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]
22 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
23 * Copyright (c) 2012, 2015 by Delphix. All rights reserved.
24 * Copyright 2015, OmniTI Computer Consulting, Inc. All rights reserved.
28 * ZFS control directory (a.k.a. ".zfs")
30 * This directory provides a common location for all ZFS meta-objects.
31 * Currently, this is only the 'snapshot' directory, but this may expand in the
32 * future. The elements are built using the GFS primitives, as the hierarchy
33 * does not actually exist on disk.
35 * For 'snapshot', we don't want to have all snapshots always mounted, because
36 * this would take up a huge amount of space in /etc/mnttab. We have three
39 * ctldir ------> snapshotdir -------> snapshot
45 * The 'snapshot' node contains just enough information to lookup '..' and act
46 * as a mountpoint for the snapshot. Whenever we lookup a specific snapshot, we
47 * perform an automount of the underlying filesystem and return the
48 * corresponding vnode.
50 * All mounts are handled automatically by the kernel, but unmounts are
51 * (currently) handled from user land. The main reason is that there is no
52 * reliable way to auto-unmount the filesystem when it's "no longer in use".
53 * When the user unmounts a filesystem, we call zfsctl_unmount(), which
54 * unmounts any snapshots within the snapshot directory.
56 * The '.zfs', '.zfs/snapshot', and all directories created under
57 * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') are all GFS nodes and
58 * share the same vfs_t as the head filesystem (what '.zfs' lives under).
60 * File systems mounted ontop of the GFS nodes '.zfs/snapshot/<snapname>'
61 * (ie: snapshots) are ZFS nodes and have their own unique vfs_t.
62 * However, vnodes within these mounted on file systems have their v_vfsp
63 * fields set to the head filesystem to make NFS happy (see
64 * zfsctl_snapdir_lookup()). We VFS_HOLD the head filesystem's vfs_t
65 * so that it cannot be freed until all snapshots have been unmounted.
68 #include <sys/types.h>
69 #include <sys/param.h>
70 #include <sys/libkern.h>
71 #include <sys/dirent.h>
72 #include <sys/zfs_context.h>
73 #include <sys/zfs_ctldir.h>
74 #include <sys/zfs_ioctl.h>
75 #include <sys/zfs_vfsops.h>
76 #include <sys/namei.h>
79 #include <sys/dsl_dataset.h>
80 #include <sys/dsl_destroy.h>
81 #include <sys/dsl_deleg.h>
82 #include <sys/mount.h>
84 #include <sys/sysproto.h>
86 #include "zfs_namecheck.h"
88 #include <sys/kernel.h>
89 #include <sys/ccompat.h>
91 /* Common access mode for all virtual directories under the ctldir */
92 const uint16_t zfsctl_ctldir_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP |
96 * "Synthetic" filesystem implementation.
100 * Assert that A implies B.
102 #define KASSERT_IMPLY(A, B, msg) KASSERT(!(A) || (B), (msg));
104 static MALLOC_DEFINE(M_SFSNODES, "sfs_nodes", "synthetic-fs nodes");
106 typedef struct sfs_node {
107 char sn_name[ZFS_MAX_DATASET_NAME_LEN];
108 uint64_t sn_parent_id;
113 * Check the parent's ID as well as the node's to account for a chance
114 * that IDs originating from different domains (snapshot IDs, artificial
115 * IDs, znode IDs) may clash.
118 sfs_compare_ids(struct vnode *vp, void *arg)
120 sfs_node_t *n1 = vp->v_data;
121 sfs_node_t *n2 = arg;
124 equal = n1->sn_id == n2->sn_id &&
125 n1->sn_parent_id == n2->sn_parent_id;
127 /* Zero means equality. */
132 sfs_vnode_get(const struct mount *mp, int flags, uint64_t parent_id,
133 uint64_t id, struct vnode **vpp)
139 search.sn_parent_id = parent_id;
140 err = vfs_hash_get(mp, (uint32_t)id, flags, curthread, vpp,
141 sfs_compare_ids, &search);
146 sfs_vnode_insert(struct vnode *vp, int flags, uint64_t parent_id,
147 uint64_t id, struct vnode **vpp)
151 KASSERT(vp->v_data != NULL, ("sfs_vnode_insert with NULL v_data"));
152 err = vfs_hash_insert(vp, (uint32_t)id, flags, curthread, vpp,
153 sfs_compare_ids, vp->v_data);
158 sfs_vnode_remove(struct vnode *vp)
163 typedef void sfs_vnode_setup_fn(vnode_t *vp, void *arg);
166 sfs_vgetx(struct mount *mp, int flags, uint64_t parent_id, uint64_t id,
167 const char *tag, struct vop_vector *vops,
168 sfs_vnode_setup_fn setup, void *arg,
174 error = sfs_vnode_get(mp, flags, parent_id, id, vpp);
175 if (error != 0 || *vpp != NULL) {
176 KASSERT_IMPLY(error == 0, (*vpp)->v_data != NULL,
177 "sfs vnode with no data");
181 /* Allocate a new vnode/inode. */
182 error = getnewvnode(tag, mp, vops, &vp);
189 * Exclusively lock the vnode vnode while it's being constructed.
