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/zfs_quota.h>
56 #include <sys/sunddi.h>
57 #include <sys/dmu_objset.h>
58 #include <sys/dsl_dir.h>
59 #include <sys/spa_boot.h>
60 #include <sys/objlist.h>
62 #include <linux/vfs_compat.h>
63 #include "zfs_comutil.h"
88 static const match_table_t zpl_tokens = {
89 { TOKEN_RO, MNTOPT_RO },
90 { TOKEN_RW, MNTOPT_RW },
91 { TOKEN_SETUID, MNTOPT_SETUID },
92 { TOKEN_NOSETUID, MNTOPT_NOSETUID },
93 { TOKEN_EXEC, MNTOPT_EXEC },
94 { TOKEN_NOEXEC, MNTOPT_NOEXEC },
95 { TOKEN_DEVICES, MNTOPT_DEVICES },
96 { TOKEN_NODEVICES, MNTOPT_NODEVICES },
97 { TOKEN_DIRXATTR, MNTOPT_DIRXATTR },
98 { TOKEN_SAXATTR, MNTOPT_SAXATTR },
99 { TOKEN_XATTR, MNTOPT_XATTR },
100 { TOKEN_NOXATTR, MNTOPT_NOXATTR },
101 { TOKEN_ATIME, MNTOPT_ATIME },
102 { TOKEN_NOATIME, MNTOPT_NOATIME },
103 { TOKEN_RELATIME, MNTOPT_RELATIME },
104 { TOKEN_NORELATIME, MNTOPT_NORELATIME },
105 { TOKEN_NBMAND, MNTOPT_NBMAND },
106 { TOKEN_NONBMAND, MNTOPT_NONBMAND },
107 { TOKEN_MNTPOINT, MNTOPT_MNTPOINT "=%s" },
108 { TOKEN_LAST, NULL },
112 zfsvfs_vfs_free(vfs_t *vfsp)
115 if (vfsp->vfs_mntpoint != NULL)
116 kmem_strfree(vfsp->vfs_mntpoint);
118 kmem_free(vfsp, sizeof (vfs_t));
123 zfsvfs_parse_option(char *option, int token, substring_t *args, vfs_t *vfsp)
127 vfsp->vfs_readonly = B_TRUE;
128 vfsp->vfs_do_readonly = B_TRUE;
131 vfsp->vfs_readonly = B_FALSE;
132 vfsp->vfs_do_readonly = B_TRUE;
135 vfsp->vfs_setuid = B_TRUE;
136 vfsp->vfs_do_setuid = B_TRUE;
139 vfsp->vfs_setuid = B_FALSE;
140 vfsp->vfs_do_setuid = B_TRUE;
143 vfsp->vfs_exec = B_TRUE;
144 vfsp->vfs_do_exec = B_TRUE;
147 vfsp->vfs_exec = B_FALSE;
148 vfsp->vfs_do_exec = B_TRUE;
151 vfsp->vfs_devices = B_TRUE;
152 vfsp->vfs_do_devices = B_TRUE;
154 case TOKEN_NODEVICES:
155 vfsp->vfs_devices = B_FALSE;
156 vfsp->vfs_do_devices = B_TRUE;
159 vfsp->vfs_xattr = ZFS_XATTR_DIR;
160 vfsp->vfs_do_xattr = B_TRUE;
163 vfsp->vfs_xattr = ZFS_XATTR_SA;
164 vfsp->vfs_do_xattr = B_TRUE;
167 vfsp->vfs_xattr = ZFS_XATTR_DIR;
168 vfsp->vfs_do_xattr = B_TRUE;
171 vfsp->vfs_xattr = ZFS_XATTR_OFF;
172 vfsp->vfs_do_xattr = B_TRUE;
175 vfsp->vfs_atime = B_TRUE;
176 vfsp->vfs_do_atime = B_TRUE;
179 vfsp->vfs_atime = B_FALSE;
180 vfsp->vfs_do_atime = B_TRUE;
183 vfsp->vfs_relatime = B_TRUE;
184 vfsp->vfs_do_relatime = B_TRUE;
186 case TOKEN_NORELATIME:
187 vfsp->vfs_relatime = B_FALSE;
188 vfsp->vfs_do_relatime = B_TRUE;
191 vfsp->vfs_nbmand = B_TRUE;
192 vfsp->vfs_do_nbmand = B_TRUE;
195 vfsp->vfs_nbmand = B_FALSE;
196 vfsp->vfs_do_nbmand = B_TRUE;
199 vfsp->vfs_mntpoint = match_strdup(&args[0]);
200 if (vfsp->vfs_mntpoint == NULL)
201 return (SET_ERROR(ENOMEM));
212 * Parse the raw mntopts and return a vfs_t describing the options.
215 zfsvfs_parse_options(char *mntopts, vfs_t **vfsp)
220 tmp_vfsp = kmem_zalloc(sizeof (vfs_t), KM_SLEEP);
222 if (mntopts != NULL) {
223 substring_t args[MAX_OPT_ARGS];
224 char *tmp_mntopts, *p, *t;
227 tmp_mntopts = t = kmem_strdup(mntopts);
228 if (tmp_mntopts == NULL)
229 return (SET_ERROR(ENOMEM));
231 while ((p = strsep(&t, ",")) != NULL) {
235 args[0].to = args[0].from = NULL;
236 token = match_token(p, zpl_tokens, args);
237 error = zfsvfs_parse_option(p, token, args, tmp_vfsp);
239 kmem_strfree(tmp_mntopts);
240 zfsvfs_vfs_free(tmp_vfsp);
245 kmem_strfree(tmp_mntopts);
254 zfs_is_readonly(zfsvfs_t *zfsvfs)
256 return (!!(zfsvfs->z_sb->s_flags & SB_RDONLY));
261 zfs_sync(struct super_block *sb, int wait, cred_t *cr)
263 zfsvfs_t *zfsvfs = sb->s_fs_info;
266 * Semantically, the only requirement is that the sync be initiated.
267 * The DMU syncs out txgs frequently, so there's nothing to do.
272 if (zfsvfs != NULL) {
274 * Sync a specific filesystem.
279 dp = dmu_objset_pool(zfsvfs->z_os);
282 * If the system is shutting down, then skip any
283 * filesystems which may exist on a suspended pool.
285 if (spa_suspended(dp->dp_spa)) {
290 if (zfsvfs->z_log != NULL)
291 zil_commit(zfsvfs->z_log, 0);
296 * Sync all ZFS filesystems. This is what happens when you
297 * run sync(1). Unlike other filesystems, ZFS honors the
298 * request by waiting for all pools to commit all dirty data.
