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;
73 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
76 * Data integrity is job one. We don't want a compromised kernel
77 * writing to the storage pool, so we never sync during panic.
84 * Sync a specific filesystem.
86 zfsvfs_t *zfsvfs = vfsp->vfs_data;
90 dp = dmu_objset_pool(zfsvfs->z_os);
93 * If the system is shutting down, then skip any
94 * filesystems which may exist on a suspended pool.
96 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
101 if (zfsvfs->z_log != NULL)
102 zil_commit(zfsvfs->z_log, 0);
107 * Sync all ZFS filesystems. This is what happens when you
108 * run sync(1M). Unlike other filesystems, ZFS honors the
109 * request by waiting for all pools to commit all dirty data.
116 EXPORT_SYMBOL(zfs_sync);
119 atime_changed_cb(void *arg, uint64_t newval)
121 zfsvfs_t *zfsvfs = arg;
123 if (newval == TRUE) {
124 zfsvfs->z_atime = TRUE;
125 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
126 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
128 zfsvfs->z_atime = FALSE;
129 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
130 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
135 xattr_changed_cb(void *arg, uint64_t newval)
137 zfsvfs_t *zfsvfs = arg;
139 if (newval == TRUE) {
140 /* XXX locking on vfs_flag? */
141 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
142 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
143 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
145 /* XXX locking on vfs_flag? */
146 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
147 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
148 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
153 blksz_changed_cb(void *arg, uint64_t newval)
155 zfsvfs_t *zfsvfs = arg;
157 if (newval < SPA_MINBLOCKSIZE ||
158 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
159 newval = SPA_MAXBLOCKSIZE;
161 zfsvfs->z_max_blksz = newval;
162 zfsvfs->z_vfs->vfs_bsize = newval;
166 readonly_changed_cb(void *arg, uint64_t newval)
168 zfsvfs_t *zfsvfs = arg;
171 /* XXX locking on vfs_flag? */
172 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
173 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
174 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
176 /* XXX locking on vfs_flag? */
177 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
178 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
179 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
184 devices_changed_cb(void *arg, uint64_t newval)
186 zfsvfs_t *zfsvfs = arg;
188 if (newval == FALSE) {
189 zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
190 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
191 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
193 zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
194 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
195 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
200 setuid_changed_cb(void *arg, uint64_t newval)
202 zfsvfs_t *zfsvfs = arg;
204 if (newval == FALSE) {
205 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
206 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
207 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
209 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
210 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
211 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
216 exec_changed_cb(void *arg, uint64_t newval)
218 zfsvfs_t *zfsvfs = arg;
220 if (newval == FALSE) {
221 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
222 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
223 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
225 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
226 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
227 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
232 * The nbmand mount option can be changed at mount time.
233 * We can't allow it to be toggled on live file systems or incorrect
234 * behavior may be seen from cifs clients
236 * This property isn't registered via dsl_prop_register(), but this callback
237 * will be called when a file system is first mounted
240 nbmand_changed_cb(void *arg, uint64_t newval)
242 zfsvfs_t *zfsvfs = arg;
243 if (newval == FALSE) {
244 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
245 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
247 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
248 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
253 snapdir_changed_cb(void *arg, uint64_t newval)
255 zfsvfs_t *zfsvfs = arg;
257 zfsvfs->z_show_ctldir = newval;
261 vscan_changed_cb(void *arg, uint64_t newval)
263 zfsvfs_t *zfsvfs = arg;
265 zfsvfs->z_vscan = newval;
269 acl_inherit_changed_cb(void *arg, uint64_t newval)
271 zfsvfs_t *zfsvfs = arg;
273 zfsvfs->z_acl_inherit = newval;
277 zfs_register_callbacks(vfs_t *vfsp)
279 struct dsl_dataset *ds = NULL;
281 zfsvfs_t *zfsvfs = NULL;
283 int readonly, do_readonly = B_FALSE;
284 int setuid, do_setuid = B_FALSE;
285 int exec, do_exec = B_FALSE;
286 int devices, do_devices = B_FALSE;
287 int xattr, do_xattr = B_FALSE;
288 int atime, do_atime = B_FALSE;
292 zfsvfs = vfsp->vfs_data;
297 * The act of registering our callbacks will destroy any mount
298 * options we may have. In order to enable temporary overrides
299 * of mount options, we stash away the current values and
300 * restore them after we register the callbacks.
