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 #ifdef CONFIG_FS_POSIX_ACL
169 zsb
->z_acl_type
= ZFS_ACLTYPE_POSIXACL
;
170 zsb
->z_sb
->s_flags
|= MS_POSIXACL
;
172 zsb
->z_acl_type
= ZFS_ACLTYPE_OFF
;
173 zsb
->z_sb
->s_flags
&= ~MS_POSIXACL
;
174 #endif /* CONFIG_FS_POSIX_ACL */
182 blksz_changed_cb(void *arg
, uint64_t newval
)
186 if (newval
< SPA_MINBLOCKSIZE
||
187 newval
> SPA_MAXBLOCKSIZE
|| !ISP2(newval
))
188 newval
= SPA_MAXBLOCKSIZE
;
190 zsb
->z_max_blksz
= newval
;
194 readonly_changed_cb(void *arg
, uint64_t newval
)
197 struct super_block
*sb
= zsb
->z_sb
;
203 sb
->s_flags
|= MS_RDONLY
;
205 sb
->s_flags
&= ~MS_RDONLY
;
209 devices_changed_cb(void *arg
, uint64_t newval
)
214 setuid_changed_cb(void *arg
, uint64_t newval
)
219 exec_changed_cb(void *arg
, uint64_t newval
)
224 nbmand_changed_cb(void *arg
, uint64_t newval
)
227 struct super_block
*sb
= zsb
->z_sb
;
233 sb
->s_flags
|= MS_MANDLOCK
;
235 sb
->s_flags
&= ~MS_MANDLOCK
;
239 snapdir_changed_cb(void *arg
, uint64_t newval
)
241 ((zfs_sb_t
*)arg
)->z_show_ctldir
= newval
;
245 vscan_changed_cb(void *arg
, uint64_t newval
)
247 ((zfs_sb_t
*)arg
)->z_vscan
= newval
;
251 acl_inherit_changed_cb(void *arg
, uint64_t newval
)
253 ((zfs_sb_t
*)arg
)->z_acl_inherit
= newval
;
257 zfs_register_callbacks(zfs_sb_t
*zsb
)
259 struct dsl_dataset
*ds
= NULL
;
260 objset_t
*os
= zsb
->z_os
;
261 boolean_t do_readonly
= B_FALSE
;
264 if (zfs_is_readonly(zsb
) || !spa_writeable(dmu_objset_spa(os
)))
265 do_readonly
= B_TRUE
;
268 * Register property callbacks.
270 * It would probably be fine to just check for i/o error from
271 * the first prop_register(), but I guess I like to go
274 ds
= dmu_objset_ds(os
);
275 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
276 error
= dsl_prop_register(ds
,
277 zfs_prop_to_name(ZFS_PROP_ATIME
), atime_changed_cb
, zsb
);
278 error
= error
? error
: dsl_prop_register(ds
,
279 zfs_prop_to_name(ZFS_PROP_XATTR
), xattr_changed_cb
, zsb
);
280 error
= error
? error
: dsl_prop_register(ds
,
281 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
), blksz_changed_cb
, zsb
);
282 error
= error
? error
: dsl_prop_register(ds
,
283 zfs_prop_to_name(ZFS_PROP_READONLY
), readonly_changed_cb
, zsb
);
284 error
= error
? error
: dsl_prop_register(ds
,
285 zfs_prop_to_name(ZFS_PROP_DEVICES
), devices_changed_cb
, zsb
);
286 error
= error
? error
: dsl_prop_register(ds
,
287 zfs_prop_to_name(ZFS_PROP_SETUID
), setuid_changed_cb
, zsb
);
288 error
= error
? error
: dsl_prop_register(ds
,
289 zfs_prop_to_name(ZFS_PROP_EXEC
), exec_changed_cb
, zsb
);
290 error
= error
? error
: dsl_prop_register(ds
,
291 zfs_prop_to_name(ZFS_PROP_SNAPDIR
), snapdir_changed_cb
, zsb
);
292 error
= error
? error
: dsl_prop_register(ds
,
293 zfs_prop_to_name(ZFS_PROP_ACLTYPE
), acltype_changed_cb
, zsb
);
294 error
= error
? error
: dsl_prop_register(ds
,
295 zfs_prop_to_name(ZFS_PROP_ACLINHERIT
), acl_inherit_changed_cb
, zsb
);
296 error
= error
? error
: dsl_prop_register(ds
,
297 zfs_prop_to_name(ZFS_PROP_VSCAN
), vscan_changed_cb
, zsb
);
298 error
= error
? error
: dsl_prop_register(ds
,
299 zfs_prop_to_name(ZFS_PROP_NBMAND
), nbmand_changed_cb
, zsb
);
300 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
305 readonly_changed_cb(zsb
, B_TRUE
);
311 * We may attempt to unregister some callbacks that are not
312 * registered, but this is OK; it will simply return ENOMSG,
313 * which we will ignore.
315 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_ATIME
),
316 atime_changed_cb
, zsb
);
317 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_XATTR
),
318 xattr_changed_cb
, zsb
);
319 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_RECORDSIZE
),
320 blksz_changed_cb
, zsb
);
321 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_READONLY
),
322 readonly_changed_cb
, zsb
);
323 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_DEVICES
),
324 devices_changed_cb
, zsb
);
325 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_SETUID
),
326 setuid_changed_cb
, zsb
);
327 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_EXEC
),
328 exec_changed_cb
, zsb
);
329 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_SNAPDIR
),
330 snapdir_changed_cb
, zsb
);
331 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_ACLTYPE
),
332 acltype_changed_cb
, zsb
);
333 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_ACLINHERIT
),
334 acl_inherit_changed_cb
, zsb
);
335 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_VSCAN
),
336 vscan_changed_cb
, zsb
);
337 (void) dsl_prop_unregister(ds
, zfs_prop_to_name(ZFS_PROP_NBMAND
),
338 nbmand_changed_cb
, zsb
);
342 EXPORT_SYMBOL(zfs_register_callbacks
);
345 zfs_space_delta_cb(dmu_object_type_t bonustype
, void *data
,
346 uint64_t *userp
, uint64_t *groupp
)
349 * Is it a valid type of object to track?
