[mlir] Fix use-after-free bugs in {RankedTensorType|VectorType}::Builder (llvm#68969)

Previously, these would set their ArrayRef members to reference their
storage SmallVectors after a copy-on-write (COW) operation. This leads
to a use-after-free if the builder is copied and the original destroyed
(as the new builder would still reference the old SmallVector).

This could easily accidentally occur in code like (annotated):
```c++
// 1. `VectorType::Builder(type)` constructs a new temporary builder
// 2. `.dropDim(0)` updates the temporary builder by reference, and returns a `VectorType::Builder&`
//    - Modifying the shape is a COW operation, so `storage` is used, and `shape` updated the reference it
// 3. Assigning the reference to `auto` copies the builder (via the default C++ copy ctor)
//    -  There's no special handling for `shape` and `storage`, so the new shape points to the old builder's `storage`
auto newType = VectorType::Builder(type).dropDim(0);
// 4. When this line is reached the original temporary builder is destroyed
//    - Actually constructing the vector type is now a use-after-free
VectorType newVectorType = VectorType(newType);
```

This is fixed with these changes by using `CopyOnWriteArrayRef<T>`,
which implements the same functionality, but ensures no
dangling references are possible if it's copied. 

---

The VectorType::Builder also set the ArrayRef<bool> scalableDims member
to a temporary SmallVector when the provided scalableDims are empty.
This again leads to a use-after-free, and is unnecessary as
VectorType::get already handles being passed an empty scalableDims
array.

These bugs were in-part caught by UBSAN, see:
https://lab.llvm.org/buildbot/#/builders/5/builds/37355

Author: MacDue

mlir/include/mlir/IR/BuiltinTypes.h

@@ -11,6 +11,7 @@
 
 #include "mlir/IR/BuiltinAttributeInterfaces.h"
 #include "mlir/IR/BuiltinTypeInterfaces.h"
+#include "mlir/Support/ADTExtras.h"
 
 namespace llvm {
 class BitVector;
@@ -274,20 +275,14 @@ class RankedTensorType::Builder {
   /// Erase a dim from shape @pos.
   Builder &dropDim(unsigned pos) {
     assert(pos < shape.size() && "overflow");
-    if (storage.empty())
-      storage.append(shape.begin(), shape.end());
-    storage.erase(storage.begin() + pos);
-    shape = {storage.data(), storage.size()};
+    shape.erase(pos);
     return *this;
   }
 
   /// Insert a val into shape @pos.
   Builder &insertDim(int64_t val, unsigned pos) {
     assert(pos <= shape.size() && "overflow");
-    if (storage.empty())
-      storage.append(shape.begin(), shape.end());
-    storage.insert(storage.begin() + pos, val);
-    shape = {storage.data(), storage.size()};
+    shape.insert(pos, val);
     return *this;
   }
 
@@ -296,9 +291,7 @@ class RankedTensorType::Builder {
   }
 
 private:
-  ArrayRef<int64_t> shape;
-  // Owning shape data for copy-on-write operations.
-  SmallVector<int64_t> storage;
+  CopyOnWriteArrayRef<int64_t> shape;
   Type elementType;
   Attribute encoding;
 };
@@ -313,27 +306,18 @@ class VectorType::Builder {
 public:
   /// Build from another VectorType.
   explicit Builder(VectorType other)
-      : shape(other.getShape()), elementType(other.getElementType()),
+      : elementType(other.getElementType()), shape(other.getShape()),
         scalableDims(other.getScalableDims()) {}
 
   /// Build from scratch.
   Builder(ArrayRef<int64_t> shape, Type elementType,
-          unsigned numScalableDims = 0, ArrayRef<bool> scalableDims = {})
-      : shape(shape), elementType(elementType) {
-    if (scalableDims.empty())
-      scalableDims = SmallVector<bool>(shape.size(), false);
-    else
-      this->scalableDims = scalableDims;
-  }
+          ArrayRef<bool> scalableDims = {})
+      : elementType(elementType), shape(shape), scalableDims(scalableDims) {}
 
   Builder &setShape(ArrayRef<int64_t> newShape,
                     ArrayRef<bool> newIsScalableDim = {}) {
-    if (newIsScalableDim.empty())
-      scalableDims = SmallVector<bool>(shape.size(), false);
-    else
-      scalableDims = newIsScalableDim;
-
     shape = newShape;
+    scalableDims = newIsScalableDim;
     return *this;
   }
 
