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) 2012, 2014 by Delphix. All rights reserved.
24 * Copyright (c) 2014 Integros [integros.com]
27 /* Portions Copyright 2010 Robert Milkowski */
29 #include <sys/types.h>
30 #include <sys/param.h>
31 #include <sys/systm.h>
32 #include <sys/sysmacros.h>
34 #include <sys/pathname.h>
35 #include <sys/vnode.h>
37 #include <sys/vfs_opreg.h>
38 #include <sys/mntent.h>
39 #include <sys/mount.h>
40 #include <sys/cmn_err.h>
41 #include "fs/fs_subr.h"
42 #include <sys/zfs_znode.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>
68 #include "zfs_comutil.h"
71 vfsops_t
*zfs_vfsops
= NULL
;
72 static major_t zfs_major
;
73 static minor_t zfs_minor
;
74 static kmutex_t zfs_dev_mtx
;
76 extern int sys_shutdown
;
78 static int zfs_mount(vfs_t
*vfsp
, vnode_t
*mvp
, struct mounta
*uap
, cred_t
*cr
);
79 static int zfs_umount(vfs_t
*vfsp
, int fflag
, cred_t
*cr
);
80 static int zfs_mountroot(vfs_t
*vfsp
, enum whymountroot
);
81 static int zfs_root(vfs_t
*vfsp
, vnode_t
**vpp
);
82 static int zfs_statvfs(vfs_t
*vfsp
, struct statvfs64
*statp
);
83 static int zfs_vget(vfs_t
*vfsp
, vnode_t
**vpp
, fid_t
*fidp
);
84 static void zfs_freevfs(vfs_t
*vfsp
);
86 static const fs_operation_def_t zfs_vfsops_template
[] = {
87 VFSNAME_MOUNT
, { .vfs_mount
= zfs_mount
},
88 VFSNAME_MOUNTROOT
, { .vfs_mountroot
= zfs_mountroot
},
89 VFSNAME_UNMOUNT
, { .vfs_unmount
= zfs_umount
},
90 VFSNAME_ROOT
, { .vfs_root
= zfs_root
},
91 VFSNAME_STATVFS
, { .vfs_statvfs
= zfs_statvfs
},
92 VFSNAME_SYNC
, { .vfs_sync
= zfs_sync
},
93 VFSNAME_VGET
, { .vfs_vget
= zfs_vget
},
94 VFSNAME_FREEVFS
, { .vfs_freevfs
= zfs_freevfs
},
98 static const fs_operation_def_t zfs_vfsops_eio_template
[] = {
99 VFSNAME_FREEVFS
, { .vfs_freevfs
= zfs_freevfs
},
104 * We need to keep a count of active fs's.
105 * This is necessary to prevent our module
106 * from being unloaded after a umount -f
108 static uint32_t zfs_active_fs_count
= 0;
110 static char *noatime_cancel
[] = { MNTOPT_ATIME
, NULL
};
111 static char *atime_cancel
[] = { MNTOPT_NOATIME
, NULL
};
112 static char *noxattr_cancel
[] = { MNTOPT_XATTR
, NULL
};
113 static char *xattr_cancel
[] = { MNTOPT_NOXATTR
, NULL
};
116 * MO_DEFAULT is not used since the default value is determined
117 * by the equivalent property.
119 static mntopt_t mntopts
[] = {
120 { MNTOPT_NOXATTR
, noxattr_cancel
, NULL
, 0, NULL
},
121 { MNTOPT_XATTR
, xattr_cancel
, NULL
, 0, NULL
},
122 { MNTOPT_NOATIME
, noatime_cancel
, NULL
, 0, NULL
},
123 { MNTOPT_ATIME
, atime_cancel
, NULL
, 0, NULL
}
126 static mntopts_t zfs_mntopts
= {
127 sizeof (mntopts
) / sizeof (mntopt_t
),
133 zfs_sync(vfs_t
*vfsp
, short flag
, cred_t
*cr
)
136 * Data integrity is job one. We don't want a compromised kernel
137 * writing to the storage pool, so we never sync during panic.
143 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
144 * to sync metadata, which they would otherwise cache indefinitely.
145 * Semantically, the only requirement is that the sync be initiated.
146 * The DMU syncs out txgs frequently, so there's nothing to do.
148 if (flag
& SYNC_ATTR
)
153 * Sync a specific filesystem.
155 zfsvfs_t
*zfsvfs
= vfsp
->vfs_data
;
159 dp
= dmu_objset_pool(zfsvfs
->z_os
);
162 * If the system is shutting down, then skip any
163 * filesystems which may exist on a suspended pool.
165 if (sys_shutdown
&& spa_suspended(dp
->dp_spa
)) {
170 if (zfsvfs
->z_log
!= NULL
)
171 zil_commit(zfsvfs
->z_log
, 0);
176 * Sync all ZFS filesystems. This is what happens when you
177 * run sync(1M). Unlike other filesystems, ZFS honors the
178 * request by waiting for all pools to commit all dirty data.
187 zfs_create_unique_device(dev_t
*dev
)
192 ASSERT3U(zfs_minor
, <=, MAXMIN32
);
193 minor_t start
= zfs_minor
;
195 mutex_enter(&zfs_dev_mtx
);
196 if (zfs_minor
>= MAXMIN32
) {
198 * If we're still using the real major
199 * keep out of /dev/zfs and /dev/zvol minor
200 * number space. If we're using a getudev()'ed
201 * major number, we can use all of its minors.
203 if (zfs_major
== ddi_name_to_major(ZFS_DRIVER
))
204 zfs_minor
= ZFS_MIN_MINOR
;
210 *dev
= makedevice(zfs_major
, zfs_minor
);
211 mutex_exit(&zfs_dev_mtx
);
212 } while (vfs_devismounted(*dev
) && zfs_minor
!= start
);
213 if (zfs_minor
== start
) {
215 * We are using all ~262,000 minor numbers for the
216 * current major number. Create a new major number.
218 if ((new_major
= getudev()) == (major_t
)-1) {
220 "zfs_mount: Can't get unique major "
224 mutex_enter(&zfs_dev_mtx
);
225 zfs_major
= new_major
;
228 mutex_exit(&zfs_dev_mtx
);
232 /* CONSTANTCONDITION */
239 atime_changed_cb(void *arg
, uint64_t newval
)
241 zfsvfs_t
*zfsvfs
= arg
;
243 if (newval
== TRUE
) {
244 zfsvfs
->z_atime
= TRUE
;
245 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_NOATIME
);
246 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_ATIME
, NULL
, 0);
248 zfsvfs
->z_atime
= FALSE
;
249 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_ATIME
);
250 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_NOATIME
, NULL
, 0);
255 xattr_changed_cb(void *arg
, uint64_t newval
)
257 zfsvfs_t
*zfsvfs
= arg
;
259 if (newval
== TRUE
) {
260 /* XXX locking on vfs_flag? */
261 zfsvfs
->z_vfs
->vfs_flag
|= VFS_XATTR
;
262 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_NOXATTR
);
263 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_XATTR
, NULL
, 0);
265 /* XXX locking on vfs_flag? */
266 zfsvfs
->z_vfs
->vfs_flag
&= ~VFS_XATTR
;
267 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_XATTR
);
268 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_NOXATTR
, NULL
, 0);
273 blksz_changed_cb(void *arg
, uint64_t newval
)
275 zfsvfs_t
*zfsvfs
= arg
;
276 ASSERT3U(newval
, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs
->z_os
)));
277 ASSERT3U(newval
, >=, SPA_MINBLOCKSIZE
);
278 ASSERT(ISP2(newval
));
280 zfsvfs
->z_max_blksz
= newval
;
281 zfsvfs
->z_vfs
->vfs_bsize
= newval
;
285 readonly_changed_cb(void *arg
, uint64_t newval
)
287 zfsvfs_t
*zfsvfs
= arg
;
290 /* XXX locking on vfs_flag? */
291 zfsvfs
->z_vfs
->vfs_flag
|= VFS_RDONLY
;
292 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_RW
);
293 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_RO
, NULL
, 0);
295 /* XXX locking on vfs_flag? */
296 zfsvfs
->z_vfs
->vfs_flag
&= ~VFS_RDONLY
;
297 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_RO
);
298 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_RW
, NULL
, 0);
303 devices_changed_cb(void *arg
, uint64_t newval
)
305 zfsvfs_t
*zfsvfs
= arg
;
307 if (newval
== FALSE
) {
308 zfsvfs
->z_vfs
->vfs_flag
|= VFS_NODEVICES
;
309 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_DEVICES
);
310 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_NODEVICES
, NULL
, 0);
312 zfsvfs
->z_vfs
->vfs_flag
&= ~VFS_NODEVICES
;
313 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_NODEVICES
);
314 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_DEVICES
, NULL
, 0);
319 setuid_changed_cb(void *arg
, uint64_t newval
)
321 zfsvfs_t
*zfsvfs
= arg
;
323 if (newval
== FALSE
) {
324 zfsvfs
->z_vfs
->vfs_flag
|= VFS_NOSETUID
;
325 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_SETUID
);
326 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_NOSETUID
, NULL
, 0);
328 zfsvfs
->z_vfs
->vfs_flag
&= ~VFS_NOSETUID
;
329 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_NOSETUID
);
330 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_SETUID
, NULL
, 0);
335 exec_changed_cb(void *arg
, uint64_t newval
)
337 zfsvfs_t
*zfsvfs
= arg
;
339 if (newval
== FALSE
) {
340 zfsvfs
->z_vfs
->vfs_flag
|= VFS_NOEXEC
;
341 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_EXEC
);
342 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_NOEXEC
, NULL
, 0);
344 zfsvfs
->z_vfs
->vfs_flag
&= ~VFS_NOEXEC
;
345 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_NOEXEC
);
346 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_EXEC
, NULL
, 0);
351 * The nbmand mount option can be changed at mount time.
