[dvblast-devel] Compile failure on x86_64

[Georgi Chorbadzhiyski]

Around 10/02/2012 04:59 PM, Alexei Takaseev scribbled:
>>>> On 10/2/12 8:11 AM, Alexei Takaseev wrote:
>>>>> System ALT Linux Sisuphus 01-10-2012
>>>>> arch x86_64
>>>>> glibc 2.16
>>>>> gcc version 4.7.2 20120920 (ALT Linux 4.7.2-alt2) (GCC)
>>>>>
>>>>> Compile failure on x86_64 with:
>>> ....
>>>>> In file included from /usr/include/string.h:656:0,
>>>>>                   from en50221.c:30:
>>>>> In function 'memcpy',
>>>>>      inlined from 'TPDUSend.constprop.35' at en50221.c:318:19,
>>>>>      inlined from 'SPDUSend' at en50221.c:500:26:
>>>>> /usr/include/bits/string3.h:51:3: error: call to
>>>>> __builtin___memcpy_chk will always overflow destination buffer
>>>>> make: *** [en50221.o] Error 1
>>>>> make: Leaving directory `/usr/src/RPM/BUILD/dvblast-2.2'
>>>>> ===========================================================
>>>>>
>>>>> On the i586 architecture there are no compilation errors.
>>>>>
>>>>> Has anyone successfully compiled with the combination of x86_64 +
>>>>> gcc 4.7 + glibc 2.16 + last dvblast of git?
>>>>
>>>> Ok, I don't see it:
>>>>
>>>>      313     case T_DATA_LAST:
>>>>      314     case T_DATA_MORE:
>>>>      315     {
>>>>      316         /* i_length <= MAX_TPDU_DATA */
>>>>      317         uint8_t *p = p_data + 3;
>>>>      318         p = SetLength( p, i_length + 1 );
>>>>      319         *p++ = i_tcid;
>>>>      320
>>>>      321         if ( i_length )
>>>>      322             memcpy( p, p_content, i_length );
>>>>      323             i_size = i_length + (p - p_data);
>>>>      324         }
>>>>      325         break;
>>>>
>>>> The memcpy will overflow if i_length is > MAX_TPDU_DATA but we
>>>> never called this function with more than MAX_TPDU_DATA set as
>>>> i_length.
>>>> Am I missing something?
>>>>
>>>> This patch probably wrong, but please try it to see if the error
>>>> goes
>>>> away:
>>>>
>>>> diff --git a/en50221.c b/en50221.c
>>>> index 0724fb3..049c5c2 100644
>>>> --- a/en50221.c
>>>> +++ b/en50221.c
>>>> @@ -205,7 +205,7 @@ static uint8_t *SetLength( uint8_t *p_data,
>>>> int
>>>> i_length )
>>>>    */
>>>>
>>>>   #define MAX_TPDU_SIZE  4096
>>>> -#define MAX_TPDU_DATA  (MAX_TPDU_SIZE - 4)
>>>> +#define MAX_TPDU_DATA  (MAX_TPDU_SIZE - 8)
>>>>
>>>>   #define DATA_INDICATOR 0x80
>>>
>>> The compilation process stopped in the same place with the same
>>> error.
>>
>> In that case as far as I can see this not a bug in dvblast.
>>
>> Try with glibc != 2.16 because the bug probably there. With 2.15
>> there are
>> no such problems.
> 
> I'm sorry, due to an error in the spec-file compilation is done without changes in en50221.c.
> 
> Successful compilation of such revisions occurred when:
> 
> commit 8ae489aa5d6e2d742e1bf6ba1f4db06549eb2712
> Author: Alexei Takaseev <taf at altlinux.org>
> Date: Tue Oct 2 22:44:01 2012 0900
> 
>      Fix build by gcc 4.7 on x86_64
> 
> diff - git a/en50221.c b/en50221.c
> index 0724fb3 .. 13a2319 100644
> --- a/en50221.c
> + + + B/en50221.c
> @ @ -205,7 +205,7 @ @ Static uint8_t * SetLength (uint8_t * p_data, int i_length)
>    * /
> 
>   # define MAX_TPDU_SIZE 4096
> - # define MAX_TPDU_DATA (MAX_TPDU_SIZE - 4)
> + # define MAX_TPDU_DATA (MAX_TPDU_SIZE - 7)
> 
>   # define DATA_INDICATOR 0x80
> 
> that is not initially considered 3 bytes.

I don't get it. You said that my patch that limited MAX_TPDU_DATA to MAX_TPDU_SIZE-8
was not working but now you are saying that limiting MAX_TPDU_DATA to MAX_TPDU_SIZE-7
is working, is that correct? Lets analyze why this "fix" is needed before blindly
applying it. I think the compiler is wrong to warn (and error) on this.

-- 
Georgi Chorbadzhiyski
http://georgi.unixsol.org/