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 | CTLFLAG_MPSAFE, 0,
78 SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0,
79 "File system owner can perform privileged operation on his file systems");
82 SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RWTUN, &zfs_debug_level, 0,
85 SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD | CTLFLAG_MPSAFE, 0,
87 static int zfs_version_acl = ZFS_ACL_VERSION;
88 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0,
90 static int zfs_version_spa = SPA_VERSION;
91 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0,
93 static int zfs_version_zpl = ZPL_VERSION;
94 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0,
97 static int zfs_quotactl(vfs_t *vfsp, int cmds, uid_t id, void *arg);
98 static int zfs_mount(vfs_t *vfsp);
99 static int zfs_umount(vfs_t *vfsp, int fflag);
100 static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
101 static int zfs_statfs(vfs_t *vfsp, struct statfs *statp);
102 static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp);
103 static int zfs_sync(vfs_t *vfsp, int waitfor);
104 static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, uint64_t *extflagsp,
105 struct ucred **credanonp, int *numsecflavors, int *secflavors);
106 static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp);
107 static void zfs_objset_close(zfsvfs_t *zfsvfs);
108 static void zfs_freevfs(vfs_t *vfsp);
110 struct vfsops zfs_vfsops = {
111 .vfs_mount = zfs_mount,
112 .vfs_unmount = zfs_umount,
113 .vfs_root = vfs_cache_root,
114 .vfs_cachedroot = zfs_root,
115 .vfs_statfs = zfs_statfs,
116 .vfs_vget = zfs_vget,
117 .vfs_sync = zfs_sync,
118 .vfs_checkexp = zfs_checkexp,
119 .vfs_fhtovp = zfs_fhtovp,
120 .vfs_quotactl = zfs_quotactl,
123 VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN);
126 * We need to keep a count of active fs's.
127 * This is necessary to prevent our module
128 * from being unloaded after a umount -f
130 static uint32_t zfs_active_fs_count = 0;
133 zfs_getquota(zfsvfs_t *zfsvfs, uid_t id, int isgroup, struct dqblk64 *dqp)
138 uint64_t usedobj, quotaobj;
139 uint64_t quota, used = 0;
142 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
143 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
145 if (quotaobj == 0 || zfsvfs->z_replay) {
149 (void)sprintf(buf, "%llx", (longlong_t)id);
150 if ((error = zap_lookup(zfsvfs->z_os, quotaobj,
151 buf, sizeof(quota), 1, "a)) != 0) {
152 dprintf("%s(%d): quotaobj lookup failed\n", __FUNCTION__, __LINE__);
156 * quota(8) uses bsoftlimit as "quoota", and hardlimit as "limit".
157 * So we set them to be the same.
159 dqp->dqb_bsoftlimit = dqp->dqb_bhardlimit = btodb(quota);
160 error = zap_lookup(zfsvfs->z_os, usedobj, buf, sizeof(used), 1, &used);
161 if (error && error != ENOENT) {
162 dprintf("%s(%d): usedobj failed; %d\n", __FUNCTION__, __LINE__, error);
165 dqp->dqb_curblocks = btodb(used);
166 dqp->dqb_ihardlimit = dqp->dqb_isoftlimit = 0;
169 * Setting this to 0 causes FreeBSD quota(8) to print
170 * the number of days since the epoch, which isn't
171 * particularly useful.
173 dqp->dqb_btime = dqp->dqb_itime = now.tv_sec;
179 zfs_quotactl(vfs_t *vfsp, int cmds, uid_t id, void *arg)
181 zfsvfs_t *zfsvfs = vfsp->vfs_data;
183 int cmd, type, error = 0;
186 zfs_userquota_prop_t quota_type;
187 struct dqblk64 dqblk = { 0 };
190 cmd = cmds >> SUBCMDSHIFT;
191 type = cmds & SUBCMDMASK;
197 id = td->td_ucred->cr_ruid;
200 id = td->td_ucred->cr_rgid;
204 if (cmd == Q_QUOTAON || cmd == Q_QUOTAOFF)
212 * ZFS_PROP_USERQUOTA,
213 * ZFS_PROP_GROUPUSED,
214 * ZFS_PROP_GROUPQUOTA
219 if (type == USRQUOTA)
220 quota_type = ZFS_PROP_USERQUOTA;
221 else if (type == GRPQUOTA)
222 quota_type = ZFS_PROP_GROUPQUOTA;
228 if (type == USRQUOTA)
229 quota_type = ZFS_PROP_USERUSED;
230 else if (type == GRPQUOTA)
231 quota_type = ZFS_PROP_GROUPUSED;
238 * Depending on the cmd, we may need to get
239 * the ruid and domain (see fuidstr_to_sid?),
240 * the fuid (how?), or other information.
241 * Create fuid using zfs_fuid_create(zfsvfs, id,
242 * ZFS_OWNER or ZFS_GROUP, cr, &fuidp)?
243 * I think I can use just the id?
245 * Look at zfs_fuid_overquota() to look up a quota.
246 * zap_lookup(something, quotaobj, fuidstring, sizeof(long long), 1, "a)
248 * See zfs_set_userquota() to set a quota.
250 if ((u_int)type >= MAXQUOTAS) {
258 error = copyout(&bitsize, arg, sizeof(int));
261 // As far as I can tell, you can't turn quotas on or off on zfs
270 error = copyin(arg, &dqblk, sizeof(dqblk));
272 error = zfs_set_userquota(zfsvfs, quota_type,
273 "", id, dbtob(dqblk.dqb_bhardlimit));
276 error = zfs_getquota(zfsvfs, id, type == GRPQUOTA, &dqblk);
278 error = copyout(&dqblk, arg, sizeof(dqblk));
291 zfs_sync(vfs_t *vfsp, int waitfor)
295 * Data integrity is job one. We don't want a compromised kernel
296 * writing to the storage pool, so we never sync during panic.
298 if (KERNEL_PANICKED())
302 * Ignore the system syncher. ZFS already commits async data
303 * at zfs_txg_timeout intervals.
305 if (waitfor == MNT_LAZY)
310 * Sync a specific filesystem.
312 zfsvfs_t *zfsvfs = vfsp->vfs_data;
316 error = vfs_stdsync(vfsp, waitfor);
321 dp = dmu_objset_pool(zfsvfs->z_os);
324 * If the system is shutting down, then skip any
325 * filesystems which may exist on a suspended pool.
327 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
332 if (zfsvfs->z_log != NULL)
333 zil_commit(zfsvfs->z_log, 0);
338 * Sync all ZFS filesystems. This is what happens when you
339 * run sync(1M). Unlike other filesystems, ZFS honors the
340 * request by waiting for all pools to commit all dirty data.
348 #ifndef __FreeBSD_kernel__
350 zfs_create_unique_device(dev_t *dev)
355 ASSERT3U(zfs_minor, <=, MAXMIN32);
356 minor_t start = zfs_minor;
358 mutex_enter(&zfs_dev_mtx);
359 if (zfs_minor >= MAXMIN32) {
361 * If we're still using the real major
362 * keep out of /dev/zfs and /dev/zvol minor
363 * number space. If we're using a getudev()'ed
364 * major number, we can use all of its minors.