352 * We can't allow it to be toggled on live file systems or incorrect
353 * behavior may be seen from cifs clients
355 * This property isn't registered via dsl_prop_register(), but this callback
356 * will be called when a file system is first mounted
359 nbmand_changed_cb(void *arg
, uint64_t newval
)
361 zfsvfs_t
*zfsvfs
= arg
;
362 if (newval
== FALSE
) {
363 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_NBMAND
);
364 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_NONBMAND
, NULL
, 0);
366 vfs_clearmntopt(zfsvfs
->z_vfs
, MNTOPT_NONBMAND
);
367 vfs_setmntopt(zfsvfs
->z_vfs
, MNTOPT_NBMAND
, NULL
, 0);
372 snapdir_changed_cb(void *arg
, uint64_t newval
)
374 zfsvfs_t
*zfsvfs
= arg
;
376 zfsvfs
->z_show_ctldir
= newval
;
380 vscan_changed_cb(void *arg
, uint64_t newval
)
382 zfsvfs_t
*zfsvfs
= arg
;
384 zfsvfs
->z_vscan
= newval
;
388 acl_mode_changed_cb(void *arg
, uint64_t newval
)
390 zfsvfs_t
*zfsvfs
= arg
;
392 zfsvfs
->z_acl_mode
= newval
;
396 acl_inherit_changed_cb(void *arg
, uint64_t newval
)
398 zfsvfs_t
*zfsvfs
= arg
;
400 zfsvfs
->z_acl_inherit
= newval
;
404 zfs_register_callbacks(vfs_t
*vfsp
)
406 struct dsl_dataset
*ds
= NULL
;
408 zfsvfs_t
*zfsvfs
= NULL
;
410 boolean_t readonly
= B_FALSE
;
411 boolean_t do_readonly
= B_FALSE
;
412 boolean_t setuid
= B_FALSE
;
413 boolean_t do_setuid
= B_FALSE
;
414 boolean_t exec
= B_FALSE
;
415 boolean_t do_exec
= B_FALSE
;
416 boolean_t devices
= B_FALSE
;
417 boolean_t do_devices
= B_FALSE
;
418 boolean_t xattr
= B_FALSE
;
419 boolean_t do_xattr
= B_FALSE
;
420 boolean_t atime
= B_FALSE
;
421 boolean_t do_atime
= B_FALSE
;
425 zfsvfs
= vfsp
->vfs_data
;
430 * The act of registering our callbacks will destroy any mount
431 * options we may have. In order to enable temporary overrides
432 * of mount options, we stash away the current values and
433 * restore them after we register the callbacks.
435 if (vfs_optionisset(vfsp
, MNTOPT_RO
, NULL
) ||
436 !spa_writeable(dmu_objset_spa(os
))) {
438 do_readonly
= B_TRUE
;
439 } else if (vfs_optionisset(vfsp
, MNTOPT_RW
, NULL
)) {
441 do_readonly
= B_TRUE
;
443 if (vfs_optionisset(vfsp
, MNTOPT_NOSUID
, NULL
)) {
449 if (vfs_optionisset(vfsp
, MNTOPT_NODEVICES
, NULL
)) {
452 } else if (vfs_optionisset(vfsp
, MNTOPT_DEVICES
, NULL
)) {
457 if (vfs_optionisset(vfsp
, MNTOPT_NOSETUID
, NULL
)) {
460 } else if (vfs_optionisset(vfsp
, MNTOPT_SETUID
, NULL
)) {
465 if (vfs_optionisset(vfsp
, MNTOPT_NOEXEC
, NULL
)) {
468 } else if (vfs_optionisset(vfsp
, MNTOPT_EXEC
, NULL
)) {
472 if (vfs_optionisset(vfsp
, MNTOPT_NOXATTR
, NULL
)) {
475 } else if (vfs_optionisset(vfsp
, MNTOPT_XATTR
, NULL
)) {
479 if (vfs_optionisset(vfsp
, MNTOPT_NOATIME
, NULL
)) {
482 } else if (vfs_optionisset(vfsp
, MNTOPT_ATIME
, NULL
)) {
488 * nbmand is a special property. It can only be changed at
491 * This is weird, but it is documented to only be changeable
494 if (vfs_optionisset(vfsp
, MNTOPT_NONBMAND
, NULL
)) {
496 } else if (vfs_optionisset(vfsp
, MNTOPT_NBMAND
, NULL
)) {
499 char osname
[MAXNAMELEN
];
501 dmu_objset_name(os
, osname
);
502 if (error
= dsl_prop_get_integer(osname
, "nbmand", &nbmand
,
509 * Register property callbacks.
511 * It would probably be fine to just check for i/o error from
512 * the first prop_register(), but I guess I like to go
515 ds
= dmu_objset_ds(os
);
516 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
517 error
= dsl_prop_register(ds
,
518 zfs_prop_to_name(ZFS_PROP_ATIME
), atime_changed_cb
, zfsvfs
);
519 error
= error
? error
: dsl_prop_register(ds
,
520 zfs_prop_to_name(ZFS_PROP_XATTR
), xattr_changed_cb
, zfsvfs
);
521 error
= error
? error
: dsl_prop_register(ds
,
522 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
), blksz_changed_cb
, zfsvfs
);
523 error
= error
? error
: dsl_prop_register(ds
,
524 zfs_prop_to_name(ZFS_PROP_READONLY
), readonly_changed_cb
, zfsvfs
);
525 error
= error
? error
: dsl_prop_register(ds
,
526 zfs_prop_to_name(ZFS_PROP_DEVICES
), devices_changed_cb
, zfsvfs
);
527 error
= error
? error
: dsl_prop_register(ds
,
528 zfs_prop_to_name(ZFS_PROP_SETUID
), setuid_changed_cb
, zfsvfs
);
529 error
= error
? error
: dsl_prop_register(ds
,
530 zfs_prop_to_name(ZFS_PROP_EXEC
), exec_changed_cb
, zfsvfs
);
531 error
= error
? error
: dsl_prop_register(ds
,
532 zfs_prop_to_name(ZFS_PROP_SNAPDIR
), snapdir_changed_cb
, zfsvfs
);
533 error
= error
? error
: dsl_prop_register(ds
,
534 zfs_prop_to_name(ZFS_PROP_ACLMODE
), acl_mode_changed_cb
, zfsvfs
);
535 error
= error
? error
: dsl_prop_register(ds
,
536 zfs_prop_to_name(ZFS_PROP_ACLINHERIT
), acl_inherit_changed_cb
,
538 error
= error
? error
: dsl_prop_register(ds
,
539 zfs_prop_to_name(ZFS_PROP_VSCAN
), vscan_changed_cb
, zfsvfs
);
540 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
545 * Invoke our callbacks to restore temporary mount options.
548 readonly_changed_cb(zfsvfs
, readonly
);
550 setuid_changed_cb(zfsvfs
, setuid
);
552 exec_changed_cb(zfsvfs
, exec
);
554 devices_changed_cb(zfsvfs
, devices
);
556 xattr_changed_cb(zfsvfs
, xattr
);
558 atime_changed_cb(zfsvfs
, atime
);
560 nbmand_changed_cb(zfsvfs
, nbmand
);
565 dsl_prop_unregister_all(ds
, zfsvfs
);
570 zfs_space_delta_cb(dmu_object_type_t bonustype
, void *data
,
571 uint64_t *userp
, uint64_t *groupp
)
574 * Is it a valid type of object to track?
