Rules applied to OvS internal ports are not represented in TC datapaths. However, it is possible to support rules matching on internal ports in TC. The start_xmit ndo of OvS internal ports directs packets back into the OvS kernel datapath where they are rematched with the ingress port now being that of the internal port. Due to this, rules matching on an internal port can be added as TC filters to an egress qdisc for these ports.
Allow rules applied to internal ports to be offloaded to TC as egress filters. Rules redirecting to an internal port are also offloaded. These are supported by the redirect ingress functionality applied in an earlier patch. Signed-off-by: John Hurley <[email protected]> Reviewed-by: Simon Horman <[email protected]> --- lib/dpif.c | 13 +++++-------- lib/netdev-linux.c | 1 + lib/netdev-tc-offloads.c | 31 +++++++++++++++++++++---------- 3 files changed, 27 insertions(+), 18 deletions(-) diff --git a/lib/dpif.c b/lib/dpif.c index 457c9bf..063ba20 100644 --- a/lib/dpif.c +++ b/lib/dpif.c @@ -101,12 +101,9 @@ static bool should_log_flow_message(const struct vlog_module *module, struct seq *tnl_conf_seq; static bool -dpif_is_internal_port(const char *type) +dpif_is_tap_port(const char *type) { - /* For userspace datapath, tap devices are the equivalent - * of internal devices in the kernel datapath, so both - * these types are 'internal' devices. */ - return !strcmp(type, "internal") || !strcmp(type, "tap"); + return !strcmp(type, "tap"); } static void @@ -359,7 +356,7 @@ do_open(const char *name, const char *type, bool create, struct dpif **dpifp) struct netdev *netdev; int err; - if (dpif_is_internal_port(dpif_port.type)) { + if (dpif_is_tap_port(dpif_port.type)) { continue; } @@ -434,7 +431,7 @@ dpif_remove_netdev_ports(struct dpif *dpif) { struct dpif_port dpif_port; DPIF_PORT_FOR_EACH (&dpif_port, &port_dump, dpif) { - if (!dpif_is_internal_port(dpif_port.type)) { + if (!dpif_is_tap_port(dpif_port.type)) { netdev_ports_remove(dpif_port.port_no, dpif->dpif_class); } } @@ -582,7 +579,7 @@ dpif_port_add(struct dpif *dpif, struct netdev *netdev, odp_port_t *port_nop) VLOG_DBG_RL(&dpmsg_rl, "%s: added %s as port %"PRIu32, dpif_name(dpif), netdev_name, port_no); - if (!dpif_is_internal_port(netdev_get_type(netdev))) { + if (!dpif_is_tap_port(netdev_get_type(netdev))) { struct dpif_port dpif_port; diff --git a/lib/netdev-linux.c b/lib/netdev-linux.c index 3e9c450..95a4a0d 100644 --- a/lib/netdev-linux.c +++ b/lib/netdev-linux.c @@ -3339,6 +3339,7 @@ const struct netdev_class netdev_tap_class = { const struct netdev_class netdev_internal_class = { NETDEV_LINUX_CLASS_COMMON, + LINUX_FLOW_OFFLOAD_API, .type = "internal", .construct = netdev_linux_construct, .get_stats = netdev_internal_get_stats, diff --git a/lib/netdev-tc-offloads.c b/lib/netdev-tc-offloads.c index 11c597a..042a3f2 100644 --- a/lib/netdev-tc-offloads.c +++ b/lib/netdev-tc-offloads.c @@ -185,11 +185,12 @@ del_ufid_tc_mapping(const ovs_u128 *ufid) /* Wrapper function to delete filter and ufid tc mapping */ static int del_filter_and_ufid_mapping(int ifindex, int prio, int handle, - uint32_t block_id, const ovs_u128 *ufid) + uint32_t block_id, const ovs_u128 *ufid, + bool egress) { int err; - err = tc_del_filter(ifindex, prio, handle, block_id, false); + err = tc_del_filter(ifindex, prio, handle, block_id, egress); del_ufid_tc_mapping(ufid); return err; @@ -346,6 +347,7 @@ get_block_id_from_netdev(struct netdev *netdev) int netdev_tc_flow_flush(struct netdev *netdev) { + bool egress = is_internal_port(netdev_get_type(netdev)); int ifindex = netdev_get_ifindex(netdev); uint32_t block_id = 0; @@ -357,13 +359,14 @@ netdev_tc_flow_flush(struct netdev *netdev) block_id = get_block_id_from_netdev(netdev); - return tc_flush(ifindex, block_id, false); + return tc_flush(ifindex, block_id, egress); } int netdev_tc_flow_dump_create(struct netdev *netdev, struct netdev_flow_dump **dump_out) { + bool egress = is_internal_port(netdev_get_type(netdev)); struct netdev_flow_dump *dump; uint32_t block_id = 0; int ifindex; @@ -379,7 +382,7 @@ netdev_tc_flow_dump_create(struct netdev *netdev, dump = xzalloc(sizeof *dump); dump->nl_dump = xzalloc(sizeof *dump->nl_dump); dump->netdev = netdev_ref(netdev); - tc_dump_flower_start(ifindex, dump->nl_dump, block_id, false); + tc_dump_flower_start(ifindex, dump->nl_dump, block_id, egress); *dump_out = dump; @@ -1080,6 +1083,7 @@ netdev_tc_flow_put(struct netdev *netdev, struct match *match, struct dpif_flow_stats *stats OVS_UNUSED) { static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 20); + bool egress = is_internal_port(netdev_get_type(netdev)); struct tc_flower flower; const struct flow *key = &match->flow; struct flow *mask = &match->wc.masks; @@ -1342,7 +1346,8 @@ netdev_tc_flow_put(struct netdev *netdev, struct match *match, handle = get_ufid_tc_mapping(ufid, &prio, NULL); if (handle && prio) { VLOG_DBG_RL(&rl, "updating old handle: %d prio: %d", handle, prio); - del_filter_and_ufid_mapping(ifindex, prio, handle, block_id, ufid); + del_filter_and_ufid_mapping(ifindex, prio, handle, block_id, ufid, + egress); } if (!prio) { @@ -1356,7 +1361,7 @@ netdev_tc_flow_put(struct netdev *netdev, struct match *match, flower.act_cookie.data = ufid; flower.act_cookie.len = sizeof *ufid; - err = tc_replace_flower(ifindex, prio, handle, &flower, block_id, false); + err = tc_replace_flower(ifindex, prio, handle, &flower, block_id, egress); if (!err) { add_ufid_tc_mapping(ufid, flower.prio, flower.handle, netdev, ifindex); } @@ -1380,6 +1385,7 @@ netdev_tc_flow_get(struct netdev *netdev OVS_UNUSED, odp_port_t in_port; int prio = 0; int ifindex; + bool egress; int handle; int err; @@ -1388,6 +1394,7 @@ netdev_tc_flow_get(struct netdev *netdev OVS_UNUSED, return ENOENT; } + egress = is_internal_port(netdev_get_type(netdev)); ifindex = netdev_get_ifindex(dev); if (ifindex < 0) { VLOG_ERR_RL(&error_rl, "flow_get: failed to get ifindex for %s: %s", @@ -1399,7 +1406,7 @@ netdev_tc_flow_get(struct netdev *netdev OVS_UNUSED, block_id = get_block_id_from_netdev(dev); VLOG_DBG_RL(&rl, "flow get (dev %s prio %d handle %d block_id %d)", netdev_get_name(dev), prio, handle, block_id); - err = tc_get_flower(ifindex, prio, handle, &flower, block_id, false); + err = tc_get_flower(ifindex, prio, handle, &flower, block_id, egress); netdev_close(dev); if (err) { VLOG_ERR_RL(&error_rl, "flow get failed (dev %s prio %d handle %d): %s", @@ -1426,6 +1433,7 @@ netdev_tc_flow_del(struct netdev *netdev OVS_UNUSED, struct netdev *dev; int prio = 0; int ifindex; + bool egress; int handle; int error; @@ -1434,6 +1442,7 @@ netdev_tc_flow_del(struct netdev *netdev OVS_UNUSED, return ENOENT; } + egress = is_internal_port(netdev_get_type(netdev)); ifindex = netdev_get_ifindex(dev); if (ifindex < 0) { VLOG_ERR_RL(&error_rl, "flow_del: failed to get ifindex for %s: %s", @@ -1446,14 +1455,15 @@ netdev_tc_flow_del(struct netdev *netdev OVS_UNUSED, if (stats) { memset(stats, 0, sizeof *stats); - if (!tc_get_flower(ifindex, prio, handle, &flower, block_id, false)) { + if (!tc_get_flower(ifindex, prio, handle, &flower, block_id, egress)) { stats->n_packets = get_32aligned_u64(&flower.stats.n_packets); stats->n_bytes = get_32aligned_u64(&flower.stats.n_bytes); stats->used = flower.lastused; } } - error = del_filter_and_ufid_mapping(ifindex, prio, handle, block_id, ufid); + error = del_filter_and_ufid_mapping(ifindex, prio, handle, block_id, ufid, + egress); netdev_close(dev); @@ -1525,6 +1535,7 @@ netdev_tc_init_flow_api(struct netdev *netdev) { static struct ovsthread_once multi_mask_once = OVSTHREAD_ONCE_INITIALIZER; static struct ovsthread_once block_once = OVSTHREAD_ONCE_INITIALIZER; + bool egress = is_internal_port(netdev_get_type(netdev)); uint32_t block_id = 0; int ifindex; int error; @@ -1550,7 +1561,7 @@ netdev_tc_init_flow_api(struct netdev *netdev) } block_id = get_block_id_from_netdev(netdev); - error = tc_add_del_qdisc(ifindex, true, block_id, false); + error = tc_add_del_qdisc(ifindex, true, block_id, egress); if (error && error != EEXIST) { VLOG_ERR("failed adding ingress qdisc required for offloading: %s", -- 2.7.4 _______________________________________________ dev mailing list [email protected] https://mail.openvswitch.org/mailman/listinfo/ovs-dev
