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.
23 * Copyright (c) 2013 by Delphix. All rights reserved.
26 /* Portions Copyright 2010 Robert Milkowski */
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/sysmacros.h>
33 #include <sys/pathname.h>
34 #include <sys/vnode.h>
36 #include <sys/vfs_opreg.h>
37 #include <sys/mntent.h>
38 #include <sys/mount.h>
39 #include <sys/cmn_err.h>
40 #include "fs/fs_subr.h"
41 #include <sys/zfs_znode.h>
42 #include <sys/zfs_vnops.h>
43 #include <sys/zfs_dir.h>
45 #include <sys/fs/zfs.h>
47 #include <sys/dsl_prop.h>
48 #include <sys/dsl_dataset.h>
49 #include <sys/dsl_deleg.h>
53 #include <sys/sa_impl.h>
54 #include <sys/varargs.h>
55 #include <sys/policy.h>
56 #include <sys/atomic.h>
57 #include <sys/mkdev.h>
58 #include <sys/modctl.h>
59 #include <sys/refstr.h>
60 #include <sys/zfs_ioctl.h>
61 #include <sys/zfs_ctldir.h>
62 #include <sys/zfs_fuid.h>
63 #include <sys/bootconf.h>
64 #include <sys/sunddi.h>
66 #include <sys/dmu_objset.h>
67 #include <sys/spa_boot.h>
69 #include "zfs_comutil.h"
74 zfs_sync(struct super_block
*sb
, int wait
, cred_t
*cr
)
76 zfs_sb_t
*zsb
= sb
->s_fs_info
;
79 * Data integrity is job one. We don't want a compromised kernel
80 * writing to the storage pool, so we never sync during panic.
82 if (unlikely(oops_in_progress
))
86 * Semantically, the only requirement is that the sync be initiated.
87 * The DMU syncs out txgs frequently, so there's nothing to do.
94 * Sync a specific filesystem.
99 dp
= dmu_objset_pool(zsb
->z_os
);
102 * If the system is shutting down, then skip any
103 * filesystems which may exist on a suspended pool.
105 if (spa_suspended(dp
->dp_spa
)) {
110 if (zsb
->z_log
!= NULL
)
111 zil_commit(zsb
->z_log
, 0);
116 * Sync all ZFS filesystems. This is what happens when you
117 * run sync(1M). Unlike other filesystems, ZFS honors the
118 * request by waiting for all pools to commit all dirty data.
125 EXPORT_SYMBOL(zfs_sync
);
128 zfs_is_readonly(zfs_sb_t
*zsb
)
130 return (!!(zsb
->z_sb
->s_flags
& MS_RDONLY
));
132 EXPORT_SYMBOL(zfs_is_readonly
);
135 atime_changed_cb(void *arg
, uint64_t newval
)
137 ((zfs_sb_t
*)arg
)->z_atime
= newval
;
141 xattr_changed_cb(void *arg
, uint64_t newval
)
145 if (newval
== ZFS_XATTR_OFF
) {
146 zsb
->z_flags
&= ~ZSB_XATTR
;
148 zsb
->z_flags
|= ZSB_XATTR
;
150 if (newval
== ZFS_XATTR_SA
)
151 zsb
->z_xattr_sa
= B_TRUE
;
153 zsb
->z_xattr_sa
= B_FALSE
;
158 acltype_changed_cb(void *arg
, uint64_t newval
)
163 case ZFS_ACLTYPE_OFF
:
164 zsb
->z_acl_type
= ZFS_ACLTYPE_OFF
;
165 zsb
->z_sb
->s_flags
&= ~MS_POSIXACL
;
167 case ZFS_ACLTYPE_POSIXACL
:
168 zsb
->z_acl_type
= ZFS_ACLTYPE_POSIXACL
;
169 zsb
->z_sb
->s_flags
|= MS_POSIXACL
;
177 blksz_changed_cb(void *arg
, uint64_t newval
)
181 if (newval
< SPA_MINBLOCKSIZE
||
182 newval
> SPA_MAXBLOCKSIZE
|| !ISP2(newval
))
183 newval
= SPA_MAXBLOCKSIZE
;
185 zsb
->z_max_blksz
= newval
;
189 readonly_changed_cb(void *arg
, uint64_t newval
)
192 struct super_block
*sb
= zsb
->z_sb
;
198 sb
->s_flags
|= MS_RDONLY
;
200 sb
->s_flags
&= ~MS_RDONLY
;
204 devices_changed_cb(void *arg
, uint64_t newval
)
209 setuid_changed_cb(void *arg
, uint64_t newval
)
214 exec_changed_cb(void *arg
, uint64_t newval
)
219 nbmand_changed_cb(void *arg
, uint64_t newval
)
222 struct super_block
*sb
= zsb
->z_sb
;
228 sb
->s_flags
|= MS_MANDLOCK
;
230 sb
->s_flags
&= ~MS_MANDLOCK
;
234 snapdir_changed_cb(void *arg
, uint64_t newval
)
236 ((zfs_sb_t
*)arg
)->z_show_ctldir
= newval
;
240 vscan_changed_cb(void *arg
, uint64_t newval
)
242 ((zfs_sb_t
*)arg
)->z_vscan
= newval
;
246 acl_inherit_changed_cb(void *arg
, uint64_t newval
)
248 ((zfs_sb_t
*)arg
)->z_acl_inherit
= newval
;
252 zfs_register_callbacks(zfs_sb_t
*zsb
)
254 struct dsl_dataset
*ds
= NULL
;
255 objset_t
*os
= zsb
->z_os
;
256 boolean_t do_readonly
= B_FALSE
;
259 if (zfs_is_readonly(zsb
) || !spa_writeable(dmu_objset_spa(os
)))
260 do_readonly
= B_TRUE
;
263 * Register property callbacks.
