Skip to content

CHANGELOG.md

Latest commit

 

History

History
124 lines (100 loc) · 5.96 KB

CHANGELOG.md

File metadata and controls

124 lines (100 loc) · 5.96 KB
Raw

Changelog

v0.1.1

Fixed

  • Process abort when a constructor is called without new. Calling an algorithm or mode constructor as a plain function (algorithm.Rijndael128() instead of new algorithm.Rijndael128()) redirected through the stored constructor with Nan::NewInstance(...).ToLocalChecked(). On current V8 that re-entrant NewInstance, on a template that uses Inherit, can return an empty MaybeLocal, and ToLocalChecked on empty is a fatal abort rather than a throw:

    FATAL ERROR: v8::ToLocalChecked Empty MaybeLocal

    The redirect now checks the MaybeLocal and reports an ordinary error instead of aborting the process. Where V8 permits the redirect it still returns a working instance, so calling with or without new both behave as before on the runtimes where they already worked.

  • Process abort on garbage collection on Node 24.19 and newer. Node 24.19.0 added environment cleanup-hook calls to node::ObjectWrap, which this addon extended. An addon object's destructor runs from a V8 weak callback with no context entered, so RemoveEnvironmentCleanupHook hit Node's own CHECK_NOT_NULL(env) and aborted the whole process:

    node::RemoveEnvironmentCleanupHook ... Assertion failed: (env) != nullptr
cryptian::AlgorithmStream<Arcfour>::~AlgorithmStream()

    It fired whenever a cipher or mode object was collected on an affected Node, so it struck ordinary programs that create and drop ciphers, and it was especially visible under test runners such as Vitest that collect aggressively. This is an upstream Node change that affected every node::ObjectWrap addon, not only this one. The addon now extends Nan::ObjectWrap, whose destructor resets its persistent handle and calls no cleanup hooks, so the null-context path cannot be reached. No API or output change.

Note

  • This does not change the separate worker_threads limitation: the addon is still registered with NODE_MODULE rather than the context-aware variant, so it cannot be loaded in a worker once the main thread has loaded it. That is tracked separately.

v0.1.0

This release corrects several faults that changed output or could crash the process. Some of the corrections change the bytes the library produces, so read the compatibility note before upgrading.

If you have data written by an earlier version

Two fixes change ciphertext for cases that were previously wrong:

  • Blowfish and CAST-128 encrypted incorrectly whenever the key contained a byte of 0x80 or higher, which is almost every real key. Data written by an affected version will not decrypt after upgrading, and it never interoperated with PHP mcrypt, OpenSSL, or anything else, because the output was wrong. The only recovery is to re-encrypt from the original source with a fixed version.

Everything else below either rejects input that used to be mishandled or fixes a crash, so it does not change any output that was previously correct.

Fixed

  • Blowfish and CAST-128 key schedules sign-extended key bytes at or above 0x80, corrupting the schedule. They now match libmcrypt for all keys.
  • Use-after-free: a mode kept a raw pointer to its algorithm without keeping the algorithm's JavaScript object alive, so a construction like new mode.cbc.Cipher(new algorithm.Rijndael128(), iv) could crash the process once the algorithm was garbage collected. The mode now holds the algorithm for its own lifetime.
  • Initialization vector length is now validated against the algorithm block size. A short vector previously caused an out-of-bounds read, and in CBC an out-of-bounds write, while appearing to succeed.
  • Stream initialization vectors: setIv had an inverted comparison, so a new vector of the same length as the previous one was silently ignored. Arcfour additionally read past the end of the vector because of an operator precedence error in its index.
  • Raw encrypt/decrypt on an algorithm now require exactly one block and throw otherwise. They previously either truncated multi-block input or, for Twofish and Serpent, returned a buffer whose trailing partial block was zeros. Empty input is a no-op.
  • Counted paddings (Pkcs5, Pkcs7, AnsiX923, Iso10126) rejected a trailing length byte of zero and no longer accept a length past the end of the data. Pkcs5 also verifies its final padding byte, which it previously skipped.
  • BlockDecrypt stream flush reported a transform failure on the stream instead of throwing an uncaught exception.
  • NotImplementedError now sets its own name and survives instanceof.

Added

  • Twofish (MCRYPT_TWOFISH), implemented from the specification and verified against published vectors and against libmcrypt.
  • Serpent (MCRYPT_SERPENT), likewise.
  • Panama (MCRYPT_PANAMA), verified against libmcrypt including its initialization-vector path.
  • A migration guide, docs/migrating-from-php-mcrypt.md, mapping every mcrypt constant onto this library and covering the naming traps.
  • Extensive test coverage: every cipher and mode is now checked against vectors generated from libmcrypt itself, the counted paddings against OpenSSL, and the remaining paddings and PCBC against their published definitions.

Changed

  • Build requires a C++20 toolchain, needed by the V8 headers in current Node releases. engines now declares Node 22 or newer, and CI covers Node 22, 24 and 26 on Linux, macOS and Windows.
  • The published package no longer contains the test suite, test vectors, or TypeScript sources, reducing it from 326 files to about 105.

Notes

  • Threeway requires setEndianCompat(true) to match mcrypt, since it defaults to the byte order of the original 3-Way. This was always the case but is now documented.
  • The raw algorithm API processes a single block; use a mode for longer data.
  • Modes are not context-aware and cannot be loaded in a worker_threads worker once the main thread has loaded the addon.