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;
}
zcrypt_keystore_node_t *
zcrypt_keystore_find_node(spa_t *spa, uint64_t dsobj,
boolean_t config_rwlock_held)
{
zcrypt_keystore_node_t search;
zcrypt_keystore_node_t *found = NULL;
rw_enter(&spa->spa_keystore->sk_lock, RW_READER);
if (avl_is_empty(&spa->spa_keystore->sk_dslkeys))
goto out;
search.skn_os = dsobj;
found = avl_find(&spa->spa_keystore->sk_dslkeys, &search, NULL);
if (found == NULL) {
int error;
dsl_pool_t *dp = spa_get_dsl(spa);
dsl_dataset_t *ds;
boolean_t need_lock;
rw_exit(&spa->spa_keystore->sk_lock);
need_lock = !dsl_pool_sync_context(dp) && !config_rwlock_held;
if (need_lock)
rw_enter(&dp->dp_config_rwlock, RW_READER);
error = dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds);
if (need_lock)
rw_exit(&dp->dp_config_rwlock);
rw_enter(&spa->spa_keystore->sk_lock, RW_READER);
if (!error) {
if (dsl_dataset_is_snapshot(ds)) {
search.skn_os =
ds->ds_dir->dd_phys->dd_head_dataset_obj;
found = avl_find(&spa->spa_keystore->sk_dslkeys,
&search, NULL);
}
dsl_dataset_rele(ds, FTAG);
}
}
out:
rw_exit(&spa->spa_keystore->sk_lock);
return (found);
}
void
spa_feature_create_zap_objects(spa_t *spa, dmu_tx_t *tx)
{
/*
* We create feature flags ZAP objects in two instances: during pool
* creation and during pool upgrade.
*/
ASSERT(dsl_pool_sync_context(spa_get_dsl(spa)) || (!spa->spa_sync_on &&
tx->tx_txg == TXG_INITIAL));
spa->spa_feat_for_read_obj = zap_create_link(spa->spa_meta_objset,
DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_FEATURES_FOR_READ, tx);
spa->spa_feat_for_write_obj = zap_create_link(spa->spa_meta_objset,
DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_FEATURES_FOR_WRITE, tx);
spa->spa_feat_desc_obj = zap_create_link(spa->spa_meta_objset,
DMU_OTN_ZAP_METADATA, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_FEATURE_DESCRIPTIONS, tx);
}
void
dsl_free_sync(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp)
{
ASSERT(dsl_pool_sync_context(dp));
zio_nowait(zio_free_sync(pio, dp->dp_spa, txg, bpp, pio->io_flags));
}
int
dsl_dir_open_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(RW_LOCK_HELD(&dp->dp_config_rwlock) ||
dsl_pool_sync_context(dp));
err = dmu_bonus_hold(dp->dp_meta_objset, ddobj, tag, &dbuf);
if (err)
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_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_open_obj(dp, dd->dd_phys->dd_parent_obj,
NULL, dd, &dd->dd_parent);
if (err)
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)
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)
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_close(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_close(dd->dd_parent, dd);
mutex_destroy(&dd->dd_lock);
kmem_free(dd, sizeof (dsl_dir_t));
dmu_buf_rele(dbuf, tag);
return (err);
}
/*
* same as dsl_open_dir, ignore the first component of name and use the
