On Mon, Apr 22, 2024 at 5:13 AM Jiri Olsa <jolsa@xxxxxxxxxx> wrote:
>
> Adding support for cookie within the session of kprobe multi
> entry and return program.
>
> The session cookie is u64 value and can be retrieved be new
> kfunc bpf_session_cookie, which returns pointer to the cookie
> value. The bpf program can use the pointer to store (on entry)
> and load (on return) the value.
>
> The cookie value is implemented via fprobe feature that allows
> to share values between entry and return ftrace fprobe callbacks.
>
> Signed-off-by: Jiri Olsa <jolsa@xxxxxxxxxx>
> ---
> kernel/bpf/verifier.c | 7 +++++++
> kernel/trace/bpf_trace.c | 19 ++++++++++++++++---
> 2 files changed, 23 insertions(+), 3 deletions(-)
>
Had the same question as Alexei, but this read-write semantics quirk
makes sense. But it's probably a bit more reliable and cleaner to
handle it by special casing this kfunc a bit earlier (see
KF_bpf_rbtree_add_impl) and setting r0_size = 8, r0_rdonly = false.
And then let generic PTR -> INT logic kick in. You'll be futzing with
register state much less.
Other than that looks good:
Acked-by: Andrii Nakryiko <andrii@xxxxxxxxxx>
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index 68cfd6fc6ad4..baaca451aebc 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -10987,6 +10987,7 @@ enum special_kfunc_type {
> KF_bpf_percpu_obj_drop_impl,
> KF_bpf_throw,
> KF_bpf_iter_css_task_new,
> + KF_bpf_session_cookie,
> };
[...]
