4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2011 Pawel Jakub Dawidek <pawel@dawidek.net>.
24 * All rights reserved.
25 * Copyright (c) 2013 by Delphix. All rights reserved.
28 /* Portions Copyright 2010 Robert Milkowski */
30 #include <sys/types.h>
31 #include <sys/param.h>
32 #include <sys/systm.h>
33 #include <sys/kernel.h>
34 #include <sys/sysmacros.h>
37 #include <sys/vnode.h>
39 #include <sys/mntent.h>
40 #include <sys/mount.h>
41 #include <sys/cmn_err.h>
42 #include <sys/zfs_znode.h>
43 #include <sys/zfs_dir.h>
45 #include <sys/fs/zfs.h>
47 #include <sys/dsl_prop.h>
48 #include <sys/dsl_dataset.h>
49 #include <sys/dsl_deleg.h>
53 #include <sys/sa_impl.h>
54 #include <sys/varargs.h>
55 #include <sys/policy.h>
56 #include <sys/atomic.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/zfs_ctldir.h>
59 #include <sys/zfs_fuid.h>
60 #include <sys/sunddi.h>
62 #include <sys/dmu_objset.h>
63 #include <sys/spa_boot.h>
65 #include "zfs_comutil.h"
67 struct mtx zfs_debug_mtx;
68 MTX_SYSINIT(zfs_debug_mtx, &zfs_debug_mtx, "zfs_debug", MTX_DEF);
70 SYSCTL_NODE(_vfs, OID_AUTO, zfs, CTLFLAG_RW, 0, "ZFS file system");
73 SYSCTL_INT(_vfs_zfs, OID_AUTO, super_owner, CTLFLAG_RW, &zfs_super_owner, 0,
74 "File system owner can perform privileged operation on his file systems");
77 TUNABLE_INT("vfs.zfs.debug", &zfs_debug_level);
78 SYSCTL_INT(_vfs_zfs, OID_AUTO, debug, CTLFLAG_RW, &zfs_debug_level, 0,
81 SYSCTL_NODE(_vfs_zfs, OID_AUTO, version, CTLFLAG_RD, 0, "ZFS versions");
82 static int zfs_version_acl = ZFS_ACL_VERSION;
83 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, acl, CTLFLAG_RD, &zfs_version_acl, 0,
85 static int zfs_version_spa = SPA_VERSION;
86 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, spa, CTLFLAG_RD, &zfs_version_spa, 0,
88 static int zfs_version_zpl = ZPL_VERSION;
89 SYSCTL_INT(_vfs_zfs_version, OID_AUTO, zpl, CTLFLAG_RD, &zfs_version_zpl, 0,
92 static int zfs_mount(vfs_t *vfsp);
93 static int zfs_umount(vfs_t *vfsp, int fflag);
94 static int zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp);
95 static int zfs_statfs(vfs_t *vfsp, struct statfs *statp);
96 static int zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp);
97 static int zfs_sync(vfs_t *vfsp, int waitfor);
98 static int zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
99 struct ucred **credanonp, int *numsecflavors, int **secflavors);
100 static int zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp);
101 static void zfs_objset_close(zfsvfs_t *zfsvfs);
102 static void zfs_freevfs(vfs_t *vfsp);
104 static struct vfsops zfs_vfsops = {
105 .vfs_mount = zfs_mount,
106 .vfs_unmount = zfs_umount,
107 .vfs_root = zfs_root,
108 .vfs_statfs = zfs_statfs,
109 .vfs_vget = zfs_vget,
110 .vfs_sync = zfs_sync,
111 .vfs_checkexp = zfs_checkexp,
112 .vfs_fhtovp = zfs_fhtovp,
115 VFS_SET(zfs_vfsops, zfs, VFCF_JAIL | VFCF_DELEGADMIN);
118 * We need to keep a count of active fs's.
119 * This is necessary to prevent our module
120 * from being unloaded after a umount -f
122 static uint32_t zfs_active_fs_count = 0;
126 zfs_sync(vfs_t *vfsp, int waitfor)
130 * Data integrity is job one. We don't want a compromised kernel
131 * writing to the storage pool, so we never sync during panic.
138 * Sync a specific filesystem.
140 zfsvfs_t *zfsvfs = vfsp->vfs_data;
144 error = vfs_stdsync(vfsp, waitfor);
149 dp = dmu_objset_pool(zfsvfs->z_os);
152 * If the system is shutting down, then skip any
153 * filesystems which may exist on a suspended pool.
155 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
160 if (zfsvfs->z_log != NULL)
161 zil_commit(zfsvfs->z_log, 0);
166 * Sync all ZFS filesystems. This is what happens when you
167 * run sync(1M). Unlike other filesystems, ZFS honors the
168 * request by waiting for all pools to commit all dirty data.
176 #ifndef __FreeBSD_kernel__
178 zfs_create_unique_device(dev_t *dev)
183 ASSERT3U(zfs_minor, <=, MAXMIN32);
184 minor_t start = zfs_minor;
186 mutex_enter(&zfs_dev_mtx);
187 if (zfs_minor >= MAXMIN32) {
189 * If we're still using the real major
190 * keep out of /dev/zfs and /dev/zvol minor
191 * number space. If we're using a getudev()'ed
192 * major number, we can use all of its minors.
194 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
195 zfs_minor = ZFS_MIN_MINOR;
201 *dev = makedevice(zfs_major, zfs_minor);
202 mutex_exit(&zfs_dev_mtx);
203 } while (vfs_devismounted(*dev) && zfs_minor != start);
204 if (zfs_minor == start) {
206 * We are using all ~262,000 minor numbers for the
207 * current major number. Create a new major number.
