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.
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 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 taskqid_t se_taskqid; /* scheduled unmount taskqid */
122 avl_node_t se_node_name; /* zfs_snapshots_by_name link */
123 avl_node_t se_node_objsetid; /* zfs_snapshots_by_objsetid link */
124 zfs_refcount_t se_refcount; /* reference count */
127 static void zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t *se, int delay);
130 * Allocate a new zfs_snapentry_t being careful to make a copy of the
131 * the snapshot name and provided mount point. No reference is taken.
133 static zfs_snapentry_t *
134 zfsctl_snapshot_alloc(char *full_name, char *full_path, spa_t *spa,
135 uint64_t objsetid, struct dentry *root_dentry)
139 se = kmem_zalloc(sizeof (zfs_snapentry_t), KM_SLEEP);
141 se->se_name = kmem_strdup(full_name);
142 se->se_path = kmem_strdup(full_path);
144 se->se_objsetid = objsetid;
145 se->se_root_dentry = root_dentry;
146 se->se_taskqid = TASKQID_INVALID;
148 zfs_refcount_create(&se->se_refcount);
154 * Free a zfs_snapentry_t the caller must ensure there are no active
158 zfsctl_snapshot_free(zfs_snapentry_t *se)
160 zfs_refcount_destroy(&se->se_refcount);
161 kmem_strfree(se->se_name);
162 kmem_strfree(se->se_path);
164 kmem_free(se, sizeof (zfs_snapentry_t));
168 * Hold a reference on the zfs_snapentry_t.
171 zfsctl_snapshot_hold(zfs_snapentry_t *se)
173 zfs_refcount_add(&se->se_refcount, NULL);
177 * Release a reference on the zfs_snapentry_t. When the number of
178 * references drops to zero the structure will be freed.
181 zfsctl_snapshot_rele(zfs_snapentry_t *se)
183 if (zfs_refcount_remove(&se->se_refcount, NULL) == 0)
184 zfsctl_snapshot_free(se);
188 * Add a zfs_snapentry_t to both the zfs_snapshots_by_name and
189 * zfs_snapshots_by_objsetid trees. While the zfs_snapentry_t is part
190 * of the trees a reference is held.
193 zfsctl_snapshot_add(zfs_snapentry_t *se)
195 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock));
196 zfsctl_snapshot_hold(se);
197 avl_add(&zfs_snapshots_by_name, se);
198 avl_add(&zfs_snapshots_by_objsetid, se);
202 * Remove a zfs_snapentry_t from both the zfs_snapshots_by_name and
203 * zfs_snapshots_by_objsetid trees. Upon removal a reference is dropped,
204 * this can result in the structure being freed if that was the last
205 * remaining reference.
208 zfsctl_snapshot_remove(zfs_snapentry_t *se)
210 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock));
211 avl_remove(&zfs_snapshots_by_name, se);
212 avl_remove(&zfs_snapshots_by_objsetid, se);
213 zfsctl_snapshot_rele(se);
217 * Snapshot name comparison function for the zfs_snapshots_by_name.
220 snapentry_compare_by_name(const void *a, const void *b)
222 const zfs_snapentry_t *se_a = a;
223 const zfs_snapentry_t *se_b = b;
226 ret = strcmp(se_a->se_name, se_b->se_name);
237 * Snapshot name comparison function for the zfs_snapshots_by_objsetid.
240 snapentry_compare_by_objsetid(const void *a, const void *b)
242 const zfs_snapentry_t *se_a = a;
243 const zfs_snapentry_t *se_b = b;
245 if (se_a->se_spa != se_b->se_spa)
246 return ((ulong_t)se_a->se_spa < (ulong_t)se_b->se_spa ? -1 : 1);
248 if (se_a->se_objsetid < se_b->se_objsetid)
250 else if (se_a->se_objsetid > se_b->se_objsetid)
257 * Find a zfs_snapentry_t in zfs_snapshots_by_name. If the snapname
258 * is found a pointer to the zfs_snapentry_t is returned and a reference
259 * taken on the structure. The caller is responsible for dropping the
260 * reference with zfsctl_snapshot_rele(). If the snapname is not found
261 * NULL will be returned.
