2020-10-27 00:04:04

by Arnd Bergmann

Subject: [PATCH] bpf: fix incorrect initialization of bpf_ctx_convert_map

From: Arnd Bergmann <[email protected]>

gcc -Wextra points out that a field may get overridden in some
configurations such as x86 allmodconfig, when the next index after the one
that has been assigned last already had a value, in this case for index
BPF_PROG_TYPE_SK_LOOKUP, which comes after BPF_PROG_TYPE_LSM in the list:

kernel/bpf/btf.c:4225:2: warning: initialized field overwritten [-Woverride-init]
4225 | 0, /* avoid empty array */
| ^
kernel/bpf/btf.c:4225:2: note: (near initialization for 'bpf_ctx_convert_map[30]')

Move the zero-initializer first instead. This avoids the warning since
nothing else uses index 0, and the last element does not have to be zero.

Fixes: e9ddbb7707ff ("bpf: Introduce SK_LOOKUP program type with a dedicated attach point")
Fixes: 4c80c7bc583a ("bpf: Fix build in minimal configurations, again")
Signed-off-by: Arnd Bergmann <[email protected]>
---
kernel/bpf/btf.c | 2 +-
1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
index ed7d02e8bc93..2a4a4aeeaac1 100644
--- a/kernel/bpf/btf.c
+++ b/kernel/bpf/btf.c
@@ -4218,11 +4218,11 @@ enum {
__ctx_convert_unused, /* to avoid empty enum in extreme .config */
};
static u8 bpf_ctx_convert_map[] = {
+ [0] = 0, /* avoid empty array */
#define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
[_id] = __ctx_convert##_id,
#include <linux/bpf_types.h>
#undef BPF_PROG_TYPE
- 0, /* avoid empty array */
};
#undef BPF_MAP_TYPE
#undef BPF_LINK_TYPE
--
2.27.0

2020-10-27 08:11:22

by Andrii Nakryiko

Subject: Re: [PATCH] bpf: fix incorrect initialization of bpf_ctx_convert_map

On Mon, Oct 26, 2020 at 2:04 PM Arnd Bergmann <[email protected]> wrote:
>
> From: Arnd Bergmann <[email protected]>
>
> gcc -Wextra points out that a field may get overridden in some
> configurations such as x86 allmodconfig, when the next index after the one
> that has been assigned last already had a value, in this case for index
> BPF_PROG_TYPE_SK_LOOKUP, which comes after BPF_PROG_TYPE_LSM in the list:
>
> kernel/bpf/btf.c:4225:2: warning: initialized field overwritten [-Woverride-init]
> 4225 | 0, /* avoid empty array */
> | ^
> kernel/bpf/btf.c:4225:2: note: (near initialization for 'bpf_ctx_convert_map[30]')
>
> Move the zero-initializer first instead. This avoids the warning since
> nothing else uses index 0, and the last element does not have to be zero.

Wouldn't it be cleaner and more explicit to add __MAX_BPF_PROG_TYPE to
enum bpf_prog_type in include/uapi/linux/bpf.h, similarly to how we do
it with enum bpf_attach_type? Then just specify the size of the array
here explicitly? Unless we are trying to save a few bytes for more
minimal configurations where some BPF program types are not used (but
still defined in an enum)?


>
> Fixes: e9ddbb7707ff ("bpf: Introduce SK_LOOKUP program type with a dedicated attach point")
> Fixes: 4c80c7bc583a ("bpf: Fix build in minimal configurations, again")
> Signed-off-by: Arnd Bergmann <[email protected]>
> ---
> kernel/bpf/btf.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> diff --git a/kernel/bpf/btf.c b/kernel/bpf/btf.c
> index ed7d02e8bc93..2a4a4aeeaac1 100644
> --- a/kernel/bpf/btf.c
> +++ b/kernel/bpf/btf.c
> @@ -4218,11 +4218,11 @@ enum {
> __ctx_convert_unused, /* to avoid empty enum in extreme .config */
> };
> static u8 bpf_ctx_convert_map[] = {
> + [0] = 0, /* avoid empty array */
> #define BPF_PROG_TYPE(_id, _name, prog_ctx_type, kern_ctx_type) \
> [_id] = __ctx_convert##_id,
> #include <linux/bpf_types.h>
> #undef BPF_PROG_TYPE
> - 0, /* avoid empty array */
> };
> #undef BPF_MAP_TYPE
> #undef BPF_LINK_TYPE
> --
> 2.27.0
>