xref: /titanic_51/usr/src/uts/common/fs/zfs/zfs_vfsops.c (revision dd51520e127b452179a2ce4ea3bd8dee949f9afe)
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  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 /* Portions Copyright 2010 Robert Milkowski */
26 
27 #include <sys/types.h>
28 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/sysmacros.h>
31 #include <sys/kmem.h>
32 #include <sys/pathname.h>
33 #include <sys/vnode.h>
34 #include <sys/vfs.h>
35 #include <sys/vfs_opreg.h>
36 #include <sys/mntent.h>
37 #include <sys/mount.h>
38 #include <sys/cmn_err.h>
39 #include "fs/fs_subr.h"
40 #include <sys/zfs_znode.h>
41 #include <sys/zfs_dir.h>
42 #include <sys/zil.h>
43 #include <sys/fs/zfs.h>
44 #include <sys/dmu.h>
45 #include <sys/dsl_prop.h>
46 #include <sys/dsl_dataset.h>
47 #include <sys/dsl_deleg.h>
48 #include <sys/spa.h>
49 #include <sys/zap.h>
50 #include <sys/sa.h>
51 #include <sys/varargs.h>
52 #include <sys/policy.h>
53 #include <sys/atomic.h>
54 #include <sys/mkdev.h>
55 #include <sys/modctl.h>
56 #include <sys/refstr.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/zfs_ctldir.h>
59 #include <sys/zfs_fuid.h>
60 #include <sys/bootconf.h>
61 #include <sys/sunddi.h>
62 #include <sys/dnlc.h>
63 #include <sys/dmu_objset.h>
64 #include <sys/spa_boot.h>
65 #include <sys/sa.h>
66 #include "zfs_comutil.h"
67 
68 int zfsfstype;
69 vfsops_t *zfs_vfsops = NULL;
70 static major_t zfs_major;
71 static minor_t zfs_minor;
72 static kmutex_t	zfs_dev_mtx;
73 
74 extern int sys_shutdown;
75 
76 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
77 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
78 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
79 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
80 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
81 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
82 static void zfs_freevfs(vfs_t *vfsp);
83 
84 static const fs_operation_def_t zfs_vfsops_template[] = {
85 	VFSNAME_MOUNT,		{ .vfs_mount = zfs_mount },
86 	VFSNAME_MOUNTROOT,	{ .vfs_mountroot = zfs_mountroot },
87 	VFSNAME_UNMOUNT,	{ .vfs_unmount = zfs_umount },
88 	VFSNAME_ROOT,		{ .vfs_root = zfs_root },
89 	VFSNAME_STATVFS,	{ .vfs_statvfs = zfs_statvfs },
90 	VFSNAME_SYNC,		{ .vfs_sync = zfs_sync },
91 	VFSNAME_VGET,		{ .vfs_vget = zfs_vget },
92 	VFSNAME_FREEVFS,	{ .vfs_freevfs = zfs_freevfs },
93 	NULL,			NULL
94 };
95 
96 static const fs_operation_def_t zfs_vfsops_eio_template[] = {
97 	VFSNAME_FREEVFS,	{ .vfs_freevfs =  zfs_freevfs },
98 	NULL,			NULL
99 };
100 
101 /*
102  * We need to keep a count of active fs's.
103  * This is necessary to prevent our module
104  * from being unloaded after a umount -f
105  */
106 static uint32_t	zfs_active_fs_count = 0;
107 
108 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
109 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
110 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
111 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
112 
113 /*
114  * MO_DEFAULT is not used since the default value is determined
115  * by the equivalent property.
116  */
117 static mntopt_t mntopts[] = {
118 	{ MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
119 	{ MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
120 	{ MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
121 	{ MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
122 };
123 
124 static mntopts_t zfs_mntopts = {
125 	sizeof (mntopts) / sizeof (mntopt_t),
126 	mntopts
127 };
128 
129 /*ARGSUSED*/
130 int
131 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
132 {
133 	/*
134 	 * Data integrity is job one.  We don't want a compromised kernel
135 	 * writing to the storage pool, so we never sync during panic.
136 	 */
137 	if (panicstr)
138 		return (0);
139 
140 	/*
141 	 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
142 	 * to sync metadata, which they would otherwise cache indefinitely.
143 	 * Semantically, the only requirement is that the sync be initiated.
144 	 * The DMU syncs out txgs frequently, so there's nothing to do.
145 	 */
146 	if (flag & SYNC_ATTR)
147 		return (0);
148 
149 	if (vfsp != NULL) {
150 		/*
151 		 * Sync a specific filesystem.
152 		 */
153 		zfsvfs_t *zfsvfs = vfsp->vfs_data;
154 		dsl_pool_t *dp;
155 
156 		ZFS_ENTER(zfsvfs);
157 		dp = dmu_objset_pool(zfsvfs->z_os);
158 
159 		/*
160 		 * If the system is shutting down, then skip any
161 		 * filesystems which may exist on a suspended pool.
162 		 */
163 		if (sys_shutdown && spa_suspended(dp->dp_spa)) {
164 			ZFS_EXIT(zfsvfs);
165 			return (0);
166 		}
167 
168 		if (zfsvfs->z_log != NULL)
169 			zil_commit(zfsvfs->z_log, 0);
170 
171 		ZFS_EXIT(zfsvfs);
172 	} else {
173 		/*
174 		 * Sync all ZFS filesystems.  This is what happens when you
175 		 * run sync(1M).  Unlike other filesystems, ZFS honors the
176 		 * request by waiting for all pools to commit all dirty data.
177 		 */
178 		spa_sync_allpools();
179 	}
180 
181 	return (0);
182 }
183 
184 static int
185 zfs_create_unique_device(dev_t *dev)
186 {
187 	major_t new_major;
188 
189 	do {
190 		ASSERT3U(zfs_minor, <=, MAXMIN32);
191 		minor_t start = zfs_minor;
192 		do {
193 			mutex_enter(&zfs_dev_mtx);
194 			if (zfs_minor >= MAXMIN32) {
195 				/*
196 				 * If we're still using the real major
197 				 * keep out of /dev/zfs and /dev/zvol minor
198 				 * number space.  If we're using a getudev()'ed
199 				 * major number, we can use all of its minors.
200 				 */
201 				if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
202 					zfs_minor = ZFS_MIN_MINOR;
203 				else
204 					zfs_minor = 0;
205 			} else {
206 				zfs_minor++;
207 			}
208 			*dev = makedevice(zfs_major, zfs_minor);
209 			mutex_exit(&zfs_dev_mtx);
210 		} while (vfs_devismounted(*dev) && zfs_minor != start);
211 		if (zfs_minor == start) {
212 			/*
213 			 * We are using all ~262,000 minor numbers for the
214 			 * current major number.  Create a new major number.
215 			 */
216 			if ((new_major = getudev()) == (major_t)-1) {
217 				cmn_err(CE_WARN,
218 				    "zfs_mount: Can't get unique major "
219 				    "device number.");
220 				return (-1);
221 			}
222 			mutex_enter(&zfs_dev_mtx);
223 			zfs_major = new_major;
224 			zfs_minor = 0;
225 
226 			mutex_exit(&zfs_dev_mtx);
227 		} else {
228 			break;
229 		}
230 		/* CONSTANTCONDITION */
231 	} while (1);
232 
233 	return (0);
234 }
235 
236 static void
237 atime_changed_cb(void *arg, uint64_t newval)
238 {
239 	zfsvfs_t *zfsvfs = arg;
240 
241 	if (newval == TRUE) {
242 		zfsvfs->z_atime = TRUE;
243 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
244 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
245 	} else {
246 		zfsvfs->z_atime = FALSE;
247 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
248 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
249 	}
250 }
251 
252 static void
253 xattr_changed_cb(void *arg, uint64_t newval)
254 {
255 	zfsvfs_t *zfsvfs = arg;
256 
257 	if (newval == TRUE) {
258 		/* XXX locking on vfs_flag? */
259 		zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
260 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
261 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
262 	} else {
263 		/* XXX locking on vfs_flag? */
264 		zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
265 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
266 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
267 	}
268 }
269 
270 static void
271 blksz_changed_cb(void *arg, uint64_t newval)
272 {
273 	zfsvfs_t *zfsvfs = arg;
274 
275 	if (newval < SPA_MINBLOCKSIZE ||
276 	    newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
277 		newval = SPA_MAXBLOCKSIZE;
278 
279 	zfsvfs->z_max_blksz = newval;
280 	zfsvfs->z_vfs->vfs_bsize = newval;
281 }
282 
283 static void
284 readonly_changed_cb(void *arg, uint64_t newval)
285 {
286 	zfsvfs_t *zfsvfs = arg;
287 
288 	if (newval) {
289 		/* XXX locking on vfs_flag? */
290 		zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
291 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
292 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
293 	} else {
294 		/* XXX locking on vfs_flag? */
295 		zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
296 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
297 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
298 	}
299 }
300 
301 static void
302 devices_changed_cb(void *arg, uint64_t newval)
303 {
304 	zfsvfs_t *zfsvfs = arg;
305 
306 	if (newval == FALSE) {
307 		zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
308 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
309 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
310 	} else {
311 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
312 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
313 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
314 	}
315 }
316 
317 static void
318 setuid_changed_cb(void *arg, uint64_t newval)
319 {
320 	zfsvfs_t *zfsvfs = arg;
321 
322 	if (newval == FALSE) {
323 		zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
324 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
325 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
326 	} else {
327 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
328 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
329 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
330 	}
331 }
332 
333 static void
334 exec_changed_cb(void *arg, uint64_t newval)
335 {
336 	zfsvfs_t *zfsvfs = arg;
337 
338 	if (newval == FALSE) {
339 		zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
340 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
341 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
342 	} else {
343 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
344 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
345 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
346 	}
347 }
348 
349 /*
350  * The nbmand mount option can be changed at mount time.