263 static zfs_snapentry_t *
264 zfsctl_snapshot_find_by_name(char *snapname)
266 zfs_snapentry_t *se, search;
268 ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock));
270 search.se_name = snapname;
271 se = avl_find(&zfs_snapshots_by_name, &search, NULL);
273 zfsctl_snapshot_hold(se);
279 * Find a zfs_snapentry_t in zfs_snapshots_by_objsetid given the objset id
280 * rather than the snapname. In all other respects it behaves the same
281 * as zfsctl_snapshot_find_by_name().
283 static zfs_snapentry_t *
284 zfsctl_snapshot_find_by_objsetid(spa_t *spa, uint64_t objsetid)
286 zfs_snapentry_t *se, search;
288 ASSERT(RW_LOCK_HELD(&zfs_snapshot_lock));
291 search.se_objsetid = objsetid;
292 se = avl_find(&zfs_snapshots_by_objsetid, &search, NULL);
294 zfsctl_snapshot_hold(se);
300 * Rename a zfs_snapentry_t in the zfs_snapshots_by_name. The structure is
301 * removed, renamed, and added back to the new correct location in the tree.
304 zfsctl_snapshot_rename(char *old_snapname, char *new_snapname)
308 ASSERT(RW_WRITE_HELD(&zfs_snapshot_lock));
310 se = zfsctl_snapshot_find_by_name(old_snapname);
312 return (SET_ERROR(ENOENT));
314 zfsctl_snapshot_remove(se);
315 kmem_strfree(se->se_name);
316 se->se_name = kmem_strdup(new_snapname);
317 zfsctl_snapshot_add(se);
318 zfsctl_snapshot_rele(se);
324 * Delayed task responsible for unmounting an expired automounted snapshot.
327 snapentry_expire(void *data)
329 zfs_snapentry_t *se = (zfs_snapentry_t *)data;
330 spa_t *spa = se->se_spa;
331 uint64_t objsetid = se->se_objsetid;
333 if (zfs_expire_snapshot <= 0) {
334 zfsctl_snapshot_rele(se);
338 se->se_taskqid = TASKQID_INVALID;
339 (void) zfsctl_snapshot_unmount(se->se_name, MNT_EXPIRE);
340 zfsctl_snapshot_rele(se);
343 * Reschedule the unmount if the zfs_snapentry_t wasn't removed.
344 * This can occur when the snapshot is busy.
346 rw_enter(&zfs_snapshot_lock, RW_READER);
347 if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) {
348 zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot);
349 zfsctl_snapshot_rele(se);
351 rw_exit(&zfs_snapshot_lock);
355 * Cancel an automatic unmount of a snapname. This callback is responsible
356 * for dropping the reference on the zfs_snapentry_t which was taken when
360 zfsctl_snapshot_unmount_cancel(zfs_snapentry_t *se)
362 if (taskq_cancel_id(system_delay_taskq, se->se_taskqid) == 0) {
363 se->se_taskqid = TASKQID_INVALID;
364 zfsctl_snapshot_rele(se);
369 * Dispatch the unmount task for delayed handling with a hold protecting it.
372 zfsctl_snapshot_unmount_delay_impl(zfs_snapentry_t *se, int delay)
374 ASSERT3S(se->se_taskqid, ==, TASKQID_INVALID);
379 zfsctl_snapshot_hold(se);
380 se->se_taskqid = taskq_dispatch_delay(system_delay_taskq,
381 snapentry_expire, se, TQ_SLEEP, ddi_get_lbolt() + delay * HZ);
385 * Schedule an automatic unmount of objset id to occur in delay seconds from
386 * now. Any previous delayed unmount will be cancelled in favor of the
387 * updated deadline. A reference is taken by zfsctl_snapshot_find_by_name()
388 * and held until the outstanding task is handled or cancelled.
