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 https://opensource.org/licenses/CDDL-1.0.
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 *
23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
24 * Copyright (C) 2011 Lawrence Livermore National Security, LLC.
25 * Produced at Lawrence Livermore National Laboratory (cf, DISCLAIMER).
26 * LLNL-CODE-403049.
27 * Rewritten for Linux by:
28 *   Rohan Puri <rohan.puri15@gmail.com>
29 *   Brian Behlendorf <behlendorf1@llnl.gov>
30 * Copyright (c) 2013 by Delphix. All rights reserved.
31 * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved.
32 * Copyright (c) 2018 George Melikov. All Rights Reserved.
33 * Copyright (c) 2019 Datto, Inc. All rights reserved.
34 * Copyright (c) 2020 The MathWorks, Inc. All rights reserved.
35 */
36
37/*
38 * ZFS control directory (a.k.a. ".zfs")
39 *
40 * This directory provides a common location for all ZFS meta-objects.
41 * Currently, this is only the 'snapshot' and 'shares' directory, but this may
42 * expand in the future.  The elements are built dynamically, as the hierarchy
43 * does not actually exist on disk.
44 *
45 * For 'snapshot', we don't want to have all snapshots always mounted, because
46 * this would take up a huge amount of space in /etc/mnttab.  We have three
47 * types of objects:
48 *
49 *	ctldir ------> snapshotdir -------> snapshot
50 *                                             |
51 *                                             |
52 *                                             V
53 *                                         mounted fs
54 *
55 * The 'snapshot' node contains just enough information to lookup '..' and act
56 * as a mountpoint for the snapshot.  Whenever we lookup a specific snapshot, we
57 * perform an automount of the underlying filesystem and return the
58 * corresponding inode.
59 *
60 * All mounts are handled automatically by an user mode helper which invokes
61 * the mount procedure.  Unmounts are handled by allowing the mount
62 * point to expire so the kernel may automatically unmount it.
63 *
64 * The '.zfs', '.zfs/snapshot', and all directories created under
65 * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') all share the same
66 * zfsvfs_t as the head filesystem (what '.zfs' lives under).
67 *
68 * File systems mounted on top of the '.zfs/snapshot/<snapname>' paths
69 * (ie: snapshots) are complete ZFS filesystems and have their own unique
70 * zfsvfs_t.  However, the fsid reported by these mounts will be the same
71 * as that used by the parent zfsvfs_t to make NFS happy.
72 */
73
74#include <sys/types.h>
75#include <sys/param.h>
76#include <sys/time.h>
77#include <sys/sysmacros.h>
78#include <sys/pathname.h>
79#include <sys/vfs.h>
80#include <sys/zfs_ctldir.h>
81#include <sys/zfs_ioctl.h>
82#include <sys/zfs_vfsops.h>
83#include <sys/zfs_vnops.h>
84#include <sys/stat.h>
85#include <sys/dmu.h>
86#include <sys/dmu_objset.h>
87#include <sys/dsl_destroy.h>
88#include <sys/dsl_deleg.h>
89#include <sys/zpl.h>
90#include <sys/mntent.h>
91#include "zfs_namecheck.h"
92
93/*
94 * Two AVL trees are maintained which contain all currently automounted
95 * snapshots.  Every automounted snapshots maps to a single zfs_snapentry_t
96 * entry which MUST:
97 *
98 *   - be attached to both trees, and
99 *   - be unique, no duplicate entries are allowed.
100 *
101 * The zfs_snapshots_by_name tree is indexed by the full dataset name
102 * while the zfs_snapshots_by_objsetid tree is indexed by the unique
103 * objsetid.  This allows for fast lookups either by name or objsetid.
104 */
105static avl_tree_t zfs_snapshots_by_name;
106static avl_tree_t zfs_snapshots_by_objsetid;
107static krwlock_t zfs_snapshot_lock;
108
109/*
110 * Control Directory Tunables (.zfs)
111 */
112int zfs_expire_snapshot = ZFSCTL_EXPIRE_SNAPSHOT;
113static int zfs_admin_snapshot = 0;
114
115typedef struct {
116	char		*se_name;	/* full snapshot name */
117	char		*se_path;	/* full mount path */
118	spa_t		*se_spa;	/* pool spa */
119	uint64_t	se_objsetid;	/* snapshot objset id */
120	struct dentry   *se_root_dentry; /* snapshot root dentry */
121	krwlock_t	se_taskqid_lock;  /* scheduled unmount taskqid lock */
122	taskqid_t	se_taskqid;	/* scheduled unmount taskqid */
123	avl_node_t	se_node_name;	/* zfs_snapshots_by_name link */
124	avl_node_t	se_node_objsetid; /* zfs_snapshots_by_objsetid link */
125	zfs_refcount_t	se_refcount;	/* reference count */
126} zfs_snapentry_t;
127
128static void zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t *se, int delay);
129
130/*
131 * Allocate a new zfs_snapentry_t being careful to make a copy of the
132 * the snapshot name and provided mount point.  No reference is taken.
133 */
134static zfs_snapentry_t *
135zfsctl_snapshot_alloc(const char *full_name, const char *full_path, spa_t *spa,
136    uint64_t objsetid, struct dentry *root_dentry)
137{
138	zfs_snapentry_t *se;
139
140	se = kmem_zalloc(sizeof (zfs_snapentry_t), KM_SLEEP);
141
142	se->se_name = kmem_strdup(full_name);
143	se->se_path = kmem_strdup(full_path);
144	se->se_spa = spa;
145	se->se_objsetid = objsetid;
146	se->se_root_dentry = root_dentry;
147	se->se_taskqid = TASKQID_INVALID;
148	rw_init(&se->se_taskqid_lock, NULL, RW_DEFAULT, NULL);
149
150	zfs_refcount_create(&se->se_refcount);
151
152	return (se);
153}
154
155/*
156 * Free a zfs_snapentry_t the caller must ensure there are no active
157 * references.
