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_dir.h>
43 #include <sys/fs/zfs.h>
45 #include <sys/dsl_prop.h>
46 #include <sys/dsl_dataset.h>
47 #include <sys/dsl_deleg.h>
51 #include <sys/varargs.h>
52 #include <sys/policy.h>
53 #include <sys/atomic.h>
54 #include <sys/mkdev.h>
55 #include <sys/modctl.h>
56 #include <sys/refstr.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/zfs_ctldir.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>
66 #include "zfs_comutil.h"
70 vfsops_t *zfs_vfsops = NULL;
71 static major_t zfs_major;
72 static minor_t zfs_minor;
73 static kmutex_t zfs_dev_mtx;
75 extern int sys_shutdown;
77 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
78 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
79 static void zfs_freevfs(vfs_t *vfsp);
81 static const fs_operation_def_t zfs_vfsops_template[] = {
82 VFSNAME_MOUNT, { .vfs_mount = zfs_mount },
83 VFSNAME_MOUNTROOT, { .vfs_mountroot = zfs_mountroot },
84 VFSNAME_UNMOUNT, { .vfs_unmount = zfs_umount },
85 VFSNAME_ROOT, { .vfs_root = zfs_root },
86 VFSNAME_STATVFS, { .vfs_statvfs = zfs_statvfs },
87 VFSNAME_SYNC, { .vfs_sync = zfs_sync },
88 VFSNAME_VGET, { .vfs_vget = zfs_vget },
89 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
93 static const fs_operation_def_t zfs_vfsops_eio_template[] = {
94 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
99 * We need to keep a count of active fs's.
100 * This is necessary to prevent our module
101 * from being unloaded after a umount -f
103 static uint32_t zfs_active_fs_count = 0;
105 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
106 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
107 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
108 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
111 * MO_DEFAULT is not used since the default value is determined
112 * by the equivalent property.
114 static mntopt_t mntopts[] = {
115 { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
116 { MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
117 { MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
118 { MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
121 static mntopts_t zfs_mntopts = {
122 sizeof (mntopts) / sizeof (mntopt_t),
128 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
131 * Data integrity is job one. We don't want a compromised kernel
132 * writing to the storage pool, so we never sync during panic.
138 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
139 * to sync metadata, which they would otherwise cache indefinitely.
140 * Semantically, the only requirement is that the sync be initiated.
141 * The DMU syncs out txgs frequently, so there's nothing to do.
143 if (flag & SYNC_ATTR)
148 * Sync a specific filesystem.
150 zfsvfs_t *zfsvfs = vfsp->vfs_data;
154 dp = dmu_objset_pool(zfsvfs->z_os);
157 * If the system is shutting down, then skip any
158 * filesystems which may exist on a suspended pool.
160 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
165 if (zfsvfs->z_log != NULL)
166 zil_commit(zfsvfs->z_log, 0);
171 * Sync all ZFS filesystems. This is what happens when you
172 * run sync(1M). Unlike other filesystems, ZFS honors the
173 * request by waiting for all pools to commit all dirty data.
180 EXPORT_SYMBOL(zfs_sync);
183 zfs_create_unique_device(dev_t *dev)
188 ASSERT3U(zfs_minor, <=, MAXMIN32);
189 minor_t start = zfs_minor;
191 mutex_enter(&zfs_dev_mtx);
192 if (zfs_minor >= MAXMIN32) {
194 * If we're still using the real major
195 * keep out of /dev/zfs and /dev/zvol minor
196 * number space. If we're using a getudev()'ed
197 * major number, we can use all of its minors.
199 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
200 zfs_minor = ZFS_MIN_MINOR;
206 *dev = makedevice(zfs_major, zfs_minor);
207 mutex_exit(&zfs_dev_mtx);
208 } while (vfs_devismounted(*dev) && zfs_minor != start);
209 if (zfs_minor == start) {
211 * We are using all ~262,000 minor numbers for the
212 * current major number. Create a new major number.
214 if ((new_major = getudev()) == (major_t)-1) {
216 "zfs_mount: Can't get unique major "
220 mutex_enter(&zfs_dev_mtx);
221 zfs_major = new_major;
224 mutex_exit(&zfs_dev_mtx);
228 /* CONSTANTCONDITION */
235 atime_changed_cb(void *arg, uint64_t newval)
237 zfsvfs_t *zfsvfs = arg;
239 if (newval == TRUE) {
240 zfsvfs->z_atime = TRUE;
241 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
242 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
244 zfsvfs->z_atime = FALSE;
245 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
246 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
251 xattr_changed_cb(void *arg, uint64_t newval)
253 zfsvfs_t *zfsvfs = arg;
255 if (newval == TRUE) {
256 /* XXX locking on vfs_flag? */
257 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
258 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
259 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
261 /* XXX locking on vfs_flag? */
262 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
263 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
264 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
269 blksz_changed_cb(void *arg, uint64_t newval)
271 zfsvfs_t *zfsvfs = arg;
273 if (newval < SPA_MINBLOCKSIZE ||
274 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
275 newval = SPA_MAXBLOCKSIZE;
277 zfsvfs->z_max_blksz = newval;
278 zfsvfs->z_vfs->vfs_bsize = newval;
282 readonly_changed_cb(void *arg, uint64_t newval)
284 zfsvfs_t *zfsvfs = arg;
287 /* XXX locking on vfs_flag? */
288 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
289 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
290 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
292 /* XXX locking on vfs_flag? */
293 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
294 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
295 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
300 devices_changed_cb(void *arg, uint64_t newval)
302 zfsvfs_t *zfsvfs = arg;
304 if (newval == FALSE) {
305 zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
306 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
307 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
309 zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
310 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
311 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
316 setuid_changed_cb(void *arg, uint64_t newval)
318 zfsvfs_t *zfsvfs = arg;
320 if (newval == FALSE) {
321 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
322 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
323 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
325 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
326 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
327 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
332 exec_changed_cb(void *arg, uint64_t newval)
334 zfsvfs_t *zfsvfs = arg;
336 if (newval == FALSE) {
337 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
338 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
339 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
341 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
342 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
343 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
348 * The nbmand mount option can be changed at mount time.