576 if (bonustype
!= DMU_OT_ZNODE
&& bonustype
!= DMU_OT_SA
)
577 return (SET_ERROR(ENOENT
));
580 * If we have a NULL data pointer
581 * then assume the id's aren't changing and
582 * return EEXIST to the dmu to let it know to
586 return (SET_ERROR(EEXIST
));
588 if (bonustype
== DMU_OT_ZNODE
) {
589 znode_phys_t
*znp
= data
;
590 *userp
= znp
->zp_uid
;
591 *groupp
= znp
->zp_gid
;
594 sa_hdr_phys_t
*sap
= data
;
595 sa_hdr_phys_t sa
= *sap
;
596 boolean_t swap
= B_FALSE
;
598 ASSERT(bonustype
== DMU_OT_SA
);
600 if (sa
.sa_magic
== 0) {
602 * This should only happen for newly created
603 * files that haven't had the znode data filled
610 if (sa
.sa_magic
== BSWAP_32(SA_MAGIC
)) {
611 sa
.sa_magic
= SA_MAGIC
;
612 sa
.sa_layout_info
= BSWAP_16(sa
.sa_layout_info
);
615 VERIFY3U(sa
.sa_magic
, ==, SA_MAGIC
);
618 hdrsize
= sa_hdrsize(&sa
);
619 VERIFY3U(hdrsize
, >=, sizeof (sa_hdr_phys_t
));
620 *userp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+
622 *groupp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+
625 *userp
= BSWAP_64(*userp
);
626 *groupp
= BSWAP_64(*groupp
);
633 fuidstr_to_sid(zfsvfs_t
*zfsvfs
, const char *fuidstr
,
634 char *domainbuf
, int buflen
, uid_t
*ridp
)
639 fuid
= strtonum(fuidstr
, NULL
);
641 domain
= zfs_fuid_find_by_idx(zfsvfs
, FUID_INDEX(fuid
));
643 (void) strlcpy(domainbuf
, domain
, buflen
);
646 *ridp
= FUID_RID(fuid
);
650 zfs_userquota_prop_to_obj(zfsvfs_t
*zfsvfs
, zfs_userquota_prop_t type
)
653 case ZFS_PROP_USERUSED
:
654 return (DMU_USERUSED_OBJECT
);
655 case ZFS_PROP_GROUPUSED
:
656 return (DMU_GROUPUSED_OBJECT
);
657 case ZFS_PROP_USERQUOTA
:
658 return (zfsvfs
->z_userquota_obj
);
659 case ZFS_PROP_GROUPQUOTA
:
660 return (zfsvfs
->z_groupquota_obj
);
666 zfs_userspace_many(zfsvfs_t
*zfsvfs
, zfs_userquota_prop_t type
,
667 uint64_t *cookiep
, void *vbuf
, uint64_t *bufsizep
)
672 zfs_useracct_t
*buf
= vbuf
;
675 if (!dmu_objset_userspace_present(zfsvfs
->z_os
))
676 return (SET_ERROR(ENOTSUP
));
678 obj
= zfs_userquota_prop_to_obj(zfsvfs
, type
);
684 for (zap_cursor_init_serialized(&zc
, zfsvfs
->z_os
, obj
, *cookiep
);
685 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
686 zap_cursor_advance(&zc
)) {
687 if ((uintptr_t)buf
- (uintptr_t)vbuf
+ sizeof (zfs_useracct_t
) >
691 fuidstr_to_sid(zfsvfs
, za
.za_name
,
692 buf
->zu_domain
, sizeof (buf
->zu_domain
), &buf
->zu_rid
);
694 buf
->zu_space
= za
.za_first_integer
;
700 ASSERT3U((uintptr_t)buf
- (uintptr_t)vbuf
, <=, *bufsizep
);
701 *bufsizep
= (uintptr_t)buf
- (uintptr_t)vbuf
;
702 *cookiep
= zap_cursor_serialize(&zc
);
703 zap_cursor_fini(&zc
);
708 * buf must be big enough (eg, 32 bytes)
711 id_to_fuidstr(zfsvfs_t
*zfsvfs
, const char *domain
, uid_t rid
,
712 char *buf
, boolean_t addok
)
717 if (domain
&& domain
[0]) {
718 domainid
= zfs_fuid_find_by_domain(zfsvfs
, domain
, NULL
, addok
);
720 return (SET_ERROR(ENOENT
));
722 fuid
= FUID_ENCODE(domainid
, rid
);
723 (void) sprintf(buf
, "%llx", (longlong_t
)fuid
);
728 zfs_userspace_one(zfsvfs_t
*zfsvfs
, zfs_userquota_prop_t type
,
729 const char *domain
, uint64_t rid
, uint64_t *valp
)
737 if (!dmu_objset_userspace_present(zfsvfs
->z_os
))
738 return (SET_ERROR(ENOTSUP
));
740 obj
= zfs_userquota_prop_to_obj(zfsvfs
, type
);
744 err
= id_to_fuidstr(zfsvfs
, domain
, rid
, buf
, B_FALSE
);
748 err
= zap_lookup(zfsvfs
->z_os
, obj
, buf
, 8, 1, valp
);
755 zfs_set_userquota(zfsvfs_t
*zfsvfs
, zfs_userquota_prop_t type
,
756 const char *domain
, uint64_t rid
, uint64_t quota
)
762 boolean_t fuid_dirtied
;
764 if (type
!= ZFS_PROP_USERQUOTA
&& type
!= ZFS_PROP_GROUPQUOTA
)
765 return (SET_ERROR(EINVAL
));
767 if (zfsvfs
->z_version
< ZPL_VERSION_USERSPACE
)
768 return (SET_ERROR(ENOTSUP
));
770 objp
= (type
== ZFS_PROP_USERQUOTA
) ? &zfsvfs
->z_userquota_obj
:
771 &zfsvfs
->z_groupquota_obj
;
773 err
= id_to_fuidstr(zfsvfs
, domain
, rid
, buf
, B_TRUE
);
776 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
778 tx
= dmu_tx_create(zfsvfs
->z_os
);
779 dmu_tx_hold_zap(tx
, *objp
? *objp
: DMU_NEW_OBJECT
, B_TRUE
, NULL
);
781 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
782 zfs_userquota_prop_prefixes
[type
]);
785 zfs_fuid_txhold(zfsvfs
, tx
);
786 err
= dmu_tx_assign(tx
, TXG_WAIT
);
792 mutex_enter(&zfsvfs
->z_lock
);
794 *objp
= zap_create(zfsvfs
->z_os
, DMU_OT_USERGROUP_QUOTA
,
796 VERIFY(0 == zap_add(zfsvfs
->z_os
, MASTER_NODE_OBJ
,
797 zfs_userquota_prop_prefixes
[type
], 8, 1, objp
, tx
));
799 mutex_exit(&zfsvfs
->z_lock
);
802 err
= zap_remove(zfsvfs
->z_os
, *objp
, buf
, tx
);
806 err
= zap_update(zfsvfs
->z_os
, *objp
, buf
, 8, 1, "a
, tx
);
810 zfs_fuid_sync(zfsvfs
, tx
);
816 zfs_fuid_overquota(zfsvfs_t
*zfsvfs
, boolean_t isgroup
, uint64_t fuid
)
819 uint64_t used
, quota
, usedobj
, quotaobj
;
822 usedobj
= isgroup
? DMU_GROUPUSED_OBJECT
: DMU_USERUSED_OBJECT
;
823 quotaobj
= isgroup
? zfsvfs
->z_groupquota_obj
: zfsvfs
->z_userquota_obj
;
825 if (quotaobj
== 0 || zfsvfs
->z_replay
)
828 (void) sprintf(buf
, "%llx", (longlong_t
)fuid
);
829 err
= zap_lookup(zfsvfs
->z_os
, quotaobj
, buf
, 8, 1, "a
);
833 err
= zap_lookup(zfsvfs
->z_os
, usedobj
, buf
, 8, 1, &used
);
836 return (used
>= quota
);
840 zfs_owner_overquota(zfsvfs_t
*zfsvfs
, znode_t
*zp
, boolean_t isgroup
)
845 quotaobj
= isgroup
? zfsvfs
->z_groupquota_obj
: zfsvfs
->z_userquota_obj
;
847 fuid
= isgroup
? zp
->z_gid
: zp
->z_uid
;
849 if (quotaobj
== 0 || zfsvfs
->z_replay
)
852 return (zfs_fuid_overquota(zfsvfs
, isgroup
, fuid
));
856 zfsvfs_create(const char *osname
, zfsvfs_t
**zfvp
)
864 zfsvfs
= kmem_zalloc(sizeof (zfsvfs_t
), KM_SLEEP
);
867 * We claim to always be readonly so we can open snapshots;
868 * other ZPL code will prevent us from writing to snapshots.
870 error
= dmu_objset_own(osname
, DMU_OST_ZFS
, B_TRUE
, zfsvfs
, &os
);
872 kmem_free(zfsvfs
, sizeof (zfsvfs_t
));
877 * Initialize the zfs-specific filesystem structure.
878 * Should probably make this a kmem cache, shuffle fields,
879 * and just bzero up to z_hold_mtx[].
