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.
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.
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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
25 /* Portions Copyright 2010 Robert Milkowski */
27 #include <sys/types.h>
28 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/sysmacros.h>
32 #include <sys/pathname.h>
33 #include <sys/vnode.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>
43 #include <sys/fs/zfs.h>
45 #include <sys/dsl_prop.h>
46 #include <sys/dsl_dataset.h>
47 #include <sys/dsl_deleg.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>
63 #include <sys/dmu_objset.h>
64 #include <sys/spa_boot.h>
66 #include "zfs_comutil.h"
69 extern int sys_shutdown;
71 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
72 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
73 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
74 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
77 * MO_DEFAULT is not used since the default value is determined
78 * by the equivalent property.
80 static mntopt_t mntopts[] = {
81 { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
82 { MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
83 { MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
84 { MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
87 static mntopts_t zfs_mntopts = {
88 sizeof (mntopts) / sizeof (mntopt_t),
94 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
97 * Data integrity is job one. We don't want a compromised kernel
98 * writing to the storage pool, so we never sync during panic.
104 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
105 * to sync metadata, which they would otherwise cache indefinitely.
106 * Semantically, the only requirement is that the sync be initiated.
107 * The DMU syncs out txgs frequently, so there's nothing to do.
109 if (flag & SYNC_ATTR)
114 * Sync a specific filesystem.
116 zfsvfs_t *zfsvfs = vfsp->vfs_data;
120 dp = dmu_objset_pool(zfsvfs->z_os);
123 * If the system is shutting down, then skip any
124 * filesystems which may exist on a suspended pool.
126 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
131 if (zfsvfs->z_log != NULL)
132 zil_commit(zfsvfs->z_log, 0);
137 * Sync all ZFS filesystems. This is what happens when you
138 * run sync(1M). Unlike other filesystems, ZFS honors the
139 * request by waiting for all pools to commit all dirty data.
146 EXPORT_SYMBOL(zfs_sync);
149 atime_changed_cb(void *arg, uint64_t newval)
151 zfsvfs_t *zfsvfs = arg;
153 if (newval == TRUE) {
154 zfsvfs->z_atime = TRUE;
155 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
156 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
158 zfsvfs->z_atime = FALSE;
159 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
160 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
165 xattr_changed_cb(void *arg, uint64_t newval)
167 zfsvfs_t *zfsvfs = arg;
169 if (newval == TRUE) {
170 /* XXX locking on vfs_flag? */
171 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
172 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
173 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
175 /* XXX locking on vfs_flag? */
176 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
177 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
178 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
183 blksz_changed_cb(void *arg, uint64_t newval)
185 zfsvfs_t *zfsvfs = arg;
187 if (newval < SPA_MINBLOCKSIZE ||
188 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
189 newval = SPA_MAXBLOCKSIZE;
191 zfsvfs->z_max_blksz = newval;
192 zfsvfs->z_vfs->vfs_bsize = newval;
196 readonly_changed_cb(void *arg, uint64_t newval)
198 zfsvfs_t *zfsvfs = arg;
201 /* XXX locking on vfs_flag? */
202 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
203 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
204 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
206 /* XXX locking on vfs_flag? */
207 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
208 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
209 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
214 devices_changed_cb(void *arg, uint64_t newval)
216 zfsvfs_t *zfsvfs = arg;
218 if (newval == FALSE) {
219 zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
220 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
221 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
223 zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
224 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
225 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
230 setuid_changed_cb(void *arg, uint64_t newval)
232 zfsvfs_t *zfsvfs = arg;
234 if (newval == FALSE) {
235 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
236 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
237 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
239 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
240 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
241 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
246 exec_changed_cb(void *arg, uint64_t newval)
248 zfsvfs_t *zfsvfs = arg;
250 if (newval == FALSE) {
251 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
252 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
253 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
255 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
256 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
257 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
262 * The nbmand mount option can be changed at mount time.
