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, 2014 by Delphix. All rights reserved.
28 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
34 #include <sys/sysmacros.h>
37 #include <sys/vnode.h>
39 #include <sys/mntent.h>
40 #include <sys/mount.h>
41 #include <sys/cmn_err.h>
42 #include <sys/zfs_znode.h>
43 #include <sys/zfs_dir.h>
45 #include <sys/fs/zfs.h>
47 #include <sys/dsl_prop.h>
48 #include <sys/dsl_dataset.h>
49 #include <sys/dsl_deleg.h>
53 #include <sys/sa_impl.h>
54 #include <sys/varargs.h>
55 #include <sys/policy.h>
56 #include <sys/atomic.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/zfs_ctldir.h>
59 #include <sys/zfs_fuid.h>
60 #include <sys/sunddi.h>
62 #include <sys/dmu_objset.h>
63 #include <sys/spa_boot.h>
65 #include "zfs_comutil.h"
67 struct mtx zfs_debug_mtx;
68 MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF);
70 SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system");
73 SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0,
74 "File system owner can perform privileged operation on his file systems");
77 TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level);
78 SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0,
81 SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions");
82 static int zfs_version_acl = ZFS_ACL_VERSION;
83 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0,
85 static int zfs_version_spa = SPA_VERSION;
86 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0,
88 static int zfs_version_zpl = ZPL_VERSION;
89 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0,
92 static int zfs_mount(vfs_t *vfsp);
93 static int zfs_umount(vfs_t *vfsp, int fflag);
94 static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
95 static int zfs_statfs(vfs_t *vfsp, struct statfs *statp);
96 static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp);
97 static int zfs_sync(vfs_t *vfsp, int waitfor);
98 static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
99 struct ucred **credanonp, int *numsecflavors, int **secflavors);
100 static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp);
101 static void zfs_objset_close(zfsvfs_t *zfsvfs);
102 static void zfs_freevfs(vfs_t *vfsp);
104 static struct vfsops zfs_vfsops = {
105 .vfs_mount = zfs_mount,
106 .vfs_unmount = zfs_umount,
107 .vfs_root = zfs_root,
108 .vfs_statfs = zfs_statfs,
109 .vfs_vget = zfs_vget,
110 .vfs_sync = zfs_sync,
111 .vfs_checkexp = zfs_checkexp,
112 .vfs_fhtovp = zfs_fhtovp,
115 VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN);
118 * We need to keep a count of active fs's.
119 * This is necessary to prevent our module
120 * from being unloaded after a umount -f
122 static uint32_t zfs_active_fs_count = 0;
126 zfs_sync(vfs_t *vfsp, int waitfor)
130 * Data integrity is job one. We don't want a compromised kernel
131 * writing to the storage pool, so we never sync during panic.
138 * Sync a specific filesystem.
140 zfsvfs_t *zfsvfs = vfsp->vfs_data;
144 error = vfs_stdsync(vfsp, waitfor);
149 dp = dmu_objset_pool(zfsvfs->z_os);
152 * If the system is shutting down, then skip any
153 * filesystems which may exist on a suspended pool.
155 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
160 if (zfsvfs->z_log != NULL)
161 zil_commit(zfsvfs->z_log, 0);
166 * Sync all ZFS filesystems. This is what happens when you
167 * run sync(1M). Unlike other filesystems, ZFS honors the
168 * request by waiting for all pools to commit all dirty data.
176 #ifndef __FreeBSD_kernel__
178 zfs_create_unique_device(dev_t *dev)
183 ASSERT3U(zfs_minor, <=, MAXMIN32);
184 minor_t start = zfs_minor;
186 mutex_enter(&zfs_dev_mtx);
187 if (zfs_minor >= MAXMIN32) {
189 * If we're still using the real major
190 * keep out of /dev/zfs and /dev/zvol minor
191 * number space. If we're using a getudev()'ed
192 * major number, we can use all of its minors.
194 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
195 zfs_minor = ZFS_MIN_MINOR;
201 *dev = makedevice(zfs_major, zfs_minor);
202 mutex_exit(&zfs_dev_mtx);
203 } while (vfs_devismounted(*dev) && zfs_minor != start);
204 if (zfs_minor == start) {
206 * We are using all ~262,000 minor numbers for the
207 * current major number. Create a new major number.
209 if ((new_major = getudev()) == (major_t)-1) {
211 "zfs_mount: Can't get unique major "
215 mutex_enter(&zfs_dev_mtx);
216 zfs_major = new_major;
219 mutex_exit(&zfs_dev_mtx);
223 /* CONSTANTCONDITION */
228 #endif /* !__FreeBSD_kernel__ */
231 atime_changed_cb(void *arg, uint64_t newval)
233 zfsvfs_t *zfsvfs = arg;
235 if (newval == TRUE) {
236 zfsvfs->z_atime = TRUE;
237 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
238 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
239 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
241 zfsvfs->z_atime = FALSE;
242 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
243 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
244 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
249 xattr_changed_cb(void *arg, uint64_t newval)
251 zfsvfs_t *zfsvfs = arg;
253 if (newval == TRUE) {
254 /* XXX locking on vfs_flag? */
256 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? */
263 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
265 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
266 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
271 blksz_changed_cb(void *arg, uint64_t newval)
273 zfsvfs_t *zfsvfs = arg;
274 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
275 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
276 ASSERT(ISP2(newval));
278 zfsvfs->z_max_blksz = newval;
279 zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
283 readonly_changed_cb(void *arg, uint64_t newval)
285 zfsvfs_t *zfsvfs = arg;
288 /* XXX locking on vfs_flag? */
289 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
290 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
291 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
293 /* XXX locking on vfs_flag? */
294 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
295 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
296 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
301 setuid_changed_cb(void *arg, uint64_t newval)
303 zfsvfs_t *zfsvfs = arg;
305 if (newval == FALSE) {
306 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
307 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
308 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
310 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
311 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
312 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
317 exec_changed_cb(void *arg, uint64_t newval)
319 zfsvfs_t *zfsvfs = arg;
321 if (newval == FALSE) {
322 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
323 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
324 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
326 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
327 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
328 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
333 * The nbmand mount option can be changed at mount time.
