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 SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RWTUN, &zfs_debug_level, 0,
80 SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions");
81 static int zfs_version_acl = ZFS_ACL_VERSION;
82 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0,
84 static int zfs_version_spa = SPA_VERSION;
85 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0,
87 static int zfs_version_zpl = ZPL_VERSION;
88 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0,
91 static int zfs_mount(vfs_t *vfsp);
92 static int zfs_umount(vfs_t *vfsp, int fflag);
93 static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
94 static int zfs_statfs(vfs_t *vfsp, struct statfs *statp);
95 static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp);
96 static int zfs_sync(vfs_t *vfsp, int waitfor);
97 static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
98 struct ucred **credanonp, int *numsecflavors, int **secflavors);
99 static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp);
100 static void zfs_objset_close(zfsvfs_t *zfsvfs);
101 static void zfs_freevfs(vfs_t *vfsp);
103 static struct vfsops zfs_vfsops = {
104 .vfs_mount = zfs_mount,
105 .vfs_unmount = zfs_umount,
106 .vfs_root = zfs_root,
107 .vfs_statfs = zfs_statfs,
108 .vfs_vget = zfs_vget,
109 .vfs_sync = zfs_sync,
110 .vfs_checkexp = zfs_checkexp,
111 .vfs_fhtovp = zfs_fhtovp,
114 VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN);
117 * We need to keep a count of active fs's.
118 * This is necessary to prevent our module
119 * from being unloaded after a umount -f
121 static uint32_t zfs_active_fs_count = 0;
125 zfs_sync(vfs_t *vfsp, int waitfor)
129 * Data integrity is job one. We don't want a compromised kernel
130 * writing to the storage pool, so we never sync during panic.
137 * Sync a specific filesystem.
139 zfsvfs_t *zfsvfs = vfsp->vfs_data;
143 error = vfs_stdsync(vfsp, waitfor);
148 dp = dmu_objset_pool(zfsvfs->z_os);
151 * If the system is shutting down, then skip any
152 * filesystems which may exist on a suspended pool.
154 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
159 if (zfsvfs->z_log != NULL)
160 zil_commit(zfsvfs->z_log, 0);
165 * Sync all ZFS filesystems. This is what happens when you
166 * run sync(1M). Unlike other filesystems, ZFS honors the
167 * request by waiting for all pools to commit all dirty data.
175 #ifndef __FreeBSD_kernel__
177 zfs_create_unique_device(dev_t *dev)
182 ASSERT3U(zfs_minor, <=, MAXMIN32);
183 minor_t start = zfs_minor;
185 mutex_enter(&zfs_dev_mtx);
186 if (zfs_minor >= MAXMIN32) {
188 * If we're still using the real major
189 * keep out of /dev/zfs and /dev/zvol minor
190 * number space. If we're using a getudev()'ed
191 * major number, we can use all of its minors.
193 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
194 zfs_minor = ZFS_MIN_MINOR;
200 *dev = makedevice(zfs_major, zfs_minor);
201 mutex_exit(&zfs_dev_mtx);
202 } while (vfs_devismounted(*dev) && zfs_minor != start);
203 if (zfs_minor == start) {
205 * We are using all ~262,000 minor numbers for the
206 * current major number. Create a new major number.
208 if ((new_major = getudev()) == (major_t)-1) {
210 "zfs_mount: Can't get unique major "
214 mutex_enter(&zfs_dev_mtx);
215 zfs_major = new_major;
218 mutex_exit(&zfs_dev_mtx);
222 /* CONSTANTCONDITION */
227 #endif /* !__FreeBSD_kernel__ */
230 atime_changed_cb(void *arg, uint64_t newval)
232 zfsvfs_t *zfsvfs = arg;
234 if (newval == TRUE) {
235 zfsvfs->z_atime = TRUE;
236 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
237 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
238 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
240 zfsvfs->z_atime = FALSE;
241 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
242 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
243 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
248 xattr_changed_cb(void *arg, uint64_t newval)
250 zfsvfs_t *zfsvfs = arg;
252 if (newval == TRUE) {
253 /* XXX locking on vfs_flag? */
255 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
257 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
258 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
260 /* XXX locking on vfs_flag? */
262 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
264 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
265 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
270 blksz_changed_cb(void *arg, uint64_t newval)
272 zfsvfs_t *zfsvfs = arg;
273 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
274 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
275 ASSERT(ISP2(newval));
277 zfsvfs->z_max_blksz = newval;
278 zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
282 readonly_changed_cb(void *arg, uint64_t newval)
284 zfsvfs_t *zfsvfs = arg;
287 /* XXX locking on vfs_flag? */
288 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
289 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
290 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
292 /* XXX locking on vfs_flag? */
293 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
294 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
295 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
300 setuid_changed_cb(void *arg, uint64_t newval)
302 zfsvfs_t *zfsvfs = arg;
304 if (newval == FALSE) {
305 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
306 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
307 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
309 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
310 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
311 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
316 exec_changed_cb(void *arg, uint64_t newval)
318 zfsvfs_t *zfsvfs = arg;
320 if (newval == FALSE) {
321 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
322 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
323 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
325 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
326 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
327 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
332 * The nbmand mount option can be changed at mount time.