263 * We can't allow it to be toggled on live file systems or incorrect
264 * behavior may be seen from cifs clients
266 * This property isn't registered via dsl_prop_register(), but this callback
267 * will be called when a file system is first mounted
270 nbmand_changed_cb(void *arg, uint64_t newval)
272 zfsvfs_t *zfsvfs = arg;
273 if (newval == FALSE) {
274 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
275 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
277 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
278 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
283 snapdir_changed_cb(void *arg, uint64_t newval)
285 zfsvfs_t *zfsvfs = arg;
287 zfsvfs->z_show_ctldir = newval;
291 vscan_changed_cb(void *arg, uint64_t newval)
293 zfsvfs_t *zfsvfs = arg;
295 zfsvfs->z_vscan = newval;
299 acl_inherit_changed_cb(void *arg, uint64_t newval)
301 zfsvfs_t *zfsvfs = arg;
303 zfsvfs->z_acl_inherit = newval;
307 zfs_register_callbacks(vfs_t *vfsp)
309 struct dsl_dataset *ds = NULL;
311 zfsvfs_t *zfsvfs = NULL;
313 int readonly, do_readonly = B_FALSE;
314 int setuid, do_setuid = B_FALSE;
315 int exec, do_exec = B_FALSE;
316 int devices, do_devices = B_FALSE;
317 int xattr, do_xattr = B_FALSE;
318 int atime, do_atime = B_FALSE;
322 zfsvfs = vfsp->vfs_data;
327 * The act of registering our callbacks will destroy any mount
328 * options we may have. In order to enable temporary overrides
329 * of mount options, we stash away the current values and
330 * restore them after we register the callbacks.
332 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
333 !spa_writeable(dmu_objset_spa(os))) {
335 do_readonly = B_TRUE;
336 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
338 do_readonly = B_TRUE;
340 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
346 if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
349 } else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
354 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
357 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
362 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
365 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
369 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
372 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
376 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
379 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
385 * nbmand is a special property. It can only be changed at
388 * This is weird, but it is documented to only be changeable
391 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
393 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
396 char osname[MAXNAMELEN];
398 dmu_objset_name(os, osname);
399 if ((error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
406 * Register property callbacks.
408 * It would probably be fine to just check for i/o error from
409 * the first prop_register(), but I guess I like to go
412 ds = dmu_objset_ds(os);
413 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs);
414 error = error ? error : dsl_prop_register(ds,
415 "xattr", xattr_changed_cb, zfsvfs);
416 error = error ? error : dsl_prop_register(ds,
417 "recordsize", blksz_changed_cb, zfsvfs);
418 error = error ? error : dsl_prop_register(ds,
419 "readonly", readonly_changed_cb, zfsvfs);
420 error = error ? error : dsl_prop_register(ds,
421 "devices", devices_changed_cb, zfsvfs);
422 error = error ? error : dsl_prop_register(ds,
423 "setuid", setuid_changed_cb, zfsvfs);
424 error = error ? error : dsl_prop_register(ds,
425 "exec", exec_changed_cb, zfsvfs);
426 error = error ? error : dsl_prop_register(ds,
427 "snapdir", snapdir_changed_cb, zfsvfs);
428 error = error ? error : dsl_prop_register(ds,
429 "aclinherit", acl_inherit_changed_cb, zfsvfs);
430 error = error ? error : dsl_prop_register(ds,
431 "vscan", vscan_changed_cb, zfsvfs);
436 * Invoke our callbacks to restore temporary mount options.
439 readonly_changed_cb(zfsvfs, readonly);
441 setuid_changed_cb(zfsvfs, setuid);
443 exec_changed_cb(zfsvfs, exec);
445 devices_changed_cb(zfsvfs, devices);
447 xattr_changed_cb(zfsvfs, xattr);
449 atime_changed_cb(zfsvfs, atime);
451 nbmand_changed_cb(zfsvfs, nbmand);
457 * We may attempt to unregister some callbacks that are not
458 * registered, but this is OK; it will simply return ENOMSG,
459 * which we will ignore.