209 if ((new_major = getudev()) == (major_t)-1) {
211 "zfs_mount: Can't get unique major "
215 mutex_enter(&zfs_dev_mtx);
216 zfs_major = new_major;
219 mutex_exit(&zfs_dev_mtx);
223 /* CONSTANTCONDITION */
228 #endif /* !__FreeBSD_kernel__ */
231 atime_changed_cb(void *arg, uint64_t newval)
233 zfsvfs_t *zfsvfs = arg;
235 if (newval == TRUE) {
236 zfsvfs->z_atime = TRUE;
237 zfsvfs->z_vfs->vfs_flag &= ~MNT_NOATIME;
238 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
239 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
241 zfsvfs->z_atime = FALSE;
242 zfsvfs->z_vfs->vfs_flag |= MNT_NOATIME;
243 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
244 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
249 xattr_changed_cb(void *arg, uint64_t newval)
251 zfsvfs_t *zfsvfs = arg;
253 if (newval == TRUE) {
254 /* XXX locking on vfs_flag? */
256 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
258 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
259 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
261 /* XXX locking on vfs_flag? */
263 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
265 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
266 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
271 blksz_changed_cb(void *arg, uint64_t newval)
273 zfsvfs_t *zfsvfs = arg;
274 ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
275 ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
276 ASSERT(ISP2(newval));
278 zfsvfs->z_max_blksz = newval;
279 zfsvfs->z_vfs->mnt_stat.f_iosize = newval;
283 readonly_changed_cb(void *arg, uint64_t newval)
285 zfsvfs_t *zfsvfs = arg;
288 /* XXX locking on vfs_flag? */
289 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
290 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
291 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
293 /* XXX locking on vfs_flag? */
294 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
295 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
296 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
301 setuid_changed_cb(void *arg, uint64_t newval)
303 zfsvfs_t *zfsvfs = arg;
305 if (newval == FALSE) {
306 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
307 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
308 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
310 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
311 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
312 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
317 exec_changed_cb(void *arg, uint64_t newval)
319 zfsvfs_t *zfsvfs = arg;
321 if (newval == FALSE) {
322 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
323 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
324 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
326 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
327 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
328 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
333 * The nbmand mount option can be changed at mount time.
334 * We can't allow it to be toggled on live file systems or incorrect
335 * behavior may be seen from cifs clients
337 * This property isn't registered via dsl_prop_register(), but this callback
338 * will be called when a file system is first mounted
341 nbmand_changed_cb(void *arg, uint64_t newval)
343 zfsvfs_t *zfsvfs = arg;
344 if (newval == FALSE) {
345 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
346 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
348 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
349 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
354 snapdir_changed_cb(void *arg, uint64_t newval)
356 zfsvfs_t *zfsvfs = arg;
358 zfsvfs->z_show_ctldir = newval;
362 vscan_changed_cb(void *arg, uint64_t newval)
364 zfsvfs_t *zfsvfs = arg;
366 zfsvfs->z_vscan = newval;
370 acl_mode_changed_cb(void *arg, uint64_t newval)
372 zfsvfs_t *zfsvfs = arg;
374 zfsvfs->z_acl_mode = newval;
378 acl_inherit_changed_cb(void *arg, uint64_t newval)
380 zfsvfs_t *zfsvfs = arg;
382 zfsvfs->z_acl_inherit = newval;
386 zfs_register_callbacks(vfs_t *vfsp)
388 struct dsl_dataset *ds = NULL;
390 zfsvfs_t *zfsvfs = NULL;
392 boolean_t readonly = B_FALSE;
393 boolean_t do_readonly = B_FALSE;
394 boolean_t setuid = B_FALSE;
395 boolean_t do_setuid = B_FALSE;
396 boolean_t exec = B_FALSE;
397 boolean_t do_exec = B_FALSE;
399 boolean_t devices = B_FALSE;
400 boolean_t do_devices = B_FALSE;
402 boolean_t xattr = B_FALSE;
403 boolean_t do_xattr = B_FALSE;
404 boolean_t atime = B_FALSE;
405 boolean_t do_atime = B_FALSE;
409 zfsvfs = vfsp->vfs_data;
414 * This function can be called for a snapshot when we update snapshot's
415 * mount point, which isn't really supported.
417 if (dmu_objset_is_snapshot(os))
421 * The act of registering our callbacks will destroy any mount
422 * options we may have. In order to enable temporary overrides
423 * of mount options, we stash away the current values and
424 * restore them after we register the callbacks.
426 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
427 !spa_writeable(dmu_objset_spa(os))) {
429 do_readonly = B_TRUE;
430 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
432 do_readonly = B_TRUE;
434 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
438 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
441 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
446 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
449 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
453 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
456 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
460 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
463 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
469 * We need to enter pool configuration here, so that we can use
470 * dsl_prop_get_int_ds() to handle the special nbmand property below.
471 * dsl_prop_get_integer() can not be used, because it has to acquire
472 * spa_namespace_lock and we can not do that because we already hold
473 * z_teardown_lock. The problem is that spa_config_sync() is called
474 * with spa_namespace_lock held and the function calls ZFS vnode
475 * operations to write the cache file and thus z_teardown_lock is
476 * acquired after spa_namespace_lock.
478 ds = dmu_objset_ds(os);
479 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
482 * nbmand is a special property. It can only be changed at
485 * This is weird, but it is documented to only be changeable
488 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
490 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
492 } else if (error = dsl_prop_get_int_ds(ds, "nbmand", &nbmand) != 0) {
493 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
498 * Register property callbacks.
500 * It would probably be fine to just check for i/o error from
501 * the first prop_register(), but I guess I like to go
504 error = dsl_prop_register(ds,
505 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
506 error = error ? error : dsl_prop_register(ds,
507 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
508 error = error ? error : dsl_prop_register(ds,
509 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
510 error = error ? error : dsl_prop_register(ds,
511 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
513 error = error ? error : dsl_prop_register(ds,
514 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
516 error = error ? error : dsl_prop_register(ds,
517 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
518 error = error ? error : dsl_prop_register(ds,
519 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
520 error = error ? error : dsl_prop_register(ds,
521 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
522 error = error ? error : dsl_prop_register(ds,
523 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
524 error = error ? error : dsl_prop_register(ds,
525 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
527 error = error ? error : dsl_prop_register(ds,
528 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
529 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
534 * Invoke our callbacks to restore temporary mount options.
537 readonly_changed_cb(zfsvfs, readonly);
539 setuid_changed_cb(zfsvfs, setuid);
541 exec_changed_cb(zfsvfs, exec);
543 xattr_changed_cb(zfsvfs, xattr);
545 atime_changed_cb(zfsvfs, atime);
547 nbmand_changed_cb(zfsvfs, nbmand);
553 * We may attempt to unregister some callbacks that are not
554 * registered, but this is OK; it will simply return ENOMSG,
555 * which we will ignore.
