Re: [PATCH bpf] bpf: fix "unresolved symbol" build error with resolve_btfids

[Yonghong Song <yhs@fb.com>]

On 9/30/20 3:59 PM, Martin KaFai Lau wrote:
> On Wed, Sep 30, 2020 at 03:50:10PM -0700, Yonghong Song wrote:
>>
>>
>> On 9/30/20 1:58 PM, Martin KaFai Lau wrote:
>>> On Wed, Sep 30, 2020 at 09:41:09AM -0700, Yonghong Song wrote:
>>>> Michal reported a build failure likes below:
>>>>      BTFIDS  vmlinux
>>>>      FAILED unresolved symbol tcp_timewait_sock
>>>>      make[1]: *** [/.../linux-5.9-rc7/Makefile:1176: vmlinux] Error 255
>>>>
>>>> This error can be triggered when config has CONFIG_NET enabled
>>>> but CONFIG_INET disabled. In this case, there is no user of
>>>> structs inet_timewait_sock and tcp_timewait_sock and hence vmlinux BTF
>>>> types are not generated for these two structures.
>>>>
>>>> To fix the problem, omit the above two types for BTF_SOCK_TYPE_xxx
>>>> macro if CONFIG_INET is not defined.
>>>>
>>>> Fixes: fce557bcef11 ("bpf: Make btf_sock_ids global")
>>>> Reported-by: Michal Kubecek <mkubecek@suse.cz>
>>>> Signed-off-by: Yonghong Song <yhs@fb.com>
>>>> ---
>>>>    include/linux/btf_ids.h | 20 ++++++++++++++++----
>>>>    1 file changed, 16 insertions(+), 4 deletions(-)
>>>>
>>>> diff --git a/include/linux/btf_ids.h b/include/linux/btf_ids.h
>>>> index 4867d549e3c1..d9a1e18d0921 100644
>>>> --- a/include/linux/btf_ids.h
>>>> +++ b/include/linux/btf_ids.h
>>>> @@ -102,24 +102,36 @@ asm(							\
>>>>     * skc_to_*_sock() helpers. All these sockets should have
>>>>     * sock_common as the first argument in its memory layout.
>>>>     */
>>>> -#define BTF_SOCK_TYPE_xxx \
>>>> +
>>>> +#define __BTF_SOCK_TYPE_xxx \
>>>>    	BTF_SOCK_TYPE(BTF_SOCK_TYPE_INET, inet_sock)			\
>>>>    	BTF_SOCK_TYPE(BTF_SOCK_TYPE_INET_CONN, inet_connection_sock)	\
>>>>    	BTF_SOCK_TYPE(BTF_SOCK_TYPE_INET_REQ, inet_request_sock)	\
>>>> -	BTF_SOCK_TYPE(BTF_SOCK_TYPE_INET_TW, inet_timewait_sock)	\
>>>>    	BTF_SOCK_TYPE(BTF_SOCK_TYPE_REQ, request_sock)			\
>>>>    	BTF_SOCK_TYPE(BTF_SOCK_TYPE_SOCK, sock)				\
>>>>    	BTF_SOCK_TYPE(BTF_SOCK_TYPE_SOCK_COMMON, sock_common)		\
>>>>    	BTF_SOCK_TYPE(BTF_SOCK_TYPE_TCP, tcp_sock)			\
>>>>    	BTF_SOCK_TYPE(BTF_SOCK_TYPE_TCP_REQ, tcp_request_sock)		\
>>>> -	BTF_SOCK_TYPE(BTF_SOCK_TYPE_TCP_TW, tcp_timewait_sock)		\
>>>>    	BTF_SOCK_TYPE(BTF_SOCK_TYPE_TCP6, tcp6_sock)			\
>>>>    	BTF_SOCK_TYPE(BTF_SOCK_TYPE_UDP, udp_sock)			\
>>>>    	BTF_SOCK_TYPE(BTF_SOCK_TYPE_UDP6, udp6_sock)
>>>> +#define __BTF_SOCK_TW_TYPE_xxx \
>>>> +	BTF_SOCK_TYPE(BTF_SOCK_TYPE_INET_TW, inet_timewait_sock)	\
>>>> +	BTF_SOCK_TYPE(BTF_SOCK_TYPE_TCP_TW, tcp_timewait_sock)
>>>> +
>>>> +#ifdef CONFIG_INET
>>>> +#define BTF_SOCK_TYPE_xxx						\
>>>> +	__BTF_SOCK_TYPE_xxx						\
>>>> +	__BTF_SOCK_TW_TYPE_xxx
>>>> +#else
>>>> +#define BTF_SOCK_TYPE_xxx	__BTF_SOCK_TYPE_xxx
>>> BTF_SOCK_TYPE_xxx is used in BTF_ID_LIST_GLOBAL(btf_sock_ids) in filter.c
>>> which does not include BTF_SOCK_TYPE_TCP_TW.
>>> However, btf_sock_ids[BTF_SOCK_TYPE_TCP_TW] is still used
>>> in bpf_skc_to_tcp_timewait_sock_proto.
>>>
>>>> +#endif
>>>> +
>>>>    enum {
>>>>    #define BTF_SOCK_TYPE(name, str) name,
>>>> -BTF_SOCK_TYPE_xxx
>>>> +__BTF_SOCK_TYPE_xxx
>>>> +__BTF_SOCK_TW_TYPE_xxx
>>> BTF_SOCK_TYPE_TCP_TW is at the end of this enum.
>>>
>>> Would btf_sock_ids[BTF_SOCK_TYPE_TCP_TW] always be 0?
>>
>> No. If CONFIG_INET is y, the above BTF_SOCK_TYPE_xxx contains
>>    __BTF_SOCK_TW_TYPE_xxx
>> and
>>    btf_sock_ids[BTF_SOCK_TYPE_TCP_TW] will be calculated properly.
>>
>> But if CONFIG_INET is n, then BTF_SOCK_TYPE_xxx will not contain
>>     __BTF_SOCK_TW_TYPE_xxx
>> so btf_sock_ids[BTF_SOCK_TYPE_TCP_TW] will have default value 0
>> as btf_sock_ids is a global.
> I could be missing something here.
> Why btf_sock_ids[BTF_SOCK_TYPE_TCP_TW] must be 0?
> How does BTF_ID_LIST_GLOBAL() know there is a
> btf_sock_ids[BTF_SOCK_TYPE_TCP_TW]?
> I would expect at least BTF_ID_UNUSED is required.
> Otherwise, the value will be whatever follow the last
> btf_sock_ids[BTF_SOCK_TYPE_UDP6].

Right, I realized later on when I prepare v2. Will send patch v2 shortly.

> 
>>
>> Will send v2 to add some comments to make it easy to understand.
>>
>>>
>>>>    #undef BTF_SOCK_TYPE
>>>>    MAX_BTF_SOCK_TYPE,
>>>>    };
>>>> -- 
>>>> 2.24.1
>>>>