xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_vfsops.c (revision 0250c53ad267726f2438e3c6556199a0bbf588a2)
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 void
406 acl_implicit_changed_cb(void *arg, uint64_t newval)
407 {
408 	zfsvfs_t *zfsvfs = arg;
409 
410 	zfsvfs->z_acl_implicit = (boolean_t)newval;
411 }
412 
413 static int
414 zfs_register_callbacks(vfs_t *vfsp)
415 {
416 	struct dsl_dataset *ds = NULL;
417 	objset_t *os = NULL;
418 	zfsvfs_t *zfsvfs = NULL;
419 	uint64_t nbmand;
420 	boolean_t readonly = B_FALSE;
421 	boolean_t do_readonly = B_FALSE;
422 	boolean_t setuid = B_FALSE;
423 	boolean_t do_setuid = B_FALSE;
424 	boolean_t exec = B_FALSE;
425 	boolean_t do_exec = B_FALSE;
426 	boolean_t devices = B_FALSE;
427 	boolean_t do_devices = B_FALSE;
428 	boolean_t xattr = B_FALSE;
429 	boolean_t do_xattr = B_FALSE;
430 	boolean_t atime = B_FALSE;
431 	boolean_t do_atime = B_FALSE;
432 	int error = 0;
433 
434 	ASSERT(vfsp);
435 	zfsvfs = vfsp->vfs_data;
436 	ASSERT(zfsvfs);
437 	os = zfsvfs->z_os;
438 
439 	/*
440 	 * The act of registering our callbacks will destroy any mount
441 	 * options we may have.  In order to enable temporary overrides
442 	 * of mount options, we stash away the current values and
443 	 * restore them after we register the callbacks.
444 	 */
445 	if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
446 	    !spa_writeable(dmu_objset_spa(os))) {
447 		readonly = B_TRUE;
448 		do_readonly = B_TRUE;
449 	} else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
450 		readonly = B_FALSE;
451 		do_readonly = B_TRUE;
452 	}
453 	if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
454 		devices = B_FALSE;
455 		setuid = B_FALSE;
456 		do_devices = B_TRUE;
457 		do_setuid = B_TRUE;
458 	} else {
459 		if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
460 			devices = B_FALSE;
461 			do_devices = B_TRUE;
462 		} else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
463 			devices = B_TRUE;
464 			do_devices = B_TRUE;
465 		}
466 
467 		if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
468 			setuid = B_FALSE;
469 			do_setuid = B_TRUE;
470 		} else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
471 			setuid = B_TRUE;
472 			do_setuid = B_TRUE;
473 		}
474 	}
475 	if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
476 		exec = B_FALSE;
477 		do_exec = B_TRUE;
478 	} else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
479 		exec = B_TRUE;
480 		do_exec = B_TRUE;
481 	}
482 	if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
483 		xattr = B_FALSE;
484 		do_xattr = B_TRUE;
485 	} else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
486 		xattr = B_TRUE;
487 		do_xattr = B_TRUE;
488 	}
489 	if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
490 		atime = B_FALSE;
491 		do_atime = B_TRUE;
492 	} else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
493 		atime = B_TRUE;
494 		do_atime = B_TRUE;
495 	}
496 
497 	/*
498 	 * nbmand is a special property.  It can only be changed at
499 	 * mount time.
500 	 *
501 	 * This is weird, but it is documented to only be changeable
502 	 * at mount time.
503 	 */
504 	if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
505 		nbmand = B_FALSE;
506 	} else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
507 		nbmand = B_TRUE;
508 	} else {
509 		char osname[ZFS_MAX_DATASET_NAME_LEN];
510 
511 		dmu_objset_name(os, osname);
512 		if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
513 		    NULL)) {
514 			return (error);
515 		}
516 	}
517 
518 	/*
519 	 * Register property callbacks.
520 	 *
521 	 * It would probably be fine to just check for i/o error from
522 	 * the first prop_register(), but I guess I like to go
523 	 * overboard...
524 	 */
525 	ds = dmu_objset_ds(os);
526 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
527 	error = dsl_prop_register(ds,
528 	    zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
529 	error = error ? error : dsl_prop_register(ds,
530 	    zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
531 	error = error ? error : dsl_prop_register(ds,
532 	    zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
533 	error = error ? error : dsl_prop_register(ds,
534 	    zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
535 	error = error ? error : dsl_prop_register(ds,
536 	    zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
537 	error = error ? error : dsl_prop_register(ds,
538 	    zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
539 	error = error ? error : dsl_prop_register(ds,
540 	    zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
541 	error = error ? error : dsl_prop_register(ds,
542 	    zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
543 	error = error ? error : dsl_prop_register(ds,
544 	    zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
545 	error = error ? error : dsl_prop_register(ds,
546 	    zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
547 	    zfsvfs);
548 	error = error ? error : dsl_prop_register(ds,
549 	    zfs_prop_to_name(ZFS_PROP_ACLIMPLICIT),
550 	    acl_implicit_changed_cb, zfsvfs);
551 	error = error ? error : dsl_prop_register(ds,
552 	    zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
553 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
554 	if (error)
555 		goto unregister;
556 
557 	/*
558 	 * Invoke our callbacks to restore temporary mount options.
559 	 */
560 	if (do_readonly)
561 		readonly_changed_cb(zfsvfs, readonly);
562 	if (do_setuid)
563 		setuid_changed_cb(zfsvfs, setuid);
564 	if (do_exec)
565 		exec_changed_cb(zfsvfs, exec);
566 	if (do_devices)
567 		devices_changed_cb(zfsvfs, devices);
568 	if (do_xattr)
569 		xattr_changed_cb(zfsvfs, xattr);
570 	if (do_atime)
571 		atime_changed_cb(zfsvfs, atime);
572 
573 	nbmand_changed_cb(zfsvfs, nbmand);
574 
575 	return (0);
576 
577 unregister:
578 	dsl_prop_unregister_all(ds, zfsvfs);
579 	return (error);
580 }
581 
582 static int
583 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
584     uint64_t *userp, uint64_t *groupp, uint64_t *projectp)
585 {
586 	sa_hdr_phys_t sa;
587 	sa_hdr_phys_t *sap = data;
588 	uint64_t flags;
589 	int hdrsize;
590 	boolean_t swap = B_FALSE;
591 
592 	/*
593 	 * Is it a valid type of object to track?
594 	 */
595 	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
596 		return (SET_ERROR(ENOENT));
597 
598 	/*
599 	 * If we have a NULL data pointer
600 	 * then assume the id's aren't changing and
601 	 * return EEXIST to the dmu to let it know to
602 	 * use the same ids
603 	 */
604 	if (data == NULL)
605 		return (SET_ERROR(EEXIST));
606 
607 	if (bonustype == DMU_OT_ZNODE) {
608 		znode_phys_t *znp = data;
609 		*userp = znp->zp_uid;
610 		*groupp = znp->zp_gid;
611 		*projectp = ZFS_DEFAULT_PROJID;
612 		return (0);
613 	}
614 
615 	if (sap->sa_magic == 0) {
616 		/*
617 		 * This should only happen for newly created files
618 		 * that haven't had the znode data filled in yet.
619 		 */
620 		*userp = 0;
621 		*groupp = 0;
622 		*projectp = ZFS_DEFAULT_PROJID;
623 		return (0);
624 	}
625 
626 	sa = *sap;
627 	if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
628 		sa.sa_magic = SA_MAGIC;
629 		sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
630 		swap = B_TRUE;
631 	} else {
632 		VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
633 	}
634 
635 	hdrsize = sa_hdrsize(&sa);
636 	VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
637 
638 	*userp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_UID_OFFSET));
639 	*groupp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_GID_OFFSET));
640 	flags = *((uint64_t *)((uintptr_t)data + hdrsize + SA_FLAGS_OFFSET));
641 	if (swap)
642 		flags = BSWAP_64(flags);
643 
644 	if (flags & ZFS_PROJID)
645 		*projectp = *((uint64_t *)((uintptr_t)data + hdrsize +
646 		    SA_PROJID_OFFSET));
647 	else
648 		*projectp = ZFS_DEFAULT_PROJID;
649 
650 	if (swap) {
651 		*userp = BSWAP_64(*userp);
652 		*groupp = BSWAP_64(*groupp);
653 		*projectp = BSWAP_64(*projectp);
654 	}
655 	return (0);
656 }
657 
658 static void
659 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
660     char *domainbuf, int buflen, uid_t *ridp)
661 {
662 	uint64_t fuid;
663 	const char *domain;
664 
665 	fuid = zfs_strtonum(fuidstr, NULL);
666 
667 	domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
668 	if (domain)
669 		(void) strlcpy(domainbuf, domain, buflen);
670 	else
671 		domainbuf[0] = '\0';
672 	*ridp = FUID_RID(fuid);
673 }
674 
675 static uint64_t
676 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
677 {
678 	switch (type) {
679 	case ZFS_PROP_USERUSED:
680 	case ZFS_PROP_USEROBJUSED:
681 		return (DMU_USERUSED_OBJECT);
682 	case ZFS_PROP_GROUPUSED:
683 	case ZFS_PROP_GROUPOBJUSED:
684 		return (DMU_GROUPUSED_OBJECT);
685 	case ZFS_PROP_PROJECTUSED:
686 	case ZFS_PROP_PROJECTOBJUSED:
687 		return (DMU_PROJECTUSED_OBJECT);
688 	case ZFS_PROP_USERQUOTA:
689 		return (zfsvfs->z_userquota_obj);
690 	case ZFS_PROP_GROUPQUOTA:
691 		return (zfsvfs->z_groupquota_obj);
692 	case ZFS_PROP_USEROBJQUOTA:
693 		return (zfsvfs->z_userobjquota_obj);
694 	case ZFS_PROP_GROUPOBJQUOTA:
695 		return (zfsvfs->z_groupobjquota_obj);
696 	case ZFS_PROP_PROJECTQUOTA:
697 		return (zfsvfs->z_projectquota_obj);
698 	case ZFS_PROP_PROJECTOBJQUOTA:
699 		return (zfsvfs->z_projectobjquota_obj);
700 	default:
701 		return (ZFS_NO_OBJECT);
702 	}
703 }
704 
705 int
706 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
707     uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
708 {
709 	int error;
710 	zap_cursor_t zc;
711 	zap_attribute_t za;
712 	zfs_useracct_t *buf = vbuf;
713 	uint64_t obj;
714 	int offset = 0;
715 
716 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
717 		return (SET_ERROR(ENOTSUP));
718 
719 	if ((type == ZFS_PROP_PROJECTQUOTA || type == ZFS_PROP_PROJECTUSED ||
720 	    type == ZFS_PROP_PROJECTOBJQUOTA ||
721 	    type == ZFS_PROP_PROJECTOBJUSED) &&
722 	    !dmu_objset_projectquota_present(zfsvfs->z_os))
723 		return (SET_ERROR(ENOTSUP));
724 
725 	if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
726 	    type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA ||
727 	    type == ZFS_PROP_PROJECTOBJUSED ||
728 	    type == ZFS_PROP_PROJECTOBJQUOTA) &&
729 	    !dmu_objset_userobjspace_present(zfsvfs->z_os))
730 		return (SET_ERROR(ENOTSUP));
731 
732 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
733 	if (obj == ZFS_NO_OBJECT) {
734 		*bufsizep = 0;
735 		return (0);
736 	}
737 
738 	if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
739 	    type == ZFS_PROP_PROJECTOBJUSED)
740 		offset = DMU_OBJACCT_PREFIX_LEN;
741 
742 	for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
743 	    (error = zap_cursor_retrieve(&zc, &za)) == 0;
744 	    zap_cursor_advance(&zc)) {
745 		if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
746 		    *bufsizep)
747 			break;
748 
749 		/*
750 		 * skip object quota (with zap name prefix DMU_OBJACCT_PREFIX)
751 		 * when dealing with block quota and vice versa.
