From 285ecd13a73ae69e775f851df95abb4390db0b94 Mon Sep 17 00:00:00 2001
From: huanghua <huanghua>
Date: Fri, 6 Feb 2009 02:27:56 +0000
Subject: [PATCH] Branch b1_6

b=3633
i=robert.read
i=oleg.drokin
i=tappro

clear the "hold for replay" flag for open request once a file is closed
---
 lustre/mdc/mdc_request.c      | 55 +++++++++++--------------------------------
 lustre/tests/replay-single.sh | 41 ++++++++++++++++++++++++++++++++
 2 files changed, 55 insertions(+), 41 deletions(-)

diff --git a/lustre/mdc/mdc_request.c b/lustre/mdc/mdc_request.c
index e80c7d8..e5bc0b2 100644
--- a/lustre/mdc/mdc_request.c
+++ b/lustre/mdc/mdc_request.c
@@ -629,36 +629,6 @@ void mdc_clear_open_replay_data(struct obd_client_handle *och)
         och->och_mod = NULL;
 }
 
-static void mdc_commit_close(struct ptlrpc_request *req)
-{
-        struct mdc_open_data *mod = req->rq_cb_data;
-        struct ptlrpc_request *open_req;
-        struct obd_import *imp = req->rq_import;
-
-        DEBUG_REQ(D_RPCTRACE, req, "close req committed");
-        if (mod == NULL)
-                return;
-
-        mod->mod_close_req = NULL;
-        req->rq_cb_data = NULL;
-        req->rq_commit_cb = NULL;
-
-        open_req = mod->mod_open_req;
-        LASSERT(open_req != NULL);
-        LASSERT(open_req != LP_POISON);
-        LASSERT(open_req->rq_type != LI_POISON);
-
-        DEBUG_REQ(D_RPCTRACE, open_req, "open req balanced");
-        LASSERT(open_req->rq_transno != 0);
-        LASSERT(open_req->rq_import == imp);
-
-        /* We no longer want to preserve this for transno-unconditional
-         * replay. */
-        spin_lock(&open_req->rq_lock);
-        open_req->rq_replay = 0;
-        spin_unlock(&open_req->rq_lock);
-}
-
 int mdc_close(struct obd_export *exp, struct obdo *oa,
               struct obd_client_handle *och, struct ptlrpc_request **request)
 {
@@ -671,7 +641,6 @@ int mdc_close(struct obd_export *exp, struct obdo *oa,
                                obd->u.cli.cl_max_mds_cookiesize };
         int rc;
         struct ptlrpc_request *req;
-        struct mdc_open_data *mod;
         ENTRY;
 
         req = ptlrpc_prep_req(class_exp2cliimp(exp), LUSTRE_MDS_VERSION,
@@ -688,17 +657,24 @@ int mdc_close(struct obd_export *exp, struct obdo *oa,
         /* Ensure that this close's handle is fixed up during replay. */
         LASSERT(och != NULL);
         LASSERT(och->och_magic == OBD_CLIENT_HANDLE_MAGIC);
-        mod = och->och_mod;
-        if (likely(mod != NULL)) {
-                if (mod->mod_open_req->rq_type == LI_POISON) {
-                        CERROR("LBUG POISONED open %p!\n", mod->mod_open_req);
+        if (likely(och->och_mod != NULL)) {
+                struct ptlrpc_request *open_req = och->och_mod->mod_open_req;;
+
+                if (open_req->rq_type == LI_POISON) {
+                        CERROR("LBUG POISONED open %p!\n", open_req);
                         LBUG();
                         ptlrpc_req_finished(req);
                         req = NULL;
                         GOTO(out, rc = -EIO);
                 }
-                mod->mod_close_req = req;
-                DEBUG_REQ(D_RPCTRACE, mod->mod_open_req, "matched open");
+                DEBUG_REQ(D_RPCTRACE, req, "close req");
+                DEBUG_REQ(D_RPCTRACE, open_req, "clear open replay");
+
+                /* We no longer want to preserve this open for replay even
+                 * though the open was committed. b=3632, b=3633 */
+                spin_lock(&open_req->rq_lock);
+                open_req->rq_replay = 0;
+                spin_unlock(&open_req->rq_lock);
         } else {
                 CDEBUG(D_RPCTRACE, "couldn't find open req; expecting error\n");
         }
@@ -706,9 +682,6 @@ int mdc_close(struct obd_export *exp, struct obdo *oa,
         mdc_close_pack(req, REQ_REC_OFF, oa, oa->o_valid, och);
 
         ptlrpc_req_set_repsize(req, 4, repsize);
-        req->rq_commit_cb = mdc_commit_close;
-        LASSERT(req->rq_cb_data == NULL);
-        req->rq_cb_data = mod;
 
         mdc_get_rpc_lock(obd->u.cli.cl_close_lock, NULL);
         rc = ptlrpc_queue_wait(req);
@@ -726,7 +699,7 @@ int mdc_close(struct obd_export *exp, struct obdo *oa,
                                   "= %d", rc);
                         if (rc > 0)
                                 rc = -rc;
-                } else if (mod == NULL) {
+                } else if (och->och_mod == NULL) {
                         CERROR("Unexpected: can't find mdc_open_data, but the "
                                "close succeeded.  Please tell <http://bugzilla.lustre.org/>.\n");
                 }
diff --git a/lustre/tests/replay-single.sh b/lustre/tests/replay-single.sh
index 15428fc..3a4e2bd 100755
--- a/lustre/tests/replay-single.sh
+++ b/lustre/tests/replay-single.sh
@@ -971,6 +971,47 @@ test_48() {
 }
 run_test 48 "MDS->OSC failure during precreate cleanup (2824)"
 
+test_49a() {
+    multiop $DIR/$tfile O_c &
+    pid=$!
+    sleep 1
+    rm -rf $DIR/$tfile
+
+#define OBD_FAIL_MDS_ALL_REPLY_NET       0x122
+    do_facet mds "lctl set_param fail_loc=0x80000122"
+    kill -USR1 $pid
+    do_facet mds "lctl set_param fail_loc=0"
+
+    replay_barrier_nodf mds
+    fail mds
+
+    wait $pid || return 1
+    $CHECKSTAT -t file $DIR/$tfile && return 2
+    return 0
+}
+run_test 49a "mds fail after close reply is dropped: open|create "
+
+test_49c() {
+    touch $DIR/$tfile
+    multiop $DIR/$tfile o_c &
+    pid=$!
+    sleep 1
+    rm -rf $DIR/$tfile
+
+#define OBD_FAIL_MDS_ALL_REPLY_NET       0x122
+    do_facet mds "lctl set_param fail_loc=0x80000122"
+    kill -USR1 $pid
+    do_facet mds "lctl set_param fail_loc=0"
+
+    replay_barrier_nodf mds
+    fail mds
+
+    wait $pid || return 1
+    $CHECKSTAT -t file $DIR/$tfile && return 2
+    return 0
+}
+run_test 49c "mds fail after close reply is dropped: open"
+
 test_50() {
     local oscdev=`do_facet mds lctl get_param -n devices | grep ${ost1_svc}-osc | awk '{print $1}' | head -1`
     [ "$oscdev" ] || return 1
-- 
1.8.3.1