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 http://www.opensolaris.org/os/licensing.
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.
35 * ZFS control directory (a.k.a. ".zfs")
37 * This directory provides a common location for all ZFS meta-objects.
38 * Currently, this is only the 'snapshot' and 'shares' directory, but this may
39 * expand in the future. The elements are built dynamically, as the hierarchy
40 * does not actually exist on disk.
42 * For 'snapshot', we don't want to have all snapshots always mounted, because
43 * this would take up a huge amount of space in /etc/mnttab. We have three
46 * ctldir ------> snapshotdir -------> snapshot
52 * The 'snapshot' node contains just enough information to lookup '..' and act
53 * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we
54 * perform an automount of the underlying filesystem and return the
55 * corresponding inode.
57 * All mounts are handled automatically by an user mode helper which invokes
58 * the mount mount procedure. Unmounts are handled by allowing the mount
59 * point to expire so the kernel may automatically unmount it.
61 * The '.zfs', '.zfs/snapshot', and all directories created under
62 * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') all share the same
63 * share the same zfsvfs_t as the head filesystem (what '.zfs' lives under).
65 * File systems mounted on top of the '.zfs/snapshot/<snapname>' paths
66 * (ie: snapshots) are complete ZFS filesystems and have their own unique
67 * zfsvfs_t. However, the fsid reported by these mounts will be the same
68 * as that used by the parent zfsvfs_t to make NFS happy.
71 #include <sys/types.h>
72 #include <sys/param.h>
74 #include <sys/sysmacros.h>
75 #include <sys/pathname.h>
77 #include <sys/zfs_ctldir.h>
78 #include <sys/zfs_ioctl.h>
79 #include <sys/zfs_vfsops.h>
80 #include <sys/zfs_vnops.h>
83 #include <sys/dmu_objset.h>
84 #include <sys/dsl_destroy.h>
85 #include <sys/dsl_deleg.h>
87 #include "zfs_namecheck.h"
90 * Two AVL trees are maintained which contain all currently automounted
91 * snapshots. Every automounted snapshots maps to a single zfs_snapentry_t
94 * - be attached to both trees, and
95 * - be unique, no duplicate entries are allowed.
97 * The zfs_snapshots_by_name tree is indexed by the full dataset name
98 * while the zfs_snapshots_by_objsetid tree is indexed by the unique
99 * objsetid. This allows for fast lookups either by name or objsetid.
101 static avl_tree_t zfs_snapshots_by_name
;
102 static avl_tree_t zfs_snapshots_by_objsetid
;
103 static krwlock_t zfs_snapshot_lock
;
106 * Control Directory Tunables (.zfs)
108 int zfs_expire_snapshot
= ZFSCTL_EXPIRE_SNAPSHOT
;
109 int zfs_admin_snapshot
= 1;
112 char *se_name
; /* full snapshot name */
113 char *se_path
; /* full mount path */
114 spa_t
*se_spa
; /* pool spa */
115 uint64_t se_objsetid
; /* snapshot objset id */
116 struct dentry
*se_root_dentry
; /* snapshot root dentry */
117 taskqid_t se_taskqid
; /* scheduled unmount taskqid */
118 avl_node_t se_node_name
; /* zfs_snapshots_by_name link */
119 avl_node_t se_node_objsetid
; /* zfs_snapshots_by_objsetid link */
120 zfs_refcount_t se_refcount
; /* reference count */
123 static void zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t
*se
, int delay
);
126 * Allocate a new zfs_snapentry_t being careful to make a copy of the
127 * the snapshot name and provided mount point. No reference is taken.
129 static zfs_snapentry_t
*
130 zfsctl_snapshot_alloc(char *full_name
, char *full_path
, spa_t
*spa
,
131 uint64_t objsetid
, struct dentry
*root_dentry
)
135 se
= kmem_zalloc(sizeof (zfs_snapentry_t
), KM_SLEEP
);
137 se
->se_name
= strdup(full_name
);
138 se
->se_path
= strdup(full_path
);
140 se
->se_objsetid
= objsetid
;
141 se
->se_root_dentry
= root_dentry
;
142 se
->se_taskqid
= TASKQID_INVALID
;
144 refcount_create(&se
->se_refcount
);
150 * Free a zfs_snapentry_t the called must ensure there are no active
154 zfsctl_snapshot_free(zfs_snapentry_t
*se
)
156 refcount_destroy(&se
->se_refcount
);
157 strfree(se
->se_name
);
158 strfree(se
->se_path
);
160 kmem_free(se
, sizeof (zfs_snapentry_t
));
164 * Hold a reference on the zfs_snapentry_t.