391 zfsctl_snapshot_unmount_delay(spa_t *spa, uint64_t objsetid, int delay)
396 rw_enter(&zfs_snapshot_lock, RW_READER);
397 if ((se = zfsctl_snapshot_find_by_objsetid(spa, objsetid)) != NULL) {
398 zfsctl_snapshot_unmount_cancel(se);
399 zfsctl_snapshot_unmount_delay_impl(se, delay);
400 zfsctl_snapshot_rele(se);
403 rw_exit(&zfs_snapshot_lock);
409 * Check if snapname is currently mounted. Returned non-zero when mounted
410 * and zero when unmounted.
413 zfsctl_snapshot_ismounted(char *snapname)
416 boolean_t ismounted = B_FALSE;
418 rw_enter(&zfs_snapshot_lock, RW_READER);
419 if ((se = zfsctl_snapshot_find_by_name(snapname)) != NULL) {
420 zfsctl_snapshot_rele(se);
423 rw_exit(&zfs_snapshot_lock);
429 * Check if the given inode is a part of the virtual .zfs directory.
432 zfsctl_is_node(struct inode *ip)
434 return (ITOZ(ip)->z_is_ctldir);
438 * Check if the given inode is a .zfs/snapshots/snapname directory.
441 zfsctl_is_snapdir(struct inode *ip)
443 return (zfsctl_is_node(ip) && (ip->i_ino <= ZFSCTL_INO_SNAPDIRS));
447 * Allocate a new inode with the passed id and ops.
449 static struct inode *
450 zfsctl_inode_alloc(zfsvfs_t *zfsvfs, uint64_t id,
451 const struct file_operations *fops, const struct inode_operations *ops)
453 inode_timespec_t now;
457 ip = new_inode(zfsvfs->z_sb);
461 now = current_time(ip);
463 ASSERT3P(zp->z_dirlocks, ==, NULL);
464 ASSERT3P(zp->z_acl_cached, ==, NULL);
465 ASSERT3P(zp->z_xattr_cached, ==, NULL);
467 zp->z_unlinked = B_FALSE;
468 zp->z_atime_dirty = B_FALSE;
469 zp->z_zn_prefetch = B_FALSE;
470 zp->z_moved = B_FALSE;
471 zp->z_is_sa = B_FALSE;
472 zp->z_is_mapped = B_FALSE;
473 zp->z_is_ctldir = B_TRUE;
474 zp->z_is_stale = B_FALSE;
483 ip->i_generation = 0;
485 ip->i_mode = (S_IFDIR | S_IRWXUGO);
486 ip->i_uid = SUID_TO_KUID(0);
487 ip->i_gid = SGID_TO_KGID(0);
488 ip->i_blkbits = SPA_MINBLOCKSHIFT;
494 #if defined(IOP_XATTR)
495 ip->i_opflags &= ~IOP_XATTR;
498 if (insert_inode_locked(ip)) {
499 unlock_new_inode(ip);
504 mutex_enter(&zfsvfs->z_znodes_lock);
505 list_insert_tail(&zfsvfs->z_all_znodes, zp);
506 zfsvfs->z_nr_znodes++;
508 mutex_exit(&zfsvfs->z_znodes_lock);
510 unlock_new_inode(ip);
516 * Lookup the inode with given id, it will be allocated if needed.
518 static struct inode *
519 zfsctl_inode_lookup(zfsvfs_t *zfsvfs, uint64_t id,
520 const struct file_operations *fops, const struct inode_operations *ops)
522 struct inode *ip = NULL;
525 ip = ilookup(zfsvfs->z_sb, (unsigned long)id);
529 /* May fail due to concurrent zfsctl_inode_alloc() */
530 ip = zfsctl_inode_alloc(zfsvfs, id, fops, ops);
537 * Create the '.zfs' directory. This directory is cached as part of the VFS
538 * structure. This results in a hold on the zfsvfs_t. The code in zfs_umount()
539 * therefore checks against a vfs_count of 2 instead of 1. This reference
540 * is removed when the ctldir is destroyed in the unmount. All other entities
541 * under the '.zfs' directory are created dynamically as needed.
