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.
26 * ZFS control directory (a.k.a. ".zfs")
28 * This directory provides a common location for all ZFS meta-objects.
29 * Currently, this is only the 'snapshot' directory, but this may expand in the
30 * future. The elements are built using the GFS primitives, as the hierarchy
31 * does not actually exist on disk.
33 * For 'snapshot', we don't want to have all snapshots always mounted, because
34 * this would take up a huge amount of space in /etc/mnttab. We have three
37 * ctldir ------> snapshotdir -------> snapshot
43 * The 'snapshot' node contains just enough information to lookup '..' and act
44 * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we
45 * perform an automount of the underlying filesystem and return the
46 * corresponding vnode.
48 * All mounts are handled automatically by the kernel, but unmounts are
49 * (currently) handled from user land. The main reason is that there is no
50 * reliable way to auto-unmount the filesystem when it's "no longer in use".
51 * When the user unmounts a filesystem, we call zfsctl_unmount(), which
52 * unmounts any snapshots within the snapshot directory.
54 * The '.zfs', '.zfs/snapshot', and all directories created under
55 * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') are all GFS nodes and
56 * share the same vfs_t as the head filesystem (what '.zfs' lives under).
58 * File systems mounted ontop of the GFS nodes '.zfs/snapshot/<snapname>'
59 * (ie: snapshots) are ZFS nodes and have their own unique vfs_t.
60 * However, vnodes within these mounted on file systems have their v_vfsp
61 * fields set to the head filesystem to make NFS happy (see
62 * zfsctl_snapdir_lookup()). We VFS_HOLD the head filesystem's vfs_t
63 * so that it cannot be freed until all snapshots have been unmounted.
68 #include <fs/fs_subr.h>
69 #include <sys/zfs_ctldir.h>
70 #include <sys/zfs_ioctl.h>
71 #include <sys/zfs_vfsops.h>
72 #include <sys/vfs_opreg.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);
115 vnodeops_t *zfsctl_ops_root;
116 vnodeops_t *zfsctl_ops_snapdir;
117 vnodeops_t *zfsctl_ops_snapshot;
118 vnodeops_t *zfsctl_ops_shares;
119 vnodeops_t *zfsctl_ops_shares_dir;
121 static const fs_operation_def_t zfsctl_tops_root[];
122 static const fs_operation_def_t zfsctl_tops_snapdir[];
123 static const fs_operation_def_t zfsctl_tops_snapshot[];
124 static const fs_operation_def_t zfsctl_tops_shares[];
126 static vnode_t *zfsctl_mknode_snapdir(vnode_t *);
127 static vnode_t *zfsctl_mknode_shares(vnode_t *);
128 static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset);
129 static int zfsctl_unmount_snap(zfs_snapentry_t *, int, cred_t *);
131 static gfs_opsvec_t zfsctl_opsvec[] = {
132 { ".zfs", zfsctl_tops_root, &zfsctl_ops_root },
133 { ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir },
134 { ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot },
135 { ".zfs/shares", zfsctl_tops_shares, &zfsctl_ops_shares_dir },
136 { ".zfs/shares/vnode", zfsctl_tops_shares, &zfsctl_ops_shares },
141 * Root directory elements. We only have two entries
142 * snapshot and shares.
144 static gfs_dirent_t zfsctl_root_entries[] = {
145 { "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE },
146 { "shares", zfsctl_mknode_shares, GFS_CACHE_VNODE },
150 /* include . and .. in the calculation */
151 #define NROOT_ENTRIES ((sizeof (zfsctl_root_entries) / \
152 sizeof (gfs_dirent_t)) + 1)
156 * Initialize the various GFS pieces we'll need to create and manipulate .zfs
157 * directories. This is called from the ZFS init routine, and initializes the
158 * vnode ops vectors that we'll be using.
