The story of the Linux kernel 3.x…

In 2005 everybody was exited about possibility of bypass ASLR on all Linux 2.6 kernels because of the new concept called VDSO (Virtual Dynamic Shared Object). More information about this story can be found at the following link:

http://www.trilithium.com/johan/2005/08/linux-gate/

 

In short, VDSO was mmap’ed by the kernel in the user space memory always at the same fixed address. Because of that well-known technique ret-to-libc (or as some ppl prefer ROP) was possible and effective to bypass existing security mitigation in the system.

… 6 years later Linus Torvalds announced the release of the new kernel version – 3.x! Now, guess what happened…

pi3-darkstar new # uname -r
3.2.12-gentoo
pi3-darkstar new # cat /proc/sys/kernel/randomize_va_space
2
pi3-darkstar new # cat /proc/self/maps|tail -2
bfa81000-bfaa2000 rw-p 00000000 00:00 0          [stack]
ffffe000-fffff000 r-xp 00000000 00:00 0          [vdso]
pi3-darkstar new # cat /proc/self/maps|tail -2
bfd5e000-bfd7f000 rw-p 00000000 00:00 0          [stack]
ffffe000-fffff000 r-xp 00000000 00:00 0          [vdso]
pi3-darkstar new # ldd /bin/ls|head -1
    linux-gate.so.1 =>  (0xffffe000)
pi3-darkstar new # ldd /bin/ls|head -1
    linux-gate.so.1 =>  (0xffffe000)
pi3-darkstar new #

 

I’m not using
dd if=/proc/self/mem of=linux-gate.dso bs=4096 skip=1048574 count=1
because I’m lame 🙂

pi3-darkstar new # echo "main(){}">dupa.c
pi3-darkstar new # gcc dupa.c -o dupa
pi3-darkstar new # gdb -q ./dupa
Reading symbols from /root/priv/projekty/pro-police/new/dupa...(no debugging symbols found)...done.
(gdb) b main
Breakpoint 1 at 0x80483b7
(gdb) r
Starting program: /root/priv/projekty/pro-police/new/dupa 

Breakpoint 1, 0x080483b7 in main ()
(gdb) dump binary memory test_dump.bin 0xffffe000 0xfffff000
(gdb) quit
A debugging session is active.

    Inferior 1 [process 20117] will be killed.

Quit anyway? (y or n) y
pi3-darkstar new # file test_dump.bin
test_dump.bin: ELF 32-bit LSB shared object, Intel 80386, version 1 (SYSV), dynamically linked, stripped
pi3-darkstar new # objdump -T ./test_dump.bin 

./test_dump.bin:     file format elf32-i386

DYNAMIC SYMBOL TABLE:
ffffe414 g    DF .text    00000014  LINUX_2.5   __kernel_vsyscall
00000000 g    DO *ABS*    00000000  LINUX_2.5   LINUX_2.5
ffffe40c g    DF .text    00000008  LINUX_2.5   __kernel_rt_sigreturn
ffffe400 g    DF .text    00000009  LINUX_2.5   __kernel_sigreturn

pi3-darkstar new # readelf -h ./test_dump.bin
ELF Header:
  Magic:   7f 45 4c 46 01 01 01 00 00 00 00 00 00 00 00 00
  Class:                             ELF32
  Data:                              2's complement, little endian
  Version:                           1 (current)
  OS/ABI:                            UNIX - System V
  ABI Version:                       0
  Type:                              DYN (Shared object file)
  Machine:                           Intel 80386
  Version:                           0x1
  Entry point address:               0xffffe414
  Start of program headers:          52 (bytes into file)
  Start of section headers:          1172 (bytes into file)
  Flags:                             0x0
  Size of this header:               52 (bytes)
  Size of program headers:           32 (bytes)
  Number of program headers:         4
  Size of section headers:           40 (bytes)
  Number of section headers:         12
pi3-darkstar new # objdump -f ./test_dump.bin 

./test_dump.bin:     file format elf32-i386
architecture: i386, flags 0x00000150:
HAS_SYMS, DYNAMIC, D_PAGED
start address 0xffffe414
^^^^^^^^^^^^^^^^^^^^^^^^

pi3-darkstar new # objdump -d --start-address=0xffffe414 ./test_dump.bin 

./test_dump.bin:     file format elf32-i386

Disassembly of section .text:

ffffe414 <__kernel_vsyscall>:
ffffe414:    51                       push   %ecx
ffffe415:    52                       push   %edx
ffffe416:    55                       push   %ebp
ffffe417:    89 e5                    mov    %esp,%ebp
ffffe419:    0f 34                    sysenter
ffffe41b:    90                       nop
ffffe41c:    90                       nop
ffffe41d:    90                       nop
ffffe41e:    90                       nop
ffffe41f:    90                       nop
ffffe420:    90                       nop
ffffe421:    90                       nop
ffffe422:    cd 80                    int    $0x80
<--------------------- Nice oldschool pop-ret :) ---------------------->
ffffe424:    5d                       pop    %ebp
ffffe425:    5a                       pop    %edx
ffffe426:    59                       pop    %ecx
ffffe427:    c3                       ret    
pi3-darkstar new #

If you look at the process memory layout and analyse every bytes from this address range you can find some useful instruction not only that which I listed in this lame write-up.

Btw. I wonder why no-one point this out before…
Btw2. Go and write reliable exploit for kernel 3.x ;p

 

 

[UPDATE]

Because my write-up wasn’t so clear this section need to be done. Problem is not in kernel 3.x by itself but in the configuration. If COMPAT_COMPAT_VDSO option was used for kernel then problem appears. Whole problem was discussed based on OpenSuse 12.1 system which enables this option by default. Nicolas Surribas in Full Disclosure list pointed out that in his case problem does not exists! After reading opensuse kernel developers list I found a problem and gentle fix:

http://lists.opensuse.org/opensuse-kernel/2012-03/msg00056.html

What about 64 bits Fedora and Ubuntu? They have fixed address range for VSYSCALL which after discussion with bliss it became as known issue: https://lkml.org/lkml/2011/8/9/274 and I didn’t know about that – my fault.

 

Summarizing:

OpenSuse 12.1 by default has this problem but latest kernel update fix it.

All 64 bits distros has VSYSCALL mmaped at fixed address range but this is known issue.

 

Thanks for everyone who was involved in this issue 😉

 

Best regards,
Adam Zabrocki