void
rrm_exit(rrmlock_t *rrl, void *tag)
{
int i;
if (rrl->locks[0].rr_writer == curthread) {
for (i = 0; i < RRM_NUM_LOCKS; i++)
rrw_exit(&rrl->locks[i], tag);
} else {
rrw_exit(&rrl->locks[RRM_TD_LOCK()], tag);
}
}
int
dmu_objset_from_ds(dsl_dataset_t *ds, objset_t **osp)
{
int err = 0;
/*
* We shouldn't be doing anything with dsl_dataset_t's unless the
* pool_config lock is held, or the dataset is long-held.
*/
ASSERT(dsl_pool_config_held(ds->ds_dir->dd_pool) ||
dsl_dataset_long_held(ds));
mutex_enter(&ds->ds_opening_lock);
if (ds->ds_objset == NULL) {
objset_t *os;
rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
err = dmu_objset_open_impl(dsl_dataset_get_spa(ds),
ds, dsl_dataset_get_blkptr(ds), &os);
rrw_exit(&ds->ds_bp_rwlock, FTAG);
if (err == 0) {
mutex_enter(&ds->ds_lock);
ASSERT(ds->ds_objset == NULL);
ds->ds_objset = os;
mutex_exit(&ds->ds_lock);
}
}
*osp = ds->ds_objset;
mutex_exit(&ds->ds_opening_lock);
return (err);
}
/*
* Called in syncing context to execute the synctask.
*/
void
dsl_sync_task_sync(dsl_sync_task_t *dst, dmu_tx_t *tx)
{
dsl_pool_t *dp = dst->dst_pool;
ASSERT0(dst->dst_error);
/*
* Check for sufficient space.
*
* When the sync task was created, the caller specified the
* type of space checking required. See the comment in
* zfs_space_check_t for details on the semantics of each
* type of space checking.
*
* We just check against what's on-disk; we don't want any
* in-flight accounting to get in our way, because open context
* may have already used up various in-core limits
* (arc_tempreserve, dsl_pool_tempreserve).
*/
if (dst->dst_space_check != ZFS_SPACE_CHECK_NONE) {
uint64_t quota = dsl_pool_adjustedsize(dp,
dst->dst_space_check == ZFS_SPACE_CHECK_RESERVED) -
metaslab_class_get_deferred(spa_normal_class(dp->dp_spa));
uint64_t used = dsl_dir_phys(dp->dp_root_dir)->dd_used_bytes;
/* MOS space is triple-dittoed, so we multiply by 3. */
if (dst->dst_space > 0 && used + dst->dst_space * 3 > quota) {
dst->dst_error = SET_ERROR(ENOSPC);
if (dst->dst_nowaiter)
kmem_free(dst, sizeof (*dst));
return;
}
}
/*
* Check for errors by calling checkfunc.
*/
rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
dst->dst_error = dst->dst_checkfunc(dst->dst_arg, tx);
if (dst->dst_error == 0)
dst->dst_syncfunc(dst->dst_arg, tx);
rrw_exit(&dp->dp_config_rwlock, FTAG);
if (dst->dst_nowaiter)
kmem_free(dst, sizeof (*dst));
}
/*
* Reopen zfsvfs_t::z_os and release VOPs.
*/
int
zfs_resume_fs(zfsvfs_t *zfsvfs, const char *osname, int mode)
{
int err;
ASSERT(RRW_WRITE_HELD(&zfsvfs->z_teardown_lock));
ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
err = dmu_objset_open(osname, DMU_OST_ZFS, mode, &zfsvfs->z_os);
if (err) {
zfsvfs->z_os = NULL;
} else {
znode_t *zp;
VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
/*
* Attempt to re-establish all the active znodes with
* their dbufs. If a zfs_rezget() fails, then we'll let
* any potential callers discover that via ZFS_ENTER_VERIFY_VP
* when they try to use their znode.
