xref: /illumos-gate/usr/src/uts/common/fs/vfs.c (revision 9164a50bf932130cbb5097a16f6986873ce0e6e5)
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 
22 /*
23  * Copyright (c) 1988, 2010, Oracle and/or its affiliates. All rights reserved.
24  * Copyright (c) 2019, Joyent, Inc.
25  * Copyright 2016 Toomas Soome <tsoome@me.com>
26  * Copyright (c) 2016, 2017 by Delphix. All rights reserved.
27  * Copyright 2016 Nexenta Systems, Inc.
28  * Copyright 2017 RackTop Systems.
29  * Copyright 2024 Oxide Computer Company
30  */
31 
32 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
33 /*	  All Rights Reserved	*/
34 
35 /*
36  * University Copyright- Copyright (c) 1982, 1986, 1988
37  * The Regents of the University of California
38  * All Rights Reserved
39  *
40  * University Acknowledgment- Portions of this document are derived from
41  * software developed by the University of California, Berkeley, and its
42  * contributors.
43  */
44 
45 #include <sys/types.h>
46 #include <sys/t_lock.h>
47 #include <sys/param.h>
48 #include <sys/errno.h>
49 #include <sys/user.h>
50 #include <sys/fstyp.h>
51 #include <sys/kmem.h>
52 #include <sys/systm.h>
53 #include <sys/proc.h>
54 #include <sys/mount.h>
55 #include <sys/vfs.h>
56 #include <sys/vfs_opreg.h>
57 #include <sys/fem.h>
58 #include <sys/mntent.h>
59 #include <sys/stat.h>
60 #include <sys/statvfs.h>
61 #include <sys/statfs.h>
62 #include <sys/cred.h>
63 #include <sys/vnode.h>
64 #include <sys/rwstlock.h>
65 #include <sys/dnlc.h>
66 #include <sys/file.h>
67 #include <sys/time.h>
68 #include <sys/atomic.h>
69 #include <sys/cmn_err.h>
70 #include <sys/buf.h>
71 #include <sys/swap.h>
72 #include <sys/debug.h>
73 #include <sys/vnode.h>
74 #include <sys/modctl.h>
75 #include <sys/ddi.h>
76 #include <sys/pathname.h>
77 #include <sys/bootconf.h>
78 #include <sys/dumphdr.h>
79 #include <sys/dc_ki.h>
80 #include <sys/poll.h>
81 #include <sys/sunddi.h>
82 #include <sys/sysmacros.h>
83 #include <sys/zone.h>
84 #include <sys/policy.h>
85 #include <sys/ctfs.h>
86 #include <sys/objfs.h>
87 #include <sys/console.h>
88 #include <sys/reboot.h>
89 #include <sys/attr.h>
90 #include <sys/zio.h>
91 #include <sys/spa.h>
92 #include <sys/lofi.h>
93 #include <sys/bootprops.h>
94 
95 #include <vm/page.h>
96 
97 #include <fs/fs_subr.h>
98 /* Private interfaces to create vopstats-related data structures */
99 extern void		initialize_vopstats(vopstats_t *);
100 extern vopstats_t	*get_fstype_vopstats(struct vfs *, struct vfssw *);
101 extern vsk_anchor_t	*get_vskstat_anchor(struct vfs *);
102 
103 static void vfs_clearmntopt_nolock(mntopts_t *, const char *, int);
104 static void vfs_setmntopt_nolock(mntopts_t *, const char *,
105     const char *, int, int);
106 static int  vfs_optionisset_nolock(const mntopts_t *, const char *, char **);
107 static void vfs_freemnttab(struct vfs *);
108 static void vfs_freeopt(mntopt_t *);
109 static void vfs_swapopttbl_nolock(mntopts_t *, mntopts_t *);
110 static void vfs_swapopttbl(mntopts_t *, mntopts_t *);
111 static void vfs_copyopttbl_extend(const mntopts_t *, mntopts_t *, int);
112 static void vfs_createopttbl_extend(mntopts_t *, const char *,
113     const mntopts_t *);
114 static char **vfs_copycancelopt_extend(char **const, int);
115 static void vfs_freecancelopt(char **);
116 static void getrootfs(char **, char **);
117 static int getmacpath(dev_info_t *, void *);
118 static void vfs_mnttabvp_setup(void);
119 
120 struct ipmnt {
121 	struct ipmnt	*mip_next;
122 	dev_t		mip_dev;
123 	struct vfs	*mip_vfsp;
124 };
125 
126 static kmutex_t		vfs_miplist_mutex;
127 static struct ipmnt	*vfs_miplist = NULL;
128 static struct ipmnt	*vfs_miplist_end = NULL;
129 
130 static kmem_cache_t *vfs_cache;	/* Pointer to VFS kmem cache */
131 
132 /*
133  * VFS global data.
134  */
135 vnode_t *rootdir;		/* pointer to root inode vnode. */
136 vnode_t *devicesdir;		/* pointer to inode of devices root */
137 vnode_t	*devdir;		/* pointer to inode of dev root */
138 
139 char *server_rootpath;		/* root path for diskless clients */
140 char *server_hostname;		/* hostname of diskless server */
141 
142 static struct vfs root;
143 static struct vfs devices;
144 static struct vfs dev;
145 struct vfs *rootvfs = &root;	/* pointer to root vfs; head of VFS list. */
146 rvfs_t *rvfs_list;		/* array of vfs ptrs for vfs hash list */
147 int vfshsz = 512;		/* # of heads/locks in vfs hash arrays */
148 				/* must be power of 2!	*/
149 timespec_t vfs_mnttab_ctime;	/* mnttab created time */
150 timespec_t vfs_mnttab_mtime;	/* mnttab last modified time */
151 char *vfs_dummyfstype = "\0";
152 struct pollhead vfs_pollhd;	/* for mnttab pollers */
153 struct vnode *vfs_mntdummyvp;	/* to fake mnttab read/write for file events */
154 int	mntfstype;		/* will be set once mnt fs is mounted */
155 
156 /*
157  * Table for generic options recognized in the VFS layer and acted
158  * on at this level before parsing file system specific options.
159  * The nosuid option is stronger than any of the devices and setuid
160  * options, so those are canceled when nosuid is seen.
161  *
162  * All options which are added here need to be added to the
163  * list of standard options in usr/src/cmd/fs.d/fslib.c as well.
164  */
165 /*
166  * VFS Mount options table
167  */
168 static char *ro_cancel[] = { MNTOPT_RW, NULL };
169 static char *rw_cancel[] = { MNTOPT_RO, NULL };
170 static char *suid_cancel[] = { MNTOPT_NOSUID, NULL };
171 static char *nosuid_cancel[] = { MNTOPT_SUID, MNTOPT_DEVICES, MNTOPT_NODEVICES,
172     MNTOPT_NOSETUID, MNTOPT_SETUID, NULL };
173 static char *devices_cancel[] = { MNTOPT_NODEVICES, NULL };
174 static char *nodevices_cancel[] = { MNTOPT_DEVICES, NULL };
175 static char *setuid_cancel[] = { MNTOPT_NOSETUID, NULL };
176 static char *nosetuid_cancel[] = { MNTOPT_SETUID, NULL };
177 static char *nbmand_cancel[] = { MNTOPT_NONBMAND, NULL };
178 static char *nonbmand_cancel[] = { MNTOPT_NBMAND, NULL };
179 static char *exec_cancel[] = { MNTOPT_NOEXEC, NULL };
180 static char *noexec_cancel[] = { MNTOPT_EXEC, NULL };
181 
182 static const mntopt_t mntopts[] = {
183 /*
184  *	option name		cancel options		default arg	flags
185  */
186 	{ MNTOPT_REMOUNT,	NULL,			NULL,
187 		MO_NODISPLAY, (void *)0 },
188 	{ MNTOPT_RO,		ro_cancel,		NULL,		0,
189 		(void *)0 },
190 	{ MNTOPT_RW,		rw_cancel,		NULL,		0,
191 		(void *)0 },
192 	{ MNTOPT_SUID,		suid_cancel,		NULL,		0,
193 		(void *)0 },
194 	{ MNTOPT_NOSUID,	nosuid_cancel,		NULL,		0,
195 		(void *)0 },
196 	{ MNTOPT_DEVICES,	devices_cancel,		NULL,		0,
197 		(void *)0 },
198 	{ MNTOPT_NODEVICES,	nodevices_cancel,	NULL,		0,
199 		(void *)0 },
200 	{ MNTOPT_SETUID,	setuid_cancel,		NULL,		0,
201 		(void *)0 },
202 	{ MNTOPT_NOSETUID,	nosetuid_cancel,	NULL,		0,
203 		(void *)0 },
204 	{ MNTOPT_NBMAND,	nbmand_cancel,		NULL,		0,
205 		(void *)0 },
206 	{ MNTOPT_NONBMAND,	nonbmand_cancel,	NULL,		0,
207 		(void *)0 },
208 	{ MNTOPT_EXEC,		exec_cancel,		NULL,		0,
209 		(void *)0 },
210 	{ MNTOPT_NOEXEC,	noexec_cancel,		NULL,		0,
211 		(void *)0 },
212 };
213 
214 const mntopts_t vfs_mntopts = {
215 	sizeof (mntopts) / sizeof (mntopt_t),
216 	(mntopt_t *)&mntopts[0]
217 };
218 
219 /*
220  * File system operation dispatch functions.
221  */
222 
223 int
224 fsop_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
225 {
226 	return (*(vfsp)->vfs_op->vfs_mount)(vfsp, mvp, uap, cr);
227 }
228 
229 int
230 fsop_unmount(vfs_t *vfsp, int flag, cred_t *cr)
231 {
232 	return (*(vfsp)->vfs_op->vfs_unmount)(vfsp, flag, cr);
233 }
234 
235 int
236 fsop_root(vfs_t *vfsp, vnode_t **vpp)
237 {
238 	refstr_t *mntpt;
239 	int ret = (*(vfsp)->vfs_op->vfs_root)(vfsp, vpp);
240 	/*
241 	 * Make sure this root has a path.  With lofs, it is possible to have
242 	 * a NULL mountpoint.
243 	 */
244 	if (ret == 0 && vfsp->vfs_mntpt != NULL &&
245 	    (*vpp)->v_path == vn_vpath_empty) {
246 		const char *path;
247 
248 		mntpt = vfs_getmntpoint(vfsp);
249 		path = refstr_value(mntpt);
250 		vn_setpath_str(*vpp, path, strlen(path));
251 		refstr_rele(mntpt);
252 	}
253 
254 	return (ret);
255 }
256 
257 int
258 fsop_statfs(vfs_t *vfsp, statvfs64_t *sp)
259 {
260 	return (*(vfsp)->vfs_op->vfs_statvfs)(vfsp, sp);
261 }
262 
263 int
264 fsop_sync(vfs_t *vfsp, short flag, cred_t *cr)
265 {
266 	return (*(vfsp)->vfs_op->vfs_sync)(vfsp, flag, cr);
267 }
268 
269 int
270 fsop_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
271 {
272 	/*
273 	 * In order to handle system attribute fids in a manner
274 	 * transparent to the underlying fs, we embed the fid for
275 	 * the sysattr parent object in the sysattr fid and tack on
276 	 * some extra bytes that only the sysattr layer knows about.
277 	 *
278 	 * This guarantees that sysattr fids are larger than other fids
279 	 * for this vfs. If the vfs supports the sysattr view interface
280 	 * (as indicated by VFSFT_SYSATTR_VIEWS), we cannot have a size
281 	 * collision with XATTR_FIDSZ.
282 	 */
283 	if (vfs_has_feature(vfsp, VFSFT_SYSATTR_VIEWS) &&
284 	    fidp->fid_len == XATTR_FIDSZ)
285 		return (xattr_dir_vget(vfsp, vpp, fidp));
286 
287 	return (*(vfsp)->vfs_op->vfs_vget)(vfsp, vpp, fidp);
288 }
289 
290 int
291 fsop_mountroot(vfs_t *vfsp, enum whymountroot reason)
292 {
293 	return (*(vfsp)->vfs_op->vfs_mountroot)(vfsp, reason);
294 }
295 
296 void
297 fsop_freefs(vfs_t *vfsp)
298 {
299 	(*(vfsp)->vfs_op->vfs_freevfs)(vfsp);
300 }
301 
302 int
303 fsop_vnstate(vfs_t *vfsp, vnode_t *vp, vntrans_t nstate)
304 {
305 	return ((*(vfsp)->vfs_op->vfs_vnstate)(vfsp, vp, nstate));
306 }
307 
308 int
309 fsop_sync_by_kind(int fstype, short flag, cred_t *cr)
310 {
311 	ASSERT((fstype >= 0) && (fstype < nfstype));
312 
313 	if (ALLOCATED_VFSSW(&vfssw[fstype]) && VFS_INSTALLED(&vfssw[fstype]))
314 		return (*vfssw[fstype].vsw_vfsops.vfs_sync) (NULL, flag, cr);
315 	else
316 		return (ENOTSUP);
317 }
318 
319 int
320 fsop_syncfs(vfs_t *vfsp, uint64_t flags, cred_t *cr)
321 {
322 	return (*(vfsp)->vfs_op->vfs_syncfs)(vfsp, flags, cr);
323 }
324 
325 /*
326  * File system initialization.  vfs_setfsops() must be called from a file
327  * system's init routine.
328  */
329 
330 static int
331 fs_copyfsops(const fs_operation_def_t *template, vfsops_t *actual,
332     int *unused_ops)
333 {
334 	static const fs_operation_trans_def_t vfs_ops_table[] = {
335 		{ VFSNAME_MOUNT, offsetof(vfsops_t, vfs_mount),
336 		    fs_nosys, fs_nosys },
337 
338 		{ VFSNAME_UNMOUNT, offsetof(vfsops_t, vfs_unmount),
339 		    fs_nosys, fs_nosys },
340 
341 		{ VFSNAME_ROOT, offsetof(vfsops_t, vfs_root),
342 		    fs_nosys, fs_nosys },
343 
344 		{ VFSNAME_STATVFS, offsetof(vfsops_t, vfs_statvfs),
345 		    fs_nosys, fs_nosys },
346 
347 		{ VFSNAME_SYNC, offsetof(vfsops_t, vfs_sync),
348 		    (fs_generic_func_p) fs_sync,
349 		    (fs_generic_func_p) fs_sync },	/* No errors allowed */
350 
351 		{ VFSNAME_VGET, offsetof(vfsops_t, vfs_vget),
352 		    fs_nosys, fs_nosys },
353 
354 		{ VFSNAME_MOUNTROOT, offsetof(vfsops_t, vfs_mountroot),
355 		    fs_nosys, fs_nosys },
356 
357 		{ VFSNAME_FREEVFS, offsetof(vfsops_t, vfs_freevfs),
358 		    (fs_generic_func_p)(uintptr_t)fs_freevfs,
359 		    /* Shouldn't fail */
360 		    (fs_generic_func_p)(uintptr_t)fs_freevfs },
361 
362 		{ VFSNAME_VNSTATE, offsetof(vfsops_t, vfs_vnstate),
363 		    (fs_generic_func_p)fs_nosys, (fs_generic_func_p)fs_nosys },
364 
365 		/*
366 		 * While it is tempting to say that a file system which does not
367 		 * implement a VFSNAME_SYNC likely doesn't need a VFSNAME_SYNCFS
368 		 * by default, implementing that policy is challenging with the
369 		 * way the fs_build_vector logic works and we'd rather a file
370 		 * system say that it doesn't support this by default rather
371 		 * than incorrectly claim to sync something that either doesn't
372 		 * make sense to sync (ala sockfs) or mislead when it didn't
373 		 * happen.
374 		 */
375 		{ VFSNAME_SYNCFS, offsetof(vfsops_t, vfs_syncfs),
376 		    (fs_generic_func_p)fs_nosys_syncfs,
377 		    (fs_generic_func_p)fs_nosys_syncfs },
378 
379 		{ NULL, 0, NULL, NULL }
380 	};
381 
382 	return (fs_build_vector(actual, unused_ops, vfs_ops_table, template));
383 }
384 
385 void
386 zfs_boot_init(void)
387 {
388 	if (strcmp(rootfs.bo_fstype, MNTTYPE_ZFS) == 0)
389 		spa_boot_init();
390 }
391 
392 int
393 vfs_setfsops(int fstype, const fs_operation_def_t *template, vfsops_t **actual)
394 {
395 	int error;
396 	int unused_ops;
397 
398 	/*
399 	 * Verify that fstype refers to a valid fs.  Note that
400 	 * 0 is valid since it's used to set "stray" ops.
401 	 */
402 	if ((fstype < 0) || (fstype >= nfstype))
403 		return (EINVAL);
404 
405 	if (!ALLOCATED_VFSSW(&vfssw[fstype]))
406 		return (EINVAL);
407 
408 	/* Set up the operations vector. */
409 
410 	error = fs_copyfsops(template, &vfssw[fstype].vsw_vfsops, &unused_ops);
411 
412 	if (error != 0)
413 		return (error);
414 
415 	vfssw[fstype].vsw_flag |= VSW_INSTALLED;
416 
417 	if (actual != NULL)
418 		*actual = &vfssw[fstype].vsw_vfsops;
419 
420 #if DEBUG
421 	if (unused_ops != 0)
422 		cmn_err(CE_WARN, "vfs_setfsops: %s: %d operations supplied "
423 		    "but not used", vfssw[fstype].vsw_name, unused_ops);
424 #endif
425 
426 	return (0);
427 }
428 
429 int
430 vfs_makefsops(const fs_operation_def_t *template, vfsops_t **actual)
431 {
432 	int error;
433 	int unused_ops;
434 
435 	*actual = (vfsops_t *)kmem_alloc(sizeof (vfsops_t), KM_SLEEP);
436 
437 	error = fs_copyfsops(template, *actual, &unused_ops);
438 	if (error != 0) {
439 		kmem_free(*actual, sizeof (vfsops_t));
440 		*actual = NULL;
441 		return (error);
442 	}
443 
444 	return (0);
445 }
446 
447 /*
448  * Free a vfsops structure created as a result of vfs_makefsops().
449  * NOTE: For a vfsops structure initialized by vfs_setfsops(), use
450  * vfs_freevfsops_by_type().
451  */
452 void
453 vfs_freevfsops(vfsops_t *vfsops)
454 {
455 	kmem_free(vfsops, sizeof (vfsops_t));
456 }
457 
458 /*
459  * Since the vfsops structure is part of the vfssw table and wasn't
460  * really allocated, we're not really freeing anything.  We keep
461  * the name for consistency with vfs_freevfsops().  We do, however,
462  * need to take care of a little bookkeeping.
463  * NOTE: For a vfsops structure created by vfs_setfsops(), use
464  * vfs_freevfsops_by_type().
465  */
466 int
467 vfs_freevfsops_by_type(int fstype)
468 {
469 
470 	/* Verify that fstype refers to a loaded fs (and not fsid 0). */
471 	if ((fstype <= 0) || (fstype >= nfstype))
472 		return (EINVAL);
473 
474 	WLOCK_VFSSW();
475 	if ((vfssw[fstype].vsw_flag & VSW_INSTALLED) == 0) {
476 		WUNLOCK_VFSSW();
477 		return (EINVAL);
478 	}
479 
480 	vfssw[fstype].vsw_flag &= ~VSW_INSTALLED;
481 	WUNLOCK_VFSSW();
482 
483 	return (0);
484 }
485 
486 /* Support routines used to reference vfs_op */
487 
488 /* Set the operations vector for a vfs */
489 void
490 vfs_setops(vfs_t *vfsp, vfsops_t *vfsops)
491 {
492 	vfsops_t	*op;
493 
494 	ASSERT(vfsp != NULL);
495 	ASSERT(vfsops != NULL);
496 
497 	op = vfsp->vfs_op;
498 	membar_consumer();
499 	if (vfsp->vfs_femhead == NULL &&
500 	    atomic_cas_ptr(&vfsp->vfs_op, op, vfsops) == op) {
501 		return;
502 	}
503 	fsem_setvfsops(vfsp, vfsops);
504 }
505 
506 /* Retrieve the operations vector for a vfs */
507 vfsops_t *
508 vfs_getops(vfs_t *vfsp)
509 {
510 	vfsops_t	*op;
511 
512 	ASSERT(vfsp != NULL);
513 
514 	op = vfsp->vfs_op;
515 	membar_consumer();
516 	if (vfsp->vfs_femhead == NULL && op == vfsp->vfs_op) {
517 		return (op);
518 	} else {
519 		return (fsem_getvfsops(vfsp));
520 	}
521 }
522 
523 /*
524  * Returns non-zero (1) if the vfsops matches that of the vfs.
525  * Returns zero (0) if not.
526  */
527 int
528 vfs_matchops(vfs_t *vfsp, vfsops_t *vfsops)
529 {
530 	return (vfs_getops(vfsp) == vfsops);
531 }
532 
533 /*
534  * Returns non-zero (1) if the file system has installed a non-default,
535  * non-error vfs_sync routine.  Returns zero (0) otherwise.
536  */
537 int
538 vfs_can_sync(vfs_t *vfsp)
539 {
540 	/* vfs_sync() routine is not the default/error function */
541 	return (vfs_getops(vfsp)->vfs_sync != fs_sync);
542 }
543 
544 /*
545  * Initialize a vfs structure.
546  */
547 void
548 vfs_init(vfs_t *vfsp, vfsops_t *op, void *data)
549 {
550 	/* Other initialization has been moved to vfs_alloc() */
551 	vfsp->vfs_count = 0;
552 	vfsp->vfs_next = vfsp;
553 	vfsp->vfs_prev = vfsp;
554 	vfsp->vfs_zone_next = vfsp;
555 	vfsp->vfs_zone_prev = vfsp;
556 	vfsp->vfs_lofi_id = 0;
557 	sema_init(&vfsp->vfs_reflock, 1, NULL, SEMA_DEFAULT, NULL);
558 	vfsimpl_setup(vfsp);
559 	vfsp->vfs_data = (data);
560 	vfs_setops((vfsp), (op));
561 }
562 
563 /*
564  * Allocate and initialize the vfs implementation private data
565  * structure, vfs_impl_t.
566  */
567 void
568 vfsimpl_setup(vfs_t *vfsp)
569 {
570 	int i;
571 
572 	if (vfsp->vfs_implp != NULL) {
573 		return;
574 	}
575 
576 	vfsp->vfs_implp = kmem_alloc(sizeof (vfs_impl_t), KM_SLEEP);
577 	/* Note that these are #define'd in vfs.h */
578 	vfsp->vfs_vskap = NULL;
579 	vfsp->vfs_fstypevsp = NULL;
580 
581 	/* Set size of counted array, then zero the array */
582 	vfsp->vfs_featureset[0] = VFS_FEATURE_MAXSZ - 1;
583 	for (i = 1; i <  VFS_FEATURE_MAXSZ; i++) {
584 		vfsp->vfs_featureset[i] = 0;
585 	}
586 }
587 
588 /*
589  * Release the vfs_impl_t structure, if it exists. Some unbundled
590  * filesystems may not use the newer version of vfs and thus
591  * would not contain this implementation private data structure.
