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.
999 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
1001 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1005 error = zfsvfs_create_impl(zfvp, zfsvfs, os);
1007 dmu_objset_disown(os, zfsvfs);
1014 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
1018 zfsvfs->z_vfs = NULL;
1019 zfsvfs->z_parent = zfsvfs;
1021 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1022 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1023 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1024 offsetof(znode_t, z_link_node));
1026 rrm_init(&zfsvfs->z_teardown_lock, B_TRUE);
1028 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1030 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1031 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1032 for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1033 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1035 error = zfsvfs_init(zfsvfs, os);
1038 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1047 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1051 error = zfs_register_callbacks(zfsvfs->z_vfs);
1055 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1058 * If we are not mounting (ie: online recv), then we don't
1059 * have to worry about replaying the log as we blocked all
1060 * operations out since we closed the ZIL.
1066 * During replay we remove the read only flag to
1067 * allow replays to succeed.
1069 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1071 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1073 zfs_unlinked_drain(zfsvfs);
1076 * Parse and replay the intent log.
1078 * Because of ziltest, this must be done after
1079 * zfs_unlinked_drain(). (Further note: ziltest
1080 * doesn't use readonly mounts, where
1081 * zfs_unlinked_drain() isn't called.) This is because
1082 * ziltest causes spa_sync() to think it's committed,
1083 * but actually it is not, so the intent log contains
1084 * many txg's worth of changes.
1086 * In particular, if object N is in the unlinked set in
1087 * the last txg to actually sync, then it could be
1088 * actually freed in a later txg and then reallocated
1089 * in a yet later txg. This would write a "create
1090 * object N" record to the intent log. Normally, this
1091 * would be fine because the spa_sync() would have
1092 * written out the fact that object N is free, before
1093 * we could write the "create object N" intent log
1096 * But when we are in ziltest mode, we advance the "open
1097 * txg" without actually spa_sync()-ing the changes to
1098 * disk. So we would see that object N is still
1099 * allocated and in the unlinked set, and there is an
1100 * intent log record saying to allocate it.
1102 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1103 if (zil_replay_disable) {
1104 zil_destroy(zfsvfs->z_log, B_FALSE);
1106 zfsvfs->z_replay = B_TRUE;
1107 zil_replay(zfsvfs->z_os, zfsvfs,
1109 zfsvfs->z_replay = B_FALSE;
1112 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1116 * Set the objset user_ptr to track its zfsvfs.
1118 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1119 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1120 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1125 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1128 zfsvfs_free(zfsvfs_t *zfsvfs)
1133 * This is a barrier to prevent the filesystem from going away in
1134 * zfs_znode_move() until we can safely ensure that the filesystem is
1135 * not unmounted. We consider the filesystem valid before the barrier
1136 * and invalid after the barrier.
1138 rw_enter(&zfsvfs_lock, RW_READER);
1139 rw_exit(&zfsvfs_lock);
1141 zfs_fuid_destroy(zfsvfs);
1143 mutex_destroy(&zfsvfs->z_znodes_lock);
1144 mutex_destroy(&zfsvfs->z_lock);
1145 list_destroy(&zfsvfs->z_all_znodes);
1146 rrm_destroy(&zfsvfs->z_teardown_lock);
1147 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1148 rw_destroy(&zfsvfs->z_fuid_lock);
1149 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1150 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1151 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1155 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1157 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1158 if (zfsvfs->z_vfs) {
1159 if (zfsvfs->z_use_fuids) {
1160 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1161 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1162 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1163 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1164 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1165 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1167 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1168 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1169 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1170 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1171 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1172 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1175 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1179 zfs_domount(vfs_t *vfsp, char *osname)
1181 uint64_t recordsize, fsid_guid;
1189 error = zfsvfs_create(osname, &zfsvfs);
1192 zfsvfs->z_vfs = vfsp;
1195 /* Initialize the generic filesystem structure. */
1196 vfsp->vfs_bcount = 0;
1197 vfsp->vfs_data = NULL;
1199 if (zfs_create_unique_device(&mount_dev) == -1) {
1200 error = SET_ERROR(ENODEV);
1203 ASSERT(vfs_devismounted(mount_dev) == 0);
1206 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1209 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1210 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1212 vfsp->vfs_data = zfsvfs;
1213 vfsp->mnt_flag |= MNT_LOCAL;
1214 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1215 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1216 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1217 vfsp->mnt_kern_flag |= MNTK_NO_IOPF; /* vn_io_fault can be used */
1220 * The fsid is 64 bits, composed of an 8-bit fs type, which
1221 * separates our fsid from any other filesystem types, and a
1222 * 56-bit objset unique ID. The objset unique ID is unique to
1223 * all objsets open on this system, provided by unique_create().
