NetBSD Problem Report #3182

Received: (qmail-queue invoked from smtpd); 2 Feb 1997 11:31:21 -0000
Message-Id: <199702021124.MAA00537@quesera.inka.de>
Date: Sun, 2 Feb 1997 12:24:27 +0100 (CET)
From: Klaus Klein <kleink@layla.inka.de>
Reply-To: kleink@layla.inka.de
To: gnats-bugs@gnats.netbsd.org
Subject: waitpid(2) and friends silently ignore invalid `options' values
X-Send-Pr-Version: 3.95

>Number:         3182
>Category:       standards
>Synopsis:       waitpid(2) and friends silently ignore invalid `options' values
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    gnats-admin
>State:          closed
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sun Feb 02 03:35:00 +0000 1997
>Closed-Date:    Thu Feb 06 21:35:52 +0000 1997
>Last-Modified:  Fri Oct 28 09:16:01 +0000 2005
>Originator:     Klaus Klein
>Release:        970201
>Organization:
private site
>Environment:
System: NetBSD quesera.inka.de 1.2C NetBSD 1.2C (QUESERA) #72: Sun Feb 2 11:27:39 CET 1997 kleink@quesera.inka.de:/usr/src/sys/arch/hp300/compile/QUESERA hp300


>Description:
Given an invalid value as `options' parameter (i.e. 4), waitpid(2), wait4(2)
and wait3(2) do not return with EINVAL set; they just check for "known"
bits set and silently ignore invalid ones.

Though I do not have the POSIX specs for waitpid() at hand to look this up,
I'm under the impression that this is a spec violation, as
  i) commercial/"compliant" implementations I've checked do return EINVAL
 ii) NIST-PCTS:151-2 requires an "options" value to be specified
     upon which EINVAL will be returned

I have no idea why this check is #ifdef'ed out in kern_exit.c:wait4();
apparently it's been so at least since lite2.

>How-To-Repeat:
_Try_ to figure out a value for EINVAL_WAITPID in NIST-PCTS:151-2. :-)

>Fix:

Index: src/lib/libc/sys/wait.2
===================================================================
RCS file: /cvsroot/netbsd/src/lib/libc/sys/wait.2,v
retrieving revision 1.1.1.1
retrieving revision 1.1.1.1.2.1
diff -b -c -r1.1.1.1 -r1.1.1.1.2.1
*** wait.2	1996/12/20 13:21:14	1.1.1.1
--- wait.2	1997/02/02 02:15:16	1.1.1.1.2.1
***************
*** 272,277 ****
--- 272,289 ----
  .Dv SA_RESTART
  flag set.
  .El
+ .Pp
+ Additionally,
+ .Fn wait4 ,
+ .Fn wait3
+ and
+ .Fn waitpid
+ will fail and return immediately if:
+ .Bl -tag -width Er
+ .It Bq Er EINVAL
+ An invalid value was specified for
+ .Fa options .
+ .El
  .Sh STANDARDS
  The
  .Fn wait
Index: src/sys/kern/kern_exit.c
===================================================================
RCS file: /cvsroot/netbsd/src/sys/kern/kern_exit.c,v
retrieving revision 1.1.1.1
retrieving revision 1.1.1.1.2.2
diff -b -c -r1.1.1.1 -r1.1.1.1.2.2
*** kern_exit.c	1996/12/20 13:28:34	1.1.1.1
--- kern_exit.c	1997/02/02 09:48:48	1.1.1.1.2.2
***************
*** 296,305 ****

  	if (SCARG(uap, pid) == 0)
  		SCARG(uap, pid) = -q->p_pgid;
- #ifdef notyet
  	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG))
  		return (EINVAL);
! #endif
  loop:
  	nfound = 0;
  	for (p = q->p_children.lh_first; p != 0; p = p->p_sibling.le_next) {
--- 296,304 ----

  	if (SCARG(uap, pid) == 0)
  		SCARG(uap, pid) = -q->p_pgid;
  	if (SCARG(uap, options) &~ (WUNTRACED|WNOHANG))
  		return (EINVAL);
! 
  loop:
  	nfound = 0;
  	for (p = q->p_children.lh_first; p != 0; p = p->p_sibling.le_next) {
>Release-Note:
>Audit-Trail:

State-Changed-From-To: open->closed
State-Changed-By: mikel
State-Changed-When: Thu Feb  6 21:35:52 1997
State-Changed-Why:
fixed as requested.  I do have a copy of POSIX.1 handy, and it is true
that waitpid() must return EINVAL if the options argument is invalid.

From: Klaus Klein <kleink@layla.inka.de>
To: kleink@quesera.inka.de
Cc: gnats-bugs@NetBSD.ORG, netbsd-bugs@NetBSD.ORG
Subject: Re: standards/3182: waitpid(2) and friends silently ignore invalid 
 `options' values
Date: Fri, 07 Feb 1997 13:44:31 +0100

 > Though I do not have the POSIX specs for waitpid() at hand to look this up,
 > I'm under the impression that this is a spec violation, as
 >   i) commercial/"compliant" implementations I've checked do return EINVAL
 >  ii) NIST-PCTS:151-2 requires an "options" value to be specified
 >      upon which EINVAL will be returned

 This time I have IEEE Std 1003.1-1988 at hand, and section 3.2.1.4 states:
 "[EINVAL]	The value of the options argument is not valid."


 -klaus
From: Geert Hendrickx <ghen@telenet.be>
To: gnats-bugs@netbsd.org
Cc: jschauma@netmeister.org
Subject: Re: pkg/3182: www/opera should be available for Linux
Date: Fri, 28 Oct 2005 11:15:54 +0200

 I would like to see this problem tackled in a more abstract way.  Packages
 could be marked as BINARY=Linux-i386, and then pkgsrc could automatically
 add ONLY_FOR_PLATFORM flags for the platforms that can run those binaries
 (either native or emulated).  

 Opera would then be flagged as BINARY=Linux-i386 FreeBSD-i386 Solaris-sparc
 and pkgsrc would pick the binary which "most closely" matches the platform
 (e.g. Linux-i386 on NetBSD and OpenBSD, but FreeBSD-i386 on DragonFly, etc).  

 Such a framework would reduce a lot of mess in the existing Linux-binary
 packages, and would also save us a lot of work when a new platform (like
 DragonFly) enters the arena, because right now, we have to add it manually
 to every (supported) Linux-binary package.  

 	Geert

>Unformatted:

NetBSD Home
NetBSD PR Database Search

(Contact us) $NetBSD: query-full-pr,v 1.51 2026/08/10 02:28:17 riastradh Exp $
$NetBSD: gnats_config.sh,v 1.10 2026/05/13 22:00:09 riastradh Exp $
Copyright © 1994-2026 The NetBSD Foundation, Inc. ALL RIGHTS RESERVED.