_E_RECVFD:
5401 	{
5402 		struct k_strrecvfd *srf;
5403 		int i, fd;
5404 
5405 		mutex_enter(&stp->sd_lock);
5406 		while (!(mp = getq(rdq))) {
5407 			if (stp->sd_flag & (STRHUP|STREOF)) {
5408 				mutex_exit(&stp->sd_lock);
5409 				return (ENXIO);
5410 			}
5411 			if ((error = strwaitq(stp, GETWAIT, (ssize_t)0,
5412 			    flag, -1, &done)) != 0 || done) {
5413 				mutex_exit(&stp->sd_lock);
5414 				return (error);
5415 			}
5416 			if ((error = i_straccess(stp, access)) != 0) {
5417 				mutex_exit(&stp->sd_lock);
5418 				return (error);
5419 			}
5420 		}
5421 		if (mp->b_datap->db_type != M_PASSFP) {
5422 			putback(stp, rdq, mp, mp->b_band);
5423 			mutex_exit(&stp->sd_lock);
5424 			return (EBADMSG);
5425 		}
5426 		mutex_exit(&stp->sd_lock);
5427 
5428 		srf = (struct k_strrecvfd *)mp->b_rptr;
5429 		if ((fd = ufalloc(0)) == -1) {
5430 			mutex_enter(&stp->sd_lock);
5431 			putback(stp, rdq, mp, mp->b_band);
5432 			mutex_exit(&stp->sd_lock);
5433 			return (EMFILE);
5434 		}
5435 		if (cmd == I_RECVFD) {
5436 			struct o_strrecvfd	ostrfd;
5437 
5438 			/* check to see if uid/gid values are too large. */
5439 
5440 			if (srf->uid > (o_uid_t)USHRT_MAX ||
5441 			    srf->gid > (o_gid_t)USHRT_MAX) {
5442 				mutex_enter(&stp->sd_lock);
5443 				putback(stp, rdq, mp, mp->b_band);
5444 				mutex_exit(&stp->sd_lock);
5445 				setf(fd, NULL);	/* release fd entry */
5446 				return (EOVERFLOW);
5447 			}
5448 
5449 			ostrfd.fd = fd;
5450 			ostrfd.uid = (o_uid_t)srf->uid;
5451 			ostrfd.gid = (o_gid_t)srf->gid;
5452 
5453 			/* Null the filler bits */
5454 			for (i = 0; i < 8; i++)
5455 				ostrfd.fill[i] = 0;
5456 
5457 			error = strcopyout(&ostrfd, (void *)arg,
5458 			    sizeof (struct o_strrecvfd), copyflag);
5459 		} else {		/* I_E_RECVFD */
5460 			struct strrecvfd	strfd;
5461 
5462 			strfd.fd = fd;
5463 			strfd.uid = srf->uid;
5464 			strfd.gid = srf->gid;
5465 
5466 			/* null the filler bits */
5467 			for (i = 0; i < 8; i++)
5468 				strfd.fill[i] = 0;
5469 
5470 			error = strcopyout(&strfd, (void *)arg,
5471 			    sizeof (struct strrecvfd), copyflag);
5472 		}
5473 
5474 		if (error) {
5475 			setf(fd, NULL);	/* release fd entry */
5476 			mutex_enter(&stp->sd_lock);
5477 			putback(stp, rdq, mp, mp->b_band);
5478 			mutex_exit(&stp->sd_lock);
5479 			return (error);
5480 		}
5481 		if (audit_active) {
5482 			audit_fdrecv(fd, srf->fp);
5483 		}
5484 
5485 		/*
5486 		 * Always increment f_count since the freemsg() below will
5487 		 * always call free_passfp() which performs a closef().
5488 		 */
5489 		mutex_enter(&srf->fp->f_tlock);
5490 		srf->fp->f_count++;
5491 		mutex_exit(&srf->fp->f_tlock);
5492 		setf(fd, srf->fp);
5493 		freemsg(mp);
5494 		return (0);
5495 	}
5496 
5497 	case I_SWROPT:
5498 		/*
5499 		 * Set/clear the write options. arg is a bit
5500 		 * mask with any of the following bits set...
5501 		 * 	SNDZERO - send zero length message
5502 		 *	SNDPIPE - send sigpipe to process if
5503 		 *		sd_werror is set and process is
5504 		 *		doing a write or putmsg.
5505 		 * The new stream head write options should reflect
5506 		 * what is in arg.
5507 		 */
5508 		if (arg & ~(SNDZERO|SNDPIPE))
5509 			return (EINVAL);
5510 
5511 		mutex_enter(&stp->sd_lock);
5512 		stp->sd_wput_opt &= ~(SW_SIGPIPE|SW_SNDZERO);
5513 		if (arg & SNDZERO)
5514 			stp->sd_wput_opt |= SW_SNDZERO;
5515 		if (arg & SNDPIPE)
5516 			stp->sd_wput_opt |= SW_SIGPIPE;
5517 		mutex_exit(&stp->sd_lock);
5518 		return (0);
5519 
5520 	case I_GWROPT:
5521 	{
5522 		int wropt = 0;
5523 
5524 		if (stp->sd_wput_opt & SW_SNDZERO)
5525 			wropt |= SNDZERO;
5526 		if (stp->sd_wput_opt & SW_SIGPIPE)
5527 			wropt |= SNDPIPE;
5528 		return (strcopyout(&wropt, (void *)arg, sizeof (wropt),
5529 		    copyflag));
5530 	}
5531 
5532 	case I_LIST:
5533 		/*
5534 		 * Returns all the modules found on this stream,
5535 		 * upto the driver. If argument is NULL, return the
5536 		 * number of modules (including driver). If argument
5537 		 * is not NULL, copy the names into the structure
5538 		 * provided.
5539 		 */
5540 
5541 	{
5542 		queue_t *q;
5543 		int num_modules, space_allocated;
5544 		STRUCT_DECL(str_list, strlist);
5545 		struct str_mlist *mlist_ptr;
5546 
5547 		if (arg == NULL) { /* Return number of modules plus driver */
5548 			q = stp->sd_wrq;
5549 			if (stp->sd_vnode->v_type == VFIFO) {
5550 				*rvalp = stp->sd_pushcnt;
5551 			} else {
5552 				*rvalp = stp->sd_pushcnt + 1;
5553 			}
5554 		} else {
5555 			STRUCT_INIT(strlist, flag);
5556 
5557 			error = strcopyin((void *)arg, STRUCT_BUF(strlist),
5558 			    STRUCT_SIZE(strlist), copyflag);
5559 			if (error)
5560 				return (error);
5561 
5562 			space_allocated = STRUCT_FGET(strlist, sl_nmods);
5563 			if ((space_allocated) <= 0)
5564 				return (EINVAL);
5565 			claimstr(stp->sd_wrq);
5566 			q = stp->sd_wrq;
5567 			num_modules = 0;
5568 			while (_SAMESTR(q) && (space_allocated != 0)) {
5569 				char *name =
5570 				    q->q_next->q_qinfo->qi_minfo->mi_idname;
5571 
5572 				mlist_ptr = STRUCT_FGETP(strlist, sl_modlist);
5573 
5574 				error = strcopyout(name, mlist_ptr,
5575 				    strlen(name) + 1, copyflag);
5576 
5577 				if (error) {
5578 					releasestr(stp->sd_wrq);
5579 					return (error);
5580 				}
5581 				q = q->q_next;
5582 				space_allocated--;
5583 				num_modules++;
5584 				mlist_ptr =
5585 				    (struct str_mlist *)((uintptr_t)mlist_ptr +
5586 				    sizeof (struct str_mlist));
5587 				STRUCT_FSETP(strlist, sl_modlist, mlist_ptr);
5588 			}
5589 			releasestr(stp->sd_wrq);
5590 			error = strcopyout(&num_modules, (void *)arg,
5591 			    sizeof (int), copyflag);
5592 		}
5593 		return (error);
5594 	}
5595 
5596 	case I_CKBAND:
5597 	{
5598 		queue_t *q;
5599 		qband_t *qbp;
5600 
5601 		if ((arg < 0) || (arg >= NBAND))
5602 			return (EINVAL);
5603 		q = _RD(stp->sd_wrq);
5604 		mutex_enter(QLOCK(q));
5605 		if (arg > (int)q->q_nband) {
5606 			*rvalp = 0;
5607 		} else {
5608 			if (arg == 0) {
5609 				if (q->q_first)
5610 					*rvalp = 1;
5611 				else
5612 					*rvalp = 0;
5613 			} else {
5614 				qbp = q->q_bandp;
5615 				while (--arg > 0)
5616 					qbp = qbp->qb_next;
5617 				if (qbp->qb_first)
5618 					*rvalp = 1;
5619 				else
5620 					*rvalp = 0;
5621 			}
5622 		}
5623 		mutex_exit(QLOCK(q));
5624 		return (0);
5625 	}
5626 
5627 	case I_GETBAND:
5628 	{
5629 		int intpri;
5630 		queue_t *q;
5631 
5632 		q = _RD(stp->sd_wrq);
5633 		mutex_enter(QLOCK(q));
5634 		mp = q->q_first;
5635 		if (!mp) {
5636 			mutex_exit(QLOCK(q));
5637 			return (ENODATA);
5638 		}
5639 		intpri = (int)mp->b_band;
5640 		error = strcopyout(&intpri, (void *)arg, sizeof (int),
5641 		    copyflag);
5642 		mutex_exit(QLOCK(q));
5643 		return (error);
5644 	}
5645 
5646 	case I_ATMARK:
5647 	{
5648 		queue_t *q;
5649 
5650 		if (arg & ~(ANYMARK|LASTMARK))
5651 			return (EINVAL);
5652 		q = _RD(stp->sd_wrq);
5653 		mutex_enter(&stp->sd_lock);
5654 		if ((stp->sd_flag & STRATMARK) && (arg == ANYMARK)) {
5655 			*rvalp = 1;
5656 		} else {
5657 			mutex_enter(QLOCK(q));
5658 			mp = q->q_first;
5659 
5660 			if (mp == NULL)
5661 				*rvalp = 0;
5662 			else if ((arg == ANYMARK) && (mp->b_flag & MSGMARK))
5663 				*rvalp = 1;
5664 			else if ((arg == LASTMARK) && (mp == stp->sd_mark))
5665 				*rvalp = 1;
5666 			else
5667 				*rvalp = 0;
5668 			mutex_exit(QLOCK(q));
5669 		}
5670 		mutex_exit(&stp->sd_lock);
5671 		return (0);
5672 	}
5673 
5674 	case I_CANPUT:
5675 	{
5676 		char band;
5677 
5678 		if ((arg < 0) || (arg >= NBAND))
5679 			return (EINVAL);
5680 		band = (char)arg;
5681 		*rvalp = bcanputnext(stp->sd_wrq, band);
5682 		return (0);
5683 	}
5684 
5685 	case I_SETCLTIME:
5686 	{
5687 		int closetime;
5688 
5689 		error = strcopyin((void *)arg, &closetime, sizeof (int),
5690 		    copyflag);
5691 		if (error)
5692 			return (error);
5693 		if (closetime < 0)
5694 			return (EINVAL);
5695 
5696 		stp->sd_closetime = closetime;
5697 		return (0);
5698 	}
5699 
5700 	case I_GETCLTIME:
5701 	{
5702 		int closetime;
5703 
5704 		closetime = stp->sd_closetime;
5705 		return (strcopyout(&closetime, (void *)arg, sizeof (int),
5706 		    copyflag));
5707 	}
5708 
5709 	case TIOCGSID:
5710 	{
5711 		pid_t sid;
5712 
5713 		mutex_enter(&stp->sd_lock);
5714 		if (stp->sd_sidp == NULL) {
5715 			mutex_exit(&stp->sd_lock);
5716 			return (ENOTTY);
5717 		}
5718 		sid = stp->sd_sidp->pid_id;
5719 		mutex_exit(&stp->sd_lock);
5720 		return (strcopyout(&sid, (void *)arg, sizeof (pid_t),
5721 		    copyflag));
5722 	}
5723 
5724 	case TIOCSPGRP:
5725 	{
5726 		pid_t pgrp;
5727 		proc_t *q;
5728 		pid_t	sid, fg_pgid, bg_pgid;
5729 
5730 		if (error = strcopyin((void *)arg, &pgrp, sizeof (pid_t),
5731 		    copyflag))
5732 			return (error);
5733 		mutex_enter(&stp->sd_lock);
5734 		mutex_enter(&pidlock);
5735 		if (stp->sd_sidp != ttoproc(curthread)->p_sessp->s_sidp) {
5736 			mutex_exit(&pidlock);
5737 			mutex_exit(&stp->sd_lock);
5738 			return (ENOTTY);
5739 		}
5740 		if (pgrp == stp->sd_pgidp->pid_id) {
5741 			mutex_exit(&pidlock);
5742 			mutex_exit(&stp->sd_lock);
5743 			return (0);
5744 		}
5745 		if (pgrp <= 0 || pgrp >= maxpid) {
5746 			mutex_exit(&pidlock);
5747 			mutex_exit(&stp->sd_lock);
5748 			return (EINVAL);
5749 		}
5750 		if ((q = pgfind(pgrp)) == NULL ||
5751 		    q->p_sessp != ttoproc(curthread)->p_sessp) {
5752 			mutex_exit(&pidlock);
5753 			mutex_exit(&stp->sd_lock);
5754 			return (EPERM);
5755 		}
5756 		sid = stp->sd_sidp->pid_id;
5757 		fg_pgid = q->p_pgrp;
5758 		bg_pgid = stp->sd_pgidp->pid_id;
5759 		CL_SET_PROCESS_GROUP(curthread, sid, bg_pgid, fg_pgid);
5760 		PID_RELE(stp->sd_pgidp);
5761 		ctty_clear_sighuped();
5762 		stp->sd_pgidp = q->p_pgidp;
5763 		PID_HOLD(stp->sd_pgidp);
5764 		mutex_exit(&pidlock);
5765 		mutex_exit(&stp->sd_lock);
5766 		return (0);
5767 	}
5768 
5769 	case TIOCGPGRP:
5770 	{
5771 		pid_t pgrp;
5772 
5773 		mutex_enter(&stp->sd_lock);
5774 		if (stp->sd_sidp == NULL) {
5775 			mutex_exit(&stp->sd_lock);
5776 			return (ENOTTY);
5777 		}
5778 		pgrp = stp->sd_pgidp->pid_id;
5779 		mutex_exit(&stp->sd_lock);
5780 		return (strcopyout(&pgrp, (void *)arg, sizeof (pid_t),
5781 		    copyflag));
5782 	}
5783 
5784 	case TIOCSCTTY:
5785 	{
5786 		return (strctty(stp));
5787 	}
5788 
5789 	case TIOCNOTTY:
5790 	{
5791 		/* freectty() always assumes curproc. */
5792 		if (freectty(B_FALSE) != 0)
5793 			return (0);
5794 		return (ENOTTY);
5795 	}
5796 
5797 	case FIONBIO:
5798 	case FIOASYNC:
5799 		return (0);	/* handled by the upper layer */
5800 	}
5801 }
5802 
5803 /*
5804  * Custom free routine used for M_PASSFP messages.
5805  */
5806 static void
5807 free_passfp(struct k_strrecvfd *srf)
5808 {
5809 	(void) closef(srf->fp);
5810 	kmem_free(srf, sizeof (struct k_strrecvfd) + sizeof (frtn_t));
5811 }
5812 
5813 /* ARGSUSED */
5814 int
5815 do_sendfp(struct stdata *stp, struct file *fp, struct cred *cr)
5816 {
5817 	queue_t *qp, *nextqp;
5818 	struct k_strrecvfd *srf;
5819 	mblk_t *mp;
5820 	frtn_t *frtnp;
5821 	size_t bufsize;
5822 	queue_t	*mate = NULL;
5823 	syncq_t	*sq = NULL;
5824 	int retval = 0;
5825 
5826 	if (stp->sd_flag & STRHUP)
5827 		return (ENXIO);
5828 
5829 	claimstr(stp->sd_wrq);
5830 
5831 	/* Fastpath, we have a pipe, and we are already mated, use it. */
5832 	if (STRMATED(stp)) {
5833 		qp = _RD(stp->sd_mate->sd_wrq);
5834 		claimstr(qp);
5835 		mate = qp;
5836 	} else { /* Not already mated. */
5837 
5838 		/*
5839 		 * Walk the stream to the end of this one.