158 */
159static void
160zfsctl_snapshot_free(zfs_snapentry_t *se)
161{
162	zfs_refcount_destroy(&se->se_refcount);
163	kmem_strfree(se->se_name);
164	kmem_strfree(se->se_path);
165	rw_destroy(&se->se_taskqid_lock);
166
167	kmem_free(se, sizeof (zfs_snapentry_t));
168}
169
170/*
171 * Hold a reference on the zfs_snapentry_t.
172 */
173static void
174zfsctl_snapshot_hold(zfs_snapentry_t *se)
175{
176	zfs_refcount_add(&se->se_refcount, NULL);
177}
178
179/*
180 * Release a reference on the zfs_snapentry_t.  When the number of
181 * references drops to zero the structure will be freed.
182 */
183static void
184zfsctl_snapshot_rele(zfs_snapentry_t *se)
185{
186	if (zfs_refcount_remove(&se->se_refcount, NULL) == 0)
187		zfsctl_snapshot_free(se);
188}
189
190/*
191 * Add a zfs_snapentry_t to both the zfs_snapshots_by_name and
192 * zfs_snapshots_by_objsetid trees.  While the zfs_snapentry_t is part
193 * of the trees a reference is held.
194 */
195static void
196zfsctl_snapshot_add(zfs_snapentry_t *se)
197{
198	ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock));
199	zfsctl_snapshot_hold(se);
200	avl_add(&zfs_snapshots_by_name, se);
201	avl_add(&zfs_snapshots_by_objsetid, se);
202}
203
204/*
205 * Remove a zfs_snapentry_t from both the zfs_snapshots_by_name and
206 * zfs_snapshots_by_objsetid trees.  Upon removal a reference is dropped,
207 * this can result in the structure being freed if that was the last
208 * remaining reference.
209 */
210static void
211zfsctl_snapshot_remove(zfs_snapentry_t *se)
212{
213	ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock));
214	avl_remove(&zfs_snapshots_by_name, se);
215	avl_remove(&zfs_snapshots_by_objsetid, se);
216	zfsctl_snapshot_rele(se);
217}
218
219/*
220 * Snapshot name comparison function for the zfs_snapshots_by_name.
221 */
222static int
223snapentry_compare_by_name(const void *a, const void *b)
224{
225	const zfs_snapentry_t *se_a = a;
226	const zfs_snapentry_t *se_b = b;
227	int ret;
228
229	ret = strcmp(se_a->se_name, se_b->se_name);
230
231	if (ret < 0)
232		return (-1);
233	else if (ret > 0)
234		return (1);
235	else
236		return (0);
237}
238
239/*
240 * Snapshot name comparison function for the zfs_snapshots_by_objsetid.
241 */
242static int
243snapentry_compare_by_objsetid(const void *a, const void *b)
244{
245	const zfs_snapentry_t *se_a = a;
246	const zfs_snapentry_t *se_b = b;
247
248	if (se_a->se_spa != se_b->se_spa)
249		return ((ulong_t)se_a->se_spa < (ulong_t)se_b->se_spa ? -1 : 1);
250
251	if (se_a->se_objsetid < se_b->se_objsetid)
252		return (-1);
253	else if (se_a->se_objsetid > se_b->se_objsetid)
254		return (1);
255	else
256		return (0);
257}
258
259/*
260 * Find a zfs_snapentry_t in zfs_snapshots_by_name.  If the snapname
261 * is found a pointer to the zfs_snapentry_t is returned and a reference
262 * taken on the structure.  The caller is responsible for dropping the
263 * reference with zfsctl_snapshot_rele().  If the snapname is not found
264 * NULL will be returned.
265 */
266static zfs_snapentry_t *
267zfsctl_snapshot_find_by_name(const char *snapname)
268{
269	zfs_snapentry_t *se, search;
270
271	ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock));
272
273	search.se_name = (char *)snapname;
274	se = avl_find(&zfs_snapshots_by_name, &search, NULL);
275	if (se)
276		zfsctl_snapshot_hold(se);
277
278	return (se);
279}
280
281/*
282 * Find a zfs_snapentry_t in zfs_snapshots_by_objsetid given the objset id
283 * rather than the snapname.  In all other respects it behaves the same
284 * as zfsctl_snapshot_find_by_name().
285 */
286static zfs_snapentry_t *
287zfsctl_snapshot_find_by_objsetid(spa_t *spa, uint64_t objsetid)
288{
289	zfs_snapentry_t *se, search;
290
291	ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock));
292
293	search.se_spa = spa;
294	search.se_objsetid = objsetid;
295	se = avl_find(&zfs_snapshots_by_objsetid, &search, NULL);
296	if (se)
297		zfsctl_snapshot_hold(se);
298
299	return (se);
300}
301
302/*
303 * Rename a zfs_snapentry_t in the zfs_snapshots_by_name.  The structure is
304 * removed, renamed, and added back to the new correct location in the tree.
305 */
306static int
307zfsctl_snapshot_rename(const char *old_snapname, const char *new_snapname)
308{
309	zfs_snapentry_t *se;
310
311	ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock));
312
313	se = zfsctl_snapshot_find_by_name(old_snapname);
314	if (se == NULL)
315		return (SET_ERROR(ENOENT));
316
317	zfsctl_snapshot_remove(se);
318	kmem_strfree(se->se_name);
319	se->se_name = kmem_strdup(new_snapname);
320	zfsctl_snapshot_add(se);
321	zfsctl_snapshot_rele(se);
322
323	return (0);
324}
325
326/*
327 * Delayed task responsible for unmounting an expired automounted snapshot.