302 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
303 !spa_writeable(dmu_objset_spa(os))) {
305 do_readonly = B_TRUE;
306 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
308 do_readonly = B_TRUE;
310 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
316 if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
319 } else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
324 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
327 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
332 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
335 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
339 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
342 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
346 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
349 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
355 * nbmand is a special property. It can only be changed at
358 * This is weird, but it is documented to only be changeable
361 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
363 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
366 char osname[MAXNAMELEN];
368 dmu_objset_name(os, osname);
369 if ((error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
376 * Register property callbacks.
378 * It would probably be fine to just check for i/o error from
379 * the first prop_register(), but I guess I like to go
382 ds = dmu_objset_ds(os);
383 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs);
384 error = error ? error : dsl_prop_register(ds,
385 "xattr", xattr_changed_cb, zfsvfs);
386 error = error ? error : dsl_prop_register(ds,
387 "recordsize", blksz_changed_cb, zfsvfs);
388 error = error ? error : dsl_prop_register(ds,
389 "readonly", readonly_changed_cb, zfsvfs);
390 error = error ? error : dsl_prop_register(ds,
391 "devices", devices_changed_cb, zfsvfs);
392 error = error ? error : dsl_prop_register(ds,
393 "setuid", setuid_changed_cb, zfsvfs);
394 error = error ? error : dsl_prop_register(ds,
395 "exec", exec_changed_cb, zfsvfs);
396 error = error ? error : dsl_prop_register(ds,
397 "snapdir", snapdir_changed_cb, zfsvfs);
398 error = error ? error : dsl_prop_register(ds,
399 "aclinherit", acl_inherit_changed_cb, zfsvfs);
400 error = error ? error : dsl_prop_register(ds,
401 "vscan", vscan_changed_cb, zfsvfs);
406 * Invoke our callbacks to restore temporary mount options.
409 readonly_changed_cb(zfsvfs, readonly);
411 setuid_changed_cb(zfsvfs, setuid);
413 exec_changed_cb(zfsvfs, exec);
415 devices_changed_cb(zfsvfs, devices);
417 xattr_changed_cb(zfsvfs, xattr);
419 atime_changed_cb(zfsvfs, atime);
421 nbmand_changed_cb(zfsvfs, nbmand);
427 * We may attempt to unregister some callbacks that are not
428 * registered, but this is OK; it will simply return ENOMSG,
429 * which we will ignore.
431 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs);
432 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs);
433 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs);
434 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs);
435 (void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zfsvfs);
436 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs);
437 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs);
438 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs);
439 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
441 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs);
445 EXPORT_SYMBOL(zfs_register_callbacks);
446 #endif /* HAVE_ZPL */
449 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
450 uint64_t *userp, uint64_t *groupp)
452 znode_phys_t *znp = data;
456 * Is it a valid type of object to track?
458 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
462 * If we have a NULL data pointer
463 * then assume the id's aren't changing and
464 * return EEXIST to the dmu to let it know to
470 if (bonustype == DMU_OT_ZNODE) {
471 *userp = znp->zp_uid;
472 *groupp = znp->zp_gid;
476 ASSERT(bonustype == DMU_OT_SA);
477 hdrsize = sa_hdrsize(data);
480 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
482 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
486 * This should only happen for newly created
487 * files that haven't had the znode data filled
499 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
500 char *domainbuf, int buflen, uid_t *ridp)
505 fuid = strtonum(fuidstr, NULL);
507 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
509 (void) strlcpy(domainbuf, domain, buflen);
512 *ridp = FUID_RID(fuid);
516 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
519 case ZFS_PROP_USERUSED:
520 return (DMU_USERUSED_OBJECT);
521 case ZFS_PROP_GROUPUSED:
522 return (DMU_GROUPUSED_OBJECT);
523 case ZFS_PROP_USERQUOTA:
524 return (zfsvfs->z_userquota_obj);
525 case ZFS_PROP_GROUPQUOTA:
526 return (zfsvfs->z_groupquota_obj);
534 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
535 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
540 zfs_useracct_t *buf = vbuf;
543 if (!dmu_objset_userspace_present(zfsvfs->z_os))
546 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
552 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
553 (error = zap_cursor_retrieve(&zc, &za)) == 0;
554 zap_cursor_advance(&zc)) {
555 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
559 fuidstr_to_sid(zfsvfs, za.za_name,
560 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
562 buf->zu_space = za.za_first_integer;
568 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
569 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
570 *cookiep = zap_cursor_serialize(&zc);
571 zap_cursor_fini(&zc);
574 EXPORT_SYMBOL(zfs_userspace_many);
577 * buf must be big enough (eg, 32 bytes)
580 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
581 char *buf, boolean_t addok)
586 if (domain && domain[0]) {
587 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
591 fuid = FUID_ENCODE(domainid, rid);
592 (void) sprintf(buf, "%llx", (longlong_t)fuid);