351 if (bonustype
!= DMU_OT_ZNODE
&& bonustype
!= DMU_OT_SA
)
352 return (SET_ERROR(ENOENT
));
355 * If we have a NULL data pointer
356 * then assume the id's aren't changing and
357 * return EEXIST to the dmu to let it know to
361 return (SET_ERROR(EEXIST
));
363 if (bonustype
== DMU_OT_ZNODE
) {
364 znode_phys_t
*znp
= data
;
365 *userp
= znp
->zp_uid
;
366 *groupp
= znp
->zp_gid
;
369 sa_hdr_phys_t
*sap
= data
;
370 sa_hdr_phys_t sa
= *sap
;
371 boolean_t swap
= B_FALSE
;
373 ASSERT(bonustype
== DMU_OT_SA
);
375 if (sa
.sa_magic
== 0) {
377 * This should only happen for newly created
378 * files that haven't had the znode data filled
385 if (sa
.sa_magic
== BSWAP_32(SA_MAGIC
)) {
386 sa
.sa_magic
= SA_MAGIC
;
387 sa
.sa_layout_info
= BSWAP_16(sa
.sa_layout_info
);
390 VERIFY3U(sa
.sa_magic
, ==, SA_MAGIC
);
393 hdrsize
= sa_hdrsize(&sa
);
394 VERIFY3U(hdrsize
, >=, sizeof (sa_hdr_phys_t
));
395 *userp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+
397 *groupp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+
400 *userp
= BSWAP_64(*userp
);
401 *groupp
= BSWAP_64(*groupp
);
408 fuidstr_to_sid(zfs_sb_t
*zsb
, const char *fuidstr
,
409 char *domainbuf
, int buflen
, uid_t
*ridp
)
414 fuid
= strtonum(fuidstr
, NULL
);
416 domain
= zfs_fuid_find_by_idx(zsb
, FUID_INDEX(fuid
));
418 (void) strlcpy(domainbuf
, domain
, buflen
);
421 *ridp
= FUID_RID(fuid
);
425 zfs_userquota_prop_to_obj(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
)
428 case ZFS_PROP_USERUSED
:
429 return (DMU_USERUSED_OBJECT
);
430 case ZFS_PROP_GROUPUSED
:
431 return (DMU_GROUPUSED_OBJECT
);
432 case ZFS_PROP_USERQUOTA
:
433 return (zsb
->z_userquota_obj
);
434 case ZFS_PROP_GROUPQUOTA
:
435 return (zsb
->z_groupquota_obj
);
437 return (SET_ERROR(ENOTSUP
));
443 zfs_userspace_many(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
444 uint64_t *cookiep
, void *vbuf
, uint64_t *bufsizep
)
449 zfs_useracct_t
*buf
= vbuf
;
452 if (!dmu_objset_userspace_present(zsb
->z_os
))
453 return (SET_ERROR(ENOTSUP
));
455 obj
= zfs_userquota_prop_to_obj(zsb
, type
);
461 for (zap_cursor_init_serialized(&zc
, zsb
->z_os
, obj
, *cookiep
);
462 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
463 zap_cursor_advance(&zc
)) {
464 if ((uintptr_t)buf
- (uintptr_t)vbuf
+ sizeof (zfs_useracct_t
) >
468 fuidstr_to_sid(zsb
, za
.za_name
,
469 buf
->zu_domain
, sizeof (buf
->zu_domain
), &buf
->zu_rid
);
471 buf
->zu_space
= za
.za_first_integer
;
477 ASSERT3U((uintptr_t)buf
- (uintptr_t)vbuf
, <=, *bufsizep
);
478 *bufsizep
= (uintptr_t)buf
- (uintptr_t)vbuf
;
479 *cookiep
= zap_cursor_serialize(&zc
);
480 zap_cursor_fini(&zc
);
483 EXPORT_SYMBOL(zfs_userspace_many
);
486 * buf must be big enough (eg, 32 bytes)
489 id_to_fuidstr(zfs_sb_t
*zsb
, const char *domain
, uid_t rid
,
490 char *buf
, boolean_t addok
)
495 if (domain
&& domain
[0]) {
496 domainid
= zfs_fuid_find_by_domain(zsb
, domain
, NULL
, addok
);
498 return (SET_ERROR(ENOENT
));
500 fuid
= FUID_ENCODE(domainid
, rid
);
501 (void) sprintf(buf
, "%llx", (longlong_t
)fuid
);
506 zfs_userspace_one(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
507 const char *domain
, uint64_t rid
, uint64_t *valp
)
515 if (!dmu_objset_userspace_present(zsb
->z_os
))
516 return (SET_ERROR(ENOTSUP
));
518 obj
= zfs_userquota_prop_to_obj(zsb
, type
);
522 err
= id_to_fuidstr(zsb
, domain
, rid
, buf
, B_FALSE
);
526 err
= zap_lookup(zsb
->z_os
, obj
, buf
, 8, 1, valp
);
531 EXPORT_SYMBOL(zfs_userspace_one
);
534 zfs_set_userquota(zfs_sb_t
*zsb
, zfs_userquota_prop_t type
,
535 const char *domain
, uint64_t rid
, uint64_t quota
)
541 boolean_t fuid_dirtied
;
543 if (type
!= ZFS_PROP_USERQUOTA
&& type
!= ZFS_PROP_GROUPQUOTA
)
544 return (SET_ERROR(EINVAL
));
546 if (zsb
->z_version
< ZPL_VERSION_USERSPACE
)
547 return (SET_ERROR(ENOTSUP
));
549 objp
= (type
== ZFS_PROP_USERQUOTA
) ? &zsb
->z_userquota_obj
:
550 &zsb
->z_groupquota_obj
;
