[OpenAFS-devel] rxk5 branch is ready; please test

[Douglas E. Engert]

In regards to the compiler failing below, I would like to question
the definition of assert in the AFS src/util/assert.h

Both gcc and the Solaris /opt/SUNWspro/bin/cc fail compiling the code.

The use of the "do {...} while"  makes the output of the define a statement,
rather then an expression. Both Solaris and Ubuntu have assert.h with
macros to define assert, and they do it so assert is defined as an expression.

Suggest something more like:

Index: src/util/assert.h
===================================================================
RCS file: /cvs/openafs/src/util/assert.h,v
retrieving revision 1.4
diff -u -r1.4 assert.h
--- src/util/assert.h   14 Jan 2003 00:46:48 -0000      1.4
+++ src/util/assert.h   14 Dec 2007 23:28:52 -0000
@@ -12,4 +12,4 @@

  void AssertionFailed(char *file, int line);

-#define assert(ex) do{if (!(ex)) AssertionFailed(__FILE__, __LINE__);}while(0)
+#define assert(ex) (void)((ex)?0:(AssertionFailed(__FILE__, __LINE__),0))



And for Problem#2:

Index: src/comerr/error_msg.c
===================================================================
RCS file: /cvs/openafs/src/comerr/error_msg.c,v
retrieving revision 1.6.14.2
diff -u -r1.6.14.2 error_msg.c
--- src/comerr/error_msg.c      26 Sep 2007 16:34:28 -0000      1.6.14.2
+++ src/comerr/error_msg.c      14 Dec 2007 23:29:55 -0000
@@ -169,7 +169,7 @@
      UNLOCK_ET_LIST;
  }

-#if 0
+#ifndef HAVE_ADD_ERROR_TABLE
  /*
   * New interfaces provided by mit krb5's com_err library
   */





Douglas E. Engert wrote:
> 
> 
> Jim Rees wrote:
>> For the past year or more, Matt Benjamin and Marcus Watts have been 
>> working
>> hard on developing rxk5, a new security mechanism for OpenAFS.  It uses
>> kerberos 5 tickets and encryption algorithms straight, and includes 
>> support
>> for all standard kerberos 5 encryption types including AES256.
>>
>> The code is currently on a branch tagged rxk5-devel-1_5_x.  Our plan 
>> is to
>> merge it to the openafs-devel-1_5_x branch, from which it will eventually
>> make its way into an official OpenAFS release at some time in the future.
>>
>> Please test this code.  Even if you don't plan to use any of the rxk5
>> features, please build it and report back here.
> 
> Trying to build with the Solaris 10 version of Kerberos. This is 
> attractive,
> as it is a vendor version, and Sun has recently released the krb5.h.
> 
> Problem #1:
> 
> The Solaris 10 /usr/include/kerberos5/krb5.h has:
>>
>>    483  #ifndef _KERNEL
>>    484  #define krb5_x(ptr,args) ((ptr)?((*(ptr)) args):(abort(),1))
>>    485  #define krb5_xc(ptr,args) ((ptr)?((*(ptr)) 
>> args):(abort(),(char*)0))
>>    486  #else
>>    487  #define krb5_x(ptr,args) ((*(ptr)) args)
>>    488  #define krb5_xc(ptr,args) ((*(ptr)) args)
>>    489  #endif
>>
>>   1276  #define krb5_cc_get_principal(context, cache, principal) 
>           krb5_x((cache)-ops->get_princ,(context, cache, principal))
> 
> It looks like the intent is to abort if a function pointer is NULL.
> 
> AFS util/assert.h has:
>  #define assert(ex) do{if (!(ex)) AssertionFailed(__FILE__, 
> __LINE__);}while(0)
> 
> Some how the pre processor is converting the abort to the assert and 
> producing
> code in ptuser.c around line 257 like: (wrapped in e-mail to make more 
> readable)
> 
>  what = "krb5_cc_get_principal";
>  code =  ( ( ( cc ) -> ops -> get_princ ) ? ( ( * ( ( cc ) -> ops -> 
> get_princ ) )
>      ( k5context , cc , & in_creds -> client ) ) : ( do { if ( ! ( 0 ))
>       AssertionFailed ( 
> "/afs/anl.gov/appl/OpenAFS-dev3/build/1.5.x-rxk5/@sys/src/ptserver/ptuser.c" 
> 
>        , 258 ) ; } while ( 0 ) , 1 ) );
>   if(code) goto Failed;
> 
> The compiler Sun compiler (/opt/SUNWspro/bin/cc -V ,cc: Sun C 5.8 Patch 
> 121015-05 2007/08/01)
> does not like the do:
> 
> "/afs/anl.gov/appl/OpenAFS-dev3/build/1.5.x-rxk5/@sys/src/ptserver/ptuser.c", 
> line 257: syntax error before or at: do
> "/afs/anl.gov/appl/OpenAFS-dev3/build/1.5.x-rxk5/@sys/src/ptserver/ptuser.c", 
> line 257: syntax error before or at: ,
> 
> 
> 
> The problem first appears the JUAFS compiles of ptserver/ptuser.c
> and  auth/rxk5_utilafs.c
> 
> A hack was applied to get the compile to proceed.
> 
> Problem #2:
> 
> Solaris does not have add_error_table or remove_error_table.
> libafsrpc.so.1.1 references these, found when trying to link
> the linktest in shlibafsrpc
> 
> 
> 
> 
> 
>> OpenAFS-devel mailing list
>> OpenAFS-devel@openafs.org
>> https://lists.openafs.org/mailman/listinfo/openafs-devel
>>
>>
> 

-- 

  Douglas E. Engert  <DEEngert@anl.gov>
  Argonne National Laboratory
  9700 South Cass Avenue
  Argonne, Illinois  60439
  (630) 252-5444