buffer: add mask() and unmask()

Add buffer.mask(source, mask, output[, offset[, length]]) and
buffer.unmask(buffer, mask), which XOR data with a repeating 4-byte
key. This is the masking that WebSocket clients apply to every frame
they send (RFC 6455, Section 5.3), and that servers undo on every frame
they receive.

Userland WebSocket implementations do this either with a byte-by-byte
JS loop (undici, and ws without optional dependencies) or with the
bufferutil native addon. bufferutil is installed for only about 3% of
ws downloads and has no linux-arm64 prebuild, so almost all users run
the JS loop. The signature matches bufferutil, so existing users can
switch with a feature check, as ws did for buffer.isUtf8().

The implementation uses a V8 fast API call and processes 8-byte words,
which the compiler vectorizes. It handles overlapping source and output
views, and treats every view type as raw bytes. It is about 20x faster
than the JS loop at 1 KiB and 40x faster at 64 KiB, and faster than
bufferutil at every size from 32 bytes up.

Signed-off-by: James M Snell <jasnell@gmail.com>

Author: jasnell

benchmark/buffers/buffer-mask.js

@@ -0,0 +1,56 @@
+'use strict';
+
+const common = require('../common.js');
+const { mask, unmask } = require('node:buffer');
+
+const bench = common.createBenchmark(main, {
+  type: ['mask', 'unmask'],
+  // 'buffer' uses buffer.mask() / buffer.unmask(). 'js' is the byte-by-byte
+  // loop that userland WebSocket implementations fall back to without a
+  // native addon, for comparison.
+  impl: ['buffer', 'js'],
+  len: [4, 16, 125, 1024, 16384, 65536],
+  n: [1e6],
+});
+
+function jsMask(source, key, output, offset, length) {
+  for (let i = 0; i < length; i++) {
+    output[offset + i] = source[i] ^ key[i & 3];
+  }
+}
+
+function jsUnmask(buffer, key) {
+  for (let i = 0; i < buffer.length; i++) {
+    buffer[i] ^= key[i & 3];
+  }
+}
+
+function main({ n, type, impl, len }) {
+  const key = Buffer.from([0x12, 0x34, 0x56, 0x78]);
+  const source = Buffer.alloc(len, 'abcdefg');
+  // Leave room for a WebSocket frame header before the payload.
+  const output = Buffer.alloc(len + 14);
+
+  switch (type) {
+    case 'mask': {
+      const fn = impl === 'buffer' ? mask : jsMask;
+      bench.start();
+      for (let i = 0; i < n; i++) {
+        fn(source, key, output, 14, len);
+      }
+      bench.end(n);
+      break;
+    }
+    case 'unmask': {
+      const fn = impl === 'buffer' ? unmask : jsUnmask;
+      bench.start();
+      for (let i = 0; i < n; i++) {
+        fn(source, key);
+      }
+      bench.end(n);
+      break;
+    }
+    default:
+      throw new Error(`Unexpected type: ${type}`);
+  }
+}

doc/api/buffer.md

@@ -5469,6 +5469,72 @@ added: v3.0.0
 
 An alias for [`buffer.constants.MAX_STRING_LENGTH`][].
 
