From 1491ffd5dd96d4d3d88ff4442c0881c5bdc771a3 Mon Sep 17 00:00:00 2001
From: tappro <tappro>
Date: Tue, 31 Mar 2009 06:11:30 +0000
Subject: [PATCH] - avoid assertion if second open|create found orphan again  
 b:18927   i:rread,johann

---
 lustre/mds/mds_unlink_open.c  |  6 ++++--
 lustre/tests/replay-dual.sh   | 17 +++++++++++++++++
 lustre/tests/replay-single.sh | 18 ++++++++++++++++++
 3 files changed, 39 insertions(+), 2 deletions(-)

diff --git a/lustre/mds/mds_unlink_open.c b/lustre/mds/mds_unlink_open.c
index d64a586..0e6d15a 100644
--- a/lustre/mds/mds_unlink_open.c
+++ b/lustre/mds/mds_unlink_open.c
@@ -362,8 +362,10 @@ int mds_check_stale_orphan(struct obd_device *obd, struct ll_fid *fid)
         if (mds_inode_is_orphan(inode)) {
                 struct dentry *orphan;
 
-                /* The exactly same inode can't be orphan */
-                LASSERT(inode->i_generation != fid->generation);
+                /* bz18927: The exactly same inode can be marked as orphan
+                 * if there was open|creat replay and this is second one */
+                if (inode->i_generation == fid->generation)
+                        GOTO(unlock_child, rc);
 
                 if (mds_orphan_open_count(inode) > 0) {
                         CERROR("Orphan "LPU64"/%u is in use!\n",
diff --git a/lustre/tests/replay-dual.sh b/lustre/tests/replay-dual.sh
index 122dc6d..eb830dd 100755
--- a/lustre/tests/replay-dual.sh
+++ b/lustre/tests/replay-dual.sh
@@ -442,6 +442,23 @@ test_20() { #16389
 }
 run_test 20 "recovery time is not increasing"
 
+test_22() { #bug 18927
+    multiop_bg_pause $MOUNT1/$tfile O_c || return 1
+    pid1=$!
+    multiop_bg_pause $MOUNT2/$tfile O_c || return 2
+    pid2=$!
+    rm -f $MOUNT1/$tfile
+    replay_barrier mds
+    fail mds
+    kill -USR1 $pid1
+    wait $pid1 || return 3
+    kill -USR1 $pid2
+    wait $pid2 || return 4
+    [ -e $MOUNT1/$tfile ] && return 5
+    return 0
+}
+run_test 22 "double open|creat in replay with open orphan from two mntp"
+
 equals_msg `basename $0`: test complete, cleaning up
 SLEEP=$((`date +%s` - $NOW))
 [ $SLEEP -lt $TIMEOUT ] && sleep $SLEEP
diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index f3ccf4a..a9872f4 100755
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -1978,6 +1978,24 @@ test_81b() {
 }
 run_test 81b "fail log_add during unlink recovery"
 
+test_82() { #bug 18927
+    multiop_bg_pause $DIR/$tfile O_c || return 1
+    pid1=$!
+    multiop_bg_pause $DIR/$tfile o:O_CREAT:O_RDONLY:_c || return 2
+    pid2=$!
+    rm -f $DIR/$tfile
+    replay_barrier mds
+    fail mds
+    kill -USR1 $pid1
+    wait $pid1 || return 3
+    kill -USR1 $pid2
+    wait $pid2 || return 4
+    [ -e $DIR/$tfile ] && return 5
+    return 0
+}
+run_test 82 "second open|creat in replay with open orphan"
+
+
 
 equals_msg `basename $0`: test complete, cleaning up
 check_and_cleanup_lustre
-- 
1.8.3.1