NetBSD Problem Report #18222

Received: (qmail 24174 invoked by uid 605); 7 Sep 2002 18:03:09 -0000
Message-Id: <20020907180306.3D3FCAC@proven.weird.com>
Date: Sat,  7 Sep 2002 14:03:06 -0400 (EDT)
From: woods@weird.com (Greg A. Woods)
Sender: gnats-bugs-owner@netbsd.org
Reply-To: woods@planix.com (Greg A. Woods)
To: gnats-bugs@gnats.netbsd.org
Subject: patch to allow a uid or gid of (-2) matching default NFS mapping for remote root users
X-Send-Pr-Version: 3.95

>Number:         18222
>Category:       bin
>Synopsis:       patch to allow a uid or gid of (-2) matching default NFS mapping for remote root users
>Confidential:   no
>Severity:       serious
>Priority:       medium
>Responsible:    bin-bug-people
>State:          open
>Class:          sw-bug
>Submitter-Id:   net
>Arrival-Date:   Sat Sep 07 18:04:01 +0000 2002
>Closed-Date:    
>Last-Modified:  Sat Sep 07 21:22:00 +0000 2002
>Originator:     Greg A. Woods
>Release:        2002/09/06
>Organization:
Planix, Inc.; Toronto, Ontario; Canada
>Environment:
System: NetBSD
>Description:

	Mountd(8) uses the vaule '-2' as the default "maroot" user-ID
	and group-ID (i.e. it is the UID and GID that root users from
	remote clients are mapped into when they access files on the NFS
	server).  Unfortunately due to checks in pw_scan(3) it's
	impossible for the NFS server administrator to add a user with
	this user-ID to the local password file (eg. so that files
	created by remote NFS client root users will show up with a
	meaningful local username).

	This is due to the current artificial ceiling on the values of
	UID_MAX and GID_MAX, used by the checks imposed in pw-scan(3).

	Given that uid_t and gid_t are defined as u_int32_t I believe
	the only integer value that is actually invalid for a user-ID or
	group-ID is UINT_MAX (i.e. -1 when converted to an int32_t).

	(even -1 may not be strictly necessary to reserve, but it's used
	as an error value in so many places that not reserving it is
	only an invitation to problems)

>How-To-Repeat:

	Attempt to add a line like the following to /etc/master.passwd
	using vipw(8):

	nfsnbody:*:4294967294:4294967294::0:0:NFS unprivileged (default maproot) user:/nonexistant:/sbin/nologin

>Fix:

	increase UID_MAX and GID_MAX to (UINT_MAX-1)

Index: syslimits.h
===================================================================
RCS file: /cvs/NetBSD/src/sys/sys/syslimits.h,v
retrieving revision 1.1.1.7
diff -c -r1.1.1.7 syslimits.h
*** syslimits.h	13 Jun 2001 07:04:40 -0000	1.1.1.7
--- syslimits.h	7 Sep 2002 17:35:12 -0000
***************
*** 43,55 ****
  #if !defined(_ANSI_SOURCE)
  #define	ARG_MAX		 (256 * 1024)	/* max bytes for an exec function */
  #define	CHILD_MAX		  160	/* max simultaneous processes */
! #define	GID_MAX		   2147483647U	/* max value for a gid_t (2^31-2) */
  #define	LINK_MAX		32767	/* max file link count */
  #define	MAX_CANON		  255	/* max bytes in term canon input line */
  #define	MAX_INPUT		  255	/* max bytes in terminal input */
  #define	NAME_MAX		  255	/* max bytes in a file name */
  #define	NGROUPS_MAX		   16	/* max supplemental group id's */
! #define	UID_MAX		   2147483647U	/* max value for a uid_t (2^31-2) */
  #define	OPEN_MAX		   64	/* max open files per process */
  #define	PATH_MAX		 1024	/* max bytes in pathname */
  #define	PIPE_BUF		  512	/* max bytes for atomic pipe writes */
--- 43,55 ----
  #if !defined(_ANSI_SOURCE)
  #define	ARG_MAX		 (256 * 1024)	/* max bytes for an exec function */
  #define	CHILD_MAX		  160	/* max simultaneous processes */
! #define	GID_MAX		 (UINT_MAX-1)	/* max value for a gid_t (2^32-1) */
  #define	LINK_MAX		32767	/* max file link count */
  #define	MAX_CANON		  255	/* max bytes in term canon input line */
  #define	MAX_INPUT		  255	/* max bytes in terminal input */
  #define	NAME_MAX		  255	/* max bytes in a file name */
  #define	NGROUPS_MAX		   16	/* max supplemental group id's */
! #define	UID_MAX		 (UINT_MAX-1)	/* max value for a uid_t (2^32-1) */
  #define	OPEN_MAX		   64	/* max open files per process */
  #define	PATH_MAX		 1024	/* max bytes in pathname */
  #define	PIPE_BUF		  512	/* max bytes for atomic pipe writes */
>Release-Note:
>Audit-Trail:

From: David Laight <david@l8s.co.uk>
To: netbsd-bugs@netbsd.org
Cc: gnats-bugs@gnats.netbsd.org
Subject: Re: bin/18222: patch to allow a uid or gid of (-2) matching default NFS mapping for remote root users
Date: Sat, 7 Sep 2002 19:23:10 +0100

 > >Synopsis:       patch to allow a uid or gid of (-2) matching default
 >		   NFS mapping for remote root users
 > 
 > 	increase UID_MAX and GID_MAX to (UINT_MAX-1)

 Wouldn't it be better to use (~(uid_t)0 - 1) and (~(gid_t)0 - 1) ?

