static void
dmu_objset_create_sync(void *arg1, void *arg2, cred_t *cr, dmu_tx_t *tx)
{
dsl_dir_t *dd = arg1;
struct oscarg *oa = arg2;
dsl_dataset_t *ds;
blkptr_t *bp;
uint64_t dsobj;
ASSERT(dmu_tx_is_syncing(tx));
dsobj = dsl_dataset_create_sync(dd, oa->lastname,
oa->clone_parent, oa->flags, cr, tx);
VERIFY(0 == dsl_dataset_hold_obj(dd->dd_pool, dsobj, FTAG, &ds));
bp = dsl_dataset_get_blkptr(ds);
if (BP_IS_HOLE(bp)) {
objset_impl_t *osi;
/* This is an empty dmu_objset; not a clone. */
osi = dmu_objset_create_impl(dsl_dataset_get_spa(ds),
ds, bp, oa->type, tx);
if (oa->userfunc)
oa->userfunc(&osi->os, oa->userarg, cr, tx);
}
spa_history_internal_log(LOG_DS_CREATE, dd->dd_pool->dp_spa,
tx, cr, "dataset = %llu", dsobj);
dsl_dataset_rele(ds, FTAG);
}
/* ARGSUSED */
int
dsl_dir_destroy_check(void *arg1, void *arg2, dmu_tx_t *tx)
{
dsl_dir_t *dd = arg1;
dsl_pool_t *dp = dd->dd_pool;
objset_t *mos = dp->dp_meta_objset;
int err;
uint64_t count;
/*
* There should be exactly two holds, both from
* dsl_dataset_destroy: one on the dd directory, and one on its
* head ds. If there are more holds, then a concurrent thread is
* performing a lookup inside this dir while we're trying to destroy
* it. To minimize this possibility, we perform this check only
* in syncing context and fail the operation if we encounter
* additional holds. The dp_config_rwlock ensures that nobody else
* opens it after we check.
*/
if (dmu_tx_is_syncing(tx) && dmu_buf_refcount(dd->dd_dbuf) > 2)
return (EBUSY);
err = zap_count(mos, dd->dd_phys->dd_child_dir_zapobj, &count);
if (err)
return (err);
if (count != 0)
return (EEXIST);
return (0);
}
/*
* The nvlist will be consumed by this call.
*/
static void
log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
dmu_tx_t *tx, const char *fmt, va_list adx)
{
char *msg;
/*
* If this is part of creating a pool, not everything is
* initialized yet, so don't bother logging the internal events.
* Likewise if the pool is not writeable.
*/
if (tx->tx_txg == TXG_INITIAL || !spa_writeable(spa)) {
fnvlist_free(nvl);
return;
}
msg = kmem_vasprintf(fmt, adx);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
strfree(msg);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(nvl, tx);
} else {
dsl_sync_task_nowait(spa_get_dsl(spa),
spa_history_log_sync, nvl, 0, ZFS_SPACE_CHECK_NONE, tx);
}
/* spa_history_log_sync() will free nvl */
}
void
vdev_indirect_births_add_entry(vdev_indirect_births_t *vib,
uint64_t max_offset, uint64_t txg, dmu_tx_t *tx)
{
vdev_indirect_birth_entry_phys_t vibe;
uint64_t old_size;
uint64_t new_size;
vdev_indirect_birth_entry_phys_t *new_entries;
ASSERT(dmu_tx_is_syncing(tx));
ASSERT(dsl_pool_sync_context(dmu_tx_pool(tx)));
ASSERT(vdev_indirect_births_verify(vib));
dmu_buf_will_dirty(vib->vib_dbuf, tx);
vibe.vibe_offset = max_offset;
vibe.vibe_phys_birth_txg = txg;
old_size = vdev_indirect_births_size_impl(vib);
dmu_write(vib->vib_objset, vib->vib_object, old_size, sizeof (vibe),
&vibe, tx);
vib->vib_phys->vib_count++;
new_size = vdev_indirect_births_size_impl(vib);
new_entries = kmem_alloc(new_size, KM_SLEEP);
if (old_size > 0) {
bcopy(vib->vib_entries, new_entries, old_size);
kmem_free(vib->vib_entries, old_size);
}
new_entries[vib->vib_phys->vib_count - 1] = vibe;
vib->vib_entries = new_entries;
}
static void
dmu_objset_create_sync(void *arg1, void *arg2, dmu_tx_t *tx)
{
dsl_dir_t *dd = arg1;
spa_t *spa = dd->dd_pool->dp_spa;
struct oscarg *oa = arg2;
uint64_t obj;
ASSERT(dmu_tx_is_syncing(tx));
obj = dsl_dataset_create_sync(dd, oa->lastname,
oa->clone_origin, oa->crypto_ctx, oa->flags, oa->cr, tx);
