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, 2018 by Delphix. All rights reserved.
26 /* Portions Copyright 2010 Robert Milkowski */
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/sysmacros.h>
32 #include <sys/pathname.h>
33 #include <sys/vnode.h>
35 #include <sys/mntent.h>
36 #include <sys/cmn_err.h>
37 #include <sys/zfs_znode.h>
38 #include <sys/zfs_vnops.h>
39 #include <sys/zfs_dir.h>
41 #include <sys/fs/zfs.h>
43 #include <sys/dsl_prop.h>
44 #include <sys/dsl_dataset.h>
45 #include <sys/dsl_deleg.h>
49 #include <sys/sa_impl.h>
50 #include <sys/policy.h>
51 #include <sys/atomic.h>
52 #include <sys/zfs_ioctl.h>
53 #include <sys/zfs_ctldir.h>
54 #include <sys/zfs_fuid.h>
55 #include <sys/sunddi.h>
56 #include <sys/dmu_objset.h>
57 #include <sys/spa_boot.h>
59 #include <linux/vfs_compat.h>
60 #include "zfs_comutil.h"
85 static const match_table_t zpl_tokens
= {
86 { TOKEN_RO
, MNTOPT_RO
},
87 { TOKEN_RW
, MNTOPT_RW
},
88 { TOKEN_SETUID
, MNTOPT_SETUID
},
89 { TOKEN_NOSETUID
, MNTOPT_NOSETUID
},
90 { TOKEN_EXEC
, MNTOPT_EXEC
},
91 { TOKEN_NOEXEC
, MNTOPT_NOEXEC
},
92 { TOKEN_DEVICES
, MNTOPT_DEVICES
},
93 { TOKEN_NODEVICES
, MNTOPT_NODEVICES
},
94 { TOKEN_DIRXATTR
, MNTOPT_DIRXATTR
},
95 { TOKEN_SAXATTR
, MNTOPT_SAXATTR
},
96 { TOKEN_XATTR
, MNTOPT_XATTR
},
97 { TOKEN_NOXATTR
, MNTOPT_NOXATTR
},
98 { TOKEN_ATIME
, MNTOPT_ATIME
},
99 { TOKEN_NOATIME
, MNTOPT_NOATIME
},
100 { TOKEN_RELATIME
, MNTOPT_RELATIME
},
101 { TOKEN_NORELATIME
, MNTOPT_NORELATIME
},
102 { TOKEN_NBMAND
, MNTOPT_NBMAND
},
103 { TOKEN_NONBMAND
, MNTOPT_NONBMAND
},
104 { TOKEN_MNTPOINT
, MNTOPT_MNTPOINT
"=%s" },
105 { TOKEN_LAST
, NULL
},
109 zfsvfs_vfs_free(vfs_t
*vfsp
)
112 if (vfsp
->vfs_mntpoint
!= NULL
)
113 strfree(vfsp
->vfs_mntpoint
);
115 kmem_free(vfsp
, sizeof (vfs_t
));
120 zfsvfs_parse_option(char *option
, int token
, substring_t
*args
, vfs_t
*vfsp
)
124 vfsp
->vfs_readonly
= B_TRUE
;
125 vfsp
->vfs_do_readonly
= B_TRUE
;
128 vfsp
->vfs_readonly
= B_FALSE
;
129 vfsp
->vfs_do_readonly
= B_TRUE
;
132 vfsp
->vfs_setuid
= B_TRUE
;
133 vfsp
->vfs_do_setuid
= B_TRUE
;
136 vfsp
->vfs_setuid
= B_FALSE
;
137 vfsp
->vfs_do_setuid
= B_TRUE
;
140 vfsp
->vfs_exec
= B_TRUE
;
141 vfsp
->vfs_do_exec
= B_TRUE
;
144 vfsp
->vfs_exec
= B_FALSE
;
145 vfsp
->vfs_do_exec
= B_TRUE
;
148 vfsp
->vfs_devices
= B_TRUE
;
149 vfsp
->vfs_do_devices
= B_TRUE
;
151 case TOKEN_NODEVICES
:
152 vfsp
->vfs_devices
= B_FALSE
;
153 vfsp
->vfs_do_devices
= B_TRUE
;
156 vfsp
->vfs_xattr
= ZFS_XATTR_DIR
;
157 vfsp
->vfs_do_xattr
= B_TRUE
;
160 vfsp
->vfs_xattr
= ZFS_XATTR_SA
;
161 vfsp
->vfs_do_xattr
= B_TRUE
;
164 vfsp
->vfs_xattr
= ZFS_XATTR_DIR
;
165 vfsp
->vfs_do_xattr
= B_TRUE
;
168 vfsp
->vfs_xattr
= ZFS_XATTR_OFF
;
169 vfsp
->vfs_do_xattr
= B_TRUE
;
172 vfsp
->vfs_atime
= B_TRUE
;
173 vfsp
->vfs_do_atime
= B_TRUE
;
176 vfsp
->vfs_atime
= B_FALSE
;
177 vfsp
->vfs_do_atime
= B_TRUE
;
180 vfsp
->vfs_relatime
= B_TRUE
;
181 vfsp
->vfs_do_relatime
= B_TRUE
;
183 case TOKEN_NORELATIME
:
184 vfsp
->vfs_relatime
= B_FALSE
;
185 vfsp
->vfs_do_relatime
= B_TRUE
;
188 vfsp
->vfs_nbmand
= B_TRUE
;
189 vfsp
->vfs_do_nbmand
= B_TRUE
;
192 vfsp
->vfs_nbmand
= B_FALSE
;
193 vfsp
->vfs_do_nbmand
= B_TRUE
;
196 vfsp
->vfs_mntpoint
= match_strdup(&args
[0]);
197 if (vfsp
->vfs_mntpoint
== NULL
)
198 return (SET_ERROR(ENOMEM
));
209 * Parse the raw mntopts and return a vfs_t describing the options.
212 zfsvfs_parse_options(char *mntopts
, vfs_t
**vfsp
)
217 tmp_vfsp
= kmem_zalloc(sizeof (vfs_t
), KM_SLEEP
);
219 if (mntopts
!= NULL
) {
220 substring_t args
[MAX_OPT_ARGS
];
221 char *tmp_mntopts
, *p
, *t
;
224 tmp_mntopts
= t
= strdup(mntopts
);
225 if (tmp_mntopts
== NULL
)
226 return (SET_ERROR(ENOMEM
));
228 while ((p
= strsep(&t
, ",")) != NULL
) {
232 args
[0].to
= args
[0].from
= NULL
;
233 token
= match_token(p
, zpl_tokens
, args
);
234 error
= zfsvfs_parse_option(p
, token
, args
, tmp_vfsp
);
236 strfree(tmp_mntopts
);
237 zfsvfs_vfs_free(tmp_vfsp
);
242 strfree(tmp_mntopts
);
251 zfs_is_readonly(zfsvfs_t
*zfsvfs
)
253 return (!!(zfsvfs
->z_sb
->s_flags
& SB_RDONLY
));
258 zfs_sync(struct super_block
*sb
, int wait
, cred_t
*cr
)
260 zfsvfs_t
*zfsvfs
= sb
->s_fs_info
;
263 * Semantically, the only requirement is that the sync be initiated.
264 * The DMU syncs out txgs frequently, so there's nothing to do.
269 if (zfsvfs
!= NULL
) {
271 * Sync a specific filesystem.
276 dp
= dmu_objset_pool(zfsvfs
->z_os
);
279 * If the system is shutting down, then skip any
280 * filesystems which may exist on a suspended pool.
282 if (spa_suspended(dp
->dp_spa
)) {
287 if (zfsvfs
->z_log
!= NULL
)
288 zil_commit(zfsvfs
->z_log
, 0);
293 * Sync all ZFS filesystems. This is what happens when you
294 * run sync(1M). Unlike other filesystems, ZFS honors the
295 * request by waiting for all pools to commit all dirty data.
304 atime_changed_cb(void *arg
, uint64_t newval
)
306 zfsvfs_t
*zfsvfs
= arg
;
307 struct super_block
*sb
= zfsvfs
->z_sb
;
312 * Update SB_NOATIME bit in VFS super block. Since atime update is
313 * determined by atime_needs_update(), atime_needs_update() needs to
314 * return false if atime is turned off, and not unconditionally return
315 * false if atime is turned on.
318 sb
->s_flags
&= ~SB_NOATIME
;
320 sb
->s_flags
|= SB_NOATIME
;
324 relatime_changed_cb(void *arg
, uint64_t newval
)
326 ((zfsvfs_t
*)arg
)->z_relatime
= newval
;
330 xattr_changed_cb(void *arg
, uint64_t newval
)
332 zfsvfs_t
*zfsvfs
= arg
;
334 if (newval
== ZFS_XATTR_OFF
) {
335 zfsvfs
->z_flags
&= ~ZSB_XATTR
;
337 zfsvfs
->z_flags
|= ZSB_XATTR
;
339 if (newval
== ZFS_XATTR_SA
)
340 zfsvfs
->z_xattr_sa
= B_TRUE
;
342 zfsvfs
->z_xattr_sa
= B_FALSE
;
347 acltype_changed_cb(void *arg
, uint64_t newval
)
349 zfsvfs_t
*zfsvfs
= arg
;
352 case ZFS_ACLTYPE_OFF
:
353 zfsvfs
->z_acl_type
= ZFS_ACLTYPE_OFF
;
354 zfsvfs
->z_sb
->s_flags
&= ~SB_POSIXACL
;
356 case ZFS_ACLTYPE_POSIXACL
:
357 #ifdef CONFIG_FS_POSIX_ACL
358 zfsvfs
->z_acl_type
= ZFS_ACLTYPE_POSIXACL
;
359 zfsvfs
->z_sb
->s_flags
|= SB_POSIXACL
;
361 zfsvfs
->z_acl_type
= ZFS_ACLTYPE_OFF
;
362 zfsvfs
->z_sb
->s_flags
&= ~SB_POSIXACL
;
363 #endif /* CONFIG_FS_POSIX_ACL */
371 blksz_changed_cb(void *arg
, uint64_t newval
)
373 zfsvfs_t
*zfsvfs
= arg
;
374 ASSERT3U(newval
, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs
->z_os
)));
375 ASSERT3U(newval
, >=, SPA_MINBLOCKSIZE
);
376 ASSERT(ISP2(newval
));
378 zfsvfs
->z_max_blksz
= newval
;
382 readonly_changed_cb(void *arg
, uint64_t newval
)
384 zfsvfs_t
*zfsvfs
= arg
;
385 struct super_block
*sb
= zfsvfs
->z_sb
;
391 sb
->s_flags
|= SB_RDONLY
;
393 sb
->s_flags
&= ~SB_RDONLY
;
397 devices_changed_cb(void *arg
, uint64_t newval
)
402 setuid_changed_cb(void *arg
, uint64_t newval
)
407 exec_changed_cb(void *arg
, uint64_t newval
)
412 nbmand_changed_cb(void *arg
, uint64_t newval
)
414 zfsvfs_t
*zfsvfs
= arg
;
415 struct super_block
*sb
= zfsvfs
->z_sb
;
421 sb
->s_flags
|= SB_MANDLOCK
;
423 sb
->s_flags
&= ~SB_MANDLOCK
;
427 snapdir_changed_cb(void *arg
, uint64_t newval
)
429 ((zfsvfs_t
*)arg
)->z_show_ctldir
= newval
;
433 vscan_changed_cb(void *arg
, uint64_t newval
)
435 ((zfsvfs_t
*)arg
)->z_vscan
= newval
;
439 acl_inherit_changed_cb(void *arg
, uint64_t newval
)
441 ((zfsvfs_t
*)arg
)->z_acl_inherit
= newval
;
445 zfs_register_callbacks(vfs_t
*vfsp
)
447 struct dsl_dataset
*ds
= NULL
;
449 zfsvfs_t
*zfsvfs
= NULL
;
453 zfsvfs
= vfsp
->vfs_data
;
458 * The act of registering our callbacks will destroy any mount
459 * options we may have. In order to enable temporary overrides
460 * of mount options, we stash away the current values and
461 * restore them after we register the callbacks.
