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.
136 * Ignore the system syncher. ZFS already commits async data
137 * at zfs_txg_timeout intervals.
139 if (waitfor == MNT_LAZY)
144 * Sync a specific filesystem.
146 zfsvfs_t *zfsvfs = vfsp->vfs_data;
150 error = vfs_stdsync(vfsp, waitfor);
155 dp = dmu_objset_pool(zfsvfs->z_os);
158 * If the system is shutting down, then skip any
159 * filesystems which may exist on a suspended pool.
161 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
166 if (zfsvfs->z_log != NULL)
167 zil_commit(zfsvfs->z_log, 0);
172 * Sync all ZFS filesystems. This is what happens when you
173 * run sync(1M). Unlike other filesystems, ZFS honors the
174 * request by waiting for all pools to commit all dirty data.
182 #ifndef __FreeBSD_kernel__
184 zfs_create_unique_device(dev_t *dev)
189 ASSERT3U(zfs_minor, <=, MAXMIN32);
190 minor_t start = zfs_minor;
192 mutex_enter(&zfs_dev_mtx);
193 if (zfs_minor >= MAXMIN32) {
195 * If we're still using the real major
196 * keep out of /dev/zfs and /dev/zvol minor
197 * number space. If we're using a getudev()'ed
198 * major number, we can use all of its minors.
200 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
201 zfs_minor = ZFS_MIN_MINOR;
207 *dev = makedevice(zfs_major, zfs_minor);
208 mutex_exit(&zfs_dev_mtx);
209 } while (vfs_devismounted(*dev) && zfs_minor != start);
210 if (zfs_minor == start) {
212 * We are using all ~262,000 minor numbers for the
213 * current major number. Create a new major number.
215 if ((new_major = getudev()) == (major_t)-1) {
217 "zfs_mount: Can't get unique major "
221 mutex_enter(&zfs_dev_mtx);
222 zfs_major = new_major;
225 mutex_exit(&zfs_dev_mtx);
229 /* CONSTANTCONDITION */
234 #endif /* !__FreeBSD_kernel__ */
237 atime_changed_cb(void *arg, uint64_t newval)
239 zfsvfs_t *zfsvfs = arg;
241 if (newval == TRUE) {
242 zfsvfs->z_atime = TRUE;
243 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
244 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
245 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
247 zfsvfs->z_atime = FALSE;
248 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
249 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
250 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
255 xattr_changed_cb(void *arg, uint64_t newval)
257 zfsvfs_t *zfsvfs = arg;
259 if (newval == TRUE) {
260 /* XXX locking on vfs_flag? */
262 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
264 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
265 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
267 /* XXX locking on vfs_flag? */
269 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
271 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
272 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
277 blksz_changed_cb(void *arg, uint64_t newval)
279 zfsvfs_t *zfsvfs = arg;
280 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
281 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
282 ASSERT(ISP2(newval));
284 zfsvfs->z_max_blksz = newval;
285 zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
289 readonly_changed_cb(void *arg, uint64_t newval)
291 zfsvfs_t *zfsvfs = arg;
294 /* XXX locking on vfs_flag? */
295 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
296 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
297 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
299 /* XXX locking on vfs_flag? */
300 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
301 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
302 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
307 setuid_changed_cb(void *arg, uint64_t newval)
309 zfsvfs_t *zfsvfs = arg;
311 if (newval == FALSE) {
312 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
313 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
314 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
316 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
317 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
318 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
323 exec_changed_cb(void *arg, uint64_t newval)
325 zfsvfs_t *zfsvfs = arg;
327 if (newval == FALSE) {
328 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
329 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
330 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
332 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
333 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
334 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
339 * The nbmand mount option can be changed at mount time.
340 * We can't allow it to be toggled on live file systems or incorrect
341 * behavior may be seen from cifs clients
343 * This property isn't registered via dsl_prop_register(), but this callback
344 * will be called when a file system is first mounted
347 nbmand_changed_cb(void *arg, uint64_t newval)
349 zfsvfs_t *zfsvfs = arg;
350 if (newval == FALSE) {
351 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
352 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
354 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
355 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
360 snapdir_changed_cb(void *arg, uint64_t newval)
362 zfsvfs_t *zfsvfs = arg;
364 zfsvfs->z_show_ctldir = newval;
368 vscan_changed_cb(void *arg, uint64_t newval)
370 zfsvfs_t *zfsvfs = arg;
372 zfsvfs->z_vscan = newval;
376 acl_mode_changed_cb(void *arg, uint64_t newval)
378 zfsvfs_t *zfsvfs = arg;
380 zfsvfs->z_acl_mode = newval;
384 acl_inherit_changed_cb(void *arg, uint64_t newval)
386 zfsvfs_t *zfsvfs = arg;
388 zfsvfs->z_acl_inherit = newval;
392 zfs_register_callbacks(vfs_t *vfsp)
394 struct dsl_dataset *ds = NULL;
396 zfsvfs_t *zfsvfs = NULL;
398 boolean_t readonly = B_FALSE;
399 boolean_t do_readonly = B_FALSE;
400 boolean_t setuid = B_FALSE;
401 boolean_t do_setuid = B_FALSE;
402 boolean_t exec = B_FALSE;
403 boolean_t do_exec = B_FALSE;
405 boolean_t devices = B_FALSE;
406 boolean_t do_devices = B_FALSE;
408 boolean_t xattr = B_FALSE;
409 boolean_t do_xattr = B_FALSE;
410 boolean_t atime = B_FALSE;
411 boolean_t do_atime = B_FALSE;
415 zfsvfs = vfsp->vfs_data;
420 * This function can be called for a snapshot when we update snapshot's
421 * mount point, which isn't really supported.
