From: alex <alex>
Date: Sat, 7 May 2005 12:23:17 +0000 (+0000)
Subject: b=4920
X-Git-Tag: v1_7_100~1366
X-Git-Url: https://git.whamcloud.com/?p=fs%2Flustre-release.git;a=commitdiff_plain;h=1f51ba95a2184b8e064ade1689480bbe4b17df95

b=4920

 - reconnect after recovery to all mds nodes. otherwise some mds will
   miss reconnect after recovery and get to recovery timeout situation
---

diff --git a/lustre/include/linux/obd.h b/lustre/include/linux/obd.h
index 684f0f7..ac5d138 100644
--- a/lustre/include/linux/obd.h
+++ b/lustre/include/linux/obd.h
@@ -717,7 +717,7 @@ struct obd_ops {
         int (*o_connect)(struct lustre_handle *conn, struct obd_device *src,
                          struct obd_uuid *cluuid, struct obd_connect_data *data,
                          unsigned long flags);
-        int (*o_connect_post)(struct obd_export *exp, unsigned long flags);
+        int (*o_connect_post)(struct obd_export *exp, unsigned, unsigned long);
         int (*o_disconnect)(struct obd_export *exp, unsigned long flags);
 
         int (*o_statfs)(struct obd_device *obd, struct obd_statfs *osfs,
diff --git a/lustre/include/linux/obd_class.h b/lustre/include/linux/obd_class.h
index 2c1e13c..9f41ed2 100644
--- a/lustre/include/linux/obd_class.h
+++ b/lustre/include/linux/obd_class.h
@@ -609,6 +609,7 @@ static inline int obd_connect(struct lustre_handle *conn,
 }
 
 static inline int obd_connect_post(struct obd_export *exp,
+                                   unsigned initial,
                                    unsigned long flags)
 {
         int rc;
@@ -618,7 +619,7 @@ static inline int obd_connect_post(struct obd_export *exp,
         if (!OBT(exp->exp_obd) || !OBP((exp->exp_obd), connect_post))
                 RETURN(0);
         OBD_COUNTER_INCREMENT(exp->exp_obd, connect_post);
-        rc = OBP(exp->exp_obd, connect_post)(exp, flags);
+        rc = OBP(exp->exp_obd, connect_post)(exp, initial, flags);
         RETURN(rc);
 }
 
diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c
index 9898297..a4ea476 100644
--- a/lustre/ldlm/ldlm_lib.c
+++ b/lustre/ldlm/ldlm_lib.c
@@ -805,7 +805,7 @@ int target_handle_connect(struct ptlrpc_request *req)
 
         class_import_put(revimp);
 
-        rc = obd_connect_post(export, connect_flags);
+        rc = obd_connect_post(export, initial_conn, connect_flags);
 out:
         if (rc)
                 req->rq_status = rc;
diff --git a/lustre/mds/handler.c b/lustre/mds/handler.c
index bf999c0..c6e48cf 100644
--- a/lustre/mds/handler.c
+++ b/lustre/mds/handler.c
@@ -562,13 +562,30 @@ out:
         return rc;
 }
 
-static int mds_connect_post(struct obd_export *exp, unsigned long flags)
+static int mds_connect_post(struct obd_export *exp, unsigned initial,
+                            unsigned long flags)
 {
         struct obd_device *obd = exp->exp_obd;
         struct mds_obd *mds = &obd->u.mds;
+        struct mds_export_data *med;
+        struct mds_client_data *mcd;
         int rc = 0;
         ENTRY;
 
+        med = &exp->exp_mds_data;
+        mcd = med->med_mcd;
+
+        if (initial) {
+                /* some one reconnect initially, we have to reset
+                 * data existing export can have. bug 6102 */
+                if (mcd->mcd_last_xid != 0)
+                        CDEBUG(D_HA, "initial reconnect to existing export\n");
+                mcd->mcd_last_transno = 0;
+                mcd->mcd_last_xid = 0;
+                mcd->mcd_last_result = 0;
+                mcd->mcd_last_data = 0;
+        }
+
         if (!(flags & OBD_OPT_MDS_CONNECTION)) {
                 if (!(exp->exp_flags & OBD_OPT_REAL_CLIENT)) {
                         atomic_inc(&mds->mds_real_clients);
@@ -3313,7 +3330,7 @@ err_llog:
         return rc;
 }
 
