xref: /titanic_51/usr/src/uts/common/os/streamio.c (revision 90bcde942a3919300ffc73f98ea903b58386c395)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*	Copyright (c) 1984, 1986, 1987, 1988, 1989 AT&T	*/
22 /*	  All Rights Reserved  	*/
23 
24 
25 /*
26  * Copyright 2008 Sun Microsystems, Inc.  All rights reserved.
27  * Use is subject to license terms.
28  */
29 
30 #pragma ident	"%Z%%M%	%I%	%E% SMI"
31 
32 #include <sys/types.h>
33 #include <sys/sysmacros.h>
34 #include <sys/param.h>
35 #include <sys/errno.h>
36 #include <sys/signal.h>
37 #include <sys/stat.h>
38 #include <sys/proc.h>
39 #include <sys/cred.h>
40 #include <sys/user.h>
41 #include <sys/vnode.h>
42 #include <sys/file.h>
43 #include <sys/stream.h>
44 #include <sys/strsubr.h>
45 #include <sys/stropts.h>
46 #include <sys/tihdr.h>
47 #include <sys/var.h>
48 #include <sys/poll.h>
49 #include <sys/termio.h>
50 #include <sys/ttold.h>
51 #include <sys/systm.h>
52 #include <sys/uio.h>
53 #include <sys/cmn_err.h>
54 #include <sys/sad.h>
55 #include <sys/netstack.h>
56 #include <sys/priocntl.h>
57 #include <sys/jioctl.h>
58 #include <sys/procset.h>
59 #include <sys/session.h>
60 #include <sys/kmem.h>
61 #include <sys/filio.h>
62 #include <sys/vtrace.h>
63 #include <sys/debug.h>
64 #include <sys/strredir.h>
65 #include <sys/fs/fifonode.h>
66 #include <sys/fs/snode.h>
67 #include <sys/strlog.h>
68 #include <sys/strsun.h>
69 #include <sys/project.h>
70 #include <sys/kbio.h>
71 #include <sys/msio.h>
72 #include <sys/tty.h>
73 #include <sys/ptyvar.h>
74 #include <sys/vuid_event.h>
75 #include <sys/modctl.h>
76 #include <sys/sunddi.h>
77 #include <sys/sunldi_impl.h>
78 #include <sys/autoconf.h>
79 #include <sys/policy.h>
80 #include <sys/dld.h>
81 #include <sys/zone.h>
82 
83 /*
84  * This define helps improve the readability of streams code while
85  * still maintaining a very old streams performance enhancement.  The
86  * performance enhancement basically involved having all callers
87  * of straccess() perform the first check that straccess() will do
88  * locally before actually calling straccess().  (There by reducing
89  * the number of unnecessary calls to straccess().)
90  */
91 #define	i_straccess(x, y)	((stp->sd_sidp == NULL) ? 0 : \
92 				    (stp->sd_vnode->v_type == VFIFO) ? 0 : \
93 				    straccess((x), (y)))
94 
95 /*
96  * what is mblk_pull_len?
97  *
98  * If a streams message consists of many short messages,
99  * a performance degradation occurs from copyout overhead.
100  * To decrease the per mblk overhead, messages that are
101  * likely to consist of many small mblks are pulled up into
102  * one continuous chunk of memory.
103  *
104  * To avoid the processing overhead of examining every
105  * mblk, a quick heuristic is used. If the first mblk in
106  * the message is shorter than mblk_pull_len, it is likely
107  * that the rest of the mblk will be short.
108  *
109  * This heuristic was decided upon after performance tests
110  * indicated that anything more complex slowed down the main
111  * code path.
112  */
113 #define	MBLK_PULL_LEN 64
114 uint32_t mblk_pull_len = MBLK_PULL_LEN;
115 
116 /*
117  * The sgttyb_handling flag controls the handling of the old BSD
118  * TIOCGETP, TIOCSETP, and TIOCSETN ioctls as follows:
119  *
120  * 0 - Emit no warnings at all and retain old, broken behavior.
121  * 1 - Emit no warnings and silently handle new semantics.
122  * 2 - Send cmn_err(CE_NOTE) when either TIOCSETP or TIOCSETN is used
123  *     (once per system invocation).  Handle with new semantics.
124  * 3 - Send SIGSYS when any TIOCGETP, TIOCSETP, or TIOCSETN call is
125  *     made (so that offenders drop core and are easy to debug).
126  *
127  * The "new semantics" are that TIOCGETP returns B38400 for
128  * sg_[io]speed if the corresponding value is over B38400, and that
129  * TIOCSET[PN] accept B38400 in these cases to mean "retain current
130  * bit rate."
131  */
132 int sgttyb_handling = 1;
133 static boolean_t sgttyb_complaint;
134 
135 /* don't push drcompat module by default on Style-2 streams */
136 static int push_drcompat = 0;
137 
138 /*
139  * id value used to distinguish between different ioctl messages
140  */
141 static uint32_t ioc_id;
142 
143 static void putback(struct stdata *, queue_t *, mblk_t *, int);
144 static void strcleanall(struct vnode *);
145 static int strwsrv(queue_t *);
146 static int strdocmd(struct stdata *, struct strcmd *, cred_t *);
147 static void struioainit(queue_t *, sodirect_t *, uio_t *);
148 
149 /*
150  * qinit and module_info structures for stream head read and write queues
151  */
152 struct module_info strm_info = { 0, "strrhead", 0, INFPSZ, STRHIGH, STRLOW };
153 struct module_info stwm_info = { 0, "strwhead", 0, 0, 0, 0 };
154 struct qinit strdata = { strrput, NULL, NULL, NULL, NULL, &strm_info };
155 struct qinit stwdata = { NULL, strwsrv, NULL, NULL, NULL, &stwm_info };
156 struct module_info fiform_info = { 0, "fifostrrhead", 0, PIPE_BUF, FIFOHIWAT,
157     FIFOLOWAT };
158 struct module_info fifowm_info = { 0, "fifostrwhead", 0, 0, 0, 0 };
159 struct qinit fifo_strdata = { strrput, NULL, NULL, NULL, NULL, &fiform_info };
160 struct qinit fifo_stwdata = { NULL, strwsrv, NULL, NULL, NULL, &fifowm_info };
161 
162 extern kmutex_t	strresources;	/* protects global resources */
163 extern kmutex_t muxifier;	/* single-threads multiplexor creation */
164 
165 static boolean_t msghasdata(mblk_t *bp);
166 #define	msgnodata(bp) (!msghasdata(bp))
167 
168 /*
169  * Stream head locking notes:
170  *	There are four monitors associated with the stream head:
171  *	1. v_stream monitor: in stropen() and strclose() v_lock
172  *		is held while the association of vnode and stream
173  *		head is established or tested for.
174  *	2. open/close/push/pop monitor: sd_lock is held while each
175  *		thread bids for exclusive access to this monitor
176  *		for opening or closing a stream.  In addition, this
177  *		monitor is entered during pushes and pops.  This
178  *		guarantees that during plumbing operations there
179  *		is only one thread trying to change the plumbing.
180  *		Any other threads present in the stream are only
181  *		using the plumbing.
182  *	3. read/write monitor: in the case of read, a thread holds
183  *		sd_lock while trying to get data from the stream
184  *		head queue.  if there is none to fulfill a read
185  *		request, it sets RSLEEP and calls cv_wait_sig() down
186  *		in strwaitq() to await the arrival of new data.
187  *		when new data arrives in strrput(), sd_lock is acquired
188  *		before testing for RSLEEP and calling cv_broadcast().
189  *		the behavior of strwrite(), strwsrv(), and WSLEEP
190  *		mirror this.
191  *	4. ioctl monitor: sd_lock is gotten to ensure that only one
192  *		thread is doing an ioctl at a time.
193  *
194  * Note, for sodirect case 3. is extended to (*sodirect_t.sod_enqueue)()
195  * call-back from below, further the sodirect support is for code paths
196  * called via kstgetmsg(), all other code paths ASSERT() that sodirect
197  * uioa generated mblk_t's (i.e. DBLK_UIOA) aren't processed.
198  */
199 
200 static int
201 push_mod(queue_t *qp, dev_t *devp, struct stdata *stp, const char *name,
202     int anchor, cred_t *crp, uint_t anchor_zoneid)
203 {
204 	int error;
205 	fmodsw_impl_t *fp;
206 
207 	if (stp->sd_flag & (STRHUP|STRDERR|STWRERR)) {
208 		error = (stp->sd_flag & STRHUP) ? ENXIO : EIO;
209 		return (error);
210 	}
211 	if (stp->sd_pushcnt >= nstrpush) {
212 		return (EINVAL);
213 	}
214 
215 	if ((fp = fmodsw_find(name, FMODSW_HOLD | FMODSW_LOAD)) == NULL) {
216 		stp->sd_flag |= STREOPENFAIL;
217 		return (EINVAL);
218 	}
219 
220 	/*
221 	 * push new module and call its open routine via qattach
222 	 */
223 	if ((error = qattach(qp, devp, 0, crp, fp, B_FALSE)) != 0)
224 		return (error);
225 
226 	/*
227 	 * Check to see if caller wants a STREAMS anchor
228 	 * put at this place in the stream, and add if so.
229 	 */
230 	mutex_enter(&stp->sd_lock);
231 	if (anchor == stp->sd_pushcnt) {
232 		stp->sd_anchor = stp->sd_pushcnt;
233 		stp->sd_anchorzone = anchor_zoneid;
234 	}
235 	mutex_exit(&stp->sd_lock);
236 
237 	return (0);
238 }
239 
240 /*
241  * Open a stream device.
242  */
243 int
244 stropen(vnode_t *vp, dev_t *devp, int flag, cred_t *crp)
245 {
246 	struct stdata *stp;
247 	queue_t *qp;
248 	int s;
249 	dev_t dummydev, savedev;
250 	struct autopush *ap;
251 	struct dlautopush dlap;
252 	int error = 0;
253 	ssize_t	rmin, rmax;
254 	int cloneopen;
255 	queue_t *brq;
256 	major_t major;
257 	str_stack_t *ss;
258 	zoneid_t zoneid;
259 	uint_t anchor;
260 
261 	if (audit_active)
262 		audit_stropen(vp, devp, flag, crp);
263 
264 	/*
265 	 * If the stream already exists, wait for any open in progress
266 	 * to complete, then call the open function of each module and
267 	 * driver in the stream.  Otherwise create the stream.
268 	 */
269 	TRACE_1(TR_FAC_STREAMS_FR, TR_STROPEN, "stropen:%p", vp);
270 retry:
271 	mutex_enter(&vp->v_lock);
272 	if ((stp = vp->v_stream) != NULL) {
273 
274 		/*
275 		 * Waiting for stream to be created to device
276 		 * due to another open.
277 		 */
278 		mutex_exit(&vp->v_lock);
279 
280 		if (STRMATED(stp)) {
281 			struct stdata *strmatep = stp->sd_mate;
282 
283 			STRLOCKMATES(stp);
284 			if (strmatep->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
285 				if (flag & (FNDELAY|FNONBLOCK)) {
286 					error = EAGAIN;
287 					mutex_exit(&strmatep->sd_lock);
288 					goto ckreturn;
289 				}
290 				mutex_exit(&stp->sd_lock);
291 				if (!cv_wait_sig(&strmatep->sd_monitor,
292 				    &strmatep->sd_lock)) {
293 					error = EINTR;
294 					mutex_exit(&strmatep->sd_lock);
295 					mutex_enter(&stp->sd_lock);
296 					goto ckreturn;
297 				}
298 				mutex_exit(&strmatep->sd_lock);
299 				goto retry;
300 			}
301 			if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
302 				if (flag & (FNDELAY|FNONBLOCK)) {
303 					error = EAGAIN;
304 					mutex_exit(&strmatep->sd_lock);
305 					goto ckreturn;
306 				}
307 				mutex_exit(&strmatep->sd_lock);
308 				if (!cv_wait_sig(&stp->sd_monitor,
309 				    &stp->sd_lock)) {
310 					error = EINTR;
311 					goto ckreturn;
312 				}
313 				mutex_exit(&stp->sd_lock);
314 				goto retry;
315 			}
316 
317 			if (stp->sd_flag & (STRDERR|STWRERR)) {
318 				error = EIO;
319 				mutex_exit(&strmatep->sd_lock);
320 				goto ckreturn;
321 			}
322 
323 			stp->sd_flag |= STWOPEN;
324 			STRUNLOCKMATES(stp);
325 		} else {
326 			mutex_enter(&stp->sd_lock);
327 			if (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
328 				if (flag & (FNDELAY|FNONBLOCK)) {
329 					error = EAGAIN;
330 					goto ckreturn;
331 				}
332 				if (!cv_wait_sig(&stp->sd_monitor,
333 				    &stp->sd_lock)) {
334 					error = EINTR;
335 					goto ckreturn;
336 				}
337 				mutex_exit(&stp->sd_lock);
338 				goto retry;  /* could be clone! */
339 			}
340 
341 			if (stp->sd_flag & (STRDERR|STWRERR)) {
342 				error = EIO;
343 				goto ckreturn;
344 			}
345 
346 			stp->sd_flag |= STWOPEN;
347 			mutex_exit(&stp->sd_lock);
348 		}
349 
350 		/*
351 		 * Open all modules and devices down stream to notify
352 		 * that another user is streaming.  For modules, set the
353 		 * last argument to MODOPEN and do not pass any open flags.
354 		 * Ignore dummydev since this is not the first open.
355 		 */
356 		claimstr(stp->sd_wrq);
357 		qp = stp->sd_wrq;
358 		while (_SAMESTR(qp)) {
359 			qp = qp->q_next;
360 			if ((error = qreopen(_RD(qp), devp, flag, crp)) != 0)
361 				break;
362 		}
363 		releasestr(stp->sd_wrq);
364 		mutex_enter(&stp->sd_lock);
365 		stp->sd_flag &= ~(STRHUP|STWOPEN|STRDERR|STWRERR);
366 		stp->sd_rerror = 0;
367 		stp->sd_werror = 0;
368 ckreturn:
369 		cv_broadcast(&stp->sd_monitor);
370 		mutex_exit(&stp->sd_lock);
371 		return (error);
372 	}
373 
374 	/*
375 	 * This vnode isn't streaming.  SPECFS already
376 	 * checked for multiple vnodes pointing to the
377 	 * same stream, so create a stream to the driver.
378 	 */
379 	qp = allocq();
380 	stp = shalloc(qp);
381 
382 	/*
383 	 * Initialize stream head.  shalloc() has given us
384 	 * exclusive access, and we have the vnode locked;
385 	 * we can do whatever we want with stp.
386 	 */
387 	stp->sd_flag = STWOPEN;
388 	stp->sd_siglist = NULL;
389 	stp->sd_pollist.ph_list = NULL;
390 	stp->sd_sigflags = 0;
391 	stp->sd_mark = NULL;
392 	stp->sd_closetime = STRTIMOUT;
393 	stp->sd_sidp = NULL;
394 	stp->sd_pgidp = NULL;
395 	stp->sd_vnode = vp;
396 	stp->sd_rerror = 0;
397 	stp->sd_werror = 0;
398 	stp->sd_wroff = 0;
399 	stp->sd_tail = 0;
400 	stp->sd_iocblk = NULL;
401 	stp->sd_cmdblk = NULL;
402 	stp->sd_pushcnt = 0;
403 	stp->sd_qn_minpsz = 0;
404 	stp->sd_qn_maxpsz = INFPSZ - 1;	/* used to check for initialization */
405 	stp->sd_maxblk = INFPSZ;
406 	stp->sd_sodirect = NULL;
407 	qp->q_ptr = _WR(qp)->q_ptr = stp;
408 	STREAM(qp) = STREAM(_WR(qp)) = stp;
409 	vp->v_stream = stp;
410 	mutex_exit(&vp->v_lock);
411 	if (vp->v_type == VFIFO) {
412 		stp->sd_flag |= OLDNDELAY;
413 		/*
414 		 * This means, both for pipes and fifos
415 		 * strwrite will send SIGPIPE if the other
416 		 * end is closed. For putmsg it depends
417 		 * on whether it is a XPG4_2 application
418 		 * or not
419 		 */
420 		stp->sd_wput_opt = SW_SIGPIPE;
421 
422 		/* setq might sleep in kmem_alloc - avoid holding locks. */
423 		setq(qp, &fifo_strdata, &fifo_stwdata, NULL, QMTSAFE,
424 		    SQ_CI|SQ_CO, B_FALSE);
425 
426 		set_qend(qp);
427 		stp->sd_strtab = fifo_getinfo();
428 		_WR(qp)->q_nfsrv = _WR(qp);
429 		qp->q_nfsrv = qp;
430 		/*
431 		 * Wake up others that are waiting for stream to be created.
432 		 */
433 		mutex_enter(&stp->sd_lock);
434 		/*
435 		 * nothing is be pushed on stream yet, so
436 		 * optimized stream head packetsizes are just that
437 		 * of the read queue
438 		 */
439 		stp->sd_qn_minpsz = qp->q_minpsz;
440 		stp->sd_qn_maxpsz = qp->q_maxpsz;
441 		stp->sd_flag &= ~STWOPEN;
442 		goto fifo_opendone;
443 	}
444 	/* setq might sleep in kmem_alloc - avoid holding locks. */
445 	setq(qp, &strdata, &stwdata, NULL, QMTSAFE, SQ_CI|SQ_CO, B_FALSE);
446 
447 	set_qend(qp);
448 
449 	/*
450 	 * Open driver and create stream to it (via qattach).
451 	 */
452 	savedev = *devp;
453 	cloneopen = (getmajor(*devp) == clone_major);
454 	if ((error = qattach(qp, devp, flag, crp, NULL, B_FALSE)) != 0) {
455 		mutex_enter(&vp->v_lock);
456 		vp->v_stream = NULL;
457 		mutex_exit(&vp->v_lock);
458 		mutex_enter(&stp->sd_lock);
459 		cv_broadcast(&stp->sd_monitor);
460 		mutex_exit(&stp->sd_lock);
461 		freeq(_RD(qp));
462 		shfree(stp);
463 		return (error);
464 	}
465 	/*
466 	 * Set sd_strtab after open in order to handle clonable drivers
467 	 */
468 	stp->sd_strtab = STREAMSTAB(getmajor(*devp));
469 
470 	/*
471 	 * Historical note: dummydev used to be be prior to the initial
472 	 * open (via qattach above), which made the value seen
473 	 * inconsistent between an I_PUSH and an autopush of a module.
474 	 */
475 	dummydev = *devp;
476 
477 	/*
478 	 * For clone open of old style (Q not associated) network driver,
479 	 * push DRMODNAME module to handle DL_ATTACH/DL_DETACH
480 	 */
481 	brq = _RD(_WR(qp)->q_next);
482 	major = getmajor(*devp);
483 	if (push_drcompat && cloneopen && NETWORK_DRV(major) &&
484 	    ((brq->q_flag & _QASSOCIATED) == 0)) {
485 		if (push_mod(qp, &dummydev, stp, DRMODNAME, 0, crp, 0) != 0)
486 			cmn_err(CE_WARN, "cannot push " DRMODNAME
487 			    " streams module");
488 	}
489 
490 	if (!NETWORK_DRV(major)) {
491 		savedev = *devp;
492 	} else {
493 		/*
494 		 * For network devices, process differently based on the
495 		 * return value from dld_autopush():
496 		 *
497 		 *   0: the passed-in device points to a GLDv3 datalink with
498 		 *   per-link autopush configuration; use that configuration
499 		 *   and ignore any per-driver autopush configuration.
500 		 *
501 		 *   1: the passed-in device points to a physical GLDv3
502 		 *   datalink without per-link autopush configuration.  The
503 		 *   passed in device was changed to refer to the actual
504 		 *   physical device (if it's not already); we use that new
505 		 *   device to look up any per-driver autopush configuration.
506 		 *
507 		 *   -1: neither of the above cases applied; use the initial
508 		 *   device to look up any per-driver autopush configuration.
509 		 */
510 		switch (dld_autopush(&savedev, &dlap)) {
511 		case 0:
512 			zoneid = crgetzoneid(crp);
513 			for (s = 0; s < dlap.dap_npush; s++) {
514 				error = push_mod(qp, &dummydev, stp,
515 				    dlap.dap_aplist[s], dlap.dap_anchor, crp,
516 				    zoneid);
517 				if (error != 0)
518 					break;
519 			}
520 			goto opendone;
521 		case 1:
522 			break;
523 		case -1:
524 			savedev = *devp;
525 			break;
526 		}
527 	}
528 	/*
529 	 * Find the autopush configuration based on "savedev". Start with the
530 	 * global zone. If not found check in the local zone.
531 	 */
532 	zoneid = GLOBAL_ZONEID;
533 retryap:
534 	ss = netstack_find_by_stackid(zoneid_to_netstackid(zoneid))->
535 	    netstack_str;
536 	if ((ap = sad_ap_find_by_dev(savedev, ss)) == NULL) {
537 		netstack_rele(ss->ss_netstack);
538 		if (zoneid == GLOBAL_ZONEID) {
539 			/*
540 			 * None found. Also look in the zone's autopush table.
541 			 */
542 			zoneid = crgetzoneid(crp);
543 			if (zoneid != GLOBAL_ZONEID)
544 				goto retryap;
545 		}
546 		goto opendone;
547 	}
548 	anchor = ap->ap_anchor;
549 	zoneid = crgetzoneid(crp);
550 	for (s = 0; s < ap->ap_npush; s++) {
551 		error = push_mod(qp, &dummydev, stp, ap->ap_list[s],
552 		    anchor, crp, zoneid);
553 		if (error != 0)
554 			break;
555 	}
556 	sad_ap_rele(ap, ss);
557 	netstack_rele(ss->ss_netstack);
558 
559 opendone:
560 
561 	/*
562 	 * let specfs know that open failed part way through
563 	 */
564 	if (error) {
565 		mutex_enter(&stp->sd_lock);
566 		stp->sd_flag |= STREOPENFAIL;
567 		mutex_exit(&stp->sd_lock);
568 	}
569 
570 	/*
571 	 * Wake up others that are waiting for stream to be created.
572 	 */
573 	mutex_enter(&stp->sd_lock);
574 	stp->sd_flag &= ~STWOPEN;
575 
576 	/*
577 	 * As a performance concern we are caching the values of
578 	 * q_minpsz and q_maxpsz of the module below the stream
579 	 * head in the stream head.
580 	 */
581 	mutex_enter(QLOCK(stp->sd_wrq->q_next));
582 	rmin = stp->sd_wrq->q_next->q_minpsz;
583 	rmax = stp->sd_wrq->q_next->q_maxpsz;
584 	mutex_exit(QLOCK(stp->sd_wrq->q_next));
585 
586 	/* do this processing here as a performance concern */
587 	if (strmsgsz != 0) {
588 		if (rmax == INFPSZ)
589 			rmax = strmsgsz;
590 		else
591 			rmax = MIN(strmsgsz, rmax);
592 	}
593 
594 	mutex_enter(QLOCK(stp->sd_wrq));
595 	stp->sd_qn_minpsz = rmin;
596 	stp->sd_qn_maxpsz = rmax;
597 	mutex_exit(QLOCK(stp->sd_wrq));
598 
599 fifo_opendone:
600 	cv_broadcast(&stp->sd_monitor);
601 	mutex_exit(&stp->sd_lock);
602 	return (error);
603 }
604 
605 static int strsink(queue_t *, mblk_t *);
606 static struct qinit deadrend = {
607 	strsink, NULL, NULL, NULL, NULL, &strm_info, NULL
608 };
609 static struct qinit deadwend = {
610 	NULL, NULL, NULL, NULL, NULL, &stwm_info, NULL
611 };
612 
613 /*
614  * Close a stream.
615  * This is called from closef() on the last close of an open stream.
616  * Strclean() will already have removed the siglist and pollist
617  * information, so all that remains is to remove all multiplexor links
618  * for the stream, pop all the modules (and the driver), and free the
619  * stream structure.
620  */
621 
622 int
623 strclose(struct vnode *vp, int flag, cred_t *crp)
624 {
625 	struct stdata *stp;
626 	queue_t *qp;
627 	int rval;
628 	int freestp = 1;
629 	queue_t *rmq;
630 
631 	if (audit_active)
632 		audit_strclose(vp, flag, crp);
633 
634 	TRACE_1(TR_FAC_STREAMS_FR,
635 	    TR_STRCLOSE, "strclose:%p", vp);
636 	ASSERT(vp->v_stream);
637 
638 	stp = vp->v_stream;
639 	ASSERT(!(stp->sd_flag & STPLEX));
640 	qp = stp->sd_wrq;
641 
642 	/*
643 	 * Needed so that strpoll will return non-zero for this fd.
644 	 * Note that with POLLNOERR STRHUP does still cause POLLHUP.
645 	 */
646 	mutex_enter(&stp->sd_lock);
647 	stp->sd_flag |= STRHUP;
648 	mutex_exit(&stp->sd_lock);
649 
650 	/*
651 	 * If the registered process or process group did not have an
652 	 * open instance of this stream then strclean would not be
653 	 * called. Thus at the time of closing all remaining siglist entries
654 	 * are removed.
655 	 */
656 	if (stp->sd_siglist != NULL)
657 		strcleanall(vp);
658 
659 	ASSERT(stp->sd_siglist == NULL);
660 	ASSERT(stp->sd_sigflags == 0);
661 
662 	if (STRMATED(stp)) {
663 		struct stdata *strmatep = stp->sd_mate;
664 		int waited = 1;
665 
666 		STRLOCKMATES(stp);
667 		while (waited) {
668 			waited = 0;
669 			while (stp->sd_flag & (STWOPEN|STRCLOSE|STRPLUMB)) {
670 				mutex_exit(&strmatep->sd_lock);
671 				cv_wait(&stp->sd_monitor, &stp->sd_lock);
672 				mutex_exit(&stp->sd_lock);
673 				STRLOCKMATES(stp);
674 				waited = 1;
675 			}
676 			while (strmatep->sd_flag &
677 			    (STWOPEN|STRCLOSE|STRPLUMB)) {
678 				mutex_exit(&stp->sd_lock);
679 				cv_wait(&strmatep->sd_monitor,
680 				    &strmatep->sd_lock);
681 				mutex_exit(&strmatep->sd_lock);
682 				STRLOCKMATES(stp);
683 				waited = 1;
684 			}
685 		}
686 		stp->sd_flag |= STRCLOSE;
687 		STRUNLOCKMATES(stp);
688 	} else {
689 		mutex_enter(&stp->sd_lock);
690 		stp->sd_flag |= STRCLOSE;
691 		mutex_exit(&stp->sd_lock);
692 	}
693 
694 	ASSERT(qp->q_first == NULL);	/* No more delayed write */
695 
696 	/* Check if an I_LINK was ever done on this stream */
697 	if (stp->sd_flag & STRHASLINKS) {
698 		netstack_t *ns;
699 		str_stack_t *ss;
700 
701 		ns = netstack_find_by_cred(crp);
702 		ASSERT(ns != NULL);
703 		ss = ns->netstack_str;
704 		ASSERT(ss != NULL);
705 
706 		(void) munlinkall(stp, LINKCLOSE|LINKNORMAL, crp, &rval, ss);
707 		netstack_rele(ss->ss_netstack);
708 	}
709 
710 	while (_SAMESTR(qp)) {
711 		/*
712 		 * Holding sd_lock prevents q_next from changing in
713 		 * this stream.
714 		 */
715 		mutex_enter(&stp->sd_lock);
716 		if (!(flag & (FNDELAY|FNONBLOCK)) && (stp->sd_closetime > 0)) {
717 
718 			/*
719 			 * sleep until awakened by strwsrv() or timeout
720 			 */
721 			for (;;) {
722 				mutex_enter(QLOCK(qp->q_next));
723 				if (!(qp->q_next->q_mblkcnt)) {
724 					mutex_exit(QLOCK(qp->q_next));
725 					break;
726 				}
727 				stp->sd_flag |= WSLEEP;
728 
729 				/* ensure strwsrv gets enabled */
730 				qp->q_next->q_flag |= QWANTW;
731 				mutex_exit(QLOCK(qp->q_next));
732 				/* get out if we timed out or recv'd a signal */
733 				if (str_cv_wait(&qp->q_wait, &stp->sd_lock,
734 				    stp->sd_closetime, 0) <= 0) {
735 					break;
736 				}
737 			}
738 			stp->sd_flag &= ~WSLEEP;
739 		}
740 		mutex_exit(&stp->sd_lock);
741 
742 		rmq = qp->q_next;
743 		if (rmq->q_flag & QISDRV) {
744 			ASSERT(!_SAMESTR(rmq));
745 			wait_sq_svc(_RD(qp)->q_syncq);
746 		}
747 
748 		qdetach(_RD(rmq), 1, flag, crp, B_FALSE);
749 	}
750 
751 	/*
752 	 * Since we call pollwakeup in close() now, the poll list should
753 	 * be empty in most cases. The only exception is the layered devices
754 	 * (e.g. the console drivers with redirection modules pushed on top
755 	 * of it).  We have to do this after calling qdetach() because
756 	 * the redirection module won't have torn down the console
757 	 * redirection until after qdetach() has been invoked.
758 	 */
759 	if (stp->sd_pollist.ph_list != NULL) {
760 		pollwakeup(&stp->sd_pollist, POLLERR);
761 		pollhead_clean(&stp->sd_pollist);
762 	}
763 	ASSERT(stp->sd_pollist.ph_list == NULL);
764 	ASSERT(stp->sd_sidp == NULL);
765 	ASSERT(stp->sd_pgidp == NULL);
766 
767 	/* Prevent qenable from re-enabling the stream head queue */
768 	disable_svc(_RD(qp));
769 
770 	/*
771 	 * Wait until service procedure of each queue is
772 	 * run, if QINSERVICE is set.
773 	 */
774 	wait_svc(_RD(qp));
775 
776 	/*
777 	 * Now, flush both queues.
778 	 */
779 	flushq(_RD(qp), FLUSHALL);
780 	flushq(qp, FLUSHALL);
781 
782 	/*
783 	 * If the write queue of the stream head is pointing to a
784 	 * read queue, we have a twisted stream.  If the read queue
785 	 * is alive, convert the stream head queues into a dead end.
786 	 * If the read queue is dead, free the dead pair.
787 	 */
788 	if (qp->q_next && !_SAMESTR(qp)) {
789 		if (qp->q_next->q_qinfo == &deadrend) {	/* half-closed pipe */
790 			flushq(qp->q_next, FLUSHALL); /* ensure no message */
791 			shfree(qp->q_next->q_stream);
792 			freeq(qp->q_next);
793 			freeq(_RD(qp));
794 		} else if (qp->q_next == _RD(qp)) {	/* fifo */
795 			freeq(_RD(qp));
796 		} else {				/* pipe */
797 			freestp = 0;
798 			/*
799 			 * The q_info pointers are never accessed when
800 			 * SQLOCK is held.
801 			 */
802 			ASSERT(qp->q_syncq == _RD(qp)->q_syncq);
803 			mutex_enter(SQLOCK(qp->q_syncq));
804 			qp->q_qinfo = &deadwend;
805 			_RD(qp)->q_qinfo = &deadrend;
806 			mutex_exit(SQLOCK(qp->q_syncq));
807 		}
808 	} else {
809 		freeq(_RD(qp)); /* free stream head queue pair */
810 	}
811 
812 	mutex_enter(&vp->v_lock);
813 	if (stp->sd_iocblk) {
814 		if (stp->sd_iocblk != (mblk_t *)-1) {
815 			freemsg(stp->sd_iocblk);
816 		}
817 		stp->sd_iocblk = NULL;
818 	}
819 	stp->sd_vnode = NULL;
820 	vp->v_stream = NULL;
821 	mutex_exit(&vp->v_lock);
822 	mutex_enter(&stp->sd_lock);
823 	freemsg(stp->sd_cmdblk);
824 	stp->sd_cmdblk = NULL;
825 	stp->sd_flag &= ~STRCLOSE;
826 	cv_broadcast(&stp->sd_monitor);
827 	mutex_exit(&stp->sd_lock);
828 
829 	if (freestp)
830 		shfree(stp);
831 	return (0);
832 }
833 
834 static int
835 strsink(queue_t *q, mblk_t *bp)
836 {
837 	struct copyresp *resp;
838 
839 	switch (bp->b_datap->db_type) {
840 	case M_FLUSH:
841 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
842 			*bp->b_rptr &= ~FLUSHR;
843 			bp->b_flag |= MSGNOLOOP;
844 			/*
845 			 * Protect against the driver passing up
846 			 * messages after it has done a qprocsoff.
847 			 */
848 			if (_OTHERQ(q)->q_next == NULL)
849 				freemsg(bp);
850 			else
851 				qreply(q, bp);
852 		} else {
853 			freemsg(bp);
854 		}
855 		break;
856 
857 	case M_COPYIN:
858 	case M_COPYOUT:
859 		if (bp->b_cont) {
860 			freemsg(bp->b_cont);
861 			bp->b_cont = NULL;
862 		}
863 		bp->b_datap->db_type = M_IOCDATA;
864 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
865 		resp = (struct copyresp *)bp->b_rptr;
866 		resp->cp_rval = (caddr_t)1;	/* failure */
867 		/*
868 		 * Protect against the driver passing up
869 		 * messages after it has done a qprocsoff.
870 		 */
871 		if (_OTHERQ(q)->q_next == NULL)
872 			freemsg(bp);
873 		else
874 			qreply(q, bp);
875 		break;
876 
877 	case M_IOCTL:
878 		if (bp->b_cont) {
879 			freemsg(bp->b_cont);
880 			bp->b_cont = NULL;
881 		}
882 		bp->b_datap->db_type = M_IOCNAK;
883 		/*
884 		 * Protect against the driver passing up
885 		 * messages after it has done a qprocsoff.
886 		 */
887 		if (_OTHERQ(q)->q_next == NULL)
888 			freemsg(bp);
889 		else
890 			qreply(q, bp);
891 		break;
892 
893 	default:
894 		freemsg(bp);
895 		break;
896 	}
897 
898 	return (0);
899 }
900 
901 /*
902  * Clean up after a process when it closes a stream.  This is called
903  * from closef for all closes, whereas strclose is called only for the
904  * last close on a stream.  The siglist is scanned for entries for the
905  * current process, and these are removed.
906  */
907 void
908 strclean(struct vnode *vp)
909 {
910 	strsig_t *ssp, *pssp, *tssp;
911 	stdata_t *stp;
912 	int update = 0;
913 
914 	TRACE_1(TR_FAC_STREAMS_FR,
915 	    TR_STRCLEAN, "strclean:%p", vp);
916 	stp = vp->v_stream;
917 	pssp = NULL;
918 	mutex_enter(&stp->sd_lock);
919 	ssp = stp->sd_siglist;
920 	while (ssp) {
921 		if (ssp->ss_pidp == curproc->p_pidp) {
922 			tssp = ssp->ss_next;
923 			if (pssp)
924 				pssp->ss_next = tssp;
925 			else
926 				stp->sd_siglist = tssp;
927 			mutex_enter(&pidlock);
928 			PID_RELE(ssp->ss_pidp);
929 			mutex_exit(&pidlock);
930 			kmem_free(ssp, sizeof (strsig_t));
931 			update = 1;
932 			ssp = tssp;
933 		} else {
934 			pssp = ssp;
935 			ssp = ssp->ss_next;
936 		}
937 	}
938 	if (update) {
939 		stp->sd_sigflags = 0;
940 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
941 			stp->sd_sigflags |= ssp->ss_events;
942 	}
943 	mutex_exit(&stp->sd_lock);
944 }
945 
946 /*
947  * Used on the last close to remove any remaining items on the siglist.
948  * These could be present on the siglist due to I_ESETSIG calls that
949  * use process groups or processed that do not have an open file descriptor
950  * for this stream (Such entries would not be removed by strclean).
951  */
952 static void
953 strcleanall(struct vnode *vp)
954 {
955 	strsig_t *ssp, *nssp;
956 	stdata_t *stp;
957 
958 	stp = vp->v_stream;
959 	mutex_enter(&stp->sd_lock);
960 	ssp = stp->sd_siglist;
961 	stp->sd_siglist = NULL;
962 	while (ssp) {
963 		nssp = ssp->ss_next;
964 		mutex_enter(&pidlock);
965 		PID_RELE(ssp->ss_pidp);
966 		mutex_exit(&pidlock);
967 		kmem_free(ssp, sizeof (strsig_t));
968 		ssp = nssp;
969 	}
970 	stp->sd_sigflags = 0;
971 	mutex_exit(&stp->sd_lock);
972 }
973 
974 /*
975  * Retrieve the next message from the logical stream head read queue
976  * using either rwnext (if sync stream) or getq_noenab.
977  * It is the callers responsibility to call qbackenable after
978  * it is finished with the message. The caller should not call
979  * qbackenable until after any putback calls to avoid spurious backenabling.
980  *
981  * Also, handle uioa initialization and process any DBLK_UIOA flaged messages.
982  */
983 mblk_t *
984 strget(struct stdata *stp, queue_t *q, struct uio *uiop, int first,
985     int *errorp)
986 {
987 	sodirect_t *sodp = stp->sd_sodirect;
988 	mblk_t *bp;
989 	int error;
990 	ssize_t rbytes = 0;
991 
992 	/* Holding sd_lock prevents the read queue from changing  */
993 	ASSERT(MUTEX_HELD(&stp->sd_lock));
994 
995 	if (uiop != NULL && stp->sd_struiordq != NULL &&
996 	    q->q_first == NULL &&
997 	    (!first || (stp->sd_wakeq & RSLEEP))) {
998 		/*
999 		 * Stream supports rwnext() for the read side.
1000 		 * If this is the first time we're called by e.g. strread
1001 		 * only do the downcall if there is a deferred wakeup
1002 		 * (registered in sd_wakeq).
1003 		 */
1004 		struiod_t uiod;
1005 
1006 		if (first)
1007 			stp->sd_wakeq &= ~RSLEEP;
1008 
1009 		(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
1010 		    sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
1011 		uiod.d_mp = 0;
1012 		/*
1013 		 * Mark that a thread is in rwnext on the read side
1014 		 * to prevent strrput from nacking ioctls immediately.
1015 		 * When the last concurrent rwnext returns
1016 		 * the ioctls are nack'ed.
1017 		 */
1018 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1019 		stp->sd_struiodnak++;
1020 		/*
1021 		 * Note: rwnext will drop sd_lock.
1022 		 */
1023 		error = rwnext(q, &uiod);
1024 		ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
1025 		mutex_enter(&stp->sd_lock);
1026 		stp->sd_struiodnak--;
1027 		while (stp->sd_struiodnak == 0 &&
1028 		    ((bp = stp->sd_struionak) != NULL)) {
1029 			stp->sd_struionak = bp->b_next;
1030 			bp->b_next = NULL;
1031 			bp->b_datap->db_type = M_IOCNAK;
1032 			/*
1033 			 * Protect against the driver passing up
1034 			 * messages after it has done a qprocsoff.
1035 			 */
1036 			if (_OTHERQ(q)->q_next == NULL)
1037 				freemsg(bp);
1038 			else {
1039 				mutex_exit(&stp->sd_lock);
1040 				qreply(q, bp);
1041 				mutex_enter(&stp->sd_lock);
1042 			}
1043 		}
1044 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1045 		if (error == 0 || error == EWOULDBLOCK) {
1046 			if ((bp = uiod.d_mp) != NULL) {
1047 				*errorp = 0;
1048 				ASSERT(MUTEX_HELD(&stp->sd_lock));
1049 				return (bp);
1050 			}
1051 			error = 0;
1052 		} else if (error == EINVAL) {
1053 			/*
1054 			 * The stream plumbing must have
1055 			 * changed while we were away, so
1056 			 * just turn off rwnext()s.
1057 			 */
1058 			error = 0;
1059 		} else if (error == EBUSY) {
1060 			/*
1061 			 * The module might have data in transit using putnext
1062 			 * Fall back on waiting + getq.
1063 			 */
1064 			error = 0;
1065 		} else {
1066 			*errorp = error;
1067 			ASSERT(MUTEX_HELD(&stp->sd_lock));
1068 			return (NULL);
1069 		}
1070 		/*
1071 		 * Try a getq in case a rwnext() generated mblk
1072 		 * has bubbled up via strrput().
1073 		 */
1074 	}
1075 	*errorp = 0;
1076 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1077 
1078 	if (sodp != NULL && (sodp->sod_state & SOD_ENABLED) &&
1079 	    (sodp->sod_uioa.uioa_state & UIOA_INIT)) {
1080 		/*
1081 		 * First kstrgetmsg() call for an uioa_t so if any
1082 		 * queued mblk_t's need to consume them before uioa
1083 		 * from below can occur.
1084 		 */
1085 		sodp->sod_uioa.uioa_state &= UIOA_CLR;
1086 		sodp->sod_uioa.uioa_state |= UIOA_ENABLED;
1087 		if (q->q_first != NULL) {
1088 			struioainit(q, sodp, uiop);
1089 		}
1090 	} else {
1091 		/*
1092 		 * If we have a valid uio, try and use this as a guide for how
1093 		 * many bytes to retrieve from the queue via getq_noenab().
1094 		 * Doing this can avoid unneccesary counting of overlong
1095 		 * messages in putback(). We currently only do this for sockets.
1096 		 */
1097 		if ((uiop != NULL) && (stp->sd_vnode->v_type == VSOCK))
1098 			rbytes = uiop->uio_resid;
1099 	}
1100 
1101 	bp = getq_noenab(q, rbytes);
1102 	if (bp != NULL && (bp->b_datap->db_flags & DBLK_UIOA)) {
1103 		/*
1104 		 * A uioa flaged mblk_t chain, already uio processed,
1105 		 * add it to the sodirect uioa pending free list.
1106 		 *
1107 		 * Note, a b_cont chain headed by a DBLK_UIOA enable
1108 		 * mblk_t must have all mblk_t(s) DBLK_UIOA enabled.
1109 		 */
1110 		mblk_t	*bpt = sodp->sod_uioaft;
1111 
1112 		ASSERT(sodp != NULL);
1113 
1114 		/*
1115 		 * Add first mblk_t of "bp" chain to current sodirect uioa
1116 		 * free list tail mblk_t, if any, else empty list so new head.
1117 		 */
1118 		if (bpt == NULL)
1119 			sodp->sod_uioafh = bp;
1120 		else
1121 			bpt->b_cont = bp;
1122 
1123 		/*
1124 		 * Walk mblk_t "bp" chain to find tail and adjust rptr of
1125 		 * each to reflect that uioamove() has consumed all data.