5840 		 * assumes that the claimstr() will prevent
5841 		 * plumbing between the stream head and the
5842 		 * driver from changing
5843 		 */
5844 		qp = stp->sd_wrq;
5845 
5846 		/*
5847 		 * Loop until we reach the end of this stream.
5848 		 * On completion, qp points to the write queue
5849 		 * at the end of the stream, or the read queue
5850 		 * at the stream head if this is a fifo.
5851 		 */
5852 		while (((qp = qp->q_next) != NULL) && _SAMESTR(qp))
5853 			;
5854 
5855 		/*
5856 		 * Just in case we get a q_next which is NULL, but
5857 		 * not at the end of the stream.  This is actually
5858 		 * broken, so we set an assert to catch it in
5859 		 * debug, and set an error and return if not debug.
5860 		 */
5861 		ASSERT(qp);
5862 		if (qp == NULL) {
5863 			releasestr(stp->sd_wrq);
5864 			return (EINVAL);
5865 		}
5866 
5867 		/*
5868 		 * Enter the syncq for the driver, so (hopefully)
5869 		 * the queue values will not change on us.
5870 		 * XXXX - This will only prevent the race IFF only
5871 		 *   the write side modifies the q_next member, and
5872 		 *   the put procedure is protected by at least
5873 		 *   MT_PERQ.
5874 		 */
5875 		if ((sq = qp->q_syncq) != NULL)
5876 			entersq(sq, SQ_PUT);
5877 
5878 		/* Now get the q_next value from this qp. */
5879 		nextqp = qp->q_next;
5880 
5881 		/*
5882 		 * If nextqp exists and the other stream is different
5883 		 * from this one claim the stream, set the mate, and
5884 		 * get the read queue at the stream head of the other
5885 		 * stream.  Assumes that nextqp was at least valid when
5886 		 * we got it.  Hopefully the entersq of the driver
5887 		 * will prevent it from changing on us.
5888 		 */
5889 		if ((nextqp != NULL) && (STREAM(nextqp) != stp)) {
5890 			ASSERT(qp->q_qinfo->qi_srvp);
5891 			ASSERT(_OTHERQ(qp)->q_qinfo->qi_srvp);
5892 			ASSERT(_OTHERQ(qp->q_next)->q_qinfo->qi_srvp);
5893 			claimstr(nextqp);
5894 
5895 			/* Make sure we still have a q_next */
5896 			if (nextqp != qp->q_next) {
5897 				releasestr(stp->sd_wrq);
5898 				releasestr(nextqp);
5899 				return (EINVAL);
5900 			}
5901 
5902 			qp = _RD(STREAM(nextqp)->sd_wrq);
5903 			mate = qp;
5904 		}
5905 		/* If we entered the synq above, leave it. */
5906 		if (sq != NULL)
5907 			leavesq(sq, SQ_PUT);
5908 	} /*  STRMATED(STP)  */
5909 
5910 	/* XXX prevents substitution of the ops vector */
5911 	if (qp->q_qinfo != &strdata && qp->q_qinfo != &fifo_strdata) {
5912 		retval = EINVAL;
5913 		goto out;
5914 	}
5915 
5916 	if (qp->q_flag & QFULL) {
5917 		retval = EAGAIN;
5918 		goto out;
5919 	}
5920 
5921 	/*
5922 	 * Since M_PASSFP messages include a file descriptor, we use
5923 	 * esballoc() and specify a custom free routine (free_passfp()) that
5924 	 * will close the descriptor as part of freeing the message.  For
5925 	 * convenience, we stash the frtn_t right after the data block.
5926 	 */
5927 	bufsize = sizeof (struct k_strrecvfd) + sizeof (frtn_t);
5928 	srf = kmem_alloc(bufsize, KM_NOSLEEP);
5929 	if (srf == NULL) {
5930 		retval = EAGAIN;
5931 		goto out;
5932 	}
5933 
5934 	frtnp = (frtn_t *)(srf + 1);
5935 	frtnp->free_arg = (caddr_t)srf;
5936 	frtnp->free_func = free_passfp;
5937 
5938 	mp = esballoc((uchar_t *)srf, bufsize, BPRI_MED, frtnp);
5939 	if (mp == NULL) {
5940 		kmem_free(srf, bufsize);
5941 		retval = EAGAIN;
5942 		goto out;
5943 	}
5944 	mp->b_wptr += sizeof (struct k_strrecvfd);
5945 	mp->b_datap->db_type = M_PASSFP;
5946 
5947 	srf->fp = fp;
5948 	srf->uid = crgetuid(curthread->t_cred);
5949 	srf->gid = crgetgid(curthread->t_cred);
5950 	mutex_enter(&fp->f_tlock);
5951 	fp->f_count++;
5952 	mutex_exit(&fp->f_tlock);
5953 
5954 	put(qp, mp);
5955 out:
5956 	releasestr(stp->sd_wrq);
5957 	if (mate)
5958 		releasestr(mate);
5959 	return (retval);
5960 }
5961 
5962 /*
5963  * Send an ioctl message downstream and wait for acknowledgement.
5964  * flags may be set to either U_TO_K or K_TO_K and a combination
5965  * of STR_NOERROR or STR_NOSIG
5966  * STR_NOSIG: Signals are essentially ignored or held and have
5967  *	no effect for the duration of the call.
5968  * STR_NOERROR: Ignores stream head read, write and hup errors.
5969  *	Additionally, if an existing ioctl times out, it is assumed
5970  *	lost and and this ioctl will continue as if the previous ioctl had
5971  *	finished.  ETIME may be returned if this ioctl times out (i.e.
5972  *	ic_timout is not INFTIM).  Non-stream head errors may be returned if
5973  *	the ioc_error indicates that the driver/module had problems,
5974  *	an EFAULT was found when accessing user data, a lack of
5975  * 	resources, etc.
5976  */
5977 int
5978 strdoioctl(
5979 	struct stdata *stp,
5980 	struct strioctl *strioc,
5981 	int fflags,		/* file flags with model info */
5982 	int flag,
5983 	cred_t *crp,
5984 	int *rvalp)
5985 {
5986 	mblk_t *bp;
5987 	struct iocblk *iocbp;
5988 	struct copyreq *reqp;
5989 	struct copyresp *resp;
5990 	int id;
5991 	int transparent = 0;
5992 	int error = 0;
5993 	int len = 0;
5994 	caddr_t taddr;
5995 	int copyflag = (flag & (U_TO_K | K_TO_K));
5996 	int sigflag = (flag & STR_NOSIG);
5997 	int errs;
5998 	uint_t waitflags;
5999 
6000 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
6001 	ASSERT((fflags & FMODELS) != 0);
6002 
6003 	TRACE_2(TR_FAC_STREAMS_FR,
6004 	    TR_STRDOIOCTL,
6005 	    "strdoioctl:stp %p strioc %p", stp, strioc);
6006 	if (strioc->ic_len == TRANSPARENT) {	/* send arg in M_DATA block */
6007 		transparent = 1;
6008 		strioc->ic_len = sizeof (intptr_t);
6009 	}
6010 
6011 	if (strioc->ic_len < 0 || (strmsgsz > 0 && strioc->ic_len > strmsgsz))
6012 		return (EINVAL);
6013 
6014 	if ((bp = allocb_cred_wait(sizeof (union ioctypes), sigflag, &error,
6015 	    crp)) == NULL)
6016 			return (error);
6017 
6018 	bzero(bp->b_wptr, sizeof (union ioctypes));
6019 
6020 	iocbp = (struct iocblk *)bp->b_wptr;
6021 	iocbp->ioc_count = strioc->ic_len;
6022 	iocbp->ioc_cmd = strioc->ic_cmd;
6023 	iocbp->ioc_flag = (fflags & FMODELS);
6024 
6025 	crhold(crp);
6026 	iocbp->ioc_cr = crp;
6027 	DB_TYPE(bp) = M_IOCTL;
6028 	DB_CPID(bp) = curproc->p_pid;
6029 	bp->b_wptr += sizeof (struct iocblk);
6030 
6031 	if (flag & STR_NOERROR)
6032 		errs = STPLEX;
6033 	else
6034 		errs = STRHUP|STRDERR|STWRERR|STPLEX;
6035 
6036 	/*
6037 	 * If there is data to copy into ioctl block, do so.
6038 	 */
6039 	if (iocbp->ioc_count > 0) {
6040 		if (transparent)
6041 			/*
6042 			 * Note: STR_NOERROR does not have an effect
6043 			 * in putiocd()
6044 			 */
6045 			id = K_TO_K | sigflag;
6046 		else
6047 			id = flag;
6048 		if ((error = putiocd(bp, strioc->ic_dp, id, crp)) != 0) {
6049 			freemsg(bp);
6050 			crfree(crp);
6051 			return (error);
6052 		}
6053 
6054 		/*
6055 		 * We could have slept copying in user pages.
6056 		 * Recheck the stream head state (the other end
6057 		 * of a pipe could have gone away).
6058 		 */
6059 		if (stp->sd_flag & errs) {
6060 			mutex_enter(&stp->sd_lock);
6061 			error = strgeterr(stp, errs, 0);
6062 			mutex_exit(&stp->sd_lock);
6063 			if (error != 0) {
6064 				freemsg(bp);
6065 				crfree(crp);
6066 				return (error);
6067 			}
6068 		}
6069 	}
6070 	if (transparent)
6071 		iocbp->ioc_count = TRANSPARENT;
6072 
6073 	/*
6074 	 * Block for up to STRTIMOUT milliseconds if there is an outstanding
6075 	 * ioctl for this stream already running.  All processes
6076 	 * sleeping here will be awakened as a result of an ACK
6077 	 * or NAK being received for the outstanding ioctl, or
6078 	 * as a result of the timer expiring on the outstanding
6079 	 * ioctl (a failure), or as a result of any waiting
6080 	 * process's timer expiring (also a failure).
6081 	 */
6082 
6083 	error = 0;
6084 	mutex_enter(&stp->sd_lock);
6085 	while (stp->sd_flag & (IOCWAIT | IOCWAITNE)) {
6086 		clock_t cv_rval;
6087 
6088 		TRACE_0(TR_FAC_STREAMS_FR,
6089 		    TR_STRDOIOCTL_WAIT,
6090 		    "strdoioctl sleeps - IOCWAIT");
6091 		cv_rval = str_cv_wait(&stp->sd_iocmonitor, &stp->sd_lock,
6092 		    STRTIMOUT, sigflag);
6093 		if (cv_rval <= 0) {
6094 			if (cv_rval == 0) {
6095 				error = EINTR;
6096 			} else {
6097 				if (flag & STR_NOERROR) {
6098 					/*
6099 					 * Terminating current ioctl in
6100 					 * progress -- assume it got lost and
6101 					 * wake up the other thread so that the
6102 					 * operation completes.
6103 					 */
6104 					if (!(stp->sd_flag & IOCWAITNE)) {
6105 						stp->sd_flag |= IOCWAITNE;
6106 						cv_broadcast(&stp->sd_monitor);
6107 					}
6108 					/*
6109 					 * Otherwise, there's a running
6110 					 * STR_NOERROR -- we have no choice
6111 					 * here but to wait forever (or until
6112 					 * interrupted).
6113 					 */
6114 				} else {
6115 					/*
6116 					 * pending ioctl has caused
6117 					 * us to time out
6118 					 */
6119 					error = ETIME;
6120 				}
6121 			}
6122 		} else if ((stp->sd_flag & errs)) {
6123 			error = strgeterr(stp, errs, 0);
6124 		}
6125 		if (error) {
6126 			mutex_exit(&stp->sd_lock);
6127 			freemsg(bp);
6128 			crfree(crp);
6129 			return (error);
6130 		}
6131 	}
6132 
6133 	/*
6134 	 * Have control of ioctl mechanism.
6135 	 * Send down ioctl packet and wait for response.
6136 	 */
6137 	if (stp->sd_iocblk != (mblk_t *)-1) {
6138 		freemsg(stp->sd_iocblk);
6139 	}
6140 	stp->sd_iocblk = NULL;
6141 
6142 	/*
6143 	 * If this is marked with 'noerror' (internal; mostly
6144 	 * I_{P,}{UN,}LINK), then make sure nobody else is able to get
6145 	 * in here by setting IOCWAITNE.
6146 	 */
6147 	waitflags = IOCWAIT;
6148 	if (flag & STR_NOERROR)
6149 		waitflags |= IOCWAITNE;
6150 
6151 	stp->sd_flag |= waitflags;
6152 
6153 	/*
6154 	 * Assign sequence number.
6155 	 */
6156 	iocbp->ioc_id = stp->sd_iocid = getiocseqno();
6157 
6158 	mutex_exit(&stp->sd_lock);
6159 
6160 	TRACE_1(TR_FAC_STREAMS_FR,
6161 	    TR_STRDOIOCTL_PUT, "strdoioctl put: stp %p", stp);
6162 	stream_willservice(stp);
6163 	putnext(stp->sd_wrq, bp);
6164 	stream_runservice(stp);
6165 
6166 	/*
6167 	 * Timed wait for acknowledgment.  The wait time is limited by the
6168 	 * timeout value, which must be a positive integer (number of
6169 	 * milliseconds) to wait, or 0 (use default value of STRTIMOUT
6170 	 * milliseconds), or -1 (wait forever).  This will be awakened
6171 	 * either by an ACK/NAK message arriving, the timer expiring, or
6172 	 * the timer expiring on another ioctl waiting for control of the
6173 	 * mechanism.
6174 	 */
6175 waitioc:
6176 	mutex_enter(&stp->sd_lock);
6177 
6178 
6179 	/*
6180 	 * If the reply has already arrived, don't sleep.  If awakened from
6181 	 * the sleep, fail only if the reply has not arrived by then.
6182 	 * Otherwise, process the reply.
6183 	 */
6184 	while (!stp->sd_iocblk) {
6185 		clock_t cv_rval;
6186 
6187 		if (stp->sd_flag & errs) {
6188 			error = strgeterr(stp, errs, 0);
6189 			if (error != 0) {
6190 				stp->sd_flag &= ~waitflags;
6191 				cv_broadcast(&stp->sd_iocmonitor);
6192 				mutex_exit(&stp->sd_lock);
6193 				crfree(crp);
6194 				return (error);
6195 			}
6196 		}
6197 
6198 		TRACE_0(TR_FAC_STREAMS_FR,
6199 		    TR_STRDOIOCTL_WAIT2,
6200 		    "strdoioctl sleeps awaiting reply");
6201 		ASSERT(error == 0);
6202 
6203 		cv_rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock,
6204 		    (strioc->ic_timout ?
6205 		    strioc->ic_timout * 1000 : STRTIMOUT), sigflag);
6206 
6207 		/*
6208 		 * There are four possible cases here: interrupt, timeout,
6209 		 * wakeup by IOCWAITNE (above), or wakeup by strrput_nondata (a
6210 		 * valid M_IOCTL reply).
6211 		 *
6212 		 * If we've been awakened by a STR_NOERROR ioctl on some other
6213 		 * thread, then sd_iocblk will still be NULL, and IOCWAITNE
6214 		 * will be set.  Pretend as if we just timed out.  Note that
6215 		 * this other thread waited at least STRTIMOUT before trying to
6216 		 * awaken our thread, so this is indistinguishable (even for
6217 		 * INFTIM) from the case where we failed with ETIME waiting on
6218 		 * IOCWAIT in the prior loop.
6219 		 */
6220 		if (cv_rval > 0 && !(flag & STR_NOERROR) &&
6221 		    stp->sd_iocblk == NULL && (stp->sd_flag & IOCWAITNE)) {
6222 			cv_rval = -1;
6223 		}
6224 
6225 		/*
6226 		 * note: STR_NOERROR does not protect
6227 		 * us here.. use ic_timout < 0
6228 		 */
6229 		if (cv_rval <= 0) {
6230 			if (cv_rval == 0) {
6231 				error = EINTR;
6232 			} else {
6233 				error =  ETIME;
6234 			}
6235 			/*
6236 			 * A message could have come in after we were scheduled
6237 			 * but before we were actually run.
