Re: [PATCH v17 11/12] selftests/bpf: Add test for bpf_verify_pkcs7_signature() kfunc

[Yonghong Song]

On 9/19/22 6:09 AM, Roberto Sassu wrote:
> On Mon, 2022-09-19 at 13:17 +0200, Roberto Sassu wrote:
> > On Thu, 2022-09-15 at 17:11 +0100, KP Singh wrote:
> > > On Fri, Sep 9, 2022 at 1:10 PM Roberto Sassu
> > > <roberto.sassu@xxxxxxxxxxxxxxx> wrote:
> > > > From: Roberto Sassu <roberto.sassu@xxxxxxxxxx>
> > >
> > > [...]
> > >
> > > > +}
> > > > diff --git
> > > > a/tools/testing/selftests/bpf/progs/test_verify_pkcs7_sig.c
> > > > b/tools/testing/selftests/bpf/progs/test_verify_pkcs7_sig.c
> > > > new file mode 100644
> > > > index 000000000000..4ceab545d99a
> > > > --- /dev/null
> > > > +++ b/tools/testing/selftests/bpf/progs/test_verify_pkcs7_sig.c
> > > > @@ -0,0 +1,100 @@
> > > > +// SPDX-License-Identifier: GPL-2.0
> > > > +
> > > > +/*
> > > > + * Copyright (C) 2022 Huawei Technologies Duesseldorf GmbH
> > > > + *
> > > > + * Author: Roberto Sassu <roberto.sassu@xxxxxxxxxx>
> > > > + */
> > > > +
> > > > +#include "vmlinux.h"
> > > > +#include <errno.h>
> > > > +#include <bpf/bpf_helpers.h>
> > > > +#include <bpf/bpf_tracing.h>
> > > > +
> > > > +#define MAX_DATA_SIZE (1024 * 1024)
> > > > +#define MAX_SIG_SIZE 1024
> > > > +
> > > > +typedef __u8 u8;
> > > > +typedef __u16 u16;
> > > > +typedef __u32 u32;
> > > > +typedef __u64 u64;
> > >
> > > I think you can avoid this and just use u32 and u64 directly.
> >
> > Thanks, yes.
> >
> > > +
> > > > +struct bpf_dynptr {
> > > > +       __u64 :64;
> > > > +       __u64 :64;
> > > > +} __attribute__((aligned(8)));
> > > > +
> > >
> > > I think you are doing this because including the uapi headers
> > > causes
> > > type conflicts.
> > > This does happen quite often. What do other folks think about doing
> > > something like
> > >
> > > #define DYNPTR(x) ((void *)x)
> > >
> > > It seems like this will be an issue anytime we use the helpers with
> > > vmlinux.h and users
> > > will always have to define this type in their tests.
> >
> > It seems it is sufficient to use struct bpf_dynptr somehow in the
> > kernel code. That causes the definition to be exported with BTF. Not
> > sure what would be the proper place to do that. When I tried, I
> > declared a unused variable.
>
> Easier:
>
> BTF_TYPE_EMIT(struct bpf_dynptr);
>
> I added it in bpf_dynptr_from_mem(), right?

Yes, you can add it to a related function. The BTF_TYPE_EMIT
will be optimized out by the compiler.

> Thanks
>
> Roberto