4 * The contents of this file are subject to the terms of the
5 * Common Development and Distribution License (the "License").
6 * You may not use this file except in compliance with the License.
8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 * or http://www.opensolaris.org/os/licensing.
10 * See the License for the specific language governing permissions
11 * and limitations under the License.
13 * When distributing Covered Code, include this CDDL HEADER in each
14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 * If applicable, add the following below this CDDL HEADER, with the
16 * fields enclosed by brackets "[]" replaced with your own identifying
17 * information: Portions Copyright [yyyy] [name of copyright owner]
22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved.
23 * Use is subject to license terms.
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 <fs/fs_subr.h>
68 #include <sys/zfs_ctldir.h>
69 #include <sys/zfs_ioctl.h>
70 #include <sys/zfs_vfsops.h>
71 #include <sys/vfs_opreg.h>
75 #include <sys/dsl_deleg.h>
76 #include <sys/mount.h>
77 #include <sys/sunddi.h>
79 #include "zfs_namecheck.h"
81 typedef struct zfsctl_node {
82 gfs_dir_t zc_gfs_private;
84 timestruc_t zc_cmtime; /* ctime and mtime, always the same */
87 typedef struct zfsctl_snapdir {
88 zfsctl_node_t sd_node;
100 snapentry_compare(const void *a, const void *b)
102 const zfs_snapentry_t *sa = a;
103 const zfs_snapentry_t *sb = b;
104 int ret = strcmp(sa->se_name, sb->se_name);
114 vnodeops_t *zfsctl_ops_root;
115 vnodeops_t *zfsctl_ops_snapdir;
116 vnodeops_t *zfsctl_ops_snapshot;
118 static const fs_operation_def_t zfsctl_tops_root[];
119 static const fs_operation_def_t zfsctl_tops_snapdir[];
120 static const fs_operation_def_t zfsctl_tops_snapshot[];
122 static vnode_t *zfsctl_mknode_snapdir(vnode_t *);
123 static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset);
124 static int zfsctl_unmount_snap(zfs_snapentry_t *, int, cred_t *);
126 static gfs_opsvec_t zfsctl_opsvec[] = {
127 { ".zfs", zfsctl_tops_root, &zfsctl_ops_root },
128 { ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir },
129 { ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot },
134 * Root directory elements. We have only a single static entry, 'snapshot'.
136 static gfs_dirent_t zfsctl_root_entries[] = {
137 { "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE },
141 /* include . and .. in the calculation */
142 #define NROOT_ENTRIES ((sizeof (zfsctl_root_entries) / \
143 sizeof (gfs_dirent_t)) + 1)
147 * Initialize the various GFS pieces we'll need to create and manipulate .zfs
148 * directories. This is called from the ZFS init routine, and initializes the
149 * vnode ops vectors that we'll be using.
154 VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0);
161 * Remove vfsctl vnode ops
164 vn_freevnodeops(zfsctl_ops_root);
165 if (zfsctl_ops_snapdir)
166 vn_freevnodeops(zfsctl_ops_snapdir);
167 if (zfsctl_ops_snapshot)
168 vn_freevnodeops(zfsctl_ops_snapshot);
170 zfsctl_ops_root = NULL;
171 zfsctl_ops_snapdir = NULL;
172 zfsctl_ops_snapshot = NULL;
176 * Return the inode number associated with the 'snapshot' directory.
180 zfsctl_root_inode_cb(vnode_t *vp, int index)
183 return (ZFSCTL_INO_SNAPDIR);
187 * Create the '.zfs' directory. This directory is cached as part of the VFS
188 * structure. This results in a hold on the vfs_t. The code in zfs_umount()
189 * therefore checks against a vfs_count of 2 instead of 1. This reference
190 * is removed when the ctldir is destroyed in the unmount.