752 		 */
753 		if ((offset > 0) != (strncmp(za.za_name, DMU_OBJACCT_PREFIX,
754 		    DMU_OBJACCT_PREFIX_LEN) == 0))
755 			continue;
756 
757 		fuidstr_to_sid(zfsvfs, za.za_name + offset,
758 		    buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
759 
760 		buf->zu_space = za.za_first_integer;
761 		buf++;
762 	}
763 	if (error == ENOENT)
764 		error = 0;
765 
766 	ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
767 	*bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
768 	*cookiep = zap_cursor_serialize(&zc);
769 	zap_cursor_fini(&zc);
770 	return (error);
771 }
772 
773 /*
774  * buf must be big enough (eg, 32 bytes)
775  */
776 static int
777 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
778     char *buf, boolean_t addok)
779 {
780 	uint64_t fuid;
781 	int domainid = 0;
782 
783 	if (domain && domain[0]) {
784 		domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
785 		if (domainid == -1)
786 			return (SET_ERROR(ENOENT));
787 	}
788 	fuid = FUID_ENCODE(domainid, rid);
789 	(void) sprintf(buf, "%llx", (longlong_t)fuid);
790 	return (0);
791 }
792 
793 int
794 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
795     const char *domain, uint64_t rid, uint64_t *valp)
796 {
797 	char buf[20 + DMU_OBJACCT_PREFIX_LEN];
798 	int offset = 0;
799 	int err;
800 	uint64_t obj;
801 
802 	*valp = 0;
803 
804 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
805 		return (SET_ERROR(ENOTSUP));
806 
807 	if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
808 	    type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA ||
809 	    type == ZFS_PROP_PROJECTOBJUSED ||
810 	    type == ZFS_PROP_PROJECTOBJQUOTA) &&
811 	    !dmu_objset_userobjspace_present(zfsvfs->z_os))
812 		return (SET_ERROR(ENOTSUP));
813 
814 	if (type == ZFS_PROP_PROJECTQUOTA || type == ZFS_PROP_PROJECTUSED ||
815 	    type == ZFS_PROP_PROJECTOBJQUOTA ||
816 	    type == ZFS_PROP_PROJECTOBJUSED) {
817 		if (!dmu_objset_projectquota_present(zfsvfs->z_os))
818 			return (SET_ERROR(ENOTSUP));
819 		if (!zpl_is_valid_projid(rid))
820 			return (SET_ERROR(EINVAL));
821 	}
822 
823 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
824 	if (obj == ZFS_NO_OBJECT)
825 		return (0);
826 
827 	if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
828 	    type == ZFS_PROP_PROJECTOBJUSED) {
829 		strncpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN);
830 		offset = DMU_OBJACCT_PREFIX_LEN;
831 	}
832 
833 	err = id_to_fuidstr(zfsvfs, domain, rid, buf + offset, B_FALSE);
834 	if (err)
835 		return (err);
836 
837 	err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
838 	if (err == ENOENT)
839 		err = 0;
840 	return (err);
841 }
842 
843 int
844 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
845     const char *domain, uint64_t rid, uint64_t quota)
846 {
847 	char buf[32];
848 	int err;
849 	dmu_tx_t *tx;
850 	uint64_t *objp;
851 	boolean_t fuid_dirtied;
852 
853 	if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
854 		return (SET_ERROR(ENOTSUP));
855 
856 	switch (type) {
857 	case ZFS_PROP_USERQUOTA:
858 		objp = &zfsvfs->z_userquota_obj;
859 		break;
860 	case ZFS_PROP_GROUPQUOTA:
861 		objp = &zfsvfs->z_groupquota_obj;
862 		break;
863 	case ZFS_PROP_USEROBJQUOTA:
864 		objp = &zfsvfs->z_userobjquota_obj;
865 		break;
866 	case ZFS_PROP_GROUPOBJQUOTA:
867 		objp = &zfsvfs->z_groupobjquota_obj;
868 		break;
869 	case ZFS_PROP_PROJECTQUOTA:
870 		if (!dmu_objset_projectquota_enabled(zfsvfs->z_os))
871 			return (SET_ERROR(ENOTSUP));
872 		if (!zpl_is_valid_projid(rid))
873 			return (SET_ERROR(EINVAL));
874 
875 		objp = &zfsvfs->z_projectquota_obj;
876 		break;
877 	case ZFS_PROP_PROJECTOBJQUOTA:
878 		if (!dmu_objset_projectquota_enabled(zfsvfs->z_os))
879 			return (SET_ERROR(ENOTSUP));
880 		if (!zpl_is_valid_projid(rid))
881 			return (SET_ERROR(EINVAL));
882 
883 		objp = &zfsvfs->z_projectobjquota_obj;
884 		break;
885 	default:
886 		return (SET_ERROR(EINVAL));
887 	}
888 
889 	err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
890 	if (err)
891 		return (err);
892 	fuid_dirtied = zfsvfs->z_fuid_dirty;
893 
894 	tx = dmu_tx_create(zfsvfs->z_os);
895 	dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
896 	if (*objp == 0) {
897 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
898 		    zfs_userquota_prop_prefixes[type]);
899 	}
900 	if (fuid_dirtied)
901 		zfs_fuid_txhold(zfsvfs, tx);
902 	err = dmu_tx_assign(tx, TXG_WAIT);
903 	if (err) {
904 		dmu_tx_abort(tx);
905 		return (err);
906 	}
907 
908 	mutex_enter(&zfsvfs->z_lock);
909 	if (*objp == 0) {
910 		*objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
911 		    DMU_OT_NONE, 0, tx);
912 		VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
913 		    zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
914 	}
915 	mutex_exit(&zfsvfs->z_lock);
916 
917 	if (quota == 0) {
918 		err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
919 		if (err == ENOENT)
920 			err = 0;
921 	} else {
922 		err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
923 	}
924 	ASSERT(err == 0);
925 	if (fuid_dirtied)
926 		zfs_fuid_sync(zfsvfs, tx);
927 	dmu_tx_commit(tx);
928 	return (err);
929 }
930 
931 boolean_t
932 zfs_id_overobjquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
933 {
934 	char buf[20 + DMU_OBJACCT_PREFIX_LEN];
935 	uint64_t used, quota, quotaobj;
936 	int err;
937 
938 	if (!dmu_objset_userobjspace_present(zfsvfs->z_os)) {
939 		if (dmu_objset_userobjspace_upgradable(zfsvfs->z_os)) {
940 			dsl_pool_config_enter(
941 			    dmu_objset_pool(zfsvfs->z_os), FTAG);
942 			dmu_objset_id_quota_upgrade(zfsvfs->z_os);
943 			dsl_pool_config_exit(
944 			    dmu_objset_pool(zfsvfs->z_os), FTAG);
945 		}
946 		return (B_FALSE);
947 	}
948 
949 	if (usedobj == DMU_PROJECTUSED_OBJECT) {
950 		if (!dmu_objset_projectquota_present(zfsvfs->z_os)) {
951 			if (dmu_objset_projectquota_upgradable(zfsvfs->z_os)) {
952 				dsl_pool_config_enter(
953 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
954 				dmu_objset_id_quota_upgrade(zfsvfs->z_os);
955 				dsl_pool_config_exit(
956 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
957 			}
958 			return (B_FALSE);
959 		}
960 		quotaobj = zfsvfs->z_projectobjquota_obj;
961 	} else if (usedobj == DMU_USERUSED_OBJECT) {
962 		quotaobj = zfsvfs->z_userobjquota_obj;
963 	} else if (usedobj == DMU_GROUPUSED_OBJECT) {
964 		quotaobj = zfsvfs->z_groupobjquota_obj;
965 	} else {
966 		return (B_FALSE);
967 	}
968 	if (quotaobj == 0 || zfsvfs->z_replay)
969 		return (B_FALSE);
970 
971 	(void) sprintf(buf, "%llx", (longlong_t)id);
972 	err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
973 	if (err != 0)
974 		return (B_FALSE);
975 
976 	(void) sprintf(buf, DMU_OBJACCT_PREFIX "%llx", (longlong_t)id);
977 	err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
978 	if (err != 0)
979 		return (B_FALSE);
980 	return (used >= quota);
981 }
982 
983 boolean_t
984 zfs_id_overblockquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
985 {
986 	char buf[20];
987 	uint64_t used, quota, quotaobj;
988 	int err;
989 
990 	if (usedobj == DMU_PROJECTUSED_OBJECT) {
991 		if (!dmu_objset_projectquota_present(zfsvfs->z_os)) {
992 			if (dmu_objset_projectquota_upgradable(zfsvfs->z_os)) {
993 				dsl_pool_config_enter(
994 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
995 				dmu_objset_id_quota_upgrade(zfsvfs->z_os);
996 				dsl_pool_config_exit(
997 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
998 			}
999 			return (B_FALSE);
1000 		}
1001 		quotaobj = zfsvfs->z_projectquota_obj;
1002 	} else if (usedobj == DMU_USERUSED_OBJECT) {
1003 		quotaobj = zfsvfs->z_userquota_obj;
1004 	} else if (usedobj == DMU_GROUPUSED_OBJECT) {
1005 		quotaobj = zfsvfs->z_groupquota_obj;
1006 	} else {
1007 		return (B_FALSE);
1008 	}
1009 	if (quotaobj == 0 || zfsvfs->z_replay)
1010 		return (B_FALSE);
1011 
1012 	(void) sprintf(buf, "%llx", (longlong_t)id);
1013 	err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
1014 	if (err != 0)
1015 		return (B_FALSE);
1016 
1017 	err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
1018 	if (err != 0)
1019 		return (B_FALSE);
1020 	return (used >= quota);
1021 }
1022 
1023 boolean_t
1024 zfs_id_overquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
1025 {
1026 	return (zfs_id_overblockquota(zfsvfs, usedobj, id) ||
1027 	    zfs_id_overobjquota(zfsvfs, usedobj, id));
1028 }
1029 
1030 /*
1031  * Associate this zfsvfs with the given objset, which must be owned.
