On 5/4/2018 7:49 AM, Evgenii Smirnov wrote:
Currently, the validity of a path is checked only on unicast ARP transmission. If Subnet Manager switchover happens and some LIDs get reassigned, driver in a network that uses only IPv6 addresses will not try to renew the path records, despite them being marked as invalid. In connected mode, remote side LID change will cause send to fail, freeing the corresponding neigh struct. Subsequent packets to this destination will trigger allocation of a new neigh struct. With this patch allocation of new neigh struct will also check the validity of the associated path and renew it if necessary. This, however, will not help in datagram mode, if the host continuously sends data to the destination with invalid path. The neigh struct alive timer will be updated, thus preventing it from reallocation. Test setup consists of two target hosts and two initiator hosts, one of the initiators is with the patch. All hosts have only IPv6 addresses from the same subnet and initiators constantly ping targets. In connected mode swapping the LIDs of target hosts and switching over SM leads to the loss of connectivity for the initiator without the patch. Initiator with the patch recovers in ~3 sec. In datagram mode initiator with the patch is able to recover only if ping is stopped for neigh_obsolete time. Signed-off-by: Evgenii Smirnov <evgenii.smirnov@xxxxxxxxxxxxxxxx>
Reviewed-by: Dennis Dalessandro <dennis.dalessandro@xxxxxxxxx> -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