351  * We can't allow it to be toggled on live file systems or incorrect
352  * behavior may be seen from cifs clients
353  *
354  * This property isn't registered via dsl_prop_register(), but this callback
355  * will be called when a file system is first mounted
356  */
357 static void
358 nbmand_changed_cb(void *arg, uint64_t newval)
359 {
360 	zfsvfs_t *zfsvfs = arg;
361 	if (newval == FALSE) {
362 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
363 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
364 	} else {
365 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
366 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
367 	}
368 }
369 
370 static void
371 snapdir_changed_cb(void *arg, uint64_t newval)
372 {
373 	zfsvfs_t *zfsvfs = arg;
374 
375 	zfsvfs->z_show_ctldir = newval;
376 }
377 
378 static void
379 vscan_changed_cb(void *arg, uint64_t newval)
380 {
381 	zfsvfs_t *zfsvfs = arg;
382 
383 	zfsvfs->z_vscan = newval;
384 }
385 
386 static void
387 acl_inherit_changed_cb(void *arg, uint64_t newval)
388 {
389 	zfsvfs_t *zfsvfs = arg;
390 
391 	zfsvfs->z_acl_inherit = newval;
392 }
393 
394 static int
395 zfs_register_callbacks(vfs_t *vfsp)
396 {
397 	struct dsl_dataset *ds = NULL;
398 	objset_t *os = NULL;
399 	zfsvfs_t *zfsvfs = NULL;
400 	uint64_t nbmand;
401 	int readonly, do_readonly = B_FALSE;
402 	int setuid, do_setuid = B_FALSE;
403 	int exec, do_exec = B_FALSE;
404 	int devices, do_devices = B_FALSE;
405 	int xattr, do_xattr = B_FALSE;
406 	int atime, do_atime = B_FALSE;
407 	int error = 0;
408 
409 	ASSERT(vfsp);
410 	zfsvfs = vfsp->vfs_data;
411 	ASSERT(zfsvfs);
412 	os = zfsvfs->z_os;
413 
414 	/*
415 	 * The act of registering our callbacks will destroy any mount
416 	 * options we may have.  In order to enable temporary overrides
417 	 * of mount options, we stash away the current values and
418 	 * restore them after we register the callbacks.
419 	 */
420 	if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
421 	    !spa_writeable(dmu_objset_spa(os))) {
422 		readonly = B_TRUE;
423 		do_readonly = B_TRUE;
424 	} else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
425 		readonly = B_FALSE;
426 		do_readonly = B_TRUE;
427 	}
428 	if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
429 		devices = B_FALSE;
430 		setuid = B_FALSE;
431 		do_devices = B_TRUE;
432 		do_setuid = B_TRUE;
433 	} else {
434 		if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
435 			devices = B_FALSE;
436 			do_devices = B_TRUE;
437 		} else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
438 			devices = B_TRUE;
439 			do_devices = B_TRUE;
440 		}
441 
442 		if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
443 			setuid = B_FALSE;
444 			do_setuid = B_TRUE;
445 		} else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
446 			setuid = B_TRUE;
447 			do_setuid = B_TRUE;
448 		}
449 	}
450 	if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
451 		exec = B_FALSE;
452 		do_exec = B_TRUE;
453 	} else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
454 		exec = B_TRUE;
455 		do_exec = B_TRUE;
456 	}
457 	if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
458 		xattr = B_FALSE;
459 		do_xattr = B_TRUE;
460 	} else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
461 		xattr = B_TRUE;
462 		do_xattr = B_TRUE;
463 	}
464 	if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
465 		atime = B_FALSE;
466 		do_atime = B_TRUE;
467 	} else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
468 		atime = B_TRUE;
469 		do_atime = B_TRUE;
470 	}
471 
472 	/*
473 	 * nbmand is a special property.  It can only be changed at
474 	 * mount time.
475 	 *
476 	 * This is weird, but it is documented to only be changeable
477 	 * at mount time.
478 	 */
479 	if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
480 		nbmand = B_FALSE;
481 	} else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
482 		nbmand = B_TRUE;
483 	} else {
484 		char osname[MAXNAMELEN];
485 
486 		dmu_objset_name(os, osname);
487 		if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
488 		    NULL)) {
489 			return (error);
490 		}
491 	}
492 
493 	/*
494 	 * Register property callbacks.
495 	 *
496 	 * It would probably be fine to just check for i/o error from
497 	 * the first prop_register(), but I guess I like to go
498 	 * overboard...
499 	 */
500 	ds = dmu_objset_ds(os);
501 	error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs);
502 	error = error ? error : dsl_prop_register(ds,
503 	    "xattr", xattr_changed_cb, zfsvfs);
504 	error = error ? error : dsl_prop_register(ds,
505 	    "recordsize", blksz_changed_cb, zfsvfs);
506 	error = error ? error : dsl_prop_register(ds,
507 	    "readonly", readonly_changed_cb, zfsvfs);
508 	error = error ? error : dsl_prop_register(ds,
509 	    "devices", devices_changed_cb, zfsvfs);
510 	error = error ? error : dsl_prop_register(ds,
511 	    "setuid", setuid_changed_cb, zfsvfs);
512 	error = error ? error : dsl_prop_register(ds,
513 	    "exec", exec_changed_cb, zfsvfs);
514 	error = error ? error : dsl_prop_register(ds,
515 	    "snapdir", snapdir_changed_cb, zfsvfs);
516 	error = error ? error : dsl_prop_register(ds,
517 	    "aclinherit", acl_inherit_changed_cb, zfsvfs);
518 	error = error ? error : dsl_prop_register(ds,
519 	    "vscan", vscan_changed_cb, zfsvfs);
520 	if (error)
521 		goto unregister;
522 
523 	/*
524 	 * Invoke our callbacks to restore temporary mount options.
525 	 */
526 	if (do_readonly)
527 		readonly_changed_cb(zfsvfs, readonly);
528 	if (do_setuid)
529 		setuid_changed_cb(zfsvfs, setuid);
530 	if (do_exec)
531 		exec_changed_cb(zfsvfs, exec);
532 	if (do_devices)
533 		devices_changed_cb(zfsvfs, devices);
534 	if (do_xattr)
535 		xattr_changed_cb(zfsvfs, xattr);
536 	if (do_atime)
537 		atime_changed_cb(zfsvfs, atime);
538 
539 	nbmand_changed_cb(zfsvfs, nbmand);
540 
541 	return (0);
542 
543 unregister:
544 	/*
545 	 * We may attempt to unregister some callbacks that are not
546 	 * registered, but this is OK; it will simply return ENOMSG,
547 	 * which we will ignore.
548 	 */
549 	(void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs);
550 	(void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs);
551 	(void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs);
552 	(void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs);
553 	(void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zfsvfs);
554 	(void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs);
555 	(void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs);
556 	(void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs);
557 	(void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
558 	    zfsvfs);
559 	(void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs);
560 	return (error);
561 
562 }
563 
564 static int
565 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
566     uint64_t *userp, uint64_t *groupp)
567 {
568 	znode_phys_t *znp = data;
569 	int error = 0;
570 
571 	/*
572 	 * Is it a valid type of object to track?
