xref: /freebsd-13-stable/sys/contrib/openzfs/module/os/linux/zfs/zfs_ctldir.c (revision b9c2c366db1beb2ed276947056f45938ad8f57ec)
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  *
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  */
105 static avl_tree_t zfs_snapshots_by_name;
106 static avl_tree_t zfs_snapshots_by_objsetid;
107 static krwlock_t zfs_snapshot_lock;
108 
109 /*
110  * Control Directory Tunables (.zfs)
111  */
112 int zfs_expire_snapshot = ZFSCTL_EXPIRE_SNAPSHOT;
113 int zfs_admin_snapshot = 0;
114 
115 typedef 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 
128 static 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  */
134 static zfs_snapentry_t *
zfsctl_snapshot_alloc(const char * full_name,const char * full_path,spa_t * spa,uint64_t objsetid,struct dentry * root_dentry)135 zfsctl_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  */
159 static void
zfsctl_snapshot_free(zfs_snapentry_t * se)160 zfsctl_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  */
173 static void
zfsctl_snapshot_hold(zfs_snapentry_t * se)174 zfsctl_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  */
183 static void
zfsctl_snapshot_rele(zfs_snapentry_t * se)184 zfsctl_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  */
195 static void
zfsctl_snapshot_add(zfs_snapentry_t * se)196 zfsctl_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  */
210 static void
zfsctl_snapshot_remove(zfs_snapentry_t * se)211 zfsctl_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  */
222 static int
snapentry_compare_by_name(const void * a,const void * b)223 snapentry_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  */
242 static int
snapentry_compare_by_objsetid(const void * a,const void * b)243 snapentry_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  */
266 static zfs_snapentry_t *
zfsctl_snapshot_find_by_name(const char * snapname)267 zfsctl_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  */
286 static zfs_snapentry_t *
zfsctl_snapshot_find_by_objsetid(spa_t * spa,uint64_t objsetid)287 zfsctl_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  */
306 static int
zfsctl_snapshot_rename(const char * old_snapname,const char * new_snapname)307 zfsctl_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  */
329 static void
snapentry_expire(void * data)330 snapentry_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  */
364 static void
zfsctl_snapshot_unmount_cancel(zfs_snapentry_t * se)365 zfsctl_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  */
386 static void
zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t * se,int delay)387 zfsctl_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  */
420 int
zfsctl_snapshot_unmount_delay(spa_t * spa,uint64_t objsetid,int delay)421 zfsctl_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  */
442 static boolean_t
zfsctl_snapshot_ismounted(const char * snapname)443 zfsctl_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  */
461 boolean_t
zfsctl_is_node(struct inode * ip)462 zfsctl_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  */
470 boolean_t
zfsctl_is_snapdir(struct inode * ip)471 zfsctl_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  */
479 static struct inode *
zfsctl_inode_alloc(zfsvfs_t * zfsvfs,uint64_t id,const struct file_operations * fops,const struct inode_operations * ops)480 zfsctl_inode_alloc(zfsvfs_t *zfsvfs, uint64_t id,
481     const struct file_operations *fops, const struct inode_operations *ops)
482 {
483 	inode_timespec_t now;
484 	struct inode *ip;
485 	znode_t *zp;
486 
487 	ip = new_inode(zfsvfs->z_sb);
488 	if (ip == NULL)
489 		return (NULL);
490 
491 	now = current_time(ip);
492 	zp = ITOZ(ip);
493 	ASSERT3P(zp->z_dirlocks, ==, NULL);
494 	ASSERT3P(zp->z_acl_cached, ==, NULL);
495 	ASSERT3P(zp->z_xattr_cached, ==, NULL);