597 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
598 const char *domain, uint64_t rid, uint64_t *valp)
606 if (!dmu_objset_userspace_present(zfsvfs->z_os))
609 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
613 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
617 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
622 EXPORT_SYMBOL(zfs_userspace_one);
625 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
626 const char *domain, uint64_t rid, uint64_t quota)
632 boolean_t fuid_dirtied;
634 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
637 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
640 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
641 &zfsvfs->z_groupquota_obj;
643 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
646 fuid_dirtied = zfsvfs->z_fuid_dirty;
648 tx = dmu_tx_create(zfsvfs->z_os);
649 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
651 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
652 zfs_userquota_prop_prefixes[type]);
655 zfs_fuid_txhold(zfsvfs, tx);
656 err = dmu_tx_assign(tx, TXG_WAIT);
662 mutex_enter(&zfsvfs->z_lock);
664 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
666 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
667 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
669 mutex_exit(&zfsvfs->z_lock);
672 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
676 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
680 zfs_fuid_sync(zfsvfs, tx);
684 EXPORT_SYMBOL(zfs_set_userquota);
687 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
690 uint64_t used, quota, usedobj, quotaobj;
693 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
694 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
696 if (quotaobj == 0 || zfsvfs->z_replay)
699 (void) sprintf(buf, "%llx", (longlong_t)fuid);
700 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
704 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
707 return (used >= quota);
709 EXPORT_SYMBOL(zfs_fuid_overquota);
712 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
717 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
719 fuid = isgroup ? zp->z_gid : zp->z_uid;
721 if (quotaobj == 0 || zfsvfs->z_replay)
724 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
726 EXPORT_SYMBOL(zfs_owner_overquota);
729 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
737 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
740 * We claim to always be readonly so we can open snapshots;
741 * other ZPL code will prevent us from writing to snapshots.
743 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
745 kmem_free(zfsvfs, sizeof (zfsvfs_t));
750 * Initialize the zfs-specific filesystem structure.
751 * Should probably make this a kmem cache, shuffle fields,
752 * and just bzero up to z_hold_mtx[].
754 zfsvfs->z_vfs = NULL;
755 zfsvfs->z_parent = zfsvfs;
756 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
757 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
760 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
763 } else if (zfsvfs->z_version >
764 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
765 (void) printk("Can't mount a version %lld file system "
766 "on a version %lld pool\n. Pool must be upgraded to mount "
767 "this file system.", (u_longlong_t)zfsvfs->z_version,
768 (u_longlong_t)spa_version(dmu_objset_spa(os)));
772 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
774 zfsvfs->z_norm = (int)zval;
776 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
778 zfsvfs->z_utf8 = (zval != 0);
780 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
782 zfsvfs->z_case = (uint_t)zval;
785 * Fold case on file systems that are always or sometimes case
788 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
789 zfsvfs->z_case == ZFS_CASE_MIXED)
790 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
792 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
793 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
795 if (zfsvfs->z_use_sa) {
796 /* should either have both of these objects or none */
797 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
803 * Pre SA versions file systems should never touch
804 * either the attribute registration or layout objects.
809 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
810 &zfsvfs->z_attr_table);
814 if (zfsvfs->z_version >= ZPL_VERSION_SA)
815 sa_register_update_callback(os, zfs_sa_upgrade);
817 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
821 ASSERT(zfsvfs->z_root != 0);
823 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
824 &zfsvfs->z_unlinkedobj);
828 error = zap_lookup(os, MASTER_NODE_OBJ,
829 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
830 8, 1, &zfsvfs->z_userquota_obj);
831 if (error && error != ENOENT)
834 error = zap_lookup(os, MASTER_NODE_OBJ,
835 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
836 8, 1, &zfsvfs->z_groupquota_obj);
837 if (error && error != ENOENT)
840 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
841 &zfsvfs->z_fuid_obj);
842 if (error && error != ENOENT)
845 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
846 &zfsvfs->z_shares_dir);
847 if (error && error != ENOENT)
850 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
851 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
852 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
853 offsetof(znode_t, z_link_node));
854 rrw_init(&zfsvfs->z_teardown_lock);
855 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
856 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
857 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
858 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
864 dmu_objset_disown(os, zfsvfs);
866 kmem_free(zfsvfs, sizeof (zfsvfs_t));
871 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
875 error = zfs_register_callbacks(zfsvfs->z_vfs);
880 * Set the objset user_ptr to track its zfsvfs.
