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 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
26 #include <sys/types.h>
27 #include <sys/param.h>
28 #include <sys/systm.h>
29 #include <sys/sysmacros.h>
31 #include <sys/pathname.h>
32 #include <sys/vnode.h>
34 #include <sys/vfs_opreg.h>
35 #include <sys/mntent.h>
36 #include <sys/mount.h>
37 #include <sys/cmn_err.h>
38 #include "fs/fs_subr.h"
39 #include <sys/zfs_znode.h>
40 #include <sys/zfs_dir.h>
42 #include <sys/fs/zfs.h>
44 #include <sys/dsl_prop.h>
45 #include <sys/dsl_dataset.h>
46 #include <sys/dsl_deleg.h>
49 #include <sys/varargs.h>
50 #include <sys/policy.h>
51 #include <sys/atomic.h>
52 #include <sys/mkdev.h>
53 #include <sys/modctl.h>
54 #include <sys/refstr.h>
55 #include <sys/zfs_ioctl.h>
56 #include <sys/zfs_ctldir.h>
57 #include <sys/zfs_fuid.h>
58 #include <sys/bootconf.h>
59 #include <sys/sunddi.h>
61 #include <sys/dmu_objset.h>
62 #include <sys/spa_boot.h>
65 vfsops_t *zfs_vfsops = NULL;
66 static major_t zfs_major;
67 static minor_t zfs_minor;
68 static kmutex_t zfs_dev_mtx;
70 extern int sys_shutdown;
72 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
73 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
74 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
75 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
76 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
77 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
78 static void zfs_freevfs(vfs_t *vfsp);
80 static const fs_operation_def_t zfs_vfsops_template[] = {
81 VFSNAME_MOUNT, { .vfs_mount = zfs_mount },
82 VFSNAME_MOUNTROOT, { .vfs_mountroot = zfs_mountroot },
83 VFSNAME_UNMOUNT, { .vfs_unmount = zfs_umount },
84 VFSNAME_ROOT, { .vfs_root = zfs_root },
85 VFSNAME_STATVFS, { .vfs_statvfs = zfs_statvfs },
86 VFSNAME_SYNC, { .vfs_sync = zfs_sync },
87 VFSNAME_VGET, { .vfs_vget = zfs_vget },
88 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
92 static const fs_operation_def_t zfs_vfsops_eio_template[] = {
93 VFSNAME_FREEVFS, { .vfs_freevfs = zfs_freevfs },
98 * We need to keep a count of active fs's.
99 * This is necessary to prevent our module
100 * from being unloaded after a umount -f
102 static uint32_t zfs_active_fs_count = 0;
104 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
105 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
106 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
107 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
110 * MO_DEFAULT is not used since the default value is determined
111 * by the equivalent property.
113 static mntopt_t mntopts[] = {
114 { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
115 { MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
116 { MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
117 { MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
120 static mntopts_t zfs_mntopts = {
121 sizeof (mntopts) / sizeof (mntopt_t),
127 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
130 * Data integrity is job one. We don't want a compromised kernel
131 * writing to the storage pool, so we never sync during panic.
137 * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
138 * to sync metadata, which they would otherwise cache indefinitely.
139 * Semantically, the only requirement is that the sync be initiated.
140 * The DMU syncs out txgs frequently, so there's nothing to do.
142 if (flag & SYNC_ATTR)
147 * Sync a specific filesystem.
149 zfsvfs_t *zfsvfs = vfsp->vfs_data;
153 dp = dmu_objset_pool(zfsvfs->z_os);
156 * If the system is shutting down, then skip any
157 * filesystems which may exist on a suspended pool.
159 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
164 if (zfsvfs->z_log != NULL)
165 zil_commit(zfsvfs->z_log, UINT64_MAX, 0);
167 txg_wait_synced(dp, 0);
171 * Sync all ZFS filesystems. This is what happens when you
172 * run sync(1M). Unlike other filesystems, ZFS honors the
173 * request by waiting for all pools to commit all dirty data.
182 zfs_create_unique_device(dev_t *dev)
187 ASSERT3U(zfs_minor, <=, MAXMIN32);
188 minor_t start = zfs_minor;
190 mutex_enter(&zfs_dev_mtx);
191 if (zfs_minor >= MAXMIN32) {
193 * If we're still using the real major
194 * keep out of /dev/zfs and /dev/zvol minor
195 * number space. If we're using a getudev()'ed
196 * major number, we can use all of its minors.
198 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
199 zfs_minor = ZFS_MIN_MINOR;
205 *dev = makedevice(zfs_major, zfs_minor);
206 mutex_exit(&zfs_dev_mtx);
207 } while (vfs_devismounted(*dev) && zfs_minor != start);
208 if (zfs_minor == start) {
210 * We are using all ~262,000 minor numbers for the
211 * current major number. Create a new major number.
