xref: /illumos-gate/usr/src/uts/common/fs/zfs/zfs_ctldir.c (revision 4b5c8e93cab28d3c65ba9d407fd8f46e3be1db1c)
1 /*
2  * CDDL HEADER START
3  *
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.
7  *
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.
12  *
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]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23  * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24  * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved.
25  */
26 
27 /*
28  * ZFS control directory (a.k.a. ".zfs")
29  *
30  * This directory provides a common location for all ZFS meta-objects.
31  * Currently, this is only the 'snapshot' directory, but this may expand in the
32  * future.  The elements are built using the GFS primitives, as the hierarchy
33  * does not actually exist on disk.
34  *
35  * For 'snapshot', we don't want to have all snapshots always mounted, because
36  * this would take up a huge amount of space in /etc/mnttab.  We have three
37  * types of objects:
38  *
39  * 	ctldir ------> snapshotdir -------> snapshot
40  *                                             |
41  *                                             |
42  *                                             V
43  *                                         mounted fs
44  *
45  * The 'snapshot' node contains just enough information to lookup '..' and act
46  * as a mountpoint for the snapshot.  Whenever we lookup a specific snapshot, we
47  * perform an automount of the underlying filesystem and return the
48  * corresponding vnode.
49  *
50  * All mounts are handled automatically by the kernel, but unmounts are
51  * (currently) handled from user land.  The main reason is that there is no
52  * reliable way to auto-unmount the filesystem when it's "no longer in use".
53  * When the user unmounts a filesystem, we call zfsctl_unmount(), which
54  * unmounts any snapshots within the snapshot directory.
55  *
56  * The '.zfs', '.zfs/snapshot', and all directories created under
57  * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') are all GFS nodes and
58  * share the same vfs_t as the head filesystem (what '.zfs' lives under).
59  *
60  * File systems mounted ontop of the GFS nodes '.zfs/snapshot/<snapname>'
61  * (ie: snapshots) are ZFS nodes and have their own unique vfs_t.
62  * However, vnodes within these mounted on file systems have their v_vfsp
63  * fields set to the head filesystem to make NFS happy (see
64  * zfsctl_snapdir_lookup()). We VFS_HOLD the head filesystem's vfs_t
65  * so that it cannot be freed until all snapshots have been unmounted.
66  */
67 
68 #include <fs/fs_subr.h>
69 #include <sys/zfs_ctldir.h>
70 #include <sys/zfs_ioctl.h>
71 #include <sys/zfs_vfsops.h>
72 #include <sys/vfs_opreg.h>
73 #include <sys/gfs.h>
74 #include <sys/stat.h>
75 #include <sys/dmu.h>
76 #include <sys/dsl_destroy.h>
77 #include <sys/dsl_deleg.h>
78 #include <sys/mount.h>
79 #include <sys/sunddi.h>
80 
81 #include "zfs_namecheck.h"
82 
83 typedef struct zfsctl_node {
84 	gfs_dir_t	zc_gfs_private;
85 	uint64_t	zc_id;
86 	timestruc_t	zc_cmtime;	/* ctime and mtime, always the same */
87 } zfsctl_node_t;
88 
89 typedef struct zfsctl_snapdir {
90 	zfsctl_node_t	sd_node;
91 	kmutex_t	sd_lock;
92 	avl_tree_t	sd_snaps;
93 } zfsctl_snapdir_t;
94 
95 typedef struct {
96 	char		*se_name;
97 	vnode_t		*se_root;
98 	avl_node_t	se_node;
99 } zfs_snapentry_t;
100 
101 static int
102 snapentry_compare(const void *a, const void *b)
103 {
104 	const zfs_snapentry_t *sa = a;
105 	const zfs_snapentry_t *sb = b;
106 	int ret = strcmp(sa->se_name, sb->se_name);
107 
108 	if (ret < 0)
109 		return (-1);
110 	else if (ret > 0)
111 		return (1);
112 	else
113 		return (0);
114 }
115 
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;
121 
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[];
126 
127 static vnode_t *zfsctl_mknode_snapdir(vnode_t *);
128 static vnode_t *zfsctl_mknode_shares(vnode_t *);
129 static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset);
130 static int zfsctl_unmount_snap(zfs_snapentry_t *, int, cred_t *);
131 
132 static gfs_opsvec_t zfsctl_opsvec[] = {
133 	{ ".zfs", zfsctl_tops_root, &zfsctl_ops_root },
134 	{ ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir },
135 	{ ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot },
136 	{ ".zfs/shares", zfsctl_tops_shares, &zfsctl_ops_shares_dir },
137 	{ ".zfs/shares/vnode", zfsctl_tops_shares, &zfsctl_ops_shares },
138 	{ NULL }
139 };
140 
141 /*
142  * Root directory elements.  We only have two entries
143  * snapshot and shares.
144  */
145 static gfs_dirent_t zfsctl_root_entries[] = {
146 	{ "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE },
147 	{ "shares", zfsctl_mknode_shares, GFS_CACHE_VNODE },
148 	{ NULL }
149 };
150 
151 /* include . and .. in the calculation */
152 #define	NROOT_ENTRIES	((sizeof (zfsctl_root_entries) / \
153     sizeof (gfs_dirent_t)) + 1)
154 
155 
156 /*
157  * Initialize the various GFS pieces we'll need to create and manipulate .zfs
158  * directories.  This is called from the ZFS init routine, and initializes the
159  * vnode ops vectors that we'll be using.
