4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
24 * All rights reserved.
25 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
26 * Copyright (c) 2014 Integros [integros.com]
27 * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
30 /* Portions Copyright 2010 Robert Milkowski */
32 #include <sys/types.h>
33 #include <sys/param.h>
34 #include <sys/systm.h>
35 #include <sys/kernel.h>
36 #include <sys/sysmacros.h>
39 #include <sys/vnode.h>
41 #include <sys/mntent.h>
42 #include <sys/mount.h>
43 #include <sys/cmn_err.h>
44 #include <sys/zfs_znode.h>
45 #include <sys/zfs_dir.h>
47 #include <sys/fs/zfs.h>
49 #include <sys/dsl_prop.h>
50 #include <sys/dsl_dataset.h>
51 #include <sys/dsl_deleg.h>
55 #include <sys/sa_impl.h>
56 #include <sys/varargs.h>
57 #include <sys/policy.h>
58 #include <sys/atomic.h>
59 #include <sys/zfs_ioctl.h>
60 #include <sys/zfs_ctldir.h>
61 #include <sys/zfs_fuid.h>
62 #include <sys/sunddi.h>
64 #include <sys/dmu_objset.h>
65 #include <sys/spa_boot.h>
67 #include <ufs/ufs/quota.h>
69 #include "zfs_comutil.h"
71 struct mtx zfs_debug_mtx;
72 MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF);
74 SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system");
77 SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0,
78 "File system owner can perform privileged operation on his file systems");
81 SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RWTUN, &zfs_debug_level, 0,
84 SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions");
85 static int zfs_version_acl = ZFS_ACL_VERSION;
86 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0,
88 static int zfs_version_spa = SPA_VERSION;
89 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0,
91 static int zfs_version_zpl = ZPL_VERSION;
92 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0,
95 static int zfs_quotactl(vfs_t *vfsp, int cmds, uid_t id, void *arg);
96 static int zfs_mount(vfs_t *vfsp);
97 static int zfs_umount(vfs_t *vfsp, int fflag);
98 static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
99 static int zfs_statfs(vfs_t *vfsp, struct statfs *statp);
100 static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp);
101 static int zfs_sync(vfs_t *vfsp, int waitfor);
102 static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
103 struct ucred **credanonp, int *numsecflavors, int **secflavors);
104 static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp);
105 static void zfs_objset_close(zfsvfs_t *zfsvfs);
106 static void zfs_freevfs(vfs_t *vfsp);
108 struct vfsops zfs_vfsops = {
109 .vfs_mount = zfs_mount,
110 .vfs_unmount = zfs_umount,
111 .vfs_root = vfs_cache_root,
112 .vfs_cachedroot = zfs_root,
113 .vfs_statfs = zfs_statfs,
114 .vfs_vget = zfs_vget,
115 .vfs_sync = zfs_sync,
116 .vfs_checkexp = zfs_checkexp,
117 .vfs_fhtovp = zfs_fhtovp,
118 .vfs_quotactl = zfs_quotactl,
121 VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN);
124 * We need to keep a count of active fs's.
125 * This is necessary to prevent our module
126 * from being unloaded after a umount -f
128 static uint32_t zfs_active_fs_count = 0;
131 zfs_getquota(zfsvfs_t *zfsvfs, uid_t id, int isgroup, struct dqblk64 *dqp)
136 uint64_t usedobj, quotaobj;
137 uint64_t quota, used = 0;
140 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
141 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
143 if (quotaobj == 0 || zfsvfs->z_replay) {
147 (void)sprintf(buf, "%llx", (longlong_t)id);
148 if ((error = zap_lookup(zfsvfs->z_os, quotaobj,
149 buf, sizeof(quota), 1, "a)) != 0) {
150 dprintf("%s(%d): quotaobj lookup failed\n", __FUNCTION__, __LINE__);
154 * quota(8) uses bsoftlimit as "quoota", and hardlimit as "limit".
155 * So we set them to be the same.
157 dqp->dqb_bsoftlimit = dqp->dqb_bhardlimit = btodb(quota);
158 error = zap_lookup(zfsvfs->z_os, usedobj, buf, sizeof(used), 1, &used);
159 if (error && error != ENOENT) {
160 dprintf("%s(%d): usedobj failed; %d\n", __FUNCTION__, __LINE__, error);
163 dqp->dqb_curblocks = btodb(used);
164 dqp->dqb_ihardlimit = dqp->dqb_isoftlimit = 0;
167 * Setting this to 0 causes FreeBSD quota(8) to print
168 * the number of days since the epoch, which isn't
169 * particularly useful.
171 dqp->dqb_btime = dqp->dqb_itime = now.tv_sec;
177 zfs_quotactl(vfs_t *vfsp, int cmds, uid_t id, void *arg)
179 zfsvfs_t *zfsvfs = vfsp->vfs_data;
181 int cmd, type, error = 0;
184 zfs_userquota_prop_t quota_type;
185 struct dqblk64 dqblk = { 0 };
188 cmd = cmds >> SUBCMDSHIFT;
189 type = cmds & SUBCMDMASK;
195 id = td->td_ucred->cr_ruid;
198 id = td->td_ucred->cr_rgid;
202 if (cmd == Q_QUOTAON || cmd == Q_QUOTAOFF)
210 * ZFS_PROP_USERQUOTA,
211 * ZFS_PROP_GROUPUSED,
212 * ZFS_PROP_GROUPQUOTA
217 if (type == USRQUOTA)
218 quota_type = ZFS_PROP_USERQUOTA;
219 else if (type == GRPQUOTA)
220 quota_type = ZFS_PROP_GROUPQUOTA;
226 if (type == USRQUOTA)
227 quota_type = ZFS_PROP_USERUSED;
228 else if (type == GRPQUOTA)
229 quota_type = ZFS_PROP_GROUPUSED;
236 * Depending on the cmd, we may need to get
237 * the ruid and domain (see fuidstr_to_sid?),
238 * the fuid (how?), or other information.
239 * Create fuid using zfs_fuid_create(zfsvfs, id,
240 * ZFS_OWNER or ZFS_GROUP, cr, &fuidp)?
241 * I think I can use just the id?
243 * Look at zfs_fuid_overquota() to look up a quota.
244 * zap_lookup(something, quotaobj, fuidstring, sizeof(long long), 1, "a)
246 * See zfs_set_userquota() to set a quota.
248 if ((u_int)type >= MAXQUOTAS) {
256 error = copyout(&bitsize, arg, sizeof(int));
259 // As far as I can tell, you can't turn quotas on or off on zfs
268 error = copyin(&dqblk, arg, sizeof(dqblk));
270 error = zfs_set_userquota(zfsvfs, quota_type,
271 "", id, dbtob(dqblk.dqb_bhardlimit));
274 error = zfs_getquota(zfsvfs, id, type == GRPQUOTA, &dqblk);
276 error = copyout(&dqblk, arg, sizeof(dqblk));
289 zfs_sync(vfs_t *vfsp, int waitfor)
293 * Data integrity is job one. We don't want a compromised kernel
294 * writing to the storage pool, so we never sync during panic.