167 zfsctl_snapshot_hold(zfs_snapentry_t
*se
)
169 zfs_refcount_add(&se
->se_refcount
, NULL
);
173 * Release a reference on the zfs_snapentry_t. When the number of
174 * references drops to zero the structure will be freed.
177 zfsctl_snapshot_rele(zfs_snapentry_t
*se
)
179 if (refcount_remove(&se
->se_refcount
, NULL
) == 0)
180 zfsctl_snapshot_free(se
);
184 * Add a zfs_snapentry_t to both the zfs_snapshots_by_name and
185 * zfs_snapshots_by_objsetid trees. While the zfs_snapentry_t is part
186 * of the trees a reference is held.
189 zfsctl_snapshot_add(zfs_snapentry_t
*se
)
191 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock
));
192 zfs_refcount_add(&se
->se_refcount
, NULL
);
193 avl_add(&zfs_snapshots_by_name
, se
);
194 avl_add(&zfs_snapshots_by_objsetid
, se
);
198 * Remove a zfs_snapentry_t from both the zfs_snapshots_by_name and
199 * zfs_snapshots_by_objsetid trees. Upon removal a reference is dropped,
200 * this can result in the structure being freed if that was the last
201 * remaining reference.
204 zfsctl_snapshot_remove(zfs_snapentry_t
*se
)
206 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock
));
207 avl_remove(&zfs_snapshots_by_name
, se
);
208 avl_remove(&zfs_snapshots_by_objsetid
, se
);
209 zfsctl_snapshot_rele(se
);
213 * Snapshot name comparison function for the zfs_snapshots_by_name.
216 snapentry_compare_by_name(const void *a
, const void *b
)
218 const zfs_snapentry_t
*se_a
= a
;
219 const zfs_snapentry_t
*se_b
= b
;
222 ret
= strcmp(se_a
->se_name
, se_b
->se_name
);
233 * Snapshot name comparison function for the zfs_snapshots_by_objsetid.
236 snapentry_compare_by_objsetid(const void *a
, const void *b
)
238 const zfs_snapentry_t
*se_a
= a
;
239 const zfs_snapentry_t
*se_b
= b
;
241 if (se_a
->se_spa
!= se_b
->se_spa
)
242 return ((ulong_t
)se_a
->se_spa
< (ulong_t
)se_b
->se_spa
? -1 : 1);
244 if (se_a
->se_objsetid
< se_b
->se_objsetid
)
246 else if (se_a
->se_objsetid
> se_b
->se_objsetid
)
253 * Find a zfs_snapentry_t in zfs_snapshots_by_name. If the snapname
254 * is found a pointer to the zfs_snapentry_t is returned and a reference
255 * taken on the structure. The caller is responsible for dropping the
256 * reference with zfsctl_snapshot_rele(). If the snapname is not found
257 * NULL will be returned.
259 static zfs_snapentry_t
*
260 zfsctl_snapshot_find_by_name(char *snapname
)
262 zfs_snapentry_t
*se
, search
;
264 ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock
));
266 search
.se_name
= snapname
;
267 se
= avl_find(&zfs_snapshots_by_name
, &search
, NULL
);
269 zfs_refcount_add(&se
->se_refcount
, NULL
);
275 * Find a zfs_snapentry_t in zfs_snapshots_by_objsetid given the objset id
276 * rather than the snapname. In all other respects it behaves the same
277 * as zfsctl_snapshot_find_by_name().
279 static zfs_snapentry_t
*
280 zfsctl_snapshot_find_by_objsetid(spa_t
*spa
, uint64_t objsetid
)
282 zfs_snapentry_t
*se
, search
;
284 ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock
));
287 search
.se_objsetid
= objsetid
;
288 se
= avl_find(&zfs_snapshots_by_objsetid
, &search
, NULL
);
290 zfs_refcount_add(&se
->se_refcount
, NULL
);
296 * Rename a zfs_snapentry_t in the zfs_snapshots_by_name. The structure is
297 * removed, renamed, and added back to the new correct location in the tree.
300 zfsctl_snapshot_rename(char *old_snapname
, char *new_snapname
)
304 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock
));
306 se
= zfsctl_snapshot_find_by_name(old_snapname
);
308 return (SET_ERROR(ENOENT
));
310 zfsctl_snapshot_remove(se
);
311 strfree(se
->se_name
);
312 se
->se_name
= strdup(new_snapname
);
313 zfsctl_snapshot_add(se
);
314 zfsctl_snapshot_rele(se
);
320 * Delayed task responsible for unmounting an expired automounted snapshot.