163 VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0);
170 * Remove vfsctl vnode ops
173 vn_freevnodeops(zfsctl_ops_root);
174 if (zfsctl_ops_snapdir)
175 vn_freevnodeops(zfsctl_ops_snapdir);
176 if (zfsctl_ops_snapshot)
177 vn_freevnodeops(zfsctl_ops_snapshot);
178 if (zfsctl_ops_shares)
179 vn_freevnodeops(zfsctl_ops_shares);
180 if (zfsctl_ops_shares_dir)
181 vn_freevnodeops(zfsctl_ops_shares_dir);
183 zfsctl_ops_root = NULL;
184 zfsctl_ops_snapdir = NULL;
185 zfsctl_ops_snapshot = NULL;
186 zfsctl_ops_shares = NULL;
187 zfsctl_ops_shares_dir = NULL;
191 zfsctl_is_node(vnode_t *vp)
193 return (vn_matchops(vp, zfsctl_ops_root) ||
194 vn_matchops(vp, zfsctl_ops_snapdir) ||
195 vn_matchops(vp, zfsctl_ops_snapshot) ||
196 vn_matchops(vp, zfsctl_ops_shares) ||
197 vn_matchops(vp, zfsctl_ops_shares_dir));
202 * Return the inode number associated with the 'snapshot' or
203 * 'shares' directory.
207 zfsctl_root_inode_cb(vnode_t *vp, int index)
209 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
214 return (ZFSCTL_INO_SNAPDIR);
216 return (zfsvfs->z_shares_dir);
220 * Create the '.zfs' directory. This directory is cached as part of the VFS
221 * structure. This results in a hold on the vfs_t. The code in zfs_umount()
222 * therefore checks against a vfs_count of 2 instead of 1. This reference
223 * is removed when the ctldir is destroyed in the unmount.
226 zfsctl_create(zfsvfs_t *zfsvfs)
232 ASSERT(zfsvfs->z_ctldir == NULL);
234 vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs,
235 zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries,
236 zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL);
238 zcp->zc_id = ZFSCTL_INO_ROOT;
240 VERIFY(VFS_ROOT(zfsvfs->z_vfs, &rvp) == 0);
241 VERIFY(0 == sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
242 &crtime, sizeof (crtime)));
243 ZFS_TIME_DECODE(&zcp->zc_cmtime, crtime);
247 * We're only faking the fact that we have a root of a filesystem for
248 * the sake of the GFS interfaces. Undo the flag manipulation it did
251 vp->v_flag &= ~(VROOT | VNOCACHE | VNOMAP | VNOSWAP | VNOMOUNT);
253 zfsvfs->z_ctldir = vp;
257 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted.
258 * There might still be more references if we were force unmounted, but only
259 * new zfs_inactive() calls can occur and they don't reference .zfs
262 zfsctl_destroy(zfsvfs_t *zfsvfs)
264 VN_RELE(zfsvfs->z_ctldir);
265 zfsvfs->z_ctldir = NULL;
269 * Given a root znode, retrieve the associated .zfs directory.
270 * Add a hold to the vnode and return it.
273 zfsctl_root(znode_t *zp)
275 ASSERT(zfs_has_ctldir(zp));
276 VN_HOLD(zp->z_zfsvfs->z_ctldir);
277 return (zp->z_zfsvfs->z_ctldir);
281 * Common open routine. Disallow any write access.
285 zfsctl_common_open(vnode_t **vpp, int flags, cred_t *cr, caller_context_t *ct)
294 * Common close routine. Nothing to do here.
298 zfsctl_common_close(vnode_t *vpp, int flags, int count, offset_t off,
299 cred_t *cr, caller_context_t *ct)
305 * Common access routine. Disallow writes.
309 zfsctl_common_access(vnode_t *vp, int mode, int flags, cred_t *cr,
310 caller_context_t *ct)
312 if (flags & V_ACE_MASK) {
313 if (mode & ACE_ALL_WRITE_PERMS)
324 * Common getattr function. Fill in basic information.
