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
)
143 if (newval
== ZFS_XATTR_OFF
) {
144 zsb
->z_flags
&= ~ZSB_XATTR
;
146 zsb
->z_flags
|= ZSB_XATTR
;
148 if (newval
== ZFS_XATTR_SA
)
149 zsb
->z_xattr_sa
= B_TRUE
;
151 zsb
->z_xattr_sa
= B_FALSE
;
156 blksz_changed_cb(void *arg
, uint64_t newval
)
160 if (newval
< SPA_MINBLOCKSIZE
||
161 newval
> SPA_MAXBLOCKSIZE
|| !ISP2(newval
))
162 newval
= SPA_MAXBLOCKSIZE
;
164 zsb
->z_max_blksz
= newval
;
168 readonly_changed_cb(void *arg
, uint64_t newval
)
171 struct super_block
*sb
= zsb
->z_sb
;
177 sb
->s_flags
|= MS_RDONLY
;
179 sb
->s_flags
&= ~MS_RDONLY
;
183 devices_changed_cb(void *arg
, uint64_t newval
)
188 setuid_changed_cb(void *arg
, uint64_t newval
)
193 exec_changed_cb(void *arg
, uint64_t newval
)
198 nbmand_changed_cb(void *arg
, uint64_t newval
)
201 struct super_block
*sb
= zsb
->z_sb
;
207 sb
->s_flags
|= MS_MANDLOCK
;
209 sb
->s_flags
&= ~MS_MANDLOCK
;
213 snapdir_changed_cb(void *arg
, uint64_t newval
)
215 ((zfs_sb_t
*)arg
)->z_show_ctldir
= newval
;
219 vscan_changed_cb(void *arg
, uint64_t newval
)
221 ((zfs_sb_t
*)arg
)->z_vscan
= newval
;
225 acl_inherit_changed_cb(void *arg
, uint64_t newval
)
227 ((zfs_sb_t
*)arg
)->z_acl_inherit
= newval
;
231 zfs_register_callbacks(zfs_sb_t
*zsb
)
233 struct dsl_dataset
*ds
= NULL
;
234 objset_t
*os
= zsb
->z_os
;
237 if (zfs_is_readonly(zsb
) || !spa_writeable(dmu_objset_spa(os
)))
238 readonly_changed_cb(zsb
, B_TRUE
);
241 * Register property callbacks.
243 * It would probably be fine to just check for i/o error from
244 * the first prop_register(), but I guess I like to go
247 ds
= dmu_objset_ds(os
);
248 error
= dsl_prop_register(ds
,
249 "atime", atime_changed_cb
, zsb
);
250 error
= error
? error
: dsl_prop_register(ds
,
251 "xattr", xattr_changed_cb
, zsb
);
252 error
= error
? error
: dsl_prop_register(ds
,
253 "recordsize", blksz_changed_cb
, zsb
);
254 error
= error
? error
: dsl_prop_register(ds
,
255 "readonly", readonly_changed_cb
, zsb
);
256 error
= error
? error
: dsl_prop_register(ds
,
257 "devices", devices_changed_cb
, zsb
);
258 error
= error
? error
: dsl_prop_register(ds
,
259 "setuid", setuid_changed_cb
, zsb
);
260 error
= error
? error
: dsl_prop_register(ds
,
261 "exec", exec_changed_cb
, zsb
);
262 error
= error
? error
: dsl_prop_register(ds
,
263 "snapdir", snapdir_changed_cb
, zsb
);
264 error
= error
? error
: dsl_prop_register(ds
,
265 "aclinherit", acl_inherit_changed_cb
, zsb
);
266 error
= error
? error
: dsl_prop_register(ds
,
267 "vscan", vscan_changed_cb
, zsb
);
268 error
= error
? error
: dsl_prop_register(ds
,
269 "nbmand", nbmand_changed_cb
, zsb
);
277 * We may attempt to unregister some callbacks that are not
278 * registered, but this is OK; it will simply return ENOMSG,
279 * which we will ignore.
281 (void) dsl_prop_unregister(ds
, "atime", atime_changed_cb
, zsb
);
282 (void) dsl_prop_unregister(ds
, "xattr", xattr_changed_cb
, zsb
);
283 (void) dsl_prop_unregister(ds
, "recordsize", blksz_changed_cb
, zsb
);
284 (void) dsl_prop_unregister(ds
, "readonly", readonly_changed_cb
, zsb
);
285 (void) dsl_prop_unregister(ds
, "devices", devices_changed_cb
, zsb
);
286 (void) dsl_prop_unregister(ds
, "setuid", setuid_changed_cb
, zsb
);
287 (void) dsl_prop_unregister(ds
, "exec", exec_changed_cb
, zsb
);
288 (void) dsl_prop_unregister(ds
, "snapdir", snapdir_changed_cb
, zsb
);
289 (void) dsl_prop_unregister(ds
, "aclinherit", acl_inherit_changed_cb
,
291 (void) dsl_prop_unregister(ds
, "vscan", vscan_changed_cb
, zsb
);
292 (void) dsl_prop_unregister(ds
, "nbmand", nbmand_changed_cb
, zsb
);
296 EXPORT_SYMBOL(zfs_register_callbacks
);
299 zfs_space_delta_cb(dmu_object_type_t bonustype
, void *data
,
300 uint64_t *userp
, uint64_t *groupp
)
302 znode_phys_t
*znp
= data
;
306 * Is it a valid type of object to track?