6238 			 */
6239 			bp = stp->sd_iocblk;
6240 			stp->sd_iocblk = NULL;
6241 			if (bp != NULL) {
6242 				if ((bp->b_datap->db_type == M_COPYIN) ||
6243 				    (bp->b_datap->db_type == M_COPYOUT)) {
6244 					mutex_exit(&stp->sd_lock);
6245 					if (bp->b_cont) {
6246 						freemsg(bp->b_cont);
6247 						bp->b_cont = NULL;
6248 					}
6249 					bp->b_datap->db_type = M_IOCDATA;
6250 					bp->b_wptr = bp->b_rptr +
6251 					    sizeof (struct copyresp);
6252 					resp = (struct copyresp *)bp->b_rptr;
6253 					resp->cp_rval =
6254 					    (caddr_t)1; /* failure */
6255 					stream_willservice(stp);
6256 					putnext(stp->sd_wrq, bp);
6257 					stream_runservice(stp);
6258 					mutex_enter(&stp->sd_lock);
6259 				} else {
6260 					freemsg(bp);
6261 				}
6262 			}
6263 			stp->sd_flag &= ~waitflags;
6264 			cv_broadcast(&stp->sd_iocmonitor);
6265 			mutex_exit(&stp->sd_lock);
6266 			crfree(crp);
6267 			return (error);
6268 		}
6269 	}
6270 	bp = stp->sd_iocblk;
6271 	/*
6272 	 * Note: it is strictly impossible to get here with sd_iocblk set to
6273 	 * -1.  This is because the initial loop above doesn't allow any new
6274 	 * ioctls into the fray until all others have passed this point.
6275 	 */
6276 	ASSERT(bp != NULL && bp != (mblk_t *)-1);
6277 	TRACE_1(TR_FAC_STREAMS_FR,
6278 	    TR_STRDOIOCTL_ACK, "strdoioctl got reply: bp %p", bp);
6279 	if ((bp->b_datap->db_type == M_IOCACK) ||
6280 	    (bp->b_datap->db_type == M_IOCNAK)) {
6281 		/* for detection of duplicate ioctl replies */
6282 		stp->sd_iocblk = (mblk_t *)-1;
6283 		stp->sd_flag &= ~waitflags;
6284 		cv_broadcast(&stp->sd_iocmonitor);
6285 		mutex_exit(&stp->sd_lock);
6286 	} else {
6287 		/*
6288 		 * flags not cleared here because we're still doing
6289 		 * copy in/out for ioctl.
6290 		 */
6291 		stp->sd_iocblk = NULL;
6292 		mutex_exit(&stp->sd_lock);
6293 	}
6294 
6295 
6296 	/*
6297 	 * Have received acknowledgment.
6298 	 */
6299 
6300 	switch (bp->b_datap->db_type) {
6301 	case M_IOCACK:
6302 		/*
6303 		 * Positive ack.
6304 		 */
6305 		iocbp = (struct iocblk *)bp->b_rptr;
6306 
6307 		/*
6308 		 * Set error if indicated.
6309 		 */
6310 		if (iocbp->ioc_error) {
6311 			error = iocbp->ioc_error;
6312 			break;
6313 		}
6314 
6315 		/*
6316 		 * Set return value.
6317 		 */
6318 		*rvalp = iocbp->ioc_rval;
6319 
6320 		/*
6321 		 * Data may have been returned in ACK message (ioc_count > 0).
6322 		 * If so, copy it out to the user's buffer.
6323 		 */
6324 		if (iocbp->ioc_count && !transparent) {
6325 			if (error = getiocd(bp, strioc->ic_dp, copyflag))
6326 				break;
6327 		}
6328 		if (!transparent) {
6329 			if (len)	/* an M_COPYOUT was used with I_STR */
6330 				strioc->ic_len = len;
6331 			else
6332 				strioc->ic_len = (int)iocbp->ioc_count;
6333 		}
6334 		break;
6335 
6336 	case M_IOCNAK:
6337 		/*
6338 		 * Negative ack.
6339 		 *
6340 		 * The only thing to do is set error as specified
6341 		 * in neg ack packet.
6342 		 */
6343 		iocbp = (struct iocblk *)bp->b_rptr;
6344 
6345 		error = (iocbp->ioc_error ? iocbp->ioc_error : EINVAL);
6346 		break;
6347 
6348 	case M_COPYIN:
6349 		/*
6350 		 * Driver or module has requested user ioctl data.
6351 		 */
6352 		reqp = (struct copyreq *)bp->b_rptr;
6353 
6354 		/*
6355 		 * M_COPYIN should *never* have a message attached, though
6356 		 * it's harmless if it does -- thus, panic on a DEBUG
6357 		 * kernel and just free it on a non-DEBUG build.
6358 		 */
6359 		ASSERT(bp->b_cont == NULL);
6360 		if (bp->b_cont != NULL) {
6361 			freemsg(bp->b_cont);
6362 			bp->b_cont = NULL;
6363 		}
6364 
6365 		error = putiocd(bp, reqp->cq_addr, flag, crp);
6366 		if (error && bp->b_cont) {
6367 			freemsg(bp->b_cont);
6368 			bp->b_cont = NULL;
6369 		}
6370 
6371 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6372 		bp->b_datap->db_type = M_IOCDATA;
6373 
6374 		mblk_setcred(bp, crp);
6375 		DB_CPID(bp) = curproc->p_pid;
6376 		resp = (struct copyresp *)bp->b_rptr;
6377 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6378 		resp->cp_flag = (fflags & FMODELS);
6379 
6380 		stream_willservice(stp);
6381 		putnext(stp->sd_wrq, bp);
6382 		stream_runservice(stp);
6383 
6384 		if (error) {
6385 			mutex_enter(&stp->sd_lock);
6386 			stp->sd_flag &= ~waitflags;
6387 			cv_broadcast(&stp->sd_iocmonitor);
6388 			mutex_exit(&stp->sd_lock);
6389 			crfree(crp);
6390 			return (error);
6391 		}
6392 
6393 		goto waitioc;
6394 
6395 	case M_COPYOUT:
6396 		/*
6397 		 * Driver or module has ioctl data for a user.
6398 		 */
6399 		reqp = (struct copyreq *)bp->b_rptr;
6400 		ASSERT(bp->b_cont != NULL);
6401 
6402 		/*
6403 		 * Always (transparent or non-transparent )
6404 		 * use the address specified in the request
6405 		 */
6406 		taddr = reqp->cq_addr;
6407 		if (!transparent)
6408 			len = (int)reqp->cq_size;
6409 
6410 		/* copyout data to the provided address */
6411 		error = getiocd(bp, taddr, copyflag);
6412 
6413 		freemsg(bp->b_cont);
6414 		bp->b_cont = NULL;
6415 
6416 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6417 		bp->b_datap->db_type = M_IOCDATA;
6418 
6419 		mblk_setcred(bp, crp);
6420 		DB_CPID(bp) = curproc->p_pid;
6421 		resp = (struct copyresp *)bp->b_rptr;
6422 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6423 		resp->cp_flag = (fflags & FMODELS);
6424 
6425 		stream_willservice(stp);
6426 		putnext(stp->sd_wrq, bp);
6427 		stream_runservice(stp);
6428 
6429 		if (error) {
6430 			mutex_enter(&stp->sd_lock);
6431 			stp->sd_flag &= ~waitflags;
6432 			cv_broadcast(&stp->sd_iocmonitor);
6433 			mutex_exit(&stp->sd_lock);
6434 			crfree(crp);
6435 			return (error);
6436 		}
6437 		goto waitioc;
6438 
6439 	default:
6440 		ASSERT(0);
6441 		mutex_enter(&stp->sd_lock);
6442 		stp->sd_flag &= ~waitflags;
6443 		cv_broadcast(&stp->sd_iocmonitor);
6444 		mutex_exit(&stp->sd_lock);
6445 		break;
6446 	}
6447 
6448 	freemsg(bp);
6449 	crfree(crp);
6450 	return (error);
6451 }
6452 
6453 /*
6454  * Send an M_CMD message downstream and wait for a reply.  This is a ptools
6455  * special used to retrieve information from modules/drivers a stream without
6456  * being subjected to flow control or interfering with pending messages on the
6457  * stream (e.g. an ioctl in flight).
6458  */
6459 int
6460 strdocmd(struct stdata *stp, struct strcmd *scp, cred_t *crp)
6461 {
6462 	mblk_t *mp;
6463 	struct cmdblk *cmdp;
6464 	int error = 0;
6465 	int errs = STRHUP|STRDERR|STWRERR|STPLEX;
6466 	clock_t rval, timeout = STRTIMOUT;
6467 
6468 	if (scp->sc_len < 0 || scp->sc_len > sizeof (scp->sc_buf) ||
6469 	    scp->sc_timeout < -1)
6470 		return (EINVAL);
6471 
6472 	if (scp->sc_timeout > 0)
6473 		timeout = scp->sc_timeout * MILLISEC;
6474 
6475 	if ((mp = allocb_cred(sizeof (struct cmdblk), crp)) == NULL)
6476 		return (ENOMEM);
6477 
6478 	crhold(crp);
6479 
6480 	cmdp = (struct cmdblk *)mp->b_wptr;
6481 	cmdp->cb_cr = crp;
6482 	cmdp->cb_cmd = scp->sc_cmd;
6483 	cmdp->cb_len = scp->sc_len;
6484 	cmdp->cb_error = 0;
6485 	mp->b_wptr += sizeof (struct cmdblk);
6486 
6487 	DB_TYPE(mp) = M_CMD;
6488 	DB_CPID(mp) = curproc->p_pid;
6489 
6490 	/*
6491 	 * Copy in the payload.
6492 	 */
6493 	if (cmdp->cb_len > 0) {
6494 		mp->b_cont = allocb_cred(sizeof (scp->sc_buf), crp);
6495 		if (mp->b_cont == NULL) {
6496 			error = ENOMEM;
6497 			goto out;
6498 		}
6499 
6500 		/* cb_len comes from sc_len, which has already been checked */
6501 		ASSERT(cmdp->cb_len <= sizeof (scp->sc_buf));
6502 		(void) bcopy(scp->sc_buf, mp->b_cont->b_wptr, cmdp->cb_len);
6503 		mp->b_cont->b_wptr += cmdp->cb_len;
6504 		DB_CPID(mp->b_cont) = curproc->p_pid;
6505 	}
6506 
6507 	/*
6508 	 * Since this mechanism is strictly for ptools, and since only one
6509 	 * process can be grabbed at a time, we simply fail if there's
6510 	 * currently an operation pending.
6511 	 */
6512 	mutex_enter(&stp->sd_lock);
6513 	if (stp->sd_flag & STRCMDWAIT) {
6514 		mutex_exit(&stp->sd_lock);
6515 		error = EBUSY;
6516 		goto out;
6517 	}
6518 	stp->sd_flag |= STRCMDWAIT;
6519 	ASSERT(stp->sd_cmdblk == NULL);
6520 	mutex_exit(&stp->sd_lock);
6521 
6522 	putnext(stp->sd_wrq, mp);
6523 	mp = NULL;
6524 
6525 	/*
6526 	 * Timed wait for acknowledgment.  If the reply has already arrived,
6527 	 * don't sleep.  If awakened from the sleep, fail only if the reply
6528 	 * has not arrived by then.  Otherwise, process the reply.
6529 	 */
6530 	mutex_enter(&stp->sd_lock);
6531 	while (stp->sd_cmdblk == NULL) {
6532 		if (stp->sd_flag & errs) {
6533 			if ((error = strgeterr(stp, errs, 0)) != 0)
6534 				goto waitout;
6535 		}
6536 
6537 		rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock, timeout, 0);
6538 		if (stp->sd_cmdblk != NULL)
6539 			break;
6540 
6541 		if (rval <= 0) {
6542 			error = (rval == 0) ? EINTR : ETIME;
6543 			goto waitout;
6544 		}
6545 	}
6546 
6547 	/*
6548 	 * We received a reply.
6549 	 */
6550 	mp = stp->sd_cmdblk;
6551 	stp->sd_cmdblk = NULL;
6552 	ASSERT(mp != NULL && DB_TYPE(mp) == M_CMD);
6553 	ASSERT(stp->sd_flag & STRCMDWAIT);
6554 	stp->sd_flag &= ~STRCMDWAIT;
6555 	mutex_exit(&stp->sd_lock);
6556 
6557 	cmdp = (struct cmdblk *)mp->b_rptr;
6558 	if ((error = cmdp->cb_error) != 0)
6559 		goto out;
6560 
6561 	/*
6562 	 * Data may have been returned in the reply (cb_len > 0).
6563 	 * If so, copy it out to the user's buffer.
6564 	 */
6565 	if (cmdp->cb_len > 0) {
6566 		if (mp->b_cont == NULL || MBLKL(mp->b_cont) < cmdp->cb_len) {
6567 			error = EPROTO;
6568 			goto out;
6569 		}
6570 
6571 		cmdp->cb_len = MIN(cmdp->cb_len, sizeof (scp->sc_buf));
6572 		(void) bcopy(mp->b_cont->b_rptr, scp->sc_buf, cmdp->cb_len);
6573 	}
6574 	scp->sc_len = cmdp->cb_len;
6575 out:
6576 	freemsg(mp);
6577 	crfree(crp);
6578 	return (error);
6579 waitout:
6580 	ASSERT(stp->sd_cmdblk == NULL);
6581 	stp->sd_flag &= ~STRCMDWAIT;
6582 	mutex_exit(&stp->sd_lock);
6583 	crfree(crp);
6584 	return (error);
6585 }
6586 
6587 /*
6588  * For the SunOS keyboard driver.
6589  * Return the next available "ioctl" sequence number.
6590  * Exported, so that streams modules can send "ioctl" messages
6591  * downstream from their open routine.
6592  */
6593 int
6594 getiocseqno(void)
6595 {
6596 	int	i;
6597 
6598 	mutex_enter(&strresources);
6599 	i = ++ioc_id;
6600 	mutex_exit(&strresources);
6601 	return (i);
6602 }
6603 
6604 /*
6605  * Get the next message from the read queue.  If the message is
6606  * priority, STRPRI will have been set by strrput().  This flag
6607  * should be reset only when the entire message at the front of the
6608  * queue as been consumed.
6609  *
6610  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
6611  */
6612 int
6613 strgetmsg(
6614 	struct vnode *vp,
6615 	struct strbuf *mctl,
6616 	struct strbuf *mdata,
6617 	unsigned char *prip,
6618 	int *flagsp,
6619 	int fmode,
6620 	rval_t *rvp)
6621 {
6622 	struct stdata *stp;
6623 	mblk_t *bp, *nbp;
6624 	mblk_t *savemp = NULL;
6625 	mblk_t *savemptail = NULL;
6626 	uint_t old_sd_flag;
6627 	int flg;
6628 	int more = 0;
6629 	int error = 0;
6630 	char first = 1;
6631 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
6632 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
6633 	unsigned char pri = 0;
6634 	queue_t *q;
6635 	int	pr = 0;			/* Partial read successful */
6636 	struct uio uios;
6637 	struct uio *uiop = &uios;
6638 	struct iovec iovs;
6639 	unsigned char type;
6640 
6641 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRGETMSG_ENTER,
6642 	    "strgetmsg:%p", vp);
6643 
6644 	ASSERT(vp->v_stream);
6645 	stp = vp->v_stream;
6646 	rvp->r_val1 = 0;
6647 
6648 	mutex_enter(&stp->sd_lock);
6649 
6650 	if ((error = i_straccess(stp, JCREAD)) != 0) {
6651 		mutex_exit(&stp->sd_lock);
6652 		return (error);
6653 	}
6654 
6655 	if (stp->sd_flag & (STRDERR|STPLEX)) {
6656 		error = strgeterr(stp, STRDERR|STPLEX, 0);
6657 		if (error != 0) {
6658 			mutex_exit(&stp->sd_lock);
6659 			return (error);
6660 		}
6661 	}
6662 	mutex_exit(&stp->sd_lock);
6663 
6664 	switch (*flagsp) {
6665 	case MSG_HIPRI:
6666 		if (*prip != 0)
6667 			return (EINVAL);
6668 		break;
6669 
6670 	case MSG_ANY:
6671 	case MSG_BAND:
6672 		break;
6673 
6674 	default:
6675 		return (EINVAL);
6676 	}
6677 	/*
6678 	 * Setup uio and iov for data part
6679 	 */
6680 	iovs.iov_base = mdata->buf;
6681 	iovs.iov_len = mdata->maxlen;
6682 	uios.uio_iov = &iovs;
6683 	uios.uio_iovcnt = 1;
6684 	uios.uio_loffset = 0;
6685 	uios.uio_segflg = UIO_USERSPACE;
6686 	uios.uio_fmode = 0;
6687 	uios.uio_extflg = UIO_COPY_CACHED;
6688 	uios.uio_resid = mdata->maxlen;
6689 	uios.uio_offset = 0;
6690 
6691 	q = _RD(stp->sd_wrq);
6692 	mutex_enter(&stp->sd_lock);
6693 	old_sd_flag = stp->sd_flag;
6694 	mark = 0;
6695 	for (;;) {
6696 		int done = 0;
6697 		mblk_t *q_first = q->q_first;
6698 
6699 		/*
6700 		 * Get the next message of appropriate priority
6701 		 * from the stream head.  If the caller is interested
6702 		 * in band or hipri messages, then they should already
6703 		 * be enqueued at the stream head.  On the other hand
6704 		 * if the caller wants normal (band 0) messages, they
6705 		 * might be deferred in a synchronous stream and they
6706 		 * will need to be pulled up.