1032  * This will cache a bunch of on-disk state from the objset in the
1033  * zfsvfs.
1034  */
1035 static int
1036 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
1037 {
1038 	int error;
1039 	uint64_t val;
1040 
1041 	zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
1042 	zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
1043 	zfsvfs->z_os = os;
1044 
1045 	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
1046 	if (error != 0)
1047 		return (error);
1048 	if (zfsvfs->z_version >
1049 	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
1050 		(void) printf("Can't mount a version %lld file system "
1051 		    "on a version %lld pool\n. Pool must be upgraded to mount "
1052 		    "this file system.", (u_longlong_t)zfsvfs->z_version,
1053 		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
1054 		return (SET_ERROR(ENOTSUP));
1055 	}
1056 	error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
1057 	if (error != 0)
1058 		return (error);
1059 	zfsvfs->z_norm = (int)val;
1060 
1061 	error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
1062 	if (error != 0)
1063 		return (error);
1064 	zfsvfs->z_utf8 = (val != 0);
1065 
1066 	error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
1067 	if (error != 0)
1068 		return (error);
1069 	zfsvfs->z_case = (uint_t)val;
1070 
1071 	/*
1072 	 * Fold case on file systems that are always or sometimes case
1073 	 * insensitive.
1074 	 */
1075 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
1076 	    zfsvfs->z_case == ZFS_CASE_MIXED)
1077 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1078 
1079 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1080 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1081 
1082 	uint64_t sa_obj = 0;
1083 	if (zfsvfs->z_use_sa) {
1084 		/* should either have both of these objects or none */
1085 		error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
1086 		    &sa_obj);
1087 		if (error != 0)
1088 			return (error);
1089 	}
1090 
1091 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1092 	    &zfsvfs->z_attr_table);
1093 	if (error != 0)
1094 		return (error);
1095 
1096 	if (zfsvfs->z_version >= ZPL_VERSION_SA)
1097 		sa_register_update_callback(os, zfs_sa_upgrade);
1098 
1099 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
1100 	    &zfsvfs->z_root);
1101 	if (error != 0)
1102 		return (error);
1103 	ASSERT(zfsvfs->z_root != 0);
1104 
1105 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
1106 	    &zfsvfs->z_unlinkedobj);
1107 	if (error != 0)
1108 		return (error);
1109 
1110 	error = zap_lookup(os, MASTER_NODE_OBJ,
1111 	    zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
1112 	    8, 1, &zfsvfs->z_userquota_obj);
1113 	if (error == ENOENT)
1114 		zfsvfs->z_userquota_obj = 0;
1115 	else if (error != 0)
1116 		return (error);
1117 
1118 	error = zap_lookup(os, MASTER_NODE_OBJ,
1119 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
1120 	    8, 1, &zfsvfs->z_groupquota_obj);
1121 	if (error == ENOENT)
1122 		zfsvfs->z_groupquota_obj = 0;
1123 	else if (error != 0)
1124 		return (error);
1125 
1126 	error = zap_lookup(os, MASTER_NODE_OBJ,
1127 	    zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA],
1128 	    8, 1, &zfsvfs->z_projectquota_obj);
1129 	if (error == ENOENT)
1130 		zfsvfs->z_projectquota_obj = 0;
1131 	else if (error != 0)
1132 		return (error);
1133 
1134 	error = zap_lookup(os, MASTER_NODE_OBJ,
1135 	    zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
1136 	    8, 1, &zfsvfs->z_userobjquota_obj);
1137 	if (error == ENOENT)
1138 		zfsvfs->z_userobjquota_obj = 0;
1139 	else if (error != 0)
1140 		return (error);
1141 
1142 	error = zap_lookup(os, MASTER_NODE_OBJ,
1143 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
1144 	    8, 1, &zfsvfs->z_groupobjquota_obj);
1145 	if (error == ENOENT)
1146 		zfsvfs->z_groupobjquota_obj = 0;
1147 	else if (error != 0)
1148 		return (error);
1149 
1150 	error = zap_lookup(os, MASTER_NODE_OBJ,
1151 	    zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA],
1152 	    8, 1, &zfsvfs->z_projectobjquota_obj);
1153 	if (error == ENOENT)
1154 		zfsvfs->z_projectobjquota_obj = 0;
1155 	else if (error != 0)
1156 		return (error);
1157 
1158 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
1159 	    &zfsvfs->z_fuid_obj);
1160 	if (error == ENOENT)
1161 		zfsvfs->z_fuid_obj = 0;
1162 	else if (error != 0)
1163 		return (error);
1164 
1165 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
1166 	    &zfsvfs->z_shares_dir);
1167 	if (error == ENOENT)
1168 		zfsvfs->z_shares_dir = 0;
1169 	else if (error != 0)
1170 		return (error);
1171 
1172 	return (0);
1173 }
1174 
1175 int
1176 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp)
1177 {
1178 	objset_t *os;
1179 	zfsvfs_t *zfsvfs;
1180 	int error;
1181 	boolean_t ro = (readonly || (strchr(osname, '@') != NULL));
1182 
1183 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1184 
1185 	error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os);
1186 	if (error != 0) {
1187 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
1188 		return (error);
1189 	}
1190 
1191 	error = zfsvfs_create_impl(zfvp, zfsvfs, os);
1192 	if (error != 0) {
1193 		dmu_objset_disown(os, B_TRUE, zfsvfs);
1194 	}
1195 	return (error);
1196 }
1197 
1198 
1199 int
1200 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
1201 {
1202 	int error;
1203 
1204 	zfsvfs->z_vfs = NULL;
1205 	zfsvfs->z_parent = zfsvfs;
1206 
1207 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1208 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1209 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1210 	    offsetof(znode_t, z_link_node));
1211 	rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1212 	rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1213 	rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1214 	for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1215 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1216 
1217 	error = zfsvfs_init(zfsvfs, os);
1218 	if (error != 0) {
1219 		*zfvp = NULL;
1220 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
1221 		return (error);
1222 	}
1223 
1224 	zfsvfs->z_drain_task = TASKQID_INVALID;
1225 	zfsvfs->z_draining = B_FALSE;
1226 	zfsvfs->z_drain_cancel = B_TRUE;
1227 
1228 	*zfvp = zfsvfs;
1229 	return (0);
1230 }
1231 
1232 static int
1233 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1234 {
1235 	int error;
1236 
1237 	error = zfs_register_callbacks(zfsvfs->z_vfs);
1238 	if (error)
1239 		return (error);
1240 
1241 	zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1242 
1243 	/*
1244 	 * If we are not mounting (ie: online recv), then we don't
1245 	 * have to worry about replaying the log as we blocked all
1246 	 * operations out since we closed the ZIL.
1247 	 */
1248 	if (mounting) {
1249 		boolean_t readonly;
1250 
1251 		/*
1252 		 * During replay we remove the read only flag to
1253 		 * allow replays to succeed.
1254 		 */
1255 		readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1256 		if (readonly != 0) {
1257 			zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1258 		} else {
1259 			zfs_unlinked_drain(zfsvfs);
1260 		}
1261 
1262 		/*
1263 		 * Parse and replay the intent log.
1264 		 *
1265 		 * Because of ziltest, this must be done after
1266 		 * zfs_unlinked_drain().  (Further note: ziltest
1267 		 * doesn't use readonly mounts, where
1268 		 * zfs_unlinked_drain() isn't called.)  This is because
1269 		 * ziltest causes spa_sync() to think it's committed,
1270 		 * but actually it is not, so the intent log contains
1271 		 * many txg's worth of changes.
1272 		 *
1273 		 * In particular, if object N is in the unlinked set in
1274 		 * the last txg to actually sync, then it could be
1275 		 * actually freed in a later txg and then reallocated
1276 		 * in a yet later txg.  This would write a "create
1277 		 * object N" record to the intent log.  Normally, this
1278 		 * would be fine because the spa_sync() would have
1279 		 * written out the fact that object N is free, before
1280 		 * we could write the "create object N" intent log
1281 		 * record.