@@ -345,25 +329,16 @@ class VectorType::Builder {
   /// Erase a dim from shape @pos.
   Builder &dropDim(unsigned pos) {
     assert(pos < shape.size() && "overflow");
-    if (storage.empty())
-      storage.append(shape.begin(), shape.end());
-    if (storageScalableDims.empty())
-      storageScalableDims.append(scalableDims.begin(), scalableDims.end());
-    storage.erase(storage.begin() + pos);
-    storageScalableDims.erase(storageScalableDims.begin() + pos);
-    shape = {storage.data(), storage.size()};
-    scalableDims =
-        ArrayRef<bool>(storageScalableDims.data(), storageScalableDims.size());
+    shape.erase(pos);
+    if (!scalableDims.empty())
+      scalableDims.erase(pos);
     return *this;
   }
 
   /// Set a dim in shape @pos to val.
   Builder &setDim(unsigned pos, int64_t val) {
-    if (storage.empty())
-      storage.append(shape.begin(), shape.end());
-    assert(pos < storage.size() && "overflow");
-    storage[pos] = val;
-    shape = {storage.data(), storage.size()};
+    assert(pos < shape.size() && "overflow");
+    shape.set(pos, val);
     return *this;
   }
 
@@ -372,13 +347,9 @@ class VectorType::Builder {
   }
 
 private:
-  ArrayRef<int64_t> shape;
-  // Owning shape data for copy-on-write operations.
-  SmallVector<int64_t> storage;
   Type elementType;
-  ArrayRef<bool> scalableDims;
-  // Owning scalableDims data for copy-on-write operations.
-  SmallVector<bool> storageScalableDims;
+  CopyOnWriteArrayRef<int64_t> shape;
+  CopyOnWriteArrayRef<bool> scalableDims;
 };
 
 /// Given an `originalShape` and a `reducedShape` assumed to be a subset of

mlir/include/mlir/Support/ADTExtras.h

@@ -0,0 +1,82 @@
+//===- ADTExtras.h - Extra ADTs for use in MLIR -----------------*- C++ -*-===//
+//
+// Part of the LLVM Project, under the Apache License v2.0 with LLVM Exceptions.
+// See https://llvm.org/LICENSE.txt for license information.
+// SPDX-License-Identifier: Apache-2.0 WITH LLVM-exception
+//
+//===----------------------------------------------------------------------===//
+
+#ifndef MLIR_SUPPORT_ADTEXTRAS_H
+#define MLIR_SUPPORT_ADTEXTRAS_H
+
+#include "llvm/ADT/ArrayRef.h"
+#include "llvm/ADT/SmallVector.h"
+
+namespace mlir {
+
+//===----------------------------------------------------------------------===//
+// CopyOnWriteArrayRef<T>
+//===----------------------------------------------------------------------===//
+
+// A wrapper around an ArrayRef<T> that copies to a SmallVector<T> on
+// modification. This is for use in the mlir::<Type>::Builders.
+template <typename T>
+class CopyOnWriteArrayRef {
+public:
+  CopyOnWriteArrayRef(ArrayRef<T> array) : nonOwning(array){};
+
+  CopyOnWriteArrayRef &operator=(ArrayRef<T> array) {
+    nonOwning = array;
+    owningStorage = {};
+    return *this;
+  }
+
+  void insert(size_t index, T value) {
+    SmallVector<T> &vector = ensureCopy();
+    vector.insert(vector.begin() + index, value);
+  }
+
+  void erase(size_t index) {
+    // Note: A copy can be avoided when just dropping the front/back dims.
+    if (isNonOwning() && index == 0) {
+      nonOwning = nonOwning.drop_front();
+    } else if (isNonOwning() && index == size() - 1) {
+      nonOwning = nonOwning.drop_back();
+    } else {
+      SmallVector<T> &vector = ensureCopy();
+      vector.erase(vector.begin() + index);
+    }
+  }
+
+  void set(size_t index, T value) { ensureCopy()[index] = value; }
+
+  size_t size() const { return ArrayRef<T>(*this).size(); }
+
+  bool empty() const { return ArrayRef<T>(*this).empty(); }
+
+  operator ArrayRef<T>() const {
+    return nonOwning.empty() ? ArrayRef<T>(owningStorage) : nonOwning;
+  }
+
+private:
+  bool isNonOwning() const { return !nonOwning.empty(); }
+
+  SmallVector<T> &ensureCopy() {
+    // Empty non-owning storage signals the array has been copied to the owning
+    // storage (or both are empty). Note: `nonOwning` should never reference
+    // `owningStorage`. This can lead to dangling references if the
+    // CopyOnWriteArrayRef<T> is copied.
+    if (isNonOwning()) {
+      owningStorage = SmallVector<T>(nonOwning);
+      nonOwning = {};
+    }
+    return owningStorage;
+  }
+
+  ArrayRef<T> nonOwning;
+  SmallVector<T> owningStorage;
+};
+
+} // namespace mlir
+
+#endif

mlir/unittests/IR/ShapedTypeTest.cpp

@@ -131,4 +131,99 @@ TEST(ShapedTypeTest, CloneVector) {
             VectorType::get(vectorNewShape, vectorNewType));
 }
 
+TEST(ShapedTypeTest, VectorTypeBuilder) {
+  MLIRContext context;
+  Type f32 = FloatType::getF32(&context);
+
+  SmallVector<int64_t> shape{2, 4, 8, 9, 1};
+  SmallVector<bool> scalableDims{true, false, true, false, false};
+  VectorType vectorType = VectorType::get(shape, f32, scalableDims);
+
+  {
+    // Drop some dims.
+    VectorType dropFrontTwoDims =
+        VectorType::Builder(vectorType).dropDim(0).dropDim(0);
+    ASSERT_EQ(vectorType.getElementType(), dropFrontTwoDims.getElementType());
+    ASSERT_EQ(vectorType.getShape().drop_front(2), dropFrontTwoDims.getShape());
+    ASSERT_EQ(vectorType.getScalableDims().drop_front(2),
+              dropFrontTwoDims.getScalableDims());
+  }
+
+  {
+    // Set some dims.
+    VectorType setTwoDims =
+        VectorType::Builder(vectorType).setDim(0, 10).setDim(3, 12);
+    ASSERT_EQ(setTwoDims.getShape(), ArrayRef<int64_t>({10, 4, 8, 12, 1}));
+    ASSERT_EQ(vectorType.getElementType(), setTwoDims.getElementType());
+    ASSERT_EQ(vectorType.getScalableDims(), setTwoDims.getScalableDims());
+  }
+
+  {
+    // Test for bug from:
+    // https://github.com/llvm/llvm-project/commit/b44b3494f60296db6aca38a14cab061d9b747a0a
+    // Constructs a temporary builder, modifies it, copies it to `builder`.
+    // This used to lead to a use-after-free. Running under sanitizers will
+    // catch any issues.
+    VectorType::Builder builder = VectorType::Builder(vectorType).setDim(0, 16);
+    VectorType newVectorType = VectorType(builder);
+    ASSERT_EQ(newVectorType.getDimSize(0), 16);
+  }
+
+  {
+    // Make builder from scratch (without scalable dims) -- this use to lead to
+    // a use-after-free see: https://github.com/llvm/llvm-project/pull/68969.
+    // Running under sanitizers will catch any issues.
+    SmallVector<int64_t> shape{1, 2, 3, 4};
+    VectorType::Builder builder(shape, f32);
+    ASSERT_EQ(VectorType(builder).getShape(), ArrayRef(shape));
+  }
+
+  {
+    // Set vector shape (without scalable dims) -- this use to lead to
+    // a use-after-free see: https://github.com/llvm/llvm-project/pull/68969.
+    // Running under sanitizers will catch any issues.
+    VectorType::Builder builder(vectorType);
+    SmallVector<int64_t> newShape{2, 2};
+    builder.setShape(newShape);
+    ASSERT_EQ(VectorType(builder).getShape(), ArrayRef(newShape));
+  }
+}
+
+TEST(ShapedTypeTest, RankedTensorTypeBuilder) {
+  MLIRContext context;
+  Type f32 = FloatType::getF32(&context);
+
+  SmallVector<int64_t> shape{2, 4, 8, 16, 32};
+  RankedTensorType tensorType = RankedTensorType::get(shape, f32);
+
+  {
+    // Drop some dims.
+    RankedTensorType dropFrontTwoDims =
+        RankedTensorType::Builder(tensorType).dropDim(0).dropDim(1).dropDim(0);
+    ASSERT_EQ(tensorType.getElementType(), dropFrontTwoDims.getElementType());
+    ASSERT_EQ(dropFrontTwoDims.getShape(), ArrayRef<int64_t>({16, 32}));
+  }
+
+  {
+    // Insert some dims.
+    RankedTensorType insertTwoDims =
+        RankedTensorType::Builder(tensorType).insertDim(7, 2).insertDim(9, 3);
+    ASSERT_EQ(tensorType.getElementType(), insertTwoDims.getElementType());
+    ASSERT_EQ(insertTwoDims.getShape(),
+              ArrayRef<int64_t>({2, 4, 7, 9, 8, 16, 32}));
+  }
+
+  {
+    // Test for bug from:
+    // https://github.com/llvm/llvm-project/commit/b44b3494f60296db6aca38a14cab061d9b747a0a
+    // Constructs a temporary builder, modifies it, copies it to `builder`.
+    // This used to lead to a use-after-free. Running under sanitizers will
+    // catch any issues.
+    RankedTensorType::Builder builder =
+        RankedTensorType::Builder(tensorType).dropDim(0);
+    RankedTensorType newTensorType = RankedTensorType(builder);
+    ASSERT_EQ(tensorType.getShape().drop_front(), newTensorType.getShape());
+  }
+}
+
 } // namespace