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"
70 vfsops_t *zfs_vfsops = NULL;
71 static major_t zfs_major;
72 static minor_t zfs_minor;
73 static kmutex_t zfs_dev_mtx;
75 extern int sys_shutdown;
77 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
78 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
79 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
80 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
81 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
82 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
83 static void zfs_freevfs(vfs_t *vfsp);
85 static const fs_operation_def_t zfs_vfsops_template[] = {
86 VFSNAME_MOUNT, { .vfs_mount = zfs_mount },
87 VFSNAME_MOUNTROOT, { .vfs_mountroot = zfs_mountroot },
88 VFSNAME_UNMOUNT, { .vfs_unmount = zfs_umount },
89 VFSNAME_ROOT, { .vfs_root = zfs_root },
90 VFSNAME_STATVFS, { .vfs_statvfs = zfs_statvfs },
91 VFSNAME_SYNC, { .vfs_sync = zfs_sync },
92 VFSNAME_VGET, { .vfs_vget = zfs_vget },
93 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
97 static const fs_operation_def_t zfs_vfsops_eio_template[] = {
98 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
103 * We need to keep a count of active fs's.
104 * This is necessary to prevent our module
105 * from being unloaded after a umount -f
107 static uint32_t zfs_active_fs_count = 0;
109 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
110 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
111 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
112 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
115 * MO_DEFAULT is not used since the default value is determined
116 * by the equivalent property.
118 static mntopt_t mntopts[] = {
119 { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
120 { MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
121 { MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
122 { MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
125 static mntopts_t zfs_mntopts = {
126 sizeof (mntopts) / sizeof (mntopt_t),
132 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
135 * Data integrity is job one. We don't want a compromised kernel
136 * writing to the storage pool, so we never sync during panic.
142 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
143 * to sync metadata, which they would otherwise cache indefinitely.
144 * Semantically, the only requirement is that the sync be initiated.
145 * The DMU syncs out txgs frequently, so there's nothing to do.
147 if (flag & SYNC_ATTR)
152 * Sync a specific filesystem.
154 zfsvfs_t *zfsvfs = vfsp->vfs_data;
158 dp = dmu_objset_pool(zfsvfs->z_os);
161 * If the system is shutting down, then skip any
162 * filesystems which may exist on a suspended pool.
164 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
169 if (zfsvfs->z_log != NULL)
170 zil_commit(zfsvfs->z_log, 0);
175 * Sync all ZFS filesystems. This is what happens when you
176 * run sync(1M). Unlike other filesystems, ZFS honors the
177 * request by waiting for all pools to commit all dirty data.
186 zfs_create_unique_device(dev_t *dev)
191 ASSERT3U(zfs_minor, <=, MAXMIN32);
192 minor_t start = zfs_minor;
194 mutex_enter(&zfs_dev_mtx);
195 if (zfs_minor >= MAXMIN32) {
197 * If we're still using the real major
198 * keep out of /dev/zfs and /dev/zvol minor
199 * number space. If we're using a getudev()'ed
200 * major number, we can use all of its minors.
202 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
203 zfs_minor = ZFS_MIN_MINOR;
209 *dev = makedevice(zfs_major, zfs_minor);
210 mutex_exit(&zfs_dev_mtx);
211 } while (vfs_devismounted(*dev) && zfs_minor != start);
212 if (zfs_minor == start) {
214 * We are using all ~262,000 minor numbers for the
215 * current major number. Create a new major number.
217 if ((new_major = getudev()) == (major_t)-1) {
219 "zfs_mount: Can't get unique major "
223 mutex_enter(&zfs_dev_mtx);
224 zfs_major = new_major;
227 mutex_exit(&zfs_dev_mtx);
231 /* CONSTANTCONDITION */
238 atime_changed_cb(void *arg, uint64_t newval)
240 zfsvfs_t *zfsvfs = arg;
242 if (newval == TRUE) {
243 zfsvfs->z_atime = TRUE;
244 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
245 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
247 zfsvfs->z_atime = FALSE;
248 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
249 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
254 xattr_changed_cb(void *arg, uint64_t newval)
256 zfsvfs_t *zfsvfs = arg;
258 if (newval == TRUE) {
259 /* XXX locking on vfs_flag? */
260 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
261 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
262 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
264 /* XXX locking on vfs_flag? */
265 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
266 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
267 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
272 blksz_changed_cb(void *arg, uint64_t newval)
274 zfsvfs_t *zfsvfs = arg;
276 if (newval < SPA_MINBLOCKSIZE ||
277 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
278 newval = SPA_MAXBLOCKSIZE;
280 zfsvfs->z_max_blksz = newval;
281 zfsvfs->z_vfs->vfs_bsize = newval;
285 readonly_changed_cb(void *arg, uint64_t newval)
287 zfsvfs_t *zfsvfs = arg;
290 /* XXX locking on vfs_flag? */
291 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
292 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
293 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
295 /* XXX locking on vfs_flag? */
296 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
297 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
298 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
303 devices_changed_cb(void *arg, uint64_t newval)
305 zfsvfs_t *zfsvfs = arg;
307 if (newval == FALSE) {
308 zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
309 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
310 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
312 zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
313 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
314 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
319 setuid_changed_cb(void *arg, uint64_t newval)
321 zfsvfs_t *zfsvfs = arg;
323 if (newval == FALSE) {
324 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
325 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
326 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
328 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
329 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
330 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
335 exec_changed_cb(void *arg, uint64_t newval)
337 zfsvfs_t *zfsvfs = arg;
339 if (newval == FALSE) {
340 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
341 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
342 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
344 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
345 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
346 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
351 * The nbmand mount option can be changed at mount time.