1282 		 *
1283 		 * But when we are in ziltest mode, we advance the "open
1284 		 * txg" without actually spa_sync()-ing the changes to
1285 		 * disk.  So we would see that object N is still
1286 		 * allocated and in the unlinked set, and there is an
1287 		 * intent log record saying to allocate it.
1288 		 */
1289 		if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1290 			if (zil_replay_disable) {
1291 				zil_destroy(zfsvfs->z_log, B_FALSE);
1292 			} else {
1293 				zfsvfs->z_replay = B_TRUE;
1294 				zil_replay(zfsvfs->z_os, zfsvfs,
1295 				    zfs_replay_vector);
1296 				zfsvfs->z_replay = B_FALSE;
1297 			}
1298 		}
1299 
1300 		/* restore readonly bit */
1301 		if (readonly != 0)
1302 			zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
1303 	}
1304 
1305 	/*
1306 	 * Set the objset user_ptr to track its zfsvfs.
1307 	 */
1308 	mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1309 	dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1310 	mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1311 
1312 	return (0);
1313 }
1314 
1315 void
1316 zfsvfs_free(zfsvfs_t *zfsvfs)
1317 {
1318 	int i;
1319 	extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1320 
1321 	/*
1322 	 * This is a barrier to prevent the filesystem from going away in
1323 	 * zfs_znode_move() until we can safely ensure that the filesystem is
1324 	 * not unmounted. We consider the filesystem valid before the barrier
1325 	 * and invalid after the barrier.
1326 	 */
1327 	rw_enter(&zfsvfs_lock, RW_READER);
1328 	rw_exit(&zfsvfs_lock);
1329 
1330 	zfs_fuid_destroy(zfsvfs);
1331 
1332 	mutex_destroy(&zfsvfs->z_znodes_lock);
1333 	mutex_destroy(&zfsvfs->z_lock);
1334 	list_destroy(&zfsvfs->z_all_znodes);
1335 	rrm_destroy(&zfsvfs->z_teardown_lock);
1336 	rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1337 	rw_destroy(&zfsvfs->z_fuid_lock);
1338 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1339 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1340 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1341 }
1342 
1343 static void
1344 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1345 {
1346 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1347 	if (zfsvfs->z_vfs) {
1348 		if (zfsvfs->z_use_fuids) {
1349 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1350 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1351 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1352 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1353 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1354 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1355 		} else {
1356 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1357 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1358 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1359 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1360 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1361 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1362 		}
1363 	}
1364 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1365 }
1366 
1367 static int
1368 zfs_domount(vfs_t *vfsp, char *osname)
1369 {
1370 	dev_t mount_dev;
1371 	uint64_t recordsize, fsid_guid;
1372 	int error = 0;
1373 	zfsvfs_t *zfsvfs;
1374 	boolean_t readonly = vfsp->vfs_flag & VFS_RDONLY ? B_TRUE : B_FALSE;
1375 
1376 	ASSERT(vfsp);
1377 	ASSERT(osname);
1378 
1379 	error = zfsvfs_create(osname, readonly, &zfsvfs);
1380 	if (error)
1381 		return (error);
1382 	zfsvfs->z_vfs = vfsp;
1383 
1384 	/* Initialize the generic filesystem structure. */
1385 	vfsp->vfs_bcount = 0;
1386 	vfsp->vfs_data = NULL;
1387 
1388 	if (zfs_create_unique_device(&mount_dev) == -1) {
1389 		error = SET_ERROR(ENODEV);
1390 		goto out;
1391 	}
1392 	ASSERT(vfs_devismounted(mount_dev) == 0);
1393 
1394 	if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1395 	    NULL))
1396 		goto out;
1397 
1398 	vfsp->vfs_dev = mount_dev;
1399 	vfsp->vfs_fstype = zfsfstype;
1400 	vfsp->vfs_bsize = recordsize;
1401 	vfsp->vfs_flag |= VFS_NOTRUNC;
1402 	vfsp->vfs_data = zfsvfs;
1403 
1404 	/*
1405 	 * The fsid is 64 bits, composed of an 8-bit fs type, which
1406 	 * separates our fsid from any other filesystem types, and a
1407 	 * 56-bit objset unique ID.  The objset unique ID is unique to
1408 	 * all objsets open on this system, provided by unique_create().
1409 	 * The 8-bit fs type must be put in the low bits of fsid[1]
1410 	 * because that's where other Solaris filesystems put it.
1411 	 */
1412 	fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1413 	ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1414 	vfsp->vfs_fsid.val[0] = fsid_guid;
1415 	vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1416 	    zfsfstype & 0xFF;
1417 
1418 	/*
1419 	 * Set features for file system.
1420 	 */
1421 	zfs_set_fuid_feature(zfsvfs);
1422 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1423 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1424 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1425 		vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1426 	} else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1427 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1428 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1429 	}
1430 	vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1431 
1432 	if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1433 		uint64_t pval;
1434 
1435 		atime_changed_cb(zfsvfs, B_FALSE);
1436 		readonly_changed_cb(zfsvfs, B_TRUE);
1437 		if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1438 			goto out;
1439 		xattr_changed_cb(zfsvfs, pval);
1440 		zfsvfs->z_issnap = B_TRUE;
1441 		zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1442 
1443 		mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1444 		dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1445 		mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1446 	} else {
1447 		error = zfsvfs_setup(zfsvfs, B_TRUE);
1448 	}
1449 
1450 	/* cache the root vnode for this mount */
1451 	znode_t *rootzp;
1452 	if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp)) {
1453 		goto out;
1454 	}
1455 	zfsvfs->z_rootdir = ZTOV(rootzp);
1456 
1457 	if (!zfsvfs->z_issnap)
1458 		zfsctl_create(zfsvfs);
1459 out:
1460 	if (error) {
1461 		dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1462 		zfsvfs_free(zfsvfs);
1463 	} else {
1464 		atomic_inc_32(&zfs_active_fs_count);
1465 	}
1466 
1467 	return (error);
1468 }
1469 
1470 void
1471 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1472 {
1473 	objset_t *os = zfsvfs->z_os;
1474 
1475 	if (!dmu_objset_is_snapshot(os))
1476 		dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1477 }
1478 
1479 /*
1480  * Convert a decimal digit string to a uint64_t integer.
1481  */
1482 static int
1483 str_to_uint64(char *str, uint64_t *objnum)
1484 {
1485 	uint64_t num = 0;
1486 
1487 	while (*str) {
1488 		if (*str < '0' || *str > '9')
1489 			return (SET_ERROR(EINVAL));
1490 
1491 		num = num*10 + *str++ - '0';
1492 	}
1493 
1494 	*objnum = num;
1495 	return (0);
1496 }
1497 
1498 /*
1499  * The boot path passed from the boot loader is in the form of
1500  * "rootpool-name/root-filesystem-object-number'. Convert this
1501  * string to a dataset name: "rootpool-name/root-filesystem-name".
1502  */
1503 static int
1504 zfs_parse_bootfs(char *bpath, char *outpath)
1505 {
1506 	char *slashp;
1507 	uint64_t objnum;
1508 	int error;
1509 
1510 	if (*bpath == 0 || *bpath == '/')
1511 		return (SET_ERROR(EINVAL));
1512 
1513 	(void) strcpy(outpath, bpath);
1514 
1515 	slashp = strchr(bpath, '/');
1516 
1517 	/* if no '/', just return the pool name */
1518 	if (slashp == NULL) {
1519 		return (0);
1520 	}
1521 
1522 	/* if not a number, just return the root dataset name */
1523 	if (str_to_uint64(slashp+1, &objnum)) {
1524 		return (0);
1525 	}
1526 
1527 	*slashp = '\0';
1528 	error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1529 	*slashp = '/';
1530 
1531 	return (error);
1532 }
1533 
1534 /*
1535  * Check that the hex label string is appropriate for the dataset being
1536  * mounted into the global_zone proper.
1537  *
1538  * Return an error if the hex label string is not default or
1539  * admin_low/admin_high.  For admin_low labels, the corresponding
1540  * dataset must be readonly.
1541  */
1542 int
1543 zfs_check_global_label(const char *dsname, const char *hexsl)
1544 {
1545 	if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1546 		return (0);
1547 	if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1548 		return (0);
1549 	if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1550 		/* must be readonly */
1551 		uint64_t rdonly;
1552 
1553 		if (dsl_prop_get_integer(dsname,
1554 		    zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1555 			return (SET_ERROR(EACCES));
1556 		return (rdonly ? 0 : EACCES);
1557 	}
1558 	return (SET_ERROR(EACCES));
1559 }
1560 
1561 static int
1562 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct statvfs64 *statp,
1563     uint32_t bshift)
1564 {
1565 	char buf[20 + DMU_OBJACCT_PREFIX_LEN];
1566 	uint64_t offset = DMU_OBJACCT_PREFIX_LEN;
1567 	uint64_t quota;
1568 	uint64_t used;
1569 	int err;
1570 
1571 	strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
1572 	err = id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset, B_FALSE);
1573 	if (err)
1574 		return (err);
1575 
1576 	if (zfsvfs->z_projectquota_obj == 0)
1577 		goto objs;
1578 
1579 	err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj,
1580 	    buf + offset, 8, 1, &quota);
1581 	if (err == ENOENT)
1582 		goto objs;
1583 	else if (err)
1584 		return (err);
1585 
1586 	err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1587 	    buf + offset, 8, 1, &used);
1588 	if (unlikely(err == ENOENT)) {
1589 		uint32_t blksize;
1590 		u_longlong_t nblocks;
1591 
1592 		/*
1593 		 * Quota accounting is async, so it is possible race case.
1594 		 * There is at least one object with the given project ID.
