sunvnet: Fix race between vnet_start_xmit() and vnet_ack()
authorSowmini Varadhan <sowmini.varadhan@oracle.com>
Sun, 9 Nov 2014 01:41:53 +0000 (20:41 -0500)
committerDavid S. Miller <davem@davemloft.net>
Tue, 11 Nov 2014 02:05:36 +0000 (21:05 -0500)
commitb0cffed54338e19e3cc46c9963478223eee0d560
tree1da52bafeb4401687b81c58d682b103d2289386a
parent6f6e741f6de5ed5eedf7d19c9c0f8c73eaa0da7d
sunvnet: Fix race between vnet_start_xmit() and vnet_ack()

When vnet_start_xmit() is concurrent with vnet_ack(), we may
have a race that looks like:

    thread 1                              thread 2
    vnet_start_xmit                       vnet_event_napi -> vnet_rx

__vnet_tx_trigger for some desc X
at this point dr->prod == X
                                        peer sends back a stopped ack for X
                                        we process X, but X == dr->prod
                                        so we bail out in vnet_ack with
                                        !idx_is_pending
update dr->prod

As a result of the fact that we never processed the stopped ack for X,
the Tx path is led to incorrectly believe that the peer is still
"started" and reading, but the peer has stopped reading, which will
ultimately end in flow-control assertions.

The fix is to synchronize the above 2 paths  on the netif_tx_lock.

Signed-off-by: Sowmini Varadhan <sowmini.varadhan@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/sun/sunvnet.c