if (oa->clone_origin == NULL) {
dsl_pool_t *dp = dd->dd_pool;
dsl_dataset_t *ds;
blkptr_t *bp;
objset_t *os;
VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, obj, FTAG, &ds));
bp = dsl_dataset_get_blkptr(ds);
ASSERT(BP_IS_HOLE(bp));
os = dmu_objset_create_impl(spa, ds, bp, oa->type,
oa->crypto_ctx, tx);
if (oa->userfunc)
oa->userfunc(os, oa->userarg, oa->cr, tx);
dsl_dataset_rele(ds, FTAG);
}
void
spa_history_internal_log(history_internal_events_t event, spa_t *spa,
dmu_tx_t *tx, cred_t *cr, const char *fmt, ...)
{
history_arg_t *hap;
char *str;
va_list adx;
hap = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
str = kmem_alloc(HIS_MAX_RECORD_LEN, KM_SLEEP);
va_start(adx, fmt);
(void) vsnprintf(str, HIS_MAX_RECORD_LEN, fmt, adx);
va_end(adx);
hap->ha_log_type = LOG_INTERNAL;
hap->ha_history_str = str;
hap->ha_event = event;
hap->ha_zone[0] = '\0';
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(spa, hap, cr, tx);
} else {
dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
spa_history_log_sync, spa, hap, 0, tx);
}
/* spa_history_log_sync() will free hap and str */
}
static void
log_internal(history_internal_events_t event, spa_t *spa,
dmu_tx_t *tx, const char *fmt, va_list adx)
{
history_arg_t *ha;
va_list adx_copy;
/*
* If this is part of creating a pool, not everything is
* initialized yet, so don't bother logging the internal events.
*/
if (tx->tx_txg == TXG_INITIAL)
return;
ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
va_copy(adx_copy, adx);
ha->ha_history_str = kmem_vasprintf(fmt, adx_copy);
va_end(adx_copy);
ha->ha_log_type = LOG_INTERNAL;
ha->ha_event = event;
ha->ha_zone = NULL;
ha->ha_uid = 0;
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(spa, ha, tx);
} else {
dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
spa_history_log_sync, spa, ha, 0, tx);
}
/* spa_history_log_sync() will free ha and strings */
}
/* called from dsl for meta-objset */
objset_t *
dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
dmu_objset_type_t type, dmu_tx_t *tx)
{
objset_t *os;
dnode_t *mdn;
ASSERT(dmu_tx_is_syncing(tx));
if (ds != NULL)
VERIFY0(dmu_objset_from_ds(ds, &os));
else
VERIFY0(dmu_objset_open_impl(spa, NULL, bp, &os));
mdn = DMU_META_DNODE(os);
dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
/*
* We don't want to have to increase the meta-dnode's nlevels
* later, because then we could do it in quescing context while
* we are also accessing it in open context.
*
* This precaution is not necessary for the MOS (ds == NULL),
* because the MOS is only updated in syncing context.
* This is most fortunate: the MOS is the only objset that
* needs to be synced multiple times as spa_sync() iterates
* to convergence, so minimizing its dn_nlevels matters.
*/
if (ds != NULL) {
int levels = 1;
/*
* Determine the number of levels necessary for the meta-dnode
* to contain DN_MAX_OBJECT dnodes.
*/
while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
(levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
DN_MAX_OBJECT * sizeof (dnode_phys_t))
levels++;
mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
mdn->dn_nlevels = levels;
}
ASSERT(type != DMU_OST_NONE);
ASSERT(type != DMU_OST_ANY);
ASSERT(type < DMU_OST_NUMTYPES);
os->os_phys->os_type = type;
if (dmu_objset_userused_enabled(os)) {
os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
os->os_flags = os->os_phys->os_flags;
}
dsl_dataset_dirty(ds, tx);
return (os);
}
/*
* This sync task completes (finishes) a condense, deleting the old
* mapping and replacing it with the new one.
