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.
25 /* Portions Copyright 2010 Robert Milkowski */
27 #include <sys/types.h>
28 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/sysmacros.h>
32 #include <sys/pathname.h>
33 #include <sys/vnode.h>
35 #include <sys/vfs_opreg.h>
36 #include <sys/mntent.h>
37 #include <sys/mount.h>
38 #include <sys/cmn_err.h>
39 #include "fs/fs_subr.h"
40 #include <sys/zfs_znode.h>
41 #include <sys/zfs_vnops.h>
42 #include <sys/zfs_dir.h>
44 #include <sys/fs/zfs.h>
46 #include <sys/dsl_prop.h>
47 #include <sys/dsl_dataset.h>
48 #include <sys/dsl_deleg.h>
52 #include <sys/varargs.h>
53 #include <sys/policy.h>
54 #include <sys/atomic.h>
55 #include <sys/mkdev.h>
56 #include <sys/modctl.h>
57 #include <sys/refstr.h>
58 #include <sys/zfs_ioctl.h>
59 #include <sys/zfs_fuid.h>
60 #include <sys/bootconf.h>
61 #include <sys/sunddi.h>
63 #include <sys/dmu_objset.h>
64 #include <sys/spa_boot.h>
67 #include "zfs_comutil.h"
72 zfs_sync(struct super_block *sb, int wait, cred_t *cr)
74 zfs_sb_t *zsb = sb->s_fs_info;
77 * Data integrity is job one. We don't want a compromised kernel
78 * writing to the storage pool, so we never sync during panic.
80 if (unlikely(oops_in_progress))
84 * Semantically, the only requirement is that the sync be initiated.
85 * The DMU syncs out txgs frequently, so there's nothing to do.
92 * Sync a specific filesystem.
97 dp = dmu_objset_pool(zsb->z_os);
100 * If the system is shutting down, then skip any
101 * filesystems which may exist on a suspended pool.
103 if (spa_suspended(dp->dp_spa)) {
108 if (zsb->z_log != NULL)
109 zil_commit(zsb->z_log, 0);
114 * Sync all ZFS filesystems. This is what happens when you
115 * run sync(1M). Unlike other filesystems, ZFS honors the
116 * request by waiting for all pools to commit all dirty data.
123 EXPORT_SYMBOL(zfs_sync);
126 atime_changed_cb(void *arg, uint64_t newval)
129 struct super_block *sb = zsb->z_sb;
130 struct vfsmount *vfs = zsb->z_vfs;
132 if (newval == TRUE) {
133 vfs->mnt_flags &= ~MNT_NOATIME;
134 sb->s_flags &= ~MS_NOATIME;
137 vfs->mnt_flags |= MNT_NOATIME;
138 sb->s_flags |= MS_NOATIME;
139 zsb->z_atime = FALSE;
144 xattr_changed_cb(void *arg, uint64_t newval)
148 if (newval == TRUE) {
149 zsb->z_flags |= ZSB_XATTR_USER;
151 zsb->z_flags &= ~ZSB_XATTR_USER;
156 blksz_changed_cb(void *arg, uint64_t newval)
160 if (newval < SPA_MINBLOCKSIZE ||
161 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
162 newval = SPA_MAXBLOCKSIZE;
164 zsb->z_max_blksz = newval;
168 readonly_changed_cb(void *arg, uint64_t newval)
171 struct super_block *sb = zsb->z_sb;
172 struct vfsmount *vfs = zsb->z_vfs;
175 vfs->mnt_flags |= MNT_READONLY;
176 sb->s_flags |= MS_RDONLY;
178 vfs->mnt_flags &= ~MNT_READONLY;
179 sb->s_flags &= ~MS_RDONLY;
184 devices_changed_cb(void *arg, uint64_t newval)
187 struct super_block *sb = zsb->z_sb;
188 struct vfsmount *vfs = zsb->z_vfs;
190 if (newval == FALSE) {
191 vfs->mnt_flags |= MNT_NODEV;
192 sb->s_flags |= MS_NODEV;
194 vfs->mnt_flags &= ~MNT_NODEV;
195 sb->s_flags &= ~MS_NODEV;
200 setuid_changed_cb(void *arg, uint64_t newval)
203 struct super_block *sb = zsb->z_sb;
204 struct vfsmount *vfs = zsb->z_vfs;
206 if (newval == FALSE) {
207 vfs->mnt_flags |= MNT_NOSUID;
208 sb->s_flags |= MS_NOSUID;
210 vfs->mnt_flags &= ~MNT_NOSUID;
211 sb->s_flags &= ~MS_NOSUID;
216 exec_changed_cb(void *arg, uint64_t newval)
219 struct super_block *sb = zsb->z_sb;
220 struct vfsmount *vfs = zsb->z_vfs;
222 if (newval == FALSE) {
223 vfs->mnt_flags |= MNT_NOEXEC;
224 sb->s_flags |= MS_NOEXEC;
226 vfs->mnt_flags &= ~MNT_NOEXEC;
227 sb->s_flags &= ~MS_NOEXEC;
232 * The nbmand mount option can be changed at mount time.
