xref: /illumos-gate/usr/src/uts/common/os/streamio.c (revision 4e5ef1cee66fbfbbd2b3e56b81e2bb5700f4a59e)
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 (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
27  * Copyright 2017 Joyent, Inc.
28  * Copyright 2019 OmniOS Community Edition (OmniOSce) Association.
29  */
30 
31 #include <sys/types.h>
32 #include <sys/sysmacros.h>
33 #include <sys/param.h>
34 #include <sys/errno.h>
35 #include <sys/signal.h>
36 #include <sys/stat.h>
37 #include <sys/proc.h>
38 #include <sys/cred.h>
39 #include <sys/user.h>
40 #include <sys/vnode.h>
41 #include <sys/file.h>
42 #include <sys/stream.h>
43 #include <sys/strsubr.h>
44 #include <sys/stropts.h>
45 #include <sys/tihdr.h>
46 #include <sys/var.h>
47 #include <sys/poll.h>
48 #include <sys/termio.h>
49 #include <sys/ttold.h>
50 #include <sys/systm.h>
51 #include <sys/uio.h>
52 #include <sys/cmn_err.h>
53 #include <sys/sad.h>
54 #include <sys/netstack.h>
55 #include <sys/priocntl.h>
56 #include <sys/jioctl.h>
57 #include <sys/procset.h>
58 #include <sys/session.h>
59 #include <sys/kmem.h>
60 #include <sys/filio.h>
61 #include <sys/vtrace.h>
62 #include <sys/debug.h>
63 #include <sys/strredir.h>
64 #include <sys/fs/fifonode.h>
65 #include <sys/fs/snode.h>
66 #include <sys/strlog.h>
67 #include <sys/strsun.h>
68 #include <sys/project.h>
69 #include <sys/kbio.h>
70 #include <sys/msio.h>
71 #include <sys/tty.h>
72 #include <sys/ptyvar.h>
73 #include <sys/vuid_event.h>
74 #include <sys/modctl.h>
75 #include <sys/sunddi.h>
76 #include <sys/sunldi_impl.h>
77 #include <sys/autoconf.h>
78 #include <sys/policy.h>
79 #include <sys/dld.h>
80 #include <sys/zone.h>
81 #include <c2/audit.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 
148 /*
149  * qinit and module_info structures for stream head read and write queues
150  */
151 struct module_info strm_info = { 0, "strrhead", 0, INFPSZ, STRHIGH, STRLOW };
152 struct module_info stwm_info = { 0, "strwhead", 0, 0, 0, 0 };
153 struct qinit strdata = { strrput, NULL, NULL, NULL, NULL, &strm_info };
154 struct qinit stwdata = { NULL, strwsrv, NULL, NULL, NULL, &stwm_info };
155 struct module_info fiform_info = { 0, "fifostrrhead", 0, PIPE_BUF, FIFOHIWAT,
156     FIFOLOWAT };
157 struct module_info fifowm_info = { 0, "fifostrwhead", 0, 0, 0, 0 };
158 struct qinit fifo_strdata = { strrput, NULL, NULL, NULL, NULL, &fiform_info };
159 struct qinit fifo_stwdata = { NULL, strwsrv, NULL, NULL, NULL, &fifowm_info };
160 
161 extern kmutex_t	strresources;	/* protects global resources */
162 extern kmutex_t muxifier;	/* single-threads multiplexor creation */
163 
164 static boolean_t msghasdata(mblk_t *bp);
165 #define	msgnodata(bp) (!msghasdata(bp))
166 
167 /*
168  * Stream head locking notes:
169  *	There are four monitors associated with the stream head:
170  *	1. v_stream monitor: in stropen() and strclose() v_lock
171  *		is held while the association of vnode and stream
172  *		head is established or tested for.
173  *	2. open/close/push/pop monitor: sd_lock is held while each
174  *		thread bids for exclusive access to this monitor
175  *		for opening or closing a stream.  In addition, this
176  *		monitor is entered during pushes and pops.  This
177  *		guarantees that during plumbing operations there
178  *		is only one thread trying to change the plumbing.
179  *		Any other threads present in the stream are only
180  *		using the plumbing.
181  *	3. read/write monitor: in the case of read, a thread holds
182  *		sd_lock while trying to get data from the stream
183  *		head queue.  if there is none to fulfill a read
184  *		request, it sets RSLEEP and calls cv_wait_sig() down
185  *		in strwaitq() to await the arrival of new data.
186  *		when new data arrives in strrput(), sd_lock is acquired
187  *		before testing for RSLEEP and calling cv_broadcast().
188  *		the behavior of strwrite(), strwsrv(), and WSLEEP
189  *		mirror this.
190  *	4. ioctl monitor: sd_lock is gotten to ensure that only one
191  *		thread is doing an ioctl at a time.
192  */
193 
194 static int
195 push_mod(queue_t *qp, dev_t *devp, struct stdata *stp, const char *name,
196     int anchor, cred_t *crp, uint_t anchor_zoneid)
197 {
198 	int error;
199 	fmodsw_impl_t *fp;
200 
201 	if (stp->sd_flag & (STRHUP|STRDERR|STWRERR)) {
202 		error = (stp->sd_flag & STRHUP) ? ENXIO : EIO;
203 		return (error);
204 	}
205 	if (stp->sd_pushcnt >= nstrpush) {
206 		return (EINVAL);
207 	}
208 
209 	if ((fp = fmodsw_find(name, FMODSW_HOLD | FMODSW_LOAD)) == NULL) {
210 		stp->sd_flag |= STREOPENFAIL;
211 		return (EINVAL);
212 	}
213 
214 	/*
215 	 * push new module and call its open routine via qattach
216 	 */
217 	if ((error = qattach(qp, devp, 0, crp, fp, B_FALSE)) != 0)
218 		return (error);
219 
220 	/*
221 	 * Check to see if caller wants a STREAMS anchor
222 	 * put at this place in the stream, and add if so.
223 	 */
224 	mutex_enter(&stp->sd_lock);
225 	if (anchor == stp->sd_pushcnt) {
226 		stp->sd_anchor = stp->sd_pushcnt;
227 		stp->sd_anchorzone = anchor_zoneid;
228 	}
229 	mutex_exit(&stp->sd_lock);
230 
231 	return (0);
232 }
233 
234 /*
235  * Open a stream device.
236  */
237 int
238 stropen(vnode_t *vp, dev_t *devp, int flag, cred_t *crp)
239 {
240 	struct stdata *stp;
241 	queue_t *qp;
242 	int s;
243 	dev_t dummydev, savedev;
244 	struct autopush *ap;
245 	struct dlautopush dlap;
246 	int error = 0;
247 	ssize_t	rmin, rmax;
248 	int cloneopen;
249 	queue_t *brq;
250 	major_t major;
251 	str_stack_t *ss;
252 	zoneid_t zoneid;
253 	uint_t anchor;
254 
255 	/*
256 	 * If the stream already exists, wait for any open in progress
257 	 * to complete, then call the open function of each module and
258 	 * driver in the stream.  Otherwise create the stream.
259 	 */
260 	TRACE_1(TR_FAC_STREAMS_FR, TR_STROPEN, "stropen:%p", vp);
261 retry:
262 	mutex_enter(&vp->v_lock);
263 	if ((stp = vp->v_stream) != NULL) {
264 
265 		/*
266 		 * Waiting for stream to be created to device
267 		 * due to another open.
268 		 */
269 		mutex_exit(&vp->v_lock);
270 
271 		if (STRMATED(stp)) {
272 			struct stdata *strmatep = stp->sd_mate;
273 
274 			STRLOCKMATES(stp);
275 			if (strmatep->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
276 				if (flag & (FNDELAY|FNONBLOCK)) {
277 					error = EAGAIN;
278 					mutex_exit(&strmatep->sd_lock);
279 					goto ckreturn;
280 				}
281 				mutex_exit(&stp->sd_lock);
282 				if (!cv_wait_sig(&strmatep->sd_monitor,
283 				    &strmatep->sd_lock)) {
284 					error = EINTR;
285 					mutex_exit(&strmatep->sd_lock);
286 					mutex_enter(&stp->sd_lock);
287 					goto ckreturn;
288 				}
289 				mutex_exit(&strmatep->sd_lock);
290 				goto retry;
291 			}
292 			if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
293 				if (flag & (FNDELAY|FNONBLOCK)) {
294 					error = EAGAIN;
295 					mutex_exit(&strmatep->sd_lock);
296 					goto ckreturn;
297 				}
298 				mutex_exit(&strmatep->sd_lock);
299 				if (!cv_wait_sig(&stp->sd_monitor,
300 				    &stp->sd_lock)) {
301 					error = EINTR;
302 					goto ckreturn;
303 				}
304 				mutex_exit(&stp->sd_lock);
305 				goto retry;
306 			}
307 
308 			if (stp->sd_flag & (STRDERR|STWRERR)) {
309 				error = EIO;
310 				mutex_exit(&strmatep->sd_lock);
311 				goto ckreturn;
312 			}
313 
314 			stp->sd_flag |= STWOPEN;
315 			STRUNLOCKMATES(stp);
316 		} else {
317 			mutex_enter(&stp->sd_lock);
318 			if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
319 				if (flag & (FNDELAY|FNONBLOCK)) {
320 					error = EAGAIN;
321 					goto ckreturn;
322 				}
323 				if (!cv_wait_sig(&stp->sd_monitor,
324 				    &stp->sd_lock)) {
325 					error = EINTR;
326 					goto ckreturn;
327 				}
328 				mutex_exit(&stp->sd_lock);
329 				goto retry;  /* could be clone! */
330 			}
331 
332 			if (stp->sd_flag & (STRDERR|STWRERR)) {
333 				error = EIO;
334 				goto ckreturn;
335 			}
336 
337 			stp->sd_flag |= STWOPEN;
338 			mutex_exit(&stp->sd_lock);
339 		}
340 
341 		/*
342 		 * Open all modules and devices down stream to notify
343 		 * that another user is streaming.  For modules, set the
344 		 * last argument to MODOPEN and do not pass any open flags.
345 		 * Ignore dummydev since this is not the first open.
346 		 */
347 		claimstr(stp->sd_wrq);
348 		qp = stp->sd_wrq;
349 		while (_SAMESTR(qp)) {
350 			qp = qp->q_next;
351 			if ((error = qreopen(_RD(qp), devp, flag, crp)) != 0)
352 				break;
353 		}
354 		releasestr(stp->sd_wrq);
355 		mutex_enter(&stp->sd_lock);
356 		stp->sd_flag &= ~(STRHUP|STWOPEN|STRDERR|STWRERR);
357 		stp->sd_rerror = 0;
358 		stp->sd_werror = 0;
359 ckreturn:
360 		cv_broadcast(&stp->sd_monitor);
361 		mutex_exit(&stp->sd_lock);
362 		return (error);
363 	}
364 
365 	/*
366 	 * This vnode isn't streaming.  SPECFS already
367 	 * checked for multiple vnodes pointing to the
368 	 * same stream, so create a stream to the driver.
369 	 */
370 	qp = allocq();
371 	stp = shalloc(qp);
372 
373 	/*
374 	 * Initialize stream head.  shalloc() has given us
375 	 * exclusive access, and we have the vnode locked;
376 	 * we can do whatever we want with stp.
377 	 */
378 	stp->sd_flag = STWOPEN;
379 	stp->sd_siglist = NULL;
380 	stp->sd_pollist.ph_list = NULL;
381 	stp->sd_sigflags = 0;
382 	stp->sd_mark = NULL;
383 	stp->sd_closetime = STRTIMOUT;
384 	stp->sd_sidp = NULL;
385 	stp->sd_pgidp = NULL;
386 	stp->sd_vnode = vp;
387 	stp->sd_pvnode = NULL;
388 	stp->sd_rerror = 0;
389 	stp->sd_werror = 0;
390 	stp->sd_wroff = 0;
391 	stp->sd_tail = 0;
392 	stp->sd_iocblk = NULL;
393 	stp->sd_cmdblk = NULL;
394 	stp->sd_pushcnt = 0;
395 	stp->sd_qn_minpsz = 0;
396 	stp->sd_qn_maxpsz = INFPSZ - 1;	/* used to check for initialization */
397 	stp->sd_maxblk = INFPSZ;
398 	qp->q_ptr = _WR(qp)->q_ptr = stp;
399 	STREAM(qp) = STREAM(_WR(qp)) = stp;
400 	vp->v_stream = stp;
401 	mutex_exit(&vp->v_lock);
402 	if (vp->v_type == VFIFO) {
403 		stp->sd_flag |= OLDNDELAY;
404 		/*
405 		 * This means, both for pipes and fifos
406 		 * strwrite will send SIGPIPE if the other
407 		 * end is closed. For putmsg it depends
408 		 * on whether it is a XPG4_2 application
409 		 * or not
410 		 */
411 		stp->sd_wput_opt = SW_SIGPIPE;
412 
413 		/* setq might sleep in kmem_alloc - avoid holding locks. */
414 		setq(qp, &fifo_strdata, &fifo_stwdata, NULL, QMTSAFE,
415 		    SQ_CI|SQ_CO, B_FALSE);
416 
417 		set_qend(qp);
418 		stp->sd_strtab = fifo_getinfo();
419 		_WR(qp)->q_nfsrv = _WR(qp);
420 		qp->q_nfsrv = qp;
421 		/*
422 		 * Wake up others that are waiting for stream to be created.
423 		 */
424 		mutex_enter(&stp->sd_lock);
425 		/*
426 		 * nothing is be pushed on stream yet, so
427 		 * optimized stream head packetsizes are just that
428 		 * of the read queue
429 		 */
430 		stp->sd_qn_minpsz = qp->q_minpsz;
431 		stp->sd_qn_maxpsz = qp->q_maxpsz;
432 		stp->sd_flag &= ~STWOPEN;
433 		goto fifo_opendone;
434 	}
435 	/* setq might sleep in kmem_alloc - avoid holding locks. */
436 	setq(qp, &strdata, &stwdata, NULL, QMTSAFE, SQ_CI|SQ_CO, B_FALSE);
437 
438 	set_qend(qp);
439 
440 	/*
441 	 * Open driver and create stream to it (via qattach).
442 	 */
443 	savedev = *devp;
444 	cloneopen = (getmajor(*devp) == clone_major);
445 	if ((error = qattach(qp, devp, flag, crp, NULL, B_FALSE)) != 0) {
446 		mutex_enter(&vp->v_lock);
447 		vp->v_stream = NULL;
448 		mutex_exit(&vp->v_lock);
449 		mutex_enter(&stp->sd_lock);
450 		cv_broadcast(&stp->sd_monitor);
451 		mutex_exit(&stp->sd_lock);
452 		freeq(_RD(qp));
453 		shfree(stp);
454 		return (error);
455 	}
456 	/*
457 	 * Set sd_strtab after open in order to handle clonable drivers
458 	 */
459 	stp->sd_strtab = STREAMSTAB(getmajor(*devp));
460 
461 	/*
462 	 * Historical note: dummydev used to be be prior to the initial
463 	 * open (via qattach above), which made the value seen
464 	 * inconsistent between an I_PUSH and an autopush of a module.
465 	 */
466 	dummydev = *devp;
467 
468 	/*
469 	 * For clone open of old style (Q not associated) network driver,
470 	 * push DRMODNAME module to handle DL_ATTACH/DL_DETACH
471 	 */
472 	brq = _RD(_WR(qp)->q_next);
473 	major = getmajor(*devp);
474 	if (push_drcompat && cloneopen && NETWORK_DRV(major) &&
475 	    ((brq->q_flag & _QASSOCIATED) == 0)) {
476 		if (push_mod(qp, &dummydev, stp, DRMODNAME, 0, crp, 0) != 0)
477 			cmn_err(CE_WARN, "cannot push " DRMODNAME
478 			    " streams module");
479 	}
480 
481 	if (!NETWORK_DRV(major)) {
482 		savedev = *devp;
483 	} else {
484 		/*
485 		 * For network devices, process differently based on the
486 		 * return value from dld_autopush():
487 		 *
488 		 *   0: the passed-in device points to a GLDv3 datalink with
489 		 *   per-link autopush configuration; use that configuration
490 		 *   and ignore any per-driver autopush configuration.
491 		 *
492 		 *   1: the passed-in device points to a physical GLDv3
493 		 *   datalink without per-link autopush configuration.  The
494 		 *   passed in device was changed to refer to the actual
495 		 *   physical device (if it's not already); we use that new
496 		 *   device to look up any per-driver autopush configuration.
497 		 *
498 		 *   -1: neither of the above cases applied; use the initial
499 		 *   device to look up any per-driver autopush configuration.
500 		 */
501 		switch (dld_autopush(&savedev, &dlap)) {
502 		case 0:
503 			zoneid = crgetzoneid(crp);
504 			for (s = 0; s < dlap.dap_npush; s++) {
505 				error = push_mod(qp, &dummydev, stp,
506 				    dlap.dap_aplist[s], dlap.dap_anchor, crp,
507 				    zoneid);
508 				if (error != 0)
509 					break;
510 			}
511 			goto opendone;
512 		case 1:
513 			break;
514 		case -1:
515 			savedev = *devp;
516 			break;
517 		}
518 	}
519 	/*
520 	 * Find the autopush configuration based on "savedev". Start with the
521 	 * global zone. If not found check in the local zone.
522 	 */
523 	zoneid = GLOBAL_ZONEID;
524 retryap:
525 	ss = netstack_find_by_stackid(zoneid_to_netstackid(zoneid))->
526 	    netstack_str;
527 	if ((ap = sad_ap_find_by_dev(savedev, ss)) == NULL) {
528 		netstack_rele(ss->ss_netstack);
529 		if (zoneid == GLOBAL_ZONEID) {
530 			/*
531 			 * None found. Also look in the zone's autopush table.
532 			 */
533 			zoneid = crgetzoneid(crp);
534 			if (zoneid != GLOBAL_ZONEID)
535 				goto retryap;
536 		}
537 		goto opendone;
538 	}
539 	anchor = ap->ap_anchor;
540 	zoneid = crgetzoneid(crp);
541 	for (s = 0; s < ap->ap_npush; s++) {
542 		error = push_mod(qp, &dummydev, stp, ap->ap_list[s],
543 		    anchor, crp, zoneid);
544 		if (error != 0)
545 			break;
546 	}
547 	sad_ap_rele(ap, ss);
548 	netstack_rele(ss->ss_netstack);
549 
550 opendone:
551 
552 	/*
553 	 * let specfs know that open failed part way through
554 	 */
555 	if (error) {
556 		mutex_enter(&stp->sd_lock);
557 		stp->sd_flag |= STREOPENFAIL;
558 		mutex_exit(&stp->sd_lock);
559 	}
560 
561 	/*
562 	 * Wake up others that are waiting for stream to be created.
563 	 */
564 	mutex_enter(&stp->sd_lock);
565 	stp->sd_flag &= ~STWOPEN;
566 
567 	/*
568 	 * As a performance concern we are caching the values of
569 	 * q_minpsz and q_maxpsz of the module below the stream
570 	 * head in the stream head.
571 	 */
572 	mutex_enter(QLOCK(stp->sd_wrq->q_next));
573 	rmin = stp->sd_wrq->q_next->q_minpsz;
574 	rmax = stp->sd_wrq->q_next->q_maxpsz;
575 	mutex_exit(QLOCK(stp->sd_wrq->q_next));
576 
577 	/* do this processing here as a performance concern */
578 	if (strmsgsz != 0) {
579 		if (rmax == INFPSZ)
580 			rmax = strmsgsz;
581 		else
582 			rmax = MIN(strmsgsz, rmax);
583 	}
584 
585 	mutex_enter(QLOCK(stp->sd_wrq));
586 	stp->sd_qn_minpsz = rmin;
587 	stp->sd_qn_maxpsz = rmax;
588 	mutex_exit(QLOCK(stp->sd_wrq));
589 
590 fifo_opendone:
591 	cv_broadcast(&stp->sd_monitor);
592 	mutex_exit(&stp->sd_lock);
593 	return (error);
594 }
595 
596 static int strsink(queue_t *, mblk_t *);
597 static struct qinit deadrend = {
598 	strsink, NULL, NULL, NULL, NULL, &strm_info, NULL
599 };
600 static struct qinit deadwend = {
601 	NULL, NULL, NULL, NULL, NULL, &stwm_info, NULL
602 };
603 
604 /*
605  * Close a stream.
606  * This is called from closef() on the last close of an open stream.
607  * Strclean() will already have removed the siglist and pollist
608  * information, so all that remains is to remove all multiplexor links
609  * for the stream, pop all the modules (and the driver), and free the
610  * stream structure.
611  */
612 
613 int
614 strclose(struct vnode *vp, int flag, cred_t *crp)
615 {
616 	struct stdata *stp;
617 	queue_t *qp;
618 	int rval;
619 	int freestp = 1;
620 	queue_t *rmq;
621 
622 	TRACE_1(TR_FAC_STREAMS_FR,
623 	    TR_STRCLOSE, "strclose:%p", vp);
624 	ASSERT(vp->v_stream);
625 
626 	stp = vp->v_stream;
627 	ASSERT(!(stp->sd_flag & STPLEX));
628 	qp = stp->sd_wrq;
629 
630 	/*
631 	 * Needed so that strpoll will return non-zero for this fd.
632 	 * Note that with POLLNOERR STRHUP does still cause POLLHUP.
633 	 */
634 	mutex_enter(&stp->sd_lock);
635 	stp->sd_flag |= STRHUP;
636 	mutex_exit(&stp->sd_lock);
637 
638 	/*
639 	 * If the registered process or process group did not have an
640 	 * open instance of this stream then strclean would not be
641 	 * called. Thus at the time of closing all remaining siglist entries
642 	 * are removed.
643 	 */
644 	if (stp->sd_siglist != NULL)
645 		strcleanall(vp);
646 
647 	ASSERT(stp->sd_siglist == NULL);
648 	ASSERT(stp->sd_sigflags == 0);
649 
650 	if (STRMATED(stp)) {
651 		struct stdata *strmatep = stp->sd_mate;
652 		int waited = 1;
653 
654 		STRLOCKMATES(stp);
655 		while (waited) {
656 			waited = 0;
657 			while (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
658 				mutex_exit(&strmatep->sd_lock);
659 				cv_wait(&stp->sd_monitor, &stp->sd_lock);
660 				mutex_exit(&stp->sd_lock);
661 				STRLOCKMATES(stp);
662 				waited = 1;
663 			}
664 			while (strmatep->sd_flag &
665 			    (STWOPEN|STRCLOSE|STRPLUMB)) {
666 				mutex_exit(&stp->sd_lock);
667 				cv_wait(&strmatep->sd_monitor,
668 				    &strmatep->sd_lock);
669 				mutex_exit(&strmatep->sd_lock);
670 				STRLOCKMATES(stp);
671 				waited = 1;
672 			}
673 		}
674 		stp->sd_flag |= STRCLOSE;
675 		STRUNLOCKMATES(stp);
676 	} else {
677 		mutex_enter(&stp->sd_lock);
678 		stp->sd_flag |= STRCLOSE;
679 		mutex_exit(&stp->sd_lock);
680 	}
681 
682 	ASSERT(qp->q_first == NULL);	/* No more delayed write */
683 
684 	/* Check if an I_LINK was ever done on this stream */
685 	if (stp->sd_flag & STRHASLINKS) {
686 		netstack_t *ns;
687 		str_stack_t *ss;
688 
689 		ns = netstack_find_by_cred(crp);
690 		ASSERT(ns != NULL);
691 		ss = ns->netstack_str;
692 		ASSERT(ss != NULL);
693 
694 		(void) munlinkall(stp, LINKCLOSE|LINKNORMAL, crp, &rval, ss);
695 		netstack_rele(ss->ss_netstack);
696 	}
697 
698 	while (_SAMESTR(qp)) {
699 		/*
700 		 * Holding sd_lock prevents q_next from changing in
701 		 * this stream.
702 		 */
703 		mutex_enter(&stp->sd_lock);
704 		if (!(flag & (FNDELAY|FNONBLOCK)) && (stp->sd_closetime > 0)) {
705 
706 			/*
707 			 * sleep until awakened by strwsrv() or timeout
708 			 */
709 			for (;;) {
710 				mutex_enter(QLOCK(qp->q_next));
711 				if (!(qp->q_next->q_mblkcnt)) {
712 					mutex_exit(QLOCK(qp->q_next));
713 					break;
714 				}
715 				stp->sd_flag |= WSLEEP;
716 
717 				/* ensure strwsrv gets enabled */
718 				qp->q_next->q_flag |= QWANTW;
719 				mutex_exit(QLOCK(qp->q_next));
720 				/* get out if we timed out or recv'd a signal */
721 				if (str_cv_wait(&qp->q_wait, &stp->sd_lock,
722 				    stp->sd_closetime, 0) <= 0) {
723 					break;
724 				}
725 			}
726 			stp->sd_flag &= ~WSLEEP;
727 		}
728 		mutex_exit(&stp->sd_lock);
729 
730 		rmq = qp->q_next;
731 		if (rmq->q_flag & QISDRV) {
732 			ASSERT(!_SAMESTR(rmq));
733 			wait_sq_svc(_RD(qp)->q_syncq);
734 		}
735 
736 		qdetach(_RD(rmq), 1, flag, crp, B_FALSE);
737 	}
738 
739 	/*
740 	 * Since we call pollwakeup in close() now, the poll list should
741 	 * be empty in most cases. The only exception is the layered devices
742 	 * (e.g. the console drivers with redirection modules pushed on top
743 	 * of it).  We have to do this after calling qdetach() because
744 	 * the redirection module won't have torn down the console
745 	 * redirection until after qdetach() has been invoked.
746 	 */
747 	if (stp->sd_pollist.ph_list != NULL) {
748 		pollwakeup(&stp->sd_pollist, POLLERR);
749 		pollhead_clean(&stp->sd_pollist);
750 	}
751 	ASSERT(stp->sd_pollist.ph_list == NULL);
752 	ASSERT(stp->sd_sidp == NULL);
753 	ASSERT(stp->sd_pgidp == NULL);
754 
755 	/* Prevent qenable from re-enabling the stream head queue */
756 	disable_svc(_RD(qp));
757 
758 	/*
759 	 * Wait until service procedure of each queue is
760 	 * run, if QINSERVICE is set.
761 	 */
762 	wait_svc(_RD(qp));
763 
764 	/*
765 	 * Now, flush both queues.
766 	 */
767 	flushq(_RD(qp), FLUSHALL);
768 	flushq(qp, FLUSHALL);
769 
770 	/*
771 	 * If the write queue of the stream head is pointing to a
772 	 * read queue, we have a twisted stream.  If the read queue
773 	 * is alive, convert the stream head queues into a dead end.
774 	 * If the read queue is dead, free the dead pair.
775 	 */
776 	if (qp->q_next && !_SAMESTR(qp)) {
777 		if (qp->q_next->q_qinfo == &deadrend) {	/* half-closed pipe */
778 			flushq(qp->q_next, FLUSHALL); /* ensure no message */
779 			shfree(qp->q_next->q_stream);
780 			freeq(qp->q_next);
781 			freeq(_RD(qp));
782 		} else if (qp->q_next == _RD(qp)) {	/* fifo */
783 			freeq(_RD(qp));
784 		} else {				/* pipe */
785 			freestp = 0;
786 			/*
787 			 * The q_info pointers are never accessed when
788 			 * SQLOCK is held.
789 			 */
790 			ASSERT(qp->q_syncq == _RD(qp)->q_syncq);
791 			mutex_enter(SQLOCK(qp->q_syncq));
792 			qp->q_qinfo = &deadwend;
793 			_RD(qp)->q_qinfo = &deadrend;
794 			mutex_exit(SQLOCK(qp->q_syncq));
795 		}
796 	} else {
797 		freeq(_RD(qp)); /* free stream head queue pair */
798 	}
799 
800 	mutex_enter(&vp->v_lock);
801 	if (stp->sd_iocblk) {
802 		if (stp->sd_iocblk != (mblk_t *)-1) {
803 			freemsg(stp->sd_iocblk);
804 		}
805 		stp->sd_iocblk = NULL;
806 	}
807 	stp->sd_vnode = stp->sd_pvnode = NULL;
808 	vp->v_stream = NULL;
809 	mutex_exit(&vp->v_lock);
810 	mutex_enter(&stp->sd_lock);
811 	freemsg(stp->sd_cmdblk);
812 	stp->sd_cmdblk = NULL;
813 	stp->sd_flag &= ~STRCLOSE;
814 	cv_broadcast(&stp->sd_monitor);
815 	mutex_exit(&stp->sd_lock);
816 
817 	if (freestp)
818 		shfree(stp);
819 	return (0);
820 }
821 
822 static int
823 strsink(queue_t *q, mblk_t *bp)
824 {
825 	struct copyresp *resp;
826 
827 	switch (bp->b_datap->db_type) {
828 	case M_FLUSH:
829 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
830 			*bp->b_rptr &= ~FLUSHR;
831 			bp->b_flag |= MSGNOLOOP;
832 			/*
833 			 * Protect against the driver passing up
834 			 * messages after it has done a qprocsoff.
835 			 */
836 			if (_OTHERQ(q)->q_next == NULL)
837 				freemsg(bp);
838 			else
839 				qreply(q, bp);
840 		} else {
841 			freemsg(bp);
842 		}
843 		break;
844 
845 	case M_COPYIN:
846 	case M_COPYOUT:
847 		if (bp->b_cont) {
848 			freemsg(bp->b_cont);
849 			bp->b_cont = NULL;
850 		}
851 		bp->b_datap->db_type = M_IOCDATA;
852 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
853 		resp = (struct copyresp *)bp->b_rptr;
854 		resp->cp_rval = (caddr_t)1;	/* failure */
855 		/*
856 		 * Protect against the driver passing up
857 		 * messages after it has done a qprocsoff.
858 		 */
859 		if (_OTHERQ(q)->q_next == NULL)
860 			freemsg(bp);
861 		else
862 			qreply(q, bp);
863 		break;
864 
865 	case M_IOCTL:
866 		if (bp->b_cont) {
867 			freemsg(bp->b_cont);
868 			bp->b_cont = NULL;
869 		}
870 		bp->b_datap->db_type = M_IOCNAK;
871 		/*
872 		 * Protect against the driver passing up
873 		 * messages after it has done a qprocsoff.
874 		 */
875 		if (_OTHERQ(q)->q_next == NULL)
876 			freemsg(bp);
877 		else
878 			qreply(q, bp);
879 		break;
880 
881 	default:
882 		freemsg(bp);
883 		break;
884 	}
885 
886 	return (0);
887 }
888 
889 /*
890  * Clean up after a process when it closes a stream.  This is called
891  * from closef for all closes, whereas strclose is called only for the
892  * last close on a stream.  The siglist is scanned for entries for the
893  * current process, and these are removed.
894  */
895 void
896 strclean(struct vnode *vp)
897 {
898 	strsig_t *ssp, *pssp, *tssp;
899 	stdata_t *stp;
900 	int update = 0;
901 
902 	TRACE_1(TR_FAC_STREAMS_FR,
903 	    TR_STRCLEAN, "strclean:%p", vp);
904 	stp = vp->v_stream;
905 	pssp = NULL;
906 	mutex_enter(&stp->sd_lock);
907 	ssp = stp->sd_siglist;
908 	while (ssp) {
909 		if (ssp->ss_pidp == curproc->p_pidp) {
910 			tssp = ssp->ss_next;
911 			if (pssp)
912 				pssp->ss_next = tssp;
913 			else
914 				stp->sd_siglist = tssp;
915 			mutex_enter(&pidlock);
916 			PID_RELE(ssp->ss_pidp);
917 			mutex_exit(&pidlock);
918 			kmem_free(ssp, sizeof (strsig_t));
919 			update = 1;
920 			ssp = tssp;
921 		} else {
922 			pssp = ssp;
923 			ssp = ssp->ss_next;
924 		}
925 	}
926 	if (update) {
927 		stp->sd_sigflags = 0;
928 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
929 			stp->sd_sigflags |= ssp->ss_events;
930 	}
931 	mutex_exit(&stp->sd_lock);
932 }
933 
934 /*
935  * Used on the last close to remove any remaining items on the siglist.
936  * These could be present on the siglist due to I_ESETSIG calls that
937  * use process groups or processed that do not have an open file descriptor
938  * for this stream (Such entries would not be removed by strclean).
939  */
940 static void
941 strcleanall(struct vnode *vp)
942 {
943 	strsig_t *ssp, *nssp;
944 	stdata_t *stp;
945 
946 	stp = vp->v_stream;
947 	mutex_enter(&stp->sd_lock);
948 	ssp = stp->sd_siglist;
949 	stp->sd_siglist = NULL;
950 	while (ssp) {
951 		nssp = ssp->ss_next;
952 		mutex_enter(&pidlock);
953 		PID_RELE(ssp->ss_pidp);
954 		mutex_exit(&pidlock);
955 		kmem_free(ssp, sizeof (strsig_t));
956 		ssp = nssp;
957 	}
958 	stp->sd_sigflags = 0;
959 	mutex_exit(&stp->sd_lock);
960 }
961 
962 /*
963  * Retrieve the next message from the logical stream head read queue
964  * using either rwnext (if sync stream) or getq_noenab.
965  * It is the callers responsibility to call qbackenable after
966  * it is finished with the message. The caller should not call
967  * qbackenable until after any putback calls to avoid spurious backenabling.
968  */
969 mblk_t *
970 strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
971     int *errorp)
972 {
973 	mblk_t *bp;
974 	int error;
975 	ssize_t rbytes = 0;
976 
977 	/* Holding sd_lock prevents the read queue from changing  */
978 	ASSERT(MUTEX_HELD(&stp->sd_lock));
979 
980 	if (uiop != NULL && stp->sd_struiordq != NULL &&
981 	    q->q_first == NULL &&
982 	    (!first || (stp->sd_wakeq & RSLEEP))) {
983 		/*
984 		 * Stream supports rwnext() for the read side.
985 		 * If this is the first time we're called by e.g. strread
986 		 * only do the downcall if there is a deferred wakeup
987 		 * (registered in sd_wakeq).
988 		 */
989 		struiod_t uiod;
990 
991 		if (first)
992 			stp->sd_wakeq &= ~RSLEEP;
993 
994 		(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
995 		    sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
996 		uiod.d_mp = 0;
997 		/*
998 		 * Mark that a thread is in rwnext on the read side
999 		 * to prevent strrput from nacking ioctls immediately.
1000 		 * When the last concurrent rwnext returns
1001 		 * the ioctls are nack'ed.
1002 		 */
1003 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1004 		stp->sd_struiodnak++;
1005 		/*
1006 		 * Note: rwnext will drop sd_lock.
1007 		 */
1008 		error = rwnext(q, &uiod);
1009 		ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
1010 		mutex_enter(&stp->sd_lock);
1011 		stp->sd_struiodnak--;
1012 		while (stp->sd_struiodnak == 0 &&
1013 		    ((bp = stp->sd_struionak) != NULL)) {
1014 			stp->sd_struionak = bp->b_next;
1015 			bp->b_next = NULL;
1016 			bp->b_datap->db_type = M_IOCNAK;
1017 			/*
1018 			 * Protect against the driver passing up
1019 			 * messages after it has done a qprocsoff.
1020 			 */
1021 			if (_OTHERQ(q)->q_next == NULL)
1022 				freemsg(bp);
1023 			else {
1024 				mutex_exit(&stp->sd_lock);
1025 				qreply(q, bp);
1026 				mutex_enter(&stp->sd_lock);
1027 			}
1028 		}
1029 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1030 		if (error == 0 || error == EWOULDBLOCK) {
1031 			if ((bp = uiod.d_mp) != NULL) {
1032 				*errorp = 0;
1033 				ASSERT(MUTEX_HELD(&stp->sd_lock));
1034 				return (bp);
1035 			}
1036 			error = 0;
1037 		} else if (error == EINVAL) {
1038 			/*
1039 			 * The stream plumbing must have
1040 			 * changed while we were away, so
1041 			 * just turn off rwnext()s.
1042 			 */
1043 			error = 0;
1044 		} else if (error == EBUSY) {
1045 			/*
1046 			 * The module might have data in transit using putnext
1047 			 * Fall back on waiting + getq.
1048 			 */
1049 			error = 0;
1050 		} else {
1051 			*errorp = error;
1052 			ASSERT(MUTEX_HELD(&stp->sd_lock));
1053 			return (NULL);
1054 		}
1055 		/*
1056 		 * Try a getq in case a rwnext() generated mblk
1057 		 * has bubbled up via strrput().
1058 		 */
1059 	}
1060 	*errorp = 0;
1061 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1062 
1063 	/*
1064 	 * If we have a valid uio, try and use this as a guide for how
1065 	 * many bytes to retrieve from the queue via getq_noenab().
1066 	 * Doing this can avoid unneccesary counting of overlong
1067 	 * messages in putback(). We currently only do this for sockets
1068 	 * and only if there is no sd_rputdatafunc hook.
1069 	 *
1070 	 * The sd_rputdatafunc hook transforms the entire message
1071 	 * before any bytes in it can be given to a client. So, rbytes
1072 	 * must be 0 if there is a hook.
1073 	 */
1074 	if ((uiop != NULL) && (stp->sd_vnode->v_type == VSOCK) &&
1075 	    (stp->sd_rputdatafunc == NULL))
1076 		rbytes = uiop->uio_resid;
1077 
1078 	return (getq_noenab(q, rbytes));
1079 }
1080 
1081 /*
1082  * Copy out the message pointed to by `bp' into the uio pointed to by `uiop'.
1083  * If the message does not fit in the uio the remainder of it is returned;
1084  * otherwise NULL is returned.  Any embedded zero-length mblk_t's are
1085  * consumed, even if uio_resid reaches zero.  On error, `*errorp' is set to
1086  * the error code, the message is consumed, and NULL is returned.
1087  */
1088 static mblk_t *
1089 struiocopyout(mblk_t *bp, struct uio *uiop, int *errorp)
1090 {
1091 	int error;
1092 	ptrdiff_t n;
1093 	mblk_t *nbp;
1094 
1095 	ASSERT(bp->b_wptr >= bp->b_rptr);
1096 
1097 	do {
1098 		if ((n = MIN(uiop->uio_resid, MBLKL(bp))) != 0) {
1099 			ASSERT(n > 0);
1100 
1101 			error = uiomove(bp->b_rptr, n, UIO_READ, uiop);
1102 			if (error != 0) {
1103 				freemsg(bp);
1104 				*errorp = error;
1105 				return (NULL);
1106 			}
1107 		}
1108 
1109 		bp->b_rptr += n;
1110 		while (bp != NULL && (bp->b_rptr >= bp->b_wptr)) {
1111 			nbp = bp;
1112 			bp = bp->b_cont;
1113 			freeb(nbp);
1114 		}
1115 	} while (bp != NULL && uiop->uio_resid > 0);
1116 
1117 	*errorp = 0;
1118 	return (bp);
1119 }
1120 
1121 /*
1122  * Read a stream according to the mode flags in sd_flag:
1123  *
1124  * (default mode)		- Byte stream, msg boundaries are ignored
1125  * RD_MSGDIS (msg discard)	- Read on msg boundaries and throw away
1126  *				any data remaining in msg
1127  * RD_MSGNODIS (msg non-discard) - Read on msg boundaries and put back
1128  *				any remaining data on head of read queue
1129  *
1130  * Consume readable messages on the front of the queue until
1131  * ttolwp(curthread)->lwp_count
1132  * is satisfied, the readable messages are exhausted, or a message
1133  * boundary is reached in a message mode.  If no data was read and
1134  * the stream was not opened with the NDELAY flag, block until data arrives.
1135  * Otherwise return the data read and update the count.
1136  *
1137  * In default mode a 0 length message signifies end-of-file and terminates
1138  * a read in progress.  The 0 length message is removed from the queue
1139  * only if it is the only message read (no data is read).
1140  *
1141  * An attempt to read an M_PROTO or M_PCPROTO message results in an
1142  * EBADMSG error return, unless either RD_PROTDAT or RD_PROTDIS are set.
1143  * If RD_PROTDAT is set, M_PROTO and M_PCPROTO messages are read as data.
1144  * If RD_PROTDIS is set, the M_PROTO and M_PCPROTO parts of the message
1145  * are unlinked from and M_DATA blocks in the message, the protos are
1146  * thrown away, and the data is read.
