/*
* Called when an unmount is requested and certain sanity checks have
* already passed. At this point no dentries or inodes have been reclaimed
* from their respective caches. We drop the extra reference on the .zfs
* control directory to allow everything to be reclaimed. All snapshots
* must already have been unmounted to reach this point.
*/
void
zfs_preumount(struct super_block *sb)
{
zfs_sb_t *zsb = sb->s_fs_info;
/* zsb is NULL when zfs_domount fails during mount */
if (zsb) {
zfsctl_destroy(sb->s_fs_info);
/*
* Wait for iput_async before entering evict_inodes in
* generic_shutdown_super. The reason we must finish before
* evict_inodes is when lazytime is on, or when zfs_purgedir
* calls zfs_zget, iput would bump i_count from 0 to 1. This
* would race with the i_count check in evict_inodes. This means
* it could destroy the inode while we are still using it.
*
* We wait for two passes. xattr directories in the first pass
* may add xattr entries in zfs_purgedir, so in the second pass
* we wait for them. We don't use taskq_wait here because it is
* a pool wide taskq. Other mounted filesystems can constantly
* do iput_async and there's no guarantee when taskq will be
* empty.
*/
taskq_wait_outstanding(dsl_pool_iput_taskq(
dmu_objset_pool(zsb->z_os)), 0);
taskq_wait_outstanding(dsl_pool_iput_taskq(
dmu_objset_pool(zsb->z_os)), 0);
}
}
/*
* Delete the entire contents of a directory. Return a count
* of the number of entries that could not be deleted. If we encounter
* an error, return a count of at least one so that the directory stays
* in the unlinked set.
*
* NOTE: this function assumes that the directory is inactive,
* so there is no need to lock its entries before deletion.
* Also, it assumes the directory contents is *only* regular
* files.
*/
static int
zfs_purgedir(znode_t *dzp)
{
zap_cursor_t zc;
zap_attribute_t zap;
znode_t *xzp;
dmu_tx_t *tx;
zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
zfs_dirlock_t dl;
int skipped = 0;
int error;
for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id);
(error = zap_cursor_retrieve(&zc, &zap)) == 0;
zap_cursor_advance(&zc)) {
error = zfs_zget(zfsvfs,
ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
if (error) {
skipped += 1;
continue;
}
/*
ASSERT((ZTOV(xzp)->v_type == VREG) ||
(ZTOV(xzp)->v_type == VLNK));
*/
tx = dmu_tx_create(zfsvfs->z_os);
dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
/* Is this really needed ? */
zfs_sa_upgrade_txholds(tx, xzp);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
//VN_RELE(ZTOV(xzp)); // async
VN_RELE_ASYNC(ZTOV(xzp), dsl_pool_vnrele_taskq(dmu_objset_pool(zfsvfs->z_os)));
skipped += 1;
continue;
}
bzero(&dl, sizeof (dl));
dl.dl_dzp = dzp;
dl.dl_name = zap.za_name;
error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
if (error)
skipped += 1;
dmu_tx_commit(tx);
//VN_RELE(ZTOV(xzp)); // async
VN_RELE_ASYNC(ZTOV(xzp), dsl_pool_vnrele_taskq(dmu_objset_pool(zfsvfs->z_os)));
}
zap_cursor_fini(&zc);
if (error != ENOENT)
skipped += 1;
return (skipped);
}
zilog_t *
zil_alloc(objset_t *os, zil_header_t *zh_phys)
{
zilog_t *zilog;
zilog = kmem_zalloc(sizeof (zilog_t), KM_SLEEP);
zilog->zl_header = zh_phys;
zilog->zl_os = os;
zilog->zl_spa = dmu_objset_spa(os);
zilog->zl_dmu_pool = dmu_objset_pool(os);
zilog->zl_destroy_txg = TXG_INITIAL - 1;
mutex_init(&zilog->zl_lock, NULL, MUTEX_DEFAULT, NULL);
list_create(&zilog->zl_itx_list, sizeof (itx_t),
offsetof(itx_t, itx_node));
list_create(&zilog->zl_lwb_list, sizeof (lwb_t),
offsetof(lwb_t, lwb_node));
mutex_init(&zilog->zl_vdev_lock, NULL, MUTEX_DEFAULT, NULL);
avl_create(&zilog->zl_vdev_tree, zil_vdev_compare,
sizeof (zil_vdev_node_t), offsetof(zil_vdev_node_t, zv_node));
cv_init(&zilog->zl_cv_writer, NULL, CV_DEFAULT, NULL);
cv_init(&zilog->zl_cv_suspend, NULL, CV_DEFAULT, NULL);
return (zilog);
}
/*
* Ensure the zap is flushed then inform the VFS of the capacity change.
*/
static int
zvol_update_volsize(uint64_t volsize, objset_t *os)
{
dmu_tx_t *tx;
int error;
uint64_t txg;
ASSERT(MUTEX_HELD(&zvol_state_lock));
tx = dmu_tx_create(os);
dmu_tx_hold_zap(tx, ZVOL_ZAP_OBJ, TRUE, NULL);
dmu_tx_mark_netfree(tx);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
return (SET_ERROR(error));
}
txg = dmu_tx_get_txg(tx);
error = zap_update(os, ZVOL_ZAP_OBJ, "size", 8, 1,
&volsize, tx);
dmu_tx_commit(tx);
txg_wait_synced(dmu_objset_pool(os), txg);
if (error == 0)
error = dmu_free_long_range(os,
ZVOL_OBJ, volsize, DMU_OBJECT_END);
return (error);
}
/*
* Return an empty bpobj, preferably the empty dummy one (dp_empty_bpobj).