496 	zp->z_id = id;
497 	zp->z_unlinked = B_FALSE;
498 	zp->z_atime_dirty = B_FALSE;
499 	zp->z_zn_prefetch = B_FALSE;
500 	zp->z_is_sa = B_FALSE;
501 #if !defined(HAVE_FILEMAP_RANGE_HAS_PAGE)
502 	zp->z_is_mapped = B_FALSE;
503 #endif
504 	zp->z_is_ctldir = B_TRUE;
505 	zp->z_sa_hdl = NULL;
506 	zp->z_blksz = 0;
507 	zp->z_seq = 0;
508 	zp->z_mapcnt = 0;
509 	zp->z_size = 0;
510 	zp->z_pflags = 0;
511 	zp->z_mode = 0;
512 	zp->z_sync_cnt = 0;
513 	zp->z_sync_writes_cnt = 0;
514 	zp->z_async_writes_cnt = 0;
515 	ip->i_generation = 0;
516 	ip->i_ino = id;
517 	ip->i_mode = (S_IFDIR | S_IRWXUGO);
518 	ip->i_uid = SUID_TO_KUID(0);
519 	ip->i_gid = SGID_TO_KGID(0);
520 	ip->i_blkbits = SPA_MINBLOCKSHIFT;
521 	zpl_inode_set_atime_to_ts(ip, now);
522 	zpl_inode_set_mtime_to_ts(ip, now);
523 	zpl_inode_set_ctime_to_ts(ip, now);
524 	ip->i_fop = fops;
525 	ip->i_op = ops;
526 #if defined(IOP_XATTR)
527 	ip->i_opflags &= ~IOP_XATTR;
528 #endif
529 
530 	if (insert_inode_locked(ip)) {
531 		unlock_new_inode(ip);
532 		iput(ip);
533 		return (NULL);
534 	}
535 
536 	mutex_enter(&zfsvfs->z_znodes_lock);
537 	list_insert_tail(&zfsvfs->z_all_znodes, zp);
538 	zfsvfs->z_nr_znodes++;
539 	membar_producer();
540 	mutex_exit(&zfsvfs->z_znodes_lock);
541 
542 	unlock_new_inode(ip);
543 
544 	return (ip);
545 }
546 
547 /*
548  * Lookup the inode with given id, it will be allocated if needed.
549  */
550 static struct inode *
zfsctl_inode_lookup(zfsvfs_t * zfsvfs,uint64_t id,const struct file_operations * fops,const struct inode_operations * ops)551 zfsctl_inode_lookup(zfsvfs_t *zfsvfs, uint64_t id,
552     const struct file_operations *fops, const struct inode_operations *ops)
553 {
554 	struct inode *ip = NULL;
555 
556 	while (ip == NULL) {
557 		ip = ilookup(zfsvfs->z_sb, (unsigned long)id);
558 		if (ip)
559 			break;
560 
561 		/* May fail due to concurrent zfsctl_inode_alloc() */
562 		ip = zfsctl_inode_alloc(zfsvfs, id, fops, ops);
563 	}
564 
565 	return (ip);
566 }
567 
568 /*
569  * Create the '.zfs' directory.  This directory is cached as part of the VFS
570  * structure.  This results in a hold on the zfsvfs_t.  The code in zfs_umount()
571  * therefore checks against a vfs_count of 2 instead of 1.  This reference
572  * is removed when the ctldir is destroyed in the unmount.  All other entities
573  * under the '.zfs' directory are created dynamically as needed.
574  *
575  * Because the dynamically created '.zfs' directory entries assume the use
576  * of 64-bit inode numbers this support must be disabled on 32-bit systems.
577  */
578 int
zfsctl_create(zfsvfs_t * zfsvfs)579 zfsctl_create(zfsvfs_t *zfsvfs)
580 {
581 	ASSERT(zfsvfs->z_ctldir == NULL);
582 
583 	zfsvfs->z_ctldir = zfsctl_inode_alloc(zfsvfs, ZFSCTL_INO_ROOT,
584 	    &zpl_fops_root, &zpl_ops_root);
585 	if (zfsvfs->z_ctldir == NULL)
586 		return (SET_ERROR(ENOENT));
587 
588 	return (0);
589 }
590 
591 /*
592  * Destroy the '.zfs' directory or remove a snapshot from zfs_snapshots_by_name.
593  * Only called when the filesystem is unmounted.
594  */
595 void
zfsctl_destroy(zfsvfs_t * zfsvfs)596 zfsctl_destroy(zfsvfs_t *zfsvfs)
597 {
598 	if (zfsvfs->z_issnap) {
599 		zfs_snapentry_t *se;
600 		spa_t *spa = zfsvfs->z_os->os_spa;
601 		uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
602 
603 		rw_enter(&zfs_snapshot_lock, RW_WRITER);
604 		se = zfsctl_snapshot_find_by_objsetid(spa, objsetid);
605 		if (se != NULL)
606 			zfsctl_snapshot_remove(se);
607 		rw_exit(&zfs_snapshot_lock);
608 		if (se != NULL) {
609 			zfsctl_snapshot_unmount_cancel(se);
610 			zfsctl_snapshot_rele(se);
611 		}
612 	} else if (zfsvfs->z_ctldir) {
613 		iput(zfsvfs->z_ctldir);
614 		zfsvfs->z_ctldir = NULL;
615 	}
616 }
617 
618 /*
619  * Given a root znode, retrieve the associated .zfs directory.
