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) 2013 by Delphix. All rights reserved.
27 * ZFS control directory (a.k.a. ".zfs")
29 * This directory provides a common location for all ZFS meta-objects.
30 * Currently, this is only the 'snapshot' directory, but this may expand in the
31 * future. The elements are built using the GFS primitives, as the hierarchy
32 * does not actually exist on disk.
34 * For 'snapshot', we don't want to have all snapshots always mounted, because
35 * this would take up a huge amount of space in /etc/mnttab. We have three
38 * ctldir ------> snapshotdir -------> snapshot
44 * The 'snapshot' node contains just enough information to lookup '..' and act
45 * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we
46 * perform an automount of the underlying filesystem and return the
47 * corresponding vnode.
49 * All mounts are handled automatically by the kernel, but unmounts are
50 * (currently) handled from user land. The main reason is that there is no
51 * reliable way to auto-unmount the filesystem when it's "no longer in use".
52 * When the user unmounts a filesystem, we call zfsctl_unmount(), which
53 * unmounts any snapshots within the snapshot directory.
55 * The '.zfs', '.zfs/snapshot', and all directories created under
56 * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') are all GFS nodes and
57 * share the same vfs_t as the head filesystem (what '.zfs' lives under).
59 * File systems mounted ontop of the GFS nodes '.zfs/snapshot/<snapname>'
60 * (ie: snapshots) are ZFS nodes and have their own unique vfs_t.
61 * However, vnodes within these mounted on file systems have their v_vfsp
62 * fields set to the head filesystem to make NFS happy (see
63 * zfsctl_snapdir_lookup()). We VFS_HOLD the head filesystem's vfs_t
64 * so that it cannot be freed until all snapshots have been unmounted.
67 #include <sys/zfs_context.h>
68 #include <sys/zfs_ctldir.h>
69 #include <sys/zfs_ioctl.h>
70 #include <sys/zfs_vfsops.h>
71 #include <sys/namei.h>
75 #include <sys/dsl_destroy.h>
76 #include <sys/dsl_deleg.h>
77 #include <sys/mount.h>
78 #include <sys/sunddi.h>
80 #include "zfs_namecheck.h"
82 typedef struct zfsctl_node {
83 gfs_dir_t zc_gfs_private;
85 timestruc_t zc_cmtime; /* ctime and mtime, always the same */
88 typedef struct zfsctl_snapdir {
89 zfsctl_node_t sd_node;
101 snapentry_compare(const void *a, const void *b)
103 const zfs_snapentry_t *sa = a;
104 const zfs_snapentry_t *sb = b;
105 int ret = strcmp(sa->se_name, sb->se_name);
116 vnodeops_t *zfsctl_ops_root;
117 vnodeops_t *zfsctl_ops_snapdir;
118 vnodeops_t *zfsctl_ops_snapshot;
119 vnodeops_t *zfsctl_ops_shares;
120 vnodeops_t *zfsctl_ops_shares_dir;
122 static const fs_operation_def_t zfsctl_tops_root[];
123 static const fs_operation_def_t zfsctl_tops_snapdir[];
124 static const fs_operation_def_t zfsctl_tops_snapshot[];
125 static const fs_operation_def_t zfsctl_tops_shares[];
127 static struct vop_vector zfsctl_ops_root;
128 static struct vop_vector zfsctl_ops_snapdir;
129 static struct vop_vector zfsctl_ops_snapshot;
130 static struct vop_vector zfsctl_ops_shares;
131 static struct vop_vector zfsctl_ops_shares_dir;
134 static vnode_t *zfsctl_mknode_snapdir(vnode_t *);
135 static vnode_t *zfsctl_mknode_shares(vnode_t *);
136 static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset);
137 static int zfsctl_unmount_snap(zfs_snapentry_t *, int, cred_t *);
140 static gfs_opsvec_t zfsctl_opsvec[] = {
141 { ".zfs", zfsctl_tops_root, &zfsctl_ops_root },
142 { ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir },
143 { ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot },
144 { ".zfs/shares", zfsctl_tops_shares, &zfsctl_ops_shares_dir },
145 { ".zfs/shares/vnode", zfsctl_tops_shares, &zfsctl_ops_shares },
151 * Root directory elements. We only have two entries
152 * snapshot and shares.
154 static gfs_dirent_t zfsctl_root_entries[] = {
155 { "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE },
156 { "shares", zfsctl_mknode_shares, GFS_CACHE_VNODE },
160 /* include . and .. in the calculation */
161 #define NROOT_ENTRIES ((sizeof (zfsctl_root_entries) / \
162 sizeof (gfs_dirent_t)) + 1)
166 * Initialize the various GFS pieces we'll need to create and manipulate .zfs
167 * directories. This is called from the ZFS init routine, and initializes the
168 * vnode ops vectors that we'll be using.