461 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs);
462 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs);
463 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs);
464 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs);
465 (void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zfsvfs);
466 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs);
467 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs);
468 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs);
469 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
471 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs);
475 EXPORT_SYMBOL(zfs_register_callbacks);
476 #endif /* HAVE_ZPL */
479 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
480 uint64_t *userp, uint64_t *groupp)
482 znode_phys_t *znp = data;
486 * Is it a valid type of object to track?
488 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
492 * If we have a NULL data pointer
493 * then assume the id's aren't changing and
494 * return EEXIST to the dmu to let it know to
500 if (bonustype == DMU_OT_ZNODE) {
501 *userp = znp->zp_uid;
502 *groupp = znp->zp_gid;
506 ASSERT(bonustype == DMU_OT_SA);
507 hdrsize = sa_hdrsize(data);
510 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
512 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
516 * This should only happen for newly created
517 * files that haven't had the znode data filled
529 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
530 char *domainbuf, int buflen, uid_t *ridp)
535 fuid = strtonum(fuidstr, NULL);
537 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
539 (void) strlcpy(domainbuf, domain, buflen);
542 *ridp = FUID_RID(fuid);
546 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
549 case ZFS_PROP_USERUSED:
550 return (DMU_USERUSED_OBJECT);
551 case ZFS_PROP_GROUPUSED:
552 return (DMU_GROUPUSED_OBJECT);
553 case ZFS_PROP_USERQUOTA:
554 return (zfsvfs->z_userquota_obj);
555 case ZFS_PROP_GROUPQUOTA:
556 return (zfsvfs->z_groupquota_obj);
564 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
565 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
570 zfs_useracct_t *buf = vbuf;
573 if (!dmu_objset_userspace_present(zfsvfs->z_os))
576 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
582 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
583 (error = zap_cursor_retrieve(&zc, &za)) == 0;
584 zap_cursor_advance(&zc)) {
585 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
589 fuidstr_to_sid(zfsvfs, za.za_name,
590 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
592 buf->zu_space = za.za_first_integer;
598 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
599 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
600 *cookiep = zap_cursor_serialize(&zc);
601 zap_cursor_fini(&zc);
604 EXPORT_SYMBOL(zfs_userspace_many);
607 * buf must be big enough (eg, 32 bytes)
610 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
611 char *buf, boolean_t addok)
616 if (domain && domain[0]) {
617 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
621 fuid = FUID_ENCODE(domainid, rid);
622 (void) sprintf(buf, "%llx", (longlong_t)fuid);
627 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
628 const char *domain, uint64_t rid, uint64_t *valp)
636 if (!dmu_objset_userspace_present(zfsvfs->z_os))
639 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
643 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
647 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
652 EXPORT_SYMBOL(zfs_userspace_one);
655 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
656 const char *domain, uint64_t rid, uint64_t quota)
662 boolean_t fuid_dirtied;
664 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
667 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
670 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
671 &zfsvfs->z_groupquota_obj;
673 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
676 fuid_dirtied = zfsvfs->z_fuid_dirty;
678 tx = dmu_tx_create(zfsvfs->z_os);
679 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
681 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
682 zfs_userquota_prop_prefixes[type]);
685 zfs_fuid_txhold(zfsvfs, tx);
686 err = dmu_tx_assign(tx, TXG_WAIT);
692 mutex_enter(&zfsvfs->z_lock);
694 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
696 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
697 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
699 mutex_exit(&zfsvfs->z_lock);
702 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
706 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
710 zfs_fuid_sync(zfsvfs, tx);
714 EXPORT_SYMBOL(zfs_set_userquota);
717 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
720 uint64_t used, quota, usedobj, quotaobj;
723 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
724 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
726 if (quotaobj == 0 || zfsvfs->z_replay)
729 (void) sprintf(buf, "%llx", (longlong_t)fuid);
730 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
734 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
737 return (used >= quota);
739 EXPORT_SYMBOL(zfs_fuid_overquota);
742 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
747 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
749 fuid = isgroup ? zp->z_gid : zp->z_uid;
751 if (quotaobj == 0 || zfsvfs->z_replay)
754 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
756 EXPORT_SYMBOL(zfs_owner_overquota);
759 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
767 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
770 * We claim to always be readonly so we can open snapshots;
771 * other ZPL code will prevent us from writing to snapshots.
