From: Alex Zhuravlev <bzzz@whamcloud.com>
Date: Fri, 17 Feb 2023 08:00:20 +0000 (+0300)
Subject: LU-16478 target: disconnected export
X-Git-Url: https://git.whamcloud.com/?a=commitdiff_plain;h=6a4687a4998a4e78fcfe5061c409af5459cc9e8a;p=fs%2Flustre-release.git

LU-16478 target: disconnected export

eviction can race with a reconnect and this in turn can lead
to a leaked export reference prevent further umount -
mdt_obd_reconnect() grabs a reference via nodemap_add_member().
call obd_disconnect() if such a case observed to balance
obd_reconnect().

Lustre-change: https://review.whamcloud.com/50041
Lustre-commit: 654d5f3fa4df2a0f7275a6da0f050a18881f4f75

Signed-off-by: Alex Zhuravlev <bzzz@whamcloud.com>
Change-Id: I3fd49429ef40ef391d58e042e091258dcb9add72
Reviewed-by: Andreas Dilger <adilger@whamcloud.com>
Reviewed-by: Sebastien Buisson <sbuisson@ddn.com>
Reviewed-on: https://review.whamcloud.com/c/ex/lustre-release/+/50427
Tested-by: jenkins <devops@whamcloud.com>
Tested-by: Maloo <maloo@whamcloud.com>
---

diff --git a/lustre/include/obd_support.h b/lustre/include/obd_support.h
index ae0c678..c3ade18 100644
--- a/lustre/include/obd_support.h
+++ b/lustre/include/obd_support.h
@@ -255,8 +255,9 @@ extern char obd_jobid_var[];
 #define OBD_FAIL_MDS_NO_LL_GETATTR	 0x170
 #define OBD_FAIL_MDS_NO_LL_OPEN		 0x171
 #define OBD_FAIL_MDS_LL_BLOCK		 0x172
-#define OBD_FAIL_MDS_LL_PCCRO		 0x173
-#define OBD_FAIL_MDS_LOD_CREATE_PAUSE	 0x174
+#define OBD_FAIL_MDS_LOD_CREATE_PAUSE	 0x173
+#define OBD_FAIL_MDS_CONNECT_VS_EVICT	 0x174
+#define OBD_FAIL_MDS_LL_PCCRO		 0x17a
 
 /* CMD */
 #define OBD_FAIL_MDS_IS_SUBDIR_NET       0x180
diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c
index 860e7fc..4bb4feb 100644
--- a/lustre/ldlm/ldlm_lib.c
+++ b/lustre/ldlm/ldlm_lib.c
@@ -1112,6 +1112,7 @@ int target_handle_connect(struct ptlrpc_request *req)
 	int size, tmpsize;
 	lnet_nid_t *client_nid = NULL;
 	struct ptlrpc_connection *pcon = NULL;
+	bool reconnected = false;
 
 	ENTRY;
 
@@ -1455,8 +1456,15 @@ dont_check_exports:
 				new_mds_mds_conn = true;
 		}
 	} else {
+		if (CFS_FAIL_CHECK(OBD_FAIL_MDS_CONNECT_VS_EVICT)) {
+			class_export_get(export);
+			class_fail_export(export);
+			class_export_put(export);
+		}
 		rc = obd_reconnect(req->rq_svc_thread->t_env,
 				   export, target, &cluuid, data, client_nid);
+		if (rc == 0)
+			reconnected = true;
 	}
 	if (rc)
 		GOTO(out, rc);
@@ -1499,6 +1507,15 @@ dont_check_exports:
 
 	if (export->exp_disconnected) {
 		spin_unlock(&export->exp_lock);
+		if (reconnected) {
+			/*
+			 * for each connect called disconnect
+			 * should be called to cleanup stuff
+			 */
+			class_export_get(export);
+			obd_disconnect(export);
+		}
+
 		GOTO(out, rc = -ENODEV);
 	}
 	if (export->exp_conn_cnt >= lustre_msg_get_conn_cnt(req->rq_reqmsg)) {
diff --git a/lustre/target/tgt_handler.c b/lustre/target/tgt_handler.c
index eebde9f..9298c3f 100644
--- a/lustre/target/tgt_handler.c
+++ b/lustre/target/tgt_handler.c
@@ -1120,6 +1120,12 @@ int tgt_obd_ping(struct tgt_session_info *tsi)
 	if (rc)
 		RETURN(err_serious(rc));
 
+	if (CFS_FAIL_CHECK(OBD_FAIL_MDS_CONNECT_VS_EVICT)) {
+		if (strstr(tsi->tsi_exp->exp_obd->obd_name, "MDT0000") &&
+		    (exp_connect_flags(tsi->tsi_exp) & OBD_CONNECT_MDS_MDS))
+			tsi->tsi_pill->rc_req->rq_no_reply = 1;
+	}
+
 	RETURN(rc);
 }
 EXPORT_SYMBOL(tgt_obd_ping);
diff --git a/lustre/tests/recovery-small.sh b/lustre/tests/recovery-small.sh
index 67731b3..94dc756 100755
--- a/lustre/tests/recovery-small.sh
+++ b/lustre/tests/recovery-small.sh
@@ -3302,6 +3302,20 @@ test_152() {
 }
 run_test 152 "QoS object allocation could be awakened in case of OST failover"
 
+test_153() {
+#define OBD_FAIL_MDS_CONNECT_VS_EVICT   0x174
+	do_facet mds1 "$LCTL set_param fail_loc=0x174"
+	# first drop ping reply from MDS and then
+	# evict on the subsequent reconnect
+	# (see target_handle_connect)
+	sleep $((TIMEOUT + 3))
+	stop mds1
+	do_facet mds1 "$LCTL set_param fail_loc=0"
+	start mds1 $(mdsdevname 1) $MDS_MOUNT_OPTS ||
+		error "Fail to start $SINGLEMDS"
+}
+run_test 153 "evict vs reconnect race"
+
 complete $SECONDS
 check_and_cleanup_lustre
 exit_status