308 if (bonustype
!= DMU_OT_ZNODE
&& bonustype
!= DMU_OT_SA
)
312 * If we have a NULL data pointer
313 * then assume the id's aren't changing and
314 * return EEXIST to the dmu to let it know to
320 if (bonustype
== DMU_OT_ZNODE
) {
321 *userp
= znp
->zp_uid
;
322 *groupp
= znp
->zp_gid
;
326 ASSERT(bonustype
== DMU_OT_SA
);
327 hdrsize
= sa_hdrsize(data
);
330 *userp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+
332 *groupp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+
336 * This should only happen for newly created
337 * files that haven't had the znode data filled
348 fuidstr_to_sid(zfs_sb_t
*zsb
, const char *fuidstr
,
349 char *domainbuf
, int buflen
, uid_t
*ridp
)
354 fuid
= strtonum(fuidstr
, NULL
);
356 domain
= zfs_fuid_find_by_idx(zsb
, FUID_INDEX(fuid
));
358 (void) strlcpy(domainbuf
, domain
, buflen
);
361 *ridp
= FUID_RID(fuid
);
365 zfs_userquota_prop_to_obj(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
)
368 case ZFS_PROP_USERUSED
:
369 return (DMU_USERUSED_OBJECT
);
370 case ZFS_PROP_GROUPUSED
:
371 return (DMU_GROUPUSED_OBJECT
);
372 case ZFS_PROP_USERQUOTA
:
373 return (zsb
->z_userquota_obj
);
374 case ZFS_PROP_GROUPQUOTA
:
375 return (zsb
->z_groupquota_obj
);
383 zfs_userspace_many(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
384 uint64_t *cookiep
, void *vbuf
, uint64_t *bufsizep
)
389 zfs_useracct_t
*buf
= vbuf
;
392 if (!dmu_objset_userspace_present(zsb
->z_os
))
395 obj
= zfs_userquota_prop_to_obj(zsb
, type
);
401 for (zap_cursor_init_serialized(&zc
, zsb
->z_os
, obj
, *cookiep
);
402 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
403 zap_cursor_advance(&zc
)) {
404 if ((uintptr_t)buf
- (uintptr_t)vbuf
+ sizeof (zfs_useracct_t
) >
408 fuidstr_to_sid(zsb
, za
.za_name
,
409 buf
->zu_domain
, sizeof (buf
->zu_domain
), &buf
->zu_rid
);
411 buf
->zu_space
= za
.za_first_integer
;
417 ASSERT3U((uintptr_t)buf
- (uintptr_t)vbuf
, <=, *bufsizep
);
418 *bufsizep
= (uintptr_t)buf
- (uintptr_t)vbuf
;
419 *cookiep
= zap_cursor_serialize(&zc
);
420 zap_cursor_fini(&zc
);
423 EXPORT_SYMBOL(zfs_userspace_many
);
426 * buf must be big enough (eg, 32 bytes)
429 id_to_fuidstr(zfs_sb_t
*zsb
, const char *domain
, uid_t rid
,
430 char *buf
, boolean_t addok
)
435 if (domain
&& domain
[0]) {
436 domainid
= zfs_fuid_find_by_domain(zsb
, domain
, NULL
, addok
);
440 fuid
= FUID_ENCODE(domainid
, rid
);
441 (void) sprintf(buf
, "%llx", (longlong_t
)fuid
);
446 zfs_userspace_one(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
447 const char *domain
, uint64_t rid
, uint64_t *valp
)
455 if (!dmu_objset_userspace_present(zsb
->z_os
))
458 obj
= zfs_userquota_prop_to_obj(zsb
, type
);
462 err
= id_to_fuidstr(zsb
, domain
, rid
, buf
, B_FALSE
);
466 err
= zap_lookup(zsb
->z_os
, obj
, buf
, 8, 1, valp
);
471 EXPORT_SYMBOL(zfs_userspace_one
);
474 zfs_set_userquota(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
475 const char *domain
, uint64_t rid
, uint64_t quota
)
481 boolean_t fuid_dirtied
;
483 if (type
!= ZFS_PROP_USERQUOTA
&& type
!= ZFS_PROP_GROUPQUOTA
)
486 if (zsb
->z_version
< ZPL_VERSION_USERSPACE
)
489 objp
= (type
== ZFS_PROP_USERQUOTA
) ? &zsb
->z_userquota_obj
:
490 &zsb
->z_groupquota_obj
;
492 err
= id_to_fuidstr(zsb
, domain
, rid
, buf
, B_TRUE
);
495 fuid_dirtied
= zsb
->z_fuid_dirty
;
497 tx
= dmu_tx_create(zsb
->z_os
);
498 dmu_tx_hold_zap(tx
, *objp
? *objp
: DMU_NEW_OBJECT
, B_TRUE
, NULL
);
500 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