557 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
558 atime_changed_cb, zfsvfs);
559 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
560 xattr_changed_cb, zfsvfs);
561 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
562 blksz_changed_cb, zfsvfs);
563 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
564 readonly_changed_cb, zfsvfs);
566 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
567 devices_changed_cb, zfsvfs);
569 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
570 setuid_changed_cb, zfsvfs);
571 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
572 exec_changed_cb, zfsvfs);
573 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
574 snapdir_changed_cb, zfsvfs);
575 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE),
576 acl_mode_changed_cb, zfsvfs);
577 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
578 acl_inherit_changed_cb, zfsvfs);
579 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
580 vscan_changed_cb, zfsvfs);
585 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
586 uint64_t *userp, uint64_t *groupp)
589 * Is it a valid type of object to track?
591 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
592 return (SET_ERROR(ENOENT));
595 * If we have a NULL data pointer
596 * then assume the id's aren't changing and
597 * return EEXIST to the dmu to let it know to
601 return (SET_ERROR(EEXIST));
603 if (bonustype == DMU_OT_ZNODE) {
604 znode_phys_t *znp = data;
605 *userp = znp->zp_uid;
606 *groupp = znp->zp_gid;
609 sa_hdr_phys_t *sap = data;
610 sa_hdr_phys_t sa = *sap;
611 boolean_t swap = B_FALSE;
613 ASSERT(bonustype == DMU_OT_SA);
615 if (sa.sa_magic == 0) {
617 * This should only happen for newly created
618 * files that haven't had the znode data filled
625 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
626 sa.sa_magic = SA_MAGIC;
627 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
630 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
633 hdrsize = sa_hdrsize(&sa);
634 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
635 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
637 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
640 *userp = BSWAP_64(*userp);
641 *groupp = BSWAP_64(*groupp);
648 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
649 char *domainbuf, int buflen, uid_t *ridp)
654 fuid = strtonum(fuidstr, NULL);
656 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
658 (void) strlcpy(domainbuf, domain, buflen);
661 *ridp = FUID_RID(fuid);
665 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
668 case ZFS_PROP_USERUSED:
669 return (DMU_USERUSED_OBJECT);
670 case ZFS_PROP_GROUPUSED:
671 return (DMU_GROUPUSED_OBJECT);
672 case ZFS_PROP_USERQUOTA:
673 return (zfsvfs->z_userquota_obj);
674 case ZFS_PROP_GROUPQUOTA:
675 return (zfsvfs->z_groupquota_obj);
681 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
682 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
687 zfs_useracct_t *buf = vbuf;
690 if (!dmu_objset_userspace_present(zfsvfs->z_os))
691 return (SET_ERROR(ENOTSUP));
693 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
699 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
700 (error = zap_cursor_retrieve(&zc, &za)) == 0;
701 zap_cursor_advance(&zc)) {
702 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
706 fuidstr_to_sid(zfsvfs, za.za_name,
707 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
709 buf->zu_space = za.za_first_integer;
715 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
716 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
717 *cookiep = zap_cursor_serialize(&zc);
718 zap_cursor_fini(&zc);
723 * buf must be big enough (eg, 32 bytes)
726 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
727 char *buf, boolean_t addok)
732 if (domain && domain[0]) {
733 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
735 return (SET_ERROR(ENOENT));
737 fuid = FUID_ENCODE(domainid, rid);
738 (void) sprintf(buf, "%llx", (longlong_t)fuid);
743 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
744 const char *domain, uint64_t rid, uint64_t *valp)
752 if (!dmu_objset_userspace_present(zfsvfs->z_os))
753 return (SET_ERROR(ENOTSUP));
755 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
759 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
763 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
770 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
771 const char *domain, uint64_t rid, uint64_t quota)
777 boolean_t fuid_dirtied;
779 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
780 return (SET_ERROR(EINVAL));
782 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
783 return (SET_ERROR(ENOTSUP));
785 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
786 &zfsvfs->z_groupquota_obj;
788 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
791 fuid_dirtied = zfsvfs->z_fuid_dirty;
793 tx = dmu_tx_create(zfsvfs->z_os);
794 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
796 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
797 zfs_userquota_prop_prefixes[type]);
800 zfs_fuid_txhold(zfsvfs, tx);
801 err = dmu_tx_assign(tx, TXG_WAIT);
807 mutex_enter(&zfsvfs->z_lock);
809 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
811 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
812 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
814 mutex_exit(&zfsvfs->z_lock);
817 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
821 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
825 zfs_fuid_sync(zfsvfs, tx);
831 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
834 uint64_t used, quota, usedobj, quotaobj;
837 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
838 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
840 if (quotaobj == 0 || zfsvfs->z_replay)
843 (void) sprintf(buf, "%llx", (longlong_t)fuid);
844 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
848 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
851 return (used >= quota);
855 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
860 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
862 fuid = isgroup ? zp->z_gid : zp->z_uid;
864 if (quotaobj == 0 || zfsvfs->z_replay)
867 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
871 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
879 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
882 * We claim to always be readonly so we can open snapshots;
883 * other ZPL code will prevent us from writing to snapshots.
885 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
887 kmem_free(zfsvfs, sizeof (zfsvfs_t));
892 * Initialize the zfs-specific filesystem structure.
893 * Should probably make this a kmem cache, shuffle fields,
894 * and just bzero up to z_hold_mtx[].
896 zfsvfs->z_vfs = NULL;
897 zfsvfs->z_parent = zfsvfs;
898 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
899 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
902 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
905 } else if (zfsvfs->z_version >
906 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
907 (void) printf("Can't mount a version %lld file system "
908 "on a version %lld pool\n. Pool must be upgraded to mount "
909 "this file system.", (u_longlong_t)zfsvfs->z_version,
910 (u_longlong_t)spa_version(dmu_objset_spa(os)));
911 error = SET_ERROR(ENOTSUP);
914 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
916 zfsvfs->z_norm = (int)zval;
918 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
920 zfsvfs->z_utf8 = (zval != 0);
922 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
924 zfsvfs->z_case = (uint_t)zval;
927 * Fold case on file systems that are always or sometimes case
930 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
931 zfsvfs->z_case == ZFS_CASE_MIXED)
932 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
934 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
935 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
937 if (zfsvfs->z_use_sa) {
938 /* should either have both of these objects or none */
939 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
945 * Pre SA versions file systems should never touch
946 * either the attribute registration or layout objects.
951 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
952 &zfsvfs->z_attr_table);
956 if (zfsvfs->z_version >= ZPL_VERSION_SA)
957 sa_register_update_callback(os, zfs_sa_upgrade);
959 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
963 ASSERT(zfsvfs->z_root != 0);
965 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
966 &zfsvfs->z_unlinkedobj);
970 error = zap_lookup(os, MASTER_NODE_OBJ,
971 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
972 8, 1, &zfsvfs->z_userquota_obj);
973 if (error && error != ENOENT)
976 error = zap_lookup(os, MASTER_NODE_OBJ,
977 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
978 8, 1, &zfsvfs->z_groupquota_obj);
979 if (error && error != ENOENT)
982 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
983 &zfsvfs->z_fuid_obj);
984 if (error && error != ENOENT)
987 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
988 &zfsvfs->z_shares_dir);
989 if (error && error != ENOENT)
992 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
993 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
994 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
995 offsetof(znode_t, z_link_node));
996 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
997 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
998 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
999 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1000 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1006 dmu_objset_disown(os, zfsvfs);
1008 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1013 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1017 error = zfs_register_callbacks(zfsvfs->z_vfs);
1022 * Set the objset user_ptr to track its zfsvfs.
1024 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1025 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1026 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1028 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1031 * If we are not mounting (ie: online recv), then we don't
1032 * have to worry about replaying the log as we blocked all
1033 * operations out since we closed the ZIL.
1039 * During replay we remove the read only flag to
1040 * allow replays to succeed.
1042 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1044 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1046 zfs_unlinked_drain(zfsvfs);
1049 * Parse and replay the intent log.
1051 * Because of ziltest, this must be done after
1052 * zfs_unlinked_drain(). (Further note: ziltest
1053 * doesn't use readonly mounts, where
1054 * zfs_unlinked_drain() isn't called.) This is because
1055 * ziltest causes spa_sync() to think it's committed,
1056 * but actually it is not, so the intent log contains
1057 * many txg's worth of changes.
1059 * In particular, if object N is in the unlinked set in
1060 * the last txg to actually sync, then it could be
1061 * actually freed in a later txg and then reallocated
1062 * in a yet later txg. This would write a "create
1063 * object N" record to the intent log. Normally, this
1064 * would be fine because the spa_sync() would have
1065 * written out the fact that object N is free, before
1066 * we could write the "create object N" intent log
1069 * But when we are in ziltest mode, we advance the "open
1070 * txg" without actually spa_sync()-ing the changes to
1071 * disk. So we would see that object N is still
1072 * allocated and in the unlinked set, and there is an
1073 * intent log record saying to allocate it.
1075 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1076 if (zil_replay_disable) {
1077 zil_destroy(zfsvfs->z_log, B_FALSE);
1079 zfsvfs->z_replay = B_TRUE;
1080 zil_replay(zfsvfs->z_os, zfsvfs,
1082 zfsvfs->z_replay = B_FALSE;
1085 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1091 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1094 zfsvfs_free(zfsvfs_t *zfsvfs)
1099 * This is a barrier to prevent the filesystem from going away in
1100 * zfs_znode_move() until we can safely ensure that the filesystem is
1101 * not unmounted. We consider the filesystem valid before the barrier
1102 * and invalid after the barrier.
1104 rw_enter(&zfsvfs_lock, RW_READER);
1105 rw_exit(&zfsvfs_lock);
1107 zfs_fuid_destroy(zfsvfs);
1109 mutex_destroy(&zfsvfs->z_znodes_lock);
1110 mutex_destroy(&zfsvfs->z_lock);
1111 list_destroy(&zfsvfs->z_all_znodes);
1112 rrm_destroy(&zfsvfs->z_teardown_lock);
1113 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1114 rw_destroy(&zfsvfs->z_fuid_lock);
1115 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1116 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1117 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1121 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1123 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1124 if (zfsvfs->z_vfs) {
1125 if (zfsvfs->z_use_fuids) {
1126 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1127 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1128 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1129 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1130 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1131 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1133 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1134 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1135 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1136 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1137 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1138 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1141 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1145 zfs_domount(vfs_t *vfsp, char *osname)
1147 uint64_t recordsize, fsid_guid;
1155 error = zfsvfs_create(osname, &zfsvfs);
1158 zfsvfs->z_vfs = vfsp;
1161 /* Initialize the generic filesystem structure. */
1162 vfsp->vfs_bcount = 0;
1163 vfsp->vfs_data = NULL;
1165 if (zfs_create_unique_device(&mount_dev) == -1) {
1166 error = SET_ERROR(ENODEV);
1169 ASSERT(vfs_devismounted(mount_dev) == 0);
1172 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1175 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1176 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1178 vfsp->vfs_data = zfsvfs;
1179 vfsp->mnt_flag |= MNT_LOCAL;
1180 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1181 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1182 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1185 * The fsid is 64 bits, composed of an 8-bit fs type, which
1186 * separates our fsid from any other filesystem types, and a
1187 * 56-bit objset unique ID. The objset unique ID is unique to
1188 * all objsets open on this system, provided by unique_create().
1189 * The 8-bit fs type must be put in the low bits of fsid[1]
1190 * because that's where other Solaris filesystems put it.
1192 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1193 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1194 vfsp->vfs_fsid.val[0] = fsid_guid;
1195 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1196 vfsp->mnt_vfc->vfc_typenum & 0xFF;
1199 * Set features for file system.
1201 zfs_set_fuid_feature(zfsvfs);
1202 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1203 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1204 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1205 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1206 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1207 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1208 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1210 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1212 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1215 atime_changed_cb(zfsvfs, B_FALSE);
1216 readonly_changed_cb(zfsvfs, B_TRUE);
1217 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1219 xattr_changed_cb(zfsvfs, pval);
1220 zfsvfs->z_issnap = B_TRUE;
1221 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1223 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1224 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1225 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1227 error = zfsvfs_setup(zfsvfs, B_TRUE);
1230 vfs_mountedfrom(vfsp, osname);
1232 if (!zfsvfs->z_issnap)
1233 zfsctl_create(zfsvfs);
1236 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1237 zfsvfs_free(zfsvfs);
1239 atomic_inc_32(&zfs_active_fs_count);
1246 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1248 objset_t *os = zfsvfs->z_os;
1249 struct dsl_dataset *ds;
1252 * Unregister properties.
