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) 2013 by Delphix. All rights reserved.
28 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
34 #include <sys/sysmacros.h>
37 #include <sys/vnode.h>
39 #include <sys/mntent.h>
40 #include <sys/mount.h>
41 #include <sys/cmn_err.h>
42 #include <sys/zfs_znode.h>
43 #include <sys/zfs_dir.h>
45 #include <sys/fs/zfs.h>
47 #include <sys/dsl_prop.h>
48 #include <sys/dsl_dataset.h>
49 #include <sys/dsl_deleg.h>
53 #include <sys/sa_impl.h>
54 #include <sys/varargs.h>
55 #include <sys/policy.h>
56 #include <sys/atomic.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/zfs_ctldir.h>
59 #include <sys/zfs_fuid.h>
60 #include <sys/sunddi.h>
62 #include <sys/dmu_objset.h>
63 #include <sys/spa_boot.h>
65 #include "zfs_comutil.h"
67 struct mtx zfs_debug_mtx;
68 MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF);
70 SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system");
73 SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0,
74 "File system owner can perform privileged operation on his file systems");
77 TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level);
78 SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0,
81 SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions");
82 static int zfs_version_acl = ZFS_ACL_VERSION;
83 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0,
85 static int zfs_version_spa = SPA_VERSION;
86 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0,
88 static int zfs_version_zpl = ZPL_VERSION;
89 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0,
92 static int zfs_mount(vfs_t *vfsp);
93 static int zfs_umount(vfs_t *vfsp, int fflag);
94 static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
95 static int zfs_statfs(vfs_t *vfsp, struct statfs *statp);
96 static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp);
97 static int zfs_sync(vfs_t *vfsp, int waitfor);
98 static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
99 struct ucred **credanonp, int *numsecflavors, int **secflavors);
100 static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp);
101 static void zfs_objset_close(zfsvfs_t *zfsvfs);
102 static void zfs_freevfs(vfs_t *vfsp);
104 static struct vfsops zfs_vfsops = {
105 .vfs_mount = zfs_mount,
106 .vfs_unmount = zfs_umount,
107 .vfs_root = zfs_root,
108 .vfs_statfs = zfs_statfs,
109 .vfs_vget = zfs_vget,
110 .vfs_sync = zfs_sync,
111 .vfs_checkexp = zfs_checkexp,
112 .vfs_fhtovp = zfs_fhtovp,
115 VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN);
118 * We need to keep a count of active fs's.
119 * This is necessary to prevent our module
120 * from being unloaded after a umount -f
122 static uint32_t zfs_active_fs_count = 0;
126 zfs_sync(vfs_t *vfsp, int waitfor)
130 * Data integrity is job one. We don't want a compromised kernel
131 * writing to the storage pool, so we never sync during panic.
138 * Sync a specific filesystem.
140 zfsvfs_t *zfsvfs = vfsp->vfs_data;
144 error = vfs_stdsync(vfsp, waitfor);
149 dp = dmu_objset_pool(zfsvfs->z_os);
152 * If the system is shutting down, then skip any
153 * filesystems which may exist on a suspended pool.
155 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
160 if (zfsvfs->z_log != NULL)
161 zil_commit(zfsvfs->z_log, 0);
166 * Sync all ZFS filesystems. This is what happens when you
167 * run sync(1M). Unlike other filesystems, ZFS honors the
168 * request by waiting for all pools to commit all dirty data.
178 zfs_create_unique_device(dev_t *dev)
183 ASSERT3U(zfs_minor, <=, MAXMIN32);
184 minor_t start = zfs_minor;
186 mutex_enter(&zfs_dev_mtx);
187 if (zfs_minor >= MAXMIN32) {
189 * If we're still using the real major
190 * keep out of /dev/zfs and /dev/zvol minor
191 * number space. If we're using a getudev()'ed
192 * major number, we can use all of its minors.
194 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
195 zfs_minor = ZFS_MIN_MINOR;
201 *dev = makedevice(zfs_major, zfs_minor);
202 mutex_exit(&zfs_dev_mtx);
203 } while (vfs_devismounted(*dev) && zfs_minor != start);
204 if (zfs_minor == start) {
206 * We are using all ~262,000 minor numbers for the
207 * current major number. Create a new major number.
209 if ((new_major = getudev()) == (major_t)-1) {
211 "zfs_mount: Can't get unique major "
215 mutex_enter(&zfs_dev_mtx);
216 zfs_major = new_major;
219 mutex_exit(&zfs_dev_mtx);
223 /* CONSTANTCONDITION */
228 #endif /* !__FreeBSD__ */
231 atime_changed_cb(void *arg, uint64_t newval)
233 zfsvfs_t *zfsvfs = arg;
235 if (newval == TRUE) {
236 zfsvfs->z_atime = TRUE;
237 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
238 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
239 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
241 zfsvfs->z_atime = FALSE;
242 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
243 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
244 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
249 xattr_changed_cb(void *arg, uint64_t newval)
251 zfsvfs_t *zfsvfs = arg;
253 if (newval == TRUE) {
254 /* XXX locking on vfs_flag? */
256 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
258 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
259 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
261 /* XXX locking on vfs_flag? */
263 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
265 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
266 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
271 blksz_changed_cb(void *arg, uint64_t newval)
273 zfsvfs_t *zfsvfs = arg;
275 if (newval < SPA_MINBLOCKSIZE ||
276 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
277 newval = SPA_MAXBLOCKSIZE;
279 zfsvfs->z_max_blksz = newval;
280 zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
284 readonly_changed_cb(void *arg, uint64_t newval)
286 zfsvfs_t *zfsvfs = arg;
289 /* XXX locking on vfs_flag? */
290 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
291 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
292 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
294 /* XXX locking on vfs_flag? */
295 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
296 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
297 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
302 setuid_changed_cb(void *arg, uint64_t newval)
304 zfsvfs_t *zfsvfs = arg;
306 if (newval == FALSE) {
307 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
308 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
309 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
311 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
312 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
313 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
318 exec_changed_cb(void *arg, uint64_t newval)
320 zfsvfs_t *zfsvfs = arg;
322 if (newval == FALSE) {
323 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
324 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
325 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
327 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
328 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
329 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
334 * The nbmand mount option can be changed at mount time.
