xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_vfsops.c (revision 4e320a8a375380e6f3963134cd6c1001e2bdb7fa)
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  * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24  * Copyright (c) 2014 Integros [integros.com]
25  * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
26  * Copyright 2019 Joyent, Inc.
27  * Copyright 2020 Joshua M. Clulow <josh@sysmgr.org>
28  * Copyright 2020 OmniOS Community Edition (OmniOSce) Association.
29  * Copyright 2022 Oxide Computer Company
30  */
31 
32 /* Portions Copyright 2010 Robert Milkowski */
33 
34 #include <sys/types.h>
35 #include <sys/param.h>
36 #include <sys/systm.h>
37 #include <sys/sysmacros.h>
38 #include <sys/kmem.h>
39 #include <sys/pathname.h>
40 #include <sys/vnode.h>
41 #include <sys/vfs.h>
42 #include <sys/vfs_opreg.h>
43 #include <sys/mntent.h>
44 #include <sys/mount.h>
45 #include <sys/cmn_err.h>
46 #include "fs/fs_subr.h"
47 #include <sys/zfs_znode.h>
48 #include <sys/zfs_dir.h>
49 #include <sys/zil.h>
50 #include <sys/fs/zfs.h>
51 #include <sys/dmu.h>
52 #include <sys/dsl_prop.h>
53 #include <sys/dsl_dataset.h>
54 #include <sys/dsl_deleg.h>
55 #include <sys/spa.h>
56 #include <sys/zap.h>
57 #include <sys/sa.h>
58 #include <sys/sa_impl.h>
59 #include <sys/varargs.h>
60 #include <sys/policy.h>
61 #include <sys/atomic.h>
62 #include <sys/mkdev.h>
63 #include <sys/modctl.h>
64 #include <sys/refstr.h>
65 #include <sys/zfs_ioctl.h>
66 #include <sys/zfs_ctldir.h>
67 #include <sys/zfs_fuid.h>
68 #include <sys/bootconf.h>
69 #include <sys/ddi.h>
70 #include <sys/sunddi.h>
71 #include <sys/dnlc.h>
72 #include <sys/dmu_objset.h>
73 #include <sys/spa_boot.h>
74 #include <sys/vdev_impl.h>
75 #include "zfs_comutil.h"
76 
77 int zfsfstype;
78 vfsops_t *zfs_vfsops = NULL;
79 static major_t zfs_major;
80 static minor_t zfs_minor;
81 static kmutex_t	zfs_dev_mtx;
82 
83 extern int sys_shutdown;
84 
85 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
86 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
87 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
88 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
89 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
90 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
91 static void zfs_freevfs(vfs_t *vfsp);
92 
93 static const fs_operation_def_t zfs_vfsops_template[] = {
94 	VFSNAME_MOUNT,		{ .vfs_mount = zfs_mount },
95 	VFSNAME_MOUNTROOT,	{ .vfs_mountroot = zfs_mountroot },
96 	VFSNAME_UNMOUNT,	{ .vfs_unmount = zfs_umount },
97 	VFSNAME_ROOT,		{ .vfs_root = zfs_root },
98 	VFSNAME_STATVFS,	{ .vfs_statvfs = zfs_statvfs },
99 	VFSNAME_SYNC,		{ .vfs_sync = zfs_sync },
100 	VFSNAME_VGET,		{ .vfs_vget = zfs_vget },
101 	VFSNAME_FREEVFS,	{ .vfs_freevfs = zfs_freevfs },
102 	NULL,			NULL
103 };
104 
105 /*
106  * We need to keep a count of active fs's.
107  * This is necessary to prevent our module
108  * from being unloaded after a umount -f
109  */
110 static uint32_t	zfs_active_fs_count = 0;
111 
112 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
113 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
114 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
115 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
116 
117 /*
118  * MO_DEFAULT is not used since the default value is determined
119  * by the equivalent property.
120  */
121 static mntopt_t mntopts[] = {
122 	{ MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
123 	{ MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
124 	{ MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
125 	{ MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
126 };
127 
128 static mntopts_t zfs_mntopts = {
129 	sizeof (mntopts) / sizeof (mntopt_t),
130 	mntopts
131 };
132 
133 /*ARGSUSED*/
134 int
135 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
136 {
137 	/*
138 	 * Data integrity is job one.  We don't want a compromised kernel
139 	 * writing to the storage pool, so we never sync during panic.
140 	 */
141 	if (panicstr)
142 		return (0);
143 
144 	/*
145 	 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
146 	 * to sync metadata, which they would otherwise cache indefinitely.
147 	 * Semantically, the only requirement is that the sync be initiated.
148 	 * The DMU syncs out txgs frequently, so there's nothing to do.
149 	 */
150 	if (flag & SYNC_ATTR)
151 		return (0);
152 
153 	if (vfsp != NULL) {
154 		/*
155 		 * Sync a specific filesystem.
156 		 */
157 		zfsvfs_t *zfsvfs = vfsp->vfs_data;
158 		dsl_pool_t *dp;
159 
160 		ZFS_ENTER(zfsvfs);
161 		dp = dmu_objset_pool(zfsvfs->z_os);
162 
163 		/*
164 		 * If the system is shutting down, then skip any
165 		 * filesystems which may exist on a suspended pool.
166 		 */
167 		if (sys_shutdown && spa_suspended(dp->dp_spa)) {
168 			ZFS_EXIT(zfsvfs);
169 			return (0);
170 		}
171 
172 		if (zfsvfs->z_log != NULL)
173 			zil_commit(zfsvfs->z_log, 0);
174 
175 		ZFS_EXIT(zfsvfs);
176 	} else {
177 		/*
178 		 * Sync all ZFS filesystems.  This is what happens when you
179 		 * run sync(8).  Unlike other filesystems, ZFS honors the
180 		 * request by waiting for all pools to commit all dirty data.
181 		 */
182 		spa_sync_allpools();
183 	}
184 
185 	return (0);
186 }
187 
188 static int
189 zfs_create_unique_device(dev_t *dev)
190 {
191 	major_t new_major;
192 
193 	do {
194 		ASSERT3U(zfs_minor, <=, MAXMIN32);
195 		minor_t start = zfs_minor;
196 		do {
197 			mutex_enter(&zfs_dev_mtx);
198 			if (zfs_minor >= MAXMIN32) {
199 				/*
200 				 * If we're still using the real major
201 				 * keep out of /dev/zfs and /dev/zvol minor
202 				 * number space.  If we're using a getudev()'ed
203 				 * major number, we can use all of its minors.
204 				 */
205 				if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
206 					zfs_minor = ZFS_MIN_MINOR;
207 				else
208 					zfs_minor = 0;
209 			} else {
210 				zfs_minor++;
211 			}
212 			*dev = makedevice(zfs_major, zfs_minor);
213 			mutex_exit(&zfs_dev_mtx);
214 		} while (vfs_devismounted(*dev) && zfs_minor != start);
215 		if (zfs_minor == start) {
216 			/*
217 			 * We are using all ~262,000 minor numbers for the
218 			 * current major number.  Create a new major number.
219 			 */
220 			if ((new_major = getudev()) == (major_t)-1) {
221 				cmn_err(CE_WARN,
222 				    "zfs_mount: Can't get unique major "
223 				    "device number.");
224 				return (-1);
225 			}
226 			mutex_enter(&zfs_dev_mtx);
227 			zfs_major = new_major;
228 			zfs_minor = 0;
229 
230 			mutex_exit(&zfs_dev_mtx);
231 		} else {
232 			break;
233 		}
234 		/* CONSTANTCONDITION */
235 	} while (1);
236 
237 	return (0);
238 }
239 
240 static void
241 atime_changed_cb(void *arg, uint64_t newval)
242 {
243 	zfsvfs_t *zfsvfs = arg;
244 
245 	if (newval == TRUE) {
246 		zfsvfs->z_atime = TRUE;
247 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
248 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
249 	} else {
250 		zfsvfs->z_atime = FALSE;
251 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
252 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
253 	}
254 }
255 
256 static void
257 xattr_changed_cb(void *arg, uint64_t newval)
258 {
259 	zfsvfs_t *zfsvfs = arg;
260 
261 	if (newval == TRUE) {
262 		/* XXX locking on vfs_flag? */
263 		zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
264 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
265 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
266 	} else {
267 		/* XXX locking on vfs_flag? */
268 		zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
269 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
270 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
271 	}
272 }
273 
274 static void
275 blksz_changed_cb(void *arg, uint64_t newval)
276 {
277 	zfsvfs_t *zfsvfs = arg;
278 	ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
279 	ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
280 	ASSERT(ISP2(newval));
281 
282 	zfsvfs->z_max_blksz = newval;
283 	zfsvfs->z_vfs->vfs_bsize = newval;
284 }
285 
286 static void
287 readonly_changed_cb(void *arg, uint64_t newval)
288 {
289 	zfsvfs_t *zfsvfs = arg;
290 
291 	if (newval) {
292 		/* XXX locking on vfs_flag? */
293 		zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
294 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
295 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
296 	} else {
297 		/* XXX locking on vfs_flag? */
298 		zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
299 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
300 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
301 	}
302 }
303 
304 static void
305 devices_changed_cb(void *arg, uint64_t newval)
306 {
307 	zfsvfs_t *zfsvfs = arg;
308 
309 	if (newval == FALSE) {
310 		zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
311 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
312 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
313 	} else {
314 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
315 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
316 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
317 	}
318 }
319 
320 static void
321 setuid_changed_cb(void *arg, uint64_t newval)
322 {
323 	zfsvfs_t *zfsvfs = arg;
324 
325 	if (newval == FALSE) {
326 		zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
327 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
328 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
329 	} else {
330 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
331 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
332 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
333 	}
334 }
335 
336 static void
337 exec_changed_cb(void *arg, uint64_t newval)
338 {
339 	zfsvfs_t *zfsvfs = arg;
340 
341 	if (newval == FALSE) {
342 		zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
343 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
344 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
345 	} else {
346 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
347 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
348 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
349 	}
350 }
351 
352 /*
353  * The nbmand mount option can be changed at mount time.