*/
uint64_t
bpobj_alloc_empty(objset_t *os, int blocksize, dmu_tx_t *tx)
{
zfeature_info_t *empty_bpobj_feat =
&spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ];
spa_t *spa = dmu_objset_spa(os);
dsl_pool_t *dp = dmu_objset_pool(os);
if (spa_feature_is_enabled(spa, empty_bpobj_feat)) {
if (!spa_feature_is_active(spa, empty_bpobj_feat)) {
ASSERT0(dp->dp_empty_bpobj);
dp->dp_empty_bpobj =
bpobj_alloc(os, SPA_MAXBLOCKSIZE, tx);
VERIFY(zap_add(os,
DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
&dp->dp_empty_bpobj, tx) == 0);
}
spa_feature_incr(spa, empty_bpobj_feat, tx);
ASSERT(dp->dp_empty_bpobj != 0);
return (dp->dp_empty_bpobj);
} else {
return (bpobj_alloc(os, blocksize, tx));
}
}
void
dmu_objset_rele(objset_t *os, void *tag)
{
dsl_pool_t *dp = dmu_objset_pool(os);
dsl_dataset_rele(os->os_dsl_dataset, tag);
dsl_pool_rele(dp, tag);
}
/* ARGSUSED */
static int
zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp,
offset_t *offp, offset_t *nextp, void *data, int flags)
{
zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
char snapname[ZFS_MAX_DATASET_NAME_LEN];
uint64_t id, cookie;
boolean_t case_conflict;
int error;
ZFS_ENTER(zfsvfs);
cookie = *offp;
dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG);
error = dmu_snapshot_list_next(zfsvfs->z_os,
sizeof (snapname), snapname, &id, &cookie, &case_conflict);
dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG);
if (error) {
ZFS_EXIT(zfsvfs);
if (error == ENOENT) {
*eofp = 1;
return (0);
}
return (error);
}
if (flags & V_RDDIR_ENTFLAGS) {
edirent_t *eodp = dp;
(void) strcpy(eodp->ed_name, snapname);
eodp->ed_ino = ZFSCTL_INO_SNAP(id);
eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0;
} else {
struct dirent64 *odp = dp;
(void) strcpy(odp->d_name, snapname);
odp->d_ino = ZFSCTL_INO_SNAP(id);
}
*nextp = cookie;
ZFS_EXIT(zfsvfs);
return (0);
}
static int osd_object_sync(const struct lu_env *env, struct dt_object *dt)
{
struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
ENTRY;
/* XXX: no other option than syncing the whole filesystem until we
* support ZIL */
txg_wait_synced(dmu_objset_pool(osd->od_objset.os), 0ULL);
RETURN(0);
}
static int
zpl_snapdir_iterate(struct file *filp, struct dir_context *ctx)
{
zfs_sb_t *zsb = ITOZSB(filp->f_path.dentry->d_inode);
fstrans_cookie_t cookie;
char snapname[MAXNAMELEN];
boolean_t case_conflict;
uint64_t id, pos;
int error = 0;
ZFS_ENTER(zsb);
cookie = spl_fstrans_mark();
if (!dir_emit_dots(filp, ctx))
goto out;
pos = ctx->pos;
while (error == 0) {
dsl_pool_config_enter(dmu_objset_pool(zsb->z_os), FTAG);
error = -dmu_snapshot_list_next(zsb->z_os, MAXNAMELEN,
snapname, &id, &pos, &case_conflict);
dsl_pool_config_exit(dmu_objset_pool(zsb->z_os), FTAG);
if (error)
goto out;
if (!dir_emit(ctx, snapname, strlen(snapname),
ZFSCTL_INO_SHARES - id, DT_DIR))
goto out;
ctx->pos = pos;
}
out:
spl_fstrans_unmark(cookie);
ZFS_EXIT(zsb);
if (error == -ENOENT)
return (0);
return (error);
}
/*ARGSUSED*/
int
zfs_sync(struct super_block *sb, int wait, cred_t *cr)
{
zfs_sb_t *zsb = sb->s_fs_info;
/*
* Data integrity is job one. We don't want a compromised kernel
* writing to the storage pool, so we never sync during panic.
*/
if (unlikely(oops_in_progress))
return (0);
/*
* Semantically, the only requirement is that the sync be initiated.
* The DMU syncs out txgs frequently, so there's nothing to do.
*/
if (!wait)
return (0);
if (zsb != NULL) {
/*
* Sync a specific filesystem.
*/
dsl_pool_t *dp;
ZFS_ENTER(zsb);
dp = dmu_objset_pool(zsb->z_os);
/*
* If the system is shutting down, then skip any
* filesystems which may exist on a suspended pool.
*/
if (spa_suspended(dp->dp_spa)) {
ZFS_EXIT(zsb);
return (0);
}
if (zsb->z_log != NULL)
zil_commit(zsb->z_log, 0);
ZFS_EXIT(zsb);
} else {
/*
* Sync all ZFS filesystems. This is what happens when you
* run sync(1M). Unlike other filesystems, ZFS honors the
* request by waiting for all pools to commit all dirty data.
*/
spa_sync_allpools();
}
return (0);
}
static int
zfs_vfs_sync(struct mount *mp, __unused int waitfor, __unused vfs_context_t context)
{
zfsvfs_t *zfsvfs = vfs_fsprivate(mp);
ZFS_ENTER(zfsvfs);
/*
* Mac OS X needs a file system modify time
*
* We use the mtime of the "com.apple.system.mtime"
* extended attribute, which is associated with the
* file system root directory.
*
* Here we sync any mtime changes to this attribute.
*/
if (zfsvfs->z_mtime_vp != NULL) {
timestruc_t mtime;
znode_t *zp;
top:
zp = VTOZ(zfsvfs->z_mtime_vp);
ZFS_TIME_DECODE(&mtime, zp->z_phys->zp_mtime);
if (zfsvfs->z_last_mtime_synced < mtime.tv_sec) {
dmu_tx_t *tx;
int error;
tx = dmu_tx_create(zfsvfs->z_os);
dmu_tx_hold_bonus(tx, zp->z_id);
error = dmu_tx_assign(tx, zfsvfs->z_assign);
if (error) {
if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) {
dmu_tx_wait(tx);
dmu_tx_abort(tx);
goto top;
}
dmu_tx_abort(tx);
} else {
dmu_buf_will_dirty(zp->z_dbuf, tx);
dmu_tx_commit(tx);
zfsvfs->z_last_mtime_synced = mtime.tv_sec;
}
}
}
if (zfsvfs->z_log != NULL)
zil_commit(zfsvfs->z_log, UINT64_MAX, 0);
else
txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
ZFS_EXIT(zfsvfs);
return (0);
}
void
bpobj_decr_empty(objset_t *os, dmu_tx_t *tx)
{
dsl_pool_t *dp = dmu_objset_pool(os);
spa_feature_decr(dmu_objset_spa(os), SPA_FEATURE_EMPTY_BPOBJ, tx);
if (!spa_feature_is_active(dmu_objset_spa(os),
SPA_FEATURE_EMPTY_BPOBJ)) {
VERIFY3U(0, ==, zap_remove(dp->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_EMPTY_BPOBJ, tx));
VERIFY3U(0, ==, dmu_object_free(os, dp->dp_empty_bpobj, tx));
dp->dp_empty_bpobj = 0;
}
static int osd_object_sync(const struct lu_env *env, struct dt_object *dt,
__u64 start, __u64 end)
{
struct osd_device *osd = osd_obj2dev(osd_dt_obj(dt));
ENTRY;
/* XXX: no other option than syncing the whole filesystem until we
* support ZIL. If the object tracked the txg that it was last
* modified in, it could pass that txg here instead of "0". Maybe
* the changes are already committed, so no wait is needed at all? */
txg_wait_synced(dmu_objset_pool(osd->od_os), 0ULL);
RETURN(0);
}
int
zfs_vnop_ioctl_fullfsync(struct vnode *vp, vfs_context_t ct, zfsvfs_t *zfsvfs)
{
int error;
error = zfs_fsync(vp, /*syncflag*/0, NULL, (caller_context_t *)ct);
if (error)
return (error);
if (zfsvfs->z_log != NULL)
zil_commit(zfsvfs->z_log, 0);
else
txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
return (0);
}
int
dmu_objset_userspace_upgrade(objset_t *os)
{
uint64_t obj;
int err = 0;
if (dmu_objset_userspace_present(os))
return (0);
if (!dmu_objset_userused_enabled(os->os))
return (ENOTSUP);
if (dmu_objset_is_snapshot(os))
return (EINVAL);
/*
* We simply need to mark every object dirty, so that it will be
* synced out and now accounted. If this is called
* concurrently, or if we already did some work before crashing,
* that's fine, since we track each object's accounted state
* independently.
