Re: [PATCH v2 02/13] bpf tools: Extract and collect map names from BPF object file

["Wangnan (F)" <wangnan0@huawei.com>]

On 2015/11/30 17:43, Namhyung Kim wrote:
> On November 30, 2015 6:27:57 PM GMT+09:00, "Wangnan (F)" <wangnan0@huawei.com> wrote:
>>
>> On 2015/11/30 16:51, Namhyung Kim wrote:
>>> On Mon, Nov 30, 2015 at 01:00:46PM +0800, Wangnan (F) wrote:
>>>> On 2015/11/30 0:14, Namhyung Kim wrote:
>>>>> Hi Wang,
>>>>>
>>>>> On Fri, Nov 27, 2015 at 08:47:36AM +0000, Wang Nan wrote:
>>>>>> This patch collects name of maps in BPF object files and saves
>> them into
>>>>>> 'maps' field in 'struct bpf_object'. 'bpf_object__get_map_by_name'
>> is
>>>>>> introduced to retrive fd and definitions of a map through its
>> name.
>>>>>> Signed-off-by: Wang Nan <wangnan0@huawei.com>
>>>>>> Signed-off-by: He Kuang <hekuang@huawei.com>
>>>>>> Cc: Alexei Starovoitov <ast@kernel.org>
>>>>>> Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
>>>>>> Cc: Masami Hiramatsu <masami.hiramatsu.pt@hitachi.com>
>>>>>> Cc: Namhyung Kim <namhyung@kernel.org>
>>>>>> Cc: Zefan Li <lizefan@huawei.com>
>>>>>> Cc: pi3orama@163.com
>>>>>> ---
>>>>>>    tools/lib/bpf/libbpf.c | 65
>> +++++++++++++++++++++++++++++++++++++++++++++++---
>>>>>>    tools/lib/bpf/libbpf.h |  3 +++
>>>>>>    2 files changed, 65 insertions(+), 3 deletions(-)
>>>>>>
>>>>>> diff --git a/tools/lib/bpf/libbpf.c b/tools/lib/bpf/libbpf.c
>>>>>> index f509825..a298614 100644
>>>>>> --- a/tools/lib/bpf/libbpf.c
>>>>>> +++ b/tools/lib/bpf/libbpf.c
>>>>>> @@ -165,6 +165,7 @@ struct bpf_program {
>>>>>>    struct bpf_map {
>>>>>>    	int fd;
>>>>>> +	char *name;
>>>>>>    	struct bpf_map_def def;
>>>>>>    	void *priv;
>>>>>>    	bpf_map_clear_priv_t clear_priv;
>>>>>> @@ -526,12 +527,46 @@ bpf_object__init_maps(struct bpf_object
>> *obj, void *data,
>>>>>>    	return 0;
>>>>>>    }
>>>>>> +static void
>>>>>> +bpf_object__init_maps_name(struct bpf_object *obj, int
>> maps_shndx)
>>>>>> +{
>>>>>> +	int i;
>>>>>> +	Elf_Data *symbols = obj->efile.symbols;
>>>>>> +
>>>>>> +	if (!symbols || maps_shndx < 0)
>>>>>> +		return;
>>>>>> +
>>>>>> +	for (i = 0; i < symbols->d_size / sizeof(GElf_Sym); i++) {
>>>>>> +		GElf_Sym sym;
>>>>>> +		size_t map_idx;
>>>>>> +		const char *map_name;
>>>>>> +
>>>>>> +		if (!gelf_getsym(symbols, i, &sym))
>>>>>> +			continue;
>>>>>> +		if (sym.st_shndx != maps_shndx)
>>>>>> +			continue;
>>>>>> +
>>>>>> +		map_name = elf_strptr(obj->efile.elf,
>>>>>> +				      obj->efile.ehdr.e_shstrndx,
>>>>>> +				      sym.st_name);
>>>>> It means that each map name is saved in section header string
>> table?
>>>> According to elf format specification:
>>>>
>>>> For an symbol table entry, the st_name field "holds an index
>>>> into the object file’s symbol string table, which holds the
>>>> character representations of the symbol names. If the value
>>>> is non-zero, it represents a string table index that gives
>>>> the symbol name. Otherwise, the symbol table entry has no
>>>> name."
>>>>
>>>> And so called "object file’s symbol string table" is a
>>>> section in the object file which index is stored into
>>>> ehdr and be loaded during gelf_getehdr(), and its index
>>>> would be set to ehdr->e_shstrndx. So I think for each map
>>>> its name should be saved in that string table.
>>> AFAIK there're two symbol string tables in a ELF file.  One for
>>> section headers (.shstrtab) and another for normal symbols (.strtab).
>>> And ehdr->e_shstrndx is the index of section header string table so
>>> your code assumes map names are saved in the section header string
>>> table, right?
>>>
>>> Thanks,
>>> Namhyung
>> In case of gcc:
>>
>> $ echo 'int func() {return 0;}' | gcc -x c -c -o ./temp.o -
>> $ readelf -h ./temp.o
>> ELF Header:
>>    Magic:   7f 45 4c 46 02 01 01 00 00 00 00 00 00 00 00 00
>>    Class:                             ELF64
>>    Data:                              2's complement, little endian