354  * We can't allow it to be toggled on live file systems or incorrect
355  * behavior may be seen from cifs clients
356  *
357  * This property isn't registered via dsl_prop_register(), but this callback
358  * will be called when a file system is first mounted
359  */
360 static void
361 nbmand_changed_cb(void *arg, uint64_t newval)
362 {
363 	zfsvfs_t *zfsvfs = arg;
364 	if (newval == FALSE) {
365 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
366 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
367 	} else {
368 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
369 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
370 	}
371 }
372 
373 static void
374 snapdir_changed_cb(void *arg, uint64_t newval)
375 {
376 	zfsvfs_t *zfsvfs = arg;
377 
378 	zfsvfs->z_show_ctldir = newval;
379 }
380 
381 static void
382 vscan_changed_cb(void *arg, uint64_t newval)
383 {
384 	zfsvfs_t *zfsvfs = arg;
385 
386 	zfsvfs->z_vscan = newval;
387 }
388 
389 static void
390 acl_mode_changed_cb(void *arg, uint64_t newval)
391 {
392 	zfsvfs_t *zfsvfs = arg;
393 
394 	zfsvfs->z_acl_mode = newval;
395 }
396 
397 static void
398 acl_inherit_changed_cb(void *arg, uint64_t newval)
399 {
400 	zfsvfs_t *zfsvfs = arg;
401 
402 	zfsvfs->z_acl_inherit = newval;
403 }
404 
405 static int
406 zfs_register_callbacks(vfs_t *vfsp)
407 {
408 	struct dsl_dataset *ds = NULL;
409 	objset_t *os = NULL;
410 	zfsvfs_t *zfsvfs = NULL;
411 	uint64_t nbmand;
412 	boolean_t readonly = B_FALSE;
413 	boolean_t do_readonly = B_FALSE;
414 	boolean_t setuid = B_FALSE;
415 	boolean_t do_setuid = B_FALSE;
416 	boolean_t exec = B_FALSE;
417 	boolean_t do_exec = B_FALSE;
418 	boolean_t devices = B_FALSE;
419 	boolean_t do_devices = B_FALSE;
420 	boolean_t xattr = B_FALSE;
421 	boolean_t do_xattr = B_FALSE;
422 	boolean_t atime = B_FALSE;
423 	boolean_t do_atime = B_FALSE;
424 	int error = 0;
425 
426 	ASSERT(vfsp);
427 	zfsvfs = vfsp->vfs_data;
428 	ASSERT(zfsvfs);
429 	os = zfsvfs->z_os;
430 
431 	/*
432 	 * The act of registering our callbacks will destroy any mount
433 	 * options we may have.  In order to enable temporary overrides
434 	 * of mount options, we stash away the current values and
435 	 * restore them after we register the callbacks.
436 	 */
437 	if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
438 	    !spa_writeable(dmu_objset_spa(os))) {
439 		readonly = B_TRUE;
440 		do_readonly = B_TRUE;
441 	} else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
442 		readonly = B_FALSE;
443 		do_readonly = B_TRUE;
444 	}
445 	if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
446 		devices = B_FALSE;
447 		setuid = B_FALSE;
448 		do_devices = B_TRUE;
449 		do_setuid = B_TRUE;
450 	} else {
451 		if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
452 			devices = B_FALSE;
453 			do_devices = B_TRUE;
454 		} else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
455 			devices = B_TRUE;
456 			do_devices = B_TRUE;
457 		}
458 
459 		if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
460 			setuid = B_FALSE;
461 			do_setuid = B_TRUE;
462 		} else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
463 			setuid = B_TRUE;
464 			do_setuid = B_TRUE;
465 		}
466 	}
467 	if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
468 		exec = B_FALSE;
469 		do_exec = B_TRUE;
470 	} else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
471 		exec = B_TRUE;
472 		do_exec = B_TRUE;
473 	}
474 	if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
475 		xattr = B_FALSE;
476 		do_xattr = B_TRUE;
477 	} else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
478 		xattr = B_TRUE;
479 		do_xattr = B_TRUE;
480 	}
481 	if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
482 		atime = B_FALSE;
483 		do_atime = B_TRUE;
484 	} else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
485 		atime = B_TRUE;
486 		do_atime = B_TRUE;
487 	}
488 
489 	/*
490 	 * nbmand is a special property.  It can only be changed at
491 	 * mount time.
492 	 *
493 	 * This is weird, but it is documented to only be changeable
494 	 * at mount time.
495 	 */
496 	if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
497 		nbmand = B_FALSE;
498 	} else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
499 		nbmand = B_TRUE;
500 	} else {
501 		char osname[ZFS_MAX_DATASET_NAME_LEN];
502 
503 		dmu_objset_name(os, osname);
504 		if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
505 		    NULL)) {
506 			return (error);
507 		}
508 	}
509 
510 	/*
511 	 * Register property callbacks.
512 	 *
513 	 * It would probably be fine to just check for i/o error from
514 	 * the first prop_register(), but I guess I like to go
515 	 * overboard...
516 	 */
517 	ds = dmu_objset_ds(os);
518 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
519 	error = dsl_prop_register(ds,
520 	    zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
521 	error = error ? error : dsl_prop_register(ds,
522 	    zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
523 	error = error ? error : dsl_prop_register(ds,
524 	    zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
525 	error = error ? error : dsl_prop_register(ds,
526 	    zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
527 	error = error ? error : dsl_prop_register(ds,
528 	    zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
529 	error = error ? error : dsl_prop_register(ds,
530 	    zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
531 	error = error ? error : dsl_prop_register(ds,
532 	    zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
533 	error = error ? error : dsl_prop_register(ds,
534 	    zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
535 	error = error ? error : dsl_prop_register(ds,
536 	    zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
537 	error = error ? error : dsl_prop_register(ds,
538 	    zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
539 	    zfsvfs);
540 	error = error ? error : dsl_prop_register(ds,
541 	    zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
542 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
543 	if (error)
544 		goto unregister;
545 
546 	/*
547 	 * Invoke our callbacks to restore temporary mount options.
548 	 */
549 	if (do_readonly)
550 		readonly_changed_cb(zfsvfs, readonly);
551 	if (do_setuid)
552 		setuid_changed_cb(zfsvfs, setuid);
553 	if (do_exec)
554 		exec_changed_cb(zfsvfs, exec);
555 	if (do_devices)
556 		devices_changed_cb(zfsvfs, devices);
557 	if (do_xattr)
558 		xattr_changed_cb(zfsvfs, xattr);
559 	if (do_atime)
560 		atime_changed_cb(zfsvfs, atime);
561 
562 	nbmand_changed_cb(zfsvfs, nbmand);
563 
564 	return (0);
565 
566 unregister:
567 	dsl_prop_unregister_all(ds, zfsvfs);
568 	return (error);
569 }
570 
571 static int
572 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
573     uint64_t *userp, uint64_t *groupp, uint64_t *projectp)
574 {
575 	sa_hdr_phys_t sa;
576 	sa_hdr_phys_t *sap = data;
577 	uint64_t flags;
578 	int hdrsize;
579 	boolean_t swap = B_FALSE;
580 
581 	/*
582 	 * Is it a valid type of object to track?
583 	 */
584 	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
585 		return (SET_ERROR(ENOENT));
586 
587 	/*
588 	 * If we have a NULL data pointer
589 	 * then assume the id's aren't changing and
590 	 * return EEXIST to the dmu to let it know to
591 	 * use the same ids
592 	 */
593 	if (data == NULL)
594 		return (SET_ERROR(EEXIST));
595 
596 	if (bonustype == DMU_OT_ZNODE) {
597 		znode_phys_t *znp = data;
598 		*userp = znp->zp_uid;
599 		*groupp = znp->zp_gid;
600 		*projectp = ZFS_DEFAULT_PROJID;
601 		return (0);
602 	}
603 
604 	if (sap->sa_magic == 0) {
605 		/*
606 		 * This should only happen for newly created files
607 		 * that haven't had the znode data filled in yet.
608 		 */
609 		*userp = 0;
610 		*groupp = 0;
611 		*projectp = ZFS_DEFAULT_PROJID;
612 		return (0);
613 	}
614 
615 	sa = *sap;
616 	if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
617 		sa.sa_magic = SA_MAGIC;
618 		sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
619 		swap = B_TRUE;
620 	} else {
621 		VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
622 	}
623 
624 	hdrsize = sa_hdrsize(&sa);
625 	VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
626 
627 	*userp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_UID_OFFSET));
628 	*groupp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_GID_OFFSET));
629 	flags = *((uint64_t *)((uintptr_t)data + hdrsize + SA_FLAGS_OFFSET));
630 	if (swap)
631 		flags = BSWAP_64(flags);
632 
633 	if (flags & ZFS_PROJID)
634 		*projectp = *((uint64_t *)((uintptr_t)data + hdrsize +
635 		    SA_PROJID_OFFSET));
636 	else
637 		*projectp = ZFS_DEFAULT_PROJID;
638 
639 	if (swap) {
640 		*userp = BSWAP_64(*userp);
641 		*groupp = BSWAP_64(*groupp);
642 		*projectp = BSWAP_64(*projectp);
643 	}
644 	return (0);
645 }
646 
647 static void
648 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
649     char *domainbuf, int buflen, uid_t *ridp)
650 {
651 	uint64_t fuid;
652 	const char *domain;
653 
654 	fuid = zfs_strtonum(fuidstr, NULL);
655 
656 	domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
657 	if (domain)
658 		(void) strlcpy(domainbuf, domain, buflen);
659 	else
660 		domainbuf[0] = '\0';
661 	*ridp = FUID_RID(fuid);
662 }
663 
664 static uint64_t
665 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
666 {
667 	switch (type) {
668 	case ZFS_PROP_USERUSED:
669 	case ZFS_PROP_USEROBJUSED:
670 		return (DMU_USERUSED_OBJECT);
671 	case ZFS_PROP_GROUPUSED:
672 	case ZFS_PROP_GROUPOBJUSED:
673 		return (DMU_GROUPUSED_OBJECT);
674 	case ZFS_PROP_PROJECTUSED:
675 	case ZFS_PROP_PROJECTOBJUSED:
676 		return (DMU_PROJECTUSED_OBJECT);
677 	case ZFS_PROP_USERQUOTA:
678 		return (zfsvfs->z_userquota_obj);
679 	case ZFS_PROP_GROUPQUOTA:
680 		return (zfsvfs->z_groupquota_obj);
681 	case ZFS_PROP_USEROBJQUOTA:
682 		return (zfsvfs->z_userobjquota_obj);
683 	case ZFS_PROP_GROUPOBJQUOTA:
684 		return (zfsvfs->z_groupobjquota_obj);
685 	case ZFS_PROP_PROJECTQUOTA:
686 		return (zfsvfs->z_projectquota_obj);
687 	case ZFS_PROP_PROJECTOBJQUOTA:
688 		return (zfsvfs->z_projectobjquota_obj);
689 	default:
690 		return (ZFS_NO_OBJECT);
691 	}
692 }
693 
694 int
695 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
696     uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
697 {
698 	int error;
699 	zap_cursor_t zc;
700 	zap_attribute_t za;
701 	zfs_useracct_t *buf = vbuf;
702 	uint64_t obj;
703 	int offset = 0;
704 
705 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
706 		return (SET_ERROR(ENOTSUP));
707 
708 	if ((type == ZFS_PROP_PROJECTQUOTA || type == ZFS_PROP_PROJECTUSED ||
709 	    type == ZFS_PROP_PROJECTOBJQUOTA ||
710 	    type == ZFS_PROP_PROJECTOBJUSED) &&
711 	    !dmu_objset_projectquota_present(zfsvfs->z_os))
712 		return (SET_ERROR(ENOTSUP));
713 
714 	if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
715 	    type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA ||
716 	    type == ZFS_PROP_PROJECTOBJUSED ||
717 	    type == ZFS_PROP_PROJECTOBJQUOTA) &&
718 	    !dmu_objset_userobjspace_present(zfsvfs->z_os))
719 		return (SET_ERROR(ENOTSUP));
720 
721 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
722 	if (obj == ZFS_NO_OBJECT) {
723 		*bufsizep = 0;
724 		return (0);
725 	}
726 
727 	if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
728 	    type == ZFS_PROP_PROJECTOBJUSED)
729 		offset = DMU_OBJACCT_PREFIX_LEN;
730 
731 	for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
732 	    (error = zap_cursor_retrieve(&zc, &za)) == 0;
733 	    zap_cursor_advance(&zc)) {
734 		if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
735 		    *bufsizep)
736 			break;
737 
738 		/*
739 		 * skip object quota (with zap name prefix DMU_OBJACCT_PREFIX)
740 		 * when dealing with block quota and vice versa.
