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 SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RWTUN, &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 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.
137 * Ignore the system syncher. ZFS already commits async data
138 * at zfs_txg_timeout intervals.
140 if (waitfor == MNT_LAZY)
145 * Sync a specific filesystem.
147 zfsvfs_t *zfsvfs = vfsp->vfs_data;
151 error = vfs_stdsync(vfsp, waitfor);
156 dp = dmu_objset_pool(zfsvfs->z_os);
159 * If the system is shutting down, then skip any
160 * filesystems which may exist on a suspended pool.
162 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
167 if (zfsvfs->z_log != NULL)
168 zil_commit(zfsvfs->z_log, 0);
173 * Sync all ZFS filesystems. This is what happens when you
174 * run sync(1M). Unlike other filesystems, ZFS honors the
175 * request by waiting for all pools to commit all dirty data.
183 #ifndef __FreeBSD_kernel__
185 zfs_create_unique_device(dev_t *dev)
190 ASSERT3U(zfs_minor, <=, MAXMIN32);
191 minor_t start = zfs_minor;
193 mutex_enter(&zfs_dev_mtx);
194 if (zfs_minor >= MAXMIN32) {
196 * If we're still using the real major
197 * keep out of /dev/zfs and /dev/zvol minor
198 * number space. If we're using a getudev()'ed
199 * major number, we can use all of its minors.
201 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
202 zfs_minor = ZFS_MIN_MINOR;
208 *dev = makedevice(zfs_major, zfs_minor);
209 mutex_exit(&zfs_dev_mtx);
210 } while (vfs_devismounted(*dev) && zfs_minor != start);
211 if (zfs_minor == start) {
213 * We are using all ~262,000 minor numbers for the
214 * current major number. Create a new major number.
216 if ((new_major = getudev()) == (major_t)-1) {
218 "zfs_mount: Can't get unique major "
222 mutex_enter(&zfs_dev_mtx);
223 zfs_major = new_major;
226 mutex_exit(&zfs_dev_mtx);
230 /* CONSTANTCONDITION */
235 #endif /* !__FreeBSD_kernel__ */
238 atime_changed_cb(void *arg, uint64_t newval)
240 zfsvfs_t *zfsvfs = arg;
242 if (newval == TRUE) {
243 zfsvfs->z_atime = TRUE;
244 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
245 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
246 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
248 zfsvfs->z_atime = FALSE;
249 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
250 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
251 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
256 xattr_changed_cb(void *arg, uint64_t newval)
258 zfsvfs_t *zfsvfs = arg;
260 if (newval == TRUE) {
261 /* XXX locking on vfs_flag? */
263 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
265 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
266 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
268 /* XXX locking on vfs_flag? */
270 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
272 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
273 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
278 blksz_changed_cb(void *arg, uint64_t newval)
280 zfsvfs_t *zfsvfs = arg;
281 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
282 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
283 ASSERT(ISP2(newval));
285 zfsvfs->z_max_blksz = newval;
286 zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
290 readonly_changed_cb(void *arg, uint64_t newval)
292 zfsvfs_t *zfsvfs = arg;
295 /* XXX locking on vfs_flag? */
296 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
297 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
298 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
300 /* XXX locking on vfs_flag? */
301 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
302 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
303 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
308 setuid_changed_cb(void *arg, uint64_t newval)
310 zfsvfs_t *zfsvfs = arg;
312 if (newval == FALSE) {
313 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
314 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
315 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
317 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
318 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
319 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
324 exec_changed_cb(void *arg, uint64_t newval)
326 zfsvfs_t *zfsvfs = arg;
328 if (newval == FALSE) {
329 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
330 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
331 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
333 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
334 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
335 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
340 * The nbmand mount option can be changed at mount time.
341 * We can't allow it to be toggled on live file systems or incorrect
342 * behavior may be seen from cifs clients
344 * This property isn't registered via dsl_prop_register(), but this callback
345 * will be called when a file system is first mounted
348 nbmand_changed_cb(void *arg, uint64_t newval)
350 zfsvfs_t *zfsvfs = arg;
351 if (newval == FALSE) {
352 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
353 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
355 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
356 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
361 snapdir_changed_cb(void *arg, uint64_t newval)
363 zfsvfs_t *zfsvfs = arg;
365 zfsvfs->z_show_ctldir = newval;
369 vscan_changed_cb(void *arg, uint64_t newval)
371 zfsvfs_t *zfsvfs = arg;
373 zfsvfs->z_vscan = newval;
377 acl_mode_changed_cb(void *arg, uint64_t newval)
379 zfsvfs_t *zfsvfs = arg;
381 zfsvfs->z_acl_mode = newval;
385 acl_inherit_changed_cb(void *arg, uint64_t newval)
387 zfsvfs_t *zfsvfs = arg;
389 zfsvfs->z_acl_inherit = newval;
393 zfs_register_callbacks(vfs_t *vfsp)
395 struct dsl_dataset *ds = NULL;
397 zfsvfs_t *zfsvfs = NULL;
399 boolean_t readonly = B_FALSE;
400 boolean_t do_readonly = B_FALSE;
401 boolean_t setuid = B_FALSE;
402 boolean_t do_setuid = B_FALSE;
403 boolean_t exec = B_FALSE;
404 boolean_t do_exec = B_FALSE;
406 boolean_t devices = B_FALSE;
407 boolean_t do_devices = B_FALSE;
409 boolean_t xattr = B_FALSE;
410 boolean_t do_xattr = B_FALSE;
411 boolean_t atime = B_FALSE;
412 boolean_t do_atime = B_FALSE;
416 zfsvfs = vfsp->vfs_data;
421 * This function can be called for a snapshot when we update snapshot's
422 * mount point, which isn't really supported.