1224 * The 8-bit fs type must be put in the low bits of fsid[1]
1225 * because that's where other Solaris filesystems put it.
1227 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1228 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1229 vfsp->vfs_fsid.val[0] = fsid_guid;
1230 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1231 vfsp->mnt_vfc->vfc_typenum & 0xFF;
1234 * Set features for file system.
1236 zfs_set_fuid_feature(zfsvfs);
1237 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1238 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1239 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1240 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1241 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1242 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1243 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1245 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1247 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1250 atime_changed_cb(zfsvfs, B_FALSE);
1251 readonly_changed_cb(zfsvfs, B_TRUE);
1252 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1254 xattr_changed_cb(zfsvfs, pval);
1255 zfsvfs->z_issnap = B_TRUE;
1256 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1258 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1259 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1260 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1262 error = zfsvfs_setup(zfsvfs, B_TRUE);
1265 vfs_mountedfrom(vfsp, osname);
1267 if (!zfsvfs->z_issnap)
1268 zfsctl_create(zfsvfs);
1271 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1272 zfsvfs_free(zfsvfs);
1274 atomic_inc_32(&zfs_active_fs_count);
1281 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1283 objset_t *os = zfsvfs->z_os;
1285 if (!dmu_objset_is_snapshot(os))
1286 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1291 * Convert a decimal digit string to a uint64_t integer.
1294 str_to_uint64(char *str, uint64_t *objnum)
1299 if (*str < '0' || *str > '9')
1300 return (SET_ERROR(EINVAL));
1302 num = num*10 + *str++ - '0';
1310 * The boot path passed from the boot loader is in the form of
1311 * "rootpool-name/root-filesystem-object-number'. Convert this
1312 * string to a dataset name: "rootpool-name/root-filesystem-name".
1315 zfs_parse_bootfs(char *bpath, char *outpath)
1321 if (*bpath == 0 || *bpath == '/')
1322 return (SET_ERROR(EINVAL));
1324 (void) strcpy(outpath, bpath);
1326 slashp = strchr(bpath, '/');
1328 /* if no '/', just return the pool name */
1329 if (slashp == NULL) {
1333 /* if not a number, just return the root dataset name */
1334 if (str_to_uint64(slashp+1, &objnum)) {
1339 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1346 * Check that the hex label string is appropriate for the dataset being
1347 * mounted into the global_zone proper.
1349 * Return an error if the hex label string is not default or
1350 * admin_low/admin_high. For admin_low labels, the corresponding
1351 * dataset must be readonly.
1354 zfs_check_global_label(const char *dsname, const char *hexsl)
1356 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1358 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1360 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1361 /* must be readonly */
1364 if (dsl_prop_get_integer(dsname,
1365 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1366 return (SET_ERROR(EACCES));
1367 return (rdonly ? 0 : EACCES);
1369 return (SET_ERROR(EACCES));
1373 * Determine whether the mount is allowed according to MAC check.
1374 * by comparing (where appropriate) label of the dataset against
1375 * the label of the zone being mounted into. If the dataset has
1376 * no label, create one.
1378 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1381 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1384 zone_t *mntzone = NULL;
1385 ts_label_t *mnt_tsl;
1388 char ds_hexsl[MAXNAMELEN];
1390 retv = EACCES; /* assume the worst */
1393 * Start by getting the dataset label if it exists.
1395 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1396 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1398 return (SET_ERROR(EACCES));
1401 * If labeling is NOT enabled, then disallow the mount of datasets
1402 * which have a non-default label already. No other label checks
1405 if (!is_system_labeled()) {
1406 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1408 return (SET_ERROR(EACCES));
1412 * Get the label of the mountpoint. If mounting into the global
1413 * zone (i.e. mountpoint is not within an active zone and the
1414 * zoned property is off), the label must be default or
1415 * admin_low/admin_high only; no other checks are needed.
1417 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1418 if (mntzone->zone_id == GLOBAL_ZONEID) {
1423 if (dsl_prop_get_integer(osname,
1424 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1425 return (SET_ERROR(EACCES));
1427 return (zfs_check_global_label(osname, ds_hexsl));
1430 * This is the case of a zone dataset being mounted
1431 * initially, before the zone has been fully created;
1432 * allow this mount into global zone.