1126 		 */
1127 		bpt = bp;
1128 		for (;;) {
1129 			bpt->b_rptr = bpt->b_wptr;
1130 			if (bpt->b_cont == NULL)
1131 				break;
1132 			bpt = bpt->b_cont;
1133 
1134 			ASSERT(bpt->b_datap->db_flags & DBLK_UIOA);
1135 		}
1136 		/* New sodirect uioa free list tail */
1137 		sodp->sod_uioaft = bpt;
1138 
1139 		/* Only 1 strget() with data returned per uioa_t */
1140 		if (sodp->sod_uioa.uioa_state & UIOA_ENABLED) {
1141 			sodp->sod_uioa.uioa_state &= UIOA_CLR;
1142 			sodp->sod_uioa.uioa_state |= UIOA_FINI;
1143 		}
1144 	}
1145 
1146 	return (bp);
1147 }
1148 
1149 /*
1150  * Copy out the message pointed to by `bp' into the uio pointed to by `uiop'.
1151  * If the message does not fit in the uio the remainder of it is returned;
1152  * otherwise NULL is returned.  Any embedded zero-length mblk_t's are
1153  * consumed, even if uio_resid reaches zero.  On error, `*errorp' is set to
1154  * the error code, the message is consumed, and NULL is returned.
1155  */
1156 static mblk_t *
1157 struiocopyout(mblk_t *bp, struct uio *uiop, int *errorp)
1158 {
1159 	int error;
1160 	ptrdiff_t n;
1161 	mblk_t *nbp;
1162 
1163 	ASSERT(bp->b_wptr >= bp->b_rptr);
1164 
1165 	do {
1166 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
1167 
1168 		if ((n = MIN(uiop->uio_resid, MBLKL(bp))) != 0) {
1169 			ASSERT(n > 0);
1170 
1171 			error = uiomove(bp->b_rptr, n, UIO_READ, uiop);
1172 			if (error != 0) {
1173 				freemsg(bp);
1174 				*errorp = error;
1175 				return (NULL);
1176 			}
1177 		}
1178 
1179 		bp->b_rptr += n;
1180 		while (bp != NULL && (bp->b_rptr >= bp->b_wptr)) {
1181 			nbp = bp;
1182 			bp = bp->b_cont;
1183 			freeb(nbp);
1184 		}
1185 	} while (bp != NULL && uiop->uio_resid > 0);
1186 
1187 	*errorp = 0;
1188 	return (bp);
1189 }
1190 
1191 /*
1192  * Read a stream according to the mode flags in sd_flag:
1193  *
1194  * (default mode)		- Byte stream, msg boundaries are ignored
1195  * RD_MSGDIS (msg discard)	- Read on msg boundaries and throw away
1196  *				any data remaining in msg
1197  * RD_MSGNODIS (msg non-discard) - Read on msg boundaries and put back
1198  *				any remaining data on head of read queue
1199  *
1200  * Consume readable messages on the front of the queue until
1201  * ttolwp(curthread)->lwp_count
1202  * is satisfied, the readable messages are exhausted, or a message
1203  * boundary is reached in a message mode.  If no data was read and
1204  * the stream was not opened with the NDELAY flag, block until data arrives.
1205  * Otherwise return the data read and update the count.
1206  *
1207  * In default mode a 0 length message signifies end-of-file and terminates
1208  * a read in progress.  The 0 length message is removed from the queue
1209  * only if it is the only message read (no data is read).
1210  *
1211  * An attempt to read an M_PROTO or M_PCPROTO message results in an
1212  * EBADMSG error return, unless either RD_PROTDAT or RD_PROTDIS are set.
1213  * If RD_PROTDAT is set, M_PROTO and M_PCPROTO messages are read as data.
1214  * If RD_PROTDIS is set, the M_PROTO and M_PCPROTO parts of the message
1215  * are unlinked from and M_DATA blocks in the message, the protos are
1216  * thrown away, and the data is read.
1217  */
1218 /* ARGSUSED */
1219 int
1220 strread(struct vnode *vp, struct uio *uiop, cred_t *crp)
1221 {
1222 	struct stdata *stp;
1223 	mblk_t *bp, *nbp;
1224 	queue_t *q;
1225 	int error = 0;
1226 	uint_t old_sd_flag;
1227 	int first;
1228 	char rflg;
1229 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
1230 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
1231 	short delim;
1232 	unsigned char pri = 0;
1233 	char waitflag;
1234 	unsigned char type;
1235 
1236 	TRACE_1(TR_FAC_STREAMS_FR,
1237 	    TR_STRREAD_ENTER, "strread:%p", vp);
1238 	ASSERT(vp->v_stream);
1239 	stp = vp->v_stream;
1240 
1241 	mutex_enter(&stp->sd_lock);
1242 
1243 	if ((error = i_straccess(stp, JCREAD)) != 0) {
1244 		mutex_exit(&stp->sd_lock);
1245 		return (error);
1246 	}
1247 
1248 	if (stp->sd_flag & (STRDERR|STPLEX)) {
1249 		error = strgeterr(stp, STRDERR|STPLEX, 0);
1250 		if (error != 0) {
1251 			mutex_exit(&stp->sd_lock);
1252 			return (error);
1253 		}
1254 	}
1255 
1256 	/*
1257 	 * Loop terminates when uiop->uio_resid == 0.
1258 	 */
1259 	rflg = 0;
1260 	waitflag = READWAIT;
1261 	q = _RD(stp->sd_wrq);
1262 	for (;;) {
1263 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1264 		old_sd_flag = stp->sd_flag;
1265 		mark = 0;
1266 		delim = 0;
1267 		first = 1;
1268 		while ((bp = strget(stp, q, uiop, first, &error)) == NULL) {
1269 			int done = 0;
1270 
1271 			ASSERT(MUTEX_HELD(&stp->sd_lock));
1272 
1273 			if (error != 0)
1274 				goto oops;
1275 
1276 			if (stp->sd_flag & (STRHUP|STREOF)) {
1277 				goto oops;
1278 			}
1279 			if (rflg && !(stp->sd_flag & STRDELIM)) {
1280 				goto oops;
1281 			}
1282 			/*
1283 			 * If a read(fd,buf,0) has been done, there is no
1284 			 * need to sleep. We always have zero bytes to
1285 			 * return.
1286 			 */
1287 			if (uiop->uio_resid == 0) {
1288 				goto oops;
1289 			}
1290 
1291 			qbackenable(q, 0);
1292 
1293 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_WAIT,
1294 			    "strread calls strwaitq:%p, %p, %p",
1295 			    vp, uiop, crp);
1296 			if ((error = strwaitq(stp, waitflag, uiop->uio_resid,
1297 			    uiop->uio_fmode, -1, &done)) != 0 || done) {
1298 				TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_DONE,
1299 				    "strread error or done:%p, %p, %p",
1300 				    vp, uiop, crp);
1301 				if ((uiop->uio_fmode & FNDELAY) &&
1302 				    (stp->sd_flag & OLDNDELAY) &&
1303 				    (error == EAGAIN))
1304 					error = 0;
1305 				goto oops;
1306 			}
1307 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRREAD_AWAKE,
1308 			    "strread awakes:%p, %p, %p", vp, uiop, crp);
1309 			if ((error = i_straccess(stp, JCREAD)) != 0) {
1310 				goto oops;
1311 			}
1312 			first = 0;
1313 		}
1314 
1315 		ASSERT(MUTEX_HELD(&stp->sd_lock));
1316 		ASSERT(bp);
1317 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
1318 		pri = bp->b_band;
1319 		/*
1320 		 * Extract any mark information. If the message is not
1321 		 * completely consumed this information will be put in the mblk
1322 		 * that is putback.
1323 		 * If MSGMARKNEXT is set and the message is completely consumed
1324 		 * the STRATMARK flag will be set below. Likewise, if
1325 		 * MSGNOTMARKNEXT is set and the message is
1326 		 * completely consumed STRNOTATMARK will be set.
1327 		 *
1328 		 * For some unknown reason strread only breaks the read at the
1329 		 * last mark.
1330 		 */
1331 		mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
1332 		ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
1333 		    (MSGMARKNEXT|MSGNOTMARKNEXT));
1334 		if (mark != 0 && bp == stp->sd_mark) {
1335 			if (rflg) {
1336 				putback(stp, q, bp, pri);
1337 				goto oops;
1338 			}
1339 			mark |= _LASTMARK;
1340 			stp->sd_mark = NULL;
1341 		}
1342 		if ((stp->sd_flag & STRDELIM) && (bp->b_flag & MSGDELIM))
1343 			delim = 1;
1344 		mutex_exit(&stp->sd_lock);
1345 
1346 		if (STREAM_NEEDSERVICE(stp))
1347 			stream_runservice(stp);
1348 
1349 		type = bp->b_datap->db_type;
1350 
1351 		switch (type) {
1352 
1353 		case M_DATA:
1354 ismdata:
1355 			if (msgnodata(bp)) {
1356 				if (mark || delim) {
1357 					freemsg(bp);
1358 				} else if (rflg) {
1359 
1360 					/*
1361 					 * If already read data put zero
1362 					 * length message back on queue else
1363 					 * free msg and return 0.
1364 					 */
1365 					bp->b_band = pri;
1366 					mutex_enter(&stp->sd_lock);
1367 					putback(stp, q, bp, pri);
1368 					mutex_exit(&stp->sd_lock);
1369 				} else {
1370 					freemsg(bp);
1371 				}
1372 				error =  0;
1373 				goto oops1;
1374 			}
1375 
1376 			rflg = 1;
1377 			waitflag |= NOINTR;
1378 			bp = struiocopyout(bp, uiop, &error);
1379 			if (error != 0)
1380 				goto oops1;
1381 
1382 			mutex_enter(&stp->sd_lock);
1383 			if (bp) {
1384 				/*
1385 				 * Have remaining data in message.
1386 				 * Free msg if in discard mode.
1387 				 */
1388 				if (stp->sd_read_opt & RD_MSGDIS) {
1389 					freemsg(bp);
1390 				} else {
1391 					bp->b_band = pri;
1392 					if ((mark & _LASTMARK) &&
1393 					    (stp->sd_mark == NULL))
1394 						stp->sd_mark = bp;
1395 					bp->b_flag |= mark & ~_LASTMARK;
1396 					if (delim)
1397 						bp->b_flag |= MSGDELIM;
1398 					if (msgnodata(bp))
1399 						freemsg(bp);
1400 					else
1401 						putback(stp, q, bp, pri);
1402 				}
1403 			} else {
1404 				/*
1405 				 * Consumed the complete message.
1406 				 * Move the MSG*MARKNEXT information
1407 				 * to the stream head just in case
1408 				 * the read queue becomes empty.
1409 				 *
1410 				 * If the stream head was at the mark
1411 				 * (STRATMARK) before we dropped sd_lock above
1412 				 * and some data was consumed then we have
1413 				 * moved past the mark thus STRATMARK is
1414 				 * cleared. However, if a message arrived in
1415 				 * strrput during the copyout above causing
1416 				 * STRATMARK to be set we can not clear that
1417 				 * flag.
1418 				 */
1419 				if (mark &
1420 				    (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
1421 					if (mark & MSGMARKNEXT) {
1422 						stp->sd_flag &= ~STRNOTATMARK;
1423 						stp->sd_flag |= STRATMARK;
1424 					} else if (mark & MSGNOTMARKNEXT) {
1425 						stp->sd_flag &= ~STRATMARK;
1426 						stp->sd_flag |= STRNOTATMARK;
1427 					} else {
1428 						stp->sd_flag &=
1429 						    ~(STRATMARK|STRNOTATMARK);
1430 					}
1431 				} else if (rflg && (old_sd_flag & STRATMARK)) {
1432 					stp->sd_flag &= ~STRATMARK;
1433 				}
1434 			}
1435 
1436 			/*
1437 			 * Check for signal messages at the front of the read
1438 			 * queue and generate the signal(s) if appropriate.
1439 			 * The only signal that can be on queue is M_SIG at
1440 			 * this point.
1441 			 */
1442 			while ((((bp = q->q_first)) != NULL) &&
1443 			    (bp->b_datap->db_type == M_SIG)) {
1444 				bp = getq_noenab(q, 0);
1445 				/*
1446 				 * sd_lock is held so the content of the
1447 				 * read queue can not change.
1448 				 */
1449 				ASSERT(bp != NULL &&
1450 				    bp->b_datap->db_type == M_SIG);
1451 				strsignal_nolock(stp, *bp->b_rptr,
1452 				    (int32_t)bp->b_band);
1453 				mutex_exit(&stp->sd_lock);
1454 				freemsg(bp);
1455 				if (STREAM_NEEDSERVICE(stp))
1456 					stream_runservice(stp);
1457 				mutex_enter(&stp->sd_lock);
1458 			}
1459 
1460 			if ((uiop->uio_resid == 0) || (mark & _LASTMARK) ||
1461 			    delim ||
1462 			    (stp->sd_read_opt & (RD_MSGDIS|RD_MSGNODIS))) {
1463 				goto oops;
1464 			}
1465 			continue;
1466 
1467 		case M_SIG:
1468 			strsignal(stp, *bp->b_rptr, (int32_t)bp->b_band);
1469 			freemsg(bp);
1470 			mutex_enter(&stp->sd_lock);
1471 			continue;
1472 
1473 		case M_PROTO:
1474 		case M_PCPROTO:
1475 			/*
1476 			 * Only data messages are readable.
1477 			 * Any others generate an error, unless
1478 			 * RD_PROTDIS or RD_PROTDAT is set.
1479 			 */
1480 			if (stp->sd_read_opt & RD_PROTDAT) {
1481 				for (nbp = bp; nbp; nbp = nbp->b_next) {
1482 					if ((nbp->b_datap->db_type ==
1483 					    M_PROTO) ||
1484 					    (nbp->b_datap->db_type ==
1485 					    M_PCPROTO)) {
1486 						nbp->b_datap->db_type = M_DATA;
1487 					} else {
1488 						break;
1489 					}
1490 				}
1491 				/*
1492 				 * clear stream head hi pri flag based on
1493 				 * first message
1494 				 */
1495 				if (type == M_PCPROTO) {
1496 					mutex_enter(&stp->sd_lock);
1497 					stp->sd_flag &= ~STRPRI;
1498 					mutex_exit(&stp->sd_lock);
1499 				}
1500 				goto ismdata;
1501 			} else if (stp->sd_read_opt & RD_PROTDIS) {
1502 				/*
1503 				 * discard non-data messages
1504 				 */
1505 				while (bp &&
1506 				    ((bp->b_datap->db_type == M_PROTO) ||
1507 				    (bp->b_datap->db_type == M_PCPROTO))) {
1508 					nbp = unlinkb(bp);
1509 					freeb(bp);
1510 					bp = nbp;
1511 				}
1512 				/*
1513 				 * clear stream head hi pri flag based on
1514 				 * first message
1515 				 */
1516 				if (type == M_PCPROTO) {
1517 					mutex_enter(&stp->sd_lock);
1518 					stp->sd_flag &= ~STRPRI;
1519 					mutex_exit(&stp->sd_lock);
1520 				}
1521 				if (bp) {
1522 					bp->b_band = pri;
1523 					goto ismdata;
1524 				} else {
1525 					break;
1526 				}
1527 			}
1528 			/* FALLTHRU */
1529 		case M_PASSFP:
1530 			if ((bp->b_datap->db_type == M_PASSFP) &&
1531 			    (stp->sd_read_opt & RD_PROTDIS)) {
1532 				freemsg(bp);
1533 				break;
1534 			}
1535 			mutex_enter(&stp->sd_lock);
1536 			putback(stp, q, bp, pri);
1537 			mutex_exit(&stp->sd_lock);
1538 			if (rflg == 0)
1539 				error = EBADMSG;
1540 			goto oops1;
1541 
1542 		default:
1543 			/*
1544 			 * Garbage on stream head read queue.
1545 			 */
1546 			cmn_err(CE_WARN, "bad %x found at stream head\n",
1547 			    bp->b_datap->db_type);
1548 			freemsg(bp);
1549 			goto oops1;
1550 		}
1551 		mutex_enter(&stp->sd_lock);
1552 	}
1553 oops:
1554 	mutex_exit(&stp->sd_lock);
1555 oops1:
1556 	qbackenable(q, pri);
1557 	return (error);
1558 #undef	_LASTMARK
1559 }
1560 
1561 /*
1562  * Default processing of M_PROTO/M_PCPROTO messages.
1563  * Determine which wakeups and signals are needed.
1564  * This can be replaced by a user-specified procedure for kernel users
1565  * of STREAMS.
1566  */
1567 /* ARGSUSED */
1568 mblk_t *
1569 strrput_proto(vnode_t *vp, mblk_t *mp,
1570     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1571     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1572 {
1573 	*wakeups = RSLEEP;
1574 	*allmsgsigs = 0;
1575 
1576 	switch (mp->b_datap->db_type) {
1577 	case M_PROTO:
1578 		if (mp->b_band == 0) {
1579 			*firstmsgsigs = S_INPUT | S_RDNORM;
1580 			*pollwakeups = POLLIN | POLLRDNORM;
1581 		} else {
1582 			*firstmsgsigs = S_INPUT | S_RDBAND;
1583 			*pollwakeups = POLLIN | POLLRDBAND;
1584 		}
1585 		break;
1586 	case M_PCPROTO:
1587 		*firstmsgsigs = S_HIPRI;
1588 		*pollwakeups = POLLPRI;
1589 		break;
1590 	}
1591 	return (mp);
1592 }
1593 
1594 /*
1595  * Default processing of everything but M_DATA, M_PROTO, M_PCPROTO and
1596  * M_PASSFP messages.
1597  * Determine which wakeups and signals are needed.
1598  * This can be replaced by a user-specified procedure for kernel users
1599  * of STREAMS.
1600  */
1601 /* ARGSUSED */
1602 mblk_t *
1603 strrput_misc(vnode_t *vp, mblk_t *mp,
1604     strwakeup_t *wakeups, strsigset_t *firstmsgsigs,
1605     strsigset_t *allmsgsigs, strpollset_t *pollwakeups)
1606 {
1607 	*wakeups = 0;
1608 	*firstmsgsigs = 0;
1609 	*allmsgsigs = 0;
1610 	*pollwakeups = 0;
1611 	return (mp);
1612 }
1613 
1614 /*
1615  * Stream read put procedure.  Called from downstream driver/module
1616  * with messages for the stream head.  Data, protocol, and in-stream
1617  * signal messages are placed on the queue, others are handled directly.
1618  */
1619 int
1620 strrput(queue_t *q, mblk_t *bp)
1621 {
1622 	struct stdata	*stp;
1623 	ulong_t		rput_opt;
1624 	strwakeup_t	wakeups;
1625 	strsigset_t	firstmsgsigs;	/* Signals if first message on queue */
1626 	strsigset_t	allmsgsigs;	/* Signals for all messages */
1627 	strsigset_t	signals;	/* Signals events to generate */
1628 	strpollset_t	pollwakeups;
1629 	mblk_t		*nextbp;
1630 	uchar_t		band = 0;
1631 	int		hipri_sig;
1632 
1633 	stp = (struct stdata *)q->q_ptr;
1634 	/*
1635 	 * Use rput_opt for optimized access to the SR_ flags except
1636 	 * SR_POLLIN. That flag has to be checked under sd_lock since it
1637 	 * is modified by strpoll().
1638 	 */
1639 	rput_opt = stp->sd_rput_opt;
1640 
1641 	ASSERT(qclaimed(q));
1642 	TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_ENTER,
1643 	    "strrput called with message type:q %p bp %p", q, bp);
1644 
1645 	/*
1646 	 * Perform initial processing and pass to the parameterized functions.
1647 	 */
1648 	ASSERT(bp->b_next == NULL);
1649 
1650 	switch (bp->b_datap->db_type) {
1651 	case M_DATA:
1652 		/*
1653 		 * sockfs is the only consumer of STREOF and when it is set,
1654 		 * it implies that the receiver is not interested in receiving
1655 		 * any more data, hence the mblk is freed to prevent unnecessary
1656 		 * message queueing at the stream head.
1657 		 */
1658 		if (stp->sd_flag == STREOF) {
1659 			freemsg(bp);
1660 			return (0);
1661 		}
1662 		if ((rput_opt & SR_IGN_ZEROLEN) &&
1663 		    bp->b_rptr == bp->b_wptr && msgnodata(bp)) {
1664 			/*
1665 			 * Ignore zero-length M_DATA messages. These might be
1666 			 * generated by some transports.
1667 			 * The zero-length M_DATA messages, even if they
1668 			 * are ignored, should effect the atmark tracking and
1669 			 * should wake up a thread sleeping in strwaitmark.
1670 			 */
1671 			mutex_enter(&stp->sd_lock);
1672 			if (bp->b_flag & MSGMARKNEXT) {
1673 				/*
1674 				 * Record the position of the mark either
1675 				 * in q_last or in STRATMARK.
1676 				 */
1677 				if (q->q_last != NULL) {
1678 					q->q_last->b_flag &= ~MSGNOTMARKNEXT;
1679 					q->q_last->b_flag |= MSGMARKNEXT;
1680 				} else {
1681 					stp->sd_flag &= ~STRNOTATMARK;
1682 					stp->sd_flag |= STRATMARK;
1683 				}
1684 			} else if (bp->b_flag & MSGNOTMARKNEXT) {
1685 				/*
1686 				 * Record that this is not the position of
1687 				 * the mark either in q_last or in
1688 				 * STRNOTATMARK.
1689 				 */
1690 				if (q->q_last != NULL) {
1691 					q->q_last->b_flag &= ~MSGMARKNEXT;
1692 					q->q_last->b_flag |= MSGNOTMARKNEXT;
1693 				} else {
1694 					stp->sd_flag &= ~STRATMARK;
1695 					stp->sd_flag |= STRNOTATMARK;
1696 				}
1697 			}
1698 			if (stp->sd_flag & RSLEEP) {
1699 				stp->sd_flag &= ~RSLEEP;
1700 				cv_broadcast(&q->q_wait);
1701 			}
1702 			mutex_exit(&stp->sd_lock);
1703 			freemsg(bp);
1704 			return (0);
1705 		}
1706 		wakeups = RSLEEP;
1707 		if (bp->b_band == 0) {
1708 			firstmsgsigs = S_INPUT | S_RDNORM;
1709 			pollwakeups = POLLIN | POLLRDNORM;
1710 		} else {
1711 			firstmsgsigs = S_INPUT | S_RDBAND;
1712 			pollwakeups = POLLIN | POLLRDBAND;
1713 		}
1714 		if (rput_opt & SR_SIGALLDATA)
1715 			allmsgsigs = firstmsgsigs;
1716 		else
1717 			allmsgsigs = 0;
1718 
1719 		mutex_enter(&stp->sd_lock);
1720 		if ((rput_opt & SR_CONSOL_DATA) &&
1721 		    (q->q_last != NULL) &&
1722 		    (bp->b_flag & (MSGMARK|MSGDELIM)) == 0) {
1723 			/*
1724 			 * Consolidate an M_DATA message onto an M_DATA,
1725 			 * M_PROTO, or M_PCPROTO by merging it with q_last.
1726 			 * The consolidation does not take place if
1727 			 * the old message is marked with either of the
1728 			 * marks or the delim flag or if the new
1729 			 * message is marked with MSGMARK. The MSGMARK
1730 			 * check is needed to handle the odd semantics of
1731 			 * MSGMARK where essentially the whole message
1732 			 * is to be treated as marked.
1733 			 * Carry any MSGMARKNEXT  and MSGNOTMARKNEXT from the
1734 			 * new message to the front of the b_cont chain.
1735 			 */
1736 			mblk_t *lbp = q->q_last;
1737 			unsigned char db_type = lbp->b_datap->db_type;
1738 
1739 			if ((db_type == M_DATA || db_type == M_PROTO ||
1740 			    db_type == M_PCPROTO) &&
1741 			    !(lbp->b_flag & (MSGDELIM|MSGMARK|MSGMARKNEXT))) {
1742 				rmvq_noenab(q, lbp);
1743 				/*
1744 				 * The first message in the b_cont list
1745 				 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
1746 				 * We need to handle the case where we
1747 				 * are appending:
1748 				 *
1749 				 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
1750 				 * 2) a MSGMARKNEXT to a plain message.
1751 				 * 3) a MSGNOTMARKNEXT to a plain message
1752 				 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
1753 				 *    message.
1754 				 *
1755 				 * Thus we never append a MSGMARKNEXT or
1756 				 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
1757 				 */
1758 				if (bp->b_flag & MSGMARKNEXT) {
1759 					lbp->b_flag |= MSGMARKNEXT;
1760 					lbp->b_flag &= ~MSGNOTMARKNEXT;
1761 					bp->b_flag &= ~MSGMARKNEXT;
1762 				} else if (bp->b_flag & MSGNOTMARKNEXT) {
1763 					lbp->b_flag |= MSGNOTMARKNEXT;
1764 					bp->b_flag &= ~MSGNOTMARKNEXT;
1765 				}
1766 
1767 				linkb(lbp, bp);
1768 				bp = lbp;
1769 				/*
1770 				 * The new message logically isn't the first
1771 				 * even though the q_first check below thinks
1772 				 * it is. Clear the firstmsgsigs to make it
1773 				 * not appear to be first.
1774 				 */
1775 				firstmsgsigs = 0;
1776 			}
1777 		}
1778 		break;
1779 
1780 	case M_PASSFP:
1781 		wakeups = RSLEEP;
1782 		allmsgsigs = 0;
1783 		if (bp->b_band == 0) {
1784 			firstmsgsigs = S_INPUT | S_RDNORM;
1785 			pollwakeups = POLLIN | POLLRDNORM;
1786 		} else {
1787 			firstmsgsigs = S_INPUT | S_RDBAND;
1788 			pollwakeups = POLLIN | POLLRDBAND;
1789 		}
1790 		mutex_enter(&stp->sd_lock);
1791 		break;
1792 
1793 	case M_PROTO:
1794 	case M_PCPROTO:
1795 		ASSERT(stp->sd_rprotofunc != NULL);
1796 		bp = (stp->sd_rprotofunc)(stp->sd_vnode, bp,
1797 		    &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1798 #define	ALLSIG	(S_INPUT|S_HIPRI|S_OUTPUT|S_MSG|S_ERROR|S_HANGUP|S_RDNORM|\
1799 		S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)
1800 #define	ALLPOLL	(POLLIN|POLLPRI|POLLOUT|POLLRDNORM|POLLWRNORM|POLLRDBAND|\
1801 		POLLWRBAND)
1802 
1803 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1804 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1805 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1806 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1807 
1808 		mutex_enter(&stp->sd_lock);
1809 		break;
1810 
1811 	default:
1812 		ASSERT(stp->sd_rmiscfunc != NULL);
1813 		bp = (stp->sd_rmiscfunc)(stp->sd_vnode, bp,
1814 		    &wakeups, &firstmsgsigs, &allmsgsigs, &pollwakeups);
1815 		ASSERT((wakeups & ~(RSLEEP|WSLEEP)) == 0);
1816 		ASSERT((firstmsgsigs & ~ALLSIG) == 0);
1817 		ASSERT((allmsgsigs & ~ALLSIG) == 0);
1818 		ASSERT((pollwakeups & ~ALLPOLL) == 0);
1819 #undef	ALLSIG
1820 #undef	ALLPOLL
1821 		mutex_enter(&stp->sd_lock);
1822 		break;
1823 	}
1824 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1825 
1826 	/* By default generate superset of signals */
1827 	signals = (firstmsgsigs | allmsgsigs);
1828 
1829 	/*
1830 	 * The  proto and misc functions can return multiple messages
1831 	 * as a b_next chain. Such messages are processed separately.
1832 	 */
1833 one_more:
1834 	hipri_sig = 0;
1835 	if (bp == NULL) {
1836 		nextbp = NULL;
1837 	} else {
1838 		nextbp = bp->b_next;
1839 		bp->b_next = NULL;
1840 
1841 		switch (bp->b_datap->db_type) {
1842 		case M_PCPROTO:
1843 			/*
1844 			 * Only one priority protocol message is allowed at the
1845 			 * stream head at a time.
1846 			 */
1847 			if (stp->sd_flag & STRPRI) {
1848 				TRACE_0(TR_FAC_STREAMS_FR, TR_STRRPUT_PROTERR,
1849 				    "M_PCPROTO already at head");
1850 				freemsg(bp);
1851 				mutex_exit(&stp->sd_lock);
1852 				goto done;
1853 			}
1854 			stp->sd_flag |= STRPRI;
1855 			hipri_sig = 1;
1856 			/* FALLTHRU */
1857 		case M_DATA:
1858 		case M_PROTO:
1859 		case M_PASSFP:
1860 			band = bp->b_band;
1861 			/*
1862 			 * Marking doesn't work well when messages
1863 			 * are marked in more than one band.  We only
1864 			 * remember the last message received, even if
1865 			 * it is placed on the queue ahead of other
1866 			 * marked messages.
1867 			 */
1868 			if (bp->b_flag & MSGMARK)
1869 				stp->sd_mark = bp;
1870 			(void) putq(q, bp);
1871 
1872 			/*
1873 			 * If message is a PCPROTO message, always use
1874 			 * firstmsgsigs to determine if a signal should be
1875 			 * sent as strrput is the only place to send
1876 			 * signals for PCPROTO. Other messages are based on
1877 			 * the STRGETINPROG flag. The flag determines if
1878 			 * strrput or (k)strgetmsg will be responsible for
1879 			 * sending the signals, in the firstmsgsigs case.
1880 			 */
1881 			if ((hipri_sig == 1) ||
1882 			    (((stp->sd_flag & STRGETINPROG) == 0) &&
1883 			    (q->q_first == bp)))
1884 				signals = (firstmsgsigs | allmsgsigs);
1885 			else
1886 				signals = allmsgsigs;
1887 			break;
1888 
1889 		default:
1890 			mutex_exit(&stp->sd_lock);
1891 			(void) strrput_nondata(q, bp);
1892 			mutex_enter(&stp->sd_lock);
1893 			break;
1894 		}
1895 	}
1896 	ASSERT(MUTEX_HELD(&stp->sd_lock));
1897 	/*
1898 	 * Wake sleeping read/getmsg and cancel deferred wakeup
1899 	 */
1900 	if (wakeups & RSLEEP)
1901 		stp->sd_wakeq &= ~RSLEEP;
1902 
1903 	wakeups &= stp->sd_flag;
1904 	if (wakeups & RSLEEP) {
1905 		stp->sd_flag &= ~RSLEEP;
1906 		cv_broadcast(&q->q_wait);
1907 	}
1908 	if (wakeups & WSLEEP) {
1909 		stp->sd_flag &= ~WSLEEP;
1910 		cv_broadcast(&_WR(q)->q_wait);
1911 	}
1912 
1913 	if (pollwakeups != 0) {
1914 		if (pollwakeups == (POLLIN | POLLRDNORM)) {
1915 			/*
1916 			 * Can't use rput_opt since it was not
1917 			 * read when sd_lock was held and SR_POLLIN is changed
1918 			 * by strpoll() under sd_lock.
1919 			 */
1920 			if (!(stp->sd_rput_opt & SR_POLLIN))
1921 				goto no_pollwake;
1922 			stp->sd_rput_opt &= ~SR_POLLIN;
1923 		}
1924 		mutex_exit(&stp->sd_lock);
1925 		pollwakeup(&stp->sd_pollist, pollwakeups);
1926 		mutex_enter(&stp->sd_lock);
1927 	}
1928 no_pollwake:
1929 
1930 	/*
1931 	 * strsendsig can handle multiple signals with a
1932 	 * single call.
1933 	 */
1934 	if (stp->sd_sigflags & signals)
1935 		strsendsig(stp->sd_siglist, signals, band, 0);
1936 	mutex_exit(&stp->sd_lock);
1937 
1938 
1939 done:
1940 	if (nextbp == NULL)
1941 		return (0);
1942 
1943 	/*
1944 	 * Any signals were handled the first time.
1945 	 * Wakeups and pollwakeups are redone to avoid any race
1946 	 * conditions - all the messages are not queued until the
1947 	 * last message has been processed by strrput.
1948 	 */
1949 	bp = nextbp;
1950 	signals = firstmsgsigs = allmsgsigs = 0;
1951 	mutex_enter(&stp->sd_lock);
1952 	goto one_more;
1953 }
1954 
1955 static void
1956 log_dupioc(queue_t *rq, mblk_t *bp)
1957 {
1958 	queue_t *wq, *qp;
1959 	char *modnames, *mnp, *dname;
1960 	size_t maxmodstr;
1961 	boolean_t islast;
1962 
1963 	/*
1964 	 * Allocate a buffer large enough to hold the names of nstrpush modules
1965 	 * and one driver, with spaces between and NUL terminator.  If we can't
1966 	 * get memory, then we'll just log the driver name.
1967 	 */
1968 	maxmodstr = nstrpush * (FMNAMESZ + 1);
1969 	mnp = modnames = kmem_alloc(maxmodstr, KM_NOSLEEP);
1970 
1971 	/* march down write side to print log message down to the driver */
1972 	wq = WR(rq);
1973 
1974 	/* make sure q_next doesn't shift around while we're grabbing data */
1975 	claimstr(wq);
1976 	qp = wq->q_next;
1977 	do {
1978 		if ((dname = qp->q_qinfo->qi_minfo->mi_idname) == NULL)
1979 			dname = "?";
1980 		islast = !SAMESTR(qp) || qp->q_next == NULL;
1981 		if (modnames == NULL) {
1982 			/*
1983 			 * If we don't have memory, then get the driver name in
1984 			 * the log where we can see it.  Note that memory
1985 			 * pressure is a possible cause of these sorts of bugs.
1986 			 */
1987 			if (islast) {
1988 				modnames = dname;
1989 				maxmodstr = 0;
1990 			}
1991 		} else {
1992 			mnp += snprintf(mnp, FMNAMESZ + 1, "%s", dname);
1993 			if (!islast)
1994 				*mnp++ = ' ';
1995 		}
1996 		qp = qp->q_next;
1997 	} while (!islast);
1998 	releasestr(wq);
1999 	/* Cannot happen unless stream head is corrupt. */
2000 	ASSERT(modnames != NULL);
2001 	(void) strlog(rq->q_qinfo->qi_minfo->mi_idnum, 0, 1,
2002 	    SL_CONSOLE|SL_TRACE|SL_ERROR,
2003 	    "Warning: stream %p received duplicate %X M_IOC%s; module list: %s",
2004 	    rq->q_ptr, ((struct iocblk *)bp->b_rptr)->ioc_cmd,
2005 	    (DB_TYPE(bp) == M_IOCACK ? "ACK" : "NAK"), modnames);
2006 	if (maxmodstr != 0)
2007 		kmem_free(modnames, maxmodstr);
2008 }
2009 
2010 int
2011 strrput_nondata(queue_t *q, mblk_t *bp)
2012 {
2013 	struct stdata *stp;
2014 	struct iocblk *iocbp;
2015 	struct stroptions *sop;
2016 	struct copyreq *reqp;
2017 	struct copyresp *resp;
2018 	unsigned char bpri;
2019 	unsigned char  flushed_already = 0;
2020 
2021 	stp = (struct stdata *)q->q_ptr;
2022 
2023 	ASSERT(!(stp->sd_flag & STPLEX));
2024 	ASSERT(qclaimed(q));
2025 
2026 	switch (bp->b_datap->db_type) {
2027 	case M_ERROR:
2028 		/*
2029 		 * An error has occurred downstream, the errno is in the first
2030 		 * bytes of the message.
2031 		 */
2032 		if ((bp->b_wptr - bp->b_rptr) == 2) {	/* New flavor */
2033 			unsigned char rw = 0;
2034 
2035 			mutex_enter(&stp->sd_lock);
2036 			if (*bp->b_rptr != NOERROR) {	/* read error */
2037 				if (*bp->b_rptr != 0) {
2038 					if (stp->sd_flag & STRDERR)
2039 						flushed_already |= FLUSHR;
2040 					stp->sd_flag |= STRDERR;
2041 					rw |= FLUSHR;
2042 				} else {
2043 					stp->sd_flag &= ~STRDERR;
2044 				}
2045 				stp->sd_rerror = *bp->b_rptr;
2046 			}
2047 			bp->b_rptr++;
2048 			if (*bp->b_rptr != NOERROR) {	/* write error */
2049 				if (*bp->b_rptr != 0) {
2050 					if (stp->sd_flag & STWRERR)
2051 						flushed_already |= FLUSHW;
2052 					stp->sd_flag |= STWRERR;
2053 					rw |= FLUSHW;
2054 				} else {
2055 					stp->sd_flag &= ~STWRERR;
2056 				}
2057 				stp->sd_werror = *bp->b_rptr;
2058 			}
2059 			if (rw) {
2060 				TRACE_2(TR_FAC_STREAMS_FR, TR_STRRPUT_WAKE,
2061 				    "strrput cv_broadcast:q %p, bp %p",
2062 				    q, bp);
2063 				cv_broadcast(&q->q_wait); /* readers */
2064 				cv_broadcast(&_WR(q)->q_wait); /* writers */
2065 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
2066 
2067 				mutex_exit(&stp->sd_lock);
2068 				pollwakeup(&stp->sd_pollist, POLLERR);
2069 				mutex_enter(&stp->sd_lock);
2070 
2071 				if (stp->sd_sigflags & S_ERROR)
2072 					strsendsig(stp->sd_siglist, S_ERROR, 0,
2073 					    ((rw & FLUSHR) ? stp->sd_rerror :
2074 					    stp->sd_werror));
2075 				mutex_exit(&stp->sd_lock);
2076 				/*
2077 				 * Send the M_FLUSH only
2078 				 * for the first M_ERROR
2079 				 * message on the stream
2080 				 */
2081 				if (flushed_already == rw) {
2082 					freemsg(bp);
2083 					return (0);
2084 				}
2085 
2086 				bp->b_datap->db_type = M_FLUSH;
2087 				*bp->b_rptr = rw;
2088 				bp->b_wptr = bp->b_rptr + 1;
2089 				/*
2090 				 * Protect against the driver
2091 				 * passing up messages after
2092 				 * it has done a qprocsoff
2093 				 */
2094 				if (_OTHERQ(q)->q_next == NULL)
2095 					freemsg(bp);
2096 				else
2097 					qreply(q, bp);
2098 				return (0);
2099 			} else
2100 				mutex_exit(&stp->sd_lock);
2101 		} else if (*bp->b_rptr != 0) {		/* Old flavor */
2102 				if (stp->sd_flag & (STRDERR|STWRERR))
2103 					flushed_already = FLUSHRW;
2104 				mutex_enter(&stp->sd_lock);
2105 				stp->sd_flag |= (STRDERR|STWRERR);
2106 				stp->sd_rerror = *bp->b_rptr;
2107 				stp->sd_werror = *bp->b_rptr;
2108 				TRACE_2(TR_FAC_STREAMS_FR,
2109 				    TR_STRRPUT_WAKE2,
2110 				    "strrput wakeup #2:q %p, bp %p", q, bp);
2111 				cv_broadcast(&q->q_wait); /* the readers */
2112 				cv_broadcast(&_WR(q)->q_wait); /* the writers */
2113 				cv_broadcast(&stp->sd_monitor); /* ioctllers */
2114 
2115 				mutex_exit(&stp->sd_lock);
2116 				pollwakeup(&stp->sd_pollist, POLLERR);
2117 				mutex_enter(&stp->sd_lock);
2118 
2119 				if (stp->sd_sigflags & S_ERROR)
2120 					strsendsig(stp->sd_siglist, S_ERROR, 0,
2121 					    (stp->sd_werror ? stp->sd_werror :
2122 					    stp->sd_rerror));
2123 				mutex_exit(&stp->sd_lock);
2124 
2125 				/*
2126 				 * Send the M_FLUSH only
2127 				 * for the first M_ERROR
2128 				 * message on the stream
2129 				 */
2130 				if (flushed_already != FLUSHRW) {
2131 					bp->b_datap->db_type = M_FLUSH;
2132 					*bp->b_rptr = FLUSHRW;
2133 					/*
2134 					 * Protect against the driver passing up
2135 					 * messages after it has done a
2136 					 * qprocsoff.
2137 					 */
2138 				if (_OTHERQ(q)->q_next == NULL)
2139 					freemsg(bp);
2140 				else
2141 					qreply(q, bp);
2142 				return (0);
2143 				}
2144 		}
2145 		freemsg(bp);
2146 		return (0);
2147 
2148 	case M_HANGUP:
2149 
2150 		freemsg(bp);
2151 		mutex_enter(&stp->sd_lock);
2152 		stp->sd_werror = ENXIO;
2153 		stp->sd_flag |= STRHUP;
2154 		stp->sd_flag &= ~(WSLEEP|RSLEEP);
2155 
2156 		/*
2157 		 * send signal if controlling tty
2158 		 */
2159 
2160 		if (stp->sd_sidp) {
2161 			prsignal(stp->sd_sidp, SIGHUP);
2162 			if (stp->sd_sidp != stp->sd_pgidp)
2163 				pgsignal(stp->sd_pgidp, SIGTSTP);
2164 		}
2165 
2166 		/*
2167 		 * wake up read, write, and exception pollers and
2168 		 * reset wakeup mechanism.
2169 		 */
2170 		cv_broadcast(&q->q_wait);	/* the readers */
2171 		cv_broadcast(&_WR(q)->q_wait);	/* the writers */
2172 		cv_broadcast(&stp->sd_monitor);	/* the ioctllers */
2173 		strhup(stp);
2174 		mutex_exit(&stp->sd_lock);
2175 		return (0);
2176 
2177 	case M_UNHANGUP:
2178 		freemsg(bp);
2179 		mutex_enter(&stp->sd_lock);
2180 		stp->sd_werror = 0;
2181 		stp->sd_flag &= ~STRHUP;
2182 		mutex_exit(&stp->sd_lock);
2183 		return (0);
2184 
2185 	case M_SIG:
2186 		/*
2187 		 * Someone downstream wants to post a signal.  The
2188 		 * signal to post is contained in the first byte of the
2189 		 * message.  If the message would go on the front of
2190 		 * the queue, send a signal to the process group
2191 		 * (if not SIGPOLL) or to the siglist processes
2192 		 * (SIGPOLL).  If something is already on the queue,
2193 		 * OR if we are delivering a delayed suspend (*sigh*
2194 		 * another "tty" hack) and there's no one sleeping already,
2195 		 * just enqueue the message.
2196 		 */
2197 		mutex_enter(&stp->sd_lock);
2198 		if (q->q_first || (*bp->b_rptr == SIGTSTP &&
2199 		    !(stp->sd_flag & RSLEEP))) {
2200 			(void) putq(q, bp);
2201 			mutex_exit(&stp->sd_lock);
2202 			return (0);
2203 		}
2204 		mutex_exit(&stp->sd_lock);
2205 		/* FALLTHRU */
2206 
2207 	case M_PCSIG:
2208 		/*
2209 		 * Don't enqueue, just post the signal.