191 lockmgr(vp->v_vnlock, LK_EXCLUSIVE, NULL);
192 error = insmntque(vp, mp);
200 error = sfs_vnode_insert(vp, flags, parent_id, id, vpp);
201 if (error != 0 || *vpp != NULL) {
202 KASSERT_IMPLY(error == 0, (*vpp)->v_data != NULL,
203 "sfs vnode with no data");
212 sfs_print_node(sfs_node_t *node)
214 printf("\tname = %s\n", node->sn_name);
215 printf("\tparent_id = %ju\n", (uintmax_t)node->sn_parent_id);
216 printf("\tid = %ju\n", (uintmax_t)node->sn_id);
220 sfs_alloc_node(size_t size, const char *name, uint64_t parent_id, uint64_t id)
222 struct sfs_node *node;
224 KASSERT(strlen(name) < sizeof (node->sn_name),
225 ("sfs node name is too long"));
226 KASSERT(size >= sizeof (*node), ("sfs node size is too small"));
227 node = malloc(size, M_SFSNODES, M_WAITOK | M_ZERO);
228 strlcpy(node->sn_name, name, sizeof (node->sn_name));
229 node->sn_parent_id = parent_id;
236 sfs_destroy_node(sfs_node_t *node)
238 free(node, M_SFSNODES);
242 sfs_reclaim_vnode(vnode_t *vp)
246 sfs_vnode_remove(vp);
253 sfs_readdir_common(uint64_t parent_id, uint64_t id, struct vop_readdir_args *ap,
254 zfs_uio_t *uio, off_t *offp)
259 /* Reset ncookies for subsequent use of vfs_read_dirent. */
260 if (ap->a_ncookies != NULL)
263 if (zfs_uio_resid(uio) < sizeof (entry))
264 return (SET_ERROR(EINVAL));
266 if (zfs_uio_offset(uio) < 0)
267 return (SET_ERROR(EINVAL));
268 if (zfs_uio_offset(uio) == 0) {
270 entry.d_type = DT_DIR;
271 entry.d_name[0] = '.';
272 entry.d_name[1] = '\0';
274 entry.d_reclen = sizeof (entry);
275 error = vfs_read_dirent(ap, &entry, zfs_uio_offset(uio));
277 return (SET_ERROR(error));
280 if (zfs_uio_offset(uio) < sizeof (entry))
281 return (SET_ERROR(EINVAL));
282 if (zfs_uio_offset(uio) == sizeof (entry)) {
283 entry.d_fileno = parent_id;
284 entry.d_type = DT_DIR;
285 entry.d_name[0] = '.';
286 entry.d_name[1] = '.';
287 entry.d_name[2] = '\0';
289 entry.d_reclen = sizeof (entry);
290 error = vfs_read_dirent(ap, &entry, zfs_uio_offset(uio));
292 return (SET_ERROR(error));
296 *offp = 2 * sizeof (entry);
302 * .zfs inode namespace
304 * We need to generate unique inode numbers for all files and directories
305 * within the .zfs pseudo-filesystem. We use the following scheme:
310 * .zfs/snapshot/<snap> objectid(snap)
312 #define ZFSCTL_INO_SNAP(id) (id)
314 static struct vop_vector zfsctl_ops_root;
315 static struct vop_vector zfsctl_ops_snapdir;
316 static struct vop_vector zfsctl_ops_snapshot;
329 zfsctl_is_node(vnode_t *vp)
331 return (vn_matchops(vp, zfsctl_ops_root) ||
332 vn_matchops(vp, zfsctl_ops_snapdir) ||
333 vn_matchops(vp, zfsctl_ops_snapshot));
337 typedef struct zfsctl_root {
345 * Create the '.zfs' directory.
348 zfsctl_create(zfsvfs_t *zfsvfs)
350 zfsctl_root_t *dot_zfs;
355 ASSERT3P(zfsvfs->z_ctldir, ==, NULL);
357 snapdir = sfs_alloc_node(sizeof (*snapdir), "snapshot", ZFSCTL_INO_ROOT,
359 dot_zfs = (zfsctl_root_t *)sfs_alloc_node(sizeof (*dot_zfs), ".zfs", 0,
361 dot_zfs->snapdir = snapdir;
363 VERIFY0(VFS_ROOT(zfsvfs->z_vfs, LK_EXCLUSIVE, &rvp));
364 VERIFY0(sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
365 &crtime, sizeof (crtime)));
366 ZFS_TIME_DECODE(&dot_zfs->cmtime, crtime);
369 zfsvfs->z_ctldir = dot_zfs;
373 * Destroy the '.zfs' directory. Only called when the filesystem is unmounted.