501 zfs_userquota_prop_prefixes
[type
]);
504 zfs_fuid_txhold(zsb
, tx
);
505 err
= dmu_tx_assign(tx
, TXG_WAIT
);
511 mutex_enter(&zsb
->z_lock
);
513 *objp
= zap_create(zsb
->z_os
, DMU_OT_USERGROUP_QUOTA
,
515 VERIFY(0 == zap_add(zsb
->z_os
, MASTER_NODE_OBJ
,
516 zfs_userquota_prop_prefixes
[type
], 8, 1, objp
, tx
));
518 mutex_exit(&zsb
->z_lock
);
521 err
= zap_remove(zsb
->z_os
, *objp
, buf
, tx
);
525 err
= zap_update(zsb
->z_os
, *objp
, buf
, 8, 1, "a
, tx
);
529 zfs_fuid_sync(zsb
, tx
);
533 EXPORT_SYMBOL(zfs_set_userquota
);
536 zfs_fuid_overquota(zfs_sb_t
*zsb
, boolean_t isgroup
, uint64_t fuid
)
539 uint64_t used
, quota
, usedobj
, quotaobj
;
542 usedobj
= isgroup
? DMU_GROUPUSED_OBJECT
: DMU_USERUSED_OBJECT
;
543 quotaobj
= isgroup
? zsb
->z_groupquota_obj
: zsb
->z_userquota_obj
;
545 if (quotaobj
== 0 || zsb
->z_replay
)
548 (void) sprintf(buf
, "%llx", (longlong_t
)fuid
);
549 err
= zap_lookup(zsb
->z_os
, quotaobj
, buf
, 8, 1, "a
);
553 err
= zap_lookup(zsb
->z_os
, usedobj
, buf
, 8, 1, &used
);
556 return (used
>= quota
);
558 EXPORT_SYMBOL(zfs_fuid_overquota
);
561 zfs_owner_overquota(zfs_sb_t
*zsb
, znode_t
*zp
, boolean_t isgroup
)
566 quotaobj
= isgroup
? zsb
->z_groupquota_obj
: zsb
->z_userquota_obj
;
568 fuid
= isgroup
? zp
->z_gid
: zp
->z_uid
;
570 if (quotaobj
== 0 || zsb
->z_replay
)
573 return (zfs_fuid_overquota(zsb
, isgroup
, fuid
));
575 EXPORT_SYMBOL(zfs_owner_overquota
);
578 zfs_sb_create(const char *osname
, zfs_sb_t
**zsbp
)
586 zsb
= kmem_zalloc(sizeof (zfs_sb_t
), KM_SLEEP
);
589 * We claim to always be readonly so we can open snapshots;
590 * other ZPL code will prevent us from writing to snapshots.
592 error
= dmu_objset_own(osname
, DMU_OST_ZFS
, B_TRUE
, zsb
, &os
);
594 kmem_free(zsb
, sizeof (zfs_sb_t
));
599 * Initialize the zfs-specific filesystem structure.
600 * Should probably make this a kmem cache, shuffle fields,
601 * and just bzero up to z_hold_mtx[].
605 zsb
->z_max_blksz
= SPA_MAXBLOCKSIZE
;
606 zsb
->z_show_ctldir
= ZFS_SNAPDIR_VISIBLE
;
609 error
= zfs_get_zplprop(os
, ZFS_PROP_VERSION
, &zsb
->z_version
);
612 } else if (zsb
->z_version
>
613 zfs_zpl_version_map(spa_version(dmu_objset_spa(os
)))) {
614 (void) printk("Can't mount a version %lld file system "
615 "on a version %lld pool\n. Pool must be upgraded to mount "
616 "this file system.", (u_longlong_t
)zsb
->z_version
,
617 (u_longlong_t
)spa_version(dmu_objset_spa(os
)));
621 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_NORMALIZE
, &zval
)) != 0)
623 zsb
->z_norm
= (int)zval
;
625 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_UTF8ONLY
, &zval
)) != 0)
627 zsb
->z_utf8
= (zval
!= 0);
629 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_CASE
, &zval
)) != 0)
631 zsb
->z_case
= (uint_t
)zval
;
634 * Fold case on file systems that are always or sometimes case
637 if (zsb
->z_case
== ZFS_CASE_INSENSITIVE
||
638 zsb
->z_case
== ZFS_CASE_MIXED
)
639 zsb
->z_norm
|= U8_TEXTPREP_TOUPPER
;
641 zsb
->z_use_fuids
= USE_FUIDS(zsb
->z_version
, zsb
->z_os
);
642 zsb
->z_use_sa
= USE_SA(zsb
->z_version
, zsb
->z_os
);
645 /* should either have both of these objects or none */
646 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SA_ATTRS
, 8, 1,
651 error
= zfs_get_zplprop(os
, ZFS_PROP_XATTR
, &zval
);
652 if ((error
== 0) && (zval
== ZFS_XATTR_SA
))
653 zsb
->z_xattr_sa
= B_TRUE
;
656 * Pre SA versions file systems should never touch
657 * either the attribute registration or layout objects.
