xref: /titanic_41/usr/src/uts/common/fs/zfs/zfs_vfsops.c (revision 675fc291908baceb17f92b0b6d961439aaddafc9)
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  */
25 
26 /* Portions Copyright 2010 Robert Milkowski */
27 
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/sysmacros.h>
32 #include <sys/kmem.h>
33 #include <sys/pathname.h>
34 #include <sys/vnode.h>
35 #include <sys/vfs.h>
36 #include <sys/vfs_opreg.h>
37 #include <sys/mntent.h>
38 #include <sys/mount.h>
39 #include <sys/cmn_err.h>
40 #include "fs/fs_subr.h"
41 #include <sys/zfs_znode.h>
42 #include <sys/zfs_dir.h>
43 #include <sys/zil.h>
44 #include <sys/fs/zfs.h>
45 #include <sys/dmu.h>
46 #include <sys/dsl_prop.h>
47 #include <sys/dsl_dataset.h>
48 #include <sys/dsl_deleg.h>
49 #include <sys/spa.h>
50 #include <sys/zap.h>
51 #include <sys/sa.h>
52 #include <sys/sa_impl.h>
53 #include <sys/varargs.h>
54 #include <sys/policy.h>
55 #include <sys/atomic.h>
56 #include <sys/mkdev.h>
57 #include <sys/modctl.h>
58 #include <sys/refstr.h>
59 #include <sys/zfs_ioctl.h>
60 #include <sys/zfs_ctldir.h>
61 #include <sys/zfs_fuid.h>
62 #include <sys/bootconf.h>
63 #include <sys/sunddi.h>
64 #include <sys/dnlc.h>
65 #include <sys/dmu_objset.h>
66 #include <sys/spa_boot.h>
67 #include <sys/zfs_events.h>
68 #include "zfs_comutil.h"
69 
70 int zfsfstype;
71 vfsops_t *zfs_vfsops = NULL;
72 static major_t zfs_major;
73 static minor_t zfs_minor;
74 static kmutex_t	zfs_dev_mtx;
75 
76 extern int sys_shutdown;
77 
78 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
79 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
80 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
81 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
82 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
83 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
84 static void zfs_freevfs(vfs_t *vfsp);
85 
86 static const fs_operation_def_t zfs_vfsops_template[] = {
87 	VFSNAME_MOUNT,		{ .vfs_mount = zfs_mount },
88 	VFSNAME_MOUNTROOT,	{ .vfs_mountroot = zfs_mountroot },
89 	VFSNAME_UNMOUNT,	{ .vfs_unmount = zfs_umount },
90 	VFSNAME_ROOT,		{ .vfs_root = zfs_root },
91 	VFSNAME_STATVFS,	{ .vfs_statvfs = zfs_statvfs },
92 	VFSNAME_SYNC,		{ .vfs_sync = zfs_sync },
93 	VFSNAME_VGET,		{ .vfs_vget = zfs_vget },
94 	VFSNAME_FREEVFS,	{ .vfs_freevfs = zfs_freevfs },
95 	NULL,			NULL
96 };
97 
98 static const fs_operation_def_t zfs_vfsops_eio_template[] = {
99 	VFSNAME_FREEVFS,	{ .vfs_freevfs =  zfs_freevfs },
100 	NULL,			NULL
101 };
102 
103 /*
104  * We need to keep a count of active fs's.
105  * This is necessary to prevent our module
106  * from being unloaded after a umount -f
107  */
108 static uint32_t	zfs_active_fs_count = 0;
109 
110 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
111 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
112 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
113 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
114 
115 /*
116  * MO_DEFAULT is not used since the default value is determined
117  * by the equivalent property.
118  */
119 static mntopt_t mntopts[] = {
120 	{ MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
121 	{ MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
122 	{ MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
123 	{ MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
124 };
125 
126 static mntopts_t zfs_mntopts = {
127 	sizeof (mntopts) / sizeof (mntopt_t),
128 	mntopts
129 };
130 
131 /*ARGSUSED*/
132 int
zfs_sync(vfs_t * vfsp,short flag,cred_t * cr)133 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
134 {
135 	/*
136 	 * Data integrity is job one.  We don't want a compromised kernel
137 	 * writing to the storage pool, so we never sync during panic.
138 	 */
139 	if (panicstr)
140 		return (0);
141 
142 	/*
143 	 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
144 	 * to sync metadata, which they would otherwise cache indefinitely.
145 	 * Semantically, the only requirement is that the sync be initiated.
146 	 * The DMU syncs out txgs frequently, so there's nothing to do.
147 	 */
148 	if (flag & SYNC_ATTR)
149 		return (0);
150 
151 	if (vfsp != NULL) {
152 		/*
153 		 * Sync a specific filesystem.
154 		 */
155 		zfsvfs_t *zfsvfs = vfsp->vfs_data;
156 		dsl_pool_t *dp;
157 
158 		ZFS_ENTER(zfsvfs);
159 		dp = dmu_objset_pool(zfsvfs->z_os);
160 
161 		/*
162 		 * If the system is shutting down, then skip any
163 		 * filesystems which may exist on a suspended pool.
164 		 */
165 		if (sys_shutdown && spa_suspended(dp->dp_spa)) {
166 			ZFS_EXIT(zfsvfs);
167 			return (0);
168 		}
169 
170 		if (zfsvfs->z_log != NULL)
171 			zil_commit(zfsvfs->z_log, 0);
172 
173 		ZFS_EXIT(zfsvfs);
174 	} else {
175 		/*
176 		 * Sync all ZFS filesystems.  This is what happens when you
177 		 * run sync(1M).  Unlike other filesystems, ZFS honors the
178 		 * request by waiting for all pools to commit all dirty data.
179 		 */
180 		spa_sync_allpools();
181 	}
182 
183 	return (0);
184 }
185 
186 static int
zfs_create_unique_device(dev_t * dev)187 zfs_create_unique_device(dev_t *dev)
188 {
189 	major_t new_major;
190 
191 	do {
192 		ASSERT3U(zfs_minor, <=, MAXMIN32);
193 		minor_t start = zfs_minor;
194 		do {
195 			mutex_enter(&zfs_dev_mtx);
196 			if (zfs_minor >= MAXMIN32) {
197 				/*
198 				 * If we're still using the real major
199 				 * keep out of /dev/zfs and /dev/zvol minor
200 				 * number space.  If we're using a getudev()'ed
201 				 * major number, we can use all of its minors.
202 				 */
203 				if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
204 					zfs_minor = ZFS_MIN_MINOR;
205 				else
206 					zfs_minor = 0;
207 			} else {
208 				zfs_minor++;
209 			}
210 			*dev = makedevice(zfs_major, zfs_minor);
211 			mutex_exit(&zfs_dev_mtx);
212 		} while (vfs_devismounted(*dev) && zfs_minor != start);
213 		if (zfs_minor == start) {
214 			/*
215 			 * We are using all ~262,000 minor numbers for the
216 			 * current major number.  Create a new major number.
217 			 */
218 			if ((new_major = getudev()) == (major_t)-1) {
219 				cmn_err(CE_WARN,
220 				    "zfs_mount: Can't get unique major "
221 				    "device number.");
222 				return (-1);
223 			}
224 			mutex_enter(&zfs_dev_mtx);
225 			zfs_major = new_major;
226 			zfs_minor = 0;
227 
228 			mutex_exit(&zfs_dev_mtx);
229 		} else {
230 			break;
231 		}
232 		/* CONSTANTCONDITION */
233 	} while (1);
234 
235 	return (0);
236 }
237 
238 static void
atime_changed_cb(void * arg,uint64_t newval)239 atime_changed_cb(void *arg, uint64_t newval)
240 {
241 	zfsvfs_t *zfsvfs = arg;
242 
243 	if (newval == TRUE) {
244 		zfsvfs->z_atime = TRUE;
245 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
246 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
247 	} else {
248 		zfsvfs->z_atime = FALSE;
249 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
250 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
251 	}
252 }
253 
254 static void
xattr_changed_cb(void * arg,uint64_t newval)255 xattr_changed_cb(void *arg, uint64_t newval)
256 {
257 	zfsvfs_t *zfsvfs = arg;
258 
259 	if (newval == TRUE) {
260 		/* XXX locking on vfs_flag? */
261 		zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
262 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
263 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
264 	} else {
265 		/* XXX locking on vfs_flag? */
266 		zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
267 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
268 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
269 	}
270 }
271 
272 static void
blksz_changed_cb(void * arg,uint64_t newval)273 blksz_changed_cb(void *arg, uint64_t newval)
274 {
275 	zfsvfs_t *zfsvfs = arg;
276 	ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
277 	ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
278 	ASSERT(ISP2(newval));
279 
280 	zfsvfs->z_max_blksz = newval;
281 	zfsvfs->z_vfs->vfs_bsize = newval;
282 }
283 
284 static void
readonly_changed_cb(void * arg,uint64_t newval)285 readonly_changed_cb(void *arg, uint64_t newval)
286 {
287 	zfsvfs_t *zfsvfs = arg;
288 
289 	if (newval) {
290 		/* XXX locking on vfs_flag? */
291 		zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
292 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
293 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
294 	} else {
295 		/* XXX locking on vfs_flag? */
296 		zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
297 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
298 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
299 	}
300 }
301 
302 static void
devices_changed_cb(void * arg,uint64_t newval)303 devices_changed_cb(void *arg, uint64_t newval)
304 {
305 	zfsvfs_t *zfsvfs = arg;
306 
307 	if (newval == FALSE) {
308 		zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
309 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
310 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
311 	} else {
312 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
313 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
314 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
315 	}
316 }
317 
318 static void
setuid_changed_cb(void * arg,uint64_t newval)319 setuid_changed_cb(void *arg, uint64_t newval)
320 {
321 	zfsvfs_t *zfsvfs = arg;
322 
323 	if (newval == FALSE) {
324 		zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
325 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
326 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
327 	} else {
328 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
329 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
330 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
331 	}
332 }
333 
334 static void
exec_changed_cb(void * arg,uint64_t newval)335 exec_changed_cb(void *arg, uint64_t newval)
336 {
337 	zfsvfs_t *zfsvfs = arg;
338 
339 	if (newval == FALSE) {
340 		zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
341 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
342 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
343 	} else {
344 		zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
345 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
346 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
347 	}
348 }
349 
350 /*
351  * The nbmand mount option can be changed at mount time.