335 * We can't allow it to be toggled on live file systems or incorrect
336 * behavior may be seen from cifs clients
338 * This property isn't registered via dsl_prop_register(), but this callback
339 * will be called when a file system is first mounted
342 nbmand_changed_cb(void *arg, uint64_t newval)
344 zfsvfs_t *zfsvfs = arg;
345 if (newval == FALSE) {
346 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
347 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
349 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
350 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
355 snapdir_changed_cb(void *arg, uint64_t newval)
357 zfsvfs_t *zfsvfs = arg;
359 zfsvfs->z_show_ctldir = newval;
363 vscan_changed_cb(void *arg, uint64_t newval)
365 zfsvfs_t *zfsvfs = arg;
367 zfsvfs->z_vscan = newval;
371 acl_mode_changed_cb(void *arg, uint64_t newval)
373 zfsvfs_t *zfsvfs = arg;
375 zfsvfs->z_acl_mode = newval;
379 acl_inherit_changed_cb(void *arg, uint64_t newval)
381 zfsvfs_t *zfsvfs = arg;
383 zfsvfs->z_acl_inherit = newval;
387 zfs_register_callbacks(vfs_t *vfsp)
389 struct dsl_dataset *ds = NULL;
391 zfsvfs_t *zfsvfs = NULL;
393 boolean_t readonly = B_FALSE;
394 boolean_t do_readonly = B_FALSE;
395 boolean_t setuid = B_FALSE;
396 boolean_t do_setuid = B_FALSE;
397 boolean_t exec = B_FALSE;
398 boolean_t do_exec = B_FALSE;
400 boolean_t devices = B_FALSE;
401 boolean_t do_devices = B_FALSE;
403 boolean_t xattr = B_FALSE;
404 boolean_t do_xattr = B_FALSE;
405 boolean_t atime = B_FALSE;
406 boolean_t do_atime = B_FALSE;
410 zfsvfs = vfsp->vfs_data;
415 * This function can be called for a snapshot when we update snapshot's
416 * mount point, which isn't really supported.
418 if (dmu_objset_is_snapshot(os))
422 * The act of registering our callbacks will destroy any mount
423 * options we may have. In order to enable temporary overrides
424 * of mount options, we stash away the current values and
425 * restore them after we register the callbacks.
427 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
428 !spa_writeable(dmu_objset_spa(os))) {
430 do_readonly = B_TRUE;
431 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
433 do_readonly = B_TRUE;
435 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
439 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
442 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
447 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
450 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
454 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
457 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
461 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
464 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
470 * nbmand is a special property. It can only be changed at
473 * This is weird, but it is documented to only be changeable
476 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
478 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
481 char osname[MAXNAMELEN];
483 dmu_objset_name(os, osname);
484 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
491 * Register property callbacks.
493 * It would probably be fine to just check for i/o error from
494 * the first prop_register(), but I guess I like to go
497 ds = dmu_objset_ds(os);
498 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
499 error = dsl_prop_register(ds,
500 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
501 error = error ? error : dsl_prop_register(ds,
502 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
503 error = error ? error : dsl_prop_register(ds,
504 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
505 error = error ? error : dsl_prop_register(ds,
506 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
508 error = error ? error : dsl_prop_register(ds,
509 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
511 error = error ? error : dsl_prop_register(ds,
512 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
513 error = error ? error : dsl_prop_register(ds,
514 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
515 error = error ? error : dsl_prop_register(ds,
516 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
517 error = error ? error : dsl_prop_register(ds,
518 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
519 error = error ? error : dsl_prop_register(ds,
520 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
522 error = error ? error : dsl_prop_register(ds,
523 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
524 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
529 * Invoke our callbacks to restore temporary mount options.
532 readonly_changed_cb(zfsvfs, readonly);
534 setuid_changed_cb(zfsvfs, setuid);
536 exec_changed_cb(zfsvfs, exec);
538 xattr_changed_cb(zfsvfs, xattr);
540 atime_changed_cb(zfsvfs, atime);
542 nbmand_changed_cb(zfsvfs, nbmand);
548 * We may attempt to unregister some callbacks that are not
549 * registered, but this is OK; it will simply return ENOMSG,
550 * which we will ignore.
552 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
553 atime_changed_cb, zfsvfs);
554 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
555 xattr_changed_cb, zfsvfs);
556 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
557 blksz_changed_cb, zfsvfs);
558 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
559 readonly_changed_cb, zfsvfs);
561 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
562 devices_changed_cb, zfsvfs);
564 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
565 setuid_changed_cb, zfsvfs);
566 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
567 exec_changed_cb, zfsvfs);
568 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
569 snapdir_changed_cb, zfsvfs);
570 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE),
571 acl_mode_changed_cb, zfsvfs);
572 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
573 acl_inherit_changed_cb, zfsvfs);
574 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
575 vscan_changed_cb, zfsvfs);
580 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
581 uint64_t *userp, uint64_t *groupp)
584 * Is it a valid type of object to track?