323 snapentry_expire(void *data
)
325 zfs_snapentry_t
*se
= (zfs_snapentry_t
*)data
;
326 spa_t
*spa
= se
->se_spa
;
327 uint64_t objsetid
= se
->se_objsetid
;
329 if (zfs_expire_snapshot
<= 0) {
330 zfsctl_snapshot_rele(se
);
334 se
->se_taskqid
= TASKQID_INVALID
;
335 (void) zfsctl_snapshot_unmount(se
->se_name
, MNT_EXPIRE
);
336 zfsctl_snapshot_rele(se
);
339 * Reschedule the unmount if the zfs_snapentry_t wasn't removed.
340 * This can occur when the snapshot is busy.
342 rw_enter(&zfs_snapshot_lock
, RW_READER
);
343 if ((se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
)) != NULL
) {
344 zfsctl_snapshot_unmount_delay_impl(se
, zfs_expire_snapshot
);
345 zfsctl_snapshot_rele(se
);
347 rw_exit(&zfs_snapshot_lock
);
351 * Cancel an automatic unmount of a snapname. This callback is responsible
352 * for dropping the reference on the zfs_snapentry_t which was taken when
356 zfsctl_snapshot_unmount_cancel(zfs_snapentry_t
*se
)
358 if (taskq_cancel_id(system_delay_taskq
, se
->se_taskqid
) == 0) {
359 se
->se_taskqid
= TASKQID_INVALID
;
360 zfsctl_snapshot_rele(se
);
365 * Dispatch the unmount task for delayed handling with a hold protecting it.
368 zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t
*se
, int delay
)
370 ASSERT3S(se
->se_taskqid
, ==, TASKQID_INVALID
);
375 zfsctl_snapshot_hold(se
);
376 se
->se_taskqid
= taskq_dispatch_delay(system_delay_taskq
,
377 snapentry_expire
, se
, TQ_SLEEP
, ddi_get_lbolt() + delay
* HZ
);
381 * Schedule an automatic unmount of objset id to occur in delay seconds from
382 * now. Any previous delayed unmount will be cancelled in favor of the
383 * updated deadline. A reference is taken by zfsctl_snapshot_find_by_name()
384 * and held until the outstanding task is handled or cancelled.
387 zfsctl_snapshot_unmount_delay(spa_t
*spa
, uint64_t objsetid
, int delay
)
392 rw_enter(&zfs_snapshot_lock
, RW_READER
);
393 if ((se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
)) != NULL
) {
394 zfsctl_snapshot_unmount_cancel(se
);
395 zfsctl_snapshot_unmount_delay_impl(se
, delay
);
396 zfsctl_snapshot_rele(se
);
399 rw_exit(&zfs_snapshot_lock
);
405 * Check if snapname is currently mounted. Returned non-zero when mounted
406 * and zero when unmounted.
409 zfsctl_snapshot_ismounted(char *snapname
)
412 boolean_t ismounted
= B_FALSE
;
414 rw_enter(&zfs_snapshot_lock
, RW_READER
);
415 if ((se
= zfsctl_snapshot_find_by_name(snapname
)) != NULL
) {
416 zfsctl_snapshot_rele(se
);
419 rw_exit(&zfs_snapshot_lock
);
425 * Check if the given inode is a part of the virtual .zfs directory.
428 zfsctl_is_node(struct inode
*ip
)
430 return (ITOZ(ip
)->z_is_ctldir
);
434 * Check if the given inode is a .zfs/snapshots/snapname directory.
437 zfsctl_is_snapdir(struct inode
*ip
)
439 return (zfsctl_is_node(ip
) && (ip
->i_ino
<= ZFSCTL_INO_SNAPDIRS
));
443 * Allocate a new inode with the passed id and ops.
