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.
25 /* Portions Copyright 2010 Robert Milkowski */
27 #include <sys/types.h>
28 #include <sys/param.h>
29 #include <sys/systm.h>
30 #include <sys/sysmacros.h>
32 #include <sys/pathname.h>
33 #include <sys/vnode.h>
35 #include <sys/vfs_opreg.h>
36 #include <sys/mntent.h>
37 #include <sys/mount.h>
38 #include <sys/cmn_err.h>
39 #include "fs/fs_subr.h"
40 #include <sys/zfs_znode.h>
41 #include <sys/zfs_dir.h>
43 #include <sys/fs/zfs.h>
45 #include <sys/dsl_prop.h>
46 #include <sys/dsl_dataset.h>
47 #include <sys/dsl_deleg.h>
51 #include <sys/varargs.h>
52 #include <sys/policy.h>
53 #include <sys/atomic.h>
54 #include <sys/mkdev.h>
55 #include <sys/modctl.h>
56 #include <sys/refstr.h>
57 #include <sys/zfs_ioctl.h>
58 #include <sys/zfs_ctldir.h>
59 #include <sys/zfs_fuid.h>
60 #include <sys/bootconf.h>
61 #include <sys/sunddi.h>
63 #include <sys/dmu_objset.h>
64 #include <sys/spa_boot.h>
66 #include "zfs_comutil.h"
69 vfsops_t *zfs_vfsops = NULL;
70 static major_t zfs_major;
71 static minor_t zfs_minor;
72 static kmutex_t zfs_dev_mtx;
74 extern int sys_shutdown;
76 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
77 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
78 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
79 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
80 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
81 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
82 static void zfs_freevfs(vfs_t *vfsp);
84 static const fs_operation_def_t zfs_vfsops_template[] = {
85 VFSNAME_MOUNT, { .vfs_mount = zfs_mount },
86 VFSNAME_MOUNTROOT, { .vfs_mountroot = zfs_mountroot },
87 VFSNAME_UNMOUNT, { .vfs_unmount = zfs_umount },
88 VFSNAME_ROOT, { .vfs_root = zfs_root },
89 VFSNAME_STATVFS, { .vfs_statvfs = zfs_statvfs },
90 VFSNAME_SYNC, { .vfs_sync = zfs_sync },
91 VFSNAME_VGET, { .vfs_vget = zfs_vget },
92 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
96 static const fs_operation_def_t zfs_vfsops_eio_template[] = {
97 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
102 * We need to keep a count of active fs's.
103 * This is necessary to prevent our module
104 * from being unloaded after a umount -f
106 static uint32_t zfs_active_fs_count = 0;
108 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
109 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
110 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
111 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
114 * MO_DEFAULT is not used since the default value is determined
115 * by the equivalent property.
117 static mntopt_t mntopts[] = {
118 { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
119 { MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
120 { MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
121 { MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
124 static mntopts_t zfs_mntopts = {
125 sizeof (mntopts) / sizeof (mntopt_t),
131 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
134 * Data integrity is job one. We don't want a compromised kernel
135 * writing to the storage pool, so we never sync during panic.
141 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
142 * to sync metadata, which they would otherwise cache indefinitely.
143 * Semantically, the only requirement is that the sync be initiated.
144 * The DMU syncs out txgs frequently, so there's nothing to do.
146 if (flag & SYNC_ATTR)
151 * Sync a specific filesystem.
153 zfsvfs_t *zfsvfs = vfsp->vfs_data;
157 dp = dmu_objset_pool(zfsvfs->z_os);
160 * If the system is shutting down, then skip any
161 * filesystems which may exist on a suspended pool.
163 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
168 if (zfsvfs->z_log != NULL)
169 zil_commit(zfsvfs->z_log, UINT64_MAX, 0);
174 * Sync all ZFS filesystems. This is what happens when you
175 * run sync(1M). Unlike other filesystems, ZFS honors the
176 * request by waiting for all pools to commit all dirty data.
185 zfs_create_unique_device(dev_t *dev)
190 ASSERT3U(zfs_minor, <=, MAXMIN32);
191 minor_t start = zfs_minor;
193 mutex_enter(&zfs_dev_mtx);
194 if (zfs_minor >= MAXMIN32) {
196 * If we're still using the real major
197 * keep out of /dev/zfs and /dev/zvol minor
198 * number space. If we're using a getudev()'ed
199 * major number, we can use all of its minors.
201 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
202 zfs_minor = ZFS_MIN_MINOR;
208 *dev = makedevice(zfs_major, zfs_minor);
209 mutex_exit(&zfs_dev_mtx);
210 } while (vfs_devismounted(*dev) && zfs_minor != start);
211 if (zfs_minor == start) {
213 * We are using all ~262,000 minor numbers for the
214 * current major number. Create a new major number.