552 err
= id_to_fuidstr(zsb
, domain
, rid
, buf
, B_TRUE
);
555 fuid_dirtied
= zsb
->z_fuid_dirty
;
557 tx
= dmu_tx_create(zsb
->z_os
);
558 dmu_tx_hold_zap(tx
, *objp
? *objp
: DMU_NEW_OBJECT
, B_TRUE
, NULL
);
560 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
561 zfs_userquota_prop_prefixes
[type
]);
564 zfs_fuid_txhold(zsb
, tx
);
565 err
= dmu_tx_assign(tx
, TXG_WAIT
);
571 mutex_enter(&zsb
->z_lock
);
573 *objp
= zap_create(zsb
->z_os
, DMU_OT_USERGROUP_QUOTA
,
575 VERIFY(0 == zap_add(zsb
->z_os
, MASTER_NODE_OBJ
,
576 zfs_userquota_prop_prefixes
[type
], 8, 1, objp
, tx
));
578 mutex_exit(&zsb
->z_lock
);
581 err
= zap_remove(zsb
->z_os
, *objp
, buf
, tx
);
585 err
= zap_update(zsb
->z_os
, *objp
, buf
, 8, 1, "a
, tx
);
589 zfs_fuid_sync(zsb
, tx
);
593 EXPORT_SYMBOL(zfs_set_userquota
);
596 zfs_fuid_overquota(zfs_sb_t
*zsb
, boolean_t isgroup
, uint64_t fuid
)
599 uint64_t used
, quota
, usedobj
, quotaobj
;
602 usedobj
= isgroup
? DMU_GROUPUSED_OBJECT
: DMU_USERUSED_OBJECT
;
603 quotaobj
= isgroup
? zsb
->z_groupquota_obj
: zsb
->z_userquota_obj
;
605 if (quotaobj
== 0 || zsb
->z_replay
)
608 (void) sprintf(buf
, "%llx", (longlong_t
)fuid
);
609 err
= zap_lookup(zsb
->z_os
, quotaobj
, buf
, 8, 1, "a
);
613 err
= zap_lookup(zsb
->z_os
, usedobj
, buf
, 8, 1, &used
);
616 return (used
>= quota
);
618 EXPORT_SYMBOL(zfs_fuid_overquota
);
621 zfs_owner_overquota(zfs_sb_t
*zsb
, znode_t
*zp
, boolean_t isgroup
)
626 quotaobj
= isgroup
? zsb
->z_groupquota_obj
: zsb
->z_userquota_obj
;
628 fuid
= isgroup
? zp
->z_gid
: zp
->z_uid
;
630 if (quotaobj
== 0 || zsb
->z_replay
)
633 return (zfs_fuid_overquota(zsb
, isgroup
, fuid
));
635 EXPORT_SYMBOL(zfs_owner_overquota
);
638 zfs_sb_create(const char *osname
, zfs_sb_t
**zsbp
)
646 zsb
= kmem_zalloc(sizeof (zfs_sb_t
), KM_SLEEP
| KM_NODEBUG
);
649 * We claim to always be readonly so we can open snapshots;
650 * other ZPL code will prevent us from writing to snapshots.
652 error
= dmu_objset_own(osname
, DMU_OST_ZFS
, B_TRUE
, zsb
, &os
);
654 kmem_free(zsb
, sizeof (zfs_sb_t
));
659 * Initialize the zfs-specific filesystem structure.
660 * Should probably make this a kmem cache, shuffle fields,
661 * and just bzero up to z_hold_mtx[].
665 zsb
->z_max_blksz
= SPA_MAXBLOCKSIZE
;
666 zsb
->z_show_ctldir
= ZFS_SNAPDIR_VISIBLE
;
669 error
= zfs_get_zplprop(os
, ZFS_PROP_VERSION
, &zsb
->z_version
);
672 } else if (zsb
->z_version
>
673 zfs_zpl_version_map(spa_version(dmu_objset_spa(os
)))) {
674 (void) printk("Can't mount a version %lld file system "
675 "on a version %lld pool\n. Pool must be upgraded to mount "
676 "this file system.", (u_longlong_t
)zsb
->z_version
,
677 (u_longlong_t
)spa_version(dmu_objset_spa(os
)));
678 error
= SET_ERROR(ENOTSUP
);
681 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_NORMALIZE
, &zval
)) != 0)
683 zsb
->z_norm
= (int)zval
;
685 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_UTF8ONLY
, &zval
)) != 0)
687 zsb
->z_utf8
= (zval
!= 0);
689 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_CASE
, &zval
)) != 0)
691 zsb
->z_case
= (uint_t
)zval
;
693 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_ACLTYPE
, &zval
)) != 0)
695 zsb
->z_acl_type
= (uint_t
)zval
;
698 * Fold case on file systems that are always or sometimes case
701 if (zsb
->z_case
== ZFS_CASE_INSENSITIVE
||
702 zsb
->z_case
== ZFS_CASE_MIXED
)
703 zsb
->z_norm
|= U8_TEXTPREP_TOUPPER
;
705 zsb
->z_use_fuids
= USE_FUIDS(zsb
->z_version
, zsb
->z_os
);
706 zsb
->z_use_sa
= USE_SA(zsb
->z_version
, zsb
->z_os
);
709 /* should either have both of these objects or none */
710 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SA_ATTRS
, 8, 1,
715 error
= zfs_get_zplprop(os
, ZFS_PROP_XATTR
, &zval
);
716 if ((error
== 0) && (zval
== ZFS_XATTR_SA
))
717 zsb
->z_xattr_sa
= B_TRUE
;
720 * Pre SA versions file systems should never touch
721 * either the attribute registration or layout objects.