349 * We can't allow it to be toggled on live file systems or incorrect
350 * behavior may be seen from cifs clients
352 * This property isn't registered via dsl_prop_register(), but this callback
353 * will be called when a file system is first mounted
356 nbmand_changed_cb(void *arg, uint64_t newval)
358 zfsvfs_t *zfsvfs = arg;
359 if (newval == FALSE) {
360 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
361 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
363 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
364 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
369 snapdir_changed_cb(void *arg, uint64_t newval)
371 zfsvfs_t *zfsvfs = arg;
373 zfsvfs->z_show_ctldir = newval;
377 vscan_changed_cb(void *arg, uint64_t newval)
379 zfsvfs_t *zfsvfs = arg;
381 zfsvfs->z_vscan = newval;
385 acl_inherit_changed_cb(void *arg, uint64_t newval)
387 zfsvfs_t *zfsvfs = arg;
389 zfsvfs->z_acl_inherit = newval;
393 zfs_register_callbacks(vfs_t *vfsp)
395 struct dsl_dataset *ds = NULL;
397 zfsvfs_t *zfsvfs = NULL;
399 int readonly, do_readonly = B_FALSE;
400 int setuid, do_setuid = B_FALSE;
401 int exec, do_exec = B_FALSE;
402 int devices, do_devices = B_FALSE;
403 int xattr, do_xattr = B_FALSE;
404 int atime, do_atime = B_FALSE;
408 zfsvfs = vfsp->vfs_data;
413 * The act of registering our callbacks will destroy any mount
414 * options we may have. In order to enable temporary overrides
415 * of mount options, we stash away the current values and
416 * restore them after we register the callbacks.
418 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
419 !spa_writeable(dmu_objset_spa(os))) {
421 do_readonly = B_TRUE;
422 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
424 do_readonly = B_TRUE;
426 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
432 if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
435 } else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
440 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
443 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
448 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
451 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
455 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
458 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
462 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
465 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
471 * nbmand is a special property. It can only be changed at
474 * This is weird, but it is documented to only be changeable
477 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
479 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
482 char osname[MAXNAMELEN];
484 dmu_objset_name(os, osname);
485 if ((error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
492 * Register property callbacks.
494 * It would probably be fine to just check for i/o error from
495 * the first prop_register(), but I guess I like to go
498 ds = dmu_objset_ds(os);
499 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs);
500 error = error ? error : dsl_prop_register(ds,
501 "xattr", xattr_changed_cb, zfsvfs);
502 error = error ? error : dsl_prop_register(ds,
503 "recordsize", blksz_changed_cb, zfsvfs);
504 error = error ? error : dsl_prop_register(ds,
505 "readonly", readonly_changed_cb, zfsvfs);
506 error = error ? error : dsl_prop_register(ds,
507 "devices", devices_changed_cb, zfsvfs);
508 error = error ? error : dsl_prop_register(ds,
509 "setuid", setuid_changed_cb, zfsvfs);
510 error = error ? error : dsl_prop_register(ds,
511 "exec", exec_changed_cb, zfsvfs);
512 error = error ? error : dsl_prop_register(ds,
513 "snapdir", snapdir_changed_cb, zfsvfs);
514 error = error ? error : dsl_prop_register(ds,
515 "aclinherit", acl_inherit_changed_cb, zfsvfs);
516 error = error ? error : dsl_prop_register(ds,
517 "vscan", vscan_changed_cb, zfsvfs);
522 * Invoke our callbacks to restore temporary mount options.
525 readonly_changed_cb(zfsvfs, readonly);
527 setuid_changed_cb(zfsvfs, setuid);
529 exec_changed_cb(zfsvfs, exec);
531 devices_changed_cb(zfsvfs, devices);
533 xattr_changed_cb(zfsvfs, xattr);
535 atime_changed_cb(zfsvfs, atime);
537 nbmand_changed_cb(zfsvfs, nbmand);
543 * We may attempt to unregister some callbacks that are not
544 * registered, but this is OK; it will simply return ENOMSG,
545 * which we will ignore.
547 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs);
548 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs);
549 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs);
550 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs);
551 (void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zfsvfs);
552 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs);
553 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs);
554 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs);
555 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
557 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs);
561 EXPORT_SYMBOL(zfs_register_callbacks);
562 #endif /* HAVE_ZPL */
565 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
566 uint64_t *userp, uint64_t *groupp)
568 znode_phys_t *znp = data;
572 * Is it a valid type of object to track?