881 zfsvfs
->z_vfs
= NULL
;
882 zfsvfs
->z_parent
= zfsvfs
;
883 zfsvfs
->z_max_blksz
= SPA_OLD_MAXBLOCKSIZE
;
884 zfsvfs
->z_show_ctldir
= ZFS_SNAPDIR_VISIBLE
;
887 error
= zfs_get_zplprop(os
, ZFS_PROP_VERSION
, &zfsvfs
->z_version
);
890 } else if (zfsvfs
->z_version
>
891 zfs_zpl_version_map(spa_version(dmu_objset_spa(os
)))) {
892 (void) printf("Can't mount a version %lld file system "
893 "on a version %lld pool\n. Pool must be upgraded to mount "
894 "this file system.", (u_longlong_t
)zfsvfs
->z_version
,
895 (u_longlong_t
)spa_version(dmu_objset_spa(os
)));
896 error
= SET_ERROR(ENOTSUP
);
899 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_NORMALIZE
, &zval
)) != 0)
901 zfsvfs
->z_norm
= (int)zval
;
903 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_UTF8ONLY
, &zval
)) != 0)
905 zfsvfs
->z_utf8
= (zval
!= 0);
907 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_CASE
, &zval
)) != 0)
909 zfsvfs
->z_case
= (uint_t
)zval
;
912 * Fold case on file systems that are always or sometimes case
915 if (zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
||
916 zfsvfs
->z_case
== ZFS_CASE_MIXED
)
917 zfsvfs
->z_norm
|= U8_TEXTPREP_TOUPPER
;
919 zfsvfs
->z_use_fuids
= USE_FUIDS(zfsvfs
->z_version
, zfsvfs
->z_os
);
920 zfsvfs
->z_use_sa
= USE_SA(zfsvfs
->z_version
, zfsvfs
->z_os
);
922 if (zfsvfs
->z_use_sa
) {
923 /* should either have both of these objects or none */
924 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SA_ATTRS
, 8, 1,
930 * Pre SA versions file systems should never touch
931 * either the attribute registration or layout objects.
936 error
= sa_setup(os
, sa_obj
, zfs_attr_table
, ZPL_END
,
937 &zfsvfs
->z_attr_table
);
941 if (zfsvfs
->z_version
>= ZPL_VERSION_SA
)
942 sa_register_update_callback(os
, zfs_sa_upgrade
);
944 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_ROOT_OBJ
, 8, 1,
948 ASSERT(zfsvfs
->z_root
!= 0);
950 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_UNLINKED_SET
, 8, 1,
951 &zfsvfs
->z_unlinkedobj
);
955 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
956 zfs_userquota_prop_prefixes
[ZFS_PROP_USERQUOTA
],
957 8, 1, &zfsvfs
->z_userquota_obj
);
958 if (error
&& error
!= ENOENT
)
961 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
962 zfs_userquota_prop_prefixes
[ZFS_PROP_GROUPQUOTA
],
963 8, 1, &zfsvfs
->z_groupquota_obj
);
964 if (error
&& error
!= ENOENT
)
967 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_FUID_TABLES
, 8, 1,
968 &zfsvfs
->z_fuid_obj
);
969 if (error
&& error
!= ENOENT
)
972 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SHARES_DIR
, 8, 1,
973 &zfsvfs
->z_shares_dir
);
974 if (error
&& error
!= ENOENT
)
977 mutex_init(&zfsvfs
->z_znodes_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
978 mutex_init(&zfsvfs
->z_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
979 list_create(&zfsvfs
->z_all_znodes
, sizeof (znode_t
),
980 offsetof(znode_t
, z_link_node
));
981 rrm_init(&zfsvfs
->z_teardown_lock
, B_FALSE
);
982 rw_init(&zfsvfs
->z_teardown_inactive_lock
, NULL
, RW_DEFAULT
, NULL
);
983 rw_init(&zfsvfs
->z_fuid_lock
, NULL
, RW_DEFAULT
, NULL
);
984 for (i
= 0; i
!= ZFS_OBJ_MTX_SZ
; i
++)
985 mutex_init(&zfsvfs
->z_hold_mtx
[i
], NULL
, MUTEX_DEFAULT
, NULL
);
991 dmu_objset_disown(os
, zfsvfs
);
993 kmem_free(zfsvfs
, sizeof (zfsvfs_t
));
998 zfsvfs_setup(zfsvfs_t
*zfsvfs
, boolean_t mounting
)
1002 error
= zfs_register_callbacks(zfsvfs
->z_vfs
);
1007 * Set the objset user_ptr to track its zfsvfs.
1009 mutex_enter(&zfsvfs
->z_os
->os_user_ptr_lock
);
1010 dmu_objset_set_user(zfsvfs
->z_os
, zfsvfs
);
1011 mutex_exit(&zfsvfs
->z_os
->os_user_ptr_lock
);
1013 zfsvfs
->z_log
= zil_open(zfsvfs
->z_os
, zfs_get_data
);
1016 * If we are not mounting (ie: online recv), then we don't
1017 * have to worry about replaying the log as we blocked all
1018 * operations out since we closed the ZIL.
1024 * During replay we remove the read only flag to
1025 * allow replays to succeed.
1027 readonly
= zfsvfs
->z_vfs
->vfs_flag
& VFS_RDONLY
;
1029 zfsvfs
->z_vfs
->vfs_flag
&= ~VFS_RDONLY
;
1031 zfs_unlinked_drain(zfsvfs
);
1034 * Parse and replay the intent log.
1036 * Because of ziltest, this must be done after
1037 * zfs_unlinked_drain(). (Further note: ziltest
1038 * doesn't use readonly mounts, where
1039 * zfs_unlinked_drain() isn't called.) This is because
1040 * ziltest causes spa_sync() to think it's committed,
1041 * but actually it is not, so the intent log contains
1042 * many txg's worth of changes.
1044 * In particular, if object N is in the unlinked set in
1045 * the last txg to actually sync, then it could be
1046 * actually freed in a later txg and then reallocated
1047 * in a yet later txg. This would write a "create
1048 * object N" record to the intent log. Normally, this
1049 * would be fine because the spa_sync() would have
1050 * written out the fact that object N is free, before
1051 * we could write the "create object N" intent log
1054 * But when we are in ziltest mode, we advance the "open
1055 * txg" without actually spa_sync()-ing the changes to
1056 * disk. So we would see that object N is still
1057 * allocated and in the unlinked set, and there is an
1058 * intent log record saying to allocate it.
1060 if (spa_writeable(dmu_objset_spa(zfsvfs
->z_os
))) {
1061 if (zil_replay_disable
) {
1062 zil_destroy(zfsvfs
->z_log
, B_FALSE
);
1064 zfsvfs
->z_replay
= B_TRUE
;
1065 zil_replay(zfsvfs
->z_os
, zfsvfs
,
1067 zfsvfs
->z_replay
= B_FALSE
;
1070 zfsvfs
->z_vfs
->vfs_flag
|= readonly
; /* restore readonly bit */
1077 zfsvfs_free(zfsvfs_t
*zfsvfs
)
1080 extern krwlock_t zfsvfs_lock
; /* in zfs_znode.c */
1083 * This is a barrier to prevent the filesystem from going away in
1084 * zfs_znode_move() until we can safely ensure that the filesystem is
1085 * not unmounted. We consider the filesystem valid before the barrier
1086 * and invalid after the barrier.
1088 rw_enter(&zfsvfs_lock
, RW_READER
);
1089 rw_exit(&zfsvfs_lock
);
1091 zfs_fuid_destroy(zfsvfs
);
1093 mutex_destroy(&zfsvfs
->z_znodes_lock
);
1094 mutex_destroy(&zfsvfs
->z_lock
);
1095 list_destroy(&zfsvfs
->z_all_znodes
);
1096 rrm_destroy(&zfsvfs
->z_teardown_lock
);
1097 rw_destroy(&zfsvfs
->z_teardown_inactive_lock
);
1098 rw_destroy(&zfsvfs
->z_fuid_lock
);
1099 for (i
= 0; i
!= ZFS_OBJ_MTX_SZ
; i
++)
1100 mutex_destroy(&zfsvfs
->z_hold_mtx
[i
]);
1101 kmem_free(zfsvfs
, sizeof (zfsvfs_t
));
1105 zfs_set_fuid_feature(zfsvfs_t
*zfsvfs
)
1107 zfsvfs
->z_use_fuids
= USE_FUIDS(zfsvfs
->z_version
, zfsvfs
->z_os
);
1108 if (zfsvfs
->z_vfs
) {
1109 if (zfsvfs
->z_use_fuids
) {
1110 vfs_set_feature(zfsvfs
->z_vfs
, VFSFT_XVATTR
);
1111 vfs_set_feature(zfsvfs
->z_vfs
, VFSFT_SYSATTR_VIEWS
);
1112 vfs_set_feature(zfsvfs
->z_vfs
, VFSFT_ACEMASKONACCESS
);
1113 vfs_set_feature(zfsvfs
->z_vfs
, VFSFT_ACLONCREATE
);
1114 vfs_set_feature(zfsvfs
->z_vfs
, VFSFT_ACCESS_FILTER
);
1115 vfs_set_feature(zfsvfs
->z_vfs
, VFSFT_REPARSE
);
1117 vfs_clear_feature(zfsvfs
->z_vfs
, VFSFT_XVATTR
);
1118 vfs_clear_feature(zfsvfs
->z_vfs
, VFSFT_SYSATTR_VIEWS
);
1119 vfs_clear_feature(zfsvfs
->z_vfs
, VFSFT_ACEMASKONACCESS
);
1120 vfs_clear_feature(zfsvfs
->z_vfs
, VFSFT_ACLONCREATE
);
1121 vfs_clear_feature(zfsvfs
->z_vfs
, VFSFT_ACCESS_FILTER
);
1122 vfs_clear_feature(zfsvfs
->z_vfs
, VFSFT_REPARSE
);
1125 zfsvfs
->z_use_sa
= USE_SA(zfsvfs
->z_version
, zfsvfs
->z_os
);
1129 zfs_domount(vfs_t
*vfsp
, char *osname
)
1132 uint64_t recordsize
, fsid_guid
;
1139 error
= zfsvfs_create(osname
, &zfsvfs
);
1142 zfsvfs
->z_vfs
= vfsp
;
1144 /* Initialize the generic filesystem structure. */
1145 vfsp
->vfs_bcount
= 0;
1146 vfsp
->vfs_data
= NULL
;
1148 if (zfs_create_unique_device(&mount_dev
) == -1) {
1149 error
= SET_ERROR(ENODEV
);
1152 ASSERT(vfs_devismounted(mount_dev
) == 0);
1154 if (error
= dsl_prop_get_integer(osname
, "recordsize", &recordsize
,
1158 vfsp
->vfs_dev
= mount_dev
;
1159 vfsp
->vfs_fstype
= zfsfstype
;
1160 vfsp
->vfs_bsize
= recordsize
;
1161 vfsp
->vfs_flag
|= VFS_NOTRUNC
;
1162 vfsp
->vfs_data
= zfsvfs
;
1165 * The fsid is 64 bits, composed of an 8-bit fs type, which
1166 * separates our fsid from any other filesystem types, and a
1167 * 56-bit objset unique ID. The objset unique ID is unique to
1168 * all objsets open on this system, provided by unique_create().
