DAOS-8082 daos: fix various leak

1. missing ivns reference put in crt_ivf_rpc_issue(). 2. free btree root allocated in tree_cache_create_internal(). 3. pool_target_addr_list is not freed in ds_pool_target_update_state(). 4. free layout of reclaim in rebuild_obj_scan_cb(). 5. Add crt_rank_abort_all() in server_fini() to abort all rpc before cleanup to avoid race. 6. Missing collective req put for a few mgmt handler. Signed-off-by: Di Wang <[email protected]>
daos-stack · Aug 2, 2021 · e452995 · e452995
1 parent 6e8af6c
commit e452995
Show file tree

Hide file tree

Showing 7 changed files with 28 additions and 36 deletions.
diff --git a/src/cart/crt_iv.c b/src/cart/crt_iv.c
@@ -1141,6 +1141,8 @@ crt_ivf_rpc_issue(d_rank_t dest_node, crt_iv_key_t *iv_key,
 			IV_DBG(iv_key, "added to kip_entry=%p\n", entry);
 			D_MUTEX_UNLOCK(&entry->kip_lock);
 			D_MUTEX_UNLOCK(&ivns_internal->cii_lock);
+			if (rc == 0)
+				IVNS_DECREF(ivns_internal);
 			return rc;
 		}
 		IV_DBG(iv_key, "kip_entry=%p present\n", entry);

diff --git a/src/engine/init.c b/src/engine/init.c
@@ -784,6 +784,7 @@ server_init(int argc, char *argv[])
 static void
 server_fini(bool force)
 {
+	crt_rank_abort_all(NULL);
 	D_INFO("Service is shutting down\n");
 	crt_unregister_event_cb(dss_crt_event_cb, NULL);
 	D_INFO("unregister event callbacks done\n");

diff --git a/src/include/daos_srv/daos_engine.h b/src/include/daos_srv/daos_engine.h
@@ -812,7 +812,7 @@ unsigned int dss_ctx_nr_get(void);
 
 /* Cache for container root */
 struct tree_cache_root {
-	struct btr_root	btr_root;
+	struct btr_root	*btr_root;
 	daos_handle_t	root_hdl;
 	unsigned int	count;
 };

diff --git a/src/mgmt/srv.c b/src/mgmt/srv.c
@@ -212,17 +212,8 @@ ds_mgmt_params_set_hdlr(crt_rpc_t *rpc)
 	tc_in->tps_value_extra = ps_in->ps_value_extra;
 
 	rc = dss_rpc_send(tc_req);
-	if (rc != 0) {
-		crt_req_decref(tc_req);
-		D_GOTO(out, rc);
-	}
 
-	out = crt_reply_get(tc_req);
-	rc = out->srv_rc;
-	if (rc != 0) {
-		crt_req_decref(tc_req);
-		D_GOTO(out, rc);
-	}
+	crt_req_decref(tc_req);
 out:
 	out = crt_reply_get(rpc);
 	out->srv_rc = rc;
@@ -262,17 +253,8 @@ ds_mgmt_profile_hdlr(crt_rpc_t *rpc)
 	tc_in->p_op = in->p_op;
 	tc_in->p_avg = in->p_avg;
 	rc = dss_rpc_send(tc_req);
-	if (rc != 0) {
-		crt_req_decref(tc_req);
-		D_GOTO(out, rc);
-	}
 
-	out = crt_reply_get(tc_req);
-	rc = out->p_rc;
-	if (rc != 0) {
-		crt_req_decref(tc_req);
-		D_GOTO(out, rc);
-	}
+	crt_req_decref(tc_req);
 out:
 	D_DEBUG(DB_MGMT, "profile hdlr: rc "DF_RC"\n", DP_RC(rc));
 	out = crt_reply_get(rpc);
@@ -310,17 +292,8 @@ ds_mgmt_mark_hdlr(crt_rpc_t *rpc)
 
 	tc_in->m_mark = in->m_mark;
 	rc = dss_rpc_send(tc_req);
-	if (rc != 0) {
-		crt_req_decref(tc_req);
-		D_GOTO(out, rc);
-	}
 
-	out = crt_reply_get(tc_req);
-	rc = out->m_rc;
-	if (rc != 0) {
-		crt_req_decref(tc_req);
-		D_GOTO(out, rc);
-	}
+	crt_req_decref(tc_req);
 out:
 	D_DEBUG(DB_MGMT, "mark hdlr: rc "DF_RC"\n", DP_RC(rc));
 	out = crt_reply_get(rpc);

diff --git a/src/object/srv_obj_migrate.c b/src/object/srv_obj_migrate.c
@@ -109,7 +109,7 @@ obj_tree_destory_cb(daos_handle_t ih, d_iov_t *key_iov,
 	if (rc)
 		D_ERROR("dbtree_destroy, cont "DF_UUID" failed: "DF_RC"\n",
 			DP_UUID(*(uuid_t *)key_iov->iov_buf), DP_RC(rc));
-
+	D_FREE_PTR(root->btr_root);
 	return rc;
 }
 
