xref: /illumos-gate/usr/src/uts/common/fs/ufs/ufs_vfsops.c (revision 08855964b9970604433f7b19dcd71cf5af5e5f14)
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 2009 Sun Microsystems, Inc.  All rights reserved.
24  * Use is subject to license terms.
25  * Copyright 2016 Nexenta Systems, Inc.
26  * Copyright (c) 2017 by Delphix. All rights reserved.
27  * Copyright 2024 Oxide Computer Company
28  */
29 
30 /*	Copyright (c) 1983, 1984, 1985, 1986, 1987, 1988, 1989 AT&T	*/
31 /*	  All Rights Reserved	*/
32 
33 /*
34  * University Copyright- Copyright (c) 1982, 1986, 1988
35  * The Regents of the University of California
36  * All Rights Reserved
37  *
38  * University Acknowledgment- Portions of this document are derived from
39  * software developed by the University of California, Berkeley, and its
40  * contributors.
41  */
42 
43 #include <sys/types.h>
44 #include <sys/t_lock.h>
45 #include <sys/param.h>
46 #include <sys/systm.h>
47 #include <sys/bitmap.h>
48 #include <sys/sysmacros.h>
49 #include <sys/kmem.h>
50 #include <sys/signal.h>
51 #include <sys/user.h>
52 #include <sys/proc.h>
53 #include <sys/disp.h>
54 #include <sys/buf.h>
55 #include <sys/pathname.h>
56 #include <sys/vfs.h>
57 #include <sys/vfs_opreg.h>
58 #include <sys/vnode.h>
59 #include <sys/file.h>
60 #include <sys/atomic.h>
61 #include <sys/uio.h>
62 #include <sys/dkio.h>
63 #include <sys/cred.h>
64 #include <sys/conf.h>
65 #include <sys/dnlc.h>
66 #include <sys/kstat.h>
67 #include <sys/acl.h>
68 #include <sys/fs/ufs_fsdir.h>
69 #include <sys/fs/ufs_fs.h>
70 #include <sys/fs/ufs_inode.h>
71 #include <sys/fs/ufs_mount.h>
72 #include <sys/fs/ufs_acl.h>
73 #include <sys/fs/ufs_panic.h>
74 #include <sys/fs/ufs_bio.h>
75 #include <sys/fs/ufs_quota.h>
76 #include <sys/fs/ufs_log.h>
77 #undef NFS
78 #include <sys/statvfs.h>
79 #include <sys/mount.h>
80 #include <sys/mntent.h>
81 #include <sys/swap.h>
82 #include <sys/errno.h>
83 #include <sys/debug.h>
84 #include "fs/fs_subr.h"
85 #include <sys/cmn_err.h>
86 #include <sys/dnlc.h>
87 #include <sys/fssnap_if.h>
88 #include <sys/sunddi.h>
89 #include <sys/bootconf.h>
90 #include <sys/policy.h>
91 #include <sys/zone.h>
92 
93 /*
94  * This is the loadable module wrapper.
95  */
96 #include <sys/modctl.h>
97 
98 int			ufsfstype;
99 vfsops_t		*ufs_vfsops;
100 static int		ufsinit(int, char *);
101 static int		mountfs();
102 extern int		highbit();
103 extern struct instats	ins;
104 extern struct vnode *common_specvp(struct vnode *vp);
105 extern vfs_t		EIO_vfs;
106 
107 struct  dquot *dquot, *dquotNDQUOT;
108 
109 /*
110  * Cylinder group summary information handling tunable.
111  * This defines when these deltas get logged.
112  * If the number of cylinders in the file system is over the
113  * tunable then we log csum updates. Otherwise the updates are only
114  * done for performance on unmount. After a panic they can be
115  * quickly constructed during mounting. See ufs_construct_si()
116  * called from ufs_getsummaryinfo().
117  *
118  * This performance feature can of course be disabled by setting
119  * ufs_ncg_log to 0, and fully enabled by setting it to 0xffffffff.
120  */
121 #define	UFS_LOG_NCG_DEFAULT 10000
122 uint32_t ufs_ncg_log = UFS_LOG_NCG_DEFAULT;
123 
124 /*
125  * ufs_clean_root indicates whether the root fs went down cleanly
126  */
127 static int ufs_clean_root = 0;
128 
129 /*
130  * UFS Mount options table
131  */
132 static char *intr_cancel[] = { MNTOPT_NOINTR, NULL };
133 static char *nointr_cancel[] = { MNTOPT_INTR, NULL };
134 static char *forcedirectio_cancel[] = { MNTOPT_NOFORCEDIRECTIO, NULL };
135 static char *noforcedirectio_cancel[] = { MNTOPT_FORCEDIRECTIO, NULL };
136 static char *largefiles_cancel[] = { MNTOPT_NOLARGEFILES, NULL };
137 static char *nolargefiles_cancel[] = { MNTOPT_LARGEFILES, NULL };
138 static char *logging_cancel[] = { MNTOPT_NOLOGGING, NULL };
139 static char *nologging_cancel[] = { MNTOPT_LOGGING, NULL };
140 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
141 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
142 static char *quota_cancel[] = { MNTOPT_NOQUOTA, NULL };
143 static char *noquota_cancel[] = { MNTOPT_QUOTA, NULL };
144 static char *dfratime_cancel[] = { MNTOPT_NODFRATIME, NULL };
145 static char *nodfratime_cancel[] = { MNTOPT_DFRATIME, NULL };
146 
147 static mntopt_t mntopts[] = {
148 /*
149  *	option name		cancel option	default arg	flags
150  *		ufs arg flag
151  */
152 	{ MNTOPT_INTR,		intr_cancel,	NULL,		MO_DEFAULT,
153 		(void *)0 },
154 	{ MNTOPT_NOINTR,	nointr_cancel,	NULL,		0,
155 		(void *)UFSMNT_NOINTR },
156 	{ MNTOPT_SYNCDIR,	NULL,		NULL,		0,
157 		(void *)UFSMNT_SYNCDIR },
158 	{ MNTOPT_FORCEDIRECTIO,	forcedirectio_cancel, NULL,	0,
159 		(void *)UFSMNT_FORCEDIRECTIO },
160 	{ MNTOPT_NOFORCEDIRECTIO, noforcedirectio_cancel, NULL, 0,
161 		(void *)UFSMNT_NOFORCEDIRECTIO },
162 	{ MNTOPT_NOSETSEC,	NULL,		NULL,		0,
163 		(void *)UFSMNT_NOSETSEC },
164 	{ MNTOPT_LARGEFILES,	largefiles_cancel, NULL,	MO_DEFAULT,
165 		(void *)UFSMNT_LARGEFILES },
166 	{ MNTOPT_NOLARGEFILES,	nolargefiles_cancel, NULL,	0,
167 		(void *)0 },
168 	{ MNTOPT_LOGGING,	logging_cancel, NULL,		MO_TAG,
169 		(void *)UFSMNT_LOGGING },
170 	{ MNTOPT_NOLOGGING,	nologging_cancel, NULL,
171 		MO_NODISPLAY|MO_DEFAULT|MO_TAG, (void *)0 },
172 	{ MNTOPT_QUOTA,		quota_cancel, NULL,		MO_IGNORE,
173 		(void *)0 },
174 	{ MNTOPT_NOQUOTA,	noquota_cancel,	NULL,
175 		MO_NODISPLAY|MO_DEFAULT, (void *)0 },
176 	{ MNTOPT_GLOBAL,	NULL,		NULL,		0,
177 		(void *)0 },
178 	{ MNTOPT_XATTR,	xattr_cancel,		NULL,		MO_DEFAULT,
179 		(void *)0 },
180 	{ MNTOPT_NOXATTR,	noxattr_cancel,		NULL,		0,
181 		(void *)0 },
182 	{ MNTOPT_NOATIME,	NULL,		NULL,		0,
183 		(void *)UFSMNT_NOATIME },
184 	{ MNTOPT_DFRATIME,	dfratime_cancel, NULL,		0,
185 		(void *)0 },
186 	{ MNTOPT_NODFRATIME,	nodfratime_cancel, NULL,
187 		MO_NODISPLAY|MO_DEFAULT, (void *)UFSMNT_NODFRATIME },
188 	{ MNTOPT_ONERROR,	NULL,		UFSMNT_ONERROR_PANIC_STR,
189 		MO_DEFAULT|MO_HASVALUE,	(void *)0 },
190 };
191 
192 static mntopts_t ufs_mntopts = {
193 	sizeof (mntopts) / sizeof (mntopt_t),
194 	mntopts
195 };
196 
197 static vfsdef_t vfw = {
198 	VFSDEF_VERSION,
199 	"ufs",
200 	ufsinit,
201 	VSW_HASPROTO|VSW_CANREMOUNT|VSW_STATS|VSW_CANLOFI|VSW_MOUNTDEV,
202 	&ufs_mntopts
203 };
204 
205 /*
206  * Module linkage information for the kernel.
207  */
208 extern struct mod_ops mod_fsops;
209 
210 static struct modlfs modlfs = {
211 	&mod_fsops, "filesystem for ufs", &vfw
212 };
213 
214 static struct modlinkage modlinkage = {
215 	MODREV_1, (void *)&modlfs, NULL
216 };
217 
218 /*
219  * An attempt has been made to make this module unloadable.  In order to
220  * test it, we need a system in which the root fs is NOT ufs.  THIS HAS NOT
221  * BEEN DONE
222  */
223 
224 extern kstat_t *ufs_inode_kstat;
225 extern uint_t ufs_lockfs_key;
226 extern void ufs_lockfs_tsd_destructor(void *);
227 extern uint_t bypass_snapshot_throttle_key;
228 
229 int
230 _init(void)
231 {
232 	/*
233 	 * Create an index into the per thread array so that any thread doing
234 	 * VOP will have a lockfs mark on it.
235 	 */
236 	tsd_create(&ufs_lockfs_key, ufs_lockfs_tsd_destructor);
237 	tsd_create(&bypass_snapshot_throttle_key, NULL);
238 	return (mod_install(&modlinkage));
239 }
240 
241 int
242 _fini(void)
243 {
244 	return (EBUSY);
245 }
246 
247 int
248 _info(struct modinfo *modinfop)
249 {
250 	return (mod_info(&modlinkage, modinfop));
251 }
252 
253 extern struct vnode *makespecvp(dev_t dev, vtype_t type);
254 
255 extern kmutex_t	ufs_scan_lock;
256 
257 static int mountfs(struct vfs *, enum whymountroot, struct vnode *, char *,
258 		struct cred *, int, void *, int);
259 
260 
261 static int
262 ufs_mount(struct vfs *vfsp, struct vnode *mvp, struct mounta *uap,
263     struct cred *cr)
264 {
265 	char *data = uap->dataptr;
266 	int datalen = uap->datalen;
267 	dev_t dev;
268 	struct vnode *lvp = NULL;
269 	struct vnode *svp = NULL;
270 	struct pathname dpn;
271 	int error;
272 	enum whymountroot why = ROOT_INIT;
273 	struct ufs_args args;
274 	int oflag, aflag;
275 	int fromspace = (uap->flags & MS_SYSSPACE) ?