6707 		 *
6708 		 * After we have dequeued a message, we might find that
6709 		 * it was a deferred M_SIG that was enqueued at the
6710 		 * stream head.  It must now be posted as part of the
6711 		 * read by calling strsignal_nolock().
6712 		 *
6713 		 * Also note that strrput does not enqueue an M_PCSIG,
6714 		 * and there cannot be more than one hipri message,
6715 		 * so there was no need to have the M_PCSIG case.
6716 		 *
6717 		 * At some time it might be nice to try and wrap the
6718 		 * functionality of kstrgetmsg() and strgetmsg() into
6719 		 * a common routine so to reduce the amount of replicated
6720 		 * code (since they are extremely similar).
6721 		 */
6722 		if (!(*flagsp & (MSG_HIPRI|MSG_BAND))) {
6723 			/* Asking for normal, band0 data */
6724 			bp = strget(stp, q, uiop, first, &error);
6725 			ASSERT(MUTEX_HELD(&stp->sd_lock));
6726 			if (bp != NULL) {
6727 				ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
6728 				if (bp->b_datap->db_type == M_SIG) {
6729 					strsignal_nolock(stp, *bp->b_rptr,
6730 					    (int32_t)bp->b_band);
6731 					continue;
6732 				} else {
6733 					break;
6734 				}
6735 			}
6736 			if (error != 0) {
6737 				goto getmout;
6738 			}
6739 
6740 		/*
6741 		 * We can't depend on the value of STRPRI here because
6742 		 * the stream head may be in transit. Therefore, we
6743 		 * must look at the type of the first message to
6744 		 * determine if a high priority messages is waiting
6745 		 */
6746 		} else if ((*flagsp & MSG_HIPRI) && q_first != NULL &&
6747 		    q_first->b_datap->db_type >= QPCTL &&
6748 		    (bp = getq_noenab(q)) != NULL) {
6749 			/* Asked for HIPRI and got one */
6750 			ASSERT(bp->b_datap->db_type >= QPCTL);
6751 			break;
6752 		} else if ((*flagsp & MSG_BAND) && q_first != NULL &&
6753 		    ((q_first->b_band >= *prip) ||
6754 		    q_first->b_datap->db_type >= QPCTL) &&
6755 		    (bp = getq_noenab(q)) != NULL) {
6756 			/*
6757 			 * Asked for at least band "prip" and got either at
6758 			 * least that band or a hipri message.
6759 			 */
6760 			ASSERT(bp->b_band >= *prip ||
6761 			    bp->b_datap->db_type >= QPCTL);
6762 			if (bp->b_datap->db_type == M_SIG) {
6763 				strsignal_nolock(stp, *bp->b_rptr,
6764 				    (int32_t)bp->b_band);
6765 				continue;
6766 			} else {
6767 				break;
6768 			}
6769 		}
6770 
6771 		/* No data. Time to sleep? */
6772 		qbackenable(q, 0);
6773 
6774 		/*
6775 		 * If STRHUP or STREOF, return 0 length control and data.
6776 		 * If resid is 0, then a read(fd,buf,0) was done. Do not
6777 		 * sleep to satisfy this request because by default we have
6778 		 * zero bytes to return.
6779 		 */
6780 		if ((stp->sd_flag & (STRHUP|STREOF)) || (mctl->maxlen == 0 &&
6781 		    mdata->maxlen == 0)) {
6782 			mctl->len = mdata->len = 0;
6783 			*flagsp = 0;
6784 			mutex_exit(&stp->sd_lock);
6785 			return (0);
6786 		}
6787 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_WAIT,
6788 		    "strgetmsg calls strwaitq:%p, %p",
6789 		    vp, uiop);
6790 		if (((error = strwaitq(stp, GETWAIT, (ssize_t)0, fmode, -1,
6791 		    &done)) != 0) || done) {
6792 			TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_DONE,
6793 			    "strgetmsg error or done:%p, %p",
6794 			    vp, uiop);
6795 			mutex_exit(&stp->sd_lock);
6796 			return (error);
6797 		}
6798 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_AWAKE,
6799 		    "strgetmsg awakes:%p, %p", vp, uiop);
6800 		if ((error = i_straccess(stp, JCREAD)) != 0) {
6801 			mutex_exit(&stp->sd_lock);
6802 			return (error);
6803 		}
6804 		first = 0;
6805 	}
6806 	ASSERT(bp != NULL);
6807 	/*
6808 	 * Extract any mark information. If the message is not completely
6809 	 * consumed this information will be put in the mblk
6810 	 * that is putback.
6811 	 * If MSGMARKNEXT is set and the message is completely consumed
6812 	 * the STRATMARK flag will be set below. Likewise, if
6813 	 * MSGNOTMARKNEXT is set and the message is
6814 	 * completely consumed STRNOTATMARK will be set.
6815 	 */
6816 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
6817 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
6818 	    (MSGMARKNEXT|MSGNOTMARKNEXT));
6819 	if (mark != 0 && bp == stp->sd_mark) {
6820 		mark |= _LASTMARK;
6821 		stp->sd_mark = NULL;
6822 	}
6823 	/*
6824 	 * keep track of the original message type and priority
6825 	 */
6826 	pri = bp->b_band;
6827 	type = bp->b_datap->db_type;
6828 	if (type == M_PASSFP) {
6829 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
6830 			stp->sd_mark = bp;
6831 		bp->b_flag |= mark & ~_LASTMARK;
6832 		putback(stp, q, bp, pri);
6833 		qbackenable(q, pri);
6834 		mutex_exit(&stp->sd_lock);
6835 		return (EBADMSG);
6836 	}
6837 	ASSERT(type != M_SIG);
6838 
6839 	/*
6840 	 * Set this flag so strrput will not generate signals. Need to
6841 	 * make sure this flag is cleared before leaving this routine
6842 	 * else signals will stop being sent.
6843 	 */
6844 	stp->sd_flag |= STRGETINPROG;
6845 	mutex_exit(&stp->sd_lock);
6846 
6847 	if (STREAM_NEEDSERVICE(stp))
6848 		stream_runservice(stp);
6849 
6850 	/*
6851 	 * Set HIPRI flag if message is priority.
6852 	 */
6853 	if (type >= QPCTL)
6854 		flg = MSG_HIPRI;
6855 	else
6856 		flg = MSG_BAND;
6857 
6858 	/*
6859 	 * First process PROTO or PCPROTO blocks, if any.
6860 	 */
6861 	if (mctl->maxlen >= 0 && type != M_DATA) {
6862 		size_t	n, bcnt;
6863 		char	*ubuf;
6864 
6865 		bcnt = mctl->maxlen;
6866 		ubuf = mctl->buf;
6867 		while (bp != NULL && bp->b_datap->db_type != M_DATA) {
6868 			if ((n = MIN(bcnt, bp->b_wptr - bp->b_rptr)) != 0 &&
6869 			    copyout(bp->b_rptr, ubuf, n)) {
6870 				error = EFAULT;
6871 				mutex_enter(&stp->sd_lock);
6872 				/*
6873 				 * clear stream head pri flag based on
6874 				 * first message type
6875 				 */
6876 				if (type >= QPCTL) {
6877 					ASSERT(type == M_PCPROTO);
6878 					stp->sd_flag &= ~STRPRI;
6879 				}
6880 				more = 0;
6881 				freemsg(bp);
6882 				goto getmout;
6883 			}
6884 			ubuf += n;
6885 			bp->b_rptr += n;
6886 			if (bp->b_rptr >= bp->b_wptr) {
6887 				nbp = bp;
6888 				bp = bp->b_cont;
6889 				freeb(nbp);
6890 			}
6891 			ASSERT(n <= bcnt);
6892 			bcnt -= n;
6893 			if (bcnt == 0)
6894 				break;
6895 		}
6896 		mctl->len = mctl->maxlen - bcnt;
6897 	} else
6898 		mctl->len = -1;
6899 
6900 	if (bp && bp->b_datap->db_type != M_DATA) {
6901 		/*
6902 		 * More PROTO blocks in msg.
6903 		 */
6904 		more |= MORECTL;
6905 		savemp = bp;
6906 		while (bp && bp->b_datap->db_type != M_DATA) {
6907 			savemptail = bp;
6908 			bp = bp->b_cont;
6909 		}
6910 		savemptail->b_cont = NULL;
6911 	}
6912 
6913 	/*
6914 	 * Now process DATA blocks, if any.
6915 	 */
6916 	if (mdata->maxlen >= 0 && bp) {
6917 		/*
6918 		 * struiocopyout will consume a potential zero-length
6919 		 * M_DATA even if uio_resid is zero.
6920 		 */
6921 		size_t oldresid = uiop->uio_resid;
6922 
6923 		bp = struiocopyout(bp, uiop, &error);
6924 		if (error != 0) {
6925 			mutex_enter(&stp->sd_lock);
6926 			/*
6927 			 * clear stream head hi pri flag based on
6928 			 * first message
6929 			 */
6930 			if (type >= QPCTL) {
6931 				ASSERT(type == M_PCPROTO);
6932 				stp->sd_flag &= ~STRPRI;
6933 			}
6934 			more = 0;
6935 			freemsg(savemp);
6936 			goto getmout;
6937 		}
6938 		/*
6939 		 * (pr == 1) indicates a partial read.
6940 		 */
6941 		if (oldresid > uiop->uio_resid)
6942 			pr = 1;
6943 		mdata->len = mdata->maxlen - uiop->uio_resid;
6944 	} else
6945 		mdata->len = -1;
6946 
6947 	if (bp) {			/* more data blocks in msg */
6948 		more |= MOREDATA;
6949 		if (savemp)
6950 			savemptail->b_cont = bp;
6951 		else
6952 			savemp = bp;
6953 	}
6954 
6955 	mutex_enter(&stp->sd_lock);
6956 	if (savemp) {
6957 		if (pr && (savemp->b_datap->db_type == M_DATA) &&
6958 		    msgnodata(savemp)) {
6959 			/*
6960 			 * Avoid queuing a zero-length tail part of
6961 			 * a message. pr=1 indicates that we read some of
6962 			 * the message.
6963 			 */
6964 			freemsg(savemp);
6965 			more &= ~MOREDATA;
6966 			/*
6967 			 * clear stream head hi pri flag based on
6968 			 * first message
6969 			 */
6970 			if (type >= QPCTL) {
6971 				ASSERT(type == M_PCPROTO);
6972 				stp->sd_flag &= ~STRPRI;
6973 			}
6974 		} else {
6975 			savemp->b_band = pri;
6976 			/*
6977 			 * If the first message was HIPRI and the one we're
6978 			 * putting back isn't, then clear STRPRI, otherwise
6979 			 * set STRPRI again.  Note that we must set STRPRI
6980 			 * again since the flush logic in strrput_nondata()
6981 			 * may have cleared it while we had sd_lock dropped.
6982 			 */
6983 			if (type >= QPCTL) {
6984 				ASSERT(type == M_PCPROTO);
6985 				if (queclass(savemp) < QPCTL)
6986 					stp->sd_flag &= ~STRPRI;
6987 				else
6988 					stp->sd_flag |= STRPRI;
6989 			} else if (queclass(savemp) >= QPCTL) {
6990 				/*
6991 				 * The first message was not a HIPRI message,
6992 				 * but the one we are about to putback is.
6993 				 * For simplicitly, we do not allow for HIPRI
6994 				 * messages to be embedded in the message
6995 				 * body, so just force it to same type as
6996 				 * first message.
6997 				 */
6998 				ASSERT(type == M_DATA || type == M_PROTO);
6999 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
7000 				savemp->b_datap->db_type = type;
7001 			}
7002 			if (mark != 0) {
7003 				savemp->b_flag |= mark & ~_LASTMARK;
7004 				if ((mark & _LASTMARK) &&
7005 				    (stp->sd_mark == NULL)) {
7006 					/*
7007 					 * If another marked message arrived
7008 					 * while sd_lock was not held sd_mark
7009 					 * would be non-NULL.
7010 					 */
7011 					stp->sd_mark = savemp;
7012 				}
7013 			}
7014 			putback(stp, q, savemp, pri);
7015 		}
7016 	} else {
7017 		/*
7018 		 * The complete message was consumed.
7019 		 *
7020 		 * If another M_PCPROTO arrived while sd_lock was not held
7021 		 * it would have been discarded since STRPRI was still set.
7022 		 *
7023 		 * Move the MSG*MARKNEXT information
7024 		 * to the stream head just in case
7025 		 * the read queue becomes empty.
7026 		 * clear stream head hi pri flag based on
7027 		 * first message
7028 		 *
7029 		 * If the stream head was at the mark
7030 		 * (STRATMARK) before we dropped sd_lock above
7031 		 * and some data was consumed then we have
7032 		 * moved past the mark thus STRATMARK is
7033 		 * cleared. However, if a message arrived in
7034 		 * strrput during the copyout above causing
7035 		 * STRATMARK to be set we can not clear that
7036 		 * flag.
7037 		 */
7038 		if (type >= QPCTL) {
7039 			ASSERT(type == M_PCPROTO);
7040 			stp->sd_flag &= ~STRPRI;
7041 		}
7042 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
7043 			if (mark & MSGMARKNEXT) {
7044 				stp->sd_flag &= ~STRNOTATMARK;
7045 				stp->sd_flag |= STRATMARK;
7046 			} else if (mark & MSGNOTMARKNEXT) {
7047 				stp->sd_flag &= ~STRATMARK;
7048 				stp->sd_flag |= STRNOTATMARK;
7049 			} else {
7050 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
7051 			}
7052 		} else if (pr && (old_sd_flag & STRATMARK)) {
7053 			stp->sd_flag &= ~STRATMARK;
7054 		}
7055 	}
7056 
7057 	*flagsp = flg;
7058 	*prip = pri;
7059 
7060 	/*
7061 	 * Getmsg cleanup processing - if the state of the queue has changed
7062 	 * some signals may need to be sent and/or poll awakened.
7063 	 */
7064 getmout:
7065 	qbackenable(q, pri);
7066 
7067 	/*
7068 	 * We dropped the stream head lock above. Send all M_SIG messages
7069 	 * before processing stream head for SIGPOLL messages.
7070 	 */
7071 	ASSERT(MUTEX_HELD(&stp->sd_lock));
7072 	while ((bp = q->q_first) != NULL &&
7073 	    (bp->b_datap->db_type == M_SIG)) {
7074 		/*
7075 		 * sd_lock is held so the content of the read queue can not
7076 		 * change.
7077 		 */
7078 		bp = getq(q);
7079 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7080 
7081 		strsignal_nolock(stp, *bp->b_rptr, (int32_t)bp->b_band);
7082 		mutex_exit(&stp->sd_lock);
7083 		freemsg(bp);
7084 		if (STREAM_NEEDSERVICE(stp))
7085 			stream_runservice(stp);
7086 		mutex_enter(&stp->sd_lock);
7087 	}
7088 
7089 	/*
7090 	 * stream head cannot change while we make the determination
7091 	 * whether or not to send a signal. Drop the flag to allow strrput
7092 	 * to send firstmsgsigs again.
