b=19756

r=adilger,shadow

release parent in reconsruct_open when done
release mfd and dentry pointed by it in mds_mfd_close on all errors.

diff --git a/lustre/ChangeLog b/lustre/ChangeLog
index e44e759..8f87c88 100644 (file)
--- a/lustre/ChangeLog
+++ b/lustre/ChangeLog
@@ -61,6 +61,15 @@ Details    : When file is opened, but open reply is lost and file is
              subsequently deleted before resend, resend processing logic 
             breaks trying to open the file again, should not try to open.
 
+Severity   : Low
+Bugzilla   : 19756
+Description: Rare dentry leakage on MDS on forceful shutdown and
+             replay-resends.
+Details    : Sometimes on transaction errors mds_mfd_close did not release
+             mfd and dentry pointed to it. reconstruct_open did not release
+            parent dentry if no matching mfd was found and child lookup was
+            used.
+
 ------------------------------------------------------------------------------
 tbd Sun Microsystems, Inc.
        * version 1.8.1

diff --git a/lustre/mds/mds_open.c b/lustre/mds/mds_open.c
index 4ce6dd8..ef3963e 100644 (file)
--- a/lustre/mds/mds_open.c
+++ b/lustre/mds/mds_open.c
@@ -579,6 +579,7 @@ static void reconstruct_open(struct mds_update_record *rec, int offset,
                 }
 
                 dchild = mds_lookup(obd, rec->ur_name, parent, rec->ur_namelen - 1);
+                l_dput(parent);
                 if (IS_ERR(dchild)) {
                         rc = PTR_ERR(dchild);
                         LCONSOLE_WARN("Child "LPU64"/%u lookup error %d."
@@ -587,7 +588,6 @@ static void reconstruct_open(struct mds_update_record *rec, int offset,
                                       rc, obd_uuid2str(&exp->exp_client_uuid),
                                       obd_export_nid2str(exp));
                         mds_export_evict(exp);
-                        l_dput(parent);
                         EXIT;
                         return;
                 }
@@ -1433,9 +1433,7 @@ found_child:
  * If we are being called from mds_disconnect() because the client has
  * disappeared, then req == NULL and we do not update last_rcvd because
  * there is nothing that could be recovered by the client at this stage
- * (it will not even _have_ an entry in last_rcvd anymore).
- *
- * Returns EAGAIN if the client needs to get more data and re-close. */
+ * (it will not even _have_ an entry in last_rcvd anymore). */
 int mds_mfd_close(struct ptlrpc_request *req, int offset,
                   struct obd_device *obd, struct mds_file_data *mfd,
                   int unlink_orphan, struct lov_mds_md *lmm, int lmm_size,
@@ -1584,10 +1582,10 @@ out:
         /* If other clients have this file open for write, rc will be > 0 */
         if (rc > 0)
                 rc = 0;
-        l_dput(mfd->mfd_dentry);
-        mds_mfd_put(mfd);
 
  cleanup:
+        l_dput(mfd->mfd_dentry);
+        mds_mfd_put(mfd);
         if (req != NULL && reply_body != NULL) {
                 rc = mds_finish_transno(mds, NULL, handle, req, rc, 0, 0);
         } else if (handle) {