276 	    UIO_SYSSPACE : UIO_USERSPACE;
277 
278 	if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
279 		return (error);
280 
281 	if (mvp->v_type != VDIR)
282 		return (ENOTDIR);
283 
284 	mutex_enter(&mvp->v_lock);
285 	if ((uap->flags & MS_REMOUNT) == 0 &&
286 	    (uap->flags & MS_OVERLAY) == 0 &&
287 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
288 		mutex_exit(&mvp->v_lock);
289 		return (EBUSY);
290 	}
291 	mutex_exit(&mvp->v_lock);
292 
293 	/*
294 	 * Get arguments
295 	 */
296 	bzero(&args, sizeof (args));
297 	if ((uap->flags & MS_DATA) && data != NULL && datalen != 0) {
298 		int copy_result = 0;
299 
300 		if (datalen > sizeof (args))
301 			return (EINVAL);
302 		if (uap->flags & MS_SYSSPACE)
303 			bcopy(data, &args, datalen);
304 		else
305 			copy_result = copyin(data, &args, datalen);
306 		if (copy_result)
307 			return (EFAULT);
308 		datalen = sizeof (struct ufs_args);
309 	} else {
310 		datalen = 0;
311 	}
312 
313 	if ((vfsp->vfs_flag & VFS_RDONLY) != 0 ||
314 	    (uap->flags & MS_RDONLY) != 0) {
315 		oflag = FREAD;
316 		aflag = VREAD;
317 	} else {
318 		oflag = FREAD | FWRITE;
319 		aflag = VREAD | VWRITE;
320 	}
321 
322 	/*
323 	 * Read in the mount point pathname
324 	 * (so we can record the directory the file system was last mounted on).
325 	 */
326 	if (error = pn_get(uap->dir, fromspace, &dpn))
327 		return (error);
328 
329 	/*
330 	 * Resolve path name of special file being mounted.
331 	 */
332 	if (error = lookupname(uap->spec, fromspace, FOLLOW, NULL, &svp)) {
333 		pn_free(&dpn);
334 		return (error);
335 	}
336 
337 	error = vfs_get_lofi(vfsp, &lvp);
338 
339 	if (error > 0) {
340 		VN_RELE(svp);
341 		pn_free(&dpn);
342 		return (error);
343 	} else if (error == 0) {
344 		dev = lvp->v_rdev;
345 
346 		if (getmajor(dev) >= devcnt) {
347 			error = ENXIO;
348 			goto out;
349 		}
350 	} else {
351 		dev = svp->v_rdev;
352 
353 		if (svp->v_type != VBLK) {
354 			VN_RELE(svp);
355 			pn_free(&dpn);
356 			return (ENOTBLK);
357 		}
358 
359 		if (getmajor(dev) >= devcnt) {
360 			error = ENXIO;
361 			goto out;
362 		}
363 
364 		/*
365 		 * In SunCluster, requests to a global device are
366 		 * satisfied by a local device. We substitute the global
367 		 * pxfs node with a local spec node here.
368 		 */
369 		if (IS_PXFSVP(svp)) {
370 			ASSERT(lvp == NULL);
371 			VN_RELE(svp);
372 			svp = makespecvp(dev, VBLK);
373 		}
374 
375 		if ((error = secpolicy_spec_open(cr, svp, oflag)) != 0) {
376 			VN_RELE(svp);
377 			pn_free(&dpn);
378 			return (error);
379 		}
380 	}
381 
382 	if (uap->flags & MS_REMOUNT)
383 		why = ROOT_REMOUNT;
384 
385 	/*
386 	 * Open device/file mounted on.  We need this to check whether
387 	 * the caller has sufficient rights to access the resource in
388 	 * question.  When bio is fixed for vnodes this can all be vnode
389 	 * operations.
390 	 */
391 	if ((error = VOP_ACCESS(svp, aflag, 0, cr, NULL)) != 0)
392 		goto out;
393 
394 	/*
395 	 * Ensure that this device isn't already mounted or in progress on a
396 	 * mount unless this is a REMOUNT request or we are told to suppress
397 	 * mount checks. Global mounts require special handling.
398 	 */
399 	if ((uap->flags & MS_NOCHECK) == 0) {
400 		if ((uap->flags & MS_GLOBAL) == 0 &&
401 		    vfs_devmounting(dev, vfsp)) {
402 			error = EBUSY;
403 			goto out;
404 		}
405 		if (vfs_devismounted(dev)) {
406 			if ((uap->flags & MS_REMOUNT) == 0) {
407 				error = EBUSY;
408 				goto out;
409 			}
410 		}
411 	}
412 
413 	/*
414 	 * If the device is a tape, mount it read only
415 	 */
416 	if (devopsp[getmajor(dev)]->devo_cb_ops->cb_flag & D_TAPE) {
417 		vfsp->vfs_flag |= VFS_RDONLY;
418 		vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
419 	}
420 	if (uap->flags & MS_RDONLY)
421 		vfsp->vfs_flag |= VFS_RDONLY;
422 
423 	/*
424 	 * Mount the filesystem, free the device vnode on error.
425 	 */
426 	error = mountfs(vfsp, why, lvp != NULL ? lvp : svp,
427 	    dpn.pn_path, cr, 0, &args, datalen);
428 
429 	if (error == 0) {
430 		vfs_set_feature(vfsp, VFSFT_SYSATTR_VIEWS);
431 
432 		/*
433 		 * If lofi, drop our reference to the original file.
434 		 */
435 		if (lvp != NULL)
436 			VN_RELE(svp);
437 	}
438 
439 out:
440 	pn_free(&dpn);
441 
442 	if (error) {
443 		if (lvp != NULL)
444 			VN_RELE(lvp);
445 		if (svp != NULL)
446 			VN_RELE(svp);
447 	}
448 	return (error);
449 }
450 
451 /*
452  * Mount root file system.
453  * "why" is ROOT_INIT on initial call ROOT_REMOUNT if called to
454  * remount the root file system, and ROOT_UNMOUNT if called to
455  * unmount the root (e.g., as part of a system shutdown).
456  *
457  * XXX - this may be partially machine-dependent; it, along with the VFS_SWAPVP
458  * operation, goes along with auto-configuration.  A mechanism should be
459  * provided by which machine-INdependent code in the kernel can say "get me the
460  * right root file system" and "get me the right initial swap area", and have
461  * that done in what may well be a machine-dependent fashion.
462  * Unfortunately, it is also file-system-type dependent (NFS gets it via
463  * bootparams calls, UFS gets it from various and sundry machine-dependent
464  * mechanisms, as SPECFS does for swap).
465  */
466 static int
467 ufs_mountroot(struct vfs *vfsp, enum whymountroot why)
468 {
469 	struct fs *fsp;
470 	int error;
471 	static int ufsrootdone = 0;
472 	dev_t rootdev;
473 	struct vnode *vp;
474 	struct vnode *devvp = 0;
475 	int ovflags;
476 	int doclkset;
477 	ufsvfs_t *ufsvfsp;
478 
479 	if (why == ROOT_INIT) {
480 		if (ufsrootdone++)
481 			return (EBUSY);
482 		rootdev = getrootdev();
483 		if (rootdev == (dev_t)NODEV)
484 			return (ENODEV);
485 		vfsp->vfs_dev = rootdev;
486 		vfsp->vfs_flag |= VFS_RDONLY;
487 	} else if (why == ROOT_REMOUNT) {
488 		vp = ((struct ufsvfs *)vfsp->vfs_data)->vfs_devvp;
489 		(void) dnlc_purge_vfsp(vfsp, 0);
490 		vp = common_specvp(vp);
491 		(void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_INVAL,
492 		    CRED(), NULL);
493 		(void) bfinval(vfsp->vfs_dev, 0);
494 		fsp = getfs(vfsp);
495 
496 		ovflags = vfsp->vfs_flag;
497 		vfsp->vfs_flag &= ~VFS_RDONLY;
498 		vfsp->vfs_flag |= VFS_REMOUNT;
499 		rootdev = vfsp->vfs_dev;
500 	} else if (why == ROOT_UNMOUNT) {
501 		if (vfs_lock(vfsp) == 0) {
502 			(void) ufs_flush(vfsp);
503 			/*
504 			 * Mark the log as fully rolled
505 			 */
506 			ufsvfsp = (ufsvfs_t *)vfsp->vfs_data;
507 			fsp = ufsvfsp->vfs_fs;
508 			if (TRANS_ISTRANS(ufsvfsp) &&
509 			    !TRANS_ISERROR(ufsvfsp) &&
510 			    (fsp->fs_rolled == FS_NEED_ROLL)) {
511 				ml_unit_t *ul = ufsvfsp->vfs_log;
512 
513 				error = ufs_putsummaryinfo(ul->un_dev,
514 				    ufsvfsp, fsp);
515 				if (error == 0) {
516 					fsp->fs_rolled = FS_ALL_ROLLED;
517 					UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp);
518 				}
519 			}
520 			vfs_unlock(vfsp);
521 		} else {
522 			ufs_update(0);
523 		}
524 
525 		vp = ((struct ufsvfs *)vfsp->vfs_data)->vfs_devvp;
526 		(void) VOP_CLOSE(vp, FREAD|FWRITE, 1,
527 		    (offset_t)0, CRED(), NULL);
528 		return (0);
529 	}
530 	error = vfs_lock(vfsp);
531 	if (error)
532 		return (error);
533 
534 	devvp = makespecvp(rootdev, VBLK);
535 
536 	/* If RO media, don't call clkset() (see below) */
537 	doclkset = 1;
538 	if (why == ROOT_INIT) {
539 		error = VOP_OPEN(&devvp, FREAD|FWRITE, CRED(), NULL);
540 		if (error == 0) {
541 			(void) VOP_CLOSE(devvp, FREAD|FWRITE, 1,
542 			    (offset_t)0, CRED(), NULL);
543 		} else {
544 			doclkset = 0;
545 		}
546 	}
547 
548 	error = mountfs(vfsp, why, devvp, "/", CRED(), 1, NULL, 0);
549 	/*
550 	 * XXX - assumes root device is not indirect, because we don't set
551 	 * rootvp.  Is rootvp used for anything?  If so, make another arg
552 	 * to mountfs.
