Analyze CVE-2022-32792: Faster Than Light, but at What Cost?

Analyze CVE-2022-32792: Faster Than Light, but at What Cost?

Introduction

Before we jump right into the vulnerability, let’s review the patch of it first.

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
From 6983e76741a1bad811783ceac0959ff9953c175d Mon Sep 17 00:00:00 2001
From: Mark Lam <[email protected]>
Date: Fri, 20 May 2022 18:33:04 +0000
Subject: [PATCH] Refine B3ReduceStrength's range for sign extension
operations. https://bugs.webkit.org/show_bug.cgi?id=240720
<rdar://problem/93536782>

Reviewed by Yusuke Suzuki and Keith Miller.

* Source/JavaScriptCore/b3/B3ReduceStrength.cpp:

Canonical link: https://commits.webkit.org/250808@main
git-svn-id: https://svn.webkit.org/repository/webkit/trunk@294563 268f45cc-cd09-0410-ab3c-d52691b4dbfc
---
Source/JavaScriptCore/b3/B3ReduceStrength.cpp | 61 ++++++++++++++++++-
1 file changed, 59 insertions(+), 2 deletions(-)

diff --git a/Source/JavaScriptCore/b3/B3ReduceStrength.cpp b/Source/JavaScriptCore/b3/B3ReduceStrength.cpp
index f30a68587876..32bcf3d81415 100644
--- a/Source/JavaScriptCore/b3/B3ReduceStrength.cpp
+++ b/Source/JavaScriptCore/b3/B3ReduceStrength.cpp
@@ -1,5 +1,5 @@
/*
- * Copyright (C) 2015-2020 Apple Inc. All rights reserved.
+ * Copyright (C) 2015-2022 Apple Inc. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
@@ -388,6 +388,61 @@ class IntRange {
}
}

+ template<typename T>
+ IntRange sExt()
+ {
+ ASSERT(m_min >= INT32_MIN);
+ ASSERT(m_max <= INT32_MAX);
+ int64_t typeMin = std::numeric_limits<T>::min();
+ int64_t typeMax = std::numeric_limits<T>::max();
+ auto min = m_min;
+ auto max = m_max;
+
+ if (typeMin <= min && min <= typeMax
+ && typeMin <= max && max <= typeMax)
+ return IntRange(min, max);
+
+ // Given type T with N bits, signed extension will turn bit N-1 as
+ // a sign bit. If bits N-1 upwards are identical for both min and max,
+ // then we're guaranteed that even after the sign extension, min and
+ // max will still be in increasing order.
+ //
+ // For example, when T is int8_t, the space of numbers from highest to
+ // lowest are as follows (in binary bits):
+ //
+ // highest 0 111 1111 ^
+ // ... |
+ // 1 0 000 0001 | top segment
+ // 0 0 000 0000 v
+ //
+ // -1 1 111 1111 ^
+ // -2 1 111 1110 | bottom segment
+ // ... |
+ // lowest 1 000 0000 v
+ //
+ // Note that if we exclude the sign bit, the range is made up of 2 segments
+ // of contiguous increasing numbers. If min and max are both in the same
+ // segment before the sign extension, then min and max will continue to be
+ // in a contiguous segment after the sign extension. Only when min and max
+ // spans across more than 1 of these segments, will min and max no longer
+ // be guaranteed to be in a contiguous range after the sign extension.
+ //
+ // Hence, we can check if bits N-1 and up are identical for the range min
+ // and max. If so, then the new min and max can be be computed by simply
+ // applying sign extension to their original values.
+
+ constexpr unsigned numberOfBits = countOfBits<T>;
+ constexpr int64_t segmentMask = (1ll << (numberOfBits - 1)) - 1;
+ constexpr int64_t topBitsMask = ~segmentMask;
+ int64_t minTopBits = topBitsMask & min;
+ int64_t maxTopBits = topBitsMask & max;
+
+ if (minTopBits == maxTopBits)
+ return IntRange(static_cast<int64_t>(static_cast<T>(min)), static_cast<int64_t>(static_cast<T>(max)));
+
+ return top<T>();
+ }
+
IntRange zExt32()
{
ASSERT(m_min >= INT32_MIN);
@@ -2765,9 +2820,11 @@ class ReduceStrength {
rangeFor(value->child(1), timeToLive - 1), value->type());

case SExt8:
+ return rangeFor(value->child(0), timeToLive - 1).sExt<int8_t>();
case SExt16:
+ return rangeFor(value->child(0), timeToLive - 1).sExt<int16_t>();
case SExt32:
- return rangeFor(value->child(0), timeToLive - 1);
+ return rangeFor(value->child(0), timeToLive - 1).sExt<int32_t>();

case ZExt32:
return rangeFor(value->child(0), timeToLive - 1).zExt32();

The patch reminds me of several similar CTF challenge of VM although the codes before patching seems innocent. To analyze the vulnerability we need some basic knowledge of JavaScriptCore’s JIT compiler.

JIT of JavaScriptCore

JavaScriptCore has four tiers for JavaScript.

  • The LLInt, or low-level interpreter
  • The Baseline JIT, also known as a bytecode templete JIT
  • The DFG JIT, or data flow graph JIT
  • The FTL JIT, or faster than light JIT. FTL JIT uses multiple IRs (DFG IR, DFG SSA IR, B3 IR, and Assembly IR).

And the patches above were applied to B3ReduceStrength. Strength Reduction is a compiler optimization technique to replace those expensive operations to equivalent but less expensive operations. For example, replace the mul

[To be continued]