int
sa_handle_get_from_db(objset_t *os, dmu_buf_t *db, void *userp,
sa_handle_type_t hdl_type, sa_handle_t **handlepp)
{
int error = 0;
dmu_object_info_t doi;
sa_handle_t *handle;
#ifdef ZFS_DEBUG
dmu_object_info_from_db(db, &doi);
ASSERT(doi.doi_bonus_type == DMU_OT_SA ||
doi.doi_bonus_type == DMU_OT_ZNODE);
#endif
/* find handle, if it exists */
/* if one doesn't exist then create a new one, and initialize it */
handle = (hdl_type == SA_HDL_SHARED) ? dmu_buf_get_user(db) : NULL;
if (handle == NULL) {
sa_handle_t *newhandle;
handle = kmem_cache_alloc(sa_cache, KM_SLEEP);
handle->sa_userp = userp;
handle->sa_bonus = db;
handle->sa_os = os;
handle->sa_spill = NULL;
error = sa_build_index(handle, SA_BONUS);
newhandle = (hdl_type == SA_HDL_SHARED) ?
dmu_buf_set_user_ie(db, handle,
NULL, sa_evict) : NULL;
if (newhandle != NULL) {
kmem_cache_free(sa_cache, handle);
handle = newhandle;
}
}
*handlepp = handle;
return (error);
}
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);
}
int
zfs_zget(zfs_sb_t *zsb, uint64_t obj_num, znode_t **zpp)
{
dmu_object_info_t doi;
dmu_buf_t *db;
znode_t *zp;
int err;
sa_handle_t *hdl;
*zpp = NULL;
again:
ZFS_OBJ_HOLD_ENTER(zsb, obj_num);
err = sa_buf_hold(zsb->z_os, obj_num, NULL, &db);
if (err) {
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (err);
}
dmu_object_info_from_db(db, &doi);
if (doi.doi_bonus_type != DMU_OT_SA &&
(doi.doi_bonus_type != DMU_OT_ZNODE ||
(doi.doi_bonus_type == DMU_OT_ZNODE &&
doi.doi_bonus_size < sizeof (znode_phys_t)))) {
sa_buf_rele(db, NULL);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (SET_ERROR(EINVAL));
}
hdl = dmu_buf_get_user(db);
if (hdl != NULL) {
zp = sa_get_userdata(hdl);
/*
* Since "SA" does immediate eviction we
* should never find a sa handle that doesn't
* know about the znode.
*/
ASSERT3P(zp, !=, NULL);
mutex_enter(&zp->z_lock);
ASSERT3U(zp->z_id, ==, obj_num);
if (zp->z_unlinked) {
err = SET_ERROR(ENOENT);
} else {
/*
* If igrab() returns NULL the VFS has independently
* determined the inode should be evicted and has
* called iput_final() to start the eviction process.
* The SA handle is still valid but because the VFS
* requires that the eviction succeed we must drop
* our locks and references to allow the eviction to
* complete. The zfs_zget() may then be retried.
*
* This unlikely case could be optimized by registering
* a sops->drop_inode() callback. The callback would
* need to detect the active SA hold thereby informing
* the VFS that this inode should not be evicted.
*/
if (igrab(ZTOI(zp)) == NULL) {
mutex_exit(&zp->z_lock);
sa_buf_rele(db, NULL);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
goto again;
}
*zpp = zp;
err = 0;
}
mutex_exit(&zp->z_lock);
sa_buf_rele(db, NULL);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (err);
}
int
zfs_zget(zfs_sb_t *zsb, uint64_t obj_num, znode_t **zpp)
{
dmu_object_info_t doi;
dmu_buf_t *db;
znode_t *zp;
int err;
sa_handle_t *hdl;
struct inode *ip;
*zpp = NULL;
again:
ip = ilookup(zsb->z_sb, obj_num);
ZFS_OBJ_HOLD_ENTER(zsb, obj_num);
err = sa_buf_hold(zsb->z_os, obj_num, NULL, &db);
if (err) {
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
iput(ip);
return (err);
}
dmu_object_info_from_db(db, &doi);
if (doi.doi_bonus_type != DMU_OT_SA &&
(doi.doi_bonus_type != DMU_OT_ZNODE ||
(doi.doi_bonus_type == DMU_OT_ZNODE &&
doi.doi_bonus_size < sizeof (znode_phys_t)))) {
sa_buf_rele(db, NULL);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
iput(ip);
return (SET_ERROR(EINVAL));
}
hdl = dmu_buf_get_user(db);
if (hdl != NULL) {
if (ip == NULL) {
/*
* ilookup returned NULL, which means
* the znode is dying - but the SA handle isn't
* quite dead yet, we need to drop any locks
* we're holding, re-schedule the task and try again.