@@ -148,14 +148,14 @@ tree_cache_create_internal(daos_handle_t toh, unsigned int tree_class,
 
 	memset(&root, 0, sizeof(root));
 	root.root_hdl = DAOS_HDL_INVAL;
+	root.btr_root = broot;
 	memset(&uma, 0, sizeof(uma));
 	uma.uma_id = UMEM_CLASS_VMEM;
 
 	rc = dbtree_create_inplace(tree_class, BTR_FEAT_DIRECT_KEY, 32,
 				   &uma, broot, &root.root_hdl);
 	if (rc) {
 		D_ERROR("failed to create rebuild tree: "DF_RC"\n", DP_RC(rc));
-		D_FREE(broot);
 		D_GOTO(out, rc);
 	}
 
@@ -171,9 +171,13 @@ tree_cache_create_internal(daos_handle_t toh, unsigned int tree_class,
 
 	*rootp = val_iov.iov_buf;
 	D_ASSERT(*rootp != NULL);
+
 out:
-	if (rc < 0 && daos_handle_is_valid(root.root_hdl))
-		dbtree_destroy(root.root_hdl, NULL);
+	if (rc < 0) {
+		if (daos_handle_is_valid(root.root_hdl))
+			dbtree_destroy(root.root_hdl, NULL);
+		D_FREE(broot);
+	}
 	return rc;
 }
 
@@ -2785,6 +2789,15 @@ migrate_cont_iter_cb(daos_handle_t ih, d_iov_t *key_iov,
 	D_DEBUG(DB_REBUILD, "iter cont "DF_UUID"/%"PRIx64" finish.\n",
 		DP_UUID(cont_uuid), ih.cookie);
 
+	rc = dbtree_destroy(root->root_hdl, NULL);
+	D_FREE_PTR(root->btr_root);
+	if (rc) {
+		/* Ignore the DRAM migrate object tree for the moment, since
+		 * it does not impact the migration on the storage anyway
+		 */
+		D_ERROR("dbtree_destroy failed: "DF_RC"\n", DP_RC(rc));
+	}
+
 	/* Snapshot fetch will yield the ULT, let's reprobe before delete  */
 	d_iov_set(&tmp_iov, cont_uuid, sizeof(uuid_t));
 	rc = dbtree_iter_probe(ih, BTR_PROBE_EQ, DAOS_INTENT_MIGRATION,

diff --git a/src/pool/srv_pool.c b/src/pool/srv_pool.c
@@ -3318,7 +3318,7 @@ ds_pool_target_update_state(uuid_t pool_uuid, d_rank_list_t *ranks,
 	crt_endpoint_t			ep;
 	struct dss_module_info		*info = dss_get_module_info();
 	crt_rpc_t			*rpc;
-	struct pool_target_addr_list	list;
+	struct pool_target_addr_list	list = { 0 };
 	struct pool_add_in		*in;
 	struct pool_add_out		*out;
 	crt_opcode_t			opcode;
@@ -3398,6 +3398,7 @@ ds_pool_target_update_state(uuid_t pool_uuid, d_rank_list_t *ranks,
 out_rpc:
 	crt_req_decref(rpc);
 out_client:
+	pool_target_addr_list_free(&list);
 	rsvc_client_fini(&client);
 	return rc;
 }

diff --git a/src/rebuild/scan.c b/src/rebuild/scan.c
@@ -157,6 +157,7 @@ rebuild_cont_send_cb(daos_handle_t ih, d_iov_t *key_iov,
 	}
 
 	rc = dbtree_destroy(root->root_hdl, NULL);
+	D_FREE_PTR(root->btr_root);
 	if (rc) {
 		D_ERROR("dbtree_destroy failed: "DF_RC"\n", DP_RC(rc));
 		return rc;
@@ -524,6 +525,7 @@ rebuild_obj_scan_cb(daos_handle_t ch, vos_iter_entry_t *ent,
 		still_needed = pl_obj_layout_contains(rpt->rt_pool->sp_map,
 						      layout, myrank, mytarget,
 						      oid.id_shard);
+		pl_obj_layout_free(layout);
 		if (!still_needed) {
 			struct rebuild_pool_tls *tls;