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_vnops.h>
42 #include <sys/zfs_dir.h>
44 #include <sys/fs/zfs.h>
46 #include <sys/dsl_prop.h>
47 #include <sys/dsl_dataset.h>
48 #include <sys/dsl_deleg.h>
52 #include <sys/varargs.h>
53 #include <sys/policy.h>
54 #include <sys/atomic.h>
55 #include <sys/mkdev.h>
56 #include <sys/modctl.h>
57 #include <sys/refstr.h>
58 #include <sys/zfs_ioctl.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>
67 #include "zfs_comutil.h"
72 zfs_sync(struct super_block
*sb
, int wait
, cred_t
*cr
)
74 zfs_sb_t
*zsb
= sb
->s_fs_info
;
77 * Data integrity is job one. We don't want a compromised kernel
78 * writing to the storage pool, so we never sync during panic.
80 if (unlikely(oops_in_progress
))
84 * Semantically, the only requirement is that the sync be initiated.
85 * The DMU syncs out txgs frequently, so there's nothing to do.
92 * Sync a specific filesystem.
97 dp
= dmu_objset_pool(zsb
->z_os
);
100 * If the system is shutting down, then skip any
101 * filesystems which may exist on a suspended pool.
103 if (spa_suspended(dp
->dp_spa
)) {
108 if (zsb
->z_log
!= NULL
)
109 zil_commit(zsb
->z_log
, 0);
114 * Sync all ZFS filesystems. This is what happens when you
115 * run sync(1M). Unlike other filesystems, ZFS honors the
116 * request by waiting for all pools to commit all dirty data.
123 EXPORT_SYMBOL(zfs_sync
);
126 zfs_is_readonly(zfs_sb_t
*zsb
)
128 return (!!(zsb
->z_sb
->s_flags
& MS_RDONLY
));
130 EXPORT_SYMBOL(zfs_is_readonly
);
133 atime_changed_cb(void *arg
, uint64_t newval
)
135 ((zfs_sb_t
*)arg
)->z_atime
= newval
;
139 xattr_changed_cb(void *arg
, uint64_t newval
)
144 zsb
->z_flags
|= ZSB_XATTR
;
146 zsb
->z_flags
&= ~ZSB_XATTR
;
150 blksz_changed_cb(void *arg
, uint64_t newval
)
154 if (newval
< SPA_MINBLOCKSIZE
||
155 newval
> SPA_MAXBLOCKSIZE
|| !ISP2(newval
))
156 newval
= SPA_MAXBLOCKSIZE
;
158 zsb
->z_max_blksz
= newval
;
162 readonly_changed_cb(void *arg
, uint64_t newval
)
165 struct super_block
*sb
= zsb
->z_sb
;
171 sb
->s_flags
|= MS_RDONLY
;
173 sb
->s_flags
&= ~MS_RDONLY
;
177 devices_changed_cb(void *arg
, uint64_t newval
)
182 setuid_changed_cb(void *arg
, uint64_t newval
)
187 exec_changed_cb(void *arg
, uint64_t newval
)
192 nbmand_changed_cb(void *arg
, uint64_t newval
)
195 struct super_block
*sb
= zsb
->z_sb
;
201 sb
->s_flags
|= MS_MANDLOCK
;
203 sb
->s_flags
&= ~MS_MANDLOCK
;
207 snapdir_changed_cb(void *arg
, uint64_t newval
)
209 ((zfs_sb_t
*)arg
)->z_show_ctldir
= newval
;
213 vscan_changed_cb(void *arg
, uint64_t newval
)
215 ((zfs_sb_t
*)arg
)->z_vscan
= newval
;
219 acl_inherit_changed_cb(void *arg
, uint64_t newval
)
221 ((zfs_sb_t
*)arg
)->z_acl_inherit
= newval
;
225 zfs_register_callbacks(zfs_sb_t
*zsb
)
227 struct dsl_dataset
*ds
= NULL
;
228 objset_t
*os
= zsb
->z_os
;
231 if (zfs_is_readonly(zsb
) || !spa_writeable(dmu_objset_spa(os
)))
232 readonly_changed_cb(zsb
, B_TRUE
);
235 * Register property callbacks.
237 * It would probably be fine to just check for i/o error from
238 * the first prop_register(), but I guess I like to go
241 ds
= dmu_objset_ds(os
);
242 error
= dsl_prop_register(ds
,
243 "atime", atime_changed_cb
, zsb
);
244 error
= error
? error
: dsl_prop_register(ds
,
245 "xattr", xattr_changed_cb
, zsb
);
246 error
= error
? error
: dsl_prop_register(ds
,
247 "recordsize", blksz_changed_cb
, zsb
);
248 error
= error
? error
: dsl_prop_register(ds
,
249 "readonly", readonly_changed_cb
, zsb
);
250 error
= error
? error
: dsl_prop_register(ds
,
251 "devices", devices_changed_cb
, zsb
);
252 error
= error
? error
: dsl_prop_register(ds
,
253 "setuid", setuid_changed_cb
, zsb
);
254 error
= error
? error
: dsl_prop_register(ds
,
255 "exec", exec_changed_cb
, zsb
);
256 error
= error
? error
: dsl_prop_register(ds
,
257 "snapdir", snapdir_changed_cb
, zsb
);
258 error
= error
? error
: dsl_prop_register(ds
,
259 "aclinherit", acl_inherit_changed_cb
, zsb
);
260 error
= error
? error
: dsl_prop_register(ds
,
261 "vscan", vscan_changed_cb
, zsb
);
262 error
= error
? error
: dsl_prop_register(ds
,
263 "nbmand", nbmand_changed_cb
, zsb
);
271 * We may attempt to unregister some callbacks that are not
272 * registered, but this is OK; it will simply return ENOMSG,
273 * which we will ignore.