366 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
367 zfs_minor = ZFS_MIN_MINOR;
373 *dev = makedevice(zfs_major, zfs_minor);
374 mutex_exit(&zfs_dev_mtx);
375 } while (vfs_devismounted(*dev) && zfs_minor != start);
376 if (zfs_minor == start) {
378 * We are using all ~262,000 minor numbers for the
379 * current major number. Create a new major number.
381 if ((new_major = getudev()) == (major_t)-1) {
383 "zfs_mount: Can't get unique major "
387 mutex_enter(&zfs_dev_mtx);
388 zfs_major = new_major;
391 mutex_exit(&zfs_dev_mtx);
395 /* CONSTANTCONDITION */
400 #endif /* !__FreeBSD_kernel__ */
403 atime_changed_cb(void *arg, uint64_t newval)
405 zfsvfs_t *zfsvfs = arg;
407 if (newval == TRUE) {
408 zfsvfs->z_atime = TRUE;
409 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
410 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
411 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
413 zfsvfs->z_atime = FALSE;
414 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
415 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
416 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
421 xattr_changed_cb(void *arg, uint64_t newval)
423 zfsvfs_t *zfsvfs = arg;
425 if (newval == TRUE) {
426 /* XXX locking on vfs_flag? */
428 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
430 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
431 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
433 /* XXX locking on vfs_flag? */
435 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
437 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
438 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
443 blksz_changed_cb(void *arg, uint64_t newval)
445 zfsvfs_t *zfsvfs = arg;
446 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
447 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
448 ASSERT(ISP2(newval));
450 zfsvfs->z_max_blksz = newval;
451 zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
455 readonly_changed_cb(void *arg, uint64_t newval)
457 zfsvfs_t *zfsvfs = arg;
460 /* XXX locking on vfs_flag? */
461 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
462 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
463 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
465 /* XXX locking on vfs_flag? */
466 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
467 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
468 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
473 setuid_changed_cb(void *arg, uint64_t newval)
475 zfsvfs_t *zfsvfs = arg;
477 if (newval == FALSE) {
478 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
479 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
480 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
482 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
483 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
484 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
489 exec_changed_cb(void *arg, uint64_t newval)
491 zfsvfs_t *zfsvfs = arg;
493 if (newval == FALSE) {
494 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
495 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
496 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
498 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
499 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
500 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
505 * The nbmand mount option can be changed at mount time.
506 * We can't allow it to be toggled on live file systems or incorrect
507 * behavior may be seen from cifs clients
509 * This property isn't registered via dsl_prop_register(), but this callback
510 * will be called when a file system is first mounted
513 nbmand_changed_cb(void *arg, uint64_t newval)
515 zfsvfs_t *zfsvfs = arg;
516 if (newval == FALSE) {
517 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
518 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
520 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
521 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
526 snapdir_changed_cb(void *arg, uint64_t newval)
528 zfsvfs_t *zfsvfs = arg;
530 zfsvfs->z_show_ctldir = newval;
534 vscan_changed_cb(void *arg, uint64_t newval)
536 zfsvfs_t *zfsvfs = arg;
538 zfsvfs->z_vscan = newval;
542 acl_mode_changed_cb(void *arg, uint64_t newval)
544 zfsvfs_t *zfsvfs = arg;
546 zfsvfs->z_acl_mode = newval;
550 acl_inherit_changed_cb(void *arg, uint64_t newval)
552 zfsvfs_t *zfsvfs = arg;
554 zfsvfs->z_acl_inherit = newval;
558 zfs_register_callbacks(vfs_t *vfsp)
560 struct dsl_dataset *ds = NULL;
562 zfsvfs_t *zfsvfs = NULL;
564 boolean_t readonly = B_FALSE;
565 boolean_t do_readonly = B_FALSE;
566 boolean_t setuid = B_FALSE;
567 boolean_t do_setuid = B_FALSE;
568 boolean_t exec = B_FALSE;
569 boolean_t do_exec = B_FALSE;
571 boolean_t devices = B_FALSE;
572 boolean_t do_devices = B_FALSE;
574 boolean_t xattr = B_FALSE;
575 boolean_t do_xattr = B_FALSE;
576 boolean_t atime = B_FALSE;
577 boolean_t do_atime = B_FALSE;
581 zfsvfs = vfsp->vfs_data;
586 * This function can be called for a snapshot when we update snapshot's
587 * mount point, which isn't really supported.
589 if (dmu_objset_is_snapshot(os))
593 * The act of registering our callbacks will destroy any mount
594 * options we may have. In order to enable temporary overrides
595 * of mount options, we stash away the current values and
596 * restore them after we register the callbacks.
598 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
599 !spa_writeable(dmu_objset_spa(os))) {
601 do_readonly = B_TRUE;
602 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
604 do_readonly = B_TRUE;
606 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
610 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
613 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
618 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
621 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
625 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
628 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
632 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
635 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
641 * We need to enter pool configuration here, so that we can use
642 * dsl_prop_get_int_ds() to handle the special nbmand property below.
643 * dsl_prop_get_integer() can not be used, because it has to acquire
644 * spa_namespace_lock and we can not do that because we already hold
645 * z_teardown_lock. The problem is that spa_write_cachefile() is called
646 * with spa_namespace_lock held and the function calls ZFS vnode
647 * operations to write the cache file and thus z_teardown_lock is
648 * acquired after spa_namespace_lock.
650 ds = dmu_objset_ds(os);
651 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
654 * nbmand is a special property. It can only be changed at
657 * This is weird, but it is documented to only be changeable
660 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
662 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
664 } else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) {
665 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
670 * Register property callbacks.
672 * It would probably be fine to just check for i/o error from
673 * the first prop_register(), but I guess I like to go
676 error = dsl_prop_register(ds,
677 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
678 error = error ? error : dsl_prop_register(ds,
679 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
680 error = error ? error : dsl_prop_register(ds,
681 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
682 error = error ? error : dsl_prop_register(ds,
683 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
685 error = error ? error : dsl_prop_register(ds,
686 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
688 error = error ? error : dsl_prop_register(ds,
689 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
690 error = error ? error : dsl_prop_register(ds,
691 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
692 error = error ? error : dsl_prop_register(ds,
693 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
694 error = error ? error : dsl_prop_register(ds,
695 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
696 error = error ? error : dsl_prop_register(ds,
697 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
699 error = error ? error : dsl_prop_register(ds,
700 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
701 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
706 * Invoke our callbacks to restore temporary mount options.
709 readonly_changed_cb(zfsvfs, readonly);
711 setuid_changed_cb(zfsvfs, setuid);
713 exec_changed_cb(zfsvfs, exec);
715 xattr_changed_cb(zfsvfs, xattr);
717 atime_changed_cb(zfsvfs, atime);
719 nbmand_changed_cb(zfsvfs, nbmand);
724 dsl_prop_unregister_all(ds, zfsvfs);
729 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
730 uint64_t *userp, uint64_t *groupp)
733 * Is it a valid type of object to track?
