Re: [PATCH bpf-next v4] libbpf: Expose API to consume one ring at a time

[Adam Sindelar <adam@wowsignal.io>]

On Thu, Jul 27, 2023 at 12:00:10PM -0700, John Fastabend wrote:
> Adam Sindelar wrote:
> > We already provide ring_buffer__epoll_fd to enable use of external
> > polling systems. However, the only API available to consume the ring
> > buffer is ring_buffer__consume, which always checks all rings. When
> > polling for many events, this can be wasteful.
> > 
> > Signed-off-by: Adam Sindelar <adam@wowsignal.io>
> > ---
> > v1->v2: Added entry to libbpf.map
> > v2->v3: Correctly set errno and handle overflow
> > v3->v4: Fixed an embarrasing typo from zealous autocomplete
> > 
> >  tools/lib/bpf/libbpf.h   |  1 +
> >  tools/lib/bpf/libbpf.map |  1 +
> >  tools/lib/bpf/ringbuf.c  | 22 ++++++++++++++++++++++
> >  3 files changed, 24 insertions(+)
> > 
> > diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
> > index 55b97b2087540..20ccc65eb3f9d 100644
> > --- a/tools/lib/bpf/libbpf.h
> > +++ b/tools/lib/bpf/libbpf.h
> > @@ -1195,6 +1195,7 @@ LIBBPF_API int ring_buffer__add(struct ring_buffer *rb, int map_fd,
> >  				ring_buffer_sample_fn sample_cb, void *ctx);
> >  LIBBPF_API int ring_buffer__poll(struct ring_buffer *rb, int timeout_ms);
> >  LIBBPF_API int ring_buffer__consume(struct ring_buffer *rb);
> > +LIBBPF_API int ring_buffer__consume_ring(struct ring_buffer *rb, uint32_t ring_id);
> >  LIBBPF_API int ring_buffer__epoll_fd(const struct ring_buffer *rb);
> >  
> >  struct user_ring_buffer_opts {
> > diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
> > index 9c7538dd5835e..42dc418b4672f 100644
> > --- a/tools/lib/bpf/libbpf.map
> > +++ b/tools/lib/bpf/libbpf.map
> > @@ -398,4 +398,5 @@ LIBBPF_1.3.0 {
> >  		bpf_prog_detach_opts;
> >  		bpf_program__attach_netfilter;
> >  		bpf_program__attach_tcx;
> > +		ring_buffer__consume_ring;
> >  } LIBBPF_1.2.0;
> > diff --git a/tools/lib/bpf/ringbuf.c b/tools/lib/bpf/ringbuf.c
> > index 02199364db136..457469fc7d71e 100644
> > --- a/tools/lib/bpf/ringbuf.c
> > +++ b/tools/lib/bpf/ringbuf.c
> > @@ -290,6 +290,28 @@ int ring_buffer__consume(struct ring_buffer *rb)
> >  	return res;
> >  }
> >  
> > +/* Consume available data from a single RINGBUF map identified by its ID.
> > + * The ring ID is returned in epoll_data by epoll_wait when called with
> > + * ring_buffer__epoll_fd.
> > + */
> > +int ring_buffer__consume_ring(struct ring_buffer *rb, uint32_t ring_id)
> > +{
> > +	struct ring *ring;
> > +	int64_t res;
> > +
> > +	if (ring_id >= rb->ring_cnt)
> > +		return libbpf_err(-EINVAL);
> > +
> > +	ring = &rb->rings[ring_id];
> > +	res = ringbuf_process_ring(ring);
> > +	if (res < 0)
> > +		return libbpf_err(res);
> > +
> > +	if (res > INT_MAX)
> > +		return INT_MAX;
> > +	return res;
> 
> Why not just return int64_t here? Then skip the INT_MAX check? I would
> just assume get the actual value if I was calling this.
> 

Mainly for consistency with the existing API. So far, the comparable
LIBBPF_API functions use int. It's hard to imagine that the number of
records would exceed ~2 billion in a single call - I think the
abberation is that ringbuf_process_ring using a 64-bit counter. If you
do exceed INT_MAX records, something is probably wrong and maybe the function
should return error instead. (But that would be outside the scope of
this patch.)

> > +}
> > +
> >  /* Poll for available data and consume records, if any are available.
> >   * Returns number of records consumed (or INT_MAX, whichever is less), or
> >   * negative number, if any of the registered callbacks returned error.
> > -- 
> > 2.39.2
> > 
> > 
> 
>