592  */
593 void
594 vfsimpl_teardown(vfs_t *vfsp)
595 {
596 	vfs_impl_t	*vip = vfsp->vfs_implp;
597 
598 	if (vip == NULL)
599 		return;
600 
601 	kmem_free(vfsp->vfs_implp, sizeof (vfs_impl_t));
602 	vfsp->vfs_implp = NULL;
603 }
604 
605 /*
606  * VFS system calls: mount, umount, syssync, statfs, fstatfs, statvfs,
607  * fstatvfs, and sysfs moved to common/syscall.
608  */
609 
610 /*
611  * Update every mounted file system.  We call the vfs_sync operation of
612  * each file system type, passing it a NULL vfsp to indicate that all
613  * mounted file systems of that type should be updated.
614  */
615 void
616 vfs_sync(int flag)
617 {
618 	struct vfssw *vswp;
619 	RLOCK_VFSSW();
620 	for (vswp = &vfssw[1]; vswp < &vfssw[nfstype]; vswp++) {
621 		if (ALLOCATED_VFSSW(vswp) && VFS_INSTALLED(vswp)) {
622 			vfs_refvfssw(vswp);
623 			RUNLOCK_VFSSW();
624 			(void) (*vswp->vsw_vfsops.vfs_sync)(NULL, flag,
625 			    CRED());
626 			vfs_unrefvfssw(vswp);
627 			RLOCK_VFSSW();
628 		}
629 	}
630 	RUNLOCK_VFSSW();
631 }
632 
633 void
634 sync(void)
635 {
636 	vfs_sync(0);
637 }
638 
639 /*
640  * External routines.
641  */
642 
643 krwlock_t vfssw_lock;	/* lock accesses to vfssw */
644 
645 /*
646  * Lock for accessing the vfs linked list.  Initialized in vfs_mountroot(),
647  * but otherwise should be accessed only via vfs_list_lock() and
648  * vfs_list_unlock().  Also used to protect the timestamp for mods to the list.
649  */
650 static krwlock_t vfslist;
651 
652 /*
653  * Mount devfs on /devices. This is done right after root is mounted
654  * to provide device access support for the system
655  */
656 static void
657 vfs_mountdevices(void)
658 {
659 	struct vfssw *vsw;
660 	struct vnode *mvp;
661 	struct mounta mounta = {	/* fake mounta for devfs_mount() */
662 		NULL,
663 		NULL,
664 		MS_SYSSPACE,
665 		NULL,
666 		NULL,
667 		0,
668 		NULL,
669 		0
670 	};
671 
672 	/*
673 	 * _init devfs module to fill in the vfssw
674 	 */
675 	if (modload("fs", "devfs") == -1)
676 		panic("Cannot _init devfs module");
677 
678 	/*
679 	 * Hold vfs
680 	 */
681 	RLOCK_VFSSW();
682 	vsw = vfs_getvfsswbyname("devfs");
683 	VFS_INIT(&devices, &vsw->vsw_vfsops, NULL);
684 	VFS_HOLD(&devices);
685 
686 	/*
687 	 * Locate mount point
688 	 */
689 	if (lookupname("/devices", UIO_SYSSPACE, FOLLOW, NULLVPP, &mvp))
690 		panic("Cannot find /devices");
691 
692 	/*
693 	 * Perform the mount of /devices
694 	 */
695 	if (VFS_MOUNT(&devices, mvp, &mounta, CRED()))
696 		panic("Cannot mount /devices");
697 
698 	RUNLOCK_VFSSW();
699 
700 	/*
701 	 * Set appropriate members and add to vfs list for mnttab display
702 	 */
703 	vfs_setresource(&devices, "/devices", 0);
704 	vfs_setmntpoint(&devices, "/devices", 0);
705 
706 	/*
707 	 * Hold the root of /devices so it won't go away
708 	 */
709 	if (VFS_ROOT(&devices, &devicesdir))
710 		panic("vfs_mountdevices: not devices root");
711 
712 	if (vfs_lock(&devices) != 0) {
713 		VN_RELE(devicesdir);
714 		cmn_err(CE_NOTE, "Cannot acquire vfs_lock of /devices");
715 		return;
716 	}
717 
718 	if (vn_vfswlock(mvp) != 0) {
719 		vfs_unlock(&devices);
720 		VN_RELE(devicesdir);
721 		cmn_err(CE_NOTE, "Cannot acquire vfswlock of /devices");
722 		return;
723 	}
724 
725 	vfs_add(mvp, &devices, 0);
726 	vn_vfsunlock(mvp);
727 	vfs_unlock(&devices);
728 	VN_RELE(devicesdir);
729 }
730 
731 /*
732  * mount the first instance of /dev  to root and remain mounted
733  */
734 static void
735 vfs_mountdev1(void)
736 {
737 	struct vfssw *vsw;
738 	struct vnode *mvp;
739 	struct mounta mounta = {	/* fake mounta for sdev_mount() */
740 		NULL,
741 		NULL,
742 		MS_SYSSPACE | MS_OVERLAY,
743 		NULL,
744 		NULL,
745 		0,
746 		NULL,
747 		0
748 	};
749 
750 	/*
751 	 * _init dev module to fill in the vfssw
752 	 */
753 	if (modload("fs", "dev") == -1)
754 		cmn_err(CE_PANIC, "Cannot _init dev module\n");
755 
756 	/*
757 	 * Hold vfs
758 	 */
759 	RLOCK_VFSSW();
760 	vsw = vfs_getvfsswbyname("dev");
761 	VFS_INIT(&dev, &vsw->vsw_vfsops, NULL);
762 	VFS_HOLD(&dev);
763 
764 	/*
765 	 * Locate mount point
766 	 */
767 	if (lookupname("/dev", UIO_SYSSPACE, FOLLOW, NULLVPP, &mvp))
768 		cmn_err(CE_PANIC, "Cannot find /dev\n");
769 
770 	/*
771 	 * Perform the mount of /dev
772 	 */
773 	if (VFS_MOUNT(&dev, mvp, &mounta, CRED()))
774 		cmn_err(CE_PANIC, "Cannot mount /dev 1\n");
775 
776 	RUNLOCK_VFSSW();
777 
778 	/*
779 	 * Set appropriate members and add to vfs list for mnttab display
780 	 */
781 	vfs_setresource(&dev, "/dev", 0);
782 	vfs_setmntpoint(&dev, "/dev", 0);
783 
784 	/*
785 	 * Hold the root of /dev so it won't go away
786 	 */
787 	if (VFS_ROOT(&dev, &devdir))
788 		cmn_err(CE_PANIC, "vfs_mountdev1: not dev root");
789 
790 	if (vfs_lock(&dev) != 0) {
791 		VN_RELE(devdir);
792 		cmn_err(CE_NOTE, "Cannot acquire vfs_lock of /dev");
793 		return;
794 	}
795 
796 	if (vn_vfswlock(mvp) != 0) {
797 		vfs_unlock(&dev);
798 		VN_RELE(devdir);
799 		cmn_err(CE_NOTE, "Cannot acquire vfswlock of /dev");
800 		return;
801 	}
802 
803 	vfs_add(mvp, &dev, 0);
804 	vn_vfsunlock(mvp);
805 	vfs_unlock(&dev);
806 	VN_RELE(devdir);
807 }
808 
809 /*
810  * Mount required filesystem. This is done right after root is mounted.
811  */
812 static void
813 vfs_mountfs(char *module, char *spec, char *path)
814 {
815 	struct vnode *mvp;
816 	struct mounta mounta;
817 	vfs_t *vfsp;
818 
819 	bzero(&mounta, sizeof (mounta));
820 	mounta.flags = MS_SYSSPACE | MS_DATA;
821 	mounta.fstype = module;
822 	mounta.spec = spec;
823 	mounta.dir = path;
824 	if (lookupname(path, UIO_SYSSPACE, FOLLOW, NULLVPP, &mvp)) {
825 		cmn_err(CE_WARN, "Cannot find %s", path);
826 		return;
827 	}
828 	if (domount(NULL, &mounta, mvp, CRED(), &vfsp))
829 		cmn_err(CE_WARN, "Cannot mount %s", path);
830 	else
831 		VFS_RELE(vfsp);
832 	VN_RELE(mvp);
833 }
834 
835 /*
836  * vfs_mountroot is called by main() to mount the root filesystem.
837  */
838 void
839 vfs_mountroot(void)
840 {
841 	struct vnode	*rvp = NULL;
842 	char		*path;
843 	size_t		plen;
844 	struct vfssw	*vswp;
845 	proc_t		*p;
846 
847 	rw_init(&vfssw_lock, NULL, RW_DEFAULT, NULL);
848 	rw_init(&vfslist, NULL, RW_DEFAULT, NULL);
849 
850 	/*
851 	 * Alloc the vfs hash bucket array and locks
852 	 */
853 	rvfs_list = kmem_zalloc(vfshsz * sizeof (rvfs_t), KM_SLEEP);
854 
855 	/*
856 	 * Call machine-dependent routine "rootconf" to choose a root
857 	 * file system type.
858 	 */
859 	if (rootconf())
860 		panic("vfs_mountroot: cannot mount root");
861 	/*
862 	 * Get vnode for '/'.  Set up rootdir, u.u_rdir and u.u_cdir
863 	 * to point to it.  These are used by lookuppn() so that it
864 	 * knows where to start from ('/' or '.').
865 	 */
866 	vfs_setmntpoint(rootvfs, "/", 0);
867 	if (VFS_ROOT(rootvfs, &rootdir))
868 		panic("vfs_mountroot: no root vnode");
869 
870 	/*
871 	 * At this point, the process tree consists of p0 and possibly some
872 	 * direct children of p0.  (i.e. there are no grandchildren)
873 	 *
874 	 * Walk through them all, setting their current directory.
875 	 */
876 	mutex_enter(&pidlock);
877 	for (p = practive; p != NULL; p = p->p_next) {
878 		ASSERT(p == &p0 || p->p_parent == &p0);
879 
880 		PTOU(p)->u_cdir = rootdir;
881 		VN_HOLD(PTOU(p)->u_cdir);
882 		PTOU(p)->u_rdir = NULL;
883 	}
884 	mutex_exit(&pidlock);
885 
886 	/*
887 	 * Setup the global zone's rootvp, now that it exists.
888 	 */
889 	global_zone->zone_rootvp = rootdir;
890 	VN_HOLD(global_zone->zone_rootvp);
891 
892 	/*
893 	 * Notify the module code that it can begin using the
894 	 * root filesystem instead of the boot program's services.
895 	 */
896 	modrootloaded = 1;
897 
898 	/*
899 	 * Special handling for a ZFS root file system.
900 	 */
901 	zfs_boot_init();
902 
903 	/*
904 	 * Set up mnttab information for root
905 	 */
906 	vfs_setresource(rootvfs, rootfs.bo_name, 0);
907 
908 	/*
909 	 * Notify cluster software that the root filesystem is available.
910 	 */
911 	clboot_mountroot();
912 
913 	/* Now that we're all done with the root FS, set up its vopstats */
914 	if ((vswp = vfs_getvfsswbyvfsops(vfs_getops(rootvfs))) != NULL) {
915 		/* Set flag for statistics collection */
916 		if (vswp->vsw_flag & VSW_STATS) {
917 			initialize_vopstats(&rootvfs->vfs_vopstats);
918 			rootvfs->vfs_flag |= VFS_STATS;
919 			rootvfs->vfs_fstypevsp =
920 			    get_fstype_vopstats(rootvfs, vswp);
921 			rootvfs->vfs_vskap = get_vskstat_anchor(rootvfs);
922 		}
923 		vfs_unrefvfssw(vswp);
924 	}
925 
926 	/*
927 	 * Mount /devices, /dev instance 1, /system/contract, /etc/mnttab,
928 	 * /etc/svc/volatile, /etc/dfs/sharetab, /system/object, and /proc.
929 	 */
930 	vfs_mountdevices();
931 	vfs_mountdev1();
932 
933 	vfs_mountfs("ctfs", "ctfs", CTFS_ROOT);
934 	vfs_mountfs("proc", "/proc", "/proc");
935 	vfs_mountfs("mntfs", "/etc/mnttab", "/etc/mnttab");
936 	vfs_mountfs("tmpfs", "/etc/svc/volatile", "/etc/svc/volatile");
937 	vfs_mountfs("objfs", "objfs", OBJFS_ROOT);
938 	vfs_mountfs("bootfs", "bootfs", "/system/boot");
939 
940 	if (getzoneid() == GLOBAL_ZONEID) {
941 		vfs_mountfs("sharefs", "sharefs", "/etc/dfs/sharetab");
942 	}
943 
944 	if (strcmp(rootfs.bo_fstype, "zfs") != 0) {
945 		/*
946 		 * Look up the root device via devfs so that a dv_node is
947 		 * created for it. The vnode is never VN_RELE()ed.
948 		 * We allocate more than MAXPATHLEN so that the
949 		 * buffer passed to i_ddi_prompath_to_devfspath() is
950 		 * exactly MAXPATHLEN (the function expects a buffer
951 		 * of that length).
952 		 */
953 		plen = strlen("/devices");
954 		path = kmem_alloc(plen + MAXPATHLEN, KM_SLEEP);
955 		(void) strcpy(path, "/devices");
956 
957 		if (i_ddi_prompath_to_devfspath(rootfs.bo_name, path + plen)
958 		    != DDI_SUCCESS ||
959 		    lookupname(path, UIO_SYSSPACE, FOLLOW, NULLVPP, &rvp)) {
960 
961 			/* NUL terminate in case "path" has garbage */
962 			path[plen + MAXPATHLEN - 1] = '\0';
963 #ifdef	DEBUG
964 			cmn_err(CE_WARN, "!Cannot lookup root device: %s",
965 			    path);
966 #endif
967 		}
968 		kmem_free(path, plen + MAXPATHLEN);
969 	}
970 
971 	vfs_mnttabvp_setup();
972 }
973 
974 /*
975  * Check to see if our "block device" is actually a file.  If so,
976  * automatically add a lofi device, and keep track of this fact.
977  */
978 static int
979 lofi_add(const char *fsname, struct vfs *vfsp,
980     mntopts_t *mntopts, struct mounta *uap)
981 {
982 	int fromspace = (uap->flags & MS_SYSSPACE) ?
983 	    UIO_SYSSPACE : UIO_USERSPACE;
984 	struct lofi_ioctl *li = NULL;
985 	struct vnode *vp = NULL;
986 	struct pathname	pn = { NULL };
987 	ldi_ident_t ldi_id;
988 	ldi_handle_t ldi_hdl;
989 	vfssw_t *vfssw;
990 	int id;
991 	int err = 0;
992 
993 	if ((vfssw = vfs_getvfssw(fsname)) == NULL)
994 		return (0);
995 
996 	if (!(vfssw->vsw_flag & VSW_CANLOFI)) {
997 		vfs_unrefvfssw(vfssw);
998 		return (0);
999 	}
1000 
1001 	vfs_unrefvfssw(vfssw);
1002 	vfssw = NULL;
1003 
1004 	if (pn_get(uap->spec, fromspace, &pn) != 0)
1005 		return (0);
1006 
1007 	if (lookupname(uap->spec, fromspace, FOLLOW, NULL, &vp) != 0)
1008 		goto out;
1009 
1010 	if (vp->v_type != VREG)
1011 		goto out;
1012 
1013 	/* OK, this is a lofi mount. */
1014 
1015 	if ((uap->flags & (MS_REMOUNT|MS_GLOBAL)) ||
1016 	    vfs_optionisset_nolock(mntopts, MNTOPT_SUID, NULL) ||
1017 	    vfs_optionisset_nolock(mntopts, MNTOPT_SETUID, NULL) ||
1018 	    vfs_optionisset_nolock(mntopts, MNTOPT_DEVICES, NULL)) {
1019 		err = EINVAL;
1020 		goto out;
1021 	}
1022 
1023 	ldi_id = ldi_ident_from_anon();
1024 	li = kmem_zalloc(sizeof (*li), KM_SLEEP);
1025 	(void) strlcpy(li->li_filename, pn.pn_path, MAXPATHLEN);
1026 
1027 	err = ldi_open_by_name("/dev/lofictl", FREAD | FWRITE, kcred,
1028 	    &ldi_hdl, ldi_id);
1029 
1030 	if (err)
1031 		goto out2;
1032 
1033 	err = ldi_ioctl(ldi_hdl, LOFI_MAP_FILE, (intptr_t)li,
1034 	    FREAD | FWRITE | FKIOCTL, kcred, &id);
1035 
1036 	(void) ldi_close(ldi_hdl, FREAD | FWRITE, kcred);
1037 
1038 	if (!err)
1039 		vfsp->vfs_lofi_id = id;
1040 
1041 out2:
1042 	ldi_ident_release(ldi_id);
1043 out:
1044 	if (li != NULL)
1045 		kmem_free(li, sizeof (*li));
1046 	if (vp != NULL)
1047 		VN_RELE(vp);
1048 	pn_free(&pn);
1049 	return (err);
1050 }
1051 
1052 static void
1053 lofi_remove(struct vfs *vfsp)
1054 {
1055 	struct lofi_ioctl *li;
1056 	ldi_ident_t ldi_id;
1057 	ldi_handle_t ldi_hdl;
1058 	int err;
1059 
1060 	if (vfsp->vfs_lofi_id == 0)
1061 		return;
1062 
1063 	ldi_id = ldi_ident_from_anon();
1064 
1065 	li = kmem_zalloc(sizeof (*li), KM_SLEEP);
1066 	li->li_id = vfsp->vfs_lofi_id;
1067 	li->li_cleanup = B_TRUE;
1068 
1069 	err = ldi_open_by_name("/dev/lofictl", FREAD | FWRITE, kcred,
1070 	    &ldi_hdl, ldi_id);
1071 
1072 	if (err)
1073 		goto out;
1074 
1075 	err = ldi_ioctl(ldi_hdl, LOFI_UNMAP_FILE_MINOR, (intptr_t)li,
1076 	    FREAD | FWRITE | FKIOCTL, kcred, NULL);
1077 
1078 	(void) ldi_close(ldi_hdl, FREAD | FWRITE, kcred);
1079 
1080 	if (!err)
1081 		vfsp->vfs_lofi_id = 0;
1082 
1083 out:
1084 	ldi_ident_release(ldi_id);
1085 	kmem_free(li, sizeof (*li));
1086 }
1087 
1088 /*
1089  * Common mount code.  Called from the system call entry point, from autofs,
1090  * nfsv4 trigger mounts, and from pxfs.
1091  *
1092  * Takes the effective file system type, mount arguments, the mount point
1093  * vnode, flags specifying whether the mount is a remount and whether it
1094  * should be entered into the vfs list, and credentials.  Fills in its vfspp
1095  * parameter with the mounted file system instance's vfs.
1096  *
1097  * Note that the effective file system type is specified as a string.  It may
1098  * be null, in which case it's determined from the mount arguments, and may
1099  * differ from the type specified in the mount arguments; this is a hook to
1100  * allow interposition when instantiating file system instances.
1101  *
1102  * The caller is responsible for releasing its own hold on the mount point
1103  * vp (this routine does its own hold when necessary).
1104  * Also note that for remounts, the mount point vp should be the vnode for
1105  * the root of the file system rather than the vnode that the file system
1106  * is mounted on top of.
1107  */
1108 int
1109 domount(char *fsname, struct mounta *uap, vnode_t *vp, struct cred *credp,
1110     struct vfs **vfspp)
1111 {
1112 	struct vfssw	*vswp;
1113 	vfsops_t	*vfsops;
1114 	struct vfs	*vfsp;
1115 	struct vnode	*bvp;
1116 	dev_t		bdev = 0;
1117 	mntopts_t	mnt_mntopts;
1118 	int		error = 0;
1119 	int		copyout_error = 0;
1120 	int		ovflags = 0;
1121 	char		*opts = uap->optptr;
1122 	char		*inargs = opts;
1123 	int		optlen = uap->optlen;
1124 	int		remount;
1125 	int		rdonly;
1126 	int		nbmand = 0;
1127 	int		delmip = 0;
1128 	int		addmip = 0;
1129 	int		splice = ((uap->flags & MS_NOSPLICE) == 0);
1130 	int		fromspace = (uap->flags & MS_SYSSPACE) ?
1131 	    UIO_SYSSPACE : UIO_USERSPACE;
1132 	char		*resource = NULL, *mountpt = NULL;
1133 	refstr_t	*oldresource, *oldmntpt;
1134 	struct pathname	pn, rpn;
1135 	vsk_anchor_t	*vskap;
1136 	char fstname[FSTYPSZ];
1137 	zone_t		*zone;
1138 
1139 	/*
1140 	 * The v_flag value for the mount point vp is permanently set
1141 	 * to VVFSLOCK so that no one bypasses the vn_vfs*locks routine
1142 	 * for mount point locking.
1143 	 */
1144 	mutex_enter(&vp->v_lock);
1145 	vp->v_flag |= VVFSLOCK;
1146 	mutex_exit(&vp->v_lock);
1147 
1148 	mnt_mntopts.mo_count = 0;
1149 	/*
1150 	 * Find the ops vector to use to invoke the file system-specific mount
1151 	 * method.  If the fsname argument is non-NULL, use it directly.
1152 	 * Otherwise, dig the file system type information out of the mount
1153 	 * arguments.
1154 	 *
1155 	 * A side effect is to hold the vfssw entry.
1156 	 *
1157 	 * Mount arguments can be specified in several ways, which are
1158 	 * distinguished by flag bit settings.  The preferred way is to set
1159 	 * MS_OPTIONSTR, indicating an 8 argument mount with the file system
1160 	 * type supplied as a character string and the last two arguments
1161 	 * being a pointer to a character buffer and the size of the buffer.
1162 	 * On entry, the buffer holds a null terminated list of options; on
1163 	 * return, the string is the list of options the file system
1164 	 * recognized. If MS_DATA is set arguments five and six point to a
1165 	 * block of binary data which the file system interprets.
1166 	 * A further wrinkle is that some callers don't set MS_FSS and MS_DATA
1167 	 * consistently with these conventions.  To handle them, we check to
1168 	 * see whether the pointer to the file system name has a numeric value
1169 	 * less than 256.  If so, we treat it as an index.
1170 	 */
1171 	if (fsname != NULL) {
1172 		if ((vswp = vfs_getvfssw(fsname)) == NULL) {
1173 			return (EINVAL);
1174 		}
1175 	} else if (uap->flags & (MS_OPTIONSTR | MS_DATA | MS_FSS)) {
1176 		size_t n;
1177 		uint_t fstype;
1178 
1179 		fsname = fstname;
1180 
1181 		if ((fstype = (uintptr_t)uap->fstype) < 256) {
1182 			RLOCK_VFSSW();
1183 			if (fstype == 0 || fstype >= nfstype ||
1184 			    !ALLOCATED_VFSSW(&vfssw[fstype])) {
1185 				RUNLOCK_VFSSW();
1186 				return (EINVAL);
1187 			}
1188 			(void) strcpy(fsname, vfssw[fstype].vsw_name);
1189 			RUNLOCK_VFSSW();
1190 			if ((vswp = vfs_getvfssw(fsname)) == NULL)
1191 				return (EINVAL);
1192 		} else {
1193 			/*
1194 			 * Handle either kernel or user address space.