*/
mutex_enter(&zfsvfs->z_znodes_lock);
for (zp = list_head(&zfsvfs->z_all_znodes); zp;
zp = list_next(&zfsvfs->z_all_znodes, zp)) {
(void) zfs_rezget(zp);
}
mutex_exit(&zfsvfs->z_znodes_lock);
}
/* release the VOPs */
rw_exit(&zfsvfs->z_teardown_inactive_lock);
rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
if (err) {
/*
* Since we couldn't reopen zfsvfs::z_os, force
* unmount this file system.
*/
if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
(void) dounmount(zfsvfs->z_vfs, MS_FORCE, curthread);
}
return (err);
}
/*
* Called in syncing context to execute the synctask.
*/
void
dsl_sync_task_sync(dsl_sync_task_t *dst, dmu_tx_t *tx)
{
dsl_pool_t *dp = dst->dst_pool;
uint64_t quota, used;
ASSERT0(dst->dst_error);
/*
* Check for sufficient space. We just check against what's
* on-disk; we don't want any in-flight accounting to get in our
* way, because open context may have already used up various
* in-core limits (arc_tempreserve, dsl_pool_tempreserve).
*/
quota = dsl_pool_adjustedsize(dp, B_FALSE) -
metaslab_class_get_deferred(spa_normal_class(dp->dp_spa));
used = dp->dp_root_dir->dd_phys->dd_used_bytes;
/* MOS space is triple-dittoed, so we multiply by 3. */
if (dst->dst_space > 0 && used + dst->dst_space * 3 > quota) {
dst->dst_error = SET_ERROR(ENOSPC);
if (dst->dst_nowaiter)
kmem_free(dst, sizeof (*dst));
return;
}
/*
* Check for errors by calling checkfunc.
*/
rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
dst->dst_error = dst->dst_checkfunc(dst->dst_arg, tx);
if (dst->dst_error == 0)
dst->dst_syncfunc(dst->dst_arg, tx);
rrw_exit(&dp->dp_config_rwlock, FTAG);
if (dst->dst_nowaiter)
kmem_free(dst, sizeof (*dst));
}
dsl_pool_t *
dsl_pool_create(spa_t *spa, nvlist_t *zplprops, uint64_t txg)
{
int err;
dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
objset_t *os;
dsl_dataset_t *ds;
uint64_t obj;
rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
/* create and open the MOS (meta-objset) */
dp->dp_meta_objset = dmu_objset_create_impl(spa,
NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx);
/* create the pool directory */
err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
ASSERT0(err);
/* Initialize scan structures */
VERIFY0(dsl_scan_init(dp, txg));
/* create and open the root dir */
dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj,
NULL, dp, &dp->dp_root_dir));
/* create and open the meta-objset dir */
(void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
VERIFY0(dsl_pool_open_special_dir(dp,
MOS_DIR_NAME, &dp->dp_mos_dir));
if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
/* create and open the free dir */
(void) dsl_dir_create_sync(dp, dp->dp_root_dir,
FREE_DIR_NAME, tx);
VERIFY0(dsl_pool_open_special_dir(dp,
FREE_DIR_NAME, &dp->dp_free_dir));
/* create and open the free_bplist */
obj = bpobj_alloc(dp->dp_meta_objset, SPA_MAXBLOCKSIZE, tx);
VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0);
VERIFY0(bpobj_open(&dp->dp_free_bpobj,
dp->dp_meta_objset, obj));
}
if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
dsl_pool_create_origin(dp, tx);
/* create the root dataset */
obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, 0, tx);
/* create the root objset */
VERIFY0(dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
os = dmu_objset_create_impl(dp->dp_spa, ds,
dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
#ifdef _KERNEL
zfs_create_fs(os, kcred, zplprops, tx);