553 	 */
554 	if (error) {
555 		vfs_unlock(vfsp);
556 		if (why == ROOT_REMOUNT)
557 			vfsp->vfs_flag = ovflags;
558 		if (rootvp) {
559 			VN_RELE(rootvp);
560 			rootvp = (struct vnode *)0;
561 		}
562 		VN_RELE(devvp);
563 		return (error);
564 	}
565 	if (why == ROOT_INIT)
566 		vfs_add((struct vnode *)0, vfsp,
567 		    (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
568 	vfs_unlock(vfsp);
569 	fsp = getfs(vfsp);
570 	clkset(doclkset ? fsp->fs_time : -1);
571 	ufsvfsp = (ufsvfs_t *)vfsp->vfs_data;
572 	if (ufsvfsp->vfs_log) {
573 		vfs_setmntopt(vfsp, MNTOPT_LOGGING, NULL, 0);
574 	}
575 	return (0);
576 }
577 
578 static int
579 remountfs(struct vfs *vfsp, dev_t dev, void *raw_argsp, int args_len)
580 {
581 	struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
582 	struct ulockfs *ulp = &ufsvfsp->vfs_ulockfs;
583 	struct buf *bp = ufsvfsp->vfs_bufp;
584 	struct fs *fsp = (struct fs *)bp->b_un.b_addr;
585 	struct fs *fspt;
586 	struct buf *tpt = 0;
587 	int error = 0;
588 	int flags = 0;
589 
590 	if (args_len == sizeof (struct ufs_args) && raw_argsp)
591 		flags = ((struct ufs_args *)raw_argsp)->flags;
592 
593 	/* cannot remount to RDONLY */
594 	if (vfsp->vfs_flag & VFS_RDONLY)
595 		return (ENOTSUP);
596 
597 	/* whoops, wrong dev */
598 	if (vfsp->vfs_dev != dev)
599 		return (EINVAL);
600 
601 	/*
602 	 * synchronize w/ufs ioctls
603 	 */
604 	mutex_enter(&ulp->ul_lock);
605 	atomic_inc_ulong(&ufs_quiesce_pend);
606 
607 	/*
608 	 * reset options
609 	 */
610 	ufsvfsp->vfs_nointr  = flags & UFSMNT_NOINTR;
611 	ufsvfsp->vfs_syncdir = flags & UFSMNT_SYNCDIR;
612 	ufsvfsp->vfs_nosetsec = flags & UFSMNT_NOSETSEC;
613 	ufsvfsp->vfs_noatime = flags & UFSMNT_NOATIME;
614 	if ((flags & UFSMNT_NODFRATIME) || ufsvfsp->vfs_noatime)
615 		ufsvfsp->vfs_dfritime &= ~UFS_DFRATIME;
616 	else	/* dfratime, default behavior */
617 		ufsvfsp->vfs_dfritime |= UFS_DFRATIME;
618 	if (flags & UFSMNT_FORCEDIRECTIO)
619 		ufsvfsp->vfs_forcedirectio = 1;
620 	else	/* default is no direct I/O */
621 		ufsvfsp->vfs_forcedirectio = 0;
622 	ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
623 
624 	/*
625 	 * set largefiles flag in ufsvfs equal to the
626 	 * value passed in by the mount command. If
627 	 * it is "nolargefiles", and the flag is set
628 	 * in the superblock, the mount fails.
629 	 */
630 	if (!(flags & UFSMNT_LARGEFILES)) {  /* "nolargefiles" */
631 		if (fsp->fs_flags & FSLARGEFILES) {
632 			error = EFBIG;
633 			goto remounterr;
634 		}
635 		ufsvfsp->vfs_lfflags &= ~UFS_LARGEFILES;
636 	} else	/* "largefiles" */
637 		ufsvfsp->vfs_lfflags |= UFS_LARGEFILES;
638 	/*
639 	 * read/write to read/write; all done
640 	 */
641 	if (fsp->fs_ronly == 0)
642 		goto remounterr;
643 
644 	/*
645 	 * fix-on-panic assumes RO->RW remount implies system-critical fs
646 	 * if it is shortly after boot; so, don't attempt to lock and fix
647 	 * (unless the user explicitly asked for another action on error)
648 	 * XXX UFSMNT_ONERROR_RDONLY rather than UFSMNT_ONERROR_PANIC
649 	 */
650 #define	BOOT_TIME_LIMIT	(180*hz)
651 	if (!(flags & UFSMNT_ONERROR_FLGMASK) &&
652 	    ddi_get_lbolt() < BOOT_TIME_LIMIT) {
653 		cmn_err(CE_WARN, "%s is required to be mounted onerror=%s",
654 		    ufsvfsp->vfs_fs->fs_fsmnt, UFSMNT_ONERROR_PANIC_STR);
655 		flags |= UFSMNT_ONERROR_PANIC;
656 	}
657 
658 	if ((error = ufsfx_mount(ufsvfsp, flags)) != 0)
659 		goto remounterr;
660 
661 	/*
662 	 * quiesce the file system
663 	 */
664 	error = ufs_quiesce(ulp);
665 	if (error)
666 		goto remounterr;
667 
668 	tpt = UFS_BREAD(ufsvfsp, ufsvfsp->vfs_dev, SBLOCK, SBSIZE);
669 	if (tpt->b_flags & B_ERROR) {
670 		error = EIO;
671 		goto remounterr;
672 	}
673 	fspt = (struct fs *)tpt->b_un.b_addr;
674 	if (((fspt->fs_magic != FS_MAGIC) &&
675 	    (fspt->fs_magic != MTB_UFS_MAGIC)) ||
676 	    (fspt->fs_magic == FS_MAGIC &&
677 	    (fspt->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
678 	    fspt->fs_version != UFS_VERSION_MIN)) ||
679 	    (fspt->fs_magic == MTB_UFS_MAGIC &&
680 	    (fspt->fs_version > MTB_UFS_VERSION_1 ||
681 	    fspt->fs_version < MTB_UFS_VERSION_MIN)) ||
682 	    fspt->fs_bsize > MAXBSIZE || fspt->fs_frag > MAXFRAG ||
683 	    fspt->fs_bsize < sizeof (struct fs) || fspt->fs_bsize < PAGESIZE) {
684 		tpt->b_flags |= B_STALE | B_AGE;
685 		error = EINVAL;
686 		goto remounterr;
687 	}
688 
689 	if (ufsvfsp->vfs_log && (ufsvfsp->vfs_log->un_flags & LDL_NOROLL)) {
690 		ufsvfsp->vfs_log->un_flags &= ~LDL_NOROLL;
691 		logmap_start_roll(ufsvfsp->vfs_log);
692 	}
693 
694 	if (TRANS_ISERROR(ufsvfsp))
695 		goto remounterr;
696 	TRANS_DOMATAMAP(ufsvfsp);
697 
698 	if ((fspt->fs_state + fspt->fs_time == FSOKAY) &&
699 	    fspt->fs_clean == FSLOG && !TRANS_ISTRANS(ufsvfsp)) {
700 		ufsvfsp->vfs_log = NULL;
701 		ufsvfsp->vfs_domatamap = 0;
702 		error = ENOSPC;
703 		goto remounterr;
704 	}
705 
706 	if (fspt->fs_state + fspt->fs_time == FSOKAY &&
707 	    (fspt->fs_clean == FSCLEAN ||
708 	    fspt->fs_clean == FSSTABLE ||
709 	    fspt->fs_clean == FSLOG)) {
710 
711 		/*
712 		 * Ensure that ufs_getsummaryinfo doesn't reconstruct
713 		 * the summary info.
714 		 */
715 		error = ufs_getsummaryinfo(vfsp->vfs_dev, ufsvfsp, fspt);
716 		if (error)
717 			goto remounterr;
718 
719 		/* preserve mount name */
720 		(void) strncpy(fspt->fs_fsmnt, fsp->fs_fsmnt, MAXMNTLEN);
721 		/* free the old cg space */
722 		kmem_free(fsp->fs_u.fs_csp, fsp->fs_cssize);
723 		/* switch in the new superblock */
724 		fspt->fs_rolled = FS_NEED_ROLL;
725 		bcopy(tpt->b_un.b_addr, bp->b_un.b_addr, fspt->fs_sbsize);
726 
727 		fsp->fs_clean = FSSTABLE;
728 	} /* superblock updated in memory */
729 	tpt->b_flags |= B_STALE | B_AGE;
730 	brelse(tpt);
731 	tpt = 0;
732 
733 	if (fsp->fs_clean != FSSTABLE) {
734 		error = ENOSPC;
735 		goto remounterr;
736 	}
737 
738 
739 	if (TRANS_ISTRANS(ufsvfsp)) {
740 		fsp->fs_clean = FSLOG;
741 		ufsvfsp->vfs_dio = 0;
742 	} else
743 		if (ufsvfsp->vfs_dio)
744 			fsp->fs_clean = FSSUSPEND;
745 
746 	TRANS_MATA_MOUNT(ufsvfsp);
747 
748 	fsp->fs_fmod = 0;
749 	fsp->fs_ronly = 0;
750 
751 	atomic_dec_ulong(&ufs_quiesce_pend);
752 	cv_broadcast(&ulp->ul_cv);
753 	mutex_exit(&ulp->ul_lock);
754 
755 	if (TRANS_ISTRANS(ufsvfsp)) {
756 
757 		/*
758 		 * start the delete thread
759 		 */
760 		ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp);
761 
762 		/*
763 		 * start the reclaim thread
764 		 */
765 		if (fsp->fs_reclaim & (FS_RECLAIM|FS_RECLAIMING)) {
766 			fsp->fs_reclaim &= ~FS_RECLAIM;
767 			fsp->fs_reclaim |=  FS_RECLAIMING;
768 			ufs_thread_start(&ufsvfsp->vfs_reclaim,
769 			    ufs_thread_reclaim, vfsp);
770 		}
771 	}
772 
773 	TRANS_SBWRITE(ufsvfsp, TOP_MOUNT);
774 
775 	return (0);
776 
777 remounterr:
778 	if (tpt)
779 		brelse(tpt);
780 	atomic_dec_ulong(&ufs_quiesce_pend);
781 	cv_broadcast(&ulp->ul_cv);
782 	mutex_exit(&ulp->ul_lock);
783 	return (error);
784 }
785 
786 /*
787  * If the device maxtransfer size is not available, we use ufs_maxmaxphys
788  * along with the system value for maxphys to determine the value for
789  * maxtransfer.
790  */
791 int ufs_maxmaxphys = (1024 * 1024);
792 
793 #include <sys/ddi.h>		/* for delay(9f) */
794 
795 int ufs_mount_error_delay = 20;	/* default to 20ms */
796 int ufs_mount_timeout = 60000;	/* default to 1 minute */
797 
798 static int
799 mountfs(struct vfs *vfsp, enum whymountroot why, struct vnode *devvp,
800     char *path, cred_t *cr, int isroot, void *raw_argsp, int args_len)
801 {
802 	dev_t dev = devvp->v_rdev;
803 	struct fs *fsp;
804 	struct ufsvfs *ufsvfsp = 0;
805 	struct buf *bp = 0;
806 	struct buf *tp = 0;
807 	struct dk_cinfo ci;
808 	int error = 0;
809 	size_t len;
810 	int needclose = 0;
811 	int needtrans = 0;
812 	struct inode *rip;
813 	struct vnode *rvp = NULL;
814 	int flags = 0;
815 	kmutex_t *ihm;
816 	int elapsed;
817 	int status;
818 	extern	int	maxphys;
819 
820 	if (args_len == sizeof (struct ufs_args) && raw_argsp)
821 		flags = ((struct ufs_args *)raw_argsp)->flags;
822 
823 	ASSERT(vfs_lock_held(vfsp));
824 
825 	if (why == ROOT_INIT) {
826 		/*
827 		 * Open block device mounted on.
828 		 * When bio is fixed for vnodes this can all be vnode
829 		 * operations.
830 		 */
831 		error = VOP_OPEN(&devvp,
832 		    (vfsp->vfs_flag & VFS_RDONLY) ? FREAD : FREAD|FWRITE,
833 		    cr, NULL);
834 		if (error)
835 			goto out;
836 		needclose = 1;
837 
838 		/*
839 		 * Refuse to go any further if this
840 		 * device is being used for swapping.