1595 		 */
1596 		sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
1597 		if (unlikely(zp->z_blksz == 0))
1598 			blksize = zfsvfs->z_max_blksz;
1599 
1600 		used = blksize * nblocks;
1601 	} else if (err) {
1602 		return (err);
1603 	}
1604 
1605 	statp->f_blocks = quota >> bshift;
1606 	statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0;
1607 	statp->f_bavail = statp->f_bfree;
1608 
1609 objs:
1610 	if (zfsvfs->z_projectobjquota_obj == 0)
1611 		return (0);
1612 
1613 	err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj,
1614 	    buf + offset, 8, 1, &quota);
1615 	if (err == ENOENT)
1616 		return (0);
1617 	else if (err)
1618 		return (err);
1619 
1620 	err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1621 	    buf, 8, 1, &used);
1622 	if (unlikely(err == ENOENT)) {
1623 		/*
1624 		 * Quota accounting is async, so it is possible race case.
1625 		 * There is at least one object with the given project ID.
1626 		 */
1627 		used = 1;
1628 	} else if (err) {
1629 		return (err);
1630 	}
1631 
1632 	statp->f_files = quota;
1633 	statp->f_ffree = (quota > used) ? (quota - used) : 0;
1634 
1635 	return (0);
1636 }
1637 
1638 /*
1639  * Determine whether the mount is allowed according to MAC check.
1640  * by comparing (where appropriate) label of the dataset against
1641  * the label of the zone being mounted into.  If the dataset has
1642  * no label, create one.
1643  *
1644  * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1645  */
1646 static int
1647 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1648 {
1649 	int		error, retv;
1650 	zone_t		*mntzone = NULL;
1651 	ts_label_t	*mnt_tsl;
1652 	bslabel_t	*mnt_sl;
1653 	bslabel_t	ds_sl;
1654 	char		ds_hexsl[MAXNAMELEN];
1655 
1656 	retv = EACCES;				/* assume the worst */
1657 
1658 	/*
1659 	 * Start by getting the dataset label if it exists.
1660 	 */
1661 	error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1662 	    1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1663 	if (error)
1664 		return (SET_ERROR(EACCES));
1665 
1666 	/*
1667 	 * If labeling is NOT enabled, then disallow the mount of datasets
1668 	 * which have a non-default label already.  No other label checks
1669 	 * are needed.
1670 	 */
1671 	if (!is_system_labeled()) {
1672 		if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1673 			return (0);
1674 		return (SET_ERROR(EACCES));
1675 	}
1676 
1677 	/*
1678 	 * Get the label of the mountpoint.  If mounting into the global
1679 	 * zone (i.e. mountpoint is not within an active zone and the
1680 	 * zoned property is off), the label must be default or
1681 	 * admin_low/admin_high only; no other checks are needed.
1682 	 */
1683 	mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1684 	if (mntzone->zone_id == GLOBAL_ZONEID) {
1685 		uint64_t zoned;
1686 
1687 		zone_rele(mntzone);
1688 
1689 		if (dsl_prop_get_integer(osname,
1690 		    zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1691 			return (SET_ERROR(EACCES));
1692 		if (!zoned)
1693 			return (zfs_check_global_label(osname, ds_hexsl));
1694 		else
1695 			/*
1696 			 * This is the case of a zone dataset being mounted
1697 			 * initially, before the zone has been fully created;
1698 			 * allow this mount into global zone.
1699 			 */
1700 			return (0);
1701 	}
1702 
1703 	mnt_tsl = mntzone->zone_slabel;
1704 	ASSERT(mnt_tsl != NULL);
1705 	label_hold(mnt_tsl);
1706 	mnt_sl = label2bslabel(mnt_tsl);
1707 
1708 	if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1709 		/*
1710 		 * The dataset doesn't have a real label, so fabricate one.
1711 		 */
1712 		char *str = NULL;
1713 
1714 		if (l_to_str_internal(mnt_sl, &str) == 0 &&
1715 		    dsl_prop_set_string(osname,
1716 		    zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1717 		    ZPROP_SRC_LOCAL, str) == 0)
1718 			retv = 0;
1719 		if (str != NULL)
1720 			kmem_free(str, strlen(str) + 1);
1721 	} else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1722 		/*
1723 		 * Now compare labels to complete the MAC check.  If the
1724 		 * labels are equal then allow access.  If the mountpoint
1725 		 * label dominates the dataset label, allow readonly access.
1726 		 * Otherwise, access is denied.
1727 		 */
1728 		if (blequal(mnt_sl, &ds_sl))
1729 			retv = 0;
1730 		else if (bldominates(mnt_sl, &ds_sl)) {
1731 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1732 			retv = 0;
1733 		}
1734 	}
1735 
1736 	label_rele(mnt_tsl);
1737 	zone_rele(mntzone);
1738 	return (retv);
1739 }
1740 
1741 /*
1742  * Load a string-valued boot property and attempt to convert it to a 64-bit
1743  * unsigned integer.  If the value is not present, or the conversion fails,
1744  * return the provided default value.
1745  */
1746 static uint64_t
1747 spa_get_bootprop_uint64(const char *name, uint64_t defval)
1748 {
1749 	char *propval;
1750 	u_longlong_t r;
1751 	int e;
1752 
1753 	if ((propval = spa_get_bootprop(name)) == NULL) {
1754 		/*
1755 		 * The property does not exist.
1756 		 */
1757 		return (defval);
1758 	}
1759 
1760 	e = ddi_strtoull(propval, NULL, 10, &r);
1761 
1762 	spa_free_bootprop(propval);
1763 
1764 	/*
1765 	 * If the conversion succeeded, return the value.  If there was any
1766 	 * kind of failure, just return the default value.
1767 	 */
1768 	return (e == 0 ? r : defval);
1769 }
1770 
1771 static int
1772 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1773 {
1774 	int error = 0;
1775 	static int zfsrootdone = 0;
1776 	zfsvfs_t *zfsvfs = NULL;
1777 	znode_t *zp = NULL;
1778 	vnode_t *vp = NULL;
1779 	char *zfs_bootfs;
1780 	char *zfs_devid;
1781 	char *zfs_rootdisk_path;
1782 	uint64_t zfs_bootpool;
1783 	uint64_t zfs_bootvdev;
1784 
1785 	ASSERT(vfsp);
1786 
1787 	/*
1788 	 * The filesystem that we mount as root is defined in the
1789 	 * boot property "zfs-bootfs" with a format of
1790 	 * "poolname/root-dataset-objnum".
1791 	 */
1792 	if (why == ROOT_INIT) {
1793 		if (zfsrootdone++)
1794 			return (SET_ERROR(EBUSY));
1795 
1796 		/*
1797 		 * the process of doing a spa_load will require the
1798 		 * clock to be set before we could (for example) do
1799 		 * something better by looking at the timestamp on
1800 		 * an uberblock, so just set it to -1.
1801 		 */
1802 		clkset(-1);
1803 
1804 		if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1805 			cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1806 			    "bootfs name");
1807 			return (SET_ERROR(EINVAL));
1808 		}
1809 		zfs_devid = spa_get_bootprop("diskdevid");
1810 
1811 		/*
1812 		 * The boot loader may also provide us with the GUID for both
1813 		 * the pool and the nominated boot vdev.  A GUID value of 0 is
1814 		 * explicitly invalid (see "spa_change_guid()"), so we use this
1815 		 * as a sentinel value when no GUID is present.
1816 		 */
1817 		zfs_bootpool = spa_get_bootprop_uint64("zfs-bootpool", 0);
1818 		zfs_bootvdev = spa_get_bootprop_uint64("zfs-bootvdev", 0);
1819 
1820 		/*
1821 		 * If we have been given a root disk override path, we want to
1822 		 * ignore device paths from the pool configuration and use only
1823 		 * the specific path we were given in the boot properties.
1824 		 */
1825 		zfs_rootdisk_path = spa_get_bootprop("zfs-rootdisk-path");
1826 
1827 		/*
1828 		 * Initialise the early boot device rescan mechanism.  A scan
1829 		 * will not actually be performed unless we need to do so in
1830 		 * order to find the correct /devices path for a relocated
1831 		 * device.
1832 		 */
1833 		vdev_disk_preroot_init(zfs_rootdisk_path);
1834 
1835 		error = spa_import_rootpool(rootfs.bo_name, zfs_devid,
1836 		    zfs_bootpool, zfs_bootvdev);
1837 
1838 		spa_free_bootprop(zfs_devid);
1839 
1840 		if (error != 0) {
1841 			spa_free_bootprop(zfs_bootfs);
1842 			spa_free_bootprop(zfs_rootdisk_path);
1843 			vdev_disk_preroot_fini();
1844 			cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1845 			    error);
1846 			return (error);
1847 		}
1848 
1849 		if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1850 			spa_free_bootprop(zfs_bootfs);
1851 			spa_free_bootprop(zfs_rootdisk_path);
1852 			vdev_disk_preroot_fini();
1853 			cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1854 			    error);
1855 			return (error);
1856 		}
1857 
1858 		spa_free_bootprop(zfs_bootfs);
1859 		spa_free_bootprop(zfs_rootdisk_path);
1860 
1861 		if ((error = vfs_lock(vfsp)) != 0) {
1862 			vdev_disk_preroot_fini();
1863 			return (error);
1864 		}
1865 
1866 		if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1867 			cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1868 			goto out;
1869 		}
1870 
1871 		/* zfs_domount has already cached the root vnode for us */
1872 		zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1873 		ASSERT(zfsvfs);
1874 		ASSERT(zfsvfs->z_rootdir);
1875 
1876 		vp = zfsvfs->z_rootdir;
1877 		mutex_enter(&vp->v_lock);
1878 		vp->v_flag |= VROOT;
1879 		mutex_exit(&vp->v_lock);
1880 
1881 		/*
1882 		 * Leave rootvp held.  The root file system is never unmounted.