*/
for (obj = 0; err == 0; err = dmu_object_next(os, &obj, FALSE, 0)) {
dmu_tx_t *tx;
dmu_buf_t *db;
int objerr;
if (issig(JUSTLOOKING) && issig(FORREAL))
return (EINTR);
objerr = dmu_bonus_hold(os, obj, FTAG, &db);
if (objerr)
continue;
tx = dmu_tx_create(os);
dmu_tx_hold_bonus(tx, obj);
objerr = dmu_tx_assign(tx, TXG_WAIT);
if (objerr) {
dmu_tx_abort(tx);
continue;
}
dmu_buf_will_dirty(db, tx);
dmu_buf_rele(db, FTAG);
dmu_tx_commit(tx);
}
os->os->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
txg_wait_synced(dmu_objset_pool(os), 0);
return (0);
}
/*ARGSUSED*/
int
zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
{
/*
* Data integrity is job one. We don't want a compromised kernel
* writing to the storage pool, so we never sync during panic.
*/
if (panicstr)
return (0);
/*
* SYNC_ATTR is used by fsflush() to force old filesystems like UFS
* to sync metadata, which they would otherwise cache indefinitely.
* Semantically, the only requirement is that the sync be initiated.
* The DMU syncs out txgs frequently, so there's nothing to do.
*/
if (flag & SYNC_ATTR)
return (0);
if (vfsp != NULL) {
/*
* Sync a specific filesystem.
*/
zfsvfs_t *zfsvfs = vfsp->vfs_data;
ZFS_ENTER(zfsvfs);
if (zfsvfs->z_log != NULL)
zil_commit(zfsvfs->z_log, UINT64_MAX, 0);
else
txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
ZFS_EXIT(zfsvfs);
} else {
/*
* Sync all ZFS filesystems. This is what happens when you
* run sync(1M). Unlike other filesystems, ZFS honors the
* request by waiting for all pools to commit all dirty data.
*/
spa_sync_allpools();
}
return (0);
}
static void
zvol_last_close(zvol_state_t *zv)
{
zil_close(zv->zv_zilog);
zv->zv_zilog = NULL;
dmu_buf_rele(zv->zv_dbuf, zvol_tag);
zv->zv_dbuf = NULL;
/*
* Evict cached data
*/
if (dsl_dataset_is_dirty(dmu_objset_ds(zv->zv_objset)) &&
!(zv->zv_flags & ZVOL_RDONLY))
txg_wait_synced(dmu_objset_pool(zv->zv_objset), 0);
(void) dmu_objset_evict_dbufs(zv->zv_objset);
dmu_objset_disown(zv->zv_objset, zvol_tag);
zv->zv_objset = NULL;
}
/*
* When we are called, os MUST refer to an objset associated with a dataset
* that is owned by 'tag'; that is, is held and long held by 'tag' and ds_owner
* == tag. We will then release and reacquire ownership of the dataset while
* holding the pool config_rwlock to avoid intervening namespace or ownership
* changes may occur.
*
* This exists solely to accommodate zfs_ioc_userspace_upgrade()'s desire to
* release the hold on its dataset and acquire a new one on the dataset of the
* same name so that it can be partially torn down and reconstructed.
*/
void
dmu_objset_refresh_ownership(objset_t *os, void *tag)
{
dsl_pool_t *dp;
dsl_dataset_t *ds, *newds;
char name[MAXNAMELEN];
ds = os->os_dsl_dataset;
VERIFY3P(ds, !=, NULL);
VERIFY3P(ds->ds_owner, ==, tag);
VERIFY(dsl_dataset_long_held(ds));
dsl_dataset_name(ds, name);
dp = dmu_objset_pool(os);
dsl_pool_config_enter(dp, FTAG);
dmu_objset_disown(os, tag);
VERIFY0(dsl_dataset_own(dp, name, tag, &newds));
VERIFY3P(newds, ==, os->os_dsl_dataset);
dsl_pool_config_exit(dp, FTAG);
}
/*ARGSUSED*/
static int
zfs_sync(vfs_t *vfsp, int waitfor)
{
/*
* Data integrity is job one. We don't want a compromised kernel
* writing to the storage pool, so we never sync during panic.
*/
if (panicstr)
return (0);
if (vfsp != NULL) {
/*
* Sync a specific filesystem.
*/
zfsvfs_t *zfsvfs = vfsp->vfs_data;
int error;
error = vfs_stdsync(vfsp, waitfor);
if (error != 0)
return (error);
ZFS_ENTER(zfsvfs);
if (zfsvfs->z_log != NULL)
zil_commit(zfsvfs->z_log, UINT64_MAX, 0);
else
txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
ZFS_EXIT(zfsvfs);
} else {
/*
* Sync all ZFS filesystems. This is what happens when you
* run sync(1M). Unlike other filesystems, ZFS honors the
* request by waiting for all pools to commit all dirty data.
*/
spa_sync_allpools();
}
return (0);
}
/*
* Teardown the zfs_sb_t.
*
* Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
* and 'z_teardown_inactive_lock' held.
*/
int
zfs_sb_teardown(zfs_sb_t *zsb, boolean_t unmounting)
{
znode_t *zp;
rrw_enter(&zsb->z_teardown_lock, RW_WRITER, FTAG);
if (!unmounting) {
/*
* We purge the parent filesystem's super block as the
* parent filesystem and all of its snapshots have their
* inode's super block set to the parent's filesystem's
* super block. Note, 'z_parent' is self referential
* for non-snapshots.
*/
shrink_dcache_sb(zsb->z_parent->z_sb);
}
/*
* If someone has not already unmounted this file system,
* drain the iput_taskq to ensure all active references to the
* zfs_sb_t have been handled only then can it be safely destroyed.
*/
if (zsb->z_os)
taskq_wait(dsl_pool_iput_taskq(dmu_objset_pool(zsb->z_os)));
/*
* Close the zil. NB: Can't close the zil while zfs_inactive
* threads are blocked as zil_close can call zfs_inactive.
*/
if (zsb->z_log) {
zil_close(zsb->z_log);
zsb->z_log = NULL;
}
rw_enter(&zsb->z_teardown_inactive_lock, RW_WRITER);
/*
* If we are not unmounting (ie: online recv) and someone already
* unmounted this file system while we were doing the switcheroo,
* or a reopen of z_os failed then just bail out now.
*/
if (!unmounting && (zsb->z_unmounted || zsb->z_os == NULL)) {
rw_exit(&zsb->z_teardown_inactive_lock);
rrw_exit(&zsb->z_teardown_lock, FTAG);
return (EIO);
}
/*
* At this point there are no VFS ops active, and any new VFS ops
* will fail with EIO since we have z_teardown_lock for writer (only
* relevant for forced unmount).