423 if (dmu_objset_is_snapshot(os))
427 * The act of registering our callbacks will destroy any mount
428 * options we may have. In order to enable temporary overrides
429 * of mount options, we stash away the current values and
430 * restore them after we register the callbacks.
432 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
433 !spa_writeable(dmu_objset_spa(os))) {
435 do_readonly = B_TRUE;
436 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
438 do_readonly = B_TRUE;
440 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
444 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
447 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
452 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
455 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
459 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
462 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
466 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
469 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
475 * nbmand is a special property. It can only be changed at
478 * This is weird, but it is documented to only be changeable
481 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
483 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
486 char osname[MAXNAMELEN];
488 dmu_objset_name(os, osname);
489 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
496 * Register property callbacks.
498 * It would probably be fine to just check for i/o error from
499 * the first prop_register(), but I guess I like to go
502 ds = dmu_objset_ds(os);
503 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
504 error = dsl_prop_register(ds,
505 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
506 error = error ? error : dsl_prop_register(ds,
507 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
508 error = error ? error : dsl_prop_register(ds,
509 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
510 error = error ? error : dsl_prop_register(ds,
511 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
513 error = error ? error : dsl_prop_register(ds,
514 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
516 error = error ? error : dsl_prop_register(ds,
517 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
518 error = error ? error : dsl_prop_register(ds,
519 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
520 error = error ? error : dsl_prop_register(ds,
521 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
522 error = error ? error : dsl_prop_register(ds,
523 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
524 error = error ? error : dsl_prop_register(ds,
525 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
527 error = error ? error : dsl_prop_register(ds,
528 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
529 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
534 * Invoke our callbacks to restore temporary mount options.
537 readonly_changed_cb(zfsvfs, readonly);
539 setuid_changed_cb(zfsvfs, setuid);
541 exec_changed_cb(zfsvfs, exec);
543 xattr_changed_cb(zfsvfs, xattr);
545 atime_changed_cb(zfsvfs, atime);
547 nbmand_changed_cb(zfsvfs, nbmand);
553 * We may attempt to unregister some callbacks that are not
554 * registered, but this is OK; it will simply return ENOMSG,
555 * which we will ignore.
557 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
558 atime_changed_cb, zfsvfs);
559 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
560 xattr_changed_cb, zfsvfs);
561 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
562 blksz_changed_cb, zfsvfs);
563 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
564 readonly_changed_cb, zfsvfs);
566 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
567 devices_changed_cb, zfsvfs);
569 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
570 setuid_changed_cb, zfsvfs);
571 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
572 exec_changed_cb, zfsvfs);
573 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
574 snapdir_changed_cb, zfsvfs);
575 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE),
576 acl_mode_changed_cb, zfsvfs);
577 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
578 acl_inherit_changed_cb, zfsvfs);
579 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
580 vscan_changed_cb, zfsvfs);
585 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
586 uint64_t *userp, uint64_t *groupp)
589 * Is it a valid type of object to track?
591 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
592 return (SET_ERROR(ENOENT));
595 * If we have a NULL data pointer
596 * then assume the id's aren't changing and
597 * return EEXIST to the dmu to let it know to
601 return (SET_ERROR(EEXIST));
603 if (bonustype == DMU_OT_ZNODE) {
604 znode_phys_t *znp = data;
605 *userp = znp->zp_uid;
606 *groupp = znp->zp_gid;
609 sa_hdr_phys_t *sap = data;
610 sa_hdr_phys_t sa = *sap;
611 boolean_t swap = B_FALSE;
613 ASSERT(bonustype == DMU_OT_SA);
615 if (sa.sa_magic == 0) {
617 * This should only happen for newly created
618 * files that haven't had the znode data filled
625 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
626 sa.sa_magic = SA_MAGIC;
627 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
630 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
633 hdrsize = sa_hdrsize(&sa);
634 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
635 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
637 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
640 *userp = BSWAP_64(*userp);
641 *groupp = BSWAP_64(*groupp);
648 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
649 char *domainbuf, int buflen, uid_t *ridp)
654 fuid = strtonum(fuidstr, NULL);
656 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
658 (void) strlcpy(domainbuf, domain, buflen);
661 *ridp = FUID_RID(fuid);
665 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
668 case ZFS_PROP_USERUSED:
669 return (DMU_USERUSED_OBJECT);
670 case ZFS_PROP_GROUPUSED:
671 return (DMU_GROUPUSED_OBJECT);
672 case ZFS_PROP_USERQUOTA:
673 return (zfsvfs->z_userquota_obj);
674 case ZFS_PROP_GROUPQUOTA:
675 return (zfsvfs->z_groupquota_obj);
681 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
682 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
687 zfs_useracct_t *buf = vbuf;
690 if (!dmu_objset_userspace_present(zfsvfs->z_os))
691 return (SET_ERROR(ENOTSUP));
693 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
699 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
700 (error = zap_cursor_retrieve(&zc, &za)) == 0;
701 zap_cursor_advance(&zc)) {
702 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
706 fuidstr_to_sid(zfsvfs, za.za_name,
707 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
709 buf->zu_space = za.za_first_integer;
715 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
716 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
717 *cookiep = zap_cursor_serialize(&zc);
718 zap_cursor_fini(&zc);
723 * buf must be big enough (eg, 32 bytes)
726 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
727 char *buf, boolean_t addok)
732 if (domain && domain[0]) {
733 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
735 return (SET_ERROR(ENOENT));
737 fuid = FUID_ENCODE(domainid, rid);
738 (void) sprintf(buf, "%llx", (longlong_t)fuid);
743 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
744 const char *domain, uint64_t rid, uint64_t *valp)
752 if (!dmu_objset_userspace_present(zfsvfs->z_os))
753 return (SET_ERROR(ENOTSUP));
755 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
759 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
763 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
770 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
771 const char *domain, uint64_t rid, uint64_t quota)
777 boolean_t fuid_dirtied;
779 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
780 return (SET_ERROR(EINVAL));
782 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
783 return (SET_ERROR(ENOTSUP));
785 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
786 &zfsvfs->z_groupquota_obj;
788 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
791 fuid_dirtied = zfsvfs->z_fuid_dirty;
793 tx = dmu_tx_create(zfsvfs->z_os);
794 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
796 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
797 zfs_userquota_prop_prefixes[type]);
800 zfs_fuid_txhold(zfsvfs, tx);
801 err = dmu_tx_assign(tx, TXG_WAIT);
807 mutex_enter(&zfsvfs->z_lock);
809 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
811 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
812 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
814 mutex_exit(&zfsvfs->z_lock);
817 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
821 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
825 zfs_fuid_sync(zfsvfs, tx);
831 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
834 uint64_t used, quota, usedobj, quotaobj;
837 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
838 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
840 if (quotaobj == 0 || zfsvfs->z_replay)
843 (void) sprintf(buf, "%llx", (longlong_t)fuid);
844 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
848 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
851 return (used >= quota);
855 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
860 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
862 fuid = isgroup ? zp->z_gid : zp->z_uid;
864 if (quotaobj == 0 || zfsvfs->z_replay)
867 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
871 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
880 * XXX: Fix struct statfs so this isn't necessary!
