xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_vfsops.c (revision b666b5bebdde1067e2d0f2b91fd28a5243311a1a)
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  */
28 
29 /* Portions Copyright 2010 Robert Milkowski */
30 
31 #include <sys/types.h>
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/sysmacros.h>
35 #include <sys/kmem.h>
36 #include <sys/pathname.h>
37 #include <sys/vnode.h>
38 #include <sys/vfs.h>
39 #include <sys/vfs_opreg.h>
40 #include <sys/mntent.h>
41 #include <sys/mount.h>
42 #include <sys/cmn_err.h>
43 #include "fs/fs_subr.h"
44 #include <sys/zfs_znode.h>
45 #include <sys/zfs_dir.h>
46 #include <sys/zil.h>
47 #include <sys/fs/zfs.h>
48 #include <sys/dmu.h>
49 #include <sys/dsl_prop.h>
50 #include <sys/dsl_dataset.h>
51 #include <sys/dsl_deleg.h>
52 #include <sys/spa.h>
53 #include <sys/zap.h>
54 #include <sys/sa.h>
55 #include <sys/sa_impl.h>
56 #include <sys/varargs.h>
57 #include <sys/policy.h>
58 #include <sys/atomic.h>
59 #include <sys/mkdev.h>
60 #include <sys/modctl.h>
61 #include <sys/refstr.h>
62 #include <sys/zfs_ioctl.h>
63 #include <sys/zfs_ctldir.h>
64 #include <sys/zfs_fuid.h>
65 #include <sys/bootconf.h>
66 #include <sys/sunddi.h>
67 #include <sys/dnlc.h>
68 #include <sys/dmu_objset.h>
69 #include <sys/spa_boot.h>
70 #include "zfs_comutil.h"
71 
72 int zfsfstype;
73 vfsops_t *zfs_vfsops = NULL;
74 static major_t zfs_major;
75 static minor_t zfs_minor;
76 static kmutex_t	zfs_dev_mtx;
77 
78 extern int sys_shutdown;
79 
80 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
81 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
82 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
83 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
84 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
85 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
86 static void zfs_freevfs(vfs_t *vfsp);
87 
88 static const fs_operation_def_t zfs_vfsops_template[] = {
89 	VFSNAME_MOUNT,		{ .vfs_mount = zfs_mount },
90 	VFSNAME_MOUNTROOT,	{ .vfs_mountroot = zfs_mountroot },
91 	VFSNAME_UNMOUNT,	{ .vfs_unmount = zfs_umount },
92 	VFSNAME_ROOT,		{ .vfs_root = zfs_root },
93 	VFSNAME_STATVFS,	{ .vfs_statvfs = zfs_statvfs },
94 	VFSNAME_SYNC,		{ .vfs_sync = zfs_sync },
95 	VFSNAME_VGET,		{ .vfs_vget = zfs_vget },
96 	VFSNAME_FREEVFS,	{ .vfs_freevfs = zfs_freevfs },
97 	NULL,			NULL
98 };
99 
100 /*
101  * We need to keep a count of active fs's.
102  * This is necessary to prevent our module
103  * from being unloaded after a umount -f
104  */
105 static uint32_t	zfs_active_fs_count = 0;
106 
107 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
108 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
109 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
110 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
111 
112 /*
113  * MO_DEFAULT is not used since the default value is determined
114  * by the equivalent property.
115  */
116 static mntopt_t mntopts[] = {
117 	{ MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
118 	{ MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
119 	{ MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
120 	{ MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
121 };
122 
123 static mntopts_t zfs_mntopts = {
124 	sizeof (mntopts) / sizeof (mntopt_t),
125 	mntopts
126 };
127 
128 /*ARGSUSED*/
129 int
130 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
131 {
132 	/*
133 	 * Data integrity is job one.  We don't want a compromised kernel
134 	 * writing to the storage pool, so we never sync during panic.
135 	 */
136 	if (panicstr)
137 		return (0);
138 
139 	/*
140 	 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
141 	 * to sync metadata, which they would otherwise cache indefinitely.
142 	 * Semantically, the only requirement is that the sync be initiated.
143 	 * The DMU syncs out txgs frequently, so there's nothing to do.
144 	 */
145 	if (flag & SYNC_ATTR)
146 		return (0);
147 
148 	if (vfsp != NULL) {
149 		/*
150 		 * Sync a specific filesystem.
151 		 */
152 		zfsvfs_t *zfsvfs = vfsp->vfs_data;
153 		dsl_pool_t *dp;
154 
155 		ZFS_ENTER(zfsvfs);
156 		dp = dmu_objset_pool(zfsvfs->z_os);
157 
158 		/*
159 		 * If the system is shutting down, then skip any
160 		 * filesystems which may exist on a suspended pool.
161 		 */
162 		if (sys_shutdown && spa_suspended(dp->dp_spa)) {
163 			ZFS_EXIT(zfsvfs);
164 			return (0);
165 		}
166 
167 		if (zfsvfs->z_log != NULL)
168 			zil_commit(zfsvfs->z_log, 0);
169 
170 		ZFS_EXIT(zfsvfs);
171 	} else {
172 		/*
173 		 * Sync all ZFS filesystems.  This is what happens when you
174 		 * run sync(1M).  Unlike other filesystems, ZFS honors the
175 		 * request by waiting for all pools to commit all dirty data.
176 		 */
177 		spa_sync_allpools();
178 	}
179 
180 	return (0);
181 }
182 
183 static int
184 zfs_create_unique_device(dev_t *dev)
185 {
186 	major_t new_major;
187 
188 	do {
189 		ASSERT3U(zfs_minor, <=, MAXMIN32);
190 		minor_t start = zfs_minor;
191 		do {
192 			mutex_enter(&zfs_dev_mtx);
193 			if (zfs_minor >= MAXMIN32) {
194 				/*
195 				 * If we're still using the real major
196 				 * keep out of /dev/zfs and /dev/zvol minor
197 				 * number space.  If we're using a getudev()'ed
198 				 * major number, we can use all of its minors.
199 				 */
200 				if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
201 					zfs_minor = ZFS_MIN_MINOR;
202 				else
203 					zfs_minor = 0;
204 			} else {
205 				zfs_minor++;
206 			}
207 			*dev = makedevice(zfs_major, zfs_minor);
208 			mutex_exit(&zfs_dev_mtx);
209 		} while (vfs_devismounted(*dev) && zfs_minor != start);
210 		if (zfs_minor == start) {
211 			/*
212 			 * We are using all ~262,000 minor numbers for the
213 			 * current major number.  Create a new major number.
214 			 */
215 			if ((new_major = getudev()) == (major_t)-1) {
216 				cmn_err(CE_WARN,
217 				    "zfs_mount: Can't get unique major "
218 				    "device number.");
219 				return (-1);
220 			}
221 			mutex_enter(&zfs_dev_mtx);
222 			zfs_major = new_major;
223 			zfs_minor = 0;
224 
225 			mutex_exit(&zfs_dev_mtx);
226 		} else {
227 			break;
228 		}
229 		/* CONSTANTCONDITION */
230 	} while (1);
231 
232 	return (0);
233 }
234 
235 static void
236 atime_changed_cb(void *arg, uint64_t newval)
237 {
238 	zfsvfs_t *zfsvfs = arg;
239 
240 	if (newval == TRUE) {
241 		zfsvfs->z_atime = TRUE;
242 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
243 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
244 	} else {
245 		zfsvfs->z_atime = FALSE;
246 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
247 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
248 	}
249 }
250 
251 static void
252 xattr_changed_cb(void *arg, uint64_t newval)
253 {
254 	zfsvfs_t *zfsvfs = arg;
255 
256 	if (newval == TRUE) {
257 		/* XXX locking on vfs_flag? */
258 		zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
259 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
260 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
261 	} else {
262 		/* XXX locking on vfs_flag? */
263 		zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
264 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
265 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
266 	}
267 }
268 
269 static void
270 blksz_changed_cb(void *arg, uint64_t newval)
271 {
272 	zfsvfs_t *zfsvfs = arg;
273 	ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
274 	ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
275 	ASSERT(ISP2(newval));
276 
277 	zfsvfs->z_max_blksz = newval;
278 	zfsvfs->z_vfs->vfs_bsize = newval;
279 }
280 
281 static void
282 readonly_changed_cb(void *arg, uint64_t newval)
283 {
284 	zfsvfs_t *zfsvfs = arg;
285 
286 	if (newval) {
287 		/* XXX locking on vfs_flag? */
288 		zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
289 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
290 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
291 	} else {
292 		/* XXX locking on vfs_flag? */
293 		zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
294 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
295 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
296 	}
297 }
298 
299 static void
300 devices_changed_cb(void *arg, uint64_t newval)
301 {
302 	zfsvfs_t *zfsvfs = arg;
303 
304 	if (newval == FALSE) {
305 		zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
306 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
307 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
308 	} else {
309 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
310 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
311 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
312 	}
313 }
314 
315 static void
316 setuid_changed_cb(void *arg, uint64_t newval)
317 {
318 	zfsvfs_t *zfsvfs = arg;
319 
320 	if (newval == FALSE) {
321 		zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
322 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
323 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
324 	} else {
325 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
326 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
327 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
328 	}
329 }
330 
331 static void
332 exec_changed_cb(void *arg, uint64_t newval)
333 {
334 	zfsvfs_t *zfsvfs = arg;
335 
336 	if (newval == FALSE) {
337 		zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
338 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
339 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
340 	} else {
341 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
342 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
343 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
344 	}
345 }
346 
347 /*
348  * The nbmand mount option can be changed at mount time.
