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 (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
25 */
26
27/*
28 * The routines defined in this file are supporting routines for FIFOFS
29 * file system type.
30 */
31#include <sys/types.h>
32#include <sys/param.h>
33#include <sys/systm.h>
34#include <sys/debug.h>
35#include <sys/errno.h>
36#include <sys/time.h>
37#include <sys/kmem.h>
38#include <sys/inline.h>
39#include <sys/file.h>
40#include <sys/proc.h>
41#include <sys/stat.h>
42#include <sys/sysmacros.h>
43#include <sys/var.h>
44#include <sys/vfs.h>
45#include <sys/vfs_opreg.h>
46#include <sys/vnode.h>
47#include <sys/mode.h>
48#include <sys/signal.h>
49#include <sys/user.h>
50#include <sys/uio.h>
51#include <sys/flock.h>
52#include <sys/stream.h>
53#include <sys/fs/fifonode.h>
54#include <sys/strsubr.h>
55#include <sys/stropts.h>
56#include <sys/cmn_err.h>
57#include <fs/fs_subr.h>
58#include <sys/ddi.h>
59
60
61#if FIFODEBUG
62int Fifo_fastmode = 1;		/* pipes/fifos will be opened in fast mode */
63int Fifo_verbose = 0;		/* msg when switching out of fast mode */
64int Fifohiwat = FIFOHIWAT;	/* Modifiable FIFO high water mark */
65#endif
66
67/*
68 * This is the loadable module wrapper.
69 */
70#include <sys/modctl.h>
71
72extern struct qinit fifo_strdata;
73
74struct vfsops *fifo_vfsops;
75
76static vfsdef_t vfw = {
77	VFSDEF_VERSION,
78	"fifofs",
79	fifoinit,
80	VSW_ZMOUNT,
81	NULL
82};
83
84/*
85 * Module linkage information for the kernel.
86 */
87extern struct mod_ops mod_fsops;
88
89static struct modlfs modlfs = {
90	&mod_fsops, "filesystem for fifo", &vfw
91};
92
93static struct modlinkage modlinkage = {
94	MODREV_1, (void *)&modlfs, NULL
95};
96
97int
98_init()
99{
100	return (mod_install(&modlinkage));
101}
102
103int
104_info(struct modinfo *modinfop)
105{
106	return (mod_info(&modlinkage, modinfop));
107}
108
109/*
110 * Define data structures within this file.
111 * XXX should the hash size be configurable ?
112 */
113#define	FIFOSHFT	5
114#define	FIFO_HASHSZ	63
115
116#if ((FIFO_HASHSZ & (FIFO_HASHSZ - 1)) == 0)
117#define	FIFOHASH(vp) (((uintptr_t)(vp) >> FIFOSHFT) & (FIFO_HASHSZ - 1))
118#else
119#define	FIFOHASH(vp) (((uintptr_t)(vp) >> FIFOSHFT) % FIFO_HASHSZ)
120#endif
121
122fifonode_t	*fifoalloc[FIFO_HASHSZ];
123dev_t		fifodev;
124struct vfs	*fifovfsp;
125int		fifofstype;
126
127kmutex_t ftable_lock;
128static kmutex_t fino_lock;
129struct kmem_cache *fnode_cache;
130struct kmem_cache *pipe_cache;
131
132static void fifoinsert(fifonode_t *);
133static fifonode_t *fifofind(vnode_t *);
134static int fifo_connld(struct vnode **, int, cred_t *);
135static void fifo_fastturnoff(fifonode_t *);
136
137static void fifo_reinit_vp(vnode_t *);
138
139static void fnode_destructor(void *, void *);
140
141/*
142 * Constructor/destructor routines for fifos and pipes.
143 *
144 * In the interest of code sharing, we define a common fifodata structure
145 * which consists of a fifolock and one or two fnodes.  A fifo contains
146 * one fnode; a pipe contains two.  The fifolock is shared by the fnodes,
147 * each of which points to it:
148 *
149 *	--> -->	---------  --- ---
150 *	|   |	| lock	|   |	|
151 *	|   |	---------   |	|
152 *	|   |	|	|  fifo	|
153 *	|   --- | fnode	|   |	|
154 *	|	|	|   |  pipe
155 *	|	---------  ---	|
156 *	|	|	|	|
157 *	------- | fnode	|	|
158 *		|	|	|
159 *		---------      ---
160 *
161 * Since the fifolock is at the beginning of the fifodata structure,
162 * the fifolock address is the same as the fifodata address.  Thus,
163 * we can determine the fifodata address from any of its member fnodes.
164 * This is essential for fifo_inactive.
165 *
166 * The fnode constructor is designed to handle any fifodata structure,
167 * deducing the number of fnodes from the total size.  Thus, the fnode
168 * constructor does most of the work for the pipe constructor.