882 * The 'osname' is used as the filesystem's special node, which means
883 * it must fit in statfs.f_mntfromname, or else it can't be
884 * enumerated, so libzfs_mnttab_find() returns NULL, which causes
885 * 'zfs unmount' to think it's not mounted when it is.
887 if (strlen(osname) >= MNAMELEN)
888 return (SET_ERROR(ENAMETOOLONG));
890 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
893 * We claim to always be readonly so we can open snapshots;
894 * other ZPL code will prevent us from writing to snapshots.
896 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
898 kmem_free(zfsvfs, sizeof (zfsvfs_t));
903 * Initialize the zfs-specific filesystem structure.
904 * Should probably make this a kmem cache, shuffle fields,
905 * and just bzero up to z_hold_mtx[].
907 zfsvfs->z_vfs = NULL;
908 zfsvfs->z_parent = zfsvfs;
909 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
910 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
913 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
916 } else if (zfsvfs->z_version >
917 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
918 (void) printf("Can't mount a version %lld file system "
919 "on a version %lld pool\n. Pool must be upgraded to mount "
920 "this file system.", (u_longlong_t)zfsvfs->z_version,
921 (u_longlong_t)spa_version(dmu_objset_spa(os)));
922 error = SET_ERROR(ENOTSUP);
925 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
927 zfsvfs->z_norm = (int)zval;
929 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
931 zfsvfs->z_utf8 = (zval != 0);
933 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
935 zfsvfs->z_case = (uint_t)zval;
938 * Fold case on file systems that are always or sometimes case
941 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
942 zfsvfs->z_case == ZFS_CASE_MIXED)
943 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
945 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
946 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
948 if (zfsvfs->z_use_sa) {
949 /* should either have both of these objects or none */
950 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
956 * Pre SA versions file systems should never touch
957 * either the attribute registration or layout objects.
962 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
963 &zfsvfs->z_attr_table);
967 if (zfsvfs->z_version >= ZPL_VERSION_SA)
968 sa_register_update_callback(os, zfs_sa_upgrade);
970 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
974 ASSERT(zfsvfs->z_root != 0);
976 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
977 &zfsvfs->z_unlinkedobj);
981 error = zap_lookup(os, MASTER_NODE_OBJ,
982 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
983 8, 1, &zfsvfs->z_userquota_obj);
984 if (error && error != ENOENT)
987 error = zap_lookup(os, MASTER_NODE_OBJ,
988 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
989 8, 1, &zfsvfs->z_groupquota_obj);
990 if (error && error != ENOENT)
993 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
994 &zfsvfs->z_fuid_obj);
995 if (error && error != ENOENT)
998 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
999 &zfsvfs->z_shares_dir);
1000 if (error && error != ENOENT)
1003 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1004 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1005 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1006 offsetof(znode_t, z_link_node));
1007 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1008 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1009 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1010 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1011 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1017 dmu_objset_disown(os, zfsvfs);
1019 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1024 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1028 error = zfs_register_callbacks(zfsvfs->z_vfs);
1033 * Set the objset user_ptr to track its zfsvfs.
1035 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1036 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1037 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1039 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1042 * If we are not mounting (ie: online recv), then we don't
1043 * have to worry about replaying the log as we blocked all
1044 * operations out since we closed the ZIL.
1050 * During replay we remove the read only flag to
1051 * allow replays to succeed.
1053 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1055 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1057 zfs_unlinked_drain(zfsvfs);
1060 * Parse and replay the intent log.
1062 * Because of ziltest, this must be done after
1063 * zfs_unlinked_drain(). (Further note: ziltest
1064 * doesn't use readonly mounts, where
1065 * zfs_unlinked_drain() isn't called.) This is because
1066 * ziltest causes spa_sync() to think it's committed,
1067 * but actually it is not, so the intent log contains
1068 * many txg's worth of changes.
1070 * In particular, if object N is in the unlinked set in
1071 * the last txg to actually sync, then it could be
1072 * actually freed in a later txg and then reallocated
1073 * in a yet later txg. This would write a "create
1074 * object N" record to the intent log. Normally, this
1075 * would be fine because the spa_sync() would have
1076 * written out the fact that object N is free, before
1077 * we could write the "create object N" intent log
1080 * But when we are in ziltest mode, we advance the "open
1081 * txg" without actually spa_sync()-ing the changes to
1082 * disk. So we would see that object N is still
1083 * allocated and in the unlinked set, and there is an
1084 * intent log record saying to allocate it.