445 static struct inode
*
446 zfsctl_inode_alloc(zfsvfs_t
*zfsvfs
, uint64_t id
,
447 const struct file_operations
*fops
, const struct inode_operations
*ops
)
449 inode_timespec_t now
;
453 ip
= new_inode(zfsvfs
->z_sb
);
457 now
= current_time(ip
);
459 ASSERT3P(zp
->z_dirlocks
, ==, NULL
);
460 ASSERT3P(zp
->z_acl_cached
, ==, NULL
);
461 ASSERT3P(zp
->z_xattr_cached
, ==, NULL
);
464 zp
->z_atime_dirty
= 0;
465 zp
->z_zn_prefetch
= 0;
475 zp
->z_is_mapped
= B_FALSE
;
476 zp
->z_is_ctldir
= B_TRUE
;
477 zp
->z_is_sa
= B_FALSE
;
478 zp
->z_is_stale
= B_FALSE
;
479 ip
->i_generation
= 0;
481 ip
->i_mode
= (S_IFDIR
| S_IRWXUGO
);
482 ip
->i_uid
= SUID_TO_KUID(0);
483 ip
->i_gid
= SGID_TO_KGID(0);
484 ip
->i_blkbits
= SPA_MINBLOCKSHIFT
;
490 #if defined(IOP_XATTR)
491 ip
->i_opflags
&= ~IOP_XATTR
;
494 if (insert_inode_locked(ip
)) {
495 unlock_new_inode(ip
);
500 mutex_enter(&zfsvfs
->z_znodes_lock
);
501 list_insert_tail(&zfsvfs
->z_all_znodes
, zp
);
502 zfsvfs
->z_nr_znodes
++;
504 mutex_exit(&zfsvfs
->z_znodes_lock
);
506 unlock_new_inode(ip
);
512 * Lookup the inode with given id, it will be allocated if needed.
514 static struct inode
*
515 zfsctl_inode_lookup(zfsvfs_t
*zfsvfs
, uint64_t id
,
516 const struct file_operations
*fops
, const struct inode_operations
*ops
)
518 struct inode
*ip
= NULL
;
521 ip
= ilookup(zfsvfs
->z_sb
, (unsigned long)id
);
525 /* May fail due to concurrent zfsctl_inode_alloc() */
526 ip
= zfsctl_inode_alloc(zfsvfs
, id
, fops
, ops
);
533 * Create the '.zfs' directory. This directory is cached as part of the VFS
534 * structure. This results in a hold on the zfsvfs_t. The code in zfs_umount()
535 * therefore checks against a vfs_count of 2 instead of 1. This reference
536 * is removed when the ctldir is destroyed in the unmount. All other entities
537 * under the '.zfs' directory are created dynamically as needed.
539 * Because the dynamically created '.zfs' directory entries assume the use
540 * of 64-bit inode numbers this support must be disabled on 32-bit systems.
543 zfsctl_create(zfsvfs_t
*zfsvfs
)
545 ASSERT(zfsvfs
->z_ctldir
== NULL
);
547 zfsvfs
->z_ctldir
= zfsctl_inode_alloc(zfsvfs
, ZFSCTL_INO_ROOT
,
548 &zpl_fops_root
, &zpl_ops_root
);
549 if (zfsvfs
->z_ctldir
== NULL
)
550 return (SET_ERROR(ENOENT
));
556 * Destroy the '.zfs' directory or remove a snapshot from zfs_snapshots_by_name.
557 * Only called when the filesystem is unmounted.
560 zfsctl_destroy(zfsvfs_t
*zfsvfs
)
562 if (zfsvfs
->z_issnap
) {
564 spa_t
*spa
= zfsvfs
->z_os
->os_spa
;
565 uint64_t objsetid
= dmu_objset_id(zfsvfs
->z_os
);
567 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
568 se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
);
570 zfsctl_snapshot_remove(se
);
571 rw_exit(&zfs_snapshot_lock
);
573 zfsctl_snapshot_unmount_cancel(se
);
574 zfsctl_snapshot_rele(se
);
576 } else if (zfsvfs
->z_ctldir
) {
577 iput(zfsvfs
->z_ctldir
);
578 zfsvfs
->z_ctldir
= NULL
;
583 * Given a root znode, retrieve the associated .zfs directory.
584 * Add a hold to the vnode and return it.
587 zfsctl_root(znode_t
*zp
)
589 ASSERT(zfs_has_ctldir(zp
));
590 igrab(ZTOZSB(zp
)->z_ctldir
);
591 return (ZTOZSB(zp
)->z_ctldir
);
595 * Generate a long fid to indicate a snapdir. We encode whether snapdir is
596 * already monunted in gen field. We do this because nfsd lookup will not
597 * trigger automount. Next time the nfsd does fh_to_dentry, we will notice
598 * this and do automount and return ESTALE to force nfsd revalidate and follow
602 zfsctl_snapdir_fid(struct inode
*ip
, fid_t
*fidp
)
604 zfid_short_t
*zfid
= (zfid_short_t
*)fidp
;
605 zfid_long_t
*zlfid
= (zfid_long_t
*)fidp
;
610 struct dentry
*dentry
;
612 if (fidp
->fid_len
< LONG_FID_LEN
) {
613 fidp
->fid_len
= LONG_FID_LEN
;
614 return (SET_ERROR(ENOSPC
));
618 objsetid
= ZFSCTL_INO_SNAPDIRS
- ip
->i_ino
;
619 zfid
->zf_len
= LONG_FID_LEN
;
621 dentry
= d_obtain_alias(igrab(ip
));
622 if (!IS_ERR(dentry
)) {
623 gen
= !!d_mountpoint(dentry
);
627 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
628 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
630 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
631 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
633 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
634 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
636 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
637 zlfid
->zf_setgen
[i
] = 0;
643 * Generate an appropriate fid for an entry in the .zfs directory.
646 zfsctl_fid(struct inode
*ip
, fid_t
*fidp
)
648 znode_t
*zp
= ITOZ(ip
);
649 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
650 uint64_t object
= zp
->z_id
;
656 if (zfsctl_is_snapdir(ip
)) {
658 return (zfsctl_snapdir_fid(ip
, fidp
));
661 if (fidp
->fid_len
< SHORT_FID_LEN
) {
662 fidp
->fid_len
= SHORT_FID_LEN
;
664 return (SET_ERROR(ENOSPC
));
667 zfid
= (zfid_short_t
*)fidp
;
669 zfid
->zf_len
= SHORT_FID_LEN
;
671 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
672 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
674 /* .zfs znodes always have a generation number of 0 */
675 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
683 * Construct a full dataset name in full_name: "pool/dataset@snap_name"
686 zfsctl_snapshot_name(zfsvfs_t
*zfsvfs
, const char *snap_name
, int len
,
689 objset_t
*os
= zfsvfs
->z_os
;
691 if (zfs_component_namecheck(snap_name
, NULL
, NULL
) != 0)
692 return (SET_ERROR(EILSEQ
));
694 dmu_objset_name(os
, full_name
);
695 if ((strlen(full_name
) + 1 + strlen(snap_name
)) >= len
)
696 return (SET_ERROR(ENAMETOOLONG
));
698 (void) strcat(full_name
, "@");
699 (void) strcat(full_name
, snap_name
);
705 * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/"
708 zfsctl_snapshot_path(struct path
*path
, int len
, char *full_path
)
710 char *path_buffer
, *path_ptr
;
711 int path_len
, error
= 0;
713 path_buffer
= kmem_alloc(len
, KM_SLEEP
);
715 path_ptr
= d_path(path
, path_buffer
, len
);
716 if (IS_ERR(path_ptr
)) {
717 error
= -PTR_ERR(path_ptr
);
721 path_len
= path_buffer
+ len
- 1 - path_ptr
;
722 if (path_len
> len
) {
723 error
= SET_ERROR(EFAULT
);
727 memcpy(full_path
, path_ptr
, path_len
);
728 full_path
[path_len
] = '\0';
730 kmem_free(path_buffer
, len
);
736 * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/"
739 zfsctl_snapshot_path_objset(zfsvfs_t
*zfsvfs
, uint64_t objsetid
,
740 int path_len
, char *full_path
)
742 objset_t
*os
= zfsvfs
->z_os
;
743 fstrans_cookie_t cookie
;
745 boolean_t case_conflict
;
746 uint64_t id
, pos
= 0;
749 if (zfsvfs
->z_vfs
->vfs_mntpoint
== NULL
)
750 return (SET_ERROR(ENOENT
));
752 cookie
= spl_fstrans_mark();
753 snapname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
756 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
757 error
= dmu_snapshot_list_next(zfsvfs
->z_os
,
758 ZFS_MAX_DATASET_NAME_LEN
, snapname
, &id
, &pos
,
760 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
768 memset(full_path
, 0, path_len
);
769 snprintf(full_path
, path_len
- 1, "%s/.zfs/snapshot/%s",
770 zfsvfs
->z_vfs
->vfs_mntpoint
, snapname
);
772 kmem_free(snapname
, ZFS_MAX_DATASET_NAME_LEN
);
773 spl_fstrans_unmark(cookie
);
779 * Special case the handling of "..".
782 zfsctl_root_lookup(struct inode
*dip
, char *name
, struct inode
**ipp
,
783 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
785 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
790 if (strcmp(name
, "..") == 0) {
791 *ipp
= dip
->i_sb
->s_root
->d_inode
;
792 } else if (strcmp(name
, ZFS_SNAPDIR_NAME
) == 0) {
793 *ipp
= zfsctl_inode_lookup(zfsvfs
, ZFSCTL_INO_SNAPDIR
,
794 &zpl_fops_snapdir
, &zpl_ops_snapdir
);
795 } else if (strcmp(name
, ZFS_SHAREDIR_NAME
) == 0) {
796 *ipp
= zfsctl_inode_lookup(zfsvfs
, ZFSCTL_INO_SHARES
,
797 &zpl_fops_shares
, &zpl_ops_shares
);
803 error
= SET_ERROR(ENOENT
);
811 * Lookup entry point for the 'snapshot' directory. Try to open the
812 * snapshot if it exist, creating the pseudo filesystem inode as necessary.
813 * Perform a mount of the associated dataset on top of the inode.
816 zfsctl_snapdir_lookup(struct inode
*dip
, char *name
, struct inode
**ipp
,
817 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
819 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
825 error
= dmu_snapshot_lookup(zfsvfs
->z_os
, name
, &id
);
831 *ipp
= zfsctl_inode_lookup(zfsvfs
, ZFSCTL_INO_SNAPDIRS
- id
,
832 &simple_dir_operations
, &simple_dir_inode_operations
);
834 error
= SET_ERROR(ENOENT
);
842 * Renaming a directory under '.zfs/snapshot' will automatically trigger
843 * a rename of the snapshot to the new given name. The rename is confined
844 * to the '.zfs/snapshot' directory snapshots cannot be moved elsewhere.
847 zfsctl_snapdir_rename(struct inode
*sdip
, char *snm
,
848 struct inode
*tdip
, char *tnm
, cred_t
*cr
, int flags
)
850 zfsvfs_t
*zfsvfs
= ITOZSB(sdip
);
851 char *to
, *from
, *real
, *fsname
;
854 if (!zfs_admin_snapshot
)
855 return (SET_ERROR(EACCES
));
859 to
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
860 from
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
861 real
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
862 fsname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
864 if (zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
) {
865 error
= dmu_snapshot_realname(zfsvfs
->z_os
, snm
, real
,
866 ZFS_MAX_DATASET_NAME_LEN
, NULL
);
869 } else if (error
!= ENOTSUP
) {
874 dmu_objset_name(zfsvfs
->z_os
, fsname
);
876 error
= zfsctl_snapshot_name(ITOZSB(sdip
), snm
,
877 ZFS_MAX_DATASET_NAME_LEN
, from
);
879 error
= zfsctl_snapshot_name(ITOZSB(tdip
), tnm
,
880 ZFS_MAX_DATASET_NAME_LEN
, to
);
882 error
= zfs_secpolicy_rename_perms(from
, to
, cr
);
887 * Cannot move snapshots out of the snapdir.
890 error
= SET_ERROR(EINVAL
);
895 * No-op when names are identical.
897 if (strcmp(snm
, tnm
) == 0) {
902 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
904 error
= dsl_dataset_rename_snapshot(fsname
, snm
, tnm
, B_FALSE
);
906 (void) zfsctl_snapshot_rename(snm
, tnm
);
908 rw_exit(&zfs_snapshot_lock
);
910 kmem_free(from
, ZFS_MAX_DATASET_NAME_LEN
);
911 kmem_free(to
, ZFS_MAX_DATASET_NAME_LEN
);
912 kmem_free(real
, ZFS_MAX_DATASET_NAME_LEN
);
913 kmem_free(fsname
, ZFS_MAX_DATASET_NAME_LEN
);
921 * Removing a directory under '.zfs/snapshot' will automatically trigger
922 * the removal of the snapshot with the given name.
925 zfsctl_snapdir_remove(struct inode
*dip
, char *name
, cred_t
*cr
, int flags
)
927 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
928 char *snapname
, *real
;
931 if (!zfs_admin_snapshot
)
932 return (SET_ERROR(EACCES
));
936 snapname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
937 real
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
939 if (zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
) {
940 error
= dmu_snapshot_realname(zfsvfs
->z_os
, name
, real
,
941 ZFS_MAX_DATASET_NAME_LEN
, NULL
);
944 } else if (error
!= ENOTSUP
) {
949 error
= zfsctl_snapshot_name(ITOZSB(dip
), name
,
950 ZFS_MAX_DATASET_NAME_LEN
, snapname
);
952 error
= zfs_secpolicy_destroy_perms(snapname
, cr
);
956 error
= zfsctl_snapshot_unmount(snapname
, MNT_FORCE
);
957 if ((error
== 0) || (error
== ENOENT
))
958 error
= dsl_destroy_snapshot(snapname
, B_FALSE
);
960 kmem_free(snapname
, ZFS_MAX_DATASET_NAME_LEN
);
961 kmem_free(real
, ZFS_MAX_DATASET_NAME_LEN
);
969 * Creating a directory under '.zfs/snapshot' will automatically trigger
970 * the creation of a new snapshot with the given name.
973 zfsctl_snapdir_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
,
974 struct inode
**ipp
, cred_t
*cr
, int flags
)
976 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
980 if (!zfs_admin_snapshot
)
981 return (SET_ERROR(EACCES
));
983 dsname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
985 if (zfs_component_namecheck(dirname
, NULL
, NULL
) != 0) {
986 error
= SET_ERROR(EILSEQ
);
990 dmu_objset_name(zfsvfs
->z_os
, dsname
);
992 error
= zfs_secpolicy_snapshot_perms(dsname
, cr
);
997 error
= dmu_objset_snapshot_one(dsname
, dirname
);
1001 error
= zfsctl_snapdir_lookup(dip
, dirname
, ipp
,
1005 kmem_free(dsname
, ZFS_MAX_DATASET_NAME_LEN
);
1011 * Attempt to unmount a snapshot by making a call to user space.
1012 * There is no assurance that this can or will succeed, is just a
1013 * best effort. In the case where it does fail, perhaps because
1014 * it's in use, the unmount will fail harmlessly.
1017 zfsctl_snapshot_unmount(char *snapname
, int flags
)
1019 char *argv
[] = { "/usr/bin/env", "umount", "-t", "zfs", "-n", NULL
,
1021 char *envp
[] = { NULL
};
1022 zfs_snapentry_t
*se
;
1025 rw_enter(&zfs_snapshot_lock
, RW_READER
);
1026 if ((se
= zfsctl_snapshot_find_by_name(snapname
)) == NULL
) {
1027 rw_exit(&zfs_snapshot_lock
);
1028 return (SET_ERROR(ENOENT
));
1030 rw_exit(&zfs_snapshot_lock
);
1032 if (flags
& MNT_FORCE
)
1034 argv
[5] = se
->se_path
;
1035 dprintf("unmount; path=%s\n", se
->se_path
);
1036 error
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
1037 zfsctl_snapshot_rele(se
);
1041 * The umount system utility will return 256 on error. We must
1042 * assume this error is because the file system is busy so it is
1043 * converted to the more sensible EBUSY.
1046 error
= SET_ERROR(EBUSY
);
1051 #define MOUNT_BUSY 0x80 /* Mount failed due to EBUSY (from mntent.h) */
1054 zfsctl_snapshot_mount(struct path
*path
, int flags
)
1056 struct dentry
*dentry
= path
->dentry
;
1057 struct inode
*ip
= dentry
->d_inode
;
1059 zfsvfs_t
*snap_zfsvfs
;
1060 zfs_snapentry_t
*se
;
1061 char *full_name
, *full_path
;
1062 char *argv
[] = { "/usr/bin/env", "mount", "-t", "zfs", "-n", NULL
, NULL
,
1064 char *envp
[] = { NULL
};
1069 return (SET_ERROR(EISDIR
));
1071 zfsvfs
= ITOZSB(ip
);
1074 full_name
= kmem_zalloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
1075 full_path
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
1077 error
= zfsctl_snapshot_name(zfsvfs
, dname(dentry
),
1078 ZFS_MAX_DATASET_NAME_LEN
, full_name
);
1082 error
= zfsctl_snapshot_path(path
, MAXPATHLEN
, full_path
);
1087 * Multiple concurrent automounts of a snapshot are never allowed.
1088 * The snapshot may be manually mounted as many times as desired.
1090 if (zfsctl_snapshot_ismounted(full_name
)) {
1096 * Attempt to mount the snapshot from user space. Normally this
1097 * would be done using the vfs_kern_mount() function, however that
1098 * function is marked GPL-only and cannot be used. On error we
1099 * careful to log the real error to the console and return EISDIR
1100 * to safely abort the automount. This should be very rare.
1102 * If the user mode helper happens to return EBUSY, a concurrent
1103 * mount is already in progress in which case the error is ignored.
1104 * Take note that if the program was executed successfully the return
1105 * value from call_usermodehelper() will be (exitcode << 8 + signal).