333 * We can't allow it to be toggled on live file systems or incorrect
334 * behavior may be seen from cifs clients
336 * This property isn't registered via dsl_prop_register(), but this callback
337 * will be called when a file system is first mounted
340 nbmand_changed_cb(void *arg, uint64_t newval)
342 zfsvfs_t *zfsvfs = arg;
343 if (newval == FALSE) {
344 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
345 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
347 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
348 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
353 snapdir_changed_cb(void *arg, uint64_t newval)
355 zfsvfs_t *zfsvfs = arg;
357 zfsvfs->z_show_ctldir = newval;
361 vscan_changed_cb(void *arg, uint64_t newval)
363 zfsvfs_t *zfsvfs = arg;
365 zfsvfs->z_vscan = newval;
369 acl_mode_changed_cb(void *arg, uint64_t newval)
371 zfsvfs_t *zfsvfs = arg;
373 zfsvfs->z_acl_mode = newval;
377 acl_inherit_changed_cb(void *arg, uint64_t newval)
379 zfsvfs_t *zfsvfs = arg;
381 zfsvfs->z_acl_inherit = newval;
385 zfs_register_callbacks(vfs_t *vfsp)
387 struct dsl_dataset *ds = NULL;
389 zfsvfs_t *zfsvfs = NULL;
391 boolean_t readonly = B_FALSE;
392 boolean_t do_readonly = B_FALSE;
393 boolean_t setuid = B_FALSE;
394 boolean_t do_setuid = B_FALSE;
395 boolean_t exec = B_FALSE;
396 boolean_t do_exec = B_FALSE;
398 boolean_t devices = B_FALSE;
399 boolean_t do_devices = B_FALSE;
401 boolean_t xattr = B_FALSE;
402 boolean_t do_xattr = B_FALSE;
403 boolean_t atime = B_FALSE;
404 boolean_t do_atime = B_FALSE;
408 zfsvfs = vfsp->vfs_data;
413 * This function can be called for a snapshot when we update snapshot's
414 * mount point, which isn't really supported.
416 if (dmu_objset_is_snapshot(os))
420 * The act of registering our callbacks will destroy any mount
421 * options we may have. In order to enable temporary overrides
422 * of mount options, we stash away the current values and
423 * restore them after we register the callbacks.
425 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
426 !spa_writeable(dmu_objset_spa(os))) {
428 do_readonly = B_TRUE;
429 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
431 do_readonly = B_TRUE;
433 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
437 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
440 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
445 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
448 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
452 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
455 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
459 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
462 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
468 * nbmand is a special property. It can only be changed at
471 * This is weird, but it is documented to only be changeable
474 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
476 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
479 char osname[MAXNAMELEN];
481 dmu_objset_name(os, osname);
482 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
489 * Register property callbacks.
491 * It would probably be fine to just check for i/o error from
492 * the first prop_register(), but I guess I like to go
495 ds = dmu_objset_ds(os);
496 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
497 error = dsl_prop_register(ds,
498 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
499 error = error ? error : dsl_prop_register(ds,
500 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
501 error = error ? error : dsl_prop_register(ds,
502 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
503 error = error ? error : dsl_prop_register(ds,
504 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
506 error = error ? error : dsl_prop_register(ds,
507 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
509 error = error ? error : dsl_prop_register(ds,
510 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
511 error = error ? error : dsl_prop_register(ds,
512 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
513 error = error ? error : dsl_prop_register(ds,
514 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
515 error = error ? error : dsl_prop_register(ds,
516 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
517 error = error ? error : dsl_prop_register(ds,
518 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
520 error = error ? error : dsl_prop_register(ds,
521 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
522 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
527 * Invoke our callbacks to restore temporary mount options.
530 readonly_changed_cb(zfsvfs, readonly);
532 setuid_changed_cb(zfsvfs, setuid);
534 exec_changed_cb(zfsvfs, exec);
536 xattr_changed_cb(zfsvfs, xattr);
538 atime_changed_cb(zfsvfs, atime);
540 nbmand_changed_cb(zfsvfs, nbmand);
546 * We may attempt to unregister some callbacks that are not
547 * registered, but this is OK; it will simply return ENOMSG,
548 * which we will ignore.
550 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
551 atime_changed_cb, zfsvfs);
552 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
553 xattr_changed_cb, zfsvfs);
554 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
555 blksz_changed_cb, zfsvfs);
556 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
557 readonly_changed_cb, zfsvfs);
559 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
560 devices_changed_cb, zfsvfs);
562 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
563 setuid_changed_cb, zfsvfs);
564 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
565 exec_changed_cb, zfsvfs);
566 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
567 snapdir_changed_cb, zfsvfs);
568 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE),
569 acl_mode_changed_cb, zfsvfs);
570 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
571 acl_inherit_changed_cb, zfsvfs);
572 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
573 vscan_changed_cb, zfsvfs);
578 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
579 uint64_t *userp, uint64_t *groupp)
582 * Is it a valid type of object to track?
584 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
585 return (SET_ERROR(ENOENT));
588 * If we have a NULL data pointer
589 * then assume the id's aren't changing and
590 * return EEXIST to the dmu to let it know to
594 return (SET_ERROR(EEXIST));
596 if (bonustype == DMU_OT_ZNODE) {
597 znode_phys_t *znp = data;
598 *userp = znp->zp_uid;
599 *groupp = znp->zp_gid;
602 sa_hdr_phys_t *sap = data;
603 sa_hdr_phys_t sa = *sap;
604 boolean_t swap = B_FALSE;
606 ASSERT(bonustype == DMU_OT_SA);
608 if (sa.sa_magic == 0) {
610 * This should only happen for newly created
611 * files that haven't had the znode data filled
618 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
619 sa.sa_magic = SA_MAGIC;
620 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
623 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
626 hdrsize = sa_hdrsize(&sa);
627 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
628 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
630 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
633 *userp = BSWAP_64(*userp);
634 *groupp = BSWAP_64(*groupp);
641 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
642 char *domainbuf, int buflen, uid_t *ridp)
647 fuid = strtonum(fuidstr, NULL);
649 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
651 (void) strlcpy(domainbuf, domain, buflen);
654 *ridp = FUID_RID(fuid);
658 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
661 case ZFS_PROP_USERUSED:
662 return (DMU_USERUSED_OBJECT);
663 case ZFS_PROP_GROUPUSED:
664 return (DMU_GROUPUSED_OBJECT);
665 case ZFS_PROP_USERQUOTA:
666 return (zfsvfs->z_userquota_obj);
667 case ZFS_PROP_GROUPQUOTA:
668 return (zfsvfs->z_groupquota_obj);
674 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
675 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
680 zfs_useracct_t *buf = vbuf;
683 if (!dmu_objset_userspace_present(zfsvfs->z_os))
684 return (SET_ERROR(ENOTSUP));
686 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
692 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