327 zfsctl_common_getattr(vnode_t *vp, vattr_t *vap)
335 * We are a purely virtual object, so we have no
336 * blocksize or allocated blocks.
341 vap->va_fsid = vp->v_vfsp->vfs_dev;
342 vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP |
346 * We live in the now (for atime).
354 zfsctl_common_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
356 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
357 zfsctl_node_t *zcp = vp->v_data;
358 uint64_t object = zcp->zc_id;
364 if (fidp->fid_len < SHORT_FID_LEN) {
365 fidp->fid_len = SHORT_FID_LEN;
370 zfid = (zfid_short_t *)fidp;
372 zfid->zf_len = SHORT_FID_LEN;
374 for (i = 0; i < sizeof (zfid->zf_object); i++)
375 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
377 /* .zfs znodes always have a generation number of 0 */
378 for (i = 0; i < sizeof (zfid->zf_gen); i++)
388 zfsctl_shares_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
390 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
396 if (zfsvfs->z_shares_dir == 0) {
401 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
402 error = VOP_FID(ZTOV(dzp), fidp, ct);
410 * .zfs inode namespace
412 * We need to generate unique inode numbers for all files and directories
413 * within the .zfs pseudo-filesystem. We use the following scheme:
418 * .zfs/snapshot/<snap> objectid(snap)
421 #define ZFSCTL_INO_SNAP(id) (id)
424 * Get root directory attributes.
428 zfsctl_root_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
429 caller_context_t *ct)
431 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
432 zfsctl_node_t *zcp = vp->v_data;
435 vap->va_nodeid = ZFSCTL_INO_ROOT;
436 vap->va_nlink = vap->va_size = NROOT_ENTRIES;
437 vap->va_mtime = vap->va_ctime = zcp->zc_cmtime;
439 zfsctl_common_getattr(vp, vap);
446 * Special case the handling of "..".
450 zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
451 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
452 int *direntflags, pathname_t *realpnp)
454 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
458 * No extended attributes allowed under .zfs
460 if (flags & LOOKUP_XATTR)
465 if (strcmp(nm, "..") == 0) {
466 err = VFS_ROOT(dvp->v_vfsp, vpp);
468 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir,
469 cr, ct, direntflags, realpnp);
478 zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
479 caller_context_t *ct)
482 * We only care about ACL_ENABLED so that libsec can
483 * display ACL correctly and not default to POSIX draft.
485 if (cmd == _PC_ACL_ENABLED) {
486 *valp = _ACL_ACE_ENABLED;
490 return (fs_pathconf(vp, cmd, valp, cr, ct));
493 static const fs_operation_def_t zfsctl_tops_root[] = {
494 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
495 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
496 { VOPNAME_IOCTL, { .error = fs_inval } },
497 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } },
498 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
499 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
500 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } },
501 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
502 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } },
503 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } },
504 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
509 zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname)
511 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
513 if (snapshot_namecheck(name, NULL, NULL) != 0)
515 dmu_objset_name(os, zname);
516 if (strlen(zname) + 1 + strlen(name) >= len)
517 return (ENAMETOOLONG);
518 (void) strcat(zname, "@");
519 (void) strcat(zname, name);
524 zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr)
526 vnode_t *svp = sep->se_root;
529 ASSERT(vn_ismntpt(svp));
531 /* this will be dropped by dounmount() */
532 if ((error = vn_vfswlock(svp)) != 0)
536 error = dounmount(vn_mountedvfs(svp), fflags, cr);
543 * We can't use VN_RELE(), as that will try to invoke
544 * zfsctl_snapdir_inactive(), which would cause us to destroy
545 * the sd_lock mutex held by our caller.
