4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012 by Delphix. All rights reserved.
26 /* Portions Copyright 2010 Robert Milkowski */
28 #include <sys/types.h>
29 #include <sys/param.h>
30 #include <sys/systm.h>
31 #include <sys/sysmacros.h>
33 #include <sys/pathname.h>
34 #include <sys/vnode.h>
36 #include <sys/vfs_opreg.h>
37 #include <sys/mntent.h>
38 #include <sys/mount.h>
39 #include <sys/cmn_err.h>
40 #include "fs/fs_subr.h"
41 #include <sys/zfs_znode.h>
42 #include <sys/zfs_dir.h>
44 #include <sys/fs/zfs.h>
46 #include <sys/dsl_prop.h>
47 #include <sys/dsl_dataset.h>
48 #include <sys/dsl_deleg.h>
52 #include <sys/sa_impl.h>
53 #include <sys/varargs.h>
54 #include <sys/policy.h>
55 #include <sys/atomic.h>
56 #include <sys/mkdev.h>
57 #include <sys/modctl.h>
58 #include <sys/refstr.h>
59 #include <sys/zfs_ioctl.h>
60 #include <sys/zfs_ctldir.h>
61 #include <sys/zfs_fuid.h>
62 #include <sys/bootconf.h>
63 #include <sys/sunddi.h>
65 #include <sys/dmu_objset.h>
66 #include <sys/spa_boot.h>
67 #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_mode_changed_cb(void *arg, uint64_t newval)
390 zfsvfs_t *zfsvfs = arg;
392 zfsvfs->z_acl_mode = newval;
396 acl_inherit_changed_cb(void *arg, uint64_t newval)
398 zfsvfs_t *zfsvfs = arg;
400 zfsvfs->z_acl_inherit = newval;
404 zfs_register_callbacks(vfs_t *vfsp)
406 struct dsl_dataset *ds = NULL;
408 zfsvfs_t *zfsvfs = NULL;
410 boolean_t readonly = B_FALSE;
411 boolean_t do_readonly = B_FALSE;
412 boolean_t setuid = B_FALSE;
413 boolean_t do_setuid = B_FALSE;
414 boolean_t exec = B_FALSE;
415 boolean_t do_exec = B_FALSE;
416 boolean_t devices = B_FALSE;
417 boolean_t do_devices = B_FALSE;
418 boolean_t xattr = B_FALSE;
419 boolean_t do_xattr = B_FALSE;
420 boolean_t atime = B_FALSE;
421 boolean_t do_atime = B_FALSE;
425 zfsvfs = vfsp->vfs_data;
430 * The act of registering our callbacks will destroy any mount
431 * options we may have. In order to enable temporary overrides
432 * of mount options, we stash away the current values and
433 * restore them after we register the callbacks.
435 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
436 !spa_writeable(dmu_objset_spa(os))) {
438 do_readonly = B_TRUE;
439 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
441 do_readonly = B_TRUE;
443 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
449 if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
452 } else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
457 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
460 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
465 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
468 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
472 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
475 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
479 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
482 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
488 * nbmand is a special property. It can only be changed at
491 * This is weird, but it is documented to only be changeable
494 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
496 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
499 char osname[MAXNAMELEN];
501 dmu_objset_name(os, osname);
502 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
509 * Register property callbacks.
511 * It would probably be fine to just check for i/o error from
512 * the first prop_register(), but I guess I like to go
515 ds = dmu_objset_ds(os);
516 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs);
517 error = error ? error : dsl_prop_register(ds,
518 "xattr", xattr_changed_cb, zfsvfs);
519 error = error ? error : dsl_prop_register(ds,
520 "recordsize", blksz_changed_cb, zfsvfs);
521 error = error ? error : dsl_prop_register(ds,
522 "readonly", readonly_changed_cb, zfsvfs);
523 error = error ? error : dsl_prop_register(ds,
524 "devices", devices_changed_cb, zfsvfs);
525 error = error ? error : dsl_prop_register(ds,
526 "setuid", setuid_changed_cb, zfsvfs);
527 error = error ? error : dsl_prop_register(ds,
528 "exec", exec_changed_cb, zfsvfs);
529 error = error ? error : dsl_prop_register(ds,
530 "snapdir", snapdir_changed_cb, zfsvfs);
531 error = error ? error : dsl_prop_register(ds,
532 "aclmode", acl_mode_changed_cb, zfsvfs);
533 error = error ? error : dsl_prop_register(ds,
534 "aclinherit", acl_inherit_changed_cb, zfsvfs);
535 error = error ? error : dsl_prop_register(ds,
536 "vscan", vscan_changed_cb, zfsvfs);
541 * Invoke our callbacks to restore temporary mount options.
