]> pilppa.com Git - linux-2.6-omap-h63xx.git/commitdiff
ocfs2: handle network errors during recovery
authorKurt Hackel <kurt.hackel@oracle.com>
Fri, 28 Apr 2006 01:06:58 +0000 (18:06 -0700)
committerMark Fasheh <mark.fasheh@oracle.com>
Mon, 26 Jun 2006 21:42:49 +0000 (14:42 -0700)
Signed-off-by: Kurt Hackel <kurt.hackel@oracle.com>
Signed-off-by: Mark Fasheh <mark.fasheh@oracle.com>
fs/ocfs2/dlm/dlmrecovery.c

index 39488763728936e706dfb952f07e3f6b5963c0e5..59c8976915a9309467c26b32320628301fae0a18 100644 (file)
@@ -757,6 +757,7 @@ static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data)
        struct list_head *iter;
        int ret;
        u8 dead_node, reco_master;
+       int skip_all_done = 0;
 
        dlm = item->dlm;
        dead_node = item->u.ral.dead_node;
@@ -793,12 +794,18 @@ static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data)
        dlm_move_reco_locks_to_list(dlm, &resources, dead_node);
 
        /* now we can begin blasting lockreses without the dlm lock */
+
+       /* any errors returned will be due to the new_master dying,
+        * the dlm_reco_thread should detect this */
        list_for_each(iter, &resources) {
                res = list_entry (iter, struct dlm_lock_resource, recovering);
                ret = dlm_send_one_lockres(dlm, res, mres, reco_master,
                                        DLM_MRES_RECOVERY);
-               if (ret < 0)
+               if (ret < 0) {
                        mlog_errno(ret);
+                       skip_all_done = 1;
+                       break;
+               }
        }
 
        /* move the resources back to the list */
@@ -806,9 +813,12 @@ static void dlm_request_all_locks_worker(struct dlm_work_item *item, void *data)
        list_splice_init(&resources, &dlm->reco.resources);
        spin_unlock(&dlm->spinlock);
 
-       ret = dlm_send_all_done_msg(dlm, dead_node, reco_master);
-       if (ret < 0)
-               mlog_errno(ret);
+       if (!skip_all_done) {
+               ret = dlm_send_all_done_msg(dlm, dead_node, reco_master);
+               if (ret < 0) {
+                       mlog_errno(ret);
+               }
+       }
 
        free_page((unsigned long)data);
 }
@@ -828,8 +838,14 @@ static int dlm_send_all_done_msg(struct dlm_ctxt *dlm, u8 dead_node, u8 send_to)
 
        ret = o2net_send_message(DLM_RECO_DATA_DONE_MSG, dlm->key, &done_msg,
                                 sizeof(done_msg), send_to, &tmpret);
-       /* negative status is ignored by the caller */
-       if (ret >= 0)
+       if (ret < 0) {
+               if (!dlm_is_host_down(ret)) {
+                       mlog_errno(ret);
+                       mlog(ML_ERROR, "%s: unknown error sending data-done "
+                            "to %u\n", dlm->name, send_to);
+                       BUG();
+               }
+       } else
                ret = tmpret;
        return ret;
 }
@@ -1109,22 +1125,25 @@ int dlm_send_one_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res,
                         * we must send it immediately. */
                        ret = dlm_send_mig_lockres_msg(dlm, mres, send_to,
                                                       res, total_locks);
-                       if (ret < 0) {
-                               // TODO
-                               mlog(ML_ERROR, "dlm_send_mig_lockres_msg "
-                                    "returned %d, TODO\n", ret);
-                               BUG();
-                       }
+                       if (ret < 0)
+                               goto error;
                }
        }
        /* flush any remaining locks */
        ret = dlm_send_mig_lockres_msg(dlm, mres, send_to, res, total_locks);
-       if (ret < 0) {
-               // TODO
-               mlog(ML_ERROR, "dlm_send_mig_lockres_msg returned %d, "
-                    "TODO\n", ret);
+       if (ret < 0)
+               goto error;
+       return ret;
+
+error:
+       mlog(ML_ERROR, "%s: dlm_send_mig_lockres_msg returned %d\n",
+            dlm->name, ret);
+       if (!dlm_is_host_down(ret))
                BUG();
-       }
+       mlog(0, "%s: node %u went down while sending %s "
+            "lockres %.*s\n", dlm->name, send_to,
+            flags & DLM_MRES_RECOVERY ?  "recovery" : "migration",
+            res->lockname.len, res->lockname.name);
        return ret;
 }