307 atime_changed_cb(void *arg, uint64_t newval)
309 zfsvfs_t *zfsvfs = arg;
310 struct super_block *sb = zfsvfs->z_sb;
315 * Update SB_NOATIME bit in VFS super block. Since atime update is
316 * determined by atime_needs_update(), atime_needs_update() needs to
317 * return false if atime is turned off, and not unconditionally return
318 * false if atime is turned on.
321 sb->s_flags &= ~SB_NOATIME;
323 sb->s_flags |= SB_NOATIME;
327 relatime_changed_cb(void *arg, uint64_t newval)
329 ((zfsvfs_t *)arg)->z_relatime = newval;
333 xattr_changed_cb(void *arg, uint64_t newval)
335 zfsvfs_t *zfsvfs = arg;
337 if (newval == ZFS_XATTR_OFF) {
338 zfsvfs->z_flags &= ~ZSB_XATTR;
340 zfsvfs->z_flags |= ZSB_XATTR;
342 if (newval == ZFS_XATTR_SA)
343 zfsvfs->z_xattr_sa = B_TRUE;
345 zfsvfs->z_xattr_sa = B_FALSE;
350 acltype_changed_cb(void *arg, uint64_t newval)
352 zfsvfs_t *zfsvfs = arg;
355 case ZFS_ACLTYPE_NFSV4:
356 case ZFS_ACLTYPE_OFF:
357 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
358 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
360 case ZFS_ACLTYPE_POSIX:
361 #ifdef CONFIG_FS_POSIX_ACL
362 zfsvfs->z_acl_type = ZFS_ACLTYPE_POSIX;
363 zfsvfs->z_sb->s_flags |= SB_POSIXACL;
365 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
366 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
367 #endif /* CONFIG_FS_POSIX_ACL */
375 blksz_changed_cb(void *arg, uint64_t newval)
377 zfsvfs_t *zfsvfs = arg;
378 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
379 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
380 ASSERT(ISP2(newval));
382 zfsvfs->z_max_blksz = newval;
386 readonly_changed_cb(void *arg, uint64_t newval)
388 zfsvfs_t *zfsvfs = arg;
389 struct super_block *sb = zfsvfs->z_sb;
395 sb->s_flags |= SB_RDONLY;
397 sb->s_flags &= ~SB_RDONLY;
401 devices_changed_cb(void *arg, uint64_t newval)
406 setuid_changed_cb(void *arg, uint64_t newval)
411 exec_changed_cb(void *arg, uint64_t newval)
416 nbmand_changed_cb(void *arg, uint64_t newval)
418 zfsvfs_t *zfsvfs = arg;
419 struct super_block *sb = zfsvfs->z_sb;
425 sb->s_flags |= SB_MANDLOCK;
427 sb->s_flags &= ~SB_MANDLOCK;
431 snapdir_changed_cb(void *arg, uint64_t newval)
433 ((zfsvfs_t *)arg)->z_show_ctldir = newval;
437 acl_mode_changed_cb(void *arg, uint64_t newval)
439 zfsvfs_t *zfsvfs = arg;
441 zfsvfs->z_acl_mode = newval;
445 acl_inherit_changed_cb(void *arg, uint64_t newval)
447 ((zfsvfs_t *)arg)->z_acl_inherit = newval;
451 zfs_register_callbacks(vfs_t *vfsp)
453 struct dsl_dataset *ds = NULL;
455 zfsvfs_t *zfsvfs = NULL;
459 zfsvfs = vfsp->vfs_data;
464 * The act of registering our callbacks will destroy any mount
465 * options we may have. In order to enable temporary overrides
466 * of mount options, we stash away the current values and
467 * restore them after we register the callbacks.
469 if (zfs_is_readonly(zfsvfs) || !spa_writeable(dmu_objset_spa(os))) {
470 vfsp->vfs_do_readonly = B_TRUE;
471 vfsp->vfs_readonly = B_TRUE;
475 * Register property callbacks.
477 * It would probably be fine to just check for i/o error from
478 * the first prop_register(), but I guess I like to go
481 ds = dmu_objset_ds(os);
482 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
483 error = dsl_prop_register(ds,
484 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
485 error = error ? error : dsl_prop_register(ds,
486 zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zfsvfs);
487 error = error ? error : dsl_prop_register(ds,
488 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
489 error = error ? error : dsl_prop_register(ds,
490 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
491 error = error ? error : dsl_prop_register(ds,
492 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
493 error = error ? error : dsl_prop_register(ds,
494 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
495 error = error ? error : dsl_prop_register(ds,
496 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
497 error = error ? error : dsl_prop_register(ds,
498 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
499 error = error ? error : dsl_prop_register(ds,
500 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
501 error = error ? error : dsl_prop_register(ds,
502 zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zfsvfs);
503 error = error ? error : dsl_prop_register(ds,
504 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
505 error = error ? error : dsl_prop_register(ds,
506 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
508 error = error ? error : dsl_prop_register(ds,
509 zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zfsvfs);
510 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
515 * Invoke our callbacks to restore temporary mount options.
517 if (vfsp->vfs_do_readonly)
518 readonly_changed_cb(zfsvfs, vfsp->vfs_readonly);
519 if (vfsp->vfs_do_setuid)
520 setuid_changed_cb(zfsvfs, vfsp->vfs_setuid);
521 if (vfsp->vfs_do_exec)
522 exec_changed_cb(zfsvfs, vfsp->vfs_exec);
523 if (vfsp->vfs_do_devices)
524 devices_changed_cb(zfsvfs, vfsp->vfs_devices);
525 if (vfsp->vfs_do_xattr)
526 xattr_changed_cb(zfsvfs, vfsp->vfs_xattr);
527 if (vfsp->vfs_do_atime)
528 atime_changed_cb(zfsvfs, vfsp->vfs_atime);
529 if (vfsp->vfs_do_relatime)
530 relatime_changed_cb(zfsvfs, vfsp->vfs_relatime);
531 if (vfsp->vfs_do_nbmand)
532 nbmand_changed_cb(zfsvfs, vfsp->vfs_nbmand);
537 dsl_prop_unregister_all(ds, zfsvfs);
542 * Takes a dataset, a property, a value and that value's setpoint as
543 * found in the ZAP. Checks if the property has been changed in the vfs.
544 * If so, val and setpoint will be overwritten with updated content.
545 * Otherwise, they are left unchanged.