216 if ((new_major = getudev()) == (major_t)-1) {
218 "zfs_mount: Can't get unique major "
222 mutex_enter(&zfs_dev_mtx);
223 zfs_major = new_major;
226 mutex_exit(&zfs_dev_mtx);
230 /* CONSTANTCONDITION */
237 atime_changed_cb(void *arg, uint64_t newval)
239 zfsvfs_t *zfsvfs = arg;
241 if (newval == TRUE) {
242 zfsvfs->z_atime = TRUE;
243 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
244 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
246 zfsvfs->z_atime = FALSE;
247 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
248 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
253 xattr_changed_cb(void *arg, uint64_t newval)
255 zfsvfs_t *zfsvfs = arg;
257 if (newval == TRUE) {
258 /* XXX locking on vfs_flag? */
259 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
260 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
261 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
263 /* XXX locking on vfs_flag? */
264 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;
275 if (newval < SPA_MINBLOCKSIZE ||
276 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
277 newval = SPA_MAXBLOCKSIZE;
279 zfsvfs->z_max_blksz = newval;
280 zfsvfs->z_vfs->vfs_bsize = newval;
284 readonly_changed_cb(void *arg, uint64_t newval)
286 zfsvfs_t *zfsvfs = arg;
289 /* XXX locking on vfs_flag? */
290 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
291 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
292 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
294 /* XXX locking on vfs_flag? */
295 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
296 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
297 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
302 devices_changed_cb(void *arg, uint64_t newval)
304 zfsvfs_t *zfsvfs = arg;
306 if (newval == FALSE) {
307 zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
308 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
309 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
311 zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
312 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
313 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
318 setuid_changed_cb(void *arg, uint64_t newval)
320 zfsvfs_t *zfsvfs = arg;
322 if (newval == FALSE) {
323 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
324 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
325 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
327 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
328 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
329 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
334 exec_changed_cb(void *arg, uint64_t newval)
336 zfsvfs_t *zfsvfs = arg;
338 if (newval == FALSE) {
339 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
340 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
341 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
343 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
344 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
345 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
350 * The nbmand mount option can be changed at mount time.
351 * We can't allow it to be toggled on live file systems or incorrect
352 * behavior may be seen from cifs clients
354 * This property isn't registered via dsl_prop_register(), but this callback
355 * will be called when a file system is first mounted
358 nbmand_changed_cb(void *arg, uint64_t newval)
360 zfsvfs_t *zfsvfs = arg;
361 if (newval == FALSE) {
362 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
363 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
365 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
366 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
371 snapdir_changed_cb(void *arg, uint64_t newval)
373 zfsvfs_t *zfsvfs = arg;
375 zfsvfs->z_show_ctldir = newval;
379 vscan_changed_cb(void *arg, uint64_t newval)
381 zfsvfs_t *zfsvfs = arg;
383 zfsvfs->z_vscan = newval;
387 acl_inherit_changed_cb(void *arg, uint64_t newval)
389 zfsvfs_t *zfsvfs = arg;
391 zfsvfs->z_acl_inherit = newval;
395 zfs_register_callbacks(vfs_t *vfsp)
397 struct dsl_dataset *ds = NULL;
399 zfsvfs_t *zfsvfs = NULL;
401 int readonly, do_readonly = B_FALSE;
402 int setuid, do_setuid = B_FALSE;
403 int exec, do_exec = B_FALSE;
404 int devices, do_devices = B_FALSE;
405 int xattr, do_xattr = B_FALSE;
406 int atime, do_atime = B_FALSE;
410 zfsvfs = vfsp->vfs_data;
415 * The act of registering our callbacks will destroy any mount
416 * options we may have. In order to enable temporary overrides
417 * of mount options, we stash away the current values and
418 * restore them after we register the callbacks.
420 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) {
422 do_readonly = B_TRUE;
423 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
425 do_readonly = B_TRUE;
427 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
433 if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
436 } else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
441 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
444 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
449 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
452 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
456 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
459 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
463 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
466 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
472 * nbmand is a special property. It can only be changed at
475 * This is weird, but it is documented to only be changeable
478 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
480 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
483 char osname[MAXNAMELEN];
485 dmu_objset_name(os, osname);
486 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
493 * Register property callbacks.
495 * It would probably be fine to just check for i/o error from
496 * the first prop_register(), but I guess I like to go
499 ds = dmu_objset_ds(os);
500 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs);
501 error = error ? error : dsl_prop_register(ds,
502 "xattr", xattr_changed_cb, zfsvfs);
503 error = error ? error : dsl_prop_register(ds,
504 "recordsize", blksz_changed_cb, zfsvfs);
505 error = error ? error : dsl_prop_register(ds,
506 "readonly", readonly_changed_cb, zfsvfs);
507 error = error ? error : dsl_prop_register(ds,
508 "devices", devices_changed_cb, zfsvfs);
509 error = error ? error : dsl_prop_register(ds,
510 "setuid", setuid_changed_cb, zfsvfs);
511 error = error ? error : dsl_prop_register(ds,
512 "exec", exec_changed_cb, zfsvfs);
513 error = error ? error : dsl_prop_register(ds,
514 "snapdir", snapdir_changed_cb, zfsvfs);
515 error = error ? error : dsl_prop_register(ds,
516 "aclinherit", acl_inherit_changed_cb, zfsvfs);
517 error = error ? error : dsl_prop_register(ds,
518 "vscan", vscan_changed_cb, zfsvfs);
523 * Invoke our callbacks to restore temporary mount options.