741 		 */
742 		if ((offset > 0) != (strncmp(za.za_name, DMU_OBJACCT_PREFIX,
743 		    DMU_OBJACCT_PREFIX_LEN) == 0))
744 			continue;
745 
746 		fuidstr_to_sid(zfsvfs, za.za_name + offset,
747 		    buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
748 
749 		buf->zu_space = za.za_first_integer;
750 		buf++;
751 	}
752 	if (error == ENOENT)
753 		error = 0;
754 
755 	ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
756 	*bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
757 	*cookiep = zap_cursor_serialize(&zc);
758 	zap_cursor_fini(&zc);
759 	return (error);
760 }
761 
762 /*
763  * buf must be big enough (eg, 32 bytes)
764  */
765 static int
766 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
767     char *buf, boolean_t addok)
768 {
769 	uint64_t fuid;
770 	int domainid = 0;
771 
772 	if (domain && domain[0]) {
773 		domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
774 		if (domainid == -1)
775 			return (SET_ERROR(ENOENT));
776 	}
777 	fuid = FUID_ENCODE(domainid, rid);
778 	(void) sprintf(buf, "%llx", (longlong_t)fuid);
779 	return (0);
780 }
781 
782 int
783 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
784     const char *domain, uint64_t rid, uint64_t *valp)
785 {
786 	char buf[20 + DMU_OBJACCT_PREFIX_LEN];
787 	int offset = 0;
788 	int err;
789 	uint64_t obj;
790 
791 	*valp = 0;
792 
793 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
794 		return (SET_ERROR(ENOTSUP));
795 
796 	if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
797 	    type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA ||
798 	    type == ZFS_PROP_PROJECTOBJUSED ||
799 	    type == ZFS_PROP_PROJECTOBJQUOTA) &&
800 	    !dmu_objset_userobjspace_present(zfsvfs->z_os))
801 		return (SET_ERROR(ENOTSUP));
802 
803 	if (type == ZFS_PROP_PROJECTQUOTA || type == ZFS_PROP_PROJECTUSED ||
804 	    type == ZFS_PROP_PROJECTOBJQUOTA ||
805 	    type == ZFS_PROP_PROJECTOBJUSED) {
806 		if (!dmu_objset_projectquota_present(zfsvfs->z_os))
807 			return (SET_ERROR(ENOTSUP));
808 		if (!zpl_is_valid_projid(rid))
809 			return (SET_ERROR(EINVAL));
810 	}
811 
812 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
813 	if (obj == ZFS_NO_OBJECT)
814 		return (0);
815 
816 	if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
817 	    type == ZFS_PROP_PROJECTOBJUSED) {
818 		strncpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN);
819 		offset = DMU_OBJACCT_PREFIX_LEN;
820 	}
821 
822 	err = id_to_fuidstr(zfsvfs, domain, rid, buf + offset, B_FALSE);
823 	if (err)
824 		return (err);
825 
826 	err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
827 	if (err == ENOENT)
828 		err = 0;
829 	return (err);
830 }
831 
832 int
833 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
834     const char *domain, uint64_t rid, uint64_t quota)
835 {
836 	char buf[32];
837 	int err;
838 	dmu_tx_t *tx;
839 	uint64_t *objp;
840 	boolean_t fuid_dirtied;
841 
842 	if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
843 		return (SET_ERROR(ENOTSUP));
844 
845 	switch (type) {
846 	case ZFS_PROP_USERQUOTA:
847 		objp = &zfsvfs->z_userquota_obj;
848 		break;
849 	case ZFS_PROP_GROUPQUOTA:
850 		objp = &zfsvfs->z_groupquota_obj;
851 		break;
852 	case ZFS_PROP_USEROBJQUOTA:
853 		objp = &zfsvfs->z_userobjquota_obj;
854 		break;
855 	case ZFS_PROP_GROUPOBJQUOTA:
856 		objp = &zfsvfs->z_groupobjquota_obj;
857 		break;
858 	case ZFS_PROP_PROJECTQUOTA:
859 		if (!dmu_objset_projectquota_enabled(zfsvfs->z_os))
860 			return (SET_ERROR(ENOTSUP));
861 		if (!zpl_is_valid_projid(rid))
862 			return (SET_ERROR(EINVAL));
863 
864 		objp = &zfsvfs->z_projectquota_obj;
865 		break;
866 	case ZFS_PROP_PROJECTOBJQUOTA:
867 		if (!dmu_objset_projectquota_enabled(zfsvfs->z_os))
868 			return (SET_ERROR(ENOTSUP));
869 		if (!zpl_is_valid_projid(rid))
870 			return (SET_ERROR(EINVAL));
871 
872 		objp = &zfsvfs->z_projectobjquota_obj;
873 		break;
874 	default:
875 		return (SET_ERROR(EINVAL));
876 	}
877 
878 	err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
879 	if (err)
880 		return (err);
881 	fuid_dirtied = zfsvfs->z_fuid_dirty;
882 
883 	tx = dmu_tx_create(zfsvfs->z_os);
884 	dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
885 	if (*objp == 0) {
886 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
887 		    zfs_userquota_prop_prefixes[type]);
888 	}
889 	if (fuid_dirtied)
890 		zfs_fuid_txhold(zfsvfs, tx);
891 	err = dmu_tx_assign(tx, TXG_WAIT);
892 	if (err) {
893 		dmu_tx_abort(tx);
894 		return (err);
895 	}
896 
897 	mutex_enter(&zfsvfs->z_lock);
898 	if (*objp == 0) {
899 		*objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
900 		    DMU_OT_NONE, 0, tx);
901 		VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
902 		    zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
903 	}
904 	mutex_exit(&zfsvfs->z_lock);
905 
906 	if (quota == 0) {
907 		err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
908 		if (err == ENOENT)
909 			err = 0;
910 	} else {
911 		err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
912 	}
913 	ASSERT(err == 0);
914 	if (fuid_dirtied)
915 		zfs_fuid_sync(zfsvfs, tx);
916 	dmu_tx_commit(tx);
917 	return (err);
918 }
919 
920 boolean_t
921 zfs_id_overobjquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
922 {
923 	char buf[20 + DMU_OBJACCT_PREFIX_LEN];
924 	uint64_t used, quota, quotaobj;
925 	int err;
926 
927 	if (!dmu_objset_userobjspace_present(zfsvfs->z_os)) {
928 		if (dmu_objset_userobjspace_upgradable(zfsvfs->z_os)) {
929 			dsl_pool_config_enter(
930 			    dmu_objset_pool(zfsvfs->z_os), FTAG);
931 			dmu_objset_id_quota_upgrade(zfsvfs->z_os);
932 			dsl_pool_config_exit(
933 			    dmu_objset_pool(zfsvfs->z_os), FTAG);
934 		}
935 		return (B_FALSE);
936 	}
937 
938 	if (usedobj == DMU_PROJECTUSED_OBJECT) {
939 		if (!dmu_objset_projectquota_present(zfsvfs->z_os)) {
940 			if (dmu_objset_projectquota_upgradable(zfsvfs->z_os)) {
941 				dsl_pool_config_enter(
942 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
943 				dmu_objset_id_quota_upgrade(zfsvfs->z_os);
944 				dsl_pool_config_exit(
945 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
946 			}
947 			return (B_FALSE);
948 		}
949 		quotaobj = zfsvfs->z_projectobjquota_obj;
950 	} else if (usedobj == DMU_USERUSED_OBJECT) {
951 		quotaobj = zfsvfs->z_userobjquota_obj;
952 	} else if (usedobj == DMU_GROUPUSED_OBJECT) {
953 		quotaobj = zfsvfs->z_groupobjquota_obj;
954 	} else {
955 		return (B_FALSE);
956 	}
957 	if (quotaobj == 0 || zfsvfs->z_replay)
958 		return (B_FALSE);
959 
960 	(void) sprintf(buf, "%llx", (longlong_t)id);
961 	err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
962 	if (err != 0)
963 		return (B_FALSE);
964 
965 	(void) sprintf(buf, DMU_OBJACCT_PREFIX "%llx", (longlong_t)id);
966 	err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
967 	if (err != 0)
968 		return (B_FALSE);
969 	return (used >= quota);
970 }
971 
972 boolean_t
973 zfs_id_overblockquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
974 {
975 	char buf[20];
976 	uint64_t used, quota, quotaobj;
977 	int err;
978 
979 	if (usedobj == DMU_PROJECTUSED_OBJECT) {
980 		if (!dmu_objset_projectquota_present(zfsvfs->z_os)) {
981 			if (dmu_objset_projectquota_upgradable(zfsvfs->z_os)) {
982 				dsl_pool_config_enter(
983 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
984 				dmu_objset_id_quota_upgrade(zfsvfs->z_os);
985 				dsl_pool_config_exit(
986 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
987 			}
988 			return (B_FALSE);
989 		}
990 		quotaobj = zfsvfs->z_projectquota_obj;
991 	} else if (usedobj == DMU_USERUSED_OBJECT) {
992 		quotaobj = zfsvfs->z_userquota_obj;
993 	} else if (usedobj == DMU_GROUPUSED_OBJECT) {
994 		quotaobj = zfsvfs->z_groupquota_obj;
995 	} else {
996 		return (B_FALSE);
997 	}
998 	if (quotaobj == 0 || zfsvfs->z_replay)
999 		return (B_FALSE);
1000 
1001 	(void) sprintf(buf, "%llx", (longlong_t)id);
1002 	err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
1003 	if (err != 0)
1004 		return (B_FALSE);
1005 
1006 	err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
1007 	if (err != 0)
1008 		return (B_FALSE);
1009 	return (used >= quota);
1010 }
1011 
1012 boolean_t
1013 zfs_id_overquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
1014 {
1015 	return (zfs_id_overblockquota(zfsvfs, usedobj, id) ||
1016 	    zfs_id_overobjquota(zfsvfs, usedobj, id));
1017 }
1018 
1019 /*
1020  * Associate this zfsvfs with the given objset, which must be owned.
1021  * This will cache a bunch of on-disk state from the objset in the
1022  * zfsvfs.
1023  */
1024 static int
1025 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
1026 {
1027 	int error;
1028 	uint64_t val;
1029 
1030 	zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
1031 	zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
1032 	zfsvfs->z_os = os;
1033 
1034 	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
1035 	if (error != 0)
1036 		return (error);
1037 	if (zfsvfs->z_version >
1038 	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
1039 		(void) printf("Can't mount a version %lld file system "
1040 		    "on a version %lld pool\n. Pool must be upgraded to mount "
1041 		    "this file system.", (u_longlong_t)zfsvfs->z_version,
1042 		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
1043 		return (SET_ERROR(ENOTSUP));
1044 	}
1045 	error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
1046 	if (error != 0)
1047 		return (error);
1048 	zfsvfs->z_norm = (int)val;
1049 
1050 	error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
1051 	if (error != 0)
1052 		return (error);
1053 	zfsvfs->z_utf8 = (val != 0);
1054 
1055 	error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
1056 	if (error != 0)
1057 		return (error);
1058 	zfsvfs->z_case = (uint_t)val;
1059 
1060 	/*
1061 	 * Fold case on file systems that are always or sometimes case
1062 	 * insensitive.