352  * We can't allow it to be toggled on live file systems or incorrect
353  * behavior may be seen from cifs clients
354  *
355  * This property isn't registered via dsl_prop_register(), but this callback
356  * will be called when a file system is first mounted
357  */
358 static void
nbmand_changed_cb(void * arg,uint64_t newval)359 nbmand_changed_cb(void *arg, uint64_t newval)
360 {
361 	zfsvfs_t *zfsvfs = arg;
362 	if (newval == FALSE) {
363 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
364 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
365 	} else {
366 		vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
367 		vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
368 	}
369 }
370 
371 static void
snapdir_changed_cb(void * arg,uint64_t newval)372 snapdir_changed_cb(void *arg, uint64_t newval)
373 {
374 	zfsvfs_t *zfsvfs = arg;
375 
376 	zfsvfs->z_show_ctldir = newval;
377 }
378 
379 static void
vscan_changed_cb(void * arg,uint64_t newval)380 vscan_changed_cb(void *arg, uint64_t newval)
381 {
382 	zfsvfs_t *zfsvfs = arg;
383 
384 	zfsvfs->z_vscan = newval;
385 }
386 
387 static void
acl_mode_changed_cb(void * arg,uint64_t newval)388 acl_mode_changed_cb(void *arg, uint64_t newval)
389 {
390 	zfsvfs_t *zfsvfs = arg;
391 
392 	zfsvfs->z_acl_mode = newval;
393 }
394 
395 static void
acl_inherit_changed_cb(void * arg,uint64_t newval)396 acl_inherit_changed_cb(void *arg, uint64_t newval)
397 {
398 	zfsvfs_t *zfsvfs = arg;
399 
400 	zfsvfs->z_acl_inherit = newval;
401 }
402 
403 static int
zfs_register_callbacks(vfs_t * vfsp)404 zfs_register_callbacks(vfs_t *vfsp)
405 {
406 	struct dsl_dataset *ds = NULL;
407 	objset_t *os = NULL;
408 	zfsvfs_t *zfsvfs = NULL;
409 	uint64_t nbmand;
410 	boolean_t readonly = B_FALSE;
411 	boolean_t do_readonly = B_FALSE;
412 	boolean_t setuid = B_FALSE;
413 	boolean_t do_setuid = B_FALSE;
414 	boolean_t exec = B_FALSE;
415 	boolean_t do_exec = B_FALSE;
416 	boolean_t devices = B_FALSE;
417 	boolean_t do_devices = B_FALSE;
418 	boolean_t xattr = B_FALSE;
419 	boolean_t do_xattr = B_FALSE;
420 	boolean_t atime = B_FALSE;
421 	boolean_t do_atime = B_FALSE;
422 	int error = 0;
423 
424 	ASSERT(vfsp);
425 	zfsvfs = vfsp->vfs_data;
426 	ASSERT(zfsvfs);
427 	os = zfsvfs->z_os;
428 
429 	/*
430 	 * The act of registering our callbacks will destroy any mount
431 	 * options we may have.  In order to enable temporary overrides
432 	 * of mount options, we stash away the current values and
433 	 * restore them after we register the callbacks.
434 	 */
435 	if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
436 	    !spa_writeable(dmu_objset_spa(os))) {
437 		readonly = B_TRUE;
438 		do_readonly = B_TRUE;
439 	} else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
440 		readonly = B_FALSE;
441 		do_readonly = B_TRUE;
442 	}
443 	if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
444 		devices = B_FALSE;
445 		setuid = B_FALSE;
446 		do_devices = B_TRUE;
447 		do_setuid = B_TRUE;
448 	} else {
449 		if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
450 			devices = B_FALSE;
451 			do_devices = B_TRUE;
452 		} else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
453 			devices = B_TRUE;
454 			do_devices = B_TRUE;
455 		}
456 
457 		if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
458 			setuid = B_FALSE;
459 			do_setuid = B_TRUE;
460 		} else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
461 			setuid = B_TRUE;
462 			do_setuid = B_TRUE;
463 		}
464 	}
465 	if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
466 		exec = B_FALSE;
467 		do_exec = B_TRUE;
468 	} else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
469 		exec = B_TRUE;
470 		do_exec = B_TRUE;
471 	}
472 	if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
473 		xattr = B_FALSE;
474 		do_xattr = B_TRUE;
475 	} else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
476 		xattr = B_TRUE;
477 		do_xattr = B_TRUE;
478 	}
479 	if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
480 		atime = B_FALSE;
481 		do_atime = B_TRUE;
482 	} else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
483 		atime = B_TRUE;
484 		do_atime = B_TRUE;
485 	}
486 
487 	/*
488 	 * nbmand is a special property.  It can only be changed at
489 	 * mount time.
490 	 *
491 	 * This is weird, but it is documented to only be changeable
492 	 * at mount time.
493 	 */
494 	if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
495 		nbmand = B_FALSE;
496 	} else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
497 		nbmand = B_TRUE;
498 	} else {
499 		char osname[ZFS_MAX_DATASET_NAME_LEN];
500 
501 		dmu_objset_name(os, osname);
502 		if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
503 		    NULL)) {
504 			return (error);
505 		}
506 	}
507 
508 	/*
509 	 * Register property callbacks.
510 	 *
511 	 * It would probably be fine to just check for i/o error from
512 	 * the first prop_register(), but I guess I like to go
513 	 * overboard...
514 	 */
515 	ds = dmu_objset_ds(os);
516 	dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
517 	error = dsl_prop_register(ds,
518 	    zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
519 	error = error ? error : dsl_prop_register(ds,
520 	    zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
521 	error = error ? error : dsl_prop_register(ds,
522 	    zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
523 	error = error ? error : dsl_prop_register(ds,
524 	    zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
525 	error = error ? error : dsl_prop_register(ds,
526 	    zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
527 	error = error ? error : dsl_prop_register(ds,
528 	    zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
529 	error = error ? error : dsl_prop_register(ds,
530 	    zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
531 	error = error ? error : dsl_prop_register(ds,
532 	    zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
533 	error = error ? error : dsl_prop_register(ds,
534 	    zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
535 	error = error ? error : dsl_prop_register(ds,
536 	    zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
537 	    zfsvfs);
538 	error = error ? error : dsl_prop_register(ds,
539 	    zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
540 	dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
541 	if (error)
542 		goto unregister;
543 
544 	/*
545 	 * Invoke our callbacks to restore temporary mount options.
546 	 */
547 	if (do_readonly)
548 		readonly_changed_cb(zfsvfs, readonly);
549 	if (do_setuid)
550 		setuid_changed_cb(zfsvfs, setuid);
551 	if (do_exec)
552 		exec_changed_cb(zfsvfs, exec);
553 	if (do_devices)
554 		devices_changed_cb(zfsvfs, devices);
555 	if (do_xattr)
556 		xattr_changed_cb(zfsvfs, xattr);
557 	if (do_atime)
558 		atime_changed_cb(zfsvfs, atime);
559 
560 	nbmand_changed_cb(zfsvfs, nbmand);
561 
562 	return (0);
563 
564 unregister:
565 	/*
566 	 * We may attempt to unregister some callbacks that are not
567 	 * registered, but this is OK; it will simply return ENOMSG,
568 	 * which we will ignore.
569 	 */
570 	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
571 	    atime_changed_cb, zfsvfs);
572 	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
573 	    xattr_changed_cb, zfsvfs);
574 	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
575 	    blksz_changed_cb, zfsvfs);
576 	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
577 	    readonly_changed_cb, zfsvfs);
578 	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
579 	    devices_changed_cb, zfsvfs);
580 	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
581 	    setuid_changed_cb, zfsvfs);
582 	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
583 	    exec_changed_cb, zfsvfs);
584 	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
585 	    snapdir_changed_cb, zfsvfs);
586 	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE),
587 	    acl_mode_changed_cb, zfsvfs);
588 	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
589 	    acl_inherit_changed_cb, zfsvfs);
590 	(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
591 	    vscan_changed_cb, zfsvfs);
592 	return (error);
593 }
594 
595 static int
zfs_space_delta_cb(dmu_object_type_t bonustype,void * data,uint64_t * userp,uint64_t * groupp)596 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
597     uint64_t *userp, uint64_t *groupp)
598 {
599 	/*
600 	 * Is it a valid type of object to track?