662 error
= sa_setup(os
, sa_obj
, zfs_attr_table
, ZPL_END
,
667 if (zsb
->z_version
>= ZPL_VERSION_SA
)
668 sa_register_update_callback(os
, zfs_sa_upgrade
);
670 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_ROOT_OBJ
, 8, 1,
674 ASSERT(zsb
->z_root
!= 0);
676 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_UNLINKED_SET
, 8, 1,
677 &zsb
->z_unlinkedobj
);
681 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
682 zfs_userquota_prop_prefixes
[ZFS_PROP_USERQUOTA
],
683 8, 1, &zsb
->z_userquota_obj
);
684 if (error
&& error
!= ENOENT
)
687 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
688 zfs_userquota_prop_prefixes
[ZFS_PROP_GROUPQUOTA
],
689 8, 1, &zsb
->z_groupquota_obj
);
690 if (error
&& error
!= ENOENT
)
693 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_FUID_TABLES
, 8, 1,
695 if (error
&& error
!= ENOENT
)
698 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SHARES_DIR
, 8, 1,
700 if (error
&& error
!= ENOENT
)
703 mutex_init(&zsb
->z_znodes_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
704 mutex_init(&zsb
->z_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
705 list_create(&zsb
->z_all_znodes
, sizeof (znode_t
),
706 offsetof(znode_t
, z_link_node
));
707 rrw_init(&zsb
->z_teardown_lock
);
708 rw_init(&zsb
->z_teardown_inactive_lock
, NULL
, RW_DEFAULT
, NULL
);
709 rw_init(&zsb
->z_fuid_lock
, NULL
, RW_DEFAULT
, NULL
);
710 for (i
= 0; i
!= ZFS_OBJ_MTX_SZ
; i
++)
711 mutex_init(&zsb
->z_hold_mtx
[i
], NULL
, MUTEX_DEFAULT
, NULL
);
717 dmu_objset_disown(os
, zsb
);
719 kmem_free(zsb
, sizeof (zfs_sb_t
));
722 EXPORT_SYMBOL(zfs_sb_create
);
725 zfs_sb_setup(zfs_sb_t
*zsb
, boolean_t mounting
)
729 error
= zfs_register_callbacks(zsb
);
734 * Set the objset user_ptr to track its zsb.
736 mutex_enter(&zsb
->z_os
->os_user_ptr_lock
);
737 dmu_objset_set_user(zsb
->z_os
, zsb
);
738 mutex_exit(&zsb
->z_os
->os_user_ptr_lock
);
740 zsb
->z_log
= zil_open(zsb
->z_os
, zfs_get_data
);
743 * If we are not mounting (ie: online recv), then we don't
744 * have to worry about replaying the log as we blocked all
745 * operations out since we closed the ZIL.
751 * During replay we remove the read only flag to
752 * allow replays to succeed.
754 readonly
= zfs_is_readonly(zsb
);
756 readonly_changed_cb(zsb
, B_FALSE
);
758 zfs_unlinked_drain(zsb
);
761 * Parse and replay the intent log.
763 * Because of ziltest, this must be done after
764 * zfs_unlinked_drain(). (Further note: ziltest
765 * doesn't use readonly mounts, where
766 * zfs_unlinked_drain() isn't called.) This is because
767 * ziltest causes spa_sync() to think it's committed,
768 * but actually it is not, so the intent log contains
769 * many txg's worth of changes.
771 * In particular, if object N is in the unlinked set in
772 * the last txg to actually sync, then it could be
773 * actually freed in a later txg and then reallocated
774 * in a yet later txg. This would write a "create
775 * object N" record to the intent log. Normally, this
776 * would be fine because the spa_sync() would have
777 * written out the fact that object N is free, before
778 * we could write the "create object N" intent log
781 * But when we are in ziltest mode, we advance the "open
782 * txg" without actually spa_sync()-ing the changes to
783 * disk. So we would see that object N is still
784 * allocated and in the unlinked set, and there is an
785 * intent log record saying to allocate it.
787 if (spa_writeable(dmu_objset_spa(zsb
->z_os
))) {
788 if (zil_replay_disable
) {
789 zil_destroy(zsb
->z_log
, B_FALSE
);
791 zsb
->z_replay
= B_TRUE
;
792 zil_replay(zsb
->z_os
, zsb
,
794 zsb
->z_replay
= B_FALSE
;
798 /* restore readonly bit */
800 readonly_changed_cb(zsb
, B_TRUE
);
805 EXPORT_SYMBOL(zfs_sb_setup
);
808 zfs_sb_free(zfs_sb_t
*zsb
)
812 zfs_fuid_destroy(zsb
);
814 mutex_destroy(&zsb
->z_znodes_lock
);
815 mutex_destroy(&zsb
->z_lock
);
816 list_destroy(&zsb
->z_all_znodes
);
817 rrw_destroy(&zsb
->z_teardown_lock
);
818 rw_destroy(&zsb
->z_teardown_inactive_lock
);
819 rw_destroy(&zsb
->z_fuid_lock
);
820 for (i
= 0; i
!= ZFS_OBJ_MTX_SZ
; i
++)
821 mutex_destroy(&zsb
->z_hold_mtx
[i
]);
822 kmem_free(zsb
, sizeof (zfs_sb_t
));
824 EXPORT_SYMBOL(zfs_sb_free
);
827 zfs_set_fuid_feature(zfs_sb_t
*zsb
)
829 zsb
->z_use_fuids
= USE_FUIDS(zsb
->z_version
, zsb
->z_os
);
830 zsb
->z_use_sa
= USE_SA(zsb
->z_version
, zsb
->z_os
);
834 zfs_unregister_callbacks(zfs_sb_t
*zsb
)
836 objset_t
*os
= zsb
->z_os
;
837 struct dsl_dataset
*ds
;
840 * Unregister properties.