463 if (zfs_is_readonly(zfsvfs
) || !spa_writeable(dmu_objset_spa(os
))) {
464 vfsp
->vfs_do_readonly
= B_TRUE
;
465 vfsp
->vfs_readonly
= B_TRUE
;
469 * Register property callbacks.
471 * It would probably be fine to just check for i/o error from
472 * the first prop_register(), but I guess I like to go
475 ds
= dmu_objset_ds(os
);
476 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
477 error
= dsl_prop_register(ds
,
478 zfs_prop_to_name(ZFS_PROP_ATIME
), atime_changed_cb
, zfsvfs
);
479 error
= error
? error
: dsl_prop_register(ds
,
480 zfs_prop_to_name(ZFS_PROP_RELATIME
), relatime_changed_cb
, zfsvfs
);
481 error
= error
? error
: dsl_prop_register(ds
,
482 zfs_prop_to_name(ZFS_PROP_XATTR
), xattr_changed_cb
, zfsvfs
);
483 error
= error
? error
: dsl_prop_register(ds
,
484 zfs_prop_to_name(ZFS_PROP_RECORDSIZE
), blksz_changed_cb
, zfsvfs
);
485 error
= error
? error
: dsl_prop_register(ds
,
486 zfs_prop_to_name(ZFS_PROP_READONLY
), readonly_changed_cb
, zfsvfs
);
487 error
= error
? error
: dsl_prop_register(ds
,
488 zfs_prop_to_name(ZFS_PROP_DEVICES
), devices_changed_cb
, zfsvfs
);
489 error
= error
? error
: dsl_prop_register(ds
,
490 zfs_prop_to_name(ZFS_PROP_SETUID
), setuid_changed_cb
, zfsvfs
);
491 error
= error
? error
: dsl_prop_register(ds
,
492 zfs_prop_to_name(ZFS_PROP_EXEC
), exec_changed_cb
, zfsvfs
);
493 error
= error
? error
: dsl_prop_register(ds
,
494 zfs_prop_to_name(ZFS_PROP_SNAPDIR
), snapdir_changed_cb
, zfsvfs
);
495 error
= error
? error
: dsl_prop_register(ds
,
496 zfs_prop_to_name(ZFS_PROP_ACLTYPE
), acltype_changed_cb
, zfsvfs
);
497 error
= error
? error
: dsl_prop_register(ds
,
498 zfs_prop_to_name(ZFS_PROP_ACLINHERIT
), acl_inherit_changed_cb
,
500 error
= error
? error
: dsl_prop_register(ds
,
501 zfs_prop_to_name(ZFS_PROP_VSCAN
), vscan_changed_cb
, zfsvfs
);
502 error
= error
? error
: dsl_prop_register(ds
,
503 zfs_prop_to_name(ZFS_PROP_NBMAND
), nbmand_changed_cb
, zfsvfs
);
504 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
509 * Invoke our callbacks to restore temporary mount options.
511 if (vfsp
->vfs_do_readonly
)
512 readonly_changed_cb(zfsvfs
, vfsp
->vfs_readonly
);
513 if (vfsp
->vfs_do_setuid
)
514 setuid_changed_cb(zfsvfs
, vfsp
->vfs_setuid
);
515 if (vfsp
->vfs_do_exec
)
516 exec_changed_cb(zfsvfs
, vfsp
->vfs_exec
);
517 if (vfsp
->vfs_do_devices
)
518 devices_changed_cb(zfsvfs
, vfsp
->vfs_devices
);
519 if (vfsp
->vfs_do_xattr
)
520 xattr_changed_cb(zfsvfs
, vfsp
->vfs_xattr
);
521 if (vfsp
->vfs_do_atime
)
522 atime_changed_cb(zfsvfs
, vfsp
->vfs_atime
);
523 if (vfsp
->vfs_do_relatime
)
524 relatime_changed_cb(zfsvfs
, vfsp
->vfs_relatime
);
525 if (vfsp
->vfs_do_nbmand
)
526 nbmand_changed_cb(zfsvfs
, vfsp
->vfs_nbmand
);
531 dsl_prop_unregister_all(ds
, zfsvfs
);
536 zfs_space_delta_cb(dmu_object_type_t bonustype
, void *data
,
537 uint64_t *userp
, uint64_t *groupp
, uint64_t *projectp
)
540 sa_hdr_phys_t
*sap
= data
;
543 boolean_t swap
= B_FALSE
;
546 * Is it a valid type of object to track?
548 if (bonustype
!= DMU_OT_ZNODE
&& bonustype
!= DMU_OT_SA
)
549 return (SET_ERROR(ENOENT
));
552 * If we have a NULL data pointer
553 * then assume the id's aren't changing and
554 * return EEXIST to the dmu to let it know to
558 return (SET_ERROR(EEXIST
));
560 if (bonustype
== DMU_OT_ZNODE
) {
561 znode_phys_t
*znp
= data
;
562 *userp
= znp
->zp_uid
;
563 *groupp
= znp
->zp_gid
;
564 *projectp
= ZFS_DEFAULT_PROJID
;
568 if (sap
->sa_magic
== 0) {
570 * This should only happen for newly created files
571 * that haven't had the znode data filled in yet.
575 *projectp
= ZFS_DEFAULT_PROJID
;
580 if (sa
.sa_magic
== BSWAP_32(SA_MAGIC
)) {
581 sa
.sa_magic
= SA_MAGIC
;
582 sa
.sa_layout_info
= BSWAP_16(sa
.sa_layout_info
);
585 VERIFY3U(sa
.sa_magic
, ==, SA_MAGIC
);
588 hdrsize
= sa_hdrsize(&sa
);
589 VERIFY3U(hdrsize
, >=, sizeof (sa_hdr_phys_t
));
591 *userp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+ SA_UID_OFFSET
));
592 *groupp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+ SA_GID_OFFSET
));
593 flags
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+ SA_FLAGS_OFFSET
));
595 flags
= BSWAP_64(flags
);
597 if (flags
& ZFS_PROJID
)
598 *projectp
= *((uint64_t *)((uintptr_t)data
+ hdrsize
+
601 *projectp
= ZFS_DEFAULT_PROJID
;
604 *userp
= BSWAP_64(*userp
);
605 *groupp
= BSWAP_64(*groupp
);
606 *projectp
= BSWAP_64(*projectp
);
612 fuidstr_to_sid(zfsvfs_t
*zfsvfs
, const char *fuidstr
,
613 char *domainbuf
, int buflen
, uid_t
*ridp
)
618 fuid
= zfs_strtonum(fuidstr
, NULL
);
620 domain
= zfs_fuid_find_by_idx(zfsvfs
, FUID_INDEX(fuid
));
622 (void) strlcpy(domainbuf
, domain
, buflen
);
625 *ridp
= FUID_RID(fuid
);
629 zfs_userquota_prop_to_obj(zfsvfs_t
*zfsvfs
, zfs_userquota_prop_t type
)
632 case ZFS_PROP_USERUSED
:
633 case ZFS_PROP_USEROBJUSED
:
634 return (DMU_USERUSED_OBJECT
);
635 case ZFS_PROP_GROUPUSED
:
636 case ZFS_PROP_GROUPOBJUSED
:
637 return (DMU_GROUPUSED_OBJECT
);
638 case ZFS_PROP_PROJECTUSED
:
639 case ZFS_PROP_PROJECTOBJUSED
:
640 return (DMU_PROJECTUSED_OBJECT
);
641 case ZFS_PROP_USERQUOTA
:
642 return (zfsvfs
->z_userquota_obj
);
643 case ZFS_PROP_GROUPQUOTA
:
644 return (zfsvfs
->z_groupquota_obj
);
645 case ZFS_PROP_USEROBJQUOTA
:
646 return (zfsvfs
->z_userobjquota_obj
);
647 case ZFS_PROP_GROUPOBJQUOTA
:
648 return (zfsvfs
->z_groupobjquota_obj
);
649 case ZFS_PROP_PROJECTQUOTA
:
650 return (zfsvfs
->z_projectquota_obj
);
651 case ZFS_PROP_PROJECTOBJQUOTA
:
652 return (zfsvfs
->z_projectobjquota_obj
);
654 return (ZFS_NO_OBJECT
);
659 zfs_userspace_many(zfsvfs_t
*zfsvfs
, zfs_userquota_prop_t type
,
660 uint64_t *cookiep
, void *vbuf
, uint64_t *bufsizep
)
665 zfs_useracct_t
*buf
= vbuf
;
669 if (!dmu_objset_userspace_present(zfsvfs
->z_os
))
670 return (SET_ERROR(ENOTSUP
));
672 if ((type
== ZFS_PROP_PROJECTQUOTA
|| type
== ZFS_PROP_PROJECTUSED
||
673 type
== ZFS_PROP_PROJECTOBJQUOTA
||
674 type
== ZFS_PROP_PROJECTOBJUSED
) &&
675 !dmu_objset_projectquota_present(zfsvfs
->z_os
))
676 return (SET_ERROR(ENOTSUP
));
678 if ((type
== ZFS_PROP_USEROBJUSED
|| type
== ZFS_PROP_GROUPOBJUSED
||
679 type
== ZFS_PROP_USEROBJQUOTA
|| type
== ZFS_PROP_GROUPOBJQUOTA
||
680 type
== ZFS_PROP_PROJECTOBJUSED
||
681 type
== ZFS_PROP_PROJECTOBJQUOTA
) &&
682 !dmu_objset_userobjspace_present(zfsvfs
->z_os
))
683 return (SET_ERROR(ENOTSUP
));
685 obj
= zfs_userquota_prop_to_obj(zfsvfs
, type
);
686 if (obj
== ZFS_NO_OBJECT
) {
691 if (type
== ZFS_PROP_USEROBJUSED
|| type
== ZFS_PROP_GROUPOBJUSED
||
692 type
== ZFS_PROP_PROJECTOBJUSED
)
693 offset
= DMU_OBJACCT_PREFIX_LEN
;
695 for (zap_cursor_init_serialized(&zc
, zfsvfs
->z_os
, obj
, *cookiep
);
696 (error
= zap_cursor_retrieve(&zc
, &za
)) == 0;
697 zap_cursor_advance(&zc
)) {
698 if ((uintptr_t)buf
- (uintptr_t)vbuf
+ sizeof (zfs_useracct_t
) >
703 * skip object quota (with zap name prefix DMU_OBJACCT_PREFIX)
704 * when dealing with block quota and vice versa.