334 * We can't allow it to be toggled on live file systems or incorrect
335 * behavior may be seen from cifs clients
337 * This property isn't registered via dsl_prop_register(), but this callback
338 * will be called when a file system is first mounted
341 nbmand_changed_cb(void *arg, uint64_t newval)
343 zfsvfs_t *zfsvfs = arg;
344 if (newval == FALSE) {
345 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
346 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
348 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
349 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
354 snapdir_changed_cb(void *arg, uint64_t newval)
356 zfsvfs_t *zfsvfs = arg;
358 zfsvfs->z_show_ctldir = newval;
362 vscan_changed_cb(void *arg, uint64_t newval)
364 zfsvfs_t *zfsvfs = arg;
366 zfsvfs->z_vscan = newval;
370 acl_mode_changed_cb(void *arg, uint64_t newval)
372 zfsvfs_t *zfsvfs = arg;
374 zfsvfs->z_acl_mode = newval;
378 acl_inherit_changed_cb(void *arg, uint64_t newval)
380 zfsvfs_t *zfsvfs = arg;
382 zfsvfs->z_acl_inherit = newval;
386 zfs_register_callbacks(vfs_t *vfsp)
388 struct dsl_dataset *ds = NULL;
390 zfsvfs_t *zfsvfs = NULL;
392 boolean_t readonly = B_FALSE;
393 boolean_t do_readonly = B_FALSE;
394 boolean_t setuid = B_FALSE;
395 boolean_t do_setuid = B_FALSE;
396 boolean_t exec = B_FALSE;
397 boolean_t do_exec = B_FALSE;
399 boolean_t devices = B_FALSE;
400 boolean_t do_devices = B_FALSE;
402 boolean_t xattr = B_FALSE;
403 boolean_t do_xattr = B_FALSE;
404 boolean_t atime = B_FALSE;
405 boolean_t do_atime = B_FALSE;
409 zfsvfs = vfsp->vfs_data;
414 * This function can be called for a snapshot when we update snapshot's
415 * mount point, which isn't really supported.
417 if (dmu_objset_is_snapshot(os))
421 * The act of registering our callbacks will destroy any mount
422 * options we may have. In order to enable temporary overrides
423 * of mount options, we stash away the current values and
424 * restore them after we register the callbacks.
426 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
427 !spa_writeable(dmu_objset_spa(os))) {
429 do_readonly = B_TRUE;
430 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
432 do_readonly = B_TRUE;
434 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
438 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
441 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
446 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
449 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
453 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
456 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
460 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
463 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
469 * We need to enter pool configuration here, so that we can use
470 * dsl_prop_get_int_ds() to handle the special nbmand property below.
471 * dsl_prop_get_integer() can not be used, because it has to acquire
472 * spa_namespace_lock and we can not do that because we already hold
473 * z_teardown_lock. The problem is that spa_config_sync() is called
474 * with spa_namespace_lock held and the function calls ZFS vnode
475 * operations to write the cache file and thus z_teardown_lock is
476 * acquired after spa_namespace_lock.
478 ds = dmu_objset_ds(os);
479 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
482 * nbmand is a special property. It can only be changed at
485 * This is weird, but it is documented to only be changeable
488 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
490 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
492 } else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) {
493 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
498 * Register property callbacks.
500 * It would probably be fine to just check for i/o error from
501 * the first prop_register(), but I guess I like to go
504 error = dsl_prop_register(ds,
505 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
506 error = error ? error : dsl_prop_register(ds,
507 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
508 error = error ? error : dsl_prop_register(ds,
509 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
510 error = error ? error : dsl_prop_register(ds,
511 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
513 error = error ? error : dsl_prop_register(ds,
514 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
516 error = error ? error : dsl_prop_register(ds,
517 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
518 error = error ? error : dsl_prop_register(ds,
519 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
520 error = error ? error : dsl_prop_register(ds,
521 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
522 error = error ? error : dsl_prop_register(ds,
523 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
524 error = error ? error : dsl_prop_register(ds,
525 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
527 error = error ? error : dsl_prop_register(ds,
528 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
529 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
534 * Invoke our callbacks to restore temporary mount options.
537 readonly_changed_cb(zfsvfs, readonly);
539 setuid_changed_cb(zfsvfs, setuid);
541 exec_changed_cb(zfsvfs, exec);
543 xattr_changed_cb(zfsvfs, xattr);
545 atime_changed_cb(zfsvfs, atime);
547 nbmand_changed_cb(zfsvfs, nbmand);
552 dsl_prop_unregister_all(ds, zfsvfs);
557 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
558 uint64_t *userp, uint64_t *groupp)
561 * Is it a valid type of object to track?