735 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
736 return (SET_ERROR(ENOENT));
739 * If we have a NULL data pointer
740 * then assume the id's aren't changing and
741 * return EEXIST to the dmu to let it know to
745 return (SET_ERROR(EEXIST));
747 if (bonustype == DMU_OT_ZNODE) {
748 znode_phys_t *znp = data;
749 *userp = znp->zp_uid;
750 *groupp = znp->zp_gid;
753 sa_hdr_phys_t *sap = data;
754 sa_hdr_phys_t sa = *sap;
755 boolean_t swap = B_FALSE;
757 ASSERT(bonustype == DMU_OT_SA);
759 if (sa.sa_magic == 0) {
761 * This should only happen for newly created
762 * files that haven't had the znode data filled
769 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
770 sa.sa_magic = SA_MAGIC;
771 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
774 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
777 hdrsize = sa_hdrsize(&sa);
778 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
779 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
781 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
784 *userp = BSWAP_64(*userp);
785 *groupp = BSWAP_64(*groupp);
792 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
793 char *domainbuf, int buflen, uid_t *ridp)
798 fuid = zfs_strtonum(fuidstr, NULL);
800 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
802 (void) strlcpy(domainbuf, domain, buflen);
805 *ridp = FUID_RID(fuid);
809 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
812 case ZFS_PROP_USERUSED:
813 return (DMU_USERUSED_OBJECT);
814 case ZFS_PROP_GROUPUSED:
815 return (DMU_GROUPUSED_OBJECT);
816 case ZFS_PROP_USERQUOTA:
817 return (zfsvfs->z_userquota_obj);
818 case ZFS_PROP_GROUPQUOTA:
819 return (zfsvfs->z_groupquota_obj);
825 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
826 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
831 zfs_useracct_t *buf = vbuf;
834 if (!dmu_objset_userspace_present(zfsvfs->z_os))
835 return (SET_ERROR(ENOTSUP));
837 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
843 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
844 (error = zap_cursor_retrieve(&zc, &za)) == 0;
845 zap_cursor_advance(&zc)) {
846 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
850 fuidstr_to_sid(zfsvfs, za.za_name,
851 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
853 buf->zu_space = za.za_first_integer;
859 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
860 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
861 *cookiep = zap_cursor_serialize(&zc);
862 zap_cursor_fini(&zc);
867 * buf must be big enough (eg, 32 bytes)
870 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
871 char *buf, boolean_t addok)
876 if (domain && domain[0]) {
877 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
879 return (SET_ERROR(ENOENT));
881 fuid = FUID_ENCODE(domainid, rid);
882 (void) sprintf(buf, "%llx", (longlong_t)fuid);
887 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
888 const char *domain, uint64_t rid, uint64_t *valp)
896 if (!dmu_objset_userspace_present(zfsvfs->z_os))
897 return (SET_ERROR(ENOTSUP));
899 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
903 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
907 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
914 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
915 const char *domain, uint64_t rid, uint64_t quota)
921 boolean_t fuid_dirtied;
923 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
924 return (SET_ERROR(EINVAL));
926 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
927 return (SET_ERROR(ENOTSUP));
929 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
930 &zfsvfs->z_groupquota_obj;
932 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
935 fuid_dirtied = zfsvfs->z_fuid_dirty;
937 tx = dmu_tx_create(zfsvfs->z_os);
938 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
940 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
941 zfs_userquota_prop_prefixes[type]);
944 zfs_fuid_txhold(zfsvfs, tx);
945 err = dmu_tx_assign(tx, TXG_WAIT);
951 mutex_enter(&zfsvfs->z_lock);
953 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
955 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
956 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
958 mutex_exit(&zfsvfs->z_lock);
961 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
965 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
969 zfs_fuid_sync(zfsvfs, tx);
975 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
978 uint64_t used, quota, usedobj, quotaobj;
981 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
982 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
984 if (quotaobj == 0 || zfsvfs->z_replay)
987 (void) sprintf(buf, "%llx", (longlong_t)fuid);
988 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
992 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
995 return (used >= quota);
999 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
1004 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
1006 fuid = isgroup ? zp->z_gid : zp->z_uid;
1008 if (quotaobj == 0 || zfsvfs->z_replay)
1011 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
1015 * Associate this zfsvfs with the given objset, which must be owned.
1016 * This will cache a bunch of on-disk state from the objset in the
1020 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
1025 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
1026 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
1029 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
1032 if (zfsvfs->z_version >
1033 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
1034 (void) printf("Can't mount a version %lld file system "
1035 "on a version %lld pool\n. Pool must be upgraded to mount "
1036 "this file system.", (u_longlong_t)zfsvfs->z_version,
1037 (u_longlong_t)spa_version(dmu_objset_spa(os)));
1038 return (SET_ERROR(ENOTSUP));
1040 error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
1043 zfsvfs->z_norm = (int)val;
1045 error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
1048 zfsvfs->z_utf8 = (val != 0);
1050 error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
1053 zfsvfs->z_case = (uint_t)val;
1056 * Fold case on file systems that are always or sometimes case
1059 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
1060 zfsvfs->z_case == ZFS_CASE_MIXED)
1061 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
1063 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1064 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1066 uint64_t sa_obj = 0;
1067 if (zfsvfs->z_use_sa) {
1068 /* should either have both of these objects or none */
1069 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
1075 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
1076 &zfsvfs->z_attr_table);
1080 if (zfsvfs->z_version >= ZPL_VERSION_SA)
1081 sa_register_update_callback(os, zfs_sa_upgrade);
1083 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
1087 ASSERT(zfsvfs->z_root != 0);
1089 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
1090 &zfsvfs->z_unlinkedobj);
1094 error = zap_lookup(os, MASTER_NODE_OBJ,
1095 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
1096 8, 1, &zfsvfs->z_userquota_obj);
1097 if (error == ENOENT)
1098 zfsvfs->z_userquota_obj = 0;
1099 else if (error != 0)
1102 error = zap_lookup(os, MASTER_NODE_OBJ,
1103 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
1104 8, 1, &zfsvfs->z_groupquota_obj);
1105 if (error == ENOENT)
1106 zfsvfs->z_groupquota_obj = 0;
1107 else if (error != 0)
1110 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
1111 &zfsvfs->z_fuid_obj);
1112 if (error == ENOENT)
1113 zfsvfs->z_fuid_obj = 0;
1114 else if (error != 0)
1117 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
1118 &zfsvfs->z_shares_dir);
1119 if (error == ENOENT)
1120 zfsvfs->z_shares_dir = 0;
1121 else if (error != 0)
1125 * Only use the name cache if we are looking for a
1126 * name on a file system that does not require normalization
1127 * or case folding. We can also look there if we happen to be
1128 * on a non-normalizing, mixed sensitivity file system IF we
1129 * are looking for the exact name (which is always the case on
1132 zfsvfs->z_use_namecache = !zfsvfs->z_norm ||
1133 ((zfsvfs->z_case == ZFS_CASE_MIXED) &&
1134 !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER));
1139 #if defined(__FreeBSD__)
1140 taskq_t *zfsvfs_taskq;
1143 zfsvfs_task_unlinked_drain(void *context, int pending __unused)
1146 zfs_unlinked_drain((zfsvfs_t *)context);
1151 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
1158 * XXX: Fix struct statfs so this isn't necessary!
