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 * nbmand is a special property. It can only be changed at
472 * This is weird, but it is documented to only be changeable
475 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
477 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
480 char osname[MAXNAMELEN];
482 dmu_objset_name(os, osname);
483 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
490 * Register property callbacks.
492 * It would probably be fine to just check for i/o error from
493 * the first prop_register(), but I guess I like to go
496 ds = dmu_objset_ds(os);
497 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
498 error = dsl_prop_register(ds,
499 zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
500 error = error ? error : dsl_prop_register(ds,
501 zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
502 error = error ? error : dsl_prop_register(ds,
503 zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
504 error = error ? error : dsl_prop_register(ds,
505 zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
507 error = error ? error : dsl_prop_register(ds,
508 zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
510 error = error ? error : dsl_prop_register(ds,
511 zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
512 error = error ? error : dsl_prop_register(ds,
513 zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
514 error = error ? error : dsl_prop_register(ds,
515 zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
516 error = error ? error : dsl_prop_register(ds,
517 zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
518 error = error ? error : dsl_prop_register(ds,
519 zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
521 error = error ? error : dsl_prop_register(ds,
522 zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
523 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
528 * Invoke our callbacks to restore temporary mount options.
531 readonly_changed_cb(zfsvfs, readonly);
533 setuid_changed_cb(zfsvfs, setuid);
535 exec_changed_cb(zfsvfs, exec);
537 xattr_changed_cb(zfsvfs, xattr);
539 atime_changed_cb(zfsvfs, atime);
541 nbmand_changed_cb(zfsvfs, nbmand);
547 * We may attempt to unregister some callbacks that are not
548 * registered, but this is OK; it will simply return ENOMSG,
549 * which we will ignore.
551 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
552 atime_changed_cb, zfsvfs);
553 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
554 xattr_changed_cb, zfsvfs);
555 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
556 blksz_changed_cb, zfsvfs);
557 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
558 readonly_changed_cb, zfsvfs);
560 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
561 devices_changed_cb, zfsvfs);
563 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
564 setuid_changed_cb, zfsvfs);
565 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
566 exec_changed_cb, zfsvfs);
567 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
568 snapdir_changed_cb, zfsvfs);
569 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLMODE),
570 acl_mode_changed_cb, zfsvfs);
571 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
572 acl_inherit_changed_cb, zfsvfs);
573 (void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
574 vscan_changed_cb, zfsvfs);
579 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
580 uint64_t *userp, uint64_t *groupp)
583 * Is it a valid type of object to track?
585 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
586 return (SET_ERROR(ENOENT));
589 * If we have a NULL data pointer
590 * then assume the id's aren't changing and
591 * return EEXIST to the dmu to let it know to
595 return (SET_ERROR(EEXIST));
597 if (bonustype == DMU_OT_ZNODE) {
598 znode_phys_t *znp = data;
599 *userp = znp->zp_uid;
600 *groupp = znp->zp_gid;
603 sa_hdr_phys_t *sap = data;
604 sa_hdr_phys_t sa = *sap;
605 boolean_t swap = B_FALSE;
607 ASSERT(bonustype == DMU_OT_SA);
609 if (sa.sa_magic == 0) {
611 * This should only happen for newly created
612 * files that haven't had the znode data filled
619 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
620 sa.sa_magic = SA_MAGIC;
621 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
624 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
627 hdrsize = sa_hdrsize(&sa);
628 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
629 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
631 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
634 *userp = BSWAP_64(*userp);
635 *groupp = BSWAP_64(*groupp);
642 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
643 char *domainbuf, int buflen, uid_t *ridp)
648 fuid = strtonum(fuidstr, NULL);
650 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
652 (void) strlcpy(domainbuf, domain, buflen);
655 *ridp = FUID_RID(fuid);
659 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
662 case ZFS_PROP_USERUSED:
663 return (DMU_USERUSED_OBJECT);
664 case ZFS_PROP_GROUPUSED:
665 return (DMU_GROUPUSED_OBJECT);
666 case ZFS_PROP_USERQUOTA:
667 return (zfsvfs->z_userquota_obj);
668 case ZFS_PROP_GROUPQUOTA:
669 return (zfsvfs->z_groupquota_obj);
675 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
676 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
681 zfs_useracct_t *buf = vbuf;
684 if (!dmu_objset_userspace_present(zfsvfs->z_os))
685 return (SET_ERROR(ENOTSUP));
687 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
693 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
694 (error = zap_cursor_retrieve(&zc, &za)) == 0;
695 zap_cursor_advance(&zc)) {
696 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
700 fuidstr_to_sid(zfsvfs, za.za_name,
701 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
703 buf->zu_space = za.za_first_integer;
709 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
710 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
711 *cookiep = zap_cursor_serialize(&zc);
712 zap_cursor_fini(&zc);
717 * buf must be big enough (eg, 32 bytes)
720 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
721 char *buf, boolean_t addok)
726 if (domain && domain[0]) {
727 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
729 return (SET_ERROR(ENOENT));
731 fuid = FUID_ENCODE(domainid, rid);
732 (void) sprintf(buf, "%llx", (longlong_t)fuid);
737 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
738 const char *domain, uint64_t rid, uint64_t *valp)
746 if (!dmu_objset_userspace_present(zfsvfs->z_os))
747 return (SET_ERROR(ENOTSUP));
749 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
753 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
757 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
764 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
765 const char *domain, uint64_t rid, uint64_t quota)
771 boolean_t fuid_dirtied;
773 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
774 return (SET_ERROR(EINVAL));
776 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
777 return (SET_ERROR(ENOTSUP));
779 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
780 &zfsvfs->z_groupquota_obj;
782 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
785 fuid_dirtied = zfsvfs->z_fuid_dirty;
787 tx = dmu_tx_create(zfsvfs->z_os);
788 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
790 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
791 zfs_userquota_prop_prefixes[type]);
794 zfs_fuid_txhold(zfsvfs, tx);
795 err = dmu_tx_assign(tx, TXG_WAIT);
801 mutex_enter(&zfsvfs->z_lock);
803 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
805 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
806 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
808 mutex_exit(&zfsvfs->z_lock);
811 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
815 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
819 zfs_fuid_sync(zfsvfs, tx);
825 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
828 uint64_t used, quota, usedobj, quotaobj;
831 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
832 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
834 if (quotaobj == 0 || zfsvfs->z_replay)
837 (void) sprintf(buf, "%llx", (longlong_t)fuid);
838 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
842 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
845 return (used >= quota);
849 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
854 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
856 fuid = isgroup ? zp->z_gid : zp->z_uid;
858 if (quotaobj == 0 || zfsvfs->z_replay)
861 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
865 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
873 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
876 * We claim to always be readonly so we can open snapshots;
877 * other ZPL code will prevent us from writing to snapshots.
