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));
408 zfid = (zfid_short_t *)fidp;
410 zfid->zf_len = SHORT_FID_LEN;
412 for (i = 0; i < sizeof (zfid->zf_object); i++)
413 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
415 /* .zfs znodes always have a generation number of 0 */
416 for (i = 0; i < sizeof (zfid->zf_gen); i++)
426 zfsctl_shares_fid(ap)
427 struct vop_fid_args /* {
432 vnode_t *vp = ap->a_vp;
433 fid_t *fidp = (void *)ap->a_fid;
434 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
440 if (zfsvfs->z_shares_dir == 0) {
442 return (SET_ERROR(ENOTSUP));
445 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
446 error = VOP_FID(ZTOV(dzp), fidp);
455 * .zfs inode namespace
457 * We need to generate unique inode numbers for all files and directories
458 * within the .zfs pseudo-filesystem. We use the following scheme:
463 * .zfs/snapshot/<snap> objectid(snap)
466 #define ZFSCTL_INO_SNAP(id) (id)
469 * Get root directory attributes.
473 zfsctl_root_getattr(ap)
474 struct vop_getattr_args /* {
477 struct ucred *a_cred;
480 struct vnode *vp = ap->a_vp;
481 struct vattr *vap = ap->a_vap;
482 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
483 zfsctl_node_t *zcp = vp->v_data;
486 vap->va_nodeid = ZFSCTL_INO_ROOT;
487 vap->va_nlink = vap->va_size = NROOT_ENTRIES;
488 vap->va_mtime = vap->va_ctime = zcp->zc_cmtime;
489 vap->va_birthtime = vap->va_ctime;
491 zfsctl_common_getattr(vp, vap);
498 * Special case the handling of "..".
502 zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
503 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
504 int *direntflags, pathname_t *realpnp)
506 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
510 * No extended attributes allowed under .zfs
512 if (flags & LOOKUP_XATTR)
513 return (SET_ERROR(EINVAL));
517 if (strcmp(nm, "..") == 0) {
519 err = VFS_ROOT(dvp->v_vfsp, LK_EXCLUSIVE, vpp);
522 * NB: can not use VFS_ROOT here as it would acquire
523 * the vnode lock of the parent (root) vnode while
524 * holding the child's (.zfs) lock.
528 err = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
533 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir,
534 cr, ct, direntflags, realpnp);
543 zfsctl_freebsd_root_lookup(ap)
544 struct vop_lookup_args /* {
546 struct vnode **a_vpp;
547 struct componentname *a_cnp;
550 vnode_t *dvp = ap->a_dvp;
551 vnode_t **vpp = ap->a_vpp;
552 cred_t *cr = ap->a_cnp->cn_cred;
553 int flags = ap->a_cnp->cn_flags;
554 int lkflags = ap->a_cnp->cn_lkflags;
555 int nameiop = ap->a_cnp->cn_nameiop;
556 char nm[NAME_MAX + 1];
559 if ((flags & ISLASTCN) && (nameiop == RENAME || nameiop == CREATE))
562 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm));
563 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1);
565 err = zfsctl_root_lookup(dvp, nm, vpp, NULL, 0, NULL, cr, NULL, NULL, NULL);
566 if (err == 0 && (nm[0] != '.' || nm[1] != '\0')) {
567 if (flags & ISDOTDOT) {
569 err = vn_lock(*vpp, lkflags);
574 vn_lock(dvp, LK_EXCLUSIVE | LK_RETRY);
576 err = vn_lock(*vpp, LK_EXCLUSIVE);
578 VERIFY3S(err, ==, ENOENT);
588 zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
589 caller_context_t *ct)
592 * We only care about ACL_ENABLED so that libsec can
593 * display ACL correctly and not default to POSIX draft.