1883 		 */
1884 		VN_HOLD(vp);
1885 		rootvp = vp;
1886 
1887 		vfs_add((struct vnode *)0, vfsp,
1888 		    (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1889 out:
1890 		vdev_disk_preroot_fini();
1891 		vfs_unlock(vfsp);
1892 		return (error);
1893 	} else if (why == ROOT_REMOUNT) {
1894 		readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1895 		vfsp->vfs_flag |= VFS_REMOUNT;
1896 
1897 		/* refresh mount options */
1898 		zfs_unregister_callbacks(vfsp->vfs_data);
1899 		return (zfs_register_callbacks(vfsp));
1900 
1901 	} else if (why == ROOT_UNMOUNT) {
1902 		zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1903 		(void) zfs_sync(vfsp, 0, 0);
1904 		return (0);
1905 	}
1906 
1907 	/*
1908 	 * if "why" is equal to anything else other than ROOT_INIT,
1909 	 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1910 	 */
1911 	return (SET_ERROR(ENOTSUP));
1912 }
1913 
1914 /*ARGSUSED*/
1915 static int
1916 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1917 {
1918 	char		*osname;
1919 	pathname_t	spn;
1920 	int		error = 0;
1921 	uio_seg_t	fromspace = (uap->flags & MS_SYSSPACE) ?
1922 	    UIO_SYSSPACE : UIO_USERSPACE;
1923 	int		canwrite;
1924 
1925 	if (mvp->v_type != VDIR)
1926 		return (SET_ERROR(ENOTDIR));
1927 
1928 	mutex_enter(&mvp->v_lock);
1929 	if ((uap->flags & MS_REMOUNT) == 0 &&
1930 	    (uap->flags & MS_OVERLAY) == 0 &&
1931 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1932 		mutex_exit(&mvp->v_lock);
1933 		return (SET_ERROR(EBUSY));
1934 	}
1935 	mutex_exit(&mvp->v_lock);
1936 
1937 	/*
1938 	 * ZFS does not support passing unparsed data in via MS_DATA.
1939 	 * Users should use the MS_OPTIONSTR interface; this means
1940 	 * that all option parsing is already done and the options struct
1941 	 * can be interrogated.
1942 	 */
1943 	if ((uap->flags & MS_DATA) && uap->datalen > 0)
1944 		return (SET_ERROR(EINVAL));
1945 
1946 	/*
1947 	 * Get the objset name (the "special" mount argument).
1948 	 */
1949 	if (error = pn_get(uap->spec, fromspace, &spn))
1950 		return (error);
1951 
1952 	osname = spn.pn_path;
1953 
1954 	/*
1955 	 * Check for mount privilege?
1956 	 *
1957 	 * If we don't have privilege then see if
1958 	 * we have local permission to allow it
1959 	 */
1960 	error = secpolicy_fs_mount(cr, mvp, vfsp);
1961 	if (error) {
1962 		if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1963 			vattr_t		vattr;
1964 
1965 			/*
1966 			 * Make sure user is the owner of the mount point
1967 			 * or has sufficient privileges.
1968 			 */
1969 
1970 			vattr.va_mask = AT_UID;
1971 
1972 			if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1973 				goto out;
1974 			}
1975 
1976 			if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1977 			    VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1978 				goto out;
1979 			}
1980 			secpolicy_fs_mount_clearopts(cr, vfsp);
1981 		} else {
1982 			goto out;
1983 		}
1984 	}
1985 
1986 	/*
1987 	 * Refuse to mount a filesystem if we are in a local zone and the
1988 	 * dataset is not visible.
1989 	 */
1990 	if (!INGLOBALZONE(curproc) &&
1991 	    (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1992 		error = SET_ERROR(EPERM);
1993 		goto out;
1994 	}
1995 
1996 	error = zfs_mount_label_policy(vfsp, osname);
1997 	if (error)
1998 		goto out;
1999 
2000 	/*
2001 	 * When doing a remount, we simply refresh our temporary properties
2002 	 * according to those options set in the current VFS options.
2003 	 */
2004 	if (uap->flags & MS_REMOUNT) {
2005 		/* refresh mount options */
2006 		zfs_unregister_callbacks(vfsp->vfs_data);
2007 		error = zfs_register_callbacks(vfsp);
2008 		goto out;
2009 	}
2010 
2011 	error = zfs_domount(vfsp, osname);
2012 
2013 	/*
2014 	 * Add an extra VFS_HOLD on our parent vfs so that it can't
2015 	 * disappear due to a forced unmount.
2016 	 */
2017 	if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
2018 		VFS_HOLD(mvp->v_vfsp);
2019 
2020 out:
2021 	pn_free(&spn);
2022 	return (error);
2023 }
2024 
2025 static int
2026 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
2027 {
2028 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2029 	dev32_t d32;
2030 	uint64_t refdbytes, availbytes, usedobjs, availobjs;
2031 	int err = 0;
2032 
2033 	ZFS_ENTER(zfsvfs);
2034 
2035 	dmu_objset_space(zfsvfs->z_os,
2036 	    &refdbytes, &availbytes, &usedobjs, &availobjs);
2037 
2038 	/*
2039 	 * The underlying storage pool actually uses multiple block sizes.
2040 	 * We report the fragsize as the smallest block size we support,
2041 	 * and we report our blocksize as the filesystem's maximum blocksize.
2042 	 */
2043 	statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
2044 	statp->f_bsize = zfsvfs->z_max_blksz;
2045 
2046 	/*
2047 	 * The following report "total" blocks of various kinds in the
2048 	 * file system, but reported in terms of f_frsize - the
2049 	 * "fragment" size.
2050 	 */
2051 
2052 	statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
2053 	statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
2054 	statp->f_bavail = statp->f_bfree; /* no root reservation */
2055 
2056 	/*
2057 	 * statvfs() should really be called statufs(), because it assumes
2058 	 * static metadata.  ZFS doesn't preallocate files, so the best
2059 	 * we can do is report the max that could possibly fit in f_files,
2060 	 * and that minus the number actually used in f_ffree.
2061 	 * For f_ffree, report the smaller of the number of object available
2062 	 * and the number of blocks (each object will take at least a block).
2063 	 */
2064 	statp->f_ffree = MIN(availobjs, statp->f_bfree);
2065 	statp->f_favail = statp->f_ffree;	/* no "root reservation" */
2066 	statp->f_files = statp->f_ffree + usedobjs;
2067 
2068 	(void) cmpldev(&d32, vfsp->vfs_dev);
2069 	statp->f_fsid = d32;
2070 
2071 	/*
2072 	 * We're a zfs filesystem.
2073 	 */
2074 	(void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
2075 
2076 	statp->f_flag = vf_to_stf(vfsp->vfs_flag);
2077 
2078 	statp->f_namemax = MAXNAMELEN - 1;
2079 
2080 	/*
2081 	 * We have all of 32 characters to stuff a string here.
2082 	 * Is there anything useful we could/should provide?
2083 	 */
2084 	bzero(statp->f_fstr, sizeof (statp->f_fstr));
2085 
2086 	if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
2087 	    dmu_objset_projectquota_present(zfsvfs->z_os)) {
2088 		znode_t *zp;
2089 
2090 		/*
2091 		 * In ZoL, zfs_statvfs is passed a Linux dentry (directory
2092 		 * entry), instead of a vfsp. The ZoL code uses the dentry
2093 		 * to get the znode from the dentry's inode. This represents
2094 		 * whatever filename was passed to the user-level statvfs
2095 		 * syscall.
2096 		 *
2097 		 * We're using the VFS root znode here, so this represents a
2098 		 * potential difference from ZoL.
2099 		 */
2100 		if (zfs_zget(zfsvfs, zfsvfs->z_root, &zp) == 0) {
2101 			uint32_t bshift = ddi_fls(statp->f_bsize) - 1;
2102 
2103 			if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid &&
2104 			    zpl_is_valid_projid(zp->z_projid))
2105 				err = zfs_statfs_project(zfsvfs, zp, statp,
2106 				    bshift);
2107 			VN_RELE(ZTOV(zp));
2108 		}
2109 	}
2110 
2111 	ZFS_EXIT(zfsvfs);
2112 	return (err);
2113 }
2114 
2115 static int
2116 zfs_root(vfs_t *vfsp, vnode_t **vpp)
2117 {
2118 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2119 	struct vnode *vp;
2120 	int error;
2121 
2122 	ZFS_ENTER(zfsvfs);
2123 
2124 	vp = zfsvfs->z_rootdir;
2125 	if (vp != NULL) {
2126 		VN_HOLD(vp);
2127 		error = 0;
2128 	} else {
2129 		/* forced unmount */
2130 		error = EIO;
2131 	}
2132 	*vpp = vp;
2133 
2134 	ZFS_EXIT(zfsvfs);
2135 	return (error);
2136 
2137 }
2138 
2139 /*
2140  * Teardown the zfsvfs::z_os.
2141  *
2142  * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
2143  * and 'z_teardown_inactive_lock' held.
2144  */
2145 static int
2146 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
2147 {
2148 	znode_t	*zp;
2149 
2150 	zfs_unlinked_drain_stop_wait(zfsvfs);
2151 
2152 	rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
2153 
2154 	if (!unmounting) {
2155 		/*
2156 		 * We purge the parent filesystem's vfsp as the parent
2157 		 * filesystem and all of its snapshots have their vnode's
2158 		 * v_vfsp set to the parent's filesystem's vfsp.  Note,
2159 		 * 'z_parent' is self referential for non-snapshots.
2160 		 */
2161 		(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2162 	}
2163 
2164 	/*
2165 	 * Close the zil. NB: Can't close the zil while zfs_inactive
2166 	 * threads are blocked as zil_close can call zfs_inactive.
2167 	 */
2168 	if (zfsvfs->z_log) {
2169 		zil_close(zfsvfs->z_log);
2170 		zfsvfs->z_log = NULL;
2171 	}
2172 
2173 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
2174 
2175 	/*
2176 	 * If we are not unmounting (ie: online recv) and someone already
2177 	 * unmounted this file system while we were doing the switcheroo,
2178 	 * or a reopen of z_os failed then just bail out now.
2179 	 */
2180 	if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
2181 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
2182 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2183 		return (SET_ERROR(EIO));
2184 	}
2185 
2186 	/*
2187 	 * At this point there are no vops active, and any new vops will
2188 	 * fail with EIO since we have z_teardown_lock for writer (only
2189 	 * relavent for forced unmount).
2190 	 *
2191 	 * Release all holds on dbufs.