300 * Ignore the system syncher. ZFS already commits async data
301 * at zfs_txg_timeout intervals.
303 if (waitfor == MNT_LAZY)
308 * Sync a specific filesystem.
310 zfsvfs_t *zfsvfs = vfsp->vfs_data;
314 error = vfs_stdsync(vfsp, waitfor);
319 dp = dmu_objset_pool(zfsvfs->z_os);
322 * If the system is shutting down, then skip any
323 * filesystems which may exist on a suspended pool.
325 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
330 if (zfsvfs->z_log != NULL)
331 zil_commit(zfsvfs->z_log, 0);
336 * Sync all ZFS filesystems. This is what happens when you
337 * run sync(1M). Unlike other filesystems, ZFS honors the
338 * request by waiting for all pools to commit all dirty data.
346 #ifndef __FreeBSD_kernel__
348 zfs_create_unique_device(dev_t *dev)
353 ASSERT3U(zfs_minor, <=, MAXMIN32);
354 minor_t start = zfs_minor;
356 mutex_enter(&zfs_dev_mtx);
357 if (zfs_minor >= MAXMIN32) {
359 * If we're still using the real major
360 * keep out of /dev/zfs and /dev/zvol minor
361 * number space. If we're using a getudev()'ed
362 * major number, we can use all of its minors.
364 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
365 zfs_minor = ZFS_MIN_MINOR;
371 *dev = makedevice(zfs_major, zfs_minor);
372 mutex_exit(&zfs_dev_mtx);
373 } while (vfs_devismounted(*dev) && zfs_minor != start);
374 if (zfs_minor == start) {
376 * We are using all ~262,000 minor numbers for the
377 * current major number. Create a new major number.
379 if ((new_major = getudev()) == (major_t)-1) {
381 "zfs_mount: Can't get unique major "
385 mutex_enter(&zfs_dev_mtx);
386 zfs_major = new_major;
389 mutex_exit(&zfs_dev_mtx);
393 /* CONSTANTCONDITION */
398 #endif /* !__FreeBSD_kernel__ */
401 atime_changed_cb(void *arg, uint64_t newval)
403 zfsvfs_t *zfsvfs = arg;
405 if (newval == TRUE) {
406 zfsvfs->z_atime = TRUE;
407 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
408 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
409 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
411 zfsvfs->z_atime = FALSE;
412 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
413 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
414 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
419 xattr_changed_cb(void *arg, uint64_t newval)
421 zfsvfs_t *zfsvfs = arg;
423 if (newval == TRUE) {
424 /* XXX locking on vfs_flag? */
426 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
428 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
429 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
431 /* XXX locking on vfs_flag? */
433 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
435 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
436 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
441 blksz_changed_cb(void *arg, uint64_t newval)
443 zfsvfs_t *zfsvfs = arg;
444 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
445 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
446 ASSERT(ISP2(newval));
448 zfsvfs->z_max_blksz = newval;
449 zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
453 readonly_changed_cb(void *arg, uint64_t newval)
455 zfsvfs_t *zfsvfs = arg;
458 /* XXX locking on vfs_flag? */
459 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
460 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
461 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
463 /* XXX locking on vfs_flag? */
464 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
465 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
466 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
471 setuid_changed_cb(void *arg, uint64_t newval)
473 zfsvfs_t *zfsvfs = arg;
475 if (newval == FALSE) {
476 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
477 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
478 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
480 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
481 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
482 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
487 exec_changed_cb(void *arg, uint64_t newval)
489 zfsvfs_t *zfsvfs = arg;
491 if (newval == FALSE) {
492 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
493 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
494 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
496 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
497 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
498 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
503 * The nbmand mount option can be changed at mount time.
504 * We can't allow it to be toggled on live file systems or incorrect
505 * behavior may be seen from cifs clients
507 * This property isn't registered via dsl_prop_register(), but this callback
508 * will be called when a file system is first mounted
511 nbmand_changed_cb(void *arg, uint64_t newval)
513 zfsvfs_t *zfsvfs = arg;
514 if (newval == FALSE) {
515 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
516 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
518 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
519 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
524 snapdir_changed_cb(void *arg, uint64_t newval)
526 zfsvfs_t *zfsvfs = arg;
528 zfsvfs->z_show_ctldir = newval;
532 vscan_changed_cb(void *arg, uint64_t newval)
534 zfsvfs_t *zfsvfs = arg;
536 zfsvfs->z_vscan = newval;
540 acl_mode_changed_cb(void *arg, uint64_t newval)
542 zfsvfs_t *zfsvfs = arg;
544 zfsvfs->z_acl_mode = newval;
548 acl_inherit_changed_cb(void *arg, uint64_t newval)
550 zfsvfs_t *zfsvfs = arg;
552 zfsvfs->z_acl_inherit = newval;
556 zfs_register_callbacks(vfs_t *vfsp)
558 struct dsl_dataset *ds = NULL;
560 zfsvfs_t *zfsvfs = NULL;
562 boolean_t readonly = B_FALSE;
563 boolean_t do_readonly = B_FALSE;
564 boolean_t setuid = B_FALSE;
565 boolean_t do_setuid = B_FALSE;
566 boolean_t exec = B_FALSE;
567 boolean_t do_exec = B_FALSE;
569 boolean_t devices = B_FALSE;
570 boolean_t do_devices = B_FALSE;
572 boolean_t xattr = B_FALSE;
573 boolean_t do_xattr = B_FALSE;
574 boolean_t atime = B_FALSE;
575 boolean_t do_atime = B_FALSE;
579 zfsvfs = vfsp->vfs_data;
584 * This function can be called for a snapshot when we update snapshot's
585 * mount point, which isn't really supported.
587 if (dmu_objset_is_snapshot(os))
591 * The act of registering our callbacks will destroy any mount
592 * options we may have. In order to enable temporary overrides
593 * of mount options, we stash away the current values and
594 * restore them after we register the callbacks.
596 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
597 !spa_writeable(dmu_objset_spa(os))) {
599 do_readonly = B_TRUE;
600 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
602 do_readonly = B_TRUE;
604 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
608 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
611 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
616 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
619 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
623 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
626 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
630 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
633 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
639 * We need to enter pool configuration here, so that we can use
640 * dsl_prop_get_int_ds() to handle the special nbmand property below.
641 * dsl_prop_get_integer() can not be used, because it has to acquire
642 * spa_namespace_lock and we can not do that because we already hold
643 * z_teardown_lock. The problem is that spa_write_cachefile() is called
644 * with spa_namespace_lock held and the function calls ZFS vnode
645 * operations to write the cache file and thus z_teardown_lock is
646 * acquired after spa_namespace_lock.
648 ds = dmu_objset_ds(os);
649 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
652 * nbmand is a special property. It can only be changed at
655 * This is weird, but it is documented to only be changeable
658 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
660 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
662 } else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) {
663 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
668 * Register property callbacks.