349  * We can't allow it to be toggled on live file systems or incorrect
350  * behavior may be seen from cifs clients
351  *
352  * This property isn't registered via dsl_prop_register(), but this callback
353  * will be called when a file system is first mounted
354  */
355 static void
356 nbmand_changed_cb(void *arg, uint64_t newval)
357 {
358 	zfsvfs_t *zfsvfs = arg;
359 	if (newval == FALSE) {
360 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
361 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
362 	} else {
363 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
364 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
365 	}
366 }
367 
368 static void
369 snapdir_changed_cb(void *arg, uint64_t newval)
370 {
371 	zfsvfs_t *zfsvfs = arg;
372 
373 	zfsvfs->z_show_ctldir = newval;
374 }
375 
376 static void
377 vscan_changed_cb(void *arg, uint64_t newval)
378 {
379 	zfsvfs_t *zfsvfs = arg;
380 
381 	zfsvfs->z_vscan = newval;
382 }
383 
384 static void
385 acl_mode_changed_cb(void *arg, uint64_t newval)
386 {
387 	zfsvfs_t *zfsvfs = arg;
388 
389 	zfsvfs->z_acl_mode = newval;
390 }
391 
392 static void
393 acl_inherit_changed_cb(void *arg, uint64_t newval)
394 {
395 	zfsvfs_t *zfsvfs = arg;
396 
397 	zfsvfs->z_acl_inherit = newval;
398 }
399 
400 static int
401 zfs_register_callbacks(vfs_t *vfsp)
402 {
403 	struct dsl_dataset *ds = NULL;
404 	objset_t *os = NULL;
405 	zfsvfs_t *zfsvfs = NULL;
406 	uint64_t nbmand;
407 	boolean_t readonly = B_FALSE;
408 	boolean_t do_readonly = B_FALSE;
409 	boolean_t setuid = B_FALSE;
410 	boolean_t do_setuid = B_FALSE;
411 	boolean_t exec = B_FALSE;
412 	boolean_t do_exec = B_FALSE;
413 	boolean_t devices = B_FALSE;
414 	boolean_t do_devices = B_FALSE;
415 	boolean_t xattr = B_FALSE;
416 	boolean_t do_xattr = B_FALSE;
417 	boolean_t atime = B_FALSE;
418 	boolean_t do_atime = B_FALSE;
419 	int error = 0;
420 
421 	ASSERT(vfsp);
422 	zfsvfs = vfsp->vfs_data;
423 	ASSERT(zfsvfs);
424 	os = zfsvfs->z_os;
425 
426 	/*
427 	 * The act of registering our callbacks will destroy any mount
428 	 * options we may have.  In order to enable temporary overrides
429 	 * of mount options, we stash away the current values and
430 	 * restore them after we register the callbacks.
431 	 */
432 	if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
433 	    !spa_writeable(dmu_objset_spa(os))) {
434 		readonly = B_TRUE;
435 		do_readonly = B_TRUE;
436 	} else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
437 		readonly = B_FALSE;
438 		do_readonly = B_TRUE;
439 	}
440 	if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
441 		devices = B_FALSE;
442 		setuid = B_FALSE;
443 		do_devices = B_TRUE;
444 		do_setuid = B_TRUE;
445 	} else {
446 		if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
447 			devices = B_FALSE;
448 			do_devices = B_TRUE;
449 		} else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
450 			devices = B_TRUE;
451 			do_devices = B_TRUE;
452 		}
453 
454 		if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
455 			setuid = B_FALSE;
456 			do_setuid = B_TRUE;
457 		} else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
458 			setuid = B_TRUE;
459 			do_setuid = B_TRUE;
460 		}
461 	}
462 	if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
463 		exec = B_FALSE;
464 		do_exec = B_TRUE;
465 	} else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
466 		exec = B_TRUE;
467 		do_exec = B_TRUE;
468 	}
469 	if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
470 		xattr = B_FALSE;
471 		do_xattr = B_TRUE;
472 	} else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
473 		xattr = B_TRUE;
474 		do_xattr = B_TRUE;
475 	}
476 	if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
477 		atime = B_FALSE;
478 		do_atime = B_TRUE;
479 	} else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
480 		atime = B_TRUE;
481 		do_atime = B_TRUE;
482 	}
483 
484 	/*
485 	 * nbmand is a special property.  It can only be changed at
486 	 * mount time.
487 	 *
488 	 * This is weird, but it is documented to only be changeable
489 	 * at mount time.
490 	 */
491 	if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
492 		nbmand = B_FALSE;
493 	} else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
494 		nbmand = B_TRUE;
495 	} else {
496 		char osname[ZFS_MAX_DATASET_NAME_LEN];
497 
498 		dmu_objset_name(os, osname);
499 		if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
500 		    NULL)) {
501 			return (error);
502 		}
503 	}
504 
505 	/*
506 	 * Register property callbacks.
507 	 *
508 	 * It would probably be fine to just check for i/o error from
509 	 * the first prop_register(), but I guess I like to go
510 	 * overboard...
511 	 */
512 	ds = dmu_objset_ds(os);
513 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
514 	error = dsl_prop_register(ds,
515 	    zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
516 	error = error ? error : dsl_prop_register(ds,
517 	    zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
518 	error = error ? error : dsl_prop_register(ds,
519 	    zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
520 	error = error ? error : dsl_prop_register(ds,
521 	    zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
522 	error = error ? error : dsl_prop_register(ds,
523 	    zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
524 	error = error ? error : dsl_prop_register(ds,
525 	    zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
526 	error = error ? error : dsl_prop_register(ds,
527 	    zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
528 	error = error ? error : dsl_prop_register(ds,
529 	    zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
530 	error = error ? error : dsl_prop_register(ds,
531 	    zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
532 	error = error ? error : dsl_prop_register(ds,
533 	    zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
534 	    zfsvfs);
535 	error = error ? error : dsl_prop_register(ds,
536 	    zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
537 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
538 	if (error)
539 		goto unregister;
540 
541 	/*
542 	 * Invoke our callbacks to restore temporary mount options.
543 	 */
544 	if (do_readonly)
545 		readonly_changed_cb(zfsvfs, readonly);
546 	if (do_setuid)
547 		setuid_changed_cb(zfsvfs, setuid);
548 	if (do_exec)
549 		exec_changed_cb(zfsvfs, exec);
550 	if (do_devices)
551 		devices_changed_cb(zfsvfs, devices);
552 	if (do_xattr)
553 		xattr_changed_cb(zfsvfs, xattr);
554 	if (do_atime)
555 		atime_changed_cb(zfsvfs, atime);
556 
557 	nbmand_changed_cb(zfsvfs, nbmand);
558 
559 	return (0);
560 
561 unregister:
562 	dsl_prop_unregister_all(ds, zfsvfs);
563 	return (error);
564 }
565 
566 static int
567 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
568     uint64_t *userp, uint64_t *groupp, uint64_t *projectp)
569 {
570 	sa_hdr_phys_t sa;
571 	sa_hdr_phys_t *sap = data;
572 	uint64_t flags;
573 	int hdrsize;
574 	boolean_t swap = B_FALSE;
575 
576 	/*
577 	 * Is it a valid type of object to track?
578 	 */
579 	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
580 		return (SET_ERROR(ENOENT));
581 
582 	/*
583 	 * If we have a NULL data pointer
584 	 * then assume the id's aren't changing and
585 	 * return EEXIST to the dmu to let it know to
586 	 * use the same ids
587 	 */
588 	if (data == NULL)
589 		return (SET_ERROR(EEXIST));
590 
591 	if (bonustype == DMU_OT_ZNODE) {
592 		znode_phys_t *znp = data;
593 		*userp = znp->zp_uid;
594 		*groupp = znp->zp_gid;
595 		*projectp = ZFS_DEFAULT_PROJID;
596 		return (0);
597 	}
598 
599 	if (sap->sa_magic == 0) {
600 		/*
601 		 * This should only happen for newly created files
602 		 * that haven't had the znode data filled in yet.
603 		 */
604 		*userp = 0;
605 		*groupp = 0;
606 		*projectp = ZFS_DEFAULT_PROJID;
607 		return (0);
608 	}
609 
610 	sa = *sap;
611 	if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
612 		sa.sa_magic = SA_MAGIC;
613 		sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
614 		swap = B_TRUE;
615 	} else {
616 		VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
617 	}
618 
619 	hdrsize = sa_hdrsize(&sa);
620 	VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
621 
622 	*userp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_UID_OFFSET));
623 	*groupp = *((uint64_t *)((uintptr_t)data + hdrsize + SA_GID_OFFSET));
624 	flags = *((uint64_t *)((uintptr_t)data + hdrsize + SA_FLAGS_OFFSET));
625 	if (swap)
626 		flags = BSWAP_64(flags);
627 
628 	if (flags & ZFS_PROJID)
629 		*projectp = *((uint64_t *)((uintptr_t)data + hdrsize +
630 		    SA_PROJID_OFFSET));
631 	else
632 		*projectp = ZFS_DEFAULT_PROJID;
633 
634 	if (swap) {
635 		*userp = BSWAP_64(*userp);
636 		*groupp = BSWAP_64(*groupp);
637 		*projectp = BSWAP_64(*projectp);
638 	}
639 	return (0);
640 }
641 
642 static void
643 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
644     char *domainbuf, int buflen, uid_t *ridp)
645 {
646 	uint64_t fuid;
647 	const char *domain;
648 
649 	fuid = zfs_strtonum(fuidstr, NULL);
650 
651 	domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
652 	if (domain)
653 		(void) strlcpy(domainbuf, domain, buflen);
654 	else
655 		domainbuf[0] = '\0';
656 	*ridp = FUID_RID(fuid);
657 }
658 
659 static uint64_t
660 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
661 {
662 	switch (type) {
663 	case ZFS_PROP_USERUSED:
664 	case ZFS_PROP_USEROBJUSED:
665 		return (DMU_USERUSED_OBJECT);
666 	case ZFS_PROP_GROUPUSED:
667 	case ZFS_PROP_GROUPOBJUSED:
668 		return (DMU_GROUPUSED_OBJECT);
669 	case ZFS_PROP_PROJECTUSED:
670 	case ZFS_PROP_PROJECTOBJUSED:
671 		return (DMU_PROJECTUSED_OBJECT);
672 	case ZFS_PROP_USERQUOTA:
673 		return (zfsvfs->z_userquota_obj);
674 	case ZFS_PROP_GROUPQUOTA:
675 		return (zfsvfs->z_groupquota_obj);
676 	case ZFS_PROP_USEROBJQUOTA:
677 		return (zfsvfs->z_userobjquota_obj);
678 	case ZFS_PROP_GROUPOBJQUOTA:
679 		return (zfsvfs->z_groupobjquota_obj);
680 	case ZFS_PROP_PROJECTQUOTA:
681 		return (zfsvfs->z_projectquota_obj);
682 	case ZFS_PROP_PROJECTOBJQUOTA:
683 		return (zfsvfs->z_projectobjquota_obj);
684 	default:
685 		return (ZFS_NO_OBJECT);
686 	}
687 }
688 
689 int
690 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
691     uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
692 {
693 	int error;
694 	zap_cursor_t zc;
695 	zap_attribute_t za;
696 	zfs_useracct_t *buf = vbuf;
697 	uint64_t obj;
698 	int offset = 0;
699 
700 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
701 		return (SET_ERROR(ENOTSUP));
702 
703 	if ((type == ZFS_PROP_PROJECTQUOTA || type == ZFS_PROP_PROJECTUSED ||
704 	    type == ZFS_PROP_PROJECTOBJQUOTA ||
705 	    type == ZFS_PROP_PROJECTOBJUSED) &&
706 	    !dmu_objset_projectquota_present(zfsvfs->z_os))
707 		return (SET_ERROR(ENOTSUP));
708 
709 	if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
710 	    type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA ||
711 	    type == ZFS_PROP_PROJECTOBJUSED ||
712 	    type == ZFS_PROP_PROJECTOBJQUOTA) &&
713 	    !dmu_objset_userobjspace_present(zfsvfs->z_os))
714 		return (SET_ERROR(ENOTSUP));
715 
716 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
717 	if (obj == ZFS_NO_OBJECT) {
718 		*bufsizep = 0;
719 		return (0);
720 	}
721 
722 	if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
723 	    type == ZFS_PROP_PROJECTOBJUSED)
724 		offset = DMU_OBJACCT_PREFIX_LEN;
725 
726 	for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
727 	    (error = zap_cursor_retrieve(&zc, &za)) == 0;
728 	    zap_cursor_advance(&zc)) {
729 		if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
730 		    *bufsizep)
731 			break;
732 
733 		/*
734 		 * skip object quota (with zap name prefix DMU_OBJACCT_PREFIX)
735 		 * when dealing with block quota and vice versa.