*/
static void
spa_condense_indirect_complete_sync(void *arg, dmu_tx_t *tx)
{
spa_condensing_indirect_t *sci = arg;
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
spa_condensing_indirect_phys_t *scip =
&spa->spa_condensing_indirect_phys;
vdev_t *vd = vdev_lookup_top(spa, scip->scip_vdev);
vdev_indirect_config_t *vic = &vd->vdev_indirect_config;
objset_t *mos = spa->spa_meta_objset;
vdev_indirect_mapping_t *old_mapping = vd->vdev_indirect_mapping;
uint64_t old_count = vdev_indirect_mapping_num_entries(old_mapping);
uint64_t new_count =
vdev_indirect_mapping_num_entries(sci->sci_new_mapping);
ASSERT(dmu_tx_is_syncing(tx));
ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
ASSERT3P(sci, ==, spa->spa_condensing_indirect);
for (int i = 0; i < TXG_SIZE; i++) {
ASSERT(list_is_empty(&sci->sci_new_mapping_entries[i]));
}
ASSERT(vic->vic_mapping_object != 0);
ASSERT3U(vd->vdev_id, ==, scip->scip_vdev);
ASSERT(scip->scip_next_mapping_object != 0);
ASSERT(scip->scip_prev_obsolete_sm_object != 0);
/*
* Reset vdev_indirect_mapping to refer to the new object.
*/
rw_enter(&vd->vdev_indirect_rwlock, RW_WRITER);
vdev_indirect_mapping_close(vd->vdev_indirect_mapping);
vd->vdev_indirect_mapping = sci->sci_new_mapping;
rw_exit(&vd->vdev_indirect_rwlock);
sci->sci_new_mapping = NULL;
vdev_indirect_mapping_free(mos, vic->vic_mapping_object, tx);
vic->vic_mapping_object = scip->scip_next_mapping_object;
scip->scip_next_mapping_object = 0;
space_map_free_obj(mos, scip->scip_prev_obsolete_sm_object, tx);
spa_feature_decr(spa, SPA_FEATURE_OBSOLETE_COUNTS, tx);
scip->scip_prev_obsolete_sm_object = 0;
scip->scip_vdev = 0;
VERIFY0(zap_remove(mos, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_CONDENSING_INDIRECT, tx));
spa_condensing_indirect_destroy(spa->spa_condensing_indirect);
spa->spa_condensing_indirect = NULL;
zfs_dbgmsg("finished condense of vdev %llu in txg %llu: "
"new mapping object %llu has %llu entries "
"(was %llu entries)",
vd->vdev_id, dmu_tx_get_txg(tx), vic->vic_mapping_object,
new_count, old_count);
vdev_config_dirty(spa->spa_root_vdev);
}
uint64_t
vdev_indirect_births_alloc(objset_t *os, dmu_tx_t *tx)
{
ASSERT(dmu_tx_is_syncing(tx));
return (dmu_object_alloc(os,
DMU_OTN_UINT64_METADATA, SPA_OLD_MAXBLOCKSIZE,
DMU_OTN_UINT64_METADATA, sizeof (vdev_indirect_birth_phys_t),
tx));
}
void
dmu_tx_add_new_object(dmu_tx_t *tx, dnode_t *dn)
{
/*
* If we're syncing, they can manipulate any object anyhow, and
* the hold on the dnode_t can cause problems.
*/
if (!dmu_tx_is_syncing(tx))
(void) dmu_tx_hold_dnode_impl(tx, dn, THT_NEWOBJECT, 0, 0);
}
/*
* Mark the DVA vdev_id:offset:size as being obsolete in the given tx. This
* wrapper is provided because the DMU does not know about vdev_t's and
* cannot directly call vdev_indirect_mark_obsolete.