1160 * The 'osname' is used as the filesystem's special node, which means
1161 * it must fit in statfs.f_mntfromname, or else it can't be
1162 * enumerated, so libzfs_mnttab_find() returns NULL, which causes
1163 * 'zfs unmount' to think it's not mounted when it is.
1165 if (strlen(osname) >= MNAMELEN)
1166 return (SET_ERROR(ENAMETOOLONG));
1168 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
1171 * We claim to always be readonly so we can open snapshots;
1172 * other ZPL code will prevent us from writing to snapshots.
1175 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
1177 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1181 error = zfsvfs_create_impl(zfvp, zfsvfs, os);
1183 dmu_objset_disown(os, zfsvfs);
1190 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
1194 zfsvfs->z_vfs = NULL;
1195 zfsvfs->z_parent = zfsvfs;
1197 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1198 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1199 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1200 offsetof(znode_t, z_link_node));
1201 #if defined(__FreeBSD__)
1202 TASK_INIT(&zfsvfs->z_unlinked_drain_task, 0,
1203 zfsvfs_task_unlinked_drain, zfsvfs);
1206 rrm_init(&zfsvfs->z_teardown_lock, B_TRUE);
1208 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1210 rms_init(&zfsvfs->z_teardown_inactive_lock, "zfs teardown inactive");
1211 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1212 for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1213 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1215 error = zfsvfs_init(zfsvfs, os);
1218 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1227 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1231 error = zfs_register_callbacks(zfsvfs->z_vfs);
1235 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1238 * If we are not mounting (ie: online recv), then we don't
1239 * have to worry about replaying the log as we blocked all
1240 * operations out since we closed the ZIL.
1246 * During replay we remove the read only flag to
1247 * allow replays to succeed.
1249 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1251 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1253 zfs_unlinked_drain(zfsvfs);
1256 * Parse and replay the intent log.
1258 * Because of ziltest, this must be done after
1259 * zfs_unlinked_drain(). (Further note: ziltest
1260 * doesn't use readonly mounts, where
1261 * zfs_unlinked_drain() isn't called.) This is because
1262 * ziltest causes spa_sync() to think it's committed,
1263 * but actually it is not, so the intent log contains
1264 * many txg's worth of changes.
1266 * In particular, if object N is in the unlinked set in
1267 * the last txg to actually sync, then it could be
1268 * actually freed in a later txg and then reallocated
1269 * in a yet later txg. This would write a "create
1270 * object N" record to the intent log. Normally, this
1271 * would be fine because the spa_sync() would have
1272 * written out the fact that object N is free, before
1273 * we could write the "create object N" intent log
1276 * But when we are in ziltest mode, we advance the "open
1277 * txg" without actually spa_sync()-ing the changes to
1278 * disk. So we would see that object N is still
1279 * allocated and in the unlinked set, and there is an
1280 * intent log record saying to allocate it.
1282 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1283 if (zil_replay_disable) {
1284 zil_destroy(zfsvfs->z_log, B_FALSE);
1286 zfsvfs->z_replay = B_TRUE;
1287 zil_replay(zfsvfs->z_os, zfsvfs,
1289 zfsvfs->z_replay = B_FALSE;
1292 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1296 * Set the objset user_ptr to track its zfsvfs.
1298 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1299 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1300 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1305 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1308 zfsvfs_free(zfsvfs_t *zfsvfs)
1313 * This is a barrier to prevent the filesystem from going away in
1314 * zfs_znode_move() until we can safely ensure that the filesystem is
1315 * not unmounted. We consider the filesystem valid before the barrier
1316 * and invalid after the barrier.
1318 rw_enter(&zfsvfs_lock, RW_READER);
1319 rw_exit(&zfsvfs_lock);
1321 zfs_fuid_destroy(zfsvfs);
1323 mutex_destroy(&zfsvfs->z_znodes_lock);
1324 mutex_destroy(&zfsvfs->z_lock);
1325 list_destroy(&zfsvfs->z_all_znodes);
1326 rrm_destroy(&zfsvfs->z_teardown_lock);
1327 rms_destroy(&zfsvfs->z_teardown_inactive_lock);
1328 rw_destroy(&zfsvfs->z_fuid_lock);
1329 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1330 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1331 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1335 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1337 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1338 if (zfsvfs->z_vfs) {
1339 if (zfsvfs->z_use_fuids) {
1340 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1341 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1342 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1343 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1344 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1345 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1347 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1348 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1349 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1350 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1351 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1352 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1355 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1359 zfs_domount(vfs_t *vfsp, char *osname)
1361 uint64_t recordsize, fsid_guid;
1369 error = zfsvfs_create(osname, &zfsvfs);
1372 zfsvfs->z_vfs = vfsp;
1375 /* Initialize the generic filesystem structure. */
1376 vfsp->vfs_bcount = 0;
1377 vfsp->vfs_data = NULL;
1379 if (zfs_create_unique_device(&mount_dev) == -1) {
1380 error = SET_ERROR(ENODEV);
1383 ASSERT(vfs_devismounted(mount_dev) == 0);
1386 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1389 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1390 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1392 vfsp->vfs_data = zfsvfs;
1393 vfsp->mnt_flag |= MNT_LOCAL;
1394 #if defined(_KERNEL) && !defined(KMEM_DEBUG)
1395 vfsp->mnt_kern_flag |= MNTK_FPLOOKUP;
1397 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1398 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1399 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1400 vfsp->mnt_kern_flag |= MNTK_NO_IOPF; /* vn_io_fault can be used */
1401 vfsp->mnt_kern_flag |= MNTK_NOMSYNC;
1402 vfsp->mnt_kern_flag |= MNTK_VMSETSIZE_BUG;
1405 * The fsid is 64 bits, composed of an 8-bit fs type, which
1406 * separates our fsid from any other filesystem types, and a
1407 * 56-bit objset unique ID. The objset unique ID is unique to
1408 * all objsets open on this system, provided by unique_create().
1409 * The 8-bit fs type must be put in the low bits of fsid[1]
1410 * because that's where other Solaris filesystems put it.
1412 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1413 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1414 vfsp->vfs_fsid.val[0] = fsid_guid;
1415 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1416 vfsp->mnt_vfc->vfc_typenum & 0xFF;
1419 * Set features for file system.