574 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
578 * If we have a NULL data pointer
579 * then assume the id's aren't changing and
580 * return EEXIST to the dmu to let it know to
586 if (bonustype == DMU_OT_ZNODE) {
587 *userp = znp->zp_uid;
588 *groupp = znp->zp_gid;
592 ASSERT(bonustype == DMU_OT_SA);
593 hdrsize = sa_hdrsize(data);
596 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
598 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
602 * This should only happen for newly created
603 * files that haven't had the znode data filled
615 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
616 char *domainbuf, int buflen, uid_t *ridp)
621 fuid = strtonum(fuidstr, NULL);
623 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
625 (void) strlcpy(domainbuf, domain, buflen);
628 *ridp = FUID_RID(fuid);
632 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
635 case ZFS_PROP_USERUSED:
636 return (DMU_USERUSED_OBJECT);
637 case ZFS_PROP_GROUPUSED:
638 return (DMU_GROUPUSED_OBJECT);
639 case ZFS_PROP_USERQUOTA:
640 return (zfsvfs->z_userquota_obj);
641 case ZFS_PROP_GROUPQUOTA:
642 return (zfsvfs->z_groupquota_obj);
650 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
651 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
656 zfs_useracct_t *buf = vbuf;
659 if (!dmu_objset_userspace_present(zfsvfs->z_os))
662 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
668 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
669 (error = zap_cursor_retrieve(&zc, &za)) == 0;
670 zap_cursor_advance(&zc)) {
671 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
675 fuidstr_to_sid(zfsvfs, za.za_name,
676 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
678 buf->zu_space = za.za_first_integer;
684 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
685 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
686 *cookiep = zap_cursor_serialize(&zc);
687 zap_cursor_fini(&zc);
690 EXPORT_SYMBOL(zfs_userspace_many);
693 * buf must be big enough (eg, 32 bytes)
696 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
697 char *buf, boolean_t addok)
702 if (domain && domain[0]) {
703 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
707 fuid = FUID_ENCODE(domainid, rid);
708 (void) sprintf(buf, "%llx", (longlong_t)fuid);
713 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
714 const char *domain, uint64_t rid, uint64_t *valp)
722 if (!dmu_objset_userspace_present(zfsvfs->z_os))
725 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
729 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
733 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
738 EXPORT_SYMBOL(zfs_userspace_one);
741 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
742 const char *domain, uint64_t rid, uint64_t quota)
748 boolean_t fuid_dirtied;
750 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
753 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
756 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
757 &zfsvfs->z_groupquota_obj;
759 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
762 fuid_dirtied = zfsvfs->z_fuid_dirty;
764 tx = dmu_tx_create(zfsvfs->z_os);
765 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
767 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
768 zfs_userquota_prop_prefixes[type]);
771 zfs_fuid_txhold(zfsvfs, tx);
772 err = dmu_tx_assign(tx, TXG_WAIT);
778 mutex_enter(&zfsvfs->z_lock);
780 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
782 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
783 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
785 mutex_exit(&zfsvfs->z_lock);
788 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
792 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
796 zfs_fuid_sync(zfsvfs, tx);
800 EXPORT_SYMBOL(zfs_set_userquota);
803 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
806 uint64_t used, quota, usedobj, quotaobj;
809 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
810 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
812 if (quotaobj == 0 || zfsvfs->z_replay)
815 (void) sprintf(buf, "%llx", (longlong_t)fuid);
816 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
820 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
823 return (used >= quota);
825 EXPORT_SYMBOL(zfs_fuid_overquota);
828 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
833 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
835 fuid = isgroup ? zp->z_gid : zp->z_uid;
837 if (quotaobj == 0 || zfsvfs->z_replay)
840 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
842 EXPORT_SYMBOL(zfs_owner_overquota);
845 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
853 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
856 * We claim to always be readonly so we can open snapshots;
857 * other ZPL code will prevent us from writing to snapshots.
859 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
861 kmem_free(zfsvfs, sizeof (zfsvfs_t));
866 * Initialize the zfs-specific filesystem structure.
867 * Should probably make this a kmem cache, shuffle fields,
868 * and just bzero up to z_hold_mtx[].
870 zfsvfs->z_vfs = NULL;
871 zfsvfs->z_parent = zfsvfs;
872 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
873 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
876 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
879 } else if (zfsvfs->z_version >
880 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
881 (void) printk("Can't mount a version %lld file system "
882 "on a version %lld pool\n. Pool must be upgraded to mount "
883 "this file system.", (u_longlong_t)zfsvfs->z_version,
884 (u_longlong_t)spa_version(dmu_objset_spa(os)));
888 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
890 zfsvfs->z_norm = (int)zval;
892 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
894 zfsvfs->z_utf8 = (zval != 0);
896 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
898 zfsvfs->z_case = (uint_t)zval;
901 * Fold case on file systems that are always or sometimes case
904 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
905 zfsvfs->z_case == ZFS_CASE_MIXED)
906 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
908 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
909 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
911 if (zfsvfs->z_use_sa) {
912 /* should either have both of these objects or none */
913 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
919 * Pre SA versions file systems should never touch
920 * either the attribute registration or layout objects.