* spa instead
*/
int
dsl_dir_open_spa(spa_t *spa, const char *name, void *tag,
dsl_dir_t **ddp, const char **tailp)
{
char buf[MAXNAMELEN];
const char *next, *nextnext = NULL;
int err;
dsl_dir_t *dd;
dsl_pool_t *dp;
uint64_t ddobj;
int openedspa = FALSE;
dprintf("%s\n", name);
err = getcomponent(name, buf, &next);
if (err)
return (err);
if (spa == NULL) {
err = spa_open(buf, &spa, FTAG);
if (err) {
dprintf("spa_open(%s) failed\n", buf);
return (err);
}
openedspa = TRUE;
/* XXX this assertion belongs in spa_open */
ASSERT(!dsl_pool_sync_context(spa_get_dsl(spa)));
}
dp = spa_get_dsl(spa);
rw_enter(&dp->dp_config_rwlock, RW_READER);
err = dsl_dir_open_obj(dp, dp->dp_root_dir_obj, NULL, tag, &dd);
if (err) {
rw_exit(&dp->dp_config_rwlock);
if (openedspa)
spa_close(spa, FTAG);
return (err);
}
while (next != NULL) {
dsl_dir_t *child_ds;
err = getcomponent(next, buf, &nextnext);
if (err)
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) {
if (err == ENOENT)
err = 0;
break;
}
err = dsl_dir_open_obj(dp, ddobj, buf, tag, &child_ds);
if (err)
break;
dsl_dir_close(dd, tag);
dd = child_ds;
next = nextnext;
}
rw_exit(&dp->dp_config_rwlock);
if (err) {
dsl_dir_close(dd, tag);
if (openedspa)
spa_close(spa, FTAG);
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_close(dd, tag);
dprintf("next=%p (%s) tail=%p\n", next, next?next:"", tailp);
err = ENOENT;
}
if (tailp)
*tailp = next;
if (openedspa)
spa_close(spa, FTAG);
*ddp = dd;
return (err);
}
/*
* Find all objsets under name, call func on each
*/
int
dmu_objset_find_spa(spa_t *spa, const char *name,
int func(spa_t *, uint64_t, const char *, void *), void *arg, int flags)
{
dsl_dir_t *dd;
dsl_pool_t *dp;
dsl_dataset_t *ds;
zap_cursor_t zc;
zap_attribute_t *attr;
char *child;
uint64_t thisobj;
int err;
if (name == NULL)
name = spa_name(spa);
err = dsl_dir_open_spa(spa, name, FTAG, &dd, NULL);
if (err)
return (err);
/* Don't visit hidden ($MOS & $ORIGIN) objsets. */
if (dd->dd_myname[0] == '$') {
dsl_dir_close(dd, FTAG);
return (0);
}
thisobj = dd->dd_phys->dd_head_dataset_obj;
attr = kmem_alloc(sizeof (zap_attribute_t), KM_SLEEP);
dp = dd->dd_pool;
/*
* Iterate over all children.
*/
if (flags & DS_FIND_CHILDREN) {
for (zap_cursor_init(&zc, dp->dp_meta_objset,
dd->dd_phys->dd_child_dir_zapobj);
zap_cursor_retrieve(&zc, attr) == 0;
(void) zap_cursor_advance(&zc)) {
ASSERT(attr->za_integer_length == sizeof (uint64_t));
ASSERT(attr->za_num_integers == 1);
child = kmem_alloc(MAXPATHLEN, KM_SLEEP);
(void) strcpy(child, name);
(void) strcat(child, "/");
(void) strcat(child, attr->za_name);
err = dmu_objset_find_spa(spa, child, func, arg, flags);
kmem_free(child, MAXPATHLEN);
if (err)
break;
}
zap_cursor_fini(&zc);
if (err) {
dsl_dir_close(dd, FTAG);
kmem_free(attr, sizeof (zap_attribute_t));
return (err);
}
}
/*
* Iterate over all snapshots.
