xref: /freebsd-14-stable/sys/contrib/openzfs/module/os/linux/zfs/zfs_ctldir.c (revision 2ec8b69480708185a273254e4e254140eb2ce633)
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  */
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 static 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,uint64_t creation)480 zfsctl_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 	zp->z_is_ctldir = B_TRUE;
504 	zp->z_sa_hdl = NULL;
505 	zp->z_blksz = 0;
506 	zp->z_seq = 0;
507 	zp->z_mapcnt = 0;
508 	zp->z_size = 0;
509 	zp->z_pflags = 0;
510 	zp->z_mode = 0;
511 	zp->z_sync_cnt = 0;
512 	zp->z_sync_writes_cnt = 0;
513 	zp->z_async_writes_cnt = 0;
514 	ip->i_generation = 0;
515 	ip->i_ino = id;
516 	ip->i_mode = (S_IFDIR | S_IRWXUGO);
517 	ip->i_uid = SUID_TO_KUID(0);
518 	ip->i_gid = SGID_TO_KGID(0);
519 	ip->i_blkbits = SPA_MINBLOCKSHIFT;
520 	zpl_inode_set_atime_to_ts(ip, now);
521 	zpl_inode_set_mtime_to_ts(ip, now);
522 	zpl_inode_set_ctime_to_ts(ip, now);
523 	ip->i_fop = fops;
524 	ip->i_op = ops;
525 #if defined(IOP_XATTR)
526 	ip->i_opflags &= ~IOP_XATTR;
527 #endif
528 
529 	if (insert_inode_locked(ip)) {
530 		unlock_new_inode(ip);
531 		iput(ip);
532 		return (NULL);
533 	}
534 
535 	mutex_enter(&zfsvfs->z_znodes_lock);
536 	list_insert_tail(&zfsvfs->z_all_znodes, zp);
537 	membar_producer();
538 	mutex_exit(&zfsvfs->z_znodes_lock);
539 
540 	unlock_new_inode(ip);
541 
542 	return (ip);
543 }
544 
545 /*
546  * Lookup the inode with given id, it will be allocated if needed.
547  */
548 static struct inode *
zfsctl_inode_lookup(zfsvfs_t * zfsvfs,uint64_t id,const struct file_operations * fops,const struct inode_operations * ops)549 zfsctl_inode_lookup(zfsvfs_t *zfsvfs, uint64_t id,
550     const struct file_operations *fops, const struct inode_operations *ops)
551 {
552 	struct inode *ip = NULL;
553 	uint64_t creation = 0;
554 	dsl_dataset_t *snap_ds;
555 	dsl_pool_t *pool;
556 
557 	while (ip == NULL) {
558 		ip = ilookup(zfsvfs->z_sb, (unsigned long)id);
559 		if (ip)
560 			break;
561 
562 		if (id <= ZFSCTL_INO_SNAPDIRS && !creation) {
563 			pool = dmu_objset_pool(zfsvfs->z_os);
564 			dsl_pool_config_enter(pool, FTAG);
565 			if (!dsl_dataset_hold_obj(pool,
566 			    ZFSCTL_INO_SNAPDIRS - id, FTAG, &snap_ds)) {
567 				creation = dsl_get_creation(snap_ds);
568 				dsl_dataset_rele(snap_ds, FTAG);
569 			}
570 			dsl_pool_config_exit(pool, FTAG);
571 		}
572 
573 		/* May fail due to concurrent zfsctl_inode_alloc() */
574 		ip = zfsctl_inode_alloc(zfsvfs, id, fops, ops, creation);
575 	}
576 
577 	return (ip);
578 }
579 
580 /*
581  * Create the '.zfs' directory.  This directory is cached as part of the VFS
582  * structure.  This results in a hold on the zfsvfs_t.  The code in zfs_umount()
583  * therefore checks against a vfs_count of 2 instead of 1.  This reference
584  * is removed when the ctldir is destroyed in the unmount.  All other entities
585  * under the '.zfs' directory are created dynamically as needed.
586  *
587  * Because the dynamically created '.zfs' directory entries assume the use
588  * of 64-bit inode numbers this support must be disabled on 32-bit systems.