374 * The nodes must not have any associated vnodes by now as they should be
378 zfsctl_destroy(zfsvfs_t *zfsvfs)
380 sfs_destroy_node(zfsvfs->z_ctldir->snapdir);
381 sfs_destroy_node((sfs_node_t *)zfsvfs->z_ctldir);
382 zfsvfs->z_ctldir = NULL;
386 zfsctl_fs_root_vnode(struct mount *mp, void *arg __unused, int flags,
389 return (VFS_ROOT(mp, flags, vpp));
393 zfsctl_common_vnode_setup(vnode_t *vp, void *arg)
395 ASSERT_VOP_ELOCKED(vp, __func__);
397 /* We support shared locking. */
404 zfsctl_root_vnode(struct mount *mp, void *arg __unused, int flags,
410 node = ((zfsvfs_t *)mp->mnt_data)->z_ctldir;
411 err = sfs_vgetx(mp, flags, 0, ZFSCTL_INO_ROOT, "zfs", &zfsctl_ops_root,
412 zfsctl_common_vnode_setup, node, vpp);
417 zfsctl_snapdir_vnode(struct mount *mp, void *arg __unused, int flags,
423 node = ((zfsvfs_t *)mp->mnt_data)->z_ctldir->snapdir;
424 err = sfs_vgetx(mp, flags, ZFSCTL_INO_ROOT, ZFSCTL_INO_SNAPDIR, "zfs",
425 &zfsctl_ops_snapdir, zfsctl_common_vnode_setup, node, vpp);
430 * Given a root znode, retrieve the associated .zfs directory.
431 * Add a hold to the vnode and return it.
434 zfsctl_root(zfsvfs_t *zfsvfs, int flags, vnode_t **vpp)
438 error = zfsctl_root_vnode(zfsvfs->z_vfs, NULL, flags, vpp);
443 * Common open routine. Disallow any write access.
446 zfsctl_common_open(struct vop_open_args *ap)
448 int flags = ap->a_mode;
451 return (SET_ERROR(EACCES));
457 * Common close routine. Nothing to do here.
461 zfsctl_common_close(struct vop_close_args *ap)
467 * Common access routine. Disallow writes.
470 zfsctl_common_access(struct vop_access_args *ap)
472 accmode_t accmode = ap->a_accmode;
474 if (accmode & VWRITE)
475 return (SET_ERROR(EACCES));
480 * Common getattr function. Fill in basic information.
483 zfsctl_common_getattr(vnode_t *vp, vattr_t *vap)
494 * We are a purely virtual object, so we have no
495 * blocksize or allocated blocks.
501 vap->va_mode = zfsctl_ctldir_mode;
504 * We live in the now (for atime).
508 /* FreeBSD: Reset chflags(2) flags. */
511 vap->va_nodeid = node->sn_id;
513 /* At least '.' and '..'. */
517 #ifndef _OPENSOLARIS_SYS_VNODE_H_
518 struct vop_fid_args {
525 zfsctl_common_fid(struct vop_fid_args *ap)
527 vnode_t *vp = ap->a_vp;
528 fid_t *fidp = (void *)ap->a_fid;
529 sfs_node_t *node = vp->v_data;
530 uint64_t object = node->sn_id;
534 zfid = (zfid_short_t *)fidp;
535 zfid->zf_len = SHORT_FID_LEN;
537 for (i = 0; i < sizeof (zfid->zf_object); i++)
538 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
540 /* .zfs nodes always have a generation number of 0 */
541 for (i = 0; i < sizeof (zfid->zf_gen); i++)
547 #ifndef _SYS_SYSPROTO_H_
548 struct vop_reclaim_args {
555 zfsctl_common_reclaim(struct vop_reclaim_args *ap)
557 vnode_t *vp = ap->a_vp;
559 (void) sfs_reclaim_vnode(vp);
563 #ifndef _SYS_SYSPROTO_H_
564 struct vop_print_args {
570 zfsctl_common_print(struct vop_print_args *ap)
572 sfs_print_node(ap->a_vp->v_data);
576 #ifndef _SYS_SYSPROTO_H_
577 struct vop_getattr_args {
580 struct ucred *a_cred;
585 * Get root directory attributes.