2192 	 */
2193 	mutex_enter(&zfsvfs->z_znodes_lock);
2194 	for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
2195 	    zp = list_next(&zfsvfs->z_all_znodes, zp))
2196 		if (zp->z_sa_hdl) {
2197 			ASSERT(ZTOV(zp)->v_count > 0);
2198 			zfs_znode_dmu_fini(zp);
2199 		}
2200 	mutex_exit(&zfsvfs->z_znodes_lock);
2201 
2202 	/*
2203 	 * If we are unmounting, set the unmounted flag and let new vops
2204 	 * unblock.  zfs_inactive will have the unmounted behavior, and all
2205 	 * other vops will fail with EIO.
2206 	 */
2207 	if (unmounting) {
2208 		/*
2209 		 * Clear the cached root vnode now that we are unmounted.
2210 		 * Its release must be performed outside the teardown locks to
2211 		 * avoid recursive lock entry via zfs_inactive().
2212 		 */
2213 		vnode_t *vp = zfsvfs->z_rootdir;
2214 		zfsvfs->z_rootdir = NULL;
2215 
2216 		zfsvfs->z_unmounted = B_TRUE;
2217 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
2218 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2219 
2220 		/* Drop the cached root vp now that it is safe */
2221 		VN_RELE(vp);
2222 	}
2223 
2224 	/*
2225 	 * z_os will be NULL if there was an error in attempting to reopen
2226 	 * zfsvfs, so just return as the properties had already been
2227 	 * unregistered and cached data had been evicted before.
2228 	 */
2229 	if (zfsvfs->z_os == NULL)
2230 		return (0);
2231 
2232 	/*
2233 	 * Unregister properties.
2234 	 */
2235 	zfs_unregister_callbacks(zfsvfs);
2236 
2237 	/*
2238 	 * Evict cached data
2239 	 */
2240 	if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
2241 	    !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
2242 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
2243 	dmu_objset_evict_dbufs(zfsvfs->z_os);
2244 
2245 	return (0);
2246 }
2247 
2248 /*ARGSUSED*/
2249 static int
2250 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
2251 {
2252 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2253 	objset_t *os;
2254 	int ret;
2255 
2256 	ret = secpolicy_fs_unmount(cr, vfsp);
2257 	if (ret) {
2258 		if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
2259 		    ZFS_DELEG_PERM_MOUNT, cr))
2260 			return (ret);
2261 	}
2262 
2263 	/*
2264 	 * We purge the parent filesystem's vfsp as the parent filesystem
2265 	 * and all of its snapshots have their vnode's v_vfsp set to the
2266 	 * parent's filesystem's vfsp.  Note, 'z_parent' is self
2267 	 * referential for non-snapshots.
2268 	 */
2269 	(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2270 
2271 	/*
2272 	 * Unmount any snapshots mounted under .zfs before unmounting the
2273 	 * dataset itself.
2274 	 */
2275 	if (zfsvfs->z_ctldir != NULL &&
2276 	    (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
2277 		return (ret);
2278 	}
2279 
2280 	if (!(fflag & MS_FORCE)) {
2281 		/*
2282 		 * Check the number of active vnodes in the file system.
2283 		 * Our count is maintained in the vfs structure, but the
2284 		 * number is off by 1 to indicate a hold on the vfs
2285 		 * structure itself.
2286 		 */
2287 		boolean_t draining;
2288 		uint_t thresh = 1;
2289 		vnode_t *ctlvp, *rvp;
2290 
2291 		/*
2292 		 * The cached vnode for the root directory of the mount also
2293 		 * maintains a hold on the vfs structure.
2294 		 */
2295 		rvp = zfsvfs->z_rootdir;
2296 		thresh++;
2297 
2298 		/*
2299 		 * The '.zfs' directory maintains a reference of its own, and
2300 		 * any active references underneath are reflected in the vnode
2301 		 * count. Allow one additional reference for it.
2302 		 */
2303 		ctlvp = zfsvfs->z_ctldir;
2304 		if (ctlvp != NULL) {
2305 			thresh++;
2306 		}
2307 
2308 		/*
2309 		 * If it's running, the asynchronous unlinked drain task needs
2310 		 * to be stopped before the number of active vnodes can be
2311 		 * reliably checked.
2312 		 */
2313 		draining = zfsvfs->z_draining;
2314 		if (draining)
2315 			zfs_unlinked_drain_stop_wait(zfsvfs);
2316 
2317 		if (vfsp->vfs_count > thresh || rvp->v_count > 1 ||
2318 		    (ctlvp != NULL && ctlvp->v_count > 1)) {
2319 			if (draining) {
2320 				/* If it was draining, restart the task */
2321 				zfs_unlinked_drain(zfsvfs);
2322 			}
2323 			return (SET_ERROR(EBUSY));
2324 		}
2325 	}
2326 
2327 	vfsp->vfs_flag |= VFS_UNMOUNTED;
2328 
2329 	VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2330 	os = zfsvfs->z_os;
2331 
2332 	/*
2333 	 * z_os will be NULL if there was an error in
2334 	 * attempting to reopen zfsvfs.
2335 	 */
2336 	if (os != NULL) {
2337 		/*
2338 		 * Unset the objset user_ptr.
2339 		 */
2340 		mutex_enter(&os->os_user_ptr_lock);
2341 		dmu_objset_set_user(os, NULL);
2342 		mutex_exit(&os->os_user_ptr_lock);
2343 
2344 		/*
2345 		 * Finally release the objset
2346 		 */
2347 		dmu_objset_disown(os, B_TRUE, zfsvfs);
2348 	}
2349 
2350 	/*
2351 	 * We can now safely destroy the '.zfs' directory node.
2352 	 */
2353 	if (zfsvfs->z_ctldir != NULL)
2354 		zfsctl_destroy(zfsvfs);
2355 
2356 	return (0);
2357 }
2358 
2359 static int
2360 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
2361 {
2362 	zfsvfs_t	*zfsvfs = vfsp->vfs_data;
2363 	znode_t		*zp;
2364 	uint64_t	object = 0;
2365 	uint64_t	fid_gen = 0;
2366 	uint64_t	gen_mask;
2367 	uint64_t	zp_gen;
2368 	int		i, err;
2369 
2370 	*vpp = NULL;
2371 
2372 	ZFS_ENTER(zfsvfs);
2373 
2374 	if (fidp->fid_len == LONG_FID_LEN) {
2375 		zfid_long_t	*zlfid = (zfid_long_t *)fidp;
2376 		uint64_t	objsetid = 0;
2377 		uint64_t	setgen = 0;
2378 
2379 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2380 			objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2381 
2382 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2383 			setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2384 
2385 		ZFS_EXIT(zfsvfs);
2386 
2387 		err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2388 		if (err)
2389 			return (SET_ERROR(EINVAL));
2390 		ZFS_ENTER(zfsvfs);
2391 	}
2392 
2393 	if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2394 		zfid_short_t	*zfid = (zfid_short_t *)fidp;
2395 
2396 		for (i = 0; i < sizeof (zfid->zf_object); i++)
2397 			object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2398 
2399 		for (i = 0; i < sizeof (zfid->zf_gen); i++)
2400 			fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2401 	} else {
2402 		ZFS_EXIT(zfsvfs);
2403 		return (SET_ERROR(EINVAL));
2404 	}
2405 
2406 	/* A zero fid_gen means we are in the .zfs control directories */
2407 	if (fid_gen == 0 &&
2408 	    (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
2409 		*vpp = zfsvfs->z_ctldir;
2410 		ASSERT(*vpp != NULL);
2411 		if (object == ZFSCTL_INO_SNAPDIR) {
2412 			VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2413 			    0, NULL, NULL, NULL, NULL, NULL) == 0);
2414 		} else {
2415 			VN_HOLD(*vpp);
2416 		}
2417 		ZFS_EXIT(zfsvfs);
2418 		return (0);
2419 	}
2420 
2421 	gen_mask = -1ULL >> (64 - 8 * i);
2422 
2423 	dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2424 	if (err = zfs_zget(zfsvfs, object, &zp)) {
2425 		ZFS_EXIT(zfsvfs);
2426 		return (err);
2427 	}
2428 	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2429 	    sizeof (uint64_t));
2430 	zp_gen = zp_gen & gen_mask;
2431 	if (zp_gen == 0)
2432 		zp_gen = 1;
2433 	if (zp->z_unlinked || zp_gen != fid_gen) {
2434 		dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2435 		VN_RELE(ZTOV(zp));
2436 		ZFS_EXIT(zfsvfs);
2437 		return (SET_ERROR(EINVAL));
2438 	}
2439 
2440 	*vpp = ZTOV(zp);
2441 	ZFS_EXIT(zfsvfs);
2442 	return (0);
2443 }
2444 
2445 /*
2446  * Block out VOPs and close zfsvfs_t::z_os
2447  *
2448  * Note, if successful, then we return with the 'z_teardown_lock' and
2449  * 'z_teardown_inactive_lock' write held.  We leave ownership of the underlying
2450  * dataset and objset intact so that they can be atomically handed off during
2451  * a subsequent rollback or recv operation and the resume thereafter.
2452  */
2453 int
2454 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2455 {
2456 	int error;
2457 
2458 	if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2459 		return (error);
2460 
2461 	return (0);
2462 }
2463 
2464 /*
2465  * Rebuild SA and release VOPs.  Note that ownership of the underlying dataset
2466  * is an invariant across any of the operations that can be performed while the
2467  * filesystem was suspended.  Whether it succeeded or failed, the preconditions
2468  * are the same: the relevant objset and associated dataset are owned by
2469  * zfsvfs, held, and long held on entry.
2470  */
2471 int
2472 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2473 {
2474 	int err;
2475 	znode_t *zp;
2476 
2477 	ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2478 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2479 
2480 	/*
2481 	 * We already own this, so just update the objset_t, as the one we
2482 	 * had before may have been evicted.