233 * We can't allow it to be toggled on live file systems or incorrect
234 * behavior may be seen from cifs clients
236 * This property isn't registered via dsl_prop_register(), but this callback
237 * will be called when a file system is first mounted
240 nbmand_changed_cb(void *arg, uint64_t newval)
243 struct super_block *sb = zsb->z_sb;
245 if (newval == TRUE) {
246 sb->s_flags |= MS_MANDLOCK;
248 sb->s_flags &= ~MS_MANDLOCK;
253 snapdir_changed_cb(void *arg, uint64_t newval)
255 ((zfs_sb_t *)arg)->z_show_ctldir = newval;
259 vscan_changed_cb(void *arg, uint64_t newval)
261 ((zfs_sb_t *)arg)->z_vscan = newval;
265 acl_inherit_changed_cb(void *arg, uint64_t newval)
267 ((zfs_sb_t *)arg)->z_acl_inherit = newval;
271 zfs_register_callbacks(zfs_sb_t *zsb)
273 struct vfsmount *vfsp = zsb->z_vfs;
274 struct dsl_dataset *ds = NULL;
275 objset_t *os = zsb->z_os;
277 boolean_t readonly = B_FALSE;
278 boolean_t setuid = B_TRUE;
279 boolean_t exec = B_TRUE;
280 boolean_t devices = B_TRUE;
281 boolean_t xattr = B_TRUE;
282 boolean_t atime = B_TRUE;
283 char osname[MAXNAMELEN];
287 * While Linux allows multiple vfs mounts per super block we have
288 * limited it artificially to one in zfs_fill_super. Thus it is
289 * safe for us to modify the vfs mount fails through the callbacks.
291 if ((vfsp->mnt_flags & MNT_READONLY) ||
292 !spa_writeable(dmu_objset_spa(os)))
295 if (vfsp->mnt_flags & MNT_NOSUID) {
299 if (vfsp->mnt_flags & MNT_NODEV)
303 if (vfsp->mnt_flags & MNT_NOEXEC)
306 if (vfsp->mnt_flags & MNT_NOATIME)
310 * nbmand is a special property which may only be changed at
311 * mount time. Unfortunately, Linux does not have a VFS mount
312 * flag instead this is a super block flag. So setting this
313 * option at mount time will have to wait until we can parse
314 * the mount option string. For now we rely on the nbmand
315 * value stored with the object set. Additional mount option
316 * string to be handled:
318 * case: sensitive|insensitive|mixed
322 dmu_objset_name(os, osname);
323 if ((error = dsl_prop_get_integer(osname, "nbmand", &nbmand, NULL)))
327 * Register property callbacks.
329 * It would probably be fine to just check for i/o error from
330 * the first prop_register(), but I guess I like to go
333 ds = dmu_objset_ds(os);
334 error = dsl_prop_register(ds,
335 "atime", atime_changed_cb, zsb);
336 error = error ? error : dsl_prop_register(ds,
337 "xattr", xattr_changed_cb, zsb);
338 error = error ? error : dsl_prop_register(ds,
339 "recordsize", blksz_changed_cb, zsb);
340 error = error ? error : dsl_prop_register(ds,
341 "readonly", readonly_changed_cb, zsb);
342 error = error ? error : dsl_prop_register(ds,
343 "devices", devices_changed_cb, zsb);
344 error = error ? error : dsl_prop_register(ds,
345 "setuid", setuid_changed_cb, zsb);
346 error = error ? error : dsl_prop_register(ds,
347 "exec", exec_changed_cb, zsb);
348 error = error ? error : dsl_prop_register(ds,
349 "snapdir", snapdir_changed_cb, zsb);
350 error = error ? error : dsl_prop_register(ds,
351 "aclinherit", acl_inherit_changed_cb, zsb);
352 error = error ? error : dsl_prop_register(ds,
353 "vscan", vscan_changed_cb, zsb);
358 * Invoke our callbacks to set required flags.
360 readonly_changed_cb(zsb, readonly);
361 setuid_changed_cb(zsb, setuid);
362 exec_changed_cb(zsb, exec);
363 devices_changed_cb(zsb, devices);
364 xattr_changed_cb(zsb, xattr);
365 atime_changed_cb(zsb, atime);
366 nbmand_changed_cb(zsb, nbmand);
372 * We may attempt to unregister some callbacks that are not
373 * registered, but this is OK; it will simply return ENOMSG,
374 * which we will ignore.
376 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zsb);
377 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zsb);
378 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zsb);
379 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zsb);
380 (void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zsb);
381 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zsb);
382 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zsb);
383 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zsb);
384 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
386 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zsb);
390 EXPORT_SYMBOL(zfs_register_callbacks);
393 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
394 uint64_t *userp, uint64_t *groupp)
396 znode_phys_t *znp = data;
400 * Is it a valid type of object to track?