588 zfsctl_root_getattr(struct vop_getattr_args *ap)
590 struct vnode *vp = ap->a_vp;
591 struct vattr *vap = ap->a_vap;
592 zfsctl_root_t *node = vp->v_data;
594 zfsctl_common_getattr(vp, vap);
595 vap->va_ctime = node->cmtime;
596 vap->va_mtime = vap->va_ctime;
597 vap->va_birthtime = vap->va_ctime;
598 vap->va_nlink += 1; /* snapdir */
599 vap->va_size = vap->va_nlink;
604 * When we lookup "." we still can be asked to lock it
605 * differently, can't we?
608 zfsctl_relock_dot(vnode_t *dvp, int ltype)
611 if (ltype != VOP_ISLOCKED(dvp)) {
612 if (ltype == LK_EXCLUSIVE)
613 vn_lock(dvp, LK_UPGRADE | LK_RETRY);
614 else /* if (ltype == LK_SHARED) */
615 vn_lock(dvp, LK_DOWNGRADE | LK_RETRY);
617 /* Relock for the "." case may left us with reclaimed vnode. */
618 if (VN_IS_DOOMED(dvp)) {
620 return (SET_ERROR(ENOENT));
627 * Special case the handling of "..".
630 zfsctl_root_lookup(struct vop_lookup_args *ap)
632 struct componentname *cnp = ap->a_cnp;
633 vnode_t *dvp = ap->a_dvp;
634 vnode_t **vpp = ap->a_vpp;
635 int flags = ap->a_cnp->cn_flags;
636 int lkflags = ap->a_cnp->cn_lkflags;
637 int nameiop = ap->a_cnp->cn_nameiop;
640 ASSERT3S(dvp->v_type, ==, VDIR);
642 if ((flags & ISLASTCN) != 0 && nameiop != LOOKUP)
643 return (SET_ERROR(ENOTSUP));
645 if (cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.') {
646 err = zfsctl_relock_dot(dvp, lkflags & LK_TYPE_MASK);
649 } else if ((flags & ISDOTDOT) != 0) {
650 err = vn_vget_ino_gen(dvp, zfsctl_fs_root_vnode, NULL,
652 } else if (strncmp(cnp->cn_nameptr, "snapshot", cnp->cn_namelen) == 0) {
653 err = zfsctl_snapdir_vnode(dvp->v_mount, NULL, lkflags, vpp);
655 err = SET_ERROR(ENOENT);
663 zfsctl_root_readdir(struct vop_readdir_args *ap)
666 vnode_t *vp = ap->a_vp;
667 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
668 zfsctl_root_t *node = vp->v_data;
670 int *eofp = ap->a_eofflag;
674 zfs_uio_init(&uio, ap->a_uio);
676 ASSERT3S(vp->v_type, ==, VDIR);
678 error = sfs_readdir_common(zfsvfs->z_root, ZFSCTL_INO_ROOT, ap, &uio,
681 if (error == ENAMETOOLONG) /* ran out of destination space */
685 if (zfs_uio_offset(&uio) != dots_offset)
686 return (SET_ERROR(EINVAL));
688 _Static_assert(sizeof (node->snapdir->sn_name) <= sizeof (entry.d_name),
689 "node->snapdir->sn_name too big for entry.d_name");
690 entry.d_fileno = node->snapdir->sn_id;
691 entry.d_type = DT_DIR;
692 strcpy(entry.d_name, node->snapdir->sn_name);
693 entry.d_namlen = strlen(entry.d_name);
694 entry.d_reclen = sizeof (entry);
695 error = vfs_read_dirent(ap, &entry, zfs_uio_offset(&uio));
697 if (error == ENAMETOOLONG)
699 return (SET_ERROR(error));
707 zfsctl_root_vptocnp(struct vop_vptocnp_args *ap)
709 static const char dotzfs_name[4] = ".zfs";
713 if (*ap->a_buflen < sizeof (dotzfs_name))
714 return (SET_ERROR(ENOMEM));
716 error = vn_vget_ino_gen(ap->a_vp, zfsctl_fs_root_vnode, NULL,
719 return (SET_ERROR(error));
723 *ap->a_buflen -= sizeof (dotzfs_name);
724 bcopy(dotzfs_name, ap->a_buf + *ap->a_buflen, sizeof (dotzfs_name));
729 zfsctl_common_pathconf(struct vop_pathconf_args *ap)
732 * We care about ACL variables so that user land utilities like ls
733 * can display them correctly. Since the ctldir's st_dev is set to be
734 * the same as the parent dataset, we must support all variables that
737 switch (ap->a_name) {
739 *ap->a_retval = MIN(LONG_MAX, ZFS_LINK_MAX);
742 case _PC_FILESIZEBITS:
746 case _PC_MIN_HOLE_SIZE:
747 *ap->a_retval = (int)SPA_MINBLOCKSIZE;
750 case _PC_ACL_EXTENDED:
758 case _PC_ACL_PATH_MAX:
759 *ap->a_retval = ACL_MAX_ENTRIES;
763 *ap->a_retval = NAME_MAX;
767 return (vop_stdpathconf(ap));
772 * Returns a trivial ACL
775 zfsctl_common_getacl(struct vop_getacl_args *ap)
779 if (ap->a_type != ACL_TYPE_NFS4)
782 acl_nfs4_sync_acl_from_mode(ap->a_aclp, zfsctl_ctldir_mode, 0);
784 * acl_nfs4_sync_acl_from_mode assumes that the owner can always modify
785 * attributes. That is not the case for the ctldir, so we must clear
786 * those bits. We also must clear ACL_READ_NAMED_ATTRS, because xattrs
787 * aren't supported by the ctldir.