265 * It would probably be fine to just check for i/o error from
266 * the first prop_register(), but I guess I like to go
269 ds
= dmu_objset_ds(os
);
270 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
271 error
= dsl_prop_register(ds
,
272 zfs_prop_to_name(ZFS_PROP_ATIME
), atime_changed_cb
, zsb
);
273 error
= error
? error
: dsl_prop_register(ds
,
274 zfs_prop_to_name(ZFS_PROP_XATTR
), xattr_changed_cb
, zsb
);
275 error
= error
? error
: dsl_prop_register(ds
,
276 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
), blksz_changed_cb
, zsb
);
277 error
= error
? error
: dsl_prop_register(ds
,
278 zfs_prop_to_name(ZFS_PROP_READONLY
), readonly_changed_cb
, zsb
);
279 error
= error
? error
: dsl_prop_register(ds
,
280 zfs_prop_to_name(ZFS_PROP_DEVICES
), devices_changed_cb
, zsb
);
281 error
= error
? error
: dsl_prop_register(ds
,
282 zfs_prop_to_name(ZFS_PROP_SETUID
), setuid_changed_cb
, zsb
);
283 error
= error
? error
: dsl_prop_register(ds
,
284 zfs_prop_to_name(ZFS_PROP_EXEC
), exec_changed_cb
, zsb
);
285 error
= error
? error
: dsl_prop_register(ds
,
286 zfs_prop_to_name(ZFS_PROP_SNAPDIR
), snapdir_changed_cb
, zsb
);
287 error
= error
? error
: dsl_prop_register(ds
,
288 zfs_prop_to_name(ZFS_PROP_ACLTYPE
), acltype_changed_cb
, zsb
);
289 error
= error
? error
: dsl_prop_register(ds
,
290 zfs_prop_to_name(ZFS_PROP_ACLINHERIT
), acl_inherit_changed_cb
, zsb
);
291 error
= error
? error
: dsl_prop_register(ds
,
292 zfs_prop_to_name(ZFS_PROP_VSCAN
), vscan_changed_cb
, zsb
);
293 error
= error
? error
: dsl_prop_register(ds
,
294 zfs_prop_to_name(ZFS_PROP_NBMAND
), nbmand_changed_cb
, zsb
);
295 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
300 readonly_changed_cb(zsb
, B_TRUE
);
306 * We may attempt to unregister some callbacks that are not
307 * registered, but this is OK; it will simply return ENOMSG,
308 * which we will ignore.
310 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_ATIME
),
311 atime_changed_cb
, zsb
);
312 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_XATTR
),
313 xattr_changed_cb
, zsb
);
314 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
315 blksz_changed_cb
, zsb
);
316 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_READONLY
),
317 readonly_changed_cb
, zsb
);
318 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_DEVICES
),
319 devices_changed_cb
, zsb
);
320 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_SETUID
),
321 setuid_changed_cb
, zsb
);
322 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_EXEC
),
323 exec_changed_cb
, zsb
);
324 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_SNAPDIR
),
325 snapdir_changed_cb
, zsb
);
326 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_ACLTYPE
),
327 acltype_changed_cb
, zsb
);
328 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_ACLINHERIT
),
329 acl_inherit_changed_cb
, zsb
);
330 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_VSCAN
),
331 vscan_changed_cb
, zsb
);
332 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_NBMAND
),
333 nbmand_changed_cb
, zsb
);
337 EXPORT_SYMBOL(zfs_register_callbacks
);
340 zfs_space_delta_cb(dmu_object_type_t bonustype
, void *data
,
341 uint64_t *userp
, uint64_t *groupp
)
344 * Is it a valid type of object to track?
346 if (bonustype
!= DMU_OT_ZNODE
&& bonustype
!= DMU_OT_SA
)
347 return (SET_ERROR(ENOENT
));
350 * If we have a NULL data pointer
351 * then assume the id's aren't changing and
352 * return EEXIST to the dmu to let it know to
356 return (SET_ERROR(EEXIST
));
358 if (bonustype
== DMU_OT_ZNODE
) {
359 znode_phys_t
*znp
= data
;
360 *userp
= znp
->zp_uid
;
361 *groupp
= znp
->zp_gid
;
364 sa_hdr_phys_t
*sap
= data
;
365 sa_hdr_phys_t sa
= *sap
;
366 boolean_t swap
= B_FALSE
;
368 ASSERT(bonustype
== DMU_OT_SA
);
370 if (sa
.sa_magic
== 0) {
372 * This should only happen for newly created
373 * files that haven't had the znode data filled
380 if (sa
.sa_magic
== BSWAP_32(SA_MAGIC
)) {
381 sa
.sa_magic
= SA_MAGIC
;
382 sa
.sa_layout_info
= BSWAP_16(sa
.sa_layout_info
);
385 VERIFY3U(sa
.sa_magic
, ==, SA_MAGIC
);
388 hdrsize
= sa_hdrsize(&sa
);
389 VERIFY3U(hdrsize
, >=, sizeof (sa_hdr_phys_t
));
390 *userp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+
392 *groupp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+
395 *userp
= BSWAP_64(*userp
);
396 *groupp
= BSWAP_64(*groupp
);
403 fuidstr_to_sid(zfs_sb_t
*zsb
, const char *fuidstr
,
404 char *domainbuf
, int buflen
, uid_t
*ridp
)
409 fuid
= strtonum(fuidstr
, NULL
);
411 domain
= zfs_fuid_find_by_idx(zsb
, FUID_INDEX(fuid
));
413 (void) strlcpy(domainbuf
, domain
, buflen
);
416 *ridp
= FUID_RID(fuid
);
420 zfs_userquota_prop_to_obj(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