#endif
dsl_dataset_rele(ds, FTAG);
dmu_tx_commit(tx);
rrw_exit(&dp->dp_config_rwlock, FTAG);
return (dp);
}
int
dsl_pool_open(dsl_pool_t *dp)
{
int err;
dsl_dir_t *dd;
dsl_dataset_t *ds;
uint64_t obj;
rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
&dp->dp_root_dir_obj);
if (err)
goto out;
err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj,
NULL, dp, &dp->dp_root_dir);
if (err)
goto out;
err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
if (err)
goto out;
if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) {
err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
if (err)
goto out;
err = dsl_dataset_hold_obj(dp, dd->dd_phys->dd_head_dataset_obj,
FTAG, &ds);
if (err == 0) {
err = dsl_dataset_hold_obj(dp,
ds->ds_phys->ds_prev_snap_obj, dp,
&dp->dp_origin_snap);
dsl_dataset_rele(ds, FTAG);
}
dsl_dir_rele(dd, dp);
if (err)
goto out;
}
if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME,
&dp->dp_free_dir);
if (err)
goto out;
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj);
if (err)
goto out;
VERIFY0(bpobj_open(&dp->dp_free_bpobj,
dp->dp_meta_objset, obj));
}
if (spa_feature_is_active(dp->dp_spa,
&spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
&dp->dp_bptree_obj);
if (err != 0)
goto out;
}
if (spa_feature_is_active(dp->dp_spa,
&spa_feature_table[SPA_FEATURE_EMPTY_BPOBJ])) {
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
&dp->dp_empty_bpobj);
if (err != 0)
goto out;
}
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1,
&dp->dp_tmp_userrefs_obj);
if (err == ENOENT)
err = 0;
if (err)
goto out;
err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg);
out:
rrw_exit(&dp->dp_config_rwlock, FTAG);
return (err);
}
dsl_pool_t *
dsl_pool_create(spa_t *spa, nvlist_t *zplprops, dsl_crypto_params_t *dcp,
uint64_t txg)
{
int err;
dsl_pool_t *dp = dsl_pool_open_impl(spa, txg);
dmu_tx_t *tx = dmu_tx_create_assigned(dp, txg);
dsl_dataset_t *ds;
uint64_t obj;
rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
/* create and open the MOS (meta-objset) */
dp->dp_meta_objset = dmu_objset_create_impl(spa,
NULL, &dp->dp_meta_rootbp, DMU_OST_META, tx);
spa->spa_meta_objset = dp->dp_meta_objset;
/* create the pool directory */
err = zap_create_claim(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_OT_OBJECT_DIRECTORY, DMU_OT_NONE, 0, tx);
ASSERT0(err);
/* Initialize scan structures */
VERIFY0(dsl_scan_init(dp, txg));
/* create and open the root dir */
dp->dp_root_dir_obj = dsl_dir_create_sync(dp, NULL, NULL, tx);
VERIFY0(dsl_dir_hold_obj(dp, dp->dp_root_dir_obj,
NULL, dp, &dp->dp_root_dir));
/* create and open the meta-objset dir */
(void) dsl_dir_create_sync(dp, dp->dp_root_dir, MOS_DIR_NAME, tx);
VERIFY0(dsl_pool_open_special_dir(dp,
MOS_DIR_NAME, &dp->dp_mos_dir));
if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
/* create and open the free dir */
(void) dsl_dir_create_sync(dp, dp->dp_root_dir,
FREE_DIR_NAME, tx);
VERIFY0(dsl_pool_open_special_dir(dp,
FREE_DIR_NAME, &dp->dp_free_dir));
/* create and open the free_bplist */
obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx);
VERIFY(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj, tx) == 0);
VERIFY0(bpobj_open(&dp->dp_free_bpobj,
dp->dp_meta_objset, obj));
}
if (spa_version(spa) >= SPA_VERSION_DSL_SCRUB)
dsl_pool_create_origin(dp, tx);
/*
* Some features may be needed when creating the root dataset, so we
* create the feature objects here.
