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.
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.
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]
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).
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.
38 * ZFS control directory (a.k.a. ".zfs")
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.
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
49 * ctldir ------> snapshotdir -------> snapshot
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.
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.
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).
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.
74 #include <sys/types.h>
75 #include <sys/param.h>
77 #include <sys/sysmacros.h>
78 #include <sys/pathname.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>
86 #include <sys/dmu_objset.h>
87 #include <sys/dsl_destroy.h>
88 #include <sys/dsl_deleg.h>
90 #include <sys/mntent.h>
91 #include "zfs_namecheck.h"
94 * Two AVL trees are maintained which contain all currently automounted
95 * snapshots. Every automounted snapshots maps to a single zfs_snapentry_t
98 * - be attached to both trees, and
99 * - be unique, no duplicate entries are allowed.
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.
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
;
110 * Control Directory Tunables (.zfs)
112 int zfs_expire_snapshot
= ZFSCTL_EXPIRE_SNAPSHOT
;
113 static int zfs_admin_snapshot
= 0;
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 */
128 static void zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t
*se
, int delay
);
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.
134 static zfs_snapentry_t
*
135 zfsctl_snapshot_alloc(const char *full_name
, const char *full_path
, spa_t
*spa
,
136 uint64_t objsetid
, struct dentry
*root_dentry
)
140 se
= kmem_zalloc(sizeof (zfs_snapentry_t
), KM_SLEEP
);
142 se
->se_name
= kmem_strdup(full_name
);
143 se
->se_path
= kmem_strdup(full_path
);
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
);
150 zfs_refcount_create(&se
->se_refcount
);
156 * Free a zfs_snapentry_t the caller must ensure there are no active
160 zfsctl_snapshot_free(zfs_snapentry_t
*se
)
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
);
167 kmem_free(se
, sizeof (zfs_snapentry_t
));
171 * Hold a reference on the zfs_snapentry_t.
174 zfsctl_snapshot_hold(zfs_snapentry_t
*se
)
176 zfs_refcount_add(&se
->se_refcount
, NULL
);
180 * Release a reference on the zfs_snapentry_t. When the number of
181 * references drops to zero the structure will be freed.
184 zfsctl_snapshot_rele(zfs_snapentry_t
*se
)
186 if (zfs_refcount_remove(&se
->se_refcount
, NULL
) == 0)
187 zfsctl_snapshot_free(se
);
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.
196 zfsctl_snapshot_add(zfs_snapentry_t
*se
)
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
);
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.
211 zfsctl_snapshot_remove(zfs_snapentry_t
*se
)
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
);
220 * Snapshot name comparison function for the zfs_snapshots_by_name.
223 snapentry_compare_by_name(const void *a
, const void *b
)
225 const zfs_snapentry_t
*se_a
= a
;
226 const zfs_snapentry_t
*se_b
= b
;
229 ret
= strcmp(se_a
->se_name
, se_b
->se_name
);
240 * Snapshot name comparison function for the zfs_snapshots_by_objsetid.
243 snapentry_compare_by_objsetid(const void *a
, const void *b
)
245 const zfs_snapentry_t
*se_a
= a
;
246 const zfs_snapentry_t
*se_b
= b
;
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);
251 if (se_a
->se_objsetid
< se_b
->se_objsetid
)
253 else if (se_a
->se_objsetid
> se_b
->se_objsetid
)
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.
266 static zfs_snapentry_t
*
267 zfsctl_snapshot_find_by_name(const char *snapname
)
269 zfs_snapentry_t
*se
, search
;
271 ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock
));
273 search
.se_name
= (char *)snapname
;
274 se
= avl_find(&zfs_snapshots_by_name
, &search
, NULL
);
276 zfsctl_snapshot_hold(se
);
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().
286 static zfs_snapentry_t
*
287 zfsctl_snapshot_find_by_objsetid(spa_t
*spa
, uint64_t objsetid
)
289 zfs_snapentry_t
*se
, search
;
291 ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock
));
294 search
.se_objsetid
= objsetid
;
295 se
= avl_find(&zfs_snapshots_by_objsetid
, &search
, NULL
);
297 zfsctl_snapshot_hold(se
);
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.