773 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
775 kmem_free(zfsvfs, sizeof (zfsvfs_t));
780 * Initialize the zfs-specific filesystem structure.
781 * Should probably make this a kmem cache, shuffle fields,
782 * and just bzero up to z_hold_mtx[].
784 zfsvfs->z_vfs = NULL;
785 zfsvfs->z_parent = zfsvfs;
786 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
787 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
790 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
793 } else if (zfsvfs->z_version >
794 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
795 (void) printk("Can't mount a version %lld file system "
796 "on a version %lld pool\n. Pool must be upgraded to mount "
797 "this file system.", (u_longlong_t)zfsvfs->z_version,
798 (u_longlong_t)spa_version(dmu_objset_spa(os)));
802 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
804 zfsvfs->z_norm = (int)zval;
806 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
808 zfsvfs->z_utf8 = (zval != 0);
810 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
812 zfsvfs->z_case = (uint_t)zval;
815 * Fold case on file systems that are always or sometimes case
818 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
819 zfsvfs->z_case == ZFS_CASE_MIXED)
820 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
822 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
823 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
825 if (zfsvfs->z_use_sa) {
826 /* should either have both of these objects or none */
827 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
833 * Pre SA versions file systems should never touch
834 * either the attribute registration or layout objects.
839 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
840 &zfsvfs->z_attr_table);
844 if (zfsvfs->z_version >= ZPL_VERSION_SA)
845 sa_register_update_callback(os, zfs_sa_upgrade);
847 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
851 ASSERT(zfsvfs->z_root != 0);
853 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
854 &zfsvfs->z_unlinkedobj);
858 error = zap_lookup(os, MASTER_NODE_OBJ,
859 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
860 8, 1, &zfsvfs->z_userquota_obj);
861 if (error && error != ENOENT)
864 error = zap_lookup(os, MASTER_NODE_OBJ,
865 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
866 8, 1, &zfsvfs->z_groupquota_obj);
867 if (error && error != ENOENT)
870 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
871 &zfsvfs->z_fuid_obj);
872 if (error && error != ENOENT)
875 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
876 &zfsvfs->z_shares_dir);
877 if (error && error != ENOENT)
880 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
881 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
882 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
883 offsetof(znode_t, z_link_node));
884 rrw_init(&zfsvfs->z_teardown_lock);
885 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
886 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
887 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
888 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
894 dmu_objset_disown(os, zfsvfs);
896 kmem_free(zfsvfs, sizeof (zfsvfs_t));
901 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
905 error = zfs_register_callbacks(zfsvfs->z_vfs);
910 * Set the objset user_ptr to track its zfsvfs.
912 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
913 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
914 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
916 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
919 * If we are not mounting (ie: online recv), then we don't
920 * have to worry about replaying the log as we blocked all
921 * operations out since we closed the ZIL.
927 * During replay we remove the read only flag to
928 * allow replays to succeed.
930 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
932 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
934 zfs_unlinked_drain(zfsvfs);
937 * Parse and replay the intent log.
939 * Because of ziltest, this must be done after
940 * zfs_unlinked_drain(). (Further note: ziltest
941 * doesn't use readonly mounts, where
942 * zfs_unlinked_drain() isn't called.) This is because
943 * ziltest causes spa_sync() to think it's committed,
944 * but actually it is not, so the intent log contains
945 * many txg's worth of changes.