1086 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1087 if (zil_replay_disable) {
1088 zil_destroy(zfsvfs->z_log, B_FALSE);
1090 zfsvfs->z_replay = B_TRUE;
1091 zil_replay(zfsvfs->z_os, zfsvfs,
1093 zfsvfs->z_replay = B_FALSE;
1096 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1102 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1105 zfsvfs_free(zfsvfs_t *zfsvfs)
1110 * This is a barrier to prevent the filesystem from going away in
1111 * zfs_znode_move() until we can safely ensure that the filesystem is
1112 * not unmounted. We consider the filesystem valid before the barrier
1113 * and invalid after the barrier.
1115 rw_enter(&zfsvfs_lock, RW_READER);
1116 rw_exit(&zfsvfs_lock);
1118 zfs_fuid_destroy(zfsvfs);
1120 mutex_destroy(&zfsvfs->z_znodes_lock);
1121 mutex_destroy(&zfsvfs->z_lock);
1122 list_destroy(&zfsvfs->z_all_znodes);
1123 rrm_destroy(&zfsvfs->z_teardown_lock);
1124 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1125 rw_destroy(&zfsvfs->z_fuid_lock);
1126 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1127 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1128 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1132 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1134 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1135 if (zfsvfs->z_vfs) {
1136 if (zfsvfs->z_use_fuids) {
1137 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1138 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1139 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1140 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1141 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1142 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1144 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1145 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1146 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1147 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1148 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1149 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1152 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1156 zfs_domount(vfs_t *vfsp, char *osname)
1158 uint64_t recordsize, fsid_guid;
1166 error = zfsvfs_create(osname, &zfsvfs);
1169 zfsvfs->z_vfs = vfsp;
1172 /* Initialize the generic filesystem structure. */
1173 vfsp->vfs_bcount = 0;
1174 vfsp->vfs_data = NULL;
1176 if (zfs_create_unique_device(&mount_dev) == -1) {
1177 error = SET_ERROR(ENODEV);
1180 ASSERT(vfs_devismounted(mount_dev) == 0);
1183 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1186 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1187 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1189 vfsp->vfs_data = zfsvfs;
1190 vfsp->mnt_flag |= MNT_LOCAL;
1191 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1192 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1193 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1196 * The fsid is 64 bits, composed of an 8-bit fs type, which
1197 * separates our fsid from any other filesystem types, and a
1198 * 56-bit objset unique ID. The objset unique ID is unique to
1199 * all objsets open on this system, provided by unique_create().
1200 * The 8-bit fs type must be put in the low bits of fsid[1]
1201 * because that's where other Solaris filesystems put it.
1203 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1204 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1205 vfsp->vfs_fsid.val[0] = fsid_guid;
1206 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1207 vfsp->mnt_vfc->vfc_typenum & 0xFF;
1210 * Set features for file system.
1212 zfs_set_fuid_feature(zfsvfs);
1213 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1214 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1215 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1216 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1217 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1218 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1219 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1221 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1223 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1226 atime_changed_cb(zfsvfs, B_FALSE);
1227 readonly_changed_cb(zfsvfs, B_TRUE);
1228 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1230 xattr_changed_cb(zfsvfs, pval);
1231 zfsvfs->z_issnap = B_TRUE;
1232 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1234 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1235 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1236 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1238 error = zfsvfs_setup(zfsvfs, B_TRUE);
1241 vfs_mountedfrom(vfsp, osname);
1243 if (!zfsvfs->z_issnap)
1244 zfsctl_create(zfsvfs);
1247 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1248 zfsvfs_free(zfsvfs);
1250 atomic_inc_32(&zfs_active_fs_count);
1257 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1259 objset_t *os = zfsvfs->z_os;
1260 struct dsl_dataset *ds;
1263 * Unregister properties.
1265 if (!dmu_objset_is_snapshot(os)) {
1266 ds = dmu_objset_ds(os);
1267 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1270 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1273 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1276 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1279 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1282 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1285 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1288 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
1291 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1292 acl_inherit_changed_cb, zfsvfs) == 0);
1294 VERIFY(dsl_prop_unregister(ds, "vscan",
1295 vscan_changed_cb, zfsvfs) == 0);
1301 * Convert a decimal digit string to a uint64_t integer.
1304 str_to_uint64(char *str, uint64_t *objnum)
1309 if (*str < '0' || *str > '9')
1310 return (SET_ERROR(EINVAL));
1312 num = num*10 + *str++ - '0';
1320 * The boot path passed from the boot loader is in the form of
1321 * "rootpool-name/root-filesystem-object-number'. Convert this
1322 * string to a dataset name: "rootpool-name/root-filesystem-name".
1325 zfs_parse_bootfs(char *bpath, char *outpath)
1331 if (*bpath == 0 || *bpath == '/')
1332 return (SET_ERROR(EINVAL));
1334 (void) strcpy(outpath, bpath);
1336 slashp = strchr(bpath, '/');
1338 /* if no '/', just return the pool name */
1339 if (slashp == NULL) {
1343 /* if not a number, just return the root dataset name */
1344 if (str_to_uint64(slashp+1, &objnum)) {
1349 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1356 * Check that the hex label string is appropriate for the dataset being
1357 * mounted into the global_zone proper.
1359 * Return an error if the hex label string is not default or
1360 * admin_low/admin_high. For admin_low labels, the corresponding
1361 * dataset must be readonly.
1364 zfs_check_global_label(const char *dsname, const char *hexsl)
1366 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1368 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1370 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1371 /* must be readonly */
1374 if (dsl_prop_get_integer(dsname,
1375 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1376 return (SET_ERROR(EACCES));
1377 return (rdonly ? 0 : EACCES);
1379 return (SET_ERROR(EACCES));
1383 * Determine whether the mount is allowed according to MAC check.
1384 * by comparing (where appropriate) label of the dataset against
1385 * the label of the zone being mounted into. If the dataset has
1386 * no label, create one.