307 zfsctl_snapshot_rename(const char *old_snapname
, const char *new_snapname
)
311 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock
));
313 se
= zfsctl_snapshot_find_by_name(old_snapname
);
315 return (SET_ERROR(ENOENT
));
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
);
327 * Delayed task responsible for unmounting an expired automounted snapshot.
330 snapentry_expire(void *data
)
332 zfs_snapentry_t
*se
= (zfs_snapentry_t
*)data
;
333 spa_t
*spa
= se
->se_spa
;
334 uint64_t objsetid
= se
->se_objsetid
;
336 if (zfs_expire_snapshot
<= 0) {
337 zfsctl_snapshot_rele(se
);
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
);
348 * Reschedule the unmount if the zfs_snapentry_t wasn't removed.
349 * This can occur when the snapshot is busy.
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
);
356 rw_exit(&zfs_snapshot_lock
);
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
365 zfsctl_snapshot_unmount_cancel(zfs_snapentry_t
*se
)
368 rw_enter(&se
->se_taskqid_lock
, RW_WRITER
);
369 err
= taskq_cancel_id(system_delay_taskq
, se
->se_taskqid
);
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.
376 se
->se_taskqid
= TASKQID_INVALID
;
377 rw_exit(&se
->se_taskqid_lock
);
379 zfsctl_snapshot_rele(se
);
384 * Dispatch the unmount task for delayed handling with a hold protecting it.
387 zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t
*se
, int delay
)
393 zfsctl_snapshot_hold(se
);
394 rw_enter(&se
->se_taskqid_lock
, RW_WRITER
);
395 ASSERT3S(se
->se_taskqid
, ==, TASKQID_INVALID
);
396 se
->se_taskqid
= taskq_dispatch_delay(system_delay_taskq
,
397 snapentry_expire
, se
, TQ_SLEEP
, ddi_get_lbolt() + delay
* HZ
);
398 rw_exit(&se
->se_taskqid_lock
);
402 * Schedule an automatic unmount of objset id to occur in delay seconds from
403 * now. Any previous delayed unmount will be cancelled in favor of the
404 * updated deadline. A reference is taken by zfsctl_snapshot_find_by_name()
405 * and held until the outstanding task is handled or cancelled.
408 zfsctl_snapshot_unmount_delay(spa_t
*spa
, uint64_t objsetid
, int delay
)
413 rw_enter(&zfs_snapshot_lock
, RW_READER
);
414 if ((se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
)) != NULL
) {
415 zfsctl_snapshot_unmount_cancel(se
);
416 zfsctl_snapshot_unmount_delay_impl(se
, delay
);
417 zfsctl_snapshot_rele(se
);
420 rw_exit(&zfs_snapshot_lock
);
426 * Check if snapname is currently mounted. Returned non-zero when mounted
427 * and zero when unmounted.
430 zfsctl_snapshot_ismounted(const char *snapname
)
433 boolean_t ismounted
= B_FALSE
;
435 rw_enter(&zfs_snapshot_lock
, RW_READER
);
436 if ((se
= zfsctl_snapshot_find_by_name(snapname
)) != NULL
) {
437 zfsctl_snapshot_rele(se
);
440 rw_exit(&zfs_snapshot_lock
);
446 * Check if the given inode is a part of the virtual .zfs directory.
449 zfsctl_is_node(struct inode
*ip
)
451 return (ITOZ(ip
)->z_is_ctldir
);
455 * Check if the given inode is a .zfs/snapshots/snapname directory.
458 zfsctl_is_snapdir(struct inode
*ip
)
460 return (zfsctl_is_node(ip
) && (ip
->i_ino
<= ZFSCTL_INO_SNAPDIRS
));
464 * Allocate a new inode with the passed id and ops.