174 VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0);
183 * Remove vfsctl vnode ops
186 vn_freevnodeops(zfsctl_ops_root);
187 if (zfsctl_ops_snapdir)
188 vn_freevnodeops(zfsctl_ops_snapdir);
189 if (zfsctl_ops_snapshot)
190 vn_freevnodeops(zfsctl_ops_snapshot);
191 if (zfsctl_ops_shares)
192 vn_freevnodeops(zfsctl_ops_shares);
193 if (zfsctl_ops_shares_dir)
194 vn_freevnodeops(zfsctl_ops_shares_dir);
196 zfsctl_ops_root = NULL;
197 zfsctl_ops_snapdir = NULL;
198 zfsctl_ops_snapshot = NULL;
199 zfsctl_ops_shares = NULL;
200 zfsctl_ops_shares_dir = NULL;
205 zfsctl_is_node(vnode_t *vp)
207 return (vn_matchops(vp, zfsctl_ops_root) ||
208 vn_matchops(vp, zfsctl_ops_snapdir) ||
209 vn_matchops(vp, zfsctl_ops_snapshot) ||
210 vn_matchops(vp, zfsctl_ops_shares) ||
211 vn_matchops(vp, zfsctl_ops_shares_dir));
216 * Return the inode number associated with the 'snapshot' or
217 * 'shares' directory.
221 zfsctl_root_inode_cb(vnode_t *vp, int index)
223 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
228 return (ZFSCTL_INO_SNAPDIR);
230 return (zfsvfs->z_shares_dir);
234 * Create the '.zfs' directory. This directory is cached as part of the VFS
235 * structure. This results in a hold on the vfs_t. The code in zfs_umount()
236 * therefore checks against a vfs_count of 2 instead of 1. This reference
237 * is removed when the ctldir is destroyed in the unmount.
240 zfsctl_create(zfsvfs_t *zfsvfs)
246 ASSERT(zfsvfs->z_ctldir == NULL);
248 vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs,
249 &zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries,
250 zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL);
252 zcp->zc_id = ZFSCTL_INO_ROOT;
254 VERIFY(VFS_ROOT(zfsvfs->z_vfs, LK_EXCLUSIVE, &rvp) == 0);
255 VERIFY(0 == sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
256 &crtime, sizeof (crtime)));
257 ZFS_TIME_DECODE(&zcp->zc_cmtime, crtime);
261 * We're only faking the fact that we have a root of a filesystem for
262 * the sake of the GFS interfaces. Undo the flag manipulation it did
265 vp->v_vflag &= ~VV_ROOT;
267 zfsvfs->z_ctldir = vp;
273 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted.
274 * There might still be more references if we were force unmounted, but only
275 * new zfs_inactive() calls can occur and they don't reference .zfs
278 zfsctl_destroy(zfsvfs_t *zfsvfs)
280 VN_RELE(zfsvfs->z_ctldir);
281 zfsvfs->z_ctldir = NULL;
285 * Given a root znode, retrieve the associated .zfs directory.
286 * Add a hold to the vnode and return it.
289 zfsctl_root(znode_t *zp)
291 ASSERT(zfs_has_ctldir(zp));
292 VN_HOLD(zp->z_zfsvfs->z_ctldir);
293 return (zp->z_zfsvfs->z_ctldir);
297 * Common open routine. Disallow any write access.
301 zfsctl_common_open(struct vop_open_args *ap)
303 int flags = ap->a_mode;
306 return (SET_ERROR(EACCES));
312 * Common close routine. Nothing to do here.
316 zfsctl_common_close(struct vop_close_args *ap)
322 * Common access routine. Disallow writes.
326 zfsctl_common_access(ap)
327 struct vop_access_args /* {
330 struct ucred *a_cred;
334 accmode_t accmode = ap->a_accmode;
337 if (flags & V_ACE_MASK) {
338 if (accmode & ACE_ALL_WRITE_PERMS)
339 return (SET_ERROR(EACCES));
342 if (accmode & VWRITE)
343 return (SET_ERROR(EACCES));
352 * Common getattr function. Fill in basic information.
355 zfsctl_common_getattr(vnode_t *vp, vattr_t *vap)
363 * We are a purely virtual object, so we have no
364 * blocksize or allocated blocks.
369 vap->va_fsid = vp->v_mount->mnt_stat.f_fsid.val[0];
370 vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP |
374 * We live in the now (for atime).