589  */
590 int
zfsctl_create(zfsvfs_t * zfsvfs)591 zfsctl_create(zfsvfs_t *zfsvfs)
592 {
593 	ASSERT(zfsvfs->z_ctldir == NULL);
594 
595 	zfsvfs->z_ctldir = zfsctl_inode_alloc(zfsvfs, ZFSCTL_INO_ROOT,
596 	    &zpl_fops_root, &zpl_ops_root, 0);
597 	if (zfsvfs->z_ctldir == NULL)
598 		return (SET_ERROR(ENOENT));
599 
600 	return (0);
601 }
602 
603 /*
604  * Destroy the '.zfs' directory or remove a snapshot from zfs_snapshots_by_name.
605  * Only called when the filesystem is unmounted.
606  */
607 void
zfsctl_destroy(zfsvfs_t * zfsvfs)608 zfsctl_destroy(zfsvfs_t *zfsvfs)
609 {
610 	if (zfsvfs->z_issnap) {
611 		zfs_snapentry_t *se;
612 		spa_t *spa = zfsvfs->z_os->os_spa;
613 		uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
614 
615 		rw_enter(&zfs_snapshot_lock, RW_WRITER);
616 		se = zfsctl_snapshot_find_by_objsetid(spa, objsetid);
617 		if (se != NULL)
618 			zfsctl_snapshot_remove(se);
619 		rw_exit(&zfs_snapshot_lock);
620 		if (se != NULL) {
621 			zfsctl_snapshot_unmount_cancel(se);
622 			zfsctl_snapshot_rele(se);
623 		}
624 	} else if (zfsvfs->z_ctldir) {
625 		iput(zfsvfs->z_ctldir);
626 		zfsvfs->z_ctldir = NULL;
627 	}
628 }
629 
630 /*
631  * Given a root znode, retrieve the associated .zfs directory.
632  * Add a hold to the vnode and return it.
633  */
634 struct inode *
zfsctl_root(znode_t * zp)635 zfsctl_root(znode_t *zp)
636 {
637 	ASSERT(zfs_has_ctldir(zp));
638 	/* Must have an existing ref, so igrab() cannot return NULL */
639 	VERIFY3P(igrab(ZTOZSB(zp)->z_ctldir), !=, NULL);
640 	return (ZTOZSB(zp)->z_ctldir);
641 }
642 
643 /*
644  * Generate a long fid to indicate a snapdir. We encode whether snapdir is
645  * already mounted in gen field. We do this because nfsd lookup will not
646  * trigger automount. Next time the nfsd does fh_to_dentry, we will notice
647  * this and do automount and return ESTALE to force nfsd revalidate and follow
648  * mount.
649  */
650 static int
zfsctl_snapdir_fid(struct inode * ip,fid_t * fidp)651 zfsctl_snapdir_fid(struct inode *ip, fid_t *fidp)
652 {
653 	zfid_short_t *zfid = (zfid_short_t *)fidp;
654 	zfid_long_t *zlfid = (zfid_long_t *)fidp;
655 	uint32_t gen = 0;
656 	uint64_t object;
657 	uint64_t objsetid;
658 	int i;
659 	struct dentry *dentry;
660 
661 	if (fidp->fid_len < LONG_FID_LEN) {
662 		fidp->fid_len = LONG_FID_LEN;
663 		return (SET_ERROR(ENOSPC));
664 	}
665 
666 	object = ip->i_ino;
667 	objsetid = ZFSCTL_INO_SNAPDIRS - ip->i_ino;
668 	zfid->zf_len = LONG_FID_LEN;
669 
670 	dentry = d_obtain_alias(igrab(ip));
671 	if (!IS_ERR(dentry)) {
672 		gen = !!d_mountpoint(dentry);
673 		dput(dentry);
674 	}
675 
676 	for (i = 0; i < sizeof (zfid->zf_object); i++)
677 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
678 
679 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
680 		zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
681 
682 	for (i = 0; i < sizeof (zlfid->zf_setid); i++)
683 		zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
684 
685 	for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
686 		zlfid->zf_setgen[i] = 0;
687 
688 	return (0);
689 }
690 
691 /*
692  * Generate an appropriate fid for an entry in the .zfs directory.