275 (void) dsl_prop_unregister(ds
, "atime", atime_changed_cb
, zsb
);
276 (void) dsl_prop_unregister(ds
, "xattr", xattr_changed_cb
, zsb
);
277 (void) dsl_prop_unregister(ds
, "recordsize", blksz_changed_cb
, zsb
);
278 (void) dsl_prop_unregister(ds
, "readonly", readonly_changed_cb
, zsb
);
279 (void) dsl_prop_unregister(ds
, "devices", devices_changed_cb
, zsb
);
280 (void) dsl_prop_unregister(ds
, "setuid", setuid_changed_cb
, zsb
);
281 (void) dsl_prop_unregister(ds
, "exec", exec_changed_cb
, zsb
);
282 (void) dsl_prop_unregister(ds
, "snapdir", snapdir_changed_cb
, zsb
);
283 (void) dsl_prop_unregister(ds
, "aclinherit", acl_inherit_changed_cb
,
285 (void) dsl_prop_unregister(ds
, "vscan", vscan_changed_cb
, zsb
);
286 (void) dsl_prop_unregister(ds
, "nbmand", nbmand_changed_cb
, zsb
);
290 EXPORT_SYMBOL(zfs_register_callbacks
);
293 zfs_space_delta_cb(dmu_object_type_t bonustype
, void *data
,
294 uint64_t *userp
, uint64_t *groupp
)
296 znode_phys_t
*znp
= data
;
300 * Is it a valid type of object to track?
302 if (bonustype
!= DMU_OT_ZNODE
&& bonustype
!= DMU_OT_SA
)
306 * If we have a NULL data pointer
307 * then assume the id's aren't changing and
308 * return EEXIST to the dmu to let it know to
314 if (bonustype
== DMU_OT_ZNODE
) {
315 *userp
= znp
->zp_uid
;
316 *groupp
= znp
->zp_gid
;
320 ASSERT(bonustype
== DMU_OT_SA
);
321 hdrsize
= sa_hdrsize(data
);
324 *userp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+
326 *groupp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+
330 * This should only happen for newly created
331 * files that haven't had the znode data filled
342 fuidstr_to_sid(zfs_sb_t
*zsb
, const char *fuidstr
,
343 char *domainbuf
, int buflen
, uid_t
*ridp
)
348 fuid
= strtonum(fuidstr
, NULL
);
350 domain
= zfs_fuid_find_by_idx(zsb
, FUID_INDEX(fuid
));
352 (void) strlcpy(domainbuf
, domain
, buflen
);
355 *ridp
= FUID_RID(fuid
);
359 zfs_userquota_prop_to_obj(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
)
362 case ZFS_PROP_USERUSED
:
363 return (DMU_USERUSED_OBJECT
);
364 case ZFS_PROP_GROUPUSED
:
365 return (DMU_GROUPUSED_OBJECT
);
366 case ZFS_PROP_USERQUOTA
:
367 return (zsb
->z_userquota_obj
);
368 case ZFS_PROP_GROUPQUOTA
:
369 return (zsb
->z_groupquota_obj
);
377 zfs_userspace_many(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
378 uint64_t *cookiep
, void *vbuf
, uint64_t *bufsizep
)
383 zfs_useracct_t
*buf
= vbuf
;
386 if (!dmu_objset_userspace_present(zsb
->z_os
))
389 obj
= zfs_userquota_prop_to_obj(zsb
, type
);
395 for (zap_cursor_init_serialized(&zc
, zsb
->z_os
, obj
, *cookiep
);
396 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
397 zap_cursor_advance(&zc
)) {
398 if ((uintptr_t)buf
- (uintptr_t)vbuf
+ sizeof (zfs_useracct_t
) >
402 fuidstr_to_sid(zsb
, za
.za_name
,
403 buf
->zu_domain
, sizeof (buf
->zu_domain
), &buf
->zu_rid
);
405 buf
->zu_space
= za
.za_first_integer
;
411 ASSERT3U((uintptr_t)buf
- (uintptr_t)vbuf
, <=, *bufsizep
);
412 *bufsizep
= (uintptr_t)buf
- (uintptr_t)vbuf
;
413 *cookiep
= zap_cursor_serialize(&zc
);
414 zap_cursor_fini(&zc
);
417 EXPORT_SYMBOL(zfs_userspace_many
);
420 * buf must be big enough (eg, 32 bytes)
423 id_to_fuidstr(zfs_sb_t
*zsb
, const char *domain