1195 			 */
1196 			if (uap->flags & MS_SYSSPACE) {
1197 				error = copystr(uap->fstype, fsname,
1198 				    FSTYPSZ, &n);
1199 			} else {
1200 				error = copyinstr(uap->fstype, fsname,
1201 				    FSTYPSZ, &n);
1202 			}
1203 			if (error) {
1204 				if (error == ENAMETOOLONG)
1205 					return (EINVAL);
1206 				return (error);
1207 			}
1208 			if ((vswp = vfs_getvfssw(fsname)) == NULL)
1209 				return (EINVAL);
1210 		}
1211 	} else {
1212 		if ((vswp = vfs_getvfsswbyvfsops(vfs_getops(rootvfs))) == NULL)
1213 			return (EINVAL);
1214 		fsname = vswp->vsw_name;
1215 	}
1216 	if (!VFS_INSTALLED(vswp))
1217 		return (EINVAL);
1218 
1219 	if ((error = secpolicy_fs_allowed_mount(fsname)) != 0)  {
1220 		vfs_unrefvfssw(vswp);
1221 		return (error);
1222 	}
1223 
1224 	vfsops = &vswp->vsw_vfsops;
1225 
1226 	vfs_copyopttbl(&vswp->vsw_optproto, &mnt_mntopts);
1227 	/*
1228 	 * Fetch mount options and parse them for generic vfs options
1229 	 */
1230 	if (uap->flags & MS_OPTIONSTR) {
1231 		/*
1232 		 * Limit the buffer size
1233 		 */
1234 		if (optlen < 0 || optlen > MAX_MNTOPT_STR) {
1235 			error = EINVAL;
1236 			goto errout;
1237 		}
1238 		if ((uap->flags & MS_SYSSPACE) == 0) {
1239 			inargs = kmem_alloc(MAX_MNTOPT_STR, KM_SLEEP);
1240 			inargs[0] = '\0';
1241 			if (optlen) {
1242 				error = copyinstr(opts, inargs, (size_t)optlen,
1243 				    NULL);
1244 				if (error) {
1245 					goto errout;
1246 				}
1247 			}
1248 		}
1249 		vfs_parsemntopts(&mnt_mntopts, inargs, 0);
1250 	}
1251 	/*
1252 	 * Flag bits override the options string.
1253 	 */
1254 	if (uap->flags & MS_REMOUNT)
1255 		vfs_setmntopt_nolock(&mnt_mntopts, MNTOPT_REMOUNT, NULL, 0, 0);
1256 	if (uap->flags & MS_RDONLY)
1257 		vfs_setmntopt_nolock(&mnt_mntopts, MNTOPT_RO, NULL, 0, 0);
1258 	if (uap->flags & MS_NOSUID)
1259 		vfs_setmntopt_nolock(&mnt_mntopts, MNTOPT_NOSUID, NULL, 0, 0);
1260 
1261 	/*
1262 	 * Check if this is a remount; must be set in the option string and
1263 	 * the file system must support a remount option.
1264 	 */
1265 	if (remount = vfs_optionisset_nolock(&mnt_mntopts,
1266 	    MNTOPT_REMOUNT, NULL)) {
1267 		if (!(vswp->vsw_flag & VSW_CANREMOUNT)) {
1268 			error = ENOTSUP;
1269 			goto errout;
1270 		}
1271 		uap->flags |= MS_REMOUNT;
1272 	}
1273 
1274 	/*
1275 	 * uap->flags and vfs_optionisset() should agree.
1276 	 */
1277 	if (rdonly = vfs_optionisset_nolock(&mnt_mntopts, MNTOPT_RO, NULL)) {
1278 		uap->flags |= MS_RDONLY;
1279 	}
1280 	if (vfs_optionisset_nolock(&mnt_mntopts, MNTOPT_NOSUID, NULL)) {
1281 		uap->flags |= MS_NOSUID;
1282 	}
1283 	nbmand = vfs_optionisset_nolock(&mnt_mntopts, MNTOPT_NBMAND, NULL);
1284 	ASSERT(splice || !remount);
1285 	/*
1286 	 * If we are splicing the fs into the namespace,
1287 	 * perform mount point checks.
1288 	 *
1289 	 * We want to resolve the path for the mount point to eliminate
1290 	 * '.' and ".." and symlinks in mount points; we can't do the
1291 	 * same for the resource string, since it would turn
1292 	 * "/dev/dsk/c0t0d0s0" into "/devices/pci@...".  We need to do
1293 	 * this before grabbing vn_vfswlock(), because otherwise we
1294 	 * would deadlock with lookuppn().
1295 	 */
1296 	if (splice) {
1297 		ASSERT(vp->v_count > 0);
1298 
1299 		/*
1300 		 * Pick up mount point and device from appropriate space.
1301 		 */
1302 		if (pn_get(uap->spec, fromspace, &pn) == 0) {
1303 			resource = kmem_alloc(pn.pn_pathlen + 1,
1304 			    KM_SLEEP);
1305 			(void) strcpy(resource, pn.pn_path);
1306 			pn_free(&pn);
1307 		}
1308 		/*
1309 		 * Do a lookupname prior to taking the
1310 		 * writelock. Mark this as completed if
1311 		 * successful for later cleanup and addition to
1312 		 * the mount in progress table.
1313 		 */
1314 		if ((vswp->vsw_flag & VSW_MOUNTDEV) &&
1315 		    (uap->flags & MS_GLOBAL) == 0 &&
1316 		    lookupname(uap->spec, fromspace,
1317 		    FOLLOW, NULL, &bvp) == 0) {
1318 			addmip = 1;
1319 		}
1320 
1321 		if ((error = pn_get(uap->dir, fromspace, &pn)) == 0) {
1322 			pathname_t *pnp;
1323 
1324 			if (*pn.pn_path != '/') {
1325 				error = EINVAL;
1326 				pn_free(&pn);
1327 				goto errout;
1328 			}
1329 			pn_alloc(&rpn);
1330 			/*
1331 			 * Kludge to prevent autofs from deadlocking with
1332 			 * itself when it calls domount().
1333 			 *
1334 			 * If autofs is calling, it is because it is doing
1335 			 * (autofs) mounts in the process of an NFS mount.  A
1336 			 * lookuppn() here would cause us to block waiting for
1337 			 * said NFS mount to complete, which can't since this
1338 			 * is the thread that was supposed to doing it.
1339 			 */
1340 			if (fromspace == UIO_USERSPACE) {
1341 				if ((error = lookuppn(&pn, &rpn, FOLLOW, NULL,
1342 				    NULL)) == 0) {
1343 					pnp = &rpn;
1344 				} else {
1345 					/*
1346 					 * The file disappeared or otherwise
1347 					 * became inaccessible since we opened
1348 					 * it; might as well fail the mount
1349 					 * since the mount point is no longer
1350 					 * accessible.
1351 					 */
1352 					pn_free(&rpn);
1353 					pn_free(&pn);
1354 					goto errout;
1355 				}
1356 			} else {
1357 				pnp = &pn;
1358 			}
1359 			mountpt = kmem_alloc(pnp->pn_pathlen + 1, KM_SLEEP);
1360 			(void) strcpy(mountpt, pnp->pn_path);
1361 
1362 			/*
1363 			 * If the addition of the zone's rootpath
1364 			 * would push us over a total path length
1365 			 * of MAXPATHLEN, we fail the mount with
1366 			 * ENAMETOOLONG, which is what we would have
1367 			 * gotten if we were trying to perform the same
1368 			 * mount in the global zone.
1369 			 *
1370 			 * strlen() doesn't count the trailing
1371 			 * '\0', but zone_rootpathlen counts both a
1372 			 * trailing '/' and the terminating '\0'.
1373 			 */
1374 			if ((curproc->p_zone->zone_rootpathlen - 1 +
1375 			    strlen(mountpt)) > MAXPATHLEN ||
1376 			    (resource != NULL &&
1377 			    (curproc->p_zone->zone_rootpathlen - 1 +
1378 			    strlen(resource)) > MAXPATHLEN)) {
1379 				error = ENAMETOOLONG;
1380 			}
1381 
1382 			pn_free(&rpn);
1383 			pn_free(&pn);
1384 		}
1385 
1386 		if (error)
1387 			goto errout;
1388 
1389 		/*
1390 		 * Prevent path name resolution from proceeding past
1391 		 * the mount point.
1392 		 */
1393 		if (vn_vfswlock(vp) != 0) {
1394 			error = EBUSY;
1395 			goto errout;
1396 		}
1397 
1398 		/*
1399 		 * Verify that it's legitimate to establish a mount on
1400 		 * the prospective mount point.
1401 		 */
1402 		if (vn_mountedvfs(vp) != NULL) {
1403 			/*
1404 			 * The mount point lock was obtained after some
1405 			 * other thread raced through and established a mount.
1406 			 */
1407 			vn_vfsunlock(vp);
1408 			error = EBUSY;
1409 			goto errout;
1410 		}
1411 		if (vp->v_flag & VNOMOUNT) {
1412 			vn_vfsunlock(vp);
1413 			error = EINVAL;
1414 			goto errout;
1415 		}
1416 	}
1417 	if ((uap->flags & (MS_DATA | MS_OPTIONSTR)) == 0) {
1418 		uap->dataptr = NULL;
1419 		uap->datalen = 0;
1420 	}
1421 
1422 	/*
1423 	 * If this is a remount, we don't want to create a new VFS.
1424 	 * Instead, we pass the existing one with a remount flag.
1425 	 */
1426 	if (remount) {
1427 		/*
1428 		 * Confirm that the mount point is the root vnode of the
1429 		 * file system that is being remounted.
1430 		 * This can happen if the user specifies a different
1431 		 * mount point directory pathname in the (re)mount command.
1432 		 *
1433 		 * Code below can only be reached if splice is true, so it's
1434 		 * safe to do vn_vfsunlock() here.
1435 		 */
1436 		if ((vp->v_flag & VROOT) == 0) {
1437 			vn_vfsunlock(vp);
1438 			error = ENOENT;
1439 			goto errout;
1440 		}
1441 		/*
1442 		 * Disallow making file systems read-only unless file system
1443 		 * explicitly allows it in its vfssw.  Ignore other flags.
1444 		 */
1445 		if (rdonly && vn_is_readonly(vp) == 0 &&
1446 		    (vswp->vsw_flag & VSW_CANRWRO) == 0) {
1447 			vn_vfsunlock(vp);
1448 			error = EINVAL;
1449 			goto errout;
1450 		}
1451 		/*
1452 		 * Disallow changing the NBMAND disposition of the file
1453 		 * system on remounts.
1454 		 */
1455 		if ((nbmand && ((vp->v_vfsp->vfs_flag & VFS_NBMAND) == 0)) ||
1456 		    (!nbmand && (vp->v_vfsp->vfs_flag & VFS_NBMAND))) {
1457 			vn_vfsunlock(vp);
1458 			error = EINVAL;
1459 			goto errout;
1460 		}
1461 		vfsp = vp->v_vfsp;
1462 		ovflags = vfsp->vfs_flag;
1463 		vfsp->vfs_flag |= VFS_REMOUNT;
1464 		vfsp->vfs_flag &= ~VFS_RDONLY;
1465 	} else {
1466 		vfsp = vfs_alloc(KM_SLEEP);
1467 		VFS_INIT(vfsp, vfsops, NULL);
1468 	}
1469 
1470 	VFS_HOLD(vfsp);
1471 
1472 	if ((error = lofi_add(fsname, vfsp, &mnt_mntopts, uap)) != 0) {
1473 		if (!remount) {
1474 			if (splice)
1475 				vn_vfsunlock(vp);
1476 			vfs_free(vfsp);
1477 		} else {
1478 			vn_vfsunlock(vp);
1479 			VFS_RELE(vfsp);
1480 		}
1481 		goto errout;
1482 	}
1483 
1484 	/*
1485 	 * PRIV_SYS_MOUNT doesn't mean you can become root.
1486 	 */
1487 	if (vfsp->vfs_lofi_id != 0) {
1488 		uap->flags |= MS_NOSUID;
1489 		vfs_setmntopt_nolock(&mnt_mntopts, MNTOPT_NOSUID, NULL, 0, 0);
1490 	}
1491 
1492 	/*
1493 	 * The vfs_reflock is not used anymore the code below explicitly
1494 	 * holds it preventing others accesing it directly.
1495 	 */
1496 	if ((sema_tryp(&vfsp->vfs_reflock) == 0) &&
1497 	    !(vfsp->vfs_flag & VFS_REMOUNT))
1498 		cmn_err(CE_WARN,
1499 		    "mount type %s couldn't get vfs_reflock", vswp->vsw_name);
1500 
1501 	/*
1502 	 * Lock the vfs. If this is a remount we want to avoid spurious umount
1503 	 * failures that happen as a side-effect of fsflush() and other mount
1504 	 * and unmount operations that might be going on simultaneously and
1505 	 * may have locked the vfs currently. To not return EBUSY immediately
1506 	 * here we use vfs_lock_wait() instead vfs_lock() for the remount case.
1507 	 */
1508 	if (!remount) {
1509 		if (error = vfs_lock(vfsp)) {
1510 			lofi_remove(vfsp);
1511 
1512 			if (splice)
1513 				vn_vfsunlock(vp);
1514 			vfs_free(vfsp);
1515 			goto errout;
1516 		}
1517 	} else {
1518 		vfs_lock_wait(vfsp);
1519 	}
1520 
1521 	/*
1522 	 * Add device to mount in progress table, global mounts require special
1523 	 * handling. It is possible that we have already done the lookupname
1524 	 * on a spliced, non-global fs. If so, we don't want to do it again
1525 	 * since we cannot do a lookupname after taking the
1526 	 * wlock above. This case is for a non-spliced, non-global filesystem.
1527 	 */
1528 	if (!addmip) {
1529 		if ((vswp->vsw_flag & VSW_MOUNTDEV) &&
1530 		    (uap->flags & MS_GLOBAL) == 0 &&
1531 		    lookupname(uap->spec, fromspace, FOLLOW, NULL, &bvp) == 0) {
1532 			addmip = 1;
1533 		}
1534 	}
1535 
1536 	if (addmip) {
1537 		vnode_t *lvp = NULL;
1538 
1539 		error = vfs_get_lofi(vfsp, &lvp);
1540 		if (error > 0) {
1541 			lofi_remove(vfsp);
1542 
1543 			if (splice)
1544 				vn_vfsunlock(vp);
1545 			vfs_unlock(vfsp);
1546 
1547 			if (remount) {
1548 				VFS_RELE(vfsp);
1549 			} else {
1550 				vfs_free(vfsp);
1551 			}
1552 
1553 			goto errout;
1554 		} else if (error == -1) {
1555 			bdev = bvp->v_rdev;
1556 			VN_RELE(bvp);
1557 		} else {
1558 			bdev = lvp->v_rdev;
1559 			VN_RELE(lvp);
1560 			VN_RELE(bvp);
1561 		}
1562 
1563 		vfs_addmip(bdev, vfsp);
1564 		addmip = 0;
1565 		delmip = 1;
1566 	}
1567 	/*
1568 	 * Invalidate cached entry for the mount point.
1569 	 */
1570 	if (splice)
1571 		dnlc_purge_vp(vp);
1572 
1573 	/*
1574 	 * If have an option string but the filesystem doesn't supply a
1575 	 * prototype options table, create a table with the global
1576 	 * options and sufficient room to accept all the options in the
1577 	 * string.  Then parse the passed in option string
1578 	 * accepting all the options in the string.  This gives us an
1579 	 * option table with all the proper cancel properties for the
1580 	 * global options.
1581 	 *
1582 	 * Filesystems that supply a prototype options table are handled
1583 	 * earlier in this function.
1584 	 */
1585 	if (uap->flags & MS_OPTIONSTR) {
1586 		if (!(vswp->vsw_flag & VSW_HASPROTO)) {
1587 			mntopts_t tmp_mntopts;
1588 
1589 			tmp_mntopts.mo_count = 0;
1590 			vfs_createopttbl_extend(&tmp_mntopts, inargs,
1591 			    &mnt_mntopts);
1592 			vfs_parsemntopts(&tmp_mntopts, inargs, 1);
1593 			vfs_swapopttbl_nolock(&mnt_mntopts, &tmp_mntopts);
1594 			vfs_freeopttbl(&tmp_mntopts);
1595 		}
1596 	}
1597 
1598 	/*
1599 	 * Serialize with zone state transitions.
1600 	 * See vfs_list_add; zone mounted into is:
1601 	 *	zone_find_by_path(refstr_value(vfsp->vfs_mntpt))
1602 	 * not the zone doing the mount (curproc->p_zone), but if we're already
1603 	 * inside a NGZ, then we know what zone we are.
1604 	 */
1605 	if (INGLOBALZONE(curproc)) {
1606 		zone = zone_find_by_path(mountpt);
1607 		ASSERT(zone != NULL);
1608 	} else {
1609 		zone = curproc->p_zone;
1610 		/*
1611 		 * zone_find_by_path does a hold, so do one here too so that
1612 		 * we can do a zone_rele after mount_completed.
1613 		 */
1614 		zone_hold(zone);
1615 	}
1616 	mount_in_progress(zone);
1617 	/*
1618 	 * Instantiate (or reinstantiate) the file system.  If appropriate,
1619 	 * splice it into the file system name space.
1620 	 *
1621 	 * We want VFS_MOUNT() to be able to override the vfs_resource
1622 	 * string if necessary (ie, mntfs), and also for a remount to
1623 	 * change the same (necessary when remounting '/' during boot).
1624 	 * So we set up vfs_mntpt and vfs_resource to what we think they
1625 	 * should be, then hand off control to VFS_MOUNT() which can
1626 	 * override this.
1627 	 *
1628 	 * For safety's sake, when changing vfs_resource or vfs_mntpt of
1629 	 * a vfs which is on the vfs list (i.e. during a remount), we must
1630 	 * never set those fields to NULL. Several bits of code make
1631 	 * assumptions that the fields are always valid.
1632 	 */
1633 	vfs_swapopttbl(&mnt_mntopts, &vfsp->vfs_mntopts);
1634 	if (remount) {
1635 		if ((oldresource = vfsp->vfs_resource) != NULL)
1636 			refstr_hold(oldresource);
1637 		if ((oldmntpt = vfsp->vfs_mntpt) != NULL)
1638 			refstr_hold(oldmntpt);
1639 	}
1640 	vfs_setresource(vfsp, resource, 0);
1641 	vfs_setmntpoint(vfsp, mountpt, 0);
1642 
1643 	/*
1644 	 * going to mount on this vnode, so notify.
1645 	 */
1646 	vnevent_mountedover(vp, NULL);
1647 	error = VFS_MOUNT(vfsp, vp, uap, credp);
1648 
1649 	if (uap->flags & MS_RDONLY)
1650 		vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1651 	if (uap->flags & MS_NOSUID)
1652 		vfs_setmntopt(vfsp, MNTOPT_NOSUID, NULL, 0);
1653 	if (uap->flags & MS_GLOBAL)
1654 		vfs_setmntopt(vfsp, MNTOPT_GLOBAL, NULL, 0);
1655 
1656 	if (error) {
1657 		lofi_remove(vfsp);
1658 
1659 		if (remount) {
1660 			/* put back pre-remount options */
1661 			vfs_swapopttbl(&mnt_mntopts, &vfsp->vfs_mntopts);
1662 			vfs_setmntpoint(vfsp, refstr_value(oldmntpt),
1663 			    VFSSP_VERBATIM);
1664 			if (oldmntpt)
1665 				refstr_rele(oldmntpt);
1666 			vfs_setresource(vfsp, refstr_value(oldresource),
1667 			    VFSSP_VERBATIM);
1668 			if (oldresource)
1669 				refstr_rele(oldresource);
1670 			vfsp->vfs_flag = ovflags;
1671 			vfs_unlock(vfsp);
1672 			VFS_RELE(vfsp);
1673 		} else {
1674 			vfs_unlock(vfsp);
1675 			vfs_freemnttab(vfsp);
1676 			vfs_free(vfsp);
1677 		}
1678 	} else {
1679 		/*
1680 		 * Set the mount time to now
1681 		 */
1682 		vfsp->vfs_mtime = ddi_get_time();
1683 		if (remount) {
1684 			vfsp->vfs_flag &= ~VFS_REMOUNT;
1685 			if (oldresource)
1686 				refstr_rele(oldresource);
1687 			if (oldmntpt)
1688 				refstr_rele(oldmntpt);
1689 		} else if (splice) {
1690 			/*
1691 			 * Link vfsp into the name space at the mount
1692 			 * point. Vfs_add() is responsible for
1693 			 * holding the mount point which will be
1694 			 * released when vfs_remove() is called.
1695 			 */
1696 			vfs_add(vp, vfsp, uap->flags);
1697 		} else {
1698 			/*
1699 			 * Hold the reference to file system which is
1700 			 * not linked into the name space.
1701 			 */
1702 			vfsp->vfs_zone = NULL;
1703 			VFS_HOLD(vfsp);
1704 			vfsp->vfs_vnodecovered = NULL;
1705 		}
1706 		/*
1707 		 * Set flags for global options encountered
1708 		 */
1709 		if (vfs_optionisset(vfsp, MNTOPT_RO, NULL))
1710 			vfsp->vfs_flag |= VFS_RDONLY;
1711 		else
1712 			vfsp->vfs_flag &= ~VFS_RDONLY;
1713 		if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
1714 			vfsp->vfs_flag |= (VFS_NOSETUID|VFS_NODEVICES);
1715 		} else {
1716 			if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL))
1717 				vfsp->vfs_flag |= VFS_NODEVICES;
1718 			else
1719 				vfsp->vfs_flag &= ~VFS_NODEVICES;
1720 			if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL))
1721 				vfsp->vfs_flag |= VFS_NOSETUID;
1722 			else
1723 				vfsp->vfs_flag &= ~VFS_NOSETUID;
1724 		}
1725 		if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL))
1726 			vfsp->vfs_flag |= VFS_NBMAND;
1727 		else
1728 			vfsp->vfs_flag &= ~VFS_NBMAND;
1729 
1730 		if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL))
1731 			vfsp->vfs_flag |= VFS_XATTR;
1732 		else
1733 			vfsp->vfs_flag &= ~VFS_XATTR;
1734 
1735 		if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL))
1736 			vfsp->vfs_flag |= VFS_NOEXEC;
1737 		else
1738 			vfsp->vfs_flag &= ~VFS_NOEXEC;
1739 
1740 		/*
1741 		 * Now construct the output option string of options
1742 		 * we recognized.