925 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
926 &zfsvfs->z_attr_table);
930 if (zfsvfs->z_version >= ZPL_VERSION_SA)
931 sa_register_update_callback(os, zfs_sa_upgrade);
933 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
937 ASSERT(zfsvfs->z_root != 0);
939 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
940 &zfsvfs->z_unlinkedobj);
944 error = zap_lookup(os, MASTER_NODE_OBJ,
945 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
946 8, 1, &zfsvfs->z_userquota_obj);
947 if (error && error != ENOENT)
950 error = zap_lookup(os, MASTER_NODE_OBJ,
951 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
952 8, 1, &zfsvfs->z_groupquota_obj);
953 if (error && error != ENOENT)
956 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
957 &zfsvfs->z_fuid_obj);
958 if (error && error != ENOENT)
961 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
962 &zfsvfs->z_shares_dir);
963 if (error && error != ENOENT)
966 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
967 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
968 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
969 offsetof(znode_t, z_link_node));
970 rrw_init(&zfsvfs->z_teardown_lock);
971 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
972 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
973 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
974 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
980 dmu_objset_disown(os, zfsvfs);
982 kmem_free(zfsvfs, sizeof (zfsvfs_t));
987 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
991 error = zfs_register_callbacks(zfsvfs->z_vfs);
996 * Set the objset user_ptr to track its zfsvfs.
998 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
999 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1000 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1002 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1005 * If we are not mounting (ie: online recv), then we don't
1006 * have to worry about replaying the log as we blocked all
1007 * operations out since we closed the ZIL.
1013 * During replay we remove the read only flag to
1014 * allow replays to succeed.
1016 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1018 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1020 zfs_unlinked_drain(zfsvfs);
1023 * Parse and replay the intent log.
1025 * Because of ziltest, this must be done after
1026 * zfs_unlinked_drain(). (Further note: ziltest
1027 * doesn't use readonly mounts, where
1028 * zfs_unlinked_drain() isn't called.) This is because
1029 * ziltest causes spa_sync() to think it's committed,
1030 * but actually it is not, so the intent log contains
1031 * many txg's worth of changes.
1033 * In particular, if object N is in the unlinked set in
1034 * the last txg to actually sync, then it could be
1035 * actually freed in a later txg and then reallocated
1036 * in a yet later txg. This would write a "create
1037 * object N" record to the intent log. Normally, this
1038 * would be fine because the spa_sync() would have
1039 * written out the fact that object N is free, before
1040 * we could write the "create object N" intent log
1043 * But when we are in ziltest mode, we advance the "open
1044 * txg" without actually spa_sync()-ing the changes to
1045 * disk. So we would see that object N is still
1046 * allocated and in the unlinked set, and there is an
1047 * intent log record saying to allocate it.
1049 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1050 if (zil_replay_disable) {
1051 zil_destroy(zfsvfs->z_log, B_FALSE);
1053 zfsvfs->z_replay = B_TRUE;
1054 zil_replay(zfsvfs->z_os, zfsvfs,
1056 zfsvfs->z_replay = B_FALSE;
1059 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1066 zfsvfs_free(zfsvfs_t *zfsvfs)
1069 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1072 * This is a barrier to prevent the filesystem from going away in
1073 * zfs_znode_move() until we can safely ensure that the filesystem is
1074 * not unmounted. We consider the filesystem valid before the barrier
1075 * and invalid after the barrier.
1077 rw_enter(&zfsvfs_lock, RW_READER);
1078 rw_exit(&zfsvfs_lock);
1080 zfs_fuid_destroy(zfsvfs);
1082 mutex_destroy(&zfsvfs->z_znodes_lock);
1083 mutex_destroy(&zfsvfs->z_lock);
1084 list_destroy(&zfsvfs->z_all_znodes);
1085 rrw_destroy(&zfsvfs->z_teardown_lock);
1086 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1087 rw_destroy(&zfsvfs->z_fuid_lock);
1088 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1089 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1090 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1094 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1096 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1097 if (zfsvfs->z_use_fuids && zfsvfs->z_vfs) {
1098 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1099 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1100 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1101 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1102 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1103 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1105 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1109 zfs_domount(vfs_t *vfsp, char *osname)
1112 uint64_t recordsize, fsid_guid;
1119 error = zfsvfs_create(osname, &zfsvfs);
1122 zfsvfs->z_vfs = vfsp;
1124 /* Initialize the generic filesystem structure. */
1125 vfsp->vfs_bcount = 0;
1126 vfsp->vfs_data = NULL;
1128 if (zfs_create_unique_device(&mount_dev) == -1) {
1132 ASSERT(vfs_devismounted(mount_dev) == 0);
1134 if ((error = dsl_prop_get_integer(osname, "recordsize",
1135 &recordsize, NULL)))
1138 vfsp->vfs_dev = mount_dev;
1139 vfsp->vfs_fstype = zfsfstype;
1140 vfsp->vfs_bsize = recordsize;
1141 vfsp->vfs_flag |= VFS_NOTRUNC;
1142 vfsp->vfs_data = zfsvfs;
1145 * The fsid is 64 bits, composed of an 8-bit fs type, which
1146 * separates our fsid from any other filesystem types, and a
1147 * 56-bit objset unique ID. The objset unique ID is unique to
1148 * all objsets open on this system, provided by unique_create().
1149 * The 8-bit fs type must be put in the low bits of fsid[1]
1150 * because that's where other Solaris filesystems put it.
1152 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1153 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1154 vfsp->vfs_fsid.val[0] = fsid_guid;
1155 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1159 * Set features for file system.
