NetBSD Problem Report #31377

From sjg@juniper.net  Fri Sep 23 16:40:26 2005
Return-Path: <sjg@juniper.net>
Received: from colo-dns-ext1.juniper.net (colo-dns-ext1.juniper.net [207.17.137.57])
	by narn.netbsd.org (Postfix) with ESMTP id 6F80E63B852
	for <gnats-bugs@gnats.NetBSD.org>; Fri, 23 Sep 2005 16:40:26 +0000 (UTC)
Message-Id: <20050923164020.6841BEF10@ran.juniper.net>
Date: Fri, 23 Sep 2005 09:40:20 -0700 (PDT)
From: sjg@juniper.net
Reply-To: sjg@juniper.net
To: gnats-bugs@netbsd.org
Subject: ksh: cannot set limit indicated by ulimit -H
X-Send-Pr-Version: 3.95

>Number:         31377
>Category:       bin
>Synopsis:       ksh: cannot set limit indicated by ulimit -H
>Confidential:   no
>Severity:       non-critical
>Priority:       medium
>Responsible:    bin-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Sep 23 16:41:00 +0000 2005
>Last-Modified:  Mon Oct 09 22:40:01 +0000 2006
>Originator:     sjg@juniper.net
>Release:        NetBSD 3.99.9
>Organization:
>Environment:


NetBSD frodo 3.99.9 NetBSD 3.99.9 (GENERIC) #1: Mon Sep 19 10:38:07 PDT 2005  sjg@swift.juniper.net:/c/sjg/work/NetBSD/current/obj/i386/obj/c/sjg/work/NetBSD/current/src/sys/arch/i386/compile/GENERIC i386
Architecture: i386
Machine: i386
>Description:

My .profile uses a function to maximize certain limits, this function 
works on all previous NetBSD as well as many other systems.  The idea
is to set a limit to the value reported by ulimit -H as in:

max_limit() {
    # Linux uses bash as /bin/sh and unlike
    # every other sh like shell it does not
    # handle -<x>H but does handle -H<x>
    # -H -<x> works on some but not all bash versions <grrr/>
    for _ul in `IFS=" -"; echo $*`
    do
        _ulmax=`ulimit -H$_ul`
        ulimit -$_ul $_ulmax
    done
}

On 3.99.9 calling

max_limit -d

fails because:

: sjg:131; umlimit -Hd 
2097152
: sjg:132; ulimit -d 2097152  
ksh: ulimit: bad limit: Invalid argument
: sjg:133; ulimit -d 2097151 
: sjg:134; 

>How-To-Repeat:

: sjg:131; umlimit -Hd 
2097152
: sjg:132; ulimit -d 2097152
ksh: ulimit: bad limit: Invalid argument
: sjg:133; ulimit -d 2097151
: sjg:134; 

>Fix:


>Audit-Trail:
From: Simon Gerraty <sjg@juniper.net>
To: tech-userland@netbsd.org
Cc: gnats-bugs@netbsd.org, sjg@juniper.net
Subject: Re: bin/31377: ksh: cannot set limit indicated by ulimit -H 
Date: Sat, 24 Sep 2005 20:24:57 -0700 (PDT)

 Ok, the problem here is that the limit for -d happens to be
 2147483648, but because ksh munges that to/from kbyte units (factor of
 1024) and is using a signed value (rlim_t), when we try to set that
 limit (after multiplying by 1024 again) we get -2147483648 which is why
 it fails. 

 I'm not sure why rlim_t is signed, but the simple patch below avoids the
 above:

 Index: c_ulimit.c
 ===================================================================
 RCS file: /cvsroot/src/bin/ksh/c_ulimit.c,v
 retrieving revision 1.7
 diff -u -p -r1.7 c_ulimit.c
 --- c_ulimit.c	7 Jul 2004 19:20:09 -0000	1.7
 +++ c_ulimit.c	25 Sep 2005 03:13:29 -0000
 @@ -194,6 +194,8 @@ c_ulimit(wp)
  			    return 1;
  			}
  			val = rval * l->factor;
 +			if (val < 0)
 +			    val = -val;
  		}
  	}
  	if (all) {

 FWIW, the problem didn't show up in earlier NetBSD releases because the
 limit was lower (1/2 the limit in 3.99)

 --sjg

From: Frederick Bruckman <fredb@immanent.net>
To: Simon Gerraty <sjg@juniper.net>
Cc: gnats-bugs@netbsd.org
Subject: Re: bin/31377: ksh: cannot set limit indicated by ulimit -H
Date: Sun, 8 Oct 2006 19:05:06 -0500 (CDT)

 Hi,

 The problem seems to be that rlim_t is a "quad_t" (i.e., 64-bits), 
 while pdksh's evaluate() only works for "long" (which is, of course, 
 only 32-bits on i386). I suggest that the right fix is to "promote" 
 evaluate() to 64-bits, adjusting all of its callers, as below...