378 /* FreeBSD: Reset chflags(2) flags. */
384 zfsctl_common_fid(ap)
385 struct vop_fid_args /* {
390 vnode_t *vp = ap->a_vp;
391 fid_t *fidp = (void *)ap->a_fid;
392 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
393 zfsctl_node_t *zcp = vp->v_data;
394 uint64_t object = zcp->zc_id;
401 if (fidp->fid_len < SHORT_FID_LEN) {
402 fidp->fid_len = SHORT_FID_LEN;
404 return (SET_ERROR(ENOSPC));
407 fidp->fid_len = SHORT_FID_LEN;
410 zfid = (zfid_short_t *)fidp;
412 zfid->zf_len = SHORT_FID_LEN;
414 for (i = 0; i < sizeof (zfid->zf_object); i++)
415 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
417 /* .zfs znodes always have a generation number of 0 */
418 for (i = 0; i < sizeof (zfid->zf_gen); i++)
428 zfsctl_shares_fid(ap)
429 struct vop_fid_args /* {
434 vnode_t *vp = ap->a_vp;
435 fid_t *fidp = (void *)ap->a_fid;
436 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
442 if (zfsvfs->z_shares_dir == 0) {
444 return (SET_ERROR(ENOTSUP));
447 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
448 error = VOP_FID(ZTOV(dzp), fidp);
457 zfsctl_common_reclaim(ap)
458 struct vop_reclaim_args /* {
463 vnode_t *vp = ap->a_vp;
466 * Destroy the vm object and flush associated pages.
468 vnode_destroy_vobject(vp);
476 * .zfs inode namespace
478 * We need to generate unique inode numbers for all files and directories
479 * within the .zfs pseudo-filesystem. We use the following scheme:
484 * .zfs/snapshot/<snap> objectid(snap)
487 #define ZFSCTL_INO_SNAP(id) (id)
490 * Get root directory attributes.
494 zfsctl_root_getattr(ap)
495 struct vop_getattr_args /* {
498 struct ucred *a_cred;
501 struct vnode *vp = ap->a_vp;
502 struct vattr *vap = ap->a_vap;
503 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
504 zfsctl_node_t *zcp = vp->v_data;
507 vap->va_nodeid = ZFSCTL_INO_ROOT;
508 vap->va_nlink = vap->va_size = NROOT_ENTRIES;
509 vap->va_mtime = vap->va_ctime = zcp->zc_cmtime;
510 vap->va_birthtime = vap->va_ctime;
512 zfsctl_common_getattr(vp, vap);
519 * Special case the handling of "..".
523 zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
524 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
525 int *direntflags, pathname_t *realpnp)
527 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
531 * No extended attributes allowed under .zfs
533 if (flags & LOOKUP_XATTR)
534 return (SET_ERROR(EINVAL));
538 if (strcmp(nm, "..") == 0) {
539 err = VFS_ROOT(dvp->v_vfsp, LK_EXCLUSIVE, vpp);
543 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir,
544 cr, ct, direntflags, realpnp);
554 zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
555 caller_context_t *ct)
558 * We only care about ACL_ENABLED so that libsec can
559 * display ACL correctly and not default to POSIX draft.
561 if (cmd == _PC_ACL_ENABLED) {
562 *valp = _ACL_ACE_ENABLED;
566 return (fs_pathconf(vp, cmd, valp, cr, ct));
571 static const fs_operation_def_t zfsctl_tops_root[] = {
572 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
573 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
574 { VOPNAME_IOCTL, { .error = fs_inval } },
575 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } },
576 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
577 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
578 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } },
579 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
580 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } },
581 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } },
582 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
588 * Special case the handling of "..".
592 zfsctl_freebsd_root_lookup(ap)
593 struct vop_lookup_args /* {
595 struct vnode **a_vpp;
596 struct componentname *a_cnp;
599 vnode_t *dvp = ap->a_dvp;
600 vnode_t **vpp = ap->a_vpp;
601 cred_t *cr = ap->a_cnp->cn_cred;
602 int flags = ap->a_cnp->cn_flags;
603 int nameiop = ap->a_cnp->cn_nameiop;
604 char nm[NAME_MAX + 1];
608 if ((flags & ISLASTCN) && (nameiop == RENAME || nameiop == CREATE))
611 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm));
612 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1);
613 err = zfsctl_root_lookup(dvp, nm, vpp, NULL, 0, NULL, cr, NULL, NULL, NULL);
614 if (err == 0 && (nm[0] != '.' || nm[1] != '\0')) {
615 ltype = VOP_ISLOCKED(dvp);
616 if (flags & ISDOTDOT) {
620 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
621 if (flags & ISDOTDOT) {
623 vn_lock(dvp, ltype| LK_RETRY);
630 static struct vop_vector zfsctl_ops_root = {
631 .vop_default = &default_vnodeops,
632 .vop_open = zfsctl_common_open,
633 .vop_close = zfsctl_common_close,
634 .vop_ioctl = VOP_EINVAL,
635 .vop_getattr = zfsctl_root_getattr,
636 .vop_access = zfsctl_common_access,
637 .vop_readdir = gfs_vop_readdir,
638 .vop_lookup = zfsctl_freebsd_root_lookup,
639 .vop_inactive = VOP_NULL,
640 .vop_reclaim = gfs_vop_reclaim,
642 .vop_pathconf = zfsctl_pathconf,
644 .vop_fid = zfsctl_common_fid,
648 * Gets the full dataset name that corresponds to the given snapshot name
650 * zfsctl_snapshot_zname("snap1") -> "mypool/myfs@snap1"
653 zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname)
655 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
657 if (zfs_component_namecheck(name, NULL, NULL) != 0)
658 return (SET_ERROR(EILSEQ));
659 dmu_objset_name(os, zname);
660 if (strlen(zname) + 1 + strlen(name) >= len)
661 return (SET_ERROR(ENAMETOOLONG));
662 (void) strcat(zname, "@");
663 (void) strcat(zname, name);
668 zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr)
670 vnode_t *svp = sep->se_root;
673 ASSERT(vn_ismntpt(svp));
675 /* this will be dropped by dounmount() */
676 if ((error = vn_vfswlock(svp)) != 0)
681 error = dounmount(vn_mountedvfs(svp), fflags, cr);
688 * We can't use VN_RELE(), as that will try to invoke
689 * zfsctl_snapdir_inactive(), which would cause us to destroy
690 * the sd_lock mutex held by our caller.
692 ASSERT(svp->v_count == 1);
693 gfs_vop_reclaim(svp, cr, NULL);
695 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
696 kmem_free(sep, sizeof (zfs_snapentry_t));
700 return (dounmount(vn_mountedvfs(svp), fflags, curthread));
706 zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm)
711 char newpath[MAXNAMELEN];
714 ASSERT(MUTEX_HELD(&sdp->sd_lock));
717 vfsp = vn_mountedvfs(sep->se_root);
718 ASSERT(vfsp != NULL);
723 * Change the name in the AVL tree.
725 avl_remove(&sdp->sd_snaps, sep);
726 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
727 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
728 (void) strcpy(sep->se_name, nm);
729 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL);
730 avl_insert(&sdp->sd_snaps, sep, where);
733 * Change the current mountpoint info:
734 * - update the tail of the mntpoint path
735 * - update the tail of the resource path
737 pathref = vfs_getmntpoint(vfsp);
738 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
739 VERIFY((tail = strrchr(newpath, '/')) != NULL);
741 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
742 (void) strcat(newpath, nm);
743 refstr_rele(pathref);
744 vfs_setmntpoint(vfsp, newpath, 0);
746 pathref = vfs_getresource(vfsp);
747 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
748 VERIFY((tail = strrchr(newpath, '@')) != NULL);
750 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
751 (void) strcat(newpath, nm);
752 refstr_rele(pathref);
753 vfs_setresource(vfsp, newpath, 0);
762 zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm,
763 cred_t *cr, caller_context_t *ct, int flags)
765 zfsctl_snapdir_t *sdp = sdvp->v_data;
766 zfs_snapentry_t search, *sep;
769 char from[MAXNAMELEN], to[MAXNAMELEN];
770 char real[MAXNAMELEN], fsname[MAXNAMELEN];
773 zfsvfs = sdvp->v_vfsp->vfs_data;
776 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
777 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real,
781 } else if (err != ENOTSUP) {
789 dmu_objset_name(zfsvfs->z_os, fsname);
791 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from);
793 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to);
795 err = zfs_secpolicy_rename_perms(from, to, cr);
800 * Cannot move snapshots out of the snapdir.
803 return (SET_ERROR(EINVAL));
805 if (strcmp(snm, tnm) == 0)
808 mutex_enter(&sdp->sd_lock);
810 search.se_name = (char *)snm;
811 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) {
812 mutex_exit(&sdp->sd_lock);
813 return (SET_ERROR(ENOENT));
816 err = dsl_dataset_rename_snapshot(fsname, snm, tnm, 0);
818 zfsctl_rename_snap(sdp, sep, tnm);
820 mutex_exit(&sdp->sd_lock);
829 zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
830 caller_context_t *ct, int flags)
832 zfsctl_snapdir_t *sdp = dvp->v_data;
833 zfs_snapentry_t *sep;
834 zfs_snapentry_t search;
836 char snapname[MAXNAMELEN];
837 char real[MAXNAMELEN];
840 zfsvfs = dvp->v_vfsp->vfs_data;
843 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
845 err = dmu_snapshot_realname(zfsvfs->z_os, name, real,
849 } else if (err != ENOTSUP) {
857 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname);
859 err = zfs_secpolicy_destroy_perms(snapname, cr);
863 mutex_enter(&sdp->sd_lock);
865 search.se_name = name;
866 sep = avl_find(&sdp->sd_snaps, &search, NULL);
868 avl_remove(&sdp->sd_snaps, sep);
869 err = zfsctl_unmount_snap(sep, MS_FORCE, cr);
871 avl_add(&sdp->sd_snaps, sep);
873 err = dsl_destroy_snapshot(snapname, B_FALSE);
875 err = SET_ERROR(ENOENT);
878 mutex_exit(&sdp->sd_lock);
885 * This creates a snapshot under '.zfs/snapshot'.