586 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
587 return (SET_ERROR(ENOENT));
590 * If we have a NULL data pointer
591 * then assume the id's aren't changing and
592 * return EEXIST to the dmu to let it know to
596 return (SET_ERROR(EEXIST));
598 if (bonustype == DMU_OT_ZNODE) {
599 znode_phys_t *znp = data;
600 *userp = znp->zp_uid;
601 *groupp = znp->zp_gid;
604 sa_hdr_phys_t *sap = data;
605 sa_hdr_phys_t sa = *sap;
606 boolean_t swap = B_FALSE;
608 ASSERT(bonustype == DMU_OT_SA);
610 if (sa.sa_magic == 0) {
612 * This should only happen for newly created
613 * files that haven't had the znode data filled
620 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
621 sa.sa_magic = SA_MAGIC;
622 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
625 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
628 hdrsize = sa_hdrsize(&sa);
629 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
630 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
632 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
635 *userp = BSWAP_64(*userp);
636 *groupp = BSWAP_64(*groupp);
643 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
644 char *domainbuf, int buflen, uid_t *ridp)
649 fuid = strtonum(fuidstr, NULL);
651 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
653 (void) strlcpy(domainbuf, domain, buflen);
656 *ridp = FUID_RID(fuid);
660 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
663 case ZFS_PROP_USERUSED:
664 return (DMU_USERUSED_OBJECT);
665 case ZFS_PROP_GROUPUSED:
666 return (DMU_GROUPUSED_OBJECT);
667 case ZFS_PROP_USERQUOTA:
668 return (zfsvfs->z_userquota_obj);
669 case ZFS_PROP_GROUPQUOTA:
670 return (zfsvfs->z_groupquota_obj);
676 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
677 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
682 zfs_useracct_t *buf = vbuf;
685 if (!dmu_objset_userspace_present(zfsvfs->z_os))
686 return (SET_ERROR(ENOTSUP));
688 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
694 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
695 (error = zap_cursor_retrieve(&zc, &za)) == 0;
696 zap_cursor_advance(&zc)) {
697 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
701 fuidstr_to_sid(zfsvfs, za.za_name,
702 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
704 buf->zu_space = za.za_first_integer;
710 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
711 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
712 *cookiep = zap_cursor_serialize(&zc);
713 zap_cursor_fini(&zc);
718 * buf must be big enough (eg, 32 bytes)
721 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
722 char *buf, boolean_t addok)
727 if (domain && domain[0]) {
728 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
730 return (SET_ERROR(ENOENT));
732 fuid = FUID_ENCODE(domainid, rid);
733 (void) sprintf(buf, "%llx", (longlong_t)fuid);
738 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
739 const char *domain, uint64_t rid, uint64_t *valp)
747 if (!dmu_objset_userspace_present(zfsvfs->z_os))
748 return (SET_ERROR(ENOTSUP));
750 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
754 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
758 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
765 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
766 const char *domain, uint64_t rid, uint64_t quota)
772 boolean_t fuid_dirtied;
774 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
775 return (SET_ERROR(EINVAL));
777 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
778 return (SET_ERROR(ENOTSUP));
780 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
781 &zfsvfs->z_groupquota_obj;
783 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
786 fuid_dirtied = zfsvfs->z_fuid_dirty;
788 tx = dmu_tx_create(zfsvfs->z_os);
789 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
791 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
792 zfs_userquota_prop_prefixes[type]);
795 zfs_fuid_txhold(zfsvfs, tx);
796 err = dmu_tx_assign(tx, TXG_WAIT);
802 mutex_enter(&zfsvfs->z_lock);
804 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
806 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
807 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
809 mutex_exit(&zfsvfs->z_lock);
812 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
816 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
820 zfs_fuid_sync(zfsvfs, tx);
826 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
829 uint64_t used, quota, usedobj, quotaobj;
832 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
833 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
835 if (quotaobj == 0 || zfsvfs->z_replay)
838 (void) sprintf(buf, "%llx", (longlong_t)fuid);
839 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
843 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
846 return (used >= quota);
850 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
855 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
857 fuid = isgroup ? zp->z_gid : zp->z_uid;
859 if (quotaobj == 0 || zfsvfs->z_replay)
862 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
866 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
874 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
877 * We claim to always be readonly so we can open snapshots;
878 * other ZPL code will prevent us from writing to snapshots.
880 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
882 kmem_free(zfsvfs, sizeof (zfsvfs_t));
887 * Initialize the zfs-specific filesystem structure.
888 * Should probably make this a kmem cache, shuffle fields,
889 * and just bzero up to z_hold_mtx[].
891 zfsvfs->z_vfs = NULL;
892 zfsvfs->z_parent = zfsvfs;
893 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
894 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
897 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
900 } else if (zfsvfs->z_version >
901 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
902 (void) printf("Can't mount a version %lld file system "
903 "on a version %lld pool\n. Pool must be upgraded to mount "
904 "this file system.", (u_longlong_t)zfsvfs->z_version,
905 (u_longlong_t)spa_version(dmu_objset_spa(os)));
906 error = SET_ERROR(ENOTSUP);
909 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
911 zfsvfs->z_norm = (int)zval;
913 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
915 zfsvfs->z_utf8 = (zval != 0);
917 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
919 zfsvfs->z_case = (uint_t)zval;
922 * Fold case on file systems that are always or sometimes case
925 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
926 zfsvfs->z_case == ZFS_CASE_MIXED)
927 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
929 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
930 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
932 if (zfsvfs->z_use_sa) {
933 /* should either have both of these objects or none */
934 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
940 * Pre SA versions file systems should never touch
941 * either the attribute registration or layout objects.
946 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
947 &zfsvfs->z_attr_table);
951 if (zfsvfs->z_version >= ZPL_VERSION_SA)
952 sa_register_update_callback(os, zfs_sa_upgrade);
954 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
958 ASSERT(zfsvfs->z_root != 0);
960 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
961 &zfsvfs->z_unlinkedobj);
965 error = zap_lookup(os, MASTER_NODE_OBJ,
966 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
967 8, 1, &zfsvfs->z_userquota_obj);
968 if (error && error != ENOENT)
971 error = zap_lookup(os, MASTER_NODE_OBJ,
972 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
973 8, 1, &zfsvfs->z_groupquota_obj);
974 if (error && error != ENOENT)
977 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
978 &zfsvfs->z_fuid_obj);
979 if (error && error != ENOENT)
982 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
983 &zfsvfs->z_shares_dir);
984 if (error && error != ENOENT)
987 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
988 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
989 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
990 offsetof(znode_t, z_link_node));
991 rrw_init(&zfsvfs->z_teardown_lock, B_FALSE);
992 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
993 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
994 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
995 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1001 dmu_objset_disown(os, zfsvfs);
1003 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1008 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1012 error = zfs_register_callbacks(zfsvfs->z_vfs);
1017 * Set the objset user_ptr to track its zfsvfs.
1019 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1020 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1021 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1023 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1026 * If we are not mounting (ie: online recv), then we don't
1027 * have to worry about replaying the log as we blocked all
1028 * operations out since we closed the ZIL.
