Message-ID: <f8b5ef05-4dd1-eb3e-c6e0-af47169d7b8e@meta.com>
Date:   Tue, 17 Jan 2023 10:29:21 -0800
From:   Yonghong Song <yhs@...a.com>
To:     Tiezhu Yang <yangtiezhu@...ngson.cn>,
        Alan Maguire <alan.maguire@...cle.com>,
        Eduard Zingerman <eddyz87@...il.com>,
        Andrii Nakryiko <andrii@...nel.org>,
        Mykola Lysenko <mykolal@...com>,
        Alexei Starovoitov <ast@...nel.org>,
        Daniel Borkmann <daniel@...earbox.net>,
        Martin KaFai Lau <martin.lau@...ux.dev>,
        Song Liu <song@...nel.org>, Yonghong Song <yhs@...com>,
        John Fastabend <john.fastabend@...il.com>,
        KP Singh <kpsingh@...nel.org>,
        Stanislav Fomichev <sdf@...gle.com>,
        Hao Luo <haoluo@...gle.com>, Jiri Olsa <jolsa@...nel.org>,
        Lorenzo Bianconi <lorenzo@...nel.org>
Cc:     bpf@...r.kernel.org, linux-kernel@...r.kernel.org
Subject: Re: [PATCH bpf-next] selftests/bpf: Fix undeclared identifier build
 errors of test_bpf_nf.c