7093 	 */
7094 	stp->sd_flag &= ~STRGETINPROG;
7095 
7096 	/*
7097 	 * If the type of message at the front of the queue changed
7098 	 * due to the receive the appropriate signals and pollwakeup events
7099 	 * are generated. The type of changes are:
7100 	 *	Processed a hipri message, q_first is not hipri.
7101 	 *	Processed a band X message, and q_first is band Y.
7102 	 * The generated signals and pollwakeups are identical to what
7103 	 * strrput() generates should the message that is now on q_first
7104 	 * arrive to an empty read queue.
7105 	 *
7106 	 * Note: only strrput will send a signal for a hipri message.
7107 	 */
7108 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7109 		strsigset_t signals = 0;
7110 		strpollset_t pollwakeups = 0;
7111 
7112 		if (flg & MSG_HIPRI) {
7113 			/*
7114 			 * Removed a hipri message. Regular data at
7115 			 * the front of  the queue.
7116 			 */
7117 			if (bp->b_band == 0) {
7118 				signals = S_INPUT | S_RDNORM;
7119 				pollwakeups = POLLIN | POLLRDNORM;
7120 			} else {
7121 				signals = S_INPUT | S_RDBAND;
7122 				pollwakeups = POLLIN | POLLRDBAND;
7123 			}
7124 		} else if (pri != bp->b_band) {
7125 			/*
7126 			 * The band is different for the new q_first.
7127 			 */
7128 			if (bp->b_band == 0) {
7129 				signals = S_RDNORM;
7130 				pollwakeups = POLLIN | POLLRDNORM;
7131 			} else {
7132 				signals = S_RDBAND;
7133 				pollwakeups = POLLIN | POLLRDBAND;
7134 			}
7135 		}
7136 
7137 		if (pollwakeups != 0) {
7138 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
7139 				if (!(stp->sd_rput_opt & SR_POLLIN))
7140 					goto no_pollwake;
7141 				stp->sd_rput_opt &= ~SR_POLLIN;
7142 			}
7143 			mutex_exit(&stp->sd_lock);
7144 			pollwakeup(&stp->sd_pollist, pollwakeups);
7145 			mutex_enter(&stp->sd_lock);
7146 		}
7147 no_pollwake:
7148 
7149 		if (stp->sd_sigflags & signals)
7150 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7151 	}
7152 	mutex_exit(&stp->sd_lock);
7153 
7154 	rvp->r_val1 = more;
7155 	return (error);
7156 #undef	_LASTMARK
7157 }
7158 
7159 /*
7160  * Get the next message from the read queue.  If the message is
7161  * priority, STRPRI will have been set by strrput().  This flag
7162  * should be reset only when the entire message at the front of the
7163  * queue as been consumed.
7164  *
7165  * If uiop is NULL all data is returned in mctlp.
7166  * Note that a NULL uiop implies that FNDELAY and FNONBLOCK are assumed
7167  * not enabled.
7168  * The timeout parameter is in milliseconds; -1 for infinity.
7169  * This routine handles the consolidation private flags:
7170  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
7171  *	MSG_DELAYERROR	Defer the error check until the queue is empty.
7172  *	MSG_HOLDSIG	Hold signals while waiting for data.
7173  *	MSG_IPEEK	Only peek at messages.
7174  *	MSG_DISCARDTAIL	Discard the tail M_DATA part of the message
7175  *			that doesn't fit.
7176  *	MSG_NOMARK	If the message is marked leave it on the queue.
7177  *
7178  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
7179  */
7180 int
7181 kstrgetmsg(
7182 	struct vnode *vp,
7183 	mblk_t **mctlp,
7184 	struct uio *uiop,
7185 	unsigned char *prip,
7186 	int *flagsp,
7187 	clock_t timout,
7188 	rval_t *rvp)
7189 {
7190 	struct stdata *stp;
7191 	mblk_t *bp, *nbp;
7192 	mblk_t *savemp = NULL;
7193 	mblk_t *savemptail = NULL;
7194 	int flags;
7195 	uint_t old_sd_flag;
7196 	int flg;
7197 	int more = 0;
7198 	int error = 0;
7199 	char first = 1;
7200 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
7201 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
7202 	unsigned char pri = 0;
7203 	queue_t *q;
7204 	int	pr = 0;			/* Partial read successful */
7205 	unsigned char type;
7206 
7207 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_ENTER,
7208 	    "kstrgetmsg:%p", vp);
7209 
7210 	ASSERT(vp->v_stream);
7211 	stp = vp->v_stream;
7212 	rvp->r_val1 = 0;
7213 
7214 	mutex_enter(&stp->sd_lock);
7215 
7216 	if ((error = i_straccess(stp, JCREAD)) != 0) {
7217 		mutex_exit(&stp->sd_lock);
7218 		return (error);
7219 	}
7220 
7221 	flags = *flagsp;
7222 	if (stp->sd_flag & (STRDERR|STPLEX)) {
7223 		if ((stp->sd_flag & STPLEX) ||
7224 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == 0) {
7225 			error = strgeterr(stp, STRDERR|STPLEX,
7226 			    (flags & MSG_IPEEK));
7227 			if (error != 0) {
7228 				mutex_exit(&stp->sd_lock);
7229 				return (error);
7230 			}
7231 		}
7232 	}
7233 	mutex_exit(&stp->sd_lock);
7234 
7235 	switch (flags & (MSG_HIPRI|MSG_ANY|MSG_BAND)) {
7236 	case MSG_HIPRI:
7237 		if (*prip != 0)
7238 			return (EINVAL);
7239 		break;
7240 
7241 	case MSG_ANY:
7242 	case MSG_BAND:
7243 		break;
7244 
7245 	default:
7246 		return (EINVAL);
7247 	}
7248 
7249 retry:
7250 	q = _RD(stp->sd_wrq);
7251 	mutex_enter(&stp->sd_lock);
7252 	old_sd_flag = stp->sd_flag;
7253 	mark = 0;
7254 	for (;;) {
7255 		int done = 0;
7256 		int waitflag;
7257 		int fmode;
7258 		mblk_t *q_first = q->q_first;
7259 
7260 		/*
7261 		 * This section of the code operates just like the code
7262 		 * in strgetmsg().  There is a comment there about what
7263 		 * is going on here.
7264 		 */
7265 		if (!(flags & (MSG_HIPRI|MSG_BAND))) {
7266 			/* Asking for normal, band0 data */
7267 			bp = strget(stp, q, uiop, first, &error);
7268 			ASSERT(MUTEX_HELD(&stp->sd_lock));
7269 			if (bp != NULL) {
7270 				if (bp->b_datap->db_type == M_SIG) {
7271 					strsignal_nolock(stp, *bp->b_rptr,
7272 					    (int32_t)bp->b_band);
7273 					continue;
7274 				} else {
7275 					break;
7276 				}
7277 			}
7278 			if (error != 0) {
7279 				goto getmout;
7280 			}
7281 		/*
7282 		 * We can't depend on the value of STRPRI here because
7283 		 * the stream head may be in transit. Therefore, we
7284 		 * must look at the type of the first message to
7285 		 * determine if a high priority messages is waiting
7286 		 */
7287 		} else if ((flags & MSG_HIPRI) && q_first != NULL &&
7288 		    q_first->b_datap->db_type >= QPCTL &&
7289 		    (bp = getq_noenab(q)) != NULL) {
7290 			ASSERT(bp->b_datap->db_type >= QPCTL);
7291 			break;
7292 		} else if ((flags & MSG_BAND) && q_first != NULL &&
7293 		    ((q_first->b_band >= *prip) ||
7294 		    q_first->b_datap->db_type >= QPCTL) &&
7295 		    (bp = getq_noenab(q)) != NULL) {
7296 			/*
7297 			 * Asked for at least band "prip" and got either at
7298 			 * least that band or a hipri message.
7299 			 */
7300 			ASSERT(bp->b_band >= *prip ||
7301 			    bp->b_datap->db_type >= QPCTL);
7302 			if (bp->b_datap->db_type == M_SIG) {
7303 				strsignal_nolock(stp, *bp->b_rptr,
7304 				    (int32_t)bp->b_band);
7305 				continue;
7306 			} else {
7307 				break;
7308 			}
7309 		}
7310 
7311 		/* No data. Time to sleep? */
7312 		qbackenable(q, 0);
7313 
7314 		/*
7315 		 * Delayed error notification?
7316 		 */
7317 		if ((stp->sd_flag & (STRDERR|STPLEX)) &&
7318 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == MSG_DELAYERROR) {
7319 			error = strgeterr(stp, STRDERR|STPLEX,
7320 			    (flags & MSG_IPEEK));
7321 			if (error != 0) {
7322 				mutex_exit(&stp->sd_lock);
7323 				return (error);
7324 			}
7325 		}
7326 
7327 		/*
7328 		 * If STRHUP or STREOF, return 0 length control and data.
7329 		 * If a read(fd,buf,0) has been done, do not sleep, just
7330 		 * return.
7331 		 *
7332 		 * If mctlp == NULL and uiop == NULL, then the code will
7333 		 * do the strwaitq. This is an understood way of saying
7334 		 * sleep "polling" until a message is received.
7335 		 */
7336 		if ((stp->sd_flag & (STRHUP|STREOF)) ||
7337 		    (uiop != NULL && uiop->uio_resid == 0)) {
7338 			if (mctlp != NULL)
7339 				*mctlp = NULL;
7340 			*flagsp = 0;
7341 			mutex_exit(&stp->sd_lock);
7342 			return (0);
7343 		}
7344 
7345 		waitflag = GETWAIT;
7346 		if (flags &
7347 		    (MSG_HOLDSIG|MSG_IGNERROR|MSG_IPEEK|MSG_DELAYERROR)) {
7348 			if (flags & MSG_HOLDSIG)
7349 				waitflag |= STR_NOSIG;
7350 			if (flags & MSG_IGNERROR)
7351 				waitflag |= STR_NOERROR;
7352 			if (flags & MSG_IPEEK)
7353 				waitflag |= STR_PEEK;
7354 			if (flags & MSG_DELAYERROR)
7355 				waitflag |= STR_DELAYERR;
7356 		}
7357 		if (uiop != NULL)
7358 			fmode = uiop->uio_fmode;
7359 		else
7360 			fmode = 0;
7361 
7362 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_WAIT,
7363 		    "kstrgetmsg calls strwaitq:%p, %p",
7364 		    vp, uiop);
7365 		if (((error = strwaitq(stp, waitflag, (ssize_t)0,
7366 		    fmode, timout, &done))) != 0 || done) {
7367 			TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_DONE,
7368 			    "kstrgetmsg error or done:%p, %p",
7369 			    vp, uiop);
7370 			mutex_exit(&stp->sd_lock);
7371 			return (error);
7372 		}
7373 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_AWAKE,
7374 		    "kstrgetmsg awakes:%p, %p", vp, uiop);
7375 		if ((error = i_straccess(stp, JCREAD)) != 0) {
7376 			mutex_exit(&stp->sd_lock);
7377 			return (error);
7378 		}
7379 		first = 0;
7380 	}
7381 	ASSERT(bp != NULL);
7382 	/*
7383 	 * Extract any mark information. If the message is not completely
7384 	 * consumed this information will be put in the mblk
7385 	 * that is putback.
7386 	 * If MSGMARKNEXT is set and the message is completely consumed
7387 	 * the STRATMARK flag will be set below. Likewise, if
7388 	 * MSGNOTMARKNEXT is set and the message is
7389 	 * completely consumed STRNOTATMARK will be set.
7390 	 */
7391 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
7392 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
7393 	    (MSGMARKNEXT|MSGNOTMARKNEXT));
7394 	pri = bp->b_band;
7395 	if (mark != 0) {
7396 		/*
7397 		 * If the caller doesn't want the mark return.
7398 		 * Used to implement MSG_WAITALL in sockets.
7399 		 */
7400 		if (flags & MSG_NOMARK) {
7401 			putback(stp, q, bp, pri);
7402 			qbackenable(q, pri);
7403 			mutex_exit(&stp->sd_lock);
7404 			return (EWOULDBLOCK);
7405 		}
7406 		if (bp == stp->sd_mark) {
7407 			mark |= _LASTMARK;
7408 			stp->sd_mark = NULL;
7409 		}
7410 	}
7411 
7412 	/*
7413 	 * keep track of the first message type
7414 	 */
7415 	type = bp->b_datap->db_type;
7416 
7417 	if (bp->b_datap->db_type == M_PASSFP) {
7418 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7419 			stp->sd_mark = bp;
7420 		bp->b_flag |= mark & ~_LASTMARK;
7421 		putback(stp, q, bp, pri);
7422 		qbackenable(q, pri);
7423 		mutex_exit(&stp->sd_lock);
7424 		return (EBADMSG);
7425 	}
7426 	ASSERT(type != M_SIG);
7427 
7428 	if (flags & MSG_IPEEK) {
7429 		/*
7430 		 * Clear any struioflag - we do the uiomove over again
7431 		 * when peeking since it simplifies the code.
7432 		 *
7433 		 * Dup the message and put the original back on the queue.
7434 		 * If dupmsg() fails, try again with copymsg() to see if
7435 		 * there is indeed a shortage of memory.  dupmsg() may fail
7436 		 * if db_ref in any of the messages reaches its limit.
7437 		 */
7438 
7439 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
7440 		if ((nbp = dupmsg(bp)) == NULL && (nbp = copymsg(bp)) == NULL) {
7441 			/*
7442 			 * Restore the state of the stream head since we
7443 			 * need to drop sd_lock (strwaitbuf is sleeping).
7444 			 */
7445 			size_t size = msgdsize(bp);
7446 
7447 			if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7448 				stp->sd_mark = bp;
7449 			bp->b_flag |= mark & ~_LASTMARK;
7450 			putback(stp, q, bp, pri);
7451 			mutex_exit(&stp->sd_lock);
7452 			error = strwaitbuf(size, BPRI_HI);
7453 			if (error) {
7454 				/*
7455 				 * There is no net change to the queue thus
7456 				 * no need to qbackenable.
7457 				 */
7458 				return (error);
7459 			}
7460 			goto retry;
7461 		}
7462 
7463 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7464 			stp->sd_mark = bp;
7465 		bp->b_flag |= mark & ~_LASTMARK;
7466 		putback(stp, q, bp, pri);
7467 		bp = nbp;
7468 	}
7469 
7470 	/*
7471 	 * Set this flag so strrput will not generate signals. Need to
7472 	 * make sure this flag is cleared before leaving this routine
7473 	 * else signals will stop being sent.
7474 	 */
7475 	stp->sd_flag |= STRGETINPROG;
7476 	mutex_exit(&stp->sd_lock);
7477 
7478 	if ((stp->sd_rputdatafunc != NULL) && (DB_TYPE(bp) == M_DATA)) {
7479 		mblk_t *tmp, *prevmp;
7480 
7481 		/*
7482 		 * Put first non-data mblk back to stream head and
7483 		 * cut the mblk chain so sd_rputdatafunc only sees
7484 		 * M_DATA mblks. We can skip the first mblk since it
7485 		 * is M_DATA according to the condition above.
7486 		 */
7487 		for (prevmp = bp, tmp = bp->b_cont; tmp != NULL;
7488 		    prevmp = tmp, tmp = tmp->b_cont) {
7489 			if (DB_TYPE(tmp) != M_DATA) {
7490 				prevmp->b_cont = NULL;
7491 				mutex_enter(&stp->sd_lock);
7492 				putback(stp, q, tmp, tmp->b_band);
7493 				mutex_exit(&stp->sd_lock);
7494 				break;
7495 			}
7496 		}
7497 
7498 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
7499 		bp = (stp->sd_rputdatafunc)(stp->sd_vnode, bp,
7500 		    NULL, NULL, NULL, NULL);
7501 
7502 		if (bp == NULL)
7503 			goto retry;
7504 	}
7505 
7506 	if (STREAM_NEEDSERVICE(stp))
7507 		stream_runservice(stp);
7508 
7509 	/*
7510 	 * Set HIPRI flag if message is priority.