466 static struct inode
*
467 zfsctl_inode_alloc(zfsvfs_t
*zfsvfs
, uint64_t id
,
468 const struct file_operations
*fops
, const struct inode_operations
*ops
)
470 inode_timespec_t now
;
474 ip
= new_inode(zfsvfs
->z_sb
);
478 now
= current_time(ip
);
480 ASSERT3P(zp
->z_dirlocks
, ==, NULL
);
481 ASSERT3P(zp
->z_acl_cached
, ==, NULL
);
482 ASSERT3P(zp
->z_xattr_cached
, ==, NULL
);
484 zp
->z_unlinked
= B_FALSE
;
485 zp
->z_atime_dirty
= B_FALSE
;
486 zp
->z_zn_prefetch
= B_FALSE
;
487 zp
->z_is_sa
= B_FALSE
;
488 zp
->z_is_mapped
= B_FALSE
;
489 zp
->z_is_ctldir
= B_TRUE
;
498 zp
->z_sync_writes_cnt
= 0;
499 zp
->z_async_writes_cnt
= 0;
500 ip
->i_generation
= 0;
502 ip
->i_mode
= (S_IFDIR
| S_IRWXUGO
);
503 ip
->i_uid
= SUID_TO_KUID(0);
504 ip
->i_gid
= SGID_TO_KGID(0);
505 ip
->i_blkbits
= SPA_MINBLOCKSHIFT
;
511 #if defined(IOP_XATTR)
512 ip
->i_opflags
&= ~IOP_XATTR
;
515 if (insert_inode_locked(ip
)) {
516 unlock_new_inode(ip
);
521 mutex_enter(&zfsvfs
->z_znodes_lock
);
522 list_insert_tail(&zfsvfs
->z_all_znodes
, zp
);
523 zfsvfs
->z_nr_znodes
++;
525 mutex_exit(&zfsvfs
->z_znodes_lock
);
527 unlock_new_inode(ip
);
533 * Lookup the inode with given id, it will be allocated if needed.
535 static struct inode
*
536 zfsctl_inode_lookup(zfsvfs_t
*zfsvfs
, uint64_t id
,
537 const struct file_operations
*fops
, const struct inode_operations
*ops
)
539 struct inode
*ip
= NULL
;
542 ip
= ilookup(zfsvfs
->z_sb
, (unsigned long)id
);
546 /* May fail due to concurrent zfsctl_inode_alloc() */
547 ip
= zfsctl_inode_alloc(zfsvfs
, id
, fops
, ops
);
554 * Create the '.zfs' directory. This directory is cached as part of the VFS
555 * structure. This results in a hold on the zfsvfs_t. The code in zfs_umount()
556 * therefore checks against a vfs_count of 2 instead of 1. This reference
557 * is removed when the ctldir is destroyed in the unmount. All other entities
558 * under the '.zfs' directory are created dynamically as needed.
560 * Because the dynamically created '.zfs' directory entries assume the use
561 * of 64-bit inode numbers this support must be disabled on 32-bit systems.
564 zfsctl_create(zfsvfs_t
*zfsvfs
)
566 ASSERT(zfsvfs
->z_ctldir
== NULL
);
568 zfsvfs
->z_ctldir
= zfsctl_inode_alloc(zfsvfs
, ZFSCTL_INO_ROOT
,
569 &zpl_fops_root
, &zpl_ops_root
);
570 if (zfsvfs
->z_ctldir
== NULL
)
571 return (SET_ERROR(ENOENT
));
577 * Destroy the '.zfs' directory or remove a snapshot from zfs_snapshots_by_name.
578 * Only called when the filesystem is unmounted.
581 zfsctl_destroy(zfsvfs_t
*zfsvfs
)
583 if (zfsvfs
->z_issnap
) {
585 spa_t
*spa
= zfsvfs
->z_os
->os_spa
;
586 uint64_t objsetid
= dmu_objset_id(zfsvfs
->z_os
);
588 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
589 se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
);
591 zfsctl_snapshot_remove(se
);
592 rw_exit(&zfs_snapshot_lock
);
594 zfsctl_snapshot_unmount_cancel(se
);
595 zfsctl_snapshot_rele(se
);
597 } else if (zfsvfs
->z_ctldir
) {
598 iput(zfsvfs
->z_ctldir
);
599 zfsvfs
->z_ctldir
= NULL
;
604 * Given a root znode, retrieve the associated .zfs directory.
