net: cls_bpf: add support for marking filters as hardware-only
authorJakub Kicinski <jakub.kicinski@netronome.com>
Wed, 21 Sep 2016 10:43:55 +0000 (11:43 +0100)
committerDavid S. Miller <davem@davemloft.net>
Wed, 21 Sep 2016 23:50:02 +0000 (19:50 -0400)
Add cls_bpf support for the TCA_CLS_FLAGS_SKIP_SW flag.

Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: David S. Miller <davem@davemloft.net>
net/sched/cls_bpf.c

index ebf01f7c1470f8ac27fd15f6f7a81a42b63e24e8..1becc2fe1bc5e33470b1dd1cb020e84c142fbbc8 100644 (file)
@@ -28,7 +28,7 @@ MODULE_DESCRIPTION("TC BPF based classifier");
 
 #define CLS_BPF_NAME_LEN       256
 #define CLS_BPF_SUPPORTED_GEN_FLAGS            \
-       TCA_CLS_FLAGS_SKIP_HW
+       (TCA_CLS_FLAGS_SKIP_HW | TCA_CLS_FLAGS_SKIP_SW)
 
 struct cls_bpf_head {
        struct list_head plist;
@@ -96,7 +96,9 @@ static int cls_bpf_classify(struct sk_buff *skb, const struct tcf_proto *tp,
 
                qdisc_skb_cb(skb)->tc_classid = prog->res.classid;
 
-               if (at_ingress) {
+               if (tc_skip_sw(prog->gen_flags)) {
+                       filter_res = prog->exts_integrated ? TC_ACT_UNSPEC : 0;
+               } else if (at_ingress) {
                        /* It is safe to push/pull even if skb_shared() */
                        __skb_push(skb, skb->mac_len);
                        bpf_compute_data_end(skb);
@@ -164,32 +166,42 @@ static int cls_bpf_offload_cmd(struct tcf_proto *tp, struct cls_bpf_prog *prog,
                                             tp->protocol, &offload);
 }
 
-static void cls_bpf_offload(struct tcf_proto *tp, struct cls_bpf_prog *prog,
-                           struct cls_bpf_prog *oldprog)
+static int cls_bpf_offload(struct tcf_proto *tp, struct cls_bpf_prog *prog,
+                          struct cls_bpf_prog *oldprog)
 {
        struct net_device *dev = tp->q->dev_queue->dev;
        struct cls_bpf_prog *obj = prog;
        enum tc_clsbpf_command cmd;
+       bool skip_sw;
+       int ret;
+
+       skip_sw = tc_skip_sw(prog->gen_flags) ||
+               (oldprog && tc_skip_sw(oldprog->gen_flags));
 
        if (oldprog && oldprog->offloaded) {
                if (tc_should_offload(dev, tp, prog->gen_flags)) {
                        cmd = TC_CLSBPF_REPLACE;
-               } else {
+               } else if (!tc_skip_sw(prog->gen_flags)) {
                        obj = oldprog;
                        cmd = TC_CLSBPF_DESTROY;
+               } else {
+                       return -EINVAL;
                }
        } else {
                if (!tc_should_offload(dev, tp, prog->gen_flags))
-                       return;
+                       return skip_sw ? -EINVAL : 0;
                cmd = TC_CLSBPF_ADD;
        }
 
-       if (cls_bpf_offload_cmd(tp, obj, cmd))
-               return;
+       ret = cls_bpf_offload_cmd(tp, obj, cmd);
+       if (ret)
+               return skip_sw ? ret : 0;
 
        obj->offloaded = true;
        if (oldprog)
                oldprog->offloaded = false;
+
+       return 0;
 }
 
 static void cls_bpf_stop_offload(struct tcf_proto *tp,
@@ -498,7 +510,11 @@ static int cls_bpf_change(struct net *net, struct sk_buff *in_skb,
        if (ret < 0)
                goto errout;
 
-       cls_bpf_offload(tp, prog, oldprog);
+       ret = cls_bpf_offload(tp, prog, oldprog);
+       if (ret) {
+               cls_bpf_delete_prog(tp, prog);
+               return ret;
+       }
 
        if (oldprog) {
                list_replace_rcu(&oldprog->link, &prog->link);