879 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
881 kmem_free(zfsvfs, sizeof (zfsvfs_t));
886 * Initialize the zfs-specific filesystem structure.
887 * Should probably make this a kmem cache, shuffle fields,
888 * and just bzero up to z_hold_mtx[].
890 zfsvfs->z_vfs = NULL;
891 zfsvfs->z_parent = zfsvfs;
892 zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
893 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
896 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
899 } else if (zfsvfs->z_version >
900 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
901 (void) printf("Can't mount a version %lld file system "
902 "on a version %lld pool\n. Pool must be upgraded to mount "
903 "this file system.", (u_longlong_t)zfsvfs->z_version,
904 (u_longlong_t)spa_version(dmu_objset_spa(os)));
905 error = SET_ERROR(ENOTSUP);
908 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
910 zfsvfs->z_norm = (int)zval;
912 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
914 zfsvfs->z_utf8 = (zval != 0);
916 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
918 zfsvfs->z_case = (uint_t)zval;
921 * Fold case on file systems that are always or sometimes case
924 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
925 zfsvfs->z_case == ZFS_CASE_MIXED)
926 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
928 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
929 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
931 if (zfsvfs->z_use_sa) {
932 /* should either have both of these objects or none */
933 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
939 * Pre SA versions file systems should never touch
940 * either the attribute registration or layout objects.
945 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
946 &zfsvfs->z_attr_table);
950 if (zfsvfs->z_version >= ZPL_VERSION_SA)
951 sa_register_update_callback(os, zfs_sa_upgrade);
953 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
957 ASSERT(zfsvfs->z_root != 0);
959 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
960 &zfsvfs->z_unlinkedobj);
964 error = zap_lookup(os, MASTER_NODE_OBJ,
965 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
966 8, 1, &zfsvfs->z_userquota_obj);
967 if (error && error != ENOENT)
970 error = zap_lookup(os, MASTER_NODE_OBJ,
971 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
972 8, 1, &zfsvfs->z_groupquota_obj);
973 if (error && error != ENOENT)
976 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
977 &zfsvfs->z_fuid_obj);
978 if (error && error != ENOENT)
981 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
982 &zfsvfs->z_shares_dir);
983 if (error && error != ENOENT)
986 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
987 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
988 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
989 offsetof(znode_t, z_link_node));
990 rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
991 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
992 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
993 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
994 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1000 dmu_objset_disown(os, zfsvfs);
1002 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1007 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1011 error = zfs_register_callbacks(zfsvfs->z_vfs);
1016 * Set the objset user_ptr to track its zfsvfs.
1018 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1019 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1020 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1022 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1025 * If we are not mounting (ie: online recv), then we don't
1026 * have to worry about replaying the log as we blocked all
1027 * operations out since we closed the ZIL.
1033 * During replay we remove the read only flag to
1034 * allow replays to succeed.
1036 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1038 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1040 zfs_unlinked_drain(zfsvfs);
1043 * Parse and replay the intent log.
1045 * Because of ziltest, this must be done after
1046 * zfs_unlinked_drain(). (Further note: ziltest
1047 * doesn't use readonly mounts, where
1048 * zfs_unlinked_drain() isn't called.) This is because
1049 * ziltest causes spa_sync() to think it's committed,
1050 * but actually it is not, so the intent log contains
1051 * many txg's worth of changes.
1053 * In particular, if object N is in the unlinked set in
1054 * the last txg to actually sync, then it could be
1055 * actually freed in a later txg and then reallocated
1056 * in a yet later txg. This would write a "create
1057 * object N" record to the intent log. Normally, this
1058 * would be fine because the spa_sync() would have
1059 * written out the fact that object N is free, before
1060 * we could write the "create object N" intent log
1063 * But when we are in ziltest mode, we advance the "open
1064 * txg" without actually spa_sync()-ing the changes to
1065 * disk. So we would see that object N is still
1066 * allocated and in the unlinked set, and there is an
1067 * intent log record saying to allocate it.