573 	 */
574 	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
575 		return (ENOENT);
576 
577 	/*
578 	 * If we have a NULL data pointer
579 	 * then assume the id's aren't changing and
580 	 * return EEXIST to the dmu to let it know to
581 	 * use the same ids
582 	 */
583 	if (data == NULL)
584 		return (EEXIST);
585 
586 	if (bonustype == DMU_OT_ZNODE) {
587 		*userp = znp->zp_uid;
588 		*groupp = znp->zp_gid;
589 	} else {
590 		int hdrsize;
591 
592 		ASSERT(bonustype == DMU_OT_SA);
593 		hdrsize = sa_hdrsize(data);
594 
595 		if (hdrsize != 0) {
596 			*userp = *((uint64_t *)((uintptr_t)data + hdrsize +
597 			    SA_UID_OFFSET));
598 			*groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
599 			    SA_GID_OFFSET));
600 		} else {
601 			/*
602 			 * This should only happen for newly created
603 			 * files that haven't had the znode data filled
604 			 * in yet.
605 			 */
606 			*userp = 0;
607 			*groupp = 0;
608 		}
609 	}
610 	return (error);
611 }
612 
613 static void
614 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
615     char *domainbuf, int buflen, uid_t *ridp)
616 {
617 	uint64_t fuid;
618 	const char *domain;
619 
620 	fuid = strtonum(fuidstr, NULL);
621 
622 	domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
623 	if (domain)
624 		(void) strlcpy(domainbuf, domain, buflen);
625 	else
626 		domainbuf[0] = '\0';
627 	*ridp = FUID_RID(fuid);
628 }
629 
630 static uint64_t
631 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
632 {
633 	switch (type) {
634 	case ZFS_PROP_USERUSED:
635 		return (DMU_USERUSED_OBJECT);
636 	case ZFS_PROP_GROUPUSED:
637 		return (DMU_GROUPUSED_OBJECT);
638 	case ZFS_PROP_USERQUOTA:
639 		return (zfsvfs->z_userquota_obj);
640 	case ZFS_PROP_GROUPQUOTA:
641 		return (zfsvfs->z_groupquota_obj);
642 	}
643 	return (0);
644 }
645 
646 int
647 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
648     uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
649 {
650 	int error;
651 	zap_cursor_t zc;
652 	zap_attribute_t za;
653 	zfs_useracct_t *buf = vbuf;
654 	uint64_t obj;
655 
656 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
657 		return (ENOTSUP);
658 
659 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
660 	if (obj == 0) {
661 		*bufsizep = 0;
662 		return (0);
663 	}
664 
665 	for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
666 	    (error = zap_cursor_retrieve(&zc, &za)) == 0;
667 	    zap_cursor_advance(&zc)) {
668 		if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
669 		    *bufsizep)
670 			break;
671 
672 		fuidstr_to_sid(zfsvfs, za.za_name,
673 		    buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
674 
675 		buf->zu_space = za.za_first_integer;
676 		buf++;
677 	}
678 	if (error == ENOENT)
679 		error = 0;
680 
681 	ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
682 	*bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
683 	*cookiep = zap_cursor_serialize(&zc);
684 	zap_cursor_fini(&zc);
685 	return (error);
686 }
687 
688 /*
689  * buf must be big enough (eg, 32 bytes)
690  */
691 static int
692 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
693     char *buf, boolean_t addok)
694 {
695 	uint64_t fuid;
696 	int domainid = 0;
697 
698 	if (domain && domain[0]) {
699 		domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
700 		if (domainid == -1)
701 			return (ENOENT);
702 	}
703 	fuid = FUID_ENCODE(domainid, rid);
704 	(void) sprintf(buf, "%llx", (longlong_t)fuid);
705 	return (0);
706 }
707 
708 int
709 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
710     const char *domain, uint64_t rid, uint64_t *valp)
711 {
712 	char buf[32];
713 	int err;
714 	uint64_t obj;
715 
716 	*valp = 0;
717 
718 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
719 		return (ENOTSUP);
720 
721 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
722 	if (obj == 0)
723 		return (0);
724 
725 	err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
726 	if (err)
727 		return (err);
728 
729 	err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
730 	if (err == ENOENT)
731 		err = 0;
732 	return (err);
733 }
734 
735 int
736 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
737     const char *domain, uint64_t rid, uint64_t quota)
738 {
739 	char buf[32];
740 	int err;
741 	dmu_tx_t *tx;
742 	uint64_t *objp;
743 	boolean_t fuid_dirtied;
744 
745 	if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
746 		return (EINVAL);
747 
748 	if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
749 		return (ENOTSUP);
750 
751 	objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
752 	    &zfsvfs->z_groupquota_obj;
753 
754 	err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
755 	if (err)
756 		return (err);
757 	fuid_dirtied = zfsvfs->z_fuid_dirty;
758 
759 	tx = dmu_tx_create(zfsvfs->z_os);
760 	dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
761 	if (*objp == 0) {
762 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
763 		    zfs_userquota_prop_prefixes[type]);
764 	}
765 	if (fuid_dirtied)
766 		zfs_fuid_txhold(zfsvfs, tx);
767 	err = dmu_tx_assign(tx, TXG_WAIT);
768 	if (err) {
769 		dmu_tx_abort(tx);
770 		return (err);
771 	}
772 
773 	mutex_enter(&zfsvfs->z_lock);
774 	if (*objp == 0) {
775 		*objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
776 		    DMU_OT_NONE, 0, tx);
777 		VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
778 		    zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
779 	}
780 	mutex_exit(&zfsvfs->z_lock);
781 
782 	if (quota == 0) {
783 		err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
784 		if (err == ENOENT)
785 			err = 0;
786 	} else {
787 		err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
788 	}
789 	ASSERT(err == 0);
790 	if (fuid_dirtied)
791 		zfs_fuid_sync(zfsvfs, tx);
792 	dmu_tx_commit(tx);
793 	return (err);
794 }
795 
796 boolean_t
797 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
798 {
799 	char buf[32];
800 	uint64_t used, quota, usedobj, quotaobj;
801 	int err;
802 
803 	usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
804 	quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
805 
806 	if (quotaobj == 0 || zfsvfs->z_replay)
807 		return (B_FALSE);
808 
809 	(void) sprintf(buf, "%llx", (longlong_t)fuid);
810 	err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
811 	if (err != 0)
812 		return (B_FALSE);
813 
814 	err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
815 	if (err != 0)
816 		return (B_FALSE);
817 	return (used >= quota);
818 }
819 
820 boolean_t
821 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
822 {
823 	uint64_t fuid;
824 	uint64_t quotaobj;
825 
826 	quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
827 
828 	fuid = isgroup ? zp->z_gid : zp->z_uid;
829 
830 	if (quotaobj == 0 || zfsvfs->z_replay)
831 		return (B_FALSE);
832 
833 	return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
834 }
835 
836 int
837 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
838 {
839 	objset_t *os;
840 	zfsvfs_t *zfsvfs;
841 	uint64_t zval;
842 	int i, error;
843 	uint64_t sa_obj;
844 
845 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
846 
847 	/*
848 	 * We claim to always be readonly so we can open snapshots;
849 	 * other ZPL code will prevent us from writing to snapshots.
850 	 */
851 	error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
852 	if (error) {
853 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
854 		return (error);
855 	}
856 
857 	/*
858 	 * Initialize the zfs-specific filesystem structure.
859 	 * Should probably make this a kmem cache, shuffle fields,
860 	 * and just bzero up to z_hold_mtx[].
861 	 */
862 	zfsvfs->z_vfs = NULL;
863 	zfsvfs->z_parent = zfsvfs;
864 	zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
865 	zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
866 	zfsvfs->z_os = os;
867 
868 	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
869 	if (error) {
870 		goto out;
871 	} else if (zfsvfs->z_version >
872 	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
873 		(void) printf("Can't mount a version %lld file system "
874 		    "on a version %lld pool\n. Pool must be upgraded to mount "
875 		    "this file system.", (u_longlong_t)zfsvfs->z_version,
876 		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
877 		error = ENOTSUP;
878 		goto out;
879 	}
880 	if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
881 		goto out;
882 	zfsvfs->z_norm = (int)zval;
883 
884 	if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
885 		goto out;
886 	zfsvfs->z_utf8 = (zval != 0);
887 
888 	if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
889 		goto out;
890 	zfsvfs->z_case = (uint_t)zval;
891 
892 	/*
893 	 * Fold case on file systems that are always or sometimes case
894 	 * insensitive.
895 	 */
896 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
897 	    zfsvfs->z_case == ZFS_CASE_MIXED)
898 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
899 
900 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
901 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
902 
903 	if (zfsvfs->z_use_sa) {
904 		/* should either have both of these objects or none */
905 		error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
906 		    &sa_obj);
907 		if (error)
908 			return (error);
909 	} else {
910 		/*
911 		 * Pre SA versions file systems should never touch
912 		 * either the attribute registration or layout objects.