543 * Because the dynamically created '.zfs' directory entries assume the use
544 * of 64-bit inode numbers this support must be disabled on 32-bit systems.
547 zfsctl_create(zfsvfs_t *zfsvfs)
549 ASSERT(zfsvfs->z_ctldir == NULL);
551 zfsvfs->z_ctldir = zfsctl_inode_alloc(zfsvfs, ZFSCTL_INO_ROOT,
552 &zpl_fops_root, &zpl_ops_root);
553 if (zfsvfs->z_ctldir == NULL)
554 return (SET_ERROR(ENOENT));
560 * Destroy the '.zfs' directory or remove a snapshot from zfs_snapshots_by_name.
561 * Only called when the filesystem is unmounted.
564 zfsctl_destroy(zfsvfs_t *zfsvfs)
566 if (zfsvfs->z_issnap) {
568 spa_t *spa = zfsvfs->z_os->os_spa;
569 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os);
571 rw_enter(&zfs_snapshot_lock, RW_WRITER);
572 se = zfsctl_snapshot_find_by_objsetid(spa, objsetid);
574 zfsctl_snapshot_remove(se);
575 rw_exit(&zfs_snapshot_lock);
577 zfsctl_snapshot_unmount_cancel(se);
578 zfsctl_snapshot_rele(se);
580 } else if (zfsvfs->z_ctldir) {
581 iput(zfsvfs->z_ctldir);
582 zfsvfs->z_ctldir = NULL;
587 * Given a root znode, retrieve the associated .zfs directory.
588 * Add a hold to the vnode and return it.
591 zfsctl_root(znode_t *zp)
593 ASSERT(zfs_has_ctldir(zp));
594 igrab(ZTOZSB(zp)->z_ctldir);
595 return (ZTOZSB(zp)->z_ctldir);
599 * Generate a long fid to indicate a snapdir. We encode whether snapdir is
600 * already mounted in gen field. We do this because nfsd lookup will not
601 * trigger automount. Next time the nfsd does fh_to_dentry, we will notice
602 * this and do automount and return ESTALE to force nfsd revalidate and follow
606 zfsctl_snapdir_fid(struct inode *ip, fid_t *fidp)
608 zfid_short_t *zfid = (zfid_short_t *)fidp;
609 zfid_long_t *zlfid = (zfid_long_t *)fidp;
614 struct dentry *dentry;
616 if (fidp->fid_len < LONG_FID_LEN) {
617 fidp->fid_len = LONG_FID_LEN;
618 return (SET_ERROR(ENOSPC));
622 objsetid = ZFSCTL_INO_SNAPDIRS - ip->i_ino;
623 zfid->zf_len = LONG_FID_LEN;
625 dentry = d_obtain_alias(igrab(ip));
626 if (!IS_ERR(dentry)) {
627 gen = !!d_mountpoint(dentry);
631 for (i = 0; i < sizeof (zfid->zf_object); i++)
632 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
634 for (i = 0; i < sizeof (zfid->zf_gen); i++)
635 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i));
637 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
638 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i));
640 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
641 zlfid->zf_setgen[i] = 0;
647 * Generate an appropriate fid for an entry in the .zfs directory.