693 (error = zap_cursor_retrieve(&zc, &za)) == 0;
694 zap_cursor_advance(&zc)) {
695 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
699 fuidstr_to_sid(zfsvfs, za.za_name,
700 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
702 buf->zu_space = za.za_first_integer;
708 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
709 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
710 *cookiep = zap_cursor_serialize(&zc);
711 zap_cursor_fini(&zc);
716 * buf must be big enough (eg, 32 bytes)
719 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
720 char *buf, boolean_t addok)
725 if (domain && domain[0]) {
726 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
728 return (SET_ERROR(ENOENT));
730 fuid = FUID_ENCODE(domainid, rid);
731 (void) sprintf(buf, "%llx", (longlong_t)fuid);
736 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
737 const char *domain, uint64_t rid, uint64_t *valp)
745 if (!dmu_objset_userspace_present(zfsvfs->z_os))
746 return (SET_ERROR(ENOTSUP));
748 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
752 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
756 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
763 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
764 const char *domain, uint64_t rid, uint64_t quota)
770 boolean_t fuid_dirtied;
772 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
773 return (SET_ERROR(EINVAL));
775 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
776 return (SET_ERROR(ENOTSUP));
778 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
779 &zfsvfs->z_groupquota_obj;
781 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
784 fuid_dirtied = zfsvfs->z_fuid_dirty;
786 tx = dmu_tx_create(zfsvfs->z_os);
787 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
789 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
790 zfs_userquota_prop_prefixes[type]);
793 zfs_fuid_txhold(zfsvfs, tx);
794 err = dmu_tx_assign(tx, TXG_WAIT);
800 mutex_enter(&zfsvfs->z_lock);
802 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
804 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
805 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
807 mutex_exit(&zfsvfs->z_lock);
810 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
814 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
818 zfs_fuid_sync(zfsvfs, tx);
824 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
827 uint64_t used, quota, usedobj, quotaobj;
830 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
831 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
833 if (quotaobj == 0 || zfsvfs->z_replay)
836 (void) sprintf(buf, "%llx", (longlong_t)fuid);
837 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
841 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
844 return (used >= quota);
848 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
853 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
855 fuid = isgroup ? zp->z_gid : zp->z_uid;
857 if (quotaobj == 0 || zfsvfs->z_replay)
860 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
864 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
873 * XXX: Fix struct statfs so this isn't necessary!
875 * The 'osname' is used as the filesystem's special node, which means
876 * it must fit in statfs.f_mntfromname, or else it can't be
877 * enumerated, so libzfs_mnttab_find() returns NULL, which causes
878 * 'zfs unmount' to think it's not mounted when it is.
880 if (strlen(osname) >= MNAMELEN)
881 return (SET_ERROR(ENAMETOOLONG));
883 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
886 * We claim to always be readonly so we can open snapshots;
887 * other ZPL code will prevent us from writing to snapshots.
889 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
891 kmem_free(zfsvfs, sizeof (zfsvfs_t));
896 * Initialize the zfs-specific filesystem structure.
897 * Should probably make this a kmem cache, shuffle fields,
898 * and just bzero up to z_hold_mtx[].
900 zfsvfs->z_vfs = NULL;
901 zfsvfs->z_parent = zfsvfs;
902 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
903 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
906 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
909 } else if (zfsvfs->z_version >
910 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
911 (void) printf("Can't mount a version %lld file system "
912 "on a version %lld pool\n. Pool must be upgraded to mount "
913 "this file system.", (u_longlong_t)zfsvfs->z_version,
914 (u_longlong_t)spa_version(dmu_objset_spa(os)));
915 error = SET_ERROR(ENOTSUP);
918 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
920 zfsvfs->z_norm = (int)zval;
922 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
924 zfsvfs->z_utf8 = (zval != 0);
926 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
928 zfsvfs->z_case = (uint_t)zval;
931 * Fold case on file systems that are always or sometimes case
934 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
935 zfsvfs->z_case == ZFS_CASE_MIXED)
936 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
938 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
939 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
941 if (zfsvfs->z_use_sa) {
942 /* should either have both of these objects or none */
943 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
949 * Pre SA versions file systems should never touch
950 * either the attribute registration or layout objects.
955 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
956 &zfsvfs->z_attr_table);
960 if (zfsvfs->z_version >= ZPL_VERSION_SA)
961 sa_register_update_callback(os, zfs_sa_upgrade);
963 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
967 ASSERT(zfsvfs->z_root != 0);
969 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
970 &zfsvfs->z_unlinkedobj);
974 error = zap_lookup(os, MASTER_NODE_OBJ,
975 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
976 8, 1, &zfsvfs->z_userquota_obj);
977 if (error && error != ENOENT)
980 error = zap_lookup(os, MASTER_NODE_OBJ,
981 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
982 8, 1, &zfsvfs->z_groupquota_obj);
983 if (error && error != ENOENT)
986 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
987 &zfsvfs->z_fuid_obj);
988 if (error && error != ENOENT)
991 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
992 &zfsvfs->z_shares_dir);
993 if (error && error != ENOENT)
996 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
997 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
998 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
999 offsetof(znode_t, z_link_node));
1000 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1001 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1002 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1003 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1004 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1010 dmu_objset_disown(os, zfsvfs);
1012 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1017 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1021 error = zfs_register_callbacks(zfsvfs->z_vfs);
1026 * Set the objset user_ptr to track its zfsvfs.
1028 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1029 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1030 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1032 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1035 * If we are not mounting (ie: online recv), then we don't
1036 * have to worry about replaying the log as we blocked all
1037 * operations out since we closed the ZIL.
1043 * During replay we remove the read only flag to
1044 * allow replays to succeed.
1046 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1048 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1050 zfs_unlinked_drain(zfsvfs);
1053 * Parse and replay the intent log.
1055 * Because of ziltest, this must be done after
1056 * zfs_unlinked_drain(). (Further note: ziltest
1057 * doesn't use readonly mounts, where
1058 * zfs_unlinked_drain() isn't called.) This is because
1059 * ziltest causes spa_sync() to think it's committed,
1060 * but actually it is not, so the intent log contains
1061 * many txg's worth of changes.