913 		 */
914 		sa_obj = 0;
915 	}
916 
917 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
918 	    &zfsvfs->z_attr_table);
919 	if (error)
920 		goto out;
921 
922 	if (zfsvfs->z_version >= ZPL_VERSION_SA)
923 		sa_register_update_callback(os, zfs_sa_upgrade);
924 
925 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
926 	    &zfsvfs->z_root);
927 	if (error)
928 		goto out;
929 	ASSERT(zfsvfs->z_root != 0);
930 
931 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
932 	    &zfsvfs->z_unlinkedobj);
933 	if (error)
934 		goto out;
935 
936 	error = zap_lookup(os, MASTER_NODE_OBJ,
937 	    zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
938 	    8, 1, &zfsvfs->z_userquota_obj);
939 	if (error && error != ENOENT)
940 		goto out;
941 
942 	error = zap_lookup(os, MASTER_NODE_OBJ,
943 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
944 	    8, 1, &zfsvfs->z_groupquota_obj);
945 	if (error && error != ENOENT)
946 		goto out;
947 
948 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
949 	    &zfsvfs->z_fuid_obj);
950 	if (error && error != ENOENT)
951 		goto out;
952 
953 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
954 	    &zfsvfs->z_shares_dir);
955 	if (error && error != ENOENT)
956 		goto out;
957 
958 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
959 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
960 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
961 	    offsetof(znode_t, z_link_node));
962 	rrw_init(&zfsvfs->z_teardown_lock);
963 	rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
964 	rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
965 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
966 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
967 
968 	*zfvp = zfsvfs;
969 	return (0);
970 
971 out:
972 	dmu_objset_disown(os, zfsvfs);
973 	*zfvp = NULL;
974 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
975 	return (error);
976 }
977 
978 static int
979 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
980 {
981 	int error;
982 
983 	error = zfs_register_callbacks(zfsvfs->z_vfs);
984 	if (error)
985 		return (error);
986 
987 	/*
988 	 * Set the objset user_ptr to track its zfsvfs.
989 	 */
990 	mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
991 	dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
992 	mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
993 
994 	zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
995 
996 	/*
997 	 * If we are not mounting (ie: online recv), then we don't
998 	 * have to worry about replaying the log as we blocked all
999 	 * operations out since we closed the ZIL.
1000 	 */
1001 	if (mounting) {
1002 		boolean_t readonly;
1003 
1004 		/*
1005 		 * During replay we remove the read only flag to
1006 		 * allow replays to succeed.
1007 		 */
1008 		readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1009 		if (readonly != 0)
1010 			zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1011 		else
1012 			zfs_unlinked_drain(zfsvfs);
1013 
1014 		/*
1015 		 * Parse and replay the intent log.
1016 		 *
1017 		 * Because of ziltest, this must be done after
1018 		 * zfs_unlinked_drain().  (Further note: ziltest
1019 		 * doesn't use readonly mounts, where
1020 		 * zfs_unlinked_drain() isn't called.)  This is because
1021 		 * ziltest causes spa_sync() to think it's committed,
1022 		 * but actually it is not, so the intent log contains
1023 		 * many txg's worth of changes.
1024 		 *
1025 		 * In particular, if object N is in the unlinked set in
1026 		 * the last txg to actually sync, then it could be
1027 		 * actually freed in a later txg and then reallocated
1028 		 * in a yet later txg.  This would write a "create
1029 		 * object N" record to the intent log.  Normally, this
1030 		 * would be fine because the spa_sync() would have
1031 		 * written out the fact that object N is free, before
1032 		 * we could write the "create object N" intent log
1033 		 * record.
1034 		 *
1035 		 * But when we are in ziltest mode, we advance the "open
1036 		 * txg" without actually spa_sync()-ing the changes to
1037 		 * disk.  So we would see that object N is still
1038 		 * allocated and in the unlinked set, and there is an
1039 		 * intent log record saying to allocate it.
1040 		 */
1041 		if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1042 			if (zil_replay_disable) {
1043 				zil_destroy(zfsvfs->z_log, B_FALSE);
1044 			} else {
1045 				zfsvfs->z_replay = B_TRUE;
1046 				zil_replay(zfsvfs->z_os, zfsvfs,
1047 				    zfs_replay_vector);
1048 				zfsvfs->z_replay = B_FALSE;
1049 			}
1050 		}
1051 		zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1052 	}
1053 
1054 	return (0);
1055 }
1056 
1057 void
1058 zfsvfs_free(zfsvfs_t *zfsvfs)
1059 {
1060 	int i;
1061 	extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1062 
1063 	/*
1064 	 * This is a barrier to prevent the filesystem from going away in
1065 	 * zfs_znode_move() until we can safely ensure that the filesystem is
1066 	 * not unmounted. We consider the filesystem valid before the barrier
1067 	 * and invalid after the barrier.
1068 	 */
1069 	rw_enter(&zfsvfs_lock, RW_READER);
1070 	rw_exit(&zfsvfs_lock);
1071 
1072 	zfs_fuid_destroy(zfsvfs);
1073 
1074 	mutex_destroy(&zfsvfs->z_znodes_lock);
1075 	mutex_destroy(&zfsvfs->z_lock);
1076 	list_destroy(&zfsvfs->z_all_znodes);
1077 	rrw_destroy(&zfsvfs->z_teardown_lock);
1078 	rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1079 	rw_destroy(&zfsvfs->z_fuid_lock);
1080 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1081 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1082 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1083 }
1084 
1085 static void
1086 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1087 {
1088 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1089 	if (zfsvfs->z_use_fuids && zfsvfs->z_vfs) {
1090 		vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1091 		vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1092 		vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1093 		vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1094 		vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1095 		vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1096 	}
1097 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1098 }
1099 
1100 static int
1101 zfs_domount(vfs_t *vfsp, char *osname)
1102 {
1103 	dev_t mount_dev;
1104 	uint64_t recordsize, fsid_guid;
1105 	int error = 0;
1106 	zfsvfs_t *zfsvfs;
1107 
1108 	ASSERT(vfsp);
1109 	ASSERT(osname);
1110 
1111 	error = zfsvfs_create(osname, &zfsvfs);
1112 	if (error)
1113 		return (error);
1114 	zfsvfs->z_vfs = vfsp;
1115 
1116 	/* Initialize the generic filesystem structure. */
1117 	vfsp->vfs_bcount = 0;
1118 	vfsp->vfs_data = NULL;
1119 
1120 	if (zfs_create_unique_device(&mount_dev) == -1) {
1121 		error = ENODEV;
1122 		goto out;
1123 	}
1124 	ASSERT(vfs_devismounted(mount_dev) == 0);
1125 
1126 	if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1127 	    NULL))
1128 		goto out;
1129 
1130 	vfsp->vfs_dev = mount_dev;
1131 	vfsp->vfs_fstype = zfsfstype;
1132 	vfsp->vfs_bsize = recordsize;
1133 	vfsp->vfs_flag |= VFS_NOTRUNC;
1134 	vfsp->vfs_data = zfsvfs;
1135 
1136 	/*
1137 	 * The fsid is 64 bits, composed of an 8-bit fs type, which
1138 	 * separates our fsid from any other filesystem types, and a
1139 	 * 56-bit objset unique ID.  The objset unique ID is unique to
1140 	 * all objsets open on this system, provided by unique_create().
1141 	 * The 8-bit fs type must be put in the low bits of fsid[1]
1142 	 * because that's where other Solaris filesystems put it.
1143 	 */
1144 	fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1145 	ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1146 	vfsp->vfs_fsid.val[0] = fsid_guid;
1147 	vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1148 	    zfsfstype & 0xFF;
1149 
1150 	/*
1151 	 * Set features for file system.
1152 	 */
1153 	zfs_set_fuid_feature(zfsvfs);
1154 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1155 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1156 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1157 		vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1158 	} else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1159 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1160 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1161 	}
1162 	vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1163 
1164 	if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1165 		uint64_t pval;
1166 
1167 		atime_changed_cb(zfsvfs, B_FALSE);
1168 		readonly_changed_cb(zfsvfs, B_TRUE);
1169 		if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1170 			goto out;
1171 		xattr_changed_cb(zfsvfs, pval);
1172 		zfsvfs->z_issnap = B_TRUE;
1173 		zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1174 
1175 		mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1176 		dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1177 		mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1178 	} else {
1179 		error = zfsvfs_setup(zfsvfs, B_TRUE);
1180 	}
1181 
1182 	if (!zfsvfs->z_issnap)
1183 		zfsctl_create(zfsvfs);
1184 out:
1185 	if (error) {
1186 		dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1187 		zfsvfs_free(zfsvfs);
1188 	} else {
1189 		atomic_add_32(&zfs_active_fs_count, 1);
1190 	}
1191 
1192 	return (error);
1193 }
1194 
1195 void
1196 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1197 {
1198 	objset_t *os = zfsvfs->z_os;
1199 	struct dsl_dataset *ds;
1200 
1201 	/*
1202 	 * Unregister properties.