 Frederick


 Index: ./bin/ksh/c_ksh.c
 ===================================================================
 RCS file: /cvsroot/src/bin/ksh/c_ksh.c,v
 retrieving revision 1.15
 diff -u -r1.15 c_ksh.c
 --- ./bin/ksh/c_ksh.c	24 Apr 2006 20:00:31 -0000	1.15
 +++ ./bin/ksh/c_ksh.c	17 Sep 2006 19:39:01 -0000
 @@ -1083,7 +1083,7 @@
   	char **wp;
   {
   	int rv = 1;
 -	long val;
 +	int64_t val;

   	if (wp[1] == (char *) 0) /* at&t ksh does this */
   		bi_errorf("no arguments");
 Index: ./bin/ksh/c_sh.c
 ===================================================================
 RCS file: /cvsroot/src/bin/ksh/c_sh.c,v
 retrieving revision 1.12
 diff -u -r1.12 c_sh.c
 --- ./bin/ksh/c_sh.c	1 Apr 2006 23:39:58 -0000	1.12
 +++ ./bin/ksh/c_sh.c	17 Sep 2006 19:39:01 -0000
 @@ -32,7 +32,7 @@
   {
   	register struct block *l = e->loc;
   	register int n;
 -	long val;
 +	int64_t val;
   	char *arg;

   	if (ksh_getopt(wp, &builtin_opt, null) == '?')
 Index: ./bin/ksh/c_test.c
 ===================================================================
 RCS file: /cvsroot/src/bin/ksh/c_test.c,v
 retrieving revision 1.6
 diff -u -r1.6 c_test.c
 --- ./bin/ksh/c_test.c	26 Jun 2005 19:09:00 -0000	1.6
 +++ ./bin/ksh/c_test.c	17 Sep 2006 19:39:01 -0000
 @@ -369,7 +369,7 @@
   	  case TO_INTLE: /* -le */
   	  case TO_INTLT: /* -lt */
   		{
 -			long v1, v2;
 +			int64_t v1, v2;

   			if (!evaluate(opnd1, &v1, KSH_RETURN_ERROR)
   			    || !evaluate(opnd2, &v2, KSH_RETURN_ERROR))
 Index: ./bin/ksh/c_ulimit.c
 ===================================================================
 RCS file: /cvsroot/src/bin/ksh/c_ulimit.c,v
 retrieving revision 1.7
 diff -u -r1.7 c_ulimit.c
 --- ./bin/ksh/c_ulimit.c	7 Jul 2004 19:20:09 -0000	1.7
 +++ ./bin/ksh/c_ulimit.c	17 Sep 2006 19:39:01 -0000
 @@ -178,7 +178,7 @@
   		if (strcmp(wp[0], "unlimited") == 0)
   			val = KSH_RLIM_INFINITY;
   		else {
 -			long rval;
 +			rlim_t rval;

   			if (!evaluate(wp[0], &rval, KSH_RETURN_ERROR))
   				return 1;
 Index: ./bin/ksh/expr.c
 ===================================================================
 RCS file: /cvsroot/src/bin/ksh/expr.c,v
 retrieving revision 1.8
 diff -u -r1.8 expr.c
 --- ./bin/ksh/expr.c	26 Jun 2005 19:09:00 -0000	1.8
 +++ ./bin/ksh/expr.c	17 Sep 2006 19:39:01 -0000
 @@ -151,7 +151,7 @@
   int
   evaluate(expr, rval, error_ok)
   	const char *expr;
 -	long *rval;
 +	int64_t *rval;
   	int error_ok;
   {
   	struct tbl v;
 Index: ./bin/ksh/proto.h
 ===================================================================
 RCS file: /cvsroot/src/bin/ksh/proto.h,v
 retrieving revision 1.7
 diff -u -r1.7 proto.h
 --- ./bin/ksh/proto.h	26 Jun 2005 19:09:00 -0000	1.7
 +++ ./bin/ksh/proto.h	17 Sep 2006 19:39:01 -0000
 @@ -78,7 +78,7 @@
   int	pr_menu		ARGS((char *const *));
   int	pr_list		ARGS((char *const *));
   /* expr.c */
 -int 	evaluate	ARGS((const char *, long *, int));
 +int 	evaluate	ARGS((const char *, int64_t *, int));
   int	v_evaluate	ARGS((struct tbl *, const char *, volatile int));
   /* history.c */
   void	init_histvec	ARGS((void));
 Index: ./bin/ksh/var.c
 ===================================================================
 RCS file: /cvsroot/src/bin/ksh/var.c,v
 retrieving revision 1.14
 diff -u -r1.14 var.c
 --- ./bin/ksh/var.c	29 Mar 2006 15:51:00 -0000	1.14
 +++ ./bin/ksh/var.c	17 Sep 2006 19:39:02 -0000
 @@ -148,7 +148,7 @@
   	p = skip_varname(n, FALSE);
   	if (p != n && *p == '[' && (len = array_ref_len(p))) {
   		char *sub, *tmp;
 -		long rval;
 +		int64_t rval;

   		/* Calculate the value of the subscript */
   		*arrayp = TRUE;

From: Simon Gerraty <sjg@juniper.net>
To: Frederick Bruckman <fredb@immanent.net>
Cc: gnats-bugs@netbsd.org
Subject: Re: bin/31377: ksh: cannot set limit indicated by ulimit -H 
Date: Mon, 09 Oct 2006 15:39:03 -0700

 Sounds reasonable.

 Thanks
 --sjg

 On Sun, 8 Oct 2006 19:05:06 -0500 (CDT), Frederick Bruckman wri
 tes:
 >Hi,
 >
 >The problem seems to be that rlim_t is a "quad_t" (i.e., 64-bi
 >ts), 
 >while pdksh's evaluate() only works for "long" (which is, of c
 >ourse, 
 >only 32-bits on i386). I suggest that the right fix is to "pro
 >mote" 
 >evaluate() to 64-bits, adjusting all of its callers, as below.
 >..
 >

>Unformatted:

Home	PR Database Search