670 * It would probably be fine to just check for i/o error from
671 * the first prop_register(), but I guess I like to go
674 error = dsl_prop_register(ds,
675 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
676 error = error ? error : dsl_prop_register(ds,
677 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
678 error = error ? error : dsl_prop_register(ds,
679 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
680 error = error ? error : dsl_prop_register(ds,
681 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
683 error = error ? error : dsl_prop_register(ds,
684 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
686 error = error ? error : dsl_prop_register(ds,
687 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
688 error = error ? error : dsl_prop_register(ds,
689 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
690 error = error ? error : dsl_prop_register(ds,
691 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
692 error = error ? error : dsl_prop_register(ds,
693 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
694 error = error ? error : dsl_prop_register(ds,
695 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
697 error = error ? error : dsl_prop_register(ds,
698 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
699 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
704 * Invoke our callbacks to restore temporary mount options.
707 readonly_changed_cb(zfsvfs, readonly);
709 setuid_changed_cb(zfsvfs, setuid);
711 exec_changed_cb(zfsvfs, exec);
713 xattr_changed_cb(zfsvfs, xattr);
715 atime_changed_cb(zfsvfs, atime);
717 nbmand_changed_cb(zfsvfs, nbmand);
722 dsl_prop_unregister_all(ds, zfsvfs);
727 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
728 uint64_t *userp, uint64_t *groupp)
731 * Is it a valid type of object to track?
733 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
734 return (SET_ERROR(ENOENT));
737 * If we have a NULL data pointer
738 * then assume the id's aren't changing and
739 * return EEXIST to the dmu to let it know to
743 return (SET_ERROR(EEXIST));
745 if (bonustype == DMU_OT_ZNODE) {
746 znode_phys_t *znp = data;
747 *userp = znp->zp_uid;
748 *groupp = znp->zp_gid;
751 sa_hdr_phys_t *sap = data;
752 sa_hdr_phys_t sa = *sap;
753 boolean_t swap = B_FALSE;
755 ASSERT(bonustype == DMU_OT_SA);
757 if (sa.sa_magic == 0) {
759 * This should only happen for newly created
760 * files that haven't had the znode data filled
767 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
768 sa.sa_magic = SA_MAGIC;
769 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
772 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
775 hdrsize = sa_hdrsize(&sa);
776 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
777 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
779 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
782 *userp = BSWAP_64(*userp);
783 *groupp = BSWAP_64(*groupp);
790 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
791 char *domainbuf, int buflen, uid_t *ridp)
796 fuid = zfs_strtonum(fuidstr, NULL);
798 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
800 (void) strlcpy(domainbuf, domain, buflen);
803 *ridp = FUID_RID(fuid);
807 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
810 case ZFS_PROP_USERUSED:
811 return (DMU_USERUSED_OBJECT);
812 case ZFS_PROP_GROUPUSED:
813 return (DMU_GROUPUSED_OBJECT);
814 case ZFS_PROP_USERQUOTA:
815 return (zfsvfs->z_userquota_obj);
816 case ZFS_PROP_GROUPQUOTA:
817 return (zfsvfs->z_groupquota_obj);
823 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
824 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
829 zfs_useracct_t *buf = vbuf;
832 if (!dmu_objset_userspace_present(zfsvfs->z_os))
833 return (SET_ERROR(ENOTSUP));
835 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
841 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
842 (error = zap_cursor_retrieve(&zc, &za)) == 0;
843 zap_cursor_advance(&zc)) {
844 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
848 fuidstr_to_sid(zfsvfs, za.za_name,
849 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
851 buf->zu_space = za.za_first_integer;
857 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
858 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
859 *cookiep = zap_cursor_serialize(&zc);
860 zap_cursor_fini(&zc);
865 * buf must be big enough (eg, 32 bytes)
868 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
869 char *buf, boolean_t addok)
874 if (domain && domain[0]) {
875 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
877 return (SET_ERROR(ENOENT));
879 fuid = FUID_ENCODE(domainid, rid);
880 (void) sprintf(buf, "%llx", (longlong_t)fuid);
885 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
886 const char *domain, uint64_t rid, uint64_t *valp)
894 if (!dmu_objset_userspace_present(zfsvfs->z_os))
895 return (SET_ERROR(ENOTSUP));
897 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
901 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
905 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
912 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
913 const char *domain, uint64_t rid, uint64_t quota)
919 boolean_t fuid_dirtied;
921 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
922 return (SET_ERROR(EINVAL));
924 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
925 return (SET_ERROR(ENOTSUP));
927 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
928 &zfsvfs->z_groupquota_obj;
930 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
933 fuid_dirtied = zfsvfs->z_fuid_dirty;
935 tx = dmu_tx_create(zfsvfs->z_os);
936 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
938 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
939 zfs_userquota_prop_prefixes[type]);
942 zfs_fuid_txhold(zfsvfs, tx);
943 err = dmu_tx_assign(tx, TXG_WAIT);
949 mutex_enter(&zfsvfs->z_lock);
951 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
953 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
954 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
956 mutex_exit(&zfsvfs->z_lock);
959 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
963 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
967 zfs_fuid_sync(zfsvfs, tx);
973 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
976 uint64_t used, quota, usedobj, quotaobj;
979 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
980 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
982 if (quotaobj == 0 || zfsvfs->z_replay)
985 (void) sprintf(buf, "%llx", (longlong_t)fuid);
986 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
990 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
993 return (used >= quota);
997 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
1002 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
1004 fuid = isgroup ? zp->z_gid : zp->z_uid;
1006 if (quotaobj == 0 || zfsvfs->z_replay)
1009 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
1013 * Associate this zfsvfs with the given objset, which must be owned.
