fix(FormData): throw exception on very large input instead of crashing

When `FormData.from()` is called with a very large ArrayBuffer (exceeding
WebKit's String::MaxLength of INT32_MAX), it would crash with an assertion
failure in WebKit's StringImpl. This fixes the issue by:

1. Adding length checks in the C++ `toString` and related functions (helpers.h)
   to check against both Bun's synthetic limit and WebKit's String::MaxLength.
   For UTF-8 tagged strings, we use simdutf to calculate the actual UTF-16
   length only when the byte length exceeds the limit.

2. Throwing an ERR_STRING_TOO_LONG exception from createFromURLQuery when the
   string is too long, instead of silently returning an empty FormData.

Changes:
- Add length checks to all UTF-8 code paths in helpers.h:
  - toString(ZigString)
  - toString(ZigString, StringPointer)
  - toStringCopy(ZigString)
  - toStringCopy(ZigString, StringPointer)
  - appendToBuilder(ZigString, StringBuilder)
- Add WTF::String::MaxLength check to non-UTF-8 paths
- Throw ERR_STRING_TOO_LONG in createFromURLQuery when string is too long
- Properly propagate JSError from toJS to fromMultipartData

Now `FormData.from(new Uint32Array(913148244))` throws:
  error: Cannot create a string longer than 2147483647 characters
         code: "ERR_STRING_TOO_LONG"

🤖 Generated with [Claude Code](https://claude.com/claude-code)

Co-Authored-By: Claude <[email protected]>

Author: Claude Bot

src/bun.js/bindings/bindings.cpp

@@ -5673,7 +5673,15 @@ CPP_DECL JSC::EncodedJSValue WebCore__DOMFormData__createFromURLQuery(JSC::JSGlo
 {
     Zig::GlobalObject* globalObject = static_cast<Zig::GlobalObject*>(arg0);
     // don't need to copy the string because it internally does.
-    auto formData = DOMFormData::create(globalObject->scriptExecutionContext(), toString(*arg1));
+    auto str = toString(*arg1);
+    // toString() in helpers.h returns an empty string when the input exceeds
+    // String::MaxLength or Bun's synthetic allocation limit. This is the only
+    // condition under which toString() returns empty for non-empty input.
+    if (str.isEmpty() && arg1->len > 0) {
+        auto scope = DECLARE_THROW_SCOPE(globalObject->vm());
+        return Bun::ERR::STRING_TOO_LONG(scope, globalObject);
+    }
+    auto formData = DOMFormData::create(globalObject->scriptExecutionContext(), WTFMove(str));
     return JSValue::encode(toJSNewlyCreated(arg0, globalObject, WTFMove(formData)));
 }
 

src/bun.js/bindings/helpers.h

@@ -2,6 +2,7 @@
 
 #include "root.h"
 #include "wtf/text/ASCIILiteral.h"
+#include "wtf/SIMDUTF.h"
 
 #include <JavaScriptCore/Error.h>
 #include <JavaScriptCore/Exception.h>
@@ -79,12 +80,24 @@ static const WTF::String toString(ZigString str)
     }
     if (isTaggedUTF8Ptr(str.ptr)) [[unlikely]] {
         ASSERT_WITH_MESSAGE(!isTaggedExternalPtr(str.ptr), "UTF8 and external ptr are mutually exclusive. The external will never be freed.");
+        // Check if the resulting UTF-16 string could possibly exceed the maximum length.
+        // For valid UTF-8, the number of UTF-16 code units is <= the number of UTF-8 bytes
+        // (ASCII is 1:1; other code points use multiple UTF-8 bytes per UTF-16 code unit).
+        // We only need to compute the actual UTF-16 length when the byte length exceeds the limit.
+        size_t maxLength = std::min(Bun__stringSyntheticAllocationLimit, static_cast<size_t>(WTF::String::MaxLength));
+        if (str.len > maxLength) [[unlikely]] {
+            // UTF-8 byte length != UTF-16 length, so use simdutf to calculate the actual UTF-16 length.
+            size_t utf16Length = simdutf::utf16_length_from_utf8(reinterpret_cast<const char*>(untag(str.ptr)), str.len);
+            if (utf16Length > maxLength) {
+                return {};
+            }
+        }
         return WTF::String::fromUTF8ReplacingInvalidSequences(std::span { untag(str.ptr), str.len });
     }
 
     if (isTaggedExternalPtr(str.ptr)) [[unlikely]] {
         // This will fail if the string is too long. Let's make it explicit instead of an ASSERT.
-        if (str.len > Bun__stringSyntheticAllocationLimit) [[unlikely]] {
+        if (str.len > Bun__stringSyntheticAllocationLimit || str.len > WTF::String::MaxLength) [[unlikely]] {
             free_global_string(nullptr, reinterpret_cast<void*>(const_cast<unsigned char*>(untag(str.ptr))), static_cast<unsigned>(str.len));
             return {};
         }
@@ -95,7 +108,7 @@ static const WTF::String toString(ZigString str)
     }
 