)
423 case ZFS_PROP_USERUSED
:
424 return (DMU_USERUSED_OBJECT
);
425 case ZFS_PROP_GROUPUSED
:
426 return (DMU_GROUPUSED_OBJECT
);
427 case ZFS_PROP_USERQUOTA
:
428 return (zsb
->z_userquota_obj
);
429 case ZFS_PROP_GROUPQUOTA
:
430 return (zsb
->z_groupquota_obj
);
432 return (SET_ERROR(ENOTSUP
));
438 zfs_userspace_many(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
439 uint64_t *cookiep
, void *vbuf
, uint64_t *bufsizep
)
444 zfs_useracct_t
*buf
= vbuf
;
447 if (!dmu_objset_userspace_present(zsb
->z_os
))
448 return (SET_ERROR(ENOTSUP
));
450 obj
= zfs_userquota_prop_to_obj(zsb
, type
);
456 for (zap_cursor_init_serialized(&zc
, zsb
->z_os
, obj
, *cookiep
);
457 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
458 zap_cursor_advance(&zc
)) {
459 if ((uintptr_t)buf
- (uintptr_t)vbuf
+ sizeof (zfs_useracct_t
) >
463 fuidstr_to_sid(zsb
, za
.za_name
,
464 buf
->zu_domain
, sizeof (buf
->zu_domain
), &buf
->zu_rid
);
466 buf
->zu_space
= za
.za_first_integer
;
472 ASSERT3U((uintptr_t)buf
- (uintptr_t)vbuf
, <=, *bufsizep
);
473 *bufsizep
= (uintptr_t)buf
- (uintptr_t)vbuf
;
474 *cookiep
= zap_cursor_serialize(&zc
);
475 zap_cursor_fini(&zc
);
478 EXPORT_SYMBOL(zfs_userspace_many
);
481 * buf must be big enough (eg, 32 bytes)
484 id_to_fuidstr(zfs_sb_t
*zsb
, const char *domain
, uid_t rid
,
485 char *buf
, boolean_t addok
)
490 if (domain
&& domain
[0]) {
491 domainid
= zfs_fuid_find_by_domain(zsb
, domain
, NULL
, addok
);
493 return (SET_ERROR(ENOENT
));
495 fuid
= FUID_ENCODE(domainid
, rid
);
496 (void) sprintf(buf
, "%llx", (longlong_t
)fuid
);
501 zfs_userspace_one(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
502 const char *domain
, uint64_t rid
, uint64_t *valp
)
510 if (!dmu_objset_userspace_present(zsb
->z_os
))
511 return (SET_ERROR(ENOTSUP
));
513 obj
= zfs_userquota_prop_to_obj(zsb
, type
);
517 err
= id_to_fuidstr(zsb
, domain
, rid
, buf
, B_FALSE
);
521 err
= zap_lookup(zsb
->z_os
, obj
, buf
, 8, 1, valp
);
526 EXPORT_SYMBOL(zfs_userspace_one
);
529 zfs_set_userquota(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
530 const char *domain
, uint64_t rid
, uint64_t quota
)
536 boolean_t fuid_dirtied
;
538 if (type
!= ZFS_PROP_USERQUOTA
&& type
!= ZFS_PROP_GROUPQUOTA
)
539 return (SET_ERROR(EINVAL
));
541 if (zsb
->z_version
< ZPL_VERSION_USERSPACE
)
542 return (SET_ERROR(ENOTSUP
));
544 objp
= (type
== ZFS_PROP_USERQUOTA
) ? &zsb
->z_userquota_obj
:
545 &zsb
->z_groupquota_obj
;
547 err
= id_to_fuidstr(zsb
, domain
, rid
, buf
, B_TRUE
);
550 fuid_dirtied
= zsb
->z_fuid_dirty
;
552 tx
= dmu_tx_create(zsb
->z_os
);
553 dmu_tx_hold_zap(tx
, *objp
? *objp
: DMU_NEW_OBJECT
, B_TRUE
, NULL
);
555 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
556 zfs_userquota_prop_prefixes
[type
]);
559 zfs_fuid_txhold(zsb
, tx
);
560 err
= dmu_tx_assign(tx
, TXG_WAIT
);
566 mutex_enter(&zsb
->z_lock
);
568 *objp
= zap_create(zsb
->z_os
, DMU_OT_USERGROUP_QUOTA
,
570 VERIFY(0 == zap_add(zsb
->z_os
, MASTER_NODE_OBJ
,
571 zfs_userquota_prop_prefixes
[type
], 8, 1, objp
, tx
));
573 mutex_exit(&zsb
->z_lock
);
576 err
= zap_remove(zsb
->z_os
, *objp
, buf
, tx
);
580 err
= zap_update(zsb
->z_os
, *objp
, buf
, 8, 1, "a
, tx
);
584 zfs_fuid_sync(zsb
, tx
);
588 EXPORT_SYMBOL(zfs_set_userquota
);
591 zfs_fuid_overquota(zfs_sb_t
*zsb
, boolean_t isgroup
, uint64_t fuid
)
594 uint64_t used
, quota
, usedobj
, quotaobj
;
597 usedobj
= isgroup
? DMU_GROUPUSED_OBJECT
: DMU_USERUSED_OBJECT
;
598 quotaobj
= isgroup
? zsb
->z_groupquota_obj
: zsb
->z_userquota_obj
;
600 if (quotaobj
== 0 || zsb
->z_replay
)
603 (void) sprintf(buf
, "%llx", (longlong_t
)fuid
);
604 err
= zap_lookup(zsb
->z_os
, quotaobj
, buf
, 8, 1, "a
);
608 err
= zap_lookup(zsb
->z_os
, usedobj
, buf
, 8, 1, &used
);
611 return (used
>= quota
);
613 EXPORT_SYMBOL(zfs_fuid_overquota
);
616 zfs_owner_overquota(zfs_sb_t
*zsb
, znode_t
*zp
, boolean_t isgroup
)
621 quotaobj
= isgroup
? zsb
->z_groupquota_obj
: zsb
->z_userquota_obj
;
623 fuid
= isgroup
? zp
->z_gid
: zp
->z_uid
;
625 if (quotaobj
== 0 || zsb
->z_replay
)
628 return (zfs_fuid_overquota(zsb
, isgroup
, fuid
));
630 EXPORT_SYMBOL(zfs_owner_overquota
);
633 zfs_sb_create(const char *osname
, zfs_sb_t
**zsbp
)
641 zsb
= kmem_zalloc(sizeof (zfs_sb_t
), KM_SLEEP
| KM_NODEBUG
);
644 * We claim to always be readonly so we can open snapshots;
645 * other ZPL code will prevent us from writing to snapshots.
647 error
= dmu_objset_own(osname
, DMU_OST_ZFS
, B_TRUE
, zsb
, &os
);
649 kmem_free(zsb
, sizeof (zfs_sb_t
));
654 * Initialize the zfs-specific filesystem structure.
655 * Should probably make this a kmem cache, shuffle fields,
656 * and just bzero up to z_hold_mtx[].
