[Click] wifi llc handling problem in Click 1.5

[Eddie Kohler]

Mathias,

Thanks for this clarification.  It looks like our CODE uses 
sizeof(wifi_llc_header) for stuff, which it should not -- it should use a 
constant.  I've changed this.  Could you cvs up and give it a try?

Eddie


Mathias Kurth wrote:
> Hi Eddie,
> 
> Maybe there is something wrong with my compiler but it reports a size of 4
> for the llc header definition.
> Here is a cut&paste example that demonstrates what I mean:
> 
> #include <stdlib.h> 
> #include <stdio.h>
> #include <stdint.h>
> #include <string.h>
> 
> // 6-byte LLC header (last byte is terminating NUL)
> #define wifi_llc_header ((const uint8_t *) "\xAA\xAA\x03\x00\x00")
> 
> int main()
> {
>   uint8_t* p;
>   int i;
>  
>   p = (uint8_t*)malloc( 200*sizeof(uint8_t) );
>   memset(p, '\x99', 200);
>   uint16_t ethtype = 256 + 1;
> 
>   memcpy(p, wifi_llc_header, sizeof(wifi_llc_header));
>   memcpy(p + 6, &ethtype, 2);
> 
>   printf("sizeof wifi_llc_header = %d\n", sizeof((char*)"hallo"));
>   for (i = 0; i < 20; i++)
>     printf("%.2x ", p[i]);
> 
>   return (0);
> }
> 
> It generates the following output:
> 
> sizeof wifi_llc_header = 4
> aa aa 03 00 99 99 01 01 99 99 99 99 99 99 99 99 99 99 99 99
> 
> This corresponds to the observations we made in our testbed. We
> captured the frames generated by click and did see that they contained
> an invalid llc header. 
> 
> Kind regards,
> Mathias
> 
> -----Ursprüngliche Nachricht-----
> Von: Eddie Kohler [mailto:kohler at cs.ucla.edu] 
> Gesendet: Sonntag, 17. September 2006 05:18
> An: Mathias Kurth
> Cc: click at amsterdam.lcs.mit.edu
> Betreff: Re: [Click] wifi llc handling problem in Click 1.5
> 
> Hi Mathias,
> 
> I think that code was written expecting that C would add a trailing null byte, 
> as it is required to.  In fact C reports that "sizeof("\xAA\xAA\x03\x00\x00")" 
> is 6.  The last byte is 0.  So I think your diagnosis is wrong although maybe 
> you are really seeing a problem.
> 
> Your diagnosis of the shift_data() problem is right, though!  The easiest 
> change is probably to shift the packet's headroom as well as its currently 
> active data area (which would shift the old ether header), and that is what 
> I've done.  This will affect performance, however, since we are shifting more 
> data per misaligned packet.  In our defense if you are having to align packets 
> you have performance problems anyway.
> 
> Report any problems you see.  Thanks.
> 
> Eddie
> 
> 
> 
> 
> Mathias Kurth wrote:
>> Hi,
>>
>> It seems there is something wrong with the wifi llc header 
>> in wifi.h. It does not consist of 6 bytes as expected:
>>
>> #define wifi_llc_header ((const uint8_t *) "\xAA\xAA\x03\x00\x00")
>>
>> The result is that the org-code is not zero and most of the 
>> client station's wifi drivers do not accept the generated frame.
>>
>> BTW: We've got another problem with the align element. It uses 
>> the method Packet::shift_data(), which moves the data part of 
>> the packet and adjusts the data and ether_header pointers according 
>> to the offset. But if ether_header() points to a valid Ethernet header, 
>> the header is not moved in memory so that ether_header() does not point to 
>> the header any more. Accessing ether_header() afterwards results 
>> obviously in trash data. 
>> Is it a bug? Or is the behavior intentional and we use the 
>> ether_header() in a wrong way?
>>
>> Kind regards,
>> Mathias 
>>
>>  
>>
>>  
>>
>>
>> _______________________________________________
>> click mailing list
>> click at amsterdam.lcs.mit.edu
>> https://amsterdam.lcs.mit.edu/mailman/listinfo/click
>