xref: /illumos-gate/usr/src/uts/common/fs/fifofs/fifosubr.c (revision da6c28aaf62fa55f0fdb8004aa40f88f23bf53f0)
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 
23 /*
24  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
25  * Use is subject to license terms.
26  */
27 
28 #pragma ident	"%Z%%M%	%I%	%E% SMI"
29 
30 /*
31  * The routines defined in this file are supporting routines for FIFOFS
32  * file system type.
33  */
34 #include <sys/types.h>
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/debug.h>
38 #include <sys/errno.h>
39 #include <sys/time.h>
40 #include <sys/kmem.h>
41 #include <sys/inline.h>
42 #include <sys/file.h>
43 #include <sys/proc.h>
44 #include <sys/stat.h>
45 #include <sys/sysmacros.h>
46 #include <sys/var.h>
47 #include <sys/vfs.h>
48 #include <sys/vfs_opreg.h>
49 #include <sys/vnode.h>
50 #include <sys/mode.h>
51 #include <sys/signal.h>
52 #include <sys/user.h>
53 #include <sys/uio.h>
54 #include <sys/flock.h>
55 #include <sys/stream.h>
56 #include <sys/fs/fifonode.h>
57 #include <sys/strsubr.h>
58 #include <sys/stropts.h>
59 #include <sys/cmn_err.h>
60 #include <fs/fs_subr.h>
61 #include <sys/ddi.h>
62 
63 
64 #if FIFODEBUG
65 int Fifo_fastmode = 1;		/* pipes/fifos will be opened in fast mode */
66 int Fifo_verbose = 0;		/* msg when switching out of fast mode */
67 int Fifohiwat = FIFOHIWAT;	/* Modifiable FIFO high water mark */
68 #endif
69 
70 /*
71  * This is the loadable module wrapper.
72  */
73 #include <sys/modctl.h>
74 
75 extern struct qinit fifo_strdata;
76 
77 struct vfsops *fifo_vfsops;
78 
79 static vfsdef_t vfw = {
80 	VFSDEF_VERSION,
81 	"fifofs",
82 	fifoinit,
83 	0,
84 	NULL
85 };
86 
87 /*
88  * Module linkage information for the kernel.
89  */
90 extern struct mod_ops mod_fsops;
91 
92 static struct modlfs modlfs = {
93 	&mod_fsops, "filesystem for fifo", &vfw
94 };
95 
96 static struct modlinkage modlinkage = {
97 	MODREV_1, (void *)&modlfs, NULL
98 };
99 
100 int
101 _init()
102 {
103 	return (mod_install(&modlinkage));
104 }
105 
106 int
107 _info(struct modinfo *modinfop)
108 {
109 	return (mod_info(&modlinkage, modinfop));
110 }
111 
112 /*
113  * Define data structures within this file.
114  * XXX should the hash size be configurable ?
115  */
116 #define	FIFOSHFT	5
117 #define	FIFO_HASHSZ	63
118 
119 #if ((FIFO_HASHSZ & (FIFO_HASHSZ - 1)) == 0)
120 #define	FIFOHASH(vp) (((uintptr_t)(vp) >> FIFOSHFT) & (FIFO_HASHSZ - 1))
121 #else
122 #define	FIFOHASH(vp) (((uintptr_t)(vp) >> FIFOSHFT) % FIFO_HASHSZ)
123 #endif
124 
125 fifonode_t	*fifoalloc[FIFO_HASHSZ];
126 dev_t		fifodev;
127 struct vfs	*fifovfsp;
128 int		fifofstype;
129 
130 kmutex_t ftable_lock;
131 static kmutex_t fino_lock;
132 struct kmem_cache *fnode_cache;
133 struct kmem_cache *pipe_cache;
134 
135 static void fifoinsert(fifonode_t *);
136 static fifonode_t *fifofind(vnode_t *);
137 static int fifo_connld(struct vnode **, int, cred_t *);
138 static void fifo_fastturnoff(fifonode_t *);
139 
140 static void fifo_reinit_vp(vnode_t *);
141 
142 /*
143  * Constructor/destructor routines for fifos and pipes.
144  *
145  * In the interest of code sharing, we define a common fifodata structure
146  * which consists of a fifolock and one or two fnodes.  A fifo contains
147  * one fnode; a pipe contains two.  The fifolock is shared by the fnodes,
148  * each of which points to it:
149  *
150  *	--> -->	---------  --- ---
151  *	|   |	| lock	|   |	|
152  *	|   |	---------   |	|
153  *	|   |	|	|  fifo	|
154  *	|   --- | fnode	|   |	|
155  *	|	|	|   |  pipe
156  *	|	---------  ---	|
157  *	|	|	|	|
158  *	------- | fnode	|	|
159  *		|	|	|
160  *		---------      ---
161  *
162  * Since the fifolock is at the beginning of the fifodata structure,
163  * the fifolock address is the same as the fifodata address.  Thus,
164  * we can determine the fifodata address from any of its member fnodes.
165  * This is essential for fifo_inactive.
166  *
167  * The fnode constructor is designed to handle any fifodata structure,
168  * deducing the number of fnodes from the total size.  Thus, the fnode
169  * constructor does most of the work for the pipe constructor.