563 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
564 return (SET_ERROR(ENOENT));
567 * If we have a NULL data pointer
568 * then assume the id's aren't changing and
569 * return EEXIST to the dmu to let it know to
573 return (SET_ERROR(EEXIST));
575 if (bonustype == DMU_OT_ZNODE) {
576 znode_phys_t *znp = data;
577 *userp = znp->zp_uid;
578 *groupp = znp->zp_gid;
581 sa_hdr_phys_t *sap = data;
582 sa_hdr_phys_t sa = *sap;
583 boolean_t swap = B_FALSE;
585 ASSERT(bonustype == DMU_OT_SA);
587 if (sa.sa_magic == 0) {
589 * This should only happen for newly created
590 * files that haven't had the znode data filled
597 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
598 sa.sa_magic = SA_MAGIC;
599 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
602 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
605 hdrsize = sa_hdrsize(&sa);
606 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
607 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
609 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
612 *userp = BSWAP_64(*userp);
613 *groupp = BSWAP_64(*groupp);
620 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
621 char *domainbuf, int buflen, uid_t *ridp)
626 fuid = strtonum(fuidstr, NULL);
628 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
630 (void) strlcpy(domainbuf, domain, buflen);
633 *ridp = FUID_RID(fuid);
637 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
640 case ZFS_PROP_USERUSED:
641 return (DMU_USERUSED_OBJECT);
642 case ZFS_PROP_GROUPUSED:
643 return (DMU_GROUPUSED_OBJECT);
644 case ZFS_PROP_USERQUOTA:
645 return (zfsvfs->z_userquota_obj);
646 case ZFS_PROP_GROUPQUOTA:
647 return (zfsvfs->z_groupquota_obj);
653 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
654 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
659 zfs_useracct_t *buf = vbuf;
662 if (!dmu_objset_userspace_present(zfsvfs->z_os))
663 return (SET_ERROR(ENOTSUP));
665 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
671 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
672 (error = zap_cursor_retrieve(&zc, &za)) == 0;
673 zap_cursor_advance(&zc)) {
674 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
678 fuidstr_to_sid(zfsvfs, za.za_name,
679 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
681 buf->zu_space = za.za_first_integer;
687 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
688 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
689 *cookiep = zap_cursor_serialize(&zc);
690 zap_cursor_fini(&zc);
695 * buf must be big enough (eg, 32 bytes)
698 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
699 char *buf, boolean_t addok)
704 if (domain && domain[0]) {
705 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
707 return (SET_ERROR(ENOENT));
709 fuid = FUID_ENCODE(domainid, rid);
710 (void) sprintf(buf, "%llx", (longlong_t)fuid);
715 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
716 const char *domain, uint64_t rid, uint64_t *valp)
724 if (!dmu_objset_userspace_present(zfsvfs->z_os))
725 return (SET_ERROR(ENOTSUP));
727 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
731 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
735 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
742 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
743 const char *domain, uint64_t rid, uint64_t quota)
749 boolean_t fuid_dirtied;
751 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
752 return (SET_ERROR(EINVAL));
754 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
755 return (SET_ERROR(ENOTSUP));
757 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
758 &zfsvfs->z_groupquota_obj;
760 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
763 fuid_dirtied = zfsvfs->z_fuid_dirty;
765 tx = dmu_tx_create(zfsvfs->z_os);
766 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
768 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
769 zfs_userquota_prop_prefixes[type]);
772 zfs_fuid_txhold(zfsvfs, tx);
773 err = dmu_tx_assign(tx, TXG_WAIT);
779 mutex_enter(&zfsvfs->z_lock);
781 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
783 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
784 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
786 mutex_exit(&zfsvfs->z_lock);
789 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
793 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
797 zfs_fuid_sync(zfsvfs, tx);
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);
827 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
832 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
834 fuid = isgroup ? zp->z_gid : zp->z_uid;
836 if (quotaobj == 0 || zfsvfs->z_replay)
839 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
843 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
851 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
854 * We claim to always be readonly so we can open snapshots;
855 * other ZPL code will prevent us from writing to snapshots.
857 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
859 kmem_free(zfsvfs, sizeof (zfsvfs_t));
864 * Initialize the zfs-specific filesystem structure.
865 * Should probably make this a kmem cache, shuffle fields,
866 * and just bzero up to z_hold_mtx[].
868 zfsvfs->z_vfs = NULL;
869 zfsvfs->z_parent = zfsvfs;
870 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
871 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
874 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
877 } else if (zfsvfs->z_version >
878 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
879 (void) printf("Can't mount a version %lld file system "
880 "on a version %lld pool\n. Pool must be upgraded to mount "
881 "this file system.", (u_longlong_t)zfsvfs->z_version,
882 (u_longlong_t)spa_version(dmu_objset_spa(os)));
883 error = SET_ERROR(ENOTSUP);
886 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
888 zfsvfs->z_norm = (int)zval;
890 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
892 zfsvfs->z_utf8 = (zval != 0);
894 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
896 zfsvfs->z_case = (uint_t)zval;
899 * Fold case on file systems that are always or sometimes case
902 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
903 zfsvfs->z_case == ZFS_CASE_MIXED)
904 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
906 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
907 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
909 if (zfsvfs->z_use_sa) {
910 /* should either have both of these objects or none */
911 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
917 * Pre SA versions file systems should never touch
918 * either the attribute registration or layout objects.
923 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
924 &zfsvfs->z_attr_table);
928 if (zfsvfs->z_version >= ZPL_VERSION_SA)
929 sa_register_update_callback(os, zfs_sa_upgrade);
931 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
935 ASSERT(zfsvfs->z_root != 0);
937 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
938 &zfsvfs->z_unlinkedobj);
942 error = zap_lookup(os, MASTER_NODE_OBJ,
943 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
944 8, 1, &zfsvfs->z_userquota_obj);
945 if (error && error != ENOENT)
948 error = zap_lookup(os, MASTER_NODE_OBJ,
949 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
950 8, 1, &zfsvfs->z_groupquota_obj);
951 if (error && error != ENOENT)
954 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
955 &zfsvfs->z_fuid_obj);
956 if (error && error != ENOENT)
959 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
960 &zfsvfs->z_shares_dir);
961 if (error && error != ENOENT)
964 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
965 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
966 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
967 offsetof(znode_t, z_link_node));
968 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
969 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
970 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
971 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
972 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
978 dmu_objset_disown(os, zfsvfs);
980 kmem_free(zfsvfs, sizeof (zfsvfs_t));
985 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
989 error = zfs_register_callbacks(zfsvfs->z_vfs);
994 * Set the objset user_ptr to track its zfsvfs.
996 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
997 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
998 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1000 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1003 * If we are not mounting (ie: online recv), then we don't
1004 * have to worry about replaying the log as we blocked all
1005 * operations out since we closed the ZIL.
1011 * During replay we remove the read only flag to
1012 * allow replays to succeed.
1014 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1016 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1018 zfs_unlinked_drain(zfsvfs);
1021 * Parse and replay the intent log.