789 for (i = 0; i < ap->a_aclp->acl_cnt; i++) {
790 struct acl_entry *entry;
791 entry = &(ap->a_aclp->acl_entry[i]);
792 entry->ae_perm &= ~(ACL_WRITE_ACL | ACL_WRITE_OWNER |
793 ACL_WRITE_ATTRIBUTES | ACL_WRITE_NAMED_ATTRS |
794 ACL_READ_NAMED_ATTRS);
800 static struct vop_vector zfsctl_ops_root = {
801 .vop_default = &default_vnodeops,
802 #if __FreeBSD_version >= 1300121
803 .vop_fplookup_vexec = VOP_EAGAIN,
805 .vop_open = zfsctl_common_open,
806 .vop_close = zfsctl_common_close,
807 .vop_ioctl = VOP_EINVAL,
808 .vop_getattr = zfsctl_root_getattr,
809 .vop_access = zfsctl_common_access,
810 .vop_readdir = zfsctl_root_readdir,
811 .vop_lookup = zfsctl_root_lookup,
812 .vop_inactive = VOP_NULL,
813 .vop_reclaim = zfsctl_common_reclaim,
814 .vop_fid = zfsctl_common_fid,
815 .vop_print = zfsctl_common_print,
816 .vop_vptocnp = zfsctl_root_vptocnp,
817 .vop_pathconf = zfsctl_common_pathconf,
818 .vop_getacl = zfsctl_common_getacl,
819 #if __FreeBSD_version >= 1400043
820 .vop_add_writecount = vop_stdadd_writecount_nomsync,
823 VFS_VOP_VECTOR_REGISTER(zfsctl_ops_root);
826 zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname)
828 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
830 dmu_objset_name(os, zname);
831 if (strlen(zname) + 1 + strlen(name) >= len)
832 return (SET_ERROR(ENAMETOOLONG));
833 (void) strcat(zname, "@");
834 (void) strcat(zname, name);
839 zfsctl_snapshot_lookup(vnode_t *vp, const char *name, uint64_t *id)
841 objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
844 err = dsl_dataset_snap_lookup(dmu_objset_ds(os), name, id);
849 * Given a vnode get a root vnode of a filesystem mounted on top of
850 * the vnode, if any. The root vnode is referenced and locked.
851 * If no filesystem is mounted then the orinal vnode remains referenced
852 * and locked. If any error happens the orinal vnode is unlocked and
856 zfsctl_mounted_here(vnode_t **vpp, int flags)
861 ASSERT_VOP_LOCKED(*vpp, __func__);
862 ASSERT3S((*vpp)->v_type, ==, VDIR);
864 if ((mp = (*vpp)->v_mountedhere) != NULL) {
865 err = vfs_busy(mp, 0);
866 KASSERT(err == 0, ("vfs_busy(mp, 0) failed with %d", err));
867 KASSERT(vrefcnt(*vpp) > 1, ("unreferenced mountpoint"));
869 err = VFS_ROOT(mp, flags, vpp);
873 return (EJUSTRETURN);
877 const char *snap_name;
879 } snapshot_setup_arg_t;
882 zfsctl_snapshot_vnode_setup(vnode_t *vp, void *arg)
884 snapshot_setup_arg_t *ssa = arg;
887 ASSERT_VOP_ELOCKED(vp, __func__);
889 node = sfs_alloc_node(sizeof (sfs_node_t),
890 ssa->snap_name, ZFSCTL_INO_SNAPDIR, ssa->snap_id);
891 zfsctl_common_vnode_setup(vp, node);
893 /* We have to support recursive locking. */
898 * Lookup entry point for the 'snapshot' directory. Try to open the
899 * snapshot if it exist, creating the pseudo filesystem vnode as necessary.
900 * Perform a mount of the associated dataset on top of the vnode.