160  */
161 void
162 zfsctl_init(void)
163 {
164 	VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0);
165 }
166 
167 void
168 zfsctl_fini(void)
169 {
170 	/*
171 	 * Remove vfsctl vnode ops
172 	 */
173 	if (zfsctl_ops_root)
174 		vn_freevnodeops(zfsctl_ops_root);
175 	if (zfsctl_ops_snapdir)
176 		vn_freevnodeops(zfsctl_ops_snapdir);
177 	if (zfsctl_ops_snapshot)
178 		vn_freevnodeops(zfsctl_ops_snapshot);
179 	if (zfsctl_ops_shares)
180 		vn_freevnodeops(zfsctl_ops_shares);
181 	if (zfsctl_ops_shares_dir)
182 		vn_freevnodeops(zfsctl_ops_shares_dir);
183 
184 	zfsctl_ops_root = NULL;
185 	zfsctl_ops_snapdir = NULL;
186 	zfsctl_ops_snapshot = NULL;
187 	zfsctl_ops_shares = NULL;
188 	zfsctl_ops_shares_dir = NULL;
189 }
190 
191 boolean_t
192 zfsctl_is_node(vnode_t *vp)
193 {
194 	return (vn_matchops(vp, zfsctl_ops_root) ||
195 	    vn_matchops(vp, zfsctl_ops_snapdir) ||
196 	    vn_matchops(vp, zfsctl_ops_snapshot) ||
197 	    vn_matchops(vp, zfsctl_ops_shares) ||
198 	    vn_matchops(vp, zfsctl_ops_shares_dir));
199 
200 }
201 
202 /*
203  * Return the inode number associated with the 'snapshot' or
204  * 'shares' directory.
205  */
206 /* ARGSUSED */
207 static ino64_t
208 zfsctl_root_inode_cb(vnode_t *vp, int index)
209 {
210 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
211 
212 	ASSERT(index <= 2);
213 
214 	if (index == 0)
215 		return (ZFSCTL_INO_SNAPDIR);
216 
217 	return (zfsvfs->z_shares_dir);
218 }
219 
220 /*
221  * Create the '.zfs' directory.  This directory is cached as part of the VFS
222  * structure.  This results in a hold on the vfs_t.  The code in zfs_umount()
223  * therefore checks against a vfs_count of 2 instead of 1.  This reference
224  * is removed when the ctldir is destroyed in the unmount.
225  */
226 void
227 zfsctl_create(zfsvfs_t *zfsvfs)
228 {
229 	vnode_t *vp, *rvp;
230 	zfsctl_node_t *zcp;
231 	uint64_t crtime[2];
232 
233 	ASSERT(zfsvfs->z_ctldir == NULL);
234 
235 	vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs,
236 	    zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries,
237 	    zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL);
238 	zcp = vp->v_data;
239 	zcp->zc_id = ZFSCTL_INO_ROOT;
240 
241 	VERIFY(VFS_ROOT(zfsvfs->z_vfs, &rvp) == 0);
242 	VERIFY(0 == sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
243 	    &crtime, sizeof (crtime)));
244 	ZFS_TIME_DECODE(&zcp->zc_cmtime, crtime);
245 	VN_RELE(rvp);
246 
247 	/*
248 	 * We're only faking the fact that we have a root of a filesystem for
249 	 * the sake of the GFS interfaces.  Undo the flag manipulation it did
250 	 * for us.
251 	 */
252 	vp->v_flag &= ~(VROOT | VNOCACHE | VNOMAP | VNOSWAP | VNOMOUNT);
253 
254 	zfsvfs->z_ctldir = vp;
255 }
256 
257 /*
258  * Destroy the '.zfs' directory.  Only called when the filesystem is unmounted.
259  * There might still be more references if we were force unmounted, but only
260  * new zfs_inactive() calls can occur and they don't reference .zfs
261  */
262 void
263 zfsctl_destroy(zfsvfs_t *zfsvfs)
264 {
265 	VN_RELE(zfsvfs->z_ctldir);
266 	zfsvfs->z_ctldir = NULL;
267 }
268 
269 /*
270  * Given a root znode, retrieve the associated .zfs directory.
271  * Add a hold to the vnode and return it.
272  */
273 vnode_t *
274 zfsctl_root(znode_t *zp)
275 {
276 	ASSERT(zfs_has_ctldir(zp));
277 	VN_HOLD(zp->z_zfsvfs->z_ctldir);
278 	return (zp->z_zfsvfs->z_ctldir);
279 }
280 
281 /*
282  * Common open routine.  Disallow any write access.
283  */
284 /* ARGSUSED */
285 static int
286 zfsctl_common_open(vnode_t **vpp, int flags, cred_t *cr, caller_context_t *ct)
287 {
288 	if (flags & FWRITE)
289 		return (SET_ERROR(EACCES));
290 
291 	return (0);
292 }
293 
294 /*
295  * Common close routine.  Nothing to do here.
296  */
297 /* ARGSUSED */
298 static int
299 zfsctl_common_close(vnode_t *vpp, int flags, int count, offset_t off,
300     cred_t *cr, caller_context_t *ct)
301 {
302 	return (0);
303 }
304 
305 /*
306  * Common access routine.  Disallow writes.
307  */
308 /* ARGSUSED */
309 static int
310 zfsctl_common_access(vnode_t *vp, int mode, int flags, cred_t *cr,
311     caller_context_t *ct)
312 {
313 	if (flags & V_ACE_MASK) {
314 		if (mode & ACE_ALL_WRITE_PERMS)
315 			return (SET_ERROR(EACCES));
316 	} else {
317 		if (mode & VWRITE)
318 			return (SET_ERROR(EACCES));
319 	}
320 
321 	return (0);
322 }
323 
324 /*
325  * Common getattr function.  Fill in basic information.
326  */
327 static void
328 zfsctl_common_getattr(vnode_t *vp, vattr_t *vap)
329 {
330 	timestruc_t	now;
331 
332 	vap->va_uid = 0;
333 	vap->va_gid = 0;
334 	vap->va_rdev = 0;
335 	/*
336 	 * We are a purely virtual object, so we have no
337 	 * blocksize or allocated blocks.
338 	 */
339 	vap->va_blksize = 0;
340 	vap->va_nblocks = 0;
341 	vap->va_seq = 0;
342 	vap->va_fsid = vp->v_vfsp->vfs_dev;
343 	vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP |
344 	    S_IROTH | S_IXOTH;
345 	vap->va_type = VDIR;
346 	/*
347 	 * We live in the now (for atime).