650 zfsctl_fid(struct inode *ip, fid_t *fidp)
652 znode_t *zp = ITOZ(ip);
653 zfsvfs_t *zfsvfs = ITOZSB(ip);
654 uint64_t object = zp->z_id;
660 if (zfsctl_is_snapdir(ip)) {
662 return (zfsctl_snapdir_fid(ip, fidp));
665 if (fidp->fid_len < SHORT_FID_LEN) {
666 fidp->fid_len = SHORT_FID_LEN;
668 return (SET_ERROR(ENOSPC));
671 zfid = (zfid_short_t *)fidp;
673 zfid->zf_len = SHORT_FID_LEN;
675 for (i = 0; i < sizeof (zfid->zf_object); i++)
676 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
678 /* .zfs znodes always have a generation number of 0 */
679 for (i = 0; i < sizeof (zfid->zf_gen); i++)
687 * Construct a full dataset name in full_name: "pool/dataset@snap_name"
690 zfsctl_snapshot_name(zfsvfs_t *zfsvfs, const char *snap_name, int len,
693 objset_t *os = zfsvfs->z_os;
695 if (zfs_component_namecheck(snap_name, NULL, NULL) != 0)
696 return (SET_ERROR(EILSEQ));
698 dmu_objset_name(os, full_name);
699 if ((strlen(full_name) + 1 + strlen(snap_name)) >= len)
700 return (SET_ERROR(ENAMETOOLONG));
702 (void) strcat(full_name, "@");
703 (void) strcat(full_name, snap_name);
709 * Returns full path in full_path: "/pool/dataset/.zfs/snapshot/snap_name/"
712 zfsctl_snapshot_path_objset(zfsvfs_t *zfsvfs, uint64_t objsetid,
713 int path_len, char *full_path)
715 objset_t *os = zfsvfs->z_os;
716 fstrans_cookie_t cookie;
718 boolean_t case_conflict;
719 uint64_t id, pos = 0;
722 if (zfsvfs->z_vfs->vfs_mntpoint == NULL)
723 return (SET_ERROR(ENOENT));
725 cookie = spl_fstrans_mark();
726 snapname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
729 dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
730 error = dmu_snapshot_list_next(zfsvfs->z_os,
731 ZFS_MAX_DATASET_NAME_LEN, snapname, &id, &pos,
733 dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
741 snprintf(full_path, path_len, "%s/.zfs/snapshot/%s",
742 zfsvfs->z_vfs->vfs_mntpoint, snapname);
744 kmem_free(snapname, ZFS_MAX_DATASET_NAME_LEN);
745 spl_fstrans_unmark(cookie);
751 * Special case the handling of "..".
754 zfsctl_root_lookup(struct inode *dip, char *name, struct inode **ipp,
755 int flags, cred_t *cr, int *direntflags, pathname_t *realpnp)
757 zfsvfs_t *zfsvfs = ITOZSB(dip);
762 if (strcmp(name, "..") == 0) {
763 *ipp = dip->i_sb->s_root->d_inode;
764 } else if (strcmp(name, ZFS_SNAPDIR_NAME) == 0) {
765 *ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SNAPDIR,
766 &zpl_fops_snapdir, &zpl_ops_snapdir);
767 } else if (strcmp(name, ZFS_SHAREDIR_NAME) == 0) {
768 *ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SHARES,
769 &zpl_fops_shares, &zpl_ops_shares);
775 error = SET_ERROR(ENOENT);
783 * Lookup entry point for the 'snapshot' directory. Try to open the
784 * snapshot if it exist, creating the pseudo filesystem inode as necessary.
787 zfsctl_snapdir_lookup(struct inode *dip, char *name, struct inode **ipp,
788 int flags, cred_t *cr, int *direntflags, pathname_t *realpnp)
790 zfsvfs_t *zfsvfs = ITOZSB(dip);
796 error = dmu_snapshot_lookup(zfsvfs->z_os, name, &id);
802 *ipp = zfsctl_inode_lookup(zfsvfs, ZFSCTL_INO_SNAPDIRS - id,
803 &simple_dir_operations, &simple_dir_inode_operations);
805 error = SET_ERROR(ENOENT);
813 * Renaming a directory under '.zfs/snapshot' will automatically trigger
814 * a rename of the snapshot to the new given name. The rename is confined
815 * to the '.zfs/snapshot' directory snapshots cannot be moved elsewhere.
