Jonathan Brossard CTO - P1 Code Security
Agenda
Virtualization : big picture
Attack surface analysis
The need for new tools
Introducing Virtual 8086 mode
•Practical fuzzing with vm86()
Virtualization : time to care !
Market shares Definitions
Virtualization : market shares
Source : Forrester Research 2009
78% of companies have production servers virtualized.
20% only have virtualized servers.
Virtualization : market shares
Source : Forrester Research 2009
VMWare is present in 98% of the companies.
Microsoft virtualization products are used by 17%.
Citrix/Xen is used by 10%.
Bottom line...
Virtualization software are so widespread that they have become more attractive targets than say web, mail or dns servers !
There is a lower variety too !
Definitions
Virtualization : Definitions
Virtualization
Virtualization is the name given to the simulation with higher level components, of lower level components.
NOTE: Virtualization of applications (as opposed to full Oses) is out of topic.
Virtualization : Definitions
Virtual Machine
A virtual machine (VM) is : "an efficient, isolated duplicate of a real machine".
--Gerald J. Popek and Robert P. Goldberg (1974). "Formal Requirements for Virtualizable Third Generation Architectures", Communications of the ACM.
Paravirtualization
Virtualization : Definitions
Paravirtualization
Requires the modification of the guest Oses (eg: Xen, UML, Qemu with kquemu, VMWare Workstation with VMWare Tools).
Opposed to « full virtualization ».
Virtualization : Definitions
There are two types of virtualizations : Virtual Machine Monitors (or Hypervisors) of type I and type II.
Type I Hypervisor
Virtualization : Definitions
Hypervisors of type I
Run on bare metal (eg: Xen,
Type II hypervisor
Virtualization : Definitions
Hypervizors of type II
Run as a process inside a host OS to virtualize guests Oses (eg: Qemu, Virtualbox, VMWare Workstation, Parallels).
Hardware assisted
virtualization
Hardware assisted
virtualization
-Takes advantage of
-x64 Only.
-The hypervizor is running in « ring
-Much like the NX bit : requires the motherboard to support it and activation in the BIOS.
Virtualization : Definitions
Isolation
Isolation of the userland part of the OS to simulate independant machines (eg:
Isolation
Attack surface analysis
Depending on your perspective...
What are the risks ?
Where to attack ?
Privilege escalation on the
host
VMware Tools HGFS Local Privilege Escalation Vulnerability
(http://labs.idefense.com/intelligence/vu
lnerabilities/display.php?id=712)
Privilege escalation on the
Guest
Attacking other guests
Vmare workstation guest isolation weaknesses (clipboard transfer)
http://www.securiteam.com/securitynew
s/5GP021FKKO.html
DoS (Host + Guests)
Escape to host
Rafal Wojtczuk (Invisible things, BHUS 2008)
IDEFENSE VMware Workstation Shared Folders Directory Traversal Vulnerability
Attack surface analysis :
usage
Hosting two companies on the same hardware is very common (shared hosting).
Getting a shell on the same machine as a given target may therefor be a matter of paying a few euros a month.
Attack surface : conclusion
Owning the Host OS from the Guest is practical : security through virtualization is a failure.
Seemingly minor bugs (local, DoS) do matter : virtualization amplifies consequences.
The need for dedicated methodologies and tools
The need for new tools :
example
How to dynamically test a virtual Hard Drive ?
How to dynamically test a virtual Hard Drive ? Naive approach
Standard API :
ssize_t read(int fd, void *buf, size_t count); ssize_t write(int fd, const void *buf, size_t
count);
This would mostly fuzz the kernel, not the Virtual Machine :(
We need something (much) lower level.
Standard (low level) attack
vectors
Ioports:
outb, outw, outl, outsb, outsw, outsl, inb, inw, inl, insb, insw, insl, outb_p, outw_p, outl_p, inb_p, inw_p, inl_p Problems: sequence, multiple ports
Ioctls:
int ioctl(int d, int request, ...) Problems : arbitrary input size !
How did we used to do it
« back in the days » ?
MS Dos : direct access to the hardware (interrupts : BIOS, HD, Display, …)
Can we get back to this ?