901 * There are four possibilities:
902 * - the snapshot node and vnode do not exist
903 * - the snapshot vnode is covered by the mounted snapshot
904 * - the snapshot vnode is not covered yet, the mount operation is in progress
905 * - the snapshot vnode is not covered, because the snapshot has been unmounted
906 * The last two states are transient and should be relatively short-lived.
909 zfsctl_snapdir_lookup(struct vop_lookup_args *ap)
911 vnode_t *dvp = ap->a_dvp;
912 vnode_t **vpp = ap->a_vpp;
913 struct componentname *cnp = ap->a_cnp;
914 char name[NAME_MAX + 1];
915 char fullname[ZFS_MAX_DATASET_NAME_LEN];
917 size_t mountpoint_len;
918 zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
920 int nameiop = cnp->cn_nameiop;
921 int lkflags = cnp->cn_lkflags;
922 int flags = cnp->cn_flags;
925 ASSERT3S(dvp->v_type, ==, VDIR);
927 if ((flags & ISLASTCN) != 0 && nameiop != LOOKUP)
928 return (SET_ERROR(ENOTSUP));
930 if (cnp->cn_namelen == 1 && *cnp->cn_nameptr == '.') {
931 err = zfsctl_relock_dot(dvp, lkflags & LK_TYPE_MASK);
936 if (flags & ISDOTDOT) {
937 err = vn_vget_ino_gen(dvp, zfsctl_root_vnode, NULL, lkflags,
942 if (cnp->cn_namelen >= sizeof (name))
943 return (SET_ERROR(ENAMETOOLONG));
945 strlcpy(name, ap->a_cnp->cn_nameptr, ap->a_cnp->cn_namelen + 1);
946 err = zfsctl_snapshot_lookup(dvp, name, &snap_id);
948 return (SET_ERROR(ENOENT));
951 snapshot_setup_arg_t ssa;
953 ssa.snap_name = name;
954 ssa.snap_id = snap_id;
955 err = sfs_vgetx(dvp->v_mount, LK_SHARED, ZFSCTL_INO_SNAPDIR,
956 snap_id, "zfs", &zfsctl_ops_snapshot,
957 zfsctl_snapshot_vnode_setup, &ssa, vpp);
961 /* Check if a new vnode has just been created. */
962 if (VOP_ISLOCKED(*vpp) == LK_EXCLUSIVE)
966 * Check if a snapshot is already mounted on top of the vnode.
968 err = zfsctl_mounted_here(vpp, lkflags);
969 if (err != EJUSTRETURN)
973 * If the vnode is not covered, then either the mount operation
974 * is in progress or the snapshot has already been unmounted
975 * but the vnode hasn't been inactivated and reclaimed yet.
976 * We can try to re-use the vnode in the latter case.
979 if (((*vpp)->v_iflag & VI_MOUNT) == 0) {
981 * Upgrade to exclusive lock in order to:
982 * - avoid race conditions
983 * - satisfy the contract of mount_snapshot()
985 err = VOP_LOCK(*vpp, LK_TRYUPGRADE | LK_INTERLOCK);
993 * In this state we can loop on uncontested locks and starve
994 * the thread doing the lengthy, non-trivial mount operation.
995 * So, yield to prevent that from happening.
998 kern_yield(PRI_USER);
1001 VERIFY0(zfsctl_snapshot_zname(dvp, name, sizeof (fullname), fullname));
1003 mountpoint_len = strlen(dvp->v_vfsp->mnt_stat.f_mntonname) +
1004 strlen("/" ZFS_CTLDIR_NAME "/snapshot/") + strlen(name) + 1;
1005 mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP);
1006 (void) snprintf(mountpoint, mountpoint_len,
1007 "%s/" ZFS_CTLDIR_NAME "/snapshot/%s",
1008 dvp->v_vfsp->mnt_stat.f_mntonname, name);
1010 err = mount_snapshot(curthread, vpp, "zfs", mountpoint, fullname, 0);
1011 kmem_free(mountpoint, mountpoint_len);
1014 * Fix up the root vnode mounted on .zfs/snapshot/<snapname>.
1016 * This is where we lie about our v_vfsp in order to
1017 * make .zfs/snapshot/<snapname> accessible over NFS
1018 * without requiring manual mounts of <snapname>.