882 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
883 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
884 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
886 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
889 * If we are not mounting (ie: online recv), then we don't
890 * have to worry about replaying the log as we blocked all
891 * operations out since we closed the ZIL.
897 * During replay we remove the read only flag to
898 * allow replays to succeed.
900 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
902 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
904 zfs_unlinked_drain(zfsvfs);
907 * Parse and replay the intent log.
909 * Because of ziltest, this must be done after
910 * zfs_unlinked_drain(). (Further note: ziltest
911 * doesn't use readonly mounts, where
912 * zfs_unlinked_drain() isn't called.) This is because
913 * ziltest causes spa_sync() to think it's committed,
914 * but actually it is not, so the intent log contains
915 * many txg's worth of changes.
917 * In particular, if object N is in the unlinked set in
918 * the last txg to actually sync, then it could be
919 * actually freed in a later txg and then reallocated
920 * in a yet later txg. This would write a "create
921 * object N" record to the intent log. Normally, this
922 * would be fine because the spa_sync() would have
923 * written out the fact that object N is free, before
924 * we could write the "create object N" intent log
927 * But when we are in ziltest mode, we advance the "open
928 * txg" without actually spa_sync()-ing the changes to
929 * disk. So we would see that object N is still
930 * allocated and in the unlinked set, and there is an
931 * intent log record saying to allocate it.
933 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
934 if (zil_replay_disable) {
935 zil_destroy(zfsvfs->z_log, B_FALSE);
937 zfsvfs->z_replay = B_TRUE;
938 zil_replay(zfsvfs->z_os, zfsvfs,
940 zfsvfs->z_replay = B_FALSE;
943 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
950 zfsvfs_free(zfsvfs_t *zfsvfs)
953 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
956 * This is a barrier to prevent the filesystem from going away in
957 * zfs_znode_move() until we can safely ensure that the filesystem is
958 * not unmounted. We consider the filesystem valid before the barrier
959 * and invalid after the barrier.
961 rw_enter(&zfsvfs_lock, RW_READER);
962 rw_exit(&zfsvfs_lock);
964 zfs_fuid_destroy(zfsvfs);
966 mutex_destroy(&zfsvfs->z_znodes_lock);
967 mutex_destroy(&zfsvfs->z_lock);
968 list_destroy(&zfsvfs->z_all_znodes);
969 rrw_destroy(&zfsvfs->z_teardown_lock);
970 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
971 rw_destroy(&zfsvfs->z_fuid_lock);
972 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
973 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
974 kmem_free(zfsvfs, sizeof (zfsvfs_t));
978 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
980 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
981 if (zfsvfs->z_use_fuids && zfsvfs->z_vfs) {
982 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
983 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
984 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
985 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
986 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
987 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
989 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
993 zfs_domount(vfs_t *vfsp, char *osname)
995 uint64_t recordsize, fsid_guid;
1002 error = zfsvfs_create(osname, &zfsvfs);
1005 zfsvfs->z_vfs = vfsp;
1007 /* Initialize the generic filesystem structure. */
1008 vfsp->vfs_bcount = 0;
1009 vfsp->vfs_data = NULL;
1011 if ((error = dsl_prop_get_integer(osname, "recordsize",
1012 &recordsize, NULL)))
1015 vfsp->vfs_bsize = recordsize;
1016 vfsp->vfs_flag |= VFS_NOTRUNC;
1017 vfsp->vfs_data = zfsvfs;
1020 * The fsid is 64 bits, composed of an 8-bit fs type, which
1021 * separates our fsid from any other filesystem types, and a
1022 * 56-bit objset unique ID. The objset unique ID is unique to
1023 * all objsets open on this system, provided by unique_create().
1024 * The 8-bit fs type must be put in the low bits of fsid[1]
1025 * because that's where other Solaris filesystems put it.