842 if (!dmu_objset_is_snapshot(os
)) {
843 ds
= dmu_objset_ds(os
);
844 VERIFY(dsl_prop_unregister(ds
, "atime", atime_changed_cb
,
847 VERIFY(dsl_prop_unregister(ds
, "xattr", xattr_changed_cb
,
850 VERIFY(dsl_prop_unregister(ds
, "recordsize", blksz_changed_cb
,
853 VERIFY(dsl_prop_unregister(ds
, "readonly", readonly_changed_cb
,
856 VERIFY(dsl_prop_unregister(ds
, "devices", devices_changed_cb
,
859 VERIFY(dsl_prop_unregister(ds
, "setuid", setuid_changed_cb
,
862 VERIFY(dsl_prop_unregister(ds
, "exec", exec_changed_cb
,
865 VERIFY(dsl_prop_unregister(ds
, "snapdir", snapdir_changed_cb
,
868 VERIFY(dsl_prop_unregister(ds
, "aclinherit",
869 acl_inherit_changed_cb
, zsb
) == 0);
871 VERIFY(dsl_prop_unregister(ds
, "vscan",
872 vscan_changed_cb
, zsb
) == 0);
874 VERIFY(dsl_prop_unregister(ds
, "nbmand",
875 nbmand_changed_cb
, zsb
) == 0);
878 EXPORT_SYMBOL(zfs_unregister_callbacks
);
882 * zfs_check_global_label:
883 * Check that the hex label string is appropriate for the dataset
884 * being mounted into the global_zone proper.
886 * Return an error if the hex label string is not default or
887 * admin_low/admin_high. For admin_low labels, the corresponding
888 * dataset must be readonly.
891 zfs_check_global_label(const char *dsname
, const char *hexsl
)
893 if (strcasecmp(hexsl
, ZFS_MLSLABEL_DEFAULT
) == 0)
895 if (strcasecmp(hexsl
, ADMIN_HIGH
) == 0)
897 if (strcasecmp(hexsl
, ADMIN_LOW
) == 0) {
898 /* must be readonly */
901 if (dsl_prop_get_integer(dsname
,
902 zfs_prop_to_name(ZFS_PROP_READONLY
), &rdonly
, NULL
))
904 return (rdonly
? 0 : EACCES
);
908 EXPORT_SYMBOL(zfs_check_global_label
);
909 #endif /* HAVE_MLSLABEL */
912 zfs_statvfs(struct dentry
*dentry
, struct kstatfs
*statp
)
914 zfs_sb_t
*zsb
= dentry
->d_sb
->s_fs_info
;
915 uint64_t refdbytes
, availbytes
, usedobjs
, availobjs
;
920 dmu_objset_space(zsb
->z_os
,
921 &refdbytes
, &availbytes
, &usedobjs
, &availobjs
);
924 * The underlying storage pool actually uses multiple block
925 * size. Under Solaris frsize (fragment size) is reported as
926 * the smallest block size we support, and bsize (block size)
927 * as the filesystem's maximum block size. Unfortunately,
928 * under Linux the fragment size and block size are often used
929 * interchangeably. Thus we are forced to report both of them
930 * as the filesystem's maximum block size.
932 statp
->f_frsize
= zsb
->z_max_blksz
;
933 statp
->f_bsize
= zsb
->z_max_blksz
;
934 bshift
= fls(statp
->f_bsize
) - 1;
937 * The following report "total" blocks of various kinds in
938 * the file system, but reported in terms of f_bsize - the
942 statp
->f_blocks
= (refdbytes
+ availbytes
) >> bshift
;
943 statp
->f_bfree
= availbytes
>> bshift
;
944 statp
->f_bavail
= statp
->f_bfree
; /* no root reservation */
947 * statvfs() should really be called statufs(), because it assumes
948 * static metadata. ZFS doesn't preallocate files, so the best
949 * we can do is report the max that could possibly fit in f_files,
950 * and that minus the number actually used in f_ffree.
951 * For f_ffree, report the smaller of the number of object available
952 * and the number of blocks (each object will take at least a block).
954 statp
->f_ffree
= MIN(availobjs
, availbytes
>> DNODE_SHIFT
);
955 statp
->f_files
= statp
->f_ffree
+ usedobjs
;
956 statp
->f_fsid
.val
[0] = dentry
->d_sb
->s_dev
;
957 statp
->f_fsid
.val
[1] = 0;
958 statp
->f_type
= ZFS_SUPER_MAGIC
;
959 statp
->f_namelen
= ZFS_MAXNAMELEN
;
962 * We have all of 40 characters to stuff a string here.
963 * Is there anything useful we could/should provide?
965 bzero(statp
->f_spare
, sizeof (statp
->f_spare
));
970 EXPORT_SYMBOL(zfs_statvfs
);
973 zfs_root(zfs_sb_t
*zsb
, struct inode
**ipp
)
980 error
= zfs_zget(zsb
, zsb
->z_root
, &rootzp
);
987 EXPORT_SYMBOL(zfs_root
);
990 * Teardown the zfs_sb_t::z_os.
992 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
993 * and 'z_teardown_inactive_lock' held.
996 zfs_sb_teardown(zfs_sb_t
*zsb
, boolean_t unmounting
)
1000 rrw_enter(&zsb
->z_teardown_lock
, RW_WRITER
, FTAG
);
1004 * We purge the parent filesystem's super block as the
1005 * parent filesystem and all of its snapshots have their
1006 * inode's super block set to the parent's filesystem's
1007 * super block. Note, 'z_parent' is self referential
1008 * for non-snapshots.