, uid_t rid
,
424 char *buf
, boolean_t addok
)
429 if (domain
&& domain
[0]) {
430 domainid
= zfs_fuid_find_by_domain(zsb
, domain
, NULL
, addok
);
434 fuid
= FUID_ENCODE(domainid
, rid
);
435 (void) sprintf(buf
, "%llx", (longlong_t
)fuid
);
440 zfs_userspace_one(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
441 const char *domain
, uint64_t rid
, uint64_t *valp
)
449 if (!dmu_objset_userspace_present(zsb
->z_os
))
452 obj
= zfs_userquota_prop_to_obj(zsb
, type
);
456 err
= id_to_fuidstr(zsb
, domain
, rid
, buf
, B_FALSE
);
460 err
= zap_lookup(zsb
->z_os
, obj
, buf
, 8, 1, valp
);
465 EXPORT_SYMBOL(zfs_userspace_one
);
468 zfs_set_userquota(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
469 const char *domain
, uint64_t rid
, uint64_t quota
)
475 boolean_t fuid_dirtied
;
477 if (type
!= ZFS_PROP_USERQUOTA
&& type
!= ZFS_PROP_GROUPQUOTA
)
480 if (zsb
->z_version
< ZPL_VERSION_USERSPACE
)
483 objp
= (type
== ZFS_PROP_USERQUOTA
) ? &zsb
->z_userquota_obj
:
484 &zsb
->z_groupquota_obj
;
486 err
= id_to_fuidstr(zsb
, domain
, rid
, buf
, B_TRUE
);
489 fuid_dirtied
= zsb
->z_fuid_dirty
;
491 tx
= dmu_tx_create(zsb
->z_os
);
492 dmu_tx_hold_zap(tx
, *objp
? *objp
: DMU_NEW_OBJECT
, B_TRUE
, NULL
);
494 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
495 zfs_userquota_prop_prefixes
[type
]);
498 zfs_fuid_txhold(zsb
, tx
);
499 err
= dmu_tx_assign(tx
, TXG_WAIT
);
505 mutex_enter(&zsb
->z_lock
);
507 *objp
= zap_create(zsb
->z_os
, DMU_OT_USERGROUP_QUOTA
,
509 VERIFY(0 == zap_add(zsb
->z_os
, MASTER_NODE_OBJ
,
510 zfs_userquota_prop_prefixes
[type
], 8, 1, objp
, tx
));
512 mutex_exit(&zsb
->z_lock
);
515 err
= zap_remove(zsb
->z_os
, *objp
, buf
, tx
);
519 err
= zap_update(zsb
->z_os
, *objp
, buf
, 8, 1, "a
, tx
);
523 zfs_fuid_sync(zsb
, tx
);
527 EXPORT_SYMBOL(zfs_set_userquota
);
530 zfs_fuid_overquota(zfs_sb_t
*zsb
, boolean_t isgroup
, uint64_t fuid
)
533 uint64_t used
, quota
, usedobj
, quotaobj
;
536 usedobj
= isgroup
? DMU_GROUPUSED_OBJECT
: DMU_USERUSED_OBJECT
;
537 quotaobj
= isgroup
? zsb
->z_groupquota_obj
: zsb
->z_userquota_obj
;
539 if (quotaobj
== 0 || zsb
->z_replay
)
542 (void) sprintf(buf
, "%llx", (longlong_t
)fuid
);
543 err
= zap_lookup(zsb
->z_os
, quotaobj
, buf
, 8, 1, "a
);
547 err
= zap_lookup(zsb
->z_os
, usedobj
, buf
, 8, 1, &used
);
550 return (used
>= quota
);
552 EXPORT_SYMBOL(zfs_fuid_overquota
);
555 zfs_owner_overquota(zfs_sb_t
*zsb
, znode_t
*zp
, boolean_t isgroup
)
560 quotaobj
= isgroup
? zsb
->z_groupquota_obj
: zsb
->z_userquota_obj
;
562 fuid
= isgroup
? zp
->z_gid
: zp
->z_uid
;
564 if (quotaobj
== 0 || zsb
->z_replay
)
567 return (zfs_fuid_overquota(zsb
, isgroup
, fuid
));
569 EXPORT_SYMBOL(zfs_owner_overquota
);
572 zfs_sb_create(const char *osname
, zfs_sb_t
**zsbp
)
580 zsb
= kmem_zalloc(sizeof (zfs_sb_t
), KM_SLEEP
);
583 * We claim to always be readonly so we can open snapshots;
584 * other ZPL code will prevent us from writing to snapshots.
586 error
= dmu_objset_own(osname
, DMU_OST_ZFS
, B_TRUE
, zsb
, &os
);
588 kmem_free(zsb
, sizeof (zfs_sb_t
));
593 * Initialize the zfs-specific filesystem structure.
594 * Should probably make this a kmem cache, shuffle fields,
595 * and just bzero up to z_hold_mtx[].