726 error
= sa_setup(os
, sa_obj
, zfs_attr_table
, ZPL_END
,
731 if (zsb
->z_version
>= ZPL_VERSION_SA
)
732 sa_register_update_callback(os
, zfs_sa_upgrade
);
734 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_ROOT_OBJ
, 8, 1,
738 ASSERT(zsb
->z_root
!= 0);
740 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_UNLINKED_SET
, 8, 1,
741 &zsb
->z_unlinkedobj
);
745 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
746 zfs_userquota_prop_prefixes
[ZFS_PROP_USERQUOTA
],
747 8, 1, &zsb
->z_userquota_obj
);
748 if (error
&& error
!= ENOENT
)
751 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
752 zfs_userquota_prop_prefixes
[ZFS_PROP_GROUPQUOTA
],
753 8, 1, &zsb
->z_groupquota_obj
);
754 if (error
&& error
!= ENOENT
)
757 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_FUID_TABLES
, 8, 1,
759 if (error
&& error
!= ENOENT
)
762 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SHARES_DIR
, 8, 1,
764 if (error
&& error
!= ENOENT
)
767 mutex_init(&zsb
->z_znodes_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
768 mutex_init(&zsb
->z_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
769 list_create(&zsb
->z_all_znodes
, sizeof (znode_t
),
770 offsetof(znode_t
, z_link_node
));
771 rrw_init(&zsb
->z_teardown_lock
, B_FALSE
);
772 rw_init(&zsb
->z_teardown_inactive_lock
, NULL
, RW_DEFAULT
, NULL
);
773 rw_init(&zsb
->z_fuid_lock
, NULL
, RW_DEFAULT
, NULL
);
774 for (i
= 0; i
!= ZFS_OBJ_MTX_SZ
; i
++)
775 mutex_init(&zsb
->z_hold_mtx
[i
], NULL
, MUTEX_DEFAULT
, NULL
);
777 avl_create(&zsb
->z_ctldir_snaps
, snapentry_compare
,
778 sizeof (zfs_snapentry_t
), offsetof(zfs_snapentry_t
, se_node
));
779 mutex_init(&zsb
->z_ctldir_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
785 dmu_objset_disown(os
, zsb
);
787 kmem_free(zsb
, sizeof (zfs_sb_t
));
790 EXPORT_SYMBOL(zfs_sb_create
);
793 zfs_sb_setup(zfs_sb_t
*zsb
, boolean_t mounting
)
797 error
= zfs_register_callbacks(zsb
);
802 * Set the objset user_ptr to track its zsb.
804 mutex_enter(&zsb
->z_os
->os_user_ptr_lock
);
805 dmu_objset_set_user(zsb
->z_os
, zsb
);
806 mutex_exit(&zsb
->z_os
->os_user_ptr_lock
);
808 zsb
->z_log
= zil_open(zsb
->z_os
, zfs_get_data
);
811 * If we are not mounting (ie: online recv), then we don't
812 * have to worry about replaying the log as we blocked all
813 * operations out since we closed the ZIL.
819 * During replay we remove the read only flag to
820 * allow replays to succeed.
822 readonly
= zfs_is_readonly(zsb
);
824 readonly_changed_cb(zsb
, B_FALSE
);
826 zfs_unlinked_drain(zsb
);
829 * Parse and replay the intent log.
831 * Because of ziltest, this must be done after
832 * zfs_unlinked_drain(). (Further note: ziltest
833 * doesn't use readonly mounts, where
834 * zfs_unlinked_drain() isn't called.) This is because
835 * ziltest causes spa_sync() to think it's committed,
836 * but actually it is not, so the intent log contains
837 * many txg's worth of changes.
839 * In particular, if object N is in the unlinked set in
840 * the last txg to actually sync, then it could be
841 * actually freed in a later txg and then reallocated
842 * in a yet later txg. This would write a "create
843 * object N" record to the intent log. Normally, this
844 * would be fine because the spa_sync() would have
845 * written out the fact that object N is free, before
846 * we could write the "create object N" intent log
849 * But when we are in ziltest mode, we advance the "open
850 * txg" without actually spa_sync()-ing the changes to
851 * disk. So we would see that object N is still
852 * allocated and in the unlinked set, and there is an
853 * intent log record saying to allocate it.
855 if (spa_writeable(dmu_objset_spa(zsb
->z_os
))) {
856 if (zil_replay_disable
) {
857 zil_destroy(zsb
->z_log
, B_FALSE
);
859 zsb
->z_replay
= B_TRUE
;
860 zil_replay(zsb
->z_os
, zsb
,
862 zsb
->z_replay
= B_FALSE
;
866 /* restore readonly bit */
868 readonly_changed_cb(zsb
, B_TRUE
);
873 EXPORT_SYMBOL(zfs_sb_setup
);
876 zfs_sb_free(zfs_sb_t
*zsb
)
880 zfs_fuid_destroy(zsb
);
882 mutex_destroy(&zsb
->z_znodes_lock
);
883 mutex_destroy(&zsb
->z_lock
);
884 list_destroy(&zsb
->z_all_znodes
);
885 rrw_destroy(&zsb
->z_teardown_lock
);
886 rw_destroy(&zsb
->z_teardown_inactive_lock
);
887 rw_destroy(&zsb
->z_fuid_lock
);
888 for (i
= 0; i
!= ZFS_OBJ_MTX_SZ
; i
++)
889 mutex_destroy(&zsb
->z_hold_mtx
[i
]);
890 mutex_destroy(&zsb
->z_ctldir_lock
);
891 avl_destroy(&zsb
->z_ctldir_snaps
);
892 kmem_free(zsb
, sizeof (zfs_sb_t
));
894 EXPORT_SYMBOL(zfs_sb_free
);
897 zfs_set_fuid_feature(zfs_sb_t
*zsb
)
899 zsb
->z_use_fuids
= USE_FUIDS(zsb
->z_version
, zsb
->z_os
);
900 zsb
->z_use_sa
= USE_SA(zsb
->z_version
, zsb
->z_os
);
904 zfs_unregister_callbacks(zfs_sb_t
*zsb
)
906 objset_t
*os
= zsb
->z_os
;
907 struct dsl_dataset
*ds
;
910 * Unregister properties.
