if (!cli->cl_conn_count) {
CERROR("disconnecting disconnected device (%s)\n",
obd->obd_name);
- GOTO(out_sem, rc = -EINVAL);
+ GOTO(out_disconnect, rc = -EINVAL);
}
cli->cl_conn_count--;
if (cli->cl_conn_count)
- GOTO(out_no_disconnect, rc = 0);
+ GOTO(out_disconnect, rc = 0);
/* Mark import deactivated now, so we don't try to reconnect if any
* of the cleanup RPCs fails (e.g. ldlm cancel, etc). We don't
cli->cl_import = NULL;
EXIT;
- out_no_disconnect:
+
+ out_disconnect:
+ /* use server style - class_disconnect should be always called for
+ * o_disconnect */
err = class_disconnect(exp);
if (!rc && err)
rc = err;
- out_sem:
+
up_write(&cli->cl_sem);
if (to_be_freed)
ldlm_namespace_free_post(to_be_freed);
target->obd_name, cluuid.uuid,
libcfs_nid2str(req->rq_peer.nid),
export, atomic_read(&export->exp_rpc_count) - 1);
+ spin_lock(&export->exp_lock);
+ if (req->rq_export->exp_conn_cnt <
+ lustre_msg_get_conn_cnt(req->rq_reqmsg))
+ /* try to abort active requests */
+ req->rq_export->exp_abort_active_req = 1;
+ spin_unlock(&export->exp_lock);
GOTO(out, rc = -EBUSY);
} else if (lustre_msg_get_conn_cnt(req->rq_reqmsg) == 1) {
CERROR("%s: NID %s (%s) reconnected with 1 conn_cnt; "
GOTO(out, rc = -EALREADY);
}
export->exp_conn_cnt = lustre_msg_get_conn_cnt(req->rq_reqmsg);
+ export->exp_abort_active_req = 0;
/* request from liblustre? Don't evict it for not pinging. */
if (lustre_msg_get_op_flags(req->rq_reqmsg) & MSG_CONNECT_LIBCLIENT) {
revimp->imp_remote_handle = conn;
revimp->imp_dlm_fake = 1;
revimp->imp_state = LUSTRE_IMP_FULL;
- /* this is a bit of a layering violation, but much less risk than
- * changing this very complex and race-prone code. bug=16839 */
- if (data->ocd_connect_flags & OBD_CONNECT_MDS)
- obd_set_info_async(export, sizeof(KEY_MDS_CONN), KEY_MDS_CONN,
- 0, NULL, NULL);
/* unknown versions will be caught in
* ptlrpc_handle_server_req_in->lustre_unpack_msg() */
} else if (!list_empty(&obd->obd_req_replay_queue)) {
req = list_entry(obd->obd_req_replay_queue.next,
struct ptlrpc_request, rq_list);
- target_exp_dequeue_req_replay(req);
list_del_init(&req->rq_list);
obd->obd_requests_queued_for_recovery--;
} else {
if (!req_replay_done(req->rq_export) ||
!lock_replay_done(req->rq_export))
reset_recovery_timer(class_exp2obd(req->rq_export),
- AT_OFF ? obd_timeout :
+ AT_OFF ? obd_timeout :
at_get(&req->rq_rqbd->rqbd_service->srv_at_estimate), 1);
+
+ /**
+ * bz18031: increase next_recovery_transno before ptlrpc_free_clone()
+ * will drop exp_rpc reference
+ */
+ if (!req_replay_done(req->rq_export)) {
+ spin_lock_bh(&req->rq_export->exp_obd->obd_processing_task_lock);
+ req->rq_export->exp_obd->obd_next_recovery_transno++;
+ spin_unlock_bh(&req->rq_export->exp_obd->obd_processing_task_lock);