620  * Add a hold to the vnode and return it.
621  */
622 struct inode *
zfsctl_root(znode_t * zp)623 zfsctl_root(znode_t *zp)
624 {
625 	ASSERT(zfs_has_ctldir(zp));
626 	/* Must have an existing ref, so igrab() cannot return NULL */
627 	VERIFY3P(igrab(ZTOZSB(zp)->z_ctldir), !=, NULL);
628 	return (ZTOZSB(zp)->z_ctldir);
629 }
630 
631 /*
632  * Generate a long fid to indicate a snapdir. We encode whether snapdir is
633  * already mounted in gen field. We do this because nfsd lookup will not
634  * trigger automount. Next time the nfsd does fh_to_dentry, we will notice
635  * this and do automount and return ESTALE to force nfsd revalidate and follow
636  * mount.
637  */
638 static int
zfsctl_snapdir_fid(struct inode * ip,fid_t * fidp)639 zfsctl_snapdir_fid(struct inode *ip, fid_t *fidp)
640 {
641 	zfid_short_t *zfid = (zfid_short_t *)fidp;
642 	zfid_long_t *zlfid = (zfid_long_t *)fidp;
643 	uint32_t gen = 0;
644 	uint64_t object;
645 	uint64_t objsetid;
646 	int i;
647 	struct dentry *dentry;
648 
649 	if (fidp->fid_len < LONG_FID_LEN) {
650 		fidp->fid_len = LONG_FID_LEN;
651 		return (SET_ERROR(ENOSPC));
652 	}
653 
654 	object = ip->i_ino;
655 	objsetid = ZFSCTL_INO_SNAPDIRS - ip->i_ino;
656 	zfid->zf_len = LONG_FID_LEN;
657 
658 	dentry = d_obtain_alias(igrab(ip));
659 	if (!IS_ERR(dentry)) {
660 		gen = !!d_mountpoint(dentry);
661 		dput(dentry);
662 	}
663 
664 	for (i = 0; i < sizeof (zfid->zf_object); i++)
665 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
666 
667 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
668 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
669 
670 	for (i = 0; i < sizeof (zlfid->zf_setid); i++)
671 		zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
672 
673 	for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
674 		zlfid->zf_setgen[i] = 0;
675 
676 	return (0);
677 }
678 
679 /*
680  * Generate an appropriate fid for an entry in the .zfs directory.
681  */
682 int
zfsctl_fid(struct inode * ip,fid_t * fidp)683 zfsctl_fid(struct inode *ip, fid_t *fidp)
684 {
685 	znode_t		*zp = ITOZ(ip);
686 	zfsvfs_t	*zfsvfs = ITOZSB(ip);
687 	uint64_t	object = zp->z_id;
688 	zfid_short_t	*zfid;
689 	int		i;
690 
691 	ZFS_ENTER(zfsvfs);
692 
693 	if (zfsctl_is_snapdir(ip)) {
694 		ZFS_EXIT(zfsvfs);
695 		return (zfsctl_snapdir_fid(ip, fidp));
696 	}
697 
698 	if (fidp->fid_len < SHORT_FID_LEN) {
699 		fidp->fid_len = SHORT_FID_LEN;
700 		ZFS_EXIT(zfsvfs);
701 		return (SET_ERROR(ENOSPC));
702 	}
703 
704 	zfid = (zfid_short_t *)fidp;
705 
706 	zfid->zf_len = SHORT_FID_LEN;
707 
708 	for (i = 0; i < sizeof (zfid->zf_object); i++)
709 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
710 
711 	/* .zfs znodes always have a generation number of 0 */
712 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
713 		zfid->zf_gen[i] = 0;
714 
715 	ZFS_EXIT(zfsvfs);
716 	return (0);
717 }
718 
719 /*
720  * Construct a full dataset name in full_name: "pool/dataset@snap_name"
721  */
722 static int
zfsctl_snapshot_name(zfsvfs_t * zfsvfs,const char * snap_name,int len,char * full_name)723 zfsctl_snapshot_name(zfsvfs_t *zfsvfs, const char *snap_name, int len,
724     char *full_name)
725 {
726 	objset_t *os = zfsvfs->z_os;
727 
728 	if (zfs_component_namecheck(snap_name, NULL, NULL) != 0)
729 		return (SET_ERROR(EILSEQ));
730 
731 	dmu_objset_name(os, full_name);
732 	if ((strlen(full_name) + 1 + strlen(snap_name)) >= len)
733 		return (SET_ERROR(ENAMETOOLONG));
734 
735 	(void) strcat(full_name, "@");
736 	(void) strcat(full_name, snap_name);