912 if (!dmu_objset_is_snapshot(os
)) {
913 ds
= dmu_objset_ds(os
);
914 VERIFY(dsl_prop_unregister(ds
, "atime", atime_changed_cb
,
917 VERIFY(dsl_prop_unregister(ds
, "xattr", xattr_changed_cb
,
920 VERIFY(dsl_prop_unregister(ds
, "recordsize", blksz_changed_cb
,
923 VERIFY(dsl_prop_unregister(ds
, "readonly", readonly_changed_cb
,
926 VERIFY(dsl_prop_unregister(ds
, "devices", devices_changed_cb
,
929 VERIFY(dsl_prop_unregister(ds
, "setuid", setuid_changed_cb
,
932 VERIFY(dsl_prop_unregister(ds
, "exec", exec_changed_cb
,
935 VERIFY(dsl_prop_unregister(ds
, "snapdir", snapdir_changed_cb
,
938 VERIFY(dsl_prop_unregister(ds
, "acltype", acltype_changed_cb
,
941 VERIFY(dsl_prop_unregister(ds
, "aclinherit",
942 acl_inherit_changed_cb
, zsb
) == 0);
944 VERIFY(dsl_prop_unregister(ds
, "vscan",
945 vscan_changed_cb
, zsb
) == 0);
947 VERIFY(dsl_prop_unregister(ds
, "nbmand",
948 nbmand_changed_cb
, zsb
) == 0);
951 EXPORT_SYMBOL(zfs_unregister_callbacks
);
955 * Check that the hex label string is appropriate for the dataset being
956 * mounted into the global_zone proper.
958 * Return an error if the hex label string is not default or
959 * admin_low/admin_high. For admin_low labels, the corresponding
960 * dataset must be readonly.
963 zfs_check_global_label(const char *dsname
, const char *hexsl
)
965 if (strcasecmp(hexsl
, ZFS_MLSLABEL_DEFAULT
) == 0)
967 if (strcasecmp(hexsl
, ADMIN_HIGH
) == 0)
969 if (strcasecmp(hexsl
, ADMIN_LOW
) == 0) {
970 /* must be readonly */
973 if (dsl_prop_get_integer(dsname
,
974 zfs_prop_to_name(ZFS_PROP_READONLY
), &rdonly
, NULL
))
975 return (SET_ERROR(EACCES
));
976 return (rdonly
? 0 : EACCES
);
978 return (SET_ERROR(EACCES
));
980 EXPORT_SYMBOL(zfs_check_global_label
);
981 #endif /* HAVE_MLSLABEL */
984 zfs_statvfs(struct dentry
*dentry
, struct kstatfs
*statp
)
986 zfs_sb_t
*zsb
= dentry
->d_sb
->s_fs_info
;
987 uint64_t refdbytes
, availbytes
, usedobjs
, availobjs
;
993 dmu_objset_space(zsb
->z_os
,
994 &refdbytes
, &availbytes
, &usedobjs
, &availobjs
);
996 fsid
= dmu_objset_fsid_guid(zsb
->z_os
);
998 * The underlying storage pool actually uses multiple block
999 * size. Under Solaris frsize (fragment size) is reported as
1000 * the smallest block size we support, and bsize (block size)
1001 * as the filesystem's maximum block size. Unfortunately,
1002 * under Linux the fragment size and block size are often used
1003 * interchangeably. Thus we are forced to report both of them
1004 * as the filesystem's maximum block size.
1006 statp
->f_frsize
= zsb
->z_max_blksz
;
1007 statp
->f_bsize
= zsb
->z_max_blksz
;
1008 bshift
= fls(statp
->f_bsize
) - 1;
1011 * The following report "total" blocks of various kinds in
1012 * the file system, but reported in terms of f_bsize - the
1016 statp
->f_blocks
= (refdbytes
+ availbytes
) >> bshift
;
1017 statp
->f_bfree
= availbytes
>> bshift
;
1018 statp
->f_bavail
= statp
->f_bfree
; /* no root reservation */
1021 * statvfs() should really be called statufs(), because it assumes
1022 * static metadata. ZFS doesn't preallocate files, so the best
1023 * we can do is report the max that could possibly fit in f_files,
1024 * and that minus the number actually used in f_ffree.
1025 * For f_ffree, report the smaller of the number of object available
1026 * and the number of blocks (each object will take at least a block).
1028 statp
->f_ffree
= MIN(availobjs
, availbytes
>> DNODE_SHIFT
);
1029 statp
->f_files
= statp
->f_ffree
+ usedobjs
;
1030 statp
->f_fsid
.val
[0] = (uint32_t)fsid
;
1031 statp
->f_fsid
.val
[1] = (uint32_t)(fsid
>> 32);
1032 statp
->f_type
= ZFS_SUPER_MAGIC
;
1033 statp
->f_namelen
= ZFS_MAXNAMELEN
;
1036 * We have all of 40 characters to stuff a string here.
1037 * Is there anything useful we could/should provide?
1039 bzero(statp
->f_spare
, sizeof (statp
->f_spare
));
1044 EXPORT_SYMBOL(zfs_statvfs
);
1047 zfs_root(zfs_sb_t
*zsb
, struct inode
**ipp
)
1054 error
= zfs_zget(zsb
, zsb
->z_root
, &rootzp
);
1056 *ipp
= ZTOI(rootzp
);
1061 EXPORT_SYMBOL(zfs_root
);
1065 zfs_sb_prune(struct super_block
*sb
, unsigned long nr_to_scan
, int *objects
)
1067 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1068 struct shrinker
*shrinker
= &sb
->s_shrink
;
1069 struct shrink_control sc
= {
1070 .nr_to_scan
= nr_to_scan
,
1071 .gfp_mask
= GFP_KERNEL
,
1075 *objects
= (*shrinker
->shrink
)(shrinker
, &sc
);
1080 EXPORT_SYMBOL(zfs_sb_prune
);
1081 #endif /* HAVE_SHRINK */
1084 * Teardown the zfs_sb_t.
1086 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1087 * and 'z_teardown_inactive_lock' held.
1090 zfs_sb_teardown(zfs_sb_t
*zsb
, boolean_t unmounting
)
1094 rrw_enter(&zsb
->z_teardown_lock
, RW_WRITER
, FTAG
);
1098 * We purge the parent filesystem's super block as the
1099 * parent filesystem and all of its snapshots have their
1100 * inode's super block set to the parent's filesystem's
1101 * super block. Note, 'z_parent' is self referential
1102 * for non-snapshots.