424 if (dmu_objset_is_snapshot(os))
428 * The act of registering our callbacks will destroy any mount
429 * options we may have. In order to enable temporary overrides
430 * of mount options, we stash away the current values and
431 * restore them after we register the callbacks.
433 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
434 !spa_writeable(dmu_objset_spa(os))) {
436 do_readonly = B_TRUE;
437 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
439 do_readonly = B_TRUE;
441 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
445 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
448 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
453 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
456 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
460 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
463 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
467 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
470 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
476 * We need to enter pool configuration here, so that we can use
477 * dsl_prop_get_int_ds() to handle the special nbmand property below.
478 * dsl_prop_get_integer() can not be used, because it has to acquire
479 * spa_namespace_lock and we can not do that because we already hold
480 * z_teardown_lock. The problem is that spa_config_sync() is called
481 * with spa_namespace_lock held and the function calls ZFS vnode
482 * operations to write the cache file and thus z_teardown_lock is
483 * acquired after spa_namespace_lock.
485 ds = dmu_objset_ds(os);
486 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
489 * nbmand is a special property. It can only be changed at
492 * This is weird, but it is documented to only be changeable
495 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
497 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
499 } else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) {
500 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
505 * Register property callbacks.
507 * It would probably be fine to just check for i/o error from
508 * the first prop_register(), but I guess I like to go
511 error = dsl_prop_register(ds,
512 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
513 error = error ? error : dsl_prop_register(ds,
514 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
515 error = error ? error : dsl_prop_register(ds,
516 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
517 error = error ? error : dsl_prop_register(ds,
518 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
520 error = error ? error : dsl_prop_register(ds,
521 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
523 error = error ? error : dsl_prop_register(ds,
524 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
525 error = error ? error : dsl_prop_register(ds,
526 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
527 error = error ? error : dsl_prop_register(ds,
528 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
529 error = error ? error : dsl_prop_register(ds,
530 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
531 error = error ? error : dsl_prop_register(ds,
532 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
534 error = error ? error : dsl_prop_register(ds,
535 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
536 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
541 * Invoke our callbacks to restore temporary mount options.
544 readonly_changed_cb(zfsvfs, readonly);
546 setuid_changed_cb(zfsvfs, setuid);
548 exec_changed_cb(zfsvfs, exec);
550 xattr_changed_cb(zfsvfs, xattr);
552 atime_changed_cb(zfsvfs, atime);
554 nbmand_changed_cb(zfsvfs, nbmand);
559 dsl_prop_unregister_all(ds, zfsvfs);
564 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
565 uint64_t *userp, uint64_t *groupp)
568 * Is it a valid type of object to track?
570 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
571 return (SET_ERROR(ENOENT));
574 * If we have a NULL data pointer
575 * then assume the id's aren't changing and
576 * return EEXIST to the dmu to let it know to
580 return (SET_ERROR(EEXIST));
582 if (bonustype == DMU_OT_ZNODE) {
583 znode_phys_t *znp = data;
584 *userp = znp->zp_uid;
585 *groupp = znp->zp_gid;
588 sa_hdr_phys_t *sap = data;
589 sa_hdr_phys_t sa = *sap;
590 boolean_t swap = B_FALSE;
592 ASSERT(bonustype == DMU_OT_SA);
594 if (sa.sa_magic == 0) {
596 * This should only happen for newly created
597 * files that haven't had the znode data filled
604 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
605 sa.sa_magic = SA_MAGIC;
606 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
609 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
612 hdrsize = sa_hdrsize(&sa);
613 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
614 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
616 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
619 *userp = BSWAP_64(*userp);
620 *groupp = BSWAP_64(*groupp);
627 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
628 char *domainbuf, int buflen, uid_t *ridp)
633 fuid = zfs_strtonum(fuidstr, NULL);
635 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
637 (void) strlcpy(domainbuf, domain, buflen);
640 *ridp = FUID_RID(fuid);
644 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
647 case ZFS_PROP_USERUSED:
648 return (DMU_USERUSED_OBJECT);
649 case ZFS_PROP_GROUPUSED:
650 return (DMU_GROUPUSED_OBJECT);
651 case ZFS_PROP_USERQUOTA:
652 return (zfsvfs->z_userquota_obj);
653 case ZFS_PROP_GROUPQUOTA:
654 return (zfsvfs->z_groupquota_obj);
660 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
661 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
666 zfs_useracct_t *buf = vbuf;
669 if (!dmu_objset_userspace_present(zfsvfs->z_os))
670 return (SET_ERROR(ENOTSUP));
672 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
678 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
679 (error = zap_cursor_retrieve(&zc, &za)) == 0;
680 zap_cursor_advance(&zc)) {
681 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
685 fuidstr_to_sid(zfsvfs, za.za_name,
686 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
688 buf->zu_space = za.za_first_integer;
694 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
695 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
696 *cookiep = zap_cursor_serialize(&zc);
697 zap_cursor_fini(&zc);
702 * buf must be big enough (eg, 32 bytes)
705 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
706 char *buf, boolean_t addok)
711 if (domain && domain[0]) {
712 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
714 return (SET_ERROR(ENOENT));
716 fuid = FUID_ENCODE(domainid, rid);
717 (void) sprintf(buf, "%llx", (longlong_t)fuid);
722 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
723 const char *domain, uint64_t rid, uint64_t *valp)
731 if (!dmu_objset_userspace_present(zfsvfs->z_os))
732 return (SET_ERROR(ENOTSUP));
734 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
738 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
742 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
749 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
750 const char *domain, uint64_t rid, uint64_t quota)
756 boolean_t fuid_dirtied;
758 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
759 return (SET_ERROR(EINVAL));
761 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
762 return (SET_ERROR(ENOTSUP));
764 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
765 &zfsvfs->z_groupquota_obj;
767 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
770 fuid_dirtied = zfsvfs->z_fuid_dirty;
772 tx = dmu_tx_create(zfsvfs->z_os);
773 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
775 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
776 zfs_userquota_prop_prefixes[type]);
779 zfs_fuid_txhold(zfsvfs, tx);
780 err = dmu_tx_assign(tx, TXG_WAIT);
786 mutex_enter(&zfsvfs->z_lock);
788 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
790 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
791 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
793 mutex_exit(&zfsvfs->z_lock);
796 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
800 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
804 zfs_fuid_sync(zfsvfs, tx);
810 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
813 uint64_t used, quota, usedobj, quotaobj;
816 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
817 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
819 if (quotaobj == 0 || zfsvfs->z_replay)
822 (void) sprintf(buf, "%llx", (longlong_t)fuid);
823 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
827 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
830 return (used >= quota);
834 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
839 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
841 fuid = isgroup ? zp->z_gid : zp->z_uid;
843 if (quotaobj == 0 || zfsvfs->z_replay)
846 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
850 * Associate this zfsvfs with the given objset, which must be owned.
