xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_vfsops.c (revision a3c49ce110f325a563c245bedc4d533adddb7211)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  */
24 
25 /* Portions Copyright 2010 Robert Milkowski */
26 
27 #include <sys/types.h>
28 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/sysmacros.h>
31 #include <sys/kmem.h>
32 #include <sys/pathname.h>
33 #include <sys/vnode.h>
34 #include <sys/vfs.h>
35 #include <sys/vfs_opreg.h>
36 #include <sys/mntent.h>
37 #include <sys/mount.h>
38 #include <sys/cmn_err.h>
39 #include "fs/fs_subr.h"
40 #include <sys/zfs_znode.h>
41 #include <sys/zfs_dir.h>
42 #include <sys/zil.h>
43 #include <sys/fs/zfs.h>
44 #include <sys/dmu.h>
45 #include <sys/dsl_prop.h>
46 #include <sys/dsl_dataset.h>
47 #include <sys/dsl_deleg.h>
48 #include <sys/spa.h>
49 #include <sys/zap.h>
50 #include <sys/sa.h>
51 #include <sys/varargs.h>
52 #include <sys/policy.h>
53 #include <sys/atomic.h>
54 #include <sys/mkdev.h>
55 #include <sys/modctl.h>
56 #include <sys/refstr.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/zfs_ctldir.h>
59 #include <sys/zfs_fuid.h>
60 #include <sys/bootconf.h>
61 #include <sys/sunddi.h>
62 #include <sys/dnlc.h>
63 #include <sys/dmu_objset.h>
64 #include <sys/spa_boot.h>
65 #include <sys/sa.h>
66 #include "zfs_comutil.h"
67 
68 int zfsfstype;
69 vfsops_t *zfs_vfsops = NULL;
70 static major_t zfs_major;
71 static minor_t zfs_minor;
72 static kmutex_t	zfs_dev_mtx;
73 
74 extern int sys_shutdown;
75 
76 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
77 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
78 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
79 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
80 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
81 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
82 static void zfs_freevfs(vfs_t *vfsp);
83 
84 static const fs_operation_def_t zfs_vfsops_template[] = {
85 	VFSNAME_MOUNT,		{ .vfs_mount = zfs_mount },
86 	VFSNAME_MOUNTROOT,	{ .vfs_mountroot = zfs_mountroot },
87 	VFSNAME_UNMOUNT,	{ .vfs_unmount = zfs_umount },
88 	VFSNAME_ROOT,		{ .vfs_root = zfs_root },
89 	VFSNAME_STATVFS,	{ .vfs_statvfs = zfs_statvfs },
90 	VFSNAME_SYNC,		{ .vfs_sync = zfs_sync },
91 	VFSNAME_VGET,		{ .vfs_vget = zfs_vget },
92 	VFSNAME_FREEVFS,	{ .vfs_freevfs = zfs_freevfs },
93 	NULL,			NULL
94 };
95 
96 static const fs_operation_def_t zfs_vfsops_eio_template[] = {
97 	VFSNAME_FREEVFS,	{ .vfs_freevfs =  zfs_freevfs },
98 	NULL,			NULL
99 };
100 
101 /*
102  * We need to keep a count of active fs's.
103  * This is necessary to prevent our module
104  * from being unloaded after a umount -f
105  */
106 static uint32_t	zfs_active_fs_count = 0;
107 
108 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
109 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
110 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
111 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
112 
113 /*
114  * MO_DEFAULT is not used since the default value is determined
115  * by the equivalent property.
116  */
117 static mntopt_t mntopts[] = {
118 	{ MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
119 	{ MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
120 	{ MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
121 	{ MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
122 };
123 
124 static mntopts_t zfs_mntopts = {
125 	sizeof (mntopts) / sizeof (mntopt_t),
126 	mntopts
127 };
128 
129 /*ARGSUSED*/
130 int
131 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
132 {
133 	/*
134 	 * Data integrity is job one.  We don't want a compromised kernel
135 	 * writing to the storage pool, so we never sync during panic.
136 	 */
137 	if (panicstr)
138 		return (0);
139 
140 	/*
141 	 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
142 	 * to sync metadata, which they would otherwise cache indefinitely.
143 	 * Semantically, the only requirement is that the sync be initiated.
144 	 * The DMU syncs out txgs frequently, so there's nothing to do.
145 	 */
146 	if (flag & SYNC_ATTR)
147 		return (0);
148 
149 	if (vfsp != NULL) {
150 		/*
151 		 * Sync a specific filesystem.
152 		 */
153 		zfsvfs_t *zfsvfs = vfsp->vfs_data;
154 		dsl_pool_t *dp;
155 
156 		ZFS_ENTER(zfsvfs);
157 		dp = dmu_objset_pool(zfsvfs->z_os);
158 
159 		/*
160 		 * If the system is shutting down, then skip any
161 		 * filesystems which may exist on a suspended pool.
162 		 */
163 		if (sys_shutdown && spa_suspended(dp->dp_spa)) {
164 			ZFS_EXIT(zfsvfs);
165 			return (0);
166 		}
167 
168 		if (zfsvfs->z_log != NULL)
169 			zil_commit(zfsvfs->z_log, 0);
170 
171 		ZFS_EXIT(zfsvfs);
172 	} else {
173 		/*
174 		 * Sync all ZFS filesystems.  This is what happens when you
175 		 * run sync(1M).  Unlike other filesystems, ZFS honors the
176 		 * request by waiting for all pools to commit all dirty data.
177 		 */
178 		spa_sync_allpools();
179 	}
180 
181 	return (0);
182 }
183 
184 static int
185 zfs_create_unique_device(dev_t *dev)
186 {
187 	major_t new_major;
188 
189 	do {
190 		ASSERT3U(zfs_minor, <=, MAXMIN32);
191 		minor_t start = zfs_minor;
192 		do {
193 			mutex_enter(&zfs_dev_mtx);
194 			if (zfs_minor >= MAXMIN32) {
195 				/*
196 				 * If we're still using the real major
197 				 * keep out of /dev/zfs and /dev/zvol minor
198 				 * number space.  If we're using a getudev()'ed
199 				 * major number, we can use all of its minors.
200 				 */
201 				if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
202 					zfs_minor = ZFS_MIN_MINOR;
203 				else
204 					zfs_minor = 0;
205 			} else {
206 				zfs_minor++;
207 			}
208 			*dev = makedevice(zfs_major, zfs_minor);
209 			mutex_exit(&zfs_dev_mtx);
210 		} while (vfs_devismounted(*dev) && zfs_minor != start);
211 		if (zfs_minor == start) {
212 			/*
213 			 * We are using all ~262,000 minor numbers for the
214 			 * current major number.  Create a new major number.
215 			 */
216 			if ((new_major = getudev()) == (major_t)-1) {
217 				cmn_err(CE_WARN,
218 				    "zfs_mount: Can't get unique major "
219 				    "device number.");
220 				return (-1);
221 			}
222 			mutex_enter(&zfs_dev_mtx);
223 			zfs_major = new_major;
224 			zfs_minor = 0;
225 
226 			mutex_exit(&zfs_dev_mtx);
227 		} else {
228 			break;
229 		}
230 		/* CONSTANTCONDITION */
231 	} while (1);
232 
233 	return (0);
234 }
235 
236 static void
237 atime_changed_cb(void *arg, uint64_t newval)
238 {
239 	zfsvfs_t *zfsvfs = arg;
240 
241 	if (newval == TRUE) {
242 		zfsvfs->z_atime = TRUE;
243 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
244 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
245 	} else {
246 		zfsvfs->z_atime = FALSE;
247 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
248 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
249 	}
250 }
251 
252 static void
253 xattr_changed_cb(void *arg, uint64_t newval)
254 {
255 	zfsvfs_t *zfsvfs = arg;
256 
257 	if (newval == TRUE) {
258 		/* XXX locking on vfs_flag? */
259 		zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
260 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
261 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
262 	} else {
263 		/* XXX locking on vfs_flag? */
264 		zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
265 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
266 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
267 	}
268 }
269 
270 static void
271 blksz_changed_cb(void *arg, uint64_t newval)
272 {
273 	zfsvfs_t *zfsvfs = arg;
274 
275 	if (newval < SPA_MINBLOCKSIZE ||
276 	    newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
277 		newval = SPA_MAXBLOCKSIZE;
278 
279 	zfsvfs->z_max_blksz = newval;
280 	zfsvfs->z_vfs->vfs_bsize = newval;
281 }
282 
283 static void
284 readonly_changed_cb(void *arg, uint64_t newval)
285 {
286 	zfsvfs_t *zfsvfs = arg;
287 
288 	if (newval) {
289 		/* XXX locking on vfs_flag? */
290 		zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
291 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
292 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
293 	} else {
294 		/* XXX locking on vfs_flag? */
295 		zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
296 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
297 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
298 	}
299 }
300 
301 static void
302 devices_changed_cb(void *arg, uint64_t newval)
303 {
304 	zfsvfs_t *zfsvfs = arg;
305 
306 	if (newval == FALSE) {
307 		zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
308 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
309 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
310 	} else {
311 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
312 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
313 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
314 	}
315 }
316 
317 static void
318 setuid_changed_cb(void *arg, uint64_t newval)
319 {
320 	zfsvfs_t *zfsvfs = arg;
321 
322 	if (newval == FALSE) {
323 		zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
324 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
325 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
326 	} else {
327 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
328 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
329 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
330 	}
331 }
332 
333 static void
334 exec_changed_cb(void *arg, uint64_t newval)
335 {
336 	zfsvfs_t *zfsvfs = arg;
337 
338 	if (newval == FALSE) {
339 		zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
340 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
341 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
342 	} else {
343 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
344 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
345 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
346 	}
347 }
348 
349 /*
350  * The nbmand mount option can be changed at mount time.