1014 * This will cache a bunch of on-disk state from the objset in the
1018 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
1023 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
1024 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
1027 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
1030 if (zfsvfs->z_version >
1031 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
1032 (void) printf("Can't mount a version %lld file system "
1033 "on a version %lld pool\n. Pool must be upgraded to mount "
1034 "this file system.", (u_longlong_t)zfsvfs->z_version,
1035 (u_longlong_t)spa_version(dmu_objset_spa(os)));
1036 return (SET_ERROR(ENOTSUP));
1038 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
1041 zfsvfs->z_norm = (int)val;
1043 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
1046 zfsvfs->z_utf8 = (val != 0);
1048 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
1051 zfsvfs->z_case = (uint_t)val;
1054 * Fold case on file systems that are always or sometimes case
1057 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
1058 zfsvfs->z_case == ZFS_CASE_MIXED)
1059 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1061 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1062 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1064 uint64_t sa_obj = 0;
1065 if (zfsvfs->z_use_sa) {
1066 /* should either have both of these objects or none */
1067 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
1073 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1074 &zfsvfs->z_attr_table);
1078 if (zfsvfs->z_version >= ZPL_VERSION_SA)
1079 sa_register_update_callback(os, zfs_sa_upgrade);
1081 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
1085 ASSERT(zfsvfs->z_root != 0);
1087 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
1088 &zfsvfs->z_unlinkedobj);
1092 error = zap_lookup(os, MASTER_NODE_OBJ,
1093 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
1094 8, 1, &zfsvfs->z_userquota_obj);
1095 if (error == ENOENT)
1096 zfsvfs->z_userquota_obj = 0;
1097 else if (error != 0)
1100 error = zap_lookup(os, MASTER_NODE_OBJ,
1101 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
1102 8, 1, &zfsvfs->z_groupquota_obj);
1103 if (error == ENOENT)
1104 zfsvfs->z_groupquota_obj = 0;
1105 else if (error != 0)
1108 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
1109 &zfsvfs->z_fuid_obj);
1110 if (error == ENOENT)
1111 zfsvfs->z_fuid_obj = 0;
1112 else if (error != 0)
1115 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
1116 &zfsvfs->z_shares_dir);
1117 if (error == ENOENT)
1118 zfsvfs->z_shares_dir = 0;
1119 else if (error != 0)
1123 * Only use the name cache if we are looking for a
1124 * name on a file system that does not require normalization
1125 * or case folding. We can also look there if we happen to be
1126 * on a non-normalizing, mixed sensitivity file system IF we
1127 * are looking for the exact name (which is always the case on
1130 zfsvfs->z_use_namecache = !zfsvfs->z_norm ||
1131 ((zfsvfs->z_case == ZFS_CASE_MIXED) &&
1132 !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER));
1137 #if defined(__FreeBSD__)
1138 taskq_t *zfsvfs_taskq;
1141 zfsvfs_task_unlinked_drain(void *context, int pending __unused)
1144 zfs_unlinked_drain((zfsvfs_t *)context);
1149 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
1156 * XXX: Fix struct statfs so this isn't necessary!
1158 * The 'osname' is used as the filesystem's special node, which means
1159 * it must fit in statfs.f_mntfromname, or else it can't be
1160 * enumerated, so libzfs_mnttab_find() returns NULL, which causes
1161 * 'zfs unmount' to think it's not mounted when it is.
1163 if (strlen(osname) >= MNAMELEN)
1164 return (SET_ERROR(ENAMETOOLONG));
1166 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1169 * We claim to always be readonly so we can open snapshots;
1170 * other ZPL code will prevent us from writing to snapshots.
1173 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
1175 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1179 error = zfsvfs_create_impl(zfvp, zfsvfs, os);
1181 dmu_objset_disown(os, zfsvfs);
1188 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
1192 zfsvfs->z_vfs = NULL;
1193 zfsvfs->z_parent = zfsvfs;
1195 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1196 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1197 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1198 offsetof(znode_t, z_link_node));
1199 #if defined(__FreeBSD__)
1200 TASK_INIT(&zfsvfs->z_unlinked_drain_task, 0,
1201 zfsvfs_task_unlinked_drain, zfsvfs);
1204 rrm_init(&zfsvfs->z_teardown_lock, B_TRUE);
1206 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1208 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1209 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1210 for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1211 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1213 error = zfsvfs_init(zfsvfs, os);
1216 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1225 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1229 error = zfs_register_callbacks(zfsvfs->z_vfs);
1233 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1236 * If we are not mounting (ie: online recv), then we don't
1237 * have to worry about replaying the log as we blocked all
1238 * operations out since we closed the ZIL.
1244 * During replay we remove the read only flag to
1245 * allow replays to succeed.
1247 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1249 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1251 zfs_unlinked_drain(zfsvfs);
1254 * Parse and replay the intent log.
1256 * Because of ziltest, this must be done after
1257 * zfs_unlinked_drain(). (Further note: ziltest
1258 * doesn't use readonly mounts, where
1259 * zfs_unlinked_drain() isn't called.) This is because
1260 * ziltest causes spa_sync() to think it's committed,
1261 * but actually it is not, so the intent log contains
1262 * many txg's worth of changes.
1264 * In particular, if object N is in the unlinked set in
1265 * the last txg to actually sync, then it could be
1266 * actually freed in a later txg and then reallocated
1267 * in a yet later txg. This would write a "create
1268 * object N" record to the intent log. Normally, this
1269 * would be fine because the spa_sync() would have
1270 * written out the fact that object N is free, before
1271 * we could write the "create object N" intent log
1274 * But when we are in ziltest mode, we advance the "open
1275 * txg" without actually spa_sync()-ing the changes to
1276 * disk. So we would see that object N is still
1277 * allocated and in the unlinked set, and there is an
1278 * intent log record saying to allocate it.
1280 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1281 if (zil_replay_disable) {
1282 zil_destroy(zfsvfs->z_log, B_FALSE);
1284 zfsvfs->z_replay = B_TRUE;
1285 zil_replay(zfsvfs->z_os, zfsvfs,
1287 zfsvfs->z_replay = B_FALSE;
1290 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1294 * Set the objset user_ptr to track its zfsvfs.
1296 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1297 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1298 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1303 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1306 zfsvfs_free(zfsvfs_t *zfsvfs)
1311 * This is a barrier to prevent the filesystem from going away in
1312 * zfs_znode_move() until we can safely ensure that the filesystem is
1313 * not unmounted. We consider the filesystem valid before the barrier
1314 * and invalid after the barrier.
1316 rw_enter(&zfsvfs_lock, RW_READER);
1317 rw_exit(&zfsvfs_lock);
1319 zfs_fuid_destroy(zfsvfs);
1321 mutex_destroy(&zfsvfs->z_znodes_lock);
1322 mutex_destroy(&zfsvfs->z_lock);
1323 list_destroy(&zfsvfs->z_all_znodes);
1324 rrm_destroy(&zfsvfs->z_teardown_lock);
1325 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1326 rw_destroy(&zfsvfs->z_fuid_lock);
1327 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1328 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1329 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1333 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1335 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1336 if (zfsvfs->z_vfs) {
1337 if (zfsvfs->z_use_fuids) {
1338 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1339 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1340 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1341 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1342 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1343 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1345 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1346 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1347 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1348 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1349 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1350 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1353 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1357 zfs_domount(vfs_t *vfsp, char *osname)
1359 uint64_t recordsize, fsid_guid;
1367 error = zfsvfs_create(osname, &zfsvfs);
1370 zfsvfs->z_vfs = vfsp;
1373 /* Initialize the generic filesystem structure. */
1374 vfsp->vfs_bcount = 0;
1375 vfsp->vfs_data = NULL;
1377 if (zfs_create_unique_device(&mount_dev) == -1) {
1378 error = SET_ERROR(ENODEV);
1381 ASSERT(vfs_devismounted(mount_dev) == 0);
1384 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1387 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1388 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1390 vfsp->vfs_data = zfsvfs;
1391 vfsp->mnt_flag |= MNT_LOCAL;
1392 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1393 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1394 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1395 vfsp->mnt_kern_flag |= MNTK_NO_IOPF; /* vn_io_fault can be used */
1398 * The fsid is 64 bits, composed of an 8-bit fs type, which
1399 * separates our fsid from any other filesystem types, and a
1400 * 56-bit objset unique ID. The objset unique ID is unique to
1401 * all objsets open on this system, provided by unique_create().