352 * We can't allow it to be toggled on live file systems or incorrect
353 * behavior may be seen from cifs clients
355 * This property isn't registered via dsl_prop_register(), but this callback
356 * will be called when a file system is first mounted
359 nbmand_changed_cb(void *arg, uint64_t newval)
361 zfsvfs_t *zfsvfs = arg;
362 if (newval == FALSE) {
363 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
364 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
366 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
367 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
372 snapdir_changed_cb(void *arg, uint64_t newval)
374 zfsvfs_t *zfsvfs = arg;
376 zfsvfs->z_show_ctldir = newval;
380 vscan_changed_cb(void *arg, uint64_t newval)
382 zfsvfs_t *zfsvfs = arg;
384 zfsvfs->z_vscan = newval;
388 acl_inherit_changed_cb(void *arg, uint64_t newval)
390 zfsvfs_t *zfsvfs = arg;
392 zfsvfs->z_acl_inherit = newval;
396 zfs_register_callbacks(vfs_t *vfsp)
398 struct dsl_dataset *ds = NULL;
400 zfsvfs_t *zfsvfs = NULL;
402 int readonly, do_readonly = B_FALSE;
403 int setuid, do_setuid = B_FALSE;
404 int exec, do_exec = B_FALSE;
405 int devices, do_devices = B_FALSE;
406 int xattr, do_xattr = B_FALSE;
407 int atime, do_atime = B_FALSE;
411 zfsvfs = vfsp->vfs_data;
416 * The act of registering our callbacks will destroy any mount
417 * options we may have. In order to enable temporary overrides
418 * of mount options, we stash away the current values and
419 * restore them after we register the callbacks.
421 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
422 !spa_writeable(dmu_objset_spa(os))) {
424 do_readonly = B_TRUE;
425 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
427 do_readonly = B_TRUE;
429 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
435 if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
438 } else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
443 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
446 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
451 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
454 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
458 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
461 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
465 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
468 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
474 * nbmand is a special property. It can only be changed at
477 * This is weird, but it is documented to only be changeable
480 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
482 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
485 char osname[MAXNAMELEN];
487 dmu_objset_name(os, osname);
488 if ((error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
495 * Register property callbacks.
497 * It would probably be fine to just check for i/o error from
498 * the first prop_register(), but I guess I like to go
501 ds = dmu_objset_ds(os);
502 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs);
503 error = error ? error : dsl_prop_register(ds,
504 "xattr", xattr_changed_cb, zfsvfs);
505 error = error ? error : dsl_prop_register(ds,
506 "recordsize", blksz_changed_cb, zfsvfs);
507 error = error ? error : dsl_prop_register(ds,
508 "readonly", readonly_changed_cb, zfsvfs);
509 error = error ? error : dsl_prop_register(ds,
510 "devices", devices_changed_cb, zfsvfs);
511 error = error ? error : dsl_prop_register(ds,
512 "setuid", setuid_changed_cb, zfsvfs);
513 error = error ? error : dsl_prop_register(ds,
514 "exec", exec_changed_cb, zfsvfs);
515 error = error ? error : dsl_prop_register(ds,
516 "snapdir", snapdir_changed_cb, zfsvfs);
517 error = error ? error : dsl_prop_register(ds,
518 "aclinherit", acl_inherit_changed_cb, zfsvfs);
519 error = error ? error : dsl_prop_register(ds,
520 "vscan", vscan_changed_cb, zfsvfs);
525 * Invoke our callbacks to restore temporary mount options.
528 readonly_changed_cb(zfsvfs, readonly);
530 setuid_changed_cb(zfsvfs, setuid);
532 exec_changed_cb(zfsvfs, exec);
534 devices_changed_cb(zfsvfs, devices);
536 xattr_changed_cb(zfsvfs, xattr);
538 atime_changed_cb(zfsvfs, atime);
540 nbmand_changed_cb(zfsvfs, nbmand);
546 * We may attempt to unregister some callbacks that are not
547 * registered, but this is OK; it will simply return ENOMSG,
548 * which we will ignore.
