Whamcloud - gitweb
LU-14397 ptlrpc: idle import vs lock enqueue race
[fs/lustre-release.git] / lustre / ptlrpc / import.c
index d2e45fa..969199b 100644 (file)
@@ -1695,7 +1695,6 @@ static struct ptlrpc_request *ptlrpc_disconnect_prep_req(struct obd_import *imp)
        req->rq_timeout = min_t(timeout_t, req->rq_timeout,
                                INITIAL_CONNECT_TIMEOUT);
 
-       import_set_state(imp, LUSTRE_IMP_CONNECTING);
        req->rq_send_state =  LUSTRE_IMP_CONNECTING;
        ptlrpc_request_set_replen(req);
 
@@ -1744,14 +1743,18 @@ int ptlrpc_disconnect_import(struct obd_import *imp, int noclose)
                        rc = -EINTR;
        }
 
+       req = ptlrpc_disconnect_prep_req(imp);
+       if (IS_ERR(req))
+               GOTO(set_state, rc = PTR_ERR(req));
+
        spin_lock(&imp->imp_lock);
-       if (imp->imp_state != LUSTRE_IMP_FULL)
+       if (imp->imp_state != LUSTRE_IMP_FULL) {
+               ptlrpc_req_finished_with_imp_lock(req);
                GOTO(out, rc);
+       }
+       import_set_state_nolock(imp, LUSTRE_IMP_CONNECTING);
        spin_unlock(&imp->imp_lock);
 
-       req = ptlrpc_disconnect_prep_req(imp);
-       if (IS_ERR(req))
-               GOTO(set_state, rc = PTR_ERR(req));
        rc = ptlrpc_queue_wait(req);
        ptlrpc_req_finished(req);
 
@@ -1832,6 +1835,21 @@ static int ptlrpc_disconnect_idle_interpret(const struct lu_env *env,
        return 0;
 }
 
+static bool ptlrpc_can_idle(struct obd_import *imp)
+{
+       struct ldlm_namespace *ns = imp->imp_obd->obd_namespace;
+
+       /* one request for disconnect rpc */
+       if (atomic_read(&imp->imp_reqs) > 1)
+               return false;
+
+       /* any lock increases ns_bref being a resource holder */
+       if (ns && atomic_read(&ns->ns_bref) > 0)
+               return false;
+
+       return true;
+}
+
 int ptlrpc_disconnect_and_idle_import(struct obd_import *imp)
 {
        struct ptlrpc_request *req;
@@ -1843,30 +1861,38 @@ int ptlrpc_disconnect_and_idle_import(struct obd_import *imp)
        if (ptlrpc_import_in_recovery(imp))
                RETURN(0);
 
+       req = ptlrpc_disconnect_prep_req(imp);
+       if (IS_ERR(req))
+               RETURN(PTR_ERR(req));
+
+       req->rq_interpret_reply = ptlrpc_disconnect_idle_interpret;
+
+       if (OBD_FAIL_PRECHECK(OBD_FAIL_PTLRPC_IDLE_RACE)) {
+               __u32 idx;
+
+               server_name2index(imp->imp_obd->obd_name, &idx, NULL);
+               if (idx == 0)
+                       OBD_RACE(OBD_FAIL_PTLRPC_IDLE_RACE);
+       }
+
        spin_lock(&imp->imp_lock);
-       if (imp->imp_state != LUSTRE_IMP_FULL) {
+       if (imp->imp_state != LUSTRE_IMP_FULL || !ptlrpc_can_idle(imp)) {
+               ptlrpc_req_finished_with_imp_lock(req);
                spin_unlock(&imp->imp_lock);
                RETURN(0);
        }
+       import_set_state_nolock(imp, LUSTRE_IMP_CONNECTING);
+       /* don't make noise at reconnection */
+       imp->imp_was_idle = 1;
        spin_unlock(&imp->imp_lock);
 
-       req = ptlrpc_disconnect_prep_req(imp);
-       if (IS_ERR(req))
-               RETURN(PTR_ERR(req));
-
        CDEBUG_LIMIT(imp->imp_idle_debug, "%s: disconnect after %llus idle\n",
                     imp->imp_obd->obd_name,
                     ktime_get_real_seconds() - imp->imp_last_reply_time);
 
-       /* don't make noise at reconnection */
-       spin_lock(&imp->imp_lock);
-       imp->imp_was_idle = 1;
-       spin_unlock(&imp->imp_lock);
-
-       req->rq_interpret_reply = ptlrpc_disconnect_idle_interpret;
        ptlrpcd_add_req(req);
 
-       RETURN(0);
+       RETURN(1);
 }
 EXPORT_SYMBOL(ptlrpc_disconnect_and_idle_import);