1063 	 */
1064 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
1065 	    zfsvfs->z_case == ZFS_CASE_MIXED)
1066 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1067 
1068 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1069 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1070 
1071 	uint64_t sa_obj = 0;
1072 	if (zfsvfs->z_use_sa) {
1073 		/* should either have both of these objects or none */
1074 		error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
1075 		    &sa_obj);
1076 		if (error != 0)
1077 			return (error);
1078 	}
1079 
1080 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1081 	    &zfsvfs->z_attr_table);
1082 	if (error != 0)
1083 		return (error);
1084 
1085 	if (zfsvfs->z_version >= ZPL_VERSION_SA)
1086 		sa_register_update_callback(os, zfs_sa_upgrade);
1087 
1088 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
1089 	    &zfsvfs->z_root);
1090 	if (error != 0)
1091 		return (error);
1092 	ASSERT(zfsvfs->z_root != 0);
1093 
1094 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
1095 	    &zfsvfs->z_unlinkedobj);
1096 	if (error != 0)
1097 		return (error);
1098 
1099 	error = zap_lookup(os, MASTER_NODE_OBJ,
1100 	    zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
1101 	    8, 1, &zfsvfs->z_userquota_obj);
1102 	if (error == ENOENT)
1103 		zfsvfs->z_userquota_obj = 0;
1104 	else if (error != 0)
1105 		return (error);
1106 
1107 	error = zap_lookup(os, MASTER_NODE_OBJ,
1108 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
1109 	    8, 1, &zfsvfs->z_groupquota_obj);
1110 	if (error == ENOENT)
1111 		zfsvfs->z_groupquota_obj = 0;
1112 	else if (error != 0)
1113 		return (error);
1114 
1115 	error = zap_lookup(os, MASTER_NODE_OBJ,
1116 	    zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA],
1117 	    8, 1, &zfsvfs->z_projectquota_obj);
1118 	if (error == ENOENT)
1119 		zfsvfs->z_projectquota_obj = 0;
1120 	else if (error != 0)
1121 		return (error);
1122 
1123 	error = zap_lookup(os, MASTER_NODE_OBJ,
1124 	    zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
1125 	    8, 1, &zfsvfs->z_userobjquota_obj);
1126 	if (error == ENOENT)
1127 		zfsvfs->z_userobjquota_obj = 0;
1128 	else if (error != 0)
1129 		return (error);
1130 
1131 	error = zap_lookup(os, MASTER_NODE_OBJ,
1132 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
1133 	    8, 1, &zfsvfs->z_groupobjquota_obj);
1134 	if (error == ENOENT)
1135 		zfsvfs->z_groupobjquota_obj = 0;
1136 	else if (error != 0)
1137 		return (error);
1138 
1139 	error = zap_lookup(os, MASTER_NODE_OBJ,
1140 	    zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA],
1141 	    8, 1, &zfsvfs->z_projectobjquota_obj);
1142 	if (error == ENOENT)
1143 		zfsvfs->z_projectobjquota_obj = 0;
1144 	else if (error != 0)
1145 		return (error);
1146 
1147 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
1148 	    &zfsvfs->z_fuid_obj);
1149 	if (error == ENOENT)
1150 		zfsvfs->z_fuid_obj = 0;
1151 	else if (error != 0)
1152 		return (error);
1153 
1154 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
1155 	    &zfsvfs->z_shares_dir);
1156 	if (error == ENOENT)
1157 		zfsvfs->z_shares_dir = 0;
1158 	else if (error != 0)
1159 		return (error);
1160 
1161 	return (0);
1162 }
1163 
1164 int
1165 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp)
1166 {
1167 	objset_t *os;
1168 	zfsvfs_t *zfsvfs;
1169 	int error;
1170 	boolean_t ro = (readonly || (strchr(osname, '@') != NULL));
1171 
1172 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1173 
1174 	error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os);
1175 	if (error != 0) {
1176 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
1177 		return (error);
1178 	}
1179 
1180 	error = zfsvfs_create_impl(zfvp, zfsvfs, os);
1181 	if (error != 0) {
1182 		dmu_objset_disown(os, B_TRUE, zfsvfs);
1183 	}
1184 	return (error);
1185 }
1186 
1187 
1188 int
1189 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
1190 {
1191 	int error;
1192 
1193 	zfsvfs->z_vfs = NULL;
1194 	zfsvfs->z_parent = zfsvfs;
1195 
1196 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1197 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1198 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1199 	    offsetof(znode_t, z_link_node));
1200 	rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1201 	rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1202 	rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1203 	for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1204 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1205 
1206 	error = zfsvfs_init(zfsvfs, os);
1207 	if (error != 0) {
1208 		*zfvp = NULL;
1209 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
1210 		return (error);
1211 	}
1212 
1213 	zfsvfs->z_drain_task = TASKQID_INVALID;
1214 	zfsvfs->z_draining = B_FALSE;
1215 	zfsvfs->z_drain_cancel = B_TRUE;
1216 
1217 	*zfvp = zfsvfs;
1218 	return (0);
1219 }
1220 
1221 static int
1222 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1223 {
1224 	int error;
1225 
1226 	error = zfs_register_callbacks(zfsvfs->z_vfs);
1227 	if (error)
1228 		return (error);
1229 
1230 	zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1231 
1232 	/*
1233 	 * If we are not mounting (ie: online recv), then we don't
1234 	 * have to worry about replaying the log as we blocked all
1235 	 * operations out since we closed the ZIL.
1236 	 */
1237 	if (mounting) {
1238 		boolean_t readonly;
1239 
1240 		/*
1241 		 * During replay we remove the read only flag to
1242 		 * allow replays to succeed.
1243 		 */
1244 		readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1245 		if (readonly != 0) {
1246 			zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1247 		} else {
1248 			zfs_unlinked_drain(zfsvfs);
1249 		}
1250 
1251 		/*
1252 		 * Parse and replay the intent log.
1253 		 *
1254 		 * Because of ziltest, this must be done after
1255 		 * zfs_unlinked_drain().  (Further note: ziltest
1256 		 * doesn't use readonly mounts, where
1257 		 * zfs_unlinked_drain() isn't called.)  This is because
1258 		 * ziltest causes spa_sync() to think it's committed,
1259 		 * but actually it is not, so the intent log contains
1260 		 * many txg's worth of changes.
1261 		 *
1262 		 * In particular, if object N is in the unlinked set in
1263 		 * the last txg to actually sync, then it could be
1264 		 * actually freed in a later txg and then reallocated
1265 		 * in a yet later txg.  This would write a "create
1266 		 * object N" record to the intent log.  Normally, this
1267 		 * would be fine because the spa_sync() would have
1268 		 * written out the fact that object N is free, before
1269 		 * we could write the "create object N" intent log
1270 		 * record.
1271 		 *
1272 		 * But when we are in ziltest mode, we advance the "open
1273 		 * txg" without actually spa_sync()-ing the changes to
1274 		 * disk.  So we would see that object N is still
1275 		 * allocated and in the unlinked set, and there is an
1276 		 * intent log record saying to allocate it.
1277 		 */
1278 		if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1279 			if (zil_replay_disable) {
1280 				zil_destroy(zfsvfs->z_log, B_FALSE);
1281 			} else {
1282 				zfsvfs->z_replay = B_TRUE;
1283 				zil_replay(zfsvfs->z_os, zfsvfs,
1284 				    zfs_replay_vector);
1285 				zfsvfs->z_replay = B_FALSE;
1286 			}
1287 		}
1288 
1289 		/* restore readonly bit */
1290 		if (readonly != 0)
1291 			zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
1292 	}
1293 
1294 	/*
1295 	 * Set the objset user_ptr to track its zfsvfs.
1296 	 */
1297 	mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1298 	dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1299 	mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1300 
1301 	return (0);
1302 }
1303 
1304 void
1305 zfsvfs_free(zfsvfs_t *zfsvfs)
1306 {
1307 	int i;
1308 	extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1309 
1310 	/*
1311 	 * This is a barrier to prevent the filesystem from going away in
1312 	 * zfs_znode_move() until we can safely ensure that the filesystem is
1313 	 * not unmounted. We consider the filesystem valid before the barrier
1314 	 * and invalid after the barrier.
1315 	 */
1316 	rw_enter(&zfsvfs_lock, RW_READER);
1317 	rw_exit(&zfsvfs_lock);
1318 
1319 	zfs_fuid_destroy(zfsvfs);
1320 
1321 	mutex_destroy(&zfsvfs->z_znodes_lock);
1322 	mutex_destroy(&zfsvfs->z_lock);
1323 	list_destroy(&zfsvfs->z_all_znodes);
1324 	rrm_destroy(&zfsvfs->z_teardown_lock);
1325 	rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1326 	rw_destroy(&zfsvfs->z_fuid_lock);
1327 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1328 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1329 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1330 }
1331 
1332 static void
1333 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1334 {
1335 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1336 	if (zfsvfs->z_vfs) {
1337 		if (zfsvfs->z_use_fuids) {
1338 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1339 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1340 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1341 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1342 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1343 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1344 		} else {
1345 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1346 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1347 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1348 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1349 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1350 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1351 		}
1352 	}
1353 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1354 }
1355 
1356 static int
1357 zfs_domount(vfs_t *vfsp, char *osname)
1358 {
1359 	dev_t mount_dev;
1360 	uint64_t recordsize, fsid_guid;
1361 	int error = 0;
1362 	zfsvfs_t *zfsvfs;
1363 	boolean_t readonly = vfsp->vfs_flag & VFS_RDONLY ? B_TRUE : B_FALSE;
1364 
1365 	ASSERT(vfsp);
1366 	ASSERT(osname);
1367 
1368 	error = zfsvfs_create(osname, readonly, &zfsvfs);
1369 	if (error)
1370 		return (error);
1371 	zfsvfs->z_vfs = vfsp;
1372 
1373 	/* Initialize the generic filesystem structure. */
1374 	vfsp->vfs_bcount = 0;
1375 	vfsp->vfs_data = NULL;
1376 
1377 	if (zfs_create_unique_device(&mount_dev) == -1) {
1378 		error = SET_ERROR(ENODEV);
1379 		goto out;
1380 	}
1381 	ASSERT(vfs_devismounted(mount_dev) == 0);
1382 
1383 	if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1384 	    NULL))
1385 		goto out;
1386 
1387 	vfsp->vfs_dev = mount_dev;
1388 	vfsp->vfs_fstype = zfsfstype;
1389 	vfsp->vfs_bsize = recordsize;
1390 	vfsp->vfs_flag |= VFS_NOTRUNC;
1391 	vfsp->vfs_data = zfsvfs;
1392 
1393 	/*
1394 	 * The fsid is 64 bits, composed of an 8-bit fs type, which
1395 	 * separates our fsid from any other filesystem types, and a
1396 	 * 56-bit objset unique ID.  The objset unique ID is unique to
1397 	 * all objsets open on this system, provided by unique_create().
1398 	 * The 8-bit fs type must be put in the low bits of fsid[1]
1399 	 * because that's where other Solaris filesystems put it.
1400 	 */
1401 	fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1402 	ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1403 	vfsp->vfs_fsid.val[0] = fsid_guid;
1404 	vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1405 	    zfsfstype & 0xFF;
1406 
1407 	/*
1408 	 * Set features for file system.