595 if (cmd == _PC_ACL_ENABLED) {
596 *valp = _ACL_ACE_ENABLED;
600 return (fs_pathconf(vp, cmd, valp, cr, ct));
605 static const fs_operation_def_t zfsctl_tops_root[] = {
606 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
607 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
608 { VOPNAME_IOCTL, { .error = fs_inval } },
609 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } },
610 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
611 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
612 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } },
613 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
614 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } },
615 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } },
616 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
621 static struct vop_vector zfsctl_ops_root = {
622 .vop_default = &default_vnodeops,
623 .vop_open = zfsctl_common_open,
624 .vop_close = zfsctl_common_close,
625 .vop_ioctl = VOP_EINVAL,
626 .vop_getattr = zfsctl_root_getattr,
627 .vop_access = zfsctl_common_access,
628 .vop_readdir = gfs_vop_readdir,
629 .vop_lookup = zfsctl_freebsd_root_lookup,
630 .vop_inactive = VOP_NULL,
631 .vop_reclaim = gfs_vop_reclaim,
633 .vop_pathconf = zfsctl_pathconf,
635 .vop_fid = zfsctl_common_fid,
639 * Gets the full dataset name that corresponds to the given snapshot name
641 * zfsctl_snapshot_zname("snap1") -> "mypool/myfs@snap1"
644 zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname)
646 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
648 if (zfs_component_namecheck(name, NULL, NULL) != 0)
649 return (SET_ERROR(EILSEQ));
650 dmu_objset_name(os, zname);
651 if (strlen(zname) + 1 + strlen(name) >= len)
652 return (SET_ERROR(ENAMETOOLONG));
653 (void) strcat(zname, "@");
654 (void) strcat(zname, name);
659 zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr)
661 vnode_t *svp = sep->se_root;
664 ASSERT(vn_ismntpt(svp));
666 /* this will be dropped by dounmount() */
667 if ((error = vn_vfswlock(svp)) != 0)
672 error = dounmount(vn_mountedvfs(svp), fflags, cr);
679 * We can't use VN_RELE(), as that will try to invoke
680 * zfsctl_snapdir_inactive(), which would cause us to destroy
681 * the sd_lock mutex held by our caller.
683 ASSERT(svp->v_count == 1);
684 gfs_vop_reclaim(svp, cr, NULL);
686 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
687 kmem_free(sep, sizeof (zfs_snapentry_t));
691 return (dounmount(vn_mountedvfs(svp), fflags, curthread));
697 zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm)
702 char newpath[MAXNAMELEN];
705 ASSERT(MUTEX_HELD(&sdp->sd_lock));
708 vfsp = vn_mountedvfs(sep->se_root);
709 ASSERT(vfsp != NULL);
714 * Change the name in the AVL tree.
716 avl_remove(&sdp->sd_snaps, sep);
717 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
718 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
719 (void) strcpy(sep->se_name, nm);
720 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL);
721 avl_insert(&sdp->sd_snaps, sep, where);
724 * Change the current mountpoint info:
725 * - update the tail of the mntpoint path
726 * - update the tail of the resource path
728 pathref = vfs_getmntpoint(vfsp);
729 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
730 VERIFY((tail = strrchr(newpath, '/')) != NULL);
732 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
733 (void) strcat(newpath, nm);
734 refstr_rele(pathref);
735 vfs_setmntpoint(vfsp, newpath, 0);
737 pathref = vfs_getresource(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_setresource(vfsp, newpath, 0);
753 zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm,
754 cred_t *cr, caller_context_t *ct, int flags)
756 zfsctl_snapdir_t *sdp = sdvp->v_data;
757 zfs_snapentry_t search, *sep;
760 char from[MAXNAMELEN], to[MAXNAMELEN];
761 char real[MAXNAMELEN], fsname[MAXNAMELEN];
764 zfsvfs = sdvp->v_vfsp->vfs_data;
767 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
768 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real,
772 } else if (err != ENOTSUP) {
780 dmu_objset_name(zfsvfs->z_os, fsname);
782 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from);
784 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to);
786 err = zfs_secpolicy_rename_perms(from, to, cr);
791 * Cannot move snapshots out of the snapdir.
794 return (SET_ERROR(EINVAL));
796 if (strcmp(snm, tnm) == 0)
799 mutex_enter(&sdp->sd_lock);
801 search.se_name = (char *)snm;
802 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) {
803 mutex_exit(&sdp->sd_lock);
804 return (SET_ERROR(ENOENT));
807 err = dsl_dataset_rename_snapshot(fsname, snm, tnm, 0);
809 zfsctl_rename_snap(sdp, sep, tnm);
811 mutex_exit(&sdp->sd_lock);
820 zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
821 caller_context_t *ct, int flags)
823 zfsctl_snapdir_t *sdp = dvp->v_data;
824 zfs_snapentry_t *sep;
825 zfs_snapentry_t search;
827 char snapname[MAXNAMELEN];
828 char real[MAXNAMELEN];
831 zfsvfs = dvp->v_vfsp->vfs_data;
834 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
836 err = dmu_snapshot_realname(zfsvfs->z_os, name, real,
840 } else if (err != ENOTSUP) {
848 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname);
850 err = zfs_secpolicy_destroy_perms(snapname, cr);
854 mutex_enter(&sdp->sd_lock);
856 search.se_name = name;
857 sep = avl_find(&sdp->sd_snaps, &search, NULL);
859 avl_remove(&sdp->sd_snaps, sep);
860 err = zfsctl_unmount_snap(sep, MS_FORCE, cr);
862 avl_add(&sdp->sd_snaps, sep);
864 err = dsl_destroy_snapshot(snapname, B_FALSE);
866 err = SET_ERROR(ENOENT);
869 mutex_exit(&sdp->sd_lock);
876 * This creates a snapshot under '.zfs/snapshot'.
880 zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp,
881 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp)
883 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
884 char name[MAXNAMELEN];
886 static enum symfollow follow = NO_FOLLOW;
887 static enum uio_seg seg = UIO_SYSSPACE;
889 if (zfs_component_namecheck(dirname, NULL, NULL) != 0)
890 return (SET_ERROR(EILSEQ));
892 dmu_objset_name(zfsvfs->z_os, name);
896 err = zfs_secpolicy_snapshot_perms(name, cr);
901 err = dmu_objset_snapshot_one(name, dirname);
904 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp);
911 zfsctl_freebsd_snapdir_mkdir(ap)
912 struct vop_mkdir_args /* {
914 struct vnode **a_vpp;
915 struct componentname *a_cnp;
920 ASSERT(ap->a_cnp->cn_flags & SAVENAME);
922 return (zfsctl_snapdir_mkdir(ap->a_dvp, ap->a_cnp->cn_nameptr, NULL,
923 ap->a_vpp, ap->a_cnp->cn_cred, NULL, 0, NULL));
927 * Lookup entry point for the 'snapshot' directory. Try to open the
928 * snapshot if it exist, creating the pseudo filesystem vnode as necessary.
929 * Perform a mount of the associated dataset on top of the vnode.
933 zfsctl_snapdir_lookup(ap)
934 struct vop_lookup_args /* {
936 struct vnode **a_vpp;
937 struct componentname *a_cnp;
940 vnode_t *dvp = ap->a_dvp;
941 vnode_t **vpp = ap->a_vpp;
942 struct componentname *cnp = ap->a_cnp;
943 char nm[NAME_MAX + 1];
944 zfsctl_snapdir_t *sdp = dvp->v_data;
946 char snapname[MAXNAMELEN];
947 char real[MAXNAMELEN];
949 zfs_snapentry_t *sep, search;
950 size_t mountpoint_len;
952 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
957 * No extended attributes allowed under .zfs
959 if (flags & LOOKUP_XATTR)
960 return (SET_ERROR(EINVAL));
961 ASSERT(ap->a_cnp->cn_namelen < sizeof(nm));
962 strlcpy(nm, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1);
964 ASSERT(dvp->v_type == VDIR);
969 * If we get a recursive call, that means we got called
970 * from the domount() code while it was trying to look up the
971 * spec (which looks like a local path for zfs). We need to
972 * add some flag to domount() to tell it not to do this lookup.
974 if (MUTEX_HELD(&sdp->sd_lock))
975 return (SET_ERROR(ENOENT));
979 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
980 if (nm[0] == '.' && nm[1] == '.' && nm[2] =='\0') {
982 VERIFY0(vn_lock(*vpp, LK_EXCLUSIVE));
983 VERIFY0(vn_lock(dvp, LK_EXCLUSIVE));
989 if (flags & FIGNORECASE) {
990 boolean_t conflict = B_FALSE;
992 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real,
993 MAXNAMELEN, &conflict);
995 strlcpy(nm, real, sizeof(nm));
996 } else if (err != ENOTSUP) {
1002 (void) strlcpy(realpnp->pn_buf, nm,
1003 realpnp->pn_bufsize);
1004 if (conflict && direntflags)
1005 *direntflags = ED_CASE_CONFLICT;
1010 mutex_enter(&sdp->sd_lock);
1011 search.se_name = (char *)nm;
1012 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) {
1013 *vpp = sep->se_root;
1015 err = traverse(vpp, LK_EXCLUSIVE | LK_RETRY);
1018 } else if (*vpp == sep->se_root) {
1020 * The snapshot was unmounted behind our backs,
1021 * try to remount it.
1023 VERIFY(zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname) == 0);
1026 mutex_exit(&sdp->sd_lock);
1032 * The requested snapshot is not currently mounted, look it up.
1034 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname);
1036 mutex_exit(&sdp->sd_lock);
1039 * handle "ls *" or "?" in a graceful manner,
1040 * forcing EILSEQ to ENOENT.
1041 * Since shell ultimately passes "*" or "?" as name to lookup
1043 return (err == EILSEQ ? ENOENT : err);
1045 if (dmu_objset_hold(snapname, FTAG, &snap) != 0) {
1046 mutex_exit(&sdp->sd_lock);
1049 return (SET_ERROR(ENOENT));
1050 #else /* !illumos */
1051 /* Translate errors and add SAVENAME when needed. */
1052 if ((cnp->cn_flags & ISLASTCN) && cnp->cn_nameiop == CREATE) {
1054 cnp->cn_flags |= SAVENAME;
1056 err = SET_ERROR(ENOENT);
1060 #endif /* !illumos */
1063 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP);
1064 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
1065 (void) strcpy(sep->se_name, nm);
1066 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap));
1067 avl_insert(&sdp->sd_snaps, sep, where);
1069 dmu_objset_rele(snap, FTAG);
1071 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) +
1072 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(nm) + 1;
1073 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP);
1074 (void) snprintf(mountpoint, mountpoint_len,
1075 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s",
1076 dvp->v_vfsp->mnt_stat.f_mntonname, nm);
1077 mutex_exit(&sdp->sd_lock);
1080 * The vnode may get reclaimed between dropping sd_lock and
1081 * getting the vnode lock.
1083 err = vn_lock(*vpp, LK_EXCLUSIVE);
1087 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, snapname, 0);
1088 kmem_free(mountpoint, mountpoint_len);
1091 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>.
1093 * This is where we lie about our v_vfsp in order to
1094 * make .zfs/snapshot/<snapname> accessible over NFS
1095 * without requiring manual mounts of <snapname>.
1097 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs);
1098 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs;
1104 * If we had an error, drop our hold on the vnode and
1105 * zfsctl_snapshot_inactive() will clean up.
1120 zfsctl_shares_lookup(ap)
1121 struct vop_lookup_args /* {
1122 struct vnode *a_dvp;
1123 struct vnode **a_vpp;
1124 struct componentname *a_cnp;
1127 vnode_t *dvp = ap->a_dvp;
1128 vnode_t **vpp = ap->a_vpp;
1129 struct componentname *cnp = ap->a_cnp;
1130 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
1131 char nm[NAME_MAX + 1];
1137 ASSERT(cnp->cn_namelen < sizeof(nm));
1138 strlcpy(nm, cnp->cn_nameptr, cnp->cn_namelen + 1);
1140 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
1141 if (nm[0] == '.' && nm[1] == '.' && nm[2] =='\0') {
1143 VERIFY0(vn_lock(*vpp, LK_EXCLUSIVE));
1144 VERIFY0(vn_lock(dvp, LK_EXCLUSIVE));
1150 if (zfsvfs->z_shares_dir == 0) {
1152 return (SET_ERROR(ENOTSUP));
1154 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0)
1155 error = VOP_LOOKUP(ZTOV(dzp), vpp, cnp);
1165 zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp,
1166 offset_t *offp, offset_t *nextp, void *data, int flags)
1168 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1169 char snapname[MAXNAMELEN];
1170 uint64_t id, cookie;
1171 boolean_t case_conflict;
1177 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG);
1178 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id,
1179 &cookie, &case_conflict);
1180 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG);
1183 if (error == ENOENT) {
1190 if (flags & V_RDDIR_ENTFLAGS) {
1191 edirent_t *eodp = dp;
1193 (void) strcpy(eodp->ed_name, snapname);
1194 eodp->ed_ino = ZFSCTL_INO_SNAP(id);
1195 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0;
1197 struct dirent64 *odp = dp;
1199 (void) strcpy(odp->d_name, snapname);
1200 odp->d_ino = ZFSCTL_INO_SNAP(id);
1211 zfsctl_shares_readdir(ap)
1212 struct vop_readdir_args /* {
1215 struct ucred *a_cred;
1221 vnode_t *vp = ap->a_vp;
1222 uio_t *uiop = ap->a_uio;
1223 cred_t *cr = ap->a_cred;
1224 int *eofp = ap->a_eofflag;
1225 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1231 if (zfsvfs->z_shares_dir == 0) {
1233 return (SET_ERROR(ENOTSUP));
1235 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1236 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY);
1237 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ap->a_ncookies, ap->a_cookies);
1238 VN_URELE(ZTOV(dzp));
1241 error = SET_ERROR(ENOENT);
1249 * pvp is the '.zfs' directory (zfsctl_node_t).
1251 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t).
1253 * This function is the callback to create a GFS vnode for '.zfs/snapshot'
1254 * when a lookup is performed on .zfs for "snapshot".
1257 zfsctl_mknode_snapdir(vnode_t *pvp)
1260 zfsctl_snapdir_t *sdp;
1262 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp, pvp->v_vfsp,
1263 &zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN,
1264 zfsctl_snapdir_readdir_cb, NULL);
1266 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR;
1267 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1268 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL);
1269 avl_create(&sdp->sd_snaps, snapentry_compare,
1270 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node));
1276 zfsctl_mknode_shares(vnode_t *pvp)
1281 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp,
1282 &zfsctl_ops_shares, NULL, NULL, MAXNAMELEN,
1285 sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1293 zfsctl_shares_getattr(ap)
1294 struct vop_getattr_args /* {
1296 struct vattr *a_vap;
1297 struct ucred *a_cred;
1298 struct thread *a_td;
1301 vnode_t *vp = ap->a_vp;
1302 vattr_t *vap = ap->a_vap;
1303 cred_t *cr = ap->a_cred;
1304 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1309 if (zfsvfs->z_shares_dir == 0) {
1311 return (SET_ERROR(ENOTSUP));
1313 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1314 vn_lock(ZTOV(dzp), LK_SHARED | LK_RETRY);
1315 error = VOP_GETATTR(ZTOV(dzp), vap, cr);
1316 VN_URELE(ZTOV(dzp));
1326 zfsctl_snapdir_getattr(ap)
1327 struct vop_getattr_args /* {
1329 struct vattr *a_vap;
1330 struct ucred *a_cred;
1333 vnode_t *vp = ap->a_vp;
1334 vattr_t *vap = ap->a_vap;
1335 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1336 zfsctl_snapdir_t *sdp = vp->v_data;
1339 zfsctl_common_getattr(vp, vap);
1340 vap->va_nodeid = gfs_file_inode(vp);
1341 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2;
1342 vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os);
1343 vap->va_birthtime = vap->va_ctime;
1351 zfsctl_snapdir_reclaim(ap)
1352 struct vop_reclaim_args /* {
1354 struct thread *a_td;