526 readonly_changed_cb(zfsvfs, readonly);
528 setuid_changed_cb(zfsvfs, setuid);
530 exec_changed_cb(zfsvfs, exec);
532 devices_changed_cb(zfsvfs, devices);
534 xattr_changed_cb(zfsvfs, xattr);
536 atime_changed_cb(zfsvfs, atime);
538 nbmand_changed_cb(zfsvfs, nbmand);
544 * We may attempt to unregister some callbacks that are not
545 * registered, but this is OK; it will simply return ENOMSG,
546 * which we will ignore.
548 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs);
549 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs);
550 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs);
551 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs);
552 (void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zfsvfs);
553 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs);
554 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs);
555 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs);
556 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
558 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs);
564 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
565 uint64_t *userp, uint64_t *groupp)
567 znode_phys_t *znp = data;
571 * Is it a valid type of object to track?
573 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
577 * If we have a NULL data pointer
578 * then assume the id's aren't changing and
579 * return EEXIST to the dmu to let it know to
585 if (bonustype == DMU_OT_ZNODE) {
586 *userp = znp->zp_uid;
587 *groupp = znp->zp_gid;
591 ASSERT(bonustype == DMU_OT_SA);
592 hdrsize = sa_hdrsize(data);
595 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
597 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
601 * This should only happen for newly created
602 * files that haven't had the znode data filled
613 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
614 char *domainbuf, int buflen, uid_t *ridp)
619 fuid = strtonum(fuidstr, NULL);
621 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
623 (void) strlcpy(domainbuf, domain, buflen);
626 *ridp = FUID_RID(fuid);
630 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
633 case ZFS_PROP_USERUSED:
634 return (DMU_USERUSED_OBJECT);
635 case ZFS_PROP_GROUPUSED:
636 return (DMU_GROUPUSED_OBJECT);
637 case ZFS_PROP_USERQUOTA:
638 return (zfsvfs->z_userquota_obj);
639 case ZFS_PROP_GROUPQUOTA:
640 return (zfsvfs->z_groupquota_obj);
646 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
647 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
652 zfs_useracct_t *buf = vbuf;
655 if (!dmu_objset_userspace_present(zfsvfs->z_os))
658 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
664 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
665 (error = zap_cursor_retrieve(&zc, &za)) == 0;
666 zap_cursor_advance(&zc)) {
667 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
671 fuidstr_to_sid(zfsvfs, za.za_name,
672 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
674 buf->zu_space = za.za_first_integer;
680 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
681 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
682 *cookiep = zap_cursor_serialize(&zc);
683 zap_cursor_fini(&zc);
688 * buf must be big enough (eg, 32 bytes)
691 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
692 char *buf, boolean_t addok)
697 if (domain && domain[0]) {
698 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
702 fuid = FUID_ENCODE(domainid, rid);
703 (void) sprintf(buf, "%llx", (longlong_t)fuid);
708 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
709 const char *domain, uint64_t rid, uint64_t *valp)
717 if (!dmu_objset_userspace_present(zfsvfs->z_os))
720 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
724 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
728 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
735 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
736 const char *domain, uint64_t rid, uint64_t quota)
742 boolean_t fuid_dirtied;
744 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
747 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
750 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
751 &zfsvfs->z_groupquota_obj;
753 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
756 fuid_dirtied = zfsvfs->z_fuid_dirty;
758 tx = dmu_tx_create(zfsvfs->z_os);
759 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
761 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
762 zfs_userquota_prop_prefixes[type]);
765 zfs_fuid_txhold(zfsvfs, tx);
766 err = dmu_tx_assign(tx, TXG_WAIT);
772 mutex_enter(&zfsvfs->z_lock);
774 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
776 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
777 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
779 mutex_exit(&zfsvfs->z_lock);
782 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
786 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
790 zfs_fuid_sync(zfsvfs, tx);
796 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
799 uint64_t used, quota, usedobj, quotaobj;
802 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
803 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
805 if (quotaobj == 0 || zfsvfs->z_replay)
808 (void) sprintf(buf, "%llx", (longlong_t)fuid);
809 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
813 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
816 return (used >= quota);
820 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
826 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
828 id = isgroup ? zp->z_gid : zp->z_uid;
830 if (quotaobj == 0 || zfsvfs->z_replay)
833 if (IS_EPHEMERAL(id)) {
834 VERIFY(0 == sa_lookup(zp->z_sa_hdl,
835 isgroup ? SA_ZPL_GID(zfsvfs) : SA_ZPL_UID(zfsvfs),
836 &fuid, sizeof (fuid)));
841 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
845 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
853 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
856 * We claim to always be readonly so we can open snapshots;
857 * other ZPL code will prevent us from writing to snapshots.
859 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
861 kmem_free(zfsvfs, sizeof (zfsvfs_t));
866 * Initialize the zfs-specific filesystem structure.
867 * Should probably make this a kmem cache, shuffle fields,
868 * and just bzero up to z_hold_mtx[].
870 zfsvfs->z_vfs = NULL;
871 zfsvfs->z_parent = zfsvfs;
872 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
873 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
876 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
879 } else if (zfsvfs->z_version >
880 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
881 (void) printf("Can't mount a version %lld file system "
882 "on a version %lld pool\n. Pool must be upgraded to mount "
883 "this file system.", (u_longlong_t)zfsvfs->z_version,
884 (u_longlong_t)spa_version(dmu_objset_spa(os)));
888 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
890 zfsvfs->z_norm = (int)zval;
892 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
894 zfsvfs->z_utf8 = (zval != 0);
896 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
898 zfsvfs->z_case = (uint_t)zval;
901 * Fold case on file systems that are always or sometimes case
904 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
905 zfsvfs->z_case == ZFS_CASE_MIXED)
906 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
908 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
909 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
911 if (zfsvfs->z_use_sa) {
912 /* should either have both of these objects or none */
913 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
919 * Pre SA versions file systems should never touch
920 * either the attribute registration or layout objects.