605 * Add a hold to the vnode and return it.
608 zfsctl_root(znode_t
*zp
)
610 ASSERT(zfs_has_ctldir(zp
));
611 /* Must have an existing ref, so igrab() cannot return NULL */
612 VERIFY3P(igrab(ZTOZSB(zp
)->z_ctldir
), !=, NULL
);
613 return (ZTOZSB(zp
)->z_ctldir
);
617 * Generate a long fid to indicate a snapdir. We encode whether snapdir is
618 * already mounted in gen field. We do this because nfsd lookup will not
619 * trigger automount. Next time the nfsd does fh_to_dentry, we will notice
620 * this and do automount and return ESTALE to force nfsd revalidate and follow
624 zfsctl_snapdir_fid(struct inode
*ip
, fid_t
*fidp
)
626 zfid_short_t
*zfid
= (zfid_short_t
*)fidp
;
627 zfid_long_t
*zlfid
= (zfid_long_t
*)fidp
;
632 struct dentry
*dentry
;
634 if (fidp
->fid_len
< LONG_FID_LEN
) {
635 fidp
->fid_len
= LONG_FID_LEN
;
636 return (SET_ERROR(ENOSPC
));
640 objsetid
= ZFSCTL_INO_SNAPDIRS
- ip
->i_ino
;
641 zfid
->zf_len
= LONG_FID_LEN
;
643 dentry
= d_obtain_alias(igrab(ip
));
644 if (!IS_ERR(dentry
)) {
645 gen
= !!d_mountpoint(dentry
);
649 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
650 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
652 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
653 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
655 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
656 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
658 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
659 zlfid
->zf_setgen
[i
] = 0;
665 * Generate an appropriate fid for an entry in the .zfs directory.
668 zfsctl_fid(struct inode
*ip
, fid_t
*fidp
)
670 znode_t
*zp
= ITOZ(ip
);
671 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
672 uint64_t object
= zp
->z_id
;
677 if ((error
= zfs_enter(zfsvfs
, FTAG
)) != 0)
680 if (zfsctl_is_snapdir(ip
)) {
681 zfs_exit(zfsvfs
, FTAG
);
682 return (zfsctl_snapdir_fid(ip
, fidp
));
685 if (fidp
->fid_len
< SHORT_FID_LEN
) {
686 fidp
->fid_len
= SHORT_FID_LEN
;
687 zfs_exit(zfsvfs
, FTAG
);
688 return (SET_ERROR(ENOSPC
));
691 zfid
= (zfid_short_t
*)fidp
;
693 zfid
->zf_len
= SHORT_FID_LEN
;
695 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
696 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
698 /* .zfs znodes always have a generation number of 0 */
699 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
702 zfs_exit(zfsvfs
, FTAG
);
707 * Construct a full dataset name in full_name: "pool/dataset@snap_name"
710 zfsctl_snapshot_name(zfsvfs_t
*zfsvfs
, const char *snap_name
, int len
,
713 objset_t
*os
= zfsvfs
->z_os
;
715 if (zfs_component_namecheck(snap_name
, NULL
, NULL
) != 0)
716 return (SET_ERROR(EILSEQ
));
718 dmu_objset_name(os
, full_name
);
719 if ((strlen(full_name
) + 1 + strlen(snap_name
)) >= len
)
720 return (SET_ERROR(ENAMETOOLONG
));
722 (void) strcat(full_name
, "@");
723 (void) strcat(full_name
, snap_name
);
729 * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/"
732 zfsctl_snapshot_path_objset(zfsvfs_t
*zfsvfs
, uint64_t objsetid
,
733 int path_len
, char *full_path
)
735 objset_t
*os
= zfsvfs
->z_os
;
736 fstrans_cookie_t cookie
;
738 boolean_t case_conflict
;
739 uint64_t id
, pos
= 0;
742 if (zfsvfs
->z_vfs
->vfs_mntpoint
== NULL
)
743 return (SET_ERROR(ENOENT
));
745 cookie
= spl_fstrans_mark();
746 snapname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
749 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
750 error
= dmu_snapshot_list_next(zfsvfs
->z_os
,
751 ZFS_MAX_DATASET_NAME_LEN
, snapname
, &id
, &pos
,
753 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
761 snprintf(full_path
, path_len
, "%s/.zfs/snapshot/%s",
762 zfsvfs
->z_vfs
->vfs_mntpoint
, snapname
);
764 kmem_free(snapname
, ZFS_MAX_DATASET_NAME_LEN
);
765 spl_fstrans_unmark(cookie
);
771 * Special case the handling of "..".