889 zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp,
890 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp)
892 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
893 char name[MAXNAMELEN];
895 static enum symfollow follow = NO_FOLLOW;
896 static enum uio_seg seg = UIO_SYSSPACE;
898 if (zfs_component_namecheck(dirname, NULL, NULL) != 0)
899 return (SET_ERROR(EILSEQ));
901 dmu_objset_name(zfsvfs->z_os, name);
905 err = zfs_secpolicy_snapshot_perms(name, cr);
910 err = dmu_objset_snapshot_one(name, dirname);
913 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp);
920 zfsctl_freebsd_snapdir_mkdir(ap)
921 struct vop_mkdir_args /* {
923 struct vnode **a_vpp;
924 struct componentname *a_cnp;
929 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
931 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL,
932 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL));
936 * Lookup entry point for the 'snapshot' directory. Try to open the
937 * snapshot if it exist, creating the pseudo filesystem vnode as necessary.
938 * Perform a mount of the associated dataset on top of the vnode.
942 zfsctl_snapdir_lookup(ap)
943 struct vop_lookup_args /* {
945 struct vnode **a_vpp;
946 struct componentname *a_cnp;
949 vnode_t *dvp = ap->a_dvp;
950 vnode_t **vpp = ap->a_vpp;
951 struct componentname *cnp = ap->a_cnp;
952 char nm[NAME_MAX + 1];
953 zfsctl_snapdir_t *sdp = dvp->v_data;
955 char snapname[MAXNAMELEN];
956 char real[MAXNAMELEN];
958 zfs_snapentry_t *sep, search;
959 size_t mountpoint_len;
961 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
963 int ltype, flags = 0;
966 * No extended attributes allowed under .zfs
968 if (flags & LOOKUP_XATTR)
969 return (SET_ERROR(EINVAL));
970 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm));
971 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1);
973 ASSERT(dvp->v_type == VDIR);
978 * If we get a recursive call, that means we got called
979 * from the domount() code while it was trying to look up the
980 * spec (which looks like a local path for zfs). We need to
981 * add some flag to domount() to tell it not to do this lookup.
983 if (MUTEX_HELD(&sdp->sd_lock))
984 return (SET_ERROR(ENOENT));
987 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
992 if (flags & FIGNORECASE) {
993 boolean_t conflict = B_FALSE;
995 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real,
996 MAXNAMELEN, &conflict);
998 strlcpy(nm, real, sizeof(nm));
999 } else if (err != ENOTSUP) {
1005 (void) strlcpy(realpnp->pn_buf, nm,
1006 realpnp->pn_bufsize);
1007 if (conflict && direntflags)
1008 *direntflags = ED_CASE_CONFLICT;
1012 mutex_enter(&sdp->sd_lock);
1013 search.se_name = (char *)nm;
1014 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) {
1015 *vpp = sep->se_root;
1017 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY);
1021 } else if (*vpp == sep->se_root) {
1023 * The snapshot was unmounted behind our backs,
1024 * try to remount it.
1026 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0);
1030 * VROOT was set during the traverse call. We need
1031 * to clear it since we're pretending to be part
1032 * of our parent's vfs.
1034 (*vpp)->v_flag &= ~VROOT;
1036 mutex_exit(&sdp->sd_lock);
1042 * The requested snapshot is not currently mounted, look it up.
1044 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname);
1046 mutex_exit(&sdp->sd_lock);
1049 * handle "ls *" or "?" in a graceful manner,
1050 * forcing EILSEQ to ENOENT.
1051 * Since shell ultimately passes "*" or "?" as name to lookup
1053 return (err == EILSEQ ? ENOENT : err);
1055 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) {
1056 mutex_exit(&sdp->sd_lock);
1059 return (SET_ERROR(ENOENT));
1060 #else /* !illumos */
1061 /* Translate errors and add SAVENAME when needed. */
1062 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) {
1064 cnp->cn_flags |= SAVENAME;
1066 err = SET_ERROR(ENOENT);
1070 #endif /* !illumos */
1073 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP);
1074 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
1075 (void) strcpy(sep->se_name, nm);
1076 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap));
1078 avl_insert(&sdp->sd_snaps, sep, where);
1080 dmu_objset_rele(snap, FTAG);
1082 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) +
1083 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1;
1084 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP);
1085 (void) snprintf(mountpoint, mountpoint_len,
1086 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s",
1087 dvp->v_vfsp->mnt_stat.f_mntonname, nm);
1088 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0);
1089 kmem_free(mountpoint, mountpoint_len);
1092 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>.
1094 * This is where we lie about our v_vfsp in order to
1095 * make .zfs/snapshot/<snapname> accessible over NFS
1096 * without requiring manual mounts of <snapname>.
1098 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs);
1099 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs;
1101 mutex_exit(&sdp->sd_lock);
1106 * If we had an error, drop our hold on the vnode and
1107 * zfsctl_snapshot_inactive() will clean up.