170  */
171 /*ARGSUSED1*/
172 static int
173 fnode_constructor(void *buf, void *cdrarg, int kmflags)
174 {
175 	fifodata_t *fdp = buf;
176 	fifolock_t *flp = &fdp->fifo_lock;
177 	fifonode_t *fnp = &fdp->fifo_fnode[0];
178 	size_t size = (uintptr_t)cdrarg;
179 
180 	mutex_init(&flp->flk_lock, NULL, MUTEX_DEFAULT, NULL);
181 	cv_init(&flp->flk_wait_cv, NULL, CV_DEFAULT, NULL);
182 	flp->flk_ocsync = 0;
183 
184 	while ((char *)fnp < (char *)buf + size) {
185 
186 		vnode_t *vp;
187 
188 		vp = vn_alloc(KM_SLEEP);
189 		fnp->fn_vnode = vp;
190 
191 		fnp->fn_lock = flp;
192 		fnp->fn_open = 0;
193 		fnp->fn_dest = fnp;
194 		fnp->fn_mp = NULL;
195 		fnp->fn_count = 0;
196 		fnp->fn_rsynccnt = 0;
197 		fnp->fn_wsynccnt = 0;
198 		fnp->fn_wwaitcnt = 0;
199 		fnp->fn_insync = 0;
200 		fnp->fn_pcredp = NULL;
201 		fnp->fn_cpid = -1;
202 		/*
203 		 * 32-bit stat(2) may fail if fn_ino isn't initialized
204 		 */
205 		fnp->fn_ino = 0;
206 
207 		cv_init(&fnp->fn_wait_cv, NULL, CV_DEFAULT, NULL);
208 
209 		vn_setops(vp, fifo_vnodeops);
210 		vp->v_stream = NULL;
211 		vp->v_type = VFIFO;
212 		vp->v_data = (caddr_t)fnp;
213 		vp->v_flag = VNOMAP | VNOSWAP;
214 		vn_exists(vp);
215 		fnp++;
216 	}
217 	return (0);
218 }
219 
220 static void
221 fnode_destructor(void *buf, void *cdrarg)
222 {
223 	fifodata_t *fdp = buf;
224 	fifolock_t *flp = &fdp->fifo_lock;
225 	fifonode_t *fnp = &fdp->fifo_fnode[0];
226 	size_t size = (uintptr_t)cdrarg;
227 
228 	mutex_destroy(&flp->flk_lock);
229 	cv_destroy(&flp->flk_wait_cv);
230 	ASSERT(flp->flk_ocsync == 0);
231 
232 	while ((char *)fnp < (char *)buf + size) {
233 
234 		vnode_t *vp = FTOV(fnp);
235 
236 		ASSERT(fnp->fn_mp == NULL);
237 		ASSERT(fnp->fn_count == 0);
238 		ASSERT(fnp->fn_lock == flp);
239 		ASSERT(fnp->fn_open == 0);
240 		ASSERT(fnp->fn_insync == 0);
241 		ASSERT(fnp->fn_rsynccnt == 0 && fnp->fn_wsynccnt == 0);
242 		ASSERT(fnp->fn_wwaitcnt == 0);
243 		ASSERT(fnp->fn_pcredp == NULL);
244 		ASSERT(vn_matchops(vp, fifo_vnodeops));
245 		ASSERT(vp->v_stream == NULL);
246 		ASSERT(vp->v_type == VFIFO);
247 		ASSERT(vp->v_data == (caddr_t)fnp);
248 		ASSERT((vp->v_flag & (VNOMAP|VNOSWAP)) == (VNOMAP|VNOSWAP));
249 
250 		cv_destroy(&fnp->fn_wait_cv);
251 		vn_invalid(vp);
252 		vn_free(vp);
253 
254 		fnp++;
255 	}
256 }
257 
258 static int
259 pipe_constructor(void *buf, void *cdrarg, int kmflags)
260 {
261 	fifodata_t *fdp = buf;
262 	fifonode_t *fnp1 = &fdp->fifo_fnode[0];
263 	fifonode_t *fnp2 = &fdp->fifo_fnode[1];
264 	vnode_t *vp1;
265 	vnode_t *vp2;
266 
267 	(void) fnode_constructor(buf, cdrarg, kmflags);
268 
269 	vp1 = FTOV(fnp1);
270 	vp2 = FTOV(fnp2);
271 
272 	vp1->v_vfsp	= vp2->v_vfsp		= fifovfsp;
273 	vp1->v_rdev	= vp2->v_rdev		= fifodev;
274 	fnp1->fn_realvp	= fnp2->fn_realvp	= NULL;
275 	fnp1->fn_dest	= fnp2;
276 	fnp2->fn_dest	= fnp1;
277 
278 	return (0);
279 }
280 
281 static void
282 pipe_destructor(void *buf, void *cdrarg)
283 {
284 #ifdef DEBUG
285 	fifodata_t *fdp = buf;
286 	fifonode_t *fnp1 = &fdp->fifo_fnode[0];
287 	fifonode_t *fnp2 = &fdp->fifo_fnode[1];
288 	vnode_t *vp1 = FTOV(fnp1);
289 	vnode_t *vp2 = FTOV(fnp2);
290 
291 	ASSERT(vp1->v_vfsp == fifovfsp);
292 	ASSERT(vp2->v_vfsp == fifovfsp);
293 	ASSERT(vp1->v_rdev == fifodev);
294 	ASSERT(vp2->v_rdev == fifodev);
295 #endif
296 	fnode_destructor(buf, cdrarg);
297 }
298 
299 /*
300  * Reinitialize a FIFO vnode (uses normal vnode reinit, but ensures that
301  * vnode type and flags are reset).
302  */
303 
304 static void fifo_reinit_vp(vnode_t *vp)
305 {
306 	vn_reinit(vp);
307 	vp->v_type = VFIFO;
308 	vp->v_flag &= VROOT;
309 	vp->v_flag |= VNOMAP | VNOSWAP;
310 }
311 
312 /*
313  * Save file system type/index, initialize vfs operations vector, get
314  * unique device number for FIFOFS and initialize the FIFOFS hash.
315  * Create and initialize a "generic" vfs pointer that will be placed
316  * in the v_vfsp field of each pipe's vnode.
