fix forever blocking umount on error

b=11756
i=nathan
i=green

diff --git a/lustre/ChangeLog b/lustre/ChangeLog
index aa9a7e4..a816357 100644 (file)
--- a/lustre/ChangeLog
+++ b/lustre/ChangeLog
@@ -115,6 +115,13 @@ Details    : Assetion hit is result of rare race between disconnect and connet
             to same nid. target_handle_connect found old connect cockie and
             tried to reconnect, but can't find export for this cockie.
 
+Severity   : normal
+Bugzilla   : 11756
+Frequency  : rare
+Description: umount blocks forever on error
+Details    : In result of wrong using obd_no_recov and obd_force flags client can
+             hand if cancel or some other requests is lost.
+
 --------------------------------------------------------------------------------
 
 2007-07-30         Cluster File Systems, Inc. <info@clusterfs.com>

diff --git a/lustre/ldlm/ldlm_lib.c b/lustre/ldlm/ldlm_lib.c
index 44feeba..f6edb49 100644 (file)
--- a/lustre/ldlm/ldlm_lib.c
+++ b/lustre/ldlm/ldlm_lib.c
@@ -464,17 +464,17 @@ int client_disconnect_export(struct obd_export *exp)
         (void)ptlrpc_pinger_del_import(imp);
 
         if (obd->obd_namespace != NULL) {
-                /* obd_no_recov == local only */
+                /* obd_force == local only */
                 ldlm_cli_cancel_unused(obd->obd_namespace, NULL,
-                                       obd->obd_no_recov ? LDLM_FL_LOCAL_ONLY:0,
+                                       obd->obd_force ? LDLM_FL_LOCAL_ONLY:0,
                                        NULL);
                 ldlm_namespace_free_prior(obd->obd_namespace);
                 to_be_freed = obd->obd_namespace;
                 obd->obd_namespace = NULL;
         }
 
-        /* Yeah, obd_no_recov also (mainly) means "forced shutdown". */
-        if (!obd->obd_no_recov)
+        /* Yeah, obd_force means "forced shutdown". */
+        if (!obd->obd_force)
                 rc = ptlrpc_disconnect_import(imp, 0);
 
         ptlrpc_invalidate_import(imp);

diff --git a/lustre/llite/file.c b/lustre/llite/file.c
index 19b4091..e55a184 100644 (file)
--- a/lustre/llite/file.c
+++ b/lustre/llite/file.c
@@ -68,7 +68,7 @@ static int ll_close_inode_openhandle(struct inode *inode,
          * canceling "open lock" and we do not call mdc_close() in this case, as
          * it will not be successful, as import is already deactivated.
          */
-        if (obd->obd_no_recov)
+        if (obd->obd_force)
                 GOTO(out, rc = 0);
 
         OBDO_ALLOC(oa);

diff --git a/lustre/llite/llite_lib.c b/lustre/llite/llite_lib.c
index ef072e5..47f4cdf 100644 (file)
--- a/lustre/llite/llite_lib.c
+++ b/lustre/llite/llite_lib.c
@@ -1016,7 +1016,7 @@ void ll_put_super(struct super_block *sb)
         if (sbi->ll_mdc_exp) {
                 obd = class_exp2obd(sbi->ll_mdc_exp);
                 if (obd) 
-                        force = obd->obd_no_recov;
+                        force = obd->obd_force;
         }
         
         /* We need to set force before the lov_disconnect in 
@@ -1824,7 +1824,7 @@ void ll_umount_begin(struct super_block *sb)
                 EXIT;
                 return;
         }
-        obd->obd_no_recov = 1;
+        obd->obd_force = 1;
         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_mdc_exp, sizeof ioc_data,
                       &ioc_data, NULL);
 
@@ -1836,7 +1836,7 @@ void ll_umount_begin(struct super_block *sb)
                 return;
         }
 