737 
738 	return (0);
739 }
740 
741 /*
742  * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/"
743  */
744 static int
zfsctl_snapshot_path_objset(zfsvfs_t * zfsvfs,uint64_t objsetid,int path_len,char * full_path)745 zfsctl_snapshot_path_objset(zfsvfs_t *zfsvfs, uint64_t objsetid,
746     int path_len, char *full_path)
747 {
748 	objset_t *os = zfsvfs->z_os;
749 	fstrans_cookie_t cookie;
750 	char *snapname;
751 	boolean_t case_conflict;
752 	uint64_t id, pos = 0;
753 	int error = 0;
754 
755 	if (zfsvfs->z_vfs->vfs_mntpoint == NULL)
756 		return (SET_ERROR(ENOENT));
757 
758 	cookie = spl_fstrans_mark();
759 	snapname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
760 
761 	while (error == 0) {
762 		dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
763 		error = dmu_snapshot_list_next(zfsvfs->z_os,
764 		    ZFS_MAX_DATASET_NAME_LEN, snapname, &id, &pos,
765 		    &case_conflict);
766 		dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
767 		if (error)
768 			goto out;
769 
770 		if (id == objsetid)
771 			break;
772 	}
773 
774 	snprintf(full_path, path_len, "%s/.zfs/snapshot/%s",
775 	    zfsvfs->z_vfs->vfs_mntpoint, snapname);
776 out:
777 	kmem_free(snapname, ZFS_MAX_DATASET_NAME_LEN);
778 	spl_fstrans_unmark(cookie);
779 
780 	return (error);
781 }
782 
783 /*
784  * Special case the handling of "..".
785  */
786 int
zfsctl_root_lookup(struct inode * dip,const char * name,struct inode ** ipp,int flags,cred_t * cr,int * direntflags,pathname_t * realpnp)787 zfsctl_root_lookup(struct inode *dip, const char *name, struct inode **ipp,
788     int flags, cred_t *cr, int *direntflags, pathname_t *realpnp)
789 {
790 	zfsvfs_t *zfsvfs = ITOZSB(dip);
791 	int error = 0;
792 
793 	ZFS_ENTER(zfsvfs);
794 
795 	if (strcmp(name, "..") == 0) {
796 		*ipp = dip->i_sb->s_root->d_inode;
797 	} else if (strcmp(name, ZFS_SNAPDIR_NAME) == 0) {
798 		*ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SNAPDIR,
799 		    &zpl_fops_snapdir, &zpl_ops_snapdir);
800 	} else if (strcmp(name, ZFS_SHAREDIR_NAME) == 0) {
801 		*ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SHARES,
802 		    &zpl_fops_shares, &zpl_ops_shares);
803 	} else {
804 		*ipp = NULL;
805 	}
806 
807 	if (*ipp == NULL)
808 		error = SET_ERROR(ENOENT);
809 
810 	ZFS_EXIT(zfsvfs);
811 
812 	return (error);
813 }
814 
815 /*
816  * Lookup entry point for the 'snapshot' directory.  Try to open the
817  * snapshot if it exist, creating the pseudo filesystem inode as necessary.
818  */
819 int
zfsctl_snapdir_lookup(struct inode * dip,const char * name,struct inode ** ipp,int flags,cred_t * cr,int * direntflags,pathname_t * realpnp)820 zfsctl_snapdir_lookup(struct inode *dip, const char *name, struct inode **ipp,
821     int flags, cred_t *cr, int *direntflags, pathname_t *realpnp)
822 {
823 	zfsvfs_t *zfsvfs = ITOZSB(dip);
824 	uint64_t id;
825 	int error;
826 
827 	ZFS_ENTER(zfsvfs);
828 
829 	error = dmu_snapshot_lookup(zfsvfs->z_os, name, &id);
830 	if (error) {
831 		ZFS_EXIT(zfsvfs);
832 		return (error);
833 	}
834 
835 	*ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SNAPDIRS - id,
836 	    &simple_dir_operations, &simple_dir_inode_operations);
837 	if (*ipp == NULL)
838 		error = SET_ERROR(ENOENT);
839 
840 	ZFS_EXIT(zfsvfs);
841 
842 	return (error);
843 }
844 
845 /*
846  * Renaming a directory under '.zfs/snapshot' will automatically trigger
847  * a rename of the snapshot to the new given name.  The rename is confined
848  * to the '.zfs/snapshot' directory snapshots cannot be moved elsewhere.