599 zsb
->z_max_blksz
= SPA_MAXBLOCKSIZE
;
600 zsb
->z_show_ctldir
= ZFS_SNAPDIR_VISIBLE
;
603 error
= zfs_get_zplprop(os
, ZFS_PROP_VERSION
, &zsb
->z_version
);
606 } else if (zsb
->z_version
>
607 zfs_zpl_version_map(spa_version(dmu_objset_spa(os
)))) {
608 (void) printk("Can't mount a version %lld file system "
609 "on a version %lld pool\n. Pool must be upgraded to mount "
610 "this file system.", (u_longlong_t
)zsb
->z_version
,
611 (u_longlong_t
)spa_version(dmu_objset_spa(os
)));
615 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_NORMALIZE
, &zval
)) != 0)
617 zsb
->z_norm
= (int)zval
;
619 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_UTF8ONLY
, &zval
)) != 0)
621 zsb
->z_utf8
= (zval
!= 0);
623 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_CASE
, &zval
)) != 0)
625 zsb
->z_case
= (uint_t
)zval
;
628 * Fold case on file systems that are always or sometimes case
631 if (zsb
->z_case
== ZFS_CASE_INSENSITIVE
||
632 zsb
->z_case
== ZFS_CASE_MIXED
)
633 zsb
->z_norm
|= U8_TEXTPREP_TOUPPER
;
635 zsb
->z_use_fuids
= USE_FUIDS(zsb
->z_version
, zsb
->z_os
);
636 zsb
->z_use_sa
= USE_SA(zsb
->z_version
, zsb
->z_os
);
639 /* should either have both of these objects or none */
640 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SA_ATTRS
, 8, 1,
646 * Pre SA versions file systems should never touch
647 * either the attribute registration or layout objects.
652 error
= sa_setup(os
, sa_obj
, zfs_attr_table
, ZPL_END
,
657 if (zsb
->z_version
>= ZPL_VERSION_SA
)
658 sa_register_update_callback(os
, zfs_sa_upgrade
);
660 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_ROOT_OBJ
, 8, 1,
664 ASSERT(zsb
->z_root
!= 0);
666 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_UNLINKED_SET
, 8, 1,
667 &zsb
->z_unlinkedobj
);
671 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
672 zfs_userquota_prop_prefixes
[ZFS_PROP_USERQUOTA
],
673 8, 1, &zsb
->z_userquota_obj
);
674 if (error
&& error
!= ENOENT
)
677 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
678 zfs_userquota_prop_prefixes
[ZFS_PROP_GROUPQUOTA
],
679 8, 1, &zsb
->z_groupquota_obj
);
680 if (error
&& error
!= ENOENT
)
683 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_FUID_TABLES
, 8, 1,
685 if (error
&& error
!= ENOENT
)
688 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SHARES_DIR
, 8, 1,
690 if (error
&& error
!= ENOENT
)
693 mutex_init(&zsb
->z_znodes_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
694 mutex_init(&zsb
->z_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
695 list_create(&zsb
->z_all_znodes
, sizeof (znode_t
),
696 offsetof(znode_t
, z_link_node
));
697 rrw_init(&zsb
->z_teardown_lock
);
698 rw_init(&zsb
->z_teardown_inactive_lock
, NULL
, RW_DEFAULT
, NULL
);
699 rw_init(&zsb
->z_fuid_lock
, NULL
, RW_DEFAULT
, NULL
);
700 for (i
= 0; i
!= ZFS_OBJ_MTX_SZ
; i
++)
701 mutex_init(&zsb
->z_hold_mtx
[i
], NULL
, MUTEX_DEFAULT
, NULL
);
707 dmu_objset_disown(os
, zsb
);
709 kmem_free(zsb
, sizeof (zfs_sb_t
));
714 zfs_sb_setup(zfs_sb_t
*zsb
, boolean_t mounting
)
718 error
= zfs_register_callbacks(zsb
);
723 * Set the objset user_ptr to track its zsb.
725 mutex_enter(&zsb
->z_os
->os_user_ptr_lock
);
726 dmu_objset_set_user(zsb
->z_os
, zsb
);
727 mutex_exit(&zsb
->z_os
->os_user_ptr_lock
);
729 zsb
->z_log
= zil_open(zsb
->z_os
, zfs_get_data
);
732 * If we are not mounting (ie: online recv), then we don't
733 * have to worry about replaying the log as we blocked all
734 * operations out since we closed the ZIL.
740 * During replay we remove the read only flag to
741 * allow replays to succeed.
743 readonly
= zfs_is_readonly(zsb
);
745 readonly_changed_cb(zsb
, B_FALSE
);
747 zfs_unlinked_drain(zsb
);
750 * Parse and replay the intent log.
752 * Because of ziltest, this must be done after
753 * zfs_unlinked_drain(). (Further note: ziltest
754 * doesn't use readonly mounts, where
755 * zfs_unlinked_drain() isn't called.) This is because
756 * ziltest causes spa_sync() to think it's committed,
757 * but actually it is not, so the intent log contains
758 * many txg's worth of changes.
760 * In particular, if object N is in the unlinked set in
761 * the last txg to actually sync, then it could be
762 * actually freed in a later txg and then reallocated
763 * in a yet later txg. This would write a "create
764 * object N" record to the intent log. Normally, this
765 * would be fine because the spa_sync() would have
766 * written out the fact that object N is free, before
767 * we could write the "create object N" intent log
770 * But when we are in ziltest mode, we advance the "open
771 * txg" without actually spa_sync()-ing the changes to
772 * disk. So we would see that object N is still
773 * allocated and in the unlinked set, and there is an
774 * intent log record saying to allocate it.
