Re: [PATCH bpf-next v3 10/10] selftests/bpf: add file dynptr tests

[Eduard Zingerman <eddyz87@gmail.com>]

On Tue, 2025-10-21 at 14:55 +0100, Mykyta Yatsenko wrote:

[...]

> > > +struct {
> > > +	__uint(type, BPF_MAP_TYPE_RINGBUF);
> > > +	__uint(max_entries, 10000000);
> > > +} ringbuf SEC(".maps");
>
> > The test case lgtm, but a question: will it be possible to use an
> > array map instead of a ringbuf?  Just to avoid the need to allocate
> > and discard the pointer.
>
> How do I use array map here? Should I set a map value to be a buffer of
> needed length (256KB) or use 1 byte value and 256K elements in the map?
> Honestly, both options seem a little awkward to me, but I'm not sure 
> maybe it is an
> expected way to get a big buffer.

I think you can declare a top level char array of needed length and
libbpf will create an array map with a single element of that size.
E.g. see arrays `arr` and `small_arr` in progs/iters.c.
If, instead, you declare an array with 1-byte elements, verifier will
reject access past offset 0 for each lookup result.

> I like allocation/discard, as it guarantees that this temporary buffer 
> is local to the
> current func execution and we need to run non-trivial deinitialization 
> anyway.

If the goal is to memset the buffer to zero, I don't think that
ringbuf guarantees that newly allocated or discarded buffer are reset.

[...]