1437 mnt_tsl = mntzone->zone_slabel;
1438 ASSERT(mnt_tsl != NULL);
1439 label_hold(mnt_tsl);
1440 mnt_sl = label2bslabel(mnt_tsl);
1442 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1444 * The dataset doesn't have a real label, so fabricate one.
1448 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1449 dsl_prop_set_string(osname,
1450 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1451 ZPROP_SRC_LOCAL, str) == 0)
1454 kmem_free(str, strlen(str) + 1);
1455 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1457 * Now compare labels to complete the MAC check. If the
1458 * labels are equal then allow access. If the mountpoint
1459 * label dominates the dataset label, allow readonly access.
1460 * Otherwise, access is denied.
1462 if (blequal(mnt_sl, &ds_sl))
1464 else if (bldominates(mnt_sl, &ds_sl)) {
1465 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1470 label_rele(mnt_tsl);
1474 #endif /* SECLABEL */
1476 #ifdef OPENSOLARIS_MOUNTROOT
1478 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1481 static int zfsrootdone = 0;
1482 zfsvfs_t *zfsvfs = NULL;
1491 * The filesystem that we mount as root is defined in the
1492 * boot property "zfs-bootfs" with a format of
1493 * "poolname/root-dataset-objnum".
1495 if (why == ROOT_INIT) {
1497 return (SET_ERROR(EBUSY));
1499 * the process of doing a spa_load will require the
1500 * clock to be set before we could (for example) do
1501 * something better by looking at the timestamp on
1502 * an uberblock, so just set it to -1.
1506 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1507 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1509 return (SET_ERROR(EINVAL));
1511 zfs_devid = spa_get_bootprop("diskdevid");
1512 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1514 spa_free_bootprop(zfs_devid);
1516 spa_free_bootprop(zfs_bootfs);
1517 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1521 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1522 spa_free_bootprop(zfs_bootfs);
1523 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1528 spa_free_bootprop(zfs_bootfs);
1530 if (error = vfs_lock(vfsp))
1533 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1534 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1538 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1540 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1541 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1546 mutex_enter(&vp->v_lock);
1547 vp->v_flag |= VROOT;
1548 mutex_exit(&vp->v_lock);
1552 * Leave rootvp held. The root file system is never unmounted.
1555 vfs_add((struct vnode *)0, vfsp,
1556 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1560 } else if (why == ROOT_REMOUNT) {
1561 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1562 vfsp->vfs_flag |= VFS_REMOUNT;
1564 /* refresh mount options */
1565 zfs_unregister_callbacks(vfsp->vfs_data);
1566 return (zfs_register_callbacks(vfsp));
1568 } else if (why == ROOT_UNMOUNT) {
1569 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1570 (void) zfs_sync(vfsp, 0, 0);
1575 * if "why" is equal to anything else other than ROOT_INIT,
1576 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1578 return (SET_ERROR(ENOTSUP));
1580 #endif /* OPENSOLARIS_MOUNTROOT */
1583 getpoolname(const char *osname, char *poolname)
1587 p = strchr(osname, '/');
1589 if (strlen(osname) >= MAXNAMELEN)
1590 return (ENAMETOOLONG);
1591 (void) strcpy(poolname, osname);
1593 if (p - osname >= MAXNAMELEN)
1594 return (ENAMETOOLONG);
1595 (void) strncpy(poolname, osname, p - osname);
1596 poolname[p - osname] = '\0';
1603 zfs_mount(vfs_t *vfsp)
1605 kthread_t *td = curthread;
1606 vnode_t *mvp = vfsp->mnt_vnodecovered;
1607 cred_t *cr = td->td_ucred;
1613 if (mvp->v_type != VDIR)
1614 return (SET_ERROR(ENOTDIR));
1616 mutex_enter(&mvp->v_lock);
1617 if ((uap->flags & MS_REMOUNT) == 0 &&
1618 (uap->flags & MS_OVERLAY) == 0 &&
1619 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1620 mutex_exit(&mvp->v_lock);
1621 return (SET_ERROR(EBUSY));
1623 mutex_exit(&mvp->v_lock);
1626 * ZFS does not support passing unparsed data in via MS_DATA.
