Branch HEAD

b=16522
i=adilger
i=wangdi

enable OBD_CONNECT_MDT flag when connecting from the MDS so that
the OSTs know that the MDS "UUID" can be reused for the same export
from a different NID, so we do not need to wait for the export to be
evicted.

diff --git a/lustre/ChangeLog b/lustre/ChangeLog
index 7976edb..843e919 100644 (file)
--- a/lustre/ChangeLog
+++ b/lustre/ChangeLog
@@ -1887,6 +1887,14 @@ Details    : With AT enabled, the recovery window can be excessively long (6000+
             INITIAL_CONNECT_TIMEOUT now) and clients report the old service
             time via pb_service_time.
 
+Severity   : normal
+Bugzilla   : 16522
+Description: Watchdog triggered on MDS failover
+Details    : enable OBD_CONNECT_MDT flag when connecting from the MDS so that
+            the OSTs know that the MDS "UUID" can be reused for the same export
+            from a different NID, so we do not need to wait for the export to be
+            evicted
+
 --------------------------------------------------------------------------------
 
 2007-08-10         Cluster File Systems, Inc. <info@clusterfs.com>

diff --git a/lustre/include/lustre/lustre_idl.h b/lustre/include/lustre/lustre_idl.h
index eb9cdca..8cac1e6 100644 (file)
--- a/lustre/include/lustre/lustre_idl.h
+++ b/lustre/include/lustre/lustre_idl.h
@@ -667,6 +667,7 @@ extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
 /* Connect flags */
 #define OBD_CONNECT_RDONLY     0x00000001ULL /* client allowed read-only access */
 #define OBD_CONNECT_INDEX      0x00000002ULL /* connect to specific LOV idx */
+#define OBD_CONNECT_MDS        0x00000004ULL /* connect from MDT to OST */
 #define OBD_CONNECT_GRANT      0x00000008ULL /* OSC acquires grant at connect */
 #define OBD_CONNECT_SRVLOCK    0x00000010ULL /* server takes locks for client */
 #define OBD_CONNECT_VERSION    0x00000020ULL /* Server supports versions in ocd */
@@ -723,7 +724,7 @@ extern void lustre_swab_ptlrpc_body(struct ptlrpc_body *pb);
                                 OBD_CONNECT_CKSUM | LRU_RESIZE_CONNECT_FLAG | \
                                 OBD_CONNECT_AT | OBD_CONNECT_CHANGE_QS | \
                                 OBD_CONNECT_RMT_CLIENT | \
-                                OBD_CONNECT_RMT_CLIENT_FORCE)
+                                OBD_CONNECT_RMT_CLIENT_FORCE | OBD_CONNECT_MDS)
 #define ECHO_CONNECT_SUPPORTED (0)
 #define MGS_CONNECT_SUPPORTED  (OBD_CONNECT_VERSION | OBD_CONNECT_AT)
 

diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c
index 0976241..a71aa3d 100644 (file)
--- a/lustre/ldlm/ldlm_lib.c
+++ b/lustre/ldlm/ldlm_lib.c
@@ -737,14 +737,25 @@ int target_handle_connect(struct ptlrpc_request *req)
                    req->rq_peer.nid != export->exp_connection->c_peer.nid) {
                 /* make darn sure this is coming from the same peer
                  * if the UUIDs matched */
-                  CWARN("%s: cookie %s seen on new NID %s when "
-                          "existing NID %s is already connected\n",
-                        target->obd_name, cluuid.uuid,
-                  libcfs_nid2str(req->rq_peer.nid),
-                  libcfs_nid2str(export->exp_connection->c_peer.nid));
-                  class_export_put(export);
-                  export = NULL;
-                  rc = -EALREADY;
+                if (data && data->ocd_connect_flags & OBD_CONNECT_MDS) {
+                        /* the MDS UUID can be reused, don't need to wait
+                         * for the export to be evicted */
+                        CWARN("%s: received MDS connection from a new NID %s,"
+                              " removing former export from NID %s\n",
+                            target->obd_name,
+                            libcfs_nid2str(req->rq_peer.nid),
+                            libcfs_nid2str(export->exp_connection->c_peer.nid));
+                        class_fail_export(export);
+                } else {
+                        CWARN("%s: cookie %s seen on new NID %s when "
+                              "existing NID %s is already connected\n",
+                            target->obd_name, cluuid.uuid,
+                            libcfs_nid2str(req->rq_peer.nid),
+                            libcfs_nid2str(export->exp_connection->c_peer.nid));
+                        rc = -EALREADY;
+                }
+                class_export_put(export);
+                export = NULL;
         } else if (export != NULL) {
                 spin_lock(&export->exp_lock);
                 export->exp_connecting = 1;

diff --git a/lustre/mds/mds_lov.c b/lustre/mds/mds_lov.c
index 7886c1c..95d10a4 100644 (file)
--- a/lustre/mds/mds_lov.c
+++ b/lustre/mds/mds_lov.c
@@ -643,7 +643,8 @@ int mds_lov_connect(struct obd_device *obd, char * lov_name)
                                   OBD_CONNECT_REQPORTAL | OBD_CONNECT_QUOTA64 |
                                   OBD_CONNECT_OSS_CAPA  | OBD_CONNECT_FID     |
                                   OBD_CONNECT_BRW_SIZE  | OBD_CONNECT_CKSUM   |
-                                  OBD_CONNECT_AT | OBD_CONNECT_CHANGE_QS;
+                                  OBD_CONNECT_CHANGE_QS | OBD_CONNECT_AT      |
+                                  OBD_CONNECT_MDS;
 #ifdef HAVE_LRU_RESIZE_SUPPORT
         data->ocd_connect_flags |= OBD_CONNECT_LRU_RESIZE;
 #endif