*
* Release all holds on dbufs.
*/
mutex_enter(&zsb->z_znodes_lock);
for (zp = list_head(&zsb->z_all_znodes); zp != NULL;
zp = list_next(&zsb->z_all_znodes, zp)) {
if (zp->z_sa_hdl) {
ASSERT(atomic_read(&ZTOI(zp)->i_count) > 0);
zfs_znode_dmu_fini(zp);
}
}
mutex_exit(&zsb->z_znodes_lock);
/*
* If we are unmounting, set the unmounted flag and let new VFS ops
* unblock. zfs_inactive will have the unmounted behavior, and all
* other VFS ops will fail with EIO.
*/
if (unmounting) {
zsb->z_unmounted = B_TRUE;
rrw_exit(&zsb->z_teardown_lock, FTAG);
rw_exit(&zsb->z_teardown_inactive_lock);
}
/*
* z_os will be NULL if there was an error in attempting to reopen
* zsb, so just return as the properties had already been
*
* unregistered and cached data had been evicted before.
*/
if (zsb->z_os == NULL)
return (0);
/*
* Unregister properties.
*/
zfs_unregister_callbacks(zsb);
/*
* Evict cached data
*/
if (dsl_dataset_is_dirty(dmu_objset_ds(zsb->z_os)) &&
!zfs_is_readonly(zsb))
txg_wait_synced(dmu_objset_pool(zsb->z_os), 0);
dmu_objset_evict_dbufs(zsb->z_os);
return (0);
}
/*
* Teardown the zfsvfs::z_os.
*
* Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
* and 'z_teardown_inactive_lock' held.
*/
static int
zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
{
znode_t *zp;
rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
if (!unmounting) {
/*
* We purge the parent filesystem's vfsp as the parent
* filesystem and all of its snapshots have their vnode's
* v_vfsp set to the parent's filesystem's vfsp. Note,
* 'z_parent' is self referential for non-snapshots.
*/
(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
#ifdef FREEBSD_NAMECACHE
cache_purgevfs(zfsvfs->z_parent->z_vfs);
#endif
}
/*
* Close the zil. NB: Can't close the zil while zfs_inactive
* threads are blocked as zil_close can call zfs_inactive.
*/
if (zfsvfs->z_log) {
zil_close(zfsvfs->z_log);
zfsvfs->z_log = NULL;
}
rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
/*
* If we are not unmounting (ie: online recv) and someone already
* unmounted this file system while we were doing the switcheroo,
* or a reopen of z_os failed then just bail out now.
*/
if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
rw_exit(&zfsvfs->z_teardown_inactive_lock);
rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
return (EIO);
}
/*
* At this point there are no vops active, and any new vops will
* fail with EIO since we have z_teardown_lock for writer (only
* relavent for forced unmount).
*
* Release all holds on dbufs.
*/
mutex_enter(&zfsvfs->z_znodes_lock);
for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
zp = list_next(&zfsvfs->z_all_znodes, zp))
if (zp->z_dbuf) {
ASSERT(ZTOV(zp)->v_count >= 0);
zfs_znode_dmu_fini(zp);
}
mutex_exit(&zfsvfs->z_znodes_lock);
/*
* If we are unmounting, set the unmounted flag and let new vops
* unblock. zfs_inactive will have the unmounted behavior, and all
* other vops will fail with EIO.
*/
if (unmounting) {
zfsvfs->z_unmounted = B_TRUE;
rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
rw_exit(&zfsvfs->z_teardown_inactive_lock);
#ifdef __FreeBSD__
/*
* Some znodes might not be fully reclaimed, wait for them.
*/
mutex_enter(&zfsvfs->z_znodes_lock);
while (list_head(&zfsvfs->z_all_znodes) != NULL) {
msleep(zfsvfs, &zfsvfs->z_znodes_lock, 0,
"zteardown", 0);
}
mutex_exit(&zfsvfs->z_znodes_lock);
#endif
}
/*
* z_os will be NULL if there was an error in attempting to reopen
* zfsvfs, so just return as the properties had already been
* unregistered and cached data had been evicted before.
*/
if (zfsvfs->z_os == NULL)
return (0);
/*
* Unregister properties.
*/
zfs_unregister_callbacks(zfsvfs);
/*
* Evict cached data
*/
if (dmu_objset_evict_dbufs(zfsvfs->z_os)) {
txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
(void) dmu_objset_evict_dbufs(zfsvfs->z_os);
}
return (0);
}
void
zfs_rmnode(znode_t *zp)
{
zfs_sb_t *zsb = ZTOZSB(zp);
objset_t *os = zsb->z_os;
znode_t *xzp = NULL;
dmu_tx_t *tx;
uint64_t acl_obj;
uint64_t xattr_obj;
uint64_t count;
int error;
ASSERT(zp->z_links == 0);
ASSERT(atomic_read(&ZTOI(zp)->i_count) == 0);
/*
* If this is an attribute directory, purge its contents.
*/
if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) {
error = zap_count(os, zp->z_id, &count);
if (error) {
zfs_znode_dmu_fini(zp);
return;
}
if (count > 0) {
taskq_t *taskq;
/*
* There are still directory entries in this xattr
* directory. Let zfs_unlinked_drain() deal with
* them to avoid deadlocking this process in the
* zfs_purgedir()->zfs_zget()->ilookup() callpath
* on the xattr inode's I_FREEING bit.
*/
taskq = dsl_pool_iput_taskq(dmu_objset_pool(os));
taskq_dispatch(taskq, (task_func_t *)
zfs_unlinked_drain, zsb, TQ_SLEEP);
zfs_znode_dmu_fini(zp);
return;
}
}
/*
* Free up all the data in the file.
*/
error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
if (error) {
/*
* Not enough space. Leave the file in the unlinked set.
*/
zfs_znode_dmu_fini(zp);
return;
}
/*
* If the file has extended attributes, we're going to unlink
* the xattr dir.
*/
error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zsb),
&xattr_obj, sizeof (xattr_obj));
if (error == 0 && xattr_obj) {
error = zfs_zget(zsb, xattr_obj, &xzp);
ASSERT(error == 0);
}
acl_obj = zfs_external_acl(zp);
/*
* Set up the final transaction.
*/
tx = dmu_tx_create(os);
dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, FALSE, NULL);
if (xzp) {
dmu_tx_hold_zap(tx, zsb->z_unlinkedobj, TRUE, NULL);
dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
}
if (acl_obj)
dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
zfs_sa_upgrade_txholds(tx, zp);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
/*
* Not enough space to delete the file. Leave it in the
* unlinked set, leaking it until the fs is remounted (at
* which point we'll call zfs_unlinked_drain() to process it).
