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) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
24 * All rights reserved.
25 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
29 /* Portions Copyright 2010 Robert Milkowski */
31 #include <sys/types.h>
32 #include <sys/param.h>
33 #include <sys/systm.h>
34 #include <sys/kernel.h>
35 #include <sys/sysmacros.h>
38 #include <sys/vnode.h>
40 #include <sys/mntent.h>
41 #include <sys/mount.h>
42 #include <sys/cmn_err.h>
43 #include <sys/zfs_znode.h>
44 #include <sys/zfs_dir.h>
46 #include <sys/fs/zfs.h>
48 #include <sys/dsl_prop.h>
49 #include <sys/dsl_dataset.h>
50 #include <sys/dsl_deleg.h>
54 #include <sys/sa_impl.h>
55 #include <sys/varargs.h>
56 #include <sys/policy.h>
57 #include <sys/atomic.h>
58 #include <sys/zfs_ioctl.h>
59 #include <sys/zfs_ctldir.h>
60 #include <sys/zfs_fuid.h>
61 #include <sys/sunddi.h>
63 #include <sys/dmu_objset.h>
64 #include <sys/spa_boot.h>
66 #include "zfs_comutil.h"
68 struct mtx zfs_debug_mtx;
69 MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF);
71 SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system");
74 SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0,
75 "File system owner can perform privileged operation on his file systems");
78 TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level);
79 SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0,
82 SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions");
83 static int zfs_version_acl = ZFS_ACL_VERSION;
84 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0,
86 static int zfs_version_spa = SPA_VERSION;
87 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0,
89 static int zfs_version_zpl = ZPL_VERSION;
90 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0,
93 static int zfs_mount(vfs_t *vfsp);
94 static int zfs_umount(vfs_t *vfsp, int fflag);
95 static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
96 static int zfs_statfs(vfs_t *vfsp, struct statfs *statp);
97 static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp);
98 static int zfs_sync(vfs_t *vfsp, int waitfor);
99 static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
100 struct ucred **credanonp, int *numsecflavors, int **secflavors);
101 static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp);
102 static void zfs_objset_close(zfsvfs_t *zfsvfs);
103 static void zfs_freevfs(vfs_t *vfsp);
105 static struct vfsops zfs_vfsops = {
106 .vfs_mount = zfs_mount,
107 .vfs_unmount = zfs_umount,
108 .vfs_root = zfs_root,
109 .vfs_statfs = zfs_statfs,
110 .vfs_vget = zfs_vget,
111 .vfs_sync = zfs_sync,
112 .vfs_checkexp = zfs_checkexp,
113 .vfs_fhtovp = zfs_fhtovp,
116 VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN);
119 * We need to keep a count of active fs's.
120 * This is necessary to prevent our module
121 * from being unloaded after a umount -f
123 static uint32_t zfs_active_fs_count = 0;
127 zfs_sync(vfs_t *vfsp, int waitfor)
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 * Ignore the system syncher. ZFS already commits async data
139 * at zfs_txg_timeout intervals.
141 if (waitfor == MNT_LAZY)
146 * Sync a specific filesystem.
148 zfsvfs_t *zfsvfs = vfsp->vfs_data;
152 error = vfs_stdsync(vfsp, waitfor);
157 dp = dmu_objset_pool(zfsvfs->z_os);
160 * If the system is shutting down, then skip any
161 * filesystems which may exist on a suspended pool.
163 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
168 if (zfsvfs->z_log != NULL)
169 zil_commit(zfsvfs->z_log, 0);
174 * Sync all ZFS filesystems. This is what happens when you
175 * run sync(1M). Unlike other filesystems, ZFS honors the
176 * request by waiting for all pools to commit all dirty data.
184 #ifndef __FreeBSD_kernel__
186 zfs_create_unique_device(dev_t *dev)
191 ASSERT3U(zfs_minor, <=, MAXMIN32);
192 minor_t start = zfs_minor;
194 mutex_enter(&zfs_dev_mtx);
195 if (zfs_minor >= MAXMIN32) {
197 * If we're still using the real major
198 * keep out of /dev/zfs and /dev/zvol minor
199 * number space. If we're using a getudev()'ed
200 * major number, we can use all of its minors.
202 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
203 zfs_minor = ZFS_MIN_MINOR;
209 *dev = makedevice(zfs_major, zfs_minor);
210 mutex_exit(&zfs_dev_mtx);
211 } while (vfs_devismounted(*dev) && zfs_minor != start);
212 if (zfs_minor == start) {
214 * We are using all ~262,000 minor numbers for the
215 * current major number. Create a new major number.
217 if ((new_major = getudev()) == (major_t)-1) {
219 "zfs_mount: Can't get unique major "
223 mutex_enter(&zfs_dev_mtx);
224 zfs_major = new_major;
227 mutex_exit(&zfs_dev_mtx);
231 /* CONSTANTCONDITION */
236 #endif /* !__FreeBSD_kernel__ */
239 atime_changed_cb(void *arg, uint64_t newval)
241 zfsvfs_t *zfsvfs = arg;
243 if (newval == TRUE) {
244 zfsvfs->z_atime = TRUE;
245 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
246 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
247 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
249 zfsvfs->z_atime = FALSE;
250 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
251 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
252 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
257 xattr_changed_cb(void *arg, uint64_t newval)
259 zfsvfs_t *zfsvfs = arg;
261 if (newval == TRUE) {
262 /* XXX locking on vfs_flag? */
264 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
266 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
267 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
269 /* XXX locking on vfs_flag? */
271 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
273 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
274 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
279 blksz_changed_cb(void *arg, uint64_t newval)
281 zfsvfs_t *zfsvfs = arg;
282 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
283 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
284 ASSERT(ISP2(newval));
286 zfsvfs->z_max_blksz = newval;
287 zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
291 readonly_changed_cb(void *arg, uint64_t newval)
293 zfsvfs_t *zfsvfs = arg;
296 /* XXX locking on vfs_flag? */
297 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
298 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
299 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
301 /* XXX locking on vfs_flag? */
302 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
303 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
304 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
309 setuid_changed_cb(void *arg, uint64_t newval)
311 zfsvfs_t *zfsvfs = arg;
313 if (newval == FALSE) {
314 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
315 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
316 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
318 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
319 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
320 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
325 exec_changed_cb(void *arg, uint64_t newval)
327 zfsvfs_t *zfsvfs = arg;
329 if (newval == FALSE) {
330 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
331 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
332 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
334 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
335 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
336 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
341 * The nbmand mount option can be changed at mount time.