550 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs);
551 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs);
552 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs);
553 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs);
554 (void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zfsvfs);
555 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs);
556 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs);
557 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs);
558 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
560 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs);
564 #endif /* HAVE_ZPL */
567 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
568 uint64_t *userp, uint64_t *groupp)
570 znode_phys_t *znp = data;
574 * Is it a valid type of object to track?
576 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
580 * If we have a NULL data pointer
581 * then assume the id's aren't changing and
582 * return EEXIST to the dmu to let it know to
588 if (bonustype == DMU_OT_ZNODE) {
589 *userp = znp->zp_uid;
590 *groupp = znp->zp_gid;
594 ASSERT(bonustype == DMU_OT_SA);
595 hdrsize = sa_hdrsize(data);
598 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
600 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
604 * This should only happen for newly created
605 * files that haven't had the znode data filled
617 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
618 char *domainbuf, int buflen, uid_t *ridp)
623 fuid = strtonum(fuidstr, NULL);
625 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
627 (void) strlcpy(domainbuf, domain, buflen);
630 *ridp = FUID_RID(fuid);
634 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
637 case ZFS_PROP_USERUSED:
638 return (DMU_USERUSED_OBJECT);
639 case ZFS_PROP_GROUPUSED:
640 return (DMU_GROUPUSED_OBJECT);
641 case ZFS_PROP_USERQUOTA:
642 return (zfsvfs->z_userquota_obj);
643 case ZFS_PROP_GROUPQUOTA:
644 return (zfsvfs->z_groupquota_obj);
652 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
653 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
658 zfs_useracct_t *buf = vbuf;
661 if (!dmu_objset_userspace_present(zfsvfs->z_os))
664 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
670 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
671 (error = zap_cursor_retrieve(&zc, &za)) == 0;
672 zap_cursor_advance(&zc)) {
673 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
677 fuidstr_to_sid(zfsvfs, za.za_name,
678 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
680 buf->zu_space = za.za_first_integer;
686 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
687 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
688 *cookiep = zap_cursor_serialize(&zc);
689 zap_cursor_fini(&zc);
694 * buf must be big enough (eg, 32 bytes)
697 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
698 char *buf, boolean_t addok)
703 if (domain && domain[0]) {
704 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
708 fuid = FUID_ENCODE(domainid, rid);
709 (void) sprintf(buf, "%llx", (longlong_t)fuid);
714 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
715 const char *domain, uint64_t rid, uint64_t *valp)
723 if (!dmu_objset_userspace_present(zfsvfs->z_os))
726 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
730 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
734 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
741 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
742 const char *domain, uint64_t rid, uint64_t quota)
748 boolean_t fuid_dirtied;
750 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
753 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
756 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
757 &zfsvfs->z_groupquota_obj;
759 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
762 fuid_dirtied = zfsvfs->z_fuid_dirty;
764 tx = dmu_tx_create(zfsvfs->z_os);
765 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
767 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
768 zfs_userquota_prop_prefixes[type]);
771 zfs_fuid_txhold(zfsvfs, tx);
772 err = dmu_tx_assign(tx, TXG_WAIT);
778 mutex_enter(&zfsvfs->z_lock);
780 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
782 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
783 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
785 mutex_exit(&zfsvfs->z_lock);
788 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
792 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
796 zfs_fuid_sync(zfsvfs, tx);
802 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
805 uint64_t used, quota, usedobj, quotaobj;
808 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
809 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
811 if (quotaobj == 0 || zfsvfs->z_replay)
814 (void) sprintf(buf, "%llx", (longlong_t)fuid);
815 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
819 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
822 return (used >= quota);
826 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
831 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
833 fuid = isgroup ? zp->z_gid : zp->z_uid;
835 if (quotaobj == 0 || zfsvfs->z_replay)
838 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
842 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
850 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
853 * We claim to always be readonly so we can open snapshots;
854 * other ZPL code will prevent us from writing to snapshots.
856 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
858 kmem_free(zfsvfs, sizeof (zfsvfs_t));
863 * Initialize the zfs-specific filesystem structure.
864 * Should probably make this a kmem cache, shuffle fields,
865 * and just bzero up to z_hold_mtx[].
867 zfsvfs->z_vfs = NULL;
868 zfsvfs->z_parent = zfsvfs;
869 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
870 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
873 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
876 } else if (zfsvfs->z_version >
877 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
878 (void) printk("Can't mount a version %lld file system "
879 "on a version %lld pool\n. Pool must be upgraded to mount "
880 "this file system.", (u_longlong_t)zfsvfs->z_version,
881 (u_longlong_t)spa_version(dmu_objset_spa(os)));
885 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
887 zfsvfs->z_norm = (int)zval;
889 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
891 zfsvfs->z_utf8 = (zval != 0);
893 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
895 zfsvfs->z_case = (uint_t)zval;
898 * Fold case on file systems that are always or sometimes case
901 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
902 zfsvfs->z_case == ZFS_CASE_MIXED)
903 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
905 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
906 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
908 if (zfsvfs->z_use_sa) {
909 /* should either have both of these objects or none */
910 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
916 * Pre SA versions file systems should never touch
917 * either the attribute registration or layout objects.