*/
sa_buf_rele(db, NULL);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
schedule();
goto again;
}
zp = sa_get_userdata(hdl);
/*
* Since "SA" does immediate eviction we
* should never find a sa handle that doesn't
* know about the znode.
*/
ASSERT3P(zp, !=, NULL);
mutex_enter(&zp->z_lock);
ASSERT3U(zp->z_id, ==, obj_num);
if (zp->z_unlinked) {
err = SET_ERROR(ENOENT);
} else {
igrab(ZTOI(zp));
*zpp = zp;
err = 0;
}
sa_buf_rele(db, NULL);
mutex_exit(&zp->z_lock);
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
iput(ip);
return (err);
}
ASSERT3P(ip, ==, NULL);
/*
* Not found create new znode/vnode but only if file exists.
*
* There is a small window where zfs_vget() could
* find this object while a file create is still in
* progress. This is checked for in zfs_znode_alloc()
*
* if zfs_znode_alloc() fails it will drop the hold on the
* bonus buffer.
*/
zp = zfs_znode_alloc(zsb, db, doi.doi_data_block_size,
doi.doi_bonus_type, obj_num, NULL, NULL);
if (zp == NULL) {
err = SET_ERROR(ENOENT);
} else {
*zpp = zp;
}
ZFS_OBJ_HOLD_EXIT(zsb, obj_num);
return (err);
}
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);
}
#endif
/* XXX assert bonus buffer size is correct */
if (dd == NULL) {
dsl_dir_t *winner;
#ifndef __APPLE__
int err;
#endif
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;
dd->dd_used_bytes = dd->dd_phys->dd_used_bytes;
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));
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) {
mutex_destroy(&dd->dd_lock);
kmem_free(dd, sizeof (dsl_dir_t));
dmu_buf_rele(dbuf, tag);
return (err);
}
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) {
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);
}
} else {
(void) strcpy(dd->dd_myname, spa_name(dp->dp_spa));
}
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);
}
int
zap_lockdir(objset_t *os, uint64_t obj, dmu_tx_t *tx,
krw_t lti, int fatreader, zap_t **zapp)
{
zap_t *zap;
dmu_buf_t *db;
krw_t lt;
int err;
*zapp = NULL;
err = dmu_buf_hold(os, obj, 0, NULL, &db);
if (err)
return (err);
#ifdef ZFS_DEBUG
{
dmu_object_info_t doi;
dmu_object_info_from_db(db, &doi);
ASSERT(dmu_ot[doi.doi_type].ot_byteswap == zap_byteswap);
}
#endif
zap = dmu_buf_get_user(db);
if (zap == NULL)
zap = mzap_open(os, obj, db);
/*
* We're checking zap_ismicro without the lock held, in order to
* tell what type of lock we want. Once we have some sort of
* lock, see if it really is the right type. In practice this
* can only be different if it was upgraded from micro to fat,
* and micro wanted WRITER but fat only needs READER.
*/
lt = (!zap->zap_ismicro && fatreader) ? RW_READER : lti;
rw_enter(&zap->zap_rwlock, lt);
if (lt != ((!zap->zap_ismicro && fatreader) ? RW_READER : lti)) {
/* it was upgraded, now we only need reader */
ASSERT(lt == RW_WRITER);
ASSERT(RW_READER ==
(!zap->zap_ismicro && fatreader) ? RW_READER : lti);
rw_downgrade(&zap->zap_rwlock);
lt = RW_READER;
}
zap->zap_objset = os;
if (lt == RW_WRITER)
dmu_buf_will_dirty(db, tx);
ASSERT3P(zap->zap_dbuf, ==, db);
ASSERT(!zap->zap_ismicro ||
zap->zap_m.zap_num_entries <= zap->zap_m.zap_num_chunks);
if (zap->zap_ismicro && tx &&
zap->zap_m.zap_num_entries == zap->zap_m.zap_num_chunks) {
uint64_t newsz = db->db_size + SPA_MINBLOCKSIZE;
if (newsz > MZAP_MAX_BLKSZ) {
dprintf("upgrading obj %llu: num_entries=%u\n",
obj, zap->zap_m.zap_num_entries);
mzap_upgrade(zap, tx);
*zapp = zap;
return (0);
}
err = dmu_object_set_blocksize(os, obj, newsz, 0, tx);
ASSERT3U(err, ==, 0);
zap->zap_m.zap_num_chunks =
db->db_size / MZAP_ENT_LEN - 1;
}