1010 shrink_dcache_sb(zsb
->z_parent
->z_sb
);
1011 (void) spl_invalidate_inodes(zsb
->z_parent
->z_sb
, 0);
1015 * Drain the iput_taskq to ensure all active references to the
1016 * zfs_sb_t have been handled only then can it be safely destroyed.
1018 taskq_wait(dsl_pool_iput_taskq(dmu_objset_pool(zsb
->z_os
)));
1021 * Close the zil. NB: Can't close the zil while zfs_inactive
1022 * threads are blocked as zil_close can call zfs_inactive.
1025 zil_close(zsb
->z_log
);
1029 rw_enter(&zsb
->z_teardown_inactive_lock
, RW_WRITER
);
1032 * If we are not unmounting (ie: online recv) and someone already
1033 * unmounted this file system while we were doing the switcheroo,
1034 * or a reopen of z_os failed then just bail out now.
1036 if (!unmounting
&& (zsb
->z_unmounted
|| zsb
->z_os
== NULL
)) {
1037 rw_exit(&zsb
->z_teardown_inactive_lock
);
1038 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1043 * At this point there are no vops active, and any new vops will
1044 * fail with EIO since we have z_teardown_lock for writer (only
1045 * relavent for forced unmount).
1047 * Release all holds on dbufs.
1049 mutex_enter(&zsb
->z_znodes_lock
);
1050 for (zp
= list_head(&zsb
->z_all_znodes
); zp
!= NULL
;
1051 zp
= list_next(&zsb
->z_all_znodes
, zp
))
1053 ASSERT(atomic_read(&ZTOI(zp
)->i_count
) > 0);
1054 zfs_znode_dmu_fini(zp
);
1056 mutex_exit(&zsb
->z_znodes_lock
);
1059 * If we are unmounting, set the unmounted flag and let new vops
1060 * unblock. zfs_inactive will have the unmounted behavior, and all
1061 * other vops will fail with EIO.
1064 zsb
->z_unmounted
= B_TRUE
;
1065 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1066 rw_exit(&zsb
->z_teardown_inactive_lock
);
1070 * z_os will be NULL if there was an error in attempting to reopen
1071 * zsb, so just return as the properties had already been
1073 * unregistered and cached data had been evicted before.
1075 if (zsb
->z_os
== NULL
)
1079 * Unregister properties.
1081 zfs_unregister_callbacks(zsb
);
1086 if (dmu_objset_is_dirty_anywhere(zsb
->z_os
))
1087 if (!zfs_is_readonly(zsb
))
1088 txg_wait_synced(dmu_objset_pool(zsb
->z_os
), 0);
1089 (void) dmu_objset_evict_dbufs(zsb
->z_os
);
1093 EXPORT_SYMBOL(zfs_sb_teardown
);
1095 #if defined(HAVE_BDI) && !defined(HAVE_BDI_SETUP_AND_REGISTER)
1096 atomic_long_t zfs_bdi_seq
= ATOMIC_LONG_INIT(0);
1097 #endif /* HAVE_BDI && !HAVE_BDI_SETUP_AND_REGISTER */
1100 zfs_domount(struct super_block
*sb
, void *data
, int silent
)
1102 zpl_mount_data_t
*zmd
= data
;
1103 const char *osname
= zmd
->z_osname
;
1105 struct inode
*root_inode
;
1106 uint64_t recordsize
;
1109 error
= zfs_sb_create(osname
, &zsb
);
1113 if ((error
= dsl_prop_get_integer(osname
, "recordsize",
1114 &recordsize
, NULL
)))
1118 sb
->s_fs_info
= zsb
;
1119 sb
->s_magic
= ZFS_SUPER_MAGIC
;
1120 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1121 sb
->s_time_gran
= 1;
1122 sb
->s_blocksize
= recordsize
;
1123 sb
->s_blocksize_bits
= ilog2(recordsize
);
1127 * 2.6.32 API change,
1128 * Added backing_device_info (BDI) per super block interfaces. A BDI
1129 * must be configured when using a non-device backed filesystem for
1130 * proper writeback. This is not required for older pdflush kernels.
1132 * NOTE: Linux read-ahead is disabled in favor of zfs read-ahead.
