drbd: fix broken state change after split-brain attach while connected
authorLars Ellenberg <lars.ellenberg@linbit.com>
Fri, 26 Feb 2010 22:15:23 +0000 (23:15 +0100)
committerPhilipp Reisner <philipp.reisner@linbit.com>
Thu, 11 Mar 2010 15:00:09 +0000 (16:00 +0100)
Situation:
we have diverging data sets, i.e. we had a split brain somewhen,
but currently are connected, one node diskless.

Then we try to attach that disk, figure it is consistent,
but has a diverging data set, we refuse to attach.

This led to strange state changes:
22:18:35 bb drbd1: peer( Unknown -> Primary ) conn( WFReportParams -> Connected) pdsk( DUnknown -> UpToDate )
22:19:30 bb drbd1: disk( Diskless -> Attaching )
22:19:30 bb drbd1: disk( Attaching -> Negotiating )
22:19:30 bb drbd1: drbd_sync_handshake:
22:19:30 bb drbd1: self 97BF25798B9D5222:F33D1F62ADE698DD:4269796F9D027C83:AC45D8B5C3C1BF93 bits:19449 flags:0
22:19:30 bb drbd1: peer 280DFB6E125465D3:F33D1F62ADE698DC:4269796F9D027C82:AC45D8B5C3C1BF93 bits:2575806 flags:0
22:19:30 bb drbd1: uuid_compare()=100 by rule 90
22:19:30 bb drbd1: Split-Brain detected, dropping connection!
22:19:30 bb drbd1: disk( Negotiating -> Diskless )

while the other side says:
22:19:30 aa drbd1: Split-Brain detected, dropping connection!
22:19:30 aa drbd1: Disk attach process on the peer node was aborted.
22:19:30 aa drbd1: conn( Connected -> TOO_LARGE ) pdsk( Diskless -> Consistent )

This should be fixed now.

Signed-off-by: Philipp Reisner <philipp.reisner@linbit.com>
Signed-off-by: Lars Ellenberg <lars.ellenberg@linbit.com>
drivers/block/drbd/drbd_receiver.c

index 8bcde4a9632bee3b9b71b9c20af7960ae92d133f..41f36a9cd407b919dd0b84d6eb444827f759d5c8 100644 (file)
@@ -2513,6 +2513,10 @@ static enum drbd_conns drbd_sync_handshake(struct drbd_conf *mdev, enum drbd_rol
        }
 
        if (hg == -100) {
+               /* FIXME this log message is not correct if we end up here
+                * after an attempted attach on a diskless node.
+                * We just refuse to attach -- well, we drop the "connection"
+                * to that disk, in a way... */
                dev_alert(DEV, "Split-Brain detected, dropping connection!\n");
                drbd_khelper(mdev, "split-brain");
                return C_MASK;
@@ -3134,12 +3138,13 @@ static int receive_state(struct drbd_conf *mdev, struct p_header *h)
 
                put_ldev(mdev);
                if (nconn == C_MASK) {
+                       nconn = C_CONNECTED;
                        if (mdev->state.disk == D_NEGOTIATING) {
                                drbd_force_state(mdev, NS(disk, D_DISKLESS));
-                               nconn = C_CONNECTED;
                        } else if (peer_state.disk == D_NEGOTIATING) {
                                dev_err(DEV, "Disk attach process on the peer node was aborted.\n");
                                peer_state.disk = D_DISKLESS;
+                               real_peer_disk = D_DISKLESS;
                        } else {
                                if (test_and_clear_bit(CONN_DRY_RUN, &mdev->flags))
                                        return FALSE;