1409 	 */
1410 	zfs_set_fuid_feature(zfsvfs);
1411 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1412 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1413 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1414 		vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1415 	} else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1416 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1417 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1418 	}
1419 	vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1420 
1421 	if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1422 		uint64_t pval;
1423 
1424 		atime_changed_cb(zfsvfs, B_FALSE);
1425 		readonly_changed_cb(zfsvfs, B_TRUE);
1426 		if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1427 			goto out;
1428 		xattr_changed_cb(zfsvfs, pval);
1429 		zfsvfs->z_issnap = B_TRUE;
1430 		zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1431 
1432 		mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1433 		dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1434 		mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1435 	} else {
1436 		error = zfsvfs_setup(zfsvfs, B_TRUE);
1437 	}
1438 
1439 	/* cache the root vnode for this mount */
1440 	znode_t *rootzp;
1441 	if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp)) {
1442 		goto out;
1443 	}
1444 	zfsvfs->z_rootdir = ZTOV(rootzp);
1445 
1446 	if (!zfsvfs->z_issnap)
1447 		zfsctl_create(zfsvfs);
1448 out:
1449 	if (error) {
1450 		dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1451 		zfsvfs_free(zfsvfs);
1452 	} else {
1453 		atomic_inc_32(&zfs_active_fs_count);
1454 	}
1455 
1456 	return (error);
1457 }
1458 
1459 void
1460 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1461 {
1462 	objset_t *os = zfsvfs->z_os;
1463 
1464 	if (!dmu_objset_is_snapshot(os))
1465 		dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1466 }
1467 
1468 /*
1469  * Convert a decimal digit string to a uint64_t integer.
1470  */
1471 static int
1472 str_to_uint64(char *str, uint64_t *objnum)
1473 {
1474 	uint64_t num = 0;
1475 
1476 	while (*str) {
1477 		if (*str < '0' || *str > '9')
1478 			return (SET_ERROR(EINVAL));
1479 
1480 		num = num*10 + *str++ - '0';
1481 	}
1482 
1483 	*objnum = num;
1484 	return (0);
1485 }
1486 
1487 /*
1488  * The boot path passed from the boot loader is in the form of
1489  * "rootpool-name/root-filesystem-object-number'. Convert this
1490  * string to a dataset name: "rootpool-name/root-filesystem-name".
1491  */
1492 static int
1493 zfs_parse_bootfs(char *bpath, char *outpath)
1494 {
1495 	char *slashp;
1496 	uint64_t objnum;
1497 	int error;
1498 
1499 	if (*bpath == 0 || *bpath == '/')
1500 		return (SET_ERROR(EINVAL));
1501 
1502 	(void) strcpy(outpath, bpath);
1503 
1504 	slashp = strchr(bpath, '/');
1505 
1506 	/* if no '/', just return the pool name */
1507 	if (slashp == NULL) {
1508 		return (0);
1509 	}
1510 
1511 	/* if not a number, just return the root dataset name */
1512 	if (str_to_uint64(slashp+1, &objnum)) {
1513 		return (0);
1514 	}
1515 
1516 	*slashp = '\0';
1517 	error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1518 	*slashp = '/';
1519 
1520 	return (error);
1521 }
1522 
1523 /*
1524  * Check that the hex label string is appropriate for the dataset being
1525  * mounted into the global_zone proper.
1526  *
1527  * Return an error if the hex label string is not default or
1528  * admin_low/admin_high.  For admin_low labels, the corresponding
1529  * dataset must be readonly.
1530  */
1531 int
1532 zfs_check_global_label(const char *dsname, const char *hexsl)
1533 {
1534 	if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1535 		return (0);
1536 	if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1537 		return (0);
1538 	if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1539 		/* must be readonly */
1540 		uint64_t rdonly;
1541 
1542 		if (dsl_prop_get_integer(dsname,
1543 		    zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1544 			return (SET_ERROR(EACCES));
1545 		return (rdonly ? 0 : EACCES);
1546 	}
1547 	return (SET_ERROR(EACCES));
1548 }
1549 
1550 static int
1551 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct statvfs64 *statp,
1552     uint32_t bshift)
1553 {
1554 	char buf[20 + DMU_OBJACCT_PREFIX_LEN];
1555 	uint64_t offset = DMU_OBJACCT_PREFIX_LEN;
1556 	uint64_t quota;
1557 	uint64_t used;
1558 	int err;
1559 
1560 	strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
1561 	err = id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset, B_FALSE);
1562 	if (err)
1563 		return (err);
1564 
1565 	if (zfsvfs->z_projectquota_obj == 0)
1566 		goto objs;
1567 
1568 	err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj,
1569 	    buf + offset, 8, 1, &quota);
1570 	if (err == ENOENT)
1571 		goto objs;
1572 	else if (err)
1573 		return (err);
1574 
1575 	err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1576 	    buf + offset, 8, 1, &used);
1577 	if (unlikely(err == ENOENT)) {
1578 		uint32_t blksize;
1579 		u_longlong_t nblocks;
1580 
1581 		/*
1582 		 * Quota accounting is async, so it is possible race case.
1583 		 * There is at least one object with the given project ID.
1584 		 */
1585 		sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
1586 		if (unlikely(zp->z_blksz == 0))
1587 			blksize = zfsvfs->z_max_blksz;
1588 
1589 		used = blksize * nblocks;
1590 	} else if (err) {
1591 		return (err);
1592 	}
1593 
1594 	statp->f_blocks = quota >> bshift;
1595 	statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0;
1596 	statp->f_bavail = statp->f_bfree;
1597 
1598 objs:
1599 	if (zfsvfs->z_projectobjquota_obj == 0)
1600 		return (0);
1601 
1602 	err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj,
1603 	    buf + offset, 8, 1, &quota);
1604 	if (err == ENOENT)
1605 		return (0);
1606 	else if (err)
1607 		return (err);
1608 
1609 	err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1610 	    buf, 8, 1, &used);
1611 	if (unlikely(err == ENOENT)) {
1612 		/*
1613 		 * Quota accounting is async, so it is possible race case.
1614 		 * There is at least one object with the given project ID.
1615 		 */
1616 		used = 1;
1617 	} else if (err) {
1618 		return (err);
1619 	}
1620 
1621 	statp->f_files = quota;
1622 	statp->f_ffree = (quota > used) ? (quota - used) : 0;
1623 
1624 	return (0);
1625 }
1626 
1627 /*
1628  * Determine whether the mount is allowed according to MAC check.
1629  * by comparing (where appropriate) label of the dataset against
1630  * the label of the zone being mounted into.  If the dataset has
1631  * no label, create one.
1632  *
1633  * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1634  */
1635 static int
1636 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1637 {
1638 	int		error, retv;
1639 	zone_t		*mntzone = NULL;
1640 	ts_label_t	*mnt_tsl;
1641 	bslabel_t	*mnt_sl;
1642 	bslabel_t	ds_sl;
1643 	char		ds_hexsl[MAXNAMELEN];
1644 
1645 	retv = EACCES;				/* assume the worst */
1646 
1647 	/*
1648 	 * Start by getting the dataset label if it exists.
1649 	 */
1650 	error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1651 	    1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1652 	if (error)
1653 		return (SET_ERROR(EACCES));
1654 
1655 	/*
1656 	 * If labeling is NOT enabled, then disallow the mount of datasets
1657 	 * which have a non-default label already.  No other label checks
1658 	 * are needed.
1659 	 */
1660 	if (!is_system_labeled()) {
1661 		if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1662 			return (0);
1663 		return (SET_ERROR(EACCES));
1664 	}
1665 
1666 	/*
1667 	 * Get the label of the mountpoint.  If mounting into the global
1668 	 * zone (i.e. mountpoint is not within an active zone and the
1669 	 * zoned property is off), the label must be default or
1670 	 * admin_low/admin_high only; no other checks are needed.
1671 	 */
1672 	mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1673 	if (mntzone->zone_id == GLOBAL_ZONEID) {
1674 		uint64_t zoned;
1675 
1676 		zone_rele(mntzone);
1677 
1678 		if (dsl_prop_get_integer(osname,
1679 		    zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1680 			return (SET_ERROR(EACCES));
1681 		if (!zoned)
1682 			return (zfs_check_global_label(osname, ds_hexsl));
1683 		else
1684 			/*
1685 			 * This is the case of a zone dataset being mounted
1686 			 * initially, before the zone has been fully created;
1687 			 * allow this mount into global zone.
1688 			 */
1689 			return (0);
1690 	}
1691 
1692 	mnt_tsl = mntzone->zone_slabel;
1693 	ASSERT(mnt_tsl != NULL);
1694 	label_hold(mnt_tsl);
1695 	mnt_sl = label2bslabel(mnt_tsl);
1696 
1697 	if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1698 		/*
1699 		 * The dataset doesn't have a real label, so fabricate one.
1700 		 */
1701 		char *str = NULL;
1702 
1703 		if (l_to_str_internal(mnt_sl, &str) == 0 &&
1704 		    dsl_prop_set_string(osname,
1705 		    zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1706 		    ZPROP_SRC_LOCAL, str) == 0)
1707 			retv = 0;
1708 		if (str != NULL)
1709 			kmem_free(str, strlen(str) + 1);
1710 	} else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1711 		/*
1712 		 * Now compare labels to complete the MAC check.  If the
1713 		 * labels are equal then allow access.  If the mountpoint
1714 		 * label dominates the dataset label, allow readonly access.
1715 		 * Otherwise, access is denied.
1716 		 */
1717 		if (blequal(mnt_sl, &ds_sl))
1718 			retv = 0;
1719 		else if (bldominates(mnt_sl, &ds_sl)) {
1720 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1721 			retv = 0;
1722 		}
1723 	}
1724 
1725 	label_rele(mnt_tsl);
1726 	zone_rele(mntzone);
1727 	return (retv);
1728 }
1729 
1730 /*
1731  * Load a string-valued boot property and attempt to convert it to a 64-bit
1732  * unsigned integer.  If the value is not present, or the conversion fails,
1733  * return the provided default value.
1734  */
1735 static uint64_t
1736 spa_get_bootprop_uint64(const char *name, uint64_t defval)
1737 {
1738 	char *propval;
1739 	u_longlong_t r;
1740 	int e;
1741 
1742 	if ((propval = spa_get_bootprop(name)) == NULL) {
1743 		/*
1744 		 * The property does not exist.
1745 		 */
1746 		return (defval);
1747 	}
1748 
1749 	e = ddi_strtoull(propval, NULL, 10, &r);
1750 
1751 	spa_free_bootprop(propval);
1752 
1753 	/*
1754 	 * If the conversion succeeded, return the value.  If there was any
1755 	 * kind of failure, just return the default value.
1756 	 */
1757 	return (e == 0 ? r : defval);
1758 }
1759 
1760 static int
1761 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1762 {
1763 	int error = 0;
1764 	static int zfsrootdone = 0;
1765 	zfsvfs_t *zfsvfs = NULL;
1766 	znode_t *zp = NULL;
1767 	vnode_t *vp = NULL;
1768 	char *zfs_bootfs;
1769 	char *zfs_devid;
1770 	char *zfs_rootdisk_path;
1771 	uint64_t zfs_bootpool;
1772 	uint64_t zfs_bootvdev;
1773 
1774 	ASSERT(vfsp);
1775 
1776 	/*
1777 	 * The filesystem that we mount as root is defined in the
1778 	 * boot property "zfs-bootfs" with a format of
1779 	 * "poolname/root-dataset-objnum".