851 * This will cache a bunch of on-disk state from the objset in the
855 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
860 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
861 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
864 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
867 if (zfsvfs->z_version >
868 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
869 (void) printf("Can't mount a version %lld file system "
870 "on a version %lld pool\n. Pool must be upgraded to mount "
871 "this file system.", (u_longlong_t)zfsvfs->z_version,
872 (u_longlong_t)spa_version(dmu_objset_spa(os)));
873 return (SET_ERROR(ENOTSUP));
875 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
878 zfsvfs->z_norm = (int)val;
880 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
883 zfsvfs->z_utf8 = (val != 0);
885 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
888 zfsvfs->z_case = (uint_t)val;
891 * Fold case on file systems that are always or sometimes case
894 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
895 zfsvfs->z_case == ZFS_CASE_MIXED)
896 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
898 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
899 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
902 if (zfsvfs->z_use_sa) {
903 /* should either have both of these objects or none */
904 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
910 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
911 &zfsvfs->z_attr_table);
915 if (zfsvfs->z_version >= ZPL_VERSION_SA)
916 sa_register_update_callback(os, zfs_sa_upgrade);
918 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
922 ASSERT(zfsvfs->z_root != 0);
924 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
925 &zfsvfs->z_unlinkedobj);
929 error = zap_lookup(os, MASTER_NODE_OBJ,
930 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
931 8, 1, &zfsvfs->z_userquota_obj);
933 zfsvfs->z_userquota_obj = 0;
937 error = zap_lookup(os, MASTER_NODE_OBJ,
938 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
939 8, 1, &zfsvfs->z_groupquota_obj);
941 zfsvfs->z_groupquota_obj = 0;
945 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
946 &zfsvfs->z_fuid_obj);
948 zfsvfs->z_fuid_obj = 0;
952 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
953 &zfsvfs->z_shares_dir);
955 zfsvfs->z_shares_dir = 0;
960 * Only use the name cache if we are looking for a
961 * name on a file system that does not require normalization
962 * or case folding. We can also look there if we happen to be
963 * on a non-normalizing, mixed sensitivity file system IF we
964 * are looking for the exact name (which is always the case on
967 zfsvfs->z_use_namecache = !zfsvfs->z_norm ||
968 ((zfsvfs->z_case == ZFS_CASE_MIXED) &&
969 !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER));
975 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
982 * XXX: Fix struct statfs so this isn't necessary!
984 * The 'osname' is used as the filesystem's special node, which means
985 * it must fit in statfs.f_mntfromname, or else it can't be
986 * enumerated, so libzfs_mnttab_find() returns NULL, which causes
987 * 'zfs unmount' to think it's not mounted when it is.
989 if (strlen(osname) >= MNAMELEN)
990 return (SET_ERROR(ENAMETOOLONG));
992 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
995 * We claim to always be readonly so we can open snapshots;
996 * other ZPL code will prevent us from writing to snapshots.
998 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
1000 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1004 zfsvfs->z_vfs = NULL;
1005 zfsvfs->z_parent = zfsvfs;
1007 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1008 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1009 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1010 offsetof(znode_t, z_link_node));
1012 rrm_init(&zfsvfs->z_teardown_lock, B_TRUE);
1014 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1016 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1017 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1018 for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1019 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1021 error = zfsvfs_init(zfsvfs, os);
1023 dmu_objset_disown(os, zfsvfs);
1025 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1034 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1038 error = zfs_register_callbacks(zfsvfs->z_vfs);
1042 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1045 * If we are not mounting (ie: online recv), then we don't
1046 * have to worry about replaying the log as we blocked all
1047 * operations out since we closed the ZIL.
1053 * During replay we remove the read only flag to
1054 * allow replays to succeed.
1056 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1058 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1060 zfs_unlinked_drain(zfsvfs);
1063 * Parse and replay the intent log.
1065 * Because of ziltest, this must be done after
1066 * zfs_unlinked_drain(). (Further note: ziltest
1067 * doesn't use readonly mounts, where
1068 * zfs_unlinked_drain() isn't called.) This is because
1069 * ziltest causes spa_sync() to think it's committed,
1070 * but actually it is not, so the intent log contains
1071 * many txg's worth of changes.
1073 * In particular, if object N is in the unlinked set in
1074 * the last txg to actually sync, then it could be
1075 * actually freed in a later txg and then reallocated
1076 * in a yet later txg. This would write a "create
1077 * object N" record to the intent log. Normally, this
1078 * would be fine because the spa_sync() would have
1079 * written out the fact that object N is free, before
1080 * we could write the "create object N" intent log
1083 * But when we are in ziltest mode, we advance the "open
1084 * txg" without actually spa_sync()-ing the changes to
1085 * disk. So we would see that object N is still
1086 * allocated and in the unlinked set, and there is an
1087 * intent log record saying to allocate it.