706 if ((offset
> 0) != (strncmp(za
.za_name
, DMU_OBJACCT_PREFIX
,
707 DMU_OBJACCT_PREFIX_LEN
) == 0))
710 fuidstr_to_sid(zfsvfs
, za
.za_name
+ offset
,
711 buf
->zu_domain
, sizeof (buf
->zu_domain
), &buf
->zu_rid
);
713 buf
->zu_space
= za
.za_first_integer
;
719 ASSERT3U((uintptr_t)buf
- (uintptr_t)vbuf
, <=, *bufsizep
);
720 *bufsizep
= (uintptr_t)buf
- (uintptr_t)vbuf
;
721 *cookiep
= zap_cursor_serialize(&zc
);
722 zap_cursor_fini(&zc
);
727 * buf must be big enough (eg, 32 bytes)
730 id_to_fuidstr(zfsvfs_t
*zfsvfs
, const char *domain
, uid_t rid
,
731 char *buf
, boolean_t addok
)
736 if (domain
&& domain
[0]) {
737 domainid
= zfs_fuid_find_by_domain(zfsvfs
, domain
, NULL
, addok
);
739 return (SET_ERROR(ENOENT
));
741 fuid
= FUID_ENCODE(domainid
, rid
);
742 (void) sprintf(buf
, "%llx", (longlong_t
)fuid
);
747 zfs_userspace_one(zfsvfs_t
*zfsvfs
, zfs_userquota_prop_t type
,
748 const char *domain
, uint64_t rid
, uint64_t *valp
)
750 char buf
[20 + DMU_OBJACCT_PREFIX_LEN
];
757 if (!dmu_objset_userspace_present(zfsvfs
->z_os
))
758 return (SET_ERROR(ENOTSUP
));
760 if ((type
== ZFS_PROP_USEROBJUSED
|| type
== ZFS_PROP_GROUPOBJUSED
||
761 type
== ZFS_PROP_USEROBJQUOTA
|| type
== ZFS_PROP_GROUPOBJQUOTA
||
762 type
== ZFS_PROP_PROJECTOBJUSED
||
763 type
== ZFS_PROP_PROJECTOBJQUOTA
) &&
764 !dmu_objset_userobjspace_present(zfsvfs
->z_os
))
765 return (SET_ERROR(ENOTSUP
));
767 if (type
== ZFS_PROP_PROJECTQUOTA
|| type
== ZFS_PROP_PROJECTUSED
||
768 type
== ZFS_PROP_PROJECTOBJQUOTA
||
769 type
== ZFS_PROP_PROJECTOBJUSED
) {
770 if (!dmu_objset_projectquota_present(zfsvfs
->z_os
))
771 return (SET_ERROR(ENOTSUP
));
772 if (!zpl_is_valid_projid(rid
))
773 return (SET_ERROR(EINVAL
));
776 obj
= zfs_userquota_prop_to_obj(zfsvfs
, type
);
777 if (obj
== ZFS_NO_OBJECT
)
780 if (type
== ZFS_PROP_USEROBJUSED
|| type
== ZFS_PROP_GROUPOBJUSED
||
781 type
== ZFS_PROP_PROJECTOBJUSED
) {
782 strlcpy(buf
, DMU_OBJACCT_PREFIX
, DMU_OBJACCT_PREFIX_LEN
+ 1);
783 offset
= DMU_OBJACCT_PREFIX_LEN
;
786 err
= id_to_fuidstr(zfsvfs
, domain
, rid
, buf
+ offset
, B_FALSE
);
790 err
= zap_lookup(zfsvfs
->z_os
, obj
, buf
, 8, 1, valp
);
797 zfs_set_userquota(zfsvfs_t
*zfsvfs
, zfs_userquota_prop_t type
,
798 const char *domain
, uint64_t rid
, uint64_t quota
)
804 boolean_t fuid_dirtied
;
806 if (zfsvfs
->z_version
< ZPL_VERSION_USERSPACE
)
807 return (SET_ERROR(ENOTSUP
));
810 case ZFS_PROP_USERQUOTA
:
811 objp
= &zfsvfs
->z_userquota_obj
;
813 case ZFS_PROP_GROUPQUOTA
:
814 objp
= &zfsvfs
->z_groupquota_obj
;
816 case ZFS_PROP_USEROBJQUOTA
:
817 objp
= &zfsvfs
->z_userobjquota_obj
;
819 case ZFS_PROP_GROUPOBJQUOTA
:
820 objp
= &zfsvfs
->z_groupobjquota_obj
;
822 case ZFS_PROP_PROJECTQUOTA
:
823 if (!dmu_objset_projectquota_enabled(zfsvfs
->z_os
))
824 return (SET_ERROR(ENOTSUP
));
825 if (!zpl_is_valid_projid(rid
))
826 return (SET_ERROR(EINVAL
));
828 objp
= &zfsvfs
->z_projectquota_obj
;
830 case ZFS_PROP_PROJECTOBJQUOTA
:
831 if (!dmu_objset_projectquota_enabled(zfsvfs
->z_os
))
832 return (SET_ERROR(ENOTSUP
));
833 if (!zpl_is_valid_projid(rid
))
834 return (SET_ERROR(EINVAL
));
836 objp
= &zfsvfs
->z_projectobjquota_obj
;
839 return (SET_ERROR(EINVAL
));
842 err
= id_to_fuidstr(zfsvfs
, domain
, rid
, buf
, B_TRUE
);
845 fuid_dirtied
= zfsvfs
->z_fuid_dirty
;
847 tx
= dmu_tx_create(zfsvfs
->z_os
);
848 dmu_tx_hold_zap(tx
, *objp
? *objp
: DMU_NEW_OBJECT
, B_TRUE
, NULL
);
850 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
851 zfs_userquota_prop_prefixes
[type
]);
854 zfs_fuid_txhold(zfsvfs
, tx
);
855 err
= dmu_tx_assign(tx
, TXG_WAIT
);
861 mutex_enter(&zfsvfs
->z_lock
);
863 *objp
= zap_create(zfsvfs
->z_os
, DMU_OT_USERGROUP_QUOTA
,
865 VERIFY(0 == zap_add(zfsvfs
->z_os
, MASTER_NODE_OBJ
,
866 zfs_userquota_prop_prefixes
[type
], 8, 1, objp
, tx
));
868 mutex_exit(&zfsvfs
->z_lock
);
871 err
= zap_remove(zfsvfs
->z_os
, *objp
, buf
, tx
);
875 err
= zap_update(zfsvfs
->z_os
, *objp
, buf
, 8, 1, "a
, tx
);
879 zfs_fuid_sync(zfsvfs
, tx
);
885 zfs_id_overobjquota(zfsvfs_t
*zfsvfs
, uint64_t usedobj
, uint64_t id
)
887 char buf
[20 + DMU_OBJACCT_PREFIX_LEN
];
888 uint64_t used
, quota
, quotaobj
;
891 if (!dmu_objset_userobjspace_present(zfsvfs
->z_os
)) {
892 if (dmu_objset_userobjspace_upgradable(zfsvfs
->z_os
)) {
893 dsl_pool_config_enter(
894 dmu_objset_pool(zfsvfs
->z_os
), FTAG
);
895 dmu_objset_id_quota_upgrade(zfsvfs
->z_os
);
896 dsl_pool_config_exit(
897 dmu_objset_pool(zfsvfs
->z_os
), FTAG
);
902 if (usedobj
== DMU_PROJECTUSED_OBJECT
) {
903 if (!dmu_objset_projectquota_present(zfsvfs
->z_os
)) {
904 if (dmu_objset_projectquota_upgradable(zfsvfs
->z_os
)) {
905 dsl_pool_config_enter(
906 dmu_objset_pool(zfsvfs
->z_os
), FTAG
);
907 dmu_objset_id_quota_upgrade(zfsvfs
->z_os
);
908 dsl_pool_config_exit(
909 dmu_objset_pool(zfsvfs
->z_os
), FTAG
);
913 quotaobj
= zfsvfs
->z_projectobjquota_obj
;
914 } else if (usedobj
== DMU_USERUSED_OBJECT
) {
915 quotaobj
= zfsvfs
->z_userobjquota_obj
;
916 } else if (usedobj
== DMU_GROUPUSED_OBJECT
) {
917 quotaobj
= zfsvfs
->z_groupobjquota_obj
;
921 if (quotaobj
== 0 || zfsvfs
->z_replay
)
924 (void) sprintf(buf
, "%llx", (longlong_t
)id
);
925 err
= zap_lookup(zfsvfs
->z_os
, quotaobj
, buf
, 8, 1, "a
);
929 (void) sprintf(buf
, DMU_OBJACCT_PREFIX
"%llx", (longlong_t
)id
);
930 err
= zap_lookup(zfsvfs
->z_os
, usedobj
, buf
, 8, 1, &used
);
933 return (used
>= quota
);
937 zfs_id_overblockquota(zfsvfs_t
*zfsvfs
, uint64_t usedobj
, uint64_t id
)
940 uint64_t used
, quota
, quotaobj
;
943 if (usedobj
== DMU_PROJECTUSED_OBJECT
) {
944 if (!dmu_objset_projectquota_present(zfsvfs
->z_os
)) {
945 if (dmu_objset_projectquota_upgradable(zfsvfs
->z_os
)) {
946 dsl_pool_config_enter(
947 dmu_objset_pool(zfsvfs
->z_os
), FTAG
);
948 dmu_objset_id_quota_upgrade(zfsvfs
->z_os
);
949 dsl_pool_config_exit(
950 dmu_objset_pool(zfsvfs
->z_os
), FTAG
);
954 quotaobj
= zfsvfs
->z_projectquota_obj
;
955 } else if (usedobj
== DMU_USERUSED_OBJECT
) {
956 quotaobj
= zfsvfs
->z_userquota_obj
;
957 } else if (usedobj
== DMU_GROUPUSED_OBJECT
) {
958 quotaobj
= zfsvfs
->z_groupquota_obj
;
962 if (quotaobj
== 0 || zfsvfs
->z_replay
)
965 (void) sprintf(buf
, "%llx", (longlong_t
)id
);
966 err
= zap_lookup(zfsvfs
->z_os
, quotaobj
, buf
, 8, 1, "a
);
970 err
= zap_lookup(zfsvfs
->z_os
, usedobj
, buf
, 8, 1, &used
);
973 return (used
>= quota
);
977 zfs_id_overquota(zfsvfs_t
*zfsvfs
, uint64_t usedobj
, uint64_t id
)
979 return (zfs_id_overblockquota(zfsvfs
, usedobj
, id
) ||
980 zfs_id_overobjquota(zfsvfs
, usedobj
, id
));
984 * Associate this zfsvfs with the given objset, which must be owned.