169 */
170static int
171fnode_constructor(void *buf, void *cdrarg, int kmflags)
172{
173	fifodata_t *fdp = buf;
174	fifolock_t *flp = &fdp->fifo_lock;
175	fifonode_t *fnp = &fdp->fifo_fnode[0];
176	size_t size = (uintptr_t)cdrarg;
177
178	mutex_init(&flp->flk_lock, NULL, MUTEX_DEFAULT, NULL);
179	cv_init(&flp->flk_wait_cv, NULL, CV_DEFAULT, NULL);
180	flp->flk_ocsync = 0;
181
182	while ((char *)fnp < (char *)buf + size) {
183
184		vnode_t *vp;
185
186		vp = vn_alloc(kmflags);
187		if (vp == NULL) {
188			fnp->fn_vnode = NULL; /* mark for destructor */
189			fnode_destructor(buf, cdrarg);
190			return (-1);
191		}
192		fnp->fn_vnode = vp;
193
194		fnp->fn_lock = flp;
195		fnp->fn_open = 0;
196		fnp->fn_dest = fnp;
197		fnp->fn_mp = NULL;
198		fnp->fn_count = 0;
199		fnp->fn_rsynccnt = 0;
200		fnp->fn_wsynccnt = 0;
201		fnp->fn_wwaitcnt = 0;
202		fnp->fn_insync = 0;
203		fnp->fn_pcredp = NULL;
204		fnp->fn_cpid = -1;
205		/*
206		 * 32-bit stat(2) may fail if fn_ino isn't initialized
207		 */
208		fnp->fn_ino = 0;
209
210		cv_init(&fnp->fn_wait_cv, NULL, CV_DEFAULT, NULL);
211
212		vn_setops(vp, fifo_vnodeops);
213		vp->v_stream = NULL;
214		vp->v_type = VFIFO;
215		vp->v_data = (caddr_t)fnp;
216		vp->v_flag = VNOMAP | VNOSWAP;
217		vn_exists(vp);
218		fnp++;
219	}
220	return (0);
221}
222
223static void
224fnode_destructor(void *buf, void *cdrarg)
225{
226	fifodata_t *fdp = buf;
227	fifolock_t *flp = &fdp->fifo_lock;
228	fifonode_t *fnp = &fdp->fifo_fnode[0];
229	size_t size = (uintptr_t)cdrarg;
230
231	mutex_destroy(&flp->flk_lock);
232	cv_destroy(&flp->flk_wait_cv);
233	ASSERT(flp->flk_ocsync == 0);
234
235	while ((char *)fnp < (char *)buf + size) {
236
237		vnode_t *vp = FTOV(fnp);
238
239		if (vp == NULL) {
240			return; /* constructor failed here */
241		}
242
243		ASSERT(fnp->fn_mp == NULL);
244		ASSERT(fnp->fn_count == 0);
245		ASSERT(fnp->fn_lock == flp);
246		ASSERT(fnp->fn_open == 0);
247		ASSERT(fnp->fn_insync == 0);
248		ASSERT(fnp->fn_rsynccnt == 0 && fnp->fn_wsynccnt == 0);
249		ASSERT(fnp->fn_wwaitcnt == 0);
250		ASSERT(fnp->fn_pcredp == NULL);
251		ASSERT(vn_matchops(vp, fifo_vnodeops));
252		ASSERT(vp->v_stream == NULL);
253		ASSERT(vp->v_type == VFIFO);
254		ASSERT(vp->v_data == (caddr_t)fnp);
255		ASSERT((vp->v_flag & (VNOMAP|VNOSWAP)) == (VNOMAP|VNOSWAP));
256
257		cv_destroy(&fnp->fn_wait_cv);
258		vn_invalid(vp);
259		vn_free(vp);
260
261		fnp++;
262	}
263}
264
265static int
266pipe_constructor(void *buf, void *cdrarg, int kmflags)
267{
268	fifodata_t *fdp = buf;
269	fifonode_t *fnp1 = &fdp->fifo_fnode[0];
270	fifonode_t *fnp2 = &fdp->fifo_fnode[1];
271	vnode_t *vp1;
272	vnode_t *vp2;
273
274	(void) fnode_constructor(buf, cdrarg, kmflags);
275
276	vp1 = FTOV(fnp1);
277	vp2 = FTOV(fnp2);
278
279	vp1->v_vfsp	= vp2->v_vfsp		= fifovfsp;
280	vp1->v_rdev	= vp2->v_rdev		= fifodev;
281	fnp1->fn_realvp	= fnp2->fn_realvp	= NULL;
282	fnp1->fn_dest	= fnp2;
283	fnp2->fn_dest	= fnp1;
284
285	return (0);
286}
287
288static void
289pipe_destructor(void *buf, void *cdrarg)
290{
291#ifdef DEBUG
292	fifodata_t *fdp = buf;
293	fifonode_t *fnp1 = &fdp->fifo_fnode[0];
294	fifonode_t *fnp2 = &fdp->fifo_fnode[1];
295	vnode_t *vp1 = FTOV(fnp1);
296	vnode_t *vp2 = FTOV(fnp2);
297
298	ASSERT(vp1->v_vfsp == fifovfsp);
299	ASSERT(vp2->v_vfsp == fifovfsp);
300	ASSERT(vp1->v_rdev == fifodev);
301	ASSERT(vp2->v_rdev == fifodev);
302#endif
303	fnode_destructor(buf, cdrarg);
304}
305
306/*
307 * Reinitialize a FIFO vnode (uses normal vnode reinit, but ensures that
308 * vnode type and flags are reset).
309 */
310
311static void fifo_reinit_vp(vnode_t *vp)
312{
313	vn_reinit(vp);
314	vp->v_type = VFIFO;
315	vp->v_flag &= VROOT;
316	vp->v_flag |= VNOMAP | VNOSWAP;
317}
318
319/*
320 * Save file system type/index, initialize vfs operations vector, get
321 * unique device number for FIFOFS and initialize the FIFOFS hash.
322 * Create and initialize a "generic" vfs pointer that will be placed
323 * in the v_vfsp field of each pipe's vnode.