1023 * Because of ziltest, this must be done after
1024 * zfs_unlinked_drain(). (Further note: ziltest
1025 * doesn't use readonly mounts, where
1026 * zfs_unlinked_drain() isn't called.) This is because
1027 * ziltest causes spa_sync() to think it's committed,
1028 * but actually it is not, so the intent log contains
1029 * many txg's worth of changes.
1031 * In particular, if object N is in the unlinked set in
1032 * the last txg to actually sync, then it could be
1033 * actually freed in a later txg and then reallocated
1034 * in a yet later txg. This would write a "create
1035 * object N" record to the intent log. Normally, this
1036 * would be fine because the spa_sync() would have
1037 * written out the fact that object N is free, before
1038 * we could write the "create object N" intent log
1041 * But when we are in ziltest mode, we advance the "open
1042 * txg" without actually spa_sync()-ing the changes to
1043 * disk. So we would see that object N is still
1044 * allocated and in the unlinked set, and there is an
1045 * intent log record saying to allocate it.
1047 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1048 if (zil_replay_disable) {
1049 zil_destroy(zfsvfs->z_log, B_FALSE);
1051 zfsvfs->z_replay = B_TRUE;
1052 zil_replay(zfsvfs->z_os, zfsvfs,
1054 zfsvfs->z_replay = B_FALSE;
1057 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1063 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1066 zfsvfs_free(zfsvfs_t *zfsvfs)
1071 * This is a barrier to prevent the filesystem from going away in
1072 * zfs_znode_move() until we can safely ensure that the filesystem is
1073 * not unmounted. We consider the filesystem valid before the barrier
1074 * and invalid after the barrier.
1076 rw_enter(&zfsvfs_lock, RW_READER);
1077 rw_exit(&zfsvfs_lock);
1079 zfs_fuid_destroy(zfsvfs);
1081 mutex_destroy(&zfsvfs->z_znodes_lock);
1082 mutex_destroy(&zfsvfs->z_lock);
1083 list_destroy(&zfsvfs->z_all_znodes);
1084 rrm_destroy(&zfsvfs->z_teardown_lock);
1085 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1086 rw_destroy(&zfsvfs->z_fuid_lock);
1087 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1088 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1089 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1093 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1095 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1096 if (zfsvfs->z_vfs) {
1097 if (zfsvfs->z_use_fuids) {
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 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1106 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1107 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1108 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1109 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1110 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1113 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1117 zfs_domount(vfs_t *vfsp, char *osname)
1119 uint64_t recordsize, fsid_guid;
1127 error = zfsvfs_create(osname, &zfsvfs);
1130 zfsvfs->z_vfs = vfsp;
1133 /* Initialize the generic filesystem structure. */
1134 vfsp->vfs_bcount = 0;
1135 vfsp->vfs_data = NULL;
1137 if (zfs_create_unique_device(&mount_dev) == -1) {
1138 error = SET_ERROR(ENODEV);
1141 ASSERT(vfs_devismounted(mount_dev) == 0);
1144 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1147 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1148 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1150 vfsp->vfs_data = zfsvfs;
1151 vfsp->mnt_flag |= MNT_LOCAL;
1152 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1153 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1154 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1157 * The fsid is 64 bits, composed of an 8-bit fs type, which
1158 * separates our fsid from any other filesystem types, and a
1159 * 56-bit objset unique ID. The objset unique ID is unique to
1160 * all objsets open on this system, provided by unique_create().
1161 * The 8-bit fs type must be put in the low bits of fsid[1]
1162 * because that's where other Solaris filesystems put it.
1164 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1165 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1166 vfsp->vfs_fsid.val[0] = fsid_guid;
1167 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1168 vfsp->mnt_vfc->vfc_typenum & 0xFF;
1171 * Set features for file system.
1173 zfs_set_fuid_feature(zfsvfs);
1174 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1175 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1176 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1177 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1178 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1179 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1180 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1182 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1184 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1187 atime_changed_cb(zfsvfs, B_FALSE);
1188 readonly_changed_cb(zfsvfs, B_TRUE);
1189 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1191 xattr_changed_cb(zfsvfs, pval);
1192 zfsvfs->z_issnap = B_TRUE;
1193 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1195 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1196 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1197 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1199 error = zfsvfs_setup(zfsvfs, B_TRUE);
1202 vfs_mountedfrom(vfsp, osname);
1204 if (!zfsvfs->z_issnap)
1205 zfsctl_create(zfsvfs);
1208 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1209 zfsvfs_free(zfsvfs);
1211 atomic_inc_32(&zfs_active_fs_count);
1218 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1220 objset_t *os = zfsvfs->z_os;
1222 if (!dmu_objset_is_snapshot(os))
1223 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1228 * Convert a decimal digit string to a uint64_t integer.
1231 str_to_uint64(char *str, uint64_t *objnum)
1236 if (*str < '0' || *str > '9')
1237 return (SET_ERROR(EINVAL));
1239 num = num*10 + *str++ - '0';
1247 * The boot path passed from the boot loader is in the form of
1248 * "rootpool-name/root-filesystem-object-number'. Convert this
1249 * string to a dataset name: "rootpool-name/root-filesystem-name".
1252 zfs_parse_bootfs(char *bpath, char *outpath)
1258 if (*bpath == 0 || *bpath == '/')
1259 return (SET_ERROR(EINVAL));
1261 (void) strcpy(outpath, bpath);
1263 slashp = strchr(bpath, '/');
1265 /* if no '/', just return the pool name */
1266 if (slashp == NULL) {
1270 /* if not a number, just return the root dataset name */
1271 if (str_to_uint64(slashp+1, &objnum)) {
1276 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1283 * Check that the hex label string is appropriate for the dataset being
1284 * mounted into the global_zone proper.
1286 * Return an error if the hex label string is not default or
1287 * admin_low/admin_high. For admin_low labels, the corresponding
1288 * dataset must be readonly.