1627 * Users should use the MS_OPTIONSTR interface; this means
1628 * that all option parsing is already done and the options struct
1629 * can be interrogated.
1631 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1632 #else /* !illumos */
1633 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS))
1634 return (SET_ERROR(EPERM));
1636 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1637 return (SET_ERROR(EINVAL));
1638 #endif /* illumos */
1641 * If full-owner-access is enabled and delegated administration is
1642 * turned on, we must set nosuid.
1644 if (zfs_super_owner &&
1645 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1646 secpolicy_fs_mount_clearopts(cr, vfsp);
1650 * Check for mount privilege?
1652 * If we don't have privilege then see if
1653 * we have local permission to allow it
1655 error = secpolicy_fs_mount(cr, mvp, vfsp);
1657 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1660 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1664 * Make sure user is the owner of the mount point
1665 * or has sufficient privileges.
1668 vattr.va_mask = AT_UID;
1670 vn_lock(mvp, LK_SHARED | LK_RETRY);
1671 if (VOP_GETATTR(mvp, &vattr, cr)) {
1676 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1677 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1684 secpolicy_fs_mount_clearopts(cr, vfsp);
1688 * Refuse to mount a filesystem if we are in a local zone and the
1689 * dataset is not visible.
1691 if (!INGLOBALZONE(curthread) &&
1692 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1693 error = SET_ERROR(EPERM);
1698 error = zfs_mount_label_policy(vfsp, osname);
1703 vfsp->vfs_flag |= MNT_NFS4ACLS;
1706 * When doing a remount, we simply refresh our temporary properties
1707 * according to those options set in the current VFS options.
1709 if (vfsp->vfs_flag & MS_REMOUNT) {
1710 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1713 * Refresh mount options with z_teardown_lock blocking I/O while
1714 * the filesystem is in an inconsistent state.
1715 * The lock also serializes this code with filesystem
1716 * manipulations between entry to zfs_suspend_fs() and return
1717 * from zfs_resume_fs().
1719 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1720 zfs_unregister_callbacks(zfsvfs);
1721 error = zfs_register_callbacks(vfsp);
1722 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1726 /* Initial root mount: try hard to import the requested root pool. */
1727 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1728 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1729 char pname[MAXNAMELEN];
1731 error = getpoolname(osname, pname);
1733 error = spa_import_rootpool(pname);
1738 error = zfs_domount(vfsp, osname);
1743 * Add an extra VFS_HOLD on our parent vfs so that it can't
1744 * disappear due to a forced unmount.
1746 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1747 VFS_HOLD(mvp->v_vfsp);
1755 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1757 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1758 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1760 statp->f_version = STATFS_VERSION;
1764 dmu_objset_space(zfsvfs->z_os,
1765 &refdbytes, &availbytes, &usedobjs, &availobjs);
1768 * The underlying storage pool actually uses multiple block sizes.
1769 * We report the fragsize as the smallest block size we support,
1770 * and we report our blocksize as the filesystem's maximum blocksize.
1772 statp->f_bsize = SPA_MINBLOCKSIZE;
1773 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1776 * The following report "total" blocks of various kinds in the
1777 * file system, but reported in terms of f_frsize - the
1781 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1782 statp->f_bfree = availbytes / statp->f_bsize;
1783 statp->f_bavail = statp->f_bfree; /* no root reservation */
1786 * statvfs() should really be called statufs(), because it assumes
1787 * static metadata. ZFS doesn't preallocate files, so the best
1788 * we can do is report the max that could possibly fit in f_files,
1789 * and that minus the number actually used in f_ffree.
1790 * For f_ffree, report the smaller of the number of object available
1791 * and the number of blocks (each object will take at least a block).
1793 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1794 statp->f_files = statp->f_ffree + usedobjs;
1797 * We're a zfs filesystem.
1799 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
1801 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
1802 sizeof(statp->f_mntfromname));
1803 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
1804 sizeof(statp->f_mntonname));
1806 statp->f_namemax = MAXNAMELEN - 1;
1813 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
1815 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1821 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1823 *vpp = ZTOV(rootzp);
1828 error = vn_lock(*vpp, flags);
1838 * Teardown the zfsvfs::z_os.
1840 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1841 * and 'z_teardown_inactive_lock' held.
1844 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1848 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1852 * We purge the parent filesystem's vfsp as the parent
1853 * filesystem and all of its snapshots have their vnode's
1854 * v_vfsp set to the parent's filesystem's vfsp. Note,
1855 * 'z_parent' is self referential for non-snapshots.