1063 * In particular, if object N is in the unlinked set in
1064 * the last txg to actually sync, then it could be
1065 * actually freed in a later txg and then reallocated
1066 * in a yet later txg. This would write a "create
1067 * object N" record to the intent log. Normally, this
1068 * would be fine because the spa_sync() would have
1069 * written out the fact that object N is free, before
1070 * we could write the "create object N" intent log
1073 * But when we are in ziltest mode, we advance the "open
1074 * txg" without actually spa_sync()-ing the changes to
1075 * disk. So we would see that object N is still
1076 * allocated and in the unlinked set, and there is an
1077 * intent log record saying to allocate it.
1079 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1080 if (zil_replay_disable) {
1081 zil_destroy(zfsvfs->z_log, B_FALSE);
1083 zfsvfs->z_replay = B_TRUE;
1084 zil_replay(zfsvfs->z_os, zfsvfs,
1086 zfsvfs->z_replay = B_FALSE;
1089 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1095 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1098 zfsvfs_free(zfsvfs_t *zfsvfs)
1103 * This is a barrier to prevent the filesystem from going away in
1104 * zfs_znode_move() until we can safely ensure that the filesystem is
1105 * not unmounted. We consider the filesystem valid before the barrier
1106 * and invalid after the barrier.
1108 rw_enter(&zfsvfs_lock, RW_READER);
1109 rw_exit(&zfsvfs_lock);
1111 zfs_fuid_destroy(zfsvfs);
1113 mutex_destroy(&zfsvfs->z_znodes_lock);
1114 mutex_destroy(&zfsvfs->z_lock);
1115 list_destroy(&zfsvfs->z_all_znodes);
1116 rrm_destroy(&zfsvfs->z_teardown_lock);
1117 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1118 rw_destroy(&zfsvfs->z_fuid_lock);
1119 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1120 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1121 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1125 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1127 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1128 if (zfsvfs->z_vfs) {
1129 if (zfsvfs->z_use_fuids) {
1130 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1131 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1132 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1133 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1134 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1135 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1137 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1138 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1139 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1140 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1141 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1142 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1145 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1149 zfs_domount(vfs_t *vfsp, char *osname)
1151 uint64_t recordsize, fsid_guid;
1159 error = zfsvfs_create(osname, &zfsvfs);
1162 zfsvfs->z_vfs = vfsp;
1165 /* Initialize the generic filesystem structure. */
1166 vfsp->vfs_bcount = 0;
1167 vfsp->vfs_data = NULL;
1169 if (zfs_create_unique_device(&mount_dev) == -1) {
1170 error = SET_ERROR(ENODEV);
1173 ASSERT(vfs_devismounted(mount_dev) == 0);
1176 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1179 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1180 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1182 vfsp->vfs_data = zfsvfs;
1183 vfsp->mnt_flag |= MNT_LOCAL;
1184 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1185 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1186 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1189 * The fsid is 64 bits, composed of an 8-bit fs type, which
1190 * separates our fsid from any other filesystem types, and a
1191 * 56-bit objset unique ID. The objset unique ID is unique to
1192 * all objsets open on this system, provided by unique_create().
1193 * The 8-bit fs type must be put in the low bits of fsid[1]
1194 * because that's where other Solaris filesystems put it.
1196 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1197 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1198 vfsp->vfs_fsid.val[0] = fsid_guid;
1199 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1200 vfsp->mnt_vfc->vfc_typenum & 0xFF;
1203 * Set features for file system.
1205 zfs_set_fuid_feature(zfsvfs);
1206 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1207 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1208 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1209 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1210 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1211 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1212 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1214 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1216 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1219 atime_changed_cb(zfsvfs, B_FALSE);
1220 readonly_changed_cb(zfsvfs, B_TRUE);
1221 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1223 xattr_changed_cb(zfsvfs, pval);
1224 zfsvfs->z_issnap = B_TRUE;
1225 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1227 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1228 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1229 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1231 error = zfsvfs_setup(zfsvfs, B_TRUE);
1234 vfs_mountedfrom(vfsp, osname);
1235 /* Grab extra reference. */
1236 VERIFY(VFS_ROOT(vfsp, LK_EXCLUSIVE, &vp) == 0);
1239 if (!zfsvfs->z_issnap)
1240 zfsctl_create(zfsvfs);
1243 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1244 zfsvfs_free(zfsvfs);
1246 atomic_inc_32(&zfs_active_fs_count);
1253 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1255 objset_t *os = zfsvfs->z_os;
1256 struct dsl_dataset *ds;
1259 * Unregister properties.
1261 if (!dmu_objset_is_snapshot(os)) {
1262 ds = dmu_objset_ds(os);
1263 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1266 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1269 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1272 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1275 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1278 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1281 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1284 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
1287 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1288 acl_inherit_changed_cb, zfsvfs) == 0);
1290 VERIFY(dsl_prop_unregister(ds, "vscan",
1291 vscan_changed_cb, zfsvfs) == 0);
1297 * Convert a decimal digit string to a uint64_t integer.
1300 str_to_uint64(char *str, uint64_t *objnum)
1305 if (*str < '0' || *str > '9')
1306 return (SET_ERROR(EINVAL));
1308 num = num*10 + *str++ - '0';
1316 * The boot path passed from the boot loader is in the form of
1317 * "rootpool-name/root-filesystem-object-number'. Convert this
1318 * string to a dataset name: "rootpool-name/root-filesystem-name".
1321 zfs_parse_bootfs(char *bpath, char *outpath)
1327 if (*bpath == 0 || *bpath == '/')
1328 return (SET_ERROR(EINVAL));
1330 (void) strcpy(outpath, bpath);
1332 slashp = strchr(bpath, '/');
1334 /* if no '/', just return the pool name */
1335 if (slashp == NULL) {
1339 /* if not a number, just return the root dataset name */
1340 if (str_to_uint64(slashp+1, &objnum)) {
1345 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1352 * Check that the hex label string is appropriate for the dataset being
1353 * mounted into the global_zone proper.