1161 zfs_set_fuid_feature(zfsvfs);
1162 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1163 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1164 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1165 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1166 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1167 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1168 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1170 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1172 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1175 atime_changed_cb(zfsvfs, B_FALSE);
1176 readonly_changed_cb(zfsvfs, B_TRUE);
1177 if ((error = dsl_prop_get_integer(osname,"xattr",&pval,NULL)))
1179 xattr_changed_cb(zfsvfs, pval);
1180 zfsvfs->z_issnap = B_TRUE;
1181 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1183 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1184 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1185 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1187 error = zfsvfs_setup(zfsvfs, B_TRUE);
1190 if (!zfsvfs->z_issnap)
1191 zfsctl_create(zfsvfs);
1194 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1195 zfsvfs_free(zfsvfs);
1197 atomic_add_32(&zfs_active_fs_count, 1);
1202 EXPORT_SYMBOL(zfs_domount);
1205 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1207 objset_t *os = zfsvfs->z_os;
1208 struct dsl_dataset *ds;
1211 * Unregister properties.
1213 if (!dmu_objset_is_snapshot(os)) {
1214 ds = dmu_objset_ds(os);
1215 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1218 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1221 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1224 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1227 VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1230 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1233 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1236 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1239 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1240 acl_inherit_changed_cb, zfsvfs) == 0);
1242 VERIFY(dsl_prop_unregister(ds, "vscan",
1243 vscan_changed_cb, zfsvfs) == 0);
1246 EXPORT_SYMBOL(zfs_unregister_callbacks);
1249 * Convert a decimal digit string to a uint64_t integer.
1252 str_to_uint64(char *str, uint64_t *objnum)
1257 if (*str < '0' || *str > '9')
1260 num = num*10 + *str++ - '0';
1268 * The boot path passed from the boot loader is in the form of
1269 * "rootpool-name/root-filesystem-object-number'. Convert this
1270 * string to a dataset name: "rootpool-name/root-filesystem-name".
1273 zfs_parse_bootfs(char *bpath, char *outpath)
1279 if (*bpath == 0 || *bpath == '/')
1282 (void) strcpy(outpath, bpath);
1284 slashp = strchr(bpath, '/');
1286 /* if no '/', just return the pool name */
1287 if (slashp == NULL) {
1291 /* if not a number, just return the root dataset name */
1292 if (str_to_uint64(slashp+1, &objnum)) {
1297 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1303 #ifdef HAVE_MLSLABEL
1305 * zfs_check_global_label:
1306 * Check that the hex label string is appropriate for the dataset
1307 * being mounted into the global_zone proper.
1309 * Return an error if the hex label string is not default or
1310 * admin_low/admin_high. For admin_low labels, the corresponding
1311 * dataset must be readonly.
1314 zfs_check_global_label(const char *dsname, const char *hexsl)
1316 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1318 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1320 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1321 /* must be readonly */
1324 if (dsl_prop_get_integer(dsname,
1325 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1327 return (rdonly ? 0 : EACCES);
1331 #endif /* HAVE_MLSLABEL */
1334 * zfs_mount_label_policy:
1335 * Determine whether the mount is allowed according to MAC check.
1336 * by comparing (where appropriate) label of the dataset against
1337 * the label of the zone being mounted into. If the dataset has
1338 * no label, create one.
1341 * 0 : access allowed
1342 * >0 : error code, such as EACCES
1345 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1348 zone_t *mntzone = NULL;
1349 ts_label_t *mnt_tsl;
1352 char ds_hexsl[MAXNAMELEN];
1354 retv = EACCES; /* assume the worst */
1357 * Start by getting the dataset label if it exists.
1359 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1360 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1365 * If labeling is NOT enabled, then disallow the mount of datasets
1366 * which have a non-default label already. No other label checks
1369 if (!is_system_labeled()) {
1370 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1376 * Get the label of the mountpoint. If mounting into the global
1377 * zone (i.e. mountpoint is not within an active zone and the
1378 * zoned property is off), the label must be default or
1379 * admin_low/admin_high only; no other checks are needed.
1381 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1382 if (mntzone->zone_id == GLOBAL_ZONEID) {
1387 if (dsl_prop_get_integer(osname,
1388 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1391 return (zfs_check_global_label(osname, ds_hexsl));
1394 * This is the case of a zone dataset being mounted
1395 * initially, before the zone has been fully created;
1396 * allow this mount into global zone.
1401 mnt_tsl = mntzone->zone_slabel;
1402 ASSERT(mnt_tsl != NULL);
1403 label_hold(mnt_tsl);
1404 mnt_sl = label2bslabel(mnt_tsl);
1406 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1408 * The dataset doesn't have a real label, so fabricate one.
1412 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1413 dsl_prop_set(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1414 ZPROP_SRC_LOCAL, 1, strlen(str) + 1, str) == 0)
1417 kmem_free(str, strlen(str) + 1);
1418 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1420 * Now compare labels to complete the MAC check. If the
1421 * labels are equal then allow access. If the mountpoint
1422 * label dominates the dataset label, allow readonly access.
1423 * Otherwise, access is denied.