841 		 */
842 		if (IS_SWAPVP(devvp)) {
843 			error = EBUSY;
844 			goto out;
845 		}
846 	}
847 
848 	/*
849 	 * check for dev already mounted on
850 	 */
851 	if (vfsp->vfs_flag & VFS_REMOUNT) {
852 		error = remountfs(vfsp, dev, raw_argsp, args_len);
853 		if (error == 0)
854 			VN_RELE(devvp);
855 		return (error);
856 	}
857 
858 	ASSERT(devvp != 0);
859 
860 	/*
861 	 * Flush back any dirty pages on the block device to
862 	 * try and keep the buffer cache in sync with the page
863 	 * cache if someone is trying to use block devices when
864 	 * they really should be using the raw device.
865 	 */
866 	(void) VOP_PUTPAGE(common_specvp(devvp), (offset_t)0,
867 	    (size_t)0, B_INVAL, cr, NULL);
868 
869 	/*
870 	 * read in superblock
871 	 */
872 	ufsvfsp = kmem_zalloc(sizeof (struct ufsvfs), KM_SLEEP);
873 	tp = UFS_BREAD(ufsvfsp, dev, SBLOCK, SBSIZE);
874 	if (tp->b_flags & B_ERROR)
875 		goto out;
876 	fsp = (struct fs *)tp->b_un.b_addr;
877 
878 	if ((fsp->fs_magic != FS_MAGIC) && (fsp->fs_magic != MTB_UFS_MAGIC)) {
879 		cmn_err(CE_NOTE,
880 		    "mount: not a UFS magic number (0x%x)", fsp->fs_magic);
881 		error = EINVAL;
882 		goto out;
883 	}
884 
885 	if ((fsp->fs_magic == FS_MAGIC) &&
886 	    (fsp->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
887 	    fsp->fs_version != UFS_VERSION_MIN)) {
888 		cmn_err(CE_NOTE,
889 		    "mount: unrecognized version of UFS on-disk format: %d",
890 		    fsp->fs_version);
891 		error = EINVAL;
892 		goto out;
893 	}
894 
895 	if ((fsp->fs_magic == MTB_UFS_MAGIC) &&
896 	    (fsp->fs_version > MTB_UFS_VERSION_1 ||
897 	    fsp->fs_version < MTB_UFS_VERSION_MIN)) {
898 		cmn_err(CE_NOTE,
899 		    "mount: unrecognized version of UFS on-disk format: %d",
900 		    fsp->fs_version);
901 		error = EINVAL;
902 		goto out;
903 	}
904 
905 #ifndef _LP64
906 	if (fsp->fs_magic == MTB_UFS_MAGIC) {
907 		/*
908 		 * Find the size of the device in sectors.  If the
909 		 * the size in sectors is greater than INT_MAX, it's
910 		 * a multi-terabyte file system, which can't be
911 		 * mounted by a 32-bit kernel.  We can't use the
912 		 * fsbtodb() macro in the next line because the macro
913 		 * casts the intermediate values to daddr_t, which is
914 		 * a 32-bit quantity in a 32-bit kernel.  Here we
915 		 * really do need the intermediate values to be held
916 		 * in 64-bit quantities because we're checking for
917 		 * overflow of a 32-bit field.
918 		 */
919 		if ((((diskaddr_t)(fsp->fs_size)) << fsp->fs_fsbtodb)
920 		    > INT_MAX) {
921 			cmn_err(CE_NOTE,
922 			    "mount: multi-terabyte UFS cannot be"
923 			    " mounted by a 32-bit kernel");
924 			error = EINVAL;
925 			goto out;
926 		}
927 
928 	}
929 #endif
930 
931 	if (fsp->fs_bsize > MAXBSIZE || fsp->fs_frag > MAXFRAG ||
932 	    fsp->fs_bsize < sizeof (struct fs) || fsp->fs_bsize < PAGESIZE) {
933 		error = EINVAL;	/* also needs translation */
934 		goto out;
935 	}
936 
937 	/*
938 	 * Allocate VFS private data.
939 	 */
940 	vfsp->vfs_bcount = 0;
941 	vfsp->vfs_data = (caddr_t)ufsvfsp;
942 	vfsp->vfs_fstype = ufsfstype;
943 	vfsp->vfs_dev = dev;
944 	vfsp->vfs_flag |= VFS_NOTRUNC;
945 	vfs_make_fsid(&vfsp->vfs_fsid, dev, ufsfstype);
946 	ufsvfsp->vfs_devvp = devvp;
947 
948 	/*
949 	 * Cross-link with vfs and add to instance list.
950 	 */
951 	ufsvfsp->vfs_vfs = vfsp;
952 	ufs_vfs_add(ufsvfsp);
953 
954 	ufsvfsp->vfs_dev = dev;
955 	ufsvfsp->vfs_bufp = tp;
956 
957 	ufsvfsp->vfs_dirsize = INODESIZE + (4 * ALLOCSIZE) + fsp->fs_fsize;
958 	ufsvfsp->vfs_minfrags =
959 	    (int)((int64_t)fsp->fs_dsize * fsp->fs_minfree / 100);
960 	/*
961 	 * if mount allows largefiles, indicate so in ufsvfs
962 	 */
963 	if (flags & UFSMNT_LARGEFILES)
964 		ufsvfsp->vfs_lfflags |= UFS_LARGEFILES;
965 	/*
966 	 * Initialize threads
967 	 */
968 	ufs_delete_init(ufsvfsp, 1);
969 	ufs_thread_init(&ufsvfsp->vfs_reclaim, 0);
970 
971 	/*
972 	 * Chicken and egg problem. The superblock may have deltas
973 	 * in the log.  So after the log is scanned we reread the
974 	 * superblock. We guarantee that the fields needed to
975 	 * scan the log will not be in the log.
976 	 */
977 	if (fsp->fs_logbno && fsp->fs_clean == FSLOG &&
978 	    (fsp->fs_state + fsp->fs_time == FSOKAY)) {
979 		error = lufs_snarf(ufsvfsp, fsp, (vfsp->vfs_flag & VFS_RDONLY));
980 		if (error) {
981 			/*
982 			 * Allow a ro mount to continue even if the
983 			 * log cannot be processed - yet.
984 			 */
985 			if (!(vfsp->vfs_flag & VFS_RDONLY)) {
986 				cmn_err(CE_WARN, "Error accessing ufs "
987 				    "log for %s; Please run fsck(8)", path);
988 				goto out;
989 			}
990 		}
991 		tp->b_flags |= (B_AGE | B_STALE);
992 		brelse(tp);
993 		tp = UFS_BREAD(ufsvfsp, dev, SBLOCK, SBSIZE);
994 		fsp = (struct fs *)tp->b_un.b_addr;
995 		ufsvfsp->vfs_bufp = tp;
996 		if (tp->b_flags & B_ERROR)
997 			goto out;
998 	}
999 
1000 	/*
1001 	 * Set logging mounted flag used by lockfs
1002 	 */
1003 	ufsvfsp->vfs_validfs = UT_MOUNTED;
1004 
1005 	/*
1006 	 * Copy the super block into a buffer in its native size.
1007 	 * Use ngeteblk to allocate the buffer
1008 	 */
1009 	bp = ngeteblk(fsp->fs_bsize);
1010 	ufsvfsp->vfs_bufp = bp;
1011 	bp->b_edev = dev;
1012 	bp->b_dev = cmpdev(dev);
1013 	bp->b_blkno = SBLOCK;
1014 	bp->b_bcount = fsp->fs_sbsize;
1015 	bcopy(tp->b_un.b_addr, bp->b_un.b_addr, fsp->fs_sbsize);
1016 	tp->b_flags |= B_STALE | B_AGE;
1017 	brelse(tp);
1018 	tp = 0;
1019 
1020 	fsp = (struct fs *)bp->b_un.b_addr;
1021 	/*
1022 	 * Mount fails if superblock flag indicates presence of large
1023 	 * files and filesystem is attempted to be mounted 'nolargefiles'.
1024 	 * The exception is for a read only mount of root, which we
1025 	 * always want to succeed, so fsck can fix potential problems.
1026 	 * The assumption is that we will remount root at some point,
1027 	 * and the remount will enforce the mount option.
1028 	 */
1029 	if (!(isroot & (vfsp->vfs_flag & VFS_RDONLY)) &&
1030 	    (fsp->fs_flags & FSLARGEFILES) &&
1031 	    !(flags & UFSMNT_LARGEFILES)) {
1032 		error = EFBIG;
1033 		goto out;
1034 	}
1035 
1036 	if (vfsp->vfs_flag & VFS_RDONLY) {
1037 		fsp->fs_ronly = 1;
1038 		fsp->fs_fmod = 0;
1039 		if (((fsp->fs_state + fsp->fs_time) == FSOKAY) &&
1040 		    ((fsp->fs_clean == FSCLEAN) ||
1041 		    (fsp->fs_clean == FSSTABLE) ||
1042 		    (fsp->fs_clean == FSLOG))) {
1043 			if (isroot) {
1044 				if (fsp->fs_clean == FSLOG) {
1045 					if (fsp->fs_rolled == FS_ALL_ROLLED) {
1046 						ufs_clean_root = 1;
1047 					}
1048 				} else {
1049 					ufs_clean_root = 1;
1050 				}
1051 			}
1052 			fsp->fs_clean = FSSTABLE;
1053 		} else {
1054 			fsp->fs_clean = FSBAD;
1055 		}
1056 	} else {
1057 
1058 		fsp->fs_fmod = 0;
1059 		fsp->fs_ronly = 0;
1060 
1061 		TRANS_DOMATAMAP(ufsvfsp);
1062 
1063 		if ((TRANS_ISERROR(ufsvfsp)) ||
1064 		    (((fsp->fs_state + fsp->fs_time) == FSOKAY) &&
1065 		    fsp->fs_clean == FSLOG && !TRANS_ISTRANS(ufsvfsp))) {
1066 			ufsvfsp->vfs_log = NULL;
1067 			ufsvfsp->vfs_domatamap = 0;
1068 			error = ENOSPC;
1069 			goto out;
1070 		}
1071 
1072 		if (((fsp->fs_state + fsp->fs_time) == FSOKAY) &&
1073 		    (fsp->fs_clean == FSCLEAN ||
1074 		    fsp->fs_clean == FSSTABLE ||
1075 		    fsp->fs_clean == FSLOG))
1076 			fsp->fs_clean = FSSTABLE;
1077 		else {
1078 			if (isroot) {
1079 				/*
1080 				 * allow root partition to be mounted even
1081 				 * when fs_state is not ok
1082 				 * will be fixed later by a remount root
1083 				 */
1084 				fsp->fs_clean = FSBAD;
1085 				ufsvfsp->vfs_log = NULL;
1086 				ufsvfsp->vfs_domatamap = 0;
1087 			} else {
1088 				error = ENOSPC;
1089 				goto out;
1090 			}
1091 		}
1092 
1093 		if (fsp->fs_clean == FSSTABLE && TRANS_ISTRANS(ufsvfsp))
1094 			fsp->fs_clean = FSLOG;
1095 	}
1096 	TRANS_MATA_MOUNT(ufsvfsp);
1097 	needtrans = 1;
1098 
1099 	vfsp->vfs_bsize = fsp->fs_bsize;
1100 
1101 	/*
1102 	 * Read in summary info
1103 	 */
1104 	if (error = ufs_getsummaryinfo(dev, ufsvfsp, fsp))
1105 		goto out;
1106 
1107 	/*
1108 	 * lastwhinetime is set to zero rather than lbolt, so that after
1109 	 * mounting if the filesystem is found to be full, then immediately the
1110 	 * "file system message" will be logged.