1169 * The 8-bit fs type must be put in the low bits of fsid[1]
1170 * because that's where other Solaris filesystems put it.
1172 fsid_guid
= dmu_objset_fsid_guid(zfsvfs
->z_os
);
1173 ASSERT((fsid_guid
& ~((1ULL<<56)-1)) == 0);
1174 vfsp
->vfs_fsid
.val
[0] = fsid_guid
;
1175 vfsp
->vfs_fsid
.val
[1] = ((fsid_guid
>>32) << 8) |
1179 * Set features for file system.
1181 zfs_set_fuid_feature(zfsvfs
);
1182 if (zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
) {
1183 vfs_set_feature(vfsp
, VFSFT_DIRENTFLAGS
);
1184 vfs_set_feature(vfsp
, VFSFT_CASEINSENSITIVE
);
1185 vfs_set_feature(vfsp
, VFSFT_NOCASESENSITIVE
);
1186 } else if (zfsvfs
->z_case
== ZFS_CASE_MIXED
) {
1187 vfs_set_feature(vfsp
, VFSFT_DIRENTFLAGS
);
1188 vfs_set_feature(vfsp
, VFSFT_CASEINSENSITIVE
);
1190 vfs_set_feature(vfsp
, VFSFT_ZEROCOPY_SUPPORTED
);
1192 if (dmu_objset_is_snapshot(zfsvfs
->z_os
)) {
1195 atime_changed_cb(zfsvfs
, B_FALSE
);
1196 readonly_changed_cb(zfsvfs
, B_TRUE
);
1197 if (error
= dsl_prop_get_integer(osname
, "xattr", &pval
, NULL
))
1199 xattr_changed_cb(zfsvfs
, pval
);
1200 zfsvfs
->z_issnap
= B_TRUE
;
1201 zfsvfs
->z_os
->os_sync
= ZFS_SYNC_DISABLED
;
1203 mutex_enter(&zfsvfs
->z_os
->os_user_ptr_lock
);
1204 dmu_objset_set_user(zfsvfs
->z_os
, zfsvfs
);
1205 mutex_exit(&zfsvfs
->z_os
->os_user_ptr_lock
);
1207 error
= zfsvfs_setup(zfsvfs
, B_TRUE
);
1210 if (!zfsvfs
->z_issnap
)
1211 zfsctl_create(zfsvfs
);
1214 dmu_objset_disown(zfsvfs
->z_os
, zfsvfs
);
1215 zfsvfs_free(zfsvfs
);
1217 atomic_inc_32(&zfs_active_fs_count
);
1224 zfs_unregister_callbacks(zfsvfs_t
*zfsvfs
)
1226 objset_t
*os
= zfsvfs
->z_os
;
1228 if (!dmu_objset_is_snapshot(os
))
1229 dsl_prop_unregister_all(dmu_objset_ds(os
), zfsvfs
);
1233 * Convert a decimal digit string to a uint64_t integer.
1236 str_to_uint64(char *str
, uint64_t *objnum
)
1241 if (*str
< '0' || *str
> '9')
1242 return (SET_ERROR(EINVAL
));
1244 num
= num
*10 + *str
++ - '0';
1252 * The boot path passed from the boot loader is in the form of
1253 * "rootpool-name/root-filesystem-object-number'. Convert this
1254 * string to a dataset name: "rootpool-name/root-filesystem-name".
1257 zfs_parse_bootfs(char *bpath
, char *outpath
)
1263 if (*bpath
== 0 || *bpath
== '/')
1264 return (SET_ERROR(EINVAL
));
1266 (void) strcpy(outpath
, bpath
);
1268 slashp
= strchr(bpath
, '/');
1270 /* if no '/', just return the pool name */
1271 if (slashp
== NULL
) {
1275 /* if not a number, just return the root dataset name */
1276 if (str_to_uint64(slashp
+1, &objnum
)) {
1281 error
= dsl_dsobj_to_dsname(bpath
, objnum
, outpath
);
1288 * Check that the hex label string is appropriate for the dataset being
1289 * mounted into the global_zone proper.
1291 * Return an error if the hex label string is not default or
1292 * admin_low/admin_high. For admin_low labels, the corresponding
1293 * dataset must be readonly.
1296 zfs_check_global_label(const char *dsname
, const char *hexsl
)
1298 if (strcasecmp(hexsl
, ZFS_MLSLABEL_DEFAULT
) == 0)
1300 if (strcasecmp(hexsl
, ADMIN_HIGH
) == 0)
1302 if (strcasecmp(hexsl
, ADMIN_LOW
) == 0) {
1303 /* must be readonly */
1306 if (dsl_prop_get_integer(dsname
,
1307 zfs_prop_to_name(ZFS_PROP_READONLY
), &rdonly
, NULL
))
1308 return (SET_ERROR(EACCES
));
1309 return (rdonly
? 0 : EACCES
);
1311 return (SET_ERROR(EACCES
));
1315 * Determine whether the mount is allowed according to MAC check.
1316 * by comparing (where appropriate) label of the dataset against
1317 * the label of the zone being mounted into. If the dataset has
1318 * no label, create one.
1320 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1323 zfs_mount_label_policy(vfs_t
*vfsp
, char *osname
)
1326 zone_t
*mntzone
= NULL
;
1327 ts_label_t
*mnt_tsl
;
1330 char ds_hexsl
[MAXNAMELEN
];
1332 retv
= EACCES
; /* assume the worst */
1335 * Start by getting the dataset label if it exists.
1337 error
= dsl_prop_get(osname
, zfs_prop_to_name(ZFS_PROP_MLSLABEL
),
1338 1, sizeof (ds_hexsl
), &ds_hexsl
, NULL
);
1340 return (SET_ERROR(EACCES
));
1343 * If labeling is NOT enabled, then disallow the mount of datasets
1344 * which have a non-default label already. No other label checks
1347 if (!is_system_labeled()) {
1348 if (strcasecmp(ds_hexsl
, ZFS_MLSLABEL_DEFAULT
) == 0)
1350 return (SET_ERROR(EACCES
));
1354 * Get the label of the mountpoint. If mounting into the global
1355 * zone (i.e. mountpoint is not within an active zone and the
1356 * zoned property is off), the label must be default or
1357 * admin_low/admin_high only; no other checks are needed.
1359 mntzone
= zone_find_by_any_path(refstr_value(vfsp
->vfs_mntpt
), B_FALSE
);
1360 if (mntzone
->zone_id
== GLOBAL_ZONEID
) {
1365 if (dsl_prop_get_integer(osname
,
1366 zfs_prop_to_name(ZFS_PROP_ZONED
), &zoned
, NULL
))
1367 return (SET_ERROR(EACCES
));
1369 return (zfs_check_global_label(osname
, ds_hexsl
));
1372 * This is the case of a zone dataset being mounted
1373 * initially, before the zone has been fully created;
1374 * allow this mount into global zone.
1379 mnt_tsl
= mntzone
->zone_slabel
;
1380 ASSERT(mnt_tsl
!= NULL
);
1381 label_hold(mnt_tsl
);
1382 mnt_sl
= label2bslabel(mnt_tsl
);
1384 if (strcasecmp(ds_hexsl
, ZFS_MLSLABEL_DEFAULT
) == 0) {
1386 * The dataset doesn't have a real label, so fabricate one.