*/
void
spa_vdev_indirect_mark_obsolete(spa_t *spa, uint64_t vdev_id, uint64_t offset,
uint64_t size, dmu_tx_t *tx)
{
vdev_t *vd = vdev_lookup_top(spa, vdev_id);
ASSERT(dmu_tx_is_syncing(tx));
/* The DMU can only remap indirect vdevs. */
ASSERT3P(vd->vdev_ops, ==, &vdev_indirect_ops);
vdev_indirect_mark_obsolete(vd, offset, size);
}
void
dmu_tx_add_new_object(dmu_tx_t *tx, objset_t *os, uint64_t object)
{
/*
* If we're syncing, they can manipulate any object anyhow, and
* the hold on the dnode_t can cause problems.
*/
if (!dmu_tx_is_syncing(tx)) {
(void) dmu_tx_hold_object_impl(tx, os,
object, THT_NEWOBJECT, 0, 0);
}
}
static int
dsl_dataset_user_hold_check(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_hold_arg_t *dduha = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
nvpair_t *pair;
if (spa_version(dp->dp_spa) < SPA_VERSION_USERREFS)
return (SET_ERROR(ENOTSUP));
if (!dmu_tx_is_syncing(tx))
return (0);
for (pair = nvlist_next_nvpair(dduha->dduha_holds, NULL);
pair != NULL; pair = nvlist_next_nvpair(dduha->dduha_holds, pair)) {
dsl_dataset_t *ds;
int error = 0;
char *htag, *name;
/* must be a snapshot */
name = nvpair_name(pair);
if (strchr(name, '@') == NULL)
error = SET_ERROR(EINVAL);
if (error == 0)
error = nvpair_value_string(pair, &htag);
if (error == 0)
error = dsl_dataset_hold(dp, name, FTAG, &ds);
if (error == 0) {
error = dsl_dataset_user_hold_check_one(ds, htag,
dduha->dduha_minor != 0, tx);
dsl_dataset_rele(ds, FTAG);
}
if (error == 0) {
fnvlist_add_string(dduha->dduha_chkholds, name, htag);
} else {
/*
* We register ENOENT errors so they can be correctly
* reported if needed, such as when all holds fail.
*/
fnvlist_add_int32(dduha->dduha_errlist, name, error);
if (error != ENOENT)
return (error);
}
}
return (0);
}
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);
}
/*
* This sync task appends entries to the new mapping object.
*/
static void
spa_condense_indirect_commit_sync(void *arg, dmu_tx_t *tx)
{
spa_condensing_indirect_t *sci = arg;
uint64_t txg = dmu_tx_get_txg(tx);
spa_t *spa = dmu_tx_pool(tx)->dp_spa;
ASSERT(dmu_tx_is_syncing(tx));
ASSERT3P(sci, ==, spa->spa_condensing_indirect);
vdev_indirect_mapping_add_entries(sci->sci_new_mapping,
&sci->sci_new_mapping_entries[txg & TXG_MASK], tx);
ASSERT(list_is_empty(&sci->sci_new_mapping_entries[txg & TXG_MASK]));
}
/*
* 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_dir_sync(dsl_dir_t *dd, dmu_tx_t *tx)
{
ASSERT(dmu_tx_is_syncing(tx));
mutex_enter(&dd->dd_lock);
ASSERT0(dd->dd_tempreserved[tx->tx_txg&TXG_MASK]);
dprintf_dd(dd, "txg=%llu towrite=%lluK\n", tx->tx_txg,
dd->dd_space_towrite[tx->tx_txg&TXG_MASK] / 1024);
dd->dd_space_towrite[tx->tx_txg&TXG_MASK] = 0;
mutex_exit(&dd->dd_lock);
/* release the hold from dsl_dir_dirty */
dmu_buf_rele(dd->dd_dbuf, dd);
}
static int
dsl_destroy_snapshot_check(void *arg, dmu_tx_t *tx)
{
dmu_snapshots_destroy_arg_t *dsda = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
nvpair_t *pair;
int error = 0;
if (!dmu_tx_is_syncing(tx))
return (0);
for (pair = nvlist_next_nvpair(dsda->dsda_snaps, NULL);
pair != NULL; pair = nvlist_next_nvpair(dsda->dsda_snaps, pair)) {
dsl_dataset_t *ds;
error = dsl_dataset_hold(dp, nvpair_name(pair),
FTAG, &ds);
/*
* If the snapshot does not exist, silently ignore it
* (it's "already destroyed").