1089 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1090 if (zil_replay_disable) {
1091 zil_destroy(zfsvfs->z_log, B_FALSE);
1093 zfsvfs->z_replay = B_TRUE;
1094 zil_replay(zfsvfs->z_os, zfsvfs,
1096 zfsvfs->z_replay = B_FALSE;
1099 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1103 * Set the objset user_ptr to track its zfsvfs.
1105 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1106 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1107 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1112 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1115 zfsvfs_free(zfsvfs_t *zfsvfs)
1120 * This is a barrier to prevent the filesystem from going away in
1121 * zfs_znode_move() until we can safely ensure that the filesystem is
1122 * not unmounted. We consider the filesystem valid before the barrier
1123 * and invalid after the barrier.
1125 rw_enter(&zfsvfs_lock, RW_READER);
1126 rw_exit(&zfsvfs_lock);
1128 zfs_fuid_destroy(zfsvfs);
1130 mutex_destroy(&zfsvfs->z_znodes_lock);
1131 mutex_destroy(&zfsvfs->z_lock);
1132 list_destroy(&zfsvfs->z_all_znodes);
1133 rrm_destroy(&zfsvfs->z_teardown_lock);
1134 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1135 rw_destroy(&zfsvfs->z_fuid_lock);
1136 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1137 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1138 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1142 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1144 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1145 if (zfsvfs->z_vfs) {
1146 if (zfsvfs->z_use_fuids) {
1147 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1148 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1149 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1150 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1151 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1152 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1154 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1155 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1156 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1157 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1158 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1159 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1162 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1166 zfs_domount(vfs_t *vfsp, char *osname)
1168 uint64_t recordsize, fsid_guid;
1176 error = zfsvfs_create(osname, &zfsvfs);
1179 zfsvfs->z_vfs = vfsp;
1182 /* Initialize the generic filesystem structure. */
1183 vfsp->vfs_bcount = 0;
1184 vfsp->vfs_data = NULL;
1186 if (zfs_create_unique_device(&mount_dev) == -1) {
1187 error = SET_ERROR(ENODEV);
1190 ASSERT(vfs_devismounted(mount_dev) == 0);
1193 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1196 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1197 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1199 vfsp->vfs_data = zfsvfs;
1200 vfsp->mnt_flag |= MNT_LOCAL;
1201 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1202 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1203 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1204 vfsp->mnt_kern_flag |= MNTK_NO_IOPF; /* vn_io_fault can be used */
1207 * The fsid is 64 bits, composed of an 8-bit fs type, which
1208 * separates our fsid from any other filesystem types, and a
1209 * 56-bit objset unique ID. The objset unique ID is unique to
1210 * all objsets open on this system, provided by unique_create().
1211 * The 8-bit fs type must be put in the low bits of fsid[1]
1212 * because that's where other Solaris filesystems put it.
1214 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1215 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1216 vfsp->vfs_fsid.val[0] = fsid_guid;
1217 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1218 vfsp->mnt_vfc->vfc_typenum & 0xFF;
1221 * Set features for file system.
1223 zfs_set_fuid_feature(zfsvfs);
1224 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1225 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1226 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1227 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1228 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1229 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1230 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1232 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1234 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1237 atime_changed_cb(zfsvfs, B_FALSE);
1238 readonly_changed_cb(zfsvfs, B_TRUE);
1239 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1241 xattr_changed_cb(zfsvfs, pval);
1242 zfsvfs->z_issnap = B_TRUE;
1243 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1245 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1246 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1247 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1249 error = zfsvfs_setup(zfsvfs, B_TRUE);
1252 vfs_mountedfrom(vfsp, osname);
1254 if (!zfsvfs->z_issnap)
1255 zfsctl_create(zfsvfs);
1258 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1259 zfsvfs_free(zfsvfs);
1261 atomic_inc_32(&zfs_active_fs_count);
1268 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1270 objset_t *os = zfsvfs->z_os;
1272 if (!dmu_objset_is_snapshot(os))
1273 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1278 * Convert a decimal digit string to a uint64_t integer.
1281 str_to_uint64(char *str, uint64_t *objnum)
1286 if (*str < '0' || *str > '9')
1287 return (SET_ERROR(EINVAL));
1289 num = num*10 + *str++ - '0';
1297 * The boot path passed from the boot loader is in the form of
1298 * "rootpool-name/root-filesystem-object-number'. Convert this
1299 * string to a dataset name: "rootpool-name/root-filesystem-name".
1302 zfs_parse_bootfs(char *bpath, char *outpath)
1308 if (*bpath == 0 || *bpath == '/')
1309 return (SET_ERROR(EINVAL));
1311 (void) strcpy(outpath, bpath);
1313 slashp = strchr(bpath, '/');
1315 /* if no '/', just return the pool name */
1316 if (slashp == NULL) {
1320 /* if not a number, just return the root dataset name */
1321 if (str_to_uint64(slashp+1, &objnum)) {
1326 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1333 * Check that the hex label string is appropriate for the dataset being
1334 * mounted into the global_zone proper.
1336 * Return an error if the hex label string is not default or
1337 * admin_low/admin_high. For admin_low labels, the corresponding
1338 * dataset must be readonly.
1341 zfs_check_global_label(const char *dsname, const char *hexsl)
1343 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1345 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1347 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1348 /* must be readonly */
1351 if (dsl_prop_get_integer(dsname,
1352 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1353 return (SET_ERROR(EACCES));
1354 return (rdonly ? 0 : EACCES);
1356 return (SET_ERROR(EACCES));
1360 * Determine whether the mount is allowed according to MAC check.