925 zfsvfs->z_attr_table = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END);
927 if (zfsvfs->z_version >= ZPL_VERSION_SA)
928 sa_register_update_callback(os, zfs_sa_upgrade);
930 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
934 ASSERT(zfsvfs->z_root != 0);
936 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
937 &zfsvfs->z_unlinkedobj);
941 error = zap_lookup(os, MASTER_NODE_OBJ,
942 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
943 8, 1, &zfsvfs->z_userquota_obj);
944 if (error && error != ENOENT)
947 error = zap_lookup(os, MASTER_NODE_OBJ,
948 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
949 8, 1, &zfsvfs->z_groupquota_obj);
950 if (error && error != ENOENT)
953 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
954 &zfsvfs->z_fuid_obj);
955 if (error && error != ENOENT)
958 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
959 &zfsvfs->z_shares_dir);
960 if (error && error != ENOENT)
963 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
964 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
965 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
966 offsetof(znode_t, z_link_node));
967 rrw_init(&zfsvfs->z_teardown_lock);
968 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
969 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
970 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
971 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
977 dmu_objset_disown(os, zfsvfs);
979 kmem_free(zfsvfs, sizeof (zfsvfs_t));
984 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
988 error = zfs_register_callbacks(zfsvfs->z_vfs);
993 * Set the objset user_ptr to track its zfsvfs.
995 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
996 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
997 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
999 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1002 * If we are not mounting (ie: online recv), then we don't
1003 * have to worry about replaying the log as we blocked all
1004 * operations out since we closed the ZIL.
1010 * During replay we remove the read only flag to
1011 * allow replays to succeed.
1013 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1015 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1017 zfs_unlinked_drain(zfsvfs);
1020 * Parse and replay the intent log.
1022 * Because of ziltest, this must be done after
1023 * zfs_unlinked_drain(). (Further note: ziltest
1024 * doesn't use readonly mounts, where
1025 * zfs_unlinked_drain() isn't called.) This is because
1026 * ziltest causes spa_sync() to think it's committed,
1027 * but actually it is not, so the intent log contains
1028 * many txg's worth of changes.
1030 * In particular, if object N is in the unlinked set in
1031 * the last txg to actually sync, then it could be
1032 * actually freed in a later txg and then reallocated
1033 * in a yet later txg. This would write a "create
1034 * object N" record to the intent log. Normally, this
1035 * would be fine because the spa_sync() would have
1036 * written out the fact that object N is free, before
1037 * we could write the "create object N" intent log
1040 * But when we are in ziltest mode, we advance the "open
1041 * txg" without actually spa_sync()-ing the changes to
1042 * disk. So we would see that object N is still
1043 * allocated and in the unlinked set, and there is an
1044 * intent log record saying to allocate it.
1046 if (zil_replay_disable) {
1047 zil_destroy(zfsvfs->z_log, B_FALSE);
1049 zfsvfs->z_replay = B_TRUE;
1050 zil_replay(zfsvfs->z_os, zfsvfs, zfs_replay_vector);
1051 zfsvfs->z_replay = B_FALSE;
1053 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1060 zfsvfs_free(zfsvfs_t *zfsvfs)
1063 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1066 * This is a barrier to prevent the filesystem from going away in
1067 * zfs_znode_move() until we can safely ensure that the filesystem is
1068 * not unmounted. We consider the filesystem valid before the barrier
1069 * and invalid after the barrier.
1071 rw_enter(&zfsvfs_lock, RW_READER);
1072 rw_exit(&zfsvfs_lock);
1074 zfs_fuid_destroy(zfsvfs);
1076 mutex_destroy(&zfsvfs->z_znodes_lock);
1077 mutex_destroy(&zfsvfs->z_lock);
1078 list_destroy(&zfsvfs->z_all_znodes);
1079 rrw_destroy(&zfsvfs->z_teardown_lock);
1080 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1081 rw_destroy(&zfsvfs->z_fuid_lock);
1082 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1083 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1084 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1088 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1090 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1091 if (zfsvfs->z_use_fuids && zfsvfs->z_vfs) {
1092 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1093 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1094 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1095 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1096 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1097 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1099 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1103 zfs_domount(vfs_t *vfsp, char *osname)
1106 uint64_t recordsize, fsid_guid;
1113 error = zfsvfs_create(osname, &zfsvfs);
1116 zfsvfs->z_vfs = vfsp;
1118 /* Initialize the generic filesystem structure. */
1119 vfsp->vfs_bcount = 0;
1120 vfsp->vfs_data = NULL;
1122 if (zfs_create_unique_device(&mount_dev) == -1) {
1126 ASSERT(vfs_devismounted(mount_dev) == 0);
1128 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1132 vfsp->vfs_dev = mount_dev;
1133 vfsp->vfs_fstype = zfsfstype;
1134 vfsp->vfs_bsize = recordsize;
1135 vfsp->vfs_flag |= VFS_NOTRUNC;
1136 vfsp->vfs_data = zfsvfs;
1139 * The fsid is 64 bits, composed of an 8-bit fs type, which
1140 * separates our fsid from any other filesystem types, and a
1141 * 56-bit objset unique ID. The objset unique ID is unique to
1142 * all objsets open on this system, provided by unique_create().
