NetBSD Problem Report #38051

From kre@munnari.OZ.AU  Mon Feb 18 12:18:07 2008
Return-Path: <kre@munnari.OZ.AU>
Received: from ( [])
	by (Postfix) with ESMTP id 0A08663B938
	for <>; Mon, 18 Feb 2008 12:18:07 +0000 (UTC)
Message-Id: <>
Date: Mon, 18 Feb 2008 19:16:16 +0700 (ICT)
From: kre@munnari.OZ.AU
Subject: security/sfs recent patch uses __NetBSD_Prereq__ incorrectly
X-Send-Pr-Version: 3.95

>Number:         38051
>Category:       pkg
>Synopsis:       security/sfs recent patch uses __NetBSD_Prereq__ incorrectly
>Confidential:   no
>Severity:       serious
>Priority:       high
>Responsible:    apb
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Mon Feb 18 12:20:02 +0000 2008
>Last-Modified:  Tue Feb 19 12:45:01 +0000 2008
>Originator:     Robert Elz
>Release:        NetBSD 3.99.15  (pkgsrc current within past 4 hours)
	Prince of Songkla University
System: NetBSD 3.99.15 NetBSD 3.99.15 (GENERIC-1.696-20060125) #8: Wed Jan 25 04:59:39 ICT 2006 i386
Architecture: i386
Machine: i386
	I compile packages using pkg_comp and libkver for a NetBSD 3.0
	environment (then they work on everything I care abuot).

	Until Saturday, security/sfs compiled file.   Then

Modified Files:
	pkgsrc/security/sfs: distinfo
Added Files:
	pkgsrc/security/sfs/patches: patch-bv

Log Message:
Deal with fourth arg to mount(2) in NetBSD.  It appeared between
4.99.23 and 4.99.24.

	That's clearly something that needed doing, but it needs
	doing correctly.

	The test added was ...
		#elif defined(__NetBSD__) && __NetBSD_Prereq__(4,99,24)

	which (as any sand person would read it) tests for 4.99.24 or
	later, as a prerequsisite.

	But what it really tests is if 4.99.24 is >= __NetBSD_Version__
	which is clearly true (for me), as __NetBSD_Version__ is 3.0.0

	I (still) suspect that the "sense" of __NetBSD_Prereq__() is
	more intended to be used in LKMs, where the appropriate think
	to know is that the kernel hasn't been made newer than the
	LKM author knows about (which might mean that any internal
	kernel interface has altered, so the LKM needs to be checked
	for continued correct design).

	Attempt to build security/sfs (on presumably just about
	anything except 4.99.24 where it will presumably work).

	I suspect the correct line should be

		#elif defined(__NetBSD__) && !__NetBSD_Prereq__(4,99,23)

	which tests that 4.99.23 < __NetBSD_Version__
		(ie: !(4.99.23 >= __NetBSD_Version))

	That is, the system being compiled against is 4.99.24
	or later, which is (apparently) where the change is required.

	Probably better is just to forget __NetBSD_Prereq__() and simply
		#elif defined(__NetBSD__) && __NetBSD_Version__ >= 499002400

	I can't test either of those is exactly what is needed, as I don't
	have 4.99.23 and 4.99.24 systems around to validate it (and
	don't really feel inclined to build them just for this).

	I have submitted this PS as high/serious, not because it is
	really urgent that security/sfs get fixed, but because it is
	urgent that there be some documentation for __NetBSD_Prereq__()
	so people actually know how to use the thing (which all relates
	back to a pending PR of mine that ought to eb closed now, since
	4.0 was released, and it was only ever relevant to what was
	released as 4.0) and a discussion on tech-kern that (to me)
	only ended up demonstrating that __NetBSD_Prereq__() is either
	misunderstood or simply useless and irrelevant (and so should
	either be documented, or deleted).

	ie: After processing this PR against security/sfs, please
	change its category to kern and leave it open instead of
	closing it, if that's possible...

	ps: with the changed patch, security/sfs does build on 3.0 at least.



Responsible-Changed-From-To: pkg-manager->apb
Responsible-Changed-When: Mon, 18 Feb 2008 18:09:30 +0000
apb committed the change in question.

From: Robert Elz <kre@munnari.OZ.AU>
Subject: Re: pkg/38051: security/sfs recent patch uses __NetBSD_Prereq__ incorrectly 
Date: Tue, 19 Feb 2008 19:41:28 +0700

     Date:        Mon, 18 Feb 2008 12:20:02 +0000 (UTC)
     Message-ID:  <>

   | 	Probably better is just to forget __NetBSD_Prereq__() and simply
   | 	test
   | 		#elif defined(__NetBSD__) && __NetBSD_Version__ >= 499002400

 I see what happened now, I'm just not running a recent enough -current (or 4.0)
 I make packages to have noticed this before.   The test must be written
 this way, and __NetBSD_Prereq__() ignored, because of the change made to
 <sys/param.h> in version 1.242 of that file, that altered the sense of the
 test (it used to be >= and is now <=)

 That means the macro is useless for testing the system version, as depending
 upon the version (the very thing you don't know) it returns different answers.

 I suspect that the original >= test was probably what was originally
 intended when __NetBSD_Prereq__() was invented - but that might just because
 of the way I rationalised its design when I saw it.   For many uses the
 current (<=) test is certainly more intuitive.

 In any case, there's no way to safely use that macro without testing the
 system version first to see how the macro result needs to be interpreted,
 which would be flat out insane...

 As for __NetBSD_Prereq__() ... it is way too late now to revert the
 change to param.h (it's been changed almost 18 months, and is that way
 in 4.0 I presume), so the only rational thing to do is simply to delete it.
 Any code using it now is broken (unless it is very very very unusual)
 and the best thing to do is to cause compilation/linker errors upon it.

 If a macro or macros are needed for testing __NetBSD_Version__ then new
 ones are needed.



NetBSD Home
NetBSD PR Database Search

(Contact us) $NetBSD: query-full-pr,v 1.39 2013/11/01 18:47:49 spz Exp $
$NetBSD:,v 1.8 2006/05/07 09:23:38 tsutsui Exp $
Copyright © 1994-2007 The NetBSD Foundation, Inc. ALL RIGHTS RESERVED.