*/
if (error == ENOENT)
continue;
if (error == 0) {
error = dsl_destroy_snapshot_check_impl(ds,
dsda->dsda_defer);
dsl_dataset_rele(ds, FTAG);
}
if (error == 0) {
fnvlist_add_boolean(dsda->dsda_successful_snaps,
nvpair_name(pair));
} else {
fnvlist_add_int32(dsda->dsda_errlist,
nvpair_name(pair), error);
}
}
pair = nvlist_next_nvpair(dsda->dsda_errlist, NULL);
if (pair != NULL)
return (fnvpair_value_int32(pair));
return (0);
}
static void
log_internal(history_internal_events_t event, spa_t *spa,
dmu_tx_t *tx, const char *fmt, va_list adx)
{
history_arg_t *ha;
/*
* If this is part of creating a pool, not everything is
* initialized yet, so don't bother logging the internal events.
*/
if (tx->tx_txg == TXG_INITIAL)
return;
ha = kmem_alloc(sizeof (history_arg_t), KM_SLEEP);
#ifdef __DARWIN__
/*
* vsnprintf(NULL, 0, ...) is broken on darwin!
*/
{
char *buf = kmem_alloc(HIS_MAX_RECORD_LEN, KM_SLEEP);
(void) vsnprintf(buf, HIS_MAX_RECORD_LEN, fmt, adx);
ha->ha_history_str = strdup(buf);
kmem_free(buf, HIS_MAX_RECORD_LEN);
}
#else
ha->ha_history_str = kmem_alloc(vsnprintf(NULL, 0, fmt, adx) + 1,
KM_SLEEP);
(void) vsprintf(ha->ha_history_str, fmt, adx);
#endif /* __DARWIN__ */
ha->ha_log_type = LOG_INTERNAL;
ha->ha_event = event;
ha->ha_zone = NULL;
ha->ha_uid = 0;
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(spa, ha, tx);
} else {
dsl_sync_task_do_nowait(spa_get_dsl(spa), NULL,
spa_history_log_sync, spa, ha, 0, tx);
}
/* spa_history_log_sync() will free ha and strings */
}
static int
dsl_dataset_user_release_check(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_release_arg_t *ddura = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
nvpair_t *pair;
int rv = 0;
if (!dmu_tx_is_syncing(tx))
return (0);
for (pair = nvlist_next_nvpair(ddura->ddura_holds, NULL); pair != NULL;
pair = nvlist_next_nvpair(ddura->ddura_holds, pair)) {
const char *name = nvpair_name(pair);
int error;
dsl_dataset_t *ds;
nvlist_t *holds;
error = nvpair_value_nvlist(pair, &holds);
if (error != 0)
return (EINVAL);
error = dsl_dataset_hold(dp, name, FTAG, &ds);
if (error == 0) {
boolean_t deleteme;
error = dsl_dataset_user_release_check_one(ds,
holds, &deleteme);
if (error == 0 && deleteme) {
fnvlist_add_boolean(ddura->ddura_todelete,
name);
}
dsl_dataset_rele(ds, FTAG);
}
if (error != 0) {
if (ddura->ddura_errlist != NULL) {
fnvlist_add_int32(ddura->ddura_errlist,
name, error);
}
rv = error;
}
}
return (rv);
}
static int
dsl_dataset_user_release_tmp_check(void *arg, dmu_tx_t *tx)
{
dsl_dataset_user_release_tmp_arg_t *ddurta = arg;
dsl_pool_t *dp = dmu_tx_pool(tx);
dsl_dataset_t *ds;
int error;
if (!dmu_tx_is_syncing(tx))
return (0);
error = dsl_dataset_hold_obj(dp, ddurta->ddurta_dsobj, FTAG, &ds);
if (error)
return (error);
error = dsl_dataset_user_release_check_one(ds,
ddurta->ddurta_holds, &ddurta->ddurta_deleteme);
dsl_dataset_rele(ds, FTAG);
return (error);
}
/*
* The nvlist will be consumed by this call.
*/
static void
log_internal(nvlist_t *nvl, const char *operation, spa_t *spa,
dmu_tx_t *tx, const char *fmt, va_list adx)
{
char *msg;
va_list adx1;
int size;
/*
* If this is part of creating a pool, not everything is
* initialized yet, so don't bother logging the internal events.
* Likewise if the pool is not writeable.
*/
if (tx->tx_txg == TXG_INITIAL || !spa_writeable(spa)) {
fnvlist_free(nvl);
return;
}
va_copy(adx1, adx);
size = vsnprintf(NULL, 0, fmt, adx1) + 1;
msg = kmem_alloc(size, KM_PUSHPAGE);
va_end(adx1);
va_copy(adx1, adx);
(void) vsprintf(msg, fmt, adx1);
va_end(adx1);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_STR, msg);
kmem_free(msg, size);
fnvlist_add_string(nvl, ZPOOL_HIST_INT_NAME, operation);
fnvlist_add_uint64(nvl, ZPOOL_HIST_TXG, tx->tx_txg);
if (dmu_tx_is_syncing(tx)) {
spa_history_log_sync(nvl, tx);
} else {
dsl_sync_task_nowait(spa_get_dsl(spa),
spa_history_log_sync, nvl, 0, tx);
}
/* spa_history_log_sync() will free nvl */
}
/* called from dsl */
void
dmu_objset_sync(objset_impl_t *os, zio_t *pio, dmu_tx_t *tx)
{
int txgoff;
zbookmark_t zb;
writeprops_t wp = { 0 };
zio_t *zio;
list_t *list;
list_t *newlist = NULL;
dbuf_dirty_record_t *dr;
dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
ASSERT(dmu_tx_is_syncing(tx));
/* XXX the write_done callback should really give us the tx... */
os->os_synctx = tx;
if (os->os_dsl_dataset == NULL) {
/*
* This is the MOS. If we have upgraded,
* spa_max_replication() could change, so reset
* os_copies here.
*/
os->os_copies = spa_max_replication(os->os_spa);
}
/*
* Create the root block IO
*/
zb.zb_objset = os->os_dsl_dataset ? os->os_dsl_dataset->ds_object : 0;
zb.zb_object = 0;
zb.zb_level = -1; /* for block ordering; it's level 0 on disk */
zb.zb_blkid = 0;
wp.wp_type = DMU_OT_OBJSET;
wp.wp_level = 0; /* on-disk BP level; see above */
wp.wp_copies = os->os_copies;
wp.wp_oschecksum = os->os_checksum;
wp.wp_oscompress = os->os_compress;
if (BP_IS_OLDER(os->os_rootbp, tx->tx_txg)) {
(void) dsl_dataset_block_kill(os->os_dsl_dataset,
os->os_rootbp, pio, tx);
}
arc_release(os->os_phys_buf, &os->os_phys_buf);
zio = arc_write(pio, os->os_spa, &wp, DMU_OS_IS_L2CACHEABLE(os),
tx->tx_txg, os->os_rootbp, os->os_phys_buf, ready, NULL, os,
ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
/*
* Sync special dnodes - the parent IO for the sync is the root block
*/
os->os_meta_dnode->dn_zio = zio;
dnode_sync(os->os_meta_dnode, tx);
os->os_phys->os_flags = os->os_flags;
if (os->os_userused_dnode &&
os->os_userused_dnode->dn_type != DMU_OT_NONE) {
os->os_userused_dnode->dn_zio = zio;
dnode_sync(os->os_userused_dnode, tx);
os->os_groupused_dnode->dn_zio = zio;
dnode_sync(os->os_groupused_dnode, tx);
}
txgoff = tx->tx_txg & TXG_MASK;
if (dmu_objset_userused_enabled(os)) {
newlist = &os->os_synced_dnodes;
/*
* We must create the list here because it uses the
* dn_dirty_link[] of this txg.
*/
list_create(newlist, sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[txgoff]));
}
dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
list = &os->os_meta_dnode->dn_dirty_records[txgoff];
while ((dr = list_head(list))) {
ASSERT(dr->dr_dbuf->db_level == 0);
list_remove(list, dr);
if (dr->dr_zio)
zio_nowait(dr->dr_zio);
}
/*
* Free intent log blocks up to this tx.
*/
zil_sync(os->os_zil, tx);
os->os_phys->os_zil_header = os->os_zil_header;
zio_nowait(zio);
}
/* called from dsl for meta-objset */
objset_t *
dmu_objset_create_impl(spa_t *spa, dsl_dataset_t *ds, blkptr_t *bp,
dmu_objset_type_t type, dsl_crypto_ctx_t *dcc, dmu_tx_t *tx)
{
objset_t *os;
dnode_t *mdn;
ASSERT(dmu_tx_is_syncing(tx));
if (ds != NULL)
VERIFY(0 == dmu_objset_from_ds(ds, &os));
else
VERIFY(0 == dmu_objset_open_impl(spa, NULL, bp, &os));
mdn = DMU_META_DNODE(os);
dnode_allocate(mdn, DMU_OT_DNODE, 1 << DNODE_BLOCK_SHIFT,
DN_MAX_INDBLKSHIFT, DMU_OT_NONE, 0, tx);
/*
* We don't want to have to increase the meta-dnode's nlevels
* later, because then we could do it in quescing context while
* we are also accessing it in open context.
*
* This precaution is not necessary for the MOS (ds == NULL),
* because the MOS is only updated in syncing context.
* This is most fortunate: the MOS is the only objset that
* needs to be synced multiple times as spa_sync() iterates
* to convergence, so minimizing its dn_nlevels matters.
*/
if (ds != NULL) {
int levels = 1;
/*
* Determine the number of levels necessary for the meta-dnode
* to contain DN_MAX_OBJECT dnodes.
*/
while ((uint64_t)mdn->dn_nblkptr << (mdn->dn_datablkshift +
(levels - 1) * (mdn->dn_indblkshift - SPA_BLKPTRSHIFT)) <
DN_MAX_OBJECT * sizeof (dnode_phys_t))
levels++;
mdn->dn_next_nlevels[tx->tx_txg & TXG_MASK] =
mdn->dn_nlevels = levels;
}
ASSERT(type != DMU_OST_NONE);
ASSERT(type != DMU_OST_ANY);
ASSERT(type < DMU_OST_NUMTYPES);
/*
* Note: although we should not be dirtying the objset outside of
* sync context, the os_type is used directly from the os_phys
* in dsl_scan so it may not be save to put os_type in the in-core
* objset_t and set it in the phys in sync context (as with os_flags)
*/
dmu_objset_dirty(os);
os->os_phys->os_type = type;
if (dmu_objset_userused_enabled(os)) {
os->os_phys->os_flags |= OBJSET_FLAG_USERACCOUNTING_COMPLETE;
// FIXME - removed in newer.
// os->os_flags = os->os_phys->os_flags;
}
if (dcc != NULL && dcc->dcc_crypt != ZIO_CRYPT_INHERIT) {
os->os_crypt = zio_crypt_select(dcc->dcc_crypt,
ZIO_CRYPT_ON_VALUE);
}
dsl_dataset_dirty(ds, tx);
return (os);
}
/* called from dsl */
void
dmu_objset_sync(objset_t *os, zio_t *pio, dmu_tx_t *tx)
{
int txgoff;
zbookmark_t zb;
zio_prop_t zp;
zio_t *zio;
list_t *list;
list_t *newlist = NULL;
dbuf_dirty_record_t *dr;
dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
ASSERT(dmu_tx_is_syncing(tx));
/* XXX the write_done callback should really give us the tx... */
os->os_synctx = tx;
if (os->os_dsl_dataset == NULL) {
/*
* This is the MOS. If we have upgraded,
* spa_max_replication() could change, so reset
* os_copies here.
*/
os->os_copies = spa_max_replication(os->os_spa);
}
/*
* Create the root block IO
*/
SET_BOOKMARK(&zb, os->os_dsl_dataset ?