947 * In particular, if object N is in the unlinked set in
948 * the last txg to actually sync, then it could be
949 * actually freed in a later txg and then reallocated
950 * in a yet later txg. This would write a "create
951 * object N" record to the intent log. Normally, this
952 * would be fine because the spa_sync() would have
953 * written out the fact that object N is free, before
954 * we could write the "create object N" intent log
957 * But when we are in ziltest mode, we advance the "open
958 * txg" without actually spa_sync()-ing the changes to
959 * disk. So we would see that object N is still
960 * allocated and in the unlinked set, and there is an
961 * intent log record saying to allocate it.
963 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
964 if (zil_replay_disable) {
965 zil_destroy(zfsvfs->z_log, B_FALSE);
967 zfsvfs->z_replay = B_TRUE;
968 zil_replay(zfsvfs->z_os, zfsvfs,
970 zfsvfs->z_replay = B_FALSE;
973 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
980 zfsvfs_free(zfsvfs_t *zfsvfs)
983 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
986 * This is a barrier to prevent the filesystem from going away in
987 * zfs_znode_move() until we can safely ensure that the filesystem is
988 * not unmounted. We consider the filesystem valid before the barrier
989 * and invalid after the barrier.
991 rw_enter(&zfsvfs_lock, RW_READER);
992 rw_exit(&zfsvfs_lock);
994 zfs_fuid_destroy(zfsvfs);
996 mutex_destroy(&zfsvfs->z_znodes_lock);
997 mutex_destroy(&zfsvfs->z_lock);
998 list_destroy(&zfsvfs->z_all_znodes);
999 rrw_destroy(&zfsvfs->z_teardown_lock);
1000 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1001 rw_destroy(&zfsvfs->z_fuid_lock);
1002 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1003 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1004 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1008 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1010 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1011 if (zfsvfs->z_use_fuids && zfsvfs->z_vfs) {
1012 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1013 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1014 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1015 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1016 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1017 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1019 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1023 zfs_domount(vfs_t *vfsp, char *osname)
1025 uint64_t recordsize, fsid_guid;
1032 error = zfsvfs_create(osname, &zfsvfs);
1035 zfsvfs->z_vfs = vfsp;
1037 /* Initialize the generic filesystem structure. */
1038 vfsp->vfs_bcount = 0;
1039 vfsp->vfs_data = NULL;
1041 if ((error = dsl_prop_get_integer(osname, "recordsize",
1042 &recordsize, NULL)))
1045 vfsp->vfs_bsize = recordsize;
1046 vfsp->vfs_flag |= VFS_NOTRUNC;
1047 vfsp->vfs_data = zfsvfs;
1050 * The fsid is 64 bits, composed of an 8-bit fs type, which
1051 * separates our fsid from any other filesystem types, and a
1052 * 56-bit objset unique ID. The objset unique ID is unique to
1053 * all objsets open on this system, provided by unique_create().
1054 * The 8-bit fs type must be put in the low bits of fsid[1]
1055 * because that's where other Solaris filesystems put it.
1057 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1058 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1059 vfsp->vfs_fsid.val[0] = fsid_guid;
1060 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8);
1063 * Set features for file system.
1065 zfs_set_fuid_feature(zfsvfs);
1066 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1067 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1068 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1069 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1070 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1071 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1072 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1074 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1076 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1079 atime_changed_cb(zfsvfs, B_FALSE);
1080 readonly_changed_cb(zfsvfs, B_TRUE);
1081 if ((error = dsl_prop_get_integer(osname,"xattr",&pval,NULL)))
1083 xattr_changed_cb(zfsvfs, pval);
1084 zfsvfs->z_issnap = B_TRUE;
1085 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1087 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1088 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1089 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1091 error = zfsvfs_setup(zfsvfs, B_TRUE);
1094 if (!zfsvfs->z_issnap)
1095 zfsctl_create(zfsvfs);
1098 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1099 zfsvfs_free(zfsvfs);
1104 EXPORT_SYMBOL(zfs_domount);
1107 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1109 objset_t *os = zfsvfs->z_os;
1110 struct dsl_dataset *ds;
1113 * Unregister properties.
