PEP: 748 Title: A Unified TLS API for Python Author: Joop van de Pol , William Woodruff Sponsor: Alyssa Coghlan Discussions-To: https://discuss.python.org/t/pre-pep-discussion-revival-of-pep-543/51263 Status: Draft Type: Standards Track Created: 27-Jun-2024 Python-Version: 3.14 Post-History: `17-Apr-2024 `__ Replaces: 543 Abstract ======== This PEP defines a standard TLS interface in the form of a collection of protocol classes. This interface will allow Python implementations and third-party libraries to provide bindings to TLS libraries other than OpenSSL. These bindings can be used by tools that expect the interface provided by the Python standard library, with the goal of reducing the dependence of the Python ecosystem on OpenSSL. Rationale ========= It has become increasingly clear that robust and user-friendly TLS support is an extremely important part of the ecosystem of any popular programming language. For most of its lifetime, this role in the Python ecosystem has primarily been served by the :mod:`ssl` module, which provides a Python API to the `OpenSSL library `_. Because the :mod:`ssl` module is distributed with the Python standard library, it has become the overwhelmingly most popular method for handling TLS in Python. A majority of Python libraries, both in the standard library and on the Python Package Index, rely on the :mod:`ssl` module for their TLS connectivity. Unfortunately, the preeminence of the :mod:`ssl` module has had a number of tied the entire Python ecosystem tightly to OpenSSL. This has forced Python users to use OpenSSL even in situations where it may provide a worse user experience than alternative TLS implementations, which imposes a cognitive burden and makes it hard to provide “platform-native” experiences. Problems -------- The fact that the :mod:`ssl` module is built into the standard library has meant that all standard-library Python networking libraries are entirely reliant on the OpenSSL that the Python implementation has been linked against. This leads to the following issues: * It is difficult to take advantage of new, higher-security TLS without recompiling Python to get a new OpenSSL. While there are third-party bindings to OpenSSL (e.g. `pyOpenSSL `_), these need to be shimmed into a format that the standard library understands, forcing projects that want to use them to maintain substantial compatibility layers. * Windows distributions of Python need to be shipped with a copy of OpenSSL. This puts the CPython development team in the position of being OpenSSL redistributors, potentially needing to ship security updates to the Windows Python distributions when OpenSSL vulnerabilities are released. * macOS distributions of Python need either to be shipped with a copy of OpenSSL or linked against the system OpenSSL library. Apple has formally deprecated linking against the system OpenSSL library, and even if they had not, that library version has been unsupported by upstream for nearly one year as of the time of writing. The CPython development team has started shipping newer OpenSSLs with the Python available from python.org, but this has the same problem as with Windows. * Users may wish to integrate with TLS libraries other than OpenSSL for other reasons, such as maintenance burden versus a system-provided implementation, or because OpenSSL is simply too large and unwieldy for their platform (e.g. for embedded Python). Those users are left with the requirement to use third-party networking libraries that can interact with their preferred TLS library or to shim their preferred library into the OpenSSL-specific :mod:`ssl` module API. Additionally, the :mod:`ssl` module as implemented today limits the ability of CPython itself to add support for alternative TLS implementations, or remove OpenSSL support entirely, should either of these become necessary or useful. The :mod:`ssl` module exposes too many OpenSSL-specific function calls and features to easily map to an alternative TLS implementation. Proposal ======== This PEP proposes to introduce a few new Protocol Classes in Python 3.14 to provide TLS functionality that is not so strongly tied to OpenSSL. It also proposes to update standard library modules to use only the interface exposed by these protocol classes wherever possible. There are three goals here: 1. To provide a common API surface for both core and third-party developers to target their TLS implementations to. This allows TLS developers to provide interfaces that can be used by most Python code, and allows network developers to have an interface that they can target that will work with a wide range of TLS implementations. 1. To provide an API that has few or no OpenSSL-specific concepts leak through. The :mod:`ssl` module today has a number of warts caused by leaking OpenSSL concepts through to the API: the new protocol classes would remove those specific concepts. 1. To provide a path for the core development team to make OpenSSL one of many possible TLS implementations, rather than requiring that it be present on a system in order for Python to have TLS support. The proposed interface is laid out below. Interfaces ---------- There are several interfaces that require standardization. Those interfaces are: 1. Configuring TLS, currently implemented by the :class:`~ssl.SSLContext` class in the :mod:`ssl` module. 1. Providing an in-memory buffer for doing in-memory encryption or decryption with no actual I/O (necessary for asynchronous I/O models), currently implemented by the :class:`~ssl.SSLObject` class in the :mod:`ssl` module. 1. Wrapping a socket object, currently implemented by the :class:`~ssl.SSLSocket` class in the :mod:`ssl` module. 1. Applying TLS configuration to the wrapping objects in (2) and (3). Currently this is also implemented by the SSLContext class in the :mod:`ssl` module. 1. Specifying TLS cipher suites. There is currently no code for doing this in the standard library: instead, the standard library uses OpenSSL cipher suite strings. 1. Specifying application-layer protocols that can be negotiated during the TLS handshake. 1. Specifying TLS versions. 1. Reporting errors to the caller, currently implemented by the :class:`~ssl.SSLError` class in the :mod:`ssl` module. 1. Specifying certificates to load, either as client or server certificates. 