On Tue, Oct 8, 2019 at 2:49 PM Daniel Borkmann <daniel@xxxxxxxxxxxxx> wrote: > > On Mon, Oct 07, 2019 at 10:47:19AM -0700, Andrii Nakryiko wrote: > > On Mon, Oct 7, 2019 at 2:43 AM Daniel Borkmann <daniel@xxxxxxxxxxxxx> wrote: > > > On Sun, Oct 06, 2019 at 08:07:36PM -0700, Andrii Nakryiko wrote: > > > > Various small fixes to BPF helper documentation comments, enabling > > > > automatic header generation with a list of BPF helpers. > > > > > > > > Signed-off-by: Andrii Nakryiko <andriin@xxxxxx> > [...] > > > I'm wondering whether it would simply be much better to always just use 'void *ctx' > > > for everything that is BPF context as it may be just confusing to people why different > > > types are chosen sometimes leading to buggy drive-by attempts to 'fix' them back into > > > struct sk_buff * et al. > > > > I'm impartial on this issue. In some cases it might be helpful to > > specify what is the expected type of the context, if it's only ever > > one type, but there are lots of helpers that accept various contexts, > > so for consistency its better to just have "void *context". > > I would favor consistency here to always have "void *context". One > additional issue I could see happening otherwise on top of the 'fix' > attempts is that if existing helpers get enabled for multiple program > types and these have different BPF context, then it might be quite > easy to forget converting struct __sk_buff * and whatnot to void * in > the helper API doc, so the auto-generated BPF helpers will continue > to have only the old type. Ok, I can create a follow-up clean up patch changing all of them to void *. There is also a weird singular case of having three declarations of bpf_get_socket_cookie() with different contexts. I assume I should just combine them into a single declaration/description, right? > > Thanks, > Daniel