*/
dmu_tx_abort(tx);
zfs_znode_dmu_fini(zp);
goto out;
}
if (xzp) {
ASSERT(error == 0);
mutex_enter(&xzp->z_lock);
xzp->z_unlinked = B_TRUE; /* mark xzp for deletion */
xzp->z_links = 0; /* no more links to it */
VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zsb),
&xzp->z_links, sizeof (xzp->z_links), tx));
mutex_exit(&xzp->z_lock);
zfs_unlinked_add(xzp, tx);
}
/* Remove this znode from the unlinked set */
VERIFY3U(0, ==,
zap_remove_int(zsb->z_os, zsb->z_unlinkedobj, zp->z_id, tx));
zfs_znode_delete(zp, tx);
dmu_tx_commit(tx);
out:
if (xzp)
iput(ZTOI(xzp));
}
/*
* This function is either called directly from reclaim, or in a delayed
* manner, so the value of zp->z_vnode may be NULL.
*/
void
zfs_rmnode(znode_t *zp)
{
zfsvfs_t *zfsvfs = zp->z_zfsvfs;
objset_t *os = zfsvfs->z_os;
znode_t *xzp = NULL;
dmu_tx_t *tx;
uint64_t acl_obj;
uint64_t xattr_obj;
int error;
ASSERT(zp->z_links == 0);
/*
* If this is an attribute directory, purge its contents.
*/
if (IFTOVT((mode_t)zp->z_mode) == VDIR &&
(zp->z_pflags & ZFS_XATTR)) {
if (!ZTOV(zp) || zfs_purgedir(zp) != 0) {
/*
* Not enough space to delete some xattrs.
* Leave it in the unlinked set.
*/
zfs_znode_dmu_fini(zp);
return;
}
}
/*
* Free up all the data in the file.
*/
error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
if (error) {
/*
* Not enough space. Leave the file in the unlinked set.
*/
zfs_znode_dmu_fini(zp);
//zfs_znode_free(zp);
return;
}
/*
* If the file has extended attributes, we're going to unlink
* the xattr dir.
*/
error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
&xattr_obj, sizeof (xattr_obj));
if (error == 0 && xattr_obj) {
error = zfs_zget(zfsvfs, xattr_obj, &xzp);
ASSERT(error == 0);
}
acl_obj = zfs_external_acl(zp);
/*
* Set up the final transaction.
*/
tx = dmu_tx_create(os);
dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
if (xzp) {
dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
}
if (acl_obj)
dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
zfs_sa_upgrade_txholds(tx, zp);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
/*
* Not enough space to delete the file. Leave it in the
* unlinked set, leaking it until the fs is remounted (at
* which point we'll call zfs_unlinked_drain() to process it).
*/
dmu_tx_abort(tx);
zfs_znode_dmu_fini(zp);
//zfs_znode_free(zp);
goto out;
}
if (xzp) {
ASSERT(error == 0);
mutex_enter(&xzp->z_lock);
xzp->z_unlinked = B_TRUE; /* mark xzp for deletion */
xzp->z_links = 0; /* no more links to it */
VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
&xzp->z_links, sizeof (xzp->z_links), tx));
mutex_exit(&xzp->z_lock);
zfs_unlinked_add(xzp, tx);
}
/* Remove this znode from the unlinked set */
VERIFY3U(0, ==,
zap_remove_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
zfs_znode_delete(zp, tx);
dmu_tx_commit(tx);
out:
if (xzp)
//VN_RELE(ZTOV(xzp)); // async
VN_RELE_ASYNC(ZTOV(xzp), dsl_pool_vnrele_taskq(dmu_objset_pool(zfsvfs->z_os)));
}
int
zvol_strategy(buf_t *bp)
{
zvol_state_t *zv = ddi_get_soft_state(zvol_state, getminor(bp->b_edev));
uint64_t off, volsize;
size_t size, resid;
char *addr;
objset_t *os;
int error = 0;
int sync;
int reading;
int txg_sync_needed = B_FALSE;
if (zv == NULL) {
bioerror(bp, ENXIO);
biodone(bp);
return (0);
}
if (getminor(bp->b_edev) == 0) {
bioerror(bp, EINVAL);
biodone(bp);
return (0);
}
if (zv->zv_readonly && !(bp->b_flags & B_READ)) {
bioerror(bp, EROFS);
biodone(bp);
return (0);
}
off = ldbtob(bp->b_blkno);
volsize = zv->zv_volsize;
os = zv->zv_objset;
ASSERT(os != NULL);
sync = !(bp->b_flags & B_ASYNC) && !(zil_disable);
bp_mapin(bp);
addr = bp->b_un.b_addr;
resid = bp->b_bcount;
/*
* There must be no buffer changes when doing a dmu_sync() because
* we can't change the data whilst calculating the checksum.
* A better approach than a per zvol rwlock would be to lock ranges.
*/
reading = bp->b_flags & B_READ;
if (reading || resid <= zvol_immediate_write_sz)
rw_enter(&zv->zv_dslock, RW_READER);
else
rw_enter(&zv->zv_dslock, RW_WRITER);
while (resid != 0 && off < volsize) {
size = MIN(resid, 1UL << 20); /* cap at 1MB per tx */
if (size > volsize - off) /* don't write past the end */
size = volsize - off;
if (reading) {
error = dmu_read(os, ZVOL_OBJ, off, size, addr);
} else {
dmu_tx_t *tx = dmu_tx_create(os);
dmu_tx_hold_write(tx, ZVOL_OBJ, off, size);
error = dmu_tx_assign(tx, TXG_WAIT);
if (error) {
dmu_tx_abort(tx);
} else {
dmu_write(os, ZVOL_OBJ, off, size, addr, tx);
if (sync) {
/* use the ZIL to commit this write */
if (zvol_log_write(zv, tx, off, size,
addr) != 0) {
txg_sync_needed = B_TRUE;
}
}
dmu_tx_commit(tx);
}
}
if (error)
break;
off += size;
addr += size;
resid -= size;
}
rw_exit(&zv->zv_dslock);
if ((bp->b_resid = resid) == bp->b_bcount)
bioerror(bp, off > volsize ? EINVAL : error);
biodone(bp);
if (sync) {
if (txg_sync_needed)
txg_wait_synced(dmu_objset_pool(os), 0);
else
zil_commit(zv->zv_zilog, UINT64_MAX, 0);
}
return (0);
}
int
zfs_register_callbacks(zfs_sb_t *zsb)
{
struct dsl_dataset *ds = NULL;
objset_t *os = zsb->z_os;
zfs_mntopts_t *zmo = zsb->z_mntopts;
int error = 0;
ASSERT(zsb);
ASSERT(zmo);
/*
* The act of registering our callbacks will destroy any mount
* options we may have. In order to enable temporary overrides
* of mount options, we stash away the current values and
* restore them after we register the callbacks.
*/
if (zfs_is_readonly(zsb) || !spa_writeable(dmu_objset_spa(os))) {
zmo->z_do_readonly = B_TRUE;
zmo->z_readonly = B_TRUE;
}
/*
* Register property callbacks.
*
* It would probably be fine to just check for i/o error from
* the first prop_register(), but I guess I like to go
* overboard...
*/
ds = dmu_objset_ds(os);
dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
error = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zsb);
dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
if (error)
goto unregister;
/*
* Invoke our callbacks to restore temporary mount options.