548 zfs_get_temporary_prop(dsl_dataset_t *ds, zfs_prop_t zfs_prop, uint64_t *val,
557 error = dmu_objset_from_ds(ds, &os);
561 if (dmu_objset_type(os) != DMU_OST_ZFS)
564 mutex_enter(&os->os_user_ptr_lock);
565 zfvp = dmu_objset_get_user(os);
566 mutex_exit(&os->os_user_ptr_lock);
574 if (vfsp->vfs_do_atime)
575 tmp = vfsp->vfs_atime;
577 case ZFS_PROP_RELATIME:
578 if (vfsp->vfs_do_relatime)
579 tmp = vfsp->vfs_relatime;
581 case ZFS_PROP_DEVICES:
582 if (vfsp->vfs_do_devices)
583 tmp = vfsp->vfs_devices;
586 if (vfsp->vfs_do_exec)
587 tmp = vfsp->vfs_exec;
589 case ZFS_PROP_SETUID:
590 if (vfsp->vfs_do_setuid)
591 tmp = vfsp->vfs_setuid;
593 case ZFS_PROP_READONLY:
594 if (vfsp->vfs_do_readonly)
595 tmp = vfsp->vfs_readonly;
598 if (vfsp->vfs_do_xattr)
599 tmp = vfsp->vfs_xattr;
601 case ZFS_PROP_NBMAND:
602 if (vfsp->vfs_do_nbmand)
603 tmp = vfsp->vfs_nbmand;
610 (void) strcpy(setpoint, "temporary");
617 * Associate this zfsvfs with the given objset, which must be owned.
618 * This will cache a bunch of on-disk state from the objset in the
622 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
627 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
628 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
631 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
634 if (zfsvfs->z_version >
635 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
636 (void) printk("Can't mount a version %lld file system "
637 "on a version %lld pool\n. Pool must be upgraded to mount "
638 "this file system.\n", (u_longlong_t)zfsvfs->z_version,
639 (u_longlong_t)spa_version(dmu_objset_spa(os)));
640 return (SET_ERROR(ENOTSUP));
642 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
645 zfsvfs->z_norm = (int)val;
647 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
650 zfsvfs->z_utf8 = (val != 0);
652 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
655 zfsvfs->z_case = (uint_t)val;
657 if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val)) != 0)
659 zfsvfs->z_acl_type = (uint_t)val;
662 * Fold case on file systems that are always or sometimes case
665 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
666 zfsvfs->z_case == ZFS_CASE_MIXED)
667 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
669 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
670 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
673 if (zfsvfs->z_use_sa) {
674 /* should either have both of these objects or none */
675 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
680 error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val);
681 if ((error == 0) && (val == ZFS_XATTR_SA))
682 zfsvfs->z_xattr_sa = B_TRUE;
685 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
689 ASSERT(zfsvfs->z_root != 0);
691 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
692 &zfsvfs->z_unlinkedobj);
696 error = zap_lookup(os, MASTER_NODE_OBJ,
697 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
698 8, 1, &zfsvfs->z_userquota_obj);
700 zfsvfs->z_userquota_obj = 0;
704 error = zap_lookup(os, MASTER_NODE_OBJ,
705 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
706 8, 1, &zfsvfs->z_groupquota_obj);
708 zfsvfs->z_groupquota_obj = 0;
712 error = zap_lookup(os, MASTER_NODE_OBJ,
713 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA],
714 8, 1, &zfsvfs->z_projectquota_obj);
716 zfsvfs->z_projectquota_obj = 0;
720 error = zap_lookup(os, MASTER_NODE_OBJ,
721 zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
722 8, 1, &zfsvfs->z_userobjquota_obj);
724 zfsvfs->z_userobjquota_obj = 0;
728 error = zap_lookup(os, MASTER_NODE_OBJ,
729 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
730 8, 1, &zfsvfs->z_groupobjquota_obj);
732 zfsvfs->z_groupobjquota_obj = 0;
736 error = zap_lookup(os, MASTER_NODE_OBJ,
737 zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA],
738 8, 1, &zfsvfs->z_projectobjquota_obj);
740 zfsvfs->z_projectobjquota_obj = 0;
744 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
745 &zfsvfs->z_fuid_obj);
747 zfsvfs->z_fuid_obj = 0;
751 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
752 &zfsvfs->z_shares_dir);
754 zfsvfs->z_shares_dir = 0;
758 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
759 &zfsvfs->z_attr_table);
763 if (zfsvfs->z_version >= ZPL_VERSION_SA)
764 sa_register_update_callback(os, zfs_sa_upgrade);
770 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp)
775 boolean_t ro = (readonly || (strchr(osname, '@') != NULL));
777 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
779 error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os);
781 kmem_free(zfsvfs, sizeof (zfsvfs_t));
785 error = zfsvfs_create_impl(zfvp, zfsvfs, os);
787 dmu_objset_disown(os, B_TRUE, zfsvfs);
794 * Note: zfsvfs is assumed to be malloc'd, and will be freed by this function
795 * on a failure. Do not pass in a statically allocated zfsvfs.
798 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
802 zfsvfs->z_vfs = NULL;
804 zfsvfs->z_parent = zfsvfs;
806 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
807 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
808 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
809 offsetof(znode_t, z_link_node));
810 ZFS_TEARDOWN_INIT(zfsvfs);
811 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
812 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
814 int size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1),
816 zfsvfs->z_hold_size = size;
817 zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
819 zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
820 for (int i = 0; i != size; i++) {
821 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
822 sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
823 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
826 error = zfsvfs_init(zfsvfs, os);
833 zfsvfs->z_drain_task = TASKQID_INVALID;
834 zfsvfs->z_draining = B_FALSE;
835 zfsvfs->z_drain_cancel = B_TRUE;
842 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
845 boolean_t readonly = zfs_is_readonly(zfsvfs);
847 error = zfs_register_callbacks(zfsvfs->z_vfs);
851 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
854 * If we are not mounting (ie: online recv), then we don't
855 * have to worry about replaying the log as we blocked all
856 * operations out since we closed the ZIL.
859 ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL);
860 dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
863 * During replay we remove the read only flag to
864 * allow replays to succeed.
867 readonly_changed_cb(zfsvfs, B_FALSE);
870 if (zap_get_stats(zfsvfs->z_os, zfsvfs->z_unlinkedobj,
872 dataset_kstats_update_nunlinks_kstat(
873 &zfsvfs->z_kstat, zs.zs_num_entries);
874 dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
875 "num_entries in unlinked set: %llu",
878 zfs_unlinked_drain(zfsvfs);
879 dsl_dir_t *dd = zfsvfs->z_os->os_dsl_dataset->ds_dir;
880 dd->dd_activity_cancelled = B_FALSE;
884 * Parse and replay the intent log.
886 * Because of ziltest, this must be done after
887 * zfs_unlinked_drain(). (Further note: ziltest
888 * doesn't use readonly mounts, where
889 * zfs_unlinked_drain() isn't called.) This is because
890 * ziltest causes spa_sync() to think it's committed,
891 * but actually it is not, so the intent log contains
892 * many txg's worth of changes.