1. Specifying which trust database should be used to validate certificates presented by a remote peer. 1. Finding a way to get hold of these interfaces at run time. For the sake of simplicity, this PEP proposes to remove interfaces (3) and (4), and replace them by a simpler interface that returns a socket which ensures that all communication through the socket is protected by TLS. In other words, this interface treats concepts such as socket initialization, the TLS handshake, Server Name Indication (SNI), etc., as an atomic part of creating a client or server connection. However, in-memory buffers are still supported, as they are useful for asynchronous communication. Obviously, (5) doesn't require a protocol class: instead, it requires a richer API for configuring supported cipher suites that can be easily updated with supported cipher suites for different implementations. (9) is a thorny problem, because in an ideal world the private keys associated with these certificates would never end up in-memory in the Python process (that is, the TLS library would collaborate with a Hardware Security Module (HSM) to provide the private key in such a way that it cannot be extracted from process memory). Thus, we need to provide an extensible model of providing certificates that allows concrete implementations the ability to provide this higher level of security, while also allowing a lower bar for those implementations that cannot. This lower bar would be the same as the status quo: that is, the certificate may be loaded from an in-memory buffer, from a file on disk, or additionally referenced by some arbitrary ID corresponding to a system certificate store. (10) also represents an issue because different TLS implementations vary wildly in how they allow users to select trust stores. Some implementations have specific trust store formats that only they can use (such as the OpenSSL CA directory format that is created by c_rehash), and others may not allow you to specify a trust store that does not include their default trust store. On the other hand, most implementations will support some form of loading custom DER- or PEM-encoded certificates. For this reason, we need to provide a model that assumes very little about the form that trust stores take, while maintaining type-compatibility with other implementations. The sections “Certificate”, “Private Keys”, and “Trust Store” below go into more detail about how this is achieved. Finally, this API will split the responsibilities currently assumed by the :class:`~ssl.SSLContext` object: specifically, the responsibility for holding and managing configuration and the responsibility for using that configuration to build buffers or sockets. This is necessary primarily for supporting functionality like Server Name Indication (SNI). In OpenSSL (and thus in the :mod:`ssl` module), the server has the ability to modify the TLS configuration in response to the client telling the server what hostname it is trying to reach. This is mostly used to change the certificate chain so as to present the correct TLS certificate chain for the given hostname. The specific mechanism by which this is done is by returning a new :class:`~ssl.SSLContext` object with the appropriate configuration as part of a user-provided SNI callback function. This is not a model that maps well to other TLS implementations, and puts a burden on users to write callback functions. Instead, we propose that the concrete implementations handle SNI transparently for every user after receiving the relevant certificates. For this reason, we split the responsibility of :class:`~ssl.SSLContext` into two separate objects, which are each split into server and client versions. The ``TLSServerConfiguration`` and ``TLSClientConfiguration`` objects act as containers for a TLS configuration: the ClientContext and ServerContext objects are instantiated with a ``TLSClientConfiguration`` and ``TLSServerConfiguration`` object, respectively, and are used to create buffers or sockets. All four objects would be immutable. .. note:: The following API declarations uniformly use type hints to aid reading. Configuration ~~~~~~~~~~~~~ The ``TLSServerConfiguration`` and ``TLSClientConfiguration`` concrete classes define objects that can hold and manage TLS configuration. The goals of these classes are as follows: 1. To provide a method of specifying TLS configuration that avoids the risk of errors in typing (this excludes the use of a simple dictionary). 1. To provide an object that can be safely compared to other configuration objects to detect changes in TLS configuration, for use with the SNI callback. These classes are not protocol classes, primarily because they are not expected to have implementation-specific behavior. The responsibility for transforming a ``TLSServerConfiguration`` or ``TLSClientConfiguration`` object into a useful set of configurations for a given TLS implementation belongs to the Context objects discussed below. These classes have one other notable property: they are immutable. This is a desirable trait for a few reasons. The most important one is that immutability by default is a good engineering practice. As a side benefit, it allows these objects to be used as dictionary keys, which is potentially useful for specific TLS implementations and their SNI configuration. On top of this, it frees implementations from needing to worry about their configuration objects being changed under their feet, which allows them to avoid needing to carefully synchronize changes between their concrete data structures and the configuration object. These objects are extendable: that is, future releases of Python may add configuration fields to these objects as they become useful. For backwards-compatibility purposes, new fields are only appended to these objects. Existing fields will never be removed, renamed, or reordered. They are split between client and server to minimize API confusion. The ``TLSClientConfiguration`` class would be defined by the following code: .. code-block:: python class TLSClientConfiguration: __slots__ = ( "_certificate_chain", "_ciphers", "_inner_protocols", "_lowest_supported_version", "_highest_supported_version", "_trust_store", ) def __init__( self, certificate_chain: SigningChain | None = None, ciphers: Sequence[CipherSuite] | None = None, inner_protocols: Sequence[NextProtocol | bytes] | None = None, lowest_supported_version: TLSVersion | None = None, highest_supported_version: TLSVersion | None = None, trust_store: TrustStore | None = None, ) -> None: if inner_protocols is None: inner_protocols = [] self._certificate_chain = certificate_chain self._ciphers = ciphers self._inner_protocols = inner_protocols self._lowest_supported_version = lowest_supported_version self._highest_supported_version = highest_supported_version self._trust_store = trust_store @property def certificate_chain(self) -> SigningChain | None: return self._certificate_chain @property def ciphers(self) -> Sequence[CipherSuite | int] | None: return self._ciphers @property def inner_protocols(self) -> Sequence[NextProtocol | bytes]: return self._inner_protocols @property def lowest_supported_version(self) -> TLSVersion | None: return self._lowest_supported_version @property def highest_supported_version(self) -> TLSVersion | None: return self._highest_supported_version @property def trust_store(self) -> TrustStore | None: return self._trust_store The ``TLSServerConfiguration`` object is similar to the client one, except that it takes a ``Sequence[SigningChain]`` as the ``certificate_chain`` parameter. Context ~~~~~~~ We define two Context protocol classes. These protocol classes define objects that allow configuration of TLS to be applied to specific connections. They can be thought of as factories for ``TLSSocket`` and ``TLSBuffer`` objects. Unlike the current :mod:`ssl` module, we provide two context classes instead of one. Specifically, we provide the ``ClientContext`` and ``ServerContext`` classes. This simplifies the APIs (for example, there is no sense in the server providing the ``server_hostname`` parameter to :meth:`~ssl.SSLContext.wrap_socket`, but because there is only one context class that parameter is still available), and ensures that implementations know as early as possible which side of a TLS connection they will serve. Additionally, it allows implementations to opt-out of one or either side of the connection. As much as possible implementers should aim to make these classes immutable: that is, they should prefer not to allow users to mutate their internal state directly, instead preferring to create new contexts from new TLSConfiguration objects. Obviously, the protocol classes cannot enforce this constraint, and so they do not attempt to. The ``ClientContext`` protocol class has the following class definition: .. code-block:: python class ClientContext(Protocol): @abstractmethod def __init__(self, configuration: TLSClientConfiguration) -> None: """Create a new client context object from a given TLS client configuration.""" ... @property @abstractmethod def configuration(self) -> TLSClientConfiguration: """Returns the TLS client configuration that was used to create the client context.""" ... @abstractmethod def connect(self, address: tuple[str | None, int]) -> TLSSocket: """Creates a TLSSocket that behaves like a socket.socket, and contains information about the TLS exchange (cipher, negotiated_protocol, negotiated_tls_version, etc.). """ ... @abstractmethod def create_buffer(self, server_hostname: str) -> TLSBuffer: """Creates a TLSBuffer that acts as an in-memory channel, and contains information about the TLS exchange (cipher, negotiated_protocol, negotiated_tls_version, etc.).""" ... The ``ServerContext`` is similar, taking a ``TLSServerConfiguration`` instead. Socket ~~~~~~ The context can be used to create sockets, which have to follow the specification of the ``TLSSocket`` protocol class. Specifically, implementations need to implement the following: * ``recv`` and ``send`` * ``listen`` and ``accept`` * ``close`` * ``getsockname`` * ``getpeername`` They also need to implement some interfaces that give information about the TLS connection, such as: * The underlying context object that was used to create this socket * The negotiated cipher * The negotiated "next" protocol * The negotiated TLS version The following code describes these functions in more detail: .. code-block:: python class TLSSocket(Protocol): """This class implements a socket.socket-like object that creates an OS socket, wraps it in an SSL context, and provides read and write methods over that channel.""" @abstractmethod def __init__(self, *args: tuple, **kwargs: tuple) -> None: """TLSSockets should not be constructed by the user. The implementation should implement a method to construct a TLSSocket object and call it in ClientContext.connect() and ServerContext.connect().""" ... @abstractmethod def recv(self, bufsize: int) -> bytes: """Receive data from the socket. The return value is a bytes object representing the data received. Should not work before the handshake is completed.""" ... @abstractmethod def send(self, bytes: bytes) -> int: """Send data to the socket. The socket must be connected to a remote socket.""" ... @abstractmethod def close(self, force: bool = False) -> None: """Shuts down the connection and mark the socket closed. If force is True, this method should send the close_notify alert and shut down the socket without waiting for the other side. If force is False, this method should send the close_notify alert and raise the WantReadError exception until a corresponding close_notify alert has been received from the other side. In either case, this method should return WantWriteError if sending the close_notify alert currently fails.""" ... @abstractmethod def listen(self, backlog: int) -> None: """Enable a server to accept connections. If backlog is specified, it specifies the number of unaccepted connections that the system will allow before refusing new connections.""" ... @abstractmethod def accept(self) -> tuple[TLSSocket, tuple[str | None, int]]: """Accept a connection. The socket must be bound to an address and listening for connections. The return value is a pair (conn, address) where conn is a new TLSSocket object usable to send and receive data on the connection, and address is the address bound to the socket on the other end of the connection.""" ... @abstractmethod def getsockname(self) -> tuple[str | None, int]: """Return the local address to which the socket is connected.""" ... @abstractmethod def getpeercert(self) -> bytes | None: """ Return the raw DER bytes of the certificate provided by the peer during the handshake, if applicable. """ ... @abstractmethod def getpeername(self) -> tuple[str | None, int]: """Return the remote address to which the socket is connected.""" ... @property @abstractmethod def context(self) -> ClientContext | ServerContext: """The ``Context`` object this socket is tied to.""" ... @abstractmethod def cipher(self) -> CipherSuite | int | None: """ Returns the CipherSuite entry for the cipher that has been negotiated on the connection. If no connection has been negotiated, returns ``None``. If the cipher negotiated is not defined in CipherSuite, returns the 16-bit integer representing that cipher directly. """ ... @abstractmethod def negotiated_protocol(self) -> NextProtocol | bytes | None: """ Returns the protocol that was selected during the TLS handshake. This selection may have been made using ALPN or some future negotiation mechanism. If the negotiated protocol is one of the protocols defined in the ``NextProtocol`` enum, the value from that enum will be returned. Otherwise, the raw bytestring of the negotiated protocol will be returned. If ``Context.set_inner_protocols()`` was not called, if the other party does not support protocol negotiation, if this socket does not support any of the peer's proposed protocols, or if the handshake has not happened yet, ``None`` is returned. """ ... @property @abstractmethod def negotiated_tls_version(self) -> TLSVersion | None: """The version of TLS that has been negotiated on this connection.""" ... Buffer ~~~~~~ The context can also be used to create buffers, which have to follow the specification of the ``TLSBuffer`` protocol class. Specifically, implementations need to implement the following: * ``read`` and ``write`` * ``do_handshake`` * ``shutdown`` * ``process_incoming`` and ``process_outgoing`` * ``incoming_bytes_buffered`` and ``outgoing_bytes_buffered`` * ``getpeercert`` Similarly to the socket case, they also need to implement some interfaces that give information about the TLS connection, such as: * The underlying context object that was used to create this buffer * The negotiated cipher * The negotiated "next" protocol * The negotiated TLS version The following code describes these functions in more detail: .. code-block:: python class TLSBuffer(Protocol): """This class implements an in memory-channel that creates two buffers, wraps them in an SSL context, and provides read and write methods over that channel.""" @abstractmethod def read(self, amt: int, buffer: Buffer | None) -> bytes | int: """ Read up to ``amt`` bytes of data from the input buffer and return the result as a ``bytes`` instance. If an optional buffer is provided, the result is written into the buffer and the number of bytes is returned instead. Once EOF is reached, all further calls to this method return the empty byte string ``b''``. May read "short": that is, fewer bytes may be returned than were requested. Raise ``WantReadError`` or ``WantWriteError`` if there is insufficient data in either the input or output buffer and the operation would have caused data to be written or read. May raise ``RaggedEOF`` if the connection has been closed without a graceful TLS