>>    Version:                           1 (current)
>>    OS/ABI:                            UNIX - System V
>>    ABI Version:                       0
>>    Type:                              REL (Relocatable file)
>>    Machine:                           Advanced Micro Devices X86-64
>>    Version:                           0x1
>>    Entry point address:               0x0
>>    Start of program headers:          0 (bytes into file)
>>    Start of section headers:          240 (bytes into file)
>>    Flags:                             0x0
>>    Size of this header:               64 (bytes)
>>    Size of program headers:           0 (bytes)
>>    Number of program headers:         0
>>    Size of section headers:           64 (bytes)
>>    Number of section headers:         11
>>    Section header string table index: 8
>>
>> Let's see what is section 8:
>>
>> $ readelf -S ./temp.o
>>    ...
>>    [ 8] .shstrtab         STRTAB           0000000000000000  00000098
>>         0000000000000054  0000000000000000           0     0     1
>>    ...
>>
>> Yes, in this case it is .shstrtab.
>>
>> However, this is what I found when using llvm:
>>
>> $ echo 'int func() {return 0;}' | x86_64-oe-linux-clang -x c -c -o
>> ./temp.o -
>> ELF Header:
>>    Magic:   7f 45 4c 46 02 01 01 03 00 00 00 00 00 00 00 00
>>    Class:                             ELF64
>>    Data:                              2's complement, little endian
>>    Version:                           1 (current)
>>    OS/ABI:                            UNIX - GNU
>>    ABI Version:                       0
>>    Type:                              REL (Relocatable file)
>>    Machine:                           Advanced Micro Devices X86-64
>>    Version:                           0x1
>>    Entry point address:               0x0
>>    Start of program headers:          0 (bytes into file)
>>    Start of section headers:          648 (bytes into file)
>>    Flags:                             0x0
>>    Size of this header:               64 (bytes)
>>    Size of program headers:           0 (bytes)
>>    Number of program headers:         0
>>    Size of section headers:           64 (bytes)
>>    Number of section headers:         10
>>    Section header string table index: 1
>>
>> $ readelf -S ./temp.o
>> There are 10 section headers, starting at offset 0x288:
>>
>> Section Headers:
>>    [Nr] Name              Type             Address           Offset
>>         Size              EntSize          Flags  Link  Info  Align
>>    [ 0]                   NULL             0000000000000000  00000000
>>         0000000000000000  0000000000000000           0     0     0
>>    [ 1] .strtab           STRTAB           0000000000000000  00000230
>>         0000000000000051  0000000000000000           0     0     1
>>
>> This time it is strtab.
>>
>> And here is the content of strtab:
>>
>> $ readelf -p .strtab ./temp.o
>>
>> String dump of section '.strtab':
>>    [     1]  .text
>>    [     7]  .comment
>>    [    10]  .bss
>>    [    15]  .note.GNU-stack
>>    [    25]  .rela.eh_frame
>>    [    34]  func
>>    [    39]  .strtab
>>    [    41]  .symtab
>>    [    49]  .data
>>    [    4f]  -
>>
>>
>> Note that I don't use BPF backend. This is a normal x86 compiling.
>>
>> So seems it is the default behavior of LLVM.
> Ah, didn't know that. So strtab has section header strings as well as normal symbol strings when compiled with LLVM, right?  It'd be great if you add comment about it.

I think technically speaking you are right, because I haven't see any 
documentation
about it, so I don't know the reason why LLVM behave like this, and 
don't know
whether it would change it in future. And it is also possible that we 
will have
other compiler to compile BPF source file. Now I'm reading readelf's 
code and try to
find a canonical way for it.

Thank you for your review.

> Thanks,
> Namhyung
>