1069 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1070 if (zil_replay_disable) {
1071 zil_destroy(zfsvfs->z_log, B_FALSE);
1073 zfsvfs->z_replay = B_TRUE;
1074 zil_replay(zfsvfs->z_os, zfsvfs,
1076 zfsvfs->z_replay = B_FALSE;
1079 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1085 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1088 zfsvfs_free(zfsvfs_t *zfsvfs)
1093 * This is a barrier to prevent the filesystem from going away in
1094 * zfs_znode_move() until we can safely ensure that the filesystem is
1095 * not unmounted. We consider the filesystem valid before the barrier
1096 * and invalid after the barrier.
1098 rw_enter(&zfsvfs_lock, RW_READER);
1099 rw_exit(&zfsvfs_lock);
1101 zfs_fuid_destroy(zfsvfs);
1103 mutex_destroy(&zfsvfs->z_znodes_lock);
1104 mutex_destroy(&zfsvfs->z_lock);
1105 list_destroy(&zfsvfs->z_all_znodes);
1106 rrm_destroy(&zfsvfs->z_teardown_lock);
1107 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1108 rw_destroy(&zfsvfs->z_fuid_lock);
1109 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1110 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1111 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1115 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1117 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1118 if (zfsvfs->z_vfs) {
1119 if (zfsvfs->z_use_fuids) {
1120 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1121 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1122 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1123 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1124 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1125 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1127 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1128 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1129 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1130 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1131 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1132 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1135 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1139 zfs_domount(vfs_t *vfsp, char *osname)
1141 uint64_t recordsize, fsid_guid;
1149 error = zfsvfs_create(osname, &zfsvfs);
1152 zfsvfs->z_vfs = vfsp;
1155 /* Initialize the generic filesystem structure. */
1156 vfsp->vfs_bcount = 0;
1157 vfsp->vfs_data = NULL;
1159 if (zfs_create_unique_device(&mount_dev) == -1) {
1160 error = SET_ERROR(ENODEV);
1163 ASSERT(vfs_devismounted(mount_dev) == 0);
1166 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1169 zfsvfs->z_vfs->vfs_bsize = SPA_MINBLOCKSIZE;
1170 zfsvfs->z_vfs->mnt_stat.f_iosize = recordsize;
1172 vfsp->vfs_data = zfsvfs;
1173 vfsp->mnt_flag |= MNT_LOCAL;
1174 vfsp->mnt_kern_flag |= MNTK_LOOKUP_SHARED;
1175 vfsp->mnt_kern_flag |= MNTK_SHARED_WRITES;
1176 vfsp->mnt_kern_flag |= MNTK_EXTENDED_SHARED;
1179 * The fsid is 64 bits, composed of an 8-bit fs type, which
1180 * separates our fsid from any other filesystem types, and a
1181 * 56-bit objset unique ID. The objset unique ID is unique to
1182 * all objsets open on this system, provided by unique_create().
1183 * The 8-bit fs type must be put in the low bits of fsid[1]
1184 * because that's where other Solaris filesystems put it.
1186 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1187 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1188 vfsp->vfs_fsid.val[0] = fsid_guid;
1189 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1190 vfsp->mnt_vfc->vfc_typenum & 0xFF;
1193 * Set features for file system.
1195 zfs_set_fuid_feature(zfsvfs);
1196 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1197 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1198 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1199 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1200 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1201 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1202 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1204 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1206 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1209 atime_changed_cb(zfsvfs, B_FALSE);
1210 readonly_changed_cb(zfsvfs, B_TRUE);
1211 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1213 xattr_changed_cb(zfsvfs, pval);
1214 zfsvfs->z_issnap = B_TRUE;
1215 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1217 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1218 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1219 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1221 error = zfsvfs_setup(zfsvfs, B_TRUE);
1224 vfs_mountedfrom(vfsp, osname);
1226 if (!zfsvfs->z_issnap)
1227 zfsctl_create(zfsvfs);
1230 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1231 zfsvfs_free(zfsvfs);
1233 atomic_inc_32(&zfs_active_fs_count);
1240 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1242 objset_t *os = zfsvfs->z_os;
1243 struct dsl_dataset *ds;
1246 * Unregister properties.
1248 if (!dmu_objset_is_snapshot(os)) {
1249 ds = dmu_objset_ds(os);
1250 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1253 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1256 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1259 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1262 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1265 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1268 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1271 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
1274 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1275 acl_inherit_changed_cb, zfsvfs) == 0);
1277 VERIFY(dsl_prop_unregister(ds, "vscan",
1278 vscan_changed_cb, zfsvfs) == 0);
1284 * Convert a decimal digit string to a uint64_t integer.
1287 str_to_uint64(char *str, uint64_t *objnum)
1292 if (*str < '0' || *str > '9')
1293 return (SET_ERROR(EINVAL));
1295 num = num*10 + *str++ - '0';
1303 * The boot path passed from the boot loader is in the form of
1304 * "rootpool-name/root-filesystem-object-number'. Convert this
1305 * string to a dataset name: "rootpool-name/root-filesystem-name".
1308 zfs_parse_bootfs(char *bpath, char *outpath)
1314 if (*bpath == 0 || *bpath == '/')
1315 return (SET_ERROR(EINVAL));
1317 (void) strcpy(outpath, bpath);
1319 slashp = strchr(bpath, '/');
1321 /* if no '/', just return the pool name */
1322 if (slashp == NULL) {
1326 /* if not a number, just return the root dataset name */
1327 if (str_to_uint64(slashp+1, &objnum)) {
1332 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1339 * Check that the hex label string is appropriate for the dataset being
1340 * mounted into the global_zone proper.