1361 * by comparing (where appropriate) label of the dataset against
1362 * the label of the zone being mounted into. If the dataset has
1363 * no label, create one.
1365 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1368 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1371 zone_t *mntzone = NULL;
1372 ts_label_t *mnt_tsl;
1375 char ds_hexsl[MAXNAMELEN];
1377 retv = EACCES; /* assume the worst */
1380 * Start by getting the dataset label if it exists.
1382 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1383 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1385 return (SET_ERROR(EACCES));
1388 * If labeling is NOT enabled, then disallow the mount of datasets
1389 * which have a non-default label already. No other label checks
1392 if (!is_system_labeled()) {
1393 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1395 return (SET_ERROR(EACCES));
1399 * Get the label of the mountpoint. If mounting into the global
1400 * zone (i.e. mountpoint is not within an active zone and the
1401 * zoned property is off), the label must be default or
1402 * admin_low/admin_high only; no other checks are needed.
1404 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1405 if (mntzone->zone_id == GLOBAL_ZONEID) {
1410 if (dsl_prop_get_integer(osname,
1411 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1412 return (SET_ERROR(EACCES));
1414 return (zfs_check_global_label(osname, ds_hexsl));
1417 * This is the case of a zone dataset being mounted
1418 * initially, before the zone has been fully created;
1419 * allow this mount into global zone.
1424 mnt_tsl = mntzone->zone_slabel;
1425 ASSERT(mnt_tsl != NULL);
1426 label_hold(mnt_tsl);
1427 mnt_sl = label2bslabel(mnt_tsl);
1429 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1431 * The dataset doesn't have a real label, so fabricate one.
1435 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1436 dsl_prop_set_string(osname,
1437 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1438 ZPROP_SRC_LOCAL, str) == 0)
1441 kmem_free(str, strlen(str) + 1);
1442 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1444 * Now compare labels to complete the MAC check. If the
1445 * labels are equal then allow access. If the mountpoint
1446 * label dominates the dataset label, allow readonly access.
1447 * Otherwise, access is denied.
1449 if (blequal(mnt_sl, &ds_sl))
1451 else if (bldominates(mnt_sl, &ds_sl)) {
1452 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1457 label_rele(mnt_tsl);
1461 #endif /* SECLABEL */
1463 #ifdef OPENSOLARIS_MOUNTROOT
1465 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1468 static int zfsrootdone = 0;
1469 zfsvfs_t *zfsvfs = NULL;
1478 * The filesystem that we mount as root is defined in the
1479 * boot property "zfs-bootfs" with a format of
1480 * "poolname/root-dataset-objnum".
1482 if (why == ROOT_INIT) {
1484 return (SET_ERROR(EBUSY));
1486 * the process of doing a spa_load will require the
1487 * clock to be set before we could (for example) do
1488 * something better by looking at the timestamp on
1489 * an uberblock, so just set it to -1.
1493 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1494 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1496 return (SET_ERROR(EINVAL));
1498 zfs_devid = spa_get_bootprop("diskdevid");
1499 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1501 spa_free_bootprop(zfs_devid);
1503 spa_free_bootprop(zfs_bootfs);
1504 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1508 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1509 spa_free_bootprop(zfs_bootfs);
1510 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1515 spa_free_bootprop(zfs_bootfs);
1517 if (error = vfs_lock(vfsp))
1520 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1521 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1525 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1527 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1528 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1533 mutex_enter(&vp->v_lock);
1534 vp->v_flag |= VROOT;
1535 mutex_exit(&vp->v_lock);
1539 * Leave rootvp held. The root file system is never unmounted.
1542 vfs_add((struct vnode *)0, vfsp,
1543 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1547 } else if (why == ROOT_REMOUNT) {
1548 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1549 vfsp->vfs_flag |= VFS_REMOUNT;
1551 /* refresh mount options */
1552 zfs_unregister_callbacks(vfsp->vfs_data);
1553 return (zfs_register_callbacks(vfsp));
1555 } else if (why == ROOT_UNMOUNT) {
1556 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1557 (void) zfs_sync(vfsp, 0, 0);
1562 * if "why" is equal to anything else other than ROOT_INIT,
1563 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1565 return (SET_ERROR(ENOTSUP));
1567 #endif /* OPENSOLARIS_MOUNTROOT */
1570 getpoolname(const char *osname, char *poolname)
1574 p = strchr(osname, '/');
1576 if (strlen(osname) >= MAXNAMELEN)
1577 return (ENAMETOOLONG);
1578 (void) strcpy(poolname, osname);
1580 if (p - osname >= MAXNAMELEN)
1581 return (ENAMETOOLONG);
1582 (void) strncpy(poolname, osname, p - osname);
1583 poolname[p - osname] = '\0';
1590 zfs_mount(vfs_t *vfsp)
1592 kthread_t *td = curthread;
1593 vnode_t *mvp = vfsp->mnt_vnodecovered;
1594 cred_t *cr = td->td_ucred;
1600 if (mvp->v_type != VDIR)
1601 return (SET_ERROR(ENOTDIR));
1603 mutex_enter(&mvp->v_lock);
1604 if ((uap->flags & MS_REMOUNT) == 0 &&
1605 (uap->flags & MS_OVERLAY) == 0 &&
1606 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1607 mutex_exit(&mvp->v_lock);
1608 return (SET_ERROR(EBUSY));
1610 mutex_exit(&mvp->v_lock);
1613 * ZFS does not support passing unparsed data in via MS_DATA.
1614 * Users should use the MS_OPTIONSTR interface; this means
1615 * that all option parsing is already done and the options struct
1616 * can be interrogated.
1618 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1619 #else /* !illumos */
1620 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS))
1621 return (SET_ERROR(EPERM));
1623 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1624 return (SET_ERROR(EINVAL));
1625 #endif /* illumos */
1628 * If full-owner-access is enabled and delegated administration is
1629 * turned on, we must set nosuid.