1203 	 */
1204 	if (!dmu_objset_is_snapshot(os)) {
1205 		ds = dmu_objset_ds(os);
1206 		VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1207 		    zfsvfs) == 0);
1208 
1209 		VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1210 		    zfsvfs) == 0);
1211 
1212 		VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1213 		    zfsvfs) == 0);
1214 
1215 		VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1216 		    zfsvfs) == 0);
1217 
1218 		VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1219 		    zfsvfs) == 0);
1220 
1221 		VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1222 		    zfsvfs) == 0);
1223 
1224 		VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1225 		    zfsvfs) == 0);
1226 
1227 		VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1228 		    zfsvfs) == 0);
1229 
1230 		VERIFY(dsl_prop_unregister(ds, "aclinherit",
1231 		    acl_inherit_changed_cb, zfsvfs) == 0);
1232 
1233 		VERIFY(dsl_prop_unregister(ds, "vscan",
1234 		    vscan_changed_cb, zfsvfs) == 0);
1235 	}
1236 }
1237 
1238 /*
1239  * Convert a decimal digit string to a uint64_t integer.
1240  */
1241 static int
1242 str_to_uint64(char *str, uint64_t *objnum)
1243 {
1244 	uint64_t num = 0;
1245 
1246 	while (*str) {
1247 		if (*str < '0' || *str > '9')
1248 			return (EINVAL);
1249 
1250 		num = num*10 + *str++ - '0';
1251 	}
1252 
1253 	*objnum = num;
1254 	return (0);
1255 }
1256 
1257 /*
1258  * The boot path passed from the boot loader is in the form of
1259  * "rootpool-name/root-filesystem-object-number'. Convert this
1260  * string to a dataset name: "rootpool-name/root-filesystem-name".
1261  */
1262 static int
1263 zfs_parse_bootfs(char *bpath, char *outpath)
1264 {
1265 	char *slashp;
1266 	uint64_t objnum;
1267 	int error;
1268 
1269 	if (*bpath == 0 || *bpath == '/')
1270 		return (EINVAL);
1271 
1272 	(void) strcpy(outpath, bpath);
1273 
1274 	slashp = strchr(bpath, '/');
1275 
1276 	/* if no '/', just return the pool name */
1277 	if (slashp == NULL) {
1278 		return (0);
1279 	}
1280 
1281 	/* if not a number, just return the root dataset name */
1282 	if (str_to_uint64(slashp+1, &objnum)) {
1283 		return (0);
1284 	}
1285 
1286 	*slashp = '\0';
1287 	error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1288 	*slashp = '/';
1289 
1290 	return (error);
1291 }
1292 
1293 /*
1294  * zfs_check_global_label:
1295  *	Check that the hex label string is appropriate for the dataset
1296  *	being mounted into the global_zone proper.
1297  *
1298  *	Return an error if the hex label string is not default or
1299  *	admin_low/admin_high.  For admin_low labels, the corresponding
1300  *	dataset must be readonly.
1301  */
1302 int
1303 zfs_check_global_label(const char *dsname, const char *hexsl)
1304 {
1305 	if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1306 		return (0);
1307 	if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1308 		return (0);
1309 	if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1310 		/* must be readonly */
1311 		uint64_t rdonly;
1312 
1313 		if (dsl_prop_get_integer(dsname,
1314 		    zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1315 			return (EACCES);
1316 		return (rdonly ? 0 : EACCES);
1317 	}
1318 	return (EACCES);
1319 }
1320 
1321 /*
1322  * zfs_mount_label_policy:
1323  *	Determine whether the mount is allowed according to MAC check.
1324  *	by comparing (where appropriate) label of the dataset against
1325  *	the label of the zone being mounted into.  If the dataset has
1326  *	no label, create one.
1327  *
1328  *	Returns:
1329  *		 0 :	access allowed
1330  *		>0 :	error code, such as EACCES
1331  */
1332 static int
1333 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1334 {
1335 	int		error, retv;
1336 	zone_t		*mntzone = NULL;
1337 	ts_label_t	*mnt_tsl;
1338 	bslabel_t	*mnt_sl;
1339 	bslabel_t	ds_sl;
1340 	char		ds_hexsl[MAXNAMELEN];
1341 
1342 	retv = EACCES;				/* assume the worst */
1343 
1344 	/*
1345 	 * Start by getting the dataset label if it exists.
1346 	 */
1347 	error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1348 	    1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1349 	if (error)
1350 		return (EACCES);
1351 
1352 	/*
1353 	 * If labeling is NOT enabled, then disallow the mount of datasets
1354 	 * which have a non-default label already.  No other label checks
1355 	 * are needed.
1356 	 */
1357 	if (!is_system_labeled()) {
1358 		if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1359 			return (0);
1360 		return (EACCES);
1361 	}
1362 
1363 	/*
1364 	 * Get the label of the mountpoint.  If mounting into the global
1365 	 * zone (i.e. mountpoint is not within an active zone and the
1366 	 * zoned property is off), the label must be default or
1367 	 * admin_low/admin_high only; no other checks are needed.
1368 	 */
1369 	mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1370 	if (mntzone->zone_id == GLOBAL_ZONEID) {
1371 		uint64_t zoned;
1372 
1373 		zone_rele(mntzone);
1374 
1375 		if (dsl_prop_get_integer(osname,
1376 		    zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1377 			return (EACCES);
1378 		if (!zoned)
1379 			return (zfs_check_global_label(osname, ds_hexsl));
1380 		else
1381 			/*
1382 			 * This is the case of a zone dataset being mounted
1383 			 * initially, before the zone has been fully created;
1384 			 * allow this mount into global zone.
1385 			 */
1386 			return (0);
1387 	}
1388 
1389 	mnt_tsl = mntzone->zone_slabel;
1390 	ASSERT(mnt_tsl != NULL);
1391 	label_hold(mnt_tsl);
1392 	mnt_sl = label2bslabel(mnt_tsl);
1393 
1394 	if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1395 		/*
1396 		 * The dataset doesn't have a real label, so fabricate one.
1397 		 */
1398 		char *str = NULL;
1399 
1400 		if (l_to_str_internal(mnt_sl, &str) == 0 &&
1401 		    dsl_prop_set(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1402 		    ZPROP_SRC_LOCAL, 1, strlen(str) + 1, str) == 0)
1403 			retv = 0;
1404 		if (str != NULL)
1405 			kmem_free(str, strlen(str) + 1);
1406 	} else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1407 		/*
1408 		 * Now compare labels to complete the MAC check.  If the
1409 		 * labels are equal then allow access.  If the mountpoint
1410 		 * label dominates the dataset label, allow readonly access.
1411 		 * Otherwise, access is denied.
1412 		 */
1413 		if (blequal(mnt_sl, &ds_sl))
1414 			retv = 0;
1415 		else if (bldominates(mnt_sl, &ds_sl)) {
1416 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1417 			retv = 0;
1418 		}
1419 	}
1420 
1421 	label_rele(mnt_tsl);
1422 	zone_rele(mntzone);
1423 	return (retv);
1424 }
1425 
1426 static int
1427 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1428 {
1429 	int error = 0;
1430 	static int zfsrootdone = 0;
1431 	zfsvfs_t *zfsvfs = NULL;
1432 	znode_t *zp = NULL;
1433 	vnode_t *vp = NULL;
1434 	char *zfs_bootfs;
1435 	char *zfs_devid;
1436 
1437 	ASSERT(vfsp);
1438 
1439 	/*
1440 	 * The filesystem that we mount as root is defined in the
1441 	 * boot property "zfs-bootfs" with a format of
1442 	 * "poolname/root-dataset-objnum".
1443 	 */
1444 	if (why == ROOT_INIT) {
1445 		if (zfsrootdone++)
1446 			return (EBUSY);
1447 		/*
1448 		 * the process of doing a spa_load will require the
1449 		 * clock to be set before we could (for example) do
1450 		 * something better by looking at the timestamp on
1451 		 * an uberblock, so just set it to -1.