1111 	 */
1112 	ufsvfsp->vfs_lastwhinetime = 0L;
1113 
1114 
1115 	mutex_init(&ufsvfsp->vfs_lock, NULL, MUTEX_DEFAULT, NULL);
1116 	(void) copystr(path, fsp->fs_fsmnt, sizeof (fsp->fs_fsmnt) - 1, &len);
1117 	bzero(fsp->fs_fsmnt + len, sizeof (fsp->fs_fsmnt) - len);
1118 
1119 	/*
1120 	 * Sanity checks for old file systems
1121 	 */
1122 	if (fsp->fs_postblformat == FS_42POSTBLFMT)
1123 		ufsvfsp->vfs_nrpos = 8;
1124 	else
1125 		ufsvfsp->vfs_nrpos = fsp->fs_nrpos;
1126 
1127 	/*
1128 	 * Initialize lockfs structure to support file system locking
1129 	 */
1130 	bzero(&ufsvfsp->vfs_ulockfs.ul_lockfs,
1131 	    sizeof (struct lockfs));
1132 	ufsvfsp->vfs_ulockfs.ul_fs_lock = ULOCKFS_ULOCK;
1133 	mutex_init(&ufsvfsp->vfs_ulockfs.ul_lock, NULL,
1134 	    MUTEX_DEFAULT, NULL);
1135 	cv_init(&ufsvfsp->vfs_ulockfs.ul_cv, NULL, CV_DEFAULT, NULL);
1136 
1137 	/*
1138 	 * We don't need to grab vfs_dqrwlock for this ufs_iget() call.
1139 	 * We are in the process of mounting the file system so there
1140 	 * is no need to grab the quota lock. If a quota applies to the
1141 	 * root inode, then it will be updated when quotas are enabled.
1142 	 *
1143 	 * However, we have an ASSERT(RW_LOCK_HELD(&ufsvfsp->vfs_dqrwlock))
1144 	 * in getinoquota() that we want to keep so grab it anyway.
1145 	 */
1146 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
1147 
1148 	error = ufs_iget_alloced(vfsp, UFSROOTINO, &rip, cr);
1149 
1150 	rw_exit(&ufsvfsp->vfs_dqrwlock);
1151 
1152 	if (error)
1153 		goto out;
1154 
1155 	/*
1156 	 * make sure root inode is a directory.  Returning ENOTDIR might
1157 	 * be confused with the mount point not being a directory, so
1158 	 * we use EIO instead.
1159 	 */
1160 	if ((rip->i_mode & IFMT) != IFDIR) {
1161 		/*
1162 		 * Mark this inode as subject for cleanup
1163 		 * to avoid stray inodes in the cache.
1164 		 */
1165 		rvp = ITOV(rip);
1166 		error = EIO;
1167 		goto out;
1168 	}
1169 
1170 	rvp = ITOV(rip);
1171 	mutex_enter(&rvp->v_lock);
1172 	rvp->v_flag |= VROOT;
1173 	mutex_exit(&rvp->v_lock);
1174 	ufsvfsp->vfs_root = rvp;
1175 	/* The buffer for the root inode does not contain a valid b_vp */
1176 	(void) bfinval(dev, 0);
1177 
1178 	/* options */
1179 	ufsvfsp->vfs_nosetsec = flags & UFSMNT_NOSETSEC;
1180 	ufsvfsp->vfs_nointr  = flags & UFSMNT_NOINTR;
1181 	ufsvfsp->vfs_syncdir = flags & UFSMNT_SYNCDIR;
1182 	ufsvfsp->vfs_noatime = flags & UFSMNT_NOATIME;
1183 	if ((flags & UFSMNT_NODFRATIME) || ufsvfsp->vfs_noatime)
1184 		ufsvfsp->vfs_dfritime &= ~UFS_DFRATIME;
1185 	else	/* dfratime, default behavior */
1186 		ufsvfsp->vfs_dfritime |= UFS_DFRATIME;
1187 	if (flags & UFSMNT_FORCEDIRECTIO)
1188 		ufsvfsp->vfs_forcedirectio = 1;
1189 	else if (flags & UFSMNT_NOFORCEDIRECTIO)
1190 		ufsvfsp->vfs_forcedirectio = 0;
1191 	ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
1192 
1193 	ufsvfsp->vfs_nindiroffset = fsp->fs_nindir - 1;
1194 	ufsvfsp->vfs_nindirshift = highbit(ufsvfsp->vfs_nindiroffset);
1195 	ufsvfsp->vfs_ioclustsz = fsp->fs_bsize * fsp->fs_maxcontig;
1196 
1197 	if (cdev_ioctl(dev, DKIOCINFO, (intptr_t)&ci,
1198 	    FKIOCTL|FNATIVE|FREAD, CRED(), &status) == 0) {
1199 		ufsvfsp->vfs_iotransz = ci.dki_maxtransfer * DEV_BSIZE;
1200 	} else {
1201 		ufsvfsp->vfs_iotransz = MIN(maxphys, ufs_maxmaxphys);
1202 	}
1203 
1204 	if (ufsvfsp->vfs_iotransz <= 0) {
1205 		ufsvfsp->vfs_iotransz = MIN(maxphys, ufs_maxmaxphys);
1206 	}
1207 
1208 	/*
1209 	 * When logging, used to reserve log space for writes and truncs
1210 	 */
1211 	ufsvfsp->vfs_avgbfree = fsp->fs_cstotal.cs_nbfree / fsp->fs_ncg;
1212 
1213 	/*
1214 	 * Determine whether to log cylinder group summary info.
1215 	 */
1216 	ufsvfsp->vfs_nolog_si = (fsp->fs_ncg < ufs_ncg_log);
1217 
1218 	if (TRANS_ISTRANS(ufsvfsp)) {
1219 		/*
1220 		 * start the delete thread
1221 		 */
1222 		ufs_thread_start(&ufsvfsp->vfs_delete, ufs_thread_delete, vfsp);
1223 
1224 		/*
1225 		 * start reclaim thread if the filesystem was not mounted
1226 		 * read only.
1227 		 */
1228 		if (!fsp->fs_ronly && (fsp->fs_reclaim &
1229 		    (FS_RECLAIM|FS_RECLAIMING))) {
1230 			fsp->fs_reclaim &= ~FS_RECLAIM;
1231 			fsp->fs_reclaim |=  FS_RECLAIMING;
1232 			ufs_thread_start(&ufsvfsp->vfs_reclaim,
1233 			    ufs_thread_reclaim, vfsp);
1234 		}
1235 
1236 		/* Mark the fs as unrolled */
1237 		fsp->fs_rolled = FS_NEED_ROLL;
1238 	} else if (!fsp->fs_ronly && (fsp->fs_reclaim &
1239 	    (FS_RECLAIM|FS_RECLAIMING))) {
1240 		/*
1241 		 * If a file system that is mounted nologging, after
1242 		 * having previously been mounted logging, becomes
1243 		 * unmounted whilst the reclaim thread is in the throes
1244 		 * of reclaiming open/deleted inodes, a subsequent mount
1245 		 * of such a file system with logging disabled could lead
1246 		 * to inodes becoming lost.  So, start reclaim now, even
1247 		 * though logging was disabled for the previous mount, to
1248 		 * tidy things up.
1249 		 */
1250 		fsp->fs_reclaim &= ~FS_RECLAIM;
1251 		fsp->fs_reclaim |=  FS_RECLAIMING;
1252 		ufs_thread_start(&ufsvfsp->vfs_reclaim,
1253 		    ufs_thread_reclaim, vfsp);
1254 	}
1255 
1256 	if (!fsp->fs_ronly) {
1257 		TRANS_SBWRITE(ufsvfsp, TOP_MOUNT);
1258 		if (error = geterror(ufsvfsp->vfs_bufp))
1259 			goto out;
1260 	}
1261 
1262 	/* fix-on-panic initialization */
1263 	if (isroot && !(flags & UFSMNT_ONERROR_FLGMASK))
1264 		flags |= UFSMNT_ONERROR_PANIC;	/* XXX ..._RDONLY */
1265 
1266 	if ((error = ufsfx_mount(ufsvfsp, flags)) != 0)
1267 		goto out;
1268 
1269 	if (why == ROOT_INIT && isroot)
1270 		rootvp = devvp;
1271 
1272 	return (0);
1273 out:
1274 	if (error == 0)
1275 		error = EIO;
1276 	if (rvp) {
1277 		/* the following sequence is similar to ufs_unmount() */
1278 
1279 		/*
1280 		 * There's a problem that ufs_iget() puts inodes into
1281 		 * the inode cache before it returns them.  If someone
1282 		 * traverses that cache and gets a reference to our
1283 		 * inode, there's a chance they'll still be using it
1284 		 * after we've destroyed it.  This is a hard race to
1285 		 * hit, but it's happened (putting in a medium delay
1286 		 * here, and a large delay in ufs_scan_inodes() for
1287 		 * inodes on the device we're bailing out on, makes
1288 		 * the race easy to demonstrate).  The symptom is some
1289 		 * other part of UFS faulting on bad inode contents,
1290 		 * or when grabbing one of the locks inside the inode,
1291 		 * etc.  The usual victim is ufs_scan_inodes() or
1292 		 * someone called by it.
1293 		 */
1294 
1295 		/*
1296 		 * First, isolate it so that no new references can be
1297 		 * gotten via the inode cache.
1298 		 */
1299 		ihm = &ih_lock[INOHASH(UFSROOTINO)];
1300 		mutex_enter(ihm);
1301 		remque(rip);
1302 		mutex_exit(ihm);
1303 
1304 		/*
1305 		 * Now wait for all outstanding references except our
1306 		 * own to drain.  This could, in theory, take forever,
1307 		 * so don't wait *too* long.  If we time out, mark
1308 		 * it stale and leak it, so we don't hit the problem
1309 		 * described above.
1310 		 *
1311 		 * Note that v_count is an int, which means we can read
1312 		 * it in one operation.  Thus, there's no need to lock
1313 		 * around our tests.
1314 		 */
1315 		elapsed = 0;
1316 		while ((rvp->v_count > 1) && (elapsed < ufs_mount_timeout)) {
1317 			delay(ufs_mount_error_delay * drv_usectohz(1000));
1318 			elapsed += ufs_mount_error_delay;
1319 		}
1320 
1321 		if (rvp->v_count > 1) {
1322 			mutex_enter(&rip->i_tlock);
1323 			rip->i_flag |= ISTALE;
1324 			mutex_exit(&rip->i_tlock);
1325 			cmn_err(CE_WARN,
1326 			    "Timed out while cleaning up after "
1327 			    "failed mount of %s", path);
1328 		} else {
1329 
1330 			/*
1331 			 * Now we're the only one with a handle left, so tear
1332 			 * it down the rest of the way.
