5269 zfs: zpool import slow
While importing a pool all objsets are enumerated twice, once to check
the zil log chains and once to claim them. On pools with many datasets
this process might take a substantial amount of time.
Speed up the process by parallelizing it utilizing a taskq. The number
of parallel tasks is limited to 4 times the number of leaf vdevs.

@@ -264,10 +264,11 @@
  */
 int dmu_objset_hold(const char *name, void *tag, objset_t **osp);
 int dmu_objset_own(const char *name, dmu_objset_type_t type,
     boolean_t readonly, void *tag, objset_t **osp);
 void dmu_objset_rele(objset_t *os, void *tag);
+void dmu_objset_rele_obj(objset_t *os, void *tag);
 void dmu_objset_disown(objset_t *os, void *tag);
 int dmu_objset_open_ds(struct dsl_dataset *ds, objset_t **osp);
 
 void dmu_objset_evict_dbufs(objset_t *os);
 int dmu_objset_create(const char *name, dmu_objset_type_t type, uint64_t flags,