1388 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1391 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1394 zone_t *mntzone = NULL;
1395 ts_label_t *mnt_tsl;
1398 char ds_hexsl[MAXNAMELEN];
1400 retv = EACCES; /* assume the worst */
1403 * Start by getting the dataset label if it exists.
1405 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1406 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1408 return (SET_ERROR(EACCES));
1411 * If labeling is NOT enabled, then disallow the mount of datasets
1412 * which have a non-default label already. No other label checks
1415 if (!is_system_labeled()) {
1416 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1418 return (SET_ERROR(EACCES));
1422 * Get the label of the mountpoint. If mounting into the global
1423 * zone (i.e. mountpoint is not within an active zone and the
1424 * zoned property is off), the label must be default or
1425 * admin_low/admin_high only; no other checks are needed.
1427 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1428 if (mntzone->zone_id == GLOBAL_ZONEID) {
1433 if (dsl_prop_get_integer(osname,
1434 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1435 return (SET_ERROR(EACCES));
1437 return (zfs_check_global_label(osname, ds_hexsl));
1440 * This is the case of a zone dataset being mounted
1441 * initially, before the zone has been fully created;
1442 * allow this mount into global zone.
1447 mnt_tsl = mntzone->zone_slabel;
1448 ASSERT(mnt_tsl != NULL);
1449 label_hold(mnt_tsl);
1450 mnt_sl = label2bslabel(mnt_tsl);
1452 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1454 * The dataset doesn't have a real label, so fabricate one.
1458 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1459 dsl_prop_set_string(osname,
1460 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1461 ZPROP_SRC_LOCAL, str) == 0)
1464 kmem_free(str, strlen(str) + 1);
1465 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1467 * Now compare labels to complete the MAC check. If the
1468 * labels are equal then allow access. If the mountpoint
1469 * label dominates the dataset label, allow readonly access.
1470 * Otherwise, access is denied.
1472 if (blequal(mnt_sl, &ds_sl))
1474 else if (bldominates(mnt_sl, &ds_sl)) {
1475 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1480 label_rele(mnt_tsl);
1484 #endif /* SECLABEL */
1486 #ifdef OPENSOLARIS_MOUNTROOT
1488 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1491 static int zfsrootdone = 0;
1492 zfsvfs_t *zfsvfs = NULL;
1501 * The filesystem that we mount as root is defined in the
1502 * boot property "zfs-bootfs" with a format of
1503 * "poolname/root-dataset-objnum".
1505 if (why == ROOT_INIT) {
1507 return (SET_ERROR(EBUSY));
1509 * the process of doing a spa_load will require the
1510 * clock to be set before we could (for example) do
1511 * something better by looking at the timestamp on
1512 * an uberblock, so just set it to -1.
1516 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1517 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1519 return (SET_ERROR(EINVAL));
1521 zfs_devid = spa_get_bootprop("diskdevid");
1522 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1524 spa_free_bootprop(zfs_devid);
1526 spa_free_bootprop(zfs_bootfs);
1527 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1531 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1532 spa_free_bootprop(zfs_bootfs);
1533 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1538 spa_free_bootprop(zfs_bootfs);
1540 if (error = vfs_lock(vfsp))
1543 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1544 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1548 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1550 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1551 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1556 mutex_enter(&vp->v_lock);
1557 vp->v_flag |= VROOT;
1558 mutex_exit(&vp->v_lock);
1562 * Leave rootvp held. The root file system is never unmounted.
1565 vfs_add((struct vnode *)0, vfsp,
1566 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1570 } else if (why == ROOT_REMOUNT) {
1571 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1572 vfsp->vfs_flag |= VFS_REMOUNT;
1574 /* refresh mount options */
1575 zfs_unregister_callbacks(vfsp->vfs_data);
1576 return (zfs_register_callbacks(vfsp));
1578 } else if (why == ROOT_UNMOUNT) {
1579 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1580 (void) zfs_sync(vfsp, 0, 0);
1585 * if "why" is equal to anything else other than ROOT_INIT,
1586 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1588 return (SET_ERROR(ENOTSUP));
1590 #endif /* OPENSOLARIS_MOUNTROOT */
1593 getpoolname(const char *osname, char *poolname)
1597 p = strchr(osname, '/');
1599 if (strlen(osname) >= MAXNAMELEN)
1600 return (ENAMETOOLONG);
1601 (void) strcpy(poolname, osname);
1603 if (p - osname >= MAXNAMELEN)
1604 return (ENAMETOOLONG);
1605 (void) strncpy(poolname, osname, p - osname);
1606 poolname[p - osname] = '\0';
1613 zfs_mount(vfs_t *vfsp)
1615 kthread_t *td = curthread;
1616 vnode_t *mvp = vfsp->mnt_vnodecovered;
1617 cred_t *cr = td->td_ucred;
1623 if (mvp->v_type != VDIR)
1624 return (SET_ERROR(ENOTDIR));
1626 mutex_enter(&mvp->v_lock);
1627 if ((uap->flags & MS_REMOUNT) == 0 &&
1628 (uap->flags & MS_OVERLAY) == 0 &&
1629 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1630 mutex_exit(&mvp->v_lock);
1631 return (SET_ERROR(EBUSY));
1633 mutex_exit(&mvp->v_lock);
1636 * ZFS does not support passing unparsed data in via MS_DATA.
1637 * Users should use the MS_OPTIONSTR interface; this means
1638 * that all option parsing is already done and the options struct
1639 * can be interrogated.
1641 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1642 #else /* !illumos */
1643 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS))
1644 return (SET_ERROR(EPERM));
1646 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1647 return (SET_ERROR(EINVAL));
1648 #endif /* illumos */
1651 * If full-owner-access is enabled and delegated administration is
1652 * turned on, we must set nosuid.