1342 * Return an error if the hex label string is not default or
1343 * admin_low/admin_high. For admin_low labels, the corresponding
1344 * dataset must be readonly.
1347 zfs_check_global_label(const char *dsname, const char *hexsl)
1349 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1351 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1353 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1354 /* must be readonly */
1357 if (dsl_prop_get_integer(dsname,
1358 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1359 return (SET_ERROR(EACCES));
1360 return (rdonly ? 0 : EACCES);
1362 return (SET_ERROR(EACCES));
1366 * Determine whether the mount is allowed according to MAC check.
1367 * by comparing (where appropriate) label of the dataset against
1368 * the label of the zone being mounted into. If the dataset has
1369 * no label, create one.
1371 * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1374 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1377 zone_t *mntzone = NULL;
1378 ts_label_t *mnt_tsl;
1381 char ds_hexsl[MAXNAMELEN];
1383 retv = EACCES; /* assume the worst */
1386 * Start by getting the dataset label if it exists.
1388 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1389 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1391 return (SET_ERROR(EACCES));
1394 * If labeling is NOT enabled, then disallow the mount of datasets
1395 * which have a non-default label already. No other label checks
1398 if (!is_system_labeled()) {
1399 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1401 return (SET_ERROR(EACCES));
1405 * Get the label of the mountpoint. If mounting into the global
1406 * zone (i.e. mountpoint is not within an active zone and the
1407 * zoned property is off), the label must be default or
1408 * admin_low/admin_high only; no other checks are needed.
1410 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1411 if (mntzone->zone_id == GLOBAL_ZONEID) {
1416 if (dsl_prop_get_integer(osname,
1417 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1418 return (SET_ERROR(EACCES));
1420 return (zfs_check_global_label(osname, ds_hexsl));
1423 * This is the case of a zone dataset being mounted
1424 * initially, before the zone has been fully created;
1425 * allow this mount into global zone.
1430 mnt_tsl = mntzone->zone_slabel;
1431 ASSERT(mnt_tsl != NULL);
1432 label_hold(mnt_tsl);
1433 mnt_sl = label2bslabel(mnt_tsl);
1435 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1437 * The dataset doesn't have a real label, so fabricate one.
1441 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1442 dsl_prop_set_string(osname,
1443 zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1444 ZPROP_SRC_LOCAL, str) == 0)
1447 kmem_free(str, strlen(str) + 1);
1448 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1450 * Now compare labels to complete the MAC check. If the
1451 * labels are equal then allow access. If the mountpoint
1452 * label dominates the dataset label, allow readonly access.
1453 * Otherwise, access is denied.
1455 if (blequal(mnt_sl, &ds_sl))
1457 else if (bldominates(mnt_sl, &ds_sl)) {
1458 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1463 label_rele(mnt_tsl);
1467 #endif /* SECLABEL */
1469 #ifdef OPENSOLARIS_MOUNTROOT
1471 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1474 static int zfsrootdone = 0;
1475 zfsvfs_t *zfsvfs = NULL;
1484 * The filesystem that we mount as root is defined in the
1485 * boot property "zfs-bootfs" with a format of
1486 * "poolname/root-dataset-objnum".
1488 if (why == ROOT_INIT) {
1490 return (SET_ERROR(EBUSY));
1492 * the process of doing a spa_load will require the
1493 * clock to be set before we could (for example) do
1494 * something better by looking at the timestamp on
1495 * an uberblock, so just set it to -1.
1499 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1500 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1502 return (SET_ERROR(EINVAL));
1504 zfs_devid = spa_get_bootprop("diskdevid");
1505 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1507 spa_free_bootprop(zfs_devid);
1509 spa_free_bootprop(zfs_bootfs);
1510 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1514 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1515 spa_free_bootprop(zfs_bootfs);
1516 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1521 spa_free_bootprop(zfs_bootfs);
1523 if (error = vfs_lock(vfsp))
1526 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1527 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1531 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1533 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1534 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1539 mutex_enter(&vp->v_lock);
1540 vp->v_flag |= VROOT;
1541 mutex_exit(&vp->v_lock);
1545 * Leave rootvp held. The root file system is never unmounted.
1548 vfs_add((struct vnode *)0, vfsp,
1549 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1553 } else if (why == ROOT_REMOUNT) {
1554 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1555 vfsp->vfs_flag |= VFS_REMOUNT;
1557 /* refresh mount options */
1558 zfs_unregister_callbacks(vfsp->vfs_data);
1559 return (zfs_register_callbacks(vfsp));
1561 } else if (why == ROOT_UNMOUNT) {
1562 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1563 (void) zfs_sync(vfsp, 0, 0);
1568 * if "why" is equal to anything else other than ROOT_INIT,
1569 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1571 return (SET_ERROR(ENOTSUP));
1573 #endif /* OPENSOLARIS_MOUNTROOT */
1576 getpoolname(const char *osname, char *poolname)
1580 p = strchr(osname, '/');
1582 if (strlen(osname) >= MAXNAMELEN)
1583 return (ENAMETOOLONG);
1584 (void) strcpy(poolname, osname);
1586 if (p - osname >= MAXNAMELEN)
1587 return (ENAMETOOLONG);
1588 (void) strncpy(poolname, osname, p - osname);
1589 poolname[p - osname] = '\0';
1596 zfs_mount(vfs_t *vfsp)
1598 kthread_t *td = curthread;
1599 vnode_t *mvp = vfsp->mnt_vnodecovered;
1600 cred_t *cr = td->td_ucred;
1606 if (mvp->v_type != VDIR)
1607 return (SET_ERROR(ENOTDIR));
1609 mutex_enter(&mvp->v_lock);
1610 if ((uap->flags & MS_REMOUNT) == 0 &&
1611 (uap->flags & MS_OVERLAY) == 0 &&
1612 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1613 mutex_exit(&mvp->v_lock);
1614 return (SET_ERROR(EBUSY));
1616 mutex_exit(&mvp->v_lock);
1619 * ZFS does not support passing unparsed data in via MS_DATA.
