static void
dsl_prop_changed_notify(dsl_pool_t *dp, uint64_t ddobj,
const char *propname, uint64_t value, int first)
{
dsl_dir_t *dd;
dsl_prop_record_t *pr;
dsl_prop_cb_record_t *cbr;
objset_t *mos = dp->dp_meta_objset;
zap_cursor_t zc;
zap_attribute_t *za;
int err;
ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
err = dsl_dir_hold_obj(dp, ddobj, NULL, FTAG, &dd);
if (err)
return;
if (!first) {
/*
* If the prop is set here, then this change is not
* being inherited here or below; stop the recursion.
*/
err = zap_contains(mos, dsl_dir_phys(dd)->dd_props_zapobj,
propname);
if (err == 0) {
dsl_dir_rele(dd, FTAG);
return;
}
ASSERT3U(err, ==, ENOENT);
}
static void
dsl_dataset_user_release_sync(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_release_arg_t *ddura = arg;
dsl_holdfunc_t *holdfunc = ddura->ddura_holdfunc;
dsl_pool_t *dp = dmu_tx_pool(tx);
ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
for (nvpair_t *pair = nvlist_next_nvpair(ddura->ddura_chkholds, NULL);
pair != NULL; pair = nvlist_next_nvpair(ddura->ddura_chkholds,
pair)) {
dsl_dataset_t *ds;
const char *name = nvpair_name(pair);
VERIFY0(holdfunc(dp, name, FTAG, &ds));
dsl_dataset_user_release_sync_one(ds,
fnvpair_value_nvlist(pair), tx);
if (nvlist_exists(ddura->ddura_todelete, name)) {
ASSERT(ds->ds_userrefs == 0 &&
ds->ds_phys->ds_num_children == 1 &&
DS_IS_DEFER_DESTROY(ds));
dsl_destroy_snapshot_sync_impl(ds, B_FALSE, tx);
}
dsl_dataset_rele(ds, FTAG);
}
}
/*
* Update all property values for ddobj & its descendants. This is used
* when renaming the dir.
*/
void
dsl_prop_notify_all(dsl_dir_t *dd)
{
dsl_pool_t *dp = dd->dd_pool;
ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
(void) dmu_objset_find_dp(dp, dd->dd_object, dsl_prop_notify_all_cb,
NULL, DS_FIND_CHILDREN);
}
void
dsl_destroy_snapshot_sync_impl(dsl_dataset_t *ds, boolean_t defer, dmu_tx_t *tx)
{
#ifdef ZFS_DEBUG
int err;
#endif
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, old_unique, used = 0, comp = 0, uncomp = 0;
dsl_dataset_t *ds_next, *ds_head, *hds;
ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
ASSERT3U(ds->ds_phys->ds_bp.blk_birth, <=, tx->tx_txg);
ASSERT(refcount_is_zero(&ds->ds_longholds));
if (defer &&
(ds->ds_userrefs > 0 || ds->ds_phys->ds_num_children > 1)) {
ASSERT(spa_version(dp->dp_spa) >= SPA_VERSION_USERREFS);
dmu_buf_will_dirty(ds->ds_dbuf, tx);
ds->ds_phys->ds_flags |= DS_FLAG_DEFER_DESTROY;
spa_history_log_internal_ds(ds, "defer_destroy", tx, "");
return;
}
ASSERT3U(ds->ds_phys->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;
if (ds->ds_phys->ds_prev_snap_obj != 0) {
ASSERT3P(ds->ds_prev, ==, NULL);
VERIFY0(dsl_dataset_hold_obj(dp,
ds->ds_phys->ds_prev_snap_obj, FTAG, &ds_prev));
after_branch_point =
(ds_prev->ds_phys->ds_next_snap_obj != obj);
dmu_buf_will_dirty(ds_prev->ds_dbuf, tx);
if (after_branch_point &&
ds_prev->ds_phys->ds_next_clones_obj != 0) {
dsl_dataset_remove_from_next_clones(ds_prev, obj, tx);
if (ds->ds_phys->ds_next_snap_obj != 0) {
VERIFY0(zap_add_int(mos,
ds_prev->ds_phys->ds_next_clones_obj,
ds->ds_phys->ds_next_snap_obj, tx));
}
}
if (!after_branch_point) {
ds_prev->ds_phys->ds_next_snap_obj =
ds->ds_phys->ds_next_snap_obj;
}
}
static int
dsl_dataset_user_release_check(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_release_arg_t *ddura;
dsl_holdfunc_t *holdfunc;
dsl_pool_t *dp;
nvpair_t *pair;
if (!dmu_tx_is_syncing(tx))
return (0);
dp = dmu_tx_pool(tx);
ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
ddura = arg;
holdfunc = ddura->ddura_holdfunc;
for (pair = nvlist_next_nvpair(ddura->ddura_holds, NULL);
pair != NULL; pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) {
int error;
dsl_dataset_t *ds;
nvlist_t *holds;
const char *snapname = nvpair_name(pair);
error = nvpair_value_nvlist(pair, &holds);
if (error != 0)
error = (SET_ERROR(EINVAL));
else
error = holdfunc(dp, snapname, FTAG, &ds);
if (error == 0) {
error = dsl_dataset_user_release_check_one(ddura, ds,
holds, snapname);
dsl_dataset_rele(ds, FTAG);
}
if (error != 0) {
if (ddura->ddura_errlist != NULL) {
fnvlist_add_int32(ddura->ddura_errlist,
snapname, error);
}
/*
* Non-existent snapshots are put on the errlist,
* but don't cause an overall failure.