1104 shrink_dcache_sb(zsb
->z_parent
->z_sb
);
1108 * If someone has not already unmounted this file system,
1109 * drain the iput_taskq to ensure all active references to the
1110 * zfs_sb_t have been handled only then can it be safely destroyed.
1113 taskq_wait(dsl_pool_iput_taskq(dmu_objset_pool(zsb
->z_os
)));
1116 * Close the zil. NB: Can't close the zil while zfs_inactive
1117 * threads are blocked as zil_close can call zfs_inactive.
1120 zil_close(zsb
->z_log
);
1124 rw_enter(&zsb
->z_teardown_inactive_lock
, RW_WRITER
);
1127 * If we are not unmounting (ie: online recv) and someone already
1128 * unmounted this file system while we were doing the switcheroo,
1129 * or a reopen of z_os failed then just bail out now.
1131 if (!unmounting
&& (zsb
->z_unmounted
|| zsb
->z_os
== NULL
)) {
1132 rw_exit(&zsb
->z_teardown_inactive_lock
);
1133 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1134 return (SET_ERROR(EIO
));
1138 * At this point there are no VFS ops active, and any new VFS ops
1139 * will fail with EIO since we have z_teardown_lock for writer (only
1140 * relevant for forced unmount).
1142 * Release all holds on dbufs.
1144 mutex_enter(&zsb
->z_znodes_lock
);
1145 for (zp
= list_head(&zsb
->z_all_znodes
); zp
!= NULL
;
1146 zp
= list_next(&zsb
->z_all_znodes
, zp
)) {
1148 zfs_znode_dmu_fini(zp
);
1150 mutex_exit(&zsb
->z_znodes_lock
);
1153 * If we are unmounting, set the unmounted flag and let new VFS ops
1154 * unblock. zfs_inactive will have the unmounted behavior, and all
1155 * other VFS ops will fail with EIO.
1158 zsb
->z_unmounted
= B_TRUE
;
1159 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1160 rw_exit(&zsb
->z_teardown_inactive_lock
);
1164 * z_os will be NULL if there was an error in attempting to reopen
1165 * zsb, so just return as the properties had already been
1167 * unregistered and cached data had been evicted before.
1169 if (zsb
->z_os
== NULL
)
1173 * Unregister properties.
1175 zfs_unregister_callbacks(zsb
);
1180 if (dsl_dataset_is_dirty(dmu_objset_ds(zsb
->z_os
)) &&
1181 !zfs_is_readonly(zsb
))
1182 txg_wait_synced(dmu_objset_pool(zsb
->z_os
), 0);
1183 dmu_objset_evict_dbufs(zsb
->z_os
);
1187 EXPORT_SYMBOL(zfs_sb_teardown
);
1189 #if defined(HAVE_BDI) && !defined(HAVE_BDI_SETUP_AND_REGISTER)
1190 atomic_long_t zfs_bdi_seq
= ATOMIC_LONG_INIT(0);
1191 #endif /* HAVE_BDI && !HAVE_BDI_SETUP_AND_REGISTER */
1194 zfs_domount(struct super_block
*sb
, void *data
, int silent
)
1196 zpl_mount_data_t
*zmd
= data
;
1197 const char *osname
= zmd
->z_osname
;
1199 struct inode
*root_inode
;
1200 uint64_t recordsize
;
1203 error
= zfs_sb_create(osname
, &zsb
);
1207 if ((error
= dsl_prop_get_integer(osname
, "recordsize",
1208 &recordsize
, NULL
)))
1212 sb
->s_fs_info
= zsb
;
1213 sb
->s_magic
= ZFS_SUPER_MAGIC
;
1214 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1215 sb
->s_time_gran
= 1;
1216 sb
->s_blocksize
= recordsize
;
1217 sb
->s_blocksize_bits
= ilog2(recordsize
);
1221 * 2.6.32 API change,
1222 * Added backing_device_info (BDI) per super block interfaces. A BDI
1223 * must be configured when using a non-device backed filesystem for
1224 * proper writeback. This is not required for older pdflush kernels.
1226 * NOTE: Linux read-ahead is disabled in favor of zfs read-ahead.
1228 zsb
->z_bdi
.ra_pages
= 0;
1229 sb
->s_bdi
= &zsb
->z_bdi
;
1231 error
= -bdi_setup_and_register(&zsb
->z_bdi
, "zfs", BDI_CAP_MAP_COPY
);
1234 #endif /* HAVE_BDI */
1236 /* Set callback operations for the file system. */
1237 sb
->s_op
= &zpl_super_operations
;
1238 sb
->s_xattr
= zpl_xattr_handlers
;
1239 sb
->s_export_op
= &zpl_export_operations
;
1241 sb
->s_d_op
= &zpl_dentry_operations
;
1242 #endif /* HAVE_S_D_OP */
1244 /* Set features for file system. */
1245 zfs_set_fuid_feature(zsb
);
1247 if (dmu_objset_is_snapshot(zsb
->z_os
)) {
1250 atime_changed_cb(zsb
, B_FALSE
);
1251 readonly_changed_cb(zsb
, B_TRUE
);
1252 if ((error
= dsl_prop_get_integer(osname
,"xattr",&pval
,NULL
)))
1254 xattr_changed_cb(zsb
, pval
);
1255 if ((error
= dsl_prop_get_integer(osname
,"acltype",&pval
,NULL
)))
1257 acltype_changed_cb(zsb
, pval
);
1258 zsb
->z_issnap
= B_TRUE
;
1259 zsb
->z_os
->os_sync
= ZFS_SYNC_DISABLED
;
1261 mutex_enter(&zsb
->z_os
->os_user_ptr_lock
);
1262 dmu_objset_set_user(zsb
->z_os
, zsb
);
1263 mutex_exit(&zsb
->z_os
->os_user_ptr_lock
);
1265 error
= zfs_sb_setup(zsb
, B_TRUE
);
1268 /* Allocate a root inode for the filesystem. */
1269 error
= zfs_root(zsb
, &root_inode
);
1271 (void) zfs_umount(sb
);
1275 /* Allocate a root dentry for the filesystem */
1276 sb
->s_root
= d_make_root(root_inode
);
1277 if (sb
->s_root
== NULL
) {
1278 (void) zfs_umount(sb
);
1279 error
= SET_ERROR(ENOMEM
);
1287 dmu_objset_disown(zsb
->z_os
, zsb
);
1293 EXPORT_SYMBOL(zfs_domount
);
1296 * Called when an unmount is requested and certain sanity checks have
1297 * already passed. At this point no dentries or inodes have been reclaimed
1298 * from their respective caches. We drop the extra reference on the .zfs
1299 * control directory to allow everything to be reclaimed. All snapshots
1300 * must already have been unmounted to reach this point.
