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 zfs_sb_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 * zfs_sb_t. However, the fsid reported by these mounts will be the same
68 * as that used by the parent zfs_sb_t to make NFS happy.
71 #include <sys/types.h>
72 #include <sys/param.h>
74 #include <sys/systm.h>
75 #include <sys/sysmacros.h>
76 #include <sys/pathname.h>
78 #include <sys/vfs_opreg.h>
79 #include <sys/zfs_ctldir.h>
80 #include <sys/zfs_ioctl.h>
81 #include <sys/zfs_vfsops.h>
82 #include <sys/zfs_vnops.h>
85 #include <sys/dmu_objset.h>
86 #include <sys/dsl_destroy.h>
87 #include <sys/dsl_deleg.h>
88 #include <sys/mount.h>
90 #include "zfs_namecheck.h"
93 * Two AVL trees are maintained which contain all currently automounted
94 * snapshots. Every automounted snapshots maps to a single zfs_snapentry_t
97 * - be attached to both trees, and
98 * - be unique, no duplicate entries are allowed.
100 * The zfs_snapshots_by_name tree is indexed by the full dataset name
101 * while the zfs_snapshots_by_objsetid tree is indexed by the unique
102 * objsetid. This allows for fast lookups either by name or objsetid.
104 static avl_tree_t zfs_snapshots_by_name
;
105 static avl_tree_t zfs_snapshots_by_objsetid
;
106 static krwlock_t zfs_snapshot_lock
;
109 * Control Directory Tunables (.zfs)
111 int zfs_expire_snapshot
= ZFSCTL_EXPIRE_SNAPSHOT
;
112 int zfs_admin_snapshot
= 1;
115 * Dedicated task queue for unmounting snapshots.
117 static taskq_t
*zfs_expire_taskq
;
120 char *se_name
; /* full snapshot name */
121 char *se_path
; /* full mount path */
122 spa_t
*se_spa
; /* pool spa */
123 uint64_t se_objsetid
; /* snapshot objset id */
124 struct dentry
*se_root_dentry
; /* snapshot root dentry */
125 taskqid_t se_taskqid
; /* scheduled unmount taskqid */
126 avl_node_t se_node_name
; /* zfs_snapshots_by_name link */
127 avl_node_t se_node_objsetid
; /* zfs_snapshots_by_objsetid link */
128 refcount_t se_refcount
; /* reference count */
131 static void zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t
*se
, int delay
);
134 * Allocate a new zfs_snapentry_t being careful to make a copy of the
135 * the snapshot name and provided mount point. No reference is taken.
137 static zfs_snapentry_t
*
138 zfsctl_snapshot_alloc(char *full_name
, char *full_path
, spa_t
*spa
,
139 uint64_t objsetid
, struct dentry
*root_dentry
)
143 se
= kmem_zalloc(sizeof (zfs_snapentry_t
), KM_SLEEP
);
145 se
->se_name
= strdup(full_name
);
146 se
->se_path
= strdup(full_path
);
148 se
->se_objsetid
= objsetid
;
149 se
->se_root_dentry
= root_dentry
;
152 refcount_create(&se
->se_refcount
);
158 * Free a zfs_snapentry_t the called must ensure there are no active
162 zfsctl_snapshot_free(zfs_snapentry_t
*se
)
164 refcount_destroy(&se
->se_refcount
);
165 strfree(se
->se_name
);
166 strfree(se
->se_path
);
168 kmem_free(se
, sizeof (zfs_snapentry_t
));
172 * Hold a reference on the zfs_snapentry_t.
175 zfsctl_snapshot_hold(zfs_snapentry_t
*se
)
177 refcount_add(&se
->se_refcount
, NULL
);
181 * Release a reference on the zfs_snapentry_t. When the number of
182 * references drops to zero the structure will be freed.
185 zfsctl_snapshot_rele(zfs_snapentry_t
*se
)
187 if (refcount_remove(&se
->se_refcount
, NULL
) == 0)
188 zfsctl_snapshot_free(se
);
192 * Add a zfs_snapentry_t to both the zfs_snapshots_by_name and
193 * zfs_snapshots_by_objsetid trees. While the zfs_snapentry_t is part
194 * of the trees a reference is held.