1027 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1028 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1029 vfsp->vfs_fsid.val[0] = fsid_guid;
1030 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8);
1033 * Set features for file system.
1035 zfs_set_fuid_feature(zfsvfs);
1036 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1037 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1038 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1039 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1040 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1041 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1042 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1044 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1046 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1049 atime_changed_cb(zfsvfs, B_FALSE);
1050 readonly_changed_cb(zfsvfs, B_TRUE);
1051 if ((error = dsl_prop_get_integer(osname,"xattr",&pval,NULL)))
1053 xattr_changed_cb(zfsvfs, pval);
1054 zfsvfs->z_issnap = B_TRUE;
1055 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1057 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1058 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1059 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1061 error = zfsvfs_setup(zfsvfs, B_TRUE);
1064 if (!zfsvfs->z_issnap)
1065 zfsctl_create(zfsvfs);
1068 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1069 zfsvfs_free(zfsvfs);
1074 EXPORT_SYMBOL(zfs_domount);
1077 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1079 objset_t *os = zfsvfs->z_os;
1080 struct dsl_dataset *ds;
1083 * Unregister properties.
1085 if (!dmu_objset_is_snapshot(os)) {
1086 ds = dmu_objset_ds(os);
1087 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1090 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1093 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1096 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1099 VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1102 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1105 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1108 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1111 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1112 acl_inherit_changed_cb, zfsvfs) == 0);
1114 VERIFY(dsl_prop_unregister(ds, "vscan",
1115 vscan_changed_cb, zfsvfs) == 0);
1118 EXPORT_SYMBOL(zfs_unregister_callbacks);
1120 #ifdef HAVE_MLSLABEL
1122 * zfs_check_global_label:
1123 * Check that the hex label string is appropriate for the dataset
1124 * being mounted into the global_zone proper.
1126 * Return an error if the hex label string is not default or
1127 * admin_low/admin_high. For admin_low labels, the corresponding
1128 * dataset must be readonly.
1131 zfs_check_global_label(const char *dsname, const char *hexsl)
1133 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1135 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1137 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1138 /* must be readonly */
1141 if (dsl_prop_get_integer(dsname,
1142 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1144 return (rdonly ? 0 : EACCES);
1148 #endif /* HAVE_MLSLABEL */
1151 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1153 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1155 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1159 dmu_objset_space(zfsvfs->z_os,
1160 &refdbytes, &availbytes, &usedobjs, &availobjs);
1163 * The underlying storage pool actually uses multiple block sizes.
1164 * We report the fragsize as the smallest block size we support,
1165 * and we report our blocksize as the filesystem's maximum blocksize.
1167 statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1168 statp->f_bsize = zfsvfs->z_max_blksz;
1171 * The following report "total" blocks of various kinds in the
1172 * file system, but reported in terms of f_frsize - the
1176 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1177 statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1178 statp->f_bavail = statp->f_bfree; /* no root reservation */
1181 * statvfs() should really be called statufs(), because it assumes
1182 * static metadata. ZFS doesn't preallocate files, so the best
1183 * we can do is report the max that could possibly fit in f_files,
1184 * and that minus the number actually used in f_ffree.
1185 * For f_ffree, report the smaller of the number of object available
1186 * and the number of blocks (each object will take at least a block).
1188 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1189 statp->f_favail = statp->f_ffree; /* no "root reservation" */
1190 statp->f_files = statp->f_ffree + usedobjs;
1192 (void) cmpldev(&d32, vfsp->vfs_dev);
1193 statp->f_fsid = d32;
1196 * We're a zfs filesystem.
1198 (void) strcpy(statp->f_basetype, MNTTYPE_ZFS);
1200 statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1202 statp->f_namemax = ZFS_MAXNAMELEN;
1205 * We have all of 32 characters to stuff a string here.
1206 * Is there anything useful we could/should provide?
1208 bzero(statp->f_fstr, sizeof (statp->f_fstr));
1213 EXPORT_SYMBOL(zfs_statvfs);
1216 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1218 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1224 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1226 *vpp = ZTOV(rootzp);
1231 EXPORT_SYMBOL(zfs_root);
1234 * Teardown the zfsvfs::z_os.
1236 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1237 * and 'z_teardown_inactive_lock' held.