1402 * The 8-bit fs type must be put in the low bits of fsid[1]
1403 * because that's where other Solaris filesystems put it.
1405 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1406 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1407 vfsp->vfs_fsid.val[0] = fsid_guid;
1408 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1409 vfsp->mnt_vfc->vfc_typenum & 0xFF;
1412 * Set features for file system.
1414 zfs_set_fuid_feature(zfsvfs);
1415 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1416 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1417 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1418 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1419 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1420 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1421 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1423 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1425 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1428 atime_changed_cb(zfsvfs, B_FALSE);
1429 readonly_changed_cb(zfsvfs, B_TRUE);
1430 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1432 xattr_changed_cb(zfsvfs, pval);
1433 zfsvfs->z_issnap = B_TRUE;
1434 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1436 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1437 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1438 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1440 error = zfsvfs_setup(zfsvfs, B_TRUE);
1443 vfs_mountedfrom(vfsp, osname);
1445 if (!zfsvfs->z_issnap)
1446 zfsctl_create(zfsvfs);
1449 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1450 zfsvfs_free(zfsvfs);
1452 atomic_inc_32(&zfs_active_fs_count);
1459 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1461 objset_t *os = zfsvfs->z_os;
1463 if (!dmu_objset_is_snapshot(os))
1464 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1469 * Convert a decimal digit string to a uint64_t integer.
1472 str_to_uint64(char *str, uint64_t *objnum)
1477 if (*str < '0' || *str > '9')
1478 return (SET_ERROR(EINVAL));
1480 num = num*10 + *str++ - '0';
1488 * The boot path passed from the boot loader is in the form of
1489 * "rootpool-name/root-filesystem-object-number'. Convert this
1490 * string to a dataset name: "rootpool-name/root-filesystem-name".
1493 zfs_parse_bootfs(char *bpath, char *outpath)
1499 if (*bpath == 0 || *bpath == '/')
1500 return (SET_ERROR(EINVAL));
1502 (void) strcpy(outpath, bpath);
1504 slashp = strchr(bpath, '/');
1506 /* if no '/', just return the pool name */
1507 if (slashp == NULL) {
1511 /* if not a number, just return the root dataset name */
1512 if (str_to_uint64(slashp+1, &objnum)) {
1517 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1524 * Check that the hex label string is appropriate for the dataset being
1525 * mounted into the global_zone proper.
1527 * Return an error if the hex label string is not default or
1528 * admin_low/admin_high. For admin_low labels, the corresponding
1529 * dataset must be readonly.
1532 zfs_check_global_label(const char *dsname, const char *hexsl)
1534 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1536 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1538 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1539 /* must be readonly */
1542 if (dsl_prop_get_integer(dsname,
1543 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1544 return (SET_ERROR(EACCES));
1545 return (rdonly ? 0 : EACCES);
1547 return (SET_ERROR(EACCES));
1551 * Determine whether the mount is allowed according to MAC check.
1552 * by comparing (where appropriate) label of the dataset against
1553 * the label of the zone being mounted into. If the dataset has
1554 * no label, create one.
1556 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1559 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1562 zone_t *mntzone = NULL;
1563 ts_label_t *mnt_tsl;
1566 char ds_hexsl[MAXNAMELEN];
1568 retv = EACCES; /* assume the worst */
1571 * Start by getting the dataset label if it exists.
1573 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1574 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1576 return (SET_ERROR(EACCES));
1579 * If labeling is NOT enabled, then disallow the mount of datasets
1580 * which have a non-default label already. No other label checks
1583 if (!is_system_labeled()) {
1584 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1586 return (SET_ERROR(EACCES));
1590 * Get the label of the mountpoint. If mounting into the global
1591 * zone (i.e. mountpoint is not within an active zone and the
1592 * zoned property is off), the label must be default or
1593 * admin_low/admin_high only; no other checks are needed.
1595 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1596 if (mntzone->zone_id == GLOBAL_ZONEID) {
1601 if (dsl_prop_get_integer(osname,
1602 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1603 return (SET_ERROR(EACCES));
1605 return (zfs_check_global_label(osname, ds_hexsl));
1608 * This is the case of a zone dataset being mounted
1609 * initially, before the zone has been fully created;
1610 * allow this mount into global zone.
1615 mnt_tsl = mntzone->zone_slabel;
1616 ASSERT(mnt_tsl != NULL);
1617 label_hold(mnt_tsl);
1618 mnt_sl = label2bslabel(mnt_tsl);
1620 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1622 * The dataset doesn't have a real label, so fabricate one.
1626 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1627 dsl_prop_set_string(osname,
1628 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1629 ZPROP_SRC_LOCAL, str) == 0)
1632 kmem_free(str, strlen(str) + 1);
1633 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1635 * Now compare labels to complete the MAC check. If the
1636 * labels are equal then allow access. If the mountpoint
1637 * label dominates the dataset label, allow readonly access.
1638 * Otherwise, access is denied.
1640 if (blequal(mnt_sl, &ds_sl))
1642 else if (bldominates(mnt_sl, &ds_sl)) {
1643 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1648 label_rele(mnt_tsl);
1652 #endif /* SECLABEL */
1654 #ifdef OPENSOLARIS_MOUNTROOT
1656 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1659 static int zfsrootdone = 0;
1660 zfsvfs_t *zfsvfs = NULL;
1669 * The filesystem that we mount as root is defined in the
1670 * boot property "zfs-bootfs" with a format of
1671 * "poolname/root-dataset-objnum".
1673 if (why == ROOT_INIT) {
1675 return (SET_ERROR(EBUSY));
1677 * the process of doing a spa_load will require the
1678 * clock to be set before we could (for example) do
1679 * something better by looking at the timestamp on
1680 * an uberblock, so just set it to -1.
1684 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1685 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1687 return (SET_ERROR(EINVAL));
1689 zfs_devid = spa_get_bootprop("diskdevid");
1690 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1692 spa_free_bootprop(zfs_devid);
1694 spa_free_bootprop(zfs_bootfs);
1695 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1699 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1700 spa_free_bootprop(zfs_bootfs);
1701 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1706 spa_free_bootprop(zfs_bootfs);
1708 if (error = vfs_lock(vfsp))
1711 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1712 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1716 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1718 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1719 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1724 mutex_enter(&vp->v_lock);
1725 vp->v_flag |= VROOT;
1726 mutex_exit(&vp->v_lock);
1730 * Leave rootvp held. The root file system is never unmounted.