1291 zfs_check_global_label(const char *dsname, const char *hexsl)
1293 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1295 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1297 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1298 /* must be readonly */
1301 if (dsl_prop_get_integer(dsname,
1302 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1303 return (SET_ERROR(EACCES));
1304 return (rdonly ? 0 : EACCES);
1306 return (SET_ERROR(EACCES));
1310 * Determine whether the mount is allowed according to MAC check.
1311 * by comparing (where appropriate) label of the dataset against
1312 * the label of the zone being mounted into. If the dataset has
1313 * no label, create one.
1315 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1318 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1321 zone_t *mntzone = NULL;
1322 ts_label_t *mnt_tsl;
1325 char ds_hexsl[MAXNAMELEN];
1327 retv = EACCES; /* assume the worst */
1330 * Start by getting the dataset label if it exists.
1332 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1333 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1335 return (SET_ERROR(EACCES));
1338 * If labeling is NOT enabled, then disallow the mount of datasets
1339 * which have a non-default label already. No other label checks
1342 if (!is_system_labeled()) {
1343 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1345 return (SET_ERROR(EACCES));
1349 * Get the label of the mountpoint. If mounting into the global
1350 * zone (i.e. mountpoint is not within an active zone and the
1351 * zoned property is off), the label must be default or
1352 * admin_low/admin_high only; no other checks are needed.
1354 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1355 if (mntzone->zone_id == GLOBAL_ZONEID) {
1360 if (dsl_prop_get_integer(osname,
1361 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1362 return (SET_ERROR(EACCES));
1364 return (zfs_check_global_label(osname, ds_hexsl));
1367 * This is the case of a zone dataset being mounted
1368 * initially, before the zone has been fully created;
1369 * allow this mount into global zone.
1374 mnt_tsl = mntzone->zone_slabel;
1375 ASSERT(mnt_tsl != NULL);
1376 label_hold(mnt_tsl);
1377 mnt_sl = label2bslabel(mnt_tsl);
1379 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1381 * The dataset doesn't have a real label, so fabricate one.
1385 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1386 dsl_prop_set_string(osname,
1387 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1388 ZPROP_SRC_LOCAL, str) == 0)
1391 kmem_free(str, strlen(str) + 1);
1392 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1394 * Now compare labels to complete the MAC check. If the
1395 * labels are equal then allow access. If the mountpoint
1396 * label dominates the dataset label, allow readonly access.
1397 * Otherwise, access is denied.
1399 if (blequal(mnt_sl, &ds_sl))
1401 else if (bldominates(mnt_sl, &ds_sl)) {
1402 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1407 label_rele(mnt_tsl);
1411 #endif /* SECLABEL */
1413 #ifdef OPENSOLARIS_MOUNTROOT
1415 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1418 static int zfsrootdone = 0;
1419 zfsvfs_t *zfsvfs = NULL;
1428 * The filesystem that we mount as root is defined in the
1429 * boot property "zfs-bootfs" with a format of
1430 * "poolname/root-dataset-objnum".
1432 if (why == ROOT_INIT) {
1434 return (SET_ERROR(EBUSY));
1436 * the process of doing a spa_load will require the
1437 * clock to be set before we could (for example) do
1438 * something better by looking at the timestamp on
1439 * an uberblock, so just set it to -1.
1443 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1444 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1446 return (SET_ERROR(EINVAL));
1448 zfs_devid = spa_get_bootprop("diskdevid");
1449 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1451 spa_free_bootprop(zfs_devid);
1453 spa_free_bootprop(zfs_bootfs);
1454 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1458 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1459 spa_free_bootprop(zfs_bootfs);
1460 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1465 spa_free_bootprop(zfs_bootfs);
1467 if (error = vfs_lock(vfsp))
1470 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1471 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1475 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1477 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1478 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1483 mutex_enter(&vp->v_lock);
1484 vp->v_flag |= VROOT;
1485 mutex_exit(&vp->v_lock);
1489 * Leave rootvp held. The root file system is never unmounted.
1492 vfs_add((struct vnode *)0, vfsp,
1493 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1497 } else if (why == ROOT_REMOUNT) {
1498 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1499 vfsp->vfs_flag |= VFS_REMOUNT;
1501 /* refresh mount options */
1502 zfs_unregister_callbacks(vfsp->vfs_data);
1503 return (zfs_register_callbacks(vfsp));
1505 } else if (why == ROOT_UNMOUNT) {
1506 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1507 (void) zfs_sync(vfsp, 0, 0);
1512 * if "why" is equal to anything else other than ROOT_INIT,
1513 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1515 return (SET_ERROR(ENOTSUP));
1517 #endif /* OPENSOLARIS_MOUNTROOT */
1520 getpoolname(const char *osname, char *poolname)
1524 p = strchr(osname, '/');
1526 if (strlen(osname) >= MAXNAMELEN)
1527 return (ENAMETOOLONG);
1528 (void) strcpy(poolname, osname);
1530 if (p - osname >= MAXNAMELEN)
1531 return (ENAMETOOLONG);
1532 (void) strncpy(poolname, osname, p - osname);
1533 poolname[p - osname] = '\0';
1540 zfs_mount(vfs_t *vfsp)
1542 kthread_t *td = curthread;
1543 vnode_t *mvp = vfsp->mnt_vnodecovered;
1544 cred_t *cr = td->td_ucred;
1550 if (mvp->v_type != VDIR)
1551 return (SET_ERROR(ENOTDIR));
1553 mutex_enter(&mvp->v_lock);
1554 if ((uap->flags & MS_REMOUNT) == 0 &&
1555 (uap->flags & MS_OVERLAY) == 0 &&
1556 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1557 mutex_exit(&mvp->v_lock);
1558 return (SET_ERROR(EBUSY));
1560 mutex_exit(&mvp->v_lock);
1563 * ZFS does not support passing unparsed data in via MS_DATA.
1564 * Users should use the MS_OPTIONSTR interface; this means
1565 * that all option parsing is already done and the options struct
1566 * can be interrogated.