197 zfsctl_snapshot_add(zfs_snapentry_t
*se
)
199 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock
));
200 refcount_add(&se
->se_refcount
, NULL
);
201 avl_add(&zfs_snapshots_by_name
, se
);
202 avl_add(&zfs_snapshots_by_objsetid
, se
);
206 * Remove a zfs_snapentry_t from both the zfs_snapshots_by_name and
207 * zfs_snapshots_by_objsetid trees. Upon removal a reference is dropped,
208 * this can result in the structure being freed if that was the last
209 * remaining reference.
212 zfsctl_snapshot_remove(zfs_snapentry_t
*se
)
214 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock
));
215 avl_remove(&zfs_snapshots_by_name
, se
);
216 avl_remove(&zfs_snapshots_by_objsetid
, se
);
217 zfsctl_snapshot_rele(se
);
221 * Snapshot name comparison function for the zfs_snapshots_by_name.
224 snapentry_compare_by_name(const void *a
, const void *b
)
226 const zfs_snapentry_t
*se_a
= a
;
227 const zfs_snapentry_t
*se_b
= b
;
230 ret
= strcmp(se_a
->se_name
, se_b
->se_name
);
241 * Snapshot name comparison function for the zfs_snapshots_by_objsetid.
244 snapentry_compare_by_objsetid(const void *a
, const void *b
)
246 const zfs_snapentry_t
*se_a
= a
;
247 const zfs_snapentry_t
*se_b
= b
;
249 if (se_a
->se_spa
!= se_b
->se_spa
)
250 return ((ulong_t
)se_a
->se_spa
< (ulong_t
)se_b
->se_spa
? -1 : 1);
252 if (se_a
->se_objsetid
< se_b
->se_objsetid
)
254 else if (se_a
->se_objsetid
> se_b
->se_objsetid
)
261 * Find a zfs_snapentry_t in zfs_snapshots_by_name. If the snapname
262 * is found a pointer to the zfs_snapentry_t is returned and a reference
263 * taken on the structure. The caller is responsible for dropping the
264 * reference with zfsctl_snapshot_rele(). If the snapname is not found
265 * NULL will be returned.
267 static zfs_snapentry_t
*
268 zfsctl_snapshot_find_by_name(char *snapname
)
270 zfs_snapentry_t
*se
, search
;
272 ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock
));
274 search
.se_name
= snapname
;
275 se
= avl_find(&zfs_snapshots_by_name
, &search
, NULL
);
277 refcount_add(&se
->se_refcount
, NULL
);
283 * Find a zfs_snapentry_t in zfs_snapshots_by_objsetid given the objset id
284 * rather than the snapname. In all other respects it behaves the same
285 * as zfsctl_snapshot_find_by_name().
287 static zfs_snapentry_t
*
288 zfsctl_snapshot_find_by_objsetid(spa_t
*spa
, uint64_t objsetid
)
290 zfs_snapentry_t
*se
, search
;
292 ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock
));
295 search
.se_objsetid
= objsetid
;
296 se
= avl_find(&zfs_snapshots_by_objsetid
, &search
, NULL
);
298 refcount_add(&se
->se_refcount
, NULL
);
304 * Rename a zfs_snapentry_t in the zfs_snapshots_by_name. The structure is
305 * removed, renamed, and added back to the new correct location in the tree.
308 zfsctl_snapshot_rename(char *old_snapname
, char *new_snapname
)
312 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock
));
314 se
= zfsctl_snapshot_find_by_name(old_snapname
);
318 zfsctl_snapshot_remove(se
);
319 strfree(se
->se_name
);
320 se
->se_name
= strdup(new_snapname
);
321 zfsctl_snapshot_add(se
);
322 zfsctl_snapshot_rele(se
);
328 * Delayed task responsible for unmounting an expired automounted snapshot.
331 snapentry_expire(void *data
)
333 zfs_snapentry_t
*se
= (zfs_snapentry_t
*)data
;
334 spa_t
*spa
= se
->se_spa
;
335 uint64_t objsetid
= se
->se_objsetid
;
337 if (zfs_expire_snapshot
<= 0) {
338 zfsctl_snapshot_rele(se
);
343 (void) zfsctl_snapshot_unmount(se
->se_name
, MNT_EXPIRE
);
344 zfsctl_snapshot_rele(se
);
347 * Reschedule the unmount if the zfs_snapentry_t wasn't removed.