213 if ((new_major = getudev()) == (major_t)-1) {
215 "zfs_mount: Can't get unique major "
219 mutex_enter(&zfs_dev_mtx);
220 zfs_major = new_major;
223 mutex_exit(&zfs_dev_mtx);
227 /* CONSTANTCONDITION */
234 atime_changed_cb(void *arg, uint64_t newval)
236 zfsvfs_t *zfsvfs = arg;
238 if (newval == TRUE) {
239 zfsvfs->z_atime = TRUE;
240 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
241 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
243 zfsvfs->z_atime = FALSE;
244 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
245 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
250 xattr_changed_cb(void *arg, uint64_t newval)
252 zfsvfs_t *zfsvfs = arg;
254 if (newval == TRUE) {
255 /* XXX locking on vfs_flag? */
256 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
257 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
258 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
260 /* XXX locking on vfs_flag? */
261 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
262 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
263 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
268 blksz_changed_cb(void *arg, uint64_t newval)
270 zfsvfs_t *zfsvfs = arg;
272 if (newval < SPA_MINBLOCKSIZE ||
273 newval > SPA_MAXBLOCKSIZE || !ISP2(newval))
274 newval = SPA_MAXBLOCKSIZE;
276 zfsvfs->z_max_blksz = newval;
277 zfsvfs->z_vfs->vfs_bsize = newval;
281 readonly_changed_cb(void *arg, uint64_t newval)
283 zfsvfs_t *zfsvfs = arg;
286 /* XXX locking on vfs_flag? */
287 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
288 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
289 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
291 /* XXX locking on vfs_flag? */
292 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
293 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
294 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
299 devices_changed_cb(void *arg, uint64_t newval)
301 zfsvfs_t *zfsvfs = arg;
303 if (newval == FALSE) {
304 zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
305 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
306 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
308 zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
309 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
310 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
315 setuid_changed_cb(void *arg, uint64_t newval)
317 zfsvfs_t *zfsvfs = arg;
319 if (newval == FALSE) {
320 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
321 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
322 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
324 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
325 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
326 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
331 exec_changed_cb(void *arg, uint64_t newval)
333 zfsvfs_t *zfsvfs = arg;
335 if (newval == FALSE) {
336 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
337 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
338 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
340 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
341 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
342 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
347 * The nbmand mount option can be changed at mount time.
348 * We can't allow it to be toggled on live file systems or incorrect
349 * behavior may be seen from cifs clients
351 * This property isn't registered via dsl_prop_register(), but this callback
352 * will be called when a file system is first mounted
355 nbmand_changed_cb(void *arg, uint64_t newval)
357 zfsvfs_t *zfsvfs = arg;
358 if (newval == FALSE) {
359 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
360 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
362 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
363 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
368 snapdir_changed_cb(void *arg, uint64_t newval)
370 zfsvfs_t *zfsvfs = arg;
372 zfsvfs->z_show_ctldir = newval;
376 vscan_changed_cb(void *arg, uint64_t newval)
378 zfsvfs_t *zfsvfs = arg;
380 zfsvfs->z_vscan = newval;
384 acl_mode_changed_cb(void *arg, uint64_t newval)
386 zfsvfs_t *zfsvfs = arg;
388 zfsvfs->z_acl_mode = newval;
392 acl_inherit_changed_cb(void *arg, uint64_t newval)
394 zfsvfs_t *zfsvfs = arg;
396 zfsvfs->z_acl_inherit = newval;
400 zfs_register_callbacks(vfs_t *vfsp)
402 struct dsl_dataset *ds = NULL;
404 zfsvfs_t *zfsvfs = NULL;
406 int readonly, do_readonly = B_FALSE;
407 int setuid, do_setuid = B_FALSE;
408 int exec, do_exec = B_FALSE;
409 int devices, do_devices = B_FALSE;
410 int xattr, do_xattr = B_FALSE;
411 int atime, do_atime = B_FALSE;
415 zfsvfs = vfsp->vfs_data;
420 * The act of registering our callbacks will destroy any mount
421 * options we may have. In order to enable temporary overrides
422 * of mount options, we stash away the current values and
423 * restore them after we register the callbacks.
425 if (vfs_optionisset(vfsp, MNTOPT_RO, NULL)) {
427 do_readonly = B_TRUE;
428 } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
430 do_readonly = B_TRUE;
432 if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
438 if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
441 } else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
446 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
449 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
454 if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
457 } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
461 if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
464 } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
468 if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
471 } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
477 * nbmand is a special property. It can only be changed at
480 * This is weird, but it is documented to only be changeable
483 if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
485 } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
488 char osname[MAXNAMELEN];
490 dmu_objset_name(os, osname);
491 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
498 * Register property callbacks.
500 * It would probably be fine to just check for i/o error from
501 * the first prop_register(), but I guess I like to go
504 ds = dmu_objset_ds(os);
505 error = dsl_prop_register(ds, "atime", atime_changed_cb, zfsvfs);
506 error = error ? error : dsl_prop_register(ds,
507 "xattr", xattr_changed_cb, zfsvfs);
508 error = error ? error : dsl_prop_register(ds,
509 "recordsize", blksz_changed_cb, zfsvfs);
510 error = error ? error : dsl_prop_register(ds,
511 "readonly", readonly_changed_cb, zfsvfs);
512 error = error ? error : dsl_prop_register(ds,
513 "devices", devices_changed_cb, zfsvfs);
514 error = error ? error : dsl_prop_register(ds,
515 "setuid", setuid_changed_cb, zfsvfs);
516 error = error ? error : dsl_prop_register(ds,
517 "exec", exec_changed_cb, zfsvfs);
518 error = error ? error : dsl_prop_register(ds,
519 "snapdir", snapdir_changed_cb, zfsvfs);
520 error = error ? error : dsl_prop_register(ds,
521 "aclmode", acl_mode_changed_cb, zfsvfs);
522 error = error ? error : dsl_prop_register(ds,
523 "aclinherit", acl_inherit_changed_cb, zfsvfs);
524 error = error ? error : dsl_prop_register(ds,
525 "vscan", vscan_changed_cb, zfsvfs);
530 * Invoke our callbacks to restore temporary mount options.
533 readonly_changed_cb(zfsvfs, readonly);
535 setuid_changed_cb(zfsvfs, setuid);
537 exec_changed_cb(zfsvfs, exec);
539 devices_changed_cb(zfsvfs, devices);
541 xattr_changed_cb(zfsvfs, xattr);
543 atime_changed_cb(zfsvfs, atime);
545 nbmand_changed_cb(zfsvfs, nbmand);
551 * We may attempt to unregister some callbacks that are not
552 * registered, but this is OK; it will simply return ENOMSG,
553 * which we will ignore.
555 (void) dsl_prop_unregister(ds, "atime", atime_changed_cb, zfsvfs);
556 (void) dsl_prop_unregister(ds, "xattr", xattr_changed_cb, zfsvfs);
557 (void) dsl_prop_unregister(ds, "recordsize", blksz_changed_cb, zfsvfs);
558 (void) dsl_prop_unregister(ds, "readonly", readonly_changed_cb, zfsvfs);
559 (void) dsl_prop_unregister(ds, "devices", devices_changed_cb, zfsvfs);
560 (void) dsl_prop_unregister(ds, "setuid", setuid_changed_cb, zfsvfs);
561 (void) dsl_prop_unregister(ds, "exec", exec_changed_cb, zfsvfs);
562 (void) dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb, zfsvfs);
563 (void) dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb, zfsvfs);
564 (void) dsl_prop_unregister(ds, "aclinherit", acl_inherit_changed_cb,
566 (void) dsl_prop_unregister(ds, "vscan", vscan_changed_cb, zfsvfs);
572 uidacct(objset_t *os, boolean_t isgroup, uint64_t fuid,
573 int64_t delta, dmu_tx_t *tx)
578 uint64_t obj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
583 (void) snprintf(buf, sizeof (buf), "%llx", (longlong_t)fuid);
584 err = zap_lookup(os, obj, buf, 8, 1, &used);
585 ASSERT(err == 0 || err == ENOENT);
586 /* no underflow/overflow */
587 ASSERT(delta > 0 || used >= -delta);
588 ASSERT(delta < 0 || used + delta > used);
591 err = zap_remove(os, obj, buf, tx);
593 err = zap_update(os, obj, buf, 8, 1, &used, tx);
598 zfs_space_delta_cb(objset_t *os, dmu_object_type_t bonustype,
599 void *oldbonus, void *newbonus,
600 uint64_t oldused, uint64_t newused, dmu_tx_t *tx)
602 znode_phys_t *oldznp = oldbonus;
603 znode_phys_t *newznp = newbonus;
605 if (bonustype != DMU_OT_ZNODE)
608 /* We charge 512 for the dnode (if it's allocated). */
609 if (oldznp->zp_gen != 0)
610 oldused += DNODE_SIZE;
611 if (newznp->zp_gen != 0)
612 newused += DNODE_SIZE;
614 if (oldznp->zp_uid == newznp->zp_uid) {
615 uidacct(os, B_FALSE, oldznp->zp_uid, newused-oldused, tx);
617 uidacct(os, B_FALSE, oldznp->zp_uid, -oldused, tx);
618 uidacct(os, B_FALSE, newznp->zp_uid, newused, tx);
621 if (oldznp->zp_gid == newznp->zp_gid) {
622 uidacct(os, B_TRUE, oldznp->zp_gid, newused-oldused, tx);
624 uidacct(os, B_TRUE, oldznp->zp_gid, -oldused, tx);
625 uidacct(os, B_TRUE, newznp->zp_gid, newused, tx);
630 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
631 char *domainbuf, int buflen, uid_t *ridp)
633 extern uint64_t strtonum(const char *str, char **nptr);
637 fuid = strtonum(fuidstr, NULL);
639 domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
641 (void) strlcpy(domainbuf, domain, buflen);
644 *ridp = FUID_RID(fuid);
648 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
651 case ZFS_PROP_USERUSED:
652 return (DMU_USERUSED_OBJECT);
653 case ZFS_PROP_GROUPUSED:
654 return (DMU_GROUPUSED_OBJECT);
655 case ZFS_PROP_USERQUOTA:
656 return (zfsvfs->z_userquota_obj);
657 case ZFS_PROP_GROUPQUOTA:
658 return (zfsvfs->z_groupquota_obj);
664 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
665 uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
670 zfs_useracct_t *buf = vbuf;
673 if (!dmu_objset_userspace_present(zfsvfs->z_os))
676 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
682 for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
683 (error = zap_cursor_retrieve(&zc, &za)) == 0;
684 zap_cursor_advance(&zc)) {
685 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
689 fuidstr_to_sid(zfsvfs, za.za_name,
690 buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
692 buf->zu_space = za.za_first_integer;
698 ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
699 *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
700 *cookiep = zap_cursor_serialize(&zc);
701 zap_cursor_fini(&zc);
706 * buf must be big enough (eg, 32 bytes)
709 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
710 char *buf, boolean_t addok)
715 if (domain && domain[0]) {
716 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
720 fuid = FUID_ENCODE(domainid, rid);
721 (void) sprintf(buf, "%llx", (longlong_t)fuid);
726 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
727 const char *domain, uint64_t rid, uint64_t *valp)
735 if (!dmu_objset_userspace_present(zfsvfs->z_os))
738 obj = zfs_userquota_prop_to_obj(zfsvfs, type);
742 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
746 err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
753 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
754 const char *domain, uint64_t rid, uint64_t quota)
760 boolean_t fuid_dirtied;
762 if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
765 if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
768 objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
769 &zfsvfs->z_groupquota_obj;
771 err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
774 fuid_dirtied = zfsvfs->z_fuid_dirty;
776 tx = dmu_tx_create(zfsvfs->z_os);
777 dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
779 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
780 zfs_userquota_prop_prefixes[type]);
783 zfs_fuid_txhold(zfsvfs, tx);
784 err = dmu_tx_assign(tx, TXG_WAIT);
790 mutex_enter(&zfsvfs->z_lock);
792 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
794 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
795 zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
797 mutex_exit(&zfsvfs->z_lock);
800 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
804 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, "a, tx);
808 zfs_fuid_sync(zfsvfs, tx);
814 zfs_usergroup_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
817 uint64_t used, quota, usedobj, quotaobj;
820 usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
821 quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
823 if (quotaobj == 0 || zfsvfs->z_replay)
826 (void) sprintf(buf, "%llx", (longlong_t)fuid);
827 err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, "a);
831 err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
834 return (used >= quota);
838 zfsvfs_create(const char *osname, int mode, zfsvfs_t **zvp)
845 if (error = dsl_prop_get_integer(osname, "readonly", &zval, NULL))
848 mode |= DS_MODE_READONLY;
850 error = dmu_objset_open(osname, DMU_OST_ZFS, mode, &os);
851 if (error == EROFS) {
852 mode |= DS_MODE_READONLY;
853 error = dmu_objset_open(osname, DMU_OST_ZFS, mode, &os);
859 * Initialize the zfs-specific filesystem structure.
860 * Should probably make this a kmem cache, shuffle fields,
861 * and just bzero up to z_hold_mtx[].
863 zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
864 zfsvfs->z_vfs = NULL;
865 zfsvfs->z_parent = zfsvfs;
866 zfsvfs->z_max_blksz = SPA_MAXBLOCKSIZE;
867 zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
870 error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
873 } else if (zfsvfs->z_version > ZPL_VERSION) {
874 (void) printf("Mismatched versions: File system "
875 "is version %llu on-disk format, which is "
876 "incompatible with this software version %lld!",
877 (u_longlong_t)zfsvfs->z_version, ZPL_VERSION);
882 if ((error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &zval)) != 0)
884 zfsvfs->z_norm = (int)zval;
886 if ((error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &zval)) != 0)
888 zfsvfs->z_utf8 = (zval != 0);
890 if ((error = zfs_get_zplprop(os, ZFS_PROP_CASE, &zval)) != 0)
892 zfsvfs->z_case = (uint_t)zval;
895 * Fold case on file systems that are always or sometimes case
898 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
899 zfsvfs->z_case == ZFS_CASE_MIXED)
900 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
902 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
904 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
908 ASSERT(zfsvfs->z_root != 0);
910 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
911 &zfsvfs->z_unlinkedobj);
915 error = zap_lookup(os, MASTER_NODE_OBJ,
916 zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
917 8, 1, &zfsvfs->z_userquota_obj);
918 if (error && error != ENOENT)
921 error = zap_lookup(os, MASTER_NODE_OBJ,
922 zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
923 8, 1, &zfsvfs->z_groupquota_obj);
924 if (error && error != ENOENT)
927 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
928 &zfsvfs->z_fuid_obj);
929 if (error && error != ENOENT)
932 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
933 &zfsvfs->z_shares_dir);
934 if (error && error != ENOENT)
937 mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
938 mutex_init(&zfsvfs->z_online_recv_lock, NULL, MUTEX_DEFAULT, NULL);
939 mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
940 list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
941 offsetof(znode_t, z_link_node));
942 rrw_init(&zfsvfs->z_teardown_lock);
943 rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
944 rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
945 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
946 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
952 dmu_objset_close(os);
954 kmem_free(zfsvfs, sizeof (zfsvfs_t));
959 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
963 error = zfs_register_callbacks(zfsvfs->z_vfs);
968 * Set the objset user_ptr to track its zfsvfs.
970 mutex_enter(&zfsvfs->z_os->os->os_user_ptr_lock);
971 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
972 mutex_exit(&zfsvfs->z_os->os->os_user_ptr_lock);
974 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
976 zil_destroy(zfsvfs->z_log, 0);
977 zfsvfs->z_log = NULL;
981 * If we are not mounting (ie: online recv), then we don't
982 * have to worry about replaying the log as we blocked all
983 * operations out since we closed the ZIL.
989 * During replay we remove the read only flag to
990 * allow replays to succeed.
992 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
994 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
996 zfs_unlinked_drain(zfsvfs);
1000 * Parse and replay the intent log.
1002 * Because of ziltest, this must be done after
1003 * zfs_unlinked_drain(). (Further note: ziltest
1004 * doesn't use readonly mounts, where
1005 * zfs_unlinked_drain() isn't called.) This is because
1006 * ziltest causes spa_sync() to think it's committed,
1007 * but actually it is not, so the intent log contains
1008 * many txg's worth of changes.
1010 * In particular, if object N is in the unlinked set in
1011 * the last txg to actually sync, then it could be
1012 * actually freed in a later txg and then reallocated
1013 * in a yet later txg. This would write a "create
1014 * object N" record to the intent log. Normally, this
1015 * would be fine because the spa_sync() would have
1016 * written out the fact that object N is free, before
1017 * we could write the "create object N" intent log
1020 * But when we are in ziltest mode, we advance the "open
1021 * txg" without actually spa_sync()-ing the changes to
1022 * disk. So we would see that object N is still
1023 * allocated and in the unlinked set, and there is an
1024 * intent log record saying to allocate it.