On 1/17/23 1:52 AM, Tiezhu Yang wrote:
> 
> 
> On 01/17/2023 02:48 PM, Yonghong Song wrote:
>>
>>
>> On 1/16/23 5:54 AM, Alan Maguire wrote:
>>> On 16/01/2023 12:30, Eduard Zingerman wrote:
>>>> On Mon, 2023-01-16 at 12:55 +0800, Tiezhu Yang wrote:
>>>>> $ make -C tools/testing/selftests/bpf/
>>>>>
>>>>>    CLNG-BPF [test_maps] test_bpf_nf.bpf.o
>>>>> progs/test_bpf_nf.c:160:42: error: use of undeclared identifier
>>>>> 'NF_NAT_MANIP_SRC'
>>>>>                  bpf_ct_set_nat_info(ct, &saddr, sport,
>>>>> NF_NAT_MANIP_SRC);
>>>>>                                                         ^
>>>>> progs/test_bpf_nf.c:163:42: error: use of undeclared identifier
>>>>> 'NF_NAT_MANIP_DST'
>>>>>                  bpf_ct_set_nat_info(ct, &daddr, dport,
>>>>> NF_NAT_MANIP_DST);
>>>>>                                                         ^
>>>>> 2 errors generated.
>>>>>
>>>>> Copy the definitions in include/net/netfilter/nf_nat.h to 
>>>>> test_bpf_nf.c
>>>>> to fix the above build errors.
>>>>>
>>>>> Fixes: b06b45e82b59 ("selftests/bpf: add tests for
>>>>> bpf_ct_set_nat_info kfunc")
>>>>> Signed-off-by: Tiezhu Yang <yangtiezhu@...ngson.cn>
>>>>> ---
>>>>>   tools/testing/selftests/bpf/progs/test_bpf_nf.c | 5 +++++
>>>>>   1 file changed, 5 insertions(+)
>>>>>
>>>>> diff --git a/tools/testing/selftests/bpf/progs/test_bpf_nf.c
>>>>> b/tools/testing/selftests/bpf/progs/test_bpf_nf.c
>>>>> index 227e85e..114f961 100644
>>>>> --- a/tools/testing/selftests/bpf/progs/test_bpf_nf.c
>>>>> +++ b/tools/testing/selftests/bpf/progs/test_bpf_nf.c
>>>>> @@ -34,6 +34,11 @@ __be16 dport = 0;
>>>>>   int test_exist_lookup = -ENOENT;
>>>>>   u32 test_exist_lookup_mark = 0;
>>>>>   +enum nf_nat_manip_type {
>>>>> +    NF_NAT_MANIP_SRC,
>>>>> +    NF_NAT_MANIP_DST
>>>>> +};
>>>>> +
>>>>
>>>> This is confusing, when I build the kernel/tests I get the declaration
>>>> the "enum nf_nat_manip_type" from the vmlinux.h (which is included
>>>> from test_bpf_nf.c).
>>>> Which means that this patch results in compilation error with my
>>>> configuration.
>>>> Is there a chance that your kernel is configured without some
>>>> necessary netfilter
>>>> configuration options? Have you tried this patch with BPF CI?
>>>>
>>>
>>> Yep; I suspect if CONFIG_NF_NAT=m , the required definitions won't
>>> make it
>>> into vmlinux.h. The reference tools/testing/seftests/bpf/config has
>>> CONFIG_NF_NAT=y so it is at least documented in the referenced config.
>>>
>>> I'd suggest going the route of
>>>
>>> commit aa67961f3243dfff26c47769f87b4d94b07ec71f
>>> Author: Martin KaFai Lau <martin.lau@...nel.org>
>>> Date:   Tue Dec 6 11:35:54 2022 -0800
>>>
>>>      selftests/bpf: Allow building bpf tests with
>>> CONFIG_XFRM_INTERFACE=[m|n]
>>>      ...and adding/using local definitons like:
>>>
>>> enum nf_nat_manip_type_local {
>>>     NF_NAT_MANIP_SRC_LOCAL,
>>>     NF_NAT_MANIP_DST_LOCAL
>>> };
>>
>> The above won't support core, and since preserve_access_index attribute
>> does not support enum for now. We need to use bpf_core_enum_value to
>> retrieve the proper value through CORE.
>>
>> could you try the following?
>>
>> enum nf_nat_manip_type___local {
>>     NF_NAT_MANIP_SRC___LOCAL,
>>     NF_NAT_MANIP_DST___LOCAL,
>> };
> 
> This is OK, it is similar with commit 1058b6a78db2 ("selftests/bpf: Do 
> not fail build if CONFIG_NF_CONNTRACK=m/n").
> 
>>
>> ...
>> bpf_ct_set_nat_info(ct, &saddr, sport, bpf_core_enum_value(enum
>> nf_nat_manip_type___local,  NF_NAT_MANIP_SRC___LOCAL));
>> ...
>>
>> bpf_ct_set_nat_info(ct, &daddr, dport, bpf_core_enum_value(enum
>> nf_nat_manip_type___local,  NF_NAT_MANIP_DST___LOCAL));
>>
>> whether it works or not? Could you also try if the
>> enumerator sequence in enum nf_nat_manip_type___local changed?
>>
>>>
>>> ...to avoid the name clash.
>>>
>>>
>>> Alan
> 
> I tested this on x86_64 fedora 36, using config-5.17.5-300.fc36.x86_64
> to generate .config, CONFIG_NF_CONNTRACK=m, CONFIG_NF_NAT=m, there are
> no definitions of NF_NAT_MANIP_SRC and NF_NAT_MANIP_DST in vmlinux.h,
> build test_bpf_nf.c failed.

Thanks for trying. Yes, I tried in my environment and it failed because
I didn't change the enum nf_nat_manip_type type for kfunc:
int bpf_ct_set_nat_info(struct nf_conn *, union nf_inet_addr *,
                         int port, enum nf_nat_manip_type) __ksym;

But when I changed 'enum nf_nat_manip_type' to
'enum nf_nat_manip_type___local', kernel complains kfunc argument
mismatch. The reason most likely is 'enum nf_nat_manip_type___local'
is not converted to 'enum nf_nat_manip_type' by libbpf.

This might be expected as preserve_access_index attribute only
supports record (struct/union) type and libbpf might just do that.
Could you double check whether this is the case or not?

Maybe it is time to implement preserve_access_index support
for enum type in clang now.

But we need to resolve the issue, even temporarily for now.
See below.