660 zsb
->z_max_blksz
= SPA_MAXBLOCKSIZE
;
661 zsb
->z_show_ctldir
= ZFS_SNAPDIR_VISIBLE
;
664 error
= zfs_get_zplprop(os
, ZFS_PROP_VERSION
, &zsb
->z_version
);
667 } else if (zsb
->z_version
>
668 zfs_zpl_version_map(spa_version(dmu_objset_spa(os
)))) {
669 (void) printk("Can't mount a version %lld file system "
670 "on a version %lld pool\n. Pool must be upgraded to mount "
671 "this file system.", (u_longlong_t
)zsb
->z_version
,
672 (u_longlong_t
)spa_version(dmu_objset_spa(os
)));
673 error
= SET_ERROR(ENOTSUP
);
676 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_NORMALIZE
, &zval
)) != 0)
678 zsb
->z_norm
= (int)zval
;
680 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_UTF8ONLY
, &zval
)) != 0)
682 zsb
->z_utf8
= (zval
!= 0);
684 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_CASE
, &zval
)) != 0)
686 zsb
->z_case
= (uint_t
)zval
;
688 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_ACLTYPE
, &zval
)) != 0)
690 zsb
->z_acl_type
= (uint_t
)zval
;
693 * Fold case on file systems that are always or sometimes case
696 if (zsb
->z_case
== ZFS_CASE_INSENSITIVE
||
697 zsb
->z_case
== ZFS_CASE_MIXED
)
698 zsb
->z_norm
|= U8_TEXTPREP_TOUPPER
;
700 zsb
->z_use_fuids
= USE_FUIDS(zsb
->z_version
, zsb
->z_os
);
701 zsb
->z_use_sa
= USE_SA(zsb
->z_version
, zsb
->z_os
);
704 /* should either have both of these objects or none */
705 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SA_ATTRS
, 8, 1,
710 error
= zfs_get_zplprop(os
, ZFS_PROP_XATTR
, &zval
);
711 if ((error
== 0) && (zval
== ZFS_XATTR_SA
))
712 zsb
->z_xattr_sa
= B_TRUE
;
715 * Pre SA versions file systems should never touch
716 * either the attribute registration or layout objects.
721 error
= sa_setup(os
, sa_obj
, zfs_attr_table
, ZPL_END
,
726 if (zsb
->z_version
>= ZPL_VERSION_SA
)
727 sa_register_update_callback(os
, zfs_sa_upgrade
);
729 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_ROOT_OBJ
, 8, 1,
733 ASSERT(zsb
->z_root
!= 0);
735 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_UNLINKED_SET
, 8, 1,
736 &zsb
->z_unlinkedobj
);
740 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
741 zfs_userquota_prop_prefixes
[ZFS_PROP_USERQUOTA
],
742 8, 1, &zsb
->z_userquota_obj
);
743 if (error
&& error
!= ENOENT
)
746 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
747 zfs_userquota_prop_prefixes
[ZFS_PROP_GROUPQUOTA
],
748 8, 1, &zsb
->z_groupquota_obj
);
749 if (error
&& error
!= ENOENT
)
752 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_FUID_TABLES
, 8, 1,
754 if (error
&& error
!= ENOENT
)
757 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SHARES_DIR
, 8, 1,
759 if (error
&& error
!= ENOENT
)
762 mutex_init(&zsb
->z_znodes_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
763 mutex_init(&zsb
->z_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
764 list_create(&zsb
->z_all_znodes
, sizeof (znode_t
),
765 offsetof(znode_t
, z_link_node
));
766 rrw_init(&zsb
->z_teardown_lock
, B_FALSE
);
767 rw_init(&zsb
->z_teardown_inactive_lock
, NULL
, RW_DEFAULT
, NULL
);
768 rw_init(&zsb
->z_fuid_lock
, NULL
, RW_DEFAULT
, NULL
);
769 for (i
= 0; i
!= ZFS_OBJ_MTX_SZ
; i
++)
770 mutex_init(&zsb
->z_hold_mtx
[i
], NULL
, MUTEX_DEFAULT
, NULL
);
772 avl_create(&zsb
->z_ctldir_snaps
, snapentry_compare
,
773 sizeof (zfs_snapentry_t
), offsetof(zfs_snapentry_t
, se_node
));
774 mutex_init(&zsb
->z_ctldir_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
780 dmu_objset_disown(os
, zsb
);
782 kmem_free(zsb
, sizeof (zfs_sb_t
));
785 EXPORT_SYMBOL(zfs_sb_create
);
788 zfs_sb_setup(zfs_sb_t
*zsb
, boolean_t mounting
)
792 error
= zfs_register_callbacks(zsb
);
797 * Set the objset user_ptr to track its zsb.
799 mutex_enter(&zsb
->z_os
->os_user_ptr_lock
);
800 dmu_objset_set_user(zsb
->z_os
, zsb
);
801 mutex_exit(&zsb
->z_os
->os_user_ptr_lock
);
803 zsb
->z_log
= zil_open(zsb
->z_os
, zfs_get_data
);
806 * If we are not mounting (ie: online recv), then we don't
807 * have to worry about replaying the log as we blocked all
808 * operations out since we closed the ZIL.
814 * During replay we remove the read only flag to
815 * allow replays to succeed.
817 readonly
= zfs_is_readonly(zsb
);
819 readonly_changed_cb(zsb
, B_FALSE
);
821 zfs_unlinked_drain(zsb
);
824 * Parse and replay the intent log.
826 * Because of ziltest, this must be done after
827 * zfs_unlinked_drain(). (Further note: ziltest
828 * doesn't use readonly mounts, where
829 * zfs_unlinked_drain() isn't called.) This is because
830 * ziltest causes spa_sync() to think it's committed,
831 * but actually it is not, so the intent log contains
832 * many txg's worth of changes.
834 * In particular, if object N is in the unlinked set in
835 * the last txg to actually sync, then it could be
836 * actually freed in a later txg and then reallocated
837 * in a yet later txg. This would write a "create
838 * object N" record to the intent log. Normally, this
839 * would be fine because the spa_sync() would have
840 * written out the fact that object N is free, before
841 * we could write the "create object N" intent log
844 * But when we are in ziltest mode, we advance the "open
845 * txg" without actually spa_sync()-ing the changes to
846 * disk. So we would see that object N is still
847 * allocated and in the unlinked set, and there is an
848 * intent log record saying to allocate it.