+### `buffer.mask(source, mask, output[, offset[, length]])`
+
+<!-- YAML
+added: REPLACEME
+-->
+
+* `source` {Buffer|TypedArray|DataView} The data to mask.
+* `mask` {Buffer|TypedArray|DataView} The 4-byte masking key.
+* `output` {Buffer|TypedArray|DataView} Where to write the masked data.
+* `offset` {integer} Byte offset in `output` at which to start writing.
+  **Default:** `0`.
+* `length` {integer} Number of bytes of `source` to mask.
+  **Default:** `source.byteLength`.
+
+XORs the first `length` bytes of `source` with `mask`, repeated, and writes the
+result to `output` starting at `offset`: byte `i` of `source` is XORed with
+byte `i % 4` of `mask`. This is the masking operation that WebSocket clients
+apply to every frame payload they send ([RFC 6455, Section 5.3][]), and that
+servers undo on every frame they receive. Applying it twice with the same
+`mask` restores the original data. `source` is not modified, unless it shares
+memory with `output`.
+
+All arguments are treated as raw bytes, whatever the view type. `source` and
+`output` may be the same view or overlapping views over the same memory; the
+result is the same as if `source` had been copied first.
+
+An error is thrown if `mask` is not exactly 4 bytes long, if `length` is
+greater than `source.byteLength`, or if `offset + length` is greater than
+`output.byteLength`. Nothing is written in that case.
+
+```mjs
+import { Buffer, mask, unmask } from 'node:buffer';
+
+const key = Buffer.from([0x37, 0xfa, 0x21, 0x3d]);
+const payload = Buffer.from('Hello');
+
+// Write a masked copy of the payload after a 6-byte frame header.
+const frame = Buffer.alloc(6 + payload.length);
+mask(payload, key, frame, 6);
+console.log(frame.subarray(6));
+// Prints: <Buffer 7f 9f 4d 51 58>
+
+const received = frame.subarray(6);
+unmask(received, key);
+console.log(received.toString());
+// Prints: Hello
+```
+
+```cjs
+const { Buffer, mask, unmask } = require('node:buffer');
+
+const key = Buffer.from([0x37, 0xfa, 0x21, 0x3d]);
+const payload = Buffer.from('Hello');
+
+// Write a masked copy of the payload after a 6-byte frame header.
+const frame = Buffer.alloc(6 + payload.length);
+mask(payload, key, frame, 6);
+console.log(frame.subarray(6));
+// Prints: <Buffer 7f 9f 4d 51 58>
+
+const received = frame.subarray(6);
+unmask(received, key);
+console.log(received.toString());
+// Prints: Hello
+```
+
 ### `buffer.resolveObjectURL(id)`
 
 <!-- YAML
@@ -5534,6 +5600,23 @@ console.log(newBuf.toString('ascii'));
 Because the Euro (`€`) sign is not representable in US-ASCII, it is replaced
 with `?` in the transcoded `Buffer`.
 
+### `buffer.unmask(buffer, mask)`
+
+<!-- YAML
+added: REPLACEME
+-->
+
+* `buffer` {Buffer|TypedArray|DataView} The data to unmask, in place.
+* `mask` {Buffer|TypedArray|DataView} The 4-byte masking key.
+
+XORs every byte of `buffer` with `mask`, repeated, in place: byte `i` is XORed
+with byte `i % 4` of `mask`. This is equivalent to
+`buffer.mask(buffer, mask, buffer)`, and is typically used to unmask a received
+WebSocket frame payload ([RFC 6455, Section 5.3][]). See [`buffer.mask()`][]
+for an example.
+
+An error is thrown if `mask` is not exactly 4 bytes long.
+
 ### Buffer constants
 
 <!-- YAML
@@ -5772,6 +5855,7 @@ or after startup, if the alignment has to hold at run time.
 [Base64]: https://en.wikipedia.org/wiki/Base64
 [ISO-8859-1]: https://en.wikipedia.org/wiki/ISO-8859-1
 [RFC 4648, Section 5]: https://tools.ietf.org/html/rfc4648#section-5
+[RFC 6455, Section 5.3]: https://datatracker.ietf.org/doc/html/rfc6455#section-5.3
 [UTF-16]: https://en.wikipedia.org/wiki/UTF-16
 [UTF-8]: https://en.wikipedia.org/wiki/UTF-8
 [WHATWG Encoding Standard]: https://encoding.spec.whatwg.org/
@@ -5815,6 +5899,7 @@ or after startup, if the alignment has to hold at run time.
 [`buffer.constants.MAX_LENGTH`]: #bufferconstantsmax_length
 [`buffer.constants.MAX_STRING_LENGTH`]: #bufferconstantsmax_string_length
 [`buffer.kMaxLength`]: #bufferkmaxlength
+[`buffer.mask()`]: #buffermasksource-mask-output-offset-length
 [`util.inspect()`]: util.md#utilinspectobject-options
 [`v8.startupSnapshot.setDeserializeMainFunction()`]: v8.md#v8startupsnapshotsetdeserializemainfunctioncallback-data
 [`v8::Uint8Array::kMaxLength`]: https://v8.github.io/api/head/classv8_1_1Uint8Array.html#a7677e3d0c9c92e4d40bef7212f5980c6

lib/buffer.js

@@ -63,6 +63,7 @@ const {
   fill: bindingFill,
   isAscii: bindingIsAscii,
   isUtf8: bindingIsUtf8,
+  mask: bindingMask,
   stringLengthUtf8: bindingStringLengthUtf8,
   indexOfBuffer,
   indexOfNumber,
@@ -1498,6 +1499,56 @@ function isAscii(input) {
   throw new ERR_INVALID_ARG_TYPE('input', ['ArrayBuffer', 'Buffer', 'TypedArray'], input);
 }
 
