Re: [Fwknop-discuss] [PATCH] PPC failures solved

[Blair Z. <bl...@or...>]

On 6/9/13 9:45 PM, Blair Zajac wrote:
> On 6/9/13 7:20 PM, Michael Rash wrote:
>> On Jun 09, 2013, Michael Rash wrote:
>>
>>> On Jun 09, 2013, Blair Zajac wrote:
>>>
>>>> On 6/9/13 3:29 PM, Blair Zajac wrote:
>>>>> I recalled that PPC is big endian so hacked the below patch in and was
>>>>> able to get fwknop to work. I wouldn't use the patch for a good
>>>>> commit,
>>>>> as it doesn't support 64-bit PPC systems and its duplicated across two
>>>>> files.
>>>>
>>>> BTW, this is on Mac OS X 10.5.8 and the OS doesn't define BYTEORDER
>>>> in a
>>>> standard header (I don't count ffi/*.h as standard headers):
>>>>
>>>> $ find /usr/include -type f | xargs grep BYTEORDER
>>>> /usr/include/ffi/fficonfig.h:# define BYTEORDER 1234
>>>> /usr/include/ffi/fficonfig.h:# define BYTEORDER 1234
>>>> /usr/include/ffi/fficonfig.h:# define BYTEORDER 4321
>>>> /usr/include/ffi/fficonfig.h:# define BYTEORDER 4321
>>>> /usr/include/libkern/_OSByteOrder.h:#ifndef _OS__OSBYTEORDER_H
>>>> /usr/include/libkern/_OSByteOrder.h:#define _OS__OSBYTEORDER_H
>>>> /usr/include/libkern/_OSByteOrder.h:#endif /* ! _OS__OSBYTEORDER_H */
>>>> /usr/include/libkern/i386/_OSByteOrder.h:#ifndef _OS__OSBYTEORDERI386_H
>>>> /usr/include/libkern/i386/_OSByteOrder.h:#define _OS__OSBYTEORDERI386_H
>>>> /usr/include/libkern/i386/_OSByteOrder.h:#endif /* !
>>>> _OS__OSBYTEORDERI386_H */
>>>> /usr/include/libkern/i386/OSByteOrder.h:#ifndef _OS_OSBYTEORDERI386_H
>>>> /usr/include/libkern/i386/OSByteOrder.h:#define _OS_OSBYTEORDERI386_H
>>>> /usr/include/libkern/i386/OSByteOrder.h:#endif /* !
>>>> _OS_OSBYTEORDERI386_H */
>>>> /usr/include/libkern/machine/OSByteOrder.h:#ifndef
>>>> _OS_OSBYTEORDERMACHINE_H
>>>> /usr/include/libkern/machine/OSByteOrder.h:#define
>>>> _OS_OSBYTEORDERMACHINE_H
>>>> /usr/include/libkern/machine/OSByteOrder.h:#endif /* !
>>>> _OS_OSBYTEORDERMACHINE_H */
>>>> /usr/include/libkern/OSByteOrder.h:#ifndef _OS_OSBYTEORDER_H
>>>> /usr/include/libkern/OSByteOrder.h:#define _OS_OSBYTEORDER_H
>>>> /usr/include/libkern/OSByteOrder.h:#endif /* ! _OS_OSBYTEORDER_H */
>>>> /usr/include/libkern/ppc/OSByteOrder.h:#ifndef _OS_OSBYTEORDERPPC_H
>>>> /usr/include/libkern/ppc/OSByteOrder.h:#define _OS_OSBYTEORDERPPC_H
>>>> /usr/include/libkern/ppc/OSByteOrder.h:#endif /* !
>>>> _OS_OSBYTEORDERPPC_H */
>>>> /usr/include/sys/sysctl.h:#define HW_BYTEORDER 4 /* int: machine byte
>>>> order */
>>>
>>> Interesting, and thanks for the bug report for PPC systems. Seems like
>>> fwknop could have a more generic way of making a guess for an endian
>>> value. There is a section of code in lib/fko_common.h that does some of
>>> this, but I think it could be extended:
>>>
>>> http://www.cipherdyne.org/cgi-bin/gitweb.cgi?p=fwknop.git;a=blob;f=lib/fko_common.h;h=24bb14c1bbc18d44c1927f1af440bf473d533269;hb=refs/heads/master#l91
>>>
>>>
>>> For example, does your system have either _BIG_ENDIAN or __BIG_ENDIAN__
>>> defined? If so, would the following patch work (which only defines
>>> BYTEORDER if all other current measures have failed and then forces a
>>> compile warning if this also fails)?:
>
> Yes, either __BIG_ENDIAN__ or __LITTLE_ENDIAN__ is defined by the
> compiler with no include files. The following prints either BIG or
> LITTLE if run through `gcc -E`
>
> #ifdef __BIG_ENDIAN__
> BIG
> #endif
>
> #ifdef __LITTLE_ENDIAN__
> LITTLE
> #endif
>
>> Or, a bit more elegantly:
>>
>> diff --git a/lib/fko_common.h b/lib/fko_common.h
>> index 24bb14c..40f1c5b 100644
>> --- a/lib/fko_common.h
>> +++ b/lib/fko_common.h
>> @@ -103,6 +103,12 @@
>> #else
>> #error unable to determine BYTEORDER
>> #endif
>> +#elif defined(_BIG_ENDIAN) || defined(__BIG_ENDIAN__)
>> + #define BYTEORDER 4321
>> +#elif defined(_LITTLE_ENDIAN) || defined(__LITTLE_ENDIAN__)
>> + #define BYTEORDER 1234
>> +#else
>> + #error unable to determine BYTEORDER
>
> This presumes a 32- bit architecture though, since there's code in
> lib/sha1.c that checks if BTYEORDER is 12345678 or 87654321. I think one
> could check for __ppc__ and __i386__ for 32-bit and __ppc64__ and
> __x86_64__ for 64-bit. If you don't have any of __ppc__, __i386__,
> __ppc64__ or __x86_64__ defined then one could error.

Odd thing is, on my 1-year old Linux box, BYTEORDER is 1234 instead of 
12345678, so maybe what I'm saying isn't correct.

In any case, I was thinking of something like this before I found that, 
no need to have multiple #error's.

--- lib/fko_common.h.orig	2013-06-09 21:58:24.000000000 -0700
+++ lib/fko_common.h	2013-06-09 22:02:07.000000000 -0700
@@ -100,9 +100,23 @@
      #define BYTEORDER 4321
    #elif defined(_LITTLE_ENDIAN)
      #define BYTEORDER 1234
-  #else
-    #error unable to determine BYTEORDER
    #endif
+#elif defined(_BIG_ENDIAN) || defined(__BIG_ENDIAN__)
+  #if defined(__i386__) || defined(__ppc__)
+    #define BYTEORDER 4321
+  #elif defined(__x86_64__) || defined(__ppc64)
+    #define BYTEORDER 87654321
+  #endif
+#elif defined(_LITTLE_ENDIAN) || defined(__LITTLE_ENDIAN__)
+  #if defined(__i386__) || defined(__ppc__)
+    #define BYTEORDER 1234
+  #elif defined(__x86_64__) || defined(__ppc64)
+    #define BYTEORDER 12345678
+  #endif
+#endif
+
+#ifndef BYTEORDER
+  #error unable to determine BYTEORDER
  #endif

  #ifdef WIN32

Blair