1134 zsb
->z_bdi
.ra_pages
= 0;
1135 sb
->s_bdi
= &zsb
->z_bdi
;
1137 error
= -bdi_setup_and_register(&zsb
->z_bdi
, "zfs", BDI_CAP_MAP_COPY
);
1140 #endif /* HAVE_BDI */
1142 /* Set callback operations for the file system. */
1143 sb
->s_op
= &zpl_super_operations
;
1144 sb
->s_xattr
= zpl_xattr_handlers
;
1145 sb
->s_export_op
= &zpl_export_operations
;
1147 /* Set features for file system. */
1148 zfs_set_fuid_feature(zsb
);
1150 if (dmu_objset_is_snapshot(zsb
->z_os
)) {
1153 atime_changed_cb(zsb
, B_FALSE
);
1154 readonly_changed_cb(zsb
, B_TRUE
);
1155 if ((error
= dsl_prop_get_integer(osname
,"xattr",&pval
,NULL
)))
1157 xattr_changed_cb(zsb
, pval
);
1158 zsb
->z_issnap
= B_TRUE
;
1159 zsb
->z_os
->os_sync
= ZFS_SYNC_DISABLED
;
1161 mutex_enter(&zsb
->z_os
->os_user_ptr_lock
);
1162 dmu_objset_set_user(zsb
->z_os
, zsb
);
1163 mutex_exit(&zsb
->z_os
->os_user_ptr_lock
);
1165 error
= zfs_sb_setup(zsb
, B_TRUE
);
1166 #ifdef HAVE_SNAPSHOT
1167 (void) zfs_snap_create(zsb
);
1168 #endif /* HAVE_SNAPSHOT */
1171 /* Allocate a root inode for the filesystem. */
1172 error
= zfs_root(zsb
, &root_inode
);
1174 (void) zfs_umount(sb
);
1178 /* Allocate a root dentry for the filesystem */
1179 sb
->s_root
= d_alloc_root(root_inode
);
1180 if (sb
->s_root
== NULL
) {
1181 (void) zfs_umount(sb
);
1187 dmu_objset_disown(zsb
->z_os
, zsb
);
1193 EXPORT_SYMBOL(zfs_domount
);
1197 zfs_umount(struct super_block
*sb
)
1199 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1202 VERIFY(zfs_sb_teardown(zsb
, B_TRUE
) == 0);
1206 bdi_destroy(sb
->s_bdi
);
1207 #endif /* HAVE_BDI */
1210 * z_os will be NULL if there was an error in
1211 * attempting to reopen zsb.
1215 * Unset the objset user_ptr.
1217 mutex_enter(&os
->os_user_ptr_lock
);
1218 dmu_objset_set_user(os
, NULL
);
1219 mutex_exit(&os
->os_user_ptr_lock
);
1222 * Finally release the objset
1224 dmu_objset_disown(os
, zsb
);
1230 EXPORT_SYMBOL(zfs_umount
);
1233 zfs_remount(struct super_block
*sb
, int *flags
, char *data
)
1236 * All namespace flags (MNT_*) and super block flags (MS_*) will
1237 * be handled by the Linux VFS. Only handle custom options here.
1241 EXPORT_SYMBOL(zfs_remount
);
1244 zfs_vget(struct super_block
*sb
, struct inode
**ipp
, fid_t
*fidp
)
1246 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1248 uint64_t object
= 0;
1249 uint64_t fid_gen
= 0;
1258 if (fidp
->fid_len
== LONG_FID_LEN
) {
1259 zfid_long_t
*zlfid
= (zfid_long_t
*)fidp
;
1260 uint64_t objsetid
= 0;
1261 uint64_t setgen
= 0;
1263 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
1264 objsetid
|= ((uint64_t)zlfid
->zf_setid
[i
]) << (8 * i
);
1266 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
1267 setgen
|= ((uint64_t)zlfid
->zf_setgen
[i
]) << (8 * i
);
1271 #ifdef HAVE_SNAPSHOT
1272 err
= zfsctl_lookup_objset(vfsp
, objsetid
, &zsb
);
1275 #endif /* HAVE_SNAPSHOT */
1279 if (fidp
->fid_len
== SHORT_FID_LEN
|| fidp
->fid_len
== LONG_FID_LEN
) {
1280 zfid_short_t
*zfid
= (zfid_short_t
*)fidp
;
1282 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
1283 object
|= ((uint64_t)zfid
->zf_object
[i
]) << (8 * i
);
1285 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
1286 fid_gen
|= ((uint64_t)zfid
->zf_gen
[i
]) << (8 * i
);
1292 #ifdef HAVE_SNAPSHOT
1293 /* A zero fid_gen means we are in the .zfs control directories */
1295 (object
== ZFSCTL_INO_ROOT
|| object
== ZFSCTL_INO_SNAPDIR
)) {
1296 *ipp
= zsb
->z_ctldir
;
1297 ASSERT(*ipp
!= NULL
);
1298 if (object
== ZFSCTL_INO_SNAPDIR
) {
1299 VERIFY(zfsctl_root_lookup(*ipp
, "snapshot", ipp
, NULL
,
1300 0, NULL
, NULL
, NULL
, NULL
, NULL
) == 0);
1307 #endif /* HAVE_SNAPSHOT */
1309 gen_mask
= -1ULL >> (64 - 8 * i
);
1311 dprintf("getting %llu [%u mask %llx]\n", object
, fid_gen
, gen_mask
);
1312 if ((err
= zfs_zget(zsb
, object
, &zp
))) {
1316 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zsb
), &zp_gen
,
1318 zp_gen
= zp_gen
& gen_mask
;
1321 if (zp
->z_unlinked
|| zp_gen
!= fid_gen
) {
1322 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen
, fid_gen
);
1330 zfs_inode_update(ITOZ(*ipp
));
1335 EXPORT_SYMBOL(zfs_vget
);
1338 * Block out VOPs and close zfs_sb_t::z_os
1340 * Note, if successful, then we return with the 'z_teardown_lock' and
1341 * 'z_teardown_inactive_lock' write held.
1344 zfs_suspend_fs(zfs_sb_t
*zsb
)
1348 if ((error
= zfs_sb_teardown(zsb
, B_FALSE
)) != 0)
1350 dmu_objset_disown(zsb
->z_os
, zsb
);
1354 EXPORT_SYMBOL(zfs_suspend_fs
);
1357 * Reopen zfs_sb_t::z_os and release VOPs.