1147  */
1148 /* ARGSUSED */
1149 int
1150 strread(struct vnode *vp, struct uio *uiop, cred_t *crp)
1151 {
1152 	struct stdata *stp;
1153 	mblk_t *bp, *nbp;
1154 	queue_t *q;
1155 	int error = 0;
1156 	uint_t old_sd_flag;
1157 	int first;
1158 	char rflg;
1159 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
1160 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
1161 	short delim;
1162 	unsigned char pri = 0;
1163 	char waitflag;
1164 	unsigned char type;
1165 
1166 	TRACE_1(TR_FAC_STREAMS_FR,
1167 	    TR_STRREAD_ENTER, "strread:%p", vp);
1168 	ASSERT(vp->v_stream);
1169 	stp = vp->v_stream;
1170 
1171 	mutex_enter(&stp->sd_lock);
1172 
1173 	if ((error = i_straccess(stp, JCREAD)) != 0) {
1174 		mutex_exit(&stp->sd_lock);
1175 		return (error);
1176 	}
1177 
1178 	if (stp->sd_flag & (STRDERR|STPLEX)) {
1179 		error = strgeterr(stp, STRDERR|STPLEX, 0);
1180 		if (error != 0) {
1181 			mutex_exit(&stp->sd_lock);
1182 			return (error);
1183 		}
1184 	}
1185 
1186 	/*
1187 	 * Loop terminates when uiop->uio_resid == 0.
1188 	 */
1189 	rflg = 0;
1190 	waitflag = READWAIT;
1191 	q = _RD(stp->sd_wrq);
1192 	for (;;) {
1193 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1194 		old_sd_flag = stp->sd_flag;
1195 		mark = 0;
1196 		delim = 0;
1197 		first = 1;
1198 		while ((bp = strget(stp, q, uiop, first, &error)) == NULL) {
1199 			int done = 0;
1200 
1201 			ASSERT(MUTEX_HELD(&stp->sd_lock));
1202 
1203 			if (error != 0)
1204 				goto oops;
1205 
1206 			if (stp->sd_flag & (STRHUP|STREOF)) {
1207 				goto oops;
1208 			}
1209 			if (rflg && !(stp->sd_flag & STRDELIM)) {
1210 				goto oops;
1211 			}
1212 			/*
1213 			 * If a read(fd,buf,0) has been done, there is no
1214 			 * need to sleep. We always have zero bytes to
1215 			 * return.
1216 			 */
1217 			if (uiop->uio_resid == 0) {
1218 				goto oops;
1219 			}
1220 
1221 			qbackenable(q, 0);
1222 
1223 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_WAIT,
1224 			    "strread calls strwaitq:%p, %p, %p",
1225 			    vp, uiop, crp);
1226 			if ((error = strwaitq(stp, waitflag, uiop->uio_resid,
1227 			    uiop->uio_fmode, -1, &done)) != 0 || done) {
1228 				TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_DONE,
1229 				    "strread error or done:%p, %p, %p",
1230 				    vp, uiop, crp);
1231 				if ((uiop->uio_fmode & FNDELAY) &&
1232 				    (stp->sd_flag & OLDNDELAY) &&
1233 				    (error == EAGAIN))
1234 					error = 0;
1235 				goto oops;
1236 			}
1237 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_AWAKE,
1238 			    "strread awakes:%p, %p, %p", vp, uiop, crp);
1239 			if ((error = i_straccess(stp, JCREAD)) != 0) {
1240 				goto oops;
1241 			}
1242 			first = 0;
1243 		}
1244 
1245 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1246 		ASSERT(bp);
1247 		pri = bp->b_band;
1248 		/*
1249 		 * Extract any mark information. If the message is not
1250 		 * completely consumed this information will be put in the mblk
1251 		 * that is putback.
1252 		 * If MSGMARKNEXT is set and the message is completely consumed
1253 		 * the STRATMARK flag will be set below. Likewise, if
1254 		 * MSGNOTMARKNEXT is set and the message is
1255 		 * completely consumed STRNOTATMARK will be set.
1256 		 *
1257 		 * For some unknown reason strread only breaks the read at the
1258 		 * last mark.
1259 		 */
1260 		mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
1261 		ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
1262 		    (MSGMARKNEXT|MSGNOTMARKNEXT));
1263 		if (mark != 0 && bp == stp->sd_mark) {
1264 			if (rflg) {
1265 				putback(stp, q, bp, pri);
1266 				goto oops;
1267 			}
1268 			mark |= _LASTMARK;
1269 			stp->sd_mark = NULL;
1270 		}
1271 		if ((stp->sd_flag & STRDELIM) && (bp->b_flag & MSGDELIM))
1272 			delim = 1;
1273 		mutex_exit(&stp->sd_lock);
1274 
1275 		if (STREAM_NEEDSERVICE(stp))
1276 			stream_runservice(stp);
1277 
1278 		type = bp->b_datap->db_type;
1279 
1280 		switch (type) {
1281 
1282 		case M_DATA:
1283 ismdata:
1284 			if (msgnodata(bp)) {
1285 				if (mark || delim) {
1286 					freemsg(bp);
1287 				} else if (rflg) {
1288 
1289 					/*
1290 					 * If already read data put zero
1291 					 * length message back on queue else
1292 					 * free msg and return 0.
1293 					 */
1294 					bp->b_band = pri;
1295 					mutex_enter(&stp->sd_lock);
1296 					putback(stp, q, bp, pri);
1297 					mutex_exit(&stp->sd_lock);
1298 				} else {
1299 					freemsg(bp);
1300 				}
1301 				error =  0;
1302 				goto oops1;
1303 			}
1304 
1305 			rflg = 1;
1306 			waitflag |= NOINTR;
1307 			bp = struiocopyout(bp, uiop, &error);
1308 			if (error != 0)
1309 				goto oops1;
1310 
1311 			mutex_enter(&stp->sd_lock);
1312 			if (bp) {
1313 				/*
1314 				 * Have remaining data in message.
1315 				 * Free msg if in discard mode.
1316 				 */
1317 				if (stp->sd_read_opt & RD_MSGDIS) {
1318 					freemsg(bp);
1319 				} else {
1320 					bp->b_band = pri;
1321 					if ((mark & _LASTMARK) &&
1322 					    (stp->sd_mark == NULL))
1323 						stp->sd_mark = bp;
1324 					bp->b_flag |= mark & ~_LASTMARK;
1325 					if (delim)
1326 						bp->b_flag |= MSGDELIM;
1327 					if (msgnodata(bp))
1328 						freemsg(bp);
1329 					else
1330 						putback(stp, q, bp, pri);
1331 				}
1332 			} else {
1333 				/*
1334 				 * Consumed the complete message.
1335 				 * Move the MSG*MARKNEXT information
1336 				 * to the stream head just in case
1337 				 * the read queue becomes empty.
1338 				 *
1339 				 * If the stream head was at the mark
1340 				 * (STRATMARK) before we dropped sd_lock above
1341 				 * and some data was consumed then we have
1342 				 * moved past the mark thus STRATMARK is
1343 				 * cleared. However, if a message arrived in
1344 				 * strrput during the copyout above causing
1345 				 * STRATMARK to be set we can not clear that
1346 				 * flag.
1347 				 */
1348 				if (mark &
1349 				    (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
1350 					if (mark & MSGMARKNEXT) {
1351 						stp->sd_flag &= ~STRNOTATMARK;
1352 						stp->sd_flag |= STRATMARK;
1353 					} else if (mark & MSGNOTMARKNEXT) {
1354 						stp->sd_flag &= ~STRATMARK;
1355 						stp->sd_flag |= STRNOTATMARK;
1356 					} else {
1357 						stp->sd_flag &=
1358 						    ~(STRATMARK|STRNOTATMARK);
1359 					}
1360 				} else if (rflg && (old_sd_flag & STRATMARK)) {
1361 					stp->sd_flag &= ~STRATMARK;
1362 				}
1363 			}
1364 
1365 			/*
1366 			 * Check for signal messages at the front of the read
1367 			 * queue and generate the signal(s) if appropriate.
1368 			 * The only signal that can be on queue is M_SIG at
1369 			 * this point.
1370 			 */
1371 			while ((((bp = q->q_first)) != NULL) &&
1372 			    (bp->b_datap->db_type == M_SIG)) {
1373 				bp = getq_noenab(q, 0);
1374 				/*
1375 				 * sd_lock is held so the content of the
1376 				 * read queue can not change.
1377 				 */
1378 				ASSERT(bp != NULL && DB_TYPE(bp) == M_SIG);
1379 				strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
1380 				mutex_exit(&stp->sd_lock);
1381 				freemsg(bp);
1382 				if (STREAM_NEEDSERVICE(stp))
1383 					stream_runservice(stp);
1384 				mutex_enter(&stp->sd_lock);
1385 			}
1386 
1387 			if ((uiop->uio_resid == 0) || (mark & _LASTMARK) ||
1388 			    delim ||
1389 			    (stp->sd_read_opt & (RD_MSGDIS|RD_MSGNODIS))) {
1390 				goto oops;
1391 			}
1392 			continue;
1393 
1394 		case M_SIG:
1395 			strsignal(stp, *bp->b_rptr, (int32_t)bp->b_band);
1396 			freemsg(bp);
1397 			mutex_enter(&stp->sd_lock);
1398 			continue;
1399 
1400 		case M_PROTO:
1401 		case M_PCPROTO:
1402 			/*
1403 			 * Only data messages are readable.
1404 			 * Any others generate an error, unless
1405 			 * RD_PROTDIS or RD_PROTDAT is set.
1406 			 */
1407 			if (stp->sd_read_opt & RD_PROTDAT) {
1408 				for (nbp = bp; nbp; nbp = nbp->b_next) {
1409 					if ((nbp->b_datap->db_type ==
1410 					    M_PROTO) ||
1411 					    (nbp->b_datap->db_type ==
1412 					    M_PCPROTO)) {
1413 						nbp->b_datap->db_type = M_DATA;
1414 					} else {
1415 						break;
1416 					}
1417 				}
1418 				/*
1419 				 * clear stream head hi pri flag based on
1420 				 * first message
1421 				 */
1422 				if (type == M_PCPROTO) {
1423 					mutex_enter(&stp->sd_lock);
1424 					stp->sd_flag &= ~STRPRI;
1425 					mutex_exit(&stp->sd_lock);
1426 				}
1427 				goto ismdata;
1428 			} else if (stp->sd_read_opt & RD_PROTDIS) {
1429 				/*
1430 				 * discard non-data messages
1431 				 */
1432 				while (bp &&
1433 				    ((bp->b_datap->db_type == M_PROTO) ||
1434 				    (bp->b_datap->db_type == M_PCPROTO))) {
1435 					nbp = unlinkb(bp);
1436 					freeb(bp);
1437 					bp = nbp;
1438 				}
1439 				/*
1440 				 * clear stream head hi pri flag based on
1441 				 * first message
1442 				 */
1443 				if (type == M_PCPROTO) {
1444 					mutex_enter(&stp->sd_lock);
1445 					stp->sd_flag &= ~STRPRI;
1446 					mutex_exit(&stp->sd_lock);
1447 				}
1448 				if (bp) {
1449 					bp->b_band = pri;
1450 					goto ismdata;
1451 				} else {
1452 					break;
1453 				}
1454 			}
1455 			/* FALLTHRU */
1456 		case M_PASSFP:
1457 			if ((bp->b_datap->db_type == M_PASSFP) &&
1458 			    (stp->sd_read_opt & RD_PROTDIS)) {
1459 				freemsg(bp);
1460 				break;
1461 			}
1462 			mutex_enter(&stp->sd_lock);
1463 			putback(stp, q, bp, pri);
1464 			mutex_exit(&stp->sd_lock);
1465 			if (rflg == 0)
1466 				error = EBADMSG;
1467 			goto oops1;
1468 
1469 		default:
1470 			/*
1471 			 * Garbage on stream head read queue.
1472 			 */
1473 			cmn_err(CE_WARN, "bad %x found at stream head\n",
1474 			    bp->b_datap->db_type);
1475 			freemsg(bp);
1476 			goto oops1;
1477 		}
1478 		mutex_enter(&stp->sd_lock);
1479 	}
1480 oops:
1481 	mutex_exit(&stp->sd_lock);
1482 oops1:
1483 	qbackenable(q, pri);
1484 	return (error);
1485 #undef	_LASTMARK
1486 }
1487 
1488 /*
1489  * Default processing of M_PROTO/M_PCPROTO messages.
1490  * Determine which wakeups and signals are needed.
1491  * This can be replaced by a user-specified procedure for kernel users
1492  * of STREAMS.
1493  */
1494 /* ARGSUSED */
1495 mblk_t *
1496 strrput_proto(vnode_t *vp, mblk_t *mp,
1497     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1498     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1499 {
1500 	*wakeups = RSLEEP;
1501 	*allmsgsigs = 0;
1502 
1503 	switch (mp->b_datap->db_type) {
1504 	case M_PROTO:
1505 		if (mp->b_band == 0) {
1506 			*firstmsgsigs = S_INPUT | S_RDNORM;
1507 			*pollwakeups = POLLIN | POLLRDNORM;
1508 		} else {
1509 			*firstmsgsigs = S_INPUT | S_RDBAND;
1510 			*pollwakeups = POLLIN | POLLRDBAND;
1511 		}
1512 		break;
1513 	case M_PCPROTO:
1514 		*firstmsgsigs = S_HIPRI;
1515 		*pollwakeups = POLLPRI;
1516 		break;
1517 	}
1518 	return (mp);
1519 }
1520 
1521 /*
1522  * Default processing of everything but M_DATA, M_PROTO, M_PCPROTO and
1523  * M_PASSFP messages.
1524  * Determine which wakeups and signals are needed.
1525  * This can be replaced by a user-specified procedure for kernel users
1526  * of STREAMS.
1527  */
1528 /* ARGSUSED */
1529 mblk_t *
1530 strrput_misc(vnode_t *vp, mblk_t *mp,
1531     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1532     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1533 {
1534 	*wakeups = 0;
1535 	*firstmsgsigs = 0;
1536 	*allmsgsigs = 0;
1537 	*pollwakeups = 0;
1538 	return (mp);
1539 }
1540 
1541 /*
1542  * Stream read put procedure.  Called from downstream driver/module
1543  * with messages for the stream head.  Data, protocol, and in-stream
1544  * signal messages are placed on the queue, others are handled directly.
1545  */
1546 int
1547 strrput(queue_t *q, mblk_t *bp)
1548 {
1549 	struct stdata	*stp;
1550 	ulong_t		rput_opt;
1551 	strwakeup_t	wakeups;
1552 	strsigset_t	firstmsgsigs;	/* Signals if first message on queue */
1553 	strsigset_t	allmsgsigs;	/* Signals for all messages */
1554 	strsigset_t	signals;	/* Signals events to generate */
1555 	strpollset_t	pollwakeups;
1556 	mblk_t		*nextbp;
1557 	uchar_t		band = 0;
1558 	int		hipri_sig;
1559 
1560 	stp = (struct stdata *)q->q_ptr;
1561 	/*
1562 	 * Use rput_opt for optimized access to the SR_ flags except
1563 	 * SR_POLLIN. That flag has to be checked under sd_lock since it
1564 	 * is modified by strpoll().
1565 	 */
1566 	rput_opt = stp->sd_rput_opt;
1567 
1568 	ASSERT(qclaimed(q));
1569 	TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_ENTER,
1570 	    "strrput called with message type:q %p bp %p", q, bp);
1571 
1572 	/*
1573 	 * Perform initial processing and pass to the parameterized functions.
1574 	 */
1575 	ASSERT(bp->b_next == NULL);
1576 
1577 	switch (bp->b_datap->db_type) {
1578 	case M_DATA:
1579 		/*
1580 		 * sockfs is the only consumer of STREOF and when it is set,
1581 		 * it implies that the receiver is not interested in receiving
1582 		 * any more data, hence the mblk is freed to prevent unnecessary
1583 		 * message queueing at the stream head.
1584 		 */
1585 		if (stp->sd_flag == STREOF) {
1586 			freemsg(bp);
1587 			return (0);
1588 		}
1589 		if ((rput_opt & SR_IGN_ZEROLEN) &&
1590 		    bp->b_rptr == bp->b_wptr && msgnodata(bp)) {
1591 			/*
1592 			 * Ignore zero-length M_DATA messages. These might be
1593 			 * generated by some transports.
1594 			 * The zero-length M_DATA messages, even if they
1595 			 * are ignored, should effect the atmark tracking and
1596 			 * should wake up a thread sleeping in strwaitmark.
1597 			 */
1598 			mutex_enter(&stp->sd_lock);
1599 			if (bp->b_flag & MSGMARKNEXT) {
1600 				/*
1601 				 * Record the position of the mark either
1602 				 * in q_last or in STRATMARK.
1603 				 */
1604 				if (q->q_last != NULL) {
1605 					q->q_last->b_flag &= ~MSGNOTMARKNEXT;
1606 					q->q_last->b_flag |= MSGMARKNEXT;
1607 				} else {
1608 					stp->sd_flag &= ~STRNOTATMARK;
1609 					stp->sd_flag |= STRATMARK;
1610 				}
1611 			} else if (bp->b_flag & MSGNOTMARKNEXT) {
1612 				/*
1613 				 * Record that this is not the position of
1614 				 * the mark either in q_last or in
1615 				 * STRNOTATMARK.
1616 				 */
1617 				if (q->q_last != NULL) {
1618 					q->q_last->b_flag &= ~MSGMARKNEXT;
1619 					q->q_last->b_flag |= MSGNOTMARKNEXT;
1620 				} else {
1621 					stp->sd_flag &= ~STRATMARK;
1622 					stp->sd_flag |= STRNOTATMARK;
1623 				}
1624 			}
1625 			if (stp->sd_flag & RSLEEP) {
1626 				stp->sd_flag &= ~RSLEEP;
1627 				cv_broadcast(&q->q_wait);
1628 			}
1629 			mutex_exit(&stp->sd_lock);
1630 			freemsg(bp);
1631 			return (0);
1632 		}
1633 		wakeups = RSLEEP;
1634 		if (bp->b_band == 0) {
1635 			firstmsgsigs = S_INPUT | S_RDNORM;
1636 			pollwakeups = POLLIN | POLLRDNORM;
1637 		} else {
1638 			firstmsgsigs = S_INPUT | S_RDBAND;
1639 			pollwakeups = POLLIN | POLLRDBAND;
1640 		}
1641 		if (rput_opt & SR_SIGALLDATA)
1642 			allmsgsigs = firstmsgsigs;
1643 		else
1644 			allmsgsigs = 0;
1645 
1646 		mutex_enter(&stp->sd_lock);
1647 		if ((rput_opt & SR_CONSOL_DATA) &&
1648 		    (q->q_last != NULL) &&
1649 		    (bp->b_flag & (MSGMARK|MSGDELIM)) == 0) {
1650 			/*
1651 			 * Consolidate an M_DATA message onto an M_DATA,
1652 			 * M_PROTO, or M_PCPROTO by merging it with q_last.
1653 			 * The consolidation does not take place if
1654 			 * the old message is marked with either of the
1655 			 * marks or the delim flag or if the new
1656 			 * message is marked with MSGMARK. The MSGMARK
1657 			 * check is needed to handle the odd semantics of
1658 			 * MSGMARK where essentially the whole message
1659 			 * is to be treated as marked.
1660 			 * Carry any MSGMARKNEXT  and MSGNOTMARKNEXT from the
1661 			 * new message to the front of the b_cont chain.
1662 			 */
1663 			mblk_t *lbp = q->q_last;
1664 			unsigned char db_type = lbp->b_datap->db_type;
1665 
1666 			if ((db_type == M_DATA || db_type == M_PROTO ||
1667 			    db_type == M_PCPROTO) &&
1668 			    !(lbp->b_flag & (MSGDELIM|MSGMARK|MSGMARKNEXT))) {
1669 				rmvq_noenab(q, lbp);
1670 				/*
1671 				 * The first message in the b_cont list
1672 				 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
1673 				 * We need to handle the case where we
1674 				 * are appending:
1675 				 *
1676 				 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
1677 				 * 2) a MSGMARKNEXT to a plain message.
1678 				 * 3) a MSGNOTMARKNEXT to a plain message
1679 				 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
1680 				 *    message.
1681 				 *
1682 				 * Thus we never append a MSGMARKNEXT or
1683 				 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
1684 				 */
1685 				if (bp->b_flag & MSGMARKNEXT) {
1686 					lbp->b_flag |= MSGMARKNEXT;
1687 					lbp->b_flag &= ~MSGNOTMARKNEXT;
1688 					bp->b_flag &= ~MSGMARKNEXT;
1689 				} else if (bp->b_flag & MSGNOTMARKNEXT) {
1690 					lbp->b_flag |= MSGNOTMARKNEXT;
1691 					bp->b_flag &= ~MSGNOTMARKNEXT;
1692 				}
1693 
1694 				linkb(lbp, bp);
1695 				bp = lbp;
1696 				/*
1697 				 * The new message logically isn't the first
1698 				 * even though the q_first check below thinks
1699 				 * it is. Clear the firstmsgsigs to make it
1700 				 * not appear to be first.
1701 				 */
1702 				firstmsgsigs = 0;
1703 			}
1704 		}
1705 		break;
1706 
1707 	case M_PASSFP:
1708 		wakeups = RSLEEP;
1709 		allmsgsigs = 0;
1710 		if (bp->b_band == 0) {
1711 			firstmsgsigs = S_INPUT | S_RDNORM;
1712 			pollwakeups = POLLIN | POLLRDNORM;
1713 		} else {
1714 			firstmsgsigs = S_INPUT | S_RDBAND;
1715 			pollwakeups = POLLIN | POLLRDBAND;
1716 		}
1717 		mutex_enter(&stp->sd_lock);
1718 		break;
1719 
1720 	case M_PROTO:
1721 	case M_PCPROTO:
1722 		ASSERT(stp->sd_rprotofunc != NULL);
1723 		bp = (stp->sd_rprotofunc)(stp->sd_vnode, bp,
1724 		    &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1725 #define	ALLSIG	(S_INPUT|S_HIPRI|S_OUTPUT|S_MSG|S_ERROR|S_HANGUP|S_RDNORM|\
1726 		S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)
1727 #define	ALLPOLL	(POLLIN|POLLPRI|POLLOUT|POLLRDNORM|POLLWRNORM|POLLRDBAND|\
1728 		POLLWRBAND)
1729 
1730 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1731 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1732 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1733 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1734 
1735 		mutex_enter(&stp->sd_lock);
1736 		break;
1737 
1738 	default:
1739 		ASSERT(stp->sd_rmiscfunc != NULL);
1740 		bp = (stp->sd_rmiscfunc)(stp->sd_vnode, bp,
1741 		    &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1742 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1743 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1744 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1745 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1746 #undef	ALLSIG
1747 #undef	ALLPOLL
1748 		mutex_enter(&stp->sd_lock);
1749 		break;
1750 	}
1751 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1752 
1753 	/* By default generate superset of signals */
1754 	signals = (firstmsgsigs | allmsgsigs);
1755 
1756 	/*
1757 	 * The  proto and misc functions can return multiple messages
1758 	 * as a b_next chain. Such messages are processed separately.
1759 	 */
1760 one_more:
1761 	hipri_sig = 0;
1762 	if (bp == NULL) {
1763 		nextbp = NULL;
1764 	} else {
1765 		nextbp = bp->b_next;
1766 		bp->b_next = NULL;
1767 
1768 		switch (bp->b_datap->db_type) {
1769 		case M_PCPROTO:
1770 			/*
1771 			 * Only one priority protocol message is allowed at the
1772 			 * stream head at a time.
1773 			 */
1774 			if (stp->sd_flag & STRPRI) {
1775 				TRACE_0(TR_FAC_STREAMS_FR, TR_STRRPUT_PROTERR,
1776 				    "M_PCPROTO already at head");
1777 				freemsg(bp);
1778 				mutex_exit(&stp->sd_lock);
1779 				goto done;
1780 			}
1781 			stp->sd_flag |= STRPRI;
1782 			hipri_sig = 1;
1783 			/* FALLTHRU */
1784 		case M_DATA:
1785 		case M_PROTO:
1786 		case M_PASSFP:
1787 			band = bp->b_band;
1788 			/*
1789 			 * Marking doesn't work well when messages
1790 			 * are marked in more than one band.  We only
1791 			 * remember the last message received, even if
1792 			 * it is placed on the queue ahead of other
1793 			 * marked messages.
1794 			 */
1795 			if (bp->b_flag & MSGMARK)
1796 				stp->sd_mark = bp;
1797 			(void) putq(q, bp);
1798 
1799 			/*
1800 			 * If message is a PCPROTO message, always use
1801 			 * firstmsgsigs to determine if a signal should be
1802 			 * sent as strrput is the only place to send
1803 			 * signals for PCPROTO. Other messages are based on
1804 			 * the STRGETINPROG flag. The flag determines if
1805 			 * strrput or (k)strgetmsg will be responsible for
1806 			 * sending the signals, in the firstmsgsigs case.
1807 			 */
1808 			if ((hipri_sig == 1) ||
1809 			    (((stp->sd_flag & STRGETINPROG) == 0) &&
1810 			    (q->q_first == bp)))
1811 				signals = (firstmsgsigs | allmsgsigs);
1812 			else
1813 				signals = allmsgsigs;
1814 			break;
1815 
1816 		default:
1817 			mutex_exit(&stp->sd_lock);
1818 			(void) strrput_nondata(q, bp);
1819 			mutex_enter(&stp->sd_lock);
1820 			break;
1821 		}
1822 	}
1823 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1824 	/*
1825 	 * Wake sleeping read/getmsg and cancel deferred wakeup
1826 	 */
1827 	if (wakeups & RSLEEP)
1828 		stp->sd_wakeq &= ~RSLEEP;
1829 
1830 	wakeups &= stp->sd_flag;
1831 	if (wakeups & RSLEEP) {
1832 		stp->sd_flag &= ~RSLEEP;
1833 		cv_broadcast(&q->q_wait);
1834 	}
1835 	if (wakeups & WSLEEP) {
1836 		stp->sd_flag &= ~WSLEEP;
1837 		cv_broadcast(&_WR(q)->q_wait);
1838 	}
1839 
1840 	if (pollwakeups != 0) {
1841 		if (pollwakeups == (POLLIN | POLLRDNORM)) {
1842 			/*
1843 			 * Can't use rput_opt since it was not
1844 			 * read when sd_lock was held and SR_POLLIN is changed
1845 			 * by strpoll() under sd_lock.
1846 			 */
1847 			if (!(stp->sd_rput_opt & SR_POLLIN))
1848 				goto no_pollwake;
1849 			stp->sd_rput_opt &= ~SR_POLLIN;
1850 		}
1851 		mutex_exit(&stp->sd_lock);
1852 		pollwakeup(&stp->sd_pollist, pollwakeups);
1853 		mutex_enter(&stp->sd_lock);
1854 	}
1855 no_pollwake:
1856 
1857 	/*
1858 	 * strsendsig can handle multiple signals with a
1859 	 * single call.
1860 	 */
1861 	if (stp->sd_sigflags & signals)
1862 		strsendsig(stp->sd_siglist, signals, band, 0);
1863 	mutex_exit(&stp->sd_lock);
1864 
1865 
1866 done:
1867 	if (nextbp == NULL)
1868 		return (0);
1869 
1870 	/*
1871 	 * Any signals were handled the first time.
1872 	 * Wakeups and pollwakeups are redone to avoid any race
1873 	 * conditions - all the messages are not queued until the
1874 	 * last message has been processed by strrput.
1875 	 */
1876 	bp = nextbp;
1877 	signals = firstmsgsigs = allmsgsigs = 0;
1878 	mutex_enter(&stp->sd_lock);
1879 	goto one_more;
1880 }
1881 
1882 static void
1883 log_dupioc(queue_t *rq, mblk_t *bp)
1884 {
1885 	queue_t *wq, *qp;
1886 	char *modnames, *mnp, *dname;
1887 	size_t maxmodstr;
1888 	boolean_t islast;
1889 
1890 	/*
1891 	 * Allocate a buffer large enough to hold the names of nstrpush modules
1892 	 * and one driver, with spaces between and NUL terminator.  If we can't
1893 	 * get memory, then we'll just log the driver name.
1894 	 */
1895 	maxmodstr = nstrpush * (FMNAMESZ + 1);
1896 	mnp = modnames = kmem_alloc(maxmodstr, KM_NOSLEEP);
1897 
1898 	/* march down write side to print log message down to the driver */
1899 	wq = WR(rq);
1900 
1901 	/* make sure q_next doesn't shift around while we're grabbing data */
1902 	claimstr(wq);
1903 	qp = wq->q_next;
1904 	do {
1905 		dname = Q2NAME(qp);
1906 		islast = !SAMESTR(qp) || qp->q_next == NULL;
1907 		if (modnames == NULL) {
1908 			/*
1909 			 * If we don't have memory, then get the driver name in
1910 			 * the log where we can see it.  Note that memory
1911 			 * pressure is a possible cause of these sorts of bugs.
1912 			 */
1913 			if (islast) {
1914 				modnames = dname;
1915 				maxmodstr = 0;
1916 			}
1917 		} else {
1918 			mnp += snprintf(mnp, FMNAMESZ + 1, "%s", dname);
1919 			if (!islast)
1920 				*mnp++ = ' ';
1921 		}
1922 		qp = qp->q_next;
1923 	} while (!islast);
1924 	releasestr(wq);
1925 	/* Cannot happen unless stream head is corrupt. */
1926 	ASSERT(modnames != NULL);
1927 	(void) strlog(rq->q_qinfo->qi_minfo->mi_idnum, 0, 1,
1928 	    SL_CONSOLE|SL_TRACE|SL_ERROR,
1929 	    "Warning: stream %p received duplicate %X M_IOC%s; module list: %s",
1930 	    rq->q_ptr, ((struct iocblk *)bp->b_rptr)->ioc_cmd,
1931 	    (DB_TYPE(bp) == M_IOCACK ? "ACK" : "NAK"), modnames);
1932 	if (maxmodstr != 0)
1933 		kmem_free(modnames, maxmodstr);
1934 }
1935 
1936 int
1937 strrput_nondata(queue_t *q, mblk_t *bp)
1938 {
1939 	struct stdata *stp;
1940 	struct iocblk *iocbp;
1941 	struct stroptions *sop;
1942 	struct copyreq *reqp;
1943 	struct copyresp *resp;
1944 	unsigned char bpri;
1945 	unsigned char  flushed_already = 0;
1946 
1947 	stp = (struct stdata *)q->q_ptr;
1948 
1949 	ASSERT(!(stp->sd_flag & STPLEX));
1950 	ASSERT(qclaimed(q));
1951 
1952 	switch (bp->b_datap->db_type) {
1953 	case M_ERROR:
1954 		/*
1955 		 * An error has occurred downstream, the errno is in the first
1956 		 * bytes of the message.
1957 		 */
1958 		if ((bp->b_wptr - bp->b_rptr) == 2) {	/* New flavor */
1959 			unsigned char rw = 0;
1960 
1961 			mutex_enter(&stp->sd_lock);
1962 			if (*bp->b_rptr != NOERROR) {	/* read error */
1963 				if (*bp->b_rptr != 0) {
1964 					if (stp->sd_flag & STRDERR)
1965 						flushed_already |= FLUSHR;
1966 					stp->sd_flag |= STRDERR;
1967 					rw |= FLUSHR;
1968 				} else {
1969 					stp->sd_flag &= ~STRDERR;
1970 				}
1971 				stp->sd_rerror = *bp->b_rptr;
1972 			}
1973 			bp->b_rptr++;
1974 			if (*bp->b_rptr != NOERROR) {	/* write error */
1975 				if (*bp->b_rptr != 0) {
1976 					if (stp->sd_flag & STWRERR)
1977 						flushed_already |= FLUSHW;
1978 					stp->sd_flag |= STWRERR;
1979 					rw |= FLUSHW;
1980 				} else {
1981 					stp->sd_flag &= ~STWRERR;
1982 				}
1983 				stp->sd_werror = *bp->b_rptr;
1984 			}
1985 			if (rw) {
1986 				TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_WAKE,
1987 				    "strrput cv_broadcast:q %p, bp %p",
1988 				    q, bp);
1989 				cv_broadcast(&q->q_wait); /* readers */
1990 				cv_broadcast(&_WR(q)->q_wait); /* writers */
1991 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
1992 
1993 				mutex_exit(&stp->sd_lock);
1994 				pollwakeup(&stp->sd_pollist, POLLERR);
1995 				mutex_enter(&stp->sd_lock);
1996 
1997 				if (stp->sd_sigflags & S_ERROR)
1998 					strsendsig(stp->sd_siglist, S_ERROR, 0,
1999 					    ((rw & FLUSHR) ? stp->sd_rerror :
2000 					    stp->sd_werror));
2001 				mutex_exit(&stp->sd_lock);
2002 				/*
2003 				 * Send the M_FLUSH only
2004 				 * for the first M_ERROR
2005 				 * message on the stream
2006 				 */
2007 				if (flushed_already == rw) {
2008 					freemsg(bp);
2009 					return (0);
2010 				}
2011 
2012 				bp->b_datap->db_type = M_FLUSH;
2013 				*bp->b_rptr = rw;
2014 				bp->b_wptr = bp->b_rptr + 1;
2015 				/*
2016 				 * Protect against the driver
2017 				 * passing up messages after
2018 				 * it has done a qprocsoff
2019 				 */
2020 				if (_OTHERQ(q)->q_next == NULL)
2021 					freemsg(bp);
2022 				else
2023 					qreply(q, bp);
2024 				return (0);
2025 			} else
2026 				mutex_exit(&stp->sd_lock);
2027 		} else if (*bp->b_rptr != 0) {		/* Old flavor */
2028 				if (stp->sd_flag & (STRDERR|STWRERR))
2029 					flushed_already = FLUSHRW;
2030 				mutex_enter(&stp->sd_lock);
2031 				stp->sd_flag |= (STRDERR|STWRERR);
2032 				stp->sd_rerror = *bp->b_rptr;
2033 				stp->sd_werror = *bp->b_rptr;
2034 				TRACE_2(TR_FAC_STREAMS_FR,
2035 				    TR_STRRPUT_WAKE2,
2036 				    "strrput wakeup #2:q %p, bp %p", q, bp);
2037 				cv_broadcast(&q->q_wait); /* the readers */
2038 				cv_broadcast(&_WR(q)->q_wait); /* the writers */
2039 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
2040 
2041 				mutex_exit(&stp->sd_lock);
2042 				pollwakeup(&stp->sd_pollist, POLLERR);
2043 				mutex_enter(&stp->sd_lock);
2044 
2045 				if (stp->sd_sigflags & S_ERROR)
2046 					strsendsig(stp->sd_siglist, S_ERROR, 0,
2047 					    (stp->sd_werror ? stp->sd_werror :
2048 					    stp->sd_rerror));
2049 				mutex_exit(&stp->sd_lock);
2050 
2051 				/*
2052 				 * Send the M_FLUSH only
2053 				 * for the first M_ERROR
2054 				 * message on the stream
2055 				 */
2056 				if (flushed_already != FLUSHRW) {
2057 					bp->b_datap->db_type = M_FLUSH;
2058 					*bp->b_rptr = FLUSHRW;
2059 					/*
2060 					 * Protect against the driver passing up
2061 					 * messages after it has done a
2062 					 * qprocsoff.
2063 					 */
2064 					if (_OTHERQ(q)->q_next == NULL)
2065 						freemsg(bp);
2066 					else
2067 						qreply(q, bp);
2068 					return (0);
2069 				}
2070 		}
2071 		freemsg(bp);
2072 		return (0);
2073 
2074 	case M_HANGUP:
2075 
2076 		freemsg(bp);
2077 		mutex_enter(&stp->sd_lock);
2078 		stp->sd_werror = ENXIO;
2079 		stp->sd_flag |= STRHUP;
2080 		stp->sd_flag &= ~(WSLEEP|RSLEEP);
2081 
2082 		/*
2083 		 * send signal if controlling tty
2084 		 */
2085 
2086 		if (stp->sd_sidp) {
2087 			prsignal(stp->sd_sidp, SIGHUP);
2088 			if (stp->sd_sidp != stp->sd_pgidp)
2089 				pgsignal(stp->sd_pgidp, SIGTSTP);
2090 		}
2091 
2092 		/*
2093 		 * wake up read, write, and exception pollers and
2094 		 * reset wakeup mechanism.
2095 		 */
2096 		cv_broadcast(&q->q_wait);	/* the readers */
2097 		cv_broadcast(&_WR(q)->q_wait);	/* the writers */
2098 		cv_broadcast(&stp->sd_monitor);	/* the ioctllers */
2099 		strhup(stp);
2100 		mutex_exit(&stp->sd_lock);
2101 		return (0);
2102 
2103 	case M_UNHANGUP:
2104 		freemsg(bp);
2105 		mutex_enter(&stp->sd_lock);
2106 		stp->sd_werror = 0;
2107 		stp->sd_flag &= ~STRHUP;
2108 		mutex_exit(&stp->sd_lock);
2109 		return (0);
2110 
2111 	case M_SIG:
2112 		/*
2113 		 * Someone downstream wants to post a signal.  The
2114 		 * signal to post is contained in the first byte of the
2115 		 * message.  If the message would go on the front of
2116 		 * the queue, send a signal to the process group
2117 		 * (if not SIGPOLL) or to the siglist processes
2118 		 * (SIGPOLL).  If something is already on the queue,
2119 		 * OR if we are delivering a delayed suspend (*sigh*
2120 		 * another "tty" hack) and there's no one sleeping already,
2121 		 * just enqueue the message.
2122 		 */
2123 		mutex_enter(&stp->sd_lock);
2124 		if (q->q_first || (*bp->b_rptr == SIGTSTP &&
2125 		    !(stp->sd_flag & RSLEEP))) {
2126 			(void) putq(q, bp);
2127 			mutex_exit(&stp->sd_lock);
2128 			return (0);
2129 		}
2130 		mutex_exit(&stp->sd_lock);
2131 		/* FALLTHRU */
2132 
2133 	case M_PCSIG:
2134 		/*
2135 		 * Don't enqueue, just post the signal.
2136 		 */
2137 		strsignal(stp, *bp->b_rptr, 0L);
2138 		freemsg(bp);
2139 		return (0);
2140 
2141 	case M_CMD:
2142 		if (MBLKL(bp) != sizeof (cmdblk_t)) {
2143 			freemsg(bp);
2144 			return (0);
2145 		}
2146 
2147 		mutex_enter(&stp->sd_lock);
2148 		if (stp->sd_flag & STRCMDWAIT) {
2149 			ASSERT(stp->sd_cmdblk == NULL);
2150 			stp->sd_cmdblk = bp;
2151 			cv_broadcast(&stp->sd_monitor);
2152 			mutex_exit(&stp->sd_lock);
2153 		} else {
2154 			mutex_exit(&stp->sd_lock);
2155 			freemsg(bp);
2156 		}
2157 		return (0);
2158 
2159 	case M_FLUSH:
2160 		/*
2161 		 * Flush queues.  The indication of which queues to flush
2162 		 * is in the first byte of the message.  If the read queue
2163 		 * is specified, then flush it.  If FLUSHBAND is set, just
2164 		 * flush the band specified by the second byte of the message.
2165 		 *
2166 		 * If a module has issued a M_SETOPT to not flush hi
2167 		 * priority messages off of the stream head, then pass this
2168 		 * flag into the flushq code to preserve such messages.
2169 		 */
2170 
2171 		if (*bp->b_rptr & FLUSHR) {
2172 			mutex_enter(&stp->sd_lock);
2173 			if (*bp->b_rptr & FLUSHBAND) {
2174 				ASSERT((bp->b_wptr - bp->b_rptr) >= 2);
2175 				flushband(q, *(bp->b_rptr + 1), FLUSHALL);
2176 			} else
2177 				flushq_common(q, FLUSHALL,
2178 				    stp->sd_read_opt & RFLUSHPCPROT);
2179 			if ((q->q_first == NULL) ||
2180 			    (q->q_first->b_datap->db_type < QPCTL))
2181 				stp->sd_flag &= ~STRPRI;
2182 			else {
2183 				ASSERT(stp->sd_flag & STRPRI);
2184 			}
2185 			mutex_exit(&stp->sd_lock);
2186 		}
2187 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
2188 			*bp->b_rptr &= ~FLUSHR;
2189 			bp->b_flag |= MSGNOLOOP;
2190 			/*
2191 			 * Protect against the driver passing up
2192 			 * messages after it has done a qprocsoff.
2193 			 */
2194 			if (_OTHERQ(q)->q_next == NULL)
2195 				freemsg(bp);
2196 			else
2197 				qreply(q, bp);
2198 			return (0);
2199 		}
2200 		freemsg(bp);
2201 		return (0);
2202 
2203 	case M_IOCACK:
2204 	case M_IOCNAK:
2205 		iocbp = (struct iocblk *)bp->b_rptr;
2206 		/*
2207 		 * If not waiting for ACK or NAK then just free msg.
2208 		 * If incorrect id sequence number then just free msg.