1568 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1569 #else /* !illumos */
1570 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS))
1571 return (SET_ERROR(EPERM));
1573 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1574 return (SET_ERROR(EINVAL));
1575 #endif /* illumos */
1578 * If full-owner-access is enabled and delegated administration is
1579 * turned on, we must set nosuid.
1581 if (zfs_super_owner &&
1582 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1583 secpolicy_fs_mount_clearopts(cr, vfsp);
1587 * Check for mount privilege?
1589 * If we don't have privilege then see if
1590 * we have local permission to allow it
1592 error = secpolicy_fs_mount(cr, mvp, vfsp);
1594 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1597 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1601 * Make sure user is the owner of the mount point
1602 * or has sufficient privileges.
1605 vattr.va_mask = AT_UID;
1607 vn_lock(mvp, LK_SHARED | LK_RETRY);
1608 if (VOP_GETATTR(mvp, &vattr, cr)) {
1613 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1614 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1621 secpolicy_fs_mount_clearopts(cr, vfsp);
1625 * Refuse to mount a filesystem if we are in a local zone and the
1626 * dataset is not visible.
1628 if (!INGLOBALZONE(curthread) &&
1629 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1630 error = SET_ERROR(EPERM);
1635 error = zfs_mount_label_policy(vfsp, osname);
1640 vfsp->vfs_flag |= MNT_NFS4ACLS;
1643 * When doing a remount, we simply refresh our temporary properties
1644 * according to those options set in the current VFS options.
1646 if (vfsp->vfs_flag & MS_REMOUNT) {
1647 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1650 * Refresh mount options with z_teardown_lock blocking I/O while
1651 * the filesystem is in an inconsistent state.
1652 * The lock also serializes this code with filesystem
1653 * manipulations between entry to zfs_suspend_fs() and return
1654 * from zfs_resume_fs().
1656 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1657 zfs_unregister_callbacks(zfsvfs);
1658 error = zfs_register_callbacks(vfsp);
1659 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1663 /* Initial root mount: try hard to import the requested root pool. */
1664 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1665 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1666 char pname[MAXNAMELEN];
1668 error = getpoolname(osname, pname);
1670 error = spa_import_rootpool(pname);
1675 error = zfs_domount(vfsp, osname);
1680 * Add an extra VFS_HOLD on our parent vfs so that it can't
1681 * disappear due to a forced unmount.
1683 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1684 VFS_HOLD(mvp->v_vfsp);
1692 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1694 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1695 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1697 statp->f_version = STATFS_VERSION;
1701 dmu_objset_space(zfsvfs->z_os,
1702 &refdbytes, &availbytes, &usedobjs, &availobjs);
1705 * The underlying storage pool actually uses multiple block sizes.
1706 * We report the fragsize as the smallest block size we support,
1707 * and we report our blocksize as the filesystem's maximum blocksize.
1709 statp->f_bsize = SPA_MINBLOCKSIZE;
1710 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1713 * The following report "total" blocks of various kinds in the
1714 * file system, but reported in terms of f_frsize - the
1718 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1719 statp->f_bfree = availbytes / statp->f_bsize;
1720 statp->f_bavail = statp->f_bfree; /* no root reservation */
1723 * statvfs() should really be called statufs(), because it assumes
1724 * static metadata. ZFS doesn't preallocate files, so the best
1725 * we can do is report the max that could possibly fit in f_files,
1726 * and that minus the number actually used in f_ffree.
1727 * For f_ffree, report the smaller of the number of object available
1728 * and the number of blocks (each object will take at least a block).
1730 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1731 statp->f_files = statp->f_ffree + usedobjs;
1734 * We're a zfs filesystem.
1736 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
1738 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
1739 sizeof(statp->f_mntfromname));
1740 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
1741 sizeof(statp->f_mntonname));
1743 statp->f_namemax = ZFS_MAXNAMELEN;
1750 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
1752 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1758 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1760 *vpp = ZTOV(rootzp);
1765 error = vn_lock(*vpp, flags);
1767 (*vpp)->v_vflag |= VV_ROOT;
1776 * Teardown the zfsvfs::z_os.
1778 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1779 * and 'z_teardown_inactive_lock' held.
1782 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1786 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1790 * We purge the parent filesystem's vfsp as the parent
1791 * filesystem and all of its snapshots have their vnode's
1792 * v_vfsp set to the parent's filesystem's vfsp. Note,
1793 * 'z_parent' is self referential for non-snapshots.
1795 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1796 #ifdef FREEBSD_NAMECACHE
1797 cache_purgevfs(zfsvfs->z_parent->z_vfs);
1802 * Close the zil. NB: Can't close the zil while zfs_inactive
1803 * threads are blocked as zil_close can call zfs_inactive.
1805 if (zfsvfs->z_log) {
1806 zil_close(zfsvfs->z_log);
1807 zfsvfs->z_log = NULL;
1810 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1813 * If we are not unmounting (ie: online recv) and someone already
1814 * unmounted this file system while we were doing the switcheroo,
1815 * or a reopen of z_os failed then just bail out now.
1817 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1818 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1819 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1820 return (SET_ERROR(EIO));
1824 * At this point there are no vops active, and any new vops will
1825 * fail with EIO since we have z_teardown_lock for writer (only
1826 * relavent for forced unmount).
1828 * Release all holds on dbufs.
1830 mutex_enter(&zfsvfs->z_znodes_lock);
1831 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1832 zp = list_next(&zfsvfs->z_all_znodes, zp))
1834 ASSERT(ZTOV(zp)->v_count >= 0);
1835 zfs_znode_dmu_fini(zp);
1837 mutex_exit(&zfsvfs->z_znodes_lock);
1840 * If we are unmounting, set the unmounted flag and let new vops
1841 * unblock. zfs_inactive will have the unmounted behavior, and all
1842 * other vops will fail with EIO.
1845 zfsvfs->z_unmounted = B_TRUE;
1846 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1847 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1851 * z_os will be NULL if there was an error in attempting to reopen
1852 * zfsvfs, so just return as the properties had already been
1853 * unregistered and cached data had been evicted before.