1143 * The 8-bit fs type must be put in the low bits of fsid[1]
1144 * because that's where other Solaris filesystems put it.
1146 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1147 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1148 vfsp->vfs_fsid.val[0] = fsid_guid;
1149 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1153 * Set features for file system.
1155 zfs_set_fuid_feature(zfsvfs);
1156 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1157 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1158 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1159 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1160 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1161 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1162 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1164 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1166 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1169 atime_changed_cb(zfsvfs, B_FALSE);
1170 readonly_changed_cb(zfsvfs, B_TRUE);
1171 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1173 xattr_changed_cb(zfsvfs, pval);
1174 zfsvfs->z_issnap = B_TRUE;
1176 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1177 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1178 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1180 error = zfsvfs_setup(zfsvfs, B_TRUE);
1183 if (!zfsvfs->z_issnap)
1184 zfsctl_create(zfsvfs);
1187 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1188 zfsvfs_free(zfsvfs);
1190 atomic_add_32(&zfs_active_fs_count, 1);
1197 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1199 objset_t *os = zfsvfs->z_os;
1200 struct dsl_dataset *ds;
1203 * Unregister properties.
1205 if (!dmu_objset_is_snapshot(os)) {
1206 ds = dmu_objset_ds(os);
1207 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1210 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1213 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1216 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1219 VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1222 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1225 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1228 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1231 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1232 acl_inherit_changed_cb, zfsvfs) == 0);
1234 VERIFY(dsl_prop_unregister(ds, "vscan",
1235 vscan_changed_cb, zfsvfs) == 0);
1240 * Convert a decimal digit string to a uint64_t integer.
1243 str_to_uint64(char *str, uint64_t *objnum)
1248 if (*str < '0' || *str > '9')
1251 num = num*10 + *str++ - '0';
1259 * The boot path passed from the boot loader is in the form of
1260 * "rootpool-name/root-filesystem-object-number'. Convert this
1261 * string to a dataset name: "rootpool-name/root-filesystem-name".
1264 zfs_parse_bootfs(char *bpath, char *outpath)
1270 if (*bpath == 0 || *bpath == '/')
1273 (void) strcpy(outpath, bpath);
1275 slashp = strchr(bpath, '/');
1277 /* if no '/', just return the pool name */
1278 if (slashp == NULL) {
1282 /* if not a number, just return the root dataset name */
1283 if (str_to_uint64(slashp+1, &objnum)) {
1288 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1295 * zfs_check_global_label:
1296 * Check that the hex label string is appropriate for the dataset
1297 * being mounted into the global_zone proper.
1299 * Return an error if the hex label string is not default or
1300 * admin_low/admin_high. For admin_low labels, the corresponding
1301 * dataset must be readonly.
1304 zfs_check_global_label(const char *dsname, const char *hexsl)
1306 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1308 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1310 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1311 /* must be readonly */
1314 if (dsl_prop_get_integer(dsname,
1315 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1317 return (rdonly ? 0 : EACCES);
1323 * zfs_mount_label_policy:
1324 * Determine whether the mount is allowed according to MAC check.
1325 * by comparing (where appropriate) label of the dataset against
1326 * the label of the zone being mounted into. If the dataset has
1327 * no label, create one.
1330 * 0 : access allowed
1331 * >0 : error code, such as EACCES
1334 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1337 zone_t *mntzone = NULL;
1338 ts_label_t *mnt_tsl;
1341 char ds_hexsl[MAXNAMELEN];
1343 retv = EACCES; /* assume the worst */
1346 * Start by getting the dataset label if it exists.
1348 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1349 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1354 * If labeling is NOT enabled, then disallow the mount of datasets
1355 * which have a non-default label already. No other label checks
1358 if (!is_system_labeled()) {
1359 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1365 * Get the label of the mountpoint. If mounting into the global
1366 * zone (i.e. mountpoint is not within an active zone and the
1367 * zoned property is off), the label must be default or
1368 * admin_low/admin_high only; no other checks are needed.
1370 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1371 if (mntzone->zone_id == GLOBAL_ZONEID) {
1376 if (dsl_prop_get_integer(osname,
1377 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1380 return (zfs_check_global_label(osname, ds_hexsl));
1383 * This is the case of a zone dataset being mounted
1384 * initially, before the zone has been fully created;
1385 * allow this mount into global zone.
1390 mnt_tsl = mntzone->zone_slabel;
1391 ASSERT(mnt_tsl != NULL);
1392 label_hold(mnt_tsl);
1393 mnt_sl = label2bslabel(mnt_tsl);
1395 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1397 * The dataset doesn't have a real label, so fabricate one.
1401 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1402 dsl_prop_set(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1403 ZPROP_SRC_LOCAL, 1, strlen(str) + 1, str) == 0)
1406 kmem_free(str, strlen(str) + 1);
1407 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1409 * Now compare labels to complete the MAC check. If the
1410 * labels are equal then allow access. If the mountpoint
1411 * label dominates the dataset label, allow readonly access.
