On Thu, Aug 25, 2022 at 1:18 PM Andrii Nakryiko
[off-list ref] wrote:

On Tue, Aug 23, 2022 at 8:01 PM Hao Luo [off-list ref] wrote:

quoted

[...]

quoted

 typedef int (*bpf_iter_attach_target_t)(struct bpf_prog *prog,

diff --git a/include/uapi/linux/bpf.h b/include/uapi/linux/bpf.h
index 934a2a8beb87..1c4e1c583880 100644
--- a/include/uapi/linux/bpf.h
+++ b/include/uapi/linux/bpf.h

@@ -87,10 +87,29 @@ struct bpf_cgroup_storage_key {
        __u32   attach_type;            /* program attach type (enum bpf_attach_type) */
 };

+enum bpf_cgroup_iter_order {
+       BPF_ITER_ORDER_UNSPEC = 0,
+       BPF_ITER_SELF_ONLY,             /* process only a single object. */
+       BPF_ITER_DESCENDANTS_PRE,       /* walk descendants in pre-order. */
+       BPF_ITER_DESCENDANTS_POST,      /* walk descendants in post-order. */
+       BPF_ITER_ANCESTORS_UP,          /* walk ancestors upward. */
+};

just skimming through this, I noticed that we have "enum
bpf_cgroup_iter_order" (good, I like) but BPF_ITER_xxx with no CGROUP
part in it (not good, don't like :). All the enumerator names have
global visibility, so it would probably be best for them to be
CGROUP-specific and roughly match the enum name itself:
BPF_CGROUP_ITER_SELF_ONLY, etc?

Ack. I will send a patch to fix this right now.