324 */
325int
326fifoinit(int fstype, char *name)
327{
328	static const fs_operation_def_t fifo_vfsops_template[] = {
329		NULL, NULL
330	};
331	int error;
332	major_t dev;
333
334	fifofstype = fstype;
335	error = vfs_setfsops(fstype, fifo_vfsops_template, &fifo_vfsops);
336	if (error != 0) {
337		cmn_err(CE_WARN, "fifoinit: bad vfs ops template");
338		return (error);
339	}
340
341	error = vn_make_ops(name, fifo_vnodeops_template, &fifo_vnodeops);
342	if (error != 0) {
343		(void) vfs_freevfsops_by_type(fstype);
344		cmn_err(CE_WARN, "fifoinit: bad vnode ops template");
345		return (error);
346	}
347
348	if ((dev = getudev()) == (major_t)-1) {
349		cmn_err(CE_WARN, "fifoinit: can't get unique device number");
350		dev = 0;
351	}
352	fifodev = makedevice(dev, 0);
353
354	fifovfsp = kmem_zalloc(sizeof (struct vfs), KM_SLEEP);
355	fifovfsp->vfs_next = NULL;
356	vfs_setops(fifovfsp, fifo_vfsops);
357	fifovfsp->vfs_vnodecovered = NULL;
358	fifovfsp->vfs_flag = 0;
359	fifovfsp->vfs_bsize = 1024;
360	fifovfsp->vfs_fstype = fifofstype;
361	vfs_make_fsid(&fifovfsp->vfs_fsid, fifodev, fifofstype);
362	fifovfsp->vfs_data = NULL;
363	fifovfsp->vfs_dev = fifodev;
364	fifovfsp->vfs_bcount = 0;
365
366	/*
367	 * It is necessary to initialize vfs_count here to 1.
368	 * This prevents the fifovfsp from getting freed when
369	 * a thread does a VFS_HOLD followed by a VFS_RELE
370	 * on the fifovfsp
371	 *
372	 * The fifovfsp should never be freed.
373	 */
374	fifovfsp->vfs_count = 1;
375
376	mutex_init(&ftable_lock, NULL, MUTEX_DEFAULT, NULL);
377	mutex_init(&fino_lock, NULL, MUTEX_DEFAULT, NULL);
378
379	/*
380	 * vnodes are cached aligned
381	 */
382	fnode_cache = kmem_cache_create("fnode_cache",
383	    sizeof (fifodata_t) - sizeof (fifonode_t), 32,
384	    fnode_constructor, fnode_destructor, NULL,
385	    (void *)(sizeof (fifodata_t) - sizeof (fifonode_t)), NULL, 0);
386
387	pipe_cache = kmem_cache_create("pipe_cache", sizeof (fifodata_t), 32,
388	    pipe_constructor, pipe_destructor, NULL,
389	    (void *)(sizeof (fifodata_t)), NULL, 0);
390
391#if FIFODEBUG
392	if (Fifohiwat < FIFOHIWAT)
393		Fifohiwat = FIFOHIWAT;
394#endif /* FIFODEBUG */
395	fifo_strdata.qi_minfo->mi_hiwat = Fifohiwat;
396
397	return (0);
398}
399
400/*
401 * Provide a shadow for a vnode.  We create a new shadow before checking for an
402 * existing one, to minimize the amount of time we need to hold ftable_lock.
403 * If a vp already has a shadow in the hash list, return its shadow.  If not,
404 * we hash the new vnode and return its pointer to the caller.
405 */
406vnode_t *
407fifovp(vnode_t *vp, cred_t *crp)
408{
409	fifonode_t *fnp;
410	fifonode_t *spec_fnp;   /* Speculative fnode ptr. */
411	fifodata_t *fdp;
412	vnode_t *newvp;
413	struct vattr va;
414	vnode_t	*rvp;
415
416	ASSERT(vp != NULL);
417
418	fdp = kmem_cache_alloc(fnode_cache, KM_SLEEP);
419
420	fdp->fifo_lock.flk_ref = 1;
421	fnp = &fdp->fifo_fnode[0];
422
423	/*
424	 * Its possible that fifo nodes on different lofs mountpoints
425	 * shadow the same real filesystem fifo node.
426	 * In this case its necessary to get and store the realvp.
427	 * This way different fifo nodes sharing the same real vnode
428	 * can use realvp for communication.
429	 */
430
431	if (VOP_REALVP(vp, &rvp, NULL) == 0)
432			vp = rvp;
433
434	fnp->fn_realvp	= vp;
435	fnp->fn_wcnt	= 0;
436	fnp->fn_rcnt	= 0;
437
438#if FIFODEBUG
439	if (! Fifo_fastmode) {
440		fnp->fn_flag	= 0;
441	} else {
442		fnp->fn_flag	= FIFOFAST;
443	}
444#else /* FIFODEBUG */
445	fnp->fn_flag	= FIFOFAST;
446#endif /* FIFODEBUG */
447
448	/*
449	 * initialize the times from vp.
450	 */
451	va.va_mask = AT_TIMES;
452	if (VOP_GETATTR(vp, &va, 0, crp, NULL) == 0) {
453		fnp->fn_atime = va.va_atime.tv_sec;
454		fnp->fn_mtime = va.va_mtime.tv_sec;
455		fnp->fn_ctime = va.va_ctime.tv_sec;
456	} else {
457		fnp->fn_atime = 0;
458		fnp->fn_mtime = 0;
459		fnp->fn_ctime = 0;
460	}
461
462	/*
463	 * Grab the VP here to avoid holding locks
464	 * whilst trying to acquire others.
465	 */
466
467	VN_HOLD(vp);
468
469	mutex_enter(&ftable_lock);
470
471	if ((spec_fnp = fifofind(vp)) != NULL) {
472		mutex_exit(&ftable_lock);
473
474		/*
475		 * Release the vnode and free up our pre-prepared fnode.
476		 * Zero the lock reference just to explicitly signal
477		 * this is unused.