1390 if (l_to_str_internal(mnt_sl
, &str
) == 0 &&
1391 dsl_prop_set_string(osname
,
1392 zfs_prop_to_name(ZFS_PROP_MLSLABEL
),
1393 ZPROP_SRC_LOCAL
, str
) == 0)
1396 kmem_free(str
, strlen(str
) + 1);
1397 } else if (hexstr_to_label(ds_hexsl
, &ds_sl
) == 0) {
1399 * Now compare labels to complete the MAC check. If the
1400 * labels are equal then allow access. If the mountpoint
1401 * label dominates the dataset label, allow readonly access.
1402 * Otherwise, access is denied.
1404 if (blequal(mnt_sl
, &ds_sl
))
1406 else if (bldominates(mnt_sl
, &ds_sl
)) {
1407 vfs_setmntopt(vfsp
, MNTOPT_RO
, NULL
, 0);
1412 label_rele(mnt_tsl
);
1418 zfs_mountroot(vfs_t
*vfsp
, enum whymountroot why
)
1421 static int zfsrootdone
= 0;
1422 zfsvfs_t
*zfsvfs
= NULL
;
1431 * The filesystem that we mount as root is defined in the
1432 * boot property "zfs-bootfs" with a format of
1433 * "poolname/root-dataset-objnum".
1435 if (why
== ROOT_INIT
) {
1437 return (SET_ERROR(EBUSY
));
1439 * the process of doing a spa_load will require the
1440 * clock to be set before we could (for example) do
1441 * something better by looking at the timestamp on
1442 * an uberblock, so just set it to -1.
1446 if ((zfs_bootfs
= spa_get_bootprop("zfs-bootfs")) == NULL
) {
1447 cmn_err(CE_NOTE
, "spa_get_bootfs: can not get "
1449 return (SET_ERROR(EINVAL
));
1451 zfs_devid
= spa_get_bootprop("diskdevid");
1452 error
= spa_import_rootpool(rootfs
.bo_name
, zfs_devid
);
1454 spa_free_bootprop(zfs_devid
);
1456 spa_free_bootprop(zfs_bootfs
);
1457 cmn_err(CE_NOTE
, "spa_import_rootpool: error %d",
1461 if (error
= zfs_parse_bootfs(zfs_bootfs
, rootfs
.bo_name
)) {
1462 spa_free_bootprop(zfs_bootfs
);
1463 cmn_err(CE_NOTE
, "zfs_parse_bootfs: error %d",
1468 spa_free_bootprop(zfs_bootfs
);
1470 if (error
= vfs_lock(vfsp
))
1473 if (error
= zfs_domount(vfsp
, rootfs
.bo_name
)) {
1474 cmn_err(CE_NOTE
, "zfs_domount: error %d", error
);
1478 zfsvfs
= (zfsvfs_t
*)vfsp
->vfs_data
;
1480 if (error
= zfs_zget(zfsvfs
, zfsvfs
->z_root
, &zp
)) {
1481 cmn_err(CE_NOTE
, "zfs_zget: error %d", error
);
1486 mutex_enter(&vp
->v_lock
);
1487 vp
->v_flag
|= VROOT
;
1488 mutex_exit(&vp
->v_lock
);
1492 * Leave rootvp held. The root file system is never unmounted.
1495 vfs_add((struct vnode
*)0, vfsp
,
1496 (vfsp
->vfs_flag
& VFS_RDONLY
) ? MS_RDONLY
: 0);
1500 } else if (why
== ROOT_REMOUNT
) {
1501 readonly_changed_cb(vfsp
->vfs_data
, B_FALSE
);
1502 vfsp
->vfs_flag
|= VFS_REMOUNT
;
1504 /* refresh mount options */
1505 zfs_unregister_callbacks(vfsp
->vfs_data
);
1506 return (zfs_register_callbacks(vfsp
));
1508 } else if (why
== ROOT_UNMOUNT
) {
1509 zfs_unregister_callbacks((zfsvfs_t
*)vfsp
->vfs_data
);
1510 (void) zfs_sync(vfsp
, 0, 0);
1515 * if "why" is equal to anything else other than ROOT_INIT,
1516 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1518 return (SET_ERROR(ENOTSUP
));
1523 zfs_mount(vfs_t
*vfsp
, vnode_t
*mvp
, struct mounta
*uap
, cred_t
*cr
)
1528 uio_seg_t fromspace
= (uap
->flags
& MS_SYSSPACE
) ?
1529 UIO_SYSSPACE
: UIO_USERSPACE
;
1532 if (mvp
->v_type
!= VDIR
)
1533 return (SET_ERROR(ENOTDIR
));
1535 mutex_enter(&mvp
->v_lock
);
1536 if ((uap
->flags
& MS_REMOUNT
) == 0 &&
1537 (uap
->flags
& MS_OVERLAY
) == 0 &&
1538 (mvp
->v_count
!= 1 || (mvp
->v_flag
& VROOT
))) {
1539 mutex_exit(&mvp
->v_lock
);
1540 return (SET_ERROR(EBUSY
));
1542 mutex_exit(&mvp
->v_lock
);
1545 * ZFS does not support passing unparsed data in via MS_DATA.
1546 * Users should use the MS_OPTIONSTR interface; this means
1547 * that all option parsing is already done and the options struct
1548 * can be interrogated.
1550 if ((uap
->flags
& MS_DATA
) && uap
->datalen
> 0)
1551 return (SET_ERROR(EINVAL
));
1554 * Get the objset name (the "special" mount argument).
1556 if (error
= pn_get(uap
->spec
, fromspace
, &spn
))
1559 osname
= spn
.pn_path
;
1562 * Check for mount privilege?
1564 * If we don't have privilege then see if
1565 * we have local permission to allow it
1567 error
= secpolicy_fs_mount(cr
, mvp
, vfsp
);
1569 if (dsl_deleg_access(osname
, ZFS_DELEG_PERM_MOUNT
, cr
) == 0) {
1573 * Make sure user is the owner of the mount point
1574 * or has sufficient privileges.
1577 vattr
.va_mask
= AT_UID
;
1579 if (VOP_GETATTR(mvp
, &vattr
, 0, cr
, NULL
)) {
1583 if (secpolicy_vnode_owner(cr
, vattr
.va_uid
) != 0 &&
1584 VOP_ACCESS(mvp
, VWRITE
, 0, cr
, NULL
) != 0) {
1587 secpolicy_fs_mount_clearopts(cr
, vfsp
);
1594 * Refuse to mount a filesystem if we are in a local zone and the
1595 * dataset is not visible.
1597 if (!INGLOBALZONE(curproc
) &&
1598 (!zone_dataset_visible(osname
, &canwrite
) || !canwrite
)) {
1599 error
= SET_ERROR(EPERM
);
1603 error
= zfs_mount_label_policy(vfsp
, osname
);
1608 * When doing a remount, we simply refresh our temporary properties
1609 * according to those options set in the current VFS options.
1611 if (uap
->flags
& MS_REMOUNT
) {
1612 /* refresh mount options */
1613 zfs_unregister_callbacks(vfsp
->vfs_data
);
1614 error
= zfs_register_callbacks(vfsp
);
1618 error
= zfs_domount(vfsp
, osname
);
1621 * Add an extra VFS_HOLD on our parent vfs so that it can't
1622 * disappear due to a forced unmount.
1624 if (error
== 0 && ((zfsvfs_t
*)vfsp
->vfs_data
)->z_issnap
)
1625 VFS_HOLD(mvp
->v_vfsp
);
1633 zfs_statvfs(vfs_t
*vfsp
, struct statvfs64
*statp
)
1635 zfsvfs_t
*zfsvfs
= vfsp
->vfs_data
;
1637 uint64_t refdbytes
, availbytes
, usedobjs
, availobjs
;
1641 dmu_objset_space(zfsvfs
->z_os
,
1642 &refdbytes
, &availbytes
, &usedobjs
, &availobjs
);
1645 * The underlying storage pool actually uses multiple block sizes.
1646 * We report the fragsize as the smallest block size we support,
1647 * and we report our blocksize as the filesystem's maximum blocksize.
1649 statp
->f_frsize
= 1UL << SPA_MINBLOCKSHIFT
;
1650 statp
->f_bsize
= zfsvfs
->z_max_blksz
;
1653 * The following report "total" blocks of various kinds in the
1654 * file system, but reported in terms of f_frsize - the
1658 statp
->f_blocks
= (refdbytes
+ availbytes
) >> SPA_MINBLOCKSHIFT
;
1659 statp
->f_bfree
= availbytes
>> SPA_MINBLOCKSHIFT
;
1660 statp
->f_bavail
= statp
->f_bfree
; /* no root reservation */
1663 * statvfs() should really be called statufs(), because it assumes
1664 * static metadata. ZFS doesn't preallocate files, so the best
1665 * we can do is report the max that could possibly fit in f_files,
1666 * and that minus the number actually used in f_ffree.
1667 * For f_ffree, report the smaller of the number of object available
1668 * and the number of blocks (each object will take at least a block).
1670 statp
->f_ffree
= MIN(availobjs
, statp
->f_bfree
);
1671 statp
->f_favail
= statp
->f_ffree
; /* no "root reservation" */
1672 statp
->f_files
= statp
->f_ffree
+ usedobjs
;
1674 (void) cmpldev(&d32
, vfsp
->vfs_dev
);
1675 statp
->f_fsid
= d32
;
1678 * We're a zfs filesystem.