+// Packs a 4-byte mask into a uint32, with byte k in bits 8k..8k+7.
+function maskKeyToUint32(key) {
+  if (!isArrayBufferView(key)) {
+    throw new ERR_INVALID_ARG_TYPE('mask', ['Buffer', 'TypedArray', 'DataView'], key);
+  }
+  if (key.byteLength !== 4) {
+    throw new ERR_INVALID_ARG_VALUE('mask', key, 'must be 4 bytes long');
+  }
+  if (!isUint8Array(key)) {
+    key = new Uint8Array(key.buffer, key.byteOffset, 4);
+  }
+  return (key[0] | (key[1] << 8) | (key[2] << 16) | (key[3] << 24)) >>> 0;
+}
+
+function mask(source, key, output, offset, length) {
+  if (!isArrayBufferView(source)) {
+    throw new ERR_INVALID_ARG_TYPE('source', ['Buffer', 'TypedArray', 'DataView'], source);
+  }
+  const maskValue = maskKeyToUint32(key);
+  if (!isArrayBufferView(output)) {
+    throw new ERR_INVALID_ARG_TYPE('output', ['Buffer', 'TypedArray', 'DataView'], output);
+  }
+  const outputLength = output.byteLength;
+  if (offset === undefined) {
+    offset = 0;
+  } else {
+    validateInteger(offset, 'offset', 0, outputLength);
+  }
+  if (length === undefined) {
+    length = source.byteLength;
+  } else {
+    validateInteger(length, 'length', 0, source.byteLength);
+  }
+  if (length > outputLength - offset) {
+    throw new ERR_OUT_OF_RANGE('length', `<= ${outputLength - offset}`, length);
+  }
+  if (length === 0) return;
+  bindingMask(source, output, offset, length, maskValue);
+}
+
+function unmask(buffer, key) {
+  if (!isArrayBufferView(buffer)) {
+    throw new ERR_INVALID_ARG_TYPE('buffer', ['Buffer', 'TypedArray', 'DataView'], buffer);
+  }
+  const maskValue = maskKeyToUint32(key);
+  const length = buffer.byteLength;
+  if (length === 0) return;
+  bindingMask(buffer, buffer, 0, length, maskValue);
+}
+
 function stringLength(input, encoding = 'utf8') {
   if (!isTypedArray(input) && !isAnyArrayBuffer(input)) {
     throw new ERR_INVALID_ARG_TYPE('input', ['ArrayBuffer', 'Buffer', 'TypedArray'], input);
@@ -1529,6 +1580,8 @@ module.exports = {
   transcode,
   isUtf8,
   isAscii,
+  mask,
+  unmask,
 
   // Legacy
   kMaxLength,

src/node_buffer.cc

@@ -1414,6 +1414,116 @@ static bool FastIsAscii(Local<Value> receiver,
 
 static CFunction fast_is_ascii(CFunction::Make(FastIsAscii));
 