328 */
329static void
330snapentry_expire(void *data)
331{
332	zfs_snapentry_t *se = (zfs_snapentry_t *)data;
333	spa_t *spa = se->se_spa;
334	uint64_t objsetid = se->se_objsetid;
335
336	if (zfs_expire_snapshot <= 0) {
337		zfsctl_snapshot_rele(se);
338		return;
339	}
340
341	rw_enter(&se->se_taskqid_lock, RW_WRITER);
342	se->se_taskqid = TASKQID_INVALID;
343	rw_exit(&se->se_taskqid_lock);
344	(void) zfsctl_snapshot_unmount(se->se_name, MNT_EXPIRE);
345	zfsctl_snapshot_rele(se);
346
347	/*
348	 * Reschedule the unmount if the zfs_snapentry_t wasn't removed.
349	 * This can occur when the snapshot is busy.
350	 */
351	rw_enter(&zfs_snapshot_lock, RW_READER);
352	if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) {
353		zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot);
354		zfsctl_snapshot_rele(se);
355	}
356	rw_exit(&zfs_snapshot_lock);
357}
358
359/*
360 * Cancel an automatic unmount of a snapname.  This callback is responsible
361 * for dropping the reference on the zfs_snapentry_t which was taken when
362 * during dispatch.
363 */
364static void
365zfsctl_snapshot_unmount_cancel(zfs_snapentry_t *se)
366{
367	int err = 0;
368	rw_enter(&se->se_taskqid_lock, RW_WRITER);
369	err = taskq_cancel_id(system_delay_taskq, se->se_taskqid);
370	/*
371	 * if we get ENOENT, the taskq couldn't be found to be
372	 * canceled, so we can just mark it as invalid because
373	 * it's already gone. If we got EBUSY, then we already
374	 * blocked until it was gone _anyway_, so we don't care.
375	 */
376	se->se_taskqid = TASKQID_INVALID;
377	rw_exit(&se->se_taskqid_lock);
378	if (err == 0) {
379		zfsctl_snapshot_rele(se);
380	}
381}
382
383/*
384 * Dispatch the unmount task for delayed handling with a hold protecting it.
385 */
386static void
387zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t *se, int delay)
388{
389
390	if (delay <= 0)
391		return;
392
393	zfsctl_snapshot_hold(se);
394	rw_enter(&se->se_taskqid_lock, RW_WRITER);
395	/*
396	 * If this condition happens, we managed to:
397	 * - dispatch once
398	 * - want to dispatch _again_ before it returned
399	 *
400	 * So let's just return - if that task fails at unmounting,
401	 * we'll eventually dispatch again, and if it succeeds,
402	 * no problem.
403	 */
404	if (se->se_taskqid != TASKQID_INVALID) {
405		rw_exit(&se->se_taskqid_lock);
406		zfsctl_snapshot_rele(se);
407		return;
408	}
409	se->se_taskqid = taskq_dispatch_delay(system_delay_taskq,
410	    snapentry_expire, se, TQ_SLEEP, ddi_get_lbolt() + delay * HZ);
411	rw_exit(&se->se_taskqid_lock);
412}
413
414/*
415 * Schedule an automatic unmount of objset id to occur in delay seconds from
416 * now.  Any previous delayed unmount will be cancelled in favor of the
417 * updated deadline.  A reference is taken by zfsctl_snapshot_find_by_name()
418 * and held until the outstanding task is handled or cancelled.
419 */
420int
421zfsctl_snapshot_unmount_delay(spa_t *spa, uint64_t objsetid, int delay)
422{
423	zfs_snapentry_t *se;
424	int error = ENOENT;
425
426	rw_enter(&zfs_snapshot_lock, RW_READER);
427	if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) {
428		zfsctl_snapshot_unmount_cancel(se);
429		zfsctl_snapshot_unmount_delay_impl(se, delay);
430		zfsctl_snapshot_rele(se);
431		error = 0;
432	}
433	rw_exit(&zfs_snapshot_lock);
434
435	return (error);
436}
437
438/*
439 * Check if snapname is currently mounted.  Returned non-zero when mounted
440 * and zero when unmounted.
441 */
442static boolean_t
443zfsctl_snapshot_ismounted(const char *snapname)
444{
445	zfs_snapentry_t *se;
446	boolean_t ismounted = B_FALSE;
447
448	rw_enter(&zfs_snapshot_lock, RW_READER);
449	if ((se = zfsctl_snapshot_find_by_name(snapname)) != NULL) {
450		zfsctl_snapshot_rele(se);
451		ismounted = B_TRUE;
452	}
453	rw_exit(&zfs_snapshot_lock);
454
455	return (ismounted);
456}
457
458/*
459 * Check if the given inode is a part of the virtual .zfs directory.
460 */
461boolean_t
462zfsctl_is_node(struct inode *ip)
463{
464	return (ITOZ(ip)->z_is_ctldir);
465}
466
467/*
468 * Check if the given inode is a .zfs/snapshots/snapname directory.
469 */
470boolean_t
471zfsctl_is_snapdir(struct inode *ip)
472{
473	return (zfsctl_is_node(ip) && (ip->i_ino <= ZFSCTL_INO_SNAPDIRS));
474}
475
476/*
477 * Allocate a new inode with the passed id and ops.