*/
if (zmo->z_do_readonly)
readonly_changed_cb(zsb, zmo->z_readonly);
if (zmo->z_do_setuid)
setuid_changed_cb(zsb, zmo->z_setuid);
if (zmo->z_do_exec)
exec_changed_cb(zsb, zmo->z_exec);
if (zmo->z_do_devices)
devices_changed_cb(zsb, zmo->z_devices);
if (zmo->z_do_xattr)
xattr_changed_cb(zsb, zmo->z_xattr);
if (zmo->z_do_atime)
atime_changed_cb(zsb, zmo->z_atime);
if (zmo->z_do_relatime)
relatime_changed_cb(zsb, zmo->z_relatime);
if (zmo->z_do_nbmand)
nbmand_changed_cb(zsb, zmo->z_nbmand);
return (0);
unregister:
/*
* We may attempt to unregister some callbacks that are not
* registered, but this is OK; it will simply return ENOMSG,
* which we will ignore.
*/
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
atime_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RELATIME),
relatime_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
xattr_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
blksz_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
readonly_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
devices_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
setuid_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
exec_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
snapdir_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLTYPE),
acltype_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
acl_inherit_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
vscan_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_NBMAND),
nbmand_changed_cb, zsb);
return (error);
}
/*
* Teardown the zfs_sb_t.
*
* Note, if 'unmounting' if FALSE, we return with the 'z_teardown_lock'
* and 'z_teardown_inactive_lock' held.
*/
int
zfs_sb_teardown(zfs_sb_t *zsb, boolean_t unmounting)
{
znode_t *zp;
/*
* If someone has not already unmounted this file system,
* drain the iput_taskq to ensure all active references to the
* zfs_sb_t have been handled only then can it be safely destroyed.
*/
if (zsb->z_os) {
/*
* If we're unmounting we have to wait for the list to
* drain completely.
*
* If we're not unmounting there's no guarantee the list
* will drain completely, but iputs run from the taskq
* may add the parents of dir-based xattrs to the taskq
* so we want to wait for these.
*
* We can safely read z_nr_znodes without locking because the
* VFS has already blocked operations which add to the
* z_all_znodes list and thus increment z_nr_znodes.
*/
int round = 0;
while (zsb->z_nr_znodes > 0) {
taskq_wait_outstanding(dsl_pool_iput_taskq(
dmu_objset_pool(zsb->z_os)), 0);
if (++round > 1 && !unmounting)
break;
}
}
rrm_enter(&zsb->z_teardown_lock, RW_WRITER, FTAG);
if (!unmounting) {
/*
* We purge the parent filesystem's super block as the
* parent filesystem and all of its snapshots have their
* inode's super block set to the parent's filesystem's
* super block. Note, 'z_parent' is self referential
* for non-snapshots.
*/
shrink_dcache_sb(zsb->z_parent->z_sb);
}
/*
* Close the zil. NB: Can't close the zil while zfs_inactive
* threads are blocked as zil_close can call zfs_inactive.
*/
if (zsb->z_log) {
zil_close(zsb->z_log);
zsb->z_log = NULL;
}
rw_enter(&zsb->z_teardown_inactive_lock, RW_WRITER);
/*
* If we are not unmounting (ie: online recv) and someone already
* unmounted this file system while we were doing the switcheroo,
* or a reopen of z_os failed then just bail out now.
*/
if (!unmounting && (zsb->z_unmounted || zsb->z_os == NULL)) {
rw_exit(&zsb->z_teardown_inactive_lock);
rrm_exit(&zsb->z_teardown_lock, FTAG);
return (SET_ERROR(EIO));
}
/*
* At this point there are no VFS ops active, and any new VFS ops
* will fail with EIO since we have z_teardown_lock for writer (only
* relevant for forced unmount).
*
* Release all holds on dbufs.
*/
if (!unmounting) {
mutex_enter(&zsb->z_znodes_lock);
for (zp = list_head(&zsb->z_all_znodes); zp != NULL;
zp = list_next(&zsb->z_all_znodes, zp)) {
if (zp->z_sa_hdl)
zfs_znode_dmu_fini(zp);
}
mutex_exit(&zsb->z_znodes_lock);
}
/*
* If we are unmounting, set the unmounted flag and let new VFS ops
* unblock. zfs_inactive will have the unmounted behavior, and all
* other VFS ops will fail with EIO.
*/
if (unmounting) {
zsb->z_unmounted = B_TRUE;
rrm_exit(&zsb->z_teardown_lock, FTAG);
rw_exit(&zsb->z_teardown_inactive_lock);
}
/*
* z_os will be NULL if there was an error in attempting to reopen
* zsb, so just return as the properties had already been
*
* unregistered and cached data had been evicted before.
*/
if (zsb->z_os == NULL)
return (0);
/*
* Unregister properties.
*/
zfs_unregister_callbacks(zsb);
/*
* Evict cached data
*/
if (dsl_dataset_is_dirty(dmu_objset_ds(zsb->z_os)) &&
!zfs_is_readonly(zsb))
txg_wait_synced(dmu_objset_pool(zsb->z_os), 0);
dmu_objset_evict_dbufs(zsb->z_os);
return (0);
}
static void
zfs_objset_close(zfsvfs_t *zfsvfs)
{
zfs_delete_t *zd = &zfsvfs->z_delete_head;
znode_t *zp, *nextzp;
objset_t *os = zfsvfs->z_os;
/*
* Stop all delete threads.
*/
(void) zfs_delete_thread_target(zfsvfs, 0);
/*
* For forced unmount, at this point all vops except zfs_inactive
* are erroring EIO. We need to now suspend zfs_inactive threads
* while we are freeing dbufs before switching zfs_inactive
* to use behaviour without a objset.
*/
rw_enter(&zfsvfs->z_um_lock, RW_WRITER);
/*
* Release all delete in progress znodes
* They will be processed when the file system remounts.
*/
mutex_enter(&zd->z_mutex);
while (zp = list_head(&zd->z_znodes)) {
list_remove(&zd->z_znodes, zp);
zp->z_dbuf_held = 0;
dmu_buf_rele(zp->z_dbuf, NULL);
}
mutex_exit(&zd->z_mutex);
/*
* Release all holds on dbufs
* Note, although we have stopped all other vop threads and
* zfs_inactive(), the dmu can callback via znode_pageout_func()
* which can zfs_znode_free() the znode.
* So we lock z_all_znodes; search the list for a held
* dbuf; drop the lock (we know zp can't disappear if we hold
* a dbuf lock; then regrab the lock and restart.
*/
mutex_enter(&zfsvfs->z_znodes_lock);
for (zp = list_head(&zfsvfs->z_all_znodes); zp; zp = nextzp) {
nextzp = list_next(&zfsvfs->z_all_znodes, zp);
if (zp->z_dbuf_held) {
/* dbufs should only be held when force unmounting */
zp->z_dbuf_held = 0;
mutex_exit(&zfsvfs->z_znodes_lock);
dmu_buf_rele(zp->z_dbuf, NULL);
/* Start again */
mutex_enter(&zfsvfs->z_znodes_lock);
nextzp = list_head(&zfsvfs->z_all_znodes);
}
}
mutex_exit(&zfsvfs->z_znodes_lock);
/*
* Unregister properties.