1115 if (!dmu_objset_is_snapshot(os)) {
1116 ds = dmu_objset_ds(os);
1117 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1120 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1123 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1126 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1129 VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1132 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1135 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1138 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1141 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1142 acl_inherit_changed_cb, zfsvfs) == 0);
1144 VERIFY(dsl_prop_unregister(ds, "vscan",
1145 vscan_changed_cb, zfsvfs) == 0);
1148 EXPORT_SYMBOL(zfs_unregister_callbacks);
1150 #ifdef HAVE_MLSLABEL
1152 * zfs_check_global_label:
1153 * Check that the hex label string is appropriate for the dataset
1154 * being mounted into the global_zone proper.
1156 * Return an error if the hex label string is not default or
1157 * admin_low/admin_high. For admin_low labels, the corresponding
1158 * dataset must be readonly.
1161 zfs_check_global_label(const char *dsname, const char *hexsl)
1163 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1165 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1167 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1168 /* must be readonly */
1171 if (dsl_prop_get_integer(dsname,
1172 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1174 return (rdonly ? 0 : EACCES);
1178 #endif /* HAVE_MLSLABEL */
1181 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1183 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1185 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1189 dmu_objset_space(zfsvfs->z_os,
1190 &refdbytes, &availbytes, &usedobjs, &availobjs);
1193 * The underlying storage pool actually uses multiple block sizes.
1194 * We report the fragsize as the smallest block size we support,
1195 * and we report our blocksize as the filesystem's maximum blocksize.
1197 statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1198 statp->f_bsize = zfsvfs->z_max_blksz;
1201 * The following report "total" blocks of various kinds in the
1202 * file system, but reported in terms of f_frsize - the
1206 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1207 statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1208 statp->f_bavail = statp->f_bfree; /* no root reservation */
1211 * statvfs() should really be called statufs(), because it assumes
1212 * static metadata. ZFS doesn't preallocate files, so the best
1213 * we can do is report the max that could possibly fit in f_files,
1214 * and that minus the number actually used in f_ffree.
1215 * For f_ffree, report the smaller of the number of object available
1216 * and the number of blocks (each object will take at least a block).
1218 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1219 statp->f_favail = statp->f_ffree; /* no "root reservation" */
1220 statp->f_files = statp->f_ffree + usedobjs;
1222 (void) cmpldev(&d32, vfsp->vfs_dev);
1223 statp->f_fsid = d32;
1226 * We're a zfs filesystem.
1228 (void) strcpy(statp->f_basetype, MNTTYPE_ZFS);
1230 statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1232 statp->f_namemax = ZFS_MAXNAMELEN;
1235 * We have all of 32 characters to stuff a string here.
1236 * Is there anything useful we could/should provide?
1238 bzero(statp->f_fstr, sizeof (statp->f_fstr));
1243 EXPORT_SYMBOL(zfs_statvfs);
1246 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1248 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1254 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1256 *vpp = ZTOV(rootzp);
1261 EXPORT_SYMBOL(zfs_root);
1264 * Teardown the zfsvfs::z_os.
1266 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1267 * and 'z_teardown_inactive_lock' held.
1270 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1274 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1278 * We purge the parent filesystem's vfsp as the parent
1279 * filesystem and all of its snapshots have their vnode's
1280 * v_vfsp set to the parent's filesystem's vfsp. Note,
1281 * 'z_parent' is self referential for non-snapshots.
1283 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1287 * Close the zil. NB: Can't close the zil while zfs_inactive
1288 * threads are blocked as zil_close can call zfs_inactive.
1290 if (zfsvfs->z_log) {
1291 zil_close(zfsvfs->z_log);
1292 zfsvfs->z_log = NULL;
1295 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1298 * If we are not unmounting (ie: online recv) and someone already
1299 * unmounted this file system while we were doing the switcheroo,
1300 * or a reopen of z_os failed then just bail out now.
1302 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1303 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1304 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1309 * At this point there are no vops active, and any new vops will
1310 * fail with EIO since we have z_teardown_lock for writer (only
1311 * relavent for forced unmount).