2210 		 */
2211 		strsignal(stp, *bp->b_rptr, 0L);
2212 		freemsg(bp);
2213 		return (0);
2214 
2215 	case M_CMD:
2216 		if (MBLKL(bp) != sizeof (cmdblk_t)) {
2217 			freemsg(bp);
2218 			return (0);
2219 		}
2220 
2221 		mutex_enter(&stp->sd_lock);
2222 		if (stp->sd_flag & STRCMDWAIT) {
2223 			ASSERT(stp->sd_cmdblk == NULL);
2224 			stp->sd_cmdblk = bp;
2225 			cv_broadcast(&stp->sd_monitor);
2226 			mutex_exit(&stp->sd_lock);
2227 		} else {
2228 			mutex_exit(&stp->sd_lock);
2229 			freemsg(bp);
2230 		}
2231 		return (0);
2232 
2233 	case M_FLUSH:
2234 		/*
2235 		 * Flush queues.  The indication of which queues to flush
2236 		 * is in the first byte of the message.  If the read queue
2237 		 * is specified, then flush it.  If FLUSHBAND is set, just
2238 		 * flush the band specified by the second byte of the message.
2239 		 *
2240 		 * If a module has issued a M_SETOPT to not flush hi
2241 		 * priority messages off of the stream head, then pass this
2242 		 * flag into the flushq code to preserve such messages.
2243 		 */
2244 
2245 		if (*bp->b_rptr & FLUSHR) {
2246 			mutex_enter(&stp->sd_lock);
2247 			if (*bp->b_rptr & FLUSHBAND) {
2248 				ASSERT((bp->b_wptr - bp->b_rptr) >= 2);
2249 				flushband(q, *(bp->b_rptr + 1), FLUSHALL);
2250 			} else
2251 				flushq_common(q, FLUSHALL,
2252 				    stp->sd_read_opt & RFLUSHPCPROT);
2253 			if ((q->q_first == NULL) ||
2254 			    (q->q_first->b_datap->db_type < QPCTL))
2255 				stp->sd_flag &= ~STRPRI;
2256 			else {
2257 				ASSERT(stp->sd_flag & STRPRI);
2258 			}
2259 			mutex_exit(&stp->sd_lock);
2260 		}
2261 		if ((*bp->b_rptr & FLUSHW) && !(bp->b_flag & MSGNOLOOP)) {
2262 			*bp->b_rptr &= ~FLUSHR;
2263 			bp->b_flag |= MSGNOLOOP;
2264 			/*
2265 			 * Protect against the driver passing up
2266 			 * messages after it has done a qprocsoff.
2267 			 */
2268 			if (_OTHERQ(q)->q_next == NULL)
2269 				freemsg(bp);
2270 			else
2271 				qreply(q, bp);
2272 			return (0);
2273 		}
2274 		freemsg(bp);
2275 		return (0);
2276 
2277 	case M_IOCACK:
2278 	case M_IOCNAK:
2279 		iocbp = (struct iocblk *)bp->b_rptr;
2280 		/*
2281 		 * If not waiting for ACK or NAK then just free msg.
2282 		 * If incorrect id sequence number then just free msg.
2283 		 * If already have ACK or NAK for user then this is a
2284 		 *    duplicate, display a warning and free the msg.
2285 		 */
2286 		mutex_enter(&stp->sd_lock);
2287 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2288 		    (stp->sd_iocid != iocbp->ioc_id)) {
2289 			/*
2290 			 * If the ACK/NAK is a dup, display a message
2291 			 * Dup is when sd_iocid == ioc_id, and
2292 			 * sd_iocblk == <valid ptr> or -1 (the former
2293 			 * is when an ioctl has been put on the stream
2294 			 * head, but has not yet been consumed, the
2295 			 * later is when it has been consumed).
2296 			 */
2297 			if ((stp->sd_iocid == iocbp->ioc_id) &&
2298 			    (stp->sd_iocblk != NULL)) {
2299 				log_dupioc(q, bp);
2300 			}
2301 			freemsg(bp);
2302 			mutex_exit(&stp->sd_lock);
2303 			return (0);
2304 		}
2305 
2306 		/*
2307 		 * Assign ACK or NAK to user and wake up.
2308 		 */
2309 		stp->sd_iocblk = bp;
2310 		cv_broadcast(&stp->sd_monitor);
2311 		mutex_exit(&stp->sd_lock);
2312 		return (0);
2313 
2314 	case M_COPYIN:
2315 	case M_COPYOUT:
2316 		reqp = (struct copyreq *)bp->b_rptr;
2317 
2318 		/*
2319 		 * If not waiting for ACK or NAK then just fail request.
2320 		 * If already have ACK, NAK, or copy request, then just
2321 		 * fail request.
2322 		 * If incorrect id sequence number then just fail request.
2323 		 */
2324 		mutex_enter(&stp->sd_lock);
2325 		if ((stp->sd_flag & IOCWAIT) == 0 || stp->sd_iocblk ||
2326 		    (stp->sd_iocid != reqp->cq_id)) {
2327 			if (bp->b_cont) {
2328 				freemsg(bp->b_cont);
2329 				bp->b_cont = NULL;
2330 			}
2331 			bp->b_datap->db_type = M_IOCDATA;
2332 			bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
2333 			resp = (struct copyresp *)bp->b_rptr;
2334 			resp->cp_rval = (caddr_t)1;	/* failure */
2335 			mutex_exit(&stp->sd_lock);
2336 			putnext(stp->sd_wrq, bp);
2337 			return (0);
2338 		}
2339 
2340 		/*
2341 		 * Assign copy request to user and wake up.
2342 		 */
2343 		stp->sd_iocblk = bp;
2344 		cv_broadcast(&stp->sd_monitor);
2345 		mutex_exit(&stp->sd_lock);
2346 		return (0);
2347 
2348 	case M_SETOPTS:
2349 		/*
2350 		 * Set stream head options (read option, write offset,
2351 		 * min/max packet size, and/or high/low water marks for
2352 		 * the read side only).
2353 		 */
2354 
2355 		bpri = 0;
2356 		sop = (struct stroptions *)bp->b_rptr;
2357 		mutex_enter(&stp->sd_lock);
2358 		if (sop->so_flags & SO_READOPT) {
2359 			switch (sop->so_readopt & RMODEMASK) {
2360 			case RNORM:
2361 				stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
2362 				break;
2363 
2364 			case RMSGD:
2365 				stp->sd_read_opt =
2366 				    ((stp->sd_read_opt & ~RD_MSGNODIS) |
2367 				    RD_MSGDIS);
2368 				break;
2369 
2370 			case RMSGN:
2371 				stp->sd_read_opt =
2372 				    ((stp->sd_read_opt & ~RD_MSGDIS) |
2373 				    RD_MSGNODIS);
2374 				break;
2375 			}
2376 			switch (sop->so_readopt & RPROTMASK) {
2377 			case RPROTNORM:
2378 				stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
2379 				break;
2380 
2381 			case RPROTDAT:
2382 				stp->sd_read_opt =
2383 				    ((stp->sd_read_opt & ~RD_PROTDIS) |
2384 				    RD_PROTDAT);
2385 				break;
2386 
2387 			case RPROTDIS:
2388 				stp->sd_read_opt =
2389 				    ((stp->sd_read_opt & ~RD_PROTDAT) |
2390 				    RD_PROTDIS);
2391 				break;
2392 			}
2393 			switch (sop->so_readopt & RFLUSHMASK) {
2394 			case RFLUSHPCPROT:
2395 				/*
2396 				 * This sets the stream head to NOT flush
2397 				 * M_PCPROTO messages.
2398 				 */
2399 				stp->sd_read_opt |= RFLUSHPCPROT;
2400 				break;
2401 			}
2402 		}
2403 		if (sop->so_flags & SO_ERROPT) {
2404 			switch (sop->so_erropt & RERRMASK) {
2405 			case RERRNORM:
2406 				stp->sd_flag &= ~STRDERRNONPERSIST;
2407 				break;
2408 			case RERRNONPERSIST:
2409 				stp->sd_flag |= STRDERRNONPERSIST;
2410 				break;
2411 			}
2412 			switch (sop->so_erropt & WERRMASK) {
2413 			case WERRNORM:
2414 				stp->sd_flag &= ~STWRERRNONPERSIST;
2415 				break;
2416 			case WERRNONPERSIST:
2417 				stp->sd_flag |= STWRERRNONPERSIST;
2418 				break;
2419 			}
2420 		}
2421 		if (sop->so_flags & SO_COPYOPT) {
2422 			if (sop->so_copyopt & ZCVMSAFE) {
2423 				stp->sd_copyflag |= STZCVMSAFE;
2424 				stp->sd_copyflag &= ~STZCVMUNSAFE;
2425 			} else if (sop->so_copyopt & ZCVMUNSAFE) {
2426 				stp->sd_copyflag |= STZCVMUNSAFE;
2427 				stp->sd_copyflag &= ~STZCVMSAFE;
2428 			}
2429 
2430 			if (sop->so_copyopt & COPYCACHED) {
2431 				stp->sd_copyflag |= STRCOPYCACHED;
2432 			}
2433 		}
2434 		if (sop->so_flags & SO_WROFF)
2435 			stp->sd_wroff = sop->so_wroff;
2436 		if (sop->so_flags & SO_TAIL)
2437 			stp->sd_tail = sop->so_tail;
2438 		if (sop->so_flags & SO_MINPSZ)
2439 			q->q_minpsz = sop->so_minpsz;
2440 		if (sop->so_flags & SO_MAXPSZ)
2441 			q->q_maxpsz = sop->so_maxpsz;
2442 		if (sop->so_flags & SO_MAXBLK)
2443 			stp->sd_maxblk = sop->so_maxblk;
2444 		if (sop->so_flags & SO_HIWAT) {
2445 			if (sop->so_flags & SO_BAND) {
2446 				if (strqset(q, QHIWAT,
2447 				    sop->so_band, sop->so_hiwat)) {
2448 					cmn_err(CE_WARN, "strrput: could not "
2449 					    "allocate qband\n");
2450 				} else {
2451 					bpri = sop->so_band;
2452 				}
2453 			} else {
2454 				q->q_hiwat = sop->so_hiwat;
2455 			}
2456 		}
2457 		if (sop->so_flags & SO_LOWAT) {
2458 			if (sop->so_flags & SO_BAND) {
2459 				if (strqset(q, QLOWAT,
2460 				    sop->so_band, sop->so_lowat)) {
2461 					cmn_err(CE_WARN, "strrput: could not "
2462 					    "allocate qband\n");
2463 				} else {
2464 					bpri = sop->so_band;
2465 				}
2466 			} else {
2467 				q->q_lowat = sop->so_lowat;
2468 			}
2469 		}
2470 		if (sop->so_flags & SO_MREADON)
2471 			stp->sd_flag |= SNDMREAD;
2472 		if (sop->so_flags & SO_MREADOFF)
2473 			stp->sd_flag &= ~SNDMREAD;
2474 		if (sop->so_flags & SO_NDELON)
2475 			stp->sd_flag |= OLDNDELAY;
2476 		if (sop->so_flags & SO_NDELOFF)
2477 			stp->sd_flag &= ~OLDNDELAY;
2478 		if (sop->so_flags & SO_ISTTY)
2479 			stp->sd_flag |= STRISTTY;
2480 		if (sop->so_flags & SO_ISNTTY)
2481 			stp->sd_flag &= ~STRISTTY;
2482 		if (sop->so_flags & SO_TOSTOP)
2483 			stp->sd_flag |= STRTOSTOP;
2484 		if (sop->so_flags & SO_TONSTOP)
2485 			stp->sd_flag &= ~STRTOSTOP;
2486 		if (sop->so_flags & SO_DELIM)
2487 			stp->sd_flag |= STRDELIM;
2488 		if (sop->so_flags & SO_NODELIM)
2489 			stp->sd_flag &= ~STRDELIM;
2490 
2491 		mutex_exit(&stp->sd_lock);
2492 		freemsg(bp);
2493 
2494 		/* Check backenable in case the water marks changed */
2495 		qbackenable(q, bpri);
2496 		return (0);
2497 
2498 	/*
2499 	 * The following set of cases deal with situations where two stream
2500 	 * heads are connected to each other (twisted streams).  These messages
2501 	 * have no meaning at the stream head.
2502 	 */
2503 	case M_BREAK:
2504 	case M_CTL:
2505 	case M_DELAY:
2506 	case M_START:
2507 	case M_STOP:
2508 	case M_IOCDATA:
2509 	case M_STARTI:
2510 	case M_STOPI:
2511 		freemsg(bp);
2512 		return (0);
2513 
2514 	case M_IOCTL:
2515 		/*
2516 		 * Always NAK this condition
2517 		 * (makes no sense)
2518 		 * If there is one or more threads in the read side
2519 		 * rwnext we have to defer the nacking until that thread
2520 		 * returns (in strget).
2521 		 */
2522 		mutex_enter(&stp->sd_lock);
2523 		if (stp->sd_struiodnak != 0) {
2524 			/*
2525 			 * Defer NAK to the streamhead. Queue at the end
2526 			 * the list.
2527 			 */
2528 			mblk_t *mp = stp->sd_struionak;
2529 
2530 			while (mp && mp->b_next)
2531 				mp = mp->b_next;
2532 			if (mp)
2533 				mp->b_next = bp;
2534 			else
2535 				stp->sd_struionak = bp;
2536 			bp->b_next = NULL;
2537 			mutex_exit(&stp->sd_lock);
2538 			return (0);
2539 		}
2540 		mutex_exit(&stp->sd_lock);
2541 
2542 		bp->b_datap->db_type = M_IOCNAK;
2543 		/*
2544 		 * Protect against the driver passing up
2545 		 * messages after it has done a qprocsoff.
2546 		 */
2547 		if (_OTHERQ(q)->q_next == NULL)
2548 			freemsg(bp);
2549 		else
2550 			qreply(q, bp);
2551 		return (0);
2552 
2553 	default:
2554 #ifdef DEBUG
2555 		cmn_err(CE_WARN,
2556 		    "bad message type %x received at stream head\n",
2557 		    bp->b_datap->db_type);
2558 #endif
2559 		freemsg(bp);
2560 		return (0);
2561 	}
2562 
2563 	/* NOTREACHED */
2564 }
2565 
2566 /*
2567  * Check if the stream pointed to by `stp' can be written to, and return an
2568  * error code if not.  If `eiohup' is set, then return EIO if STRHUP is set.
2569  * If `sigpipeok' is set and the SW_SIGPIPE option is enabled on the stream,
2570  * then always return EPIPE and send a SIGPIPE to the invoking thread.
2571  */
2572 static int
2573 strwriteable(struct stdata *stp, boolean_t eiohup, boolean_t sigpipeok)
2574 {
2575 	int error;
2576 
2577 	ASSERT(MUTEX_HELD(&stp->sd_lock));
2578 
2579 	/*
2580 	 * For modem support, POSIX states that on writes, EIO should
2581 	 * be returned if the stream has been hung up.
2582 	 */
2583 	if (eiohup && (stp->sd_flag & (STPLEX|STRHUP)) == STRHUP)
2584 		error = EIO;
2585 	else
2586 		error = strgeterr(stp, STRHUP|STPLEX|STWRERR, 0);
2587 
2588 	if (error != 0) {
2589 		if (!(stp->sd_flag & STPLEX) &&
2590 		    (stp->sd_wput_opt & SW_SIGPIPE) && sigpipeok) {
2591 			tsignal(curthread, SIGPIPE);
2592 			error = EPIPE;
2593 		}
2594 	}
2595 
2596 	return (error);
2597 }
2598 
2599 /*
2600  * Copyin and send data down a stream.
2601  * The caller will allocate and copyin any control part that precedes the
2602  * message and pass than in as mctl.
2603  *
2604  * Caller should *not* hold sd_lock.
2605  * When EWOULDBLOCK is returned the caller has to redo the canputnext
2606  * under sd_lock in order to avoid missing a backenabling wakeup.
2607  *
2608  * Use iosize = -1 to not send any M_DATA. iosize = 0 sends zero-length M_DATA.
2609  *
2610  * Set MSG_IGNFLOW in flags to ignore flow control for hipri messages.
2611  * For sync streams we can only ignore flow control by reverting to using
2612  * putnext.
2613  *
2614  * If sd_maxblk is less than *iosize this routine might return without
2615  * transferring all of *iosize. In all cases, on return *iosize will contain
2616  * the amount of data that was transferred.
2617  */
2618 static int
2619 strput(struct stdata *stp, mblk_t *mctl, struct uio *uiop, ssize_t *iosize,
2620     int b_flag, int pri, int flags)
2621 {
2622 	struiod_t uiod;
2623 	mblk_t *mp;
2624 	queue_t *wqp = stp->sd_wrq;
2625 	int error = 0;
2626 	ssize_t count = *iosize;
2627 	cred_t *cr;
2628 
2629 	ASSERT(MUTEX_NOT_HELD(&stp->sd_lock));
2630 
2631 	if (uiop != NULL && count >= 0)
2632 		flags |= stp->sd_struiowrq ? STRUIO_POSTPONE : 0;
2633 
2634 	if (!(flags & STRUIO_POSTPONE)) {
2635 		/*
2636 		 * Use regular canputnext, strmakedata, putnext sequence.
2637 		 */
2638 		if (pri == 0) {
2639 			if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2640 				freemsg(mctl);
2641 				return (EWOULDBLOCK);
2642 			}
2643 		} else {
2644 			if (!(flags & MSG_IGNFLOW) && !bcanputnext(wqp, pri)) {
2645 				freemsg(mctl);
2646 				return (EWOULDBLOCK);
2647 			}
2648 		}
2649 
2650 		if ((error = strmakedata(iosize, uiop, stp, flags,
2651 		    &mp)) != 0) {
2652 			freemsg(mctl);
2653 			/*
2654 			 * need to change return code to ENOMEM
2655 			 * so that this is not confused with
2656 			 * flow control, EAGAIN.
2657 			 */
2658 
2659 			if (error == EAGAIN)
2660 				return (ENOMEM);
2661 			else
2662 				return (error);
2663 		}
2664 		if (mctl != NULL) {
2665 			if (mctl->b_cont == NULL)
2666 				mctl->b_cont = mp;
2667 			else if (mp != NULL)
2668 				linkb(mctl, mp);
2669 			mp = mctl;
2670 			/*
2671 			 * Note that for interrupt thread, the CRED() is
2672 			 * NULL. Don't bother with the pid either.
2673 			 */
2674 			if ((cr = CRED()) != NULL) {
2675 				mblk_setcred(mp, cr);
2676 				DB_CPID(mp) = curproc->p_pid;
2677 			}
2678 		} else if (mp == NULL)
2679 			return (0);
2680 
2681 		mp->b_flag |= b_flag;
2682 		mp->b_band = (uchar_t)pri;
2683 
2684 		if (flags & MSG_IGNFLOW) {
2685 			/*
2686 			 * XXX Hack: Don't get stuck running service
2687 			 * procedures. This is needed for sockfs when
2688 			 * sending the unbind message out of the rput
2689 			 * 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 	 * Stream supports rwnext() for the write side.
2702 	 */
2703 	if ((error = strmakedata(iosize, uiop, stp, flags, &mp)) != 0) {
2704 		freemsg(mctl);
2705 		/*
2706 		 * map EAGAIN to ENOMEM since EAGAIN means "flow controlled".
2707 		 */
2708 		return (error == EAGAIN ? ENOMEM : error);
2709 	}
2710 	if (mctl != NULL) {
2711 		if (mctl->b_cont == NULL)
2712 			mctl->b_cont = mp;
2713 		else if (mp != NULL)
2714 			linkb(mctl, mp);
2715 		mp = mctl;
2716 		/*
2717 		 * Note that for interrupt thread, the CRED() is
2718 		 * NULL.  Don't bother with the pid either.
2719 		 */
2720 		if ((cr = CRED()) != NULL) {
2721 			mblk_setcred(mp, cr);
2722 			DB_CPID(mp) = curproc->p_pid;
2723 		}
2724 	} else if (mp == NULL) {
2725 		return (0);
2726 	}
2727 
2728 	mp->b_flag |= b_flag;
2729 	mp->b_band = (uchar_t)pri;
2730 
2731 	(void) uiodup(uiop, &uiod.d_uio, uiod.d_iov,
2732 	    sizeof (uiod.d_iov) / sizeof (*uiod.d_iov));
2733 	uiod.d_uio.uio_offset = 0;
2734 	uiod.d_mp = mp;
2735 	error = rwnext(wqp, &uiod);
2736 	if (! uiod.d_mp) {
2737 		uioskip(uiop, *iosize);
2738 		return (error);
2739 	}
2740 	ASSERT(mp == uiod.d_mp);
2741 	if (error == EINVAL) {
2742 		/*
2743 		 * The stream plumbing must have changed while
2744 		 * we were away, so just turn off rwnext()s.
2745 		 */
2746 		error = 0;
2747 	} else if (error == EBUSY || error == EWOULDBLOCK) {
2748 		/*
2749 		 * Couldn't enter a perimeter or took a page fault,
2750 		 * so fall-back to putnext().
2751 		 */
2752 		error = 0;
2753 	} else {
2754 		freemsg(mp);
2755 		return (error);
2756 	}
2757 	/* Have to check canput before consuming data from the uio */
2758 	if (pri == 0) {
2759 		if (!canputnext(wqp) && !(flags & MSG_IGNFLOW)) {
2760 			freemsg(mp);
2761 			return (EWOULDBLOCK);
2762 		}
2763 	} else {
2764 		if (!bcanputnext(wqp, pri) && !(flags & MSG_IGNFLOW)) {
2765 			freemsg(mp);
2766 			return (EWOULDBLOCK);
2767 		}
2768 	}
2769 	ASSERT(mp == uiod.d_mp);
2770 	/* Copyin data from the uio */
2771 	if ((error = struioget(wqp, mp, &uiod, 0)) != 0) {
2772 		freemsg(mp);
2773 		return (error);
2774 	}
2775 	uioskip(uiop, *iosize);
2776 	if (flags & MSG_IGNFLOW) {
2777 		/*
2778 		 * XXX Hack: Don't get stuck running service procedures.
2779 		 * This is needed for sockfs when sending the unbind message
2780 		 * out of the rput procedure - we don't want a put procedure
2781 		 * to run service procedures.
2782 		 */
2783 		putnext(wqp, mp);
2784 	} else {
2785 		stream_willservice(stp);
2786 		putnext(wqp, mp);
2787 		stream_runservice(stp);
2788 	}
2789 	return (0);
2790 }
2791 
2792 /*
2793  * Write attempts to break the write request into messages conforming
2794  * with the minimum and maximum packet sizes set downstream.
2795  *
2796  * Write will not block if downstream queue is full and
2797  * O_NDELAY is set, otherwise it will block waiting for the queue to get room.
2798  *
2799  * A write of zero bytes gets packaged into a zero length message and sent
2800  * downstream like any other message.
2801  *
2802  * If buffers of the requested sizes are not available, the write will
2803  * sleep until the buffers become available.
2804  *
2805  * Write (if specified) will supply a write offset in a message if it
2806  * makes sense. This can be specified by downstream modules as part of
2807  * a M_SETOPTS message.  Write will not supply the write offset if it
2808  * cannot supply any data in a buffer.  In other words, write will never
2809  * send down an empty packet due to a write offset.
2810  */
2811 /* ARGSUSED2 */
2812 int
2813 strwrite(struct vnode *vp, struct uio *uiop, cred_t *crp)
2814 {
2815 	return (strwrite_common(vp, uiop, crp, 0));
2816 }
2817 
2818 /* ARGSUSED2 */
2819 int
2820 strwrite_common(struct vnode *vp, struct uio *uiop, cred_t *crp, int wflag)
2821 {
2822 	struct stdata *stp;
2823 	struct queue *wqp;
2824 	ssize_t rmin, rmax;
2825 	ssize_t iosize;
2826 	int waitflag;
2827 	int tempmode;
2828 	int error = 0;
2829 	int b_flag;
2830 
2831 	ASSERT(vp->v_stream);
2832 	stp = vp->v_stream;
2833 
2834 	mutex_enter(&stp->sd_lock);
2835 
2836 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
2837 		mutex_exit(&stp->sd_lock);
2838 		return (error);
2839 	}
2840 
2841 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
2842 		error = strwriteable(stp, B_TRUE, B_TRUE);
2843 		if (error != 0) {
2844 			mutex_exit(&stp->sd_lock);
2845 			return (error);
2846 		}
2847 	}
2848 
2849 	mutex_exit(&stp->sd_lock);
2850 
2851 	wqp = stp->sd_wrq;
2852 
2853 	/* get these values from them cached in the stream head */
2854 	rmin = stp->sd_qn_minpsz;
2855 	rmax = stp->sd_qn_maxpsz;
2856 
2857 	/*
2858 	 * Check the min/max packet size constraints.  If min packet size
2859 	 * is non-zero, the write cannot be split into multiple messages
2860 	 * and still guarantee the size constraints.
2861 	 */
2862 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_IN, "strwrite in:q %p", wqp);
2863 
2864 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
2865 	if (rmax == 0) {
2866 		return (0);
2867 	}
2868 	if (rmin > 0) {
2869 		if (uiop->uio_resid < rmin) {
2870 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2871 			    "strwrite out:q %p out %d error %d",
2872 			    wqp, 0, ERANGE);
2873 			return (ERANGE);
2874 		}
2875 		if ((rmax != INFPSZ) && (uiop->uio_resid > rmax)) {
2876 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2877 			    "strwrite out:q %p out %d error %d",
2878 			    wqp, 1, ERANGE);
2879 			return (ERANGE);
2880 		}
2881 	}
2882 
2883 	/*
2884 	 * Do until count satisfied or error.
2885 	 */
2886 	waitflag = WRITEWAIT | wflag;
2887 	if (stp->sd_flag & OLDNDELAY)
2888 		tempmode = uiop->uio_fmode & ~FNDELAY;
2889 	else
2890 		tempmode = uiop->uio_fmode;
2891 
2892 	if (rmax == INFPSZ)
2893 		rmax = uiop->uio_resid;
2894 
2895 	/*
2896 	 * Note that tempmode does not get used in strput/strmakedata
2897 	 * but only in strwaitq. The other routines use uio_fmode
2898 	 * unmodified.
2899 	 */
2900 
2901 	/* LINTED: constant in conditional context */
2902 	while (1) {	/* breaks when uio_resid reaches zero */
2903 		/*
2904 		 * Determine the size of the next message to be
2905 		 * packaged.  May have to break write into several
2906 		 * messages based on max packet size.
2907 		 */
2908 		iosize = MIN(uiop->uio_resid, rmax);
2909 
2910 		/*
2911 		 * Put block downstream when flow control allows it.
2912 		 */
2913 		if ((stp->sd_flag & STRDELIM) && (uiop->uio_resid == iosize))
2914 			b_flag = MSGDELIM;
2915 		else
2916 			b_flag = 0;
2917 
2918 		for (;;) {
2919 			int done = 0;
2920 
2921 			error = strput(stp, NULL, uiop, &iosize, b_flag, 0, 0);
2922 			if (error == 0)
2923 				break;
2924 			if (error != EWOULDBLOCK)
2925 				goto out;
2926 
2927 			mutex_enter(&stp->sd_lock);
2928 			/*
2929 			 * Check for a missed wakeup.
2930 			 * Needed since strput did not hold sd_lock across
2931 			 * the canputnext.
2932 			 */
2933 			if (canputnext(wqp)) {
2934 				/* Try again */
2935 				mutex_exit(&stp->sd_lock);
2936 				continue;
2937 			}
2938 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAIT,
2939 			    "strwrite wait:q %p wait", wqp);
2940 			if ((error = strwaitq(stp, waitflag, (ssize_t)0,
2941 			    tempmode, -1, &done)) != 0 || done) {
2942 				mutex_exit(&stp->sd_lock);
2943 				if ((vp->v_type == VFIFO) &&
2944 				    (uiop->uio_fmode & FNDELAY) &&
2945 				    (error == EAGAIN))
2946 					error = 0;
2947 				goto out;
2948 			}
2949 			TRACE_1(TR_FAC_STREAMS_FR, TR_STRWRITE_WAKE,
2950 			    "strwrite wake:q %p awakes", wqp);
2951 			if ((error = i_straccess(stp, JCWRITE)) != 0) {
2952 				mutex_exit(&stp->sd_lock);
2953 				goto out;
2954 			}
2955 			mutex_exit(&stp->sd_lock);
2956 		}
2957 		waitflag |= NOINTR;
2958 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRWRITE_RESID,
2959 		    "strwrite resid:q %p uiop %p", wqp, uiop);
2960 		if (uiop->uio_resid) {
2961 			/* Recheck for errors - needed for sockets */
2962 			if ((stp->sd_wput_opt & SW_RECHECK_ERR) &&
2963 			    (stp->sd_flag & (STWRERR|STRHUP|STPLEX))) {
2964 				mutex_enter(&stp->sd_lock);
2965 				error = strwriteable(stp, B_FALSE, B_TRUE);
2966 				mutex_exit(&stp->sd_lock);
2967 				if (error != 0)
2968 					return (error);
2969 			}
2970 			continue;
2971 		}
2972 		break;
2973 	}
2974 out:
2975 	/*
2976 	 * For historical reasons, applications expect EAGAIN when a data
2977 	 * mblk_t cannot be allocated, so change ENOMEM back to EAGAIN.
2978 	 */
2979 	if (error == ENOMEM)
2980 		error = EAGAIN;
2981 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRWRITE_OUT,
2982 	    "strwrite out:q %p out %d error %d", wqp, 2, error);
2983 	return (error);
2984 }
2985 
2986 /*
2987  * Stream head write service routine.
2988  * Its job is to wake up any sleeping writers when a queue
2989  * downstream needs data (part of the flow control in putq and getq).
2990  * It also must wake anyone sleeping on a poll().
2991  * For stream head right below mux module, it must also invoke put procedure
2992  * of next downstream module.
2993  */
2994 int
2995 strwsrv(queue_t *q)
2996 {
2997 	struct stdata *stp;
2998 	queue_t *tq;
2999 	qband_t *qbp;
3000 	int i;
3001 	qband_t *myqbp;
3002 	int isevent;
3003 	unsigned char	qbf[NBAND];	/* band flushing backenable flags */
3004 
3005 	TRACE_1(TR_FAC_STREAMS_FR,
3006 	    TR_STRWSRV, "strwsrv:q %p", q);
3007 	stp = (struct stdata *)q->q_ptr;
3008 	ASSERT(qclaimed(q));
3009 	mutex_enter(&stp->sd_lock);
3010 	ASSERT(!(stp->sd_flag & STPLEX));
3011 
3012 	if (stp->sd_flag & WSLEEP) {
3013 		stp->sd_flag &= ~WSLEEP;
3014 		cv_broadcast(&q->q_wait);
3015 	}
3016 	mutex_exit(&stp->sd_lock);
3017 
3018 	/* The other end of a stream pipe went away. */
3019 	if ((tq = q->q_next) == NULL) {
3020 		return (0);
3021 	}
3022 
3023 	/* Find the next module forward that has a service procedure */
3024 	claimstr(q);
3025 	tq = q->q_nfsrv;
3026 	ASSERT(tq != NULL);
3027 
3028 	if ((q->q_flag & QBACK)) {
3029 		if ((tq->q_flag & QFULL)) {
3030 			mutex_enter(QLOCK(tq));
3031 			if (!(tq->q_flag & QFULL)) {
3032 				mutex_exit(QLOCK(tq));
3033 				goto wakeup;
3034 			}
3035 			/*
3036 			 * The queue must have become full again. Set QWANTW
3037 			 * again so strwsrv will be back enabled when
3038 			 * the queue becomes non-full next time.
3039 			 */
3040 			tq->q_flag |= QWANTW;
3041 			mutex_exit(QLOCK(tq));
3042 		} else {
3043 		wakeup:
3044 			pollwakeup(&stp->sd_pollist, POLLWRNORM);
3045 			mutex_enter(&stp->sd_lock);
3046 			if (stp->sd_sigflags & S_WRNORM)
3047 				strsendsig(stp->sd_siglist, S_WRNORM, 0, 0);
3048 			mutex_exit(&stp->sd_lock);
3049 		}
3050 	}
3051 
3052 	isevent = 0;
3053 	i = 1;
3054 	bzero((caddr_t)qbf, NBAND);
3055 	mutex_enter(QLOCK(tq));
3056 	if ((myqbp = q->q_bandp) != NULL)
3057 		for (qbp = tq->q_bandp; qbp && myqbp; qbp = qbp->qb_next) {
3058 			ASSERT(myqbp);
3059 			if ((myqbp->qb_flag & QB_BACK)) {
3060 				if (qbp->qb_flag & QB_FULL) {
3061 					/*
3062 					 * The band must have become full again.
3063 					 * Set QB_WANTW again so strwsrv will
3064 					 * be back enabled when the band becomes
3065 					 * non-full next time.
3066 					 */
3067 					qbp->qb_flag |= QB_WANTW;
3068 				} else {
3069 					isevent = 1;
3070 					qbf[i] = 1;
3071 				}
3072 			}
3073 			myqbp = myqbp->qb_next;
3074 			i++;
3075 		}
3076 	mutex_exit(QLOCK(tq));
3077 
3078 	if (isevent) {
3079 		for (i = tq->q_nband; i; i--) {
3080 			if (qbf[i]) {
3081 				pollwakeup(&stp->sd_pollist, POLLWRBAND);
3082 				mutex_enter(&stp->sd_lock);
3083 				if (stp->sd_sigflags & S_WRBAND)
3084 					strsendsig(stp->sd_siglist, S_WRBAND,
3085 					    (uchar_t)i, 0);
3086 				mutex_exit(&stp->sd_lock);
3087 			}
3088 		}
3089 	}
3090 
3091 	releasestr(q);
3092 	return (0);
3093 }
3094 
3095 /*
3096  * Special case of strcopyin/strcopyout for copying
3097  * struct strioctl that can deal with both data
3098  * models.
3099  */
3100 
3101 #ifdef	_LP64
3102 
3103 static int
3104 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3105 {
3106 	struct	strioctl32 strioc32;
3107 	struct	strioctl *striocp;
3108 
3109 	if (copyflag & U_TO_K) {
3110 		ASSERT((copyflag & K_TO_K) == 0);
3111 
3112 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
3113 			if (copyin(from, &strioc32, sizeof (strioc32)))
3114 				return (EFAULT);
3115 
3116 			striocp = (struct strioctl *)to;
3117 			striocp->ic_cmd	= strioc32.ic_cmd;
3118 			striocp->ic_timout = strioc32.ic_timout;
3119 			striocp->ic_len	= strioc32.ic_len;
3120 			striocp->ic_dp	= (char *)(uintptr_t)strioc32.ic_dp;
3121 
3122 		} else { /* NATIVE data model */
3123 			if (copyin(from, to, sizeof (struct strioctl))) {
3124 				return (EFAULT);
3125 			} else {
3126 				return (0);
3127 			}
3128 		}
3129 	} else {
3130 		ASSERT(copyflag & K_TO_K);
3131 		bcopy(from, to, sizeof (struct strioctl));
3132 	}
3133 	return (0);
3134 }
3135 
3136 static int
3137 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3138 {
3139 	struct	strioctl32 strioc32;
3140 	struct	strioctl *striocp;
3141 
3142 	if (copyflag & U_TO_K) {
3143 		ASSERT((copyflag & K_TO_K) == 0);
3144 
3145 		if ((flag & FMODELS) == DATAMODEL_ILP32) {
3146 			striocp = (struct strioctl *)from;
3147 			strioc32.ic_cmd	= striocp->ic_cmd;
3148 			strioc32.ic_timout = striocp->ic_timout;
3149 			strioc32.ic_len	= striocp->ic_len;
3150 			strioc32.ic_dp	= (caddr32_t)(uintptr_t)striocp->ic_dp;
3151 			ASSERT((char *)(uintptr_t)strioc32.ic_dp ==
3152 			    striocp->ic_dp);
3153 
3154 			if (copyout(&strioc32, to, sizeof (strioc32)))
3155 				return (EFAULT);
3156 
3157 		} else { /* NATIVE data model */
3158 			if (copyout(from, to, sizeof (struct strioctl))) {
3159 				return (EFAULT);
3160 			} else {
3161 				return (0);
3162 			}
3163 		}
3164 	} else {
3165 		ASSERT(copyflag & K_TO_K);
3166 		bcopy(from, to, sizeof (struct strioctl));
3167 	}
3168 	return (0);
3169 }
3170 
3171 #else	/* ! _LP64 */
3172 
3173 /* ARGSUSED2 */
3174 static int
3175 strcopyin_strioctl(void *from, void *to, int flag, int copyflag)
3176 {
3177 	return (strcopyin(from, to, sizeof (struct strioctl), copyflag));
3178 }
3179 
3180 /* ARGSUSED2 */
3181 static int
3182 strcopyout_strioctl(void *from, void *to, int flag, int copyflag)
3183 {
3184 	return (strcopyout(from, to, sizeof (struct strioctl), copyflag));
3185 }
3186 
3187 #endif	/* _LP64 */
3188 
3189 /*
3190  * Determine type of job control semantics expected by user.  The
3191  * possibilities are:
3192  *	JCREAD	- Behaves like read() on fd; send SIGTTIN
3193  *	JCWRITE	- Behaves like write() on fd; send SIGTTOU if TOSTOP set
3194  *	JCSETP	- Sets a value in the stream; send SIGTTOU, ignore TOSTOP
3195  *	JCGETP	- Gets a value in the stream; no signals.
3196  * See straccess in strsubr.c for usage of these values.
3197  *
3198  * This routine also returns -1 for I_STR as a special case; the
3199  * caller must call again with the real ioctl number for
3200  * classification.
3201  */
3202 static int
3203 job_control_type(int cmd)
3204 {
3205 	switch (cmd) {
3206 	case I_STR:
3207 		return (-1);
3208 
3209 	case I_RECVFD:
3210 	case I_E_RECVFD:
3211 		return (JCREAD);
3212 
3213 	case I_FDINSERT:
3214 	case I_SENDFD:
3215 		return (JCWRITE);
3216 
3217 	case TCSETA:
3218 	case TCSETAW:
3219 	case TCSETAF:
3220 	case TCSBRK:
3221 	case TCXONC:
3222 	case TCFLSH:
3223 	case TCDSET:	/* Obsolete */
3224 	case TIOCSWINSZ:
3225 	case TCSETS:
3226 	case TCSETSW:
3227 	case TCSETSF:
3228 	case TIOCSETD:
3229 	case TIOCHPCL:
3230 	case TIOCSETP:
3231 	case TIOCSETN:
3232 	case TIOCEXCL:
3233 	case TIOCNXCL:
3234 	case TIOCFLUSH:
3235 	case TIOCSETC:
3236 	case TIOCLBIS:
3237 	case TIOCLBIC:
3238 	case TIOCLSET:
3239 	case TIOCSBRK:
3240 	case TIOCCBRK:
3241 	case TIOCSDTR:
3242 	case TIOCCDTR:
3243 	case TIOCSLTC:
3244 	case TIOCSTOP:
3245 	case TIOCSTART:
3246 	case TIOCSTI:
3247 	case TIOCSPGRP:
3248 	case TIOCMSET:
3249 	case TIOCMBIS:
3250 	case TIOCMBIC:
3251 	case TIOCREMOTE:
3252 	case TIOCSIGNAL:
3253 	case LDSETT:
3254 	case LDSMAP:	/* Obsolete */
3255 	case DIOCSETP:
3256 	case I_FLUSH:
3257 	case I_SRDOPT:
3258 	case I_SETSIG:
3259 	case I_SWROPT:
3260 	case I_FLUSHBAND:
3261 	case I_SETCLTIME:
3262 	case I_SERROPT:
3263 	case I_ESETSIG:
3264 	case FIONBIO:
3265 	case FIOASYNC:
3266 	case FIOSETOWN:
3267 	case JBOOT:	/* Obsolete */
3268 	case JTERM:	/* Obsolete */
3269 	case JTIMOM:	/* Obsolete */
3270 	case JZOMBOOT:	/* Obsolete */
3271 	case JAGENT:	/* Obsolete */
3272 	case JTRUN:	/* Obsolete */
3273 	case JXTPROTO:	/* Obsolete */
3274 	case TIOCSETLD:
3275 		return (JCSETP);
3276 	}
3277 
3278 	return (JCGETP);
3279 }
3280 
3281 /*
3282  * ioctl for streams
3283  */
3284 int
3285 strioctl(struct vnode *vp, int cmd, intptr_t arg, int flag, int copyflag,
3286     cred_t *crp, int *rvalp)
3287 {
3288 	struct stdata *stp;
3289 	struct strcmd *scp;
3290 	struct strioctl strioc;
3291 	struct uio uio;
3292 	struct iovec iov;
3293 	int access;
3294 	mblk_t *mp;
3295 	int error = 0;
3296 	int done = 0;
3297 	ssize_t	rmin, rmax;
3298 	queue_t *wrq;
3299 	queue_t *rdq;
3300 	boolean_t kioctl = B_FALSE;
3301 
3302 	if (flag & FKIOCTL) {
3303 		copyflag = K_TO_K;
3304 		kioctl = B_TRUE;
3305 	}
3306 	ASSERT(vp->v_stream);
3307 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
3308 	stp = vp->v_stream;
3309 
3310 	TRACE_3(TR_FAC_STREAMS_FR, TR_IOCTL_ENTER,
3311 	    "strioctl:stp %p cmd %X arg %lX", stp, cmd, arg);
3312 
3313 	if (audit_active)
3314 		audit_strioctl(vp, cmd, arg, flag, copyflag, crp, rvalp);
3315 
3316 	/*
3317 	 * If the copy is kernel to kernel, make sure that the FNATIVE
3318 	 * flag is set.  After this it would be a serious error to have
3319 	 * no model flag.
3320 	 */
3321 	if (copyflag == K_TO_K)
3322 		flag = (flag & ~FMODELS) | FNATIVE;
3323 
3324 	ASSERT((flag & FMODELS) != 0);
3325 
3326 	wrq = stp->sd_wrq;
3327 	rdq = _RD(wrq);
3328 
3329 	access = job_control_type(cmd);
3330 
3331 	/* We should never see these here, should be handled by iwscn */
3332 	if (cmd == SRIOCSREDIR || cmd == SRIOCISREDIR)
3333 		return (EINVAL);
3334 
3335 	mutex_enter(&stp->sd_lock);
3336 	if ((access != -1) && ((error = i_straccess(stp, access)) != 0)) {
3337 		mutex_exit(&stp->sd_lock);
3338 		return (error);
3339 	}
3340 	mutex_exit(&stp->sd_lock);
3341 
3342 	/*
3343 	 * Check for sgttyb-related ioctls first, and complain as
3344 	 * necessary.
3345 	 */
3346 	switch (cmd) {
3347 	case TIOCGETP:
3348 	case TIOCSETP:
3349 	case TIOCSETN:
3350 		if (sgttyb_handling >= 2 && !sgttyb_complaint) {
3351 			sgttyb_complaint = B_TRUE;
3352 			cmn_err(CE_NOTE,
3353 			    "application used obsolete TIOC[GS]ET");
3354 		}
3355 		if (sgttyb_handling >= 3) {
3356 			tsignal(curthread, SIGSYS);
3357 			return (EIO);
3358 		}
3359 		break;
3360 	}
3361 
3362 	mutex_enter(&stp->sd_lock);
3363 
3364 	switch (cmd) {
3365 	case I_RECVFD:
3366 	case I_E_RECVFD:
3367 	case I_PEEK:
3368 	case I_NREAD:
3369 	case FIONREAD:
3370 	case FIORDCHK:
3371 	case I_ATMARK:
3372 	case FIONBIO:
3373 	case FIOASYNC:
3374 		if (stp->sd_flag & (STRDERR|STPLEX)) {
3375 			error = strgeterr(stp, STRDERR|STPLEX, 0);
3376 			if (error != 0) {
3377 				mutex_exit(&stp->sd_lock);
3378 				return (error);
3379 			}
3380 		}
3381 		break;
3382 
3383 	default:
3384 		if (stp->sd_flag & (STRDERR|STWRERR|STPLEX)) {
3385 			error = strgeterr(stp, STRDERR|STWRERR|STPLEX, 0);
3386 			if (error != 0) {
3387 				mutex_exit(&stp->sd_lock);
3388 				return (error);
3389 			}
3390 		}
3391 	}
3392 
3393 	mutex_exit(&stp->sd_lock);
3394 
3395 	switch (cmd) {
3396 	default:
3397 		/*
3398 		 * The stream head has hardcoded knowledge of a
3399 		 * miscellaneous collection of terminal-, keyboard- and
3400 		 * mouse-related ioctls, enumerated below.  This hardcoded
3401 		 * knowledge allows the stream head to automatically
3402 		 * convert transparent ioctl requests made by userland
3403 		 * programs into I_STR ioctls which many old STREAMS
3404 		 * modules and drivers require.
3405 		 *
3406 		 * No new ioctls should ever be added to this list.
3407 		 * Instead, the STREAMS module or driver should be written
3408 		 * to either handle transparent ioctls or require any
3409 		 * userland programs to use I_STR ioctls (by returning
3410 		 * EINVAL to any transparent ioctl requests).
3411 		 *
3412 		 * More importantly, removing ioctls from this list should
3413 		 * be done with the utmost care, since our STREAMS modules
3414 		 * and drivers *count* on the stream head performing this
3415 		 * conversion, and thus may panic while processing
3416 		 * transparent ioctl request for one of these ioctls (keep
3417 		 * in mind that third party modules and drivers may have
3418 		 * similar problems).
3419 		 */
3420 		if (((cmd & IOCTYPE) == LDIOC) ||
3421 		    ((cmd & IOCTYPE) == tIOC) ||
3422 		    ((cmd & IOCTYPE) == TIOC) ||
3423 		    ((cmd & IOCTYPE) == KIOC) ||
3424 		    ((cmd & IOCTYPE) == MSIOC) ||
3425 		    ((cmd & IOCTYPE) == VUIOC)) {
3426 			/*
3427 			 * The ioctl is a tty ioctl - set up strioc buffer
3428 			 * and call strdoioctl() to do the work.
3429 			 */
3430 			if (stp->sd_flag & STRHUP)
3431 				return (ENXIO);
3432 			strioc.ic_cmd = cmd;
3433 			strioc.ic_timout = INFTIM;
3434 
3435 			switch (cmd) {
3436 
3437 			case TCXONC:
3438 			case TCSBRK:
3439 			case TCFLSH:
3440 			case TCDSET:
3441 				{
3442 				int native_arg = (int)arg;
3443 				strioc.ic_len = sizeof (int);
3444 				strioc.ic_dp = (char *)&native_arg;
3445 				return (strdoioctl(stp, &strioc, flag,
3446 				    K_TO_K, crp, rvalp));
3447 				}
3448 
3449 			case TCSETA:
3450 			case TCSETAW:
3451 			case TCSETAF:
3452 				strioc.ic_len = sizeof (struct termio);
3453 				strioc.ic_dp = (char *)arg;
3454 				return (strdoioctl(stp, &strioc, flag,
3455 				    copyflag, crp, rvalp));
3456 
3457 			case TCSETS:
3458 			case TCSETSW:
3459 			case TCSETSF:
3460 				strioc.ic_len = sizeof (struct termios);
3461 				strioc.ic_dp = (char *)arg;
3462 				return (strdoioctl(stp, &strioc, flag,
3463 				    copyflag, crp, rvalp));
3464 
3465 			case LDSETT:
3466 				strioc.ic_len = sizeof (struct termcb);
3467 				strioc.ic_dp = (char *)arg;
3468 				return (strdoioctl(stp, &strioc, flag,
3469 				    copyflag, crp, rvalp));
3470 
3471 			case TIOCSETP:
3472 				strioc.ic_len = sizeof (struct sgttyb);
3473 				strioc.ic_dp = (char *)arg;
3474 				return (strdoioctl(stp, &strioc, flag,
3475 				    copyflag, crp, rvalp));
3476 
3477 			case TIOCSTI:
3478 				if ((flag & FREAD) == 0 &&
3479 				    secpolicy_sti(crp) != 0) {
3480 					return (EPERM);
3481 				}
3482 				mutex_enter(&stp->sd_lock);
3483 				mutex_enter(&curproc->p_splock);
3484 				if (stp->sd_sidp != curproc->p_sessp->s_sidp &&
3485 				    secpolicy_sti(crp) != 0) {
3486 					mutex_exit(&curproc->p_splock);
3487 					mutex_exit(&stp->sd_lock);
3488 					return (EACCES);
3489 				}
3490 				mutex_exit(&curproc->p_splock);
3491 				mutex_exit(&stp->sd_lock);
3492 
3493 				strioc.ic_len = sizeof (char);
3494 				strioc.ic_dp = (char *)arg;
3495 				return (strdoioctl(stp, &strioc, flag,
3496 				    copyflag, crp, rvalp));
3497 
3498 			case TIOCSWINSZ:
3499 				strioc.ic_len = sizeof (struct winsize);
3500 				strioc.ic_dp = (char *)arg;
3501 				return (strdoioctl(stp, &strioc, flag,
3502 				    copyflag, crp, rvalp));
3503 
3504 			case TIOCSSIZE:
3505 				strioc.ic_len = sizeof (struct ttysize);
3506 				strioc.ic_dp = (char *)arg;
3507 				return (strdoioctl(stp, &strioc, flag,
3508 				    copyflag, crp, rvalp));
3509 
3510 			case TIOCSSOFTCAR:
3511 			case KIOCTRANS:
3512 			case KIOCTRANSABLE:
3513 			case KIOCCMD:
3514 			case KIOCSDIRECT:
3515 			case KIOCSCOMPAT:
3516 			case KIOCSKABORTEN:
3517 			case KIOCSRPTDELAY:
3518 			case KIOCSRPTRATE:
3519 			case VUIDSFORMAT:
3520 			case TIOCSPPS:
3521 				strioc.ic_len = sizeof (int);
3522 				strioc.ic_dp = (char *)arg;
3523 				return (strdoioctl(stp, &strioc, flag,
3524 				    copyflag, crp, rvalp));
3525 
3526 			case KIOCSETKEY:
3527 			case KIOCGETKEY:
3528 				strioc.ic_len = sizeof (struct kiockey);
3529 				strioc.ic_dp = (char *)arg;
3530 				return (strdoioctl(stp, &strioc, flag,
3531 				    copyflag, crp, rvalp));
3532 
3533 			case KIOCSKEY:
3534 			case KIOCGKEY:
3535 				strioc.ic_len = sizeof (struct kiockeymap);
3536 				strioc.ic_dp = (char *)arg;
3537 				return (strdoioctl(stp, &strioc, flag,
3538 				    copyflag, crp, rvalp));
3539 
3540 			case KIOCSLED:
3541 				/* arg is a pointer to char */
3542 				strioc.ic_len = sizeof (char);
3543 				strioc.ic_dp = (char *)arg;
3544 				return (strdoioctl(stp, &strioc, flag,
3545 				    copyflag, crp, rvalp));
3546 
3547 			case MSIOSETPARMS:
3548 				strioc.ic_len = sizeof (Ms_parms);
3549 				strioc.ic_dp = (char *)arg;
3550 				return (strdoioctl(stp, &strioc, flag,
3551 				    copyflag, crp, rvalp));
3552 
3553 			case VUIDSADDR:
3554 			case VUIDGADDR:
3555 				strioc.ic_len = sizeof (struct vuid_addr_probe);
3556 				strioc.ic_dp = (char *)arg;
3557 				return (strdoioctl(stp, &strioc, flag,
3558 				    copyflag, crp, rvalp));
3559 
3560 			/*
3561 			 * These M_IOCTL's don't require any data to be sent
3562 			 * downstream, and the driver will allocate and link
3563 			 * on its own mblk_t upon M_IOCACK -- thus we set
3564 			 * ic_len to zero and set ic_dp to arg so we know
3565 			 * where to copyout to later.
3566 			 */
3567 			case TIOCGSOFTCAR:
3568 			case TIOCGWINSZ:
3569 			case TIOCGSIZE:
3570 			case KIOCGTRANS:
3571 			case KIOCGTRANSABLE:
3572 			case KIOCTYPE:
3573 			case KIOCGDIRECT:
3574 			case KIOCGCOMPAT:
3575 			case KIOCLAYOUT:
3576 			case KIOCGLED:
3577 			case MSIOGETPARMS:
3578 			case MSIOBUTTONS:
3579 			case VUIDGFORMAT:
3580 			case TIOCGPPS:
3581 			case TIOCGPPSEV:
3582 			case TCGETA:
3583 			case TCGETS:
3584 			case LDGETT:
3585 			case TIOCGETP:
3586 			case KIOCGRPTDELAY:
3587 			case KIOCGRPTRATE:
3588 				strioc.ic_len = 0;
3589 				strioc.ic_dp = (char *)arg;
3590 				return (strdoioctl(stp, &strioc, flag,
3591 				    copyflag, crp, rvalp));
3592 			}
3593 		}
3594 
3595 		/*
3596 		 * Unknown cmd - send it down as a transparent ioctl.
3597 		 */
3598 		strioc.ic_cmd = cmd;
3599 		strioc.ic_timout = INFTIM;
3600 		strioc.ic_len = TRANSPARENT;
3601 		strioc.ic_dp = (char *)&arg;
3602 
3603 		return (strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp));
3604 
3605 	case I_STR:
3606 		/*
3607 		 * Stream ioctl.  Read in an strioctl buffer from the user
3608 		 * along with any data specified and send it downstream.
3609 		 * Strdoioctl will wait allow only one ioctl message at
3610 		 * a time, and waits for the acknowledgement.
3611 		 */
3612 
3613 		if (stp->sd_flag & STRHUP)
3614 			return (ENXIO);
3615 
3616 		error = strcopyin_strioctl((void *)arg, &strioc, flag,
3617 		    copyflag);
3618 		if (error != 0)
3619 			return (error);
3620 
3621 		if ((strioc.ic_len < 0) || (strioc.ic_timout < -1))
3622 			return (EINVAL);
3623 
3624 		access = job_control_type(strioc.ic_cmd);
3625 		mutex_enter(&stp->sd_lock);
3626 		if ((access != -1) &&
3627 		    ((error = i_straccess(stp, access)) != 0)) {
3628 			mutex_exit(&stp->sd_lock);
3629 			return (error);
3630 		}
3631 		mutex_exit(&stp->sd_lock);
3632 
3633 		/*
3634 		 * The I_STR facility provides a trap door for malicious
3635 		 * code to send down bogus streamio(7I) ioctl commands to
3636 		 * unsuspecting STREAMS modules and drivers which expect to
3637 		 * only get these messages from the stream head.
3638 		 * Explicitly prohibit any streamio ioctls which can be
3639 		 * passed downstream by the stream head.  Note that we do
3640 		 * not block all streamio ioctls because the ioctl
3641 		 * numberspace is not well managed and thus it's possible
3642 		 * that a module or driver's ioctl numbers may accidentally
3643 		 * collide with them.
3644 		 */
3645 		switch (strioc.ic_cmd) {
3646 		case I_LINK:
3647 		case I_PLINK:
3648 		case I_UNLINK:
3649 		case I_PUNLINK:
3650 		case _I_GETPEERCRED:
3651 		case _I_PLINK_LH:
3652 			return (EINVAL);
3653 		}
3654 
3655 		error = strdoioctl(stp, &strioc, flag, copyflag, crp, rvalp);
3656 		if (error == 0) {
3657 			error = strcopyout_strioctl(&strioc, (void *)arg,
3658 			    flag, copyflag);
3659 		}
3660 		return (error);
3661 
3662 	case _I_CMD:
3663 		/*
3664 		 * Like I_STR, but without using M_IOC* messages and without
3665 		 * copyins/copyouts beyond the passed-in argument.
3666 		 */
3667 		if (stp->sd_flag & STRHUP)
3668 			return (ENXIO);
3669 
3670 		if ((scp = kmem_alloc(sizeof (strcmd_t), KM_NOSLEEP)) == NULL)
3671 			return (ENOMEM);
3672 
3673 		if (copyin((void *)arg, scp, sizeof (strcmd_t))) {
3674 			kmem_free(scp, sizeof (strcmd_t));
3675 			return (EFAULT);
3676 		}
3677 
3678 		access = job_control_type(scp->sc_cmd);
3679 		mutex_enter(&stp->sd_lock);
3680 		if (access != -1 && (error = i_straccess(stp, access)) != 0) {
3681 			mutex_exit(&stp->sd_lock);
3682 			kmem_free(scp, sizeof (strcmd_t));
3683 			return (error);
3684 		}
3685 		mutex_exit(&stp->sd_lock);
3686 
3687 		*rvalp = 0;
3688 		if ((error = strdocmd(stp, scp, crp)) == 0) {
3689 			if (copyout(scp, (void *)arg, sizeof (strcmd_t)))
3690 				error = EFAULT;
3691 		}
3692 		kmem_free(scp, sizeof (strcmd_t));
3693 		return (error);
3694 
3695 	case I_NREAD:
3696 		/*
3697 		 * Return number of bytes of data in first message
3698 		 * in queue in "arg" and return the number of messages
3699 		 * in queue in return value.
3700 		 */
3701 	{
3702 		size_t	size;
3703 		int	retval;
3704 		int	count = 0;
3705 
3706 		mutex_enter(QLOCK(rdq));
3707 
3708 		size = msgdsize(rdq->q_first);
3709 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3710 			count++;
3711 
3712 		mutex_exit(QLOCK(rdq));
3713 		if (stp->sd_struiordq) {
3714 			infod_t infod;
3715 
3716 			infod.d_cmd = INFOD_COUNT;
3717 			infod.d_count = 0;
3718 			if (count == 0) {
3719 				infod.d_cmd |= INFOD_FIRSTBYTES;
3720 				infod.d_bytes = 0;
3721 			}
3722 			infod.d_res = 0;
3723 			(void) infonext(rdq, &infod);
3724 			count += infod.d_count;
3725 			if (infod.d_res & INFOD_FIRSTBYTES)
3726 				size = infod.d_bytes;
3727 		}
3728 
3729 		/*
3730 		 * Drop down from size_t to the "int" required by the
3731 		 * interface.  Cap at INT_MAX.
3732 		 */
3733 		retval = MIN(size, INT_MAX);
3734 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3735 		    copyflag);
3736 		if (!error)
3737 			*rvalp = count;
3738 		return (error);
3739 	}
3740 
3741 	case FIONREAD:
3742 		/*
3743 		 * Return number of bytes of data in all data messages
3744 		 * in queue in "arg".
3745 		 */
3746 	{
3747 		size_t	size = 0;
3748 		int	retval;
3749 
3750 		mutex_enter(QLOCK(rdq));
3751 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3752 			size += msgdsize(mp);
3753 		mutex_exit(QLOCK(rdq));
3754 
3755 		if (stp->sd_struiordq) {
3756 			infod_t infod;
3757 
3758 			infod.d_cmd = INFOD_BYTES;
3759 			infod.d_res = 0;
3760 			infod.d_bytes = 0;
3761 			(void) infonext(rdq, &infod);
3762 			size += infod.d_bytes;
3763 		}
3764 
3765 		/*
3766 		 * Drop down from size_t to the "int" required by the
3767 		 * interface.  Cap at INT_MAX.
3768 		 */
3769 		retval = MIN(size, INT_MAX);
3770 		error = strcopyout(&retval, (void *)arg, sizeof (retval),
3771 		    copyflag);
3772 
3773 		*rvalp = 0;
3774 		return (error);
3775 	}
3776 	case FIORDCHK:
3777 		/*
3778 		 * FIORDCHK does not use arg value (like FIONREAD),
3779 		 * instead a count is returned. I_NREAD value may
3780 		 * not be accurate but safe. The real thing to do is
3781 		 * to add the msgdsizes of all data  messages until
3782 		 * a non-data message.
3783 		 */
3784 	{
3785 		size_t size = 0;
3786 
3787 		mutex_enter(QLOCK(rdq));
3788 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
3789 			size += msgdsize(mp);
3790 		mutex_exit(QLOCK(rdq));
3791 
3792 		if (stp->sd_struiordq) {
3793 			infod_t infod;
3794 
3795 			infod.d_cmd = INFOD_BYTES;
3796 			infod.d_res = 0;
3797 			infod.d_bytes = 0;
3798 			(void) infonext(rdq, &infod);
3799 			size += infod.d_bytes;
3800 		}
3801 
3802 		/*
3803 		 * Since ioctl returns an int, and memory sizes under
3804 		 * LP64 may not fit, we return INT_MAX if the count was
3805 		 * actually greater.
3806 		 */
3807 		*rvalp = MIN(size, INT_MAX);
3808 		return (0);
3809 	}
3810 
3811 	case I_FIND:
3812 		/*
3813 		 * Get module name.
3814 		 */
3815 	{
3816 		char mname[FMNAMESZ + 1];
3817 		queue_t *q;
3818 
3819 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3820 		    mname, FMNAMESZ + 1, NULL);
3821 		if (error)
3822 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3823 
3824 		/*
3825 		 * Return EINVAL if we're handed a bogus module name.
3826 		 */
3827 		if (fmodsw_find(mname, FMODSW_LOAD) == NULL) {
3828 			TRACE_0(TR_FAC_STREAMS_FR,
3829 			    TR_I_CANT_FIND, "couldn't I_FIND");
3830 			return (EINVAL);
3831 		}
3832 
3833 		*rvalp = 0;
3834 
3835 		/* Look downstream to see if module is there. */
3836 		claimstr(stp->sd_wrq);
3837 		for (q = stp->sd_wrq->q_next; q; q = q->q_next) {
3838 			if (q->q_flag&QREADR) {
3839 				q = NULL;
3840 				break;
3841 			}
3842 			if (strcmp(mname, q->q_qinfo->qi_minfo->mi_idname) == 0)
3843 				break;
3844 		}
3845 		releasestr(stp->sd_wrq);
3846 
3847 		*rvalp = (q ? 1 : 0);
3848 		return (error);
3849 	}
3850 
3851 	case I_PUSH:
3852 	case __I_PUSH_NOCTTY:
3853 		/*
3854 		 * Push a module.
3855 		 * For the case __I_PUSH_NOCTTY push a module but
3856 		 * do not allocate controlling tty. See bugid 4025044
3857 		 */
3858 
3859 	{
3860 		char mname[FMNAMESZ + 1];
3861 		fmodsw_impl_t *fp;
3862 		dev_t dummydev;
3863 
3864 		if (stp->sd_flag & STRHUP)
3865 			return (ENXIO);
3866 
3867 		/*
3868 		 * Get module name and look up in fmodsw.
3869 		 */
3870 		error = (copyflag & U_TO_K ? copyinstr : copystr)((void *)arg,
3871 		    mname, FMNAMESZ + 1, NULL);
3872 		if (error)
3873 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
3874 
3875 		if ((fp = fmodsw_find(mname, FMODSW_HOLD | FMODSW_LOAD)) ==
3876 		    NULL)
3877 			return (EINVAL);
3878 
3879 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_PUSH,
3880 		    "I_PUSH:fp %p stp %p", fp, stp);
3881 
3882 		if (error = strstartplumb(stp, flag, cmd)) {
3883 			fmodsw_rele(fp);
3884 			return (error);
3885 		}
3886 
3887 		/*
3888 		 * See if any more modules can be pushed on this stream.
3889 		 * Note that this check must be done after strstartplumb()
3890 		 * since otherwise multiple threads issuing I_PUSHes on
3891 		 * the same stream will be able to exceed nstrpush.
3892 		 */
3893 		mutex_enter(&stp->sd_lock);
3894 		if (stp->sd_pushcnt >= nstrpush) {
3895 			fmodsw_rele(fp);
3896 			strendplumb(stp);
3897 			mutex_exit(&stp->sd_lock);
3898 			return (EINVAL);
3899 		}
3900 		mutex_exit(&stp->sd_lock);
3901 
3902 		/*
3903 		 * Push new module and call its open routine
3904 		 * via qattach().  Modules don't change device
3905 		 * numbers, so just ignore dummydev here.
3906 		 */
3907 		dummydev = vp->v_rdev;
3908 		if ((error = qattach(rdq, &dummydev, 0, crp, fp,
3909 		    B_FALSE)) == 0) {
3910 			if (vp->v_type == VCHR && /* sorry, no pipes allowed */
3911 			    (cmd == I_PUSH) && (stp->sd_flag & STRISTTY)) {
3912 				/*
3913 				 * try to allocate it as a controlling terminal
3914 				 */
3915 				(void) strctty(stp);
3916 			}
3917 		}
3918 
3919 		mutex_enter(&stp->sd_lock);
3920 
3921 		/*
3922 		 * As a performance concern we are caching the values of
3923 		 * q_minpsz and q_maxpsz of the module below the stream
3924 		 * head in the stream head.
3925 		 */
3926 		mutex_enter(QLOCK(stp->sd_wrq->q_next));
3927 		rmin = stp->sd_wrq->q_next->q_minpsz;
3928 		rmax = stp->sd_wrq->q_next->q_maxpsz;
3929 		mutex_exit(QLOCK(stp->sd_wrq->q_next));
3930 
3931 		/* Do this processing here as a performance concern */
3932 		if (strmsgsz != 0) {
3933 			if (rmax == INFPSZ)
3934 				rmax = strmsgsz;
3935 			else  {
3936 				if (vp->v_type == VFIFO)
3937 					rmax = MIN(PIPE_BUF, rmax);
3938 				else	rmax = MIN(strmsgsz, rmax);
3939 			}
3940 		}
3941 
3942 		mutex_enter(QLOCK(wrq));
3943 		stp->sd_qn_minpsz = rmin;
3944 		stp->sd_qn_maxpsz = rmax;
3945 		mutex_exit(QLOCK(wrq));
3946 
3947 		strendplumb(stp);
3948 		mutex_exit(&stp->sd_lock);
3949 		return (error);
3950 	}
3951 
3952 	case I_POP:
3953 	{
3954 		queue_t	*q;
3955 
3956 		if (stp->sd_flag & STRHUP)
3957 			return (ENXIO);
3958 		if (!wrq->q_next)	/* for broken pipes */
3959 			return (EINVAL);
3960 
3961 		if (error = strstartplumb(stp, flag, cmd))
3962 			return (error);
3963 
3964 		/*
3965 		 * If there is an anchor on this stream and popping
3966 		 * the current module would attempt to pop through the
3967 		 * anchor, then disallow the pop unless we have sufficient
3968 		 * privileges; take the cheapest (non-locking) check
3969 		 * first.
3970 		 */
3971 		if (secpolicy_ip_config(crp, B_TRUE) != 0 ||
3972 		    (stp->sd_anchorzone != crgetzoneid(crp))) {
3973 			mutex_enter(&stp->sd_lock);
3974 			/*
3975 			 * Anchors only apply if there's at least one
3976 			 * module on the stream (sd_pushcnt > 0).
3977 			 */
3978 			if (stp->sd_pushcnt > 0 &&
3979 			    stp->sd_pushcnt == stp->sd_anchor &&
3980 			    stp->sd_vnode->v_type != VFIFO) {
3981 				strendplumb(stp);
3982 				mutex_exit(&stp->sd_lock);
3983 				if (stp->sd_anchorzone != crgetzoneid(crp))
3984 					return (EINVAL);
3985 				/* Audit and report error */
3986 				return (secpolicy_ip_config(crp, B_FALSE));
3987 			}
3988 			mutex_exit(&stp->sd_lock);
3989 		}
3990 
3991 		q = wrq->q_next;
3992 		TRACE_2(TR_FAC_STREAMS_FR, TR_I_POP,
3993 		    "I_POP:%p from %p", q, stp);
3994 		if (q->q_next == NULL || (q->q_flag & (QREADR|QISDRV))) {
3995 			error = EINVAL;
3996 		} else {
3997 			qdetach(_RD(q), 1, flag, crp, B_FALSE);
3998 			error = 0;
3999 		}
4000 		mutex_enter(&stp->sd_lock);
4001 
4002 		/*
4003 		 * As a performance concern we are caching the values of
4004 		 * q_minpsz and q_maxpsz of the module below the stream
4005 		 * head in the stream head.
4006 		 */
4007 		mutex_enter(QLOCK(wrq->q_next));
4008 		rmin = wrq->q_next->q_minpsz;
4009 		rmax = wrq->q_next->q_maxpsz;
4010 		mutex_exit(QLOCK(wrq->q_next));
4011 
4012 		/* Do this processing here as a performance concern */
4013 		if (strmsgsz != 0) {
4014 			if (rmax == INFPSZ)
4015 				rmax = strmsgsz;
4016 			else  {
4017 				if (vp->v_type == VFIFO)
4018 					rmax = MIN(PIPE_BUF, rmax);
4019 				else	rmax = MIN(strmsgsz, rmax);
4020 			}
4021 		}
4022 
4023 		mutex_enter(QLOCK(wrq));
4024 		stp->sd_qn_minpsz = rmin;
4025 		stp->sd_qn_maxpsz = rmax;
4026 		mutex_exit(QLOCK(wrq));
4027 
4028 		/* If we popped through the anchor, then reset the anchor. */
4029 		if (stp->sd_pushcnt < stp->sd_anchor) {
4030 			stp->sd_anchor = 0;
4031 			stp->sd_anchorzone = 0;
4032 		}
4033 		strendplumb(stp);
4034 		mutex_exit(&stp->sd_lock);
4035 		return (error);
4036 	}
4037 
4038 	case _I_MUXID2FD:
4039 	{
4040 		/*
4041 		 * Create a fd for a I_PLINK'ed lower stream with a given
4042 		 * muxid.  With the fd, application can send down ioctls,
4043 		 * like I_LIST, to the previously I_PLINK'ed stream.  Note
4044 		 * that after getting the fd, the application has to do an
4045 		 * I_PUNLINK on the muxid before it can do any operation
4046 		 * on the lower stream.  This is required by spec1170.
4047 		 *
4048 		 * The fd used to do this ioctl should point to the same
4049 		 * controlling device used to do the I_PLINK.  If it uses
4050 		 * a different stream or an invalid muxid, I_MUXID2FD will
4051 		 * fail.  The error code is set to EINVAL.
4052 		 *
4053 		 * The intended use of this interface is the following.
4054 		 * An application I_PLINK'ed a stream and exits.  The fd
4055 		 * to the lower stream is gone.  Another application
4056 		 * wants to get a fd to the lower stream, it uses I_MUXID2FD.
4057 		 */
4058 		int muxid = (int)arg;
4059 		int fd;
4060 		linkinfo_t *linkp;
4061 		struct file *fp;
4062 		netstack_t *ns;
4063 		str_stack_t *ss;
4064 
4065 		/*
4066 		 * Do not allow the wildcard muxid.  This ioctl is not
4067 		 * intended to find arbitrary link.
4068 		 */
4069 		if (muxid == 0) {
4070 			return (EINVAL);
4071 		}
4072 
4073 		ns = netstack_find_by_cred(crp);
4074 		ASSERT(ns != NULL);
4075 		ss = ns->netstack_str;
4076 		ASSERT(ss != NULL);
4077 
4078 		mutex_enter(&muxifier);
4079 		linkp = findlinks(vp->v_stream, muxid, LINKPERSIST, ss);
4080 		if (linkp == NULL) {
4081 			mutex_exit(&muxifier);
4082 			netstack_rele(ss->ss_netstack);
4083 			return (EINVAL);
4084 		}
4085 
4086 		if ((fd = ufalloc(0)) == -1) {
4087 			mutex_exit(&muxifier);
4088 			netstack_rele(ss->ss_netstack);
4089 			return (EMFILE);
4090 		}
4091 		fp = linkp->li_fpdown;
4092 		mutex_enter(&fp->f_tlock);
4093 		fp->f_count++;
4094 		mutex_exit(&fp->f_tlock);
4095 		mutex_exit(&muxifier);
4096 		setf(fd, fp);
4097 		*rvalp = fd;
4098 		netstack_rele(ss->ss_netstack);
4099 		return (0);
4100 	}
4101 
4102 	case _I_INSERT:
4103 	{
4104 		/*
4105 		 * To insert a module to a given position in a stream.
4106 		 * In the first release, only allow privileged user
4107 		 * to use this ioctl. Furthermore, the insert is only allowed
4108 		 * below an anchor if the zoneid is the same as the zoneid
4109 		 * which created the anchor.
4110 		 *
4111 		 * Note that we do not plan to support this ioctl
4112 		 * on pipes in the first release.  We want to learn more
4113 		 * about the implications of these ioctls before extending
4114 		 * their support.  And we do not think these features are
4115 		 * valuable for pipes.
4116 		 *
4117 		 * Neither do we support O/C hot stream.  Note that only
4118 		 * the upper streams of TCP/IP stack are O/C hot streams.
4119 		 * The lower IP stream is not.
4120 		 * When there is a O/C cold barrier, we only allow inserts
4121 		 * above the barrier.
4122 		 */
4123 		STRUCT_DECL(strmodconf, strmodinsert);
4124 		char mod_name[FMNAMESZ + 1];
4125 		fmodsw_impl_t *fp;
4126 		dev_t dummydev;
4127 		queue_t *tmp_wrq;
4128 		int pos;
4129 		boolean_t is_insert;
4130 
4131 		STRUCT_INIT(strmodinsert, flag);
4132 		if (stp->sd_flag & STRHUP)
4133 			return (ENXIO);
4134 		if (STRMATED(stp))
4135 			return (EINVAL);
4136 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4137 			return (error);
4138 		if (stp->sd_anchor != 0 &&
4139 		    stp->sd_anchorzone != crgetzoneid(crp))
4140 			return (EINVAL);
4141 
4142 		error = strcopyin((void *)arg, STRUCT_BUF(strmodinsert),
4143 		    STRUCT_SIZE(strmodinsert), copyflag);
4144 		if (error)
4145 			return (error);
4146 
4147 		/*
4148 		 * Get module name and look up in fmodsw.
4149 		 */
4150 		error = (copyflag & U_TO_K ? copyinstr :
4151 		    copystr)(STRUCT_FGETP(strmodinsert, mod_name),
4152 		    mod_name, FMNAMESZ + 1, NULL);
4153 		if (error)
4154 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4155 
4156 		if ((fp = fmodsw_find(mod_name, FMODSW_HOLD | FMODSW_LOAD)) ==
4157 		    NULL)
4158 			return (EINVAL);
4159 
4160 		if (error = strstartplumb(stp, flag, cmd)) {
4161 			fmodsw_rele(fp);
4162 			return (error);
4163 		}
4164 
4165 		/*
4166 		 * Is this _I_INSERT just like an I_PUSH?  We need to know
4167 		 * this because we do some optimizations if this is a
4168 		 * module being pushed.
4169 		 */
4170 		pos = STRUCT_FGET(strmodinsert, pos);
4171 		is_insert = (pos != 0);
4172 
4173 		/*
4174 		 * Make sure pos is valid.  Even though it is not an I_PUSH,
4175 		 * we impose the same limit on the number of modules in a
4176 		 * stream.
4177 		 */
4178 		mutex_enter(&stp->sd_lock);
4179 		if (stp->sd_pushcnt >= nstrpush || pos < 0 ||
4180 		    pos > stp->sd_pushcnt) {
4181 			fmodsw_rele(fp);
4182 			strendplumb(stp);
4183 			mutex_exit(&stp->sd_lock);
4184 			return (EINVAL);
4185 		}
4186 		if (stp->sd_anchor != 0) {
4187 			/*
4188 			 * Is this insert below the anchor?
4189 			 * Pushcnt hasn't been increased yet hence
4190 			 * we test for greater than here, and greater or
4191 			 * equal after qattach.
4192 			 */
4193 			if (pos > (stp->sd_pushcnt - stp->sd_anchor) &&
4194 			    stp->sd_anchorzone != crgetzoneid(crp)) {
4195 				fmodsw_rele(fp);
4196 				strendplumb(stp);
4197 				mutex_exit(&stp->sd_lock);
4198 				return (EPERM);
4199 			}
4200 		}
4201 
4202 		mutex_exit(&stp->sd_lock);
4203 
4204 		/*
4205 		 * First find the correct position this module to
4206 		 * be inserted.  We don't need to call claimstr()
4207 		 * as the stream should not be changing at this point.
4208 		 *
4209 		 * Insert new module and call its open routine
4210 		 * via qattach().  Modules don't change device
4211 		 * numbers, so just ignore dummydev here.
4212 		 */
4213 		for (tmp_wrq = stp->sd_wrq; pos > 0;
4214 		    tmp_wrq = tmp_wrq->q_next, pos--) {
4215 			ASSERT(SAMESTR(tmp_wrq));
4216 		}
4217 		dummydev = vp->v_rdev;
4218 		if ((error = qattach(_RD(tmp_wrq), &dummydev, 0, crp,
4219 		    fp, is_insert)) != 0) {
4220 			mutex_enter(&stp->sd_lock);
4221 			strendplumb(stp);
4222 			mutex_exit(&stp->sd_lock);
4223 			return (error);
4224 		}
4225 
4226 		mutex_enter(&stp->sd_lock);
4227 
4228 		/*
4229 		 * As a performance concern we are caching the values of
4230 		 * q_minpsz and q_maxpsz of the module below the stream
4231 		 * head in the stream head.
4232 		 */
4233 		if (!is_insert) {
4234 			mutex_enter(QLOCK(stp->sd_wrq->q_next));
4235 			rmin = stp->sd_wrq->q_next->q_minpsz;
4236 			rmax = stp->sd_wrq->q_next->q_maxpsz;
4237 			mutex_exit(QLOCK(stp->sd_wrq->q_next));
4238 
4239 			/* Do this processing here as a performance concern */
4240 			if (strmsgsz != 0) {
4241 				if (rmax == INFPSZ) {
4242 					rmax = strmsgsz;
4243 				} else  {
4244 					rmax = MIN(strmsgsz, rmax);
4245 				}
4246 			}
4247 
4248 			mutex_enter(QLOCK(wrq));
4249 			stp->sd_qn_minpsz = rmin;
4250 			stp->sd_qn_maxpsz = rmax;
4251 			mutex_exit(QLOCK(wrq));
4252 		}
4253 
4254 		/*
4255 		 * Need to update the anchor value if this module is
4256 		 * inserted below the anchor point.
4257 		 */
4258 		if (stp->sd_anchor != 0) {
4259 			pos = STRUCT_FGET(strmodinsert, pos);
4260 			if (pos >= (stp->sd_pushcnt - stp->sd_anchor))
4261 				stp->sd_anchor++;
4262 		}
4263 
4264 		strendplumb(stp);
4265 		mutex_exit(&stp->sd_lock);
4266 		return (0);
4267 	}
4268 
4269 	case _I_REMOVE:
4270 	{
4271 		/*
4272 		 * To remove a module with a given name in a stream.  The
4273 		 * caller of this ioctl needs to provide both the name and
4274 		 * the position of the module to be removed.  This eliminates
4275 		 * the ambiguity of removal if a module is inserted/pushed
4276 		 * multiple times in a stream.  In the first release, only
4277 		 * allow privileged user to use this ioctl.
4278 		 * Furthermore, the remove is only allowed
4279 		 * below an anchor if the zoneid is the same as the zoneid
4280 		 * which created the anchor.
4281 		 *
4282 		 * Note that we do not plan to support this ioctl
4283 		 * on pipes in the first release.  We want to learn more
4284 		 * about the implications of these ioctls before extending
4285 		 * their support.  And we do not think these features are
4286 		 * valuable for pipes.
4287 		 *
4288 		 * Neither do we support O/C hot stream.  Note that only
4289 		 * the upper streams of TCP/IP stack are O/C hot streams.
4290 		 * The lower IP stream is not.
4291 		 * When there is a O/C cold barrier we do not allow removal
4292 		 * below the barrier.
4293 		 *
4294 		 * Also note that _I_REMOVE cannot be used to remove a
4295 		 * driver or the stream head.
4296 		 */
4297 		STRUCT_DECL(strmodconf, strmodremove);
4298 		queue_t	*q;
4299 		int pos;
4300 		char mod_name[FMNAMESZ + 1];
4301 		boolean_t is_remove;
4302 
4303 		STRUCT_INIT(strmodremove, flag);
4304 		if (stp->sd_flag & STRHUP)
4305 			return (ENXIO);
4306 		if (STRMATED(stp))
4307 			return (EINVAL);
4308 		if ((error = secpolicy_net_config(crp, B_FALSE)) != 0)
4309 			return (error);
4310 		if (stp->sd_anchor != 0 &&
4311 		    stp->sd_anchorzone != crgetzoneid(crp))
4312 			return (EINVAL);
4313 
4314 		error = strcopyin((void *)arg, STRUCT_BUF(strmodremove),
4315 		    STRUCT_SIZE(strmodremove), copyflag);
4316 		if (error)
4317 			return (error);
4318 
4319 		error = (copyflag & U_TO_K ? copyinstr :
4320 		    copystr)(STRUCT_FGETP(strmodremove, mod_name),
4321 		    mod_name, FMNAMESZ + 1, NULL);
4322 		if (error)
4323 			return ((error == ENAMETOOLONG) ? EINVAL : EFAULT);
4324 
4325 		if ((error = strstartplumb(stp, flag, cmd)) != 0)
4326 			return (error);
4327 
4328 		/*
4329 		 * Match the name of given module to the name of module at
4330 		 * the given position.
4331 		 */
4332 		pos = STRUCT_FGET(strmodremove, pos);
4333 
4334 		is_remove = (pos != 0);
4335 		for (q = stp->sd_wrq->q_next; SAMESTR(q) && pos > 0;
4336 		    q = q->q_next, pos--)
4337 			;
4338 		if (pos > 0 || ! SAMESTR(q) ||
4339 		    strncmp(q->q_qinfo->qi_minfo->mi_idname, mod_name,
4340 		    strlen(q->q_qinfo->qi_minfo->mi_idname)) != 0) {
4341 			mutex_enter(&stp->sd_lock);
4342 			strendplumb(stp);
4343 			mutex_exit(&stp->sd_lock);
4344 			return (EINVAL);
4345 		}
4346 
4347 		/*
4348 		 * If the position is at or below an anchor, then the zoneid
4349 		 * must match the zoneid that created the anchor.
4350 		 */
4351 		if (stp->sd_anchor != 0) {
4352 			pos = STRUCT_FGET(strmodremove, pos);
4353 			if (pos >= (stp->sd_pushcnt - stp->sd_anchor) &&
4354 			    stp->sd_anchorzone != crgetzoneid(crp)) {
4355 				mutex_enter(&stp->sd_lock);
4356 				strendplumb(stp);
4357 				mutex_exit(&stp->sd_lock);
4358 				return (EPERM);
4359 			}
4360 		}
4361 
4362 
4363 		ASSERT(!(q->q_flag & QREADR));
4364 		qdetach(_RD(q), 1, flag, crp, is_remove);
4365 
4366 		mutex_enter(&stp->sd_lock);
4367 
4368 		/*
4369 		 * As a performance concern we are caching the values of
4370 		 * q_minpsz and q_maxpsz of the module below the stream
4371 		 * head in the stream head.
4372 		 */
4373 		if (!is_remove) {
4374 			mutex_enter(QLOCK(wrq->q_next));
4375 			rmin = wrq->q_next->q_minpsz;
4376 			rmax = wrq->q_next->q_maxpsz;
4377 			mutex_exit(QLOCK(wrq->q_next));
4378 
4379 			/* Do this processing here as a performance concern */
4380 			if (strmsgsz != 0) {
4381 				if (rmax == INFPSZ)
4382 					rmax = strmsgsz;
4383 				else  {
4384 					if (vp->v_type == VFIFO)
4385 						rmax = MIN(PIPE_BUF, rmax);
4386 					else	rmax = MIN(strmsgsz, rmax);
4387 				}
4388 			}
4389 
4390 			mutex_enter(QLOCK(wrq));
4391 			stp->sd_qn_minpsz = rmin;
4392 			stp->sd_qn_maxpsz = rmax;
4393 			mutex_exit(QLOCK(wrq));
4394 		}
4395 
4396 		/*
4397 		 * Need to update the anchor value if this module is removed
4398 		 * at or below the anchor point.  If the removed module is at
4399 		 * the anchor point, remove the anchor for this stream if
4400 		 * there is no module above the anchor point.  Otherwise, if
4401 		 * the removed module is below the anchor point, decrement the
4402 		 * anchor point by 1.
4403 		 */
4404 		if (stp->sd_anchor != 0) {
4405 			pos = STRUCT_FGET(strmodremove, pos);
4406 			if (pos == stp->sd_pushcnt - stp->sd_anchor + 1)
4407 				stp->sd_anchor = 0;
4408 			else if (pos > (stp->sd_pushcnt - stp->sd_anchor + 1))
4409 				stp->sd_anchor--;
4410 		}
4411 
4412 		strendplumb(stp);
4413 		mutex_exit(&stp->sd_lock);
4414 		return (0);
4415 	}
4416 
4417 	case I_ANCHOR:
4418 		/*
4419 		 * Set the anchor position on the stream to reside at
4420 		 * the top module (in other words, the top module
4421 		 * cannot be popped).  Anchors with a FIFO make no
4422 		 * obvious sense, so they're not allowed.