shutdown. Whether this is an exception that should be ignored or not is up to the specific application. As at any time a re-negotiation is possible, a call to ``read()`` can also cause write operations. """ ... @abstractmethod def write(self, buf: Buffer) -> int: """ Write ``buf`` in encrypted form to the output buffer and return the number of bytes written. The ``buf`` argument must be an object supporting the buffer interface. Raise ``WantReadError`` or ``WantWriteError`` if there is insufficient data in either the input or output buffer and the operation would have caused data to be written or read. In either case, users should endeavour to resolve that situation and then re-call this method. When re-calling this method users *should* re-use the exact same ``buf`` object, as some implementations require that the exact same buffer be used. This operation may write "short": that is, fewer bytes may be written than were in the buffer. As at any time a re-negotiation is possible, a call to ``write()`` can also cause read operations. """ ... @abstractmethod def do_handshake(self) -> None: """ Performs the TLS handshake. Also performs certificate validation and hostname verification. """ ... @abstractmethod def cipher(self) -> CipherSuite | int | None: """ Returns the CipherSuite entry for the cipher that has been negotiated on the connection. If no connection has been negotiated, returns ``None``. If the cipher negotiated is not defined in CipherSuite, returns the 16-bit integer representing that cipher directly. """ ... @abstractmethod def negotiated_protocol(self) -> NextProtocol | bytes | None: """ Returns the protocol that was selected during the TLS handshake. This selection may have been made using ALPN, NPN, or some future negotiation mechanism. If the negotiated protocol is one of the protocols defined in the ``NextProtocol`` enum, the value from that enum will be returned. Otherwise, the raw bytestring of the negotiated protocol will be returned. If ``Context.set_inner_protocols()`` was not called, if the other party does not support protocol negotiation, if this socket does not support any of the peer's proposed protocols, or if the handshake has not happened yet, ``None`` is returned. """ ... @property @abstractmethod def context(self) -> ClientContext | ServerContext: """ The ``Context`` object this buffer is tied to. """ ... @property @abstractmethod def negotiated_tls_version(self) -> TLSVersion | None: """ The version of TLS that has been negotiated on this connection. """ ... @abstractmethod def shutdown(self) -> None: """ Performs a clean TLS shut down. This should generally be used whenever possible to signal to the remote peer that the content is finished. """ ... @abstractmethod def process_incoming(self, data_from_network: bytes) -> None: """ Receives some TLS data from the network and stores it in an internal buffer. If the internal buffer is overfull, this method will raise ``WantReadError`` and store no data. At this point, the user must call ``read`` to remove some data from the internal buffer before repeating this call. """ ... @abstractmethod def incoming_bytes_buffered(self) -> int: """ Returns how many bytes are in the incoming buffer waiting to be processed. """ ... @abstractmethod def process_outgoing(self, amount_bytes_for_network: int) -> bytes: """ Returns the next ``amt`` bytes of data that should be written to the network from the outgoing data buffer, removing it from the internal buffer. """ ... @abstractmethod def outgoing_bytes_buffered(self) -> int: """ Returns how many bytes are in the outgoing buffer waiting to be sent. """ ... @abstractmethod def getpeercert(self) -> bytes | None: """ Return the raw DER bytes of the certificate provided by the peer during the handshake, if applicable. """ ... Cipher Suites ~~~~~~~~~~~~~ Supporting cipher suites in a truly library-agnostic fashion is a remarkably difficult undertaking. Different TLS implementations often have radically different APIs for specifying cipher suites, but more problematically these APIs frequently differ in capability as well as in style. Below are examples of different cipher suite selection APIs. These examples are not intended to obligate implementation against each API, only to illuminate the constraints imposed by each. OpenSSL ^^^^^^^ OpenSSL uses a well-known cipher string format. This format has been adopted as a configuration language by most products that use OpenSSL, including Python. This format is relatively easy to read, but has a number of downsides: it is a string, which makes it easy to provide bad inputs; it lacks much detailed validation, meaning that it is possible to configure OpenSSL in a way that doesn't allow it to negotiate any cipher at all; and it allows specifying cipher suites in a number of different ways that make it tricky to parse. The biggest problem with this format is that there is no formal specification for it, meaning that the only way to parse a given string the way OpenSSL would is to get OpenSSL to parse it. OpenSSL's cipher strings can look like this: .. code-block:: python "ECDH+AESGCM:ECDH+CHACHA20:DH+AESGCM:DH+CHACHA20:ECDH+AES256:DH+AES256:ECDH+AES128:DH+AES:RSA+AESGCM:RSA+AES:!aNULL:!eNULL:!MD5" This string demonstrates some of the complexity of the OpenSSL format. For example, it is possible for one entry to specify multiple cipher suites: the entry ``ECDH+AESGCM`` means “all ciphers suites that include both elliptic-curve Diffie-Hellman key exchange and AES in Galois Counter Mode”. More explicitly, that will expand to four cipher suites: .. code-block:: python "ECDHE-ECDSA-AES256-GCM-SHA384:ECDHE-RSA-AES256-GCM-SHA384:ECDHE-ECDSA-AES128-GCM-SHA256:ECDHE-RSA-AES128-GCM-SHA256" That makes parsing a complete OpenSSL cipher string extremely tricky. Add to the fact that there are other meta-characters, such as “!” (exclude all cipher suites that match this criterion, even if they would otherwise be included: “!MD5” means that no cipher suites using the MD5 hash algorithm should be included), “-” (exclude matching ciphers if they were already included, but allow them to be re-added later if they get included again), and “+” (include the matching ciphers, but place them at the end of the list), and you get an extremely complex format to parse. On top of this complexity it should be noted that the actual result depends on the OpenSSL version, as an OpenSSL cipher string is valid so long as it contains at least one cipher that OpenSSL recognizes. OpenSSL also uses different names for its ciphers than the names used in the relevant specifications. See the manual page for ``ciphers(1)`` for more details. The actual API inside OpenSSL for the cipher string is simple: .. code-block:: c char *cipher_list = ; int rc = SSL_CTX_set_cipher_list(context, cipher_list); This means that any format that is used by this module must be able to be converted to an OpenSSL cipher string for use with OpenSSL. Network Framework ^^^^^^^^^^^^^^^^^ Network Framework is the macOS (10.15+) system TLS library. This library is substantially more restricted than OpenSSL in many ways, as it has a much more restricted class of users. One of these substantial restrictions is in controlling supported cipher suites. Ciphers in Network Framework are represented by a Objective-C ``uint16_t`` enum. This enum has one entry per cipher suite, with no aggregate entries, meaning that