1313 * Release all holds on dbufs.
1315 mutex_enter(&zfsvfs->z_znodes_lock);
1316 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1317 zp = list_next(&zfsvfs->z_all_znodes, zp))
1319 ASSERT(ZTOV(zp)->v_count > 0);
1320 zfs_znode_dmu_fini(zp);
1322 mutex_exit(&zfsvfs->z_znodes_lock);
1325 * If we are unmounting, set the unmounted flag and let new vops
1326 * unblock. zfs_inactive will have the unmounted behavior, and all
1327 * other vops will fail with EIO.
1330 zfsvfs->z_unmounted = B_TRUE;
1331 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1332 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1336 * z_os will be NULL if there was an error in attempting to reopen
1337 * zfsvfs, so just return as the properties had already been
1338 * unregistered and cached data had been evicted before.
1340 if (zfsvfs->z_os == NULL)
1344 * Unregister properties.
1346 zfs_unregister_callbacks(zfsvfs);
1351 if (dmu_objset_is_dirty_anywhere(zfsvfs->z_os))
1352 if (!(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1353 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1354 (void) dmu_objset_evict_dbufs(zfsvfs->z_os);
1361 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1363 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1367 ret = secpolicy_fs_unmount(cr, vfsp);
1369 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1370 ZFS_DELEG_PERM_MOUNT, cr))
1375 * We purge the parent filesystem's vfsp as the parent filesystem
1376 * and all of its snapshots have their vnode's v_vfsp set to the
1377 * parent's filesystem's vfsp. Note, 'z_parent' is self
1378 * referential for non-snapshots.
1380 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1383 * Unmount any snapshots mounted under .zfs before unmounting the
1386 if (zfsvfs->z_ctldir != NULL &&
1387 (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1391 if (!(fflag & MS_FORCE)) {
1393 * Check the number of active vnodes in the file system.
1394 * Our count is maintained in the vfs structure, but the
1395 * number is off by 1 to indicate a hold on the vfs
1398 * The '.zfs' directory maintains a reference of its
1399 * own, and any active references underneath are
1400 * reflected in the vnode count.
1402 if (zfsvfs->z_ctldir == NULL) {
1403 if (vfsp->vfs_count > 1)
1406 if (vfsp->vfs_count > 2 ||
1407 zfsvfs->z_ctldir->v_count > 1)
1412 vfsp->vfs_flag |= VFS_UNMOUNTED;
1414 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1418 * z_os will be NULL if there was an error in
1419 * attempting to reopen zfsvfs.
1423 * Unset the objset user_ptr.
1425 mutex_enter(&os->os_user_ptr_lock);
1426 dmu_objset_set_user(os, NULL);
1427 mutex_exit(&os->os_user_ptr_lock);
1430 * Finally release the objset
1432 dmu_objset_disown(os, zfsvfs);
1436 * We can now safely destroy the '.zfs' directory node.
1438 if (zfsvfs->z_ctldir != NULL)
1439 zfsctl_destroy(zfsvfs);
1443 EXPORT_SYMBOL(zfs_umount);
1446 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1448 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1450 uint64_t object = 0;
1451 uint64_t fid_gen = 0;
1460 if (fidp->fid_len == LONG_FID_LEN) {
1461 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1462 uint64_t objsetid = 0;
1463 uint64_t setgen = 0;
1465 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1466 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1468 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1469 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1473 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1479 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1480 zfid_short_t *zfid = (zfid_short_t *)fidp;
1482 for (i = 0; i < sizeof (zfid->zf_object); i++)
1483 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1485 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1486 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1492 /* A zero fid_gen means we are in the .zfs control directories */
1494 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1495 *vpp = zfsvfs->z_ctldir;
1496 ASSERT(*vpp != NULL);
1497 if (object == ZFSCTL_INO_SNAPDIR) {
1498 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
1499 0, NULL, NULL, NULL, NULL, NULL) == 0);
1507 gen_mask = -1ULL >> (64 - 8 * i);
1509 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1510 if ((err = zfs_zget(zfsvfs, object, &zp))) {
1514 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1516 zp_gen = zp_gen & gen_mask;
1519 if (zp->z_unlinked || zp_gen != fid_gen) {
1520 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1528 zfs_inode_update(VTOZ(*vpp));
1533 EXPORT_SYMBOL(zfs_vget);
1536 * Block out VOPs and close zfsvfs_t::z_os
1538 * Note, if successful, then we return with the 'z_teardown_lock' and
1539 * 'z_teardown_inactive_lock' write held.