1654 if (zfs_super_owner &&
1655 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1656 secpolicy_fs_mount_clearopts(cr, vfsp);
1660 * Check for mount privilege?
1662 * If we don't have privilege then see if
1663 * we have local permission to allow it
1665 error = secpolicy_fs_mount(cr, mvp, vfsp);
1667 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1670 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1674 * Make sure user is the owner of the mount point
1675 * or has sufficient privileges.
1678 vattr.va_mask = AT_UID;
1680 vn_lock(mvp, LK_SHARED | LK_RETRY);
1681 if (VOP_GETATTR(mvp, &vattr, cr)) {
1686 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1687 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1694 secpolicy_fs_mount_clearopts(cr, vfsp);
1698 * Refuse to mount a filesystem if we are in a local zone and the
1699 * dataset is not visible.
1701 if (!INGLOBALZONE(curthread) &&
1702 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1703 error = SET_ERROR(EPERM);
1708 error = zfs_mount_label_policy(vfsp, osname);
1713 vfsp->vfs_flag |= MNT_NFS4ACLS;
1716 * When doing a remount, we simply refresh our temporary properties
1717 * according to those options set in the current VFS options.
1719 if (vfsp->vfs_flag & MS_REMOUNT) {
1720 /* refresh mount options */
1721 zfs_unregister_callbacks(vfsp->vfs_data);
1722 error = zfs_register_callbacks(vfsp);
1726 /* Initial root mount: try hard to import the requested root pool. */
1727 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1728 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1729 char pname[MAXNAMELEN];
1731 error = getpoolname(osname, pname);
1733 error = spa_import_rootpool(pname);
1738 error = zfs_domount(vfsp, osname);
1743 * Add an extra VFS_HOLD on our parent vfs so that it can't
1744 * disappear due to a forced unmount.
1746 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1747 VFS_HOLD(mvp->v_vfsp);
1755 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1757 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1758 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1760 statp->f_version = STATFS_VERSION;
1764 dmu_objset_space(zfsvfs->z_os,
1765 &refdbytes, &availbytes, &usedobjs, &availobjs);
1768 * The underlying storage pool actually uses multiple block sizes.
1769 * We report the fragsize as the smallest block size we support,
1770 * and we report our blocksize as the filesystem's maximum blocksize.
1772 statp->f_bsize = SPA_MINBLOCKSIZE;
1773 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1776 * The following report "total" blocks of various kinds in the
1777 * file system, but reported in terms of f_frsize - the
1781 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1782 statp->f_bfree = availbytes / statp->f_bsize;
1783 statp->f_bavail = statp->f_bfree; /* no root reservation */
1786 * statvfs() should really be called statufs(), because it assumes
1787 * static metadata. ZFS doesn't preallocate files, so the best
1788 * we can do is report the max that could possibly fit in f_files,
1789 * and that minus the number actually used in f_ffree.
1790 * For f_ffree, report the smaller of the number of object available
1791 * and the number of blocks (each object will take at least a block).
1793 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1794 statp->f_files = statp->f_ffree + usedobjs;
1797 * We're a zfs filesystem.
1799 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
1801 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
1802 sizeof(statp->f_mntfromname));
1803 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
1804 sizeof(statp->f_mntonname));
1806 statp->f_namemax = ZFS_MAXNAMELEN;
1813 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
1815 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1821 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1823 *vpp = ZTOV(rootzp);
1828 error = vn_lock(*vpp, flags);
1830 (*vpp)->v_vflag |= VV_ROOT;
1839 * Teardown the zfsvfs::z_os.
1841 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1842 * and 'z_teardown_inactive_lock' held.
1845 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1849 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1853 * We purge the parent filesystem's vfsp as the parent
1854 * filesystem and all of its snapshots have their vnode's
1855 * v_vfsp set to the parent's filesystem's vfsp. Note,
1856 * 'z_parent' is self referential for non-snapshots.
1858 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1859 #ifdef FREEBSD_NAMECACHE
1860 cache_purgevfs(zfsvfs->z_parent->z_vfs);
1865 * Close the zil. NB: Can't close the zil while zfs_inactive
1866 * threads are blocked as zil_close can call zfs_inactive.
1868 if (zfsvfs->z_log) {
1869 zil_close(zfsvfs->z_log);
1870 zfsvfs->z_log = NULL;
1873 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1876 * If we are not unmounting (ie: online recv) and someone already
1877 * unmounted this file system while we were doing the switcheroo,
1878 * or a reopen of z_os failed then just bail out now.
1880 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1881 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1882 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1883 return (SET_ERROR(EIO));
1887 * At this point there are no vops active, and any new vops will
1888 * fail with EIO since we have z_teardown_lock for writer (only
1889 * relavent for forced unmount).
1891 * Release all holds on dbufs.
1893 mutex_enter(&zfsvfs->z_znodes_lock);
1894 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1895 zp = list_next(&zfsvfs->z_all_znodes, zp))
1897 ASSERT(ZTOV(zp)->v_count >= 0);
1898 zfs_znode_dmu_fini(zp);
1900 mutex_exit(&zfsvfs->z_znodes_lock);
1903 * If we are unmounting, set the unmounted flag and let new vops
1904 * unblock. zfs_inactive will have the unmounted behavior, and all
1905 * other vops will fail with EIO.
1908 zfsvfs->z_unmounted = B_TRUE;
1909 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1910 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1914 * z_os will be NULL if there was an error in attempting to reopen
1915 * zfsvfs, so just return as the properties had already been
1916 * unregistered and cached data had been evicted before.
1918 if (zfsvfs->z_os == NULL)
1922 * Unregister properties.