818 zfsctl_snapdir_rename(struct inode *sdip, char *snm,
819 struct inode *tdip, char *tnm, cred_t *cr, int flags)
821 zfsvfs_t *zfsvfs = ITOZSB(sdip);
822 char *to, *from, *real, *fsname;
825 if (!zfs_admin_snapshot)
826 return (SET_ERROR(EACCES));
830 to = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
831 from = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
832 real = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
833 fsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
835 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
836 error = dmu_snapshot_realname(zfsvfs->z_os, snm, real,
837 ZFS_MAX_DATASET_NAME_LEN, NULL);
840 } else if (error != ENOTSUP) {
845 dmu_objset_name(zfsvfs->z_os, fsname);
847 error = zfsctl_snapshot_name(ITOZSB(sdip), snm,
848 ZFS_MAX_DATASET_NAME_LEN, from);
850 error = zfsctl_snapshot_name(ITOZSB(tdip), tnm,
851 ZFS_MAX_DATASET_NAME_LEN, to);
853 error = zfs_secpolicy_rename_perms(from, to, cr);
858 * Cannot move snapshots out of the snapdir.
861 error = SET_ERROR(EINVAL);
866 * No-op when names are identical.
868 if (strcmp(snm, tnm) == 0) {
873 rw_enter(&zfs_snapshot_lock, RW_WRITER);
875 error = dsl_dataset_rename_snapshot(fsname, snm, tnm, B_FALSE);
877 (void) zfsctl_snapshot_rename(snm, tnm);
879 rw_exit(&zfs_snapshot_lock);
881 kmem_free(from, ZFS_MAX_DATASET_NAME_LEN);
882 kmem_free(to, ZFS_MAX_DATASET_NAME_LEN);
883 kmem_free(real, ZFS_MAX_DATASET_NAME_LEN);
884 kmem_free(fsname, ZFS_MAX_DATASET_NAME_LEN);
892 * Removing a directory under '.zfs/snapshot' will automatically trigger
893 * the removal of the snapshot with the given name.
896 zfsctl_snapdir_remove(struct inode *dip, char *name, cred_t *cr, int flags)
898 zfsvfs_t *zfsvfs = ITOZSB(dip);
899 char *snapname, *real;
902 if (!zfs_admin_snapshot)
903 return (SET_ERROR(EACCES));
907 snapname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
908 real = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
910 if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
911 error = dmu_snapshot_realname(zfsvfs->z_os, name, real,
912 ZFS_MAX_DATASET_NAME_LEN, NULL);
915 } else if (error != ENOTSUP) {
920 error = zfsctl_snapshot_name(ITOZSB(dip), name,
921 ZFS_MAX_DATASET_NAME_LEN, snapname);
923 error = zfs_secpolicy_destroy_perms(snapname, cr);
927 error = zfsctl_snapshot_unmount(snapname, MNT_FORCE);
928 if ((error == 0) || (error == ENOENT))
929 error = dsl_destroy_snapshot(snapname, B_FALSE);
931 kmem_free(snapname, ZFS_MAX_DATASET_NAME_LEN);
932 kmem_free(real, ZFS_MAX_DATASET_NAME_LEN);
940 * Creating a directory under '.zfs/snapshot' will automatically trigger
941 * the creation of a new snapshot with the given name.
944 zfsctl_snapdir_mkdir(struct inode *dip, char *dirname, vattr_t *vap,
945 struct inode **ipp, cred_t *cr, int flags)
947 zfsvfs_t *zfsvfs = ITOZSB(dip);
951 if (!zfs_admin_snapshot)
952 return (SET_ERROR(EACCES));
954 dsname = kmem_alloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
956 if (zfs_component_namecheck(dirname, NULL, NULL) != 0) {
957 error = SET_ERROR(EILSEQ);
961 dmu_objset_name(zfsvfs->z_os, dsname);
963 error = zfs_secpolicy_snapshot_perms(dsname, cr);
968 error = dmu_objset_snapshot_one(dsname, dirname);
972 error = zfsctl_snapdir_lookup(dip, dirname, ipp,
976 kmem_free(dsname, ZFS_MAX_DATASET_NAME_LEN);
982 * Flush everything out of the kernel's export table and such.