1542 zfs_suspend_fs(zfsvfs_t *zfsvfs)
1546 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
1548 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1552 EXPORT_SYMBOL(zfs_suspend_fs);
1555 * Reopen zfsvfs_t::z_os and release VOPs.
1558 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
1562 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
1563 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1565 err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zfsvfs,
1568 zfsvfs->z_os = NULL;
1571 uint64_t sa_obj = 0;
1573 err2 = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
1574 ZFS_SA_ATTRS, 8, 1, &sa_obj);
1576 if ((err || err2) && zfsvfs->z_version >= ZPL_VERSION_SA)
1580 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
1581 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
1584 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
1587 * Attempt to re-establish all the active znodes with
1588 * their dbufs. If a zfs_rezget() fails, then we'll let
1589 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
1590 * when they try to use their znode.
1592 mutex_enter(&zfsvfs->z_znodes_lock);
1593 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
1594 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1595 (void) zfs_rezget(zp);
1597 mutex_exit(&zfsvfs->z_znodes_lock);
1602 /* release the VOPs */
1603 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1604 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1608 * Since we couldn't reopen zfsvfs::z_os, force
1609 * unmount this file system.
1611 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
1612 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
1616 EXPORT_SYMBOL(zfs_resume_fs);
1619 zfs_freevfs(vfs_t *vfsp)
1621 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1623 zfsvfs_free(zfsvfs);
1625 #endif /* HAVE_ZPL */
1633 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
1645 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
1648 objset_t *os = zfsvfs->z_os;
1651 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
1654 if (newvers < zfsvfs->z_version)
1657 if (zfs_spa_version_map(newvers) >
1658 spa_version(dmu_objset_spa(zfsvfs->z_os)))
1661 tx = dmu_tx_create(os);
1662 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
1663 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1664 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
1666 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1668 error = dmu_tx_assign(tx, TXG_WAIT);
1674 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
1675 8, 1, &newvers, tx);
1682 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1685 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
1687 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1688 DMU_OT_NONE, 0, tx);
1690 error = zap_add(os, MASTER_NODE_OBJ,
1691 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1692 ASSERT3U(error, ==, 0);
1694 VERIFY(0 == sa_set_sa_object(os, sa_obj));
1695 sa_register_update_callback(os, zfs_sa_upgrade);
1698 spa_history_log_internal(LOG_DS_UPGRADE,
1699 dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu",
1700 zfsvfs->z_version, newvers, dmu_objset_id(os));
1704 zfsvfs->z_version = newvers;
1706 if (zfsvfs->z_version >= ZPL_VERSION_FUID)
1707 zfs_set_fuid_feature(zfsvfs);
1711 EXPORT_SYMBOL(zfs_set_version);
1712 #endif /* HAVE_ZPL */
1715 * Read a property stored within the master node.
1718 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
1724 * Look up the file system's value for the property. For the
1725 * version property, we look up a slightly different string.
1727 if (prop == ZFS_PROP_VERSION)
1728 pname = ZPL_VERSION_STR;
1730 pname = zfs_prop_to_name(prop);
1733 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
1735 if (error == ENOENT) {
1736 /* No value set, use the default value */
1738 case ZFS_PROP_VERSION:
1739 *value = ZPL_VERSION;
1741 case ZFS_PROP_NORMALIZE:
1742 case ZFS_PROP_UTF8ONLY:
1746 *value = ZFS_CASE_SENSITIVE;