478		 */
479		VN_RELE(vp);
480		fdp->fifo_lock.flk_ref = 0;
481		kmem_cache_free(fnode_cache, fdp);
482
483		return (FTOV(spec_fnp));
484	}
485
486	newvp = FTOV(fnp);
487	fifo_reinit_vp(newvp);
488	/*
489	 * Since the fifo vnode's v_vfsp needs to point to the
490	 * underlying filesystem's vfsp we need to bump up the
491	 * underlying filesystem's vfs reference count.
492	 * The count is decremented when the fifo node is
493	 * inactivated.
494	 */
495
496	VFS_HOLD(vp->v_vfsp);
497	newvp->v_vfsp = vp->v_vfsp;
498	newvp->v_rdev = vp->v_rdev;
499	newvp->v_flag |= (vp->v_flag & VROOT);
500
501	fifoinsert(fnp);
502	mutex_exit(&ftable_lock);
503
504	return (newvp);
505}
506
507/*
508 * Create a pipe end by...
509 * allocating a vnode-fifonode pair and initializing the fifonode.
510 */
511void
512makepipe(vnode_t **vpp1, vnode_t **vpp2)
513{
514	fifonode_t *fnp1;
515	fifonode_t *fnp2;
516	vnode_t *nvp1;
517	vnode_t *nvp2;
518	fifodata_t *fdp;
519	time_t now;
520
521	fdp = kmem_cache_alloc(pipe_cache, KM_SLEEP);
522	fdp->fifo_lock.flk_ref = 2;
523	fnp1 = &fdp->fifo_fnode[0];
524	fnp2 = &fdp->fifo_fnode[1];
525
526	fnp1->fn_wcnt	= fnp2->fn_wcnt		= 1;
527	fnp1->fn_rcnt	= fnp2->fn_rcnt		= 1;
528#if FIFODEBUG
529	if (! Fifo_fastmode) {
530		fnp1->fn_flag	= fnp2->fn_flag		= ISPIPE;
531	} else {
532		fnp1->fn_flag	= fnp2->fn_flag		= ISPIPE | FIFOFAST;
533	}
534#else /* FIFODEBUG */
535	fnp1->fn_flag	= fnp2->fn_flag		= ISPIPE | FIFOFAST;
536#endif /* FIFODEBUG */
537	now = gethrestime_sec();
538	fnp1->fn_atime	= fnp2->fn_atime	= now;
539	fnp1->fn_mtime	= fnp2->fn_mtime	= now;
540	fnp1->fn_ctime	= fnp2->fn_ctime	= now;
541
542	*vpp1 = nvp1 = FTOV(fnp1);
543	*vpp2 = nvp2 = FTOV(fnp2);
544
545	fifo_reinit_vp(nvp1);		/* Reinitialize vnodes for reuse... */
546	fifo_reinit_vp(nvp2);
547	nvp1->v_vfsp = fifovfsp; 	/* Need to re-establish VFS & device */
548	nvp2->v_vfsp = fifovfsp; 	/* before we can reuse this vnode. */
549	nvp1->v_rdev = fifodev;
550	nvp2->v_rdev = fifodev;
551}
552
553/*
554 * Attempt to establish a unique pipe id.  Only un-named pipes use this
555 * routine.
556 */
557ino_t
558fifogetid(void)
559{
560	static ino_t fifo_ino = 0;
561	ino_t fino;
562
563	mutex_enter(&fino_lock);
564	fino = fifo_ino++;
565	mutex_exit(&fino_lock);
566	return (fino);
567}
568
569
570/*
571 * Stream a pipe/FIFO.
572 * The FIFOCONNLD flag is used when CONNLD has been pushed on the stream.
573 * If the flag is set, a new vnode is created by calling fifo_connld().
574 * Connld logic was moved to fifo_connld() to speed up the open
575 * operation, simplify the connld/fifo interaction, and remove inherent
576 * race conditions between the connld module and fifos.
577 * This routine is single threaded for two reasons.
578 * 1) connld requests are synchronous; that is, they must block
579 *    until the server does an I_RECVFD (oh, well).  Single threading is
580 *    the simplest way to accomplish this.
581 * 2) fifo_close() must not send M_HANGUP or M_ERROR while we are
582 *    in stropen. Stropen() has a tendency to reset things and
583 *    we would like streams to remember that a hangup occurred.
584 */
585int
586fifo_stropen(vnode_t **vpp, int flag, cred_t *crp, int dotwist, int lockheld)
587{
588	int error = 0;
589	vnode_t *oldvp = *vpp;
590	fifonode_t *fnp = VTOF(*vpp);
591	dev_t pdev = 0;
592	int firstopen = 0;
593	fifolock_t *fn_lock;
594
595	fn_lock = fnp->fn_lock;
596	if (!lockheld)
597		mutex_enter(&fn_lock->flk_lock);
598	ASSERT(MUTEX_HELD(&fnp->fn_lock->flk_lock));
599
600	/*
601	 * FIFO is in the process of opening. Wait for it
602	 * to complete before starting another open on it
603	 * This prevents races associated with connld open
604	 */
605	while (fnp->fn_flag & FIFOOPEN) {
606		if (!cv_wait_sig(&fnp->fn_wait_cv, &fn_lock->flk_lock)) {
607			fifo_cleanup(oldvp, flag);
608			if (!lockheld)
609				mutex_exit(&fn_lock->flk_lock);
610			return (EINTR);
611		}
612	}
613
614	/*
615	 * The other end of the pipe is almost closed so
616	 * reject any other open on this end of the pipe
617	 * This only happens with a pipe mounted under namefs
618	 */
619	if ((fnp->fn_flag & (FIFOCLOSE|ISPIPE)) == (FIFOCLOSE|ISPIPE)) {
620		fifo_cleanup(oldvp, flag);
621		cv_broadcast(&fnp->fn_wait_cv);
622		if (!lockheld)
623			mutex_exit(&fn_lock->flk_lock);
624		return (ENXIO);
625	}
626
627	fnp->fn_flag |= FIFOOPEN;
628
629	/*
630	 * can't allow close to happen while we are
631	 * in the middle of stropen().