894 * In particular, if object N is in the unlinked set in
895 * the last txg to actually sync, then it could be
896 * actually freed in a later txg and then reallocated
897 * in a yet later txg. This would write a "create
898 * object N" record to the intent log. Normally, this
899 * would be fine because the spa_sync() would have
900 * written out the fact that object N is free, before
901 * we could write the "create object N" intent log
904 * But when we are in ziltest mode, we advance the "open
905 * txg" without actually spa_sync()-ing the changes to
906 * disk. So we would see that object N is still
907 * allocated and in the unlinked set, and there is an
908 * intent log record saying to allocate it.
910 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
911 if (zil_replay_disable) {
912 zil_destroy(zfsvfs->z_log, B_FALSE);
914 zfsvfs->z_replay = B_TRUE;
915 zil_replay(zfsvfs->z_os, zfsvfs,
917 zfsvfs->z_replay = B_FALSE;
921 /* restore readonly bit */
923 readonly_changed_cb(zfsvfs, B_TRUE);
927 * Set the objset user_ptr to track its zfsvfs.
929 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
930 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
931 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
937 zfsvfs_free(zfsvfs_t *zfsvfs)
939 int i, size = zfsvfs->z_hold_size;
941 zfs_fuid_destroy(zfsvfs);
943 mutex_destroy(&zfsvfs->z_znodes_lock);
944 mutex_destroy(&zfsvfs->z_lock);
945 list_destroy(&zfsvfs->z_all_znodes);
946 ZFS_TEARDOWN_DESTROY(zfsvfs);
947 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
948 rw_destroy(&zfsvfs->z_fuid_lock);
949 for (i = 0; i != size; i++) {
950 avl_destroy(&zfsvfs->z_hold_trees[i]);
951 mutex_destroy(&zfsvfs->z_hold_locks[i]);
953 vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
954 vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
955 zfsvfs_vfs_free(zfsvfs->z_vfs);
956 dataset_kstats_destroy(&zfsvfs->z_kstat);
957 kmem_free(zfsvfs, sizeof (zfsvfs_t));
961 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
963 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
964 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
968 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
970 objset_t *os = zfsvfs->z_os;
972 if (!dmu_objset_is_snapshot(os))
973 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
978 * Check that the hex label string is appropriate for the dataset being
979 * mounted into the global_zone proper.
981 * Return an error if the hex label string is not default or
982 * admin_low/admin_high. For admin_low labels, the corresponding
983 * dataset must be readonly.
986 zfs_check_global_label(const char *dsname, const char *hexsl)
988 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
990 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
992 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
993 /* must be readonly */
996 if (dsl_prop_get_integer(dsname,
997 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
998 return (SET_ERROR(EACCES));
999 return (rdonly ? 0 : SET_ERROR(EACCES));
1001 return (SET_ERROR(EACCES));
1003 #endif /* HAVE_MLSLABEL */
1006 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct kstatfs *statp,
1009 char buf[20 + DMU_OBJACCT_PREFIX_LEN];
1010 uint64_t offset = DMU_OBJACCT_PREFIX_LEN;
1015 strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
1016 err = zfs_id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset,
1017 sizeof (buf) - offset, B_FALSE);
1021 if (zfsvfs->z_projectquota_obj == 0)
1024 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj,
1025 buf + offset, 8, 1, "a);
1031 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1032 buf + offset, 8, 1, &used);
1033 if (unlikely(err == ENOENT)) {
1035 u_longlong_t nblocks;
1038 * Quota accounting is async, so it is possible race case.
1039 * There is at least one object with the given project ID.
1041 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
1042 if (unlikely(zp->z_blksz == 0))
1043 blksize = zfsvfs->z_max_blksz;
1045 used = blksize * nblocks;
1050 statp->f_blocks = quota >> bshift;
1051 statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0;
1052 statp->f_bavail = statp->f_bfree;
1055 if (zfsvfs->z_projectobjquota_obj == 0)
1058 err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj,
1059 buf + offset, 8, 1, "a);
1065 err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1067 if (unlikely(err == ENOENT)) {
1069 * Quota accounting is async, so it is possible race case.
1070 * There is at least one object with the given project ID.
1077 statp->f_files = quota;
1078 statp->f_ffree = (quota > used) ? (quota - used) : 0;
1084 zfs_statvfs(struct inode *ip, struct kstatfs *statp)
1086 zfsvfs_t *zfsvfs = ITOZSB(ip);
1087 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1092 dmu_objset_space(zfsvfs->z_os,
1093 &refdbytes, &availbytes, &usedobjs, &availobjs);
1095 uint64_t fsid = dmu_objset_fsid_guid(zfsvfs->z_os);
1097 * The underlying storage pool actually uses multiple block
1098 * size. Under Solaris frsize (fragment size) is reported as
1099 * the smallest block size we support, and bsize (block size)
1100 * as the filesystem's maximum block size. Unfortunately,
1101 * under Linux the fragment size and block size are often used
1102 * interchangeably. Thus we are forced to report both of them
1103 * as the filesystem's maximum block size.
1105 statp->f_frsize = zfsvfs->z_max_blksz;
1106 statp->f_bsize = zfsvfs->z_max_blksz;
1107 uint32_t bshift = fls(statp->f_bsize) - 1;
1110 * The following report "total" blocks of various kinds in
1111 * the file system, but reported in terms of f_bsize - the
1115 /* Round up so we never have a filesystem using 0 blocks. */
1116 refdbytes = P2ROUNDUP(refdbytes, statp->f_bsize);
1117 statp->f_blocks = (refdbytes + availbytes) >> bshift;
1118 statp->f_bfree = availbytes >> bshift;
1119 statp->f_bavail = statp->f_bfree; /* no root reservation */
1122 * statvfs() should really be called statufs(), because it assumes
1123 * static metadata. ZFS doesn't preallocate files, so the best
1124 * we can do is report the max that could possibly fit in f_files,
1125 * and that minus the number actually used in f_ffree.
1126 * For f_ffree, report the smaller of the number of objects available
1127 * and the number of blocks (each object will take at least a block).
1129 statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT);
1130 statp->f_files = statp->f_ffree + usedobjs;
1131 statp->f_fsid.val[0] = (uint32_t)fsid;
1132 statp->f_fsid.val[1] = (uint32_t)(fsid >> 32);
1133 statp->f_type = ZFS_SUPER_MAGIC;
1134 statp->f_namelen = MAXNAMELEN - 1;
1137 * We have all of 40 characters to stuff a string here.