850 if (spa_writeable(dmu_objset_spa(zsb
->z_os
))) {
851 if (zil_replay_disable
) {
852 zil_destroy(zsb
->z_log
, B_FALSE
);
854 zsb
->z_replay
= B_TRUE
;
855 zil_replay(zsb
->z_os
, zsb
,
857 zsb
->z_replay
= B_FALSE
;
861 /* restore readonly bit */
863 readonly_changed_cb(zsb
, B_TRUE
);
868 EXPORT_SYMBOL(zfs_sb_setup
);
871 zfs_sb_free(zfs_sb_t
*zsb
)
875 zfs_fuid_destroy(zsb
);
877 mutex_destroy(&zsb
->z_znodes_lock
);
878 mutex_destroy(&zsb
->z_lock
);
879 list_destroy(&zsb
->z_all_znodes
);
880 rrw_destroy(&zsb
->z_teardown_lock
);
881 rw_destroy(&zsb
->z_teardown_inactive_lock
);
882 rw_destroy(&zsb
->z_fuid_lock
);
883 for (i
= 0; i
!= ZFS_OBJ_MTX_SZ
; i
++)
884 mutex_destroy(&zsb
->z_hold_mtx
[i
]);
885 mutex_destroy(&zsb
->z_ctldir_lock
);
886 avl_destroy(&zsb
->z_ctldir_snaps
);
887 kmem_free(zsb
, sizeof (zfs_sb_t
));
889 EXPORT_SYMBOL(zfs_sb_free
);
892 zfs_set_fuid_feature(zfs_sb_t
*zsb
)
894 zsb
->z_use_fuids
= USE_FUIDS(zsb
->z_version
, zsb
->z_os
);
895 zsb
->z_use_sa
= USE_SA(zsb
->z_version
, zsb
->z_os
);
899 zfs_unregister_callbacks(zfs_sb_t
*zsb
)
901 objset_t
*os
= zsb
->z_os
;
902 struct dsl_dataset
*ds
;
905 * Unregister properties.
907 if (!dmu_objset_is_snapshot(os
)) {
908 ds
= dmu_objset_ds(os
);
909 VERIFY(dsl_prop_unregister(ds
, "atime", atime_changed_cb
,
912 VERIFY(dsl_prop_unregister(ds
, "xattr", xattr_changed_cb
,
915 VERIFY(dsl_prop_unregister(ds
, "recordsize", blksz_changed_cb
,
918 VERIFY(dsl_prop_unregister(ds
, "readonly", readonly_changed_cb
,
921 VERIFY(dsl_prop_unregister(ds
, "devices", devices_changed_cb
,
924 VERIFY(dsl_prop_unregister(ds
, "setuid", setuid_changed_cb
,
927 VERIFY(dsl_prop_unregister(ds
, "exec", exec_changed_cb
,
930 VERIFY(dsl_prop_unregister(ds
, "snapdir", snapdir_changed_cb
,
933 VERIFY(dsl_prop_unregister(ds
, "acltype", acltype_changed_cb
,
936 VERIFY(dsl_prop_unregister(ds
, "aclinherit",
937 acl_inherit_changed_cb
, zsb
) == 0);
939 VERIFY(dsl_prop_unregister(ds
, "vscan",
940 vscan_changed_cb
, zsb
) == 0);
942 VERIFY(dsl_prop_unregister(ds
, "nbmand",
943 nbmand_changed_cb
, zsb
) == 0);
946 EXPORT_SYMBOL(zfs_unregister_callbacks
);
950 * zfs_check_global_label:
951 * Check that the hex label string is appropriate for the dataset
952 * being mounted into the global_zone proper.
954 * Return an error if the hex label string is not default or
955 * admin_low/admin_high. For admin_low labels, the corresponding
956 * dataset must be readonly.
959 zfs_check_global_label(const char *dsname
, const char *hexsl
)
961 if (strcasecmp(hexsl
, ZFS_MLSLABEL_DEFAULT
) == 0)
963 if (strcasecmp(hexsl
, ADMIN_HIGH
) == 0)
965 if (strcasecmp(hexsl
, ADMIN_LOW
) == 0) {
966 /* must be readonly */
969 if (dsl_prop_get_integer(dsname
,
970 zfs_prop_to_name(ZFS_PROP_READONLY
), &rdonly
, NULL
))
971 return (SET_ERROR(EACCES
));
972 return (rdonly
? 0 : EACCES
);
974 return (SET_ERROR(EACCES
));
976 EXPORT_SYMBOL(zfs_check_global_label
);
977 #endif /* HAVE_MLSLABEL */
980 zfs_statvfs(struct dentry
*dentry
, struct kstatfs
*statp
)
982 zfs_sb_t
*zsb
= dentry
->d_sb
->s_fs_info
;
983 uint64_t refdbytes
, availbytes
, usedobjs
, availobjs
;
989 dmu_objset_space(zsb
->z_os
,
990 &refdbytes
, &availbytes
, &usedobjs
, &availobjs
);
992 fsid
= dmu_objset_fsid_guid(zsb
->z_os
);
994 * The underlying storage pool actually uses multiple block
995 * size. Under Solaris frsize (fragment size) is reported as
996 * the smallest block size we support, and bsize (block size)
997 * as the filesystem's maximum block size. Unfortunately,
998 * under Linux the fragment size and block size are often used
999 * interchangeably. Thus we are forced to report both of them
1000 * as the filesystem's maximum block size.
1002 statp
->f_frsize
= zsb
->z_max_blksz
;
1003 statp
->f_bsize
= zsb
->z_max_blksz
;
1004 bshift
= fls(statp
->f_bsize
) - 1;
1007 * The following report "total" blocks of various kinds in
1008 * the file system, but reported in terms of f_bsize - the
1012 statp
->f_blocks
= (refdbytes
+ availbytes
) >> bshift
;
1013 statp
->f_bfree
= availbytes
>> bshift
;
1014 statp
->f_bavail
= statp
->f_bfree
; /* no root reservation */
1017 * statvfs() should really be called statufs(), because it assumes
1018 * static metadata. ZFS doesn't preallocate files, so the best
1019 * we can do is report the max that could possibly fit in f_files,
1020 * and that minus the number actually used in f_ffree.
1021 * For f_ffree, report the smaller of the number of object available
1022 * and the number of blocks (each object will take at least a block).
1024 statp
->f_ffree
= MIN(availobjs
, availbytes
>> DNODE_SHIFT
);
1025 statp
->f_files
= statp
->f_ffree
+ usedobjs
;
1026 statp
->f_fsid
.val
[0] = (uint32_t)fsid
;
1027 statp
->f_fsid
.val
[1] = (uint32_t)(fsid
>> 32);
1028 statp
->f_type
= ZFS_SUPER_MAGIC
;
1029 statp
->f_namelen
= ZFS_MAXNAMELEN
;
1032 * We have all of 40 characters to stuff a string here.
1033 * Is there anything useful we could/should provide?
