Re: [PATCH bpf-next v2] libbpf: Introduce 'btf_custom_path' to 'bpf_obj_open_opts'.

[Andrii Nakryiko]

On Thu, Jul 8, 2021 at 6:07 AM Shuyi Cheng <chengshuyi@xxxxxxxxxxxxxxxxx> wrote:
>
> In order to enable the older kernel to use the CO-RE feature, load the
> vmlinux btf of the specified path.
>
> Learn from Andrii's comments in [0], add the btf_custom_path parameter
> to bpf_obj_open_opts, you can directly use the skeleton's
> <objname>_bpf__open_opts function to pass in the btf_custom_path
> parameter.
>
> Prior to this, there was also a developer who provided a patch with
> similar functions. It is a pity that the follow-up did not continue to
> advance. See [1].
>
>         [0]https://lore.kernel.org/bpf/CAEf4BzbJZLjNoiK8_VfeVg_Vrg=9iYFv+po-38SMe=UzwDKJ=Q@xxxxxxxxxxxxxx/#t
>         [1]https://yhbt.net/lore/all/CAEf4Bzbgw49w2PtowsrzKQNcxD4fZRE6AKByX-5-dMo-+oWHHA@xxxxxxxxxxxxxx/
>
> Cc: Andrii Nakryiko <andrii@xxxxxxxxxx>
> Cc: Daniel Borkmann <daniel@xxxxxxxxxxxxx>
> Signed-off-by: Shuyi Cheng <chengshuyi@xxxxxxxxxxxxxxxxx>
> ---
> v1: https://lore.kernel.org/bpf/CAEf4BzaGjEC4t1OefDo11pj2-HfNy0BLhs_G2UREjRNTmb2u=A@xxxxxxxxxxxxxx/t/#m4d9f7c6761fbd2b436b5dfe491cd864b70225804
> v1->v2:
> -- Change custom_btf_path to btf_custom_path.
> -- If the length of btf_custom_path of bpf_obj_open_opts is too long,
>    return ERR_PTR(-ENAMETOOLONG).
> -- Add `custom BTF is in addition to vmlinux BTF`
>    with btf_custom_path field.
>
>  tools/lib/bpf/libbpf.c | 27 ++++++++++++++++++++++++---
>  tools/lib/bpf/libbpf.h |  6 +++++-
>  2 files changed, 29 insertions(+), 4 deletions(-)
>
> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
> index 1e04ce7..aed156c 100644
> --- a/tools/lib/bpf/libbpf.c
> +++ b/tools/lib/bpf/libbpf.c
> @@ -494,6 +494,10 @@ struct bpf_object {
>         struct btf *btf;
>         struct btf_ext *btf_ext;
>
> +       /* custom BTF is in addition to vmlinux BTF (i.e., Use the CO-RE
> +        * feature in the old kernel).
> +        */
> +       char *btf_custom_path;
>         /* Parse and load BTF vmlinux if any of the programs in the object need
>          * it at load time.
>          */
> @@ -2679,8 +2683,15 @@ static int bpf_object__load_vmlinux_btf(struct bpf_object *obj, bool force)
>         if (!force && !obj_needs_vmlinux_btf(obj))
>                 return 0;
>
> -       obj->btf_vmlinux = libbpf_find_kernel_btf();
> -       err = libbpf_get_error(obj->btf_vmlinux);
> +       if (obj->btf_custom_path) {
> +               obj->btf_vmlinux = btf__parse(obj->btf_custom_path, NULL);

I don't think this is right. Keep in mind the general case where
vmlinux BTF might be available, but user wants to perform CO-RE using
custom BTF (like we do in our selftests in core_reloc.c). In such
cases all the other features that rely on vmlinux BTF would use real
vmlinux BTF, but CO-RE relocations would use custom BTF. See the
original patch you are referencing, it loaded btf_override separately
from obj->btf_vmlinux.

> +               err = libbpf_get_error(obj->btf_vmlinux);
> +               pr_debug("loading custom vmlinux BTF '%s': %d\n", obj->btf_custom_path, err);
> +       } else {
> +               obj->btf_vmlinux = libbpf_find_kernel_btf();
> +               err = libbpf_get_error(obj->btf_vmlinux);
> +       }
> +
>         if (err) {
>                 pr_warn("Error loading vmlinux BTF: %d\n", err);
>                 obj->btf_vmlinux = NULL;
> @@ -7554,7 +7565,7 @@ int bpf_program__load(struct bpf_program *prog, char *license, __u32 kern_ver)
>  __bpf_object__open(const char *path, const void *obj_buf, size_t obj_buf_sz,
>                    const struct bpf_object_open_opts *opts)
>  {
> -       const char *obj_name, *kconfig;
> +       const char *obj_name, *kconfig, *btf_tmp_path;
>         struct bpf_program *prog;
>         struct bpf_object *obj;
>         char tmp_name[64];
> @@ -7584,6 +7595,15 @@ int bpf_program__load(struct bpf_program *prog, char *license, __u32 kern_ver)
>         obj = bpf_object__new(path, obj_buf, obj_buf_sz, obj_name);
>         if (IS_ERR(obj))
>                 return obj;
> +
> +       btf_tmp_path = OPTS_GET(opts, btf_custom_path, NULL);
> +       if (btf_tmp_path) {
> +               if (strlen(btf_tmp_path) >= PATH_MAX)
> +                       return ERR_PTR(-ENAMETOOLONG);

we are leaking obj here

> +               obj->btf_custom_path = strdup(btf_tmp_path);
> +               if (!obj->btf_custom_path)
> +                       return ERR_PTR(-ENOMEM);

and here

> +       }
>
>         kconfig = OPTS_GET(opts, kconfig, NULL);
>         if (kconfig) {

And a few lines below. You didn't introduce this bug, I just spotted
it while reviewing your patch, but it would be nice to fix it as well.

> @@ -8702,6 +8722,7 @@ void bpf_object__close(struct bpf_object *obj)
>         for (i = 0; i < obj->nr_maps; i++)
>                 bpf_map__destroy(&obj->maps[i]);
>
> +       zfree(&obj->btf_custom_path);
>         zfree(&obj->kconfig);
>         zfree(&obj->externs);
>         obj->nr_extern = 0;
> diff --git a/tools/lib/bpf/libbpf.h b/tools/lib/bpf/libbpf.h
> index 6e61342..5002d1f 100644
> --- a/tools/lib/bpf/libbpf.h
> +++ b/tools/lib/bpf/libbpf.h
> @@ -94,8 +94,12 @@ struct bpf_object_open_opts {
>          * system Kconfig for CONFIG_xxx externs.
>          */
>         const char *kconfig;
> +       /* custom BTF is in addition to vmlinux BTF (i.e., Use the CO-RE
> +        * feature in the old kernel).
> +        */
> +       char *btf_custom_path;
>  };
> -#define bpf_object_open_opts__last_field kconfig
> +#define bpf_object_open_opts__last_field btf_custom_path
>
>  LIBBPF_API struct bpf_object *bpf_object__open(const char *path);
>  LIBBPF_API struct bpf_object *
> --
> 1.8.3.1
>