774 zfsctl_root_lookup(struct inode
*dip
, const char *name
, struct inode
**ipp
,
775 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
777 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
780 if ((error
= zfs_enter(zfsvfs
, FTAG
)) != 0)
783 if (strcmp(name
, "..") == 0) {
784 *ipp
= dip
->i_sb
->s_root
->d_inode
;
785 } else if (strcmp(name
, ZFS_SNAPDIR_NAME
) == 0) {
786 *ipp
= zfsctl_inode_lookup(zfsvfs
, ZFSCTL_INO_SNAPDIR
,
787 &zpl_fops_snapdir
, &zpl_ops_snapdir
);
788 } else if (strcmp(name
, ZFS_SHAREDIR_NAME
) == 0) {
789 *ipp
= zfsctl_inode_lookup(zfsvfs
, ZFSCTL_INO_SHARES
,
790 &zpl_fops_shares
, &zpl_ops_shares
);
796 error
= SET_ERROR(ENOENT
);
798 zfs_exit(zfsvfs
, FTAG
);
804 * Lookup entry point for the 'snapshot' directory. Try to open the
805 * snapshot if it exist, creating the pseudo filesystem inode as necessary.
808 zfsctl_snapdir_lookup(struct inode
*dip
, const char *name
, struct inode
**ipp
,
809 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
811 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
815 if ((error
= zfs_enter(zfsvfs
, FTAG
)) != 0)
818 error
= dmu_snapshot_lookup(zfsvfs
->z_os
, name
, &id
);
820 zfs_exit(zfsvfs
, FTAG
);
824 *ipp
= zfsctl_inode_lookup(zfsvfs
, ZFSCTL_INO_SNAPDIRS
- id
,
825 &simple_dir_operations
, &simple_dir_inode_operations
);
827 error
= SET_ERROR(ENOENT
);
829 zfs_exit(zfsvfs
, FTAG
);
835 * Renaming a directory under '.zfs/snapshot' will automatically trigger
836 * a rename of the snapshot to the new given name. The rename is confined
837 * to the '.zfs/snapshot' directory snapshots cannot be moved elsewhere.
840 zfsctl_snapdir_rename(struct inode
*sdip
, const char *snm
,
841 struct inode
*tdip
, const char *tnm
, cred_t
*cr
, int flags
)
843 zfsvfs_t
*zfsvfs
= ITOZSB(sdip
);
844 char *to
, *from
, *real
, *fsname
;
847 if (!zfs_admin_snapshot
)
848 return (SET_ERROR(EACCES
));
850 if ((error
= zfs_enter(zfsvfs
, FTAG
)) != 0)
853 to
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
854 from
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
855 real
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
856 fsname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
858 if (zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
) {
859 error
= dmu_snapshot_realname(zfsvfs
->z_os
, snm
, real
,
860 ZFS_MAX_DATASET_NAME_LEN
, NULL
);
863 } else if (error
!= ENOTSUP
) {
868 dmu_objset_name(zfsvfs
->z_os
, fsname
);
870 error
= zfsctl_snapshot_name(ITOZSB(sdip
), snm
,
871 ZFS_MAX_DATASET_NAME_LEN
, from
);
873 error
= zfsctl_snapshot_name(ITOZSB(tdip
), tnm
,
874 ZFS_MAX_DATASET_NAME_LEN
, to
);
876 error
= zfs_secpolicy_rename_perms(from
, to
, cr
);
881 * Cannot move snapshots out of the snapdir.