1733 vfs_add((struct vnode *)0, vfsp,
1734 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1738 } else if (why == ROOT_REMOUNT) {
1739 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1740 vfsp->vfs_flag |= VFS_REMOUNT;
1742 /* refresh mount options */
1743 zfs_unregister_callbacks(vfsp->vfs_data);
1744 return (zfs_register_callbacks(vfsp));
1746 } else if (why == ROOT_UNMOUNT) {
1747 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1748 (void) zfs_sync(vfsp, 0, 0);
1753 * if "why" is equal to anything else other than ROOT_INIT,
1754 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1756 return (SET_ERROR(ENOTSUP));
1758 #endif /* OPENSOLARIS_MOUNTROOT */
1761 getpoolname(const char *osname, char *poolname)
1765 p = strchr(osname, '/');
1767 if (strlen(osname) >= MAXNAMELEN)
1768 return (ENAMETOOLONG);
1769 (void) strcpy(poolname, osname);
1771 if (p - osname >= MAXNAMELEN)
1772 return (ENAMETOOLONG);
1773 (void) strncpy(poolname, osname, p - osname);
1774 poolname[p - osname] = '\0';
1781 zfs_mount(vfs_t *vfsp)
1783 kthread_t *td = curthread;
1784 vnode_t *mvp = vfsp->mnt_vnodecovered;
1785 cred_t *cr = td->td_ucred;
1791 if (mvp->v_type != VDIR)
1792 return (SET_ERROR(ENOTDIR));
1794 mutex_enter(&mvp->v_lock);
1795 if ((uap->flags & MS_REMOUNT) == 0 &&
1796 (uap->flags & MS_OVERLAY) == 0 &&
1797 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1798 mutex_exit(&mvp->v_lock);
1799 return (SET_ERROR(EBUSY));
1801 mutex_exit(&mvp->v_lock);
1804 * ZFS does not support passing unparsed data in via MS_DATA.
1805 * Users should use the MS_OPTIONSTR interface; this means
1806 * that all option parsing is already done and the options struct
1807 * can be interrogated.
1809 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1810 return (SET_ERROR(EINVAL));
1813 * Get the objset name (the "special" mount argument).
1815 if (error = pn_get(uap->spec, fromspace, &spn))
1818 osname = spn.pn_path;
1819 #else /* !illumos */
1820 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1821 return (SET_ERROR(EINVAL));
1824 * If full-owner-access is enabled and delegated administration is
1825 * turned on, we must set nosuid.
1827 if (zfs_super_owner &&
1828 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1829 secpolicy_fs_mount_clearopts(cr, vfsp);
1831 #endif /* illumos */
1834 * Check for mount privilege?
1836 * If we don't have privilege then see if
1837 * we have local permission to allow it
1839 error = secpolicy_fs_mount(cr, mvp, vfsp);
1841 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1844 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1848 * Make sure user is the owner of the mount point
1849 * or has sufficient privileges.
1852 vattr.va_mask = AT_UID;
1854 vn_lock(mvp, LK_SHARED | LK_RETRY);
1855 if (VOP_GETATTR(mvp, &vattr, cr)) {
1860 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1861 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1868 secpolicy_fs_mount_clearopts(cr, vfsp);
1872 * Refuse to mount a filesystem if we are in a local zone and the
1873 * dataset is not visible.
1875 if (!INGLOBALZONE(curthread) &&
1876 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1877 error = SET_ERROR(EPERM);
1882 error = zfs_mount_label_policy(vfsp, osname);
1887 vfsp->vfs_flag |= MNT_NFS4ACLS;
1890 * When doing a remount, we simply refresh our temporary properties
1891 * according to those options set in the current VFS options.
1893 if (vfsp->vfs_flag & MS_REMOUNT) {
1894 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1897 * Refresh mount options with z_teardown_lock blocking I/O while
1898 * the filesystem is in an inconsistent state.
1899 * The lock also serializes this code with filesystem
1900 * manipulations between entry to zfs_suspend_fs() and return
1901 * from zfs_resume_fs().
1903 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1904 zfs_unregister_callbacks(zfsvfs);
1905 error = zfs_register_callbacks(vfsp);
1906 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1910 /* Initial root mount: try hard to import the requested root pool. */
1911 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1912 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1913 char pname[MAXNAMELEN];
1915 error = getpoolname(osname, pname);
1917 error = spa_import_rootpool(pname);
1922 error = zfs_domount(vfsp, osname);
1927 * Add an extra VFS_HOLD on our parent vfs so that it can't
1928 * disappear due to a forced unmount.
1930 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1931 VFS_HOLD(mvp->v_vfsp);
1939 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1941 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1942 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1944 statp->f_version = STATFS_VERSION;
1948 dmu_objset_space(zfsvfs->z_os,
1949 &refdbytes, &availbytes, &usedobjs, &availobjs);
1952 * The underlying storage pool actually uses multiple block sizes.
1953 * We report the fragsize as the smallest block size we support,
1954 * and we report our blocksize as the filesystem's maximum blocksize.
1956 statp->f_bsize = SPA_MINBLOCKSIZE;
1957 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1960 * The following report "total" blocks of various kinds in the
1961 * file system, but reported in terms of f_frsize - the
1965 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1966 statp->f_bfree = availbytes / statp->f_bsize;
1967 statp->f_bavail = statp->f_bfree; /* no root reservation */
1970 * statvfs() should really be called statufs(), because it assumes
1971 * static metadata. ZFS doesn't preallocate files, so the best
1972 * we can do is report the max that could possibly fit in f_files,
1973 * and that minus the number actually used in f_ffree.
1974 * For f_ffree, report the smaller of the number of object available
1975 * and the number of blocks (each object will take at least a block).
1977 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1978 statp->f_files = statp->f_ffree + usedobjs;
1981 * We're a zfs filesystem.
1983 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
1985 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
1986 sizeof(statp->f_mntfromname));
1987 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
1988 sizeof(statp->f_mntonname));
1990 statp->f_namemax = MAXNAMELEN - 1;
1997 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
1999 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2005 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
2007 *vpp = ZTOV(rootzp);
2012 error = vn_lock(*vpp, flags);
2022 * Teardown the zfsvfs::z_os.
2024 * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
2025 * and 'z_teardown_inactive_lock' held.
2028 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
2032 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
2036 * We purge the parent filesystem's vfsp as the parent
2037 * filesystem and all of its snapshots have their vnode's
2038 * v_vfsp set to the parent's filesystem's vfsp. Note,
2039 * 'z_parent' is self referential for non-snapshots.
2041 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2042 #ifdef FREEBSD_NAMECACHE
2043 cache_purgevfs(zfsvfs->z_parent->z_vfs, true);
2048 * Close the zil. NB: Can't close the zil while zfs_inactive
2049 * threads are blocked as zil_close can call zfs_inactive.
2051 if (zfsvfs->z_log) {
2052 zil_close(zfsvfs->z_log);
2053 zfsvfs->z_log = NULL;
2056 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
2059 * If we are not unmounting (ie: online recv) and someone already
2060 * unmounted this file system while we were doing the switcheroo,
2061 * or a reopen of z_os failed then just bail out now.
2063 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
2064 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2065 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2066 return (SET_ERROR(EIO));
2070 * At this point there are no vops active, and any new vops will
2071 * fail with EIO since we have z_teardown_lock for writer (only
2072 * relavent for forced unmount).
2074 * Release all holds on dbufs.