983 * This is needed as once the snapshot is used over NFS, its
984 * entries in svc_export and svc_expkey caches hold reference
985 * to the snapshot mount point. There is no known way of flushing
986 * only the entries related to the snapshot.
991 char *argv[] = { "/usr/sbin/exportfs", "-f", NULL };
992 char *envp[] = { NULL };
994 (void) call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
998 * Attempt to unmount a snapshot by making a call to user space.
999 * There is no assurance that this can or will succeed, is just a
1000 * best effort. In the case where it does fail, perhaps because
1001 * it's in use, the unmount will fail harmlessly.
1004 zfsctl_snapshot_unmount(char *snapname, int flags)
1006 char *argv[] = { "/usr/bin/env", "umount", "-t", "zfs", "-n", NULL,
1008 char *envp[] = { NULL };
1009 zfs_snapentry_t *se;
1012 rw_enter(&zfs_snapshot_lock, RW_READER);
1013 if ((se = zfsctl_snapshot_find_by_name(snapname)) == NULL) {
1014 rw_exit(&zfs_snapshot_lock);
1015 return (SET_ERROR(ENOENT));
1017 rw_exit(&zfs_snapshot_lock);
1021 if (flags & MNT_FORCE)
1023 argv[5] = se->se_path;
1024 dprintf("unmount; path=%s\n", se->se_path);
1025 error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1026 zfsctl_snapshot_rele(se);
1030 * The umount system utility will return 256 on error. We must
1031 * assume this error is because the file system is busy so it is
1032 * converted to the more sensible EBUSY.
1035 error = SET_ERROR(EBUSY);
1041 zfsctl_snapshot_mount(struct path *path, int flags)
1043 struct dentry *dentry = path->dentry;
1044 struct inode *ip = dentry->d_inode;
1046 zfsvfs_t *snap_zfsvfs;
1047 zfs_snapentry_t *se;
1048 char *full_name, *full_path;
1049 char *argv[] = { "/usr/bin/env", "mount", "-t", "zfs", "-n", NULL, NULL,
1051 char *envp[] = { NULL };
1056 return (SET_ERROR(EISDIR));
1058 zfsvfs = ITOZSB(ip);
1061 full_name = kmem_zalloc(ZFS_MAX_DATASET_NAME_LEN, KM_SLEEP);
1062 full_path = kmem_zalloc(MAXPATHLEN, KM_SLEEP);
1064 error = zfsctl_snapshot_name(zfsvfs, dname(dentry),
1065 ZFS_MAX_DATASET_NAME_LEN, full_name);
1070 * Construct a mount point path from sb of the ctldir inode and dirent
1071 * name, instead of from d_path(), so that chroot'd process doesn't fail
1074 snprintf(full_path, MAXPATHLEN, "%s/.zfs/snapshot/%s",
1075 zfsvfs->z_vfs->vfs_mntpoint ? zfsvfs->z_vfs->vfs_mntpoint : "",
1079 * Multiple concurrent automounts of a snapshot are never allowed.
1080 * The snapshot may be manually mounted as many times as desired.
1082 if (zfsctl_snapshot_ismounted(full_name)) {
1088 * Attempt to mount the snapshot from user space. Normally this
1089 * would be done using the vfs_kern_mount() function, however that
1090 * function is marked GPL-only and cannot be used. On error we
1091 * careful to log the real error to the console and return EISDIR
1092 * to safely abort the automount. This should be very rare.
1094 * If the user mode helper happens to return EBUSY, a concurrent
1095 * mount is already in progress in which case the error is ignored.
1096 * Take note that if the program was executed successfully the return
1097 * value from call_usermodehelper() will be (exitcode << 8 + signal).
1099 dprintf("mount; name=%s path=%s\n", full_name, full_path);
1100 argv[5] = full_name;
1101 argv[6] = full_path;
1102 error = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
1104 if (!(error & MOUNT_BUSY << 8)) {
1105 zfs_dbgmsg("Unable to automount %s error=%d",
1107 error = SET_ERROR(EISDIR);
1110 * EBUSY, this could mean a concurrent mount, or the
1111 * snapshot has already been mounted at completely
1112 * different place. We return 0 so VFS will retry. For
1113 * the latter case the VFS will retry several times
1114 * and return ELOOP, which is probably not a very good
1123 * Follow down in to the mounted snapshot and set MNT_SHRINKABLE
1124 * to identify this as an automounted filesystem.