351  * We can't allow it to be toggled on live file systems or incorrect
352  * behavior may be seen from cifs clients
353  *
354  * This property isn't registered via dsl_prop_register(), but this callback
355  * will be called when a file system is first mounted
356  */
357 static void
358 nbmand_changed_cb(void *arg, uint64_t newval)
359 {
360 	zfsvfs_t *zfsvfs = arg;
361 	if (newval == FALSE) {
362 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
363 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
364 	} else {
365 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
366 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
367 	}
368 }
369 
370 static void
371 snapdir_changed_cb(void *arg, uint64_t newval)
372 {
373 	zfsvfs_t *zfsvfs = arg;
374 
375 	zfsvfs->z_show_ctldir = newval;
376 }
377 
378 static void
379 vscan_changed_cb(void *arg, uint64_t newval)
380 {
381 	zfsvfs_t *zfsvfs = arg;
382 
383 	zfsvfs->z_vscan = newval;
384 }
385 
386 static void
387 acl_mode_changed_cb(void *arg, uint64_t newval)
388 {
389 	zfsvfs_t *zfsvfs = arg;
390 
391 	zfsvfs->z_acl_mode = newval;
392 }
393 
394 static void
395 acl_inherit_changed_cb(void *arg, uint64_t newval)
396 {
397 	zfsvfs_t *zfsvfs = arg;
398 
399 	zfsvfs->z_acl_inherit = newval;
400 }
401 
402 static int
403 zfs_register_callbacks(vfs_t *vfsp)
404 {
405 	struct dsl_dataset *ds = NULL;
406 	objset_t *os = NULL;
407 	zfsvfs_t *zfsvfs = NULL;
408 	uint64_t nbmand;
409 	int readonly, do_readonly = B_FALSE;
410 	int setuid, do_setuid = B_FALSE;
411 	int exec, do_exec = B_FALSE;
412 	int devices, do_devices = B_FALSE;
413 	int xattr, do_xattr = B_FALSE;
414 	int atime, do_atime = B_FALSE;
415 	int error = 0;
416 
417 	ASSERT(vfsp);
418 	zfsvfs = vfsp->vfs_data;
419 	ASSERT(zfsvfs);
420 	os = zfsvfs->z_os;
421 
422 	/*
423 	 * The act of registering our callbacks will destroy any mount
424 	 * options we may have.  In order to enable temporary overrides
425 	 * of mount options, we stash away the current values and
426 	 * restore them after we register the callbacks.
427 	 */
428 	if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
429 	    !spa_writeable(dmu_objset_spa(os))) {
430 		readonly = B_TRUE;
431 		do_readonly = B_TRUE;
432 	} else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
433 		readonly = B_FALSE;
434 		do_readonly = B_TRUE;
435 	}
436 	if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
437 		devices = B_FALSE;
438 		setuid = B_FALSE;
439 		do_devices = B_TRUE;
440 		do_setuid = B_TRUE;
441 	} else {
442 		if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
443 			devices = B_FALSE;
444 			do_devices = B_TRUE;
445 		} else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
446 			devices = B_TRUE;
447 			do_devices = B_TRUE;
448 		}
449 
450 		if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
451 			setuid = B_FALSE;
452 			do_setuid = B_TRUE;
453 		} else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
454 			setuid = B_TRUE;
455 			do_setuid = B_TRUE;
456 		}
457 	}
458 	if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
459 		exec = B_FALSE;
460 		do_exec = B_TRUE;
461 	} else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
462 		exec = B_TRUE;
463 		do_exec = B_TRUE;
464 	}
465 	if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
466 		xattr = B_FALSE;
467 		do_xattr = B_TRUE;
468 	} else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
469 		xattr = B_TRUE;
470 		do_xattr = B_TRUE;
471 	}
472 	if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
473 		atime = B_FALSE;
474 		do_atime = B_TRUE;
475 	} else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
476 		atime = B_TRUE;
477 		do_atime = B_TRUE;
478 	}
479 
480 	/*
481 	 * nbmand is a special property.  It can only be changed at
482 	 * mount time.
483 	 *
484 	 * This is weird, but it is documented to only be changeable
485 	 * at mount time.
486 	 */
487 	if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
488 		nbmand = B_FALSE;
489 	} else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
490 		nbmand = B_TRUE;
491 	} else {
492 		char osname[MAXNAMELEN];
493 
494 		dmu_objset_name(os, osname);
495 		if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
496 		    NULL)) {
497 			return (error);
498 		}
499 	}
500 
501 	/*
502 	 * Register property callbacks.
503 	 *
504 	 * It would probably be fine to just check for i/o error from
505 	 * the first prop_register(), but I guess I like to go
506 	 * overboard...
507 	 */
508 	ds = dmu_objset_ds(os);
509 	error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs);
510 	error = error ? error : dsl_prop_register(ds,
511 	    "xattr", xattr_changed_cb, zfsvfs);
512 	error = error ? error : dsl_prop_register(ds,
513 	    "recordsize", blksz_changed_cb, zfsvfs);
514 	error = error ? error : dsl_prop_register(ds,
515 	    "readonly", readonly_changed_cb, zfsvfs);
516 	error = error ? error : dsl_prop_register(ds,
517 	    "devices", devices_changed_cb, zfsvfs);
518 	error = error ? error : dsl_prop_register(ds,
519 	    "setuid", setuid_changed_cb, zfsvfs);
520 	error = error ? error : dsl_prop_register(ds,
521 	    "exec", exec_changed_cb, zfsvfs);
522 	error = error ? error : dsl_prop_register(ds,
523 	    "snapdir", snapdir_changed_cb, zfsvfs);
524 	error = error ? error : dsl_prop_register(ds,
525 	    "aclmode", acl_mode_changed_cb, zfsvfs);
526 	error = error ? error : dsl_prop_register(ds,
527 	    "aclinherit", acl_inherit_changed_cb, zfsvfs);
528 	error = error ? error : dsl_prop_register(ds,
529 	    "vscan", vscan_changed_cb, zfsvfs);
530 	if (error)
531 		goto unregister;
532 
533 	/*
534 	 * Invoke our callbacks to restore temporary mount options.
535 	 */
536 	if (do_readonly)
537 		readonly_changed_cb(zfsvfs, readonly);
538 	if (do_setuid)
539 		setuid_changed_cb(zfsvfs, setuid);
540 	if (do_exec)
541 		exec_changed_cb(zfsvfs, exec);
542 	if (do_devices)
543 		devices_changed_cb(zfsvfs, devices);
544 	if (do_xattr)
545 		xattr_changed_cb(zfsvfs, xattr);
546 	if (do_atime)
547 		atime_changed_cb(zfsvfs, atime);
548 
549 	nbmand_changed_cb(zfsvfs, nbmand);
550 
551 	return (0);
552 
553 unregister:
554 	/*
555 	 * We may attempt to unregister some callbacks that are not
556 	 * registered, but this is OK; it will simply return ENOMSG,
557 	 * which we will ignore.
558 	 */
559 	(void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs);
560 	(void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs);
561 	(void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs);
562 	(void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs);
563 	(void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zfsvfs);
564 	(void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs);
565 	(void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs);
566 	(void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs);
567 	(void) dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, zfsvfs);
568 	(void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
569 	    zfsvfs);
570 	(void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs);
571 	return (error);
572 
573 }
574 
575 static int
576 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
577     uint64_t *userp, uint64_t *groupp)
578 {
579 	znode_phys_t *znp = data;
580 	int error = 0;
581 
582 	/*
583 	 * Is it a valid type of object to track?
