Implement optional constant-time decryption flow to hinder Bleichenbacher-like
attacks against RSA- and ElGamal public-key encrypted session keys.
Changes:
- Add `config.constantTimePKCS1Decryption` to enable the constant-time
processing (defaults to `false`). The constant-time option is off by default
since it has measurable performance impact on message decryption, and it is
only helpful in specific application scenarios (more info below).
- Add `config.constantTimePKCS1DecryptionSupportedSymmetricAlgorithms`
(defaults to the AES algorithms). The set of supported ciphers is restricted by
default since the number of algorithms negatively affects performance.
Bleichenbacher-like attacks are of concern for applications where both of the
following conditions are met:
1. new/incoming messages are automatically decrypted (without user
interaction);
2. an attacker can determine how long it takes to decrypt each message (e.g.
due to decryption errors being logged remotely).
Breaking change:
when generating new subkeys through `key.addSubkey()`, we now check
`config.rejectCurves` and prevent adding subkeys using the corresponding
curves.
By default, `config.rejectCurves` includes the brainpool curves
(`brainpoolP256r1`, `brainpoolP384r1`, `brainpoolP512r1`) and the Bitcoin curve
(`secp256k1`).
This is a follow up to #1395 , which introduced the same check to
`openpgp.generateKey`.
In several packet classes, we used to store string identifiers for public-key,
aead, cipher or hash algorithms. To make the code consistent and to avoid
having to convert to/from string values, we now always store integer values
instead, e.g. `enums.symmetric.aes128` is used instead of `'aes128'`.
This is not expected to be a breaking change for most library users. Note that
the type of `Key.getAlgorithmInfo()` and of the session key objects returned
and accepted by top-level functions remain unchanged.
Affected classes (type changes for some properties and method's arguments):
- `PublicKeyPacket`, `PublicSubkeyPacket`, `SecretKeyPacket`,
`SecretSubkeyPacket`
- `SymEncryptedIntegrityProtectedDataPacket`, `AEADEncryptedDataPacket`,
`SymmetricallyEncryptedDataPacket`
- `LiteralDataPacket`, `CompressedDataPacket`
- `PublicKeyEncryptedSessionKey`, `SymEncryptedSessionKeyPacket`
- `SignaturePacket`
Other potentially breaking changes:
- Removed property `AEADEncryptedDataPacket.aeadAlgo`, since it was redudant
given `.aeadAlgorithm`.
- Renamed `AEADEncryptedDataPacket.cipherAlgo` -> `.cipherAlgorithm`
Remove line separators (U+2028) and paragraph separators (U+2029)
from randomly generated test messages. These messages cause the
test to fail due to the difference in handling them between
multiline regexes and OpenPGP.js-internal functions.
Using `openpgp.reformatKey` with the default `date` option would render
messages signed with the original key unverifiable by OpenPGP.js v5 (not v4),
since the signing key would not be considered valid at the time of signing (due
to its self-certification signature being in the future, compared to the
message signature creation time).
This commit adds `config.allowInsecureVerificationWithReformattedKeys` (false
by default) to make it possible to still verify such messages with the
reformatted key provided the key is valid at the `date` specified for
verification (which defaults to the current time).
Support build processes that replace `process.env.NODE_ENV` with a
constant string (such as webpack and Vite) by using
`process.env.NODE_ENV` directly, instead of `globalThis.process &&
globalThis.process.env.NODE_ENV`, but do so inside a try/catch in case
`process` is not defined.
Co-authored-by: Daniel Huigens <d.huigens@protonmail.com>
Breaking changes:
- throw error on key generation if the requested public key algorithm is
included in `config.rejectPublicKeyAlgorithms`;
- add `config.rejectCurves` to blacklist a set of ECC curves, to prevent keys
using those curves from being generated, or being used to
encrypt/decrypt/sign/verify messages.
By default, `config.rejectCurves` includes the brainpool curves
(`brainpoolP256r1`, `brainpoolP384r1`, `brainpoolP512r1`) and the Bitcoin curve
(`secp256k1`). This is because it's unclear whether these curves will be
standardised[1], and we prefer to blacklist them already, rather than introduce
a breaking change after release.
[1] https://gitlab.com/openpgp-wg/rfc4880bis/-/merge_requests/47#note_634199141
Breaking change: passing non-existing configuration options to top-level
functions now results in an error, to make users aware that the given
configuration cannot be applied.