1357 vnode_t *vp = ap->a_vp;
1358 zfsctl_snapdir_t *sdp = vp->v_data;
1359 zfs_snapentry_t *sep;
1361 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0);
1362 mutex_destroy(&sdp->sd_lock);
1363 avl_destroy(&sdp->sd_snaps);
1364 gfs_vop_reclaim(ap);
1370 static const fs_operation_def_t zfsctl_tops_snapdir[] = {
1371 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
1372 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
1373 { VOPNAME_IOCTL, { .error = fs_inval } },
1374 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } },
1375 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
1376 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } },
1377 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } },
1378 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } },
1379 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
1380 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } },
1381 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
1382 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } },
1383 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
1387 static const fs_operation_def_t zfsctl_tops_shares[] = {
1388 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
1389 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
1390 { VOPNAME_IOCTL, { .error = fs_inval } },
1391 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_shares_getattr } },
1392 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
1393 { VOPNAME_READDIR, { .vop_readdir = zfsctl_shares_readdir } },
1394 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_shares_lookup } },
1395 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
1396 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } },
1397 { VOPNAME_FID, { .vop_fid = zfsctl_shares_fid } },
1401 static struct vop_vector zfsctl_ops_snapdir = {
1402 .vop_default = &default_vnodeops,
1403 .vop_open = zfsctl_common_open,
1404 .vop_close = zfsctl_common_close,
1405 .vop_ioctl = VOP_EINVAL,
1406 .vop_getattr = zfsctl_snapdir_getattr,
1407 .vop_access = zfsctl_common_access,
1408 .vop_mkdir = zfsctl_freebsd_snapdir_mkdir,
1409 .vop_readdir = gfs_vop_readdir,
1410 .vop_lookup = zfsctl_snapdir_lookup,
1411 .vop_inactive = VOP_NULL,
1412 .vop_reclaim = zfsctl_snapdir_reclaim,
1413 .vop_fid = zfsctl_common_fid,
1416 static struct vop_vector zfsctl_ops_shares = {
1417 .vop_default = &default_vnodeops,
1418 .vop_open = zfsctl_common_open,
1419 .vop_close = zfsctl_common_close,
1420 .vop_ioctl = VOP_EINVAL,
1421 .vop_getattr = zfsctl_shares_getattr,
1422 .vop_access = zfsctl_common_access,
1423 .vop_readdir = zfsctl_shares_readdir,
1424 .vop_lookup = zfsctl_shares_lookup,
1425 .vop_inactive = VOP_NULL,
1426 .vop_reclaim = gfs_vop_reclaim,
1427 .vop_fid = zfsctl_shares_fid,
1432 * pvp is the GFS vnode '.zfs/snapshot'.
1434 * This creates a GFS node under '.zfs/snapshot' representing each
1435 * snapshot. This newly created GFS node is what we mount snapshot
1439 zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset)
1444 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp, pvp->v_vfsp,
1445 &zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL);
1447 zcp->zc_id = objset;
1454 zfsctl_snapshot_inactive(ap)
1455 struct vop_inactive_args /* {
1457 struct thread *a_td;
1460 vnode_t *vp = ap->a_vp;
1462 vrecycle(vp, curthread);
1467 zfsctl_snapshot_reclaim(ap)
1468 struct vop_reclaim_args /* {
1470 struct thread *a_td;
1473 vnode_t *vp = ap->a_vp;
1474 cred_t *cr = ap->a_td->td_ucred;
1475 zfsctl_snapdir_t *sdp;
1476 zfs_snapentry_t *sep, *next;
1480 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0);
1482 /* this may already have been unmounted */
1487 if (!(locked = MUTEX_HELD(&sdp->sd_lock)))
1488 mutex_enter(&sdp->sd_lock);
1490 ASSERT(!vn_ismntpt(vp));
1492 sep = avl_first(&sdp->sd_snaps);
1493 while (sep != NULL) {
1494 next = AVL_NEXT(&sdp->sd_snaps, sep);
1496 if (sep->se_root == vp) {
1497 avl_remove(&sdp->sd_snaps, sep);
1498 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
1499 kmem_free(sep, sizeof (zfs_snapentry_t));
1504 ASSERT(sep != NULL);
1507 mutex_exit(&sdp->sd_lock);
1511 * Dispose of the vnode for the snapshot mount point.
1512 * This is safe to do because once this entry has been removed
1513 * from the AVL tree, it can't be found again, so cannot become
1514 * "active". If we lookup the same name again we will end up
1515 * creating a new vnode.
1517 gfs_vop_reclaim(ap);
1523 zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap)
1525 zfsvfs_t *zfsvfs = ap->a_vp->v_vfsp->vfs_data;
1527 zfsctl_snapdir_t *sdp;
1528 zfs_snapentry_t *sep;
1531 ASSERT(zfsvfs->z_ctldir != NULL);
1532 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1533 NULL, 0, NULL, kcred, NULL, NULL, NULL);
1538 mutex_enter(&sdp->sd_lock);
1539 sep = avl_first(&sdp->sd_snaps);
1540 while (sep != NULL) {
1544 sep = AVL_NEXT(&sdp->sd_snaps, sep);
1547 mutex_exit(&sdp->sd_lock);
1552 len = strlen(sep->se_name);
1553 *ap->a_buflen -= len;
1554 bcopy(sep->se_name, ap->a_buf + *ap->a_buflen, len);
1555 mutex_exit(&sdp->sd_lock);
1565 * These VP's should never see the light of day. They should always
1568 static struct vop_vector zfsctl_ops_snapshot = {
1569 .vop_default = &default_vnodeops,
1570 .vop_inactive = zfsctl_snapshot_inactive,
1571 .vop_reclaim = zfsctl_snapshot_reclaim,
1572 .vop_vptocnp = zfsctl_snapshot_vptocnp,
1576 zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp)
1578 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1580 zfsctl_snapdir_t *sdp;
1582 zfs_snapentry_t *sep;
1585 ASSERT(zfsvfs->z_ctldir != NULL);
1586 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1587 NULL, 0, NULL, kcred, NULL, NULL, NULL);
1592 mutex_enter(&sdp->sd_lock);
1593 sep = avl_first(&sdp->sd_snaps);
1594 while (sep != NULL) {
1597 if (zcp->zc_id == objsetid)
1600 sep = AVL_NEXT(&sdp->sd_snaps, sep);
1606 * Return the mounted root rather than the covered mount point.
1607 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid>
1608 * and returns the ZFS vnode mounted on top of the GFS node.
1609 * This ZFS vnode is the root of the vfs for objset 'objsetid'.
1611 error = traverse(&vp, LK_SHARED | LK_RETRY);
1613 if (vp == sep->se_root) {
1614 VN_RELE(vp); /* release covered vp */
1615 error = SET_ERROR(EINVAL);
1617 *zfsvfsp = VTOZ(vp)->z_zfsvfs;
1618 VN_URELE(vp); /* put snapshot's root vp */
1621 mutex_exit(&sdp->sd_lock);
1623 error = SET_ERROR(EINVAL);
1624 mutex_exit(&sdp->sd_lock);
1633 * Unmount any snapshots for the given filesystem. This is called from
1634 * zfs_umount() - if we have a ctldir, then go through and unmount all the
1638 zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr)
1640 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1642 zfsctl_snapdir_t *sdp;
1643 zfs_snapentry_t *sep, *next;
1646 ASSERT(zfsvfs->z_ctldir != NULL);
1647 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1648 NULL, 0, NULL, cr, NULL, NULL, NULL);
1653 mutex_enter(&sdp->sd_lock);
1655 sep = avl_first(&sdp->sd_snaps);
1656 while (sep != NULL) {
1657 next = AVL_NEXT(&sdp->sd_snaps, sep);
1660 * If this snapshot is not mounted, then it must
1661 * have just been unmounted by somebody else, and
1662 * will be cleaned up by zfsctl_snapdir_inactive().
1664 if (vn_ismntpt(sep->se_root)) {
1665 error = zfsctl_unmount_snap(sep, fflags, cr);
1670 * Before reinserting snapshot to the tree,
1671 * check if it was actually removed. For example
1672 * when snapshot mount point is busy, we will
1673 * have an error here, but there will be no need
1674 * to reinsert snapshot.
1676 if (avl_find(&sdp->sd_snaps, sep, &where) == NULL)
1677 avl_insert(&sdp->sd_snaps, sep, where);
1684 mutex_exit(&sdp->sd_lock);