*/
if (spa_version(spa) >= SPA_VERSION_FEATURES)
spa_feature_create_zap_objects(spa, tx);
if (dcp != NULL && dcp->cp_crypt != ZIO_CRYPT_OFF &&
dcp->cp_crypt != ZIO_CRYPT_INHERIT)
spa_feature_enable(spa, SPA_FEATURE_ENCRYPTION, tx);
/* create the root dataset */
obj = dsl_dataset_create_sync_dd(dp->dp_root_dir, NULL, dcp, 0, tx);
/* create the root objset */
VERIFY0(dsl_dataset_hold_obj_flags(dp, obj,
DS_HOLD_FLAG_DECRYPT, FTAG, &ds));
#ifdef _KERNEL
{
objset_t *os;
rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
os = dmu_objset_create_impl(dp->dp_spa, ds,
dsl_dataset_get_blkptr(ds), DMU_OST_ZFS, tx);
rrw_exit(&ds->ds_bp_rwlock, FTAG);
zfs_create_fs(os, kcred, zplprops, tx);
}
#endif
dsl_dataset_rele_flags(ds, DS_HOLD_FLAG_DECRYPT, FTAG);
dmu_tx_commit(tx);
rrw_exit(&dp->dp_config_rwlock, FTAG);
return (dp);
}
int
dsl_pool_open(dsl_pool_t *dp)
{
int err;
dsl_dir_t *dd;
dsl_dataset_t *ds;
uint64_t obj;
rrw_enter(&dp->dp_config_rwlock, RW_WRITER, FTAG);
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_ROOT_DATASET, sizeof (uint64_t), 1,
&dp->dp_root_dir_obj);
if (err)
goto out;
err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj,
NULL, dp, &dp->dp_root_dir);
if (err)
goto out;
err = dsl_pool_open_special_dir(dp, MOS_DIR_NAME, &dp->dp_mos_dir);
if (err)
goto out;
if (spa_version(dp->dp_spa) >= SPA_VERSION_ORIGIN) {
err = dsl_pool_open_special_dir(dp, ORIGIN_DIR_NAME, &dd);
if (err)
goto out;
err = dsl_dataset_hold_obj(dp,
dsl_dir_phys(dd)->dd_head_dataset_obj, FTAG, &ds);
if (err == 0) {
err = dsl_dataset_hold_obj(dp,
dsl_dataset_phys(ds)->ds_prev_snap_obj, dp,
&dp->dp_origin_snap);
dsl_dataset_rele(ds, FTAG);
}
dsl_dir_rele(dd, dp);
if (err)
goto out;
}
if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
err = dsl_pool_open_special_dir(dp, FREE_DIR_NAME,
&dp->dp_free_dir);
if (err)
goto out;
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_FREE_BPOBJ, sizeof (uint64_t), 1, &obj);
if (err)
goto out;
VERIFY0(bpobj_open(&dp->dp_free_bpobj,
dp->dp_meta_objset, obj));
}
if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_OBSOLETE_COUNTS)) {
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_OBSOLETE_BPOBJ, sizeof (uint64_t), 1, &obj);
if (err == 0) {
VERIFY0(bpobj_open(&dp->dp_obsolete_bpobj,
dp->dp_meta_objset, obj));
} else if (err == ENOENT) {
/*
* We might not have created the remap bpobj yet.
*/
err = 0;
} else {
goto out;
}
}
/*
* Note: errors ignored, because the these special dirs, used for
* space accounting, are only created on demand.