693  */
694 int
zfsctl_fid(struct inode * ip,fid_t * fidp)695 zfsctl_fid(struct inode *ip, fid_t *fidp)
696 {
697 	znode_t		*zp = ITOZ(ip);
698 	zfsvfs_t	*zfsvfs = ITOZSB(ip);
699 	uint64_t	object = zp->z_id;
700 	zfid_short_t	*zfid;
701 	int		i;
702 	int		error;
703 
704 	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
705 		return (error);
706 
707 	if (zfsctl_is_snapdir(ip)) {
708 		zfs_exit(zfsvfs, FTAG);
709 		return (zfsctl_snapdir_fid(ip, fidp));
710 	}
711 
712 	if (fidp->fid_len < SHORT_FID_LEN) {
713 		fidp->fid_len = SHORT_FID_LEN;
714 		zfs_exit(zfsvfs, FTAG);
715 		return (SET_ERROR(ENOSPC));
716 	}
717 
718 	zfid = (zfid_short_t *)fidp;
719 
720 	zfid->zf_len = SHORT_FID_LEN;
721 
722 	for (i = 0; i < sizeof (zfid->zf_object); i++)
723 		zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
724 
725 	/* .zfs znodes always have a generation number of 0 */
726 	for (i = 0; i < sizeof (zfid->zf_gen); i++)
727 		zfid->zf_gen[i] = 0;
728 
729 	zfs_exit(zfsvfs, FTAG);
730 	return (0);
731 }
732 
733 /*
734  * Construct a full dataset name in full_name: "pool/dataset@snap_name"
735  */
736 static int
zfsctl_snapshot_name(zfsvfs_t * zfsvfs,const char * snap_name,int len,char * full_name)737 zfsctl_snapshot_name(zfsvfs_t *zfsvfs, const char *snap_name, int len,
738     char *full_name)
739 {
740 	objset_t *os = zfsvfs->z_os;
741 
742 	if (zfs_component_namecheck(snap_name, NULL, NULL) != 0)
743 		return (SET_ERROR(EILSEQ));
744 
745 	dmu_objset_name(os, full_name);
746 	if ((strlen(full_name) + 1 + strlen(snap_name)) >= len)
747 		return (SET_ERROR(ENAMETOOLONG));
748 
749 	(void) strcat(full_name, "@");
750 	(void) strcat(full_name, snap_name);
751 
752 	return (0);
753 }
754 
755 /*
756  * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/"
757  */
758 static int
zfsctl_snapshot_path_objset(zfsvfs_t * zfsvfs,uint64_t objsetid,int path_len,char * full_path)759 zfsctl_snapshot_path_objset(zfsvfs_t *zfsvfs, uint64_t objsetid,
760     int path_len, char *full_path)
761 {
762 	objset_t *os = zfsvfs->z_os;
763 	fstrans_cookie_t cookie;
764 	char *snapname;
765 	boolean_t case_conflict;
766 	uint64_t id, pos = 0;
767 	int error = 0;
768 
769 	cookie = spl_fstrans_mark();
770 	snapname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
771 
772 	while (error == 0) {
773 		dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
774 		error = dmu_snapshot_list_next(zfsvfs->z_os,
775 		    ZFS_MAX_DATASET_NAME_LEN, snapname, &id, &pos,
776 		    &case_conflict);
777 		dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
778 		if (error)
779 			goto out;
780 
781 		if (id == objsetid)
782 			break;
783 	}
784 
785 	mutex_enter(&zfsvfs->z_vfs->vfs_mntpt_lock);
786 	if (zfsvfs->z_vfs->vfs_mntpoint != NULL) {
787 		snprintf(full_path, path_len, "%s/.zfs/snapshot/%s",
788 		    zfsvfs->z_vfs->vfs_mntpoint, snapname);
789 	} else
790 		error = SET_ERROR(ENOENT);
791 	mutex_exit(&zfsvfs->z_vfs->vfs_mntpt_lock);
792 
793 out:
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  */
803 int
zfsctl_root_lookup(struct inode * dip,const char * name,struct inode ** ipp,int flags,cred_t * cr,int * direntflags,pathname_t * realpnp)804 zfsctl_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  */
837 int