544 readonly_changed_cb(zfsvfs, readonly);
546 setuid_changed_cb(zfsvfs, setuid);
548 exec_changed_cb(zfsvfs, exec);
550 devices_changed_cb(zfsvfs, devices);
552 xattr_changed_cb(zfsvfs, xattr);
554 atime_changed_cb(zfsvfs, atime);
556 nbmand_changed_cb(zfsvfs, nbmand);
562 * We may attempt to unregister some callbacks that are not
563 * registered, but this is OK; it will simply return ENOMSG,
564 * which we will ignore.
566 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs);
567 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs);
568 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs);
569 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs);
570 (void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zfsvfs);
571 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs);
572 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs);
573 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs);
574 (void) dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, zfsvfs);
575 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
577 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs);
583 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
584 uint64_t *userp, uint64_t *groupp)
589 * Is it a valid type of object to track?
591 if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
595 * If we have a NULL data pointer
596 * then assume the id's aren't changing and
597 * return EEXIST to the dmu to let it know to
603 if (bonustype == DMU_OT_ZNODE) {
604 znode_phys_t *znp = data;
605 *userp = znp->zp_uid;
606 *groupp = znp->zp_gid;
609 sa_hdr_phys_t *sap = data;
610 sa_hdr_phys_t sa = *sap;
611 boolean_t swap = B_FALSE;
613 ASSERT(bonustype == DMU_OT_SA);
615 if (sa.sa_magic == 0) {
617 * This should only happen for newly created
618 * files that haven't had the znode data filled
625 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
626 sa.sa_magic = SA_MAGIC;
627 sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
630 VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
633 hdrsize = sa_hdrsize(&sa);
634 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
635 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
637 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
640 *userp = BSWAP_64(*userp);
641 *groupp = BSWAP_64(*groupp);
648 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
649 char *domainbuf, int buflen, uid_t *ridp)
654 fuid = strtonum(fuidstr, NULL);
656 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
658 (void) strlcpy(domainbuf, domain, buflen);
661 *ridp = FUID_RID(fuid);
665 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
668 case ZFS_PROP_USERUSED:
669 return (DMU_USERUSED_OBJECT);
670 case ZFS_PROP_GROUPUSED:
671 return (DMU_GROUPUSED_OBJECT);
672 case ZFS_PROP_USERQUOTA:
673 return (zfsvfs->z_userquota_obj);
674 case ZFS_PROP_GROUPQUOTA:
675 return (zfsvfs->z_groupquota_obj);
681 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
682 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
687 zfs_useracct_t *buf = vbuf;
690 if (!dmu_objset_userspace_present(zfsvfs->z_os))
693 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
699 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
700 (error = zap_cursor_retrieve(&zc, &za)) == 0;
701 zap_cursor_advance(&zc)) {
702 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
706 fuidstr_to_sid(zfsvfs, za.za_name,
707 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
709 buf->zu_space = za.za_first_integer;
715 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
716 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
717 *cookiep = zap_cursor_serialize(&zc);
718 zap_cursor_fini(&zc);
723 * buf must be big enough (eg, 32 bytes)
726 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
727 char *buf, boolean_t addok)
732 if (domain && domain[0]) {
733 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
737 fuid = FUID_ENCODE(domainid, rid);
738 (void) sprintf(buf, "%llx", (longlong_t)fuid);
743 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
744 const char *domain, uint64_t rid, uint64_t *valp)
752 if (!dmu_objset_userspace_present(zfsvfs->z_os))
755 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
759 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
763 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
770 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
771 const char *domain, uint64_t rid, uint64_t quota)
777 boolean_t fuid_dirtied;
779 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
782 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
785 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
786 &zfsvfs->z_groupquota_obj;
788 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
791 fuid_dirtied = zfsvfs->z_fuid_dirty;
793 tx = dmu_tx_create(zfsvfs->z_os);
794 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
796 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
797 zfs_userquota_prop_prefixes[type]);
800 zfs_fuid_txhold(zfsvfs, tx);
801 err = dmu_tx_assign(tx, TXG_WAIT);
807 mutex_enter(&zfsvfs->z_lock);
809 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
811 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
812 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
814 mutex_exit(&zfsvfs->z_lock);
817 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
821 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
825 zfs_fuid_sync(zfsvfs, tx);
831 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
834 uint64_t used, quota, usedobj, quotaobj;
837 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
838 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
840 if (quotaobj == 0 || zfsvfs->z_replay)
843 (void) sprintf(buf, "%llx", (longlong_t)fuid);
844 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
848 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
851 return (used >= quota);
855 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
860 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
862 fuid = isgroup ? zp->z_gid : zp->z_uid;
864 if (quotaobj == 0 || zfsvfs->z_replay)
867 return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
871 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
879 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
882 * We claim to always be readonly so we can open snapshots;
883 * other ZPL code will prevent us from writing to snapshots.