478 */
479static struct inode *
480zfsctl_inode_alloc(zfsvfs_t *zfsvfs, uint64_t id,
481    const struct file_operations *fops, const struct inode_operations *ops,
482    uint64_t creation)
483{
484	struct inode *ip;
485	znode_t *zp;
486	inode_timespec_t now = {.tv_sec = creation};
487
488	ip = new_inode(zfsvfs->z_sb);
489	if (ip == NULL)
490		return (NULL);
491
492	if (!creation)
493		now = current_time(ip);
494	zp = ITOZ(ip);
495	ASSERT3P(zp->z_dirlocks, ==, NULL);
496	ASSERT3P(zp->z_acl_cached, ==, NULL);
497	ASSERT3P(zp->z_xattr_cached, ==, NULL);
498	zp->z_id = id;
499	zp->z_unlinked = B_FALSE;
500	zp->z_atime_dirty = B_FALSE;
501	zp->z_zn_prefetch = B_FALSE;
502	zp->z_is_sa = B_FALSE;
503#if !defined(HAVE_FILEMAP_RANGE_HAS_PAGE)
504	zp->z_is_mapped = B_FALSE;
505#endif
506	zp->z_is_ctldir = B_TRUE;
507	zp->z_sa_hdl = NULL;
508	zp->z_blksz = 0;
509	zp->z_seq = 0;
510	zp->z_mapcnt = 0;
511	zp->z_size = 0;
512	zp->z_pflags = 0;
513	zp->z_mode = 0;
514	zp->z_sync_cnt = 0;
515	zp->z_sync_writes_cnt = 0;
516	zp->z_async_writes_cnt = 0;
517	ip->i_generation = 0;
518	ip->i_ino = id;
519	ip->i_mode = (S_IFDIR | S_IRWXUGO);
520	ip->i_uid = SUID_TO_KUID(0);
521	ip->i_gid = SGID_TO_KGID(0);
522	ip->i_blkbits = SPA_MINBLOCKSHIFT;
523	zpl_inode_set_atime_to_ts(ip, now);
524	zpl_inode_set_mtime_to_ts(ip, now);
525	zpl_inode_set_ctime_to_ts(ip, now);
526	ip->i_fop = fops;
527	ip->i_op = ops;
528#if defined(IOP_XATTR)
529	ip->i_opflags &= ~IOP_XATTR;
530#endif
531
532	if (insert_inode_locked(ip)) {
533		unlock_new_inode(ip);
534		iput(ip);
535		return (NULL);
536	}
537
538	mutex_enter(&zfsvfs->z_znodes_lock);
539	list_insert_tail(&zfsvfs->z_all_znodes, zp);
540	membar_producer();
541	mutex_exit(&zfsvfs->z_znodes_lock);
542
543	unlock_new_inode(ip);
544
545	return (ip);
546}
547
548/*
549 * Lookup the inode with given id, it will be allocated if needed.
550 */
551static struct inode *
552zfsctl_inode_lookup(zfsvfs_t *zfsvfs, uint64_t id,
553    const struct file_operations *fops, const struct inode_operations *ops)
554{
555	struct inode *ip = NULL;
556	uint64_t creation = 0;
557	dsl_dataset_t *snap_ds;
558	dsl_pool_t *pool;
559
560	while (ip == NULL) {
561		ip = ilookup(zfsvfs->z_sb, (unsigned long)id);
562		if (ip)
563			break;
564
565		if (id <= ZFSCTL_INO_SNAPDIRS && !creation) {
566			pool = dmu_objset_pool(zfsvfs->z_os);
567			dsl_pool_config_enter(pool, FTAG);
568			if (!dsl_dataset_hold_obj(pool,
569			    ZFSCTL_INO_SNAPDIRS - id, FTAG, &snap_ds)) {
570				creation = dsl_get_creation(snap_ds);
571				dsl_dataset_rele(snap_ds, FTAG);
572			}
573			dsl_pool_config_exit(pool, FTAG);
574		}
575
576		/* May fail due to concurrent zfsctl_inode_alloc() */
577		ip = zfsctl_inode_alloc(zfsvfs, id, fops, ops, creation);
578	}
579
580	return (ip);
581}
582
583/*
584 * Create the '.zfs' directory.  This directory is cached as part of the VFS
585 * structure.  This results in a hold on the zfsvfs_t.  The code in zfs_umount()
586 * therefore checks against a vfs_count of 2 instead of 1.  This reference
587 * is removed when the ctldir is destroyed in the unmount.  All other entities
588 * under the '.zfs' directory are created dynamically as needed.
589 *
590 * Because the dynamically created '.zfs' directory entries assume the use
591 * of 64-bit inode numbers this support must be disabled on 32-bit systems.
592 */
593int
594zfsctl_create(zfsvfs_t *zfsvfs)
595{
596	ASSERT(zfsvfs->z_ctldir == NULL);
597
598	zfsvfs->z_ctldir = zfsctl_inode_alloc(zfsvfs, ZFSCTL_INO_ROOT,
599	    &zpl_fops_root, &zpl_ops_root, 0);
600	if (zfsvfs->z_ctldir == NULL)
601		return (SET_ERROR(ENOENT));
602
603	return (0);
604}
605
606/*
607 * Destroy the '.zfs' directory or remove a snapshot from zfs_snapshots_by_name.
608 * Only called when the filesystem is unmounted.
609 */
610void
611zfsctl_destroy(zfsvfs_t *zfsvfs)
612{
613	if (zfsvfs->z_issnap) {
614		zfs_snapentry_t *se;
615		spa_t *spa = zfsvfs->z_os->os_spa;
616		uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
617
618		rw_enter(&zfs_snapshot_lock, RW_WRITER);
619		se = zfsctl_snapshot_find_by_objsetid(spa, objsetid);
620		if (se != NULL)
621			zfsctl_snapshot_remove(se);
622		rw_exit(&zfs_snapshot_lock);
623		if (se != NULL) {
624			zfsctl_snapshot_unmount_cancel(se);
625			zfsctl_snapshot_rele(se);
626		}
627	} else if (zfsvfs->z_ctldir) {
628		iput(zfsvfs->z_ctldir);
629		zfsvfs->z_ctldir = NULL;
630	}
631}
632
633/*
634 * Given a root znode, retrieve the associated .zfs directory.
635 * Add a hold to the vnode and return it.
636 */
637struct inode *
638zfsctl_root(znode_t *zp)
639{
640	ASSERT(zfs_has_ctldir(zp));
641	/* Must have an existing ref, so igrab() cannot return NULL */
642	VERIFY3P(igrab(ZTOZSB(zp)->z_ctldir), !=, NULL);
643	return (ZTOZSB(zp)->z_ctldir);
644}
645
646/*
647 * Generate a long fid to indicate a snapdir. We encode whether snapdir is
648 * already mounted in gen field. We do this because nfsd lookup will not
649 * trigger automount. Next time the nfsd does fh_to_dentry, we will notice
650 * this and do automount and return ESTALE to force nfsd revalidate and follow
651 * mount.
