NetBSD Problem Report #7761

Received: (qmail 20508 invoked from network); 11 Jun 1999 17:02:07 -0000
Message-Id: <199906111702.TAA11298@calvin.augustsson.net>
Date: Fri, 11 Jun 1999 19:02:05 +0200 (CEST)
From: Lennart Augustsson <augustss@augustsson.net>
Reply-To: augustss@augustsson.net
To: gnats-bugs@gnats.netbsd.org
Subject: getcwd() fails mysteriously
X-Send-Pr-Version: 3.95

>Number:         7761
>Category:       kern
>Synopsis:       getcwd() fails mysteriously
>Confidential:   no
>Severity:       serious
>Priority:       low
>Responsible:    sommerfeld
>State:          closed
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Fri Jun 11 10:05:01 +0000 1999
>Closed-Date:    Sat Jun 19 18:02:15 +0000 1999
>Last-Modified:  Sat Jun 19 18:02:55 +0000 1999
>Originator:     Lennart Augustsson
>Release:        NetBSD 1.4
>Organization:
>Environment:
System: NetBSD calvin 1.4 NetBSD 1.4 (CALVIN) #0: Sun May 9 18:41:36 CEST 1999 augustss@calvin:/usr/src/sys/arch/i386/compile/CALVIN i386


>Description:
	If the size argument to getcwd() is large enough the call fails
	with ERANGE.  From what I can tell from the man page this is
	wrong, and totally unintuitive.
	Furthermore, it used to work before getcwd() went into the kernel.
>How-To-Repeat:
	Run the following program.  With a smaller b (e.g. 1000) it works.

#include <stdio.h>
#include <unistd.h>
#include <err.h>

char b[10000];

main()
{
	char *r;
	r = getcwd(b, sizeof b);
	if (r == 0)
		err(1, 0);
	printf("path = '%s'\n", b);
}

>Fix:
	Dunno.
>Release-Note:
>Audit-Trail:

From: Jaromir Dolecek <dolecek@ics.muni.cz>
To: augustss@augustsson.net
Cc: gnats-bugs@gnats.netbsd.org
Subject: Re: kern/7761: getcwd() fails mysteriously
Date: Sat, 12 Jun 1999 17:00:21 +0200 (MEST)

 Lennart Augustsson wrote:
 > 	Run the following program.  With a smaller b (e.g. 1000) it works.

 Well, according to vfs_getcwd.c, the call unconditionally returns
 ERANGE when (len > 4*MAXPATHLEN). There seems to be a another bug in
 the code -- sys___getcwd() uses only half of the specified length
 (vfs_getcwd.c:getcwd_common() is called with limit == len/2).

 Seems like algorithm of sys___getcwd() should be changed to:

 1. allocate buffer with size min(len, MAXPATHLEN)
 2. find the cwd; if the buffer size is not big enough, return ERANGE
 3. otherwise, copyout the cwd into supplied buffer.

 Also, I'm not sure why getcwd_common() is called with limit == len/2.

 Bill Sommerfeld should be the definitive authority about it.

 -- 
 Jaromir Dolecek <dolecek@ics.muni.cz>      http://www.ics.muni.cz/~dolecek/
 "The only way how to get rid temptation is to yield to it." -- Oscar Wilde
Responsible-Changed-From-To: kern-bug-people->sommerfeld 
Responsible-Changed-By: sommerfeld 
Responsible-Changed-When: Sun Jun 13 11:49:16 PDT 1999 
Responsible-Changed-Why:  
it's my bug.. 

From: Bill Sommerfeld <sommerfeld@orchard.arlington.ma.us>
To: augustss@augustsson.net, dolecek@ics.muni.cz
Cc: gnats-bugs@netbsd.org
Subject: Re: kern/7761 
Date: Sun, 13 Jun 1999 17:39:43 -0400

 A few clarifications:

  1) Jaromir's analysis is faulty.  The getcwd_common() function takes
 two limits: buffer length, and number of directories to traverse
 before giving up (for loop detection..).

 Each entryname takes up at least 2 characters in the buffer: one for
 the name, one for the '/' separator, so an N byte buffer can hold at
 most N/2 directory levels.

 The full buffer passed in is usable, as a run of the regression test
 code in regress/sys/kern/getcwd would clearly show.

  2) pathnames longer than MAXPATHLEN are supported for compatibility
 with the old getcwd() routine.  I suggest changing either the kernel
 or userspace to (silently) clip the passed in buffer length to
 MAXPATHLEN*4 rather than rewriting it to copy out a component at time
 and then returning an offset or some such to the start of the
 pathname.

 I probably won't be able to deal with this for the next week.  if
 someone else codes a fix in the meantime, send a patch my way and i'll
 take a look at it..

 					- Bill
State-Changed-From-To: open->closed 
State-Changed-By: sommerfeld 
State-Changed-When: Sat Jun 19 11:02:15 PDT 1999 
State-Changed-Why:  
Fixed; kernel now silently rounds down length to MAXPATHLEN*4.. 
>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.