*/
(void) dsl_pool_open_special_dir(dp, LEAK_DIR_NAME,
&dp->dp_leak_dir);
if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_ASYNC_DESTROY)) {
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_BPTREE_OBJ, sizeof (uint64_t), 1,
&dp->dp_bptree_obj);
if (err != 0)
goto out;
}
if (spa_feature_is_active(dp->dp_spa, SPA_FEATURE_EMPTY_BPOBJ)) {
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_EMPTY_BPOBJ, sizeof (uint64_t), 1,
&dp->dp_empty_bpobj);
if (err != 0)
goto out;
}
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_TMP_USERREFS, sizeof (uint64_t), 1,
&dp->dp_tmp_userrefs_obj);
if (err == ENOENT)
err = 0;
if (err)
goto out;
err = dsl_scan_init(dp, dp->dp_tx.tx_open_txg);
out:
rrw_exit(&dp->dp_config_rwlock, FTAG);
return (err);
}
/*
* Reopen zfs_sb_t and release VFS ops.
*/
int
zfs_resume_fs(zfs_sb_t *zsb, const char *osname)
{
int err, err2;
ASSERT(RRW_WRITE_HELD(&zsb->z_teardown_lock));
ASSERT(RW_WRITE_HELD(&zsb->z_teardown_inactive_lock));
err = dmu_objset_own(osname, DMU_OST_ZFS, B_FALSE, zsb, &zsb->z_os);
if (err) {
zsb->z_os = NULL;
} else {
znode_t *zp;
uint64_t sa_obj = 0;
err2 = zap_lookup(zsb->z_os, MASTER_NODE_OBJ,
ZFS_SA_ATTRS, 8, 1, &sa_obj);
if ((err || err2) && zsb->z_version >= ZPL_VERSION_SA)
goto bail;
if ((err = sa_setup(zsb->z_os, sa_obj,
zfs_attr_table, ZPL_END, &zsb->z_attr_table)) != 0)
goto bail;
VERIFY(zfs_sb_setup(zsb, B_FALSE) == 0);
zsb->z_rollback_time = jiffies;
/*
* Attempt to re-establish all the active inodes with their
* dbufs. If a zfs_rezget() fails, then we unhash the inode
* and mark it stale. This prevents a collision if a new
* inode/object is created which must use the same inode
* number. The stale inode will be be released when the
* VFS prunes the dentry holding the remaining references
* on the stale inode.
*/
mutex_enter(&zsb->z_znodes_lock);
for (zp = list_head(&zsb->z_all_znodes); zp;
zp = list_next(&zsb->z_all_znodes, zp)) {
err2 = zfs_rezget(zp);
if (err2) {
remove_inode_hash(ZTOI(zp));
zp->z_is_stale = B_TRUE;
}
}
mutex_exit(&zsb->z_znodes_lock);
}
bail:
/* release the VFS ops */
rw_exit(&zsb->z_teardown_inactive_lock);
rrw_exit(&zsb->z_teardown_lock, FTAG);
if (err) {
/*
* Since we couldn't reopen zfs_sb_t or, setup the
* sa framework, force unmount this file system.
*/
if (zsb->z_os)
(void) zfs_umount(zsb->z_sb);
}
return (err);
}
void
dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx)
{
int err;
int after_branch_point = FALSE;
dsl_pool_t *dp = ds->ds_dir->dd_pool;
objset_t *mos = dp->dp_meta_objset;
dsl_dataset_t *ds_prev = NULL;
uint64_t obj;
ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
rrw_enter(&ds->ds_bp_rwlock, RW_READER, FTAG);
ASSERT3U(dsl_dataset_phys(ds)->ds_bp.blk_birth, <=, tx->tx_txg);
rrw_exit(&ds->ds_bp_rwlock, FTAG);
ASSERT(refcount_is_zero(&ds->ds_longholds));
if (defer &&
(ds->ds_userrefs > 0 ||
dsl_dataset_phys(ds)->ds_num_children > 1)) {
ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
dmu_buf_will_dirty(ds->ds_dbuf, tx);
dsl_dataset_phys(ds)->ds_flags |= DS_FLAG_DEFER_DESTROY;
spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
return;
}
ASSERT3U(dsl_dataset_phys(ds)->ds_num_children, <=, 1);
/* We need to log before removing it from the namespace. */
spa_history_log_internal_ds(ds, "destroy", tx, "");
dsl_scan_ds_destroyed(ds, tx);
obj = ds->ds_object;
boolean_t book_exists = dsl_bookmark_ds_destroyed(ds, tx);
for (spa_feature_t f = 0; f < SPA_FEATURES; f++) {
if (dsl_dataset_feature_is_active(ds, f))
dsl_dataset_deactivate_feature(ds, f, tx);
}
if (dsl_dataset_phys(ds)->ds_prev_snap_obj != 0) {
ASSERT3P(ds->ds_prev, ==, NULL);
VERIFY0(dsl_dataset_hold_obj(dp,
dsl_dataset_phys(ds)->ds_prev_snap_obj, FTAG, &ds_prev));
after_branch_point =
(dsl_dataset_phys(ds_prev)->ds_next_snap_obj != obj);
dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
if (after_branch_point &&
dsl_dataset_phys(ds_prev)->ds_next_clones_obj != 0) {
dsl_dataset_remove_from_next_clones(ds_prev, obj, tx);
if (dsl_dataset_phys(ds)->ds_next_snap_obj != 0) {
VERIFY0(zap_add_int(mos,
dsl_dataset_phys(ds_prev)->
ds_next_clones_obj,
dsl_dataset_phys(ds)->ds_next_snap_obj,
tx));
}
}
if (!after_branch_point) {
dsl_dataset_phys(ds_prev)->ds_next_snap_obj =
dsl_dataset_phys(ds)->ds_next_snap_obj;
}
}
/*
* Reopen zfs_sb_t and release VFS ops.
*/
int
zfs_resume_fs(zfs_sb_t *zsb, const char *osname)
{
int err, err2;
znode_t *zp;
uint64_t sa_obj = 0;
ASSERT(RRW_WRITE_HELD(&zsb->z_teardown_lock));
ASSERT(RW_WRITE_HELD(&zsb->z_teardown_inactive_lock));
/*
* We already own this, so just hold and rele it to update the
* objset_t, as the one we had before may have been evicted.
*/
VERIFY0(dmu_objset_hold(osname, zsb, &zsb->z_os));
VERIFY3P(zsb->z_os->os_dsl_dataset->ds_owner, ==, zsb);
VERIFY(dsl_dataset_long_held(zsb->z_os->os_dsl_dataset));
dmu_objset_rele(zsb->z_os, zsb);
/*
* Make sure version hasn't changed
*/
err = zfs_get_zplprop(zsb->z_os, ZFS_PROP_VERSION,
&zsb->z_version);
if (err)
goto bail;
err = zap_lookup(zsb->z_os, MASTER_NODE_OBJ,
ZFS_SA_ATTRS, 8, 1, &sa_obj);
if (err && zsb->z_version >= ZPL_VERSION_SA)
goto bail;
if ((err = sa_setup(zsb->z_os, sa_obj,
zfs_attr_table, ZPL_END, &zsb->z_attr_table)) != 0)
goto bail;
if (zsb->z_version >= ZPL_VERSION_SA)
sa_register_update_callback(zsb->z_os,
zfs_sa_upgrade);
VERIFY(zfs_sb_setup(zsb, B_FALSE) == 0);
zfs_set_fuid_feature(zsb);
zsb->z_rollback_time = jiffies;
/*
* Attempt to re-establish all the active inodes with their
* dbufs. If a zfs_rezget() fails, then we unhash the inode
* and mark it stale. This prevents a collision if a new
* inode/object is created which must use the same inode
* number. The stale inode will be be released when the
* VFS prunes the dentry holding the remaining references
* on the stale inode.