652 */
653static int
654zfsctl_snapdir_fid(struct inode *ip, fid_t *fidp)
655{
656	zfid_short_t *zfid = (zfid_short_t *)fidp;
657	zfid_long_t *zlfid = (zfid_long_t *)fidp;
658	uint32_t gen = 0;
659	uint64_t object;
660	uint64_t objsetid;
661	int i;
662	struct dentry *dentry;
663
664	if (fidp->fid_len < LONG_FID_LEN) {
665		fidp->fid_len = LONG_FID_LEN;
666		return (SET_ERROR(ENOSPC));
667	}
668
669	object = ip->i_ino;
670	objsetid = ZFSCTL_INO_SNAPDIRS - ip->i_ino;
671	zfid->zf_len = LONG_FID_LEN;
672
673	dentry = d_obtain_alias(igrab(ip));
674	if (!IS_ERR(dentry)) {
675		gen = !!d_mountpoint(dentry);
676		dput(dentry);
677	}
678
679	for (i = 0; i < sizeof (zfid->zf_object); i++)
680		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
681
682	for (i = 0; i < sizeof (zfid->zf_gen); i++)
683		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
684
685	for (i = 0; i < sizeof (zlfid->zf_setid); i++)
686		zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
687
688	for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
689		zlfid->zf_setgen[i] = 0;
690
691	return (0);
692}
693
694/*
695 * Generate an appropriate fid for an entry in the .zfs directory.
696 */
697int
698zfsctl_fid(struct inode *ip, fid_t *fidp)
699{
700	znode_t		*zp = ITOZ(ip);
701	zfsvfs_t	*zfsvfs = ITOZSB(ip);
702	uint64_t	object = zp->z_id;
703	zfid_short_t	*zfid;
704	int		i;
705	int		error;
706
707	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
708		return (error);
709
710	if (zfsctl_is_snapdir(ip)) {
711		zfs_exit(zfsvfs, FTAG);
712		return (zfsctl_snapdir_fid(ip, fidp));
713	}
714
715	if (fidp->fid_len < SHORT_FID_LEN) {
716		fidp->fid_len = SHORT_FID_LEN;
717		zfs_exit(zfsvfs, FTAG);
718		return (SET_ERROR(ENOSPC));
719	}
720
721	zfid = (zfid_short_t *)fidp;
722
723	zfid->zf_len = SHORT_FID_LEN;
724
725	for (i = 0; i < sizeof (zfid->zf_object); i++)
726		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
727
728	/* .zfs znodes always have a generation number of 0 */
729	for (i = 0; i < sizeof (zfid->zf_gen); i++)
730		zfid->zf_gen[i] = 0;
731
732	zfs_exit(zfsvfs, FTAG);
733	return (0);
734}
735
736/*
737 * Construct a full dataset name in full_name: "pool/dataset@snap_name"
738 */
739static int
740zfsctl_snapshot_name(zfsvfs_t *zfsvfs, const char *snap_name, int len,
741    char *full_name)
742{
743	objset_t *os = zfsvfs->z_os;
744
745	if (zfs_component_namecheck(snap_name, NULL, NULL) != 0)
746		return (SET_ERROR(EILSEQ));
747
748	dmu_objset_name(os, full_name);
749	if ((strlen(full_name) + 1 + strlen(snap_name)) >= len)
750		return (SET_ERROR(ENAMETOOLONG));
751
752	(void) strcat(full_name, "@");
753	(void) strcat(full_name, snap_name);
754
755	return (0);
756}
757
758/*
759 * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/"
760 */
761static int
762zfsctl_snapshot_path_objset(zfsvfs_t *zfsvfs, uint64_t objsetid,
763    int path_len, char *full_path)
764{
765	objset_t *os = zfsvfs->z_os;
766	fstrans_cookie_t cookie;
767	char *snapname;
768	boolean_t case_conflict;
769	uint64_t id, pos = 0;
770	int error = 0;
771
772	if (zfsvfs->z_vfs->vfs_mntpoint == NULL)
773		return (SET_ERROR(ENOENT));
774
775	cookie = spl_fstrans_mark();
776	snapname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
777
778	while (error == 0) {
779		dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
780		error = dmu_snapshot_list_next(zfsvfs->z_os,
781		    ZFS_MAX_DATASET_NAME_LEN, snapname, &id, &pos,
782		    &case_conflict);
783		dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
784		if (error)
785			goto out;
786
787		if (id == objsetid)
788			break;
789	}
790
791	snprintf(full_path, path_len, "%s/.zfs/snapshot/%s",
792	    zfsvfs->z_vfs->vfs_mntpoint, snapname);
793out:
794	kmem_free(snapname, ZFS_MAX_DATASET_NAME_LEN);
795	spl_fstrans_unmark(cookie);
796
797	return (error);
798}
799
800/*
801 * Special case the handling of "..".
802 */
803int
804zfsctl_root_lookup(struct inode *dip, const char *name, struct inode **ipp,
805    int flags, cred_t *cr, int *direntflags, pathname_t *realpnp)
806{
807	zfsvfs_t *zfsvfs = ITOZSB(dip);
808	int error = 0;
809
810	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
811		return (error);
812
813	if (strcmp(name, "..") == 0) {
814		*ipp = dip->i_sb->s_root->d_inode;
815	} else if (strcmp(name, ZFS_SNAPDIR_NAME) == 0) {
816		*ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SNAPDIR,
817		    &zpl_fops_snapdir, &zpl_ops_snapdir);
818	} else if (strcmp(name, ZFS_SHAREDIR_NAME) == 0) {
819		*ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SHARES,
820		    &zpl_fops_shares, &zpl_ops_shares);
821	} else {
822		*ipp = NULL;
823	}
824
825	if (*ipp == NULL)
826		error = SET_ERROR(ENOENT);
827
828	zfs_exit(zfsvfs, FTAG);
829
830	return (error);
831}
832
833/*
834 * Lookup entry point for the 'snapshot' directory.  Try to open the
835 * snapshot if it exist, creating the pseudo filesystem inode as necessary.