1355 * Return an error if the hex label string is not default or
1356 * admin_low/admin_high. For admin_low labels, the corresponding
1357 * dataset must be readonly.
1360 zfs_check_global_label(const char *dsname, const char *hexsl)
1362 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1364 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1366 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1367 /* must be readonly */
1370 if (dsl_prop_get_integer(dsname,
1371 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1372 return (SET_ERROR(EACCES));
1373 return (rdonly ? 0 : EACCES);
1375 return (SET_ERROR(EACCES));
1379 * Determine whether the mount is allowed according to MAC check.
1380 * by comparing (where appropriate) label of the dataset against
1381 * the label of the zone being mounted into. If the dataset has
1382 * no label, create one.
1384 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1387 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1390 zone_t *mntzone = NULL;
1391 ts_label_t *mnt_tsl;
1394 char ds_hexsl[MAXNAMELEN];
1396 retv = EACCES; /* assume the worst */
1399 * Start by getting the dataset label if it exists.
1401 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1402 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1404 return (SET_ERROR(EACCES));
1407 * If labeling is NOT enabled, then disallow the mount of datasets
1408 * which have a non-default label already. No other label checks
1411 if (!is_system_labeled()) {
1412 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1414 return (SET_ERROR(EACCES));
1418 * Get the label of the mountpoint. If mounting into the global
1419 * zone (i.e. mountpoint is not within an active zone and the
1420 * zoned property is off), the label must be default or
1421 * admin_low/admin_high only; no other checks are needed.
1423 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1424 if (mntzone->zone_id == GLOBAL_ZONEID) {
1429 if (dsl_prop_get_integer(osname,
1430 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1431 return (SET_ERROR(EACCES));
1433 return (zfs_check_global_label(osname, ds_hexsl));
1436 * This is the case of a zone dataset being mounted
1437 * initially, before the zone has been fully created;
1438 * allow this mount into global zone.
1443 mnt_tsl = mntzone->zone_slabel;
1444 ASSERT(mnt_tsl != NULL);
1445 label_hold(mnt_tsl);
1446 mnt_sl = label2bslabel(mnt_tsl);
1448 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1450 * The dataset doesn't have a real label, so fabricate one.
1454 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1455 dsl_prop_set_string(osname,
1456 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1457 ZPROP_SRC_LOCAL, str) == 0)
1460 kmem_free(str, strlen(str) + 1);
1461 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1463 * Now compare labels to complete the MAC check. If the
1464 * labels are equal then allow access. If the mountpoint
1465 * label dominates the dataset label, allow readonly access.
1466 * Otherwise, access is denied.
1468 if (blequal(mnt_sl, &ds_sl))
1470 else if (bldominates(mnt_sl, &ds_sl)) {
1471 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1476 label_rele(mnt_tsl);
1480 #endif /* SECLABEL */
1482 #ifdef OPENSOLARIS_MOUNTROOT
1484 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1487 static int zfsrootdone = 0;
1488 zfsvfs_t *zfsvfs = NULL;
1497 * The filesystem that we mount as root is defined in the
1498 * boot property "zfs-bootfs" with a format of
1499 * "poolname/root-dataset-objnum".
1501 if (why == ROOT_INIT) {
1503 return (SET_ERROR(EBUSY));
1505 * the process of doing a spa_load will require the
1506 * clock to be set before we could (for example) do
1507 * something better by looking at the timestamp on
1508 * an uberblock, so just set it to -1.
1512 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1513 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1515 return (SET_ERROR(EINVAL));
1517 zfs_devid = spa_get_bootprop("diskdevid");
1518 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1520 spa_free_bootprop(zfs_devid);
1522 spa_free_bootprop(zfs_bootfs);
1523 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1527 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1528 spa_free_bootprop(zfs_bootfs);
1529 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1534 spa_free_bootprop(zfs_bootfs);
1536 if (error = vfs_lock(vfsp))
1539 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1540 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1544 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1546 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1547 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1552 mutex_enter(&vp->v_lock);
1553 vp->v_flag |= VROOT;
1554 mutex_exit(&vp->v_lock);
1558 * Leave rootvp held. The root file system is never unmounted.
1561 vfs_add((struct vnode *)0, vfsp,
1562 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1566 } else if (why == ROOT_REMOUNT) {
1567 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1568 vfsp->vfs_flag |= VFS_REMOUNT;
1570 /* refresh mount options */
1571 zfs_unregister_callbacks(vfsp->vfs_data);
1572 return (zfs_register_callbacks(vfsp));
1574 } else if (why == ROOT_UNMOUNT) {
1575 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1576 (void) zfs_sync(vfsp, 0, 0);
1581 * if "why" is equal to anything else other than ROOT_INIT,
1582 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1584 return (SET_ERROR(ENOTSUP));
1586 #endif /* OPENSOLARIS_MOUNTROOT */
1589 getpoolname(const char *osname, char *poolname)
1593 p = strchr(osname, '/');
1595 if (strlen(osname) >= MAXNAMELEN)
1596 return (ENAMETOOLONG);
1597 (void) strcpy(poolname, osname);
1599 if (p - osname >= MAXNAMELEN)
1600 return (ENAMETOOLONG);
1601 (void) strncpy(poolname, osname, p - osname);
1602 poolname[p - osname] = '\0';
1609 zfs_mount(vfs_t *vfsp)
1611 kthread_t *td = curthread;
1612 vnode_t *mvp = vfsp->mnt_vnodecovered;
1613 cred_t *cr = td->td_ucred;
1619 if (mvp->v_type != VDIR)
1620 return (SET_ERROR(ENOTDIR));
1622 mutex_enter(&mvp->v_lock);
1623 if ((uap->flags & MS_REMOUNT) == 0 &&
1624 (uap->flags & MS_OVERLAY) == 0 &&
1625 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1626 mutex_exit(&mvp->v_lock);
1627 return (SET_ERROR(EBUSY));
1629 mutex_exit(&mvp->v_lock);
1632 * ZFS does not support passing unparsed data in via MS_DATA.