1780 	 */
1781 	if (why == ROOT_INIT) {
1782 		if (zfsrootdone++)
1783 			return (SET_ERROR(EBUSY));
1784 
1785 		/*
1786 		 * the process of doing a spa_load will require the
1787 		 * clock to be set before we could (for example) do
1788 		 * something better by looking at the timestamp on
1789 		 * an uberblock, so just set it to -1.
1790 		 */
1791 		clkset(-1);
1792 
1793 		if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1794 			cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1795 			    "bootfs name");
1796 			return (SET_ERROR(EINVAL));
1797 		}
1798 		zfs_devid = spa_get_bootprop("diskdevid");
1799 
1800 		/*
1801 		 * The boot loader may also provide us with the GUID for both
1802 		 * the pool and the nominated boot vdev.  A GUID value of 0 is
1803 		 * explicitly invalid (see "spa_change_guid()"), so we use this
1804 		 * as a sentinel value when no GUID is present.
1805 		 */
1806 		zfs_bootpool = spa_get_bootprop_uint64("zfs-bootpool", 0);
1807 		zfs_bootvdev = spa_get_bootprop_uint64("zfs-bootvdev", 0);
1808 
1809 		/*
1810 		 * If we have been given a root disk override path, we want to
1811 		 * ignore device paths from the pool configuration and use only
1812 		 * the specific path we were given in the boot properties.
1813 		 */
1814 		zfs_rootdisk_path = spa_get_bootprop("zfs-rootdisk-path");
1815 
1816 		/*
1817 		 * Initialise the early boot device rescan mechanism.  A scan
1818 		 * will not actually be performed unless we need to do so in
1819 		 * order to find the correct /devices path for a relocated
1820 		 * device.
1821 		 */
1822 		vdev_disk_preroot_init(zfs_rootdisk_path);
1823 
1824 		error = spa_import_rootpool(rootfs.bo_name, zfs_devid,
1825 		    zfs_bootpool, zfs_bootvdev);
1826 
1827 		spa_free_bootprop(zfs_devid);
1828 
1829 		if (error != 0) {
1830 			spa_free_bootprop(zfs_bootfs);
1831 			spa_free_bootprop(zfs_rootdisk_path);
1832 			vdev_disk_preroot_fini();
1833 			cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1834 			    error);
1835 			return (error);
1836 		}
1837 
1838 		if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1839 			spa_free_bootprop(zfs_bootfs);
1840 			spa_free_bootprop(zfs_rootdisk_path);
1841 			vdev_disk_preroot_fini();
1842 			cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1843 			    error);
1844 			return (error);
1845 		}
1846 
1847 		spa_free_bootprop(zfs_bootfs);
1848 		spa_free_bootprop(zfs_rootdisk_path);
1849 
1850 		if ((error = vfs_lock(vfsp)) != 0) {
1851 			vdev_disk_preroot_fini();
1852 			return (error);
1853 		}
1854 
1855 		if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1856 			cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1857 			goto out;
1858 		}
1859 
1860 		/* zfs_domount has already cached the root vnode for us */
1861 		zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1862 		ASSERT(zfsvfs);
1863 		ASSERT(zfsvfs->z_rootdir);
1864 
1865 		vp = zfsvfs->z_rootdir;
1866 		mutex_enter(&vp->v_lock);
1867 		vp->v_flag |= VROOT;
1868 		mutex_exit(&vp->v_lock);
1869 
1870 		/*
1871 		 * Leave rootvp held.  The root file system is never unmounted.
1872 		 */
1873 		VN_HOLD(vp);
1874 		rootvp = vp;
1875 
1876 		vfs_add((struct vnode *)0, vfsp,
1877 		    (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1878 out:
1879 		vdev_disk_preroot_fini();
1880 		vfs_unlock(vfsp);
1881 		return (error);
1882 	} else if (why == ROOT_REMOUNT) {
1883 		readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1884 		vfsp->vfs_flag |= VFS_REMOUNT;
1885 
1886 		/* refresh mount options */
1887 		zfs_unregister_callbacks(vfsp->vfs_data);
1888 		return (zfs_register_callbacks(vfsp));
1889 
1890 	} else if (why == ROOT_UNMOUNT) {
1891 		zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1892 		(void) zfs_sync(vfsp, 0, 0);
1893 		return (0);
1894 	}
1895 
1896 	/*
1897 	 * if "why" is equal to anything else other than ROOT_INIT,
1898 	 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1899 	 */
1900 	return (SET_ERROR(ENOTSUP));
1901 }
1902 
1903 /*ARGSUSED*/
1904 static int
1905 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1906 {
1907 	char		*osname;
1908 	pathname_t	spn;
1909 	int		error = 0;
1910 	uio_seg_t	fromspace = (uap->flags & MS_SYSSPACE) ?
1911 	    UIO_SYSSPACE : UIO_USERSPACE;
1912 	int		canwrite;
1913 
1914 	if (mvp->v_type != VDIR)
1915 		return (SET_ERROR(ENOTDIR));
1916 
1917 	mutex_enter(&mvp->v_lock);
1918 	if ((uap->flags & MS_REMOUNT) == 0 &&
1919 	    (uap->flags & MS_OVERLAY) == 0 &&
1920 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1921 		mutex_exit(&mvp->v_lock);
1922 		return (SET_ERROR(EBUSY));
1923 	}
1924 	mutex_exit(&mvp->v_lock);
1925 
1926 	/*
1927 	 * ZFS does not support passing unparsed data in via MS_DATA.
1928 	 * Users should use the MS_OPTIONSTR interface; this means
1929 	 * that all option parsing is already done and the options struct
1930 	 * can be interrogated.
1931 	 */
1932 	if ((uap->flags & MS_DATA) && uap->datalen > 0)
1933 		return (SET_ERROR(EINVAL));
1934 
1935 	/*
1936 	 * Get the objset name (the "special" mount argument).
1937 	 */
1938 	if (error = pn_get(uap->spec, fromspace, &spn))
1939 		return (error);
1940 
1941 	osname = spn.pn_path;
1942 
1943 	/*
1944 	 * Check for mount privilege?
1945 	 *
1946 	 * If we don't have privilege then see if
1947 	 * we have local permission to allow it
1948 	 */
1949 	error = secpolicy_fs_mount(cr, mvp, vfsp);
1950 	if (error) {
1951 		if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1952 			vattr_t		vattr;
1953 
1954 			/*
1955 			 * Make sure user is the owner of the mount point
1956 			 * or has sufficient privileges.
1957 			 */
1958 
1959 			vattr.va_mask = AT_UID;
1960 
1961 			if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1962 				goto out;
1963 			}
1964 
1965 			if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1966 			    VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1967 				goto out;
1968 			}
1969 			secpolicy_fs_mount_clearopts(cr, vfsp);
1970 		} else {
1971 			goto out;
1972 		}
1973 	}
1974 
1975 	/*
1976 	 * Refuse to mount a filesystem if we are in a local zone and the
1977 	 * dataset is not visible.
1978 	 */
1979 	if (!INGLOBALZONE(curproc) &&
1980 	    (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1981 		error = SET_ERROR(EPERM);
1982 		goto out;
1983 	}
1984 
1985 	error = zfs_mount_label_policy(vfsp, osname);
1986 	if (error)
1987 		goto out;
1988 
1989 	/*
1990 	 * When doing a remount, we simply refresh our temporary properties
1991 	 * according to those options set in the current VFS options.
1992 	 */
1993 	if (uap->flags & MS_REMOUNT) {
1994 		/* refresh mount options */
1995 		zfs_unregister_callbacks(vfsp->vfs_data);
1996 		error = zfs_register_callbacks(vfsp);
1997 		goto out;
1998 	}
1999 
2000 	error = zfs_domount(vfsp, osname);
2001 
2002 	/*
2003 	 * Add an extra VFS_HOLD on our parent vfs so that it can't
2004 	 * disappear due to a forced unmount.
2005 	 */
2006 	if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
2007 		VFS_HOLD(mvp->v_vfsp);
2008 
2009 out:
2010 	pn_free(&spn);
2011 	return (error);
2012 }
2013 
2014 static int
2015 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
2016 {
2017 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2018 	dev32_t d32;
2019 	uint64_t refdbytes, availbytes, usedobjs, availobjs;
2020 	int err = 0;
2021 
2022 	ZFS_ENTER(zfsvfs);
2023 
2024 	dmu_objset_space(zfsvfs->z_os,
2025 	    &refdbytes, &availbytes, &usedobjs, &availobjs);
2026 
2027 	/*
2028 	 * The underlying storage pool actually uses multiple block sizes.
2029 	 * We report the fragsize as the smallest block size we support,
2030 	 * and we report our blocksize as the filesystem's maximum blocksize.
2031 	 */
2032 	statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
2033 	statp->f_bsize = zfsvfs->z_max_blksz;
2034 
2035 	/*
2036 	 * The following report "total" blocks of various kinds in the
2037 	 * file system, but reported in terms of f_frsize - the
2038 	 * "fragment" size.
2039 	 */
2040 
2041 	statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
2042 	statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
2043 	statp->f_bavail = statp->f_bfree; /* no root reservation */
2044 
2045 	/*
2046 	 * statvfs() should really be called statufs(), because it assumes
2047 	 * static metadata.  ZFS doesn't preallocate files, so the best
2048 	 * we can do is report the max that could possibly fit in f_files,
2049 	 * and that minus the number actually used in f_ffree.
2050 	 * For f_ffree, report the smaller of the number of object available
2051 	 * and the number of blocks (each object will take at least a block).
2052 	 */
2053 	statp->f_ffree = MIN(availobjs, statp->f_bfree);
2054 	statp->f_favail = statp->f_ffree;	/* no "root reservation" */
2055 	statp->f_files = statp->f_ffree + usedobjs;
2056 
2057 	(void) cmpldev(&d32, vfsp->vfs_dev);
2058 	statp->f_fsid = d32;
2059 
2060 	/*
2061 	 * We're a zfs filesystem.
2062 	 */
2063 	(void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
2064 
2065 	statp->f_flag = vf_to_stf(vfsp->vfs_flag);
2066 
2067 	statp->f_namemax = MAXNAMELEN - 1;
2068 
2069 	/*
2070 	 * We have all of 32 characters to stuff a string here.
2071 	 * Is there anything useful we could/should provide?
2072 	 */
2073 	bzero(statp->f_fstr, sizeof (statp->f_fstr));
2074 
2075 	if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
2076 	    dmu_objset_projectquota_present(zfsvfs->z_os)) {
2077 		znode_t *zp;
2078 
2079 		/*
2080 		 * In ZoL, zfs_statvfs is passed a Linux dentry (directory
2081 		 * entry), instead of a vfsp. The ZoL code uses the dentry
2082 		 * to get the znode from the dentry's inode. This represents
2083 		 * whatever filename was passed to the user-level statvfs
2084 		 * syscall.
2085 		 *
2086 		 * We're using the VFS root znode here, so this represents a
2087 		 * potential difference from ZoL.
