static void
zhack_do_feature_stat(int argc, char **argv)
{
spa_t *spa;
objset_t *os;
char *target;
argc--;
argv++;
if (argc < 1) {
(void) fprintf(stderr, "error: missing pool name\n");
usage();
}
target = argv[0];
zhack_spa_open(target, B_TRUE, FTAG, &spa);
os = spa->spa_meta_objset;
dump_obj(os, spa->spa_feat_for_read_obj, "for_read");
dump_obj(os, spa->spa_feat_for_write_obj, "for_write");
dump_obj(os, spa->spa_feat_desc_obj, "descriptions");
if (spa_feature_is_active(spa, SPA_FEATURE_ENABLED_TXG)) {
dump_obj(os, spa->spa_feat_enabled_txg_obj, "enabled_txg");
}
dump_mos(spa);
spa_close(spa, FTAG);
}
/*
* 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
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;
}
/*
* Sync task to begin the condensing process.
*/
void
spa_condense_indirect_start_sync(vdev_t *vd, dmu_tx_t *tx)
{
spa_t *spa = vd->vdev_spa;
spa_condensing_indirect_phys_t *scip =
&spa->spa_condensing_indirect_phys;
ASSERT0(scip->scip_next_mapping_object);
ASSERT0(scip->scip_prev_obsolete_sm_object);
ASSERT0(scip->scip_vdev);
ASSERT(dmu_tx_is_syncing(tx));
ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
ASSERT(spa_feature_is_active(spa, SPA_FEATURE_OBSOLETE_COUNTS));
ASSERT(vdev_indirect_mapping_num_entries(vd->vdev_indirect_mapping));
uint64_t obsolete_sm_obj = vdev_obsolete_sm_object(vd);
ASSERT(obsolete_sm_obj != 0);
scip->scip_vdev = vd->vdev_id;
scip->scip_next_mapping_object =
vdev_indirect_mapping_alloc(spa->spa_meta_objset, tx);
scip->scip_prev_obsolete_sm_object = obsolete_sm_obj;
/*
* We don't need to allocate a new space map object, since
* vdev_indirect_sync_obsolete will allocate one when needed.
*/
space_map_close(vd->vdev_obsolete_sm);
vd->vdev_obsolete_sm = NULL;
VERIFY0(zap_remove(spa->spa_meta_objset, vd->vdev_top_zap,
VDEV_TOP_ZAP_INDIRECT_OBSOLETE_SM, tx));
VERIFY0(zap_add(spa->spa_dsl_pool->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_CONDENSING_INDIRECT, sizeof (uint64_t),
sizeof (*scip) / sizeof (uint64_t), scip, tx));
ASSERT3P(spa->spa_condensing_indirect, ==, NULL);
spa->spa_condensing_indirect = spa_condensing_indirect_create(spa);
zfs_dbgmsg("starting condense of vdev %llu in txg %llu: "
"posm=%llu nm=%llu",
vd->vdev_id, dmu_tx_get_txg(tx),
(u_longlong_t)scip->scip_prev_obsolete_sm_object,
(u_longlong_t)scip->scip_next_mapping_object);
zthr_wakeup(spa->spa_condense_zthr);
}
void
dsl_pool_create_obsolete_bpobj(dsl_pool_t *dp, dmu_tx_t *tx)
{
uint64_t obj;
/*
* Currently, we only create the obsolete_bpobj where there are
* indirect vdevs with referenced mappings.
*/
ASSERT(spa_feature_is_active(dp->dp_spa, SPA_FEATURE_DEVICE_REMOVAL));
/* create and open the obsolete_bpobj */
obj = bpobj_alloc(dp->dp_meta_objset, SPA_OLD_MAXBLOCKSIZE, tx);
VERIFY0(bpobj_open(&dp->dp_obsolete_bpobj, dp->dp_meta_objset, obj));
VERIFY0(zap_add(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_OBSOLETE_BPOBJ, sizeof (uint64_t), 1, &obj, tx));
spa_feature_incr(dp->dp_spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
}
static void
free_blocks(dnode_t *dn, blkptr_t *bp, int num, dmu_tx_t *tx)
{
dsl_dataset_t *ds = dn->dn_objset->os_dsl_dataset;
uint64_t bytesfreed = 0;
int i;
dprintf("ds=%p obj=%llx num=%d\n", ds, dn->dn_object, num);
for (i = 0; i < num; i++, bp++) {
uint64_t lsize, lvl;
dmu_object_type_t type;
if (BP_IS_HOLE(bp))
continue;
bytesfreed += dsl_dataset_block_kill(ds, bp, tx, B_FALSE);
ASSERT3U(bytesfreed, <=, DN_USED_BYTES(dn->dn_phys));
/*
* Save some useful information on the holes being
* punched, including logical size, type, and indirection
* level. Retaining birth time enables detection of when
* holes are punched for reducing the number of free
* records transmitted during a zfs send.
*/
lsize = BP_GET_LSIZE(bp);
type = BP_GET_TYPE(bp);
lvl = BP_GET_LEVEL(bp);
bzero(bp, sizeof (blkptr_t));
if (spa_feature_is_active(dn->dn_objset->os_spa,
SPA_FEATURE_HOLE_BIRTH)) {
BP_SET_LSIZE(bp, lsize);
BP_SET_TYPE(bp, type);
BP_SET_LEVEL(bp, lvl);
BP_SET_BIRTH(bp, dmu_tx_get_txg(tx), 0);
}
}
dnode_diduse_space(dn, -bytesfreed);
}
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);
}
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);
}
static void
zhack_do_feature_ref(int argc, char **argv)
{
char c;
char *target;
boolean_t decr = B_FALSE;
spa_t *spa;
objset_t *mos;
zfeature_info_t feature;
zfeature_info_t *nodeps[] = { NULL };
/*
* fi_desc does not matter here because it was written to disk
* when the feature was enabled, but we need to properly set the
* feature for read or write based on the information we read off
* disk later.
*/
feature.fi_uname = "zhack";
feature.fi_mos = B_FALSE;
feature.fi_desc = NULL;
feature.fi_depends = nodeps;
optind = 1;
while ((c = getopt(argc, argv, "md")) != -1) {
switch (c) {
case 'm':
feature.fi_mos = B_TRUE;
break;
case 'd':
decr = B_TRUE;
break;
default:
usage();
break;
}
}
argc -= optind;
argv += optind;
if (argc < 2) {
(void) fprintf(stderr, "error: missing feature or pool name\n");
usage();
}
target = argv[0];
feature.fi_guid = argv[1];
if (!zfeature_is_valid_guid(feature.fi_guid))
fatal("invalid feature guid: %s", feature.fi_guid);
zhack_spa_open(target, B_FALSE, FTAG, &spa);
mos = spa->spa_meta_objset;
if (0 == zfeature_lookup_guid(feature.fi_guid, NULL))
fatal("'%s' is a real feature, will not change refcount");
if (0 == zap_contains(mos, spa->spa_feat_for_read_obj,
feature.fi_guid)) {
feature.fi_can_readonly = B_FALSE;
} else if (0 == zap_contains(mos, spa->spa_feat_for_write_obj,
feature.fi_guid)) {
feature.fi_can_readonly = B_TRUE;
} else {
fatal("feature is not enabled: %s", feature.fi_guid);
}
if (decr && !spa_feature_is_active(spa, &feature))
fatal("feature refcount already 0: %s", feature.fi_guid);
VERIFY0(dsl_sync_task(spa_name(spa), NULL,
decr ? feature_decr_sync : feature_incr_sync, &feature, 5));
spa_close(spa, FTAG);
}