632	 * M_HANGUP and M_ERROR could leave the stream in a strange state
633	 */
634	while (fn_lock->flk_ocsync)
635		cv_wait(&fn_lock->flk_wait_cv, &fn_lock->flk_lock);
636
637	fn_lock->flk_ocsync = 1;
638
639	if (fnp->fn_flag & FIFOCONNLD) {
640		/*
641		 * This is a reopen, so we should release the fifo lock
642		 * just in case some strange module pushed on connld
643		 * has some odd side effect.
644		 * Note: this stropen is on the oldvp.  It will
645		 * have no impact on the connld vp returned and
646		 * strclose() will only be called when we release
647		 * flk_ocsync
648		 */
649		mutex_exit(&fn_lock->flk_lock);
650		if ((error = stropen(oldvp, &pdev, flag, crp)) != 0) {
651			mutex_enter(&fn_lock->flk_lock);
652			fifo_cleanup(oldvp, flag);
653			fn_lock->flk_ocsync = 0;
654			cv_broadcast(&fn_lock->flk_wait_cv);
655			goto out;
656		}
657		/*
658		 * streams open done, allow close on other end if
659		 * required.  Do this now.. it could
660		 * be a very long time before fifo_connld returns.
661		 */
662		mutex_enter(&fn_lock->flk_lock);
663		/*
664		 * we need to fake an open here so that if this
665		 * end of the pipe closes, we don't loose the
666		 * stream head (kind of like single threading
667		 * open and close for this end of the pipe)
668		 * We'll need to call fifo_close() to do clean
669		 * up in case this end of the pipe was closed
670		 * down while we were in fifo_connld()
671		 */
672		ASSERT(fnp->fn_open > 0);
673		fnp->fn_open++;
674		fn_lock->flk_ocsync = 0;
675		cv_broadcast(&fn_lock->flk_wait_cv);
676		mutex_exit(&fn_lock->flk_lock);
677		/*
678		 * Connld has been pushed onto the pipe
679		 * Create new pipe on behalf of connld
680		 */
681		if (error = fifo_connld(vpp, flag, crp)) {
682			(void) fifo_close(oldvp, flag, 1, 0, crp, NULL);
683			mutex_enter(&fn_lock->flk_lock);
684			goto out;
685		}
686		/*
687		 * undo fake open.  We need to call fifo_close
688		 * because some other thread could have done
689		 * a close and detach of the named pipe while
690		 * we were in fifo_connld(), so
691		 * we want to make sure the close completes (yuk)
692		 */
693		(void) fifo_close(oldvp, flag, 1, 0, crp, NULL);
694		/*
695		 * fifo_connld has changed the vp, so we
696		 * need to re-initialize locals
697		 */
698		fnp = VTOF(*vpp);
699		fn_lock = fnp->fn_lock;
700		mutex_enter(&fn_lock->flk_lock);
701	} else {
702		/*
703		 * release lock in case there are modules pushed that
704		 * could have some strange side effect
705		 */
706
707		mutex_exit(&fn_lock->flk_lock);
708
709		/*
710		 * If this is the first open of a fifo (dotwist
711		 * will be non-zero) we will need to twist the queues.
712		 */
713		if (oldvp->v_stream == NULL)
714			firstopen = 1;
715
716
717		/*
718		 * normal open of pipe/fifo
719		 */
720
721		if ((error = stropen(oldvp, &pdev, flag, crp)) != 0) {
722			mutex_enter(&fn_lock->flk_lock);
723			fifo_cleanup(oldvp, flag);
724			ASSERT(fnp->fn_open != 0 || oldvp->v_stream == NULL);
725			fn_lock->flk_ocsync = 0;
726			cv_broadcast(&fn_lock->flk_wait_cv);
727			goto out;
728		}
729		mutex_enter(&fn_lock->flk_lock);
730
731		/*
732		 * twist the ends of the fifo together
733		 */
734		if (dotwist && firstopen)
735			strmate(*vpp, *vpp);
736
737		/*
738		 * Show that this open has succeeded
739		 * and allow closes or other opens to proceed
740		 */
741		fnp->fn_open++;
742		fn_lock->flk_ocsync = 0;
743		cv_broadcast(&fn_lock->flk_wait_cv);
744	}
745out:
746	fnp->fn_flag &= ~FIFOOPEN;
747	if (error == 0) {
748		fnp->fn_flag |= FIFOISOPEN;
749		/*
750		 * If this is a FIFO and has the close flag set
751		 * and there are now writers, clear the close flag
752		 * Note: close flag only gets set when last writer
753		 * on a FIFO goes away.
754		 */
755		if (((fnp->fn_flag & (ISPIPE|FIFOCLOSE)) == FIFOCLOSE) &&
756		    fnp->fn_wcnt > 0)
757			fnp->fn_flag &= ~FIFOCLOSE;
758	}
759	cv_broadcast(&fnp->fn_wait_cv);
760	if (!lockheld)
761		mutex_exit(&fn_lock->flk_lock);
762	return (error);
763}
764
765/*
766 * Clean up the state of a FIFO and/or mounted pipe in the
767 * event that a fifo_open() was interrupted while the
768 * process was blocked.