776 if (spa_writeable(dmu_objset_spa(zsb
->z_os
))) {
777 if (zil_replay_disable
) {
778 zil_destroy(zsb
->z_log
, B_FALSE
);
780 zsb
->z_replay
= B_TRUE
;
781 zil_replay(zsb
->z_os
, zsb
,
783 zsb
->z_replay
= B_FALSE
;
787 /* restore readonly bit */
789 readonly_changed_cb(zsb
, B_TRUE
);
796 zfs_sb_free(zfs_sb_t
*zsb
)
800 zfs_fuid_destroy(zsb
);
802 mutex_destroy(&zsb
->z_znodes_lock
);
803 mutex_destroy(&zsb
->z_lock
);
804 list_destroy(&zsb
->z_all_znodes
);
805 rrw_destroy(&zsb
->z_teardown_lock
);
806 rw_destroy(&zsb
->z_teardown_inactive_lock
);
807 rw_destroy(&zsb
->z_fuid_lock
);
808 for (i
= 0; i
!= ZFS_OBJ_MTX_SZ
; i
++)
809 mutex_destroy(&zsb
->z_hold_mtx
[i
]);
810 kmem_free(zsb
, sizeof (zfs_sb_t
));
814 zfs_set_fuid_feature(zfs_sb_t
*zsb
)
816 zsb
->z_use_fuids
= USE_FUIDS(zsb
->z_version
, zsb
->z_os
);
817 zsb
->z_use_sa
= USE_SA(zsb
->z_version
, zsb
->z_os
);
821 zfs_unregister_callbacks(zfs_sb_t
*zsb
)
823 objset_t
*os
= zsb
->z_os
;
824 struct dsl_dataset
*ds
;
827 * Unregister properties.
829 if (!dmu_objset_is_snapshot(os
)) {
830 ds
= dmu_objset_ds(os
);
831 VERIFY(dsl_prop_unregister(ds
, "atime", atime_changed_cb
,
834 VERIFY(dsl_prop_unregister(ds
, "xattr", xattr_changed_cb
,
837 VERIFY(dsl_prop_unregister(ds
, "recordsize", blksz_changed_cb
,
840 VERIFY(dsl_prop_unregister(ds
, "readonly", readonly_changed_cb
,
843 VERIFY(dsl_prop_unregister(ds
, "devices", devices_changed_cb
,
846 VERIFY(dsl_prop_unregister(ds
, "setuid", setuid_changed_cb
,
849 VERIFY(dsl_prop_unregister(ds
, "exec", exec_changed_cb
,
852 VERIFY(dsl_prop_unregister(ds
, "snapdir", snapdir_changed_cb
,
855 VERIFY(dsl_prop_unregister(ds
, "aclinherit",
856 acl_inherit_changed_cb
, zsb
) == 0);
858 VERIFY(dsl_prop_unregister(ds
, "vscan",
859 vscan_changed_cb
, zsb
) == 0);
861 VERIFY(dsl_prop_unregister(ds
, "nbmand",
862 nbmand_changed_cb
, zsb
) == 0);
865 EXPORT_SYMBOL(zfs_unregister_callbacks
);
869 * zfs_check_global_label:
870 * Check that the hex label string is appropriate for the dataset
871 * being mounted into the global_zone proper.
873 * Return an error if the hex label string is not default or
874 * admin_low/admin_high. For admin_low labels, the corresponding
875 * dataset must be readonly.
878 zfs_check_global_label(const char *dsname
, const char *hexsl
)
880 if (strcasecmp(hexsl
, ZFS_MLSLABEL_DEFAULT
) == 0)
882 if (strcasecmp(hexsl
, ADMIN_HIGH
) == 0)
884 if (strcasecmp(hexsl
, ADMIN_LOW
) == 0) {
885 /* must be readonly */
888 if (dsl_prop_get_integer(dsname
,
889 zfs_prop_to_name(ZFS_PROP_READONLY
), &rdonly
, NULL
))
891 return (rdonly
? 0 : EACCES
);
895 #endif /* HAVE_MLSLABEL */
898 zfs_statvfs(struct dentry
*dentry
, struct kstatfs
*statp
)
900 zfs_sb_t
*zsb
= dentry
->d_sb
->s_fs_info
;
901 uint64_t refdbytes
, availbytes
, usedobjs
, availobjs
;
906 dmu_objset_space(zsb
->z_os
,
907 &refdbytes
, &availbytes
, &usedobjs
, &availobjs
);
910 * The underlying storage pool actually uses multiple block
911 * size. Under Solaris frsize (fragment size) is reported as
912 * the smallest block size we support, and bsize (block size)
913 * as the filesystem's maximum block size. Unfortunately,
914 * under Linux the fragment size and block size are often used
915 * interchangeably. Thus we are forced to report both of them
916 * as the filesystem's maximum block size.
918 statp
->f_frsize
= zsb
->z_max_blksz
;
919 statp
->f_bsize
= zsb
->z_max_blksz
;
920 bshift
= fls(statp
->f_bsize
) - 1;
923 * The following report "total" blocks of various kinds in
924 * the file system, but reported in terms of f_bsize - the
928 statp
->f_blocks
= (refdbytes
+ availbytes
) >> bshift
;
929 statp
->f_bfree
= availbytes
>> bshift
;
930 statp
->f_bavail
= statp
->f_bfree
; /* no root reservation */
933 * statvfs() should really be called statufs(), because it assumes
934 * static metadata. ZFS doesn't preallocate files, so the best
935 * we can do is report the max that could possibly fit in f_files,
936 * and that minus the number actually used in f_ffree.
937 * For f_ffree, report the smaller of the number of object available
938 * and the number of blocks (each object will take at least a block).