1743 		 */
1744 		if (uap->flags & MS_OPTIONSTR) {
1745 			vfs_list_read_lock();
1746 			copyout_error = vfs_buildoptionstr(
1747 			    &vfsp->vfs_mntopts, inargs, optlen);
1748 			vfs_list_unlock();
1749 			if (copyout_error == 0 &&
1750 			    (uap->flags & MS_SYSSPACE) == 0) {
1751 				copyout_error = copyoutstr(inargs, opts,
1752 				    optlen, NULL);
1753 			}
1754 		}
1755 
1756 		/*
1757 		 * If this isn't a remount, set up the vopstats before
1758 		 * anyone can touch this. We only allow spliced file
1759 		 * systems (file systems which are in the namespace) to
1760 		 * have the VFS_STATS flag set.
1761 		 * NOTE: PxFS mounts the underlying file system with
1762 		 * MS_NOSPLICE set and copies those vfs_flags to its private
1763 		 * vfs structure. As a result, PxFS should never have
1764 		 * the VFS_STATS flag or else we might access the vfs
1765 		 * statistics-related fields prior to them being
1766 		 * properly initialized.
1767 		 */
1768 		if (!remount && (vswp->vsw_flag & VSW_STATS) && splice) {
1769 			initialize_vopstats(&vfsp->vfs_vopstats);
1770 			/*
1771 			 * We need to set vfs_vskap to NULL because there's
1772 			 * a chance it won't be set below.  This is checked
1773 			 * in teardown_vopstats() so we can't have garbage.
1774 			 */
1775 			vfsp->vfs_vskap = NULL;
1776 			vfsp->vfs_flag |= VFS_STATS;
1777 			vfsp->vfs_fstypevsp = get_fstype_vopstats(vfsp, vswp);
1778 		}
1779 
1780 		if (vswp->vsw_flag & VSW_XID)
1781 			vfsp->vfs_flag |= VFS_XID;
1782 
1783 		vfs_unlock(vfsp);
1784 	}
1785 	mount_completed(zone);
1786 	zone_rele(zone);
1787 	if (splice)
1788 		vn_vfsunlock(vp);
1789 
1790 	if ((error == 0) && (copyout_error == 0)) {
1791 		if (!remount) {
1792 			/*
1793 			 * Don't call get_vskstat_anchor() while holding
1794 			 * locks since it allocates memory and calls
1795 			 * VFS_STATVFS().  For NFS, the latter can generate
1796 			 * an over-the-wire call.
1797 			 */
1798 			vskap = get_vskstat_anchor(vfsp);
1799 			/* Only take the lock if we have something to do */
1800 			if (vskap != NULL) {
1801 				vfs_lock_wait(vfsp);
1802 				if (vfsp->vfs_flag & VFS_STATS) {
1803 					vfsp->vfs_vskap = vskap;
1804 				}
1805 				vfs_unlock(vfsp);
1806 			}
1807 		}
1808 		/* Return vfsp to caller. */
1809 		*vfspp = vfsp;
1810 	}
1811 errout:
1812 	vfs_freeopttbl(&mnt_mntopts);
1813 	if (resource != NULL)
1814 		kmem_free(resource, strlen(resource) + 1);
1815 	if (mountpt != NULL)
1816 		kmem_free(mountpt, strlen(mountpt) + 1);
1817 	/*
1818 	 * It is possible we errored prior to adding to mount in progress
1819 	 * table. Must free vnode we acquired with successful lookupname.
1820 	 */
1821 	if (addmip)
1822 		VN_RELE(bvp);
1823 	if (delmip)
1824 		vfs_delmip(vfsp);
1825 	ASSERT(vswp != NULL);
1826 	vfs_unrefvfssw(vswp);
1827 	if (inargs != opts)
1828 		kmem_free(inargs, MAX_MNTOPT_STR);
1829 	if (copyout_error) {
1830 		lofi_remove(vfsp);
1831 		VFS_RELE(vfsp);
1832 		error = copyout_error;
1833 	}
1834 	return (error);
1835 }
1836 
1837 static void
1838 vfs_setpath(
1839     struct vfs *vfsp,		/* vfs being updated */
1840     refstr_t **refp,		/* Ref-count string to contain the new path */
1841     const char *newpath,	/* Path to add to refp (above) */
1842     uint32_t flag)		/* flag */
1843 {
1844 	size_t len;
1845 	refstr_t *ref;
1846 	zone_t *zone = curproc->p_zone;
1847 	char *sp;
1848 	int have_list_lock = 0;
1849 
1850 	ASSERT(!VFS_ON_LIST(vfsp) || vfs_lock_held(vfsp));
1851 
1852 	/*
1853 	 * New path must be less than MAXPATHLEN because mntfs
1854 	 * will only display up to MAXPATHLEN bytes. This is currently
1855 	 * safe, because domount() uses pn_get(), and other callers
1856 	 * similarly cap the size to fewer than MAXPATHLEN bytes.
1857 	 */
1858 
1859 	ASSERT(strlen(newpath) < MAXPATHLEN);
1860 
1861 	/* mntfs requires consistency while vfs list lock is held */
1862 
1863 	if (VFS_ON_LIST(vfsp)) {
1864 		have_list_lock = 1;
1865 		vfs_list_lock();
1866 	}
1867 
1868 	if (*refp != NULL)
1869 		refstr_rele(*refp);
1870 
1871 	/*
1872 	 * If we are in a non-global zone then we prefix the supplied path,
1873 	 * newpath, with the zone's root path, with two exceptions. The first
1874 	 * is where we have been explicitly directed to avoid doing so; this
1875 	 * will be the case following a failed remount, where the path supplied
1876 	 * will be a saved version which must now be restored. The second
1877 	 * exception is where newpath is not a pathname but a descriptive name,
1878 	 * e.g. "procfs".
1879 	 */
1880 	if (zone == global_zone || (flag & VFSSP_VERBATIM) || *newpath != '/') {
1881 		ref = refstr_alloc(newpath);
1882 		goto out;
1883 	}
1884 
1885 	/*
1886 	 * Truncate the trailing '/' in the zoneroot, and merge
1887 	 * in the zone's rootpath with the "newpath" (resource
1888 	 * or mountpoint) passed in.
1889 	 *
1890 	 * The size of the required buffer is thus the size of
1891 	 * the buffer required for the passed-in newpath
1892 	 * (strlen(newpath) + 1), plus the size of the buffer
1893 	 * required to hold zone_rootpath (zone_rootpathlen)
1894 	 * minus one for one of the now-superfluous NUL
1895 	 * terminations, minus one for the trailing '/'.
1896 	 *
1897 	 * That gives us:
1898 	 *
1899 	 * (strlen(newpath) + 1) + zone_rootpathlen - 1 - 1
1900 	 *
1901 	 * Which is what we have below.
1902 	 */
1903 
1904 	len = strlen(newpath) + zone->zone_rootpathlen - 1;
1905 	sp = kmem_alloc(len, KM_SLEEP);
1906 
1907 	/*
1908 	 * Copy everything including the trailing slash, which
1909 	 * we then overwrite with the NUL character.
1910 	 */
1911 
1912 	(void) strcpy(sp, zone->zone_rootpath);
1913 	sp[zone->zone_rootpathlen - 2] = '\0';
1914 	(void) strcat(sp, newpath);
1915 
1916 	ref = refstr_alloc(sp);
1917 	kmem_free(sp, len);
1918 out:
1919 	*refp = ref;
1920 
1921 	if (have_list_lock) {
1922 		vfs_mnttab_modtimeupd();
1923 		vfs_list_unlock();
1924 	}
1925 }
1926 
1927 /*
1928  * Record a mounted resource name in a vfs structure.
1929  * If vfsp is already mounted, caller must hold the vfs lock.
1930  */
1931 void
1932 vfs_setresource(struct vfs *vfsp, const char *resource, uint32_t flag)
1933 {
1934 	if (resource == NULL || resource[0] == '\0')
1935 		resource = VFS_NORESOURCE;
1936 	vfs_setpath(vfsp, &vfsp->vfs_resource, resource, flag);
1937 }
1938 
1939 /*
1940  * Record a mount point name in a vfs structure.
1941  * If vfsp is already mounted, caller must hold the vfs lock.
1942  */
1943 void
1944 vfs_setmntpoint(struct vfs *vfsp, const char *mntpt, uint32_t flag)
1945 {
1946 	if (mntpt == NULL || mntpt[0] == '\0')
1947 		mntpt = VFS_NOMNTPT;
1948 	vfs_setpath(vfsp, &vfsp->vfs_mntpt, mntpt, flag);
1949 }
1950 
1951 /* Returns the vfs_resource. Caller must call refstr_rele() when finished. */
1952 
1953 refstr_t *
1954 vfs_getresource(const struct vfs *vfsp)
1955 {
1956 	refstr_t *resource;
1957 
1958 	vfs_list_read_lock();
1959 	resource = vfsp->vfs_resource;
1960 	refstr_hold(resource);
1961 	vfs_list_unlock();
1962 
1963 	return (resource);
1964 }
1965 
1966 /* Returns the vfs_mntpt. Caller must call refstr_rele() when finished. */
1967 
1968 refstr_t *
1969 vfs_getmntpoint(const struct vfs *vfsp)
1970 {
1971 	refstr_t *mntpt;
1972 
1973 	vfs_list_read_lock();
1974 	mntpt = vfsp->vfs_mntpt;
1975 	refstr_hold(mntpt);
1976 	vfs_list_unlock();
1977 
1978 	return (mntpt);
1979 }
1980 
1981 /*
1982  * Create an empty options table with enough empty slots to hold all
1983  * The options in the options string passed as an argument.
1984  * Potentially prepend another options table.
1985  *
1986  * Note: caller is responsible for locking the vfs list, if needed,
1987  *       to protect mops.
1988  */
1989 static void
1990 vfs_createopttbl_extend(mntopts_t *mops, const char *opts,
1991     const mntopts_t *mtmpl)
1992 {
1993 	const char *s = opts;
1994 	uint_t count;
1995 
1996 	if (opts == NULL || *opts == '\0') {
1997 		count = 0;
1998 	} else {
1999 		count = 1;
2000 
2001 		/*
2002 		 * Count number of options in the string
2003 		 */
2004 		for (s = strchr(s, ','); s != NULL; s = strchr(s, ',')) {
2005 			count++;
2006 			s++;
2007 		}
2008 	}
2009 	vfs_copyopttbl_extend(mtmpl, mops, count);
2010 }
2011 
2012 /*
2013  * Create an empty options table with enough empty slots to hold all
2014  * The options in the options string passed as an argument.
2015  *
2016  * This function is *not* for general use by filesystems.
2017  *
2018  * Note: caller is responsible for locking the vfs list, if needed,
2019  *       to protect mops.
2020  */
2021 void
2022 vfs_createopttbl(mntopts_t *mops, const char *opts)
2023 {
2024 	vfs_createopttbl_extend(mops, opts, NULL);
2025 }
2026 
2027 
2028 /*
2029  * Swap two mount options tables
2030  */
2031 static void
2032 vfs_swapopttbl_nolock(mntopts_t *optbl1, mntopts_t *optbl2)
2033 {
2034 	uint_t tmpcnt;
2035 	mntopt_t *tmplist;
2036 
2037 	tmpcnt = optbl2->mo_count;
2038 	tmplist = optbl2->mo_list;
2039 	optbl2->mo_count = optbl1->mo_count;
2040 	optbl2->mo_list = optbl1->mo_list;
2041 	optbl1->mo_count = tmpcnt;
2042 	optbl1->mo_list = tmplist;
2043 }
2044 
2045 static void
2046 vfs_swapopttbl(mntopts_t *optbl1, mntopts_t *optbl2)
2047 {
2048 	vfs_list_lock();
2049 	vfs_swapopttbl_nolock(optbl1, optbl2);
2050 	vfs_mnttab_modtimeupd();
2051 	vfs_list_unlock();
2052 }
2053 
2054 static char **
2055 vfs_copycancelopt_extend(char **const moc, int extend)
2056 {
2057 	int i = 0;
2058 	int j;
2059 	char **result;
2060 
2061 	if (moc != NULL) {
2062 		for (; moc[i] != NULL; i++)
2063 			/* count number of options to cancel */;
2064 	}
2065 
2066 	if (i + extend == 0)
2067 		return (NULL);
2068 
2069 	result = kmem_alloc((i + extend + 1) * sizeof (char *), KM_SLEEP);
2070 
2071 	for (j = 0; j < i; j++) {
2072 		result[j] = kmem_alloc(strlen(moc[j]) + 1, KM_SLEEP);
2073 		(void) strcpy(result[j], moc[j]);
2074 	}
2075 	for (; j <= i + extend; j++)
2076 		result[j] = NULL;
2077 
2078 	return (result);
2079 }
2080 
2081 static void
2082 vfs_copyopt(const mntopt_t *s, mntopt_t *d)
2083 {
2084 	char *sp, *dp;
2085 
2086 	d->mo_flags = s->mo_flags;
2087 	d->mo_data = s->mo_data;
2088 	sp = s->mo_name;
2089 	if (sp != NULL) {
2090 		dp = kmem_alloc(strlen(sp) + 1, KM_SLEEP);
2091 		(void) strcpy(dp, sp);
2092 		d->mo_name = dp;
2093 	} else {
2094 		d->mo_name = NULL; /* should never happen */
2095 	}
2096 
2097 	d->mo_cancel = vfs_copycancelopt_extend(s->mo_cancel, 0);
2098 
2099 	sp = s->mo_arg;
2100 	if (sp != NULL) {
2101 		dp = kmem_alloc(strlen(sp) + 1, KM_SLEEP);
2102 		(void) strcpy(dp, sp);
2103 		d->mo_arg = dp;
2104 	} else {
2105 		d->mo_arg = NULL;
2106 	}
2107 }
2108 
2109 /*
2110  * Copy a mount options table, possibly allocating some spare
2111  * slots at the end.  It is permissible to copy_extend the NULL table.
2112  */
2113 static void
2114 vfs_copyopttbl_extend(const mntopts_t *smo, mntopts_t *dmo, int extra)
2115 {
2116 	uint_t i, count;
2117 	mntopt_t *motbl;
2118 
2119 	/*
2120 	 * Clear out any existing stuff in the options table being initialized
2121 	 */
2122 	vfs_freeopttbl(dmo);
2123 	count = (smo == NULL) ? 0 : smo->mo_count;
2124 	if ((count + extra) == 0)	/* nothing to do */
2125 		return;
2126 	dmo->mo_count = count + extra;
2127 	motbl = kmem_zalloc((count + extra) * sizeof (mntopt_t), KM_SLEEP);
2128 	dmo->mo_list = motbl;
2129 	for (i = 0; i < count; i++) {
2130 		vfs_copyopt(&smo->mo_list[i], &motbl[i]);
2131 	}
2132 	for (i = count; i < count + extra; i++) {
2133 		motbl[i].mo_flags = MO_EMPTY;
2134 	}
2135 }
2136 
2137 /*
2138  * Copy a mount options table.
2139  *
2140  * This function is *not* for general use by filesystems.
2141  *
2142  * Note: caller is responsible for locking the vfs list, if needed,
2143  *       to protect smo and dmo.
2144  */
2145 void
2146 vfs_copyopttbl(const mntopts_t *smo, mntopts_t *dmo)
2147 {
2148 	vfs_copyopttbl_extend(smo, dmo, 0);
2149 }
2150 
2151 static char **
2152 vfs_mergecancelopts(const mntopt_t *mop1, const mntopt_t *mop2)
2153 {
2154 	int c1 = 0;
2155 	int c2 = 0;
2156 	char **result;
2157 	char **sp1, **sp2, **dp;
2158 
2159 	/*
2160 	 * First we count both lists of cancel options.
2161 	 * If either is NULL or has no elements, we return a copy of
2162 	 * the other.
2163 	 */
2164 	if (mop1->mo_cancel != NULL) {
2165 		for (; mop1->mo_cancel[c1] != NULL; c1++)
2166 			/* count cancel options in mop1 */;
2167 	}
2168 
2169 	if (c1 == 0)
2170 		return (vfs_copycancelopt_extend(mop2->mo_cancel, 0));
2171 
2172 	if (mop2->mo_cancel != NULL) {
2173 		for (; mop2->mo_cancel[c2] != NULL; c2++)
2174 			/* count cancel options in mop2 */;
2175 	}
2176 
2177 	result = vfs_copycancelopt_extend(mop1->mo_cancel, c2);
2178 
2179 	if (c2 == 0)
2180 		return (result);
2181 
2182 	/*
2183 	 * When we get here, we've got two sets of cancel options;
2184 	 * we need to merge the two sets.  We know that the result
2185 	 * array has "c1+c2+1" entries and in the end we might shrink
2186 	 * it.
2187 	 * Result now has a copy of the c1 entries from mop1; we'll
2188 	 * now lookup all the entries of mop2 in mop1 and copy it if
2189 	 * it is unique.
2190 	 * This operation is O(n^2) but it's only called once per
2191 	 * filesystem per duplicate option.  This is a situation
2192 	 * which doesn't arise with the filesystems in ON and
2193 	 * n is generally 1.
2194 	 */
2195 
2196 	dp = &result[c1];
2197 	for (sp2 = mop2->mo_cancel; *sp2 != NULL; sp2++) {
2198 		for (sp1 = mop1->mo_cancel; *sp1 != NULL; sp1++) {
2199 			if (strcmp(*sp1, *sp2) == 0)
2200 				break;
2201 		}
2202 		if (*sp1 == NULL) {
2203 			/*
2204 			 * Option *sp2 not found in mop1, so copy it.
2205 			 * The calls to vfs_copycancelopt_extend()
2206 			 * guarantee that there's enough room.
2207 			 */
2208 			*dp = kmem_alloc(strlen(*sp2) + 1, KM_SLEEP);
2209 			(void) strcpy(*dp++, *sp2);
2210 		}
2211 	}
2212 	if (dp != &result[c1+c2]) {
2213 		size_t bytes = (dp - result + 1) * sizeof (char *);
2214 		char **nres = kmem_alloc(bytes, KM_SLEEP);
2215 
2216 		bcopy(result, nres, bytes);
2217 		kmem_free(result, (c1 + c2 + 1) * sizeof (char *));
2218 		result = nres;
2219 	}
2220 	return (result);
2221 }
2222 
2223 /*
2224  * Merge two mount option tables (outer and inner) into one.  This is very
2225  * similar to "merging" global variables and automatic variables in C.
2226  *
2227  * This isn't (and doesn't have to be) fast.
2228  *
2229  * This function is *not* for general use by filesystems.
2230  *
2231  * Note: caller is responsible for locking the vfs list, if needed,
2232  *       to protect omo, imo & dmo.
2233  */
2234 void
2235 vfs_mergeopttbl(const mntopts_t *omo, const mntopts_t *imo, mntopts_t *dmo)
2236 {
2237 	uint_t i, count;
2238 	mntopt_t *mop, *motbl;
2239 	uint_t freeidx;
2240 
2241 	/*
2242 	 * First determine how much space we need to allocate.
2243 	 */
2244 	count = omo->mo_count;
2245 	for (i = 0; i < imo->mo_count; i++) {
2246 		if (imo->mo_list[i].mo_flags & MO_EMPTY)
2247 			continue;
2248 		if (vfs_hasopt(omo, imo->mo_list[i].mo_name) == NULL)
2249 			count++;
2250 	}
2251 	ASSERT(count >= omo->mo_count &&
2252 	    count <= omo->mo_count + imo->mo_count);
2253 	motbl = kmem_alloc(count * sizeof (mntopt_t), KM_SLEEP);
2254 	for (i = 0; i < omo->mo_count; i++)
2255 		vfs_copyopt(&omo->mo_list[i], &motbl[i]);
2256 	freeidx = omo->mo_count;
2257 	for (i = 0; i < imo->mo_count; i++) {
2258 		if (imo->mo_list[i].mo_flags & MO_EMPTY)
2259 			continue;
2260 		if ((mop = vfs_hasopt(omo, imo->mo_list[i].mo_name)) != NULL) {
2261 			char **newcanp;
2262 			uint_t index = mop - omo->mo_list;
2263 
2264 			newcanp = vfs_mergecancelopts(mop, &motbl[index]);
2265 
2266 			vfs_freeopt(&motbl[index]);
2267 			vfs_copyopt(&imo->mo_list[i], &motbl[index]);
2268 
2269 			vfs_freecancelopt(motbl[index].mo_cancel);
2270 			motbl[index].mo_cancel = newcanp;
2271 		} else {
2272 			/*
2273 			 * If it's a new option, just copy it over to the first
2274 			 * free location.
2275 			 */
2276 			vfs_copyopt(&imo->mo_list[i], &motbl[freeidx++]);
2277 		}
2278 	}
2279 	dmo->mo_count = count;
2280 	dmo->mo_list = motbl;
2281 }
2282 
2283 /*
2284  * Functions to set and clear mount options in a mount options table.
2285  */
2286 
2287 /*
2288  * Clear a mount option, if it exists.
2289  *
2290  * The update_mnttab arg indicates whether mops is part of a vfs that is on
2291  * the vfs list.
2292  */
2293 static void
2294 vfs_clearmntopt_nolock(mntopts_t *mops, const char *opt, int update_mnttab)
2295 {
2296 	struct mntopt *mop;
2297 	uint_t i, count;
2298 
2299 	ASSERT(!update_mnttab || RW_WRITE_HELD(&vfslist));
2300 
2301 	count = mops->mo_count;
2302 	for (i = 0; i < count; i++) {
2303 		mop = &mops->mo_list[i];
2304 
2305 		if (mop->mo_flags & MO_EMPTY)
2306 			continue;
2307 		if (strcmp(opt, mop->mo_name))
2308 			continue;
2309 		mop->mo_flags &= ~MO_SET;
2310 		if (mop->mo_arg != NULL) {
2311 			kmem_free(mop->mo_arg, strlen(mop->mo_arg) + 1);
2312 		}
2313 		mop->mo_arg = NULL;
2314 		if (update_mnttab)
2315 			vfs_mnttab_modtimeupd();
2316 		break;
2317 	}
2318 }
2319 
2320 void
2321 vfs_clearmntopt(struct vfs *vfsp, const char *opt)
2322 {
2323 	int gotlock = 0;
2324 
2325 	if (VFS_ON_LIST(vfsp)) {
2326 		gotlock = 1;
2327 		vfs_list_lock();
2328 	}
2329 	vfs_clearmntopt_nolock(&vfsp->vfs_mntopts, opt, gotlock);
2330 	if (gotlock)
2331 		vfs_list_unlock();
2332 }
2333 
2334 
2335 /*
2336  * Set a mount option on.  If it's not found in the table, it's silently
2337  * ignored.  If the option has MO_IGNORE set, it is still set unless the
2338  * VFS_NOFORCEOPT bit is set in the flags.  Also, VFS_DISPLAY/VFS_NODISPLAY flag
2339  * bits can be used to toggle the MO_NODISPLAY bit for the option.
2340  * If the VFS_CREATEOPT flag bit is set then the first option slot with
2341  * MO_EMPTY set is created as the option passed in.
2342  *
2343  * The update_mnttab arg indicates whether mops is part of a vfs that is on
2344  * the vfs list.