1425 if (blequal(mnt_sl, &ds_sl))
1427 else if (bldominates(mnt_sl, &ds_sl)) {
1428 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1433 label_rele(mnt_tsl);
1439 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1442 static int zfsrootdone = 0;
1443 zfsvfs_t *zfsvfs = NULL;
1452 * The filesystem that we mount as root is defined in the
1453 * boot property "zfs-bootfs" with a format of
1454 * "poolname/root-dataset-objnum".
1456 if (why == ROOT_INIT) {
1460 * the process of doing a spa_load will require the
1461 * clock to be set before we could (for example) do
1462 * something better by looking at the timestamp on
1463 * an uberblock, so just set it to -1.
1467 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1468 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1472 zfs_devid = spa_get_bootprop("diskdevid");
1473 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1475 spa_free_bootprop(zfs_devid);
1477 spa_free_bootprop(zfs_bootfs);
1478 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1482 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1483 spa_free_bootprop(zfs_bootfs);
1484 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1489 spa_free_bootprop(zfs_bootfs);
1491 if (error = vfs_lock(vfsp))
1494 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1495 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1499 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1501 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1502 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1507 mutex_enter(&vp->v_lock);
1508 vp->v_flag |= VROOT;
1509 mutex_exit(&vp->v_lock);
1513 * Leave rootvp held. The root file system is never unmounted.
1516 vfs_add((struct vnode *)0, vfsp,
1517 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1521 } else if (why == ROOT_REMOUNT) {
1522 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1523 vfsp->vfs_flag |= VFS_REMOUNT;
1525 /* refresh mount options */
1526 zfs_unregister_callbacks(vfsp->vfs_data);
1527 return (zfs_register_callbacks(vfsp));
1529 } else if (why == ROOT_UNMOUNT) {
1530 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1531 (void) zfs_sync(vfsp, 0, 0);
1536 * if "why" is equal to anything else other than ROOT_INIT,
1537 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1544 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1549 uio_seg_t fromspace = (uap->flags & MS_SYSSPACE) ?
1550 UIO_SYSSPACE : UIO_USERSPACE;
1553 if (mvp->v_type != VDIR)
1556 mutex_enter(&mvp->v_lock);
1557 if ((uap->flags & MS_REMOUNT) == 0 &&
1558 (uap->flags & MS_OVERLAY) == 0 &&
1559 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1560 mutex_exit(&mvp->v_lock);
1563 mutex_exit(&mvp->v_lock);
1566 * ZFS does not support passing unparsed data in via MS_DATA.
1567 * Users should use the MS_OPTIONSTR interface; this means
1568 * that all option parsing is already done and the options struct
1569 * can be interrogated.
1571 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1575 * Get the objset name (the "special" mount argument).
1577 if ((error = pn_get(uap->spec, fromspace, &spn)))
1580 osname = spn.pn_path;
1583 * Check for mount privilege?
1585 * If we don't have privilege then see if
1586 * we have local permission to allow it
1588 error = secpolicy_fs_mount(cr, mvp, vfsp);
1590 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1594 * Make sure user is the owner of the mount point
1595 * or has sufficient privileges.
1598 vattr.va_mask = AT_UID;
1600 if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1604 if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1605 VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1608 secpolicy_fs_mount_clearopts(cr, vfsp);
1615 * Refuse to mount a filesystem if we are in a local zone and the
1616 * dataset is not visible.
1618 if (!INGLOBALZONE(curproc) &&
1619 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1624 error = zfs_mount_label_policy(vfsp, osname);
1629 * When doing a remount, we simply refresh our temporary properties
1630 * according to those options set in the current VFS options.
1632 if (uap->flags & MS_REMOUNT) {
1633 /* refresh mount options */
1634 zfs_unregister_callbacks(vfsp->vfs_data);
1635 error = zfs_register_callbacks(vfsp);
1639 error = zfs_domount(vfsp, osname);
1642 * Add an extra VFS_HOLD on our parent vfs so that it can't
1643 * disappear due to a forced unmount.
1645 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1646 VFS_HOLD(mvp->v_vfsp);
1654 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1656 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1658 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1662 dmu_objset_space(zfsvfs->z_os,
1663 &refdbytes, &availbytes, &usedobjs, &availobjs);
1666 * The underlying storage pool actually uses multiple block sizes.
1667 * We report the fragsize as the smallest block size we support,
1668 * and we report our blocksize as the filesystem's maximum blocksize.
1670 statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1671 statp->f_bsize = zfsvfs->z_max_blksz;
1674 * The following report "total" blocks of various kinds in the
1675 * file system, but reported in terms of f_frsize - the
1679 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1680 statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1681 statp->f_bavail = statp->f_bfree; /* no root reservation */
1684 * statvfs() should really be called statufs(), because it assumes
1685 * static metadata. ZFS doesn't preallocate files, so the best
1686 * we can do is report the max that could possibly fit in f_files,
1687 * and that minus the number actually used in f_ffree.
1688 * For f_ffree, report the smaller of the number of object available
1689 * and the number of blocks (each object will take at least a block).
1691 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1692 statp->f_favail = statp->f_ffree; /* no "root reservation" */
1693 statp->f_files = statp->f_ffree + usedobjs;
1695 (void) cmpldev(&d32, vfsp->vfs_dev);
1696 statp->f_fsid = d32;
1699 * We're a zfs filesystem.
1701 (void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1703 statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1705 statp->f_namemax = ZFS_MAXNAMELEN;
1708 * We have all of 32 characters to stuff a string here.
1709 * Is there anything useful we could/should provide?
1711 bzero(statp->f_fstr, sizeof (statp->f_fstr));
1716 EXPORT_SYMBOL(zfs_statvfs);
1719 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1721 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1727 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1729 *vpp = ZTOV(rootzp);
1734 EXPORT_SYMBOL(zfs_root);
1737 * Teardown the zfsvfs::z_os.
1739 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1740 * and 'z_teardown_inactive_lock' held.
1743 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1747 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1751 * We purge the parent filesystem's vfsp as the parent
1752 * filesystem and all of its snapshots have their vnode's
1753 * v_vfsp set to the parent's filesystem's vfsp. Note,
1754 * 'z_parent' is self referential for non-snapshots.
1756 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1760 * Close the zil. NB: Can't close the zil while zfs_inactive
1761 * threads are blocked as zil_close can call zfs_inactive.
1763 if (zfsvfs->z_log) {
1764 zil_close(zfsvfs->z_log);
1765 zfsvfs->z_log = NULL;
1768 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1771 * If we are not unmounting (ie: online recv) and someone already
1772 * unmounted this file system while we were doing the switcheroo,
1773 * or a reopen of z_os failed then just bail out now.
1775 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1776 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1777 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1782 * At this point there are no vops active, and any new vops will
1783 * fail with EIO since we have z_teardown_lock for writer (only
1784 * relavent for forced unmount).
1786 * Release all holds on dbufs.
1788 mutex_enter(&zfsvfs->z_znodes_lock);
1789 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1790 zp = list_next(&zfsvfs->z_all_znodes, zp))
1792 ASSERT(ZTOV(zp)->v_count > 0);
1793 zfs_znode_dmu_fini(zp);
1795 mutex_exit(&zfsvfs->z_znodes_lock);
1798 * If we are unmounting, set the unmounted flag and let new vops
1799 * unblock. zfs_inactive will have the unmounted behavior, and all
1800 * other vops will fail with EIO.
1803 zfsvfs->z_unmounted = B_TRUE;
1804 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1805 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1809 * z_os will be NULL if there was an error in attempting to reopen
1810 * zfsvfs, so just return as the properties had already been
1811 * unregistered and cached data had been evicted before.
1813 if (zfsvfs->z_os == NULL)
1817 * Unregister properties.
1819 zfs_unregister_callbacks(zfsvfs);
1824 if (dmu_objset_is_dirty_anywhere(zfsvfs->z_os))
1825 if (!(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1826 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1827 (void) dmu_objset_evict_dbufs(zfsvfs->z_os);
1834 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1836 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1840 ret = secpolicy_fs_unmount(cr, vfsp);
1842 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1843 ZFS_DELEG_PERM_MOUNT, cr))
1848 * We purge the parent filesystem's vfsp as the parent filesystem
1849 * and all of its snapshots have their vnode's v_vfsp set to the
1850 * parent's filesystem's vfsp. Note, 'z_parent' is self
1851 * referential for non-snapshots.
1853 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1856 * Unmount any snapshots mounted under .zfs before unmounting the
1859 if (zfsvfs->z_ctldir != NULL &&
1860 (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1864 if (!(fflag & MS_FORCE)) {
1866 * Check the number of active vnodes in the file system.
1867 * Our count is maintained in the vfs structure, but the
1868 * number is off by 1 to indicate a hold on the vfs
1871 * The '.zfs' directory maintains a reference of its
1872 * own, and any active references underneath are
1873 * reflected in the vnode count.
1875 if (zfsvfs->z_ctldir == NULL) {
1876 if (vfsp->vfs_count > 1)
1879 if (vfsp->vfs_count > 2 ||
1880 zfsvfs->z_ctldir->v_count > 1)
1885 vfsp->vfs_flag |= VFS_UNMOUNTED;
1887 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1891 * z_os will be NULL if there was an error in
1892 * attempting to reopen zfsvfs.
1896 * Unset the objset user_ptr.
1898 mutex_enter(&os->os_user_ptr_lock);
1899 dmu_objset_set_user(os, NULL);
1900 mutex_exit(&os->os_user_ptr_lock);
1903 * Finally release the objset
1905 dmu_objset_disown(os, zfsvfs);
1909 * We can now safely destroy the '.zfs' directory node.
1911 if (zfsvfs->z_ctldir != NULL)
1912 zfsctl_destroy(zfsvfs);
1916 EXPORT_SYMBOL(zfs_umount);
1919 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1921 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1923 uint64_t object = 0;
1924 uint64_t fid_gen = 0;
1933 if (fidp->fid_len == LONG_FID_LEN) {
1934 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1935 uint64_t objsetid = 0;
1936 uint64_t setgen = 0;
1938 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1939 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1941 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1942 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1946 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1952 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1953 zfid_short_t *zfid = (zfid_short_t *)fidp;
1955 for (i = 0; i < sizeof (zfid->zf_object); i++)
1956 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1958 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1959 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1965 /* A zero fid_gen means we are in the .zfs control directories */
1967 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1968 *vpp = zfsvfs->z_ctldir;
1969 ASSERT(*vpp != NULL);
1970 if (object == ZFSCTL_INO_SNAPDIR) {
1971 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
1972 0, NULL, NULL, NULL, NULL, NULL) == 0);
1980 gen_mask = -1ULL >> (64 - 8 * i);
1982 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1983 if ((err = zfs_zget(zfsvfs, object, &zp))) {
1987 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1989 zp_gen = zp_gen & gen_mask;
1992 if (zp->z_unlinked || zp_gen != fid_gen) {
1993 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2003 EXPORT_SYMBOL(zfs_vget);
2006 * Block out VOPs and close zfsvfs_t::z_os
2008 * Note, if successful, then we return with the 'z_teardown_lock' and
2009 * 'z_teardown_inactive_lock' write held.
2012 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2016 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2018 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
2022 EXPORT_SYMBOL(zfs_suspend_fs);
2025 * Reopen zfsvfs_t::z_os and release VOPs.
2028 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2032 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
2033 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2035 err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zfsvfs,
2038 zfsvfs->z_os = NULL;
2041 uint64_t sa_obj = 0;
2043 err2 = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2044 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2046 if ((err || err2) && zfsvfs->z_version >= ZPL_VERSION_SA)
2050 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2051 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
2054 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2057 * Attempt to re-establish all the active znodes with
2058 * their dbufs. If a zfs_rezget() fails, then we'll let
2059 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2060 * when they try to use their znode.
2062 mutex_enter(&zfsvfs->z_znodes_lock);
2063 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2064 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2065 (void) zfs_rezget(zp);
2067 mutex_exit(&zfsvfs->z_znodes_lock);
2072 /* release the VOPs */
2073 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2074 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
2078 * Since we couldn't reopen zfsvfs::z_os, force
2079 * unmount this file system.
2081 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2082 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2086 EXPORT_SYMBOL(zfs_resume_fs);
2089 zfs_freevfs(vfs_t *vfsp)
2091 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2094 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2095 * from zfs_mount(). Release it here. If we came through
2096 * zfs_mountroot() instead, we didn't grab an extra hold, so
2097 * skip the VFS_RELE for rootvfs.
2099 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2100 VFS_RELE(zfsvfs->z_parent->z_vfs);
2102 zfsvfs_free(zfsvfs);
2104 atomic_add_32(&zfs_active_fs_count, -1);
2108 * VFS_INIT() initialization. Note that there is no VFS_FINI(),
2109 * so we can't safely do any non-idempotent initialization here.
2110 * Leave that to zfs_init() and zfs_fini(), which are called
2111 * from the module's _init() and _fini() entry points.
2115 zfs_vfsinit(int fstype, char *name)
2122 * Setup vfsops and vnodeops tables.
2124 error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2126 cmn_err(CE_WARN, "zfs: bad vfs ops template");
2129 error = zfs_create_op_tables();
2131 zfs_remove_op_tables();
2132 cmn_err(CE_WARN, "zfs: bad vnode ops template");
2133 (void) vfs_freevfsops_by_type(zfsfstype);
2137 mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2140 * Unique major number for all zfs mounts.
2141 * If we run out of 32-bit minors, we'll getudev() another major.
2143 zfs_major = ddi_name_to_major(ZFS_DRIVER);
2144 zfs_minor = ZFS_MIN_MINOR;
2148 #endif /* HAVE_ZPL */
2155 * Initialize .zfs directory structures
2160 * Initialize znode cache, vnode ops, etc...
2163 #endif /* HAVE_ZPL */
2165 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2174 #endif /* HAVE_ZPL */
2179 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2182 objset_t *os = zfsvfs->z_os;
2185 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2188 if (newvers < zfsvfs->z_version)
2191 if (zfs_spa_version_map(newvers) >
2192 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2195 tx = dmu_tx_create(os);
2196 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2197 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2198 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2200 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2202 error = dmu_tx_assign(tx, TXG_WAIT);
2208 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2209 8, 1, &newvers, tx);
2216 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2219 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2221 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2222 DMU_OT_NONE, 0, tx);
2224 error = zap_add(os, MASTER_NODE_OBJ,
2225 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2226 ASSERT3U(error, ==, 0);
2228 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2229 sa_register_update_callback(os, zfs_sa_upgrade);
2232 spa_history_log_internal(LOG_DS_UPGRADE,
2233 dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu",
2234 zfsvfs->z_version, newvers, dmu_objset_id(os));
2238 zfsvfs->z_version = newvers;
2240 if (zfsvfs->z_version >= ZPL_VERSION_FUID)
2241 zfs_set_fuid_feature(zfsvfs);
2245 EXPORT_SYMBOL(zfs_set_version);
2246 #endif /* HAVE_ZPL */
2249 * Read a property stored within the master node.
2252 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2258 * Look up the file system's value for the property. For the
2259 * version property, we look up a slightly different string.
2261 if (prop == ZFS_PROP_VERSION)
2262 pname = ZPL_VERSION_STR;
2264 pname = zfs_prop_to_name(prop);
2267 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2269 if (error == ENOENT) {
2270 /* No value set, use the default value */
2272 case ZFS_PROP_VERSION:
2273 *value = ZPL_VERSION;
2275 case ZFS_PROP_NORMALIZE:
2276 case ZFS_PROP_UTF8ONLY:
2280 *value = ZFS_CASE_SENSITIVE;
2291 static vfsdef_t vfw = {
2295 VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2300 struct modlfs zfs_modlfs = {
2301 &mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2303 #endif /* HAVE_ZPL */