601 	 */
602 	if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
603 		return (SET_ERROR(ENOENT));
604 
605 	/*
606 	 * If we have a NULL data pointer
607 	 * then assume the id's aren't changing and
608 	 * return EEXIST to the dmu to let it know to
609 	 * use the same ids
610 	 */
611 	if (data == NULL)
612 		return (SET_ERROR(EEXIST));
613 
614 	if (bonustype == DMU_OT_ZNODE) {
615 		znode_phys_t *znp = data;
616 		*userp = znp->zp_uid;
617 		*groupp = znp->zp_gid;
618 	} else {
619 		int hdrsize;
620 		sa_hdr_phys_t *sap = data;
621 		sa_hdr_phys_t sa = *sap;
622 		boolean_t swap = B_FALSE;
623 
624 		ASSERT(bonustype == DMU_OT_SA);
625 
626 		if (sa.sa_magic == 0) {
627 			/*
628 			 * This should only happen for newly created
629 			 * files that haven't had the znode data filled
630 			 * in yet.
631 			 */
632 			*userp = 0;
633 			*groupp = 0;
634 			return (0);
635 		}
636 		if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
637 			sa.sa_magic = SA_MAGIC;
638 			sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
639 			swap = B_TRUE;
640 		} else {
641 			VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
642 		}
643 
644 		hdrsize = sa_hdrsize(&sa);
645 		VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
646 		*userp = *((uint64_t *)((uintptr_t)data + hdrsize +
647 		    SA_UID_OFFSET));
648 		*groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
649 		    SA_GID_OFFSET));
650 		if (swap) {
651 			*userp = BSWAP_64(*userp);
652 			*groupp = BSWAP_64(*groupp);
653 		}
654 	}
655 	return (0);
656 }
657 
658 static void
fuidstr_to_sid(zfsvfs_t * zfsvfs,const char * fuidstr,char * domainbuf,int buflen,uid_t * ridp)659 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
660     char *domainbuf, int buflen, uid_t *ridp)
661 {
662 	uint64_t fuid;
663 	const char *domain;
664 
665 	fuid = strtonum(fuidstr, NULL);
666 
667 	domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
668 	if (domain)
669 		(void) strlcpy(domainbuf, domain, buflen);
670 	else
671 		domainbuf[0] = '\0';
672 	*ridp = FUID_RID(fuid);
673 }
674 
675 static uint64_t
zfs_userquota_prop_to_obj(zfsvfs_t * zfsvfs,zfs_userquota_prop_t type)676 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
677 {
678 	switch (type) {
679 	case ZFS_PROP_USERUSED:
680 		return (DMU_USERUSED_OBJECT);
681 	case ZFS_PROP_GROUPUSED:
682 		return (DMU_GROUPUSED_OBJECT);
683 	case ZFS_PROP_USERQUOTA:
684 		return (zfsvfs->z_userquota_obj);
685 	case ZFS_PROP_GROUPQUOTA:
686 		return (zfsvfs->z_groupquota_obj);
687 	}
688 	return (0);
689 }
690 
691 int
zfs_userspace_many(zfsvfs_t * zfsvfs,zfs_userquota_prop_t type,uint64_t * cookiep,void * vbuf,uint64_t * bufsizep)692 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
693     uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
694 {
695 	int error;
696 	zap_cursor_t zc;
697 	zap_attribute_t za;
698 	zfs_useracct_t *buf = vbuf;
699 	uint64_t obj;
700 
701 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
702 		return (SET_ERROR(ENOTSUP));
703 
704 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
705 	if (obj == 0) {
706 		*bufsizep = 0;
707 		return (0);
708 	}
709 
710 	for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
711 	    (error = zap_cursor_retrieve(&zc, &za)) == 0;
712 	    zap_cursor_advance(&zc)) {
713 		if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
714 		    *bufsizep)
715 			break;
716 
717 		fuidstr_to_sid(zfsvfs, za.za_name,
718 		    buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
719 
720 		buf->zu_space = za.za_first_integer;
721 		buf++;
722 	}
723 	if (error == ENOENT)
724 		error = 0;
725 
726 	ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
727 	*bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
728 	*cookiep = zap_cursor_serialize(&zc);
729 	zap_cursor_fini(&zc);
730 	return (error);
731 }
732 
733 /*
734  * buf must be big enough (eg, 32 bytes)
735  */
736 static int
id_to_fuidstr(zfsvfs_t * zfsvfs,const char * domain,uid_t rid,char * buf,boolean_t addok)737 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
738     char *buf, boolean_t addok)
739 {
740 	uint64_t fuid;
741 	int domainid = 0;
742 
743 	if (domain && domain[0]) {
744 		domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
745 		if (domainid == -1)
746 			return (SET_ERROR(ENOENT));
747 	}
748 	fuid = FUID_ENCODE(domainid, rid);
749 	(void) sprintf(buf, "%llx", (longlong_t)fuid);
750 	return (0);
751 }
752 
753 int
zfs_userspace_one(zfsvfs_t * zfsvfs,zfs_userquota_prop_t type,const char * domain,uint64_t rid,uint64_t * valp)754 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
755     const char *domain, uint64_t rid, uint64_t *valp)
756 {
757 	char buf[32];
758 	int err;
759 	uint64_t obj;
760 
761 	*valp = 0;
762 
763 	if (!dmu_objset_userspace_present(zfsvfs->z_os))
764 		return (SET_ERROR(ENOTSUP));
765 
766 	obj = zfs_userquota_prop_to_obj(zfsvfs, type);
767 	if (obj == 0)
768 		return (0);
769 
770 	err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
771 	if (err)
772 		return (err);
773 
774 	err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
775 	if (err == ENOENT)
776 		err = 0;
777 	return (err);
778 }
779 
780 int
zfs_set_userquota(zfsvfs_t * zfsvfs,zfs_userquota_prop_t type,const char * domain,uint64_t rid,uint64_t quota)781 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
782     const char *domain, uint64_t rid, uint64_t quota)
783 {
784 	char buf[32];
785 	int err;
786 	dmu_tx_t *tx;
787 	uint64_t *objp;
788 	boolean_t fuid_dirtied;
789 
790 	if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
791 		return (SET_ERROR(EINVAL));
792 
793 	if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
794 		return (SET_ERROR(ENOTSUP));
795 
796 	objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
797 	    &zfsvfs->z_groupquota_obj;
798 
799 	err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
800 	if (err)
801 		return (err);
802 	fuid_dirtied = zfsvfs->z_fuid_dirty;
803 
804 	tx = dmu_tx_create(zfsvfs->z_os);
805 	dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
806 	if (*objp == 0) {
807 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
808 		    zfs_userquota_prop_prefixes[type]);
809 	}
810 	if (fuid_dirtied)
811 		zfs_fuid_txhold(zfsvfs, tx);
812 	err = dmu_tx_assign(tx, TXG_WAIT);
813 	if (err) {
814 		dmu_tx_abort(tx);
815 		return (err);
816 	}
817 
818 	mutex_enter(&zfsvfs->z_lock);
819 	if (*objp == 0) {
820 		*objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
821 		    DMU_OT_NONE, 0, tx);
822 		VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
823 		    zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
824 	}
825 	mutex_exit(&zfsvfs->z_lock);
826 
827 	if (quota == 0) {
828 		err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
829 		if (err == ENOENT)
830 			err = 0;
831 	} else {
832 		err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
833 	}
834 	ASSERT(err == 0);
835 	if (fuid_dirtied)
836 		zfs_fuid_sync(zfsvfs, tx);
837 	dmu_tx_commit(tx);
838 	return (err);
839 }
840 
841 boolean_t
zfs_fuid_overquota(zfsvfs_t * zfsvfs,boolean_t isgroup,uint64_t fuid)842 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
843 {
844 	char buf[32];
845 	uint64_t used, quota, usedobj, quotaobj;
846 	int err;
847 
848 	usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
849 	quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
850 
851 	if (quotaobj == 0 || zfsvfs->z_replay)
852 		return (B_FALSE);
853 
854 	(void) sprintf(buf, "%llx", (longlong_t)fuid);
855 	err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
856 	if (err != 0)
857 		return (B_FALSE);
858 
859 	err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
860 	if (err != 0)
861 		return (B_FALSE);
862 	return (used >= quota);
863 }
864 
865 boolean_t
zfs_owner_overquota(zfsvfs_t * zfsvfs,znode_t * zp,boolean_t isgroup)866 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
867 {
868 	uint64_t fuid;
869 	uint64_t quotaobj;
870 
871 	quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
872 
873 	fuid = isgroup ? zp->z_gid : zp->z_uid;
874 
875 	if (quotaobj == 0 || zfsvfs->z_replay)
876 		return (B_FALSE);
877 
878 	return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
879 }
880 
881 int
zfsvfs_create(const char * osname,zfsvfs_t ** zfvp)882 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
883 {
884 	objset_t *os;
885 	zfsvfs_t *zfsvfs;
886 	uint64_t zval;
887 	int i, error;
888 	uint64_t sa_obj;
889 
890 	zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
891 
892 	/*
893 	 * We claim to always be readonly so we can open snapshots;
894 	 * other ZPL code will prevent us from writing to snapshots.
895 	 */
896 	error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
897 	if (error) {
898 		kmem_free(zfsvfs, sizeof (zfsvfs_t));
899 		return (error);
900 	}
901 
902 	/*
903 	 * Initialize the zfs-specific filesystem structure.
904 	 * Should probably make this a kmem cache, shuffle fields,
905 	 * and just bzero up to z_hold_mtx[].