1303 zfs_preumount(struct super_block
*sb
)
1305 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1307 if (zsb
!= NULL
&& zsb
->z_ctldir
!= NULL
)
1308 zfsctl_destroy(zsb
);
1310 EXPORT_SYMBOL(zfs_preumount
);
1313 * Called once all other unmount released tear down has occurred.
1314 * It is our responsibility to release any remaining infrastructure.
1318 zfs_umount(struct super_block
*sb
)
1320 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1323 VERIFY(zfs_sb_teardown(zsb
, B_TRUE
) == 0);
1327 bdi_destroy(sb
->s_bdi
);
1328 #endif /* HAVE_BDI */
1331 * z_os will be NULL if there was an error in
1332 * attempting to reopen zsb.
1336 * Unset the objset user_ptr.
1338 mutex_enter(&os
->os_user_ptr_lock
);
1339 dmu_objset_set_user(os
, NULL
);
1340 mutex_exit(&os
->os_user_ptr_lock
);
1343 * Finally release the objset
1345 dmu_objset_disown(os
, zsb
);
1351 EXPORT_SYMBOL(zfs_umount
);
1354 zfs_remount(struct super_block
*sb
, int *flags
, char *data
)
1357 * All namespace flags (MNT_*) and super block flags (MS_*) will
1358 * be handled by the Linux VFS. Only handle custom options here.
1362 EXPORT_SYMBOL(zfs_remount
);
1365 zfs_vget(struct super_block
*sb
, struct inode
**ipp
, fid_t
*fidp
)
1367 zfs_sb_t
*zsb
= sb
->s_fs_info
;
1369 uint64_t object
= 0;
1370 uint64_t fid_gen
= 0;
1379 if (fidp
->fid_len
== LONG_FID_LEN
) {
1380 zfid_long_t
*zlfid
= (zfid_long_t
*)fidp
;
1381 uint64_t objsetid
= 0;
1382 uint64_t setgen
= 0;
1384 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
1385 objsetid
|= ((uint64_t)zlfid
->zf_setid
[i
]) << (8 * i
);
1387 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
1388 setgen
|= ((uint64_t)zlfid
->zf_setgen
[i
]) << (8 * i
);
1392 err
= zfsctl_lookup_objset(sb
, objsetid
, &zsb
);
1394 return (SET_ERROR(EINVAL
));
1399 if (fidp
->fid_len
== SHORT_FID_LEN
|| fidp
->fid_len
== LONG_FID_LEN
) {
1400 zfid_short_t
*zfid
= (zfid_short_t
*)fidp
;
1402 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
1403 object
|= ((uint64_t)zfid
->zf_object
[i
]) << (8 * i
);
1405 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
1406 fid_gen
|= ((uint64_t)zfid
->zf_gen
[i
]) << (8 * i
);
1409 return (SET_ERROR(EINVAL
));
1412 /* A zero fid_gen means we are in the .zfs control directories */
1414 (object
== ZFSCTL_INO_ROOT
|| object
== ZFSCTL_INO_SNAPDIR
)) {
1415 *ipp
= zsb
->z_ctldir
;
1416 ASSERT(*ipp
!= NULL
);
1417 if (object
== ZFSCTL_INO_SNAPDIR
) {
1418 VERIFY(zfsctl_root_lookup(*ipp
, "snapshot", ipp
,
1419 0, kcred
, NULL
, NULL
) == 0);
1427 gen_mask
= -1ULL >> (64 - 8 * i
);
1429 dprintf("getting %llu [%u mask %llx]\n", object
, fid_gen
, gen_mask
);
1430 if ((err
= zfs_zget(zsb
, object
, &zp
))) {
1434 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zsb
), &zp_gen
,
1436 zp_gen
= zp_gen
& gen_mask
;
1439 if (zp
->z_unlinked
|| zp_gen
!= fid_gen
) {
1440 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen
, fid_gen
);
1443 return (SET_ERROR(EINVAL
));
1448 zfs_inode_update(ITOZ(*ipp
));
1453 EXPORT_SYMBOL(zfs_vget
);
1456 * Block out VFS ops and close zfs_sb_t
1458 * Note, if successful, then we return with the 'z_teardown_lock' and
1459 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
1460 * dataset and objset intact so that they can be atomically handed off during
1461 * a subsequent rollback or recv operation and the resume thereafter.
1464 zfs_suspend_fs(zfs_sb_t
*zsb
)
1468 if ((error
= zfs_sb_teardown(zsb
, B_FALSE
)) != 0)
1473 EXPORT_SYMBOL(zfs_suspend_fs
);
1476 * Reopen zfs_sb_t and release VFS ops.
1479 zfs_resume_fs(zfs_sb_t
*zsb
, const char *osname
)
1483 uint64_t sa_obj
= 0;
1485 ASSERT(RRW_WRITE_HELD(&zsb
->z_teardown_lock
));
1486 ASSERT(RW_WRITE_HELD(&zsb
->z_teardown_inactive_lock
));
1489 * We already own this, so just hold and rele it to update the
1490 * objset_t, as the one we had before may have been evicted.