836 */
837int
838zfsctl_snapdir_lookup(struct inode *dip, const char *name, struct inode **ipp,
839    int flags, cred_t *cr, int *direntflags, pathname_t *realpnp)
840{
841	zfsvfs_t *zfsvfs = ITOZSB(dip);
842	uint64_t id;
843	int error;
844
845	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
846		return (error);
847
848	error = dmu_snapshot_lookup(zfsvfs->z_os, name, &id);
849	if (error) {
850		zfs_exit(zfsvfs, FTAG);
851		return (error);
852	}
853
854	*ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SNAPDIRS - id,
855	    &simple_dir_operations, &simple_dir_inode_operations);
856	if (*ipp == NULL)
857		error = SET_ERROR(ENOENT);
858
859	zfs_exit(zfsvfs, FTAG);
860
861	return (error);
862}
863
864/*
865 * Renaming a directory under '.zfs/snapshot' will automatically trigger
866 * a rename of the snapshot to the new given name.  The rename is confined
867 * to the '.zfs/snapshot' directory snapshots cannot be moved elsewhere.
868 */
869int
870zfsctl_snapdir_rename(struct inode *sdip, const char *snm,
871    struct inode *tdip, const char *tnm, cred_t *cr, int flags)
872{
873	zfsvfs_t *zfsvfs = ITOZSB(sdip);
874	char *to, *from, *real, *fsname;
875	int error;
876
877	if (!zfs_admin_snapshot)
878		return (SET_ERROR(EACCES));
879
880	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
881		return (error);
882
883	to = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
884	from = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
885	real = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
886	fsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
887
888	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
889		error = dmu_snapshot_realname(zfsvfs->z_os, snm, real,
890		    ZFS_MAX_DATASET_NAME_LEN, NULL);
891		if (error == 0) {
892			snm = real;
893		} else if (error != ENOTSUP) {
894			goto out;
895		}
896	}
897
898	dmu_objset_name(zfsvfs->z_os, fsname);
899
900	error = zfsctl_snapshot_name(ITOZSB(sdip), snm,
901	    ZFS_MAX_DATASET_NAME_LEN, from);
902	if (error == 0)
903		error = zfsctl_snapshot_name(ITOZSB(tdip), tnm,
904		    ZFS_MAX_DATASET_NAME_LEN, to);
905	if (error == 0)
906		error = zfs_secpolicy_rename_perms(from, to, cr);
907	if (error != 0)
908		goto out;
909
910	/*
911	 * Cannot move snapshots out of the snapdir.
912	 */
913	if (sdip != tdip) {
914		error = SET_ERROR(EINVAL);
915		goto out;
916	}
917
918	/*
919	 * No-op when names are identical.
920	 */
921	if (strcmp(snm, tnm) == 0) {
922		error = 0;
923		goto out;
924	}
925
926	rw_enter(&zfs_snapshot_lock, RW_WRITER);
927
928	error = dsl_dataset_rename_snapshot(fsname, snm, tnm, B_FALSE);
929	if (error == 0)
930		(void) zfsctl_snapshot_rename(snm, tnm);
931
932	rw_exit(&zfs_snapshot_lock);
933out:
934	kmem_free(from, ZFS_MAX_DATASET_NAME_LEN);
935	kmem_free(to, ZFS_MAX_DATASET_NAME_LEN);
936	kmem_free(real, ZFS_MAX_DATASET_NAME_LEN);
937	kmem_free(fsname, ZFS_MAX_DATASET_NAME_LEN);
938
939	zfs_exit(zfsvfs, FTAG);
940
941	return (error);
942}
943
944/*
945 * Removing a directory under '.zfs/snapshot' will automatically trigger
946 * the removal of the snapshot with the given name.
947 */
948int
949zfsctl_snapdir_remove(struct inode *dip, const char *name, cred_t *cr,
950    int flags)
951{
952	zfsvfs_t *zfsvfs = ITOZSB(dip);
953	char *snapname, *real;
954	int error;
955
956	if (!zfs_admin_snapshot)
957		return (SET_ERROR(EACCES));
958
959	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
960		return (error);
961
962	snapname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
963	real = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
964
965	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
966		error = dmu_snapshot_realname(zfsvfs->z_os, name, real,
967		    ZFS_MAX_DATASET_NAME_LEN, NULL);
968		if (error == 0) {
969			name = real;
970		} else if (error != ENOTSUP) {
971			goto out;
972		}
973	}
974
975	error = zfsctl_snapshot_name(ITOZSB(dip), name,
976	    ZFS_MAX_DATASET_NAME_LEN, snapname);
977	if (error == 0)
978		error = zfs_secpolicy_destroy_perms(snapname, cr);
979	if (error != 0)
980		goto out;
981
982	error = zfsctl_snapshot_unmount(snapname, MNT_FORCE);
983	if ((error == 0) || (error == ENOENT))
984		error = dsl_destroy_snapshot(snapname, B_FALSE);
985out:
986	kmem_free(snapname, ZFS_MAX_DATASET_NAME_LEN);
987	kmem_free(real, ZFS_MAX_DATASET_NAME_LEN);
988
989	zfs_exit(zfsvfs, FTAG);
990
991	return (error);
992}
993
994/*
995 * Creating a directory under '.zfs/snapshot' will automatically trigger
996 * the creation of a new snapshot with the given name.
