Patch address a racing migrate request message and an exit domain message.
Instead of blocking exit domains for the duration of the migrate, we ignore
failure to deliver that message. This is because an exiting domain should
not have any active locks and thus has no role to play in the migration.
Signed-off-by: Sunil Mushran <sunil.mushran@oracle.com>
Signed-off-by: Mark Fasheh <mfasheh@suse.com>
struct dlm_node_iter *iter)
{
struct dlm_migrate_request migrate;
struct dlm_node_iter *iter)
{
struct dlm_migrate_request migrate;
+ int ret, skip, status = 0;
int nodenum;
memset(&migrate, 0, sizeof(migrate));
int nodenum;
memset(&migrate, 0, sizeof(migrate));
nodenum == new_master)
continue;
nodenum == new_master)
continue;
+ /* We could race exit domain. If exited, skip. */
+ spin_lock(&dlm->spinlock);
+ skip = (!test_bit(nodenum, dlm->domain_map));
+ spin_unlock(&dlm->spinlock);
+ if (skip) {
+ clear_bit(nodenum, iter->node_map);
+ continue;
+ }
+
ret = o2net_send_message(DLM_MIGRATE_REQUEST_MSG, dlm->key,
&migrate, sizeof(migrate), nodenum,
&status);
ret = o2net_send_message(DLM_MIGRATE_REQUEST_MSG, dlm->key,
&migrate, sizeof(migrate), nodenum,
&status);
- if (ret < 0)
- mlog_errno(ret);
- else if (status < 0) {
+ if (ret < 0) {
+ mlog(0, "migrate_request returned %d!\n", ret);
+ if (!dlm_is_host_down(ret)) {
+ mlog(ML_ERROR, "unhandled error=%d!\n", ret);
+ BUG();
+ }
+ clear_bit(nodenum, iter->node_map);
+ ret = 0;
+ } else if (status < 0) {
mlog(0, "migrate request (node %u) returned %d!\n",
nodenum, status);
ret = status;
mlog(0, "migrate request (node %u) returned %d!\n",
nodenum, status);
ret = status;