2022-10-04 13:56:39

by Roberto Sassu

[permalink] [raw]
Subject: [RESEND][PATCH 0/6] Add _opts variant for bpf_*_get_fd_by_id()

From: Roberto Sassu <[email protected]>

NOTE: resending with libbpf_get_fd_opts test added to deny list for s390x.

Add the _opts variant for bpf_*_get_fd_by_id() functions, to be able to
pass to the kernel more options, when requesting a fd of an eBPF object to
the kernel.

Pass the options through a newly introduced structure, bpf_get_fd_opts,
which currently contains open_flags (the other two members are for
compatibility and for padding).

open_flags allows the caller to request specific permissions to access a
map (e.g. read-only). This is useful for example in the situation where a
map is write-protected.

Besides patches 2-6, which introduce the new variants and the data
structure, patch 1 fixes the LIBBPF_1.0.0 declaration in libbpf.map.

Roberto Sassu (6):
libbpf: Fix LIBBPF_1.0.0 declaration in libbpf.map
libbpf: Define bpf_get_fd_opts and introduce
bpf_map_get_fd_by_id_opts()
libbpf: Introduce bpf_prog_get_fd_by_id_opts()
libbpf: Introduce bpf_btf_get_fd_by_id_opts()
libbpf: Introduce bpf_link_get_fd_by_id_opts()
selftests/bpf: Add tests for _opts variants of bpf_*_get_fd_by_id()

tools/lib/bpf/bpf.c | 47 +++++++++-
tools/lib/bpf/bpf.h | 16 ++++
tools/lib/bpf/libbpf.map | 6 +-
tools/testing/selftests/bpf/DENYLIST.s390x | 1 +
.../bpf/prog_tests/libbpf_get_fd_opts.c | 88 +++++++++++++++++++
.../bpf/progs/test_libbpf_get_fd_opts.c | 36 ++++++++
6 files changed, 189 insertions(+), 5 deletions(-)
create mode 100644 tools/testing/selftests/bpf/prog_tests/libbpf_get_fd_opts.c
create mode 100644 tools/testing/selftests/bpf/progs/test_libbpf_get_fd_opts.c

--
2.25.1


2022-10-04 14:03:25

by Roberto Sassu

[permalink] [raw]
Subject: [RESEND][PATCH 4/6] libbpf: Introduce bpf_btf_get_fd_by_id_opts()

From: Roberto Sassu <[email protected]>

Introduce bpf_btf_get_fd_by_id_opts(), for symmetry with
bpf_map_get_fd_by_id_opts(), to let the caller pass the newly introduced
data structure bpf_get_fd_opts. Keep the existing bpf_btf_get_fd_by_id(),
and call bpf_btf_get_fd_by_id_opts() with NULL as opts argument, to prevent
setting open_flags.

Currently, the kernel does not support non-zero open_flags for
bpf_btf_get_fd_by_id_opts(), and a call with them will result in an error
returned by the bpf() system call. The caller should always pass zero
open_flags.

Signed-off-by: Roberto Sassu <[email protected]>
---
tools/lib/bpf/bpf.c | 12 +++++++++++-
tools/lib/bpf/bpf.h | 2 ++
tools/lib/bpf/libbpf.map | 1 +
3 files changed, 14 insertions(+), 1 deletion(-)

diff --git a/tools/lib/bpf/bpf.c b/tools/lib/bpf/bpf.c
index 92464c2d1da7..f43e8c8afbd3 100644
--- a/tools/lib/bpf/bpf.c
+++ b/tools/lib/bpf/bpf.c
@@ -980,19 +980,29 @@ int bpf_map_get_fd_by_id(__u32 id)
return bpf_map_get_fd_by_id_opts(id, NULL);
}

-int bpf_btf_get_fd_by_id(__u32 id)
+int bpf_btf_get_fd_by_id_opts(__u32 id,
+ const struct bpf_get_fd_opts *opts)
{
const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
union bpf_attr attr;
int fd;

+ if (!OPTS_VALID(opts, bpf_get_fd_opts))
+ return libbpf_err(-EINVAL);
+
memset(&attr, 0, attr_sz);
attr.btf_id = id;
+ attr.open_flags = OPTS_GET(opts, open_flags, 0);

fd = sys_bpf_fd(BPF_BTF_GET_FD_BY_ID, &attr, attr_sz);
return libbpf_err_errno(fd);
}