906 	 */
907 	zfsvfs->z_vfs = NULL;
908 	zfsvfs->z_parent = zfsvfs;
909 	zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
910 	zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
911 	zfsvfs->z_os = os;
912 
913 	error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
914 	if (error) {
915 		goto out;
916 	} else if (zfsvfs->z_version >
917 	    zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
918 		(void) printf("Can't mount a version %lld file system "
919 		    "on a version %lld pool\n. Pool must be upgraded to mount "
920 		    "this file system.", (u_longlong_t)zfsvfs->z_version,
921 		    (u_longlong_t)spa_version(dmu_objset_spa(os)));
922 		error = SET_ERROR(ENOTSUP);
923 		goto out;
924 	}
925 	if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
926 		goto out;
927 	zfsvfs->z_norm = (int)zval;
928 
929 	if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
930 		goto out;
931 	zfsvfs->z_utf8 = (zval != 0);
932 
933 	if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
934 		goto out;
935 	zfsvfs->z_case = (uint_t)zval;
936 
937 	/*
938 	 * Fold case on file systems that are always or sometimes case
939 	 * insensitive.
940 	 */
941 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
942 	    zfsvfs->z_case == ZFS_CASE_MIXED)
943 		zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
944 
945 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
946 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
947 
948 	if (zfsvfs->z_use_sa) {
949 		/* should either have both of these objects or none */
950 		error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
951 		    &sa_obj);
952 		if (error)
953 			goto out;
954 	} else {
955 		/*
956 		 * Pre SA versions file systems should never touch
957 		 * either the attribute registration or layout objects.
958 		 */
959 		sa_obj = 0;
960 	}
961 
962 	error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
963 	    &zfsvfs->z_attr_table);
964 	if (error)
965 		goto out;
966 
967 	if (zfsvfs->z_version >= ZPL_VERSION_SA)
968 		sa_register_update_callback(os, zfs_sa_upgrade);
969 
970 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
971 	    &zfsvfs->z_root);
972 	if (error)
973 		goto out;
974 	ASSERT(zfsvfs->z_root != 0);
975 
976 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
977 	    &zfsvfs->z_unlinkedobj);
978 	if (error)
979 		goto out;
980 
981 	error = zap_lookup(os, MASTER_NODE_OBJ,
982 	    zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
983 	    8, 1, &zfsvfs->z_userquota_obj);
984 	if (error && error != ENOENT)
985 		goto out;
986 
987 	error = zap_lookup(os, MASTER_NODE_OBJ,
988 	    zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
989 	    8, 1, &zfsvfs->z_groupquota_obj);
990 	if (error && error != ENOENT)
991 		goto out;
992 
993 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
994 	    &zfsvfs->z_fuid_obj);
995 	if (error && error != ENOENT)
996 		goto out;
997 
998 	error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
999 	    &zfsvfs->z_shares_dir);
1000 	if (error && error != ENOENT)
1001 		goto out;
1002 
1003 	mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1004 	mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1005 	list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1006 	    offsetof(znode_t, z_link_node));
1007 	rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1008 	rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1009 	rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1010 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1011 		mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1012 
1013 	*zfvp = zfsvfs;
1014 	return (0);
1015 
1016 out:
1017 	dmu_objset_disown(os, zfsvfs);
1018 	*zfvp = NULL;
1019 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1020 	return (error);
1021 }
1022 
1023 static int
zfsvfs_setup(zfsvfs_t * zfsvfs,boolean_t mounting)1024 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1025 {
1026 	int error;
1027 
1028 	error = zfs_register_callbacks(zfsvfs->z_vfs);
1029 	if (error)
1030 		return (error);
1031 
1032 	/*
1033 	 * Set the objset user_ptr to track its zfsvfs.
1034 	 */
1035 	mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1036 	dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1037 	mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1038 
1039 	zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1040 
1041 	/*
1042 	 * If we are not mounting (ie: online recv), then we don't
1043 	 * have to worry about replaying the log as we blocked all
1044 	 * operations out since we closed the ZIL.
1045 	 */
1046 	if (mounting) {
1047 		boolean_t readonly;
1048 
1049 		/*
1050 		 * During replay we remove the read only flag to
1051 		 * allow replays to succeed.
1052 		 */
1053 		readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1054 		if (readonly != 0)
1055 			zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1056 		else
1057 			zfs_unlinked_drain(zfsvfs);
1058 
1059 		/*
1060 		 * Parse and replay the intent log.
1061 		 *
1062 		 * Because of ziltest, this must be done after
1063 		 * zfs_unlinked_drain().  (Further note: ziltest
1064 		 * doesn't use readonly mounts, where
1065 		 * zfs_unlinked_drain() isn't called.)  This is because
1066 		 * ziltest causes spa_sync() to think it's committed,
1067 		 * but actually it is not, so the intent log contains
1068 		 * many txg's worth of changes.
1069 		 *
1070 		 * In particular, if object N is in the unlinked set in
1071 		 * the last txg to actually sync, then it could be
1072 		 * actually freed in a later txg and then reallocated
1073 		 * in a yet later txg.  This would write a "create
1074 		 * object N" record to the intent log.  Normally, this
1075 		 * would be fine because the spa_sync() would have
1076 		 * written out the fact that object N is free, before
1077 		 * we could write the "create object N" intent log
1078 		 * record.
1079 		 *
1080 		 * But when we are in ziltest mode, we advance the "open
1081 		 * txg" without actually spa_sync()-ing the changes to
1082 		 * disk.  So we would see that object N is still
1083 		 * allocated and in the unlinked set, and there is an
1084 		 * intent log record saying to allocate it.
1085 		 */
1086 		if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1087 			if (zil_replay_disable) {
1088 				zil_destroy(zfsvfs->z_log, B_FALSE);
1089 			} else {
1090 				zfsvfs->z_replay = B_TRUE;
1091 				zil_replay(zfsvfs->z_os, zfsvfs,
1092 				    zfs_replay_vector);
1093 				zfsvfs->z_replay = B_FALSE;
1094 			}
1095 		}
1096 		zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1097 	}
1098 
1099 	return (0);
1100 }
1101 
1102 void
zfsvfs_free(zfsvfs_t * zfsvfs)1103 zfsvfs_free(zfsvfs_t *zfsvfs)
1104 {
1105 	int i;
1106 	extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1107 
1108 	/*
1109 	 * This is a barrier to prevent the filesystem from going away in
1110 	 * zfs_znode_move() until we can safely ensure that the filesystem is
1111 	 * not unmounted. We consider the filesystem valid before the barrier
1112 	 * and invalid after the barrier.
1113 	 */
1114 	rw_enter(&zfsvfs_lock, RW_READER);
1115 	rw_exit(&zfsvfs_lock);
1116 
1117 	zfs_fuid_destroy(zfsvfs);
1118 
1119 	mutex_destroy(&zfsvfs->z_znodes_lock);
1120 	mutex_destroy(&zfsvfs->z_lock);
1121 	list_destroy(&zfsvfs->z_all_znodes);
1122 	rrm_destroy(&zfsvfs->z_teardown_lock);
1123 	rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1124 	rw_destroy(&zfsvfs->z_fuid_lock);
1125 	for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1126 		mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1127 	kmem_free(zfsvfs, sizeof (zfsvfs_t));
1128 }
1129 
1130 static void
zfs_set_fuid_feature(zfsvfs_t * zfsvfs)1131 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1132 {
1133 	zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1134 	if (zfsvfs->z_vfs) {
1135 		if (zfsvfs->z_use_fuids) {
1136 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1137 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1138 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1139 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1140 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1141 			vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1142 		} else {
1143 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1144 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1145 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1146 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1147 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1148 			vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1149 		}
1150 	}
1151 	zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1152 }
1153 
1154 static int
zfs_domount(vfs_t * vfsp,char * osname)1155 zfs_domount(vfs_t *vfsp, char *osname)
1156 {
1157 	dev_t mount_dev;
1158 	uint64_t recordsize, fsid_guid;
1159 	int error = 0;
1160 	zfsvfs_t *zfsvfs;
1161 
1162 	ASSERT(vfsp);
1163 	ASSERT(osname);
1164 
1165 	error = zfsvfs_create(osname, &zfsvfs);
1166 	if (error)
1167 		return (error);
1168 	zfsvfs->z_vfs = vfsp;
1169 
1170 	/* Initialize the generic filesystem structure. */
1171 	vfsp->vfs_bcount = 0;
1172 	vfsp->vfs_data = NULL;
1173 
1174 	if (zfs_create_unique_device(&mount_dev) == -1) {
1175 		error = SET_ERROR(ENODEV);
1176 		goto out;
1177 	}
1178 	ASSERT(vfs_devismounted(mount_dev) == 0);
1179 
1180 	if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1181 	    NULL))
1182 		goto out;
1183 
1184 	vfsp->vfs_dev = mount_dev;
1185 	vfsp->vfs_fstype = zfsfstype;
1186 	vfsp->vfs_bsize = recordsize;
1187 	vfsp->vfs_flag |= VFS_NOTRUNC;
1188 	vfsp->vfs_data = zfsvfs;
1189 
1190 	/*
1191 	 * The fsid is 64 bits, composed of an 8-bit fs type, which
1192 	 * separates our fsid from any other filesystem types, and a
1193 	 * 56-bit objset unique ID.  The objset unique ID is unique to
1194 	 * all objsets open on this system, provided by unique_create().
1195 	 * The 8-bit fs type must be put in the low bits of fsid[1]
1196 	 * because that's where other Solaris filesystems put it.
1197 	 */
1198 	fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1199 	ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1200 	vfsp->vfs_fsid.val[0] = fsid_guid;
1201 	vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1202 	    zfsfstype & 0xFF;
1203 
1204 	/*
1205 	 * Set features for file system.