849  */
850 int
zfsctl_snapdir_rename(struct inode * sdip,const char * snm,struct inode * tdip,const char * tnm,cred_t * cr,int flags)851 zfsctl_snapdir_rename(struct inode *sdip, const char *snm,
852     struct inode *tdip, const char *tnm, cred_t *cr, int flags)
853 {
854 	zfsvfs_t *zfsvfs = ITOZSB(sdip);
855 	char *to, *from, *real, *fsname;
856 	int error;
857 
858 	if (!zfs_admin_snapshot)
859 		return (SET_ERROR(EACCES));
860 
861 	ZFS_ENTER(zfsvfs);
862 
863 	to = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
864 	from = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
865 	real = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
866 	fsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
867 
868 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
869 		error = dmu_snapshot_realname(zfsvfs->z_os, snm, real,
870 		    ZFS_MAX_DATASET_NAME_LEN, NULL);
871 		if (error == 0) {
872 			snm = real;
873 		} else if (error != ENOTSUP) {
874 			goto out;
875 		}
876 	}
877 
878 	dmu_objset_name(zfsvfs->z_os, fsname);
879 
880 	error = zfsctl_snapshot_name(ITOZSB(sdip), snm,
881 	    ZFS_MAX_DATASET_NAME_LEN, from);
882 	if (error == 0)
883 		error = zfsctl_snapshot_name(ITOZSB(tdip), tnm,
884 		    ZFS_MAX_DATASET_NAME_LEN, to);
885 	if (error == 0)
886 		error = zfs_secpolicy_rename_perms(from, to, cr);
887 	if (error != 0)
888 		goto out;
889 
890 	/*
891 	 * Cannot move snapshots out of the snapdir.
892 	 */
893 	if (sdip != tdip) {
894 		error = SET_ERROR(EINVAL);
895 		goto out;
896 	}
897 
898 	/*
899 	 * No-op when names are identical.
900 	 */
901 	if (strcmp(snm, tnm) == 0) {
902 		error = 0;
903 		goto out;
904 	}
905 
906 	rw_enter(&zfs_snapshot_lock, RW_WRITER);
907 
908 	error = dsl_dataset_rename_snapshot(fsname, snm, tnm, B_FALSE);
909 	if (error == 0)
910 		(void) zfsctl_snapshot_rename(snm, tnm);
911 
912 	rw_exit(&zfs_snapshot_lock);
913 out:
914 	kmem_free(from, ZFS_MAX_DATASET_NAME_LEN);
915 	kmem_free(to, ZFS_MAX_DATASET_NAME_LEN);
916 	kmem_free(real, ZFS_MAX_DATASET_NAME_LEN);
917 	kmem_free(fsname, ZFS_MAX_DATASET_NAME_LEN);
918 
919 	ZFS_EXIT(zfsvfs);
920 
921 	return (error);
922 }
923 
924 /*
925  * Removing a directory under '.zfs/snapshot' will automatically trigger
926  * the removal of the snapshot with the given name.
927  */
928 int
zfsctl_snapdir_remove(struct inode * dip,const char * name,cred_t * cr,int flags)929 zfsctl_snapdir_remove(struct inode *dip, const char *name, cred_t *cr,
930     int flags)
931 {
932 	zfsvfs_t *zfsvfs = ITOZSB(dip);
933 	char *snapname, *real;
934 	int error;
935 
936 	if (!zfs_admin_snapshot)
937 		return (SET_ERROR(EACCES));
938 
939 	ZFS_ENTER(zfsvfs);
940 
941 	snapname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
942 	real = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
943 
944 	if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
945 		error = dmu_snapshot_realname(zfsvfs->z_os, name, real,
946 		    ZFS_MAX_DATASET_NAME_LEN, NULL);
947 		if (error == 0) {
948 			name = real;
949 		} else if (error != ENOTSUP) {
950 			goto out;
951 		}
952 	}
953 
954 	error = zfsctl_snapshot_name(ITOZSB(dip), name,
955 	    ZFS_MAX_DATASET_NAME_LEN, snapname);
956 	if (error == 0)
957 		error = zfs_secpolicy_destroy_perms(snapname, cr);
958 	if (error != 0)
959 		goto out;
960 
961 	error = zfsctl_snapshot_unmount(snapname, MNT_FORCE);
962 	if ((error == 0) || (error == ENOENT))
963 		error = dsl_destroy_snapshot(snapname, B_FALSE);
964 out:
965 	kmem_free(snapname, ZFS_MAX_DATASET_NAME_LEN);
966 	kmem_free(real, ZFS_MAX_DATASET_NAME_LEN);
967 
968 	ZFS_EXIT(zfsvfs);
969 
970 	return (error);
971 }
972 
973 /*
974  * Creating a directory under '.zfs/snapshot' will automatically trigger
975  * the creation of a new snapshot with the given name.