1240 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1244 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1248 * We purge the parent filesystem's vfsp as the parent
1249 * filesystem and all of its snapshots have their vnode's
1250 * v_vfsp set to the parent's filesystem's vfsp. Note,
1251 * 'z_parent' is self referential for non-snapshots.
1253 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1257 * Close the zil. NB: Can't close the zil while zfs_inactive
1258 * threads are blocked as zil_close can call zfs_inactive.
1260 if (zfsvfs->z_log) {
1261 zil_close(zfsvfs->z_log);
1262 zfsvfs->z_log = NULL;
1265 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1268 * If we are not unmounting (ie: online recv) and someone already
1269 * unmounted this file system while we were doing the switcheroo,
1270 * or a reopen of z_os failed then just bail out now.
1272 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1273 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1274 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1279 * At this point there are no vops active, and any new vops will
1280 * fail with EIO since we have z_teardown_lock for writer (only
1281 * relavent for forced unmount).
1283 * Release all holds on dbufs.
1285 mutex_enter(&zfsvfs->z_znodes_lock);
1286 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1287 zp = list_next(&zfsvfs->z_all_znodes, zp))
1289 ASSERT(ZTOV(zp)->v_count > 0);
1290 zfs_znode_dmu_fini(zp);
1292 mutex_exit(&zfsvfs->z_znodes_lock);
1295 * If we are unmounting, set the unmounted flag and let new vops
1296 * unblock. zfs_inactive will have the unmounted behavior, and all
1297 * other vops will fail with EIO.
1300 zfsvfs->z_unmounted = B_TRUE;
1301 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1302 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1306 * z_os will be NULL if there was an error in attempting to reopen
1307 * zfsvfs, so just return as the properties had already been
1308 * unregistered and cached data had been evicted before.
1310 if (zfsvfs->z_os == NULL)
1314 * Unregister properties.
1316 zfs_unregister_callbacks(zfsvfs);
1321 if (dmu_objset_is_dirty_anywhere(zfsvfs->z_os))
1322 if (!(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1323 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1324 (void) dmu_objset_evict_dbufs(zfsvfs->z_os);
1331 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1333 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1337 ret = secpolicy_fs_unmount(cr, vfsp);
1339 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1340 ZFS_DELEG_PERM_MOUNT, cr))
1345 * We purge the parent filesystem's vfsp as the parent filesystem
1346 * and all of its snapshots have their vnode's v_vfsp set to the
1347 * parent's filesystem's vfsp. Note, 'z_parent' is self
1348 * referential for non-snapshots.
1350 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1353 * Unmount any snapshots mounted under .zfs before unmounting the
1356 if (zfsvfs->z_ctldir != NULL &&
1357 (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1361 if (!(fflag & MS_FORCE)) {
1363 * Check the number of active vnodes in the file system.
1364 * Our count is maintained in the vfs structure, but the
1365 * number is off by 1 to indicate a hold on the vfs
1368 * The '.zfs' directory maintains a reference of its
1369 * own, and any active references underneath are
1370 * reflected in the vnode count.
1372 if (zfsvfs->z_ctldir == NULL) {
1373 if (vfsp->vfs_count > 1)
1376 if (vfsp->vfs_count > 2 ||
1377 zfsvfs->z_ctldir->v_count > 1)
1382 vfsp->vfs_flag |= VFS_UNMOUNTED;
1384 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1388 * z_os will be NULL if there was an error in
1389 * attempting to reopen zfsvfs.
1393 * Unset the objset user_ptr.
1395 mutex_enter(&os->os_user_ptr_lock);
1396 dmu_objset_set_user(os, NULL);
1397 mutex_exit(&os->os_user_ptr_lock);
1400 * Finally release the objset
1402 dmu_objset_disown(os, zfsvfs);
1406 * We can now safely destroy the '.zfs' directory node.