193 zfsctl_create(zfsvfs_t *zfsvfs)
198 ASSERT(zfsvfs->z_ctldir == NULL);
200 vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs,
201 zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries,
202 zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL);
204 zcp->zc_id = ZFSCTL_INO_ROOT;
206 VERIFY(VFS_ROOT(zfsvfs->z_vfs, &rvp) == 0);
207 ZFS_TIME_DECODE(&zcp->zc_cmtime, VTOZ(rvp)->z_phys->zp_crtime);
211 * We're only faking the fact that we have a root of a filesystem for
212 * the sake of the GFS interfaces. Undo the flag manipulation it did
215 vp->v_flag &= ~(VROOT | VNOCACHE | VNOMAP | VNOSWAP | VNOMOUNT);
217 zfsvfs->z_ctldir = vp;
221 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted.
222 * There might still be more references if we were force unmounted, but only
223 * new zfs_inactive() calls can occur and they don't reference .zfs
226 zfsctl_destroy(zfsvfs_t *zfsvfs)
228 VN_RELE(zfsvfs->z_ctldir);
229 zfsvfs->z_ctldir = NULL;
233 * Given a root znode, retrieve the associated .zfs directory.
234 * Add a hold to the vnode and return it.
237 zfsctl_root(znode_t *zp)
239 ASSERT(zfs_has_ctldir(zp));
240 VN_HOLD(zp->z_zfsvfs->z_ctldir);
241 return (zp->z_zfsvfs->z_ctldir);
245 * Common open routine. Disallow any write access.
249 zfsctl_common_open(vnode_t **vpp, int flags, cred_t *cr, caller_context_t *ct)
258 * Common close routine. Nothing to do here.
262 zfsctl_common_close(vnode_t *vpp, int flags, int count, offset_t off,
263 cred_t *cr, caller_context_t *ct)
269 * Common access routine. Disallow writes.
273 zfsctl_common_access(vnode_t *vp, int mode, int flags, cred_t *cr,
274 caller_context_t *ct)
276 if (flags & V_ACE_MASK) {
277 if (mode & ACE_ALL_WRITE_PERMS)
288 * Common getattr function. Fill in basic information.
291 zfsctl_common_getattr(vnode_t *vp, vattr_t *vap)
293 zfsctl_node_t *zcp = vp->v_data;
300 * We are a purly virtual object, so we have no
301 * blocksize or allocated blocks.
306 vap->va_fsid = vp->v_vfsp->vfs_dev;
307 vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP |
311 * We live in the now (for atime).
315 vap->va_mtime = vap->va_ctime = zcp->zc_cmtime;
320 zfsctl_common_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
322 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
323 zfsctl_node_t *zcp = vp->v_data;
324 uint64_t object = zcp->zc_id;
330 if (fidp->fid_len < SHORT_FID_LEN) {
331 fidp->fid_len = SHORT_FID_LEN;
336 zfid = (zfid_short_t *)fidp;
338 zfid->zf_len = SHORT_FID_LEN;
340 for (i = 0; i < sizeof (zfid->zf_object); i++)
341 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
343 /* .zfs znodes always have a generation number of 0 */
344 for (i = 0; i < sizeof (zfid->zf_gen); i++)
352 * .zfs inode namespace
354 * We need to generate unique inode numbers for all files and directories
355 * within the .zfs pseudo-filesystem. We use the following scheme:
360 * .zfs/snapshot/<snap> objectid(snap)
363 #define ZFSCTL_INO_SNAP(id) (id)
366 * Get root directory attributes.
370 zfsctl_root_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
371 caller_context_t *ct)
373 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
376 vap->va_nodeid = ZFSCTL_INO_ROOT;
377 vap->va_nlink = vap->va_size = NROOT_ENTRIES;
379 zfsctl_common_getattr(vp, vap);
386 * Special case the handling of "..".
390 zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
391 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
392 int *direntflags, pathname_t *realpnp)
394 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
398 * No extended attributes allowed under .zfs
400 if (flags & LOOKUP_XATTR)
405 if (strcmp(nm, "..") == 0) {
406 err = VFS_ROOT(dvp->v_vfsp, vpp);
408 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir,
409 cr, ct, direntflags, realpnp);
418 zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
419 caller_context_t *ct)
422 * We only care about ACL_ENABLED so that libsec can
423 * display ACL correctly and not default to POSIX draft.
425 if (cmd == _PC_ACL_ENABLED) {
426 *valp = _ACL_ACE_ENABLED;
430 return (fs_pathconf(vp, cmd, valp, cr, ct));
433 static const fs_operation_def_t zfsctl_tops_root[] = {
434 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
435 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
436 { VOPNAME_IOCTL, { .error = fs_inval } },
437 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_root_getattr } },
438 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
439 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
440 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_root_lookup } },
441 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
442 { VOPNAME_INACTIVE, { .vop_inactive = gfs_vop_inactive } },
443 { VOPNAME_PATHCONF, { .vop_pathconf = zfsctl_pathconf } },
444 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
449 zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname)
451 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
453 if (snapshot_namecheck(name, NULL, NULL) != 0)
455 dmu_objset_name(os, zname);
456 if (strlen(zname) + 1 + strlen(name) >= len)
457 return (ENAMETOOLONG);
458 (void) strcat(zname, "@");
459 (void) strcat(zname, name);
464 zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr)
466 vnode_t *svp = sep->se_root;
469 ASSERT(vn_ismntpt(svp));
471 /* this will be dropped by dounmount() */
472 if ((error = vn_vfswlock(svp)) != 0)
476 error = dounmount(vn_mountedvfs(svp), fflags, cr);
481 VFS_RELE(svp->v_vfsp);
483 * We can't use VN_RELE(), as that will try to invoke
484 * zfsctl_snapdir_inactive(), which would cause us to destroy
485 * the sd_lock mutex held by our caller.
487 ASSERT(svp->v_count == 1);
488 gfs_vop_inactive(svp, cr, NULL);
490 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
491 kmem_free(sep, sizeof (zfs_snapentry_t));
497 zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm)
502 char newpath[MAXNAMELEN];
505 ASSERT(MUTEX_HELD(&sdp->sd_lock));
508 vfsp = vn_mountedvfs(sep->se_root);
509 ASSERT(vfsp != NULL);
514 * Change the name in the AVL tree.
516 avl_remove(&sdp->sd_snaps, sep);
517 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
518 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
519 (void) strcpy(sep->se_name, nm);
520 VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL);
521 avl_insert(&sdp->sd_snaps, sep, where);
524 * Change the current mountpoint info:
525 * - update the tail of the mntpoint path
526 * - update the tail of the resource path
528 pathref = vfs_getmntpoint(vfsp);
529 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
530 VERIFY((tail = strrchr(newpath, '/')) != NULL);
532 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
533 (void) strcat(newpath, nm);
534 refstr_rele(pathref);
535 vfs_setmntpoint(vfsp, newpath);
537 pathref = vfs_getresource(vfsp);
538 (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
539 VERIFY((tail = strrchr(newpath, '@')) != NULL);
541 ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
542 (void) strcat(newpath, nm);
543 refstr_rele(pathref);
544 vfs_setresource(vfsp, newpath);
551 zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm,
552 cred_t *cr, caller_context_t *ct, int flags)
554 zfsctl_snapdir_t *sdp = sdvp->v_data;
555 zfs_snapentry_t search, *sep;
558 char from[MAXNAMELEN], to[MAXNAMELEN];
559 char real[MAXNAMELEN];
562 zfsvfs = sdvp->v_vfsp->vfs_data;
565 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
566 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real,
570 } else if (err != ENOTSUP) {
578 err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from);
580 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to);
582 err = zfs_secpolicy_rename_perms(from, to, cr);
587 * Cannot move snapshots out of the snapdir.
592 if (strcmp(snm, tnm) == 0)
595 mutex_enter(&sdp->sd_lock);
597 search.se_name = (char *)snm;
598 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) {
599 mutex_exit(&sdp->sd_lock);
603 err = dmu_objset_rename(from, to, B_FALSE);
605 zfsctl_rename_snap(sdp, sep, tnm);
607 mutex_exit(&sdp->sd_lock);
614 zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
615 caller_context_t *ct, int flags)
617 zfsctl_snapdir_t *sdp = dvp->v_data;
618 zfs_snapentry_t *sep;
619 zfs_snapentry_t search;
621 char snapname[MAXNAMELEN];
622 char real[MAXNAMELEN];
625 zfsvfs = dvp->v_vfsp->vfs_data;
628 if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
630 err = dmu_snapshot_realname(zfsvfs->z_os, name, real,
634 } else if (err != ENOTSUP) {
642 err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname);
644 err = zfs_secpolicy_destroy_perms(snapname, cr);
648 mutex_enter(&sdp->sd_lock);
650 search.se_name = name;
651 sep = avl_find(&sdp->sd_snaps, &search, NULL);
653 avl_remove(&sdp->sd_snaps, sep);
654 err = zfsctl_unmount_snap(sep, MS_FORCE, cr);
656 avl_add(&sdp->sd_snaps, sep);
658 err = dmu_objset_destroy(snapname);
663 mutex_exit(&sdp->sd_lock);
669 * This creates a snapshot under '.zfs/snapshot'.
673 zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp,
674 cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp)
676 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
677 char name[MAXNAMELEN];
679 static enum symfollow follow = NO_FOLLOW;
680 static enum uio_seg seg = UIO_SYSSPACE;
682 if (snapshot_namecheck(dirname, NULL, NULL) != 0)
685 dmu_objset_name(zfsvfs->z_os, name);
689 err = zfs_secpolicy_snapshot_perms(name, cr);
694 err = dmu_objset_snapshot(name, dirname, B_FALSE);
697 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp);
704 * Lookup entry point for the 'snapshot' directory. Try to open the
705 * snapshot if it exist, creating the pseudo filesystem vnode as necessary.
706 * Perform a mount of the associated dataset on top of the vnode.
710 zfsctl_snapdir_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
711 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
712 int *direntflags, pathname_t *realpnp)
714 zfsctl_snapdir_t *sdp = dvp->v_data;
716 char snapname[MAXNAMELEN];
717 char real[MAXNAMELEN];
719 zfs_snapentry_t *sep, search;
722 size_t mountpoint_len;
724 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
728 * No extended attributes allowed under .zfs
730 if (flags & LOOKUP_XATTR)
733 ASSERT(dvp->v_type == VDIR);
735 if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0)
739 * If we get a recursive call, that means we got called
740 * from the domount() code while it was trying to look up the
741 * spec (which looks like a local path for zfs). We need to
742 * add some flag to domount() to tell it not to do this lookup.
744 if (MUTEX_HELD(&sdp->sd_lock))
749 if (flags & FIGNORECASE) {
750 boolean_t conflict = B_FALSE;
752 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real,
753 MAXNAMELEN, &conflict);
756 } else if (err != ENOTSUP) {
761 (void) strlcpy(realpnp->pn_buf, nm,
762 realpnp->pn_bufsize);
763 if (conflict && direntflags)
764 *direntflags = ED_CASE_CONFLICT;
767 mutex_enter(&sdp->sd_lock);
768 search.se_name = (char *)nm;
769 if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) {
776 } else if (*vpp == sep->se_root) {
778 * The snapshot was unmounted behind our backs,
784 * VROOT was set during the traverse call. We need
785 * to clear it since we're pretending to be part
786 * of our parent's vfs.
788 (*vpp)->v_flag &= ~VROOT;
790 mutex_exit(&sdp->sd_lock);
796 * The requested snapshot is not currently mounted, look it up.
798 err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname);
800 mutex_exit(&sdp->sd_lock);
803 * handle "ls *" or "?" in a graceful manner,
804 * forcing EILSEQ to ENOENT.
805 * Since shell ultimately passes "*" or "?" as name to lookup
807 return (err == EILSEQ ? ENOENT : err);
809 if (dmu_objset_open(snapname, DMU_OST_ZFS,
810 DS_MODE_USER | DS_MODE_READONLY, &snap) != 0) {
811 mutex_exit(&sdp->sd_lock);
816 sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP);
817 sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
818 (void) strcpy(sep->se_name, nm);
819 *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap));
820 avl_insert(&sdp->sd_snaps, sep, where);
822 dmu_objset_close(snap);
824 mountpoint_len = strlen(refstr_value(dvp->v_vfsp->vfs_mntpt)) +
825 strlen("/.zfs/snapshot/") + strlen(nm) + 1;
826 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP);
827 (void) snprintf(mountpoint, mountpoint_len, "%s/.zfs/snapshot/%s",
828 refstr_value(dvp->v_vfsp->vfs_mntpt), nm);
830 margs.spec = snapname;
831 margs.dir = mountpoint;
832 margs.flags = MS_SYSSPACE | MS_NOMNTTAB;
833 margs.fstype = "zfs";
834 margs.dataptr = NULL;
839 err = domount("zfs", &margs, *vpp, kcred, &vfsp);
840 kmem_free(mountpoint, mountpoint_len);
844 * Return the mounted root rather than the covered mount point.
845 * Takes the GFS vnode at .zfs/snapshot/<snapname> and returns
846 * the ZFS vnode mounted on top of the GFS node. This ZFS
847 * vnode is the root the newly created vfsp.
855 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>.
857 * This is where we lie about our v_vfsp in order to
858 * make .zfs/snapshot/<snapname> accessible over NFS
859 * without requiring manual mounts of <snapname>.
861 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs);
862 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs;
863 (*vpp)->v_vfsp = zfsvfs->z_vfs;
864 (*vpp)->v_flag &= ~VROOT;
866 mutex_exit(&sdp->sd_lock);
870 * If we had an error, drop our hold on the vnode and
871 * zfsctl_snapshot_inactive() will clean up.
882 zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp,
883 offset_t *offp, offset_t *nextp, void *data, int flags)
885 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
886 char snapname[MAXNAMELEN];
888 boolean_t case_conflict;
894 error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id,
895 &cookie, &case_conflict);
898 if (error == ENOENT) {
905 if (flags & V_RDDIR_ENTFLAGS) {
906 edirent_t *eodp = dp;
908 (void) strcpy(eodp->ed_name, snapname);
909 eodp->ed_ino = ZFSCTL_INO_SNAP(id);
910 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0;
912 struct dirent64 *odp = dp;
914 (void) strcpy(odp->d_name, snapname);
915 odp->d_ino = ZFSCTL_INO_SNAP(id);
925 * pvp is the '.zfs' directory (zfsctl_node_t).
926 * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t).
928 * This function is the callback to create a GFS vnode for '.zfs/snapshot'
929 * when a lookup is performed on .zfs for "snapshot".
932 zfsctl_mknode_snapdir(vnode_t *pvp)
935 zfsctl_snapdir_t *sdp;
937 vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp,
938 zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN,
939 zfsctl_snapdir_readdir_cb, NULL);
941 sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR;
942 sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
943 mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL);
944 avl_create(&sdp->sd_snaps, snapentry_compare,
945 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node));
951 zfsctl_snapdir_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
952 caller_context_t *ct)
954 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
955 zfsctl_snapdir_t *sdp = vp->v_data;
958 zfsctl_common_getattr(vp, vap);
959 vap->va_nodeid = gfs_file_inode(vp);
960 vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2;
968 zfsctl_snapdir_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
970 zfsctl_snapdir_t *sdp = vp->v_data;
973 private = gfs_dir_inactive(vp);
974 if (private != NULL) {
975 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0);
976 mutex_destroy(&sdp->sd_lock);
977 avl_destroy(&sdp->sd_snaps);
978 kmem_free(private, sizeof (zfsctl_snapdir_t));
982 static const fs_operation_def_t zfsctl_tops_snapdir[] = {
983 { VOPNAME_OPEN, { .vop_open = zfsctl_common_open } },
984 { VOPNAME_CLOSE, { .vop_close = zfsctl_common_close } },
985 { VOPNAME_IOCTL, { .error = fs_inval } },
986 { VOPNAME_GETATTR, { .vop_getattr = zfsctl_snapdir_getattr } },
987 { VOPNAME_ACCESS, { .vop_access = zfsctl_common_access } },
988 { VOPNAME_RENAME, { .vop_rename = zfsctl_snapdir_rename } },
989 { VOPNAME_RMDIR, { .vop_rmdir = zfsctl_snapdir_remove } },
990 { VOPNAME_MKDIR, { .vop_mkdir = zfsctl_snapdir_mkdir } },
991 { VOPNAME_READDIR, { .vop_readdir = gfs_vop_readdir } },
992 { VOPNAME_LOOKUP, { .vop_lookup = zfsctl_snapdir_lookup } },
993 { VOPNAME_SEEK, { .vop_seek = fs_seek } },
994 { VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapdir_inactive } },
995 { VOPNAME_FID, { .vop_fid = zfsctl_common_fid } },
1000 * pvp is the GFS vnode '.zfs/snapshot'.
1002 * This creates a GFS node under '.zfs/snapshot' representing each
1003 * snapshot. This newly created GFS node is what we mount snapshot
1007 zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset)
1012 vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp,
1013 zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL);
1015 zcp->zc_id = objset;
1016 VFS_HOLD(vp->v_vfsp);
1022 zfsctl_snapshot_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1024 zfsctl_snapdir_t *sdp;
1025 zfs_snapentry_t *sep, *next;
1028 VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0);
1031 mutex_enter(&sdp->sd_lock);
1033 if (vp->v_count > 1) {
1034 mutex_exit(&sdp->sd_lock);
1037 ASSERT(!vn_ismntpt(vp));
1039 sep = avl_first(&sdp->sd_snaps);
1040 while (sep != NULL) {
1041 next = AVL_NEXT(&sdp->sd_snaps, sep);
1043 if (sep->se_root == vp) {
1044 avl_remove(&sdp->sd_snaps, sep);
1045 kmem_free(sep->se_name, strlen(sep->se_name) + 1);
1046 kmem_free(sep, sizeof (zfs_snapentry_t));
1051 ASSERT(sep != NULL);
1053 mutex_exit(&sdp->sd_lock);
1055 VFS_RELE(vp->v_vfsp);
1058 * Dispose of the vnode for the snapshot mount point.
1059 * This is safe to do because once this entry has been removed
1060 * from the AVL tree, it can't be found again, so cannot become
1061 * "active". If we lookup the same name again we will end up
1062 * creating a new vnode.
1064 gfs_vop_inactive(vp, cr, ct);
1069 * These VP's should never see the light of day. They should always
1072 static const fs_operation_def_t zfsctl_tops_snapshot[] = {
1073 VOPNAME_INACTIVE, { .vop_inactive = zfsctl_snapshot_inactive },
1078 zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp)
1080 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1082 zfsctl_snapdir_t *sdp;
1084 zfs_snapentry_t *sep;
1087 ASSERT(zfsvfs->z_ctldir != NULL);
1088 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1089 NULL, 0, NULL, kcred, NULL, NULL, NULL);
1094 mutex_enter(&sdp->sd_lock);
1095 sep = avl_first(&sdp->sd_snaps);
1096 while (sep != NULL) {
1099 if (zcp->zc_id == objsetid)
1102 sep = AVL_NEXT(&sdp->sd_snaps, sep);
1108 * Return the mounted root rather than the covered mount point.
1109 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid>
1110 * and returns the ZFS vnode mounted on top of the GFS node.
1111 * This ZFS vnode is the root of the vfs for objset 'objsetid'.
1113 error = traverse(&vp);
1115 if (vp == sep->se_root)
1118 *zfsvfsp = VTOZ(vp)->z_zfsvfs;
1120 mutex_exit(&sdp->sd_lock);
1124 mutex_exit(&sdp->sd_lock);
1133 * Unmount any snapshots for the given filesystem. This is called from
1134 * zfs_umount() - if we have a ctldir, then go through and unmount all the
1138 zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr)
1140 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1142 zfsctl_snapdir_t *sdp;
1143 zfs_snapentry_t *sep, *next;
1146 ASSERT(zfsvfs->z_ctldir != NULL);
1147 error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1148 NULL, 0, NULL, cr, NULL, NULL, NULL);
1153 mutex_enter(&sdp->sd_lock);
1155 sep = avl_first(&sdp->sd_snaps);
1156 while (sep != NULL) {
1157 next = AVL_NEXT(&sdp->sd_snaps, sep);
1160 * If this snapshot is not mounted, then it must
1161 * have just been unmounted by somebody else, and
1162 * will be cleaned up by zfsctl_snapdir_inactive().
1164 if (vn_ismntpt(sep->se_root)) {
1165 avl_remove(&sdp->sd_snaps, sep);
1166 error = zfsctl_unmount_snap(sep, fflags, cr);
1168 avl_add(&sdp->sd_snaps, sep);
1175 mutex_exit(&sdp->sd_lock);