1412 * Otherwise, access is denied.
1414 if (blequal(mnt_sl, &ds_sl))
1416 else if (bldominates(mnt_sl, &ds_sl)) {
1417 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1422 label_rele(mnt_tsl);
1428 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1431 static int zfsrootdone = 0;
1432 zfsvfs_t *zfsvfs = NULL;
1441 * The filesystem that we mount as root is defined in the
1442 * boot property "zfs-bootfs" with a format of
1443 * "poolname/root-dataset-objnum".
1445 if (why == ROOT_INIT) {
1449 * the process of doing a spa_load will require the
1450 * clock to be set before we could (for example) do
1451 * something better by looking at the timestamp on
1452 * an uberblock, so just set it to -1.
1456 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1457 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1461 zfs_devid = spa_get_bootprop("diskdevid");
1462 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1464 spa_free_bootprop(zfs_devid);
1466 spa_free_bootprop(zfs_bootfs);
1467 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1471 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1472 spa_free_bootprop(zfs_bootfs);
1473 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1478 spa_free_bootprop(zfs_bootfs);
1480 if (error = vfs_lock(vfsp))
1483 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1484 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1488 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1490 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1491 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1496 mutex_enter(&vp->v_lock);
1497 vp->v_flag |= VROOT;
1498 mutex_exit(&vp->v_lock);
1502 * Leave rootvp held. The root file system is never unmounted.
1505 vfs_add((struct vnode *)0, vfsp,
1506 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1510 } else if (why == ROOT_REMOUNT) {
1511 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1512 vfsp->vfs_flag |= VFS_REMOUNT;
1514 /* refresh mount options */
1515 zfs_unregister_callbacks(vfsp->vfs_data);
1516 return (zfs_register_callbacks(vfsp));
1518 } else if (why == ROOT_UNMOUNT) {
1519 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1520 (void) zfs_sync(vfsp, 0, 0);
1525 * if "why" is equal to anything else other than ROOT_INIT,
1526 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1533 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1538 uio_seg_t fromspace = (uap->flags & MS_SYSSPACE) ?
1539 UIO_SYSSPACE : UIO_USERSPACE;
1542 if (mvp->v_type != VDIR)
1545 mutex_enter(&mvp->v_lock);
1546 if ((uap->flags & MS_REMOUNT) == 0 &&
1547 (uap->flags & MS_OVERLAY) == 0 &&
1548 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1549 mutex_exit(&mvp->v_lock);
1552 mutex_exit(&mvp->v_lock);
1555 * ZFS does not support passing unparsed data in via MS_DATA.
1556 * Users should use the MS_OPTIONSTR interface; this means
1557 * that all option parsing is already done and the options struct
1558 * can be interrogated.
1560 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1564 * Get the objset name (the "special" mount argument).
1566 if (error = pn_get(uap->spec, fromspace, &spn))
1569 osname = spn.pn_path;
1572 * Check for mount privilege?
1574 * If we don't have privilege then see if
1575 * we have local permission to allow it
1577 error = secpolicy_fs_mount(cr, mvp, vfsp);
1579 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1583 * Make sure user is the owner of the mount point
1584 * or has sufficient privileges.
1587 vattr.va_mask = AT_UID;
1589 if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1593 if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1594 VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1597 secpolicy_fs_mount_clearopts(cr, vfsp);
1604 * Refuse to mount a filesystem if we are in a local zone and the
1605 * dataset is not visible.
1607 if (!INGLOBALZONE(curproc) &&
1608 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1613 error = zfs_mount_label_policy(vfsp, osname);
1618 * When doing a remount, we simply refresh our temporary properties
1619 * according to those options set in the current VFS options.
1621 if (uap->flags & MS_REMOUNT) {
1622 /* refresh mount options */
1623 zfs_unregister_callbacks(vfsp->vfs_data);
1624 error = zfs_register_callbacks(vfsp);
1628 error = zfs_domount(vfsp, osname);
1631 * Add an extra VFS_HOLD on our parent vfs so that it can't
1632 * disappear due to a forced unmount.
1634 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1635 VFS_HOLD(mvp->v_vfsp);
1643 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1645 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1647 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1651 dmu_objset_space(zfsvfs->z_os,
1652 &refdbytes, &availbytes, &usedobjs, &availobjs);
1655 * The underlying storage pool actually uses multiple block sizes.
1656 * We report the fragsize as the smallest block size we support,
1657 * and we report our blocksize as the filesystem's maximum blocksize.
1659 statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1660 statp->f_bsize = zfsvfs->z_max_blksz;
1663 * The following report "total" blocks of various kinds in the
1664 * file system, but reported in terms of f_frsize - the
1668 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1669 statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1670 statp->f_bavail = statp->f_bfree; /* no root reservation */
1673 * statvfs() should really be called statufs(), because it assumes
1674 * static metadata. ZFS doesn't preallocate files, so the best
1675 * we can do is report the max that could possibly fit in f_files,
1676 * and that minus the number actually used in f_ffree.
1677 * For f_ffree, report the smaller of the number of object available
1678 * and the number of blocks (each object will take at least a block).