1857 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1858 #ifdef FREEBSD_NAMECACHE
1859 cache_purgevfs(zfsvfs->z_parent->z_vfs, true);
1864 * Close the zil. NB: Can't close the zil while zfs_inactive
1865 * threads are blocked as zil_close can call zfs_inactive.
1867 if (zfsvfs->z_log) {
1868 zil_close(zfsvfs->z_log);
1869 zfsvfs->z_log = NULL;
1872 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1875 * If we are not unmounting (ie: online recv) and someone already
1876 * unmounted this file system while we were doing the switcheroo,
1877 * or a reopen of z_os failed then just bail out now.
1879 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1880 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1881 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1882 return (SET_ERROR(EIO));
1886 * At this point there are no vops active, and any new vops will
1887 * fail with EIO since we have z_teardown_lock for writer (only
1888 * relavent for forced unmount).
1890 * Release all holds on dbufs.
1892 mutex_enter(&zfsvfs->z_znodes_lock);
1893 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1894 zp = list_next(&zfsvfs->z_all_znodes, zp))
1896 ASSERT(ZTOV(zp)->v_count >= 0);
1897 zfs_znode_dmu_fini(zp);
1899 mutex_exit(&zfsvfs->z_znodes_lock);
1902 * If we are unmounting, set the unmounted flag and let new vops
1903 * unblock. zfs_inactive will have the unmounted behavior, and all
1904 * other vops will fail with EIO.
1907 zfsvfs->z_unmounted = B_TRUE;
1908 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1909 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1913 * z_os will be NULL if there was an error in attempting to reopen
1914 * zfsvfs, so just return as the properties had already been
1915 * unregistered and cached data had been evicted before.
1917 if (zfsvfs->z_os == NULL)
1921 * Unregister properties.
1923 zfs_unregister_callbacks(zfsvfs);
1928 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1929 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1930 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1931 dmu_objset_evict_dbufs(zfsvfs->z_os);
1938 zfs_umount(vfs_t *vfsp, int fflag)
1940 kthread_t *td = curthread;
1941 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1943 cred_t *cr = td->td_ucred;
1946 ret = secpolicy_fs_unmount(cr, vfsp);
1948 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1949 ZFS_DELEG_PERM_MOUNT, cr))
1954 * We purge the parent filesystem's vfsp as the parent filesystem
1955 * and all of its snapshots have their vnode's v_vfsp set to the
1956 * parent's filesystem's vfsp. Note, 'z_parent' is self
1957 * referential for non-snapshots.
1959 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1962 * Unmount any snapshots mounted under .zfs before unmounting the
1965 if (zfsvfs->z_ctldir != NULL) {
1966 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
1970 if (fflag & MS_FORCE) {
1972 * Mark file system as unmounted before calling
1973 * vflush(FORCECLOSE). This way we ensure no future vnops
1974 * will be called and risk operating on DOOMED vnodes.
1976 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1977 zfsvfs->z_unmounted = B_TRUE;
1978 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1982 * Flush all the files.
1984 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
1989 if (!(fflag & MS_FORCE)) {
1991 * Check the number of active vnodes in the file system.
1992 * Our count is maintained in the vfs structure, but the
1993 * number is off by 1 to indicate a hold on the vfs
1996 * The '.zfs' directory maintains a reference of its
1997 * own, and any active references underneath are
1998 * reflected in the vnode count.
2000 if (zfsvfs->z_ctldir == NULL) {
2001 if (vfsp->vfs_count > 1)
2002 return (SET_ERROR(EBUSY));
2004 if (vfsp->vfs_count > 2 ||
2005 zfsvfs->z_ctldir->v_count > 1)
2006 return (SET_ERROR(EBUSY));
2011 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2015 * z_os will be NULL if there was an error in
2016 * attempting to reopen zfsvfs.
2020 * Unset the objset user_ptr.
2022 mutex_enter(&os->os_user_ptr_lock);
2023 dmu_objset_set_user(os, NULL);
2024 mutex_exit(&os->os_user_ptr_lock);
2027 * Finally release the objset
2029 dmu_objset_disown(os, zfsvfs);
2033 * We can now safely destroy the '.zfs' directory node.
2035 if (zfsvfs->z_ctldir != NULL)
2036 zfsctl_destroy(zfsvfs);
2043 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2045 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2050 * zfs_zget() can't operate on virtual entries like .zfs/ or
2051 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2052 * This will make NFS to switch to LOOKUP instead of using VGET.