1924 zfs_unregister_callbacks(zfsvfs);
1929 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1930 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1931 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1932 dmu_objset_evict_dbufs(zfsvfs->z_os);
1939 zfs_umount(vfs_t *vfsp, int fflag)
1941 kthread_t *td = curthread;
1942 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1944 cred_t *cr = td->td_ucred;
1947 ret = secpolicy_fs_unmount(cr, vfsp);
1949 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1950 ZFS_DELEG_PERM_MOUNT, cr))
1955 * We purge the parent filesystem's vfsp as the parent filesystem
1956 * and all of its snapshots have their vnode's v_vfsp set to the
1957 * parent's filesystem's vfsp. Note, 'z_parent' is self
1958 * referential for non-snapshots.
1960 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1963 * Unmount any snapshots mounted under .zfs before unmounting the
1966 if (zfsvfs->z_ctldir != NULL) {
1967 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
1969 ret = vflush(vfsp, 0, 0, td);
1970 ASSERT(ret == EBUSY);
1971 if (!(fflag & MS_FORCE)) {
1972 if (zfsvfs->z_ctldir->v_count > 1)
1974 ASSERT(zfsvfs->z_ctldir->v_count == 1);
1976 zfsctl_destroy(zfsvfs);
1977 ASSERT(zfsvfs->z_ctldir == NULL);
1980 if (fflag & MS_FORCE) {
1982 * Mark file system as unmounted before calling
1983 * vflush(FORCECLOSE). This way we ensure no future vnops
1984 * will be called and risk operating on DOOMED vnodes.
1986 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1987 zfsvfs->z_unmounted = B_TRUE;
1988 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1992 * Flush all the files.
1994 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
1996 if (!zfsvfs->z_issnap) {
1997 zfsctl_create(zfsvfs);
1998 ASSERT(zfsvfs->z_ctldir != NULL);
2004 if (!(fflag & MS_FORCE)) {
2006 * Check the number of active vnodes in the file system.
2007 * Our count is maintained in the vfs structure, but the
2008 * number is off by 1 to indicate a hold on the vfs
2011 * The '.zfs' directory maintains a reference of its
2012 * own, and any active references underneath are
2013 * reflected in the vnode count.
2015 if (zfsvfs->z_ctldir == NULL) {
2016 if (vfsp->vfs_count > 1)
2017 return (SET_ERROR(EBUSY));
2019 if (vfsp->vfs_count > 2 ||
2020 zfsvfs->z_ctldir->v_count > 1)
2021 return (SET_ERROR(EBUSY));
2026 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2030 * z_os will be NULL if there was an error in
2031 * attempting to reopen zfsvfs.
2035 * Unset the objset user_ptr.
2037 mutex_enter(&os->os_user_ptr_lock);
2038 dmu_objset_set_user(os, NULL);
2039 mutex_exit(&os->os_user_ptr_lock);
2042 * Finally release the objset
2044 dmu_objset_disown(os, zfsvfs);
2048 * We can now safely destroy the '.zfs' directory node.
2050 if (zfsvfs->z_ctldir != NULL)
2051 zfsctl_destroy(zfsvfs);
2052 if (zfsvfs->z_issnap) {
2053 vnode_t *svp = vfsp->mnt_vnodecovered;
2055 if (svp->v_count >= 2)
2064 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2066 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2071 * zfs_zget() can't operate on virtual entries like .zfs/ or
2072 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2073 * This will make NFS to switch to LOOKUP instead of using VGET.
2075 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2076 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2077 return (EOPNOTSUPP);
2080 err = zfs_zget(zfsvfs, ino, &zp);
2081 if (err == 0 && zp->z_unlinked) {
2089 err = vn_lock(*vpp, flags);
2096 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
2097 struct ucred **credanonp, int *numsecflavors, int **secflavors)
2099 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2102 * If this is regular file system vfsp is the same as
2103 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2104 * zfsvfs->z_parent->z_vfs represents parent file system
2105 * which we have to use here, because only this file system
2106 * has mnt_export configured.
2108 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2109 credanonp, numsecflavors, secflavors));
2112 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2113 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2116 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2118 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2120 uint64_t object = 0;
2121 uint64_t fid_gen = 0;
2131 * On FreeBSD we can get snapshot's mount point or its parent file
2132 * system mount point depending if snapshot is already mounted or not.
2134 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2135 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2136 uint64_t objsetid = 0;
2137 uint64_t setgen = 0;
2139 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2140 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2142 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2143 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2147 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2149 return (SET_ERROR(EINVAL));
2153 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2154 zfid_short_t *zfid = (zfid_short_t *)fidp;
2156 for (i = 0; i < sizeof (zfid->zf_object); i++)
2157 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2159 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2160 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2163 return (SET_ERROR(EINVAL));
2167 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2168 * directory tree. If the object == zfsvfs->z_shares_dir, then
2169 * we are in the .zfs/shares directory tree.
2171 if ((fid_gen == 0 &&
2172 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2173 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2174 *vpp = zfsvfs->z_ctldir;
2175 ASSERT(*vpp != NULL);
2176 if (object == ZFSCTL_INO_SNAPDIR) {
2177 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2178 0, NULL, NULL, NULL, NULL, NULL) == 0);
2179 } else if (object == zfsvfs->z_shares_dir) {
2180 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL,
2181 0, NULL, NULL, NULL, NULL, NULL) == 0);
2186 err = vn_lock(*vpp, flags);
2192 gen_mask = -1ULL >> (64 - 8 * i);
2194 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2195 if (err = zfs_zget(zfsvfs, object, &zp)) {
2199 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2201 zp_gen = zp_gen & gen_mask;
2204 if (zp->z_unlinked || zp_gen != fid_gen) {
2205 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2208 return (SET_ERROR(EINVAL));
2213 err = vn_lock(*vpp, flags | LK_RETRY);
2215 vnode_create_vobject(*vpp, zp->z_size, curthread);
2222 * Block out VOPs and close zfsvfs_t::z_os
2224 * Note, if successful, then we return with the 'z_teardown_lock' and
2225 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2226 * dataset and objset intact so that they can be atomically handed off during
2227 * a subsequent rollback or recv operation and the resume thereafter.