584 	 */
585 	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
586 		return (ENOENT);
587 
588 	/*
589 	 * If we have a NULL data pointer
590 	 * then assume the id's aren't changing and
591 	 * return EEXIST to the dmu to let it know to
592 	 * use the same ids
593 	 */
594 	if (data == NULL)
595 		return (EEXIST);
596 
597 	if (bonustype == DMU_OT_ZNODE) {
598 		*userp = znp->zp_uid;
599 		*groupp = znp->zp_gid;
600 	} else {
601 		int hdrsize;
602 
603 		ASSERT(bonustype == DMU_OT_SA);
604 		hdrsize = sa_hdrsize(data);
605 
606 		if (hdrsize != 0) {
607 			*userp = *((uint64_t *)((uintptr_t)data + hdrsize +
608 			    SA_UID_OFFSET));
609 			*groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
610 			    SA_GID_OFFSET));
611 		} else {
612 			/*
613 			 * This should only happen for newly created
614 			 * files that haven't had the znode data filled
615 			 * in yet.
616 			 */
617 			*userp = 0;
618 			*groupp = 0;
619 		}
620 	}
621 	return (error);
622 }
623 
624 static void
625 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
626     char *domainbuf, int buflen, uid_t *ridp)
627 {
628 	uint64_t fuid;
629 	const char *domain;
630 
631 	fuid = strtonum(fuidstr, NULL);
632 
633 	domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
634 	if (domain)
635 		(void) strlcpy(domainbuf, domain, buflen);
636 	else
637 		domainbuf[0] = '\0';
638 	*ridp = FUID_RID(fuid);
639 }
640 
641 static uint64_t
642 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
643 {
644 	switch (type) {
645 	case ZFS_PROP_USERUSED:
646 		return (DMU_USERUSED_OBJECT);
647 	case ZFS_PROP_GROUPUSED:
648 		return (DMU_GROUPUSED_OBJECT);
649 	case ZFS_PROP_USERQUOTA:
650 		return (zfsvfs->z_userquota_obj);
651 	case ZFS_PROP_GROUPQUOTA:
652 		return (zfsvfs->z_groupquota_obj);
653 	}
654 	return (0);
655 }
656 
657 int
658 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
659     uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
660 {
661 	int error;
662 	zap_cursor_t zc;
663 	zap_attribute_t za;
664 	zfs_useracct_t *buf = vbuf;
665 	uint64_t obj;
666 
667 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
668 		return (ENOTSUP);
669 
670 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
671 	if (obj == 0) {
672 		*bufsizep = 0;
673 		return (0);
674 	}
675 
676 	for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
677 	    (error = zap_cursor_retrieve(&zc, &za)) == 0;
678 	    zap_cursor_advance(&zc)) {
679 		if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
680 		    *bufsizep)
681 			break;
682 
683 		fuidstr_to_sid(zfsvfs, za.za_name,
684 		    buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
685 
686 		buf->zu_space = za.za_first_integer;
687 		buf++;
688 	}
689 	if (error == ENOENT)
690 		error = 0;
691 
692 	ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
693 	*bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
694 	*cookiep = zap_cursor_serialize(&zc);
695 	zap_cursor_fini(&zc);
696 	return (error);
697 }
698 
699 /*
700  * buf must be big enough (eg, 32 bytes)
701  */
702 static int
703 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
704     char *buf, boolean_t addok)
705 {
706 	uint64_t fuid;
707 	int domainid = 0;
708 
709 	if (domain && domain[0]) {
710 		domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
711 		if (domainid == -1)
712 			return (ENOENT);
713 	}
714 	fuid = FUID_ENCODE(domainid, rid);
715 	(void) sprintf(buf, "%llx", (longlong_t)fuid);
716 	return (0);
717 }
718 
719 int
720 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
721     const char *domain, uint64_t rid, uint64_t *valp)
722 {
723 	char buf[32];
724 	int err;
725 	uint64_t obj;
726 
727 	*valp = 0;
728 
729 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
730 		return (ENOTSUP);
731 
732 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
733 	if (obj == 0)
734 		return (0);
735 
736 	err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
737 	if (err)
738 		return (err);
739 
740 	err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
741 	if (err == ENOENT)
742 		err = 0;
743 	return (err);
744 }
745 
746 int
747 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
748     const char *domain, uint64_t rid, uint64_t quota)
749 {
750 	char buf[32];
751 	int err;
752 	dmu_tx_t *tx;
753 	uint64_t *objp;
754 	boolean_t fuid_dirtied;
755 
756 	if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
757 		return (EINVAL);
758 
759 	if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
760 		return (ENOTSUP);
761 
762 	objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
763 	    &zfsvfs->z_groupquota_obj;
764 
765 	err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
766 	if (err)
767 		return (err);
768 	fuid_dirtied = zfsvfs->z_fuid_dirty;
769 
770 	tx = dmu_tx_create(zfsvfs->z_os);
771 	dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
772 	if (*objp == 0) {
773 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
774 		    zfs_userquota_prop_prefixes[type]);
775 	}
776 	if (fuid_dirtied)
777 		zfs_fuid_txhold(zfsvfs, tx);
778 	err = dmu_tx_assign(tx, TXG_WAIT);
779 	if (err) {
780 		dmu_tx_abort(tx);
781 		return (err);
782 	}
783 
784 	mutex_enter(&zfsvfs->z_lock);
785 	if (*objp == 0) {
786 		*objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
787 		    DMU_OT_NONE, 0, tx);
788 		VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
789 		    zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
790 	}
791 	mutex_exit(&zfsvfs->z_lock);
792 
793 	if (quota == 0) {
794 		err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
795 		if (err == ENOENT)
796 			err = 0;
797 	} else {
798 		err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
799 	}
800 	ASSERT(err == 0);
801 	if (fuid_dirtied)
802 		zfs_fuid_sync(zfsvfs, tx);
803 	dmu_tx_commit(tx);
804 	return (err);
805 }
806 
807 boolean_t
808 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
809 {
810 	char buf[32];
811 	uint64_t used, quota, usedobj, quotaobj;
812 	int err;
813 
814 	usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
815 	quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
816 
817 	if (quotaobj == 0 || zfsvfs->z_replay)
818 		return (B_FALSE);
819 
820 	(void) sprintf(buf, "%llx", (longlong_t)fuid);
821 	err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
822 	if (err != 0)
823 		return (B_FALSE);
824 
825 	err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
826 	if (err != 0)
827 		return (B_FALSE);
828 	return (used >= quota);
829 }
830 
831 boolean_t
832 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
833 {
834 	uint64_t fuid;
835 	uint64_t quotaobj;
836 
837 	quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
838 
839 	fuid = isgroup ? zp->z_gid : zp->z_uid;
840 
841 	if (quotaobj == 0 || zfsvfs->z_replay)
842 		return (B_FALSE);
843 
844 	return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
845 }
846 
847 int
848 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
849 {
850 	objset_t *os;
851 	zfsvfs_t *zfsvfs;
852 	uint64_t zval;
853 	int i, error;
854 	uint64_t sa_obj;
855 
856 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
857 
858 	/*
859 	 * We claim to always be readonly so we can open snapshots;
860 	 * other ZPL code will prevent us from writing to snapshots.
861 	 */
862 	error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
863 	if (error) {
864 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
865 		return (error);
866 	}
867 
868 	/*
869 	 * Initialize the zfs-specific filesystem structure.
870 	 * Should probably make this a kmem cache, shuffle fields,
871 	 * and just bzero up to z_hold_mtx[].
872 	 */
873 	zfsvfs->z_vfs = NULL;
874 	zfsvfs->z_parent = zfsvfs;
875 	zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
876 	zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
877 	zfsvfs->z_os = os;
878 
879 	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
880 	if (error) {
881 		goto out;
882 	} else if (zfsvfs->z_version >
883 	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
884 		(void) printf("Can't mount a version %lld file system "
885 		    "on a version %lld pool\n. Pool must be upgraded to mount "
886 		    "this file system.", (u_longlong_t)zfsvfs->z_version,
887 		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
888 		error = ENOTSUP;
889 		goto out;
890 	}
891 	if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
892 		goto out;
893 	zfsvfs->z_norm = (int)zval;
894 
895 	if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
896 		goto out;
897 	zfsvfs->z_utf8 = (zval != 0);
898 
899 	if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
900 		goto out;
901 	zfsvfs->z_case = (uint_t)zval;
902 
903 	/*
904 	 * Fold case on file systems that are always or sometimes case
905 	 * insensitive.
906 	 */
907 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
908 	    zfsvfs->z_case == ZFS_CASE_MIXED)
909 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
910 
911 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
912 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
913 
914 	if (zfsvfs->z_use_sa) {
915 		/* should either have both of these objects or none */
916 		error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
917 		    &sa_obj);
918 		if (error)
919 			return (error);
920 	} else {
921 		/*
922 		 * Pre SA versions file systems should never touch
923 		 * either the attribute registration or layout objects.