> 
> $ grep -w CONFIG_NF_CONNTRACK /boot/config-5.17.5-300.fc36.x86_64
> CONFIG_NF_CONNTRACK=m
> $ grep -w CONFIG_NF_NAT /boot/config-5.17.5-300.fc36.x86_64
> CONFIG_NF_NAT=m
> 
> I tested with various configs, the definitions of NF_NAT_MANIP_SRC and
> NF_NAT_MANIP_DST in vmlinux.h only depend on CONFIG_NF_CONNTRACK=y.
> 
> (1) CONFIG_NF_CONNTRACK=m, CONFIG_NF_NAT=m, no definitions
> $ grep -w CONFIG_NF_CONNTRACK .config
> CONFIG_NF_CONNTRACK=m
> $ grep -w CONFIG_NF_NAT .config
> CONFIG_NF_NAT=m
> $ grep NF_NAT_MANIP_SRC tools/testing/selftests/bpf/tools/include/vmlinux.h
> $ grep NF_NAT_MANIP_DST tools/testing/selftests/bpf/tools/include/vmlinux.h
> $
> 
> (2) CONFIG_NF_CONNTRACK=m, CONFIG_NF_NAT=y, no definitions
> This case is unable, because CONFIG_NF_NAT depends on CONFIG_NF_CONNTRACK.
> 
> (3) CONFIG_NF_CONNTRACK=m, CONFIG_NF_NAT=n, no definitions
> $ grep -w CONFIG_NF_CONNTRACK .config
> CONFIG_NF_CONNTRACK=m
> $ grep -w CONFIG_NF_NAT .config
> # CONFIG_NF_NAT is not set
> $ grep NF_NAT_MANIP_SRC tools/testing/selftests/bpf/tools/include/vmlinux.h
> $ grep NF_NAT_MANIP_DST tools/testing/selftests/bpf/tools/include/vmlinux.h
> $
> 
> (4) CONFIG_NF_CONNTRACK=y, CONFIG_NF_NAT=m, have definitions
> $ grep -w CONFIG_NF_CONNTRACK .config
> CONFIG_NF_CONNTRACK=y
> $ grep -w CONFIG_NF_NAT .config
> CONFIG_NF_NAT=m
> $ grep NF_NAT_MANIP_SRC tools/testing/selftests/bpf/tools/include/vmlinux.h
>      NF_NAT_MANIP_SRC = 0,
> $ grep NF_NAT_MANIP_DST tools/testing/selftests/bpf/tools/include/vmlinux.h
>      NF_NAT_MANIP_DST = 1,
> 
> (5) CONFIG_NF_CONNTRACK=y, CONFIG_NF_NAT=y, have definitions
> $ grep -w CONFIG_NF_CONNTRACK .config
> CONFIG_NF_CONNTRACK=y
> $ grep -w CONFIG_NF_NAT .config
> CONFIG_NF_NAT=y
> $ grep NF_NAT_MANIP_SRC tools/testing/selftests/bpf/tools/include/vmlinux.h
>      NF_NAT_MANIP_SRC = 0,
> $ grep NF_NAT_MANIP_DST tools/testing/selftests/bpf/tools/include/vmlinux.h
>      NF_NAT_MANIP_DST = 1,
> 
> (6) CONFIG_NF_CONNTRACK=y, CONFIG_NF_NAT=n, have definitions
> $ grep -w CONFIG_NF_CONNTRACK .config
> CONFIG_NF_CONNTRACK=y
> $ grep -w CONFIG_NF_NAT .config
> # CONFIG_NF_NAT is not set
> $ grep NF_NAT_MANIP_SRC tools/testing/selftests/bpf/tools/include/vmlinux.h
>      NF_NAT_MANIP_SRC = 0,
> $ grep NF_NAT_MANIP_DST tools/testing/selftests/bpf/tools/include/vmlinux.h
>      NF_NAT_MANIP_DST = 1,
> 
> (7) CONFIG_NF_CONNTRACK=n, CONFIG_NF_NAT=n, no definitions
> $ grep -w CONFIG_NF_CONNTRACK .config
> # CONFIG_NF_CONNTRACK is not set
> $ grep -w CONFIG_NF_NAT .config
> $ grep NF_NAT_MANIP_SRC tools/testing/selftests/bpf/tools/include/vmlinux.h
> $ grep NF_NAT_MANIP_DST tools/testing/selftests/bpf/tools/include/vmlinux.h
> $
> 
> (8) CONFIG_NF_CONNTRACK=n, CONFIG_NF_NAT=y, no definitions
> This case is unable, because CONFIG_NF_NAT depends on CONFIG_NF_CONNTRACK.
> 
> Here is an alternative change to check whether CONFIG_NF_CONNTRACK
> is m, enum nf_nat_manip_type___local is simple, which one is better?
> 
> $ git diff tools/testing/selftests/bpf/
> diff --git a/tools/testing/selftests/bpf/Makefile 
> b/tools/testing/selftests/bpf/Makefile
> index 22533a18705e..f3cf02046c20 100644
> --- a/tools/testing/selftests/bpf/Makefile
> +++ b/tools/testing/selftests/bpf/Makefile
> @@ -325,7 +325,7 @@ endif
> 
>   CLANG_SYS_INCLUDES = $(call 
> get_sys_includes,$(CLANG),$(CLANG_TARGET_ARCH))
>   BPF_CFLAGS = -g -Werror -D__TARGET_ARCH_$(SRCARCH) $(MENDIAN)          \
> -            -I$(INCLUDE_DIR) -I$(CURDIR) -I$(APIDIR)                   \
> +            -I$(INCLUDE_DIR) -I$(CURDIR) -I$(APIDIR) -I$(TOOLSINCDIR)  \
>               -I$(abspath $(OUTPUT)/../usr/include)
> 
>   CLANG_CFLAGS = $(CLANG_SYS_INCLUDES) \
> diff --git a/tools/testing/selftests/bpf/progs/test_bpf_nf.c 
> b/tools/testing/selftests/bpf/progs/test_bpf_nf.c
> index 227e85e85dda..f2101807072f 100644
> --- a/tools/testing/selftests/bpf/progs/test_bpf_nf.c
> +++ b/tools/testing/selftests/bpf/progs/test_bpf_nf.c
> @@ -2,6 +2,7 @@
>   #include <vmlinux.h>
>   #include <bpf/bpf_helpers.h>
>   #include <bpf/bpf_endian.h>
> +#include <linux/kconfig.h>
> 
>   #define EAFNOSUPPORT 97
>   #define EPROTO 71
> @@ -34,6 +35,13 @@ __be16 dport = 0;
>   int test_exist_lookup = -ENOENT;
>   u32 test_exist_lookup_mark = 0;
> 
> +#if IS_MODULE(CONFIG_NF_CONNTRACK)
> +enum nf_nat_manip_type {
> +       NF_NAT_MANIP_SRC,
> +       NF_NAT_MANIP_DST
> +};
> +#endif
> +

The above change does not work for me. The complication failed with

   CLNG-BPF [test_maps] 
btf__core_reloc_nesting___err_missing_container.bpf.o
   CLNG-BPF [test_maps] test_sysctl_loop2.bpf.o
progs/test_bpf_nf.c:168:42: error: use of undeclared identifier 
'NF_NAT_MANIP_SRC'
                 bpf_ct_set_nat_info(ct, &saddr, sport, NF_NAT_MANIP_SRC);
                                                        ^
progs/test_bpf_nf.c:171:42: error: use of undeclared identifier 
'NF_NAT_MANIP_DST'
                 bpf_ct_set_nat_info(ct, &daddr, dport, NF_NAT_MANIP_DST);
                                                        ^
2 errors generated.


Apparently, IS_MODULE(...) is recognized but it returns 0.
I do have CONFIG_NF_CONNTRACK=m in my config.
Note that I build the kernel with KBUILD_OUTPUT=<another dir>
(make -j LLVM=1) so vmlinux is in a different place.
While I build selftest
with 'make -C tools/testing/selftests/bpf -j LLVM=1' which
is in-tree.

>   struct nf_conn;
> 
>   struct bpf_ct_opts___local {
> 
> Note that when unset CONFIG_NF_CONNTRACK, there are much more
> build errors, I do not know whether it is necessary to fix it
> and how to fix it properly. Here, I only consider the failed
> case CONFIG_NF_CONNTRACK=m.
> 
> Thanks,
> Tiezhu
> 

Powered by blists - more mailing lists