940 statp
->f_ffree
= MIN(availobjs
, statp
->f_bfree
);
941 statp
->f_files
= statp
->f_ffree
+ usedobjs
;
942 statp
->f_fsid
.val
[0] = dentry
->d_sb
->s_dev
;
943 statp
->f_fsid
.val
[1] = 0;
944 statp
->f_type
= ZFS_SUPER_MAGIC
;
945 statp
->f_namelen
= ZFS_MAXNAMELEN
;
948 * We have all of 40 characters to stuff a string here.
949 * Is there anything useful we could/should provide?
951 bzero(statp
->f_spare
, sizeof (statp
->f_spare
));
956 EXPORT_SYMBOL(zfs_statvfs
);
959 zfs_root(zfs_sb_t
*zsb
, struct inode
**ipp
)
966 error
= zfs_zget(zsb
, zsb
->z_root
, &rootzp
);
973 EXPORT_SYMBOL(zfs_root
);
976 * Teardown the zfs_sb_t::z_os.
978 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
979 * and 'z_teardown_inactive_lock' held.
982 zfsvfs_teardown(zfs_sb_t
*zsb
, boolean_t unmounting
)
986 rrw_enter(&zsb
->z_teardown_lock
, RW_WRITER
, FTAG
);
990 * We purge the parent filesystem's super block as the
991 * parent filesystem and all of its snapshots have their
992 * inode's super block set to the parent's filesystem's
993 * super block. Note, 'z_parent' is self referential
996 shrink_dcache_sb(zsb
->z_parent
->z_sb
);
997 (void) spl_invalidate_inodes(zsb
->z_parent
->z_sb
, 0);
1001 * Drain the iput_taskq to ensure all active references to the
1002 * zfs_sb_t have been handled only then can it be safely destroyed.
1004 taskq_wait(dsl_pool_iput_taskq(dmu_objset_pool(zsb
->z_os
)));
1007 * Close the zil. NB: Can't close the zil while zfs_inactive
1008 * threads are blocked as zil_close can call zfs_inactive.
1011 zil_close(zsb
->z_log
);
1015 rw_enter(&zsb
->z_teardown_inactive_lock
, RW_WRITER
);
1018 * If we are not unmounting (ie: online recv) and someone already
1019 * unmounted this file system while we were doing the switcheroo,
1020 * or a reopen of z_os failed then just bail out now.
1022 if (!unmounting
&& (zsb
->z_unmounted
|| zsb
->z_os
== NULL
)) {
1023 rw_exit(&zsb
->z_teardown_inactive_lock
);
1024 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1029 * At this point there are no vops active, and any new vops will
1030 * fail with EIO since we have z_teardown_lock for writer (only
1031 * relavent for forced unmount).
1033 * Release all holds on dbufs.
1035 mutex_enter(&zsb
->z_znodes_lock
);
1036 for (zp
= list_head(&zsb
->z_all_znodes
); zp
!= NULL
;
1037 zp
= list_next(&zsb
->z_all_znodes
, zp
))
1039 ASSERT(atomic_read(&ZTOI(zp
)->i_count
) > 0);
1040 zfs_znode_dmu_fini(zp
);
1042 mutex_exit(&zsb
->z_znodes_lock
);
1045 * If we are unmounting, set the unmounted flag and let new vops
1046 * unblock. zfs_inactive will have the unmounted behavior, and all
1047 * other vops will fail with EIO.
1050 zsb
->z_unmounted
= B_TRUE
;
1051 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1052 rw_exit(&zsb
->z_teardown_inactive_lock
);
1056 * z_os will be NULL if there was an error in attempting to reopen
1057 * zsb, so just return as the properties had already been
1059 * unregistered and cached data had been evicted before.
1061 if (zsb
->z_os
== NULL
)
1065 * Unregister properties.
1067 zfs_unregister_callbacks(zsb
);
1072 if (dmu_objset_is_dirty_anywhere(zsb
->z_os
))
1073 if (!zfs_is_readonly(zsb
))
1074 txg_wait_synced(dmu_objset_pool(zsb
->z_os
), 0);
1075 (void) dmu_objset_evict_dbufs(zsb
->z_os
);
1081 zfs_domount(struct super_block
*sb
, void *data
, int silent
)
1083 zpl_mount_data_t
*zmd
= data
;
1084 const char *osname
= zmd
->z_osname
;
1086 struct inode
*root_inode
;
1087 uint64_t recordsize
;
1090 error
= zfs_sb_create(osname
, &zsb
);
1094 if ((error
= dsl_prop_get_integer(osname
, "recordsize",
1095 &recordsize
, NULL
)))
1099 sb
->s_fs_info
= zsb
;
1100 sb
->s_magic
= ZFS_SUPER_MAGIC
;
1101 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1102 sb
->s_time_gran
= 1;
1103 sb
->s_blocksize
= recordsize
;
1104 sb
->s_blocksize_bits
= ilog2(recordsize
);
1106 /* Set callback operations for the file system. */
1107 sb
->s_op
= &zpl_super_operations
;
1108 sb
->s_xattr
= zpl_xattr_handlers
;
1109 sb
->s_export_op
= &zpl_export_operations
;
1111 /* Set features for file system. */
1112 zfs_set_fuid_feature(zsb
);
1114 if (dmu_objset_is_snapshot(zsb
->z_os
)) {
1117 atime_changed_cb(zsb
, B_FALSE
);
1118 readonly_changed_cb(zsb
, B_TRUE
);
1119 if ((error
= dsl_prop_get_integer(osname
,"xattr",&pval
,NULL
)))
1121 xattr_changed_cb(zsb
, pval
);
1122 zsb
->z_issnap
= B_TRUE
;
1123 zsb
->z_os
->os_sync
= ZFS_SYNC_DISABLED
;
1125 mutex_enter(&zsb
->z_os
->os_user_ptr_lock
);
1126 dmu_objset_set_user(zsb
->z_os
, zsb
);
1127 mutex_exit(&zsb
->z_os
->os_user_ptr_lock
);
1129 error
= zfs_sb_setup(zsb
, B_TRUE
);
1130 #ifdef HAVE_SNAPSHOT
1131 (void) zfs_snap_create(zsb
);
1132 #endif /* HAVE_SNAPSHOT */
1135 /* Allocate a root inode for the filesystem. */
1136 error
= zfs_root(zsb
, &root_inode
);
1138 (void) zfs_umount(sb
);
1142 /* Allocate a root dentry for the filesystem */
1143 sb
->s_root
= d_alloc_root(root_inode
);
1144 if (sb
->s_root
== NULL
) {
1145 (void) zfs_umount(sb
);
1151 dmu_objset_disown(zsb
->z_os
, zsb
);
1157 EXPORT_SYMBOL(zfs_domount
);
1161 zfs_umount(struct super_block
*sb
)
1163 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1166 VERIFY(zfsvfs_teardown(zsb
, B_TRUE
) == 0);
1170 * z_os will be NULL if there was an error in
1171 * attempting to reopen zsb.