924 		 */
925 		sa_obj = 0;
926 	}
927 
928 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
929 	    &zfsvfs->z_attr_table);
930 	if (error)
931 		goto out;
932 
933 	if (zfsvfs->z_version >= ZPL_VERSION_SA)
934 		sa_register_update_callback(os, zfs_sa_upgrade);
935 
936 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
937 	    &zfsvfs->z_root);
938 	if (error)
939 		goto out;
940 	ASSERT(zfsvfs->z_root != 0);
941 
942 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
943 	    &zfsvfs->z_unlinkedobj);
944 	if (error)
945 		goto out;
946 
947 	error = zap_lookup(os, MASTER_NODE_OBJ,
948 	    zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
949 	    8, 1, &zfsvfs->z_userquota_obj);
950 	if (error && error != ENOENT)
951 		goto out;
952 
953 	error = zap_lookup(os, MASTER_NODE_OBJ,
954 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
955 	    8, 1, &zfsvfs->z_groupquota_obj);
956 	if (error && error != ENOENT)
957 		goto out;
958 
959 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
960 	    &zfsvfs->z_fuid_obj);
961 	if (error && error != ENOENT)
962 		goto out;
963 
964 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
965 	    &zfsvfs->z_shares_dir);
966 	if (error && error != ENOENT)
967 		goto out;
968 
969 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
970 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
971 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
972 	    offsetof(znode_t, z_link_node));
973 	rrw_init(&zfsvfs->z_teardown_lock);
974 	rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
975 	rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
976 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
977 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
978 
979 	*zfvp = zfsvfs;
980 	return (0);
981 
982 out:
983 	dmu_objset_disown(os, zfsvfs);
984 	*zfvp = NULL;
985 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
986 	return (error);
987 }
988 
989 static int
990 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
991 {
992 	int error;
993 
994 	error = zfs_register_callbacks(zfsvfs->z_vfs);
995 	if (error)
996 		return (error);
997 
998 	/*
999 	 * Set the objset user_ptr to track its zfsvfs.
1000 	 */
1001 	mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1002 	dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1003 	mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1004 
1005 	zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1006 
1007 	/*
1008 	 * If we are not mounting (ie: online recv), then we don't
1009 	 * have to worry about replaying the log as we blocked all
1010 	 * operations out since we closed the ZIL.
1011 	 */
1012 	if (mounting) {
1013 		boolean_t readonly;
1014 
1015 		/*
1016 		 * During replay we remove the read only flag to
1017 		 * allow replays to succeed.
1018 		 */
1019 		readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1020 		if (readonly != 0)
1021 			zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1022 		else
1023 			zfs_unlinked_drain(zfsvfs);
1024 
1025 		/*
1026 		 * Parse and replay the intent log.
1027 		 *
1028 		 * Because of ziltest, this must be done after
1029 		 * zfs_unlinked_drain().  (Further note: ziltest
1030 		 * doesn't use readonly mounts, where
1031 		 * zfs_unlinked_drain() isn't called.)  This is because
1032 		 * ziltest causes spa_sync() to think it's committed,
1033 		 * but actually it is not, so the intent log contains
1034 		 * many txg's worth of changes.
1035 		 *
1036 		 * In particular, if object N is in the unlinked set in
1037 		 * the last txg to actually sync, then it could be
1038 		 * actually freed in a later txg and then reallocated
1039 		 * in a yet later txg.  This would write a "create
1040 		 * object N" record to the intent log.  Normally, this
1041 		 * would be fine because the spa_sync() would have
1042 		 * written out the fact that object N is free, before
1043 		 * we could write the "create object N" intent log
1044 		 * record.
1045 		 *
1046 		 * But when we are in ziltest mode, we advance the "open
1047 		 * txg" without actually spa_sync()-ing the changes to
1048 		 * disk.  So we would see that object N is still
1049 		 * allocated and in the unlinked set, and there is an
1050 		 * intent log record saying to allocate it.
1051 		 */
1052 		if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1053 			if (zil_replay_disable) {
1054 				zil_destroy(zfsvfs->z_log, B_FALSE);
1055 			} else {
1056 				zfsvfs->z_replay = B_TRUE;
1057 				zil_replay(zfsvfs->z_os, zfsvfs,
1058 				    zfs_replay_vector);
1059 				zfsvfs->z_replay = B_FALSE;
1060 			}
1061 		}
1062 		zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1063 	}
1064 
1065 	return (0);
1066 }
1067 
1068 void
1069 zfsvfs_free(zfsvfs_t *zfsvfs)
1070 {
1071 	int i;
1072 	extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1073 
1074 	/*
1075 	 * This is a barrier to prevent the filesystem from going away in
1076 	 * zfs_znode_move() until we can safely ensure that the filesystem is
1077 	 * not unmounted. We consider the filesystem valid before the barrier
1078 	 * and invalid after the barrier.
1079 	 */
1080 	rw_enter(&zfsvfs_lock, RW_READER);
1081 	rw_exit(&zfsvfs_lock);
1082 
1083 	zfs_fuid_destroy(zfsvfs);
1084 
1085 	mutex_destroy(&zfsvfs->z_znodes_lock);
1086 	mutex_destroy(&zfsvfs->z_lock);
1087 	list_destroy(&zfsvfs->z_all_znodes);
1088 	rrw_destroy(&zfsvfs->z_teardown_lock);
1089 	rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1090 	rw_destroy(&zfsvfs->z_fuid_lock);
1091 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1092 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1093 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1094 }
1095 
1096 static void
1097 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1098 {
1099 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1100 	if (zfsvfs->z_vfs) {
1101 		if (zfsvfs->z_use_fuids) {
1102 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1103 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1104 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1105 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1106 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1107 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1108 		} else {
1109 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1110 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1111 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1112 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1113 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1114 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1115 		}
1116 	}
1117 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1118 }
1119 
1120 static int
1121 zfs_domount(vfs_t *vfsp, char *osname)
1122 {
1123 	dev_t mount_dev;
1124 	uint64_t recordsize, fsid_guid;
1125 	int error = 0;
1126 	zfsvfs_t *zfsvfs;
1127 
1128 	ASSERT(vfsp);
1129 	ASSERT(osname);
1130 
1131 	error = zfsvfs_create(osname, &zfsvfs);
1132 	if (error)
1133 		return (error);
1134 	zfsvfs->z_vfs = vfsp;
1135 
1136 	/* Initialize the generic filesystem structure. */
1137 	vfsp->vfs_bcount = 0;
1138 	vfsp->vfs_data = NULL;
1139 
1140 	if (zfs_create_unique_device(&mount_dev) == -1) {
1141 		error = ENODEV;
1142 		goto out;
1143 	}
1144 	ASSERT(vfs_devismounted(mount_dev) == 0);
1145 
1146 	if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1147 	    NULL))
1148 		goto out;
1149 
1150 	vfsp->vfs_dev = mount_dev;
1151 	vfsp->vfs_fstype = zfsfstype;
1152 	vfsp->vfs_bsize = recordsize;
1153 	vfsp->vfs_flag |= VFS_NOTRUNC;
1154 	vfsp->vfs_data = zfsvfs;
1155 
1156 	/*
1157 	 * The fsid is 64 bits, composed of an 8-bit fs type, which
1158 	 * separates our fsid from any other filesystem types, and a
1159 	 * 56-bit objset unique ID.  The objset unique ID is unique to
1160 	 * all objsets open on this system, provided by unique_create().
1161 	 * The 8-bit fs type must be put in the low bits of fsid[1]
1162 	 * because that's where other Solaris filesystems put it.
1163 	 */
1164 	fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1165 	ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1166 	vfsp->vfs_fsid.val[0] = fsid_guid;
1167 	vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1168 	    zfsfstype & 0xFF;
1169 
1170 	/*
1171 	 * Set features for file system.
1172 	 */
1173 	zfs_set_fuid_feature(zfsvfs);
1174 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1175 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1176 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1177 		vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1178 	} else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1179 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1180 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1181 	}
1182 	vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1183 
1184 	if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1185 		uint64_t pval;
1186 
1187 		atime_changed_cb(zfsvfs, B_FALSE);
1188 		readonly_changed_cb(zfsvfs, B_TRUE);
1189 		if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1190 			goto out;
1191 		xattr_changed_cb(zfsvfs, pval);
1192 		zfsvfs->z_issnap = B_TRUE;
1193 		zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1194 
1195 		mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1196 		dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1197 		mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1198 	} else {
1199 		error = zfsvfs_setup(zfsvfs, B_TRUE);
1200 	}
1201 
1202 	if (!zfsvfs->z_issnap)
1203 		zfsctl_create(zfsvfs);
1204 out:
1205 	if (error) {
1206 		dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1207 		zfsvfs_free(zfsvfs);
1208 	} else {
1209 		atomic_add_32(&zfs_active_fs_count, 1);
1210 	}
1211 
1212 	return (error);
1213 }
1214 
1215 void
1216 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1217 {
1218 	objset_t *os = zfsvfs->z_os;
1219 	struct dsl_dataset *ds;
1220 
1221 	/*
1222 	 * Unregister properties.
1223 	 */
1224 	if (!dmu_objset_is_snapshot(os)) {
1225 		ds = dmu_objset_ds(os);
1226 		VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1227 		    zfsvfs) == 0);
1228 
1229 		VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1230 		    zfsvfs) == 0);
1231 
1232 		VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1233 		    zfsvfs) == 0);
1234 
1235 		VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1236 		    zfsvfs) == 0);
1237 
1238 		VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1239 		    zfsvfs) == 0);
1240 
1241 		VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1242 		    zfsvfs) == 0);
1243 
1244 		VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1245 		    zfsvfs) == 0);
1246 
1247 		VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1248 		    zfsvfs) == 0);
1249 
1250 		VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
1251 		    zfsvfs) == 0);
1252 
1253 		VERIFY(dsl_prop_unregister(ds, "aclinherit",
1254 		    acl_inherit_changed_cb, zfsvfs) == 0);
1255 
1256 		VERIFY(dsl_prop_unregister(ds, "vscan",
1257 		    vscan_changed_cb, zfsvfs) == 0);
1258 	}
1259 }
1260 
1261 /*
1262  * Convert a decimal digit string to a uint64_t integer.
1263  */
1264 static int
1265 str_to_uint64(char *str, uint64_t *objnum)
1266 {
1267 	uint64_t num = 0;
1268 
1269 	while (*str) {
1270 		if (*str < '0' || *str > '9')
1271 			return (EINVAL);
1272 
1273 		num = num*10 + *str++ - '0';
1274 	}
1275 
1276 	*objnum = num;
1277 	return (0);
1278 }
1279 
1280 /*
1281  * The boot path passed from the boot loader is in the form of
1282  * "rootpool-name/root-filesystem-object-number'. Convert this
1283  * string to a dataset name: "rootpool-name/root-filesystem-name".
1284  */
1285 static int
1286 zfs_parse_bootfs(char *bpath, char *outpath)
1287 {
1288 	char *slashp;
1289 	uint64_t objnum;
1290 	int error;
1291 
1292 	if (*bpath == 0 || *bpath == '/')
1293 		return (EINVAL);
1294 
1295 	(void) strcpy(outpath, bpath);
1296 
1297 	slashp = strchr(bpath, '/');
1298 
1299 	/* if no '/', just return the pool name */
1300 	if (slashp == NULL) {
1301 		return (0);
1302 	}
1303 
1304 	/* if not a number, just return the root dataset name */
1305 	if (str_to_uint64(slashp+1, &objnum)) {
1306 		return (0);
1307 	}
1308 
1309 	*slashp = '\0';
1310 	error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1311 	*slashp = '/';
1312 
1313 	return (error);
1314 }
1315 
1316 /*
1317  * zfs_check_global_label:
1318  *	Check that the hex label string is appropriate for the dataset
1319  *	being mounted into the global_zone proper.
1320  *
1321  *	Return an error if the hex label string is not default or
1322  *	admin_low/admin_high.  For admin_low labels, the corresponding
1323  *	dataset must be readonly.
1324  */
1325 int
1326 zfs_check_global_label(const char *dsname, const char *hexsl)
1327 {
1328 	if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1329 		return (0);
1330 	if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1331 		return (0);
1332 	if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1333 		/* must be readonly */
1334 		uint64_t rdonly;
1335 
1336 		if (dsl_prop_get_integer(dsname,
1337 		    zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1338 			return (EACCES);
1339 		return (rdonly ? 0 : EACCES);
1340 	}
1341 	return (EACCES);
1342 }
1343 
1344 /*
1345  * zfs_mount_label_policy:
1346  *	Determine whether the mount is allowed according to MAC check.
1347  *	by comparing (where appropriate) label of the dataset against
1348  *	the label of the zone being mounted into.  If the dataset has
1349  *	no label, create one.
1350  *
1351  *	Returns:
1352  *		 0 :	access allowed
1353  *		>0 :	error code, such as EACCES
1354  */
1355 static int
1356 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1357 {
1358 	int		error, retv;
1359 	zone_t		*mntzone = NULL;
1360 	ts_label_t	*mnt_tsl;
1361 	bslabel_t	*mnt_sl;
1362 	bslabel_t	ds_sl;
1363 	char		ds_hexsl[MAXNAMELEN];
1364 
1365 	retv = EACCES;				/* assume the worst */
1366 
1367 	/*
1368 	 * Start by getting the dataset label if it exists.
1369 	 */
1370 	error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1371 	    1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1372 	if (error)
1373 		return (EACCES);
1374 
1375 	/*
1376 	 * If labeling is NOT enabled, then disallow the mount of datasets
1377 	 * which have a non-default label already.  No other label checks
1378 	 * are needed.
1379 	 */
1380 	if (!is_system_labeled()) {
1381 		if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1382 			return (0);
1383 		return (EACCES);
1384 	}
1385 
1386 	/*
1387 	 * Get the label of the mountpoint.  If mounting into the global
1388 	 * zone (i.e. mountpoint is not within an active zone and the
1389 	 * zoned property is off), the label must be default or
1390 	 * admin_low/admin_high only; no other checks are needed.
1391 	 */
1392 	mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1393 	if (mntzone->zone_id == GLOBAL_ZONEID) {
1394 		uint64_t zoned;
1395 
1396 		zone_rele(mntzone);
1397 
1398 		if (dsl_prop_get_integer(osname,
1399 		    zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1400 			return (EACCES);
1401 		if (!zoned)
1402 			return (zfs_check_global_label(osname, ds_hexsl));
1403 		else
1404 			/*
1405 			 * This is the case of a zone dataset being mounted
1406 			 * initially, before the zone has been fully created;
1407 			 * allow this mount into global zone.
1408 			 */
1409 			return (0);
1410 	}
1411 
1412 	mnt_tsl = mntzone->zone_slabel;
1413 	ASSERT(mnt_tsl != NULL);
1414 	label_hold(mnt_tsl);
1415 	mnt_sl = label2bslabel(mnt_tsl);
1416 
1417 	if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1418 		/*
1419 		 * The dataset doesn't have a real label, so fabricate one.
1420 		 */
1421 		char *str = NULL;
1422 
1423 		if (l_to_str_internal(mnt_sl, &str) == 0 &&
1424 		    dsl_prop_set(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1425 		    ZPROP_SRC_LOCAL, 1, strlen(str) + 1, str) == 0)
1426 			retv = 0;
1427 		if (str != NULL)
1428 			kmem_free(str, strlen(str) + 1);
1429 	} else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1430 		/*
1431 		 * Now compare labels to complete the MAC check.  If the
1432 		 * labels are equal then allow access.  If the mountpoint
1433 		 * label dominates the dataset label, allow readonly access.
1434 		 * Otherwise, access is denied.
1435 		 */
1436 		if (blequal(mnt_sl, &ds_sl))
1437 			retv = 0;
1438 		else if (bldominates(mnt_sl, &ds_sl)) {
1439 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1440 			retv = 0;
1441 		}
1442 	}
1443 
1444 	label_rele(mnt_tsl);
1445 	zone_rele(mntzone);
1446 	return (retv);
1447 }
1448 
1449 static int
1450 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1451 {
1452 	int error = 0;
1453 	static int zfsrootdone = 0;
1454 	zfsvfs_t *zfsvfs = NULL;
1455 	znode_t *zp = NULL;
1456 	vnode_t *vp = NULL;
1457 	char *zfs_bootfs;
1458 	char *zfs_devid;
1459 
1460 	ASSERT(vfsp);
1461 
1462 	/*
1463 	 * The filesystem that we mount as root is defined in the
1464 	 * boot property "zfs-bootfs" with a format of
1465 	 * "poolname/root-dataset-objnum".
1466 	 */
1467 	if (why == ROOT_INIT) {
1468 		if (zfsrootdone++)
1469 			return (EBUSY);
1470 		/*
1471 		 * the process of doing a spa_load will require the
1472 		 * clock to be set before we could (for example) do
1473 		 * something better by looking at the timestamp on
1474 		 * an uberblock, so just set it to -1.
1475 		 */
1476 		clkset(-1);
1477 
1478 		if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1479 			cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1480 			    "bootfs name");
1481 			return (EINVAL);
1482 		}
1483 		zfs_devid = spa_get_bootprop("diskdevid");
1484 		error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1485 		if (zfs_devid)
1486 			spa_free_bootprop(zfs_devid);
1487 		if (error) {
1488 			spa_free_bootprop(zfs_bootfs);
1489 			cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1490 			    error);
1491 			return (error);
1492 		}
1493 		if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1494 			spa_free_bootprop(zfs_bootfs);
1495 			cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1496 			    error);
1497 			return (error);
1498 		}
1499 
1500 		spa_free_bootprop(zfs_bootfs);
1501 
1502 		if (error = vfs_lock(vfsp))
1503 			return (error);
1504 
1505 		if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1506 			cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1507 			goto out;
1508 		}
1509 
1510 		zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1511 		ASSERT(zfsvfs);
1512 		if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1513 			cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1514 			goto out;
1515 		}
1516 
1517 		vp = ZTOV(zp);
1518 		mutex_enter(&vp->v_lock);
1519 		vp->v_flag |= VROOT;
1520 		mutex_exit(&vp->v_lock);
1521 		rootvp = vp;
1522 
1523 		/*
1524 		 * Leave rootvp held.  The root file system is never unmounted.
1525 		 */
1526 
1527 		vfs_add((struct vnode *)0, vfsp,
1528 		    (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1529 out:
1530 		vfs_unlock(vfsp);
1531 		return (error);
1532 	} else if (why == ROOT_REMOUNT) {
1533 		readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1534 		vfsp->vfs_flag |= VFS_REMOUNT;
1535 
1536 		/* refresh mount options */
1537 		zfs_unregister_callbacks(vfsp->vfs_data);
1538 		return (zfs_register_callbacks(vfsp));
1539 
1540 	} else if (why == ROOT_UNMOUNT) {
1541 		zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1542 		(void) zfs_sync(vfsp, 0, 0);
1543 		return (0);
1544 	}
1545 
1546 	/*
1547 	 * if "why" is equal to anything else other than ROOT_INIT,
1548 	 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1549 	 */
1550 	return (ENOTSUP);
1551 }
1552 
1553 /*ARGSUSED*/
1554 static int
1555 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1556 {
1557 	char		*osname;
1558 	pathname_t	spn;
1559 	int		error = 0;
1560 	uio_seg_t	fromspace = (uap->flags & MS_SYSSPACE) ?
1561 	    UIO_SYSSPACE : UIO_USERSPACE;
1562 	int		canwrite;
1563 
1564 	if (mvp->v_type != VDIR)
1565 		return (ENOTDIR);
1566 
1567 	mutex_enter(&mvp->v_lock);
1568 	if ((uap->flags & MS_REMOUNT) == 0 &&
1569 	    (uap->flags & MS_OVERLAY) == 0 &&
1570 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1571 		mutex_exit(&mvp->v_lock);
1572 		return (EBUSY);
1573 	}
1574 	mutex_exit(&mvp->v_lock);
1575 
1576 	/*
1577 	 * ZFS does not support passing unparsed data in via MS_DATA.
1578 	 * Users should use the MS_OPTIONSTR interface; this means
1579 	 * that all option parsing is already done and the options struct
1580 	 * can be interrogated.
1581 	 */
1582 	if ((uap->flags & MS_DATA) && uap->datalen > 0)
1583 		return (EINVAL);
1584 
1585 	/*
1586 	 * Get the objset name (the "special" mount argument).
1587 	 */
1588 	if (error = pn_get(uap->spec, fromspace, &spn))
1589 		return (error);
1590 
1591 	osname = spn.pn_path;
1592 
1593 	/*
1594 	 * Check for mount privilege?
1595 	 *
1596 	 * If we don't have privilege then see if
1597 	 * we have local permission to allow it
1598 	 */
1599 	error = secpolicy_fs_mount(cr, mvp, vfsp);
1600 	if (error) {
1601 		if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1602 			vattr_t		vattr;
1603 
1604 			/*
1605 			 * Make sure user is the owner of the mount point
1606 			 * or has sufficient privileges.
1607 			 */
1608 
1609 			vattr.va_mask = AT_UID;
1610 
1611 			if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1612 				goto out;
1613 			}
1614 
1615 			if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1616 			    VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1617 				goto out;
1618 			}
1619 			secpolicy_fs_mount_clearopts(cr, vfsp);
1620 		} else {
1621 			goto out;
1622 		}
1623 	}
1624 
1625 	/*
1626 	 * Refuse to mount a filesystem if we are in a local zone and the
1627 	 * dataset is not visible.
1628 	 */
1629 	if (!INGLOBALZONE(curproc) &&
1630 	    (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1631 		error = EPERM;
1632 		goto out;
1633 	}
1634 
1635 	error = zfs_mount_label_policy(vfsp, osname);
1636 	if (error)
1637 		goto out;
1638 
1639 	/*
1640 	 * When doing a remount, we simply refresh our temporary properties
1641 	 * according to those options set in the current VFS options.
1642 	 */
1643 	if (uap->flags & MS_REMOUNT) {
1644 		/* refresh mount options */
1645 		zfs_unregister_callbacks(vfsp->vfs_data);
1646 		error = zfs_register_callbacks(vfsp);
1647 		goto out;
1648 	}
1649 
1650 	error = zfs_domount(vfsp, osname);
1651 
1652 	/*
1653 	 * Add an extra VFS_HOLD on our parent vfs so that it can't
1654 	 * disappear due to a forced unmount.
1655 	 */
1656 	if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1657 		VFS_HOLD(mvp->v_vfsp);
1658 
1659 out:
1660 	pn_free(&spn);
1661 	return (error);
1662 }
1663 
1664 static int
1665 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1666 {
1667 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1668 	dev32_t d32;
1669 	uint64_t refdbytes, availbytes, usedobjs, availobjs;
1670 
1671 	ZFS_ENTER(zfsvfs);
1672 
1673 	dmu_objset_space(zfsvfs->z_os,
1674 	    &refdbytes, &availbytes, &usedobjs, &availobjs);
1675 
1676 	/*
1677 	 * The underlying storage pool actually uses multiple block sizes.
1678 	 * We report the fragsize as the smallest block size we support,
1679 	 * and we report our blocksize as the filesystem's maximum blocksize.
1680 	 */
1681 	statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1682 	statp->f_bsize = zfsvfs->z_max_blksz;
1683 
1684 	/*
1685 	 * The following report "total" blocks of various kinds in the
1686 	 * file system, but reported in terms of f_frsize - the
1687 	 * "fragment" size.
1688 	 */
1689 
1690 	statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1691 	statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1692 	statp->f_bavail = statp->f_bfree; /* no root reservation */
1693 
1694 	/*
1695 	 * statvfs() should really be called statufs(), because it assumes
1696 	 * static metadata.  ZFS doesn't preallocate files, so the best
1697 	 * we can do is report the max that could possibly fit in f_files,
1698 	 * and that minus the number actually used in f_ffree.
1699 	 * For f_ffree, report the smaller of the number of object available
1700 	 * and the number of blocks (each object will take at least a block).
1701 	 */
1702 	statp->f_ffree = MIN(availobjs, statp->f_bfree);
1703 	statp->f_favail = statp->f_ffree;	/* no "root reservation" */
1704 	statp->f_files = statp->f_ffree + usedobjs;
1705 
1706 	(void) cmpldev(&d32, vfsp->vfs_dev);
1707 	statp->f_fsid = d32;
1708 
1709 	/*
1710 	 * We're a zfs filesystem.
1711 	 */
1712 	(void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1713 
1714 	statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1715 
1716 	statp->f_namemax = ZFS_MAXNAMELEN;
1717 
1718 	/*
1719 	 * We have all of 32 characters to stuff a string here.
1720 	 * Is there anything useful we could/should provide?
1721 	 */
1722 	bzero(statp->f_fstr, sizeof (statp->f_fstr));
1723 
1724 	ZFS_EXIT(zfsvfs);
1725 	return (0);
1726 }
1727 
1728 static int
1729 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1730 {
1731 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1732 	znode_t *rootzp;
1733 	int error;
1734 
1735 	ZFS_ENTER(zfsvfs);
1736 
1737 	error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1738 	if (error == 0)
1739 		*vpp = ZTOV(rootzp);
1740 
1741 	ZFS_EXIT(zfsvfs);
1742 	return (error);
1743 }
1744 
1745 /*
1746  * Teardown the zfsvfs::z_os.
1747  *
1748  * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1749  * and 'z_teardown_inactive_lock' held.
1750  */
1751 static int
1752 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1753 {
1754 	znode_t	*zp;
1755 
1756 	rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1757 
1758 	if (!unmounting) {
1759 		/*
1760 		 * We purge the parent filesystem's vfsp as the parent
1761 		 * filesystem and all of its snapshots have their vnode's
1762 		 * v_vfsp set to the parent's filesystem's vfsp.  Note,
1763 		 * 'z_parent' is self referential for non-snapshots.
1764 		 */
1765 		(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1766 	}
1767 
1768 	/*
1769 	 * Close the zil. NB: Can't close the zil while zfs_inactive
1770 	 * threads are blocked as zil_close can call zfs_inactive.
1771 	 */
1772 	if (zfsvfs->z_log) {
1773 		zil_close(zfsvfs->z_log);
1774 		zfsvfs->z_log = NULL;
1775 	}
1776 
1777 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1778 
1779 	/*
1780 	 * If we are not unmounting (ie: online recv) and someone already
1781 	 * unmounted this file system while we were doing the switcheroo,
1782 	 * or a reopen of z_os failed then just bail out now.
1783 	 */
1784 	if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1785 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
1786 		rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1787 		return (EIO);
1788 	}
1789 
1790 	/*
1791 	 * At this point there are no vops active, and any new vops will
1792 	 * fail with EIO since we have z_teardown_lock for writer (only
1793 	 * relavent for forced unmount).
1794 	 *
1795 	 * Release all holds on dbufs.
1796 	 */
1797 	mutex_enter(&zfsvfs->z_znodes_lock);
1798 	for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1799 	    zp = list_next(&zfsvfs->z_all_znodes, zp))
1800 		if (zp->z_sa_hdl) {
1801 			ASSERT(ZTOV(zp)->v_count > 0);
1802 			zfs_znode_dmu_fini(zp);
1803 		}
1804 	mutex_exit(&zfsvfs->z_znodes_lock);
1805 
1806 	/*
1807 	 * If we are unmounting, set the unmounted flag and let new vops
1808 	 * unblock.  zfs_inactive will have the unmounted behavior, and all
1809 	 * other vops will fail with EIO.
1810 	 */
1811 	if (unmounting) {
1812 		zfsvfs->z_unmounted = B_TRUE;
1813 		rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1814 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
1815 	}
1816 
1817 	/*
1818 	 * z_os will be NULL if there was an error in attempting to reopen
1819 	 * zfsvfs, so just return as the properties had already been
1820 	 * unregistered and cached data had been evicted before.
1821 	 */
1822 	if (zfsvfs->z_os == NULL)
1823 		return (0);
1824 
1825 	/*
1826 	 * Unregister properties.
1827 	 */
1828 	zfs_unregister_callbacks(zfsvfs);
1829 
1830 	/*
1831 	 * Evict cached data
1832 	 */
1833 	if (dmu_objset_is_dirty_anywhere(zfsvfs->z_os))
1834 		if (!(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1835 			txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1836 	(void) dmu_objset_evict_dbufs(zfsvfs->z_os);
1837 
1838 	return (0);
1839 }
1840 
1841 /*ARGSUSED*/
1842 static int
1843 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1844 {
1845 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1846 	objset_t *os;
1847 	int ret;
1848 
1849 	ret = secpolicy_fs_unmount(cr, vfsp);
1850 	if (ret) {
1851 		if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1852 		    ZFS_DELEG_PERM_MOUNT, cr))
1853 			return (ret);
1854 	}
1855 
1856 	/*
1857 	 * We purge the parent filesystem's vfsp as the parent filesystem
1858 	 * and all of its snapshots have their vnode's v_vfsp set to the
1859 	 * parent's filesystem's vfsp.  Note, 'z_parent' is self
1860 	 * referential for non-snapshots.
1861 	 */
1862 	(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1863 
1864 	/*
1865 	 * Unmount any snapshots mounted under .zfs before unmounting the
1866 	 * dataset itself.
1867 	 */
1868 	if (zfsvfs->z_ctldir != NULL &&
1869 	    (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1870 		return (ret);
1871 	}
1872 
1873 	if (!(fflag & MS_FORCE)) {
1874 		/*
1875 		 * Check the number of active vnodes in the file system.
1876 		 * Our count is maintained in the vfs structure, but the
1877 		 * number is off by 1 to indicate a hold on the vfs
1878 		 * structure itself.
1879 		 *
1880 		 * The '.zfs' directory maintains a reference of its
1881 		 * own, and any active references underneath are
1882 		 * reflected in the vnode count.
1883 		 */
1884 		if (zfsvfs->z_ctldir == NULL) {
1885 			if (vfsp->vfs_count > 1)
1886 				return (EBUSY);
1887 		} else {
1888 			if (vfsp->vfs_count > 2 ||
1889 			    zfsvfs->z_ctldir->v_count > 1)
1890 				return (EBUSY);
1891 		}
1892 	}
1893 
1894 	vfsp->vfs_flag |= VFS_UNMOUNTED;
1895 
1896 	VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1897 	os = zfsvfs->z_os;
1898 
1899 	/*
1900 	 * z_os will be NULL if there was an error in
1901 	 * attempting to reopen zfsvfs.
1902 	 */
1903 	if (os != NULL) {
1904 		/*
1905 		 * Unset the objset user_ptr.
1906 		 */
1907 		mutex_enter(&os->os_user_ptr_lock);
1908 		dmu_objset_set_user(os, NULL);
1909 		mutex_exit(&os->os_user_ptr_lock);
1910 
1911 		/*
1912 		 * Finally release the objset
1913 		 */
1914 		dmu_objset_disown(os, zfsvfs);
1915 	}
1916 
1917 	/*
1918 	 * We can now safely destroy the '.zfs' directory node.
1919 	 */
1920 	if (zfsvfs->z_ctldir != NULL)
1921 		zfsctl_destroy(zfsvfs);
1922 
1923 	return (0);
1924 }
1925 
1926 static int
1927 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1928 {
1929 	zfsvfs_t	*zfsvfs = vfsp->vfs_data;
1930 	znode_t		*zp;
1931 	uint64_t	object = 0;
1932 	uint64_t	fid_gen = 0;
1933 	uint64_t	gen_mask;
1934 	uint64_t	zp_gen;
1935 	int 		i, err;
1936 
1937 	*vpp = NULL;
1938 
1939 	ZFS_ENTER(zfsvfs);
1940 
1941 	if (fidp->fid_len == LONG_FID_LEN) {
1942 		zfid_long_t	*zlfid = (zfid_long_t *)fidp;
1943 		uint64_t	objsetid = 0;
1944 		uint64_t	setgen = 0;
1945 
1946 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1947 			objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1948 
1949 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1950 			setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1951 
1952 		ZFS_EXIT(zfsvfs);
1953 
1954 		err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1955 		if (err)
1956 			return (EINVAL);
1957 		ZFS_ENTER(zfsvfs);
1958 	}
1959 
1960 	if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1961 		zfid_short_t	*zfid = (zfid_short_t *)fidp;
1962 
1963 		for (i = 0; i < sizeof (zfid->zf_object); i++)
1964 			object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1965 
1966 		for (i = 0; i < sizeof (zfid->zf_gen); i++)
1967 			fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1968 	} else {
1969 		ZFS_EXIT(zfsvfs);
1970 		return (EINVAL);
1971 	}
1972 
1973 	/* A zero fid_gen means we are in the .zfs control directories */
1974 	if (fid_gen == 0 &&
1975 	    (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1976 		*vpp = zfsvfs->z_ctldir;
1977 		ASSERT(*vpp != NULL);
1978 		if (object == ZFSCTL_INO_SNAPDIR) {
1979 			VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
1980 			    0, NULL, NULL, NULL, NULL, NULL) == 0);
1981 		} else {
1982 			VN_HOLD(*vpp);
1983 		}
1984 		ZFS_EXIT(zfsvfs);
1985 		return (0);
1986 	}
1987 
1988 	gen_mask = -1ULL >> (64 - 8 * i);
1989 
1990 	dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1991 	if (err = zfs_zget(zfsvfs, object, &zp)) {
1992 		ZFS_EXIT(zfsvfs);
1993 		return (err);
1994 	}
1995 	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1996 	    sizeof (uint64_t));
1997 	zp_gen = zp_gen & gen_mask;
1998 	if (zp_gen == 0)
1999 		zp_gen = 1;
2000 	if (zp->z_unlinked || zp_gen != fid_gen) {
2001 		dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2002 		VN_RELE(ZTOV(zp));
2003 		ZFS_EXIT(zfsvfs);
2004 		return (EINVAL);
2005 	}
2006 
2007 	*vpp = ZTOV(zp);
2008 	ZFS_EXIT(zfsvfs);
2009 	return (0);
2010 }
2011 
2012 /*
2013  * Block out VOPs and close zfsvfs_t::z_os
2014  *
2015  * Note, if successful, then we return with the 'z_teardown_lock' and
2016  * 'z_teardown_inactive_lock' write held.