922 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
923 &zfsvfs->z_attr_table);
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 (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1047 if (zil_replay_disable) {
1048 zil_destroy(zfsvfs->z_log, B_FALSE);
1050 zfsvfs->z_replay = B_TRUE;
1051 zil_replay(zfsvfs->z_os, zfsvfs,
1053 zfsvfs->z_replay = B_FALSE;
1056 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1063 zfsvfs_free(zfsvfs_t *zfsvfs)
1066 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1069 * This is a barrier to prevent the filesystem from going away in
1070 * zfs_znode_move() until we can safely ensure that the filesystem is
1071 * not unmounted. We consider the filesystem valid before the barrier
1072 * and invalid after the barrier.
1074 rw_enter(&zfsvfs_lock, RW_READER);
1075 rw_exit(&zfsvfs_lock);
1077 zfs_fuid_destroy(zfsvfs);
1079 mutex_destroy(&zfsvfs->z_znodes_lock);
1080 mutex_destroy(&zfsvfs->z_lock);
1081 list_destroy(&zfsvfs->z_all_znodes);
1082 rrw_destroy(&zfsvfs->z_teardown_lock);
1083 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1084 rw_destroy(&zfsvfs->z_fuid_lock);
1085 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1086 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1087 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1091 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1093 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1094 if (zfsvfs->z_use_fuids && zfsvfs->z_vfs) {
1095 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1096 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1097 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1098 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1099 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1100 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1102 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1106 zfs_domount(vfs_t *vfsp, char *osname)
1109 uint64_t recordsize, fsid_guid;
1116 error = zfsvfs_create(osname, &zfsvfs);
1119 zfsvfs->z_vfs = vfsp;
1121 /* Initialize the generic filesystem structure. */
1122 vfsp->vfs_bcount = 0;
1123 vfsp->vfs_data = NULL;
1125 if (zfs_create_unique_device(&mount_dev) == -1) {
1129 ASSERT(vfs_devismounted(mount_dev) == 0);
1131 if ((error = dsl_prop_get_integer(osname, "recordsize",
1132 &recordsize, NULL)))
1135 vfsp->vfs_dev = mount_dev;
1136 vfsp->vfs_fstype = zfsfstype;
1137 vfsp->vfs_bsize = recordsize;
1138 vfsp->vfs_flag |= VFS_NOTRUNC;
1139 vfsp->vfs_data = zfsvfs;
1142 * The fsid is 64 bits, composed of an 8-bit fs type, which
1143 * separates our fsid from any other filesystem types, and a
1144 * 56-bit objset unique ID. The objset unique ID is unique to
1145 * all objsets open on this system, provided by unique_create().
1146 * The 8-bit fs type must be put in the low bits of fsid[1]
1147 * because that's where other Solaris filesystems put it.
1149 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1150 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1151 vfsp->vfs_fsid.val[0] = fsid_guid;
1152 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1156 * Set features for file system.
1158 zfs_set_fuid_feature(zfsvfs);
1159 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1160 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1161 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1162 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1163 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1164 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1165 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1167 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1169 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1172 atime_changed_cb(zfsvfs, B_FALSE);
1173 readonly_changed_cb(zfsvfs, B_TRUE);
1174 if ((error = dsl_prop_get_integer(osname,"xattr",&pval,NULL)))
1176 xattr_changed_cb(zfsvfs, pval);
1177 zfsvfs->z_issnap = B_TRUE;
1178 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1180 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1181 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1182 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1184 error = zfsvfs_setup(zfsvfs, B_TRUE);
1187 if (!zfsvfs->z_issnap)
1188 zfsctl_create(zfsvfs);
1191 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1192 zfsvfs_free(zfsvfs);
1194 atomic_add_32(&zfs_active_fs_count, 1);
1201 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1203 objset_t *os = zfsvfs->z_os;
1204 struct dsl_dataset *ds;
1207 * Unregister properties.
1209 if (!dmu_objset_is_snapshot(os)) {
1210 ds = dmu_objset_ds(os);
1211 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1214 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1217 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1220 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1223 VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1226 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1229 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1232 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1235 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1236 acl_inherit_changed_cb, zfsvfs) == 0);
1238 VERIFY(dsl_prop_unregister(ds, "vscan",
1239 vscan_changed_cb, zfsvfs) == 0);
1244 * Convert a decimal digit string to a uint64_t integer.