1122 zfsctl_shares_lookup(ap)
1123 struct vop_lookup_args /* {
1124 struct vnode *a_dvp;
1125 struct vnode **a_vpp;
1126 struct componentname *a_cnp;
1129 vnode_t *dvp = ap->a_dvp;
1130 vnode_t **vpp = ap->a_vpp;
1131 struct componentname *cnp = ap->a_cnp;
1132 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
1133 char nm[NAME_MAX + 1];
1139 ASSERT(cnp->cn_namelen < sizeof(nm));
1140 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1);
1142 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
1147 if (zfsvfs->z_shares_dir == 0) {
1149 return (SET_ERROR(ENOTSUP));
1151 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0)
1152 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp);
1162 zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp,
1163 offset_t *offp, offset_t *nextp, void *data, int flags)
1165 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1166 char snapname[MAXNAMELEN];
1167 uint64_t id, cookie;
1168 boolean_t case_conflict;
1174 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG);
1175 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id,
1176 &cookie, &case_conflict);
1177 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG);
1180 if (error == ENOENT) {
1187 if (flags & V_RDDIR_ENTFLAGS) {
1188 edirent_t *eodp = dp;
1190 (void) strcpy(eodp->ed_name, snapname);
1191 eodp->ed_ino = ZFSCTL_INO_SNAP(id);
1192 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0;
1194 struct dirent64 *odp = dp;
1196 (void) strcpy(odp->d_name, snapname);
1197 odp->d_ino = ZFSCTL_INO_SNAP(id);
1208 zfsctl_shares_readdir(ap)
1209 struct vop_readdir_args /* {
1212 struct ucred *a_cred;
1218 vnode_t *vp = ap->a_vp;
1219 uio_t *uiop = ap->a_uio;
1220 cred_t *cr = ap->a_cred;
1221 int *eofp = ap->a_eofflag;
1222 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1228 if (zfsvfs->z_shares_dir == 0) {
1230 return (SET_ERROR(ENOTSUP));
1232 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1233 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY);
1234 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies);
1235 VN_URELE(ZTOV(dzp));
1238 error = SET_ERROR(ENOENT);
1246 * pvp is the '.zfs' directory (zfsctl_node_t).
1248 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t).
1250 * This function is the callback to create a GFS vnode for '.zfs/snapshot'
1251 * when a lookup is performed on .zfs for "snapshot".
1254 zfsctl_mknode_snapdir(vnode_t *pvp)
1257 zfsctl_snapdir_t *sdp;
1259 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp,
1260 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN,
1261 zfsctl_snapdir_readdir_cb, NULL);
1263 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR;
1264 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1265 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL);
1266 avl_create(&sdp->sd_snaps, snapentry_compare,
1267 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node));
1273 zfsctl_mknode_shares(vnode_t *pvp)
1278 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp,
1279 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN,
1282 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1290 zfsctl_shares_getattr(ap)
1291 struct vop_getattr_args /* {
1293 struct vattr *a_vap;
1294 struct ucred *a_cred;
1295 struct thread *a_td;
1298 vnode_t *vp = ap->a_vp;
1299 vattr_t *vap = ap->a_vap;
1300 cred_t *cr = ap->a_cred;
1301 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1306 if (zfsvfs->z_shares_dir == 0) {
1308 return (SET_ERROR(ENOTSUP));
1310 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1311 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY);
1312 error = VOP_GETATTR(ZTOV(dzp), vap, cr);
1313 VN_URELE(ZTOV(dzp));
1323 zfsctl_snapdir_getattr(ap)
1324 struct vop_getattr_args /* {
1326 struct vattr *a_vap;
1327 struct ucred *a_cred;
1330 vnode_t *vp = ap->a_vp;
1331 vattr_t *vap = ap->a_vap;
1332 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1333 zfsctl_snapdir_t *sdp = vp->v_data;
1336 zfsctl_common_getattr(vp, vap);
1337 vap->va_nodeid = gfs_file_inode(vp);
1338 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2;
1339 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os);
1340 vap->va_birthtime = vap->va_ctime;
1348 zfsctl_snapdir_inactive(ap)
1349 struct vop_inactive_args /* {
1351 struct thread *a_td;
1354 vnode_t *vp = ap->a_vp;
1355 zfsctl_snapdir_t *sdp = vp->v_data;
1356 zfs_snapentry_t *sep;
1359 * On forced unmount we have to free snapshots from here.