1421 zfs_set_fuid_feature(zfsvfs);
1422 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1423 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1424 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1425 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1426 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1427 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1428 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1430 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1432 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1435 atime_changed_cb(zfsvfs, B_FALSE);
1436 readonly_changed_cb(zfsvfs, B_TRUE);
1437 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1439 xattr_changed_cb(zfsvfs, pval);
1440 zfsvfs->z_issnap = B_TRUE;
1441 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1443 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1444 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1445 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1447 error = zfsvfs_setup(zfsvfs, B_TRUE);
1450 vfs_mountedfrom(vfsp, osname);
1452 if (!zfsvfs->z_issnap)
1453 zfsctl_create(zfsvfs);
1456 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1457 zfsvfs_free(zfsvfs);
1459 atomic_inc_32(&zfs_active_fs_count);
1466 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1468 objset_t *os = zfsvfs->z_os;
1470 if (!dmu_objset_is_snapshot(os))
1471 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1476 * Convert a decimal digit string to a uint64_t integer.
1479 str_to_uint64(char *str, uint64_t *objnum)
1484 if (*str < '0' || *str > '9')
1485 return (SET_ERROR(EINVAL));
1487 num = num*10 + *str++ - '0';
1495 * The boot path passed from the boot loader is in the form of
1496 * "rootpool-name/root-filesystem-object-number'. Convert this
1497 * string to a dataset name: "rootpool-name/root-filesystem-name".
1500 zfs_parse_bootfs(char *bpath, char *outpath)
1506 if (*bpath == 0 || *bpath == '/')
1507 return (SET_ERROR(EINVAL));
1509 (void) strcpy(outpath, bpath);
1511 slashp = strchr(bpath, '/');
1513 /* if no '/', just return the pool name */
1514 if (slashp == NULL) {
1518 /* if not a number, just return the root dataset name */
1519 if (str_to_uint64(slashp+1, &objnum)) {
1524 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1531 * Check that the hex label string is appropriate for the dataset being
1532 * mounted into the global_zone proper.
1534 * Return an error if the hex label string is not default or
1535 * admin_low/admin_high. For admin_low labels, the corresponding
1536 * dataset must be readonly.
1539 zfs_check_global_label(const char *dsname, const char *hexsl)
1541 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1543 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1545 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1546 /* must be readonly */
1549 if (dsl_prop_get_integer(dsname,
1550 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1551 return (SET_ERROR(EACCES));
1552 return (rdonly ? 0 : EACCES);
1554 return (SET_ERROR(EACCES));
1558 * Determine whether the mount is allowed according to MAC check.
1559 * by comparing (where appropriate) label of the dataset against
1560 * the label of the zone being mounted into. If the dataset has
1561 * no label, create one.
1563 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1566 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1569 zone_t *mntzone = NULL;
1570 ts_label_t *mnt_tsl;
1573 char ds_hexsl[MAXNAMELEN];
1575 retv = EACCES; /* assume the worst */
1578 * Start by getting the dataset label if it exists.
1580 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1581 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1583 return (SET_ERROR(EACCES));
1586 * If labeling is NOT enabled, then disallow the mount of datasets
1587 * which have a non-default label already. No other label checks
1590 if (!is_system_labeled()) {
1591 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1593 return (SET_ERROR(EACCES));
1597 * Get the label of the mountpoint. If mounting into the global
1598 * zone (i.e. mountpoint is not within an active zone and the
1599 * zoned property is off), the label must be default or
1600 * admin_low/admin_high only; no other checks are needed.
1602 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1603 if (mntzone->zone_id == GLOBAL_ZONEID) {
1608 if (dsl_prop_get_integer(osname,
1609 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1610 return (SET_ERROR(EACCES));
1612 return (zfs_check_global_label(osname, ds_hexsl));
1615 * This is the case of a zone dataset being mounted
1616 * initially, before the zone has been fully created;
1617 * allow this mount into global zone.
1622 mnt_tsl = mntzone->zone_slabel;
1623 ASSERT(mnt_tsl != NULL);
1624 label_hold(mnt_tsl);
1625 mnt_sl = label2bslabel(mnt_tsl);
1627 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1629 * The dataset doesn't have a real label, so fabricate one.
1633 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1634 dsl_prop_set_string(osname,
1635 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1636 ZPROP_SRC_LOCAL, str) == 0)
1639 kmem_free(str, strlen(str) + 1);
1640 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1642 * Now compare labels to complete the MAC check. If the
1643 * labels are equal then allow access. If the mountpoint
1644 * label dominates the dataset label, allow readonly access.
1645 * Otherwise, access is denied.
1647 if (blequal(mnt_sl, &ds_sl))
1649 else if (bldominates(mnt_sl, &ds_sl)) {
1650 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1655 label_rele(mnt_tsl);
1659 #endif /* SECLABEL */
1661 #ifdef OPENSOLARIS_MOUNTROOT
1663 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1666 static int zfsrootdone = 0;
1667 zfsvfs_t *zfsvfs = NULL;
1676 * The filesystem that we mount as root is defined in the
1677 * boot property "zfs-bootfs" with a format of
1678 * "poolname/root-dataset-objnum".
1680 if (why == ROOT_INIT) {
1682 return (SET_ERROR(EBUSY));
1684 * the process of doing a spa_load will require the
1685 * clock to be set before we could (for example) do
1686 * something better by looking at the timestamp on
1687 * an uberblock, so just set it to -1.
1691 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1692 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1694 return (SET_ERROR(EINVAL));
1696 zfs_devid = spa_get_bootprop("diskdevid");
1697 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1699 spa_free_bootprop(zfs_devid);
1701 spa_free_bootprop(zfs_bootfs);
1702 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1706 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1707 spa_free_bootprop(zfs_bootfs);
1708 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1713 spa_free_bootprop(zfs_bootfs);
1715 if (error = vfs_lock(vfsp))
1718 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1719 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1723 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1725 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1726 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1731 mutex_enter(&vp->v_lock);
1732 vp->v_flag |= VROOT;
1733 mutex_exit(&vp->v_lock);
1737 * Leave rootvp held. The root file system is never unmounted.
1740 vfs_add((struct vnode *)0, vfsp,
1741 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1745 } else if (why == ROOT_REMOUNT) {
1746 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1747 vfsp->vfs_flag |= VFS_REMOUNT;
1749 /* refresh mount options */
1750 zfs_unregister_callbacks(vfsp->vfs_data);
1751 return (zfs_register_callbacks(vfsp));
1753 } else if (why == ROOT_UNMOUNT) {
1754 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1755 (void) zfs_sync(vfsp, 0, 0);
1760 * if "why" is equal to anything else other than ROOT_INIT,
1761 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1763 return (SET_ERROR(ENOTSUP));
1765 #endif /* OPENSOLARIS_MOUNTROOT */
1768 getpoolname(const char *osname, char *poolname)
1772 p = strchr(osname, '/');
1774 if (strlen(osname) >= MAXNAMELEN)
1775 return (ENAMETOOLONG);
1776 (void) strcpy(poolname, osname);
1778 if (p - osname >= MAXNAMELEN)
1779 return (ENAMETOOLONG);
1780 (void) strncpy(poolname, osname, p - osname);
1781 poolname[p - osname] = '\0';
1787 fetch_osname_options(char *name, bool *checkpointrewind)
1790 if (name[0] == '!') {
1791 *checkpointrewind = true;
1792 memmove(name, name + 1, strlen(name));
1794 *checkpointrewind = false;
1800 zfs_mount(vfs_t *vfsp)
1802 kthread_t *td = curthread;
1803 vnode_t *mvp = vfsp->mnt_vnodecovered;
1804 cred_t *cr = td->td_ucred;
1808 bool checkpointrewind;
1811 if (mvp->v_type != VDIR)
1812 return (SET_ERROR(ENOTDIR));
1814 mutex_enter(&mvp->v_lock);
1815 if ((uap->flags & MS_REMOUNT) == 0 &&
1816 (uap->flags & MS_OVERLAY) == 0 &&
1817 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1818 mutex_exit(&mvp->v_lock);
1819 return (SET_ERROR(EBUSY));
1821 mutex_exit(&mvp->v_lock);
1824 * ZFS does not support passing unparsed data in via MS_DATA.