*/
mutex_enter(&zsb->z_znodes_lock);
for (zp = list_head(&zsb->z_all_znodes); zp;
zp = list_next(&zsb->z_all_znodes, zp)) {
err2 = zfs_rezget(zp);
if (err2) {
remove_inode_hash(ZTOI(zp));
zp->z_is_stale = B_TRUE;
}
}
mutex_exit(&zsb->z_znodes_lock);
bail:
/* release the VFS ops */
rw_exit(&zsb->z_teardown_inactive_lock);
rrw_exit(&zsb->z_teardown_lock, FTAG);
if (err) {
/*
* Since we couldn't setup the sa framework, try to force
* unmount this file system.
*/
if (zsb->z_os)
(void) zfs_umount(zsb->z_sb);
}
return (err);
}
/*ARGSUSED*/
static int
zfs_umount(vfs_t *vfsp, int fflag)
{
zfsvfs_t *zfsvfs = vfsp->vfs_data;
objset_t *os;
cred_t *cr = curthread->td_ucred;
int ret;
ret = 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);
/*
* Unmount any snapshots mounted under .zfs before unmounting the
* dataset itself.
*/
if (zfsvfs->z_ctldir != NULL) {
if ((ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0)
return (ret);
ret = vflush(vfsp, 0, 0, curthread);
ASSERT(ret == EBUSY);
if (!(fflag & MS_FORCE)) {
if (zfsvfs->z_ctldir->v_count > 1)
return (EBUSY);
ASSERT(zfsvfs->z_ctldir->v_count == 1);
}
zfsctl_destroy(zfsvfs);
ASSERT(zfsvfs->z_ctldir == NULL);
}
if (fflag & MS_FORCE) {
/*
* Mark file system as unmounted before calling
* vflush(FORCECLOSE). This way we ensure no future vnops
* will be called and risk operating on DOOMED vnodes.
*/
rrw_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
zfsvfs->z_unmounted = B_TRUE;
rrw_exit(&zfsvfs->z_teardown_lock, FTAG);
}
/*
* Flush all the files.
*/
ret = vflush(vfsp, 1, (fflag & MS_FORCE) ? FORCECLOSE : 0, curthread);
if (ret != 0) {
if (!zfsvfs->z_issnap) {
zfsctl_create(zfsvfs);
ASSERT(zfsvfs->z_ctldir != NULL);
}
return (ret);
}
if (!(fflag & MS_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 (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);
}
} else {
MNT_ILOCK(vfsp);
vfsp->mnt_kern_flag |= MNTK_UNMOUNTF;
MNT_IUNLOCK(vfsp);
}
VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
os = zfsvfs->z_os;
/*
* z_os will be NULL if there was an error in
* attempting to reopen zfsvfs.
*/
if (os != NULL) {
/*
* Unset the objset user_ptr.
*/
mutex_enter(&os->os->os_user_ptr_lock);
dmu_objset_set_user(os, NULL);
mutex_exit(&os->os->os_user_ptr_lock);
/*
* Finally release the objset
*/
dmu_objset_close(os);
}
/*
* We can now safely destroy the '.zfs' directory node.
*/
if (zfsvfs->z_ctldir != NULL)
zfsctl_destroy(zfsvfs);
if (zfsvfs->z_issnap) {
vnode_t *svp = vfsp->mnt_vnodecovered;
if (svp->v_count >= 2)
VN_RELE(svp);
}
zfs_freevfs(vfsp);
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);
}
int
dsl_crypto_key_load(const char *dsname, zcrypt_key_t *wrappingkey)
{
dsl_dataset_t *ds;
uint64_t crypt;
int error;
dsl_pool_t *dp;
error = dsl_pool_hold(dsname, FTAG, &dp);
if (error != 0)
return (error);
if ((error = dsl_dataset_hold(dp, dsname, FTAG, &ds)) != 0) {
dsl_pool_rele(dp, FTAG);
return (error);
}
/*
* This is key load not key change so if ds->ds_key is already
* set we fail.
*/
if (zcrypt_keystore_find_node(dsl_dataset_get_spa(ds),
ds->ds_object, B_FALSE) != NULL) {
dsl_dataset_rele(ds, FTAG);
dsl_pool_rele(dp, FTAG);
return (EEXIST);
}
/*
* Find out what size of key we expect.