885 error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
887 kmem_free(zfsvfs, sizeof (zfsvfs_t));
892 * Initialize the zfs-specific filesystem structure.
893 * Should probably make this a kmem cache, shuffle fields,
894 * and just bzero up to z_hold_mtx[].
896 zfsvfs->z_vfs = NULL;
897 zfsvfs->z_parent = zfsvfs;
898 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
899 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
902 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
905 } else if (zfsvfs->z_version >
906 zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
907 (void) printf("Can't mount a version %lld file system "
908 "on a version %lld pool\n. Pool must be upgraded to mount "
909 "this file system.", (u_longlong_t)zfsvfs->z_version,
910 (u_longlong_t)spa_version(dmu_objset_spa(os)));
914 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
916 zfsvfs->z_norm = (int)zval;
918 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
920 zfsvfs->z_utf8 = (zval != 0);
922 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
924 zfsvfs->z_case = (uint_t)zval;
927 * Fold case on file systems that are always or sometimes case
930 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
931 zfsvfs->z_case == ZFS_CASE_MIXED)
932 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
934 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
935 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
937 if (zfsvfs->z_use_sa) {
938 /* should either have both of these objects or none */
939 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
945 * Pre SA versions file systems should never touch
946 * either the attribute registration or layout objects.
951 error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
952 &zfsvfs->z_attr_table);
956 if (zfsvfs->z_version >= ZPL_VERSION_SA)
957 sa_register_update_callback(os, zfs_sa_upgrade);
959 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
963 ASSERT(zfsvfs->z_root != 0);
965 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
966 &zfsvfs->z_unlinkedobj);
970 error = zap_lookup(os, MASTER_NODE_OBJ,
971 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
972 8, 1, &zfsvfs->z_userquota_obj);
973 if (error && error != ENOENT)
976 error = zap_lookup(os, MASTER_NODE_OBJ,
977 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
978 8, 1, &zfsvfs->z_groupquota_obj);
979 if (error && error != ENOENT)
982 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
983 &zfsvfs->z_fuid_obj);
984 if (error && error != ENOENT)
987 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
988 &zfsvfs->z_shares_dir);
989 if (error && error != ENOENT)
992 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
993 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
994 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
995 offsetof(znode_t, z_link_node));
996 rrw_init(&zfsvfs->z_teardown_lock);
997 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
998 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
999 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1000 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1006 dmu_objset_disown(os, zfsvfs);
1008 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1013 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1017 error = zfs_register_callbacks(zfsvfs->z_vfs);
1022 * Set the objset user_ptr to track its zfsvfs.
1024 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1025 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1026 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1028 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1031 * If we are not mounting (ie: online recv), then we don't
1032 * have to worry about replaying the log as we blocked all
1033 * operations out since we closed the ZIL.
1039 * During replay we remove the read only flag to
1040 * allow replays to succeed.
1042 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1044 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1046 zfs_unlinked_drain(zfsvfs);
1049 * Parse and replay the intent log.
1051 * Because of ziltest, this must be done after
1052 * zfs_unlinked_drain(). (Further note: ziltest
1053 * doesn't use readonly mounts, where
1054 * zfs_unlinked_drain() isn't called.) This is because
1055 * ziltest causes spa_sync() to think it's committed,
1056 * but actually it is not, so the intent log contains
1057 * many txg's worth of changes.
1059 * In particular, if object N is in the unlinked set in
1060 * the last txg to actually sync, then it could be
1061 * actually freed in a later txg and then reallocated
1062 * in a yet later txg. This would write a "create
1063 * object N" record to the intent log. Normally, this
1064 * would be fine because the spa_sync() would have
1065 * written out the fact that object N is free, before
1066 * we could write the "create object N" intent log
1069 * But when we are in ziltest mode, we advance the "open
1070 * txg" without actually spa_sync()-ing the changes to
1071 * disk. So we would see that object N is still
1072 * allocated and in the unlinked set, and there is an
1073 * intent log record saying to allocate it.
1075 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1076 if (zil_replay_disable) {
1077 zil_destroy(zfsvfs->z_log, B_FALSE);
1079 zfsvfs->z_replay = B_TRUE;
1080 zil_replay(zfsvfs->z_os, zfsvfs,
1082 zfsvfs->z_replay = B_FALSE;
1085 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1092 zfsvfs_free(zfsvfs_t *zfsvfs)
1095 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1098 * This is a barrier to prevent the filesystem from going away in
1099 * zfs_znode_move() until we can safely ensure that the filesystem is
1100 * not unmounted. We consider the filesystem valid before the barrier
1101 * and invalid after the barrier.