1254 if (!dmu_objset_is_snapshot(os)) {
1255 ds = dmu_objset_ds(os);
1256 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1259 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1262 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1265 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1268 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1271 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1274 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1277 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
1280 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1281 acl_inherit_changed_cb, zfsvfs) == 0);
1283 VERIFY(dsl_prop_unregister(ds, "vscan",
1284 vscan_changed_cb, zfsvfs) == 0);
1290 * Convert a decimal digit string to a uint64_t integer.
1293 str_to_uint64(char *str, uint64_t *objnum)
1298 if (*str < '0' || *str > '9')
1299 return (SET_ERROR(EINVAL));
1301 num = num*10 + *str++ - '0';
1309 * The boot path passed from the boot loader is in the form of
1310 * "rootpool-name/root-filesystem-object-number'. Convert this
1311 * string to a dataset name: "rootpool-name/root-filesystem-name".
1314 zfs_parse_bootfs(char *bpath, char *outpath)
1320 if (*bpath == 0 || *bpath == '/')
1321 return (SET_ERROR(EINVAL));
1323 (void) strcpy(outpath, bpath);
1325 slashp = strchr(bpath, '/');
1327 /* if no '/', just return the pool name */
1328 if (slashp == NULL) {
1332 /* if not a number, just return the root dataset name */
1333 if (str_to_uint64(slashp+1, &objnum)) {
1338 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1345 * Check that the hex label string is appropriate for the dataset being
1346 * mounted into the global_zone proper.
1348 * Return an error if the hex label string is not default or
1349 * admin_low/admin_high. For admin_low labels, the corresponding
1350 * dataset must be readonly.
1353 zfs_check_global_label(const char *dsname, const char *hexsl)
1355 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1357 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1359 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1360 /* must be readonly */
1363 if (dsl_prop_get_integer(dsname,
1364 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1365 return (SET_ERROR(EACCES));
1366 return (rdonly ? 0 : EACCES);
1368 return (SET_ERROR(EACCES));
1372 * Determine whether the mount is allowed according to MAC check.
1373 * by comparing (where appropriate) label of the dataset against
1374 * the label of the zone being mounted into. If the dataset has
1375 * no label, create one.
1377 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1380 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1383 zone_t *mntzone = NULL;
1384 ts_label_t *mnt_tsl;
1387 char ds_hexsl[MAXNAMELEN];
1389 retv = EACCES; /* assume the worst */
1392 * Start by getting the dataset label if it exists.
1394 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1395 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1397 return (SET_ERROR(EACCES));
1400 * If labeling is NOT enabled, then disallow the mount of datasets
1401 * which have a non-default label already. No other label checks
1404 if (!is_system_labeled()) {
1405 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1407 return (SET_ERROR(EACCES));
1411 * Get the label of the mountpoint. If mounting into the global
1412 * zone (i.e. mountpoint is not within an active zone and the
1413 * zoned property is off), the label must be default or
1414 * admin_low/admin_high only; no other checks are needed.
1416 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1417 if (mntzone->zone_id == GLOBAL_ZONEID) {
1422 if (dsl_prop_get_integer(osname,
1423 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1424 return (SET_ERROR(EACCES));
1426 return (zfs_check_global_label(osname, ds_hexsl));
1429 * This is the case of a zone dataset being mounted
1430 * initially, before the zone has been fully created;
1431 * allow this mount into global zone.
1436 mnt_tsl = mntzone->zone_slabel;
1437 ASSERT(mnt_tsl != NULL);
1438 label_hold(mnt_tsl);
1439 mnt_sl = label2bslabel(mnt_tsl);
1441 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1443 * The dataset doesn't have a real label, so fabricate one.
1447 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1448 dsl_prop_set_string(osname,
1449 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1450 ZPROP_SRC_LOCAL, str) == 0)
1453 kmem_free(str, strlen(str) + 1);
1454 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1456 * Now compare labels to complete the MAC check. If the
1457 * labels are equal then allow access. If the mountpoint
1458 * label dominates the dataset label, allow readonly access.
1459 * Otherwise, access is denied.
1461 if (blequal(mnt_sl, &ds_sl))
1463 else if (bldominates(mnt_sl, &ds_sl)) {
1464 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1469 label_rele(mnt_tsl);
1473 #endif /* SECLABEL */
1475 #ifdef OPENSOLARIS_MOUNTROOT
1477 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1480 static int zfsrootdone = 0;
1481 zfsvfs_t *zfsvfs = NULL;
1490 * The filesystem that we mount as root is defined in the
1491 * boot property "zfs-bootfs" with a format of
1492 * "poolname/root-dataset-objnum".
1494 if (why == ROOT_INIT) {
1496 return (SET_ERROR(EBUSY));
1498 * the process of doing a spa_load will require the
1499 * clock to be set before we could (for example) do
1500 * something better by looking at the timestamp on
1501 * an uberblock, so just set it to -1.
1505 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1506 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1508 return (SET_ERROR(EINVAL));
1510 zfs_devid = spa_get_bootprop("diskdevid");
1511 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1513 spa_free_bootprop(zfs_devid);
1515 spa_free_bootprop(zfs_bootfs);
1516 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1520 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1521 spa_free_bootprop(zfs_bootfs);
1522 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1527 spa_free_bootprop(zfs_bootfs);
1529 if (error = vfs_lock(vfsp))
1532 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1533 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1537 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1539 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1540 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1545 mutex_enter(&vp->v_lock);
1546 vp->v_flag |= VROOT;
1547 mutex_exit(&vp->v_lock);
1551 * Leave rootvp held. The root file system is never unmounted.
1554 vfs_add((struct vnode *)0, vfsp,
1555 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1559 } else if (why == ROOT_REMOUNT) {
1560 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1561 vfsp->vfs_flag |= VFS_REMOUNT;
1563 /* refresh mount options */
1564 zfs_unregister_callbacks(vfsp->vfs_data);
1565 return (zfs_register_callbacks(vfsp));
1567 } else if (why == ROOT_UNMOUNT) {
1568 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1569 (void) zfs_sync(vfsp, 0, 0);
1574 * if "why" is equal to anything else other than ROOT_INIT,
1575 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1577 return (SET_ERROR(ENOTSUP));
1579 #endif /* OPENSOLARIS_MOUNTROOT */
1582 getpoolname(const char *osname, char *poolname)
1586 p = strchr(osname, '/');
1588 if (strlen(osname) >= MAXNAMELEN)
1589 return (ENAMETOOLONG);
1590 (void) strcpy(poolname, osname);
1592 if (p - osname >= MAXNAMELEN)
1593 return (ENAMETOOLONG);
1594 (void) strncpy(poolname, osname, p - osname);
1595 poolname[p - osname] = '\0';
1602 zfs_mount(vfs_t *vfsp)
1604 kthread_t *td = curthread;
1605 vnode_t *mvp = vfsp->mnt_vnodecovered;
1606 cred_t *cr = td->td_ucred;
1612 if (mvp->v_type != VDIR)
1613 return (SET_ERROR(ENOTDIR));
1615 mutex_enter(&mvp->v_lock);
1616 if ((uap->flags & MS_REMOUNT) == 0 &&
1617 (uap->flags & MS_OVERLAY) == 0 &&
1618 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1619 mutex_exit(&mvp->v_lock);
1620 return (SET_ERROR(EBUSY));
1622 mutex_exit(&mvp->v_lock);
1625 * ZFS does not support passing unparsed data in via MS_DATA.
1626 * Users should use the MS_OPTIONSTR interface; this means
1627 * that all option parsing is already done and the options struct
1628 * can be interrogated.
1630 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1632 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS))
1633 return (SET_ERROR(EPERM));
1635 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1636 return (SET_ERROR(EINVAL));
1637 #endif /* ! illumos */
1640 * If full-owner-access is enabled and delegated administration is
1641 * turned on, we must set nosuid.
1643 if (zfs_super_owner &&
1644 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1645 secpolicy_fs_mount_clearopts(cr, vfsp);
1649 * Check for mount privilege?
1651 * If we don't have privilege then see if
1652 * we have local permission to allow it
1654 error = secpolicy_fs_mount(cr, mvp, vfsp);
1656 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1659 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1663 * Make sure user is the owner of the mount point
1664 * or has sufficient privileges.
1667 vattr.va_mask = AT_UID;
1669 vn_lock(mvp, LK_SHARED | LK_RETRY);
1670 if (VOP_GETATTR(mvp, &vattr, cr)) {
1675 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1676 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1683 secpolicy_fs_mount_clearopts(cr, vfsp);
1687 * Refuse to mount a filesystem if we are in a local zone and the
1688 * dataset is not visible.
1690 if (!INGLOBALZONE(curthread) &&
1691 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1692 error = SET_ERROR(EPERM);
1697 error = zfs_mount_label_policy(vfsp, osname);
1702 vfsp->vfs_flag |= MNT_NFS4ACLS;
1705 * When doing a remount, we simply refresh our temporary properties
1706 * according to those options set in the current VFS options.
1708 if (vfsp->vfs_flag & MS_REMOUNT) {
1709 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1712 * Refresh mount options with z_teardown_lock blocking I/O while
1713 * the filesystem is in an inconsistent state.
1714 * The lock also serializes this code with filesystem
1715 * manipulations between entry to zfs_suspend_fs() and return
1716 * from zfs_resume_fs().
1718 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1719 zfs_unregister_callbacks(zfsvfs);
1720 error = zfs_register_callbacks(vfsp);
1721 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1725 /* Initial root mount: try hard to import the requested root pool. */
1726 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1727 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1728 char pname[MAXNAMELEN];
1730 error = getpoolname(osname, pname);
1732 error = spa_import_rootpool(pname);
1737 error = zfs_domount(vfsp, osname);
1742 * Add an extra VFS_HOLD on our parent vfs so that it can't
1743 * disappear due to a forced unmount.
1745 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1746 VFS_HOLD(mvp->v_vfsp);
1754 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1756 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1757 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1759 statp->f_version = STATFS_VERSION;
1763 dmu_objset_space(zfsvfs->z_os,
1764 &refdbytes, &availbytes, &usedobjs, &availobjs);
1767 * The underlying storage pool actually uses multiple block sizes.
1768 * We report the fragsize as the smallest block size we support,
1769 * and we report our blocksize as the filesystem's maximum blocksize.
1771 statp->f_bsize = SPA_MINBLOCKSIZE;
1772 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1775 * The following report "total" blocks of various kinds in the
1776 * file system, but reported in terms of f_frsize - the
1780 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1781 statp->f_bfree = availbytes / statp->f_bsize;
1782 statp->f_bavail = statp->f_bfree; /* no root reservation */
1785 * statvfs() should really be called statufs(), because it assumes
1786 * static metadata. ZFS doesn't preallocate files, so the best
1787 * we can do is report the max that could possibly fit in f_files,
1788 * and that minus the number actually used in f_ffree.
1789 * For f_ffree, report the smaller of the number of object available
1790 * and the number of blocks (each object will take at least a block).
1792 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1793 statp->f_files = statp->f_ffree + usedobjs;
1796 * We're a zfs filesystem.
1798 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
1800 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
1801 sizeof(statp->f_mntfromname));
1802 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
1803 sizeof(statp->f_mntonname));
1805 statp->f_namemax = ZFS_MAXNAMELEN;
1812 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
1814 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1820 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1822 *vpp = ZTOV(rootzp);
1827 error = vn_lock(*vpp, flags);
1829 (*vpp)->v_vflag |= VV_ROOT;
1838 * Teardown the zfsvfs::z_os.
1840 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1841 * and 'z_teardown_inactive_lock' held.
1844 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1848 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1852 * We purge the parent filesystem's vfsp as the parent
1853 * filesystem and all of its snapshots have their vnode's
1854 * v_vfsp set to the parent's filesystem's vfsp. Note,
1855 * 'z_parent' is self referential for non-snapshots.
1857 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1858 #ifdef FREEBSD_NAMECACHE
1859 cache_purgevfs(zfsvfs->z_parent->z_vfs);
1864 * Close the zil. NB: Can't close the zil while zfs_inactive
1865 * threads are blocked as zil_close can call zfs_inactive.
1867 if (zfsvfs->z_log) {
1868 zil_close(zfsvfs->z_log);
1869 zfsvfs->z_log = NULL;
1872 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1875 * If we are not unmounting (ie: online recv) and someone already
1876 * unmounted this file system while we were doing the switcheroo,
1877 * or a reopen of z_os failed then just bail out now.
1879 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1880 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1881 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1882 return (SET_ERROR(EIO));
1886 * At this point there are no vops active, and any new vops will
1887 * fail with EIO since we have z_teardown_lock for writer (only
1888 * relavent for forced unmount).