1138 * Is there anything useful we could/should provide?
1140 bzero(statp->f_spare, sizeof (statp->f_spare));
1142 if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
1143 dmu_objset_projectquota_present(zfsvfs->z_os)) {
1144 znode_t *zp = ITOZ(ip);
1146 if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid &&
1147 zpl_is_valid_projid(zp->z_projid))
1148 err = zfs_statfs_project(zfsvfs, zp, statp, bshift);
1156 zfs_root(zfsvfs_t *zfsvfs, struct inode **ipp)
1163 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1165 *ipp = ZTOI(rootzp);
1172 * Linux kernels older than 3.1 do not support a per-filesystem shrinker.
1173 * To accommodate this we must improvise and manually walk the list of znodes
1174 * attempting to prune dentries in order to be able to drop the inodes.
1176 * To avoid scanning the same znodes multiple times they are always rotated
1177 * to the end of the z_all_znodes list. New znodes are inserted at the
1178 * end of the list so we're always scanning the oldest znodes first.
1181 zfs_prune_aliases(zfsvfs_t *zfsvfs, unsigned long nr_to_scan)
1183 znode_t **zp_array, *zp;
1184 int max_array = MIN(nr_to_scan, PAGE_SIZE * 8 / sizeof (znode_t *));
1188 zp_array = kmem_zalloc(max_array * sizeof (znode_t *), KM_SLEEP);
1190 mutex_enter(&zfsvfs->z_znodes_lock);
1191 while ((zp = list_head(&zfsvfs->z_all_znodes)) != NULL) {
1193 if ((i++ > nr_to_scan) || (j >= max_array))
1196 ASSERT(list_link_active(&zp->z_link_node));
1197 list_remove(&zfsvfs->z_all_znodes, zp);
1198 list_insert_tail(&zfsvfs->z_all_znodes, zp);
1200 /* Skip active znodes and .zfs entries */
1201 if (MUTEX_HELD(&zp->z_lock) || zp->z_is_ctldir)
1204 if (igrab(ZTOI(zp)) == NULL)
1210 mutex_exit(&zfsvfs->z_znodes_lock);
1212 for (i = 0; i < j; i++) {
1215 ASSERT3P(zp, !=, NULL);
1216 d_prune_aliases(ZTOI(zp));
1218 if (atomic_read(&ZTOI(zp)->i_count) == 1)
1224 kmem_free(zp_array, max_array * sizeof (znode_t *));
1230 * The ARC has requested that the filesystem drop entries from the dentry
1231 * and inode caches. This can occur when the ARC needs to free meta data
1232 * blocks but can't because they are all pinned by entries in these caches.
1235 zfs_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
1237 zfsvfs_t *zfsvfs = sb->s_fs_info;
1239 struct shrinker *shrinker = &sb->s_shrink;
1240 struct shrink_control sc = {
1241 .nr_to_scan = nr_to_scan,
1242 .gfp_mask = GFP_KERNEL,
1247 #if defined(HAVE_SPLIT_SHRINKER_CALLBACK) && \
1248 defined(SHRINK_CONTROL_HAS_NID) && \
1249 defined(SHRINKER_NUMA_AWARE)
1250 if (sb->s_shrink.flags & SHRINKER_NUMA_AWARE) {
1252 for_each_online_node(sc.nid) {
1253 *objects += (*shrinker->scan_objects)(shrinker, &sc);
1256 *objects = (*shrinker->scan_objects)(shrinker, &sc);
1259 #elif defined(HAVE_SPLIT_SHRINKER_CALLBACK)
1260 *objects = (*shrinker->scan_objects)(shrinker, &sc);
1261 #elif defined(HAVE_SINGLE_SHRINKER_CALLBACK)
1262 *objects = (*shrinker->shrink)(shrinker, &sc);
1263 #elif defined(HAVE_D_PRUNE_ALIASES)
1264 #define D_PRUNE_ALIASES_IS_DEFAULT
1265 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1267 #error "No available dentry and inode cache pruning mechanism."
1270 #if defined(HAVE_D_PRUNE_ALIASES) && !defined(D_PRUNE_ALIASES_IS_DEFAULT)
1271 #undef D_PRUNE_ALIASES_IS_DEFAULT
1273 * Fall back to zfs_prune_aliases if the kernel's per-superblock
1274 * shrinker couldn't free anything, possibly due to the inodes being
1275 * allocated in a different memcg.
1278 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1283 dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
1284 "pruning, nr_to_scan=%lu objects=%d error=%d\n",
1285 nr_to_scan, *objects, error);
1291 * Teardown the zfsvfs_t.
1293 * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
1294 * and 'z_teardown_inactive_lock' held.
1297 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1301 zfs_unlinked_drain_stop_wait(zfsvfs);
1304 * If someone has not already unmounted this file system,
1305 * drain the zrele_taskq to ensure all active references to the
1306 * zfsvfs_t have been handled only then can it be safely destroyed.
1310 * If we're unmounting we have to wait for the list to
1313 * If we're not unmounting there's no guarantee the list
1314 * will drain completely, but iputs run from the taskq
1315 * may add the parents of dir-based xattrs to the taskq
1316 * so we want to wait for these.
1318 * We can safely read z_nr_znodes without locking because the
1319 * VFS has already blocked operations which add to the
1320 * z_all_znodes list and thus increment z_nr_znodes.
1323 while (zfsvfs->z_nr_znodes > 0) {
1324 taskq_wait_outstanding(dsl_pool_zrele_taskq(
1325 dmu_objset_pool(zfsvfs->z_os)), 0);
1326 if (++round > 1 && !unmounting)
1331 ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG);
1335 * We purge the parent filesystem's super block as the
1336 * parent filesystem and all of its snapshots have their
1337 * inode's super block set to the parent's filesystem's
1338 * super block. Note, 'z_parent' is self referential
1339 * for non-snapshots.
1341 shrink_dcache_sb(zfsvfs->z_parent->z_sb);
1345 * Close the zil. NB: Can't close the zil while zfs_inactive
1346 * threads are blocked as zil_close can call zfs_inactive.
1348 if (zfsvfs->z_log) {
1349 zil_close(zfsvfs->z_log);
1350 zfsvfs->z_log = NULL;
1353 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1356 * If we are not unmounting (ie: online recv) and someone already
1357 * unmounted this file system while we were doing the switcheroo,
1358 * or a reopen of z_os failed then just bail out now.
1360 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1361 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1362 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1363 return (SET_ERROR(EIO));
1367 * At this point there are no VFS ops active, and any new VFS ops
1368 * will fail with EIO since we have z_teardown_lock for writer (only
1369 * relevant for forced unmount).