1026 zfsvfs->z_replay = B_TRUE;
1027 zil_replay(zfsvfs->z_os, zfsvfs, zfs_replay_vector);
1028 zfsvfs->z_replay = B_FALSE;
1030 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1037 zfsvfs_free(zfsvfs_t *zfsvfs)
1040 extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1043 * This is a barrier to prevent the filesystem from going away in
1044 * zfs_znode_move() until we can safely ensure that the filesystem is
1045 * not unmounted. We consider the filesystem valid before the barrier
1046 * and invalid after the barrier.
1048 rw_enter(&zfsvfs_lock, RW_READER);
1049 rw_exit(&zfsvfs_lock);
1051 zfs_fuid_destroy(zfsvfs);
1053 mutex_destroy(&zfsvfs->z_znodes_lock);
1054 mutex_destroy(&zfsvfs->z_online_recv_lock);
1055 mutex_destroy(&zfsvfs->z_lock);
1056 list_destroy(&zfsvfs->z_all_znodes);
1057 rrw_destroy(&zfsvfs->z_teardown_lock);
1058 rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1059 rw_destroy(&zfsvfs->z_fuid_lock);
1060 for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1061 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1062 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1066 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1068 zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1069 if (zfsvfs->z_use_fuids && zfsvfs->z_vfs) {
1070 vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1071 vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1072 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1073 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1074 vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1079 zfs_domount(vfs_t *vfsp, char *osname)
1082 uint64_t recordsize, fsid_guid;
1089 error = zfsvfs_create(osname, DS_MODE_OWNER, &zfsvfs);
1092 zfsvfs->z_vfs = vfsp;
1094 /* Initialize the generic filesystem structure. */
1095 vfsp->vfs_bcount = 0;
1096 vfsp->vfs_data = NULL;
1098 if (zfs_create_unique_device(&mount_dev) == -1) {
1102 ASSERT(vfs_devismounted(mount_dev) == 0);
1104 if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1108 vfsp->vfs_dev = mount_dev;
1109 vfsp->vfs_fstype = zfsfstype;
1110 vfsp->vfs_bsize = recordsize;
1111 vfsp->vfs_flag |= VFS_NOTRUNC;
1112 vfsp->vfs_data = zfsvfs;
1115 * The fsid is 64 bits, composed of an 8-bit fs type, which
1116 * separates our fsid from any other filesystem types, and a
1117 * 56-bit objset unique ID. The objset unique ID is unique to
1118 * all objsets open on this system, provided by unique_create().
1119 * The 8-bit fs type must be put in the low bits of fsid[1]
1120 * because that's where other Solaris filesystems put it.
1122 fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1123 ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1124 vfsp->vfs_fsid.val[0] = fsid_guid;
1125 vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1129 * Set features for file system.
1131 zfs_set_fuid_feature(zfsvfs);
1132 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1133 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1134 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1135 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1136 } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1137 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1138 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1141 if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1144 atime_changed_cb(zfsvfs, B_FALSE);
1145 readonly_changed_cb(zfsvfs, B_TRUE);
1146 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1148 xattr_changed_cb(zfsvfs, pval);
1149 zfsvfs->z_issnap = B_TRUE;
1151 mutex_enter(&zfsvfs->z_os->os->os_user_ptr_lock);
1152 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1153 mutex_exit(&zfsvfs->z_os->os->os_user_ptr_lock);
1155 error = zfsvfs_setup(zfsvfs, B_TRUE);
1158 if (!zfsvfs->z_issnap)
1159 zfsctl_create(zfsvfs);
1162 dmu_objset_close(zfsvfs->z_os);
1163 zfsvfs_free(zfsvfs);
1165 atomic_add_32(&zfs_active_fs_count, 1);
1172 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1174 objset_t *os = zfsvfs->z_os;
1175 struct dsl_dataset *ds;
1178 * Unregister properties.
1180 if (!dmu_objset_is_snapshot(os)) {
1181 ds = dmu_objset_ds(os);
1182 VERIFY(dsl_prop_unregister(ds, "atime", atime_changed_cb,
1185 VERIFY(dsl_prop_unregister(ds, "xattr", xattr_changed_cb,
1188 VERIFY(dsl_prop_unregister(ds, "recordsize", blksz_changed_cb,
1191 VERIFY(dsl_prop_unregister(ds, "readonly", readonly_changed_cb,
1194 VERIFY(dsl_prop_unregister(ds, "devices", devices_changed_cb,
1197 VERIFY(dsl_prop_unregister(ds, "setuid", setuid_changed_cb,
1200 VERIFY(dsl_prop_unregister(ds, "exec", exec_changed_cb,
1203 VERIFY(dsl_prop_unregister(ds, "snapdir", snapdir_changed_cb,
1206 VERIFY(dsl_prop_unregister(ds, "aclmode", acl_mode_changed_cb,
1209 VERIFY(dsl_prop_unregister(ds, "aclinherit",
1210 acl_inherit_changed_cb, zfsvfs) == 0);
1212 VERIFY(dsl_prop_unregister(ds, "vscan",
1213 vscan_changed_cb, zfsvfs) == 0);
1218 * Convert a decimal digit string to a uint64_t integer.
