On Mon, Apr 22, 2019 at 5:58 PM Daniel Borkmann <daniel@xxxxxxxxxxxxx> wrote:
>
> On 04/19/2019 03:18 AM, Andrii Nakryiko wrote:
> > On Tue, Apr 9, 2019 at 2:20 PM Daniel Borkmann <daniel@xxxxxxxxxxxxx> wrote:
> >>
> [...]
> >> + def->type = BPF_MAP_TYPE_ARRAY;
> >> + def->key_size = sizeof(int);
> >> + def->value_size = data->d_size;
> >> + def->max_entries = 1;
> >> + def->map_flags = type == LIBBPF_MAP_RODATA ?
> >> + BPF_F_RDONLY_PROG : 0;
> >
> > This is breaking BPF programs (even those that don't use global data,
> > as they still have .rodata section, though I haven't investigated its
> > contents) on kernels that don't yet support BPF_F_RDONLY_PROG flag
> > yet. We probably need to probe support for that flag first, before
> > using it. Just giving heads up, as I just discovered it trying to sync
> > libbpf on github.
>
> Thanks for reporting! On a quick look test_progs (modulo global data test)
> seems to pass with an slightly older kernel. I'll retry with a latest LLVM git
> tree tomorrow with our test suite. Did you see a specific one failing or do you
> have a reproducer in case it's something not covered where I could look into?
You need to add something like this to trigger .rodata section
generation (BPF code doesn't have to use that struct, it just needs to
be present):
const struct { int x, y; } bla = {};
This will cause libbpf to create a map for .rodata and specify
BPF_F_RDONLY_PROG flag, which on older kernels will be rejected.