1825 * Users should use the MS_OPTIONSTR interface; this means
1826 * that all option parsing is already done and the options struct
1827 * can be interrogated.
1829 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1830 return (SET_ERROR(EINVAL));
1833 * Get the objset name (the "special" mount argument).
1835 if (error = pn_get(uap->spec, fromspace, &spn))
1838 osname = spn.pn_path;
1839 #else /* !illumos */
1840 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1841 return (SET_ERROR(EINVAL));
1844 * If full-owner-access is enabled and delegated administration is
1845 * turned on, we must set nosuid.
1847 if (zfs_super_owner &&
1848 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1849 secpolicy_fs_mount_clearopts(cr, vfsp);
1851 #endif /* illumos */
1852 fetch_osname_options(osname, &checkpointrewind);
1855 * Check for mount privilege?
1857 * If we don't have privilege then see if
1858 * we have local permission to allow it
1860 error = secpolicy_fs_mount(cr, mvp, vfsp);
1862 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1865 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1869 * Make sure user is the owner of the mount point
1870 * or has sufficient privileges.
1873 vattr.va_mask = AT_UID;
1875 vn_lock(mvp, LK_SHARED | LK_RETRY);
1876 if (VOP_GETATTR(mvp, &vattr, cr)) {
1881 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1882 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1889 secpolicy_fs_mount_clearopts(cr, vfsp);
1893 * Refuse to mount a filesystem if we are in a local zone and the
1894 * dataset is not visible.
1896 if (!INGLOBALZONE(curthread) &&
1897 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1898 error = SET_ERROR(EPERM);
1903 error = zfs_mount_label_policy(vfsp, osname);
1908 vfsp->vfs_flag |= MNT_NFS4ACLS;
1911 * When doing a remount, we simply refresh our temporary properties
1912 * according to those options set in the current VFS options.
1914 if (vfsp->vfs_flag & MS_REMOUNT) {
1915 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1918 * Refresh mount options with z_teardown_lock blocking I/O while
1919 * the filesystem is in an inconsistent state.
1920 * The lock also serializes this code with filesystem
1921 * manipulations between entry to zfs_suspend_fs() and return
1922 * from zfs_resume_fs().
1924 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1925 zfs_unregister_callbacks(zfsvfs);
1926 error = zfs_register_callbacks(vfsp);
1927 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1931 /* Initial root mount: try hard to import the requested root pool. */
1932 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1933 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1934 char pname[MAXNAMELEN];
1936 error = getpoolname(osname, pname);
1938 error = spa_import_rootpool(pname, checkpointrewind);
1943 error = zfs_domount(vfsp, osname);
1948 * Add an extra VFS_HOLD on our parent vfs so that it can't
1949 * disappear due to a forced unmount.
1951 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1952 VFS_HOLD(mvp->v_vfsp);
1960 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1962 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1963 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1965 statp->f_version = STATFS_VERSION;
1969 dmu_objset_space(zfsvfs->z_os,
1970 &refdbytes, &availbytes, &usedobjs, &availobjs);
1973 * The underlying storage pool actually uses multiple block sizes.
1974 * We report the fragsize as the smallest block size we support,
1975 * and we report our blocksize as the filesystem's maximum blocksize.
1977 statp->f_bsize = SPA_MINBLOCKSIZE;
1978 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1981 * The following report "total" blocks of various kinds in the
1982 * file system, but reported in terms of f_frsize - the
1986 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1987 statp->f_bfree = availbytes / statp->f_bsize;
1988 statp->f_bavail = statp->f_bfree; /* no root reservation */
1991 * statvfs() should really be called statufs(), because it assumes
1992 * static metadata. ZFS doesn't preallocate files, so the best
1993 * we can do is report the max that could possibly fit in f_files,
1994 * and that minus the number actually used in f_ffree.
1995 * For f_ffree, report the smaller of the number of object available
1996 * and the number of blocks (each object will take at least a block).
1998 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1999 statp->f_files = statp->f_ffree + usedobjs;
2002 * We're a zfs filesystem.
2004 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
2006 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
2007 sizeof(statp->f_mntfromname));
2008 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
2009 sizeof(statp->f_mntonname));
2011 statp->f_namemax = MAXNAMELEN - 1;
2018 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
2020 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2026 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
2028 *vpp = ZTOV(rootzp);
2033 error = vn_lock(*vpp, flags);
2043 * Teardown the zfsvfs::z_os.
2045 * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
2046 * and 'z_teardown_inactive_lock' held.
2049 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
2053 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
2057 * We purge the parent filesystem's vfsp as the parent
2058 * filesystem and all of its snapshots have their vnode's
2059 * v_vfsp set to the parent's filesystem's vfsp. Note,
2060 * 'z_parent' is self referential for non-snapshots.
2062 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2063 #ifdef FREEBSD_NAMECACHE
2064 cache_purgevfs(zfsvfs->z_parent->z_vfs, true);
2069 * Close the zil. NB: Can't close the zil while zfs_inactive
2070 * threads are blocked as zil_close can call zfs_inactive.
2072 if (zfsvfs->z_log) {
2073 zil_close(zfsvfs->z_log);
2074 zfsvfs->z_log = NULL;
2077 ZFS_WLOCK_TEARDOWN_INACTIVE(zfsvfs);
2080 * If we are not unmounting (ie: online recv) and someone already
2081 * unmounted this file system while we were doing the switcheroo,
2082 * or a reopen of z_os failed then just bail out now.
2084 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
2085 ZFS_WUNLOCK_TEARDOWN_INACTIVE(zfsvfs);
2086 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2087 return (SET_ERROR(EIO));
2091 * At this point there are no vops active, and any new vops will
2092 * fail with EIO since we have z_teardown_lock for writer (only
2093 * relavent for forced unmount).
2095 * Release all holds on dbufs.
2097 mutex_enter(&zfsvfs->z_znodes_lock);
2098 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
2099 zp = list_next(&zfsvfs->z_all_znodes, zp))
2101 ASSERT(ZTOV(zp)->v_count >= 0);
2102 zfs_znode_dmu_fini(zp);
2104 mutex_exit(&zfsvfs->z_znodes_lock);
2107 * If we are unmounting, set the unmounted flag and let new vops
2108 * unblock. zfs_inactive will have the unmounted behavior, and all
2109 * other vops will fail with EIO.