736 		 */
737 		if ((offset > 0) != (strncmp(za.za_name, DMU_OBJACCT_PREFIX,
738 		    DMU_OBJACCT_PREFIX_LEN) == 0))
739 			continue;
740 
741 		fuidstr_to_sid(zfsvfs, za.za_name + offset,
742 		    buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
743 
744 		buf->zu_space = za.za_first_integer;
745 		buf++;
746 	}
747 	if (error == ENOENT)
748 		error = 0;
749 
750 	ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
751 	*bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
752 	*cookiep = zap_cursor_serialize(&zc);
753 	zap_cursor_fini(&zc);
754 	return (error);
755 }
756 
757 /*
758  * buf must be big enough (eg, 32 bytes)
759  */
760 static int
761 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
762     char *buf, boolean_t addok)
763 {
764 	uint64_t fuid;
765 	int domainid = 0;
766 
767 	if (domain && domain[0]) {
768 		domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
769 		if (domainid == -1)
770 			return (SET_ERROR(ENOENT));
771 	}
772 	fuid = FUID_ENCODE(domainid, rid);
773 	(void) sprintf(buf, "%llx", (longlong_t)fuid);
774 	return (0);
775 }
776 
777 int
778 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
779     const char *domain, uint64_t rid, uint64_t *valp)
780 {
781 	char buf[20 + DMU_OBJACCT_PREFIX_LEN];
782 	int offset = 0;
783 	int err;
784 	uint64_t obj;
785 
786 	*valp = 0;
787 
788 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
789 		return (SET_ERROR(ENOTSUP));
790 
791 	if ((type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
792 	    type == ZFS_PROP_USEROBJQUOTA || type == ZFS_PROP_GROUPOBJQUOTA ||
793 	    type == ZFS_PROP_PROJECTOBJUSED ||
794 	    type == ZFS_PROP_PROJECTOBJQUOTA) &&
795 	    !dmu_objset_userobjspace_present(zfsvfs->z_os))
796 		return (SET_ERROR(ENOTSUP));
797 
798 	if (type == ZFS_PROP_PROJECTQUOTA || type == ZFS_PROP_PROJECTUSED ||
799 	    type == ZFS_PROP_PROJECTOBJQUOTA ||
800 	    type == ZFS_PROP_PROJECTOBJUSED) {
801 		if (!dmu_objset_projectquota_present(zfsvfs->z_os))
802 			return (SET_ERROR(ENOTSUP));
803 		if (!zpl_is_valid_projid(rid))
804 			return (SET_ERROR(EINVAL));
805 	}
806 
807 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
808 	if (obj == ZFS_NO_OBJECT)
809 		return (0);
810 
811 	if (type == ZFS_PROP_USEROBJUSED || type == ZFS_PROP_GROUPOBJUSED ||
812 	    type == ZFS_PROP_PROJECTOBJUSED) {
813 		strncpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN);
814 		offset = DMU_OBJACCT_PREFIX_LEN;
815 	}
816 
817 	err = id_to_fuidstr(zfsvfs, domain, rid, buf + offset, B_FALSE);
818 	if (err)
819 		return (err);
820 
821 	err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
822 	if (err == ENOENT)
823 		err = 0;
824 	return (err);
825 }
826 
827 int
828 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
829     const char *domain, uint64_t rid, uint64_t quota)
830 {
831 	char buf[32];
832 	int err;
833 	dmu_tx_t *tx;
834 	uint64_t *objp;
835 	boolean_t fuid_dirtied;
836 
837 	if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
838 		return (SET_ERROR(ENOTSUP));
839 
840 	switch (type) {
841 	case ZFS_PROP_USERQUOTA:
842 		objp = &zfsvfs->z_userquota_obj;
843 		break;
844 	case ZFS_PROP_GROUPQUOTA:
845 		objp = &zfsvfs->z_groupquota_obj;
846 		break;
847 	case ZFS_PROP_USEROBJQUOTA:
848 		objp = &zfsvfs->z_userobjquota_obj;
849 		break;
850 	case ZFS_PROP_GROUPOBJQUOTA:
851 		objp = &zfsvfs->z_groupobjquota_obj;
852 		break;
853 	case ZFS_PROP_PROJECTQUOTA:
854 		if (!dmu_objset_projectquota_enabled(zfsvfs->z_os))
855 			return (SET_ERROR(ENOTSUP));
856 		if (!zpl_is_valid_projid(rid))
857 			return (SET_ERROR(EINVAL));
858 
859 		objp = &zfsvfs->z_projectquota_obj;
860 		break;
861 	case ZFS_PROP_PROJECTOBJQUOTA:
862 		if (!dmu_objset_projectquota_enabled(zfsvfs->z_os))
863 			return (SET_ERROR(ENOTSUP));
864 		if (!zpl_is_valid_projid(rid))
865 			return (SET_ERROR(EINVAL));
866 
867 		objp = &zfsvfs->z_projectobjquota_obj;
868 		break;
869 	default:
870 		return (SET_ERROR(EINVAL));
871 	}
872 
873 	err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
874 	if (err)
875 		return (err);
876 	fuid_dirtied = zfsvfs->z_fuid_dirty;
877 
878 	tx = dmu_tx_create(zfsvfs->z_os);
879 	dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
880 	if (*objp == 0) {
881 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
882 		    zfs_userquota_prop_prefixes[type]);
883 	}
884 	if (fuid_dirtied)
885 		zfs_fuid_txhold(zfsvfs, tx);
886 	err = dmu_tx_assign(tx, TXG_WAIT);
887 	if (err) {
888 		dmu_tx_abort(tx);
889 		return (err);
890 	}
891 
892 	mutex_enter(&zfsvfs->z_lock);
893 	if (*objp == 0) {
894 		*objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
895 		    DMU_OT_NONE, 0, tx);
896 		VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
897 		    zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
898 	}
899 	mutex_exit(&zfsvfs->z_lock);
900 
901 	if (quota == 0) {
902 		err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
903 		if (err == ENOENT)
904 			err = 0;
905 	} else {
906 		err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
907 	}
908 	ASSERT(err == 0);
909 	if (fuid_dirtied)
910 		zfs_fuid_sync(zfsvfs, tx);
911 	dmu_tx_commit(tx);
912 	return (err);
913 }
914 
915 boolean_t
916 zfs_id_overobjquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
917 {
918 	char buf[20 + DMU_OBJACCT_PREFIX_LEN];
919 	uint64_t used, quota, quotaobj;
920 	int err;
921 
922 	if (!dmu_objset_userobjspace_present(zfsvfs->z_os)) {
923 		if (dmu_objset_userobjspace_upgradable(zfsvfs->z_os))
924 			dmu_objset_id_quota_upgrade(zfsvfs->z_os);
925 		return (B_FALSE);
926 	}
927 
928 	if (usedobj == DMU_PROJECTUSED_OBJECT) {
929 		if (!dmu_objset_projectquota_present(zfsvfs->z_os)) {
930 			if (dmu_objset_projectquota_upgradable(zfsvfs->z_os)) {
931 				dsl_pool_config_enter(
932 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
933 				dmu_objset_id_quota_upgrade(zfsvfs->z_os);
934 				dsl_pool_config_exit(
935 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
936 			}
937 			return (B_FALSE);
938 		}
939 		quotaobj = zfsvfs->z_projectobjquota_obj;
940 	} else if (usedobj == DMU_USERUSED_OBJECT) {
941 		quotaobj = zfsvfs->z_userobjquota_obj;
942 	} else if (usedobj == DMU_GROUPUSED_OBJECT) {
943 		quotaobj = zfsvfs->z_groupobjquota_obj;
944 	} else {
945 		return (B_FALSE);
946 	}
947 	if (quotaobj == 0 || zfsvfs->z_replay)
948 		return (B_FALSE);
949 
950 	(void) sprintf(buf, "%llx", (longlong_t)id);
951 	err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
952 	if (err != 0)
953 		return (B_FALSE);
954 
955 	(void) sprintf(buf, DMU_OBJACCT_PREFIX "%llx", (longlong_t)id);
956 	err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
957 	if (err != 0)
958 		return (B_FALSE);
959 	return (used >= quota);
960 }
961 
962 boolean_t
963 zfs_id_overblockquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
964 {
965 	char buf[20];
966 	uint64_t used, quota, quotaobj;
967 	int err;
968 
969 	if (usedobj == DMU_PROJECTUSED_OBJECT) {
970 		if (!dmu_objset_projectquota_present(zfsvfs->z_os)) {
971 			if (dmu_objset_projectquota_upgradable(zfsvfs->z_os)) {
972 				dsl_pool_config_enter(
973 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
974 				dmu_objset_id_quota_upgrade(zfsvfs->z_os);
975 				dsl_pool_config_exit(
976 				    dmu_objset_pool(zfsvfs->z_os), FTAG);
977 			}
978 			return (B_FALSE);
979 		}
980 		quotaobj = zfsvfs->z_projectquota_obj;
981 	} else if (usedobj == DMU_USERUSED_OBJECT) {
982 		quotaobj = zfsvfs->z_userquota_obj;
983 	} else if (usedobj == DMU_GROUPUSED_OBJECT) {
984 		quotaobj = zfsvfs->z_groupquota_obj;
985 	} else {
986 		return (B_FALSE);
987 	}
988 	if (quotaobj == 0 || zfsvfs->z_replay)
989 		return (B_FALSE);
990 
991 	(void) sprintf(buf, "%llx", (longlong_t)id);
992 	err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
993 	if (err != 0)
994 		return (B_FALSE);
995 
996 	err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
997 	if (err != 0)
998 		return (B_FALSE);
999 	return (used >= quota);
1000 }
1001 
1002 boolean_t
1003 zfs_id_overquota(zfsvfs_t *zfsvfs, uint64_t usedobj, uint64_t id)
1004 {
1005 	return (zfs_id_overblockquota(zfsvfs, usedobj, id) ||
1006 	    zfs_id_overobjquota(zfsvfs, usedobj, id));
1007 }
1008 
1009 /*
1010  * Associate this zfsvfs with the given objset, which must be owned.