348 * This can occur when the snapshot is busy.
350 rw_enter(&zfs_snapshot_lock
, RW_READER
);
351 if ((se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
)) != NULL
) {
352 zfsctl_snapshot_unmount_delay_impl(se
, zfs_expire_snapshot
);
353 zfsctl_snapshot_rele(se
);
355 rw_exit(&zfs_snapshot_lock
);
359 * Cancel an automatic unmount of a snapname. This callback is responsible
360 * for dropping the reference on the zfs_snapentry_t which was taken when
364 zfsctl_snapshot_unmount_cancel(zfs_snapentry_t
*se
)
366 ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock
));
368 if (taskq_cancel_id(zfs_expire_taskq
, se
->se_taskqid
) == 0) {
370 zfsctl_snapshot_rele(se
);
375 * Dispatch the unmount task for delayed handling with a hold protecting it.
378 zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t
*se
, int delay
)
380 ASSERT3S(se
->se_taskqid
, ==, -1);
385 zfsctl_snapshot_hold(se
);
386 se
->se_taskqid
= taskq_dispatch_delay(zfs_expire_taskq
,
387 snapentry_expire
, se
, TQ_SLEEP
, ddi_get_lbolt() + delay
* HZ
);
391 * Schedule an automatic unmount of objset id to occur in delay seconds from
392 * now. Any previous delayed unmount will be cancelled in favor of the
393 * updated deadline. A reference is taken by zfsctl_snapshot_find_by_name()
394 * and held until the outstanding task is handled or cancelled.
397 zfsctl_snapshot_unmount_delay(spa_t
*spa
, uint64_t objsetid
, int delay
)
402 rw_enter(&zfs_snapshot_lock
, RW_READER
);
403 if ((se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
)) != NULL
) {
404 zfsctl_snapshot_unmount_cancel(se
);
405 zfsctl_snapshot_unmount_delay_impl(se
, delay
);
406 zfsctl_snapshot_rele(se
);
409 rw_exit(&zfs_snapshot_lock
);
415 * Check if snapname is currently mounted. Returned non-zero when mounted
416 * and zero when unmounted.
419 zfsctl_snapshot_ismounted(char *snapname
)
422 boolean_t ismounted
= B_FALSE
;
424 rw_enter(&zfs_snapshot_lock
, RW_READER
);
425 if ((se
= zfsctl_snapshot_find_by_name(snapname
)) != NULL
) {
426 zfsctl_snapshot_rele(se
);
429 rw_exit(&zfs_snapshot_lock
);
435 * Check if the given inode is a part of the virtual .zfs directory.
438 zfsctl_is_node(struct inode
*ip
)
440 return (ITOZ(ip
)->z_is_ctldir
);
444 * Check if the given inode is a .zfs/snapshots/snapname directory.
447 zfsctl_is_snapdir(struct inode
*ip
)
449 return (zfsctl_is_node(ip
) && (ip
->i_ino
<= ZFSCTL_INO_SNAPDIRS
));
453 * Allocate a new inode with the passed id and ops.
455 static struct inode
*
456 zfsctl_inode_alloc(zfs_sb_t
*zsb
, uint64_t id
,
457 const struct file_operations
*fops
, const struct inode_operations
*ops
)
459 struct timespec now
= current_fs_time(zsb
->z_sb
);
463 ip
= new_inode(zsb
->z_sb
);
468 ASSERT3P(zp
->z_dirlocks
, ==, NULL
);
469 ASSERT3P(zp
->z_acl_cached
, ==, NULL
);
470 ASSERT3P(zp
->z_xattr_cached
, ==, NULL
);
473 zp
->z_atime_dirty
= 0;
474 zp
->z_zn_prefetch
= 0;
484 zp
->z_is_mapped
= B_FALSE
;
485 zp
->z_is_ctldir
= B_TRUE
;
486 zp
->z_is_sa
= B_FALSE
;
487 zp
->z_is_stale
= B_FALSE
;
488 ip
->i_generation
= 0;
490 ip
->i_mode
= (S_IFDIR
| S_IRWXUGO
);
491 ip
->i_uid
= SUID_TO_KUID(0);
492 ip
->i_gid
= SGID_TO_KGID(0);
493 ip
->i_blkbits
= SPA_MINBLOCKSHIFT
;
500 if (insert_inode_locked(ip
)) {
501 unlock_new_inode(ip
);
506 mutex_enter(&zsb
->z_znodes_lock
);
507 list_insert_tail(&zsb
->z_all_znodes
, zp
);
510 mutex_exit(&zsb
->z_znodes_lock
);
512 unlock_new_inode(ip
);
518 * Lookup the inode with given id, it will be allocated if needed.
520 static struct inode
*
521 zfsctl_inode_lookup(zfs_sb_t
*zsb
, uint64_t id
,
522 const struct file_operations
*fops
, const struct inode_operations
*ops
)
524 struct inode
*ip
= NULL
;
527 ip
= ilookup(zsb
->z_sb
, (unsigned long)id
);
531 /* May fail due to concurrent zfsctl_inode_alloc() */
532 ip
= zfsctl_inode_alloc(zsb
, id
, fops
, ops
);
539 * Create the '.zfs' directory. This directory is cached as part of the VFS
540 * structure. This results in a hold on the zfs_sb_t. The code in zfs_umount()
541 * therefore checks against a vfs_count of 2 instead of 1. This reference
542 * is removed when the ctldir is destroyed in the unmount. All other entities
543 * under the '.zfs' directory are created dynamically as needed.
545 * Because the dynamically created '.zfs' directory entries assume the use
546 * of 64-bit inode numbers this support must be disabled on 32-bit systems.
549 zfsctl_create(zfs_sb_t
*zsb
)
551 #if defined(CONFIG_64BIT)
552 ASSERT(zsb
->z_ctldir
== NULL
);
554 zsb
->z_ctldir
= zfsctl_inode_alloc(zsb
, ZFSCTL_INO_ROOT
,
555 &zpl_fops_root
, &zpl_ops_root
);
556 if (zsb
->z_ctldir
== NULL
)
557 return (SET_ERROR(ENOENT
));
561 return (SET_ERROR(EOPNOTSUPP
));
562 #endif /* CONFIG_64BIT */
566 * Destroy the '.zfs' directory or remove a snapshot from zfs_snapshots_by_name.
567 * Only called when the filesystem is unmounted.
570 zfsctl_destroy(zfs_sb_t
*zsb
)
574 spa_t
*spa
= zsb
->z_os
->os_spa
;
575 uint64_t objsetid
= dmu_objset_id(zsb
->z_os
);
577 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
578 if ((se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
))
580 zfsctl_snapshot_unmount_cancel(se
);
581 zfsctl_snapshot_remove(se
);
582 zfsctl_snapshot_rele(se
);
584 rw_exit(&zfs_snapshot_lock
);
585 } else if (zsb
->z_ctldir
) {
587 zsb
->z_ctldir
= NULL
;
592 * Given a root znode, retrieve the associated .zfs directory.
593 * Add a hold to the vnode and return it.
596 zfsctl_root(znode_t
*zp
)
598 ASSERT(zfs_has_ctldir(zp
));
599 igrab(ZTOZSB(zp
)->z_ctldir
);
600 return (ZTOZSB(zp
)->z_ctldir
);
603 * Generate a long fid which includes the root object and objset of a
604 * snapshot but not the generation number. For the root object the
605 * generation number is ignored when zero to avoid needing to open
606 * the dataset when generating fids for the snapshot names.
609 zfsctl_snapdir_fid(struct inode
*ip
, fid_t
*fidp
)
611 zfs_sb_t
*zsb
= ITOZSB(ip
);
612 zfid_short_t
*zfid
= (zfid_short_t
*)fidp
;
613 zfid_long_t
*zlfid
= (zfid_long_t
*)fidp
;
619 object
= zsb
->z_root
;
620 objsetid
= ZFSCTL_INO_SNAPDIRS
- ip
->i_ino
;
621 zfid
->zf_len
= LONG_FID_LEN
;
623 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
624 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
626 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
627 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
629 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
630 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
632 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
633 zlfid
->zf_setgen
[i
] = 0;
639 * Generate an appropriate fid for an entry in the .zfs directory.
642 zfsctl_fid(struct inode
*ip
, fid_t
*fidp
)
644 znode_t
*zp
= ITOZ(ip
);
645 zfs_sb_t
*zsb
= ITOZSB(ip
);
646 uint64_t object
= zp
->z_id
;
652 if (fidp
->fid_len
< SHORT_FID_LEN
) {
653 fidp
->fid_len
= SHORT_FID_LEN
;
655 return (SET_ERROR(ENOSPC
));
658 if (zfsctl_is_snapdir(ip
)) {
660 return (zfsctl_snapdir_fid(ip
, fidp
));
663 zfid
= (zfid_short_t
*)fidp
;
665 zfid
->zf_len
= SHORT_FID_LEN
;
667 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
668 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
670 /* .zfs znodes always have a generation number of 0 */
671 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
679 * Construct a full dataset name in full_name: "pool/dataset@snap_name"
682 zfsctl_snapshot_name(zfs_sb_t
*zsb
, const char *snap_name
, int len
,
685 objset_t
*os
= zsb
->z_os
;
687 if (zfs_component_namecheck(snap_name
, NULL
, NULL
) != 0)
688 return (SET_ERROR(EILSEQ
));
690 dmu_objset_name(os
, full_name
);
691 if ((strlen(full_name
) + 1 + strlen(snap_name
)) >= len
)
692 return (SET_ERROR(ENAMETOOLONG
));
694 (void) strcat(full_name
, "@");
695 (void) strcat(full_name
, snap_name
);
701 * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/"
704 zfsctl_snapshot_path(struct path
*path
, int len
, char *full_path
)
706 char *path_buffer
, *path_ptr
;
707 int path_len
, error
= 0;
709 path_buffer
= kmem_alloc(len
, KM_SLEEP
);
711 path_ptr
= d_path(path
, path_buffer
, len
);
712 if (IS_ERR(path_ptr
)) {
713 error
= -PTR_ERR(path_ptr
);
717 path_len
= path_buffer
+ len
- 1 - path_ptr
;
718 if (path_len
> len
) {
719 error
= SET_ERROR(EFAULT
);
723 memcpy(full_path
, path_ptr
, path_len
);
724 full_path
[path_len
] = '\0';
726 kmem_free(path_buffer
, len
);
732 * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/"
735 zfsctl_snapshot_path_objset(zfs_sb_t
*zsb
, uint64_t objsetid
,
736 int path_len
, char *full_path
)
738 objset_t
*os
= zsb
->z_os
;
739 fstrans_cookie_t cookie
;
741 boolean_t case_conflict
;
742 uint64_t id
, pos
= 0;
745 if (zsb
->z_mntopts
->z_mntpoint
== NULL
)
748 cookie
= spl_fstrans_mark();
749 snapname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
752 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
753 error
= dmu_snapshot_list_next(zsb
->z_os
,
754 ZFS_MAX_DATASET_NAME_LEN
, snapname
, &id
, &pos
,
756 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
764 memset(full_path
, 0, path_len
);
765 snprintf(full_path
, path_len
- 1, "%s/.zfs/snapshot/%s",
766 zsb
->z_mntopts
->z_mntpoint
, snapname
);
768 kmem_free(snapname
, ZFS_MAX_DATASET_NAME_LEN
);
769 spl_fstrans_unmark(cookie
);
775 * Special case the handling of "..".
778 zfsctl_root_lookup(struct inode
*dip
, char *name
, struct inode
**ipp
,
779 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
781 zfs_sb_t
*zsb
= ITOZSB(dip
);
786 if (strcmp(name
, "..") == 0) {
787 *ipp
= dip
->i_sb
->s_root
->d_inode
;
788 } else if (strcmp(name
, ZFS_SNAPDIR_NAME
) == 0) {
789 *ipp
= zfsctl_inode_lookup(zsb
, ZFSCTL_INO_SNAPDIR
,
790 &zpl_fops_snapdir
, &zpl_ops_snapdir
);
791 } else if (strcmp(name
, ZFS_SHAREDIR_NAME
) == 0) {
792 *ipp
= zfsctl_inode_lookup(zsb
, ZFSCTL_INO_SHARES
,
793 &zpl_fops_shares
, &zpl_ops_shares
);
799 error
= SET_ERROR(ENOENT
);
807 * Lookup entry point for the 'snapshot' directory. Try to open the
808 * snapshot if it exist, creating the pseudo filesystem inode as necessary.
809 * Perform a mount of the associated dataset on top of the inode.
812 zfsctl_snapdir_lookup(struct inode
*dip
, char *name
, struct inode
**ipp
,
813 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
815 zfs_sb_t
*zsb
= ITOZSB(dip
);
821 error
= dmu_snapshot_lookup(zsb
->z_os
, name
, &id
);
827 *ipp
= zfsctl_inode_lookup(zsb
, ZFSCTL_INO_SNAPDIRS
- id
,
828 &simple_dir_operations
, &simple_dir_inode_operations
);
830 error
= SET_ERROR(ENOENT
);
838 * Renaming a directory under '.zfs/snapshot' will automatically trigger
839 * a rename of the snapshot to the new given name. The rename is confined
840 * to the '.zfs/snapshot' directory snapshots cannot be moved elsewhere.
843 zfsctl_snapdir_rename(struct inode
*sdip
, char *snm
,
844 struct inode
*tdip
, char *tnm
, cred_t
*cr
, int flags
)
846 zfs_sb_t
*zsb
= ITOZSB(sdip
);
847 char *to
, *from
, *real
, *fsname
;
850 if (!zfs_admin_snapshot
)
855 to
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
856 from
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
857 real
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
858 fsname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
860 if (zsb
->z_case
== ZFS_CASE_INSENSITIVE
) {
861 error
= dmu_snapshot_realname(zsb
->z_os
, snm
, real
,
862 ZFS_MAX_DATASET_NAME_LEN
, NULL
);
865 } else if (error
!= ENOTSUP
) {
870 dmu_objset_name(zsb
->z_os
, fsname
);
872 error
= zfsctl_snapshot_name(ITOZSB(sdip
), snm
,
873 ZFS_MAX_DATASET_NAME_LEN
, from
);
875 error
= zfsctl_snapshot_name(ITOZSB(tdip
), tnm
,
876 ZFS_MAX_DATASET_NAME_LEN
, to
);
878 error
= zfs_secpolicy_rename_perms(from
, to
, cr
);
883 * Cannot move snapshots out of the snapdir.
886 error
= SET_ERROR(EINVAL
);
891 * No-op when names are identical.
893 if (strcmp(snm
, tnm
) == 0) {
898 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
900 error
= dsl_dataset_rename_snapshot(fsname
, snm
, tnm
, B_FALSE
);
902 (void) zfsctl_snapshot_rename(snm
, tnm
);
904 rw_exit(&zfs_snapshot_lock
);
906 kmem_free(from
, ZFS_MAX_DATASET_NAME_LEN
);
907 kmem_free(to
, ZFS_MAX_DATASET_NAME_LEN
);
908 kmem_free(real
, ZFS_MAX_DATASET_NAME_LEN
);
909 kmem_free(fsname
, ZFS_MAX_DATASET_NAME_LEN
);
917 * Removing a directory under '.zfs/snapshot' will automatically trigger
918 * the removal of the snapshot with the given name.
921 zfsctl_snapdir_remove(struct inode
*dip
, char *name
, cred_t
*cr
, int flags
)
923 zfs_sb_t
*zsb
= ITOZSB(dip
);
924 char *snapname
, *real
;
927 if (!zfs_admin_snapshot
)
932 snapname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
933 real
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
935 if (zsb
->z_case
== ZFS_CASE_INSENSITIVE
) {
936 error
= dmu_snapshot_realname(zsb
->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
);
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
, char *dirname
, vattr_t
*vap
,
970 struct inode
**ipp
, cred_t
*cr
, int flags
)
972 zfs_sb_t
*zsb
= ITOZSB(dip
);
976 if (!zfs_admin_snapshot
)
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(zsb
->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 * Attempt to unmount a snapshot by making a call to user space.
1008 * There is no assurance that this can or will succeed, is just a
1009 * best effort. In the case where it does fail, perhaps because
1010 * it's in use, the unmount will fail harmlessly.
1012 #define SET_UNMOUNT_CMD \
1013 "exec 0</dev/null " \
1016 "umount -t zfs -n %s'%s'"
1019 zfsctl_snapshot_unmount(char *snapname
, int flags
)
1021 char *argv
[] = { "/bin/sh", "-c", NULL
, NULL
};
1022 char *envp
[] = { NULL
};
1023 zfs_snapentry_t
*se
;
1026 rw_enter(&zfs_snapshot_lock
, RW_READER
);
1027 if ((se
= zfsctl_snapshot_find_by_name(snapname
)) == NULL
) {
1028 rw_exit(&zfs_snapshot_lock
);
1031 rw_exit(&zfs_snapshot_lock
);
1033 argv
[2] = kmem_asprintf(SET_UNMOUNT_CMD
,
1034 flags
& MNT_FORCE
? "-f " : "", se
->se_path
);
1035 dprintf("unmount; path=%s\n", se
->se_path
);
1036 error
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
1038 zfsctl_snapshot_rele(se
);
1042 * The umount system utility will return 256 on error. We must
1043 * assume this error is because the file system is busy so it is
1044 * converted to the more sensible EBUSY.
1047 error
= SET_ERROR(EBUSY
);
1052 #define MOUNT_BUSY 0x80 /* Mount failed due to EBUSY (from mntent.h) */
1053 #define SET_MOUNT_CMD \
1054 "exec 0</dev/null " \
1057 "mount -t zfs -n '%s' '%s'"
1060 zfsctl_snapshot_mount(struct path
*path
, int flags
)
1062 struct dentry
*dentry
= path
->dentry
;
1063 struct inode
*ip
= dentry
->d_inode
;
1066 zfs_snapentry_t
*se
;
1067 char *full_name
, *full_path
;
1068 char *argv
[] = { "/bin/sh", "-c", NULL
, NULL
};
1069 char *envp
[] = { NULL
};
1079 full_name
= kmem_zalloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
1080 full_path
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
1082 error
= zfsctl_snapshot_name(zsb
, dname(dentry
),
1083 ZFS_MAX_DATASET_NAME_LEN
, full_name
);
1087 error
= zfsctl_snapshot_path(path
, MAXPATHLEN
, full_path
);
1092 * Multiple concurrent automounts of a snapshot are never allowed.
1093 * The snapshot may be manually mounted as many times as desired.
1095 if (zfsctl_snapshot_ismounted(full_name
)) {
1101 * Attempt to mount the snapshot from user space. Normally this
1102 * would be done using the vfs_kern_mount() function, however that
1103 * function is marked GPL-only and cannot be used. On error we
1104 * careful to log the real error to the console and return EISDIR
1105 * to safely abort the automount. This should be very rare.
1107 * If the user mode helper happens to return EBUSY, a concurrent
1108 * mount is already in progress in which case the error is ignored.
1109 * Take note that if the program was executed successfully the return
1110 * value from call_usermodehelper() will be (exitcode << 8 + signal).
1112 dprintf("mount; name=%s path=%s\n", full_name
, full_path
);
1113 argv
[2] = kmem_asprintf(SET_MOUNT_CMD
, full_name
, full_path
);
1114 error
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
1117 if (!(error
& MOUNT_BUSY
<< 8)) {
1118 cmn_err(CE_WARN
, "Unable to automount %s/%s: %d",
1119 full_path
, full_name
, error
);
1120 error
= SET_ERROR(EISDIR
);
1123 * EBUSY, this could mean a concurrent mount, or the
1124 * snapshot has already been mounted at completely
1125 * different place. We return 0 so VFS will retry. For
1126 * the latter case the VFS will retry several times
1127 * and return ELOOP, which is probably not a very good
1136 * Follow down in to the mounted snapshot and set MNT_SHRINKABLE
1137 * to identify this as an automounted filesystem.
1141 if (zpl_follow_down_one(&spath
)) {
1142 snap_zsb
= ITOZSB(spath
.dentry
->d_inode
);
1143 snap_zsb
->z_parent
= zsb
;
1144 dentry
= spath
.dentry
;
1145 spath
.mnt
->mnt_flags
|= MNT_SHRINKABLE
;
1147 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
1148 se
= zfsctl_snapshot_alloc(full_name
, full_path
,
1149 snap_zsb
->z_os
->os_spa
, dmu_objset_id(snap_zsb
->z_os
),
1151 zfsctl_snapshot_add(se
);
1152 zfsctl_snapshot_unmount_delay_impl(se
, zfs_expire_snapshot
);
1153 rw_exit(&zfs_snapshot_lock
);
1157 kmem_free(full_name
, ZFS_MAX_DATASET_NAME_LEN
);
1158 kmem_free(full_path
, MAXPATHLEN
);
1166 * Given the objset id of the snapshot return its zfs_sb_t as zsbp.