1680 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1681 statp->f_favail = statp->f_ffree; /* no "root reservation" */
1682 statp->f_files = statp->f_ffree + usedobjs;
1684 (void) cmpldev(&d32, vfsp->vfs_dev);
1685 statp->f_fsid = d32;
1688 * We're a zfs filesystem.
1690 (void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1692 statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1694 statp->f_namemax = ZFS_MAXNAMELEN;
1697 * We have all of 32 characters to stuff a string here.
1698 * Is there anything useful we could/should provide?
1700 bzero(statp->f_fstr, sizeof (statp->f_fstr));
1707 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1709 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1715 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1717 *vpp = ZTOV(rootzp);
1724 * Teardown the zfsvfs::z_os.
1726 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1727 * and 'z_teardown_inactive_lock' held.
1730 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1734 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1738 * We purge the parent filesystem's vfsp as the parent
1739 * filesystem and all of its snapshots have their vnode's
1740 * v_vfsp set to the parent's filesystem's vfsp. Note,
1741 * 'z_parent' is self referential for non-snapshots.
1743 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1747 * Close the zil. NB: Can't close the zil while zfs_inactive
1748 * threads are blocked as zil_close can call zfs_inactive.
1750 if (zfsvfs->z_log) {
1751 zil_close(zfsvfs->z_log);
1752 zfsvfs->z_log = NULL;
1755 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1758 * If we are not unmounting (ie: online recv) and someone already
1759 * unmounted this file system while we were doing the switcheroo,
1760 * or a reopen of z_os failed then just bail out now.
1762 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1763 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1764 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1769 * At this point there are no vops active, and any new vops will
1770 * fail with EIO since we have z_teardown_lock for writer (only
1771 * relavent for forced unmount).
1773 * Release all holds on dbufs.
1775 mutex_enter(&zfsvfs->z_znodes_lock);
1776 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1777 zp = list_next(&zfsvfs->z_all_znodes, zp))
1779 ASSERT(ZTOV(zp)->v_count > 0);
1780 zfs_znode_dmu_fini(zp);
1782 mutex_exit(&zfsvfs->z_znodes_lock);
1785 * If we are unmounting, set the unmounted flag and let new vops
1786 * unblock. zfs_inactive will have the unmounted behavior, and all
1787 * other vops will fail with EIO.
1790 zfsvfs->z_unmounted = B_TRUE;
1791 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1792 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1796 * z_os will be NULL if there was an error in attempting to reopen
1797 * zfsvfs, so just return as the properties had already been
1798 * unregistered and cached data had been evicted before.
1800 if (zfsvfs->z_os == NULL)
1804 * Unregister properties.
1806 zfs_unregister_callbacks(zfsvfs);
1811 if (dmu_objset_evict_dbufs(zfsvfs->z_os)) {
1812 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1813 (void) dmu_objset_evict_dbufs(zfsvfs->z_os);
1821 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1823 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1827 ret = secpolicy_fs_unmount(cr, vfsp);
1829 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1830 ZFS_DELEG_PERM_MOUNT, cr))
1835 * We purge the parent filesystem's vfsp as the parent filesystem
1836 * and all of its snapshots have their vnode's v_vfsp set to the
1837 * parent's filesystem's vfsp. Note, 'z_parent' is self
1838 * referential for non-snapshots.
1840 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1843 * Unmount any snapshots mounted under .zfs before unmounting the
1846 if (zfsvfs->z_ctldir != NULL &&
1847 (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1851 if (!(fflag & MS_FORCE)) {
1853 * Check the number of active vnodes in the file system.
1854 * Our count is maintained in the vfs structure, but the
1855 * number is off by 1 to indicate a hold on the vfs
1858 * The '.zfs' directory maintains a reference of its
1859 * own, and any active references underneath are
1860 * reflected in the vnode count.
1862 if (zfsvfs->z_ctldir == NULL) {
1863 if (vfsp->vfs_count > 1)
1866 if (vfsp->vfs_count > 2 ||
1867 zfsvfs->z_ctldir->v_count > 1)
1872 vfsp->vfs_flag |= VFS_UNMOUNTED;
1874 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1878 * z_os will be NULL if there was an error in
1879 * attempting to reopen zfsvfs.
1883 * Unset the objset user_ptr.
1885 mutex_enter(&os->os_user_ptr_lock);
1886 dmu_objset_set_user(os, NULL);
1887 mutex_exit(&os->os_user_ptr_lock);
1890 * Finally release the objset
1892 dmu_objset_disown(os, zfsvfs);
1896 * We can now safely destroy the '.zfs' directory node.