2230 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2234 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2241 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2242 * is an invariant across any of the operations that can be performed while the
2243 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2244 * are the same: the relevant objset and associated dataset are owned by
2245 * zfsvfs, held, and long held on entry.
2248 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2252 uint64_t sa_obj = 0;
2254 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2255 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2258 * We already own this, so just hold and rele it to update the
2259 * objset_t, as the one we had before may have been evicted.
2261 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os));
2262 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs);
2263 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset));
2264 dmu_objset_rele(zfsvfs->z_os, zfsvfs);
2267 * Make sure version hasn't changed
2270 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
2271 &zfsvfs->z_version);
2276 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2277 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2279 if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
2282 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2283 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
2286 if (zfsvfs->z_version >= ZPL_VERSION_SA)
2287 sa_register_update_callback(zfsvfs->z_os,
2290 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2292 zfs_set_fuid_feature(zfsvfs);
2295 * Attempt to re-establish all the active znodes with
2296 * their dbufs. If a zfs_rezget() fails, then we'll let
2297 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2298 * when they try to use their znode.
2300 mutex_enter(&zfsvfs->z_znodes_lock);
2301 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2302 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2303 (void) zfs_rezget(zp);
2305 mutex_exit(&zfsvfs->z_znodes_lock);
2308 /* release the VOPs */
2309 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2310 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2314 * Since we couldn't setup the sa framework, try to force
2315 * unmount this file system.
2317 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2318 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2324 zfs_freevfs(vfs_t *vfsp)
2326 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2330 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2331 * from zfs_mount(). Release it here. If we came through
2332 * zfs_mountroot() instead, we didn't grab an extra hold, so
2333 * skip the VFS_RELE for rootvfs.
2335 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2336 VFS_RELE(zfsvfs->z_parent->z_vfs);
2339 zfsvfs_free(zfsvfs);
2341 atomic_dec_32(&zfs_active_fs_count);
2345 static int desiredvnodes_backup;
2349 zfs_vnodes_adjust(void)
2352 int newdesiredvnodes;
2354 desiredvnodes_backup = desiredvnodes;
2357 * We calculate newdesiredvnodes the same way it is done in
2358 * vntblinit(). If it is equal to desiredvnodes, it means that
2359 * it wasn't tuned by the administrator and we can tune it down.
2361 newdesiredvnodes = min(maxproc + vm_cnt.v_page_count / 4, 2 *
2362 vm_kmem_size / (5 * (sizeof(struct vm_object) +
2363 sizeof(struct vnode))));
2364 if (newdesiredvnodes == desiredvnodes)
2365 desiredvnodes = (3 * newdesiredvnodes) / 4;
2370 zfs_vnodes_adjust_back(void)
2374 desiredvnodes = desiredvnodes_backup;
2382 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2385 * Initialize .zfs directory structures
2390 * Initialize znode cache, vnode ops, etc...
2395 * Reduce number of vnodes. Originally number of vnodes is calculated
2396 * with UFS inode in mind. We reduce it here, because it's too big for
2399 zfs_vnodes_adjust();
2401 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2409 zfs_vnodes_adjust_back();
2415 return (zfs_active_fs_count != 0);
2419 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2422 objset_t *os = zfsvfs->z_os;
2425 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2426 return (SET_ERROR(EINVAL));
2428 if (newvers < zfsvfs->z_version)
2429 return (SET_ERROR(EINVAL));
2431 if (zfs_spa_version_map(newvers) >
2432 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2433 return (SET_ERROR(ENOTSUP));
2435 tx = dmu_tx_create(os);
2436 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2437 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2438 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2440 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2442 error = dmu_tx_assign(tx, TXG_WAIT);
2448 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2449 8, 1, &newvers, tx);
2456 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2459 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2461 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2462 DMU_OT_NONE, 0, tx);
2464 error = zap_add(os, MASTER_NODE_OBJ,
2465 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2468 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2469 sa_register_update_callback(os, zfs_sa_upgrade);
2472 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2473 "from %llu to %llu", zfsvfs->z_version, newvers);
2477 zfsvfs->z_version = newvers;
2479 zfs_set_fuid_feature(zfsvfs);
2485 * Read a property stored within the master node.
2488 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2494 * Look up the file system's value for the property. For the
2495 * version property, we look up a slightly different string.
2497 if (prop == ZFS_PROP_VERSION)
2498 pname = ZPL_VERSION_STR;
2500 pname = zfs_prop_to_name(prop);
2503 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2505 if (error == ENOENT) {
2506 /* No value set, use the default value */
2508 case ZFS_PROP_VERSION:
2509 *value = ZPL_VERSION;
2511 case ZFS_PROP_NORMALIZE:
2512 case ZFS_PROP_UTF8ONLY:
2516 *value = ZFS_CASE_SENSITIVE;
2528 zfsvfs_update_fromname(const char *oldname, const char *newname)
2530 char tmpbuf[MAXPATHLEN];
2535 oldlen = strlen(oldname);
2537 mtx_lock(&mountlist_mtx);
2538 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2539 fromname = mp->mnt_stat.f_mntfromname;
2540 if (strcmp(fromname, oldname) == 0) {
2541 (void)strlcpy(fromname, newname,
2542 sizeof(mp->mnt_stat.f_mntfromname));
2545 if (strncmp(fromname, oldname, oldlen) == 0 &&
2546 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2547 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
2548 newname, fromname + oldlen);
2549 (void)strlcpy(fromname, tmpbuf,
2550 sizeof(mp->mnt_stat.f_mntfromname));
2554 mtx_unlock(&mountlist_mtx);