1128 if (follow_down_one(&spath)) {
1129 snap_zfsvfs = ITOZSB(spath.dentry->d_inode);
1130 snap_zfsvfs->z_parent = zfsvfs;
1131 dentry = spath.dentry;
1132 spath.mnt->mnt_flags |= MNT_SHRINKABLE;
1134 rw_enter(&zfs_snapshot_lock, RW_WRITER);
1135 se = zfsctl_snapshot_alloc(full_name, full_path,
1136 snap_zfsvfs->z_os->os_spa, dmu_objset_id(snap_zfsvfs->z_os),
1138 zfsctl_snapshot_add(se);
1139 zfsctl_snapshot_unmount_delay_impl(se, zfs_expire_snapshot);
1140 rw_exit(&zfs_snapshot_lock);
1144 kmem_free(full_name, ZFS_MAX_DATASET_NAME_LEN);
1145 kmem_free(full_path, MAXPATHLEN);
1153 * Get the snapdir inode from fid
1156 zfsctl_snapdir_vget(struct super_block *sb, uint64_t objsetid, int gen,
1162 struct dentry *dentry;
1164 mnt = kmem_alloc(MAXPATHLEN, KM_SLEEP);
1166 error = zfsctl_snapshot_path_objset(sb->s_fs_info, objsetid,
1171 /* Trigger automount */
1172 error = -kern_path(mnt, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &path);
1178 * Get the snapdir inode. Note, we don't want to use the above
1179 * path because it contains the root of the snapshot rather
1182 *ipp = ilookup(sb, ZFSCTL_INO_SNAPDIRS - objsetid);
1184 error = SET_ERROR(ENOENT);
1188 /* check gen, see zfsctl_snapdir_fid */
1189 dentry = d_obtain_alias(igrab(*ipp));
1190 if (gen != (!IS_ERR(dentry) && d_mountpoint(dentry))) {
1193 error = SET_ERROR(ENOENT);
1195 if (!IS_ERR(dentry))
1198 kmem_free(mnt, MAXPATHLEN);
1203 zfsctl_shares_lookup(struct inode *dip, char *name, struct inode **ipp,
1204 int flags, cred_t *cr, int *direntflags, pathname_t *realpnp)
1206 zfsvfs_t *zfsvfs = ITOZSB(dip);
1213 if (zfsvfs->z_shares_dir == 0) {
1215 return (SET_ERROR(ENOTSUP));
1218 if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1219 error = zfs_lookup(dzp, name, &zp, 0, cr, NULL, NULL);
1229 * Initialize the various pieces we'll need to create and manipulate .zfs
1230 * directories. Currently this is unused but available.
1235 avl_create(&zfs_snapshots_by_name, snapentry_compare_by_name,
1236 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t,
1238 avl_create(&zfs_snapshots_by_objsetid, snapentry_compare_by_objsetid,
1239 sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t,
1241 rw_init(&zfs_snapshot_lock, NULL, RW_DEFAULT, NULL);
1245 * Cleanup the various pieces we needed for .zfs directories. In particular
1246 * ensure the expiry timer is canceled safely.
1251 avl_destroy(&zfs_snapshots_by_name);
1252 avl_destroy(&zfs_snapshots_by_objsetid);
1253 rw_destroy(&zfs_snapshot_lock);
1256 module_param(zfs_admin_snapshot, int, 0644);
1257 MODULE_PARM_DESC(zfs_admin_snapshot, "Enable mkdir/rmdir/mv in .zfs/snapshot");
1259 module_param(zfs_expire_snapshot, int, 0644);
1260 MODULE_PARM_DESC(zfs_expire_snapshot, "Seconds to expire .zfs/snapshot");