1221 str_to_uint64(char *str, uint64_t *objnum)
1226 if (*str < '0' || *str > '9')
1229 num = num*10 + *str++ - '0';
1237 * The boot path passed from the boot loader is in the form of
1238 * "rootpool-name/root-filesystem-object-number'. Convert this
1239 * string to a dataset name: "rootpool-name/root-filesystem-name".
1242 zfs_parse_bootfs(char *bpath, char *outpath)
1248 if (*bpath == 0 || *bpath == '/')
1251 (void) strcpy(outpath, bpath);
1253 slashp = strchr(bpath, '/');
1255 /* if no '/', just return the pool name */
1256 if (slashp == NULL) {
1260 /* if not a number, just return the root dataset name */
1261 if (str_to_uint64(slashp+1, &objnum)) {
1266 error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1273 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1276 static int zfsrootdone = 0;
1277 zfsvfs_t *zfsvfs = NULL;
1286 * The filesystem that we mount as root is defined in the
1287 * boot property "zfs-bootfs" with a format of
1288 * "poolname/root-dataset-objnum".
1290 if (why == ROOT_INIT) {
1294 * the process of doing a spa_load will require the
1295 * clock to be set before we could (for example) do
1296 * something better by looking at the timestamp on
1297 * an uberblock, so just set it to -1.
1301 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1302 cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1306 zfs_devid = spa_get_bootprop("diskdevid");
1307 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1309 spa_free_bootprop(zfs_devid);
1311 spa_free_bootprop(zfs_bootfs);
1312 cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1316 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1317 spa_free_bootprop(zfs_bootfs);
1318 cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1323 spa_free_bootprop(zfs_bootfs);
1325 if (error = vfs_lock(vfsp))
1328 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1329 cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1333 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1335 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1336 cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1341 mutex_enter(&vp->v_lock);
1342 vp->v_flag |= VROOT;
1343 mutex_exit(&vp->v_lock);
1347 * Leave rootvp held. The root file system is never unmounted.
1350 vfs_add((struct vnode *)0, vfsp,
1351 (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1355 } else if (why == ROOT_REMOUNT) {
1356 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1357 vfsp->vfs_flag |= VFS_REMOUNT;
1359 /* refresh mount options */
1360 zfs_unregister_callbacks(vfsp->vfs_data);
1361 return (zfs_register_callbacks(vfsp));
1363 } else if (why == ROOT_UNMOUNT) {
1364 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1365 (void) zfs_sync(vfsp, 0, 0);
1370 * if "why" is equal to anything else other than ROOT_INIT,
1371 * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1378 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1383 uio_seg_t fromspace = (uap->flags & MS_SYSSPACE) ?
1384 UIO_SYSSPACE : UIO_USERSPACE;
1387 if (mvp->v_type != VDIR)
1390 mutex_enter(&mvp->v_lock);
1391 if ((uap->flags & MS_REMOUNT) == 0 &&
1392 (uap->flags & MS_OVERLAY) == 0 &&
1393 (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1394 mutex_exit(&mvp->v_lock);
1397 mutex_exit(&mvp->v_lock);
1400 * ZFS does not support passing unparsed data in via MS_DATA.
1401 * Users should use the MS_OPTIONSTR interface; this means
1402 * that all option parsing is already done and the options struct
1403 * can be interrogated.
1405 if ((uap->flags & MS_DATA) && uap->datalen > 0)
1409 * Get the objset name (the "special" mount argument).
1411 if (error = pn_get(uap->spec, fromspace, &spn))
1414 osname = spn.pn_path;
1417 * Check for mount privilege?
1419 * If we don't have privilege then see if
1420 * we have local permission to allow it
1422 error = secpolicy_fs_mount(cr, mvp, vfsp);
1424 error = dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr);
1429 * Make sure user is the owner of the mount point
1430 * or has sufficient privileges.
1433 vattr.va_mask = AT_UID;
1435 if (error = VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1439 if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1440 VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1445 secpolicy_fs_mount_clearopts(cr, vfsp);
1452 * Refuse to mount a filesystem if we are in a local zone and the
1453 * dataset is not visible.
1455 if (!INGLOBALZONE(curproc) &&
1456 (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1462 * When doing a remount, we simply refresh our temporary properties
1463 * according to those options set in the current VFS options.
1465 if (uap->flags & MS_REMOUNT) {
1466 /* refresh mount options */
1467 zfs_unregister_callbacks(vfsp->vfs_data);
1468 error = zfs_register_callbacks(vfsp);
1472 error = zfs_domount(vfsp, osname);
1475 * Add an extra VFS_HOLD on our parent vfs so that it can't
1476 * disappear due to a forced unmount.
1478 if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1479 VFS_HOLD(mvp->v_vfsp);
1487 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1489 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1491 uint64_t refdbytes, availbytes, usedobjs, availobjs;
1495 dmu_objset_space(zfsvfs->z_os,
1496 &refdbytes, &availbytes, &usedobjs, &availobjs);
1499 * The underlying storage pool actually uses multiple block sizes.
1500 * We report the fragsize as the smallest block size we support,
1501 * and we report our blocksize as the filesystem's maximum blocksize.
1503 statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1504 statp->f_bsize = zfsvfs->z_max_blksz;
1507 * The following report "total" blocks of various kinds in the
1508 * file system, but reported in terms of f_frsize - the
1512 statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1513 statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1514 statp->f_bavail = statp->f_bfree; /* no root reservation */
1517 * statvfs() should really be called statufs(), because it assumes
1518 * static metadata. ZFS doesn't preallocate files, so the best
1519 * we can do is report the max that could possibly fit in f_files,
1520 * and that minus the number actually used in f_ffree.
1521 * For f_ffree, report the smaller of the number of object available
1522 * and the number of blocks (each object will take at least a block).
1524 statp->f_ffree = MIN(availobjs, statp->f_bfree);
1525 statp->f_favail = statp->f_ffree; /* no "root reservation" */
1526 statp->f_files = statp->f_ffree + usedobjs;
1528 (void) cmpldev(&d32, vfsp->vfs_dev);
1529 statp->f_fsid = d32;
1532 * We're a zfs filesystem.
1534 (void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1536 statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1538 statp->f_namemax = ZFS_MAXNAMELEN;
1541 * We have all of 32 characters to stuff a string here.
1542 * Is there anything useful we could/should provide?
1544 bzero(statp->f_fstr, sizeof (statp->f_fstr));
1551 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1553 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1559 error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1561 *vpp = ZTOV(rootzp);
1568 * Teardown the zfsvfs::z_os.
1570 * Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
1571 * and 'z_teardown_inactive_lock' held.
1574 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1578 rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1582 * We purge the parent filesystem's vfsp as the parent
1583 * filesystem and all of its snapshots have their vnode's
1584 * v_vfsp set to the parent's filesystem's vfsp. Note,
1585 * 'z_parent' is self referential for non-snapshots.
1587 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1591 * Close the zil. NB: Can't close the zil while zfs_inactive
1592 * threads are blocked as zil_close can call zfs_inactive.
1594 if (zfsvfs->z_log) {
1595 zil_close(zfsvfs->z_log);
1596 zfsvfs->z_log = NULL;
1599 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1602 * If we are not unmounting (ie: online recv) and someone already
1603 * unmounted this file system while we were doing the switcheroo,
1604 * or a reopen of z_os failed then just bail out now.
1606 if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1607 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1608 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1613 * At this point there are no vops active, and any new vops will
1614 * fail with EIO since we have z_teardown_lock for writer (only
1615 * relavent for forced unmount).
1617 * Release all holds on dbufs.
1619 mutex_enter(&zfsvfs->z_znodes_lock);
1620 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1621 zp = list_next(&zfsvfs->z_all_znodes, zp))
1623 ASSERT(ZTOV(zp)->v_count > 0);
1624 zfs_znode_dmu_fini(zp);
1626 mutex_exit(&zfsvfs->z_znodes_lock);
1629 * If we are unmounting, set the unmounted flag and let new vops
1630 * unblock. zfs_inactive will have the unmounted behavior, and all
1631 * other vops will fail with EIO.
1634 zfsvfs->z_unmounted = B_TRUE;
1635 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1636 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1640 * z_os will be NULL if there was an error in attempting to reopen
1641 * zfsvfs, so just return as the properties had already been
1642 * unregistered and cached data had been evicted before.
1644 if (zfsvfs->z_os == NULL)
1648 * Unregister properties.
1650 zfs_unregister_callbacks(zfsvfs);
1655 if (dmu_objset_evict_dbufs(zfsvfs->z_os)) {
1656 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1657 (void) dmu_objset_evict_dbufs(zfsvfs->z_os);
1665 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1667 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1671 ret = secpolicy_fs_unmount(cr, vfsp);
1673 ret = dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1674 ZFS_DELEG_PERM_MOUNT, cr);
1680 * We purge the parent filesystem's vfsp as the parent filesystem
1681 * and all of its snapshots have their vnode's v_vfsp set to the
1682 * parent's filesystem's vfsp. Note, 'z_parent' is self
1683 * referential for non-snapshots.
1685 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1688 * Unmount any snapshots mounted under .zfs before unmounting the
1691 if (zfsvfs->z_ctldir != NULL &&
1692 (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1696 if (!(fflag & MS_FORCE)) {
1698 * Check the number of active vnodes in the file system.
1699 * Our count is maintained in the vfs structure, but the
1700 * number is off by 1 to indicate a hold on the vfs
1703 * The '.zfs' directory maintains a reference of its
1704 * own, and any active references underneath are
1705 * reflected in the vnode count.
1707 if (zfsvfs->z_ctldir == NULL) {
1708 if (vfsp->vfs_count > 1)
1711 if (vfsp->vfs_count > 2 ||
1712 zfsvfs->z_ctldir->v_count > 1)
1717 vfsp->vfs_flag |= VFS_UNMOUNTED;
1719 VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1723 * z_os will be NULL if there was an error in
1724 * attempting to reopen zfsvfs.
1728 * Unset the objset user_ptr.
1730 mutex_enter(&os->os->os_user_ptr_lock);
1731 dmu_objset_set_user(os, NULL);
1732 mutex_exit(&os->os->os_user_ptr_lock);
1735 * Finally release the objset
1737 dmu_objset_close(os);
1741 * We can now safely destroy the '.zfs' directory node.