769 */
770void
771fifo_cleanup(vnode_t *vp, int flag)
772{
773	fifonode_t *fnp = VTOF(vp);
774
775	ASSERT(MUTEX_HELD(&fnp->fn_lock->flk_lock));
776
777	cleanlocks(vp, curproc->p_pid, 0);
778	cleanshares(vp, curproc->p_pid);
779	if (flag & FREAD) {
780		fnp->fn_rcnt--;
781	}
782	if (flag & FWRITE) {
783		fnp->fn_wcnt--;
784	}
785	cv_broadcast(&fnp->fn_wait_cv);
786}
787
788
789/*
790 * Insert a fifonode-vnode pair onto the fifoalloc hash list.
791 */
792static void
793fifoinsert(fifonode_t *fnp)
794{
795	int idx = FIFOHASH(fnp->fn_realvp);
796
797	/*
798	 * We don't need to hold fn_lock since we're holding ftable_lock and
799	 * this routine is only called right after we've allocated an fnode.
800	 * FIFO is inserted at head of NULL terminated doubly linked list.
801	 */
802
803	ASSERT(MUTEX_HELD(&ftable_lock));
804	fnp->fn_backp = NULL;
805	fnp->fn_nextp = fifoalloc[idx];
806	fifoalloc[idx] = fnp;
807	if (fnp->fn_nextp)
808		fnp->fn_nextp->fn_backp = fnp;
809}
810
811/*
812 * Find a fifonode-vnode pair on the fifoalloc hash list.
813 * vp is a vnode to be shadowed. If it's on the hash list,
814 * it already has a shadow, therefore return its corresponding
815 * fifonode.
816 */
817static fifonode_t *
818fifofind(vnode_t *vp)
819{
820	fifonode_t *fnode;
821
822	ASSERT(MUTEX_HELD(&ftable_lock));
823	for (fnode = fifoalloc[FIFOHASH(vp)]; fnode; fnode = fnode->fn_nextp) {
824		if (fnode->fn_realvp == vp) {
825			VN_HOLD(FTOV(fnode));
826			return (fnode);
827		}
828	}
829	return (NULL);
830}
831
832/*
833 * Remove a fifonode-vnode pair from the fifoalloc hash list.
834 * This routine is called from the fifo_inactive() routine when a
835 * FIFO is being released.
836 * If the link to be removed is the only link, set fifoalloc to NULL.
837 */
838void
839fiforemove(fifonode_t *fnp)
840{
841	int idx = FIFOHASH(fnp->fn_realvp);
842	fifonode_t *fnode;
843
844	ASSERT(MUTEX_HELD(&ftable_lock));
845	fnode = fifoalloc[idx];
846	/*
847	 * fast path... only 1 FIFO in this list entry
848	 */
849	if (fnode != NULL && fnode == fnp &&
850	    !fnode->fn_nextp && !fnode->fn_backp) {
851		fifoalloc[idx] = NULL;
852	} else {
853
854		for (;  fnode;  fnode = fnode->fn_nextp) {
855			if (fnode == fnp) {
856				/*
857				 * if we are first entry
858				 */
859				if (fnp == fifoalloc[idx])
860					fifoalloc[idx] = fnp->fn_nextp;
861				if (fnode->fn_nextp)
862					fnode->fn_nextp->fn_backp =
863					    fnode->fn_backp;
864				if (fnode->fn_backp)
865					fnode->fn_backp->fn_nextp =
866					    fnode->fn_nextp;
867				break;
868			}
869		}
870	}
871}
872
873/*
874 * Flush all data from a fifo's message queue
875 */
876
877void
878fifo_fastflush(fifonode_t *fnp)
879{
880	mblk_t *bp;
881	ASSERT(MUTEX_HELD(&fnp->fn_lock->flk_lock));
882
883	if ((bp = fnp->fn_mp) != NULL) {
884		fnp->fn_mp = NULL;
885		fnp->fn_count = 0;
886		freemsg(bp);
887	}
888	fifo_wakewriter(fnp->fn_dest, fnp->fn_lock);
889}
890
891/*
892 * Note:  This routine is single threaded
893 *  Protected by FIFOOPEN flag (i.e. flk_lock is not held)
894 *  Upon successful completion, the original fifo is unlocked
895 *  and FIFOOPEN is cleared for the original vpp.
896 *  The new fifo returned has FIFOOPEN set.
897 */
898static int
899fifo_connld(struct vnode **vpp, int flag, cred_t *crp)
900{
901	struct vnode *vp1;
902	struct vnode *vp2;
903	struct fifonode *oldfnp;
904	struct fifonode *fn_dest;
905	int error;
906	struct file *filep;
907	struct fifolock *fn_lock;
908	cred_t *c;
909
910	/*
911	 * Get two vnodes that will represent the pipe ends for the new pipe.
912	 */
913	makepipe(&vp1, &vp2);
914
915	/*
916	 * Allocate a file descriptor and file pointer for one of the pipe
917	 * ends. The file descriptor will be used to send that pipe end to
918	 * the process on the other end of this stream. Note that we get
919	 * the file structure only, there is no file list entry allocated.
920	 */
921	if (error = falloc(vp1, FWRITE|FREAD, &filep, NULL)) {
922		VN_RELE(vp1);
923		VN_RELE(vp2);
924		return (error);
925	}
926	mutex_exit(&filep->f_tlock);
927	oldfnp = VTOF(*vpp);
928	fn_lock = oldfnp->fn_lock;
929	fn_dest = oldfnp->fn_dest;
930
931	/*
932	 * Create two new stream heads and attach them to the two vnodes for
933	 * the new pipe.
