drbd: don't send out P_BARRIER with stale information
authorLars Ellenberg <lars.ellenberg@linbit.com>
Mon, 20 Aug 2012 09:05:23 +0000 (11:05 +0200)
committerPhilipp Reisner <philipp.reisner@linbit.com>
Fri, 9 Nov 2012 13:08:19 +0000 (14:08 +0100)
commit4eb9b3cba00471a01699cceb0f4b1f0cb8111ee2
tree55ee26423ac7d71efc76d8436622563ea272a549
parent08332d73250eec349b055843a503d45a9b5c13b6
drbd: don't send out P_BARRIER with stale information

We must only send P_BARRIER for epochs we actually sent P_DATA in.

If we (re-)establish a connection, we reinitialized the
send.current_epoch_nr, but forgot to reset send.current_epoch_writes.

This could result in a spurious P_BARRIER with stale epoch information,
and a disconnect/reconnect cycle once the then "unexpected"
P_BARRIER_ACK is received:
  BAD! BarrierAck #28823 received, expected #28829!

Introduce re_init_if_first_write() and maybe_send_barrier() helpers,
and call them appropriately for read/write/set-out-of-sync requests.

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