1175 * Unset the objset user_ptr.
1177 mutex_enter(&os
->os_user_ptr_lock
);
1178 dmu_objset_set_user(os
, NULL
);
1179 mutex_exit(&os
->os_user_ptr_lock
);
1182 * Finally release the objset
1184 dmu_objset_disown(os
, zsb
);
1190 EXPORT_SYMBOL(zfs_umount
);
1193 zfs_remount(struct super_block
*sb
, int *flags
, char *data
)
1196 * All namespace flags (MNT_*) and super block flags (MS_*) will
1197 * be handled by the Linux VFS. Only handle custom options here.
1201 EXPORT_SYMBOL(zfs_remount
);
1204 zfs_vget(struct super_block
*sb
, struct inode
**ipp
, fid_t
*fidp
)
1206 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1208 uint64_t object
= 0;
1209 uint64_t fid_gen
= 0;
1218 if (fidp
->fid_len
== LONG_FID_LEN
) {
1219 zfid_long_t
*zlfid
= (zfid_long_t
*)fidp
;
1220 uint64_t objsetid
= 0;
1221 uint64_t setgen
= 0;
1223 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
1224 objsetid
|= ((uint64_t)zlfid
->zf_setid
[i
]) << (8 * i
);
1226 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
1227 setgen
|= ((uint64_t)zlfid
->zf_setgen
[i
]) << (8 * i
);
1231 #ifdef HAVE_SNAPSHOT
1232 err
= zfsctl_lookup_objset(vfsp
, objsetid
, &zsb
);
1235 #endif /* HAVE_SNAPSHOT */
1239 if (fidp
->fid_len
== SHORT_FID_LEN
|| fidp
->fid_len
== LONG_FID_LEN
) {
1240 zfid_short_t
*zfid
= (zfid_short_t
*)fidp
;
1242 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
1243 object
|= ((uint64_t)zfid
->zf_object
[i
]) << (8 * i
);
1245 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
1246 fid_gen
|= ((uint64_t)zfid
->zf_gen
[i
]) << (8 * i
);
1252 #ifdef HAVE_SNAPSHOT
1253 /* A zero fid_gen means we are in the .zfs control directories */
1255 (object
== ZFSCTL_INO_ROOT
|| object
== ZFSCTL_INO_SNAPDIR
)) {
1256 *ipp
= zsb
->z_ctldir
;
1257 ASSERT(*ipp
!= NULL
);
1258 if (object
== ZFSCTL_INO_SNAPDIR
) {
1259 VERIFY(zfsctl_root_lookup(*ipp
, "snapshot", ipp
, NULL
,
1260 0, NULL
, NULL
, NULL
, NULL
, NULL
) == 0);
1267 #endif /* HAVE_SNAPSHOT */
1269 gen_mask
= -1ULL >> (64 - 8 * i
);
1271 dprintf("getting %llu [%u mask %llx]\n", object
, fid_gen
, gen_mask
);
1272 if ((err
= zfs_zget(zsb
, object
, &zp
))) {
1276 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zsb
), &zp_gen
,
1278 zp_gen
= zp_gen
& gen_mask
;
1281 if (zp
->z_unlinked
|| zp_gen
!= fid_gen
) {
1282 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen
, fid_gen
);
1290 zfs_inode_update(ITOZ(*ipp
));
1295 EXPORT_SYMBOL(zfs_vget
);
1298 * Block out VOPs and close zfs_sb_t::z_os
1300 * Note, if successful, then we return with the 'z_teardown_lock' and
1301 * 'z_teardown_inactive_lock' write held.
1304 zfs_suspend_fs(zfs_sb_t
*zsb
)
1308 if ((error
= zfsvfs_teardown(zsb
, B_FALSE
)) != 0)
1310 dmu_objset_disown(zsb
->z_os
, zsb
);
1314 EXPORT_SYMBOL(zfs_suspend_fs
);
1317 * Reopen zfs_sb_t::z_os and release VOPs.