1855 if (zfsvfs->z_os == NULL)
1859 * Unregister properties.
1861 zfs_unregister_callbacks(zfsvfs);
1866 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1867 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1868 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1869 dmu_objset_evict_dbufs(zfsvfs->z_os);
1876 zfs_umount(vfs_t *vfsp, int fflag)
1878 kthread_t *td = curthread;
1879 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1881 cred_t *cr = td->td_ucred;
1884 ret = secpolicy_fs_unmount(cr, vfsp);
1886 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1887 ZFS_DELEG_PERM_MOUNT, cr))
1892 * We purge the parent filesystem's vfsp as the parent filesystem
1893 * and all of its snapshots have their vnode's v_vfsp set to the
1894 * parent's filesystem's vfsp. Note, 'z_parent' is self
1895 * referential for non-snapshots.
1897 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1900 * Unmount any snapshots mounted under .zfs before unmounting the
1903 if (zfsvfs->z_ctldir != NULL) {
1904 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
1906 ret = vflush(vfsp, 0, 0, td);
1907 ASSERT(ret == EBUSY);
1908 if (!(fflag & MS_FORCE)) {
1909 if (zfsvfs->z_ctldir->v_count > 1)
1911 ASSERT(zfsvfs->z_ctldir->v_count == 1);
1913 zfsctl_destroy(zfsvfs);
1914 ASSERT(zfsvfs->z_ctldir == NULL);
1917 if (fflag & MS_FORCE) {
1919 * Mark file system as unmounted before calling
1920 * vflush(FORCECLOSE). This way we ensure no future vnops
1921 * will be called and risk operating on DOOMED vnodes.
1923 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1924 zfsvfs->z_unmounted = B_TRUE;
1925 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1929 * Flush all the files.
1931 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
1933 if (!zfsvfs->z_issnap) {
1934 zfsctl_create(zfsvfs);
1935 ASSERT(zfsvfs->z_ctldir != NULL);
1941 if (!(fflag & MS_FORCE)) {
1943 * Check the number of active vnodes in the file system.
1944 * Our count is maintained in the vfs structure, but the
1945 * number is off by 1 to indicate a hold on the vfs
1948 * The '.zfs' directory maintains a reference of its
1949 * own, and any active references underneath are
1950 * reflected in the vnode count.
1952 if (zfsvfs->z_ctldir == NULL) {
1953 if (vfsp->vfs_count > 1)
1954 return (SET_ERROR(EBUSY));
1956 if (vfsp->vfs_count > 2 ||
1957 zfsvfs->z_ctldir->v_count > 1)
1958 return (SET_ERROR(EBUSY));
1963 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1967 * z_os will be NULL if there was an error in
1968 * attempting to reopen zfsvfs.
1972 * Unset the objset user_ptr.
1974 mutex_enter(&os->os_user_ptr_lock);
1975 dmu_objset_set_user(os, NULL);
1976 mutex_exit(&os->os_user_ptr_lock);
1979 * Finally release the objset
1981 dmu_objset_disown(os, zfsvfs);
1985 * We can now safely destroy the '.zfs' directory node.
1987 if (zfsvfs->z_ctldir != NULL)
1988 zfsctl_destroy(zfsvfs);
1989 if (zfsvfs->z_issnap) {
1990 vnode_t *svp = vfsp->mnt_vnodecovered;
1992 if (svp->v_count >= 2)
2001 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2003 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2008 * zfs_zget() can't operate on virtual entries like .zfs/ or
2009 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2010 * This will make NFS to switch to LOOKUP instead of using VGET.
2012 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2013 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2014 return (EOPNOTSUPP);
2017 err = zfs_zget(zfsvfs, ino, &zp);
2018 if (err == 0 && zp->z_unlinked) {
2026 err = vn_lock(*vpp, flags);
2033 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
2034 struct ucred **credanonp, int *numsecflavors, int **secflavors)
2036 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2039 * If this is regular file system vfsp is the same as
2040 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2041 * zfsvfs->z_parent->z_vfs represents parent file system
2042 * which we have to use here, because only this file system
2043 * has mnt_export configured.
2045 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2046 credanonp, numsecflavors, secflavors));
2049 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2050 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2053 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2055 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2057 uint64_t object = 0;
2058 uint64_t fid_gen = 0;
2068 * On FreeBSD we can get snapshot's mount point or its parent file
2069 * system mount point depending if snapshot is already mounted or not.
2071 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2072 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2073 uint64_t objsetid = 0;
2074 uint64_t setgen = 0;
2076 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2077 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2079 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2080 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2084 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2086 return (SET_ERROR(EINVAL));
2090 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2091 zfid_short_t *zfid = (zfid_short_t *)fidp;
2093 for (i = 0; i < sizeof (zfid->zf_object); i++)
2094 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2096 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2097 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2100 return (SET_ERROR(EINVAL));
2104 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2105 * directory tree. If the object == zfsvfs->z_shares_dir, then
2106 * we are in the .zfs/shares directory tree.
2108 if ((fid_gen == 0 &&
2109 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2110 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2111 *vpp = zfsvfs->z_ctldir;
2112 ASSERT(*vpp != NULL);
2113 if (object == ZFSCTL_INO_SNAPDIR) {
2114 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2115 0, NULL, NULL, NULL, NULL, NULL) == 0);
2116 } else if (object == zfsvfs->z_shares_dir) {
2117 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL,
2118 0, NULL, NULL, NULL, NULL, NULL) == 0);
2123 err = vn_lock(*vpp, flags);
2129 gen_mask = -1ULL >> (64 - 8 * i);
2131 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2132 if (err = zfs_zget(zfsvfs, object, &zp)) {
2136 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2138 zp_gen = zp_gen & gen_mask;
2141 if (zp->z_unlinked || zp_gen != fid_gen) {
2142 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2145 return (SET_ERROR(EINVAL));
2150 err = vn_lock(*vpp, flags | LK_RETRY);
2152 vnode_create_vobject(*vpp, zp->z_size, curthread);
2159 * Block out VOPs and close zfsvfs_t::z_os
2161 * Note, if successful, then we return with the 'z_teardown_lock' and
2162 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2163 * dataset and objset intact so that they can be atomically handed off during
2164 * a subsequent rollback or recv operation and the resume thereafter.