985 * This will cache a bunch of on-disk state from the objset in the
989 zfsvfs_init(zfsvfs_t
*zfsvfs
, objset_t
*os
)
994 zfsvfs
->z_max_blksz
= SPA_OLD_MAXBLOCKSIZE
;
995 zfsvfs
->z_show_ctldir
= ZFS_SNAPDIR_VISIBLE
;
998 error
= zfs_get_zplprop(os
, ZFS_PROP_VERSION
, &zfsvfs
->z_version
);
1001 if (zfsvfs
->z_version
>
1002 zfs_zpl_version_map(spa_version(dmu_objset_spa(os
)))) {
1003 (void) printk("Can't mount a version %lld file system "
1004 "on a version %lld pool\n. Pool must be upgraded to mount "
1005 "this file system.\n", (u_longlong_t
)zfsvfs
->z_version
,
1006 (u_longlong_t
)spa_version(dmu_objset_spa(os
)));
1007 return (SET_ERROR(ENOTSUP
));
1009 error
= zfs_get_zplprop(os
, ZFS_PROP_NORMALIZE
, &val
);
1012 zfsvfs
->z_norm
= (int)val
;
1014 error
= zfs_get_zplprop(os
, ZFS_PROP_UTF8ONLY
, &val
);
1017 zfsvfs
->z_utf8
= (val
!= 0);
1019 error
= zfs_get_zplprop(os
, ZFS_PROP_CASE
, &val
);
1022 zfsvfs
->z_case
= (uint_t
)val
;
1024 if ((error
= zfs_get_zplprop(os
, ZFS_PROP_ACLTYPE
, &val
)) != 0)
1026 zfsvfs
->z_acl_type
= (uint_t
)val
;
1029 * Fold case on file systems that are always or sometimes case
1032 if (zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
||
1033 zfsvfs
->z_case
== ZFS_CASE_MIXED
)
1034 zfsvfs
->z_norm
|= U8_TEXTPREP_TOUPPER
;
1036 zfsvfs
->z_use_fuids
= USE_FUIDS(zfsvfs
->z_version
, zfsvfs
->z_os
);
1037 zfsvfs
->z_use_sa
= USE_SA(zfsvfs
->z_version
, zfsvfs
->z_os
);
1039 uint64_t sa_obj
= 0;
1040 if (zfsvfs
->z_use_sa
) {
1041 /* should either have both of these objects or none */
1042 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SA_ATTRS
, 8, 1,
1047 error
= zfs_get_zplprop(os
, ZFS_PROP_XATTR
, &val
);
1048 if ((error
== 0) && (val
== ZFS_XATTR_SA
))
1049 zfsvfs
->z_xattr_sa
= B_TRUE
;
1052 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_ROOT_OBJ
, 8, 1,
1056 ASSERT(zfsvfs
->z_root
!= 0);
1058 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_UNLINKED_SET
, 8, 1,
1059 &zfsvfs
->z_unlinkedobj
);
1063 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
1064 zfs_userquota_prop_prefixes
[ZFS_PROP_USERQUOTA
],
1065 8, 1, &zfsvfs
->z_userquota_obj
);
1066 if (error
== ENOENT
)
1067 zfsvfs
->z_userquota_obj
= 0;
1068 else if (error
!= 0)
1071 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
1072 zfs_userquota_prop_prefixes
[ZFS_PROP_GROUPQUOTA
],
1073 8, 1, &zfsvfs
->z_groupquota_obj
);
1074 if (error
== ENOENT
)
1075 zfsvfs
->z_groupquota_obj
= 0;
1076 else if (error
!= 0)
1079 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
1080 zfs_userquota_prop_prefixes
[ZFS_PROP_PROJECTQUOTA
],
1081 8, 1, &zfsvfs
->z_projectquota_obj
);
1082 if (error
== ENOENT
)
1083 zfsvfs
->z_projectquota_obj
= 0;
1084 else if (error
!= 0)
1087 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
1088 zfs_userquota_prop_prefixes
[ZFS_PROP_USEROBJQUOTA
],
1089 8, 1, &zfsvfs
->z_userobjquota_obj
);
1090 if (error
== ENOENT
)
1091 zfsvfs
->z_userobjquota_obj
= 0;
1092 else if (error
!= 0)
1095 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
1096 zfs_userquota_prop_prefixes
[ZFS_PROP_GROUPOBJQUOTA
],
1097 8, 1, &zfsvfs
->z_groupobjquota_obj
);
1098 if (error
== ENOENT
)
1099 zfsvfs
->z_groupobjquota_obj
= 0;
1100 else if (error
!= 0)
1103 error
= zap_lookup(os
, MASTER_NODE_OBJ
,
1104 zfs_userquota_prop_prefixes
[ZFS_PROP_PROJECTOBJQUOTA
],
1105 8, 1, &zfsvfs
->z_projectobjquota_obj
);
1106 if (error
== ENOENT
)
1107 zfsvfs
->z_projectobjquota_obj
= 0;
1108 else if (error
!= 0)
1111 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_FUID_TABLES
, 8, 1,
1112 &zfsvfs
->z_fuid_obj
);
1113 if (error
== ENOENT
)
1114 zfsvfs
->z_fuid_obj
= 0;
1115 else if (error
!= 0)
1118 error
= zap_lookup(os
, MASTER_NODE_OBJ
, ZFS_SHARES_DIR
, 8, 1,
1119 &zfsvfs
->z_shares_dir
);
1120 if (error
== ENOENT
)
1121 zfsvfs
->z_shares_dir
= 0;
1122 else if (error
!= 0)
1125 error
= sa_setup(os
, sa_obj
, zfs_attr_table
, ZPL_END
,
1126 &zfsvfs
->z_attr_table
);
1130 if (zfsvfs
->z_version
>= ZPL_VERSION_SA
)
1131 sa_register_update_callback(os
, zfs_sa_upgrade
);
1137 zfsvfs_create(const char *osname
, boolean_t readonly
, zfsvfs_t
**zfvp
)
1142 boolean_t ro
= (readonly
|| (strchr(osname
, '@') != NULL
));
1144 zfsvfs
= kmem_zalloc(sizeof (zfsvfs_t
), KM_SLEEP
);
1146 error
= dmu_objset_own(osname
, DMU_OST_ZFS
, ro
, B_TRUE
, zfsvfs
, &os
);
1148 kmem_free(zfsvfs
, sizeof (zfsvfs_t
));
1152 error
= zfsvfs_create_impl(zfvp
, zfsvfs
, os
);
1154 dmu_objset_disown(os
, B_TRUE
, zfsvfs
);
1161 * Note: zfsvfs is assumed to be malloc'd, and will be freed by this function
1162 * on a failure. Do not pass in a statically allocated zfsvfs.
1165 zfsvfs_create_impl(zfsvfs_t
**zfvp
, zfsvfs_t
*zfsvfs
, objset_t
*os
)
1169 zfsvfs
->z_vfs
= NULL
;
1170 zfsvfs
->z_sb
= NULL
;
1171 zfsvfs
->z_parent
= zfsvfs
;
1173 mutex_init(&zfsvfs
->z_znodes_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1174 mutex_init(&zfsvfs
->z_lock
, NULL
, MUTEX_DEFAULT
, NULL
);
1175 list_create(&zfsvfs
->z_all_znodes
, sizeof (znode_t
),
1176 offsetof(znode_t
, z_link_node
));
1177 rrm_init(&zfsvfs
->z_teardown_lock
, B_FALSE
);
1178 rw_init(&zfsvfs
->z_teardown_inactive_lock
, NULL
, RW_DEFAULT
, NULL
);
1179 rw_init(&zfsvfs
->z_fuid_lock
, NULL
, RW_DEFAULT
, NULL
);
1181 int size
= MIN(1 << (highbit64(zfs_object_mutex_size
) - 1),
1183 zfsvfs
->z_hold_size
= size
;
1184 zfsvfs
->z_hold_trees
= vmem_zalloc(sizeof (avl_tree_t
) * size
,
1186 zfsvfs
->z_hold_locks
= vmem_zalloc(sizeof (kmutex_t
) * size
, KM_SLEEP
);
1187 for (int i
= 0; i
!= size
; i
++) {
1188 avl_create(&zfsvfs
->z_hold_trees
[i
], zfs_znode_hold_compare
,
1189 sizeof (znode_hold_t
), offsetof(znode_hold_t
, zh_node
));
1190 mutex_init(&zfsvfs
->z_hold_locks
[i
], NULL
, MUTEX_DEFAULT
, NULL
);
1193 error
= zfsvfs_init(zfsvfs
, os
);
1196 zfsvfs_free(zfsvfs
);
1200 zfsvfs
->z_drain_task
= TASKQID_INVALID
;
1201 zfsvfs
->z_draining
= B_FALSE
;
1202 zfsvfs
->z_drain_cancel
= B_TRUE
;
1209 zfsvfs_setup(zfsvfs_t
*zfsvfs
, boolean_t mounting
)
1212 boolean_t readonly
= zfs_is_readonly(zfsvfs
);
1214 error
= zfs_register_callbacks(zfsvfs
->z_vfs
);
1218 zfsvfs
->z_log
= zil_open(zfsvfs
->z_os
, zfs_get_data
);
1221 * If we are not mounting (ie: online recv), then we don't
1222 * have to worry about replaying the log as we blocked all
1223 * operations out since we closed the ZIL.