1633 * Users should use the MS_OPTIONSTR interface; this means
1634 * that all option parsing is already done and the options struct
1635 * can be interrogated.
1637 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1639 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS))
1640 return (SET_ERROR(EPERM));
1642 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1643 return (SET_ERROR(EINVAL));
1644 #endif /* ! illumos */
1647 * If full-owner-access is enabled and delegated administration is
1648 * turned on, we must set nosuid.
1650 if (zfs_super_owner &&
1651 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1652 secpolicy_fs_mount_clearopts(cr, vfsp);
1656 * Check for mount privilege?
1658 * If we don't have privilege then see if
1659 * we have local permission to allow it
1661 error = secpolicy_fs_mount(cr, mvp, vfsp);
1663 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1666 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1670 * Make sure user is the owner of the mount point
1671 * or has sufficient privileges.
1674 vattr.va_mask = AT_UID;
1676 vn_lock(mvp, LK_SHARED | LK_RETRY);
1677 if (VOP_GETATTR(mvp, &vattr, cr)) {
1682 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1683 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1690 secpolicy_fs_mount_clearopts(cr, vfsp);
1694 * Refuse to mount a filesystem if we are in a local zone and the
1695 * dataset is not visible.
1697 if (!INGLOBALZONE(curthread) &&
1698 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1699 error = SET_ERROR(EPERM);
1704 error = zfs_mount_label_policy(vfsp, osname);
1709 vfsp->vfs_flag |= MNT_NFS4ACLS;
1712 * When doing a remount, we simply refresh our temporary properties
1713 * according to those options set in the current VFS options.
1715 if (vfsp->vfs_flag & MS_REMOUNT) {
1716 /* refresh mount options */
1717 zfs_unregister_callbacks(vfsp->vfs_data);
1718 error = zfs_register_callbacks(vfsp);
1722 /* Initial root mount: try hard to import the requested root pool. */
1723 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1724 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1725 char pname[MAXNAMELEN];
1727 error = getpoolname(osname, pname);
1729 error = spa_import_rootpool(pname);
1734 error = zfs_domount(vfsp, osname);
1739 * Add an extra VFS_HOLD on our parent vfs so that it can't
1740 * disappear due to a forced unmount.
1742 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1743 VFS_HOLD(mvp->v_vfsp);
1751 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1753 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1754 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1756 statp->f_version = STATFS_VERSION;
1760 dmu_objset_space(zfsvfs->z_os,
1761 &refdbytes, &availbytes, &usedobjs, &availobjs);
1764 * The underlying storage pool actually uses multiple block sizes.
1765 * We report the fragsize as the smallest block size we support,
1766 * and we report our blocksize as the filesystem's maximum blocksize.
1768 statp->f_bsize = SPA_MINBLOCKSIZE;
1769 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1772 * The following report "total" blocks of various kinds in the
1773 * file system, but reported in terms of f_frsize - the
1777 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1778 statp->f_bfree = availbytes / statp->f_bsize;
1779 statp->f_bavail = statp->f_bfree; /* no root reservation */
1782 * statvfs() should really be called statufs(), because it assumes
1783 * static metadata. ZFS doesn't preallocate files, so the best
1784 * we can do is report the max that could possibly fit in f_files,
1785 * and that minus the number actually used in f_ffree.
1786 * For f_ffree, report the smaller of the number of object available
1787 * and the number of blocks (each object will take at least a block).
1789 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1790 statp->f_files = statp->f_ffree + usedobjs;
1793 * We're a zfs filesystem.
1795 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
1797 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
1798 sizeof(statp->f_mntfromname));
1799 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
1800 sizeof(statp->f_mntonname));
1802 statp->f_namemax = ZFS_MAXNAMELEN;
1809 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
1811 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1815 ZFS_ENTER_NOERROR(zfsvfs);
1817 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1819 *vpp = ZTOV(rootzp);
1824 error = vn_lock(*vpp, flags);
1826 (*vpp)->v_vflag |= VV_ROOT;
1835 * Teardown the zfsvfs::z_os.
1837 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1838 * and 'z_teardown_inactive_lock' held.
1841 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1845 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1849 * We purge the parent filesystem's vfsp as the parent
1850 * filesystem and all of its snapshots have their vnode's
1851 * v_vfsp set to the parent's filesystem's vfsp. Note,
1852 * 'z_parent' is self referential for non-snapshots.
1854 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1855 #ifdef FREEBSD_NAMECACHE
1856 cache_purgevfs(zfsvfs->z_parent->z_vfs);
1861 * Close the zil. NB: Can't close the zil while zfs_inactive
1862 * threads are blocked as zil_close can call zfs_inactive.
1864 if (zfsvfs->z_log) {
1865 zil_close(zfsvfs->z_log);
1866 zfsvfs->z_log = NULL;
1869 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1872 * If we are not unmounting (ie: online recv) and someone already
1873 * unmounted this file system while we were doing the switcheroo,
1874 * or a reopen of z_os failed then just bail out now.
1876 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1877 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1878 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1879 return (SET_ERROR(EIO));
1883 * At this point there are no vops active, and any new vops will
1884 * fail with EIO since we have z_teardown_lock for writer (only
1885 * relavent for forced unmount).
1887 * Release all holds on dbufs.
1889 mutex_enter(&zfsvfs->z_znodes_lock);
1890 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1891 zp = list_next(&zfsvfs->z_all_znodes, zp))
1893 ASSERT(ZTOV(zp)->v_count >= 0);
1894 zfs_znode_dmu_fini(zp);
1896 mutex_exit(&zfsvfs->z_znodes_lock);
1899 * If we are unmounting, set the unmounted flag and let new vops
1900 * unblock. zfs_inactive will have the unmounted behavior, and all
1901 * other vops will fail with EIO.