1743 if (zfsvfs->z_ctldir != NULL)
1744 zfsctl_destroy(zfsvfs);
1750 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1752 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1754 uint64_t object = 0;
1755 uint64_t fid_gen = 0;
1764 if (fidp->fid_len == LONG_FID_LEN) {
1765 zfid_long_t *zlfid = (zfid_long_t *)fidp;
1766 uint64_t objsetid = 0;
1767 uint64_t setgen = 0;
1769 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1770 objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1772 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1773 setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1777 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1783 if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1784 zfid_short_t *zfid = (zfid_short_t *)fidp;
1786 for (i = 0; i < sizeof (zfid->zf_object); i++)
1787 object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1789 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1790 fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1796 /* A zero fid_gen means we are in the .zfs control directories */
1798 (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1799 *vpp = zfsvfs->z_ctldir;
1800 ASSERT(*vpp != NULL);
1801 if (object == ZFSCTL_INO_SNAPDIR) {
1802 VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
1803 0, NULL, NULL, NULL, NULL, NULL) == 0);
1811 gen_mask = -1ULL >> (64 - 8 * i);
1813 dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1814 if (err = zfs_zget(zfsvfs, object, &zp)) {
1818 zp_gen = zp->z_phys->zp_gen & gen_mask;
1821 if (zp->z_unlinked || zp_gen != fid_gen) {
1822 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1834 * Block out VOPs and close zfsvfs_t::z_os
1836 * Note, if successful, then we return with the 'z_teardown_lock' and
1837 * 'z_teardown_inactive_lock' write held.
1840 zfs_suspend_fs(zfsvfs_t *zfsvfs, char *name, int *modep)
1844 if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
1847 *modep = zfsvfs->z_os->os_mode;
1849 dmu_objset_name(zfsvfs->z_os, name);
1850 dmu_objset_close(zfsvfs->z_os);
1856 * Reopen zfsvfs_t::z_os and release VOPs.
1859 zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname, int mode)
1863 ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
1864 ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1866 err = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os);
1868 zfsvfs->z_os = NULL;
1872 VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
1875 * Attempt to re-establish all the active znodes with
1876 * their dbufs. If a zfs_rezget() fails, then we'll let
1877 * any potential callers discover that via ZFS_ENTER_VERIFY_VP
1878 * when they try to use their znode.
1880 mutex_enter(&zfsvfs->z_znodes_lock);
1881 for (zp = list_head(&zfsvfs->z_all_znodes); zp;
1882 zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1883 (void) zfs_rezget(zp);
1885 mutex_exit(&zfsvfs->z_znodes_lock);
1889 /* release the VOPs */
1890 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1891 rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
1895 * Since we couldn't reopen zfsvfs::z_os, force
1896 * unmount this file system.
1898 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
1899 (void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
1905 zfs_freevfs(vfs_t *vfsp)
1907 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1910 * If this is a snapshot, we have an extra VFS_HOLD on our parent
1911 * from zfs_mount(). Release it here.
1913 if (zfsvfs->z_issnap)
1914 VFS_RELE(zfsvfs->z_parent->z_vfs);
1916 zfsvfs_free(zfsvfs);
1918 atomic_add_32(&zfs_active_fs_count, -1);
1922 * VFS_INIT() initialization. Note that there is no VFS_FINI(),
1923 * so we can't safely do any non-idempotent initialization here.
1924 * Leave that to zfs_init() and zfs_fini(), which are called
1925 * from the module's _init() and _fini() entry points.
1929 zfs_vfsinit(int fstype, char *name)
1936 * Setup vfsops and vnodeops tables.
1938 error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
1940 cmn_err(CE_WARN, "zfs: bad vfs ops template");
1943 error = zfs_create_op_tables();
1945 zfs_remove_op_tables();
1946 cmn_err(CE_WARN, "zfs: bad vnode ops template");
1947 (void) vfs_freevfsops_by_type(zfsfstype);
1951 mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
1954 * Unique major number for all zfs mounts.
1955 * If we run out of 32-bit minors, we'll getudev() another major.
1957 zfs_major = ddi_name_to_major(ZFS_DRIVER);
1958 zfs_minor = ZFS_MIN_MINOR;
1967 * Initialize .zfs directory structures
1972 * Initialize znode cache, vnode ops, etc...
1976 dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
1989 return (zfs_active_fs_count != 0);
1993 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
1996 objset_t *os = zfsvfs->z_os;
1999 if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2002 if (newvers < zfsvfs->z_version)
2005 tx = dmu_tx_create(os);
2006 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2007 error = dmu_tx_assign(tx, TXG_WAIT);
2012 error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2013 8, 1, &newvers, tx);
2020 spa_history_internal_log(LOG_DS_UPGRADE,
2021 dmu_objset_spa(os), tx, CRED(),
2022 "oldver=%llu newver=%llu dataset = %llu",
2023 zfsvfs->z_version, newvers, dmu_objset_id(os));
2027 zfsvfs->z_version = newvers;
2029 if (zfsvfs->z_version >= ZPL_VERSION_FUID)
2030 zfs_set_fuid_feature(zfsvfs);
2036 * Read a property stored within the master node.
2039 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2045 * Look up the file system's value for the property. For the
2046 * version property, we look up a slightly different string.
2048 if (prop == ZFS_PROP_VERSION)
2049 pname = ZPL_VERSION_STR;
2051 pname = zfs_prop_to_name(prop);
2054 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2056 if (error == ENOENT) {
2057 /* No value set, use the default value */
2059 case ZFS_PROP_VERSION:
2060 *value = ZPL_VERSION;
2062 case ZFS_PROP_NORMALIZE:
2063 case ZFS_PROP_UTF8ONLY:
2067 *value = ZFS_CASE_SENSITIVE;
2077 static vfsdef_t vfw = {
2081 VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2086 struct modlfs zfs_modlfs = {
2087 &mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw