xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_ctldir.c (revision ecd6cf800b63704be73fb264c3f5b6e0dafc068d)
1fa9e4066Sahrens /*
2fa9e4066Sahrens  * CDDL HEADER START
3fa9e4066Sahrens  *
4fa9e4066Sahrens  * The contents of this file are subject to the terms of the
50f2dc02eSek  * Common Development and Distribution License (the "License").
60f2dc02eSek  * You may not use this file except in compliance with the License.
7fa9e4066Sahrens  *
8fa9e4066Sahrens  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9fa9e4066Sahrens  * or http://www.opensolaris.org/os/licensing.
10fa9e4066Sahrens  * See the License for the specific language governing permissions
11fa9e4066Sahrens  * and limitations under the License.
12fa9e4066Sahrens  *
13fa9e4066Sahrens  * When distributing Covered Code, include this CDDL HEADER in each
14fa9e4066Sahrens  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15fa9e4066Sahrens  * If applicable, add the following below this CDDL HEADER, with the
16fa9e4066Sahrens  * fields enclosed by brackets "[]" replaced with your own identifying
17fa9e4066Sahrens  * information: Portions Copyright [yyyy] [name of copyright owner]
18fa9e4066Sahrens  *
19fa9e4066Sahrens  * CDDL HEADER END
20fa9e4066Sahrens  */
21fa9e4066Sahrens /*
22aa59c4cbSrsb  * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
23fa9e4066Sahrens  * Use is subject to license terms.
24fa9e4066Sahrens  */
25fa9e4066Sahrens 
26fa9e4066Sahrens #pragma ident	"%Z%%M%	%I%	%E% SMI"
27fa9e4066Sahrens 
28fa9e4066Sahrens /*
29fa9e4066Sahrens  * ZFS control directory (a.k.a. ".zfs")
30fa9e4066Sahrens  *
31fa9e4066Sahrens  * This directory provides a common location for all ZFS meta-objects.
32fa9e4066Sahrens  * Currently, this is only the 'snapshot' directory, but this may expand in the
33fa9e4066Sahrens  * future.  The elements are built using the GFS primitives, as the hierarchy
34fa9e4066Sahrens  * does not actually exist on disk.
35fa9e4066Sahrens  *
36fa9e4066Sahrens  * For 'snapshot', we don't want to have all snapshots always mounted, because
37fa9e4066Sahrens  * this would take up a huge amount of space in /etc/mnttab.  We have three
38fa9e4066Sahrens  * types of objects:
39fa9e4066Sahrens  *
40fa9e4066Sahrens  * 	ctldir ------> snapshotdir -------> snapshot
41fa9e4066Sahrens  *                                             |
42fa9e4066Sahrens  *                                             |
43fa9e4066Sahrens  *                                             V
44fa9e4066Sahrens  *                                         mounted fs
45fa9e4066Sahrens  *
46fa9e4066Sahrens  * The 'snapshot' node contains just enough information to lookup '..' and act
47fa9e4066Sahrens  * as a mountpoint for the snapshot.  Whenever we lookup a specific snapshot, we
48fa9e4066Sahrens  * perform an automount of the underlying filesystem and return the
49fa9e4066Sahrens  * corresponding vnode.
50fa9e4066Sahrens  *
51fa9e4066Sahrens  * All mounts are handled automatically by the kernel, but unmounts are
52fa9e4066Sahrens  * (currently) handled from user land.  The main reason is that there is no
53fa9e4066Sahrens  * reliable way to auto-unmount the filesystem when it's "no longer in use".
54fa9e4066Sahrens  * When the user unmounts a filesystem, we call zfsctl_unmount(), which
55fa9e4066Sahrens  * unmounts any snapshots within the snapshot directory.
56fa9e4066Sahrens  */
57fa9e4066Sahrens 
58fa9e4066Sahrens #include <fs/fs_subr.h>
59fa9e4066Sahrens #include <sys/zfs_ctldir.h>
60fa9e4066Sahrens #include <sys/zfs_ioctl.h>
61fa9e4066Sahrens #include <sys/zfs_vfsops.h>
62aa59c4cbSrsb #include <sys/vfs_opreg.h>
63fa9e4066Sahrens #include <sys/gfs.h>
64fa9e4066Sahrens #include <sys/stat.h>
65fa9e4066Sahrens #include <sys/dmu.h>
66*ecd6cf80Smarks #include <sys/dsl_deleg.h>
67fa9e4066Sahrens #include <sys/mount.h>
68fa9e4066Sahrens 
69fa9e4066Sahrens typedef struct {
70fa9e4066Sahrens 	char		*se_name;
71fa9e4066Sahrens 	vnode_t		*se_root;
72fa9e4066Sahrens 	avl_node_t	se_node;
73fa9e4066Sahrens } zfs_snapentry_t;
74fa9e4066Sahrens 
75fa9e4066Sahrens static int
76fa9e4066Sahrens snapentry_compare(const void *a, const void *b)
77fa9e4066Sahrens {
78fa9e4066Sahrens 	const zfs_snapentry_t *sa = a;
79fa9e4066Sahrens 	const zfs_snapentry_t *sb = b;
80fa9e4066Sahrens 	int ret = strcmp(sa->se_name, sb->se_name);
81fa9e4066Sahrens 
82fa9e4066Sahrens 	if (ret < 0)
83fa9e4066Sahrens 		return (-1);
84fa9e4066Sahrens 	else if (ret > 0)
85fa9e4066Sahrens 		return (1);
86fa9e4066Sahrens 	else
87fa9e4066Sahrens 		return (0);
88fa9e4066Sahrens }
89fa9e4066Sahrens 
90fa9e4066Sahrens vnodeops_t *zfsctl_ops_root;
91fa9e4066Sahrens vnodeops_t *zfsctl_ops_snapdir;
92fa9e4066Sahrens vnodeops_t *zfsctl_ops_snapshot;
93fa9e4066Sahrens 
94fa9e4066Sahrens static const fs_operation_def_t zfsctl_tops_root[];
95fa9e4066Sahrens static const fs_operation_def_t zfsctl_tops_snapdir[];
96fa9e4066Sahrens static const fs_operation_def_t zfsctl_tops_snapshot[];
97fa9e4066Sahrens 
98fa9e4066Sahrens static vnode_t *zfsctl_mknode_snapdir(vnode_t *);
99fa9e4066Sahrens static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset);
100fa9e4066Sahrens 
101fa9e4066Sahrens static gfs_opsvec_t zfsctl_opsvec[] = {
102fa9e4066Sahrens 	{ ".zfs", zfsctl_tops_root, &zfsctl_ops_root },
103fa9e4066Sahrens 	{ ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir },
104fa9e4066Sahrens 	{ ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot },
105fa9e4066Sahrens 	{ NULL }
106fa9e4066Sahrens };
107fa9e4066Sahrens 
108fa9e4066Sahrens typedef struct zfsctl_node {
109fa9e4066Sahrens 	gfs_dir_t	zc_gfs_private;
110fa9e4066Sahrens 	uint64_t	zc_id;
111a32fded1Sek 	timestruc_t	zc_cmtime;	/* ctime and mtime, always the same */
112fa9e4066Sahrens } zfsctl_node_t;
113fa9e4066Sahrens 
114fa9e4066Sahrens typedef struct zfsctl_snapdir {
115fa9e4066Sahrens 	zfsctl_node_t	sd_node;
116fa9e4066Sahrens 	kmutex_t	sd_lock;
117fa9e4066Sahrens 	avl_tree_t	sd_snaps;
118fa9e4066Sahrens } zfsctl_snapdir_t;
119fa9e4066Sahrens 
120fa9e4066Sahrens /*
121fa9e4066Sahrens  * Root directory elements.  We have only a single static entry, 'snapshot'.
122fa9e4066Sahrens  */
123fa9e4066Sahrens static gfs_dirent_t zfsctl_root_entries[] = {
124fa9e4066Sahrens 	{ "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE },
125fa9e4066Sahrens 	{ NULL }
126fa9e4066Sahrens };
127fa9e4066Sahrens 
128fa9e4066Sahrens /* include . and .. in the calculation */
129fa9e4066Sahrens #define	NROOT_ENTRIES	((sizeof (zfsctl_root_entries) / \
130fa9e4066Sahrens     sizeof (gfs_dirent_t)) + 1)
131fa9e4066Sahrens 
132fa9e4066Sahrens 
133fa9e4066Sahrens /*
134fa9e4066Sahrens  * Initialize the various GFS pieces we'll need to create and manipulate .zfs
135fa9e4066Sahrens  * directories.  This is called from the ZFS init routine, and initializes the
136fa9e4066Sahrens  * vnode ops vectors that we'll be using.
137fa9e4066Sahrens  */
138fa9e4066Sahrens void
139fa9e4066Sahrens zfsctl_init(void)
140fa9e4066Sahrens {
141fa9e4066Sahrens 	VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0);
142fa9e4066Sahrens }
143fa9e4066Sahrens 
144fa9e4066Sahrens void
145fa9e4066Sahrens zfsctl_fini(void)
146fa9e4066Sahrens {
147fa9e4066Sahrens 	/*
148fa9e4066Sahrens 	 * Remove vfsctl vnode ops
149fa9e4066Sahrens 	 */
150fa9e4066Sahrens 	if (zfsctl_ops_root)
151fa9e4066Sahrens 		vn_freevnodeops(zfsctl_ops_root);
152fa9e4066Sahrens 	if (zfsctl_ops_snapdir)
153fa9e4066Sahrens 		vn_freevnodeops(zfsctl_ops_snapdir);
154fa9e4066Sahrens 	if (zfsctl_ops_snapshot)
155fa9e4066Sahrens 		vn_freevnodeops(zfsctl_ops_snapshot);
156fa9e4066Sahrens 
157fa9e4066Sahrens 	zfsctl_ops_root = NULL;
158fa9e4066Sahrens 	zfsctl_ops_snapdir = NULL;
159fa9e4066Sahrens 	zfsctl_ops_snapshot = NULL;
160fa9e4066Sahrens }
161fa9e4066Sahrens 
162fa9e4066Sahrens /*
163fa9e4066Sahrens  * Return the inode number associated with the 'snapshot' directory.
164fa9e4066Sahrens  */
165fa9e4066Sahrens /* ARGSUSED */
166fa9e4066Sahrens static ino64_t
167fa9e4066Sahrens zfsctl_root_inode_cb(vnode_t *vp, int index)
168fa9e4066Sahrens {
169fa9e4066Sahrens 	ASSERT(index == 0);
170fa9e4066Sahrens 	return (ZFSCTL_INO_SNAPDIR);
171fa9e4066Sahrens }
172fa9e4066Sahrens 
173fa9e4066Sahrens /*
174fa9e4066Sahrens  * Create the '.zfs' directory.  This directory is cached as part of the VFS
175fa9e4066Sahrens  * structure.  This results in a hold on the vfs_t.  The code in zfs_umount()
176fa9e4066Sahrens  * therefore checks against a vfs_count of 2 instead of 1.  This reference
177fa9e4066Sahrens  * is removed when the ctldir is destroyed in the unmount.
178fa9e4066Sahrens  */
179fa9e4066Sahrens void
180fa9e4066Sahrens zfsctl_create(zfsvfs_t *zfsvfs)
181fa9e4066Sahrens {
182a32fded1Sek 	vnode_t *vp, *rvp;
183fa9e4066Sahrens 	zfsctl_node_t *zcp;
184fa9e4066Sahrens 
185fa9e4066Sahrens 	ASSERT(zfsvfs->z_ctldir == NULL);
186fa9e4066Sahrens 
187fa9e4066Sahrens 	vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs,
188fa9e4066Sahrens 	    zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries,
189fa9e4066Sahrens 	    zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL);
190fa9e4066Sahrens 	zcp = vp->v_data;
191fa9e4066Sahrens 	zcp->zc_id = ZFSCTL_INO_ROOT;
192fa9e4066Sahrens 
193a32fded1Sek 	VERIFY(VFS_ROOT(zfsvfs->z_vfs, &rvp) == 0);
194a32fded1Sek 	ZFS_TIME_DECODE(&zcp->zc_cmtime, VTOZ(rvp)->z_phys->zp_crtime);
195a32fded1Sek 	VN_RELE(rvp);
196a32fded1Sek 
197fa9e4066Sahrens 	/*
198fa9e4066Sahrens 	 * We're only faking the fact that we have a root of a filesystem for
199fa9e4066Sahrens 	 * the sake of the GFS interfaces.  Undo the flag manipulation it did
200fa9e4066Sahrens 	 * for us.
201fa9e4066Sahrens 	 */
202fa9e4066Sahrens 	vp->v_flag &= ~(VROOT | VNOCACHE | VNOMAP | VNOSWAP | VNOMOUNT);
203fa9e4066Sahrens 
204fa9e4066Sahrens 	zfsvfs->z_ctldir = vp;
205fa9e4066Sahrens }
206fa9e4066Sahrens 
207fa9e4066Sahrens /*
2088afd4dd6Sperrin  * Destroy the '.zfs' directory.  Only called when the filesystem is unmounted.
2098afd4dd6Sperrin  * There might still be more references if we were force unmounted, but only
2108afd4dd6Sperrin  * new zfs_inactive() calls can occur and they don't reference .zfs
211fa9e4066Sahrens  */
212fa9e4066Sahrens void
213fa9e4066Sahrens zfsctl_destroy(zfsvfs_t *zfsvfs)
214fa9e4066Sahrens {
215fa9e4066Sahrens 	VN_RELE(zfsvfs->z_ctldir);
216fa9e4066Sahrens 	zfsvfs->z_ctldir = NULL;
217fa9e4066Sahrens }
218fa9e4066Sahrens 
219fa9e4066Sahrens /*
220fa9e4066Sahrens  * Given a root znode, retrieve the associated .zfs directory.
221fa9e4066Sahrens  * Add a hold to the vnode and return it.
222fa9e4066Sahrens  */
223fa9e4066Sahrens vnode_t *
224fa9e4066Sahrens zfsctl_root(znode_t *zp)
225fa9e4066Sahrens {
226fa9e4066Sahrens 	ASSERT(zfs_has_ctldir(zp));
227fa9e4066Sahrens 	VN_HOLD(zp->z_zfsvfs->z_ctldir);
228fa9e4066Sahrens 	return (zp->z_zfsvfs->z_ctldir);
229fa9e4066Sahrens }
230fa9e4066Sahrens 
231fa9e4066Sahrens /*
232fa9e4066Sahrens  * Common open routine.  Disallow any write access.
233fa9e4066Sahrens  */
234fa9e4066Sahrens /* ARGSUSED */
235fa9e4066Sahrens static int
236fa9e4066Sahrens zfsctl_common_open(vnode_t **vpp, int flags, cred_t *cr)
237fa9e4066Sahrens {
238fa9e4066Sahrens 	if (flags & FWRITE)
239fa9e4066Sahrens 		return (EACCES);
240fa9e4066Sahrens 
241fa9e4066Sahrens 	return (0);
242fa9e4066Sahrens }
243fa9e4066Sahrens 
244fa9e4066Sahrens /*
245fa9e4066Sahrens  * Common close routine.  Nothing to do here.
246fa9e4066Sahrens  */
247fa9e4066Sahrens /* ARGSUSED */
248fa9e4066Sahrens static int
249fa9e4066Sahrens zfsctl_common_close(vnode_t *vpp, int flags, int count, offset_t off,
250fa9e4066Sahrens     cred_t *cr)
251fa9e4066Sahrens {
252fa9e4066Sahrens 	return (0);
253fa9e4066Sahrens }
254fa9e4066Sahrens 
255fa9e4066Sahrens /*
256fa9e4066Sahrens  * Common access routine.  Disallow writes.
257fa9e4066Sahrens  */
258fa9e4066Sahrens /* ARGSUSED */
259fa9e4066Sahrens static int
260fa9e4066Sahrens zfsctl_common_access(vnode_t *vp, int mode, int flags, cred_t *cr)
261fa9e4066Sahrens {
262fa9e4066Sahrens 	if (mode & VWRITE)
263fa9e4066Sahrens 		return (EACCES);
264fa9e4066Sahrens 
265fa9e4066Sahrens 	return (0);
266fa9e4066Sahrens }
267fa9e4066Sahrens 
268fa9e4066Sahrens /*
269fa9e4066Sahrens  * Common getattr function.  Fill in basic information.
270fa9e4066Sahrens  */
271fa9e4066Sahrens static void
272fa9e4066Sahrens zfsctl_common_getattr(vnode_t *vp, vattr_t *vap)
273fa9e4066Sahrens {
274a32fded1Sek 	zfsctl_node_t	*zcp = vp->v_data;
275a32fded1Sek 	timestruc_t	now;
276fa9e4066Sahrens 
277fa9e4066Sahrens 	vap->va_uid = 0;
278fa9e4066Sahrens 	vap->va_gid = 0;
279fa9e4066Sahrens 	vap->va_rdev = 0;
280fa9e4066Sahrens 	/*
281fa9e4066Sahrens 	 * We are a purly virtual object, so we have no
282fa9e4066Sahrens 	 * blocksize or allocated blocks.
283fa9e4066Sahrens 	 */
284fa9e4066Sahrens 	vap->va_blksize = 0;
285fa9e4066Sahrens 	vap->va_nblocks = 0;
286fa9e4066Sahrens 	vap->va_seq = 0;
287fa9e4066Sahrens 	vap->va_fsid = vp->v_vfsp->vfs_dev;
288fa9e4066Sahrens 	vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP |
289fa9e4066Sahrens 	    S_IROTH | S_IXOTH;
290fa9e4066Sahrens 	vap->va_type = VDIR;
291fa9e4066Sahrens 	/*
292a32fded1Sek 	 * We live in the now (for atime).
293fa9e4066Sahrens 	 */
294fa9e4066Sahrens 	gethrestime(&now);
295a32fded1Sek 	vap->va_atime = now;
296a32fded1Sek 	vap->va_mtime = vap->va_ctime = zcp->zc_cmtime;
297fa9e4066Sahrens }
298fa9e4066Sahrens 
299fa9e4066Sahrens static int
300fa9e4066Sahrens zfsctl_common_fid(vnode_t *vp, fid_t *fidp)
301fa9e4066Sahrens {
302fa9e4066Sahrens 	zfsvfs_t	*zfsvfs = vp->v_vfsp->vfs_data;
303fa9e4066Sahrens 	zfsctl_node_t	*zcp = vp->v_data;
304fa9e4066Sahrens 	uint64_t	object = zcp->zc_id;
305fa9e4066Sahrens 	zfid_short_t	*zfid;
306fa9e4066Sahrens 	int		i;
307fa9e4066Sahrens 
308fa9e4066Sahrens 	ZFS_ENTER(zfsvfs);
309fa9e4066Sahrens 
310fa9e4066Sahrens 	if (fidp->fid_len < SHORT_FID_LEN) {
311fa9e4066Sahrens 		fidp->fid_len = SHORT_FID_LEN;
3120f2dc02eSek 		ZFS_EXIT(zfsvfs);
313fa9e4066Sahrens 		return (ENOSPC);
314fa9e4066Sahrens 	}
315fa9e4066Sahrens 
316fa9e4066Sahrens 	zfid = (zfid_short_t *)fidp;
317fa9e4066Sahrens 
318fa9e4066Sahrens 	zfid->zf_len = SHORT_FID_LEN;
319fa9e4066Sahrens 
320fa9e4066Sahrens 	for (i = 0; i < sizeof (zfid->zf_object); i++)
321fa9e4066Sahrens 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
322fa9e4066Sahrens 
323fa9e4066Sahrens 	/* .zfs znodes always have a generation number of 0 */
324fa9e4066Sahrens 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
325fa9e4066Sahrens 		zfid->zf_gen[i] = 0;
326fa9e4066Sahrens 
327fa9e4066Sahrens 	ZFS_EXIT(zfsvfs);
328fa9e4066Sahrens 	return (0);
329fa9e4066Sahrens }
330fa9e4066Sahrens 
331fa9e4066Sahrens /*
332fa9e4066Sahrens  * .zfs inode namespace
333fa9e4066Sahrens  *
334fa9e4066Sahrens  * We need to generate unique inode numbers for all files and directories
335fa9e4066Sahrens  * within the .zfs pseudo-filesystem.  We use the following scheme:
336fa9e4066Sahrens  *
337fa9e4066Sahrens  * 	ENTRY			ZFSCTL_INODE
338fa9e4066Sahrens  * 	.zfs			1
339fa9e4066Sahrens  * 	.zfs/snapshot		2
340fa9e4066Sahrens  * 	.zfs/snapshot/<snap>	objectid(snap)
341fa9e4066Sahrens  */
342fa9e4066Sahrens 
343fa9e4066Sahrens #define	ZFSCTL_INO_SNAP(id)	(id)
344fa9e4066Sahrens 
345fa9e4066Sahrens /*
346fa9e4066Sahrens  * Get root directory attributes.
347fa9e4066Sahrens  */
348fa9e4066Sahrens /* ARGSUSED */
349fa9e4066Sahrens static int
350fa9e4066Sahrens zfsctl_root_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr)
351fa9e4066Sahrens {
352fa9e4066Sahrens 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
353fa9e4066Sahrens 
354fa9e4066Sahrens 	ZFS_ENTER(zfsvfs);
355fa9e4066Sahrens 	vap->va_nodeid = ZFSCTL_INO_ROOT;
356fa9e4066Sahrens 	vap->va_nlink = vap->va_size = NROOT_ENTRIES;
357fa9e4066Sahrens 
358fa9e4066Sahrens 	zfsctl_common_getattr(vp, vap);
359fa9e4066Sahrens 	ZFS_EXIT(zfsvfs);
360fa9e4066Sahrens 
361fa9e4066Sahrens 	return (0);
362fa9e4066Sahrens }
363fa9e4066Sahrens 
364fa9e4066Sahrens /*
365fa9e4066Sahrens  * Special case the handling of "..".
366fa9e4066Sahrens  */
367fa9e4066Sahrens /* ARGSUSED */
368fa9e4066Sahrens int
369fa9e4066Sahrens zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
370fa9e4066Sahrens     int flags, vnode_t *rdir, cred_t *cr)
371fa9e4066Sahrens {
372fa9e4066Sahrens 	zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
373fa9e4066Sahrens 	int err;
374fa9e4066Sahrens 
375fa9e4066Sahrens 	ZFS_ENTER(zfsvfs);
376fa9e4066Sahrens 
377fa9e4066Sahrens 	if (strcmp(nm, "..") == 0) {
378fa9e4066Sahrens 		err = VFS_ROOT(dvp->v_vfsp, vpp);
379fa9e4066Sahrens 	} else {
380fa9e4066Sahrens 		err = gfs_dir_lookup(dvp, nm, vpp);
381fa9e4066Sahrens 	}
382fa9e4066Sahrens 
383fa9e4066Sahrens 	ZFS_EXIT(zfsvfs);
384fa9e4066Sahrens 
385fa9e4066Sahrens 	return (err);
386fa9e4066Sahrens }
387fa9e4066Sahrens 
388fa9e4066Sahrens static const fs_operation_def_t zfsctl_tops_root[] = {
389aa59c4cbSrsb 	{ VOPNAME_OPEN,		{ .vop_open = zfsctl_common_open }	},
390aa59c4cbSrsb 	{ VOPNAME_CLOSE,	{ .vop_close = zfsctl_common_close }	},
391aa59c4cbSrsb 	{ VOPNAME_IOCTL,	{ .error = fs_inval }			},
392aa59c4cbSrsb 	{ VOPNAME_GETATTR,	{ .vop_getattr = zfsctl_root_getattr }	},
393aa59c4cbSrsb 	{ VOPNAME_ACCESS,	{ .vop_access = zfsctl_common_access }	},
394aa59c4cbSrsb 	{ VOPNAME_READDIR,	{ .vop_readdir = gfs_vop_readdir } 	},
395aa59c4cbSrsb 	{ VOPNAME_LOOKUP,	{ .vop_lookup = zfsctl_root_lookup }	},
396aa59c4cbSrsb 	{ VOPNAME_SEEK,		{ .vop_seek = fs_seek }			},
397aa59c4cbSrsb 	{ VOPNAME_INACTIVE,	{ .vop_inactive = gfs_vop_inactive }	},
398aa59c4cbSrsb 	{ VOPNAME_FID,		{ .vop_fid = zfsctl_common_fid	}	},
399fa9e4066Sahrens 	{ NULL }
400fa9e4066Sahrens };
401fa9e4066Sahrens 
402fa9e4066Sahrens static int
403fa9e4066Sahrens zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname)
404fa9e4066Sahrens {
405fa9e4066Sahrens 	objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
406fa9e4066Sahrens 
407fa9e4066Sahrens 	dmu_objset_name(os, zname);
4085422900bSmaybee 	if (strlen(zname) + 1 + strlen(name) >= len)
409fa9e4066Sahrens 		return (ENAMETOOLONG);
4105422900bSmaybee 	(void) strcat(zname, "@");
411fa9e4066Sahrens 	(void) strcat(zname, name);
412fa9e4066Sahrens 	return (0);
413fa9e4066Sahrens }
414fa9e4066Sahrens 
415*ecd6cf80Smarks int
416fa9e4066Sahrens zfsctl_unmount_snap(vnode_t *dvp, const char *name, int force, cred_t *cr)
417fa9e4066Sahrens {
418fa9e4066Sahrens 	zfsctl_snapdir_t *sdp = dvp->v_data;
419fa9e4066Sahrens 	zfs_snapentry_t search, *sep;
420fa9e4066Sahrens 	avl_index_t where;
421fa9e4066Sahrens 	int err;
422fa9e4066Sahrens 
423fa9e4066Sahrens 	ASSERT(MUTEX_HELD(&sdp->sd_lock));
424fa9e4066Sahrens 
425fa9e4066Sahrens 	search.se_name = (char *)name;
426fa9e4066Sahrens 	if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL)
427fa9e4066Sahrens 		return (ENOENT);
428fa9e4066Sahrens 
429fa9e4066Sahrens 	ASSERT(vn_ismntpt(sep->se_root));
430fa9e4066Sahrens 
431fa9e4066Sahrens 	/* this will be dropped by dounmount() */
432fa9e4066Sahrens 	if ((err = vn_vfswlock(sep->se_root)) != 0)
433fa9e4066Sahrens 		return (err);
434fa9e4066Sahrens 
435fa9e4066Sahrens 	VN_HOLD(sep->se_root);
436bbf4a8dfSmaybee 	err = dounmount(vn_mountedvfs(sep->se_root), force, kcred);
437bbf4a8dfSmaybee 	if (err) {
438bbf4a8dfSmaybee 		VN_RELE(sep->se_root);
439fa9e4066Sahrens 		return (err);
440bbf4a8dfSmaybee 	}
441fa9e4066Sahrens 	ASSERT(sep->se_root->v_count == 1);
442fa9e4066Sahrens 	gfs_vop_inactive(sep->se_root, cr);
443fa9e4066Sahrens 
444fa9e4066Sahrens 	avl_remove(&sdp->sd_snaps, sep);
445fa9e4066Sahrens 	kmem_free(sep->se_name, strlen(sep->se_name) + 1);
446fa9e4066Sahrens 	kmem_free(sep, sizeof (zfs_snapentry_t));
447fa9e4066Sahrens 
448fa9e4066Sahrens 	return (0);
449fa9e4066Sahrens }
450fa9e4066Sahrens 
451fa9e4066Sahrens 
4525422900bSmaybee static void
453fa9e4066Sahrens zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm)
454fa9e4066Sahrens {
455fa9e4066Sahrens 	avl_index_t where;
456fa9e4066Sahrens 	vfs_t *vfsp;
457fa9e4066Sahrens 	refstr_t *pathref;
458fa9e4066Sahrens 	char newpath[MAXNAMELEN];
459fa9e4066Sahrens 	char *tail;
460fa9e4066Sahrens 
461fa9e4066Sahrens 	ASSERT(MUTEX_HELD(&sdp->sd_lock));
462fa9e4066Sahrens 	ASSERT(sep != NULL);
463fa9e4066Sahrens 
464fa9e4066Sahrens 	vfsp = vn_mountedvfs(sep->se_root);
465fa9e4066Sahrens 	ASSERT(vfsp != NULL);
466fa9e4066Sahrens 
4675422900bSmaybee 	vfs_lock_wait(vfsp);
468fa9e4066Sahrens 
469fa9e4066Sahrens 	/*
470fa9e4066Sahrens 	 * Change the name in the AVL tree.
471fa9e4066Sahrens 	 */
472fa9e4066Sahrens 	avl_remove(&sdp->sd_snaps, sep);
473fa9e4066Sahrens 	kmem_free(sep->se_name, strlen(sep->se_name) + 1);
474fa9e4066Sahrens 	sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
475fa9e4066Sahrens 	(void) strcpy(sep->se_name, nm);
476fa9e4066Sahrens 	VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL);
477fa9e4066Sahrens 	avl_insert(&sdp->sd_snaps, sep, where);
478fa9e4066Sahrens 
479fa9e4066Sahrens 	/*
480fa9e4066Sahrens 	 * Change the current mountpoint info:
481fa9e4066Sahrens 	 * 	- update the tail of the mntpoint path
482fa9e4066Sahrens 	 *	- update the tail of the resource path
483fa9e4066Sahrens 	 */
484fa9e4066Sahrens 	pathref = vfs_getmntpoint(vfsp);
4850b69c2f0Sahrens 	(void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
4860b69c2f0Sahrens 	VERIFY((tail = strrchr(newpath, '/')) != NULL);
4870b69c2f0Sahrens 	*(tail+1) = '\0';
4880b69c2f0Sahrens 	ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
489fa9e4066Sahrens 	(void) strcat(newpath, nm);
490fa9e4066Sahrens 	refstr_rele(pathref);
491fa9e4066Sahrens 	vfs_setmntpoint(vfsp, newpath);
492fa9e4066Sahrens 
493fa9e4066Sahrens 	pathref = vfs_getresource(vfsp);
4940b69c2f0Sahrens 	(void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
4950b69c2f0Sahrens 	VERIFY((tail = strrchr(newpath, '@')) != NULL);
4960b69c2f0Sahrens 	*(tail+1) = '\0';
4970b69c2f0Sahrens 	ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
498fa9e4066Sahrens 	(void) strcat(newpath, nm);
499fa9e4066Sahrens 	refstr_rele(pathref);
500fa9e4066Sahrens 	vfs_setresource(vfsp, newpath);
501fa9e4066Sahrens 
502fa9e4066Sahrens 	vfs_unlock(vfsp);
503fa9e4066Sahrens }
504fa9e4066Sahrens 
505fa9e4066Sahrens static int
506fa9e4066Sahrens zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm,
507fa9e4066Sahrens     cred_t *cr)
508fa9e4066Sahrens {
509fa9e4066Sahrens 	zfsctl_snapdir_t *sdp = sdvp->v_data;
510fa9e4066Sahrens 	zfs_snapentry_t search, *sep;
511fa9e4066Sahrens 	avl_index_t where;
512fa9e4066Sahrens 	char from[MAXNAMELEN], to[MAXNAMELEN];
513fa9e4066Sahrens 	int err;
514fa9e4066Sahrens 
5155422900bSmaybee 	err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from);
5165422900bSmaybee 	if (err)
5175422900bSmaybee 		return (err);
518*ecd6cf80Smarks 
519*ecd6cf80Smarks 	err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to);
520fa9e4066Sahrens 	if (err)
521fa9e4066Sahrens 		return (err);
522fa9e4066Sahrens 
523*ecd6cf80Smarks 	if (err = zfs_secpolicy_rename_perms(from, to, cr))
524*ecd6cf80Smarks 		return (err);
525fa9e4066Sahrens 	/*
526fa9e4066Sahrens 	 * Cannot move snapshots out of the snapdir.