348 	 */
349 	gethrestime(&now);
350 	vap->va_atime = now;
351 }
352 
353 /*ARGSUSED*/
354 static int
355 zfsctl_common_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
356 {
357 	zfsvfs_t	*zfsvfs = vp->v_vfsp->vfs_data;
358 	zfsctl_node_t	*zcp = vp->v_data;
359 	uint64_t	object = zcp->zc_id;
360 	zfid_short_t	*zfid;
361 	int		i;
362 
363 	ZFS_ENTER(zfsvfs);
364 
365 	if (fidp->fid_len < SHORT_FID_LEN) {
366 		fidp->fid_len = SHORT_FID_LEN;
367 		ZFS_EXIT(zfsvfs);
368 		return (SET_ERROR(ENOSPC));
369 	}
370 
371 	zfid = (zfid_short_t *)fidp;
372 
373 	zfid->zf_len = SHORT_FID_LEN;
374 
375 	for (i = 0; i < sizeof (zfid->zf_object); i++)
376 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
377 
378 	/* .zfs znodes always have a generation number of 0 */
379 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
380 		zfid->zf_gen[i] = 0;
381 
382 	ZFS_EXIT(zfsvfs);
383 	return (0);
384 }
385 
386 
387 /*ARGSUSED*/
388 static int
389 zfsctl_shares_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
390 {
391 	zfsvfs_t	*zfsvfs = vp->v_vfsp->vfs_data;
392 	znode_t		*dzp;
393 	int		error;
394 
395 	ZFS_ENTER(zfsvfs);
396 
397 	if (zfsvfs->z_shares_dir == 0) {
398 		ZFS_EXIT(zfsvfs);
399 		return (SET_ERROR(ENOTSUP));
400 	}
401 
402 	if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
403 		error = VOP_FID(ZTOV(dzp), fidp, ct);
404 		VN_RELE(ZTOV(dzp));
405 	}
406 
407 	ZFS_EXIT(zfsvfs);
408 	return (error);
409 }
410 /*
411  * .zfs inode namespace
412  *
413  * We need to generate unique inode numbers for all files and directories
414  * within the .zfs pseudo-filesystem.  We use the following scheme:
415  *
416  * 	ENTRY			ZFSCTL_INODE
417  * 	.zfs			1
418  * 	.zfs/snapshot		2
419  * 	.zfs/snapshot/<snap>	objectid(snap)
420  */
421 
422 #define	ZFSCTL_INO_SNAP(id)	(id)
423 
424 /*
425  * Get root directory attributes.
426  */
427 /* ARGSUSED */
428 static int
429 zfsctl_root_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
430     caller_context_t *ct)
431 {
432 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
433 	zfsctl_node_t *zcp = vp->v_data;
434 
435 	ZFS_ENTER(zfsvfs);
436 	vap->va_nodeid = ZFSCTL_INO_ROOT;
437 	vap->va_nlink = vap->va_size = NROOT_ENTRIES;
438 	vap->va_mtime = vap->va_ctime = zcp->zc_cmtime;
439 
440 	zfsctl_common_getattr(vp, vap);
441 	ZFS_EXIT(zfsvfs);
442 
443 	return (0);
444 }
445 
446 /*
447  * Special case the handling of "..".
448  */
449 /* ARGSUSED */
450 int
451 zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
452     int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
453     int *direntflags, pathname_t *realpnp)
454 {
455 	zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
456 	int err;
457 
458 	/*
459 	 * No extended attributes allowed under .zfs
460 	 */
461 	if (flags & LOOKUP_XATTR)
462 		return (SET_ERROR(EINVAL));
463 
464 	ZFS_ENTER(zfsvfs);
465 
466 	if (strcmp(nm, "..") == 0) {
467 		err = VFS_ROOT(dvp->v_vfsp, vpp);
468 	} else {
469 		err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir,
470 		    cr, ct, direntflags, realpnp);
471 	}
472 
473 	ZFS_EXIT(zfsvfs);
474 
475 	return (err);
476 }
477 
478 static int
479 zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
480     caller_context_t *ct)
481 {
482 	/*
483 	 * We only care about ACL_ENABLED so that libsec can
484 	 * display ACL correctly and not default to POSIX draft.
485 	 */
486 	if (cmd == _PC_ACL_ENABLED) {
487 		*valp = _ACL_ACE_ENABLED;
488 		return (0);
489 	}
490 
491 	return (fs_pathconf(vp, cmd, valp, cr, ct));
492 }
493 
494 static const fs_operation_def_t zfsctl_tops_root[] = {
495 	{ VOPNAME_OPEN,		{ .vop_open = zfsctl_common_open }	},
496 	{ VOPNAME_CLOSE,	{ .vop_close = zfsctl_common_close }	},
497 	{ VOPNAME_IOCTL,	{ .error = fs_inval }			},
498 	{ VOPNAME_GETATTR,	{ .vop_getattr = zfsctl_root_getattr }	},
499 	{ VOPNAME_ACCESS,	{ .vop_access = zfsctl_common_access }	},
500 	{ VOPNAME_READDIR,	{ .vop_readdir = gfs_vop_readdir } 	},
501 	{ VOPNAME_LOOKUP,	{ .vop_lookup = zfsctl_root_lookup }	},
502 	{ VOPNAME_SEEK,		{ .vop_seek = fs_seek }			},
503 	{ VOPNAME_INACTIVE,	{ .vop_inactive = gfs_vop_inactive }	},
504 	{ VOPNAME_PATHCONF,	{ .vop_pathconf = zfsctl_pathconf }	},
505 	{ VOPNAME_FID,		{ .vop_fid = zfsctl_common_fid	}	},
506 	{ NULL }
507 };
508 
509 /*
510  * Gets the full dataset name that corresponds to the given snapshot name
511  * Example:
512  * 	zfsctl_snapshot_zname("snap1") -> "mypool/myfs@snap1"
513  */
514 static int
515 zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname)
516 {
517 	objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
518 
519 	if (zfs_component_namecheck(name, NULL, NULL) != 0)
520 		return (SET_ERROR(EILSEQ));
521 	dmu_objset_name(os, zname);
522 	if (strlen(zname) + 1 + strlen(name) >= len)
523 		return (SET_ERROR(ENAMETOOLONG));
524 	(void) strcat(zname, "@");
525 	(void) strcat(zname, name);
526 	return (0);
527 }
528 
529 static int
530 zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr)
531 {
532 	vnode_t *svp = sep->se_root;
533 	int error;
534 
535 	ASSERT(vn_ismntpt(svp));
536 
537 	/* this will be dropped by dounmount() */
538 	if ((error = vn_vfswlock(svp)) != 0)
539 		return (error);
540 
541 	VN_HOLD(svp);
542 	error = dounmount(vn_mountedvfs(svp), fflags, cr);
543 	if (error) {
544 		VN_RELE(svp);
545 		return (error);
546 	}
547 
548 	/*
549 	 * We can't use VN_RELE(), as that will try to invoke
550 	 * zfsctl_snapdir_inactive(), which would cause us to destroy
551 	 * the sd_lock mutex held by our caller.