1034 * During replay we remove the read only flag to
1035 * allow replays to succeed.
1037 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1039 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1041 zfs_unlinked_drain(zfsvfs);
1044 * Parse and replay the intent log.
1046 * Because of ziltest, this must be done after
1047 * zfs_unlinked_drain(). (Further note: ziltest
1048 * doesn't use readonly mounts, where
1049 * zfs_unlinked_drain() isn't called.) This is because
1050 * ziltest causes spa_sync() to think it's committed,
1051 * but actually it is not, so the intent log contains
1052 * many txg's worth of changes.
1054 * In particular, if object N is in the unlinked set in
1055 * the last txg to actually sync, then it could be
1056 * actually freed in a later txg and then reallocated
1057 * in a yet later txg. This would write a "create
1058 * object N" record to the intent log. Normally, this
1059 * would be fine because the spa_sync() would have
1060 * written out the fact that object N is free, before
1061 * we could write the "create object N" intent log
1064 * But when we are in ziltest mode, we advance the "open
1065 * txg" without actually spa_sync()-ing the changes to
1066 * disk. So we would see that object N is still
1067 * allocated and in the unlinked set, and there is an
1068 * intent log record saying to allocate it.
1070 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1071 if (zil_replay_disable) {
1072 zil_destroy(zfsvfs->z_log, B_FALSE);
1074 zfsvfs->z_replay = B_TRUE;
1075 zil_replay(zfsvfs->z_os, zfsvfs,
1077 zfsvfs->z_replay = B_FALSE;
1080 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1086 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1089 zfsvfs_free(zfsvfs_t *zfsvfs)
1094 * This is a barrier to prevent the filesystem from going away in
1095 * zfs_znode_move() until we can safely ensure that the filesystem is
1096 * not unmounted. We consider the filesystem valid before the barrier
1097 * and invalid after the barrier.
1099 rw_enter(&zfsvfs_lock, RW_READER);
1100 rw_exit(&zfsvfs_lock);
1102 zfs_fuid_destroy(zfsvfs);
1104 mutex_destroy(&zfsvfs->z_znodes_lock);
1105 mutex_destroy(&zfsvfs->z_lock);
1106 list_destroy(&zfsvfs->z_all_znodes);
1107 rrw_destroy(&zfsvfs->z_teardown_lock);
1108 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1109 rw_destroy(&zfsvfs->z_fuid_lock);
1110 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1111 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1112 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1116 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1118 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1119 if (zfsvfs->z_vfs) {
1120 if (zfsvfs->z_use_fuids) {
1121 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1122 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1123 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1124 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1125 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1126 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1128 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1129 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1130 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1131 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1132 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1133 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1136 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1140 zfs_domount(vfs_t *vfsp, char *osname)
1142 uint64_t recordsize, fsid_guid;
1150 error = zfsvfs_create(osname, &zfsvfs);
1153 zfsvfs->z_vfs = vfsp;
1156 /* Initialize the generic filesystem structure. */
1157 vfsp->vfs_bcount = 0;
1158 vfsp->vfs_data = NULL;
1160 if (zfs_create_unique_device(&mount_dev) == -1) {
1161 error = SET_ERROR(ENODEV);
1164 ASSERT(vfs_devismounted(mount_dev) == 0);
1167 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1170 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1171 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1173 vfsp->vfs_data = zfsvfs;
1174 vfsp->mnt_flag |= MNT_LOCAL;
1175 vfsp->mnt_kern_flag |= MNTK_MPSAFE;
1176 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1177 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1178 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1181 * The fsid is 64 bits, composed of an 8-bit fs type, which
1182 * separates our fsid from any other filesystem types, and a
1183 * 56-bit objset unique ID. The objset unique ID is unique to
1184 * all objsets open on this system, provided by unique_create().
1185 * The 8-bit fs type must be put in the low bits of fsid[1]
1186 * because that's where other Solaris filesystems put it.
1188 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1189 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1190 vfsp->vfs_fsid.val[0] = fsid_guid;
1191 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1192 vfsp->mnt_vfc->vfc_typenum & 0xFF;
1195 * Set features for file system.
1197 zfs_set_fuid_feature(zfsvfs);
1198 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1199 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1200 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1201 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1202 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1203 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1204 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1206 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1208 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1211 atime_changed_cb(zfsvfs, B_FALSE);
1212 readonly_changed_cb(zfsvfs, B_TRUE);
1213 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1215 xattr_changed_cb(zfsvfs, pval);
1216 zfsvfs->z_issnap = B_TRUE;
1217 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1219 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1220 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1221 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1223 error = zfsvfs_setup(zfsvfs, B_TRUE);
1226 vfs_mountedfrom(vfsp, osname);
1227 /* Grab extra reference. */
1228 VERIFY(VFS_ROOT(vfsp, LK_EXCLUSIVE, &vp) == 0);
1231 if (!zfsvfs->z_issnap)
1232 zfsctl_create(zfsvfs);
1235 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1236 zfsvfs_free(zfsvfs);
1238 atomic_add_32(&zfs_active_fs_count, 1);
1245 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1247 objset_t *os = zfsvfs->z_os;
1248 struct dsl_dataset *ds;
1251 * Unregister properties.
1253 if (!dmu_objset_is_snapshot(os)) {
1254 ds = dmu_objset_ds(os);
1255 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1258 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1261 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1264 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1267 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1270 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1273 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1276 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
1279 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1280 acl_inherit_changed_cb, zfsvfs) == 0);
1282 VERIFY(dsl_prop_unregister(ds, "vscan",
1283 vscan_changed_cb, zfsvfs) == 0);
1289 * Convert a decimal digit string to a uint64_t integer.
1292 str_to_uint64(char *str, uint64_t *objnum)
1297 if (*str < '0' || *str > '9')
1298 return (SET_ERROR(EINVAL));
1300 num = num*10 + *str++ - '0';
1308 * The boot path passed from the boot loader is in the form of
1309 * "rootpool-name/root-filesystem-object-number'. Convert this
1310 * string to a dataset name: "rootpool-name/root-filesystem-name".
