LU-1129 obdfilter: handle race condition of recreating objects

During OST recovery, a race can happen while handling replayed
OST_WRITE request during the MDS->OST orphan recovery period to
recreate missing objects, which can trigger ASSERTION(diff >= 0)
failure.

This patch handles the above issue by adding obd->obd_recovering
into the assertion to check whether the OST is in recovery or not.
If it's in recovery and diff < 0, then no assertion failure occurs,
the object has been recreated. If the OST is not in recovery and
diff < 0, then the assertion failure occurs.

Test-Parameters: envdefinitions=SLOW=yes,ENABLE_QUOTA=yes testlist=sanity-quota
Signed-off-by: Jian Yu <jian.yu@intel.com>
Change-Id: I346875378802385a95b0832b76d19f9957910cdf
Reviewed-on: http://review.whamcloud.com/5013
Tested-by: Hudson
Tested-by: Maloo <whamcloud.maloo@gmail.com>
Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com>
Reviewed-by: Fan Yong <fan.yong@intel.com>
Reviewed-by: Johann Lombardi <johann.lombardi@intel.com>

diff --git a/lustre/obdfilter/filter.c b/lustre/obdfilter/filter.c
index 828475c..cb2cf77 100644 (file)
--- a/lustre/obdfilter/filter.c
+++ b/lustre/obdfilter/filter.c
@@ -3221,8 +3221,15 @@ static int filter_handle_precreate(struct obd_export *exp, struct obdo *oa,
                 CDEBUG(D_RPCTRACE, "filter_last_id() = "LPU64" -> diff = %d\n",
                        filter_last_id(filter, group), diff);
 
-                LASSERTF(diff >= 0,"%s: "LPU64" - "LPU64" = %d\n",obd->obd_name,
-                         oa->o_id, filter_last_id(filter, group), diff);
+               /*
+                * Check obd->obd_recovering to handle the race condition
+                * while recreating missing precreated objects through
+                * filter_preprw_write() and mds_lov_clear_orphans()
+                * at the same time.
+                */
+               LASSERTF(ergo(!obd->obd_recovering, diff >= 0),
+                        "%s: "LPU64" - "LPU64" = %d\n", obd->obd_name,
+                        oa->o_id, filter_last_id(filter, group), diff);
         }
 
         if (diff > 0) {