2088 		 */
2089 		if (zfs_zget(zfsvfs, zfsvfs->z_root, &zp) == 0) {
2090 			uint32_t bshift = ddi_fls(statp->f_bsize) - 1;
2091 
2092 			if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid &&
2093 			    zpl_is_valid_projid(zp->z_projid))
2094 				err = zfs_statfs_project(zfsvfs, zp, statp,
2095 				    bshift);
2096 			VN_RELE(ZTOV(zp));
2097 		}
2098 	}
2099 
2100 	ZFS_EXIT(zfsvfs);
2101 	return (err);
2102 }
2103 
2104 static int
2105 zfs_root(vfs_t *vfsp, vnode_t **vpp)
2106 {
2107 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2108 	struct vnode *vp;
2109 	int error;
2110 
2111 	ZFS_ENTER(zfsvfs);
2112 
2113 	vp = zfsvfs->z_rootdir;
2114 	if (vp != NULL) {
2115 		VN_HOLD(vp);
2116 		error = 0;
2117 	} else {
2118 		/* forced unmount */
2119 		error = EIO;
2120 	}
2121 	*vpp = vp;
2122 
2123 	ZFS_EXIT(zfsvfs);
2124 	return (error);
2125 
2126 }
2127 
2128 /*
2129  * Teardown the zfsvfs::z_os.
2130  *
2131  * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
2132  * and 'z_teardown_inactive_lock' held.
2133  */
2134 static int
2135 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
2136 {
2137 	znode_t	*zp;
2138 
2139 	zfs_unlinked_drain_stop_wait(zfsvfs);
2140 
2141 	rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
2142 
2143 	if (!unmounting) {
2144 		/*
2145 		 * We purge the parent filesystem's vfsp as the parent
2146 		 * filesystem and all of its snapshots have their vnode's
2147 		 * v_vfsp set to the parent's filesystem's vfsp.  Note,
2148 		 * 'z_parent' is self referential for non-snapshots.
2149 		 */
2150 		(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2151 	}
2152 
2153 	/*
2154 	 * Close the zil. NB: Can't close the zil while zfs_inactive
2155 	 * threads are blocked as zil_close can call zfs_inactive.
2156 	 */
2157 	if (zfsvfs->z_log) {
2158 		zil_close(zfsvfs->z_log);
2159 		zfsvfs->z_log = NULL;
2160 	}
2161 
2162 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
2163 
2164 	/*
2165 	 * If we are not unmounting (ie: online recv) and someone already
2166 	 * unmounted this file system while we were doing the switcheroo,
2167 	 * or a reopen of z_os failed then just bail out now.
2168 	 */
2169 	if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
2170 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
2171 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2172 		return (SET_ERROR(EIO));
2173 	}
2174 
2175 	/*
2176 	 * At this point there are no vops active, and any new vops will
2177 	 * fail with EIO since we have z_teardown_lock for writer (only
2178 	 * relavent for forced unmount).
2179 	 *
2180 	 * Release all holds on dbufs.
2181 	 */
2182 	mutex_enter(&zfsvfs->z_znodes_lock);
2183 	for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
2184 	    zp = list_next(&zfsvfs->z_all_znodes, zp))
2185 		if (zp->z_sa_hdl) {
2186 			ASSERT(ZTOV(zp)->v_count > 0);
2187 			zfs_znode_dmu_fini(zp);
2188 		}
2189 	mutex_exit(&zfsvfs->z_znodes_lock);
2190 
2191 	/*
2192 	 * If we are unmounting, set the unmounted flag and let new vops
2193 	 * unblock.  zfs_inactive will have the unmounted behavior, and all
2194 	 * other vops will fail with EIO.
2195 	 */
2196 	if (unmounting) {
2197 		/*
2198 		 * Clear the cached root vnode now that we are unmounted.
2199 		 * Its release must be performed outside the teardown locks to
2200 		 * avoid recursive lock entry via zfs_inactive().
2201 		 */
2202 		vnode_t *vp = zfsvfs->z_rootdir;
2203 		zfsvfs->z_rootdir = NULL;
2204 
2205 		zfsvfs->z_unmounted = B_TRUE;
2206 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
2207 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2208 
2209 		/* Drop the cached root vp now that it is safe */
2210 		VN_RELE(vp);
2211 	}
2212 
2213 	/*
2214 	 * z_os will be NULL if there was an error in attempting to reopen
2215 	 * zfsvfs, so just return as the properties had already been
2216 	 * unregistered and cached data had been evicted before.
2217 	 */
2218 	if (zfsvfs->z_os == NULL)
2219 		return (0);
2220 
2221 	/*
2222 	 * Unregister properties.
2223 	 */
2224 	zfs_unregister_callbacks(zfsvfs);
2225 
2226 	/*
2227 	 * Evict cached data
2228 	 */
2229 	if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
2230 	    !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
2231 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
2232 	dmu_objset_evict_dbufs(zfsvfs->z_os);
2233 
2234 	return (0);
2235 }
2236 
2237 /*ARGSUSED*/
2238 static int
2239 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
2240 {
2241 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2242 	objset_t *os;
2243 	int ret;
2244 
2245 	ret = secpolicy_fs_unmount(cr, vfsp);
2246 	if (ret) {
2247 		if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
2248 		    ZFS_DELEG_PERM_MOUNT, cr))
2249 			return (ret);
2250 	}
2251 
2252 	/*
2253 	 * We purge the parent filesystem's vfsp as the parent filesystem
2254 	 * and all of its snapshots have their vnode's v_vfsp set to the
2255 	 * parent's filesystem's vfsp.  Note, 'z_parent' is self
2256 	 * referential for non-snapshots.
2257 	 */
2258 	(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2259 
2260 	/*
2261 	 * Unmount any snapshots mounted under .zfs before unmounting the
2262 	 * dataset itself.
2263 	 */
2264 	if (zfsvfs->z_ctldir != NULL &&
2265 	    (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
2266 		return (ret);
2267 	}
2268 
2269 	if (!(fflag & MS_FORCE)) {
2270 		/*
2271 		 * Check the number of active vnodes in the file system.
2272 		 * Our count is maintained in the vfs structure, but the
2273 		 * number is off by 1 to indicate a hold on the vfs
2274 		 * structure itself.
2275 		 */
2276 		boolean_t draining;
2277 		uint_t thresh = 1;
2278 		vnode_t *ctlvp, *rvp;
2279 
2280 		/*
2281 		 * The cached vnode for the root directory of the mount also
2282 		 * maintains a hold on the vfs structure.
2283 		 */
2284 		rvp = zfsvfs->z_rootdir;
2285 		thresh++;
2286 
2287 		/*
2288 		 * The '.zfs' directory maintains a reference of its own, and
2289 		 * any active references underneath are reflected in the vnode
2290 		 * count. Allow one additional reference for it.
2291 		 */
2292 		ctlvp = zfsvfs->z_ctldir;
2293 		if (ctlvp != NULL) {
2294 			thresh++;
2295 		}
2296 
2297 		/*
2298 		 * If it's running, the asynchronous unlinked drain task needs
2299 		 * to be stopped before the number of active vnodes can be
2300 		 * reliably checked.
2301 		 */
2302 		draining = zfsvfs->z_draining;
2303 		if (draining)
2304 			zfs_unlinked_drain_stop_wait(zfsvfs);
2305 
2306 		if (vfsp->vfs_count > thresh || rvp->v_count > 1 ||
2307 		    (ctlvp != NULL && ctlvp->v_count > 1)) {
2308 			if (draining) {
2309 				/* If it was draining, restart the task */
2310 				zfs_unlinked_drain(zfsvfs);
2311 			}
2312 			return (SET_ERROR(EBUSY));
2313 		}
2314 	}
2315 
2316 	vfsp->vfs_flag |= VFS_UNMOUNTED;
2317 
2318 	VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2319 	os = zfsvfs->z_os;
2320 
2321 	/*
2322 	 * z_os will be NULL if there was an error in
2323 	 * attempting to reopen zfsvfs.
2324 	 */
2325 	if (os != NULL) {
2326 		/*
2327 		 * Unset the objset user_ptr.
2328 		 */
2329 		mutex_enter(&os->os_user_ptr_lock);
2330 		dmu_objset_set_user(os, NULL);
2331 		mutex_exit(&os->os_user_ptr_lock);
2332 
2333 		/*
2334 		 * Finally release the objset
2335 		 */
2336 		dmu_objset_disown(os, B_TRUE, zfsvfs);
2337 	}
2338 
2339 	/*
2340 	 * We can now safely destroy the '.zfs' directory node.
2341 	 */
2342 	if (zfsvfs->z_ctldir != NULL)
2343 		zfsctl_destroy(zfsvfs);
2344 
2345 	return (0);
2346 }
2347 
2348 static int
2349 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
2350 {
2351 	zfsvfs_t	*zfsvfs = vfsp->vfs_data;
2352 	znode_t		*zp;
2353 	uint64_t	object = 0;
2354 	uint64_t	fid_gen = 0;
2355 	uint64_t	gen_mask;
2356 	uint64_t	zp_gen;
2357 	int		i, err;
2358 
2359 	*vpp = NULL;
2360 
2361 	ZFS_ENTER(zfsvfs);
2362 
2363 	if (fidp->fid_len == LONG_FID_LEN) {
2364 		zfid_long_t	*zlfid = (zfid_long_t *)fidp;
2365 		uint64_t	objsetid = 0;
2366 		uint64_t	setgen = 0;
2367 
2368 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2369 			objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2370 
2371 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2372 			setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2373 
2374 		ZFS_EXIT(zfsvfs);
2375 
2376 		err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2377 		if (err)
2378 			return (SET_ERROR(EINVAL));
2379 		ZFS_ENTER(zfsvfs);
2380 	}
2381 
2382 	if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2383 		zfid_short_t	*zfid = (zfid_short_t *)fidp;
2384 
2385 		for (i = 0; i < sizeof (zfid->zf_object); i++)
2386 			object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2387 
2388 		for (i = 0; i < sizeof (zfid->zf_gen); i++)
2389 			fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2390 	} else {
2391 		ZFS_EXIT(zfsvfs);
2392 		return (SET_ERROR(EINVAL));
2393 	}
2394 
2395 	/* A zero fid_gen means we are in the .zfs control directories */
2396 	if (fid_gen == 0 &&
2397 	    (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
2398 		*vpp = zfsvfs->z_ctldir;
2399 		ASSERT(*vpp != NULL);
2400 		if (object == ZFSCTL_INO_SNAPDIR) {
2401 			VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2402 			    0, NULL, NULL, NULL, NULL, NULL) == 0);
2403 		} else {
2404 			VN_HOLD(*vpp);
2405 		}
2406 		ZFS_EXIT(zfsvfs);
2407 		return (0);
2408 	}
2409 
2410 	gen_mask = -1ULL >> (64 - 8 * i);
2411 
2412 	dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2413 	if (err = zfs_zget(zfsvfs, object, &zp)) {
2414 		ZFS_EXIT(zfsvfs);
2415 		return (err);
2416 	}
2417 	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2418 	    sizeof (uint64_t));
2419 	zp_gen = zp_gen & gen_mask;
2420 	if (zp_gen == 0)
2421 		zp_gen = 1;
2422 	if (zp->z_unlinked || zp_gen != fid_gen) {
2423 		dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2424 		VN_RELE(ZTOV(zp));
2425 		ZFS_EXIT(zfsvfs);
2426 		return (SET_ERROR(EINVAL));
2427 	}
2428 
2429 	*vpp = ZTOV(zp);
2430 	ZFS_EXIT(zfsvfs);
2431 	return (0);
2432 }
2433 
2434 /*
2435  * Block out VOPs and close zfsvfs_t::z_os
2436  *
2437  * Note, if successful, then we return with the 'z_teardown_lock' and
2438  * 'z_teardown_inactive_lock' write held.  We leave ownership of the underlying
2439  * dataset and objset intact so that they can be atomically handed off during
2440  * a subsequent rollback or recv operation and the resume thereafter.