1890 * Release all holds on dbufs.
1892 mutex_enter(&zfsvfs->z_znodes_lock);
1893 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1894 zp = list_next(&zfsvfs->z_all_znodes, zp))
1896 ASSERT(ZTOV(zp)->v_count >= 0);
1897 zfs_znode_dmu_fini(zp);
1899 mutex_exit(&zfsvfs->z_znodes_lock);
1902 * If we are unmounting, set the unmounted flag and let new vops
1903 * unblock. zfs_inactive will have the unmounted behavior, and all
1904 * other vops will fail with EIO.
1907 zfsvfs->z_unmounted = B_TRUE;
1908 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1909 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1913 * z_os will be NULL if there was an error in attempting to reopen
1914 * zfsvfs, so just return as the properties had already been
1915 * unregistered and cached data had been evicted before.
1917 if (zfsvfs->z_os == NULL)
1921 * Unregister properties.
1923 zfs_unregister_callbacks(zfsvfs);
1928 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1929 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1930 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1931 dmu_objset_evict_dbufs(zfsvfs->z_os);
1938 zfs_umount(vfs_t *vfsp, int fflag)
1940 kthread_t *td = curthread;
1941 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1943 cred_t *cr = td->td_ucred;
1946 ret = secpolicy_fs_unmount(cr, vfsp);
1948 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1949 ZFS_DELEG_PERM_MOUNT, cr))
1954 * We purge the parent filesystem's vfsp as the parent filesystem
1955 * and all of its snapshots have their vnode's v_vfsp set to the
1956 * parent's filesystem's vfsp. Note, 'z_parent' is self
1957 * referential for non-snapshots.
1959 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1962 * Unmount any snapshots mounted under .zfs before unmounting the
1965 if (zfsvfs->z_ctldir != NULL) {
1966 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
1968 ret = vflush(vfsp, 0, 0, td);
1969 ASSERT(ret == EBUSY);
1970 if (!(fflag & MS_FORCE)) {
1971 if (zfsvfs->z_ctldir->v_count > 1)
1973 ASSERT(zfsvfs->z_ctldir->v_count == 1);
1975 zfsctl_destroy(zfsvfs);
1976 ASSERT(zfsvfs->z_ctldir == NULL);
1979 if (fflag & MS_FORCE) {
1981 * Mark file system as unmounted before calling
1982 * vflush(FORCECLOSE). This way we ensure no future vnops
1983 * will be called and risk operating on DOOMED vnodes.
1985 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1986 zfsvfs->z_unmounted = B_TRUE;
1987 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1991 * Flush all the files.
1993 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
1995 if (!zfsvfs->z_issnap) {
1996 zfsctl_create(zfsvfs);
1997 ASSERT(zfsvfs->z_ctldir != NULL);
2003 if (!(fflag & MS_FORCE)) {
2005 * Check the number of active vnodes in the file system.
2006 * Our count is maintained in the vfs structure, but the
2007 * number is off by 1 to indicate a hold on the vfs
2010 * The '.zfs' directory maintains a reference of its
2011 * own, and any active references underneath are
2012 * reflected in the vnode count.
2014 if (zfsvfs->z_ctldir == NULL) {
2015 if (vfsp->vfs_count > 1)
2016 return (SET_ERROR(EBUSY));
2018 if (vfsp->vfs_count > 2 ||
2019 zfsvfs->z_ctldir->v_count > 1)
2020 return (SET_ERROR(EBUSY));
2025 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2029 * z_os will be NULL if there was an error in
2030 * attempting to reopen zfsvfs.
2034 * Unset the objset user_ptr.
2036 mutex_enter(&os->os_user_ptr_lock);
2037 dmu_objset_set_user(os, NULL);
2038 mutex_exit(&os->os_user_ptr_lock);
2041 * Finally release the objset
2043 dmu_objset_disown(os, zfsvfs);
2047 * We can now safely destroy the '.zfs' directory node.
2049 if (zfsvfs->z_ctldir != NULL)
2050 zfsctl_destroy(zfsvfs);
2051 if (zfsvfs->z_issnap) {
2052 vnode_t *svp = vfsp->mnt_vnodecovered;
2054 if (svp->v_count >= 2)
2063 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2065 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2070 * zfs_zget() can't operate on virtual entries like .zfs/ or
2071 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2072 * This will make NFS to switch to LOOKUP instead of using VGET.
2074 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2075 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2076 return (EOPNOTSUPP);
2079 err = zfs_zget(zfsvfs, ino, &zp);
2080 if (err == 0 && zp->z_unlinked) {
2088 err = vn_lock(*vpp, flags);
2095 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
2096 struct ucred **credanonp, int *numsecflavors, int **secflavors)
2098 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2101 * If this is regular file system vfsp is the same as
2102 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2103 * zfsvfs->z_parent->z_vfs represents parent file system
2104 * which we have to use here, because only this file system
2105 * has mnt_export configured.
2107 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2108 credanonp, numsecflavors, secflavors));
2111 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2112 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2115 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2117 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2119 uint64_t object = 0;
2120 uint64_t fid_gen = 0;
2130 * On FreeBSD we can get snapshot's mount point or its parent file
2131 * system mount point depending if snapshot is already mounted or not.
2133 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2134 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2135 uint64_t objsetid = 0;
2136 uint64_t setgen = 0;
2138 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2139 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2141 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2142 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2146 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2148 return (SET_ERROR(EINVAL));
2152 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2153 zfid_short_t *zfid = (zfid_short_t *)fidp;
2155 for (i = 0; i < sizeof (zfid->zf_object); i++)
2156 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2158 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2159 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2162 return (SET_ERROR(EINVAL));
2166 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2167 * directory tree. If the object == zfsvfs->z_shares_dir, then
2168 * we are in the .zfs/shares directory tree.