1904 zfsvfs->z_unmounted = B_TRUE;
1905 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1906 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1910 * z_os will be NULL if there was an error in attempting to reopen
1911 * zfsvfs, so just return as the properties had already been
1912 * unregistered and cached data had been evicted before.
1914 if (zfsvfs->z_os == NULL)
1918 * Unregister properties.
1920 zfs_unregister_callbacks(zfsvfs);
1925 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1926 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1927 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1928 dmu_objset_evict_dbufs(zfsvfs->z_os);
1935 zfs_umount(vfs_t *vfsp, int fflag)
1937 kthread_t *td = curthread;
1938 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1940 cred_t *cr = td->td_ucred;
1943 ret = secpolicy_fs_unmount(cr, vfsp);
1945 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1946 ZFS_DELEG_PERM_MOUNT, cr))
1951 * We purge the parent filesystem's vfsp as the parent filesystem
1952 * and all of its snapshots have their vnode's v_vfsp set to the
1953 * parent's filesystem's vfsp. Note, 'z_parent' is self
1954 * referential for non-snapshots.
1956 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1959 * Unmount any snapshots mounted under .zfs before unmounting the
1962 if (zfsvfs->z_ctldir != NULL) {
1963 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
1965 ret = vflush(vfsp, 0, 0, td);
1966 ASSERT(ret == EBUSY);
1967 if (!(fflag & MS_FORCE)) {
1968 if (zfsvfs->z_ctldir->v_count > 1)
1970 ASSERT(zfsvfs->z_ctldir->v_count == 1);
1972 zfsctl_destroy(zfsvfs);
1973 ASSERT(zfsvfs->z_ctldir == NULL);
1976 if (fflag & MS_FORCE) {
1978 * Mark file system as unmounted before calling
1979 * vflush(FORCECLOSE). This way we ensure no future vnops
1980 * will be called and risk operating on DOOMED vnodes.
1982 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1983 zfsvfs->z_unmounted = B_TRUE;
1984 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1988 * Flush all the files.
1990 ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
1992 if (!zfsvfs->z_issnap) {
1993 zfsctl_create(zfsvfs);
1994 ASSERT(zfsvfs->z_ctldir != NULL);
2000 if (!(fflag & MS_FORCE)) {
2002 * Check the number of active vnodes in the file system.
2003 * Our count is maintained in the vfs structure, but the
2004 * number is off by 1 to indicate a hold on the vfs
2007 * The '.zfs' directory maintains a reference of its
2008 * own, and any active references underneath are
2009 * reflected in the vnode count.
2011 if (zfsvfs->z_ctldir == NULL) {
2012 if (vfsp->vfs_count > 1)
2013 return (SET_ERROR(EBUSY));
2015 if (vfsp->vfs_count > 2 ||
2016 zfsvfs->z_ctldir->v_count > 1)
2017 return (SET_ERROR(EBUSY));
2022 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2026 * z_os will be NULL if there was an error in
2027 * attempting to reopen zfsvfs.
2031 * Unset the objset user_ptr.
2033 mutex_enter(&os->os_user_ptr_lock);
2034 dmu_objset_set_user(os, NULL);
2035 mutex_exit(&os->os_user_ptr_lock);
2038 * Finally release the objset
2040 dmu_objset_disown(os, zfsvfs);
2044 * We can now safely destroy the '.zfs' directory node.
2046 if (zfsvfs->z_ctldir != NULL)
2047 zfsctl_destroy(zfsvfs);
2048 if (zfsvfs->z_issnap) {
2049 vnode_t *svp = vfsp->mnt_vnodecovered;
2051 if (svp->v_count >= 2)
2060 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2062 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2067 * zfs_zget() can't operate on virtual entries like .zfs/ or
2068 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2069 * This will make NFS to switch to LOOKUP instead of using VGET.
2071 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2072 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2073 return (EOPNOTSUPP);
2076 err = zfs_zget(zfsvfs, ino, &zp);
2077 if (err == 0 && zp->z_unlinked) {
2085 err = vn_lock(*vpp, flags);
2092 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
2093 struct ucred **credanonp, int *numsecflavors, int **secflavors)
2095 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2098 * If this is regular file system vfsp is the same as
2099 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2100 * zfsvfs->z_parent->z_vfs represents parent file system
2101 * which we have to use here, because only this file system
2102 * has mnt_export configured.
2104 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2105 credanonp, numsecflavors, secflavors));
2108 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2109 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2112 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2114 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2116 uint64_t object = 0;
2117 uint64_t fid_gen = 0;
2127 * On FreeBSD we can get snapshot's mount point or its parent file
2128 * system mount point depending if snapshot is already mounted or not.
2130 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2131 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2132 uint64_t objsetid = 0;
2133 uint64_t setgen = 0;
2135 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2136 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2138 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2139 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2143 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2145 return (SET_ERROR(EINVAL));
2149 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2150 zfid_short_t *zfid = (zfid_short_t *)fidp;
2152 for (i = 0; i < sizeof (zfid->zf_object); i++)
2153 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2155 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2156 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2159 return (SET_ERROR(EINVAL));
2163 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2164 * directory tree. If the object == zfsvfs->z_shares_dir, then
2165 * we are in the .zfs/shares directory tree.
2167 if ((fid_gen == 0 &&
2168 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2169 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2170 *vpp = zfsvfs->z_ctldir;
2171 ASSERT(*vpp != NULL);
2172 if (object == ZFSCTL_INO_SNAPDIR) {
2173 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2174 0, NULL, NULL, NULL, NULL, NULL) == 0);
2175 } else if (object == zfsvfs->z_shares_dir) {
2176 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL,
2177 0, NULL, NULL, NULL, NULL, NULL) == 0);
2182 err = vn_lock(*vpp, flags);
2188 gen_mask = -1ULL >> (64 - 8 * i);
2190 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2191 if (err = zfs_zget(zfsvfs, object, &zp)) {
2195 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2197 zp_gen = zp_gen & gen_mask;
2200 if (zp->z_unlinked || zp_gen != fid_gen) {
2201 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2204 return (SET_ERROR(EINVAL));
2209 err = vn_lock(*vpp, flags | LK_RETRY);
2211 vnode_create_vobject(*vpp, zp->z_size, curthread);
2218 * Block out VOPs and close zfsvfs_t::z_os
2220 * Note, if successful, then we return with the 'z_teardown_lock' and
2221 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2222 * dataset and objset intact so that they can be atomically handed off during
2223 * a subsequent rollback or recv operation and the resume thereafter.