2441  */
2442 int
2443 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2444 {
2445 	int error;
2446 
2447 	if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2448 		return (error);
2449 
2450 	return (0);
2451 }
2452 
2453 /*
2454  * Rebuild SA and release VOPs.  Note that ownership of the underlying dataset
2455  * is an invariant across any of the operations that can be performed while the
2456  * filesystem was suspended.  Whether it succeeded or failed, the preconditions
2457  * are the same: the relevant objset and associated dataset are owned by
2458  * zfsvfs, held, and long held on entry.
2459  */
2460 int
2461 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2462 {
2463 	int err;
2464 	znode_t *zp;
2465 
2466 	ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2467 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2468 
2469 	/*
2470 	 * We already own this, so just update the objset_t, as the one we
2471 	 * had before may have been evicted.
2472 	 */
2473 	objset_t *os;
2474 	VERIFY3P(ds->ds_owner, ==, zfsvfs);
2475 	VERIFY(dsl_dataset_long_held(ds));
2476 	VERIFY0(dmu_objset_from_ds(ds, &os));
2477 
2478 	err = zfsvfs_init(zfsvfs, os);
2479 	if (err != 0)
2480 		goto bail;
2481 
2482 	VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2483 
2484 	zfs_set_fuid_feature(zfsvfs);
2485 
2486 	/*
2487 	 * Attempt to re-establish all the active znodes with
2488 	 * their dbufs.  If a zfs_rezget() fails, then we'll let
2489 	 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2490 	 * when they try to use their znode.
2491 	 */
2492 	mutex_enter(&zfsvfs->z_znodes_lock);
2493 	for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2494 	    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2495 		(void) zfs_rezget(zp);
2496 	}
2497 	mutex_exit(&zfsvfs->z_znodes_lock);
2498 
2499 	if (((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) &&
2500 	    !zfsvfs->z_unmounted) {
2501 		/*
2502 		 * zfs_suspend_fs() could have interrupted freeing
2503 		 * of dnodes. We need to restart this freeing so
2504 		 * that we don't "leak" the space.
2505 		 */
2506 		zfs_unlinked_drain(zfsvfs);
2507 	}
2508 
2509 bail:
2510 	/* release the VOPs */
2511 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
2512 	rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2513 
2514 	if (err) {
2515 		/*
2516 		 * Since we couldn't setup the sa framework, try to force
2517 		 * unmount this file system.
2518 		 */
2519 		if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2520 			(void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2521 	}
2522 	return (err);
2523 }
2524 
2525 static void
2526 zfs_freevfs(vfs_t *vfsp)
2527 {
2528 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2529 
2530 	/*
2531 	 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2532 	 * from zfs_mount().  Release it here.  If we came through
2533 	 * zfs_mountroot() instead, we didn't grab an extra hold, so
2534 	 * skip the VFS_RELE for rootvfs.
2535 	 */
2536 	if (zfsvfs->z_issnap && (vfsp != rootvfs))
2537 		VFS_RELE(zfsvfs->z_parent->z_vfs);
2538 
2539 	zfsvfs_free(zfsvfs);
2540 
2541 	atomic_dec_32(&zfs_active_fs_count);
2542 }
2543 
2544 /*
2545  * VFS_INIT() initialization.  Note that there is no VFS_FINI(),
2546  * so we can't safely do any non-idempotent initialization here.
2547  * Leave that to zfs_init() and zfs_fini(), which are called
2548  * from the module's _init() and _fini() entry points.
2549  */
2550 /*ARGSUSED*/
2551 static int
2552 zfs_vfsinit(int fstype, char *name)
2553 {
2554 	int error;
2555 
2556 	zfsfstype = fstype;
2557 
2558 	/*
2559 	 * Setup vfsops and vnodeops tables.
2560 	 */
2561 	error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2562 	if (error != 0) {
2563 		cmn_err(CE_WARN, "zfs: bad vfs ops template");
2564 	}
2565 
2566 	error = zfs_create_op_tables();
2567 	if (error) {
2568 		zfs_remove_op_tables();
2569 		cmn_err(CE_WARN, "zfs: bad vnode ops template");
2570 		(void) vfs_freevfsops_by_type(zfsfstype);
2571 		return (error);
2572 	}
2573 
2574 	mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2575 
2576 	/*
2577 	 * Unique major number for all zfs mounts.
2578 	 * If we run out of 32-bit minors, we'll getudev() another major.
2579 	 */
2580 	zfs_major = ddi_name_to_major(ZFS_DRIVER);
2581 	zfs_minor = ZFS_MIN_MINOR;
2582 
2583 	return (0);
2584 }
2585 
2586 void
2587 zfs_init(void)
2588 {
2589 	/*
2590 	 * Initialize .zfs directory structures
2591 	 */
2592 	zfsctl_init();
2593 
2594 	/*
2595 	 * Initialize znode cache, vnode ops, etc...
2596 	 */
2597 	zfs_znode_init();
2598 
2599 	dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2600 }
2601 
2602 void
2603 zfs_fini(void)
2604 {
2605 	zfsctl_fini();
2606 	zfs_znode_fini();
2607 }
2608 
2609 int
2610 zfs_busy(void)
2611 {
2612 	return (zfs_active_fs_count != 0);
2613 }
2614 
2615 int
2616 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2617 {
2618 	int error;
2619 	objset_t *os = zfsvfs->z_os;
2620 	dmu_tx_t *tx;
2621 
2622 	if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2623 		return (SET_ERROR(EINVAL));
2624 
2625 	if (newvers < zfsvfs->z_version)
2626 		return (SET_ERROR(EINVAL));
2627 
2628 	if (zfs_spa_version_map(newvers) >
2629 	    spa_version(dmu_objset_spa(zfsvfs->z_os)))
2630 		return (SET_ERROR(ENOTSUP));
2631 
2632 	tx = dmu_tx_create(os);
2633 	dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2634 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2635 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2636 		    ZFS_SA_ATTRS);
2637 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2638 	}
2639 	error = dmu_tx_assign(tx, TXG_WAIT);
2640 	if (error) {
2641 		dmu_tx_abort(tx);
2642 		return (error);
2643 	}
2644 
2645 	error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2646 	    8, 1, &newvers, tx);
2647 
2648 	if (error) {
2649 		dmu_tx_commit(tx);
2650 		return (error);
2651 	}
2652 
2653 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2654 		uint64_t sa_obj;
2655 
2656 		ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2657 		    SPA_VERSION_SA);
2658 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2659 		    DMU_OT_NONE, 0, tx);
2660 
2661 		error = zap_add(os, MASTER_NODE_OBJ,
2662 		    ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2663 		ASSERT0(error);
2664 
2665 		VERIFY(0 == sa_set_sa_object(os, sa_obj));
2666 		sa_register_update_callback(os, zfs_sa_upgrade);
2667 	}
2668 
2669 	spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2670 	    "from %llu to %llu", zfsvfs->z_version, newvers);
2671 
2672 	dmu_tx_commit(tx);
2673 
2674 	zfsvfs->z_version = newvers;
2675 	os->os_version = newvers;
2676 
2677 	zfs_set_fuid_feature(zfsvfs);
2678 
2679 	return (0);
2680 }
2681 
2682 /*
2683  * Read a property stored within the master node.
2684  */
2685 int
2686 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2687 {
2688 	uint64_t *cached_copy = NULL;
2689 
2690 	/*
2691 	 * Figure out where in the objset_t the cached copy would live, if it
2692 	 * is available for the requested property.
2693 	 */
2694 	if (os != NULL) {
2695 		switch (prop) {
2696 		case ZFS_PROP_VERSION:
2697 			cached_copy = &os->os_version;
2698 			break;
2699 		case ZFS_PROP_NORMALIZE:
2700 			cached_copy = &os->os_normalization;
2701 			break;
2702 		case ZFS_PROP_UTF8ONLY:
2703 			cached_copy = &os->os_utf8only;
2704 			break;
2705 		case ZFS_PROP_CASE:
2706 			cached_copy = &os->os_casesensitivity;
2707 			break;
2708 		default:
2709 			break;
2710 		}
2711 	}
2712 	if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) {
2713 		*value = *cached_copy;
2714 		return (0);
2715 	}
2716 
2717 	/*
2718 	 * If the property wasn't cached, look up the file system's value for
2719 	 * the property. For the version property, we look up a slightly
2720 	 * different string.
2721 	 */
2722 	const char *pname;
2723 	int error = ENOENT;
2724 	if (prop == ZFS_PROP_VERSION) {
2725 		pname = ZPL_VERSION_STR;
2726 	} else {
2727 		pname = zfs_prop_to_name(prop);
2728 	}
2729 
2730 	if (os != NULL) {
2731 		ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2732 		error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2733 	}
2734 
2735 	if (error == ENOENT) {
2736 		/* No value set, use the default value */
2737 		switch (prop) {
2738 		case ZFS_PROP_VERSION:
2739 			*value = ZPL_VERSION;
2740 			break;
2741 		case ZFS_PROP_NORMALIZE:
2742 		case ZFS_PROP_UTF8ONLY:
2743 			*value = 0;
2744 			break;
2745 		case ZFS_PROP_CASE:
2746 			*value = ZFS_CASE_SENSITIVE;
2747 			break;
2748 		default:
2749 			return (error);
2750 		}
2751 		error = 0;
2752 	}
2753 
2754 	/*
2755 	 * If one of the methods for getting the property value above worked,
2756 	 * copy it into the objset_t's cache.
2757 	 */
2758 	if (error == 0 && cached_copy != NULL) {
2759 		*cached_copy = *value;
2760 	}
2761 
2762 	return (error);
2763 }
2764 
2765 /*
2766  * Return true if the coresponding vfs's unmounted flag is set.
2767  * Otherwise return false.
2768  * If this function returns true we know VFS unmount has been initiated.
2769  */
2770 boolean_t
2771 zfs_get_vfs_flag_unmounted(objset_t *os)
2772 {
2773 	zfsvfs_t *zfvp;
2774 	boolean_t unmounted = B_FALSE;
2775 
2776 	ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2777 
2778 	mutex_enter(&os->os_user_ptr_lock);
2779 	zfvp = dmu_objset_get_user(os);
2780 	if (zfvp != NULL && zfvp->z_vfs != NULL &&
2781 	    (zfvp->z_vfs->vfs_flag & VFS_UNMOUNTED))
2782 		unmounted = B_TRUE;
2783 	mutex_exit(&os->os_user_ptr_lock);
2784 
2785 	return (unmounted);
2786 }
2787 
2788 static vfsdef_t vfw = {
2789 	VFSDEF_VERSION,
2790 	MNTTYPE_ZFS,
2791 	zfs_vfsinit,
2792 	VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2793 	    VSW_XID|VSW_ZMOUNT,
2794 	&zfs_mntopts
2795 };
2796 
2797 struct modlfs zfs_modlfs = {
2798 	&mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2799 };
2800