zfsctl_snapdir_lookup(struct inode * dip,const char * name,struct inode ** ipp,int flags,cred_t * cr,int * direntflags,pathname_t * realpnp)838 zfsctl_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  */
869 int
zfsctl_snapdir_rename(struct inode * sdip,const char * snm,struct inode * tdip,const char * tnm,cred_t * cr,int flags)870 zfsctl_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);
933 out:
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  */
948 int
zfsctl_snapdir_remove(struct inode * dip,const char * name,cred_t * cr,int flags)949 zfsctl_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);
985 out:
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  */
998 int
zfsctl_snapdir_mkdir(struct inode * dip,const char * dirname,vattr_t * vap,struct inode ** ipp,cred_t * cr,int flags)999 zfsctl_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 	}
1030 out:
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  */
1043 static void
exportfs_flush(void)1044 exportfs_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  * Returns the path in char format for given struct path. Uses
1054  * d_path exported by kernel to convert struct path to char
1055  * format. Returns the correct path for mountpoints and chroot
1056  * environments.
1057  *
1058  * If chroot environment has directories that are mounted with
1059  * --bind or --rbind flag, d_path returns the complete path inside
1060  * chroot environment but does not return the absolute path, i.e.
1061  * the path to chroot environment is missing.
1062  */
1063 static int
get_root_path(struct path * path,char * buff,int len)1064 get_root_path(struct path *path, char *buff, int len)
1065 {
1066 	char *path_buffer, *path_ptr;
1067 	int error = 0;
1068 
1069 	path_get(path);
1070 	path_buffer = kmem_zalloc(len, KM_SLEEP);
1071 	path_ptr = d_path(path, path_buffer, len);
1072 	if (IS_ERR(path_ptr))
1073 		error = SET_ERROR(-PTR_ERR(path_ptr));
1074 	else
1075 		strcpy(buff, path_ptr);
1076 
1077 	kmem_free(path_buffer, len);
1078 	path_put(path);
1079 	return (error);
1080 }
1081 
1082 /*
1083  * Returns if the current process root is chrooted or not. Linux
1084  * kernel exposes the task_struct for current process and init.
1085  * Since init process root points to actual root filesystem when
1086  * Linux runtime is reached, we can compare the current process
1087  * root with init process root to determine if root of the current
1088  * process is different from init, which can reliably determine if
1089  * current process is in chroot context or not.
1090  */
1091 static int
is_current_chrooted(void)1092 is_current_chrooted(void)
1093 {
1094 	struct task_struct *curr = current, *global = &init_task;
1095 	struct path cr_root, gl_root;
1096 
1097 	task_lock(curr);
1098 	get_fs_root(curr->fs, &cr_root);
1099 	task_unlock(curr);
1100 
1101 	task_lock(global);
1102 	get_fs_root(global->fs, &gl_root);
1103 	task_unlock(global);
1104 
1105 	int chrooted = !path_equal(&cr_root, &gl_root);
1106 	path_put(&gl_root);
1107 	path_put(&cr_root);
1108 
1109 	return (chrooted);
1110 }
1111 
1112 /*
1113  * Attempt to unmount a snapshot by making a call to user space.
1114  * There is no assurance that this can or will succeed, is just a
1115  * best effort.  In the case where it does fail, perhaps because
1116  * it's in use, the unmount will fail harmlessly.