2345  */
2346 static void
2347 vfs_setmntopt_nolock(mntopts_t *mops, const char *opt,
2348     const char *arg, int flags, int update_mnttab)
2349 {
2350 	mntopt_t *mop;
2351 	uint_t i, count;
2352 	char *sp;
2353 
2354 	ASSERT(!update_mnttab || RW_WRITE_HELD(&vfslist));
2355 
2356 	if (flags & VFS_CREATEOPT) {
2357 		if (vfs_hasopt(mops, opt) != NULL) {
2358 			flags &= ~VFS_CREATEOPT;
2359 		}
2360 	}
2361 	count = mops->mo_count;
2362 	for (i = 0; i < count; i++) {
2363 		mop = &mops->mo_list[i];
2364 
2365 		if (mop->mo_flags & MO_EMPTY) {
2366 			if ((flags & VFS_CREATEOPT) == 0)
2367 				continue;
2368 			sp = kmem_alloc(strlen(opt) + 1, KM_SLEEP);
2369 			(void) strcpy(sp, opt);
2370 			mop->mo_name = sp;
2371 			if (arg != NULL)
2372 				mop->mo_flags = MO_HASVALUE;
2373 			else
2374 				mop->mo_flags = 0;
2375 		} else if (strcmp(opt, mop->mo_name)) {
2376 			continue;
2377 		}
2378 		if ((mop->mo_flags & MO_IGNORE) && (flags & VFS_NOFORCEOPT))
2379 			break;
2380 		if (arg != NULL && (mop->mo_flags & MO_HASVALUE) != 0) {
2381 			sp = kmem_alloc(strlen(arg) + 1, KM_SLEEP);
2382 			(void) strcpy(sp, arg);
2383 		} else {
2384 			sp = NULL;
2385 		}
2386 		if (mop->mo_arg != NULL)
2387 			kmem_free(mop->mo_arg, strlen(mop->mo_arg) + 1);
2388 		mop->mo_arg = sp;
2389 		if (flags & VFS_DISPLAY)
2390 			mop->mo_flags &= ~MO_NODISPLAY;
2391 		if (flags & VFS_NODISPLAY)
2392 			mop->mo_flags |= MO_NODISPLAY;
2393 		mop->mo_flags |= MO_SET;
2394 		if (mop->mo_cancel != NULL) {
2395 			char **cp;
2396 
2397 			for (cp = mop->mo_cancel; *cp != NULL; cp++)
2398 				vfs_clearmntopt_nolock(mops, *cp, 0);
2399 		}
2400 		if (update_mnttab)
2401 			vfs_mnttab_modtimeupd();
2402 		break;
2403 	}
2404 }
2405 
2406 void
2407 vfs_setmntopt(struct vfs *vfsp, const char *opt, const char *arg, int flags)
2408 {
2409 	int gotlock = 0;
2410 
2411 	if (VFS_ON_LIST(vfsp)) {
2412 		gotlock = 1;
2413 		vfs_list_lock();
2414 	}
2415 	vfs_setmntopt_nolock(&vfsp->vfs_mntopts, opt, arg, flags, gotlock);
2416 	if (gotlock)
2417 		vfs_list_unlock();
2418 }
2419 
2420 
2421 /*
2422  * Add a "tag" option to a mounted file system's options list.
2423  *
2424  * Note: caller is responsible for locking the vfs list, if needed,
2425  *       to protect mops.
2426  */
2427 static mntopt_t *
2428 vfs_addtag(mntopts_t *mops, const char *tag)
2429 {
2430 	uint_t count;
2431 	mntopt_t *mop, *motbl;
2432 
2433 	count = mops->mo_count + 1;
2434 	motbl = kmem_zalloc(count * sizeof (mntopt_t), KM_SLEEP);
2435 	if (mops->mo_count) {
2436 		size_t len = (count - 1) * sizeof (mntopt_t);
2437 
2438 		bcopy(mops->mo_list, motbl, len);
2439 		kmem_free(mops->mo_list, len);
2440 	}
2441 	mops->mo_count = count;
2442 	mops->mo_list = motbl;
2443 	mop = &motbl[count - 1];
2444 	mop->mo_flags = MO_TAG;
2445 	mop->mo_name = kmem_alloc(strlen(tag) + 1, KM_SLEEP);
2446 	(void) strcpy(mop->mo_name, tag);
2447 	return (mop);
2448 }
2449 
2450 /*
2451  * Allow users to set arbitrary "tags" in a vfs's mount options.
2452  * Broader use within the kernel is discouraged.
2453  */
2454 int
2455 vfs_settag(uint_t major, uint_t minor, const char *mntpt, const char *tag,
2456     cred_t *cr)
2457 {
2458 	vfs_t *vfsp;
2459 	mntopts_t *mops;
2460 	mntopt_t *mop;
2461 	int found = 0;
2462 	dev_t dev = makedevice(major, minor);
2463 	int err = 0;
2464 	char *buf = kmem_alloc(MAX_MNTOPT_STR, KM_SLEEP);
2465 
2466 	/*
2467 	 * Find the desired mounted file system
2468 	 */
2469 	vfs_list_lock();
2470 	vfsp = rootvfs;
2471 	do {
2472 		if (vfsp->vfs_dev == dev &&
2473 		    strcmp(mntpt, refstr_value(vfsp->vfs_mntpt)) == 0) {
2474 			found = 1;
2475 			break;
2476 		}
2477 		vfsp = vfsp->vfs_next;
2478 	} while (vfsp != rootvfs);
2479 
2480 	if (!found) {
2481 		err = EINVAL;
2482 		goto out;
2483 	}
2484 	err = secpolicy_fs_config(cr, vfsp);
2485 	if (err != 0)
2486 		goto out;
2487 
2488 	mops = &vfsp->vfs_mntopts;
2489 	/*
2490 	 * Add tag if it doesn't already exist
2491 	 */
2492 	if ((mop = vfs_hasopt(mops, tag)) == NULL) {
2493 		int len;
2494 
2495 		(void) vfs_buildoptionstr(mops, buf, MAX_MNTOPT_STR);
2496 		len = strlen(buf);
2497 		if (len + strlen(tag) + 2 > MAX_MNTOPT_STR) {
2498 			err = ENAMETOOLONG;
2499 			goto out;
2500 		}
2501 		mop = vfs_addtag(mops, tag);
2502 	}
2503 	if ((mop->mo_flags & MO_TAG) == 0) {
2504 		err = EINVAL;
2505 		goto out;
2506 	}
2507 	vfs_setmntopt_nolock(mops, tag, NULL, 0, 1);
2508 out:
2509 	vfs_list_unlock();
2510 	kmem_free(buf, MAX_MNTOPT_STR);
2511 	return (err);
2512 }
2513 
2514 /*
2515  * Allow users to remove arbitrary "tags" in a vfs's mount options.
2516  * Broader use within the kernel is discouraged.
2517  */
2518 int
2519 vfs_clrtag(uint_t major, uint_t minor, const char *mntpt, const char *tag,
2520     cred_t *cr)
2521 {
2522 	vfs_t *vfsp;
2523 	mntopt_t *mop;
2524 	int found = 0;
2525 	dev_t dev = makedevice(major, minor);
2526 	int err = 0;
2527 
2528 	/*
2529 	 * Find the desired mounted file system
2530 	 */
2531 	vfs_list_lock();
2532 	vfsp = rootvfs;
2533 	do {
2534 		if (vfsp->vfs_dev == dev &&
2535 		    strcmp(mntpt, refstr_value(vfsp->vfs_mntpt)) == 0) {
2536 			found = 1;
2537 			break;
2538 		}
2539 		vfsp = vfsp->vfs_next;
2540 	} while (vfsp != rootvfs);
2541 
2542 	if (!found) {
2543 		err = EINVAL;
2544 		goto out;
2545 	}
2546 	err = secpolicy_fs_config(cr, vfsp);
2547 	if (err != 0)
2548 		goto out;
2549 
2550 	if ((mop = vfs_hasopt(&vfsp->vfs_mntopts, tag)) == NULL) {
2551 		err = EINVAL;
2552 		goto out;
2553 	}
2554 	if ((mop->mo_flags & MO_TAG) == 0) {
2555 		err = EINVAL;
2556 		goto out;
2557 	}
2558 	vfs_clearmntopt_nolock(&vfsp->vfs_mntopts, tag, 1);
2559 out:
2560 	vfs_list_unlock();
2561 	return (err);
2562 }
2563 
2564 /*
2565  * Function to parse an option string and fill in a mount options table.
2566  * Unknown options are silently ignored.  The input option string is modified
2567  * by replacing separators with nulls.  If the create flag is set, options
2568  * not found in the table are just added on the fly.  The table must have
2569  * an option slot marked MO_EMPTY to add an option on the fly.
2570  *
2571  * This function is *not* for general use by filesystems.
2572  *
2573  * Note: caller is responsible for locking the vfs list, if needed,
2574  *       to protect mops..
2575  */
2576 void
2577 vfs_parsemntopts(mntopts_t *mops, char *osp, int create)
2578 {
2579 	char *s = osp, *p, *nextop, *valp, *cp, *ep;
2580 	int setflg = VFS_NOFORCEOPT;
2581 
2582 	if (osp == NULL)
2583 		return;
2584 	while (*s != '\0') {
2585 		p = strchr(s, ',');	/* find next option */
2586 		if (p == NULL) {
2587 			cp = NULL;
2588 			p = s + strlen(s);
2589 		} else {
2590 			cp = p;		/* save location of comma */
2591 			*p++ = '\0';	/* mark end and point to next option */
2592 		}
2593 		nextop = p;
2594 		p = strchr(s, '=');	/* look for value */
2595 		if (p == NULL) {
2596 			valp = NULL;	/* no value supplied */
2597 		} else {
2598 			ep = p;		/* save location of equals */
2599 			*p++ = '\0';	/* end option and point to value */
2600 			valp = p;
2601 		}
2602 		/*
2603 		 * set option into options table
2604 		 */
2605 		if (create)
2606 			setflg |= VFS_CREATEOPT;
2607 		vfs_setmntopt_nolock(mops, s, valp, setflg, 0);
2608 		if (cp != NULL)
2609 			*cp = ',';	/* restore the comma */
2610 		if (valp != NULL)
2611 			*ep = '=';	/* restore the equals */
2612 		s = nextop;
2613 	}
2614 }
2615 
2616 /*
2617  * Function to inquire if an option exists in a mount options table.
2618  * Returns a pointer to the option if it exists, else NULL.
2619  *
2620  * This function is *not* for general use by filesystems.
2621  *
2622  * Note: caller is responsible for locking the vfs list, if needed,
2623  *       to protect mops.
2624  */
2625 struct mntopt *
2626 vfs_hasopt(const mntopts_t *mops, const char *opt)
2627 {
2628 	struct mntopt *mop;
2629 	uint_t i, count;
2630 
2631 	count = mops->mo_count;
2632 	for (i = 0; i < count; i++) {
2633 		mop = &mops->mo_list[i];
2634 
2635 		if (mop->mo_flags & MO_EMPTY)
2636 			continue;
2637 		if (strcmp(opt, mop->mo_name) == 0)
2638 			return (mop);
2639 	}
2640 	return (NULL);
2641 }
2642 
2643 /*
2644  * Function to inquire if an option is set in a mount options table.
2645  * Returns non-zero if set and fills in the arg pointer with a pointer to
2646  * the argument string or NULL if there is no argument string.
2647  */
2648 static int
2649 vfs_optionisset_nolock(const mntopts_t *mops, const char *opt, char **argp)
2650 {
2651 	struct mntopt *mop;
2652 	uint_t i, count;
2653 
2654 	count = mops->mo_count;
2655 	for (i = 0; i < count; i++) {
2656 		mop = &mops->mo_list[i];
2657 
2658 		if (mop->mo_flags & MO_EMPTY)
2659 			continue;
2660 		if (strcmp(opt, mop->mo_name))
2661 			continue;
2662 		if ((mop->mo_flags & MO_SET) == 0)
2663 			return (0);
2664 		if (argp != NULL && (mop->mo_flags & MO_HASVALUE) != 0)
2665 			*argp = mop->mo_arg;
2666 		return (1);
2667 	}
2668 	return (0);
2669 }
2670 
2671 
2672 int
2673 vfs_optionisset(const struct vfs *vfsp, const char *opt, char **argp)
2674 {
2675 	int ret;
2676 
2677 	vfs_list_read_lock();
2678 	ret = vfs_optionisset_nolock(&vfsp->vfs_mntopts, opt, argp);
2679 	vfs_list_unlock();
2680 	return (ret);
2681 }
2682 
2683 
2684 /*
2685  * Construct a comma separated string of the options set in the given
2686  * mount table, return the string in the given buffer.  Return non-zero if
2687  * the buffer would overflow.
2688  *
2689  * This function is *not* for general use by filesystems.
2690  *
2691  * Note: caller is responsible for locking the vfs list, if needed,
2692  *       to protect mp.
2693  */
2694 int
2695 vfs_buildoptionstr(const mntopts_t *mp, char *buf, int len)
2696 {
2697 	char *cp;
2698 	uint_t i;
2699 
2700 	buf[0] = '\0';
2701 	cp = buf;
2702 	for (i = 0; i < mp->mo_count; i++) {
2703 		struct mntopt *mop;
2704 
2705 		mop = &mp->mo_list[i];
2706 		if (mop->mo_flags & MO_SET) {
2707 			int optlen, comma = 0;
2708 
2709 			if (buf[0] != '\0')
2710 				comma = 1;
2711 			optlen = strlen(mop->mo_name);
2712 			if (strlen(buf) + comma + optlen + 1 > len)
2713 				goto err;
2714 			if (comma)
2715 				*cp++ = ',';
2716 			(void) strcpy(cp, mop->mo_name);
2717 			cp += optlen;
2718 			/*
2719 			 * Append option value if there is one
2720 			 */
2721 			if (mop->mo_arg != NULL) {
2722 				int arglen;
2723 
2724 				arglen = strlen(mop->mo_arg);
2725 				if (strlen(buf) + arglen + 2 > len)
2726 					goto err;
2727 				*cp++ = '=';
2728 				(void) strcpy(cp, mop->mo_arg);
2729 				cp += arglen;
2730 			}
2731 		}
2732 	}
2733 	return (0);
2734 err:
2735 	return (EOVERFLOW);
2736 }
2737 
2738 static void
2739 vfs_freecancelopt(char **moc)
2740 {
2741 	if (moc != NULL) {
2742 		int ccnt = 0;
2743 		char **cp;
2744 
2745 		for (cp = moc; *cp != NULL; cp++) {
2746 			kmem_free(*cp, strlen(*cp) + 1);
2747 			ccnt++;
2748 		}
2749 		kmem_free(moc, (ccnt + 1) * sizeof (char *));
2750 	}
2751 }
2752 
2753 static void
2754 vfs_freeopt(mntopt_t *mop)
2755 {
2756 	if (mop->mo_name != NULL)
2757 		kmem_free(mop->mo_name, strlen(mop->mo_name) + 1);
2758 
2759 	vfs_freecancelopt(mop->mo_cancel);
2760 
2761 	if (mop->mo_arg != NULL)
2762 		kmem_free(mop->mo_arg, strlen(mop->mo_arg) + 1);
2763 }
2764 
2765 /*
2766  * Free a mount options table
2767  *
2768  * This function is *not* for general use by filesystems.
2769  *
2770  * Note: caller is responsible for locking the vfs list, if needed,
2771  *       to protect mp.
2772  */
2773 void
2774 vfs_freeopttbl(mntopts_t *mp)
2775 {
2776 	uint_t i, count;
2777 
2778 	count = mp->mo_count;
2779 	for (i = 0; i < count; i++) {
2780 		vfs_freeopt(&mp->mo_list[i]);
2781 	}
2782 	if (count) {
2783 		kmem_free(mp->mo_list, sizeof (mntopt_t) * count);
2784 		mp->mo_count = 0;
2785 		mp->mo_list = NULL;
2786 	}
2787 }
2788 
2789 
2790 /* ARGSUSED */
2791 static int
2792 vfs_mntdummyread(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cred,
2793     caller_context_t *ct)
2794 {
2795 	return (0);
2796 }
2797 
2798 /* ARGSUSED */
2799 static int
2800 vfs_mntdummywrite(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cred,
2801     caller_context_t *ct)
2802 {
2803 	return (0);
2804 }
2805 
2806 /*
2807  * The dummy vnode is currently used only by file events notification
2808  * module which is just interested in the timestamps.
2809  */
2810 /* ARGSUSED */
2811 static int
2812 vfs_mntdummygetattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
2813     caller_context_t *ct)
2814 {
2815 	bzero(vap, sizeof (vattr_t));
2816 	vap->va_type = VREG;
2817 	vap->va_nlink = 1;
2818 	vap->va_ctime = vfs_mnttab_ctime;
2819 	/*
2820 	 * it is ok to just copy mtime as the time will be monotonically
2821 	 * increasing.
2822 	 */
2823 	vap->va_mtime = vfs_mnttab_mtime;
2824 	vap->va_atime = vap->va_mtime;
2825 	return (0);
2826 }
2827 
2828 static void
2829 vfs_mnttabvp_setup(void)
2830 {
2831 	vnode_t *tvp;
2832 	vnodeops_t *vfs_mntdummyvnops;
2833 	const fs_operation_def_t mnt_dummyvnodeops_template[] = {
2834 		VOPNAME_READ,		{ .vop_read = vfs_mntdummyread },
2835 		VOPNAME_WRITE,		{ .vop_write = vfs_mntdummywrite },
2836 		VOPNAME_GETATTR,	{ .vop_getattr = vfs_mntdummygetattr },
2837 		VOPNAME_VNEVENT,	{ .vop_vnevent = fs_vnevent_support },
2838 		NULL,			NULL
2839 	};
2840 
2841 	if (vn_make_ops("mnttab", mnt_dummyvnodeops_template,
2842 	    &vfs_mntdummyvnops) != 0) {
2843 		cmn_err(CE_WARN, "vfs_mnttabvp_setup: vn_make_ops failed");
2844 		/* Shouldn't happen, but not bad enough to panic */
2845 		return;
2846 	}
2847 
2848 	/*
2849 	 * A global dummy vnode is allocated to represent mntfs files.
2850 	 * The mntfs file (/etc/mnttab) can be monitored for file events
2851 	 * and receive an event when mnttab changes. Dummy VOP calls
2852 	 * will be made on this vnode. The file events notification module
2853 	 * intercepts this vnode and delivers relevant events.
2854 	 */
2855 	tvp = vn_alloc(KM_SLEEP);
2856 	tvp->v_flag = VNOMOUNT|VNOMAP|VNOSWAP|VNOCACHE;
2857 	vn_setops(tvp, vfs_mntdummyvnops);
2858 	tvp->v_type = VREG;
2859 	/*
2860 	 * The mnt dummy ops do not reference v_data.
2861 	 * No other module intercepting this vnode should either.
2862 	 * Just set it to point to itself.
2863 	 */
2864 	tvp->v_data = (caddr_t)tvp;
2865 	tvp->v_vfsp = rootvfs;
2866 	vfs_mntdummyvp = tvp;
2867 }
2868 
2869 /*
2870  * performs fake read/write ops
2871  */
2872 static void
2873 vfs_mnttab_rwop(int rw)
2874 {
2875 	struct uio	uio;
2876 	struct iovec	iov;
2877 	char	buf[1];
2878 
2879 	if (vfs_mntdummyvp == NULL)
2880 		return;
2881 
2882 	bzero(&uio, sizeof (uio));
2883 	bzero(&iov, sizeof (iov));
2884 	iov.iov_base = buf;
2885 	iov.iov_len = 0;
2886 	uio.uio_iov = &iov;
2887 	uio.uio_iovcnt = 1;
2888 	uio.uio_loffset = 0;
2889 	uio.uio_segflg = UIO_SYSSPACE;
2890 	uio.uio_resid = 0;
2891 	if (rw) {
2892 		(void) VOP_WRITE(vfs_mntdummyvp, &uio, 0, kcred, NULL);
2893 	} else {
2894 		(void) VOP_READ(vfs_mntdummyvp, &uio, 0, kcred, NULL);
2895 	}
2896 }
2897 
2898 /*
2899  * Generate a write operation.
2900  */
2901 void
2902 vfs_mnttab_writeop(void)
2903 {
2904 	vfs_mnttab_rwop(1);
2905 }
2906 
2907 /*
2908  * Generate a read operation.
2909  */
2910 void
2911 vfs_mnttab_readop(void)
2912 {
2913 	vfs_mnttab_rwop(0);
2914 }
2915 
2916 /*
2917  * Free any mnttab information recorded in the vfs struct.
2918  * The vfs must not be on the vfs list.
2919  */
2920 static void
2921 vfs_freemnttab(struct vfs *vfsp)
2922 {
2923 	ASSERT(!VFS_ON_LIST(vfsp));
2924 
2925 	/*
2926 	 * Free device and mount point information
2927 	 */
2928 	if (vfsp->vfs_mntpt != NULL) {
2929 		refstr_rele(vfsp->vfs_mntpt);
2930 		vfsp->vfs_mntpt = NULL;
2931 	}
2932 	if (vfsp->vfs_resource != NULL) {
2933 		refstr_rele(vfsp->vfs_resource);
2934 		vfsp->vfs_resource = NULL;
2935 	}
2936 	/*
2937 	 * Now free mount options information
2938 	 */
2939 	vfs_freeopttbl(&vfsp->vfs_mntopts);
2940 }
2941 
2942 /*
2943  * Return the last mnttab modification time
2944  */
2945 void
2946 vfs_mnttab_modtime(timespec_t *ts)
2947 {
2948 	ASSERT(RW_LOCK_HELD(&vfslist));
2949 	*ts = vfs_mnttab_mtime;
2950 }
2951 
2952 /*
2953  * See if mnttab is changed
2954  */
2955 void
2956 vfs_mnttab_poll(timespec_t *old, struct pollhead **phpp)
2957 {
2958 	int changed;
2959 
2960 	*phpp = (struct pollhead *)NULL;
2961 
2962 	/*
2963 	 * Note: don't grab vfs list lock before accessing vfs_mnttab_mtime.
2964 	 * Can lead to deadlock against vfs_mnttab_modtimeupd(). It is safe
2965 	 * to not grab the vfs list lock because tv_sec is monotonically
2966 	 * increasing.
2967 	 */
2968 
2969 	changed = (old->tv_nsec != vfs_mnttab_mtime.tv_nsec) ||
2970 	    (old->tv_sec != vfs_mnttab_mtime.tv_sec);
2971 	if (!changed) {
2972 		*phpp = &vfs_pollhd;
2973 	}
2974 }
2975 
2976 /* Provide a unique and monotonically-increasing timestamp. */
2977 void
2978 vfs_mono_time(timespec_t *ts)
2979 {
2980 	static volatile hrtime_t hrt;		/* The saved time. */
2981 	hrtime_t	newhrt, oldhrt;		/* For effecting the CAS. */
2982 	timespec_t	newts;
2983 
2984 	/*
2985 	 * Try gethrestime() first, but be prepared to fabricate a sensible
2986 	 * answer at the first sign of any trouble.
