On Mon, Apr 26, 2021 at 10:14:45AM -0700, Andrii Nakryiko wrote:
> On Thu, Apr 22, 2021 at 5:27 PM Alexei Starovoitov
> <alexei.starovoitov@xxxxxxxxx> wrote:
> >
> > From: Alexei Starovoitov <ast@xxxxxxxxxx>
> >
> > Add support for FD_IDX make libbpf prefer that approach to loading programs.
> >
> > Signed-off-by: Alexei Starovoitov <ast@xxxxxxxxxx>
> > ---
> > tools/lib/bpf/bpf.c | 1 +
> > tools/lib/bpf/libbpf.c | 70 +++++++++++++++++++++++++++++----
> > tools/lib/bpf/libbpf_internal.h | 1 +
> > 3 files changed, 65 insertions(+), 7 deletions(-)
> >
>
> [...]
>
> > +static int probe_kern_fd_idx(void)
> > +{
> > + struct bpf_load_program_attr attr;
> > + struct bpf_insn insns[] = {
> > + BPF_LD_IMM64_RAW(BPF_REG_0, BPF_PSEUDO_MAP_IDX, 0),
> > + BPF_EXIT_INSN(),
> > + };
> > +
> > + memset(&attr, 0, sizeof(attr));
> > + attr.prog_type = BPF_PROG_TYPE_SOCKET_FILTER;
> > + attr.insns = insns;
> > + attr.insns_cnt = ARRAY_SIZE(insns);
> > + attr.license = "GPL";
> > +
> > + probe_fd(bpf_load_program_xattr(&attr, NULL, 0));
>
> probe_fd() calls close(fd) internally, which technically can interfere
> with errno, though close() shouldn't be called because syscall has to
> fail on correct kernels... So this should work, but I feel like
> open-coding that logic is better than ignoring probe_fd() result.
It will fail on all kernels.
That probe_fd was a left over of earlier detection approach where it would
proceed to load all the way, but then I switched to:
> > + return errno == EPROTO;
since such style of probing is much cheaper for the kernel and user space.
But point taken. Will open code it.
> > +}
> > +
>
> [...]
>
> > @@ -7239,6 +7279,8 @@ bpf_object__load_progs(struct bpf_object *obj, int log_level)
> > struct bpf_program *prog;
> > size_t i;
> > int err;
> > + struct bpf_map *map;
> > + int *fd_array = NULL;
> >
> > for (i = 0; i < obj->nr_programs; i++) {
> > prog = &obj->programs[i];
> > @@ -7247,6 +7289,16 @@ bpf_object__load_progs(struct bpf_object *obj, int log_level)
> > return err;
> > }
> >
> > + if (kernel_supports(FEAT_FD_IDX) && obj->nr_maps) {
> > + fd_array = malloc(sizeof(int) * obj->nr_maps);
> > + if (!fd_array)
> > + return -ENOMEM;
> > + for (i = 0; i < obj->nr_maps; i++) {
> > + map = &obj->maps[i];
> > + fd_array[i] = map->fd;
>
> nit: obj->maps[i].fd will keep it a single line
>
> > + }
> > + }
> > +
> > for (i = 0; i < obj->nr_programs; i++) {
> > prog = &obj->programs[i];
> > if (prog_is_subprog(obj, prog))
> > @@ -7256,10 +7308,14 @@ bpf_object__load_progs(struct bpf_object *obj, int log_level)
> > continue;
> > }
> > prog->log_level |= log_level;
> > + prog->fd_array = fd_array;
>
> you are not freeing this memory on success, as far as I can see.
hmm. there is free on success below.
> And
> given multiple programs are sharing fd_array, it's a bit problematic
> for prog to have fd_array. This is per-object properly, so let's add
> it at bpf_object level and clean it up on bpf_object__close()? And by
> assigning to obj->fd_array at malloc() site, you won't need to do all
> the error-handling free()s below.
hmm. that sounds worse.
why add another 8 byte to bpf_object that won't be used
until this last step of bpf_object__load_progs.
And only for the duration of this loading.
It's cheaper to have this alloc here with two free()s below.
>
> > err = bpf_program__load(prog, obj->license, obj->kern_version);
> > - if (err)
> > + if (err) {
> > + free(fd_array);
> > return err;
> > + }
> > }
> > + free(fd_array);
> > return 0;
> > }
> >
> > diff --git a/tools/lib/bpf/libbpf_internal.h b/tools/lib/bpf/libbpf_internal.h
> > index 6017902c687e..9114c7085f2a 100644
> > --- a/tools/lib/bpf/libbpf_internal.h
> > +++ b/tools/lib/bpf/libbpf_internal.h
> > @@ -204,6 +204,7 @@ struct bpf_prog_load_params {
> > __u32 log_level;
> > char *log_buf;
> > size_t log_buf_sz;
> > + int *fd_array;
> > };
> >
> > int libbpf__bpf_prog_load(const struct bpf_prog_load_params *load_attr);
> > --
> > 2.30.2
> >
--