7511 	 */
7512 	if (type >= QPCTL)
7513 		flg = MSG_HIPRI;
7514 	else
7515 		flg = MSG_BAND;
7516 
7517 	/*
7518 	 * First process PROTO or PCPROTO blocks, if any.
7519 	 */
7520 	if (mctlp != NULL && type != M_DATA) {
7521 		mblk_t *nbp;
7522 
7523 		*mctlp = bp;
7524 		while (bp->b_cont && bp->b_cont->b_datap->db_type != M_DATA)
7525 			bp = bp->b_cont;
7526 		nbp = bp->b_cont;
7527 		bp->b_cont = NULL;
7528 		bp = nbp;
7529 	}
7530 
7531 	if (bp && bp->b_datap->db_type != M_DATA) {
7532 		/*
7533 		 * More PROTO blocks in msg. Will only happen if mctlp is NULL.
7534 		 */
7535 		more |= MORECTL;
7536 		savemp = bp;
7537 		while (bp && bp->b_datap->db_type != M_DATA) {
7538 			savemptail = bp;
7539 			bp = bp->b_cont;
7540 		}
7541 		savemptail->b_cont = NULL;
7542 	}
7543 
7544 	/*
7545 	 * Now process DATA blocks, if any.
7546 	 */
7547 	if (uiop == NULL) {
7548 		/* Append data to tail of mctlp */
7549 
7550 		ASSERT(bp == NULL || !(bp->b_datap->db_flags & DBLK_UIOA));
7551 		if (mctlp != NULL) {
7552 			mblk_t **mpp = mctlp;
7553 
7554 			while (*mpp != NULL)
7555 				mpp = &((*mpp)->b_cont);
7556 			*mpp = bp;
7557 			bp = NULL;
7558 		}
7559 	} else if (bp && (bp->b_datap->db_flags & DBLK_UIOA)) {
7560 		/*
7561 		 * A uioa mblk_t chain, as uio processing has already
7562 		 * been done we simple skip over processing.
7563 		 */
7564 		bp = NULL;
7565 		pr = 0;
7566 
7567 	} else if (uiop->uio_resid >= 0 && bp) {
7568 		size_t oldresid = uiop->uio_resid;
7569 
7570 		/*
7571 		 * If a streams message is likely to consist
7572 		 * of many small mblks, it is pulled up into
7573 		 * one continuous chunk of memory.
7574 		 * see longer comment at top of page
7575 		 * by mblk_pull_len declaration.
7576 		 */
7577 
7578 		if (MBLKL(bp) < mblk_pull_len) {
7579 			(void) pullupmsg(bp, -1);
7580 		}
7581 
7582 		bp = struiocopyout(bp, uiop, &error);
7583 		if (error != 0) {
7584 			if (mctlp != NULL) {
7585 				freemsg(*mctlp);
7586 				*mctlp = NULL;
7587 			} else
7588 				freemsg(savemp);
7589 			mutex_enter(&stp->sd_lock);
7590 			/*
7591 			 * clear stream head hi pri flag based on
7592 			 * first message
7593 			 */
7594 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7595 				ASSERT(type == M_PCPROTO);
7596 				stp->sd_flag &= ~STRPRI;
7597 			}
7598 			more = 0;
7599 			goto getmout;
7600 		}
7601 		/*
7602 		 * (pr == 1) indicates a partial read.
7603 		 */
7604 		if (oldresid > uiop->uio_resid)
7605 			pr = 1;
7606 	}
7607 
7608 	if (bp) {			/* more data blocks in msg */
7609 		more |= MOREDATA;
7610 		if (savemp)
7611 			savemptail->b_cont = bp;
7612 		else
7613 			savemp = bp;
7614 	}
7615 
7616 	mutex_enter(&stp->sd_lock);
7617 	if (savemp) {
7618 		if (flags & (MSG_IPEEK|MSG_DISCARDTAIL)) {
7619 			/*
7620 			 * When MSG_DISCARDTAIL is set or
7621 			 * when peeking discard any tail. When peeking this
7622 			 * is the tail of the dup that was copied out - the
7623 			 * message has already been putback on the queue.
7624 			 * Return MOREDATA to the caller even though the data
7625 			 * is discarded. This is used by sockets (to
7626 			 * set MSG_TRUNC).
7627 			 */
7628 			freemsg(savemp);
7629 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7630 				ASSERT(type == M_PCPROTO);
7631 				stp->sd_flag &= ~STRPRI;
7632 			}
7633 		} else if (pr && (savemp->b_datap->db_type == M_DATA) &&
7634 		    msgnodata(savemp)) {
7635 			/*
7636 			 * Avoid queuing a zero-length tail part of
7637 			 * a message. pr=1 indicates that we read some of
7638 			 * the message.
7639 			 */
7640 			freemsg(savemp);
7641 			more &= ~MOREDATA;
7642 			if (type >= QPCTL) {
7643 				ASSERT(type == M_PCPROTO);
7644 				stp->sd_flag &= ~STRPRI;
7645 			}
7646 		} else {
7647 			savemp->b_band = pri;
7648 			/*
7649 			 * If the first message was HIPRI and the one we're
7650 			 * putting back isn't, then clear STRPRI, otherwise
7651 			 * set STRPRI again.  Note that we must set STRPRI
7652 			 * again since the flush logic in strrput_nondata()
7653 			 * may have cleared it while we had sd_lock dropped.
7654 			 */
7655 
7656 			ASSERT(!(savemp->b_datap->db_flags & DBLK_UIOA));
7657 			if (type >= QPCTL) {
7658 				ASSERT(type == M_PCPROTO);
7659 				if (queclass(savemp) < QPCTL)
7660 					stp->sd_flag &= ~STRPRI;
7661 				else
7662 					stp->sd_flag |= STRPRI;
7663 			} else if (queclass(savemp) >= QPCTL) {
7664 				/*
7665 				 * The first message was not a HIPRI message,
7666 				 * but the one we are about to putback is.
7667 				 * For simplicitly, we do not allow for HIPRI
7668 				 * messages to be embedded in the message
7669 				 * body, so just force it to same type as
7670 				 * first message.
7671 				 */
7672 				ASSERT(type == M_DATA || type == M_PROTO);
7673 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
7674 				savemp->b_datap->db_type = type;
7675 			}
7676 			if (mark != 0) {
7677 				if ((mark & _LASTMARK) &&
7678 				    (stp->sd_mark == NULL)) {
7679 					/*
7680 					 * If another marked message arrived
7681 					 * while sd_lock was not held sd_mark
7682 					 * would be non-NULL.
7683 					 */
7684 					stp->sd_mark = savemp;
7685 				}
7686 				savemp->b_flag |= mark & ~_LASTMARK;
7687 			}
7688 			putback(stp, q, savemp, pri);
7689 		}
7690 	} else if (!(flags & MSG_IPEEK)) {
7691 		/*
7692 		 * The complete message was consumed.
7693 		 *
7694 		 * If another M_PCPROTO arrived while sd_lock was not held
7695 		 * it would have been discarded since STRPRI was still set.
7696 		 *
7697 		 * Move the MSG*MARKNEXT information
7698 		 * to the stream head just in case
7699 		 * the read queue becomes empty.
7700 		 * clear stream head hi pri flag based on
7701 		 * first message
7702 		 *
7703 		 * If the stream head was at the mark
7704 		 * (STRATMARK) before we dropped sd_lock above
7705 		 * and some data was consumed then we have
7706 		 * moved past the mark thus STRATMARK is
7707 		 * cleared. However, if a message arrived in
7708 		 * strrput during the copyout above causing
7709 		 * STRATMARK to be set we can not clear that
7710 		 * flag.
7711 		 * XXX A "perimeter" would help by single-threading strrput,
7712 		 * strread, strgetmsg and kstrgetmsg.
7713 		 */
7714 		if (type >= QPCTL) {
7715 			ASSERT(type == M_PCPROTO);
7716 			stp->sd_flag &= ~STRPRI;
7717 		}
7718 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
7719 			if (mark & MSGMARKNEXT) {
7720 				stp->sd_flag &= ~STRNOTATMARK;
7721 				stp->sd_flag |= STRATMARK;
7722 			} else if (mark & MSGNOTMARKNEXT) {
7723 				stp->sd_flag &= ~STRATMARK;
7724 				stp->sd_flag |= STRNOTATMARK;
7725 			} else {
7726 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
7727 			}
7728 		} else if (pr && (old_sd_flag & STRATMARK)) {
7729 			stp->sd_flag &= ~STRATMARK;
7730 		}
7731 	}
7732 
7733 	*flagsp = flg;
7734 	*prip = pri;
7735 
7736 	/*
7737 	 * Getmsg cleanup processing - if the state of the queue has changed
7738 	 * some signals may need to be sent and/or poll awakened.
7739 	 */
7740 getmout:
7741 	qbackenable(q, pri);
7742 
7743 	/*
7744 	 * We dropped the stream head lock above. Send all M_SIG messages
7745 	 * before processing stream head for SIGPOLL messages.
7746 	 */
7747 	ASSERT(MUTEX_HELD(&stp->sd_lock));
7748 	while ((bp = q->q_first) != NULL &&
7749 	    (bp->b_datap->db_type == M_SIG)) {
7750 		/*
7751 		 * sd_lock is held so the content of the read queue can not
7752 		 * change.
7753 		 */
7754 		bp = getq(q);
7755 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7756 
7757 		strsignal_nolock(stp, *bp->b_rptr, (int32_t)bp->b_band);
7758 		mutex_exit(&stp->sd_lock);
7759 		freemsg(bp);
7760 		if (STREAM_NEEDSERVICE(stp))
7761 			stream_runservice(stp);
7762 		mutex_enter(&stp->sd_lock);
7763 	}
7764 
7765 	/*
7766 	 * stream head cannot change while we make the determination
7767 	 * whether or not to send a signal. Drop the flag to allow strrput
7768 	 * to send firstmsgsigs again.
7769 	 */
7770 	stp->sd_flag &= ~STRGETINPROG;
7771 
7772 	/*
7773 	 * If the type of message at the front of the queue changed
7774 	 * due to the receive the appropriate signals and pollwakeup events
7775 	 * are generated. The type of changes are:
7776 	 *	Processed a hipri message, q_first is not hipri.
7777 	 *	Processed a band X message, and q_first is band Y.
7778 	 * The generated signals and pollwakeups are identical to what
7779 	 * strrput() generates should the message that is now on q_first
7780 	 * arrive to an empty read queue.
7781 	 *
7782 	 * Note: only strrput will send a signal for a hipri message.
7783 	 */
7784 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7785 		strsigset_t signals = 0;
7786 		strpollset_t pollwakeups = 0;
7787 
7788 		if (flg & MSG_HIPRI) {
7789 			/*
7790 			 * Removed a hipri message. Regular data at
7791 			 * the front of  the queue.
7792 			 */
7793 			if (bp->b_band == 0) {
7794 				signals = S_INPUT | S_RDNORM;
7795 				pollwakeups = POLLIN | POLLRDNORM;
7796 			} else {
7797 				signals = S_INPUT | S_RDBAND;
7798 				pollwakeups = POLLIN | POLLRDBAND;
7799 			}
7800 		} else if (pri != bp->b_band) {
7801 			/*
7802 			 * The band is different for the new q_first.
7803 			 */
7804 			if (bp->b_band == 0) {
7805 				signals = S_RDNORM;
7806 				pollwakeups = POLLIN | POLLRDNORM;
7807 			} else {
7808 				signals = S_RDBAND;
7809 				pollwakeups = POLLIN | POLLRDBAND;
7810 			}
7811 		}
7812 
7813 		if (pollwakeups != 0) {
7814 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
7815 				if (!(stp->sd_rput_opt & SR_POLLIN))
7816 					goto no_pollwake;
7817 				stp->sd_rput_opt &= ~SR_POLLIN;
7818 			}
7819 			mutex_exit(&stp->sd_lock);
7820 			pollwakeup(&stp->sd_pollist, pollwakeups);
7821 			mutex_enter(&stp->sd_lock);
7822 		}
7823 no_pollwake:
7824 
7825 		if (stp->sd_sigflags & signals)
7826 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7827 	}
7828 	mutex_exit(&stp->sd_lock);
7829 
7830 	rvp->r_val1 = more;
7831 	return (error);
7832 #undef	_LASTMARK
7833 }
7834 
7835 /*
7836  * Put a message downstream.
7837  *
7838  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7839  */
7840 int
7841 strputmsg(
7842 	struct vnode *vp,
7843 	struct strbuf *mctl,
7844 	struct strbuf *mdata,
7845 	unsigned char pri,
7846 	int flag,
7847 	int fmode)
7848 {
7849 	struct stdata *stp;
7850 	queue_t *wqp;
7851 	mblk_t *mp;
7852 	ssize_t msgsize;
7853 	ssize_t rmin, rmax;
7854 	int error;
7855 	struct uio uios;
7856 	struct uio *uiop = &uios;
7857 	struct iovec iovs;
7858 	int xpg4 = 0;
7859 
7860 	ASSERT(vp->v_stream);
7861 	stp = vp->v_stream;
7862 	wqp = stp->sd_wrq;
7863 
7864 	/*
7865 	 * If it is an XPG4 application, we need to send
7866 	 * SIGPIPE below
7867 	 */
7868 
7869 	xpg4 = (flag & MSG_XPG4) ? 1 : 0;
7870 	flag &= ~MSG_XPG4;
7871 
7872 	if (audit_active)
7873 		audit_strputmsg(vp, mctl, mdata, pri, flag, fmode);
7874 
7875 	mutex_enter(&stp->sd_lock);
7876 
7877 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
7878 		mutex_exit(&stp->sd_lock);
7879 		return (error);
7880 	}
7881 
7882 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7883 		error = strwriteable(stp, B_FALSE, xpg4);
7884 		if (error != 0) {
7885 			mutex_exit(&stp->sd_lock);
7886 			return (error);
7887 		}
7888 	}
7889 
7890 	mutex_exit(&stp->sd_lock);
7891 
7892 	/*
7893 	 * Check for legal flag value.
7894 	 */
7895 	switch (flag) {
7896 	case MSG_HIPRI:
7897 		if ((mctl->len < 0) || (pri != 0))
7898 			return (EINVAL);
7899 		break;
7900 	case MSG_BAND:
7901 		break;
7902 
7903 	default:
7904 		return (EINVAL);
7905 	}
7906 
7907 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_IN,
7908 	    "strputmsg in:stp %p", stp);
7909 
7910 	/* get these values from those cached in the stream head */
7911 	rmin = stp->sd_qn_minpsz;
7912 	rmax = stp->sd_qn_maxpsz;
7913 
7914 	/*
7915 	 * Make sure ctl and data sizes together fall within the
7916 	 * limits of the max and min receive packet sizes and do
7917 	 * not exceed system limit.
7918 	 */
7919 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
7920 	if (rmax == 0) {
7921 		return (ERANGE);
7922 	}
7923 	/*
7924 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
7925 	 * Needed to prevent partial failures in the strmakedata loop.
7926 	 */
7927 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
7928 		rmax = stp->sd_maxblk;
7929 
7930 	if ((msgsize = mdata->len) < 0) {
7931 		msgsize = 0;
7932 		rmin = 0;	/* no range check for NULL data part */
7933 	}
7934 	if ((msgsize < rmin) ||
7935 	    ((msgsize > rmax) && (rmax != INFPSZ)) ||
7936 	    (mctl->len > strctlsz)) {
7937 		return (ERANGE);
7938 	}
7939 
7940 	/*
7941 	 * Setup uio and iov for data part
7942 	 */
7943 	iovs.iov_base = mdata->buf;
7944 	iovs.iov_len = msgsize;
7945 	uios.uio_iov = &iovs;
7946 	uios.uio_iovcnt = 1;
7947 	uios.uio_loffset = 0;
7948 	uios.uio_segflg = UIO_USERSPACE;
7949 	uios.uio_fmode = fmode;
7950 	uios.uio_extflg = UIO_COPY_DEFAULT;
7951 	uios.uio_resid = msgsize;
7952 	uios.uio_offset = 0;
7953 
7954 	/* Ignore flow control in strput for HIPRI */
7955 	if (flag & MSG_HIPRI)
7956 		flag |= MSG_IGNFLOW;
7957 
7958 	for (;;) {
7959 		int done = 0;
7960 
7961 		/*
7962 		 * strput will always free the ctl mblk - even when strput
7963 		 * fails.