2170 if ((fid_gen == 0 &&
2171 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2172 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2173 *vpp = zfsvfs->z_ctldir;
2174 ASSERT(*vpp != NULL);
2175 if (object == ZFSCTL_INO_SNAPDIR) {
2176 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2177 0, NULL, NULL, NULL, NULL, NULL) == 0);
2178 } else if (object == zfsvfs->z_shares_dir) {
2179 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL,
2180 0, NULL, NULL, NULL, NULL, NULL) == 0);
2185 err = vn_lock(*vpp, flags);
2191 gen_mask = -1ULL >> (64 - 8 * i);
2193 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2194 if (err = zfs_zget(zfsvfs, object, &zp)) {
2198 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2200 zp_gen = zp_gen & gen_mask;
2203 if (zp->z_unlinked || zp_gen != fid_gen) {
2204 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2207 return (SET_ERROR(EINVAL));
2212 err = vn_lock(*vpp, flags | LK_RETRY);
2214 vnode_create_vobject(*vpp, zp->z_size, curthread);
2221 * Block out VOPs and close zfsvfs_t::z_os
2223 * Note, if successful, then we return with the 'z_teardown_lock' and
2224 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2225 * dataset and objset intact so that they can be atomically handed off during
2226 * a subsequent rollback or recv operation and the resume thereafter.
2229 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2233 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2240 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2241 * is an invariant across any of the operations that can be performed while the
2242 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2243 * are the same: the relevant objset and associated dataset are owned by
2244 * zfsvfs, held, and long held on entry.
2247 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2251 uint64_t sa_obj = 0;
2253 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2254 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2257 * We already own this, so just hold and rele it to update the
2258 * objset_t, as the one we had before may have been evicted.
2260 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os));
2261 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs);
2262 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset));
2263 dmu_objset_rele(zfsvfs->z_os, zfsvfs);
2266 * Make sure version hasn't changed
2269 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
2270 &zfsvfs->z_version);
2275 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2276 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2278 if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
2281 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2282 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
2285 if (zfsvfs->z_version >= ZPL_VERSION_SA)
2286 sa_register_update_callback(zfsvfs->z_os,
2289 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2291 zfs_set_fuid_feature(zfsvfs);
2294 * Attempt to re-establish all the active znodes with
2295 * their dbufs. If a zfs_rezget() fails, then we'll let
2296 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2297 * when they try to use their znode.
2299 mutex_enter(&zfsvfs->z_znodes_lock);
2300 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2301 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2302 (void) zfs_rezget(zp);
2304 mutex_exit(&zfsvfs->z_znodes_lock);
2307 /* release the VOPs */
2308 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2309 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2313 * Since we couldn't setup the sa framework, try to force
2314 * unmount this file system.
2316 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) {
2317 vfs_ref(zfsvfs->z_vfs);
2318 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2325 zfs_freevfs(vfs_t *vfsp)
2327 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2331 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2332 * from zfs_mount(). Release it here. If we came through
2333 * zfs_mountroot() instead, we didn't grab an extra hold, so
2334 * skip the VFS_RELE for rootvfs.
2336 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2337 VFS_RELE(zfsvfs->z_parent->z_vfs);
2340 zfsvfs_free(zfsvfs);
2342 atomic_dec_32(&zfs_active_fs_count);
2346 static int desiredvnodes_backup;
2350 zfs_vnodes_adjust(void)
2353 int newdesiredvnodes;
2355 desiredvnodes_backup = desiredvnodes;
2358 * We calculate newdesiredvnodes the same way it is done in
2359 * vntblinit(). If it is equal to desiredvnodes, it means that
2360 * it wasn't tuned by the administrator and we can tune it down.
2362 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 *
2363 vm_kmem_size / (5 * (sizeof(struct vm_object) +
2364 sizeof(struct vnode))));
2365 if (newdesiredvnodes == desiredvnodes)
2366 desiredvnodes = (3 * newdesiredvnodes) / 4;
2371 zfs_vnodes_adjust_back(void)
2375 desiredvnodes = desiredvnodes_backup;
2383 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2386 * Initialize .zfs directory structures
2391 * Initialize znode cache, vnode ops, etc...
2396 * Reduce number of vnodes. Originally number of vnodes is calculated
2397 * with UFS inode in mind. We reduce it here, because it's too big for
2400 zfs_vnodes_adjust();
2402 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2410 zfs_vnodes_adjust_back();
2416 return (zfs_active_fs_count != 0);
2420 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2423 objset_t *os = zfsvfs->z_os;
2426 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2427 return (SET_ERROR(EINVAL));
2429 if (newvers < zfsvfs->z_version)
2430 return (SET_ERROR(EINVAL));
2432 if (zfs_spa_version_map(newvers) >
2433 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2434 return (SET_ERROR(ENOTSUP));
2436 tx = dmu_tx_create(os);
2437 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2438 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2439 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2441 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2443 error = dmu_tx_assign(tx, TXG_WAIT);
2449 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2450 8, 1, &newvers, tx);
2457 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2460 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2462 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2463 DMU_OT_NONE, 0, tx);
2465 error = zap_add(os, MASTER_NODE_OBJ,
2466 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2469 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2470 sa_register_update_callback(os, zfs_sa_upgrade);
2473 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2474 "from %llu to %llu", zfsvfs->z_version, newvers);
2478 zfsvfs->z_version = newvers;
2480 zfs_set_fuid_feature(zfsvfs);
2486 * Read a property stored within the master node.
2489 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2495 * Look up the file system's value for the property. For the
2496 * version property, we look up a slightly different string.
2498 if (prop == ZFS_PROP_VERSION)
2499 pname = ZPL_VERSION_STR;
2501 pname = zfs_prop_to_name(prop);
2504 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2506 if (error == ENOENT) {
2507 /* No value set, use the default value */
2509 case ZFS_PROP_VERSION:
2510 *value = ZPL_VERSION;
2512 case ZFS_PROP_NORMALIZE:
2513 case ZFS_PROP_UTF8ONLY:
2517 *value = ZFS_CASE_SENSITIVE;
2529 zfsvfs_update_fromname(const char *oldname, const char *newname)
2531 char tmpbuf[MAXPATHLEN];
2536 oldlen = strlen(oldname);
2538 mtx_lock(&mountlist_mtx);
2539 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2540 fromname = mp->mnt_stat.f_mntfromname;
2541 if (strcmp(fromname, oldname) == 0) {
2542 (void)strlcpy(fromname, newname,
2543 sizeof(mp->mnt_stat.f_mntfromname));
2546 if (strncmp(fromname, oldname, oldlen) == 0 &&
2547 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2548 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
2549 newname, fromname + oldlen);
2550 (void)strlcpy(fromname, tmpbuf,
2551 sizeof(mp->mnt_stat.f_mntfromname));
2555 mtx_unlock(&mountlist_mtx);