402 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
406 * If we have a NULL data pointer
407 * then assume the id's aren't changing and
408 * return EEXIST to the dmu to let it know to
414 if (bonustype == DMU_OT_ZNODE) {
415 *userp = znp->zp_uid;
416 *groupp = znp->zp_gid;
420 ASSERT(bonustype == DMU_OT_SA);
421 hdrsize = sa_hdrsize(data);
424 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
426 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
430 * This should only happen for newly created
431 * files that haven't had the znode data filled
442 fuidstr_to_sid(zfs_sb_t *zsb, const char *fuidstr,
443 char *domainbuf, int buflen, uid_t *ridp)
448 fuid = strtonum(fuidstr, NULL);
450 domain = zfs_fuid_find_by_idx(zsb, FUID_INDEX(fuid));
452 (void) strlcpy(domainbuf, domain, buflen);
455 *ridp = FUID_RID(fuid);
459 zfs_userquota_prop_to_obj(zfs_sb_t *zsb, zfs_userquota_prop_t type)
462 case ZFS_PROP_USERUSED:
463 return (DMU_USERUSED_OBJECT);
464 case ZFS_PROP_GROUPUSED:
465 return (DMU_GROUPUSED_OBJECT);
466 case ZFS_PROP_USERQUOTA:
467 return (zsb->z_userquota_obj);
468 case ZFS_PROP_GROUPQUOTA:
469 return (zsb->z_groupquota_obj);
477 zfs_userspace_many(zfs_sb_t *zsb, zfs_userquota_prop_t type,
478 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
483 zfs_useracct_t *buf = vbuf;
486 if (!dmu_objset_userspace_present(zsb->z_os))
489 obj = zfs_userquota_prop_to_obj(zsb, type);
495 for (zap_cursor_init_serialized(&zc, zsb->z_os, obj, *cookiep);
496 (error = zap_cursor_retrieve(&zc, &za)) == 0;
497 zap_cursor_advance(&zc)) {
498 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
502 fuidstr_to_sid(zsb, za.za_name,
503 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
505 buf->zu_space = za.za_first_integer;
511 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
512 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
513 *cookiep = zap_cursor_serialize(&zc);
514 zap_cursor_fini(&zc);
517 EXPORT_SYMBOL(zfs_userspace_many);
520 * buf must be big enough (eg, 32 bytes)
523 id_to_fuidstr(zfs_sb_t *zsb, const char *domain, uid_t rid,
524 char *buf, boolean_t addok)
529 if (domain && domain[0]) {
530 domainid = zfs_fuid_find_by_domain(zsb, domain, NULL, addok);
534 fuid = FUID_ENCODE(domainid, rid);
535 (void) sprintf(buf, "%llx", (longlong_t)fuid);
540 zfs_userspace_one(zfs_sb_t *zsb, zfs_userquota_prop_t type,
541 const char *domain, uint64_t rid, uint64_t *valp)
549 if (!dmu_objset_userspace_present(zsb->z_os))
552 obj = zfs_userquota_prop_to_obj(zsb, type);
556 err = id_to_fuidstr(zsb, domain, rid, buf, B_FALSE);
560 err = zap_lookup(zsb->z_os, obj, buf, 8, 1, valp);
565 EXPORT_SYMBOL(zfs_userspace_one);
568 zfs_set_userquota(zfs_sb_t *zsb, zfs_userquota_prop_t type,
569 const char *domain, uint64_t rid, uint64_t quota)
575 boolean_t fuid_dirtied;
577 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
580 if (zsb->z_version < ZPL_VERSION_USERSPACE)
583 objp = (type == ZFS_PROP_USERQUOTA) ? &zsb->z_userquota_obj :
584 &zsb->z_groupquota_obj;
586 err = id_to_fuidstr(zsb, domain, rid, buf, B_TRUE);
589 fuid_dirtied = zsb->z_fuid_dirty;
591 tx = dmu_tx_create(zsb->z_os);
592 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
594 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
595 zfs_userquota_prop_prefixes[type]);
598 zfs_fuid_txhold(zsb, tx);
599 err = dmu_tx_assign(tx, TXG_WAIT);
605 mutex_enter(&zsb->z_lock);
607 *objp = zap_create(zsb->z_os, DMU_OT_USERGROUP_QUOTA,
609 VERIFY(0 == zap_add(zsb->z_os, MASTER_NODE_OBJ,
610 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
612 mutex_exit(&zsb->z_lock);
615 err = zap_remove(zsb->z_os, *objp, buf, tx);
619 err = zap_update(zsb->z_os, *objp, buf, 8, 1, "a, tx);
623 zfs_fuid_sync(zsb, tx);
627 EXPORT_SYMBOL(zfs_set_userquota);
630 zfs_fuid_overquota(zfs_sb_t *zsb, boolean_t isgroup, uint64_t fuid)
633 uint64_t used, quota, usedobj, quotaobj;
636 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
637 quotaobj = isgroup ? zsb->z_groupquota_obj : zsb->z_userquota_obj;
639 if (quotaobj == 0 || zsb->z_replay)
642 (void) sprintf(buf, "%llx", (longlong_t)fuid);
643 err = zap_lookup(zsb->z_os, quotaobj, buf, 8, 1, "a);
647 err = zap_lookup(zsb->z_os, usedobj, buf, 8, 1, &used);
650 return (used >= quota);
652 EXPORT_SYMBOL(zfs_fuid_overquota);
655 zfs_owner_overquota(zfs_sb_t *zsb, znode_t *zp, boolean_t isgroup)
660 quotaobj = isgroup ? zsb->z_groupquota_obj : zsb->z_userquota_obj;
662 fuid = isgroup ? zp->z_gid : zp->z_uid;
664 if (quotaobj == 0 || zsb->z_replay)
667 return (zfs_fuid_overquota(zsb, isgroup, fuid));
669 EXPORT_SYMBOL(zfs_owner_overquota);
672 zfs_sb_create(const char *osname, zfs_sb_t **zsbp)
680 zsb = kmem_zalloc(sizeof (zfs_sb_t), KM_SLEEP);
683 * We claim to always be readonly so we can open snapshots;
684 * other ZPL code will prevent us from writing to snapshots.