1247 str_to_uint64(char *str, uint64_t *objnum)
1252 if (*str < '0' || *str > '9')
1255 num = num*10 + *str++ - '0';
1263 * The boot path passed from the boot loader is in the form of
1264 * "rootpool-name/root-filesystem-object-number'. Convert this
1265 * string to a dataset name: "rootpool-name/root-filesystem-name".
1268 zfs_parse_bootfs(char *bpath, char *outpath)
1274 if (*bpath == 0 || *bpath == '/')
1277 (void) strcpy(outpath, bpath);
1279 slashp = strchr(bpath, '/');
1281 /* if no '/', just return the pool name */
1282 if (slashp == NULL) {
1286 /* if not a number, just return the root dataset name */
1287 if (str_to_uint64(slashp+1, &objnum)) {
1292 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1299 * zfs_check_global_label:
1300 * Check that the hex label string is appropriate for the dataset
1301 * being mounted into the global_zone proper.
1303 * Return an error if the hex label string is not default or
1304 * admin_low/admin_high. For admin_low labels, the corresponding
1305 * dataset must be readonly.
1308 zfs_check_global_label(const char *dsname, const char *hexsl)
1310 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1312 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1314 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1315 /* must be readonly */
1318 if (dsl_prop_get_integer(dsname,
1319 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1321 return (rdonly ? 0 : EACCES);
1327 * zfs_mount_label_policy:
1328 * Determine whether the mount is allowed according to MAC check.
1329 * by comparing (where appropriate) label of the dataset against
1330 * the label of the zone being mounted into. If the dataset has
1331 * no label, create one.
1334 * 0 : access allowed
1335 * >0 : error code, such as EACCES
1338 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1341 zone_t *mntzone = NULL;
1342 ts_label_t *mnt_tsl;
1345 char ds_hexsl[MAXNAMELEN];
1347 retv = EACCES; /* assume the worst */
1350 * Start by getting the dataset label if it exists.
1352 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1353 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1358 * If labeling is NOT enabled, then disallow the mount of datasets
1359 * which have a non-default label already. No other label checks
1362 if (!is_system_labeled()) {
1363 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1369 * Get the label of the mountpoint. If mounting into the global
1370 * zone (i.e. mountpoint is not within an active zone and the
1371 * zoned property is off), the label must be default or
1372 * admin_low/admin_high only; no other checks are needed.
1374 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1375 if (mntzone->zone_id == GLOBAL_ZONEID) {
1380 if (dsl_prop_get_integer(osname,
1381 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1384 return (zfs_check_global_label(osname, ds_hexsl));
1387 * This is the case of a zone dataset being mounted
1388 * initially, before the zone has been fully created;
1389 * allow this mount into global zone.
1394 mnt_tsl = mntzone->zone_slabel;
1395 ASSERT(mnt_tsl != NULL);
1396 label_hold(mnt_tsl);
1397 mnt_sl = label2bslabel(mnt_tsl);
1399 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1401 * The dataset doesn't have a real label, so fabricate one.
1405 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1406 dsl_prop_set(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1407 ZPROP_SRC_LOCAL, 1, strlen(str) + 1, str) == 0)
1410 kmem_free(str, strlen(str) + 1);
1411 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1413 * Now compare labels to complete the MAC check. If the
1414 * labels are equal then allow access. If the mountpoint
1415 * label dominates the dataset label, allow readonly access.
1416 * Otherwise, access is denied.
1418 if (blequal(mnt_sl, &ds_sl))
1420 else if (bldominates(mnt_sl, &ds_sl)) {
1421 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1426 label_rele(mnt_tsl);
1432 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1435 static int zfsrootdone = 0;
1436 zfsvfs_t *zfsvfs = NULL;
1445 * The filesystem that we mount as root is defined in the
1446 * boot property "zfs-bootfs" with a format of
1447 * "poolname/root-dataset-objnum".
1449 if (why == ROOT_INIT) {
1453 * the process of doing a spa_load will require the
1454 * clock to be set before we could (for example) do
1455 * something better by looking at the timestamp on
1456 * an uberblock, so just set it to -1.
1460 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1461 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1465 zfs_devid = spa_get_bootprop("diskdevid");
1466 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1468 spa_free_bootprop(zfs_devid);
1470 spa_free_bootprop(zfs_bootfs);
1471 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1475 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1476 spa_free_bootprop(zfs_bootfs);
1477 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1482 spa_free_bootprop(zfs_bootfs);
1484 if (error = vfs_lock(vfsp))
1487 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1488 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1492 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1494 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1495 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1500 mutex_enter(&vp->v_lock);
1501 vp->v_flag |= VROOT;
1502 mutex_exit(&vp->v_lock);
1506 * Leave rootvp held. The root file system is never unmounted.
1509 vfs_add((struct vnode *)0, vfsp,
1510 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1514 } else if (why == ROOT_REMOUNT) {
1515 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1516 vfsp->vfs_flag |= VFS_REMOUNT;
1518 /* refresh mount options */
1519 zfs_unregister_callbacks(vfsp->vfs_data);
1520 return (zfs_register_callbacks(vfsp));
1522 } else if (why == ROOT_UNMOUNT) {
1523 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1524 (void) zfs_sync(vfsp, 0, 0);
1529 * if "why" is equal to anything else other than ROOT_INIT,
1530 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1537 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1542 uio_seg_t fromspace = (uap->flags & MS_SYSSPACE) ?
1543 UIO_SYSSPACE : UIO_USERSPACE;
1546 if (mvp->v_type != VDIR)
1549 mutex_enter(&mvp->v_lock);
1550 if ((uap->flags & MS_REMOUNT) == 0 &&
1551 (uap->flags & MS_OVERLAY) == 0 &&
1552 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1553 mutex_exit(&mvp->v_lock);
1556 mutex_exit(&mvp->v_lock);
1559 * ZFS does not support passing unparsed data in via MS_DATA.
1560 * Users should use the MS_OPTIONSTR interface; this means
1561 * that all option parsing is already done and the options struct
1562 * can be interrogated.
1564 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1568 * Get the objset name (the "special" mount argument).
1570 if ((error = pn_get(uap->spec, fromspace, &spn)))
1573 osname = spn.pn_path;
1576 * Check for mount privilege?
1578 * If we don't have privilege then see if
1579 * we have local permission to allow it
1581 error = secpolicy_fs_mount(cr, mvp, vfsp);
1583 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1587 * Make sure user is the owner of the mount point
1588 * or has sufficient privileges.
1591 vattr.va_mask = AT_UID;
1593 if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1597 if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1598 VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1601 secpolicy_fs_mount_clearopts(cr, vfsp);
1608 * Refuse to mount a filesystem if we are in a local zone and the
1609 * dataset is not visible.
1611 if (!INGLOBALZONE(curproc) &&
1612 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1617 error = zfs_mount_label_policy(vfsp, osname);
1622 * When doing a remount, we simply refresh our temporary properties
1623 * according to those options set in the current VFS options.
1625 if (uap->flags & MS_REMOUNT) {
1626 /* refresh mount options */
1627 zfs_unregister_callbacks(vfsp->vfs_data);
1628 error = zfs_register_callbacks(vfsp);
1632 error = zfs_domount(vfsp, osname);
1635 * Add an extra VFS_HOLD on our parent vfs so that it can't
1636 * disappear due to a forced unmount.
1638 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1639 VFS_HOLD(mvp->v_vfsp);
1647 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1649 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1651 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1655 dmu_objset_space(zfsvfs->z_os,
1656 &refdbytes, &availbytes, &usedobjs, &availobjs);
1659 * The underlying storage pool actually uses multiple block sizes.
1660 * We report the fragsize as the smallest block size we support,
1661 * and we report our blocksize as the filesystem's maximum blocksize.
1663 statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1664 statp->f_bsize = zfsvfs->z_max_blksz;
1667 * The following report "total" blocks of various kinds in the
1668 * file system, but reported in terms of f_frsize - the
1672 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1673 statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1674 statp->f_bavail = statp->f_bfree; /* no root reservation */
1677 * statvfs() should really be called statufs(), because it assumes
1678 * static metadata. ZFS doesn't preallocate files, so the best
1679 * we can do is report the max that could possibly fit in f_files,
1680 * and that minus the number actually used in f_ffree.
1681 * For f_ffree, report the smaller of the number of object available
1682 * and the number of blocks (each object will take at least a block).
1684 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1685 statp->f_favail = statp->f_ffree; /* no "root reservation" */
1686 statp->f_files = statp->f_ffree + usedobjs;
1688 (void) cmpldev(&d32, vfsp->vfs_dev);
1689 statp->f_fsid = d32;
1692 * We're a zfs filesystem.
1694 (void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1696 statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1698 statp->f_namemax = ZFS_MAXNAMELEN;
1701 * We have all of 32 characters to stuff a string here.
1702 * Is there anything useful we could/should provide?
1704 bzero(statp->f_fstr, sizeof (statp->f_fstr));
1711 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1713 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1719 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1721 *vpp = ZTOV(rootzp);
1728 * Teardown the zfsvfs::z_os.
1730 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1731 * and 'z_teardown_inactive_lock' held.
1734 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1738 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1742 * We purge the parent filesystem's vfsp as the parent
1743 * filesystem and all of its snapshots have their vnode's
1744 * v_vfsp set to the parent's filesystem's vfsp. Note,
1745 * 'z_parent' is self referential for non-snapshots.