1103 rw_enter(&zfsvfs_lock, RW_READER);
1104 rw_exit(&zfsvfs_lock);
1106 zfs_fuid_destroy(zfsvfs);
1108 mutex_destroy(&zfsvfs->z_znodes_lock);
1109 mutex_destroy(&zfsvfs->z_lock);
1110 list_destroy(&zfsvfs->z_all_znodes);
1111 rrw_destroy(&zfsvfs->z_teardown_lock);
1112 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1113 rw_destroy(&zfsvfs->z_fuid_lock);
1114 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1115 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1116 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1120 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1122 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1123 if (zfsvfs->z_vfs) {
1124 if (zfsvfs->z_use_fuids) {
1125 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1126 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1127 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1128 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1129 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1130 vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1132 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1133 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1134 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1135 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1136 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1137 vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1140 zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1144 zfs_domount(vfs_t *vfsp, char *osname)
1147 uint64_t recordsize, fsid_guid;
1154 error = zfsvfs_create(osname, &zfsvfs);
1157 zfsvfs->z_vfs = vfsp;
1159 /* Initialize the generic filesystem structure. */
1160 vfsp->vfs_bcount = 0;
1161 vfsp->vfs_data = NULL;
1163 if (zfs_create_unique_device(&mount_dev) == -1) {
1167 ASSERT(vfs_devismounted(mount_dev) == 0);
1169 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1173 vfsp->vfs_dev = mount_dev;
1174 vfsp->vfs_fstype = zfsfstype;
1175 vfsp->vfs_bsize = recordsize;
1176 vfsp->vfs_flag |= VFS_NOTRUNC;
1177 vfsp->vfs_data = zfsvfs;
1180 * The fsid is 64 bits, composed of an 8-bit fs type, which
1181 * separates our fsid from any other filesystem types, and a
1182 * 56-bit objset unique ID. The objset unique ID is unique to
1183 * all objsets open on this system, provided by unique_create().
1184 * The 8-bit fs type must be put in the low bits of fsid[1]
1185 * because that's where other Solaris filesystems put it.
1187 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1188 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1189 vfsp->vfs_fsid.val[0] = fsid_guid;
1190 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1194 * Set features for file system.
1196 zfs_set_fuid_feature(zfsvfs);
1197 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1198 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1199 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1200 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1201 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1202 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1203 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1205 vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1207 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1210 atime_changed_cb(zfsvfs, B_FALSE);
1211 readonly_changed_cb(zfsvfs, B_TRUE);
1212 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1214 xattr_changed_cb(zfsvfs, pval);
1215 zfsvfs->z_issnap = B_TRUE;
1216 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1218 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1219 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1220 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1222 error = zfsvfs_setup(zfsvfs, B_TRUE);
1225 if (!zfsvfs->z_issnap)
1226 zfsctl_create(zfsvfs);
1229 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1230 zfsvfs_free(zfsvfs);
1232 atomic_add_32(&zfs_active_fs_count, 1);
1239 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1241 objset_t *os = zfsvfs->z_os;
1242 struct dsl_dataset *ds;
1245 * Unregister properties.
1247 if (!dmu_objset_is_snapshot(os)) {
1248 ds = dmu_objset_ds(os);
1249 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1252 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1255 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1258 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1261 VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1264 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1267 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1270 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1273 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
1276 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1277 acl_inherit_changed_cb, zfsvfs) == 0);
1279 VERIFY(dsl_prop_unregister(ds, "vscan",
1280 vscan_changed_cb, zfsvfs) == 0);
1285 * Convert a decimal digit string to a uint64_t integer.
1288 str_to_uint64(char *str, uint64_t *objnum)
1293 if (*str < '0' || *str > '9')
1296 num = num*10 + *str++ - '0';
1304 * The boot path passed from the boot loader is in the form of
1305 * "rootpool-name/root-filesystem-object-number'. Convert this
1306 * string to a dataset name: "rootpool-name/root-filesystem-name".
1309 zfs_parse_bootfs(char *bpath, char *outpath)
1315 if (*bpath == 0 || *bpath == '/')
1318 (void) strcpy(outpath, bpath);
1320 slashp = strchr(bpath, '/');
1322 /* if no '/', just return the pool name */
1323 if (slashp == NULL) {
1327 /* if not a number, just return the root dataset name */
1328 if (str_to_uint64(slashp+1, &objnum)) {
1333 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1340 * zfs_check_global_label:
1341 * Check that the hex label string is appropriate for the dataset
1342 * being mounted into the global_zone proper.