1313 zfs_parse_bootfs(char *bpath, char *outpath)
1319 if (*bpath == 0 || *bpath == '/')
1320 return (SET_ERROR(EINVAL));
1322 (void) strcpy(outpath, bpath);
1324 slashp = strchr(bpath, '/');
1326 /* if no '/', just return the pool name */
1327 if (slashp == NULL) {
1331 /* if not a number, just return the root dataset name */
1332 if (str_to_uint64(slashp+1, &objnum)) {
1337 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1344 * Check that the hex label string is appropriate for the dataset being
1345 * mounted into the global_zone proper.
1347 * Return an error if the hex label string is not default or
1348 * admin_low/admin_high. For admin_low labels, the corresponding
1349 * dataset must be readonly.
1352 zfs_check_global_label(const char *dsname, const char *hexsl)
1354 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1356 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1358 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1359 /* must be readonly */
1362 if (dsl_prop_get_integer(dsname,
1363 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1364 return (SET_ERROR(EACCES));
1365 return (rdonly ? 0 : EACCES);
1367 return (SET_ERROR(EACCES));
1371 * Determine whether the mount is allowed according to MAC check.
1372 * by comparing (where appropriate) label of the dataset against
1373 * the label of the zone being mounted into. If the dataset has
1374 * no label, create one.
1376 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1379 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1382 zone_t *mntzone = NULL;
1383 ts_label_t *mnt_tsl;
1386 char ds_hexsl[MAXNAMELEN];
1388 retv = EACCES; /* assume the worst */
1391 * Start by getting the dataset label if it exists.
1393 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1394 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1396 return (SET_ERROR(EACCES));
1399 * If labeling is NOT enabled, then disallow the mount of datasets
1400 * which have a non-default label already. No other label checks
1403 if (!is_system_labeled()) {
1404 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1406 return (SET_ERROR(EACCES));
1410 * Get the label of the mountpoint. If mounting into the global
1411 * zone (i.e. mountpoint is not within an active zone and the
1412 * zoned property is off), the label must be default or
1413 * admin_low/admin_high only; no other checks are needed.
1415 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1416 if (mntzone->zone_id == GLOBAL_ZONEID) {
1421 if (dsl_prop_get_integer(osname,
1422 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1423 return (SET_ERROR(EACCES));
1425 return (zfs_check_global_label(osname, ds_hexsl));
1428 * This is the case of a zone dataset being mounted
1429 * initially, before the zone has been fully created;
1430 * allow this mount into global zone.
1435 mnt_tsl = mntzone->zone_slabel;
1436 ASSERT(mnt_tsl != NULL);
1437 label_hold(mnt_tsl);
1438 mnt_sl = label2bslabel(mnt_tsl);
1440 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1442 * The dataset doesn't have a real label, so fabricate one.
1446 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1447 dsl_prop_set_string(osname,
1448 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1449 ZPROP_SRC_LOCAL, str) == 0)
1452 kmem_free(str, strlen(str) + 1);
1453 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1455 * Now compare labels to complete the MAC check. If the
1456 * labels are equal then allow access. If the mountpoint
1457 * label dominates the dataset label, allow readonly access.
1458 * Otherwise, access is denied.
1460 if (blequal(mnt_sl, &ds_sl))
1462 else if (bldominates(mnt_sl, &ds_sl)) {
1463 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1468 label_rele(mnt_tsl);
1472 #endif /* SECLABEL */
1474 #ifdef OPENSOLARIS_MOUNTROOT
1476 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1479 static int zfsrootdone = 0;
1480 zfsvfs_t *zfsvfs = NULL;
1489 * The filesystem that we mount as root is defined in the
1490 * boot property "zfs-bootfs" with a format of
1491 * "poolname/root-dataset-objnum".
1493 if (why == ROOT_INIT) {
1495 return (SET_ERROR(EBUSY));
1497 * the process of doing a spa_load will require the
1498 * clock to be set before we could (for example) do
1499 * something better by looking at the timestamp on
1500 * an uberblock, so just set it to -1.
1504 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1505 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1507 return (SET_ERROR(EINVAL));
1509 zfs_devid = spa_get_bootprop("diskdevid");
1510 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1512 spa_free_bootprop(zfs_devid);
1514 spa_free_bootprop(zfs_bootfs);
1515 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1519 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1520 spa_free_bootprop(zfs_bootfs);
1521 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1526 spa_free_bootprop(zfs_bootfs);
1528 if (error = vfs_lock(vfsp))
1531 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1532 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1536 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1538 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1539 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1544 mutex_enter(&vp->v_lock);
1545 vp->v_flag |= VROOT;
1546 mutex_exit(&vp->v_lock);
1550 * Leave rootvp held. The root file system is never unmounted.
1553 vfs_add((struct vnode *)0, vfsp,
1554 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1558 } else if (why == ROOT_REMOUNT) {
1559 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1560 vfsp->vfs_flag |= VFS_REMOUNT;
1562 /* refresh mount options */
1563 zfs_unregister_callbacks(vfsp->vfs_data);
1564 return (zfs_register_callbacks(vfsp));
1566 } else if (why == ROOT_UNMOUNT) {
1567 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1568 (void) zfs_sync(vfsp, 0, 0);
1573 * if "why" is equal to anything else other than ROOT_INIT,
1574 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1576 return (SET_ERROR(ENOTSUP));
1578 #endif /* OPENSOLARIS_MOUNTROOT */
1581 getpoolname(const char *osname, char *poolname)
1585 p = strchr(osname, '/');
1587 if (strlen(osname) >= MAXNAMELEN)
1588 return (ENAMETOOLONG);
1589 (void) strcpy(poolname, osname);
1591 if (p - osname >= MAXNAMELEN)
1592 return (ENAMETOOLONG);
1593 (void) strncpy(poolname, osname, p - osname);
1594 poolname[p - osname] = '\0';
1601 zfs_mount(vfs_t *vfsp)
1603 kthread_t *td = curthread;
1604 vnode_t *mvp = vfsp->mnt_vnodecovered;
1605 cred_t *cr = td->td_ucred;
1611 if (mvp->v_type != VDIR)
1612 return (SET_ERROR(ENOTDIR));
1614 mutex_enter(&mvp->v_lock);
1615 if ((uap->flags & MS_REMOUNT) == 0 &&
1616 (uap->flags & MS_OVERLAY) == 0 &&
1617 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1618 mutex_exit(&mvp->v_lock);
1619 return (SET_ERROR(EBUSY));
1621 mutex_exit(&mvp->v_lock);
1624 * ZFS does not support passing unparsed data in via MS_DATA.