1333 			 */
1334 			if (ufs_rmidle(rip))
1335 				VN_RELE(rvp);
1336 			ufs_si_del(rip);
1337 			rip->i_ufsvfs = NULL;
1338 			rvp->v_vfsp = NULL;
1339 			rvp->v_type = VBAD;
1340 			VN_RELE(rvp);
1341 		}
1342 	}
1343 	if (needtrans) {
1344 		TRANS_MATA_UMOUNT(ufsvfsp);
1345 	}
1346 	if (ufsvfsp) {
1347 		ufs_vfs_remove(ufsvfsp);
1348 		ufs_thread_exit(&ufsvfsp->vfs_delete);
1349 		ufs_thread_exit(&ufsvfsp->vfs_reclaim);
1350 		mutex_destroy(&ufsvfsp->vfs_lock);
1351 		if (ufsvfsp->vfs_log) {
1352 			lufs_unsnarf(ufsvfsp);
1353 		}
1354 		kmem_free(ufsvfsp, sizeof (struct ufsvfs));
1355 	}
1356 	if (bp) {
1357 		bp->b_flags |= (B_STALE|B_AGE);
1358 		brelse(bp);
1359 	}
1360 	if (tp) {
1361 		tp->b_flags |= (B_STALE|B_AGE);
1362 		brelse(tp);
1363 	}
1364 	if (needclose) {
1365 		(void) VOP_CLOSE(devvp, (vfsp->vfs_flag & VFS_RDONLY) ?
1366 		    FREAD : FREAD|FWRITE, 1, (offset_t)0, cr, NULL);
1367 		bflush(dev);
1368 		(void) bfinval(dev, 1);
1369 	}
1370 	return (error);
1371 }
1372 
1373 /*
1374  * vfs operations
1375  */
1376 static int
1377 ufs_unmount(struct vfs *vfsp, int fflag, struct cred *cr)
1378 {
1379 	dev_t		dev		= vfsp->vfs_dev;
1380 	struct ufsvfs	*ufsvfsp	= (struct ufsvfs *)vfsp->vfs_data;
1381 	struct fs	*fs		= ufsvfsp->vfs_fs;
1382 	struct ulockfs	*ulp		= &ufsvfsp->vfs_ulockfs;
1383 	struct vnode	*bvp, *vp;
1384 	struct buf	*bp;
1385 	struct inode	*ip, *inext, *rip;
1386 	union ihead	*ih;
1387 	int		error, flag, i;
1388 	struct lockfs	lockfs;
1389 	int		poll_events = POLLPRI;
1390 	extern struct pollhead ufs_pollhd;
1391 	refstr_t	*mountpoint;
1392 
1393 	ASSERT(vfs_lock_held(vfsp));
1394 
1395 	if (secpolicy_fs_unmount(cr, vfsp) != 0)
1396 		return (EPERM);
1397 	/*
1398 	 * Forced unmount is now supported through the
1399 	 * lockfs protocol.
1400 	 */
1401 	if (fflag & MS_FORCE) {
1402 		/*
1403 		 * Mark the filesystem as being unmounted now in
1404 		 * case of a forcible umount before we take any
1405 		 * locks inside UFS to prevent racing with a VFS_VGET()
1406 		 * request. Throw these VFS_VGET() requests away for
1407 		 * the duration of the forcible umount so they won't
1408 		 * use stale or even freed data later on when we're done.
1409 		 * It may happen that the VFS has had a additional hold
1410 		 * placed on it by someone other than UFS and thus will
1411 		 * not get freed immediately once we're done with the
1412 		 * umount by dounmount() - use VFS_UNMOUNTED to inform
1413 		 * users of this still-alive VFS that its corresponding
1414 		 * filesystem being gone so they can detect that and error
1415 		 * out.
1416 		 */
1417 		vfsp->vfs_flag |= VFS_UNMOUNTED;
1418 
1419 		ufs_thread_suspend(&ufsvfsp->vfs_delete);
1420 		mutex_enter(&ulp->ul_lock);
1421 		/*
1422 		 * If file system is already hard locked,
1423 		 * unmount the file system, otherwise
1424 		 * hard lock it before unmounting.
1425 		 */
1426 		if (!ULOCKFS_IS_HLOCK(ulp)) {
1427 			atomic_inc_ulong(&ufs_quiesce_pend);
1428 			lockfs.lf_lock = LOCKFS_HLOCK;
1429 			lockfs.lf_flags = 0;
1430 			lockfs.lf_key = ulp->ul_lockfs.lf_key + 1;
1431 			lockfs.lf_comlen = 0;
1432 			lockfs.lf_comment = NULL;
1433 			ufs_freeze(ulp, &lockfs);
1434 			ULOCKFS_SET_BUSY(ulp);
1435 			LOCKFS_SET_BUSY(&ulp->ul_lockfs);
1436 			(void) ufs_quiesce(ulp);
1437 			(void) ufs_flush(vfsp);
1438 			(void) ufs_thaw(vfsp, ufsvfsp, ulp);
1439 			atomic_dec_ulong(&ufs_quiesce_pend);
1440 			ULOCKFS_CLR_BUSY(ulp);
1441 			LOCKFS_CLR_BUSY(&ulp->ul_lockfs);
1442 			poll_events |= POLLERR;
1443 			pollwakeup(&ufs_pollhd, poll_events);
1444 		}
1445 		ufs_thread_continue(&ufsvfsp->vfs_delete);
1446 		mutex_exit(&ulp->ul_lock);
1447 	}
1448 
1449 	/* let all types of writes go through */
1450 	ufsvfsp->vfs_iotstamp = ddi_get_lbolt();
1451 
1452 	/* coordinate with global hlock thread */
1453 	if (TRANS_ISTRANS(ufsvfsp) && (ufsvfsp->vfs_validfs == UT_HLOCKING)) {
1454 		/*
1455 		 * last possibility for a forced umount to fail hence clear
1456 		 * VFS_UNMOUNTED if appropriate.
1457 		 */
1458 		if (fflag & MS_FORCE)
1459 			vfsp->vfs_flag &= ~VFS_UNMOUNTED;
1460 		return (EAGAIN);
1461 	}
1462 
1463 	ufsvfsp->vfs_validfs = UT_UNMOUNTED;
1464 
1465 	/* kill the reclaim thread */
1466 	ufs_thread_exit(&ufsvfsp->vfs_reclaim);
1467 
1468 	/* suspend the delete thread */
1469 	ufs_thread_suspend(&ufsvfsp->vfs_delete);
1470 
1471 	/*
1472 	 * drain the delete and idle queues
1473 	 */
1474 	ufs_delete_drain(vfsp, -1, 1);
1475 	ufs_idle_drain(vfsp);
1476 
1477 	/*
1478 	 * use the lockfs protocol to prevent new ops from starting
1479 	 * a forcible umount can not fail beyond this point as
1480 	 * we hard-locked the filesystem and drained all current consumers
1481 	 * before.
1482 	 */
1483 	mutex_enter(&ulp->ul_lock);
1484 
1485 	/*
1486 	 * if the file system is busy; return EBUSY
1487 	 */
1488 	if (ulp->ul_vnops_cnt || ulp->ul_falloc_cnt || ULOCKFS_IS_SLOCK(ulp)) {
1489 		error = EBUSY;
1490 		goto out;
1491 	}
1492 
1493 	/*
1494 	 * if this is not a forced unmount (!hard/error locked), then
1495 	 * get rid of every inode except the root and quota inodes
1496 	 * also, commit any outstanding transactions
1497 	 */
1498 	if (!ULOCKFS_IS_HLOCK(ulp) && !ULOCKFS_IS_ELOCK(ulp))
1499 		if (error = ufs_flush(vfsp))
1500 			goto out;
1501 
1502 	/*
1503 	 * ignore inodes in the cache if fs is hard locked or error locked
1504 	 */
1505 	rip = VTOI(ufsvfsp->vfs_root);
1506 	if (!ULOCKFS_IS_HLOCK(ulp) && !ULOCKFS_IS_ELOCK(ulp)) {
1507 		/*
1508 		 * Otherwise, only the quota and root inodes are in the cache.
1509 		 *
1510 		 * Avoid racing with ufs_update() and ufs_sync().
1511 		 */
1512 		mutex_enter(&ufs_scan_lock);
1513 
1514 		for (i = 0, ih = ihead; i < inohsz; i++, ih++) {
1515 			mutex_enter(&ih_lock[i]);
1516 			for (ip = ih->ih_chain[0];
1517 			    ip != (struct inode *)ih;
1518 			    ip = ip->i_forw) {
1519 				if (ip->i_ufsvfs != ufsvfsp)
1520 					continue;
1521 				if (ip == ufsvfsp->vfs_qinod)
1522 					continue;
1523 				if (ip == rip && ITOV(ip)->v_count == 1)
1524 					continue;
1525 				mutex_exit(&ih_lock[i]);
1526 				mutex_exit(&ufs_scan_lock);
1527 				error = EBUSY;
1528 				goto out;
1529 			}
1530 			mutex_exit(&ih_lock[i]);
1531 		}
1532 		mutex_exit(&ufs_scan_lock);
1533 	}
1534 
1535 	/*
1536 	 * if a snapshot exists and this is a forced unmount, then delete
1537 	 * the snapshot.  Otherwise return EBUSY.  This will insure the
1538 	 * snapshot always belongs to a valid file system.
1539 	 */
1540 	if (ufsvfsp->vfs_snapshot) {
1541 		if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) {
1542 			(void) fssnap_delete(&ufsvfsp->vfs_snapshot);
1543 		} else {
1544 			error = EBUSY;
1545 			goto out;
1546 		}
1547 	}
1548 
1549 	/*
1550 	 * Close the quota file and invalidate anything left in the quota
1551 	 * cache for this file system.  Pass kcred to allow all quota
1552 	 * manipulations.
1553 	 */
1554 	(void) closedq(ufsvfsp, kcred);
1555 	invalidatedq(ufsvfsp);
1556 	/*
1557 	 * drain the delete and idle queues
1558 	 */
1559 	ufs_delete_drain(vfsp, -1, 0);
1560 	ufs_idle_drain(vfsp);
1561 
1562 	/*
1563 	 * discard the inodes for this fs (including root, shadow, and quota)
1564 	 */
1565 	for (i = 0, ih = ihead; i < inohsz; i++, ih++) {
1566 		mutex_enter(&ih_lock[i]);
1567 		for (inext = 0, ip = ih->ih_chain[0];
1568 		    ip != (struct inode *)ih;
1569 		    ip = inext) {
1570 			inext = ip->i_forw;
1571 			if (ip->i_ufsvfs != ufsvfsp)
1572 				continue;
1573 
1574 			/*
1575 			 * We've found the inode in the cache and as we
1576 			 * hold the hash mutex the inode can not
1577 			 * disappear from underneath us.
1578 			 * We also know it must have at least a vnode
1579 			 * reference count of 1.
1580 			 * We perform an additional VN_HOLD so the VN_RELE
1581 			 * in case we take the inode off the idle queue
1582 			 * can not be the last one.
1583 			 * It is safe to grab the writer contents lock here
1584 			 * to prevent a race with ufs_iinactive() putting
1585 			 * inodes into the idle queue while we operate on
1586 			 * this inode.
1587 			 */
1588 			rw_enter(&ip->i_contents, RW_WRITER);
1589 
1590 			vp = ITOV(ip);
1591 			VN_HOLD(vp)
1592 			remque(ip);
1593 			if (ufs_rmidle(ip))
1594 				VN_RELE(vp);
1595 			ufs_si_del(ip);
1596 			/*
1597 			 * rip->i_ufsvfsp is needed by bflush()
1598 			 */
1599 			if (ip != rip)
1600 				ip->i_ufsvfs = NULL;
1601 			/*
1602 			 * Set vnode's vfsops to dummy ops, which return
1603 			 * EIO. This is needed to forced unmounts to work
1604 			 * with lofs/nfs properly.