1344 * Return an error if the hex label string is not default or
1345 * admin_low/admin_high. For admin_low labels, the corresponding
1346 * dataset must be readonly.
1349 zfs_check_global_label(const char *dsname, const char *hexsl)
1351 if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1353 if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1355 if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1356 /* must be readonly */
1359 if (dsl_prop_get_integer(dsname,
1360 zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1362 return (rdonly ? 0 : EACCES);
1368 * zfs_mount_label_policy:
1369 * Determine whether the mount is allowed according to MAC check.
1370 * by comparing (where appropriate) label of the dataset against
1371 * the label of the zone being mounted into. If the dataset has
1372 * no label, create one.
1375 * 0 : access allowed
1376 * >0 : error code, such as EACCES
1379 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1382 zone_t *mntzone = NULL;
1383 ts_label_t *mnt_tsl;
1386 char ds_hexsl[MAXNAMELEN];
1388 retv = EACCES; /* assume the worst */
1391 * Start by getting the dataset label if it exists.
1393 error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1394 1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1399 * If labeling is NOT enabled, then disallow the mount of datasets
1400 * which have a non-default label already. No other label checks
1403 if (!is_system_labeled()) {
1404 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1410 * Get the label of the mountpoint. If mounting into the global
1411 * zone (i.e. mountpoint is not within an active zone and the
1412 * zoned property is off), the label must be default or
1413 * admin_low/admin_high only; no other checks are needed.
1415 mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1416 if (mntzone->zone_id == GLOBAL_ZONEID) {
1421 if (dsl_prop_get_integer(osname,
1422 zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1425 return (zfs_check_global_label(osname, ds_hexsl));
1428 * This is the case of a zone dataset being mounted
1429 * initially, before the zone has been fully created;
1430 * allow this mount into global zone.
1435 mnt_tsl = mntzone->zone_slabel;
1436 ASSERT(mnt_tsl != NULL);
1437 label_hold(mnt_tsl);
1438 mnt_sl = label2bslabel(mnt_tsl);
1440 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1442 * The dataset doesn't have a real label, so fabricate one.
1446 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1447 dsl_prop_set(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1448 ZPROP_SRC_LOCAL, 1, strlen(str) + 1, str) == 0)
1451 kmem_free(str, strlen(str) + 1);
1452 } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1454 * Now compare labels to complete the MAC check. If the
1455 * labels are equal then allow access. If the mountpoint
1456 * label dominates the dataset label, allow readonly access.
1457 * Otherwise, access is denied.
1459 if (blequal(mnt_sl, &ds_sl))
1461 else if (bldominates(mnt_sl, &ds_sl)) {
1462 vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1467 label_rele(mnt_tsl);
1473 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1476 static int zfsrootdone = 0;
1477 zfsvfs_t *zfsvfs = NULL;
1486 * The filesystem that we mount as root is defined in the
1487 * boot property "zfs-bootfs" with a format of
1488 * "poolname/root-dataset-objnum".
1490 if (why == ROOT_INIT) {
1494 * the process of doing a spa_load will require the
1495 * clock to be set before we could (for example) do
1496 * something better by looking at the timestamp on
1497 * an uberblock, so just set it to -1.
1501 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1502 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1506 zfs_devid = spa_get_bootprop("diskdevid");
1507 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1509 spa_free_bootprop(zfs_devid);
1511 spa_free_bootprop(zfs_bootfs);
1512 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1516 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1517 spa_free_bootprop(zfs_bootfs);
1518 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1523 spa_free_bootprop(zfs_bootfs);
1525 if (error = vfs_lock(vfsp))
1528 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1529 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1533 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1535 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1536 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1541 mutex_enter(&vp->v_lock);
1542 vp->v_flag |= VROOT;
1543 mutex_exit(&vp->v_lock);
1547 * Leave rootvp held. The root file system is never unmounted.
1550 vfs_add((struct vnode *)0, vfsp,
1551 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1555 } else if (why == ROOT_REMOUNT) {
1556 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1557 vfsp->vfs_flag |= VFS_REMOUNT;
1559 /* refresh mount options */
1560 zfs_unregister_callbacks(vfsp->vfs_data);
1561 return (zfs_register_callbacks(vfsp));
1563 } else if (why == ROOT_UNMOUNT) {
1564 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1565 (void) zfs_sync(vfsp, 0, 0);
1570 * if "why" is equal to anything else other than ROOT_INIT,
1571 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1578 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1583 uio_seg_t fromspace = (uap->flags & MS_SYSSPACE) ?
1584 UIO_SYSSPACE : UIO_USERSPACE;
1587 if (mvp->v_type != VDIR)
1590 mutex_enter(&mvp->v_lock);
1591 if ((uap->flags & MS_REMOUNT) == 0 &&
1592 (uap->flags & MS_OVERLAY) == 0 &&
1593 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1594 mutex_exit(&mvp->v_lock);
1597 mutex_exit(&mvp->v_lock);
1600 * ZFS does not support passing unparsed data in via MS_DATA.
1601 * Users should use the MS_OPTIONSTR interface; this means
1602 * that all option parsing is already done and the options struct
1603 * can be interrogated.
1605 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1609 * Get the objset name (the "special" mount argument).
1611 if (error = pn_get(uap->spec, fromspace, &spn))
1614 osname = spn.pn_path;
1617 * Check for mount privilege?
1619 * If we don't have privilege then see if
1620 * we have local permission to allow it
1622 error = secpolicy_fs_mount(cr, mvp, vfsp);
1624 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1628 * Make sure user is the owner of the mount point
1629 * or has sufficient privileges.
1632 vattr.va_mask = AT_UID;
1634 if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1638 if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1639 VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1642 secpolicy_fs_mount_clearopts(cr, vfsp);
1649 * Refuse to mount a filesystem if we are in a local zone and the
1650 * dataset is not visible.
1652 if (!INGLOBALZONE(curproc) &&
1653 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1658 error = zfs_mount_label_policy(vfsp, osname);
1663 * When doing a remount, we simply refresh our temporary properties
1664 * according to those options set in the current VFS options.
1666 if (uap->flags & MS_REMOUNT) {
1667 /* refresh mount options */
1668 zfs_unregister_callbacks(vfsp->vfs_data);
1669 error = zfs_register_callbacks(vfsp);
1673 error = zfs_domount(vfsp, osname);
1676 * Add an extra VFS_HOLD on our parent vfs so that it can't
1677 * disappear due to a forced unmount.
1679 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1680 VFS_HOLD(mvp->v_vfsp);
1688 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1690 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1692 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1696 dmu_objset_space(zfsvfs->z_os,
1697 &refdbytes, &availbytes, &usedobjs, &availobjs);
1700 * The underlying storage pool actually uses multiple block sizes.
1701 * We report the fragsize as the smallest block size we support,
1702 * and we report our blocksize as the filesystem's maximum blocksize.
1704 statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1705 statp->f_bsize = zfsvfs->z_max_blksz;
1708 * The following report "total" blocks of various kinds in the
1709 * file system, but reported in terms of f_frsize - the
1713 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1714 statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1715 statp->f_bavail = statp->f_bfree; /* no root reservation */
1718 * statvfs() should really be called statufs(), because it assumes
1719 * static metadata. ZFS doesn't preallocate files, so the best
1720 * we can do is report the max that could possibly fit in f_files,
1721 * and that minus the number actually used in f_ffree.
1722 * For f_ffree, report the smaller of the number of object available
1723 * and the number of blocks (each object will take at least a block).
1725 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1726 statp->f_favail = statp->f_ffree; /* no "root reservation" */
1727 statp->f_files = statp->f_ffree + usedobjs;
1729 (void) cmpldev(&d32, vfsp->vfs_dev);
1730 statp->f_fsid = d32;
1733 * We're a zfs filesystem.
1735 (void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1737 statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1739 statp->f_namemax = ZFS_MAXNAMELEN;
1742 * We have all of 32 characters to stuff a string here.
1743 * Is there anything useful we could/should provide?
1745 bzero(statp->f_fstr, sizeof (statp->f_fstr));
1752 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1754 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1760 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1762 *vpp = ZTOV(rootzp);
1769 * Teardown the zfsvfs::z_os.
1771 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1772 * and 'z_teardown_inactive_lock' held.
1775 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1779 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1783 * We purge the parent filesystem's vfsp as the parent
1784 * filesystem and all of its snapshots have their vnode's
1785 * v_vfsp set to the parent's filesystem's vfsp. Note,
1786 * 'z_parent' is self referential for non-snapshots.
1788 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1792 * Close the zil. NB: Can't close the zil while zfs_inactive
1793 * threads are blocked as zil_close can call zfs_inactive.
1795 if (zfsvfs->z_log) {
1796 zil_close(zfsvfs->z_log);
1797 zfsvfs->z_log = NULL;
1800 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1803 * If we are not unmounting (ie: online recv) and someone already
1804 * unmounted this file system while we were doing the switcheroo,
1805 * or a reopen of z_os failed then just bail out now.
1807 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1808 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1809 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1814 * At this point there are no vops active, and any new vops will
1815 * fail with EIO since we have z_teardown_lock for writer (only
1816 * relavent for forced unmount).
1818 * Release all holds on dbufs.
1820 mutex_enter(&zfsvfs->z_znodes_lock);
1821 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1822 zp = list_next(&zfsvfs->z_all_znodes, zp))
1824 ASSERT(ZTOV(zp)->v_count > 0);
1825 zfs_znode_dmu_fini(zp);
1827 mutex_exit(&zfsvfs->z_znodes_lock);
1830 * If we are unmounting, set the unmounted flag and let new vops
1831 * unblock. zfs_inactive will have the unmounted behavior, and all
1832 * other vops will fail with EIO.
1835 zfsvfs->z_unmounted = B_TRUE;
1836 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1837 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1841 * z_os will be NULL if there was an error in attempting to reopen
1842 * zfsvfs, so just return as the properties had already been
1843 * unregistered and cached data had been evicted before.
1845 if (zfsvfs->z_os == NULL)
1849 * Unregister properties.
1851 zfs_unregister_callbacks(zfsvfs);
1856 if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1857 !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1858 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1859 (void) dmu_objset_evict_dbufs(zfsvfs->z_os);
1866 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1868 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1872 ret = secpolicy_fs_unmount(cr, vfsp);
1874 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1875 ZFS_DELEG_PERM_MOUNT, cr))
1880 * We purge the parent filesystem's vfsp as the parent filesystem
1881 * and all of its snapshots have their vnode's v_vfsp set to the
1882 * parent's filesystem's vfsp. Note, 'z_parent' is self
1883 * referential for non-snapshots.
1885 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1888 * Unmount any snapshots mounted under .zfs before unmounting the
1891 if (zfsvfs->z_ctldir != NULL &&
1892 (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1896 if (!(fflag & MS_FORCE)) {
1898 * Check the number of active vnodes in the file system.
1899 * Our count is maintained in the vfs structure, but the
1900 * number is off by 1 to indicate a hold on the vfs
1903 * The '.zfs' directory maintains a reference of its
1904 * own, and any active references underneath are
1905 * reflected in the vnode count.
1907 if (zfsvfs->z_ctldir == NULL) {
1908 if (vfsp->vfs_count > 1)
1911 if (vfsp->vfs_count > 2 ||
1912 zfsvfs->z_ctldir->v_count > 1)
1917 vfsp->vfs_flag |= VFS_UNMOUNTED;
1919 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1923 * z_os will be NULL if there was an error in
1924 * attempting to reopen zfsvfs.
1928 * Unset the objset user_ptr.
1930 mutex_enter(&os->os_user_ptr_lock);
1931 dmu_objset_set_user(os, NULL);
1932 mutex_exit(&os->os_user_ptr_lock);
1935 * Finally release the objset
1937 dmu_objset_disown(os, zfsvfs);
1941 * We can now safely destroy the '.zfs' directory node.
1943 if (zfsvfs->z_ctldir != NULL)
1944 zfsctl_destroy(zfsvfs);
1950 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1952 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1954 uint64_t object = 0;
1955 uint64_t fid_gen = 0;
1964 if (fidp->fid_len == LONG_FID_LEN) {
1965 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1966 uint64_t objsetid = 0;
1967 uint64_t setgen = 0;
1969 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1970 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1972 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1973 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1977 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1983 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1984 zfid_short_t *zfid = (zfid_short_t *)fidp;
1986 for (i = 0; i < sizeof (zfid->zf_object); i++)
1987 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1989 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1990 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1996 /* A zero fid_gen means we are in the .zfs control directories */
1998 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1999 *vpp = zfsvfs->z_ctldir;
2000 ASSERT(*vpp != NULL);
2001 if (object == ZFSCTL_INO_SNAPDIR) {
2002 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
2003 0, NULL, NULL, NULL, NULL, NULL) == 0);
2011 gen_mask = -1ULL >> (64 - 8 * i);
2013 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
2014 if (err = zfs_zget(zfsvfs, object, &zp)) {
2018 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
2020 zp_gen = zp_gen & gen_mask;
2023 if (zp->z_unlinked || zp_gen != fid_gen) {
2024 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
2036 * Block out VOPs and close zfsvfs_t::z_os
2038 * Note, if successful, then we return with the 'z_teardown_lock' and
2039 * 'z_teardown_inactive_lock' write held.
2042 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2046 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2048 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
2054 * Reopen zfsvfs_t::z_os and release VOPs.
2057 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname)
2061 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
2062 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2064 err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zfsvfs,
2067 zfsvfs->z_os = NULL;
2070 uint64_t sa_obj = 0;
2073 * Make sure version hasn't changed
2076 err = zfs_get_zplprop(zfsvfs->z_os, ZFS_PROP_VERSION,
2077 &zfsvfs->z_version);
2082 err = zap_lookup(zfsvfs->z_os, MASTER_NODE_OBJ,
2083 ZFS_SA_ATTRS, 8, 1, &sa_obj);
2085 if (err && zfsvfs->z_version >= ZPL_VERSION_SA)
2088 if ((err = sa_setup(zfsvfs->z_os, sa_obj,
2089 zfs_attr_table, ZPL_END, &zfsvfs->z_attr_table)) != 0)
2092 if (zfsvfs->z_version >= ZPL_VERSION_SA)
2093 sa_register_update_callback(zfsvfs->z_os,
2096 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2098 zfs_set_fuid_feature(zfsvfs);
2101 * Attempt to re-establish all the active znodes with
2102 * their dbufs. If a zfs_rezget() fails, then we'll let
2103 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2104 * when they try to use their znode.
2106 mutex_enter(&zfsvfs->z_znodes_lock);
2107 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2108 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2109 (void) zfs_rezget(zp);
2111 mutex_exit(&zfsvfs->z_znodes_lock);
2115 /* release the VOPs */
2116 rw_exit(&zfsvfs->z_teardown_inactive_lock);
2117 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
2121 * Since we couldn't reopen zfsvfs::z_os, or
2122 * setup the sa framework force unmount this file system.
2124 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2125 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2131 zfs_freevfs(vfs_t *vfsp)
2133 zfsvfs_t *zfsvfs = vfsp->vfs_data;
2136 * If this is a snapshot, we have an extra VFS_HOLD on our parent
2137 * from zfs_mount(). Release it here. If we came through
2138 * zfs_mountroot() instead, we didn't grab an extra hold, so
2139 * skip the VFS_RELE for rootvfs.
2141 if (zfsvfs->z_issnap && (vfsp != rootvfs))
2142 VFS_RELE(zfsvfs->z_parent->z_vfs);
2144 zfsvfs_free(zfsvfs);
2146 atomic_add_32(&zfs_active_fs_count, -1);
2150 * VFS_INIT() initialization. Note that there is no VFS_FINI(),
2151 * so we can't safely do any non-idempotent initialization here.
2152 * Leave that to zfs_init() and zfs_fini(), which are called
2153 * from the module's _init() and _fini() entry points.
2157 zfs_vfsinit(int fstype, char *name)
2164 * Setup vfsops and vnodeops tables.
2166 error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2168 cmn_err(CE_WARN, "zfs: bad vfs ops template");
2171 error = zfs_create_op_tables();
2173 zfs_remove_op_tables();
2174 cmn_err(CE_WARN, "zfs: bad vnode ops template");
2175 (void) vfs_freevfsops_by_type(zfsfstype);
2179 mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2182 * Unique major number for all zfs mounts.
2183 * If we run out of 32-bit minors, we'll getudev() another major.
2185 zfs_major = ddi_name_to_major(ZFS_DRIVER);
2186 zfs_minor = ZFS_MIN_MINOR;
2195 * Initialize .zfs directory structures
2200 * Initialize znode cache, vnode ops, etc...
2204 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2217 return (zfs_active_fs_count != 0);
2221 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2224 objset_t *os = zfsvfs->z_os;
2227 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2230 if (newvers < zfsvfs->z_version)
2233 if (zfs_spa_version_map(newvers) >
2234 spa_version(dmu_objset_spa(zfsvfs->z_os)))
2237 tx = dmu_tx_create(os);
2238 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2239 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2240 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2242 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2244 error = dmu_tx_assign(tx, TXG_WAIT);
2250 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2251 8, 1, &newvers, tx);
2258 if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2261 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2263 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2264 DMU_OT_NONE, 0, tx);
2266 error = zap_add(os, MASTER_NODE_OBJ,
2267 ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2270 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2271 sa_register_update_callback(os, zfs_sa_upgrade);
2274 spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2275 "from %llu to %llu", zfsvfs->z_version, newvers);
2279 zfsvfs->z_version = newvers;
2281 zfs_set_fuid_feature(zfsvfs);
2287 * Read a property stored within the master node.
2290 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2296 * Look up the file system's value for the property. For the
2297 * version property, we look up a slightly different string.
2299 if (prop == ZFS_PROP_VERSION)
2300 pname = ZPL_VERSION_STR;
2302 pname = zfs_prop_to_name(prop);
2305 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2307 if (error == ENOENT) {
2308 /* No value set, use the default value */
2310 case ZFS_PROP_VERSION:
2311 *value = ZPL_VERSION;
2313 case ZFS_PROP_NORMALIZE:
2314 case ZFS_PROP_UTF8ONLY:
2318 *value = ZFS_CASE_SENSITIVE;
2328 static vfsdef_t vfw = {
2332 VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2337 struct modlfs zfs_modlfs = {
2338 &mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw