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);
}
int
dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
{
uint64_t obj;
int err;
err = zap_lookup(dp->dp_meta_objset,
dp->dp_root_dir->dd_phys->dd_child_dir_zapobj,
name, sizeof (obj), 1, &obj);
if (err)
return (err);
return (dsl_dir_hold_obj(dp, obj, name, dp, ddp));
}
int
dsl_pool_open_special_dir(dsl_pool_t *dp, const char *name, dsl_dir_t **ddp)
{
uint64_t obj;
int err;
#if 0
printf(" +dsl_pool_open_special_dir: '%s'\n", name ? name : "(null)");
printf(" dp %p\n", dp);
printf(" dp->meta %p\n", dp->dp_meta_objset);
printf(" dp->root %p\n", dp->dp_root_dir);
printf(" dp->root->phys %p\n", dp->dp_root_dir->dd_phys);
printf(" dp->root->phys->zap %p\n", dp->dp_root_dir->dd_phys->dd_child_dir_zapobj);
#endif
err = zap_lookup(dp->dp_meta_objset,
dsl_dir_phys(dp->dp_root_dir)->dd_child_dir_zapobj,
name, sizeof (obj), 1, &obj);
if (err)
return (err);
return (dsl_dir_hold_obj(dp, obj, name, dp, ddp));
}
int
dsl_dir_hold_obj(dsl_pool_t *dp, uint64_t ddobj,
const char *tail, void *tag, dsl_dir_t **ddp)
{
dmu_buf_t *dbuf;
dsl_dir_t *dd;
int err;
ASSERT(dsl_pool_config_held(dp));
err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
if (err != 0)
return (err);
dd = dmu_buf_get_user(dbuf);
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(dbuf, &doi);
ASSERT3U(doi.doi_bonus_type, ==, DMU_OT_DSL_DIR);
ASSERT3U(doi.doi_bonus_size, >=, sizeof (dsl_dir_phys_t));
}
#endif
if (dd == NULL) {
dsl_dir_t *winner;
dd = kmem_zalloc(sizeof (dsl_dir_t), KM_SLEEP);
dd->dd_object = ddobj;
dd->dd_dbuf = dbuf;
dd->dd_pool = dp;
dd->dd_phys = dbuf->db_data;
mutex_init(&dd->dd_lock, NULL, MUTEX_DEFAULT, NULL);
list_create(&dd->dd_prop_cbs, sizeof (dsl_prop_cb_record_t),
offsetof(dsl_prop_cb_record_t, cbr_node));
dsl_dir_snap_cmtime_update(dd);
if (dd->dd_phys->dd_parent_obj) {
err = dsl_dir_hold_obj(dp, dd->dd_phys->dd_parent_obj,
NULL, dd, &dd->dd_parent);
if (err != 0)
goto errout;
if (tail) {
#ifdef ZFS_DEBUG
uint64_t foundobj;
err = zap_lookup(dp->dp_meta_objset,
dd->dd_parent->dd_phys->dd_child_dir_zapobj,
tail, sizeof (foundobj), 1, &foundobj);
ASSERT(err || foundobj == ddobj);
#endif
(void) strcpy(dd->dd_myname, tail);
} else {
err = zap_value_search(dp->dp_meta_objset,
dd->dd_parent->dd_phys->dd_child_dir_zapobj,
ddobj, 0, dd->dd_myname);
}
if (err != 0)
goto errout;
} else {
(void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
}
if (dsl_dir_is_clone(dd)) {
dmu_buf_t *origin_bonus;
dsl_dataset_phys_t *origin_phys;
/*
* We can't open the origin dataset, because
* that would require opening this dsl_dir.
* Just look at its phys directly instead.
*/
err = dmu_bonus_hold(dp->dp_meta_objset,
dd->dd_phys->dd_origin_obj, FTAG, &origin_bonus);
if (err != 0)
goto errout;
origin_phys = origin_bonus->db_data;
dd->dd_origin_txg =
origin_phys->ds_creation_txg;
dmu_buf_rele(origin_bonus, FTAG);
}
winner = dmu_buf_set_user_ie(dbuf, dd, &dd->dd_phys,
dsl_dir_evict);
if (winner) {
if (dd->dd_parent)
dsl_dir_rele(dd->dd_parent, dd);
mutex_destroy(&dd->dd_lock);
kmem_free(dd, sizeof (dsl_dir_t));
dd = winner;
} else {
spa_open_ref(dp->dp_spa, dd);
}
}
/*
* The dsl_dir_t has both open-to-close and instantiate-to-evict
* holds on the spa. We need the open-to-close holds because
* otherwise the spa_refcnt wouldn't change when we open a
* dir which the spa also has open, so we could incorrectly
* think it was OK to unload/export/destroy the pool. We need
* the instantiate-to-evict hold because the dsl_dir_t has a
* pointer to the dd_pool, which has a pointer to the spa_t.
*/
spa_open_ref(dp->dp_spa, tag);
ASSERT3P(dd->dd_pool, ==, dp);
ASSERT3U(dd->dd_object, ==, ddobj);
ASSERT3P(dd->dd_dbuf, ==, dbuf);
*ddp = dd;
return (0);
errout:
if (dd->dd_parent)
dsl_dir_rele(dd->dd_parent, dd);
mutex_destroy(&dd->dd_lock);
kmem_free(dd, sizeof (dsl_dir_t));
dmu_buf_rele(dbuf, tag);
return (err);
}
/*
* Return the dsl_dir_t, and possibly the last component which couldn't
* be found in *tail. The name must be in the specified dsl_pool_t. This
* thread must hold the dp_config_rwlock for the pool. Returns NULL if the
* path is bogus, or if tail==NULL and we couldn't parse the whole name.
* (*tail)[0] == '@' means that the last component is a snapshot.
*/
int
dsl_dir_hold(dsl_pool_t *dp, const char *name, void *tag,
dsl_dir_t **ddp, const char **tailp)
{
char buf[MAXNAMELEN];
const char *spaname, *next, *nextnext = NULL;
int err;
dsl_dir_t *dd;
uint64_t ddobj;
err = getcomponent(name, buf, &next);
if (err != 0)
return (err);
/* Make sure the name is in the specified pool. */
spaname = spa_name(dp->dp_spa);
if (strcmp(buf, spaname) != 0)
return (SET_ERROR(EINVAL));
ASSERT(dsl_pool_config_held(dp));
err = dsl_dir_hold_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
if (err != 0) {
return (err);
}
while (next != NULL) {
dsl_dir_t *child_ds;
err = getcomponent(next, buf, &nextnext);
if (err != 0)
break;
ASSERT(next[0] != '\0');
if (next[0] == '@')
break;
dprintf("looking up %s in obj%lld\n",
buf, dd->dd_phys->dd_child_dir_zapobj);
err = zap_lookup(dp->dp_meta_objset,
dd->dd_phys->dd_child_dir_zapobj,
buf, sizeof (ddobj), 1, &ddobj);
if (err != 0) {
if (err == ENOENT)
err = 0;
break;
}
err = dsl_dir_hold_obj(dp, ddobj, buf, tag, &child_ds);
if (err != 0)
break;
dsl_dir_rele(dd, tag);
dd = child_ds;
next = nextnext;
}
if (err != 0) {
dsl_dir_rele(dd, tag);
return (err);
}
/*
* It's an error if there's more than one component left, or
* tailp==NULL and there's any component left.
*/
if (next != NULL &&
(tailp == NULL || (nextnext && nextnext[0] != '\0'))) {
/* bad path name */
dsl_dir_rele(dd, tag);
dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
err = SET_ERROR(ENOENT);
}
if (tailp != NULL)
*tailp = next;
*ddp = dd;
return (err);
}
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);
}