-        obd->obd_no_recov = 1;
+        obd->obd_force = 1;
         obd_iocontrol(IOC_OSC_SET_ACTIVE, sbi->ll_osc_exp, sizeof ioc_data,
                       &ioc_data, NULL);
 

diff --git a/lustre/ptlrpc/pinger.c b/lustre/ptlrpc/pinger.c
index d75d1e9..52d9574 100644 (file)
--- a/lustre/ptlrpc/pinger.c
+++ b/lustre/ptlrpc/pinger.c
@@ -269,6 +269,9 @@ int ptlrpc_pinger_add_import(struct obd_import *imp)
         mutex_down(&pinger_sem);
         CDEBUG(D_HA, "adding pingable import %s->%s\n",
                imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
+        /* if we add to pinger we want recovery on this import */
+        imp->imp_obd->obd_no_recov = 0;
+
         ptlrpc_update_next_ping(imp);
         /* XXX sort, blah blah */
         list_add_tail(&imp->imp_pinger_chain, &pinger_imports);
@@ -290,6 +293,8 @@ int ptlrpc_pinger_del_import(struct obd_import *imp)
         list_del_init(&imp->imp_pinger_chain);
         CDEBUG(D_HA, "removing pingable import %s->%s\n",
                imp->imp_obd->obd_uuid.uuid, obd2cli_tgt(imp->imp_obd));
+        /* if we remove from pinger we don't want recovery on this import */
+        imp->imp_obd->obd_no_recov = 1;
         class_import_put(imp);
         mutex_up(&pinger_sem);
         RETURN(0);

diff --git a/lustre/tests/conf-sanity.sh b/lustre/tests/conf-sanity.sh
index f207d81..b40c7c0 100644 (file)
--- a/lustre/tests/conf-sanity.sh
+++ b/lustre/tests/conf-sanity.sh
@@ -87,6 +87,16 @@ stop_ost2() {
        stop ost2 -f  || return 93
 }
 
+start_client() {
+       echo "start client on `facet_active_host client`"
+       start client || return 99 
+}
+
+stop_client() {
+       echo "stop client on `facet_active_host client`"
+       stop client || return 100 
+}
+
 mount_client() {
        local MOUNTPATH=$1
        echo "mount $FSNAME on ${MOUNTPATH}....."
@@ -109,8 +119,12 @@ umount_client() {
 }
 
 manual_umount_client(){
+       local rc
+       local FORCE=$1
        echo "manual umount lustre on ${MOUNT}...."
-       do_facet client "umount -d $MOUNT"
+       do_facet client "umount -d ${FORCE} $MOUNT"
+       rc=$?
+       return $rc
 }
 
 setup() {
@@ -1079,6 +1093,7 @@ test_32a() {
 
         # mount a second time to make sure we didnt leave upgrade flag on
         $TUNEFS --dryrun $TMP/$tdir/mds || error "tunefs failed"
+       load_modules
         start mds $TMP/$tdir/mds "-o loop,exclude=lustre-OST0000" || return 12
         cleanup_nocli
 
@@ -1153,5 +1168,66 @@ run_test 33 "Mount ost with a large index number"
 umount_client $MOUNT   
 cleanup_nocli
 
+test_33() {
+        setup
+
+        do_facet client dd if=/dev/zero of=$MOUNT/24 bs=1024k count=1
+        # Drop lock cancelation reply during umount
+       #define OBD_FAIL_LDLM_CANCEL             0x304
+        do_facet client sysctl -w lustre.fail_loc=0x80000304
+        #sysctl -w lnet.debug=-1
+        umount_client $MOUNT
+        cleanup
+}
+run_test 33 "Drop cancel during umount"
+
+test_34a() {
+        setup
+       do_facet client multiop $DIR/file O_c &
+
+       manual_umount_client
+       rc=$?
+       do_facet client killall -USR1 multiop
+       if [ $rc -eq 0 ]; then
+               error "umount not fail!"
+       fi
+       sleep 1
+        cleanup
+}
+run_test 34a "umount with opened file should be fail"
+
+
+test_34b() {
+       setup
+       touch $DIR/$tfile || return 1
+       stop_mds --force || return 2
+
+       manual_umount_client --force
+       rc=$?
+       if [ $rc -ne 0 ]; then
+               error "mtab after failed umount - rc $rc"
+       fi
+
+       cleanup
+       return 0        
+}
+run_test 34b "force umount with failed mds should be normal"
+
+test_34c() {
+       setup
+       touch $DIR/$tfile || return 1
+       stop_ost --force || return 2
+
+       manual_umount_client --force
+       rc=$?
+       if [ $rc -ne 0 ]; then
+               error "mtab after failed umount - rc $rc"
+       fi
+
+       cleanup
+       return 0        
+}
+run_test 34c "force umount with failed mds should be normal"
+
 equals_msg "Done"
 echo "$0: completed"