[Drbd-dev] DRBD8: ASSERT( mdev->bc->md.uuid[Bitmap] == 0 ) in drbd_main.c:drbd_uuid_new_current()

[Drbd-dev] DRBD8: ASSERT( mdev->bc->md.uuid[Bitmap] == 0 ) in drbd_main.c:drbd_uuid_new_current()

[Montrose, Ernest]

[-- Attachment #1.1: Type: text/plain, Size: 625 bytes --]

Hi all,
We are seeing another problem that occurs with the following sequence of
events:
drbdadm secondary vm2.root.fs
drbdadm detach vm2.root.fs
drbdadm primary vm2.root.fs
drbdadm secondary vm2.root.fs
drbdadm primary vm2.root.fs <----- This causes the assert
drbdadm attach vm2.root.fs  <------ This causes an instant Split brain
but only because of the assert above.
 
I think what's happening is that we are generating a new bitmap UUID
every time the diskless peer goes Primary.
I see no reason why we have to do this more then once.  I include a tiny
patch to do it just one time. 
 
Thanks,
 
EM--

[-- Attachment #1.2: Type: text/html, Size: 7322 bytes --]

[-- Attachment #2: bitmapassert.patch --]
[-- Type: application/octet-stream, Size: 500 bytes --]

Index: drbd/drbd_main.c
===================================================================
--- drbd/drbd_main.c	(revision 16853)
+++ drbd/drbd_main.c	(working copy)
@@ -933,7 +933,8 @@
 
 	if( ns.pdsk < Inconsistent ) {
 		/* Diskless Peer becomes primary */
-		if (os.peer == Secondary && ns.peer == Primary ) {
+		if (os.peer == Secondary && ns.peer == Primary &&
+                  mdev->bc->md.uuid[Bitmap] == 0 ) {
 			drbd_uuid_new_current(mdev);
 		}
 		/* Diskless Peer becomes secondary */

Re: [Drbd-dev] DRBD8: ASSERT( mdev->bc->md.uuid[Bitmap] == 0 ) in drbd_main.c:drbd_uuid_new_current()

[Philipp Reisner]

On Tuesday 31 July 2007 23:18:05 Montrose, Ernest wrote:
> Hi all,
> We are seeing another problem that occurs with the following sequence of
> events:
> drbdadm secondary vm2.root.fs
> drbdadm detach vm2.root.fs
> drbdadm primary vm2.root.fs
> drbdadm secondary vm2.root.fs
> drbdadm primary vm2.root.fs <----- This causes the assert
> drbdadm attach vm2.root.fs  <------ This causes an instant Split brain
> but only because of the assert above.
>
> I think what's happening is that we are generating a new bitmap UUID
> every time the diskless peer goes Primary.
> I see no reason why we have to do this more then once.  I include a tiny
> patch to do it just one time.
>

You are absolutely right.

http://lists.linbit.com/pipermail/drbd-cvs/2007-August/001612.html

-Phil
-- 
: Dipl-Ing Philipp Reisner                      Tel +43-1-8178292-50 :
: LINBIT Information Technologies GmbH          Fax +43-1-8178292-82 :
: Vivenotgasse 48, 1120 Vienna, Austria        http://www.linbit.com :