+ target_exp_dequeue_req_replay(req);
+ }
ptlrpc_free_clone(req);
RETURN(0);
}
handle_recovery_req(thread, req,
trd->trd_recovery_handler);
obd->obd_replayed_requests++;
- spin_lock_bh(&obd->obd_processing_task_lock);
- obd->obd_next_recovery_transno++;
- spin_unlock_bh(&obd->obd_processing_task_lock);
}
/* If some clients haven't replayed requests in time, evict them */
void target_recovery_init(struct lu_target *lut, svc_handler_t handler)
{
struct obd_device *obd = lut->lut_obd;
- if (obd->obd_max_recoverable_clients == 0)
+ if (obd->obd_max_recoverable_clients == 0) {
+ /** Update server last boot epoch */
+ lut_boot_epoch_update(lut);
return;
+ }
CWARN("RECOVERY: service %s, %d recoverable clients, "
"last_transno "LPU64"\n", obd->obd_name,
rs->rs_xid = req->rq_xid;
rs->rs_transno = req->rq_transno;
rs->rs_export = exp;
+ rs->rs_opc = lustre_msg_get_opc(rs->rs_msg);
spin_lock(&exp->exp_uncommitted_replies_lock);
CDEBUG(D_NET, "rs transno = "LPU64", last committed = "LPU64"\n",
LASSERT(req->rq_export);
- OBD_ALLOC(qdata, sizeof(struct qunit_data));
- if (!qdata)
- RETURN(-ENOMEM);
- rc = quota_get_qdata(req, qdata, QUOTA_REQUEST, QUOTA_EXPORT);
- if (rc < 0) {
+ qdata = quota_get_qdata(req, QUOTA_REQUEST, QUOTA_EXPORT);
+ if (IS_ERR(qdata)) {
+ rc = PTR_ERR(qdata);
CDEBUG(D_ERROR, "Can't unpack qunit_data(rc: %d)\n", rc);
+ req->rq_status = rc;
GOTO(out, rc);
}
if (!obd->obd_observer || !obd->obd_observer->obd_observer) {
CERROR("Can't find the observer, it is recovering\n");
req->rq_status = -EAGAIN;
- GOTO(send_reply, rc = -EAGAIN);
+ GOTO(out, rc);
}
master_obd = obd->obd_observer->obd_observer;
CDEBUG(D_QUOTA, "quota_type not processed yet, return "
"-EAGAIN\n");
req->rq_status = -EAGAIN;
- rc = ptlrpc_reply(req);
GOTO(out, rc);
}
CDEBUG(D_QUOTA, "quota_ctxt is not ready yet, return "
"-EAGAIN\n");
req->rq_status = -EAGAIN;
- rc = ptlrpc_reply(req);
GOTO(out, rc);
}
up_read(&obt->obt_rwsem);
if (rc && rc != -EDQUOT)
CDEBUG(rc == -EBUSY ? D_QUOTA : D_ERROR,
- "dqacq failed! (rc:%d)\n", rc);
+ "dqacq/dqrel failed! (rc:%d)\n", rc);
req->rq_status = rc;
- /* there are three forms of qunit(historic causes), so we need to
- * adjust the same form to different forms slaves needed */
rc = quota_copy_qdata(req, qdata, QUOTA_REPLY, QUOTA_EXPORT);
if (rc < 0) {
- CDEBUG(D_ERROR, "Can't pack qunit_data(rc: %d)\n", rc);
+ CERROR("Can't pack qunit_data(rc: %d)\n", rc);
GOTO(out, rc);
}
/* Block the quota req. b=14840 */
OBD_FAIL_TIMEOUT(OBD_FAIL_MDS_BLOCK_QUOTA_REQ, obd_timeout);
-send_reply:
- rc = ptlrpc_reply(req);
+ EXIT;
+
out:
- OBD_FREE(qdata, sizeof(struct qunit_data));
- RETURN(rc);
+ rc = ptlrpc_reply(req);
+ return rc;
#else
return 0;
#endif /* !__KERNEL__ */