4423 		 */
4424 		mutex_enter(&stp->sd_lock);
4425 
4426 		if (stp->sd_vnode->v_type == VFIFO) {
4427 			mutex_exit(&stp->sd_lock);
4428 			return (EINVAL);
4429 		}
4430 		/* Only allow the same zoneid to update the anchor */
4431 		if (stp->sd_anchor != 0 &&
4432 		    stp->sd_anchorzone != crgetzoneid(crp)) {
4433 			mutex_exit(&stp->sd_lock);
4434 			return (EINVAL);
4435 		}
4436 		stp->sd_anchor = stp->sd_pushcnt;
4437 		stp->sd_anchorzone = crgetzoneid(crp);
4438 		mutex_exit(&stp->sd_lock);
4439 		return (0);
4440 
4441 	case I_LOOK:
4442 		/*
4443 		 * Get name of first module downstream.
4444 		 * If no module, return an error.
4445 		 */
4446 	{
4447 		claimstr(wrq);
4448 		if (_SAMESTR(wrq) && wrq->q_next->q_next) {
4449 			char *name = wrq->q_next->q_qinfo->qi_minfo->mi_idname;
4450 			error = strcopyout(name, (void *)arg, strlen(name) + 1,
4451 			    copyflag);
4452 			releasestr(wrq);
4453 			return (error);
4454 		}
4455 		releasestr(wrq);
4456 		return (EINVAL);
4457 	}
4458 
4459 	case I_LINK:
4460 	case I_PLINK:
4461 		/*
4462 		 * Link a multiplexor.
4463 		 */
4464 		error = mlink(vp, cmd, (int)arg, crp, rvalp, 0);
4465 		return (error);
4466 
4467 	case _I_PLINK_LH:
4468 		/*
4469 		 * Link a multiplexor: Call must originate from kernel.
4470 		 */
4471 		if (kioctl)
4472 			return (ldi_mlink_lh(vp, cmd, arg, crp, rvalp));
4473 
4474 		return (EINVAL);
4475 	case I_UNLINK:
4476 	case I_PUNLINK:
4477 		/*
4478 		 * Unlink a multiplexor.
4479 		 * If arg is -1, unlink all links for which this is the
4480 		 * controlling stream.  Otherwise, arg is an index number
4481 		 * for a link to be removed.
4482 		 */
4483 	{
4484 		struct linkinfo *linkp;
4485 		int native_arg = (int)arg;
4486 		int type;
4487 		netstack_t *ns;
4488 		str_stack_t *ss;
4489 
4490 		TRACE_1(TR_FAC_STREAMS_FR,
4491 		    TR_I_UNLINK, "I_UNLINK/I_PUNLINK:%p", stp);
4492 		if (vp->v_type == VFIFO) {
4493 			return (EINVAL);
4494 		}
4495 		if (cmd == I_UNLINK)
4496 			type = LINKNORMAL;
4497 		else	/* I_PUNLINK */
4498 			type = LINKPERSIST;
4499 		if (native_arg == 0) {
4500 			return (EINVAL);
4501 		}
4502 		ns = netstack_find_by_cred(crp);
4503 		ASSERT(ns != NULL);
4504 		ss = ns->netstack_str;
4505 		ASSERT(ss != NULL);
4506 
4507 		if (native_arg == MUXID_ALL)
4508 			error = munlinkall(stp, type, crp, rvalp, ss);
4509 		else {
4510 			mutex_enter(&muxifier);
4511 			if (!(linkp = findlinks(stp, (int)arg, type, ss))) {
4512 				/* invalid user supplied index number */
4513 				mutex_exit(&muxifier);
4514 				netstack_rele(ss->ss_netstack);
4515 				return (EINVAL);
4516 			}
4517 			/* munlink drops the muxifier lock */
4518 			error = munlink(stp, linkp, type, crp, rvalp, ss);
4519 		}
4520 		netstack_rele(ss->ss_netstack);
4521 		return (error);
4522 	}
4523 
4524 	case I_FLUSH:
4525 		/*
4526 		 * send a flush message downstream
4527 		 * flush message can indicate
4528 		 * FLUSHR - flush read queue
4529 		 * FLUSHW - flush write queue
4530 		 * FLUSHRW - flush read/write queue
4531 		 */
4532 		if (stp->sd_flag & STRHUP)
4533 			return (ENXIO);
4534 		if (arg & ~FLUSHRW)
4535 			return (EINVAL);
4536 
4537 		for (;;) {
4538 			if (putnextctl1(stp->sd_wrq, M_FLUSH, (int)arg)) {
4539 				break;
4540 			}
4541 			if (error = strwaitbuf(1, BPRI_HI)) {
4542 				return (error);
4543 			}
4544 		}
4545 
4546 		/*
4547 		 * Send down an unsupported ioctl and wait for the nack
4548 		 * in order to allow the M_FLUSH to propagate back
4549 		 * up to the stream head.
4550 		 * Replaces if (qready()) runqueues();
4551 		 */
4552 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4553 		strioc.ic_timout = 0;
4554 		strioc.ic_len = 0;
4555 		strioc.ic_dp = NULL;
4556 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4557 		*rvalp = 0;
4558 		return (0);
4559 
4560 	case I_FLUSHBAND:
4561 	{
4562 		struct bandinfo binfo;
4563 
4564 		error = strcopyin((void *)arg, &binfo, sizeof (binfo),
4565 		    copyflag);
4566 		if (error)
4567 			return (error);
4568 		if (stp->sd_flag & STRHUP)
4569 			return (ENXIO);
4570 		if (binfo.bi_flag & ~FLUSHRW)
4571 			return (EINVAL);
4572 		while (!(mp = allocb(2, BPRI_HI))) {
4573 			if (error = strwaitbuf(2, BPRI_HI))
4574 				return (error);
4575 		}
4576 		mp->b_datap->db_type = M_FLUSH;
4577 		*mp->b_wptr++ = binfo.bi_flag | FLUSHBAND;
4578 		*mp->b_wptr++ = binfo.bi_pri;
4579 		putnext(stp->sd_wrq, mp);
4580 		/*
4581 		 * Send down an unsupported ioctl and wait for the nack
4582 		 * in order to allow the M_FLUSH to propagate back
4583 		 * up to the stream head.
4584 		 * Replaces if (qready()) runqueues();
4585 		 */
4586 		strioc.ic_cmd = -1;	/* The unsupported ioctl */
4587 		strioc.ic_timout = 0;
4588 		strioc.ic_len = 0;
4589 		strioc.ic_dp = NULL;
4590 		(void) strdoioctl(stp, &strioc, flag, K_TO_K, crp, rvalp);
4591 		*rvalp = 0;
4592 		return (0);
4593 	}
4594 
4595 	case I_SRDOPT:
4596 		/*
4597 		 * Set read options
4598 		 *
4599 		 * RNORM - default stream mode
4600 		 * RMSGN - message no discard
4601 		 * RMSGD - message discard
4602 		 * RPROTNORM - fail read with EBADMSG for M_[PC]PROTOs
4603 		 * RPROTDAT - convert M_[PC]PROTOs to M_DATAs
4604 		 * RPROTDIS - discard M_[PC]PROTOs and retain M_DATAs
4605 		 */
4606 		if (arg & ~(RMODEMASK | RPROTMASK))
4607 			return (EINVAL);
4608 
4609 		if ((arg & (RMSGD|RMSGN)) == (RMSGD|RMSGN))
4610 			return (EINVAL);
4611 
4612 		mutex_enter(&stp->sd_lock);
4613 		switch (arg & RMODEMASK) {
4614 		case RNORM:
4615 			stp->sd_read_opt &= ~(RD_MSGDIS | RD_MSGNODIS);
4616 			break;
4617 		case RMSGD:
4618 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGNODIS) |
4619 			    RD_MSGDIS;
4620 			break;
4621 		case RMSGN:
4622 			stp->sd_read_opt = (stp->sd_read_opt & ~RD_MSGDIS) |
4623 			    RD_MSGNODIS;
4624 			break;
4625 		}
4626 
4627 		switch (arg & RPROTMASK) {
4628 		case RPROTNORM:
4629 			stp->sd_read_opt &= ~(RD_PROTDAT | RD_PROTDIS);
4630 			break;
4631 
4632 		case RPROTDAT:
4633 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDIS) |
4634 			    RD_PROTDAT);
4635 			break;
4636 
4637 		case RPROTDIS:
4638 			stp->sd_read_opt = ((stp->sd_read_opt & ~RD_PROTDAT) |
4639 			    RD_PROTDIS);
4640 			break;
4641 		}
4642 		mutex_exit(&stp->sd_lock);
4643 		return (0);
4644 
4645 	case I_GRDOPT:
4646 		/*
4647 		 * Get read option and return the value
4648 		 * to spot pointed to by arg
4649 		 */
4650 	{
4651 		int rdopt;
4652 
4653 		rdopt = ((stp->sd_read_opt & RD_MSGDIS) ? RMSGD :
4654 		    ((stp->sd_read_opt & RD_MSGNODIS) ? RMSGN : RNORM));
4655 		rdopt |= ((stp->sd_read_opt & RD_PROTDAT) ? RPROTDAT :
4656 		    ((stp->sd_read_opt & RD_PROTDIS) ? RPROTDIS : RPROTNORM));
4657 
4658 		return (strcopyout(&rdopt, (void *)arg, sizeof (int),
4659 		    copyflag));
4660 	}
4661 
4662 	case I_SERROPT:
4663 		/*
4664 		 * Set error options
4665 		 *
4666 		 * RERRNORM - persistent read errors
4667 		 * RERRNONPERSIST - non-persistent read errors
4668 		 * WERRNORM - persistent write errors
4669 		 * WERRNONPERSIST - non-persistent write errors
4670 		 */
4671 		if (arg & ~(RERRMASK | WERRMASK))
4672 			return (EINVAL);
4673 
4674 		mutex_enter(&stp->sd_lock);
4675 		switch (arg & RERRMASK) {
4676 		case RERRNORM:
4677 			stp->sd_flag &= ~STRDERRNONPERSIST;
4678 			break;
4679 		case RERRNONPERSIST:
4680 			stp->sd_flag |= STRDERRNONPERSIST;
4681 			break;
4682 		}
4683 		switch (arg & WERRMASK) {
4684 		case WERRNORM:
4685 			stp->sd_flag &= ~STWRERRNONPERSIST;
4686 			break;
4687 		case WERRNONPERSIST:
4688 			stp->sd_flag |= STWRERRNONPERSIST;
4689 			break;
4690 		}
4691 		mutex_exit(&stp->sd_lock);
4692 		return (0);
4693 
4694 	case I_GERROPT:
4695 		/*
4696 		 * Get error option and return the value
4697 		 * to spot pointed to by arg
4698 		 */
4699 	{
4700 		int erropt = 0;
4701 
4702 		erropt |= (stp->sd_flag & STRDERRNONPERSIST) ? RERRNONPERSIST :
4703 		    RERRNORM;
4704 		erropt |= (stp->sd_flag & STWRERRNONPERSIST) ? WERRNONPERSIST :
4705 		    WERRNORM;
4706 		return (strcopyout(&erropt, (void *)arg, sizeof (int),
4707 		    copyflag));
4708 	}
4709 
4710 	case I_SETSIG:
4711 		/*
4712 		 * Register the calling proc to receive the SIGPOLL
4713 		 * signal based on the events given in arg.  If
4714 		 * arg is zero, remove the proc from register list.
4715 		 */
4716 	{
4717 		strsig_t *ssp, *pssp;
4718 		struct pid *pidp;
4719 
4720 		pssp = NULL;
4721 		pidp = curproc->p_pidp;
4722 		/*
4723 		 * Hold sd_lock to prevent traversal of sd_siglist while
4724 		 * it is modified.
4725 		 */
4726 		mutex_enter(&stp->sd_lock);
4727 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pidp != pidp);
4728 		    pssp = ssp, ssp = ssp->ss_next)
4729 			;
4730 
4731 		if (arg) {
4732 			if (arg & ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4733 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4734 				mutex_exit(&stp->sd_lock);
4735 				return (EINVAL);
4736 			}
4737 			if ((arg & S_BANDURG) && !(arg & S_RDBAND)) {
4738 				mutex_exit(&stp->sd_lock);
4739 				return (EINVAL);
4740 			}
4741 
4742 			/*
4743 			 * If proc not already registered, add it
4744 			 * to list.
4745 			 */
4746 			if (!ssp) {
4747 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4748 				ssp->ss_pidp = pidp;
4749 				ssp->ss_pid = pidp->pid_id;
4750 				ssp->ss_next = NULL;
4751 				if (pssp)
4752 					pssp->ss_next = ssp;
4753 				else
4754 					stp->sd_siglist = ssp;
4755 				mutex_enter(&pidlock);
4756 				PID_HOLD(pidp);
4757 				mutex_exit(&pidlock);
4758 			}
4759 
4760 			/*
4761 			 * Set events.
4762 			 */
4763 			ssp->ss_events = (int)arg;
4764 		} else {
4765 			/*
4766 			 * Remove proc from register list.
4767 			 */
4768 			if (ssp) {
4769 				mutex_enter(&pidlock);
4770 				PID_RELE(pidp);
4771 				mutex_exit(&pidlock);
4772 				if (pssp)
4773 					pssp->ss_next = ssp->ss_next;
4774 				else
4775 					stp->sd_siglist = ssp->ss_next;
4776 				kmem_free(ssp, sizeof (strsig_t));
4777 			} else {
4778 				mutex_exit(&stp->sd_lock);
4779 				return (EINVAL);
4780 			}
4781 		}
4782 
4783 		/*
4784 		 * Recalculate OR of sig events.
4785 		 */
4786 		stp->sd_sigflags = 0;
4787 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4788 			stp->sd_sigflags |= ssp->ss_events;
4789 		mutex_exit(&stp->sd_lock);
4790 		return (0);
4791 	}
4792 
4793 	case I_GETSIG:
4794 		/*
4795 		 * Return (in arg) the current registration of events
4796 		 * for which the calling proc is to be signaled.
4797 		 */
4798 	{
4799 		struct strsig *ssp;
4800 		struct pid  *pidp;
4801 
4802 		pidp = curproc->p_pidp;
4803 		mutex_enter(&stp->sd_lock);
4804 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4805 			if (ssp->ss_pidp == pidp) {
4806 				error = strcopyout(&ssp->ss_events, (void *)arg,
4807 				    sizeof (int), copyflag);
4808 				mutex_exit(&stp->sd_lock);
4809 				return (error);
4810 			}
4811 		mutex_exit(&stp->sd_lock);
4812 		return (EINVAL);
4813 	}
4814 
4815 	case I_ESETSIG:
4816 		/*
4817 		 * Register the ss_pid to receive the SIGPOLL
4818 		 * signal based on the events is ss_events arg.  If
4819 		 * ss_events is zero, remove the proc from register list.
4820 		 */
4821 	{
4822 		struct strsig *ssp, *pssp;
4823 		struct proc *proc;
4824 		struct pid  *pidp;
4825 		pid_t pid;
4826 		struct strsigset ss;
4827 
4828 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4829 		if (error)
4830 			return (error);
4831 
4832 		pid = ss.ss_pid;
4833 
4834 		if (ss.ss_events != 0) {
4835 			/*
4836 			 * Permissions check by sending signal 0.
4837 			 * Note that when kill fails it does a set_errno
4838 			 * causing the system call to fail.
4839 			 */
4840 			error = kill(pid, 0);
4841 			if (error) {
4842 				return (error);
4843 			}
4844 		}
4845 		mutex_enter(&pidlock);
4846 		if (pid == 0)
4847 			proc = curproc;
4848 		else if (pid < 0)
4849 			proc = pgfind(-pid);
4850 		else
4851 			proc = prfind(pid);
4852 		if (proc == NULL) {
4853 			mutex_exit(&pidlock);
4854 			return (ESRCH);
4855 		}
4856 		if (pid < 0)
4857 			pidp = proc->p_pgidp;
4858 		else
4859 			pidp = proc->p_pidp;
4860 		ASSERT(pidp);
4861 		/*
4862 		 * Get a hold on the pid structure while referencing it.
4863 		 * There is a separate PID_HOLD should it be inserted
4864 		 * in the list below.
4865 		 */
4866 		PID_HOLD(pidp);
4867 		mutex_exit(&pidlock);
4868 
4869 		pssp = NULL;
4870 		/*
4871 		 * Hold sd_lock to prevent traversal of sd_siglist while
4872 		 * it is modified.
4873 		 */
4874 		mutex_enter(&stp->sd_lock);
4875 		for (ssp = stp->sd_siglist; ssp && (ssp->ss_pid != pid);
4876 		    pssp = ssp, ssp = ssp->ss_next)
4877 			;
4878 
4879 		if (ss.ss_events) {
4880 			if (ss.ss_events &
4881 			    ~(S_INPUT|S_HIPRI|S_MSG|S_HANGUP|S_ERROR|
4882 			    S_RDNORM|S_WRNORM|S_RDBAND|S_WRBAND|S_BANDURG)) {
4883 				mutex_exit(&stp->sd_lock);
4884 				mutex_enter(&pidlock);
4885 				PID_RELE(pidp);
4886 				mutex_exit(&pidlock);
4887 				return (EINVAL);
4888 			}
4889 			if ((ss.ss_events & S_BANDURG) &&
4890 			    !(ss.ss_events & S_RDBAND)) {
4891 				mutex_exit(&stp->sd_lock);
4892 				mutex_enter(&pidlock);
4893 				PID_RELE(pidp);
4894 				mutex_exit(&pidlock);
4895 				return (EINVAL);
4896 			}
4897 
4898 			/*
4899 			 * If proc not already registered, add it
4900 			 * to list.
4901 			 */
4902 			if (!ssp) {
4903 				ssp = kmem_alloc(sizeof (strsig_t), KM_SLEEP);
4904 				ssp->ss_pidp = pidp;
4905 				ssp->ss_pid = pid;
4906 				ssp->ss_next = NULL;
4907 				if (pssp)
4908 					pssp->ss_next = ssp;
4909 				else
4910 					stp->sd_siglist = ssp;
4911 				mutex_enter(&pidlock);
4912 				PID_HOLD(pidp);
4913 				mutex_exit(&pidlock);
4914 			}
4915 
4916 			/*
4917 			 * Set events.
4918 			 */
4919 			ssp->ss_events = ss.ss_events;
4920 		} else {
4921 			/*
4922 			 * Remove proc from register list.
4923 			 */
4924 			if (ssp) {
4925 				mutex_enter(&pidlock);
4926 				PID_RELE(pidp);
4927 				mutex_exit(&pidlock);
4928 				if (pssp)
4929 					pssp->ss_next = ssp->ss_next;
4930 				else
4931 					stp->sd_siglist = ssp->ss_next;
4932 				kmem_free(ssp, sizeof (strsig_t));
4933 			} else {
4934 				mutex_exit(&stp->sd_lock);
4935 				mutex_enter(&pidlock);
4936 				PID_RELE(pidp);
4937 				mutex_exit(&pidlock);
4938 				return (EINVAL);
4939 			}
4940 		}
4941 
4942 		/*
4943 		 * Recalculate OR of sig events.
4944 		 */
4945 		stp->sd_sigflags = 0;
4946 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4947 			stp->sd_sigflags |= ssp->ss_events;
4948 		mutex_exit(&stp->sd_lock);
4949 		mutex_enter(&pidlock);
4950 		PID_RELE(pidp);
4951 		mutex_exit(&pidlock);
4952 		return (0);
4953 	}
4954 
4955 	case I_EGETSIG:
4956 		/*
4957 		 * Return (in arg) the current registration of events
4958 		 * for which the calling proc is to be signaled.
4959 		 */
4960 	{
4961 		struct strsig *ssp;
4962 		struct proc *proc;
4963 		pid_t pid;
4964 		struct pid  *pidp;
4965 		struct strsigset ss;
4966 
4967 		error = strcopyin((void *)arg, &ss, sizeof (ss), copyflag);
4968 		if (error)
4969 			return (error);
4970 
4971 		pid = ss.ss_pid;
4972 		mutex_enter(&pidlock);
4973 		if (pid == 0)
4974 			proc = curproc;
4975 		else if (pid < 0)
4976 			proc = pgfind(-pid);
4977 		else
4978 			proc = prfind(pid);
4979 		if (proc == NULL) {
4980 			mutex_exit(&pidlock);
4981 			return (ESRCH);
4982 		}
4983 		if (pid < 0)
4984 			pidp = proc->p_pgidp;
4985 		else
4986 			pidp = proc->p_pidp;
4987 
4988 		/* Prevent the pidp from being reassigned */
4989 		PID_HOLD(pidp);
4990 		mutex_exit(&pidlock);
4991 
4992 		mutex_enter(&stp->sd_lock);
4993 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
4994 			if (ssp->ss_pid == pid) {
4995 				ss.ss_pid = ssp->ss_pid;
4996 				ss.ss_events = ssp->ss_events;
4997 				error = strcopyout(&ss, (void *)arg,
4998 				    sizeof (struct strsigset), copyflag);
4999 				mutex_exit(&stp->sd_lock);
5000 				mutex_enter(&pidlock);
5001 				PID_RELE(pidp);
5002 				mutex_exit(&pidlock);
5003 				return (error);
5004 			}
5005 		mutex_exit(&stp->sd_lock);
5006 		mutex_enter(&pidlock);
5007 		PID_RELE(pidp);
5008 		mutex_exit(&pidlock);
5009 		return (EINVAL);
5010 	}
5011 
5012 	case I_PEEK:
5013 	{
5014 		STRUCT_DECL(strpeek, strpeek);
5015 		size_t n;
5016 		mblk_t *fmp, *tmp_mp = NULL;
5017 
5018 		STRUCT_INIT(strpeek, flag);
5019 
5020 		error = strcopyin((void *)arg, STRUCT_BUF(strpeek),
5021 		    STRUCT_SIZE(strpeek), copyflag);
5022 		if (error)
5023 			return (error);
5024 
5025 		mutex_enter(QLOCK(rdq));
5026 		/*
5027 		 * Skip the invalid messages
5028 		 */
5029 		for (mp = rdq->q_first; mp != NULL; mp = mp->b_next)
5030 			if (mp->b_datap->db_type != M_SIG)
5031 				break;
5032 
5033 		/*
5034 		 * If user has requested to peek at a high priority message
5035 		 * and first message is not, return 0
5036 		 */
5037 		if (mp != NULL) {
5038 			if ((STRUCT_FGET(strpeek, flags) & RS_HIPRI) &&
5039 			    queclass(mp) == QNORM) {
5040 				*rvalp = 0;
5041 				mutex_exit(QLOCK(rdq));
5042 				return (0);
5043 			}
5044 		} else if (stp->sd_struiordq == NULL ||
5045 		    (STRUCT_FGET(strpeek, flags) & RS_HIPRI)) {
5046 			/*
5047 			 * No mblks to look at at the streamhead and
5048 			 * 1). This isn't a synch stream or
5049 			 * 2). This is a synch stream but caller wants high
5050 			 *	priority messages which is not supported by
5051 			 *	the synch stream. (it only supports QNORM)
5052 			 */
5053 			*rvalp = 0;
5054 			mutex_exit(QLOCK(rdq));
5055 			return (0);
5056 		}
5057 
5058 		fmp = mp;
5059 
5060 		if (mp && mp->b_datap->db_type == M_PASSFP) {
5061 			mutex_exit(QLOCK(rdq));
5062 			return (EBADMSG);
5063 		}
5064 
5065 		ASSERT(mp == NULL || mp->b_datap->db_type == M_PCPROTO ||
5066 		    mp->b_datap->db_type == M_PROTO ||
5067 		    mp->b_datap->db_type == M_DATA);
5068 
5069 		if (mp && mp->b_datap->db_type == M_PCPROTO) {
5070 			STRUCT_FSET(strpeek, flags, RS_HIPRI);
5071 		} else {
5072 			STRUCT_FSET(strpeek, flags, 0);
5073 		}
5074 
5075 
5076 		if (mp && ((tmp_mp = dupmsg(mp)) == NULL)) {
5077 			mutex_exit(QLOCK(rdq));
5078 			return (ENOSR);
5079 		}
5080 		mutex_exit(QLOCK(rdq));
5081 
5082 		/*
5083 		 * set mp = tmp_mp, so that I_PEEK processing can continue.
5084 		 * tmp_mp is used to free the dup'd message.
5085 		 */
5086 		mp = tmp_mp;
5087 
5088 		uio.uio_fmode = 0;
5089 		uio.uio_extflg = UIO_COPY_CACHED;
5090 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5091 		    UIO_SYSSPACE;
5092 		uio.uio_limit = 0;
5093 		/*
5094 		 * First process PROTO blocks, if any.
5095 		 * If user doesn't want to get ctl info by setting maxlen <= 0,
5096 		 * then set len to -1/0 and skip control blocks part.
5097 		 */
5098 		if (STRUCT_FGET(strpeek, ctlbuf.maxlen) < 0)
5099 			STRUCT_FSET(strpeek, ctlbuf.len, -1);
5100 		else if (STRUCT_FGET(strpeek, ctlbuf.maxlen) == 0)
5101 			STRUCT_FSET(strpeek, ctlbuf.len, 0);
5102 		else {
5103 			int	ctl_part = 0;
5104 
5105 			iov.iov_base = STRUCT_FGETP(strpeek, ctlbuf.buf);
5106 			iov.iov_len = STRUCT_FGET(strpeek, ctlbuf.maxlen);
5107 			uio.uio_iov = &iov;
5108 			uio.uio_resid = iov.iov_len;
5109 			uio.uio_loffset = 0;
5110 			uio.uio_iovcnt = 1;
5111 			while (mp && mp->b_datap->db_type != M_DATA &&
5112 			    uio.uio_resid >= 0) {
5113 				ASSERT(STRUCT_FGET(strpeek, flags) == 0 ?
5114 				    mp->b_datap->db_type == M_PROTO :
5115 				    mp->b_datap->db_type == M_PCPROTO);
5116 
5117 				if ((n = MIN(uio.uio_resid,
5118 				    mp->b_wptr - mp->b_rptr)) != 0 &&
5119 				    (error = uiomove((char *)mp->b_rptr, n,
5120 				    UIO_READ, &uio)) != 0) {
5121 					freemsg(tmp_mp);
5122 					return (error);
5123 				}
5124 				ctl_part = 1;
5125 				mp = mp->b_cont;
5126 			}
5127 			/* No ctl message */
5128 			if (ctl_part == 0)
5129 				STRUCT_FSET(strpeek, ctlbuf.len, -1);
5130 			else
5131 				STRUCT_FSET(strpeek, ctlbuf.len,
5132 				    STRUCT_FGET(strpeek, ctlbuf.maxlen) -
5133 				    uio.uio_resid);
5134 		}
5135 
5136 		/*
5137 		 * Now process DATA blocks, if any.
5138 		 * If user doesn't want to get data info by setting maxlen <= 0,
5139 		 * then set len to -1/0 and skip data blocks part.
5140 		 */
5141 		if (STRUCT_FGET(strpeek, databuf.maxlen) < 0)
5142 			STRUCT_FSET(strpeek, databuf.len, -1);
5143 		else if (STRUCT_FGET(strpeek, databuf.maxlen) == 0)
5144 			STRUCT_FSET(strpeek, databuf.len, 0);
5145 		else {
5146 			int	data_part = 0;
5147 
5148 			iov.iov_base = STRUCT_FGETP(strpeek, databuf.buf);
5149 			iov.iov_len = STRUCT_FGET(strpeek, databuf.maxlen);
5150 			uio.uio_iov = &iov;
5151 			uio.uio_resid = iov.iov_len;
5152 			uio.uio_loffset = 0;
5153 			uio.uio_iovcnt = 1;
5154 			while (mp && uio.uio_resid) {
5155 				if (mp->b_datap->db_type == M_DATA) {
5156 					if ((n = MIN(uio.uio_resid,
5157 					    mp->b_wptr - mp->b_rptr)) != 0 &&
5158 					    (error = uiomove((char *)mp->b_rptr,
5159 					    n, UIO_READ, &uio)) != 0) {
5160 						freemsg(tmp_mp);
5161 						return (error);
5162 					}
5163 					data_part = 1;
5164 				}
5165 				ASSERT(data_part == 0 ||
5166 				    mp->b_datap->db_type == M_DATA);
5167 				mp = mp->b_cont;
5168 			}
5169 			/* No data message */
5170 			if (data_part == 0)
5171 				STRUCT_FSET(strpeek, databuf.len, -1);
5172 			else
5173 				STRUCT_FSET(strpeek, databuf.len,
5174 				    STRUCT_FGET(strpeek, databuf.maxlen) -
5175 				    uio.uio_resid);
5176 		}
5177 		freemsg(tmp_mp);
5178 
5179 		/*
5180 		 * It is a synch stream and user wants to get
5181 		 * data (maxlen > 0).
5182 		 * uio setup is done by the codes that process DATA
5183 		 * blocks above.
5184 		 */
5185 		if ((fmp == NULL) && STRUCT_FGET(strpeek, databuf.maxlen) > 0) {
5186 			infod_t infod;
5187 
5188 			infod.d_cmd = INFOD_COPYOUT;
5189 			infod.d_res = 0;
5190 			infod.d_uiop = &uio;
5191 			error = infonext(rdq, &infod);
5192 			if (error == EINVAL || error == EBUSY)
5193 				error = 0;
5194 			if (error)
5195 				return (error);
5196 			STRUCT_FSET(strpeek, databuf.len, STRUCT_FGET(strpeek,
5197 			    databuf.maxlen) - uio.uio_resid);
5198 			if (STRUCT_FGET(strpeek, databuf.len) == 0) {
5199 				/*
5200 				 * No data found by the infonext().
5201 				 */
5202 				STRUCT_FSET(strpeek, databuf.len, -1);
5203 			}
5204 		}
5205 		error = strcopyout(STRUCT_BUF(strpeek), (void *)arg,
5206 		    STRUCT_SIZE(strpeek), copyflag);
5207 		if (error) {
5208 			return (error);
5209 		}
5210 		/*
5211 		 * If there is no message retrieved, set return code to 0
5212 		 * otherwise, set it to 1.
5213 		 */
5214 		if (STRUCT_FGET(strpeek, ctlbuf.len) == -1 &&
5215 		    STRUCT_FGET(strpeek, databuf.len) == -1)
5216 			*rvalp = 0;
5217 		else
5218 			*rvalp = 1;
5219 		return (0);
5220 	}
5221 
5222 	case I_FDINSERT:
5223 	{
5224 		STRUCT_DECL(strfdinsert, strfdinsert);
5225 		struct file *resftp;
5226 		struct stdata *resstp;
5227 		t_uscalar_t	ival;
5228 		ssize_t msgsize;
5229 		struct strbuf mctl;
5230 
5231 		STRUCT_INIT(strfdinsert, flag);
5232 		if (stp->sd_flag & STRHUP)
5233 			return (ENXIO);
5234 		/*
5235 		 * STRDERR, STWRERR and STPLEX tested above.
5236 		 */
5237 		error = strcopyin((void *)arg, STRUCT_BUF(strfdinsert),
5238 		    STRUCT_SIZE(strfdinsert), copyflag);
5239 		if (error)
5240 			return (error);
5241 
5242 		if (STRUCT_FGET(strfdinsert, offset) < 0 ||
5243 		    (STRUCT_FGET(strfdinsert, offset) %
5244 		    sizeof (t_uscalar_t)) != 0)
5245 			return (EINVAL);
5246 		if ((resftp = getf(STRUCT_FGET(strfdinsert, fildes))) != NULL) {
5247 			if ((resstp = resftp->f_vnode->v_stream) == NULL) {
5248 				releasef(STRUCT_FGET(strfdinsert, fildes));
5249 				return (EINVAL);
5250 			}
5251 		} else
5252 			return (EINVAL);
5253 
5254 		mutex_enter(&resstp->sd_lock);
5255 		if (resstp->sd_flag & (STRDERR|STWRERR|STRHUP|STPLEX)) {
5256 			error = strgeterr(resstp,
5257 			    STRDERR|STWRERR|STRHUP|STPLEX, 0);
5258 			if (error != 0) {
5259 				mutex_exit(&resstp->sd_lock);
5260 				releasef(STRUCT_FGET(strfdinsert, fildes));
5261 				return (error);
5262 			}
5263 		}
5264 		mutex_exit(&resstp->sd_lock);
5265 
5266 #ifdef	_ILP32
5267 		{
5268 			queue_t	*q;
5269 			queue_t	*mate = NULL;
5270 
5271 			/* get read queue of stream terminus */
5272 			claimstr(resstp->sd_wrq);
5273 			for (q = resstp->sd_wrq->q_next; q->q_next != NULL;
5274 			    q = q->q_next)
5275 				if (!STRMATED(resstp) && STREAM(q) != resstp &&
5276 				    mate == NULL) {
5277 					ASSERT(q->q_qinfo->qi_srvp);
5278 					ASSERT(_OTHERQ(q)->q_qinfo->qi_srvp);
5279 					claimstr(q);
5280 					mate = q;
5281 				}
5282 			q = _RD(q);
5283 			if (mate)
5284 				releasestr(mate);
5285 			releasestr(resstp->sd_wrq);
5286 			ival = (t_uscalar_t)q;
5287 		}
5288 #else
5289 		ival = (t_uscalar_t)getminor(resftp->f_vnode->v_rdev);
5290 #endif	/* _ILP32 */
5291 
5292 		if (STRUCT_FGET(strfdinsert, ctlbuf.len) <
5293 		    STRUCT_FGET(strfdinsert, offset) + sizeof (t_uscalar_t)) {
5294 			releasef(STRUCT_FGET(strfdinsert, fildes));
5295 			return (EINVAL);
5296 		}
5297 
5298 		/*
5299 		 * Check for legal flag value.
5300 		 */
5301 		if (STRUCT_FGET(strfdinsert, flags) & ~RS_HIPRI) {
5302 			releasef(STRUCT_FGET(strfdinsert, fildes));
5303 			return (EINVAL);
5304 		}
5305 
5306 		/* get these values from those cached in the stream head */
5307 		mutex_enter(QLOCK(stp->sd_wrq));
5308 		rmin = stp->sd_qn_minpsz;
5309 		rmax = stp->sd_qn_maxpsz;
5310 		mutex_exit(QLOCK(stp->sd_wrq));
5311 
5312 		/*
5313 		 * Make sure ctl and data sizes together fall within
5314 		 * the limits of the max and min receive packet sizes
5315 		 * and do not exceed system limit.  A negative data
5316 		 * length means that no data part is to be sent.
5317 		 */
5318 		ASSERT((rmax >= 0) || (rmax == INFPSZ));
5319 		if (rmax == 0) {
5320 			releasef(STRUCT_FGET(strfdinsert, fildes));
5321 			return (ERANGE);
5322 		}
5323 		if ((msgsize = STRUCT_FGET(strfdinsert, databuf.len)) < 0)
5324 			msgsize = 0;
5325 		if ((msgsize < rmin) ||
5326 		    ((msgsize > rmax) && (rmax != INFPSZ)) ||
5327 		    (STRUCT_FGET(strfdinsert, ctlbuf.len) > strctlsz)) {
5328 			releasef(STRUCT_FGET(strfdinsert, fildes));
5329 			return (ERANGE);
5330 		}
5331 
5332 		mutex_enter(&stp->sd_lock);
5333 		while (!(STRUCT_FGET(strfdinsert, flags) & RS_HIPRI) &&
5334 		    !canputnext(stp->sd_wrq)) {
5335 			if ((error = strwaitq(stp, WRITEWAIT, (ssize_t)0,
5336 			    flag, -1, &done)) != 0 || done) {
5337 				mutex_exit(&stp->sd_lock);
5338 				releasef(STRUCT_FGET(strfdinsert, fildes));
5339 				return (error);
5340 			}
5341 			if ((error = i_straccess(stp, access)) != 0) {
5342 				mutex_exit(&stp->sd_lock);
5343 				releasef(
5344 				    STRUCT_FGET(strfdinsert, fildes));
5345 				return (error);
5346 			}
5347 		}
5348 		mutex_exit(&stp->sd_lock);
5349 
5350 		/*
5351 		 * Copy strfdinsert.ctlbuf into native form of
5352 		 * ctlbuf to pass down into strmakemsg().
5353 		 */
5354 		mctl.maxlen = STRUCT_FGET(strfdinsert, ctlbuf.maxlen);
5355 		mctl.len = STRUCT_FGET(strfdinsert, ctlbuf.len);
5356 		mctl.buf = STRUCT_FGETP(strfdinsert, ctlbuf.buf);
5357 
5358 		iov.iov_base = STRUCT_FGETP(strfdinsert, databuf.buf);
5359 		iov.iov_len = STRUCT_FGET(strfdinsert, databuf.len);
5360 		uio.uio_iov = &iov;
5361 		uio.uio_iovcnt = 1;
5362 		uio.uio_loffset = 0;
5363 		uio.uio_segflg = (copyflag == U_TO_K) ? UIO_USERSPACE :
5364 		    UIO_SYSSPACE;
5365 		uio.uio_fmode = 0;
5366 		uio.uio_extflg = UIO_COPY_CACHED;
5367 		uio.uio_resid = iov.iov_len;
5368 		if ((error = strmakemsg(&mctl,
5369 		    &msgsize, &uio, stp,
5370 		    STRUCT_FGET(strfdinsert, flags), &mp)) != 0 || !mp) {
5371 			STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5372 			releasef(STRUCT_FGET(strfdinsert, fildes));
5373 			return (error);
5374 		}
5375 
5376 		STRUCT_FSET(strfdinsert, databuf.len, msgsize);
5377 
5378 		/*
5379 		 * Place the possibly reencoded queue pointer 'offset' bytes
5380 		 * from the start of the control portion of the message.
5381 		 */
5382 		*((t_uscalar_t *)(mp->b_rptr +
5383 		    STRUCT_FGET(strfdinsert, offset))) = ival;
5384 
5385 		/*
5386 		 * Put message downstream.
5387 		 */
5388 		stream_willservice(stp);
5389 		putnext(stp->sd_wrq, mp);
5390 		stream_runservice(stp);
5391 		releasef(STRUCT_FGET(strfdinsert, fildes));
5392 		return (error);
5393 	}
5394 
5395 	case I_SENDFD:
5396 	{
5397 		struct file *fp;
5398 
5399 		if ((fp = getf((int)arg)) == NULL)
5400 			return (EBADF);
5401 		error = do_sendfp(stp, fp, crp);
5402 		if (audit_active) {
5403 			audit_fdsend((int)arg, fp, error);
5404 		}
5405 		releasef((int)arg);
5406 		return (error);
5407 	}
5408 
5409 	case I_RECVFD:
5410 	case I_E_RECVFD:
5411 	{
5412 		struct k_strrecvfd *srf;
5413 		int i, fd;
5414 
5415 		mutex_enter(&stp->sd_lock);
5416 		while (!(mp = getq(rdq))) {
5417 			if (stp->sd_flag & (STRHUP|STREOF)) {
5418 				mutex_exit(&stp->sd_lock);
5419 				return (ENXIO);
5420 			}
5421 			if ((error = strwaitq(stp, GETWAIT, (ssize_t)0,
5422 			    flag, -1, &done)) != 0 || done) {
5423 				mutex_exit(&stp->sd_lock);
5424 				return (error);
5425 			}
5426 			if ((error = i_straccess(stp, access)) != 0) {
5427 				mutex_exit(&stp->sd_lock);
5428 				return (error);
5429 			}
5430 		}
5431 		if (mp->b_datap->db_type != M_PASSFP) {
5432 			putback(stp, rdq, mp, mp->b_band);
5433 			mutex_exit(&stp->sd_lock);
5434 			return (EBADMSG);
5435 		}
5436 		mutex_exit(&stp->sd_lock);
5437 
5438 		srf = (struct k_strrecvfd *)mp->b_rptr;
5439 		if ((fd = ufalloc(0)) == -1) {
5440 			mutex_enter(&stp->sd_lock);
5441 			putback(stp, rdq, mp, mp->b_band);
5442 			mutex_exit(&stp->sd_lock);
5443 			return (EMFILE);
5444 		}
5445 		if (cmd == I_RECVFD) {
5446 			struct o_strrecvfd	ostrfd;
5447 
5448 			/* check to see if uid/gid values are too large. */
5449 
5450 			if (srf->uid > (o_uid_t)USHRT_MAX ||
5451 			    srf->gid > (o_gid_t)USHRT_MAX) {
5452 				mutex_enter(&stp->sd_lock);
5453 				putback(stp, rdq, mp, mp->b_band);
5454 				mutex_exit(&stp->sd_lock);
5455 				setf(fd, NULL);	/* release fd entry */
5456 				return (EOVERFLOW);
5457 			}
5458 
5459 			ostrfd.fd = fd;
5460 			ostrfd.uid = (o_uid_t)srf->uid;
5461 			ostrfd.gid = (o_gid_t)srf->gid;
5462 
5463 			/* Null the filler bits */
5464 			for (i = 0; i < 8; i++)
5465 				ostrfd.fill[i] = 0;
5466 
5467 			error = strcopyout(&ostrfd, (void *)arg,
5468 			    sizeof (struct o_strrecvfd), copyflag);
5469 		} else {		/* I_E_RECVFD */
5470 			struct strrecvfd	strfd;
5471 
5472 			strfd.fd = fd;
5473 			strfd.uid = srf->uid;
5474 			strfd.gid = srf->gid;
5475 
5476 			/* null the filler bits */
5477 			for (i = 0; i < 8; i++)
5478 				strfd.fill[i] = 0;
5479 
5480 			error = strcopyout(&strfd, (void *)arg,
5481 			    sizeof (struct strrecvfd), copyflag);
5482 		}
5483 
5484 		if (error) {
5485 			setf(fd, NULL);	/* release fd entry */
5486 			mutex_enter(&stp->sd_lock);
5487 			putback(stp, rdq, mp, mp->b_band);
5488 			mutex_exit(&stp->sd_lock);
5489 			return (error);
5490 		}
5491 		if (audit_active) {
5492 			audit_fdrecv(fd, srf->fp);
5493 		}
5494 
5495 		/*
5496 		 * Always increment f_count since the freemsg() below will
5497 		 * always call free_passfp() which performs a closef().
5498 		 */
5499 		mutex_enter(&srf->fp->f_tlock);
5500 		srf->fp->f_count++;
5501 		mutex_exit(&srf->fp->f_tlock);
5502 		setf(fd, srf->fp);
5503 		freemsg(mp);
5504 		return (0);
5505 	}
5506 
5507 	case I_SWROPT:
5508 		/*
5509 		 * Set/clear the write options. arg is a bit
5510 		 * mask with any of the following bits set...
5511 		 * 	SNDZERO - send zero length message
5512 		 *	SNDPIPE - send sigpipe to process if
5513 		 *		sd_werror is set and process is
5514 		 *		doing a write or putmsg.
5515 		 * The new stream head write options should reflect
5516 		 * what is in arg.
5517 		 */
5518 		if (arg & ~(SNDZERO|SNDPIPE))
5519 			return (EINVAL);
5520 
5521 		mutex_enter(&stp->sd_lock);
5522 		stp->sd_wput_opt &= ~(SW_SIGPIPE|SW_SNDZERO);
5523 		if (arg & SNDZERO)
5524 			stp->sd_wput_opt |= SW_SNDZERO;
5525 		if (arg & SNDPIPE)
5526 			stp->sd_wput_opt |= SW_SIGPIPE;
5527 		mutex_exit(&stp->sd_lock);
5528 		return (0);
5529 
5530 	case I_GWROPT:
5531 	{
5532 		int wropt = 0;
5533 
5534 		if (stp->sd_wput_opt & SW_SNDZERO)
5535 			wropt |= SNDZERO;
5536 		if (stp->sd_wput_opt & SW_SIGPIPE)
5537 			wropt |= SNDPIPE;
5538 		return (strcopyout(&wropt, (void *)arg, sizeof (wropt),
5539 		    copyflag));
5540 	}
5541 
5542 	case I_LIST:
5543 		/*
5544 		 * Returns all the modules found on this stream,
5545 		 * upto the driver. If argument is NULL, return the
5546 		 * number of modules (including driver). If argument
5547 		 * is not NULL, copy the names into the structure
5548 		 * provided.
5549 		 */
5550 
5551 	{
5552 		queue_t *q;
5553 		int num_modules, space_allocated;
5554 		STRUCT_DECL(str_list, strlist);
5555 		struct str_mlist *mlist_ptr;
5556 
5557 		if (arg == NULL) { /* Return number of modules plus driver */
5558 			q = stp->sd_wrq;
5559 			if (stp->sd_vnode->v_type == VFIFO) {
5560 				*rvalp = stp->sd_pushcnt;
5561 			} else {
5562 				*rvalp = stp->sd_pushcnt + 1;
5563 			}
5564 		} else {
5565 			STRUCT_INIT(strlist, flag);
5566 
5567 			error = strcopyin((void *)arg, STRUCT_BUF(strlist),
5568 			    STRUCT_SIZE(strlist), copyflag);
5569 			if (error)
5570 				return (error);
5571 
5572 			space_allocated = STRUCT_FGET(strlist, sl_nmods);
5573 			if ((space_allocated) <= 0)
5574 				return (EINVAL);
5575 			claimstr(stp->sd_wrq);
5576 			q = stp->sd_wrq;
5577 			num_modules = 0;
5578 			while (_SAMESTR(q) && (space_allocated != 0)) {
5579 				char *name =
5580 				    q->q_next->q_qinfo->qi_minfo->mi_idname;
5581 
5582 				mlist_ptr = STRUCT_FGETP(strlist, sl_modlist);
5583 
5584 				error = strcopyout(name, mlist_ptr,
5585 				    strlen(name) + 1, copyflag);
5586 
5587 				if (error) {
5588 					releasestr(stp->sd_wrq);
5589 					return (error);
5590 				}
5591 				q = q->q_next;
5592 				space_allocated--;
5593 				num_modules++;
5594 				mlist_ptr =
5595 				    (struct str_mlist *)((uintptr_t)mlist_ptr +
5596 				    sizeof (struct str_mlist));
5597 				STRUCT_FSETP(strlist, sl_modlist, mlist_ptr);
5598 			}
5599 			releasestr(stp->sd_wrq);
5600 			error = strcopyout(&num_modules, (void *)arg,
5601 			    sizeof (int), copyflag);
5602 		}
5603 		return (error);
5604 	}
5605 
5606 	case I_CKBAND:
5607 	{
5608 		queue_t *q;
5609 		qband_t *qbp;
5610 
5611 		if ((arg < 0) || (arg >= NBAND))
5612 			return (EINVAL);
5613 		q = _RD(stp->sd_wrq);
5614 		mutex_enter(QLOCK(q));
5615 		if (arg > (int)q->q_nband) {
5616 			*rvalp = 0;
5617 		} else {
5618 			if (arg == 0) {
5619 				if (q->q_first)
5620 					*rvalp = 1;
5621 				else
5622 					*rvalp = 0;
5623 			} else {
5624 				qbp = q->q_bandp;
5625 				while (--arg > 0)
5626 					qbp = qbp->qb_next;
5627 				if (qbp->qb_first)
5628 					*rvalp = 1;
5629 				else
5630 					*rvalp = 0;
5631 			}
5632 		}
5633 		mutex_exit(QLOCK(q));
5634 		return (0);
5635 	}
5636 
5637 	case I_GETBAND:
5638 	{
5639 		int intpri;
5640 		queue_t *q;
5641 
5642 		q = _RD(stp->sd_wrq);
5643 		mutex_enter(QLOCK(q));
5644 		mp = q->q_first;
5645 		if (!mp) {
5646 			mutex_exit(QLOCK(q));
5647 			return (ENODATA);
5648 		}
5649 		intpri = (int)mp->b_band;
5650 		error = strcopyout(&intpri, (void *)arg, sizeof (int),
5651 		    copyflag);
5652 		mutex_exit(QLOCK(q));
5653 		return (error);
5654 	}
5655 
5656 	case I_ATMARK:
5657 	{
5658 		queue_t *q;
5659 
5660 		if (arg & ~(ANYMARK|LASTMARK))
5661 			return (EINVAL);
5662 		q = _RD(stp->sd_wrq);
5663 		mutex_enter(&stp->sd_lock);
5664 		if ((stp->sd_flag & STRATMARK) && (arg == ANYMARK)) {
5665 			*rvalp = 1;
5666 		} else {
5667 			mutex_enter(QLOCK(q));
5668 			mp = q->q_first;
5669 
5670 			if (mp == NULL)
5671 				*rvalp = 0;
5672 			else if ((arg == ANYMARK) && (mp->b_flag & MSGMARK))
5673 				*rvalp = 1;
5674 			else if ((arg == LASTMARK) && (mp == stp->sd_mark))
5675 				*rvalp = 1;
5676 			else
5677 				*rvalp = 0;
5678 			mutex_exit(QLOCK(q));
5679 		}
5680 		mutex_exit(&stp->sd_lock);
5681 		return (0);
5682 	}
5683 
5684 	case I_CANPUT:
5685 	{
5686 		char band;
5687 
5688 		if ((arg < 0) || (arg >= NBAND))
5689 			return (EINVAL);
5690 		band = (char)arg;
5691 		*rvalp = bcanputnext(stp->sd_wrq, band);
5692 		return (0);
5693 	}
5694 
5695 	case I_SETCLTIME:
5696 	{
5697 		int closetime;
5698 
5699 		error = strcopyin((void *)arg, &closetime, sizeof (int),
5700 		    copyflag);
5701 		if (error)
5702 			return (error);
5703 		if (closetime < 0)
5704 			return (EINVAL);
5705 
5706 		stp->sd_closetime = closetime;
5707 		return (0);
5708 	}
5709 
5710 	case I_GETCLTIME:
5711 	{
5712 		int closetime;
5713 
5714 		closetime = stp->sd_closetime;
5715 		return (strcopyout(&closetime, (void *)arg, sizeof (int),
5716 		    copyflag));
5717 	}
5718 
5719 	case TIOCGSID:
5720 	{
5721 		pid_t sid;
5722 
5723 		mutex_enter(&stp->sd_lock);
5724 		if (stp->sd_sidp == NULL) {
5725 			mutex_exit(&stp->sd_lock);
5726 			return (ENOTTY);
5727 		}
5728 		sid = stp->sd_sidp->pid_id;
5729 		mutex_exit(&stp->sd_lock);
5730 		return (strcopyout(&sid, (void *)arg, sizeof (pid_t),
5731 		    copyflag));
5732 	}
5733 
5734 	case TIOCSPGRP:
5735 	{
5736 		pid_t pgrp;
5737 		proc_t *q;
5738 		pid_t	sid, fg_pgid, bg_pgid;
5739 
5740 		if (error = strcopyin((void *)arg, &pgrp, sizeof (pid_t),
5741 		    copyflag))
5742 			return (error);
5743 		mutex_enter(&stp->sd_lock);
5744 		mutex_enter(&pidlock);
5745 		if (stp->sd_sidp != ttoproc(curthread)->p_sessp->s_sidp) {
5746 			mutex_exit(&pidlock);
5747 			mutex_exit(&stp->sd_lock);
5748 			return (ENOTTY);
5749 		}
5750 		if (pgrp == stp->sd_pgidp->pid_id) {
5751 			mutex_exit(&pidlock);
5752 			mutex_exit(&stp->sd_lock);
5753 			return (0);
5754 		}
5755 		if (pgrp <= 0 || pgrp >= maxpid) {
5756 			mutex_exit(&pidlock);
5757 			mutex_exit(&stp->sd_lock);
5758 			return (EINVAL);
5759 		}
5760 		if ((q = pgfind(pgrp)) == NULL ||
5761 		    q->p_sessp != ttoproc(curthread)->p_sessp) {
5762 			mutex_exit(&pidlock);
5763 			mutex_exit(&stp->sd_lock);
5764 			return (EPERM);
5765 		}
5766 		sid = stp->sd_sidp->pid_id;
5767 		fg_pgid = q->p_pgrp;
5768 		bg_pgid = stp->sd_pgidp->pid_id;
5769 		CL_SET_PROCESS_GROUP(curthread, sid, bg_pgid, fg_pgid);
5770 		PID_RELE(stp->sd_pgidp);
5771 		ctty_clear_sighuped();
5772 		stp->sd_pgidp = q->p_pgidp;
5773 		PID_HOLD(stp->sd_pgidp);
5774 		mutex_exit(&pidlock);
5775 		mutex_exit(&stp->sd_lock);
5776 		return (0);
5777 	}
5778 
5779 	case TIOCGPGRP:
5780 	{
5781 		pid_t pgrp;
5782 
5783 		mutex_enter(&stp->sd_lock);
5784 		if (stp->sd_sidp == NULL) {
5785 			mutex_exit(&stp->sd_lock);
5786 			return (ENOTTY);
5787 		}
5788 		pgrp = stp->sd_pgidp->pid_id;
5789 		mutex_exit(&stp->sd_lock);
5790 		return (strcopyout(&pgrp, (void *)arg, sizeof (pid_t),
5791 		    copyflag));
5792 	}
5793 
5794 	case TIOCSCTTY:
5795 	{
5796 		return (strctty(stp));
5797 	}
5798 
5799 	case TIOCNOTTY:
5800 	{
5801 		/* freectty() always assumes curproc. */
5802 		if (freectty(B_FALSE) != 0)
5803 			return (0);
5804 		return (ENOTTY);
5805 	}
5806 
5807 	case FIONBIO:
5808 	case FIOASYNC:
5809 		return (0);	/* handled by the upper layer */
5810 	}
5811 }
5812 
5813 /*
5814  * Custom free routine used for M_PASSFP messages.
5815  */
5816 static void
5817 free_passfp(struct k_strrecvfd *srf)
5818 {
5819 	(void) closef(srf->fp);
5820 	kmem_free(srf, sizeof (struct k_strrecvfd) + sizeof (frtn_t));
5821 }
5822 
5823 /* ARGSUSED */
5824 int
5825 do_sendfp(struct stdata *stp, struct file *fp, struct cred *cr)
5826 {
5827 	queue_t *qp, *nextqp;
5828 	struct k_strrecvfd *srf;
5829 	mblk_t *mp;
5830 	frtn_t *frtnp;
5831 	size_t bufsize;
5832 	queue_t	*mate = NULL;
5833 	syncq_t	*sq = NULL;
5834 	int retval = 0;
5835 
5836 	if (stp->sd_flag & STRHUP)
5837 		return (ENXIO);
5838 
5839 	claimstr(stp->sd_wrq);
5840 
5841 	/* Fastpath, we have a pipe, and we are already mated, use it. */
5842 	if (STRMATED(stp)) {
5843 		qp = _RD(stp->sd_mate->sd_wrq);
5844 		claimstr(qp);
5845 		mate = qp;
5846 	} else { /* Not already mated. */
5847 
5848 		/*
5849 		 * Walk the stream to the end of this one.
5850 		 * assumes that the claimstr() will prevent
5851 		 * plumbing between the stream head and the
5852 		 * driver from changing
5853 		 */
5854 		qp = stp->sd_wrq;
5855 
5856 		/*
5857 		 * Loop until we reach the end of this stream.
5858 		 * On completion, qp points to the write queue
5859 		 * at the end of the stream, or the read queue
5860 		 * at the stream head if this is a fifo.
5861 		 */
5862 		while (((qp = qp->q_next) != NULL) && _SAMESTR(qp))
5863 			;
5864 
5865 		/*
5866 		 * Just in case we get a q_next which is NULL, but
5867 		 * not at the end of the stream.  This is actually
5868 		 * broken, so we set an assert to catch it in
5869 		 * debug, and set an error and return if not debug.
5870 		 */
5871 		ASSERT(qp);
5872 		if (qp == NULL) {
5873 			releasestr(stp->sd_wrq);
5874 			return (EINVAL);
5875 		}
5876 
5877 		/*
5878 		 * Enter the syncq for the driver, so (hopefully)
5879 		 * the queue values will not change on us.
5880 		 * XXXX - This will only prevent the race IFF only
5881 		 *   the write side modifies the q_next member, and
5882 		 *   the put procedure is protected by at least
5883 		 *   MT_PERQ.
5884 		 */
5885 		if ((sq = qp->q_syncq) != NULL)
5886 			entersq(sq, SQ_PUT);
5887 
5888 		/* Now get the q_next value from this qp. */
5889 		nextqp = qp->q_next;
5890 
5891 		/*
5892 		 * If nextqp exists and the other stream is different
5893 		 * from this one claim the stream, set the mate, and
5894 		 * get the read queue at the stream head of the other
5895 		 * stream.  Assumes that nextqp was at least valid when
5896 		 * we got it.  Hopefully the entersq of the driver
5897 		 * will prevent it from changing on us.
5898 		 */
5899 		if ((nextqp != NULL) && (STREAM(nextqp) != stp)) {
5900 			ASSERT(qp->q_qinfo->qi_srvp);
5901 			ASSERT(_OTHERQ(qp)->q_qinfo->qi_srvp);
5902 			ASSERT(_OTHERQ(qp->q_next)->q_qinfo->qi_srvp);
5903 			claimstr(nextqp);
5904 
5905 			/* Make sure we still have a q_next */
5906 			if (nextqp != qp->q_next) {
5907 				releasestr(stp->sd_wrq);
5908 				releasestr(nextqp);
5909 				return (EINVAL);
5910 			}
5911 
5912 			qp = _RD(STREAM(nextqp)->sd_wrq);
5913 			mate = qp;
5914 		}
5915 		/* If we entered the synq above, leave it. */
5916 		if (sq != NULL)
5917 			leavesq(sq, SQ_PUT);
5918 	} /*  STRMATED(STP)  */
5919 
5920 	/* XXX prevents substitution of the ops vector */
5921 	if (qp->q_qinfo != &strdata && qp->q_qinfo != &fifo_strdata) {
5922 		retval = EINVAL;
5923 		goto out;
5924 	}
5925 
5926 	if (qp->q_flag & QFULL) {
5927 		retval = EAGAIN;
5928 		goto out;
5929 	}
5930 
5931 	/*
5932 	 * Since M_PASSFP messages include a file descriptor, we use
5933 	 * esballoc() and specify a custom free routine (free_passfp()) that
5934 	 * will close the descriptor as part of freeing the message.  For
5935 	 * convenience, we stash the frtn_t right after the data block.
5936 	 */
5937 	bufsize = sizeof (struct k_strrecvfd) + sizeof (frtn_t);
5938 	srf = kmem_alloc(bufsize, KM_NOSLEEP);
5939 	if (srf == NULL) {
5940 		retval = EAGAIN;
5941 		goto out;
5942 	}
5943 
5944 	frtnp = (frtn_t *)(srf + 1);
5945 	frtnp->free_arg = (caddr_t)srf;
5946 	frtnp->free_func = free_passfp;
5947 
5948 	mp = esballoc((uchar_t *)srf, bufsize, BPRI_MED, frtnp);
5949 	if (mp == NULL) {
5950 		kmem_free(srf, bufsize);
5951 		retval = EAGAIN;
5952 		goto out;
5953 	}
5954 	mp->b_wptr += sizeof (struct k_strrecvfd);
5955 	mp->b_datap->db_type = M_PASSFP;
5956 
5957 	srf->fp = fp;
5958 	srf->uid = crgetuid(curthread->t_cred);
5959 	srf->gid = crgetgid(curthread->t_cred);
5960 	mutex_enter(&fp->f_tlock);
5961 	fp->f_count++;
5962 	mutex_exit(&fp->f_tlock);
5963 
5964 	put(qp, mp);
5965 out:
5966 	releasestr(stp->sd_wrq);
5967 	if (mate)
5968 		releasestr(mate);
5969 	return (retval);
5970 }
5971 
5972 /*
5973  * Send an ioctl message downstream and wait for acknowledgement.
5974  * flags may be set to either U_TO_K or K_TO_K and a combination
5975  * of STR_NOERROR or STR_NOSIG
5976  * STR_NOSIG: Signals are essentially ignored or held and have
5977  *	no effect for the duration of the call.
5978  * STR_NOERROR: Ignores stream head read, write and hup errors.
5979  *	Additionally, if an existing ioctl times out, it is assumed
5980  *	lost and and this ioctl will continue as if the previous ioctl had
5981  *	finished.  ETIME may be returned if this ioctl times out (i.e.
5982  *	ic_timout is not INFTIM).  Non-stream head errors may be returned if
5983  *	the ioc_error indicates that the driver/module had problems,
5984  *	an EFAULT was found when accessing user data, a lack of
5985  * 	resources, etc.
5986  */
5987 int
5988 strdoioctl(
5989 	struct stdata *stp,
5990 	struct strioctl *strioc,
5991 	int fflags,		/* file flags with model info */
5992 	int flag,
5993 	cred_t *crp,
5994 	int *rvalp)
5995 {
5996 	mblk_t *bp;
5997 	struct iocblk *iocbp;
5998 	struct copyreq *reqp;
5999 	struct copyresp *resp;
6000 	int id;
6001 	int transparent = 0;
6002 	int error = 0;
6003 	int len = 0;
6004 	caddr_t taddr;
6005 	int copyflag = (flag & (U_TO_K | K_TO_K));
6006 	int sigflag = (flag & STR_NOSIG);
6007 	int errs;
6008 	uint_t waitflags;
6009 
6010 	ASSERT(copyflag == U_TO_K || copyflag == K_TO_K);
6011 	ASSERT((fflags & FMODELS) != 0);
6012 
6013 	TRACE_2(TR_FAC_STREAMS_FR,
6014 	    TR_STRDOIOCTL,
6015 	    "strdoioctl:stp %p strioc %p", stp, strioc);
6016 	if (strioc->ic_len == TRANSPARENT) {	/* send arg in M_DATA block */
6017 		transparent = 1;
6018 		strioc->ic_len = sizeof (intptr_t);
6019 	}
6020 
6021 	if (strioc->ic_len < 0 || (strmsgsz > 0 && strioc->ic_len > strmsgsz))
6022 		return (EINVAL);
6023 
6024 	if ((bp = allocb_cred_wait(sizeof (union ioctypes), sigflag, &error,
6025 	    crp)) == NULL)
6026 			return (error);
6027 
6028 	bzero(bp->b_wptr, sizeof (union ioctypes));
6029 
6030 	iocbp = (struct iocblk *)bp->b_wptr;
6031 	iocbp->ioc_count = strioc->ic_len;
6032 	iocbp->ioc_cmd = strioc->ic_cmd;
6033 	iocbp->ioc_flag = (fflags & FMODELS);
6034 
6035 	crhold(crp);
6036 	iocbp->ioc_cr = crp;
6037 	DB_TYPE(bp) = M_IOCTL;
6038 	DB_CPID(bp) = curproc->p_pid;
6039 	bp->b_wptr += sizeof (struct iocblk);
6040 
6041 	if (flag & STR_NOERROR)
6042 		errs = STPLEX;
6043 	else
6044 		errs = STRHUP|STRDERR|STWRERR|STPLEX;
6045 
6046 	/*
6047 	 * If there is data to copy into ioctl block, do so.
6048 	 */
6049 	if (iocbp->ioc_count > 0) {
6050 		if (transparent)
6051 			/*
6052 			 * Note: STR_NOERROR does not have an effect
6053 			 * in putiocd()
6054 			 */
6055 			id = K_TO_K | sigflag;
6056 		else
6057 			id = flag;
6058 		if ((error = putiocd(bp, strioc->ic_dp, id, crp)) != 0) {
6059 			freemsg(bp);
6060 			crfree(crp);
6061 			return (error);
6062 		}
6063 
6064 		/*
6065 		 * We could have slept copying in user pages.
6066 		 * Recheck the stream head state (the other end
6067 		 * of a pipe could have gone away).
6068 		 */
6069 		if (stp->sd_flag & errs) {
6070 			mutex_enter(&stp->sd_lock);
6071 			error = strgeterr(stp, errs, 0);
6072 			mutex_exit(&stp->sd_lock);
6073 			if (error != 0) {
6074 				freemsg(bp);
6075 				crfree(crp);
6076 				return (error);
6077 			}
6078 		}
6079 	}
6080 	if (transparent)
6081 		iocbp->ioc_count = TRANSPARENT;
6082 
6083 	/*
6084 	 * Block for up to STRTIMOUT milliseconds if there is an outstanding
6085 	 * ioctl for this stream already running.  All processes
6086 	 * sleeping here will be awakened as a result of an ACK
6087 	 * or NAK being received for the outstanding ioctl, or
6088 	 * as a result of the timer expiring on the outstanding
6089 	 * ioctl (a failure), or as a result of any waiting
6090 	 * process's timer expiring (also a failure).
6091 	 */
6092 
6093 	error = 0;
6094 	mutex_enter(&stp->sd_lock);
6095 	while (stp->sd_flag & (IOCWAIT | IOCWAITNE)) {
6096 		clock_t cv_rval;
6097 
6098 		TRACE_0(TR_FAC_STREAMS_FR,
6099 		    TR_STRDOIOCTL_WAIT,
6100 		    "strdoioctl sleeps - IOCWAIT");
6101 		cv_rval = str_cv_wait(&stp->sd_iocmonitor, &stp->sd_lock,
6102 		    STRTIMOUT, sigflag);
6103 		if (cv_rval <= 0) {
6104 			if (cv_rval == 0) {
6105 				error = EINTR;
6106 			} else {
6107 				if (flag & STR_NOERROR) {
6108 					/*
6109 					 * Terminating current ioctl in
6110 					 * progress -- assume it got lost and
6111 					 * wake up the other thread so that the
6112 					 * operation completes.
6113 					 */
6114 					if (!(stp->sd_flag & IOCWAITNE)) {
6115 						stp->sd_flag |= IOCWAITNE;
6116 						cv_broadcast(&stp->sd_monitor);
6117 					}
6118 					/*
6119 					 * Otherwise, there's a running
6120 					 * STR_NOERROR -- we have no choice
6121 					 * here but to wait forever (or until
6122 					 * interrupted).
6123 					 */
6124 				} else {
6125 					/*
6126 					 * pending ioctl has caused
6127 					 * us to time out
6128 					 */
6129 					error = ETIME;
6130 				}
6131 			}
6132 		} else if ((stp->sd_flag & errs)) {
6133 			error = strgeterr(stp, errs, 0);
6134 		}
6135 		if (error) {
6136 			mutex_exit(&stp->sd_lock);
6137 			freemsg(bp);
6138 			crfree(crp);
6139 			return (error);
6140 		}
6141 	}
6142 
6143 	/*
6144 	 * Have control of ioctl mechanism.
6145 	 * Send down ioctl packet and wait for response.
6146 	 */
6147 	if (stp->sd_iocblk != (mblk_t *)-1) {
6148 		freemsg(stp->sd_iocblk);
6149 	}
6150 	stp->sd_iocblk = NULL;
6151 
6152 	/*
6153 	 * If this is marked with 'noerror' (internal; mostly
6154 	 * I_{P,}{UN,}LINK), then make sure nobody else is able to get
6155 	 * in here by setting IOCWAITNE.
6156 	 */
6157 	waitflags = IOCWAIT;
6158 	if (flag & STR_NOERROR)
6159 		waitflags |= IOCWAITNE;
6160 
6161 	stp->sd_flag |= waitflags;
6162 
6163 	/*
6164 	 * Assign sequence number.
6165 	 */
6166 	iocbp->ioc_id = stp->sd_iocid = getiocseqno();
6167 
6168 	mutex_exit(&stp->sd_lock);
6169 
6170 	TRACE_1(TR_FAC_STREAMS_FR,
6171 	    TR_STRDOIOCTL_PUT, "strdoioctl put: stp %p", stp);
6172 	stream_willservice(stp);
6173 	putnext(stp->sd_wrq, bp);
6174 	stream_runservice(stp);
6175 
6176 	/*
6177 	 * Timed wait for acknowledgment.  The wait time is limited by the
6178 	 * timeout value, which must be a positive integer (number of
6179 	 * milliseconds) to wait, or 0 (use default value of STRTIMOUT
6180 	 * milliseconds), or -1 (wait forever).  This will be awakened
6181 	 * either by an ACK/NAK message arriving, the timer expiring, or
6182 	 * the timer expiring on another ioctl waiting for control of the
6183 	 * mechanism.
6184 	 */
6185 waitioc:
6186 	mutex_enter(&stp->sd_lock);
6187 
6188 
6189 	/*
6190 	 * If the reply has already arrived, don't sleep.  If awakened from
6191 	 * the sleep, fail only if the reply has not arrived by then.
6192 	 * Otherwise, process the reply.
6193 	 */
6194 	while (!stp->sd_iocblk) {
6195 		clock_t cv_rval;
6196 
6197 		if (stp->sd_flag & errs) {
6198 			error = strgeterr(stp, errs, 0);
6199 			if (error != 0) {
6200 				stp->sd_flag &= ~waitflags;
6201 				cv_broadcast(&stp->sd_iocmonitor);
6202 				mutex_exit(&stp->sd_lock);
6203 				crfree(crp);
6204 				return (error);
6205 			}
6206 		}
6207 
6208 		TRACE_0(TR_FAC_STREAMS_FR,
6209 		    TR_STRDOIOCTL_WAIT2,
6210 		    "strdoioctl sleeps awaiting reply");
6211 		ASSERT(error == 0);
6212 
6213 		cv_rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock,
6214 		    (strioc->ic_timout ?
6215 		    strioc->ic_timout * 1000 : STRTIMOUT), sigflag);
6216 
6217 		/*
6218 		 * There are four possible cases here: interrupt, timeout,
6219 		 * wakeup by IOCWAITNE (above), or wakeup by strrput_nondata (a
6220 		 * valid M_IOCTL reply).
6221 		 *
6222 		 * If we've been awakened by a STR_NOERROR ioctl on some other
6223 		 * thread, then sd_iocblk will still be NULL, and IOCWAITNE
6224 		 * will be set.  Pretend as if we just timed out.  Note that
6225 		 * this other thread waited at least STRTIMOUT before trying to
6226 		 * awaken our thread, so this is indistinguishable (even for
6227 		 * INFTIM) from the case where we failed with ETIME waiting on
6228 		 * IOCWAIT in the prior loop.
6229 		 */
6230 		if (cv_rval > 0 && !(flag & STR_NOERROR) &&
6231 		    stp->sd_iocblk == NULL && (stp->sd_flag & IOCWAITNE)) {
6232 			cv_rval = -1;
6233 		}
6234 
6235 		/*
6236 		 * note: STR_NOERROR does not protect
6237 		 * us here.. use ic_timout < 0
6238 		 */
6239 		if (cv_rval <= 0) {
6240 			if (cv_rval == 0) {
6241 				error = EINTR;
6242 			} else {
6243 				error =  ETIME;
6244 			}
6245 			/*
6246 			 * A message could have come in after we were scheduled
6247 			 * but before we were actually run.
6248 			 */
6249 			bp = stp->sd_iocblk;
6250 			stp->sd_iocblk = NULL;
6251 			if (bp != NULL) {
6252 				if ((bp->b_datap->db_type == M_COPYIN) ||
6253 				    (bp->b_datap->db_type == M_COPYOUT)) {
6254 					mutex_exit(&stp->sd_lock);
6255 					if (bp->b_cont) {
6256 						freemsg(bp->b_cont);
6257 						bp->b_cont = NULL;
6258 					}
6259 					bp->b_datap->db_type = M_IOCDATA;
6260 					bp->b_wptr = bp->b_rptr +
6261 					    sizeof (struct copyresp);
6262 					resp = (struct copyresp *)bp->b_rptr;
6263 					resp->cp_rval =
6264 					    (caddr_t)1; /* failure */
6265 					stream_willservice(stp);
6266 					putnext(stp->sd_wrq, bp);
6267 					stream_runservice(stp);
6268 					mutex_enter(&stp->sd_lock);
6269 				} else {
6270 					freemsg(bp);
6271 				}
6272 			}
6273 			stp->sd_flag &= ~waitflags;
6274 			cv_broadcast(&stp->sd_iocmonitor);
6275 			mutex_exit(&stp->sd_lock);
6276 			crfree(crp);
6277 			return (error);
6278 		}
6279 	}
6280 	bp = stp->sd_iocblk;
6281 	/*
6282 	 * Note: it is strictly impossible to get here with sd_iocblk set to
6283 	 * -1.  This is because the initial loop above doesn't allow any new
6284 	 * ioctls into the fray until all others have passed this point.
6285 	 */
6286 	ASSERT(bp != NULL && bp != (mblk_t *)-1);
6287 	TRACE_1(TR_FAC_STREAMS_FR,
6288 	    TR_STRDOIOCTL_ACK, "strdoioctl got reply: bp %p", bp);
6289 	if ((bp->b_datap->db_type == M_IOCACK) ||
6290 	    (bp->b_datap->db_type == M_IOCNAK)) {
6291 		/* for detection of duplicate ioctl replies */
6292 		stp->sd_iocblk = (mblk_t *)-1;
6293 		stp->sd_flag &= ~waitflags;
6294 		cv_broadcast(&stp->sd_iocmonitor);
6295 		mutex_exit(&stp->sd_lock);
6296 	} else {
6297 		/*
6298 		 * flags not cleared here because we're still doing
6299 		 * copy in/out for ioctl.
6300 		 */
6301 		stp->sd_iocblk = NULL;
6302 		mutex_exit(&stp->sd_lock);
6303 	}
6304 
6305 
6306 	/*
6307 	 * Have received acknowledgment.
6308 	 */
6309 
6310 	switch (bp->b_datap->db_type) {
6311 	case M_IOCACK:
6312 		/*
6313 		 * Positive ack.
6314 		 */
6315 		iocbp = (struct iocblk *)bp->b_rptr;
6316 
6317 		/*
6318 		 * Set error if indicated.
6319 		 */
6320 		if (iocbp->ioc_error) {
6321 			error = iocbp->ioc_error;
6322 			break;
6323 		}
6324 
6325 		/*
6326 		 * Set return value.
6327 		 */
6328 		*rvalp = iocbp->ioc_rval;
6329 
6330 		/*
6331 		 * Data may have been returned in ACK message (ioc_count > 0).
6332 		 * If so, copy it out to the user's buffer.
6333 		 */
6334 		if (iocbp->ioc_count && !transparent) {
6335 			if (error = getiocd(bp, strioc->ic_dp, copyflag))
6336 				break;
6337 		}
6338 		if (!transparent) {
6339 			if (len)	/* an M_COPYOUT was used with I_STR */
6340 				strioc->ic_len = len;
6341 			else
6342 				strioc->ic_len = (int)iocbp->ioc_count;
6343 		}
6344 		break;
6345 
6346 	case M_IOCNAK:
6347 		/*
6348 		 * Negative ack.
6349 		 *
6350 		 * The only thing to do is set error as specified
6351 		 * in neg ack packet.
6352 		 */
6353 		iocbp = (struct iocblk *)bp->b_rptr;
6354 
6355 		error = (iocbp->ioc_error ? iocbp->ioc_error : EINVAL);
6356 		break;
6357 
6358 	case M_COPYIN:
6359 		/*
6360 		 * Driver or module has requested user ioctl data.
6361 		 */
6362 		reqp = (struct copyreq *)bp->b_rptr;
6363 
6364 		/*
6365 		 * M_COPYIN should *never* have a message attached, though
6366 		 * it's harmless if it does -- thus, panic on a DEBUG
6367 		 * kernel and just free it on a non-DEBUG build.
6368 		 */
6369 		ASSERT(bp->b_cont == NULL);
6370 		if (bp->b_cont != NULL) {
6371 			freemsg(bp->b_cont);
6372 			bp->b_cont = NULL;
6373 		}
6374 
6375 		error = putiocd(bp, reqp->cq_addr, flag, crp);
6376 		if (error && bp->b_cont) {
6377 			freemsg(bp->b_cont);
6378 			bp->b_cont = NULL;
6379 		}
6380 
6381 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6382 		bp->b_datap->db_type = M_IOCDATA;
6383 
6384 		mblk_setcred(bp, crp);
6385 		DB_CPID(bp) = curproc->p_pid;
6386 		resp = (struct copyresp *)bp->b_rptr;
6387 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6388 		resp->cp_flag = (fflags & FMODELS);
6389 
6390 		stream_willservice(stp);
6391 		putnext(stp->sd_wrq, bp);
6392 		stream_runservice(stp);
6393 
6394 		if (error) {
6395 			mutex_enter(&stp->sd_lock);
6396 			stp->sd_flag &= ~waitflags;
6397 			cv_broadcast(&stp->sd_iocmonitor);
6398 			mutex_exit(&stp->sd_lock);
6399 			crfree(crp);
6400 			return (error);
6401 		}
6402 
6403 		goto waitioc;
6404 
6405 	case M_COPYOUT:
6406 		/*
6407 		 * Driver or module has ioctl data for a user.
6408 		 */
6409 		reqp = (struct copyreq *)bp->b_rptr;
6410 		ASSERT(bp->b_cont != NULL);
6411 
6412 		/*
6413 		 * Always (transparent or non-transparent )
6414 		 * use the address specified in the request
6415 		 */
6416 		taddr = reqp->cq_addr;
6417 		if (!transparent)
6418 			len = (int)reqp->cq_size;
6419 
6420 		/* copyout data to the provided address */
6421 		error = getiocd(bp, taddr, copyflag);
6422 
6423 		freemsg(bp->b_cont);
6424 		bp->b_cont = NULL;
6425 
6426 		bp->b_wptr = bp->b_rptr + sizeof (struct copyresp);
6427 		bp->b_datap->db_type = M_IOCDATA;
6428 
6429 		mblk_setcred(bp, crp);
6430 		DB_CPID(bp) = curproc->p_pid;
6431 		resp = (struct copyresp *)bp->b_rptr;
6432 		resp->cp_rval = (caddr_t)(uintptr_t)error;
6433 		resp->cp_flag = (fflags & FMODELS);
6434 
6435 		stream_willservice(stp);
6436 		putnext(stp->sd_wrq, bp);
6437 		stream_runservice(stp);
6438 
6439 		if (error) {
6440 			mutex_enter(&stp->sd_lock);
6441 			stp->sd_flag &= ~waitflags;
6442 			cv_broadcast(&stp->sd_iocmonitor);
6443 			mutex_exit(&stp->sd_lock);
6444 			crfree(crp);
6445 			return (error);
6446 		}
6447 		goto waitioc;
6448 
6449 	default:
6450 		ASSERT(0);
6451 		mutex_enter(&stp->sd_lock);
6452 		stp->sd_flag &= ~waitflags;
6453 		cv_broadcast(&stp->sd_iocmonitor);
6454 		mutex_exit(&stp->sd_lock);
6455 		break;
6456 	}
6457 
6458 	freemsg(bp);
6459 	crfree(crp);
6460 	return (error);
6461 }
6462 
6463 /*
6464  * Send an M_CMD message downstream and wait for a reply.  This is a ptools
6465  * special used to retrieve information from modules/drivers a stream without
6466  * being subjected to flow control or interfering with pending messages on the
6467  * stream (e.g. an ioctl in flight).
6468  */
6469 int
6470 strdocmd(struct stdata *stp, struct strcmd *scp, cred_t *crp)
6471 {
6472 	mblk_t *mp;
6473 	struct cmdblk *cmdp;
6474 	int error = 0;
6475 	int errs = STRHUP|STRDERR|STWRERR|STPLEX;
6476 	clock_t rval, timeout = STRTIMOUT;
6477 
6478 	if (scp->sc_len < 0 || scp->sc_len > sizeof (scp->sc_buf) ||
6479 	    scp->sc_timeout < -1)
6480 		return (EINVAL);
6481 
6482 	if (scp->sc_timeout > 0)
6483 		timeout = scp->sc_timeout * MILLISEC;
6484 
6485 	if ((mp = allocb_cred(sizeof (struct cmdblk), crp)) == NULL)
6486 		return (ENOMEM);
6487 
6488 	crhold(crp);
6489 
6490 	cmdp = (struct cmdblk *)mp->b_wptr;
6491 	cmdp->cb_cr = crp;
6492 	cmdp->cb_cmd = scp->sc_cmd;
6493 	cmdp->cb_len = scp->sc_len;
6494 	cmdp->cb_error = 0;
6495 	mp->b_wptr += sizeof (struct cmdblk);
6496 
6497 	DB_TYPE(mp) = M_CMD;
6498 	DB_CPID(mp) = curproc->p_pid;
6499 
6500 	/*
6501 	 * Copy in the payload.
6502 	 */
6503 	if (cmdp->cb_len > 0) {
6504 		mp->b_cont = allocb_cred(sizeof (scp->sc_buf), crp);
6505 		if (mp->b_cont == NULL) {
6506 			error = ENOMEM;
6507 			goto out;
6508 		}
6509 
6510 		/* cb_len comes from sc_len, which has already been checked */
6511 		ASSERT(cmdp->cb_len <= sizeof (scp->sc_buf));
6512 		(void) bcopy(scp->sc_buf, mp->b_cont->b_wptr, cmdp->cb_len);
6513 		mp->b_cont->b_wptr += cmdp->cb_len;
6514 		DB_CPID(mp->b_cont) = curproc->p_pid;
6515 	}
6516 
6517 	/*
6518 	 * Since this mechanism is strictly for ptools, and since only one
6519 	 * process can be grabbed at a time, we simply fail if there's
6520 	 * currently an operation pending.
6521 	 */
6522 	mutex_enter(&stp->sd_lock);
6523 	if (stp->sd_flag & STRCMDWAIT) {
6524 		mutex_exit(&stp->sd_lock);
6525 		error = EBUSY;
6526 		goto out;
6527 	}
6528 	stp->sd_flag |= STRCMDWAIT;
6529 	ASSERT(stp->sd_cmdblk == NULL);
6530 	mutex_exit(&stp->sd_lock);
6531 
6532 	putnext(stp->sd_wrq, mp);
6533 	mp = NULL;
6534 
6535 	/*
6536 	 * Timed wait for acknowledgment.  If the reply has already arrived,
6537 	 * don't sleep.  If awakened from the sleep, fail only if the reply
6538 	 * has not arrived by then.  Otherwise, process the reply.
6539 	 */
6540 	mutex_enter(&stp->sd_lock);
6541 	while (stp->sd_cmdblk == NULL) {
6542 		if (stp->sd_flag & errs) {
6543 			if ((error = strgeterr(stp, errs, 0)) != 0)
6544 				goto waitout;
6545 		}
6546 
6547 		rval = str_cv_wait(&stp->sd_monitor, &stp->sd_lock, timeout, 0);
6548 		if (stp->sd_cmdblk != NULL)
6549 			break;
6550 
6551 		if (rval <= 0) {
6552 			error = (rval == 0) ? EINTR : ETIME;
6553 			goto waitout;
6554 		}
6555 	}
6556 
6557 	/*
6558 	 * We received a reply.
6559 	 */
6560 	mp = stp->sd_cmdblk;
6561 	stp->sd_cmdblk = NULL;
6562 	ASSERT(mp != NULL && DB_TYPE(mp) == M_CMD);
6563 	ASSERT(stp->sd_flag & STRCMDWAIT);
6564 	stp->sd_flag &= ~STRCMDWAIT;
6565 	mutex_exit(&stp->sd_lock);
6566 
6567 	cmdp = (struct cmdblk *)mp->b_rptr;
6568 	if ((error = cmdp->cb_error) != 0)
6569 		goto out;
6570 
6571 	/*
6572 	 * Data may have been returned in the reply (cb_len > 0).
6573 	 * If so, copy it out to the user's buffer.
6574 	 */
6575 	if (cmdp->cb_len > 0) {
6576 		if (mp->b_cont == NULL || MBLKL(mp->b_cont) < cmdp->cb_len) {
6577 			error = EPROTO;
6578 			goto out;
6579 		}
6580 
6581 		cmdp->cb_len = MIN(cmdp->cb_len, sizeof (scp->sc_buf));
6582 		(void) bcopy(mp->b_cont->b_rptr, scp->sc_buf, cmdp->cb_len);
6583 	}
6584 	scp->sc_len = cmdp->cb_len;
6585 out:
6586 	freemsg(mp);
6587 	crfree(crp);
6588 	return (error);
6589 waitout:
6590 	ASSERT(stp->sd_cmdblk == NULL);
6591 	stp->sd_flag &= ~STRCMDWAIT;
6592 	mutex_exit(&stp->sd_lock);
6593 	crfree(crp);
6594 	return (error);
6595 }
6596 
6597 /*
6598  * For the SunOS keyboard driver.
6599  * Return the next available "ioctl" sequence number.
6600  * Exported, so that streams modules can send "ioctl" messages
6601  * downstream from their open routine.
6602  */
6603 int
6604 getiocseqno(void)
6605 {
6606 	int	i;
6607 
6608 	mutex_enter(&strresources);
6609 	i = ++ioc_id;
6610 	mutex_exit(&strresources);
6611 	return (i);
6612 }
6613 
6614 /*
6615  * Get the next message from the read queue.  If the message is
6616  * priority, STRPRI will have been set by strrput().  This flag
6617  * should be reset only when the entire message at the front of the
6618  * queue as been consumed.
6619  *
6620  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
6621  */
6622 int
6623 strgetmsg(
6624 	struct vnode *vp,
6625 	struct strbuf *mctl,
6626 	struct strbuf *mdata,
6627 	unsigned char *prip,
6628 	int *flagsp,
6629 	int fmode,
6630 	rval_t *rvp)
6631 {
6632 	struct stdata *stp;
6633 	mblk_t *bp, *nbp;
6634 	mblk_t *savemp = NULL;
6635 	mblk_t *savemptail = NULL;
6636 	uint_t old_sd_flag;
6637 	int flg;
6638 	int more = 0;
6639 	int error = 0;
6640 	char first = 1;
6641 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
6642 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
6643 	unsigned char pri = 0;
6644 	queue_t *q;
6645 	int	pr = 0;			/* Partial read successful */
6646 	struct uio uios;
6647 	struct uio *uiop = &uios;
6648 	struct iovec iovs;
6649 	unsigned char type;
6650 
6651 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRGETMSG_ENTER,
6652 	    "strgetmsg:%p", vp);
6653 
6654 	ASSERT(vp->v_stream);
6655 	stp = vp->v_stream;
6656 	rvp->r_val1 = 0;
6657 
6658 	mutex_enter(&stp->sd_lock);
6659 
6660 	if ((error = i_straccess(stp, JCREAD)) != 0) {
6661 		mutex_exit(&stp->sd_lock);
6662 		return (error);
6663 	}
6664 
6665 	if (stp->sd_flag & (STRDERR|STPLEX)) {
6666 		error = strgeterr(stp, STRDERR|STPLEX, 0);
6667 		if (error != 0) {
6668 			mutex_exit(&stp->sd_lock);
6669 			return (error);
6670 		}
6671 	}
6672 	mutex_exit(&stp->sd_lock);
6673 
6674 	switch (*flagsp) {
6675 	case MSG_HIPRI:
6676 		if (*prip != 0)
6677 			return (EINVAL);
6678 		break;
6679 
6680 	case MSG_ANY:
6681 	case MSG_BAND:
6682 		break;
6683 
6684 	default:
6685 		return (EINVAL);
6686 	}
6687 	/*
6688 	 * Setup uio and iov for data part
6689 	 */
6690 	iovs.iov_base = mdata->buf;
6691 	iovs.iov_len = mdata->maxlen;
6692 	uios.uio_iov = &iovs;
6693 	uios.uio_iovcnt = 1;
6694 	uios.uio_loffset = 0;
6695 	uios.uio_segflg = UIO_USERSPACE;
6696 	uios.uio_fmode = 0;
6697 	uios.uio_extflg = UIO_COPY_CACHED;
6698 	uios.uio_resid = mdata->maxlen;
6699 	uios.uio_offset = 0;
6700 
6701 	q = _RD(stp->sd_wrq);
6702 	mutex_enter(&stp->sd_lock);
6703 	old_sd_flag = stp->sd_flag;
6704 	mark = 0;
6705 	for (;;) {
6706 		int done = 0;
6707 		mblk_t *q_first = q->q_first;
6708 
6709 		/*
6710 		 * Get the next message of appropriate priority
6711 		 * from the stream head.  If the caller is interested
6712 		 * in band or hipri messages, then they should already
6713 		 * be enqueued at the stream head.  On the other hand
6714 		 * if the caller wants normal (band 0) messages, they
6715 		 * might be deferred in a synchronous stream and they
6716 		 * will need to be pulled up.
6717 		 *
6718 		 * After we have dequeued a message, we might find that
6719 		 * it was a deferred M_SIG that was enqueued at the
6720 		 * stream head.  It must now be posted as part of the
6721 		 * read by calling strsignal_nolock().
6722 		 *
6723 		 * Also note that strrput does not enqueue an M_PCSIG,
6724 		 * and there cannot be more than one hipri message,
6725 		 * so there was no need to have the M_PCSIG case.
6726 		 *
6727 		 * At some time it might be nice to try and wrap the
6728 		 * functionality of kstrgetmsg() and strgetmsg() into
6729 		 * a common routine so to reduce the amount of replicated
6730 		 * code (since they are extremely similar).
6731 		 */
6732 		if (!(*flagsp & (MSG_HIPRI|MSG_BAND))) {
6733 			/* Asking for normal, band0 data */
6734 			bp = strget(stp, q, uiop, first, &error);
6735 			ASSERT(MUTEX_HELD(&stp->sd_lock));
6736 			if (bp != NULL) {
6737 				ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
6738 				if (bp->b_datap->db_type == M_SIG) {
6739 					strsignal_nolock(stp, *bp->b_rptr,
6740 					    (int32_t)bp->b_band);
6741 					continue;
6742 				} else {
6743 					break;
6744 				}
6745 			}
6746 			if (error != 0) {
6747 				goto getmout;
6748 			}
6749 
6750 		/*
6751 		 * We can't depend on the value of STRPRI here because
6752 		 * the stream head may be in transit. Therefore, we
6753 		 * must look at the type of the first message to
6754 		 * determine if a high priority messages is waiting
6755 		 */
6756 		} else if ((*flagsp & MSG_HIPRI) && q_first != NULL &&
6757 		    q_first->b_datap->db_type >= QPCTL &&
6758 		    (bp = getq_noenab(q, 0)) != NULL) {
6759 			/* Asked for HIPRI and got one */
6760 			ASSERT(bp->b_datap->db_type >= QPCTL);
6761 			break;
6762 		} else if ((*flagsp & MSG_BAND) && q_first != NULL &&
6763 		    ((q_first->b_band >= *prip) ||
6764 		    q_first->b_datap->db_type >= QPCTL) &&
6765 		    (bp = getq_noenab(q, 0)) != NULL) {
6766 			/*
6767 			 * Asked for at least band "prip" and got either at
6768 			 * least that band or a hipri message.
6769 			 */
6770 			ASSERT(bp->b_band >= *prip ||
6771 			    bp->b_datap->db_type >= QPCTL);
6772 			if (bp->b_datap->db_type == M_SIG) {
6773 				strsignal_nolock(stp, *bp->b_rptr,
6774 				    (int32_t)bp->b_band);
6775 				continue;
6776 			} else {
6777 				break;
6778 			}
6779 		}
6780 
6781 		/* No data. Time to sleep? */
6782 		qbackenable(q, 0);
6783 
6784 		/*
6785 		 * If STRHUP or STREOF, return 0 length control and data.
6786 		 * If resid is 0, then a read(fd,buf,0) was done. Do not
6787 		 * sleep to satisfy this request because by default we have
6788 		 * zero bytes to return.
6789 		 */
6790 		if ((stp->sd_flag & (STRHUP|STREOF)) || (mctl->maxlen == 0 &&
6791 		    mdata->maxlen == 0)) {
6792 			mctl->len = mdata->len = 0;
6793 			*flagsp = 0;
6794 			mutex_exit(&stp->sd_lock);
6795 			return (0);
6796 		}
6797 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_WAIT,
6798 		    "strgetmsg calls strwaitq:%p, %p",
6799 		    vp, uiop);
6800 		if (((error = strwaitq(stp, GETWAIT, (ssize_t)0, fmode, -1,
6801 		    &done)) != 0) || done) {
6802 			TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_DONE,
6803 			    "strgetmsg error or done:%p, %p",
6804 			    vp, uiop);
6805 			mutex_exit(&stp->sd_lock);
6806 			return (error);
6807 		}
6808 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRGETMSG_AWAKE,
6809 		    "strgetmsg awakes:%p, %p", vp, uiop);
6810 		if ((error = i_straccess(stp, JCREAD)) != 0) {
6811 			mutex_exit(&stp->sd_lock);
6812 			return (error);
6813 		}
6814 		first = 0;
6815 	}
6816 	ASSERT(bp != NULL);
6817 	/*
6818 	 * Extract any mark information. If the message is not completely
6819 	 * consumed this information will be put in the mblk
6820 	 * that is putback.
6821 	 * If MSGMARKNEXT is set and the message is completely consumed
6822 	 * the STRATMARK flag will be set below. Likewise, if
6823 	 * MSGNOTMARKNEXT is set and the message is
6824 	 * completely consumed STRNOTATMARK will be set.
6825 	 */
6826 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
6827 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
6828 	    (MSGMARKNEXT|MSGNOTMARKNEXT));
6829 	if (mark != 0 && bp == stp->sd_mark) {
6830 		mark |= _LASTMARK;
6831 		stp->sd_mark = NULL;
6832 	}
6833 	/*
6834 	 * keep track of the original message type and priority
6835 	 */
6836 	pri = bp->b_band;
6837 	type = bp->b_datap->db_type;
6838 	if (type == M_PASSFP) {
6839 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
6840 			stp->sd_mark = bp;
6841 		bp->b_flag |= mark & ~_LASTMARK;
6842 		putback(stp, q, bp, pri);
6843 		qbackenable(q, pri);
6844 		mutex_exit(&stp->sd_lock);
6845 		return (EBADMSG);
6846 	}
6847 	ASSERT(type != M_SIG);
6848 
6849 	/*
6850 	 * Set this flag so strrput will not generate signals. Need to
6851 	 * make sure this flag is cleared before leaving this routine
6852 	 * else signals will stop being sent.
6853 	 */
6854 	stp->sd_flag |= STRGETINPROG;
6855 	mutex_exit(&stp->sd_lock);
6856 
6857 	if (STREAM_NEEDSERVICE(stp))
6858 		stream_runservice(stp);
6859 
6860 	/*
6861 	 * Set HIPRI flag if message is priority.
6862 	 */
6863 	if (type >= QPCTL)
6864 		flg = MSG_HIPRI;
6865 	else
6866 		flg = MSG_BAND;
6867 
6868 	/*
6869 	 * First process PROTO or PCPROTO blocks, if any.
6870 	 */
6871 	if (mctl->maxlen >= 0 && type != M_DATA) {
6872 		size_t	n, bcnt;
6873 		char	*ubuf;
6874 
6875 		bcnt = mctl->maxlen;
6876 		ubuf = mctl->buf;
6877 		while (bp != NULL && bp->b_datap->db_type != M_DATA) {
6878 			if ((n = MIN(bcnt, bp->b_wptr - bp->b_rptr)) != 0 &&
6879 			    copyout(bp->b_rptr, ubuf, n)) {
6880 				error = EFAULT;
6881 				mutex_enter(&stp->sd_lock);
6882 				/*
6883 				 * clear stream head pri flag based on
6884 				 * first message type
6885 				 */
6886 				if (type >= QPCTL) {
6887 					ASSERT(type == M_PCPROTO);
6888 					stp->sd_flag &= ~STRPRI;
6889 				}
6890 				more = 0;
6891 				freemsg(bp);
6892 				goto getmout;
6893 			}
6894 			ubuf += n;
6895 			bp->b_rptr += n;
6896 			if (bp->b_rptr >= bp->b_wptr) {
6897 				nbp = bp;
6898 				bp = bp->b_cont;
6899 				freeb(nbp);
6900 			}
6901 			ASSERT(n <= bcnt);
6902 			bcnt -= n;
6903 			if (bcnt == 0)
6904 				break;
6905 		}
6906 		mctl->len = mctl->maxlen - bcnt;
6907 	} else
6908 		mctl->len = -1;
6909 
6910 	if (bp && bp->b_datap->db_type != M_DATA) {
6911 		/*
6912 		 * More PROTO blocks in msg.
6913 		 */
6914 		more |= MORECTL;
6915 		savemp = bp;
6916 		while (bp && bp->b_datap->db_type != M_DATA) {
6917 			savemptail = bp;
6918 			bp = bp->b_cont;
6919 		}
6920 		savemptail->b_cont = NULL;
6921 	}
6922 
6923 	/*
6924 	 * Now process DATA blocks, if any.
6925 	 */
6926 	if (mdata->maxlen >= 0 && bp) {
6927 		/*
6928 		 * struiocopyout will consume a potential zero-length
6929 		 * M_DATA even if uio_resid is zero.
6930 		 */
6931 		size_t oldresid = uiop->uio_resid;
6932 
6933 		bp = struiocopyout(bp, uiop, &error);
6934 		if (error != 0) {
6935 			mutex_enter(&stp->sd_lock);
6936 			/*
6937 			 * clear stream head hi pri flag based on
6938 			 * first message
6939 			 */
6940 			if (type >= QPCTL) {
6941 				ASSERT(type == M_PCPROTO);
6942 				stp->sd_flag &= ~STRPRI;
6943 			}
6944 			more = 0;
6945 			freemsg(savemp);
6946 			goto getmout;
6947 		}
6948 		/*
6949 		 * (pr == 1) indicates a partial read.
6950 		 */
6951 		if (oldresid > uiop->uio_resid)
6952 			pr = 1;
6953 		mdata->len = mdata->maxlen - uiop->uio_resid;
6954 	} else
6955 		mdata->len = -1;
6956 
6957 	if (bp) {			/* more data blocks in msg */
6958 		more |= MOREDATA;
6959 		if (savemp)
6960 			savemptail->b_cont = bp;
6961 		else
6962 			savemp = bp;
6963 	}
6964 
6965 	mutex_enter(&stp->sd_lock);
6966 	if (savemp) {
6967 		if (pr && (savemp->b_datap->db_type == M_DATA) &&
6968 		    msgnodata(savemp)) {
6969 			/*
6970 			 * Avoid queuing a zero-length tail part of
6971 			 * a message. pr=1 indicates that we read some of
6972 			 * the message.
6973 			 */
6974 			freemsg(savemp);
6975 			more &= ~MOREDATA;
6976 			/*
6977 			 * clear stream head hi pri flag based on
6978 			 * first message
6979 			 */
6980 			if (type >= QPCTL) {
6981 				ASSERT(type == M_PCPROTO);
6982 				stp->sd_flag &= ~STRPRI;
6983 			}
6984 		} else {
6985 			savemp->b_band = pri;
6986 			/*
6987 			 * If the first message was HIPRI and the one we're
6988 			 * putting back isn't, then clear STRPRI, otherwise
6989 			 * set STRPRI again.  Note that we must set STRPRI
6990 			 * again since the flush logic in strrput_nondata()
6991 			 * may have cleared it while we had sd_lock dropped.
6992 			 */
6993 			if (type >= QPCTL) {
6994 				ASSERT(type == M_PCPROTO);
6995 				if (queclass(savemp) < QPCTL)
6996 					stp->sd_flag &= ~STRPRI;
6997 				else
6998 					stp->sd_flag |= STRPRI;
6999 			} else if (queclass(savemp) >= QPCTL) {
7000 				/*
7001 				 * The first message was not a HIPRI message,
7002 				 * but the one we are about to putback is.
7003 				 * For simplicitly, we do not allow for HIPRI
7004 				 * messages to be embedded in the message
7005 				 * body, so just force it to same type as
7006 				 * first message.
7007 				 */
7008 				ASSERT(type == M_DATA || type == M_PROTO);
7009 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
7010 				savemp->b_datap->db_type = type;
7011 			}
7012 			if (mark != 0) {
7013 				savemp->b_flag |= mark & ~_LASTMARK;
7014 				if ((mark & _LASTMARK) &&
7015 				    (stp->sd_mark == NULL)) {
7016 					/*
7017 					 * If another marked message arrived
7018 					 * while sd_lock was not held sd_mark
7019 					 * would be non-NULL.
7020 					 */
7021 					stp->sd_mark = savemp;
7022 				}
7023 			}
7024 			putback(stp, q, savemp, pri);
7025 		}
7026 	} else {
7027 		/*
7028 		 * The complete message was consumed.
7029 		 *
7030 		 * If another M_PCPROTO arrived while sd_lock was not held
7031 		 * it would have been discarded since STRPRI was still set.
7032 		 *
7033 		 * Move the MSG*MARKNEXT information
7034 		 * to the stream head just in case
7035 		 * the read queue becomes empty.
7036 		 * clear stream head hi pri flag based on
7037 		 * first message
7038 		 *
7039 		 * If the stream head was at the mark
7040 		 * (STRATMARK) before we dropped sd_lock above
7041 		 * and some data was consumed then we have
7042 		 * moved past the mark thus STRATMARK is
7043 		 * cleared. However, if a message arrived in
7044 		 * strrput during the copyout above causing
7045 		 * STRATMARK to be set we can not clear that
7046 		 * flag.
7047 		 */
7048 		if (type >= QPCTL) {
7049 			ASSERT(type == M_PCPROTO);
7050 			stp->sd_flag &= ~STRPRI;
7051 		}
7052 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
7053 			if (mark & MSGMARKNEXT) {
7054 				stp->sd_flag &= ~STRNOTATMARK;
7055 				stp->sd_flag |= STRATMARK;
7056 			} else if (mark & MSGNOTMARKNEXT) {
7057 				stp->sd_flag &= ~STRATMARK;
7058 				stp->sd_flag |= STRNOTATMARK;
7059 			} else {
7060 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
7061 			}
7062 		} else if (pr && (old_sd_flag & STRATMARK)) {
7063 			stp->sd_flag &= ~STRATMARK;
7064 		}
7065 	}
7066 
7067 	*flagsp = flg;
7068 	*prip = pri;
7069 
7070 	/*
7071 	 * Getmsg cleanup processing - if the state of the queue has changed
7072 	 * some signals may need to be sent and/or poll awakened.
7073 	 */
7074 getmout:
7075 	qbackenable(q, pri);
7076 
7077 	/*
7078 	 * We dropped the stream head lock above. Send all M_SIG messages
7079 	 * before processing stream head for SIGPOLL messages.
7080 	 */
7081 	ASSERT(MUTEX_HELD(&stp->sd_lock));
7082 	while ((bp = q->q_first) != NULL &&
7083 	    (bp->b_datap->db_type == M_SIG)) {
7084 		/*
7085 		 * sd_lock is held so the content of the read queue can not
7086 		 * change.
7087 		 */
7088 		bp = getq(q);
7089 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7090 
7091 		strsignal_nolock(stp, *bp->b_rptr, (int32_t)bp->b_band);
7092 		mutex_exit(&stp->sd_lock);
7093 		freemsg(bp);
7094 		if (STREAM_NEEDSERVICE(stp))
7095 			stream_runservice(stp);
7096 		mutex_enter(&stp->sd_lock);
7097 	}
7098 
7099 	/*
7100 	 * stream head cannot change while we make the determination
7101 	 * whether or not to send a signal. Drop the flag to allow strrput
7102 	 * to send firstmsgsigs again.
7103 	 */
7104 	stp->sd_flag &= ~STRGETINPROG;
7105 
7106 	/*
7107 	 * If the type of message at the front of the queue changed
7108 	 * due to the receive the appropriate signals and pollwakeup events
7109 	 * are generated. The type of changes are:
7110 	 *	Processed a hipri message, q_first is not hipri.
7111 	 *	Processed a band X message, and q_first is band Y.
7112 	 * The generated signals and pollwakeups are identical to what
7113 	 * strrput() generates should the message that is now on q_first
7114 	 * arrive to an empty read queue.
7115 	 *
7116 	 * Note: only strrput will send a signal for a hipri message.
7117 	 */
7118 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7119 		strsigset_t signals = 0;
7120 		strpollset_t pollwakeups = 0;
7121 
7122 		if (flg & MSG_HIPRI) {
7123 			/*
7124 			 * Removed a hipri message. Regular data at
7125 			 * the front of  the queue.
7126 			 */
7127 			if (bp->b_band == 0) {
7128 				signals = S_INPUT | S_RDNORM;
7129 				pollwakeups = POLLIN | POLLRDNORM;
7130 			} else {
7131 				signals = S_INPUT | S_RDBAND;
7132 				pollwakeups = POLLIN | POLLRDBAND;
7133 			}
7134 		} else if (pri != bp->b_band) {
7135 			/*
7136 			 * The band is different for the new q_first.
7137 			 */
7138 			if (bp->b_band == 0) {
7139 				signals = S_RDNORM;
7140 				pollwakeups = POLLIN | POLLRDNORM;
7141 			} else {
7142 				signals = S_RDBAND;
7143 				pollwakeups = POLLIN | POLLRDBAND;
7144 			}
7145 		}
7146 
7147 		if (pollwakeups != 0) {
7148 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
7149 				if (!(stp->sd_rput_opt & SR_POLLIN))
7150 					goto no_pollwake;
7151 				stp->sd_rput_opt &= ~SR_POLLIN;
7152 			}
7153 			mutex_exit(&stp->sd_lock);
7154 			pollwakeup(&stp->sd_pollist, pollwakeups);
7155 			mutex_enter(&stp->sd_lock);
7156 		}
7157 no_pollwake:
7158 
7159 		if (stp->sd_sigflags & signals)
7160 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7161 	}
7162 	mutex_exit(&stp->sd_lock);
7163 
7164 	rvp->r_val1 = more;
7165 	return (error);
7166 #undef	_LASTMARK
7167 }
7168 
7169 /*
7170  * Get the next message from the read queue.  If the message is
7171  * priority, STRPRI will have been set by strrput().  This flag
7172  * should be reset only when the entire message at the front of the
7173  * queue as been consumed.
7174  *
7175  * If uiop is NULL all data is returned in mctlp.
7176  * Note that a NULL uiop implies that FNDELAY and FNONBLOCK are assumed
7177  * not enabled.
7178  * The timeout parameter is in milliseconds; -1 for infinity.
7179  * This routine handles the consolidation private flags:
7180  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
7181  *	MSG_DELAYERROR	Defer the error check until the queue is empty.
7182  *	MSG_HOLDSIG	Hold signals while waiting for data.
7183  *	MSG_IPEEK	Only peek at messages.
7184  *	MSG_DISCARDTAIL	Discard the tail M_DATA part of the message
7185  *			that doesn't fit.
7186  *	MSG_NOMARK	If the message is marked leave it on the queue.
7187  *
7188  * NOTE: strgetmsg and kstrgetmsg have much of the logic in common.
7189  */
7190 int
7191 kstrgetmsg(
7192 	struct vnode *vp,
7193 	mblk_t **mctlp,
7194 	struct uio *uiop,
7195 	unsigned char *prip,
7196 	int *flagsp,
7197 	clock_t timout,
7198 	rval_t *rvp)
7199 {
7200 	struct stdata *stp;
7201 	mblk_t *bp, *nbp;
7202 	mblk_t *savemp = NULL;
7203 	mblk_t *savemptail = NULL;
7204 	int flags;
7205 	uint_t old_sd_flag;
7206 	int flg;
7207 	int more = 0;
7208 	int error = 0;
7209 	char first = 1;
7210 	uint_t mark;		/* Contains MSG*MARK and _LASTMARK */
7211 #define	_LASTMARK	0x8000	/* Distinct from MSG*MARK */
7212 	unsigned char pri = 0;
7213 	queue_t *q;
7214 	int	pr = 0;			/* Partial read successful */
7215 	unsigned char type;
7216 
7217 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_ENTER,
7218 	    "kstrgetmsg:%p", vp);
7219 
7220 	ASSERT(vp->v_stream);
7221 	stp = vp->v_stream;
7222 	rvp->r_val1 = 0;
7223 
7224 	mutex_enter(&stp->sd_lock);
7225 
7226 	if ((error = i_straccess(stp, JCREAD)) != 0) {
7227 		mutex_exit(&stp->sd_lock);
7228 		return (error);
7229 	}
7230 
7231 	flags = *flagsp;
7232 	if (stp->sd_flag & (STRDERR|STPLEX)) {
7233 		if ((stp->sd_flag & STPLEX) ||
7234 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == 0) {
7235 			error = strgeterr(stp, STRDERR|STPLEX,
7236 			    (flags & MSG_IPEEK));
7237 			if (error != 0) {
7238 				mutex_exit(&stp->sd_lock);
7239 				return (error);
7240 			}
7241 		}
7242 	}
7243 	mutex_exit(&stp->sd_lock);
7244 
7245 	switch (flags & (MSG_HIPRI|MSG_ANY|MSG_BAND)) {
7246 	case MSG_HIPRI:
7247 		if (*prip != 0)
7248 			return (EINVAL);
7249 		break;
7250 
7251 	case MSG_ANY:
7252 	case MSG_BAND:
7253 		break;
7254 
7255 	default:
7256 		return (EINVAL);
7257 	}
7258 
7259 retry:
7260 	q = _RD(stp->sd_wrq);
7261 	mutex_enter(&stp->sd_lock);
7262 	old_sd_flag = stp->sd_flag;
7263 	mark = 0;
7264 	for (;;) {
7265 		int done = 0;
7266 		int waitflag;
7267 		int fmode;
7268 		mblk_t *q_first = q->q_first;
7269 
7270 		/*
7271 		 * This section of the code operates just like the code
7272 		 * in strgetmsg().  There is a comment there about what
7273 		 * is going on here.
7274 		 */
7275 		if (!(flags & (MSG_HIPRI|MSG_BAND))) {
7276 			/* Asking for normal, band0 data */
7277 			bp = strget(stp, q, uiop, first, &error);
7278 			ASSERT(MUTEX_HELD(&stp->sd_lock));
7279 			if (bp != NULL) {
7280 				if (bp->b_datap->db_type == M_SIG) {
7281 					strsignal_nolock(stp, *bp->b_rptr,
7282 					    (int32_t)bp->b_band);
7283 					continue;
7284 				} else {
7285 					break;
7286 				}
7287 			}
7288 			if (error != 0) {
7289 				goto getmout;
7290 			}
7291 		/*
7292 		 * We can't depend on the value of STRPRI here because
7293 		 * the stream head may be in transit. Therefore, we
7294 		 * must look at the type of the first message to
7295 		 * determine if a high priority messages is waiting
7296 		 */
7297 		} else if ((flags & MSG_HIPRI) && q_first != NULL &&
7298 		    q_first->b_datap->db_type >= QPCTL &&
7299 		    (bp = getq_noenab(q, 0)) != NULL) {
7300 			ASSERT(bp->b_datap->db_type >= QPCTL);
7301 			break;
7302 		} else if ((flags & MSG_BAND) && q_first != NULL &&
7303 		    ((q_first->b_band >= *prip) ||
7304 		    q_first->b_datap->db_type >= QPCTL) &&
7305 		    (bp = getq_noenab(q, 0)) != NULL) {
7306 			/*
7307 			 * Asked for at least band "prip" and got either at
7308 			 * least that band or a hipri message.
7309 			 */
7310 			ASSERT(bp->b_band >= *prip ||
7311 			    bp->b_datap->db_type >= QPCTL);
7312 			if (bp->b_datap->db_type == M_SIG) {
7313 				strsignal_nolock(stp, *bp->b_rptr,
7314 				    (int32_t)bp->b_band);
7315 				continue;
7316 			} else {
7317 				break;
7318 			}
7319 		}
7320 
7321 		/* No data. Time to sleep? */
7322 		qbackenable(q, 0);
7323 
7324 		/*
7325 		 * Delayed error notification?
7326 		 */
7327 		if ((stp->sd_flag & (STRDERR|STPLEX)) &&
7328 		    (flags & (MSG_IGNERROR|MSG_DELAYERROR)) == MSG_DELAYERROR) {
7329 			error = strgeterr(stp, STRDERR|STPLEX,
7330 			    (flags & MSG_IPEEK));
7331 			if (error != 0) {
7332 				mutex_exit(&stp->sd_lock);
7333 				return (error);
7334 			}
7335 		}
7336 
7337 		/*
7338 		 * If STRHUP or STREOF, return 0 length control and data.
7339 		 * If a read(fd,buf,0) has been done, do not sleep, just
7340 		 * return.
7341 		 *
7342 		 * If mctlp == NULL and uiop == NULL, then the code will
7343 		 * do the strwaitq. This is an understood way of saying
7344 		 * sleep "polling" until a message is received.
7345 		 */
7346 		if ((stp->sd_flag & (STRHUP|STREOF)) ||
7347 		    (uiop != NULL && uiop->uio_resid == 0)) {
7348 			if (mctlp != NULL)
7349 				*mctlp = NULL;
7350 			*flagsp = 0;
7351 			mutex_exit(&stp->sd_lock);
7352 			return (0);
7353 		}
7354 
7355 		waitflag = GETWAIT;
7356 		if (flags &
7357 		    (MSG_HOLDSIG|MSG_IGNERROR|MSG_IPEEK|MSG_DELAYERROR)) {
7358 			if (flags & MSG_HOLDSIG)
7359 				waitflag |= STR_NOSIG;
7360 			if (flags & MSG_IGNERROR)
7361 				waitflag |= STR_NOERROR;
7362 			if (flags & MSG_IPEEK)
7363 				waitflag |= STR_PEEK;
7364 			if (flags & MSG_DELAYERROR)
7365 				waitflag |= STR_DELAYERR;
7366 		}
7367 		if (uiop != NULL)
7368 			fmode = uiop->uio_fmode;
7369 		else
7370 			fmode = 0;
7371 
7372 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_WAIT,
7373 		    "kstrgetmsg calls strwaitq:%p, %p",
7374 		    vp, uiop);
7375 		if (((error = strwaitq(stp, waitflag, (ssize_t)0,
7376 		    fmode, timout, &done))) != 0 || done) {
7377 			TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_DONE,
7378 			    "kstrgetmsg error or done:%p, %p",
7379 			    vp, uiop);
7380 			mutex_exit(&stp->sd_lock);
7381 			return (error);
7382 		}
7383 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRGETMSG_AWAKE,
7384 		    "kstrgetmsg awakes:%p, %p", vp, uiop);
7385 		if ((error = i_straccess(stp, JCREAD)) != 0) {
7386 			mutex_exit(&stp->sd_lock);
7387 			return (error);
7388 		}
7389 		first = 0;
7390 	}
7391 	ASSERT(bp != NULL);
7392 	/*
7393 	 * Extract any mark information. If the message is not completely
7394 	 * consumed this information will be put in the mblk
7395 	 * that is putback.
7396 	 * If MSGMARKNEXT is set and the message is completely consumed
7397 	 * the STRATMARK flag will be set below. Likewise, if
7398 	 * MSGNOTMARKNEXT is set and the message is
7399 	 * completely consumed STRNOTATMARK will be set.
7400 	 */
7401 	mark = bp->b_flag & (MSGMARK | MSGMARKNEXT | MSGNOTMARKNEXT);
7402 	ASSERT((mark & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
7403 	    (MSGMARKNEXT|MSGNOTMARKNEXT));
7404 	pri = bp->b_band;
7405 	if (mark != 0) {
7406 		/*
7407 		 * If the caller doesn't want the mark return.
7408 		 * Used to implement MSG_WAITALL in sockets.
7409 		 */
7410 		if (flags & MSG_NOMARK) {
7411 			putback(stp, q, bp, pri);
7412 			qbackenable(q, pri);
7413 			mutex_exit(&stp->sd_lock);
7414 			return (EWOULDBLOCK);
7415 		}
7416 		if (bp == stp->sd_mark) {
7417 			mark |= _LASTMARK;
7418 			stp->sd_mark = NULL;
7419 		}
7420 	}
7421 
7422 	/*
7423 	 * keep track of the first message type
7424 	 */
7425 	type = bp->b_datap->db_type;
7426 
7427 	if (bp->b_datap->db_type == M_PASSFP) {
7428 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7429 			stp->sd_mark = bp;
7430 		bp->b_flag |= mark & ~_LASTMARK;
7431 		putback(stp, q, bp, pri);
7432 		qbackenable(q, pri);
7433 		mutex_exit(&stp->sd_lock);
7434 		return (EBADMSG);
7435 	}
7436 	ASSERT(type != M_SIG);
7437 
7438 	if (flags & MSG_IPEEK) {
7439 		/*
7440 		 * Clear any struioflag - we do the uiomove over again
7441 		 * when peeking since it simplifies the code.
7442 		 *
7443 		 * Dup the message and put the original back on the queue.
7444 		 * If dupmsg() fails, try again with copymsg() to see if
7445 		 * there is indeed a shortage of memory.  dupmsg() may fail
7446 		 * if db_ref in any of the messages reaches its limit.
7447 		 */
7448 
7449 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
7450 		if ((nbp = dupmsg(bp)) == NULL && (nbp = copymsg(bp)) == NULL) {
7451 			/*
7452 			 * Restore the state of the stream head since we
7453 			 * need to drop sd_lock (strwaitbuf is sleeping).
7454 			 */
7455 			size_t size = msgdsize(bp);
7456 
7457 			if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7458 				stp->sd_mark = bp;
7459 			bp->b_flag |= mark & ~_LASTMARK;
7460 			putback(stp, q, bp, pri);
7461 			mutex_exit(&stp->sd_lock);
7462 			error = strwaitbuf(size, BPRI_HI);
7463 			if (error) {
7464 				/*
7465 				 * There is no net change to the queue thus
7466 				 * no need to qbackenable.
7467 				 */
7468 				return (error);
7469 			}
7470 			goto retry;
7471 		}
7472 
7473 		if ((mark & _LASTMARK) && (stp->sd_mark == NULL))
7474 			stp->sd_mark = bp;
7475 		bp->b_flag |= mark & ~_LASTMARK;
7476 		putback(stp, q, bp, pri);
7477 		bp = nbp;
7478 	}
7479 
7480 	/*
7481 	 * Set this flag so strrput will not generate signals. Need to
7482 	 * make sure this flag is cleared before leaving this routine
7483 	 * else signals will stop being sent.
7484 	 */
7485 	stp->sd_flag |= STRGETINPROG;
7486 	mutex_exit(&stp->sd_lock);
7487 
7488 	if ((stp->sd_rputdatafunc != NULL) && (DB_TYPE(bp) == M_DATA)) {
7489 		mblk_t *tmp, *prevmp;
7490 
7491 		/*
7492 		 * Put first non-data mblk back to stream head and
7493 		 * cut the mblk chain so sd_rputdatafunc only sees
7494 		 * M_DATA mblks. We can skip the first mblk since it
7495 		 * is M_DATA according to the condition above.
7496 		 */
7497 		for (prevmp = bp, tmp = bp->b_cont; tmp != NULL;
7498 		    prevmp = tmp, tmp = tmp->b_cont) {
7499 			if (DB_TYPE(tmp) != M_DATA) {
7500 				prevmp->b_cont = NULL;
7501 				mutex_enter(&stp->sd_lock);
7502 				putback(stp, q, tmp, tmp->b_band);
7503 				mutex_exit(&stp->sd_lock);
7504 				break;
7505 			}
7506 		}
7507 
7508 		ASSERT(!(bp->b_datap->db_flags & DBLK_UIOA));
7509 		bp = (stp->sd_rputdatafunc)(stp->sd_vnode, bp,
7510 		    NULL, NULL, NULL, NULL);
7511 
7512 		if (bp == NULL)
7513 			goto retry;
7514 	}
7515 
7516 	if (STREAM_NEEDSERVICE(stp))
7517 		stream_runservice(stp);
7518 
7519 	/*
7520 	 * Set HIPRI flag if message is priority.
7521 	 */
7522 	if (type >= QPCTL)
7523 		flg = MSG_HIPRI;
7524 	else
7525 		flg = MSG_BAND;
7526 
7527 	/*
7528 	 * First process PROTO or PCPROTO blocks, if any.
7529 	 */
7530 	if (mctlp != NULL && type != M_DATA) {
7531 		mblk_t *nbp;
7532 
7533 		*mctlp = bp;
7534 		while (bp->b_cont && bp->b_cont->b_datap->db_type != M_DATA)
7535 			bp = bp->b_cont;
7536 		nbp = bp->b_cont;
7537 		bp->b_cont = NULL;
7538 		bp = nbp;
7539 	}
7540 
7541 	if (bp && bp->b_datap->db_type != M_DATA) {
7542 		/*
7543 		 * More PROTO blocks in msg. Will only happen if mctlp is NULL.
7544 		 */
7545 		more |= MORECTL;
7546 		savemp = bp;
7547 		while (bp && bp->b_datap->db_type != M_DATA) {
7548 			savemptail = bp;
7549 			bp = bp->b_cont;
7550 		}
7551 		savemptail->b_cont = NULL;
7552 	}
7553 
7554 	/*
7555 	 * Now process DATA blocks, if any.
7556 	 */
7557 	if (uiop == NULL) {
7558 		/* Append data to tail of mctlp */
7559 
7560 		ASSERT(bp == NULL || !(bp->b_datap->db_flags & DBLK_UIOA));
7561 		if (mctlp != NULL) {
7562 			mblk_t **mpp = mctlp;
7563 
7564 			while (*mpp != NULL)
7565 				mpp = &((*mpp)->b_cont);
7566 			*mpp = bp;
7567 			bp = NULL;
7568 		}
7569 	} else if (bp && (bp->b_datap->db_flags & DBLK_UIOA)) {
7570 		/*
7571 		 * A uioa mblk_t chain, as uio processing has already
7572 		 * been done we simple skip over processing.
7573 		 */
7574 		bp = NULL;
7575 		pr = 0;
7576 
7577 	} else if (uiop->uio_resid >= 0 && bp) {
7578 		size_t oldresid = uiop->uio_resid;
7579 
7580 		/*
7581 		 * If a streams message is likely to consist
7582 		 * of many small mblks, it is pulled up into
7583 		 * one continuous chunk of memory.
7584 		 * see longer comment at top of page
7585 		 * by mblk_pull_len declaration.
7586 		 */
7587 
7588 		if (MBLKL(bp) < mblk_pull_len) {
7589 			(void) pullupmsg(bp, -1);
7590 		}
7591 
7592 		bp = struiocopyout(bp, uiop, &error);
7593 		if (error != 0) {
7594 			if (mctlp != NULL) {
7595 				freemsg(*mctlp);
7596 				*mctlp = NULL;
7597 			} else
7598 				freemsg(savemp);
7599 			mutex_enter(&stp->sd_lock);
7600 			/*
7601 			 * clear stream head hi pri flag based on
7602 			 * first message
7603 			 */
7604 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7605 				ASSERT(type == M_PCPROTO);
7606 				stp->sd_flag &= ~STRPRI;
7607 			}
7608 			more = 0;
7609 			goto getmout;
7610 		}
7611 		/*
7612 		 * (pr == 1) indicates a partial read.
7613 		 */
7614 		if (oldresid > uiop->uio_resid)
7615 			pr = 1;
7616 	}
7617 
7618 	if (bp) {			/* more data blocks in msg */
7619 		more |= MOREDATA;
7620 		if (savemp)
7621 			savemptail->b_cont = bp;
7622 		else
7623 			savemp = bp;
7624 	}
7625 
7626 	mutex_enter(&stp->sd_lock);
7627 	if (savemp) {
7628 		if (flags & (MSG_IPEEK|MSG_DISCARDTAIL)) {
7629 			/*
7630 			 * When MSG_DISCARDTAIL is set or
7631 			 * when peeking discard any tail. When peeking this
7632 			 * is the tail of the dup that was copied out - the
7633 			 * message has already been putback on the queue.
7634 			 * Return MOREDATA to the caller even though the data
7635 			 * is discarded. This is used by sockets (to
7636 			 * set MSG_TRUNC).
7637 			 */
7638 			freemsg(savemp);
7639 			if (!(flags & MSG_IPEEK) && (type >= QPCTL)) {
7640 				ASSERT(type == M_PCPROTO);
7641 				stp->sd_flag &= ~STRPRI;
7642 			}
7643 		} else if (pr && (savemp->b_datap->db_type == M_DATA) &&
7644 		    msgnodata(savemp)) {
7645 			/*
7646 			 * Avoid queuing a zero-length tail part of
7647 			 * a message. pr=1 indicates that we read some of
7648 			 * the message.
7649 			 */
7650 			freemsg(savemp);
7651 			more &= ~MOREDATA;
7652 			if (type >= QPCTL) {
7653 				ASSERT(type == M_PCPROTO);
7654 				stp->sd_flag &= ~STRPRI;
7655 			}
7656 		} else {
7657 			savemp->b_band = pri;
7658 			/*
7659 			 * If the first message was HIPRI and the one we're
7660 			 * putting back isn't, then clear STRPRI, otherwise
7661 			 * set STRPRI again.  Note that we must set STRPRI
7662 			 * again since the flush logic in strrput_nondata()
7663 			 * may have cleared it while we had sd_lock dropped.
7664 			 */
7665 
7666 			ASSERT(!(savemp->b_datap->db_flags & DBLK_UIOA));
7667 			if (type >= QPCTL) {
7668 				ASSERT(type == M_PCPROTO);
7669 				if (queclass(savemp) < QPCTL)
7670 					stp->sd_flag &= ~STRPRI;
7671 				else
7672 					stp->sd_flag |= STRPRI;
7673 			} else if (queclass(savemp) >= QPCTL) {
7674 				/*
7675 				 * The first message was not a HIPRI message,
7676 				 * but the one we are about to putback is.
7677 				 * For simplicitly, we do not allow for HIPRI
7678 				 * messages to be embedded in the message
7679 				 * body, so just force it to same type as
7680 				 * first message.
7681 				 */
7682 				ASSERT(type == M_DATA || type == M_PROTO);
7683 				ASSERT(savemp->b_datap->db_type == M_PCPROTO);
7684 				savemp->b_datap->db_type = type;
7685 			}
7686 			if (mark != 0) {
7687 				if ((mark & _LASTMARK) &&
7688 				    (stp->sd_mark == NULL)) {
7689 					/*
7690 					 * If another marked message arrived
7691 					 * while sd_lock was not held sd_mark
7692 					 * would be non-NULL.
7693 					 */
7694 					stp->sd_mark = savemp;
7695 				}
7696 				savemp->b_flag |= mark & ~_LASTMARK;
7697 			}
7698 			putback(stp, q, savemp, pri);
7699 		}
7700 	} else if (!(flags & MSG_IPEEK)) {
7701 		/*
7702 		 * The complete message was consumed.
7703 		 *
7704 		 * If another M_PCPROTO arrived while sd_lock was not held
7705 		 * it would have been discarded since STRPRI was still set.
7706 		 *
7707 		 * Move the MSG*MARKNEXT information
7708 		 * to the stream head just in case
7709 		 * the read queue becomes empty.
7710 		 * clear stream head hi pri flag based on
7711 		 * first message
7712 		 *
7713 		 * If the stream head was at the mark
7714 		 * (STRATMARK) before we dropped sd_lock above
7715 		 * and some data was consumed then we have
7716 		 * moved past the mark thus STRATMARK is
7717 		 * cleared. However, if a message arrived in
7718 		 * strrput during the copyout above causing
7719 		 * STRATMARK to be set we can not clear that
7720 		 * flag.
7721 		 * XXX A "perimeter" would help by single-threading strrput,
7722 		 * strread, strgetmsg and kstrgetmsg.