1206 	 */
1207 	zfs_set_fuid_feature(zfsvfs);
1208 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1209 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1210 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1211 		vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1212 	} else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1213 		vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1214 		vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1215 	}
1216 	vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1217 
1218 	if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1219 		uint64_t pval;
1220 
1221 		atime_changed_cb(zfsvfs, B_FALSE);
1222 		readonly_changed_cb(zfsvfs, B_TRUE);
1223 		if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1224 			goto out;
1225 		xattr_changed_cb(zfsvfs, pval);
1226 		zfsvfs->z_issnap = B_TRUE;
1227 		zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1228 
1229 		mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1230 		dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1231 		mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1232 	} else {
1233 		error = zfsvfs_setup(zfsvfs, B_TRUE);
1234 	}
1235 
1236 	if (!zfsvfs->z_issnap)
1237 		zfsctl_create(zfsvfs);
1238 out:
1239 	if (error) {
1240 		dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1241 		zfsvfs_free(zfsvfs);
1242 	} else {
1243 		atomic_inc_32(&zfs_active_fs_count);
1244 	}
1245 
1246 	return (error);
1247 }
1248 
1249 void
zfs_unregister_callbacks(zfsvfs_t * zfsvfs)1250 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1251 {
1252 	objset_t *os = zfsvfs->z_os;
1253 	struct dsl_dataset *ds;
1254 
1255 	/*
1256 	 * Unregister properties.
1257 	 */
1258 	if (!dmu_objset_is_snapshot(os)) {
1259 		ds = dmu_objset_ds(os);
1260 		VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1261 		    zfsvfs) == 0);
1262 
1263 		VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1264 		    zfsvfs) == 0);
1265 
1266 		VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1267 		    zfsvfs) == 0);
1268 
1269 		VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1270 		    zfsvfs) == 0);
1271 
1272 		VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1273 		    zfsvfs) == 0);
1274 
1275 		VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1276 		    zfsvfs) == 0);
1277 
1278 		VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1279 		    zfsvfs) == 0);
1280 
1281 		VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1282 		    zfsvfs) == 0);
1283 
1284 		VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
1285 		    zfsvfs) == 0);
1286 
1287 		VERIFY(dsl_prop_unregister(ds, "aclinherit",
1288 		    acl_inherit_changed_cb, zfsvfs) == 0);
1289 
1290 		VERIFY(dsl_prop_unregister(ds, "vscan",
1291 		    vscan_changed_cb, zfsvfs) == 0);
1292 	}
1293 }
1294 
1295 /*
1296  * Convert a decimal digit string to a uint64_t integer.
1297  */
1298 static int
str_to_uint64(char * str,uint64_t * objnum)1299 str_to_uint64(char *str, uint64_t *objnum)
1300 {
1301 	uint64_t num = 0;
1302 
1303 	while (*str) {
1304 		if (*str < '0' || *str > '9')
1305 			return (SET_ERROR(EINVAL));
1306 
1307 		num = num*10 + *str++ - '0';
1308 	}
1309 
1310 	*objnum = num;
1311 	return (0);
1312 }
1313 
1314 /*
1315  * The boot path passed from the boot loader is in the form of
1316  * "rootpool-name/root-filesystem-object-number'. Convert this
1317  * string to a dataset name: "rootpool-name/root-filesystem-name".
1318  */
1319 static int
zfs_parse_bootfs(char * bpath,char * outpath)1320 zfs_parse_bootfs(char *bpath, char *outpath)
1321 {
1322 	char *slashp;
1323 	uint64_t objnum;
1324 	int error;
1325 
1326 	if (*bpath == 0 || *bpath == '/')
1327 		return (SET_ERROR(EINVAL));
1328 
1329 	(void) strcpy(outpath, bpath);
1330 
1331 	slashp = strchr(bpath, '/');
1332 
1333 	/* if no '/', just return the pool name */
1334 	if (slashp == NULL) {
1335 		return (0);
1336 	}
1337 
1338 	/* if not a number, just return the root dataset name */
1339 	if (str_to_uint64(slashp+1, &objnum)) {
1340 		return (0);
1341 	}
1342 
1343 	*slashp = '\0';
1344 	error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1345 	*slashp = '/';
1346 
1347 	return (error);
1348 }
1349 
1350 /*
1351  * Check that the hex label string is appropriate for the dataset being
1352  * mounted into the global_zone proper.
1353  *
1354  * Return an error if the hex label string is not default or
1355  * admin_low/admin_high.  For admin_low labels, the corresponding
1356  * dataset must be readonly.
1357  */
1358 int
zfs_check_global_label(const char * dsname,const char * hexsl)1359 zfs_check_global_label(const char *dsname, const char *hexsl)
1360 {
1361 	if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1362 		return (0);
1363 	if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1364 		return (0);
1365 	if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1366 		/* must be readonly */
1367 		uint64_t rdonly;
1368 
1369 		if (dsl_prop_get_integer(dsname,
1370 		    zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1371 			return (SET_ERROR(EACCES));
1372 		return (rdonly ? 0 : EACCES);
1373 	}
1374 	return (SET_ERROR(EACCES));
1375 }
1376 
1377 /*
1378  * Determine whether the mount is allowed according to MAC check.
1379  * by comparing (where appropriate) label of the dataset against
1380  * the label of the zone being mounted into.  If the dataset has
1381  * no label, create one.
1382  *
1383  * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1384  */
1385 static int
zfs_mount_label_policy(vfs_t * vfsp,char * osname)1386 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1387 {
1388 	int		error, retv;
1389 	zone_t		*mntzone = NULL;
1390 	ts_label_t	*mnt_tsl;
1391 	bslabel_t	*mnt_sl;
1392 	bslabel_t	ds_sl;
1393 	char		ds_hexsl[MAXNAMELEN];
1394 
1395 	retv = EACCES;				/* assume the worst */
1396 
1397 	/*
1398 	 * Start by getting the dataset label if it exists.
1399 	 */
1400 	error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1401 	    1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1402 	if (error)
1403 		return (SET_ERROR(EACCES));
1404 
1405 	/*
1406 	 * If labeling is NOT enabled, then disallow the mount of datasets
1407 	 * which have a non-default label already.  No other label checks
1408 	 * are needed.
1409 	 */
1410 	if (!is_system_labeled()) {
1411 		if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1412 			return (0);
1413 		return (SET_ERROR(EACCES));
1414 	}
1415 
1416 	/*
1417 	 * Get the label of the mountpoint.  If mounting into the global
1418 	 * zone (i.e. mountpoint is not within an active zone and the
1419 	 * zoned property is off), the label must be default or
1420 	 * admin_low/admin_high only; no other checks are needed.
1421 	 */
1422 	mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1423 	if (mntzone->zone_id == GLOBAL_ZONEID) {
1424 		uint64_t zoned;
1425 
1426 		zone_rele(mntzone);
1427 
1428 		if (dsl_prop_get_integer(osname,
1429 		    zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1430 			return (SET_ERROR(EACCES));
1431 		if (!zoned)
1432 			return (zfs_check_global_label(osname, ds_hexsl));
1433 		else
1434 			/*
1435 			 * This is the case of a zone dataset being mounted
1436 			 * initially, before the zone has been fully created;
1437 			 * allow this mount into global zone.
1438 			 */
1439 			return (0);
1440 	}
1441 
1442 	mnt_tsl = mntzone->zone_slabel;
1443 	ASSERT(mnt_tsl != NULL);
1444 	label_hold(mnt_tsl);
1445 	mnt_sl = label2bslabel(mnt_tsl);
1446 
1447 	if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1448 		/*
1449 		 * The dataset doesn't have a real label, so fabricate one.
1450 		 */
1451 		char *str = NULL;
1452 
1453 		if (l_to_str_internal(mnt_sl, &str) == 0 &&
1454 		    dsl_prop_set_string(osname,
1455 		    zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1456 		    ZPROP_SRC_LOCAL, str) == 0)
1457 			retv = 0;
1458 		if (str != NULL)
1459 			kmem_free(str, strlen(str) + 1);
1460 	} else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1461 		/*
1462 		 * Now compare labels to complete the MAC check.  If the
1463 		 * labels are equal then allow access.  If the mountpoint
1464 		 * label dominates the dataset label, allow readonly access.
1465 		 * Otherwise, access is denied.
1466 		 */
1467 		if (blequal(mnt_sl, &ds_sl))
1468 			retv = 0;
1469 		else if (bldominates(mnt_sl, &ds_sl)) {
1470 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1471 			retv = 0;
1472 		}
1473 	}
1474 
1475 	label_rele(mnt_tsl);
1476 	zone_rele(mntzone);
1477 	return (retv);
1478 }
1479 
1480 static int
zfs_mountroot(vfs_t * vfsp,enum whymountroot why)1481 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1482 {
1483 	int error = 0;
1484 	static int zfsrootdone = 0;
1485 	zfsvfs_t *zfsvfs = NULL;
1486 	znode_t *zp = NULL;
1487 	vnode_t *vp = NULL;
1488 	char *zfs_bootfs;
1489 	char *zfs_devid;
1490 
1491 	ASSERT(vfsp);
1492 
1493 	/*
1494 	 * The filesystem that we mount as root is defined in the
1495 	 * boot property "zfs-bootfs" with a format of
1496 	 * "poolname/root-dataset-objnum".
1497 	 */
1498 	if (why == ROOT_INIT) {
1499 		if (zfsrootdone++)
1500 			return (SET_ERROR(EBUSY));
1501 		/*
1502 		 * the process of doing a spa_load will require the
1503 		 * clock to be set before we could (for example) do
1504 		 * something better by looking at the timestamp on
1505 		 * an uberblock, so just set it to -1.