     // This will fail if the string is too long. Let's make it explicit instead of an ASSERT.
-    if (str.len > Bun__stringSyntheticAllocationLimit) [[unlikely]] {
+    if (str.len > Bun__stringSyntheticAllocationLimit || str.len > WTF::String::MaxLength) [[unlikely]] {
         return {};
     }
 
@@ -121,11 +134,19 @@ static const WTF::String toString(ZigString str, StringPointer ptr)
         return WTF::String();
     }
     if (isTaggedUTF8Ptr(str.ptr)) [[unlikely]] {
+        // Check if the resulting UTF-16 string could possibly exceed the maximum length.
+        size_t maxLength = std::min(Bun__stringSyntheticAllocationLimit, static_cast<size_t>(WTF::String::MaxLength));
+        if (ptr.len > maxLength) [[unlikely]] {
+            size_t utf16Length = simdutf::utf16_length_from_utf8(reinterpret_cast<const char*>(&untag(str.ptr)[ptr.off]), ptr.len);
+            if (utf16Length > maxLength) {
+                return {};
+            }
+        }
         return WTF::String::fromUTF8ReplacingInvalidSequences(std::span { &untag(str.ptr)[ptr.off], ptr.len });
     }
 
     // This will fail if the string is too long. Let's make it explicit instead of an ASSERT.
-    if (str.len > Bun__stringSyntheticAllocationLimit) [[unlikely]] {
+    if (ptr.len > Bun__stringSyntheticAllocationLimit || ptr.len > WTF::String::MaxLength) [[unlikely]] {
         return {};
     }
 
@@ -141,11 +162,19 @@ static const WTF::String toStringCopy(ZigString str, StringPointer ptr)
         return WTF::String();
     }
     if (isTaggedUTF8Ptr(str.ptr)) [[unlikely]] {
+        // Check if the resulting UTF-16 string could possibly exceed the maximum length.
+        size_t maxLength = std::min(Bun__stringSyntheticAllocationLimit, static_cast<size_t>(WTF::String::MaxLength));
+        if (ptr.len > maxLength) [[unlikely]] {
+            size_t utf16Length = simdutf::utf16_length_from_utf8(reinterpret_cast<const char*>(&untag(str.ptr)[ptr.off]), ptr.len);
+            if (utf16Length > maxLength) {
+                return {};
+            }
+        }
         return WTF::String::fromUTF8ReplacingInvalidSequences(std::span { &untag(str.ptr)[ptr.off], ptr.len });
     }
 
     // This will fail if the string is too long. Let's make it explicit instead of an ASSERT.
-    if (str.len > Bun__stringSyntheticAllocationLimit) [[unlikely]] {
+    if (ptr.len > Bun__stringSyntheticAllocationLimit || ptr.len > WTF::String::MaxLength) [[unlikely]] {
         return {};
     }
 
@@ -161,6 +190,14 @@ static const WTF::String toStringCopy(ZigString str)
         return WTF::String();
     }
     if (isTaggedUTF8Ptr(str.ptr)) [[unlikely]] {
+        // Check if the resulting UTF-16 string could possibly exceed the maximum length.
+        size_t maxLength = std::min(Bun__stringSyntheticAllocationLimit, static_cast<size_t>(WTF::String::MaxLength));
+        if (str.len > maxLength) [[unlikely]] {
+            size_t utf16Length = simdutf::utf16_length_from_utf8(reinterpret_cast<const char*>(untag(str.ptr)), str.len);
+            if (utf16Length > maxLength) {
+                return {};
+            }
+        }
         return WTF::String::fromUTF8ReplacingInvalidSequences(std::span { untag(str.ptr), str.len });
     }
 