1360 zfs_resume_fs(zfs_sb_t
*zsb
, const char *osname
)
1364 ASSERT(RRW_WRITE_HELD(&zsb
->z_teardown_lock
));
1365 ASSERT(RW_WRITE_HELD(&zsb
->z_teardown_inactive_lock
));
1367 err
= dmu_objset_own(osname
, DMU_OST_ZFS
, B_FALSE
, zsb
, &zsb
->z_os
);
1372 uint64_t sa_obj
= 0;
1374 err2
= zap_lookup(zsb
->z_os
, MASTER_NODE_OBJ
,
1375 ZFS_SA_ATTRS
, 8, 1, &sa_obj
);
1377 if ((err
|| err2
) && zsb
->z_version
>= ZPL_VERSION_SA
)
1381 if ((err
= sa_setup(zsb
->z_os
, sa_obj
,
1382 zfs_attr_table
, ZPL_END
, &zsb
->z_attr_table
)) != 0)
1385 VERIFY(zfs_sb_setup(zsb
, B_FALSE
) == 0);
1388 * Attempt to re-establish all the active znodes with
1389 * their dbufs. If a zfs_rezget() fails, then we'll let
1390 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
1391 * when they try to use their znode.
1393 mutex_enter(&zsb
->z_znodes_lock
);
1394 for (zp
= list_head(&zsb
->z_all_znodes
); zp
;
1395 zp
= list_next(&zsb
->z_all_znodes
, zp
)) {
1396 (void) zfs_rezget(zp
);
1398 mutex_exit(&zsb
->z_znodes_lock
);
1403 /* release the VOPs */
1404 rw_exit(&zsb
->z_teardown_inactive_lock
);
1405 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1409 * Since we couldn't reopen zfs_sb_t::z_os, force
1410 * unmount this file system.
1412 (void) zfs_umount(zsb
->z_sb
);
1416 EXPORT_SYMBOL(zfs_resume_fs
);
1419 zfs_set_version(zfs_sb_t
*zsb
, uint64_t newvers
)
1422 objset_t
*os
= zsb
->z_os
;
1425 if (newvers
< ZPL_VERSION_INITIAL
|| newvers
> ZPL_VERSION
)
1428 if (newvers
< zsb
->z_version
)
1431 if (zfs_spa_version_map(newvers
) >
1432 spa_version(dmu_objset_spa(zsb
->z_os
)))
1435 tx
= dmu_tx_create(os
);
1436 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_FALSE
, ZPL_VERSION_STR
);
1437 if (newvers
>= ZPL_VERSION_SA
&& !zsb
->z_use_sa
) {
1438 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
1440 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1442 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1448 error
= zap_update(os
, MASTER_NODE_OBJ
, ZPL_VERSION_STR
,
1449 8, 1, &newvers
, tx
);
1456 if (newvers
>= ZPL_VERSION_SA
&& !zsb
->z_use_sa
) {
1459 ASSERT3U(spa_version(dmu_objset_spa(zsb
->z_os
)), >=,
1461 sa_obj
= zap_create(os
, DMU_OT_SA_MASTER_NODE
,
1462 DMU_OT_NONE
, 0, tx
);
1464 error
= zap_add(os
, MASTER_NODE_OBJ
,
1465 ZFS_SA_ATTRS
, 8, 1, &sa_obj
, tx
);
1466 ASSERT3U(error
, ==, 0);
1468 VERIFY(0 == sa_set_sa_object(os
, sa_obj
));
1469 sa_register_update_callback(os
, zfs_sa_upgrade
);
1472 spa_history_log_internal(LOG_DS_UPGRADE
,
1473 dmu_objset_spa(os
), tx
, "oldver=%llu newver=%llu dataset = %llu",
1474 zsb
->z_version
, newvers
, dmu_objset_id(os
));
1478 zsb
->z_version
= newvers
;
1480 if (zsb
->z_version
>= ZPL_VERSION_FUID
)
1481 zfs_set_fuid_feature(zsb
);
1485 EXPORT_SYMBOL(zfs_set_version
);
1488 * Read a property stored within the master node.
1491 zfs_get_zplprop(objset_t
*os
, zfs_prop_t prop
, uint64_t *value
)
1497 * Look up the file system's value for the property. For the
1498 * version property, we look up a slightly different string.
1500 if (prop
== ZFS_PROP_VERSION
)
1501 pname
= ZPL_VERSION_STR
;
1503 pname
= zfs_prop_to_name(prop
);
1506 error
= zap_lookup(os
, MASTER_NODE_OBJ
, pname
, 8, 1, value
);
1508 if (error
== ENOENT
) {
1509 /* No value set, use the default value */
1511 case ZFS_PROP_VERSION
:
1512 *value
= ZPL_VERSION
;
1514 case ZFS_PROP_NORMALIZE
:
1515 case ZFS_PROP_UTF8ONLY
:
1519 *value
= ZFS_CASE_SENSITIVE
;
1528 EXPORT_SYMBOL(zfs_get_zplprop
);
1534 dmu_objset_register_type(DMU_OST_ZFS
, zfs_space_delta_cb
);
1535 register_filesystem(&zpl_fs_type
);
1541 unregister_filesystem(&zpl_fs_type
);