1625 * Users should use the MS_OPTIONSTR interface; this means
1626 * that all option parsing is already done and the options struct
1627 * can be interrogated.
1629 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1631 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS))
1632 return (SET_ERROR(EPERM));
1634 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1635 return (SET_ERROR(EINVAL));
1636 #endif /* ! illumos */
1639 * If full-owner-access is enabled and delegated administration is
1640 * turned on, we must set nosuid.
1642 if (zfs_super_owner &&
1643 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1644 secpolicy_fs_mount_clearopts(cr, vfsp);
1648 * Check for mount privilege?
1650 * If we don't have privilege then see if
1651 * we have local permission to allow it
1653 error = secpolicy_fs_mount(cr, mvp, vfsp);
1655 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1658 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1662 * Make sure user is the owner of the mount point
1663 * or has sufficient privileges.
1666 vattr.va_mask = AT_UID;
1668 vn_lock(mvp, LK_SHARED | LK_RETRY);
1669 if (VOP_GETATTR(mvp, &vattr, cr)) {
1674 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1675 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1682 secpolicy_fs_mount_clearopts(cr, vfsp);
1686 * Refuse to mount a filesystem if we are in a local zone and the
1687 * dataset is not visible.
1689 if (!INGLOBALZONE(curthread) &&
1690 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1691 error = SET_ERROR(EPERM);
1696 error = zfs_mount_label_policy(vfsp, osname);
1701 vfsp->vfs_flag |= MNT_NFS4ACLS;
1704 * When doing a remount, we simply refresh our temporary properties
1705 * according to those options set in the current VFS options.
1707 if (vfsp->vfs_flag & MS_REMOUNT) {
1708 /* refresh mount options */
1709 zfs_unregister_callbacks(vfsp->vfs_data);
1710 error = zfs_register_callbacks(vfsp);
1714 /* Initial root mount: try hard to import the requested root pool. */
1715 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1716 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1717 char pname[MAXNAMELEN];
1719 error = getpoolname(osname, pname);
1721 error = spa_import_rootpool(pname);
1726 error = zfs_domount(vfsp, osname);
1731 * Add an extra VFS_HOLD on our parent vfs so that it can't
1732 * disappear due to a forced unmount.
1734 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1735 VFS_HOLD(mvp->v_vfsp);
1743 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1745 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1746 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1748 statp->f_version = STATFS_VERSION;
1752 dmu_objset_space(zfsvfs->z_os,
1753 &refdbytes, &availbytes, &usedobjs, &availobjs);
1756 * The underlying storage pool actually uses multiple block sizes.
1757 * We report the fragsize as the smallest block size we support,
1758 * and we report our blocksize as the filesystem's maximum blocksize.
1760 statp->f_bsize = SPA_MINBLOCKSIZE;
1761 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1764 * The following report "total" blocks of various kinds in the
1765 * file system, but reported in terms of f_frsize - the
1769 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1770 statp->f_bfree = availbytes / statp->f_bsize;
1771 statp->f_bavail = statp->f_bfree; /* no root reservation */
1774 * statvfs() should really be called statufs(), because it assumes
1775 * static metadata. ZFS doesn't preallocate files, so the best
1776 * we can do is report the max that could possibly fit in f_files,
1777 * and that minus the number actually used in f_ffree.
1778 * For f_ffree, report the smaller of the number of object available
1779 * and the number of blocks (each object will take at least a block).
1781 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1782 statp->f_files = statp->f_ffree + usedobjs;
1785 * We're a zfs filesystem.
1787 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
1789 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
1790 sizeof(statp->f_mntfromname));
1791 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
1792 sizeof(statp->f_mntonname));
1794 statp->f_namemax = ZFS_MAXNAMELEN;
1801 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
1803 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1807 ZFS_ENTER_NOERROR(zfsvfs);
1809 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1811 *vpp = ZTOV(rootzp);
1816 error = vn_lock(*vpp, flags);
1818 (*vpp)->v_vflag |= VV_ROOT;
1827 * Teardown the zfsvfs::z_os.
1829 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1830 * and 'z_teardown_inactive_lock' held.
1833 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1837 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1841 * We purge the parent filesystem's vfsp as the parent
1842 * filesystem and all of its snapshots have their vnode's
1843 * v_vfsp set to the parent's filesystem's vfsp. Note,
1844 * 'z_parent' is self referential for non-snapshots.
1846 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1847 #ifdef FREEBSD_NAMECACHE
1848 cache_purgevfs(zfsvfs->z_parent->z_vfs);
1853 * Close the zil. NB: Can't close the zil while zfs_inactive
1854 * threads are blocked as zil_close can call zfs_inactive.
1856 if (zfsvfs->z_log) {
1857 zil_close(zfsvfs->z_log);
1858 zfsvfs->z_log = NULL;
1861 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1864 * If we are not unmounting (ie: online recv) and someone already
1865 * unmounted this file system while we were doing the switcheroo,
1866 * or a reopen of z_os failed then just bail out now.
1868 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1869 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1870 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1871 return (SET_ERROR(EIO));
1875 * At this point there are no vops active, and any new vops will
1876 * fail with EIO since we have z_teardown_lock for writer (only
1877 * relavent for forced unmount).
1879 * Release all holds on dbufs.
1881 mutex_enter(&zfsvfs->z_znodes_lock);
1882 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1883 zp = list_next(&zfsvfs->z_all_znodes, zp))
1885 ASSERT(ZTOV(zp)->v_count >= 0);
1886 zfs_znode_dmu_fini(zp);
1888 mutex_exit(&zfsvfs->z_znodes_lock);
1891 * If we are unmounting, set the unmounted flag and let new vops
1892 * unblock. zfs_inactive will have the unmounted behavior, and all
1893 * other vops will fail with EIO.