*/
if (!dmu_objset_is_snapshot(os))
zfs_unregister_callbacks(zfsvfs);
/*
* Switch zfs_inactive to behaviour without an objset.
* It just tosses cached pages and frees the znode & vnode.
* Then re-enable zfs_inactive threads in that new behaviour.
*/
zfsvfs->z_unmounted2 = B_TRUE;
rw_exit(&zfsvfs->z_um_lock); /* re-enable any zfs_inactive threads */
/*
* Close the zil. Can't close the zil while zfs_inactive
* threads are blocked as zil_close can call zfs_inactive.
*/
if (zfsvfs->z_log) {
zil_close(zfsvfs->z_log);
zfsvfs->z_log = NULL;
}
/*
* Evict all dbufs so that cached znodes will be freed
*/
if (dmu_objset_evict_dbufs(os, 1)) {
txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
(void) dmu_objset_evict_dbufs(os, 0);
}
/*
* Finally close the objset
*/
dmu_objset_close(os);
/*
* We can now safely destroy the '.zfs' directory node.
*/
if (zfsvfs->z_ctldir != NULL)
zfsctl_destroy(zfsvfs);
}
/*ARGSUSED*/
static int
zfs_vfs_unmount(struct mount *mp, int mntflags, vfs_context_t context)
{
zfsvfs_t *zfsvfs = vfs_fsprivate(mp);
objset_t *os = zfsvfs->z_os;
znode_t *zp, *nextzp;
int ret, i;
int flags;
/*XXX NOEL: delegation admin stuffs, add back if we use delg. admin */
#if 0
ret = 0; /* UNDEFINED: secpolicy_fs_unmount(cr, vfsp); */
if (ret) {
ret = dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
ZFS_DELEG_PERM_MOUNT, cr);
if (ret)
return (ret);
}
/*
* We purge the parent filesystem's vfsp as the parent filesystem
* and all of its snapshots have their vnode's v_vfsp set to the
* parent's filesystem's vfsp. Note, 'z_parent' is self
* referential for non-snapshots.
*/
(void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
#endif
/*
* Unmount any snapshots mounted under .zfs before unmounting the
* dataset itself.
*/
#if 0
if (zfsvfs->z_ctldir != NULL &&
(ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
return (ret);
#endif
flags = SKIPSYSTEM;
if (mntflags & MNT_FORCE)
flags |= FORCECLOSE;
ret = vflush(mp, NULLVP, flags);
/*
* Mac OS X needs a file system modify time
*
* We use the mtime of the "com.apple.system.mtime"
* extended attribute, which is associated with the
* file system root directory.
*
* Here we need to release the ref we took on z_mtime_vp during mount.
*/
if ((ret == 0) || (mntflags & MNT_FORCE)) {
if (zfsvfs->z_mtime_vp != NULL) {
struct vnode *mvp;
mvp = zfsvfs->z_mtime_vp;
zfsvfs->z_mtime_vp = NULL;
if (vnode_get(mvp) == 0) {
vnode_rele(mvp);
vnode_recycle(mvp);
vnode_put(mvp);
}
}
}
if (!(mntflags & MNT_FORCE)) {
/*
* Check the number of active vnodes in the file system.
* Our count is maintained in the vfs structure, but the
* number is off by 1 to indicate a hold on the vfs
* structure itself.
*
* The '.zfs' directory maintains a reference of its
* own, and any active references underneath are
* reflected in the vnode count.
*/
if (ret)
return (EBUSY);
#if 0
if (zfsvfs->z_ctldir == NULL) {
if (vfsp->vfs_count > 1)
return (EBUSY);
} else {
if (vfsp->vfs_count > 2 ||
zfsvfs->z_ctldir->v_count > 1) {
return (EBUSY);
}
}
#endif
}
rw_enter(&zfsvfs->z_unmount_lock, RW_WRITER);
rw_enter(&zfsvfs->z_unmount_inactive_lock, RW_WRITER);
/*
* At this point there are no vops active, and any new vops will
* fail with EIO since we have z_unmount_lock for writer (only
* relavent for forced unmount).
*
* Release all holds on dbufs.
* Note, the dmu can still callback via znode_pageout_func()
* which can zfs_znode_free() the znode. So we lock
* z_all_znodes; search the list for a held dbuf; drop the lock
* (we know zp can't disappear if we hold a dbuf lock) then
* regrab the lock and restart.
*/
mutex_enter(&zfsvfs->z_znodes_lock);
for (zp = list_head(&zfsvfs->z_all_znodes); zp; zp = nextzp) {
nextzp = list_next(&zfsvfs->z_all_znodes, zp);
if (zp->z_dbuf_held) {
/* dbufs should only be held when force unmounting */
zp->z_dbuf_held = 0;
mutex_exit(&zfsvfs->z_znodes_lock);
dmu_buf_rele(zp->z_dbuf, NULL);
/* Start again */
mutex_enter(&zfsvfs->z_znodes_lock);
nextzp = list_head(&zfsvfs->z_all_znodes);
}
}
mutex_exit(&zfsvfs->z_znodes_lock);
/*
* Set the unmounted flag and let new vops unblock.
* zfs_inactive will have the unmounted behavior, and all other
* vops will fail with EIO.
*/
zfsvfs->z_unmounted = B_TRUE;
rw_exit(&zfsvfs->z_unmount_lock);
rw_exit(&zfsvfs->z_unmount_inactive_lock);
/*
* Unregister properties.
*/
#ifndef __APPLE__
if (!dmu_objset_is_snapshot(os))
zfs_unregister_callbacks(zfsvfs);
#endif
/*
* Close the zil. NB: Can't close the zil while zfs_inactive
* threads are blocked as zil_close can call zfs_inactive.
*/
if (zfsvfs->z_log) {
zil_close(zfsvfs->z_log);
zfsvfs->z_log = NULL;
}
/*
* Evict all dbufs so that cached znodes will be freed
*/
if (dmu_objset_evict_dbufs(os, B_TRUE)) {
txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
(void) dmu_objset_evict_dbufs(os, B_FALSE);
}
/*
* Finally close the objset
*/
dmu_objset_close(os);
/*
* We can now safely destroy the '.zfs' directory node.