1631 if (zfs_super_owner &&
1632 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1633 secpolicy_fs_mount_clearopts(cr, vfsp);
1637 * Check for mount privilege?
1639 * If we don't have privilege then see if
1640 * we have local permission to allow it
1642 error = secpolicy_fs_mount(cr, mvp, vfsp);
1644 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1647 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1651 * Make sure user is the owner of the mount point
1652 * or has sufficient privileges.
1655 vattr.va_mask = AT_UID;
1657 vn_lock(mvp, LK_SHARED | LK_RETRY);
1658 if (VOP_GETATTR(mvp, &vattr, cr)) {
1663 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1664 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1671 secpolicy_fs_mount_clearopts(cr, vfsp);
1675 * Refuse to mount a filesystem if we are in a local zone and the
1676 * dataset is not visible.
1678 if (!INGLOBALZONE(curthread) &&
1679 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1680 error = SET_ERROR(EPERM);
1685 error = zfs_mount_label_policy(vfsp, osname);
1690 vfsp->vfs_flag |= MNT_NFS4ACLS;
1693 * When doing a remount, we simply refresh our temporary properties
1694 * according to those options set in the current VFS options.
1696 if (vfsp->vfs_flag & MS_REMOUNT) {
1697 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1700 * Refresh mount options with z_teardown_lock blocking I/O while
1701 * the filesystem is in an inconsistent state.
1702 * The lock also serializes this code with filesystem
1703 * manipulations between entry to zfs_suspend_fs() and return
1704 * from zfs_resume_fs().
1706 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1707 zfs_unregister_callbacks(zfsvfs);
1708 error = zfs_register_callbacks(vfsp);
1709 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1713 /* Initial root mount: try hard to import the requested root pool. */
1714 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1715 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1716 char pname[MAXNAMELEN];
1718 error = getpoolname(osname, pname);
1720 error = spa_import_rootpool(pname);
1725 error = zfs_domount(vfsp, osname);
1730 * Add an extra VFS_HOLD on our parent vfs so that it can't
1731 * disappear due to a forced unmount.
1733 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1734 VFS_HOLD(mvp->v_vfsp);
1742 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1744 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1745 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1747 statp->f_version = STATFS_VERSION;
1751 dmu_objset_space(zfsvfs->z_os,
1752 &refdbytes, &availbytes, &usedobjs, &availobjs);
1755 * The underlying storage pool actually uses multiple block sizes.
1756 * We report the fragsize as the smallest block size we support,
1757 * and we report our blocksize as the filesystem's maximum blocksize.
1759 statp->f_bsize = SPA_MINBLOCKSIZE;
1760 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1763 * The following report "total" blocks of various kinds in the
1764 * file system, but reported in terms of f_frsize - the
1768 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1769 statp->f_bfree = availbytes / statp->f_bsize;
1770 statp->f_bavail = statp->f_bfree; /* no root reservation */
1773 * statvfs() should really be called statufs(), because it assumes
1774 * static metadata. ZFS doesn't preallocate files, so the best
1775 * we can do is report the max that could possibly fit in f_files,
1776 * and that minus the number actually used in f_ffree.
1777 * For f_ffree, report the smaller of the number of object available
1778 * and the number of blocks (each object will take at least a block).
1780 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1781 statp->f_files = statp->f_ffree + usedobjs;
1784 * We're a zfs filesystem.
1786 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
1788 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
1789 sizeof(statp->f_mntfromname));
1790 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
1791 sizeof(statp->f_mntonname));
1793 statp->f_namemax = MAXNAMELEN - 1;
1800 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
1802 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1808 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1810 *vpp = ZTOV(rootzp);
1815 error = vn_lock(*vpp, flags);
1825 * Teardown the zfsvfs::z_os.
1827 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1828 * and 'z_teardown_inactive_lock' held.
1831 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1835 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1839 * We purge the parent filesystem's vfsp as the parent
1840 * filesystem and all of its snapshots have their vnode's
1841 * v_vfsp set to the parent's filesystem's vfsp. Note,
1842 * 'z_parent' is self referential for non-snapshots.
1844 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1845 #ifdef FREEBSD_NAMECACHE
1846 cache_purgevfs(zfsvfs->z_parent->z_vfs, true);
1851 * Close the zil. NB: Can't close the zil while zfs_inactive
1852 * threads are blocked as zil_close can call zfs_inactive.
1854 if (zfsvfs->z_log) {
1855 zil_close(zfsvfs->z_log);
1856 zfsvfs->z_log = NULL;
1859 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1862 * If we are not unmounting (ie: online recv) and someone already
1863 * unmounted this file system while we were doing the switcheroo,
1864 * or a reopen of z_os failed then just bail out now.
1866 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1867 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1868 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1869 return (SET_ERROR(EIO));
1873 * At this point there are no vops active, and any new vops will
1874 * fail with EIO since we have z_teardown_lock for writer (only
1875 * relavent for forced unmount).
1877 * Release all holds on dbufs.
1879 mutex_enter(&zfsvfs->z_znodes_lock);
1880 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1881 zp = list_next(&zfsvfs->z_all_znodes, zp))
1883 ASSERT(ZTOV(zp)->v_count >= 0);
1884 zfs_znode_dmu_fini(zp);
1886 mutex_exit(&zfsvfs->z_znodes_lock);
1889 * If we are unmounting, set the unmounted flag and let new vops
1890 * unblock. zfs_inactive will have the unmounted behavior, and all
1891 * other vops will fail with EIO.
1894 zfsvfs->z_unmounted = B_TRUE;
1895 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1896 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1900 * z_os will be NULL if there was an error in attempting to reopen
1901 * zfsvfs, so just return as the properties had already been
1902 * unregistered and cached data had been evicted before.