it is not possible to reproduce the meaning of an OpenSSL cipher string like ``“ECDH+AESGCM”`` without hand-coding which categories each enum member falls into. However, the names of most of the enum members are in line with the formal names of the cipher suites: that is, the cipher suite that OpenSSL calls ``“ECDHE-ECDSA-AES256-GCM-SHA384”`` is called ``“tls_ciphersuite_ECDHE_ECDSA_WITH_AES_256_GCM_SHA384”`` in Network Framework. The API for configuring cipher suites inside Network Framework is simple: .. code-block:: c void sec_protocol_options_append_tls_ciphersuite(sec_protocol_options_t options, tls_ciphersuite_t ciphersuite); SChannel ^^^^^^^^ SChannel is the Windows system TLS library. SChannel has extremely restrictive support for controlling available TLS cipher suites, and additionally adopts a third method of expressing what TLS cipher suites are supported. Specifically, SChannel defines a set of ``ALG_ID`` constants (C unsigned ints). Each of these constants does not refer to an entire cipher suite, but instead an individual algorithm. Some examples are ``CALG_3DES`` and ``CALG_AES_256``, which refer to the bulk encryption algorithm used in a cipher suite, ``CALG_ECDH_EPHEM`` and ``CALG_RSA_KEYX`` which refer to part of the key exchange algorithm used in a cipher suite, ``CALG_SHA_256`` and ``CALG_SHA_384`` which refer to the message authentication code used in a cipher suite, and ``CALG_ECDSA`` and ``CALG_RSA_SIGN`` which refer to the signing portions of the key exchange algorithm. In earlier versions of the SChannel API, these constants were used to define the algorithms that could be used. The latest version, however, uses these constants to prohibit which algorithms can be used. This can be thought of as the half of OpenSSL's functionality that Network Framework doesn't have: Network Framework only allows specifying exact cipher suites (and a limited number of pre-defined cipher suite groups), whereas SChannel only allows specifying parts of the cipher suite, while OpenSSL allows both. Determining which cipher suites are allowed on a given connection is done by providing a pointer to an array of these ``ALG_ID`` constants. This means that any suitable API must allow the Python code to determine which ``ALG_ID`` constants must be provided. Network Security Services (NSS) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ NSS is Mozilla's crypto and TLS library. It's used in Firefox, Thunderbird, and as an alternative to OpenSSL in multiple libraries, e.g. curl. By default, NSS comes with secure configuration of allowed ciphers. On some platforms such as Fedora, the list of enabled ciphers is globally configured in a system policy. Generally, applications should not modify cipher suites unless they have specific reasons to do so. NSS has both process global and per-connection settings for cipher suites. It does not have a concept of :class:`~ssl.SSLContext` like OpenSSL. A :class:`~ssl.SSLContext`-like behavior can be easily emulated. Specifically, ciphers can be enabled or disabled globally with ``SSL_CipherPrefSetDefault(PRInt32 cipher, PRBool enabled)``, and ``SSL_CipherPrefSet(PRFileDesc *fd, PRInt32 cipher, PRBool enabled)`` for a connection. The cipher ``PRInt32`` number is a signed 32-bit integer that directly corresponds to an registered IANA id, e.g. ``0x1301`` is ``TLS_AES_128_GCM_SHA256``. Contrary to OpenSSL, the preference order of ciphers is fixed and cannot be modified at runtime. Like Network Framework, NSS has no API for aggregated entries. Some consumers of NSS have implemented custom mappings from OpenSSL cipher names and rules to NSS ciphers, e.g. ``mod_nss``. Proposed Interface ^^^^^^^^^^^^^^^^^^ The proposed interface for the new module is influenced by the combined set of limitations of the above implementations. Specifically, as every implementation except OpenSSL requires that each individual cipher be provided, there is no option but to provide that lowest common denominator approach. The simplest approach is to provide an enumerated type that includes a large subset of the cipher suites defined for TLS. The values of the enum members will be their two-octet cipher identifier as used in the TLS handshake, stored as a 16 bit integer. The names of the enum members will be their IANA-registered cipher suite names. As of now, the `IANA cipher suite registry `_ contains over 320 cipher suites. A large portion of the cipher suites are irrelevant for TLS connections to network services. Other suites specify deprecated and insecure algorithms that are no longer provided by recent versions of implementations. The enum contains the five fixed cipher suites defined for TLS v1.3. For TLS v1.2, it only contains the cipher suites that correspond to the TLS v1.3 cipher suites, with ECDHE key exchange (for perfect forward secrecy) and ECDSA or RSA signatures, which are an additional ten cipher suites. In addition to this enum, the interface defines a default cipher suite list for TLS v1.2, which includes only those defined cipher suites based on AES-GCM or ChaCha20-Poly1305. The default cipher suite list for TLS v1.3 will comprise the five cipher suites defined in the specification. The current enum is quite restricted, including only cipher suites that provide forward secrecy. Because the enum doesn't contain every defined cipher, and also to allow for forward-looking applications, all parts of this API that accept ``CipherSuite`` objects will also accept raw 16-bit integers directly. .. code-block:: python class CipherSuite(IntEnum): """ Known cipher suites. See: """ TLS_AES_128_GCM_SHA256 = 0x1301 TLS_AES_256_GCM_SHA384 = 0x1302 TLS_CHACHA20_POLY1305_SHA256 = 0x1303 TLS_AES_128_CCM_SHA256 = 0x1304 TLS_AES_128_CCM_8_SHA256 = 0x1305 TLS_ECDHE_ECDSA_WITH_AES_128_GCM_SHA256 = 0xC02B TLS_ECDHE_ECDSA_WITH_AES_256_GCM_SHA384 = 0xC02C TLS_ECDHE_RSA_WITH_AES_128_GCM_SHA256 = 0xC02F TLS_ECDHE_RSA_WITH_AES_256_GCM_SHA384 = 0xC030 TLS_ECDHE_ECDSA_WITH_AES_128_CCM = 0xC0AC TLS_ECDHE_ECDSA_WITH_AES_256_CCM = 0xC0AD TLS_ECDHE_ECDSA_WITH_AES_128_CCM_8 = 0xC0AE TLS_ECDHE_ECDSA_WITH_AES_256_CCM_8 = 0xC0AF TLS_ECDHE_RSA_WITH_CHACHA20_POLY1305_SHA256 = 0xCCA8 TLS_ECDHE_ECDSA_WITH_CHACHA20_POLY1305_SHA256 = 0xCCA9 For Network Framework, these enum members directly refer to the values of the cipher suite constants. For example, Network Framework defines the cipher suite enum member ``tls_ciphersuite_ECDHE_ECDSA_WITH_AES_256_GCM_SHA384`` as having the value ``0xC02C``. Not coincidentally, that is identical to its value in the above enum. This makes mapping between Network Framework and the above enum very easy indeed. For SChannel there is no easy direct mapping, due to the fact that SChannel configures ciphers, instead of cipher suites. This represents an ongoing concern with SChannel, which is that it is very difficult to configure in a specific manner compared to other TLS implementations. For the purposes of this PEP, any SChannel implementation will need to determine which ciphers to choose based on the enum members. This may be more open than the actual cipher suite list actually wants to allow, or it may be more restrictive, depending on the choices of the implementation. This PEP recommends that it be more restrictive, but of course this cannot be enforced. Finally, we expect that for most users, secure defaults will be enough. When specifying no list of ciphers, the