*
* For now the wrapping key size (and type) matches the size
* of the dataset key, this may not always be the case
* (particularly if we ever support wrapping dataset keys
* with asymmetric keys (eg RSA)).
*
* When alternate wrapping keys are added it maybe done using
* a index property.
*/
rrw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER, FTAG);
error = dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_ENCRYPTION),
8, 1, &crypt, NULL/*, DSL_PROP_GET_EFFECTIVE*/);
rrw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock, FTAG);
if (error != 0) {
dsl_dataset_rele(ds, FTAG);
dsl_pool_rele(dp, FTAG);
return (error);
}
if (crypt == ZIO_CRYPT_OFF) {
dsl_dataset_rele(ds, FTAG);
dsl_pool_rele(dp, FTAG);
return (ENOTSUP);
}
ASSERT(crypt != ZIO_CRYPT_INHERIT);
error = dsl_keychain_load(ds, crypt, wrappingkey);
dsl_dataset_rele(ds, FTAG);
dsl_pool_rele(dp, FTAG);
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;
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);
}
static int
dsl_crypto_key_change_find(const char *dsname, void *arg)
{
struct wkey_change_arg *ca = arg;
struct kcnode *kcn;
dsl_dataset_t *ds;
objset_t *os;
uint64_t crypt;
char caource[MAXNAMELEN];
char setpoint[MAXNAMELEN];
int err;
dsl_pool_t *dp;
err = dsl_pool_hold(dsname, FTAG, &dp);
if (err != 0)
return (err);
kcn = kmem_alloc(sizeof (struct kcnode), KM_SLEEP);
if ((err = dsl_dataset_hold(dp, dsname, kcn, &ds)) != 0) {
kmem_free(kcn, sizeof (struct kcnode));
dsl_pool_rele(dp, FTAG);
return (err);
}
if ((err = dmu_objset_from_ds(ds, &os)) != 0) {
dsl_dataset_rele(ds, kcn);
dsl_pool_rele(dp, FTAG);
kmem_free(kcn, sizeof (struct kcnode));
return (err);
}
/*
* Check that this child dataset of ca->parent
* is actually inheriting keysource from ca->parent and
* not somewhere else (eg local, or some other dataset).
*/
rrw_enter(&ds->ds_dir->dd_pool->dp_config_rwlock, RW_READER, FTAG);
VERIFY(dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_ENCRYPTION),
8, 1, &crypt, NULL/*, DSL_PROP_GET_EFFECTIVE*/) == 0);
VERIFY(dsl_prop_get_ds(ds, zfs_prop_to_name(ZFS_PROP_KEYSOURCE), 1,
sizeof (caource), &caource, setpoint/*, DSL_PROP_GET_EFFECTIVE*/) == 0);
rrw_exit(&ds->ds_dir->dd_pool->dp_config_rwlock, FTAG);
if (crypt == ZIO_CRYPT_OFF ||
((strcmp(ca->ca_parent, setpoint) != 0 &&
strcmp(ca->ca_parent, dsname) != 0))) {
dsl_dataset_rele(ds, kcn);
dsl_pool_rele(dp, FTAG);
kmem_free(kcn, sizeof (struct kcnode));
return (0);
}
//dsl_sync_task_create(ca->ca_dstg, dsl_crypto_key_change_check,
// dsl_crypto_key_change_sync, ds, arg, 1);
ca->ca_ds = ds;
err = dsl_sync_task(dsname, dsl_crypto_key_change_check,
dsl_crypto_key_change_sync, arg,
1, ZFS_SPACE_CHECK_NONE);
kcn->kc_ds = ds;
list_insert_tail(&ca->ca_nodes, kcn);
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;
/*
* 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;
}
}
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);
}
/*
* 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 (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;
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);
}