- Add some more verbose logging of the cases that get clients into recovery.

[fs/lustre-release.git] / lustre / ptlrpc / rpc.c

diff --git a/lustre/ptlrpc/rpc.c b/lustre/ptlrpc/rpc.c
index dad8744..ceefc33 100644 (file)
--- a/lustre/ptlrpc/rpc.c
+++ b/lustre/ptlrpc/rpc.c
@@ -64,23 +64,83 @@ int connmgr_cleanup(struct obd_device *dev)
         RETURN(0);
 }
 
-/* should this be in llite? */
-
-int connmgr_iocontrol(long cmd, struct lustre_handle *conn, int len, void *karg,
+int connmgr_iocontrol(long cmd, struct lustre_handle *hdl, int len, void *karg,
                       void *uarg)
 {
-        struct obd_device *obd = class_conn2obd(conn);
+        struct ptlrpc_connection *conn = NULL;
+        struct obd_device *obd = class_conn2obd(hdl);
         struct recovd_obd *recovd = &obd->u.recovd;
+        struct obd_ioctl_data *data = karg;
+        struct list_head *tmp;
+        int rc = 0;
 
         ENTRY;
-        if (cmd == OBD_IOC_RECOVD_NEWCONN) { 
-                spin_lock(&recovd->recovd_lock);
-                /* XXX shaver flag upcall answer */
-                wake_up(&recovd->recovd_waitq);
+
+        if (cmd != OBD_IOC_RECOVD_NEWCONN && cmd != OBD_IOC_RECOVD_FAILCONN)
+                RETURN(-EINVAL); /* XXX ENOSYS? */
+        
+        /* Find the connection that's been rebuilt or has failed. */
+        spin_lock(&recovd->recovd_lock);
+        list_for_each(tmp, &recovd->recovd_troubled_items) {
+                conn = list_entry(tmp, struct ptlrpc_connection,
+                                  c_recovd_data.rd_managed_chain);
+
+                LASSERT(conn->c_recovd_data.rd_recovd == recovd); /* sanity */
+
+                if (!strcmp(conn->c_remote_uuid, data->ioc_inlbuf1))
+                        break;
+                conn = NULL;
+        }
+
+        if (!conn) {
+                if (cmd == OBD_IOC_RECOVD_NEWCONN)
+                        GOTO(out, rc = -EINVAL);
+                /* XXX macroize/inline and share with loop above */
+                list_for_each(tmp, &recovd->recovd_managed_items) {
+                        conn = list_entry(tmp, struct ptlrpc_connection,
+                                          c_recovd_data.rd_managed_chain);
+                        
+                        LASSERT(conn->c_recovd_data.rd_recovd == recovd);
+                        
+                        if (!strcmp(conn->c_remote_uuid, data->ioc_inlbuf1))
+                                break;
+                        conn = NULL;
+                }
+                if (!conn)
+                        GOTO(out, rc = -EINVAL);
+        }
+
+        if (cmd == OBD_IOC_RECOVD_FAILCONN) {
                 spin_unlock(&recovd->recovd_lock);
-                EXIT;
+                recovd_conn_fail(conn);
+                spin_lock(&recovd->recovd_lock);
+
+                /* Jump straight to the "failed" phase of recovery. */
+                conn->c_recovd_data.rd_phase = RD_FAILED;
+                goto out;
         }
-        return 0;
+
+        /* else (NEWCONN) */
+        if (conn->c_recovd_data.rd_phase != RD_PREPARING)
+                GOTO(out, rc = -EALREADY);
+
+        spin_lock(&conn->c_lock);
+        if (data->ioc_inllen2) {
+                CERROR("conn %p UUID change %s -> %s\n",
+                       conn, conn->c_remote_uuid, data->ioc_inlbuf2);
+                strcpy(conn->c_remote_uuid, data->ioc_inlbuf2);
+        } else {
+                CERROR("conn %p UUID %s reconnected\n", conn,
+                       conn->c_remote_uuid);
+        }
+        ptlrpc_readdress_connection(conn, conn->c_remote_uuid);
+        spin_unlock(&conn->c_lock);
+        
+        conn->c_recovd_data.rd_phase = RD_PREPARED;
+        wake_up(&recovd->recovd_waitq);
+ out:
+        spin_unlock(&recovd->recovd_lock);
+        RETURN(rc);
 }
 
 
@@ -89,6 +149,8 @@ static struct obd_ops recovd_obd_ops = {
         o_setup:       connmgr_setup,
         o_cleanup:     connmgr_cleanup,
         o_iocontrol:   connmgr_iocontrol,
+        o_connect:     class_connect,
+        o_disconnect:  class_disconnect
 };
 
 static int __init ptlrpc_init(void)
@@ -165,6 +227,9 @@ EXPORT_SYMBOL(lustre_msg_size);
 EXPORT_SYMBOL(lustre_unpack_msg);
 EXPORT_SYMBOL(lustre_msg_buf);
 
+EXPORT_SYMBOL(ll_recover);
+
+
 MODULE_AUTHOR("Cluster File Systems, Inc <info@clusterfs.com>");
 MODULE_DESCRIPTION("Lustre Request Processor v1.0");
 MODULE_LICENSE("GPL");