2226 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2230 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2237 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2238 * is an invariant across any of the operations that can be performed while the
2239 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2240 * are the same: the relevant objset and associated dataset are owned by
2241 * zfsvfs, held, and long held on entry.
2244 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2248 uint64_t sa_obj = 0;
2250 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2251 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2254 * We already own this, so just hold and rele it to update the
2255 * objset_t, as the one we had before may have been evicted.
2257 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os));
2258 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs);
2259 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset));
2260 dmu_objset_rele(zfsvfs->z_os, zfsvfs);
2263 * Make sure version hasn't changed
2266 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
2267 &zfsvfs->z_version);
2272 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2273 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2275 if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
2278 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2279 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
2282 if (zfsvfs->z_version >= ZPL_VERSION_SA)
2283 sa_register_update_callback(zfsvfs->z_os,
2286 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2288 zfs_set_fuid_feature(zfsvfs);
2291 * Attempt to re-establish all the active znodes with
2292 * their dbufs. If a zfs_rezget() fails, then we'll let
2293 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2294 * when they try to use their znode.
2296 mutex_enter(&zfsvfs->z_znodes_lock);
2297 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2298 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2299 (void) zfs_rezget(zp);
2301 mutex_exit(&zfsvfs->z_znodes_lock);
2304 /* release the VOPs */
2305 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2306 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2310 * Since we couldn't setup the sa framework, try to force
2311 * unmount this file system.
2313 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2314 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2320 zfs_freevfs(vfs_t *vfsp)
2322 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2326 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2327 * from zfs_mount(). Release it here. If we came through
2328 * zfs_mountroot() instead, we didn't grab an extra hold, so
2329 * skip the VFS_RELE for rootvfs.
2331 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2332 VFS_RELE(zfsvfs->z_parent->z_vfs);
2335 zfsvfs_free(zfsvfs);
2337 atomic_dec_32(&zfs_active_fs_count);
2341 static int desiredvnodes_backup;
2345 zfs_vnodes_adjust(void)
2348 int newdesiredvnodes;
2350 desiredvnodes_backup = desiredvnodes;
2353 * We calculate newdesiredvnodes the same way it is done in
2354 * vntblinit(). If it is equal to desiredvnodes, it means that
2355 * it wasn't tuned by the administrator and we can tune it down.
2357 newdesiredvnodes = min(maxproc + vm_cnt.v_page_count / 4, 2 *
2358 vm_kmem_size / (5 * (sizeof(struct vm_object) +
2359 sizeof(struct vnode))));
2360 if (newdesiredvnodes == desiredvnodes)
2361 desiredvnodes = (3 * newdesiredvnodes) / 4;
2366 zfs_vnodes_adjust_back(void)
2370 desiredvnodes = desiredvnodes_backup;
2378 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2381 * Initialize .zfs directory structures
2386 * Initialize znode cache, vnode ops, etc...
2391 * Reduce number of vnodes. Originally number of vnodes is calculated
2392 * with UFS inode in mind. We reduce it here, because it's too big for
2395 zfs_vnodes_adjust();
2397 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2405 zfs_vnodes_adjust_back();
2411 return (zfs_active_fs_count != 0);
2415 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2418 objset_t *os = zfsvfs->z_os;
2421 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2422 return (SET_ERROR(EINVAL));
2424 if (newvers < zfsvfs->z_version)
2425 return (SET_ERROR(EINVAL));
2427 if (zfs_spa_version_map(newvers) >
2428 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2429 return (SET_ERROR(ENOTSUP));
2431 tx = dmu_tx_create(os);
2432 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2433 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2434 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2436 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2438 error = dmu_tx_assign(tx, TXG_WAIT);
2444 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2445 8, 1, &newvers, tx);
2452 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2455 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2457 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2458 DMU_OT_NONE, 0, tx);
2460 error = zap_add(os, MASTER_NODE_OBJ,
2461 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2464 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2465 sa_register_update_callback(os, zfs_sa_upgrade);
2468 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2469 "from %llu to %llu", zfsvfs->z_version, newvers);
2473 zfsvfs->z_version = newvers;
2475 zfs_set_fuid_feature(zfsvfs);
2481 * Read a property stored within the master node.
2484 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2490 * Look up the file system's value for the property. For the
2491 * version property, we look up a slightly different string.
2493 if (prop == ZFS_PROP_VERSION)
2494 pname = ZPL_VERSION_STR;
2496 pname = zfs_prop_to_name(prop);
2499 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2501 if (error == ENOENT) {
2502 /* No value set, use the default value */
2504 case ZFS_PROP_VERSION:
2505 *value = ZPL_VERSION;
2507 case ZFS_PROP_NORMALIZE:
2508 case ZFS_PROP_UTF8ONLY:
2512 *value = ZFS_CASE_SENSITIVE;
2524 zfsvfs_update_fromname(const char *oldname, const char *newname)
2526 char tmpbuf[MAXPATHLEN];
2531 oldlen = strlen(oldname);
2533 mtx_lock(&mountlist_mtx);
2534 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2535 fromname = mp->mnt_stat.f_mntfromname;
2536 if (strcmp(fromname, oldname) == 0) {
2537 (void)strlcpy(fromname, newname,
2538 sizeof(mp->mnt_stat.f_mntfromname));
2541 if (strncmp(fromname, oldname, oldlen) == 0 &&
2542 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2543 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
2544 newname, fromname + oldlen);
2545 (void)strlcpy(fromname, tmpbuf,
2546 sizeof(mp->mnt_stat.f_mntfromname));
2550 mtx_unlock(&mountlist_mtx);