1904 if (zfsvfs->z_os == NULL)
1908 * Unregister properties.
1910 zfs_unregister_callbacks(zfsvfs);
1915 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1916 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1917 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1918 dmu_objset_evict_dbufs(zfsvfs->z_os);
1925 zfs_umount(vfs_t *vfsp, int fflag)
1927 kthread_t *td = curthread;
1928 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1930 cred_t *cr = td->td_ucred;
1933 ret = secpolicy_fs_unmount(cr, vfsp);
1935 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1936 ZFS_DELEG_PERM_MOUNT, cr))
1941 * We purge the parent filesystem's vfsp as the parent filesystem
1942 * and all of its snapshots have their vnode's v_vfsp set to the
1943 * parent's filesystem's vfsp. Note, 'z_parent' is self
1944 * referential for non-snapshots.
1946 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1949 * Unmount any snapshots mounted under .zfs before unmounting the
1952 if (zfsvfs->z_ctldir != NULL) {
1953 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
1957 if (fflag & MS_FORCE) {
1959 * Mark file system as unmounted before calling
1960 * vflush(FORCECLOSE). This way we ensure no future vnops
1961 * will be called and risk operating on DOOMED vnodes.
1963 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1964 zfsvfs->z_unmounted = B_TRUE;
1965 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1969 * Flush all the files.
1971 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
1976 if (!(fflag & MS_FORCE)) {
1978 * Check the number of active vnodes in the file system.
1979 * Our count is maintained in the vfs structure, but the
1980 * number is off by 1 to indicate a hold on the vfs
1983 * The '.zfs' directory maintains a reference of its
1984 * own, and any active references underneath are
1985 * reflected in the vnode count.
1987 if (zfsvfs->z_ctldir == NULL) {
1988 if (vfsp->vfs_count > 1)
1989 return (SET_ERROR(EBUSY));
1991 if (vfsp->vfs_count > 2 ||
1992 zfsvfs->z_ctldir->v_count > 1)
1993 return (SET_ERROR(EBUSY));
1998 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2002 * z_os will be NULL if there was an error in
2003 * attempting to reopen zfsvfs.
2007 * Unset the objset user_ptr.
2009 mutex_enter(&os->os_user_ptr_lock);
2010 dmu_objset_set_user(os, NULL);
2011 mutex_exit(&os->os_user_ptr_lock);
2014 * Finally release the objset
2016 dmu_objset_disown(os, zfsvfs);
2020 * We can now safely destroy the '.zfs' directory node.
2022 if (zfsvfs->z_ctldir != NULL)
2023 zfsctl_destroy(zfsvfs);
2030 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2032 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2037 * zfs_zget() can't operate on virtual entries like .zfs/ or
2038 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2039 * This will make NFS to switch to LOOKUP instead of using VGET.
2041 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2042 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2043 return (EOPNOTSUPP);
2046 err = zfs_zget(zfsvfs, ino, &zp);
2047 if (err == 0 && zp->z_unlinked) {
2055 err = vn_lock(*vpp, flags);
2062 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
2063 struct ucred **credanonp, int *numsecflavors, int **secflavors)
2065 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2068 * If this is regular file system vfsp is the same as
2069 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2070 * zfsvfs->z_parent->z_vfs represents parent file system
2071 * which we have to use here, because only this file system
2072 * has mnt_export configured.
2074 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2075 credanonp, numsecflavors, secflavors));
2078 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2079 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2082 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2084 struct componentname cn;
2085 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2088 uint64_t object = 0;
2089 uint64_t fid_gen = 0;
2099 * On FreeBSD we can get snapshot's mount point or its parent file
2100 * system mount point depending if snapshot is already mounted or not.
2102 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2103 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2104 uint64_t objsetid = 0;
2105 uint64_t setgen = 0;
2107 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2108 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2110 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2111 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2115 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2117 return (SET_ERROR(EINVAL));
2121 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2122 zfid_short_t *zfid = (zfid_short_t *)fidp;
2124 for (i = 0; i < sizeof (zfid->zf_object); i++)
2125 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2127 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2128 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2131 return (SET_ERROR(EINVAL));
2135 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2136 * directory tree. If the object == zfsvfs->z_shares_dir, then
2137 * we are in the .zfs/shares directory tree.
2139 if ((fid_gen == 0 &&
2140 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2141 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2143 VERIFY0(zfsctl_root(zfsvfs, LK_SHARED, &dvp));
2144 if (object == ZFSCTL_INO_SNAPDIR) {
2145 cn.cn_nameptr = "snapshot";
2146 cn.cn_namelen = strlen(cn.cn_nameptr);
2147 cn.cn_nameiop = LOOKUP;
2148 cn.cn_flags = ISLASTCN | LOCKLEAF;
2149 cn.cn_lkflags = flags;
2150 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2152 } else if (object == zfsvfs->z_shares_dir) {
2154 * XXX This branch must not be taken,
2155 * if it is, then the lookup below will
2158 cn.cn_nameptr = "shares";
2159 cn.cn_namelen = strlen(cn.cn_nameptr);
2160 cn.cn_nameiop = LOOKUP;
2161 cn.cn_flags = ISLASTCN;
2162 cn.cn_lkflags = flags;
2163 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2171 gen_mask = -1ULL >> (64 - 8 * i);
2173 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2174 if (err = zfs_zget(zfsvfs, object, &zp)) {
2178 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2180 zp_gen = zp_gen & gen_mask;
2183 if (zp->z_unlinked || zp_gen != fid_gen) {
2184 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2187 return (SET_ERROR(EINVAL));
2192 err = vn_lock(*vpp, flags);
2194 vnode_create_vobject(*vpp, zp->z_size, curthread);
2201 * Block out VOPs and close zfsvfs_t::z_os
2203 * Note, if successful, then we return with the 'z_teardown_lock' and
2204 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2205 * dataset and objset intact so that they can be atomically handed off during
2206 * a subsequent rollback or recv operation and the resume thereafter.