7964 		 */
7965 		if ((error = strmakectl(mctl, flag, fmode, &mp)) != 0) {
7966 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7967 			    "strputmsg out:stp %p out %d error %d",
7968 			    stp, 1, error);
7969 			return (error);
7970 		}
7971 		/*
7972 		 * Verify that the whole message can be transferred by
7973 		 * strput.
7974 		 */
7975 		ASSERT(stp->sd_maxblk == INFPSZ ||
7976 		    stp->sd_maxblk >= mdata->len);
7977 
7978 		msgsize = mdata->len;
7979 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
7980 		mdata->len = msgsize;
7981 
7982 		if (error == 0)
7983 			break;
7984 
7985 		if (error != EWOULDBLOCK)
7986 			goto out;
7987 
7988 		mutex_enter(&stp->sd_lock);
7989 		/*
7990 		 * Check for a missed wakeup.
7991 		 * Needed since strput did not hold sd_lock across
7992 		 * the canputnext.
7993 		 */
7994 		if (bcanputnext(wqp, pri)) {
7995 			/* Try again */
7996 			mutex_exit(&stp->sd_lock);
7997 			continue;
7998 		}
7999 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAIT,
8000 		    "strputmsg wait:stp %p waits pri %d", stp, pri);
8001 		if (((error = strwaitq(stp, WRITEWAIT, (ssize_t)0, fmode, -1,
8002 		    &done)) != 0) || done) {
8003 			mutex_exit(&stp->sd_lock);
8004 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
8005 			    "strputmsg out:q %p out %d error %d",
8006 			    stp, 0, error);
8007 			return (error);
8008 		}
8009 		TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAKE,
8010 		    "strputmsg wake:stp %p wakes", stp);
8011 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
8012 			mutex_exit(&stp->sd_lock);
8013 			return (error);
8014 		}
8015 		mutex_exit(&stp->sd_lock);
8016 	}
8017 out:
8018 	/*
8019 	 * For historic reasons, applications expect EAGAIN
8020 	 * when data mblk could not be allocated. so change
8021 	 * ENOMEM back to EAGAIN
8022 	 */
8023 	if (error == ENOMEM)
8024 		error = EAGAIN;
8025 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
8026 	    "strputmsg out:stp %p out %d error %d", stp, 2, error);
8027 	return (error);
8028 }
8029 
8030 /*
8031  * Put a message downstream.
8032  * Can send only an M_PROTO/M_PCPROTO by passing in a NULL uiop.
8033  * The fmode flag (NDELAY, NONBLOCK) is the or of the flags in the uio
8034  * and the fmode parameter.
8035  *
8036  * This routine handles the consolidation private flags:
8037  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
8038  *	MSG_HOLDSIG	Hold signals while waiting for data.
8039  *	MSG_IGNFLOW	Don't check streams flow control.
8040  *
8041  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
8042  */
8043 int
8044 kstrputmsg(
8045 	struct vnode *vp,
8046 	mblk_t *mctl,
8047 	struct uio *uiop,
8048 	ssize_t msgsize,
8049 	unsigned char pri,
8050 	int flag,
8051 	int fmode)
8052 {
8053 	struct stdata *stp;
8054 	queue_t *wqp;
8055 	ssize_t rmin, rmax;
8056 	int error;
8057 
8058 	ASSERT(vp->v_stream);
8059 	stp = vp->v_stream;
8060 	wqp = stp->sd_wrq;
8061 	if (audit_active)
8062 		audit_strputmsg(vp, NULL, NULL, pri, flag, fmode);
8063 	if (mctl == NULL)
8064 		return (EINVAL);
8065 
8066 	mutex_enter(&stp->sd_lock);
8067 
8068 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
8069 		mutex_exit(&stp->sd_lock);
8070 		freemsg(mctl);
8071 		return (error);
8072 	}
8073 
8074 	if ((stp->sd_flag & STPLEX) || !(flag & MSG_IGNERROR)) {
8075 		if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
8076 			error = strwriteable(stp, B_FALSE, B_TRUE);
8077 			if (error != 0) {
8078 				mutex_exit(&stp->sd_lock);
8079 				freemsg(mctl);
8080 				return (error);
8081 			}
8082 		}
8083 	}
8084 
8085 	mutex_exit(&stp->sd_lock);
8086 
8087 	/*
8088 	 * Check for legal flag value.
8089 	 */
8090 	switch (flag & (MSG_HIPRI|MSG_BAND|MSG_ANY)) {
8091 	case MSG_HIPRI:
8092 		if (pri != 0) {
8093 			freemsg(mctl);
8094 			return (EINVAL);
8095 		}
8096 		break;
8097 	case MSG_BAND:
8098 		break;
8099 	default:
8100 		freemsg(mctl);
8101 		return (EINVAL);
8102 	}
8103 
8104 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_IN,
8105 	    "kstrputmsg in:stp %p", stp);
8106 
8107 	/* get these values from those cached in the stream head */
8108 	rmin = stp->sd_qn_minpsz;
8109 	rmax = stp->sd_qn_maxpsz;
8110 
8111 	/*
8112 	 * Make sure ctl and data sizes together fall within the
8113 	 * limits of the max and min receive packet sizes and do
8114 	 * not exceed system limit.
8115 	 */
8116 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
8117 	if (rmax == 0) {
8118 		freemsg(mctl);
8119 		return (ERANGE);
8120 	}
8121 	/*
8122 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
8123 	 * Needed to prevent partial failures in the strmakedata loop.
8124 	 */
8125 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
8126 		rmax = stp->sd_maxblk;
8127 
8128 	if (uiop == NULL) {
8129 		msgsize = -1;
8130 		rmin = -1;	/* no range check for NULL data part */
8131 	} else {
8132 		/* Use uio flags as well as the fmode parameter flags */
8133 		fmode |= uiop->uio_fmode;
8134 
8135 		if ((msgsize < rmin) ||
8136 		    ((msgsize > rmax) && (rmax != INFPSZ))) {
8137 			freemsg(mctl);
8138 			return (ERANGE);
8139 		}
8140 	}
8141 
8142 	/* Ignore flow control in strput for HIPRI */
8143 	if (flag & MSG_HIPRI)
8144 		flag |= MSG_IGNFLOW;
8145 
8146 	for (;;) {
8147 		int done = 0;
8148 		int waitflag;
8149 		mblk_t *mp;
8150 
8151 		/*
8152 		 * strput will always free the ctl mblk - even when strput
8153 		 * fails. If MSG_IGNFLOW is set then any error returned
8154 		 * will cause us to break the loop, so we don't need a copy
8155 		 * of the message. If MSG_IGNFLOW is not set, then we can
8156 		 * get hit by flow control and be forced to try again. In
8157 		 * this case we need to have a copy of the message. We
8158 		 * do this using copymsg since the message may get modified
8159 		 * by something below us.
8160 		 *
8161 		 * We've observed that many TPI providers do not check db_ref
8162 		 * on the control messages but blindly reuse them for the
8163 		 * T_OK_ACK/T_ERROR_ACK. Thus using copymsg is more
8164 		 * friendly to such providers than using dupmsg. Also, note
8165 		 * that sockfs uses MSG_IGNFLOW for all TPI control messages.
8166 		 * Only data messages are subject to flow control, hence
8167 		 * subject to this copymsg.
8168 		 */
8169 		if (flag & MSG_IGNFLOW) {
8170 			mp = mctl;
8171 			mctl = NULL;
8172 		} else {
8173 			do {
8174 				/*
8175 				 * If a message has a free pointer, the message
8176 				 * must be dupmsg to maintain this pointer.
8177 				 * Code using this facility must be sure
8178 				 * that modules below will not change the
8179 				 * contents of the dblk without checking db_ref
8180 				 * first. If db_ref is > 1, then the module
8181 				 * needs to do a copymsg first. Otherwise,
8182 				 * the contents of the dblk may become
8183 				 * inconsistent because the freesmg/freeb below
8184 				 * may end up calling atomic_add_32_nv.
8185 				 * The atomic_add_32_nv in freeb (accessing
8186 				 * all of db_ref, db_type, db_flags, and
8187 				 * db_struioflag) does not prevent other threads
8188 				 * from concurrently trying to modify e.g.
8189 				 * db_type.
8190 				 */
8191 				if (mctl->b_datap->db_frtnp != NULL)
8192 					mp = dupmsg(mctl);
8193 				else
8194 					mp = copymsg(mctl);
8195 
8196 				if (mp != NULL)
8197 					break;
8198 
8199 				error = strwaitbuf(msgdsize(mctl), BPRI_MED);
8200 				if (error) {
8201 					freemsg(mctl);
8202 					return (error);
8203 				}
8204 			} while (mp == NULL);
8205 		}
8206 		/*
8207 		 * Verify that all of msgsize can be transferred by
8208 		 * strput.
8209 		 */
8210 		ASSERT(stp->sd_maxblk == INFPSZ || stp->sd_maxblk >= msgsize);
8211 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
8212 		if (error == 0)
8213 			break;
8214 
8215 		if (error != EWOULDBLOCK)
8216 			goto out;
8217 
8218 		/*
8219 		 * IF MSG_IGNFLOW is set we should have broken out of loop
8220 		 * above.
8221 		 */
8222 		ASSERT(!(flag & MSG_IGNFLOW));
8223 		mutex_enter(&stp->sd_lock);
8224 		/*
8225 		 * Check for a missed wakeup.
8226 		 * Needed since strput did not hold sd_lock across
8227 		 * the canputnext.
8228 		 */
8229 		if (bcanputnext(wqp, pri)) {
8230 			/* Try again */
8231 			mutex_exit(&stp->sd_lock);
8232 			continue;
8233 		}
8234 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAIT,
8235 		    "kstrputmsg wait:stp %p waits pri %d", stp, pri);
8236 
8237 		waitflag = WRITEWAIT;
8238 		if (flag & (MSG_HOLDSIG|MSG_IGNERROR)) {
8239 			if (flag & MSG_HOLDSIG)
8240 				waitflag |= STR_NOSIG;
8241 			if (flag & MSG_IGNERROR)
8242 				waitflag |= STR_NOERROR;
8243 		}
8244 		if (((error = strwaitq(stp, waitflag,
8245 		    (ssize_t)0, fmode, -1, &done)) != 0) || done) {
8246 			mutex_exit(&stp->sd_lock);
8247 			TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8248 			    "kstrputmsg out:stp %p out %d error %d",
8249 			    stp, 0, error);
8250 			freemsg(mctl);
8251 			return (error);
8252 		}
8253 		TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAKE,
8254 		    "kstrputmsg wake:stp %p wakes", stp);
8255 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
8256 			mutex_exit(&stp->sd_lock);
8257 			freemsg(mctl);
8258 			return (error);
8259 		}
8260 		mutex_exit(&stp->sd_lock);
8261 	}
8262 out:
8263 	freemsg(mctl);
8264 	/*
8265 	 * For historic reasons, applications expect EAGAIN
8266 	 * when data mblk could not be allocated. so change
8267 	 * ENOMEM back to EAGAIN
8268 	 */
8269 	if (error == ENOMEM)
8270 		error = EAGAIN;
8271 	TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8272 	    "kstrputmsg out:stp %p out %d error %d", stp, 2, error);
8273 	return (error);
8274 }
8275 
8276 /*
8277  * Determines whether the necessary conditions are set on a stream
8278  * for it to be readable, writeable, or have exceptions.
8279  *
8280  * strpoll handles the consolidation private events:
8281  *	POLLNOERR	Do not return POLLERR even if there are stream
8282  *			head errors.
8283  *			Used by sockfs.
8284  *	POLLRDDATA	Do not return POLLIN unless at least one message on
8285  *			the queue contains one or more M_DATA mblks. Thus
8286  *			when this flag is set a queue with only
8287  *			M_PROTO/M_PCPROTO mblks does not return POLLIN.
8288  *			Used by sockfs to ignore T_EXDATA_IND messages.
8289  *
8290  * Note: POLLRDDATA assumes that synch streams only return messages with
8291  * an M_DATA attached (i.e. not messages consisting of only
8292  * an M_PROTO/M_PCPROTO part).
8293  */
8294 int
8295 strpoll(
8296 	struct stdata *stp,
8297 	short events_arg,
8298 	int anyyet,
8299 	short *reventsp,
8300 	struct pollhead **phpp)
8301 {
8302 	int events = (ushort_t)events_arg;
8303 	int retevents = 0;
8304 	mblk_t *mp;
8305 	qband_t *qbp;
8306 	long sd_flags = stp->sd_flag;
8307 	int headlocked = 0;
8308 
8309 	/*
8310 	 * For performance, a single 'if' tests for most possible edge
8311 	 * conditions in one shot
8312 	 */
8313 	if (sd_flags & (STPLEX | STRDERR | STWRERR)) {
8314 		if (sd_flags & STPLEX) {
8315 			*reventsp = POLLNVAL;
8316 			return (EINVAL);
8317 		}
8318 		if (((events & (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) &&
8319 		    (sd_flags & STRDERR)) ||
8320 		    ((events & (POLLOUT | POLLWRNORM | POLLWRBAND)) &&
8321 		    (sd_flags & STWRERR))) {
8322 			if (!(events & POLLNOERR)) {
8323 				*reventsp = POLLERR;
8324 				return (0);
8325 			}
8326 		}
8327 	}
8328 	if (sd_flags & STRHUP) {
8329 		retevents |= POLLHUP;
8330 	} else if (events & (POLLWRNORM | POLLWRBAND)) {
8331 		queue_t *tq;
8332 		queue_t	*qp = stp->sd_wrq;
8333 
8334 		claimstr(qp);
8335 		/* Find next module forward that has a service procedure */
8336 		tq = qp->q_next->q_nfsrv;
8337 		ASSERT(tq != NULL);
8338 
8339 		polllock(&stp->sd_pollist, QLOCK(tq));
8340 		if (events & POLLWRNORM) {
8341 			queue_t *sqp;
8342 
8343 			if (tq->q_flag & QFULL)
8344 				/* ensure backq svc procedure runs */
8345 				tq->q_flag |= QWANTW;
8346 			else if ((sqp = stp->sd_struiowrq) != NULL) {
8347 				/* Check sync stream barrier write q */
8348 				mutex_exit(QLOCK(tq));
8349 				polllock(&stp->sd_pollist, QLOCK(sqp));
8350 				if (sqp->q_flag & QFULL)
8351 					/* ensure pollwakeup() is done */
8352 					sqp->q_flag |= QWANTWSYNC;
8353 				else
8354 					retevents |= POLLOUT;
8355 				/* More write events to process ??? */
8356 				if (! (events & POLLWRBAND)) {
8357 					mutex_exit(QLOCK(sqp));
8358 					releasestr(qp);
8359 					goto chkrd;
8360 				}
8361 				mutex_exit(QLOCK(sqp));
8362 				polllock(&stp->sd_pollist, QLOCK(tq));
8363 			} else
8364 				retevents |= POLLOUT;
8365 		}
8366 		if (events & POLLWRBAND) {
8367 			qbp = tq->q_bandp;
8368 			if (qbp) {
8369 				while (qbp) {
8370 					if (qbp->qb_flag & QB_FULL)
8371 						qbp->qb_flag |= QB_WANTW;
8372 					else
8373 						retevents |= POLLWRBAND;
8374 					qbp = qbp->qb_next;
8375 				}
8376 			} else {
8377 				retevents |= POLLWRBAND;
8378 			}
8379 		}
8380 		mutex_exit(QLOCK(tq));
8381 		releasestr(qp);
8382 	}
8383 chkrd:
8384 	if (sd_flags & STRPRI) {
8385 		retevents |= (events & POLLPRI);
8386 	} else if (events & (POLLRDNORM | POLLRDBAND | POLLIN)) {
8387 		queue_t	*qp = _RD(stp->sd_wrq);
8388 		int normevents = (events & (POLLIN | POLLRDNORM));
8389 
8390 		/*
8391 		 * Note: Need to do polllock() here since ps_lock may be
8392 		 * held. See bug 4191544.
8393 		 */
8394 		polllock(&stp->sd_pollist, &stp->sd_lock);
8395 		headlocked = 1;
8396 		mp = qp->q_first;
8397 		while (mp) {
8398 			/*
8399 			 * For POLLRDDATA we scan b_cont and b_next until we
8400 			 * find an M_DATA.
8401 			 */
8402 			if ((events & POLLRDDATA) &&
8403 			    mp->b_datap->db_type != M_DATA) {
8404 				mblk_t *nmp = mp->b_cont;
8405 
8406 				while (nmp != NULL &&
8407 				    nmp->b_datap->db_type != M_DATA)
8408 					nmp = nmp->b_cont;
8409 				if (nmp == NULL) {
8410 					mp = mp->b_next;
8411 					continue;
8412 				}
8413 			}
8414 			if (mp->b_band == 0)
8415 				retevents |= normevents;
8416 			else
8417 				retevents |= (events & (POLLIN | POLLRDBAND));
8418 			break;
8419 		}
8420 		if (! (retevents & normevents) &&
8421 		    (stp->sd_wakeq & RSLEEP)) {
8422 			/*
8423 			 * Sync stream barrier read queue has data.
8424 			 */
8425 			retevents |= normevents;
8426 		}
8427 		/* Treat eof as normal data */
8428 		if (sd_flags & STREOF)
8429 			retevents |= normevents;
8430 	}
8431 
8432 	*reventsp = (short)retevents;
8433 	if (retevents) {
8434 		if (headlocked)
8435 			mutex_exit(&stp->sd_lock);
8436 		return (0);
8437 	}
8438 
8439 	/*
8440 	 * If poll() has not found any events yet, set up event cell
8441 	 * to wake up the poll if a requested event occurs on this
8442 	 * stream.  Check for collisions with outstanding poll requests.
8443 	 */
8444 	if (!anyyet) {
8445 		*phpp = &stp->sd_pollist;
8446 		if (headlocked == 0) {
8447 			polllock(&stp->sd_pollist, &stp->sd_lock);
8448 			headlocked = 1;
8449 		}
8450 		stp->sd_rput_opt |= SR_POLLIN;
8451 	}
8452 	if (headlocked)
8453 		mutex_exit(&stp->sd_lock);
8454 	return (0);
8455 }
8456 
8457 /*
8458  * The purpose of putback() is to assure sleeping polls/reads
8459  * are awakened when there are no new messages arriving at the,
8460  * stream head, and a message is placed back on the read queue.
8461  *
8462  * sd_lock must be held when messages are placed back on stream
8463  * head.  (getq() holds sd_lock when it removes messages from
8464  * the queue)
8465  */
8466 
8467 static void
8468 putback(struct stdata *stp, queue_t *q, mblk_t *bp, int band)
8469 {
8470 	mblk_t	*qfirst = q->q_first;
8471 	ASSERT(MUTEX_HELD(&stp->sd_lock));
8472 
8473 	if ((stp->sd_rput_opt & SR_CONSOL_DATA) &&
8474 	    (qfirst != NULL) &&
8475 	    (qfirst->b_datap->db_type == M_DATA) &&
8476 	    ((qfirst->b_flag & (MSGMARK|MSGDELIM)) == 0)) {
8477 		/*
8478 		 * We use the same logic as defined in strrput()
8479 		 * but in reverse as we are putting back onto the
8480 		 * queue and want to retain byte ordering.
8481 		 * Consolidate an M_DATA message onto an M_DATA,
8482 		 * M_PROTO, or M_PCPROTO by merging it with q_first.
8483 		 * The consolidation does not take place if the message
8484 		 * we are returning to the queue is marked with either
8485 		 * of the marks or the delim flag or if q_first
8486 		 * is marked with MSGMARK. The MSGMARK check is needed to
8487 		 * handle the odd semantics of MSGMARK where essentially
8488 		 * the whole message is to be treated as marked.
8489 		 * Carry any MSGMARKNEXT and MSGNOTMARKNEXT from q_first
8490 		 * to the front of the b_cont chain.
8491 		 */
8492 		unsigned char db_type = bp->b_datap->db_type;
8493 
8494 		if ((db_type == M_DATA || db_type == M_PROTO ||
8495 		    db_type == M_PCPROTO) &&
8496 		    !(bp->b_flag & (MSGMARK|MSGDELIM|MSGMARKNEXT))) {
8497 			rmvq_noenab(q, qfirst);
8498 			/*
8499 			 * The first message in the b_cont list
8500 			 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
8501 			 * We need to handle the case where we
8502 			 * are appending:
8503 			 *
8504 			 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
8505 			 * 2) a MSGMARKNEXT to a plain message.
8506 			 * 3) a MSGNOTMARKNEXT to a plain message
8507 			 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
8508 			 *    message.
8509 			 *
8510 			 * Thus we never append a MSGMARKNEXT or
8511 			 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
8512 			 */
8513 			if (qfirst->b_flag & MSGMARKNEXT) {
8514 				bp->b_flag |= MSGMARKNEXT;
8515 				bp->b_flag &= ~MSGNOTMARKNEXT;
8516 				qfirst->b_flag &= ~MSGMARKNEXT;
8517 			} else if (qfirst->b_flag & MSGNOTMARKNEXT) {
8518 				bp->b_flag |= MSGNOTMARKNEXT;
8519 				qfirst->b_flag &= ~MSGNOTMARKNEXT;
8520 			}
8521 
8522 			linkb(bp, qfirst);
8523 		}
8524 	}
8525 	(void) putbq(q, bp);
8526 
8527 	/*
8528 	 * A message may have come in when the sd_lock was dropped in the
8529 	 * calling routine. If this is the case and STR*ATMARK info was
8530 	 * received, need to move that from the stream head to the q_last
8531 	 * so that SIOCATMARK can return the proper value.
8532 	 */
8533 	if (stp->sd_flag & (STRATMARK | STRNOTATMARK)) {
8534 		unsigned short *flagp = &q->q_last->b_flag;
8535 		uint_t b_flag = (uint_t)*flagp;
8536 
8537 		if (stp->sd_flag & STRATMARK) {
8538 			b_flag &= ~MSGNOTMARKNEXT;
8539 			b_flag |= MSGMARKNEXT;
8540 			stp->sd_flag &= ~STRATMARK;
8541 		} else {
8542 			b_flag &= ~MSGMARKNEXT;
8543 			b_flag |= MSGNOTMARKNEXT;
8544 			stp->sd_flag &= ~STRNOTATMARK;
8545 		}
8546 		*flagp = (unsigned short) b_flag;
8547 	}
8548 
8549 #ifdef	DEBUG
8550 	/*
8551 	 * Make sure that the flags are not messed up.
8552 	 */
8553 	{
8554 		mblk_t *mp;
8555 		mp = q->q_last;
8556 		while (mp != NULL) {
8557 			ASSERT((mp->b_flag & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
8558 			    (MSGMARKNEXT|MSGNOTMARKNEXT));
8559 			mp = mp->b_cont;
8560 		}
8561 	}
8562 #endif
8563 	if (q->q_first == bp) {
8564 		short pollevents;
8565 
8566 		if (stp->sd_flag & RSLEEP) {
8567 			stp->sd_flag &= ~RSLEEP;
8568 			cv_broadcast(&q->q_wait);
8569 		}
8570 		if (stp->sd_flag & STRPRI) {
8571 			pollevents = POLLPRI;
8572 		} else {
8573 			if (band == 0) {
8574 				if (!(stp->sd_rput_opt & SR_POLLIN))
8575 					return;
8576 				stp->sd_rput_opt &= ~SR_POLLIN;
8577 				pollevents = POLLIN | POLLRDNORM;
8578 			} else {
8579 				pollevents = POLLIN | POLLRDBAND;
8580 			}
8581 		}
8582 		mutex_exit(&stp->sd_lock);
8583 		pollwakeup(&stp->sd_pollist, pollevents);
8584 		mutex_enter(&stp->sd_lock);
8585 	}
8586 }
8587 
8588 /*
8589  * Return the held vnode attached to the stream head of a
8590  * given queue
8591  * It is the responsibility of the calling routine to ensure
8592  * that the queue does not go away (e.g. pop).
8593  */
8594 vnode_t *
8595 strq2vp(queue_t *qp)
8596 {
8597 	vnode_t *vp;
8598 	vp = STREAM(qp)->sd_vnode;
8599 	ASSERT(vp != NULL);
8600 	VN_HOLD(vp);
8601 	return (vp);
8602 }
8603 
8604 /*
8605  * return the stream head write queue for the given vp
8606  * It is the responsibility of the calling routine to ensure
8607  * that the stream or vnode do not close.
8608  */
8609 queue_t *
8610 strvp2wq(vnode_t *vp)
8611 {
8612 	ASSERT(vp->v_stream != NULL);
8613 	return (vp->v_stream->sd_wrq);
8614 }
8615 
8616 /*
8617  * pollwakeup stream head
8618  * It is the responsibility of the calling routine to ensure
8619  * that the stream or vnode do not close.
8620  */
8621 void
8622 strpollwakeup(vnode_t *vp, short event)
8623 {
8624 	ASSERT(vp->v_stream);
8625 	pollwakeup(&vp->v_stream->sd_pollist, event);
8626 }
8627 
8628 /*
8629  * Mate the stream heads of two vnodes together. If the two vnodes are the
8630  * same, we just make the write-side point at the read-side -- otherwise,
8631  * we do a full mate.  Only works on vnodes associated with streams that are
8632  * still being built and thus have only a stream head.
8633  */
8634 void
8635 strmate(vnode_t *vp1, vnode_t *vp2)
8636 {
8637 	queue_t *wrq1 = strvp2wq(vp1);
8638 	queue_t *wrq2 = strvp2wq(vp2);
8639 
8640 	/*
8641 	 * Verify that there are no modules on the stream yet.  We also
8642 	 * rely on the stream head always having a service procedure to
8643 	 * avoid tweaking q_nfsrv.
8644 	 */
8645 	ASSERT(wrq1->q_next == NULL && wrq2->q_next == NULL);
8646 	ASSERT(wrq1->q_qinfo->qi_srvp != NULL);
8647 	ASSERT(wrq2->q_qinfo->qi_srvp != NULL);
8648 
8649 	/*
8650 	 * If the queues are the same, just twist; otherwise do a full mate.
8651 	 */
8652 	if (wrq1 == wrq2) {
8653 		wrq1->q_next = _RD(wrq1);
8654 	} else {
8655 		wrq1->q_next = _RD(wrq2);
8656 		wrq2->q_next = _RD(wrq1);
8657 		STREAM(wrq1)->sd_mate = STREAM(wrq2);
8658 		STREAM(wrq1)->sd_flag |= STRMATE;
8659 		STREAM(wrq2)->sd_mate = STREAM(wrq1);
8660 		STREAM(wrq2)->sd_flag |= STRMATE;
8661 	}
8662 }
8663 
8664 /*
8665  * XXX will go away when console is correctly fixed.
8666  * Clean up the console PIDS, from previous I_SETSIG,
8667  * called only for cnopen which never calls strclean().
8668  */
8669 void
8670 str_cn_clean(struct vnode *vp)
8671 {
8672 	strsig_t *ssp, *pssp, *tssp;
8673 	struct stdata *stp;
8674 	struct pid  *pidp;
8675 	int update = 0;
8676 
8677 	ASSERT(vp->v_stream);
8678 	stp = vp->v_stream;
8679 	pssp = NULL;
8680 	mutex_enter(&stp->sd_lock);
8681 	ssp = stp->sd_siglist;
8682 	while (ssp) {
8683 		mutex_enter(&pidlock);
8684 		pidp = ssp->ss_pidp;
8685 		/*
8686 		 * Get rid of PID if the proc is gone.
8687 		 */
8688 		if (pidp->pid_prinactive) {
8689 			tssp = ssp->ss_next;
8690 			if (pssp)
8691 				pssp->ss_next = tssp;
8692 			else
8693 				stp->sd_siglist = tssp;
8694 			ASSERT(pidp->pid_ref <= 1);
8695 			PID_RELE(ssp->ss_pidp);
8696 			mutex_exit(&pidlock);
8697 			kmem_free(ssp, sizeof (strsig_t));
8698 			update = 1;
8699 			ssp = tssp;
8700 			continue;
8701 		} else
8702 			mutex_exit(&pidlock);
8703 		pssp = ssp;
8704 		ssp = ssp->ss_next;
8705 	}
8706 	if (update) {
8707 		stp->sd_sigflags = 0;
8708 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
8709 			stp->sd_sigflags |= ssp->ss_events;
8710 	}
8711 	mutex_exit(&stp->sd_lock);
8712 }
8713 
8714 /*
8715  * Return B_TRUE if there is data in the message, B_FALSE otherwise.
8716  */
8717 static boolean_t
8718 msghasdata(mblk_t *bp)
8719 {
8720 	for (; bp; bp = bp->b_cont)
8721 		if (bp->b_datap->db_type == M_DATA) {
8722 			ASSERT(bp->b_wptr >= bp->b_rptr);
8723 			if (bp->b_wptr > bp->b_rptr)
8724 				return (B_TRUE);
8725 		}
8726 	return (B_FALSE);
8727 }
8728 
8729 /*
8730  * Called on the first strget() of a sodirect/uioa enabled streamhead,
8731  * if any mblk_t(s) enqueued they must first be uioamove()d before uioa
8732  * can be enabled for the underlying transport's use.
8733  */
8734 void
8735 struioainit(queue_t *q, sodirect_t *sodp, uio_t *uiop)
8736 {
8737 	uioa_t	*uioap = (uioa_t *)uiop;
8738 	mblk_t	*bp = q->q_first;
8739 	mblk_t	*lbp = NULL;
8740 	mblk_t	*nbp, *wbp;
8741 	int	len;
8742 	int	error;
8743 
8744 	ASSERT(MUTEX_HELD(sodp->sod_lock));
8745 	ASSERT(&sodp->sod_uioa == uioap);
8746 
8747 	/*
8748 	 * Walk the b_next/b_prev doubly linked list of b_cont chain(s)
8749 	 * and schedule any M_DATA mblk_t's for uio asynchronous move.
8750 	 */
8751 	do {
8752 		/* Next mblk_t chain */
8753 		nbp = bp->b_next;
8754 		/* Walk the chain */
8755 		wbp = bp;
8756 		do {
8757 			if (wbp->b_datap->db_type != M_DATA) {
8758 				/* Not M_DATA, no more uioa */
8759 				goto nouioa;
8760 			}
8761 			if ((len = wbp->b_wptr - wbp->b_rptr) > 0) {
8762 				/* Have a M_DATA mblk_t with data */
8763 				if (len > uioap->uio_resid) {
8764 					/* Not enough uio sapce */
8765 					goto nouioa;
8766 				}
8767 				error = uioamove(wbp->b_rptr, len,
8768 				    UIO_READ, uioap);
8769 				if (!error) {
8770 					/* Scheduled, mark dblk_t as such */
8771 					wbp->b_datap->db_flags |= DBLK_UIOA;
8772 				} else {
8773 					/* Error of some sort, no more uioa */
8774 					uioap->uioa_state &= UIOA_CLR;
8775 					uioap->uioa_state |= UIOA_FINI;
8776 					return;
8777 				}
8778 			}
8779 			/* Save last wbp processed */
8780 			lbp = wbp;
8781 		} while ((wbp = wbp->b_cont) != NULL);
8782 	} while ((bp = nbp) != NULL);
8783 
8784 	return;
8785 
8786 nouioa:
8787 	/* No more uioa */
8788 	uioap->uioa_state &= UIOA_CLR;
8789 	uioap->uioa_state |= UIOA_FINI;
8790 
8791 	/*
8792 	 * If we processed 1 or more mblk_t(s) then we need to split the
8793 	 * current mblk_t chain in 2 so that all the uioamove()ed mblk_t(s)
8794 	 * are in the current chain and the rest are in the following new
8795 	 * chain.
8796 	 */
8797 	if (lbp != NULL) {
8798 		/* New end of current chain */
8799 		lbp->b_cont = NULL;
8800 
8801 		/* Insert new chain wbp after bp */
8802 		if ((wbp->b_next = nbp) != NULL)
8803 			nbp->b_prev = wbp;
8804 		else
8805 			q->q_last = wbp;
8806 		wbp->b_prev = bp;
8807 		bp->b_next = wbp;
8808 	}
8809 }
8810