1506 		 */
1507 		clkset(-1);
1508 
1509 		if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1510 			cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1511 			    "bootfs name");
1512 			return (SET_ERROR(EINVAL));
1513 		}
1514 		zfs_devid = spa_get_bootprop("diskdevid");
1515 		error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1516 		if (zfs_devid)
1517 			spa_free_bootprop(zfs_devid);
1518 		if (error) {
1519 			spa_free_bootprop(zfs_bootfs);
1520 			cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1521 			    error);
1522 			return (error);
1523 		}
1524 		if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1525 			spa_free_bootprop(zfs_bootfs);
1526 			cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1527 			    error);
1528 			return (error);
1529 		}
1530 
1531 		spa_free_bootprop(zfs_bootfs);
1532 
1533 		if (error = vfs_lock(vfsp))
1534 			return (error);
1535 
1536 		if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1537 			cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1538 			goto out;
1539 		}
1540 
1541 		zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1542 		ASSERT(zfsvfs);
1543 		if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1544 			cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1545 			goto out;
1546 		}
1547 
1548 		vp = ZTOV(zp);
1549 		mutex_enter(&vp->v_lock);
1550 		vp->v_flag |= VROOT;
1551 		mutex_exit(&vp->v_lock);
1552 		rootvp = vp;
1553 
1554 		/*
1555 		 * Leave rootvp held.  The root file system is never unmounted.
1556 		 */
1557 
1558 		vfs_add((struct vnode *)0, vfsp,
1559 		    (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1560 out:
1561 		vfs_unlock(vfsp);
1562 		return (error);
1563 	} else if (why == ROOT_REMOUNT) {
1564 		readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1565 		vfsp->vfs_flag |= VFS_REMOUNT;
1566 
1567 		/* refresh mount options */
1568 		zfs_unregister_callbacks(vfsp->vfs_data);
1569 		return (zfs_register_callbacks(vfsp));
1570 
1571 	} else if (why == ROOT_UNMOUNT) {
1572 		zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1573 		(void) zfs_sync(vfsp, 0, 0);
1574 		return (0);
1575 	}
1576 
1577 	/*
1578 	 * if "why" is equal to anything else other than ROOT_INIT,
1579 	 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1580 	 */
1581 	return (SET_ERROR(ENOTSUP));
1582 }
1583 
1584 /*ARGSUSED*/
1585 static int
zfs_mount(vfs_t * vfsp,vnode_t * mvp,struct mounta * uap,cred_t * cr)1586 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1587 {
1588 	char		*osname;
1589 	pathname_t	spn;
1590 	int		error = 0;
1591 	uio_seg_t	fromspace = (uap->flags & MS_SYSSPACE) ?
1592 	    UIO_SYSSPACE : UIO_USERSPACE;
1593 	int		canwrite;
1594 
1595 	if (mvp->v_type != VDIR)
1596 		return (SET_ERROR(ENOTDIR));
1597 
1598 	mutex_enter(&mvp->v_lock);
1599 	if ((uap->flags & MS_REMOUNT) == 0 &&
1600 	    (uap->flags & MS_OVERLAY) == 0 &&
1601 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1602 		mutex_exit(&mvp->v_lock);
1603 		return (SET_ERROR(EBUSY));
1604 	}
1605 	mutex_exit(&mvp->v_lock);
1606 
1607 	/*
1608 	 * ZFS does not support passing unparsed data in via MS_DATA.
1609 	 * Users should use the MS_OPTIONSTR interface; this means
1610 	 * that all option parsing is already done and the options struct
1611 	 * can be interrogated.
1612 	 */
1613 	if ((uap->flags & MS_DATA) && uap->datalen > 0)
1614 		return (SET_ERROR(EINVAL));
1615 
1616 	/*
1617 	 * Get the objset name (the "special" mount argument).
1618 	 */
1619 	if (error = pn_get(uap->spec, fromspace, &spn))
1620 		return (error);
1621 
1622 	osname = spn.pn_path;
1623 
1624 	/*
1625 	 * Check for mount privilege?
1626 	 *
1627 	 * If we don't have privilege then see if
1628 	 * we have local permission to allow it
1629 	 */
1630 	error = secpolicy_fs_mount(cr, mvp, vfsp);
1631 	if (error) {
1632 		if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1633 			vattr_t		vattr;
1634 
1635 			/*
1636 			 * Make sure user is the owner of the mount point
1637 			 * or has sufficient privileges.
1638 			 */
1639 
1640 			vattr.va_mask = AT_UID;
1641 
1642 			if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1643 				goto out;
1644 			}
1645 
1646 			if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1647 			    VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1648 				goto out;
1649 			}
1650 			secpolicy_fs_mount_clearopts(cr, vfsp);
1651 		} else {
1652 			goto out;
1653 		}
1654 	}
1655 
1656 	/*
1657 	 * Refuse to mount a filesystem if we are in a local zone and the
1658 	 * dataset is not visible.
1659 	 */
1660 	if (!INGLOBALZONE(curproc) &&
1661 	    (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1662 		error = SET_ERROR(EPERM);
1663 		goto out;
1664 	}
1665 
1666 	error = zfs_mount_label_policy(vfsp, osname);
1667 	if (error)
1668 		goto out;
1669 
1670 	/*
1671 	 * When doing a remount, we simply refresh our temporary properties
1672 	 * according to those options set in the current VFS options.
1673 	 */
1674 	if (uap->flags & MS_REMOUNT) {
1675 		/* refresh mount options */
1676 		zfs_unregister_callbacks(vfsp->vfs_data);
1677 		error = zfs_register_callbacks(vfsp);
1678 		goto out;
1679 	}
1680 
1681 	error = zfs_domount(vfsp, osname);
1682 
1683 	/*
1684 	 * Add an extra VFS_HOLD on our parent vfs so that it can't
1685 	 * disappear due to a forced unmount.
1686 	 */
1687 	if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1688 		VFS_HOLD(mvp->v_vfsp);
1689 
1690 out:
1691 	if (error == 0) {
1692 		rw_enter(&rz_zev_rwlock, RW_READER);
1693 		if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_zfs_mount)
1694 			rz_zev_callbacks->rz_zev_zfs_mount(vfsp, mvp, osname,
1695 				uap->flags & MS_REMOUNT ?  B_TRUE : B_FALSE);
1696 		rw_exit(&rz_zev_rwlock);
1697 	}
1698 	pn_free(&spn);
1699 	return (error);
1700 }
1701 
1702 static int
zfs_statvfs(vfs_t * vfsp,struct statvfs64 * statp)1703 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1704 {
1705 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1706 	dev32_t d32;
1707 	uint64_t refdbytes, availbytes, usedobjs, availobjs;
1708 
1709 	ZFS_ENTER(zfsvfs);
1710 
1711 	dmu_objset_space(zfsvfs->z_os,
1712 	    &refdbytes, &availbytes, &usedobjs, &availobjs);
1713 
1714 	/*
1715 	 * The underlying storage pool actually uses multiple block sizes.
1716 	 * We report the fragsize as the smallest block size we support,
1717 	 * and we report our blocksize as the filesystem's maximum blocksize.
1718 	 */
1719 	statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1720 	statp->f_bsize = zfsvfs->z_max_blksz;
1721 
1722 	/*
1723 	 * The following report "total" blocks of various kinds in the
1724 	 * file system, but reported in terms of f_frsize - the
1725 	 * "fragment" size.
1726 	 */
1727 
1728 	statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1729 	statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1730 	statp->f_bavail = statp->f_bfree; /* no root reservation */
1731 
1732 	/*
1733 	 * statvfs() should really be called statufs(), because it assumes
1734 	 * static metadata.  ZFS doesn't preallocate files, so the best
1735 	 * we can do is report the max that could possibly fit in f_files,
1736 	 * and that minus the number actually used in f_ffree.
1737 	 * For f_ffree, report the smaller of the number of object available
1738 	 * and the number of blocks (each object will take at least a block).
1739 	 */
1740 	statp->f_ffree = MIN(availobjs, statp->f_bfree);
1741 	statp->f_favail = statp->f_ffree;	/* no "root reservation" */
1742 	statp->f_files = statp->f_ffree + usedobjs;
1743 
1744 	(void) cmpldev(&d32, vfsp->vfs_dev);
1745 	statp->f_fsid = d32;
1746 
1747 	/*
1748 	 * We're a zfs filesystem.
1749 	 */
1750 	(void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1751 
1752 	statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1753 
1754 	statp->f_namemax = MAXNAMELEN - 1;
1755 
1756 	/*
1757 	 * We have all of 32 characters to stuff a string here.
1758 	 * Is there anything useful we could/should provide?
1759 	 */
1760 	bzero(statp->f_fstr, sizeof (statp->f_fstr));
1761 
1762 	ZFS_EXIT(zfsvfs);
1763 	return (0);
1764 }
1765 
1766 static int
zfs_root(vfs_t * vfsp,vnode_t ** vpp)1767 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1768 {
1769 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1770 	znode_t *rootzp;
1771 	int error;
1772 
1773 	ZFS_ENTER(zfsvfs);
1774 
1775 	error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1776 	if (error == 0)
1777 		*vpp = ZTOV(rootzp);
1778 
1779 	ZFS_EXIT(zfsvfs);
1780 	return (error);
1781 }
1782 
1783 /*
1784  * Teardown the zfsvfs::z_os.
1785  *
1786  * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1787  * and 'z_teardown_inactive_lock' held.