1020 ASSERT3P(VTOZ(*vpp)->z_zfsvfs, !=, zfsvfs);
1021 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs;
1023 /* Clear the root flag (set via VFS_ROOT) as well. */
1024 (*vpp)->v_vflag &= ~VV_ROOT;
1033 zfsctl_snapdir_readdir(struct vop_readdir_args *ap)
1035 char snapname[ZFS_MAX_DATASET_NAME_LEN];
1036 struct dirent entry;
1037 vnode_t *vp = ap->a_vp;
1038 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1040 int *eofp = ap->a_eofflag;
1044 zfs_uio_init(&uio, ap->a_uio);
1046 ASSERT3S(vp->v_type, ==, VDIR);
1048 error = sfs_readdir_common(ZFSCTL_INO_ROOT, ZFSCTL_INO_SNAPDIR, ap,
1049 &uio, &dots_offset);
1051 if (error == ENAMETOOLONG) /* ran out of destination space */
1061 cookie = zfs_uio_offset(&uio) - dots_offset;
1063 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG);
1064 error = dmu_snapshot_list_next(zfsvfs->z_os, sizeof (snapname),
1065 snapname, &id, &cookie, NULL);
1066 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG);
1068 if (error == ENOENT) {
1077 entry.d_fileno = id;
1078 entry.d_type = DT_DIR;
1079 strcpy(entry.d_name, snapname);
1080 entry.d_namlen = strlen(entry.d_name);
1081 entry.d_reclen = sizeof (entry);
1082 error = vfs_read_dirent(ap, &entry, zfs_uio_offset(&uio));
1084 if (error == ENAMETOOLONG)
1087 return (SET_ERROR(error));
1089 zfs_uio_setoffset(&uio, cookie + dots_offset);
1091 __builtin_unreachable();
1095 zfsctl_snapdir_getattr(struct vop_getattr_args *ap)
1097 vnode_t *vp = ap->a_vp;
1098 vattr_t *vap = ap->a_vap;
1099 zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1101 uint64_t snap_count;
1105 ds = dmu_objset_ds(zfsvfs->z_os);
1106 zfsctl_common_getattr(vp, vap);
1107 vap->va_ctime = dmu_objset_snap_cmtime(zfsvfs->z_os);
1108 vap->va_mtime = vap->va_ctime;
1109 vap->va_birthtime = vap->va_ctime;
1110 if (dsl_dataset_phys(ds)->ds_snapnames_zapobj != 0) {
1111 err = zap_count(dmu_objset_pool(ds->ds_objset)->dp_meta_objset,
1112 dsl_dataset_phys(ds)->ds_snapnames_zapobj, &snap_count);
1117 vap->va_nlink += snap_count;
1119 vap->va_size = vap->va_nlink;
1125 static struct vop_vector zfsctl_ops_snapdir = {
1126 .vop_default = &default_vnodeops,
1127 #if __FreeBSD_version >= 1300121
1128 .vop_fplookup_vexec = VOP_EAGAIN,
1130 .vop_open = zfsctl_common_open,
1131 .vop_close = zfsctl_common_close,
1132 .vop_getattr = zfsctl_snapdir_getattr,
1133 .vop_access = zfsctl_common_access,
1134 .vop_readdir = zfsctl_snapdir_readdir,
1135 .vop_lookup = zfsctl_snapdir_lookup,
1136 .vop_reclaim = zfsctl_common_reclaim,
1137 .vop_fid = zfsctl_common_fid,
1138 .vop_print = zfsctl_common_print,
1139 .vop_pathconf = zfsctl_common_pathconf,
1140 .vop_getacl = zfsctl_common_getacl,
1141 #if __FreeBSD_version >= 1400043
1142 .vop_add_writecount = vop_stdadd_writecount_nomsync,
1145 VFS_VOP_VECTOR_REGISTER(zfsctl_ops_snapdir);
1149 zfsctl_snapshot_inactive(struct vop_inactive_args *ap)
1151 vnode_t *vp = ap->a_vp;
1153 VERIFY3S(vrecycle(vp), ==, 1);
1158 zfsctl_snapshot_reclaim(struct vop_reclaim_args *ap)
1160 vnode_t *vp = ap->a_vp;
1161 void *data = vp->v_data;
1163 sfs_reclaim_vnode(vp);
1164 sfs_destroy_node(data);
1169 zfsctl_snapshot_vptocnp(struct vop_vptocnp_args *ap)
1181 len = strlen(node->sn_name);
1182 if (*ap->a_buflen < len)
1183 return (SET_ERROR(ENOMEM));
1186 * Prevent unmounting of the snapshot while the vnode lock
1187 * is not held. That is not strictly required, but allows
1188 * us to assert that an uncovered snapshot vnode is never
1191 mp = vp->v_mountedhere;
1193 return (SET_ERROR(ENOENT));
1194 error = vfs_busy(mp, 0);
1195 KASSERT(error == 0, ("vfs_busy(mp, 0) failed with %d", error));
1198 * We can vput the vnode as we can now depend on the reference owned
1199 * by the busied mp. But we also need to hold the vnode, because
1200 * the reference may go after vfs_unbusy() which has to be called
1201 * before we can lock the vnode again.