1011  * This will cache a bunch of on-disk state from the objset in the
1012  * zfsvfs.
1013  */
1014 static int
1015 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
1016 {
1017 	int error;
1018 	uint64_t val;
1019 
1020 	zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
1021 	zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
1022 	zfsvfs->z_os = os;
1023 
1024 	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
1025 	if (error != 0)
1026 		return (error);
1027 	if (zfsvfs->z_version >
1028 	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
1029 		(void) printf("Can't mount a version %lld file system "
1030 		    "on a version %lld pool\n. Pool must be upgraded to mount "
1031 		    "this file system.", (u_longlong_t)zfsvfs->z_version,
1032 		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
1033 		return (SET_ERROR(ENOTSUP));
1034 	}
1035 	error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
1036 	if (error != 0)
1037 		return (error);
1038 	zfsvfs->z_norm = (int)val;
1039 
1040 	error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
1041 	if (error != 0)
1042 		return (error);
1043 	zfsvfs->z_utf8 = (val != 0);
1044 
1045 	error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
1046 	if (error != 0)
1047 		return (error);
1048 	zfsvfs->z_case = (uint_t)val;
1049 
1050 	/*
1051 	 * Fold case on file systems that are always or sometimes case
1052 	 * insensitive.
1053 	 */
1054 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
1055 	    zfsvfs->z_case == ZFS_CASE_MIXED)
1056 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1057 
1058 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1059 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1060 
1061 	uint64_t sa_obj = 0;
1062 	if (zfsvfs->z_use_sa) {
1063 		/* should either have both of these objects or none */
1064 		error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
1065 		    &sa_obj);
1066 		if (error != 0)
1067 			return (error);
1068 	}
1069 
1070 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1071 	    &zfsvfs->z_attr_table);
1072 	if (error != 0)
1073 		return (error);
1074 
1075 	if (zfsvfs->z_version >= ZPL_VERSION_SA)
1076 		sa_register_update_callback(os, zfs_sa_upgrade);
1077 
1078 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
1079 	    &zfsvfs->z_root);
1080 	if (error != 0)
1081 		return (error);
1082 	ASSERT(zfsvfs->z_root != 0);
1083 
1084 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
1085 	    &zfsvfs->z_unlinkedobj);
1086 	if (error != 0)
1087 		return (error);
1088 
1089 	error = zap_lookup(os, MASTER_NODE_OBJ,
1090 	    zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
1091 	    8, 1, &zfsvfs->z_userquota_obj);
1092 	if (error == ENOENT)
1093 		zfsvfs->z_userquota_obj = 0;
1094 	else if (error != 0)
1095 		return (error);
1096 
1097 	error = zap_lookup(os, MASTER_NODE_OBJ,
1098 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
1099 	    8, 1, &zfsvfs->z_groupquota_obj);
1100 	if (error == ENOENT)
1101 		zfsvfs->z_groupquota_obj = 0;
1102 	else if (error != 0)
1103 		return (error);
1104 
1105 	error = zap_lookup(os, MASTER_NODE_OBJ,
1106 	    zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA],
1107 	    8, 1, &zfsvfs->z_projectquota_obj);
1108 	if (error == ENOENT)
1109 		zfsvfs->z_projectquota_obj = 0;
1110 	else if (error != 0)
1111 		return (error);
1112 
1113 	error = zap_lookup(os, MASTER_NODE_OBJ,
1114 	    zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
1115 	    8, 1, &zfsvfs->z_userobjquota_obj);
1116 	if (error == ENOENT)
1117 		zfsvfs->z_userobjquota_obj = 0;
1118 	else if (error != 0)
1119 		return (error);
1120 
1121 	error = zap_lookup(os, MASTER_NODE_OBJ,
1122 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
1123 	    8, 1, &zfsvfs->z_groupobjquota_obj);
1124 	if (error == ENOENT)
1125 		zfsvfs->z_groupobjquota_obj = 0;
1126 	else if (error != 0)
1127 		return (error);
1128 
1129 	error = zap_lookup(os, MASTER_NODE_OBJ,
1130 	    zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA],
1131 	    8, 1, &zfsvfs->z_projectobjquota_obj);
1132 	if (error == ENOENT)
1133 		zfsvfs->z_projectobjquota_obj = 0;
1134 	else if (error != 0)
1135 		return (error);
1136 
1137 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
1138 	    &zfsvfs->z_fuid_obj);
1139 	if (error == ENOENT)
1140 		zfsvfs->z_fuid_obj = 0;
1141 	else if (error != 0)
1142 		return (error);
1143 
1144 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
1145 	    &zfsvfs->z_shares_dir);
1146 	if (error == ENOENT)
1147 		zfsvfs->z_shares_dir = 0;
1148 	else if (error != 0)
1149 		return (error);
1150 
1151 	return (0);
1152 }
1153 
1154 int
1155 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp)
1156 {
1157 	objset_t *os;
1158 	zfsvfs_t *zfsvfs;
1159 	int error;
1160 	boolean_t ro = (readonly || (strchr(osname, '@') != NULL));
1161 
1162 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1163 
1164 	error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os);
1165 	if (error != 0) {
1166 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
1167 		return (error);
1168 	}
1169 
1170 	error = zfsvfs_create_impl(zfvp, zfsvfs, os);
1171 	if (error != 0) {
1172 		dmu_objset_disown(os, B_TRUE, zfsvfs);
1173 	}
1174 	return (error);
1175 }
1176 
1177 
1178 int
1179 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
1180 {
1181 	int error;
1182 
1183 	zfsvfs->z_vfs = NULL;
1184 	zfsvfs->z_parent = zfsvfs;
1185 
1186 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1187 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1188 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1189 	    offsetof(znode_t, z_link_node));
1190 	rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1191 	rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1192 	rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1193 	for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1194 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1195 
1196 	error = zfsvfs_init(zfsvfs, os);
1197 	if (error != 0) {
1198 		*zfvp = NULL;
1199 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
1200 		return (error);
1201 	}
1202 
1203 	zfsvfs->z_drain_task = TASKQID_INVALID;
1204 	zfsvfs->z_draining = B_FALSE;
1205 	zfsvfs->z_drain_cancel = B_TRUE;
1206 
1207 	*zfvp = zfsvfs;
1208 	return (0);
1209 }
1210 
1211 static int
1212 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1213 {
1214 	int error;
1215 
1216 	error = zfs_register_callbacks(zfsvfs->z_vfs);
1217 	if (error)
1218 		return (error);
1219 
1220 	zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1221 
1222 	/*
1223 	 * If we are not mounting (ie: online recv), then we don't
1224 	 * have to worry about replaying the log as we blocked all
1225 	 * operations out since we closed the ZIL.
1226 	 */
1227 	if (mounting) {
1228 		boolean_t readonly;
1229 
1230 		/*
1231 		 * During replay we remove the read only flag to
1232 		 * allow replays to succeed.
1233 		 */
1234 		readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1235 		if (readonly != 0) {
1236 			zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1237 		} else {
1238 			zfs_unlinked_drain(zfsvfs);
1239 		}
1240 
1241 		/*
1242 		 * Parse and replay the intent log.
1243 		 *
1244 		 * Because of ziltest, this must be done after
1245 		 * zfs_unlinked_drain().  (Further note: ziltest
1246 		 * doesn't use readonly mounts, where
1247 		 * zfs_unlinked_drain() isn't called.)  This is because
1248 		 * ziltest causes spa_sync() to think it's committed,
1249 		 * but actually it is not, so the intent log contains
1250 		 * many txg's worth of changes.
1251 		 *
1252 		 * In particular, if object N is in the unlinked set in
1253 		 * the last txg to actually sync, then it could be
1254 		 * actually freed in a later txg and then reallocated
1255 		 * in a yet later txg.  This would write a "create
1256 		 * object N" record to the intent log.  Normally, this
1257 		 * would be fine because the spa_sync() would have
1258 		 * written out the fact that object N is free, before
1259 		 * we could write the "create object N" intent log
1260 		 * record.
1261 		 *
1262 		 * But when we are in ziltest mode, we advance the "open
1263 		 * txg" without actually spa_sync()-ing the changes to
1264 		 * disk.  So we would see that object N is still
1265 		 * allocated and in the unlinked set, and there is an
1266 		 * intent log record saying to allocate it.
1267 		 */
1268 		if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1269 			if (zil_replay_disable) {
1270 				zil_destroy(zfsvfs->z_log, B_FALSE);
1271 			} else {
1272 				zfsvfs->z_replay = B_TRUE;
1273 				zil_replay(zfsvfs->z_os, zfsvfs,
1274 				    zfs_replay_vector);
1275 				zfsvfs->z_replay = B_FALSE;
1276 			}
1277 		}
1278 
1279 		/* restore readonly bit */
1280 		if (readonly != 0)
1281 			zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
1282 	}
1283 
1284 	/*
1285 	 * Set the objset user_ptr to track its zfsvfs.
1286 	 */
1287 	mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1288 	dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1289 	mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1290 
1291 	return (0);
1292 }
1293 
1294 void
1295 zfsvfs_free(zfsvfs_t *zfsvfs)
1296 {
1297 	int i;
1298 	extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1299 
1300 	/*
1301 	 * This is a barrier to prevent the filesystem from going away in
1302 	 * zfs_znode_move() until we can safely ensure that the filesystem is
1303 	 * not unmounted. We consider the filesystem valid before the barrier
1304 	 * and invalid after the barrier.