1452 		 */
1453 		clkset(-1);
1454 
1455 		if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1456 			cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1457 			    "bootfs name");
1458 			return (EINVAL);
1459 		}
1460 		zfs_devid = spa_get_bootprop("diskdevid");
1461 		error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1462 		if (zfs_devid)
1463 			spa_free_bootprop(zfs_devid);
1464 		if (error) {
1465 			spa_free_bootprop(zfs_bootfs);
1466 			cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1467 			    error);
1468 			return (error);
1469 		}
1470 		if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1471 			spa_free_bootprop(zfs_bootfs);
1472 			cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1473 			    error);
1474 			return (error);
1475 		}
1476 
1477 		spa_free_bootprop(zfs_bootfs);
1478 
1479 		if (error = vfs_lock(vfsp))
1480 			return (error);
1481 
1482 		if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1483 			cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1484 			goto out;
1485 		}
1486 
1487 		zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1488 		ASSERT(zfsvfs);
1489 		if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1490 			cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1491 			goto out;
1492 		}
1493 
1494 		vp = ZTOV(zp);
1495 		mutex_enter(&vp->v_lock);
1496 		vp->v_flag |= VROOT;
1497 		mutex_exit(&vp->v_lock);
1498 		rootvp = vp;
1499 
1500 		/*
1501 		 * Leave rootvp held.  The root file system is never unmounted.
1502 		 */
1503 
1504 		vfs_add((struct vnode *)0, vfsp,
1505 		    (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1506 out:
1507 		vfs_unlock(vfsp);
1508 		return (error);
1509 	} else if (why == ROOT_REMOUNT) {
1510 		readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1511 		vfsp->vfs_flag |= VFS_REMOUNT;
1512 
1513 		/* refresh mount options */
1514 		zfs_unregister_callbacks(vfsp->vfs_data);
1515 		return (zfs_register_callbacks(vfsp));
1516 
1517 	} else if (why == ROOT_UNMOUNT) {
1518 		zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1519 		(void) zfs_sync(vfsp, 0, 0);
1520 		return (0);
1521 	}
1522 
1523 	/*
1524 	 * if "why" is equal to anything else other than ROOT_INIT,
1525 	 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1526 	 */
1527 	return (ENOTSUP);
1528 }
1529 
1530 /*ARGSUSED*/
1531 static int
1532 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1533 {
1534 	char		*osname;
1535 	pathname_t	spn;
1536 	int		error = 0;
1537 	uio_seg_t	fromspace = (uap->flags & MS_SYSSPACE) ?
1538 	    UIO_SYSSPACE : UIO_USERSPACE;
1539 	int		canwrite;
1540 
1541 	if (mvp->v_type != VDIR)
1542 		return (ENOTDIR);
1543 
1544 	mutex_enter(&mvp->v_lock);
1545 	if ((uap->flags & MS_REMOUNT) == 0 &&
1546 	    (uap->flags & MS_OVERLAY) == 0 &&
1547 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1548 		mutex_exit(&mvp->v_lock);
1549 		return (EBUSY);
1550 	}
1551 	mutex_exit(&mvp->v_lock);
1552 
1553 	/*
1554 	 * ZFS does not support passing unparsed data in via MS_DATA.
1555 	 * Users should use the MS_OPTIONSTR interface; this means
1556 	 * that all option parsing is already done and the options struct
1557 	 * can be interrogated.
1558 	 */
1559 	if ((uap->flags & MS_DATA) && uap->datalen > 0)
1560 		return (EINVAL);
1561 
1562 	/*
1563 	 * Get the objset name (the "special" mount argument).
1564 	 */
1565 	if (error = pn_get(uap->spec, fromspace, &spn))
1566 		return (error);
1567 
1568 	osname = spn.pn_path;
1569 
1570 	/*
1571 	 * Check for mount privilege?
1572 	 *
1573 	 * If we don't have privilege then see if
1574 	 * we have local permission to allow it
1575 	 */
1576 	error = secpolicy_fs_mount(cr, mvp, vfsp);
1577 	if (error) {
1578 		if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1579 			vattr_t		vattr;
1580 
1581 			/*
1582 			 * Make sure user is the owner of the mount point
1583 			 * or has sufficient privileges.
1584 			 */
1585 
1586 			vattr.va_mask = AT_UID;
1587 
1588 			if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1589 				goto out;
1590 			}
1591 
1592 			if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1593 			    VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1594 				goto out;
1595 			}
1596 			secpolicy_fs_mount_clearopts(cr, vfsp);
1597 		} else {
1598 			goto out;
1599 		}
1600 	}
1601 
1602 	/*
1603 	 * Refuse to mount a filesystem if we are in a local zone and the
1604 	 * dataset is not visible.
1605 	 */
1606 	if (!INGLOBALZONE(curproc) &&
1607 	    (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1608 		error = EPERM;
1609 		goto out;
1610 	}
1611 
1612 	error = zfs_mount_label_policy(vfsp, osname);
1613 	if (error)
1614 		goto out;
1615 
1616 	/*
1617 	 * When doing a remount, we simply refresh our temporary properties
1618 	 * according to those options set in the current VFS options.
1619 	 */
1620 	if (uap->flags & MS_REMOUNT) {
1621 		/* refresh mount options */
1622 		zfs_unregister_callbacks(vfsp->vfs_data);
1623 		error = zfs_register_callbacks(vfsp);
1624 		goto out;
1625 	}
1626 
1627 	error = zfs_domount(vfsp, osname);
1628 
1629 	/*
1630 	 * Add an extra VFS_HOLD on our parent vfs so that it can't
1631 	 * disappear due to a forced unmount.
1632 	 */
1633 	if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1634 		VFS_HOLD(mvp->v_vfsp);
1635 
1636 out:
1637 	pn_free(&spn);
1638 	return (error);
1639 }
1640 
1641 static int
1642 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1643 {
1644 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1645 	dev32_t d32;
1646 	uint64_t refdbytes, availbytes, usedobjs, availobjs;
1647 
1648 	ZFS_ENTER(zfsvfs);
1649 
1650 	dmu_objset_space(zfsvfs->z_os,
1651 	    &refdbytes, &availbytes, &usedobjs, &availobjs);
1652 
1653 	/*
1654 	 * The underlying storage pool actually uses multiple block sizes.
1655 	 * We report the fragsize as the smallest block size we support,
1656 	 * and we report our blocksize as the filesystem's maximum blocksize.
1657 	 */
1658 	statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1659 	statp->f_bsize = zfsvfs->z_max_blksz;
1660 
1661 	/*
1662 	 * The following report "total" blocks of various kinds in the
1663 	 * file system, but reported in terms of f_frsize - the
1664 	 * "fragment" size.
1665 	 */
1666 
1667 	statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1668 	statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1669 	statp->f_bavail = statp->f_bfree; /* no root reservation */
1670 
1671 	/*
1672 	 * statvfs() should really be called statufs(), because it assumes
1673 	 * static metadata.  ZFS doesn't preallocate files, so the best
1674 	 * we can do is report the max that could possibly fit in f_files,
1675 	 * and that minus the number actually used in f_ffree.
1676 	 * For f_ffree, report the smaller of the number of object available
1677 	 * and the number of blocks (each object will take at least a block).
1678 	 */
1679 	statp->f_ffree = MIN(availobjs, statp->f_bfree);
1680 	statp->f_favail = statp->f_ffree;	/* no "root reservation" */
1681 	statp->f_files = statp->f_ffree + usedobjs;
1682 
1683 	(void) cmpldev(&d32, vfsp->vfs_dev);
1684 	statp->f_fsid = d32;
1685 
1686 	/*
1687 	 * We're a zfs filesystem.
1688 	 */
1689 	(void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1690 
1691 	statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1692 
1693 	statp->f_namemax = ZFS_MAXNAMELEN;
1694 
1695 	/*
1696 	 * We have all of 32 characters to stuff a string here.
1697 	 * Is there anything useful we could/should provide?
1698 	 */
1699 	bzero(statp->f_fstr, sizeof (statp->f_fstr));
1700 
1701 	ZFS_EXIT(zfsvfs);
1702 	return (0);
1703 }
1704 
1705 static int
1706 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1707 {
1708 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1709 	znode_t *rootzp;
1710 	int error;
1711 
1712 	ZFS_ENTER(zfsvfs);
1713 
1714 	error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1715 	if (error == 0)
1716 		*vpp = ZTOV(rootzp);
1717 
1718 	ZFS_EXIT(zfsvfs);
1719 	return (error);
1720 }
1721 
1722 /*
1723  * Teardown the zfsvfs::z_os.
1724  *
1725  * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1726  * and 'z_teardown_inactive_lock' held.