1226 ASSERT3P(zfsvfs
->z_kstat
.dk_kstats
, ==, NULL
);
1227 dataset_kstats_create(&zfsvfs
->z_kstat
, zfsvfs
->z_os
);
1230 * During replay we remove the read only flag to
1231 * allow replays to succeed.
1233 if (readonly
!= 0) {
1234 readonly_changed_cb(zfsvfs
, B_FALSE
);
1237 if (zap_get_stats(zfsvfs
->z_os
, zfsvfs
->z_unlinkedobj
,
1239 dataset_kstats_update_nunlinks_kstat(
1240 &zfsvfs
->z_kstat
, zs
.zs_num_entries
);
1242 dprintf_ds(zfsvfs
->z_os
->os_dsl_dataset
,
1243 "num_entries in unlinked set: %llu",
1245 zfs_unlinked_drain(zfsvfs
);
1249 * Parse and replay the intent log.
1251 * Because of ziltest, this must be done after
1252 * zfs_unlinked_drain(). (Further note: ziltest
1253 * doesn't use readonly mounts, where
1254 * zfs_unlinked_drain() isn't called.) This is because
1255 * ziltest causes spa_sync() to think it's committed,
1256 * but actually it is not, so the intent log contains
1257 * many txg's worth of changes.
1259 * In particular, if object N is in the unlinked set in
1260 * the last txg to actually sync, then it could be
1261 * actually freed in a later txg and then reallocated
1262 * in a yet later txg. This would write a "create
1263 * object N" record to the intent log. Normally, this
1264 * would be fine because the spa_sync() would have
1265 * written out the fact that object N is free, before
1266 * we could write the "create object N" intent log
1269 * But when we are in ziltest mode, we advance the "open
1270 * txg" without actually spa_sync()-ing the changes to
1271 * disk. So we would see that object N is still
1272 * allocated and in the unlinked set, and there is an
1273 * intent log record saying to allocate it.
1275 if (spa_writeable(dmu_objset_spa(zfsvfs
->z_os
))) {
1276 if (zil_replay_disable
) {
1277 zil_destroy(zfsvfs
->z_log
, B_FALSE
);
1279 zfsvfs
->z_replay
= B_TRUE
;
1280 zil_replay(zfsvfs
->z_os
, zfsvfs
,
1282 zfsvfs
->z_replay
= B_FALSE
;
1286 /* restore readonly bit */
1288 readonly_changed_cb(zfsvfs
, B_TRUE
);
1292 * Set the objset user_ptr to track its zfsvfs.
1294 mutex_enter(&zfsvfs
->z_os
->os_user_ptr_lock
);
1295 dmu_objset_set_user(zfsvfs
->z_os
, zfsvfs
);
1296 mutex_exit(&zfsvfs
->z_os
->os_user_ptr_lock
);
1302 zfsvfs_free(zfsvfs_t
*zfsvfs
)
1304 int i
, size
= zfsvfs
->z_hold_size
;
1306 zfs_fuid_destroy(zfsvfs
);
1308 mutex_destroy(&zfsvfs
->z_znodes_lock
);
1309 mutex_destroy(&zfsvfs
->z_lock
);
1310 list_destroy(&zfsvfs
->z_all_znodes
);
1311 rrm_destroy(&zfsvfs
->z_teardown_lock
);
1312 rw_destroy(&zfsvfs
->z_teardown_inactive_lock
);
1313 rw_destroy(&zfsvfs
->z_fuid_lock
);
1314 for (i
= 0; i
!= size
; i
++) {
1315 avl_destroy(&zfsvfs
->z_hold_trees
[i
]);
1316 mutex_destroy(&zfsvfs
->z_hold_locks
[i
]);
1318 vmem_free(zfsvfs
->z_hold_trees
, sizeof (avl_tree_t
) * size
);
1319 vmem_free(zfsvfs
->z_hold_locks
, sizeof (kmutex_t
) * size
);
1320 zfsvfs_vfs_free(zfsvfs
->z_vfs
);
1321 dataset_kstats_destroy(&zfsvfs
->z_kstat
);
1322 kmem_free(zfsvfs
, sizeof (zfsvfs_t
));
1326 zfs_set_fuid_feature(zfsvfs_t
*zfsvfs
)
1328 zfsvfs
->z_use_fuids
= USE_FUIDS(zfsvfs
->z_version
, zfsvfs
->z_os
);
1329 zfsvfs
->z_use_sa
= USE_SA(zfsvfs
->z_version
, zfsvfs
->z_os
);
1333 zfs_unregister_callbacks(zfsvfs_t
*zfsvfs
)
1335 objset_t
*os
= zfsvfs
->z_os
;
1337 if (!dmu_objset_is_snapshot(os
))
1338 dsl_prop_unregister_all(dmu_objset_ds(os
), zfsvfs
);
1341 #ifdef HAVE_MLSLABEL
1343 * Check that the hex label string is appropriate for the dataset being
1344 * mounted into the global_zone proper.
1346 * Return an error if the hex label string is not default or
1347 * admin_low/admin_high. For admin_low labels, the corresponding
1348 * dataset must be readonly.
1351 zfs_check_global_label(const char *dsname
, const char *hexsl
)
1353 if (strcasecmp(hexsl
, ZFS_MLSLABEL_DEFAULT
) == 0)
1355 if (strcasecmp(hexsl
, ADMIN_HIGH
) == 0)
1357 if (strcasecmp(hexsl
, ADMIN_LOW
) == 0) {
1358 /* must be readonly */
1361 if (dsl_prop_get_integer(dsname
,
1362 zfs_prop_to_name(ZFS_PROP_READONLY
), &rdonly
, NULL
))
1363 return (SET_ERROR(EACCES
));
1364 return (rdonly
? 0 : EACCES
);
1366 return (SET_ERROR(EACCES
));
1368 #endif /* HAVE_MLSLABEL */
1371 zfs_statfs_project(zfsvfs_t
*zfsvfs
, znode_t
*zp
, struct kstatfs
*statp
,
1374 char buf
[20 + DMU_OBJACCT_PREFIX_LEN
];
1375 uint64_t offset
= DMU_OBJACCT_PREFIX_LEN
;
1380 strlcpy(buf
, DMU_OBJACCT_PREFIX
, DMU_OBJACCT_PREFIX_LEN
+ 1);
1381 err
= id_to_fuidstr(zfsvfs
, NULL
, zp
->z_projid
, buf
+ offset
, B_FALSE
);
1385 if (zfsvfs
->z_projectquota_obj
== 0)
1388 err
= zap_lookup(zfsvfs
->z_os
, zfsvfs
->z_projectquota_obj
,
1389 buf
+ offset
, 8, 1, "a
);
1395 err
= zap_lookup(zfsvfs
->z_os
, DMU_PROJECTUSED_OBJECT
,
1396 buf
+ offset
, 8, 1, &used
);
1397 if (unlikely(err
== ENOENT
)) {
1399 u_longlong_t nblocks
;
1402 * Quota accounting is async, so it is possible race case.
1403 * There is at least one object with the given project ID.
1405 sa_object_size(zp
->z_sa_hdl
, &blksize
, &nblocks
);
1406 if (unlikely(zp
->z_blksz
== 0))
1407 blksize
= zfsvfs
->z_max_blksz
;
1409 used
= blksize
* nblocks
;
1414 statp
->f_blocks
= quota
>> bshift
;
1415 statp
->f_bfree
= (quota
> used
) ? ((quota
- used
) >> bshift
) : 0;
1416 statp
->f_bavail
= statp
->f_bfree
;
1419 if (zfsvfs
->z_projectobjquota_obj
== 0)
1422 err
= zap_lookup(zfsvfs
->z_os
, zfsvfs
->z_projectobjquota_obj
,
1423 buf
+ offset
, 8, 1, "a
);
1429 err
= zap_lookup(zfsvfs
->z_os
, DMU_PROJECTUSED_OBJECT
,
1431 if (unlikely(err
== ENOENT
)) {
1433 * Quota accounting is async, so it is possible race case.
1434 * There is at least one object with the given project ID.
1441 statp
->f_files
= quota
;
1442 statp
->f_ffree
= (quota
> used
) ? (quota
- used
) : 0;
1448 zfs_statvfs(struct dentry
*dentry
, struct kstatfs
*statp
)
1450 zfsvfs_t
*zfsvfs
= dentry
->d_sb
->s_fs_info
;
1451 uint64_t refdbytes
, availbytes
, usedobjs
, availobjs
;
1456 dmu_objset_space(zfsvfs
->z_os
,
1457 &refdbytes
, &availbytes
, &usedobjs
, &availobjs
);
1459 uint64_t fsid
= dmu_objset_fsid_guid(zfsvfs
->z_os
);
1461 * The underlying storage pool actually uses multiple block
1462 * size. Under Solaris frsize (fragment size) is reported as
1463 * the smallest block size we support, and bsize (block size)
1464 * as the filesystem's maximum block size. Unfortunately,
1465 * under Linux the fragment size and block size are often used
1466 * interchangeably. Thus we are forced to report both of them
1467 * as the filesystem's maximum block size.
1469 statp
->f_frsize
= zfsvfs
->z_max_blksz
;
1470 statp
->f_bsize
= zfsvfs
->z_max_blksz
;
1471 uint32_t bshift
= fls(statp
->f_bsize
) - 1;
1474 * The following report "total" blocks of various kinds in
1475 * the file system, but reported in terms of f_bsize - the
1479 /* Round up so we never have a filesytem using 0 blocks. */
1480 refdbytes
= P2ROUNDUP(refdbytes
, statp
->f_bsize
);
1481 statp
->f_blocks
= (refdbytes
+ availbytes
) >> bshift
;
1482 statp
->f_bfree
= availbytes
>> bshift
;
1483 statp
->f_bavail
= statp
->f_bfree
; /* no root reservation */
1486 * statvfs() should really be called statufs(), because it assumes
1487 * static metadata. ZFS doesn't preallocate files, so the best
1488 * we can do is report the max that could possibly fit in f_files,
1489 * and that minus the number actually used in f_ffree.