2209 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2213 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2220 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2221 * is an invariant across any of the operations that can be performed while the
2222 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2223 * are the same: the relevant objset and associated dataset are owned by
2224 * zfsvfs, held, and long held on entry.
2227 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2232 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2233 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2236 * We already own this, so just update the objset_t, as the one we
2237 * had before may have been evicted.
2240 VERIFY3P(ds->ds_owner, ==, zfsvfs);
2241 VERIFY(dsl_dataset_long_held(ds));
2242 VERIFY0(dmu_objset_from_ds(ds, &os));
2244 err = zfsvfs_init(zfsvfs, os);
2248 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2250 zfs_set_fuid_feature(zfsvfs);
2253 * Attempt to re-establish all the active znodes with
2254 * their dbufs. If a zfs_rezget() fails, then we'll let
2255 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2256 * when they try to use their znode.
2258 mutex_enter(&zfsvfs->z_znodes_lock);
2259 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2260 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2261 (void) zfs_rezget(zp);
2263 mutex_exit(&zfsvfs->z_znodes_lock);
2266 /* release the VOPs */
2267 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2268 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2272 * Since we couldn't setup the sa framework, try to force
2273 * unmount this file system.
2275 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) {
2276 vfs_ref(zfsvfs->z_vfs);
2277 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2284 zfs_freevfs(vfs_t *vfsp)
2286 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2290 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2291 * from zfs_mount(). Release it here. If we came through
2292 * zfs_mountroot() instead, we didn't grab an extra hold, so
2293 * skip the VFS_RELE for rootvfs.
2295 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2296 VFS_RELE(zfsvfs->z_parent->z_vfs);
2299 zfsvfs_free(zfsvfs);
2301 atomic_dec_32(&zfs_active_fs_count);
2305 static int desiredvnodes_backup;
2309 zfs_vnodes_adjust(void)
2312 int newdesiredvnodes;
2314 desiredvnodes_backup = desiredvnodes;
2317 * We calculate newdesiredvnodes the same way it is done in
2318 * vntblinit(). If it is equal to desiredvnodes, it means that
2319 * it wasn't tuned by the administrator and we can tune it down.
2321 newdesiredvnodes = min(maxproc + vm_cnt.v_page_count / 4, 2 *
2322 vm_kmem_size / (5 * (sizeof(struct vm_object) +
2323 sizeof(struct vnode))));
2324 if (newdesiredvnodes == desiredvnodes)
2325 desiredvnodes = (3 * newdesiredvnodes) / 4;
2330 zfs_vnodes_adjust_back(void)
2334 desiredvnodes = desiredvnodes_backup;
2342 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2345 * Initialize .zfs directory structures
2350 * Initialize znode cache, vnode ops, etc...
2355 * Reduce number of vnodes. Originally number of vnodes is calculated
2356 * with UFS inode in mind. We reduce it here, because it's too big for
2359 zfs_vnodes_adjust();
2361 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2369 zfs_vnodes_adjust_back();
2375 return (zfs_active_fs_count != 0);
2379 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2382 objset_t *os = zfsvfs->z_os;
2385 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2386 return (SET_ERROR(EINVAL));
2388 if (newvers < zfsvfs->z_version)
2389 return (SET_ERROR(EINVAL));
2391 if (zfs_spa_version_map(newvers) >
2392 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2393 return (SET_ERROR(ENOTSUP));
2395 tx = dmu_tx_create(os);
2396 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2397 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2398 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2400 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2402 error = dmu_tx_assign(tx, TXG_WAIT);
2408 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2409 8, 1, &newvers, tx);
2416 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2419 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2421 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2422 DMU_OT_NONE, 0, tx);
2424 error = zap_add(os, MASTER_NODE_OBJ,
2425 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2428 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2429 sa_register_update_callback(os, zfs_sa_upgrade);
2432 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2433 "from %llu to %llu", zfsvfs->z_version, newvers);
2437 zfsvfs->z_version = newvers;
2439 zfs_set_fuid_feature(zfsvfs);
2445 * Read a property stored within the master node.
2448 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2454 * Look up the file system's value for the property. For the
2455 * version property, we look up a slightly different string.
2457 if (prop == ZFS_PROP_VERSION)
2458 pname = ZPL_VERSION_STR;
2460 pname = zfs_prop_to_name(prop);
2463 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2464 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2467 if (error == ENOENT) {
2468 /* No value set, use the default value */
2470 case ZFS_PROP_VERSION:
2471 *value = ZPL_VERSION;
2473 case ZFS_PROP_NORMALIZE:
2474 case ZFS_PROP_UTF8ONLY:
2478 *value = ZFS_CASE_SENSITIVE;
2490 zfsvfs_update_fromname(const char *oldname, const char *newname)
2492 char tmpbuf[MAXPATHLEN];
2497 oldlen = strlen(oldname);
2499 mtx_lock(&mountlist_mtx);
2500 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2501 fromname = mp->mnt_stat.f_mntfromname;
2502 if (strcmp(fromname, oldname) == 0) {
2503 (void)strlcpy(fromname, newname,
2504 sizeof(mp->mnt_stat.f_mntfromname));
2507 if (strncmp(fromname, oldname, oldlen) == 0 &&
2508 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2509 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
2510 newname, fromname + oldlen);
2511 (void)strlcpy(fromname, tmpbuf,
2512 sizeof(mp->mnt_stat.f_mntfromname));
2516 mtx_unlock(&mountlist_mtx);