884 error
= SET_ERROR(EINVAL
);
889 * No-op when names are identical.
891 if (strcmp(snm
, tnm
) == 0) {
896 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
898 error
= dsl_dataset_rename_snapshot(fsname
, snm
, tnm
, B_FALSE
);
900 (void) zfsctl_snapshot_rename(snm
, tnm
);
902 rw_exit(&zfs_snapshot_lock
);
904 kmem_free(from
, ZFS_MAX_DATASET_NAME_LEN
);
905 kmem_free(to
, ZFS_MAX_DATASET_NAME_LEN
);
906 kmem_free(real
, ZFS_MAX_DATASET_NAME_LEN
);
907 kmem_free(fsname
, ZFS_MAX_DATASET_NAME_LEN
);
909 zfs_exit(zfsvfs
, FTAG
);
915 * Removing a directory under '.zfs/snapshot' will automatically trigger
916 * the removal of the snapshot with the given name.
919 zfsctl_snapdir_remove(struct inode
*dip
, const char *name
, cred_t
*cr
,
922 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
923 char *snapname
, *real
;
926 if (!zfs_admin_snapshot
)
927 return (SET_ERROR(EACCES
));
929 if ((error
= zfs_enter(zfsvfs
, FTAG
)) != 0)
932 snapname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
933 real
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
935 if (zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
) {
936 error
= dmu_snapshot_realname(zfsvfs
->z_os
, name
, real
,
937 ZFS_MAX_DATASET_NAME_LEN
, NULL
);
940 } else if (error
!= ENOTSUP
) {
945 error
= zfsctl_snapshot_name(ITOZSB(dip
), name
,
946 ZFS_MAX_DATASET_NAME_LEN
, snapname
);
948 error
= zfs_secpolicy_destroy_perms(snapname
, cr
);
952 error
= zfsctl_snapshot_unmount(snapname
, MNT_FORCE
);
953 if ((error
== 0) || (error
== ENOENT
))
954 error
= dsl_destroy_snapshot(snapname
, B_FALSE
);
956 kmem_free(snapname
, ZFS_MAX_DATASET_NAME_LEN
);
957 kmem_free(real
, ZFS_MAX_DATASET_NAME_LEN
);
959 zfs_exit(zfsvfs
, FTAG
);
965 * Creating a directory under '.zfs/snapshot' will automatically trigger
966 * the creation of a new snapshot with the given name.
969 zfsctl_snapdir_mkdir(struct inode
*dip
, const char *dirname
, vattr_t
*vap
,
970 struct inode
**ipp
, cred_t
*cr
, int flags
)
972 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
976 if (!zfs_admin_snapshot
)
977 return (SET_ERROR(EACCES
));
979 dsname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
981 if (zfs_component_namecheck(dirname
, NULL
, NULL
) != 0) {
982 error
= SET_ERROR(EILSEQ
);
986 dmu_objset_name(zfsvfs
->z_os
, dsname
);
988 error
= zfs_secpolicy_snapshot_perms(dsname
, cr
);
993 error
= dmu_objset_snapshot_one(dsname
, dirname
);
997 error
= zfsctl_snapdir_lookup(dip
, dirname
, ipp
,
1001 kmem_free(dsname
, ZFS_MAX_DATASET_NAME_LEN
);
1007 * Flush everything out of the kernel's export table and such.
1008 * This is needed as once the snapshot is used over NFS, its
1009 * entries in svc_export and svc_expkey caches hold reference
1010 * to the snapshot mount point. There is no known way of flushing
1011 * only the entries related to the snapshot.
1014 exportfs_flush(void)
1016 char *argv
[] = { "/usr/sbin/exportfs", "-f", NULL
};
1017 char *envp
[] = { NULL
};
1019 (void) call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
1023 * Attempt to unmount a snapshot by making a call to user space.
1024 * There is no assurance that this can or will succeed, is just a
1025 * best effort. In the case where it does fail, perhaps because
1026 * it's in use, the unmount will fail harmlessly.
1029 zfsctl_snapshot_unmount(const char *snapname
, int flags
)
1031 char *argv
[] = { "/usr/bin/env", "umount", "-t", "zfs", "-n", NULL
,
1033 char *envp
[] = { NULL
};
1034 zfs_snapentry_t
*se
;
1037 rw_enter(&zfs_snapshot_lock
, RW_READER
);
1038 if ((se
= zfsctl_snapshot_find_by_name(snapname
)) == NULL
) {
1039 rw_exit(&zfs_snapshot_lock
);
1040 return (SET_ERROR(ENOENT
));
1042 rw_exit(&zfs_snapshot_lock
);
1046 if (flags
& MNT_FORCE
)
1048 argv
[5] = se
->se_path
;
1049 dprintf("unmount; path=%s\n", se
->se_path
);
1050 error
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
1051 zfsctl_snapshot_rele(se
);
1055 * The umount system utility will return 256 on error. We must
1056 * assume this error is because the file system is busy so it is
1057 * converted to the more sensible EBUSY.
1060 error
= SET_ERROR(EBUSY
);
1066 zfsctl_snapshot_mount(struct path
*path
, int flags
)
1068 struct dentry
*dentry
= path
->dentry
;
1069 struct inode
*ip
= dentry
->d_inode
;
1071 zfsvfs_t
*snap_zfsvfs
;
1072 zfs_snapentry_t
*se
;
1073 char *full_name
, *full_path
;
1074 char *argv
[] = { "/usr/bin/env", "mount", "-t", "zfs", "-n", NULL
, NULL
,
1076 char *envp
[] = { NULL
};
1081 return (SET_ERROR(EISDIR
));
1083 zfsvfs
= ITOZSB(ip
);
1084 if ((error
= zfs_enter(zfsvfs
, FTAG
)) != 0)
1087 full_name
= kmem_zalloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
1088 full_path
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
1090 error
= zfsctl_snapshot_name(zfsvfs
, dname(dentry
),
1091 ZFS_MAX_DATASET_NAME_LEN
, full_name
);
1096 * Construct a mount point path from sb of the ctldir inode and dirent
1097 * name, instead of from d_path(), so that chroot'd process doesn't fail
1100 snprintf(full_path
, MAXPATHLEN
, "%s/.zfs/snapshot/%s",
1101 zfsvfs
->z_vfs
->vfs_mntpoint
? zfsvfs
->z_vfs
->vfs_mntpoint
: "",
1105 * Multiple concurrent automounts of a snapshot are never allowed.
1106 * The snapshot may be manually mounted as many times as desired.
1108 if (zfsctl_snapshot_ismounted(full_name
)) {
1114 * Attempt to mount the snapshot from user space. Normally this
1115 * would be done using the vfs_kern_mount() function, however that
1116 * function is marked GPL-only and cannot be used. On error we
1117 * careful to log the real error to the console and return EISDIR
1118 * to safely abort the automount. This should be very rare.
1120 * If the user mode helper happens to return EBUSY, a concurrent
1121 * mount is already in progress in which case the error is ignored.
1122 * Take note that if the program was executed successfully the return
1123 * value from call_usermodehelper() will be (exitcode << 8 + signal).
1125 dprintf("mount; name=%s path=%s\n", full_name
, full_path
);
1126 argv
[5] = full_name
;
1127 argv
[6] = full_path
;
1128 error
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
1130 if (!(error
& MOUNT_BUSY
<< 8)) {
1131 zfs_dbgmsg("Unable to automount %s error=%d",
1133 error
= SET_ERROR(EISDIR
);
1136 * EBUSY, this could mean a concurrent mount, or the
1137 * snapshot has already been mounted at completely
1138 * different place. We return 0 so VFS will retry. For
1139 * the latter case the VFS will retry several times
1140 * and return ELOOP, which is probably not a very good
1149 * Follow down in to the mounted snapshot and set MNT_SHRINKABLE
1150 * to identify this as an automounted filesystem.