1305 	 */
1306 	rw_enter(&zfsvfs_lock, RW_READER);
1307 	rw_exit(&zfsvfs_lock);
1308 
1309 	zfs_fuid_destroy(zfsvfs);
1310 
1311 	mutex_destroy(&zfsvfs->z_znodes_lock);
1312 	mutex_destroy(&zfsvfs->z_lock);
1313 	list_destroy(&zfsvfs->z_all_znodes);
1314 	rrm_destroy(&zfsvfs->z_teardown_lock);
1315 	rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1316 	rw_destroy(&zfsvfs->z_fuid_lock);
1317 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1318 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1319 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1320 }
1321 
1322 static void
1323 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1324 {
1325 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1326 	if (zfsvfs->z_vfs) {
1327 		if (zfsvfs->z_use_fuids) {
1328 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1329 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1330 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1331 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1332 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1333 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1334 		} else {
1335 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1336 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1337 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1338 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1339 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1340 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1341 		}
1342 	}
1343 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1344 }
1345 
1346 static int
1347 zfs_domount(vfs_t *vfsp, char *osname)
1348 {
1349 	dev_t mount_dev;
1350 	uint64_t recordsize, fsid_guid;
1351 	int error = 0;
1352 	zfsvfs_t *zfsvfs;
1353 	boolean_t readonly = vfsp->vfs_flag & VFS_RDONLY ? B_TRUE : B_FALSE;
1354 
1355 	ASSERT(vfsp);
1356 	ASSERT(osname);
1357 
1358 	error = zfsvfs_create(osname, readonly, &zfsvfs);
1359 	if (error)
1360 		return (error);
1361 	zfsvfs->z_vfs = vfsp;
1362 
1363 	/* Initialize the generic filesystem structure. */
1364 	vfsp->vfs_bcount = 0;
1365 	vfsp->vfs_data = NULL;
1366 
1367 	if (zfs_create_unique_device(&mount_dev) == -1) {
1368 		error = SET_ERROR(ENODEV);
1369 		goto out;
1370 	}
1371 	ASSERT(vfs_devismounted(mount_dev) == 0);
1372 
1373 	if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1374 	    NULL))
1375 		goto out;
1376 
1377 	vfsp->vfs_dev = mount_dev;
1378 	vfsp->vfs_fstype = zfsfstype;
1379 	vfsp->vfs_bsize = recordsize;
1380 	vfsp->vfs_flag |= VFS_NOTRUNC;
1381 	vfsp->vfs_data = zfsvfs;
1382 
1383 	/*
1384 	 * The fsid is 64 bits, composed of an 8-bit fs type, which
1385 	 * separates our fsid from any other filesystem types, and a
1386 	 * 56-bit objset unique ID.  The objset unique ID is unique to
1387 	 * all objsets open on this system, provided by unique_create().
1388 	 * The 8-bit fs type must be put in the low bits of fsid[1]
1389 	 * because that's where other Solaris filesystems put it.
1390 	 */
1391 	fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1392 	ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1393 	vfsp->vfs_fsid.val[0] = fsid_guid;
1394 	vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1395 	    zfsfstype & 0xFF;
1396 
1397 	/*
1398 	 * Set features for file system.
1399 	 */
1400 	zfs_set_fuid_feature(zfsvfs);
1401 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1402 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1403 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1404 		vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1405 	} else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1406 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1407 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1408 	}
1409 	vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1410 
1411 	if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1412 		uint64_t pval;
1413 
1414 		atime_changed_cb(zfsvfs, B_FALSE);
1415 		readonly_changed_cb(zfsvfs, B_TRUE);
1416 		if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1417 			goto out;
1418 		xattr_changed_cb(zfsvfs, pval);
1419 		zfsvfs->z_issnap = B_TRUE;
1420 		zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1421 
1422 		mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1423 		dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1424 		mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1425 	} else {
1426 		error = zfsvfs_setup(zfsvfs, B_TRUE);
1427 	}
1428 
1429 	if (!zfsvfs->z_issnap)
1430 		zfsctl_create(zfsvfs);
1431 out:
1432 	if (error) {
1433 		dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1434 		zfsvfs_free(zfsvfs);
1435 	} else {
1436 		atomic_inc_32(&zfs_active_fs_count);
1437 	}
1438 
1439 	return (error);
1440 }
1441 
1442 void
1443 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1444 {
1445 	objset_t *os = zfsvfs->z_os;
1446 
1447 	if (!dmu_objset_is_snapshot(os))
1448 		dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1449 }
1450 
1451 /*
1452  * Convert a decimal digit string to a uint64_t integer.
1453  */
1454 static int
1455 str_to_uint64(char *str, uint64_t *objnum)
1456 {
1457 	uint64_t num = 0;
1458 
1459 	while (*str) {
1460 		if (*str < '0' || *str > '9')
1461 			return (SET_ERROR(EINVAL));
1462 
1463 		num = num*10 + *str++ - '0';
1464 	}
1465 
1466 	*objnum = num;
1467 	return (0);
1468 }
1469 
1470 /*
1471  * The boot path passed from the boot loader is in the form of
1472  * "rootpool-name/root-filesystem-object-number'. Convert this
1473  * string to a dataset name: "rootpool-name/root-filesystem-name".
1474  */
1475 static int
1476 zfs_parse_bootfs(char *bpath, char *outpath)
1477 {
1478 	char *slashp;
1479 	uint64_t objnum;
1480 	int error;
1481 
1482 	if (*bpath == 0 || *bpath == '/')
1483 		return (SET_ERROR(EINVAL));
1484 
1485 	(void) strcpy(outpath, bpath);
1486 
1487 	slashp = strchr(bpath, '/');
1488 
1489 	/* if no '/', just return the pool name */
1490 	if (slashp == NULL) {
1491 		return (0);
1492 	}
1493 
1494 	/* if not a number, just return the root dataset name */
1495 	if (str_to_uint64(slashp+1, &objnum)) {
1496 		return (0);
1497 	}
1498 
1499 	*slashp = '\0';
1500 	error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1501 	*slashp = '/';
1502 
1503 	return (error);
1504 }
1505 
1506 /*
1507  * Check that the hex label string is appropriate for the dataset being
1508  * mounted into the global_zone proper.
1509  *
1510  * Return an error if the hex label string is not default or
1511  * admin_low/admin_high.  For admin_low labels, the corresponding
1512  * dataset must be readonly.
1513  */
1514 int
1515 zfs_check_global_label(const char *dsname, const char *hexsl)
1516 {
1517 	if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1518 		return (0);
1519 	if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1520 		return (0);
1521 	if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1522 		/* must be readonly */
1523 		uint64_t rdonly;
1524 
1525 		if (dsl_prop_get_integer(dsname,
1526 		    zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1527 			return (SET_ERROR(EACCES));
1528 		return (rdonly ? 0 : EACCES);
1529 	}
1530 	return (SET_ERROR(EACCES));
1531 }
1532 
1533 static int
1534 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct statvfs64 *statp,
1535     uint32_t bshift)
1536 {
1537 	char buf[20 + DMU_OBJACCT_PREFIX_LEN];
1538 	uint64_t offset = DMU_OBJACCT_PREFIX_LEN;
1539 	uint64_t quota;
1540 	uint64_t used;
1541 	int err;
1542 
1543 	strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
1544 	err = id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset, B_FALSE);
1545 	if (err)
1546 		return (err);
1547 
1548 	if (zfsvfs->z_projectquota_obj == 0)
1549 		goto objs;
1550 
1551 	err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj,
1552 	    buf + offset, 8, 1, &quota);
1553 	if (err == ENOENT)
1554 		goto objs;
1555 	else if (err)
1556 		return (err);
1557 
1558 	err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1559 	    buf + offset, 8, 1, &used);
1560 	if (unlikely(err == ENOENT)) {
1561 		uint32_t blksize;
1562 		u_longlong_t nblocks;
1563 
1564 		/*
1565 		 * Quota accounting is async, so it is possible race case.
1566 		 * There is at least one object with the given project ID.
1567 		 */
1568 		sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
1569 		if (unlikely(zp->z_blksz == 0))
1570 			blksize = zfsvfs->z_max_blksz;
1571 
1572 		used = blksize * nblocks;
1573 	} else if (err) {
1574 		return (err);
1575 	}
1576 
1577 	statp->f_blocks = quota >> bshift;
1578 	statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0;
1579 	statp->f_bavail = statp->f_bfree;
1580 
1581 objs:
1582 	if (zfsvfs->z_projectobjquota_obj == 0)
1583 		return (0);
1584 
1585 	err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj,
1586 	    buf + offset, 8, 1, &quota);
1587 	if (err == ENOENT)
1588 		return (0);
1589 	else if (err)
1590 		return (err);
1591 
1592 	err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1593 	    buf, 8, 1, &used);
1594 	if (unlikely(err == ENOENT)) {
1595 		/*
1596 		 * Quota accounting is async, so it is possible race case.
1597 		 * There is at least one object with the given project ID.
1598 		 */
1599 		used = 1;
1600 	} else if (err) {
1601 		return (err);
1602 	}
1603 
1604 	statp->f_files = quota;
1605 	statp->f_ffree = (quota > used) ? (quota - used) : 0;
1606 
1607 	return (0);
1608 }
1609 
1610 /*
1611  * Determine whether the mount is allowed according to MAC check.
1612  * by comparing (where appropriate) label of the dataset against
1613  * the label of the zone being mounted into.  If the dataset has
1614  * no label, create one.
1615  *
1616  * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1617  */
1618 static int
1619 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1620 {
1621 	int		error, retv;
1622 	zone_t		*mntzone = NULL;
1623 	ts_label_t	*mnt_tsl;
1624 	bslabel_t	*mnt_sl;
1625 	bslabel_t	ds_sl;
1626 	char		ds_hexsl[MAXNAMELEN];
1627 
1628 	retv = EACCES;				/* assume the worst */
1629 
1630 	/*
1631 	 * Start by getting the dataset label if it exists.
1632 	 */
1633 	error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1634 	    1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1635 	if (error)
1636 		return (SET_ERROR(EACCES));
1637 
1638 	/*
1639 	 * If labeling is NOT enabled, then disallow the mount of datasets
1640 	 * which have a non-default label already.  No other label checks
1641 	 * are needed.
1642 	 */
1643 	if (!is_system_labeled()) {
1644 		if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1645 			return (0);
1646 		return (SET_ERROR(EACCES));
1647 	}
1648 
1649 	/*
1650 	 * Get the label of the mountpoint.  If mounting into the global
1651 	 * zone (i.e. mountpoint is not within an active zone and the
1652 	 * zoned property is off), the label must be default or
1653 	 * admin_low/admin_high only; no other checks are needed.
1654 	 */
1655 	mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1656 	if (mntzone->zone_id == GLOBAL_ZONEID) {
1657 		uint64_t zoned;
1658 
1659 		zone_rele(mntzone);
1660 
1661 		if (dsl_prop_get_integer(osname,
1662 		    zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1663 			return (SET_ERROR(EACCES));
1664 		if (!zoned)
1665 			return (zfs_check_global_label(osname, ds_hexsl));
1666 		else
1667 			/*
1668 			 * This is the case of a zone dataset being mounted
1669 			 * initially, before the zone has been fully created;
1670 			 * allow this mount into global zone.
1671 			 */
1672 			return (0);
1673 	}
1674 
1675 	mnt_tsl = mntzone->zone_slabel;
1676 	ASSERT(mnt_tsl != NULL);
1677 	label_hold(mnt_tsl);
1678 	mnt_sl = label2bslabel(mnt_tsl);
1679 
1680 	if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1681 		/*
1682 		 * The dataset doesn't have a real label, so fabricate one.