*/
if (flags & DS_FIND_SNAPSHOTS) {
if (!dsl_pool_sync_context(dp))
rw_enter(&dp->dp_config_rwlock, RW_READER);
err = dsl_dataset_hold_obj(dp, thisobj, FTAG, &ds);
if (!dsl_pool_sync_context(dp))
rw_exit(&dp->dp_config_rwlock);
if (err == 0) {
uint64_t snapobj = ds->ds_phys->ds_snapnames_zapobj;
dsl_dataset_rele(ds, FTAG);
for (zap_cursor_init(&zc, dp->dp_meta_objset, snapobj);
zap_cursor_retrieve(&zc, attr) == 0;
(void) zap_cursor_advance(&zc)) {
ASSERT(attr->za_integer_length ==
sizeof (uint64_t));
ASSERT(attr->za_num_integers == 1);
child = kmem_alloc(MAXPATHLEN, KM_SLEEP);
(void) strcpy(child, name);
(void) strcat(child, "@");
(void) strcat(child, attr->za_name);
err = func(spa, attr->za_first_integer,
child, arg);
kmem_free(child, MAXPATHLEN);
if (err)
break;
}
zap_cursor_fini(&zc);
}
}
dsl_dir_close(dd, FTAG);
kmem_free(attr, sizeof (zap_attribute_t));
if (err)
return (err);
/*
* Apply to self if appropriate.
*/
err = func(spa, thisobj, name, arg);
return (err);
}
boolean_t
vdev_indirect_should_condense(vdev_t *vd)
{
vdev_indirect_mapping_t *vim = vd->vdev_indirect_mapping;
spa_t *spa = vd->vdev_spa;
ASSERT(dsl_pool_sync_context(spa->spa_dsl_pool));
if (!zfs_condense_indirect_vdevs_enable)
return (B_FALSE);
/*
* We can only condense one indirect vdev at a time.
*/
if (spa->spa_condensing_indirect != NULL)
return (B_FALSE);
if (spa_shutting_down(spa))
return (B_FALSE);
/*
* The mapping object size must not change while we are
* condensing, so we can only condense indirect vdevs
* (not vdevs that are still in the middle of being removed).
*/
if (vd->vdev_ops != &vdev_indirect_ops)
return (B_FALSE);
/*
* If nothing new has been marked obsolete, there is no
* point in condensing.
*/
if (vd->vdev_obsolete_sm == NULL) {
ASSERT0(vdev_obsolete_sm_object(vd));
return (B_FALSE);
}
ASSERT(vd->vdev_obsolete_sm != NULL);
ASSERT3U(vdev_obsolete_sm_object(vd), ==,
space_map_object(vd->vdev_obsolete_sm));
uint64_t bytes_mapped = vdev_indirect_mapping_bytes_mapped(vim);
uint64_t bytes_obsolete = space_map_allocated(vd->vdev_obsolete_sm);
uint64_t mapping_size = vdev_indirect_mapping_size(vim);
uint64_t obsolete_sm_size = space_map_length(vd->vdev_obsolete_sm);
ASSERT3U(bytes_obsolete, <=, bytes_mapped);
/*
* If a high percentage of the bytes that are mapped have become
* obsolete, condense (unless the mapping is already small enough).
* This has a good chance of reducing the amount of memory used
* by the mapping.
*/
if (bytes_obsolete * 100 / bytes_mapped >=
zfs_indirect_condense_obsolete_pct &&
mapping_size > zfs_condense_min_mapping_bytes) {
zfs_dbgmsg("should condense vdev %llu because obsolete "
"spacemap covers %d%% of %lluMB mapping",
(u_longlong_t)vd->vdev_id,
(int)(bytes_obsolete * 100 / bytes_mapped),
(u_longlong_t)bytes_mapped / 1024 / 1024);
return (B_TRUE);
}
/*
* If the obsolete space map takes up too much space on disk,
* condense in order to free up this disk space.
*/
if (obsolete_sm_size >= zfs_condense_max_obsolete_bytes) {
zfs_dbgmsg("should condense vdev %llu because obsolete sm "
"length %lluMB >= max size %lluMB",
(u_longlong_t)vd->vdev_id,
(u_longlong_t)obsolete_sm_size / 1024 / 1024,
(u_longlong_t)zfs_condense_max_obsolete_bytes /
1024 / 1024);
return (B_TRUE);
}
return (B_FALSE);
}