*/
if (error != ENOENT)
return (error);
}
}
return (0);
}
static void
dsl_dataset_user_hold_sync_one_impl(nvlist_t *tmpholds, dsl_dataset_t *ds,
const char *htag, minor_t minor, uint64_t now, dmu_tx_t *tx)
{
dsl_pool_t *dp = ds->ds_dir->dd_pool;
objset_t *mos = dp->dp_meta_objset;
uint64_t zapobj;
ASSERT(RRW_WRITE_HELD(&dp->dp_config_rwlock));
if (ds->ds_phys->ds_userrefs_obj == 0) {
/*
* This is the first user hold for this dataset. Create
* the userrefs zap object.
*/
dmu_buf_will_dirty(ds->ds_dbuf, tx);
zapobj = ds->ds_phys->ds_userrefs_obj =
zap_create(mos, DMU_OT_USERREFS, DMU_OT_NONE, 0, tx);
} else {
zapobj = ds->ds_phys->ds_userrefs_obj;
}
ds->ds_userrefs++;
VERIFY0(zap_add(mos, zapobj, htag, 8, 1, &now, tx));
if (minor != 0) {
char name[MAXNAMELEN];
nvlist_t *tags;
VERIFY0(dsl_pool_user_hold(dp, ds->ds_object,
htag, now, tx));
(void) snprintf(name, sizeof (name), "%llx",
(u_longlong_t)ds->ds_object);
if (nvlist_lookup_nvlist(tmpholds, name, &tags) != 0) {
VERIFY0(nvlist_alloc(&tags, NV_UNIQUE_NAME,
KM_PUSHPAGE));
fnvlist_add_boolean(tags, htag);
fnvlist_add_nvlist(tmpholds, name, tags);
fnvlist_free(tags);
} else {
fnvlist_add_boolean(tags, htag);
}
}
spa_history_log_internal_ds(ds, "hold", tx,
"tag=%s temp=%d refs=%llu",
htag, minor != 0, ds->ds_userrefs);
}
/*
* 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);
}
/*
* dsl_crypto_key_change
*
* The old key must already be present in memory since the user interface
* doesn't provide away to prompt or retrieve the old key.
*/
static void
dsl_crypto_key_change_sync(void *arg1, dmu_tx_t *tx)
{
struct wkey_change_arg *ca = arg1;
dsl_dataset_t *ds = ca->ca_ds;
size_t wkeylen;
char *wkeybuf = NULL;
zcrypt_key_t *txgkey;
zap_cursor_t zc;
zap_attribute_t za;
objset_t *mos;
uint64_t keychain_zapobj;
spa_t *spa;
zcrypt_keystore_node_t *zkn;
ASSERT(RRW_WRITE_HELD(&ds->ds_dir->dd_pool->dp_config_rwlock));
mos = ds->ds_dir->dd_pool->dp_meta_objset;
keychain_zapobj = dsl_dir_phys(ds->ds_dir)->dd_keychain_obj;
/*
* To allow for the case were the keychains of child datasets
* are not loaded (ie an explicit 'zfs key -u tank/fs/sub' had
* been done some time before doing 'zfs key -c tank/fs') we itterate
* over the zap objects on disk rather than copying from the
* in memory keystore node.
*/
for (zap_cursor_init(&zc, mos, keychain_zapobj);
zap_cursor_retrieve(&zc, &za) == 0;
zap_cursor_advance(&zc)) {
wkeylen = za.za_num_integers;
wkeybuf = kmem_alloc(wkeylen, KM_SLEEP);
VERIFY(zap_lookup_uint64(mos, keychain_zapobj,
(uint64_t *)za.za_name, 1, 1, wkeylen, wkeybuf) == 0);
VERIFY(zcrypt_unwrap_key(ca->ca_old_key,
ds->ds_objset->os_crypt, wkeybuf, wkeylen, &txgkey) == 0);
kmem_free(wkeybuf, wkeylen);
VERIFY(zcrypt_wrap_key(ca->ca_new_key, txgkey,
&wkeybuf, &wkeylen,
zio_crypt_select_wrap(ds->ds_objset->os_crypt)) == 0);
zcrypt_key_free(txgkey);
VERIFY(zap_update_uint64(mos, keychain_zapobj,
(uint64_t *)za.za_name, 1, 1, wkeylen, wkeybuf, tx) == 0);
kmem_free(wkeybuf, wkeylen);
}
zap_cursor_fini(&zc);
spa = dsl_dataset_get_spa(ds);
/*
* If the wrapping key is loaded switch the in memory copy now.
*/
zkn = zcrypt_keystore_find_node(spa, ds->ds_object, B_FALSE);
if (zkn != NULL) {
zcrypt_key_free(zkn->skn_wrapkey);
zkn->skn_wrapkey = zcrypt_key_copy(ca->ca_new_key);
}
spa_history_log_internal(spa, "key change", tx,
"succeeded dataset = %llu", ds->ds_object);
}
/*
* 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);
}