1683 		 */
1684 		char *str = NULL;
1685 
1686 		if (l_to_str_internal(mnt_sl, &str) == 0 &&
1687 		    dsl_prop_set_string(osname,
1688 		    zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1689 		    ZPROP_SRC_LOCAL, str) == 0)
1690 			retv = 0;
1691 		if (str != NULL)
1692 			kmem_free(str, strlen(str) + 1);
1693 	} else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1694 		/*
1695 		 * Now compare labels to complete the MAC check.  If the
1696 		 * labels are equal then allow access.  If the mountpoint
1697 		 * label dominates the dataset label, allow readonly access.
1698 		 * Otherwise, access is denied.
1699 		 */
1700 		if (blequal(mnt_sl, &ds_sl))
1701 			retv = 0;
1702 		else if (bldominates(mnt_sl, &ds_sl)) {
1703 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1704 			retv = 0;
1705 		}
1706 	}
1707 
1708 	label_rele(mnt_tsl);
1709 	zone_rele(mntzone);
1710 	return (retv);
1711 }
1712 
1713 static int
1714 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1715 {
1716 	int error = 0;
1717 	static int zfsrootdone = 0;
1718 	zfsvfs_t *zfsvfs = NULL;
1719 	znode_t *zp = NULL;
1720 	vnode_t *vp = NULL;
1721 	char *zfs_bootfs;
1722 	char *zfs_devid;
1723 
1724 	ASSERT(vfsp);
1725 
1726 	/*
1727 	 * The filesystem that we mount as root is defined in the
1728 	 * boot property "zfs-bootfs" with a format of
1729 	 * "poolname/root-dataset-objnum".
1730 	 */
1731 	if (why == ROOT_INIT) {
1732 		if (zfsrootdone++)
1733 			return (SET_ERROR(EBUSY));
1734 		/*
1735 		 * the process of doing a spa_load will require the
1736 		 * clock to be set before we could (for example) do
1737 		 * something better by looking at the timestamp on
1738 		 * an uberblock, so just set it to -1.
1739 		 */
1740 		clkset(-1);
1741 
1742 		if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1743 			cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1744 			    "bootfs name");
1745 			return (SET_ERROR(EINVAL));
1746 		}
1747 		zfs_devid = spa_get_bootprop("diskdevid");
1748 		error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1749 		if (zfs_devid)
1750 			spa_free_bootprop(zfs_devid);
1751 		if (error) {
1752 			spa_free_bootprop(zfs_bootfs);
1753 			cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1754 			    error);
1755 			return (error);
1756 		}
1757 		if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1758 			spa_free_bootprop(zfs_bootfs);
1759 			cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1760 			    error);
1761 			return (error);
1762 		}
1763 
1764 		spa_free_bootprop(zfs_bootfs);
1765 
1766 		if (error = vfs_lock(vfsp))
1767 			return (error);
1768 
1769 		if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1770 			cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1771 			goto out;
1772 		}
1773 
1774 		zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1775 		ASSERT(zfsvfs);
1776 		if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1777 			cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1778 			goto out;
1779 		}
1780 
1781 		vp = ZTOV(zp);
1782 		mutex_enter(&vp->v_lock);
1783 		vp->v_flag |= VROOT;
1784 		mutex_exit(&vp->v_lock);
1785 		rootvp = vp;
1786 
1787 		/*
1788 		 * Leave rootvp held.  The root file system is never unmounted.
1789 		 */
1790 
1791 		vfs_add((struct vnode *)0, vfsp,
1792 		    (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1793 out:
1794 		vfs_unlock(vfsp);
1795 		return (error);
1796 	} else if (why == ROOT_REMOUNT) {
1797 		readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1798 		vfsp->vfs_flag |= VFS_REMOUNT;
1799 
1800 		/* refresh mount options */
1801 		zfs_unregister_callbacks(vfsp->vfs_data);
1802 		return (zfs_register_callbacks(vfsp));
1803 
1804 	} else if (why == ROOT_UNMOUNT) {
1805 		zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1806 		(void) zfs_sync(vfsp, 0, 0);
1807 		return (0);
1808 	}
1809 
1810 	/*
1811 	 * if "why" is equal to anything else other than ROOT_INIT,
1812 	 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1813 	 */
1814 	return (SET_ERROR(ENOTSUP));
1815 }
1816 
1817 /*ARGSUSED*/
1818 static int
1819 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1820 {
1821 	char		*osname;
1822 	pathname_t	spn;
1823 	int		error = 0;
1824 	uio_seg_t	fromspace = (uap->flags & MS_SYSSPACE) ?
1825 	    UIO_SYSSPACE : UIO_USERSPACE;
1826 	int		canwrite;
1827 
1828 	if (mvp->v_type != VDIR)
1829 		return (SET_ERROR(ENOTDIR));
1830 
1831 	mutex_enter(&mvp->v_lock);
1832 	if ((uap->flags & MS_REMOUNT) == 0 &&
1833 	    (uap->flags & MS_OVERLAY) == 0 &&
1834 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1835 		mutex_exit(&mvp->v_lock);
1836 		return (SET_ERROR(EBUSY));
1837 	}
1838 	mutex_exit(&mvp->v_lock);
1839 
1840 	/*
1841 	 * ZFS does not support passing unparsed data in via MS_DATA.
1842 	 * Users should use the MS_OPTIONSTR interface; this means
1843 	 * that all option parsing is already done and the options struct
1844 	 * can be interrogated.
1845 	 */
1846 	if ((uap->flags & MS_DATA) && uap->datalen > 0)
1847 		return (SET_ERROR(EINVAL));
1848 
1849 	/*
1850 	 * Get the objset name (the "special" mount argument).
1851 	 */
1852 	if (error = pn_get(uap->spec, fromspace, &spn))
1853 		return (error);
1854 
1855 	osname = spn.pn_path;
1856 
1857 	/*
1858 	 * Check for mount privilege?
1859 	 *
1860 	 * If we don't have privilege then see if
1861 	 * we have local permission to allow it
1862 	 */
1863 	error = secpolicy_fs_mount(cr, mvp, vfsp);
1864 	if (error) {
1865 		if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1866 			vattr_t		vattr;
1867 
1868 			/*
1869 			 * Make sure user is the owner of the mount point
1870 			 * or has sufficient privileges.
1871 			 */
1872 
1873 			vattr.va_mask = AT_UID;
1874 
1875 			if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1876 				goto out;
1877 			}
1878 
1879 			if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1880 			    VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1881 				goto out;
1882 			}
1883 			secpolicy_fs_mount_clearopts(cr, vfsp);
1884 		} else {
1885 			goto out;
1886 		}
1887 	}
1888 
1889 	/*
1890 	 * Refuse to mount a filesystem if we are in a local zone and the
1891 	 * dataset is not visible.
1892 	 */
1893 	if (!INGLOBALZONE(curproc) &&
1894 	    (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1895 		error = SET_ERROR(EPERM);
1896 		goto out;
1897 	}
1898 
1899 	error = zfs_mount_label_policy(vfsp, osname);
1900 	if (error)
1901 		goto out;
1902 
1903 	/*
1904 	 * When doing a remount, we simply refresh our temporary properties
1905 	 * according to those options set in the current VFS options.
1906 	 */
1907 	if (uap->flags & MS_REMOUNT) {
1908 		/* refresh mount options */
1909 		zfs_unregister_callbacks(vfsp->vfs_data);
1910 		error = zfs_register_callbacks(vfsp);
1911 		goto out;
1912 	}
1913 
1914 	error = zfs_domount(vfsp, osname);
1915 
1916 	/*
1917 	 * Add an extra VFS_HOLD on our parent vfs so that it can't
1918 	 * disappear due to a forced unmount.
1919 	 */
1920 	if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1921 		VFS_HOLD(mvp->v_vfsp);
1922 
1923 out:
1924 	pn_free(&spn);
1925 	return (error);
1926 }
1927 
1928 static int
1929 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1930 {
1931 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1932 	dev32_t d32;
1933 	uint64_t refdbytes, availbytes, usedobjs, availobjs;
1934 	int err = 0;
1935 
1936 	ZFS_ENTER(zfsvfs);
1937 
1938 	dmu_objset_space(zfsvfs->z_os,
1939 	    &refdbytes, &availbytes, &usedobjs, &availobjs);
1940 
1941 	/*
1942 	 * The underlying storage pool actually uses multiple block sizes.
1943 	 * We report the fragsize as the smallest block size we support,
1944 	 * and we report our blocksize as the filesystem's maximum blocksize.
1945 	 */
1946 	statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1947 	statp->f_bsize = zfsvfs->z_max_blksz;
1948 
1949 	/*
1950 	 * The following report "total" blocks of various kinds in the
1951 	 * file system, but reported in terms of f_frsize - the
1952 	 * "fragment" size.
1953 	 */
1954 
1955 	statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1956 	statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1957 	statp->f_bavail = statp->f_bfree; /* no root reservation */
1958 
1959 	/*
1960 	 * statvfs() should really be called statufs(), because it assumes
1961 	 * static metadata.  ZFS doesn't preallocate files, so the best
1962 	 * we can do is report the max that could possibly fit in f_files,
1963 	 * and that minus the number actually used in f_ffree.
1964 	 * For f_ffree, report the smaller of the number of object available
1965 	 * and the number of blocks (each object will take at least a block).
1966 	 */
1967 	statp->f_ffree = MIN(availobjs, statp->f_bfree);
1968 	statp->f_favail = statp->f_ffree;	/* no "root reservation" */
1969 	statp->f_files = statp->f_ffree + usedobjs;
1970 
1971 	(void) cmpldev(&d32, vfsp->vfs_dev);
1972 	statp->f_fsid = d32;
1973 
1974 	/*
1975 	 * We're a zfs filesystem.
1976 	 */
1977 	(void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1978 
1979 	statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1980 
1981 	statp->f_namemax = MAXNAMELEN - 1;
1982 
1983 	/*
1984 	 * We have all of 32 characters to stuff a string here.
1985 	 * Is there anything useful we could/should provide?
1986 	 */
1987 	bzero(statp->f_fstr, sizeof (statp->f_fstr));
1988 
1989 	if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
1990 	    dmu_objset_projectquota_present(zfsvfs->z_os)) {
1991 		znode_t *zp;
1992 
1993 		/*
1994 		 * In ZoL, zfs_statvfs is passed a Linux dentry (directory
1995 		 * entry), instead of a vfsp. The ZoL code uses the dentry
1996 		 * to get the znode from the dentry's inode. This represents
1997 		 * whatever filename was passed to the user-level statvfs
1998 		 * syscall.
1999 		 *
2000 		 * We're using the VFS root znode here, so this represents a
2001 		 * potential difference from ZoL.