1680 (void) strcpy(statp
->f_basetype
, vfssw
[vfsp
->vfs_fstype
].vsw_name
);
1682 statp
->f_flag
= vf_to_stf(vfsp
->vfs_flag
);
1684 statp
->f_namemax
= ZFS_MAXNAMELEN
;
1687 * We have all of 32 characters to stuff a string here.
1688 * Is there anything useful we could/should provide?
1690 bzero(statp
->f_fstr
, sizeof (statp
->f_fstr
));
1697 zfs_root(vfs_t
*vfsp
, vnode_t
**vpp
)
1699 zfsvfs_t
*zfsvfs
= vfsp
->vfs_data
;
1705 error
= zfs_zget(zfsvfs
, zfsvfs
->z_root
, &rootzp
);
1707 *vpp
= ZTOV(rootzp
);
1714 * Teardown the zfsvfs::z_os.
1716 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1717 * and 'z_teardown_inactive_lock' held.
1720 zfsvfs_teardown(zfsvfs_t
*zfsvfs
, boolean_t unmounting
)
1724 rrm_enter(&zfsvfs
->z_teardown_lock
, RW_WRITER
, FTAG
);
1728 * We purge the parent filesystem's vfsp as the parent
1729 * filesystem and all of its snapshots have their vnode's
1730 * v_vfsp set to the parent's filesystem's vfsp. Note,
1731 * 'z_parent' is self referential for non-snapshots.
1733 (void) dnlc_purge_vfsp(zfsvfs
->z_parent
->z_vfs
, 0);
1737 * Close the zil. NB: Can't close the zil while zfs_inactive
1738 * threads are blocked as zil_close can call zfs_inactive.
1740 if (zfsvfs
->z_log
) {
1741 zil_close(zfsvfs
->z_log
);
1742 zfsvfs
->z_log
= NULL
;
1745 rw_enter(&zfsvfs
->z_teardown_inactive_lock
, RW_WRITER
);
1748 * If we are not unmounting (ie: online recv) and someone already
1749 * unmounted this file system while we were doing the switcheroo,
1750 * or a reopen of z_os failed then just bail out now.
1752 if (!unmounting
&& (zfsvfs
->z_unmounted
|| zfsvfs
->z_os
== NULL
)) {
1753 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
1754 rrm_exit(&zfsvfs
->z_teardown_lock
, FTAG
);
1755 return (SET_ERROR(EIO
));
1759 * At this point there are no vops active, and any new vops will
1760 * fail with EIO since we have z_teardown_lock for writer (only
1761 * relavent for forced unmount).
1763 * Release all holds on dbufs.
1765 mutex_enter(&zfsvfs
->z_znodes_lock
);
1766 for (zp
= list_head(&zfsvfs
->z_all_znodes
); zp
!= NULL
;
1767 zp
= list_next(&zfsvfs
->z_all_znodes
, zp
))
1769 ASSERT(ZTOV(zp
)->v_count
> 0);
1770 zfs_znode_dmu_fini(zp
);
1772 mutex_exit(&zfsvfs
->z_znodes_lock
);
1775 * If we are unmounting, set the unmounted flag and let new vops
1776 * unblock. zfs_inactive will have the unmounted behavior, and all
1777 * other vops will fail with EIO.
1780 zfsvfs
->z_unmounted
= B_TRUE
;
1781 rrm_exit(&zfsvfs
->z_teardown_lock
, FTAG
);
1782 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
1786 * z_os will be NULL if there was an error in attempting to reopen
1787 * zfsvfs, so just return as the properties had already been
1788 * unregistered and cached data had been evicted before.
1790 if (zfsvfs
->z_os
== NULL
)
1794 * Unregister properties.
1796 zfs_unregister_callbacks(zfsvfs
);
1801 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs
->z_os
)) &&
1802 !(zfsvfs
->z_vfs
->vfs_flag
& VFS_RDONLY
))
1803 txg_wait_synced(dmu_objset_pool(zfsvfs
->z_os
), 0);
1804 dmu_objset_evict_dbufs(zfsvfs
->z_os
);
1811 zfs_umount(vfs_t
*vfsp
, int fflag
, cred_t
*cr
)
1813 zfsvfs_t
*zfsvfs
= vfsp
->vfs_data
;
1817 ret
= secpolicy_fs_unmount(cr
, vfsp
);
1819 if (dsl_deleg_access((char *)refstr_value(vfsp
->vfs_resource
),
1820 ZFS_DELEG_PERM_MOUNT
, cr
))
1825 * We purge the parent filesystem's vfsp as the parent filesystem
1826 * and all of its snapshots have their vnode's v_vfsp set to the
1827 * parent's filesystem's vfsp. Note, 'z_parent' is self
1828 * referential for non-snapshots.
1830 (void) dnlc_purge_vfsp(zfsvfs
->z_parent
->z_vfs
, 0);
1833 * Unmount any snapshots mounted under .zfs before unmounting the
1836 if (zfsvfs
->z_ctldir
!= NULL
&&
1837 (ret
= zfsctl_umount_snapshots(vfsp
, fflag
, cr
)) != 0) {
1841 if (!(fflag
& MS_FORCE
)) {
1843 * Check the number of active vnodes in the file system.
1844 * Our count is maintained in the vfs structure, but the
1845 * number is off by 1 to indicate a hold on the vfs
1848 * The '.zfs' directory maintains a reference of its
1849 * own, and any active references underneath are
1850 * reflected in the vnode count.
1852 if (zfsvfs
->z_ctldir
== NULL
) {
1853 if (vfsp
->vfs_count
> 1)
1854 return (SET_ERROR(EBUSY
));
1856 if (vfsp
->vfs_count
> 2 ||
1857 zfsvfs
->z_ctldir
->v_count
> 1)
1858 return (SET_ERROR(EBUSY
));
1862 vfsp
->vfs_flag
|= VFS_UNMOUNTED
;
1864 VERIFY(zfsvfs_teardown(zfsvfs
, B_TRUE
) == 0);
1868 * z_os will be NULL if there was an error in
1869 * attempting to reopen zfsvfs.
1873 * Unset the objset user_ptr.
1875 mutex_enter(&os
->os_user_ptr_lock
);
1876 dmu_objset_set_user(os
, NULL
);
1877 mutex_exit(&os
->os_user_ptr_lock
);
1880 * Finally release the objset
1882 dmu_objset_disown(os
, zfsvfs
);
1886 * We can now safely destroy the '.zfs' directory node.
1888 if (zfsvfs
->z_ctldir
!= NULL
)
1889 zfsctl_destroy(zfsvfs
);
1895 zfs_vget(vfs_t
*vfsp
, vnode_t
**vpp
, fid_t
*fidp
)
1897 zfsvfs_t
*zfsvfs
= vfsp
->vfs_data
;
1899 uint64_t object
= 0;
1900 uint64_t fid_gen
= 0;
1909 if (fidp
->fid_len
== LONG_FID_LEN
) {
1910 zfid_long_t
*zlfid
= (zfid_long_t
*)fidp
;
1911 uint64_t objsetid
= 0;
1912 uint64_t setgen
= 0;
1914 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
1915 objsetid
|= ((uint64_t)zlfid
->zf_setid
[i
]) << (8 * i
);
1917 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
1918 setgen
|= ((uint64_t)zlfid
->zf_setgen
[i
]) << (8 * i
);
1922 err
= zfsctl_lookup_objset(vfsp
, objsetid
, &zfsvfs
);
1924 return (SET_ERROR(EINVAL
));
1928 if (fidp
->fid_len
== SHORT_FID_LEN
|| fidp
->fid_len
== LONG_FID_LEN
) {
1929 zfid_short_t
*zfid
= (zfid_short_t
*)fidp
;
1931 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
1932 object
|= ((uint64_t)zfid
->zf_object
[i
]) << (8 * i
);
1934 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
1935 fid_gen
|= ((uint64_t)zfid
->zf_gen
[i
]) << (8 * i
);
1938 return (SET_ERROR(EINVAL
));
1941 /* A zero fid_gen means we are in the .zfs control directories */
1943 (object
== ZFSCTL_INO_ROOT
|| object
== ZFSCTL_INO_SNAPDIR
)) {
1944 *vpp
= zfsvfs
->z_ctldir
;
1945 ASSERT(*vpp
!= NULL
);
1946 if (object
== ZFSCTL_INO_SNAPDIR
) {
1947 VERIFY(zfsctl_root_lookup(*vpp
, "snapshot", vpp
, NULL
,
1948 0, NULL
, NULL
, NULL
, NULL
, NULL
) == 0);
1956 gen_mask
= -1ULL >> (64 - 8 * i
);
1958 dprintf("getting %llu [%u mask %llx]\n", object
, fid_gen
, gen_mask
);
1959 if (err
= zfs_zget(zfsvfs
, object
, &zp
)) {
1963 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zfsvfs
), &zp_gen
,
1965 zp_gen
= zp_gen
& gen_mask
;
1968 if (zp
->z_unlinked
|| zp_gen
!= fid_gen
) {
1969 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen
, fid_gen
);
1972 return (SET_ERROR(EINVAL
));
1981 * Block out VOPs and close zfsvfs_t::z_os
1983 * Note, if successful, then we return with the 'z_teardown_lock' and
1984 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
1985 * dataset and objset intact so that they can be atomically handed off during
1986 * a subsequent rollback or recv operation and the resume thereafter.