1490 * For f_ffree, report the smaller of the number of objects available
1491 * and the number of blocks (each object will take at least a block).
1493 statp
->f_ffree
= MIN(availobjs
, availbytes
>> DNODE_SHIFT
);
1494 statp
->f_files
= statp
->f_ffree
+ usedobjs
;
1495 statp
->f_fsid
.val
[0] = (uint32_t)fsid
;
1496 statp
->f_fsid
.val
[1] = (uint32_t)(fsid
>> 32);
1497 statp
->f_type
= ZFS_SUPER_MAGIC
;
1498 statp
->f_namelen
= MAXNAMELEN
- 1;
1501 * We have all of 40 characters to stuff a string here.
1502 * Is there anything useful we could/should provide?
1504 bzero(statp
->f_spare
, sizeof (statp
->f_spare
));
1506 if (dmu_objset_projectquota_enabled(zfsvfs
->z_os
) &&
1507 dmu_objset_projectquota_present(zfsvfs
->z_os
)) {
1508 znode_t
*zp
= ITOZ(dentry
->d_inode
);
1510 if (zp
->z_pflags
& ZFS_PROJINHERIT
&& zp
->z_projid
&&
1511 zpl_is_valid_projid(zp
->z_projid
))
1512 err
= zfs_statfs_project(zfsvfs
, zp
, statp
, bshift
);
1520 zfs_root(zfsvfs_t
*zfsvfs
, struct inode
**ipp
)
1527 error
= zfs_zget(zfsvfs
, zfsvfs
->z_root
, &rootzp
);
1529 *ipp
= ZTOI(rootzp
);
1535 #ifdef HAVE_D_PRUNE_ALIASES
1537 * Linux kernels older than 3.1 do not support a per-filesystem shrinker.
1538 * To accommodate this we must improvise and manually walk the list of znodes
1539 * attempting to prune dentries in order to be able to drop the inodes.
1541 * To avoid scanning the same znodes multiple times they are always rotated
1542 * to the end of the z_all_znodes list. New znodes are inserted at the
1543 * end of the list so we're always scanning the oldest znodes first.
1546 zfs_prune_aliases(zfsvfs_t
*zfsvfs
, unsigned long nr_to_scan
)
1548 znode_t
**zp_array
, *zp
;
1549 int max_array
= MIN(nr_to_scan
, PAGE_SIZE
* 8 / sizeof (znode_t
*));
1553 zp_array
= kmem_zalloc(max_array
* sizeof (znode_t
*), KM_SLEEP
);
1555 mutex_enter(&zfsvfs
->z_znodes_lock
);
1556 while ((zp
= list_head(&zfsvfs
->z_all_znodes
)) != NULL
) {
1558 if ((i
++ > nr_to_scan
) || (j
>= max_array
))
1561 ASSERT(list_link_active(&zp
->z_link_node
));
1562 list_remove(&zfsvfs
->z_all_znodes
, zp
);
1563 list_insert_tail(&zfsvfs
->z_all_znodes
, zp
);
1565 /* Skip active znodes and .zfs entries */
1566 if (MUTEX_HELD(&zp
->z_lock
) || zp
->z_is_ctldir
)
1569 if (igrab(ZTOI(zp
)) == NULL
)
1575 mutex_exit(&zfsvfs
->z_znodes_lock
);
1577 for (i
= 0; i
< j
; i
++) {
1580 ASSERT3P(zp
, !=, NULL
);
1581 d_prune_aliases(ZTOI(zp
));
1583 if (atomic_read(&ZTOI(zp
)->i_count
) == 1)
1589 kmem_free(zp_array
, max_array
* sizeof (znode_t
*));
1593 #endif /* HAVE_D_PRUNE_ALIASES */
1596 * The ARC has requested that the filesystem drop entries from the dentry
1597 * and inode caches. This can occur when the ARC needs to free meta data
1598 * blocks but can't because they are all pinned by entries in these caches.
1601 zfs_prune(struct super_block
*sb
, unsigned long nr_to_scan
, int *objects
)
1603 zfsvfs_t
*zfsvfs
= sb
->s_fs_info
;
1605 #if defined(HAVE_SHRINK) || defined(HAVE_SPLIT_SHRINKER_CALLBACK)
1606 struct shrinker
*shrinker
= &sb
->s_shrink
;
1607 struct shrink_control sc
= {
1608 .nr_to_scan
= nr_to_scan
,
1609 .gfp_mask
= GFP_KERNEL
,
1615 #if defined(HAVE_SPLIT_SHRINKER_CALLBACK) && \
1616 defined(SHRINK_CONTROL_HAS_NID) && \
1617 defined(SHRINKER_NUMA_AWARE)
1618 if (sb
->s_shrink
.flags
& SHRINKER_NUMA_AWARE
) {
1620 for_each_online_node(sc
.nid
) {
1621 *objects
+= (*shrinker
->scan_objects
)(shrinker
, &sc
);
1624 *objects
= (*shrinker
->scan_objects
)(shrinker
, &sc
);
1627 #elif defined(HAVE_SPLIT_SHRINKER_CALLBACK)
1628 *objects
= (*shrinker
->scan_objects
)(shrinker
, &sc
);
1629 #elif defined(HAVE_SHRINK)
1630 *objects
= (*shrinker
->shrink
)(shrinker
, &sc
);
1631 #elif defined(HAVE_D_PRUNE_ALIASES)
1632 #define D_PRUNE_ALIASES_IS_DEFAULT
1633 *objects
= zfs_prune_aliases(zfsvfs
, nr_to_scan
);
1635 #error "No available dentry and inode cache pruning mechanism."
1638 #if defined(HAVE_D_PRUNE_ALIASES) && !defined(D_PRUNE_ALIASES_IS_DEFAULT)
1639 #undef D_PRUNE_ALIASES_IS_DEFAULT
1641 * Fall back to zfs_prune_aliases if the kernel's per-superblock
1642 * shrinker couldn't free anything, possibly due to the inodes being
1643 * allocated in a different memcg.
1646 *objects
= zfs_prune_aliases(zfsvfs
, nr_to_scan
);
1651 dprintf_ds(zfsvfs
->z_os
->os_dsl_dataset
,
1652 "pruning, nr_to_scan=%lu objects=%d error=%d\n",
1653 nr_to_scan
, *objects
, error
);
1659 * Teardown the zfsvfs_t.
1661 * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
1662 * and 'z_teardown_inactive_lock' held.
1665 zfsvfs_teardown(zfsvfs_t
*zfsvfs
, boolean_t unmounting
)
1669 zfs_unlinked_drain_stop_wait(zfsvfs
);
1672 * If someone has not already unmounted this file system,
1673 * drain the iput_taskq to ensure all active references to the
1674 * zfsvfs_t have been handled only then can it be safely destroyed.
1678 * If we're unmounting we have to wait for the list to
1681 * If we're not unmounting there's no guarantee the list
1682 * will drain completely, but iputs run from the taskq
1683 * may add the parents of dir-based xattrs to the taskq
1684 * so we want to wait for these.
1686 * We can safely read z_nr_znodes without locking because the
1687 * VFS has already blocked operations which add to the
1688 * z_all_znodes list and thus increment z_nr_znodes.
1691 while (zfsvfs
->z_nr_znodes
> 0) {
1692 taskq_wait_outstanding(dsl_pool_iput_taskq(
1693 dmu_objset_pool(zfsvfs
->z_os
)), 0);
1694 if (++round
> 1 && !unmounting
)
1699 rrm_enter(&zfsvfs
->z_teardown_lock
, RW_WRITER
, FTAG
);
1703 * We purge the parent filesystem's super block as the
1704 * parent filesystem and all of its snapshots have their
1705 * inode's super block set to the parent's filesystem's
1706 * super block. Note, 'z_parent' is self referential
1707 * for non-snapshots.
1709 shrink_dcache_sb(zfsvfs
->z_parent
->z_sb
);
1713 * Close the zil. NB: Can't close the zil while zfs_inactive
1714 * threads are blocked as zil_close can call zfs_inactive.
1716 if (zfsvfs
->z_log
) {
1717 zil_close(zfsvfs
->z_log
);
1718 zfsvfs
->z_log
= NULL
;
1721 rw_enter(&zfsvfs
->z_teardown_inactive_lock
, RW_WRITER
);
1724 * If we are not unmounting (ie: online recv) and someone already
1725 * unmounted this file system while we were doing the switcheroo,
1726 * or a reopen of z_os failed then just bail out now.
1728 if (!unmounting
&& (zfsvfs
->z_unmounted
|| zfsvfs
->z_os
== NULL
)) {
1729 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
1730 rrm_exit(&zfsvfs
->z_teardown_lock
, FTAG
);
1731 return (SET_ERROR(EIO
));
1735 * At this point there are no VFS ops active, and any new VFS ops
1736 * will fail with EIO since we have z_teardown_lock for writer (only
1737 * relevant for forced unmount).
1739 * Release all holds on dbufs.
1742 mutex_enter(&zfsvfs
->z_znodes_lock
);
1743 for (zp
= list_head(&zfsvfs
->z_all_znodes
); zp
!= NULL
;
1744 zp
= list_next(&zfsvfs
->z_all_znodes
, zp
)) {
1746 zfs_znode_dmu_fini(zp
);
1748 mutex_exit(&zfsvfs
->z_znodes_lock
);
1752 * If we are unmounting, set the unmounted flag and let new VFS ops
1753 * unblock. zfs_inactive will have the unmounted behavior, and all
1754 * other VFS ops will fail with EIO.
1757 zfsvfs
->z_unmounted
= B_TRUE
;
1758 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
1759 rrm_exit(&zfsvfs
->z_teardown_lock
, FTAG
);
1763 * z_os will be NULL if there was an error in attempting to reopen
1764 * zfsvfs, so just return as the properties had already been
1766 * unregistered and cached data had been evicted before.
1768 if (zfsvfs
->z_os
== NULL
)
1772 * Unregister properties.
1774 zfs_unregister_callbacks(zfsvfs
);
1777 * Evict cached data. We must write out any dirty data before
1778 * disowning the dataset.
