zlib: re-enable CRC32 SIMD optimization

deps/zlib/zlib.gyp

@@ -111,33 +111,41 @@
             }],
           ],
         }, # zlib_arm_crc32
-        # {
-        #   'target_name': 'zlib_crc32_simd',
-        #   'type': 'static_library',
-        #   'conditions': [
-        #     ['OS!="win" or clang==1', {
-        #       'cflags': [
-        #         '-msse4.2',
-        #         '-mpclmul',
-        #       ],
-        #       'xcode_settings': {
-        #         'OTHER_CFLAGS': [
-        #           '-msse4.2',
-        #           '-mpclmul',
-        #         ],
-        #       },
-        #     }]
-        #   ],
-        #   'defines': [ 'CRC32_SIMD_SSE42_PCLMUL' ],
-        #   'include_dirs': [ '<(ZLIB_ROOT)' ],
-        #   'direct_dependent_settings': {
-        #     'defines': [ 'CRC32_SIMD_SSE42_PCLMUL' ],
-        #     'include_dirs': [ '<(ZLIB_ROOT)' ],
-        #   },
-        #   'sources': [
-        #     '<!@pymod_do_main(GN-scraper "<(ZLIB_ROOT)/BUILD.gn" "\\"zlib_crc32_simd\\".*?sources = ")',
-        #   ],
-        # }, # zlib_crc32_simd
+        {
+          'target_name': 'zlib_crc32_simd',
+          'type': 'static_library',
+          'conditions': [
+            ['OS!="win" or clang==1', {
+              'cflags': [
+                '-msse4.2',
+                '-mpclmul',
+              ],
+              'xcode_settings': {
+                'OTHER_CFLAGS': [
+                  '-msse4.2',
+                  '-mpclmul',
+                ],
+              },
+              'msvs_settings': {
+                'VCCLCompilerTool': {
+                  'AdditionalOptions': [
+                    '-msse4.2',
+                    '-mpclmul',
+                  ],
+                },
+              },
+            }]
+          ],
+          'defines': [ 'CRC32_SIMD_SSE42_PCLMUL' ],
+          'include_dirs': [ '<(ZLIB_ROOT)' ],
+          'direct_dependent_settings': {
+            'defines': [ 'CRC32_SIMD_SSE42_PCLMUL' ],
+            'include_dirs': [ '<(ZLIB_ROOT)' ],
+          },
+          'sources': [
+            '<!@pymod_do_main(GN-scraper "<(ZLIB_ROOT)/BUILD.gn" "\\"zlib_crc32_simd\\".*?sources = ")',
+          ],
+        }, # zlib_crc32_simd
         {
           'target_name': 'zlib_data_chunk_simd',
           'type': 'static_library',
@@ -212,9 +220,7 @@
             ['target_arch in "ia32 x64" and OS!="ios"', {
               'dependencies': [
                 'zlib_adler32_simd',
-                # Disabled due to memory corruption.
-                # See https://github.com/nodejs/node/issues/45268.
-                # 'zlib_crc32_simd',
+                'zlib_crc32_simd',
               ],
               'defines': [ 'DEFLATE_SLIDE_HASH_SSE2' ],
               'conditions': [

src/node_zlib.cc

@@ -2184,6 +2184,12 @@ void Initialize(Local<Object> target,
                 Local<Value> unused,
                 Local<Context> context,
                 void* priv) {
+  // The bundled zlib only detects CPU features in deflateInit2() and when
+  // crc32() or adler32() is called with a NULL buffer. zlib.crc32() always
+  // passes a buffer, so trigger detection here to let it use the SIMD path
+  // before any stream has been created.
+  crc32(0, nullptr, 0);
+
   Environment* env = Environment::GetCurrent(context);
 
   MakeClass<ZlibStream>::Make(env, target, "Zlib");

test/parallel/test-zlib-gzip-stored-output-boundary.js

@@ -0,0 +1,30 @@
+'use strict';
+
+// Regression test for https://github.com/nodejs/node/issues/45268.
+// With gzip and level 0, deflate_stored() copies input directly into the
+// output buffer through the SIMD CRC32 copy routine. An old version of that
+// routine stored 16 bytes for the final (len % 16) bytes and wrote past the
+// end of the output buffer, corrupting the heap. Exercise output buffers that
+// end at every offset modulo 16 and check the round trip.
+
+require('../common');
+const assert = require('assert');
+const zlib = require('zlib');
+
+for (let i = 0; i < 400; i++) {
+  const size = 40000 + i * 517;
+  const input = Buffer.alloc(size, i & 0xff);
+  for (let j = 0; j < size; j += 97) input[j] = j & 0xff;
+  const chunkSize = 65536 + (i % 17);
+  const gz = zlib.gzipSync(input, { level: 0, chunkSize });
+  assert.deepStrictEqual(zlib.gunzipSync(gz), input);
+}
+
+// Small inputs finished straight into a nearly full output buffer.
+for (let len = 1; len <= 64; len++) {
+  const input = Buffer.alloc(len, 0x61);
+  for (let chunkSize = 64; chunkSize <= 96; chunkSize++) {
+    const gz = zlib.gzipSync(input, { level: 0, chunkSize });
+    assert.deepStrictEqual(zlib.gunzipSync(gz), input);
+  }
+}