552 	 */
553 	ASSERT(svp->v_count == 1);
554 	gfs_vop_inactive(svp, cr, NULL);
555 
556 	kmem_free(sep->se_name, strlen(sep->se_name) + 1);
557 	kmem_free(sep, sizeof (zfs_snapentry_t));
558 
559 	return (0);
560 }
561 
562 static void
563 zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm)
564 {
565 	avl_index_t where;
566 	vfs_t *vfsp;
567 	refstr_t *pathref;
568 	char newpath[MAXNAMELEN];
569 	char *tail;
570 
571 	ASSERT(MUTEX_HELD(&sdp->sd_lock));
572 	ASSERT(sep != NULL);
573 
574 	vfsp = vn_mountedvfs(sep->se_root);
575 	ASSERT(vfsp != NULL);
576 
577 	vfs_lock_wait(vfsp);
578 
579 	/*
580 	 * Change the name in the AVL tree.
581 	 */
582 	avl_remove(&sdp->sd_snaps, sep);
583 	kmem_free(sep->se_name, strlen(sep->se_name) + 1);
584 	sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
585 	(void) strcpy(sep->se_name, nm);
586 	VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL);
587 	avl_insert(&sdp->sd_snaps, sep, where);
588 
589 	/*
590 	 * Change the current mountpoint info:
591 	 * 	- update the tail of the mntpoint path
592 	 *	- update the tail of the resource path
593 	 */
594 	pathref = vfs_getmntpoint(vfsp);
595 	(void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
596 	VERIFY((tail = strrchr(newpath, '/')) != NULL);
597 	*(tail+1) = '\0';
598 	ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
599 	(void) strcat(newpath, nm);
600 	refstr_rele(pathref);
601 	vfs_setmntpoint(vfsp, newpath, 0);
602 
603 	pathref = vfs_getresource(vfsp);
604 	(void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
605 	VERIFY((tail = strrchr(newpath, '@')) != NULL);
606 	*(tail+1) = '\0';
607 	ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
608 	(void) strcat(newpath, nm);
609 	refstr_rele(pathref);
610 	vfs_setresource(vfsp, newpath, 0);
611 
612 	vfs_unlock(vfsp);
613 }
614 
615 /*ARGSUSED*/
616 static int
617 zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm,
618     cred_t *cr, caller_context_t *ct, int flags)
619 {
620 	zfsctl_snapdir_t *sdp = sdvp->v_data;
621 	zfs_snapentry_t search, *sep;
622 	zfsvfs_t *zfsvfs;
623 	avl_index_t where;
624 	char from[ZFS_MAX_DATASET_NAME_LEN], to[ZFS_MAX_DATASET_NAME_LEN];
625 	char real[ZFS_MAX_DATASET_NAME_LEN], fsname[ZFS_MAX_DATASET_NAME_LEN];
626 	int err;
627 
628 	zfsvfs = sdvp->v_vfsp->vfs_data;
629 	ZFS_ENTER(zfsvfs);
630 
631 	if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
632 		err = dmu_snapshot_realname(zfsvfs->z_os, snm, real,
633 		    sizeof (real), NULL);
634 		if (err == 0) {
635 			snm = real;
636 		} else if (err != ENOTSUP) {
637 			ZFS_EXIT(zfsvfs);
638 			return (err);
639 		}
640 	}
641 
642 	ZFS_EXIT(zfsvfs);
643 
644 	dmu_objset_name(zfsvfs->z_os, fsname);
645 
646 	err = zfsctl_snapshot_zname(sdvp, snm, sizeof (from), from);
647 	if (err == 0)
648 		err = zfsctl_snapshot_zname(tdvp, tnm, sizeof (to), to);
649 	if (err == 0)
650 		err = zfs_secpolicy_rename_perms(from, to, cr);
651 	if (err != 0)
652 		return (err);
653 
654 	/*
655 	 * Cannot move snapshots out of the snapdir.
656 	 */
657 	if (sdvp != tdvp)
658 		return (SET_ERROR(EINVAL));
659 
660 	if (strcmp(snm, tnm) == 0)
661 		return (0);
662 
663 	mutex_enter(&sdp->sd_lock);
664 
665 	search.se_name = (char *)snm;
666 	if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) {
667 		mutex_exit(&sdp->sd_lock);
668 		return (SET_ERROR(ENOENT));
669 	}
670 
671 	err = dsl_dataset_rename_snapshot(fsname, snm, tnm, B_FALSE);
672 	if (err == 0)
673 		zfsctl_rename_snap(sdp, sep, tnm);
674 
675 	mutex_exit(&sdp->sd_lock);
676 
677 	return (err);
678 }
679 
680 /* ARGSUSED */
681 static int
682 zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
683     caller_context_t *ct, int flags)
684 {
685 	zfsctl_snapdir_t *sdp = dvp->v_data;
686 	zfs_snapentry_t *sep;
687 	zfs_snapentry_t search;
688 	zfsvfs_t *zfsvfs;
689 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
690 	char real[ZFS_MAX_DATASET_NAME_LEN];
691 	int err;
692 
693 	zfsvfs = dvp->v_vfsp->vfs_data;
694 	ZFS_ENTER(zfsvfs);
695 
696 	if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
697 
698 		err = dmu_snapshot_realname(zfsvfs->z_os, name, real,
699 		    sizeof (real), NULL);
700 		if (err == 0) {
701 			name = real;
702 		} else if (err != ENOTSUP) {
703 			ZFS_EXIT(zfsvfs);
704 			return (err);
705 		}
706 	}
707 
708 	ZFS_EXIT(zfsvfs);
709 
710 	err = zfsctl_snapshot_zname(dvp, name, sizeof (snapname), snapname);
711 	if (err == 0)
712 		err = zfs_secpolicy_destroy_perms(snapname, cr);
713 	if (err != 0)
714 		return (err);
715 
716 	mutex_enter(&sdp->sd_lock);
717 
718 	search.se_name = name;
719 	sep = avl_find(&sdp->sd_snaps, &search, NULL);
720 	if (sep) {
721 		avl_remove(&sdp->sd_snaps, sep);
722 		err = zfsctl_unmount_snap(sep, MS_FORCE, cr);
723 		if (err != 0)
724 			avl_add(&sdp->sd_snaps, sep);
725 		else
726 			err = dsl_destroy_snapshot(snapname, B_FALSE);
727 	} else {
728 		err = SET_ERROR(ENOENT);
729 	}
730 
731 	mutex_exit(&sdp->sd_lock);
732 
733 	return (err);
734 }
735 
736 /*
737  * This creates a snapshot under '.zfs/snapshot'.