1620 * Users should use the MS_OPTIONSTR interface; this means
1621 * that all option parsing is already done and the options struct
1622 * can be interrogated.
1624 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1626 if (!prison_allow(td->td_ucred, PR_ALLOW_MOUNT_ZFS))
1627 return (SET_ERROR(EPERM));
1629 if (vfs_getopt(vfsp->mnt_optnew, "from", (void **)&osname, NULL))
1630 return (SET_ERROR(EINVAL));
1631 #endif /* ! illumos */
1634 * If full-owner-access is enabled and delegated administration is
1635 * turned on, we must set nosuid.
1637 if (zfs_super_owner &&
1638 dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != ECANCELED) {
1639 secpolicy_fs_mount_clearopts(cr, vfsp);
1643 * Check for mount privilege?
1645 * If we don't have privilege then see if
1646 * we have local permission to allow it
1648 error = secpolicy_fs_mount(cr, mvp, vfsp);
1650 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) != 0)
1653 if (!(vfsp->vfs_flag & MS_REMOUNT)) {
1657 * Make sure user is the owner of the mount point
1658 * or has sufficient privileges.
1661 vattr.va_mask = AT_UID;
1663 vn_lock(mvp, LK_SHARED | LK_RETRY);
1664 if (VOP_GETATTR(mvp, &vattr, cr)) {
1669 if (secpolicy_vnode_owner(mvp, cr, vattr.va_uid) != 0 &&
1670 VOP_ACCESS(mvp, VWRITE, cr, td) != 0) {
1677 secpolicy_fs_mount_clearopts(cr, vfsp);
1681 * Refuse to mount a filesystem if we are in a local zone and the
1682 * dataset is not visible.
1684 if (!INGLOBALZONE(curthread) &&
1685 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1686 error = SET_ERROR(EPERM);
1691 error = zfs_mount_label_policy(vfsp, osname);
1696 vfsp->vfs_flag |= MNT_NFS4ACLS;
1699 * When doing a remount, we simply refresh our temporary properties
1700 * according to those options set in the current VFS options.
1702 if (vfsp->vfs_flag & MS_REMOUNT) {
1703 /* refresh mount options */
1704 zfs_unregister_callbacks(vfsp->vfs_data);
1705 error = zfs_register_callbacks(vfsp);
1709 /* Initial root mount: try hard to import the requested root pool. */
1710 if ((vfsp->vfs_flag & MNT_ROOTFS) != 0 &&
1711 (vfsp->vfs_flag & MNT_UPDATE) == 0) {
1712 char pname[MAXNAMELEN];
1714 error = getpoolname(osname, pname);
1716 error = spa_import_rootpool(pname);
1721 error = zfs_domount(vfsp, osname);
1726 * Add an extra VFS_HOLD on our parent vfs so that it can't
1727 * disappear due to a forced unmount.
1729 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1730 VFS_HOLD(mvp->v_vfsp);
1738 zfs_statfs(vfs_t *vfsp, struct statfs *statp)
1740 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1741 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1743 statp->f_version = STATFS_VERSION;
1747 dmu_objset_space(zfsvfs->z_os,
1748 &refdbytes, &availbytes, &usedobjs, &availobjs);
1751 * The underlying storage pool actually uses multiple block sizes.
1752 * We report the fragsize as the smallest block size we support,
1753 * and we report our blocksize as the filesystem's maximum blocksize.
1755 statp->f_bsize = SPA_MINBLOCKSIZE;
1756 statp->f_iosize = zfsvfs->z_vfs->mnt_stat.f_iosize;
1759 * The following report "total" blocks of various kinds in the
1760 * file system, but reported in terms of f_frsize - the
1764 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1765 statp->f_bfree = availbytes / statp->f_bsize;
1766 statp->f_bavail = statp->f_bfree; /* no root reservation */
1769 * statvfs() should really be called statufs(), because it assumes
1770 * static metadata. ZFS doesn't preallocate files, so the best
1771 * we can do is report the max that could possibly fit in f_files,
1772 * and that minus the number actually used in f_ffree.
1773 * For f_ffree, report the smaller of the number of object available
1774 * and the number of blocks (each object will take at least a block).
1776 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1777 statp->f_files = statp->f_ffree + usedobjs;
1780 * We're a zfs filesystem.
1782 (void) strlcpy(statp->f_fstypename, "zfs", sizeof(statp->f_fstypename));
1784 strlcpy(statp->f_mntfromname, vfsp->mnt_stat.f_mntfromname,
1785 sizeof(statp->f_mntfromname));
1786 strlcpy(statp->f_mntonname, vfsp->mnt_stat.f_mntonname,
1787 sizeof(statp->f_mntonname));
1789 statp->f_namemax = ZFS_MAXNAMELEN;
1796 zfs_root(vfs_t *vfsp, int flags, vnode_t **vpp)
1798 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1804 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1806 *vpp = ZTOV(rootzp);
1811 error = vn_lock(*vpp, flags);
1813 (*vpp)->v_vflag |= VV_ROOT;
1822 * Teardown the zfsvfs::z_os.