1747 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1751 * Close the zil. NB: Can't close the zil while zfs_inactive
1752 * threads are blocked as zil_close can call zfs_inactive.
1754 if (zfsvfs->z_log) {
1755 zil_close(zfsvfs->z_log);
1756 zfsvfs->z_log = NULL;
1759 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1762 * If we are not unmounting (ie: online recv) and someone already
1763 * unmounted this file system while we were doing the switcheroo,
1764 * or a reopen of z_os failed then just bail out now.
1766 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1767 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1768 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1773 * At this point there are no vops active, and any new vops will
1774 * fail with EIO since we have z_teardown_lock for writer (only
1775 * relavent for forced unmount).
1777 * Release all holds on dbufs.
1779 mutex_enter(&zfsvfs->z_znodes_lock);
1780 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1781 zp = list_next(&zfsvfs->z_all_znodes, zp))
1783 ASSERT(ZTOV(zp)->v_count > 0);
1784 zfs_znode_dmu_fini(zp);
1786 mutex_exit(&zfsvfs->z_znodes_lock);
1789 * If we are unmounting, set the unmounted flag and let new vops
1790 * unblock. zfs_inactive will have the unmounted behavior, and all
1791 * other vops will fail with EIO.
1794 zfsvfs->z_unmounted = B_TRUE;
1795 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1796 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1800 * z_os will be NULL if there was an error in attempting to reopen
1801 * zfsvfs, so just return as the properties had already been
1802 * unregistered and cached data had been evicted before.
1804 if (zfsvfs->z_os == NULL)
1808 * Unregister properties.
1810 zfs_unregister_callbacks(zfsvfs);
1815 if (dmu_objset_is_dirty_anywhere(zfsvfs->z_os))
1816 if (!(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1817 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1818 (void) dmu_objset_evict_dbufs(zfsvfs->z_os);
1825 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1827 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1831 ret = secpolicy_fs_unmount(cr, vfsp);
1833 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1834 ZFS_DELEG_PERM_MOUNT, cr))
1839 * We purge the parent filesystem's vfsp as the parent filesystem
1840 * and all of its snapshots have their vnode's v_vfsp set to the
1841 * parent's filesystem's vfsp. Note, 'z_parent' is self
1842 * referential for non-snapshots.
1844 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1847 * Unmount any snapshots mounted under .zfs before unmounting the
1850 if (zfsvfs->z_ctldir != NULL &&
1851 (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1855 if (!(fflag & MS_FORCE)) {
1857 * Check the number of active vnodes in the file system.
1858 * Our count is maintained in the vfs structure, but the
1859 * number is off by 1 to indicate a hold on the vfs
1862 * The '.zfs' directory maintains a reference of its
1863 * own, and any active references underneath are
1864 * reflected in the vnode count.
1866 if (zfsvfs->z_ctldir == NULL) {
1867 if (vfsp->vfs_count > 1)
1870 if (vfsp->vfs_count > 2 ||
1871 zfsvfs->z_ctldir->v_count > 1)
1876 vfsp->vfs_flag |= VFS_UNMOUNTED;
1878 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1882 * z_os will be NULL if there was an error in
1883 * attempting to reopen zfsvfs.
1887 * Unset the objset user_ptr.
1889 mutex_enter(&os->os_user_ptr_lock);
1890 dmu_objset_set_user(os, NULL);
1891 mutex_exit(&os->os_user_ptr_lock);
1894 * Finally release the objset
1896 dmu_objset_disown(os, zfsvfs);
1900 * We can now safely destroy the '.zfs' directory node.
1902 if (zfsvfs->z_ctldir != NULL)
1903 zfsctl_destroy(zfsvfs);
1909 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1911 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1913 uint64_t object = 0;
1914 uint64_t fid_gen = 0;
1923 if (fidp->fid_len == LONG_FID_LEN) {
1924 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1925 uint64_t objsetid = 0;
1926 uint64_t setgen = 0;
1928 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1929 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1931 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1932 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1936 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1942 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1943 zfid_short_t *zfid = (zfid_short_t *)fidp;
1945 for (i = 0; i < sizeof (zfid->zf_object); i++)
1946 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1948 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1949 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1955 /* A zero fid_gen means we are in the .zfs control directories */
1957 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1958 *vpp = zfsvfs->z_ctldir;
1959 ASSERT(*vpp != NULL);
1960 if (object == ZFSCTL_INO_SNAPDIR) {
1961 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
1962 0, NULL, NULL, NULL, NULL, NULL) == 0);
1970 gen_mask = -1ULL >> (64 - 8 * i);
1972 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1973 if ((err = zfs_zget(zfsvfs, object, &zp))) {
1977 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1979 zp_gen = zp_gen & gen_mask;
1982 if (zp->z_unlinked || zp_gen != fid_gen) {
1983 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1995 * Block out VOPs and close zfsvfs_t::z_os
1997 * Note, if successful, then we return with the 'z_teardown_lock' and
1998 * 'z_teardown_inactive_lock' write held.
2001 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2005 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2007 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
2013 * Reopen zfsvfs_t::z_os and release VOPs.
2016 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2020 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
2021 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2023 err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zfsvfs,
2026 zfsvfs->z_os = NULL;
2029 uint64_t sa_obj = 0;
2031 err2 = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2032 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2034 if ((err || err2) && zfsvfs->z_version >= ZPL_VERSION_SA)
2038 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2039 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
2042 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2045 * Attempt to re-establish all the active znodes with
2046 * their dbufs. If a zfs_rezget() fails, then we'll let
2047 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2048 * when they try to use their znode.
2050 mutex_enter(&zfsvfs->z_znodes_lock);
2051 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2052 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2053 (void) zfs_rezget(zp);
2055 mutex_exit(&zfsvfs->z_znodes_lock);
2060 /* release the VOPs */
2061 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2062 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
2066 * Since we couldn't reopen zfsvfs::z_os, force
2067 * unmount this file system.
2069 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2070 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2076 zfs_freevfs(vfs_t *vfsp)
2078 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2081 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2082 * from zfs_mount(). Release it here. If we came through
2083 * zfs_mountroot() instead, we didn't grab an extra hold, so
2084 * skip the VFS_RELE for rootvfs.
2086 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2087 VFS_RELE(zfsvfs->z_parent->z_vfs);
2089 zfsvfs_free(zfsvfs);
2091 atomic_add_32(&zfs_active_fs_count, -1);
2095 * VFS_INIT() initialization. Note that there is no VFS_FINI(),
2096 * so we can't safely do any non-idempotent initialization here.
2097 * Leave that to zfs_init() and zfs_fini(), which are called
2098 * from the module's _init() and _fini() entry points.
2102 zfs_vfsinit(int fstype, char *name)
2109 * Setup vfsops and vnodeops tables.
2111 error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2113 cmn_err(CE_WARN, "zfs: bad vfs ops template");
2116 error = zfs_create_op_tables();
2118 zfs_remove_op_tables();
2119 cmn_err(CE_WARN, "zfs: bad vnode ops template");
2120 (void) vfs_freevfsops_by_type(zfsfstype);
2124 mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2127 * Unique major number for all zfs mounts.
2128 * If we run out of 32-bit minors, we'll getudev() another major.
2130 zfs_major = ddi_name_to_major(ZFS_DRIVER);
2131 zfs_minor = ZFS_MIN_MINOR;
2135 #endif /* HAVE_ZPL */
2142 * Initialize .zfs directory structures
2147 * Initialize znode cache, vnode ops, etc...
2150 #endif /* HAVE_ZPL */
2152 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2161 #endif /* HAVE_ZPL */
2166 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2169 objset_t *os = zfsvfs->z_os;
2172 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2175 if (newvers < zfsvfs->z_version)
2178 if (zfs_spa_version_map(newvers) >
2179 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2182 tx = dmu_tx_create(os);
2183 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2184 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2185 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2187 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2189 error = dmu_tx_assign(tx, TXG_WAIT);
2195 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2196 8, 1, &newvers, tx);
2203 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2206 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2208 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2209 DMU_OT_NONE, 0, tx);
2211 error = zap_add(os, MASTER_NODE_OBJ,
2212 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2213 ASSERT3U(error, ==, 0);
2215 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2216 sa_register_update_callback(os, zfs_sa_upgrade);
2219 spa_history_log_internal(LOG_DS_UPGRADE,
2220 dmu_objset_spa(os), tx, "oldver=%llu newver=%llu dataset = %llu",
2221 zfsvfs->z_version, newvers, dmu_objset_id(os));
2225 zfsvfs->z_version = newvers;
2227 if (zfsvfs->z_version >= ZPL_VERSION_FUID)
2228 zfs_set_fuid_feature(zfsvfs);
2232 #endif /* HAVE_ZPL */
2235 * Read a property stored within the master node.
2238 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2244 * Look up the file system's value for the property. For the
2245 * version property, we look up a slightly different string.
2247 if (prop == ZFS_PROP_VERSION)
2248 pname = ZPL_VERSION_STR;
2250 pname = zfs_prop_to_name(prop);
2253 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2255 if (error == ENOENT) {
2256 /* No value set, use the default value */
2258 case ZFS_PROP_VERSION:
2259 *value = ZPL_VERSION;
2261 case ZFS_PROP_NORMALIZE:
2262 case ZFS_PROP_UTF8ONLY:
2266 *value = ZFS_CASE_SENSITIVE;
2277 static vfsdef_t vfw = {
2281 VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2286 struct modlfs zfs_modlfs = {
2287 &mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2289 #endif /* HAVE_ZPL */