1361 mutex_enter(&sdp->sd_lock);
1362 while ((sep = avl_first(&sdp->sd_snaps)) != NULL) {
1363 avl_remove(&sdp->sd_snaps, sep);
1364 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
1365 kmem_free(sep, sizeof (zfs_snapentry_t));
1367 mutex_exit(&sdp->sd_lock);
1368 gfs_dir_inactive(vp);
1369 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0);
1370 mutex_destroy(&sdp->sd_lock);
1371 avl_destroy(&sdp->sd_snaps);
1372 kmem_free(sdp, sizeof (zfsctl_snapdir_t));
1378 static const fs_operation_def_t zfsctl_tops_snapdir[] = {
1379 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
1380 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
1381 { VOPNAME_IOCTL, { .error = fs_inval } },
1382 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } },
1383 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
1384 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } },
1385 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } },
1386 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } },
1387 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
1388 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } },
1389 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
1390 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } },
1391 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
1395 static const fs_operation_def_t zfsctl_tops_shares[] = {
1396 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
1397 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
1398 { VOPNAME_IOCTL, { .error = fs_inval } },
1399 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } },
1400 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
1401 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } },
1402 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } },
1403 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
1404 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } },
1405 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } },
1409 static struct vop_vector zfsctl_ops_snapdir = {
1410 .vop_default = &default_vnodeops,
1411 .vop_open = zfsctl_common_open,
1412 .vop_close = zfsctl_common_close,
1413 .vop_ioctl = VOP_EINVAL,
1414 .vop_getattr = zfsctl_snapdir_getattr,
1415 .vop_access = zfsctl_common_access,
1416 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir,
1417 .vop_readdir = gfs_vop_readdir,
1418 .vop_lookup = zfsctl_snapdir_lookup,
1419 .vop_inactive = zfsctl_snapdir_inactive,
1420 .vop_reclaim = zfsctl_common_reclaim,
1421 .vop_fid = zfsctl_common_fid,
1424 static struct vop_vector zfsctl_ops_shares = {
1425 .vop_default = &default_vnodeops,
1426 .vop_open = zfsctl_common_open,
1427 .vop_close = zfsctl_common_close,
1428 .vop_ioctl = VOP_EINVAL,
1429 .vop_getattr = zfsctl_shares_getattr,
1430 .vop_access = zfsctl_common_access,
1431 .vop_readdir = zfsctl_shares_readdir,
1432 .vop_lookup = zfsctl_shares_lookup,
1433 .vop_inactive = VOP_NULL,
1434 .vop_reclaim = gfs_vop_reclaim,
1435 .vop_fid = zfsctl_shares_fid,
1440 * pvp is the GFS vnode '.zfs/snapshot'.
1442 * This creates a GFS node under '.zfs/snapshot' representing each
1443 * snapshot. This newly created GFS node is what we mount snapshot
1447 zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset)
1452 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp,
1453 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL);
1456 zcp->zc_id = objset;
1464 zfsctl_snapshot_reclaim(ap)
1465 struct vop_inactive_args /* {
1467 struct thread *a_td;
1470 vnode_t *vp = ap->a_vp;
1471 cred_t *cr = ap->a_td->td_ucred;
1472 struct vop_reclaim_args iap;
1473 zfsctl_snapdir_t *sdp;
1474 zfs_snapentry_t *sep, *next;
1478 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0);
1481 /* this may already have been unmounted */
1486 if (!(locked = MUTEX_HELD(&sdp->sd_lock)))
1487 mutex_enter(&sdp->sd_lock);
1489 ASSERT(!vn_ismntpt(vp));
1491 sep = avl_first(&sdp->sd_snaps);
1492 while (sep != NULL) {
1493 next = AVL_NEXT(&sdp->sd_snaps, sep);
1495 if (sep->se_root == vp) {
1496 avl_remove(&sdp->sd_snaps, sep);
1497 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
1498 kmem_free(sep, sizeof (zfs_snapentry_t));
1503 ASSERT(sep != NULL);
1506 mutex_exit(&sdp->sd_lock);
1510 * Dispose of the vnode for the snapshot mount point.
1511 * This is safe to do because once this entry has been removed
1512 * from the AVL tree, it can't be found again, so cannot become
1513 * "active". If we lookup the same name again we will end up
1514 * creating a new vnode.