2209 		 * If already have ACK or NAK for user then this is a
2210 		 *    duplicate, display a warning and free the msg.
2211 		 */
2212 		mutex_enter(&stp->sd_lock);
2213 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2214 		    (stp->sd_iocid != iocbp->ioc_id)) {
2215 			/*
2216 			 * If the ACK/NAK is a dup, display a message
2217 			 * Dup is when sd_iocid == ioc_id, and
2218 			 * sd_iocblk == <valid ptr> or -1 (the former
2219 			 * is when an ioctl has been put on the stream
2220 			 * head, but has not yet been consumed, the
2221 			 * later is when it has been consumed).
2222 			 */
2223 			if ((stp->sd_iocid == iocbp->ioc_id) &&
2224 			    (stp->sd_iocblk != NULL)) {
2225 				log_dupioc(q, bp);
2226 			}
2227 			freemsg(bp);
2228 			mutex_exit(&stp->sd_lock);
2229 			return (0);
2230 		}
2231 
2232 		/*
2233 		 * Assign ACK or NAK to user and wake up.
2234 		 */
2235 		stp->sd_iocblk = bp;
2236 		cv_broadcast(&stp->sd_monitor);
2237 		mutex_exit(&stp->sd_lock);
2238 		return (0);
2239 
2240 	case M_COPYIN:
2241 	case M_COPYOUT:
2242 		reqp = (struct copyreq *)bp->b_rptr;
2243 
2244 		/*
2245 		 * If not waiting for ACK or NAK then just fail request.
2246 		 * If already have ACK, NAK, or copy request, then just
2247 		 * fail request.
2248 		 * If incorrect id sequence number then just fail request.
2249 		 */
2250 		mutex_enter(&stp->sd_lock);
2251 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2252 		    (stp->sd_iocid != reqp->cq_id)) {
2253 			if (bp->b_cont) {
2254 				freemsg(bp->b_cont);
2255 				bp->b_cont = NULL;
2256 			}
2257 			bp->b_datap->db_type = M_IOCDATA;
2258 			bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
2259 			resp = (struct copyresp *)bp->b_rptr;
2260 			resp->cp_rval = (caddr_t)1;	/* failure */
2261 			mutex_exit(&stp->sd_lock);
2262 			putnext(stp->sd_wrq, bp);
2263 			return (0);
2264 		}
2265 
2266 		/*
2267 		 * Assign copy request to user and wake up.
2268 		 */
2269 		stp->sd_iocblk = bp;
2270 		cv_broadcast(&stp->sd_monitor);
2271 		mutex_exit(&stp->sd_lock);
2272 		return (0);
2273 
2274 	case M_SETOPTS:
2275 		/*
2276 		 * Set stream head options (read option, write offset,
2277 		 * min/max packet size, and/or high/low water marks for
2278 		 * the read side only).
2279 		 */
2280 
2281 		bpri = 0;
2282 		sop = (struct stroptions *)bp->b_rptr;
2283 		mutex_enter(&stp->sd_lock);
2284 		if (sop->so_flags & SO_READOPT) {
2285 			switch (sop->so_readopt & RMODEMASK) {
2286 			case RNORM:
2287 				stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
2288 				break;
2289 
2290 			case RMSGD:
2291 				stp->sd_read_opt =
2292 				    ((stp->sd_read_opt & ~RD_MSGNODIS) |
2293 				    RD_MSGDIS);
2294 				break;
2295 
2296 			case RMSGN:
2297 				stp->sd_read_opt =
2298 				    ((stp->sd_read_opt & ~RD_MSGDIS) |
2299 				    RD_MSGNODIS);
2300 				break;
2301 			}
2302 			switch (sop->so_readopt & RPROTMASK) {
2303 			case RPROTNORM:
2304 				stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
2305 				break;
2306 
2307 			case RPROTDAT:
2308 				stp->sd_read_opt =
2309 				    ((stp->sd_read_opt & ~RD_PROTDIS) |
2310 				    RD_PROTDAT);
2311 				break;
2312 
2313 			case RPROTDIS:
2314 				stp->sd_read_opt =
2315 				    ((stp->sd_read_opt & ~RD_PROTDAT) |
2316 				    RD_PROTDIS);
2317 				break;
2318 			}
2319 			switch (sop->so_readopt & RFLUSHMASK) {
2320 			case RFLUSHPCPROT:
2321 				/*
2322 				 * This sets the stream head to NOT flush
2323 				 * M_PCPROTO messages.
2324 				 */
2325 				stp->sd_read_opt |= RFLUSHPCPROT;
2326 				break;
2327 			}
2328 		}
2329 		if (sop->so_flags & SO_ERROPT) {
2330 			switch (sop->so_erropt & RERRMASK) {
2331 			case RERRNORM:
2332 				stp->sd_flag &= ~STRDERRNONPERSIST;
2333 				break;
2334 			case RERRNONPERSIST:
2335 				stp->sd_flag |= STRDERRNONPERSIST;
2336 				break;
2337 			}
2338 			switch (sop->so_erropt & WERRMASK) {
2339 			case WERRNORM:
2340 				stp->sd_flag &= ~STWRERRNONPERSIST;
2341 				break;
2342 			case WERRNONPERSIST:
2343 				stp->sd_flag |= STWRERRNONPERSIST;
2344 				break;
2345 			}
2346 		}
2347 		if (sop->so_flags & SO_COPYOPT) {
2348 			if (sop->so_copyopt & ZCVMSAFE) {
2349 				stp->sd_copyflag |= STZCVMSAFE;
2350 				stp->sd_copyflag &= ~STZCVMUNSAFE;
2351 			} else if (sop->so_copyopt & ZCVMUNSAFE) {
2352 				stp->sd_copyflag |= STZCVMUNSAFE;
2353 				stp->sd_copyflag &= ~STZCVMSAFE;
2354 			}
2355 
2356 			if (sop->so_copyopt & COPYCACHED) {
2357 				stp->sd_copyflag |= STRCOPYCACHED;
2358 			}
2359 		}
2360 		if (sop->so_flags & SO_WROFF)
2361 			stp->sd_wroff = sop->so_wroff;
2362 		if (sop->so_flags & SO_TAIL)
2363 			stp->sd_tail = sop->so_tail;
2364 		if (sop->so_flags & SO_MINPSZ)
2365 			q->q_minpsz = sop->so_minpsz;
2366 		if (sop->so_flags & SO_MAXPSZ)
2367 			q->q_maxpsz = sop->so_maxpsz;
2368 		if (sop->so_flags & SO_MAXBLK)
2369 			stp->sd_maxblk = sop->so_maxblk;
2370 		if (sop->so_flags & SO_HIWAT) {
2371 			if (sop->so_flags & SO_BAND) {
2372 				if (strqset(q, QHIWAT,
2373 				    sop->so_band, sop->so_hiwat)) {
2374 					cmn_err(CE_WARN, "strrput: could not "
2375 					    "allocate qband\n");
2376 				} else {
2377 					bpri = sop->so_band;
2378 				}
2379 			} else {
2380 				q->q_hiwat = sop->so_hiwat;
2381 			}
2382 		}
2383 		if (sop->so_flags & SO_LOWAT) {
2384 			if (sop->so_flags & SO_BAND) {
2385 				if (strqset(q, QLOWAT,
2386 				    sop->so_band, sop->so_lowat)) {
2387 					cmn_err(CE_WARN, "strrput: could not "
2388 					    "allocate qband\n");
2389 				} else {
2390 					bpri = sop->so_band;
2391 				}
2392 			} else {
2393 				q->q_lowat = sop->so_lowat;
2394 			}
2395 		}
2396 		if (sop->so_flags & SO_MREADON)
2397 			stp->sd_flag |= SNDMREAD;
2398 		if (sop->so_flags & SO_MREADOFF)
2399 			stp->sd_flag &= ~SNDMREAD;
2400 		if (sop->so_flags & SO_NDELON)
2401 			stp->sd_flag |= OLDNDELAY;
2402 		if (sop->so_flags & SO_NDELOFF)
2403 			stp->sd_flag &= ~OLDNDELAY;
2404 		if (sop->so_flags & SO_ISTTY)
2405 			stp->sd_flag |= STRISTTY;
2406 		if (sop->so_flags & SO_ISNTTY)
2407 			stp->sd_flag &= ~STRISTTY;
2408 		if (sop->so_flags & SO_TOSTOP)
2409 			stp->sd_flag |= STRTOSTOP;
2410 		if (sop->so_flags & SO_TONSTOP)
2411 			stp->sd_flag &= ~STRTOSTOP;
2412 		if (sop->so_flags & SO_DELIM)
2413 			stp->sd_flag |= STRDELIM;
2414 		if (sop->so_flags & SO_NODELIM)
2415 			stp->sd_flag &= ~STRDELIM;
2416 
2417 		mutex_exit(&stp->sd_lock);
2418 		freemsg(bp);
2419 
2420 		/* Check backenable in case the water marks changed */
2421 		qbackenable(q, bpri);
2422 		return (0);
2423 
2424 	/*
2425 	 * The following set of cases deal with situations where two stream
2426 	 * heads are connected to each other (twisted streams).  These messages
2427 	 * have no meaning at the stream head.
2428 	 */
2429 	case M_BREAK:
2430 	case M_CTL:
2431 	case M_DELAY:
2432 	case M_START:
2433 	case M_STOP:
2434 	case M_IOCDATA:
2435 	case M_STARTI:
2436 	case M_STOPI:
2437 		freemsg(bp);
2438 		return (0);
2439 
2440 	case M_IOCTL:
2441 		/*
2442 		 * Always NAK this condition
2443 		 * (makes no sense)
2444 		 * If there is one or more threads in the read side
2445 		 * rwnext we have to defer the nacking until that thread
2446 		 * returns (in strget).
2447 		 */
2448 		mutex_enter(&stp->sd_lock);
2449 		if (stp->sd_struiodnak != 0) {
2450 			/*
2451 			 * Defer NAK to the streamhead. Queue at the end
2452 			 * the list.
2453 			 */
2454 			mblk_t *mp = stp->sd_struionak;
2455 
2456 			while (mp && mp->b_next)
2457 				mp = mp->b_next;
2458 			if (mp)
2459 				mp->b_next = bp;
2460 			else
2461 				stp->sd_struionak = bp;
2462 			bp->b_next = NULL;
2463 			mutex_exit(&stp->sd_lock);
2464 			return (0);
2465 		}
2466 		mutex_exit(&stp->sd_lock);
2467 
2468 		bp->b_datap->db_type = M_IOCNAK;
2469 		/*
2470 		 * Protect against the driver passing up
2471 		 * messages after it has done a qprocsoff.
2472 		 */
2473 		if (_OTHERQ(q)->q_next == NULL)
2474 			freemsg(bp);
2475 		else
2476 			qreply(q, bp);
2477 		return (0);
2478 
2479 	default:
2480 #ifdef DEBUG
2481 		cmn_err(CE_WARN,
2482 		    "bad message type %x received at stream head\n",
2483 		    bp->b_datap->db_type);
2484 #endif
2485 		freemsg(bp);
2486 		return (0);
2487 	}
2488 
2489 	/* NOTREACHED */
2490 }
2491 
2492 /*
2493  * Check if the stream pointed to by `stp' can be written to, and return an
2494  * error code if not.  If `eiohup' is set, then return EIO if STRHUP is set.
2495  * If `sigpipeok' is set and the SW_SIGPIPE option is enabled on the stream,
2496  * then always return EPIPE and send a SIGPIPE to the invoking thread.
2497  */
2498 static int
2499 strwriteable(struct stdata *stp, boolean_t eiohup, boolean_t sigpipeok)
2500 {
2501 	int error;
2502 
2503 	ASSERT(MUTEX_HELD(&stp->sd_lock));
2504 
2505 	/*
2506 	 * For modem support, POSIX states that on writes, EIO should
2507 	 * be returned if the stream has been hung up.
2508 	 */
2509 	if (eiohup && (stp->sd_flag & (STPLEX|STRHUP)) == STRHUP)
2510 		error = EIO;
2511 	else
2512 		error = strgeterr(stp, STRHUP|STPLEX|STWRERR, 0);
2513 
2514 	if (error != 0) {
2515 		if (!(stp->sd_flag & STPLEX) &&
2516 		    (stp->sd_wput_opt & SW_SIGPIPE) && sigpipeok) {
2517 			tsignal(curthread, SIGPIPE);
2518 			error = EPIPE;
2519 		}
2520 	}
2521 
2522 	return (error);
2523 }
2524 
2525 /*
2526  * Copyin and send data down a stream.
2527  * The caller will allocate and copyin any control part that precedes the
2528  * message and pass that in as mctl.
2529  *
2530  * Caller should *not* hold sd_lock.
2531  * When EWOULDBLOCK is returned the caller has to redo the canputnext
2532  * under sd_lock in order to avoid missing a backenabling wakeup.
2533  *
2534  * Use iosize = -1 to not send any M_DATA. iosize = 0 sends zero-length M_DATA.
2535  *
2536  * Set MSG_IGNFLOW in flags to ignore flow control for hipri messages.
2537  * For sync streams we can only ignore flow control by reverting to using
2538  * putnext.
2539  *
2540  * If sd_maxblk is less than *iosize this routine might return without
2541  * transferring all of *iosize. In all cases, on return *iosize will contain
2542  * the amount of data that was transferred.
2543  */
2544 static int
2545 strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
2546     int b_flag, int pri, int flags)
2547 {
2548 	struiod_t uiod;
2549 	mblk_t *mp;
2550 	queue_t *wqp = stp->sd_wrq;
2551 	int error = 0;
2552 	ssize_t count = *iosize;
2553 
2554 	ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
2555 
2556 	if (uiop != NULL && count >= 0)
2557 		flags |= stp->sd_struiowrq ? STRUIO_POSTPONE : 0;
2558 
2559 	if (!(flags & STRUIO_POSTPONE)) {
2560 		/*
2561 		 * Use regular canputnext, strmakedata, putnext sequence.
2562 		 */
2563 		if (pri == 0) {
2564 			if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2565 				freemsg(mctl);
2566 				return (EWOULDBLOCK);
2567 			}
2568 		} else {
2569 			if (!(flags & MSG_IGNFLOW) && !bcanputnext(wqp, pri)) {
2570 				freemsg(mctl);
2571 				return (EWOULDBLOCK);
2572 			}
2573 		}
2574 
2575 		if ((error = strmakedata(iosize, uiop, stp, flags,
2576 		    &mp)) != 0) {
2577 			freemsg(mctl);
2578 			/*
2579 			 * need to change return code to ENOMEM
2580 			 * so that this is not confused with
2581 			 * flow control, EAGAIN.
2582 			 */
2583 
2584 			if (error == EAGAIN)
2585 				return (ENOMEM);
2586 			else
2587 				return (error);
2588 		}
2589 		if (mctl != NULL) {
2590 			if (mctl->b_cont == NULL)
2591 				mctl->b_cont = mp;
2592 			else if (mp != NULL)
2593 				linkb(mctl, mp);
2594 			mp = mctl;
2595 		} else if (mp == NULL)
2596 			return (0);
2597 
2598 		mp->b_flag |= b_flag;
2599 		mp->b_band = (uchar_t)pri;
2600 
2601 		if (flags & MSG_IGNFLOW) {
2602 			/*
2603 			 * XXX Hack: Don't get stuck running service
2604 			 * procedures. This is needed for sockfs when
2605 			 * sending the unbind message out of the rput
2606 			 * procedure - we don't want a put procedure
2607 			 * to run service procedures.
2608 			 */
2609 			putnext(wqp, mp);
2610 		} else {
2611 			stream_willservice(stp);
2612 			putnext(wqp, mp);
2613 			stream_runservice(stp);
2614 		}
2615 		return (0);
2616 	}
2617 	/*
2618 	 * Stream supports rwnext() for the write side.
2619 	 */
2620 	if ((error = strmakedata(iosize, uiop, stp, flags, &mp)) != 0) {
2621 		freemsg(mctl);
2622 		/*
2623 		 * map EAGAIN to ENOMEM since EAGAIN means "flow controlled".
2624 		 */
2625 		return (error == EAGAIN ? ENOMEM : error);
2626 	}
2627 	if (mctl != NULL) {
2628 		if (mctl->b_cont == NULL)
2629 			mctl->b_cont = mp;
2630 		else if (mp != NULL)
2631 			linkb(mctl, mp);
2632 		mp = mctl;
2633 	} else if (mp == NULL) {
2634 		return (0);
2635 	}
2636 
2637 	mp->b_flag |= b_flag;
2638 	mp->b_band = (uchar_t)pri;
2639 
2640 	(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
2641 	    sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
2642 	uiod.d_uio.uio_offset = 0;
2643 	uiod.d_mp = mp;
2644 	error = rwnext(wqp, &uiod);
2645 	if (! uiod.d_mp) {
2646 		uioskip(uiop, *iosize);
2647 		return (error);
2648 	}
2649 	ASSERT(mp == uiod.d_mp);
2650 	if (error == EINVAL) {
2651 		/*
2652 		 * The stream plumbing must have changed while
2653 		 * we were away, so just turn off rwnext()s.
2654 		 */
2655 		error = 0;
2656 	} else if (error == EBUSY || error == EWOULDBLOCK) {
2657 		/*
2658 		 * Couldn't enter a perimeter or took a page fault,
2659 		 * so fall-back to putnext().
2660 		 */
2661 		error = 0;
2662 	} else {
2663 		freemsg(mp);
2664 		return (error);
2665 	}
2666 	/* Have to check canput before consuming data from the uio */
2667 	if (pri == 0) {
2668 		if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2669 			freemsg(mp);
2670 			return (EWOULDBLOCK);
2671 		}
2672 	} else {
2673 		if (!bcanputnext(wqp, pri) && !(flags & MSG_IGNFLOW)) {
2674 			freemsg(mp);
2675 			return (EWOULDBLOCK);
2676 		}
2677 	}
2678 	ASSERT(mp == uiod.d_mp);
2679 	/* Copyin data from the uio */
2680 	if ((error = struioget(wqp, mp, &uiod, 0)) != 0) {
2681 		freemsg(mp);
2682 		return (error);
2683 	}
2684 	uioskip(uiop, *iosize);
2685 	if (flags & MSG_IGNFLOW) {
2686 		/*
2687 		 * XXX Hack: Don't get stuck running service procedures.
2688 		 * This is needed for sockfs when sending the unbind message
2689 		 * out of the rput procedure - we don't want a put procedure
2690 		 * to run service procedures.
2691 		 */
2692 		putnext(wqp, mp);
2693 	} else {
2694 		stream_willservice(stp);
2695 		putnext(wqp, mp);
2696 		stream_runservice(stp);
2697 	}
2698 	return (0);
2699 }
2700 
2701 /*
2702  * Write attempts to break the write request into messages conforming
2703  * with the minimum and maximum packet sizes set downstream.
2704  *
2705  * Write will not block if downstream queue is full and
2706  * O_NDELAY is set, otherwise it will block waiting for the queue to get room.
2707  *
2708  * A write of zero bytes gets packaged into a zero length message and sent
2709  * downstream like any other message.
2710  *
2711  * If buffers of the requested sizes are not available, the write will
2712  * sleep until the buffers become available.
2713  *
2714  * Write (if specified) will supply a write offset in a message if it
2715  * makes sense. This can be specified by downstream modules as part of
2716  * a M_SETOPTS message.  Write will not supply the write offset if it
2717  * cannot supply any data in a buffer.  In other words, write will never
2718  * send down an empty packet due to a write offset.
2719  */
2720 /* ARGSUSED2 */
2721 int
2722 strwrite(struct vnode *vp, struct uio *uiop, cred_t *crp)
2723 {
2724 	return (strwrite_common(vp, uiop, crp, 0));
2725 }
2726 
2727 /* ARGSUSED2 */
2728 int
2729 strwrite_common(struct vnode *vp, struct uio *uiop, cred_t *crp, int wflag)
2730 {
2731 	struct stdata *stp;
2732 	struct queue *wqp;
2733 	ssize_t rmin, rmax;
2734 	ssize_t iosize;
2735 	int waitflag;
2736 	int tempmode;
2737 	int error = 0;
2738 	int b_flag;
2739 
2740 	ASSERT(vp->v_stream);
2741 	stp = vp->v_stream;
2742 
2743 	mutex_enter(&stp->sd_lock);
2744 
2745 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
2746 		mutex_exit(&stp->sd_lock);
2747 		return (error);
2748 	}
2749 
2750 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
2751 		error = strwriteable(stp, B_TRUE, B_TRUE);
2752 		if (error != 0) {
2753 			mutex_exit(&stp->sd_lock);
2754 			return (error);
2755 		}
2756 	}
2757 
2758 	mutex_exit(&stp->sd_lock);
2759 
2760 	wqp = stp->sd_wrq;
2761 
2762 	/* get these values from them cached in the stream head */
2763 	rmin = stp->sd_qn_minpsz;
2764 	rmax = stp->sd_qn_maxpsz;
2765 
2766 	/*
2767 	 * Check the min/max packet size constraints.  If min packet size
2768 	 * is non-zero, the write cannot be split into multiple messages
2769 	 * and still guarantee the size constraints.
2770 	 */
2771 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_IN, "strwrite in:q %p", wqp);
2772 
2773 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
2774 	if (rmax == 0) {
2775 		return (0);
2776 	}
2777 	if (rmin > 0) {
2778 		if (uiop->uio_resid < rmin) {
2779 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2780 			    "strwrite out:q %p out %d error %d",
2781 			    wqp, 0, ERANGE);
2782 			return (ERANGE);
2783 		}
2784 		if ((rmax != INFPSZ) && (uiop->uio_resid > rmax)) {
2785 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2786 			    "strwrite out:q %p out %d error %d",
2787 			    wqp, 1, ERANGE);
2788 			return (ERANGE);
2789 		}
2790 	}
2791 
2792 	/*
2793 	 * Do until count satisfied or error.
2794 	 */
2795 	waitflag = WRITEWAIT | wflag;
2796 	if (stp->sd_flag & OLDNDELAY)
2797 		tempmode = uiop->uio_fmode & ~FNDELAY;
2798 	else
2799 		tempmode = uiop->uio_fmode;
2800 
2801 	if (rmax == INFPSZ)
2802 		rmax = uiop->uio_resid;
2803 
2804 	/*
2805 	 * Note that tempmode does not get used in strput/strmakedata
2806 	 * but only in strwaitq. The other routines use uio_fmode
2807 	 * unmodified.
2808 	 */
2809 
2810 	/* LINTED: constant in conditional context */
2811 	while (1) {	/* breaks when uio_resid reaches zero */
2812 		/*
2813 		 * Determine the size of the next message to be
2814 		 * packaged.  May have to break write into several
2815 		 * messages based on max packet size.
2816 		 */
2817 		iosize = MIN(uiop->uio_resid, rmax);
2818 
2819 		/*
2820 		 * Put block downstream when flow control allows it.
2821 		 */
2822 		if ((stp->sd_flag & STRDELIM) && (uiop->uio_resid == iosize))
2823 			b_flag = MSGDELIM;
2824 		else
2825 			b_flag = 0;
2826 
2827 		for (;;) {
2828 			int done = 0;
2829 
2830 			error = strput(stp, NULL, uiop, &iosize, b_flag, 0, 0);
2831 			if (error == 0)
2832 				break;
2833 			if (error != EWOULDBLOCK)
2834 				goto out;
2835 
2836 			mutex_enter(&stp->sd_lock);
2837 			/*
2838 			 * Check for a missed wakeup.
2839 			 * Needed since strput did not hold sd_lock across
2840 			 * the canputnext.
2841 			 */
2842 			if (canputnext(wqp)) {
2843 				/* Try again */
2844 				mutex_exit(&stp->sd_lock);
2845 				continue;
2846 			}
2847 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAIT,
2848 			    "strwrite wait:q %p wait", wqp);
2849 			if ((error = strwaitq(stp, waitflag, (ssize_t)0,
2850 			    tempmode, -1, &done)) != 0 || done) {
2851 				mutex_exit(&stp->sd_lock);
2852 				if ((vp->v_type == VFIFO) &&
2853 				    (uiop->uio_fmode & FNDELAY) &&
2854 				    (error == EAGAIN))
2855 					error = 0;
2856 				goto out;
2857 			}
2858 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAKE,
2859 			    "strwrite wake:q %p awakes", wqp);
2860 			if ((error = i_straccess(stp, JCWRITE)) != 0) {
2861 				mutex_exit(&stp->sd_lock);
2862 				goto out;
2863 			}
2864 			mutex_exit(&stp->sd_lock);
2865 		}
2866 		waitflag |= NOINTR;
2867 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRWRITE_RESID,
2868 		    "strwrite resid:q %p uiop %p", wqp, uiop);
2869 		if (uiop->uio_resid) {
2870 			/* Recheck for errors - needed for sockets */
2871 			if ((stp->sd_wput_opt & SW_RECHECK_ERR) &&
2872 			    (stp->sd_flag & (STWRERR|STRHUP|STPLEX))) {
2873 				mutex_enter(&stp->sd_lock);
2874 				error = strwriteable(stp, B_FALSE, B_TRUE);
2875 				mutex_exit(&stp->sd_lock);
2876 				if (error != 0)
2877 					return (error);
2878 			}
2879 			continue;
2880 		}
2881 		break;
2882 	}
2883 out:
2884 	/*
2885 	 * For historical reasons, applications expect EAGAIN when a data
2886 	 * mblk_t cannot be allocated, so change ENOMEM back to EAGAIN.
2887 	 */
2888 	if (error == ENOMEM)
2889 		error = EAGAIN;
2890 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2891 	    "strwrite out:q %p out %d error %d", wqp, 2, error);
2892 	return (error);
2893 }
2894 
2895 /*
2896  * Stream head write service routine.
2897  * Its job is to wake up any sleeping writers when a queue
2898  * downstream needs data (part of the flow control in putq and getq).
2899  * It also must wake anyone sleeping on a poll().
2900  * For stream head right below mux module, it must also invoke put procedure
2901  * of next downstream module.
2902  */
2903 int
2904 strwsrv(queue_t *q)
2905 {
2906 	struct stdata *stp;
2907 	queue_t *tq;
2908 	qband_t *qbp;
2909 	int i;
2910 	qband_t *myqbp;
2911 	int isevent;
2912 	unsigned char	qbf[NBAND];	/* band flushing backenable flags */
2913 
2914 	TRACE_1(TR_FAC_STREAMS_FR,
2915 	    TR_STRWSRV, "strwsrv:q %p", q);
2916 	stp = (struct stdata *)q->q_ptr;
2917 	ASSERT(qclaimed(q));
2918 	mutex_enter(&stp->sd_lock);
2919 	ASSERT(!(stp->sd_flag & STPLEX));
2920 
2921 	if (stp->sd_flag & WSLEEP) {
2922 		stp->sd_flag &= ~WSLEEP;
2923 		cv_broadcast(&q->q_wait);
2924 	}
2925 	mutex_exit(&stp->sd_lock);
2926 
2927 	/* The other end of a stream pipe went away. */
2928 	if ((tq = q->q_next) == NULL) {
2929 		return (0);
2930 	}
2931 
2932 	/* Find the next module forward that has a service procedure */
2933 	claimstr(q);
2934 	tq = q->q_nfsrv;
2935 	ASSERT(tq != NULL);
2936 
2937 	if ((q->q_flag & QBACK)) {
2938 		if ((tq->q_flag & QFULL)) {
2939 			mutex_enter(QLOCK(tq));
2940 			if (!(tq->q_flag & QFULL)) {
2941 				mutex_exit(QLOCK(tq));
2942 				goto wakeup;
2943 			}
2944 			/*
2945 			 * The queue must have become full again. Set QWANTW
2946 			 * again so strwsrv will be back enabled when
2947 			 * the queue becomes non-full next time.
2948 			 */
2949 			tq->q_flag |= QWANTW;
2950 			mutex_exit(QLOCK(tq));
2951 		} else {
2952 		wakeup:
2953 			pollwakeup(&stp->sd_pollist, POLLWRNORM);
2954 			mutex_enter(&stp->sd_lock);
2955 			if (stp->sd_sigflags & S_WRNORM)
2956 				strsendsig(stp->sd_siglist, S_WRNORM, 0, 0);
2957 			mutex_exit(&stp->sd_lock);
2958 		}
2959 	}
2960 
2961 	isevent = 0;
2962 	i = 1;
2963 	bzero((caddr_t)qbf, NBAND);
2964 	mutex_enter(QLOCK(tq));
2965 	if ((myqbp = q->q_bandp) != NULL)
2966 		for (qbp = tq->q_bandp; qbp && myqbp; qbp = qbp->qb_next) {
2967 			ASSERT(myqbp);
2968 			if ((myqbp->qb_flag & QB_BACK)) {
2969 				if (qbp->qb_flag & QB_FULL) {
2970 					/*
2971 					 * The band must have become full again.
2972 					 * Set QB_WANTW again so strwsrv will
2973 					 * be back enabled when the band becomes
2974 					 * non-full next time.
2975 					 */
2976 					qbp->qb_flag |= QB_WANTW;
2977 				} else {
2978 					isevent = 1;
2979 					qbf[i] = 1;
2980 				}
2981 			}
2982 			myqbp = myqbp->qb_next;
2983 			i++;
2984 		}
2985 	mutex_exit(QLOCK(tq));
2986 
2987 	if (isevent) {
2988 		for (i = tq->q_nband; i; i--) {
2989 			if (qbf[i]) {
2990 				pollwakeup(&stp->sd_pollist, POLLWRBAND);
2991 				mutex_enter(&stp->sd_lock);
2992 				if (stp->sd_sigflags & S_WRBAND)
2993 					strsendsig(stp->sd_siglist, S_WRBAND,
2994 					    (uchar_t)i, 0);
2995 				mutex_exit(&stp->sd_lock);
2996 			}
2997 		}
2998 	}
2999 
3000 	releasestr(q);
3001 	return (0);
3002 }
3003 
3004 /*
3005  * Special case of strcopyin/strcopyout for copying
3006  * struct strioctl that can deal with both data
3007  * models.
3008  */
3009 
3010 #ifdef	_LP64
3011 
3012 static int
3013 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3014 {
3015 	struct	strioctl32 strioc32;
3016 	struct	strioctl *striocp;
3017 
3018 	if (copyflag & U_TO_K) {
3019 		ASSERT((copyflag & K_TO_K) == 0);
3020 
3021 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
3022 			if (copyin(from, &strioc32, sizeof (strioc32)))
3023 				return (EFAULT);
3024 
3025 			striocp = (struct strioctl *)to;
3026 			striocp->ic_cmd	= strioc32.ic_cmd;
3027 			striocp->ic_timout = strioc32.ic_timout;
3028 			striocp->ic_len	= strioc32.ic_len;
3029 			striocp->ic_dp	= (char *)(uintptr_t)strioc32.ic_dp;
3030 
3031 		} else { /* NATIVE data model */
3032 			if (copyin(from, to, sizeof (struct strioctl))) {
3033 				return (EFAULT);
3034 			} else {
3035 				return (0);
3036 			}
3037 		}
3038 	} else {
3039 		ASSERT(copyflag & K_TO_K);
3040 		bcopy(from, to, sizeof (struct strioctl));
3041 	}
3042 	return (0);
3043 }
3044 
3045 static int
3046 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3047 {
3048 	struct	strioctl32 strioc32;
3049 	struct	strioctl *striocp;
3050 
3051 	if (copyflag & U_TO_K) {
3052 		ASSERT((copyflag & K_TO_K) == 0);
3053 
3054 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
3055 			striocp = (struct strioctl *)from;
3056 			strioc32.ic_cmd	= striocp->ic_cmd;
3057 			strioc32.ic_timout = striocp->ic_timout;
3058 			strioc32.ic_len	= striocp->ic_len;
3059 			strioc32.ic_dp	= (caddr32_t)(uintptr_t)striocp->ic_dp;
3060 			ASSERT((char *)(uintptr_t)strioc32.ic_dp ==
3061 			    striocp->ic_dp);
3062 
3063 			if (copyout(&strioc32, to, sizeof (strioc32)))
3064 				return (EFAULT);
3065 
3066 		} else { /* NATIVE data model */
3067 			if (copyout(from, to, sizeof (struct strioctl))) {
3068 				return (EFAULT);
3069 			} else {
3070 				return (0);
3071 			}
3072 		}
3073 	} else {
3074 		ASSERT(copyflag & K_TO_K);
3075 		bcopy(from, to, sizeof (struct strioctl));
3076 	}
3077 	return (0);
3078 }
3079 
3080 #else	/* ! _LP64 */
3081 
3082 /* ARGSUSED2 */
3083 static int
3084 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3085 {
3086 	return (strcopyin(from, to, sizeof (struct strioctl), copyflag));
3087 }
3088 
3089 /* ARGSUSED2 */
3090 static int
3091 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3092 {
3093 	return (strcopyout(from, to, sizeof (struct strioctl), copyflag));
3094 }
3095 
3096 #endif	/* _LP64 */
3097 
3098 /*
3099  * Determine type of job control semantics expected by user.  The
3100  * possibilities are:
3101  *	JCREAD	- Behaves like read() on fd; send SIGTTIN
3102  *	JCWRITE	- Behaves like write() on fd; send SIGTTOU if TOSTOP set
3103  *	JCSETP	- Sets a value in the stream; send SIGTTOU, ignore TOSTOP
3104  *	JCGETP	- Gets a value in the stream; no signals.
3105  * See straccess in strsubr.c for usage of these values.
3106  *
3107  * This routine also returns -1 for I_STR as a special case; the
3108  * caller must call again with the real ioctl number for
3109  * classification.
3110  */
3111 static int
3112 job_control_type(int cmd)
3113 {
3114 	switch (cmd) {
3115 	case I_STR:
3116 		return (-1);
3117 
3118 	case I_RECVFD:
3119 	case I_E_RECVFD:
3120 		return (JCREAD);
3121 
3122 	case I_FDINSERT:
3123 	case I_SENDFD:
3124 		return (JCWRITE);
3125 
3126 	case TCSETA:
3127 	case TCSETAW:
3128 	case TCSETAF:
3129 	case TCSBRK:
3130 	case TCXONC:
3131 	case TCFLSH:
3132 	case TCDSET:	/* Obsolete */
3133 	case TIOCSWINSZ:
3134 	case TCSETS:
3135 	case TCSETSW:
3136 	case TCSETSF:
3137 	case TIOCSETD:
3138 	case TIOCHPCL:
3139 	case TIOCSETP:
3140 	case TIOCSETN:
3141 	case TIOCEXCL:
3142 	case TIOCNXCL:
3143 	case TIOCFLUSH:
3144 	case TIOCSETC:
3145 	case TIOCLBIS:
3146 	case TIOCLBIC:
3147 	case TIOCLSET:
3148 	case TIOCSBRK:
3149 	case TIOCCBRK:
3150 	case TIOCSDTR:
3151 	case TIOCCDTR:
3152 	case TIOCSLTC:
3153 	case TIOCSTOP:
3154 	case TIOCSTART:
3155 	case TIOCSTI:
3156 	case TIOCSPGRP:
3157 	case TIOCMSET:
3158 	case TIOCMBIS:
3159 	case TIOCMBIC:
3160 	case TIOCREMOTE:
3161 	case TIOCSIGNAL:
3162 	case LDSETT:
3163 	case LDSMAP:	/* Obsolete */
3164 	case DIOCSETP:
3165 	case I_FLUSH:
3166 	case I_SRDOPT:
3167 	case I_SETSIG:
3168 	case I_SWROPT:
3169 	case I_FLUSHBAND:
3170 	case I_SETCLTIME:
3171 	case I_SERROPT:
3172 	case I_ESETSIG:
3173 	case FIONBIO:
3174 	case FIOASYNC:
3175 	case FIOSETOWN:
3176 	case JBOOT:	/* Obsolete */
3177 	case JTERM:	/* Obsolete */
3178 	case JTIMOM:	/* Obsolete */
3179 	case JZOMBOOT:	/* Obsolete */
3180 	case JAGENT:	/* Obsolete */
3181 	case JTRUN:	/* Obsolete */
3182 	case JXTPROTO:	/* Obsolete */
3183 		return (JCSETP);
3184 	}
3185 
3186 	return (JCGETP);
3187 }
3188 
3189 /*
3190  * ioctl for streams
3191  */
3192 int
3193 strioctl(struct vnode *vp, int cmd, intptr_t arg, int flag, int copyflag,
3194     cred_t *crp, int *rvalp)
3195 {
3196 	struct stdata *stp;
3197 	struct strcmd *scp;
3198 	struct strioctl strioc;
3199 	struct uio uio;
3200 	struct iovec iov;
3201 	int access;
3202 	mblk_t *mp;
3203 	int error = 0;
3204 	int done = 0;
3205 	ssize_t	rmin, rmax;
3206 	queue_t *wrq;
3207 	queue_t *rdq;
3208 	boolean_t kioctl = B_FALSE;
3209 	uint32_t auditing = AU_AUDITING();
3210 
3211 	if (flag & FKIOCTL) {
3212 		copyflag = K_TO_K;
3213 		kioctl = B_TRUE;
3214 	}
3215 	ASSERT(vp->v_stream);
3216 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
3217 	stp = vp->v_stream;
3218 
3219 	TRACE_3(TR_FAC_STREAMS_FR, TR_IOCTL_ENTER,
3220 	    "strioctl:stp %p cmd %X arg %lX", stp, cmd, arg);
3221 
3222 	/*
3223 	 * If the copy is kernel to kernel, make sure that the FNATIVE
3224 	 * flag is set.  After this it would be a serious error to have
3225 	 * no model flag.
3226 	 */
3227 	if (copyflag == K_TO_K)
3228 		flag = (flag & ~FMODELS) | FNATIVE;
3229 
3230 	ASSERT((flag & FMODELS) != 0);
3231 
3232 	wrq = stp->sd_wrq;
3233 	rdq = _RD(wrq);
3234 
3235 	access = job_control_type(cmd);
3236 
3237 	/* We should never see these here, should be handled by iwscn */
3238 	if (cmd == SRIOCSREDIR || cmd == SRIOCISREDIR)
3239 		return (EINVAL);
3240 
3241 	mutex_enter(&stp->sd_lock);
3242 	if ((access != -1) && ((error = i_straccess(stp, access)) != 0)) {
3243 		mutex_exit(&stp->sd_lock);
3244 		return (error);
3245 	}
3246 	mutex_exit(&stp->sd_lock);
3247 
3248 	/*
3249 	 * Check for sgttyb-related ioctls first, and complain as
3250 	 * necessary.
3251 	 */
3252 	switch (cmd) {
3253 	case TIOCGETP:
3254 	case TIOCSETP:
3255 	case TIOCSETN:
3256 		if (sgttyb_handling >= 2 && !sgttyb_complaint) {
3257 			sgttyb_complaint = B_TRUE;
3258 			cmn_err(CE_NOTE,
3259 			    "application used obsolete TIOC[GS]ET");
3260 		}
3261 		if (sgttyb_handling >= 3) {
3262 			tsignal(curthread, SIGSYS);
3263 			return (EIO);
3264 		}
3265 		break;
3266 	}
3267 
3268 	mutex_enter(&stp->sd_lock);
3269 
3270 	switch (cmd) {
3271 	case I_RECVFD:
3272 	case I_E_RECVFD:
3273 	case I_PEEK:
3274 	case I_NREAD:
3275 	case FIONREAD:
3276 	case FIORDCHK:
3277 	case I_ATMARK:
3278 	case FIONBIO:
3279 	case FIOASYNC:
3280 		if (stp->sd_flag & (STRDERR|STPLEX)) {
3281 			error = strgeterr(stp, STRDERR|STPLEX, 0);
3282 			if (error != 0) {
3283 				mutex_exit(&stp->sd_lock);
3284 				return (error);
3285 			}
3286 		}
3287 		break;
3288 
3289 	default:
3290 		if (stp->sd_flag & (STRDERR|STWRERR|STPLEX)) {
3291 			error = strgeterr(stp, STRDERR|STWRERR|STPLEX, 0);
3292 			if (error != 0) {
3293 				mutex_exit(&stp->sd_lock);
3294 				return (error);
3295 			}
3296 		}
3297 	}
3298 
3299 	mutex_exit(&stp->sd_lock);
3300 
3301 	switch (cmd) {
3302 	default:
3303 		/*
3304 		 * The stream head has hardcoded knowledge of a
3305 		 * miscellaneous collection of terminal-, keyboard- and
3306 		 * mouse-related ioctls, enumerated below.  This hardcoded
3307 		 * knowledge allows the stream head to automatically
3308 		 * convert transparent ioctl requests made by userland
3309 		 * programs into I_STR ioctls which many old STREAMS
3310 		 * modules and drivers require.
3311 		 *
3312 		 * No new ioctls should ever be added to this list.
3313 		 * Instead, the STREAMS module or driver should be written
3314 		 * to either handle transparent ioctls or require any
3315 		 * userland programs to use I_STR ioctls (by returning
3316 		 * EINVAL to any transparent ioctl requests).
3317 		 *
3318 		 * More importantly, removing ioctls from this list should
3319 		 * be done with the utmost care, since our STREAMS modules
3320 		 * and drivers *count* on the stream head performing this
3321 		 * conversion, and thus may panic while processing
3322 		 * transparent ioctl request for one of these ioctls (keep
3323 		 * in mind that third party modules and drivers may have
3324 		 * similar problems).
3325 		 */
3326 		if (((cmd & IOCTYPE) == LDIOC) ||
3327 		    ((cmd & IOCTYPE) == tIOC) ||
3328 		    ((cmd & IOCTYPE) == TIOC) ||
3329 		    ((cmd & IOCTYPE) == KIOC) ||
3330 		    ((cmd & IOCTYPE) == MSIOC) ||
3331 		    ((cmd & IOCTYPE) == VUIOC)) {
3332 			/*
3333 			 * The ioctl is a tty ioctl - set up strioc buffer
3334 			 * and call strdoioctl() to do the work.
3335 			 */
3336 			if (stp->sd_flag & STRHUP)
3337 				return (ENXIO);
3338 			strioc.ic_cmd = cmd;
3339 			strioc.ic_timout = INFTIM;
3340 
3341 			switch (cmd) {
3342 
3343 			case TCXONC:
3344 			case TCSBRK:
3345 			case TCFLSH:
3346 			case TCDSET:
3347 				{
3348 				int native_arg = (int)arg;
3349 				strioc.ic_len = sizeof (int);
3350 				strioc.ic_dp = (char *)&native_arg;
3351 				return (strdoioctl(stp, &strioc, flag,
3352 				    K_TO_K, crp, rvalp));
3353 				}
3354 
3355 			case TCSETA:
3356 			case TCSETAW:
3357 			case TCSETAF:
3358 				strioc.ic_len = sizeof (struct termio);
3359 				strioc.ic_dp = (char *)arg;
3360 				return (strdoioctl(stp, &strioc, flag,
3361 				    copyflag, crp, rvalp));
3362 
3363 			case TCSETS:
3364 			case TCSETSW:
3365 			case TCSETSF:
3366 				strioc.ic_len = sizeof (struct termios);
3367 				strioc.ic_dp = (char *)arg;
3368 				return (strdoioctl(stp, &strioc, flag,
3369 				    copyflag, crp, rvalp));
3370 
3371 			case LDSETT:
3372 				strioc.ic_len = sizeof (struct termcb);
3373 				strioc.ic_dp = (char *)arg;
3374 				return (strdoioctl(stp, &strioc, flag,
3375 				    copyflag, crp, rvalp));
3376 
3377 			case TIOCSETP:
3378 				strioc.ic_len = sizeof (struct sgttyb);
3379 				strioc.ic_dp = (char *)arg;
3380 				return (strdoioctl(stp, &strioc, flag,
3381 				    copyflag, crp, rvalp));
3382 
3383 			case TIOCSTI:
3384 				if ((flag & FREAD) == 0 &&
3385 				    secpolicy_sti(crp) != 0) {
3386 					return (EPERM);
3387 				}
3388 				mutex_enter(&stp->sd_lock);
3389 				mutex_enter(&curproc->p_splock);
3390 				if (stp->sd_sidp != curproc->p_sessp->s_sidp &&
3391 				    secpolicy_sti(crp) != 0) {
3392 					mutex_exit(&curproc->p_splock);
3393 					mutex_exit(&stp->sd_lock);
3394 					return (EACCES);
3395 				}
3396 				mutex_exit(&curproc->p_splock);
3397 				mutex_exit(&stp->sd_lock);
3398 
3399 				strioc.ic_len = sizeof (char);
3400 				strioc.ic_dp = (char *)arg;
3401 				return (strdoioctl(stp, &strioc, flag,
3402 				    copyflag, crp, rvalp));
3403 
3404 			case TIOCSWINSZ:
3405 				strioc.ic_len = sizeof (struct winsize);
3406 				strioc.ic_dp = (char *)arg;
3407 				return (strdoioctl(stp, &strioc, flag,
3408 				    copyflag, crp, rvalp));
3409 
3410 			case TIOCSSIZE:
3411 				strioc.ic_len = sizeof (struct ttysize);
3412 				strioc.ic_dp = (char *)arg;
3413 				return (strdoioctl(stp, &strioc, flag,
3414 				    copyflag, crp, rvalp));
3415 
3416 			case TIOCSSOFTCAR:
3417 			case KIOCTRANS:
3418 			case KIOCTRANSABLE:
3419 			case KIOCCMD:
3420 			case KIOCSDIRECT:
3421 			case KIOCSCOMPAT:
3422 			case KIOCSKABORTEN:
3423 			case KIOCSRPTCOUNT:
3424 			case KIOCSRPTDELAY:
3425 			case KIOCSRPTRATE:
3426 			case VUIDSFORMAT:
3427 			case TIOCSPPS:
3428 				strioc.ic_len = sizeof (int);
3429 				strioc.ic_dp = (char *)arg;
3430 				return (strdoioctl(stp, &strioc, flag,
3431 				    copyflag, crp, rvalp));
3432 
3433 			case KIOCSETKEY:
3434 			case KIOCGETKEY:
3435 				strioc.ic_len = sizeof (struct kiockey);
3436 				strioc.ic_dp = (char *)arg;
3437 				return (strdoioctl(stp, &strioc, flag,
3438 				    copyflag, crp, rvalp));
3439 
3440 			case KIOCSKEY:
3441 			case KIOCGKEY:
3442 				strioc.ic_len = sizeof (struct kiockeymap);
3443 				strioc.ic_dp = (char *)arg;
3444 				return (strdoioctl(stp, &strioc, flag,
3445 				    copyflag, crp, rvalp));
3446 
3447 			case KIOCSLED:
3448 				/* arg is a pointer to char */
3449 				strioc.ic_len = sizeof (char);
3450 				strioc.ic_dp = (char *)arg;
3451 				return (strdoioctl(stp, &strioc, flag,
3452 				    copyflag, crp, rvalp));
3453 
3454 			case MSIOSETPARMS:
3455 				strioc.ic_len = sizeof (Ms_parms);
3456 				strioc.ic_dp = (char *)arg;
3457 				return (strdoioctl(stp, &strioc, flag,
3458 				    copyflag, crp, rvalp));
3459 
3460 			case VUIDSADDR:
3461 			case VUIDGADDR:
3462 				strioc.ic_len = sizeof (struct vuid_addr_probe);
3463 				strioc.ic_dp = (char *)arg;
3464 				return (strdoioctl(stp, &strioc, flag,
3465 				    copyflag, crp, rvalp));
3466 
3467 			/*
3468 			 * These M_IOCTL's don't require any data to be sent
3469 			 * downstream, and the driver will allocate and link
3470 			 * on its own mblk_t upon M_IOCACK -- thus we set
3471 			 * ic_len to zero and set ic_dp to arg so we know
3472 			 * where to copyout to later.
3473 			 */
3474 			case TIOCGSOFTCAR:
3475 			case TIOCGWINSZ:
3476 			case TIOCGSIZE:
3477 			case KIOCGTRANS:
3478 			case KIOCGTRANSABLE:
3479 			case KIOCTYPE:
3480 			case KIOCGDIRECT:
3481 			case KIOCGCOMPAT:
3482 			case KIOCLAYOUT:
3483 			case KIOCGLED:
3484 			case MSIOGETPARMS:
3485 			case MSIOBUTTONS:
3486 			case VUIDGFORMAT:
3487 			case TIOCGPPS:
3488 			case TIOCGPPSEV:
3489 			case TCGETA:
3490 			case TCGETS:
3491 			case LDGETT:
3492 			case TIOCGETP:
3493 			case KIOCGRPTCOUNT:
3494 			case KIOCGRPTDELAY:
3495 			case KIOCGRPTRATE:
3496 				strioc.ic_len = 0;
3497 				strioc.ic_dp = (char *)arg;
3498 				return (strdoioctl(stp, &strioc, flag,
3499 				    copyflag, crp, rvalp));
3500 			}
3501 		}
3502 
3503 		/*
3504 		 * Unknown cmd - send it down as a transparent ioctl.
3505 		 */
3506 		strioc.ic_cmd = cmd;
3507 		strioc.ic_timout = INFTIM;
3508 		strioc.ic_len = TRANSPARENT;
3509 		strioc.ic_dp = (char *)&arg;
3510 
3511 		return (strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp));
3512 
3513 	case I_STR:
3514 		/*
3515 		 * Stream ioctl.  Read in an strioctl buffer from the user
3516 		 * along with any data specified and send it downstream.
3517 		 * Strdoioctl will wait allow only one ioctl message at
3518 		 * a time, and waits for the acknowledgement.
3519 		 */
3520 
3521 		if (stp->sd_flag & STRHUP)
3522 			return (ENXIO);
3523 
3524 		error = strcopyin_strioctl((void *)arg, &strioc, flag,
3525 		    copyflag);
3526 		if (error != 0)
3527 			return (error);
3528 
3529 		if ((strioc.ic_len < 0) || (strioc.ic_timout < -1))
3530 			return (EINVAL);
3531 
3532 		access = job_control_type(strioc.ic_cmd);
3533 		mutex_enter(&stp->sd_lock);
3534 		if ((access != -1) &&
3535 		    ((error = i_straccess(stp, access)) != 0)) {
3536 			mutex_exit(&stp->sd_lock);
3537 			return (error);
3538 		}
3539 		mutex_exit(&stp->sd_lock);
3540 
3541 		/*
3542 		 * The I_STR facility provides a trap door for malicious
3543 		 * code to send down bogus streamio(7I) ioctl commands to
3544 		 * unsuspecting STREAMS modules and drivers which expect to
3545 		 * only get these messages from the stream head.
3546 		 * Explicitly prohibit any streamio ioctls which can be
3547 		 * passed downstream by the stream head.  Note that we do
3548 		 * not block all streamio ioctls because the ioctl
3549 		 * numberspace is not well managed and thus it's possible
3550 		 * that a module or driver's ioctl numbers may accidentally
3551 		 * collide with them.
3552 		 */
3553 		switch (strioc.ic_cmd) {
3554 		case I_LINK:
3555 		case I_PLINK:
3556 		case I_UNLINK:
3557 		case I_PUNLINK:
3558 		case _I_GETPEERCRED:
3559 		case _I_PLINK_LH:
3560 			return (EINVAL);
3561 		}
3562 
3563 		error = strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp);
3564 		if (error == 0) {
3565 			error = strcopyout_strioctl(&strioc, (void *)arg,
3566 			    flag, copyflag);
3567 		}
3568 		return (error);
3569 
3570 	case _I_CMD:
3571 		/*
3572 		 * Like I_STR, but without using M_IOC* messages and without
3573 		 * copyins/copyouts beyond the passed-in argument.
3574 		 */
3575 		if (stp->sd_flag & STRHUP)
3576 			return (ENXIO);
3577 
3578 		if (copyflag == U_TO_K) {
3579 			if ((scp = kmem_alloc(sizeof (strcmd_t),
3580 			    KM_NOSLEEP)) == NULL) {
3581 				return (ENOMEM);
3582 			}
3583 
3584 			if (copyin((void *)arg, scp, sizeof (strcmd_t))) {
3585 				kmem_free(scp, sizeof (strcmd_t));
3586 				return (EFAULT);
3587 			}
3588 		} else {
3589 			scp = (strcmd_t *)arg;
3590 		}
3591 
3592 		access = job_control_type(scp->sc_cmd);
3593 		mutex_enter(&stp->sd_lock);
3594 		if (access != -1 && (error = i_straccess(stp, access)) != 0) {
3595 			mutex_exit(&stp->sd_lock);
3596 			if (copyflag == U_TO_K)
3597 				kmem_free(scp, sizeof (strcmd_t));
3598 			return (error);
3599 		}
3600 		mutex_exit(&stp->sd_lock);
3601 
3602 		*rvalp = 0;
3603 		if ((error = strdocmd(stp, scp, crp)) == 0) {
3604 			if (copyflag == U_TO_K &&
3605 			    copyout(scp, (void *)arg, sizeof (strcmd_t))) {
3606 				error = EFAULT;
3607 			}
3608 		}
3609 		if (copyflag == U_TO_K)
3610 			kmem_free(scp, sizeof (strcmd_t));
3611 		return (error);
3612 
3613 	case I_NREAD:
3614 		/*
3615 		 * Return number of bytes of data in first message
3616 		 * in queue in "arg" and return the number of messages
3617 		 * in queue in return value.
3618 		 */
3619 	{
3620 		size_t	size;
3621 		int	retval;
3622 		int	count = 0;
3623 
3624 		mutex_enter(QLOCK(rdq));
3625 
3626 		size = msgdsize(rdq->q_first);
3627 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3628 			count++;
3629 
3630 		mutex_exit(QLOCK(rdq));
3631 		if (stp->sd_struiordq) {
3632 			infod_t infod;
3633 
3634 			infod.d_cmd = INFOD_COUNT;
3635 			infod.d_count = 0;
3636 			if (count == 0) {
3637 				infod.d_cmd |= INFOD_FIRSTBYTES;
3638 				infod.d_bytes = 0;
3639 			}
3640 			infod.d_res = 0;
3641 			(void) infonext(rdq, &infod);
3642 			count += infod.d_count;
3643 			if (infod.d_res & INFOD_FIRSTBYTES)
3644 				size = infod.d_bytes;
3645 		}
3646 
3647 		/*
3648 		 * Drop down from size_t to the "int" required by the
3649 		 * interface.  Cap at INT_MAX.
3650 		 */
3651 		retval = MIN(size, INT_MAX);
3652 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3653 		    copyflag);
3654 		if (!error)
3655 			*rvalp = count;
3656 		return (error);
3657 	}
3658 
3659 	case FIONREAD:
3660 		/*
3661 		 * Return number of bytes of data in all data messages
3662 		 * in queue in "arg".
3663 		 */
3664 	{
3665 		size_t	size = 0;
3666 		int	retval;
3667 
3668 		mutex_enter(QLOCK(rdq));
3669 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3670 			size += msgdsize(mp);
3671 		mutex_exit(QLOCK(rdq));
3672 
3673 		if (stp->sd_struiordq) {
3674 			infod_t infod;
3675 
3676 			infod.d_cmd = INFOD_BYTES;
3677 			infod.d_res = 0;
3678 			infod.d_bytes = 0;
3679 			(void) infonext(rdq, &infod);
3680 			size += infod.d_bytes;
3681 		}
3682 
3683 		/*
3684 		 * Drop down from size_t to the "int" required by the
3685 		 * interface.  Cap at INT_MAX.
3686 		 */
3687 		retval = MIN(size, INT_MAX);
3688 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3689 		    copyflag);
3690 
3691 		*rvalp = 0;
3692 		return (error);
3693 	}
3694 	case FIORDCHK:
3695 		/*
3696 		 * FIORDCHK does not use arg value (like FIONREAD),
3697 		 * instead a count is returned. I_NREAD value may
3698 		 * not be accurate but safe. The real thing to do is
3699 		 * to add the msgdsizes of all data  messages until
3700 		 * a non-data message.
3701 		 */
3702 	{
3703 		size_t size = 0;
3704 
3705 		mutex_enter(QLOCK(rdq));
3706 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3707 			size += msgdsize(mp);
3708 		mutex_exit(QLOCK(rdq));
3709 
3710 		if (stp->sd_struiordq) {
3711 			infod_t infod;
3712 
3713 			infod.d_cmd = INFOD_BYTES;
3714 			infod.d_res = 0;
3715 			infod.d_bytes = 0;
3716 			(void) infonext(rdq, &infod);
3717 			size += infod.d_bytes;
3718 		}
3719 
3720 		/*
3721 		 * Since ioctl returns an int, and memory sizes under
3722 		 * LP64 may not fit, we return INT_MAX if the count was
3723 		 * actually greater.
3724 		 */
3725 		*rvalp = MIN(size, INT_MAX);
3726 		return (0);
3727 	}
3728 
3729 	case I_FIND:
3730 		/*
3731 		 * Get module name.
3732 		 */
3733 	{
3734 		char mname[FMNAMESZ + 1];
3735 		queue_t *q;
3736 
3737 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3738 		    mname, FMNAMESZ + 1, NULL);
3739 		if (error)
3740 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3741 
3742 		/*
3743 		 * Return EINVAL if we're handed a bogus module name.
3744 		 */
3745 		if (fmodsw_find(mname, FMODSW_LOAD) == NULL) {
3746 			TRACE_0(TR_FAC_STREAMS_FR,
3747 			    TR_I_CANT_FIND, "couldn't I_FIND");
3748 			return (EINVAL);
3749 		}
3750 
3751 		*rvalp = 0;
3752 
3753 		/* Look downstream to see if module is there. */
3754 		claimstr(stp->sd_wrq);
3755 		for (q = stp->sd_wrq->q_next; q; q = q->q_next) {
3756 			if (q->q_flag & QREADR) {
3757 				q = NULL;
3758 				break;
3759 			}
3760 			if (strcmp(mname, Q2NAME(q)) == 0)
3761 				break;
3762 		}
3763 		releasestr(stp->sd_wrq);
3764 
3765 		*rvalp = (q ? 1 : 0);
3766 		return (error);
3767 	}
3768 
3769 	case I_PUSH:
3770 	case __I_PUSH_NOCTTY:
3771 		/*
3772 		 * Push a module.
3773 		 * For the case __I_PUSH_NOCTTY push a module but
3774 		 * do not allocate controlling tty. See bugid 4025044
3775 		 */
3776 
3777 	{
3778 		char mname[FMNAMESZ + 1];
3779 		fmodsw_impl_t *fp;
3780 		dev_t dummydev;
3781 
3782 		if (stp->sd_flag & STRHUP)
3783 			return (ENXIO);
3784 
3785 		/*
3786 		 * Get module name and look up in fmodsw.
3787 		 */
3788 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3789 		    mname, FMNAMESZ + 1, NULL);
3790 		if (error)
3791 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3792 
3793 		if ((fp = fmodsw_find(mname, FMODSW_HOLD | FMODSW_LOAD)) ==
3794 		    NULL)
3795 			return (EINVAL);
3796 
3797 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_PUSH,
3798 		    "I_PUSH:fp %p stp %p", fp, stp);
3799 
3800 		/*
3801 		 * If the module is flagged as single-instance, then check
3802 		 * to see if the module is already pushed. If it is, return
3803 		 * as if the push was successful.
3804 		 */
3805 		if (fp->f_qflag & _QSINGLE_INSTANCE) {
3806 			queue_t *q;
3807 
3808 			claimstr(stp->sd_wrq);
3809 			for (q = stp->sd_wrq->q_next; q; q = q->q_next) {
3810 				if (q->q_flag & QREADR) {
3811 					q = NULL;
3812 					break;
3813 				}
3814 				if (strcmp(mname, Q2NAME(q)) == 0)
3815 					break;
3816 			}
3817 			releasestr(stp->sd_wrq);
3818 			if (q != NULL) {
3819 				fmodsw_rele(fp);
3820 				return (0);
3821 			}
3822 		}
3823 
3824 		if (error = strstartplumb(stp, flag, cmd)) {
3825 			fmodsw_rele(fp);
3826 			return (error);
3827 		}
3828 
3829 		/*
3830 		 * See if any more modules can be pushed on this stream.
3831 		 * Note that this check must be done after strstartplumb()
3832 		 * since otherwise multiple threads issuing I_PUSHes on
3833 		 * the same stream will be able to exceed nstrpush.
3834 		 */
3835 		mutex_enter(&stp->sd_lock);
3836 		if (stp->sd_pushcnt >= nstrpush) {
3837 			fmodsw_rele(fp);
3838 			strendplumb(stp);
3839 			mutex_exit(&stp->sd_lock);
3840 			return (EINVAL);
3841 		}
3842 		mutex_exit(&stp->sd_lock);
3843 
3844 		/*
3845 		 * Push new module and call its open routine
3846 		 * via qattach().  Modules don't change device
3847 		 * numbers, so just ignore dummydev here.
3848 		 */
3849 		dummydev = vp->v_rdev;
3850 		if ((error = qattach(rdq, &dummydev, 0, crp, fp,
3851 		    B_FALSE)) == 0) {
3852 			if (vp->v_type == VCHR && /* sorry, no pipes allowed */
3853 			    (cmd == I_PUSH) && (stp->sd_flag & STRISTTY)) {
3854 				/*
3855 				 * try to allocate it as a controlling terminal
3856 				 */
3857 				(void) strctty(stp);
3858 			}
3859 		}
3860 
3861 		mutex_enter(&stp->sd_lock);
3862 
3863 		/*
3864 		 * As a performance concern we are caching the values of
3865 		 * q_minpsz and q_maxpsz of the module below the stream
3866 		 * head in the stream head.
3867 		 */
3868 		mutex_enter(QLOCK(stp->sd_wrq->q_next));
3869 		rmin = stp->sd_wrq->q_next->q_minpsz;
3870 		rmax = stp->sd_wrq->q_next->q_maxpsz;
3871 		mutex_exit(QLOCK(stp->sd_wrq->q_next));
3872 
3873 		/* Do this processing here as a performance concern */
3874 		if (strmsgsz != 0) {
3875 			if (rmax == INFPSZ)
3876 				rmax = strmsgsz;
3877 			else  {
3878 				if (vp->v_type == VFIFO)
3879 					rmax = MIN(PIPE_BUF, rmax);
3880 				else	rmax = MIN(strmsgsz, rmax);
3881 			}
3882 		}
3883 
3884 		mutex_enter(QLOCK(wrq));
3885 		stp->sd_qn_minpsz = rmin;
3886 		stp->sd_qn_maxpsz = rmax;
3887 		mutex_exit(QLOCK(wrq));
3888 
3889 		strendplumb(stp);
3890 		mutex_exit(&stp->sd_lock);
3891 		return (error);
3892 	}
3893 
3894 	case I_POP:
3895 	{
3896 		queue_t	*q;
3897 
3898 		if (stp->sd_flag & STRHUP)
3899 			return (ENXIO);
3900 		if (!wrq->q_next)	/* for broken pipes */
3901 			return (EINVAL);
3902 
3903 		if (error = strstartplumb(stp, flag, cmd))
3904 			return (error);
3905 
3906 		/*
3907 		 * If there is an anchor on this stream and popping
3908 		 * the current module would attempt to pop through the
3909 		 * anchor, then disallow the pop unless we have sufficient
3910 		 * privileges; take the cheapest (non-locking) check
3911 		 * first.
3912 		 */
3913 		if (secpolicy_ip_config(crp, B_TRUE) != 0 ||
3914 		    (stp->sd_anchorzone != crgetzoneid(crp))) {
3915 			mutex_enter(&stp->sd_lock);
3916 			/*
3917 			 * Anchors only apply if there's at least one
3918 			 * module on the stream (sd_pushcnt > 0).
3919 			 */
3920 			if (stp->sd_pushcnt > 0 &&
3921 			    stp->sd_pushcnt == stp->sd_anchor &&
3922 			    stp->sd_vnode->v_type != VFIFO) {
3923 				strendplumb(stp);
3924 				mutex_exit(&stp->sd_lock);
3925 				if (stp->sd_anchorzone != crgetzoneid(crp))
3926 					return (EINVAL);
3927 				/* Audit and report error */
3928 				return (secpolicy_ip_config(crp, B_FALSE));
3929 			}
3930 			mutex_exit(&stp->sd_lock);
3931 		}
3932 
3933 		q = wrq->q_next;
3934 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_POP,
3935 		    "I_POP:%p from %p", q, stp);
3936 		if (q->q_next == NULL || (q->q_flag & (QREADR|QISDRV))) {
3937 			error = EINVAL;
3938 		} else {
3939 			qdetach(_RD(q), 1, flag, crp, B_FALSE);
3940 			error = 0;
3941 		}
3942 		mutex_enter(&stp->sd_lock);
3943 
3944 		/*
3945 		 * As a performance concern we are caching the values of
3946 		 * q_minpsz and q_maxpsz of the module below the stream
3947 		 * head in the stream head.
3948 		 */
3949 		mutex_enter(QLOCK(wrq->q_next));
3950 		rmin = wrq->q_next->q_minpsz;
3951 		rmax = wrq->q_next->q_maxpsz;
3952 		mutex_exit(QLOCK(wrq->q_next));
3953 
3954 		/* Do this processing here as a performance concern */
3955 		if (strmsgsz != 0) {
3956 			if (rmax == INFPSZ)
3957 				rmax = strmsgsz;
3958 			else  {
3959 				if (vp->v_type == VFIFO)
3960 					rmax = MIN(PIPE_BUF, rmax);
3961 				else	rmax = MIN(strmsgsz, rmax);
3962 			}
3963 		}
3964 
3965 		mutex_enter(QLOCK(wrq));
3966 		stp->sd_qn_minpsz = rmin;
3967 		stp->sd_qn_maxpsz = rmax;
3968 		mutex_exit(QLOCK(wrq));
3969 
3970 		/* If we popped through the anchor, then reset the anchor. */
3971 		if (stp->sd_pushcnt < stp->sd_anchor) {
3972 			stp->sd_anchor = 0;
3973 			stp->sd_anchorzone = 0;
3974 		}
3975 		strendplumb(stp);
3976 		mutex_exit(&stp->sd_lock);
3977 		return (error);
3978 	}
3979 
3980 	case _I_MUXID2FD:
3981 	{
3982 		/*
3983 		 * Create a fd for a I_PLINK'ed lower stream with a given
3984 		 * muxid.  With the fd, application can send down ioctls,
3985 		 * like I_LIST, to the previously I_PLINK'ed stream.  Note
3986 		 * that after getting the fd, the application has to do an
3987 		 * I_PUNLINK on the muxid before it can do any operation
3988 		 * on the lower stream.  This is required by spec1170.
3989 		 *
3990 		 * The fd used to do this ioctl should point to the same
3991 		 * controlling device used to do the I_PLINK.  If it uses
3992 		 * a different stream or an invalid muxid, I_MUXID2FD will
3993 		 * fail.  The error code is set to EINVAL.
3994 		 *
3995 		 * The intended use of this interface is the following.
3996 		 * An application I_PLINK'ed a stream and exits.  The fd
3997 		 * to the lower stream is gone.  Another application
3998 		 * wants to get a fd to the lower stream, it uses I_MUXID2FD.
3999 		 */
4000 		int muxid = (int)arg;
4001 		int fd;
4002 		linkinfo_t *linkp;
4003 		struct file *fp;
4004 		netstack_t *ns;
4005 		str_stack_t *ss;
4006 
4007 		/*
4008 		 * Do not allow the wildcard muxid.  This ioctl is not
4009 		 * intended to find arbitrary link.
4010 		 */
4011 		if (muxid == 0) {
4012 			return (EINVAL);
4013 		}
4014 
4015 		ns = netstack_find_by_cred(crp);
4016 		ASSERT(ns != NULL);
4017 		ss = ns->netstack_str;
4018 		ASSERT(ss != NULL);
4019 
4020 		mutex_enter(&muxifier);
4021 		linkp = findlinks(vp->v_stream, muxid, LINKPERSIST, ss);
4022 		if (linkp == NULL) {
4023 			mutex_exit(&muxifier);
4024 			netstack_rele(ss->ss_netstack);
4025 			return (EINVAL);
4026 		}
4027 
4028 		if ((fd = ufalloc(0)) == -1) {
4029 			mutex_exit(&muxifier);
4030 			netstack_rele(ss->ss_netstack);
4031 			return (EMFILE);
4032 		}
4033 		fp = linkp->li_fpdown;
4034 		mutex_enter(&fp->f_tlock);
4035 		fp->f_count++;
4036 		mutex_exit(&fp->f_tlock);
4037 		mutex_exit(&muxifier);
4038 		setf(fd, fp);
4039 		*rvalp = fd;
4040 		netstack_rele(ss->ss_netstack);
4041 		return (0);
4042 	}
4043 
4044 	case _I_INSERT:
4045 	{
4046 		/*
4047 		 * To insert a module to a given position in a stream.
4048 		 * In the first release, only allow privileged user
4049 		 * to use this ioctl. Furthermore, the insert is only allowed
4050 		 * below an anchor if the zoneid is the same as the zoneid
4051 		 * which created the anchor.
4052 		 *
4053 		 * Note that we do not plan to support this ioctl
4054 		 * on pipes in the first release.  We want to learn more
4055 		 * about the implications of these ioctls before extending
4056 		 * their support.  And we do not think these features are
4057 		 * valuable for pipes.
4058 		 */
4059 		STRUCT_DECL(strmodconf, strmodinsert);
4060 		char mod_name[FMNAMESZ + 1];
4061 		fmodsw_impl_t *fp;
4062 		dev_t dummydev;
4063 		queue_t *tmp_wrq;
4064 		int pos;
4065 		boolean_t is_insert;
4066 
4067 		STRUCT_INIT(strmodinsert, flag);
4068 		if (stp->sd_flag & STRHUP)
4069 			return (ENXIO);
4070 		if (STRMATED(stp))
4071 			return (EINVAL);
4072 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4073 			return (error);
4074 		if (stp->sd_anchor != 0 &&
4075 		    stp->sd_anchorzone != crgetzoneid(crp))
4076 			return (EINVAL);
4077 
4078 		error = strcopyin((void *)arg, STRUCT_BUF(strmodinsert),
4079 		    STRUCT_SIZE(strmodinsert), copyflag);
4080 		if (error)
4081 			return (error);
4082 
4083 		/*
4084 		 * Get module name and look up in fmodsw.
4085 		 */
4086 		error = (copyflag & U_TO_K ? copyinstr :
4087 		    copystr)(STRUCT_FGETP(strmodinsert, mod_name),
4088 		    mod_name, FMNAMESZ + 1, NULL);
4089 		if (error)
4090 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4091 
4092 		if ((fp = fmodsw_find(mod_name, FMODSW_HOLD | FMODSW_LOAD)) ==
4093 		    NULL)
4094 			return (EINVAL);
4095 
4096 		if (error = strstartplumb(stp, flag, cmd)) {
4097 			fmodsw_rele(fp);
4098 			return (error);
4099 		}
4100 
4101 		/*
4102 		 * Is this _I_INSERT just like an I_PUSH?  We need to know
4103 		 * this because we do some optimizations if this is a
4104 		 * module being pushed.
4105 		 */
4106 		pos = STRUCT_FGET(strmodinsert, pos);
4107 		is_insert = (pos != 0);
4108 
4109 		/*
4110 		 * Make sure pos is valid.  Even though it is not an I_PUSH,
4111 		 * we impose the same limit on the number of modules in a
4112 		 * stream.
4113 		 */
4114 		mutex_enter(&stp->sd_lock);
4115 		if (stp->sd_pushcnt >= nstrpush || pos < 0 ||
4116 		    pos > stp->sd_pushcnt) {
4117 			fmodsw_rele(fp);
4118 			strendplumb(stp);
4119 			mutex_exit(&stp->sd_lock);
4120 			return (EINVAL);
4121 		}
4122 		if (stp->sd_anchor != 0) {
4123 			/*
4124 			 * Is this insert below the anchor?
4125 			 * Pushcnt hasn't been increased yet hence
4126 			 * we test for greater than here, and greater or
4127 			 * equal after qattach.
4128 			 */
4129 			if (pos > (stp->sd_pushcnt - stp->sd_anchor) &&
4130 			    stp->sd_anchorzone != crgetzoneid(crp)) {
4131 				fmodsw_rele(fp);
4132 				strendplumb(stp);
4133 				mutex_exit(&stp->sd_lock);
4134 				return (EPERM);
4135 			}
4136 		}
4137 
4138 		mutex_exit(&stp->sd_lock);
4139 
4140 		/*
4141 		 * First find the correct position this module to
4142 		 * be inserted.  We don't need to call claimstr()
4143 		 * as the stream should not be changing at this point.
4144 		 *
4145 		 * Insert new module and call its open routine
4146 		 * via qattach().  Modules don't change device
4147 		 * numbers, so just ignore dummydev here.
4148 		 */
4149 		for (tmp_wrq = stp->sd_wrq; pos > 0;
4150 		    tmp_wrq = tmp_wrq->q_next, pos--) {
4151 			ASSERT(SAMESTR(tmp_wrq));
4152 		}
4153 		dummydev = vp->v_rdev;
4154 		if ((error = qattach(_RD(tmp_wrq), &dummydev, 0, crp,
4155 		    fp, is_insert)) != 0) {
4156 			mutex_enter(&stp->sd_lock);
4157 			strendplumb(stp);
4158 			mutex_exit(&stp->sd_lock);
4159 			return (error);
4160 		}
4161 
4162 		mutex_enter(&stp->sd_lock);
4163 
4164 		/*
4165 		 * As a performance concern we are caching the values of
4166 		 * q_minpsz and q_maxpsz of the module below the stream
4167 		 * head in the stream head.
4168 		 */
4169 		if (!is_insert) {
4170 			mutex_enter(QLOCK(stp->sd_wrq->q_next));
4171 			rmin = stp->sd_wrq->q_next->q_minpsz;
4172 			rmax = stp->sd_wrq->q_next->q_maxpsz;
4173 			mutex_exit(QLOCK(stp->sd_wrq->q_next));
4174 
4175 			/* Do this processing here as a performance concern */
4176 			if (strmsgsz != 0) {
4177 				if (rmax == INFPSZ) {
4178 					rmax = strmsgsz;
4179 				} else  {
4180 					rmax = MIN(strmsgsz, rmax);
4181 				}
4182 			}
4183 
4184 			mutex_enter(QLOCK(wrq));
4185 			stp->sd_qn_minpsz = rmin;
4186 			stp->sd_qn_maxpsz = rmax;
4187 			mutex_exit(QLOCK(wrq));
4188 		}
4189 
4190 		/*
4191 		 * Need to update the anchor value if this module is
4192 		 * inserted below the anchor point.
4193 		 */
4194 		if (stp->sd_anchor != 0) {
4195 			pos = STRUCT_FGET(strmodinsert, pos);
4196 			if (pos >= (stp->sd_pushcnt - stp->sd_anchor))
4197 				stp->sd_anchor++;
4198 		}
4199 
4200 		strendplumb(stp);
4201 		mutex_exit(&stp->sd_lock);
4202 		return (0);
4203 	}
4204 
4205 	case _I_REMOVE:
4206 	{
4207 		/*
4208 		 * To remove a module with a given name in a stream.  The
4209 		 * caller of this ioctl needs to provide both the name and
4210 		 * the position of the module to be removed.  This eliminates
4211 		 * the ambiguity of removal if a module is inserted/pushed
4212 		 * multiple times in a stream.  In the first release, only
4213 		 * allow privileged user to use this ioctl.
4214 		 * Furthermore, the remove is only allowed
4215 		 * below an anchor if the zoneid is the same as the zoneid
4216 		 * which created the anchor.
4217 		 *
4218 		 * Note that we do not plan to support this ioctl
4219 		 * on pipes in the first release.  We want to learn more
4220 		 * about the implications of these ioctls before extending
4221 		 * their support.  And we do not think these features are
4222 		 * valuable for pipes.
4223 		 *
4224 		 * Also note that _I_REMOVE cannot be used to remove a
4225 		 * driver or the stream head.
4226 		 */
4227 		STRUCT_DECL(strmodconf, strmodremove);
4228 		queue_t	*q;
4229 		int pos;
4230 		char mod_name[FMNAMESZ + 1];
4231 		boolean_t is_remove;
4232 
4233 		STRUCT_INIT(strmodremove, flag);
4234 		if (stp->sd_flag & STRHUP)
4235 			return (ENXIO);
4236 		if (STRMATED(stp))
4237 			return (EINVAL);
4238 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4239 			return (error);
4240 		if (stp->sd_anchor != 0 &&
4241 		    stp->sd_anchorzone != crgetzoneid(crp))
4242 			return (EINVAL);
4243 
4244 		error = strcopyin((void *)arg, STRUCT_BUF(strmodremove),
4245 		    STRUCT_SIZE(strmodremove), copyflag);
4246 		if (error)
4247 			return (error);
4248 
4249 		error = (copyflag & U_TO_K ? copyinstr :
4250 		    copystr)(STRUCT_FGETP(strmodremove, mod_name),
4251 		    mod_name, FMNAMESZ + 1, NULL);
4252 		if (error)
4253 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4254 
4255 		if ((error = strstartplumb(stp, flag, cmd)) != 0)
4256 			return (error);
4257 
4258 		/*
4259 		 * Match the name of given module to the name of module at
4260 		 * the given position.
4261 		 */
4262 		pos = STRUCT_FGET(strmodremove, pos);
4263 
4264 		is_remove = (pos != 0);
4265 		for (q = stp->sd_wrq->q_next; SAMESTR(q) && pos > 0;
4266 		    q = q->q_next, pos--)
4267 			;
4268 		if (pos > 0 || !SAMESTR(q) ||
4269 		    strcmp(Q2NAME(q), mod_name) != 0) {
4270 			mutex_enter(&stp->sd_lock);
4271 			strendplumb(stp);
4272 			mutex_exit(&stp->sd_lock);
4273 			return (EINVAL);
4274 		}
4275 
4276 		/*
4277 		 * If the position is at or below an anchor, then the zoneid
4278 		 * must match the zoneid that created the anchor.
4279 		 */
4280 		if (stp->sd_anchor != 0) {
4281 			pos = STRUCT_FGET(strmodremove, pos);
4282 			if (pos >= (stp->sd_pushcnt - stp->sd_anchor) &&
4283 			    stp->sd_anchorzone != crgetzoneid(crp)) {
4284 				mutex_enter(&stp->sd_lock);
4285 				strendplumb(stp);
4286 				mutex_exit(&stp->sd_lock);
4287 				return (EPERM);
4288 			}
4289 		}
4290 
4291 
4292 		ASSERT(!(q->q_flag & QREADR));
4293 		qdetach(_RD(q), 1, flag, crp, is_remove);
4294 
4295 		mutex_enter(&stp->sd_lock);
4296 
4297 		/*
4298 		 * As a performance concern we are caching the values of
4299 		 * q_minpsz and q_maxpsz of the module below the stream
4300 		 * head in the stream head.
4301 		 */
4302 		if (!is_remove) {
4303 			mutex_enter(QLOCK(wrq->q_next));
4304 			rmin = wrq->q_next->q_minpsz;
4305 			rmax = wrq->q_next->q_maxpsz;
4306 			mutex_exit(QLOCK(wrq->q_next));
4307 
4308 			/* Do this processing here as a performance concern */
4309 			if (strmsgsz != 0) {
4310 				if (rmax == INFPSZ)
4311 					rmax = strmsgsz;
4312 				else  {
4313 					if (vp->v_type == VFIFO)
4314 						rmax = MIN(PIPE_BUF, rmax);
4315 					else	rmax = MIN(strmsgsz, rmax);
4316 				}
4317 			}
4318 
4319 			mutex_enter(QLOCK(wrq));
4320 			stp->sd_qn_minpsz = rmin;
4321 			stp->sd_qn_maxpsz = rmax;
4322 			mutex_exit(QLOCK(wrq));
4323 		}
4324 
4325 		/*
4326 		 * Need to update the anchor value if this module is removed
4327 		 * at or below the anchor point.  If the removed module is at
4328 		 * the anchor point, remove the anchor for this stream if
4329 		 * there is no module above the anchor point.  Otherwise, if
4330 		 * the removed module is below the anchor point, decrement the
4331 		 * anchor point by 1.
4332 		 */
4333 		if (stp->sd_anchor != 0) {
4334 			pos = STRUCT_FGET(strmodremove, pos);
4335 			if (pos == stp->sd_pushcnt - stp->sd_anchor + 1)
4336 				stp->sd_anchor = 0;
4337 			else if (pos > (stp->sd_pushcnt - stp->sd_anchor + 1))
4338 				stp->sd_anchor--;
4339 		}
4340 
4341 		strendplumb(stp);
4342 		mutex_exit(&stp->sd_lock);
4343 		return (0);
4344 	}
4345 
4346 	case I_ANCHOR:
4347 		/*
4348 		 * Set the anchor position on the stream to reside at
4349 		 * the top module (in other words, the top module
4350 		 * cannot be popped).  Anchors with a FIFO make no
4351 		 * obvious sense, so they're not allowed.
4352 		 */
4353 		mutex_enter(&stp->sd_lock);
4354 
4355 		if (stp->sd_vnode->v_type == VFIFO) {
4356 			mutex_exit(&stp->sd_lock);
4357 			return (EINVAL);
4358 		}
4359 		/* Only allow the same zoneid to update the anchor */
4360 		if (stp->sd_anchor != 0 &&
4361 		    stp->sd_anchorzone != crgetzoneid(crp)) {
4362 			mutex_exit(&stp->sd_lock);
4363 			return (EINVAL);
4364 		}
4365 		stp->sd_anchor = stp->sd_pushcnt;
4366 		stp->sd_anchorzone = crgetzoneid(crp);
4367 		mutex_exit(&stp->sd_lock);
4368 		return (0);
4369 
4370 	case I_LOOK:
4371 		/*
4372 		 * Get name of first module downstream.
4373 		 * If no module, return an error.
4374 		 */
4375 		claimstr(wrq);
4376 		if (_SAMESTR(wrq) && wrq->q_next->q_next != NULL) {
4377 			char *name = Q2NAME(wrq->q_next);
4378 
4379 			error = strcopyout(name, (void *)arg, strlen(name) + 1,
4380 			    copyflag);
4381 			releasestr(wrq);
4382 			return (error);
4383 		}
4384 		releasestr(wrq);
4385 		return (EINVAL);
4386 
4387 	case I_LINK:
4388 	case I_PLINK:
4389 		/*
4390 		 * Link a multiplexor.
4391 		 */
4392 		return (mlink(vp, cmd, (int)arg, crp, rvalp, 0));
4393 
4394 	case _I_PLINK_LH:
4395 		/*
4396 		 * Link a multiplexor: Call must originate from kernel.
4397 		 */
4398 		if (kioctl)
4399 			return (ldi_mlink_lh(vp, cmd, arg, crp, rvalp));
4400 
4401 		return (EINVAL);
4402 	case I_UNLINK:
4403 	case I_PUNLINK:
4404 		/*
4405 		 * Unlink a multiplexor.
4406 		 * If arg is -1, unlink all links for which this is the
4407 		 * controlling stream.  Otherwise, arg is an index number
4408 		 * for a link to be removed.
4409 		 */
4410 	{
4411 		struct linkinfo *linkp;
4412 		int native_arg = (int)arg;
4413 		int type;
4414 		netstack_t *ns;
4415 		str_stack_t *ss;
4416 
4417 		TRACE_1(TR_FAC_STREAMS_FR,
4418 		    TR_I_UNLINK, "I_UNLINK/I_PUNLINK:%p", stp);
4419 		if (vp->v_type == VFIFO) {
4420 			return (EINVAL);
4421 		}
4422 		if (cmd == I_UNLINK)
4423 			type = LINKNORMAL;
4424 		else	/* I_PUNLINK */
4425 			type = LINKPERSIST;
4426 		if (native_arg == 0) {
4427 			return (EINVAL);
4428 		}
4429 		ns = netstack_find_by_cred(crp);
4430 		ASSERT(ns != NULL);
4431 		ss = ns->netstack_str;
4432 		ASSERT(ss != NULL);
4433 
4434 		if (native_arg == MUXID_ALL)
4435 			error = munlinkall(stp, type, crp, rvalp, ss);
4436 		else {
4437 			mutex_enter(&muxifier);
4438 			if (!(linkp = findlinks(stp, (int)arg, type, ss))) {
4439 				/* invalid user supplied index number */
4440 				mutex_exit(&muxifier);
4441 				netstack_rele(ss->ss_netstack);
4442 				return (EINVAL);
4443 			}
4444 			/* munlink drops the muxifier lock */
4445 			error = munlink(stp, linkp, type, crp, rvalp, ss);
4446 		}
4447 		netstack_rele(ss->ss_netstack);
4448 		return (error);
4449 	}
4450 
4451 	case I_FLUSH:
4452 		/*
4453 		 * send a flush message downstream
4454 		 * flush message can indicate
4455 		 * FLUSHR - flush read queue
4456 		 * FLUSHW - flush write queue
4457 		 * FLUSHRW - flush read/write queue
4458 		 */
4459 		if (stp->sd_flag & STRHUP)
4460 			return (ENXIO);
4461 		if (arg & ~FLUSHRW)
4462 			return (EINVAL);
4463 
4464 		for (;;) {
4465 			if (putnextctl1(stp->sd_wrq, M_FLUSH, (int)arg)) {
4466 				break;
4467 			}
4468 			if (error = strwaitbuf(1, BPRI_HI)) {
4469 				return (error);
4470 			}
4471 		}
4472 
4473 		/*
4474 		 * Send down an unsupported ioctl and wait for the nack
4475 		 * in order to allow the M_FLUSH to propagate back
4476 		 * up to the stream head.
4477 		 * Replaces if (qready()) runqueues();
4478 		 */
4479 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4480 		strioc.ic_timout = 0;
4481 		strioc.ic_len = 0;
4482 		strioc.ic_dp = NULL;
4483 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4484 		*rvalp = 0;
4485 		return (0);
4486 
4487 	case I_FLUSHBAND:
4488 	{
4489 		struct bandinfo binfo;
4490 
4491 		error = strcopyin((void *)arg, &binfo, sizeof (binfo),
4492 		    copyflag);
4493 		if (error)
4494 			return (error);
4495 		if (stp->sd_flag & STRHUP)
4496 			return (ENXIO);
4497 		if (binfo.bi_flag & ~FLUSHRW)
4498 			return (EINVAL);
4499 		while (!(mp = allocb(2, BPRI_HI))) {
4500 			if (error = strwaitbuf(2, BPRI_HI))
4501 				return (error);
4502 		}
4503 		mp->b_datap->db_type = M_FLUSH;
4504 		*mp->b_wptr++ = binfo.bi_flag | FLUSHBAND;
4505 		*mp->b_wptr++ = binfo.bi_pri;
4506 		putnext(stp->sd_wrq, mp);
4507 		/*
4508 		 * Send down an unsupported ioctl and wait for the nack
4509 		 * in order to allow the M_FLUSH to propagate back
4510 		 * up to the stream head.
4511 		 * Replaces if (qready()) runqueues();
4512 		 */
4513 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4514 		strioc.ic_timout = 0;
4515 		strioc.ic_len = 0;
4516 		strioc.ic_dp = NULL;
4517 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4518 		*rvalp = 0;
4519 		return (0);
4520 	}
4521 
4522 	case I_SRDOPT:
4523 		/*
4524 		 * Set read options
4525 		 *
4526 		 * RNORM - default stream mode
4527 		 * RMSGN - message no discard
4528 		 * RMSGD - message discard
4529 		 * RPROTNORM - fail read with EBADMSG for M_[PC]PROTOs
4530 		 * RPROTDAT - convert M_[PC]PROTOs to M_DATAs
4531 		 * RPROTDIS - discard M_[PC]PROTOs and retain M_DATAs
4532 		 */
4533 		if (arg & ~(RMODEMASK | RPROTMASK))
4534 			return (EINVAL);
4535 
4536 		if ((arg & (RMSGD|RMSGN)) == (RMSGD|RMSGN))
4537 			return (EINVAL);
4538 
4539 		mutex_enter(&stp->sd_lock);
4540 		switch (arg & RMODEMASK) {
4541 		case RNORM:
4542 			stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
4543 			break;
4544 		case RMSGD:
4545 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGNODIS) |
4546 			    RD_MSGDIS;
4547 			break;
4548 		case RMSGN:
4549 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGDIS) |
4550 			    RD_MSGNODIS;
4551 			break;
4552 		}
4553 
4554 		switch (arg & RPROTMASK) {
4555 		case RPROTNORM:
4556 			stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
4557 			break;
4558 
4559 		case RPROTDAT:
4560 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDIS) |
4561 			    RD_PROTDAT);
4562 			break;
4563 
4564 		case RPROTDIS:
4565 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDAT) |
4566 			    RD_PROTDIS);
4567 			break;
4568 		}
4569 		mutex_exit(&stp->sd_lock);
4570 		return (0);
4571 
4572 	case I_GRDOPT:
4573 		/*
4574 		 * Get read option and return the value
4575 		 * to spot pointed to by arg
4576 		 */
4577 	{
4578 		int rdopt;
4579 
4580 		rdopt = ((stp->sd_read_opt & RD_MSGDIS) ? RMSGD :
4581 		    ((stp->sd_read_opt & RD_MSGNODIS) ? RMSGN : RNORM));
4582 		rdopt |= ((stp->sd_read_opt & RD_PROTDAT) ? RPROTDAT :
4583 		    ((stp->sd_read_opt & RD_PROTDIS) ? RPROTDIS : RPROTNORM));
4584 
4585 		return (strcopyout(&rdopt, (void *)arg, sizeof (int),
4586 		    copyflag));
4587 	}
4588 
4589 	case I_SERROPT:
4590 		/*
4591 		 * Set error options
4592 		 *
4593 		 * RERRNORM - persistent read errors
4594 		 * RERRNONPERSIST - non-persistent read errors
4595 		 * WERRNORM - persistent write errors
4596 		 * WERRNONPERSIST - non-persistent write errors
4597 		 */
4598 		if (arg & ~(RERRMASK | WERRMASK))
4599 			return (EINVAL);
4600 
4601 		mutex_enter(&stp->sd_lock);
4602 		switch (arg & RERRMASK) {
4603 		case RERRNORM:
4604 			stp->sd_flag &= ~STRDERRNONPERSIST;
4605 			break;
4606 		case RERRNONPERSIST:
4607 			stp->sd_flag |= STRDERRNONPERSIST;
4608 			break;
4609 		}
4610 		switch (arg & WERRMASK) {
4611 		case WERRNORM:
4612 			stp->sd_flag &= ~STWRERRNONPERSIST;
4613 			break;
4614 		case WERRNONPERSIST:
4615 			stp->sd_flag |= STWRERRNONPERSIST;
4616 			break;
4617 		}
4618 		mutex_exit(&stp->sd_lock);
4619 		return (0);
4620 
4621 	case I_GERROPT:
4622 		/*
4623 		 * Get error option and return the value
4624 		 * to spot pointed to by arg
4625 		 */
4626 	{
4627 		int erropt = 0;
4628 
4629 		erropt |= (stp->sd_flag & STRDERRNONPERSIST) ? RERRNONPERSIST :
4630 		    RERRNORM;
4631 		erropt |= (stp->sd_flag & STWRERRNONPERSIST) ? WERRNONPERSIST :
4632 		    WERRNORM;
4633 		return (strcopyout(&erropt, (void *)arg, sizeof (int),
4634 		    copyflag));
4635 	}
4636 
4637 	case I_SETSIG:
4638 		/*
4639 		 * Register the calling proc to receive the SIGPOLL
4640 		 * signal based on the events given in arg.  If
4641 		 * arg is zero, remove the proc from register list.
4642 		 */
4643 	{
4644 		strsig_t *ssp, *pssp;
4645 		struct pid *pidp;
4646 
4647 		pssp = NULL;
4648 		pidp = curproc->p_pidp;
4649 		/*
4650 		 * Hold sd_lock to prevent traversal of sd_siglist while
4651 		 * it is modified.
4652 		 */
4653 		mutex_enter(&stp->sd_lock);
4654 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pidp != pidp);
4655 		    pssp = ssp, ssp = ssp->ss_next)
4656 			;
4657 
4658 		if (arg) {
4659 			if (arg & ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4660 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4661 				mutex_exit(&stp->sd_lock);
4662 				return (EINVAL);
4663 			}
4664 			if ((arg & S_BANDURG) && !(arg & S_RDBAND)) {
4665 				mutex_exit(&stp->sd_lock);
4666 				return (EINVAL);
4667 			}
4668 
4669 			/*
4670 			 * If proc not already registered, add it
4671 			 * to list.
4672 			 */
4673 			if (!ssp) {
4674 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4675 				ssp->ss_pidp = pidp;
4676 				ssp->ss_pid = pidp->pid_id;
4677 				ssp->ss_next = NULL;
4678 				if (pssp)
4679 					pssp->ss_next = ssp;
4680 				else
4681 					stp->sd_siglist = ssp;
4682 				mutex_enter(&pidlock);
4683 				PID_HOLD(pidp);
4684 				mutex_exit(&pidlock);
4685 			}
4686 
4687 			/*
4688 			 * Set events.
4689 			 */
4690 			ssp->ss_events = (int)arg;
4691 		} else {
4692 			/*
4693 			 * Remove proc from register list.
4694 			 */
4695 			if (ssp) {
4696 				mutex_enter(&pidlock);
4697 				PID_RELE(pidp);
4698 				mutex_exit(&pidlock);
4699 				if (pssp)
4700 					pssp->ss_next = ssp->ss_next;
4701 				else
4702 					stp->sd_siglist = ssp->ss_next;
4703 				kmem_free(ssp, sizeof (strsig_t));
4704 			} else {
4705 				mutex_exit(&stp->sd_lock);
4706 				return (EINVAL);
4707 			}
4708 		}
4709 
4710 		/*
4711 		 * Recalculate OR of sig events.
4712 		 */
4713 		stp->sd_sigflags = 0;
4714 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4715 			stp->sd_sigflags |= ssp->ss_events;
4716 		mutex_exit(&stp->sd_lock);
4717 		return (0);
4718 	}
4719 
4720 	case I_GETSIG:
4721 		/*
4722 		 * Return (in arg) the current registration of events
4723 		 * for which the calling proc is to be signaled.
4724 		 */
4725 	{
4726 		struct strsig *ssp;
4727 		struct pid  *pidp;
4728 
4729 		pidp = curproc->p_pidp;
4730 		mutex_enter(&stp->sd_lock);
4731 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4732 			if (ssp->ss_pidp == pidp) {
4733 				error = strcopyout(&ssp->ss_events, (void *)arg,
4734 				    sizeof (int), copyflag);
4735 				mutex_exit(&stp->sd_lock);
4736 				return (error);
4737 			}
4738 		mutex_exit(&stp->sd_lock);
4739 		return (EINVAL);
4740 	}
4741 
4742 	case I_ESETSIG:
4743 		/*
4744 		 * Register the ss_pid to receive the SIGPOLL
4745 		 * signal based on the events is ss_events arg.  If
4746 		 * ss_events is zero, remove the proc from register list.
4747 		 */
4748 	{
4749 		struct strsig *ssp, *pssp;
4750 		struct proc *proc;
4751 		struct pid  *pidp;
4752 		pid_t pid;
4753 		struct strsigset ss;
4754 
4755 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4756 		if (error)
4757 			return (error);
4758 
4759 		pid = ss.ss_pid;
4760 
4761 		if (ss.ss_events != 0) {
4762 			/*
4763 			 * Permissions check by sending signal 0.
4764 			 * Note that when kill fails it does a set_errno
4765 			 * causing the system call to fail.
4766 			 */
4767 			error = kill(pid, 0);
4768 			if (error) {
4769 				return (error);
4770 			}
4771 		}
4772 		mutex_enter(&pidlock);
4773 		if (pid == 0)
4774 			proc = curproc;
4775 		else if (pid < 0)
4776 			proc = pgfind(-pid);
4777 		else
4778 			proc = prfind(pid);
4779 		if (proc == NULL) {
4780 			mutex_exit(&pidlock);
4781 			return (ESRCH);
4782 		}
4783 		if (pid < 0)
4784 			pidp = proc->p_pgidp;
4785 		else
4786 			pidp = proc->p_pidp;
4787 		ASSERT(pidp);
4788 		/*
4789 		 * Get a hold on the pid structure while referencing it.
4790 		 * There is a separate PID_HOLD should it be inserted
4791 		 * in the list below.
4792 		 */
4793 		PID_HOLD(pidp);
4794 		mutex_exit(&pidlock);
4795 
4796 		pssp = NULL;
4797 		/*
4798 		 * Hold sd_lock to prevent traversal of sd_siglist while
4799 		 * it is modified.
4800 		 */
4801 		mutex_enter(&stp->sd_lock);
4802 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pid != pid);
4803 		    pssp = ssp, ssp = ssp->ss_next)
4804 			;
4805 
4806 		if (ss.ss_events) {
4807 			if (ss.ss_events &
4808 			    ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4809 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4810 				mutex_exit(&stp->sd_lock);
4811 				mutex_enter(&pidlock);
4812 				PID_RELE(pidp);
4813 				mutex_exit(&pidlock);
4814 				return (EINVAL);
4815 			}
4816 			if ((ss.ss_events & S_BANDURG) &&
4817 			    !(ss.ss_events & S_RDBAND)) {
4818 				mutex_exit(&stp->sd_lock);
4819 				mutex_enter(&pidlock);
4820 				PID_RELE(pidp);
4821 				mutex_exit(&pidlock);
4822 				return (EINVAL);
4823 			}
4824 
4825 			/*
4826 			 * If proc not already registered, add it
4827 			 * to list.
4828 			 */
4829 			if (!ssp) {
4830 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4831 				ssp->ss_pidp = pidp;
4832 				ssp->ss_pid = pid;
4833 				ssp->ss_next = NULL;
4834 				if (pssp)
4835 					pssp->ss_next = ssp;
4836 				else
4837 					stp->sd_siglist = ssp;
4838 				mutex_enter(&pidlock);
4839 				PID_HOLD(pidp);
4840 				mutex_exit(&pidlock);
4841 			}
4842 
4843 			/*
4844 			 * Set events.
4845 			 */
4846 			ssp->ss_events = ss.ss_events;
4847 		} else {
4848 			/*
4849 			 * Remove proc from register list.
4850 			 */
4851 			if (ssp) {
4852 				mutex_enter(&pidlock);
4853 				PID_RELE(pidp);
4854 				mutex_exit(&pidlock);
4855 				if (pssp)
4856 					pssp->ss_next = ssp->ss_next;
4857 				else
4858 					stp->sd_siglist = ssp->ss_next;
4859 				kmem_free(ssp, sizeof (strsig_t));
4860 			} else {
4861 				mutex_exit(&stp->sd_lock);
4862 				mutex_enter(&pidlock);
4863 				PID_RELE(pidp);
4864 				mutex_exit(&pidlock);
4865 				return (EINVAL);
4866 			}
4867 		}
4868 
4869 		/*
4870 		 * Recalculate OR of sig events.
4871 		 */
4872 		stp->sd_sigflags = 0;
4873 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4874 			stp->sd_sigflags |= ssp->ss_events;
4875 		mutex_exit(&stp->sd_lock);
4876 		mutex_enter(&pidlock);
4877 		PID_RELE(pidp);
4878 		mutex_exit(&pidlock);
4879 		return (0);
4880 	}
4881 
4882 	case I_EGETSIG:
4883 		/*
4884 		 * Return (in arg) the current registration of events
4885 		 * for which the calling proc is to be signaled.
4886 		 */
4887 	{
4888 		struct strsig *ssp;
4889 		struct proc *proc;
4890 		pid_t pid;
4891 		struct pid  *pidp;
4892 		struct strsigset ss;
4893 
4894 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4895 		if (error)
4896 			return (error);
4897 
4898 		pid = ss.ss_pid;
4899 		mutex_enter(&pidlock);
4900 		if (pid == 0)
4901 			proc = curproc;
4902 		else if (pid < 0)
4903 			proc = pgfind(-pid);
4904 		else
4905 			proc = prfind(pid);
4906 		if (proc == NULL) {
4907 			mutex_exit(&pidlock);
4908 			return (ESRCH);
4909 		}
4910 		if (pid < 0)
4911 			pidp = proc->p_pgidp;
4912 		else
4913 			pidp = proc->p_pidp;
4914 
4915 		/* Prevent the pidp from being reassigned */
4916 		PID_HOLD(pidp);
4917 		mutex_exit(&pidlock);
4918 
4919 		mutex_enter(&stp->sd_lock);
4920 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4921 			if (ssp->ss_pid == pid) {
4922 				ss.ss_pid = ssp->ss_pid;
4923 				ss.ss_events = ssp->ss_events;
4924 				error = strcopyout(&ss, (void *)arg,
4925 				    sizeof (struct strsigset), copyflag);
4926 				mutex_exit(&stp->sd_lock);
4927 				mutex_enter(&pidlock);
4928 				PID_RELE(pidp);
4929 				mutex_exit(&pidlock);
4930 				return (error);
4931 			}
4932 		mutex_exit(&stp->sd_lock);
4933 		mutex_enter(&pidlock);
4934 		PID_RELE(pidp);
4935 		mutex_exit(&pidlock);
4936 		return (EINVAL);
4937 	}
4938 
4939 	case I_PEEK:
4940 	{
4941 		STRUCT_DECL(strpeek, strpeek);
4942 		size_t n;
4943 		mblk_t *fmp, *tmp_mp = NULL;
4944 
4945 		STRUCT_INIT(strpeek, flag);
4946 
4947 		error = strcopyin((void *)arg, STRUCT_BUF(strpeek),
4948 		    STRUCT_SIZE(strpeek), copyflag);
4949 		if (error)
4950 			return (error);
4951 
4952 		mutex_enter(QLOCK(rdq));
4953 		/*
4954 		 * Skip the invalid messages
4955 		 */
4956 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
4957 			if (mp->b_datap->db_type != M_SIG)
4958 				break;
4959 
4960 		/*
4961 		 * If user has requested to peek at a high priority message
4962 		 * and first message is not, return 0
4963 		 */
4964 		if (mp != NULL) {
4965 			if ((STRUCT_FGET(strpeek, flags) & RS_HIPRI) &&
4966 			    queclass(mp) == QNORM) {
4967 				*rvalp = 0;
4968 				mutex_exit(QLOCK(rdq));
4969 				return (0);
4970 			}
4971 		} else if (stp->sd_struiordq == NULL ||
4972 		    (STRUCT_FGET(strpeek, flags) & RS_HIPRI)) {
4973 			/*
4974 			 * No mblks to look at at the streamhead and
4975 			 * 1). This isn't a synch stream or
4976 			 * 2). This is a synch stream but caller wants high
4977 			 *	priority messages which is not supported by
4978 			 *	the synch stream. (it only supports QNORM)
4979 			 */
4980 			*rvalp = 0;
4981 			mutex_exit(QLOCK(rdq));
4982 			return (0);
4983 		}
4984 
4985 		fmp = mp;
4986 
4987 		if (mp && mp->b_datap->db_type == M_PASSFP) {
4988 			mutex_exit(QLOCK(rdq));
4989 			return (EBADMSG);
4990 		}
4991 
4992 		ASSERT(mp == NULL || mp->b_datap->db_type == M_PCPROTO ||
4993 		    mp->b_datap->db_type == M_PROTO ||
4994 		    mp->b_datap->db_type == M_DATA);
4995 
4996 		if (mp && mp->b_datap->db_type == M_PCPROTO) {
4997 			STRUCT_FSET(strpeek, flags, RS_HIPRI);
4998 		} else {
4999 			STRUCT_FSET(strpeek, flags, 0);
5000 		}
5001 
5002 
5003 		if (mp && ((tmp_mp = dupmsg(mp)) == NULL)) {
5004 			mutex_exit(QLOCK(rdq));
5005 			return (ENOSR);
5006 		}
5007 		mutex_exit(QLOCK(rdq));
5008 
5009 		/*
5010 		 * set mp = tmp_mp, so that I_PEEK processing can continue.
5011 		 * tmp_mp is used to free the dup'd message.
5012 		 */
5013 		mp = tmp_mp;
5014 
5015 		uio.uio_fmode = 0;
5016 		uio.uio_extflg = UIO_COPY_CACHED;
5017 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5018 		    UIO_SYSSPACE;
5019 		uio.uio_limit = 0;
5020 		/*
5021 		 * First process PROTO blocks, if any.
5022 		 * If user doesn't want to get ctl info by setting maxlen <= 0,
5023 		 * then set len to -1/0 and skip control blocks part.
5024 		 */
5025 		if (STRUCT_FGET(strpeek, ctlbuf.maxlen) < 0)
5026 			STRUCT_FSET(strpeek, ctlbuf.len, -1);
5027 		else if (STRUCT_FGET(strpeek, ctlbuf.maxlen) == 0)
5028 			STRUCT_FSET(strpeek, ctlbuf.len, 0);
5029 		else {
5030 			int	ctl_part = 0;
5031 
5032 			iov.iov_base = STRUCT_FGETP(strpeek, ctlbuf.buf);
5033 			iov.iov_len = STRUCT_FGET(strpeek, ctlbuf.maxlen);
5034 			uio.uio_iov = &iov;
5035 			uio.uio_resid = iov.iov_len;
5036 			uio.uio_loffset = 0;
5037 			uio.uio_iovcnt = 1;
5038 			while (mp && mp->b_datap->db_type != M_DATA &&
5039 			    uio.uio_resid >= 0) {
5040 				ASSERT(STRUCT_FGET(strpeek, flags) == 0 ?
5041 				    mp->b_datap->db_type == M_PROTO :
5042 				    mp->b_datap->db_type == M_PCPROTO);
5043 
5044 				if ((n = MIN(uio.uio_resid,
5045 				    mp->b_wptr - mp->b_rptr)) != 0 &&
5046 				    (error = uiomove((char *)mp->b_rptr, n,
5047 				    UIO_READ, &uio)) != 0) {
5048 					freemsg(tmp_mp);
5049 					return (error);
5050 				}
5051 				ctl_part = 1;
5052 				mp = mp->b_cont;
5053 			}
5054 			/* No ctl message */
5055 			if (ctl_part == 0)
5056 				STRUCT_FSET(strpeek, ctlbuf.len, -1);
5057 			else
5058 				STRUCT_FSET(strpeek, ctlbuf.len,
5059 				    STRUCT_FGET(strpeek, ctlbuf.maxlen) -
5060 				    uio.uio_resid);
5061 		}
5062 
5063 		/*
5064 		 * Now process DATA blocks, if any.
5065 		 * If user doesn't want to get data info by setting maxlen <= 0,
5066 		 * then set len to -1/0 and skip data blocks part.
5067 		 */
5068 		if (STRUCT_FGET(strpeek, databuf.maxlen) < 0)
5069 			STRUCT_FSET(strpeek, databuf.len, -1);
5070 		else if (STRUCT_FGET(strpeek, databuf.maxlen) == 0)
5071 			STRUCT_FSET(strpeek, databuf.len, 0);
5072 		else {
5073 			int	data_part = 0;
5074 
5075 			iov.iov_base = STRUCT_FGETP(strpeek, databuf.buf);
5076 			iov.iov_len = STRUCT_FGET(strpeek, databuf.maxlen);
5077 			uio.uio_iov = &iov;
5078 			uio.uio_resid = iov.iov_len;
5079 			uio.uio_loffset = 0;
5080 			uio.uio_iovcnt = 1;
5081 			while (mp && uio.uio_resid) {
5082 				if (mp->b_datap->db_type == M_DATA) {
5083 					if ((n = MIN(uio.uio_resid,
5084 					    mp->b_wptr - mp->b_rptr)) != 0 &&
5085 					    (error = uiomove((char *)mp->b_rptr,
5086 					    n, UIO_READ, &uio)) != 0) {
5087 						freemsg(tmp_mp);
5088 						return (error);
5089 					}
5090 					data_part = 1;
5091 				}
5092 				ASSERT(data_part == 0 ||
5093 				    mp->b_datap->db_type == M_DATA);
5094 				mp = mp->b_cont;
5095 			}
5096 			/* No data message */
5097 			if (data_part == 0)
5098 				STRUCT_FSET(strpeek, databuf.len, -1);
5099 			else
5100 				STRUCT_FSET(strpeek, databuf.len,
5101 				    STRUCT_FGET(strpeek, databuf.maxlen) -
5102 				    uio.uio_resid);
5103 		}
5104 		freemsg(tmp_mp);
5105 
5106 		/*
5107 		 * It is a synch stream and user wants to get
5108 		 * data (maxlen > 0).
5109 		 * uio setup is done by the codes that process DATA
5110 		 * blocks above.
5111 		 */
5112 		if ((fmp == NULL) && STRUCT_FGET(strpeek, databuf.maxlen) > 0) {
5113 			infod_t infod;
5114 
5115 			infod.d_cmd = INFOD_COPYOUT;
5116 			infod.d_res = 0;
5117 			infod.d_uiop = &uio;
5118 			error = infonext(rdq, &infod);
5119 			if (error == EINVAL || error == EBUSY)
5120 				error = 0;
5121 			if (error)
5122 				return (error);
5123 			STRUCT_FSET(strpeek, databuf.len, STRUCT_FGET(strpeek,
5124 			    databuf.maxlen) - uio.uio_resid);
5125 			if (STRUCT_FGET(strpeek, databuf.len) == 0) {
5126 				/*
5127 				 * No data found by the infonext().
5128 				 */
5129 				STRUCT_FSET(strpeek, databuf.len, -1);
5130 			}
5131 		}
5132 		error = strcopyout(STRUCT_BUF(strpeek), (void *)arg,
5133 		    STRUCT_SIZE(strpeek), copyflag);
5134 		if (error) {
5135 			return (error);
5136 		}
5137 		/*
5138 		 * If there is no message retrieved, set return code to 0
5139 		 * otherwise, set it to 1.
5140 		 */
5141 		if (STRUCT_FGET(strpeek, ctlbuf.len) == -1 &&
5142 		    STRUCT_FGET(strpeek, databuf.len) == -1)
5143 			*rvalp = 0;
5144 		else
5145 			*rvalp = 1;
5146 		return (0);
5147 	}
5148 
5149 	case I_FDINSERT:
5150 	{
5151 		STRUCT_DECL(strfdinsert, strfdinsert);
5152 		struct file *resftp;
5153 		struct stdata *resstp;
5154 		t_uscalar_t	ival;
5155 		ssize_t msgsize;
5156 		struct strbuf mctl;
5157 
5158 		STRUCT_INIT(strfdinsert, flag);
5159 		if (stp->sd_flag & STRHUP)
5160 			return (ENXIO);
5161 		/*
5162 		 * STRDERR, STWRERR and STPLEX tested above.
5163 		 */
5164 		error = strcopyin((void *)arg, STRUCT_BUF(strfdinsert),
5165 		    STRUCT_SIZE(strfdinsert), copyflag);
5166 		if (error)
5167 			return (error);
5168 
5169 		if (STRUCT_FGET(strfdinsert, offset) < 0 ||
5170 		    (STRUCT_FGET(strfdinsert, offset) %
5171 		    sizeof (t_uscalar_t)) != 0)
5172 			return (EINVAL);
5173 		if ((resftp = getf(STRUCT_FGET(strfdinsert, fildes))) != NULL) {
5174 			if ((resstp = resftp->f_vnode->v_stream) == NULL) {
5175 				releasef(STRUCT_FGET(strfdinsert, fildes));
5176 				return (EINVAL);
5177 			}
5178 		} else
5179 			return (EINVAL);
5180 
5181 		mutex_enter(&resstp->sd_lock);
5182 		if (resstp->sd_flag & (STRDERR|STWRERR|STRHUP|STPLEX)) {
5183 			error = strgeterr(resstp,
5184 			    STRDERR|STWRERR|STRHUP|STPLEX, 0);
5185 			if (error != 0) {
5186 				mutex_exit(&resstp->sd_lock);
5187 				releasef(STRUCT_FGET(strfdinsert, fildes));
5188 				return (error);
5189 			}
5190 		}
5191 		mutex_exit(&resstp->sd_lock);
5192 
5193 #ifdef	_ILP32
5194 		{
5195 			queue_t	*q;
5196 			queue_t	*mate = NULL;
5197 
5198 			/* get read queue of stream terminus */
5199 			claimstr(resstp->sd_wrq);
5200 			for (q = resstp->sd_wrq->q_next; q->q_next != NULL;
5201 			    q = q->q_next)
5202 				if (!STRMATED(resstp) && STREAM(q) != resstp &&
5203 				    mate == NULL) {
5204 					ASSERT(q->q_qinfo->qi_srvp);
5205 					ASSERT(_OTHERQ(q)->q_qinfo->qi_srvp);
5206 					claimstr(q);
5207 					mate = q;
5208 				}
5209 			q = _RD(q);
5210 			if (mate)
5211 				releasestr(mate);
5212 			releasestr(resstp->sd_wrq);
5213 			ival = (t_uscalar_t)q;
5214 		}
5215 #else
5216 		ival = (t_uscalar_t)getminor(resftp->f_vnode->v_rdev);
5217 #endif	/* _ILP32 */
5218 
5219 		if (STRUCT_FGET(strfdinsert, ctlbuf.len) <
5220 		    STRUCT_FGET(strfdinsert, offset) + sizeof (t_uscalar_t)) {
5221 			releasef(STRUCT_FGET(strfdinsert, fildes));
5222 			return (EINVAL);
5223 		}
5224 
5225 		/*
5226 		 * Check for legal flag value.
5227 		 */
5228 		if (STRUCT_FGET(strfdinsert, flags) & ~RS_HIPRI) {
5229 			releasef(STRUCT_FGET(strfdinsert, fildes));
5230 			return (EINVAL);
5231 		}
5232 
5233 		/* get these values from those cached in the stream head */
5234 		mutex_enter(QLOCK(stp->sd_wrq));
5235 		rmin = stp->sd_qn_minpsz;
5236 		rmax = stp->sd_qn_maxpsz;
5237 		mutex_exit(QLOCK(stp->sd_wrq));
5238 
5239 		/*
5240 		 * Make sure ctl and data sizes together fall within
5241 		 * the limits of the max and min receive packet sizes
5242 		 * and do not exceed system limit.  A negative data
5243 		 * length means that no data part is to be sent.
5244 		 */
5245 		ASSERT((rmax >= 0) || (rmax == INFPSZ));
5246 		if (rmax == 0) {
5247 			releasef(STRUCT_FGET(strfdinsert, fildes));
5248 			return (ERANGE);
5249 		}
5250 		if ((msgsize = STRUCT_FGET(strfdinsert, databuf.len)) < 0)
5251 			msgsize = 0;
5252 		if ((msgsize < rmin) ||
5253 		    ((msgsize > rmax) && (rmax != INFPSZ)) ||
5254 		    (STRUCT_FGET(strfdinsert, ctlbuf.len) > strctlsz)) {
5255 			releasef(STRUCT_FGET(strfdinsert, fildes));
5256 			return (ERANGE);
5257 		}
5258 
5259 		mutex_enter(&stp->sd_lock);
5260 		while (!(STRUCT_FGET(strfdinsert, flags) & RS_HIPRI) &&
5261 		    !canputnext(stp->sd_wrq)) {
5262 			if ((error = strwaitq(stp, WRITEWAIT, (ssize_t)0,
5263 			    flag, -1, &done)) != 0 || done) {
5264 				mutex_exit(&stp->sd_lock);
5265 				releasef(STRUCT_FGET(strfdinsert, fildes));
5266 				return (error);
5267 			}
5268 			if ((error = i_straccess(stp, access)) != 0) {
5269 				mutex_exit(&stp->sd_lock);
5270 				releasef(
5271 				    STRUCT_FGET(strfdinsert, fildes));
5272 				return (error);
5273 			}
5274 		}
5275 		mutex_exit(&stp->sd_lock);
5276 
5277 		/*
5278 		 * Copy strfdinsert.ctlbuf into native form of
5279 		 * ctlbuf to pass down into strmakemsg().
5280 		 */
5281 		mctl.maxlen = STRUCT_FGET(strfdinsert, ctlbuf.maxlen);
5282 		mctl.len = STRUCT_FGET(strfdinsert, ctlbuf.len);
5283 		mctl.buf = STRUCT_FGETP(strfdinsert, ctlbuf.buf);
5284 
5285 		iov.iov_base = STRUCT_FGETP(strfdinsert, databuf.buf);
5286 		iov.iov_len = STRUCT_FGET(strfdinsert, databuf.len);
5287 		uio.uio_iov = &iov;
5288 		uio.uio_iovcnt = 1;
5289 		uio.uio_loffset = 0;
5290 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5291 		    UIO_SYSSPACE;
5292 		uio.uio_fmode = 0;
5293 		uio.uio_extflg = UIO_COPY_CACHED;
5294 		uio.uio_resid = iov.iov_len;
5295 		if ((error = strmakemsg(&mctl,
5296 		    &msgsize, &uio, stp,
5297 		    STRUCT_FGET(strfdinsert, flags), &mp)) != 0 || !mp) {
5298 			STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5299 			releasef(STRUCT_FGET(strfdinsert, fildes));
5300 			return (error);
5301 		}
5302 
5303 		STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5304 
5305 		/*
5306 		 * Place the possibly reencoded queue pointer 'offset' bytes
5307 		 * from the start of the control portion of the message.
5308 		 */
5309 		*((t_uscalar_t *)(mp->b_rptr +
5310 		    STRUCT_FGET(strfdinsert, offset))) = ival;
5311 
5312 		/*
5313 		 * Put message downstream.
5314 		 */
5315 		stream_willservice(stp);
5316 		putnext(stp->sd_wrq, mp);
5317 		stream_runservice(stp);
5318 		releasef(STRUCT_FGET(strfdinsert, fildes));
5319 		return (error);
5320 	}
5321 
5322 	case I_SENDFD:
5323 	{
5324 		struct file *fp;
5325 
5326 		if ((fp = getf((int)arg)) == NULL)
5327 			return (EBADF);
5328 		error = do_sendfp(stp, fp, crp);
5329 		if (auditing) {
5330 			audit_fdsend((int)arg, fp, error);
5331 		}
5332 		releasef((int)arg);
5333 		return (error);
5334 	}
5335 
5336 	case I_RECVFD:
5337 	case I_E_RECVFD:
5338 	{
5339 		struct k_strrecvfd *srf;
5340 		int i, fd;
5341 
5342 		mutex_enter(&stp->sd_lock);
5343 		while (!(mp = getq(rdq))) {
5344 			if (stp->sd_flag & (STRHUP|STREOF)) {
5345 				mutex_exit(&stp->sd_lock);
5346 				return (ENXIO);
5347 			}
5348 			if ((error = strwaitq(stp, GETWAIT, (ssize_t)0,
5349 			    flag, -1, &done)) != 0 || done) {
5350 				mutex_exit(&stp->sd_lock);
5351 				return (error);
5352 			}
5353 			if ((error = i_straccess(stp, access)) != 0) {
5354 				mutex_exit(&stp->sd_lock);
5355 				return (error);
5356 			}
5357 		}
5358 		if (mp->b_datap->db_type != M_PASSFP) {
5359 			putback(stp, rdq, mp, mp->b_band);
5360 			mutex_exit(&stp->sd_lock);
5361 			return (EBADMSG);
5362 		}
5363 		mutex_exit(&stp->sd_lock);
5364 
5365 		srf = (struct k_strrecvfd *)mp->b_rptr;
5366 		if ((fd = ufalloc(0)) == -1) {
5367 			mutex_enter(&stp->sd_lock);
5368 			putback(stp, rdq, mp, mp->b_band);
5369 			mutex_exit(&stp->sd_lock);
5370 			return (EMFILE);
5371 		}
5372 		if (cmd == I_RECVFD) {
5373 			struct o_strrecvfd	ostrfd;
5374 
5375 			/* check to see if uid/gid values are too large. */
5376 
5377 			if (srf->uid > (o_uid_t)USHRT_MAX ||
5378 			    srf->gid > (o_gid_t)USHRT_MAX) {
5379 				mutex_enter(&stp->sd_lock);
5380 				putback(stp, rdq, mp, mp->b_band);
5381 				mutex_exit(&stp->sd_lock);
5382 				setf(fd, NULL);	/* release fd entry */
5383 				return (EOVERFLOW);
5384 			}
5385 
5386 			ostrfd.fd = fd;
5387 			ostrfd.uid = (o_uid_t)srf->uid;
5388 			ostrfd.gid = (o_gid_t)srf->gid;
5389 
5390 			/* Null the filler bits */
5391 			for (i = 0; i < 8; i++)
5392 				ostrfd.fill[i] = 0;
5393 
5394 			error = strcopyout(&ostrfd, (void *)arg,
5395 			    sizeof (struct o_strrecvfd), copyflag);
5396 		} else {		/* I_E_RECVFD */
5397 			struct strrecvfd	strfd;
5398 
5399 			strfd.fd = fd;
5400 			strfd.uid = srf->uid;
5401 			strfd.gid = srf->gid;
5402 
5403 			/* null the filler bits */
5404 			for (i = 0; i < 8; i++)
5405 				strfd.fill[i] = 0;
5406 
5407 			error = strcopyout(&strfd, (void *)arg,
5408 			    sizeof (struct strrecvfd), copyflag);
5409 		}
5410 
5411 		if (error) {
5412 			setf(fd, NULL);	/* release fd entry */
5413 			mutex_enter(&stp->sd_lock);
5414 			putback(stp, rdq, mp, mp->b_band);
5415 			mutex_exit(&stp->sd_lock);
5416 			return (error);
5417 		}
5418 		if (auditing) {
5419 			audit_fdrecv(fd, srf->fp);
5420 		}
5421 
5422 		/*
5423 		 * Always increment f_count since the freemsg() below will
5424 		 * always call free_passfp() which performs a closef().
5425 		 */
5426 		mutex_enter(&srf->fp->f_tlock);
5427 		srf->fp->f_count++;
5428 		mutex_exit(&srf->fp->f_tlock);
5429 		setf(fd, srf->fp);
5430 		freemsg(mp);
5431 		return (0);
5432 	}
5433 
5434 	case I_SWROPT:
5435 		/*
5436 		 * Set/clear the write options. arg is a bit
5437 		 * mask with any of the following bits set...
5438 		 *	SNDZERO - send zero length message
5439 		 *	SNDPIPE - send sigpipe to process if
5440 		 *		sd_werror is set and process is
5441 		 *		doing a write or putmsg.
5442 		 * The new stream head write options should reflect
5443 		 * what is in arg.
5444 		 */
5445 		if (arg & ~(SNDZERO|SNDPIPE))
5446 			return (EINVAL);
5447 
5448 		mutex_enter(&stp->sd_lock);
5449 		stp->sd_wput_opt &= ~(SW_SIGPIPE|SW_SNDZERO);
5450 		if (arg & SNDZERO)
5451 			stp->sd_wput_opt |= SW_SNDZERO;
5452 		if (arg & SNDPIPE)
5453 			stp->sd_wput_opt |= SW_SIGPIPE;
5454 		mutex_exit(&stp->sd_lock);
5455 		return (0);
5456 
5457 	case I_GWROPT:
5458 	{
5459 		int wropt = 0;
5460 
5461 		if (stp->sd_wput_opt & SW_SNDZERO)
5462 			wropt |= SNDZERO;
5463 		if (stp->sd_wput_opt & SW_SIGPIPE)
5464 			wropt |= SNDPIPE;
5465 		return (strcopyout(&wropt, (void *)arg, sizeof (wropt),
5466 		    copyflag));
5467 	}
5468 
5469 	case I_LIST:
5470 		/*
5471 		 * Returns all the modules found on this stream,
5472 		 * upto the driver. If argument is NULL, return the
5473 		 * number of modules (including driver). If argument
5474 		 * is not NULL, copy the names into the structure
5475 		 * provided.
5476 		 */
5477 
5478 	{
5479 		queue_t *q;
5480 		char *qname;
5481 		int i, nmods;
5482 		struct str_mlist *mlist;
5483 		STRUCT_DECL(str_list, strlist);
5484 
5485 		if (arg == 0) { /* Return number of modules plus driver */
5486 			if (stp->sd_vnode->v_type == VFIFO)
5487 				*rvalp = stp->sd_pushcnt;
5488 			else
5489 				*rvalp = stp->sd_pushcnt + 1;
5490 			return (0);
5491 		}
5492 
5493 		STRUCT_INIT(strlist, flag);
5494 
5495 		error = strcopyin((void *)arg, STRUCT_BUF(strlist),
5496 		    STRUCT_SIZE(strlist), copyflag);
5497 		if (error != 0)
5498 			return (error);
5499 
5500 		mlist = STRUCT_FGETP(strlist, sl_modlist);
5501 		nmods = STRUCT_FGET(strlist, sl_nmods);
5502 		if (nmods <= 0)
5503 			return (EINVAL);
5504 
5505 		claimstr(stp->sd_wrq);
5506 		q = stp->sd_wrq;
5507 		for (i = 0; i < nmods && _SAMESTR(q); i++, q = q->q_next) {
5508 			qname = Q2NAME(q->q_next);
5509 			error = strcopyout(qname, &mlist[i], strlen(qname) + 1,
5510 			    copyflag);
5511 			if (error != 0) {
5512 				releasestr(stp->sd_wrq);
5513 				return (error);
5514 			}
5515 		}
5516 		releasestr(stp->sd_wrq);
5517 		return (strcopyout(&i, (void *)arg, sizeof (int), copyflag));
5518 	}
5519 
5520 	case I_CKBAND:
5521 	{
5522 		queue_t *q;
5523 		qband_t *qbp;
5524 
5525 		if ((arg < 0) || (arg >= NBAND))
5526 			return (EINVAL);
5527 		q = _RD(stp->sd_wrq);
5528 		mutex_enter(QLOCK(q));
5529 		if (arg > (int)q->q_nband) {
5530 			*rvalp = 0;
5531 		} else {
5532 			if (arg == 0) {
5533 				if (q->q_first)
5534 					*rvalp = 1;
5535 				else
5536 					*rvalp = 0;
5537 			} else {
5538 				qbp = q->q_bandp;
5539 				while (--arg > 0)
5540 					qbp = qbp->qb_next;
5541 				if (qbp->qb_first)
5542 					*rvalp = 1;
5543 				else
5544 					*rvalp = 0;
5545 			}
5546 		}
5547 		mutex_exit(QLOCK(q));
5548 		return (0);
5549 	}
5550 
5551 	case I_GETBAND:
5552 	{
5553 		int intpri;
5554 		queue_t *q;
5555 
5556 		q = _RD(stp->sd_wrq);
5557 		mutex_enter(QLOCK(q));
5558 		mp = q->q_first;
5559 		if (!mp) {
5560 			mutex_exit(QLOCK(q));
5561 			return (ENODATA);
5562 		}
5563 		intpri = (int)mp->b_band;
5564 		error = strcopyout(&intpri, (void *)arg, sizeof (int),
5565 		    copyflag);
5566 		mutex_exit(QLOCK(q));
5567 		return (error);
5568 	}
5569 
5570 	case I_ATMARK:
5571 	{
5572 		queue_t *q;
5573 
5574 		if (arg & ~(ANYMARK|LASTMARK))
5575 			return (EINVAL);
5576 		q = _RD(stp->sd_wrq);
5577 		mutex_enter(&stp->sd_lock);
5578 		if ((stp->sd_flag & STRATMARK) && (arg == ANYMARK)) {
5579 			*rvalp = 1;
5580 		} else {
5581 			mutex_enter(QLOCK(q));
5582 			mp = q->q_first;
5583 
5584 			if (mp == NULL)
5585 				*rvalp = 0;
5586 			else if ((arg == ANYMARK) && (mp->b_flag & MSGMARK))
5587 				*rvalp = 1;
5588 			else if ((arg == LASTMARK) && (mp == stp->sd_mark))
5589 				*rvalp = 1;
5590 			else
5591 				*rvalp = 0;
5592 			mutex_exit(QLOCK(q));
5593 		}
5594 		mutex_exit(&stp->sd_lock);
5595 		return (0);
5596 	}
5597 
5598 	case I_CANPUT:
5599 	{
5600 		char band;
5601 
5602 		if ((arg < 0) || (arg >= NBAND))
5603 			return (EINVAL);
5604 		band = (char)arg;
5605 		*rvalp = bcanputnext(stp->sd_wrq, band);
5606 		return (0);
5607 	}
5608 
5609 	case I_SETCLTIME:
5610 	{
5611 		int closetime;
5612 
5613 		error = strcopyin((void *)arg, &closetime, sizeof (int),
5614 		    copyflag);
5615 		if (error)
5616 			return (error);
5617 		if (closetime < 0)
5618 			return (EINVAL);
5619 
5620 		stp->sd_closetime = closetime;
5621 		return (0);
5622 	}
5623 
5624 	case I_GETCLTIME:
5625 	{
5626 		int closetime;
5627 
5628 		closetime = stp->sd_closetime;
5629 		return (strcopyout(&closetime, (void *)arg, sizeof (int),
5630 		    copyflag));
5631 	}
5632 
5633 	case TIOCGSID:
5634 	{
5635 		pid_t sid;
5636 
5637 		mutex_enter(&stp->sd_lock);
5638 		if (stp->sd_sidp == NULL) {
5639 			mutex_exit(&stp->sd_lock);
5640 			return (ENOTTY);
5641 		}
5642 		sid = stp->sd_sidp->pid_id;
5643 		mutex_exit(&stp->sd_lock);
5644 		return (strcopyout(&sid, (void *)arg, sizeof (pid_t),
5645 		    copyflag));
5646 	}
5647 
5648 	case TIOCSPGRP:
5649 	{
5650 		pid_t pgrp;
5651 		proc_t *q;
5652 		pid_t	sid, fg_pgid, bg_pgid;
5653 
5654 		if (error = strcopyin((void *)arg, &pgrp, sizeof (pid_t),
5655 		    copyflag))
5656 			return (error);
5657 		mutex_enter(&stp->sd_lock);
5658 		mutex_enter(&pidlock);
5659 		if (stp->sd_sidp != ttoproc(curthread)->p_sessp->s_sidp) {
5660 			mutex_exit(&pidlock);
5661 			mutex_exit(&stp->sd_lock);
5662 			return (ENOTTY);
5663 		}
5664 		if (pgrp == stp->sd_pgidp->pid_id) {
5665 			mutex_exit(&pidlock);
5666 			mutex_exit(&stp->sd_lock);
5667 			return (0);
5668 		}
5669 		if (pgrp <= 0 || pgrp >= maxpid) {
5670 			mutex_exit(&pidlock);
5671 			mutex_exit(&stp->sd_lock);
5672 			return (EINVAL);
5673 		}
5674 		if ((q = pgfind(pgrp)) == NULL ||
5675 		    q->p_sessp != ttoproc(curthread)->p_sessp) {
5676 			mutex_exit(&pidlock);
5677 			mutex_exit(&stp->sd_lock);
5678 			return (EPERM);
5679 		}
5680 		sid = stp->sd_sidp->pid_id;
5681 		fg_pgid = q->p_pgrp;
5682 		bg_pgid = stp->sd_pgidp->pid_id;
5683 		CL_SET_PROCESS_GROUP(curthread, sid, bg_pgid, fg_pgid);
5684 		PID_RELE(stp->sd_pgidp);
5685 		ctty_clear_sighuped();
5686 		stp->sd_pgidp = q->p_pgidp;
5687 		PID_HOLD(stp->sd_pgidp);
5688 		mutex_exit(&pidlock);
5689 		mutex_exit(&stp->sd_lock);
5690 		return (0);
5691 	}
5692 
5693 	case TIOCGPGRP:
5694 	{
5695 		pid_t pgrp;
5696 
5697 		mutex_enter(&stp->sd_lock);
5698 		if (stp->sd_sidp == NULL) {
5699 			mutex_exit(&stp->sd_lock);
5700 			return (ENOTTY);
5701 		}
5702 		pgrp = stp->sd_pgidp->pid_id;
5703 		mutex_exit(&stp->sd_lock);
5704 		return (strcopyout(&pgrp, (void *)arg, sizeof (pid_t),
5705 		    copyflag));
5706 	}
5707 
5708 	case TIOCSCTTY:
5709 	{
5710 		return (strctty(stp));
5711 	}
5712 
5713 	case TIOCNOTTY:
5714 	{
5715 		/* freectty() always assumes curproc. */
5716 		if (freectty(B_FALSE) != 0)
5717 			return (0);
5718 		return (ENOTTY);
5719 	}
5720 
5721 	case FIONBIO:
5722 	case FIOASYNC:
5723 		return (0);	/* handled by the upper layer */
5724 	}
5725 }
5726 
5727 /*
5728  * Custom free routine used for M_PASSFP messages.
5729  */
5730 static void
5731 free_passfp(struct k_strrecvfd *srf)
5732 {
5733 	(void) closef(srf->fp);
5734 	kmem_free(srf, sizeof (struct k_strrecvfd) + sizeof (frtn_t));
5735 }
5736 
5737 /* ARGSUSED */
5738 int
5739 do_sendfp(struct stdata *stp, struct file *fp, struct cred *cr)
5740 {
5741 	queue_t *qp, *nextqp;
5742 	struct k_strrecvfd *srf;
5743 	mblk_t *mp;
5744 	frtn_t *frtnp;
5745 	size_t bufsize;
5746 	queue_t	*mate = NULL;
5747 	syncq_t	*sq = NULL;
5748 	int retval = 0;
5749 
5750 	if (stp->sd_flag & STRHUP)
5751 		return (ENXIO);
5752 
5753 	claimstr(stp->sd_wrq);
5754 
5755 	/* Fastpath, we have a pipe, and we are already mated, use it. */
5756 	if (STRMATED(stp)) {
5757 		qp = _RD(stp->sd_mate->sd_wrq);
5758 		claimstr(qp);
5759 		mate = qp;
5760 	} else { /* Not already mated. */
5761 
5762 		/*
5763 		 * Walk the stream to the end of this one.
5764 		 * assumes that the claimstr() will prevent
5765 		 * plumbing between the stream head and the
5766 		 * driver from changing
5767 		 */
5768 		qp = stp->sd_wrq;
5769 
5770 		/*
5771 		 * Loop until we reach the end of this stream.
5772 		 * On completion, qp points to the write queue
5773 		 * at the end of the stream, or the read queue
5774 		 * at the stream head if this is a fifo.
5775 		 */
5776 		while (((qp = qp->q_next) != NULL) && _SAMESTR(qp))
5777 			;
5778 
5779 		/*
5780 		 * Just in case we get a q_next which is NULL, but
5781 		 * not at the end of the stream.  This is actually
5782 		 * broken, so we set an assert to catch it in
5783 		 * debug, and set an error and return if not debug.
5784 		 */
5785 		ASSERT(qp);
5786 		if (qp == NULL) {
5787 			releasestr(stp->sd_wrq);
5788 			return (EINVAL);
5789 		}
5790 
5791 		/*
5792 		 * Enter the syncq for the driver, so (hopefully)
5793 		 * the queue values will not change on us.
5794 		 * XXXX - This will only prevent the race IFF only
5795 		 *   the write side modifies the q_next member, and
5796 		 *   the put procedure is protected by at least
5797 		 *   MT_PERQ.
5798 		 */
5799 		if ((sq = qp->q_syncq) != NULL)
5800 			entersq(sq, SQ_PUT);
5801 
5802 		/* Now get the q_next value from this qp. */
5803 		nextqp = qp->q_next;
5804 
5805 		/*
5806 		 * If nextqp exists and the other stream is different
5807 		 * from this one claim the stream, set the mate, and
5808 		 * get the read queue at the stream head of the other
5809 		 * stream.  Assumes that nextqp was at least valid when
5810 		 * we got it.  Hopefully the entersq of the driver
5811 		 * will prevent it from changing on us.
5812 		 */
5813 		if ((nextqp != NULL) && (STREAM(nextqp) != stp)) {
5814 			ASSERT(qp->q_qinfo->qi_srvp);
5815 			ASSERT(_OTHERQ(qp)->q_qinfo->qi_srvp);
5816 			ASSERT(_OTHERQ(qp->q_next)->q_qinfo->qi_srvp);
5817 			claimstr(nextqp);
5818 
5819 			/* Make sure we still have a q_next */
5820 			if (nextqp != qp->q_next) {
5821 				releasestr(stp->sd_wrq);
5822 				releasestr(nextqp);
5823 				return (EINVAL);
5824 			}
5825 
5826 			qp = _RD(STREAM(nextqp)->sd_wrq);
5827 			mate = qp;
5828 		}
5829 		/* If we entered the synq above, leave it. */
5830 		if (sq != NULL)
5831 			leavesq(sq, SQ_PUT);
5832 	} /*  STRMATED(STP)  */
5833 
5834 	/* XXX prevents substitution of the ops vector */
5835 	if (qp->q_qinfo != &strdata && qp->q_qinfo != &fifo_strdata) {
5836 		retval = EINVAL;
5837 		goto out;
5838 	}
5839 
5840 	if (qp->q_flag & QFULL) {
5841 		retval = EAGAIN;
5842 		goto out;
5843 	}
5844 
5845 	/*
5846 	 * Since M_PASSFP messages include a file descriptor, we use
5847 	 * esballoc() and specify a custom free routine (free_passfp()) that
5848 	 * will close the descriptor as part of freeing the message.  For
5849 	 * convenience, we stash the frtn_t right after the data block.
5850 	 */
5851 	bufsize = sizeof (struct k_strrecvfd) + sizeof (frtn_t);
5852 	srf = kmem_alloc(bufsize, KM_NOSLEEP);
5853 	if (srf == NULL) {
5854 		retval = EAGAIN;
5855 		goto out;
5856 	}
5857 
5858 	frtnp = (frtn_t *)(srf + 1);
5859 	frtnp->free_arg = (caddr_t)srf;
5860 	frtnp->free_func = free_passfp;
5861 
5862 	mp = esballoc((uchar_t *)srf, bufsize, BPRI_MED, frtnp);
5863 	if (mp == NULL) {
5864 		kmem_free(srf, bufsize);
5865 		retval = EAGAIN;
5866 		goto out;
5867 	}
5868 	mp->b_wptr += sizeof (struct k_strrecvfd);
5869 	mp->b_datap->db_type = M_PASSFP;
5870 
5871 	srf->fp = fp;
5872 	srf->uid = crgetuid(curthread->t_cred);
5873 	srf->gid = crgetgid(curthread->t_cred);
5874 	mutex_enter(&fp->f_tlock);
5875 	fp->f_count++;
5876 	mutex_exit(&fp->f_tlock);
5877 
5878 	put(qp, mp);
5879 out:
5880 	releasestr(stp->sd_wrq);
5881 	if (mate)
5882 		releasestr(mate);
5883 	return (retval);
5884 }
5885 
5886 /*
5887  * Send an ioctl message downstream and wait for acknowledgement.
5888  * flags may be set to either U_TO_K or K_TO_K and a combination
5889  * of STR_NOERROR or STR_NOSIG
5890  * STR_NOSIG: Signals are essentially ignored or held and have
5891  *	no effect for the duration of the call.
5892  * STR_NOERROR: Ignores stream head read, write and hup errors.
5893  *	Additionally, if an existing ioctl times out, it is assumed
5894  *	lost and and this ioctl will continue as if the previous ioctl had
5895  *	finished.  ETIME may be returned if this ioctl times out (i.e.
5896  *	ic_timout is not INFTIM).  Non-stream head errors may be returned if
5897  *	the ioc_error indicates that the driver/module had problems,
5898  *	an EFAULT was found when accessing user data, a lack of
5899  *	resources, etc.
5900  */
5901 int
5902 strdoioctl(
5903 	struct stdata *stp,
5904 	struct strioctl *strioc,
5905 	int fflags,		/* file flags with model info */
5906 	int flag,
5907 	cred_t *crp,
5908 	int *rvalp)
5909 {
5910 	mblk_t *bp;
5911 	struct iocblk *iocbp;
5912 	struct copyreq *reqp;
5913 	struct copyresp *resp;
5914 	int id;
5915 	int transparent = 0;
5916 	int error = 0;
5917 	int len = 0;
5918 	caddr_t taddr;
5919 	int copyflag = (flag & (U_TO_K | K_TO_K));
5920 	int sigflag = (flag & STR_NOSIG);
5921 	int errs;
5922 	uint_t waitflags;
5923 	boolean_t set_iocwaitne = B_FALSE;
5924 
5925 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
5926 	ASSERT((fflags & FMODELS) != 0);
5927 
5928 	TRACE_2(TR_FAC_STREAMS_FR,
5929 	    TR_STRDOIOCTL,
5930 	    "strdoioctl:stp %p strioc %p", stp, strioc);
5931 	if (strioc->ic_len == TRANSPARENT) {	/* send arg in M_DATA block */
5932 		transparent = 1;
5933 		strioc->ic_len = sizeof (intptr_t);
5934 	}
5935 
5936 	if (strioc->ic_len < 0 || (strmsgsz > 0 && strioc->ic_len > strmsgsz))
5937 		return (EINVAL);
5938 
5939 	if ((bp = allocb_cred_wait(sizeof (union ioctypes), sigflag, &error,
5940 	    crp, curproc->p_pid)) == NULL)
5941 			return (error);
5942 
5943 	bzero(bp->b_wptr, sizeof (union ioctypes));
5944 
5945 	iocbp = (struct iocblk *)bp->b_wptr;
5946 	iocbp->ioc_count = strioc->ic_len;
5947 	iocbp->ioc_cmd = strioc->ic_cmd;
5948 	iocbp->ioc_flag = (fflags & FMODELS);
5949 
5950 	crhold(crp);
5951 	iocbp->ioc_cr = crp;
5952 	DB_TYPE(bp) = M_IOCTL;
5953 	bp->b_wptr += sizeof (struct iocblk);
5954 
5955 	if (flag & STR_NOERROR)
5956 		errs = STPLEX;
5957 	else
5958 		errs = STRHUP|STRDERR|STWRERR|STPLEX;
5959 
5960 	/*
5961 	 * If there is data to copy into ioctl block, do so.
5962 	 */
5963 	if (iocbp->ioc_count > 0) {
5964 		if (transparent)
5965 			/*
5966 			 * Note: STR_NOERROR does not have an effect
5967 			 * in putiocd()
5968 			 */
5969 			id = K_TO_K | sigflag;
5970 		else
5971 			id = flag;
5972 		if ((error = putiocd(bp, strioc->ic_dp, id, crp)) != 0) {
5973 			freemsg(bp);
5974 			crfree(crp);
5975 			return (error);
5976 		}
5977 
5978 		/*
5979 		 * We could have slept copying in user pages.
5980 		 * Recheck the stream head state (the other end
5981 		 * of a pipe could have gone away).
5982 		 */
5983 		if (stp->sd_flag & errs) {
5984 			mutex_enter(&stp->sd_lock);
5985 			error = strgeterr(stp, errs, 0);
5986 			mutex_exit(&stp->sd_lock);
5987 			if (error != 0) {
5988 				freemsg(bp);
5989 				crfree(crp);
5990 				return (error);
5991 			}
5992 		}
5993 	}
5994 	if (transparent)
5995 		iocbp->ioc_count = TRANSPARENT;
5996 
5997 	/*
5998 	 * Block for up to STRTIMOUT milliseconds if there is an outstanding
5999 	 * ioctl for this stream already running.  All processes
6000 	 * sleeping here will be awakened as a result of an ACK
6001 	 * or NAK being received for the outstanding ioctl, or
6002 	 * as a result of the timer expiring on the outstanding
6003 	 * ioctl (a failure), or as a result of any waiting
6004 	 * process's timer expiring (also a failure).
6005 	 */
6006 
6007 	error = 0;
6008 	mutex_enter(&stp->sd_lock);
6009 	while ((stp->sd_flag & IOCWAIT) ||
6010 	    (!set_iocwaitne && (stp->sd_flag & IOCWAITNE))) {
6011 		clock_t cv_rval;
6012 
6013 		TRACE_0(TR_FAC_STREAMS_FR,
6014 		    TR_STRDOIOCTL_WAIT,
6015 		    "strdoioctl sleeps - IOCWAIT");
6016 		cv_rval = str_cv_wait(&stp->sd_iocmonitor, &stp->sd_lock,
6017 		    STRTIMOUT, sigflag);
6018 		if (cv_rval <= 0) {
6019 			if (cv_rval == 0) {
6020 				error = EINTR;
6021 			} else {
6022 				if (flag & STR_NOERROR) {
6023 					/*
6024 					 * Terminating current ioctl in
6025 					 * progress -- assume it got lost and
6026 					 * wake up the other thread so that the
6027 					 * operation completes.
6028 					 */
6029 					if (!(stp->sd_flag & IOCWAITNE)) {
6030 						set_iocwaitne = B_TRUE;
6031 						stp->sd_flag |= IOCWAITNE;
6032 						cv_broadcast(&stp->sd_monitor);
6033 					}
6034 					/*
6035 					 * Otherwise, there's a running
6036 					 * STR_NOERROR -- we have no choice
6037 					 * here but to wait forever (or until
6038 					 * interrupted).
6039 					 */
6040 				} else {
6041 					/*
6042 					 * pending ioctl has caused
6043 					 * us to time out
6044 					 */
6045 					error = ETIME;
6046 				}
6047 			}
6048 		} else if ((stp->sd_flag & errs)) {
6049 			error = strgeterr(stp, errs, 0);
6050 		}
6051 		if (error) {
6052 			mutex_exit(&stp->sd_lock);
6053 			freemsg(bp);
6054 			crfree(crp);
6055 			return (error);
6056 		}
6057 	}
6058 
6059 	/*
6060 	 * Have control of ioctl mechanism.
6061 	 * Send down ioctl packet and wait for response.
6062 	 */
6063 	if (stp->sd_iocblk != (mblk_t *)-1) {
6064 		freemsg(stp->sd_iocblk);
6065 	}
6066 	stp->sd_iocblk = NULL;
6067 
6068 	/*
6069 	 * If this is marked with 'noerror' (internal; mostly
6070 	 * I_{P,}{UN,}LINK), then make sure nobody else is able to get
6071 	 * in here by setting IOCWAITNE.
6072 	 */
6073 	waitflags = IOCWAIT;
6074 	if (flag & STR_NOERROR)
6075 		waitflags |= IOCWAITNE;
6076 
6077 	stp->sd_flag |= waitflags;
6078 
6079 	/*
6080 	 * Assign sequence number.
6081 	 */
6082 	iocbp->ioc_id = stp->sd_iocid = getiocseqno();
6083 
6084 	mutex_exit(&stp->sd_lock);
6085 
6086 	TRACE_1(TR_FAC_STREAMS_FR,
6087 	    TR_STRDOIOCTL_PUT, "strdoioctl put: stp %p", stp);
6088 	stream_willservice(stp);
6089 	putnext(stp->sd_wrq, bp);
6090 	stream_runservice(stp);
6091 
6092 	/*
6093 	 * Timed wait for acknowledgment.  The wait time is limited by the
6094 	 * timeout value, which must be a positive integer (number of
6095 	 * milliseconds) to wait, or 0 (use default value of STRTIMOUT
6096 	 * milliseconds), or -1 (wait forever).  This will be awakened
6097 	 * either by an ACK/NAK message arriving, the timer expiring, or
6098 	 * the timer expiring on another ioctl waiting for control of the
6099 	 * mechanism.
6100 	 */
6101 waitioc:
6102 	mutex_enter(&stp->sd_lock);
6103 
6104 
6105 	/*
6106 	 * If the reply has already arrived, don't sleep.  If awakened from
6107 	 * the sleep, fail only if the reply has not arrived by then.
6108 	 * Otherwise, process the reply.
6109 	 */
6110 	while (!stp->sd_iocblk) {
6111 		clock_t cv_rval;
6112 
6113 		if (stp->sd_flag & errs) {
6114 			error = strgeterr(stp, errs, 0);
6115 			if (error != 0) {
6116 				stp->sd_flag &= ~waitflags;
6117 				cv_broadcast(&stp->sd_iocmonitor);
6118 				mutex_exit(&stp->sd_lock);
6119 				crfree(crp);
6120 				return (error);
6121 			}
6122 		}
6123 
6124 		TRACE_0(TR_FAC_STREAMS_FR,
6125 		    TR_STRDOIOCTL_WAIT2,
6126 		    "strdoioctl sleeps awaiting reply");
6127 		ASSERT(error == 0);
6128 
6129 		cv_rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock,
6130 		    (strioc->ic_timout ?
6131 		    strioc->ic_timout * 1000 : STRTIMOUT), sigflag);
6132 
6133 		/*
6134 		 * There are four possible cases here: interrupt, timeout,
6135 		 * wakeup by IOCWAITNE (above), or wakeup by strrput_nondata (a
6136 		 * valid M_IOCTL reply).
6137 		 *
6138 		 * If we've been awakened by a STR_NOERROR ioctl on some other
6139 		 * thread, then sd_iocblk will still be NULL, and IOCWAITNE
6140 		 * will be set.  Pretend as if we just timed out.  Note that
6141 		 * this other thread waited at least STRTIMOUT before trying to
6142 		 * awaken our thread, so this is indistinguishable (even for
6143 		 * INFTIM) from the case where we failed with ETIME waiting on
6144 		 * IOCWAIT in the prior loop.
6145 		 */
6146 		if (cv_rval > 0 && !(flag & STR_NOERROR) &&
6147 		    stp->sd_iocblk == NULL && (stp->sd_flag & IOCWAITNE)) {
6148 			cv_rval = -1;
6149 		}
6150 
6151 		/*
6152 		 * note: STR_NOERROR does not protect
6153 		 * us here.. use ic_timout < 0
6154 		 */
6155 		if (cv_rval <= 0) {
6156 			if (cv_rval == 0) {
6157 				error = EINTR;
6158 			} else {
6159 				error =  ETIME;
6160 			}
6161 			/*
6162 			 * A message could have come in after we were scheduled
6163 			 * but before we were actually run.
6164 			 */
6165 			bp = stp->sd_iocblk;
6166 			stp->sd_iocblk = NULL;
6167 			if (bp != NULL) {
6168 				if ((bp->b_datap->db_type == M_COPYIN) ||
6169 				    (bp->b_datap->db_type == M_COPYOUT)) {
6170 					mutex_exit(&stp->sd_lock);
6171 					if (bp->b_cont) {
6172 						freemsg(bp->b_cont);
6173 						bp->b_cont = NULL;
6174 					}
6175 					bp->b_datap->db_type = M_IOCDATA;
6176 					bp->b_wptr = bp->b_rptr +
6177 					    sizeof (struct copyresp);
6178 					resp = (struct copyresp *)bp->b_rptr;
6179 					resp->cp_rval =
6180 					    (caddr_t)1; /* failure */
6181 					stream_willservice(stp);
6182 					putnext(stp->sd_wrq, bp);
6183 					stream_runservice(stp);
6184 					mutex_enter(&stp->sd_lock);
6185 				} else {
6186 					freemsg(bp);
6187 				}
6188 			}
6189 			stp->sd_flag &= ~waitflags;
6190 			cv_broadcast(&stp->sd_iocmonitor);
6191 			mutex_exit(&stp->sd_lock);
6192 			crfree(crp);
6193 			return (error);
6194 		}
6195 	}
6196 	bp = stp->sd_iocblk;
6197 	/*
6198 	 * Note: it is strictly impossible to get here with sd_iocblk set to
6199 	 * -1.  This is because the initial loop above doesn't allow any new
6200 	 * ioctls into the fray until all others have passed this point.
6201 	 */
6202 	ASSERT(bp != NULL && bp != (mblk_t *)-1);
6203 	TRACE_1(TR_FAC_STREAMS_FR,
6204 	    TR_STRDOIOCTL_ACK, "strdoioctl got reply: bp %p", bp);
6205 	if ((bp->b_datap->db_type == M_IOCACK) ||
6206 	    (bp->b_datap->db_type == M_IOCNAK)) {
6207 		/* for detection of duplicate ioctl replies */
6208 		stp->sd_iocblk = (mblk_t *)-1;
6209 		stp->sd_flag &= ~waitflags;
6210 		cv_broadcast(&stp->sd_iocmonitor);
6211 		mutex_exit(&stp->sd_lock);
6212 	} else {
6213 		/*
6214 		 * flags not cleared here because we're still doing
6215 		 * copy in/out for ioctl.
6216 		 */
6217 		stp->sd_iocblk = NULL;
6218 		mutex_exit(&stp->sd_lock);
6219 	}
6220 
6221 
6222 	/*
6223 	 * Have received acknowledgment.
6224 	 */
6225 
6226 	switch (bp->b_datap->db_type) {
6227 	case M_IOCACK:
6228 		/*
6229 		 * Positive ack.
6230 		 */
6231 		iocbp = (struct iocblk *)bp->b_rptr;
6232 
6233 		/*
6234 		 * Set error if indicated.
6235 		 */
6236 		if (iocbp->ioc_error) {
6237 			error = iocbp->ioc_error;
6238 			break;
6239 		}
6240 
6241 		/*
6242 		 * Set return value.
6243 		 */
6244 		*rvalp = iocbp->ioc_rval;
6245 
6246 		/*
6247 		 * Data may have been returned in ACK message (ioc_count > 0).
6248 		 * If so, copy it out to the user's buffer.
6249 		 */
6250 		if (iocbp->ioc_count && !transparent) {
6251 			if (error = getiocd(bp, strioc->ic_dp, copyflag))
6252 				break;
6253 		}
6254 		if (!transparent) {
6255 			if (len)	/* an M_COPYOUT was used with I_STR */
6256 				strioc->ic_len = len;
6257 			else
6258 				strioc->ic_len = (int)iocbp->ioc_count;
6259 		}
6260 		break;
6261 
6262 	case M_IOCNAK:
6263 		/*
6264 		 * Negative ack.
6265 		 *
6266 		 * The only thing to do is set error as specified
6267 		 * in neg ack packet.
6268 		 */
6269 		iocbp = (struct iocblk *)bp->b_rptr;
6270 
6271 		error = (iocbp->ioc_error ? iocbp->ioc_error : EINVAL);
6272 		break;
6273 
6274 	case M_COPYIN:
6275 		/*
6276 		 * Driver or module has requested user ioctl data.
6277 		 */
6278 		reqp = (struct copyreq *)bp->b_rptr;
6279 
6280 		/*
6281 		 * M_COPYIN should *never* have a message attached, though
6282 		 * it's harmless if it does -- thus, panic on a DEBUG
6283 		 * kernel and just free it on a non-DEBUG build.
6284 		 */
6285 		ASSERT(bp->b_cont == NULL);
6286 		if (bp->b_cont != NULL) {
6287 			freemsg(bp->b_cont);
6288 			bp->b_cont = NULL;
6289 		}
6290 
6291 		error = putiocd(bp, reqp->cq_addr, flag, crp);
6292 		if (error && bp->b_cont) {
6293 			freemsg(bp->b_cont);
6294 			bp->b_cont = NULL;
6295 		}
6296 
6297 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6298 		bp->b_datap->db_type = M_IOCDATA;
6299 
6300 		mblk_setcred(bp, crp, curproc->p_pid);
6301 		resp = (struct copyresp *)bp->b_rptr;
6302 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6303 		resp->cp_flag = (fflags & FMODELS);
6304 
6305 		stream_willservice(stp);
6306 		putnext(stp->sd_wrq, bp);
6307 		stream_runservice(stp);
6308 
6309 		if (error) {
6310 			mutex_enter(&stp->sd_lock);
6311 			stp->sd_flag &= ~waitflags;
6312 			cv_broadcast(&stp->sd_iocmonitor);
6313 			mutex_exit(&stp->sd_lock);
6314 			crfree(crp);
6315 			return (error);
6316 		}
6317 
6318 		goto waitioc;
6319 
6320 	case M_COPYOUT:
6321 		/*
6322 		 * Driver or module has ioctl data for a user.
6323 		 */
6324 		reqp = (struct copyreq *)bp->b_rptr;
6325 		ASSERT(bp->b_cont != NULL);
6326 
6327 		/*
6328 		 * Always (transparent or non-transparent )
6329 		 * use the address specified in the request
6330 		 */
6331 		taddr = reqp->cq_addr;
6332 		if (!transparent)
6333 			len = (int)reqp->cq_size;
6334 
6335 		/* copyout data to the provided address */
6336 		error = getiocd(bp, taddr, copyflag);
6337 
6338 		freemsg(bp->b_cont);
6339 		bp->b_cont = NULL;
6340 
6341 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6342 		bp->b_datap->db_type = M_IOCDATA;
6343 
6344 		mblk_setcred(bp, crp, curproc->p_pid);
6345 		resp = (struct copyresp *)bp->b_rptr;
6346 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6347 		resp->cp_flag = (fflags & FMODELS);
6348 
6349 		stream_willservice(stp);
6350 		putnext(stp->sd_wrq, bp);
6351 		stream_runservice(stp);
6352 
6353 		if (error) {
6354 			mutex_enter(&stp->sd_lock);
6355 			stp->sd_flag &= ~waitflags;
6356 			cv_broadcast(&stp->sd_iocmonitor);
6357 			mutex_exit(&stp->sd_lock);
6358 			crfree(crp);
6359 			return (error);
6360 		}
6361 		goto waitioc;
6362 
6363 	default:
6364 		ASSERT(0);
6365 		mutex_enter(&stp->sd_lock);
6366 		stp->sd_flag &= ~waitflags;
6367 		cv_broadcast(&stp->sd_iocmonitor);
6368 		mutex_exit(&stp->sd_lock);
6369 		break;
6370 	}
6371 
6372 	freemsg(bp);
6373 	crfree(crp);
6374 	return (error);
6375 }
6376 
6377 /*
6378  * Send an M_CMD message downstream and wait for a reply.  This is a ptools
6379  * special used to retrieve information from modules/drivers a stream without
6380  * being subjected to flow control or interfering with pending messages on the
6381  * stream (e.g. an ioctl in flight).
6382  */
6383 int
6384 strdocmd(struct stdata *stp, struct strcmd *scp, cred_t *crp)
6385 {
6386 	mblk_t *mp;
6387 	struct cmdblk *cmdp;
6388 	int error = 0;
6389 	int errs = STRHUP|STRDERR|STWRERR|STPLEX;
6390 	clock_t rval, timeout = STRTIMOUT;
6391 
6392 	if (scp->sc_len < 0 || scp->sc_len > sizeof (scp->sc_buf) ||
6393 	    scp->sc_timeout < -1)
6394 		return (EINVAL);
6395 
6396 	if (scp->sc_timeout > 0)
6397 		timeout = scp->sc_timeout * MILLISEC;
6398 
6399 	if ((mp = allocb_cred(sizeof (struct cmdblk), crp,
6400 	    curproc->p_pid)) == NULL)
6401 		return (ENOMEM);
6402 
6403 	crhold(crp);
6404 
6405 	cmdp = (struct cmdblk *)mp->b_wptr;
6406 	cmdp->cb_cr = crp;
6407 	cmdp->cb_cmd = scp->sc_cmd;
6408 	cmdp->cb_len = scp->sc_len;
6409 	cmdp->cb_error = 0;
6410 	mp->b_wptr += sizeof (struct cmdblk);
6411 
6412 	DB_TYPE(mp) = M_CMD;
6413 	DB_CPID(mp) = curproc->p_pid;
6414 
6415 	/*
6416 	 * Copy in the payload.
6417 	 */
6418 	if (cmdp->cb_len > 0) {
6419 		mp->b_cont = allocb_cred(sizeof (scp->sc_buf), crp,
6420 		    curproc->p_pid);
6421 		if (mp->b_cont == NULL) {
6422 			error = ENOMEM;
6423 			goto out;
6424 		}
6425 
6426 		/* cb_len comes from sc_len, which has already been checked */
6427 		ASSERT(cmdp->cb_len <= sizeof (scp->sc_buf));
6428 		(void) bcopy(scp->sc_buf, mp->b_cont->b_wptr, cmdp->cb_len);
6429 		mp->b_cont->b_wptr += cmdp->cb_len;
6430 		DB_CPID(mp->b_cont) = curproc->p_pid;
6431 	}
6432 
6433 	/*
6434 	 * Since this mechanism is strictly for ptools, and since only one
6435 	 * process can be grabbed at a time, we simply fail if there's
6436 	 * currently an operation pending.
6437 	 */
6438 	mutex_enter(&stp->sd_lock);
6439 	if (stp->sd_flag & STRCMDWAIT) {
6440 		mutex_exit(&stp->sd_lock);
6441 		error = EBUSY;
6442 		goto out;
6443 	}
6444 	stp->sd_flag |= STRCMDWAIT;
6445 	ASSERT(stp->sd_cmdblk == NULL);
6446 	mutex_exit(&stp->sd_lock);
6447 
6448 	putnext(stp->sd_wrq, mp);
6449 	mp = NULL;
6450 
6451 	/*
6452 	 * Timed wait for acknowledgment.  If the reply has already arrived,
6453 	 * don't sleep.  If awakened from the sleep, fail only if the reply
6454 	 * has not arrived by then.  Otherwise, process the reply.
6455 	 */
6456 	mutex_enter(&stp->sd_lock);
6457 	while (stp->sd_cmdblk == NULL) {
6458 		if (stp->sd_flag & errs) {
6459 			if ((error = strgeterr(stp, errs, 0)) != 0)
6460 				goto waitout;
6461 		}
6462 
6463 		rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock, timeout, 0);
6464 		if (stp->sd_cmdblk != NULL)
6465 			break;
6466 
6467 		if (rval <= 0) {
6468 			error = (rval == 0) ? EINTR : ETIME;
6469 			goto waitout;
6470 		}
6471 	}
6472 
6473 	/*
6474 	 * We received a reply.
6475 	 */
6476 	mp = stp->sd_cmdblk;
6477 	stp->sd_cmdblk = NULL;
6478 	ASSERT(mp != NULL && DB_TYPE(mp) == M_CMD);
6479 	ASSERT(stp->sd_flag & STRCMDWAIT);
6480 	stp->sd_flag &= ~STRCMDWAIT;
6481 	mutex_exit(&stp->sd_lock);
6482 
6483 	cmdp = (struct cmdblk *)mp->b_rptr;
6484 	if ((error = cmdp->cb_error) != 0)
6485 		goto out;
6486 
6487 	/*
6488 	 * Data may have been returned in the reply (cb_len > 0).
6489 	 * If so, copy it out to the user's buffer.
6490 	 */
6491 	if (cmdp->cb_len > 0) {
6492 		if (mp->b_cont == NULL || MBLKL(mp->b_cont) < cmdp->cb_len) {
6493 			error = EPROTO;
6494 			goto out;
6495 		}
6496 
6497 		cmdp->cb_len = MIN(cmdp->cb_len, sizeof (scp->sc_buf));
6498 		(void) bcopy(mp->b_cont->b_rptr, scp->sc_buf, cmdp->cb_len);
6499 	}
6500 	scp->sc_len = cmdp->cb_len;
6501 out:
6502 	freemsg(mp);
6503 	crfree(crp);
6504 	return (error);
6505 waitout:
6506 	ASSERT(stp->sd_cmdblk == NULL);
6507 	stp->sd_flag &= ~STRCMDWAIT;
6508 	mutex_exit(&stp->sd_lock);
6509 	crfree(crp);
6510 	return (error);
6511 }
6512 
6513 /*
6514  * For the SunOS keyboard driver.
6515  * Return the next available "ioctl" sequence number.
6516  * Exported, so that streams modules can send "ioctl" messages
6517  * downstream from their open routine.
6518  */
6519 int
6520 getiocseqno(void)
6521 {
6522 	int	i;
6523 
6524 	mutex_enter(&strresources);
6525 	i = ++ioc_id;
6526 	mutex_exit(&strresources);
6527 	return (i);
6528 }
6529 
6530 /*
6531  * Get the next message from the read queue.  If the message is
6532  * priority, STRPRI will have been set by strrput().  This flag
6533  * should be reset only when the entire message at the front of the
6534  * queue as been consumed.
6535  *
6536  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
6537  */
6538 int
6539 strgetmsg(
6540 	struct vnode *vp,
6541 	struct strbuf *mctl,
6542 	struct strbuf *mdata,
6543 	unsigned char *prip,
6544 	int *flagsp,
6545 	int fmode,
6546 	rval_t *rvp)
6547 {
6548 	struct stdata *stp;
6549 	mblk_t *bp, *nbp;
6550 	mblk_t *savemp = NULL;
6551 	mblk_t *savemptail = NULL;
6552 	uint_t old_sd_flag;
6553 	int flg;
6554 	int more = 0;
6555 	int error = 0;
6556 	char first = 1;
6557 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
6558 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
6559 	unsigned char pri = 0;
6560 	queue_t *q;
6561 	int	pr = 0;			/* Partial read successful */
6562 	struct uio uios;
6563 	struct uio *uiop = &uios;
6564 	struct iovec iovs;
6565 	unsigned char type;
6566 
6567 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRGETMSG_ENTER,
6568 	    "strgetmsg:%p", vp);
6569 
6570 	ASSERT(vp->v_stream);
6571 	stp = vp->v_stream;
6572 	rvp->r_val1 = 0;
6573 
6574 	mutex_enter(&stp->sd_lock);
6575 
6576 	if ((error = i_straccess(stp, JCREAD)) != 0) {
6577 		mutex_exit(&stp->sd_lock);
6578 		return (error);
6579 	}
6580 
6581 	if (stp->sd_flag & (STRDERR|STPLEX)) {
6582 		error = strgeterr(stp, STRDERR|STPLEX, 0);
6583 		if (error != 0) {
6584 			mutex_exit(&stp->sd_lock);
6585 			return (error);
6586 		}
6587 	}
6588 	mutex_exit(&stp->sd_lock);
6589 
6590 	switch (*flagsp) {
6591 	case MSG_HIPRI:
6592 		if (*prip != 0)
6593 			return (EINVAL);
6594 		break;
6595 
6596 	case MSG_ANY:
6597 	case MSG_BAND:
6598 		break;
6599 
6600 	default:
6601 		return (EINVAL);
6602 	}
6603 	/*
6604 	 * Setup uio and iov for data part
6605 	 */
6606 	iovs.iov_base = mdata->buf;
6607 	iovs.iov_len = mdata->maxlen;
6608 	uios.uio_iov = &iovs;
6609 	uios.uio_iovcnt = 1;
6610 	uios.uio_loffset = 0;
6611 	uios.uio_segflg = UIO_USERSPACE;
6612 	uios.uio_fmode = 0;
6613 	uios.uio_extflg = UIO_COPY_CACHED;
6614 	uios.uio_resid = mdata->maxlen;
6615 	uios.uio_offset = 0;
6616 
6617 	q = _RD(stp->sd_wrq);
6618 	mutex_enter(&stp->sd_lock);
6619 	old_sd_flag = stp->sd_flag;
6620 	mark = 0;
6621 	for (;;) {
6622 		int done = 0;
6623 		mblk_t *q_first = q->q_first;
6624 
6625 		/*
6626 		 * Get the next message of appropriate priority
6627 		 * from the stream head.  If the caller is interested
6628 		 * in band or hipri messages, then they should already
6629 		 * be enqueued at the stream head.  On the other hand
6630 		 * if the caller wants normal (band 0) messages, they
6631 		 * might be deferred in a synchronous stream and they
6632 		 * will need to be pulled up.
6633 		 *
6634 		 * After we have dequeued a message, we might find that
6635 		 * it was a deferred M_SIG that was enqueued at the
6636 		 * stream head.  It must now be posted as part of the
6637 		 * read by calling strsignal_nolock().
6638 		 *
6639 		 * Also note that strrput does not enqueue an M_PCSIG,
6640 		 * and there cannot be more than one hipri message,
6641 		 * so there was no need to have the M_PCSIG case.
6642 		 *
6643 		 * At some time it might be nice to try and wrap the
6644 		 * functionality of kstrgetmsg() and strgetmsg() into
6645 		 * a common routine so to reduce the amount of replicated
6646 		 * code (since they are extremely similar).
6647 		 */
6648 		if (!(*flagsp & (MSG_HIPRI|MSG_BAND))) {
6649 			/* Asking for normal, band0 data */
6650 			bp = strget(stp, q, uiop, first, &error);
6651 			ASSERT(MUTEX_HELD(&stp->sd_lock));
6652 			if (bp != NULL) {
6653 				if (DB_TYPE(bp) == M_SIG) {
6654 					strsignal_nolock(stp, *bp->b_rptr,
6655 					    bp->b_band);
6656 					freemsg(bp);
6657 					continue;
6658 				} else {
6659 					break;
6660 				}
6661 			}
6662 			if (error != 0)
6663 				goto getmout;
6664 
6665 		/*
6666 		 * We can't depend on the value of STRPRI here because
6667 		 * the stream head may be in transit. Therefore, we
6668 		 * must look at the type of the first message to
6669 		 * determine if a high priority messages is waiting
6670 		 */
6671 		} else if ((*flagsp & MSG_HIPRI) && q_first != NULL &&
6672 		    DB_TYPE(q_first) >= QPCTL &&
6673 		    (bp = getq_noenab(q, 0)) != NULL) {
6674 			/* Asked for HIPRI and got one */
6675 			ASSERT(DB_TYPE(bp) >= QPCTL);
6676 			break;
6677 		} else if ((*flagsp & MSG_BAND) && q_first != NULL &&
6678 		    ((q_first->b_band >= *prip) || DB_TYPE(q_first) >= QPCTL) &&
6679 		    (bp = getq_noenab(q, 0)) != NULL) {
6680 			/*
6681 			 * Asked for at least band "prip" and got either at
6682 			 * least that band or a hipri message.
6683 			 */
6684 			ASSERT(bp->b_band >= *prip || DB_TYPE(bp) >= QPCTL);
6685 			if (DB_TYPE(bp) == M_SIG) {
6686 				strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
6687 				freemsg(bp);
6688 				continue;
6689 			} else {
6690 				break;
6691 			}
6692 		}
6693 
6694 		/* No data. Time to sleep? */
6695 		qbackenable(q, 0);
6696 
6697 		/*
6698 		 * If STRHUP or STREOF, return 0 length control and data.
6699 		 * If resid is 0, then a read(fd,buf,0) was done. Do not
6700 		 * sleep to satisfy this request because by default we have
6701 		 * zero bytes to return.
6702 		 */
6703 		if ((stp->sd_flag & (STRHUP|STREOF)) || (mctl->maxlen == 0 &&
6704 		    mdata->maxlen == 0)) {
6705 			mctl->len = mdata->len = 0;
6706 			*flagsp = 0;
6707 			mutex_exit(&stp->sd_lock);
6708 			return (0);
6709 		}
6710 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_WAIT,
6711 		    "strgetmsg calls strwaitq:%p, %p",
6712 		    vp, uiop);
6713 		if (((error = strwaitq(stp, GETWAIT, (ssize_t)0, fmode, -1,
6714 		    &done)) != 0) || done) {
6715 			TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_DONE,
6716 			    "strgetmsg error or done:%p, %p",
6717 			    vp, uiop);
6718 			mutex_exit(&stp->sd_lock);
6719 			return (error);
6720 		}
6721 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_AWAKE,
6722 		    "strgetmsg awakes:%p, %p", vp, uiop);
6723 		if ((error = i_straccess(stp, JCREAD)) != 0) {
6724 			mutex_exit(&stp->sd_lock);
6725 			return (error);
6726 		}
6727 		first = 0;
6728 	}
6729 	ASSERT(bp != NULL);
6730 	/*
6731 	 * Extract any mark information. If the message is not completely
6732 	 * consumed this information will be put in the mblk
6733 	 * that is putback.
6734 	 * If MSGMARKNEXT is set and the message is completely consumed
6735 	 * the STRATMARK flag will be set below. Likewise, if
6736 	 * MSGNOTMARKNEXT is set and the message is
6737 	 * completely consumed STRNOTATMARK will be set.
6738 	 */
6739 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
6740 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
6741 	    (MSGMARKNEXT|MSGNOTMARKNEXT));
6742 	if (mark != 0 && bp == stp->sd_mark) {
6743 		mark |= _LASTMARK;
6744 		stp->sd_mark = NULL;
6745 	}
6746 	/*
6747 	 * keep track of the original message type and priority
6748 	 */
6749 	pri = bp->b_band;
6750 	type = bp->b_datap->db_type;
6751 	if (type == M_PASSFP) {
6752 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
6753 			stp->sd_mark = bp;
6754 		bp->b_flag |= mark & ~_LASTMARK;
6755 		putback(stp, q, bp, pri);
6756 		qbackenable(q, pri);
6757 		mutex_exit(&stp->sd_lock);
6758 		return (EBADMSG);
6759 	}
6760 	ASSERT(type != M_SIG);
6761 
6762 	/*
6763 	 * Set this flag so strrput will not generate signals. Need to
6764 	 * make sure this flag is cleared before leaving this routine
6765 	 * else signals will stop being sent.
6766 	 */
6767 	stp->sd_flag |= STRGETINPROG;
6768 	mutex_exit(&stp->sd_lock);
6769 
6770 	if (STREAM_NEEDSERVICE(stp))
6771 		stream_runservice(stp);
6772 
6773 	/*
6774 	 * Set HIPRI flag if message is priority.
6775 	 */
6776 	if (type >= QPCTL)
6777 		flg = MSG_HIPRI;
6778 	else
6779 		flg = MSG_BAND;
6780 
6781 	/*
6782 	 * First process PROTO or PCPROTO blocks, if any.
6783 	 */
6784 	if (mctl->maxlen >= 0 && type != M_DATA) {
6785 		size_t	n, bcnt;
6786 		char	*ubuf;
6787 
6788 		bcnt = mctl->maxlen;
6789 		ubuf = mctl->buf;
6790 		while (bp != NULL && bp->b_datap->db_type != M_DATA) {
6791 			if ((n = MIN(bcnt, bp->b_wptr - bp->b_rptr)) != 0 &&
6792 			    copyout(bp->b_rptr, ubuf, n)) {
6793 				error = EFAULT;
6794 				mutex_enter(&stp->sd_lock);
6795 				/*
6796 				 * clear stream head pri flag based on
6797 				 * first message type
6798 				 */
6799 				if (type >= QPCTL) {
6800 					ASSERT(type == M_PCPROTO);
6801 					stp->sd_flag &= ~STRPRI;
6802 				}
6803 				more = 0;
6804 				freemsg(bp);
6805 				goto getmout;
6806 			}
6807 			ubuf += n;
6808 			bp->b_rptr += n;
6809 			if (bp->b_rptr >= bp->b_wptr) {
6810 				nbp = bp;
6811 				bp = bp->b_cont;
6812 				freeb(nbp);
6813 			}
6814 			ASSERT(n <= bcnt);
6815 			bcnt -= n;
6816 			if (bcnt == 0)
6817 				break;
6818 		}
6819 		mctl->len = mctl->maxlen - bcnt;
6820 	} else
6821 		mctl->len = -1;
6822 
6823 	if (bp && bp->b_datap->db_type != M_DATA) {
6824 		/*
6825 		 * More PROTO blocks in msg.
6826 		 */
6827 		more |= MORECTL;
6828 		savemp = bp;
6829 		while (bp && bp->b_datap->db_type != M_DATA) {
6830 			savemptail = bp;
6831 			bp = bp->b_cont;
6832 		}
6833 		savemptail->b_cont = NULL;
6834 	}
6835 
6836 	/*
6837 	 * Now process DATA blocks, if any.
6838 	 */
6839 	if (mdata->maxlen >= 0 && bp) {
6840 		/*
6841 		 * struiocopyout will consume a potential zero-length
6842 		 * M_DATA even if uio_resid is zero.
6843 		 */
6844 		size_t oldresid = uiop->uio_resid;
6845 
6846 		bp = struiocopyout(bp, uiop, &error);
6847 		if (error != 0) {
6848 			mutex_enter(&stp->sd_lock);
6849 			/*
6850 			 * clear stream head hi pri flag based on
6851 			 * first message
6852 			 */
6853 			if (type >= QPCTL) {
6854 				ASSERT(type == M_PCPROTO);
6855 				stp->sd_flag &= ~STRPRI;
6856 			}
6857 			more = 0;
6858 			freemsg(savemp);
6859 			goto getmout;
6860 		}
6861 		/*
6862 		 * (pr == 1) indicates a partial read.
6863 		 */
6864 		if (oldresid > uiop->uio_resid)
6865 			pr = 1;
6866 		mdata->len = mdata->maxlen - uiop->uio_resid;
6867 	} else
6868 		mdata->len = -1;
6869 
6870 	if (bp) {			/* more data blocks in msg */
6871 		more |= MOREDATA;
6872 		if (savemp)
6873 			savemptail->b_cont = bp;
6874 		else
6875 			savemp = bp;
6876 	}
6877 
6878 	mutex_enter(&stp->sd_lock);
6879 	if (savemp) {
6880 		if (pr && (savemp->b_datap->db_type == M_DATA) &&
6881 		    msgnodata(savemp)) {
6882 			/*
6883 			 * Avoid queuing a zero-length tail part of
6884 			 * a message. pr=1 indicates that we read some of
6885 			 * the message.
6886 			 */
6887 			freemsg(savemp);
6888 			more &= ~MOREDATA;
6889 			/*
6890 			 * clear stream head hi pri flag based on
6891 			 * first message
6892 			 */
6893 			if (type >= QPCTL) {
6894 				ASSERT(type == M_PCPROTO);
6895 				stp->sd_flag &= ~STRPRI;
6896 			}
6897 		} else {
6898 			savemp->b_band = pri;
6899 			/*
6900 			 * If the first message was HIPRI and the one we're
6901 			 * putting back isn't, then clear STRPRI, otherwise
6902 			 * set STRPRI again.  Note that we must set STRPRI
6903 			 * again since the flush logic in strrput_nondata()
6904 			 * may have cleared it while we had sd_lock dropped.
6905 			 */
6906 			if (type >= QPCTL) {
6907 				ASSERT(type == M_PCPROTO);
6908 				if (queclass(savemp) < QPCTL)
6909 					stp->sd_flag &= ~STRPRI;
6910 				else
6911 					stp->sd_flag |= STRPRI;
6912 			} else if (queclass(savemp) >= QPCTL) {
6913 				/*
6914 				 * The first message was not a HIPRI message,
6915 				 * but the one we are about to putback is.
6916 				 * For simplicitly, we do not allow for HIPRI
6917 				 * messages to be embedded in the message
6918 				 * body, so just force it to same type as
6919 				 * first message.
6920 				 */
6921 				ASSERT(type == M_DATA || type == M_PROTO);
6922 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
6923 				savemp->b_datap->db_type = type;
6924 			}
6925 			if (mark != 0) {
6926 				savemp->b_flag |= mark & ~_LASTMARK;
6927 				if ((mark & _LASTMARK) &&
6928 				    (stp->sd_mark == NULL)) {
6929 					/*
6930 					 * If another marked message arrived
6931 					 * while sd_lock was not held sd_mark
6932 					 * would be non-NULL.
6933 					 */
6934 					stp->sd_mark = savemp;
6935 				}
6936 			}
6937 			putback(stp, q, savemp, pri);
6938 		}
6939 	} else {
6940 		/*
6941 		 * The complete message was consumed.
6942 		 *
6943 		 * If another M_PCPROTO arrived while sd_lock was not held
6944 		 * it would have been discarded since STRPRI was still set.
6945 		 *
6946 		 * Move the MSG*MARKNEXT information
6947 		 * to the stream head just in case
6948 		 * the read queue becomes empty.
6949 		 * clear stream head hi pri flag based on
6950 		 * first message
6951 		 *
6952 		 * If the stream head was at the mark
6953 		 * (STRATMARK) before we dropped sd_lock above
6954 		 * and some data was consumed then we have
6955 		 * moved past the mark thus STRATMARK is
6956 		 * cleared. However, if a message arrived in
6957 		 * strrput during the copyout above causing
6958 		 * STRATMARK to be set we can not clear that
6959 		 * flag.
6960 		 */
6961 		if (type >= QPCTL) {
6962 			ASSERT(type == M_PCPROTO);
6963 			stp->sd_flag &= ~STRPRI;
6964 		}
6965 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
6966 			if (mark & MSGMARKNEXT) {
6967 				stp->sd_flag &= ~STRNOTATMARK;
6968 				stp->sd_flag |= STRATMARK;
6969 			} else if (mark & MSGNOTMARKNEXT) {
6970 				stp->sd_flag &= ~STRATMARK;
6971 				stp->sd_flag |= STRNOTATMARK;
6972 			} else {
6973 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
6974 			}
6975 		} else if (pr && (old_sd_flag & STRATMARK)) {
6976 			stp->sd_flag &= ~STRATMARK;
6977 		}
6978 	}
6979 
6980 	*flagsp = flg;
6981 	*prip = pri;
6982 
6983 	/*
6984 	 * Getmsg cleanup processing - if the state of the queue has changed
6985 	 * some signals may need to be sent and/or poll awakened.
6986 	 */
6987 getmout:
6988 	qbackenable(q, pri);
6989 
6990 	/*
6991 	 * We dropped the stream head lock above. Send all M_SIG messages
6992 	 * before processing stream head for SIGPOLL messages.
6993 	 */
6994 	ASSERT(MUTEX_HELD(&stp->sd_lock));
6995 	while ((bp = q->q_first) != NULL &&
6996 	    (bp->b_datap->db_type == M_SIG)) {
6997 		/*
6998 		 * sd_lock is held so the content of the read queue can not
6999 		 * change.
7000 		 */
7001 		bp = getq(q);
7002 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7003 
7004 		strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7005 		mutex_exit(&stp->sd_lock);
7006 		freemsg(bp);
7007 		if (STREAM_NEEDSERVICE(stp))
7008 			stream_runservice(stp);
7009 		mutex_enter(&stp->sd_lock);
7010 	}
7011 
7012 	/*
7013 	 * stream head cannot change while we make the determination
7014 	 * whether or not to send a signal. Drop the flag to allow strrput
7015 	 * to send firstmsgsigs again.
7016 	 */
7017 	stp->sd_flag &= ~STRGETINPROG;
7018 
7019 	/*
7020 	 * If the type of message at the front of the queue changed
7021 	 * due to the receive the appropriate signals and pollwakeup events
7022 	 * are generated. The type of changes are:
7023 	 *	Processed a hipri message, q_first is not hipri.
7024 	 *	Processed a band X message, and q_first is band Y.
7025 	 * The generated signals and pollwakeups are identical to what
7026 	 * strrput() generates should the message that is now on q_first
7027 	 * arrive to an empty read queue.
7028 	 *
7029 	 * Note: only strrput will send a signal for a hipri message.
7030 	 */
7031 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7032 		strsigset_t signals = 0;
7033 		strpollset_t pollwakeups = 0;
7034 
7035 		if (flg & MSG_HIPRI) {
7036 			/*
7037 			 * Removed a hipri message. Regular data at
7038 			 * the front of  the queue.
7039 			 */
7040 			if (bp->b_band == 0) {
7041 				signals = S_INPUT | S_RDNORM;
7042 				pollwakeups = POLLIN | POLLRDNORM;
7043 			} else {
7044 				signals = S_INPUT | S_RDBAND;
7045 				pollwakeups = POLLIN | POLLRDBAND;
7046 			}
7047 		} else if (pri != bp->b_band) {
7048 			/*
7049 			 * The band is different for the new q_first.
7050 			 */
7051 			if (bp->b_band == 0) {
7052 				signals = S_RDNORM;
7053 				pollwakeups = POLLIN | POLLRDNORM;
7054 			} else {
7055 				signals = S_RDBAND;
7056 				pollwakeups = POLLIN | POLLRDBAND;
7057 			}
7058 		}
7059 
7060 		if (pollwakeups != 0) {
7061 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
7062 				if (!(stp->sd_rput_opt & SR_POLLIN))
7063 					goto no_pollwake;
7064 				stp->sd_rput_opt &= ~SR_POLLIN;
7065 			}
7066 			mutex_exit(&stp->sd_lock);
7067 			pollwakeup(&stp->sd_pollist, pollwakeups);
7068 			mutex_enter(&stp->sd_lock);
7069 		}
7070 no_pollwake:
7071 
7072 		if (stp->sd_sigflags & signals)
7073 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7074 	}
7075 	mutex_exit(&stp->sd_lock);
7076 
7077 	rvp->r_val1 = more;
7078 	return (error);
7079 #undef	_LASTMARK
7080 }
7081 
7082 /*
7083  * Get the next message from the read queue.  If the message is
7084  * priority, STRPRI will have been set by strrput().  This flag
7085  * should be reset only when the entire message at the front of the
7086  * queue as been consumed.
7087  *
7088  * If uiop is NULL all data is returned in mctlp.
7089  * Note that a NULL uiop implies that FNDELAY and FNONBLOCK are assumed
7090  * not enabled.
7091  * The timeout parameter is in milliseconds; -1 for infinity.
7092  * This routine handles the consolidation private flags:
7093  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
7094  *	MSG_DELAYERROR	Defer the error check until the queue is empty.
7095  *	MSG_HOLDSIG	Hold signals while waiting for data.
7096  *	MSG_IPEEK	Only peek at messages.
7097  *	MSG_DISCARDTAIL	Discard the tail M_DATA part of the message
7098  *			that doesn't fit.
7099  *	MSG_NOMARK	If the message is marked leave it on the queue.
7100  *
7101  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
7102  */
7103 int
7104 kstrgetmsg(
7105 	struct vnode *vp,
7106 	mblk_t **mctlp,
7107 	struct uio *uiop,
7108 	unsigned char *prip,
7109 	int *flagsp,
7110 	clock_t timout,
7111 	rval_t *rvp)
7112 {
7113 	struct stdata *stp;
7114 	mblk_t *bp, *nbp;
7115 	mblk_t *savemp = NULL;
7116 	mblk_t *savemptail = NULL;
7117 	int flags;
7118 	uint_t old_sd_flag;
7119 	int flg;
7120 	int more = 0;
7121 	int error = 0;
7122 	char first = 1;
7123 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
7124 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
7125 	unsigned char pri = 0;
7126 	queue_t *q;
7127 	int	pr = 0;			/* Partial read successful */
7128 	unsigned char type;
7129 
7130 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_ENTER,
7131 	    "kstrgetmsg:%p", vp);
7132 
7133 	ASSERT(vp->v_stream);
7134 	stp = vp->v_stream;
7135 	rvp->r_val1 = 0;
7136 
7137 	mutex_enter(&stp->sd_lock);
7138 
7139 	if ((error = i_straccess(stp, JCREAD)) != 0) {
7140 		mutex_exit(&stp->sd_lock);
7141 		return (error);
7142 	}
7143 
7144 	flags = *flagsp;
7145 	if (stp->sd_flag & (STRDERR|STPLEX)) {
7146 		if ((stp->sd_flag & STPLEX) ||
7147 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == 0) {
7148 			error = strgeterr(stp, STRDERR|STPLEX,
7149 			    (flags & MSG_IPEEK));
7150 			if (error != 0) {
7151 				mutex_exit(&stp->sd_lock);
7152 				return (error);
7153 			}
7154 		}
7155 	}
7156 	mutex_exit(&stp->sd_lock);
7157 
7158 	switch (flags & (MSG_HIPRI|MSG_ANY|MSG_BAND)) {
7159 	case MSG_HIPRI:
7160 		if (*prip != 0)
7161 			return (EINVAL);
7162 		break;
7163 
7164 	case MSG_ANY:
7165 	case MSG_BAND:
7166 		break;
7167 
7168 	default:
7169 		return (EINVAL);
7170 	}
7171 
7172 retry:
7173 	q = _RD(stp->sd_wrq);
7174 	mutex_enter(&stp->sd_lock);
7175 	old_sd_flag = stp->sd_flag;
7176 	mark = 0;
7177 	for (;;) {
7178 		int done = 0;
7179 		int waitflag;
7180 		int fmode;
7181 		mblk_t *q_first = q->q_first;
7182 
7183 		/*
7184 		 * This section of the code operates just like the code
7185 		 * in strgetmsg().  There is a comment there about what
7186 		 * is going on here.
7187 		 */
7188 		if (!(flags & (MSG_HIPRI|MSG_BAND))) {
7189 			/* Asking for normal, band0 data */
7190 			bp = strget(stp, q, uiop, first, &error);
7191 			ASSERT(MUTEX_HELD(&stp->sd_lock));
7192 			if (bp != NULL) {
7193 				if (DB_TYPE(bp) == M_SIG) {
7194 					strsignal_nolock(stp, *bp->b_rptr,
7195 					    bp->b_band);
7196 					freemsg(bp);
7197 					continue;
7198 				} else {
7199 					break;
7200 				}
7201 			}
7202 			if (error != 0) {
7203 				goto getmout;
7204 			}
7205 		/*
7206 		 * We can't depend on the value of STRPRI here because
7207 		 * the stream head may be in transit. Therefore, we
7208 		 * must look at the type of the first message to
7209 		 * determine if a high priority messages is waiting
7210 		 */
7211 		} else if ((flags & MSG_HIPRI) && q_first != NULL &&
7212 		    DB_TYPE(q_first) >= QPCTL &&
7213 		    (bp = getq_noenab(q, 0)) != NULL) {
7214 			ASSERT(DB_TYPE(bp) >= QPCTL);
7215 			break;
7216 		} else if ((flags & MSG_BAND) && q_first != NULL &&
7217 		    ((q_first->b_band >= *prip) || DB_TYPE(q_first) >= QPCTL) &&
7218 		    (bp = getq_noenab(q, 0)) != NULL) {
7219 			/*
7220 			 * Asked for at least band "prip" and got either at
7221 			 * least that band or a hipri message.
7222 			 */
7223 			ASSERT(bp->b_band >= *prip || DB_TYPE(bp) >= QPCTL);
7224 			if (DB_TYPE(bp) == M_SIG) {
7225 				strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7226 				freemsg(bp);
7227 				continue;
7228 			} else {
7229 				break;
7230 			}
7231 		}
7232 
7233 		/* No data. Time to sleep? */
7234 		qbackenable(q, 0);
7235 
7236 		/*
7237 		 * Delayed error notification?
7238 		 */
7239 		if ((stp->sd_flag & (STRDERR|STPLEX)) &&
7240 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == MSG_DELAYERROR) {
7241 			error = strgeterr(stp, STRDERR|STPLEX,
7242 			    (flags & MSG_IPEEK));
7243 			if (error != 0) {
7244 				mutex_exit(&stp->sd_lock);
7245 				return (error);
7246 			}
7247 		}
7248 
7249 		/*
7250 		 * If STRHUP or STREOF, return 0 length control and data.
7251 		 * If a read(fd,buf,0) has been done, do not sleep, just
7252 		 * return.
7253 		 *
7254 		 * If mctlp == NULL and uiop == NULL, then the code will
7255 		 * do the strwaitq. This is an understood way of saying
7256 		 * sleep "polling" until a message is received.
7257 		 */
7258 		if ((stp->sd_flag & (STRHUP|STREOF)) ||
7259 		    (uiop != NULL && uiop->uio_resid == 0)) {
7260 			if (mctlp != NULL)
7261 				*mctlp = NULL;
7262 			*flagsp = 0;
7263 			mutex_exit(&stp->sd_lock);
7264 			return (0);
7265 		}
7266 
7267 		waitflag = GETWAIT;
7268 		if (flags &
7269 		    (MSG_HOLDSIG|MSG_IGNERROR|MSG_IPEEK|MSG_DELAYERROR)) {
7270 			if (flags & MSG_HOLDSIG)
7271 				waitflag |= STR_NOSIG;
7272 			if (flags & MSG_IGNERROR)
7273 				waitflag |= STR_NOERROR;
7274 			if (flags & MSG_IPEEK)
7275 				waitflag |= STR_PEEK;
7276 			if (flags & MSG_DELAYERROR)
7277 				waitflag |= STR_DELAYERR;
7278 		}
7279 		if (uiop != NULL)
7280 			fmode = uiop->uio_fmode;
7281 		else
7282 			fmode = 0;
7283 
7284 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_WAIT,
7285 		    "kstrgetmsg calls strwaitq:%p, %p",
7286 		    vp, uiop);
7287 		if (((error = strwaitq(stp, waitflag, (ssize_t)0,
7288 		    fmode, timout, &done))) != 0 || done) {
7289 			TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_DONE,
7290 			    "kstrgetmsg error or done:%p, %p",
7291 			    vp, uiop);
7292 			mutex_exit(&stp->sd_lock);
7293 			return (error);
7294 		}
7295 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_AWAKE,
7296 		    "kstrgetmsg awakes:%p, %p", vp, uiop);
7297 		if ((error = i_straccess(stp, JCREAD)) != 0) {
7298 			mutex_exit(&stp->sd_lock);
7299 			return (error);
7300 		}
7301 		first = 0;
7302 	}
7303 	ASSERT(bp != NULL);
7304 	/*
7305 	 * Extract any mark information. If the message is not completely
7306 	 * consumed this information will be put in the mblk
7307 	 * that is putback.
7308 	 * If MSGMARKNEXT is set and the message is completely consumed
7309 	 * the STRATMARK flag will be set below. Likewise, if
7310 	 * MSGNOTMARKNEXT is set and the message is
7311 	 * completely consumed STRNOTATMARK will be set.
7312 	 */
7313 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
7314 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
7315 	    (MSGMARKNEXT|MSGNOTMARKNEXT));
7316 	pri = bp->b_band;
7317 	if (mark != 0) {
7318 		/*
7319 		 * If the caller doesn't want the mark return.
7320 		 * Used to implement MSG_WAITALL in sockets.
7321 		 */
7322 		if (flags & MSG_NOMARK) {
7323 			putback(stp, q, bp, pri);
7324 			qbackenable(q, pri);
7325 			mutex_exit(&stp->sd_lock);
7326 			return (EWOULDBLOCK);
7327 		}
7328 		if (bp == stp->sd_mark) {
7329 			mark |= _LASTMARK;
7330 			stp->sd_mark = NULL;
7331 		}
7332 	}
7333 
7334 	/*
7335 	 * keep track of the first message type
7336 	 */
7337 	type = bp->b_datap->db_type;
7338 
7339 	if (bp->b_datap->db_type == M_PASSFP) {
7340 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7341 			stp->sd_mark = bp;
7342 		bp->b_flag |= mark & ~_LASTMARK;
7343 		putback(stp, q, bp, pri);
7344 		qbackenable(q, pri);
7345 		mutex_exit(&stp->sd_lock);
7346 		return (EBADMSG);
7347 	}
7348 	ASSERT(type != M_SIG);
7349 
7350 	if (flags & MSG_IPEEK) {
7351 		/*
7352 		 * Clear any struioflag - we do the uiomove over again
7353 		 * when peeking since it simplifies the code.
7354 		 *
7355 		 * Dup the message and put the original back on the queue.
7356 		 * If dupmsg() fails, try again with copymsg() to see if
7357 		 * there is indeed a shortage of memory.  dupmsg() may fail
7358 		 * if db_ref in any of the messages reaches its limit.
7359 		 */
7360 
7361 		if ((nbp = dupmsg(bp)) == NULL && (nbp = copymsg(bp)) == NULL) {
7362 			/*
7363 			 * Restore the state of the stream head since we
7364 			 * need to drop sd_lock (strwaitbuf is sleeping).
7365 			 */
7366 			size_t size = msgdsize(bp);
7367 
7368 			if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7369 				stp->sd_mark = bp;
7370 			bp->b_flag |= mark & ~_LASTMARK;
7371 			putback(stp, q, bp, pri);
7372 			mutex_exit(&stp->sd_lock);
7373 			error = strwaitbuf(size, BPRI_HI);
7374 			if (error) {
7375 				/*
7376 				 * There is no net change to the queue thus
7377 				 * no need to qbackenable.
7378 				 */
7379 				return (error);
7380 			}
7381 			goto retry;
7382 		}
7383 
7384 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7385 			stp->sd_mark = bp;
7386 		bp->b_flag |= mark & ~_LASTMARK;
7387 		putback(stp, q, bp, pri);
7388 		bp = nbp;
7389 	}
7390 
7391 	/*
7392 	 * Set this flag so strrput will not generate signals. Need to
7393 	 * make sure this flag is cleared before leaving this routine
7394 	 * else signals will stop being sent.
7395 	 */
7396 	stp->sd_flag |= STRGETINPROG;
7397 	mutex_exit(&stp->sd_lock);
7398 
7399 	if ((stp->sd_rputdatafunc != NULL) && (DB_TYPE(bp) == M_DATA)) {
7400 		mblk_t *tmp, *prevmp;
7401 
7402 		/*
7403 		 * Put first non-data mblk back to stream head and
7404 		 * cut the mblk chain so sd_rputdatafunc only sees
7405 		 * M_DATA mblks. We can skip the first mblk since it
7406 		 * is M_DATA according to the condition above.
7407 		 */
7408 		for (prevmp = bp, tmp = bp->b_cont; tmp != NULL;
7409 		    prevmp = tmp, tmp = tmp->b_cont) {
7410 			if (DB_TYPE(tmp) != M_DATA) {
7411 				prevmp->b_cont = NULL;
7412 				mutex_enter(&stp->sd_lock);
7413 				putback(stp, q, tmp, tmp->b_band);
7414 				mutex_exit(&stp->sd_lock);
7415 				break;
7416 			}
7417 		}
7418 
7419 		bp = (stp->sd_rputdatafunc)(stp->sd_vnode, bp,
7420 		    NULL, NULL, NULL, NULL);
7421 
7422 		if (bp == NULL)
7423 			goto retry;
7424 	}
7425 
7426 	if (STREAM_NEEDSERVICE(stp))
7427 		stream_runservice(stp);
7428 
7429 	/*
7430 	 * Set HIPRI flag if message is priority.
7431 	 */
7432 	if (type >= QPCTL)
7433 		flg = MSG_HIPRI;
7434 	else
7435 		flg = MSG_BAND;
7436 
7437 	/*
7438 	 * First process PROTO or PCPROTO blocks, if any.
7439 	 */
7440 	if (mctlp != NULL && type != M_DATA) {
7441 		mblk_t *nbp;
7442 
7443 		*mctlp = bp;
7444 		while (bp->b_cont && bp->b_cont->b_datap->db_type != M_DATA)
7445 			bp = bp->b_cont;
7446 		nbp = bp->b_cont;
7447 		bp->b_cont = NULL;
7448 		bp = nbp;
7449 	}
7450 
7451 	if (bp && bp->b_datap->db_type != M_DATA) {
7452 		/*
7453 		 * More PROTO blocks in msg. Will only happen if mctlp is NULL.
7454 		 */
7455 		more |= MORECTL;
7456 		savemp = bp;
7457 		while (bp && bp->b_datap->db_type != M_DATA) {
7458 			savemptail = bp;
7459 			bp = bp->b_cont;
7460 		}
7461 		savemptail->b_cont = NULL;
7462 	}
7463 
7464 	/*
7465 	 * Now process DATA blocks, if any.
7466 	 */
7467 	if (uiop == NULL) {
7468 		/* Append data to tail of mctlp */
7469 
7470 		if (mctlp != NULL) {
7471 			mblk_t **mpp = mctlp;
7472 
7473 			while (*mpp != NULL)
7474 				mpp = &((*mpp)->b_cont);
7475 			*mpp = bp;
7476 			bp = NULL;
7477 		}
7478 	} else if (uiop->uio_resid >= 0 && bp) {
7479 		size_t oldresid = uiop->uio_resid;
7480 
7481 		/*
7482 		 * If a streams message is likely to consist
7483 		 * of many small mblks, it is pulled up into
7484 		 * one continuous chunk of memory.
7485 		 * The size of the first mblk may be bogus because
7486 		 * successive read() calls on the socket reduce
7487 		 * the size of this mblk until it is exhausted
7488 		 * and then the code walks on to the next. Thus
7489 		 * the size of the mblk may not be the original size
7490 		 * that was passed up, it's simply a remainder
7491 		 * and hence can be very small without any
7492 		 * implication that the packet is badly fragmented.
7493 		 * So the size of the possible second mblk is
7494 		 * used to spot a badly fragmented packet.
7495 		 * see longer comment at top of page
7496 		 * by mblk_pull_len declaration.
7497 		 */
7498 
7499 		if (bp->b_cont != NULL && MBLKL(bp->b_cont) < mblk_pull_len) {
7500 			(void) pullupmsg(bp, -1);
7501 		}
7502 
7503 		bp = struiocopyout(bp, uiop, &error);
7504 		if (error != 0) {
7505 			if (mctlp != NULL) {
7506 				freemsg(*mctlp);
7507 				*mctlp = NULL;
7508 			} else
7509 				freemsg(savemp);
7510 			mutex_enter(&stp->sd_lock);
7511 			/*
7512 			 * clear stream head hi pri flag based on
7513 			 * first message
7514 			 */
7515 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7516 				ASSERT(type == M_PCPROTO);
7517 				stp->sd_flag &= ~STRPRI;
7518 			}
7519 			more = 0;
7520 			goto getmout;
7521 		}
7522 		/*
7523 		 * (pr == 1) indicates a partial read.
7524 		 */
7525 		if (oldresid > uiop->uio_resid)
7526 			pr = 1;
7527 	}
7528 
7529 	if (bp) {			/* more data blocks in msg */
7530 		more |= MOREDATA;
7531 		if (savemp)
7532 			savemptail->b_cont = bp;
7533 		else
7534 			savemp = bp;
7535 	}
7536 
7537 	mutex_enter(&stp->sd_lock);
7538 	if (savemp) {
7539 		if (flags & (MSG_IPEEK|MSG_DISCARDTAIL)) {
7540 			/*
7541 			 * When MSG_DISCARDTAIL is set or
7542 			 * when peeking discard any tail. When peeking this
7543 			 * is the tail of the dup that was copied out - the
7544 			 * message has already been putback on the queue.
7545 			 * Return MOREDATA to the caller even though the data
7546 			 * is discarded. This is used by sockets (to
7547 			 * set MSG_TRUNC).
7548 			 */
7549 			freemsg(savemp);
7550 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7551 				ASSERT(type == M_PCPROTO);
7552 				stp->sd_flag &= ~STRPRI;
7553 			}
7554 		} else if (pr && (savemp->b_datap->db_type == M_DATA) &&
7555 		    msgnodata(savemp)) {
7556 			/*
7557 			 * Avoid queuing a zero-length tail part of
7558 			 * a message. pr=1 indicates that we read some of
7559 			 * the message.
7560 			 */
7561 			freemsg(savemp);
7562 			more &= ~MOREDATA;
7563 			if (type >= QPCTL) {
7564 				ASSERT(type == M_PCPROTO);
7565 				stp->sd_flag &= ~STRPRI;
7566 			}
7567 		} else {
7568 			savemp->b_band = pri;
7569 			/*
7570 			 * If the first message was HIPRI and the one we're
7571 			 * putting back isn't, then clear STRPRI, otherwise
7572 			 * set STRPRI again.  Note that we must set STRPRI
7573 			 * again since the flush logic in strrput_nondata()
7574 			 * may have cleared it while we had sd_lock dropped.
7575 			 */
7576 
7577 			if (type >= QPCTL) {
7578 				ASSERT(type == M_PCPROTO);
7579 				if (queclass(savemp) < QPCTL)
7580 					stp->sd_flag &= ~STRPRI;
7581 				else
7582 					stp->sd_flag |= STRPRI;
7583 			} else if (queclass(savemp) >= QPCTL) {
7584 				/*
7585 				 * The first message was not a HIPRI message,
7586 				 * but the one we are about to putback is.
7587 				 * For simplicitly, we do not allow for HIPRI
7588 				 * messages to be embedded in the message
7589 				 * body, so just force it to same type as
7590 				 * first message.
7591 				 */
7592 				ASSERT(type == M_DATA || type == M_PROTO);
7593 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
7594 				savemp->b_datap->db_type = type;
7595 			}
7596 			if (mark != 0) {
7597 				if ((mark & _LASTMARK) &&
7598 				    (stp->sd_mark == NULL)) {
7599 					/*
7600 					 * If another marked message arrived
7601 					 * while sd_lock was not held sd_mark
7602 					 * would be non-NULL.
7603 					 */
7604 					stp->sd_mark = savemp;
7605 				}
7606 				savemp->b_flag |= mark & ~_LASTMARK;
7607 			}
7608 			putback(stp, q, savemp, pri);
7609 		}
7610 	} else if (!(flags & MSG_IPEEK)) {
7611 		/*
7612 		 * The complete message was consumed.
7613 		 *
7614 		 * If another M_PCPROTO arrived while sd_lock was not held
7615 		 * it would have been discarded since STRPRI was still set.
7616 		 *
7617 		 * Move the MSG*MARKNEXT information
7618 		 * to the stream head just in case
7619 		 * the read queue becomes empty.
7620 		 * clear stream head hi pri flag based on
7621 		 * first message
7622 		 *
7623 		 * If the stream head was at the mark
7624 		 * (STRATMARK) before we dropped sd_lock above
7625 		 * and some data was consumed then we have
7626 		 * moved past the mark thus STRATMARK is
7627 		 * cleared. However, if a message arrived in
7628 		 * strrput during the copyout above causing
7629 		 * STRATMARK to be set we can not clear that
7630 		 * flag.
7631 		 * XXX A "perimeter" would help by single-threading strrput,
7632 		 * strread, strgetmsg and kstrgetmsg.
7633 		 */
7634 		if (type >= QPCTL) {
7635 			ASSERT(type == M_PCPROTO);
7636 			stp->sd_flag &= ~STRPRI;
7637 		}
7638 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
7639 			if (mark & MSGMARKNEXT) {
7640 				stp->sd_flag &= ~STRNOTATMARK;
7641 				stp->sd_flag |= STRATMARK;
7642 			} else if (mark & MSGNOTMARKNEXT) {
7643 				stp->sd_flag &= ~STRATMARK;
7644 				stp->sd_flag |= STRNOTATMARK;
7645 			} else {
7646 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
7647 			}
7648 		} else if (pr && (old_sd_flag & STRATMARK)) {
7649 			stp->sd_flag &= ~STRATMARK;
7650 		}
7651 	}
7652 
7653 	*flagsp = flg;
7654 	*prip = pri;
7655 
7656 	/*
7657 	 * Getmsg cleanup processing - if the state of the queue has changed
7658 	 * some signals may need to be sent and/or poll awakened.
7659 	 */
7660 getmout:
7661 	qbackenable(q, pri);
7662 
7663 	/*
7664 	 * We dropped the stream head lock above. Send all M_SIG messages
7665 	 * before processing stream head for SIGPOLL messages.
7666 	 */
7667 	ASSERT(MUTEX_HELD(&stp->sd_lock));
7668 	while ((bp = q->q_first) != NULL &&
7669 	    (bp->b_datap->db_type == M_SIG)) {
7670 		/*
7671 		 * sd_lock is held so the content of the read queue can not
7672 		 * change.
7673 		 */
7674 		bp = getq(q);
7675 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7676 
7677 		strsignal_nolock(stp, *bp->b_rptr, bp->b_band);
7678 		mutex_exit(&stp->sd_lock);
7679 		freemsg(bp);
7680 		if (STREAM_NEEDSERVICE(stp))
7681 			stream_runservice(stp);
7682 		mutex_enter(&stp->sd_lock);
7683 	}
7684 
7685 	/*
7686 	 * stream head cannot change while we make the determination
7687 	 * whether or not to send a signal. Drop the flag to allow strrput
7688 	 * to send firstmsgsigs again.
7689 	 */
7690 	stp->sd_flag &= ~STRGETINPROG;
7691 
7692 	/*
7693 	 * If the type of message at the front of the queue changed
7694 	 * due to the receive the appropriate signals and pollwakeup events
7695 	 * are generated. The type of changes are:
7696 	 *	Processed a hipri message, q_first is not hipri.
7697 	 *	Processed a band X message, and q_first is band Y.
7698 	 * The generated signals and pollwakeups are identical to what
7699 	 * strrput() generates should the message that is now on q_first
7700 	 * arrive to an empty read queue.
7701 	 *
7702 	 * Note: only strrput will send a signal for a hipri message.
7703 	 */
7704 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7705 		strsigset_t signals = 0;
7706 		strpollset_t pollwakeups = 0;
7707 
7708 		if (flg & MSG_HIPRI) {
7709 			/*
7710 			 * Removed a hipri message. Regular data at
7711 			 * the front of  the queue.
7712 			 */
7713 			if (bp->b_band == 0) {
7714 				signals = S_INPUT | S_RDNORM;
7715 				pollwakeups = POLLIN | POLLRDNORM;
7716 			} else {
7717 				signals = S_INPUT | S_RDBAND;
7718 				pollwakeups = POLLIN | POLLRDBAND;
7719 			}
7720 		} else if (pri != bp->b_band) {
7721 			/*
7722 			 * The band is different for the new q_first.
7723 			 */
7724 			if (bp->b_band == 0) {
7725 				signals = S_RDNORM;
7726 				pollwakeups = POLLIN | POLLRDNORM;
7727 			} else {
7728 				signals = S_RDBAND;
7729 				pollwakeups = POLLIN | POLLRDBAND;
7730 			}
7731 		}
7732 
7733 		if (pollwakeups != 0) {
7734 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
7735 				if (!(stp->sd_rput_opt & SR_POLLIN))
7736 					goto no_pollwake;
7737 				stp->sd_rput_opt &= ~SR_POLLIN;
7738 			}
7739 			mutex_exit(&stp->sd_lock);
7740 			pollwakeup(&stp->sd_pollist, pollwakeups);
7741 			mutex_enter(&stp->sd_lock);
7742 		}
7743 no_pollwake:
7744 
7745 		if (stp->sd_sigflags & signals)
7746 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7747 	}
7748 	mutex_exit(&stp->sd_lock);
7749 
7750 	rvp->r_val1 = more;
7751 	return (error);
7752 #undef	_LASTMARK
7753 }
7754 
7755 /*
7756  * Put a message downstream.
7757  *
7758  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7759  */
7760 int
7761 strputmsg(
7762 	struct vnode *vp,
7763 	struct strbuf *mctl,
7764 	struct strbuf *mdata,
7765 	unsigned char pri,
7766 	int flag,
7767 	int fmode)
7768 {
7769 	struct stdata *stp;
7770 	queue_t *wqp;
7771 	mblk_t *mp;
7772 	ssize_t msgsize;
7773 	ssize_t rmin, rmax;
7774 	int error;
7775 	struct uio uios;
7776 	struct uio *uiop = &uios;
7777 	struct iovec iovs;
7778 	int xpg4 = 0;
7779 
7780 	ASSERT(vp->v_stream);
7781 	stp = vp->v_stream;
7782 	wqp = stp->sd_wrq;
7783 
7784 	/*
7785 	 * If it is an XPG4 application, we need to send
7786 	 * SIGPIPE below
7787 	 */
7788 
7789 	xpg4 = (flag & MSG_XPG4) ? 1 : 0;
7790 	flag &= ~MSG_XPG4;
7791 
7792 	if (AU_AUDITING())
7793 		audit_strputmsg(vp, mctl, mdata, pri, flag, fmode);
7794 
7795 	mutex_enter(&stp->sd_lock);
7796 
7797 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
7798 		mutex_exit(&stp->sd_lock);
7799 		return (error);
7800 	}
7801 
7802 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7803 		error = strwriteable(stp, B_FALSE, xpg4);
7804 		if (error != 0) {
7805 			mutex_exit(&stp->sd_lock);
7806 			return (error);
7807 		}
7808 	}
7809 
7810 	mutex_exit(&stp->sd_lock);
7811 
7812 	/*
7813 	 * Check for legal flag value.
7814 	 */
7815 	switch (flag) {
7816 	case MSG_HIPRI:
7817 		if ((mctl->len < 0) || (pri != 0))
7818 			return (EINVAL);
7819 		break;
7820 	case MSG_BAND:
7821 		break;
7822 
7823 	default:
7824 		return (EINVAL);
7825 	}
7826 
7827 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_IN,
7828 	    "strputmsg in:stp %p", stp);
7829 
7830 	/* get these values from those cached in the stream head */
7831 	rmin = stp->sd_qn_minpsz;
7832 	rmax = stp->sd_qn_maxpsz;
7833 
7834 	/*
7835 	 * Make sure ctl and data sizes together fall within the
7836 	 * limits of the max and min receive packet sizes and do
7837 	 * not exceed system limit.
7838 	 */
7839 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
7840 	if (rmax == 0) {
7841 		return (ERANGE);
7842 	}
7843 	/*
7844 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
7845 	 * Needed to prevent partial failures in the strmakedata loop.
7846 	 */
7847 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
7848 		rmax = stp->sd_maxblk;
7849 
7850 	if ((msgsize = mdata->len) < 0) {
7851 		msgsize = 0;
7852 		rmin = 0;	/* no range check for NULL data part */
7853 	}
7854 	if ((msgsize < rmin) ||
7855 	    ((msgsize > rmax) && (rmax != INFPSZ)) ||
7856 	    (mctl->len > strctlsz)) {
7857 		return (ERANGE);
7858 	}
7859 
7860 	/*
7861 	 * Setup uio and iov for data part
7862 	 */
7863 	iovs.iov_base = mdata->buf;
7864 	iovs.iov_len = msgsize;
7865 	uios.uio_iov = &iovs;
7866 	uios.uio_iovcnt = 1;
7867 	uios.uio_loffset = 0;
7868 	uios.uio_segflg = UIO_USERSPACE;
7869 	uios.uio_fmode = fmode;
7870 	uios.uio_extflg = UIO_COPY_DEFAULT;
7871 	uios.uio_resid = msgsize;
7872 	uios.uio_offset = 0;
7873 
7874 	/* Ignore flow control in strput for HIPRI */
7875 	if (flag & MSG_HIPRI)
7876 		flag |= MSG_IGNFLOW;
7877 
7878 	for (;;) {
7879 		int done = 0;
7880 
7881 		/*
7882 		 * strput will always free the ctl mblk - even when strput
7883 		 * fails.
7884 		 */
7885 		if ((error = strmakectl(mctl, flag, fmode, &mp)) != 0) {
7886 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7887 			    "strputmsg out:stp %p out %d error %d",
7888 			    stp, 1, error);
7889 			return (error);
7890 		}
7891 		/*
7892 		 * Verify that the whole message can be transferred by
7893 		 * strput.
7894 		 */
7895 		ASSERT(stp->sd_maxblk == INFPSZ ||
7896 		    stp->sd_maxblk >= mdata->len);
7897 
7898 		msgsize = mdata->len;
7899 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
7900 		mdata->len = msgsize;
7901 
7902 		if (error == 0)
7903 			break;
7904 
7905 		if (error != EWOULDBLOCK)
7906 			goto out;
7907 
7908 		mutex_enter(&stp->sd_lock);
7909 		/*
7910 		 * Check for a missed wakeup.
7911 		 * Needed since strput did not hold sd_lock across
7912 		 * the canputnext.
7913 		 */
7914 		if (bcanputnext(wqp, pri)) {
7915 			/* Try again */
7916 			mutex_exit(&stp->sd_lock);
7917 			continue;
7918 		}
7919 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAIT,
7920 		    "strputmsg wait:stp %p waits pri %d", stp, pri);
7921 		if (((error = strwaitq(stp, WRITEWAIT, (ssize_t)0, fmode, -1,
7922 		    &done)) != 0) || done) {
7923 			mutex_exit(&stp->sd_lock);
7924 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7925 			    "strputmsg out:q %p out %d error %d",
7926 			    stp, 0, error);
7927 			return (error);
7928 		}
7929 		TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAKE,
7930 		    "strputmsg wake:stp %p wakes", stp);
7931 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
7932 			mutex_exit(&stp->sd_lock);
7933 			return (error);
7934 		}
7935 		mutex_exit(&stp->sd_lock);
7936 	}
7937 out:
7938 	/*
7939 	 * For historic reasons, applications expect EAGAIN
7940 	 * when data mblk could not be allocated. so change
7941 	 * ENOMEM back to EAGAIN
7942 	 */
7943 	if (error == ENOMEM)
7944 		error = EAGAIN;
7945 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7946 	    "strputmsg out:stp %p out %d error %d", stp, 2, error);
7947 	return (error);
7948 }
7949 
7950 /*
7951  * Put a message downstream.
7952  * Can send only an M_PROTO/M_PCPROTO by passing in a NULL uiop.
7953  * The fmode flag (NDELAY, NONBLOCK) is the or of the flags in the uio
7954  * and the fmode parameter.
7955  *
7956  * This routine handles the consolidation private flags:
7957  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
7958  *	MSG_HOLDSIG	Hold signals while waiting for data.
7959  *	MSG_IGNFLOW	Don't check streams flow control.
7960  *
7961  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7962  */
7963 int
7964 kstrputmsg(
7965 	struct vnode *vp,
7966 	mblk_t *mctl,
7967 	struct uio *uiop,
7968 	ssize_t msgsize,
7969 	unsigned char pri,
7970 	int flag,
7971 	int fmode)
7972 {
7973 	struct stdata *stp;
7974 	queue_t *wqp;
7975 	ssize_t rmin, rmax;
7976 	int error;
7977 
7978 	ASSERT(vp->v_stream);
7979 	stp = vp->v_stream;
7980 	wqp = stp->sd_wrq;
7981 	if (AU_AUDITING())
7982 		audit_strputmsg(vp, NULL, NULL, pri, flag, fmode);
7983 	if (mctl == NULL)
7984 		return (EINVAL);
7985 
7986 	mutex_enter(&stp->sd_lock);
7987 
7988 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
7989 		mutex_exit(&stp->sd_lock);
7990 		freemsg(mctl);
7991 		return (error);
7992 	}
7993 
7994 	if ((stp->sd_flag & STPLEX) || !(flag & MSG_IGNERROR)) {
7995 		if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7996 			error = strwriteable(stp, B_FALSE, B_TRUE);
7997 			if (error != 0) {
7998 				mutex_exit(&stp->sd_lock);
7999 				freemsg(mctl);
8000 				return (error);
8001 			}
8002 		}
8003 	}
8004 
8005 	mutex_exit(&stp->sd_lock);
8006 
8007 	/*
8008 	 * Check for legal flag value.
8009 	 */
8010 	switch (flag & (MSG_HIPRI|MSG_BAND|MSG_ANY)) {
8011 	case MSG_HIPRI:
8012 		if (pri != 0) {
8013 			freemsg(mctl);
8014 			return (EINVAL);
8015 		}
8016 		break;
8017 	case MSG_BAND:
8018 		break;
8019 	default:
8020 		freemsg(mctl);
8021 		return (EINVAL);
8022 	}
8023 
8024 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_IN,
8025 	    "kstrputmsg in:stp %p", stp);
8026 
8027 	/* get these values from those cached in the stream head */
8028 	rmin = stp->sd_qn_minpsz;
8029 	rmax = stp->sd_qn_maxpsz;
8030 
8031 	/*
8032 	 * Make sure ctl and data sizes together fall within the
8033 	 * limits of the max and min receive packet sizes and do
8034 	 * not exceed system limit.
8035 	 */
8036 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
8037 	if (rmax == 0) {
8038 		freemsg(mctl);
8039 		return (ERANGE);
8040 	}
8041 	/*
8042 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
8043 	 * Needed to prevent partial failures in the strmakedata loop.
8044 	 */
8045 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
8046 		rmax = stp->sd_maxblk;
8047 
8048 	if (uiop == NULL) {
8049 		msgsize = -1;
8050 		rmin = -1;	/* no range check for NULL data part */
8051 	} else {
8052 		/* Use uio flags as well as the fmode parameter flags */
8053 		fmode |= uiop->uio_fmode;
8054 
8055 		if ((msgsize < rmin) ||
8056 		    ((msgsize > rmax) && (rmax != INFPSZ))) {
8057 			freemsg(mctl);
8058 			return (ERANGE);
8059 		}
8060 	}
8061 
8062 	/* Ignore flow control in strput for HIPRI */
8063 	if (flag & MSG_HIPRI)
8064 		flag |= MSG_IGNFLOW;
8065 
8066 	for (;;) {
8067 		int done = 0;
8068 		int waitflag;
8069 		mblk_t *mp;
8070 
8071 		/*
8072 		 * strput will always free the ctl mblk - even when strput
8073 		 * fails. If MSG_IGNFLOW is set then any error returned
8074 		 * will cause us to break the loop, so we don't need a copy
8075 		 * of the message. If MSG_IGNFLOW is not set, then we can
8076 		 * get hit by flow control and be forced to try again. In
8077 		 * this case we need to have a copy of the message. We
8078 		 * do this using copymsg since the message may get modified
8079 		 * by something below us.
8080 		 *
8081 		 * We've observed that many TPI providers do not check db_ref
8082 		 * on the control messages but blindly reuse them for the
8083 		 * T_OK_ACK/T_ERROR_ACK. Thus using copymsg is more
8084 		 * friendly to such providers than using dupmsg. Also, note
8085 		 * that sockfs uses MSG_IGNFLOW for all TPI control messages.
8086 		 * Only data messages are subject to flow control, hence
8087 		 * subject to this copymsg.
8088 		 */
8089 		if (flag & MSG_IGNFLOW) {
8090 			mp = mctl;
8091 			mctl = NULL;
8092 		} else {
8093 			do {
8094 				/*
8095 				 * If a message has a free pointer, the message
8096 				 * must be dupmsg to maintain this pointer.
8097 				 * Code using this facility must be sure
8098 				 * that modules below will not change the
8099 				 * contents of the dblk without checking db_ref
8100 				 * first. If db_ref is > 1, then the module
8101 				 * needs to do a copymsg first. Otherwise,
8102 				 * the contents of the dblk may become
8103 				 * inconsistent because the freesmg/freeb below
8104 				 * may end up calling atomic_add_32_nv.
8105 				 * The atomic_add_32_nv in freeb (accessing
8106 				 * all of db_ref, db_type, db_flags, and
8107 				 * db_struioflag) does not prevent other threads
8108 				 * from concurrently trying to modify e.g.
8109 				 * db_type.
8110 				 */
8111 				if (mctl->b_datap->db_frtnp != NULL)
8112 					mp = dupmsg(mctl);
8113 				else
8114 					mp = copymsg(mctl);
8115 
8116 				if (mp != NULL)
8117 					break;
8118 
8119 				error = strwaitbuf(msgdsize(mctl), BPRI_MED);
8120 				if (error) {
8121 					freemsg(mctl);
8122 					return (error);
8123 				}
8124 			} while (mp == NULL);
8125 		}
8126 		/*
8127 		 * Verify that all of msgsize can be transferred by
8128 		 * strput.
8129 		 */
8130 		ASSERT(stp->sd_maxblk == INFPSZ || stp->sd_maxblk >= msgsize);
8131 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
8132 		if (error == 0)
8133 			break;
8134 
8135 		if (error != EWOULDBLOCK)
8136 			goto out;
8137 
8138 		/*
8139 		 * IF MSG_IGNFLOW is set we should have broken out of loop
8140 		 * above.
8141 		 */
8142 		ASSERT(!(flag & MSG_IGNFLOW));
8143 		mutex_enter(&stp->sd_lock);
8144 		/*
8145 		 * Check for a missed wakeup.
8146 		 * Needed since strput did not hold sd_lock across
8147 		 * the canputnext.
8148 		 */
8149 		if (bcanputnext(wqp, pri)) {
8150 			/* Try again */
8151 			mutex_exit(&stp->sd_lock);
8152 			continue;
8153 		}
8154 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAIT,
8155 		    "kstrputmsg wait:stp %p waits pri %d", stp, pri);
8156 
8157 		waitflag = WRITEWAIT;
8158 		if (flag & (MSG_HOLDSIG|MSG_IGNERROR)) {
8159 			if (flag & MSG_HOLDSIG)
8160 				waitflag |= STR_NOSIG;
8161 			if (flag & MSG_IGNERROR)
8162 				waitflag |= STR_NOERROR;
8163 		}
8164 		if (((error = strwaitq(stp, waitflag,
8165 		    (ssize_t)0, fmode, -1, &done)) != 0) || done) {
8166 			mutex_exit(&stp->sd_lock);
8167 			TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8168 			    "kstrputmsg out:stp %p out %d error %d",
8169 			    stp, 0, error);
8170 			freemsg(mctl);
8171 			return (error);
8172 		}
8173 		TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAKE,
8174 		    "kstrputmsg wake:stp %p wakes", stp);
8175 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
8176 			mutex_exit(&stp->sd_lock);
8177 			freemsg(mctl);
8178 			return (error);
8179 		}
8180 		mutex_exit(&stp->sd_lock);
8181 	}
8182 out:
8183 	freemsg(mctl);
8184 	/*
8185 	 * For historic reasons, applications expect EAGAIN
8186 	 * when data mblk could not be allocated. so change
8187 	 * ENOMEM back to EAGAIN
8188 	 */
8189 	if (error == ENOMEM)
8190 		error = EAGAIN;
8191 	TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8192 	    "kstrputmsg out:stp %p out %d error %d", stp, 2, error);
8193 	return (error);
8194 }
8195 
8196 /*
8197  * Determines whether the necessary conditions are set on a stream
8198  * for it to be readable, writeable, or have exceptions.
8199  *
8200  * strpoll handles the consolidation private events:
8201  *	POLLNOERR	Do not return POLLERR even if there are stream
8202  *			head errors.
8203  *			Used by sockfs.
8204  *	POLLRDDATA	Do not return POLLIN unless at least one message on
8205  *			the queue contains one or more M_DATA mblks. Thus
8206  *			when this flag is set a queue with only
8207  *			M_PROTO/M_PCPROTO mblks does not return POLLIN.
8208  *			Used by sockfs to ignore T_EXDATA_IND messages.
8209  *
8210  * Note: POLLRDDATA assumes that synch streams only return messages with
8211  * an M_DATA attached (i.e. not messages consisting of only
8212  * an M_PROTO/M_PCPROTO part).
8213  */
8214 int
8215 strpoll(struct stdata *stp, short events_arg, int anyyet, short *reventsp,
8216     struct pollhead **phpp)
8217 {
8218 	int events = (ushort_t)events_arg;
8219 	int retevents = 0;
8220 	mblk_t *mp;
8221 	qband_t *qbp;
8222 	long sd_flags = stp->sd_flag;
8223 	int headlocked = 0;
8224 
8225 	/*
8226 	 * For performance, a single 'if' tests for most possible edge
8227 	 * conditions in one shot
8228 	 */
8229 	if (sd_flags & (STPLEX | STRDERR | STWRERR)) {
8230 		if (sd_flags & STPLEX) {
8231 			*reventsp = POLLNVAL;
8232 			return (EINVAL);
8233 		}
8234 		if (((events & (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) &&
8235 		    (sd_flags & STRDERR)) ||
8236 		    ((events & (POLLOUT | POLLWRNORM | POLLWRBAND)) &&
8237 		    (sd_flags & STWRERR))) {
8238 			if (!(events & POLLNOERR)) {
8239 				*reventsp = POLLERR;
8240 				return (0);
8241 			}
8242 		}
8243 	}
8244 	if (sd_flags & STRHUP) {
8245 		retevents |= POLLHUP;
8246 	} else if (events & (POLLWRNORM | POLLWRBAND)) {
8247 		queue_t *tq;
8248 		queue_t	*qp = stp->sd_wrq;
8249 
8250 		claimstr(qp);
8251 		/* Find next module forward that has a service procedure */
8252 		tq = qp->q_next->q_nfsrv;
8253 		ASSERT(tq != NULL);
8254 
8255 		if (polllock(&stp->sd_pollist, QLOCK(tq)) != 0) {
8256 			releasestr(qp);
8257 			*reventsp = POLLNVAL;
8258 			return (0);
8259 		}
8260 		if (events & POLLWRNORM) {
8261 			queue_t *sqp;
8262 
8263 			if (tq->q_flag & QFULL)
8264 				/* ensure backq svc procedure runs */
8265 				tq->q_flag |= QWANTW;
8266 			else if ((sqp = stp->sd_struiowrq) != NULL) {
8267 				/* Check sync stream barrier write q */
8268 				mutex_exit(QLOCK(tq));
8269 				if (polllock(&stp->sd_pollist,
8270 				    QLOCK(sqp)) != 0) {
8271 					releasestr(qp);
8272 					*reventsp = POLLNVAL;
8273 					return (0);
8274 				}
8275 				if (sqp->q_flag & QFULL)
8276 					/* ensure pollwakeup() is done */
8277 					sqp->q_flag |= QWANTWSYNC;
8278 				else
8279 					retevents |= POLLOUT;
8280 				/* More write events to process ??? */
8281 				if (! (events & POLLWRBAND)) {
8282 					mutex_exit(QLOCK(sqp));
8283 					releasestr(qp);
8284 					goto chkrd;
8285 				}
8286 				mutex_exit(QLOCK(sqp));
8287 				if (polllock(&stp->sd_pollist,
8288 				    QLOCK(tq)) != 0) {
8289 					releasestr(qp);
8290 					*reventsp = POLLNVAL;
8291 					return (0);
8292 				}
8293 			} else
8294 				retevents |= POLLOUT;
8295 		}
8296 		if (events & POLLWRBAND) {
8297 			qbp = tq->q_bandp;
8298 			if (qbp) {
8299 				while (qbp) {
8300 					if (qbp->qb_flag & QB_FULL)
8301 						qbp->qb_flag |= QB_WANTW;
8302 					else
8303 						retevents |= POLLWRBAND;
8304 					qbp = qbp->qb_next;
8305 				}
8306 			} else {
8307 				retevents |= POLLWRBAND;
8308 			}
8309 		}
8310 		mutex_exit(QLOCK(tq));
8311 		releasestr(qp);
8312 	}
8313 chkrd:
8314 	if (sd_flags & STRPRI) {
8315 		retevents |= (events & POLLPRI);
8316 	} else if (events & (POLLRDNORM | POLLRDBAND | POLLIN)) {
8317 		queue_t	*qp = _RD(stp->sd_wrq);
8318 		int normevents = (events & (POLLIN | POLLRDNORM));
8319 
8320 		/*
8321 		 * Note: Need to do polllock() here since ps_lock may be
8322 		 * held. See bug 4191544.
8323 		 */
8324 		if (polllock(&stp->sd_pollist, &stp->sd_lock) != 0) {
8325 			*reventsp = POLLNVAL;
8326 			return (0);
8327 		}
8328 		headlocked = 1;
8329 		mp = qp->q_first;
8330 		while (mp) {
8331 			/*
8332 			 * For POLLRDDATA we scan b_cont and b_next until we
8333 			 * find an M_DATA.
8334 			 */
8335 			if ((events & POLLRDDATA) &&
8336 			    mp->b_datap->db_type != M_DATA) {
8337 				mblk_t *nmp = mp->b_cont;
8338 
8339 				while (nmp != NULL &&
8340 				    nmp->b_datap->db_type != M_DATA)
8341 					nmp = nmp->b_cont;
8342 				if (nmp == NULL) {
8343 					mp = mp->b_next;
8344 					continue;
8345 				}
8346 			}
8347 			if (mp->b_band == 0)
8348 				retevents |= normevents;
8349 			else
8350 				retevents |= (events & (POLLIN | POLLRDBAND));
8351 			break;
8352 		}
8353 		if (!(retevents & normevents) && (stp->sd_wakeq & RSLEEP)) {
8354 			/*
8355 			 * Sync stream barrier read queue has data.
8356 			 */
8357 			retevents |= normevents;
8358 		}
8359 		/* Treat eof as normal data */
8360 		if (sd_flags & STREOF)
8361 			retevents |= normevents;
8362 	}
8363 
8364 	/*
8365 	 * Pass back a pollhead if no events are pending or if edge-triggering
8366 	 * has been configured on this resource.
8367 	 */
8368 	if ((retevents == 0 && !anyyet) || (events & POLLET)) {
8369 		*phpp = &stp->sd_pollist;
8370 		if (headlocked == 0) {
8371 			if (polllock(&stp->sd_pollist, &stp->sd_lock) != 0) {
8372 				*reventsp = POLLNVAL;
8373 				return (0);
8374 			}
8375 			headlocked = 1;
8376 		}
8377 		stp->sd_rput_opt |= SR_POLLIN;
8378 	}
8379 
8380 	*reventsp = (short)retevents;
8381 	if (headlocked)
8382 		mutex_exit(&stp->sd_lock);
8383 	return (0);
8384 }
8385 
8386 /*
8387  * The purpose of putback() is to assure sleeping polls/reads
8388  * are awakened when there are no new messages arriving at the,
8389  * stream head, and a message is placed back on the read queue.
8390  *
8391  * sd_lock must be held when messages are placed back on stream
8392  * head.  (getq() holds sd_lock when it removes messages from
8393  * the queue)
8394  */
8395 
8396 static void
8397 putback(struct stdata *stp, queue_t *q, mblk_t *bp, int band)
8398 {
8399 	mblk_t	*qfirst;
8400 	ASSERT(MUTEX_HELD(&stp->sd_lock));
8401 
8402 	/*
8403 	 * As a result of lock-step ordering around q_lock and sd_lock,
8404 	 * it's possible for function calls like putnext() and
8405 	 * canputnext() to get an inaccurate picture of how much
8406 	 * data is really being processed at the stream head.
8407 	 * We only consolidate with existing messages on the queue
8408 	 * if the length of the message we want to put back is smaller
8409 	 * than the queue hiwater mark.
8410 	 */
8411 	if ((stp->sd_rput_opt & SR_CONSOL_DATA) &&
8412 	    (DB_TYPE(bp) == M_DATA) && ((qfirst = q->q_first) != NULL) &&
8413 	    (DB_TYPE(qfirst) == M_DATA) &&
8414 	    ((qfirst->b_flag & (MSGMARK|MSGDELIM)) == 0) &&
8415 	    ((bp->b_flag & (MSGMARK|MSGDELIM|MSGMARKNEXT)) == 0) &&
8416 	    (mp_cont_len(bp, NULL) < q->q_hiwat)) {
8417 		/*
8418 		 * We use the same logic as defined in strrput()
8419 		 * but in reverse as we are putting back onto the
8420 		 * queue and want to retain byte ordering.
8421 		 * Consolidate M_DATA messages with M_DATA ONLY.
8422 		 * strrput() allows the consolidation of M_DATA onto
8423 		 * M_PROTO | M_PCPROTO but not the other way round.
8424 		 *
8425 		 * The consolidation does not take place if the message
8426 		 * we are returning to the queue is marked with either
8427 		 * of the marks or the delim flag or if q_first
8428 		 * is marked with MSGMARK. The MSGMARK check is needed to
8429 		 * handle the odd semantics of MSGMARK where essentially
8430 		 * the whole message is to be treated as marked.
8431 		 * Carry any MSGMARKNEXT and MSGNOTMARKNEXT from q_first
8432 		 * to the front of the b_cont chain.
8433 		 */
8434 		rmvq_noenab(q, qfirst);
8435 
8436 		/*
8437 		 * The first message in the b_cont list
8438 		 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
8439 		 * We need to handle the case where we
8440 		 * are appending:
8441 		 *
8442 		 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
8443 		 * 2) a MSGMARKNEXT to a plain message.
8444 		 * 3) a MSGNOTMARKNEXT to a plain message
8445 		 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
8446 		 *    message.
8447 		 *
8448 		 * Thus we never append a MSGMARKNEXT or
8449 		 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
8450 		 */
8451 		if (qfirst->b_flag & MSGMARKNEXT) {
8452 			bp->b_flag |= MSGMARKNEXT;
8453 			bp->b_flag &= ~MSGNOTMARKNEXT;
8454 			qfirst->b_flag &= ~MSGMARKNEXT;
8455 		} else if (qfirst->b_flag & MSGNOTMARKNEXT) {
8456 			bp->b_flag |= MSGNOTMARKNEXT;
8457 			qfirst->b_flag &= ~MSGNOTMARKNEXT;
8458 		}
8459 
8460 		linkb(bp, qfirst);
8461 	}
8462 	(void) putbq(q, bp);
8463 
8464 	/*
8465 	 * A message may have come in when the sd_lock was dropped in the
8466 	 * calling routine. If this is the case and STR*ATMARK info was
8467 	 * received, need to move that from the stream head to the q_last
8468 	 * so that SIOCATMARK can return the proper value.
8469 	 */
8470 	if (stp->sd_flag & (STRATMARK | STRNOTATMARK)) {
8471 		unsigned short *flagp = &q->q_last->b_flag;
8472 		uint_t b_flag = (uint_t)*flagp;
8473 
8474 		if (stp->sd_flag & STRATMARK) {
8475 			b_flag &= ~MSGNOTMARKNEXT;
8476 			b_flag |= MSGMARKNEXT;
8477 			stp->sd_flag &= ~STRATMARK;
8478 		} else {
8479 			b_flag &= ~MSGMARKNEXT;
8480 			b_flag |= MSGNOTMARKNEXT;
8481 			stp->sd_flag &= ~STRNOTATMARK;
8482 		}
8483 		*flagp = (unsigned short) b_flag;
8484 	}
8485 
8486 #ifdef	DEBUG
8487 	/*
8488 	 * Make sure that the flags are not messed up.
8489 	 */
8490 	{
8491 		mblk_t *mp;
8492 		mp = q->q_last;
8493 		while (mp != NULL) {
8494 			ASSERT((mp->b_flag & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
8495 			    (MSGMARKNEXT|MSGNOTMARKNEXT));
8496 			mp = mp->b_cont;
8497 		}
8498 	}
8499 #endif
8500 	if (q->q_first == bp) {
8501 		short pollevents;
8502 
8503 		if (stp->sd_flag & RSLEEP) {
8504 			stp->sd_flag &= ~RSLEEP;
8505 			cv_broadcast(&q->q_wait);
8506 		}
8507 		if (stp->sd_flag & STRPRI) {
8508 			pollevents = POLLPRI;
8509 		} else {
8510 			if (band == 0) {
8511 				if (!(stp->sd_rput_opt & SR_POLLIN))
8512 					return;
8513 				stp->sd_rput_opt &= ~SR_POLLIN;
8514 				pollevents = POLLIN | POLLRDNORM;
8515 			} else {
8516 				pollevents = POLLIN | POLLRDBAND;
8517 			}
8518 		}
8519 		mutex_exit(&stp->sd_lock);
8520 		pollwakeup(&stp->sd_pollist, pollevents);
8521 		mutex_enter(&stp->sd_lock);
8522 	}
8523 }
8524 
8525 /*
8526  * Return the held vnode attached to the stream head of a
8527  * given queue
8528  * It is the responsibility of the calling routine to ensure
8529  * that the queue does not go away (e.g. pop).
8530  */
8531 vnode_t *
8532 strq2vp(queue_t *qp)
8533 {
8534 	vnode_t *vp;
8535 	vp = STREAM(qp)->sd_vnode;
8536 	ASSERT(vp != NULL);
8537 	VN_HOLD(vp);
8538 	return (vp);
8539 }
8540 
8541 /*
8542  * return the stream head write queue for the given vp
8543  * It is the responsibility of the calling routine to ensure
8544  * that the stream or vnode do not close.
8545  */
8546 queue_t *
8547 strvp2wq(vnode_t *vp)
8548 {
8549 	ASSERT(vp->v_stream != NULL);
8550 	return (vp->v_stream->sd_wrq);
8551 }
8552 
8553 /*
8554  * pollwakeup stream head
8555  * It is the responsibility of the calling routine to ensure
8556  * that the stream or vnode do not close.
8557  */
8558 void
8559 strpollwakeup(vnode_t *vp, short event)
8560 {
8561 	ASSERT(vp->v_stream);
8562 	pollwakeup(&vp->v_stream->sd_pollist, event);
8563 }
8564 
8565 /*
8566  * Mate the stream heads of two vnodes together. If the two vnodes are the
8567  * same, we just make the write-side point at the read-side -- otherwise,
8568  * we do a full mate.  Only works on vnodes associated with streams that are
8569  * still being built and thus have only a stream head.
8570  */
8571 void
8572 strmate(vnode_t *vp1, vnode_t *vp2)
8573 {
8574 	queue_t *wrq1 = strvp2wq(vp1);
8575 	queue_t *wrq2 = strvp2wq(vp2);
8576 
8577 	/*
8578 	 * Verify that there are no modules on the stream yet.  We also
8579 	 * rely on the stream head always having a service procedure to
8580 	 * avoid tweaking q_nfsrv.
8581 	 */
8582 	ASSERT(wrq1->q_next == NULL && wrq2->q_next == NULL);
8583 	ASSERT(wrq1->q_qinfo->qi_srvp != NULL);
8584 	ASSERT(wrq2->q_qinfo->qi_srvp != NULL);
8585 
8586 	/*
8587 	 * If the queues are the same, just twist; otherwise do a full mate.
8588 	 */
8589 	if (wrq1 == wrq2) {
8590 		wrq1->q_next = _RD(wrq1);
8591 	} else {
8592 		wrq1->q_next = _RD(wrq2);
8593 		wrq2->q_next = _RD(wrq1);
8594 		STREAM(wrq1)->sd_mate = STREAM(wrq2);
8595 		STREAM(wrq1)->sd_flag |= STRMATE;
8596 		STREAM(wrq2)->sd_mate = STREAM(wrq1);
8597 		STREAM(wrq2)->sd_flag |= STRMATE;
8598 	}
8599 }
8600 
8601 /*
8602  * XXX will go away when console is correctly fixed.
8603  * Clean up the console PIDS, from previous I_SETSIG,
8604  * called only for cnopen which never calls strclean().
8605  */
8606 void
8607 str_cn_clean(struct vnode *vp)
8608 {
8609 	strsig_t *ssp, *pssp, *tssp;
8610 	struct stdata *stp;
8611 	struct pid  *pidp;
8612 	int update = 0;
8613 
8614 	ASSERT(vp->v_stream);
8615 	stp = vp->v_stream;
8616 	pssp = NULL;
8617 	mutex_enter(&stp->sd_lock);
8618 	ssp = stp->sd_siglist;
8619 	while (ssp) {
8620 		mutex_enter(&pidlock);
8621 		pidp = ssp->ss_pidp;
8622 		/*
8623 		 * Get rid of PID if the proc is gone.
8624 		 */
8625 		if (pidp->pid_prinactive) {
8626 			tssp = ssp->ss_next;
8627 			if (pssp)
8628 				pssp->ss_next = tssp;
8629 			else
8630 				stp->sd_siglist = tssp;
8631 			ASSERT(pidp->pid_ref <= 1);
8632 			PID_RELE(ssp->ss_pidp);
8633 			mutex_exit(&pidlock);
8634 			kmem_free(ssp, sizeof (strsig_t));
8635 			update = 1;
8636 			ssp = tssp;
8637 			continue;
8638 		} else
8639 			mutex_exit(&pidlock);
8640 		pssp = ssp;
8641 		ssp = ssp->ss_next;
8642 	}
8643 	if (update) {
8644 		stp->sd_sigflags = 0;
8645 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
8646 			stp->sd_sigflags |= ssp->ss_events;
8647 	}
8648 	mutex_exit(&stp->sd_lock);
8649 }
8650 
8651 /*
8652  * Return B_TRUE if there is data in the message, B_FALSE otherwise.
8653  */
8654 static boolean_t
8655 msghasdata(mblk_t *bp)
8656 {
8657 	for (; bp; bp = bp->b_cont)
8658 		if (bp->b_datap->db_type == M_DATA) {
8659 			ASSERT(bp->b_wptr >= bp->b_rptr);
8660 			if (bp->b_wptr > bp->b_rptr)
8661 				return (B_TRUE);
8662 		}
8663 	return (B_FALSE);
8664 }
8665