1727  */
1728 static int
1729 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1730 {
1731 	znode_t	*zp;
1732 
1733 	rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1734 
1735 	if (!unmounting) {
1736 		/*
1737 		 * We purge the parent filesystem's vfsp as the parent
1738 		 * filesystem and all of its snapshots have their vnode's
1739 		 * v_vfsp set to the parent's filesystem's vfsp.  Note,
1740 		 * 'z_parent' is self referential for non-snapshots.
1741 		 */
1742 		(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1743 	}
1744 
1745 	/*
1746 	 * Close the zil. NB: Can't close the zil while zfs_inactive
1747 	 * threads are blocked as zil_close can call zfs_inactive.
1748 	 */
1749 	if (zfsvfs->z_log) {
1750 		zil_close(zfsvfs->z_log);
1751 		zfsvfs->z_log = NULL;
1752 	}
1753 
1754 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1755 
1756 	/*
1757 	 * If we are not unmounting (ie: online recv) and someone already
1758 	 * unmounted this file system while we were doing the switcheroo,
1759 	 * or a reopen of z_os failed then just bail out now.
1760 	 */
1761 	if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1762 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
1763 		rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1764 		return (EIO);
1765 	}
1766 
1767 	/*
1768 	 * At this point there are no vops active, and any new vops will
1769 	 * fail with EIO since we have z_teardown_lock for writer (only
1770 	 * relavent for forced unmount).
1771 	 *
1772 	 * Release all holds on dbufs.
1773 	 */
1774 	mutex_enter(&zfsvfs->z_znodes_lock);
1775 	for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1776 	    zp = list_next(&zfsvfs->z_all_znodes, zp))
1777 		if (zp->z_sa_hdl) {
1778 			ASSERT(ZTOV(zp)->v_count > 0);
1779 			zfs_znode_dmu_fini(zp);
1780 		}
1781 	mutex_exit(&zfsvfs->z_znodes_lock);
1782 
1783 	/*
1784 	 * If we are unmounting, set the unmounted flag and let new vops
1785 	 * unblock.  zfs_inactive will have the unmounted behavior, and all
1786 	 * other vops will fail with EIO.
1787 	 */
1788 	if (unmounting) {
1789 		zfsvfs->z_unmounted = B_TRUE;
1790 		rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1791 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
1792 	}
1793 
1794 	/*
1795 	 * z_os will be NULL if there was an error in attempting to reopen
1796 	 * zfsvfs, so just return as the properties had already been
1797 	 * unregistered and cached data had been evicted before.
1798 	 */
1799 	if (zfsvfs->z_os == NULL)
1800 		return (0);
1801 
1802 	/*
1803 	 * Unregister properties.
1804 	 */
1805 	zfs_unregister_callbacks(zfsvfs);
1806 
1807 	/*
1808 	 * Evict cached data
1809 	 */
1810 	if (dmu_objset_is_dirty_anywhere(zfsvfs->z_os))
1811 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1812 			txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1813 	(void) dmu_objset_evict_dbufs(zfsvfs->z_os);
1814 
1815 	return (0);
1816 }
1817 
1818 /*ARGSUSED*/
1819 static int
1820 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1821 {
1822 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1823 	objset_t *os;
1824 	int ret;
1825 
1826 	ret = secpolicy_fs_unmount(cr, vfsp);
1827 	if (ret) {
1828 		if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1829 		    ZFS_DELEG_PERM_MOUNT, cr))
1830 			return (ret);
1831 	}
1832 
1833 	/*
1834 	 * We purge the parent filesystem's vfsp as the parent filesystem
1835 	 * and all of its snapshots have their vnode's v_vfsp set to the
1836 	 * parent's filesystem's vfsp.  Note, 'z_parent' is self
1837 	 * referential for non-snapshots.
1838 	 */
1839 	(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1840 
1841 	/*
1842 	 * Unmount any snapshots mounted under .zfs before unmounting the
1843 	 * dataset itself.
1844 	 */
1845 	if (zfsvfs->z_ctldir != NULL &&
1846 	    (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1847 		return (ret);
1848 	}
1849 
1850 	if (!(fflag & MS_FORCE)) {
1851 		/*
1852 		 * Check the number of active vnodes in the file system.
1853 		 * Our count is maintained in the vfs structure, but the
1854 		 * number is off by 1 to indicate a hold on the vfs
1855 		 * structure itself.
1856 		 *
1857 		 * The '.zfs' directory maintains a reference of its
1858 		 * own, and any active references underneath are
1859 		 * reflected in the vnode count.
1860 		 */
1861 		if (zfsvfs->z_ctldir == NULL) {
1862 			if (vfsp->vfs_count > 1)
1863 				return (EBUSY);
1864 		} else {
1865 			if (vfsp->vfs_count > 2 ||
1866 			    zfsvfs->z_ctldir->v_count > 1)
1867 				return (EBUSY);
1868 		}
1869 	}
1870 
1871 	vfsp->vfs_flag |= VFS_UNMOUNTED;
1872 
1873 	VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1874 	os = zfsvfs->z_os;
1875 
1876 	/*
1877 	 * z_os will be NULL if there was an error in
1878 	 * attempting to reopen zfsvfs.
1879 	 */
1880 	if (os != NULL) {
1881 		/*
1882 		 * Unset the objset user_ptr.
1883 		 */
1884 		mutex_enter(&os->os_user_ptr_lock);
1885 		dmu_objset_set_user(os, NULL);
1886 		mutex_exit(&os->os_user_ptr_lock);
1887 
1888 		/*
1889 		 * Finally release the objset
1890 		 */
1891 		dmu_objset_disown(os, zfsvfs);
1892 	}
1893 
1894 	/*
1895 	 * We can now safely destroy the '.zfs' directory node.
1896 	 */
1897 	if (zfsvfs->z_ctldir != NULL)
1898 		zfsctl_destroy(zfsvfs);
1899 
1900 	return (0);
1901 }
1902 
1903 static int
1904 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1905 {
1906 	zfsvfs_t	*zfsvfs = vfsp->vfs_data;
1907 	znode_t		*zp;
1908 	uint64_t	object = 0;
1909 	uint64_t	fid_gen = 0;
1910 	uint64_t	gen_mask;
1911 	uint64_t	zp_gen;
1912 	int 		i, err;
1913 
1914 	*vpp = NULL;
1915 
1916 	ZFS_ENTER(zfsvfs);
1917 
1918 	if (fidp->fid_len == LONG_FID_LEN) {
1919 		zfid_long_t	*zlfid = (zfid_long_t *)fidp;
1920 		uint64_t	objsetid = 0;
1921 		uint64_t	setgen = 0;
1922 
1923 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1924 			objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1925 
1926 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1927 			setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1928 
1929 		ZFS_EXIT(zfsvfs);
1930 
1931 		err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1932 		if (err)
1933 			return (EINVAL);
1934 		ZFS_ENTER(zfsvfs);
1935 	}
1936 
1937 	if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1938 		zfid_short_t	*zfid = (zfid_short_t *)fidp;
1939 
1940 		for (i = 0; i < sizeof (zfid->zf_object); i++)
1941 			object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1942 
1943 		for (i = 0; i < sizeof (zfid->zf_gen); i++)
1944 			fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1945 	} else {
1946 		ZFS_EXIT(zfsvfs);
1947 		return (EINVAL);
1948 	}
1949 
1950 	/* A zero fid_gen means we are in the .zfs control directories */
1951 	if (fid_gen == 0 &&
1952 	    (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1953 		*vpp = zfsvfs->z_ctldir;
1954 		ASSERT(*vpp != NULL);
1955 		if (object == ZFSCTL_INO_SNAPDIR) {
1956 			VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
1957 			    0, NULL, NULL, NULL, NULL, NULL) == 0);
1958 		} else {
1959 			VN_HOLD(*vpp);
1960 		}
1961 		ZFS_EXIT(zfsvfs);
1962 		return (0);
1963 	}
1964 
1965 	gen_mask = -1ULL >> (64 - 8 * i);
1966 
1967 	dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1968 	if (err = zfs_zget(zfsvfs, object, &zp)) {
1969 		ZFS_EXIT(zfsvfs);
1970 		return (err);
1971 	}
1972 	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1973 	    sizeof (uint64_t));
1974 	zp_gen = zp_gen & gen_mask;
1975 	if (zp_gen == 0)
1976 		zp_gen = 1;
1977 	if (zp->z_unlinked || zp_gen != fid_gen) {
1978 		dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1979 		VN_RELE(ZTOV(zp));
1980 		ZFS_EXIT(zfsvfs);
1981 		return (EINVAL);
1982 	}
1983 
1984 	*vpp = ZTOV(zp);
1985 	ZFS_EXIT(zfsvfs);
1986 	return (0);
1987 }
1988 
1989 /*
1990  * Block out VOPs and close zfsvfs_t::z_os
1991  *
1992  * Note, if successful, then we return with the 'z_teardown_lock' and
1993  * 'z_teardown_inactive_lock' write held.