@@ -188,6 +225,14 @@ static void appendToBuilder(ZigString str, WTF::StringBuilder& builder)
         return;
     }
     if (isTaggedUTF8Ptr(str.ptr)) [[unlikely]] {
+        // Check if the resulting UTF-16 string could possibly exceed the maximum length.
+        size_t maxLength = std::min(Bun__stringSyntheticAllocationLimit, static_cast<size_t>(WTF::String::MaxLength));
+        if (str.len > maxLength) [[unlikely]] {
+            size_t utf16Length = simdutf::utf16_length_from_utf8(reinterpret_cast<const char*>(untag(str.ptr)), str.len);
+            if (utf16Length > maxLength) {
+                return;
+            }
+        }
         WTF::String converted = WTF::String::fromUTF8ReplacingInvalidSequences(std::span { untag(str.ptr), str.len });
         builder.append(converted);
         return;

src/url.zig

@@ -980,13 +980,18 @@ pub const FormData = struct {
         };
     };
 
-    pub fn toJS(globalThis: *jsc.JSGlobalObject, input: []const u8, encoding: Encoding) !jsc.JSValue {
+    pub fn toJS(globalThis: *jsc.JSGlobalObject, input: []const u8, encoding: Encoding) bun.JSError!jsc.JSValue {
         switch (encoding) {
             .URLEncoded => {
                 var str = jsc.ZigString.fromUTF8(strings.withoutUTF8BOM(input));
-                return jsc.DOMFormData.createFromURLQuery(globalThis, &str);
+                const result = jsc.DOMFormData.createFromURLQuery(globalThis, &str);
+                // Check if an exception was thrown (e.g., string too long)
+                if (result == .zero) {
+                    return error.JSError;
+                }
+                return result;
             },
-            .Multipart => |boundary| return toJSFromMultipartData(globalThis, input, boundary),
+            .Multipart => |boundary| return try toJSFromMultipartData(globalThis, input, boundary),
         }
     }
 
@@ -1041,7 +1046,7 @@ pub const FormData = struct {
             return globalThis.throwInvalidArguments("input must be a string or ArrayBufferView", .{});
         }
 
-        return FormData.toJS(globalThis, input, encoding) catch |err| return globalThis.throwError(err, "while parsing FormData");
+        return try FormData.toJS(globalThis, input, encoding);
     }
 
     comptime {
@@ -1053,12 +1058,12 @@ pub const FormData = struct {
         globalThis: *jsc.JSGlobalObject,
         input: []const u8,
         boundary: []const u8,
-    ) !jsc.JSValue {
+    ) bun.JSError!jsc.JSValue {
         const form_data_value = jsc.DOMFormData.create(globalThis);
         form_data_value.ensureStillAlive();
         const form = jsc.DOMFormData.fromJS(form_data_value) orelse {
             log("failed to create DOMFormData.fromJS", .{});
-            return error.@"failed to parse multipart data";
+            return globalThis.throwError(error.@"failed to parse multipart data", "while parsing FormData");
         };
         const Wrapper = struct {
             globalThis: *jsc.JSGlobalObject,
@@ -1131,7 +1136,7 @@ pub const FormData = struct {
 
             forEachMultipartEntry(input, boundary, *Wrapper, &wrap, Wrapper.onEntry) catch |err| {
                 log("failed to parse multipart data", .{});
-                return err;
+                return globalThis.throwError(err, "while parsing FormData");
             };
         }
 

test/js/web/html/FormData.test.ts

@@ -277,6 +277,23 @@ describe("FormData", () => {
     expect(fd.toJSON()).toEqual({ "1": "1" });
   });
 
+  test("FormData.from throws on very large input instead of crashing", () => {
+    // This test verifies that FormData.from throws an exception instead of crashing
+    // when given input larger than WebKit's String::MaxLength (INT32_MAX ~= 2GB).
+    // We use a smaller test case with the synthetic limit to avoid actually allocating 2GB+.
+    const { setSyntheticAllocationLimitForTesting } = require("bun:internal-for-testing");
+    // Set a small limit so we can test the boundary without allocating gigabytes
+    const originalLimit = setSyntheticAllocationLimitForTesting(1024 * 1024); // 1MB limit
+    try {
+      // Create a buffer larger than the limit
+      const largeBuffer = new Uint8Array(2 * 1024 * 1024); // 2MB
+      // @ts-expect-error - FormData.from is a Bun extension
+      expect(() => FormData.from(largeBuffer)).toThrow("Cannot create a string longer than");
+    } finally {
+      setSyntheticAllocationLimitForTesting(originalLimit);
+    }
+  });
+
   it("should throw on bad boundary", async () => {
     const response = new Response('foo\r\nContent-Disposition: form-data; name="foo"\r\n\r\nbar\r\n', {
       headers: {