Change Details

`lagg_input_infiniband()` do not hold reference of lagg interface. It is possible that the lagg interface get destroyed and become invalid right after it is returned from `lagg_input_infiniband()` and may trigger an user-after-free. Widen NET_EPOCH coverage so that `infiniband_input()` will have consistent view of network stack. MFC after: 1 week

From static code analysis, some device drivers (cxgbe, mthca, mlx4 and qlnx) do not enter net epoch before `lagg_input_infiniband()`. If IPoIB interface is a member of lagg(4), after return from `lagg_input_infiniban()` the receiving interface of mbuf is set to lagg(4) interface. When concurrently destroying the interface, there's a small window that the interface get destroyed and become invalid before re-enter net epoch, thus leading use-after-free. MFC after: 1 week

`lagg_input_infiniband()` do not hold reference of lagg interface.From static code analysis, some device drivers (cxgbe, mthca, mlx4 and qlnx) do not enter net epoch before `lagg_input_infiniband()`. If IPoIB interface is a member of lagg(4), after return from `lagg_input_infiniban()` the receiving interface of mbuf is set to lagg(4) interface. When concurrently destroying the interface, It is possiblethere's a small window that the lagg interface get destroyed and become invalid right after it is returned from `lagg_input_infiniband()` and may trigger an user-after-free. Widen NET_EPOCH coverage so that `infiniband_input()` will have consistent view of network stackbefore re-enter net epoch, thus leading use-after-free. MFC after: 1 week