Configuration options related to parsing have been changed to make it possible
to try to read messages containing malformed packets. Changes:
- rename `config.tolerant` to `config.ignoreUnsupportedPackets`. This still
defaults to `true`.
- Add `config.ignoreMalformedPackets` to ignore packets that fail to parse
(when possible). This option was not available before and it defaults to `false`.
The `format` option in `openpgp.generateKey, reformatKey, revokeKey, encrypt,
sign, encryptSessionKey` now expects the value `'armored'` instead of `'armor'`
to output armored data. The other format options (i.e. `'binary'` and
`'object'`) remain unchanged.
Breaking changes:
- a new `format` option has been added to `openpgp.encrypt`, `sign` and
`encryptSessionKey` to select the format of the output message. `format`
replaces the existing `armor` option, and accepts three values:
* if `format: 'armor'` (default), an armored signed/encrypted message is
returned (same as `armor: true`).
* if `format: 'binary'`, a binary signed/encrypted message is returned (same
as `armor: false`).
* if `format: 'object'`, a Message or Signature object is returned (this was
not supported before).
This change is to uniform the output format selection across all top-level
functions (following up to #1345).
- All top-level functions now throw if unrecognised options are passed, to make
library users aware that those options are not being applied.
This change is to make the code more consistent between the streaming and
non-streaming cases.
The validity of a signature (or the corresponding verification error) can be
determined through the existing `verified` property.
The tests fail in Safari 14.1 due to unhandled rejections that have no clear
cause. The problem is likely related to the native TransformStream
implementation added in Safari 14.1 (in fact, using a polyfilled
TransformStream solves all issues).
API changes:
- `Key.isPublic()` has been removed, since it was redundant and it would
introduce TypeScript issues. Call `!Key.isPrivate()` instead.
TypeScript changes:
- the `openpgp.readKey(s)` functions are now declared as returning a `Key`
instead of a `PublicKey`. This is just a readability improvement to make it
clearer that the result could also be a `PrivateKey`.
- All `Key` methods that return a key object now have the narrowest possible
return type.
- The `Key.isPrivate()` method can now be used for type inference, allowing the
compiler to distinguish between `PrivateKey` and `PublicKey`.
Calling `key.isPrivate()` is the recommended way of distinguishing between a
`PrivateKey` and `PublicKey` at runtime, over using `key instanceof ...`, since
the latter depends on the specifics of the `Key` class hierarchy.
- Support passing a single Key ID directly to the `encryption/signingKeyIDs`
options of `openpgp.encrypt`, `sign`, `generateSessionKey` and
`encryptSessionKey`.
- Add type definitions for `openpgp.encryptSessionKey` and `decryptSessionKeys`.
- `openpgp.generateKey`, `reformatKey` and `revokeKey` take a new `format`
option, whose possible values are: `'armor', 'binary', 'object'` (default is
`'armor'`).
- `generateKey` and `reformatKey` now return an object of the form `{
publicKey, privateKey, revocationCertificate }`, where the type of `publicKey`
and `privateKey` depends on `options.format`:
* if `format: 'armor'` then `privateKey, publicKey` are armored strings;
* if `format: 'binary'` then `privateKey, publicKey` are `Uint8Array`;
* if `format: 'object'` then `privateKey, publicKey` are `PrivateKey` and
`PublicKey` objects respectively;
- `revokeKey` now returns `{ publicKey, privateKey }`, where:
* if a `PrivateKey` is passed as `key` input, `privateKey, publicKey` are of the
requested format;
* if a `PublicKey` is passed as `key` input, `publicKey` is of the requested format,
while `privateKey` is `null` (previously, in this case the `privateKey` field
was not defined).
Breaking changes:
- In `revokeKey`, if no `format` option is specified, the returned `publicKey,
privateKey` are armored strings (they used to be objects).
- In `generateKey` and `reformatKey`, the `key` value is no longer returned.
- For all three functions, the `publicKeyArmored` and `privateKeyArmored`
values are no longer returned.
The following fields are now `readonly` instead of `private`:
- `Key.keyPacket`
- `Subkey.keyPacket` and `Subkey.mainKey`
- `Signature.packets`
- `Message.packets`
- When parsing, throw on unexpected packets even if `config.tolerant = true`
(e.g. if a Public Key packet is found when reading a signature).
- Always ignore Trust and Marker packets on parsing.
- Fix#1145: correctly verify signatures that include Marker packets when
`config.tolerant = false`.