2002 		 */
2003 		if (zfs_zget(zfsvfs, zfsvfs->z_root, &zp) == 0) {
2004 			uint32_t bshift = ddi_fls(statp->f_bsize) - 1;
2005 
2006 			if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid &&
2007 			    zpl_is_valid_projid(zp->z_projid))
2008 				err = zfs_statfs_project(zfsvfs, zp, statp,
2009 				    bshift);
2010 			VN_RELE(ZTOV(zp));
2011 		}
2012 	}
2013 
2014 	ZFS_EXIT(zfsvfs);
2015 	return (err);
2016 }
2017 
2018 static int
2019 zfs_root(vfs_t *vfsp, vnode_t **vpp)
2020 {
2021 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2022 	znode_t *rootzp;
2023 	int error;
2024 
2025 	ZFS_ENTER(zfsvfs);
2026 
2027 	error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
2028 	if (error == 0)
2029 		*vpp = ZTOV(rootzp);
2030 
2031 	ZFS_EXIT(zfsvfs);
2032 	return (error);
2033 }
2034 
2035 /*
2036  * Teardown the zfsvfs::z_os.
2037  *
2038  * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
2039  * and 'z_teardown_inactive_lock' held.
2040  */
2041 static int
2042 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
2043 {
2044 	znode_t	*zp;
2045 
2046 	zfs_unlinked_drain_stop_wait(zfsvfs);
2047 
2048 	rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
2049 
2050 	if (!unmounting) {
2051 		/*
2052 		 * We purge the parent filesystem's vfsp as the parent
2053 		 * filesystem and all of its snapshots have their vnode's
2054 		 * v_vfsp set to the parent's filesystem's vfsp.  Note,
2055 		 * 'z_parent' is self referential for non-snapshots.
2056 		 */
2057 		(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2058 	}
2059 
2060 	/*
2061 	 * Close the zil. NB: Can't close the zil while zfs_inactive
2062 	 * threads are blocked as zil_close can call zfs_inactive.
2063 	 */
2064 	if (zfsvfs->z_log) {
2065 		zil_close(zfsvfs->z_log);
2066 		zfsvfs->z_log = NULL;
2067 	}
2068 
2069 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
2070 
2071 	/*
2072 	 * If we are not unmounting (ie: online recv) and someone already
2073 	 * unmounted this file system while we were doing the switcheroo,
2074 	 * or a reopen of z_os failed then just bail out now.
2075 	 */
2076 	if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
2077 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
2078 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2079 		return (SET_ERROR(EIO));
2080 	}
2081 
2082 	/*
2083 	 * At this point there are no vops active, and any new vops will
2084 	 * fail with EIO since we have z_teardown_lock for writer (only
2085 	 * relavent for forced unmount).
2086 	 *
2087 	 * Release all holds on dbufs.
2088 	 */
2089 	mutex_enter(&zfsvfs->z_znodes_lock);
2090 	for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
2091 	    zp = list_next(&zfsvfs->z_all_znodes, zp))
2092 		if (zp->z_sa_hdl) {
2093 			ASSERT(ZTOV(zp)->v_count > 0);
2094 			zfs_znode_dmu_fini(zp);
2095 		}
2096 	mutex_exit(&zfsvfs->z_znodes_lock);
2097 
2098 	/*
2099 	 * If we are unmounting, set the unmounted flag and let new vops
2100 	 * unblock.  zfs_inactive will have the unmounted behavior, and all
2101 	 * other vops will fail with EIO.
2102 	 */
2103 	if (unmounting) {
2104 		zfsvfs->z_unmounted = B_TRUE;
2105 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
2106 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2107 	}
2108 
2109 	/*
2110 	 * z_os will be NULL if there was an error in attempting to reopen
2111 	 * zfsvfs, so just return as the properties had already been
2112 	 * unregistered and cached data had been evicted before.
2113 	 */
2114 	if (zfsvfs->z_os == NULL)
2115 		return (0);
2116 
2117 	/*
2118 	 * Unregister properties.
2119 	 */
2120 	zfs_unregister_callbacks(zfsvfs);
2121 
2122 	/*
2123 	 * Evict cached data
2124 	 */
2125 	if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
2126 	    !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
2127 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
2128 	dmu_objset_evict_dbufs(zfsvfs->z_os);
2129 
2130 	return (0);
2131 }
2132 
2133 /*ARGSUSED*/
2134 static int
2135 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
2136 {
2137 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2138 	objset_t *os;
2139 	int ret;
2140 
2141 	ret = secpolicy_fs_unmount(cr, vfsp);
2142 	if (ret) {
2143 		if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
2144 		    ZFS_DELEG_PERM_MOUNT, cr))
2145 			return (ret);
2146 	}
2147 
2148 	/*
2149 	 * We purge the parent filesystem's vfsp as the parent filesystem
2150 	 * and all of its snapshots have their vnode's v_vfsp set to the
2151 	 * parent's filesystem's vfsp.  Note, 'z_parent' is self
2152 	 * referential for non-snapshots.
2153 	 */
2154 	(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2155 
2156 	/*
2157 	 * Unmount any snapshots mounted under .zfs before unmounting the
2158 	 * dataset itself.
2159 	 */
2160 	if (zfsvfs->z_ctldir != NULL &&
2161 	    (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
2162 		return (ret);
2163 	}
2164 
2165 	if (!(fflag & MS_FORCE)) {
2166 		/*
2167 		 * Check the number of active vnodes in the file system.
2168 		 * Our count is maintained in the vfs structure, but the
2169 		 * number is off by 1 to indicate a hold on the vfs
2170 		 * structure itself.
2171 		 *
2172 		 * The '.zfs' directory maintains a reference of its
2173 		 * own, and any active references underneath are
2174 		 * reflected in the vnode count.
2175 		 */
2176 		if (zfsvfs->z_ctldir == NULL) {
2177 			if (vfsp->vfs_count > 1)
2178 				return (SET_ERROR(EBUSY));
2179 		} else {
2180 			if (vfsp->vfs_count > 2 ||
2181 			    zfsvfs->z_ctldir->v_count > 1)
2182 				return (SET_ERROR(EBUSY));
2183 		}
2184 	}
2185 
2186 	vfsp->vfs_flag |= VFS_UNMOUNTED;
2187 
2188 	VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2189 	os = zfsvfs->z_os;
2190 
2191 	/*
2192 	 * z_os will be NULL if there was an error in
2193 	 * attempting to reopen zfsvfs.
2194 	 */
2195 	if (os != NULL) {
2196 		/*
2197 		 * Unset the objset user_ptr.
2198 		 */
2199 		mutex_enter(&os->os_user_ptr_lock);
2200 		dmu_objset_set_user(os, NULL);
2201 		mutex_exit(&os->os_user_ptr_lock);
2202 
2203 		/*
2204 		 * Finally release the objset
2205 		 */
2206 		dmu_objset_disown(os, B_TRUE, zfsvfs);
2207 	}
2208 
2209 	/*
2210 	 * We can now safely destroy the '.zfs' directory node.
2211 	 */
2212 	if (zfsvfs->z_ctldir != NULL)
2213 		zfsctl_destroy(zfsvfs);
2214 
2215 	return (0);
2216 }
2217 
2218 static int
2219 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
2220 {
2221 	zfsvfs_t	*zfsvfs = vfsp->vfs_data;
2222 	znode_t		*zp;
2223 	uint64_t	object = 0;
2224 	uint64_t	fid_gen = 0;
2225 	uint64_t	gen_mask;
2226 	uint64_t	zp_gen;
2227 	int		i, err;
2228 
2229 	*vpp = NULL;
2230 
2231 	ZFS_ENTER(zfsvfs);
2232 
2233 	if (fidp->fid_len == LONG_FID_LEN) {
2234 		zfid_long_t	*zlfid = (zfid_long_t *)fidp;
2235 		uint64_t	objsetid = 0;
2236 		uint64_t	setgen = 0;
2237 
2238 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2239 			objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2240 
2241 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2242 			setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2243 
2244 		ZFS_EXIT(zfsvfs);
2245 
2246 		err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2247 		if (err)
2248 			return (SET_ERROR(EINVAL));
2249 		ZFS_ENTER(zfsvfs);
2250 	}
2251 
2252 	if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2253 		zfid_short_t	*zfid = (zfid_short_t *)fidp;
2254 
2255 		for (i = 0; i < sizeof (zfid->zf_object); i++)
2256 			object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2257 
2258 		for (i = 0; i < sizeof (zfid->zf_gen); i++)
2259 			fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2260 	} else {
2261 		ZFS_EXIT(zfsvfs);
2262 		return (SET_ERROR(EINVAL));
2263 	}
2264 
2265 	/* A zero fid_gen means we are in the .zfs control directories */
2266 	if (fid_gen == 0 &&
2267 	    (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
2268 		*vpp = zfsvfs->z_ctldir;
2269 		ASSERT(*vpp != NULL);
2270 		if (object == ZFSCTL_INO_SNAPDIR) {
2271 			VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2272 			    0, NULL, NULL, NULL, NULL, NULL) == 0);
2273 		} else {
2274 			VN_HOLD(*vpp);
2275 		}
2276 		ZFS_EXIT(zfsvfs);
2277 		return (0);
2278 	}
2279 
2280 	gen_mask = -1ULL >> (64 - 8 * i);
2281 
2282 	dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2283 	if (err = zfs_zget(zfsvfs, object, &zp)) {
2284 		ZFS_EXIT(zfsvfs);
2285 		return (err);
2286 	}
2287 	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2288 	    sizeof (uint64_t));
2289 	zp_gen = zp_gen & gen_mask;
2290 	if (zp_gen == 0)
2291 		zp_gen = 1;
2292 	if (zp->z_unlinked || zp_gen != fid_gen) {
2293 		dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2294 		VN_RELE(ZTOV(zp));
2295 		ZFS_EXIT(zfsvfs);
2296 		return (SET_ERROR(EINVAL));
2297 	}
2298 
2299 	*vpp = ZTOV(zp);
2300 	ZFS_EXIT(zfsvfs);
2301 	return (0);
2302 }
2303 
2304 /*
2305  * Block out VOPs and close zfsvfs_t::z_os
2306  *
2307  * Note, if successful, then we return with the 'z_teardown_lock' and
2308  * 'z_teardown_inactive_lock' write held.  We leave ownership of the underlying
2309  * dataset and objset intact so that they can be atomically handed off during
2310  * a subsequent rollback or recv operation and the resume thereafter.
2311  */
2312 int
2313 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2314 {
2315 	int error;
2316 
2317 	if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2318 		return (error);
2319 
2320 	return (0);
2321 }
2322 
2323 /*
2324  * Rebuild SA and release VOPs.  Note that ownership of the underlying dataset
2325  * is an invariant across any of the operations that can be performed while the
2326  * filesystem was suspended.  Whether it succeeded or failed, the preconditions
2327  * are the same: the relevant objset and associated dataset are owned by
2328  * zfsvfs, held, and long held on entry.
2329  */
2330 int
2331 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2332 {
2333 	int err;
2334 	znode_t *zp;
2335 
2336 	ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2337 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2338 
2339 	/*
2340 	 * We already own this, so just update the objset_t, as the one we
2341 	 * had before may have been evicted.
2342 	 */
2343 	objset_t *os;
2344 	VERIFY3P(ds->ds_owner, ==, zfsvfs);
2345 	VERIFY(dsl_dataset_long_held(ds));
2346 	VERIFY0(dmu_objset_from_ds(ds, &os));
2347 
2348 	err = zfsvfs_init(zfsvfs, os);
2349 	if (err != 0)
2350 		goto bail;
2351 
2352 	VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2353 
2354 	zfs_set_fuid_feature(zfsvfs);
2355 
2356 	/*
2357 	 * Attempt to re-establish all the active znodes with
2358 	 * their dbufs.  If a zfs_rezget() fails, then we'll let
2359 	 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2360 	 * when they try to use their znode.
2361 	 */
2362 	mutex_enter(&zfsvfs->z_znodes_lock);
2363 	for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2364 	    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2365 		(void) zfs_rezget(zp);
2366 	}
2367 	mutex_exit(&zfsvfs->z_znodes_lock);
2368 
2369 	if (((zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) == 0) &&
2370 	    !zfsvfs->z_unmounted) {
2371 		/*
2372 		 * zfs_suspend_fs() could have interrupted freeing
2373 		 * of dnodes. We need to restart this freeing so
2374 		 * that we don't "leak" the space.
2375 		 */
2376 		zfs_unlinked_drain(zfsvfs);
2377 	}
2378 
2379 bail:
2380 	/* release the VOPs */
2381 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
2382 	rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2383 
2384 	if (err) {
2385 		/*
2386 		 * Since we couldn't setup the sa framework, try to force
2387 		 * unmount this file system.
2388 		 */
2389 		if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2390 			(void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2391 	}
2392 	return (err);
2393 }
2394 
2395 static void
2396 zfs_freevfs(vfs_t *vfsp)
2397 {
2398 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2399 
2400 	/*
2401 	 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2402 	 * from zfs_mount().  Release it here.  If we came through
2403 	 * zfs_mountroot() instead, we didn't grab an extra hold, so
2404 	 * skip the VFS_RELE for rootvfs.
2405 	 */
2406 	if (zfsvfs->z_issnap && (vfsp != rootvfs))
2407 		VFS_RELE(zfsvfs->z_parent->z_vfs);
2408 
2409 	zfsvfs_free(zfsvfs);
2410 
2411 	atomic_dec_32(&zfs_active_fs_count);
2412 }
2413 
2414 /*
2415  * VFS_INIT() initialization.  Note that there is no VFS_FINI(),
2416  * so we can't safely do any non-idempotent initialization here.
2417  * Leave that to zfs_init() and zfs_fini(), which are called
2418  * from the module's _init() and _fini() entry points.
2419  */
2420 /*ARGSUSED*/
2421 static int
2422 zfs_vfsinit(int fstype, char *name)
2423 {
2424 	int error;
2425 
2426 	zfsfstype = fstype;
2427 
2428 	/*
2429 	 * Setup vfsops and vnodeops tables.
2430 	 */
2431 	error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2432 	if (error != 0) {
2433 		cmn_err(CE_WARN, "zfs: bad vfs ops template");
2434 	}
2435 
2436 	error = zfs_create_op_tables();
2437 	if (error) {
2438 		zfs_remove_op_tables();
2439 		cmn_err(CE_WARN, "zfs: bad vnode ops template");
2440 		(void) vfs_freevfsops_by_type(zfsfstype);
2441 		return (error);
2442 	}
2443 
2444 	mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2445 
2446 	/*
2447 	 * Unique major number for all zfs mounts.
2448 	 * If we run out of 32-bit minors, we'll getudev() another major.
2449 	 */
2450 	zfs_major = ddi_name_to_major(ZFS_DRIVER);
2451 	zfs_minor = ZFS_MIN_MINOR;
2452 
2453 	return (0);
2454 }
2455 
2456 void
2457 zfs_init(void)
2458 {
2459 	/*
2460 	 * Initialize .zfs directory structures
2461 	 */
2462 	zfsctl_init();
2463 
2464 	/*
2465 	 * Initialize znode cache, vnode ops, etc...
2466 	 */
2467 	zfs_znode_init();
2468 
2469 	dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2470 }
2471 
2472 void
2473 zfs_fini(void)
2474 {
2475 	zfsctl_fini();
2476 	zfs_znode_fini();
2477 }
2478 
2479 int
2480 zfs_busy(void)
2481 {
2482 	return (zfs_active_fs_count != 0);
2483 }
2484 
2485 int
2486 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2487 {
2488 	int error;
2489 	objset_t *os = zfsvfs->z_os;
2490 	dmu_tx_t *tx;
2491 
2492 	if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2493 		return (SET_ERROR(EINVAL));
2494 
2495 	if (newvers < zfsvfs->z_version)
2496 		return (SET_ERROR(EINVAL));
2497 
2498 	if (zfs_spa_version_map(newvers) >
2499 	    spa_version(dmu_objset_spa(zfsvfs->z_os)))
2500 		return (SET_ERROR(ENOTSUP));
2501 
2502 	tx = dmu_tx_create(os);
2503 	dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2504 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2505 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2506 		    ZFS_SA_ATTRS);
2507 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2508 	}
2509 	error = dmu_tx_assign(tx, TXG_WAIT);
2510 	if (error) {
2511 		dmu_tx_abort(tx);
2512 		return (error);
2513 	}
2514 
2515 	error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2516 	    8, 1, &newvers, tx);
2517 
2518 	if (error) {
2519 		dmu_tx_commit(tx);
2520 		return (error);
2521 	}
2522 
2523 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2524 		uint64_t sa_obj;
2525 
2526 		ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2527 		    SPA_VERSION_SA);
2528 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2529 		    DMU_OT_NONE, 0, tx);
2530 
2531 		error = zap_add(os, MASTER_NODE_OBJ,
2532 		    ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2533 		ASSERT0(error);
2534 
2535 		VERIFY(0 == sa_set_sa_object(os, sa_obj));
2536 		sa_register_update_callback(os, zfs_sa_upgrade);
2537 	}
2538 
2539 	spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2540 	    "from %llu to %llu", zfsvfs->z_version, newvers);
2541 
2542 	dmu_tx_commit(tx);
2543 
2544 	zfsvfs->z_version = newvers;
2545 	os->os_version = newvers;
2546 
2547 	zfs_set_fuid_feature(zfsvfs);
2548 
2549 	return (0);
2550 }
2551 
2552 /*
2553  * Read a property stored within the master node.
2554  */
2555 int
2556 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2557 {
2558 	uint64_t *cached_copy = NULL;
2559 
2560 	/*
2561 	 * Figure out where in the objset_t the cached copy would live, if it
2562 	 * is available for the requested property.
2563 	 */
2564 	if (os != NULL) {
2565 		switch (prop) {
2566 		case ZFS_PROP_VERSION:
2567 			cached_copy = &os->os_version;
2568 			break;
2569 		case ZFS_PROP_NORMALIZE:
2570 			cached_copy = &os->os_normalization;
2571 			break;
2572 		case ZFS_PROP_UTF8ONLY:
2573 			cached_copy = &os->os_utf8only;
2574 			break;
2575 		case ZFS_PROP_CASE:
2576 			cached_copy = &os->os_casesensitivity;
2577 			break;
2578 		default:
2579 			break;
2580 		}
2581 	}
2582 	if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) {
2583 		*value = *cached_copy;
2584 		return (0);
2585 	}
2586 
2587 	/*
2588 	 * If the property wasn't cached, look up the file system's value for
2589 	 * the property. For the version property, we look up a slightly
2590 	 * different string.
2591 	 */
2592 	const char *pname;
2593 	int error = ENOENT;
2594 	if (prop == ZFS_PROP_VERSION) {
2595 		pname = ZPL_VERSION_STR;
2596 	} else {
2597 		pname = zfs_prop_to_name(prop);
2598 	}
2599 
2600 	if (os != NULL) {
2601 		ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2602 		error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2603 	}
2604 
2605 	if (error == ENOENT) {
2606 		/* No value set, use the default value */
2607 		switch (prop) {
2608 		case ZFS_PROP_VERSION:
2609 			*value = ZPL_VERSION;
2610 			break;
2611 		case ZFS_PROP_NORMALIZE:
2612 		case ZFS_PROP_UTF8ONLY:
2613 			*value = 0;
2614 			break;
2615 		case ZFS_PROP_CASE:
2616 			*value = ZFS_CASE_SENSITIVE;
2617 			break;
2618 		default:
2619 			return (error);
2620 		}
2621 		error = 0;
2622 	}
2623 
2624 	/*
2625 	 * If one of the methods for getting the property value above worked,
2626 	 * copy it into the objset_t's cache.
2627 	 */
2628 	if (error == 0 && cached_copy != NULL) {
2629 		*cached_copy = *value;
2630 	}
2631 
2632 	return (error);
2633 }
2634 
2635 /*
2636  * Return true if the coresponding vfs's unmounted flag is set.
2637  * Otherwise return false.
2638  * If this function returns true we know VFS unmount has been initiated.
2639  */
2640 boolean_t
2641 zfs_get_vfs_flag_unmounted(objset_t *os)
2642 {
2643 	zfsvfs_t *zfvp;
2644 	boolean_t unmounted = B_FALSE;
2645 
2646 	ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2647 
2648 	mutex_enter(&os->os_user_ptr_lock);
2649 	zfvp = dmu_objset_get_user(os);
2650 	if (zfvp != NULL && zfvp->z_vfs != NULL &&
2651 	    (zfvp->z_vfs->vfs_flag & VFS_UNMOUNTED))
2652 		unmounted = B_TRUE;
2653 	mutex_exit(&os->os_user_ptr_lock);
2654 
2655 	return (unmounted);
2656 }
2657 
2658 static vfsdef_t vfw = {
2659 	VFSDEF_VERSION,
2660 	MNTTYPE_ZFS,
2661 	zfs_vfsinit,
2662 	VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2663 	    VSW_XID|VSW_ZMOUNT,
2664 	&zfs_mntopts
2665 };
2666 
2667 struct modlfs zfs_modlfs = {
2668 	&mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2669 };
2670