implementations should use secure defaults (possibly derived from system recommended settings). Protocol Negotiation ~~~~~~~~~~~~~~~~~~~~ ALPN allows for protocol negotiation as part of the HTTP/2 handshake. While ALPN is at a fundamental level built on top of bytestrings, string-based APIs are frequently problematic as they allow for errors in typing that can be hard to detect. For this reason, this module will define a type that protocol negotiation implementations can pass and be passed. This type would wrap a bytestring to allow for aliases for well-known protocols. This allows us to avoid the problems inherent in typos for well-known protocols, while allowing the full extensibility of the protocol negotiation layer if needed by letting users pass byte strings directly. .. code-block:: python class NextProtocol(Enum): """The underlying negotiated ("next") protocol.""" H2 = b"h2" H2C = b"h2c" HTTP1 = b"http/1.1" WEBRTC = b"webrtc" C_WEBRTC = b"c-webrtc" FTP = b"ftp" STUN = b"stun.nat-discovery" TURN = b"stun.turn" TLS Versions ~~~~~~~~~~~~ It is often useful to be able to restrict the versions of TLS you're willing to support. There are many security advantages in refusing to use old versions of TLS, and some misbehaving servers will mishandle TLS clients advertising support for newer versions. The following enumerated type can be used to gate TLS versions. Forward-looking applications should almost never set a maximum TLS version unless they absolutely must, as a TLS implementation that is newer than the Python that uses it may support TLS versions that are not in this enumerated type. Additionally, this enumerated type defines two additional flags that can always be used to request either the lowest or highest TLS version supported by an implementation. As for cipher suites, we expect that for most users, secure defaults will be enough. When specifying no list of TLS versions, the implementations should use secure defaults (possibly derived from system recommended settings). .. code-block:: python class TLSVersion(Enum): """ TLS versions. The `MINIMUM_SUPPORTED` and `MAXIMUM_SUPPORTED` variants are "open ended", and refer to the "lowest mutually supported" and "highest mutually supported" TLS versions, respectively. """ MINIMUM_SUPPORTED = "MINIMUM_SUPPORTED" TLSv1_2 = "TLSv1.2" TLSv1_3 = "TLSv1.3" MAXIMUM_SUPPORTED = "MAXIMUM_SUPPORTED" Errors ~~~~~~ This module would define four base classes for use with error handling. Unlike many of the other classes defined here, these classes are not abstract, as they have no behavior. They exist simply to signal certain common behaviors. TLS implementations should subclass these exceptions in their own packages, but needn't define any behavior for them. In general, concrete implementations should subclass these exceptions rather than throw them directly. This makes it moderately easier to determine which concrete TLS implementation is in use during debugging of unexpected errors. However, this is not mandatory. The definitions of the errors are below: .. code-block:: python class TLSError(Exception): """ The base exception for all TLS related errors from any implementation. Catching this error should be sufficient to catch *all* TLS errors, regardless of what implementation is used. """ class WantWriteError(TLSError): """ A special signaling exception used only when non-blocking or buffer-only I/O is used. This error signals that the requested operation cannot complete until more data is written to the network, or until the output buffer is drained. This error is should only be raised when it is completely impossible to write any data. If a partial write is achievable then this should not be raised. """ class WantReadError(TLSError): """ A special signaling exception used only when non-blocking or buffer-only I/O is used. This error signals that the requested operation cannot complete until more data is read from the network, or until more data is available in the input buffer. This error should only be raised when it is completely impossible to write any data. If a partial write is achievable then this should not be raised. """ class RaggedEOF(TLSError): """A special signaling exception used when a TLS connection has been closed gracelessly: that is, when a TLS CloseNotify was not received from the peer before the underlying TCP socket reached EOF. This is a so-called "ragged EOF". This exception is not guaranteed to be raised in the face of a ragged EOF: some implementations may not be able to detect or report the ragged EOF. This exception is not always a problem. Ragged EOFs are a concern only when protocols are vulnerable to length truncation attacks. Any protocol that can detect length truncation attacks at the application layer (e.g. HTTP/1.1 and HTTP/2) is not vulnerable to this kind of attack and so can ignore this exception. """ class ConfigurationError(TLSError): """An special exception that implementations can use when the provided configuration uses features not supported by that implementation.""" Certificates ~~~~~~~~~~~~ This module would define a concrete certificate class. This class would have almost no behavior, as the goal of this module is not to provide all possible relevant cryptographic functionality that could be provided by X.509 certificates. Instead, all we need is the ability to signal the source of a certificate to a concrete implementation. For that reason, this certificate class defines three attributes, corresponding to the three envisioned constructors: certificates from files, certificates from memory, or certificates from arbitrary identifiers. It is possible that implementations do not support all of these constructors, and they can communicate this to users as described in the “Runtime” section below. Certificates from arbitrary identifiers, in particular, are expected to be useful primarily to users seeking to build integrations on top of HSMs, TPMs, SSMs, and similar. Specifically, this class does not parse any provided input to validate that it is a correct certificate, and also does not provide any form of introspection into a particular certificate. TLS implementations are not required to provide such introspection either. Peer certificates that are received during the handshake are provided as raw DER bytes. .. code-block:: python class Certificate: """Object representing a certificate used in TLS.""" __slots__ = ( "_buffer", "_path", "_id", ) def __init__( self, buffer: bytes | None = None, path: os.PathLike[str] | None = None, id: bytes | None = None ): """ Creates a Certificate object from a path, buffer, or ID. If none of these is given, an exception is raised. """ if buffer is None and path is None and id is None: raise ValueError("Certificate cannot be empty.") self._buffer = buffer self._path = path self._id = id @classmethod def from_buffer(cls, buffer: bytes) -> Certificate: """ Creates a Certificate object from a byte buffer. This byte buffer may be either PEM-encoded or DER-encoded. If the buffer is PEM encoded it *must* begin with the standard PEM preamble (a series of dashes followed by the ASCII bytes "BEGIN CERTIFICATE" and another series of dashes). In the absence of that preamble, the implementation may assume that the certificate is DER-encoded instead. """ return cls(buffer=buffer) @classmethod def from_file(cls, path: os.PathLike[str]) -> Certificate: """ Creates a