2987 	 */
2988 	gethrestime(&newts);
2989 	newhrt = ts2hrt(&newts);
2990 	for (;;) {
2991 		oldhrt = hrt;
2992 		if (newhrt <= hrt)
2993 			newhrt = hrt + 1;
2994 		if (atomic_cas_64((uint64_t *)&hrt, oldhrt, newhrt) == oldhrt)
2995 			break;
2996 	}
2997 	hrt2ts(newhrt, ts);
2998 }
2999 
3000 /*
3001  * Update the mnttab modification time and wake up any waiters for
3002  * mnttab changes
3003  */
3004 void
3005 vfs_mnttab_modtimeupd()
3006 {
3007 	hrtime_t oldhrt, newhrt;
3008 
3009 	ASSERT(RW_WRITE_HELD(&vfslist));
3010 	oldhrt = ts2hrt(&vfs_mnttab_mtime);
3011 	gethrestime(&vfs_mnttab_mtime);
3012 	newhrt = ts2hrt(&vfs_mnttab_mtime);
3013 	if (oldhrt == (hrtime_t)0)
3014 		vfs_mnttab_ctime = vfs_mnttab_mtime;
3015 	/*
3016 	 * Attempt to provide unique mtime (like uniqtime but not).
3017 	 */
3018 	if (newhrt == oldhrt) {
3019 		newhrt++;
3020 		hrt2ts(newhrt, &vfs_mnttab_mtime);
3021 	}
3022 	pollwakeup(&vfs_pollhd, (short)POLLRDBAND);
3023 	vfs_mnttab_writeop();
3024 }
3025 
3026 int
3027 dounmount(struct vfs *vfsp, int flag, cred_t *cr)
3028 {
3029 	vnode_t *coveredvp;
3030 	int error;
3031 	extern void teardown_vopstats(vfs_t *);
3032 
3033 	/*
3034 	 * Get covered vnode. This will be NULL if the vfs is not linked
3035 	 * into the file system name space (i.e., domount() with MNT_NOSPICE).
3036 	 */
3037 	coveredvp = vfsp->vfs_vnodecovered;
3038 	ASSERT(coveredvp == NULL || vn_vfswlock_held(coveredvp));
3039 
3040 	/*
3041 	 * Purge all dnlc entries for this vfs.
3042 	 */
3043 	(void) dnlc_purge_vfsp(vfsp, 0);
3044 
3045 	/* For forcible umount, skip VFS_SYNC() since it may hang */
3046 	if ((flag & MS_FORCE) == 0)
3047 		(void) VFS_SYNC(vfsp, 0, cr);
3048 
3049 	/*
3050 	 * Lock the vfs to maintain fs status quo during unmount.  This
3051 	 * has to be done after the sync because ufs_update tries to acquire
3052 	 * the vfs_reflock.
3053 	 */
3054 	vfs_lock_wait(vfsp);
3055 
3056 	if (error = VFS_UNMOUNT(vfsp, flag, cr)) {
3057 		vfs_unlock(vfsp);
3058 		if (coveredvp != NULL)
3059 			vn_vfsunlock(coveredvp);
3060 	} else if (coveredvp != NULL) {
3061 		teardown_vopstats(vfsp);
3062 		/*
3063 		 * vfs_remove() will do a VN_RELE(vfsp->vfs_vnodecovered)
3064 		 * when it frees vfsp so we do a VN_HOLD() so we can
3065 		 * continue to use coveredvp afterwards.
3066 		 */
3067 		VN_HOLD(coveredvp);
3068 		vfs_remove(vfsp);
3069 		vn_vfsunlock(coveredvp);
3070 		VN_RELE(coveredvp);
3071 	} else {
3072 		teardown_vopstats(vfsp);
3073 		/*
3074 		 * Release the reference to vfs that is not linked
3075 		 * into the name space.
3076 		 */
3077 		vfs_unlock(vfsp);
3078 		VFS_RELE(vfsp);
3079 	}
3080 	return (error);
3081 }
3082 
3083 
3084 /*
3085  * Vfs_unmountall() is called by uadmin() to unmount all
3086  * mounted file systems (except the root file system) during shutdown.
3087  * It follows the existing locking protocol when traversing the vfs list
3088  * to sync and unmount vfses. Even though there should be no
3089  * other thread running while the system is shutting down, it is prudent
3090  * to still follow the locking protocol.
3091  */
3092 void
3093 vfs_unmountall(void)
3094 {
3095 	struct vfs *vfsp;
3096 	struct vfs *prev_vfsp = NULL;
3097 	int error;
3098 
3099 	/*
3100 	 * Toss all dnlc entries now so that the per-vfs sync
3101 	 * and unmount operations don't have to slog through
3102 	 * a bunch of uninteresting vnodes over and over again.
3103 	 */
3104 	dnlc_purge();
3105 
3106 	vfs_list_lock();
3107 	for (vfsp = rootvfs->vfs_prev; vfsp != rootvfs; vfsp = prev_vfsp) {
3108 		prev_vfsp = vfsp->vfs_prev;
3109 
3110 		if (vfs_lock(vfsp) != 0)
3111 			continue;
3112 		error = vn_vfswlock(vfsp->vfs_vnodecovered);
3113 		vfs_unlock(vfsp);
3114 		if (error)
3115 			continue;
3116 
3117 		vfs_list_unlock();
3118 
3119 		(void) VFS_SYNC(vfsp, SYNC_CLOSE, CRED());
3120 		(void) dounmount(vfsp, 0, CRED());
3121 
3122 		/*
3123 		 * Since we dropped the vfslist lock above we must
3124 		 * verify that next_vfsp still exists, else start over.
3125 		 */
3126 		vfs_list_lock();
3127 		for (vfsp = rootvfs->vfs_prev;
3128 		    vfsp != rootvfs; vfsp = vfsp->vfs_prev)
3129 			if (vfsp == prev_vfsp)
3130 				break;
3131 		if (vfsp == rootvfs && prev_vfsp != rootvfs)
3132 			prev_vfsp = rootvfs->vfs_prev;
3133 	}
3134 	vfs_list_unlock();
3135 }
3136 
3137 /*
3138  * Called to add an entry to the end of the vfs mount in progress list
3139  */
3140 void
3141 vfs_addmip(dev_t dev, struct vfs *vfsp)
3142 {
3143 	struct ipmnt *mipp;
3144 
3145 	mipp = (struct ipmnt *)kmem_alloc(sizeof (struct ipmnt), KM_SLEEP);
3146 	mipp->mip_next = NULL;
3147 	mipp->mip_dev = dev;
3148 	mipp->mip_vfsp = vfsp;
3149 	mutex_enter(&vfs_miplist_mutex);
3150 	if (vfs_miplist_end != NULL)
3151 		vfs_miplist_end->mip_next = mipp;
3152 	else
3153 		vfs_miplist = mipp;
3154 	vfs_miplist_end = mipp;
3155 	mutex_exit(&vfs_miplist_mutex);
3156 }
3157 
3158 /*
3159  * Called to remove an entry from the mount in progress list
3160  * Either because the mount completed or it failed.
3161  */
3162 void
3163 vfs_delmip(struct vfs *vfsp)
3164 {
3165 	struct ipmnt *mipp, *mipprev;
3166 
3167 	mutex_enter(&vfs_miplist_mutex);
3168 	mipprev = NULL;
3169 	for (mipp = vfs_miplist;
3170 	    mipp && mipp->mip_vfsp != vfsp; mipp = mipp->mip_next) {
3171 		mipprev = mipp;
3172 	}
3173 	if (mipp == NULL)
3174 		return; /* shouldn't happen */
3175 	if (mipp == vfs_miplist_end)
3176 		vfs_miplist_end = mipprev;
3177 	if (mipprev == NULL)
3178 		vfs_miplist = mipp->mip_next;
3179 	else
3180 		mipprev->mip_next = mipp->mip_next;
3181 	mutex_exit(&vfs_miplist_mutex);
3182 	kmem_free(mipp, sizeof (struct ipmnt));
3183 }
3184 
3185 /*
3186  * vfs_add is called by a specific filesystem's mount routine to add
3187  * the new vfs into the vfs list/hash and to cover the mounted-on vnode.
3188  * The vfs should already have been locked by the caller.
3189  *
3190  * coveredvp is NULL if this is the root.
3191  */
3192 void
3193 vfs_add(vnode_t *coveredvp, struct vfs *vfsp, int mflag)
3194 {
3195 	int newflag;
3196 
3197 	ASSERT(vfs_lock_held(vfsp));
3198 	VFS_HOLD(vfsp);
3199 	newflag = vfsp->vfs_flag;
3200 	if (mflag & MS_RDONLY)
3201 		newflag |= VFS_RDONLY;
3202 	else
3203 		newflag &= ~VFS_RDONLY;
3204 	if (mflag & MS_NOSUID)
3205 		newflag |= (VFS_NOSETUID|VFS_NODEVICES);
3206 	else
3207 		newflag &= ~(VFS_NOSETUID|VFS_NODEVICES);
3208 	if (mflag & MS_NOMNTTAB)
3209 		newflag |= VFS_NOMNTTAB;
3210 	else
3211 		newflag &= ~VFS_NOMNTTAB;
3212 
3213 	if (coveredvp != NULL) {
3214 		ASSERT(vn_vfswlock_held(coveredvp));
3215 		coveredvp->v_vfsmountedhere = vfsp;
3216 		VN_HOLD(coveredvp);
3217 	}
3218 	vfsp->vfs_vnodecovered = coveredvp;
3219 	vfsp->vfs_flag = newflag;
3220 
3221 	vfs_list_add(vfsp);
3222 }
3223 
3224 /*
3225  * Remove a vfs from the vfs list, null out the pointer from the
3226  * covered vnode to the vfs (v_vfsmountedhere), and null out the pointer
3227  * from the vfs to the covered vnode (vfs_vnodecovered). Release the
3228  * reference to the vfs and to the covered vnode.
3229  *
3230  * Called from dounmount after it's confirmed with the file system
3231  * that the unmount is legal.
3232  */
3233 void
3234 vfs_remove(struct vfs *vfsp)
3235 {
3236 	vnode_t *vp;
3237 
3238 	ASSERT(vfs_lock_held(vfsp));
3239 
3240 	/*
3241 	 * Can't unmount root.  Should never happen because fs will
3242 	 * be busy.
3243 	 */
3244 	if (vfsp == rootvfs)
3245 		panic("vfs_remove: unmounting root");
3246 
3247 	vfs_list_remove(vfsp);
3248 
3249 	/*
3250 	 * Unhook from the file system name space.
3251 	 */
3252 	vp = vfsp->vfs_vnodecovered;
3253 	ASSERT(vn_vfswlock_held(vp));
3254 	vp->v_vfsmountedhere = NULL;
3255 	vfsp->vfs_vnodecovered = NULL;
3256 	VN_RELE(vp);
3257 
3258 	/*
3259 	 * Release lock and wakeup anybody waiting.
3260 	 */
3261 	vfs_unlock(vfsp);
3262 	VFS_RELE(vfsp);
3263 }
3264 
3265 /*
3266  * Lock a filesystem to prevent access to it while mounting,
3267  * unmounting and syncing.  Return EBUSY immediately if lock
3268  * can't be acquired.
3269  */
3270 int
3271 vfs_lock(vfs_t *vfsp)
3272 {
3273 	vn_vfslocks_entry_t *vpvfsentry;
3274 
3275 	vpvfsentry = vn_vfslocks_getlock(vfsp);
3276 	if (rwst_tryenter(&vpvfsentry->ve_lock, RW_WRITER))
3277 		return (0);
3278 
3279 	vn_vfslocks_rele(vpvfsentry);
3280 	return (EBUSY);
3281 }
3282 
3283 int
3284 vfs_rlock(vfs_t *vfsp)
3285 {
3286 	vn_vfslocks_entry_t *vpvfsentry;
3287 
3288 	vpvfsentry = vn_vfslocks_getlock(vfsp);
3289 
3290 	if (rwst_tryenter(&vpvfsentry->ve_lock, RW_READER))
3291 		return (0);
3292 
3293 	vn_vfslocks_rele(vpvfsentry);
3294 	return (EBUSY);
3295 }
3296 
3297 void
3298 vfs_lock_wait(vfs_t *vfsp)
3299 {
3300 	vn_vfslocks_entry_t *vpvfsentry;
3301 
3302 	vpvfsentry = vn_vfslocks_getlock(vfsp);
3303 	rwst_enter(&vpvfsentry->ve_lock, RW_WRITER);
3304 }
3305 
3306 void
3307 vfs_rlock_wait(vfs_t *vfsp)
3308 {
3309 	vn_vfslocks_entry_t *vpvfsentry;
3310 
3311 	vpvfsentry = vn_vfslocks_getlock(vfsp);
3312 	rwst_enter(&vpvfsentry->ve_lock, RW_READER);
3313 }
3314 
3315 /*
3316  * Unlock a locked filesystem.
3317  */
3318 void
3319 vfs_unlock(vfs_t *vfsp)
3320 {
3321 	vn_vfslocks_entry_t *vpvfsentry;
3322 
3323 	/*
3324 	 * vfs_unlock will mimic sema_v behaviour to fix 4748018.
3325 	 * And these changes should remain for the patch changes as it is.
3326 	 */
3327 	if (panicstr)
3328 		return;
3329 
3330 	/*
3331 	 * ve_refcount needs to be dropped twice here.
3332 	 * 1. To release refernce after a call to vfs_locks_getlock()
3333 	 * 2. To release the reference from the locking routines like
3334 	 *    vfs_rlock_wait/vfs_wlock_wait/vfs_wlock etc,.
3335 	 */
3336 
3337 	vpvfsentry = vn_vfslocks_getlock(vfsp);
3338 	vn_vfslocks_rele(vpvfsentry);
3339 
3340 	rwst_exit(&vpvfsentry->ve_lock);
3341 	vn_vfslocks_rele(vpvfsentry);
3342 }
3343 
3344 /*
3345  * Utility routine that allows a filesystem to construct its
3346  * fsid in "the usual way" - by munging some underlying dev_t and
3347  * the filesystem type number into the 64-bit fsid.  Note that
3348  * this implicitly relies on dev_t persistence to make filesystem
3349  * id's persistent.
3350  *
3351  * There's nothing to prevent an individual fs from constructing its
3352  * fsid in a different way, and indeed they should.
3353  *
3354  * Since we want fsids to be 32-bit quantities (so that they can be
3355  * exported identically by either 32-bit or 64-bit APIs, as well as
3356  * the fact that fsid's are "known" to NFS), we compress the device
3357  * number given down to 32-bits, and panic if that isn't possible.
3358  */
3359 void
3360 vfs_make_fsid(fsid_t *fsi, dev_t dev, int val)
3361 {
3362 	if (!cmpldev((dev32_t *)&fsi->val[0], dev))
3363 		panic("device number too big for fsid!");
3364 	fsi->val[1] = val;
3365 }
3366 
3367 int
3368 vfs_lock_held(vfs_t *vfsp)
3369 {
3370 	int held;
3371 	vn_vfslocks_entry_t *vpvfsentry;
3372 
3373 	/*
3374 	 * vfs_lock_held will mimic sema_held behaviour
3375 	 * if panicstr is set. And these changes should remain
3376 	 * for the patch changes as it is.
3377 	 */
3378 	if (panicstr)
3379 		return (1);
3380 
3381 	vpvfsentry = vn_vfslocks_getlock(vfsp);
3382 	held = rwst_lock_held(&vpvfsentry->ve_lock, RW_WRITER);
3383 
3384 	vn_vfslocks_rele(vpvfsentry);
3385 	return (held);
3386 }
3387 
3388 struct _kthread *
3389 vfs_lock_owner(vfs_t *vfsp)
3390 {
3391 	struct _kthread *owner;
3392 	vn_vfslocks_entry_t *vpvfsentry;
3393 
3394 	/*
3395 	 * vfs_wlock_held will mimic sema_held behaviour
3396 	 * if panicstr is set. And these changes should remain
3397 	 * for the patch changes as it is.
3398 	 */
3399 	if (panicstr)
3400 		return (NULL);
3401 
3402 	vpvfsentry = vn_vfslocks_getlock(vfsp);
3403 	owner = rwst_owner(&vpvfsentry->ve_lock);
3404 
3405 	vn_vfslocks_rele(vpvfsentry);
3406 	return (owner);
3407 }
3408 
3409 /*
3410  * vfs list locking.
3411  *
3412  * Rather than manipulate the vfslist lock directly, we abstract into lock
3413  * and unlock routines to allow the locking implementation to be changed for
3414  * clustering.
3415  *
3416  * Whenever the vfs list is modified through its hash links, the overall list
3417  * lock must be obtained before locking the relevant hash bucket.  But to see
3418  * whether a given vfs is on the list, it suffices to obtain the lock for the
3419  * hash bucket without getting the overall list lock.  (See getvfs() below.)
3420  */
3421 
3422 void
3423 vfs_list_lock()
3424 {
3425 	rw_enter(&vfslist, RW_WRITER);
3426 }
3427 
3428 void
3429 vfs_list_read_lock()
3430 {
3431 	rw_enter(&vfslist, RW_READER);
3432 }
3433 
3434 void
3435 vfs_list_unlock()
3436 {
3437 	rw_exit(&vfslist);
3438 }
3439 
3440 /*
3441  * Low level worker routines for adding entries to and removing entries from
3442  * the vfs list.
3443  */
3444 
3445 static void
3446 vfs_hash_add(struct vfs *vfsp, int insert_at_head)
3447 {
3448 	int vhno;
3449 	struct vfs **hp;
3450 	dev_t dev;
3451 
3452 	ASSERT(RW_WRITE_HELD(&vfslist));
3453 
3454 	dev = expldev(vfsp->vfs_fsid.val[0]);
3455 	vhno = VFSHASH(getmajor(dev), getminor(dev));
3456 
3457 	mutex_enter(&rvfs_list[vhno].rvfs_lock);
3458 
3459 	/*
3460 	 * Link into the hash table, inserting it at the end, so that LOFS
3461 	 * with the same fsid as UFS (or other) file systems will not hide the
3462 	 * UFS.
3463 	 */
3464 	if (insert_at_head) {
3465 		vfsp->vfs_hash = rvfs_list[vhno].rvfs_head;
3466 		rvfs_list[vhno].rvfs_head = vfsp;
3467 	} else {
3468 		for (hp = &rvfs_list[vhno].rvfs_head; *hp != NULL;
3469 		    hp = &(*hp)->vfs_hash)
3470 			continue;
3471 		/*
3472 		 * hp now contains the address of the pointer to update
3473 		 * to effect the insertion.
3474 		 */
3475 		vfsp->vfs_hash = NULL;
3476 		*hp = vfsp;
3477 	}
3478 
3479 	rvfs_list[vhno].rvfs_len++;
3480 	mutex_exit(&rvfs_list[vhno].rvfs_lock);
3481 }
3482 
3483 
3484 static void
3485 vfs_hash_remove(struct vfs *vfsp)
3486 {
3487 	int vhno;
3488 	struct vfs *tvfsp;
3489 	dev_t dev;
3490 
3491 	ASSERT(RW_WRITE_HELD(&vfslist));
3492 
3493 	dev = expldev(vfsp->vfs_fsid.val[0]);
3494 	vhno = VFSHASH(getmajor(dev), getminor(dev));
3495 
3496 	mutex_enter(&rvfs_list[vhno].rvfs_lock);
3497 
3498 	/*
3499 	 * Remove from hash.
3500 	 */
3501 	if (rvfs_list[vhno].rvfs_head == vfsp) {
3502 		rvfs_list[vhno].rvfs_head = vfsp->vfs_hash;
3503 		rvfs_list[vhno].rvfs_len--;
3504 		goto foundit;
3505 	}
3506 	for (tvfsp = rvfs_list[vhno].rvfs_head; tvfsp != NULL;
3507 	    tvfsp = tvfsp->vfs_hash) {
3508 		if (tvfsp->vfs_hash == vfsp) {
3509 			tvfsp->vfs_hash = vfsp->vfs_hash;
3510 			rvfs_list[vhno].rvfs_len--;
3511 			goto foundit;
3512 		}
3513 	}
3514 	cmn_err(CE_WARN, "vfs_list_remove: vfs not found in hash");
3515 
3516 foundit:
3517 
3518 	mutex_exit(&rvfs_list[vhno].rvfs_lock);
3519 }
3520 
3521 
3522 void
3523 vfs_list_add(struct vfs *vfsp)
3524 {
3525 	zone_t *zone;
3526 
3527 	/*
3528 	 * Typically, the vfs_t will have been created on behalf of the file
3529 	 * system in vfs_init, where it will have been provided with a
3530 	 * vfs_impl_t. This, however, might be lacking if the vfs_t was created
3531 	 * by an unbundled file system. We therefore check for such an example
3532 	 * before stamping the vfs_t with its creation time for the benefit of
3533 	 * mntfs.
3534 	 */
3535 	if (vfsp->vfs_implp == NULL)
3536 		vfsimpl_setup(vfsp);
3537 	vfs_mono_time(&vfsp->vfs_hrctime);
3538 
3539 	/*
3540 	 * The zone that owns the mount is the one that performed the mount.
3541 	 * Note that this isn't necessarily the same as the zone mounted into.
3542 	 * The corresponding zone_rele_ref() will be done when the vfs_t
3543 	 * is being free'd.
3544 	 */
3545 	vfsp->vfs_zone = curproc->p_zone;
3546 	zone_init_ref(&vfsp->vfs_implp->vi_zone_ref);
3547 	zone_hold_ref(vfsp->vfs_zone, &vfsp->vfs_implp->vi_zone_ref,
3548 	    ZONE_REF_VFS);
3549 
3550 	/*
3551 	 * Find the zone mounted into, and put this mount on its vfs list.
3552 	 */
3553 	zone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
3554 	ASSERT(zone != NULL);
3555 	/*
3556 	 * Special casing for the root vfs.  This structure is allocated
3557 	 * statically and hooked onto rootvfs at link time.  During the
3558 	 * vfs_mountroot call at system startup time, the root file system's
3559 	 * VFS_MOUNTROOT routine will call vfs_add with this root vfs struct
3560 	 * as argument.  The code below must detect and handle this special
3561 	 * case.  The only apparent justification for this special casing is
3562 	 * to ensure that the root file system appears at the head of the
3563 	 * list.
3564 	 *
3565 	 * XXX:	I'm assuming that it's ok to do normal list locking when
3566 	 *	adding the entry for the root file system (this used to be
3567 	 *	done with no locks held).
3568 	 */
3569 	vfs_list_lock();
3570 	/*
3571 	 * Link into the vfs list proper.
3572 	 */
3573 	if (vfsp == &root) {
3574 		/*
3575 		 * Assert: This vfs is already on the list as its first entry.
3576 		 * Thus, there's nothing to do.
3577 		 */
3578 		ASSERT(rootvfs == vfsp);
3579 		/*
3580 		 * Add it to the head of the global zone's vfslist.