1605 			 */
1606 			if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp))
1607 				vp->v_vfsp = &EIO_vfs;
1608 			else
1609 				vp->v_vfsp = NULL;
1610 			vp->v_type = VBAD;
1611 
1612 			rw_exit(&ip->i_contents);
1613 
1614 			VN_RELE(vp);
1615 		}
1616 		mutex_exit(&ih_lock[i]);
1617 	}
1618 	ufs_si_cache_flush(dev);
1619 
1620 	/*
1621 	 * kill the delete thread and drain the idle queue
1622 	 */
1623 	ufs_thread_exit(&ufsvfsp->vfs_delete);
1624 	ufs_idle_drain(vfsp);
1625 
1626 	bp = ufsvfsp->vfs_bufp;
1627 	bvp = ufsvfsp->vfs_devvp;
1628 	flag = !fs->fs_ronly;
1629 	if (flag) {
1630 		bflush(dev);
1631 		if (fs->fs_clean != FSBAD) {
1632 			if (fs->fs_clean == FSSTABLE)
1633 				fs->fs_clean = FSCLEAN;
1634 			fs->fs_reclaim &= ~FS_RECLAIM;
1635 		}
1636 		if (TRANS_ISTRANS(ufsvfsp) &&
1637 		    !TRANS_ISERROR(ufsvfsp) &&
1638 		    !ULOCKFS_IS_HLOCK(ulp) &&
1639 		    (fs->fs_rolled == FS_NEED_ROLL)) {
1640 			/*
1641 			 * ufs_flush() above has flushed the last Moby.
1642 			 * This is needed to ensure the following superblock
1643 			 * update really is the last metadata update
1644 			 */
1645 			error = ufs_putsummaryinfo(dev, ufsvfsp, fs);
1646 			if (error == 0) {
1647 				fs->fs_rolled = FS_ALL_ROLLED;
1648 			}
1649 		}
1650 		TRANS_SBUPDATE(ufsvfsp, vfsp, TOP_SBUPDATE_UNMOUNT);
1651 		/*
1652 		 * push this last transaction
1653 		 */
1654 		curthread->t_flag |= T_DONTBLOCK;
1655 		TRANS_BEGIN_SYNC(ufsvfsp, TOP_COMMIT_UNMOUNT, TOP_COMMIT_SIZE,
1656 		    error);
1657 		if (!error)
1658 			TRANS_END_SYNC(ufsvfsp, error, TOP_COMMIT_UNMOUNT,
1659 			    TOP_COMMIT_SIZE);
1660 		curthread->t_flag &= ~T_DONTBLOCK;
1661 	}
1662 
1663 	TRANS_MATA_UMOUNT(ufsvfsp);
1664 	lufs_unsnarf(ufsvfsp);		/* Release the in-memory structs */
1665 	ufsfx_unmount(ufsvfsp);		/* fix-on-panic bookkeeping */
1666 	kmem_free(fs->fs_u.fs_csp, fs->fs_cssize);
1667 
1668 	bp->b_flags |= B_STALE|B_AGE;
1669 	ufsvfsp->vfs_bufp = NULL;	/* don't point at freed buf */
1670 	brelse(bp);			/* free the superblock buf */
1671 
1672 	(void) VOP_PUTPAGE(common_specvp(bvp), (offset_t)0, (size_t)0,
1673 	    B_INVAL, cr, NULL);
1674 	(void) VOP_CLOSE(bvp, flag, 1, (offset_t)0, cr, NULL);
1675 	bflush(dev);
1676 	(void) bfinval(dev, 1);
1677 	VN_RELE(bvp);
1678 
1679 	/*
1680 	 * It is now safe to NULL out the ufsvfs pointer and discard
1681 	 * the root inode.
1682 	 */
1683 	rip->i_ufsvfs = NULL;
1684 	VN_RELE(ITOV(rip));
1685 
1686 	/* free up lockfs comment structure, if any */
1687 	if (ulp->ul_lockfs.lf_comlen && ulp->ul_lockfs.lf_comment)
1688 		kmem_free(ulp->ul_lockfs.lf_comment, ulp->ul_lockfs.lf_comlen);
1689 
1690 	/*
1691 	 * Remove from instance list.
1692 	 */
1693 	ufs_vfs_remove(ufsvfsp);
1694 
1695 	/*
1696 	 * For a forcible unmount, threads may be asleep in
1697 	 * ufs_lockfs_begin/ufs_check_lockfs.  These threads will need
1698 	 * the ufsvfs structure so we don't free it, yet.  ufs_update
1699 	 * will free it up after awhile.
1700 	 */
1701 	if (ULOCKFS_IS_HLOCK(ulp) || ULOCKFS_IS_ELOCK(ulp)) {
1702 		extern kmutex_t		ufsvfs_mutex;
1703 		extern struct ufsvfs	*ufsvfslist;
1704 
1705 		mutex_enter(&ufsvfs_mutex);
1706 		ufsvfsp->vfs_dontblock = 1;
1707 		ufsvfsp->vfs_next = ufsvfslist;
1708 		ufsvfslist = ufsvfsp;
1709 		mutex_exit(&ufsvfs_mutex);
1710 		/* wakeup any suspended threads */
1711 		cv_broadcast(&ulp->ul_cv);
1712 		mutex_exit(&ulp->ul_lock);
1713 	} else {
1714 		mutex_destroy(&ufsvfsp->vfs_lock);
1715 		kmem_free(ufsvfsp, sizeof (struct ufsvfs));
1716 	}
1717 
1718 	/*
1719 	 * Now mark the filesystem as unmounted since we're done with it.
1720 	 */
1721 	vfsp->vfs_flag |= VFS_UNMOUNTED;
1722 
1723 	return (0);
1724 out:
1725 	/* open the fs to new ops */
1726 	cv_broadcast(&ulp->ul_cv);
1727 	mutex_exit(&ulp->ul_lock);
1728 
1729 	if (TRANS_ISTRANS(ufsvfsp)) {
1730 		/* allow the delete thread to continue */
1731 		ufs_thread_continue(&ufsvfsp->vfs_delete);
1732 		/* restart the reclaim thread */
1733 		ufs_thread_start(&ufsvfsp->vfs_reclaim, ufs_thread_reclaim,
1734 		    vfsp);
1735 		/* coordinate with global hlock thread */
1736 		ufsvfsp->vfs_validfs = UT_MOUNTED;
1737 		/* check for trans errors during umount */
1738 		ufs_trans_onerror();
1739 
1740 		/*
1741 		 * if we have a separate /usr it will never unmount
1742 		 * when halting. In order to not re-read all the
1743 		 * cylinder group summary info on mounting after
1744 		 * reboot the logging of summary info is re-enabled
1745 		 * and the super block written out.
1746 		 */
1747 		mountpoint = vfs_getmntpoint(vfsp);
1748 		if ((fs->fs_si == FS_SI_OK) &&
1749 		    (strcmp("/usr", refstr_value(mountpoint)) == 0)) {
1750 			ufsvfsp->vfs_nolog_si = 0;
1751 			UFS_BWRITE2(NULL, ufsvfsp->vfs_bufp);
1752 		}
1753 		refstr_rele(mountpoint);
1754 	}
1755 
1756 	return (error);
1757 }
1758 
1759 static int
1760 ufs_root(struct vfs *vfsp, struct vnode **vpp)
1761 {
1762 	struct ufsvfs *ufsvfsp;
1763 	struct vnode *vp;
1764 
1765 	if (!vfsp)
1766 		return (EIO);
1767 
1768 	ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1769 	if (!ufsvfsp || !ufsvfsp->vfs_root)
1770 		return (EIO);	/* forced unmount */
1771 
1772 	vp = ufsvfsp->vfs_root;
1773 	VN_HOLD(vp);
1774 	*vpp = vp;
1775 	return (0);
1776 }
1777 
1778 /*
1779  * Get file system statistics.
1780  */
1781 static int
1782 ufs_statvfs(struct vfs *vfsp, struct statvfs64 *sp)
1783 {
1784 	struct fs *fsp;
1785 	struct ufsvfs *ufsvfsp;
1786 	int blk, i;
1787 	long max_avail, used;
1788 	dev32_t d32;
1789 
1790 	if (vfsp->vfs_flag & VFS_UNMOUNTED)
1791 		return (EIO);
1792 
1793 	ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1794 	fsp = ufsvfsp->vfs_fs;
1795 	if ((fsp->fs_magic != FS_MAGIC) && (fsp->fs_magic != MTB_UFS_MAGIC))
1796 		return (EINVAL);
1797 	if (fsp->fs_magic == FS_MAGIC &&
1798 	    (fsp->fs_version != UFS_EFISTYLE4NONEFI_VERSION_2 &&
1799 	    fsp->fs_version != UFS_VERSION_MIN))
1800 		return (EINVAL);
1801 	if (fsp->fs_magic == MTB_UFS_MAGIC &&
1802 	    (fsp->fs_version > MTB_UFS_VERSION_1 ||
1803 	    fsp->fs_version < MTB_UFS_VERSION_MIN))
1804 		return (EINVAL);
1805 
1806 	/*
1807 	 * get the basic numbers
1808 	 */
1809 	(void) bzero(sp, sizeof (*sp));
1810 
1811 	sp->f_bsize = fsp->fs_bsize;
1812 	sp->f_frsize = fsp->fs_fsize;
1813 	sp->f_blocks = (fsblkcnt64_t)fsp->fs_dsize;
1814 	sp->f_bfree = (fsblkcnt64_t)fsp->fs_cstotal.cs_nbfree * fsp->fs_frag +
1815 	    fsp->fs_cstotal.cs_nffree;
1816 
1817 	sp->f_files = (fsfilcnt64_t)fsp->fs_ncg * fsp->fs_ipg;
1818 	sp->f_ffree = (fsfilcnt64_t)fsp->fs_cstotal.cs_nifree;
1819 
1820 	/*
1821 	 * Adjust the numbers based on things waiting to be deleted.
1822 	 * modifies f_bfree and f_ffree.  Afterwards, everything we
1823 	 * come up with will be self-consistent.  By definition, this
1824 	 * is a point-in-time snapshot, so the fact that the delete
1825 	 * thread's probably already invalidated the results is not a
1826 	 * problem.  Note that if the delete thread is ever extended to
1827 	 * non-logging ufs, this adjustment must always be made.
1828 	 */
1829 	if (TRANS_ISTRANS(ufsvfsp))
1830 		ufs_delete_adjust_stats(ufsvfsp, sp);
1831 
1832 	/*
1833 	 * avail = MAX(max_avail - used, 0)
1834 	 */
1835 	max_avail = fsp->fs_dsize - ufsvfsp->vfs_minfrags;
1836 
1837 	used = (fsp->fs_dsize - sp->f_bfree);
1838 
1839 	if (max_avail > used)
1840 		sp->f_bavail = (fsblkcnt64_t)max_avail - used;
1841 	else
1842 		sp->f_bavail = (fsblkcnt64_t)0;
1843 
1844 	sp->f_favail = sp->f_ffree;
1845 	(void) cmpldev(&d32, vfsp->vfs_dev);
1846 	sp->f_fsid = d32;
1847 	(void) strcpy(sp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1848 	sp->f_flag = vf_to_stf(vfsp->vfs_flag);
1849 
1850 	/* keep coordinated with ufs_l_pathconf() */
1851 	sp->f_namemax = MAXNAMLEN;
1852 
1853 	if (fsp->fs_cpc == 0) {
1854 		bzero(sp->f_fstr, 14);
1855 		return (0);
1856 	}
1857 	blk = fsp->fs_spc * fsp->fs_cpc / NSPF(fsp);
1858 	for (i = 0; i < blk; i += fsp->fs_frag) /* CSTYLED */
1859 		/* void */;
1860 	i -= fsp->fs_frag;
1861 	blk = i / fsp->fs_frag;
1862 	bcopy(&(fs_rotbl(fsp)[blk]), sp->f_fstr, 14);
1863 	return (0);
1864 }
1865 
1866 /*
1867  * Flush any pending I/O to file system vfsp.