1320 zfs_resume_fs(zfs_sb_t
*zsb
, const char *osname
)
1324 ASSERT(RRW_WRITE_HELD(&zsb
->z_teardown_lock
));
1325 ASSERT(RW_WRITE_HELD(&zsb
->z_teardown_inactive_lock
));
1327 err
= dmu_objset_own(osname
, DMU_OST_ZFS
, B_FALSE
, zsb
, &zsb
->z_os
);
1332 uint64_t sa_obj
= 0;
1334 err2
= zap_lookup(zsb
->z_os
, MASTER_NODE_OBJ
,
1335 ZFS_SA_ATTRS
, 8, 1, &sa_obj
);
1337 if ((err
|| err2
) && zsb
->z_version
>= ZPL_VERSION_SA
)
1341 if ((err
= sa_setup(zsb
->z_os
, sa_obj
,
1342 zfs_attr_table
, ZPL_END
, &zsb
->z_attr_table
)) != 0)
1345 VERIFY(zfs_sb_setup(zsb
, B_FALSE
) == 0);
1348 * Attempt to re-establish all the active znodes with
1349 * their dbufs. If a zfs_rezget() fails, then we'll let
1350 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
1351 * when they try to use their znode.
1353 mutex_enter(&zsb
->z_znodes_lock
);
1354 for (zp
= list_head(&zsb
->z_all_znodes
); zp
;
1355 zp
= list_next(&zsb
->z_all_znodes
, zp
)) {
1356 (void) zfs_rezget(zp
);
1358 mutex_exit(&zsb
->z_znodes_lock
);
1363 /* release the VOPs */
1364 rw_exit(&zsb
->z_teardown_inactive_lock
);
1365 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1369 * Since we couldn't reopen zfs_sb_t::z_os, force
1370 * unmount this file system.
1372 (void) zfs_umount(zsb
->z_sb
);
1376 EXPORT_SYMBOL(zfs_resume_fs
);
1379 zfs_set_version(zfs_sb_t
*zsb
, uint64_t newvers
)
1382 objset_t
*os
= zsb
->z_os
;
1385 if (newvers
< ZPL_VERSION_INITIAL
|| newvers
> ZPL_VERSION
)
1388 if (newvers
< zsb
->z_version
)
1391 if (zfs_spa_version_map(newvers
) >
1392 spa_version(dmu_objset_spa(zsb
->z_os
)))
1395 tx
= dmu_tx_create(os
);
1396 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_FALSE
, ZPL_VERSION_STR
);
1397 if (newvers
>= ZPL_VERSION_SA
&& !zsb
->z_use_sa
) {
1398 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
1400 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1402 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1408 error
= zap_update(os
, MASTER_NODE_OBJ
, ZPL_VERSION_STR
,
1409 8, 1, &newvers
, tx
);
1416 if (newvers
>= ZPL_VERSION_SA
&& !zsb
->z_use_sa
) {
1419 ASSERT3U(spa_version(dmu_objset_spa(zsb
->z_os
)), >=,
1421 sa_obj
= zap_create(os
, DMU_OT_SA_MASTER_NODE
,
1422 DMU_OT_NONE
, 0, tx
);
1424 error
= zap_add(os
, MASTER_NODE_OBJ
,
1425 ZFS_SA_ATTRS
, 8, 1, &sa_obj
, tx
);
1426 ASSERT3U(error
, ==, 0);
1428 VERIFY(0 == sa_set_sa_object(os
, sa_obj
));
1429 sa_register_update_callback(os
, zfs_sa_upgrade
);
1432 spa_history_log_internal(LOG_DS_UPGRADE
,
1433 dmu_objset_spa(os
), tx
, "oldver=%llu newver=%llu dataset = %llu",
1434 zsb
->z_version
, newvers
, dmu_objset_id(os
));
1438 zsb
->z_version
= newvers
;
1440 if (zsb
->z_version
>= ZPL_VERSION_FUID
)
1441 zfs_set_fuid_feature(zsb
);
1445 EXPORT_SYMBOL(zfs_set_version
);
1448 * Read a property stored within the master node.
1451 zfs_get_zplprop(objset_t
*os
, zfs_prop_t prop
, uint64_t *value
)
1457 * Look up the file system's value for the property. For the
1458 * version property, we look up a slightly different string.
1460 if (prop
== ZFS_PROP_VERSION
)
1461 pname
= ZPL_VERSION_STR
;
1463 pname
= zfs_prop_to_name(prop
);
1466 error
= zap_lookup(os
, MASTER_NODE_OBJ
, pname
, 8, 1, value
);
1468 if (error
== ENOENT
) {
1469 /* No value set, use the default value */
1471 case ZFS_PROP_VERSION
:
1472 *value
= ZPL_VERSION
;
1474 case ZFS_PROP_NORMALIZE
:
1475 case ZFS_PROP_UTF8ONLY
:
1479 *value
= ZFS_CASE_SENSITIVE
;
1493 dmu_objset_register_type(DMU_OST_ZFS
, zfs_space_delta_cb
);
1494 register_filesystem(&zpl_fs_type
);
1500 unregister_filesystem(&zpl_fs_type
);