Certificate object from a file on disk. The file on disk should contain a series of bytes corresponding to a certificate that may be either PEM-encoded or DER-encoded. If the bytes are PEM encoded it *must* begin with the standard PEM preamble (a series of dashes followed by the ASCII bytes "BEGIN CERTIFICATE" and another series of dashes). In the absence of that preamble, the implementation may assume that the certificate is DER-encoded instead. """ return cls(path=path) @classmethod def from_id(cls, id: bytes) -> Certificate: """ Creates a Certificate object from an arbitrary identifier. This may be useful for implementations that rely on system certificate stores. """ return cls(id=id) Private Keys ~~~~~~~~~~~~ This module would define a concrete private key class. Much like the ``Certificate`` class, this class has three attributes to correspond to the three constructors, and further has all the caveats of the ``Certificate`` class. .. code-block:: python class PrivateKey: """Object representing a private key corresponding to a public key for a certificate used in TLS.""" __slots__ = ( "_buffer", "_path", "_id", ) def __init__( self, buffer: bytes | None = None, path: os.PathLike | None = None, id: bytes | None = None ): """ Creates a PrivateKey object from a path, buffer, or ID. If none of these is given, an exception is raised. """ if buffer is None and path is None and id is None: raise ValueError("PrivateKey cannot be empty.") self._buffer = buffer self._path = path self._id = id @classmethod def from_buffer(cls, buffer: bytes) -> PrivateKey: """ Creates a PrivateKey object from a byte buffer. This byte buffer may be either PEM-encoded or DER-encoded. If the buffer is PEM encoded it *must* begin with the standard PEM preamble (a series of dashes followed by the ASCII bytes "BEGIN", the key type, and another series of dashes). In the absence of that preamble, the implementation may assume that the private key is DER-encoded instead. """ return cls(buffer=buffer) @classmethod def from_file(cls, path: os.PathLike) -> PrivateKey: """ Creates a PrivateKey object from a file on disk. The file on disk should contain a series of bytes corresponding to a certificate that may be either PEM-encoded or DER-encoded. If the bytes are PEM encoded it *must* begin with the standard PEM preamble (a series of dashes followed by the ASCII bytes "BEGIN", the key type, and another series of dashes). In the absence of that preamble, the implementation may assume that the certificate is DER-encoded instead. """ return cls(path=path) @classmethod def from_id(cls, id: bytes) -> PrivateKey: """ Creates a PrivateKey object from an arbitrary identifier. This may be useful for implementations that rely on system private key stores. """ return cls(id=id) Signing Chain ~~~~~~~~~~~~~ In order to authenticate themselves, TLS participants need to provide a leaf certificate with a chain leading up to some root certificate that is trusted by the other side. Servers always need to authenticate themselves to clients, but clients can also authenticate themselves to servers during client authentication. Additionally, the leaf certificate must be accompanied by a private key, which can either be stored in a separate object, or together with the leaf certificate itself. This module defines the collection of these objects as a ``SigningChain`` as detailed below: .. code-block:: python class SigningChain: """Object representing a certificate chain used in TLS.""" leaf: tuple[Certificate, PrivateKey | None] chain: list[Certificate] def __init__( self, leaf: tuple[Certificate, PrivateKey | None], chain: Sequence[Certificate] | None = None, ): """Initializes a SigningChain object.""" self.leaf = leaf if chain is None: chain = [] self.chain = list(chain) As shown in the configuration classes above, a client can have one signing chain in the case of client authentication or none otherwise. A server can have a sequence of signing chains, which is useful when it is responsible for multiple domains. Trust Store ~~~~~~~~~~~ As discussed above, loading a trust store represents an issue because different TLS implementations vary wildly in how they allow users to select trust stores. For this reason, we need to provide a model that assumes very little about the form that trust stores take. This problem is the same as the one that the ``Certificate`` and ``PrivateKey`` types need to solve. For this reason, we use the exact same model, by creating a concrete class that captures the various means of how users could define a trust store. A given TLS implementation is not required to handle all possible trust stores. However, it is strongly recommended that a given TLS implementation handles the ``system`` constructor if at all possible, as this is the most common validation trust store that is used. TLS implementations can communicate unsupported options as described in the “Runtime” section below. .. code-block:: python class TrustStore: """ The trust store that is used to verify certificate validity. """ __slots__ = ( "_buffer", "_path", "_id", ) def __init__( self, buffer: bytes | None = None, path: os.PathLike | None = None, id: bytes | None = None ): """ Creates a TrustStore object from a path, buffer, or ID. If none of these is given, the default system trust store is used. """ self._buffer = buffer self._path = path self._id = id @classmethod def system(cls) -> TrustStore: """ Returns a TrustStore object that represents the system trust database. """ return cls() @classmethod def from_buffer(cls, buffer: bytes) -> TrustStore: """ Initializes a trust store from a buffer of PEM-encoded certificates. """ return cls(buffer=buffer) @classmethod def from_file(cls, path: os.PathLike) -> TrustStore: """ Initializes a trust store from a single file containing PEMs. """ return cls(path=path) @classmethod def from_id(cls, id: bytes) -> TrustStore: """ Initializes a trust store from an arbitrary identifier. """ return cls(id=id) Runtime Access ~~~~~~~~~~~~~~ A not-uncommon use case is for library users to want to specify the TLS implementation to use while allowing the library to configure the details of the actual TLS connection. For example, users of :pypi:`requests` may want to be able to select between OpenSSL or a platform-native solution on Windows and macOS, or between OpenSSL and NSS on some Linux platforms. These users, however, may not care about exactly how their TLS configuration is done. This poses two problems: given an arbitrary concrete implementation, how can a library: * Work out whether the implementation supports particular constructors for certificates or trust stores (e.g. from arbitrary identifiers)? * Get the correct types for the two context classes? Constructing certificate and trust store objects should be possible outside of the implementation. Therefore, the implementations need to provide a way for users to verify whether the implementation is compatible with user-constructed certificates and trust stores. Therefore, each implementation should implement a ``validate_config`` method that takes a ``TLSClientConfiguration`` or ``TLSServerConfiguration`` object and raises an exception if unsupported constructors were used. For the types, there are two options: either all concrete implementations can be required to fit into a specific naming scheme, or we can provide an API that makes it possible to grab these