On Mon, Nov 19, 2018 at 04:04:16PM +0100, Jiri Pirko wrote:

Mon, Nov 19, 2018 at 03:48:50PM CET, pablo@netfilter.org wrote:

quoted

On Mon, Nov 19, 2018 at 02:57:05PM +0100, Jiri Pirko wrote:

quoted

Mon, Nov 19, 2018 at 01:15:12AM CET, pablo@netfilter.org wrote:

[...]

quoted

quoted

diff --git a/include/net/pkt_cls.h b/include/net/pkt_cls.h
index 7d7aefa5fcd2..7f9a8d5ca945 100644
--- a/include/net/pkt_cls.h
+++ b/include/net/pkt_cls.h

@@ -758,6 +758,12 @@ enum tc_fl_command {

	TC_CLSFLOWER_TMPLT_DESTROY,
};

+struct tc_cls_flower_stats {
+	u64	pkts;
+	u64	bytes;
+	u64	lastused;
+};
+
struct tc_cls_flower_offload {
	struct tc_cls_common_offload common;
	enum tc_fl_command command;

@@ -765,6 +771,7 @@ struct tc_cls_flower_offload {

	struct flow_rule rule;
	struct tcf_exts *exts;
	u32 classid;
+	struct tc_cls_flower_stats stats;
};

static inline struct flow_rule *

@@ -773,6 +780,14 @@ tc_cls_flower_offload_flow_rule(struct tc_cls_flower_offload *tc_flow_cmd)

	return &tc_flow_cmd->rule;
}

+static inline void tc_cls_flower_stats_update(struct tc_cls_flower_offload *cls_flower,
+					      u64 pkts, u64 bytes, u64 lastused)
+{
+	cls_flower->stats.pkts		= pkts;
+	cls_flower->stats.bytes		= bytes;
+	cls_flower->stats.lastused	= lastused;

Why do you need to store the values here in struct tc_cls_flower_offload?
Why don't you just call tcf_exts_stats_update()? Basically,
tc_cls_flower_stats_update() should be just wrapper around
tcf_exts_stats_update() so that drivers wouldn't use ->exts directly, as
you will remove them in follow-up patches, no?

Patch 07/12 stops exposing tc action exts to drivers, so we need a
structure (struct tc_cls_flower_stats) to convey this statistics back
to the cls_flower frontend.

Hmm, shouldn't these stats be rather flow_rule related than flower
related?

I can rename tc_cls_flower_stats to struct flow_stats and place this
in include/net/flow.h, given flow_rule is unset / not present from the
TC_CLSFLOWER_STATS path.

Thanks for reviewing!