976  */
977 int
zfsctl_snapdir_mkdir(struct inode * dip,const char * dirname,vattr_t * vap,struct inode ** ipp,cred_t * cr,int flags)978 zfsctl_snapdir_mkdir(struct inode *dip, const char *dirname, vattr_t *vap,
979     struct inode **ipp, cred_t *cr, int flags)
980 {
981 	zfsvfs_t *zfsvfs = ITOZSB(dip);
982 	char *dsname;
983 	int error;
984 
985 	if (!zfs_admin_snapshot)
986 		return (SET_ERROR(EACCES));
987 
988 	dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
989 
990 	if (zfs_component_namecheck(dirname, NULL, NULL) != 0) {
991 		error = SET_ERROR(EILSEQ);
992 		goto out;
993 	}
994 
995 	dmu_objset_name(zfsvfs->z_os, dsname);
996 
997 	error = zfs_secpolicy_snapshot_perms(dsname, cr);
998 	if (error != 0)
999 		goto out;
1000 
1001 	if (error == 0) {
1002 		error = dmu_objset_snapshot_one(dsname, dirname);
1003 		if (error != 0)
1004 			goto out;
1005 
1006 		error = zfsctl_snapdir_lookup(dip, dirname, ipp,
1007 		    0, cr, NULL, NULL);
1008 	}
1009 out:
1010 	kmem_free(dsname, ZFS_MAX_DATASET_NAME_LEN);
1011 
1012 	return (error);
1013 }
1014 
1015 /*
1016  * Flush everything out of the kernel's export table and such.
1017  * This is needed as once the snapshot is used over NFS, its
1018  * entries in svc_export and svc_expkey caches hold reference
1019  * to the snapshot mount point. There is no known way of flushing
1020  * only the entries related to the snapshot.
1021  */
1022 static void
exportfs_flush(void)1023 exportfs_flush(void)
1024 {
1025 	char *argv[] = { "/usr/sbin/exportfs", "-f", NULL };
1026 	char *envp[] = { NULL };
1027 
1028 	(void) call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1029 }
1030 
1031 /*
1032  * Attempt to unmount a snapshot by making a call to user space.
1033  * There is no assurance that this can or will succeed, is just a
1034  * best effort.  In the case where it does fail, perhaps because
1035  * it's in use, the unmount will fail harmlessly.
1036  */
1037 int
zfsctl_snapshot_unmount(const char * snapname,int flags)1038 zfsctl_snapshot_unmount(const char *snapname, int flags)
1039 {
1040 	char *argv[] = { "/usr/bin/env", "umount", "-t", "zfs", "-n", NULL,
1041 	    NULL };
1042 	char *envp[] = { NULL };
1043 	zfs_snapentry_t *se;
1044 	int error;
1045 
1046 	rw_enter(&zfs_snapshot_lock, RW_READER);
1047 	if ((se = zfsctl_snapshot_find_by_name(snapname)) == NULL) {
1048 		rw_exit(&zfs_snapshot_lock);
1049 		return (SET_ERROR(ENOENT));
1050 	}
1051 	rw_exit(&zfs_snapshot_lock);
1052 
1053 	exportfs_flush();
1054 
1055 	if (flags & MNT_FORCE)
1056 		argv[4] = "-fn";
1057 	argv[5] = se->se_path;
1058 	dprintf("unmount; path=%s\n", se->se_path);
1059 	error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1060 	zfsctl_snapshot_rele(se);
1061 
1062 
1063 	/*
1064 	 * The umount system utility will return 256 on error.  We must
1065 	 * assume this error is because the file system is busy so it is
1066 	 * converted to the more sensible EBUSY.