1035 bzero(statp
->f_spare
, sizeof (statp
->f_spare
));
1040 EXPORT_SYMBOL(zfs_statvfs
);
1043 zfs_root(zfs_sb_t
*zsb
, struct inode
**ipp
)
1050 error
= zfs_zget(zsb
, zsb
->z_root
, &rootzp
);
1052 *ipp
= ZTOI(rootzp
);
1057 EXPORT_SYMBOL(zfs_root
);
1061 zfs_sb_prune(struct super_block
*sb
, unsigned long nr_to_scan
, int *objects
)
1063 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1064 struct shrinker
*shrinker
= &sb
->s_shrink
;
1065 struct shrink_control sc
= {
1066 .nr_to_scan
= nr_to_scan
,
1067 .gfp_mask
= GFP_KERNEL
,
1071 *objects
= (*shrinker
->shrink
)(shrinker
, &sc
);
1076 EXPORT_SYMBOL(zfs_sb_prune
);
1077 #endif /* HAVE_SHRINK */
1080 * Teardown the zfs_sb_t.
1082 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1083 * and 'z_teardown_inactive_lock' held.
1086 zfs_sb_teardown(zfs_sb_t
*zsb
, boolean_t unmounting
)
1090 rrw_enter(&zsb
->z_teardown_lock
, RW_WRITER
, FTAG
);
1094 * We purge the parent filesystem's super block as the
1095 * parent filesystem and all of its snapshots have their
1096 * inode's super block set to the parent's filesystem's
1097 * super block. Note, 'z_parent' is self referential
1098 * for non-snapshots.
1100 shrink_dcache_sb(zsb
->z_parent
->z_sb
);
1104 * If someone has not already unmounted this file system,
1105 * drain the iput_taskq to ensure all active references to the
1106 * zfs_sb_t have been handled only then can it be safely destroyed.
1109 taskq_wait(dsl_pool_iput_taskq(dmu_objset_pool(zsb
->z_os
)));
1112 * Close the zil. NB: Can't close the zil while zfs_inactive
1113 * threads are blocked as zil_close can call zfs_inactive.
1116 zil_close(zsb
->z_log
);
1120 rw_enter(&zsb
->z_teardown_inactive_lock
, RW_WRITER
);
1123 * If we are not unmounting (ie: online recv) and someone already
1124 * unmounted this file system while we were doing the switcheroo,
1125 * or a reopen of z_os failed then just bail out now.
1127 if (!unmounting
&& (zsb
->z_unmounted
|| zsb
->z_os
== NULL
)) {
1128 rw_exit(&zsb
->z_teardown_inactive_lock
);
1129 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1130 return (SET_ERROR(EIO
));
1134 * At this point there are no VFS ops active, and any new VFS ops
1135 * will fail with EIO since we have z_teardown_lock for writer (only
1136 * relevant for forced unmount).
1138 * Release all holds on dbufs.
1140 mutex_enter(&zsb
->z_znodes_lock
);
1141 for (zp
= list_head(&zsb
->z_all_znodes
); zp
!= NULL
;
1142 zp
= list_next(&zsb
->z_all_znodes
, zp
)) {
1144 ASSERT(atomic_read(&ZTOI(zp
)->i_count
) > 0);
1145 zfs_znode_dmu_fini(zp
);
1148 mutex_exit(&zsb
->z_znodes_lock
);
1151 * If we are unmounting, set the unmounted flag and let new VFS ops
1152 * unblock. zfs_inactive will have the unmounted behavior, and all
1153 * other VFS ops will fail with EIO.
1156 zsb
->z_unmounted
= B_TRUE
;
1157 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1158 rw_exit(&zsb
->z_teardown_inactive_lock
);
1162 * z_os will be NULL if there was an error in attempting to reopen
1163 * zsb, so just return as the properties had already been
1165 * unregistered and cached data had been evicted before.
1167 if (zsb
->z_os
== NULL
)
1171 * Unregister properties.
1173 zfs_unregister_callbacks(zsb
);
1178 if (dsl_dataset_is_dirty(dmu_objset_ds(zsb
->z_os
)) &&
1179 !zfs_is_readonly(zsb
))
1180 txg_wait_synced(dmu_objset_pool(zsb
->z_os
), 0);
1181 dmu_objset_evict_dbufs(zsb
->z_os
);
1185 EXPORT_SYMBOL(zfs_sb_teardown
);
1187 #if defined(HAVE_BDI) && !defined(HAVE_BDI_SETUP_AND_REGISTER)
1188 atomic_long_t zfs_bdi_seq
= ATOMIC_LONG_INIT(0);
1189 #endif /* HAVE_BDI && !HAVE_BDI_SETUP_AND_REGISTER */
1192 zfs_domount(struct super_block
*sb
, void *data
, int silent
)
1194 zpl_mount_data_t
*zmd
= data
;
1195 const char *osname
= zmd
->z_osname
;
1197 struct inode
*root_inode
;
1198 uint64_t recordsize
;
1201 error
= zfs_sb_create(osname
, &zsb
);
1205 if ((error
= dsl_prop_get_integer(osname
, "recordsize",
1206 &recordsize
, NULL
)))
1210 sb
->s_fs_info
= zsb
;
1211 sb
->s_magic
= ZFS_SUPER_MAGIC
;
1212 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1213 sb
->s_time_gran
= 1;
1214 sb
->s_blocksize
= recordsize
;
1215 sb
->s_blocksize_bits
= ilog2(recordsize
);
1219 * 2.6.32 API change,
1220 * Added backing_device_info (BDI) per super block interfaces. A BDI
1221 * must be configured when using a non-device backed filesystem for
1222 * proper writeback. This is not required for older pdflush kernels.
1224 * NOTE: Linux read-ahead is disabled in favor of zfs read-ahead.