997 */
998int
999zfsctl_snapdir_mkdir(struct inode *dip, const char *dirname, vattr_t *vap,
1000    struct inode **ipp, cred_t *cr, int flags)
1001{
1002	zfsvfs_t *zfsvfs = ITOZSB(dip);
1003	char *dsname;
1004	int error;
1005
1006	if (!zfs_admin_snapshot)
1007		return (SET_ERROR(EACCES));
1008
1009	dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
1010
1011	if (zfs_component_namecheck(dirname, NULL, NULL) != 0) {
1012		error = SET_ERROR(EILSEQ);
1013		goto out;
1014	}
1015
1016	dmu_objset_name(zfsvfs->z_os, dsname);
1017
1018	error = zfs_secpolicy_snapshot_perms(dsname, cr);
1019	if (error != 0)
1020		goto out;
1021
1022	if (error == 0) {
1023		error = dmu_objset_snapshot_one(dsname, dirname);
1024		if (error != 0)
1025			goto out;
1026
1027		error = zfsctl_snapdir_lookup(dip, dirname, ipp,
1028		    0, cr, NULL, NULL);
1029	}
1030out:
1031	kmem_free(dsname, ZFS_MAX_DATASET_NAME_LEN);
1032
1033	return (error);
1034}
1035
1036/*
1037 * Flush everything out of the kernel's export table and such.
1038 * This is needed as once the snapshot is used over NFS, its
1039 * entries in svc_export and svc_expkey caches hold reference
1040 * to the snapshot mount point. There is no known way of flushing
1041 * only the entries related to the snapshot.
1042 */
1043static void
1044exportfs_flush(void)
1045{
1046	char *argv[] = { "/usr/sbin/exportfs", "-f", NULL };
1047	char *envp[] = { NULL };
1048
1049	(void) call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1050}
1051
1052/*
1053 * Attempt to unmount a snapshot by making a call to user space.
1054 * There is no assurance that this can or will succeed, is just a
1055 * best effort.  In the case where it does fail, perhaps because
1056 * it's in use, the unmount will fail harmlessly.
1057 */
1058int
1059zfsctl_snapshot_unmount(const char *snapname, int flags)
1060{
1061	char *argv[] = { "/usr/bin/env", "umount", "-t", "zfs", "-n", NULL,
1062	    NULL };
1063	char *envp[] = { NULL };
1064	zfs_snapentry_t *se;
1065	int error;
1066
1067	rw_enter(&zfs_snapshot_lock, RW_READER);
1068	if ((se = zfsctl_snapshot_find_by_name(snapname)) == NULL) {
1069		rw_exit(&zfs_snapshot_lock);
1070		return (SET_ERROR(ENOENT));
1071	}
1072	rw_exit(&zfs_snapshot_lock);
1073
1074	exportfs_flush();
1075
1076	if (flags & MNT_FORCE)
1077		argv[4] = "-fn";
1078	argv[5] = se->se_path;
1079	dprintf("unmount; path=%s\n", se->se_path);
1080	error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1081	zfsctl_snapshot_rele(se);
1082
1083
1084	/*
1085	 * The umount system utility will return 256 on error.  We must
1086	 * assume this error is because the file system is busy so it is
1087	 * converted to the more sensible EBUSY.
1088	 */
1089	if (error)
1090		error = SET_ERROR(EBUSY);
1091
1092	return (error);
1093}
1094
1095int
1096zfsctl_snapshot_mount(struct path *path, int flags)
1097{
1098	struct dentry *dentry = path->dentry;
1099	struct inode *ip = dentry->d_inode;
1100	zfsvfs_t *zfsvfs;
1101	zfsvfs_t *snap_zfsvfs;
1102	zfs_snapentry_t *se;
1103	char *full_name, *full_path;
1104	char *argv[] = { "/usr/bin/env", "mount", "-t", "zfs", "-n", NULL, NULL,
1105	    NULL };
1106	char *envp[] = { NULL };
1107	int error;
1108	struct path spath;
1109
1110	if (ip == NULL)
1111		return (SET_ERROR(EISDIR));
1112
1113	zfsvfs = ITOZSB(ip);
1114	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
1115		return (error);
1116
1117	full_name = kmem_zalloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
1118	full_path = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1119
1120	error = zfsctl_snapshot_name(zfsvfs, dname(dentry),
1121	    ZFS_MAX_DATASET_NAME_LEN, full_name);
1122	if (error)
1123		goto error;
1124
1125	/*
1126	 * Construct a mount point path from sb of the ctldir inode and dirent
1127	 * name, instead of from d_path(), so that chroot'd process doesn't fail
1128	 * on mount.zfs(8).
1129	 */
1130	snprintf(full_path, MAXPATHLEN, "%s/.zfs/snapshot/%s",
1131	    zfsvfs->z_vfs->vfs_mntpoint ? zfsvfs->z_vfs->vfs_mntpoint : "",
1132	    dname(dentry));
1133
1134	/*
1135	 * Multiple concurrent automounts of a snapshot are never allowed.
1136	 * The snapshot may be manually mounted as many times as desired.
1137	 */
1138	if (zfsctl_snapshot_ismounted(full_name)) {
1139		error = 0;
1140		goto error;
1141	}
1142
1143	/*
1144	 * Attempt to mount the snapshot from user space.  Normally this
1145	 * would be done using the vfs_kern_mount() function, however that
1146	 * function is marked GPL-only and cannot be used.  On error we
1147	 * careful to log the real error to the console and return EISDIR
1148	 * to safely abort the automount.  This should be very rare.
1149	 *
1150	 * If the user mode helper happens to return EBUSY, a concurrent
1151	 * mount is already in progress in which case the error is ignored.
1152	 * Take note that if the program was executed successfully the return
1153	 * value from call_usermodehelper() will be (exitcode << 8 + signal).