1788  */
1789 static int
zfsvfs_teardown(zfsvfs_t * zfsvfs,boolean_t unmounting)1790 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1791 {
1792 	znode_t	*zp;
1793 
1794 	rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1795 
1796 	if (!unmounting) {
1797 		/*
1798 		 * We purge the parent filesystem's vfsp as the parent
1799 		 * filesystem and all of its snapshots have their vnode's
1800 		 * v_vfsp set to the parent's filesystem's vfsp.  Note,
1801 		 * 'z_parent' is self referential for non-snapshots.
1802 		 */
1803 		(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1804 	}
1805 
1806 	/*
1807 	 * Close the zil. NB: Can't close the zil while zfs_inactive
1808 	 * threads are blocked as zil_close can call zfs_inactive.
1809 	 */
1810 	if (zfsvfs->z_log) {
1811 		zil_close(zfsvfs->z_log);
1812 		zfsvfs->z_log = NULL;
1813 	}
1814 
1815 	rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1816 
1817 	/*
1818 	 * If we are not unmounting (ie: online recv) and someone already
1819 	 * unmounted this file system while we were doing the switcheroo,
1820 	 * or a reopen of z_os failed then just bail out now.
1821 	 */
1822 	if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1823 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
1824 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1825 		return (SET_ERROR(EIO));
1826 	}
1827 
1828 	/*
1829 	 * At this point there are no vops active, and any new vops will
1830 	 * fail with EIO since we have z_teardown_lock for writer (only
1831 	 * relavent for forced unmount).
1832 	 *
1833 	 * Release all holds on dbufs.
1834 	 */
1835 	mutex_enter(&zfsvfs->z_znodes_lock);
1836 	for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1837 	    zp = list_next(&zfsvfs->z_all_znodes, zp))
1838 		if (zp->z_sa_hdl) {
1839 			ASSERT(ZTOV(zp)->v_count > 0);
1840 			zfs_znode_dmu_fini(zp);
1841 		}
1842 	mutex_exit(&zfsvfs->z_znodes_lock);
1843 
1844 	/*
1845 	 * If we are unmounting, set the unmounted flag and let new vops
1846 	 * unblock.  zfs_inactive will have the unmounted behavior, and all
1847 	 * other vops will fail with EIO.
1848 	 */
1849 	if (unmounting) {
1850 		zfsvfs->z_unmounted = B_TRUE;
1851 		rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1852 		rw_exit(&zfsvfs->z_teardown_inactive_lock);
1853 	}
1854 
1855 	/*
1856 	 * z_os will be NULL if there was an error in attempting to reopen
1857 	 * zfsvfs, so just return as the properties had already been
1858 	 * unregistered and cached data had been evicted before.
1859 	 */
1860 	if (zfsvfs->z_os == NULL)
1861 		return (0);
1862 
1863 	/*
1864 	 * Unregister properties.
1865 	 */
1866 	zfs_unregister_callbacks(zfsvfs);
1867 
1868 	/*
1869 	 * Evict cached data
1870 	 */
1871 	if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1872 	    !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1873 		txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1874 	dmu_objset_evict_dbufs(zfsvfs->z_os);
1875 
1876 	return (0);
1877 }
1878 
1879 /*ARGSUSED*/
1880 static int
zfs_umount(vfs_t * vfsp,int fflag,cred_t * cr)1881 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1882 {
1883 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1884 	objset_t *os;
1885 	int ret;
1886 
1887 	ret = secpolicy_fs_unmount(cr, vfsp);
1888 	if (ret) {
1889 		if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1890 		    ZFS_DELEG_PERM_MOUNT, cr))
1891 			return (ret);
1892 	}
1893 
1894 	/*
1895 	 * We purge the parent filesystem's vfsp as the parent filesystem
1896 	 * and all of its snapshots have their vnode's v_vfsp set to the
1897 	 * parent's filesystem's vfsp.  Note, 'z_parent' is self
1898 	 * referential for non-snapshots.
1899 	 */
1900 	(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1901 
1902 	/*
1903 	 * Unmount any snapshots mounted under .zfs before unmounting the
1904 	 * dataset itself.
1905 	 */
1906 	if (zfsvfs->z_ctldir != NULL &&
1907 	    (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1908 		return (ret);
1909 	}
1910 
1911 	if (!(fflag & MS_FORCE)) {
1912 		/*
1913 		 * Check the number of active vnodes in the file system.
1914 		 * Our count is maintained in the vfs structure, but the
1915 		 * number is off by 1 to indicate a hold on the vfs
1916 		 * structure itself.
1917 		 *
1918 		 * The '.zfs' directory maintains a reference of its
1919 		 * own, and any active references underneath are
1920 		 * reflected in the vnode count.
1921 		 */
1922 		if (zfsvfs->z_ctldir == NULL) {
1923 			if (vfsp->vfs_count > 1)
1924 				return (SET_ERROR(EBUSY));
1925 		} else {
1926 			if (vfsp->vfs_count > 2 ||
1927 			    zfsvfs->z_ctldir->v_count > 1)
1928 				return (SET_ERROR(EBUSY));
1929 		}
1930 	}
1931 
1932 	vfsp->vfs_flag |= VFS_UNMOUNTED;
1933 
1934 	rw_enter(&rz_zev_rwlock, RW_READER);
1935 	if (rz_zev_callbacks && rz_zev_callbacks->rz_zev_zfs_umount)
1936 		rz_zev_callbacks->rz_zev_zfs_umount(vfsp);
1937 	rw_exit(&rz_zev_rwlock);
1938 
1939 	VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1940 	os = zfsvfs->z_os;
1941 
1942 	/*
1943 	 * z_os will be NULL if there was an error in
1944 	 * attempting to reopen zfsvfs.
1945 	 */
1946 	if (os != NULL) {
1947 		/*
1948 		 * Unset the objset user_ptr.
1949 		 */
1950 		mutex_enter(&os->os_user_ptr_lock);
1951 		dmu_objset_set_user(os, NULL);
1952 		mutex_exit(&os->os_user_ptr_lock);
1953 
1954 		/*
1955 		 * Finally release the objset
1956 		 */
1957 		dmu_objset_disown(os, zfsvfs);
1958 	}
1959 
1960 	/*
1961 	 * We can now safely destroy the '.zfs' directory node.
1962 	 */
1963 	if (zfsvfs->z_ctldir != NULL)
1964 		zfsctl_destroy(zfsvfs);
1965 
1966 	return (0);
1967 }
1968 
1969 static int
zfs_vget(vfs_t * vfsp,vnode_t ** vpp,fid_t * fidp)1970 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1971 {
1972 	zfsvfs_t	*zfsvfs = vfsp->vfs_data;
1973 	znode_t		*zp;
1974 	uint64_t	object = 0;
1975 	uint64_t	fid_gen = 0;
1976 	uint64_t	gen_mask;
1977 	uint64_t	zp_gen;
1978 	int 		i, err;
1979 
1980 	*vpp = NULL;
1981 
1982 	ZFS_ENTER(zfsvfs);
1983 
1984 	if (fidp->fid_len == LONG_FID_LEN) {
1985 		zfid_long_t	*zlfid = (zfid_long_t *)fidp;
1986 		uint64_t	objsetid = 0;
1987 		uint64_t	setgen = 0;
1988 
1989 		for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1990 			objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1991 
1992 		for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1993 			setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1994 
1995 		ZFS_EXIT(zfsvfs);
1996 
1997 		err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1998 		if (err)
1999 			return (SET_ERROR(EINVAL));
2000 		ZFS_ENTER(zfsvfs);
2001 	}
2002 
2003 	if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2004 		zfid_short_t	*zfid = (zfid_short_t *)fidp;
2005 
2006 		for (i = 0; i < sizeof (zfid->zf_object); i++)
2007 			object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2008 
2009 		for (i = 0; i < sizeof (zfid->zf_gen); i++)
2010 			fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2011 	} else {
2012 		ZFS_EXIT(zfsvfs);
2013 		return (SET_ERROR(EINVAL));
2014 	}
2015 
2016 	/* A zero fid_gen means we are in the .zfs control directories */
2017 	if (fid_gen == 0 &&
2018 	    (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
2019 		*vpp = zfsvfs->z_ctldir;
2020 		ASSERT(*vpp != NULL);
2021 		if (object == ZFSCTL_INO_SNAPDIR) {
2022 			VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2023 			    0, NULL, NULL, NULL, NULL, NULL) == 0);
2024 		} else {
2025 			VN_HOLD(*vpp);
2026 		}
2027 		ZFS_EXIT(zfsvfs);
2028 		return (0);
2029 	}
2030 
2031 	gen_mask = -1ULL >> (64 - 8 * i);
2032 
2033 	dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2034 	if (err = zfs_zget(zfsvfs, object, &zp)) {
2035 		ZFS_EXIT(zfsvfs);
2036 		return (err);
2037 	}
2038 	(void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2039 	    sizeof (uint64_t));
2040 	zp_gen = zp_gen & gen_mask;
2041 	if (zp_gen == 0)
2042 		zp_gen = 1;
2043 	if (zp->z_unlinked || zp_gen != fid_gen) {
2044 		dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2045 		VN_RELE(ZTOV(zp));
2046 		ZFS_EXIT(zfsvfs);
2047 		return (SET_ERROR(EINVAL));
2048 	}
2049 
2050 	*vpp = ZTOV(zp);
2051 	ZFS_EXIT(zfsvfs);
2052 	return (0);
2053 }
2054 
2055 /*
2056  * Block out VOPs and close zfsvfs_t::z_os
2057  *
2058  * Note, if successful, then we return with the 'z_teardown_lock' and
2059  * 'z_teardown_inactive_lock' write held.  We leave ownership of the underlying
2060  * dataset and objset intact so that they can be atomically handed off during
2061  * a subsequent rollback or recv operation and the resume thereafter.