1226 zsb
->z_bdi
.ra_pages
= 0;
1227 sb
->s_bdi
= &zsb
->z_bdi
;
1229 error
= -bdi_setup_and_register(&zsb
->z_bdi
, "zfs", BDI_CAP_MAP_COPY
);
1232 #endif /* HAVE_BDI */
1234 /* Set callback operations for the file system. */
1235 sb
->s_op
= &zpl_super_operations
;
1236 sb
->s_xattr
= zpl_xattr_handlers
;
1237 sb
->s_export_op
= &zpl_export_operations
;
1239 sb
->s_d_op
= &zpl_dentry_operations
;
1240 #endif /* HAVE_S_D_OP */
1242 /* Set features for file system. */
1243 zfs_set_fuid_feature(zsb
);
1245 if (dmu_objset_is_snapshot(zsb
->z_os
)) {
1248 atime_changed_cb(zsb
, B_FALSE
);
1249 readonly_changed_cb(zsb
, B_TRUE
);
1250 if ((error
= dsl_prop_get_integer(osname
,"xattr",&pval
,NULL
)))
1252 xattr_changed_cb(zsb
, pval
);
1253 if ((error
= dsl_prop_get_integer(osname
,"acltype",&pval
,NULL
)))
1255 acltype_changed_cb(zsb
, pval
);
1256 zsb
->z_issnap
= B_TRUE
;
1257 zsb
->z_os
->os_sync
= ZFS_SYNC_DISABLED
;
1259 mutex_enter(&zsb
->z_os
->os_user_ptr_lock
);
1260 dmu_objset_set_user(zsb
->z_os
, zsb
);
1261 mutex_exit(&zsb
->z_os
->os_user_ptr_lock
);
1263 error
= zfs_sb_setup(zsb
, B_TRUE
);
1266 /* Allocate a root inode for the filesystem. */
1267 error
= zfs_root(zsb
, &root_inode
);
1269 (void) zfs_umount(sb
);
1273 /* Allocate a root dentry for the filesystem */
1274 sb
->s_root
= d_make_root(root_inode
);
1275 if (sb
->s_root
== NULL
) {
1276 (void) zfs_umount(sb
);
1277 error
= SET_ERROR(ENOMEM
);
1285 dmu_objset_disown(zsb
->z_os
, zsb
);
1291 EXPORT_SYMBOL(zfs_domount
);
1294 * Called when an unmount is requested and certain sanity checks have
1295 * already passed. At this point no dentries or inodes have been reclaimed
1296 * from their respective caches. We drop the extra reference on the .zfs
1297 * control directory to allow everything to be reclaimed. All snapshots
1298 * must already have been unmounted to reach this point.
1301 zfs_preumount(struct super_block
*sb
)
1303 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1305 if (zsb
!= NULL
&& zsb
->z_ctldir
!= NULL
)
1306 zfsctl_destroy(zsb
);
1308 EXPORT_SYMBOL(zfs_preumount
);
1311 * Called once all other unmount released tear down has occurred.
1312 * It is our responsibility to release any remaining infrastructure.
1316 zfs_umount(struct super_block
*sb
)
1318 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1321 VERIFY(zfs_sb_teardown(zsb
, B_TRUE
) == 0);
1325 bdi_destroy(sb
->s_bdi
);
1326 #endif /* HAVE_BDI */
1329 * z_os will be NULL if there was an error in
1330 * attempting to reopen zsb.
1334 * Unset the objset user_ptr.
1336 mutex_enter(&os
->os_user_ptr_lock
);
1337 dmu_objset_set_user(os
, NULL
);
1338 mutex_exit(&os
->os_user_ptr_lock
);
1341 * Finally release the objset
1343 dmu_objset_disown(os
, zsb
);
1349 EXPORT_SYMBOL(zfs_umount
);
1352 zfs_remount(struct super_block
*sb
, int *flags
, char *data
)
1355 * All namespace flags (MNT_*) and super block flags (MS_*) will
1356 * be handled by the Linux VFS. Only handle custom options here.
1360 EXPORT_SYMBOL(zfs_remount
);
1363 zfs_vget(struct super_block
*sb
, struct inode
**ipp
, fid_t
*fidp
)
1365 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1367 uint64_t object
= 0;
1368 uint64_t fid_gen
= 0;
1377 if (fidp
->fid_len
== LONG_FID_LEN
) {
1378 zfid_long_t
*zlfid
= (zfid_long_t
*)fidp
;
1379 uint64_t objsetid
= 0;
1380 uint64_t setgen
= 0;
1382 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
1383 objsetid
|= ((uint64_t)zlfid
->zf_setid
[i
]) << (8 * i
);
1385 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
1386 setgen
|= ((uint64_t)zlfid
->zf_setgen
[i
]) << (8 * i
);
1390 err
= zfsctl_lookup_objset(sb
, objsetid
, &zsb
);
1392 return (SET_ERROR(EINVAL
));
1397 if (fidp
->fid_len
== SHORT_FID_LEN
|| fidp
->fid_len
== LONG_FID_LEN
) {
1398 zfid_short_t
*zfid
= (zfid_short_t
*)fidp
;
1400 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
1401 object
|= ((uint64_t)zfid
->zf_object
[i
]) << (8 * i
);
1403 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
1404 fid_gen
|= ((uint64_t)zfid
->zf_gen
[i
]) << (8 * i
);
1407 return (SET_ERROR(EINVAL
));
1410 /* A zero fid_gen means we are in the .zfs control directories */
1412 (object
== ZFSCTL_INO_ROOT
|| object
== ZFSCTL_INO_SNAPDIR
)) {
1413 *ipp
= zsb
->z_ctldir
;
1414 ASSERT(*ipp
!= NULL
);
1415 if (object
== ZFSCTL_INO_SNAPDIR
) {
1416 VERIFY(zfsctl_root_lookup(*ipp
, "snapshot", ipp
,
1417 0, kcred
, NULL
, NULL
) == 0);
1425 gen_mask
= -1ULL >> (64 - 8 * i
);
1427 dprintf("getting %llu [%u mask %llx]\n", object
, fid_gen
, gen_mask
);
1428 if ((err
= zfs_zget(zsb
, object
, &zp
))) {
1432 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zsb
), &zp_gen
,
1434 zp_gen
= zp_gen
& gen_mask
;
1437 if (zp
->z_unlinked
|| zp_gen
!= fid_gen
) {
1438 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen
, fid_gen
);
1441 return (SET_ERROR(EINVAL
));
1446 zfs_inode_update(ITOZ(*ipp
));
1451 EXPORT_SYMBOL(zfs_vget
);
1454 * Block out VFS ops and close zfs_sb_t
1456 * Note, if successful, then we return with the 'z_teardown_lock' and
1457 * 'z_teardown_inactive_lock' write held.