1154 if (follow_down_one(&spath
)) {
1155 snap_zfsvfs
= ITOZSB(spath
.dentry
->d_inode
);
1156 snap_zfsvfs
->z_parent
= zfsvfs
;
1157 dentry
= spath
.dentry
;
1158 spath
.mnt
->mnt_flags
|= MNT_SHRINKABLE
;
1160 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
1161 se
= zfsctl_snapshot_alloc(full_name
, full_path
,
1162 snap_zfsvfs
->z_os
->os_spa
, dmu_objset_id(snap_zfsvfs
->z_os
),
1164 zfsctl_snapshot_add(se
);
1165 zfsctl_snapshot_unmount_delay_impl(se
, zfs_expire_snapshot
);
1166 rw_exit(&zfs_snapshot_lock
);
1170 kmem_free(full_name
, ZFS_MAX_DATASET_NAME_LEN
);
1171 kmem_free(full_path
, MAXPATHLEN
);
1173 zfs_exit(zfsvfs
, FTAG
);
1179 * Get the snapdir inode from fid
1182 zfsctl_snapdir_vget(struct super_block
*sb
, uint64_t objsetid
, int gen
,
1188 struct dentry
*dentry
;
1190 mnt
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
1192 error
= zfsctl_snapshot_path_objset(sb
->s_fs_info
, objsetid
,
1197 /* Trigger automount */
1198 error
= -kern_path(mnt
, LOOKUP_FOLLOW
|LOOKUP_DIRECTORY
, &path
);
1204 * Get the snapdir inode. Note, we don't want to use the above
1205 * path because it contains the root of the snapshot rather
1208 *ipp
= ilookup(sb
, ZFSCTL_INO_SNAPDIRS
- objsetid
);
1210 error
= SET_ERROR(ENOENT
);
1214 /* check gen, see zfsctl_snapdir_fid */
1215 dentry
= d_obtain_alias(igrab(*ipp
));
1216 if (gen
!= (!IS_ERR(dentry
) && d_mountpoint(dentry
))) {
1219 error
= SET_ERROR(ENOENT
);
1221 if (!IS_ERR(dentry
))
1224 kmem_free(mnt
, MAXPATHLEN
);
1229 zfsctl_shares_lookup(struct inode
*dip
, char *name
, struct inode
**ipp
,
1230 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1232 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1237 if ((error
= zfs_enter(zfsvfs
, FTAG
)) != 0)
1240 if (zfsvfs
->z_shares_dir
== 0) {
1241 zfs_exit(zfsvfs
, FTAG
);
1242 return (SET_ERROR(ENOTSUP
));
1245 if ((error
= zfs_zget(zfsvfs
, zfsvfs
->z_shares_dir
, &dzp
)) == 0) {
1246 error
= zfs_lookup(dzp
, name
, &zp
, 0, cr
, NULL
, NULL
);
1250 zfs_exit(zfsvfs
, FTAG
);
1256 * Initialize the various pieces we'll need to create and manipulate .zfs
1257 * directories. Currently this is unused but available.
1262 avl_create(&zfs_snapshots_by_name
, snapentry_compare_by_name
,
1263 sizeof (zfs_snapentry_t
), offsetof(zfs_snapentry_t
,
1265 avl_create(&zfs_snapshots_by_objsetid
, snapentry_compare_by_objsetid
,
1266 sizeof (zfs_snapentry_t
), offsetof(zfs_snapentry_t
,
1268 rw_init(&zfs_snapshot_lock
, NULL
, RW_DEFAULT
, NULL
);
1272 * Cleanup the various pieces we needed for .zfs directories. In particular
1273 * ensure the expiry timer is canceled safely.
1278 avl_destroy(&zfs_snapshots_by_name
);
1279 avl_destroy(&zfs_snapshots_by_objsetid
);
1280 rw_destroy(&zfs_snapshot_lock
);
1283 module_param(zfs_admin_snapshot
, int, 0644);
1284 MODULE_PARM_DESC(zfs_admin_snapshot
, "Enable mkdir/rmdir/mv in .zfs/snapshot");
1286 module_param(zfs_expire_snapshot
, int, 0644);
1287 MODULE_PARM_DESC(zfs_expire_snapshot
, "Seconds to expire .zfs/snapshot");