1517 gfs_vop_reclaim(&iap);
1523 zfsctl_traverse_begin(vnode_t **vpp, int lktype)
1527 /* Snapshot should be already mounted, but just in case. */
1528 if (vn_mountedvfs(*vpp) == NULL)
1530 return (traverse(vpp, lktype));
1534 zfsctl_traverse_end(vnode_t *vp, int err)
1544 zfsctl_snapshot_getattr(ap)
1545 struct vop_getattr_args /* {
1547 struct vattr *a_vap;
1548 struct ucred *a_cred;
1551 vnode_t *vp = ap->a_vp;
1554 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY);
1556 err = VOP_GETATTR(vp, ap->a_vap, ap->a_cred);
1557 zfsctl_traverse_end(vp, err);
1562 zfsctl_snapshot_fid(ap)
1563 struct vop_fid_args /* {
1568 vnode_t *vp = ap->a_vp;
1571 err = zfsctl_traverse_begin(&vp, LK_SHARED | LK_RETRY);
1573 err = VOP_VPTOFH(vp, (void *)ap->a_fid);
1574 zfsctl_traverse_end(vp, err);
1579 zfsctl_snapshot_lookup(ap)
1580 struct vop_lookup_args /* {
1581 struct vnode *a_dvp;
1582 struct vnode **a_vpp;
1583 struct componentname *a_cnp;
1586 vnode_t *dvp = ap->a_dvp;
1587 vnode_t **vpp = ap->a_vpp;
1588 struct componentname *cnp = ap->a_cnp;
1589 cred_t *cr = ap->a_cnp->cn_cred;
1590 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
1593 if (cnp->cn_namelen != 2 || cnp->cn_nameptr[0] != '.' ||
1594 cnp->cn_nameptr[1] != '.') {
1598 ASSERT(dvp->v_type == VDIR);
1599 ASSERT(zfsvfs->z_ctldir != NULL);
1601 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", vpp,
1602 NULL, 0, NULL, cr, NULL, NULL, NULL);
1604 int ltype = VOP_ISLOCKED(dvp);
1607 vn_lock(*vpp, LK_EXCLUSIVE | LK_RETRY);
1609 vn_lock(dvp, ltype | LK_RETRY);
1616 zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap)
1618 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data;
1620 zfsctl_snapdir_t *sdp;
1621 zfs_snapentry_t *sep;
1624 ASSERT(zfsvfs->z_ctldir != NULL);
1625 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1626 NULL, 0, NULL, kcred, NULL, NULL, NULL);
1631 mutex_enter(&sdp->sd_lock);
1632 sep = avl_first(&sdp->sd_snaps);
1633 while (sep != NULL) {
1637 sep = AVL_NEXT(&sdp->sd_snaps, sep);
1640 mutex_exit(&sdp->sd_lock);
1645 len = strlen(sep->se_name);
1646 *ap->a_buflen -= len;
1647 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len);
1648 mutex_exit(&sdp->sd_lock);
1658 * These VP's should never see the light of day. They should always
1661 static struct vop_vector zfsctl_ops_snapshot = {
1662 .vop_default = &default_vnodeops,
1663 .vop_inactive = VOP_NULL,
1664 .vop_lookup = zfsctl_snapshot_lookup,
1665 .vop_reclaim = zfsctl_snapshot_reclaim,
1666 .vop_getattr = zfsctl_snapshot_getattr,
1667 .vop_fid = zfsctl_snapshot_fid,
1668 .vop_vptocnp = zfsctl_snapshot_vptocnp,
1672 zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp)
1674 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1676 zfsctl_snapdir_t *sdp;
1678 zfs_snapentry_t *sep;
1681 ASSERT(zfsvfs->z_ctldir != NULL);
1682 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1683 NULL, 0, NULL, kcred, NULL, NULL, NULL);
1688 mutex_enter(&sdp->sd_lock);
1689 sep = avl_first(&sdp->sd_snaps);
1690 while (sep != NULL) {
1693 if (zcp->zc_id == objsetid)
1696 sep = AVL_NEXT(&sdp->sd_snaps, sep);
1702 * Return the mounted root rather than the covered mount point.
1703 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid>
1704 * and returns the ZFS vnode mounted on top of the GFS node.
1705 * This ZFS vnode is the root of the vfs for objset 'objsetid'.
1707 error = traverse(&vp, LK_SHARED | LK_RETRY);
1709 if (vp == sep->se_root)
1710 error = SET_ERROR(EINVAL);
1712 *zfsvfsp = VTOZ(vp)->z_zfsvfs;
1714 mutex_exit(&sdp->sd_lock);
1720 error = SET_ERROR(EINVAL);
1721 mutex_exit(&sdp->sd_lock);
1730 * Unmount any snapshots for the given filesystem. This is called from
1731 * zfs_umount() - if we have a ctldir, then go through and unmount all the
1735 zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr)
1737 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1739 zfsctl_snapdir_t *sdp;
1740 zfs_snapentry_t *sep, *next;
1743 ASSERT(zfsvfs->z_ctldir != NULL);
1744 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1745 NULL, 0, NULL, cr, NULL, NULL, NULL);
1750 mutex_enter(&sdp->sd_lock);
1752 sep = avl_first(&sdp->sd_snaps);
1753 while (sep != NULL) {
1754 next = AVL_NEXT(&sdp->sd_snaps, sep);
1757 * If this snapshot is not mounted, then it must
1758 * have just been unmounted by somebody else, and
1759 * will be cleaned up by zfsctl_snapdir_inactive().
1761 if (vn_ismntpt(sep->se_root)) {
1762 error = zfsctl_unmount_snap(sep, fflags, cr);
1767 * Before reinserting snapshot to the tree,
1768 * check if it was actually removed. For example
1769 * when snapshot mount point is busy, we will
1770 * have an error here, but there will be no need
1771 * to reinsert snapshot.
1773 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL)
1774 avl_insert(&sdp->sd_snaps, sep, where);
1781 mutex_exit(&sdp->sd_lock);