317  */
318 int
319 fifoinit(int fstype, char *name)
320 {
321 	static const fs_operation_def_t fifo_vfsops_template[] = {
322 		NULL, NULL
323 	};
324 	int error;
325 	major_t dev;
326 
327 	fifofstype = fstype;
328 	error = vfs_setfsops(fstype, fifo_vfsops_template, &fifo_vfsops);
329 	if (error != 0) {
330 		cmn_err(CE_WARN, "fifoinit: bad vfs ops template");
331 		return (error);
332 	}
333 
334 	error = vn_make_ops(name, fifo_vnodeops_template, &fifo_vnodeops);
335 	if (error != 0) {
336 		(void) vfs_freevfsops_by_type(fstype);
337 		cmn_err(CE_WARN, "fifoinit: bad vnode ops template");
338 		return (error);
339 	}
340 
341 	if ((dev = getudev()) == (major_t)-1) {
342 		cmn_err(CE_WARN, "fifoinit: can't get unique device number");
343 		dev = 0;
344 	}
345 	fifodev = makedevice(dev, 0);
346 
347 	fifovfsp = kmem_zalloc(sizeof (struct vfs), KM_SLEEP);
348 	fifovfsp->vfs_next = NULL;
349 	vfs_setops(fifovfsp, fifo_vfsops);
350 	fifovfsp->vfs_vnodecovered = NULL;
351 	fifovfsp->vfs_flag = 0;
352 	fifovfsp->vfs_bsize = 1024;
353 	fifovfsp->vfs_fstype = fifofstype;
354 	vfs_make_fsid(&fifovfsp->vfs_fsid, fifodev, fifofstype);
355 	fifovfsp->vfs_data = NULL;
356 	fifovfsp->vfs_dev = fifodev;
357 	fifovfsp->vfs_bcount = 0;
358 
359 	mutex_init(&ftable_lock, NULL, MUTEX_DEFAULT, NULL);
360 	mutex_init(&fino_lock, NULL, MUTEX_DEFAULT, NULL);
361 
362 	/*
363 	 * vnodes are cached aligned
364 	 */
365 	fnode_cache = kmem_cache_create("fnode_cache",
366 		sizeof (fifodata_t) - sizeof (fifonode_t), 32,
367 		fnode_constructor, fnode_destructor, NULL,
368 		(void *)(sizeof (fifodata_t) - sizeof (fifonode_t)), NULL, 0);
369 
370 	pipe_cache = kmem_cache_create("pipe_cache", sizeof (fifodata_t), 32,
371 		pipe_constructor, pipe_destructor, NULL,
372 		(void *)(sizeof (fifodata_t)), NULL, 0);
373 
374 #if FIFODEBUG
375 	if (Fifohiwat < FIFOHIWAT)
376 		Fifohiwat = FIFOHIWAT;
377 #endif /* FIFODEBUG */
378 	fifo_strdata.qi_minfo->mi_hiwat = Fifohiwat;
379 
380 	return (0);
381 }
382 
383 /*
384  * Provide a shadow for a vnode.  We create a new shadow before checking for an
385  * existing one, to minimize the amount of time we need to hold ftable_lock.
386  * If a vp already has a shadow in the hash list, return its shadow.  If not,
387  * we hash the new vnode and return its pointer to the caller.
388  */
389 vnode_t *
390 fifovp(vnode_t *vp, cred_t *crp)
391 {
392 	fifonode_t *fnp;
393 	fifonode_t *spec_fnp;   /* Speculative fnode ptr. */
394 	fifodata_t *fdp;
395 	vnode_t *newvp;
396 	struct vattr va;
397 	vnode_t	*rvp;
398 
399 	ASSERT(vp != NULL);
400 
401 	fdp = kmem_cache_alloc(fnode_cache, KM_SLEEP);
402 
403 	fdp->fifo_lock.flk_ref = 1;
404 	fnp = &fdp->fifo_fnode[0];
405 
406 	/*
407 	 * Its possible that fifo nodes on different lofs mountpoints
408 	 * shadow the same real filesystem fifo node.
409 	 * In this case its necessary to get and store the realvp.
410 	 * This way different fifo nodes sharing the same real vnode
411 	 * can use realvp for communication.
412 	 */
413 
414 	if (VOP_REALVP(vp, &rvp, NULL) == 0)
415 			vp = rvp;
416 
417 	fnp->fn_realvp	= vp;
418 	fnp->fn_wcnt	= 0;
419 	fnp->fn_rcnt	= 0;
420 
421 #if FIFODEBUG
422 	if (! Fifo_fastmode) {
423 		fnp->fn_flag	= 0;
424 	} else {
425 		fnp->fn_flag	= FIFOFAST;
426 	}
427 #else /* FIFODEBUG */
428 	fnp->fn_flag	= FIFOFAST;
429 #endif /* FIFODEBUG */
430 
431 	/*
432 	 * initialize the times from vp.
433 	 */
434 	va.va_mask = AT_TIMES;
435 	if (VOP_GETATTR(vp, &va, 0, crp, NULL) == 0) {
436 		fnp->fn_atime = va.va_atime.tv_sec;
437 		fnp->fn_mtime = va.va_mtime.tv_sec;
438 		fnp->fn_ctime = va.va_ctime.tv_sec;
439 	} else {
440 		fnp->fn_atime = 0;
441 		fnp->fn_mtime = 0;
442 		fnp->fn_ctime = 0;
443 	}
444 
445 	/*
446 	 * Grab the VP here to avoid holding locks
447 	 * whilst trying to acquire others.
448 	 */
449 
450 	VN_HOLD(vp);
451 
452 	mutex_enter(&ftable_lock);
453 
454 	if ((spec_fnp = fifofind(vp)) != NULL) {
455 		mutex_exit(&ftable_lock);
456 
457 		/*
458 		 * Release the vnode and free up our pre-prepared fnode.
459 		 * Zero the lock reference just to explicitly signal
460 		 * this is unused.
461 		 */
462 		VN_RELE(vp);
463 		fdp->fifo_lock.flk_ref = 0;
464 		kmem_cache_free(fnode_cache, fdp);
465 
466 		return (FTOV(spec_fnp));
467 	}
468 
469 	newvp = FTOV(fnp);
470 	fifo_reinit_vp(newvp);
471 	newvp->v_vfsp = vp->v_vfsp;
472 	newvp->v_rdev = vp->v_rdev;
473 	newvp->v_flag |= (vp->v_flag & VROOT);
474 
475 	fifoinsert(fnp);
476 	mutex_exit(&ftable_lock);
477 
478 	return (newvp);
479 }
480 
481 /*
482  * Create a pipe end by...
483  * allocating a vnode-fifonode pair and initializing the fifonode.
484  */
485 void
486 makepipe(vnode_t **vpp1, vnode_t **vpp2)
487 {
488 	fifonode_t *fnp1;
489 	fifonode_t *fnp2;
490 	vnode_t *nvp1;
491 	vnode_t *nvp2;
492 	fifodata_t *fdp;
493 	time_t now;
494 
495 	fdp = kmem_cache_alloc(pipe_cache, KM_SLEEP);
496 	fdp->fifo_lock.flk_ref = 2;
497 	fnp1 = &fdp->fifo_fnode[0];
498 	fnp2 = &fdp->fifo_fnode[1];
499 
500 	fnp1->fn_wcnt	= fnp2->fn_wcnt		= 1;
501 	fnp1->fn_rcnt	= fnp2->fn_rcnt		= 1;
502 #if FIFODEBUG
503 	if (! Fifo_fastmode) {
504 		fnp1->fn_flag	= fnp2->fn_flag		= ISPIPE;
505 	} else {
506 		fnp1->fn_flag	= fnp2->fn_flag		= ISPIPE | FIFOFAST;
507 	}
508 #else /* FIFODEBUG */
509 	fnp1->fn_flag	= fnp2->fn_flag		= ISPIPE | FIFOFAST;
510 #endif /* FIFODEBUG */
511 	now = gethrestime_sec();
512 	fnp1->fn_atime	= fnp2->fn_atime	= now;
513 	fnp1->fn_mtime	= fnp2->fn_mtime	= now;
514 	fnp1->fn_ctime	= fnp2->fn_ctime	= now;
515 
516 	*vpp1 = nvp1 = FTOV(fnp1);
517 	*vpp2 = nvp2 = FTOV(fnp2);
518 
519 	fifo_reinit_vp(nvp1);		/* Reinitialize vnodes for reuse... */
520 	fifo_reinit_vp(nvp2);
521 	nvp1->v_vfsp = fifovfsp; 	/* Need to re-establish VFS & device */
522 	nvp2->v_vfsp = fifovfsp; 	/* before we can reuse this vnode. */
523 	nvp1->v_rdev = fifodev;
524 	nvp2->v_rdev = fifodev;
525 }
526 
527 /*
528  * Attempt to establish a unique pipe id.  Only un-named pipes use this
529  * routine.
530  */
531 ino_t
532 fifogetid(void)
533 {
534 	static ino_t fifo_ino = 0;
535 	ino_t fino;
536 
537 	mutex_enter(&fino_lock);
538 	fino = fifo_ino++;
539 	mutex_exit(&fino_lock);
540 	return (fino);
541 }
542 
543 
544 /*
545  * Stream a pipe/FIFO.
546  * The FIFOCONNLD flag is used when CONNLD has been pushed on the stream.
547  * If the flag is set, a new vnode is created by calling fifo_connld().
548  * Connld logic was moved to fifo_connld() to speed up the open
549  * operation, simplify the connld/fifo interaction, and remove inherent
550  * race conditions between the connld module and fifos.
551  * This routine is single threaded for two reasons.
552  * 1) connld requests are synchronous; that is, they must block
553  *    until the server does an I_RECVFD (oh, well).  Single threading is
554  *    the simplest way to accomplish this.
555  * 2) fifo_close() must not send M_HANGUP or M_ERROR while we are
556  *    in stropen. Stropen() has a tendency to reset things and
557  *    we would like streams to remember that a hangup occurred.
558  */
559 int
560 fifo_stropen(vnode_t **vpp, int flag, cred_t *crp, int dotwist, int lockheld)
561 {
562 	int error = 0;
563 	vnode_t *oldvp = *vpp;
564 	fifonode_t *fnp = VTOF(*vpp);
565 	dev_t pdev = 0;
566 	int firstopen = 0;
567 	fifolock_t *fn_lock;
568 
569 	fn_lock = fnp->fn_lock;
570 	if (!lockheld)
571 		mutex_enter(&fn_lock->flk_lock);
572 	ASSERT(MUTEX_HELD(&fnp->fn_lock->flk_lock));
573 
574 	/*
575 	 * FIFO is in the process of opening. Wait for it
576 	 * to complete before starting another open on it
577 	 * This prevents races associated with connld open
578 	 */
579 	while (fnp->fn_flag & FIFOOPEN) {
580 		if (!cv_wait_sig(&fnp->fn_wait_cv, &fn_lock->flk_lock)) {
581 			fifo_cleanup(oldvp, flag);
582 			if (!lockheld)
583 				mutex_exit(&fn_lock->flk_lock);
584 			return (EINTR);
585 		}
586 	}
587 
588 	/*
589 	 * The other end of the pipe is almost closed so
590 	 * reject any other open on this end of the pipe
591 	 * This only happens with a pipe mounted under namefs
592 	 */
593 	if ((fnp->fn_flag & (FIFOCLOSE|ISPIPE)) == (FIFOCLOSE|ISPIPE)) {
594 		fifo_cleanup(oldvp, flag);
595 		cv_broadcast(&fnp->fn_wait_cv);
596 		if (!lockheld)
597 			mutex_exit(&fn_lock->flk_lock);
598 		return (ENXIO);
599 	}
600 
601 	fnp->fn_flag |= FIFOOPEN;
602 
603 	/*
604 	 * can't allow close to happen while we are
605 	 * in the middle of stropen().
606 	 * M_HANGUP and M_ERROR could leave the stream in a strange state
607 	 */
608 	while (fn_lock->flk_ocsync)
609 		cv_wait(&fn_lock->flk_wait_cv, &fn_lock->flk_lock);
610 
611 	fn_lock->flk_ocsync = 1;
612 
613 	if (fnp->fn_flag & FIFOCONNLD) {
614 		/*
615 		 * This is a reopen, so we should release the fifo lock
616 		 * just in case some strange module pushed on connld
617 		 * has some odd side effect.
618 		 * Note: this stropen is on the oldvp.  It will
619 		 * have no impact on the connld vp returned and
620 		 * strclose() will only be called when we release
621 		 * flk_ocsync
622 		 */
623 		mutex_exit(&fn_lock->flk_lock);
624 		if ((error = stropen(oldvp, &pdev, flag, crp)) != 0) {
625 			mutex_enter(&fn_lock->flk_lock);
626 			fifo_cleanup(oldvp, flag);
627 			fn_lock->flk_ocsync = 0;
628 			cv_broadcast(&fn_lock->flk_wait_cv);
629 			goto out;
630 		}
631 		/*
632 		 * streams open done, allow close on other end if
633 		 * required.  Do this now.. it could
634 		 * be a very long time before fifo_connld returns.
635 		 */
636 		mutex_enter(&fn_lock->flk_lock);
637 		/*
638 		 * we need to fake an open here so that if this
639 		 * end of the pipe closes, we don't loose the
640 		 * stream head (kind of like single threading
641 		 * open and close for this end of the pipe)
642 		 * We'll need to call fifo_close() to do clean
643 		 * up in case this end of the pipe was closed
644 		 * down while we were in fifo_connld()
645 		 */
646 		ASSERT(fnp->fn_open > 0);
647 		fnp->fn_open++;
648 		fn_lock->flk_ocsync = 0;
649 		cv_broadcast(&fn_lock->flk_wait_cv);
650 		mutex_exit(&fn_lock->flk_lock);
651 		/*
652 		 * Connld has been pushed onto the pipe
653 		 * Create new pipe on behalf of connld
654 		 */
655 		if (error = fifo_connld(vpp, flag, crp)) {
656 			(void) fifo_close(oldvp, flag, 1, 0, crp, NULL);
657 			mutex_enter(&fn_lock->flk_lock);
658 			goto out;
659 		}
660 		/*
661 		 * undo fake open.  We need to call fifo_close
662 		 * because some other thread could have done
663 		 * a close and detach of the named pipe while
664 		 * we were in fifo_connld(), so
665 		 * we want to make sure the close completes (yuk)
666 		 */
667 		(void) fifo_close(oldvp, flag, 1, 0, crp, NULL);
668 		/*
669 		 * fifo_connld has changed the vp, so we
670 		 * need to re-initialize locals
671 		 */
672 		fnp = VTOF(*vpp);
673 		fn_lock = fnp->fn_lock;
674 		mutex_enter(&fn_lock->flk_lock);
675 	} else {
676 		/*
677 		 * release lock in case there are modules pushed that
678 		 * could have some strange side effect
679 		 */
680 
681 		mutex_exit(&fn_lock->flk_lock);
682 
683 		/*
684 		 * If this is the first open of a fifo (dotwist
685 		 * will be non-zero) we will need to twist the queues.
686 		 */
687 		if (oldvp->v_stream == NULL)
688 			firstopen = 1;
689 
690 
691 		/*
692 		 * normal open of pipe/fifo
693 		 */
694 
695 		if ((error = stropen(oldvp, &pdev, flag, crp)) != 0) {
696 			mutex_enter(&fn_lock->flk_lock);
697 			fifo_cleanup(oldvp, flag);
698 			ASSERT(fnp->fn_open != 0 || oldvp->v_stream == NULL);
699 			fn_lock->flk_ocsync = 0;
700 			cv_broadcast(&fn_lock->flk_wait_cv);
701 			goto out;
702 		}
703 		mutex_enter(&fn_lock->flk_lock);
704 
705 		/*
706 		 * twist the ends of the fifo together
707 		 */
708 		if (dotwist && firstopen)
709 			strmate(*vpp, *vpp);
710 
711 		/*
712 		 * Show that this open has succeeded
713 		 * and allow closes or other opens to proceed
714 		 */
715 		fnp->fn_open++;
716 		fn_lock->flk_ocsync = 0;
717 		cv_broadcast(&fn_lock->flk_wait_cv);
718 	}
719 out:
720 	fnp->fn_flag &= ~FIFOOPEN;
721 	if (error == 0) {
722 		fnp->fn_flag |= FIFOISOPEN;
723 		/*
724 		 * If this is a FIFO and has the close flag set
725 		 * and there are now writers, clear the close flag
726 		 * Note: close flag only gets set when last writer
727 		 * on a FIFO goes away.
728 		 */
729 		if (((fnp->fn_flag & (ISPIPE|FIFOCLOSE)) == FIFOCLOSE) &&
730 		    fnp->fn_wcnt > 0)
731 			fnp->fn_flag &= ~FIFOCLOSE;
732 	}
733 	cv_broadcast(&fnp->fn_wait_cv);
734 	if (!lockheld)
735 		mutex_exit(&fn_lock->flk_lock);
736 	return (error);
737 }
738 
739 /*
740  * Clean up the state of a FIFO and/or mounted pipe in the
741  * event that a fifo_open() was interrupted while the
742  * process was blocked.
743  */
744 void
745 fifo_cleanup(vnode_t *vp, int flag)
746 {
747 	fifonode_t *fnp = VTOF(vp);
748 
749 	ASSERT(MUTEX_HELD(&fnp->fn_lock->flk_lock));
750 
751 	cleanlocks(vp, curproc->p_pid, 0);
752 	cleanshares(vp, curproc->p_pid);
753 	if (flag & FREAD) {
754 		fnp->fn_rcnt--;
755 	}
756 	if (flag & FWRITE) {
757 		fnp->fn_wcnt--;
758 	}
759 	cv_broadcast(&fnp->fn_wait_cv);
760 }
761 
762 
763 /*
764  * Insert a fifonode-vnode pair onto the fifoalloc hash list.
765  */
766 static void
767 fifoinsert(fifonode_t *fnp)
768 {
769 	int idx = FIFOHASH(fnp->fn_realvp);
770 
771 	/*
772 	 * We don't need to hold fn_lock since we're holding ftable_lock and
773 	 * this routine is only called right after we've allocated an fnode.
774 	 * FIFO is inserted at head of NULL terminated doubly linked list.
775 	 */
776 
777 	ASSERT(MUTEX_HELD(&ftable_lock));
778 	fnp->fn_backp = NULL;
779 	fnp->fn_nextp = fifoalloc[idx];
780 	fifoalloc[idx] = fnp;
781 	if (fnp->fn_nextp)
782 		fnp->fn_nextp->fn_backp = fnp;
783 }
784 
785 /*
786  * Find a fifonode-vnode pair on the fifoalloc hash list.
787  * vp is a vnode to be shadowed. If it's on the hash list,
788  * it already has a shadow, therefore return its corresponding
789  * fifonode.
790  */
791 static fifonode_t *
792 fifofind(vnode_t *vp)
793 {
794 	fifonode_t *fnode;
795 
796 	ASSERT(MUTEX_HELD(&ftable_lock));
797 	for (fnode = fifoalloc[FIFOHASH(vp)]; fnode; fnode = fnode->fn_nextp) {
798 		if (fnode->fn_realvp == vp) {
799 			VN_HOLD(FTOV(fnode));
800 			return (fnode);
801 		}
802 	}
803 	return (NULL);
804 }
805 
806 /*
807  * Remove a fifonode-vnode pair from the fifoalloc hash list.
808  * This routine is called from the fifo_inactive() routine when a
809  * FIFO is being released.
810  * If the link to be removed is the only link, set fifoalloc to NULL.
811  */
812 void
813 fiforemove(fifonode_t *fnp)
814 {
815 	int idx = FIFOHASH(fnp->fn_realvp);
816 	fifonode_t *fnode;
817 
818 	ASSERT(MUTEX_HELD(&ftable_lock));
819 	fnode = fifoalloc[idx];
820 	/*
821 	 * fast path... only 1 FIFO in this list entry
822 	 */
823 	if (fnode != NULL && fnode == fnp &&
824 		!fnode->fn_nextp && !fnode->fn_backp) {
825 			fifoalloc[idx] = NULL;
826 	} else {
827 
828 		for (;  fnode;  fnode = fnode->fn_nextp) {
829 			if (fnode == fnp) {
830 				/*
831 				 * if we are first entry
832 				 */
833 				if (fnp == fifoalloc[idx])
834 					fifoalloc[idx] = fnp->fn_nextp;
835 				if (fnode->fn_nextp)
836 					fnode->fn_nextp->fn_backp =
837 						fnode->fn_backp;
838 				if (fnode->fn_backp)
839 					fnode->fn_backp->fn_nextp =
840 						fnode->fn_nextp;
841 				break;
842 			}
843 		}
844 	}
845 }
846 
847 /*
848  * Flush all data from a fifo's message queue
849  */
850 
851 void
852 fifo_fastflush(fifonode_t *fnp)
853 {
854 	mblk_t *bp;
855 	ASSERT(MUTEX_HELD(&fnp->fn_lock->flk_lock));
856 
857 	if ((bp = fnp->fn_mp) != NULL) {
858 		fnp->fn_mp = NULL;
859 		fnp->fn_count = 0;
860 		freemsg(bp);
861 	}
862 	fifo_wakewriter(fnp->fn_dest, fnp->fn_lock);
863 }
864 
865 /*
866  * Note:  This routine is single threaded
867  *  Protected by FIFOOPEN flag (i.e. flk_lock is not held)
868  *  Upon successful completion, the original fifo is unlocked
869  *  and FIFOOPEN is cleared for the original vpp.
870  *  The new fifo returned has FIFOOPEN set.
871  */
872 static int
873 fifo_connld(struct vnode **vpp, int flag, cred_t *crp)
874 {
875 	struct vnode *vp1;
876 	struct vnode *vp2;
877 	struct fifonode *oldfnp;
878 	struct fifonode *fn_dest;
879 	int error;
880 	struct file *filep;
881 	struct fifolock *fn_lock;
882 	cred_t *c;
883 
884 	/*
885 	 * Get two vnodes that will represent the pipe ends for the new pipe.
886 	 */
887 	makepipe(&vp1, &vp2);
888 
889 	/*
890 	 * Allocate a file descriptor and file pointer for one of the pipe
891 	 * ends. The file descriptor will be used to send that pipe end to
892 	 * the process on the other end of this stream. Note that we get
893 	 * the file structure only, there is no file list entry allocated.
894 	 */
895 	if (error = falloc(vp1, FWRITE|FREAD, &filep, NULL)) {
896 		VN_RELE(vp1);
897 		VN_RELE(vp2);
898 		return (error);
899 	}
900 	mutex_exit(&filep->f_tlock);
901 	oldfnp = VTOF(*vpp);
902 	fn_lock = oldfnp->fn_lock;
903 	fn_dest = oldfnp->fn_dest;
904 
905 	/*
906 	 * Create two new stream heads and attach them to the two vnodes for
907 	 * the new pipe.