1492 VERIFY0(dmu_objset_hold(osname
, zsb
, &zsb
->z_os
));
1493 VERIFY3P(zsb
->z_os
->os_dsl_dataset
->ds_owner
, ==, zsb
);
1494 VERIFY(dsl_dataset_long_held(zsb
->z_os
->os_dsl_dataset
));
1495 dmu_objset_rele(zsb
->z_os
, zsb
);
1498 * Make sure version hasn't changed
1501 err
= zfs_get_zplprop(zsb
->z_os
, ZFS_PROP_VERSION
,
1507 err
= zap_lookup(zsb
->z_os
, MASTER_NODE_OBJ
,
1508 ZFS_SA_ATTRS
, 8, 1, &sa_obj
);
1510 if (err
&& zsb
->z_version
>= ZPL_VERSION_SA
)
1513 if ((err
= sa_setup(zsb
->z_os
, sa_obj
,
1514 zfs_attr_table
, ZPL_END
, &zsb
->z_attr_table
)) != 0)
1517 if (zsb
->z_version
>= ZPL_VERSION_SA
)
1518 sa_register_update_callback(zsb
->z_os
,
1521 VERIFY(zfs_sb_setup(zsb
, B_FALSE
) == 0);
1523 zfs_set_fuid_feature(zsb
);
1524 zsb
->z_rollback_time
= jiffies
;
1527 * Attempt to re-establish all the active inodes with their
1528 * dbufs. If a zfs_rezget() fails, then we unhash the inode
1529 * and mark it stale. This prevents a collision if a new
1530 * inode/object is created which must use the same inode
1531 * number. The stale inode will be be released when the
1532 * VFS prunes the dentry holding the remaining references
1533 * on the stale inode.
1535 mutex_enter(&zsb
->z_znodes_lock
);
1536 for (zp
= list_head(&zsb
->z_all_znodes
); zp
;
1537 zp
= list_next(&zsb
->z_all_znodes
, zp
)) {
1538 err2
= zfs_rezget(zp
);
1540 remove_inode_hash(ZTOI(zp
));
1541 zp
->z_is_stale
= B_TRUE
;
1544 mutex_exit(&zsb
->z_znodes_lock
);
1547 /* release the VFS ops */
1548 rw_exit(&zsb
->z_teardown_inactive_lock
);
1549 rrw_exit(&zsb
->z_teardown_lock
, FTAG
);
1553 * Since we couldn't setup the sa framework, try to force
1554 * unmount this file system.
1557 (void) zfs_umount(zsb
->z_sb
);
1561 EXPORT_SYMBOL(zfs_resume_fs
);
1564 zfs_set_version(zfs_sb_t
*zsb
, uint64_t newvers
)
1567 objset_t
*os
= zsb
->z_os
;
1570 if (newvers
< ZPL_VERSION_INITIAL
|| newvers
> ZPL_VERSION
)
1571 return (SET_ERROR(EINVAL
));
1573 if (newvers
< zsb
->z_version
)
1574 return (SET_ERROR(EINVAL
));
1576 if (zfs_spa_version_map(newvers
) >
1577 spa_version(dmu_objset_spa(zsb
->z_os
)))
1578 return (SET_ERROR(ENOTSUP
));
1580 tx
= dmu_tx_create(os
);
1581 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_FALSE
, ZPL_VERSION_STR
);
1582 if (newvers
>= ZPL_VERSION_SA
&& !zsb
->z_use_sa
) {
1583 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
1585 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
1587 error
= dmu_tx_assign(tx
, TXG_WAIT
);
1593 error
= zap_update(os
, MASTER_NODE_OBJ
, ZPL_VERSION_STR
,
1594 8, 1, &newvers
, tx
);
1601 if (newvers
>= ZPL_VERSION_SA
&& !zsb
->z_use_sa
) {
1604 ASSERT3U(spa_version(dmu_objset_spa(zsb
->z_os
)), >=,
1606 sa_obj
= zap_create(os
, DMU_OT_SA_MASTER_NODE
,
1607 DMU_OT_NONE
, 0, tx
);
1609 error
= zap_add(os
, MASTER_NODE_OBJ
,
1610 ZFS_SA_ATTRS
, 8, 1, &sa_obj
, tx
);
1613 VERIFY(0 == sa_set_sa_object(os
, sa_obj
));
1614 sa_register_update_callback(os
, zfs_sa_upgrade
);
1617 spa_history_log_internal_ds(dmu_objset_ds(os
), "upgrade", tx
,
1618 "from %llu to %llu", zsb
->z_version
, newvers
);
1622 zsb
->z_version
= newvers
;
1624 zfs_set_fuid_feature(zsb
);
1628 EXPORT_SYMBOL(zfs_set_version
);
1631 * Read a property stored within the master node.
1634 zfs_get_zplprop(objset_t
*os
, zfs_prop_t prop
, uint64_t *value
)
1637 int error
= SET_ERROR(ENOENT
);
1640 * Look up the file system's value for the property. For the
1641 * version property, we look up a slightly different string.
1643 if (prop
== ZFS_PROP_VERSION
)
1644 pname
= ZPL_VERSION_STR
;
1646 pname
= zfs_prop_to_name(prop
);
1649 error
= zap_lookup(os
, MASTER_NODE_OBJ
, pname
, 8, 1, value
);
1651 if (error
== ENOENT
) {
1652 /* No value set, use the default value */
1654 case ZFS_PROP_VERSION
:
1655 *value
= ZPL_VERSION
;
1657 case ZFS_PROP_NORMALIZE
:
1658 case ZFS_PROP_UTF8ONLY
:
1662 *value
= ZFS_CASE_SENSITIVE
;
1664 case ZFS_PROP_ACLTYPE
:
1665 *value
= ZFS_ACLTYPE_OFF
;
1674 EXPORT_SYMBOL(zfs_get_zplprop
);
1681 dmu_objset_register_type(DMU_OST_ZFS
, zfs_space_delta_cb
);
1682 register_filesystem(&zpl_fs_type
);
1683 (void) arc_add_prune_callback(zpl_prune_sbs
, NULL
);
1689 taskq_wait(system_taskq
);
1690 unregister_filesystem(&zpl_fs_type
);