2017  */
2018 int
2019 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2020 {
2021 	int error;
2022 
2023 	if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2024 		return (error);
2025 	dmu_objset_disown(zfsvfs->z_os, zfsvfs);
2026 
2027 	return (0);
2028 }
2029 
2030 /*
2031  * Reopen zfsvfs_t::z_os and release VOPs.
2032  */
2033 int
2034 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2035 {
2036 	int err;
2037 
2038 	ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
2039 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2040 
2041 	err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zfsvfs,
2042 	    &zfsvfs->z_os);
2043 	if (err) {
2044 		zfsvfs->z_os = NULL;
2045 	} else {
2046 		znode_t *zp;
2047 		uint64_t sa_obj = 0;
2048 
2049 		/*
2050 		 * Make sure version hasn't changed
2051 		 */
2052 
2053 		err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
2054 		    &zfsvfs->z_version);
2055 
2056 		if (err)
2057 			goto bail;
2058 
2059 		err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2060 		    ZFS_SA_ATTRS, 8, 1, &sa_obj);
2061 
2062 		if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
2063 			goto bail;
2064 
2065 		if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2066 		    zfs_attr_table,  ZPL_END, &zfsvfs->z_attr_table)) != 0)
2067 			goto bail;
2068 
2069 		if (zfsvfs->z_version >= ZPL_VERSION_SA)
2070 			sa_register_update_callback(zfsvfs->z_os,
2071 			    zfs_sa_upgrade);
2072 
2073 		VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2074 
2075 		zfs_set_fuid_feature(zfsvfs);
2076 
2077 		/*
2078 		 * Attempt to re-establish all the active znodes with
2079 		 * their dbufs.  If a zfs_rezget() fails, then we'll let
2080 		 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2081 		 * when they try to use their znode.
2082 		 */
2083 		mutex_enter(&zfsvfs->z_znodes_lock);
2084 		for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2085 		    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2086 			(void) zfs_rezget(zp);
2087 		}
2088 		mutex_exit(&zfsvfs->z_znodes_lock);
2089 	}
2090 
2091 bail:
2092 	/* release the VOPs */
2093 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
2094 	rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
2095 
2096 	if (err) {
2097 		/*
2098 		 * Since we couldn't reopen zfsvfs::z_os, or
2099 		 * setup the sa framework force unmount this file system.
2100 		 */
2101 		if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2102 			(void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2103 	}
2104 	return (err);
2105 }
2106 
2107 static void
2108 zfs_freevfs(vfs_t *vfsp)
2109 {
2110 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2111 
2112 	/*
2113 	 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2114 	 * from zfs_mount().  Release it here.  If we came through
2115 	 * zfs_mountroot() instead, we didn't grab an extra hold, so
2116 	 * skip the VFS_RELE for rootvfs.
2117 	 */
2118 	if (zfsvfs->z_issnap && (vfsp != rootvfs))
2119 		VFS_RELE(zfsvfs->z_parent->z_vfs);
2120 
2121 	zfsvfs_free(zfsvfs);
2122 
2123 	atomic_add_32(&zfs_active_fs_count, -1);
2124 }
2125 
2126 /*
2127  * VFS_INIT() initialization.  Note that there is no VFS_FINI(),
2128  * so we can't safely do any non-idempotent initialization here.
2129  * Leave that to zfs_init() and zfs_fini(), which are called
2130  * from the module's _init() and _fini() entry points.
2131  */
2132 /*ARGSUSED*/
2133 static int
2134 zfs_vfsinit(int fstype, char *name)
2135 {
2136 	int error;
2137 
2138 	zfsfstype = fstype;
2139 
2140 	/*
2141 	 * Setup vfsops and vnodeops tables.
2142 	 */
2143 	error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2144 	if (error != 0) {
2145 		cmn_err(CE_WARN, "zfs: bad vfs ops template");
2146 	}
2147 
2148 	error = zfs_create_op_tables();
2149 	if (error) {
2150 		zfs_remove_op_tables();
2151 		cmn_err(CE_WARN, "zfs: bad vnode ops template");
2152 		(void) vfs_freevfsops_by_type(zfsfstype);
2153 		return (error);
2154 	}
2155 
2156 	mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2157 
2158 	/*
2159 	 * Unique major number for all zfs mounts.
2160 	 * If we run out of 32-bit minors, we'll getudev() another major.
2161 	 */
2162 	zfs_major = ddi_name_to_major(ZFS_DRIVER);
2163 	zfs_minor = ZFS_MIN_MINOR;
2164 
2165 	return (0);
2166 }
2167 
2168 void
2169 zfs_init(void)
2170 {
2171 	/*
2172 	 * Initialize .zfs directory structures
2173 	 */
2174 	zfsctl_init();
2175 
2176 	/*
2177 	 * Initialize znode cache, vnode ops, etc...
2178 	 */
2179 	zfs_znode_init();
2180 
2181 	dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2182 }
2183 
2184 void
2185 zfs_fini(void)
2186 {
2187 	zfsctl_fini();
2188 	zfs_znode_fini();
2189 }
2190 
2191 int
2192 zfs_busy(void)
2193 {
2194 	return (zfs_active_fs_count != 0);
2195 }
2196 
2197 int
2198 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2199 {
2200 	int error;
2201 	objset_t *os = zfsvfs->z_os;
2202 	dmu_tx_t *tx;
2203 
2204 	if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2205 		return (EINVAL);
2206 
2207 	if (newvers < zfsvfs->z_version)
2208 		return (EINVAL);
2209 
2210 	if (zfs_spa_version_map(newvers) >
2211 	    spa_version(dmu_objset_spa(zfsvfs->z_os)))
2212 		return (ENOTSUP);
2213 
2214 	tx = dmu_tx_create(os);
2215 	dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2216 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2217 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2218 		    ZFS_SA_ATTRS);
2219 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2220 	}
2221 	error = dmu_tx_assign(tx, TXG_WAIT);
2222 	if (error) {
2223 		dmu_tx_abort(tx);
2224 		return (error);
2225 	}
2226 
2227 	error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2228 	    8, 1, &newvers, tx);
2229 
2230 	if (error) {
2231 		dmu_tx_commit(tx);
2232 		return (error);
2233 	}
2234 
2235 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2236 		uint64_t sa_obj;
2237 
2238 		ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2239 		    SPA_VERSION_SA);
2240 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2241 		    DMU_OT_NONE, 0, tx);
2242 
2243 		error = zap_add(os, MASTER_NODE_OBJ,
2244 		    ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2245 		ASSERT3U(error, ==, 0);
2246 
2247 		VERIFY(0 == sa_set_sa_object(os, sa_obj));
2248 		sa_register_update_callback(os, zfs_sa_upgrade);
2249 	}
2250 
2251 	spa_history_log_internal(LOG_DS_UPGRADE,
2252 	    dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu",
2253 	    zfsvfs->z_version, newvers, dmu_objset_id(os));
2254 
2255 	dmu_tx_commit(tx);
2256 
2257 	zfsvfs->z_version = newvers;
2258 
2259 	zfs_set_fuid_feature(zfsvfs);
2260 
2261 	return (0);
2262 }
2263 
2264 /*
2265  * Read a property stored within the master node.
2266  */
2267 int
2268 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2269 {
2270 	const char *pname;
2271 	int error = ENOENT;
2272 
2273 	/*
2274 	 * Look up the file system's value for the property.  For the
2275 	 * version property, we look up a slightly different string.
2276 	 */
2277 	if (prop == ZFS_PROP_VERSION)
2278 		pname = ZPL_VERSION_STR;
2279 	else
2280 		pname = zfs_prop_to_name(prop);
2281 
2282 	if (os != NULL)
2283 		error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2284 
2285 	if (error == ENOENT) {
2286 		/* No value set, use the default value */
2287 		switch (prop) {
2288 		case ZFS_PROP_VERSION:
2289 			*value = ZPL_VERSION;
2290 			break;
2291 		case ZFS_PROP_NORMALIZE:
2292 		case ZFS_PROP_UTF8ONLY:
2293 			*value = 0;
2294 			break;
2295 		case ZFS_PROP_CASE:
2296 			*value = ZFS_CASE_SENSITIVE;
2297 			break;
2298 		default:
2299 			return (error);
2300 		}
2301 		error = 0;
2302 	}
2303 	return (error);
2304 }
2305 
2306 static vfsdef_t vfw = {
2307 	VFSDEF_VERSION,
2308 	MNTTYPE_ZFS,
2309 	zfs_vfsinit,
2310 	VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2311 	    VSW_XID|VSW_ZMOUNT,
2312 	&zfs_mntopts
2313 };
2314 
2315 struct modlfs zfs_modlfs = {
2316 	&mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2317 };
2318