objects. This PEP proposes that we use the second approach. This grants the greatest freedom to concrete implementations to structure their code as they see fit, requiring only that they provide a single object that has the appropriate properties in place. Users can then pass this implementation object to libraries that support it, and those libraries can take care of configuring and using the concrete implementation. All concrete implementations must provide a method of obtaining a ``TLSImplementation`` object. The ``TLSImplementation`` object can be a global singleton or can be created by a callable if there is an advantage in doing that. The ``TLSImplementation`` object has the following definition: .. code-block:: python class TLSImplementation(Generic[_ClientContext, _ServerContext]): __slots__ = ( "_client_context", "_server_context", "_validate_config", ) def __init__( self, client_context: type[_ClientContext], server_context: type[_ServerContext], validate_config: Callable[[TLSClientConfiguration | TLSServerConfiguration], None], ) -> None: self._client_context = client_context self._server_context = server_context self._validate_config = validate_config The first two properties must provide the concrete implementation of the relevant Protocol class. For example, for the client context: .. code-block:: python @property def client_context(self) -> type[_ClientContext]: """The concrete implementation of the PEP 543 Client Context object, if this TLS implementation supports being the client on a TLS connection. """ return self._client_context This ensures that code like this will work for any implementation: .. code-block:: python client_config = TLSClientConfiguration() client_context = implementation.client_context(client_config) The third property must provide a function that verifies whether a given TLS configuration contains implementation-compatible certificates, private keys, and a trust store: .. code-block:: python @property def validate_config(self) -> Callable[[TLSClientConfiguration | TLSServerConfiguration], None]: """A function that reveals whether this TLS implementation supports a particular TLS configuration. """ return self._validate_config Note that this function only needs to verify that supported constructors were used for the certificates, private keys, and trust store. It does not need to parse or retrieve the objects to validate them further. Insecure Usage -------------- All of the above assumes that users want to use the module in a secure way. Sometimes, users want to do imprudent things like disable certificate validation for testing purposes. To this end, we propose a separate ``insecure`` module that allows users to do this. This module contains insecure variants of the configuration, context, and implementation objects, which allow to disable certificate validation as well as the server hostname check. This functionality is placed in a separate module to make it as hard as possible for legitimate users to accidentally use the insecure functionality. Additionally, it defines a new warning called ``SecurityWarning``, and loudly warns at every step of the way when trying to create an insecure connection. This module is only intended for testing purposes. In real-world situations where a user wants to connect to some IoT device which only has a self-signed certificate, it is strongly recommended to add this certificate into a custom trust store, rather than using the insecure module to disable certificate validation. Changes to the Standard Library =============================== The portions of the standard library that interact with TLS should be revised to use these Protocol classes. This will allow them to function with other TLS implementations. This includes the following modules: * :mod:`asyncio` * :mod:`ftplib` * :mod:`http` * :mod:`imaplib` * :mod:`nntplib` * :mod:`poplib` * :mod:`smtplib` * :mod:`urllib` Migration of the ssl module --------------------------- Naturally, we will need to extend the :mod:`ssl` module itself to conform to these Protocol classes. This extension will take the form of new classes, potentially in an entirely new module. This will allow applications that take advantage of the current :mod:`ssl` module to continue to do so, while enabling the new APIs for applications and libraries that want to use them. In general, migrating from the :mod:`ssl` module to the new Protocol classes is not expected to be one-to-one. This is normally acceptable: most tools that use the :mod:`ssl` module hide it from the user, and so refactoring to use the new module should be invisible. However, a specific problem comes from libraries or applications that leak exceptions from the :mod:`ssl` module, either as part of their defined API or by accident (which is easily done). Users of those tools may have written code that tolerates and handles exceptions from the :mod:`ssl` module being raised: migrating to the protocol classes presented here would potentially cause the exceptions defined above to be thrown instead, and existing ``except`` blocks will not catch them. For this reason, part of the migration of the :mod:`ssl` module would require that the exceptions in the :mod:`ssl` module alias those defined above. That is, they would require the following statements to all succeed: .. code-block:: python assert ssl.SSLError is tls.TLSError assert ssl.SSLWantReadError is tls.WantReadError assert ssl.SSLWantWriteError is tls.WantWriteError The exact mechanics of how this will be done are beyond the scope of this PEP, as they are made more complex due to the fact that the current ssl exceptions are defined in C code, but more details can be found in `an email sent to the Security-SIG by Christian Heimes `_. Future ====== Major future TLS features may require revisions of these protocol classes. These revisions should be made cautiously: many implementations may not be able to move forward swiftly, and will be invalidated by changes in these protocol classes. This is acceptable, but wherever possible features that are specific to individual implementations should not be added to the protocol classes. The protocol classes should restrict themselves to high-level descriptions of IETF-specified features. However, well-justified extensions to this API absolutely should be made. The focus of this API is to provide a unifying lowest-common-denominator configuration option for the Python community. TLS is not a static target, and as TLS evolves so must this API. Credits ======= This PEP is adapted substantially from :pep:`543`, which was withdrawn in 2020. :pep:`543` was authored by Cory Benfield and Christian Heimes, and received extensive review from a number of individuals in the community who have substantially helped shape it. Detailed review for both :pep:`543` and this PEP was provided by: * Alex Chan * Alex Gaynor * Antoine Pitrou * Ashwini Oruganti * Donald Stufft * Ethan Furman * Glyph * Hynek Schlawack * Jim J Jewett * Nathaniel J. Smith * Alyssa Coghlan * Paul Kehrer * Steve Dower * Steven Fackler * Wes Turner * Will Bond * Cory Benfield * Marc-André Lemburg * Seth M. Larson * Victor Stinner * Ronald Oussoren Further review of :pep:`543` was provided by the Security-SIG and python-ideas mailing lists. Copyright ========= This document is placed in the public domain or under the CC0-1.0-Universal license, whichever is more permissive.