Introducing the Virtual 8086 mode
Introducing the Virtual 8086 mode
Introduced with Intel 386 (1985)
Introducing the Virtual 8086 mode
Intel x86 cpus support 3 modes
-Protected mode
-Real mode
-System Management Mode (SMM)
Introducing the
Virtual 8086 mode
Protected mode
This mode is the native state of the processor. Among the capabilities of protected mode is the ability to directly execute
Introducing the
Virtual 8086 mode
This mode implements the programming environment of the Intel 8086 processor with extensions (such as the ability to switch to protected or system management mode). The processor is placed in
Introducing the
Virtual 8086 mode
System management mode (SMM)
This mode provides an operating system or executive with a transparent mechanism for implementing platform specific functions such as power management and system security. The processor enters SMM when the external SMM interrupt pin (SMI#) is activated or an SMI is received from the advanced programmable interrupt controller (APIC).
Nice things about Real
mode / Virtual 8086 mode
Direct access to hardware via interruptions !
example:
Mov ah, 0x42 ; read sector from drive Mov ch, 0x01 ; Track
Mov cl, 0x02 ; Sector Mov dh, 0x03 ; Head
Mov dl, 0x80 ; Drive (here first HD) Mov bx, offset buff ; es:bx is destination
Int 0x13 ; hard disk operation
Complexity
ax*bx*cx*dx (per interruption)
Id est: [0;65535]^4 ~ 1.8 * 10^19
=> still huge
=> much better than ioctl()'s arbitrary input length !
Introducing the Virtual 8086 mode
Problem is... is this even possible inside a virtual machine ?
Introducing the
Virtual 8086 mode
A closer look at the boot sequence...
Introducing the
Virtual 8086 mode
The kernel boots in (16b) real mode, and then switches to protected mode (32b).
The cpu normally doesn't get back to
real mode untill next reboot.
Introducing the
Virtual 8086 mode
Corollary
The hypervisor could run under any mode. protected mode in practice (being it ring0, ring1 or ring3).
All of the guests run only in protected
mode.
Now how to swith to Virtual 8086
mode ? It this even possible ?
Leaving protected mode ?
(Ascii Art : Courtesy of phrack 65)
Setting the VM flag in CR0 under protected mode would get us to Virtual Mode
Removing the PE flag from CR0 would get us back to real mode
Leaving protected mode ?
static const unsigned char real_mode_switch [] =
{
0x66, 0x0f, 0x20, 0xc0, |
/* |
movl |
%cr0,%eax |
*/ |
|
||
0x66, 0x83, 0xe0, 0x11, |
/* |
andl $0x00000011,%eax */ |
|||||
0x66, 0x0d, 0x00, 0x00, 0x00, 0x60, |
|
/* |
orl |
$0x60000000,%eax |
|||
*/ |
/* |
movl |
%eax,%cr0 |
*/ |
|
||
0x66, 0x0f, 0x22, 0xc0, |
|
||||||
0x66, 0x0f, 0x22, 0xd8, |
/* |
movl |
%eax,%cr3 |
*/ |
|
||
0x66, 0x0f, 0x20, 0xc3, |
/* |
movl |
%cr0,%ebx |
*/ |
|
||
0x66, 0x81, 0xe3, 0x00, 0x00, 0x00, 0x60, |
/* |
andl |
|
|
|||
$0x60000000,%ebx */ |
/* |
jz |
f |
|
*/ |
|
|
0x74, 0x02, |
|
*/ |
|
||||
0x0f, 0x09, |
/* |
wbinvd |
|
|
|||
0x24, 0x10, |
/* f: andb |
$0x10,al |
*/ |
*/ |
|||
0x66, 0x0f, 0x22, 0xc0 |
/* |
movl %eax,%cr0 |
|||||
};
Trouble is...
This obviously won't work inside a virtual machine !
Because
IS THIS « GAME OVER » ?
Actually not quite ...
Truth is : we don't need to
switch back to real
mode/virtual 8086 mode !
Most Operating systems offer a way to run 16b applications (eg: MS DOS) under protected mode by emulating a switch to Virtual 8086 Mode.
Notably Windows (x86) and Linux (x86).
The Windows case
NTVDM : ntvdm.exe
« Windows 16b Virtual Machine »
The Linux case
The linux kernel provides an emulation of real mode in the form of two syscalls:
#define __NR_vm86old |
113 |
#define __NR_vm86 |
166 |
The Linux case
#include <sys/vm86.h>
int vm86old(struct vm86_struct *info);
int vm86(unsigned long fn, struct vm86plus_struct *v86);
struct vm86_struct {
struct vm86_regs regs;
unsigned long flags; unsigned long screen_bitmap; unsigned long cpu_type; struct revectored_struct
int_revectored; struct revectored_struct int21_revectored;
};
The Linux case
struct vm86_regs { long ebx; long ecx; long edx; long esi; long edi; long ebp; long eax;
(…)
unsigned short es, __esh; unsigned short ds, __dsh; unsigned short fs, __fsh; unsigned short gs, __gsh;
};
In a nutshell
-The switch to Virtual mode is entirely emulated by the kernel (this will work inside a VM)
-We can still program using old school interruptions (easy !)