1117  */
1118 int
zfsctl_snapshot_unmount(const char * snapname,int flags)1119 zfsctl_snapshot_unmount(const char *snapname, int flags)
1120 {
1121 	char *argv[] = { "/usr/bin/env", "umount", "-t", "zfs", "-n", NULL,
1122 	    NULL };
1123 	char *envp[] = { NULL };
1124 	zfs_snapentry_t *se;
1125 	int error;
1126 
1127 	rw_enter(&zfs_snapshot_lock, RW_READER);
1128 	if ((se = zfsctl_snapshot_find_by_name(snapname)) == NULL) {
1129 		rw_exit(&zfs_snapshot_lock);
1130 		return (SET_ERROR(ENOENT));
1131 	}
1132 	rw_exit(&zfs_snapshot_lock);
1133 
1134 	exportfs_flush();
1135 
1136 	if (flags & MNT_FORCE)
1137 		argv[4] = "-fn";
1138 	argv[5] = se->se_path;
1139 	dprintf("unmount; path=%s\n", se->se_path);
1140 	error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1141 	zfsctl_snapshot_rele(se);
1142 
1143 
1144 	/*
1145 	 * The umount system utility will return 256 on error.  We must
1146 	 * assume this error is because the file system is busy so it is
1147 	 * converted to the more sensible EBUSY.
1148 	 */
1149 	if (error)
1150 		error = SET_ERROR(EBUSY);
1151 
1152 	return (error);
1153 }
1154 
1155 int
zfsctl_snapshot_mount(struct path * path,int flags)1156 zfsctl_snapshot_mount(struct path *path, int flags)
1157 {
1158 	struct dentry *dentry = path->dentry;
1159 	struct inode *ip = dentry->d_inode;
1160 	zfsvfs_t *zfsvfs;
1161 	zfsvfs_t *snap_zfsvfs;
1162 	zfs_snapentry_t *se;
1163 	char *full_name, *full_path;
1164 	char *argv[] = { "/usr/bin/env", "mount", "-t", "zfs", "-n", NULL, NULL,
1165 	    NULL };
1166 	char *envp[] = { NULL };
1167 	int error;
1168 	struct path spath;
1169 
1170 	if (ip == NULL)
1171 		return (SET_ERROR(EISDIR));
1172 
1173 	zfsvfs = ITOZSB(ip);
1174 	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
1175 		return (error);
1176 
1177 	full_name = kmem_zalloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
1178 	full_path = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1179 
1180 	error = zfsctl_snapshot_name(zfsvfs, dname(dentry),
1181 	    ZFS_MAX_DATASET_NAME_LEN, full_name);
1182 	if (error)
1183 		goto error;
1184 
1185 	if (is_current_chrooted() == 0) {
1186 		/*
1187 		 * Current process is not in chroot context
1188 		 */
1189 
1190 		char *m = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1191 		struct path mnt_path;
1192 		mnt_path.mnt = path->mnt;
1193 		mnt_path.dentry = path->mnt->mnt_root;
1194 
1195 		/*
1196 		 * Get path to current mountpoint
1197 		 */
1198 		error = get_root_path(&mnt_path, m, MAXPATHLEN);
1199 		if (error != 0) {
1200 			kmem_free(m, MAXPATHLEN);
1201 			goto error;
1202 		}
1203 		mutex_enter(&zfsvfs->z_vfs->vfs_mntpt_lock);
1204 		if (zfsvfs->z_vfs->vfs_mntpoint != NULL) {
1205 			/*
1206 			 * If current mnountpoint and vfs_mntpoint are not same,
1207 			 * store current mountpoint in vfs_mntpoint.
1208 			 */
1209 			if (strcmp(zfsvfs->z_vfs->vfs_mntpoint, m) != 0) {
1210 				kmem_strfree(zfsvfs->z_vfs->vfs_mntpoint);
1211 				zfsvfs->z_vfs->vfs_mntpoint = kmem_strdup(m);
1212 			}
1213 		} else
1214 			zfsvfs->z_vfs->vfs_mntpoint = kmem_strdup(m);
1215 		mutex_exit(&zfsvfs->z_vfs->vfs_mntpt_lock);
1216 		kmem_free(m, MAXPATHLEN);
1217 	}
1218 
1219 	/*
1220 	 * Construct a mount point path from sb of the ctldir inode and dirent
1221 	 * name, instead of from d_path(), so that chroot'd process doesn't fail
1222 	 * on mount.zfs(8).