-int mds_postrecov(struct obd_device *obd)
+int mds_postrecov_common(struct obd_device *obd)
 {
         struct mds_obd *mds = &obd->u.mds;
         struct llog_ctxt *ctxt;
@@ -3371,6 +3388,16 @@ err_llog:
         goto out;
 }
 
+int mds_postrecov(struct obd_device *obd)
+{
+        int rc;
+        ENTRY;
+        rc = mds_postrecov_common(obd);
+        if (rc == 0)
+                rc = mds_md_reconnect(obd);
+        RETURN(rc);
+}
+
 int mds_dt_clean(struct obd_device *obd)
 {
         struct mds_obd *mds = &obd->u.mds;
diff --git a/lustre/mds/mds_internal.h b/lustre/mds/mds_internal.h
index 2dc75a7..b01baaf 100644
--- a/lustre/mds/mds_internal.h
+++ b/lustre/mds/mds_internal.h
@@ -176,6 +176,7 @@ int mds_handle(struct ptlrpc_request *req);
 extern struct lvfs_callback_ops mds_lvfs_ops;
 int mds_dt_clean(struct obd_device *obd);
 int mds_postrecov(struct obd_device *obd);
+int mds_postrecov_common(struct obd_device *obd);
 extern struct lvfs_callback_ops mds_lvfs_ops;
 
 extern int mds_iocontrol(unsigned int cmd,
@@ -259,6 +260,7 @@ void mds_unlock_slave_objs(struct obd_device *, struct dentry *,
 int mds_lock_and_check_slave(int, struct ptlrpc_request *, struct lustre_handle *);
 int mds_convert_mea_ea(struct obd_device *, struct inode *, struct lov_mds_md *, int);
 int mds_is_dir_empty(struct obd_device *, struct dentry *);
+int mds_md_reconnect(struct obd_device *obd);
 
 /* mds_lsd.c */
 struct upcall_cache *__mds_get_global_lsd_cache(void);
diff --git a/lustre/mds/mds_lmv.c b/lustre/mds/mds_lmv.c
index d988f6a..c675654 100644
--- a/lustre/mds/mds_lmv.c
+++ b/lustre/mds/mds_lmv.c
@@ -208,6 +208,28 @@ int mds_md_disconnect(struct obd_device *obd, int flags)
         RETURN(rc);
 }
 
+int mds_md_reconnect(struct obd_device *obd)
+{
+        struct mds_obd *mds = &obd->u.mds;
+        struct obd_statfs osfs;
+        int err;
+        ENTRY;
+
+        /* We don't know state of connections to another MDSes
+         * before the failure. If MDS we were connected to before
+         * the failure gets failed, then it will wait for us to
+         * reconnect and will timed recovery out. bug 4920 */
+        if (mds->mds_md_connected == 0)
+                RETURN(0);
+        if (mds->mds_md_obd == NULL)
+                RETURN(0);
+
+        err = obd_statfs(mds->mds_md_obd, &osfs, jiffies - HZ);
+        if (err)
+                CERROR("can't reconnect to MDSes after recovery: %d\n", err);
+        RETURN(0);
+}
+
 int mds_md_get_attr(struct obd_device *obd, struct inode *inode,
                     struct mea **mea, int *mea_size)
 {
diff --git a/lustre/mds/mds_lov.c b/lustre/mds/mds_lov.c
index d5070b2..fbc6155 100644
--- a/lustre/mds/mds_lov.c
+++ b/lustre/mds/mds_lov.c
@@ -312,7 +312,7 @@ int mds_dt_connect(struct obd_device *obd, char * lov_name)
          * it can use the obd_recovering flag to determine when the
          * the OBD is full available. */
         if (!obd->obd_recovering)
-                rc = mds_postrecov(obd);
+                rc = mds_postrecov_common(obd);
         RETURN(rc);
 
 err_reg:
diff --git a/lustre/obdfilter/filter.c b/lustre/obdfilter/filter.c
index 16a5a70..c5dbac8 100644
--- a/lustre/obdfilter/filter.c
+++ b/lustre/obdfilter/filter.c
@@ -1634,7 +1634,8 @@ static int filter_cleanup(struct obd_device *obd, int flags)
         RETURN(0);
 }
 
-static int filter_connect_post(struct obd_export *exp, unsigned long connect_flags)
+static int filter_connect_post(struct obd_export *exp, unsigned initial,
+                               unsigned long connect_flags)
 {
         struct obd_device *obd = exp->exp_obd;
         struct filter_export_data *fed;