547 ASSERT(svp->v_count == 1);
548 gfs_vop_inactive(svp, cr, NULL);
550 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
551 kmem_free(sep, sizeof (zfs_snapentry_t));
557 zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm)
562 char newpath[MAXNAMELEN];
565 ASSERT(MUTEX_HELD(&sdp->sd_lock));
568 vfsp = vn_mountedvfs(sep->se_root);
569 ASSERT(vfsp != NULL);
574 * Change the name in the AVL tree.
576 avl_remove(&sdp->sd_snaps, sep);
577 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
578 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
579 (void) strcpy(sep->se_name, nm);
580 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL);
581 avl_insert(&sdp->sd_snaps, sep, where);
584 * Change the current mountpoint info:
585 * - update the tail of the mntpoint path
586 * - update the tail of the resource path
588 pathref = vfs_getmntpoint(vfsp);
589 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
590 VERIFY((tail = strrchr(newpath, '/')) != NULL);
592 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
593 (void) strcat(newpath, nm);
594 refstr_rele(pathref);
595 vfs_setmntpoint(vfsp, newpath, 0);
597 pathref = vfs_getresource(vfsp);
598 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
599 VERIFY((tail = strrchr(newpath, '@')) != NULL);
601 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
602 (void) strcat(newpath, nm);
603 refstr_rele(pathref);
604 vfs_setresource(vfsp, newpath, 0);
611 zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm,
612 cred_t *cr, caller_context_t *ct, int flags)
614 zfsctl_snapdir_t *sdp = sdvp->v_data;
615 zfs_snapentry_t search, *sep;
618 char from[MAXNAMELEN], to[MAXNAMELEN];
619 char real[MAXNAMELEN];
622 zfsvfs = sdvp->v_vfsp->vfs_data;
625 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
626 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real,
630 } else if (err != ENOTSUP) {
638 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from);
640 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to);
642 err = zfs_secpolicy_rename_perms(from, to, cr);
647 * Cannot move snapshots out of the snapdir.
652 if (strcmp(snm, tnm) == 0)
655 mutex_enter(&sdp->sd_lock);
657 search.se_name = (char *)snm;
658 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) {
659 mutex_exit(&sdp->sd_lock);
663 err = dmu_objset_rename(from, to, B_FALSE);
665 zfsctl_rename_snap(sdp, sep, tnm);
667 mutex_exit(&sdp->sd_lock);
674 zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
675 caller_context_t *ct, int flags)
677 zfsctl_snapdir_t *sdp = dvp->v_data;
678 zfs_snapentry_t *sep;
679 zfs_snapentry_t search;
681 char snapname[MAXNAMELEN];
682 char real[MAXNAMELEN];
685 zfsvfs = dvp->v_vfsp->vfs_data;
688 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
690 err = dmu_snapshot_realname(zfsvfs->z_os, name, real,
694 } else if (err != ENOTSUP) {
702 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname);
704 err = zfs_secpolicy_destroy_perms(snapname, cr);
708 mutex_enter(&sdp->sd_lock);
710 search.se_name = name;
711 sep = avl_find(&sdp->sd_snaps, &search, NULL);
713 avl_remove(&sdp->sd_snaps, sep);
714 err = zfsctl_unmount_snap(sep, MS_FORCE, cr);
716 avl_add(&sdp->sd_snaps, sep);
718 err = dmu_objset_destroy(snapname, B_FALSE);
723 mutex_exit(&sdp->sd_lock);
729 * This creates a snapshot under '.zfs/snapshot'.
733 zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp,
734 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp)
736 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
737 char name[MAXNAMELEN];
739 static enum symfollow follow = NO_FOLLOW;
740 static enum uio_seg seg = UIO_SYSSPACE;
742 if (snapshot_namecheck(dirname, NULL, NULL) != 0)
745 dmu_objset_name(zfsvfs->z_os, name);
749 err = zfs_secpolicy_snapshot_perms(name, cr);
754 err = dmu_objset_snapshot(name, dirname, NULL, NULL,
755 B_FALSE, B_FALSE, -1);
758 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp);
765 * Lookup entry point for the 'snapshot' directory. Try to open the
766 * snapshot if it exist, creating the pseudo filesystem vnode as necessary.
767 * Perform a mount of the associated dataset on top of the vnode.
771 zfsctl_snapdir_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
772 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
773 int *direntflags, pathname_t *realpnp)
775 zfsctl_snapdir_t *sdp = dvp->v_data;
777 char snapname[MAXNAMELEN];
778 char real[MAXNAMELEN];
780 zfs_snapentry_t *sep, search;
783 size_t mountpoint_len;
785 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
789 * No extended attributes allowed under .zfs
791 if (flags & LOOKUP_XATTR)
794 ASSERT(dvp->v_type == VDIR);
797 * If we get a recursive call, that means we got called
798 * from the domount() code while it was trying to look up the
799 * spec (which looks like a local path for zfs). We need to
800 * add some flag to domount() to tell it not to do this lookup.
802 if (MUTEX_HELD(&sdp->sd_lock))
807 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
812 if (flags & FIGNORECASE) {
813 boolean_t conflict = B_FALSE;
815 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real,
816 MAXNAMELEN, &conflict);
819 } else if (err != ENOTSUP) {
824 (void) strlcpy(realpnp->pn_buf, nm,
825 realpnp->pn_bufsize);
826 if (conflict && direntflags)
827 *direntflags = ED_CASE_CONFLICT;
830 mutex_enter(&sdp->sd_lock);
831 search.se_name = (char *)nm;
832 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) {
839 } else if (*vpp == sep->se_root) {
841 * The snapshot was unmounted behind our backs,
847 * VROOT was set during the traverse call. We need
848 * to clear it since we're pretending to be part
849 * of our parent's vfs.
851 (*vpp)->v_flag &= ~VROOT;
853 mutex_exit(&sdp->sd_lock);
859 * The requested snapshot is not currently mounted, look it up.
861 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname);
863 mutex_exit(&sdp->sd_lock);
866 * handle "ls *" or "?" in a graceful manner,
867 * forcing EILSEQ to ENOENT.
868 * Since shell ultimately passes "*" or "?" as name to lookup
870 return (err == EILSEQ ? ENOENT : err);
872 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) {
873 mutex_exit(&sdp->sd_lock);
878 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP);
879 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
880 (void) strcpy(sep->se_name, nm);
881 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap));
882 avl_insert(&sdp->sd_snaps, sep, where);
884 dmu_objset_rele(snap, FTAG);
886 mountpoint_len = strlen(refstr_value(dvp->v_vfsp->vfs_mntpt)) +
887 strlen("/.zfs/snapshot/") + strlen(nm) + 1;
888 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP);
889 (void) snprintf(mountpoint, mountpoint_len, "%s/.zfs/snapshot/%s",
890 refstr_value(dvp->v_vfsp->vfs_mntpt), nm);
892 margs.spec = snapname;
893 margs.dir = mountpoint;
894 margs.flags = MS_SYSSPACE | MS_NOMNTTAB;
895 margs.fstype = "zfs";
896 margs.dataptr = NULL;
901 err = domount("zfs", &margs, *vpp, kcred, &vfsp);
902 kmem_free(mountpoint, mountpoint_len);
906 * Return the mounted root rather than the covered mount point.
907 * Takes the GFS vnode at .zfs/snapshot/<snapname> and returns
908 * the ZFS vnode mounted on top of the GFS node. This ZFS
909 * vnode is the root of the newly created vfsp.
917 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>.
919 * This is where we lie about our v_vfsp in order to
920 * make .zfs/snapshot/<snapname> accessible over NFS
921 * without requiring manual mounts of <snapname>.
923 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs);
924 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs;
925 (*vpp)->v_vfsp = zfsvfs->z_vfs;
926 (*vpp)->v_flag &= ~VROOT;
928 mutex_exit(&sdp->sd_lock);
932 * If we had an error, drop our hold on the vnode and
933 * zfsctl_snapshot_inactive() will clean up.
944 zfsctl_shares_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
945 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
946 int *direntflags, pathname_t *realpnp)
948 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
954 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
959 if (zfsvfs->z_shares_dir == 0) {
963 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0)
964 error = VOP_LOOKUP(ZTOV(dzp), nm, vpp, pnp,
965 flags, rdir, cr, ct, direntflags, realpnp);
975 zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp,
976 offset_t *offp, offset_t *nextp, void *data, int flags)
978 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
979 char snapname[MAXNAMELEN];
981 boolean_t case_conflict;
987 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id,
988 &cookie, &case_conflict);
991 if (error == ENOENT) {
998 if (flags & V_RDDIR_ENTFLAGS) {
999 edirent_t *eodp = dp;
1001 (void) strcpy(eodp->ed_name, snapname);
1002 eodp->ed_ino = ZFSCTL_INO_SNAP(id);
1003 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0;
1005 struct dirent64 *odp = dp;
1007 (void) strcpy(odp->d_name, snapname);
1008 odp->d_ino = ZFSCTL_INO_SNAP(id);
1019 zfsctl_shares_readdir(vnode_t *vp, uio_t *uiop, cred_t *cr, int *eofp,
1020 caller_context_t *ct, int flags)
1022 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1028 if (zfsvfs->z_shares_dir == 0) {
1032 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1033 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ct, flags);
1045 * pvp is the '.zfs' directory (zfsctl_node_t).
1046 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t).
1048 * This function is the callback to create a GFS vnode for '.zfs/snapshot'
1049 * when a lookup is performed on .zfs for "snapshot".
1052 zfsctl_mknode_snapdir(vnode_t *pvp)
1055 zfsctl_snapdir_t *sdp;
1057 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp,
1058 zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN,
1059 zfsctl_snapdir_readdir_cb, NULL);
1061 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR;
1062 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1063 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL);
1064 avl_create(&sdp->sd_snaps, snapentry_compare,
1065 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node));
1070 zfsctl_mknode_shares(vnode_t *pvp)
1075 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp,
1076 zfsctl_ops_shares, NULL, NULL, MAXNAMELEN,
1079 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1086 zfsctl_shares_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1087 caller_context_t *ct)
1089 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1094 if (zfsvfs->z_shares_dir == 0) {
1098 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1099 error = VOP_GETATTR(ZTOV(dzp), vap, flags, cr, ct);
1110 zfsctl_snapdir_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1111 caller_context_t *ct)
1113 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1114 zfsctl_snapdir_t *sdp = vp->v_data;
1117 zfsctl_common_getattr(vp, vap);
1118 vap->va_nodeid = gfs_file_inode(vp);
1119 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2;
1120 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os);
1128 zfsctl_snapdir_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1130 zfsctl_snapdir_t *sdp = vp->v_data;
1133 private = gfs_dir_inactive(vp);
1134 if (private != NULL) {
1135 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0);
1136 mutex_destroy(&sdp->sd_lock);
1137 avl_destroy(&sdp->sd_snaps);
1138 kmem_free(private, sizeof (zfsctl_snapdir_t));
1142 static const fs_operation_def_t zfsctl_tops_snapdir[] = {
1143 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
1144 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
1145 { VOPNAME_IOCTL, { .error = fs_inval } },
1146 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } },
1147 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
1148 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } },
1149 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } },
1150 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } },
1151 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
1152 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } },
1153 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
1154 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } },
1155 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
1159 static const fs_operation_def_t zfsctl_tops_shares[] = {
1160 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
1161 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
1162 { VOPNAME_IOCTL, { .error = fs_inval } },
1163 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } },
1164 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
1165 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } },
1166 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } },
1167 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
1168 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } },
1169 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } },
1174 * pvp is the GFS vnode '.zfs/snapshot'.
1176 * This creates a GFS node under '.zfs/snapshot' representing each
1177 * snapshot. This newly created GFS node is what we mount snapshot
1181 zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset)
1186 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp,
1187 zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL);
1189 zcp->zc_id = objset;
1195 zfsctl_snapshot_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1197 zfsctl_snapdir_t *sdp;
1198 zfs_snapentry_t *sep, *next;
1201 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0);
1204 mutex_enter(&sdp->sd_lock);
1206 if (vp->v_count > 1) {
1207 mutex_exit(&sdp->sd_lock);
1210 ASSERT(!vn_ismntpt(vp));
1212 sep = avl_first(&sdp->sd_snaps);
1213 while (sep != NULL) {
1214 next = AVL_NEXT(&sdp->sd_snaps, sep);
1216 if (sep->se_root == vp) {
1217 avl_remove(&sdp->sd_snaps, sep);
1218 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
1219 kmem_free(sep, sizeof (zfs_snapentry_t));
1224 ASSERT(sep != NULL);
1226 mutex_exit(&sdp->sd_lock);
1230 * Dispose of the vnode for the snapshot mount point.
1231 * This is safe to do because once this entry has been removed
1232 * from the AVL tree, it can't be found again, so cannot become
1233 * "active". If we lookup the same name again we will end up
1234 * creating a new vnode.
1236 gfs_vop_inactive(vp, cr, ct);
1241 * These VP's should never see the light of day. They should always
1244 static const fs_operation_def_t zfsctl_tops_snapshot[] = {
1245 VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapshot_inactive },
1250 zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp)
1252 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1254 zfsctl_snapdir_t *sdp;
1256 zfs_snapentry_t *sep;
1259 ASSERT(zfsvfs->z_ctldir != NULL);
1260 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1261 NULL, 0, NULL, kcred, NULL, NULL, NULL);
1266 mutex_enter(&sdp->sd_lock);
1267 sep = avl_first(&sdp->sd_snaps);
1268 while (sep != NULL) {
1271 if (zcp->zc_id == objsetid)
1274 sep = AVL_NEXT(&sdp->sd_snaps, sep);
1280 * Return the mounted root rather than the covered mount point.
1281 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid>
1282 * and returns the ZFS vnode mounted on top of the GFS node.
1283 * This ZFS vnode is the root of the vfs for objset 'objsetid'.
1285 error = traverse(&vp);
1287 if (vp == sep->se_root)
1290 *zfsvfsp = VTOZ(vp)->z_zfsvfs;
1292 mutex_exit(&sdp->sd_lock);
1296 mutex_exit(&sdp->sd_lock);
1305 * Unmount any snapshots for the given filesystem. This is called from
1306 * zfs_umount() - if we have a ctldir, then go through and unmount all the
1310 zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr)
1312 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1314 zfsctl_snapdir_t *sdp;
1315 zfs_snapentry_t *sep, *next;
1318 ASSERT(zfsvfs->z_ctldir != NULL);
1319 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1320 NULL, 0, NULL, cr, NULL, NULL, NULL);
1325 mutex_enter(&sdp->sd_lock);
1327 sep = avl_first(&sdp->sd_snaps);
1328 while (sep != NULL) {
1329 next = AVL_NEXT(&sdp->sd_snaps, sep);
1332 * If this snapshot is not mounted, then it must
1333 * have just been unmounted by somebody else, and
1334 * will be cleaned up by zfsctl_snapdir_inactive().
1336 if (vn_ismntpt(sep->se_root)) {
1337 avl_remove(&sdp->sd_snaps, sep);
1338 error = zfsctl_unmount_snap(sep, fflags, cr);
1340 avl_add(&sdp->sd_snaps, sep);
1347 mutex_exit(&sdp->sd_lock);
1352 #endif /* HAVE_ZPL */