2167 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2171 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2178 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2179 * is an invariant across any of the operations that can be performed while the
2180 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2181 * are the same: the relevant objset and associated dataset are owned by
2182 * zfsvfs, held, and long held on entry.
2185 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2189 uint64_t sa_obj = 0;
2191 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2192 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2195 * We already own this, so just hold and rele it to update the
2196 * objset_t, as the one we had before may have been evicted.
2198 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os));
2199 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs);
2200 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset));
2201 dmu_objset_rele(zfsvfs->z_os, zfsvfs);
2204 * Make sure version hasn't changed
2207 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
2208 &zfsvfs->z_version);
2213 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2214 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2216 if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
2219 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2220 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
2223 if (zfsvfs->z_version >= ZPL_VERSION_SA)
2224 sa_register_update_callback(zfsvfs->z_os,
2227 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2229 zfs_set_fuid_feature(zfsvfs);
2232 * Attempt to re-establish all the active znodes with
2233 * their dbufs. If a zfs_rezget() fails, then we'll let
2234 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2235 * when they try to use their znode.
2237 mutex_enter(&zfsvfs->z_znodes_lock);
2238 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2239 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2240 (void) zfs_rezget(zp);
2242 mutex_exit(&zfsvfs->z_znodes_lock);
2245 /* release the VOPs */
2246 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2247 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2251 * Since we couldn't setup the sa framework, try to force
2252 * unmount this file system.
2254 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) {
2255 vfs_ref(zfsvfs->z_vfs);
2256 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2263 zfs_freevfs(vfs_t *vfsp)
2265 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2269 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2270 * from zfs_mount(). Release it here. If we came through
2271 * zfs_mountroot() instead, we didn't grab an extra hold, so
2272 * skip the VFS_RELE for rootvfs.
2274 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2275 VFS_RELE(zfsvfs->z_parent->z_vfs);
2278 zfsvfs_free(zfsvfs);
2280 atomic_dec_32(&zfs_active_fs_count);
2284 static int desiredvnodes_backup;
2288 zfs_vnodes_adjust(void)
2291 int newdesiredvnodes;
2293 desiredvnodes_backup = desiredvnodes;
2296 * We calculate newdesiredvnodes the same way it is done in
2297 * vntblinit(). If it is equal to desiredvnodes, it means that
2298 * it wasn't tuned by the administrator and we can tune it down.
2300 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 *
2301 vm_kmem_size / (5 * (sizeof(struct vm_object) +
2302 sizeof(struct vnode))));
2303 if (newdesiredvnodes == desiredvnodes)
2304 desiredvnodes = (3 * newdesiredvnodes) / 4;
2309 zfs_vnodes_adjust_back(void)
2313 desiredvnodes = desiredvnodes_backup;
2321 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2324 * Initialize .zfs directory structures
2329 * Initialize znode cache, vnode ops, etc...
2334 * Reduce number of vnodes. Originally number of vnodes is calculated
2335 * with UFS inode in mind. We reduce it here, because it's too big for
2338 zfs_vnodes_adjust();
2340 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2348 zfs_vnodes_adjust_back();
2354 return (zfs_active_fs_count != 0);
2358 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2361 objset_t *os = zfsvfs->z_os;
2364 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2365 return (SET_ERROR(EINVAL));
2367 if (newvers < zfsvfs->z_version)
2368 return (SET_ERROR(EINVAL));
2370 if (zfs_spa_version_map(newvers) >
2371 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2372 return (SET_ERROR(ENOTSUP));
2374 tx = dmu_tx_create(os);
2375 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2376 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2377 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2379 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2381 error = dmu_tx_assign(tx, TXG_WAIT);
2387 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2388 8, 1, &newvers, tx);
2395 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2398 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2400 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2401 DMU_OT_NONE, 0, tx);
2403 error = zap_add(os, MASTER_NODE_OBJ,
2404 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2407 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2408 sa_register_update_callback(os, zfs_sa_upgrade);
2411 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2412 "from %llu to %llu", zfsvfs->z_version, newvers);
2416 zfsvfs->z_version = newvers;
2418 zfs_set_fuid_feature(zfsvfs);
2424 * Read a property stored within the master node.
2427 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2433 * Look up the file system's value for the property. For the
2434 * version property, we look up a slightly different string.
2436 if (prop == ZFS_PROP_VERSION)
2437 pname = ZPL_VERSION_STR;
2439 pname = zfs_prop_to_name(prop);
2442 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2444 if (error == ENOENT) {
2445 /* No value set, use the default value */
2447 case ZFS_PROP_VERSION:
2448 *value = ZPL_VERSION;
2450 case ZFS_PROP_NORMALIZE:
2451 case ZFS_PROP_UTF8ONLY:
2455 *value = ZFS_CASE_SENSITIVE;
2467 zfsvfs_update_fromname(const char *oldname, const char *newname)
2469 char tmpbuf[MAXPATHLEN];
2474 oldlen = strlen(oldname);
2476 mtx_lock(&mountlist_mtx);
2477 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2478 fromname = mp->mnt_stat.f_mntfromname;
2479 if (strcmp(fromname, oldname) == 0) {
2480 (void)strlcpy(fromname, newname,
2481 sizeof(mp->mnt_stat.f_mntfromname));
2484 if (strncmp(fromname, oldname, oldlen) == 0 &&
2485 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2486 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
2487 newname, fromname + oldlen);
2488 (void)strlcpy(fromname, tmpbuf,
2489 sizeof(mp->mnt_stat.f_mntfromname));
2493 mtx_unlock(&mountlist_mtx);