[Drbd-dev] [PATCH] drbd: fix a bug of do_change_disk_state when attach

[Xu Rui]

PING
From: Rui Xu <rui.xu at easystack.cn>
Date: 2022-02-23 16:44:42
To:  philipp.reisner at linbit.com,drbd-dev at lists.linbit.com
Cc:  dongsheng.yang at easystack.cn,Rui Xu <rui.xu at easystack.cn>
Subject: [PATCH] drbd: fix a bug of do_change_disk_state when attach>There is a bug in do_change_disk_state when attach disk to
>drbd. Consider this scenario:
>
>Primary: node-1, Secondary: node-2, node-3
>
>(1) Network failure occurs on node-1 so that node-2 and
>node-3 will lose connection to node-1
>
>(2) Detach the disk of drbd on node-2 use 'drbdadm detach'
>
>(3) Attach the disk of drbd on node-2 use 'drbdadm attach'
>
>we can see that disk state of node-2 will stay in the negotiating
>and miss a handshake with node-3, what causes this is when we attach,
>disk state will going to negotiating from attaching, we will determine
>whether a two-pc is required in do_change_disk_state, since we lose the
>connection with node-1 and node-1 was first connection, connection->agreeed
>_pro_version will less than 110, at last, we will not do a two-pc, which
>leads a problem that node-3 will not send p_state to node-2, then node-2 miss
>a handshake with node-3 and the disk state of node-2 will not change.
>
>Fix it by using supports_two_phase_commit to determine whether a two-pc is
>really required.
>
>Signed-off-by: Rui Xu <rui.xu at easystack.cn>
>---
> drbd/drbd_state.c | 5 +----
> 1 file changed, 1 insertion(+), 4 deletions(-)
>
>diff --git a/drbd/drbd_state.c b/drbd/drbd_state.c
>index aeaf36a..2cb501d 100644
>--- a/drbd/drbd_state.c
>+++ b/drbd/drbd_state.c
>@@ -5173,10 +5173,7 @@ static bool do_change_disk_state(struct change_context *context, enum change_pha
> 	if (device->disk_state[NOW] == D_ATTACHING &&
> 	    context->val.disk == D_NEGOTIATING) {
> 		if (device_has_peer_devices_with_disk(device)) {
>-			struct drbd_connection *connection =
>-				first_connection(device->resource);
>-			cluster_wide_state_change =
>-				connection && connection->agreed_pro_version >= 110;
>+			cluster_wide_state_change = supports_two_phase_commit(device->resource);
> 		} else {
> 			/* very last part of attach */
> 			context->val.disk = disk_state_from_md(device);
>-- 
>1.8.3.1
>