1371 * Release all holds on dbufs. We also grab an extra reference to all
1372 * the remaining inodes so that the kernel does not attempt to free
1373 * any inodes of a suspended fs. This can cause deadlocks since the
1374 * zfs_resume_fs() process may involve starting threads, which might
1375 * attempt to free unreferenced inodes to free up memory for the new
1379 mutex_enter(&zfsvfs->z_znodes_lock);
1380 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1381 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1383 zfs_znode_dmu_fini(zp);
1384 if (igrab(ZTOI(zp)) != NULL)
1385 zp->z_suspended = B_TRUE;
1388 mutex_exit(&zfsvfs->z_znodes_lock);
1392 * If we are unmounting, set the unmounted flag and let new VFS ops
1393 * unblock. zfs_inactive will have the unmounted behavior, and all
1394 * other VFS ops will fail with EIO.
1397 zfsvfs->z_unmounted = B_TRUE;
1398 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1399 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1403 * z_os will be NULL if there was an error in attempting to reopen
1404 * zfsvfs, so just return as the properties had already been
1406 * unregistered and cached data had been evicted before.
1408 if (zfsvfs->z_os == NULL)
1412 * Unregister properties.
1414 zfs_unregister_callbacks(zfsvfs);
1417 * Evict cached data. We must write out any dirty data before
1418 * disowning the dataset.
1420 objset_t *os = zfsvfs->z_os;
1421 boolean_t os_dirty = B_FALSE;
1422 for (int t = 0; t < TXG_SIZE; t++) {
1423 if (dmu_objset_is_dirty(os, t)) {
1428 if (!zfs_is_readonly(zfsvfs) && os_dirty) {
1429 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1431 dmu_objset_evict_dbufs(zfsvfs->z_os);
1432 dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1433 dsl_dir_cancel_waiters(dd);
1438 #if defined(HAVE_SUPER_SETUP_BDI_NAME)
1439 atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0);
1443 zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
1445 const char *osname = zm->mnt_osname;
1446 struct inode *root_inode = NULL;
1447 uint64_t recordsize;
1449 zfsvfs_t *zfsvfs = NULL;
1455 error = zfsvfs_parse_options(zm->mnt_data, &vfs);
1459 error = zfsvfs_create(osname, vfs->vfs_readonly, &zfsvfs);
1461 zfsvfs_vfs_free(vfs);
1465 if ((error = dsl_prop_get_integer(osname, "recordsize",
1466 &recordsize, NULL))) {
1467 zfsvfs_vfs_free(vfs);
1471 vfs->vfs_data = zfsvfs;
1472 zfsvfs->z_vfs = vfs;
1474 sb->s_fs_info = zfsvfs;
1475 sb->s_magic = ZFS_SUPER_MAGIC;
1476 sb->s_maxbytes = MAX_LFS_FILESIZE;
1477 sb->s_time_gran = 1;
1478 sb->s_blocksize = recordsize;
1479 sb->s_blocksize_bits = ilog2(recordsize);
1481 error = -zpl_bdi_setup(sb, "zfs");
1485 sb->s_bdi->ra_pages = 0;
1487 /* Set callback operations for the file system. */
1488 sb->s_op = &zpl_super_operations;
1489 sb->s_xattr = zpl_xattr_handlers;
1490 sb->s_export_op = &zpl_export_operations;
1491 sb->s_d_op = &zpl_dentry_operations;
1493 /* Set features for file system. */
1494 zfs_set_fuid_feature(zfsvfs);
1496 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1499 atime_changed_cb(zfsvfs, B_FALSE);
1500 readonly_changed_cb(zfsvfs, B_TRUE);
1501 if ((error = dsl_prop_get_integer(osname,
1502 "xattr", &pval, NULL)))
1504 xattr_changed_cb(zfsvfs, pval);
1505 if ((error = dsl_prop_get_integer(osname,
1506 "acltype", &pval, NULL)))
1508 acltype_changed_cb(zfsvfs, pval);
1509 zfsvfs->z_issnap = B_TRUE;
1510 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1511 zfsvfs->z_snap_defer_time = jiffies;
1513 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1514 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1515 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1517 if ((error = zfsvfs_setup(zfsvfs, B_TRUE)))
1521 /* Allocate a root inode for the filesystem. */
1522 error = zfs_root(zfsvfs, &root_inode);
1524 (void) zfs_umount(sb);
1528 /* Allocate a root dentry for the filesystem */
1529 sb->s_root = d_make_root(root_inode);
1530 if (sb->s_root == NULL) {
1531 (void) zfs_umount(sb);
1532 error = SET_ERROR(ENOMEM);
1536 if (!zfsvfs->z_issnap)
1537 zfsctl_create(zfsvfs);
1539 zfsvfs->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb);
1542 if (zfsvfs != NULL) {
1543 dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1544 zfsvfs_free(zfsvfs);
1547 * make sure we don't have dangling sb->s_fs_info which
1548 * zfs_preumount will use.
1550 sb->s_fs_info = NULL;
1557 * Called when an unmount is requested and certain sanity checks have
1558 * already passed. At this point no dentries or inodes have been reclaimed
1559 * from their respective caches. We drop the extra reference on the .zfs
1560 * control directory to allow everything to be reclaimed. All snapshots
1561 * must already have been unmounted to reach this point.
1564 zfs_preumount(struct super_block *sb)
1566 zfsvfs_t *zfsvfs = sb->s_fs_info;
1568 /* zfsvfs is NULL when zfs_domount fails during mount */
1570 zfs_unlinked_drain_stop_wait(zfsvfs);
1571 zfsctl_destroy(sb->s_fs_info);
1573 * Wait for zrele_async before entering evict_inodes in
1574 * generic_shutdown_super. The reason we must finish before
1575 * evict_inodes is when lazytime is on, or when zfs_purgedir
1576 * calls zfs_zget, zrele would bump i_count from 0 to 1. This
1577 * would race with the i_count check in evict_inodes. This means
1578 * it could destroy the inode while we are still using it.
1580 * We wait for two passes. xattr directories in the first pass
1581 * may add xattr entries in zfs_purgedir, so in the second pass
1582 * we wait for them. We don't use taskq_wait here because it is
1583 * a pool wide taskq. Other mounted filesystems can constantly
1584 * do zrele_async and there's no guarantee when taskq will be
1587 taskq_wait_outstanding(dsl_pool_zrele_taskq(
1588 dmu_objset_pool(zfsvfs->z_os)), 0);
1589 taskq_wait_outstanding(dsl_pool_zrele_taskq(
1590 dmu_objset_pool(zfsvfs->z_os)), 0);
1595 * Called once all other unmount released tear down has occurred.
1596 * It is our responsibility to release any remaining infrastructure.