527fa9e4066Sahrens 	 */
528fa9e4066Sahrens 	if (sdvp != tdvp)
529fa9e4066Sahrens 		return (EINVAL);
530fa9e4066Sahrens 
531fa9e4066Sahrens 	if (strcmp(snm, tnm) == 0)
532fa9e4066Sahrens 		return (0);
533fa9e4066Sahrens 
534fa9e4066Sahrens 	mutex_enter(&sdp->sd_lock);
535fa9e4066Sahrens 
536fa9e4066Sahrens 	search.se_name = (char *)snm;
5375422900bSmaybee 	if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) {
5385422900bSmaybee 		mutex_exit(&sdp->sd_lock);
5395422900bSmaybee 		return (ENOENT);
540fa9e4066Sahrens 	}
541fa9e4066Sahrens 
542cdf5b4caSmmusante 	err = dmu_objset_rename(from, to, B_FALSE);
5435422900bSmaybee 	if (err == 0)
5445422900bSmaybee 		zfsctl_rename_snap(sdp, sep, tnm);
545fa9e4066Sahrens 
546fa9e4066Sahrens 	mutex_exit(&sdp->sd_lock);
547fa9e4066Sahrens 
548fa9e4066Sahrens 	return (err);
549fa9e4066Sahrens }
550fa9e4066Sahrens 
551fa9e4066Sahrens /* ARGSUSED */
552fa9e4066Sahrens static int
553fa9e4066Sahrens zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr)
554fa9e4066Sahrens {
555fa9e4066Sahrens 	zfsctl_snapdir_t *sdp = dvp->v_data;
556fa9e4066Sahrens 	char snapname[MAXNAMELEN];
557fa9e4066Sahrens 	int err;
558fa9e4066Sahrens 
5595422900bSmaybee 	err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname);
5605422900bSmaybee 	if (err)
5615422900bSmaybee 		return (err);
562*ecd6cf80Smarks 
563*ecd6cf80Smarks 	if (err = zfs_secpolicy_destroy_perms(snapname, cr))
564fa9e4066Sahrens 		return (err);
565fa9e4066Sahrens 
566fa9e4066Sahrens 	mutex_enter(&sdp->sd_lock);
567fa9e4066Sahrens 
568*ecd6cf80Smarks 	err = zfsctl_unmount_snap(dvp, name, MS_FORCE, cr);
569fa9e4066Sahrens 	if (err) {
570fa9e4066Sahrens 		mutex_exit(&sdp->sd_lock);
571fa9e4066Sahrens 		return (err);
572fa9e4066Sahrens 	}
573fa9e4066Sahrens 
574fa9e4066Sahrens 	err = dmu_objset_destroy(snapname);
575fa9e4066Sahrens 
576fa9e4066Sahrens 	mutex_exit(&sdp->sd_lock);
577fa9e4066Sahrens 
578fa9e4066Sahrens 	return (err);
579fa9e4066Sahrens }
580fa9e4066Sahrens 
581*ecd6cf80Smarks /* ARGSUSED */
582*ecd6cf80Smarks static int
583*ecd6cf80Smarks zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t  **vpp,
584*ecd6cf80Smarks     cred_t *cr)
585*ecd6cf80Smarks {
586*ecd6cf80Smarks 	zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
587*ecd6cf80Smarks 	char name[MAXNAMELEN];
588*ecd6cf80Smarks 	int err;
589*ecd6cf80Smarks 	static enum symfollow follow = NO_FOLLOW;
590*ecd6cf80Smarks 	static enum uio_seg seg = UIO_SYSSPACE;
591*ecd6cf80Smarks 
592*ecd6cf80Smarks 	dmu_objset_name(zfsvfs->z_os, name);
593*ecd6cf80Smarks 
594*ecd6cf80Smarks 	*vpp = NULL;
595*ecd6cf80Smarks 
596*ecd6cf80Smarks 	err = zfs_secpolicy_snapshot_perms(name, cr);
597*ecd6cf80Smarks 	if (err)
598*ecd6cf80Smarks 		return (err);
599*ecd6cf80Smarks 
600*ecd6cf80Smarks 	if (err == 0) {
601*ecd6cf80Smarks 		err = dmu_objset_snapshot(name, dirname, B_FALSE);
602*ecd6cf80Smarks 		if (err)
603*ecd6cf80Smarks 			return (err);
604*ecd6cf80Smarks 		err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp);
605*ecd6cf80Smarks 	}
606*ecd6cf80Smarks 
607*ecd6cf80Smarks 	return (err);
608*ecd6cf80Smarks }
609*ecd6cf80Smarks 
610fa9e4066Sahrens /*
611fa9e4066Sahrens  * Lookup entry point for the 'snapshot' directory.  Try to open the
612fa9e4066Sahrens  * snapshot if it exist, creating the pseudo filesystem vnode as necessary.
613fa9e4066Sahrens  * Perform a mount of the associated dataset on top of the vnode.
614fa9e4066Sahrens  */
615fa9e4066Sahrens /* ARGSUSED */
616fa9e4066Sahrens static int
617fa9e4066Sahrens zfsctl_snapdir_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
618fa9e4066Sahrens     int flags, vnode_t *rdir, cred_t *cr)
619fa9e4066Sahrens {
620fa9e4066Sahrens 	zfsctl_snapdir_t *sdp = dvp->v_data;
621fa9e4066Sahrens 	objset_t *snap;
622fa9e4066Sahrens 	char snapname[MAXNAMELEN];
623fa9e4066Sahrens 	char *mountpoint;
624fa9e4066Sahrens 	zfs_snapentry_t *sep, search;
625fa9e4066Sahrens 	struct mounta margs;
626fa9e4066Sahrens 	vfs_t *vfsp;
627fa9e4066Sahrens 	size_t mountpoint_len;
628fa9e4066Sahrens 	avl_index_t where;
629fa9e4066Sahrens 	zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
630fa9e4066Sahrens 	int err;
631fa9e4066Sahrens 
632fa9e4066Sahrens 	ASSERT(dvp->v_type == VDIR);
633fa9e4066Sahrens 
634fa9e4066Sahrens 	if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0)
635fa9e4066Sahrens 		return (0);
636fa9e4066Sahrens 
637fa9e4066Sahrens 	/*
638fa9e4066Sahrens 	 * If we get a recursive call, that means we got called
639fa9e4066Sahrens 	 * from the domount() code while it was trying to look up the
640fa9e4066Sahrens 	 * spec (which looks like a local path for zfs).  We need to
641fa9e4066Sahrens 	 * add some flag to domount() to tell it not to do this lookup.
642fa9e4066Sahrens 	 */
643fa9e4066Sahrens 	if (MUTEX_HELD(&sdp->sd_lock))
644fa9e4066Sahrens 		return (ENOENT);
645fa9e4066Sahrens 
646fa9e4066Sahrens 	ZFS_ENTER(zfsvfs);
647fa9e4066Sahrens 
648fa9e4066Sahrens 	mutex_enter(&sdp->sd_lock);
649fa9e4066Sahrens 	search.se_name = (char *)nm;
650fa9e4066Sahrens 	if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) {
651fa9e4066Sahrens 		*vpp = sep->se_root;
652fa9e4066Sahrens 		VN_HOLD(*vpp);
653bbf4a8dfSmaybee 		err = traverse(vpp);
654bbf4a8dfSmaybee 		if (err) {
655bbf4a8dfSmaybee 			VN_RELE(*vpp);
656bbf4a8dfSmaybee 			*vpp = NULL;
657bbf4a8dfSmaybee 		} else if (*vpp == sep->se_root) {
658bbf4a8dfSmaybee 			/*
659bbf4a8dfSmaybee 			 * The snapshot was unmounted behind our backs,
660bbf4a8dfSmaybee 			 * try to remount it.
661bbf4a8dfSmaybee 			 */
662fa9e4066Sahrens 			goto domount;
663bb963f1cSmaybee 		}
664fa9e4066Sahrens 		mutex_exit(&sdp->sd_lock);
665fa9e4066Sahrens 		ZFS_EXIT(zfsvfs);
666bbf4a8dfSmaybee 		return (err);
667fa9e4066Sahrens 	}
668fa9e4066Sahrens 
669fa9e4066Sahrens 	/*
670fa9e4066Sahrens 	 * The requested snapshot is not currently mounted, look it up.
671fa9e4066Sahrens 	 */
6725422900bSmaybee 	err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname);
6735422900bSmaybee 	if (err) {
6745422900bSmaybee 		mutex_exit(&sdp->sd_lock);
6755422900bSmaybee 		ZFS_EXIT(zfsvfs);
6765422900bSmaybee 		return (err);
6775422900bSmaybee 	}
678fa9e4066Sahrens 	if (dmu_objset_open(snapname, DMU_OST_ZFS,
679fa9e4066Sahrens 	    DS_MODE_STANDARD | DS_MODE_READONLY, &snap) != 0) {
680fa9e4066Sahrens 		mutex_exit(&sdp->sd_lock);
681fa9e4066Sahrens 		ZFS_EXIT(zfsvfs);
682fa9e4066Sahrens 		return (ENOENT);
683fa9e4066Sahrens 	}
684fa9e4066Sahrens 
685fa9e4066Sahrens 	sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP);
686fa9e4066Sahrens 	sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
687fa9e4066Sahrens 	(void) strcpy(sep->se_name, nm);
688fa9e4066Sahrens 	*vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap));
689fa9e4066Sahrens 	avl_insert(&sdp->sd_snaps, sep, where);
690fa9e4066Sahrens 
691fa9e4066Sahrens 	dmu_objset_close(snap);
692fa9e4066Sahrens domount:
693fa9e4066Sahrens 	mountpoint_len = strlen(refstr_value(dvp->v_vfsp->vfs_mntpt)) +
694fa9e4066Sahrens 	    strlen("/.zfs/snapshot/") + strlen(nm) + 1;
695fa9e4066Sahrens 	mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP);
696fa9e4066Sahrens 	(void) snprintf(mountpoint, mountpoint_len, "%s/.zfs/snapshot/%s",
697fa9e4066Sahrens 	    refstr_value(dvp->v_vfsp->vfs_mntpt), nm);
698fa9e4066Sahrens 
699fa9e4066Sahrens 	margs.spec = snapname;
700fa9e4066Sahrens 	margs.dir = mountpoint;
701fa9e4066Sahrens 	margs.flags = MS_SYSSPACE | MS_NOMNTTAB;
702fa9e4066Sahrens 	margs.fstype = "zfs";
703fa9e4066Sahrens 	margs.dataptr = NULL;
704fa9e4066Sahrens 	margs.datalen = 0;
705fa9e4066Sahrens 	margs.optptr = NULL;
706fa9e4066Sahrens 	margs.optlen = 0;
707fa9e4066Sahrens 
708fa9e4066Sahrens 	err = domount("zfs", &margs, *vpp, kcred, &vfsp);
709fa9e4066Sahrens 	kmem_free(mountpoint, mountpoint_len);
710fa9e4066Sahrens 
71195c9592aSmaybee 	if (err == 0) {
71295c9592aSmaybee 		/*
71395c9592aSmaybee 		 * Return the mounted root rather than the covered mount point.
71495c9592aSmaybee 		 */
71595c9592aSmaybee 		VFS_RELE(vfsp);
71695c9592aSmaybee 		err = traverse(vpp);
71795c9592aSmaybee 	}
718fa9e4066Sahrens 
71995c9592aSmaybee 	if (err == 0) {
72095c9592aSmaybee 		/*
72195c9592aSmaybee 		 * Fix up the root vnode.
72295c9592aSmaybee 		 */
72395c9592aSmaybee 		ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs);
72495c9592aSmaybee 		VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs;
72595c9592aSmaybee 		(*vpp)->v_vfsp = zfsvfs->z_vfs;
72695c9592aSmaybee 		(*vpp)->v_flag &= ~VROOT;
72795c9592aSmaybee 	}
728fa9e4066Sahrens 	mutex_exit(&sdp->sd_lock);
729fa9e4066Sahrens 	ZFS_EXIT(zfsvfs);
730fa9e4066Sahrens 
731bb963f1cSmaybee 	/*
732bb963f1cSmaybee 	 * If we had an error, drop our hold on the vnode and
733bb963f1cSmaybee 	 * zfsctl_snapshot_inactive() will clean up.
734bb963f1cSmaybee 	 */
735bb963f1cSmaybee 	if (err) {
73695c9592aSmaybee 		VN_RELE(*vpp);
737bb963f1cSmaybee 		*vpp = NULL;
738bb963f1cSmaybee 	}
73995c9592aSmaybee 	return (err);
740fa9e4066Sahrens }
741fa9e4066Sahrens 
742fa9e4066Sahrens /* ARGSUSED */
743fa9e4066Sahrens static int
744fa9e4066Sahrens zfsctl_snapdir_readdir_cb(vnode_t *vp, struct dirent64 *dp, int *eofp,
745fa9e4066Sahrens     offset_t *offp, offset_t *nextp, void *data)
746fa9e4066Sahrens {
747fa9e4066Sahrens 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
748fa9e4066Sahrens 	char snapname[MAXNAMELEN];
749fa9e4066Sahrens 	uint64_t id, cookie;
750fa9e4066Sahrens 
751fa9e4066Sahrens 	ZFS_ENTER(zfsvfs);
752fa9e4066Sahrens 
753fa9e4066Sahrens 	cookie = *offp;
754fa9e4066Sahrens 	if (dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id,
755fa9e4066Sahrens 	    &cookie) == ENOENT) {
756fa9e4066Sahrens 		*eofp = 1;
757fa9e4066Sahrens 		ZFS_EXIT(zfsvfs);
758fa9e4066Sahrens 		return (0);
759fa9e4066Sahrens 	}
760fa9e4066Sahrens 
761fa9e4066Sahrens 	(void) strcpy(dp->d_name, snapname);
762fa9e4066Sahrens 	dp->d_ino = ZFSCTL_INO_SNAP(id);
763fa9e4066Sahrens 	*nextp = cookie;
764fa9e4066Sahrens 
765fa9e4066Sahrens 	ZFS_EXIT(zfsvfs);
766fa9e4066Sahrens 
767fa9e4066Sahrens 	return (0);
768fa9e4066Sahrens }
769fa9e4066Sahrens 
770fa9e4066Sahrens vnode_t *
771fa9e4066Sahrens zfsctl_mknode_snapdir(vnode_t *pvp)
772fa9e4066Sahrens {
773fa9e4066Sahrens 	vnode_t *vp;
774fa9e4066Sahrens 	zfsctl_snapdir_t *sdp;
775fa9e4066Sahrens 
776fa9e4066Sahrens 	vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp,
777fa9e4066Sahrens 	    zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN,
778fa9e4066Sahrens 	    zfsctl_snapdir_readdir_cb, NULL);
779fa9e4066Sahrens 	sdp = vp->v_data;
780fa9e4066Sahrens 	sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR;
781a32fded1Sek 	sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
782fa9e4066Sahrens 	mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL);
783fa9e4066Sahrens 	avl_create(&sdp->sd_snaps, snapentry_compare,
784fa9e4066Sahrens 	    sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node));
785fa9e4066Sahrens 	return (vp);
786fa9e4066Sahrens }
787fa9e4066Sahrens 
788fa9e4066Sahrens /* ARGSUSED */
789fa9e4066Sahrens static int
790fa9e4066Sahrens zfsctl_snapdir_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr)
791fa9e4066Sahrens {
792fa9e4066Sahrens 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
793fa9e4066Sahrens 	zfsctl_snapdir_t *sdp = vp->v_data;
794fa9e4066Sahrens 
795fa9e4066Sahrens 	ZFS_ENTER(zfsvfs);
796fa9e4066Sahrens 	zfsctl_common_getattr(vp, vap);
797fa9e4066Sahrens 	vap->va_nodeid = gfs_file_inode(vp);
798fa9e4066Sahrens 	vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2;
799fa9e4066Sahrens 	ZFS_EXIT(zfsvfs);
800fa9e4066Sahrens 
801fa9e4066Sahrens 	return (0);
802fa9e4066Sahrens }
803fa9e4066Sahrens 
804bb963f1cSmaybee /* ARGSUSED */
805fa9e4066Sahrens static void
806fa9e4066Sahrens zfsctl_snapdir_inactive(vnode_t *vp, cred_t *cr)
807fa9e4066Sahrens {
808fa9e4066Sahrens 	zfsctl_snapdir_t *sdp = vp->v_data;
809bb963f1cSmaybee 	void *private;
810bb963f1cSmaybee 
811bb963f1cSmaybee 	private = gfs_dir_inactive(vp);
812bb963f1cSmaybee 	if (private != NULL) {
813bb963f1cSmaybee 		ASSERT(avl_numnodes(&sdp->sd_snaps) == 0);
814bb963f1cSmaybee 		mutex_destroy(&sdp->sd_lock);
815bb963f1cSmaybee 		avl_destroy(&sdp->sd_snaps);
816bb963f1cSmaybee 		kmem_free(private, sizeof (zfsctl_snapdir_t));
817bb963f1cSmaybee 	}
818fa9e4066Sahrens }
819fa9e4066Sahrens 
820fa9e4066Sahrens static const fs_operation_def_t zfsctl_tops_snapdir[] = {
821aa59c4cbSrsb 	{ VOPNAME_OPEN,		{ .vop_open = zfsctl_common_open }	},
822aa59c4cbSrsb 	{ VOPNAME_CLOSE,	{ .vop_close = zfsctl_common_close }	},
823aa59c4cbSrsb 	{ VOPNAME_IOCTL,	{ .error = fs_inval }			},
824aa59c4cbSrsb 	{ VOPNAME_GETATTR,	{ .vop_getattr = zfsctl_snapdir_getattr } },
825aa59c4cbSrsb 	{ VOPNAME_ACCESS,	{ .vop_access = zfsctl_common_access }	},
826aa59c4cbSrsb 	{ VOPNAME_RENAME,	{ .vop_rename = zfsctl_snapdir_rename }	},
827aa59c4cbSrsb 	{ VOPNAME_RMDIR,	{ .vop_rmdir = zfsctl_snapdir_remove }	},
828*ecd6cf80Smarks 	{ VOPNAME_MKDIR,	{ .vop_mkdir = zfsctl_snapdir_mkdir }	},
829aa59c4cbSrsb 	{ VOPNAME_READDIR,	{ .vop_readdir = gfs_vop_readdir }	},
830aa59c4cbSrsb 	{ VOPNAME_LOOKUP,	{ .vop_lookup = zfsctl_snapdir_lookup }	},
831aa59c4cbSrsb 	{ VOPNAME_SEEK,		{ .vop_seek = fs_seek }			},
832aa59c4cbSrsb 	{ VOPNAME_INACTIVE,	{ .vop_inactive = zfsctl_snapdir_inactive } },
833aa59c4cbSrsb 	{ VOPNAME_FID,		{ .vop_fid = zfsctl_common_fid }	},
834fa9e4066Sahrens 	{ NULL }
835fa9e4066Sahrens };
836fa9e4066Sahrens 
837fa9e4066Sahrens static vnode_t *
838fa9e4066Sahrens zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset)
839fa9e4066Sahrens {
840fa9e4066Sahrens 	vnode_t *vp;
841fa9e4066Sahrens 	zfsctl_node_t *zcp;
842fa9e4066Sahrens 
843fa9e4066Sahrens 	vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp,
844fa9e4066Sahrens 	    zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL);
845fa9e4066Sahrens 	zcp = vp->v_data;
846fa9e4066Sahrens 	zcp->zc_id = objset;
847fa9e4066Sahrens 
848fa9e4066Sahrens 	return (vp);
849fa9e4066Sahrens }
850fa9e4066Sahrens 
851fa9e4066Sahrens static void
852fa9e4066Sahrens zfsctl_snapshot_inactive(vnode_t *vp, cred_t *cr)
853fa9e4066Sahrens {
854fa9e4066Sahrens 	zfsctl_snapdir_t *sdp;
855fa9e4066Sahrens 	zfs_snapentry_t *sep, *next;
856fa9e4066Sahrens 	vnode_t *dvp;
857fa9e4066Sahrens 
858fa9e4066Sahrens 	VERIFY(gfs_dir_lookup(vp, "..", &dvp) == 0);
859fa9e4066Sahrens 	sdp = dvp->v_data;
860fa9e4066Sahrens 
861fa9e4066Sahrens 	mutex_enter(&sdp->sd_lock);
862fa9e4066Sahrens 
863fa9e4066Sahrens 	if (vp->v_count > 1) {
864fa9e4066Sahrens 		mutex_exit(&sdp->sd_lock);
865fa9e4066Sahrens 		return;
866fa9e4066Sahrens 	}
867fa9e4066Sahrens 	ASSERT(!vn_ismntpt(vp));
868fa9e4066Sahrens 
869fa9e4066Sahrens 	sep = avl_first(&sdp->sd_snaps);
870fa9e4066Sahrens 	while (sep != NULL) {
871fa9e4066Sahrens 		next = AVL_NEXT(&sdp->sd_snaps, sep);
872fa9e4066Sahrens 
873fa9e4066Sahrens 		if (sep->se_root == vp) {
874fa9e4066Sahrens 			avl_remove(&sdp->sd_snaps, sep);
875fa9e4066Sahrens 			kmem_free(sep->se_name, strlen(sep->se_name) + 1);
876fa9e4066Sahrens 			kmem_free(sep, sizeof (zfs_snapentry_t));
877fa9e4066Sahrens 			break;
878fa9e4066Sahrens 		}
879fa9e4066Sahrens 		sep = next;
880fa9e4066Sahrens 	}
881fa9e4066Sahrens 	ASSERT(sep != NULL);
882fa9e4066Sahrens 
883fa9e4066Sahrens 	mutex_exit(&sdp->sd_lock);
884fa9e4066Sahrens 	VN_RELE(dvp);
885fa9e4066Sahrens 
886bb963f1cSmaybee 	/*
887bb963f1cSmaybee 	 * Dispose of the vnode for the snapshot mount point.
888bb963f1cSmaybee 	 * This is safe to do because once this entry has been removed
889bb963f1cSmaybee 	 * from the AVL tree, it can't be found again, so cannot become
890bb963f1cSmaybee 	 * "active".  If we lookup the same name again we will end up
891bb963f1cSmaybee 	 * creating a new vnode.
892bb963f1cSmaybee 	 */
893fa9e4066Sahrens 	gfs_vop_inactive(vp, cr);
894fa9e4066Sahrens }
895fa9e4066Sahrens 
896fa9e4066Sahrens 
897fa9e4066Sahrens /*
898fa9e4066Sahrens  * These VP's should never see the light of day.  They should always
899fa9e4066Sahrens  * be covered.
900fa9e4066Sahrens  */
901fa9e4066Sahrens static const fs_operation_def_t zfsctl_tops_snapshot[] = {
902aa59c4cbSrsb 	VOPNAME_INACTIVE, { .vop_inactive =  zfsctl_snapshot_inactive },
903fa9e4066Sahrens 	NULL, NULL
904fa9e4066Sahrens };
905fa9e4066Sahrens 
906fa9e4066Sahrens int
907fa9e4066Sahrens zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp)
908fa9e4066Sahrens {
909fa9e4066Sahrens 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
910fa9e4066Sahrens 	vnode_t *dvp, *vp;
911fa9e4066Sahrens 	zfsctl_snapdir_t *sdp;
912fa9e4066Sahrens 	zfsctl_node_t *zcp;
913fa9e4066Sahrens 	zfs_snapentry_t *sep;
914fa9e4066Sahrens 	int error;
915fa9e4066Sahrens 
916fa9e4066Sahrens 	ASSERT(zfsvfs->z_ctldir != NULL);
917fa9e4066Sahrens 	error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
918fa9e4066Sahrens 	    NULL, 0, NULL, kcred);
919fa9e4066Sahrens 	if (error != 0)
920fa9e4066Sahrens 		return (error);
921fa9e4066Sahrens 	sdp = dvp->v_data;
922fa9e4066Sahrens 
923fa9e4066Sahrens 	mutex_enter(&sdp->sd_lock);
924fa9e4066Sahrens 	sep = avl_first(&sdp->sd_snaps);
925fa9e4066Sahrens 	while (sep != NULL) {
926fa9e4066Sahrens 		vp = sep->se_root;
927fa9e4066Sahrens 		zcp = vp->v_data;
928fa9e4066Sahrens 		if (zcp->zc_id == objsetid)
929fa9e4066Sahrens 			break;
930fa9e4066Sahrens 
931fa9e4066Sahrens 		sep = AVL_NEXT(&sdp->sd_snaps, sep);
932fa9e4066Sahrens 	}
933fa9e4066Sahrens 
934fa9e4066Sahrens 	if (sep != NULL) {
935fa9e4066Sahrens 		VN_HOLD(vp);
936fa9e4066Sahrens 		error = traverse(&vp);
937bbf4a8dfSmaybee 		if (error == 0) {
938bbf4a8dfSmaybee 			if (vp == sep->se_root)
939bbf4a8dfSmaybee 				error = EINVAL;
940bbf4a8dfSmaybee 			else
941bbf4a8dfSmaybee 				*zfsvfsp = VTOZ(vp)->z_zfsvfs;
942bbf4a8dfSmaybee 		}
943fea70b54Snd 		mutex_exit(&sdp->sd_lock);
944fa9e4066Sahrens 		VN_RELE(vp);
945fa9e4066Sahrens 	} else {
946fa9e4066Sahrens 		error = EINVAL;
947fea70b54Snd 		mutex_exit(&sdp->sd_lock);
948fa9e4066Sahrens 	}
949fa9e4066Sahrens 
950fa9e4066Sahrens 	VN_RELE(dvp);
951fa9e4066Sahrens 
952fa9e4066Sahrens 	return (error);
953fa9e4066Sahrens }
954fa9e4066Sahrens 
955fa9e4066Sahrens /*
956fa9e4066Sahrens  * Unmount any snapshots for the given filesystem.  This is called from
957fa9e4066Sahrens  * zfs_umount() - if we have a ctldir, then go through and unmount all the
958fa9e4066Sahrens  * snapshots.
959fa9e4066Sahrens  */
960fa9e4066Sahrens int
961fa9e4066Sahrens zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr)
962fa9e4066Sahrens {
963fa9e4066Sahrens 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
964fa9e4066Sahrens 	vnode_t *dvp, *svp;
965fa9e4066Sahrens 	zfsctl_snapdir_t *sdp;
966fa9e4066Sahrens 	zfs_snapentry_t *sep, *next;
967fa9e4066Sahrens 	int error;
968fa9e4066Sahrens 
969fa9e4066Sahrens 	ASSERT(zfsvfs->z_ctldir != NULL);
970fa9e4066Sahrens 	error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
971fa9e4066Sahrens 	    NULL, 0, NULL, cr);
972fa9e4066Sahrens 	if (error != 0)
973fa9e4066Sahrens 		return (error);
974fa9e4066Sahrens 	sdp = dvp->v_data;
975fa9e4066Sahrens 
976fa9e4066Sahrens 	mutex_enter(&sdp->sd_lock);
977fa9e4066Sahrens 
978fa9e4066Sahrens 	sep = avl_first(&sdp->sd_snaps);
979fa9e4066Sahrens 	while (sep != NULL) {
980fa9e4066Sahrens 		svp = sep->se_root;
981fa9e4066Sahrens 		next = AVL_NEXT(&sdp->sd_snaps, sep);
982fa9e4066Sahrens 
983fa9e4066Sahrens 		/*
984fa9e4066Sahrens 		 * If this snapshot is not mounted, then it must
985fa9e4066Sahrens 		 * have just been unmounted by somebody else, and
986fa9e4066Sahrens 		 * will be cleaned up by zfsctl_snapdir_inactive().
987fa9e4066Sahrens 		 */
988fa9e4066Sahrens 		if (vn_ismntpt(svp)) {
989fa9e4066Sahrens 			if ((error = vn_vfswlock(svp)) != 0)
990fa9e4066Sahrens 				goto out;
991fa9e4066Sahrens 
992fa9e4066Sahrens 			VN_HOLD(svp);
993fa9e4066Sahrens 			error = dounmount(vn_mountedvfs(svp), fflags, cr);
994fa9e4066Sahrens 			if (error) {
995fa9e4066Sahrens 				VN_RELE(svp);
996fa9e4066Sahrens 				goto out;
997fa9e4066Sahrens 			}
998fa9e4066Sahrens 
999fa9e4066Sahrens 			avl_remove(&sdp->sd_snaps, sep);
1000fa9e4066Sahrens 			kmem_free(sep->se_name, strlen(sep->se_name) + 1);
1001fa9e4066Sahrens 			kmem_free(sep, sizeof (zfs_snapentry_t));
1002fa9e4066Sahrens 
1003fa9e4066Sahrens 			/*
1004fa9e4066Sahrens 			 * We can't use VN_RELE(), as that will try to
1005fa9e4066Sahrens 			 * invoke zfsctl_snapdir_inactive(), and that
1006fa9e4066Sahrens 			 * would lead to an attempt to re-grab the sd_lock.
1007fa9e4066Sahrens 			 */
1008fa9e4066Sahrens 			ASSERT3U(svp->v_count, ==, 1);
1009fa9e4066Sahrens 			gfs_vop_inactive(svp, cr);
1010fa9e4066Sahrens 		}
1011fa9e4066Sahrens 		sep = next;
1012fa9e4066Sahrens 	}
1013fa9e4066Sahrens out:
1014fa9e4066Sahrens 	mutex_exit(&sdp->sd_lock);
1015fa9e4066Sahrens 	VN_RELE(dvp);
1016fa9e4066Sahrens 
1017fa9e4066Sahrens 	return (error);
1018fa9e4066Sahrens }
1019