1154	 */
1155	dprintf("mount; name=%s path=%s\n", full_name, full_path);
1156	argv[5] = full_name;
1157	argv[6] = full_path;
1158	error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1159	if (error) {
1160		if (!(error & MOUNT_BUSY << 8)) {
1161			zfs_dbgmsg("Unable to automount %s error=%d",
1162			    full_path, error);
1163			error = SET_ERROR(EISDIR);
1164		} else {
1165			/*
1166			 * EBUSY, this could mean a concurrent mount, or the
1167			 * snapshot has already been mounted at completely
1168			 * different place. We return 0 so VFS will retry. For
1169			 * the latter case the VFS will retry several times
1170			 * and return ELOOP, which is probably not a very good
1171			 * behavior.
1172			 */
1173			error = 0;
1174		}
1175		goto error;
1176	}
1177
1178	/*
1179	 * Follow down in to the mounted snapshot and set MNT_SHRINKABLE
1180	 * to identify this as an automounted filesystem.
1181	 */
1182	spath = *path;
1183	path_get(&spath);
1184	if (follow_down_one(&spath)) {
1185		snap_zfsvfs = ITOZSB(spath.dentry->d_inode);
1186		snap_zfsvfs->z_parent = zfsvfs;
1187		dentry = spath.dentry;
1188		spath.mnt->mnt_flags |= MNT_SHRINKABLE;
1189
1190		rw_enter(&zfs_snapshot_lock, RW_WRITER);
1191		se = zfsctl_snapshot_alloc(full_name, full_path,
1192		    snap_zfsvfs->z_os->os_spa, dmu_objset_id(snap_zfsvfs->z_os),
1193		    dentry);
1194		zfsctl_snapshot_add(se);
1195		zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot);
1196		rw_exit(&zfs_snapshot_lock);
1197	}
1198	path_put(&spath);
1199error:
1200	kmem_free(full_name, ZFS_MAX_DATASET_NAME_LEN);
1201	kmem_free(full_path, MAXPATHLEN);
1202
1203	zfs_exit(zfsvfs, FTAG);
1204
1205	return (error);
1206}
1207
1208/*
1209 * Get the snapdir inode from fid
1210 */
1211int
1212zfsctl_snapdir_vget(struct super_block *sb, uint64_t objsetid, int gen,
1213    struct inode **ipp)
1214{
1215	int error;
1216	struct path path;
1217	char *mnt;
1218	struct dentry *dentry;
1219
1220	mnt = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1221
1222	error = zfsctl_snapshot_path_objset(sb->s_fs_info, objsetid,
1223	    MAXPATHLEN, mnt);
1224	if (error)
1225		goto out;
1226
1227	/* Trigger automount */
1228	error = -kern_path(mnt, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &path);
1229	if (error)
1230		goto out;
1231
1232	path_put(&path);
1233	/*
1234	 * Get the snapdir inode. Note, we don't want to use the above
1235	 * path because it contains the root of the snapshot rather
1236	 * than the snapdir.
1237	 */
1238	*ipp = ilookup(sb, ZFSCTL_INO_SNAPDIRS - objsetid);
1239	if (*ipp == NULL) {
1240		error = SET_ERROR(ENOENT);
1241		goto out;
1242	}
1243
1244	/* check gen, see zfsctl_snapdir_fid */
1245	dentry = d_obtain_alias(igrab(*ipp));
1246	if (gen != (!IS_ERR(dentry) && d_mountpoint(dentry))) {
1247		iput(*ipp);
1248		*ipp = NULL;
1249		error = SET_ERROR(ENOENT);
1250	}
1251	if (!IS_ERR(dentry))
1252		dput(dentry);
1253out:
1254	kmem_free(mnt, MAXPATHLEN);
1255	return (error);
1256}
1257
1258int
1259zfsctl_shares_lookup(struct inode *dip, char *name, struct inode **ipp,
1260    int flags, cred_t *cr, int *direntflags, pathname_t *realpnp)
1261{
1262	zfsvfs_t *zfsvfs = ITOZSB(dip);
1263	znode_t *zp;
1264	znode_t *dzp;
1265	int error;
1266
1267	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
1268		return (error);
1269
1270	if (zfsvfs->z_shares_dir == 0) {
1271		zfs_exit(zfsvfs, FTAG);
1272		return (SET_ERROR(ENOTSUP));
1273	}
1274
1275	if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1276		error = zfs_lookup(dzp, name, &zp, 0, cr, NULL, NULL);
1277		zrele(dzp);
1278	}
1279
1280	zfs_exit(zfsvfs, FTAG);
1281
1282	return (error);
1283}
1284
1285/*
1286 * Initialize the various pieces we'll need to create and manipulate .zfs
1287 * directories.  Currently this is unused but available.
1288 */
1289void
1290zfsctl_init(void)
1291{
1292	avl_create(&zfs_snapshots_by_name, snapentry_compare_by_name,
1293	    sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t,
1294	    se_node_name));
1295	avl_create(&zfs_snapshots_by_objsetid, snapentry_compare_by_objsetid,
1296	    sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t,
1297	    se_node_objsetid));
1298	rw_init(&zfs_snapshot_lock, NULL, RW_DEFAULT, NULL);
1299}
1300
1301/*
1302 * Cleanup the various pieces we needed for .zfs directories.  In particular
1303 * ensure the expiry timer is canceled safely.
1304 */
1305void
1306zfsctl_fini(void)
1307{
1308	avl_destroy(&zfs_snapshots_by_name);
1309	avl_destroy(&zfs_snapshots_by_objsetid);
1310	rw_destroy(&zfs_snapshot_lock);
1311}
1312
1313module_param(zfs_admin_snapshot, int, 0644);
1314MODULE_PARM_DESC(zfs_admin_snapshot, "Enable mkdir/rmdir/mv in .zfs/snapshot");
1315
1316module_param(zfs_expire_snapshot, int, 0644);
1317MODULE_PARM_DESC(zfs_expire_snapshot, "Seconds to expire .zfs/snapshot");
1318