2062  */
2063 int
zfs_suspend_fs(zfsvfs_t * zfsvfs)2064 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2065 {
2066 	int error;
2067 
2068 	if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2069 		return (error);
2070 
2071 	return (0);
2072 }
2073 
2074 /*
2075  * Rebuild SA and release VOPs.  Note that ownership of the underlying dataset
2076  * is an invariant across any of the operations that can be performed while the
2077  * filesystem was suspended.  Whether it succeeded or failed, the preconditions
2078  * are the same: the relevant objset and associated dataset are owned by
2079  * zfsvfs, held, and long held on entry.
2080  */
2081 int
zfs_resume_fs(zfsvfs_t * zfsvfs,const char * osname)2082 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2083 {
2084 	int err;
2085 	znode_t *zp;
2086 	uint64_t sa_obj = 0;
2087 
2088 	ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2089 	ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2090 
2091 	/*
2092 	 * We already own this, so just hold and rele it to update the
2093 	 * objset_t, as the one we had before may have been evicted.
2094 	 */
2095 	VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os));
2096 	VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs);
2097 	VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset));
2098 	dmu_objset_rele(zfsvfs->z_os, zfsvfs);
2099 
2100 	/*
2101 	 * Make sure version hasn't changed
2102 	 */
2103 
2104 	err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
2105 	    &zfsvfs->z_version);
2106 
2107 	if (err)
2108 		goto bail;
2109 
2110 	err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2111 	    ZFS_SA_ATTRS, 8, 1, &sa_obj);
2112 
2113 	if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
2114 		goto bail;
2115 
2116 	if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2117 	    zfs_attr_table,  ZPL_END, &zfsvfs->z_attr_table)) != 0)
2118 		goto bail;
2119 
2120 	if (zfsvfs->z_version >= ZPL_VERSION_SA)
2121 		sa_register_update_callback(zfsvfs->z_os,
2122 		    zfs_sa_upgrade);
2123 
2124 	VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2125 
2126 	zfs_set_fuid_feature(zfsvfs);
2127 
2128 	/*
2129 	 * Attempt to re-establish all the active znodes with
2130 	 * their dbufs.  If a zfs_rezget() fails, then we'll let
2131 	 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2132 	 * when they try to use their znode.
2133 	 */
2134 	mutex_enter(&zfsvfs->z_znodes_lock);
2135 	for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2136 	    zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2137 		(void) zfs_rezget(zp);
2138 	}
2139 	mutex_exit(&zfsvfs->z_znodes_lock);
2140 
2141 bail:
2142 	/* release the VOPs */
2143 	rw_exit(&zfsvfs->z_teardown_inactive_lock);
2144 	rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2145 
2146 	if (err) {
2147 		/*
2148 		 * Since we couldn't setup the sa framework, try to force
2149 		 * unmount this file system.
2150 		 */
2151 		if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2152 			(void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2153 	}
2154 	return (err);
2155 }
2156 
2157 static void
zfs_freevfs(vfs_t * vfsp)2158 zfs_freevfs(vfs_t *vfsp)
2159 {
2160 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
2161 
2162 	/*
2163 	 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2164 	 * from zfs_mount().  Release it here.  If we came through
2165 	 * zfs_mountroot() instead, we didn't grab an extra hold, so
2166 	 * skip the VFS_RELE for rootvfs.
2167 	 */
2168 	if (zfsvfs->z_issnap && (vfsp != rootvfs))
2169 		VFS_RELE(zfsvfs->z_parent->z_vfs);
2170 
2171 	zfsvfs_free(zfsvfs);
2172 
2173 	atomic_dec_32(&zfs_active_fs_count);
2174 }
2175 
2176 /*
2177  * VFS_INIT() initialization.  Note that there is no VFS_FINI(),
2178  * so we can't safely do any non-idempotent initialization here.
2179  * Leave that to zfs_init() and zfs_fini(), which are called
2180  * from the module's _init() and _fini() entry points.
2181  */
2182 /*ARGSUSED*/
2183 static int
zfs_vfsinit(int fstype,char * name)2184 zfs_vfsinit(int fstype, char *name)
2185 {
2186 	int error;
2187 
2188 	zfsfstype = fstype;
2189 
2190 	/*
2191 	 * Setup vfsops and vnodeops tables.
2192 	 */
2193 	error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2194 	if (error != 0) {
2195 		cmn_err(CE_WARN, "zfs: bad vfs ops template");
2196 	}
2197 
2198 	error = zfs_create_op_tables();
2199 	if (error) {
2200 		zfs_remove_op_tables();
2201 		cmn_err(CE_WARN, "zfs: bad vnode ops template");
2202 		(void) vfs_freevfsops_by_type(zfsfstype);
2203 		return (error);
2204 	}
2205 
2206 	mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2207 
2208 	/*
2209 	 * Unique major number for all zfs mounts.
2210 	 * If we run out of 32-bit minors, we'll getudev() another major.
2211 	 */
2212 	zfs_major = ddi_name_to_major(ZFS_DRIVER);
2213 	zfs_minor = ZFS_MIN_MINOR;
2214 
2215 	return (0);
2216 }
2217 
2218 void
zfs_init(void)2219 zfs_init(void)
2220 {
2221 	/*
2222 	 * Initialize .zfs directory structures
2223 	 */
2224 	zfsctl_init();
2225 
2226 	/*
2227 	 * Initialize znode cache, vnode ops, etc...
2228 	 */
2229 	zfs_znode_init();
2230 
2231 	dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2232 }
2233 
2234 void
zfs_fini(void)2235 zfs_fini(void)
2236 {
2237 	zfsctl_fini();
2238 	zfs_znode_fini();
2239 }
2240 
2241 int
zfs_busy(void)2242 zfs_busy(void)
2243 {
2244 	return (zfs_active_fs_count != 0);
2245 }
2246 
2247 int
zfs_set_version(zfsvfs_t * zfsvfs,uint64_t newvers)2248 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2249 {
2250 	int error;
2251 	objset_t *os = zfsvfs->z_os;
2252 	dmu_tx_t *tx;
2253 
2254 	if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2255 		return (SET_ERROR(EINVAL));
2256 
2257 	if (newvers < zfsvfs->z_version)
2258 		return (SET_ERROR(EINVAL));
2259 
2260 	if (zfs_spa_version_map(newvers) >
2261 	    spa_version(dmu_objset_spa(zfsvfs->z_os)))
2262 		return (SET_ERROR(ENOTSUP));
2263 
2264 	tx = dmu_tx_create(os);
2265 	dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2266 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2267 		dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2268 		    ZFS_SA_ATTRS);
2269 		dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2270 	}
2271 	error = dmu_tx_assign(tx, TXG_WAIT);
2272 	if (error) {
2273 		dmu_tx_abort(tx);
2274 		return (error);
2275 	}
2276 
2277 	error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2278 	    8, 1, &newvers, tx);
2279 
2280 	if (error) {
2281 		dmu_tx_commit(tx);
2282 		return (error);
2283 	}
2284 
2285 	if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2286 		uint64_t sa_obj;
2287 
2288 		ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2289 		    SPA_VERSION_SA);
2290 		sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2291 		    DMU_OT_NONE, 0, tx);
2292 
2293 		error = zap_add(os, MASTER_NODE_OBJ,
2294 		    ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2295 		ASSERT0(error);
2296 
2297 		VERIFY(0 == sa_set_sa_object(os, sa_obj));
2298 		sa_register_update_callback(os, zfs_sa_upgrade);
2299 	}
2300 
2301 	spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2302 	    "from %llu to %llu", zfsvfs->z_version, newvers);
2303 
2304 	dmu_tx_commit(tx);
2305 
2306 	zfsvfs->z_version = newvers;
2307 
2308 	zfs_set_fuid_feature(zfsvfs);
2309 
2310 	return (0);
2311 }
2312 
2313 /*
2314  * Read a property stored within the master node.
2315  */
2316 int
zfs_get_zplprop(objset_t * os,zfs_prop_t prop,uint64_t * value)2317 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2318 {
2319 	const char *pname;
2320 	int error = ENOENT;
2321 
2322 	/*
2323 	 * Look up the file system's value for the property.  For the
2324 	 * version property, we look up a slightly different string.
2325 	 */
2326 	if (prop == ZFS_PROP_VERSION)
2327 		pname = ZPL_VERSION_STR;
2328 	else
2329 		pname = zfs_prop_to_name(prop);
2330 
2331 	if (os != NULL)
2332 		error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2333 
2334 	if (error == ENOENT) {
2335 		/* No value set, use the default value */
2336 		switch (prop) {
2337 		case ZFS_PROP_VERSION:
2338 			*value = ZPL_VERSION;
2339 			break;
2340 		case ZFS_PROP_NORMALIZE:
2341 		case ZFS_PROP_UTF8ONLY:
2342 			*value = 0;
2343 			break;
2344 		case ZFS_PROP_CASE:
2345 			*value = ZFS_CASE_SENSITIVE;
2346 			break;
2347 		default:
2348 			return (error);
2349 		}
2350 		error = 0;
2351 	}
2352 	return (error);
2353 }
2354 
2355 static vfsdef_t vfw = {
2356 	VFSDEF_VERSION,
2357 	MNTTYPE_ZFS,
2358 	zfs_vfsinit,
2359 	VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2360 	    VSW_XID|VSW_ZMOUNT,
2361 	&zfs_mntopts
2362 };
2363 
2364 struct modlfs zfs_modlfs = {
2365 	&mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2366 };
2367