2054 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2055 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2056 return (EOPNOTSUPP);
2059 err = zfs_zget(zfsvfs, ino, &zp);
2060 if (err == 0 && zp->z_unlinked) {
2068 err = vn_lock(*vpp, flags);
2075 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
2076 struct ucred **credanonp, int *numsecflavors, int **secflavors)
2078 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2081 * If this is regular file system vfsp is the same as
2082 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2083 * zfsvfs->z_parent->z_vfs represents parent file system
2084 * which we have to use here, because only this file system
2085 * has mnt_export configured.
2087 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2088 credanonp, numsecflavors, secflavors));
2091 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2092 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2095 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2097 struct componentname cn;
2098 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2101 uint64_t object = 0;
2102 uint64_t fid_gen = 0;
2112 * On FreeBSD we can get snapshot's mount point or its parent file
2113 * system mount point depending if snapshot is already mounted or not.
2115 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2116 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2117 uint64_t objsetid = 0;
2118 uint64_t setgen = 0;
2120 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2121 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2123 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2124 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2128 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2130 return (SET_ERROR(EINVAL));
2134 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2135 zfid_short_t *zfid = (zfid_short_t *)fidp;
2137 for (i = 0; i < sizeof (zfid->zf_object); i++)
2138 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2140 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2141 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2144 return (SET_ERROR(EINVAL));
2148 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2149 * directory tree. If the object == zfsvfs->z_shares_dir, then
2150 * we are in the .zfs/shares directory tree.
2152 if ((fid_gen == 0 &&
2153 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2154 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2156 VERIFY0(zfsctl_root(zfsvfs, LK_SHARED, &dvp));
2157 if (object == ZFSCTL_INO_SNAPDIR) {
2158 cn.cn_nameptr = "snapshot";
2159 cn.cn_namelen = strlen(cn.cn_nameptr);
2160 cn.cn_nameiop = LOOKUP;
2161 cn.cn_flags = ISLASTCN | LOCKLEAF;
2162 cn.cn_lkflags = flags;
2163 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2165 } else if (object == zfsvfs->z_shares_dir) {
2167 * XXX This branch must not be taken,
2168 * if it is, then the lookup below will
2171 cn.cn_nameptr = "shares";
2172 cn.cn_namelen = strlen(cn.cn_nameptr);
2173 cn.cn_nameiop = LOOKUP;
2174 cn.cn_flags = ISLASTCN;
2175 cn.cn_lkflags = flags;
2176 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2184 gen_mask = -1ULL >> (64 - 8 * i);
2186 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2187 if (err = zfs_zget(zfsvfs, object, &zp)) {
2191 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2193 zp_gen = zp_gen & gen_mask;
2196 if (zp->z_unlinked || zp_gen != fid_gen) {
2197 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2200 return (SET_ERROR(EINVAL));
2205 err = vn_lock(*vpp, flags);
2207 vnode_create_vobject(*vpp, zp->z_size, curthread);
2214 * Block out VOPs and close zfsvfs_t::z_os
2216 * Note, if successful, then we return with the 'z_teardown_lock' and
2217 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2218 * dataset and objset intact so that they can be atomically handed off during
2219 * a subsequent rollback or recv operation and the resume thereafter.
2222 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2226 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2233 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2234 * is an invariant across any of the operations that can be performed while the
2235 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2236 * are the same: the relevant objset and associated dataset are owned by
2237 * zfsvfs, held, and long held on entry.
2240 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2245 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2246 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2249 * We already own this, so just update the objset_t, as the one we
2250 * had before may have been evicted.
2253 VERIFY3P(ds->ds_owner, ==, zfsvfs);
2254 VERIFY(dsl_dataset_long_held(ds));
2255 VERIFY0(dmu_objset_from_ds(ds, &os));
2257 err = zfsvfs_init(zfsvfs, os);
2261 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2263 zfs_set_fuid_feature(zfsvfs);
2266 * Attempt to re-establish all the active znodes with
2267 * their dbufs. If a zfs_rezget() fails, then we'll let
2268 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2269 * when they try to use their znode.
2271 mutex_enter(&zfsvfs->z_znodes_lock);
2272 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2273 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2274 (void) zfs_rezget(zp);
2276 mutex_exit(&zfsvfs->z_znodes_lock);
2279 /* release the VOPs */
2280 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2281 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2285 * Since we couldn't setup the sa framework, try to force
2286 * unmount this file system.