1780 if (!zfs_is_readonly(zfsvfs
))
1781 txg_wait_synced(dmu_objset_pool(zfsvfs
->z_os
), 0);
1782 dmu_objset_evict_dbufs(zfsvfs
->z_os
);
1787 #if !defined(HAVE_2ARGS_BDI_SETUP_AND_REGISTER) && \
1788 !defined(HAVE_3ARGS_BDI_SETUP_AND_REGISTER)
1789 atomic_long_t zfs_bdi_seq
= ATOMIC_LONG_INIT(0);
1793 zfs_domount(struct super_block
*sb
, zfs_mnt_t
*zm
, int silent
)
1795 const char *osname
= zm
->mnt_osname
;
1796 struct inode
*root_inode
;
1797 uint64_t recordsize
;
1799 zfsvfs_t
*zfsvfs
= NULL
;
1805 error
= zfsvfs_parse_options(zm
->mnt_data
, &vfs
);
1809 error
= zfsvfs_create(osname
, vfs
->vfs_readonly
, &zfsvfs
);
1811 zfsvfs_vfs_free(vfs
);
1815 if ((error
= dsl_prop_get_integer(osname
, "recordsize",
1816 &recordsize
, NULL
))) {
1817 zfsvfs_vfs_free(vfs
);
1821 vfs
->vfs_data
= zfsvfs
;
1822 zfsvfs
->z_vfs
= vfs
;
1824 sb
->s_fs_info
= zfsvfs
;
1825 sb
->s_magic
= ZFS_SUPER_MAGIC
;
1826 sb
->s_maxbytes
= MAX_LFS_FILESIZE
;
1827 sb
->s_time_gran
= 1;
1828 sb
->s_blocksize
= recordsize
;
1829 sb
->s_blocksize_bits
= ilog2(recordsize
);
1831 error
= -zpl_bdi_setup(sb
, "zfs");
1835 sb
->s_bdi
->ra_pages
= 0;
1837 /* Set callback operations for the file system. */
1838 sb
->s_op
= &zpl_super_operations
;
1839 sb
->s_xattr
= zpl_xattr_handlers
;
1840 sb
->s_export_op
= &zpl_export_operations
;
1842 sb
->s_d_op
= &zpl_dentry_operations
;
1843 #endif /* HAVE_S_D_OP */
1845 /* Set features for file system. */
1846 zfs_set_fuid_feature(zfsvfs
);
1848 if (dmu_objset_is_snapshot(zfsvfs
->z_os
)) {
1851 atime_changed_cb(zfsvfs
, B_FALSE
);
1852 readonly_changed_cb(zfsvfs
, B_TRUE
);
1853 if ((error
= dsl_prop_get_integer(osname
,
1854 "xattr", &pval
, NULL
)))
1856 xattr_changed_cb(zfsvfs
, pval
);
1857 if ((error
= dsl_prop_get_integer(osname
,
1858 "acltype", &pval
, NULL
)))
1860 acltype_changed_cb(zfsvfs
, pval
);
1861 zfsvfs
->z_issnap
= B_TRUE
;
1862 zfsvfs
->z_os
->os_sync
= ZFS_SYNC_DISABLED
;
1863 zfsvfs
->z_snap_defer_time
= jiffies
;
1865 mutex_enter(&zfsvfs
->z_os
->os_user_ptr_lock
);
1866 dmu_objset_set_user(zfsvfs
->z_os
, zfsvfs
);
1867 mutex_exit(&zfsvfs
->z_os
->os_user_ptr_lock
);
1869 if ((error
= zfsvfs_setup(zfsvfs
, B_TRUE
)))
1873 /* Allocate a root inode for the filesystem. */
1874 error
= zfs_root(zfsvfs
, &root_inode
);
1876 (void) zfs_umount(sb
);
1880 /* Allocate a root dentry for the filesystem */
1881 sb
->s_root
= d_make_root(root_inode
);
1882 if (sb
->s_root
== NULL
) {
1883 (void) zfs_umount(sb
);
1884 error
= SET_ERROR(ENOMEM
);
1888 if (!zfsvfs
->z_issnap
)
1889 zfsctl_create(zfsvfs
);
1891 zfsvfs
->z_arc_prune
= arc_add_prune_callback(zpl_prune_sb
, sb
);
1894 if (zfsvfs
!= NULL
) {
1895 dmu_objset_disown(zfsvfs
->z_os
, B_TRUE
, zfsvfs
);
1896 zfsvfs_free(zfsvfs
);
1899 * make sure we don't have dangling sb->s_fs_info which
1900 * zfs_preumount will use.
1902 sb
->s_fs_info
= NULL
;
1909 * Called when an unmount is requested and certain sanity checks have
1910 * already passed. At this point no dentries or inodes have been reclaimed
1911 * from their respective caches. We drop the extra reference on the .zfs
1912 * control directory to allow everything to be reclaimed. All snapshots
1913 * must already have been unmounted to reach this point.
1916 zfs_preumount(struct super_block
*sb
)
1918 zfsvfs_t
*zfsvfs
= sb
->s_fs_info
;
1920 /* zfsvfs is NULL when zfs_domount fails during mount */
1922 zfs_unlinked_drain_stop_wait(zfsvfs
);
1923 zfsctl_destroy(sb
->s_fs_info
);
1925 * Wait for iput_async before entering evict_inodes in
1926 * generic_shutdown_super. The reason we must finish before
1927 * evict_inodes is when lazytime is on, or when zfs_purgedir
1928 * calls zfs_zget, iput would bump i_count from 0 to 1. This
1929 * would race with the i_count check in evict_inodes. This means
1930 * it could destroy the inode while we are still using it.
1932 * We wait for two passes. xattr directories in the first pass
1933 * may add xattr entries in zfs_purgedir, so in the second pass
1934 * we wait for them. We don't use taskq_wait here because it is
1935 * a pool wide taskq. Other mounted filesystems can constantly
1936 * do iput_async and there's no guarantee when taskq will be
1939 taskq_wait_outstanding(dsl_pool_iput_taskq(
1940 dmu_objset_pool(zfsvfs
->z_os
)), 0);
1941 taskq_wait_outstanding(dsl_pool_iput_taskq(
1942 dmu_objset_pool(zfsvfs
->z_os
)), 0);
1947 * Called once all other unmount released tear down has occurred.
1948 * It is our responsibility to release any remaining infrastructure.
1952 zfs_umount(struct super_block
*sb
)
1954 zfsvfs_t
*zfsvfs
= sb
->s_fs_info
;
1957 if (zfsvfs
->z_arc_prune
!= NULL
)
1958 arc_remove_prune_callback(zfsvfs
->z_arc_prune
);
1959 VERIFY(zfsvfs_teardown(zfsvfs
, B_TRUE
) == 0);
1961 zpl_bdi_destroy(sb
);
1964 * z_os will be NULL if there was an error in
1965 * attempting to reopen zfsvfs.
1969 * Unset the objset user_ptr.
1971 mutex_enter(&os
->os_user_ptr_lock
);
1972 dmu_objset_set_user(os
, NULL
);
1973 mutex_exit(&os
->os_user_ptr_lock
);
1976 * Finally release the objset
1978 dmu_objset_disown(os
, B_TRUE
, zfsvfs
);
1981 zfsvfs_free(zfsvfs
);
1986 zfs_remount(struct super_block
*sb
, int *flags
, zfs_mnt_t
*zm
)
1988 zfsvfs_t
*zfsvfs
= sb
->s_fs_info
;
1990 boolean_t issnap
= dmu_objset_is_snapshot(zfsvfs
->z_os
);
1993 if ((issnap
|| !spa_writeable(dmu_objset_spa(zfsvfs
->z_os
))) &&
1994 !(*flags
& SB_RDONLY
)) {
1995 *flags
|= SB_RDONLY
;
1999 error
= zfsvfs_parse_options(zm
->mnt_data
, &vfsp
);
2003 if (!zfs_is_readonly(zfsvfs
) && (*flags
& SB_RDONLY
))
2004 txg_wait_synced(dmu_objset_pool(zfsvfs
->z_os
), 0);
2006 zfs_unregister_callbacks(zfsvfs
);
2007 zfsvfs_vfs_free(zfsvfs
->z_vfs
);
2009 vfsp
->vfs_data
= zfsvfs
;
2010 zfsvfs
->z_vfs
= vfsp
;
2012 (void) zfs_register_callbacks(vfsp
);
2018 zfs_vget(struct super_block
*sb
, struct inode
**ipp
, fid_t
*fidp
)
2020 zfsvfs_t
*zfsvfs
= sb
->s_fs_info
;
2022 uint64_t object
= 0;
2023 uint64_t fid_gen
= 0;
2030 if (fidp
->fid_len
== SHORT_FID_LEN
|| fidp
->fid_len
== LONG_FID_LEN
) {
2031 zfid_short_t
*zfid
= (zfid_short_t
*)fidp
;
2033 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
2034 object
|= ((uint64_t)zfid
->zf_object
[i
]) << (8 * i
);
2036 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
2037 fid_gen
|= ((uint64_t)zfid
->zf_gen
[i
]) << (8 * i
);
2039 return (SET_ERROR(EINVAL
));
2042 /* LONG_FID_LEN means snapdirs */
2043 if (fidp
->fid_len
== LONG_FID_LEN
) {
2044 zfid_long_t
*zlfid
= (zfid_long_t
*)fidp
;
2045 uint64_t objsetid
= 0;
2046 uint64_t setgen
= 0;
2048 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
2049 objsetid
|= ((uint64_t)zlfid
->zf_setid
[i
]) << (8 * i
);
2051 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
2052 setgen
|= ((uint64_t)zlfid
->zf_setgen
[i
]) << (8 * i
);
2054 if (objsetid
!= ZFSCTL_INO_SNAPDIRS
- object
) {
2055 dprintf("snapdir fid: objsetid (%llu) != "
2056 "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n",
2057 objsetid
, ZFSCTL_INO_SNAPDIRS
, object
);
2059 return (SET_ERROR(EINVAL
));
2062 if (fid_gen
> 1 || setgen
!= 0) {
2063 dprintf("snapdir fid: fid_gen (%llu) and setgen "
2064 "(%llu)\n", fid_gen
, setgen
);
2065 return (SET_ERROR(EINVAL
));
2068 return (zfsctl_snapdir_vget(sb
, objsetid
, fid_gen
, ipp
));
2072 /* A zero fid_gen means we are in the .zfs control directories */
2074 (object
== ZFSCTL_INO_ROOT
|| object
== ZFSCTL_INO_SNAPDIR
)) {
2075 *ipp
= zfsvfs
->z_ctldir
;
2076 ASSERT(*ipp
!= NULL
);
2077 if (object
== ZFSCTL_INO_SNAPDIR
) {
2078 VERIFY(zfsctl_root_lookup(*ipp
, "snapshot", ipp
,
2079 0, kcred
, NULL
, NULL
) == 0);
2087 gen_mask
= -1ULL >> (64 - 8 * i
);
2089 dprintf("getting %llu [%llu mask %llx]\n", object
, fid_gen
, gen_mask
);
2090 if ((err
= zfs_zget(zfsvfs
, object
, &zp
))) {
2095 /* Don't export xattr stuff */
2096 if (zp
->z_pflags
& ZFS_XATTR
) {
2099 return (SET_ERROR(ENOENT
));
2102 (void) sa_lookup(zp
->z_sa_hdl
, SA_ZPL_GEN(zfsvfs
), &zp_gen
,
2104 zp_gen
= zp_gen
& gen_mask
;
2107 if ((fid_gen
== 0) && (zfsvfs
->z_root
== object
))
2109 if (zp
->z_unlinked
|| zp_gen
!= fid_gen
) {
2110 dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen
,
2114 return (SET_ERROR(ENOENT
));
2119 zfs_inode_update(ITOZ(*ipp
));
2126 * Block out VFS ops and close zfsvfs_t
2128 * Note, if successful, then we return with the 'z_teardown_lock' and
2129 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2130 * dataset and objset intact so that they can be atomically handed off during
2131 * a subsequent rollback or recv operation and the resume thereafter.