*/
#if 0
if (zfsvfs->z_ctldir != NULL)
zfsctl_destroy(zfsvfs);
#endif
/*
* Note that this work is normally done in zfs_freevfs, but since
* there is no VOP_FREEVFS in OSX, we free VFS items here
*/
OSDecrementAtomic((SInt32 *)&zfs_active_fs_count);
for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
mutex_destroy(&zfsvfs->z_hold_mtx[i]);
mutex_destroy(&zfsvfs->z_znodes_lock);
list_destroy(&zfsvfs->z_all_znodes);
rw_destroy(&zfsvfs->z_unmount_lock);
rw_destroy(&zfsvfs->z_unmount_inactive_lock);
return (0);
}
struct vnode* vnode_getparent(struct vnode *vp); /* sys/vnode_internal.h */
static int
zfs_vget_internal(zfsvfs_t *zfsvfs, ino64_t ino, struct vnode **vpp)
{
struct vnode *vp;
struct vnode *dvp = NULL;
znode_t *zp;
int error;
*vpp = NULL;
/*
* On Mac OS X we always export the root directory id as 2
* and its parent as 1
*/
if (ino == 2 || ino == 1)
ino = zfsvfs->z_root;
if ((error = zfs_zget(zfsvfs, ino, &zp)))
goto out;
/* Don't expose EA objects! */
if (zp->z_phys->zp_flags & ZFS_XATTR) {
vnode_put(ZTOV(zp));
error = ENOENT;
goto out;
}
*vpp = vp = ZTOV(zp);
if (vnode_isvroot(vp))
goto out;
/*
* If this znode didn't just come from the cache then
* it won't have a valid identity (parent and name).
*
* Manually fix its identity here (normally done by namei lookup).
*/
if ((dvp = vnode_getparent(vp)) == NULL) {
if (zp->z_phys->zp_parent != 0 &&
zfs_vget_internal(zfsvfs, zp->z_phys->zp_parent, &dvp)) {
goto out;
}
if ( vnode_isdir(dvp) ) {
char objname[ZAP_MAXNAMELEN]; /* 256 bytes */
int flags = VNODE_UPDATE_PARENT;
/* Look for znode's name in its parent's zap */
if ( zap_value_search(zfsvfs->z_os,
zp->z_phys->zp_parent,
zp->z_id,
ZFS_DIRENT_OBJ(-1ULL),
objname) == 0 ) {
flags |= VNODE_UPDATE_NAME;
}
/* Update the znode's parent and name */
vnode_update_identity(vp, dvp, objname, 0, 0, flags);
}
}
/* All done with znode's parent */
vnode_put(dvp);
out:
return (error);
}
/*
* Get a vnode from a file id (ignoring the generation)
*
* Use by NFS Server (readdirplus) and VFS (build_path)
*/
static int
zfs_vfs_vget(struct mount *mp, ino64_t ino, struct vnode **vpp, __unused vfs_context_t context)
{
zfsvfs_t *zfsvfs = vfs_fsprivate(mp);
int error;
ZFS_ENTER(zfsvfs);
/*
* On Mac OS X we always export the root directory id as 2.
* So we don't expect to see the real root directory id
* from zfs_vfs_vget KPI (unless of course the real id was
* already 2).
*/
if ((ino == zfsvfs->z_root) && (zfsvfs->z_root != 2)) {
ZFS_EXIT(zfsvfs);
return (ENOENT);
}
error = zfs_vget_internal(zfsvfs, ino, vpp);
ZFS_EXIT(zfsvfs);
return (error);
}
/*
* dsl_crypto_key_unload
*
* Remove the key from the in memory keystore.
*
* First we have to remove the minor node for a ZVOL or unmount
* the filesystem. This is so that we flush all pending IO for it to disk
* so we won't need to encrypt anything with this key. Anything in flight
* should already have a lock on the keys it needs.
* We can't assume that userland has already successfully unmounted the
* dataset though in many cases it will have.
*
* If the key can't be removed return the failure back to our caller.
*/
int
dsl_crypto_key_unload(const char *dsname)
{
dsl_dataset_t *ds;
objset_t *os;
int error;
spa_t *spa;
dsl_pool_t *dp;
#ifdef _KERNEL
dmu_objset_type_t os_type;
//vfs_t *vfsp;
struct vfsmount *vfsp;
#endif /* _KERNEL */
error = dsl_pool_hold(dsname, FTAG, &dp);
if (error != 0)
return (error);
/* XXX - should we use own_exclusive() here? */
if ((error = dsl_dataset_hold(dp, dsname, FTAG, &ds)) != 0) {
dsl_pool_rele(dp, FTAG);
return (error);
}
if ((error = dmu_objset_from_ds(ds, &os)) != 0) {
dsl_dataset_rele(ds, FTAG);
dsl_pool_rele(dp, FTAG);
return (error);
}
#ifdef _KERNEL
/*
* Make sure that the device node has gone for ZVOLs
* and that filesystems are umounted.
*/
#if 0 // FIXME
os_type = dmu_objset_type(os);
if (os_type == DMU_OST_ZVOL) {
error = zvol_remove_minor(dsname);
if (error == ENXIO)
error = 0;
} else if (os_type == DMU_OST_ZFS) {
vfsp = zfs_get_vfs(dsname);
if (vfsp != NULL) {
error = vn_vfswlock(vfsp->vfs_vnodecovered);
VFS_RELE(vfsp);
if (error == 0)
error = dounmount(vfsp, 0, CRED());
}
}
if (error != 0) {
dsl_dataset_rele(ds, FTAG);
return (error);
}
#endif
#endif /* _KERNEL */
/*
* Make sure all dbufs are synced.
*
* It is essential for encrypted datasets to ensure that
* there is no further pending IO before removing the key.
*/
if (dmu_objset_is_dirty(os, 0)) // FIXME, 0?
txg_wait_synced(dmu_objset_pool(os), 0);
dmu_objset_evict_dbufs(os);
spa = dsl_dataset_get_spa(ds);
error = zcrypt_keystore_remove(spa, ds->ds_object);
dsl_dataset_rele(ds, FTAG);
dsl_pool_rele(dp, FTAG);
return (error);
}
int
zfs_register_callbacks(zfs_sb_t *zsb)
{
struct dsl_dataset *ds = NULL;
objset_t *os = zsb->z_os;
boolean_t do_readonly = B_FALSE;
int error = 0;
if (zfs_is_readonly(zsb) || !spa_writeable(dmu_objset_spa(os)))
do_readonly = B_TRUE;
/*
* Register property callbacks.
*
* It would probably be fine to just check for i/o error from
* the first prop_register(), but I guess I like to go
* overboard...
*/
ds = dmu_objset_ds(os);
dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
error = dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zsb);
error = error ? error : dsl_prop_register(ds,
zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zsb);
dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
if (error)
goto unregister;
if (do_readonly)
readonly_changed_cb(zsb, B_TRUE);
return (0);
unregister:
/*
* We may attempt to unregister some callbacks that are not
* registered, but this is OK; it will simply return ENOMSG,
* which we will ignore.
*/
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ATIME),
atime_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_XATTR),
xattr_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_RECORDSIZE),
blksz_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_READONLY),
readonly_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_DEVICES),
devices_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SETUID),
setuid_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_EXEC),
exec_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_SNAPDIR),
snapdir_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_ACLINHERIT),
acl_inherit_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_VSCAN),
vscan_changed_cb, zsb);
(void) dsl_prop_unregister(ds, zfs_prop_to_name(ZFS_PROP_NBMAND),
nbmand_changed_cb, zsb);
return (error);
}