-Those interruptions are delivered to the hardware (id est: either the emulated one, or the real one).
=> We just got a « bare metal (possibly virtualized) hardware interface »
The x64 case...
The x64 case
X64 cpus in 64b long mode can't swith to Virtual mode.
That's too bad : we'd like to fuzz latest Vmware ESX or Microsoft HyperV (necessarily under x64).
But under virtualization, the switch to VM86 mode is being
emulated by the kernel...
The x64 case
Using kernel patches, we can add VM86 capabilities to a x64 GNU/Linux kernel.
EG:
What's not possible in real hardware becomes possible under a virtualized environment !
Practical use : Fuzzing
using vm86()
Practical use : Fuzzing
using vm86()
Looking at the IVT allows us to fuzz all the hardware know after BIOS Post, efficently (no calls to empty/dummy interrupts).
Practical use : Fuzzing
using vm86()
Exemple bugs !
Practical use : Fuzzing
using vm86()
Bugs in hypervizors...
Virtualbox
00:21:13.603 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
00:21:13.603 !!
00:21:13.603 !! Guru Meditation
00:21:13.603 !! TRAP=0e ERRCD=0000000000000000 CR2=00000000000c0000 EIP=ff215e33 Type=0
00:21:13.603 !! EIP in VMMGC.gc (ff1dd000) at rva 38e33 near symbols: 00:21:13.603 !! ff215df0 rva 00038df0 off 00000043
_ZL10disCoreOneP12_DISCPUSTATEyPj
00:21:13.603 !! ff216040 rva 00039040 off
00:21:13.603 !! fff8:ff215e33 0f b6 10 |
movzx edx, byte [eax] |
00:21:13.603 !! |
|
00:21:13.603 !! |
|
00:21:13.603 !! |
|
00:21:13.603 Hypervisor CPUM state: se |
|
00:21:13.603 .eax=000c0000 .ebx=fed69cfc .ecx=00000000 .edx=00000000
.esi=00000001 .edi=fec01000
Virtualbox (take 2)
00:02:51.129 !!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
00:02:51.129 !!
00:02:51.129 !! Guru Meditation
00:02:51.129 !! TRAP=0e ERRCD=0000000000000000 CR2=00000000000ab000 EIP=ff215e33 Type=0
00:02:51.129 !! EIP in VMMGC.gc (ff1dd000) at rva 38e33 near symbols: 00:02:51.129 !! ff215df0 rva 00038df0 off 00000043
_ZL10disCoreOneP12_DISCPUSTATEyPj
00:02:51.129 !! ff216040 rva 00039040 off
00:02:51.129 !! fff8:ff215e33 0f b6 10 movzx edx, byte [eax] 00:02:51.129 !!
00:02:51.129 !! 00:02:51.129 !!
00:02:51.129 Hypervisor CPUM state: se
00:02:51.129 .eax=000ab000 .ebx=fed69cfc .ecx=00000000 .edx=00000000
.esi=00000001 .edi=fec01000
00:02:51.129 .eip=ff215e33 .esp=fed69c7c .ebp=fed69cc4 .iopl=0 rf nv up di nt zr ac pe cy
More (guest) bugs
Virtual PC
Parallels (Guest)
Inhibit Mask=0
CS=FF63 [0000FFFF 0000F30F] V=1 SS=FFD3 [0000FFFF 00CF9300] V=1 DS=0018 [0000FFFF 00CFF300] V=1 ES=0018 [0000FFFF 00CFF300] V=1 FS=FF9B [0000FFFF 00CF9300] V=1 GS=0018 [0000FFFF 00CF9300] V=1
EAX=000000A9 EBX=00005148 ECX=0000F686 EDX=0000000B
ESI=00002D72 EDI=000007E4 EBP=00002E99 ESP=00000FFA
EIP=0000FE96 EFLAGS=00023202
What about x64 ?
Attacking Microsoft HyperV
DEMOS
DEMO
Adding layers of virtualization is actually a bad idea : the only way is to secure the software is to properly test it for security bugs...
Thank you for coming
Questions ?