2112 zfsvfs->z_unmounted = B_TRUE;
2113 ZFS_WUNLOCK_TEARDOWN_INACTIVE(zfsvfs);
2114 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2118 * z_os will be NULL if there was an error in attempting to reopen
2119 * zfsvfs, so just return as the properties had already been
2120 * unregistered and cached data had been evicted before.
2122 if (zfsvfs->z_os == NULL)
2126 * Unregister properties.
2128 zfs_unregister_callbacks(zfsvfs);
2133 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
2134 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
2135 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
2136 dmu_objset_evict_dbufs(zfsvfs->z_os);
2143 zfs_umount(vfs_t *vfsp, int fflag)
2145 kthread_t *td = curthread;
2146 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2148 cred_t *cr = td->td_ucred;
2151 ret = secpolicy_fs_unmount(cr, vfsp);
2153 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
2154 ZFS_DELEG_PERM_MOUNT, cr))
2159 * We purge the parent filesystem's vfsp as the parent filesystem
2160 * and all of its snapshots have their vnode's v_vfsp set to the
2161 * parent's filesystem's vfsp. Note, 'z_parent' is self
2162 * referential for non-snapshots.
2164 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
2167 * Unmount any snapshots mounted under .zfs before unmounting the
2170 if (zfsvfs->z_ctldir != NULL) {
2171 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
2175 if (fflag & MS_FORCE) {
2177 * Mark file system as unmounted before calling
2178 * vflush(FORCECLOSE). This way we ensure no future vnops
2179 * will be called and risk operating on DOOMED vnodes.
2181 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
2182 zfsvfs->z_unmounted = B_TRUE;
2183 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2187 * Flush all the files.
2189 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
2194 if (!(fflag & MS_FORCE)) {
2196 * Check the number of active vnodes in the file system.
2197 * Our count is maintained in the vfs structure, but the
2198 * number is off by 1 to indicate a hold on the vfs
2201 * The '.zfs' directory maintains a reference of its
2202 * own, and any active references underneath are
2203 * reflected in the vnode count.
2205 if (zfsvfs->z_ctldir == NULL) {
2206 if (vfsp->vfs_count > 1)
2207 return (SET_ERROR(EBUSY));
2209 if (vfsp->vfs_count > 2 ||
2210 zfsvfs->z_ctldir->v_count > 1)
2211 return (SET_ERROR(EBUSY));
2216 while (taskqueue_cancel(zfsvfs_taskq->tq_queue,
2217 &zfsvfs->z_unlinked_drain_task, NULL) != 0)
2218 taskqueue_drain(zfsvfs_taskq->tq_queue,
2219 &zfsvfs->z_unlinked_drain_task);
2221 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2225 * z_os will be NULL if there was an error in
2226 * attempting to reopen zfsvfs.
2230 * Unset the objset user_ptr.
2232 mutex_enter(&os->os_user_ptr_lock);
2233 dmu_objset_set_user(os, NULL);
2234 mutex_exit(&os->os_user_ptr_lock);
2237 * Finally release the objset
2239 dmu_objset_disown(os, zfsvfs);
2243 * We can now safely destroy the '.zfs' directory node.
2245 if (zfsvfs->z_ctldir != NULL)
2246 zfsctl_destroy(zfsvfs);
2253 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2255 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2260 * zfs_zget() can't operate on virtual entries like .zfs/ or
2261 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2262 * This will make NFS to switch to LOOKUP instead of using VGET.
2264 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2265 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2266 return (EOPNOTSUPP);
2269 err = zfs_zget(zfsvfs, ino, &zp);
2270 if (err == 0 && zp->z_unlinked) {
2278 err = vn_lock(*vpp, flags);
2288 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, uint64_t *extflagsp,
2289 struct ucred **credanonp, int *numsecflavors, int *secflavors)
2291 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2294 * If this is regular file system vfsp is the same as
2295 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2296 * zfsvfs->z_parent->z_vfs represents parent file system
2297 * which we have to use here, because only this file system
2298 * has mnt_export configured.
2300 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2301 credanonp, numsecflavors, secflavors));
2304 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2305 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2308 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2310 struct componentname cn;
2311 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2314 uint64_t object = 0;
2315 uint64_t fid_gen = 0;
2325 * On FreeBSD we can get snapshot's mount point or its parent file
2326 * system mount point depending if snapshot is already mounted or not.
2328 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2329 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2330 uint64_t objsetid = 0;
2331 uint64_t setgen = 0;
2333 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2334 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2336 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2337 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2341 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2343 return (SET_ERROR(EINVAL));
2347 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2348 zfid_short_t *zfid = (zfid_short_t *)fidp;
2350 for (i = 0; i < sizeof (zfid->zf_object); i++)
2351 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2353 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2354 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2357 return (SET_ERROR(EINVAL));
2361 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2362 * directory tree. If the object == zfsvfs->z_shares_dir, then
2363 * we are in the .zfs/shares directory tree.
2365 if ((fid_gen == 0 &&
2366 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2367 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2369 VERIFY0(zfsctl_root(zfsvfs, LK_SHARED, &dvp));
2370 if (object == ZFSCTL_INO_SNAPDIR) {
2371 cn.cn_nameptr = "snapshot";
2372 cn.cn_namelen = strlen(cn.cn_nameptr);
2373 cn.cn_nameiop = LOOKUP;
2374 cn.cn_flags = ISLASTCN | LOCKLEAF;
2375 cn.cn_lkflags = flags;
2376 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2378 } else if (object == zfsvfs->z_shares_dir) {
2380 * XXX This branch must not be taken,
2381 * if it is, then the lookup below will
2384 cn.cn_nameptr = "shares";
2385 cn.cn_namelen = strlen(cn.cn_nameptr);
2386 cn.cn_nameiop = LOOKUP;
2387 cn.cn_flags = ISLASTCN;
2388 cn.cn_lkflags = flags;
2389 VERIFY0(VOP_LOOKUP(dvp, vpp, &cn));
2397 gen_mask = -1ULL >> (64 - 8 * i);
2399 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2400 if (err = zfs_zget(zfsvfs, object, &zp)) {
2404 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2406 zp_gen = zp_gen & gen_mask;
2409 if (zp->z_unlinked || zp_gen != fid_gen) {
2410 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2413 return (SET_ERROR(EINVAL));
2418 err = vn_lock(*vpp, flags);
2420 vnode_create_vobject(*vpp, zp->z_size, curthread);
2427 * Block out VOPs and close zfsvfs_t::z_os
2429 * Note, if successful, then we return with the 'z_teardown_lock' and
2430 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2431 * dataset and objset intact so that they can be atomically handed off during
2432 * a subsequent rollback or recv operation and the resume thereafter.
2435 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2439 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2446 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2447 * is an invariant across any of the operations that can be performed while the
2448 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2449 * are the same: the relevant objset and associated dataset are owned by
2450 * zfsvfs, held, and long held on entry.