908 	 */
909 	if ((error = fifo_stropen(&vp1, FREAD|FWRITE, filep->f_cred, 0, 0)) !=
910 	    0 ||
911 	    (error = fifo_stropen(&vp2, flag, filep->f_cred, 0, 0)) != 0) {
912 #if DEBUG
913 		cmn_err(CE_NOTE, "fifo stropen failed error 0x%x",
914 			error);
915 #endif
916 		/*
917 		 * this will call fifo_close and VN_RELE on vp1
918 		 */
919 		(void) closef(filep);
920 		VN_RELE(vp2);
921 		return (error);
922 	}
923 
924 	/*
925 	 * twist the ends of the pipe together
926 	 */
927 	strmate(vp1, vp2);
928 
929 	/*
930 	 * Set our end to busy in open
931 	 * Note: Don't need lock around this because we're the only
932 	 * one who knows about it
933 	 */
934 	VTOF(vp2)->fn_flag |= FIFOOPEN;
935 
936 	mutex_enter(&fn_lock->flk_lock);
937 
938 	fn_dest->fn_flag |= FIFOSEND;
939 	/*
940 	 * check to make sure neither end of pipe has gone away
941 	 */
942 	if (!(fn_dest->fn_flag & FIFOISOPEN)) {
943 		error = ENXIO;
944 		fn_dest->fn_flag &= ~FIFOSEND;
945 		mutex_exit(&fn_lock->flk_lock);
946 		/*
947 		 * this will call fifo_close and VN_RELE on vp1
948 		 */
949 		goto out;
950 	}
951 	mutex_exit(&fn_lock->flk_lock);
952 
953 	/*
954 	 * Tag the sender's credential on the pipe descriptor.
955 	 */
956 	crhold(VTOF(vp1)->fn_pcredp = crp);
957 	VTOF(vp1)->fn_cpid = curproc->p_pid;
958 
959 	/*
960 	 * send the file descriptor to other end of pipe
961 	 */
962 	if (error = do_sendfp((*vpp)->v_stream, filep, crp)) {
963 		mutex_enter(&fn_lock->flk_lock);
964 		fn_dest->fn_flag &= ~FIFOSEND;
965 		mutex_exit(&fn_lock->flk_lock);
966 		/*
967 		 * this will call fifo_close and VN_RELE on vp1
968 		 */
969 		goto out;
970 	}
971 
972 	mutex_enter(&fn_lock->flk_lock);
973 	/*
974 	 * Wait for other end to receive file descriptor
975 	 * FIFOCLOSE indicates that one or both sides of the pipe
976 	 * have gone away.
977 	 */
978 	while ((fn_dest->fn_flag & (FIFOCLOSE | FIFOSEND)) == FIFOSEND) {
979 		if (!cv_wait_sig(&oldfnp->fn_wait_cv, &fn_lock->flk_lock)) {
980 			error = EINTR;
981 			fn_dest->fn_flag &= ~FIFOSEND;
982 			mutex_exit(&fn_lock->flk_lock);
983 			goto out;
984 		}
985 	}
986 	/*
987 	 * If either end of pipe has gone away and the other end did not
988 	 * receive pipe, reject the connld open
989 	 */
990 	if ((fn_dest->fn_flag & FIFOSEND)) {
991 		error = ENXIO;
992 		fn_dest->fn_flag &= ~FIFOSEND;
993 		mutex_exit(&fn_lock->flk_lock);
994 		goto out;
995 	}
996 
997 	oldfnp->fn_flag &= ~FIFOOPEN;
998 	cv_broadcast(&oldfnp->fn_wait_cv);
999 	mutex_exit(&fn_lock->flk_lock);
1000 
1001 	VN_RELE(*vpp);
1002 	*vpp = vp2;
1003 	(void) closef(filep);
1004 	return (0);
1005 out:
1006 	c = filep->f_cred;
1007 	crhold(c);
1008 	(void) closef(filep);
1009 	VTOF(vp2)->fn_flag &= ~FIFOOPEN;
1010 	(void) fifo_close(vp2, flag, 1, (offset_t)0, c, NULL);
1011 	crfree(c);
1012 	VN_RELE(vp2);
1013 	return (error);
1014 }
1015 
1016 /*
1017  * Disable fastpath mode.
1018  */
1019 void
1020 fifo_fastoff(fifonode_t *fnp)
1021 {
1022 	ASSERT(MUTEX_HELD(&fnp->fn_lock->flk_lock));
1023 	ASSERT(FTOV(fnp)->v_stream);
1024 
1025 	/* FIFOSTAYFAST is set => FIFOFAST is set */
1026 	while ((fnp->fn_flag & FIFOSTAYFAST) || ((fnp->fn_flag & ISPIPE) &&
1027 	    (fnp->fn_dest->fn_flag & FIFOSTAYFAST))) {
1028 		ASSERT(fnp->fn_flag & FIFOFAST);
1029 		/* indicate someone is waiting to turn into stream mode */
1030 		fnp->fn_flag |= FIFOWAITMODE;
1031 		cv_wait(&fnp->fn_wait_cv, &fnp->fn_lock->flk_lock);
1032 		fnp->fn_flag &= ~FIFOWAITMODE;
1033 	}
1034 
1035 	/* as we may have relased the lock, test the FIFOFAST flag here */
1036 	if (!(fnp->fn_flag & FIFOFAST))
1037 		return;
1038 #if FIFODEBUG
1039 	if (Fifo_verbose)
1040 		cmn_err(CE_NOTE, "Fifo reverting to streams mode\n");
1041 #endif
1042 
1043 	fifo_fastturnoff(fnp);
1044 	if (fnp->fn_flag & ISPIPE) {
1045 		fifo_fastturnoff(fnp->fn_dest);
1046 	}
1047 }
1048 
1049 
1050 /*
1051  * flk_lock must be held while calling fifo_fastturnoff() to
1052  * preserve data ordering (no reads or writes allowed)
1053  */
1054 
1055 static void
1056 fifo_fastturnoff(fifonode_t *fnp)
1057 {
1058 	fifonode_t *fn_dest = fnp->fn_dest;
1059 	mblk_t	*fn_mp;
1060 	int	fn_flag;
1061 
1062 	ASSERT(MUTEX_HELD(&fnp->fn_lock->flk_lock));
1063 	/*
1064 	 * Note: This end can't be closed if there
1065 	 * is stuff in fn_mp
1066 	 */
1067 	if ((fn_mp = fnp->fn_mp) != NULL) {
1068 		ASSERT(fnp->fn_flag & FIFOISOPEN);
1069 		ASSERT(FTOV(fnp)->v_stream != NULL);
1070 		ASSERT(FTOV(fnp)->v_stream->sd_wrq != NULL);
1071 		ASSERT(RD(FTOV(fnp)->v_stream->sd_wrq) != NULL);
1072 		ASSERT(strvp2wq(FTOV(fnp)) != NULL);
1073 		fnp->fn_mp = NULL;
1074 		fnp->fn_count = 0;
1075 		/*
1076 		 * Don't need to drop flk_lock across the put()
1077 		 * since we're just moving the message from the fifo
1078 		 * node to the STREAM head...