1896 zfsvfs->z_unmounted = B_TRUE;
1897 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1898 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1902 * z_os will be NULL if there was an error in attempting to reopen
1903 * zfsvfs, so just return as the properties had already been
1904 * unregistered and cached data had been evicted before.
1906 if (zfsvfs->z_os == NULL)
1910 * Unregister properties.
1912 zfs_unregister_callbacks(zfsvfs);
1917 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1918 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1919 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1920 dmu_objset_evict_dbufs(zfsvfs->z_os);
1927 zfs_umount(vfs_t *vfsp, int fflag)
1929 kthread_t *td = curthread;
1930 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1932 cred_t *cr = td->td_ucred;
1935 ret = secpolicy_fs_unmount(cr, vfsp);
1937 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1938 ZFS_DELEG_PERM_MOUNT, cr))
1943 * We purge the parent filesystem's vfsp as the parent filesystem
1944 * and all of its snapshots have their vnode's v_vfsp set to the
1945 * parent's filesystem's vfsp. Note, 'z_parent' is self
1946 * referential for non-snapshots.
1948 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1951 * Unmount any snapshots mounted under .zfs before unmounting the
1954 if (zfsvfs->z_ctldir != NULL) {
1955 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
1957 ret = vflush(vfsp, 0, 0, td);
1958 ASSERT(ret == EBUSY);
1959 if (!(fflag & MS_FORCE)) {
1960 if (zfsvfs->z_ctldir->v_count > 1)
1962 ASSERT(zfsvfs->z_ctldir->v_count == 1);
1964 zfsctl_destroy(zfsvfs);
1965 ASSERT(zfsvfs->z_ctldir == NULL);
1968 if (fflag & MS_FORCE) {
1970 * Mark file system as unmounted before calling
1971 * vflush(FORCECLOSE). This way we ensure no future vnops
1972 * will be called and risk operating on DOOMED vnodes.
1974 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1975 zfsvfs->z_unmounted = B_TRUE;
1976 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1980 * Flush all the files.
1982 ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
1984 if (!zfsvfs->z_issnap) {
1985 zfsctl_create(zfsvfs);
1986 ASSERT(zfsvfs->z_ctldir != NULL);
1992 if (!(fflag & MS_FORCE)) {
1994 * Check the number of active vnodes in the file system.
1995 * Our count is maintained in the vfs structure, but the
1996 * number is off by 1 to indicate a hold on the vfs
1999 * The '.zfs' directory maintains a reference of its
2000 * own, and any active references underneath are
2001 * reflected in the vnode count.
2003 if (zfsvfs->z_ctldir == NULL) {
2004 if (vfsp->vfs_count > 1)
2005 return (SET_ERROR(EBUSY));
2007 if (vfsp->vfs_count > 2 ||
2008 zfsvfs->z_ctldir->v_count > 1)
2009 return (SET_ERROR(EBUSY));
2014 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2018 * z_os will be NULL if there was an error in
2019 * attempting to reopen zfsvfs.
2023 * Unset the objset user_ptr.
2025 mutex_enter(&os->os_user_ptr_lock);
2026 dmu_objset_set_user(os, NULL);
2027 mutex_exit(&os->os_user_ptr_lock);
2030 * Finally release the objset
2032 dmu_objset_disown(os, zfsvfs);
2036 * We can now safely destroy the '.zfs' directory node.
2038 if (zfsvfs->z_ctldir != NULL)
2039 zfsctl_destroy(zfsvfs);
2040 if (zfsvfs->z_issnap) {
2041 vnode_t *svp = vfsp->mnt_vnodecovered;
2043 if (svp->v_count >= 2)
2052 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2054 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2059 * zfs_zget() can't operate on virtual entries like .zfs/ or
2060 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2061 * This will make NFS to switch to LOOKUP instead of using VGET.
2063 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2064 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2065 return (EOPNOTSUPP);
2068 err = zfs_zget(zfsvfs, ino, &zp);
2069 if (err == 0 && zp->z_unlinked) {
2077 err = vn_lock(*vpp, flags);
2081 (*vpp)->v_hash = ino;
2086 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
2087 struct ucred **credanonp, int *numsecflavors, int **secflavors)
2089 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2092 * If this is regular file system vfsp is the same as
2093 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2094 * zfsvfs->z_parent->z_vfs represents parent file system
2095 * which we have to use here, because only this file system
2096 * has mnt_export configured.
2098 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2099 credanonp, numsecflavors, secflavors));
2102 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2103 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2106 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2108 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2110 uint64_t object = 0;
2111 uint64_t fid_gen = 0;
2121 * On FreeBSD we can get snapshot's mount point or its parent file
2122 * system mount point depending if snapshot is already mounted or not.
2124 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2125 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2126 uint64_t objsetid = 0;
2127 uint64_t setgen = 0;
2129 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2130 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2132 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2133 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2137 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2139 return (SET_ERROR(EINVAL));
2143 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2144 zfid_short_t *zfid = (zfid_short_t *)fidp;
2146 for (i = 0; i < sizeof (zfid->zf_object); i++)
2147 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2149 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2150 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2153 return (SET_ERROR(EINVAL));
2157 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2158 * directory tree. If the object == zfsvfs->z_shares_dir, then
2159 * we are in the .zfs/shares directory tree.
2161 if ((fid_gen == 0 &&
2162 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2163 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2164 *vpp = zfsvfs->z_ctldir;
2165 ASSERT(*vpp != NULL);
2166 if (object == ZFSCTL_INO_SNAPDIR) {
2167 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2168 0, NULL, NULL, NULL, NULL, NULL) == 0);
2169 } else if (object == zfsvfs->z_shares_dir) {
2170 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL,
2171 0, NULL, NULL, NULL, NULL, NULL) == 0);
2176 err = vn_lock(*vpp, flags);
2182 gen_mask = -1ULL >> (64 - 8 * i);
2184 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2185 if (err = zfs_zget(zfsvfs, object, &zp)) {
2189 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2191 zp_gen = zp_gen & gen_mask;
2194 if (zp->z_unlinked || zp_gen != fid_gen) {
2195 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2198 return (SET_ERROR(EINVAL));
2203 err = vn_lock(*vpp, flags | LK_RETRY);
2205 vnode_create_vobject(*vpp, zp->z_size, curthread);
2212 * Block out VOPs and close zfsvfs_t::z_os
2214 * Note, if successful, then we return with the 'z_teardown_lock' and
2215 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2216 * dataset and objset intact so that they can be atomically handed off during
2217 * a subsequent rollback or recv operation and the resume thereafter.