1989 zfs_suspend_fs(zfsvfs_t
*zfsvfs
)
1993 if ((error
= zfsvfs_teardown(zfsvfs
, B_FALSE
)) != 0)
2000 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2001 * is an invariant across any of the operations that can be performed while the
2002 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2003 * are the same: the relevant objset and associated dataset are owned by
2004 * zfsvfs, held, and long held on entry.
2007 zfs_resume_fs(zfsvfs_t
*zfsvfs
, const char *osname
)
2011 uint64_t sa_obj
= 0;
2013 ASSERT(RRM_WRITE_HELD(&zfsvfs
->z_teardown_lock
));
2014 ASSERT(RW_WRITE_HELD(&zfsvfs
->z_teardown_inactive_lock
));
2017 * We already own this, so just hold and rele it to update the
2018 * objset_t, as the one we had before may have been evicted.
2020 VERIFY0(dmu_objset_hold(osname
, zfsvfs
, &zfsvfs
->z_os
));
2021 VERIFY3P(zfsvfs
->z_os
->os_dsl_dataset
->ds_owner
, ==, zfsvfs
);
2022 VERIFY(dsl_dataset_long_held(zfsvfs
->z_os
->os_dsl_dataset
));
2023 dmu_objset_rele(zfsvfs
->z_os
, zfsvfs
);
2026 * Make sure version hasn't changed
2029 err
= zfs_get_zplprop(zfsvfs
->z_os
, ZFS_PROP_VERSION
,
2030 &zfsvfs
->z_version
);
2035 err
= zap_lookup(zfsvfs
->z_os
, MASTER_NODE_OBJ
,
2036 ZFS_SA_ATTRS
, 8, 1, &sa_obj
);
2038 if (err
&& zfsvfs
->z_version
>= ZPL_VERSION_SA
)
2041 if ((err
= sa_setup(zfsvfs
->z_os
, sa_obj
,
2042 zfs_attr_table
, ZPL_END
, &zfsvfs
->z_attr_table
)) != 0)
2045 if (zfsvfs
->z_version
>= ZPL_VERSION_SA
)
2046 sa_register_update_callback(zfsvfs
->z_os
,
2049 VERIFY(zfsvfs_setup(zfsvfs
, B_FALSE
) == 0);
2051 zfs_set_fuid_feature(zfsvfs
);
2054 * Attempt to re-establish all the active znodes with
2055 * their dbufs. If a zfs_rezget() fails, then we'll let
2056 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2057 * when they try to use their znode.
2059 mutex_enter(&zfsvfs
->z_znodes_lock
);
2060 for (zp
= list_head(&zfsvfs
->z_all_znodes
); zp
;
2061 zp
= list_next(&zfsvfs
->z_all_znodes
, zp
)) {
2062 (void) zfs_rezget(zp
);
2064 mutex_exit(&zfsvfs
->z_znodes_lock
);
2067 /* release the VOPs */
2068 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
2069 rrm_exit(&zfsvfs
->z_teardown_lock
, FTAG
);
2073 * Since we couldn't setup the sa framework, try to force
2074 * unmount this file system.
2076 if (vn_vfswlock(zfsvfs
->z_vfs
->vfs_vnodecovered
) == 0)
2077 (void) dounmount(zfsvfs
->z_vfs
, MS_FORCE
, CRED());
2083 zfs_freevfs(vfs_t
*vfsp
)
2085 zfsvfs_t
*zfsvfs
= vfsp
->vfs_data
;
2088 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2089 * from zfs_mount(). Release it here. If we came through
2090 * zfs_mountroot() instead, we didn't grab an extra hold, so
2091 * skip the VFS_RELE for rootvfs.
2093 if (zfsvfs
->z_issnap
&& (vfsp
!= rootvfs
))
2094 VFS_RELE(zfsvfs
->z_parent
->z_vfs
);
2096 zfsvfs_free(zfsvfs
);
2098 atomic_dec_32(&zfs_active_fs_count
);
2102 * VFS_INIT() initialization. Note that there is no VFS_FINI(),
2103 * so we can't safely do any non-idempotent initialization here.
2104 * Leave that to zfs_init() and zfs_fini(), which are called
2105 * from the module's _init() and _fini() entry points.
2109 zfs_vfsinit(int fstype
, char *name
)
2116 * Setup vfsops and vnodeops tables.
2118 error
= vfs_setfsops(fstype
, zfs_vfsops_template
, &zfs_vfsops
);
2120 cmn_err(CE_WARN
, "zfs: bad vfs ops template");
2123 error
= zfs_create_op_tables();
2125 zfs_remove_op_tables();
2126 cmn_err(CE_WARN
, "zfs: bad vnode ops template");
2127 (void) vfs_freevfsops_by_type(zfsfstype
);
2131 mutex_init(&zfs_dev_mtx
, NULL
, MUTEX_DEFAULT
, NULL
);
2134 * Unique major number for all zfs mounts.
2135 * If we run out of 32-bit minors, we'll getudev() another major.
2137 zfs_major
= ddi_name_to_major(ZFS_DRIVER
);
2138 zfs_minor
= ZFS_MIN_MINOR
;
2147 * Initialize .zfs directory structures
2152 * Initialize znode cache, vnode ops, etc...
2156 dmu_objset_register_type(DMU_OST_ZFS
, zfs_space_delta_cb
);
2169 return (zfs_active_fs_count
!= 0);
2173 zfs_set_version(zfsvfs_t
*zfsvfs
, uint64_t newvers
)
2176 objset_t
*os
= zfsvfs
->z_os
;
2179 if (newvers
< ZPL_VERSION_INITIAL
|| newvers
> ZPL_VERSION
)
2180 return (SET_ERROR(EINVAL
));
2182 if (newvers
< zfsvfs
->z_version
)
2183 return (SET_ERROR(EINVAL
));
2185 if (zfs_spa_version_map(newvers
) >
2186 spa_version(dmu_objset_spa(zfsvfs
->z_os
)))
2187 return (SET_ERROR(ENOTSUP
));
2189 tx
= dmu_tx_create(os
);
2190 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_FALSE
, ZPL_VERSION_STR
);
2191 if (newvers
>= ZPL_VERSION_SA
&& !zfsvfs
->z_use_sa
) {
2192 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
2194 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
2196 error
= dmu_tx_assign(tx
, TXG_WAIT
);
2202 error
= zap_update(os
, MASTER_NODE_OBJ
, ZPL_VERSION_STR
,
2203 8, 1, &newvers
, tx
);
2210 if (newvers
>= ZPL_VERSION_SA
&& !zfsvfs
->z_use_sa
) {
2213 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs
->z_os
)), >=,
2215 sa_obj
= zap_create(os
, DMU_OT_SA_MASTER_NODE
,
2216 DMU_OT_NONE
, 0, tx
);
2218 error
= zap_add(os
, MASTER_NODE_OBJ
,
2219 ZFS_SA_ATTRS
, 8, 1, &sa_obj
, tx
);
2222 VERIFY(0 == sa_set_sa_object(os
, sa_obj
));
2223 sa_register_update_callback(os
, zfs_sa_upgrade
);
2226 spa_history_log_internal_ds(dmu_objset_ds(os
), "upgrade", tx
,
2227 "from %llu to %llu", zfsvfs
->z_version
, newvers
);
2231 zfsvfs
->z_version
= newvers
;
2233 zfs_set_fuid_feature(zfsvfs
);
2239 * Read a property stored within the master node.
2242 zfs_get_zplprop(objset_t
*os
, zfs_prop_t prop
, uint64_t *value
)
2248 * Look up the file system's value for the property. For the
2249 * version property, we look up a slightly different string.
2251 if (prop
== ZFS_PROP_VERSION
)
2252 pname
= ZPL_VERSION_STR
;
2254 pname
= zfs_prop_to_name(prop
);
2257 error
= zap_lookup(os
, MASTER_NODE_OBJ
, pname
, 8, 1, value
);
2259 if (error
== ENOENT
) {
2260 /* No value set, use the default value */
2262 case ZFS_PROP_VERSION
:
2263 *value
= ZPL_VERSION
;
2265 case ZFS_PROP_NORMALIZE
:
2266 case ZFS_PROP_UTF8ONLY
:
2270 *value
= ZFS_CASE_SENSITIVE
;
2280 static vfsdef_t vfw
= {
2284 VSW_HASPROTO
|VSW_CANRWRO
|VSW_CANREMOUNT
|VSW_VOLATILEDEV
|VSW_STATS
|
2289 struct modlfs zfs_modlfs
= {
2290 &mod_fsops
, "ZFS filesystem version " SPA_VERSION_STRING
, &vfw