686 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zsb, &os);
688 kmem_free(zsb, sizeof (zfs_sb_t));
693 * Initialize the zfs-specific filesystem structure.
694 * Should probably make this a kmem cache, shuffle fields,
695 * and just bzero up to z_hold_mtx[].
699 zsb->z_max_blksz = SPA_MAXBLOCKSIZE;
700 zsb->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
703 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zsb->z_version);
706 } else if (zsb->z_version >
707 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
708 (void) printk("Can't mount a version %lld file system "
709 "on a version %lld pool\n. Pool must be upgraded to mount "
710 "this file system.", (u_longlong_t)zsb->z_version,
711 (u_longlong_t)spa_version(dmu_objset_spa(os)));
715 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
717 zsb->z_norm = (int)zval;
719 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
721 zsb->z_utf8 = (zval != 0);
723 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
725 zsb->z_case = (uint_t)zval;
728 * Fold case on file systems that are always or sometimes case
731 if (zsb->z_case == ZFS_CASE_INSENSITIVE ||
732 zsb->z_case == ZFS_CASE_MIXED)
733 zsb->z_norm |= U8_TEXTPREP_TOUPPER;
735 zsb->z_use_fuids = USE_FUIDS(zsb->z_version, zsb->z_os);
736 zsb->z_use_sa = USE_SA(zsb->z_version, zsb->z_os);
739 /* should either have both of these objects or none */
740 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
746 * Pre SA versions file systems should never touch
747 * either the attribute registration or layout objects.
752 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
757 if (zsb->z_version >= ZPL_VERSION_SA)
758 sa_register_update_callback(os, zfs_sa_upgrade);
760 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
764 ASSERT(zsb->z_root != 0);
766 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
767 &zsb->z_unlinkedobj);
771 error = zap_lookup(os, MASTER_NODE_OBJ,
772 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
773 8, 1, &zsb->z_userquota_obj);
774 if (error && error != ENOENT)
777 error = zap_lookup(os, MASTER_NODE_OBJ,
778 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
779 8, 1, &zsb->z_groupquota_obj);
780 if (error && error != ENOENT)
783 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
785 if (error && error != ENOENT)
788 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
790 if (error && error != ENOENT)
793 mutex_init(&zsb->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
794 mutex_init(&zsb->z_lock, NULL, MUTEX_DEFAULT, NULL);
795 list_create(&zsb->z_all_znodes, sizeof (znode_t),
796 offsetof(znode_t, z_link_node));
797 rrw_init(&zsb->z_teardown_lock);
798 rw_init(&zsb->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
799 rw_init(&zsb->z_fuid_lock, NULL, RW_DEFAULT, NULL);
800 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
801 mutex_init(&zsb->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
807 dmu_objset_disown(os, zsb);
809 kmem_free(zsb, sizeof (zfs_sb_t));
814 zfs_sb_setup(zfs_sb_t *zsb, boolean_t mounting)
818 error = zfs_register_callbacks(zsb);
823 * Set the objset user_ptr to track its zsb.
825 mutex_enter(&zsb->z_os->os_user_ptr_lock);
826 dmu_objset_set_user(zsb->z_os, zsb);
827 mutex_exit(&zsb->z_os->os_user_ptr_lock);
829 zsb->z_log = zil_open(zsb->z_os, zfs_get_data);
832 * If we are not mounting (ie: online recv), then we don't
833 * have to worry about replaying the log as we blocked all
834 * operations out since we closed the ZIL.
840 * During replay we remove the read only flag to
841 * allow replays to succeed.
843 readonly = zsb->z_vfs->mnt_flags & MNT_READONLY;
845 zsb->z_vfs->mnt_flags &= ~MNT_READONLY;
847 zfs_unlinked_drain(zsb);
850 * Parse and replay the intent log.