1223 	 */
1224 	mutex_enter(&zfsvfs->z_vfs->vfs_mntpt_lock);
1225 	snprintf(full_path, MAXPATHLEN, "%s/.zfs/snapshot/%s",
1226 	    zfsvfs->z_vfs->vfs_mntpoint ? zfsvfs->z_vfs->vfs_mntpoint : "",
1227 	    dname(dentry));
1228 	mutex_exit(&zfsvfs->z_vfs->vfs_mntpt_lock);
1229 
1230 	/*
1231 	 * Multiple concurrent automounts of a snapshot are never allowed.
1232 	 * The snapshot may be manually mounted as many times as desired.
1233 	 */
1234 	if (zfsctl_snapshot_ismounted(full_name)) {
1235 		error = 0;
1236 		goto error;
1237 	}
1238 
1239 	/*
1240 	 * Attempt to mount the snapshot from user space.  Normally this
1241 	 * would be done using the vfs_kern_mount() function, however that
1242 	 * function is marked GPL-only and cannot be used.  On error we
1243 	 * careful to log the real error to the console and return EISDIR
1244 	 * to safely abort the automount.  This should be very rare.
1245 	 *
1246 	 * If the user mode helper happens to return EBUSY, a concurrent
1247 	 * mount is already in progress in which case the error is ignored.
1248 	 * Take note that if the program was executed successfully the return
1249 	 * value from call_usermodehelper() will be (exitcode << 8 + signal).
1250 	 */
1251 	dprintf("mount; name=%s path=%s\n", full_name, full_path);
1252 	argv[5] = full_name;
1253 	argv[6] = full_path;
1254 	error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1255 	if (error) {
1256 		if (!(error & MOUNT_BUSY << 8)) {
1257 			zfs_dbgmsg("Unable to automount %s error=%d",
1258 			    full_path, error);
1259 			error = SET_ERROR(EISDIR);
1260 		} else {
1261 			/*
1262 			 * EBUSY, this could mean a concurrent mount, or the
1263 			 * snapshot has already been mounted at completely
1264 			 * different place. We return 0 so VFS will retry. For
1265 			 * the latter case the VFS will retry several times
1266 			 * and return ELOOP, which is probably not a very good
1267 			 * behavior.
1268 			 */
1269 			error = 0;
1270 		}
1271 		goto error;
1272 	}
1273 
1274 	/*
1275 	 * Follow down in to the mounted snapshot and set MNT_SHRINKABLE
1276 	 * to identify this as an automounted filesystem.
1277 	 */
1278 	spath = *path;
1279 	path_get(&spath);
1280 	if (follow_down_one(&spath)) {
1281 		snap_zfsvfs = ITOZSB(spath.dentry->d_inode);
1282 		snap_zfsvfs->z_parent = zfsvfs;
1283 		dentry = spath.dentry;
1284 		spath.mnt->mnt_flags |= MNT_SHRINKABLE;
1285 
1286 		rw_enter(&zfs_snapshot_lock, RW_WRITER);
1287 		se = zfsctl_snapshot_alloc(full_name, full_path,
1288 		    snap_zfsvfs->z_os->os_spa, dmu_objset_id(snap_zfsvfs->z_os),
1289 		    dentry);
1290 		zfsctl_snapshot_add(se);
1291 		zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot);
1292 		rw_exit(&zfs_snapshot_lock);
1293 	}
1294 	path_put(&spath);
1295 error:
1296 	kmem_free(full_name, ZFS_MAX_DATASET_NAME_LEN);
1297 	kmem_free(full_path, MAXPATHLEN);
1298 
1299 	zfs_exit(zfsvfs, FTAG);
1300 
1301 	return (error);
1302 }
1303 
1304 /*
1305  * Get the snapdir inode from fid
1306  */
1307 int
zfsctl_snapdir_vget(struct super_block * sb,uint64_t objsetid,int gen,struct inode ** ipp)1308 zfsctl_snapdir_vget(struct super_block *sb, uint64_t objsetid, int gen,
1309     struct inode **ipp)
1310 {
1311 	int error;
1312 	struct path path;
1313 	char *mnt;
1314 	struct dentry *dentry;
1315 
1316 	mnt = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1317 
1318 	error = zfsctl_snapshot_path_objset(sb->s_fs_info, objsetid,
1319 	    MAXPATHLEN, mnt);
1320 	if (error)
1321 		goto out;
1322 
1323 	/* Trigger automount */
1324 	error = -kern_path(mnt, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &path);
1325 	if (error)
1326 		goto out;
1327 
1328 	path_put(&path);
1329 	/*
1330 	 * Get the snapdir inode. Note, we don't want to use the above
1331 	 * path because it contains the root of the snapshot rather
1332 	 * than the snapdir.
