Re: [PATCH bpf-next v2 01/13] bpf: Add dynptr type for skb metadata

On Fri, Jul 18, 2025 at 12:19 PM -07, Martin KaFai Lau wrote:

On 7/18/25 3:01 AM, Jakub Sitnicki wrote:

quoted

On Thu, Jul 17, 2025 at 05:06 PM -07, Martin KaFai Lau wrote:

quoted

On 7/16/25 9:16 AM, Jakub Sitnicki wrote:

quoted

+__bpf_kfunc int bpf_dynptr_from_skb_meta(struct __sk_buff *skb, u64 flags,
+					 struct bpf_dynptr *ptr__uninit)
+{
+	return dynptr_from_skb_meta(skb, flags, ptr__uninit, false);
+}
+
   __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_md *x, u64 flags,
   				    struct bpf_dynptr *ptr__uninit)
   {

@@ -12165,8 +12190,15 @@ int bpf_dynptr_from_skb_rdonly(struct __sk_buff *skb, u64 flags,
   	return 0;
   }
   +int bpf_dynptr_from_skb_meta_rdonly(struct __sk_buff *skb, u64 flags,
+				    struct bpf_dynptr *ptr__uninit)
+{
+	return dynptr_from_skb_meta(skb, flags, ptr__uninit, true);
+}
+
   BTF_KFUNCS_START(bpf_kfunc_check_set_skb)
   BTF_ID_FLAGS(func, bpf_dynptr_from_skb, KF_TRUSTED_ARGS)
+BTF_ID_FLAGS(func, bpf_dynptr_from_skb_meta, KF_TRUSTED_ARGS)

I looked at the high level of the set. I have a quick question.

Have you considered to create another bpf_kfunc_check_set_xxx that is only for
the tc and tracing prog type? No need to expose this kfunc to other prog types

After some more thoughts, lets target it for tc only. I think skb_meta is not
available in most of the tracepoints now. Lets wait until the skb_meta will be
supported in other hooks/layers first.

Makes sense. I was planning to drop patch 5 ("net: Clear skb metadata on
handover from device to protocol"), which means
skb_shinfo(skb)->meta_len could be invalid at some tracepoints.