1868  * The ufs_update() routine will only flush *all* ufs files.
1869  * If vfsp is non-NULL, only sync this ufs (in preparation
1870  * for a umount).
1871  */
1872 /*ARGSUSED*/
1873 static int
1874 ufs_sync(struct vfs *vfsp, short flag, struct cred *cr)
1875 {
1876 	struct ufsvfs *ufsvfsp;
1877 	struct fs *fs;
1878 	int cheap = flag & SYNC_ATTR;
1879 	int error;
1880 
1881 	/*
1882 	 * SYNC_CLOSE means we're rebooting.  Toss everything
1883 	 * on the idle queue so we don't have to slog through
1884 	 * a bunch of uninteresting inodes over and over again.
1885 	 */
1886 	if (flag & SYNC_CLOSE)
1887 		ufs_idle_drain(NULL);
1888 
1889 	if (vfsp == NULL) {
1890 		ufs_update(flag);
1891 		return (0);
1892 	}
1893 
1894 	/* Flush a single ufs */
1895 	if (!vfs_matchops(vfsp, ufs_vfsops) || vfs_lock(vfsp) != 0)
1896 		return (0);
1897 
1898 	ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1899 	if (!ufsvfsp)
1900 		return (EIO);
1901 	fs = ufsvfsp->vfs_fs;
1902 	mutex_enter(&ufsvfsp->vfs_lock);
1903 
1904 	if (ufsvfsp->vfs_dio &&
1905 	    fs->fs_ronly == 0 &&
1906 	    fs->fs_clean != FSBAD &&
1907 	    fs->fs_clean != FSLOG) {
1908 		/* turn off fast-io on unmount, so no fsck needed (4029401) */
1909 		ufsvfsp->vfs_dio = 0;
1910 		fs->fs_clean = FSACTIVE;
1911 		fs->fs_fmod = 1;
1912 	}
1913 
1914 	/* Write back modified superblock */
1915 	if (fs->fs_fmod == 0) {
1916 		mutex_exit(&ufsvfsp->vfs_lock);
1917 	} else {
1918 		if (fs->fs_ronly != 0) {
1919 			mutex_exit(&ufsvfsp->vfs_lock);
1920 			vfs_unlock(vfsp);
1921 			return (ufs_fault(ufsvfsp->vfs_root,
1922 			    "fs = %s update: ro fs mod\n", fs->fs_fsmnt));
1923 		}
1924 		fs->fs_fmod = 0;
1925 		mutex_exit(&ufsvfsp->vfs_lock);
1926 
1927 		TRANS_SBUPDATE(ufsvfsp, vfsp, TOP_SBUPDATE_UPDATE);
1928 	}
1929 	vfs_unlock(vfsp);
1930 
1931 	/*
1932 	 * Avoid racing with ufs_update() and ufs_unmount().
1933 	 *
1934 	 */
1935 	mutex_enter(&ufs_scan_lock);
1936 
1937 	(void) ufs_scan_inodes(1, ufs_sync_inode,
1938 	    (void *)(uintptr_t)cheap, ufsvfsp);
1939 
1940 	mutex_exit(&ufs_scan_lock);
1941 
1942 	bflush((dev_t)vfsp->vfs_dev);
1943 
1944 	/*
1945 	 * commit any outstanding async transactions
1946 	 */
1947 	curthread->t_flag |= T_DONTBLOCK;
1948 	TRANS_BEGIN_SYNC(ufsvfsp, TOP_COMMIT_UPDATE, TOP_COMMIT_SIZE, error);
1949 	if (!error) {
1950 		TRANS_END_SYNC(ufsvfsp, error, TOP_COMMIT_UPDATE,
1951 		    TOP_COMMIT_SIZE);
1952 	}
1953 	curthread->t_flag &= ~T_DONTBLOCK;
1954 
1955 	return (0);
1956 }
1957 
1958 
1959 void
1960 sbupdate(struct vfs *vfsp)
1961 {
1962 	struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
1963 	struct fs *fs = ufsvfsp->vfs_fs;
1964 	struct buf *bp;
1965 	int blks;
1966 	caddr_t space;
1967 	int i;
1968 	size_t size;
1969 
1970 	/*
1971 	 * for ulockfs processing, limit the superblock writes
1972 	 */
1973 	if ((ufsvfsp->vfs_ulockfs.ul_sbowner) &&
1974 	    (curthread != ufsvfsp->vfs_ulockfs.ul_sbowner)) {
1975 		/* process later */
1976 		fs->fs_fmod = 1;
1977 		return;
1978 	}
1979 	ULOCKFS_SET_MOD((&ufsvfsp->vfs_ulockfs));
1980 
1981 	if (TRANS_ISTRANS(ufsvfsp)) {
1982 		mutex_enter(&ufsvfsp->vfs_lock);
1983 		ufs_sbwrite(ufsvfsp);
1984 		mutex_exit(&ufsvfsp->vfs_lock);
1985 		return;
1986 	}
1987 
1988 	blks = howmany(fs->fs_cssize, fs->fs_fsize);
1989 	space = (caddr_t)fs->fs_u.fs_csp;
1990 	for (i = 0; i < blks; i += fs->fs_frag) {
1991 		size = fs->fs_bsize;
1992 		if (i + fs->fs_frag > blks)
1993 			size = (blks - i) * fs->fs_fsize;
1994 		bp = UFS_GETBLK(ufsvfsp, ufsvfsp->vfs_dev,
1995 		    (daddr_t)(fsbtodb(fs, fs->fs_csaddr + i)),
1996 		    fs->fs_bsize);
1997 		bcopy(space, bp->b_un.b_addr, size);
1998 		space += size;
1999 		bp->b_bcount = size;
2000 		UFS_BRWRITE(ufsvfsp, bp);
2001 	}
2002 	mutex_enter(&ufsvfsp->vfs_lock);
2003 	ufs_sbwrite(ufsvfsp);
2004 	mutex_exit(&ufsvfsp->vfs_lock);
2005 }
2006 
2007 int ufs_vget_idle_count = 2;	/* Number of inodes to idle each time */
2008 static int
2009 ufs_vget(struct vfs *vfsp, struct vnode **vpp, struct fid *fidp)
2010 {
2011 	int error = 0;
2012 	struct ufid *ufid;
2013 	struct inode *ip;
2014 	struct ufsvfs *ufsvfsp = (struct ufsvfs *)vfsp->vfs_data;
2015 	struct ulockfs *ulp;
2016 
2017 	/*
2018 	 * Check for unmounted filesystem.
2019 	 */
2020 	if (vfsp->vfs_flag & VFS_UNMOUNTED) {
2021 		error = EIO;
2022 		goto errout;
2023 	}
2024 
2025 	/*
2026 	 * Keep the idle queue from getting too long by
2027 	 * idling an inode before attempting to allocate another.
2028 	 *    This operation must be performed before entering
2029 	 *    lockfs or a transaction.
2030 	 */
2031 	if (ufs_idle_q.uq_ne > ufs_idle_q.uq_hiwat)
2032 		if ((curthread->t_flag & T_DONTBLOCK) == 0) {
2033 			ins.in_vidles.value.ul += ufs_vget_idle_count;
2034 			ufs_idle_some(ufs_vget_idle_count);
2035 		}
2036 
2037 	ufid = (struct ufid *)fidp;
2038 
2039 	if (error = ufs_lockfs_begin(ufsvfsp, &ulp, ULOCKFS_VGET_MASK))
2040 		goto errout;
2041 
2042 	rw_enter(&ufsvfsp->vfs_dqrwlock, RW_READER);
2043 
2044 	error = ufs_iget(vfsp, ufid->ufid_ino, &ip, CRED());
2045 
2046 	rw_exit(&ufsvfsp->vfs_dqrwlock);
2047 
2048 	ufs_lockfs_end(ulp);
2049 
2050 	if (error)
2051 		goto errout;
2052 
2053 	/*
2054 	 * Check if the inode has been deleted or freed or is in transient state
2055 	 * since the last VFS_VGET() request for it, release it and don't return
2056 	 * it to the caller, presumably NFS, as it's no longer valid.
2057 	 */
2058 	if (ip->i_gen != ufid->ufid_gen || ip->i_mode == 0 ||
2059 	    (ip->i_nlink <= 0)) {
2060 		VN_RELE(ITOV(ip));
2061 		error = EINVAL;
2062 		goto errout;
2063 	}
2064 
2065 	*vpp = ITOV(ip);
2066 	return (0);
2067 
2068 errout:
2069 	*vpp = NULL;
2070 	return (error);
2071 }
2072 
2073 static int
2074 ufs_syncfs(vfs_t *vfsp, uint64_t flags, cred_t *cr)
2075 {
2076 	if (flags != 0) {
2077 		return (ENOTSUP);
2078 	}
2079 
2080 	return (ufs_fioffs(vfsp, cr));
2081 }
2082 
2083 static int
2084 ufsinit(int fstype, char *name)
2085 {
2086 	static const fs_operation_def_t ufs_vfsops_template[] = {
2087 		VFSNAME_MOUNT,		{ .vfs_mount = ufs_mount },
2088 		VFSNAME_UNMOUNT,	{ .vfs_unmount = ufs_unmount },
2089 		VFSNAME_ROOT,		{ .vfs_root = ufs_root },
2090 		VFSNAME_STATVFS,	{ .vfs_statvfs = ufs_statvfs },
2091 		VFSNAME_SYNC,		{ .vfs_sync = ufs_sync },
2092 		VFSNAME_VGET,		{ .vfs_vget = ufs_vget },
2093 		VFSNAME_MOUNTROOT,	{ .vfs_mountroot = ufs_mountroot },
2094 		VFSNAME_SYNCFS,		{ .vfs_syncfs = ufs_syncfs },
2095 		NULL,			NULL
2096 	};
2097 	int error;
2098 
2099 	ufsfstype = fstype;
2100 
2101 	error = vfs_setfsops(fstype, ufs_vfsops_template, &ufs_vfsops);
2102 	if (error != 0) {
2103 		cmn_err(CE_WARN, "ufsinit: bad vfs ops template");
2104 		return (error);
2105 	}
2106 
2107 	error = vn_make_ops(name, ufs_vnodeops_template, &ufs_vnodeops);
2108 	if (error != 0) {
2109 		(void) vfs_freevfsops_by_type(fstype);
2110 		cmn_err(CE_WARN, "ufsinit: bad vnode ops template");
2111 		return (error);
2112 	}
2113 
2114 	ufs_iinit();
2115 	return (0);
2116 }
2117