2288 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) {
2289 vfs_ref(zfsvfs->z_vfs);
2290 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2297 zfs_freevfs(vfs_t *vfsp)
2299 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2303 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2304 * from zfs_mount(). Release it here. If we came through
2305 * zfs_mountroot() instead, we didn't grab an extra hold, so
2306 * skip the VFS_RELE for rootvfs.
2308 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2309 VFS_RELE(zfsvfs->z_parent->z_vfs);
2312 zfsvfs_free(zfsvfs);
2314 atomic_dec_32(&zfs_active_fs_count);
2318 static int desiredvnodes_backup;
2322 zfs_vnodes_adjust(void)
2325 int newdesiredvnodes;
2327 desiredvnodes_backup = desiredvnodes;
2330 * We calculate newdesiredvnodes the same way it is done in
2331 * vntblinit(). If it is equal to desiredvnodes, it means that
2332 * it wasn't tuned by the administrator and we can tune it down.
2334 newdesiredvnodes = min(maxproc + vm_cnt.v_page_count / 4, 2 *
2335 vm_kmem_size / (5 * (sizeof(struct vm_object) +
2336 sizeof(struct vnode))));
2337 if (newdesiredvnodes == desiredvnodes)
2338 desiredvnodes = (3 * newdesiredvnodes) / 4;
2343 zfs_vnodes_adjust_back(void)
2347 desiredvnodes = desiredvnodes_backup;
2355 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2358 * Initialize .zfs directory structures
2363 * Initialize znode cache, vnode ops, etc...
2368 * Reduce number of vnodes. Originally number of vnodes is calculated
2369 * with UFS inode in mind. We reduce it here, because it's too big for
2372 zfs_vnodes_adjust();
2374 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2382 zfs_vnodes_adjust_back();
2388 return (zfs_active_fs_count != 0);
2392 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2395 objset_t *os = zfsvfs->z_os;
2398 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2399 return (SET_ERROR(EINVAL));
2401 if (newvers < zfsvfs->z_version)
2402 return (SET_ERROR(EINVAL));
2404 if (zfs_spa_version_map(newvers) >
2405 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2406 return (SET_ERROR(ENOTSUP));
2408 tx = dmu_tx_create(os);
2409 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2410 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2411 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2413 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2415 error = dmu_tx_assign(tx, TXG_WAIT);
2421 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2422 8, 1, &newvers, tx);
2429 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2432 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2434 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2435 DMU_OT_NONE, 0, tx);
2437 error = zap_add(os, MASTER_NODE_OBJ,
2438 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2441 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2442 sa_register_update_callback(os, zfs_sa_upgrade);
2445 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2446 "from %llu to %llu", zfsvfs->z_version, newvers);
2450 zfsvfs->z_version = newvers;
2452 zfs_set_fuid_feature(zfsvfs);
2458 * Read a property stored within the master node.
2461 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2467 * Look up the file system's value for the property. For the
2468 * version property, we look up a slightly different string.
2470 if (prop == ZFS_PROP_VERSION)
2471 pname = ZPL_VERSION_STR;
2473 pname = zfs_prop_to_name(prop);
2476 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2477 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2480 if (error == ENOENT) {
2481 /* No value set, use the default value */
2483 case ZFS_PROP_VERSION:
2484 *value = ZPL_VERSION;
2486 case ZFS_PROP_NORMALIZE:
2487 case ZFS_PROP_UTF8ONLY:
2491 *value = ZFS_CASE_SENSITIVE;
2503 zfsvfs_update_fromname(const char *oldname, const char *newname)
2505 char tmpbuf[MAXPATHLEN];
2510 oldlen = strlen(oldname);
2512 mtx_lock(&mountlist_mtx);
2513 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2514 fromname = mp->mnt_stat.f_mntfromname;
2515 if (strcmp(fromname, oldname) == 0) {
2516 (void)strlcpy(fromname, newname,
2517 sizeof(mp->mnt_stat.f_mntfromname));
2520 if (strncmp(fromname, oldname, oldlen) == 0 &&
2521 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2522 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
2523 newname, fromname + oldlen);
2524 (void)strlcpy(fromname, tmpbuf,
2525 sizeof(mp->mnt_stat.f_mntfromname));
2529 mtx_unlock(&mountlist_mtx);