+// Copies `length` bytes from `source` to `destination`, XORing byte i with
+// byte (i % 4) of `mask`, as done for WebSocket frame payloads (RFC 6455,
+// Section 5.3). The mask bytes are packed little-endian into a uint32, so mask
+// byte k is (mask >> (8 * k)) & 0xff regardless of platform endianness.
+// `source` and `destination` may be the same memory or otherwise overlap.
+static void MaskImpl(const uint8_t* source,
+                     uint8_t* destination,
+                     size_t length,
+                     uint32_t mask) {
+  uint8_t pattern[8];
+  for (size_t i = 0; i < 8; i++) {
+    pattern[i] = static_cast<uint8_t>(mask >> (8 * (i & 3)));
+  }
+  uint64_t pattern64;
+  memcpy(&pattern64, pattern, sizeof(pattern64));
+
+  // A forward pass is safe when the destination starts at or before the
+  // source: every source chunk is read before any overlapping byte is
+  // written. Otherwise, take a copy of the source first.
+  std::unique_ptr<uint8_t[]> copy;
+  if (destination > source && destination < source + length) [[unlikely]] {
+    copy.reset(new uint8_t[length]);
+    memcpy(copy.get(), source, length);
+    source = copy.get();
+  }
+
+  // Chunks start at multiples of 8, so the mask phase of every chunk is 0.
+  // memcpy() is used for unaligned loads and stores, and lets the compiler
+  // vectorize the loop.
+  size_t i = 0;
+  for (; i + 32 <= length; i += 32) {
+    uint64_t a, b, c, d;
+    memcpy(&a, source + i, 8);
+    memcpy(&b, source + i + 8, 8);
+    memcpy(&c, source + i + 16, 8);
+    memcpy(&d, source + i + 24, 8);
+    a ^= pattern64;
+    b ^= pattern64;
+    c ^= pattern64;
+    d ^= pattern64;
+    memcpy(destination + i, &a, 8);
+    memcpy(destination + i + 8, &b, 8);
+    memcpy(destination + i + 16, &c, 8);
+    memcpy(destination + i + 24, &d, 8);
+  }
+  for (; i + 8 <= length; i += 8) {
+    uint64_t v;
+    memcpy(&v, source + i, sizeof(v));
+    v ^= pattern64;
+    memcpy(destination + i, &v, sizeof(v));
+  }
+  for (; i < length; i++) destination[i] = source[i] ^ pattern[i & 3];
+}
+
+// Arguments are validated in JS: source and destination are ArrayBufferViews,
+// offset + length <= destination.byteLength and length <= source.byteLength.
+static void MaskArgs(Local<Value> source_obj,
+                     Local<Value> destination_obj,
+                     size_t offset,
+                     size_t length,
+                     uint32_t mask) {
+  if (length == 0) return;
+  SPREAD_BUFFER_ARG(destination_obj, destination);
+  CHECK_LE(offset, destination_length);
+  CHECK_LE(length, destination_length - offset);
+  uint8_t* dest = reinterpret_cast<uint8_t*>(destination_data) + offset;
+  if (source_obj == destination_obj) {
+    MaskImpl(
+        reinterpret_cast<const uint8_t*>(destination_data), dest, length, mask);
+    return;
+  }
+  SPREAD_BUFFER_ARG(source_obj, source);
+  CHECK_LE(length, source_length);
+  MaskImpl(reinterpret_cast<const uint8_t*>(source_data), dest, length, mask);
+}
+
+// mask(source, destination, offset, length, mask)
+static void Mask(const FunctionCallbackInfo<Value>& args) {
+  CHECK_EQ(args.Length(), 5);
+  CHECK(args[2]->IsNumber());
+  CHECK(args[3]->IsNumber());
+  CHECK(args[4]->IsUint32());
+  // Offsets and lengths can exceed uint32 for buffers larger than 4 GiB, so
+  // they are passed as doubles (exact for integers < 2^53).
+  MaskArgs(args[0],
+           args[1],
+           static_cast<size_t>(args[2].As<Number>()->Value()),
+           static_cast<size_t>(args[3].As<Number>()->Value()),
+           args[4].As<Uint32>()->Value());
+}
+
+static void FastMask(Local<Value> receiver,
+                     Local<Value> source_obj,
+                     Local<Value> destination_obj,
+                     double offset,
+                     double length,
+                     uint32_t mask,
+                     // NOLINTNEXTLINE(runtime/references)
+                     FastApiCallbackOptions& options) {
+  TRACK_V8_FAST_API_CALL("buffer.mask");
+  HandleScope scope(options.isolate);
+  MaskArgs(source_obj,
+           destination_obj,
+           static_cast<size_t>(offset),
+           static_cast<size_t>(length),
+           mask);
+}
+
+static CFunction fast_mask(CFunction::Make(FastMask));
+
 // Number of UTF-16 code units produced by decoding [p, end) as UTF-8 with
 // WHATWG "maximal subpart" U+FFFD replacement, matching the fallback that
 // StringBytes::Encode takes for invalid input (v8::String::NewFromUtf8).
@@ -1928,6 +2038,7 @@ void Initialize(Local<Object> target,
   SetFastMethodNoSideEffect(context, target, "isUtf8", IsUtf8, &fast_is_utf8);
   SetFastMethodNoSideEffect(
       context, target, "isAscii", IsAscii, &fast_is_ascii);
+  SetFastMethod(context, target, "mask", Mask, &fast_mask);
   SetFastMethodNoSideEffect(context,
                             target,
                             "stringLengthUtf8",
@@ -2008,6 +2119,8 @@ void RegisterExternalReferences(ExternalReferenceRegistry* registry) {
   registry->Register(fast_is_utf8);
   registry->Register(IsAscii);
   registry->Register(fast_is_ascii);
+  registry->Register(Mask);
+  registry->Register(fast_mask);
   registry->Register(StringLengthUtf8);
   registry->Register(fast_string_length_utf8);