1600 zfs_umount(struct super_block *sb)
1602 zfsvfs_t *zfsvfs = sb->s_fs_info;
1605 if (zfsvfs->z_arc_prune != NULL)
1606 arc_remove_prune_callback(zfsvfs->z_arc_prune);
1607 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1609 zpl_bdi_destroy(sb);
1612 * z_os will be NULL if there was an error in
1613 * attempting to reopen zfsvfs.
1617 * Unset the objset user_ptr.
1619 mutex_enter(&os->os_user_ptr_lock);
1620 dmu_objset_set_user(os, NULL);
1621 mutex_exit(&os->os_user_ptr_lock);
1624 * Finally release the objset
1626 dmu_objset_disown(os, B_TRUE, zfsvfs);
1629 zfsvfs_free(zfsvfs);
1634 zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm)
1636 zfsvfs_t *zfsvfs = sb->s_fs_info;
1638 boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os);
1641 if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) &&
1642 !(*flags & SB_RDONLY)) {
1643 *flags |= SB_RDONLY;
1647 error = zfsvfs_parse_options(zm->mnt_data, &vfsp);
1651 if (!zfs_is_readonly(zfsvfs) && (*flags & SB_RDONLY))
1652 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1654 zfs_unregister_callbacks(zfsvfs);
1655 zfsvfs_vfs_free(zfsvfs->z_vfs);
1657 vfsp->vfs_data = zfsvfs;
1658 zfsvfs->z_vfs = vfsp;
1660 (void) zfs_register_callbacks(vfsp);
1666 zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp)
1668 zfsvfs_t *zfsvfs = sb->s_fs_info;
1670 uint64_t object = 0;
1671 uint64_t fid_gen = 0;
1678 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1679 zfid_short_t *zfid = (zfid_short_t *)fidp;
1681 for (i = 0; i < sizeof (zfid->zf_object); i++)
1682 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1684 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1685 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1687 return (SET_ERROR(EINVAL));
1690 /* LONG_FID_LEN means snapdirs */
1691 if (fidp->fid_len == LONG_FID_LEN) {
1692 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1693 uint64_t objsetid = 0;
1694 uint64_t setgen = 0;
1696 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1697 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1699 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1700 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1702 if (objsetid != ZFSCTL_INO_SNAPDIRS - object) {
1703 dprintf("snapdir fid: objsetid (%llu) != "
1704 "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n",
1705 objsetid, ZFSCTL_INO_SNAPDIRS, object);
1707 return (SET_ERROR(EINVAL));
1710 if (fid_gen > 1 || setgen != 0) {
1711 dprintf("snapdir fid: fid_gen (%llu) and setgen "
1712 "(%llu)\n", fid_gen, setgen);
1713 return (SET_ERROR(EINVAL));
1716 return (zfsctl_snapdir_vget(sb, objsetid, fid_gen, ipp));
1720 /* A zero fid_gen means we are in the .zfs control directories */
1722 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1723 *ipp = zfsvfs->z_ctldir;
1724 ASSERT(*ipp != NULL);
1725 if (object == ZFSCTL_INO_SNAPDIR) {
1726 VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp,
1727 0, kcred, NULL, NULL) == 0);
1730 * Must have an existing ref, so igrab()
1731 * cannot return NULL
1733 VERIFY3P(igrab(*ipp), !=, NULL);
1739 gen_mask = -1ULL >> (64 - 8 * i);
1741 dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask);
1742 if ((err = zfs_zget(zfsvfs, object, &zp))) {
1747 /* Don't export xattr stuff */
1748 if (zp->z_pflags & ZFS_XATTR) {
1751 return (SET_ERROR(ENOENT));
1754 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1756 zp_gen = zp_gen & gen_mask;
1759 if ((fid_gen == 0) && (zfsvfs->z_root == object))
1761 if (zp->z_unlinked || zp_gen != fid_gen) {
1762 dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen,
1766 return (SET_ERROR(ENOENT));
1771 zfs_znode_update_vfs(ITOZ(*ipp));
1778 * Block out VFS ops and close zfsvfs_t
1780 * Note, if successful, then we return with the 'z_teardown_lock' and
1781 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
1782 * dataset and objset intact so that they can be atomically handed off during
1783 * a subsequent rollback or recv operation and the resume thereafter.
1786 zfs_suspend_fs(zfsvfs_t *zfsvfs)
1790 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
1797 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
1798 * is an invariant across any of the operations that can be performed while the
1799 * filesystem was suspended. Whether it succeeded or failed, the preconditions
1800 * are the same: the relevant objset and associated dataset are owned by
1801 * zfsvfs, held, and long held on entry.
1804 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1809 ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
1810 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1813 * We already own this, so just update the objset_t, as the one we
1814 * had before may have been evicted.
1817 VERIFY3P(ds->ds_owner, ==, zfsvfs);
1818 VERIFY(dsl_dataset_long_held(ds));
1819 dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
1820 dsl_pool_config_enter(dp, FTAG);
1821 VERIFY0(dmu_objset_from_ds(ds, &os));
1822 dsl_pool_config_exit(dp, FTAG);
1824 err = zfsvfs_init(zfsvfs, os);
1828 ds->ds_dir->dd_activity_cancelled = B_FALSE;
1829 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
1831 zfs_set_fuid_feature(zfsvfs);
1832 zfsvfs->z_rollback_time = jiffies;
1835 * Attempt to re-establish all the active inodes with their
1836 * dbufs. If a zfs_rezget() fails, then we unhash the inode
1837 * and mark it stale. This prevents a collision if a new
1838 * inode/object is created which must use the same inode
1839 * number. The stale inode will be be released when the
1840 * VFS prunes the dentry holding the remaining references
1841 * on the stale inode.
1843 mutex_enter(&zfsvfs->z_znodes_lock);
1844 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
1845 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1846 err2 = zfs_rezget(zp);
1848 remove_inode_hash(ZTOI(zp));
1849 zp->z_is_stale = B_TRUE;
1852 /* see comment in zfs_suspend_fs() */
1853 if (zp->z_suspended) {
1854 zfs_zrele_async(zp);
1855 zp->z_suspended = B_FALSE;
1858 mutex_exit(&zfsvfs->z_znodes_lock);
1860 if (!zfs_is_readonly(zfsvfs) && !zfsvfs->z_unmounted) {
1862 * zfs_suspend_fs() could have interrupted freeing
1863 * of dnodes. We need to restart this freeing so
1864 * that we don't "leak" the space.
1866 zfs_unlinked_drain(zfsvfs);
1870 * Most of the time zfs_suspend_fs is used for changing the contents
1871 * of the underlying dataset. ZFS rollback and receive operations
1872 * might create files for which negative dentries are present in
1873 * the cache. Since walking the dcache would require a lot of GPL-only
1874 * code duplication, it's much easier on these rather rare occasions
1875 * just to flush the whole dcache for the given dataset/filesystem.
1877 shrink_dcache_sb(zfsvfs->z_sb);
1881 zfsvfs->z_unmounted = B_TRUE;
1883 /* release the VFS ops */
1884 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1885 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1889 * Since we couldn't setup the sa framework, try to force
1890 * unmount this file system.
1893 (void) zfs_umount(zfsvfs->z_sb);
1899 * Release VOPs and unmount a suspended filesystem.
1902 zfs_end_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1904 ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
1905 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1908 * We already own this, so just hold and rele it to update the
1909 * objset_t, as the one we had before may have been evicted.
1912 VERIFY3P(ds->ds_owner, ==, zfsvfs);
1913 VERIFY(dsl_dataset_long_held(ds));
1914 dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
1915 dsl_pool_config_enter(dp, FTAG);
1916 VERIFY0(dmu_objset_from_ds(ds, &os));
1917 dsl_pool_config_exit(dp, FTAG);
1920 /* release the VOPs */
1921 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1922 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1925 * Try to force unmount this file system.
1927 (void) zfs_umount(zfsvfs->z_sb);
1928 zfsvfs->z_unmounted = B_TRUE;
1933 * Automounted snapshots rely on periodic revalidation
1934 * to defer snapshots from being automatically unmounted.
1938 zfs_exit_fs(zfsvfs_t *zfsvfs)
1940 if (!zfsvfs->z_issnap)
1943 if (time_after(jiffies, zfsvfs->z_snap_defer_time +
1944 MAX(zfs_expire_snapshot * HZ / 2, HZ))) {
1945 zfsvfs->z_snap_defer_time = jiffies;
1946 zfsctl_snapshot_unmount_delay(zfsvfs->z_os->os_spa,
1947 dmu_objset_id(zfsvfs->z_os),
1948 zfs_expire_snapshot);
1953 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
1956 objset_t *os = zfsvfs->z_os;
1959 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
1960 return (SET_ERROR(EINVAL));
1962 if (newvers < zfsvfs->z_version)
1963 return (SET_ERROR(EINVAL));
1965 if (zfs_spa_version_map(newvers) >
1966 spa_version(dmu_objset_spa(zfsvfs->z_os)))
1967 return (SET_ERROR(ENOTSUP));
1969 tx = dmu_tx_create(os);
1970 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
1971 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1972 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
1974 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1976 error = dmu_tx_assign(tx, TXG_WAIT);
1982 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
1983 8, 1, &newvers, tx);
1990 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1993 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
1995 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1996 DMU_OT_NONE, 0, tx);
1998 error = zap_add(os, MASTER_NODE_OBJ,
1999 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2002 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2003 sa_register_update_callback(os, zfs_sa_upgrade);
2006 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2007 "from %llu to %llu", zfsvfs->z_version, newvers);
2011 zfsvfs->z_version = newvers;
2012 os->os_version = newvers;
2014 zfs_set_fuid_feature(zfsvfs);
2020 * Read a property stored within the master node.
2023 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2025 uint64_t *cached_copy = NULL;
2028 * Figure out where in the objset_t the cached copy would live, if it
2029 * is available for the requested property.
2033 case ZFS_PROP_VERSION:
2034 cached_copy = &os->os_version;
2036 case ZFS_PROP_NORMALIZE:
2037 cached_copy = &os->os_normalization;
2039 case ZFS_PROP_UTF8ONLY:
2040 cached_copy = &os->os_utf8only;
2043 cached_copy = &os->os_casesensitivity;
2049 if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) {
2050 *value = *cached_copy;
2055 * If the property wasn't cached, look up the file system's value for
2056 * the property. For the version property, we look up a slightly
2061 if (prop == ZFS_PROP_VERSION)
2062 pname = ZPL_VERSION_STR;
2064 pname = zfs_prop_to_name(prop);
2067 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2068 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2071 if (error == ENOENT) {
2072 /* No value set, use the default value */
2074 case ZFS_PROP_VERSION:
2075 *value = ZPL_VERSION;
2077 case ZFS_PROP_NORMALIZE:
2078 case ZFS_PROP_UTF8ONLY:
2082 *value = ZFS_CASE_SENSITIVE;
2084 case ZFS_PROP_ACLTYPE:
2085 *value = ZFS_ACLTYPE_OFF;
2094 * If one of the methods for getting the property value above worked,
2095 * copy it into the objset_t's cache.
2097 if (error == 0 && cached_copy != NULL) {
2098 *cached_copy = *value;
2105 * Return true if the corresponding vfs's unmounted flag is set.
2106 * Otherwise return false.
2107 * If this function returns true we know VFS unmount has been initiated.
2110 zfs_get_vfs_flag_unmounted(objset_t *os)
2113 boolean_t unmounted = B_FALSE;
2115 ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2117 mutex_enter(&os->os_user_ptr_lock);
2118 zfvp = dmu_objset_get_user(os);
2119 if (zfvp != NULL && zfvp->z_unmounted)
2121 mutex_exit(&os->os_user_ptr_lock);
2128 zfsvfs_update_fromname(const char *oldname, const char *newname)
2131 * We don't need to do anything here, the devname is always current by
2132 * virtue of zfsvfs->z_sb->s_op->show_devname.
2141 dmu_objset_register_type(DMU_OST_ZFS, zpl_get_file_info);
2142 register_filesystem(&zpl_fs_type);
2149 * we don't use outstanding because zpl_posix_acl_free might add more.
2151 taskq_wait(system_delay_taskq);
2152 taskq_wait(system_taskq);
2153 unregister_filesystem(&zpl_fs_type);
2158 #if defined(_KERNEL)
2159 EXPORT_SYMBOL(zfs_suspend_fs);
2160 EXPORT_SYMBOL(zfs_resume_fs);
2161 EXPORT_SYMBOL(zfs_set_version);
2162 EXPORT_SYMBOL(zfsvfs_create);
2163 EXPORT_SYMBOL(zfsvfs_free);
2164 EXPORT_SYMBOL(zfs_is_readonly);
2165 EXPORT_SYMBOL(zfs_domount);
2166 EXPORT_SYMBOL(zfs_preumount);
2167 EXPORT_SYMBOL(zfs_umount);
2168 EXPORT_SYMBOL(zfs_remount);
2169 EXPORT_SYMBOL(zfs_statvfs);
2170 EXPORT_SYMBOL(zfs_vget);
2171 EXPORT_SYMBOL(zfs_prune);