1203 locked = VOP_ISLOCKED(vp);
1204 #if __FreeBSD_version >= 1300045
1205 enum vgetstate vs = vget_prep(vp);
1211 /* Look up .zfs/snapshot, our parent. */
1212 error = zfsctl_snapdir_vnode(vp->v_mount, NULL, LK_SHARED, &dvp);
1216 *ap->a_buflen -= len;
1217 bcopy(node->sn_name, ap->a_buf + *ap->a_buflen, len);
1220 #if __FreeBSD_version >= 1300045
1221 vget_finish(vp, locked | LK_RETRY, vs);
1223 vget(vp, locked | LK_VNHELD | LK_RETRY, curthread);
1229 * These VP's should never see the light of day. They should always
1232 static struct vop_vector zfsctl_ops_snapshot = {
1233 .vop_default = NULL, /* ensure very restricted access */
1234 #if __FreeBSD_version >= 1300121
1235 .vop_fplookup_vexec = VOP_EAGAIN,
1237 .vop_inactive = zfsctl_snapshot_inactive,
1238 #if __FreeBSD_version >= 1300045
1239 .vop_need_inactive = vop_stdneed_inactive,
1241 .vop_reclaim = zfsctl_snapshot_reclaim,
1242 .vop_vptocnp = zfsctl_snapshot_vptocnp,
1243 .vop_lock1 = vop_stdlock,
1244 .vop_unlock = vop_stdunlock,
1245 .vop_islocked = vop_stdislocked,
1246 .vop_advlockpurge = vop_stdadvlockpurge, /* called by vgone */
1247 .vop_print = zfsctl_common_print,
1248 #if __FreeBSD_version >= 1400043
1249 .vop_add_writecount = vop_stdadd_writecount_nomsync,
1252 VFS_VOP_VECTOR_REGISTER(zfsctl_ops_snapshot);
1255 zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp)
1257 zfsvfs_t *zfsvfs __unused = vfsp->vfs_data;
1261 ASSERT3P(zfsvfs->z_ctldir, !=, NULL);
1263 error = sfs_vnode_get(vfsp, LK_EXCLUSIVE,
1264 ZFSCTL_INO_SNAPDIR, objsetid, &vp);
1265 if (error == 0 && vp != NULL) {
1267 * XXX Probably need to at least reference, if not busy, the mp.
1269 if (vp->v_mountedhere != NULL)
1270 *zfsvfsp = vp->v_mountedhere->mnt_data;
1273 if (*zfsvfsp == NULL)
1274 return (SET_ERROR(EINVAL));
1279 * Unmount any snapshots for the given filesystem. This is called from
1280 * zfs_umount() - if we have a ctldir, then go through and unmount all the
1284 zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr)
1286 char snapname[ZFS_MAX_DATASET_NAME_LEN];
1287 zfsvfs_t *zfsvfs = vfsp->vfs_data;
1293 ASSERT3P(zfsvfs->z_ctldir, !=, NULL);
1299 dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG);
1300 error = dmu_snapshot_list_next(zfsvfs->z_os, sizeof (snapname),
1301 snapname, &id, &cookie, NULL);
1302 dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG);
1304 if (error == ENOENT)
1310 error = sfs_vnode_get(vfsp, LK_EXCLUSIVE,
1311 ZFSCTL_INO_SNAPDIR, id, &vp);
1312 if (error != 0 || vp == NULL)
1315 mp = vp->v_mountedhere;
1318 * v_mountedhere being NULL means that the
1319 * (uncovered) vnode is in a transient state
1320 * (mounting or unmounting), so loop until it
1330 continue; /* no mountpoint, nothing to do */
1333 * The mount-point vnode is kept locked to avoid spurious EBUSY
1334 * from a concurrent umount.
1335 * The vnode lock must have recursive locking enabled.
1338 error = dounmount(mp, fflags, curthread);
1339 KASSERT_IMPLY(error == 0, vrefcnt(vp) == 1,
1340 ("extra references after unmount"));
1345 KASSERT_IMPLY((fflags & MS_FORCE) != 0, error == 0,
1346 ("force unmounting failed"));
1351 zfsctl_snapshot_unmount(const char *snapname, int flags __unused)
1354 zfsvfs_t *zfsvfs = NULL;
1356 if (strchr(snapname, '@') == NULL)
1359 int err = getzfsvfs(snapname, &zfsvfs);
1361 ASSERT3P(zfsvfs, ==, NULL);
1364 vfsp = zfsvfs->z_vfs;
1366 ASSERT(!dsl_pool_config_held(dmu_objset_pool(zfsvfs->z_os)));
1370 return (dounmount(vfsp, MS_FORCE, curthread));