2483 	 */
2484 	objset_t *os;
2485 	VERIFY3P(ds->ds_owner, ==, zfsvfs);
2486 	VERIFY(dsl_dataset_long_held(ds));
2487 	VERIFY0(dmu_objset_from_ds(ds, &os));
2488 
2489 	err = zfsvfs_init(zfsvfs, os);
2490 	if (err != 0)
2491 		goto bail;
2492 
2493 	VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2494 
2495 	zfs_set_fuid_feature(zfsvfs);
2496 
2497 	/*
2498 	 * Attempt to re-establish all the active znodes with
2499 	 * their dbufs.  If a zfs_rezget() fails, then we'll let
2500 	 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2501 	 * when they try to use their znode.
2502 	 */
2503 	mutex_enter(&zfsvfs->z_znodes_lock);
2504 	for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2505 	    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2506 		(void) zfs_rezget(zp);
2507 	}
2508 	mutex_exit(&zfsvfs->z_znodes_lock);
2509 
2510 	if (((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) &&
2511 	    !zfsvfs->z_unmounted) {
2512 		/*
2513 		 * zfs_suspend_fs() could have interrupted freeing
2514 		 * of dnodes. We need to restart this freeing so
2515 		 * that we don't "leak" the space.
2516 		 */
2517 		zfs_unlinked_drain(zfsvfs);
2518 	}
2519 
2520 bail:
2521 	/* release the VOPs */
2522 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
2523 	rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2524 
2525 	if (err) {
2526 		/*
2527 		 * Since we couldn't setup the sa framework, try to force
2528 		 * unmount this file system.
2529 		 */
2530 		if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2531 			(void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2532 	}
2533 	return (err);
2534 }
2535 
2536 static void
2537 zfs_freevfs(vfs_t *vfsp)
2538 {
2539 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2540 
2541 	/*
2542 	 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2543 	 * from zfs_mount().  Release it here.  If we came through
2544 	 * zfs_mountroot() instead, we didn't grab an extra hold, so
2545 	 * skip the VFS_RELE for rootvfs.
2546 	 */
2547 	if (zfsvfs->z_issnap && (vfsp != rootvfs))
2548 		VFS_RELE(zfsvfs->z_parent->z_vfs);
2549 
2550 	zfsvfs_free(zfsvfs);
2551 
2552 	atomic_dec_32(&zfs_active_fs_count);
2553 }
2554 
2555 /*
2556  * VFS_INIT() initialization.  Note that there is no VFS_FINI(),
2557  * so we can't safely do any non-idempotent initialization here.
2558  * Leave that to zfs_init() and zfs_fini(), which are called
2559  * from the module's _init() and _fini() entry points.
2560  */
2561 /*ARGSUSED*/
2562 static int
2563 zfs_vfsinit(int fstype, char *name)
2564 {
2565 	int error;
2566 
2567 	zfsfstype = fstype;
2568 
2569 	/*
2570 	 * Setup vfsops and vnodeops tables.
2571 	 */
2572 	error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2573 	if (error != 0) {
2574 		cmn_err(CE_WARN, "zfs: bad vfs ops template");
2575 	}
2576 
2577 	error = zfs_create_op_tables();
2578 	if (error) {
2579 		zfs_remove_op_tables();
2580 		cmn_err(CE_WARN, "zfs: bad vnode ops template");
2581 		(void) vfs_freevfsops_by_type(zfsfstype);
2582 		return (error);
2583 	}
2584 
2585 	mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2586 
2587 	/*
2588 	 * Unique major number for all zfs mounts.
2589 	 * If we run out of 32-bit minors, we'll getudev() another major.
2590 	 */
2591 	zfs_major = ddi_name_to_major(ZFS_DRIVER);
2592 	zfs_minor = ZFS_MIN_MINOR;
2593 
2594 	return (0);
2595 }
2596 
2597 void
2598 zfs_init(void)
2599 {
2600 	/*
2601 	 * Initialize .zfs directory structures
2602 	 */
2603 	zfsctl_init();
2604 
2605 	/*
2606 	 * Initialize znode cache, vnode ops, etc...
2607 	 */
2608 	zfs_znode_init();
2609 
2610 	dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2611 }
2612 
2613 void
2614 zfs_fini(void)
2615 {
2616 	zfsctl_fini();
2617 	zfs_znode_fini();
2618 }
2619 
2620 int
2621 zfs_busy(void)
2622 {
2623 	return (zfs_active_fs_count != 0);
2624 }
2625 
2626 int
2627 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2628 {
2629 	int error;
2630 	objset_t *os = zfsvfs->z_os;
2631 	dmu_tx_t *tx;
2632 
2633 	if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2634 		return (SET_ERROR(EINVAL));
2635 
2636 	if (newvers < zfsvfs->z_version)
2637 		return (SET_ERROR(EINVAL));
2638 
2639 	if (zfs_spa_version_map(newvers) >
2640 	    spa_version(dmu_objset_spa(zfsvfs->z_os)))
2641 		return (SET_ERROR(ENOTSUP));
2642 
2643 	tx = dmu_tx_create(os);
2644 	dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2645 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2646 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2647 		    ZFS_SA_ATTRS);
2648 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2649 	}
2650 	error = dmu_tx_assign(tx, TXG_WAIT);
2651 	if (error) {
2652 		dmu_tx_abort(tx);
2653 		return (error);
2654 	}
2655 
2656 	error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2657 	    8, 1, &newvers, tx);
2658 
2659 	if (error) {
2660 		dmu_tx_commit(tx);
2661 		return (error);
2662 	}
2663 
2664 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2665 		uint64_t sa_obj;
2666 
2667 		ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2668 		    SPA_VERSION_SA);
2669 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2670 		    DMU_OT_NONE, 0, tx);
2671 
2672 		error = zap_add(os, MASTER_NODE_OBJ,
2673 		    ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2674 		ASSERT0(error);
2675 
2676 		VERIFY(0 == sa_set_sa_object(os, sa_obj));
2677 		sa_register_update_callback(os, zfs_sa_upgrade);
2678 	}
2679 
2680 	spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2681 	    "from %llu to %llu", zfsvfs->z_version, newvers);
2682 
2683 	dmu_tx_commit(tx);
2684 
2685 	zfsvfs->z_version = newvers;
2686 	os->os_version = newvers;
2687 
2688 	zfs_set_fuid_feature(zfsvfs);
2689 
2690 	return (0);
2691 }
2692 
2693 /*
2694  * Read a property stored within the master node.
2695  */
2696 int
2697 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2698 {
2699 	uint64_t *cached_copy = NULL;
2700 
2701 	/*
2702 	 * Figure out where in the objset_t the cached copy would live, if it
2703 	 * is available for the requested property.
2704 	 */
2705 	if (os != NULL) {
2706 		switch (prop) {
2707 		case ZFS_PROP_VERSION:
2708 			cached_copy = &os->os_version;
2709 			break;
2710 		case ZFS_PROP_NORMALIZE:
2711 			cached_copy = &os->os_normalization;
2712 			break;
2713 		case ZFS_PROP_UTF8ONLY:
2714 			cached_copy = &os->os_utf8only;
2715 			break;
2716 		case ZFS_PROP_CASE:
2717 			cached_copy = &os->os_casesensitivity;
2718 			break;
2719 		default:
2720 			break;
2721 		}
2722 	}
2723 	if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) {
2724 		*value = *cached_copy;
2725 		return (0);
2726 	}
2727 
2728 	/*
2729 	 * If the property wasn't cached, look up the file system's value for
2730 	 * the property. For the version property, we look up a slightly
2731 	 * different string.
2732 	 */
2733 	const char *pname;
2734 	int error = ENOENT;
2735 	if (prop == ZFS_PROP_VERSION) {
2736 		pname = ZPL_VERSION_STR;
2737 	} else {
2738 		pname = zfs_prop_to_name(prop);
2739 	}
2740 
2741 	if (os != NULL) {
2742 		ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2743 		error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2744 	}
2745 
2746 	if (error == ENOENT) {
2747 		/* No value set, use the default value */
2748 		switch (prop) {
2749 		case ZFS_PROP_VERSION:
2750 			*value = ZPL_VERSION;
2751 			break;
2752 		case ZFS_PROP_NORMALIZE:
2753 		case ZFS_PROP_UTF8ONLY:
2754 			*value = 0;
2755 			break;
2756 		case ZFS_PROP_CASE:
2757 			*value = ZFS_CASE_SENSITIVE;
2758 			break;
2759 		default:
2760 			return (error);
2761 		}
2762 		error = 0;
2763 	}
2764 
2765 	/*
2766 	 * If one of the methods for getting the property value above worked,
2767 	 * copy it into the objset_t's cache.
2768 	 */
2769 	if (error == 0 && cached_copy != NULL) {
2770 		*cached_copy = *value;
2771 	}
2772 
2773 	return (error);
2774 }
2775 
2776 /*
2777  * Return true if the coresponding vfs's unmounted flag is set.
2778  * Otherwise return false.
2779  * If this function returns true we know VFS unmount has been initiated.
2780  */
2781 boolean_t
2782 zfs_get_vfs_flag_unmounted(objset_t *os)
2783 {
2784 	zfsvfs_t *zfvp;
2785 	boolean_t unmounted = B_FALSE;
2786 
2787 	ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2788 
2789 	mutex_enter(&os->os_user_ptr_lock);
2790 	zfvp = dmu_objset_get_user(os);
2791 	if (zfvp != NULL && zfvp->z_vfs != NULL &&
2792 	    (zfvp->z_vfs->vfs_flag & VFS_UNMOUNTED))
2793 		unmounted = B_TRUE;
2794 	mutex_exit(&os->os_user_ptr_lock);
2795 
2796 	return (unmounted);
2797 }
2798 
2799 static vfsdef_t vfw = {
2800 	VFSDEF_VERSION,
2801 	MNTTYPE_ZFS,
2802 	zfs_vfsinit,
2803 	VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2804 	    VSW_XID|VSW_ZMOUNT,
2805 	&zfs_mntopts
2806 };
2807 
2808 struct modlfs zfs_modlfs = {
2809 	&mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2810 };
2811