1898 if (zfsvfs->z_ctldir != NULL)
1899 zfsctl_destroy(zfsvfs);
1905 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1907 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1909 uint64_t object = 0;
1910 uint64_t fid_gen = 0;
1919 if (fidp->fid_len == LONG_FID_LEN) {
1920 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1921 uint64_t objsetid = 0;
1922 uint64_t setgen = 0;
1924 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1925 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1927 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1928 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1932 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1938 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1939 zfid_short_t *zfid = (zfid_short_t *)fidp;
1941 for (i = 0; i < sizeof (zfid->zf_object); i++)
1942 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1944 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1945 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1951 /* A zero fid_gen means we are in the .zfs control directories */
1953 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1954 *vpp = zfsvfs->z_ctldir;
1955 ASSERT(*vpp != NULL);
1956 if (object == ZFSCTL_INO_SNAPDIR) {
1957 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
1958 0, NULL, NULL, NULL, NULL, NULL) == 0);
1966 gen_mask = -1ULL >> (64 - 8 * i);
1968 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1969 if (err = zfs_zget(zfsvfs, object, &zp)) {
1973 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1975 zp_gen = zp_gen & gen_mask;
1978 if (zp->z_unlinked || zp_gen != fid_gen) {
1979 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1991 * Block out VOPs and close zfsvfs_t::z_os
1993 * Note, if successful, then we return with the 'z_teardown_lock' and
1994 * 'z_teardown_inactive_lock' write held.
1997 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2001 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2003 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
2009 * Reopen zfsvfs_t::z_os and release VOPs.
2012 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2016 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
2017 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2019 err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zfsvfs,
2022 zfsvfs->z_os = NULL;
2025 uint64_t sa_obj = 0;
2027 err2 = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2028 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2030 if ((err || err2) && zfsvfs->z_version >= ZPL_VERSION_SA)
2034 zfsvfs->z_attr_table = sa_setup(zfsvfs->z_os, sa_obj,
2035 zfs_attr_table, ZPL_END);
2037 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2040 * Attempt to re-establish all the active znodes with
2041 * their dbufs. If a zfs_rezget() fails, then we'll let
2042 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2043 * when they try to use their znode.
2045 mutex_enter(&zfsvfs->z_znodes_lock);
2046 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2047 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2048 (void) zfs_rezget(zp);
2050 mutex_exit(&zfsvfs->z_znodes_lock);
2055 /* release the VOPs */
2056 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2057 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
2061 * Since we couldn't reopen zfsvfs::z_os, force
2062 * unmount this file system.
2064 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2065 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2071 zfs_freevfs(vfs_t *vfsp)
2073 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2076 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2077 * from zfs_mount(). Release it here. If we came through
2078 * zfs_mountroot() instead, we didn't grab an extra hold, so
2079 * skip the VFS_RELE for rootvfs.
2081 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2082 VFS_RELE(zfsvfs->z_parent->z_vfs);
2084 zfsvfs_free(zfsvfs);
2086 atomic_add_32(&zfs_active_fs_count, -1);
2090 * VFS_INIT() initialization. Note that there is no VFS_FINI(),
2091 * so we can't safely do any non-idempotent initialization here.
2092 * Leave that to zfs_init() and zfs_fini(), which are called
2093 * from the module's _init() and _fini() entry points.
2097 zfs_vfsinit(int fstype, char *name)
2104 * Setup vfsops and vnodeops tables.
2106 error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2108 cmn_err(CE_WARN, "zfs: bad vfs ops template");
2111 error = zfs_create_op_tables();
2113 zfs_remove_op_tables();
2114 cmn_err(CE_WARN, "zfs: bad vnode ops template");
2115 (void) vfs_freevfsops_by_type(zfsfstype);
2119 mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2122 * Unique major number for all zfs mounts.
2123 * If we run out of 32-bit minors, we'll getudev() another major.
2125 zfs_major = ddi_name_to_major(ZFS_DRIVER);
2126 zfs_minor = ZFS_MIN_MINOR;
2135 * Initialize .zfs directory structures
2140 * Initialize znode cache, vnode ops, etc...
2144 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2157 return (zfs_active_fs_count != 0);
2161 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2164 objset_t *os = zfsvfs->z_os;
2167 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2170 if (newvers < zfsvfs->z_version)
2173 if (zfs_spa_version_map(newvers) >
2174 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2177 tx = dmu_tx_create(os);
2178 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2179 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2180 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2182 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2184 error = dmu_tx_assign(tx, TXG_WAIT);
2190 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2191 8, 1, &newvers, tx);
2198 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2201 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2203 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2204 DMU_OT_NONE, 0, tx);
2206 error = zap_add(os, MASTER_NODE_OBJ,
2207 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2208 ASSERT3U(error, ==, 0);
2210 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2211 sa_register_update_callback(os, zfs_sa_upgrade);
2214 spa_history_log_internal(LOG_DS_UPGRADE,
2215 dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu",
2216 zfsvfs->z_version, newvers, dmu_objset_id(os));
2220 zfsvfs->z_version = newvers;
2222 if (zfsvfs->z_version >= ZPL_VERSION_FUID)
2223 zfs_set_fuid_feature(zfsvfs);
2229 * Read a property stored within the master node.
2232 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2238 * Look up the file system's value for the property. For the
2239 * version property, we look up a slightly different string.
2241 if (prop == ZFS_PROP_VERSION)
2242 pname = ZPL_VERSION_STR;
2244 pname = zfs_prop_to_name(prop);
2247 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2249 if (error == ENOENT) {
2250 /* No value set, use the default value */
2252 case ZFS_PROP_VERSION:
2253 *value = ZPL_VERSION;
2255 case ZFS_PROP_NORMALIZE:
2256 case ZFS_PROP_UTF8ONLY:
2260 *value = ZFS_CASE_SENSITIVE;
2270 static vfsdef_t vfw = {
2274 VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2279 struct modlfs zfs_modlfs = {
2280 &mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw