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.
32 * Copyright (c) 2018 George Melikov. All Rights Reserved.
36 * ZFS control directory (a.k.a. ".zfs")
38 * This directory provides a common location for all ZFS meta-objects.
39 * Currently, this is only the 'snapshot' and 'shares' directory, but this may
40 * expand in the future. The elements are built dynamically, as the hierarchy
41 * does not actually exist on disk.
43 * For 'snapshot', we don't want to have all snapshots always mounted, because
44 * this would take up a huge amount of space in /etc/mnttab. We have three
47 * ctldir ------> snapshotdir -------> snapshot
53 * The 'snapshot' node contains just enough information to lookup '..' and act
54 * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we
55 * perform an automount of the underlying filesystem and return the
56 * corresponding inode.
58 * All mounts are handled automatically by an user mode helper which invokes
59 * the mount mount procedure. Unmounts are handled by allowing the mount
60 * point to expire so the kernel may automatically unmount it.
62 * The '.zfs', '.zfs/snapshot', and all directories created under
63 * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') all share the same
64 * share the same zfsvfs_t as the head filesystem (what '.zfs' lives under).
66 * File systems mounted on top of the '.zfs/snapshot/<snapname>' paths
67 * (ie: snapshots) are complete ZFS filesystems and have their own unique
68 * zfsvfs_t. However, the fsid reported by these mounts will be the same
69 * as that used by the parent zfsvfs_t to make NFS happy.
72 #include <sys/types.h>
73 #include <sys/param.h>
75 #include <sys/sysmacros.h>
76 #include <sys/pathname.h>
78 #include <sys/zfs_ctldir.h>
79 #include <sys/zfs_ioctl.h>
80 #include <sys/zfs_vfsops.h>
81 #include <sys/zfs_vnops.h>
84 #include <sys/dmu_objset.h>
85 #include <sys/dsl_destroy.h>
86 #include <sys/dsl_deleg.h>
88 #include "zfs_namecheck.h"
91 * Two AVL trees are maintained which contain all currently automounted
92 * snapshots. Every automounted snapshots maps to a single zfs_snapentry_t
95 * - be attached to both trees, and
96 * - be unique, no duplicate entries are allowed.
98 * The zfs_snapshots_by_name tree is indexed by the full dataset name
99 * while the zfs_snapshots_by_objsetid tree is indexed by the unique
100 * objsetid. This allows for fast lookups either by name or objsetid.
102 static avl_tree_t zfs_snapshots_by_name
;
103 static avl_tree_t zfs_snapshots_by_objsetid
;
104 static krwlock_t zfs_snapshot_lock
;
107 * Control Directory Tunables (.zfs)
109 int zfs_expire_snapshot
= ZFSCTL_EXPIRE_SNAPSHOT
;
110 int zfs_admin_snapshot
= 0;
113 char *se_name
; /* full snapshot name */
114 char *se_path
; /* full mount path */
115 spa_t
*se_spa
; /* pool spa */
116 uint64_t se_objsetid
; /* snapshot objset id */
117 struct dentry
*se_root_dentry
; /* snapshot root dentry */
118 taskqid_t se_taskqid
; /* scheduled unmount taskqid */
119 avl_node_t se_node_name
; /* zfs_snapshots_by_name link */
120 avl_node_t se_node_objsetid
; /* zfs_snapshots_by_objsetid link */
121 zfs_refcount_t se_refcount
; /* reference count */
124 static void zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t
*se
, int delay
);
127 * Allocate a new zfs_snapentry_t being careful to make a copy of the
128 * the snapshot name and provided mount point. No reference is taken.
130 static zfs_snapentry_t
*
131 zfsctl_snapshot_alloc(char *full_name
, char *full_path
, spa_t
*spa
,
132 uint64_t objsetid
, struct dentry
*root_dentry
)
136 se
= kmem_zalloc(sizeof (zfs_snapentry_t
), KM_SLEEP
);
138 se
->se_name
= strdup(full_name
);
139 se
->se_path
= strdup(full_path
);
141 se
->se_objsetid
= objsetid
;
142 se
->se_root_dentry
= root_dentry
;
143 se
->se_taskqid
= TASKQID_INVALID
;
145 zfs_refcount_create(&se
->se_refcount
);
151 * Free a zfs_snapentry_t the called must ensure there are no active
155 zfsctl_snapshot_free(zfs_snapentry_t
*se
)
157 zfs_refcount_destroy(&se
->se_refcount
);
158 strfree(se
->se_name
);
159 strfree(se
->se_path
);
161 kmem_free(se
, sizeof (zfs_snapentry_t
));
165 * Hold a reference on the zfs_snapentry_t.
168 zfsctl_snapshot_hold(zfs_snapentry_t
*se
)
170 zfs_refcount_add(&se
->se_refcount
, NULL
);
174 * Release a reference on the zfs_snapentry_t. When the number of
175 * references drops to zero the structure will be freed.
178 zfsctl_snapshot_rele(zfs_snapentry_t
*se
)
180 if (zfs_refcount_remove(&se
->se_refcount
, NULL
) == 0)
181 zfsctl_snapshot_free(se
);
185 * Add a zfs_snapentry_t to both the zfs_snapshots_by_name and
186 * zfs_snapshots_by_objsetid trees. While the zfs_snapentry_t is part
187 * of the trees a reference is held.
190 zfsctl_snapshot_add(zfs_snapentry_t
*se
)
192 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock
));
193 zfs_refcount_add(&se
->se_refcount
, NULL
);
194 avl_add(&zfs_snapshots_by_name
, se
);
195 avl_add(&zfs_snapshots_by_objsetid
, se
);
199 * Remove a zfs_snapentry_t from both the zfs_snapshots_by_name and
200 * zfs_snapshots_by_objsetid trees. Upon removal a reference is dropped,
201 * this can result in the structure being freed if that was the last
202 * remaining reference.
205 zfsctl_snapshot_remove(zfs_snapentry_t
*se
)
207 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock
));
208 avl_remove(&zfs_snapshots_by_name
, se
);
209 avl_remove(&zfs_snapshots_by_objsetid
, se
);
210 zfsctl_snapshot_rele(se
);
214 * Snapshot name comparison function for the zfs_snapshots_by_name.
217 snapentry_compare_by_name(const void *a
, const void *b
)
219 const zfs_snapentry_t
*se_a
= a
;
220 const zfs_snapentry_t
*se_b
= b
;
223 ret
= strcmp(se_a
->se_name
, se_b
->se_name
);
234 * Snapshot name comparison function for the zfs_snapshots_by_objsetid.
237 snapentry_compare_by_objsetid(const void *a
, const void *b
)
239 const zfs_snapentry_t
*se_a
= a
;
240 const zfs_snapentry_t
*se_b
= b
;
242 if (se_a
->se_spa
!= se_b
->se_spa
)
243 return ((ulong_t
)se_a
->se_spa
< (ulong_t
)se_b
->se_spa
? -1 : 1);
245 if (se_a
->se_objsetid
< se_b
->se_objsetid
)
247 else if (se_a
->se_objsetid
> se_b
->se_objsetid
)
254 * Find a zfs_snapentry_t in zfs_snapshots_by_name. If the snapname
255 * is found a pointer to the zfs_snapentry_t is returned and a reference
256 * taken on the structure. The caller is responsible for dropping the
257 * reference with zfsctl_snapshot_rele(). If the snapname is not found
258 * NULL will be returned.
260 static zfs_snapentry_t
*
261 zfsctl_snapshot_find_by_name(char *snapname
)
263 zfs_snapentry_t
*se
, search
;
265 ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock
));
267 search
.se_name
= snapname
;
268 se
= avl_find(&zfs_snapshots_by_name
, &search
, NULL
);
270 zfs_refcount_add(&se
->se_refcount
, NULL
);
276 * Find a zfs_snapentry_t in zfs_snapshots_by_objsetid given the objset id
277 * rather than the snapname. In all other respects it behaves the same
278 * as zfsctl_snapshot_find_by_name().
280 static zfs_snapentry_t
*
281 zfsctl_snapshot_find_by_objsetid(spa_t
*spa
, uint64_t objsetid
)
283 zfs_snapentry_t
*se
, search
;
285 ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock
));
288 search
.se_objsetid
= objsetid
;
289 se
= avl_find(&zfs_snapshots_by_objsetid
, &search
, NULL
);
291 zfs_refcount_add(&se
->se_refcount
, NULL
);
297 * Rename a zfs_snapentry_t in the zfs_snapshots_by_name. The structure is
298 * removed, renamed, and added back to the new correct location in the tree.
301 zfsctl_snapshot_rename(char *old_snapname
, char *new_snapname
)
305 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock
));
307 se
= zfsctl_snapshot_find_by_name(old_snapname
);
309 return (SET_ERROR(ENOENT
));
311 zfsctl_snapshot_remove(se
);
312 strfree(se
->se_name
);
313 se
->se_name
= strdup(new_snapname
);
314 zfsctl_snapshot_add(se
);
315 zfsctl_snapshot_rele(se
);
321 * Delayed task responsible for unmounting an expired automounted snapshot.
324 snapentry_expire(void *data
)
326 zfs_snapentry_t
*se
= (zfs_snapentry_t
*)data
;
327 spa_t
*spa
= se
->se_spa
;
328 uint64_t objsetid
= se
->se_objsetid
;
330 if (zfs_expire_snapshot
<= 0) {
331 zfsctl_snapshot_rele(se
);
335 se
->se_taskqid
= TASKQID_INVALID
;
336 (void) zfsctl_snapshot_unmount(se
->se_name
, MNT_EXPIRE
);
337 zfsctl_snapshot_rele(se
);
340 * Reschedule the unmount if the zfs_snapentry_t wasn't removed.
341 * This can occur when the snapshot is busy.
343 rw_enter(&zfs_snapshot_lock
, RW_READER
);
344 if ((se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
)) != NULL
) {
345 zfsctl_snapshot_unmount_delay_impl(se
, zfs_expire_snapshot
);
346 zfsctl_snapshot_rele(se
);
348 rw_exit(&zfs_snapshot_lock
);
352 * Cancel an automatic unmount of a snapname. This callback is responsible
353 * for dropping the reference on the zfs_snapentry_t which was taken when
357 zfsctl_snapshot_unmount_cancel(zfs_snapentry_t
*se
)
359 if (taskq_cancel_id(system_delay_taskq
, se
->se_taskqid
) == 0) {
360 se
->se_taskqid
= TASKQID_INVALID
;
361 zfsctl_snapshot_rele(se
);
366 * Dispatch the unmount task for delayed handling with a hold protecting it.
369 zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t
*se
, int delay
)
371 ASSERT3S(se
->se_taskqid
, ==, TASKQID_INVALID
);
376 zfsctl_snapshot_hold(se
);
377 se
->se_taskqid
= taskq_dispatch_delay(system_delay_taskq
,
378 snapentry_expire
, se
, TQ_SLEEP
, ddi_get_lbolt() + delay
* HZ
);
382 * Schedule an automatic unmount of objset id to occur in delay seconds from
383 * now. Any previous delayed unmount will be cancelled in favor of the
384 * updated deadline. A reference is taken by zfsctl_snapshot_find_by_name()
385 * and held until the outstanding task is handled or cancelled.
388 zfsctl_snapshot_unmount_delay(spa_t
*spa
, uint64_t objsetid
, int delay
)
393 rw_enter(&zfs_snapshot_lock
, RW_READER
);
394 if ((se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
)) != NULL
) {
395 zfsctl_snapshot_unmount_cancel(se
);
396 zfsctl_snapshot_unmount_delay_impl(se
, delay
);
397 zfsctl_snapshot_rele(se
);
400 rw_exit(&zfs_snapshot_lock
);
406 * Check if snapname is currently mounted. Returned non-zero when mounted
407 * and zero when unmounted.
410 zfsctl_snapshot_ismounted(char *snapname
)
413 boolean_t ismounted
= B_FALSE
;
415 rw_enter(&zfs_snapshot_lock
, RW_READER
);
416 if ((se
= zfsctl_snapshot_find_by_name(snapname
)) != NULL
) {
417 zfsctl_snapshot_rele(se
);
420 rw_exit(&zfs_snapshot_lock
);
426 * Check if the given inode is a part of the virtual .zfs directory.
429 zfsctl_is_node(struct inode
*ip
)
431 return (ITOZ(ip
)->z_is_ctldir
);
435 * Check if the given inode is a .zfs/snapshots/snapname directory.
438 zfsctl_is_snapdir(struct inode
*ip
)
440 return (zfsctl_is_node(ip
) && (ip
->i_ino
<= ZFSCTL_INO_SNAPDIRS
));
444 * Allocate a new inode with the passed id and ops.
446 static struct inode
*
447 zfsctl_inode_alloc(zfsvfs_t
*zfsvfs
, uint64_t id
,
448 const struct file_operations
*fops
, const struct inode_operations
*ops
)
450 inode_timespec_t now
;
454 ip
= new_inode(zfsvfs
->z_sb
);
458 now
= current_time(ip
);
460 ASSERT3P(zp
->z_dirlocks
, ==, NULL
);
461 ASSERT3P(zp
->z_acl_cached
, ==, NULL
);
462 ASSERT3P(zp
->z_xattr_cached
, ==, NULL
);
465 zp
->z_atime_dirty
= 0;
466 zp
->z_zn_prefetch
= 0;
476 zp
->z_is_mapped
= B_FALSE
;
477 zp
->z_is_ctldir
= B_TRUE
;
478 zp
->z_is_sa
= B_FALSE
;
479 zp
->z_is_stale
= B_FALSE
;
480 ip
->i_generation
= 0;
482 ip
->i_mode
= (S_IFDIR
| S_IRWXUGO
);
483 ip
->i_uid
= SUID_TO_KUID(0);
484 ip
->i_gid
= SGID_TO_KGID(0);
485 ip
->i_blkbits
= SPA_MINBLOCKSHIFT
;
491 #if defined(IOP_XATTR)
492 ip
->i_opflags
&= ~IOP_XATTR
;
495 if (insert_inode_locked(ip
)) {
496 unlock_new_inode(ip
);
501 mutex_enter(&zfsvfs
->z_znodes_lock
);
502 list_insert_tail(&zfsvfs
->z_all_znodes
, zp
);
503 zfsvfs
->z_nr_znodes
++;
505 mutex_exit(&zfsvfs
->z_znodes_lock
);
507 unlock_new_inode(ip
);
513 * Lookup the inode with given id, it will be allocated if needed.
515 static struct inode
*
516 zfsctl_inode_lookup(zfsvfs_t
*zfsvfs
, uint64_t id
,
517 const struct file_operations
*fops
, const struct inode_operations
*ops
)
519 struct inode
*ip
= NULL
;
522 ip
= ilookup(zfsvfs
->z_sb
, (unsigned long)id
);
526 /* May fail due to concurrent zfsctl_inode_alloc() */
527 ip
= zfsctl_inode_alloc(zfsvfs
, id
, fops
, ops
);
534 * Create the '.zfs' directory. This directory is cached as part of the VFS
535 * structure. This results in a hold on the zfsvfs_t. The code in zfs_umount()
536 * therefore checks against a vfs_count of 2 instead of 1. This reference
537 * is removed when the ctldir is destroyed in the unmount. All other entities
538 * under the '.zfs' directory are created dynamically as needed.
540 * Because the dynamically created '.zfs' directory entries assume the use
541 * of 64-bit inode numbers this support must be disabled on 32-bit systems.
544 zfsctl_create(zfsvfs_t
*zfsvfs
)
546 ASSERT(zfsvfs
->z_ctldir
== NULL
);
548 zfsvfs
->z_ctldir
= zfsctl_inode_alloc(zfsvfs
, ZFSCTL_INO_ROOT
,
549 &zpl_fops_root
, &zpl_ops_root
);
550 if (zfsvfs
->z_ctldir
== NULL
)
551 return (SET_ERROR(ENOENT
));
557 * Destroy the '.zfs' directory or remove a snapshot from zfs_snapshots_by_name.
558 * Only called when the filesystem is unmounted.
561 zfsctl_destroy(zfsvfs_t
*zfsvfs
)
563 if (zfsvfs
->z_issnap
) {
565 spa_t
*spa
= zfsvfs
->z_os
->os_spa
;
566 uint64_t objsetid
= dmu_objset_id(zfsvfs
->z_os
);
568 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
569 se
= zfsctl_snapshot_find_by_objsetid(spa
, objsetid
);
571 zfsctl_snapshot_remove(se
);
572 rw_exit(&zfs_snapshot_lock
);
574 zfsctl_snapshot_unmount_cancel(se
);
575 zfsctl_snapshot_rele(se
);
577 } else if (zfsvfs
->z_ctldir
) {
578 iput(zfsvfs
->z_ctldir
);
579 zfsvfs
->z_ctldir
= NULL
;
584 * Given a root znode, retrieve the associated .zfs directory.
585 * Add a hold to the vnode and return it.
588 zfsctl_root(znode_t
*zp
)
590 ASSERT(zfs_has_ctldir(zp
));
591 igrab(ZTOZSB(zp
)->z_ctldir
);
592 return (ZTOZSB(zp
)->z_ctldir
);
596 * Generate a long fid to indicate a snapdir. We encode whether snapdir is
597 * already monunted in gen field. We do this because nfsd lookup will not
598 * trigger automount. Next time the nfsd does fh_to_dentry, we will notice
599 * this and do automount and return ESTALE to force nfsd revalidate and follow
603 zfsctl_snapdir_fid(struct inode
*ip
, fid_t
*fidp
)
605 zfid_short_t
*zfid
= (zfid_short_t
*)fidp
;
606 zfid_long_t
*zlfid
= (zfid_long_t
*)fidp
;
611 struct dentry
*dentry
;
613 if (fidp
->fid_len
< LONG_FID_LEN
) {
614 fidp
->fid_len
= LONG_FID_LEN
;
615 return (SET_ERROR(ENOSPC
));
619 objsetid
= ZFSCTL_INO_SNAPDIRS
- ip
->i_ino
;
620 zfid
->zf_len
= LONG_FID_LEN
;
622 dentry
= d_obtain_alias(igrab(ip
));
623 if (!IS_ERR(dentry
)) {
624 gen
= !!d_mountpoint(dentry
);
628 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
629 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
631 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
632 zfid
->zf_gen
[i
] = (uint8_t)(gen
>> (8 * i
));
634 for (i
= 0; i
< sizeof (zlfid
->zf_setid
); i
++)
635 zlfid
->zf_setid
[i
] = (uint8_t)(objsetid
>> (8 * i
));
637 for (i
= 0; i
< sizeof (zlfid
->zf_setgen
); i
++)
638 zlfid
->zf_setgen
[i
] = 0;
644 * Generate an appropriate fid for an entry in the .zfs directory.
647 zfsctl_fid(struct inode
*ip
, fid_t
*fidp
)
649 znode_t
*zp
= ITOZ(ip
);
650 zfsvfs_t
*zfsvfs
= ITOZSB(ip
);
651 uint64_t object
= zp
->z_id
;
657 if (zfsctl_is_snapdir(ip
)) {
659 return (zfsctl_snapdir_fid(ip
, fidp
));
662 if (fidp
->fid_len
< SHORT_FID_LEN
) {
663 fidp
->fid_len
= SHORT_FID_LEN
;
665 return (SET_ERROR(ENOSPC
));
668 zfid
= (zfid_short_t
*)fidp
;
670 zfid
->zf_len
= SHORT_FID_LEN
;
672 for (i
= 0; i
< sizeof (zfid
->zf_object
); i
++)
673 zfid
->zf_object
[i
] = (uint8_t)(object
>> (8 * i
));
675 /* .zfs znodes always have a generation number of 0 */
676 for (i
= 0; i
< sizeof (zfid
->zf_gen
); i
++)
684 * Construct a full dataset name in full_name: "pool/dataset@snap_name"
687 zfsctl_snapshot_name(zfsvfs_t
*zfsvfs
, const char *snap_name
, int len
,
690 objset_t
*os
= zfsvfs
->z_os
;
692 if (zfs_component_namecheck(snap_name
, NULL
, NULL
) != 0)
693 return (SET_ERROR(EILSEQ
));
695 dmu_objset_name(os
, full_name
);
696 if ((strlen(full_name
) + 1 + strlen(snap_name
)) >= len
)
697 return (SET_ERROR(ENAMETOOLONG
));
699 (void) strcat(full_name
, "@");
700 (void) strcat(full_name
, snap_name
);
706 * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/"
709 zfsctl_snapshot_path(struct path
*path
, int len
, char *full_path
)
711 char *path_buffer
, *path_ptr
;
712 int path_len
, error
= 0;
714 path_buffer
= kmem_alloc(len
, KM_SLEEP
);
716 path_ptr
= d_path(path
, path_buffer
, len
);
717 if (IS_ERR(path_ptr
)) {
718 error
= -PTR_ERR(path_ptr
);
722 path_len
= path_buffer
+ len
- 1 - path_ptr
;
723 if (path_len
> len
) {
724 error
= SET_ERROR(EFAULT
);
728 memcpy(full_path
, path_ptr
, path_len
);
729 full_path
[path_len
] = '\0';
731 kmem_free(path_buffer
, len
);
737 * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/"
740 zfsctl_snapshot_path_objset(zfsvfs_t
*zfsvfs
, uint64_t objsetid
,
741 int path_len
, char *full_path
)
743 objset_t
*os
= zfsvfs
->z_os
;
744 fstrans_cookie_t cookie
;
746 boolean_t case_conflict
;
747 uint64_t id
, pos
= 0;
750 if (zfsvfs
->z_vfs
->vfs_mntpoint
== NULL
)
751 return (SET_ERROR(ENOENT
));
753 cookie
= spl_fstrans_mark();
754 snapname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
757 dsl_pool_config_enter(dmu_objset_pool(os
), FTAG
);
758 error
= dmu_snapshot_list_next(zfsvfs
->z_os
,
759 ZFS_MAX_DATASET_NAME_LEN
, snapname
, &id
, &pos
,
761 dsl_pool_config_exit(dmu_objset_pool(os
), FTAG
);
769 memset(full_path
, 0, path_len
);
770 snprintf(full_path
, path_len
- 1, "%s/.zfs/snapshot/%s",
771 zfsvfs
->z_vfs
->vfs_mntpoint
, snapname
);
773 kmem_free(snapname
, ZFS_MAX_DATASET_NAME_LEN
);
774 spl_fstrans_unmark(cookie
);
780 * Special case the handling of "..".
783 zfsctl_root_lookup(struct inode
*dip
, char *name
, struct inode
**ipp
,
784 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
786 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
791 if (strcmp(name
, "..") == 0) {
792 *ipp
= dip
->i_sb
->s_root
->d_inode
;
793 } else if (strcmp(name
, ZFS_SNAPDIR_NAME
) == 0) {
794 *ipp
= zfsctl_inode_lookup(zfsvfs
, ZFSCTL_INO_SNAPDIR
,
795 &zpl_fops_snapdir
, &zpl_ops_snapdir
);
796 } else if (strcmp(name
, ZFS_SHAREDIR_NAME
) == 0) {
797 *ipp
= zfsctl_inode_lookup(zfsvfs
, ZFSCTL_INO_SHARES
,
798 &zpl_fops_shares
, &zpl_ops_shares
);
804 error
= SET_ERROR(ENOENT
);
812 * Lookup entry point for the 'snapshot' directory. Try to open the
813 * snapshot if it exist, creating the pseudo filesystem inode as necessary.
814 * Perform a mount of the associated dataset on top of the inode.
817 zfsctl_snapdir_lookup(struct inode
*dip
, char *name
, struct inode
**ipp
,
818 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
820 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
826 error
= dmu_snapshot_lookup(zfsvfs
->z_os
, name
, &id
);
832 *ipp
= zfsctl_inode_lookup(zfsvfs
, ZFSCTL_INO_SNAPDIRS
- id
,
833 &simple_dir_operations
, &simple_dir_inode_operations
);
835 error
= SET_ERROR(ENOENT
);
843 * Renaming a directory under '.zfs/snapshot' will automatically trigger
844 * a rename of the snapshot to the new given name. The rename is confined
845 * to the '.zfs/snapshot' directory snapshots cannot be moved elsewhere.
848 zfsctl_snapdir_rename(struct inode
*sdip
, char *snm
,
849 struct inode
*tdip
, char *tnm
, cred_t
*cr
, int flags
)
851 zfsvfs_t
*zfsvfs
= ITOZSB(sdip
);
852 char *to
, *from
, *real
, *fsname
;
855 if (!zfs_admin_snapshot
)
856 return (SET_ERROR(EACCES
));
860 to
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
861 from
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
862 real
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
863 fsname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
865 if (zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
) {
866 error
= dmu_snapshot_realname(zfsvfs
->z_os
, snm
, real
,
867 ZFS_MAX_DATASET_NAME_LEN
, NULL
);
870 } else if (error
!= ENOTSUP
) {
875 dmu_objset_name(zfsvfs
->z_os
, fsname
);
877 error
= zfsctl_snapshot_name(ITOZSB(sdip
), snm
,
878 ZFS_MAX_DATASET_NAME_LEN
, from
);
880 error
= zfsctl_snapshot_name(ITOZSB(tdip
), tnm
,
881 ZFS_MAX_DATASET_NAME_LEN
, to
);
883 error
= zfs_secpolicy_rename_perms(from
, to
, cr
);
888 * Cannot move snapshots out of the snapdir.
891 error
= SET_ERROR(EINVAL
);
896 * No-op when names are identical.
898 if (strcmp(snm
, tnm
) == 0) {
903 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
905 error
= dsl_dataset_rename_snapshot(fsname
, snm
, tnm
, B_FALSE
);
907 (void) zfsctl_snapshot_rename(snm
, tnm
);
909 rw_exit(&zfs_snapshot_lock
);
911 kmem_free(from
, ZFS_MAX_DATASET_NAME_LEN
);
912 kmem_free(to
, ZFS_MAX_DATASET_NAME_LEN
);
913 kmem_free(real
, ZFS_MAX_DATASET_NAME_LEN
);
914 kmem_free(fsname
, ZFS_MAX_DATASET_NAME_LEN
);
922 * Removing a directory under '.zfs/snapshot' will automatically trigger
923 * the removal of the snapshot with the given name.
926 zfsctl_snapdir_remove(struct inode
*dip
, char *name
, cred_t
*cr
, int flags
)
928 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
929 char *snapname
, *real
;
932 if (!zfs_admin_snapshot
)
933 return (SET_ERROR(EACCES
));
937 snapname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
938 real
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
940 if (zfsvfs
->z_case
== ZFS_CASE_INSENSITIVE
) {
941 error
= dmu_snapshot_realname(zfsvfs
->z_os
, name
, real
,
942 ZFS_MAX_DATASET_NAME_LEN
, NULL
);
945 } else if (error
!= ENOTSUP
) {
950 error
= zfsctl_snapshot_name(ITOZSB(dip
), name
,
951 ZFS_MAX_DATASET_NAME_LEN
, snapname
);
953 error
= zfs_secpolicy_destroy_perms(snapname
, cr
);
957 error
= zfsctl_snapshot_unmount(snapname
, MNT_FORCE
);
958 if ((error
== 0) || (error
== ENOENT
))
959 error
= dsl_destroy_snapshot(snapname
, B_FALSE
);
961 kmem_free(snapname
, ZFS_MAX_DATASET_NAME_LEN
);
962 kmem_free(real
, ZFS_MAX_DATASET_NAME_LEN
);
970 * Creating a directory under '.zfs/snapshot' will automatically trigger
971 * the creation of a new snapshot with the given name.
974 zfsctl_snapdir_mkdir(struct inode
*dip
, char *dirname
, vattr_t
*vap
,
975 struct inode
**ipp
, cred_t
*cr
, int flags
)
977 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
981 if (!zfs_admin_snapshot
)
982 return (SET_ERROR(EACCES
));
984 dsname
= kmem_alloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
986 if (zfs_component_namecheck(dirname
, NULL
, NULL
) != 0) {
987 error
= SET_ERROR(EILSEQ
);
991 dmu_objset_name(zfsvfs
->z_os
, dsname
);
993 error
= zfs_secpolicy_snapshot_perms(dsname
, cr
);
998 error
= dmu_objset_snapshot_one(dsname
, dirname
);
1002 error
= zfsctl_snapdir_lookup(dip
, dirname
, ipp
,
1006 kmem_free(dsname
, ZFS_MAX_DATASET_NAME_LEN
);
1012 * Attempt to unmount a snapshot by making a call to user space.
1013 * There is no assurance that this can or will succeed, is just a
1014 * best effort. In the case where it does fail, perhaps because
1015 * it's in use, the unmount will fail harmlessly.
1018 zfsctl_snapshot_unmount(char *snapname
, int flags
)
1020 char *argv
[] = { "/usr/bin/env", "umount", "-t", "zfs", "-n", 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
);
1029 return (SET_ERROR(ENOENT
));
1031 rw_exit(&zfs_snapshot_lock
);
1033 if (flags
& MNT_FORCE
)
1035 argv
[5] = se
->se_path
;
1036 dprintf("unmount; path=%s\n", se
->se_path
);
1037 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) */
1055 zfsctl_snapshot_mount(struct path
*path
, int flags
)
1057 struct dentry
*dentry
= path
->dentry
;
1058 struct inode
*ip
= dentry
->d_inode
;
1060 zfsvfs_t
*snap_zfsvfs
;
1061 zfs_snapentry_t
*se
;
1062 char *full_name
, *full_path
;
1063 char *argv
[] = { "/usr/bin/env", "mount", "-t", "zfs", "-n", NULL
, NULL
,
1065 char *envp
[] = { NULL
};
1070 return (SET_ERROR(EISDIR
));
1072 zfsvfs
= ITOZSB(ip
);
1075 full_name
= kmem_zalloc(ZFS_MAX_DATASET_NAME_LEN
, KM_SLEEP
);
1076 full_path
= kmem_zalloc(MAXPATHLEN
, KM_SLEEP
);
1078 error
= zfsctl_snapshot_name(zfsvfs
, dname(dentry
),
1079 ZFS_MAX_DATASET_NAME_LEN
, full_name
);
1083 error
= zfsctl_snapshot_path(path
, MAXPATHLEN
, full_path
);
1088 * Multiple concurrent automounts of a snapshot are never allowed.
1089 * The snapshot may be manually mounted as many times as desired.
1091 if (zfsctl_snapshot_ismounted(full_name
)) {
1097 * Attempt to mount the snapshot from user space. Normally this
1098 * would be done using the vfs_kern_mount() function, however that
1099 * function is marked GPL-only and cannot be used. On error we
1100 * careful to log the real error to the console and return EISDIR
1101 * to safely abort the automount. This should be very rare.
1103 * If the user mode helper happens to return EBUSY, a concurrent
1104 * mount is already in progress in which case the error is ignored.
1105 * Take note that if the program was executed successfully the return
1106 * value from call_usermodehelper() will be (exitcode << 8 + signal).
1108 dprintf("mount; name=%s path=%s\n", full_name
, full_path
);
1109 argv
[5] = full_name
;
1110 argv
[6] = full_path
;
1111 error
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
1113 if (!(error
& MOUNT_BUSY
<< 8)) {
1114 cmn_err(CE_WARN
, "Unable to automount %s/%s: %d",
1115 full_path
, full_name
, error
);
1116 error
= SET_ERROR(EISDIR
);
1119 * EBUSY, this could mean a concurrent mount, or the
1120 * snapshot has already been mounted at completely
1121 * different place. We return 0 so VFS will retry. For
1122 * the latter case the VFS will retry several times
1123 * and return ELOOP, which is probably not a very good
1132 * Follow down in to the mounted snapshot and set MNT_SHRINKABLE
1133 * to identify this as an automounted filesystem.
1137 if (zpl_follow_down_one(&spath
)) {
1138 snap_zfsvfs
= ITOZSB(spath
.dentry
->d_inode
);
1139 snap_zfsvfs
->z_parent
= zfsvfs
;
1140 dentry
= spath
.dentry
;
1141 spath
.mnt
->mnt_flags
|= MNT_SHRINKABLE
;
1143 rw_enter(&zfs_snapshot_lock
, RW_WRITER
);
1144 se
= zfsctl_snapshot_alloc(full_name
, full_path
,
1145 snap_zfsvfs
->z_os
->os_spa
, dmu_objset_id(snap_zfsvfs
->z_os
),
1147 zfsctl_snapshot_add(se
);
1148 zfsctl_snapshot_unmount_delay_impl(se
, zfs_expire_snapshot
);
1149 rw_exit(&zfs_snapshot_lock
);
1153 kmem_free(full_name
, ZFS_MAX_DATASET_NAME_LEN
);
1154 kmem_free(full_path
, MAXPATHLEN
);
1162 * Get the snapdir inode from fid
1165 zfsctl_snapdir_vget(struct super_block
*sb
, uint64_t objsetid
, int gen
,
1171 struct dentry
*dentry
;
1173 mnt
= kmem_alloc(MAXPATHLEN
, KM_SLEEP
);
1175 error
= zfsctl_snapshot_path_objset(sb
->s_fs_info
, objsetid
,
1180 /* Trigger automount */
1181 error
= -kern_path(mnt
, LOOKUP_FOLLOW
|LOOKUP_DIRECTORY
, &path
);
1187 * Get the snapdir inode. Note, we don't want to use the above
1188 * path because it contains the root of the snapshot rather
1191 *ipp
= ilookup(sb
, ZFSCTL_INO_SNAPDIRS
- objsetid
);
1193 error
= SET_ERROR(ENOENT
);
1197 /* check gen, see zfsctl_snapdir_fid */
1198 dentry
= d_obtain_alias(igrab(*ipp
));
1199 if (gen
!= (!IS_ERR(dentry
) && d_mountpoint(dentry
))) {
1202 error
= SET_ERROR(ENOENT
);
1204 if (!IS_ERR(dentry
))
1207 kmem_free(mnt
, MAXPATHLEN
);
1212 zfsctl_shares_lookup(struct inode
*dip
, char *name
, struct inode
**ipp
,
1213 int flags
, cred_t
*cr
, int *direntflags
, pathname_t
*realpnp
)
1215 zfsvfs_t
*zfsvfs
= ITOZSB(dip
);
1222 if (zfsvfs
->z_shares_dir
== 0) {
1224 return (SET_ERROR(ENOTSUP
));
1227 if ((error
= zfs_zget(zfsvfs
, zfsvfs
->z_shares_dir
, &dzp
)) == 0) {
1228 error
= zfs_lookup(ZTOI(dzp
), name
, &ip
, 0, cr
, NULL
, NULL
);
1238 * Initialize the various pieces we'll need to create and manipulate .zfs
1239 * directories. Currently this is unused but available.
1244 avl_create(&zfs_snapshots_by_name
, snapentry_compare_by_name
,
1245 sizeof (zfs_snapentry_t
), offsetof(zfs_snapentry_t
,
1247 avl_create(&zfs_snapshots_by_objsetid
, snapentry_compare_by_objsetid
,
1248 sizeof (zfs_snapentry_t
), offsetof(zfs_snapentry_t
,
1250 rw_init(&zfs_snapshot_lock
, NULL
, RW_DEFAULT
, NULL
);
1254 * Cleanup the various pieces we needed for .zfs directories. In particular
1255 * ensure the expiry timer is canceled safely.
1260 avl_destroy(&zfs_snapshots_by_name
);
1261 avl_destroy(&zfs_snapshots_by_objsetid
);
1262 rw_destroy(&zfs_snapshot_lock
);
1265 module_param(zfs_admin_snapshot
, int, 0644);
1266 MODULE_PARM_DESC(zfs_admin_snapshot
, "Enable mkdir/rmdir/mv in .zfs/snapshot");
1268 module_param(zfs_expire_snapshot
, int, 0644);
1269 MODULE_PARM_DESC(zfs_expire_snapshot
, "Seconds to expire .zfs/snapshot");