3581 		 */
3582 		ASSERT(zone == global_zone);
3583 		ASSERT(zone->zone_vfslist == NULL);
3584 		zone->zone_vfslist = vfsp;
3585 	} else {
3586 		/*
3587 		 * Link to end of list using vfs_prev (as rootvfs is now a
3588 		 * doubly linked circular list) so list is in mount order for
3589 		 * mnttab use.
3590 		 */
3591 		rootvfs->vfs_prev->vfs_next = vfsp;
3592 		vfsp->vfs_prev = rootvfs->vfs_prev;
3593 		rootvfs->vfs_prev = vfsp;
3594 		vfsp->vfs_next = rootvfs;
3595 
3596 		/*
3597 		 * Do it again for the zone-private list (which may be NULL).
3598 		 */
3599 		if (zone->zone_vfslist == NULL) {
3600 			ASSERT(zone != global_zone);
3601 			zone->zone_vfslist = vfsp;
3602 		} else {
3603 			zone->zone_vfslist->vfs_zone_prev->vfs_zone_next = vfsp;
3604 			vfsp->vfs_zone_prev = zone->zone_vfslist->vfs_zone_prev;
3605 			zone->zone_vfslist->vfs_zone_prev = vfsp;
3606 			vfsp->vfs_zone_next = zone->zone_vfslist;
3607 		}
3608 	}
3609 
3610 	/*
3611 	 * Link into the hash table, inserting it at the end, so that LOFS
3612 	 * with the same fsid as UFS (or other) file systems will not hide
3613 	 * the UFS.
3614 	 */
3615 	vfs_hash_add(vfsp, 0);
3616 
3617 	/*
3618 	 * update the mnttab modification time
3619 	 */
3620 	vfs_mnttab_modtimeupd();
3621 	vfs_list_unlock();
3622 	zone_rele(zone);
3623 }
3624 
3625 void
3626 vfs_list_remove(struct vfs *vfsp)
3627 {
3628 	zone_t *zone;
3629 
3630 	zone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
3631 	ASSERT(zone != NULL);
3632 	/*
3633 	 * Callers are responsible for preventing attempts to unmount the
3634 	 * root.
3635 	 */
3636 	ASSERT(vfsp != rootvfs);
3637 
3638 	vfs_list_lock();
3639 
3640 	/*
3641 	 * Remove from hash.
3642 	 */
3643 	vfs_hash_remove(vfsp);
3644 
3645 	/*
3646 	 * Remove from vfs list.
3647 	 */
3648 	vfsp->vfs_prev->vfs_next = vfsp->vfs_next;
3649 	vfsp->vfs_next->vfs_prev = vfsp->vfs_prev;
3650 	vfsp->vfs_next = vfsp->vfs_prev = NULL;
3651 
3652 	/*
3653 	 * Remove from zone-specific vfs list.
3654 	 */
3655 	if (zone->zone_vfslist == vfsp)
3656 		zone->zone_vfslist = vfsp->vfs_zone_next;
3657 
3658 	if (vfsp->vfs_zone_next == vfsp) {
3659 		ASSERT(vfsp->vfs_zone_prev == vfsp);
3660 		ASSERT(zone->zone_vfslist == vfsp);
3661 		zone->zone_vfslist = NULL;
3662 	}
3663 
3664 	vfsp->vfs_zone_prev->vfs_zone_next = vfsp->vfs_zone_next;
3665 	vfsp->vfs_zone_next->vfs_zone_prev = vfsp->vfs_zone_prev;
3666 	vfsp->vfs_zone_next = vfsp->vfs_zone_prev = NULL;
3667 
3668 	/*
3669 	 * update the mnttab modification time
3670 	 */
3671 	vfs_mnttab_modtimeupd();
3672 	vfs_list_unlock();
3673 	zone_rele(zone);
3674 }
3675 
3676 struct vfs *
3677 getvfs(fsid_t *fsid)
3678 {
3679 	struct vfs *vfsp;
3680 	int val0 = fsid->val[0];
3681 	int val1 = fsid->val[1];
3682 	dev_t dev = expldev(val0);
3683 	int vhno = VFSHASH(getmajor(dev), getminor(dev));
3684 	kmutex_t *hmp = &rvfs_list[vhno].rvfs_lock;
3685 
3686 	mutex_enter(hmp);
3687 	for (vfsp = rvfs_list[vhno].rvfs_head; vfsp; vfsp = vfsp->vfs_hash) {
3688 		if (vfsp->vfs_fsid.val[0] == val0 &&
3689 		    vfsp->vfs_fsid.val[1] == val1) {
3690 			VFS_HOLD(vfsp);
3691 			mutex_exit(hmp);
3692 			return (vfsp);
3693 		}
3694 	}
3695 	mutex_exit(hmp);
3696 	return (NULL);
3697 }
3698 
3699 /*
3700  * Search the vfs mount in progress list for a specified device/vfs entry.
3701  * Returns 0 if the first entry in the list that the device matches has the
3702  * given vfs pointer as well.  If the device matches but a different vfs
3703  * pointer is encountered in the list before the given vfs pointer then
3704  * a 1 is returned.
3705  */
3706 
3707 int
3708 vfs_devmounting(dev_t dev, struct vfs *vfsp)
3709 {
3710 	int retval = 0;
3711 	struct ipmnt *mipp;
3712 
3713 	mutex_enter(&vfs_miplist_mutex);
3714 	for (mipp = vfs_miplist; mipp != NULL; mipp = mipp->mip_next) {
3715 		if (mipp->mip_dev == dev) {
3716 			if (mipp->mip_vfsp != vfsp)
3717 				retval = 1;
3718 			break;
3719 		}
3720 	}
3721 	mutex_exit(&vfs_miplist_mutex);
3722 	return (retval);
3723 }
3724 
3725 /*
3726  * Search the vfs list for a specified device.  Returns 1, if entry is found
3727  * or 0 if no suitable entry is found.
3728  */
3729 
3730 int
3731 vfs_devismounted(dev_t dev)
3732 {
3733 	struct vfs *vfsp;
3734 	int found;
3735 
3736 	vfs_list_read_lock();
3737 	vfsp = rootvfs;
3738 	found = 0;
3739 	do {
3740 		if (vfsp->vfs_dev == dev) {
3741 			found = 1;
3742 			break;
3743 		}
3744 		vfsp = vfsp->vfs_next;
3745 	} while (vfsp != rootvfs);
3746 
3747 	vfs_list_unlock();
3748 	return (found);
3749 }
3750 
3751 /*
3752  * Search the vfs list for a specified device.  Returns a pointer to it
3753  * or NULL if no suitable entry is found. The caller of this routine
3754  * is responsible for releasing the returned vfs pointer.
3755  */
3756 struct vfs *
3757 vfs_dev2vfsp(dev_t dev)
3758 {
3759 	struct vfs *vfsp;
3760 	int found;
3761 
3762 	vfs_list_read_lock();
3763 	vfsp = rootvfs;
3764 	found = 0;
3765 	do {
3766 		/*
3767 		 * The following could be made more efficient by making
3768 		 * the entire loop use vfs_zone_next if the call is from
3769 		 * a zone.  The only callers, however, ustat(2) and
3770 		 * umount2(2), don't seem to justify the added
3771 		 * complexity at present.
3772 		 */
3773 		if (vfsp->vfs_dev == dev &&
3774 		    ZONE_PATH_VISIBLE(refstr_value(vfsp->vfs_mntpt),
3775 		    curproc->p_zone)) {
3776 			VFS_HOLD(vfsp);
3777 			found = 1;
3778 			break;
3779 		}
3780 		vfsp = vfsp->vfs_next;
3781 	} while (vfsp != rootvfs);
3782 	vfs_list_unlock();
3783 	return (found ? vfsp: NULL);
3784 }
3785 
3786 /*
3787  * Search the vfs list for a specified mntpoint.  Returns a pointer to it
3788  * or NULL if no suitable entry is found. The caller of this routine
3789  * is responsible for releasing the returned vfs pointer.
3790  *
3791  * Note that if multiple mntpoints match, the last one matching is
3792  * returned in an attempt to return the "top" mount when overlay
3793  * mounts are covering the same mount point.  This is accomplished by starting
3794  * at the end of the list and working our way backwards, stopping at the first
3795  * matching mount.
3796  */
3797 struct vfs *
3798 vfs_mntpoint2vfsp(const char *mp)
3799 {
3800 	struct vfs *vfsp;
3801 	struct vfs *retvfsp = NULL;
3802 	zone_t *zone = curproc->p_zone;
3803 	struct vfs *list;
3804 
3805 	vfs_list_read_lock();
3806 	if (getzoneid() == GLOBAL_ZONEID) {
3807 		/*
3808 		 * The global zone may see filesystems in any zone.
3809 		 */
3810 		vfsp = rootvfs->vfs_prev;
3811 		do {
3812 			if (strcmp(refstr_value(vfsp->vfs_mntpt), mp) == 0) {
3813 				retvfsp = vfsp;
3814 				break;
3815 			}
3816 			vfsp = vfsp->vfs_prev;
3817 		} while (vfsp != rootvfs->vfs_prev);
3818 	} else if ((list = zone->zone_vfslist) != NULL) {
3819 		const char *mntpt;
3820 
3821 		vfsp = list->vfs_zone_prev;
3822 		do {
3823 			mntpt = refstr_value(vfsp->vfs_mntpt);
3824 			mntpt = ZONE_PATH_TRANSLATE(mntpt, zone);
3825 			if (strcmp(mntpt, mp) == 0) {
3826 				retvfsp = vfsp;
3827 				break;
3828 			}
3829 			vfsp = vfsp->vfs_zone_prev;
3830 		} while (vfsp != list->vfs_zone_prev);
3831 	}
3832 	if (retvfsp)
3833 		VFS_HOLD(retvfsp);
3834 	vfs_list_unlock();
3835 	return (retvfsp);
3836 }
3837 
3838 /*
3839  * Search the vfs list for a specified vfsops.
3840  * if vfs entry is found then return 1, else 0.
3841  */
3842 int
3843 vfs_opsinuse(vfsops_t *ops)
3844 {
3845 	struct vfs *vfsp;
3846 	int found;
3847 
3848 	vfs_list_read_lock();
3849 	vfsp = rootvfs;
3850 	found = 0;
3851 	do {
3852 		if (vfs_getops(vfsp) == ops) {
3853 			found = 1;
3854 			break;
3855 		}
3856 		vfsp = vfsp->vfs_next;
3857 	} while (vfsp != rootvfs);
3858 	vfs_list_unlock();
3859 	return (found);
3860 }
3861 
3862 /*
3863  * Allocate an entry in vfssw for a file system type
3864  */
3865 struct vfssw *
3866 allocate_vfssw(const char *type)
3867 {
3868 	struct vfssw *vswp;
3869 
3870 	if (type[0] == '\0' || strlen(type) + 1 > _ST_FSTYPSZ) {
3871 		/*
3872 		 * The vfssw table uses the empty string to identify an
3873 		 * available entry; we cannot add any type which has
3874 		 * a leading NUL. The string length is limited to
3875 		 * the size of the st_fstype array in struct stat.
3876 		 */
3877 		return (NULL);
3878 	}
3879 
3880 	ASSERT(VFSSW_WRITE_LOCKED());
3881 	for (vswp = &vfssw[1]; vswp < &vfssw[nfstype]; vswp++)
3882 		if (!ALLOCATED_VFSSW(vswp)) {
3883 			vswp->vsw_name = kmem_alloc(strlen(type) + 1, KM_SLEEP);
3884 			(void) strcpy(vswp->vsw_name, type);
3885 			ASSERT(vswp->vsw_count == 0);
3886 			vswp->vsw_count = 1;
3887 			mutex_init(&vswp->vsw_lock, NULL, MUTEX_DEFAULT, NULL);
3888 			return (vswp);
3889 		}
3890 	return (NULL);
3891 }
3892 
3893 /*
3894  * Impose additional layer of translation between vfstype names
3895  * and module names in the filesystem.
3896  */
3897 static const char *
3898 vfs_to_modname(const char *vfstype)
3899 {
3900 	if (strcmp(vfstype, "proc") == 0) {
3901 		vfstype = "procfs";
3902 	} else if (strcmp(vfstype, "fd") == 0) {
3903 		vfstype = "fdfs";
3904 	} else if (strncmp(vfstype, "nfs", 3) == 0) {
3905 		vfstype = "nfs";
3906 	}
3907 
3908 	return (vfstype);
3909 }
3910 
3911 /*
3912  * Find a vfssw entry given a file system type name.
3913  * Try to autoload the filesystem if it's not found.
3914  * If it's installed, return the vfssw locked to prevent unloading.
3915  */
3916 struct vfssw *
3917 vfs_getvfssw(const char *type)
3918 {
3919 	struct vfssw *vswp;
3920 	const char *modname;
3921 
3922 	RLOCK_VFSSW();
3923 	vswp = vfs_getvfsswbyname(type);
3924 	modname = vfs_to_modname(type);
3925 
3926 	if (rootdir == NULL) {
3927 		/*
3928 		 * If we haven't yet loaded the root file system, then our
3929 		 * _init won't be called until later. Allocate vfssw entry,
3930 		 * because mod_installfs won't be called.
3931 		 */
3932 		if (vswp == NULL) {
3933 			RUNLOCK_VFSSW();
3934 			WLOCK_VFSSW();
3935 			if ((vswp = vfs_getvfsswbyname(type)) == NULL) {
3936 				if ((vswp = allocate_vfssw(type)) == NULL) {
3937 					WUNLOCK_VFSSW();
3938 					return (NULL);
3939 				}
3940 			}
3941 			WUNLOCK_VFSSW();
3942 			RLOCK_VFSSW();
3943 		}
3944 		if (!VFS_INSTALLED(vswp)) {
3945 			RUNLOCK_VFSSW();
3946 			(void) modloadonly("fs", modname);
3947 		} else
3948 			RUNLOCK_VFSSW();
3949 		return (vswp);
3950 	}
3951 
3952 	/*
3953 	 * Try to load the filesystem.  Before calling modload(), we drop
3954 	 * our lock on the VFS switch table, and pick it up after the
3955 	 * module is loaded.  However, there is a potential race:  the
3956 	 * module could be unloaded after the call to modload() completes
3957 	 * but before we pick up the lock and drive on.  Therefore,
3958 	 * we keep reloading the module until we've loaded the module
3959 	 * _and_ we have the lock on the VFS switch table.
3960 	 */
3961 	while (vswp == NULL || !VFS_INSTALLED(vswp)) {
3962 		RUNLOCK_VFSSW();
3963 		if (modload("fs", modname) == -1)
3964 			return (NULL);
3965 		RLOCK_VFSSW();
3966 		if (vswp == NULL)
3967 			if ((vswp = vfs_getvfsswbyname(type)) == NULL)
3968 				break;
3969 	}
3970 	RUNLOCK_VFSSW();
3971 
3972 	return (vswp);
3973 }
3974 
3975 /*
3976  * Find a vfssw entry given a file system type name.
3977  */
3978 struct vfssw *
3979 vfs_getvfsswbyname(const char *type)
3980 {
3981 	struct vfssw *vswp;
3982 
3983 	ASSERT(VFSSW_LOCKED());
3984 	if (type == NULL || *type == '\0')
3985 		return (NULL);
3986 
3987 	for (vswp = &vfssw[1]; vswp < &vfssw[nfstype]; vswp++) {
3988 		if (strcmp(type, vswp->vsw_name) == 0) {
3989 			vfs_refvfssw(vswp);
3990 			return (vswp);
3991 		}
3992 	}
3993 
3994 	return (NULL);
3995 }
3996 
3997 /*
3998  * Find a vfssw entry given a set of vfsops.
3999  */
4000 struct vfssw *
4001 vfs_getvfsswbyvfsops(vfsops_t *vfsops)
4002 {
4003 	struct vfssw *vswp;
4004 
4005 	RLOCK_VFSSW();
4006 	for (vswp = &vfssw[1]; vswp < &vfssw[nfstype]; vswp++) {
4007 		if (ALLOCATED_VFSSW(vswp) && &vswp->vsw_vfsops == vfsops) {
4008 			vfs_refvfssw(vswp);
4009 			RUNLOCK_VFSSW();
4010 			return (vswp);
4011 		}
4012 	}
4013 	RUNLOCK_VFSSW();
4014 
4015 	return (NULL);
4016 }
4017 
4018 /*
4019  * Reference a vfssw entry.
4020  */
4021 void
4022 vfs_refvfssw(struct vfssw *vswp)
4023 {
4024 
4025 	mutex_enter(&vswp->vsw_lock);
4026 	vswp->vsw_count++;
4027 	mutex_exit(&vswp->vsw_lock);
4028 }
4029 
4030 /*
4031  * Unreference a vfssw entry.
4032  */
4033 void
4034 vfs_unrefvfssw(struct vfssw *vswp)
4035 {
4036 
4037 	mutex_enter(&vswp->vsw_lock);
4038 	vswp->vsw_count--;
4039 	mutex_exit(&vswp->vsw_lock);
4040 }
4041 
4042 static int sync_retries = 20;	/* number of retries when not making progress */
4043 static int sync_triesleft;	/* portion of sync_retries remaining */
4044 
4045 static pgcnt_t old_pgcnt, new_pgcnt;
4046 static int new_bufcnt, old_bufcnt;
4047 
4048 /*
4049  * Sync all of the mounted filesystems, and then wait for the actual i/o to
4050  * complete.  We wait by counting the number of dirty pages and buffers,
4051  * pushing them out using bio_busy() and page_busy(), and then counting again.
4052  * This routine is used during the uadmin A_SHUTDOWN code.  It should only
4053  * be used after some higher-level mechanism has quiesced the system so that
4054  * new writes are not being initiated while we are waiting for completion.
4055  *
4056  * To ensure finite running time, our algorithm uses sync_triesleft (a progress
4057  * counter used by the vfs_syncall() loop below). It is declared above so
4058  * it can be found easily in the debugger.
4059  *
4060  * The sync_triesleft counter is updated by vfs_syncall() itself.  If we make
4061  * sync_retries consecutive calls to bio_busy() and page_busy() without
4062  * decreasing either the number of dirty buffers or dirty pages below the
4063  * lowest count we have seen so far, we give up and return from vfs_syncall().
4064  *
4065  * Each loop iteration ends with a call to delay() one second to allow time for
4066  * i/o completion and to permit the user time to read our progress messages.
4067  */
4068 void
4069 vfs_syncall(void)
4070 {
4071 	if (rootdir == NULL && !modrootloaded)
4072 		return; /* no filesystems have been loaded yet */
4073 
4074 	printf("syncing file systems...");
4075 	sync();
4076 
4077 	sync_triesleft = sync_retries;
4078 
4079 	old_bufcnt = new_bufcnt = INT_MAX;
4080 	old_pgcnt = new_pgcnt = ULONG_MAX;
4081 
4082 	while (sync_triesleft > 0) {
4083 		old_bufcnt = MIN(old_bufcnt, new_bufcnt);
4084 		old_pgcnt = MIN(old_pgcnt, new_pgcnt);
4085 
4086 		new_bufcnt = bio_busy(B_TRUE);
4087 		new_pgcnt = page_busy(B_TRUE);
4088 
4089 		if (new_bufcnt == 0 && new_pgcnt == 0)
4090 			break;
4091 
4092 		if (new_bufcnt < old_bufcnt || new_pgcnt < old_pgcnt)
4093 			sync_triesleft = sync_retries;
4094 		else
4095 			sync_triesleft--;
4096 
4097 		if (new_bufcnt)
4098 			printf(" [%d]", new_bufcnt);
4099 		if (new_pgcnt)
4100 			printf(" %lu", new_pgcnt);
4101 
4102 		delay(hz);
4103 	}
4104 
4105 	if (new_bufcnt != 0 || new_pgcnt != 0)
4106 		printf(" done (not all i/o completed)\n");
4107 	else
4108 		printf(" done\n");
4109 
4110 	delay(hz);
4111 }
4112 
4113 /*
4114  * Map VFS flags to statvfs flags.  These shouldn't really be separate
4115  * flags at all.
4116  */
4117 uint_t
4118 vf_to_stf(uint_t vf)
4119 {
4120 	uint_t stf = 0;
4121 
4122 	if (vf & VFS_RDONLY)
4123 		stf |= ST_RDONLY;
4124 	if (vf & VFS_NOSETUID)
4125 		stf |= ST_NOSUID;
4126 	if (vf & VFS_NOTRUNC)
4127 		stf |= ST_NOTRUNC;
4128 
4129 	return (stf);
4130 }
4131 
4132 /*
4133  * Entries for (illegal) fstype 0.
4134  */
4135 /* ARGSUSED */
4136 int
4137 vfsstray_sync(struct vfs *vfsp, short arg, struct cred *cr)
4138 {
4139 	cmn_err(CE_PANIC, "stray vfs operation");
4140 	return (0);
4141 }
4142 
4143 /*
4144  * Entries for (illegal) fstype 0.
4145  */
4146 int
4147 vfsstray(void)
4148 {
4149 	cmn_err(CE_PANIC, "stray vfs operation");
4150 	return (0);
4151 }
4152 
4153 /*
4154  * Support for dealing with forced UFS unmount and its interaction with
4155  * LOFS. Could be used by any filesystem.
4156  * See bug 1203132.
4157  */
4158 int
4159 vfs_EIO(void)
4160 {
4161 	return (EIO);
4162 }
4163 
4164 /*
4165  * We've gotta define the op for sync separately, since the compiler gets
4166  * confused if we mix and match ANSI and normal style prototypes when
4167  * a "short" argument is present and spits out a warning.