2076 mutex_enter(&zfsvfs->z_znodes_lock);
2077 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
2078 zp = list_next(&zfsvfs->z_all_znodes, zp))
2080 ASSERT(ZTOV(zp)->v_count >= 0);
2081 zfs_znode_dmu_fini(zp);
2083 mutex_exit(&zfsvfs->z_znodes_lock);
2086 * If we are unmounting, set the unmounted flag and let new vops
2087 * unblock. zfs_inactive will have the unmounted behavior, and all
2088 * other vops will fail with EIO.
2091 zfsvfs->z_unmounted = B_TRUE;
2092 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2093 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2097 * z_os will be NULL if there was an error in attempting to reopen
2098 * zfsvfs, so just return as the properties had already been
2099 * unregistered and cached data had been evicted before.
2101 if (zfsvfs->z_os == NULL)
2105 * Unregister properties.
2107 zfs_unregister_callbacks(zfsvfs);
2112 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
2113 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
2114 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
2115 dmu_objset_evict_dbufs(zfsvfs->z_os);
2122 zfs_umount(vfs_t *vfsp, int fflag)
2124 kthread_t *td = curthread;
2125 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2127 cred_t *cr = td->td_ucred;
2130 ret = secpolicy_fs_unmount(cr, vfsp);
2132 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
2133 ZFS_DELEG_PERM_MOUNT, cr))
2138 * We purge the parent filesystem's vfsp as the parent filesystem
2139 * and all of its snapshots have their vnode's v_vfsp set to the
2140 * parent's filesystem's vfsp. Note, 'z_parent' is self
2141 * referential for non-snapshots.
2143 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2146 * Unmount any snapshots mounted under .zfs before unmounting the
2149 if (zfsvfs->z_ctldir != NULL) {
2150 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
2154 if (fflag & MS_FORCE) {
2156 * Mark file system as unmounted before calling
2157 * vflush(FORCECLOSE). This way we ensure no future vnops
2158 * will be called and risk operating on DOOMED vnodes.
2160 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
2161 zfsvfs->z_unmounted = B_TRUE;
2162 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2166 * Flush all the files.
2168 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
2173 if (!(fflag & MS_FORCE)) {
2175 * Check the number of active vnodes in the file system.
2176 * Our count is maintained in the vfs structure, but the
2177 * number is off by 1 to indicate a hold on the vfs
2180 * The '.zfs' directory maintains a reference of its
2181 * own, and any active references underneath are
2182 * reflected in the vnode count.
2184 if (zfsvfs->z_ctldir == NULL) {
2185 if (vfsp->vfs_count > 1)
2186 return (SET_ERROR(EBUSY));
2188 if (vfsp->vfs_count > 2 ||
2189 zfsvfs->z_ctldir->v_count > 1)
2190 return (SET_ERROR(EBUSY));
2195 while (taskqueue_cancel(zfsvfs_taskq->tq_queue,
2196 &zfsvfs->z_unlinked_drain_task, NULL) != 0)
2197 taskqueue_drain(zfsvfs_taskq->tq_queue,
2198 &zfsvfs->z_unlinked_drain_task);
2200 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2204 * z_os will be NULL if there was an error in
2205 * attempting to reopen zfsvfs.
2209 * Unset the objset user_ptr.
2211 mutex_enter(&os->os_user_ptr_lock);
2212 dmu_objset_set_user(os, NULL);
2213 mutex_exit(&os->os_user_ptr_lock);
2216 * Finally release the objset
2218 dmu_objset_disown(os, zfsvfs);
2222 * We can now safely destroy the '.zfs' directory node.
2224 if (zfsvfs->z_ctldir != NULL)
2225 zfsctl_destroy(zfsvfs);
2232 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2234 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2239 * zfs_zget() can't operate on virtual entries like .zfs/ or
2240 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2241 * This will make NFS to switch to LOOKUP instead of using VGET.
2243 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2244 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2245 return (EOPNOTSUPP);
2248 err = zfs_zget(zfsvfs, ino, &zp);
2249 if (err == 0 && zp->z_unlinked) {
2257 err = vn_lock(*vpp, flags);
2267 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
2268 struct ucred **credanonp, int *numsecflavors, int **secflavors)
2270 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2273 * If this is regular file system vfsp is the same as
2274 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2275 * zfsvfs->z_parent->z_vfs represents parent file system
2276 * which we have to use here, because only this file system
2277 * has mnt_export configured.
2279 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2280 credanonp, numsecflavors, secflavors));
2283 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2284 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2287 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2289 struct componentname cn;
2290 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2293 uint64_t object = 0;
2294 uint64_t fid_gen = 0;
2304 * On FreeBSD we can get snapshot's mount point or its parent file
2305 * system mount point depending if snapshot is already mounted or not.
2307 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2308 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2309 uint64_t objsetid = 0;
2310 uint64_t setgen = 0;
2312 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2313 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2315 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2316 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2320 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2322 return (SET_ERROR(EINVAL));
2326 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2327 zfid_short_t *zfid = (zfid_short_t *)fidp;
2329 for (i = 0; i < sizeof (zfid->zf_object); i++)
2330 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2332 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2333 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2336 return (SET_ERROR(EINVAL));
2340 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2341 * directory tree. If the object == zfsvfs->z_shares_dir, then
2342 * we are in the .zfs/shares directory tree.
2344 if ((fid_gen == 0 &&
2345 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2346 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2348 VERIFY0(zfsctl_root(zfsvfs, LK_SHARED, &dvp));
2349 if (object == ZFSCTL_INO_SNAPDIR) {
2350 cn.cn_nameptr = "snapshot";
2351 cn.cn_namelen = strlen(cn.cn_nameptr);
2352 cn.cn_nameiop = LOOKUP;
2353 cn.cn_flags = ISLASTCN | LOCKLEAF;
2354 cn.cn_lkflags = flags;
2355 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2357 } else if (object == zfsvfs->z_shares_dir) {
2359 * XXX This branch must not be taken,
2360 * if it is, then the lookup below will
2363 cn.cn_nameptr = "shares";
2364 cn.cn_namelen = strlen(cn.cn_nameptr);
2365 cn.cn_nameiop = LOOKUP;
2366 cn.cn_flags = ISLASTCN;
2367 cn.cn_lkflags = flags;
2368 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2376 gen_mask = -1ULL >> (64 - 8 * i);
2378 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2379 if (err = zfs_zget(zfsvfs, object, &zp)) {
2383 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2385 zp_gen = zp_gen & gen_mask;
2388 if (zp->z_unlinked || zp_gen != fid_gen) {
2389 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2392 return (SET_ERROR(EINVAL));
2397 err = vn_lock(*vpp, flags);
2399 vnode_create_vobject(*vpp, zp->z_size, curthread);
2406 * Block out VOPs and close zfsvfs_t::z_os
2408 * Note, if successful, then we return with the 'z_teardown_lock' and
2409 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2410 * dataset and objset intact so that they can be atomically handed off during
2411 * a subsequent rollback or recv operation and the resume thereafter.
2414 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2418 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2425 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2426 * is an invariant across any of the operations that can be performed while the
2427 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2428 * are the same: the relevant objset and associated dataset are owned by
2429 * zfsvfs, held, and long held on entry.
