Re: [PATCH bpf-next v2 15/16] bpf: Realign skb metadata for TC progs using data_meta

On Mon, Jan 5, 2026 at 11:14 AM Alexei Starovoitov
[off-list ref] wrote:

On Mon, Jan 5, 2026 at 4:15 AM Jakub Sitnicki [off-list ref] wrote:

quoted

+__bpf_kfunc_start_defs();
+
+__bpf_kfunc void bpf_skb_meta_realign(struct __sk_buff *skb_)
+{
+       struct sk_buff *skb = (typeof(skb))skb_;
+       u8 *meta_end = skb_metadata_end(skb);
+       u8 meta_len = skb_metadata_len(skb);
+       u8 *meta;
+       int gap;
+
+       gap = skb_mac_header(skb) - meta_end;
+       if (!meta_len || !gap)
+               return;
+
+       if (WARN_ONCE(gap < 0, "skb metadata end past mac header")) {
+               skb_metadata_clear(skb);
+               return;
+       }
+
+       meta = meta_end - meta_len;
+       memmove(meta + gap, meta, meta_len);
+       skb_shinfo(skb)->meta_end += gap;
+
+       bpf_compute_data_pointers(skb);
+}
+
+__bpf_kfunc_end_defs();
+
+BTF_KFUNCS_START(tc_cls_act_hidden_ids)
+BTF_ID_FLAGS(func, bpf_skb_meta_realign)
+BTF_KFUNCS_END(tc_cls_act_hidden_ids)
+
+BTF_ID_LIST_SINGLE(bpf_skb_meta_realign_ids, func, bpf_skb_meta_realign)
+
 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, u32 pkt_access_flags,
                               const struct bpf_prog *prog)
 {
-       return bpf_unclone_prologue(insn_buf, pkt_access_flags, prog,
-                                   TC_ACT_SHOT);
+       struct bpf_insn *insn = insn_buf;
+       int cnt;
+
+       if (pkt_access_flags & PA_F_DATA_META_LOAD) {
+               /* Realign skb metadata for access through data_meta pointer.
+                *
+                * r6 = r1; // r6 will be "u64 *ctx"
+                * r0 = bpf_skb_meta_realign(r1); // r0 is undefined
+                * r1 = r6;
+                */
+               *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1);
+               *insn++ = BPF_CALL_KFUNC(0, bpf_skb_meta_realign_ids[0]);
+               *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6);
+       }

I see that we already did this hack with bpf_qdisc_init_prologue()
and bpf_qdisc_reset_destroy_epilogue().
Not sure why we went that route back then.

imo much cleaner to do BPF_EMIT_CALL() and wrap
BPF_CALL_1(bpf_skb_meta_realign, struct sk_buff *, skb)

BPF_CALL_x doesn't make it an uapi helper.
It's still a hidden kernel function,
while this kfunc stuff looks wrong, since kfunc isn't really hidden.

I suspect progs can call this bpf_skb_meta_realign() explicitly,
just like they can call bpf_qdisc_init_prologue() ?

qdisc prologue and epilogue qdisc kfuncs should be hidden from users.
The kfunc filter, bpf_qdisc_kfunc_filter(), determines what kfunc are
actually exposed.

BPF_CALL_x is simpler as there is no need for a kfunc filter to hide
it. However, IMO for qdisc they don't make too much difference since
bpf qdisc already needs the filter to limit .enqueue and .dequeue
specific kfunc.

Am I missing anything?

cc Amery, Martin.