os->os_dsl_dataset->ds_object : DMU_META_OBJSET,
ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
arc_release(os->os_phys_buf, &os->os_phys_buf);
dmu_write_policy(os, NULL, 0, 0, &zp);
zio = arc_write(pio, os->os_spa, tx->tx_txg,
os->os_rootbp, os->os_phys_buf, DMU_OS_IS_L2CACHEABLE(os), &zp,
dmu_objset_write_ready, dmu_objset_write_done, os,
ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED, &zb);
/*
* Sync special dnodes - the parent IO for the sync is the root block
*/
DMU_META_DNODE(os)->dn_zio = zio;
dnode_sync(DMU_META_DNODE(os), tx);
os->os_phys->os_flags = os->os_flags;
if (DMU_USERUSED_DNODE(os) &&
DMU_USERUSED_DNODE(os)->dn_type != DMU_OT_NONE) {
DMU_USERUSED_DNODE(os)->dn_zio = zio;
dnode_sync(DMU_USERUSED_DNODE(os), tx);
DMU_GROUPUSED_DNODE(os)->dn_zio = zio;
dnode_sync(DMU_GROUPUSED_DNODE(os), tx);
}
txgoff = tx->tx_txg & TXG_MASK;
if (dmu_objset_userused_enabled(os)) {
newlist = &os->os_synced_dnodes;
/*
* We must create the list here because it uses the
* dn_dirty_link[] of this txg.
*/
list_create(newlist, sizeof (dnode_t),
offsetof(dnode_t, dn_dirty_link[txgoff]));
}
dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], newlist, tx);
dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], newlist, tx);
list = &DMU_META_DNODE(os)->dn_dirty_records[txgoff];
while (dr = list_head(list)) {
ASSERT0(dr->dr_dbuf->db_level);
list_remove(list, dr);
if (dr->dr_zio)
zio_nowait(dr->dr_zio);
}
/*
* Free intent log blocks up to this tx.
*/
zil_sync(os->os_zil, tx);
os->os_phys->os_zil_header = os->os_zil_header;
zio_nowait(zio);
}
/* called from dsl */
void
dmu_objset_sync(objset_impl_t *os, zio_t *pio, dmu_tx_t *tx)
{
int txgoff;
zbookmark_t zb;
writeprops_t wp = { 0 };
zio_t *zio;
list_t *list;
dbuf_dirty_record_t *dr;
dprintf_ds(os->os_dsl_dataset, "txg=%llu\n", tx->tx_txg);
ASSERT(dmu_tx_is_syncing(tx));
/* XXX the write_done callback should really give us the tx... */
os->os_synctx = tx;
if (os->os_dsl_dataset == NULL) {
/*
* This is the MOS. If we have upgraded,
* spa_max_replication() could change, so reset
* os_copies here.
*/
os->os_copies = spa_max_replication(os->os_spa);
}
/*
* Create the root block IO
*/
zb.zb_objset = os->os_dsl_dataset ? os->os_dsl_dataset->ds_object : 0;
zb.zb_object = 0;
zb.zb_level = -1;
zb.zb_blkid = 0;
if (BP_IS_OLDER(os->os_rootbp, tx->tx_txg)) {
(void) dsl_dataset_block_kill(os->os_dsl_dataset,
os->os_rootbp, pio, tx);
}
wp.wp_type = DMU_OT_OBJSET;
wp.wp_copies = os->os_copies;
wp.wp_level = (uint8_t)-1;
wp.wp_oschecksum = os->os_checksum;
wp.wp_oscompress = os->os_compress;
arc_release(os->os_phys_buf, &os->os_phys_buf);
zio = arc_write(pio, os->os_spa, &wp,
tx->tx_txg, os->os_rootbp, os->os_phys_buf, ready, NULL, os,
ZIO_PRIORITY_ASYNC_WRITE, ZIO_FLAG_MUSTSUCCEED | ZIO_FLAG_METADATA,
&zb);
/*
* Sync meta-dnode - the parent IO for the sync is the root block
*/
os->os_meta_dnode->dn_zio = zio;
dnode_sync(os->os_meta_dnode, tx);
txgoff = tx->tx_txg & TXG_MASK;
dmu_objset_sync_dnodes(&os->os_free_dnodes[txgoff], tx);
dmu_objset_sync_dnodes(&os->os_dirty_dnodes[txgoff], tx);
list = &os->os_meta_dnode->dn_dirty_records[txgoff];
while (dr = list_head(list)) {
ASSERT(dr->dr_dbuf->db_level == 0);
list_remove(list, dr);
if (dr->dr_zio)
zio_nowait(dr->dr_zio);
}
/*
* Free intent log blocks up to this tx.
*/
zil_sync(os->os_zil, tx);
os->os_phys->os_zil_header = os->os_zil_header;
zio_nowait(zio);
}