934	 */
935	if ((error = fifo_stropen(&vp1, FREAD|FWRITE, filep->f_cred, 0, 0)) !=
936	    0 ||
937	    (error = fifo_stropen(&vp2, flag, filep->f_cred, 0, 0)) != 0) {
938#if DEBUG
939		cmn_err(CE_NOTE, "fifo stropen failed error 0x%x", error);
940#endif
941		/*
942		 * this will call fifo_close and VN_RELE on vp1
943		 */
944		(void) closef(filep);
945		VN_RELE(vp2);
946		return (error);
947	}
948
949	/*
950	 * twist the ends of the pipe together
951	 */
952	strmate(vp1, vp2);
953
954	/*
955	 * Set our end to busy in open
956	 * Note: Don't need lock around this because we're the only
957	 * one who knows about it
958	 */
959	VTOF(vp2)->fn_flag |= FIFOOPEN;
960
961	mutex_enter(&fn_lock->flk_lock);
962
963	fn_dest->fn_flag |= FIFOSEND;
964	/*
965	 * check to make sure neither end of pipe has gone away
966	 */
967	if (!(fn_dest->fn_flag & FIFOISOPEN)) {
968		error = ENXIO;
969		fn_dest->fn_flag &= ~FIFOSEND;
970		mutex_exit(&fn_lock->flk_lock);
971		/*
972		 * this will call fifo_close and VN_RELE on vp1
973		 */
974		goto out;
975	}
976	mutex_exit(&fn_lock->flk_lock);
977
978	/*
979	 * Tag the sender's credential on the pipe descriptor.
980	 */
981	crhold(VTOF(vp1)->fn_pcredp = crp);
982	VTOF(vp1)->fn_cpid = curproc->p_pid;
983
984	/*
985	 * send the file descriptor to other end of pipe
986	 */
987	if (error = do_sendfp((*vpp)->v_stream, filep, crp)) {
988		mutex_enter(&fn_lock->flk_lock);
989		fn_dest->fn_flag &= ~FIFOSEND;
990		mutex_exit(&fn_lock->flk_lock);
991		/*
992		 * this will call fifo_close and VN_RELE on vp1
993		 */
994		goto out;
995	}
996
997	mutex_enter(&fn_lock->flk_lock);
998	/*
999	 * Wait for other end to receive file descriptor
1000	 * FIFOCLOSE indicates that one or both sides of the pipe
1001	 * have gone away.
1002	 */
1003	while ((fn_dest->fn_flag & (FIFOCLOSE | FIFOSEND)) == FIFOSEND) {
1004		if (!cv_wait_sig(&oldfnp->fn_wait_cv, &fn_lock->flk_lock)) {
1005			error = EINTR;
1006			fn_dest->fn_flag &= ~FIFOSEND;
1007			mutex_exit(&fn_lock->flk_lock);
1008			goto out;
1009		}
1010	}
1011	/*
1012	 * If either end of pipe has gone away and the other end did not
1013	 * receive pipe, reject the connld open
1014	 */
1015	if ((fn_dest->fn_flag & FIFOSEND)) {
1016		error = ENXIO;
1017		fn_dest->fn_flag &= ~FIFOSEND;
1018		mutex_exit(&fn_lock->flk_lock);
1019		goto out;
1020	}
1021
1022	oldfnp->fn_flag &= ~FIFOOPEN;
1023	cv_broadcast(&oldfnp->fn_wait_cv);
1024	mutex_exit(&fn_lock->flk_lock);
1025
1026	VN_RELE(*vpp);
1027	*vpp = vp2;
1028	(void) closef(filep);
1029	return (0);
1030out:
1031	c = filep->f_cred;
1032	crhold(c);
1033	(void) closef(filep);
1034	VTOF(vp2)->fn_flag &= ~FIFOOPEN;
1035	(void) fifo_close(vp2, flag, 1, (offset_t)0, c, NULL);
1036	crfree(c);
1037	VN_RELE(vp2);
1038	return (error);
1039}
1040
1041/*
1042 * Disable fastpath mode.
1043 */
1044void
1045fifo_fastoff(fifonode_t *fnp)
1046{
1047	ASSERT(MUTEX_HELD(&fnp->fn_lock->flk_lock));
1048	ASSERT(FTOV(fnp)->v_stream);
1049
1050	/* FIFOSTAYFAST is set => FIFOFAST is set */
1051	while ((fnp->fn_flag & FIFOSTAYFAST) || ((fnp->fn_flag & ISPIPE) &&
1052	    (fnp->fn_dest->fn_flag & FIFOSTAYFAST))) {
1053		ASSERT(fnp->fn_flag & FIFOFAST);
1054		/* indicate someone is waiting to turn into stream mode */
1055		fnp->fn_flag |= FIFOWAITMODE;
1056		cv_wait(&fnp->fn_wait_cv, &fnp->fn_lock->flk_lock);
1057		fnp->fn_flag &= ~FIFOWAITMODE;
1058	}
1059
1060	/* as we may have relased the lock, test the FIFOFAST flag here */
1061	if (!(fnp->fn_flag & FIFOFAST))
1062		return;
1063#if FIFODEBUG
1064	if (Fifo_verbose)
1065		cmn_err(CE_NOTE, "Fifo reverting to streams mode\n");
1066#endif
1067
1068	fifo_fastturnoff(fnp);
1069	if (fnp->fn_flag & ISPIPE) {
1070		fifo_fastturnoff(fnp->fn_dest);
1071	}
1072}
1073
1074
1075/*
1076 * flk_lock must be held while calling fifo_fastturnoff() to
1077 * preserve data ordering (no reads or writes allowed)
1078 */
1079
1080static void
1081fifo_fastturnoff(fifonode_t *fnp)
1082{
1083	fifonode_t *fn_dest = fnp->fn_dest;
1084	mblk_t	*fn_mp;
1085	int	fn_flag;
1086
1087	ASSERT(MUTEX_HELD(&fnp->fn_lock->flk_lock));
1088	/*
1089	 * Note: This end can't be closed if there
1090	 * is stuff in fn_mp
1091	 */
1092	if ((fn_mp = fnp->fn_mp) != NULL) {
1093		ASSERT(fnp->fn_flag & FIFOISOPEN);
1094		ASSERT(FTOV(fnp)->v_stream != NULL);
1095		ASSERT(FTOV(fnp)->v_stream->sd_wrq != NULL);
1096		ASSERT(RD(FTOV(fnp)->v_stream->sd_wrq) != NULL);
1097		ASSERT(strvp2wq(FTOV(fnp)) != NULL);
1098		fnp->fn_mp = NULL;
1099		fnp->fn_count = 0;
1100		/*
1101		 * Don't need to drop flk_lock across the put()
1102		 * since we're just moving the message from the fifo
1103		 * node to the STREAM head...