1333 	 */
1334 	*ipp = ilookup(sb, ZFSCTL_INO_SNAPDIRS - objsetid);
1335 	if (*ipp == NULL) {
1336 		error = SET_ERROR(ENOENT);
1337 		goto out;
1338 	}
1339 
1340 	/* check gen, see zfsctl_snapdir_fid */
1341 	dentry = d_obtain_alias(igrab(*ipp));
1342 	if (gen != (!IS_ERR(dentry) && d_mountpoint(dentry))) {
1343 		iput(*ipp);
1344 		*ipp = NULL;
1345 		error = SET_ERROR(ENOENT);
1346 	}
1347 	if (!IS_ERR(dentry))
1348 		dput(dentry);
1349 out:
1350 	kmem_free(mnt, MAXPATHLEN);
1351 	return (error);
1352 }
1353 
1354 int
zfsctl_shares_lookup(struct inode * dip,char * name,struct inode ** ipp,int flags,cred_t * cr,int * direntflags,pathname_t * realpnp)1355 zfsctl_shares_lookup(struct inode *dip, char *name, struct inode **ipp,
1356     int flags, cred_t *cr, int *direntflags, pathname_t *realpnp)
1357 {
1358 	zfsvfs_t *zfsvfs = ITOZSB(dip);
1359 	znode_t *zp;
1360 	znode_t *dzp;
1361 	int error;
1362 
1363 	if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
1364 		return (error);
1365 
1366 	if (zfsvfs->z_shares_dir == 0) {
1367 		zfs_exit(zfsvfs, FTAG);
1368 		return (SET_ERROR(ENOTSUP));
1369 	}
1370 
1371 	if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1372 		error = zfs_lookup(dzp, name, &zp, 0, cr, NULL, NULL);
1373 		zrele(dzp);
1374 	}
1375 
1376 	zfs_exit(zfsvfs, FTAG);
1377 
1378 	return (error);
1379 }
1380 
1381 /*
1382  * Initialize the various pieces we'll need to create and manipulate .zfs
1383  * directories.  Currently this is unused but available.
1384  */
1385 void
zfsctl_init(void)1386 zfsctl_init(void)
1387 {
1388 	avl_create(&zfs_snapshots_by_name, snapentry_compare_by_name,
1389 	    sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t,
1390 	    se_node_name));
1391 	avl_create(&zfs_snapshots_by_objsetid, snapentry_compare_by_objsetid,
1392 	    sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t,
1393 	    se_node_objsetid));
1394 	rw_init(&zfs_snapshot_lock, NULL, RW_DEFAULT, NULL);
1395 }
1396 
1397 /*
1398  * Cleanup the various pieces we needed for .zfs directories.  In particular
1399  * ensure the expiry timer is canceled safely.
1400  */
1401 void
zfsctl_fini(void)1402 zfsctl_fini(void)
1403 {
1404 	avl_destroy(&zfs_snapshots_by_name);
1405 	avl_destroy(&zfs_snapshots_by_objsetid);
1406 	rw_destroy(&zfs_snapshot_lock);
1407 }
1408 
1409 module_param(zfs_admin_snapshot, int, 0644);
1410 MODULE_PARM_DESC(zfs_admin_snapshot, "Enable mkdir/rmdir/mv in .zfs/snapshot");
1411 
1412 module_param(zfs_expire_snapshot, int, 0644);
1413 MODULE_PARM_DESC(zfs_expire_snapshot, "Seconds to expire .zfs/snapshot");
1414