4168  */
4169 /*ARGSUSED*/
4170 int
4171 vfs_EIO_sync(struct vfs *vfsp, short arg, struct cred *cr)
4172 {
4173 	return (EIO);
4174 }
4175 
4176 vfs_t EIO_vfs;
4177 vfsops_t *EIO_vfsops;
4178 
4179 /*
4180  * Called from startup() to initialize all loaded vfs's
4181  */
4182 void
4183 vfsinit(void)
4184 {
4185 	struct vfssw *vswp;
4186 	int error;
4187 	extern int vopstats_enabled;
4188 	extern void vopstats_startup();
4189 
4190 	static const fs_operation_def_t EIO_vfsops_template[] = {
4191 		VFSNAME_MOUNT,		{ .error = vfs_EIO },
4192 		VFSNAME_UNMOUNT,	{ .error = vfs_EIO },
4193 		VFSNAME_ROOT,		{ .error = vfs_EIO },
4194 		VFSNAME_STATVFS,	{ .error = vfs_EIO },
4195 		VFSNAME_SYNC,		{ .vfs_sync = vfs_EIO_sync },
4196 		VFSNAME_VGET,		{ .error = vfs_EIO },
4197 		VFSNAME_MOUNTROOT,	{ .error = vfs_EIO },
4198 		VFSNAME_FREEVFS,	{ .error = vfs_EIO },
4199 		VFSNAME_VNSTATE,	{ .error = vfs_EIO },
4200 		VFSNAME_SYNCFS,		{ .error = vfs_EIO },
4201 		NULL, NULL
4202 	};
4203 
4204 	static const fs_operation_def_t stray_vfsops_template[] = {
4205 		VFSNAME_MOUNT,		{ .error = vfsstray },
4206 		VFSNAME_UNMOUNT,	{ .error = vfsstray },
4207 		VFSNAME_ROOT,		{ .error = vfsstray },
4208 		VFSNAME_STATVFS,	{ .error = vfsstray },
4209 		VFSNAME_SYNC,		{ .vfs_sync = vfsstray_sync },
4210 		VFSNAME_VGET,		{ .error = vfsstray },
4211 		VFSNAME_MOUNTROOT,	{ .error = vfsstray },
4212 		VFSNAME_FREEVFS,	{ .error = vfsstray },
4213 		VFSNAME_VNSTATE,	{ .error = vfsstray },
4214 		VFSNAME_SYNCFS,		{ .error = vfsstray },
4215 		NULL, NULL
4216 	};
4217 
4218 	/* Create vfs cache */
4219 	vfs_cache = kmem_cache_create("vfs_cache", sizeof (struct vfs),
4220 	    sizeof (uintptr_t), NULL, NULL, NULL, NULL, NULL, 0);
4221 
4222 	/* Initialize the vnode cache (file systems may use it during init). */
4223 	vn_create_cache();
4224 
4225 	/* Setup event monitor framework */
4226 	fem_init();
4227 
4228 	/* Initialize the dummy stray file system type. */
4229 	error = vfs_setfsops(0, stray_vfsops_template, NULL);
4230 
4231 	/* Initialize the dummy EIO file system. */
4232 	error = vfs_makefsops(EIO_vfsops_template, &EIO_vfsops);
4233 	if (error != 0) {
4234 		cmn_err(CE_WARN, "vfsinit: bad EIO vfs ops template");
4235 		/* Shouldn't happen, but not bad enough to panic */
4236 	}
4237 
4238 	VFS_INIT(&EIO_vfs, EIO_vfsops, (caddr_t)NULL);
4239 
4240 	/*
4241 	 * Default EIO_vfs.vfs_flag to VFS_UNMOUNTED so a lookup
4242 	 * on this vfs can immediately notice it's invalid.
4243 	 */
4244 	EIO_vfs.vfs_flag |= VFS_UNMOUNTED;
4245 
4246 	/*
4247 	 * Call the init routines of non-loadable filesystems only.
4248 	 * Filesystems which are loaded as separate modules will be
4249 	 * initialized by the module loading code instead.
4250 	 */
4251 
4252 	for (vswp = &vfssw[1]; vswp < &vfssw[nfstype]; vswp++) {
4253 		RLOCK_VFSSW();
4254 		if (vswp->vsw_init != NULL)
4255 			(void) (*vswp->vsw_init)(vswp - vfssw, vswp->vsw_name);
4256 		RUNLOCK_VFSSW();
4257 	}
4258 
4259 	vopstats_startup();
4260 
4261 	if (vopstats_enabled) {
4262 		/* EIO_vfs can collect stats, but we don't retrieve them */
4263 		initialize_vopstats(&EIO_vfs.vfs_vopstats);
4264 		EIO_vfs.vfs_fstypevsp = NULL;
4265 		EIO_vfs.vfs_vskap = NULL;
4266 		EIO_vfs.vfs_flag |= VFS_STATS;
4267 	}
4268 
4269 	xattr_init();
4270 
4271 	reparse_point_init();
4272 }
4273 
4274 vfs_t *
4275 vfs_alloc(int kmflag)
4276 {
4277 	vfs_t *vfsp;
4278 
4279 	vfsp = kmem_cache_alloc(vfs_cache, kmflag);
4280 
4281 	/*
4282 	 * Do the simplest initialization here.
4283 	 * Everything else gets done in vfs_init()
4284 	 */
4285 	bzero(vfsp, sizeof (vfs_t));
4286 	return (vfsp);
4287 }
4288 
4289 void
4290 vfs_free(vfs_t *vfsp)
4291 {
4292 	/*
4293 	 * One would be tempted to assert that "vfsp->vfs_count == 0".
4294 	 * The problem is that this gets called out of domount() with
4295 	 * a partially initialized vfs and a vfs_count of 1.  This is
4296 	 * also called from vfs_rele() with a vfs_count of 0.  We can't
4297 	 * call VFS_RELE() from domount() if VFS_MOUNT() hasn't successfully
4298 	 * returned.  This is because VFS_MOUNT() fully initializes the
4299 	 * vfs structure and its associated data.  VFS_RELE() will call
4300 	 * VFS_FREEVFS() which may panic the system if the data structures
4301 	 * aren't fully initialized from a successful VFS_MOUNT()).
4302 	 */
4303 
4304 	/* If FEM was in use, make sure everything gets cleaned up */
4305 	if (vfsp->vfs_femhead) {
4306 		ASSERT(vfsp->vfs_femhead->femh_list == NULL);
4307 		mutex_destroy(&vfsp->vfs_femhead->femh_lock);
4308 		kmem_free(vfsp->vfs_femhead, sizeof (*(vfsp->vfs_femhead)));
4309 		vfsp->vfs_femhead = NULL;
4310 	}
4311 
4312 	if (vfsp->vfs_implp)
4313 		vfsimpl_teardown(vfsp);
4314 	sema_destroy(&vfsp->vfs_reflock);
4315 	kmem_cache_free(vfs_cache, vfsp);
4316 }
4317 
4318 /*
4319  * Increments the vfs reference count by one atomically.
4320  */
4321 void
4322 vfs_hold(vfs_t *vfsp)
4323 {
4324 	atomic_inc_32(&vfsp->vfs_count);
4325 	ASSERT(vfsp->vfs_count != 0);
4326 }
4327 
4328 /*
4329  * Decrements the vfs reference count by one atomically. When
4330  * vfs reference count becomes zero, it calls the file system
4331  * specific vfs_freevfs() to free up the resources.
4332  */
4333 void
4334 vfs_rele(vfs_t *vfsp)
4335 {
4336 	ASSERT(vfsp->vfs_count != 0);
4337 	if (atomic_dec_32_nv(&vfsp->vfs_count) == 0) {
4338 		VFS_FREEVFS(vfsp);
4339 		lofi_remove(vfsp);
4340 		if (vfsp->vfs_zone)
4341 			zone_rele_ref(&vfsp->vfs_implp->vi_zone_ref,
4342 			    ZONE_REF_VFS);
4343 		vfs_freemnttab(vfsp);
4344 		vfs_free(vfsp);
4345 	}
4346 }
4347 
4348 /*
4349  * Generic operations vector support.
4350  *
4351  * This is used to build operations vectors for both the vfs and vnode.
4352  * It's normally called only when a file system is loaded.
4353  *
4354  * There are many possible algorithms for this, including the following:
4355  *
4356  *   (1) scan the list of known operations; for each, see if the file system
4357  *       includes an entry for it, and fill it in as appropriate.
4358  *
4359  *   (2) set up defaults for all known operations.  scan the list of ops
4360  *       supplied by the file system; for each which is both supplied and
4361  *       known, fill it in.
4362  *
4363  *   (3) sort the lists of known ops & supplied ops; scan the list, filling
4364  *       in entries as we go.
4365  *
4366  * we choose (1) for simplicity, and because performance isn't critical here.
4367  * note that (2) could be sped up using a precomputed hash table on known ops.
4368  * (3) could be faster than either, but only if the lists were very large or
4369  * supplied in sorted order.
4370  *
4371  */
4372 
4373 int
4374 fs_build_vector(void *vector, int *unused_ops,
4375     const fs_operation_trans_def_t *translation,
4376     const fs_operation_def_t *operations)
4377 {
4378 	int i, num_trans, num_ops, used;
4379 
4380 	/*
4381 	 * Count the number of translations and the number of supplied
4382 	 * operations.
4383 	 */
4384 
4385 	{
4386 		const fs_operation_trans_def_t *p;
4387 
4388 		for (num_trans = 0, p = translation;
4389 		    p->name != NULL;
4390 		    num_trans++, p++)
4391 			;
4392 	}
4393 
4394 	{
4395 		const fs_operation_def_t *p;
4396 
4397 		for (num_ops = 0, p = operations;
4398 		    p->name != NULL;
4399 		    num_ops++, p++)
4400 			;
4401 	}
4402 
4403 	/* Walk through each operation known to our caller.  There will be */
4404 	/* one entry in the supplied "translation table" for each. */
4405 
4406 	used = 0;
4407 
4408 	for (i = 0; i < num_trans; i++) {
4409 		int j, found;
4410 		char *curname;
4411 		fs_generic_func_p result;
4412 		fs_generic_func_p *location;
4413 
4414 		curname = translation[i].name;
4415 
4416 		/* Look for a matching operation in the list supplied by the */
4417 		/* file system. */
4418 
4419 		found = 0;
4420 
4421 		for (j = 0; j < num_ops; j++) {
4422 			if (strcmp(operations[j].name, curname) == 0) {
4423 				used++;
4424 				found = 1;
4425 				break;
4426 			}
4427 		}
4428 
4429 		/*
4430 		 * If the file system is using a "placeholder" for default
4431 		 * or error functions, grab the appropriate function out of
4432 		 * the translation table.  If the file system didn't supply
4433 		 * this operation at all, use the default function.
4434 		 */
4435 
4436 		if (found) {
4437 			result = operations[j].func.fs_generic;
4438 			if (result == fs_default) {
4439 				result = translation[i].defaultFunc;
4440 			} else if (result == fs_error) {
4441 				result = translation[i].errorFunc;
4442 			} else if (result == NULL) {
4443 				/* Null values are PROHIBITED */
4444 				return (EINVAL);
4445 			}
4446 		} else {
4447 			result = translation[i].defaultFunc;
4448 		}
4449 
4450 		/* Now store the function into the operations vector. */
4451 
4452 		location = (fs_generic_func_p *)
4453 		    (((char *)vector) + translation[i].offset);
4454 
4455 		*location = result;
4456 	}
4457 
4458 	*unused_ops = num_ops - used;
4459 
4460 	return (0);
4461 }
4462 
4463 /* Placeholder functions, should never be called. */
4464 
4465 int
4466 fs_error(void)
4467 {
4468 	cmn_err(CE_PANIC, "fs_error called");
4469 	return (0);
4470 }
4471 
4472 int
4473 fs_default(void)
4474 {
4475 	cmn_err(CE_PANIC, "fs_default called");
4476 	return (0);
4477 }
4478 
4479 #ifdef __sparc
4480 
4481 /*
4482  * Part of the implementation of booting off a mirrored root
4483  * involves a change of dev_t for the root device.  To
4484  * accomplish this, first remove the existing hash table
4485  * entry for the root device, convert to the new dev_t,
4486  * then re-insert in the hash table at the head of the list.
4487  */
4488 void
4489 vfs_root_redev(vfs_t *vfsp, dev_t ndev, int fstype)
4490 {
4491 	vfs_list_lock();
4492 
4493 	vfs_hash_remove(vfsp);
4494 
4495 	vfsp->vfs_dev = ndev;
4496 	vfs_make_fsid(&vfsp->vfs_fsid, ndev, fstype);
4497 
4498 	vfs_hash_add(vfsp, 1);
4499 
4500 	vfs_list_unlock();
4501 }
4502 
4503 #else /* x86 NEWBOOT */
4504 
4505 #if defined(__x86)
4506 extern int hvmboot_rootconf();
4507 #endif /* __x86 */
4508 
4509 extern ib_boot_prop_t *iscsiboot_prop;
4510 
4511 int
4512 rootconf()
4513 {
4514 	int error;
4515 	struct vfssw *vsw;
4516 	extern void pm_init();
4517 	char *fstyp, *fsmod;
4518 	int ret = -1;
4519 
4520 	getrootfs(&fstyp, &fsmod);
4521 
4522 #if defined(__x86)
4523 	/*
4524 	 * hvmboot_rootconf() is defined in the hvm_bootstrap misc module,
4525 	 * which lives in /platform/i86hvm, and hence is only available when
4526 	 * booted in an x86 hvm environment.  If the hvm_bootstrap misc module
4527 	 * is not available then the modstub for this function will return 0.
4528 	 * If the hvm_bootstrap misc module is available it will be loaded
4529 	 * and hvmboot_rootconf() will be invoked.
4530 	 */
4531 	if (error = hvmboot_rootconf())
4532 		return (error);
4533 #endif /* __x86 */
4534 
4535 	if (error = clboot_rootconf())
4536 		return (error);
4537 
4538 	if (modload("fs", fsmod) == -1)
4539 		panic("Cannot _init %s module", fsmod);
4540 
4541 	RLOCK_VFSSW();
4542 	vsw = vfs_getvfsswbyname(fstyp);
4543 	RUNLOCK_VFSSW();
4544 	if (vsw == NULL) {
4545 		cmn_err(CE_CONT, "Cannot find %s filesystem\n", fstyp);
4546 		return (ENXIO);
4547 	}
4548 	VFS_INIT(rootvfs, &vsw->vsw_vfsops, 0);
4549 	VFS_HOLD(rootvfs);
4550 
4551 	/* always mount readonly first */
4552 	rootvfs->vfs_flag |= VFS_RDONLY;
4553 
4554 	pm_init();
4555 
4556 	if (netboot && iscsiboot_prop) {
4557 		cmn_err(CE_WARN, "NFS boot and iSCSI boot"
4558 		    " shouldn't happen in the same time");
4559 		return (EINVAL);
4560 	}
4561 
4562 	if (netboot || iscsiboot_prop) {
4563 		ret = strplumb();
4564 		if (ret != 0) {
4565 			cmn_err(CE_WARN, "Cannot plumb network device %d", ret);
4566 			return (EFAULT);
4567 		}
4568 	}
4569 
4570 	if ((ret == 0) && iscsiboot_prop) {
4571 		ret = modload("drv", "iscsi");
4572 		/* -1 indicates fail */
4573 		if (ret == -1) {
4574 			cmn_err(CE_WARN, "Failed to load iscsi module");
4575 			iscsi_boot_prop_free();
4576 			return (EINVAL);
4577 		} else {
4578 			if (!i_ddi_attach_pseudo_node("iscsi")) {
4579 				cmn_err(CE_WARN,
4580 				    "Failed to attach iscsi driver");
4581 				iscsi_boot_prop_free();
4582 				return (ENODEV);
4583 			}
4584 		}
4585 	}
4586 
4587 	error = VFS_MOUNTROOT(rootvfs, ROOT_INIT);
4588 	vfs_unrefvfssw(vsw);
4589 	rootdev = rootvfs->vfs_dev;
4590 
4591 	if (error)
4592 		cmn_err(CE_CONT, "Cannot mount root on %s fstype %s\n",
4593 		    rootfs.bo_name, fstyp);
4594 	else
4595 		cmn_err(CE_CONT, "?root on %s fstype %s\n",
4596 		    rootfs.bo_name, fstyp);
4597 	return (error);
4598 }
4599 
4600 /*
4601  * XXX this is called by nfs only and should probably be removed
4602  * If booted with ASKNAME, prompt on the console for a filesystem
4603  * name and return it.
4604  */
4605 void
4606 getfsname(char *askfor, char *name, size_t namelen)
4607 {
4608 	if (boothowto & RB_ASKNAME) {
4609 		printf("%s name: ", askfor);
4610 		console_gets(name, namelen);
4611 	}
4612 }
4613 
4614 /*
4615  * Init the root filesystem type (rootfs.bo_fstype) from the "fstype"
4616  * property.
4617  *
4618  * Filesystem types starting with the prefix "nfs" are diskless clients;
4619  * init the root filename name (rootfs.bo_name), too.
4620  *
4621  * If we are booting via NFS we currently have these options:
4622  *	nfs -	dynamically choose NFS V2, V3, or V4 (default)
4623  *	nfs2 -	force NFS V2
4624  *	nfs3 -	force NFS V3
4625  *	nfs4 -	force NFS V4
4626  * Because we need to maintain backward compatibility with the naming
4627  * convention that the NFS V2 filesystem name is "nfs" (see vfs_conf.c)
4628  * we need to map "nfs" => "nfsdyn" and "nfs2" => "nfs".  The dynamic
4629  * nfs module will map the type back to either "nfs", "nfs3", or "nfs4".
4630  * This is only for root filesystems, all other uses will expect
4631  * that "nfs" == NFS V2.
4632  */
4633 static void
4634 getrootfs(char **fstypp, char **fsmodp)
4635 {
4636 	char *propstr = NULL;
4637 
4638 	/*
4639 	 * Check fstype property; for diskless it should be one of "nfs",
4640 	 * "nfs2", "nfs3" or "nfs4".
4641 	 */
4642 	if (ddi_prop_lookup_string(DDI_DEV_T_ANY, ddi_root_node(),
4643 	    DDI_PROP_DONTPASS, "fstype", &propstr)
4644 	    == DDI_SUCCESS) {
4645 		(void) strncpy(rootfs.bo_fstype, propstr, BO_MAXFSNAME);
4646 		ddi_prop_free(propstr);
4647 
4648 	/*
4649 	 * if the boot property 'fstype' is not set, but 'zfs-bootfs' is set,
4650 	 * assume the type of this root filesystem is 'zfs'.
4651 	 */
4652 	} else if (ddi_prop_lookup_string(DDI_DEV_T_ANY, ddi_root_node(),
4653 	    DDI_PROP_DONTPASS, "zfs-bootfs", &propstr)
4654 	    == DDI_SUCCESS) {
4655 		(void) strncpy(rootfs.bo_fstype, "zfs", BO_MAXFSNAME);
4656 		ddi_prop_free(propstr);
4657 	}
4658 
4659 	if (strncmp(rootfs.bo_fstype, "nfs", 3) != 0) {
4660 		*fstypp = *fsmodp = rootfs.bo_fstype;
4661 		return;
4662 	}
4663 
4664 	++netboot;
4665 
4666 	if (strcmp(rootfs.bo_fstype, "nfs2") == 0)
4667 		(void) strcpy(rootfs.bo_fstype, "nfs");
4668 	else if (strcmp(rootfs.bo_fstype, "nfs") == 0)
4669 		(void) strcpy(rootfs.bo_fstype, "nfsdyn");
4670 
4671 	/*
4672 	 * check if path to network interface is specified in bootpath
4673 	 * or by a hypervisor domain configuration file.
4674 	 * XXPV - enable strlumb_get_netdev_path()
4675 	 */
4676 	if (ddi_prop_exists(DDI_DEV_T_ANY, ddi_root_node(), DDI_PROP_DONTPASS,
4677 	    "xpv-nfsroot")) {
4678 		(void) strcpy(rootfs.bo_name, "/xpvd/xnf@0");
4679 	} else if (ddi_prop_lookup_string(DDI_DEV_T_ANY, ddi_root_node(),
4680 	    DDI_PROP_DONTPASS, "bootpath", &propstr)
4681 	    == DDI_SUCCESS) {
4682 		(void) strncpy(rootfs.bo_name, propstr, BO_MAXOBJNAME);
4683 		ddi_prop_free(propstr);
4684 	} else {
4685 		rootfs.bo_name[0] = '\0';
4686 	}
4687 	*fstypp = rootfs.bo_fstype;
4688 	*fsmodp = "nfs";
4689 }
4690 #endif
4691 
4692 /*
4693  * VFS feature routines
4694  */
4695 
4696 #define	VFTINDEX(feature)	(((feature) >> 32) & 0xFFFFFFFF)
4697 #define	VFTBITS(feature)	((feature) & 0xFFFFFFFFLL)
4698 
4699 /* Register a feature in the vfs */
4700 void
4701 vfs_set_feature(vfs_t *vfsp, vfs_feature_t feature)
4702 {
4703 	/* Note that vfs_featureset[] is found in *vfsp->vfs_implp */
4704 	if (vfsp->vfs_implp == NULL)
4705 		return;
4706 
4707 	vfsp->vfs_featureset[VFTINDEX(feature)] |= VFTBITS(feature);
4708 }
4709 
4710 void
4711 vfs_clear_feature(vfs_t *vfsp, vfs_feature_t feature)
4712 {
4713 	/* Note that vfs_featureset[] is found in *vfsp->vfs_implp */
4714 	if (vfsp->vfs_implp == NULL)
4715 		return;
4716 	vfsp->vfs_featureset[VFTINDEX(feature)] &= VFTBITS(~feature);
4717 }
4718 
4719 /*
4720  * Query a vfs for a feature.
4721  * Returns 1 if feature is present, 0 if not
4722  */
4723 int
4724 vfs_has_feature(vfs_t *vfsp, vfs_feature_t feature)
4725 {
4726 	int	ret = 0;
4727 
4728 	/* Note that vfs_featureset[] is found in *vfsp->vfs_implp */
4729 	if (vfsp->vfs_implp == NULL)
4730 		return (ret);
4731 
4732 	if (vfsp->vfs_featureset[VFTINDEX(feature)] & VFTBITS(feature))
4733 		ret = 1;
4734 
4735 	return (ret);
4736 }
4737 
4738 /*
4739  * Propagate feature set from one vfs to another
4740  */
4741 void
4742 vfs_propagate_features(vfs_t *from, vfs_t *to)
4743 {
4744 	int i;
4745 
4746 	if (to->vfs_implp == NULL || from->vfs_implp == NULL)
4747 		return;
4748 
4749 	for (i = 1; i <= to->vfs_featureset[0]; i++) {
4750 		to->vfs_featureset[i] = from->vfs_featureset[i];
4751 	}
4752 }
4753 
4754 #define	LOFINODE_PATH "/dev/lofi/%d"
4755 
4756 /*
4757  * Return the vnode for the lofi node if there's a lofi mount in place.
4758  * Returns -1 when there's no lofi node, 0 on success, and > 0 on
4759  * failure.
4760  */
4761 int
4762 vfs_get_lofi(vfs_t *vfsp, vnode_t **vpp)
4763 {
4764 	char *path = NULL;
4765 	int strsize;
4766 	int err;
4767 
4768 	if (vfsp->vfs_lofi_id == 0) {
4769 		*vpp = NULL;
4770 		return (-1);
4771 	}
4772 
4773 	strsize = snprintf(NULL, 0, LOFINODE_PATH, vfsp->vfs_lofi_id);
4774 	path = kmem_alloc(strsize + 1, KM_SLEEP);
4775 	(void) snprintf(path, strsize + 1, LOFINODE_PATH, vfsp->vfs_lofi_id);
4776 
4777 	/*
4778 	 * We may be inside a zone, so we need to use the /dev path, but
4779 	 * it's created asynchronously, so we wait here.
4780 	 */
4781 	for (;;) {
4782 		err = lookupname(path, UIO_SYSSPACE, FOLLOW, NULLVPP, vpp);
4783 
4784 		if (err != ENOENT)
4785 			break;
4786 
4787 		if ((err = delay_sig(hz / 8)) == EINTR)
4788 			break;
4789 	}
4790 
4791 	if (err)
4792 		*vpp = NULL;
4793 
4794 	kmem_free(path, strsize + 1);
4795 	return (err);
4796 }
4797