1824 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1825 * and 'z_teardown_inactive_lock' held.
1828 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1832 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1836 * We purge the parent filesystem's vfsp as the parent
1837 * filesystem and all of its snapshots have their vnode's
1838 * v_vfsp set to the parent's filesystem's vfsp. Note,
1839 * 'z_parent' is self referential for non-snapshots.
1841 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1842 #ifdef FREEBSD_NAMECACHE
1843 cache_purgevfs(zfsvfs->z_parent->z_vfs);
1848 * Close the zil. NB: Can't close the zil while zfs_inactive
1849 * threads are blocked as zil_close can call zfs_inactive.
1851 if (zfsvfs->z_log) {
1852 zil_close(zfsvfs->z_log);
1853 zfsvfs->z_log = NULL;
1856 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1859 * If we are not unmounting (ie: online recv) and someone already
1860 * unmounted this file system while we were doing the switcheroo,
1861 * or a reopen of z_os failed then just bail out now.
1863 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1864 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1865 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1866 return (SET_ERROR(EIO));
1870 * At this point there are no vops active, and any new vops will
1871 * fail with EIO since we have z_teardown_lock for writer (only
1872 * relavent for forced unmount).
1874 * Release all holds on dbufs.
1876 mutex_enter(&zfsvfs->z_znodes_lock);
1877 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1878 zp = list_next(&zfsvfs->z_all_znodes, zp))
1880 ASSERT(ZTOV(zp)->v_count >= 0);
1881 zfs_znode_dmu_fini(zp);
1883 mutex_exit(&zfsvfs->z_znodes_lock);
1886 * If we are unmounting, set the unmounted flag and let new vops
1887 * unblock. zfs_inactive will have the unmounted behavior, and all
1888 * other vops will fail with EIO.
1891 zfsvfs->z_unmounted = B_TRUE;
1892 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1893 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1897 * z_os will be NULL if there was an error in attempting to reopen
1898 * zfsvfs, so just return as the properties had already been
1899 * unregistered and cached data had been evicted before.
1901 if (zfsvfs->z_os == NULL)
1905 * Unregister properties.
1907 zfs_unregister_callbacks(zfsvfs);
1912 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1913 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1914 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1915 dmu_objset_evict_dbufs(zfsvfs->z_os);
1922 zfs_umount(vfs_t *vfsp, int fflag)
1924 kthread_t *td = curthread;
1925 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1927 cred_t *cr = td->td_ucred;
1930 ret = secpolicy_fs_unmount(cr, vfsp);
1932 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1933 ZFS_DELEG_PERM_MOUNT, cr))
1938 * We purge the parent filesystem's vfsp as the parent filesystem
1939 * and all of its snapshots have their vnode's v_vfsp set to the
1940 * parent's filesystem's vfsp. Note, 'z_parent' is self
1941 * referential for non-snapshots.
1943 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1946 * Unmount any snapshots mounted under .zfs before unmounting the
1949 if (zfsvfs->z_ctldir != NULL) {
1950 if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
1952 ret = vflush(vfsp, 0, 0, td);
1953 ASSERT(ret == EBUSY);
1954 if (!(fflag & MS_FORCE)) {
1955 if (zfsvfs->z_ctldir->v_count > 1)
1957 ASSERT(zfsvfs->z_ctldir->v_count == 1);
1959 zfsctl_destroy(zfsvfs);
1960 ASSERT(zfsvfs->z_ctldir == NULL);
1963 if (fflag & MS_FORCE) {
1965 * Mark file system as unmounted before calling
1966 * vflush(FORCECLOSE). This way we ensure no future vnops
1967 * will be called and risk operating on DOOMED vnodes.
1969 rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1970 zfsvfs->z_unmounted = B_TRUE;
1971 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1975 * Flush all the files.
1977 ret = vflush(vfsp, 0, (fflag & MS_FORCE) ? FORCECLOSE : 0, td);
1979 if (!zfsvfs->z_issnap) {
1980 zfsctl_create(zfsvfs);
1981 ASSERT(zfsvfs->z_ctldir != NULL);
1987 if (!(fflag & MS_FORCE)) {
1989 * Check the number of active vnodes in the file system.
1990 * Our count is maintained in the vfs structure, but the
1991 * number is off by 1 to indicate a hold on the vfs
1994 * The '.zfs' directory maintains a reference of its
1995 * own, and any active references underneath are
1996 * reflected in the vnode count.
1998 if (zfsvfs->z_ctldir == NULL) {
1999 if (vfsp->vfs_count > 1)
2000 return (SET_ERROR(EBUSY));
2002 if (vfsp->vfs_count > 2 ||
2003 zfsvfs->z_ctldir->v_count > 1)
2004 return (SET_ERROR(EBUSY));
2009 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
2013 * z_os will be NULL if there was an error in
2014 * attempting to reopen zfsvfs.
2018 * Unset the objset user_ptr.
2020 mutex_enter(&os->os_user_ptr_lock);
2021 dmu_objset_set_user(os, NULL);
2022 mutex_exit(&os->os_user_ptr_lock);
2025 * Finally release the objset
2027 dmu_objset_disown(os, zfsvfs);
2031 * We can now safely destroy the '.zfs' directory node.
2033 if (zfsvfs->z_ctldir != NULL)
2034 zfsctl_destroy(zfsvfs);
2035 if (zfsvfs->z_issnap) {
2036 vnode_t *svp = vfsp->mnt_vnodecovered;
2038 if (svp->v_count >= 2)
2047 zfs_vget(vfs_t *vfsp, ino_t ino, int flags, vnode_t **vpp)
2049 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2054 * zfs_zget() can't operate on virtual entries like .zfs/ or
2055 * .zfs/snapshot/ directories, that's why we return EOPNOTSUPP.
2056 * This will make NFS to switch to LOOKUP instead of using VGET.
2058 if (ino == ZFSCTL_INO_ROOT || ino == ZFSCTL_INO_SNAPDIR ||
2059 (zfsvfs->z_shares_dir != 0 && ino == zfsvfs->z_shares_dir))
2060 return (EOPNOTSUPP);
2063 err = zfs_zget(zfsvfs, ino, &zp);
2064 if (err == 0 && zp->z_unlinked) {
2072 err = vn_lock(*vpp, flags);
2079 zfs_checkexp(vfs_t *vfsp, struct sockaddr *nam, int *extflagsp,
2080 struct ucred **credanonp, int *numsecflavors, int **secflavors)
2082 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2085 * If this is regular file system vfsp is the same as
2086 * zfsvfs->z_parent->z_vfs, but if it is snapshot,
2087 * zfsvfs->z_parent->z_vfs represents parent file system
2088 * which we have to use here, because only this file system
2089 * has mnt_export configured.
2091 return (vfs_stdcheckexp(zfsvfs->z_parent->z_vfs, nam, extflagsp,
2092 credanonp, numsecflavors, secflavors));
2095 CTASSERT(SHORT_FID_LEN <= sizeof(struct fid));
2096 CTASSERT(LONG_FID_LEN <= sizeof(struct fid));
2099 zfs_fhtovp(vfs_t *vfsp, fid_t *fidp, int flags, vnode_t **vpp)
2101 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2103 uint64_t object = 0;
2104 uint64_t fid_gen = 0;
2114 * On FreeBSD we can get snapshot's mount point or its parent file
2115 * system mount point depending if snapshot is already mounted or not.
2117 if (zfsvfs->z_parent == zfsvfs && fidp->fid_len == LONG_FID_LEN) {
2118 zfid_long_t *zlfid = (zfid_long_t *)fidp;
2119 uint64_t objsetid = 0;
2120 uint64_t setgen = 0;
2122 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
2123 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
2125 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
2126 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
2130 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
2132 return (SET_ERROR(EINVAL));
2136 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
2137 zfid_short_t *zfid = (zfid_short_t *)fidp;
2139 for (i = 0; i < sizeof (zfid->zf_object); i++)
2140 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
2142 for (i = 0; i < sizeof (zfid->zf_gen); i++)
2143 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
2146 return (SET_ERROR(EINVAL));
2150 * A zero fid_gen means we are in .zfs or the .zfs/snapshot
2151 * directory tree. If the object == zfsvfs->z_shares_dir, then
2152 * we are in the .zfs/shares directory tree.
2154 if ((fid_gen == 0 &&
2155 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) ||
2156 (zfsvfs->z_shares_dir != 0 && object == zfsvfs->z_shares_dir)) {
2157 *vpp = zfsvfs->z_ctldir;
2158 ASSERT(*vpp != NULL);
2159 if (object == ZFSCTL_INO_SNAPDIR) {
2160 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2161 0, NULL, NULL, NULL, NULL, NULL) == 0);
2162 } else if (object == zfsvfs->z_shares_dir) {
2163 VERIFY(zfsctl_root_lookup(*vpp, "shares", vpp, NULL,
2164 0, NULL, NULL, NULL, NULL, NULL) == 0);
2169 err = vn_lock(*vpp, flags);
2175 gen_mask = -1ULL >> (64 - 8 * i);
2177 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2178 if (err = zfs_zget(zfsvfs, object, &zp)) {
2182 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2184 zp_gen = zp_gen & gen_mask;
2187 if (zp->z_unlinked || zp_gen != fid_gen) {
2188 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2191 return (SET_ERROR(EINVAL));
2196 err = vn_lock(*vpp, flags | LK_RETRY);
2198 vnode_create_vobject(*vpp, zp->z_size, curthread);
2205 * Block out VOPs and close zfsvfs_t::z_os
2207 * Note, if successful, then we return with the 'z_teardown_lock' and
2208 * 'z_teardown_inactive_lock' write held. We leave ownership of the underlying
2209 * dataset and objset intact so that they can be atomically handed off during
2210 * a subsequent rollback or recv operation and the resume thereafter.
2213 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2217 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2224 * Rebuild SA and release VOPs. Note that ownership of the underlying dataset
2225 * is an invariant across any of the operations that can be performed while the
2226 * filesystem was suspended. Whether it succeeded or failed, the preconditions
2227 * are the same: the relevant objset and associated dataset are owned by
2228 * zfsvfs, held, and long held on entry.
2231 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2235 uint64_t sa_obj = 0;
2237 ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2238 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2241 * We already own this, so just hold and rele it to update the
2242 * objset_t, as the one we had before may have been evicted.