1994  */
1995 int
1996 zfs_suspend_fs(zfsvfs_t *zfsvfs)
1997 {
1998 	int error;
1999 
2000 	if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2001 		return (error);
2002 	dmu_objset_disown(zfsvfs->z_os, zfsvfs);
2003 
2004 	return (0);
2005 }
2006 
2007 /*
2008  * Reopen zfsvfs_t::z_os and release VOPs.
2009  */
2010 int
2011 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2012 {
2013 	int err, err2;
2014 
2015 	ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
2016 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2017 
2018 	err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zfsvfs,
2019 	    &zfsvfs->z_os);
2020 	if (err) {
2021 		zfsvfs->z_os = NULL;
2022 	} else {
2023 		znode_t *zp;
2024 		uint64_t sa_obj = 0;
2025 
2026 		err2 = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2027 		    ZFS_SA_ATTRS, 8, 1, &sa_obj);
2028 
2029 		if ((err || err2) && zfsvfs->z_version >= ZPL_VERSION_SA)
2030 			goto bail;
2031 
2032 
2033 		if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2034 		    zfs_attr_table,  ZPL_END, &zfsvfs->z_attr_table)) != 0)
2035 			goto bail;
2036 
2037 		VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2038 
2039 		/*
2040 		 * Attempt to re-establish all the active znodes with
2041 		 * their dbufs.  If a zfs_rezget() fails, then we'll let
2042 		 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2043 		 * when they try to use their znode.
2044 		 */
2045 		mutex_enter(&zfsvfs->z_znodes_lock);
2046 		for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2047 		    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2048 			(void) zfs_rezget(zp);
2049 		}
2050 		mutex_exit(&zfsvfs->z_znodes_lock);
2051 
2052 	}
2053 
2054 bail:
2055 	/* release the VOPs */
2056 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
2057 	rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
2058 
2059 	if (err) {
2060 		/*
2061 		 * Since we couldn't reopen zfsvfs::z_os, force
2062 		 * unmount this file system.
2063 		 */
2064 		if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2065 			(void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2066 	}
2067 	return (err);
2068 }
2069 
2070 static void
2071 zfs_freevfs(vfs_t *vfsp)
2072 {
2073 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2074 
2075 	/*
2076 	 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2077 	 * from zfs_mount().  Release it here.  If we came through
2078 	 * zfs_mountroot() instead, we didn't grab an extra hold, so
2079 	 * skip the VFS_RELE for rootvfs.
2080 	 */
2081 	if (zfsvfs->z_issnap && (vfsp != rootvfs))
2082 		VFS_RELE(zfsvfs->z_parent->z_vfs);
2083 
2084 	zfsvfs_free(zfsvfs);
2085 
2086 	atomic_add_32(&zfs_active_fs_count, -1);
2087 }
2088 
2089 /*
2090  * VFS_INIT() initialization.  Note that there is no VFS_FINI(),
2091  * so we can't safely do any non-idempotent initialization here.
2092  * Leave that to zfs_init() and zfs_fini(), which are called
2093  * from the module's _init() and _fini() entry points.
2094  */
2095 /*ARGSUSED*/
2096 static int
2097 zfs_vfsinit(int fstype, char *name)
2098 {
2099 	int error;
2100 
2101 	zfsfstype = fstype;
2102 
2103 	/*
2104 	 * Setup vfsops and vnodeops tables.
2105 	 */
2106 	error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2107 	if (error != 0) {
2108 		cmn_err(CE_WARN, "zfs: bad vfs ops template");
2109 	}
2110 
2111 	error = zfs_create_op_tables();
2112 	if (error) {
2113 		zfs_remove_op_tables();
2114 		cmn_err(CE_WARN, "zfs: bad vnode ops template");
2115 		(void) vfs_freevfsops_by_type(zfsfstype);
2116 		return (error);
2117 	}
2118 
2119 	mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2120 
2121 	/*
2122 	 * Unique major number for all zfs mounts.
2123 	 * If we run out of 32-bit minors, we'll getudev() another major.
2124 	 */
2125 	zfs_major = ddi_name_to_major(ZFS_DRIVER);
2126 	zfs_minor = ZFS_MIN_MINOR;
2127 
2128 	return (0);
2129 }
2130 
2131 void
2132 zfs_init(void)
2133 {
2134 	/*
2135 	 * Initialize .zfs directory structures
2136 	 */
2137 	zfsctl_init();
2138 
2139 	/*
2140 	 * Initialize znode cache, vnode ops, etc...
2141 	 */
2142 	zfs_znode_init();
2143 
2144 	dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2145 }
2146 
2147 void
2148 zfs_fini(void)
2149 {
2150 	zfsctl_fini();
2151 	zfs_znode_fini();
2152 }
2153 
2154 int
2155 zfs_busy(void)
2156 {
2157 	return (zfs_active_fs_count != 0);
2158 }
2159 
2160 int
2161 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2162 {
2163 	int error;
2164 	objset_t *os = zfsvfs->z_os;
2165 	dmu_tx_t *tx;
2166 
2167 	if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2168 		return (EINVAL);
2169 
2170 	if (newvers < zfsvfs->z_version)
2171 		return (EINVAL);
2172 
2173 	if (zfs_spa_version_map(newvers) >
2174 	    spa_version(dmu_objset_spa(zfsvfs->z_os)))
2175 		return (ENOTSUP);
2176 
2177 	tx = dmu_tx_create(os);
2178 	dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2179 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2180 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2181 		    ZFS_SA_ATTRS);
2182 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2183 	}
2184 	error = dmu_tx_assign(tx, TXG_WAIT);
2185 	if (error) {
2186 		dmu_tx_abort(tx);
2187 		return (error);
2188 	}
2189 
2190 	error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2191 	    8, 1, &newvers, tx);
2192 
2193 	if (error) {
2194 		dmu_tx_commit(tx);
2195 		return (error);
2196 	}
2197 
2198 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2199 		uint64_t sa_obj;
2200 
2201 		ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2202 		    SPA_VERSION_SA);
2203 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2204 		    DMU_OT_NONE, 0, tx);
2205 
2206 		error = zap_add(os, MASTER_NODE_OBJ,
2207 		    ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2208 		ASSERT3U(error, ==, 0);
2209 
2210 		VERIFY(0 == sa_set_sa_object(os, sa_obj));
2211 		sa_register_update_callback(os, zfs_sa_upgrade);
2212 	}
2213 
2214 	spa_history_log_internal(LOG_DS_UPGRADE,
2215 	    dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu",
2216 	    zfsvfs->z_version, newvers, dmu_objset_id(os));
2217 
2218 	dmu_tx_commit(tx);
2219 
2220 	zfsvfs->z_version = newvers;
2221 
2222 	if (zfsvfs->z_version >= ZPL_VERSION_FUID)
2223 		zfs_set_fuid_feature(zfsvfs);
2224 
2225 	return (0);
2226 }
2227 
2228 /*
2229  * Read a property stored within the master node.
2230  */
2231 int
2232 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2233 {
2234 	const char *pname;
2235 	int error = ENOENT;
2236 
2237 	/*
2238 	 * Look up the file system's value for the property.  For the
2239 	 * version property, we look up a slightly different string.
2240 	 */
2241 	if (prop == ZFS_PROP_VERSION)
2242 		pname = ZPL_VERSION_STR;
2243 	else
2244 		pname = zfs_prop_to_name(prop);
2245 
2246 	if (os != NULL)
2247 		error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2248 
2249 	if (error == ENOENT) {
2250 		/* No value set, use the default value */
2251 		switch (prop) {
2252 		case ZFS_PROP_VERSION:
2253 			*value = ZPL_VERSION;
2254 			break;
2255 		case ZFS_PROP_NORMALIZE:
2256 		case ZFS_PROP_UTF8ONLY:
2257 			*value = 0;
2258 			break;
2259 		case ZFS_PROP_CASE:
2260 			*value = ZFS_CASE_SENSITIVE;
2261 			break;
2262 		default:
2263 			return (error);
2264 		}
2265 		error = 0;
2266 	}
2267 	return (error);
2268 }
2269 
2270 static vfsdef_t vfw = {
2271 	VFSDEF_VERSION,
2272 	MNTTYPE_ZFS,
2273 	zfs_vfsinit,
2274 	VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2275 	    VSW_XID|VSW_ZMOUNT,
2276 	&zfs_mntopts
2277 };
2278 
2279 struct modlfs zfs_modlfs = {
2280 	&mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2281 };
2282