852 * Because of ziltest, this must be done after
853 * zfs_unlinked_drain(). (Further note: ziltest
854 * doesn't use readonly mounts, where
855 * zfs_unlinked_drain() isn't called.) This is because
856 * ziltest causes spa_sync() to think it's committed,
857 * but actually it is not, so the intent log contains
858 * many txg's worth of changes.
860 * In particular, if object N is in the unlinked set in
861 * the last txg to actually sync, then it could be
862 * actually freed in a later txg and then reallocated
863 * in a yet later txg. This would write a "create
864 * object N" record to the intent log. Normally, this
865 * would be fine because the spa_sync() would have
866 * written out the fact that object N is free, before
867 * we could write the "create object N" intent log
870 * But when we are in ziltest mode, we advance the "open
871 * txg" without actually spa_sync()-ing the changes to
872 * disk. So we would see that object N is still
873 * allocated and in the unlinked set, and there is an
874 * intent log record saying to allocate it.
876 if (spa_writeable(dmu_objset_spa(zsb->z_os))) {
877 if (zil_replay_disable) {
878 zil_destroy(zsb->z_log, B_FALSE);
880 zsb->z_replay = B_TRUE;
881 zil_replay(zsb->z_os, zsb,
883 zsb->z_replay = B_FALSE;
886 zsb->z_vfs->mnt_flags |= readonly; /* restore readonly bit */
893 zfs_sb_free(zfs_sb_t *zsb)
897 zfs_fuid_destroy(zsb);
899 mutex_destroy(&zsb->z_znodes_lock);
900 mutex_destroy(&zsb->z_lock);
901 list_destroy(&zsb->z_all_znodes);
902 rrw_destroy(&zsb->z_teardown_lock);
903 rw_destroy(&zsb->z_teardown_inactive_lock);
904 rw_destroy(&zsb->z_fuid_lock);
905 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
906 mutex_destroy(&zsb->z_hold_mtx[i]);
907 kmem_free(zsb, sizeof (zfs_sb_t));
911 zfs_set_fuid_feature(zfs_sb_t *zsb)
913 zsb->z_use_fuids = USE_FUIDS(zsb->z_version, zsb->z_os);
914 zsb->z_use_sa = USE_SA(zsb->z_version, zsb->z_os);
918 zfs_unregister_callbacks(zfs_sb_t *zsb)
920 objset_t *os = zsb->z_os;
921 struct dsl_dataset *ds;
924 * Unregister properties.
926 if (!dmu_objset_is_snapshot(os)) {
927 ds = dmu_objset_ds(os);
928 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
931 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
934 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
937 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
940 VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
943 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
946 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
949 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
952 VERIFY(dsl_prop_unregister(ds, "aclinherit",
953 acl_inherit_changed_cb, zsb) == 0);
955 VERIFY(dsl_prop_unregister(ds, "vscan",
956 vscan_changed_cb, zsb) == 0);
959 EXPORT_SYMBOL(zfs_unregister_callbacks);
963 * zfs_check_global_label:
964 * Check that the hex label string is appropriate for the dataset
965 * being mounted into the global_zone proper.
967 * Return an error if the hex label string is not default or
968 * admin_low/admin_high. For admin_low labels, the corresponding
969 * dataset must be readonly.
972 zfs_check_global_label(const char *dsname, const char *hexsl)
974 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
976 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
978 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
979 /* must be readonly */
982 if (dsl_prop_get_integer(dsname,
983 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
985 return (rdonly ? 0 : EACCES);
989 #endif /* HAVE_MLSLABEL */
992 zfs_statvfs(struct dentry *dentry, struct kstatfs *statp)
994 zfs_sb_t *zsb = dentry->d_sb->s_fs_info;
995 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1000 dmu_objset_space(zsb->z_os,
1001 &refdbytes, &availbytes, &usedobjs, &availobjs);
1004 * The underlying storage pool actually uses multiple block
1005 * size. Under Solaris frsize (fragment size) is reported as
1006 * the smallest block size we support, and bsize (block size)
1007 * as the filesystem's maximum block size. Unfortunately,
1008 * under Linux the fragment size and block size are often used
1009 * interchangeably. Thus we are forced to report both of them
1010 * as the filesystem's maximum block size.
1012 statp->f_frsize = zsb->z_max_blksz;
1013 statp->f_bsize = zsb->z_max_blksz;
1014 bshift = fls(statp->f_bsize) - 1;
1017 * The following report "total" blocks of various kinds in
1018 * the file system, but reported in terms of f_bsize - the
1022 statp->f_blocks = (refdbytes + availbytes) >> bshift;
1023 statp->f_bfree = availbytes >> bshift;
1024 statp->f_bavail = statp->f_bfree; /* no root reservation */
1027 * statvfs() should really be called statufs(), because it assumes
1028 * static metadata. ZFS doesn't preallocate files, so the best
1029 * we can do is report the max that could possibly fit in f_files,
1030 * and that minus the number actually used in f_ffree.
1031 * For f_ffree, report the smaller of the number of object available
1032 * and the number of blocks (each object will take at least a block).