342 * We can't allow it to be toggled on live file systems or incorrect
343 * behavior may be seen from cifs clients
345 * This property isn't registered via dsl_prop_register(), but this callback
346 * will be called when a file system is first mounted
349 nbmand_changed_cb(void *arg, uint64_t newval)
351 zfsvfs_t *zfsvfs = arg;
352 if (newval == FALSE) {
353 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
354 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
356 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
357 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
362 snapdir_changed_cb(void *arg, uint64_t newval)
364 zfsvfs_t *zfsvfs = arg;
366 zfsvfs->z_show_ctldir = newval;
370 vscan_changed_cb(void *arg, uint64_t newval)
372 zfsvfs_t *zfsvfs = arg;
374 zfsvfs->z_vscan = newval;
378 acl_mode_changed_cb(void *arg, uint64_t newval)
380 zfsvfs_t *zfsvfs = arg;
382 zfsvfs->z_acl_mode = newval;
386 acl_inherit_changed_cb(void *arg, uint64_t newval)
388 zfsvfs_t *zfsvfs = arg;
390 zfsvfs->z_acl_inherit = newval;
394 zfs_register_callbacks(vfs_t *vfsp)
396 struct dsl_dataset *ds = NULL;
398 zfsvfs_t *zfsvfs = NULL;
400 boolean_t readonly = B_FALSE;
401 boolean_t do_readonly = B_FALSE;
402 boolean_t setuid = B_FALSE;
403 boolean_t do_setuid = B_FALSE;
404 boolean_t exec = B_FALSE;
405 boolean_t do_exec = B_FALSE;
407 boolean_t devices = B_FALSE;
408 boolean_t do_devices = B_FALSE;
410 boolean_t xattr = B_FALSE;
411 boolean_t do_xattr = B_FALSE;
412 boolean_t atime = B_FALSE;
413 boolean_t do_atime = B_FALSE;
417 zfsvfs = vfsp->vfs_data;
422 * This function can be called for a snapshot when we update snapshot's
423 * mount point, which isn't really supported.
425 if (dmu_objset_is_snapshot(os))
429 * The act of registering our callbacks will destroy any mount
430 * options we may have. In order to enable temporary overrides
431 * of mount options, we stash away the current values and
432 * restore them after we register the callbacks.
434 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
435 !spa_writeable(dmu_objset_spa(os))) {
437 do_readonly = B_TRUE;
438 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
440 do_readonly = B_TRUE;
442 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
446 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
449 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
454 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
457 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
461 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
464 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
468 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
471 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
477 * We need to enter pool configuration here, so that we can use
478 * dsl_prop_get_int_ds() to handle the special nbmand property below.
479 * dsl_prop_get_integer() can not be used, because it has to acquire
480 * spa_namespace_lock and we can not do that because we already hold
481 * z_teardown_lock. The problem is that spa_config_sync() is called
482 * with spa_namespace_lock held and the function calls ZFS vnode
483 * operations to write the cache file and thus z_teardown_lock is
484 * acquired after spa_namespace_lock.
486 ds = dmu_objset_ds(os);
487 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
490 * nbmand is a special property. It can only be changed at
493 * This is weird, but it is documented to only be changeable
496 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
498 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
500 } else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) {
501 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
506 * Register property callbacks.
508 * It would probably be fine to just check for i/o error from
509 * the first prop_register(), but I guess I like to go
512 error = dsl_prop_register(ds,
513 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
514 error = error ? error : dsl_prop_register(ds,
515 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
516 error = error ? error : dsl_prop_register(ds,
517 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
518 error = error ? error : dsl_prop_register(ds,
519 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
521 error = error ? error : dsl_prop_register(ds,
522 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
524 error = error ? error : dsl_prop_register(ds,
525 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
526 error = error ? error : dsl_prop_register(ds,
527 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
528 error = error ? error : dsl_prop_register(ds,
529 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
530 error = error ? error : dsl_prop_register(ds,
531 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
532 error = error ? error : dsl_prop_register(ds,
533 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
535 error = error ? error : dsl_prop_register(ds,
536 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
537 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
542 * Invoke our callbacks to restore temporary mount options.
545 readonly_changed_cb(zfsvfs, readonly);
547 setuid_changed_cb(zfsvfs, setuid);
549 exec_changed_cb(zfsvfs, exec);
551 xattr_changed_cb(zfsvfs, xattr);
553 atime_changed_cb(zfsvfs, atime);
555 nbmand_changed_cb(zfsvfs, nbmand);
560 dsl_prop_unregister_all(ds, zfsvfs);
565 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
566 uint64_t *userp, uint64_t *groupp)
569 * Is it a valid type of object to track?
571 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
572 return (SET_ERROR(ENOENT));
575 * If we have a NULL data pointer
576 * then assume the id's aren't changing and
577 * return EEXIST to the dmu to let it know to
581 return (SET_ERROR(EEXIST));
583 if (bonustype == DMU_OT_ZNODE) {
584 znode_phys_t *znp = data;
585 *userp = znp->zp_uid;
586 *groupp = znp->zp_gid;
589 sa_hdr_phys_t *sap = data;
590 sa_hdr_phys_t sa = *sap;
591 boolean_t swap = B_FALSE;
593 ASSERT(bonustype == DMU_OT_SA);
595 if (sa.sa_magic == 0) {
597 * This should only happen for newly created
598 * files that haven't had the znode data filled
605 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
606 sa.sa_magic = SA_MAGIC;
607 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
610 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
613 hdrsize = sa_hdrsize(&sa);
614 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
615 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
617 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
620 *userp = BSWAP_64(*userp);
621 *groupp = BSWAP_64(*groupp);
628 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
629 char *domainbuf, int buflen, uid_t *ridp)
634 fuid = strtonum(fuidstr, NULL);
636 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
638 (void) strlcpy(domainbuf, domain, buflen);
641 *ridp = FUID_RID(fuid);
645 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
648 case ZFS_PROP_USERUSED:
649 return (DMU_USERUSED_OBJECT);
650 case ZFS_PROP_GROUPUSED:
651 return (DMU_GROUPUSED_OBJECT);
652 case ZFS_PROP_USERQUOTA:
653 return (zfsvfs->z_userquota_obj);
654 case ZFS_PROP_GROUPQUOTA:
655 return (zfsvfs->z_groupquota_obj);
661 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
662 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
667 zfs_useracct_t *buf = vbuf;
670 if (!dmu_objset_userspace_present(zfsvfs->z_os))
671 return (SET_ERROR(ENOTSUP));
673 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
679 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
680 (error = zap_cursor_retrieve(&zc, &za)) == 0;
681 zap_cursor_advance(&zc)) {
682 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
686 fuidstr_to_sid(zfsvfs, za.za_name,
687 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
689 buf->zu_space = za.za_first_integer;
695 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
696 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
697 *cookiep = zap_cursor_serialize(&zc);
698 zap_cursor_fini(&zc);
703 * buf must be big enough (eg, 32 bytes)
706 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
707 char *buf, boolean_t addok)
712 if (domain && domain[0]) {
713 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
715 return (SET_ERROR(ENOENT));
717 fuid = FUID_ENCODE(domainid, rid);
718 (void) sprintf(buf, "%llx", (longlong_t)fuid);
723 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
724 const char *domain, uint64_t rid, uint64_t *valp)
732 if (!dmu_objset_userspace_present(zfsvfs->z_os))
733 return (SET_ERROR(ENOTSUP));
735 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
739 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
743 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
750 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
751 const char *domain, uint64_t rid, uint64_t quota)
757 boolean_t fuid_dirtied;
759 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
760 return (SET_ERROR(EINVAL));
762 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
763 return (SET_ERROR(ENOTSUP));
765 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
766 &zfsvfs->z_groupquota_obj;
768 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
771 fuid_dirtied = zfsvfs->z_fuid_dirty;
773 tx = dmu_tx_create(zfsvfs->z_os);
774 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
776 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
777 zfs_userquota_prop_prefixes[type]);
780 zfs_fuid_txhold(zfsvfs, tx);
781 err = dmu_tx_assign(tx, TXG_WAIT);
787 mutex_enter(&zfsvfs->z_lock);
789 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
791 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
792 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
794 mutex_exit(&zfsvfs->z_lock);
797 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
801 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
805 zfs_fuid_sync(zfsvfs, tx);
811 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
814 uint64_t used, quota, usedobj, quotaobj;
817 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
818 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
820 if (quotaobj == 0 || zfsvfs->z_replay)
823 (void) sprintf(buf, "%llx", (longlong_t)fuid);
824 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
828 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
831 return (used >= quota);
835 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
840 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
842 fuid = isgroup ? zp->z_gid : zp->z_uid;
844 if (quotaobj == 0 || zfsvfs->z_replay)
847 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
851 * Associate this zfsvfs with the given objset, which must be owned.
852 * This will cache a bunch of on-disk state from the objset in the
856 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
861 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
862 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
865 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
868 if (zfsvfs->z_version >
869 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
870 (void) printf("Can't mount a version %lld file system "
871 "on a version %lld pool\n. Pool must be upgraded to mount "
872 "this file system.", (u_longlong_t)zfsvfs->z_version,
873 (u_longlong_t)spa_version(dmu_objset_spa(os)));
874 return (SET_ERROR(ENOTSUP));
876 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
879 zfsvfs->z_norm = (int)val;
881 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
884 zfsvfs->z_utf8 = (val != 0);
886 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
889 zfsvfs->z_case = (uint_t)val;
892 * Fold case on file systems that are always or sometimes case
895 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
896 zfsvfs->z_case == ZFS_CASE_MIXED)
897 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
899 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
900 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
903 if (zfsvfs->z_use_sa) {
904 /* should either have both of these objects or none */
905 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
911 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
912 &zfsvfs->z_attr_table);
916 if (zfsvfs->z_version >= ZPL_VERSION_SA)
917 sa_register_update_callback(os, zfs_sa_upgrade);
919 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
923 ASSERT(zfsvfs->z_root != 0);
925 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
926 &zfsvfs->z_unlinkedobj);
930 error = zap_lookup(os, MASTER_NODE_OBJ,
931 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
932 8, 1, &zfsvfs->z_userquota_obj);
934 zfsvfs->z_userquota_obj = 0;
938 error = zap_lookup(os, MASTER_NODE_OBJ,
939 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
940 8, 1, &zfsvfs->z_groupquota_obj);
942 zfsvfs->z_groupquota_obj = 0;
946 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
947 &zfsvfs->z_fuid_obj);
949 zfsvfs->z_fuid_obj = 0;
953 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
954 &zfsvfs->z_shares_dir);
956 zfsvfs->z_shares_dir = 0;
961 * Only use the name cache if we are looking for a
962 * name on a file system that does not require normalization
963 * or case folding. We can also look there if we happen to be
964 * on a non-normalizing, mixed sensitivity file system IF we
965 * are looking for the exact name (which is always the case on
968 zfsvfs->z_use_namecache = !zfsvfs->z_norm ||
969 ((zfsvfs->z_case == ZFS_CASE_MIXED) &&
970 !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER));
976 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
983 * XXX: Fix struct statfs so this isn't necessary!
985 * The 'osname' is used as the filesystem's special node, which means
986 * it must fit in statfs.f_mntfromname, or else it can't be
987 * enumerated, so libzfs_mnttab_find() returns NULL, which causes
988 * 'zfs unmount' to think it's not mounted when it is.
990 if (strlen(osname) >= MNAMELEN)
991 return (SET_ERROR(ENAMETOOLONG));
993 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
996 * We claim to always be readonly so we can open snapshots;
997 * other ZPL code will prevent us from writing to snapshots.
999 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
1001 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1005 zfsvfs->z_vfs = NULL;
1006 zfsvfs->z_parent = zfsvfs;
1008 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1009 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1010 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1011 offsetof(znode_t, z_link_node));
1013 rrm_init(&zfsvfs->z_teardown_lock, B_TRUE);
1015 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1017 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1018 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1019 for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1020 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1022 error = zfsvfs_init(zfsvfs, os);
1024 dmu_objset_disown(os, zfsvfs);
1026 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1035 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1039 error = zfs_register_callbacks(zfsvfs->z_vfs);
1043 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1046 * If we are not mounting (ie: online recv), then we don't
1047 * have to worry about replaying the log as we blocked all
1048 * operations out since we closed the ZIL.
1054 * During replay we remove the read only flag to
1055 * allow replays to succeed.
1057 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1059 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1061 zfs_unlinked_drain(zfsvfs);
1064 * Parse and replay the intent log.
1066 * Because of ziltest, this must be done after
1067 * zfs_unlinked_drain(). (Further note: ziltest
1068 * doesn't use readonly mounts, where
1069 * zfs_unlinked_drain() isn't called.) This is because
1070 * ziltest causes spa_sync() to think it's committed,
1071 * but actually it is not, so the intent log contains
1072 * many txg's worth of changes.
1074 * In particular, if object N is in the unlinked set in
1075 * the last txg to actually sync, then it could be
1076 * actually freed in a later txg and then reallocated
1077 * in a yet later txg. This would write a "create
1078 * object N" record to the intent log. Normally, this
1079 * would be fine because the spa_sync() would have
1080 * written out the fact that object N is free, before
1081 * we could write the "create object N" intent log
1084 * But when we are in ziltest mode, we advance the "open
1085 * txg" without actually spa_sync()-ing the changes to
1086 * disk. So we would see that object N is still
1087 * allocated and in the unlinked set, and there is an
1088 * intent log record saying to allocate it.
1090 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1091 if (zil_replay_disable) {
1092 zil_destroy(zfsvfs->z_log, B_FALSE);
1094 zfsvfs->z_replay = B_TRUE;
1095 zil_replay(zfsvfs->z_os, zfsvfs,
1097 zfsvfs->z_replay = B_FALSE;
1100 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1104 * Set the objset user_ptr to track its zfsvfs.
1106 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1107 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1108 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1113 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1116 zfsvfs_free(zfsvfs_t *zfsvfs)
1121 * This is a barrier to prevent the filesystem from going away in
1122 * zfs_znode_move() until we can safely ensure that the filesystem is
1123 * not unmounted. We consider the filesystem valid before the barrier
1124 * and invalid after the barrier.
1126 rw_enter(&zfsvfs_lock, RW_READER);
1127 rw_exit(&zfsvfs_lock);
1129 zfs_fuid_destroy(zfsvfs);
1131 mutex_destroy(&zfsvfs->z_znodes_lock);
1132 mutex_destroy(&zfsvfs->z_lock);
1133 list_destroy(&zfsvfs->z_all_znodes);
1134 rrm_destroy(&zfsvfs->z_teardown_lock);
1135 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1136 rw_destroy(&zfsvfs->z_fuid_lock);
1137 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1138 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1139 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1143 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1145 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1146 if (zfsvfs->z_vfs) {
1147 if (zfsvfs->z_use_fuids) {
1148 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1149 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1150 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1151 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1152 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1153 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1155 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1156 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1157 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1158 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1159 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1160 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1163 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1167 zfs_domount(vfs_t *vfsp, char *osname)
1169 uint64_t recordsize, fsid_guid;
1177 error = zfsvfs_create(osname, &zfsvfs);
1180 zfsvfs->z_vfs = vfsp;
1183 /* Initialize the generic filesystem structure. */
1184 vfsp->vfs_bcount = 0;
1185 vfsp->vfs_data = NULL;
1187 if (zfs_create_unique_device(&mount_dev) == -1) {
1188 error = SET_ERROR(ENODEV);
1191 ASSERT(vfs_devismounted(mount_dev) == 0);
1194 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1197 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1198 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1200 vfsp->vfs_data = zfsvfs;
1201 vfsp->mnt_flag |= MNT_LOCAL;
1202 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1203 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1204 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1205 vfsp->mnt_kern_flag |= MNTK_NO_IOPF; /* vn_io_fault can be used */
1208 * The fsid is 64 bits, composed of an 8-bit fs type, which
1209 * separates our fsid from any other filesystem types, and a
1210 * 56-bit objset unique ID. The objset unique ID is unique to
1211 * all objsets open on this system, provided by unique_create().
1212 * The 8-bit fs type must be put in the low bits of fsid[1]
1213 * because that's where other Solaris filesystems put it.
1215 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1216 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1217 vfsp->vfs_fsid.val[0] = fsid_guid;
1218 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1219 vfsp->mnt_vfc->vfc_typenum & 0xFF;
1222 * Set features for file system.
1224 zfs_set_fuid_feature(zfsvfs);
1225 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1226 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1227 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1228 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1229 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1230 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1231 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1233 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1235 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1238 atime_changed_cb(zfsvfs, B_FALSE);
1239 readonly_changed_cb(zfsvfs, B_TRUE);
1240 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1242 xattr_changed_cb(zfsvfs, pval);
1243 zfsvfs->z_issnap = B_TRUE;
1244 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1246 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1247 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1248 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1250 error = zfsvfs_setup(zfsvfs, B_TRUE);
1253 vfs_mountedfrom(vfsp, osname);
1255 if (!zfsvfs->z_issnap)
1256 zfsctl_create(zfsvfs);
1259 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1260 zfsvfs_free(zfsvfs);
1262 atomic_inc_32(&zfs_active_fs_count);
1269 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1271 objset_t *os = zfsvfs->z_os;
1273 if (!dmu_objset_is_snapshot(os))
1274 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1279 * Convert a decimal digit string to a uint64_t integer.
1282 str_to_uint64(char *str, uint64_t *objnum)
1287 if (*str < '0' || *str > '9')
1288 return (SET_ERROR(EINVAL));
1290 num = num*10 + *str++ - '0';
1298 * The boot path passed from the boot loader is in the form of
1299 * "rootpool-name/root-filesystem-object-number'. Convert this
1300 * string to a dataset name: "rootpool-name/root-filesystem-name".
1303 zfs_parse_bootfs(char *bpath, char *outpath)
1309 if (*bpath == 0 || *bpath == '/')
1310 return (SET_ERROR(EINVAL));
1312 (void) strcpy(outpath, bpath);
1314 slashp = strchr(bpath, '/');
1316 /* if no '/', just return the pool name */
1317 if (slashp == NULL) {
1321 /* if not a number, just return the root dataset name */
1322 if (str_to_uint64(slashp+1, &objnum)) {
1327 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1334 * Check that the hex label string is appropriate for the dataset being
1335 * mounted into the global_zone proper.
1337 * Return an error if the hex label string is not default or
1338 * admin_low/admin_high. For admin_low labels, the corresponding
1339 * dataset must be readonly.
1342 zfs_check_global_label(const char *dsname, const char *hexsl)
1344 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1346 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1348 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1349 /* must be readonly */
1352 if (dsl_prop_get_integer(dsname,
1353 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1354 return (SET_ERROR(EACCES));
1355 return (rdonly ? 0 : EACCES);
1357 return (SET_ERROR(EACCES));
1361 * Determine whether the mount is allowed according to MAC check.
1362 * by comparing (where appropriate) label of the dataset against
1363 * the label of the zone being mounted into. If the dataset has
1364 * no label, create one.
1366 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1369 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1372 zone_t *mntzone = NULL;
1373 ts_label_t *mnt_tsl;
1376 char ds_hexsl[MAXNAMELEN];
1378 retv = EACCES; /* assume the worst */
1381 * Start by getting the dataset label if it exists.
1383 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1384 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1386 return (SET_ERROR(EACCES));
1389 * If labeling is NOT enabled, then disallow the mount of datasets
1390 * which have a non-default label already. No other label checks
1393 if (!is_system_labeled()) {
1394 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1396 return (SET_ERROR(EACCES));
1400 * Get the label of the mountpoint. If mounting into the global
1401 * zone (i.e. mountpoint is not within an active zone and the
1402 * zoned property is off), the label must be default or
1403 * admin_low/admin_high only; no other checks are needed.
1405 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1406 if (mntzone->zone_id == GLOBAL_ZONEID) {
1411 if (dsl_prop_get_integer(osname,
1412 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1413 return (SET_ERROR(EACCES));
1415 return (zfs_check_global_label(osname, ds_hexsl));
1418 * This is the case of a zone dataset being mounted
1419 * initially, before the zone has been fully created;
1420 * allow this mount into global zone.
1425 mnt_tsl = mntzone->zone_slabel;
1426 ASSERT(mnt_tsl != NULL);
1427 label_hold(mnt_tsl);
1428 mnt_sl = label2bslabel(mnt_tsl);
1430 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1432 * The dataset doesn't have a real label, so fabricate one.
1436 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1437 dsl_prop_set_string(osname,
1438 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1439 ZPROP_SRC_LOCAL, str) == 0)
1442 kmem_free(str, strlen(str) + 1);
1443 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1445 * Now compare labels to complete the MAC check. If the
1446 * labels are equal then allow access. If the mountpoint
1447 * label dominates the dataset label, allow readonly access.
1448 * Otherwise, access is denied.
1450 if (blequal(mnt_sl, &ds_sl))
1452 else if (bldominates(mnt_sl, &ds_sl)) {
1453 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1458 label_rele(mnt_tsl);
1462 #endif /* SECLABEL */
1464 #ifdef OPENSOLARIS_MOUNTROOT
1466 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1469 static int zfsrootdone = 0;
1470 zfsvfs_t *zfsvfs = NULL;
1479 * The filesystem that we mount as root is defined in the
1480 * boot property "zfs-bootfs" with a format of
1481 * "poolname/root-dataset-objnum".
1483 if (why == ROOT_INIT) {
1485 return (SET_ERROR(EBUSY));
1487 * the process of doing a spa_load will require the
1488 * clock to be set before we could (for example) do
1489 * something better by looking at the timestamp on
1490 * an uberblock, so just set it to -1.
1494 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1495 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1497 return (SET_ERROR(EINVAL));
1499 zfs_devid = spa_get_bootprop("diskdevid");
1500 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1502 spa_free_bootprop(zfs_devid);
1504 spa_free_bootprop(zfs_bootfs);
1505 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1509 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1510 spa_free_bootprop(zfs_bootfs);
1511 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1516 spa_free_bootprop(zfs_bootfs);
1518 if (error = vfs_lock(vfsp))
1521 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1522 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1526 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1528 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1529 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1534 mutex_enter(&vp->v_lock);
1535 vp->v_flag |= VROOT;
1536 mutex_exit(&vp->v_lock);
1540 * Leave rootvp held. The root file system is never unmounted.
1543 vfs_add((struct vnode *)0, vfsp,
1544 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1548 } else if (why == ROOT_REMOUNT) {
1549 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1550 vfsp->vfs_flag |= VFS_REMOUNT;
1552 /* refresh mount options */
1553 zfs_unregister_callbacks(vfsp->vfs_data);
1554 return (zfs_register_callbacks(vfsp));
1556 } else if (why == ROOT_UNMOUNT) {
1557 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1558 (void) zfs_sync(vfsp, 0, 0);
1563 * if "why" is equal to anything else other than ROOT_INIT,
1564 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1566 return (SET_ERROR(ENOTSUP));
1568 #endif /* OPENSOLARIS_MOUNTROOT */
1571 getpoolname(const char *osname, char *poolname)
1575 p = strchr(osname, '/');
1577 if (strlen(osname) >= MAXNAMELEN)
1578 return (ENAMETOOLONG);
1579 (void) strcpy(poolname, osname);
1581 if (p - osname >= MAXNAMELEN)
1582 return (ENAMETOOLONG);
1583 (void) strncpy(poolname, osname, p - osname);
1584 poolname[p - osname] = '\0';
1591 zfs_mount(vfs_t *vfsp)
1593 kthread_t *td = curthread;
1594 vnode_t *mvp = vfsp->mnt_vnodecovered;
1595 cred_t *cr = td->td_ucred;
1601 if (mvp->v_type != VDIR)
1602 return (SET_ERROR(ENOTDIR));
1604 mutex_enter(&mvp->v_lock);
1605 if ((uap->flags & MS_REMOUNT) == 0 &&
1606 (uap->flags & MS_OVERLAY) == 0 &&
1607 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1608 mutex_exit(&mvp->v_lock);
1609 return (SET_ERROR(EBUSY));
1611 mutex_exit(&mvp->v_lock);
1614 * ZFS does not support passing unparsed data in via MS_DATA.
1615 * Users should use the MS_OPTIONSTR interface; this means
1616 * that all option parsing is already done and the options struct
1617 * can be interrogated.
1619 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1620 return (SET_ERROR(EINVAL));
1623 * Get the objset name (the "special" mount argument).
1625 if (error = pn_get(uap->spec, fromspace, &spn))
1628 osname = spn.pn_path;
1629 #else /* !illumos */
1630 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS))
1631 return (SET_ERROR(EPERM));
1633 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1634 return (SET_ERROR(EINVAL));
1637 * If full-owner-access is enabled and delegated administration is
1638 * turned on, we must set nosuid.
1640 if (zfs_super_owner &&
1641 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1642 secpolicy_fs_mount_clearopts(cr, vfsp);
1644 #endif /* illumos */
1647 * Check for mount privilege?
1649 * If we don't have privilege then see if
1650 * we have local permission to allow it
1652 error = secpolicy_fs_mount(cr, mvp, vfsp);
1654 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1657 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1661 * Make sure user is the owner of the mount point
1662 * or has sufficient privileges.
1665 vattr.va_mask = AT_UID;
1667 vn_lock(mvp, LK_SHARED | LK_RETRY);
1668 if (VOP_GETATTR(mvp, &vattr, cr)) {
1673 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1674 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1681 secpolicy_fs_mount_clearopts(cr, vfsp);
1685 * Refuse to mount a filesystem if we are in a local zone and the
1686 * dataset is not visible.
1688 if (!INGLOBALZONE(curthread) &&
1689 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1690 error = SET_ERROR(EPERM);
1695 error = zfs_mount_label_policy(vfsp, osname);
1700 vfsp->vfs_flag |= MNT_NFS4ACLS;
1703 * When doing a remount, we simply refresh our temporary properties
1704 * according to those options set in the current VFS options.
1706 if (vfsp->vfs_flag & MS_REMOUNT) {
1707 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1710 * Refresh mount options with z_teardown_lock blocking I/O while
1711 * the filesystem is in an inconsistent state.
1712 * The lock also serializes this code with filesystem
1713 * manipulations between entry to zfs_suspend_fs() and return
1714 * from zfs_resume_fs().
1716 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1717 zfs_unregister_callbacks(zfsvfs);
1718 error = zfs_register_callbacks(vfsp);
1719 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1723 /* Initial root mount: try hard to import the requested root pool. */
1724 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1725 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1726 char pname[MAXNAMELEN];
1728 error = getpoolname(osname, pname);
1730 error = spa_import_rootpool(pname);
1735 error = zfs_domount(vfsp, osname);
1740 * Add an extra VFS_HOLD on our parent vfs so that it can't
1741 * disappear due to a forced unmount.
1743 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1744 VFS_HOLD(mvp->v_vfsp);
1752 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1754 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1755 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1757 statp->f_version = STATFS_VERSION;
1761 dmu_objset_space(zfsvfs->z_os,
1762 &refdbytes, &availbytes, &usedobjs, &availobjs);
1765 * The underlying storage pool actually uses multiple block sizes.
1766 * We report the fragsize as the smallest block size we support,
1767 * and we report our blocksize as the filesystem's maximum blocksize.
1769 statp->f_bsize = SPA_MINBLOCKSIZE;
1770 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1773 * The following report "total" blocks of various kinds in the
1774 * file system, but reported in terms of f_frsize - the
1778 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1779 statp->f_bfree = availbytes / statp->f_bsize;
1780 statp->f_bavail = statp->f_bfree; /* no root reservation */
1783 * statvfs() should really be called statufs(), because it assumes
1784 * static metadata. ZFS doesn't preallocate files, so the best
1785 * we can do is report the max that could possibly fit in f_files,
1786 * and that minus the number actually used in f_ffree.
1787 * For f_ffree, report the smaller of the number of object available
1788 * and the number of blocks (each object will take at least a block).
1790 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1791 statp->f_files = statp->f_ffree + usedobjs;
1794 * We're a zfs filesystem.
1796 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
1798 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
1799 sizeof(statp->f_mntfromname));
1800 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
1801 sizeof(statp->f_mntonname));
1803 statp->f_namemax = MAXNAMELEN - 1;
1810 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
1812 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1818 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1820 *vpp = ZTOV(rootzp);
1825 error = vn_lock(*vpp, flags);
1835 * Teardown the zfsvfs::z_os.
1837 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1838 * and 'z_teardown_inactive_lock' held.
1841 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1845 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1849 * We purge the parent filesystem's vfsp as the parent
1850 * filesystem and all of its snapshots have their vnode's
1851 * v_vfsp set to the parent's filesystem's vfsp. Note,
1852 * 'z_parent' is self referential for non-snapshots.
1854 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1855 #ifdef FREEBSD_NAMECACHE
1856 cache_purgevfs(zfsvfs->z_parent->z_vfs);
1861 * Close the zil. NB: Can't close the zil while zfs_inactive
1862 * threads are blocked as zil_close can call zfs_inactive.
1864 if (zfsvfs->z_log) {
1865 zil_close(zfsvfs->z_log);
1866 zfsvfs->z_log = NULL;
1869 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1872 * If we are not unmounting (ie: online recv) and someone already
1873 * unmounted this file system while we were doing the switcheroo,
1874 * or a reopen of z_os failed then just bail out now.
1876 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1877 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1878 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1879 return (SET_ERROR(EIO));
1883 * At this point there are no vops active, and any new vops will
1884 * fail with EIO since we have z_teardown_lock for writer (only
1885 * relavent for forced unmount).
1887 * Release all holds on dbufs.
1889 mutex_enter(&zfsvfs->z_znodes_lock);
1890 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1891 zp = list_next(&zfsvfs->z_all_znodes, zp))
1893 ASSERT(ZTOV(zp)->v_count >= 0);
1894 zfs_znode_dmu_fini(zp);
1896 mutex_exit(&zfsvfs->z_znodes_lock);
1899 * If we are unmounting, set the unmounted flag and let new vops
1900 * unblock. zfs_inactive will have the unmounted behavior, and all
1901 * other vops will fail with EIO.
1904 zfsvfs->z_unmounted = B_TRUE;
1905 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1906 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1910 * z_os will be NULL if there was an error in attempting to reopen
1911 * zfsvfs, so just return as the properties had already been
1912 * unregistered and cached data had been evicted before.
1914 if (zfsvfs->z_os == NULL)
1918 * Unregister properties.
1920 zfs_unregister_callbacks(zfsvfs);
1925 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1926 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1927 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1928 dmu_objset_evict_dbufs(zfsvfs->z_os);
1935 zfs_umount(vfs_t *vfsp, int fflag)
1937 kthread_t *td = curthread;
1938 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1940 cred_t *cr = td->td_ucred;
1943 ret = secpolicy_fs_unmount(cr, vfsp);
1945 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1946 ZFS_DELEG_PERM_MOUNT, cr))
1951 * We purge the parent filesystem's vfsp as the parent filesystem
1952 * and all of its snapshots have their vnode's v_vfsp set to the
1953 * parent's filesystem's vfsp. Note, 'z_parent' is self
1954 * referential for non-snapshots.
1956 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1959 * Unmount any snapshots mounted under .zfs before unmounting the
1962 if (zfsvfs->z_ctldir != NULL) {
1963 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
1967 if (fflag & MS_FORCE) {
1969 * Mark file system as unmounted before calling
1970 * vflush(FORCECLOSE). This way we ensure no future vnops
1971 * will be called and risk operating on DOOMED vnodes.
1973 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1974 zfsvfs->z_unmounted = B_TRUE;
1975 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1979 * Flush all the files.
1981 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
1986 if (!(fflag & MS_FORCE)) {
1988 * Check the number of active vnodes in the file system.
1989 * Our count is maintained in the vfs structure, but the
1990 * number is off by 1 to indicate a hold on the vfs
1993 * The '.zfs' directory maintains a reference of its
1994 * own, and any active references underneath are
1995 * reflected in the vnode count.
1997 if (zfsvfs->z_ctldir == NULL) {
1998 if (vfsp->vfs_count > 1)
1999 return (SET_ERROR(EBUSY));
2001 if (vfsp->vfs_count > 2 ||
2002 zfsvfs->z_ctldir->v_count > 1)
2003 return (SET_ERROR(EBUSY));
2008 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2012 * z_os will be NULL if there was an error in
2013 * attempting to reopen zfsvfs.
2017 * Unset the objset user_ptr.
2019 mutex_enter(&os->os_user_ptr_lock);
2020 dmu_objset_set_user(os, NULL);
2021 mutex_exit(&os->os_user_ptr_lock);
2024 * Finally release the objset
2026 dmu_objset_disown(os, zfsvfs);
2030 * We can now safely destroy the '.zfs' directory node.
2032 if (zfsvfs->z_ctldir != NULL)
2033 zfsctl_destroy(zfsvfs);
2040 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2042 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2047 * zfs_zget() can't operate on virtual entries like .zfs/ or
2048 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2049 * This will make NFS to switch to LOOKUP instead of using VGET.
2051 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2052 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2053 return (EOPNOTSUPP);
2056 err = zfs_zget(zfsvfs, ino, &zp);
2057 if (err == 0 && zp->z_unlinked) {
2065 err = vn_lock(*vpp, flags);
2072 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
2073 struct ucred **credanonp, int *numsecflavors, int **secflavors)
2075 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2078 * If this is regular file system vfsp is the same as
2079 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2080 * zfsvfs->z_parent->z_vfs represents parent file system
2081 * which we have to use here, because only this file system
2082 * has mnt_export configured.
2084 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2085 credanonp, numsecflavors, secflavors));
2088 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2089 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2092 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2094 struct componentname cn;
2095 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2098 uint64_t object = 0;
2099 uint64_t fid_gen = 0;
2109 * On FreeBSD we can get snapshot's mount point or its parent file
2110 * system mount point depending if snapshot is already mounted or not.
2112 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2113 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2114 uint64_t objsetid = 0;
2115 uint64_t setgen = 0;
2117 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2118 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2120 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2121 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2125 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2127 return (SET_ERROR(EINVAL));
2131 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2132 zfid_short_t *zfid = (zfid_short_t *)fidp;
2134 for (i = 0; i < sizeof (zfid->zf_object); i++)
2135 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2137 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2138 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2141 return (SET_ERROR(EINVAL));
2145 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2146 * directory tree. If the object == zfsvfs->z_shares_dir, then
2147 * we are in the .zfs/shares directory tree.
2149 if ((fid_gen == 0 &&
2150 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2151 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2153 VERIFY0(zfsctl_root(zfsvfs, LK_SHARED, &dvp));
2154 if (object == ZFSCTL_INO_SNAPDIR) {
2155 cn.cn_nameptr = "snapshot";
2156 cn.cn_namelen = strlen(cn.cn_nameptr);
2157 cn.cn_nameiop = LOOKUP;
2158 cn.cn_flags = ISLASTCN | LOCKLEAF;
2159 cn.cn_lkflags = flags;
2160 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2162 } else if (object == zfsvfs->z_shares_dir) {
2164 * XXX This branch must not be taken,
2165 * if it is, then the lookup below will
2168 cn.cn_nameptr = "shares";
2169 cn.cn_namelen = strlen(cn.cn_nameptr);
2170 cn.cn_nameiop = LOOKUP;
2171 cn.cn_flags = ISLASTCN;
2172 cn.cn_lkflags = flags;
2173 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2181 gen_mask = -1ULL >> (64 - 8 * i);
2183 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2184 if (err = zfs_zget(zfsvfs, object, &zp)) {
2188 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2190 zp_gen = zp_gen & gen_mask;
2193 if (zp->z_unlinked || zp_gen != fid_gen) {
2194 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2197 return (SET_ERROR(EINVAL));
2202 err = vn_lock(*vpp, flags);
2204 vnode_create_vobject(*vpp, zp->z_size, curthread);
2211 * Block out VOPs and close zfsvfs_t::z_os
2213 * Note, if successful, then we return with the 'z_teardown_lock' and
2214 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2215 * dataset and objset intact so that they can be atomically handed off during
2216 * a subsequent rollback or recv operation and the resume thereafter.
2219 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2223 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2230 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2231 * is an invariant across any of the operations that can be performed while the
2232 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2233 * are the same: the relevant objset and associated dataset are owned by
2234 * zfsvfs, held, and long held on entry.
2237 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2242 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2243 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2246 * We already own this, so just update the objset_t, as the one we
2247 * had before may have been evicted.
2250 VERIFY3P(ds->ds_owner, ==, zfsvfs);
2251 VERIFY(dsl_dataset_long_held(ds));
2252 VERIFY0(dmu_objset_from_ds(ds, &os));
2254 err = zfsvfs_init(zfsvfs, os);
2258 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2260 zfs_set_fuid_feature(zfsvfs);
2263 * Attempt to re-establish all the active znodes with
2264 * their dbufs. If a zfs_rezget() fails, then we'll let
2265 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2266 * when they try to use their znode.
2268 mutex_enter(&zfsvfs->z_znodes_lock);
2269 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2270 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2271 (void) zfs_rezget(zp);
2273 mutex_exit(&zfsvfs->z_znodes_lock);
2276 /* release the VOPs */
2277 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2278 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2282 * Since we couldn't setup the sa framework, try to force
2283 * unmount this file system.
2285 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) {
2286 vfs_ref(zfsvfs->z_vfs);
2287 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2294 zfs_freevfs(vfs_t *vfsp)
2296 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2300 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2301 * from zfs_mount(). Release it here. If we came through
2302 * zfs_mountroot() instead, we didn't grab an extra hold, so
2303 * skip the VFS_RELE for rootvfs.
2305 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2306 VFS_RELE(zfsvfs->z_parent->z_vfs);
2309 zfsvfs_free(zfsvfs);
2311 atomic_dec_32(&zfs_active_fs_count);
2315 static int desiredvnodes_backup;
2319 zfs_vnodes_adjust(void)
2322 int newdesiredvnodes;
2324 desiredvnodes_backup = desiredvnodes;
2327 * We calculate newdesiredvnodes the same way it is done in
2328 * vntblinit(). If it is equal to desiredvnodes, it means that
2329 * it wasn't tuned by the administrator and we can tune it down.
2331 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 *
2332 vm_kmem_size / (5 * (sizeof(struct vm_object) +
2333 sizeof(struct vnode))));
2334 if (newdesiredvnodes == desiredvnodes)
2335 desiredvnodes = (3 * newdesiredvnodes) / 4;
2340 zfs_vnodes_adjust_back(void)
2344 desiredvnodes = desiredvnodes_backup;
2352 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2355 * Initialize .zfs directory structures
2360 * Initialize znode cache, vnode ops, etc...
2365 * Reduce number of vnodes. Originally number of vnodes is calculated
2366 * with UFS inode in mind. We reduce it here, because it's too big for
2369 zfs_vnodes_adjust();
2371 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2379 zfs_vnodes_adjust_back();
2385 return (zfs_active_fs_count != 0);
2389 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2392 objset_t *os = zfsvfs->z_os;
2395 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2396 return (SET_ERROR(EINVAL));
2398 if (newvers < zfsvfs->z_version)
2399 return (SET_ERROR(EINVAL));
2401 if (zfs_spa_version_map(newvers) >
2402 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2403 return (SET_ERROR(ENOTSUP));
2405 tx = dmu_tx_create(os);
2406 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2407 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2408 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2410 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2412 error = dmu_tx_assign(tx, TXG_WAIT);
2418 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2419 8, 1, &newvers, tx);
2426 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2429 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2431 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2432 DMU_OT_NONE, 0, tx);
2434 error = zap_add(os, MASTER_NODE_OBJ,
2435 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2438 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2439 sa_register_update_callback(os, zfs_sa_upgrade);
2442 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2443 "from %llu to %llu", zfsvfs->z_version, newvers);
2447 zfsvfs->z_version = newvers;
2449 zfs_set_fuid_feature(zfsvfs);
2455 * Read a property stored within the master node.
2458 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2464 * Look up the file system's value for the property. For the
2465 * version property, we look up a slightly different string.
2467 if (prop == ZFS_PROP_VERSION)
2468 pname = ZPL_VERSION_STR;
2470 pname = zfs_prop_to_name(prop);
2473 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2475 if (error == ENOENT) {
2476 /* No value set, use the default value */
2478 case ZFS_PROP_VERSION:
2479 *value = ZPL_VERSION;
2481 case ZFS_PROP_NORMALIZE:
2482 case ZFS_PROP_UTF8ONLY:
2486 *value = ZFS_CASE_SENSITIVE;
2498 zfsvfs_update_fromname(const char *oldname, const char *newname)
2500 char tmpbuf[MAXPATHLEN];
2505 oldlen = strlen(oldname);
2507 mtx_lock(&mountlist_mtx);
2508 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2509 fromname = mp->mnt_stat.f_mntfromname;
2510 if (strcmp(fromname, oldname) == 0) {
2511 (void)strlcpy(fromname, newname,
2512 sizeof(mp->mnt_stat.f_mntfromname));
2515 if (strncmp(fromname, oldname, oldlen) == 0 &&
2516 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2517 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
2518 newname, fromname + oldlen);
2519 (void)strlcpy(fromname, tmpbuf,
2520 sizeof(mp->mnt_stat.f_mntfromname));
2524 mtx_unlock(&mountlist_mtx);