7723 		 */
7724 		if (type >= QPCTL) {
7725 			ASSERT(type == M_PCPROTO);
7726 			stp->sd_flag &= ~STRPRI;
7727 		}
7728 		if (mark & (MSGMARKNEXT|MSGNOTMARKNEXT|MSGMARK)) {
7729 			if (mark & MSGMARKNEXT) {
7730 				stp->sd_flag &= ~STRNOTATMARK;
7731 				stp->sd_flag |= STRATMARK;
7732 			} else if (mark & MSGNOTMARKNEXT) {
7733 				stp->sd_flag &= ~STRATMARK;
7734 				stp->sd_flag |= STRNOTATMARK;
7735 			} else {
7736 				stp->sd_flag &= ~(STRATMARK|STRNOTATMARK);
7737 			}
7738 		} else if (pr && (old_sd_flag & STRATMARK)) {
7739 			stp->sd_flag &= ~STRATMARK;
7740 		}
7741 	}
7742 
7743 	*flagsp = flg;
7744 	*prip = pri;
7745 
7746 	/*
7747 	 * Getmsg cleanup processing - if the state of the queue has changed
7748 	 * some signals may need to be sent and/or poll awakened.
7749 	 */
7750 getmout:
7751 	qbackenable(q, pri);
7752 
7753 	/*
7754 	 * We dropped the stream head lock above. Send all M_SIG messages
7755 	 * before processing stream head for SIGPOLL messages.
7756 	 */
7757 	ASSERT(MUTEX_HELD(&stp->sd_lock));
7758 	while ((bp = q->q_first) != NULL &&
7759 	    (bp->b_datap->db_type == M_SIG)) {
7760 		/*
7761 		 * sd_lock is held so the content of the read queue can not
7762 		 * change.
7763 		 */
7764 		bp = getq(q);
7765 		ASSERT(bp != NULL && bp->b_datap->db_type == M_SIG);
7766 
7767 		strsignal_nolock(stp, *bp->b_rptr, (int32_t)bp->b_band);
7768 		mutex_exit(&stp->sd_lock);
7769 		freemsg(bp);
7770 		if (STREAM_NEEDSERVICE(stp))
7771 			stream_runservice(stp);
7772 		mutex_enter(&stp->sd_lock);
7773 	}
7774 
7775 	/*
7776 	 * stream head cannot change while we make the determination
7777 	 * whether or not to send a signal. Drop the flag to allow strrput
7778 	 * to send firstmsgsigs again.
7779 	 */
7780 	stp->sd_flag &= ~STRGETINPROG;
7781 
7782 	/*
7783 	 * If the type of message at the front of the queue changed
7784 	 * due to the receive the appropriate signals and pollwakeup events
7785 	 * are generated. The type of changes are:
7786 	 *	Processed a hipri message, q_first is not hipri.
7787 	 *	Processed a band X message, and q_first is band Y.
7788 	 * The generated signals and pollwakeups are identical to what
7789 	 * strrput() generates should the message that is now on q_first
7790 	 * arrive to an empty read queue.
7791 	 *
7792 	 * Note: only strrput will send a signal for a hipri message.
7793 	 */
7794 	if ((bp = q->q_first) != NULL && !(stp->sd_flag & STRPRI)) {
7795 		strsigset_t signals = 0;
7796 		strpollset_t pollwakeups = 0;
7797 
7798 		if (flg & MSG_HIPRI) {
7799 			/*
7800 			 * Removed a hipri message. Regular data at
7801 			 * the front of  the queue.
7802 			 */
7803 			if (bp->b_band == 0) {
7804 				signals = S_INPUT | S_RDNORM;
7805 				pollwakeups = POLLIN | POLLRDNORM;
7806 			} else {
7807 				signals = S_INPUT | S_RDBAND;
7808 				pollwakeups = POLLIN | POLLRDBAND;
7809 			}
7810 		} else if (pri != bp->b_band) {
7811 			/*
7812 			 * The band is different for the new q_first.
7813 			 */
7814 			if (bp->b_band == 0) {
7815 				signals = S_RDNORM;
7816 				pollwakeups = POLLIN | POLLRDNORM;
7817 			} else {
7818 				signals = S_RDBAND;
7819 				pollwakeups = POLLIN | POLLRDBAND;
7820 			}
7821 		}
7822 
7823 		if (pollwakeups != 0) {
7824 			if (pollwakeups == (POLLIN | POLLRDNORM)) {
7825 				if (!(stp->sd_rput_opt & SR_POLLIN))
7826 					goto no_pollwake;
7827 				stp->sd_rput_opt &= ~SR_POLLIN;
7828 			}
7829 			mutex_exit(&stp->sd_lock);
7830 			pollwakeup(&stp->sd_pollist, pollwakeups);
7831 			mutex_enter(&stp->sd_lock);
7832 		}
7833 no_pollwake:
7834 
7835 		if (stp->sd_sigflags & signals)
7836 			strsendsig(stp->sd_siglist, signals, bp->b_band, 0);
7837 	}
7838 	mutex_exit(&stp->sd_lock);
7839 
7840 	rvp->r_val1 = more;
7841 	return (error);
7842 #undef	_LASTMARK
7843 }
7844 
7845 /*
7846  * Put a message downstream.
7847  *
7848  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
7849  */
7850 int
7851 strputmsg(
7852 	struct vnode *vp,
7853 	struct strbuf *mctl,
7854 	struct strbuf *mdata,
7855 	unsigned char pri,
7856 	int flag,
7857 	int fmode)
7858 {
7859 	struct stdata *stp;
7860 	queue_t *wqp;
7861 	mblk_t *mp;
7862 	ssize_t msgsize;
7863 	ssize_t rmin, rmax;
7864 	int error;
7865 	struct uio uios;
7866 	struct uio *uiop = &uios;
7867 	struct iovec iovs;
7868 	int xpg4 = 0;
7869 
7870 	ASSERT(vp->v_stream);
7871 	stp = vp->v_stream;
7872 	wqp = stp->sd_wrq;
7873 
7874 	/*
7875 	 * If it is an XPG4 application, we need to send
7876 	 * SIGPIPE below
7877 	 */
7878 
7879 	xpg4 = (flag & MSG_XPG4) ? 1 : 0;
7880 	flag &= ~MSG_XPG4;
7881 
7882 	if (audit_active)
7883 		audit_strputmsg(vp, mctl, mdata, pri, flag, fmode);
7884 
7885 	mutex_enter(&stp->sd_lock);
7886 
7887 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
7888 		mutex_exit(&stp->sd_lock);
7889 		return (error);
7890 	}
7891 
7892 	if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
7893 		error = strwriteable(stp, B_FALSE, xpg4);
7894 		if (error != 0) {
7895 			mutex_exit(&stp->sd_lock);
7896 			return (error);
7897 		}
7898 	}
7899 
7900 	mutex_exit(&stp->sd_lock);
7901 
7902 	/*
7903 	 * Check for legal flag value.
7904 	 */
7905 	switch (flag) {
7906 	case MSG_HIPRI:
7907 		if ((mctl->len < 0) || (pri != 0))
7908 			return (EINVAL);
7909 		break;
7910 	case MSG_BAND:
7911 		break;
7912 
7913 	default:
7914 		return (EINVAL);
7915 	}
7916 
7917 	TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_IN,
7918 	    "strputmsg in:stp %p", stp);
7919 
7920 	/* get these values from those cached in the stream head */
7921 	rmin = stp->sd_qn_minpsz;
7922 	rmax = stp->sd_qn_maxpsz;
7923 
7924 	/*
7925 	 * Make sure ctl and data sizes together fall within the
7926 	 * limits of the max and min receive packet sizes and do
7927 	 * not exceed system limit.
7928 	 */
7929 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
7930 	if (rmax == 0) {
7931 		return (ERANGE);
7932 	}
7933 	/*
7934 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
7935 	 * Needed to prevent partial failures in the strmakedata loop.
7936 	 */
7937 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
7938 		rmax = stp->sd_maxblk;
7939 
7940 	if ((msgsize = mdata->len) < 0) {
7941 		msgsize = 0;
7942 		rmin = 0;	/* no range check for NULL data part */
7943 	}
7944 	if ((msgsize < rmin) ||
7945 	    ((msgsize > rmax) && (rmax != INFPSZ)) ||
7946 	    (mctl->len > strctlsz)) {
7947 		return (ERANGE);
7948 	}
7949 
7950 	/*
7951 	 * Setup uio and iov for data part
7952 	 */
7953 	iovs.iov_base = mdata->buf;
7954 	iovs.iov_len = msgsize;
7955 	uios.uio_iov = &iovs;
7956 	uios.uio_iovcnt = 1;
7957 	uios.uio_loffset = 0;
7958 	uios.uio_segflg = UIO_USERSPACE;
7959 	uios.uio_fmode = fmode;
7960 	uios.uio_extflg = UIO_COPY_DEFAULT;
7961 	uios.uio_resid = msgsize;
7962 	uios.uio_offset = 0;
7963 
7964 	/* Ignore flow control in strput for HIPRI */
7965 	if (flag & MSG_HIPRI)
7966 		flag |= MSG_IGNFLOW;
7967 
7968 	for (;;) {
7969 		int done = 0;
7970 
7971 		/*
7972 		 * strput will always free the ctl mblk - even when strput
7973 		 * fails.
7974 		 */
7975 		if ((error = strmakectl(mctl, flag, fmode, &mp)) != 0) {
7976 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
7977 			    "strputmsg out:stp %p out %d error %d",
7978 			    stp, 1, error);
7979 			return (error);
7980 		}
7981 		/*
7982 		 * Verify that the whole message can be transferred by
7983 		 * strput.
7984 		 */
7985 		ASSERT(stp->sd_maxblk == INFPSZ ||
7986 		    stp->sd_maxblk >= mdata->len);
7987 
7988 		msgsize = mdata->len;
7989 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
7990 		mdata->len = msgsize;
7991 
7992 		if (error == 0)
7993 			break;
7994 
7995 		if (error != EWOULDBLOCK)
7996 			goto out;
7997 
7998 		mutex_enter(&stp->sd_lock);
7999 		/*
8000 		 * Check for a missed wakeup.
8001 		 * Needed since strput did not hold sd_lock across
8002 		 * the canputnext.
8003 		 */
8004 		if (bcanputnext(wqp, pri)) {
8005 			/* Try again */
8006 			mutex_exit(&stp->sd_lock);
8007 			continue;
8008 		}
8009 		TRACE_2(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAIT,
8010 		    "strputmsg wait:stp %p waits pri %d", stp, pri);
8011 		if (((error = strwaitq(stp, WRITEWAIT, (ssize_t)0, fmode, -1,
8012 		    &done)) != 0) || done) {
8013 			mutex_exit(&stp->sd_lock);
8014 			TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
8015 			    "strputmsg out:q %p out %d error %d",
8016 			    stp, 0, error);
8017 			return (error);
8018 		}
8019 		TRACE_1(TR_FAC_STREAMS_FR, TR_STRPUTMSG_WAKE,
8020 		    "strputmsg wake:stp %p wakes", stp);
8021 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
8022 			mutex_exit(&stp->sd_lock);
8023 			return (error);
8024 		}
8025 		mutex_exit(&stp->sd_lock);
8026 	}
8027 out:
8028 	/*
8029 	 * For historic reasons, applications expect EAGAIN
8030 	 * when data mblk could not be allocated. so change
8031 	 * ENOMEM back to EAGAIN
8032 	 */
8033 	if (error == ENOMEM)
8034 		error = EAGAIN;
8035 	TRACE_3(TR_FAC_STREAMS_FR, TR_STRPUTMSG_OUT,
8036 	    "strputmsg out:stp %p out %d error %d", stp, 2, error);
8037 	return (error);
8038 }
8039 
8040 /*
8041  * Put a message downstream.
8042  * Can send only an M_PROTO/M_PCPROTO by passing in a NULL uiop.
8043  * The fmode flag (NDELAY, NONBLOCK) is the or of the flags in the uio
8044  * and the fmode parameter.
8045  *
8046  * This routine handles the consolidation private flags:
8047  *	MSG_IGNERROR	Ignore any stream head error except STPLEX.
8048  *	MSG_HOLDSIG	Hold signals while waiting for data.
8049  *	MSG_IGNFLOW	Don't check streams flow control.
8050  *
8051  * NOTE: strputmsg and kstrputmsg have much of the logic in common.
8052  */
8053 int
8054 kstrputmsg(
8055 	struct vnode *vp,
8056 	mblk_t *mctl,
8057 	struct uio *uiop,
8058 	ssize_t msgsize,
8059 	unsigned char pri,
8060 	int flag,
8061 	int fmode)
8062 {
8063 	struct stdata *stp;
8064 	queue_t *wqp;
8065 	ssize_t rmin, rmax;
8066 	int error;
8067 
8068 	ASSERT(vp->v_stream);
8069 	stp = vp->v_stream;
8070 	wqp = stp->sd_wrq;
8071 	if (audit_active)
8072 		audit_strputmsg(vp, NULL, NULL, pri, flag, fmode);
8073 	if (mctl == NULL)
8074 		return (EINVAL);
8075 
8076 	mutex_enter(&stp->sd_lock);
8077 
8078 	if ((error = i_straccess(stp, JCWRITE)) != 0) {
8079 		mutex_exit(&stp->sd_lock);
8080 		freemsg(mctl);
8081 		return (error);
8082 	}
8083 
8084 	if ((stp->sd_flag & STPLEX) || !(flag & MSG_IGNERROR)) {
8085 		if (stp->sd_flag & (STWRERR|STRHUP|STPLEX)) {
8086 			error = strwriteable(stp, B_FALSE, B_TRUE);
8087 			if (error != 0) {
8088 				mutex_exit(&stp->sd_lock);
8089 				freemsg(mctl);
8090 				return (error);
8091 			}
8092 		}
8093 	}
8094 
8095 	mutex_exit(&stp->sd_lock);
8096 
8097 	/*
8098 	 * Check for legal flag value.
8099 	 */
8100 	switch (flag & (MSG_HIPRI|MSG_BAND|MSG_ANY)) {
8101 	case MSG_HIPRI:
8102 		if (pri != 0) {
8103 			freemsg(mctl);
8104 			return (EINVAL);
8105 		}
8106 		break;
8107 	case MSG_BAND:
8108 		break;
8109 	default:
8110 		freemsg(mctl);
8111 		return (EINVAL);
8112 	}
8113 
8114 	TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_IN,
8115 	    "kstrputmsg in:stp %p", stp);
8116 
8117 	/* get these values from those cached in the stream head */
8118 	rmin = stp->sd_qn_minpsz;
8119 	rmax = stp->sd_qn_maxpsz;
8120 
8121 	/*
8122 	 * Make sure ctl and data sizes together fall within the
8123 	 * limits of the max and min receive packet sizes and do
8124 	 * not exceed system limit.
8125 	 */
8126 	ASSERT((rmax >= 0) || (rmax == INFPSZ));
8127 	if (rmax == 0) {
8128 		freemsg(mctl);
8129 		return (ERANGE);
8130 	}
8131 	/*
8132 	 * Use the MAXIMUM of sd_maxblk and q_maxpsz.
8133 	 * Needed to prevent partial failures in the strmakedata loop.
8134 	 */
8135 	if (stp->sd_maxblk != INFPSZ && rmax != INFPSZ && rmax < stp->sd_maxblk)
8136 		rmax = stp->sd_maxblk;
8137 
8138 	if (uiop == NULL) {
8139 		msgsize = -1;
8140 		rmin = -1;	/* no range check for NULL data part */
8141 	} else {
8142 		/* Use uio flags as well as the fmode parameter flags */
8143 		fmode |= uiop->uio_fmode;
8144 
8145 		if ((msgsize < rmin) ||
8146 		    ((msgsize > rmax) && (rmax != INFPSZ))) {
8147 			freemsg(mctl);
8148 			return (ERANGE);
8149 		}
8150 	}
8151 
8152 	/* Ignore flow control in strput for HIPRI */
8153 	if (flag & MSG_HIPRI)
8154 		flag |= MSG_IGNFLOW;
8155 
8156 	for (;;) {
8157 		int done = 0;
8158 		int waitflag;
8159 		mblk_t *mp;
8160 
8161 		/*
8162 		 * strput will always free the ctl mblk - even when strput
8163 		 * fails. If MSG_IGNFLOW is set then any error returned
8164 		 * will cause us to break the loop, so we don't need a copy
8165 		 * of the message. If MSG_IGNFLOW is not set, then we can
8166 		 * get hit by flow control and be forced to try again. In
8167 		 * this case we need to have a copy of the message. We
8168 		 * do this using copymsg since the message may get modified
8169 		 * by something below us.
8170 		 *
8171 		 * We've observed that many TPI providers do not check db_ref
8172 		 * on the control messages but blindly reuse them for the
8173 		 * T_OK_ACK/T_ERROR_ACK. Thus using copymsg is more
8174 		 * friendly to such providers than using dupmsg. Also, note
8175 		 * that sockfs uses MSG_IGNFLOW for all TPI control messages.
8176 		 * Only data messages are subject to flow control, hence
8177 		 * subject to this copymsg.
8178 		 */
8179 		if (flag & MSG_IGNFLOW) {
8180 			mp = mctl;
8181 			mctl = NULL;
8182 		} else {
8183 			do {
8184 				/*
8185 				 * If a message has a free pointer, the message
8186 				 * must be dupmsg to maintain this pointer.
8187 				 * Code using this facility must be sure
8188 				 * that modules below will not change the
8189 				 * contents of the dblk without checking db_ref
8190 				 * first. If db_ref is > 1, then the module
8191 				 * needs to do a copymsg first. Otherwise,
8192 				 * the contents of the dblk may become
8193 				 * inconsistent because the freesmg/freeb below
8194 				 * may end up calling atomic_add_32_nv.
8195 				 * The atomic_add_32_nv in freeb (accessing
8196 				 * all of db_ref, db_type, db_flags, and
8197 				 * db_struioflag) does not prevent other threads
8198 				 * from concurrently trying to modify e.g.
8199 				 * db_type.
8200 				 */
8201 				if (mctl->b_datap->db_frtnp != NULL)
8202 					mp = dupmsg(mctl);
8203 				else
8204 					mp = copymsg(mctl);
8205 
8206 				if (mp != NULL)
8207 					break;
8208 
8209 				error = strwaitbuf(msgdsize(mctl), BPRI_MED);
8210 				if (error) {
8211 					freemsg(mctl);
8212 					return (error);
8213 				}
8214 			} while (mp == NULL);
8215 		}
8216 		/*
8217 		 * Verify that all of msgsize can be transferred by
8218 		 * strput.
8219 		 */
8220 		ASSERT(stp->sd_maxblk == INFPSZ || stp->sd_maxblk >= msgsize);
8221 		error = strput(stp, mp, uiop, &msgsize, 0, pri, flag);
8222 		if (error == 0)
8223 			break;
8224 
8225 		if (error != EWOULDBLOCK)
8226 			goto out;
8227 
8228 		/*
8229 		 * IF MSG_IGNFLOW is set we should have broken out of loop
8230 		 * above.
8231 		 */
8232 		ASSERT(!(flag & MSG_IGNFLOW));
8233 		mutex_enter(&stp->sd_lock);
8234 		/*
8235 		 * Check for a missed wakeup.
8236 		 * Needed since strput did not hold sd_lock across
8237 		 * the canputnext.
8238 		 */
8239 		if (bcanputnext(wqp, pri)) {
8240 			/* Try again */
8241 			mutex_exit(&stp->sd_lock);
8242 			continue;
8243 		}
8244 		TRACE_2(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAIT,
8245 		    "kstrputmsg wait:stp %p waits pri %d", stp, pri);
8246 
8247 		waitflag = WRITEWAIT;
8248 		if (flag & (MSG_HOLDSIG|MSG_IGNERROR)) {
8249 			if (flag & MSG_HOLDSIG)
8250 				waitflag |= STR_NOSIG;
8251 			if (flag & MSG_IGNERROR)
8252 				waitflag |= STR_NOERROR;
8253 		}
8254 		if (((error = strwaitq(stp, waitflag,
8255 		    (ssize_t)0, fmode, -1, &done)) != 0) || done) {
8256 			mutex_exit(&stp->sd_lock);
8257 			TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8258 			    "kstrputmsg out:stp %p out %d error %d",
8259 			    stp, 0, error);
8260 			freemsg(mctl);
8261 			return (error);
8262 		}
8263 		TRACE_1(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_WAKE,
8264 		    "kstrputmsg wake:stp %p wakes", stp);
8265 		if ((error = i_straccess(stp, JCWRITE)) != 0) {
8266 			mutex_exit(&stp->sd_lock);
8267 			freemsg(mctl);
8268 			return (error);
8269 		}
8270 		mutex_exit(&stp->sd_lock);
8271 	}
8272 out:
8273 	freemsg(mctl);
8274 	/*
8275 	 * For historic reasons, applications expect EAGAIN
8276 	 * when data mblk could not be allocated. so change
8277 	 * ENOMEM back to EAGAIN
8278 	 */
8279 	if (error == ENOMEM)
8280 		error = EAGAIN;
8281 	TRACE_3(TR_FAC_STREAMS_FR, TR_KSTRPUTMSG_OUT,
8282 	    "kstrputmsg out:stp %p out %d error %d", stp, 2, error);
8283 	return (error);
8284 }
8285 
8286 /*
8287  * Determines whether the necessary conditions are set on a stream
8288  * for it to be readable, writeable, or have exceptions.
8289  *
8290  * strpoll handles the consolidation private events:
8291  *	POLLNOERR	Do not return POLLERR even if there are stream
8292  *			head errors.
8293  *			Used by sockfs.
8294  *	POLLRDDATA	Do not return POLLIN unless at least one message on
8295  *			the queue contains one or more M_DATA mblks. Thus
8296  *			when this flag is set a queue with only
8297  *			M_PROTO/M_PCPROTO mblks does not return POLLIN.
8298  *			Used by sockfs to ignore T_EXDATA_IND messages.
8299  *
8300  * Note: POLLRDDATA assumes that synch streams only return messages with
8301  * an M_DATA attached (i.e. not messages consisting of only
8302  * an M_PROTO/M_PCPROTO part).
8303  */
8304 int
8305 strpoll(
8306 	struct stdata *stp,
8307 	short events_arg,
8308 	int anyyet,
8309 	short *reventsp,
8310 	struct pollhead **phpp)
8311 {
8312 	int events = (ushort_t)events_arg;
8313 	int retevents = 0;
8314 	mblk_t *mp;
8315 	qband_t *qbp;
8316 	long sd_flags = stp->sd_flag;
8317 	int headlocked = 0;
8318 
8319 	/*
8320 	 * For performance, a single 'if' tests for most possible edge
8321 	 * conditions in one shot
8322 	 */
8323 	if (sd_flags & (STPLEX | STRDERR | STWRERR)) {
8324 		if (sd_flags & STPLEX) {
8325 			*reventsp = POLLNVAL;
8326 			return (EINVAL);
8327 		}
8328 		if (((events & (POLLIN | POLLRDNORM | POLLRDBAND | POLLPRI)) &&
8329 		    (sd_flags & STRDERR)) ||
8330 		    ((events & (POLLOUT | POLLWRNORM | POLLWRBAND)) &&
8331 		    (sd_flags & STWRERR))) {
8332 			if (!(events & POLLNOERR)) {
8333 				*reventsp = POLLERR;
8334 				return (0);
8335 			}
8336 		}
8337 	}
8338 	if (sd_flags & STRHUP) {
8339 		retevents |= POLLHUP;
8340 	} else if (events & (POLLWRNORM | POLLWRBAND)) {
8341 		queue_t *tq;
8342 		queue_t	*qp = stp->sd_wrq;
8343 
8344 		claimstr(qp);
8345 		/* Find next module forward that has a service procedure */
8346 		tq = qp->q_next->q_nfsrv;
8347 		ASSERT(tq != NULL);
8348 
8349 		polllock(&stp->sd_pollist, QLOCK(tq));
8350 		if (events & POLLWRNORM) {
8351 			queue_t *sqp;
8352 
8353 			if (tq->q_flag & QFULL)
8354 				/* ensure backq svc procedure runs */
8355 				tq->q_flag |= QWANTW;
8356 			else if ((sqp = stp->sd_struiowrq) != NULL) {
8357 				/* Check sync stream barrier write q */
8358 				mutex_exit(QLOCK(tq));
8359 				polllock(&stp->sd_pollist, QLOCK(sqp));
8360 				if (sqp->q_flag & QFULL)
8361 					/* ensure pollwakeup() is done */
8362 					sqp->q_flag |= QWANTWSYNC;
8363 				else
8364 					retevents |= POLLOUT;
8365 				/* More write events to process ??? */
8366 				if (! (events & POLLWRBAND)) {
8367 					mutex_exit(QLOCK(sqp));
8368 					releasestr(qp);
8369 					goto chkrd;
8370 				}
8371 				mutex_exit(QLOCK(sqp));
8372 				polllock(&stp->sd_pollist, QLOCK(tq));
8373 			} else
8374 				retevents |= POLLOUT;
8375 		}
8376 		if (events & POLLWRBAND) {
8377 			qbp = tq->q_bandp;
8378 			if (qbp) {
8379 				while (qbp) {
8380 					if (qbp->qb_flag & QB_FULL)
8381 						qbp->qb_flag |= QB_WANTW;
8382 					else
8383 						retevents |= POLLWRBAND;
8384 					qbp = qbp->qb_next;
8385 				}
8386 			} else {
8387 				retevents |= POLLWRBAND;
8388 			}
8389 		}
8390 		mutex_exit(QLOCK(tq));
8391 		releasestr(qp);
8392 	}
8393 chkrd:
8394 	if (sd_flags & STRPRI) {
8395 		retevents |= (events & POLLPRI);
8396 	} else if (events & (POLLRDNORM | POLLRDBAND | POLLIN)) {
8397 		queue_t	*qp = _RD(stp->sd_wrq);
8398 		int normevents = (events & (POLLIN | POLLRDNORM));
8399 
8400 		/*
8401 		 * Note: Need to do polllock() here since ps_lock may be
8402 		 * held. See bug 4191544.
8403 		 */
8404 		polllock(&stp->sd_pollist, &stp->sd_lock);
8405 		headlocked = 1;
8406 		mp = qp->q_first;
8407 		while (mp) {
8408 			/*
8409 			 * For POLLRDDATA we scan b_cont and b_next until we
8410 			 * find an M_DATA.
8411 			 */
8412 			if ((events & POLLRDDATA) &&
8413 			    mp->b_datap->db_type != M_DATA) {
8414 				mblk_t *nmp = mp->b_cont;
8415 
8416 				while (nmp != NULL &&
8417 				    nmp->b_datap->db_type != M_DATA)
8418 					nmp = nmp->b_cont;
8419 				if (nmp == NULL) {
8420 					mp = mp->b_next;
8421 					continue;
8422 				}
8423 			}
8424 			if (mp->b_band == 0)
8425 				retevents |= normevents;
8426 			else
8427 				retevents |= (events & (POLLIN | POLLRDBAND));
8428 			break;
8429 		}
8430 		if (! (retevents & normevents) &&
8431 		    (stp->sd_wakeq & RSLEEP)) {
8432 			/*
8433 			 * Sync stream barrier read queue has data.
8434 			 */
8435 			retevents |= normevents;
8436 		}
8437 		/* Treat eof as normal data */
8438 		if (sd_flags & STREOF)
8439 			retevents |= normevents;
8440 	}
8441 
8442 	*reventsp = (short)retevents;
8443 	if (retevents) {
8444 		if (headlocked)
8445 			mutex_exit(&stp->sd_lock);
8446 		return (0);
8447 	}
8448 
8449 	/*
8450 	 * If poll() has not found any events yet, set up event cell
8451 	 * to wake up the poll if a requested event occurs on this
8452 	 * stream.  Check for collisions with outstanding poll requests.
8453 	 */
8454 	if (!anyyet) {
8455 		*phpp = &stp->sd_pollist;
8456 		if (headlocked == 0) {
8457 			polllock(&stp->sd_pollist, &stp->sd_lock);
8458 			headlocked = 1;
8459 		}
8460 		stp->sd_rput_opt |= SR_POLLIN;
8461 	}
8462 	if (headlocked)
8463 		mutex_exit(&stp->sd_lock);
8464 	return (0);
8465 }
8466 
8467 /*
8468  * The purpose of putback() is to assure sleeping polls/reads
8469  * are awakened when there are no new messages arriving at the,
8470  * stream head, and a message is placed back on the read queue.
8471  *
8472  * sd_lock must be held when messages are placed back on stream
8473  * head.  (getq() holds sd_lock when it removes messages from
8474  * the queue)
8475  */
8476 
8477 static void
8478 putback(struct stdata *stp, queue_t *q, mblk_t *bp, int band)
8479 {
8480 	mblk_t	*qfirst;
8481 	ASSERT(MUTEX_HELD(&stp->sd_lock));
8482 
8483 	/*
8484 	 * As a result of lock-step ordering around q_lock and sd_lock,
8485 	 * it's possible for function calls like putnext() and
8486 	 * canputnext() to get an inaccurate picture of how much
8487 	 * data is really being processed at the stream head.
8488 	 * We only consolidate with existing messages on the queue
8489 	 * if the length of the message we want to put back is smaller
8490 	 * than the queue hiwater mark.
8491 	 */
8492 	if ((stp->sd_rput_opt & SR_CONSOL_DATA) &&
8493 	    (DB_TYPE(bp) == M_DATA) && ((qfirst = q->q_first) != NULL) &&
8494 	    (DB_TYPE(qfirst) == M_DATA) &&
8495 	    ((qfirst->b_flag & (MSGMARK|MSGDELIM)) == 0) &&
8496 	    ((bp->b_flag & (MSGMARK|MSGDELIM|MSGMARKNEXT)) == 0) &&
8497 	    (mp_cont_len(bp, NULL) < q->q_hiwat)) {
8498 		/*
8499 		 * We use the same logic as defined in strrput()
8500 		 * but in reverse as we are putting back onto the
8501 		 * queue and want to retain byte ordering.
8502 		 * Consolidate M_DATA messages with M_DATA ONLY.
8503 		 * strrput() allows the consolidation of M_DATA onto
8504 		 * M_PROTO | M_PCPROTO but not the other way round.
8505 		 *
8506 		 * The consolidation does not take place if the message
8507 		 * we are returning to the queue is marked with either
8508 		 * of the marks or the delim flag or if q_first
8509 		 * is marked with MSGMARK. The MSGMARK check is needed to
8510 		 * handle the odd semantics of MSGMARK where essentially
8511 		 * the whole message is to be treated as marked.
8512 		 * Carry any MSGMARKNEXT and MSGNOTMARKNEXT from q_first
8513 		 * to the front of the b_cont chain.
8514 		 */
8515 		rmvq_noenab(q, qfirst);
8516 
8517 		/*
8518 		 * The first message in the b_cont list
8519 		 * tracks MSGMARKNEXT and MSGNOTMARKNEXT.
8520 		 * We need to handle the case where we
8521 		 * are appending:
8522 		 *
8523 		 * 1) a MSGMARKNEXT to a MSGNOTMARKNEXT.
8524 		 * 2) a MSGMARKNEXT to a plain message.
8525 		 * 3) a MSGNOTMARKNEXT to a plain message
8526 		 * 4) a MSGNOTMARKNEXT to a MSGNOTMARKNEXT
8527 		 *    message.
8528 		 *
8529 		 * Thus we never append a MSGMARKNEXT or
8530 		 * MSGNOTMARKNEXT to a MSGMARKNEXT message.
8531 		 */
8532 		if (qfirst->b_flag & MSGMARKNEXT) {
8533 			bp->b_flag |= MSGMARKNEXT;
8534 			bp->b_flag &= ~MSGNOTMARKNEXT;
8535 			qfirst->b_flag &= ~MSGMARKNEXT;
8536 		} else if (qfirst->b_flag & MSGNOTMARKNEXT) {
8537 			bp->b_flag |= MSGNOTMARKNEXT;
8538 			qfirst->b_flag &= ~MSGNOTMARKNEXT;
8539 		}
8540 
8541 		linkb(bp, qfirst);
8542 	}
8543 	(void) putbq(q, bp);
8544 
8545 	/*
8546 	 * A message may have come in when the sd_lock was dropped in the
8547 	 * calling routine. If this is the case and STR*ATMARK info was
8548 	 * received, need to move that from the stream head to the q_last
8549 	 * so that SIOCATMARK can return the proper value.
8550 	 */
8551 	if (stp->sd_flag & (STRATMARK | STRNOTATMARK)) {
8552 		unsigned short *flagp = &q->q_last->b_flag;
8553 		uint_t b_flag = (uint_t)*flagp;
8554 
8555 		if (stp->sd_flag & STRATMARK) {
8556 			b_flag &= ~MSGNOTMARKNEXT;
8557 			b_flag |= MSGMARKNEXT;
8558 			stp->sd_flag &= ~STRATMARK;
8559 		} else {
8560 			b_flag &= ~MSGMARKNEXT;
8561 			b_flag |= MSGNOTMARKNEXT;
8562 			stp->sd_flag &= ~STRNOTATMARK;
8563 		}
8564 		*flagp = (unsigned short) b_flag;
8565 	}
8566 
8567 #ifdef	DEBUG
8568 	/*
8569 	 * Make sure that the flags are not messed up.
8570 	 */
8571 	{
8572 		mblk_t *mp;
8573 		mp = q->q_last;
8574 		while (mp != NULL) {
8575 			ASSERT((mp->b_flag & (MSGMARKNEXT|MSGNOTMARKNEXT)) !=
8576 			    (MSGMARKNEXT|MSGNOTMARKNEXT));
8577 			mp = mp->b_cont;
8578 		}
8579 	}
8580 #endif
8581 	if (q->q_first == bp) {
8582 		short pollevents;
8583 
8584 		if (stp->sd_flag & RSLEEP) {
8585 			stp->sd_flag &= ~RSLEEP;
8586 			cv_broadcast(&q->q_wait);
8587 		}
8588 		if (stp->sd_flag & STRPRI) {
8589 			pollevents = POLLPRI;
8590 		} else {
8591 			if (band == 0) {
8592 				if (!(stp->sd_rput_opt & SR_POLLIN))
8593 					return;
8594 				stp->sd_rput_opt &= ~SR_POLLIN;
8595 				pollevents = POLLIN | POLLRDNORM;
8596 			} else {
8597 				pollevents = POLLIN | POLLRDBAND;
8598 			}
8599 		}
8600 		mutex_exit(&stp->sd_lock);
8601 		pollwakeup(&stp->sd_pollist, pollevents);
8602 		mutex_enter(&stp->sd_lock);
8603 	}
8604 }
8605 
8606 /*
8607  * Return the held vnode attached to the stream head of a
8608  * given queue
8609  * It is the responsibility of the calling routine to ensure
8610  * that the queue does not go away (e.g. pop).
8611  */
8612 vnode_t *
8613 strq2vp(queue_t *qp)
8614 {
8615 	vnode_t *vp;
8616 	vp = STREAM(qp)->sd_vnode;
8617 	ASSERT(vp != NULL);
8618 	VN_HOLD(vp);
8619 	return (vp);
8620 }
8621 
8622 /*
8623  * return the stream head write queue for the given vp
8624  * It is the responsibility of the calling routine to ensure
8625  * that the stream or vnode do not close.
8626  */
8627 queue_t *
8628 strvp2wq(vnode_t *vp)
8629 {
8630 	ASSERT(vp->v_stream != NULL);
8631 	return (vp->v_stream->sd_wrq);
8632 }
8633 
8634 /*
8635  * pollwakeup stream head
8636  * It is the responsibility of the calling routine to ensure
8637  * that the stream or vnode do not close.
8638  */
8639 void
8640 strpollwakeup(vnode_t *vp, short event)
8641 {
8642 	ASSERT(vp->v_stream);
8643 	pollwakeup(&vp->v_stream->sd_pollist, event);
8644 }
8645 
8646 /*
8647  * Mate the stream heads of two vnodes together. If the two vnodes are the
8648  * same, we just make the write-side point at the read-side -- otherwise,
8649  * we do a full mate.  Only works on vnodes associated with streams that are
8650  * still being built and thus have only a stream head.
8651  */
8652 void
8653 strmate(vnode_t *vp1, vnode_t *vp2)
8654 {
8655 	queue_t *wrq1 = strvp2wq(vp1);
8656 	queue_t *wrq2 = strvp2wq(vp2);
8657 
8658 	/*
8659 	 * Verify that there are no modules on the stream yet.  We also
8660 	 * rely on the stream head always having a service procedure to
8661 	 * avoid tweaking q_nfsrv.
8662 	 */
8663 	ASSERT(wrq1->q_next == NULL && wrq2->q_next == NULL);
8664 	ASSERT(wrq1->q_qinfo->qi_srvp != NULL);
8665 	ASSERT(wrq2->q_qinfo->qi_srvp != NULL);
8666 
8667 	/*
8668 	 * If the queues are the same, just twist; otherwise do a full mate.
8669 	 */
8670 	if (wrq1 == wrq2) {
8671 		wrq1->q_next = _RD(wrq1);
8672 	} else {
8673 		wrq1->q_next = _RD(wrq2);
8674 		wrq2->q_next = _RD(wrq1);
8675 		STREAM(wrq1)->sd_mate = STREAM(wrq2);
8676 		STREAM(wrq1)->sd_flag |= STRMATE;
8677 		STREAM(wrq2)->sd_mate = STREAM(wrq1);
8678 		STREAM(wrq2)->sd_flag |= STRMATE;
8679 	}
8680 }
8681 
8682 /*
8683  * XXX will go away when console is correctly fixed.
8684  * Clean up the console PIDS, from previous I_SETSIG,
8685  * called only for cnopen which never calls strclean().
8686  */
8687 void
8688 str_cn_clean(struct vnode *vp)
8689 {
8690 	strsig_t *ssp, *pssp, *tssp;
8691 	struct stdata *stp;
8692 	struct pid  *pidp;
8693 	int update = 0;
8694 
8695 	ASSERT(vp->v_stream);
8696 	stp = vp->v_stream;
8697 	pssp = NULL;
8698 	mutex_enter(&stp->sd_lock);
8699 	ssp = stp->sd_siglist;
8700 	while (ssp) {
8701 		mutex_enter(&pidlock);
8702 		pidp = ssp->ss_pidp;
8703 		/*
8704 		 * Get rid of PID if the proc is gone.
8705 		 */
8706 		if (pidp->pid_prinactive) {
8707 			tssp = ssp->ss_next;
8708 			if (pssp)
8709 				pssp->ss_next = tssp;
8710 			else
8711 				stp->sd_siglist = tssp;
8712 			ASSERT(pidp->pid_ref <= 1);
8713 			PID_RELE(ssp->ss_pidp);
8714 			mutex_exit(&pidlock);
8715 			kmem_free(ssp, sizeof (strsig_t));
8716 			update = 1;
8717 			ssp = tssp;
8718 			continue;
8719 		} else
8720 			mutex_exit(&pidlock);
8721 		pssp = ssp;
8722 		ssp = ssp->ss_next;
8723 	}
8724 	if (update) {
8725 		stp->sd_sigflags = 0;
8726 		for (ssp = stp->sd_siglist; ssp; ssp = ssp->ss_next)
8727 			stp->sd_sigflags |= ssp->ss_events;
8728 	}
8729 	mutex_exit(&stp->sd_lock);
8730 }
8731 
8732 /*
8733  * Return B_TRUE if there is data in the message, B_FALSE otherwise.
8734  */
8735 static boolean_t
8736 msghasdata(mblk_t *bp)
8737 {
8738 	for (; bp; bp = bp->b_cont)
8739 		if (bp->b_datap->db_type == M_DATA) {
8740 			ASSERT(bp->b_wptr >= bp->b_rptr);
8741 			if (bp->b_wptr > bp->b_rptr)
8742 				return (B_TRUE);
8743 		}
8744 	return (B_FALSE);
8745 }
8746 
8747 /*
8748  * Called on the first strget() of a sodirect/uioa enabled streamhead,
8749  * if any mblk_t(s) enqueued they must first be uioamove()d before uioa
8750  * can be enabled for the underlying transport's use.
8751  */
8752 void
8753 struioainit(queue_t *q, sodirect_t *sodp, uio_t *uiop)
8754 {
8755 	uioa_t	*uioap = (uioa_t *)uiop;
8756 	mblk_t	*bp = q->q_first;
8757 	mblk_t	*lbp = NULL;
8758 	mblk_t	*nbp, *wbp;
8759 	int	len;
8760 	int	error;
8761 
8762 	ASSERT(MUTEX_HELD(sodp->sod_lock));
8763 	ASSERT(&sodp->sod_uioa == uioap);
8764 
8765 	/*
8766 	 * Walk the b_next/b_prev doubly linked list of b_cont chain(s)
8767 	 * and schedule any M_DATA mblk_t's for uio asynchronous move.
8768 	 */
8769 	do {
8770 		/* Next mblk_t chain */
8771 		nbp = bp->b_next;
8772 		/* Walk the chain */
8773 		wbp = bp;
8774 		do {
8775 			if (wbp->b_datap->db_type != M_DATA) {
8776 				/* Not M_DATA, no more uioa */
8777 				goto nouioa;
8778 			}
8779 			if ((len = wbp->b_wptr - wbp->b_rptr) > 0) {
8780 				/* Have a M_DATA mblk_t with data */
8781 				if (len > uioap->uio_resid) {
8782 					/* Not enough uio sapce */
8783 					goto nouioa;
8784 				}
8785 				error = uioamove(wbp->b_rptr, len,
8786 				    UIO_READ, uioap);
8787 				if (!error) {
8788 					/* Scheduled, mark dblk_t as such */
8789 					wbp->b_datap->db_flags |= DBLK_UIOA;
8790 				} else {
8791 					/* Error of some sort, no more uioa */
8792 					uioap->uioa_state &= UIOA_CLR;
8793 					uioap->uioa_state |= UIOA_FINI;
8794 					return;
8795 				}
8796 			}
8797 			/* Save last wbp processed */
8798 			lbp = wbp;
8799 		} while ((wbp = wbp->b_cont) != NULL);
8800 	} while ((bp = nbp) != NULL);
8801 
8802 	return;
8803 
8804 nouioa:
8805 	/* No more uioa */
8806 	uioap->uioa_state &= UIOA_CLR;
8807 	uioap->uioa_state |= UIOA_FINI;
8808 
8809 	/*
8810 	 * If we processed 1 or more mblk_t(s) then we need to split the
8811 	 * current mblk_t chain in 2 so that all the uioamove()ed mblk_t(s)
8812 	 * are in the current chain and the rest are in the following new
8813 	 * chain.
8814 	 */
8815 	if (lbp != NULL) {
8816 		/* New end of current chain */
8817 		lbp->b_cont = NULL;
8818 
8819 		/* Insert new chain wbp after bp */
8820 		if ((wbp->b_next = nbp) != NULL)
8821 			nbp->b_prev = wbp;
8822 		else
8823 			q->q_last = wbp;
8824 		wbp->b_prev = bp;
8825 		bp->b_next = wbp;
8826 	}
8827 }
8828