1460 zfs_suspend_fs(zfs_sb_t
*zsb
)
1464 if ((error
= zfs_sb_teardown(zsb
, B_FALSE
)) != 0)
1467 dmu_objset_disown(zsb
->z_os
, zsb
);
1471 EXPORT_SYMBOL(zfs_suspend_fs
);
1474 * Reopen zfs_sb_t and release VFS ops.
1477 zfs_resume_fs(zfs_sb_t
*zsb
, const char *osname
)
1481 ASSERT(RRW_WRITE_HELD(&zsb
->z_teardown_lock
));
1482 ASSERT(RW_WRITE_HELD(&zsb
->z_teardown_inactive_lock
));
1484 err
= dmu_objset_own(osname
, DMU_OST_ZFS
, B_FALSE
, zsb
, &zsb
->z_os
);
1489 uint64_t sa_obj
= 0;
1492 * Make sure version hasn't changed
1495 err
= zfs_get_zplprop(zsb
->z_os
, ZFS_PROP_VERSION
,
1501 err
= zap_lookup(zsb
->z_os
, MASTER_NODE_OBJ
,
1502 ZFS_SA_ATTRS
, 8, 1, &sa_obj
);
1504 if (err
&& zsb
->z_version
>= ZPL_VERSION_SA
)
1507 if ((err
= sa_setup(zsb
->z_os
, sa_obj
,
1508 zfs_attr_table
, ZPL_END
, &zsb
->z_attr_table
)) != 0)
1511 if (zsb
->z_version
>= ZPL_VERSION_SA
)
1512 sa_register_update_callback(zsb
->z_os
,
1515 VERIFY(zfs_sb_setup(zsb
, B_FALSE
) == 0);
1517 zfs_set_fuid_feature(zsb
);
1518 zsb
->z_rollback_time
= jiffies
;
1521 * Attempt to re-establish all the active inodes with their
1522 * dbufs. If a zfs_rezget() fails, then we unhash the inode
1523 * and mark it stale. This prevents a collision if a new
1524 * inode/object is created which must use the same inode
1525 * number. The stale inode will be be released when the
1526 * VFS prunes the dentry holding the remaining references
1527 * on the stale inode.
1529 mutex_enter(&zsb
->z_znodes_lock
);
1530 for (zp
= list_head(&zsb
->z_all_znodes
); zp
;
1531 zp
= list_next(&zsb
->z_all_znodes
, zp
)) {
1532 err2
= zfs_rezget(zp
);
1534 remove_inode_hash(ZTOI(zp
));
1535 zp
->z_is_stale
= B_TRUE
;
1538 mutex_exit(&zsb
->z_znodes_lock
);
1542 /* release the VFS ops */
1543 rw_exit(&zsb
->z_teardown_inactive_lock
);
1544 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1548 * Since we couldn't reopen zfs_sb_t or, or
1549 * setup the sa framework force unmount this file system.
1552 (void) zfs_umount(zsb
->z_sb
);
1556 EXPORT_SYMBOL(zfs_resume_fs
);
1559 zfs_set_version(zfs_sb_t
*zsb
, uint64_t newvers
)
1562 objset_t
*os
= zsb
->z_os
;
1565 if (newvers
< ZPL_VERSION_INITIAL
|| newvers
> ZPL_VERSION
)
1566 return (SET_ERROR(EINVAL
));
1568 if (newvers
< zsb
->z_version
)
1569 return (SET_ERROR(EINVAL
));
1571 if (zfs_spa_version_map(newvers
) >
1572 spa_version(dmu_objset_spa(zsb
->z_os
)))
1573 return (SET_ERROR(ENOTSUP
));
1575 tx
= dmu_tx_create(os
);
1576 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_FALSE
, ZPL_VERSION_STR
);
1577 if (newvers
>= ZPL_VERSION_SA
&& !zsb
->z_use_sa
) {
1578 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
1580 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1582 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1588 error
= zap_update(os
, MASTER_NODE_OBJ
, ZPL_VERSION_STR
,
1589 8, 1, &newvers
, tx
);
1596 if (newvers
>= ZPL_VERSION_SA
&& !zsb
->z_use_sa
) {
1599 ASSERT3U(spa_version(dmu_objset_spa(zsb
->z_os
)), >=,
1601 sa_obj
= zap_create(os
, DMU_OT_SA_MASTER_NODE
,
1602 DMU_OT_NONE
, 0, tx
);
1604 error
= zap_add(os
, MASTER_NODE_OBJ
,
1605 ZFS_SA_ATTRS
, 8, 1, &sa_obj
, tx
);
1608 VERIFY(0 == sa_set_sa_object(os
, sa_obj
));
1609 sa_register_update_callback(os
, zfs_sa_upgrade
);
1612 spa_history_log_internal_ds(dmu_objset_ds(os
), "upgrade", tx
,
1613 "from %llu to %llu", zsb
->z_version
, newvers
);
1617 zsb
->z_version
= newvers
;
1619 zfs_set_fuid_feature(zsb
);
1623 EXPORT_SYMBOL(zfs_set_version
);
1626 * Read a property stored within the master node.
1629 zfs_get_zplprop(objset_t
*os
, zfs_prop_t prop
, uint64_t *value
)
1632 int error
= SET_ERROR(ENOENT
);
1635 * Look up the file system's value for the property. For the
1636 * version property, we look up a slightly different string.
1638 if (prop
== ZFS_PROP_VERSION
)
1639 pname
= ZPL_VERSION_STR
;
1641 pname
= zfs_prop_to_name(prop
);
1644 error
= zap_lookup(os
, MASTER_NODE_OBJ
, pname
, 8, 1, value
);
1646 if (error
== ENOENT
) {
1647 /* No value set, use the default value */
1649 case ZFS_PROP_VERSION
:
1650 *value
= ZPL_VERSION
;
1652 case ZFS_PROP_NORMALIZE
:
1653 case ZFS_PROP_UTF8ONLY
:
1657 *value
= ZFS_CASE_SENSITIVE
;
1659 case ZFS_PROP_ACLTYPE
:
1660 *value
= ZFS_ACLTYPE_OFF
;
1669 EXPORT_SYMBOL(zfs_get_zplprop
);
1676 dmu_objset_register_type(DMU_OST_ZFS
, zfs_space_delta_cb
);
1677 register_filesystem(&zpl_fs_type
);
1678 (void) arc_add_prune_callback(zpl_prune_sbs
, NULL
);
1684 taskq_wait(system_taskq
);
1685 unregister_filesystem(&zpl_fs_type
);