1408 if (zfsvfs->z_ctldir != NULL)
1409 zfsctl_destroy(zfsvfs);
1413 EXPORT_SYMBOL(zfs_umount);
1416 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1418 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1420 uint64_t object = 0;
1421 uint64_t fid_gen = 0;
1430 if (fidp->fid_len == LONG_FID_LEN) {
1431 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1432 uint64_t objsetid = 0;
1433 uint64_t setgen = 0;
1435 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1436 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1438 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1439 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1443 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1449 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1450 zfid_short_t *zfid = (zfid_short_t *)fidp;
1452 for (i = 0; i < sizeof (zfid->zf_object); i++)
1453 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1455 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1456 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1462 /* A zero fid_gen means we are in the .zfs control directories */
1464 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1465 *vpp = zfsvfs->z_ctldir;
1466 ASSERT(*vpp != NULL);
1467 if (object == ZFSCTL_INO_SNAPDIR) {
1468 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
1469 0, NULL, NULL, NULL, NULL, NULL) == 0);
1477 gen_mask = -1ULL >> (64 - 8 * i);
1479 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1480 if ((err = zfs_zget(zfsvfs, object, &zp))) {
1484 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1486 zp_gen = zp_gen & gen_mask;
1489 if (zp->z_unlinked || zp_gen != fid_gen) {
1490 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1498 zfs_inode_update(VTOZ(*vpp));
1503 EXPORT_SYMBOL(zfs_vget);
1506 * Block out VOPs and close zfsvfs_t::z_os
1508 * Note, if successful, then we return with the 'z_teardown_lock' and
1509 * 'z_teardown_inactive_lock' write held.
1512 zfs_suspend_fs(zfsvfs_t *zfsvfs)
1516 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
1518 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1522 EXPORT_SYMBOL(zfs_suspend_fs);
1525 * Reopen zfsvfs_t::z_os and release VOPs.
1528 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
1532 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
1533 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1535 err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zfsvfs,
1538 zfsvfs->z_os = NULL;
1541 uint64_t sa_obj = 0;
1543 err2 = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
1544 ZFS_SA_ATTRS, 8, 1, &sa_obj);
1546 if ((err || err2) && zfsvfs->z_version >= ZPL_VERSION_SA)
1550 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
1551 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
1554 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
1557 * Attempt to re-establish all the active znodes with
1558 * their dbufs. If a zfs_rezget() fails, then we'll let
1559 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
1560 * when they try to use their znode.
1562 mutex_enter(&zfsvfs->z_znodes_lock);
1563 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
1564 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1565 (void) zfs_rezget(zp);
1567 mutex_exit(&zfsvfs->z_znodes_lock);
1572 /* release the VOPs */
1573 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1574 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1578 * Since we couldn't reopen zfsvfs::z_os, force
1579 * unmount this file system.
1581 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
1582 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
1586 EXPORT_SYMBOL(zfs_resume_fs);
1589 zfs_freevfs(vfs_t *vfsp)
1591 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1593 zfsvfs_free(zfsvfs);
1595 #endif /* HAVE_ZPL */
1603 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
1615 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
1618 objset_t *os = zfsvfs->z_os;
1621 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
1624 if (newvers < zfsvfs->z_version)
1627 if (zfs_spa_version_map(newvers) >
1628 spa_version(dmu_objset_spa(zfsvfs->z_os)))
1631 tx = dmu_tx_create(os);
1632 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
1633 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1634 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
1636 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1638 error = dmu_tx_assign(tx, TXG_WAIT);
1644 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
1645 8, 1, &newvers, tx);
1652 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1655 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
1657 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1658 DMU_OT_NONE, 0, tx);
1660 error = zap_add(os, MASTER_NODE_OBJ,
1661 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1662 ASSERT3U(error, ==, 0);
1664 VERIFY(0 == sa_set_sa_object(os, sa_obj));
1665 sa_register_update_callback(os, zfs_sa_upgrade);
1668 spa_history_log_internal(LOG_DS_UPGRADE,
1669 dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu",
1670 zfsvfs->z_version, newvers, dmu_objset_id(os));
1674 zfsvfs->z_version = newvers;
1676 if (zfsvfs->z_version >= ZPL_VERSION_FUID)
1677 zfs_set_fuid_feature(zfsvfs);
1681 EXPORT_SYMBOL(zfs_set_version);
1682 #endif /* HAVE_ZPL */
1685 * Read a property stored within the master node.
1688 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
1694 * Look up the file system's value for the property. For the
1695 * version property, we look up a slightly different string.
1697 if (prop == ZFS_PROP_VERSION)
1698 pname = ZPL_VERSION_STR;
1700 pname = zfs_prop_to_name(prop);
1703 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
1705 if (error == ENOENT) {
1706 /* No value set, use the default value */
1708 case ZFS_PROP_VERSION:
1709 *value = ZPL_VERSION;
1711 case ZFS_PROP_NORMALIZE:
1712 case ZFS_PROP_UTF8ONLY:
1716 *value = ZFS_CASE_SENSITIVE;