1079 		 */
1080 		put(RD(strvp2wq(FTOV(fnp))), fn_mp);
1081 	}
1082 
1083 	/*
1084 	 * Need to re-issue any pending poll requests
1085 	 * so that the STREAMS framework sees them
1086 	 * Writers would be waiting on fnp and readers on fn_dest
1087 	 */
1088 	if ((fnp->fn_flag & (FIFOISOPEN | FIFOPOLLW)) ==
1089 	    (FIFOISOPEN | FIFOPOLLW)) {
1090 		strpollwakeup(FTOV(fnp), POLLWRNORM);
1091 	}
1092 	fn_flag = fn_dest->fn_flag;
1093 	if ((fn_flag & FIFOISOPEN) == FIFOISOPEN) {
1094 		if ((fn_flag & (FIFOPOLLR | FIFOPOLLRBAND))) {
1095 			strpollwakeup(FTOV(fn_dest), POLLIN|POLLRDNORM);
1096 		}
1097 	}
1098 	/*
1099 	 * wake up any sleeping processes so they can notice we went
1100 	 * to streams mode
1101 	 */
1102 	fnp->fn_flag &= ~(FIFOFAST|FIFOWANTW|FIFOWANTR);
1103 	cv_broadcast(&fnp->fn_wait_cv);
1104 }
1105 
1106 /*
1107  * Alternative version of fifo_fastoff()
1108  * optimized for putmsg/getmsg.
1109  */
1110 void
1111 fifo_vfastoff(vnode_t *vp)
1112 {
1113 	fifonode_t	*fnp = VTOF(vp);
1114 
1115 	mutex_enter(&fnp->fn_lock->flk_lock);
1116 	if (!(fnp->fn_flag & FIFOFAST)) {
1117 		mutex_exit(&fnp->fn_lock->flk_lock);
1118 		return;
1119 	}
1120 	fifo_fastoff(fnp);
1121 	mutex_exit(&fnp->fn_lock->flk_lock);
1122 }
1123 
1124 /*
1125  * Wake any sleeping writers, poll and send signals if necessary
1126  * This module is only called when we drop below the hi water mark
1127  * FIFOWANTW indicates that a process is sleeping in fifo_write()
1128  * FIFOHIWATW indicates that we have either attempted a poll or
1129  * non-blocking write and were over the high water mark
1130  * This routine assumes a low water mark of 0.
1131  */
1132 
1133 void
1134 fifo_wakewriter(fifonode_t *fn_dest, fifolock_t *fn_lock)
1135 {
1136 	int fn_dflag = fn_dest->fn_flag;
1137 
1138 	ASSERT(MUTEX_HELD(&fn_lock->flk_lock));
1139 	ASSERT(fn_dest->fn_dest->fn_count < Fifohiwat);
1140 	if ((fn_dflag & FIFOWANTW)) {
1141 		cv_broadcast(&fn_dest->fn_wait_cv);
1142 	}
1143 	if ((fn_dflag & (FIFOHIWATW | FIFOISOPEN)) ==
1144 	    (FIFOHIWATW | FIFOISOPEN)) {
1145 		if (fn_dflag & FIFOPOLLW)
1146 			strpollwakeup(FTOV(fn_dest), POLLWRNORM);
1147 		if (fn_dflag & FIFOSETSIG)
1148 			str_sendsig(FTOV(fn_dest), S_WRNORM, 0, 0);
1149 	}
1150 	/*
1151 	 * FIFOPOLLW can't be set without setting FIFOHIWAT
1152 	 * This allows us to clear both here.
1153 	 */
1154 	fn_dest->fn_flag = fn_dflag & ~(FIFOWANTW | FIFOHIWATW | FIFOPOLLW);
1155 }
1156 
1157 /*
1158  * wake up any sleeping readers, poll or send signal if needed
1159  * FIFOWANTR indicates that a process is waiting in fifo_read() for data
1160  * FIFOSETSIG indicates that SIGPOLL should be sent to process
1161  * FIFOPOLLR indicates that a poll request for reading on the fifo was made
1162  */
1163 
1164 void
1165 fifo_wakereader(fifonode_t *fn_dest, fifolock_t *fn_lock)
1166 {
1167 	int fn_dflag = fn_dest->fn_flag;
1168 
1169 	ASSERT(MUTEX_HELD(&fn_lock->flk_lock));
1170 	if (fn_dflag & FIFOWANTR) {
1171 		cv_broadcast(&fn_dest->fn_wait_cv);
1172 	}
1173 	if (fn_dflag & FIFOISOPEN) {
1174 		if (fn_dflag & FIFOPOLLR)
1175 			strpollwakeup(FTOV(fn_dest), POLLIN | POLLRDNORM);
1176 		if (fn_dflag & FIFOSETSIG)
1177 			str_sendsig(FTOV(fn_dest), S_INPUT | S_RDNORM, 0, 0);
1178 	}
1179 	fn_dest->fn_flag = fn_dflag & ~(FIFOWANTR | FIFOPOLLR);
1180 }
1181