1067 	 */
1068 	if (error)
1069 		error = SET_ERROR(EBUSY);
1070 
1071 	return (error);
1072 }
1073 
1074 int
zfsctl_snapshot_mount(struct path * path,int flags)1075 zfsctl_snapshot_mount(struct path *path, int flags)
1076 {
1077 	struct dentry *dentry = path->dentry;
1078 	struct inode *ip = dentry->d_inode;
1079 	zfsvfs_t *zfsvfs;
1080 	zfsvfs_t *snap_zfsvfs;
1081 	zfs_snapentry_t *se;
1082 	char *full_name, *full_path;
1083 	char *argv[] = { "/usr/bin/env", "mount", "-t", "zfs", "-n", NULL, NULL,
1084 	    NULL };
1085 	char *envp[] = { NULL };
1086 	int error;
1087 	struct path spath;
1088 
1089 	if (ip == NULL)
1090 		return (SET_ERROR(EISDIR));
1091 
1092 	zfsvfs = ITOZSB(ip);
1093 	ZFS_ENTER(zfsvfs);
1094 
1095 	full_name = kmem_zalloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
1096 	full_path = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1097 
1098 	error = zfsctl_snapshot_name(zfsvfs, dname(dentry),
1099 	    ZFS_MAX_DATASET_NAME_LEN, full_name);
1100 	if (error)
1101 		goto error;
1102 
1103 	/*
1104 	 * Construct a mount point path from sb of the ctldir inode and dirent
1105 	 * name, instead of from d_path(), so that chroot'd process doesn't fail
1106 	 * on mount.zfs(8).
1107 	 */
1108 	snprintf(full_path, MAXPATHLEN, "%s/.zfs/snapshot/%s",
1109 	    zfsvfs->z_vfs->vfs_mntpoint ? zfsvfs->z_vfs->vfs_mntpoint : "",
1110 	    dname(dentry));
1111 
1112 	/*
1113 	 * Multiple concurrent automounts of a snapshot are never allowed.
1114 	 * The snapshot may be manually mounted as many times as desired.
1115 	 */
1116 	if (zfsctl_snapshot_ismounted(full_name)) {
1117 		error = 0;
1118 		goto error;
1119 	}
1120 
1121 	/*
1122 	 * Attempt to mount the snapshot from user space.  Normally this
1123 	 * would be done using the vfs_kern_mount() function, however that
1124 	 * function is marked GPL-only and cannot be used.  On error we
1125 	 * careful to log the real error to the console and return EISDIR
1126 	 * to safely abort the automount.  This should be very rare.
1127 	 *
1128 	 * If the user mode helper happens to return EBUSY, a concurrent
1129 	 * mount is already in progress in which case the error is ignored.
1130 	 * Take note that if the program was executed successfully the return
1131 	 * value from call_usermodehelper() will be (exitcode << 8 + signal).
1132 	 */
1133 	dprintf("mount; name=%s path=%s\n", full_name, full_path);
1134 	argv[5] = full_name;
1135 	argv[6] = full_path;
1136 	error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1137 	if (error) {
1138 		if (!(error & MOUNT_BUSY << 8)) {
1139 			zfs_dbgmsg("Unable to automount %s error=%d",
1140 			    full_path, error);
1141 			error = SET_ERROR(EISDIR);
1142 		} else {
1143 			/*
1144 			 * EBUSY, this could mean a concurrent mount, or the
1145 			 * snapshot has already been mounted at completely
1146 			 * different place. We return 0 so VFS will retry. For
1147 			 * the latter case the VFS will retry several times
1148 			 * and return ELOOP, which is probably not a very good
1149 			 * behavior.
1150 			 */
1151 			error = 0;
1152 		}
1153 		goto error;
1154 	}
1155 
1156 	/*
1157 	 * Follow down in to the mounted snapshot and set MNT_SHRINKABLE
1158 	 * to identify this as an automounted filesystem.
1159 	 */
1160 	spath = *path;
1161 	path_get(&spath);
1162 	if (follow_down_one(&spath)) {
1163 		snap_zfsvfs = ITOZSB(spath.dentry->d_inode);
1164 		snap_zfsvfs->z_parent = zfsvfs;
1165 		dentry = spath.dentry;
1166 		spath.mnt->mnt_flags |= MNT_SHRINKABLE;
1167 
1168 		rw_enter(&zfs_snapshot_lock, RW_WRITER);
1169 		se = zfsctl_snapshot_alloc(full_name, full_path,
1170 		    snap_zfsvfs->z_os->os_spa, dmu_objset_id(snap_zfsvfs->z_os),
1171 		    dentry);
1172 		zfsctl_snapshot_add(se);
1173 		zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot);
1174 		rw_exit(&zfs_snapshot_lock);
1175 	}
1176 	path_put(&spath);
1177 error:
1178 	kmem_free(full_name, ZFS_MAX_DATASET_NAME_LEN);
1179 	kmem_free(full_path, MAXPATHLEN);
1180 
1181 	ZFS_EXIT(zfsvfs);
1182 
1183 	return (error);
1184 }
1185 
1186 /*
1187  * Get the snapdir inode from fid
1188  */
1189 int
zfsctl_snapdir_vget(struct super_block * sb,uint64_t objsetid,int gen,struct inode ** ipp)1190 zfsctl_snapdir_vget(struct super_block *sb, uint64_t objsetid, int gen,
1191     struct inode **ipp)
1192 {
1193 	int error;
1194 	struct path path;
1195 	char *mnt;
1196 	struct dentry *dentry;
1197 
1198 	mnt = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1199 
1200 	error = zfsctl_snapshot_path_objset(sb->s_fs_info, objsetid,
1201 	    MAXPATHLEN, mnt);
1202 	if (error)
1203 		goto out;
1204 
1205 	/* Trigger automount */
1206 	error = -kern_path(mnt, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &path);
1207 	if (error)
1208 		goto out;
1209 
1210 	path_put(&path);
1211 	/*
1212 	 * Get the snapdir inode. Note, we don't want to use the above
1213 	 * path because it contains the root of the snapshot rather
1214 	 * than the snapdir.