2220 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2224 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2231 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2232 * is an invariant across any of the operations that can be performed while the
2233 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2234 * are the same: the relevant objset and associated dataset are owned by
2235 * zfsvfs, held, and long held on entry.
2238 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2242 uint64_t sa_obj = 0;
2244 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
2245 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2248 * We already own this, so just hold and rele it to update the
2249 * objset_t, as the one we had before may have been evicted.
2251 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os));
2252 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs);
2253 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset));
2254 dmu_objset_rele(zfsvfs->z_os, zfsvfs);
2257 * Make sure version hasn't changed
2260 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
2261 &zfsvfs->z_version);
2266 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2267 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2269 if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
2272 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2273 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
2276 if (zfsvfs->z_version >= ZPL_VERSION_SA)
2277 sa_register_update_callback(zfsvfs->z_os,
2280 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2282 zfs_set_fuid_feature(zfsvfs);
2285 * Attempt to re-establish all the active znodes with
2286 * their dbufs. If a zfs_rezget() fails, then we'll let
2287 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2288 * when they try to use their znode.
2290 mutex_enter(&zfsvfs->z_znodes_lock);
2291 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2292 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2293 (void) zfs_rezget(zp);
2295 mutex_exit(&zfsvfs->z_znodes_lock);
2298 /* release the VOPs */
2299 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2300 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
2304 * Since we couldn't setup the sa framework, try to force
2305 * unmount this file system.
2307 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2308 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2314 zfs_freevfs(vfs_t *vfsp)
2316 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2320 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2321 * from zfs_mount(). Release it here. If we came through
2322 * zfs_mountroot() instead, we didn't grab an extra hold, so
2323 * skip the VFS_RELE for rootvfs.
2325 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2326 VFS_RELE(zfsvfs->z_parent->z_vfs);
2329 zfsvfs_free(zfsvfs);
2331 atomic_add_32(&zfs_active_fs_count, -1);
2335 static int desiredvnodes_backup;
2339 zfs_vnodes_adjust(void)
2342 int newdesiredvnodes;
2344 desiredvnodes_backup = desiredvnodes;
2347 * We calculate newdesiredvnodes the same way it is done in
2348 * vntblinit(). If it is equal to desiredvnodes, it means that
2349 * it wasn't tuned by the administrator and we can tune it down.
2351 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 *
2352 vm_kmem_size / (5 * (sizeof(struct vm_object) +
2353 sizeof(struct vnode))));
2354 if (newdesiredvnodes == desiredvnodes)
2355 desiredvnodes = (3 * newdesiredvnodes) / 4;
2360 zfs_vnodes_adjust_back(void)
2364 desiredvnodes = desiredvnodes_backup;
2372 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2375 * Initialize .zfs directory structures
2380 * Initialize znode cache, vnode ops, etc...
2385 * Reduce number of vnodes. Originally number of vnodes is calculated
2386 * with UFS inode in mind. We reduce it here, because it's too big for
2389 zfs_vnodes_adjust();
2391 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2399 zfs_vnodes_adjust_back();
2405 return (zfs_active_fs_count != 0);
2409 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2412 objset_t *os = zfsvfs->z_os;
2415 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2416 return (SET_ERROR(EINVAL));
2418 if (newvers < zfsvfs->z_version)
2419 return (SET_ERROR(EINVAL));
2421 if (zfs_spa_version_map(newvers) >
2422 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2423 return (SET_ERROR(ENOTSUP));
2425 tx = dmu_tx_create(os);
2426 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2427 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2428 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2430 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2432 error = dmu_tx_assign(tx, TXG_WAIT);
2438 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2439 8, 1, &newvers, tx);
2446 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2449 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2451 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2452 DMU_OT_NONE, 0, tx);
2454 error = zap_add(os, MASTER_NODE_OBJ,
2455 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2458 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2459 sa_register_update_callback(os, zfs_sa_upgrade);
2462 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2463 "from %llu to %llu", zfsvfs->z_version, newvers);
2467 zfsvfs->z_version = newvers;
2469 zfs_set_fuid_feature(zfsvfs);
2475 * Read a property stored within the master node.
2478 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2484 * Look up the file system's value for the property. For the
2485 * version property, we look up a slightly different string.
2487 if (prop == ZFS_PROP_VERSION)
2488 pname = ZPL_VERSION_STR;
2490 pname = zfs_prop_to_name(prop);
2493 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2495 if (error == ENOENT) {
2496 /* No value set, use the default value */
2498 case ZFS_PROP_VERSION:
2499 *value = ZPL_VERSION;
2501 case ZFS_PROP_NORMALIZE:
2502 case ZFS_PROP_UTF8ONLY:
2506 *value = ZFS_CASE_SENSITIVE;
2518 zfsvfs_update_fromname(const char *oldname, const char *newname)
2520 char tmpbuf[MAXPATHLEN];
2525 oldlen = strlen(oldname);
2527 mtx_lock(&mountlist_mtx);
2528 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2529 fromname = mp->mnt_stat.f_mntfromname;
2530 if (strcmp(fromname, oldname) == 0) {
2531 (void)strlcpy(fromname, newname,
2532 sizeof(mp->mnt_stat.f_mntfromname));
2535 if (strncmp(fromname, oldname, oldlen) == 0 &&
2536 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2537 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
2538 newname, fromname + oldlen);
2539 (void)strlcpy(fromname, tmpbuf,
2540 sizeof(mp->mnt_stat.f_mntfromname));
2544 mtx_unlock(&mountlist_mtx);