1169 zfsctl_lookup_objset(struct super_block
*sb
, uint64_t objsetid
, zfs_sb_t
**zsbp
)
1171 zfs_snapentry_t
*se
;
1173 spa_t
*spa
= ((zfs_sb_t
*)(sb
->s_fs_info
))->z_os
->os_spa
;
1176 * Verify that the snapshot is mounted then lookup the mounted root
1177 * rather than the covered mount point. This may fail if the
1178 * snapshot has just been unmounted by an unrelated user space
1179 * process. This race cannot occur to an expired mount point
1180 * because we hold the zfs_snapshot_lock to prevent the race.
1182 rw_enter(&zfs_snapshot_lock
, RW_READER
);
1183 if ((se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
)) != NULL
) {
1186 zsb
= ITOZSB(se
->se_root_dentry
->d_inode
);
1187 ASSERT3U(dmu_objset_id(zsb
->z_os
), ==, objsetid
);
1189 if (time_after(jiffies
, zsb
->z_snap_defer_time
+
1190 MAX(zfs_expire_snapshot
* HZ
/ 2, HZ
))) {
1191 zsb
->z_snap_defer_time
= jiffies
;
1192 zfsctl_snapshot_unmount_cancel(se
);
1193 zfsctl_snapshot_unmount_delay_impl(se
,
1194 zfs_expire_snapshot
);
1198 zfsctl_snapshot_rele(se
);
1199 error
= SET_ERROR(0);
1201 error
= SET_ERROR(ENOENT
);
1203 rw_exit(&zfs_snapshot_lock
);
1206 * Automount the snapshot given the objset id by constructing the
1207 * full mount point and performing a traversal.
1209 if (error
== ENOENT
) {
1213 mnt
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
1214 error
= zfsctl_snapshot_path_objset(sb
->s_fs_info
, objsetid
,
1217 kmem_free(mnt
, MAXPATHLEN
);
1218 return (SET_ERROR(error
));
1221 error
= kern_path(mnt
, LOOKUP_FOLLOW
|LOOKUP_DIRECTORY
, &path
);
1223 *zsbp
= ITOZSB(path
.dentry
->d_inode
);
1227 kmem_free(mnt
, MAXPATHLEN
);
1234 zfsctl_shares_lookup(struct inode
*dip
, char *name
, struct inode
**ipp
,
1235 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1237 zfs_sb_t
*zsb
= ITOZSB(dip
);
1244 if (zsb
->z_shares_dir
== 0) {
1246 return (SET_ERROR(ENOTSUP
));
1249 if ((error
= zfs_zget(zsb
, zsb
->z_shares_dir
, &dzp
)) == 0) {
1250 error
= zfs_lookup(ZTOI(dzp
), name
, &ip
, 0, cr
, NULL
, NULL
);
1260 * Initialize the various pieces we'll need to create and manipulate .zfs
1261 * directories. Currently this is unused but available.
1266 avl_create(&zfs_snapshots_by_name
, snapentry_compare_by_name
,
1267 sizeof (zfs_snapentry_t
), offsetof(zfs_snapentry_t
,
1269 avl_create(&zfs_snapshots_by_objsetid
, snapentry_compare_by_objsetid
,
1270 sizeof (zfs_snapentry_t
), offsetof(zfs_snapentry_t
,
1272 rw_init(&zfs_snapshot_lock
, NULL
, RW_DEFAULT
, NULL
);
1274 zfs_expire_taskq
= taskq_create("z_unmount", 1, defclsyspri
,
1275 1, 8, TASKQ_PREPOPULATE
);
1279 * Cleanup the various pieces we needed for .zfs directories. In particular
1280 * ensure the expiry timer is canceled safely.
1285 taskq_destroy(zfs_expire_taskq
);
1287 avl_destroy(&zfs_snapshots_by_name
);
1288 avl_destroy(&zfs_snapshots_by_objsetid
);
1289 rw_destroy(&zfs_snapshot_lock
);
1292 module_param(zfs_admin_snapshot
, int, 0644);
1293 MODULE_PARM_DESC(zfs_admin_snapshot
, "Enable mkdir/rmdir/mv in .zfs/snapshot");
1295 module_param(zfs_expire_snapshot
, int, 0644);
1296 MODULE_PARM_DESC(zfs_expire_snapshot
, "Seconds to expire .zfs/snapshot");