2134 zfs_suspend_fs(zfsvfs_t
*zfsvfs
)
2138 if ((error
= zfsvfs_teardown(zfsvfs
, B_FALSE
)) != 0)
2145 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2146 * is an invariant across any of the operations that can be performed while the
2147 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2148 * are the same: the relevant objset and associated dataset are owned by
2149 * zfsvfs, held, and long held on entry.
2152 zfs_resume_fs(zfsvfs_t
*zfsvfs
, dsl_dataset_t
*ds
)
2157 ASSERT(RRM_WRITE_HELD(&zfsvfs
->z_teardown_lock
));
2158 ASSERT(RW_WRITE_HELD(&zfsvfs
->z_teardown_inactive_lock
));
2161 * We already own this, so just update the objset_t, as the one we
2162 * had before may have been evicted.
2165 VERIFY3P(ds
->ds_owner
, ==, zfsvfs
);
2166 VERIFY(dsl_dataset_long_held(ds
));
2167 VERIFY0(dmu_objset_from_ds(ds
, &os
));
2169 err
= zfsvfs_init(zfsvfs
, os
);
2173 VERIFY(zfsvfs_setup(zfsvfs
, B_FALSE
) == 0);
2175 zfs_set_fuid_feature(zfsvfs
);
2176 zfsvfs
->z_rollback_time
= jiffies
;
2179 * Attempt to re-establish all the active inodes with their
2180 * dbufs. If a zfs_rezget() fails, then we unhash the inode
2181 * and mark it stale. This prevents a collision if a new
2182 * inode/object is created which must use the same inode
2183 * number. The stale inode will be be released when the
2184 * VFS prunes the dentry holding the remaining references
2185 * on the stale inode.
2187 mutex_enter(&zfsvfs
->z_znodes_lock
);
2188 for (zp
= list_head(&zfsvfs
->z_all_znodes
); zp
;
2189 zp
= list_next(&zfsvfs
->z_all_znodes
, zp
)) {
2190 err2
= zfs_rezget(zp
);
2192 remove_inode_hash(ZTOI(zp
));
2193 zp
->z_is_stale
= B_TRUE
;
2196 mutex_exit(&zfsvfs
->z_znodes_lock
);
2198 if (!zfs_is_readonly(zfsvfs
) && !zfsvfs
->z_unmounted
) {
2200 * zfs_suspend_fs() could have interrupted freeing
2201 * of dnodes. We need to restart this freeing so
2202 * that we don't "leak" the space.
2204 zfs_unlinked_drain(zfsvfs
);
2208 /* release the VFS ops */
2209 rw_exit(&zfsvfs
->z_teardown_inactive_lock
);
2210 rrm_exit(&zfsvfs
->z_teardown_lock
, FTAG
);
2214 * Since we couldn't setup the sa framework, try to force
2215 * unmount this file system.
2218 (void) zfs_umount(zfsvfs
->z_sb
);
2224 zfs_set_version(zfsvfs_t
*zfsvfs
, uint64_t newvers
)
2227 objset_t
*os
= zfsvfs
->z_os
;
2230 if (newvers
< ZPL_VERSION_INITIAL
|| newvers
> ZPL_VERSION
)
2231 return (SET_ERROR(EINVAL
));
2233 if (newvers
< zfsvfs
->z_version
)
2234 return (SET_ERROR(EINVAL
));
2236 if (zfs_spa_version_map(newvers
) >
2237 spa_version(dmu_objset_spa(zfsvfs
->z_os
)))
2238 return (SET_ERROR(ENOTSUP
));
2240 tx
= dmu_tx_create(os
);
2241 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_FALSE
, ZPL_VERSION_STR
);
2242 if (newvers
>= ZPL_VERSION_SA
&& !zfsvfs
->z_use_sa
) {
2243 dmu_tx_hold_zap(tx
, MASTER_NODE_OBJ
, B_TRUE
,
2245 dmu_tx_hold_zap(tx
, DMU_NEW_OBJECT
, FALSE
, NULL
);
2247 error
= dmu_tx_assign(tx
, TXG_WAIT
);
2253 error
= zap_update(os
, MASTER_NODE_OBJ
, ZPL_VERSION_STR
,
2254 8, 1, &newvers
, tx
);
2261 if (newvers
>= ZPL_VERSION_SA
&& !zfsvfs
->z_use_sa
) {
2264 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs
->z_os
)), >=,
2266 sa_obj
= zap_create(os
, DMU_OT_SA_MASTER_NODE
,
2267 DMU_OT_NONE
, 0, tx
);
2269 error
= zap_add(os
, MASTER_NODE_OBJ
,
2270 ZFS_SA_ATTRS
, 8, 1, &sa_obj
, tx
);
2273 VERIFY(0 == sa_set_sa_object(os
, sa_obj
));
2274 sa_register_update_callback(os
, zfs_sa_upgrade
);
2277 spa_history_log_internal_ds(dmu_objset_ds(os
), "upgrade", tx
,
2278 "from %llu to %llu", zfsvfs
->z_version
, newvers
);
2282 zfsvfs
->z_version
= newvers
;
2283 os
->os_version
= newvers
;
2285 zfs_set_fuid_feature(zfsvfs
);
2291 * Read a property stored within the master node.
2294 zfs_get_zplprop(objset_t
*os
, zfs_prop_t prop
, uint64_t *value
)
2296 uint64_t *cached_copy
= NULL
;
2299 * Figure out where in the objset_t the cached copy would live, if it
2300 * is available for the requested property.
2304 case ZFS_PROP_VERSION
:
2305 cached_copy
= &os
->os_version
;
2307 case ZFS_PROP_NORMALIZE
:
2308 cached_copy
= &os
->os_normalization
;
2310 case ZFS_PROP_UTF8ONLY
:
2311 cached_copy
= &os
->os_utf8only
;
2314 cached_copy
= &os
->os_casesensitivity
;
2320 if (cached_copy
!= NULL
&& *cached_copy
!= OBJSET_PROP_UNINITIALIZED
) {
2321 *value
= *cached_copy
;
2326 * If the property wasn't cached, look up the file system's value for
2327 * the property. For the version property, we look up a slightly
2332 if (prop
== ZFS_PROP_VERSION
)
2333 pname
= ZPL_VERSION_STR
;
2335 pname
= zfs_prop_to_name(prop
);
2338 ASSERT3U(os
->os_phys
->os_type
, ==, DMU_OST_ZFS
);
2339 error
= zap_lookup(os
, MASTER_NODE_OBJ
, pname
, 8, 1, value
);
2342 if (error
== ENOENT
) {
2343 /* No value set, use the default value */
2345 case ZFS_PROP_VERSION
:
2346 *value
= ZPL_VERSION
;
2348 case ZFS_PROP_NORMALIZE
:
2349 case ZFS_PROP_UTF8ONLY
:
2353 *value
= ZFS_CASE_SENSITIVE
;
2355 case ZFS_PROP_ACLTYPE
:
2356 *value
= ZFS_ACLTYPE_OFF
;
2365 * If one of the methods for getting the property value above worked,
2366 * copy it into the objset_t's cache.
2368 if (error
== 0 && cached_copy
!= NULL
) {
2369 *cached_copy
= *value
;
2376 * Return true if the coresponding vfs's unmounted flag is set.
2377 * Otherwise return false.
2378 * If this function returns true we know VFS unmount has been initiated.
2381 zfs_get_vfs_flag_unmounted(objset_t
*os
)
2384 boolean_t unmounted
= B_FALSE
;
2386 ASSERT(dmu_objset_type(os
) == DMU_OST_ZFS
);
2388 mutex_enter(&os
->os_user_ptr_lock
);
2389 zfvp
= dmu_objset_get_user(os
);
2390 if (zfvp
!= NULL
&& zfvp
->z_unmounted
)
2392 mutex_exit(&os
->os_user_ptr_lock
);
2402 dmu_objset_register_type(DMU_OST_ZFS
, zfs_space_delta_cb
);
2403 register_filesystem(&zpl_fs_type
);
2410 * we don't use outstanding because zpl_posix_acl_free might add more.
2412 taskq_wait(system_delay_taskq
);
2413 taskq_wait(system_taskq
);
2414 unregister_filesystem(&zpl_fs_type
);
2419 #if defined(_KERNEL)
2420 EXPORT_SYMBOL(zfs_suspend_fs
);
2421 EXPORT_SYMBOL(zfs_resume_fs
);
2422 EXPORT_SYMBOL(zfs_userspace_one
);
2423 EXPORT_SYMBOL(zfs_userspace_many
);
2424 EXPORT_SYMBOL(zfs_set_userquota
);
2425 EXPORT_SYMBOL(zfs_id_overblockquota
);
2426 EXPORT_SYMBOL(zfs_id_overobjquota
);
2427 EXPORT_SYMBOL(zfs_id_overquota
);
2428 EXPORT_SYMBOL(zfs_set_version
);
2429 EXPORT_SYMBOL(zfsvfs_create
);
2430 EXPORT_SYMBOL(zfsvfs_free
);
2431 EXPORT_SYMBOL(zfs_is_readonly
);
2432 EXPORT_SYMBOL(zfs_domount
);
2433 EXPORT_SYMBOL(zfs_preumount
);
2434 EXPORT_SYMBOL(zfs_umount
);
2435 EXPORT_SYMBOL(zfs_remount
);
2436 EXPORT_SYMBOL(zfs_statvfs
);
2437 EXPORT_SYMBOL(zfs_vget
);
2438 EXPORT_SYMBOL(zfs_prune
);