by mail.netbsd.org with SMTP; 5 Sep 2000 23:27:26 -0000
	by coconut.itojun.org (8.9.3+3.2W/3.7W) with ESMTP id IAA24775;
	Wed, 6 Sep 2000 08:27:20 +0900 (JST)
To: matthew green <mrg@eterna.com.au>
cc: thorpej@netbsd.org, tech-security@netbsd.org
In-reply-to: mrg's message of Wed, 06 Sep 2000 10:11:21 +1100.
      <25398.968195481@eterna.com.au>
Subject: Re: openssl build on 64bit arch 
From: itojun@iijlab.net
MIME-Version: 1.0
Content-Type: multipart/mixed; boundary="----- =_aaaaaaaaaa0"
Content-ID: <24763.968196384.0@coconut.itojun.org>
Date: Wed, 06 Sep 2000 08:27:20 +0900
Message-ID: <24773.968196440@coconut.itojun.org>

------- =_aaaaaaaaaa0
Content-Type: text/plain; charset="us-ascii"
Content-ID: <24763.968196384.1@coconut.itojun.org>
Content-Transfer-Encoding: 7bit

>i suggest you contact the openssl people ASAP and find out
>their input before deciding what to do.

	I see, I did.

itojun

------- =_aaaaaaaaaa0
Content-Type: message/rfc822

To: openssl-dev@openssl.org
Subject: behavior on top of LP64
From: itojun@iijlab.net
Date: Wed, 06 Sep 2000 08:25:23 +0900
Message-ID: <24732.968196323@coconut.itojun.org>

	I'd like to know about the intended behavior on 64bit architecture
	machines.

	in many places, it seems to me that "unsigned long" is assumed to be
	32bit.  For example, in des.h and blowfish.h, it seems that the code
	assume that unsigned long is 32bit.  otherwise, we can't pass a pointer
	to bytestream into bf_encrypt and encrypt/decrypt functions become less
	useful (NOTE: I understand alignment constraint difference in pointer
	cast - please put that aside for now).
		/* this one does not work if we have 64bit unsigned long */
		char *p = { 0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00 };
		char q[8];
		BF_KEY k;
		des_encrypt((BF_LONG *)p, (BF_LONG *)q, &k);
	it is also true for internal source code level.  it is assumed that
	unsigned long is 32bit.  I had troubles making openssl work on
	alpha and possibly sparc64 (on top of NetBSD).

	also, it bothers me that include files has many #ifdefs.  if some third
	party application defines a symbol which is used in openssl headers,
	the API openssl headers present will get different from the real
	API in the compiled binary.

	I'm trying to go through all of these and rewrite as much as possible,
	including:
	- nuke all visible #ifdef in headers in include/openssl, to get
	  stable behavior.  #ifdef is in local headers (des_locl.h) or
	  Makefiles.
	- use types like uint32_t or u_int32_t, when we want a type with
	  fixed size.  if these types are not available, find them at
	  configuration time.
	it may involve some updates into build framework (like Configure and
	Makefile.ssl).
	is it worth a effort, or do I misunderstand something?

itojun

------- =_aaaaaaaaaa0--