1034 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1035 statp->f_files = statp->f_ffree + usedobjs;
1036 statp->f_fsid.val[0] = dentry->d_sb->s_dev;
1037 statp->f_fsid.val[1] = 0;
1038 statp->f_type = ZFS_SUPER_MAGIC;
1039 statp->f_namelen = ZFS_MAXNAMELEN;
1042 * We have all of 40 characters to stuff a string here.
1043 * Is there anything useful we could/should provide?
1045 bzero(statp->f_spare, sizeof (statp->f_spare));
1050 EXPORT_SYMBOL(zfs_statvfs);
1053 zfs_root(zfs_sb_t *zsb, struct inode **ipp)
1060 error = zfs_zget(zsb, zsb->z_root, &rootzp);
1062 *ipp = ZTOI(rootzp);
1067 EXPORT_SYMBOL(zfs_root);
1070 * Teardown the zfs_sb_t::z_os.
1072 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1073 * and 'z_teardown_inactive_lock' held.
1076 zfsvfs_teardown(zfs_sb_t *zsb, boolean_t unmounting)
1080 rrw_enter(&zsb->z_teardown_lock, RW_WRITER, FTAG);
1084 * We purge the parent filesystem's super block as the
1085 * parent filesystem and all of its snapshots have their
1086 * inode's super block set to the parent's filesystem's
1087 * super block. Note, 'z_parent' is self referential
1088 * for non-snapshots.
1090 shrink_dcache_sb(zsb->z_parent->z_sb);
1091 (void) spl_invalidate_inodes(zsb->z_parent->z_sb, 0);
1095 * Drain the iput_taskq to ensure all active references to the
1096 * zfs_sb_t have been handled only then can it be safely destroyed.
1098 taskq_wait(dsl_pool_iput_taskq(dmu_objset_pool(zsb->z_os)));
1101 * Close the zil. NB: Can't close the zil while zfs_inactive
1102 * threads are blocked as zil_close can call zfs_inactive.
1105 zil_close(zsb->z_log);
1109 rw_enter(&zsb->z_teardown_inactive_lock, RW_WRITER);
1112 * If we are not unmounting (ie: online recv) and someone already
1113 * unmounted this file system while we were doing the switcheroo,
1114 * or a reopen of z_os failed then just bail out now.
1116 if (!unmounting && (zsb->z_unmounted || zsb->z_os == NULL)) {
1117 rw_exit(&zsb->z_teardown_inactive_lock);
1118 rrw_exit(&zsb->z_teardown_lock, FTAG);
1123 * At this point there are no vops active, and any new vops will
1124 * fail with EIO since we have z_teardown_lock for writer (only
1125 * relavent for forced unmount).
1127 * Release all holds on dbufs.
1129 mutex_enter(&zsb->z_znodes_lock);
1130 for (zp = list_head(&zsb->z_all_znodes); zp != NULL;
1131 zp = list_next(&zsb->z_all_znodes, zp))
1133 ASSERT(atomic_read(&ZTOI(zp)->i_count) > 0);
1134 zfs_znode_dmu_fini(zp);
1136 mutex_exit(&zsb->z_znodes_lock);
1139 * If we are unmounting, set the unmounted flag and let new vops
1140 * unblock. zfs_inactive will have the unmounted behavior, and all
1141 * other vops will fail with EIO.
1144 zsb->z_unmounted = B_TRUE;
1145 rrw_exit(&zsb->z_teardown_lock, FTAG);
1146 rw_exit(&zsb->z_teardown_inactive_lock);
1150 * z_os will be NULL if there was an error in attempting to reopen
1151 * zsb, so just return as the properties had already been
1153 * unregistered and cached data had been evicted before.
1155 if (zsb->z_os == NULL)
1159 * Unregister properties.
1161 zfs_unregister_callbacks(zsb);
1166 if (dmu_objset_is_dirty_anywhere(zsb->z_os))
1167 if (!(zsb->z_vfs->mnt_flags & MNT_READONLY))
1168 txg_wait_synced(dmu_objset_pool(zsb->z_os), 0);
1169 (void) dmu_objset_evict_dbufs(zsb->z_os);
1175 zfs_domount(struct super_block *sb, void *data, int silent)
1177 zpl_mount_data_t *zmd = data;
1178 const char *osname = zmd->z_osname;
1180 struct inode *root_inode;
1181 uint64_t recordsize;
1185 * Linux allows multiple vfs mounts per super block. However, the
1186 * zfs_sb_t only contains a pointer for a single vfs mount. This
1187 * back reference in the long term could be extended to a list of
1188 * vfs mounts if a hook were added to the kernel to notify us when
1189 * a vfsmount is destroyed. Until then we must limit the number
1190 * of mounts per super block to one.