2453 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2458 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2459 ASSERT(ZFS_TEARDOWN_INACTIVE_WLOCKED(zfsvfs));
2462 * We already own this, so just update the objset_t, as the one we
2463 * had before may have been evicted.
2466 VERIFY3P(ds->ds_owner, ==, zfsvfs);
2467 VERIFY(dsl_dataset_long_held(ds));
2468 VERIFY0(dmu_objset_from_ds(ds, &os));
2470 err = zfsvfs_init(zfsvfs, os);
2474 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2476 zfs_set_fuid_feature(zfsvfs);
2479 * Attempt to re-establish all the active znodes with
2480 * their dbufs. If a zfs_rezget() fails, then we'll let
2481 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2482 * when they try to use their znode.
2484 mutex_enter(&zfsvfs->z_znodes_lock);
2485 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2486 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2487 (void) zfs_rezget(zp);
2489 mutex_exit(&zfsvfs->z_znodes_lock);
2492 /* release the VOPs */
2493 ZFS_WUNLOCK_TEARDOWN_INACTIVE(zfsvfs);
2494 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2498 * Since we couldn't setup the sa framework, try to force
2499 * unmount this file system.
2501 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) {
2502 vfs_ref(zfsvfs->z_vfs);
2503 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2510 zfs_freevfs(vfs_t *vfsp)
2512 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2516 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2517 * from zfs_mount(). Release it here. If we came through
2518 * zfs_mountroot() instead, we didn't grab an extra hold, so
2519 * skip the VFS_RELE for rootvfs.
2521 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2522 VFS_RELE(zfsvfs->z_parent->z_vfs);
2525 zfsvfs_free(zfsvfs);
2527 atomic_dec_32(&zfs_active_fs_count);
2531 static int desiredvnodes_backup;
2535 zfs_vnodes_adjust(void)
2538 int newdesiredvnodes;
2540 desiredvnodes_backup = desiredvnodes;
2543 * We calculate newdesiredvnodes the same way it is done in
2544 * vntblinit(). If it is equal to desiredvnodes, it means that
2545 * it wasn't tuned by the administrator and we can tune it down.
2547 newdesiredvnodes = min(maxproc + vm_cnt.v_page_count / 4, 2 *
2548 vm_kmem_size / (5 * (sizeof(struct vm_object) +
2549 sizeof(struct vnode))));
2550 if (newdesiredvnodes == desiredvnodes)
2551 desiredvnodes = (3 * newdesiredvnodes) / 4;
2556 zfs_vnodes_adjust_back(void)
2560 desiredvnodes = desiredvnodes_backup;
2568 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2571 * Initialize .zfs directory structures
2576 * Initialize znode cache, vnode ops, etc...
2581 * Reduce number of vnodes. Originally number of vnodes is calculated
2582 * with UFS inode in mind. We reduce it here, because it's too big for
2585 zfs_vnodes_adjust();
2587 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2588 #if defined(__FreeBSD__)
2589 zfsvfs_taskq = taskq_create("zfsvfs", 1, minclsyspri, 0, 0, 0);
2596 #if defined(__FreeBSD__)
2597 taskq_destroy(zfsvfs_taskq);
2601 zfs_vnodes_adjust_back();
2607 return (zfs_active_fs_count != 0);
2611 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2614 objset_t *os = zfsvfs->z_os;
2617 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2618 return (SET_ERROR(EINVAL));
2620 if (newvers < zfsvfs->z_version)
2621 return (SET_ERROR(EINVAL));
2623 if (zfs_spa_version_map(newvers) >
2624 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2625 return (SET_ERROR(ENOTSUP));
2627 tx = dmu_tx_create(os);
2628 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2629 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2630 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2632 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2634 error = dmu_tx_assign(tx, TXG_WAIT);
2640 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2641 8, 1, &newvers, tx);
2648 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2651 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2653 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2654 DMU_OT_NONE, 0, tx);
2656 error = zap_add(os, MASTER_NODE_OBJ,
2657 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2660 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2661 sa_register_update_callback(os, zfs_sa_upgrade);
2664 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2665 "from %llu to %llu", zfsvfs->z_version, newvers);
2669 zfsvfs->z_version = newvers;
2670 os->os_version = newvers;
2672 zfs_set_fuid_feature(zfsvfs);
2678 * Read a property stored within the master node.
2681 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2683 uint64_t *cached_copy = NULL;
2686 * Figure out where in the objset_t the cached copy would live, if it
2687 * is available for the requested property.
2691 case ZFS_PROP_VERSION:
2692 cached_copy = &os->os_version;
2694 case ZFS_PROP_NORMALIZE:
2695 cached_copy = &os->os_normalization;
2697 case ZFS_PROP_UTF8ONLY:
2698 cached_copy = &os->os_utf8only;
2701 cached_copy = &os->os_casesensitivity;
2707 if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) {
2708 *value = *cached_copy;
2713 * If the property wasn't cached, look up the file system's value for
2714 * the property. For the version property, we look up a slightly
2719 if (prop == ZFS_PROP_VERSION) {
2720 pname = ZPL_VERSION_STR;
2722 pname = zfs_prop_to_name(prop);
2726 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2727 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2730 if (error == ENOENT) {
2731 /* No value set, use the default value */
2733 case ZFS_PROP_VERSION:
2734 *value = ZPL_VERSION;
2736 case ZFS_PROP_NORMALIZE:
2737 case ZFS_PROP_UTF8ONLY:
2741 *value = ZFS_CASE_SENSITIVE;
2750 * If one of the methods for getting the property value above worked,
2751 * copy it into the objset_t's cache.
2753 if (error == 0 && cached_copy != NULL) {
2754 *cached_copy = *value;
2761 * Return true if the coresponding vfs's unmounted flag is set.
2762 * Otherwise return false.
2763 * If this function returns true we know VFS unmount has been initiated.
2766 zfs_get_vfs_flag_unmounted(objset_t *os)
2769 boolean_t unmounted = B_FALSE;
2771 ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2773 mutex_enter(&os->os_user_ptr_lock);
2774 zfvp = dmu_objset_get_user(os);
2775 if (zfvp != NULL && zfvp->z_vfs != NULL &&
2776 (zfvp->z_vfs->mnt_kern_flag & MNTK_UNMOUNT))
2778 mutex_exit(&os->os_user_ptr_lock);
2785 zfsvfs_update_fromname(const char *oldname, const char *newname)
2787 char tmpbuf[MAXPATHLEN];
2792 oldlen = strlen(oldname);
2794 mtx_lock(&mountlist_mtx);
2795 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2796 fromname = mp->mnt_stat.f_mntfromname;
2797 if (strcmp(fromname, oldname) == 0) {
2798 (void)strlcpy(fromname, newname,
2799 sizeof(mp->mnt_stat.f_mntfromname));
2802 if (strncmp(fromname, oldname, oldlen) == 0 &&
2803 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2804 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
2805 newname, fromname + oldlen);
2806 (void)strlcpy(fromname, tmpbuf,
2807 sizeof(mp->mnt_stat.f_mntfromname));
2811 mtx_unlock(&mountlist_mtx);