2432 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2437 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2438 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2441 * We already own this, so just update the objset_t, as the one we
2442 * had before may have been evicted.
2445 VERIFY3P(ds->ds_owner, ==, zfsvfs);
2446 VERIFY(dsl_dataset_long_held(ds));
2447 VERIFY0(dmu_objset_from_ds(ds, &os));
2449 err = zfsvfs_init(zfsvfs, os);
2453 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2455 zfs_set_fuid_feature(zfsvfs);
2458 * Attempt to re-establish all the active znodes with
2459 * their dbufs. If a zfs_rezget() fails, then we'll let
2460 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2461 * when they try to use their znode.
2463 mutex_enter(&zfsvfs->z_znodes_lock);
2464 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2465 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2466 (void) zfs_rezget(zp);
2468 mutex_exit(&zfsvfs->z_znodes_lock);
2471 /* release the VOPs */
2472 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2473 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2477 * Since we couldn't setup the sa framework, try to force
2478 * unmount this file system.
2480 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) {
2481 vfs_ref(zfsvfs->z_vfs);
2482 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2489 zfs_freevfs(vfs_t *vfsp)
2491 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2495 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2496 * from zfs_mount(). Release it here. If we came through
2497 * zfs_mountroot() instead, we didn't grab an extra hold, so
2498 * skip the VFS_RELE for rootvfs.
2500 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2501 VFS_RELE(zfsvfs->z_parent->z_vfs);
2504 zfsvfs_free(zfsvfs);
2506 atomic_dec_32(&zfs_active_fs_count);
2510 static int desiredvnodes_backup;
2514 zfs_vnodes_adjust(void)
2517 int newdesiredvnodes;
2519 desiredvnodes_backup = desiredvnodes;
2522 * We calculate newdesiredvnodes the same way it is done in
2523 * vntblinit(). If it is equal to desiredvnodes, it means that
2524 * it wasn't tuned by the administrator and we can tune it down.
2526 newdesiredvnodes = min(maxproc + vm_cnt.v_page_count / 4, 2 *
2527 vm_kmem_size / (5 * (sizeof(struct vm_object) +
2528 sizeof(struct vnode))));
2529 if (newdesiredvnodes == desiredvnodes)
2530 desiredvnodes = (3 * newdesiredvnodes) / 4;
2535 zfs_vnodes_adjust_back(void)
2539 desiredvnodes = desiredvnodes_backup;
2547 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2550 * Initialize .zfs directory structures
2555 * Initialize znode cache, vnode ops, etc...
2560 * Reduce number of vnodes. Originally number of vnodes is calculated
2561 * with UFS inode in mind. We reduce it here, because it's too big for
2564 zfs_vnodes_adjust();
2566 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2567 #if defined(__FreeBSD__)
2568 zfsvfs_taskq = taskq_create("zfsvfs", 1, minclsyspri, 0, 0, 0);
2575 #if defined(__FreeBSD__)
2576 taskq_destroy(zfsvfs_taskq);
2580 zfs_vnodes_adjust_back();
2586 return (zfs_active_fs_count != 0);
2590 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2593 objset_t *os = zfsvfs->z_os;
2596 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2597 return (SET_ERROR(EINVAL));
2599 if (newvers < zfsvfs->z_version)
2600 return (SET_ERROR(EINVAL));
2602 if (zfs_spa_version_map(newvers) >
2603 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2604 return (SET_ERROR(ENOTSUP));
2606 tx = dmu_tx_create(os);
2607 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2608 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2609 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2611 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2613 error = dmu_tx_assign(tx, TXG_WAIT);
2619 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2620 8, 1, &newvers, tx);
2627 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2630 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2632 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2633 DMU_OT_NONE, 0, tx);
2635 error = zap_add(os, MASTER_NODE_OBJ,
2636 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2639 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2640 sa_register_update_callback(os, zfs_sa_upgrade);
2643 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2644 "from %llu to %llu", zfsvfs->z_version, newvers);
2648 zfsvfs->z_version = newvers;
2649 os->os_version = newvers;
2651 zfs_set_fuid_feature(zfsvfs);
2657 * Read a property stored within the master node.
2660 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2662 uint64_t *cached_copy = NULL;
2665 * Figure out where in the objset_t the cached copy would live, if it
2666 * is available for the requested property.
2670 case ZFS_PROP_VERSION:
2671 cached_copy = &os->os_version;
2673 case ZFS_PROP_NORMALIZE:
2674 cached_copy = &os->os_normalization;
2676 case ZFS_PROP_UTF8ONLY:
2677 cached_copy = &os->os_utf8only;
2680 cached_copy = &os->os_casesensitivity;
2686 if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) {
2687 *value = *cached_copy;
2692 * If the property wasn't cached, look up the file system's value for
2693 * the property. For the version property, we look up a slightly
2698 if (prop == ZFS_PROP_VERSION) {
2699 pname = ZPL_VERSION_STR;
2701 pname = zfs_prop_to_name(prop);
2705 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2706 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2709 if (error == ENOENT) {
2710 /* No value set, use the default value */
2712 case ZFS_PROP_VERSION:
2713 *value = ZPL_VERSION;
2715 case ZFS_PROP_NORMALIZE:
2716 case ZFS_PROP_UTF8ONLY:
2720 *value = ZFS_CASE_SENSITIVE;
2729 * If one of the methods for getting the property value above worked,
2730 * copy it into the objset_t's cache.
2732 if (error == 0 && cached_copy != NULL) {
2733 *cached_copy = *value;
2740 * Return true if the coresponding vfs's unmounted flag is set.
2741 * Otherwise return false.
2742 * If this function returns true we know VFS unmount has been initiated.
2745 zfs_get_vfs_flag_unmounted(objset_t *os)
2748 boolean_t unmounted = B_FALSE;
2750 ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2752 mutex_enter(&os->os_user_ptr_lock);
2753 zfvp = dmu_objset_get_user(os);
2754 if (zfvp != NULL && zfvp->z_vfs != NULL &&
2755 (zfvp->z_vfs->mnt_kern_flag & MNTK_UNMOUNT))
2757 mutex_exit(&os->os_user_ptr_lock);
2764 zfsvfs_update_fromname(const char *oldname, const char *newname)
2766 char tmpbuf[MAXPATHLEN];
2771 oldlen = strlen(oldname);
2773 mtx_lock(&mountlist_mtx);
2774 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2775 fromname = mp->mnt_stat.f_mntfromname;
2776 if (strcmp(fromname, oldname) == 0) {
2777 (void)strlcpy(fromname, newname,
2778 sizeof(mp->mnt_stat.f_mntfromname));
2781 if (strncmp(fromname, oldname, oldlen) == 0 &&
2782 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2783 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
2784 newname, fromname + oldlen);
2785 (void)strlcpy(fromname, tmpbuf,
2786 sizeof(mp->mnt_stat.f_mntfromname));
2790 mtx_unlock(&mountlist_mtx);