1192 if (atomic_read(&sb->s_active) > 1)
1195 error = zfs_sb_create(osname, &zsb);
1199 if ((error = dsl_prop_get_integer(osname, "recordsize",
1200 &recordsize, NULL)))
1204 zsb->z_vfs = zmd->z_vfs;
1205 sb->s_fs_info = zsb;
1206 sb->s_magic = ZFS_SUPER_MAGIC;
1207 sb->s_maxbytes = MAX_LFS_FILESIZE;
1208 sb->s_time_gran = 1;
1209 sb->s_blocksize = recordsize;
1210 sb->s_blocksize_bits = ilog2(recordsize);
1212 /* Set callback operations for the file system. */
1213 sb->s_op = &zpl_super_operations;
1214 sb->s_xattr = zpl_xattr_handlers;
1215 sb->s_export_op = &zpl_export_operations;
1217 /* Set features for file system. */
1218 zfs_set_fuid_feature(zsb);
1220 if (dmu_objset_is_snapshot(zsb->z_os)) {
1223 atime_changed_cb(zsb, B_FALSE);
1224 readonly_changed_cb(zsb, B_TRUE);
1225 if ((error = dsl_prop_get_integer(osname,"xattr",&pval,NULL)))
1227 xattr_changed_cb(zsb, pval);
1228 zsb->z_issnap = B_TRUE;
1229 zsb->z_os->os_sync = ZFS_SYNC_DISABLED;
1231 mutex_enter(&zsb->z_os->os_user_ptr_lock);
1232 dmu_objset_set_user(zsb->z_os, zsb);
1233 mutex_exit(&zsb->z_os->os_user_ptr_lock);
1235 error = zfs_sb_setup(zsb, B_TRUE);
1236 #ifdef HAVE_SNAPSHOT
1237 (void) zfs_snap_create(zsb);
1238 #endif /* HAVE_SNAPSHOT */
1241 /* Allocate a root inode for the filesystem. */
1242 error = zfs_root(zsb, &root_inode);
1244 (void) zfs_umount(sb);
1248 /* Allocate a root dentry for the filesystem */
1249 sb->s_root = d_alloc_root(root_inode);
1250 if (sb->s_root == NULL) {
1251 (void) zfs_umount(sb);
1257 dmu_objset_disown(zsb->z_os, zsb);
1263 EXPORT_SYMBOL(zfs_domount);
1267 zfs_umount(struct super_block *sb)
1269 zfs_sb_t *zsb = sb->s_fs_info;
1272 VERIFY(zfsvfs_teardown(zsb, B_TRUE) == 0);
1276 * z_os will be NULL if there was an error in
1277 * attempting to reopen zsb.
1281 * Unset the objset user_ptr.
1283 mutex_enter(&os->os_user_ptr_lock);
1284 dmu_objset_set_user(os, NULL);
1285 mutex_exit(&os->os_user_ptr_lock);
1288 * Finally release the objset
1290 dmu_objset_disown(os, zsb);
1296 EXPORT_SYMBOL(zfs_umount);
1299 zfs_remount(struct super_block *sb, int *flags, char *data)
1301 zfs_sb_t *zsb = sb->s_fs_info;
1302 boolean_t readonly = B_FALSE;
1303 boolean_t setuid = B_TRUE;
1304 boolean_t exec = B_TRUE;
1305 boolean_t devices = B_TRUE;
1306 boolean_t atime = B_TRUE;
1308 if (*flags & MS_RDONLY)
1311 if (*flags & MS_NOSUID) {
1315 if (*flags & MS_NODEV)
1319 if (*flags & MS_NOEXEC)
1322 if (*flags & MS_NOATIME)
1326 * Invoke our callbacks to set required flags.
1328 readonly_changed_cb(zsb, readonly);
1329 setuid_changed_cb(zsb, setuid);
1330 exec_changed_cb(zsb, exec);
1331 devices_changed_cb(zsb, devices);
1332 atime_changed_cb(zsb, atime);
1336 EXPORT_SYMBOL(zfs_remount);
1339 zfs_vget(struct vfsmount *vfsp, struct inode **ipp, fid_t *fidp)
1341 zfs_sb_t *zsb = VTOZSB(vfsp);
1343 uint64_t object = 0;
1344 uint64_t fid_gen = 0;
1353 if (fidp->fid_len == LONG_FID_LEN) {
1354 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1355 uint64_t objsetid = 0;
1356 uint64_t setgen = 0;
1358 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1359 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1361 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1362 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1366 #ifdef HAVE_SNAPSHOT
1367 err = zfsctl_lookup_objset(vfsp, objsetid, &zsb);
1370 #endif /* HAVE_SNAPSHOT */
1374 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1375 zfid_short_t *zfid = (zfid_short_t *)fidp;
1377 for (i = 0; i < sizeof (zfid->zf_object); i++)
1378 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1380 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1381 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1387 #ifdef HAVE_SNAPSHOT
1388 /* A zero fid_gen means we are in the .zfs control directories */
1390 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1391 *ipp = zsb->z_ctldir;
1392 ASSERT(*ipp != NULL);
1393 if (object == ZFSCTL_INO_SNAPDIR) {
1394 VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp, NULL,
1395 0, NULL, NULL, NULL, NULL, NULL) == 0);
1402 #endif /* HAVE_SNAPSHOT */
1404 gen_mask = -1ULL >> (64 - 8 * i);
1406 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1407 if ((err = zfs_zget(zsb, object, &zp))) {
1411 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zsb), &zp_gen,
1413 zp_gen = zp_gen & gen_mask;
1416 if (zp->z_unlinked || zp_gen != fid_gen) {
1417 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1425 zfs_inode_update(ITOZ(*ipp));
1430 EXPORT_SYMBOL(zfs_vget);
1433 * Block out VOPs and close zfs_sb_t::z_os
1435 * Note, if successful, then we return with the 'z_teardown_lock' and
1436 * 'z_teardown_inactive_lock' write held.
1439 zfs_suspend_fs(zfs_sb_t *zsb)
1443 if ((error = zfsvfs_teardown(zsb, B_FALSE)) != 0)
1445 dmu_objset_disown(zsb->z_os, zsb);
1449 EXPORT_SYMBOL(zfs_suspend_fs);
1452 * Reopen zfs_sb_t::z_os and release VOPs.
1455 zfs_resume_fs(zfs_sb_t *zsb, const char *osname)
1459 ASSERT(RRW_WRITE_HELD(&zsb->z_teardown_lock));
1460 ASSERT(RW_WRITE_HELD(&zsb->z_teardown_inactive_lock));
1462 err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zsb, &zsb->z_os);
1467 uint64_t sa_obj = 0;
1469 err2 = zap_lookup(zsb->z_os, MASTER_NODE_OBJ,
1470 ZFS_SA_ATTRS, 8, 1, &sa_obj);
1472 if ((err || err2) && zsb->z_version >= ZPL_VERSION_SA)
1476 if ((err = sa_setup(zsb->z_os, sa_obj,
1477 zfs_attr_table, ZPL_END, &zsb->z_attr_table)) != 0)
1480 VERIFY(zfs_sb_setup(zsb, B_FALSE) == 0);
1483 * Attempt to re-establish all the active znodes with
1484 * their dbufs. If a zfs_rezget() fails, then we'll let
1485 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
1486 * when they try to use their znode.
1488 mutex_enter(&zsb->z_znodes_lock);
1489 for (zp = list_head(&zsb->z_all_znodes); zp;
1490 zp = list_next(&zsb->z_all_znodes, zp)) {
1491 (void) zfs_rezget(zp);
1493 mutex_exit(&zsb->z_znodes_lock);
1498 /* release the VOPs */
1499 rw_exit(&zsb->z_teardown_inactive_lock);
1500 rrw_exit(&zsb->z_teardown_lock, FTAG);
1504 * Since we couldn't reopen zfs_sb_t::z_os, force
1505 * unmount this file system.
1507 (void) zfs_umount(zsb->z_sb);
1511 EXPORT_SYMBOL(zfs_resume_fs);
1514 zfs_set_version(zfs_sb_t *zsb, uint64_t newvers)
1517 objset_t *os = zsb->z_os;
1520 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
1523 if (newvers < zsb->z_version)
1526 if (zfs_spa_version_map(newvers) >
1527 spa_version(dmu_objset_spa(zsb->z_os)))
1530 tx = dmu_tx_create(os);
1531 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
1532 if (newvers >= ZPL_VERSION_SA && !zsb->z_use_sa) {
1533 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
1535 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1537 error = dmu_tx_assign(tx, TXG_WAIT);
1543 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
1544 8, 1, &newvers, tx);
1551 if (newvers >= ZPL_VERSION_SA && !zsb->z_use_sa) {
1554 ASSERT3U(spa_version(dmu_objset_spa(zsb->z_os)), >=,
1556 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
1557 DMU_OT_NONE, 0, tx);
1559 error = zap_add(os, MASTER_NODE_OBJ,
1560 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
1561 ASSERT3U(error, ==, 0);
1563 VERIFY(0 == sa_set_sa_object(os, sa_obj));
1564 sa_register_update_callback(os, zfs_sa_upgrade);
1567 spa_history_log_internal(LOG_DS_UPGRADE,
1568 dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu",
1569 zsb->z_version, newvers, dmu_objset_id(os));
1573 zsb->z_version = newvers;
1575 if (zsb->z_version >= ZPL_VERSION_FUID)
1576 zfs_set_fuid_feature(zsb);
1580 EXPORT_SYMBOL(zfs_set_version);
1583 * Read a property stored within the master node.
1586 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
1592 * Look up the file system's value for the property. For the
1593 * version property, we look up a slightly different string.
1595 if (prop == ZFS_PROP_VERSION)
1596 pname = ZPL_VERSION_STR;
1598 pname = zfs_prop_to_name(prop);
1601 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
1603 if (error == ENOENT) {
1604 /* No value set, use the default value */
1606 case ZFS_PROP_VERSION:
1607 *value = ZPL_VERSION;
1609 case ZFS_PROP_NORMALIZE:
1610 case ZFS_PROP_UTF8ONLY:
1614 *value = ZFS_CASE_SENSITIVE;
1628 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
1629 register_filesystem(&zpl_fs_type);
1635 unregister_filesystem(&zpl_fs_type);