2244 VERIFY0(dmu_objset_hold(osname, zfsvfs, &zfsvfs->z_os));
2245 VERIFY3P(zfsvfs->z_os->os_dsl_dataset->ds_owner, ==, zfsvfs);
2246 VERIFY(dsl_dataset_long_held(zfsvfs->z_os->os_dsl_dataset));
2247 dmu_objset_rele(zfsvfs->z_os, zfsvfs);
2250 * Make sure version hasn't changed
2253 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
2254 &zfsvfs->z_version);
2259 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2260 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2262 if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
2265 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2266 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
2269 if (zfsvfs->z_version >= ZPL_VERSION_SA)
2270 sa_register_update_callback(zfsvfs->z_os,
2273 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2275 zfs_set_fuid_feature(zfsvfs);
2278 * Attempt to re-establish all the active znodes with
2279 * their dbufs. If a zfs_rezget() fails, then we'll let
2280 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2281 * when they try to use their znode.
2283 mutex_enter(&zfsvfs->z_znodes_lock);
2284 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2285 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2286 (void) zfs_rezget(zp);
2288 mutex_exit(&zfsvfs->z_znodes_lock);
2291 /* release the VOPs */
2292 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2293 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2297 * Since we couldn't setup the sa framework, try to force
2298 * unmount this file system.
2300 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0) {
2301 vfs_ref(zfsvfs->z_vfs);
2302 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
2309 zfs_freevfs(vfs_t *vfsp)
2311 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2315 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2316 * from zfs_mount(). Release it here. If we came through
2317 * zfs_mountroot() instead, we didn't grab an extra hold, so
2318 * skip the VFS_RELE for rootvfs.
2320 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2321 VFS_RELE(zfsvfs->z_parent->z_vfs);
2324 zfsvfs_free(zfsvfs);
2326 atomic_dec_32(&zfs_active_fs_count);
2330 static int desiredvnodes_backup;
2334 zfs_vnodes_adjust(void)
2337 int newdesiredvnodes;
2339 desiredvnodes_backup = desiredvnodes;
2342 * We calculate newdesiredvnodes the same way it is done in
2343 * vntblinit(). If it is equal to desiredvnodes, it means that
2344 * it wasn't tuned by the administrator and we can tune it down.
2346 newdesiredvnodes = min(maxproc + cnt.v_page_count / 4, 2 *
2347 vm_kmem_size / (5 * (sizeof(struct vm_object) +
2348 sizeof(struct vnode))));
2349 if (newdesiredvnodes == desiredvnodes)
2350 desiredvnodes = (3 * newdesiredvnodes) / 4;
2355 zfs_vnodes_adjust_back(void)
2359 desiredvnodes = desiredvnodes_backup;
2367 printf("ZFS filesystem version: " ZPL_VERSION_STRING "\n");
2370 * Initialize .zfs directory structures
2375 * Initialize znode cache, vnode ops, etc...
2380 * Reduce number of vnodes. Originally number of vnodes is calculated
2381 * with UFS inode in mind. We reduce it here, because it's too big for
2384 zfs_vnodes_adjust();
2386 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2394 zfs_vnodes_adjust_back();
2400 return (zfs_active_fs_count != 0);
2404 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2407 objset_t *os = zfsvfs->z_os;
2410 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2411 return (SET_ERROR(EINVAL));
2413 if (newvers < zfsvfs->z_version)
2414 return (SET_ERROR(EINVAL));
2416 if (zfs_spa_version_map(newvers) >
2417 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2418 return (SET_ERROR(ENOTSUP));
2420 tx = dmu_tx_create(os);
2421 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2422 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2423 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2425 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2427 error = dmu_tx_assign(tx, TXG_WAIT);
2433 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2434 8, 1, &newvers, tx);
2441 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2444 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2446 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2447 DMU_OT_NONE, 0, tx);
2449 error = zap_add(os, MASTER_NODE_OBJ,
2450 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2453 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2454 sa_register_update_callback(os, zfs_sa_upgrade);
2457 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2458 "from %llu to %llu", zfsvfs->z_version, newvers);
2462 zfsvfs->z_version = newvers;
2464 zfs_set_fuid_feature(zfsvfs);
2470 * Read a property stored within the master node.
2473 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2479 * Look up the file system's value for the property. For the
2480 * version property, we look up a slightly different string.
2482 if (prop == ZFS_PROP_VERSION)
2483 pname = ZPL_VERSION_STR;
2485 pname = zfs_prop_to_name(prop);
2488 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2490 if (error == ENOENT) {
2491 /* No value set, use the default value */
2493 case ZFS_PROP_VERSION:
2494 *value = ZPL_VERSION;
2496 case ZFS_PROP_NORMALIZE:
2497 case ZFS_PROP_UTF8ONLY:
2501 *value = ZFS_CASE_SENSITIVE;
2513 zfsvfs_update_fromname(const char *oldname, const char *newname)
2515 char tmpbuf[MAXPATHLEN];
2520 oldlen = strlen(oldname);
2522 mtx_lock(&mountlist_mtx);
2523 TAILQ_FOREACH(mp, &mountlist, mnt_list) {
2524 fromname = mp->mnt_stat.f_mntfromname;
2525 if (strcmp(fromname, oldname) == 0) {
2526 (void)strlcpy(fromname, newname,
2527 sizeof(mp->mnt_stat.f_mntfromname));
2530 if (strncmp(fromname, oldname, oldlen) == 0 &&
2531 (fromname[oldlen] == '/' || fromname[oldlen] == '@')) {
2532 (void)snprintf(tmpbuf, sizeof(tmpbuf), "%s%s",
2533 newname, fromname + oldlen);
2534 (void)strlcpy(fromname, tmpbuf,
2535 sizeof(mp->mnt_stat.f_mntfromname));
2539 mtx_unlock(&mountlist_mtx);