1107 dprintf("mount; name=%s path=%s\n", full_name
, full_path
);
1108 argv
[5] = full_name
;
1109 argv
[6] = full_path
;
1110 error
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
1112 if (!(error
& MOUNT_BUSY
<< 8)) {
1113 cmn_err(CE_WARN
, "Unable to automount %s/%s: %d",
1114 full_path
, full_name
, error
);
1115 error
= SET_ERROR(EISDIR
);
1118 * EBUSY, this could mean a concurrent mount, or the
1119 * snapshot has already been mounted at completely
1120 * different place. We return 0 so VFS will retry. For
1121 * the latter case the VFS will retry several times
1122 * and return ELOOP, which is probably not a very good
1131 * Follow down in to the mounted snapshot and set MNT_SHRINKABLE
1132 * to identify this as an automounted filesystem.
1136 if (zpl_follow_down_one(&spath
)) {
1137 snap_zfsvfs
= ITOZSB(spath
.dentry
->d_inode
);
1138 snap_zfsvfs
->z_parent
= zfsvfs
;
1139 dentry
= spath
.dentry
;
1140 spath
.mnt
->mnt_flags
|= MNT_SHRINKABLE
;
1142 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
1143 se
= zfsctl_snapshot_alloc(full_name
, full_path
,
1144 snap_zfsvfs
->z_os
->os_spa
, dmu_objset_id(snap_zfsvfs
->z_os
),
1146 zfsctl_snapshot_add(se
);
1147 zfsctl_snapshot_unmount_delay_impl(se
, zfs_expire_snapshot
);
1148 rw_exit(&zfs_snapshot_lock
);
1152 kmem_free(full_name
, ZFS_MAX_DATASET_NAME_LEN
);
1153 kmem_free(full_path
, MAXPATHLEN
);
1161 * Get the snapdir inode from fid
1164 zfsctl_snapdir_vget(struct super_block
*sb
, uint64_t objsetid
, int gen
,
1170 struct dentry
*dentry
;
1172 mnt
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
1174 error
= zfsctl_snapshot_path_objset(sb
->s_fs_info
, objsetid
,
1179 /* Trigger automount */
1180 error
= -kern_path(mnt
, LOOKUP_FOLLOW
|LOOKUP_DIRECTORY
, &path
);
1186 * Get the snapdir inode. Note, we don't want to use the above
1187 * path because it contains the root of the snapshot rather
1190 *ipp
= ilookup(sb
, ZFSCTL_INO_SNAPDIRS
- objsetid
);
1192 error
= SET_ERROR(ENOENT
);
1196 /* check gen, see zfsctl_snapdir_fid */
1197 dentry
= d_obtain_alias(igrab(*ipp
));
1198 if (gen
!= (!IS_ERR(dentry
) && d_mountpoint(dentry
))) {
1201 error
= SET_ERROR(ENOENT
);
1203 if (!IS_ERR(dentry
))
1206 kmem_free(mnt
, MAXPATHLEN
);
1211 zfsctl_shares_lookup(struct inode
*dip
, char *name
, struct inode
**ipp
,
1212 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1214 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1221 if (zfsvfs
->z_shares_dir
== 0) {
1223 return (SET_ERROR(ENOTSUP
));
1226 if ((error
= zfs_zget(zfsvfs
, zfsvfs
->z_shares_dir
, &dzp
)) == 0) {
1227 error
= zfs_lookup(ZTOI(dzp
), name
, &ip
, 0, cr
, NULL
, NULL
);
1237 * Initialize the various pieces we'll need to create and manipulate .zfs
1238 * directories. Currently this is unused but available.
1243 avl_create(&zfs_snapshots_by_name
, snapentry_compare_by_name
,
1244 sizeof (zfs_snapentry_t
), offsetof(zfs_snapentry_t
,
1246 avl_create(&zfs_snapshots_by_objsetid
, snapentry_compare_by_objsetid
,
1247 sizeof (zfs_snapentry_t
), offsetof(zfs_snapentry_t
,
1249 rw_init(&zfs_snapshot_lock
, NULL
, RW_DEFAULT
, NULL
);
1253 * Cleanup the various pieces we needed for .zfs directories. In particular
1254 * ensure the expiry timer is canceled safely.
1259 avl_destroy(&zfs_snapshots_by_name
);
1260 avl_destroy(&zfs_snapshots_by_objsetid
);
1261 rw_destroy(&zfs_snapshot_lock
);
1264 module_param(zfs_admin_snapshot
, int, 0644);
1265 MODULE_PARM_DESC(zfs_admin_snapshot
, "Enable mkdir/rmdir/mv in .zfs/snapshot");
1267 module_param(zfs_expire_snapshot
, int, 0644);
1268 MODULE_PARM_DESC(zfs_expire_snapshot
, "Seconds to expire .zfs/snapshot");