1104		 */
1105		put(RD(strvp2wq(FTOV(fnp))), fn_mp);
1106	}
1107
1108	/*
1109	 * Need to re-issue any pending poll requests
1110	 * so that the STREAMS framework sees them
1111	 * Writers would be waiting on fnp and readers on fn_dest
1112	 */
1113	if ((fnp->fn_flag & (FIFOISOPEN | FIFOPOLLW)) ==
1114	    (FIFOISOPEN | FIFOPOLLW)) {
1115		strpollwakeup(FTOV(fnp), POLLWRNORM);
1116	}
1117	fn_flag = fn_dest->fn_flag;
1118	if ((fn_flag & FIFOISOPEN) == FIFOISOPEN) {
1119		if ((fn_flag & (FIFOPOLLR | FIFOPOLLRBAND))) {
1120			strpollwakeup(FTOV(fn_dest), POLLIN|POLLRDNORM);
1121		}
1122	}
1123	/*
1124	 * wake up any sleeping processes so they can notice we went
1125	 * to streams mode
1126	 */
1127	fnp->fn_flag &= ~(FIFOFAST|FIFOWANTW|FIFOWANTR);
1128	cv_broadcast(&fnp->fn_wait_cv);
1129}
1130
1131/*
1132 * Alternative version of fifo_fastoff()
1133 * optimized for putmsg/getmsg.
1134 */
1135void
1136fifo_vfastoff(vnode_t *vp)
1137{
1138	fifonode_t	*fnp = VTOF(vp);
1139
1140	mutex_enter(&fnp->fn_lock->flk_lock);
1141	if (!(fnp->fn_flag & FIFOFAST)) {
1142		mutex_exit(&fnp->fn_lock->flk_lock);
1143		return;
1144	}
1145	fifo_fastoff(fnp);
1146	mutex_exit(&fnp->fn_lock->flk_lock);
1147}
1148
1149/*
1150 * Wake any sleeping writers, poll and send signals if necessary
1151 * This module is only called when we drop below the hi water mark
1152 * FIFOWANTW indicates that a process is sleeping in fifo_write()
1153 * FIFOHIWATW indicates that we have either attempted a poll or
1154 * non-blocking write and were over the high water mark
1155 * This routine assumes a low water mark of 0.
1156 */
1157
1158void
1159fifo_wakewriter(fifonode_t *fn_dest, fifolock_t *fn_lock)
1160{
1161	int fn_dflag = fn_dest->fn_flag;
1162
1163	ASSERT(MUTEX_HELD(&fn_lock->flk_lock));
1164	ASSERT(fn_dest->fn_dest->fn_count < Fifohiwat);
1165	if ((fn_dflag & FIFOWANTW)) {
1166		cv_broadcast(&fn_dest->fn_wait_cv);
1167	}
1168	if ((fn_dflag & (FIFOHIWATW | FIFOISOPEN)) ==
1169	    (FIFOHIWATW | FIFOISOPEN)) {
1170		if (fn_dflag & FIFOPOLLW)
1171			strpollwakeup(FTOV(fn_dest), POLLWRNORM);
1172		if (fn_dflag & FIFOSETSIG)
1173			str_sendsig(FTOV(fn_dest), S_WRNORM, 0, 0);
1174	}
1175	/*
1176	 * FIFOPOLLW can't be set without setting FIFOHIWAT
1177	 * This allows us to clear both here.
1178	 */
1179	fn_dest->fn_flag = fn_dflag & ~(FIFOWANTW | FIFOHIWATW | FIFOPOLLW);
1180}
1181
1182/*
1183 * wake up any sleeping readers, poll or send signal if needed
1184 * FIFOWANTR indicates that a process is waiting in fifo_read() for data
1185 * FIFOSETSIG indicates that SIGPOLL should be sent to process
1186 * FIFOPOLLR indicates that a poll request for reading on the fifo was made
1187 */
1188
1189void
1190fifo_wakereader(fifonode_t *fn_dest, fifolock_t *fn_lock)
1191{
1192	int fn_dflag = fn_dest->fn_flag;
1193
1194	ASSERT(MUTEX_HELD(&fn_lock->flk_lock));
1195	if (fn_dflag & FIFOWANTR) {
1196		cv_broadcast(&fn_dest->fn_wait_cv);
1197	}
1198	if (fn_dflag & FIFOISOPEN) {
1199		if (fn_dflag & FIFOPOLLR)
1200			strpollwakeup(FTOV(fn_dest), POLLIN | POLLRDNORM);
1201		if (fn_dflag & FIFOSETSIG)
1202			str_sendsig(FTOV(fn_dest), S_INPUT | S_RDNORM, 0, 0);
1203	}
1204	fn_dest->fn_flag = fn_dflag & ~(FIFOWANTR | FIFOPOLLR);
1205}
1206