738  */
739 /* ARGSUSED */
740 static int
741 zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t  **vpp,
742     cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp)
743 {
744 	zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
745 	char name[ZFS_MAX_DATASET_NAME_LEN];
746 	int err;
747 	static enum symfollow follow = NO_FOLLOW;
748 	static enum uio_seg seg = UIO_SYSSPACE;
749 
750 	if (zfs_component_namecheck(dirname, NULL, NULL) != 0)
751 		return (SET_ERROR(EILSEQ));
752 
753 	dmu_objset_name(zfsvfs->z_os, name);
754 
755 	*vpp = NULL;
756 
757 	err = zfs_secpolicy_snapshot_perms(name, cr);
758 	if (err != 0)
759 		return (err);
760 
761 	if (err == 0) {
762 		err = dmu_objset_snapshot_one(name, dirname);
763 		if (err != 0)
764 			return (err);
765 		err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp);
766 	}
767 
768 	return (err);
769 }
770 
771 /*
772  * Lookup entry point for the 'snapshot' directory.  Try to open the
773  * snapshot if it exist, creating the pseudo filesystem vnode as necessary.
774  * Perform a mount of the associated dataset on top of the vnode.
775  */
776 /* ARGSUSED */
777 static int
778 zfsctl_snapdir_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
779     int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
780     int *direntflags, pathname_t *realpnp)
781 {
782 	zfsctl_snapdir_t *sdp = dvp->v_data;
783 	objset_t *snap;
784 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
785 	char real[ZFS_MAX_DATASET_NAME_LEN];
786 	char *mountpoint;
787 	zfs_snapentry_t *sep, search;
788 	struct mounta margs;
789 	vfs_t *vfsp;
790 	size_t mountpoint_len;
791 	avl_index_t where;
792 	zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
793 	int err;
794 
795 	/*
796 	 * No extended attributes allowed under .zfs
797 	 */
798 	if (flags & LOOKUP_XATTR)
799 		return (SET_ERROR(EINVAL));
800 
801 	ASSERT(dvp->v_type == VDIR);
802 
803 	/*
804 	 * If we get a recursive call, that means we got called
805 	 * from the domount() code while it was trying to look up the
806 	 * spec (which looks like a local path for zfs).  We need to
807 	 * add some flag to domount() to tell it not to do this lookup.
808 	 */
809 	if (MUTEX_HELD(&sdp->sd_lock))
810 		return (SET_ERROR(ENOENT));
811 
812 	ZFS_ENTER(zfsvfs);
813 
814 	if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
815 		ZFS_EXIT(zfsvfs);
816 		return (0);
817 	}
818 
819 	if (flags & FIGNORECASE) {
820 		boolean_t conflict = B_FALSE;
821 
822 		err = dmu_snapshot_realname(zfsvfs->z_os, nm, real,
823 		    sizeof (real), &conflict);
824 		if (err == 0) {
825 			nm = real;
826 		} else if (err != ENOTSUP) {
827 			ZFS_EXIT(zfsvfs);
828 			return (err);
829 		}
830 		if (realpnp)
831 			(void) strlcpy(realpnp->pn_buf, nm,
832 			    realpnp->pn_bufsize);
833 		if (conflict && direntflags)
834 			*direntflags = ED_CASE_CONFLICT;
835 	}
836 
837 	mutex_enter(&sdp->sd_lock);
838 	search.se_name = (char *)nm;
839 	if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) {
840 		*vpp = sep->se_root;
841 		VN_HOLD(*vpp);
842 		err = traverse(vpp);
843 		if (err != 0) {
844 			VN_RELE(*vpp);
845 			*vpp = NULL;
846 		} else if (*vpp == sep->se_root) {
847 			/*
848 			 * The snapshot was unmounted behind our backs,
849 			 * try to remount it.
850 			 */
851 			goto domount;
852 		} else {
853 			/*
854 			 * VROOT was set during the traverse call.  We need
855 			 * to clear it since we're pretending to be part
856 			 * of our parent's vfs.
857 			 */
858 			(*vpp)->v_flag &= ~VROOT;
859 		}
860 		mutex_exit(&sdp->sd_lock);
861 		ZFS_EXIT(zfsvfs);
862 		return (err);
863 	}
864 
865 	/*
866 	 * The requested snapshot is not currently mounted, look it up.
867 	 */
868 	err = zfsctl_snapshot_zname(dvp, nm, sizeof (snapname), snapname);
869 	if (err != 0) {
870 		mutex_exit(&sdp->sd_lock);
871 		ZFS_EXIT(zfsvfs);
872 		/*
873 		 * handle "ls *" or "?" in a graceful manner,
874 		 * forcing EILSEQ to ENOENT.
875 		 * Since shell ultimately passes "*" or "?" as name to lookup
876 		 */
877 		return (err == EILSEQ ? ENOENT : err);
878 	}
879 	if (dmu_objset_hold(snapname, FTAG, &snap) != 0) {
880 		mutex_exit(&sdp->sd_lock);
881 		ZFS_EXIT(zfsvfs);
882 		return (SET_ERROR(ENOENT));
883 	}
884 
885 	sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP);
886 	sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
887 	(void) strcpy(sep->se_name, nm);
888 	*vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap));
889 	avl_insert(&sdp->sd_snaps, sep, where);
890 
891 	dmu_objset_rele(snap, FTAG);
892 domount:
893 	mountpoint_len = strlen(refstr_value(dvp->v_vfsp->vfs_mntpt)) +
894 	    strlen("/.zfs/snapshot/") + strlen(nm) + 1;
895 	mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP);
896 	(void) snprintf(mountpoint, mountpoint_len, "%s/.zfs/snapshot/%s",
897 	    refstr_value(dvp->v_vfsp->vfs_mntpt), nm);
898 
899 	margs.spec = snapname;
900 	margs.dir = mountpoint;
901 	margs.flags = MS_SYSSPACE | MS_NOMNTTAB;
902 	margs.fstype = "zfs";
903 	margs.dataptr = NULL;
904 	margs.datalen = 0;
905 	margs.optptr = NULL;
906 	margs.optlen = 0;
907 
908 	err = domount("zfs", &margs, *vpp, kcred, &vfsp);
909 	kmem_free(mountpoint, mountpoint_len);
910 
911 	if (err == 0) {
912 		/*
913 		 * Return the mounted root rather than the covered mount point.
914 		 * Takes the GFS vnode at .zfs/snapshot/<snapname> and returns
915 		 * the ZFS vnode mounted on top of the GFS node.  This ZFS
916 		 * vnode is the root of the newly created vfsp.
917 		 */
918 		VFS_RELE(vfsp);
919 		err = traverse(vpp);
920 	}
921 
922 	if (err == 0) {
923 		/*
924 		 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>.
925 		 *
926 		 * This is where we lie about our v_vfsp in order to
927 		 * make .zfs/snapshot/<snapname> accessible over NFS
928 		 * without requiring manual mounts of <snapname>.
929 		 */
930 		ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs);
931 		VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs;
932 		(*vpp)->v_vfsp = zfsvfs->z_vfs;
933 		(*vpp)->v_flag &= ~VROOT;
934 	}
935 	mutex_exit(&sdp->sd_lock);
936 	ZFS_EXIT(zfsvfs);
937 
938 	/*
939 	 * If we had an error, drop our hold on the vnode and
940 	 * zfsctl_snapshot_inactive() will clean up.
941 	 */
942 	if (err != 0) {
943 		VN_RELE(*vpp);
944 		*vpp = NULL;
945 	}
946 	return (err);
947 }
948 
949 /* ARGSUSED */
950 static int
951 zfsctl_shares_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
952     int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
953     int *direntflags, pathname_t *realpnp)
954 {
955 	zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
956 	znode_t *dzp;
957 	int error;
958 
959 	ZFS_ENTER(zfsvfs);
960 
961 	if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
962 		ZFS_EXIT(zfsvfs);
963 		return (0);
964 	}
965 
966 	if (zfsvfs->z_shares_dir == 0) {
967 		ZFS_EXIT(zfsvfs);
968 		return (SET_ERROR(ENOTSUP));
969 	}
970 	if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
971 		error = VOP_LOOKUP(ZTOV(dzp), nm, vpp, pnp,
972 		    flags, rdir, cr, ct, direntflags, realpnp);
973 		VN_RELE(ZTOV(dzp));
974 	}
975 
976 	ZFS_EXIT(zfsvfs);
977 
978 	return (error);
979 }
980 
981 /* ARGSUSED */
982 static int
983 zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp,
984     offset_t *offp, offset_t *nextp, void *data, int flags)
985 {
986 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
987 	char snapname[ZFS_MAX_DATASET_NAME_LEN];
988 	uint64_t id, cookie;
989 	boolean_t case_conflict;
990 	int error;
991 
992 	ZFS_ENTER(zfsvfs);
993 
994 	cookie = *offp;
995 	dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG);
996 	error = dmu_snapshot_list_next(zfsvfs->z_os,
997 	    sizeof (snapname), snapname, &id, &cookie, &case_conflict);
998 	dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG);
999 	if (error) {
1000 		ZFS_EXIT(zfsvfs);
1001 		if (error == ENOENT) {
1002 			*eofp = 1;
1003 			return (0);
1004 		}
1005 		return (error);
1006 	}
1007 
1008 	if (flags & V_RDDIR_ENTFLAGS) {
1009 		edirent_t *eodp = dp;
1010 
1011 		(void) strcpy(eodp->ed_name, snapname);
1012 		eodp->ed_ino = ZFSCTL_INO_SNAP(id);
1013 		eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0;
1014 	} else {
1015 		struct dirent64 *odp = dp;
1016 
1017 		(void) strcpy(odp->d_name, snapname);
1018 		odp->d_ino = ZFSCTL_INO_SNAP(id);
1019 	}
1020 	*nextp = cookie;
1021 
1022 	ZFS_EXIT(zfsvfs);
1023 
1024 	return (0);
1025 }
1026 
1027 /* ARGSUSED */
1028 static int
1029 zfsctl_shares_readdir(vnode_t *vp, uio_t *uiop, cred_t *cr, int *eofp,
1030     caller_context_t *ct, int flags)
1031 {
1032 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1033 	znode_t *dzp;
1034 	int error;
1035 
1036 	ZFS_ENTER(zfsvfs);
1037 
1038 	if (zfsvfs->z_shares_dir == 0) {
1039 		ZFS_EXIT(zfsvfs);
1040 		return (SET_ERROR(ENOTSUP));
1041 	}
1042 	if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1043 		error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ct, flags);
1044 		VN_RELE(ZTOV(dzp));
1045 	} else {
1046 		*eofp = 1;
1047 		error = SET_ERROR(ENOENT);
1048 	}
1049 
1050 	ZFS_EXIT(zfsvfs);
1051 	return (error);
1052 }
1053 
1054 /*
1055  * pvp is the '.zfs' directory (zfsctl_node_t).
1056  *
1057  * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t).
1058  *
1059  * This function is the callback to create a GFS vnode for '.zfs/snapshot'
1060  * when a lookup is performed on .zfs for "snapshot".
1061  */
1062 vnode_t *
1063 zfsctl_mknode_snapdir(vnode_t *pvp)
1064 {
1065 	vnode_t *vp;
1066 	zfsctl_snapdir_t *sdp;
1067 
1068 	vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp,
1069 	    zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN,
1070 	    zfsctl_snapdir_readdir_cb, NULL);
1071 	sdp = vp->v_data;
1072 	sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR;
1073 	sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1074 	mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL);
1075 	avl_create(&sdp->sd_snaps, snapentry_compare,
1076 	    sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node));
1077 	return (vp);
1078 }
1079 
1080 vnode_t *
1081 zfsctl_mknode_shares(vnode_t *pvp)
1082 {
1083 	vnode_t *vp;
1084 	zfsctl_node_t *sdp;
1085 
1086 	vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp,
1087 	    zfsctl_ops_shares, NULL, NULL, MAXNAMELEN,
1088 	    NULL, NULL);
1089 	sdp = vp->v_data;
1090 	sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1091 	return (vp);
1092 
1093 }
1094 
1095 /* ARGSUSED */
1096 static int
1097 zfsctl_shares_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1098     caller_context_t *ct)
1099 {
1100 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1101 	znode_t *dzp;
1102 	int error;
1103 
1104 	ZFS_ENTER(zfsvfs);
1105 	if (zfsvfs->z_shares_dir == 0) {
1106 		ZFS_EXIT(zfsvfs);
1107 		return (SET_ERROR(ENOTSUP));
1108 	}
1109 	if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1110 		error = VOP_GETATTR(ZTOV(dzp), vap, flags, cr, ct);
1111 		VN_RELE(ZTOV(dzp));
1112 	}
1113 	ZFS_EXIT(zfsvfs);
1114 	return (error);
1115 
1116 
1117 }
1118 
1119 /* ARGSUSED */
1120 static int
1121 zfsctl_snapdir_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1122     caller_context_t *ct)
1123 {
1124 	zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1125 	zfsctl_snapdir_t *sdp = vp->v_data;
1126 
1127 	ZFS_ENTER(zfsvfs);
1128 	zfsctl_common_getattr(vp, vap);
1129 	vap->va_nodeid = gfs_file_inode(vp);
1130 	vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2;
1131 	vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os);
1132 	ZFS_EXIT(zfsvfs);
1133 
1134 	return (0);
1135 }
1136 
1137 /* ARGSUSED */
1138 static void
1139 zfsctl_snapdir_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1140 {
1141 	zfsctl_snapdir_t *sdp = vp->v_data;
1142 	void *private;
1143 
1144 	private = gfs_dir_inactive(vp);
1145 	if (private != NULL) {
1146 		ASSERT(avl_numnodes(&sdp->sd_snaps) == 0);
1147 		mutex_destroy(&sdp->sd_lock);
1148 		avl_destroy(&sdp->sd_snaps);
1149 		kmem_free(private, sizeof (zfsctl_snapdir_t));
1150 	}
1151 }
1152 
1153 static const fs_operation_def_t zfsctl_tops_snapdir[] = {
1154 	{ VOPNAME_OPEN,		{ .vop_open = zfsctl_common_open }	},
1155 	{ VOPNAME_CLOSE,	{ .vop_close = zfsctl_common_close }	},
1156 	{ VOPNAME_IOCTL,	{ .error = fs_inval }			},
1157 	{ VOPNAME_GETATTR,	{ .vop_getattr = zfsctl_snapdir_getattr } },
1158 	{ VOPNAME_ACCESS,	{ .vop_access = zfsctl_common_access }	},
1159 	{ VOPNAME_RENAME,	{ .vop_rename = zfsctl_snapdir_rename }	},
1160 	{ VOPNAME_RMDIR,	{ .vop_rmdir = zfsctl_snapdir_remove }	},
1161 	{ VOPNAME_MKDIR,	{ .vop_mkdir = zfsctl_snapdir_mkdir }	},
1162 	{ VOPNAME_READDIR,	{ .vop_readdir = gfs_vop_readdir }	},
1163 	{ VOPNAME_LOOKUP,	{ .vop_lookup = zfsctl_snapdir_lookup }	},
1164 	{ VOPNAME_SEEK,		{ .vop_seek = fs_seek }			},
1165 	{ VOPNAME_INACTIVE,	{ .vop_inactive = zfsctl_snapdir_inactive } },
1166 	{ VOPNAME_FID,		{ .vop_fid = zfsctl_common_fid }	},
1167 	{ NULL }
1168 };
1169 
1170 static const fs_operation_def_t zfsctl_tops_shares[] = {
1171 	{ VOPNAME_OPEN,		{ .vop_open = zfsctl_common_open }	},
1172 	{ VOPNAME_CLOSE,	{ .vop_close = zfsctl_common_close }	},
1173 	{ VOPNAME_IOCTL,	{ .error = fs_inval }			},
1174 	{ VOPNAME_GETATTR,	{ .vop_getattr = zfsctl_shares_getattr } },
1175 	{ VOPNAME_ACCESS,	{ .vop_access = zfsctl_common_access }	},
1176 	{ VOPNAME_READDIR,	{ .vop_readdir = zfsctl_shares_readdir } },
1177 	{ VOPNAME_LOOKUP,	{ .vop_lookup = zfsctl_shares_lookup }	},
1178 	{ VOPNAME_SEEK,		{ .vop_seek = fs_seek }			},
1179 	{ VOPNAME_INACTIVE,	{ .vop_inactive = gfs_vop_inactive } },
1180 	{ VOPNAME_FID,		{ .vop_fid = zfsctl_shares_fid } },
1181 	{ NULL }
1182 };
1183 
1184 /*
1185  * pvp is the GFS vnode '.zfs/snapshot'.
1186  *
1187  * This creates a GFS node under '.zfs/snapshot' representing each
1188  * snapshot.  This newly created GFS node is what we mount snapshot
1189  * vfs_t's ontop of.
1190  */
1191 static vnode_t *
1192 zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset)
1193 {
1194 	vnode_t *vp;
1195 	zfsctl_node_t *zcp;
1196 
1197 	vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp,
1198 	    zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL);
1199 	zcp = vp->v_data;
1200 	zcp->zc_id = objset;
1201 
1202 	return (vp);
1203 }
1204 
1205 static void
1206 zfsctl_snapshot_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1207 {
1208 	zfsctl_snapdir_t *sdp;
1209 	zfs_snapentry_t *sep, *next;
1210 	vnode_t *dvp;
1211 
1212 	VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0);
1213 	sdp = dvp->v_data;
1214 
1215 	mutex_enter(&sdp->sd_lock);
1216 
1217 	mutex_enter(&vp->v_lock);
1218 	if (vp->v_count > 1) {
1219 		vp->v_count--;
1220 		mutex_exit(&vp->v_lock);
1221 		mutex_exit(&sdp->sd_lock);
1222 		VN_RELE(dvp);
1223 		return;
1224 	}
1225 	mutex_exit(&vp->v_lock);
1226 	ASSERT(!vn_ismntpt(vp));
1227 
1228 	sep = avl_first(&sdp->sd_snaps);
1229 	while (sep != NULL) {
1230 		next = AVL_NEXT(&sdp->sd_snaps, sep);
1231 
1232 		if (sep->se_root == vp) {
1233 			avl_remove(&sdp->sd_snaps, sep);
1234 			kmem_free(sep->se_name, strlen(sep->se_name) + 1);
1235 			kmem_free(sep, sizeof (zfs_snapentry_t));
1236 			break;
1237 		}
1238 		sep = next;
1239 	}
1240 	ASSERT(sep != NULL);
1241 
1242 	mutex_exit(&sdp->sd_lock);
1243 	VN_RELE(dvp);
1244 
1245 	/*
1246 	 * Dispose of the vnode for the snapshot mount point.
1247 	 * This is safe to do because once this entry has been removed
1248 	 * from the AVL tree, it can't be found again, so cannot become
1249 	 * "active".  If we lookup the same name again we will end up
1250 	 * creating a new vnode.
1251 	 */
1252 	gfs_vop_inactive(vp, cr, ct);
1253 }
1254 
1255 
1256 /*
1257  * These VP's should never see the light of day.  They should always
1258  * be covered.
1259  */
1260 static const fs_operation_def_t zfsctl_tops_snapshot[] = {
1261 	VOPNAME_INACTIVE, { .vop_inactive =  zfsctl_snapshot_inactive },
1262 	NULL, NULL
1263 };
1264 
1265 int
1266 zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp)
1267 {
1268 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1269 	vnode_t *dvp, *vp;
1270 	zfsctl_snapdir_t *sdp;
1271 	zfsctl_node_t *zcp;
1272 	zfs_snapentry_t *sep;
1273 	int error;
1274 
1275 	ASSERT(zfsvfs->z_ctldir != NULL);
1276 	error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1277 	    NULL, 0, NULL, kcred, NULL, NULL, NULL);
1278 	if (error != 0)
1279 		return (error);
1280 	sdp = dvp->v_data;
1281 
1282 	mutex_enter(&sdp->sd_lock);
1283 	sep = avl_first(&sdp->sd_snaps);
1284 	while (sep != NULL) {
1285 		vp = sep->se_root;
1286 		zcp = vp->v_data;
1287 		if (zcp->zc_id == objsetid)
1288 			break;
1289 
1290 		sep = AVL_NEXT(&sdp->sd_snaps, sep);
1291 	}
1292 
1293 	if (sep != NULL) {
1294 		VN_HOLD(vp);
1295 		/*
1296 		 * Return the mounted root rather than the covered mount point.
1297 		 * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid>
1298 		 * and returns the ZFS vnode mounted on top of the GFS node.
1299 		 * This ZFS vnode is the root of the vfs for objset 'objsetid'.
1300 		 */
1301 		error = traverse(&vp);
1302 		if (error == 0) {
1303 			if (vp == sep->se_root)
1304 				error = SET_ERROR(EINVAL);
1305 			else
1306 				*zfsvfsp = VTOZ(vp)->z_zfsvfs;
1307 		}
1308 		mutex_exit(&sdp->sd_lock);
1309 		VN_RELE(vp);
1310 	} else {
1311 		error = SET_ERROR(EINVAL);
1312 		mutex_exit(&sdp->sd_lock);
1313 	}
1314 
1315 	VN_RELE(dvp);
1316 
1317 	return (error);
1318 }
1319 
1320 /*
1321  * Unmount any snapshots for the given filesystem.  This is called from
1322  * zfs_umount() - if we have a ctldir, then go through and unmount all the
1323  * snapshots.
1324  */
1325 int
1326 zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr)
1327 {
1328 	zfsvfs_t *zfsvfs = vfsp->vfs_data;
1329 	vnode_t *dvp;
1330 	zfsctl_snapdir_t *sdp;
1331 	zfs_snapentry_t *sep, *next;
1332 	int error;
1333 
1334 	ASSERT(zfsvfs->z_ctldir != NULL);
1335 	error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1336 	    NULL, 0, NULL, cr, NULL, NULL, NULL);
1337 	if (error != 0)
1338 		return (error);
1339 	sdp = dvp->v_data;
1340 
1341 	mutex_enter(&sdp->sd_lock);
1342 
1343 	sep = avl_first(&sdp->sd_snaps);
1344 	while (sep != NULL) {
1345 		next = AVL_NEXT(&sdp->sd_snaps, sep);
1346 
1347 		/*
1348 		 * If this snapshot is not mounted, then it must
1349 		 * have just been unmounted by somebody else, and
1350 		 * will be cleaned up by zfsctl_snapdir_inactive().
1351 		 */
1352 		if (vn_ismntpt(sep->se_root)) {
1353 			avl_remove(&sdp->sd_snaps, sep);
1354 			error = zfsctl_unmount_snap(sep, fflags, cr);
1355 			if (error) {
1356 				avl_add(&sdp->sd_snaps, sep);
1357 				break;
1358 			}
1359 		}
1360 		sep = next;
1361 	}
1362 
1363 	mutex_exit(&sdp->sd_lock);
1364 	VN_RELE(dvp);
1365 
1366 	return (error);
1367 }
1368