1215 	 */
1216 	*ipp = ilookup(sb, ZFSCTL_INO_SNAPDIRS - objsetid);
1217 	if (*ipp == NULL) {
1218 		error = SET_ERROR(ENOENT);
1219 		goto out;
1220 	}
1221 
1222 	/* check gen, see zfsctl_snapdir_fid */
1223 	dentry = d_obtain_alias(igrab(*ipp));
1224 	if (gen != (!IS_ERR(dentry) && d_mountpoint(dentry))) {
1225 		iput(*ipp);
1226 		*ipp = NULL;
1227 		error = SET_ERROR(ENOENT);
1228 	}
1229 	if (!IS_ERR(dentry))
1230 		dput(dentry);
1231 out:
1232 	kmem_free(mnt, MAXPATHLEN);
1233 	return (error);
1234 }
1235 
1236 int
zfsctl_shares_lookup(struct inode * dip,char * name,struct inode ** ipp,int flags,cred_t * cr,int * direntflags,pathname_t * realpnp)1237 zfsctl_shares_lookup(struct inode *dip, char *name, struct inode **ipp,
1238     int flags, cred_t *cr, int *direntflags, pathname_t *realpnp)
1239 {
1240 	zfsvfs_t *zfsvfs = ITOZSB(dip);
1241 	znode_t *zp;
1242 	znode_t *dzp;
1243 	int error;
1244 
1245 	ZFS_ENTER(zfsvfs);
1246 
1247 	if (zfsvfs->z_shares_dir == 0) {
1248 		ZFS_EXIT(zfsvfs);
1249 		return (SET_ERROR(ENOTSUP));
1250 	}
1251 
1252 	if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1253 		error = zfs_lookup(dzp, name, &zp, 0, cr, NULL, NULL);
1254 		zrele(dzp);
1255 	}
1256 
1257 	ZFS_EXIT(zfsvfs);
1258 
1259 	return (error);
1260 }
1261 
1262 /*
1263  * Initialize the various pieces we'll need to create and manipulate .zfs
1264  * directories.  Currently this is unused but available.
1265  */
1266 void
zfsctl_init(void)1267 zfsctl_init(void)
1268 {
1269 	avl_create(&zfs_snapshots_by_name, snapentry_compare_by_name,
1270 	    sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t,
1271 	    se_node_name));
1272 	avl_create(&zfs_snapshots_by_objsetid, snapentry_compare_by_objsetid,
1273 	    sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t,
1274 	    se_node_objsetid));
1275 	rw_init(&zfs_snapshot_lock, NULL, RW_DEFAULT, NULL);
1276 }
1277 
1278 /*
1279  * Cleanup the various pieces we needed for .zfs directories.  In particular
1280  * ensure the expiry timer is canceled safely.
1281  */
1282 void
zfsctl_fini(void)1283 zfsctl_fini(void)
1284 {
1285 	avl_destroy(&zfs_snapshots_by_name);
1286 	avl_destroy(&zfs_snapshots_by_objsetid);
1287 	rw_destroy(&zfs_snapshot_lock);
1288 }
1289 
1290 module_param(zfs_admin_snapshot, int, 0644);
1291 MODULE_PARM_DESC(zfs_admin_snapshot, "Enable mkdir/rmdir/mv in .zfs/snapshot");
1292 
1293 module_param(zfs_expire_snapshot, int, 0644);
1294 MODULE_PARM_DESC(zfs_expire_snapshot, "Seconds to expire .zfs/snapshot");
1295