+int bpf_btf_get_fd_by_id(__u32 id)
+{
+ return bpf_btf_get_fd_by_id_opts(id, NULL);
+}
+
int bpf_link_get_fd_by_id(__u32 id)
{
const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
diff --git a/tools/lib/bpf/bpf.h b/tools/lib/bpf/bpf.h
index 595d342fb7f9..559af0b84299 100644
--- a/tools/lib/bpf/bpf.h
+++ b/tools/lib/bpf/bpf.h
@@ -379,6 +379,8 @@ LIBBPF_API int bpf_prog_get_fd_by_id(__u32 id);
LIBBPF_API int bpf_map_get_fd_by_id_opts(__u32 id,
const struct bpf_get_fd_opts *opts);
LIBBPF_API int bpf_map_get_fd_by_id(__u32 id);
+LIBBPF_API int bpf_btf_get_fd_by_id_opts(__u32 id,
+ const struct bpf_get_fd_opts *opts);
LIBBPF_API int bpf_btf_get_fd_by_id(__u32 id);
LIBBPF_API int bpf_link_get_fd_by_id(__u32 id);
LIBBPF_API int bpf_obj_get_info_by_fd(int bpf_fd, void *info, __u32 *info_len);
diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
index c3604eaa220d..7011d5eec67b 100644
--- a/tools/lib/bpf/libbpf.map
+++ b/tools/lib/bpf/libbpf.map
@@ -371,6 +371,7 @@ LIBBPF_1.0.0 {

LIBBPF_1.1.0 {
global:
+ bpf_btf_get_fd_by_id_opts;
bpf_map_get_fd_by_id_opts;
bpf_prog_get_fd_by_id_opts;
user_ring_buffer__discard;
--
2.25.1

2022-10-05 23:08:18

by Andrii Nakryiko

[permalink] [raw]
Subject: Re: [RESEND][PATCH 4/6] libbpf: Introduce bpf_btf_get_fd_by_id_opts()

On Tue, Oct 4, 2022 at 6:18 AM Roberto Sassu
<[email protected]> wrote:
>
> From: Roberto Sassu <[email protected]>
>
> Introduce bpf_btf_get_fd_by_id_opts(), for symmetry with
> bpf_map_get_fd_by_id_opts(), to let the caller pass the newly introduced
> data structure bpf_get_fd_opts. Keep the existing bpf_btf_get_fd_by_id(),
> and call bpf_btf_get_fd_by_id_opts() with NULL as opts argument, to prevent
> setting open_flags.
>
> Currently, the kernel does not support non-zero open_flags for
> bpf_btf_get_fd_by_id_opts(), and a call with them will result in an error
> returned by the bpf() system call. The caller should always pass zero
> open_flags.
>
> Signed-off-by: Roberto Sassu <[email protected]>
> ---
> tools/lib/bpf/bpf.c | 12 +++++++++++-
> tools/lib/bpf/bpf.h | 2 ++
> tools/lib/bpf/libbpf.map | 1 +
> 3 files changed, 14 insertions(+), 1 deletion(-)
>
> diff --git a/tools/lib/bpf/bpf.c b/tools/lib/bpf/bpf.c
> index 92464c2d1da7..f43e8c8afbd3 100644
> --- a/tools/lib/bpf/bpf.c
> +++ b/tools/lib/bpf/bpf.c
> @@ -980,19 +980,29 @@ int bpf_map_get_fd_by_id(__u32 id)
> return bpf_map_get_fd_by_id_opts(id, NULL);
> }
>
> -int bpf_btf_get_fd_by_id(__u32 id)
> +int bpf_btf_get_fd_by_id_opts(__u32 id,
> + const struct bpf_get_fd_opts *opts)
> {
> const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
> union bpf_attr attr;
> int fd;
>
> + if (!OPTS_VALID(opts, bpf_get_fd_opts))
> + return libbpf_err(-EINVAL);
> +
> memset(&attr, 0, attr_sz);
> attr.btf_id = id;
> + attr.open_flags = OPTS_GET(opts, open_flags, 0);
>
> fd = sys_bpf_fd(BPF_BTF_GET_FD_BY_ID, &attr, attr_sz);
> return libbpf_err_errno(fd);
> }
>
> +int bpf_btf_get_fd_by_id(__u32 id)
> +{
> + return bpf_btf_get_fd_by_id_opts(id, NULL);
> +}
> +
> int bpf_link_get_fd_by_id(__u32 id)
> {
> const size_t attr_sz = offsetofend(union bpf_attr, open_flags);
> diff --git a/tools/lib/bpf/bpf.h b/tools/lib/bpf/bpf.h
> index 595d342fb7f9..559af0b84299 100644
> --- a/tools/lib/bpf/bpf.h
> +++ b/tools/lib/bpf/bpf.h
> @@ -379,6 +379,8 @@ LIBBPF_API int bpf_prog_get_fd_by_id(__u32 id);
> LIBBPF_API int bpf_map_get_fd_by_id_opts(__u32 id,
> const struct bpf_get_fd_opts *opts);
> LIBBPF_API int bpf_map_get_fd_by_id(__u32 id);
> +LIBBPF_API int bpf_btf_get_fd_by_id_opts(__u32 id,
> + const struct bpf_get_fd_opts *opts);

It's subjective, but opts variants of APIs, being more "advanced"
usually are listed after their simpler non-opts variants. Can you
please add new _opts APIs after their non-opts variants?

> LIBBPF_API int bpf_btf_get_fd_by_id(__u32 id);
> LIBBPF_API int bpf_link_get_fd_by_id(__u32 id);
> LIBBPF_API int bpf_obj_get_info_by_fd(int bpf_fd, void *info, __u32 *info_len);
> diff --git a/tools/lib/bpf/libbpf.map b/tools/lib/bpf/libbpf.map
> index c3604eaa220d..7011d5eec67b 100644
> --- a/tools/lib/bpf/libbpf.map
> +++ b/tools/lib/bpf/libbpf.map
> @@ -371,6 +371,7 @@ LIBBPF_1.0.0 {
>
> LIBBPF_1.1.0 {
> global:
> + bpf_btf_get_fd_by_id_opts;
> bpf_map_get_fd_by_id_opts;
> bpf_prog_get_fd_by_id_opts;
> user_ring_buffer__discard;
> --
> 2.25.1
>

2022-10-05 23:48:24

by Andrii Nakryiko

[permalink] [raw]
Subject: Re: [RESEND][PATCH 0/6] Add _opts variant for bpf_*_get_fd_by_id()

On Tue, Oct 4, 2022 at 6:18 AM Roberto Sassu
<[email protected]> wrote:
>
> From: Roberto Sassu <[email protected]>
>
> NOTE: resending with libbpf_get_fd_opts test added to deny list for s390x.

it's not a resend if you change something in patches, it's a new
version, please mark it appropriately with v2 instead of [RESEND]

>
> Add the _opts variant for bpf_*_get_fd_by_id() functions, to be able to
> pass to the kernel more options, when requesting a fd of an eBPF object to
> the kernel.
>
> Pass the options through a newly introduced structure, bpf_get_fd_opts,
> which currently contains open_flags (the other two members are for
> compatibility and for padding).
>
> open_flags allows the caller to request specific permissions to access a
> map (e.g. read-only). This is useful for example in the situation where a
> map is write-protected.
>
> Besides patches 2-6, which introduce the new variants and the data
> structure, patch 1 fixes the LIBBPF_1.0.0 declaration in libbpf.map.
>
> Roberto Sassu (6):
> libbpf: Fix LIBBPF_1.0.0 declaration in libbpf.map
> libbpf: Define bpf_get_fd_opts and introduce
> bpf_map_get_fd_by_id_opts()
> libbpf: Introduce bpf_prog_get_fd_by_id_opts()
> libbpf: Introduce bpf_btf_get_fd_by_id_opts()
> libbpf: Introduce bpf_link_get_fd_by_id_opts()
> selftests/bpf: Add tests for _opts variants of bpf_*_get_fd_by_id()
>
> tools/lib/bpf/bpf.c | 47 +++++++++-
> tools/lib/bpf/bpf.h | 16 ++++
> tools/lib/bpf/libbpf.map | 6 +-
> tools/testing/selftests/bpf/DENYLIST.s390x | 1 +
> .../bpf/prog_tests/libbpf_get_fd_opts.c | 88 +++++++++++++++++++
> .../bpf/progs/test_libbpf_get_fd_opts.c | 36 ++++++++
> 6 files changed, 189 insertions(+), 5 deletions(-)
> create mode 100644 tools/testing/selftests/bpf/prog_tests/libbpf_get_fd_opts.c
> create mode 100644 tools/testing/selftests/bpf/progs/test_libbpf_get_fd_opts.c
>
> --
> 2.25.1
>