 	David

 -- 
 David Laight: david@l8s.co.uk

From: David Laight <david@l8s.co.uk>
To: netbsd-bugs@netbsd.org
Cc: gnats-bugs@gnats.netbsd.org
Subject: Re: bin/18222: patch to allow a uid or gid of (-2) matching default NFS mapping for remote root users
Date: Sat, 7 Sep 2002 19:44:09 +0100

 On Sat, Sep 07, 2002 at 07:23:10PM +0100, David Laight wrote:
 > > >Synopsis:       patch to allow a uid or gid of (-2) matching default
 > >		   NFS mapping for remote root users
 > > 
 > > 	increase UID_MAX and GID_MAX to (UINT_MAX-1)
 > 
 > Wouldn't it be better to use (~(uid_t)0 - 1) and (~(gid_t)0 - 1) ?

 A quick look at the history of syslimits.h (brought on by the fact
 that the comments for these values didn't match the value) shows
 that the limit has been 2^32-2 before.

 It was changed because setreu/gid() allows -1 (meaning don't
 change) which has to be within the domain of u/gid_t.

 This is another of those types [1] whose domain is -1..MAXINT-1
 that C doesn't quite copy with.

 	David

 [1] includes pid_t and the return value from lseek(), probably
 really size_t in genaral.

 -- 
 David Laight: david@l8s.co.uk

From: woods@weird.com (Greg A. Woods)
To: David Laight <david@l8s.co.uk>
Cc: netbsd-bugs@netbsd.org, gnats-bugs@gnats.netbsd.org
Subject: Re: bin/18222: patch to allow a uid or gid of (-2) matching default NFS mapping for remote root users
Date: Sat,  7 Sep 2002 15:51:08 -0400 (EDT)

 [ On Saturday, September 7, 2002 at 19:23:10 (+0100), David Laight wrote: ]
 > Subject: Re: bin/18222: patch to allow a uid or gid of (-2) matching default NFS mapping for remote root users
 >
 > > >Synopsis:       patch to allow a uid or gid of (-2) matching default
 > >		   NFS mapping for remote root users
 > > 
 > > 	increase UID_MAX and GID_MAX to (UINT_MAX-1)
 > 
 > Wouldn't it be better to use (~(uid_t)0 - 1) and (~(gid_t)0 - 1) ?

 Yes, I suppose it would!  Thanks!

 -- 
 								Greg A. Woods

 +1 416 218-0098;            <g.a.woods@ieee.org>;           <woods@robohack.ca>
 Planix, Inc. <woods@planix.com>; VE3TCP; Secrets of the Weird <woods@weird.com>

From: woods@weird.com (Greg A. Woods)
To: netbsd-bugs@netbsd.org, gnats-bugs@gnats.netbsd.org
Cc:  
Subject: Re: bin/18222: patch to allow a uid or gid of (-2) matching default NFS mapping for remote root users
Date: Sat,  7 Sep 2002 17:01:00 -0400 (EDT)

 [ On Saturday, September 7, 2002 at 19:44:09 (+0100), David Laight wrote: ]
 > Subject: Re: bin/18222: patch to allow a uid or gid of (-2) matching default NFS mapping for remote root users
 >
 > On Sat, Sep 07, 2002 at 07:23:10PM +0100, David Laight wrote:
 > > > >Synopsis:       patch to allow a uid or gid of (-2) matching default
 > > >		   NFS mapping for remote root users
 > > > 
 > > > 	increase UID_MAX and GID_MAX to (UINT_MAX-1)
 > > 
 > > Wouldn't it be better to use (~(uid_t)0 - 1) and (~(gid_t)0 - 1) ?
 > 
 > A quick look at the history of syslimits.h (brought on by the fact
 > that the comments for these values didn't match the value) shows
 > that the limit has been 2^32-2 before.
 > 
 > It was changed because setreu/gid() allows -1 (meaning don't
 > change) which has to be within the domain of u/gid_t.

 Yes, I remember that -- but the change was ultra-conservative and
 results in my "nfsanon" user being "invalid".  My PR is in part in
 (delayed) response to that change.

 2^32-1 is the correct limit to allow for -2 as a valid value and -1 as a
 special meaning to things like setreuid() [which I don't really care
 about anywas as I've disabled it on my own systems for security reasons ;-)].

 > This is another of those types [1] whose domain is -1..MAXINT-1
 > that C doesn't quite copy with.

 It's not C that's at fault -- it's bad/lazy API design.  :-)

 -- 
 								Greg A. Woods

 +1 416 218-0098;            <g.a.woods@ieee.org>;           <woods@robohack.ca>
 Planix, Inc. <woods@planix.com>; VE3TCP; Secrets of the Weird <woods@weird.com>

From: David Laight <david@l8s.co.uk>
To: netbsd-bugs@netbsd.org, gnats-bugs@gnats.netbsd.org
Cc:  
Subject: Re: bin/18222: patch to allow a uid or gid of (-2) matching default NFS mapping for remote root users
Date: Sat, 7 Sep 2002 22:24:38 +0100

 I've looked at the posix/ieee definitions of uid_t, gid_t and pid_t on
 http://www.opengroup.org/onlinepubs/007904975/basedefs/sys/types.h.html
 uid_t and gid_t are defined to be unsigned, and pid_t signed.
 setreuid() is there, and requires -1 to be valid for an unsigned
 parameter. C's 'Integral Convertions' DTRT (by K&R2) [1] but I bet
 gcc will give a warning.

 NetBSD sys/ansi.h seems to have an unsigned pid_t....

 	David

 [1] by definition (unsigned long long)(signed char)-1 == ~0ull

 -- 
 David Laight: david@l8s.co.uk
>Unformatted:

Home	PR Database Search