openssl/doc/man3/SSL_read_early_data.pod

300 lines
15 KiB
Text
Raw Normal View History

=pod
=head1 NAME
SSL_set_max_early_data,
SSL_CTX_set_max_early_data,
SSL_get_max_early_data,
SSL_CTX_get_max_early_data,
SSL_SESSION_get_max_early_data,
SSL_SESSION_set_max_early_data,
SSL_write_early_data,
SSL_read_early_data,
SSL_get_early_data_status
- functions for sending and receiving early data
=head1 SYNOPSIS
#include <openssl/ssl.h>
int SSL_CTX_set_max_early_data(SSL_CTX *ctx, uint32_t max_early_data);
uint32_t SSL_CTX_get_max_early_data(const SSL_CTX *ctx);
int SSL_set_max_early_data(SSL *s, uint32_t max_early_data);
uint32_t SSL_get_max_early_data(const SSL *s);
uint32_t SSL_SESSION_get_max_early_data(const SSL_SESSION *s);
int SSL_SESSION_set_max_early_data(SSL_SESSION *s, uint32_t max_early_data);
int SSL_write_early_data(SSL *s, const void *buf, size_t num, size_t *written);
int SSL_read_early_data(SSL *s, void *buf, size_t num, size_t *readbytes);
int SSL_get_early_data_status(const SSL *s);
=head1 DESCRIPTION
These functions are used to send and receive early data where TLSv1.3 has been
negotiated. Early data can be sent by the client immediately after its initial
ClientHello without having to wait for the server to complete the handshake.
Early data can only be sent if a session has previously been established with
the server, and the server is known to support it. Additionally these functions
can be used to send data from the server to the client when the client has not
yet completed the authentication stage of the handshake.
Early data has weaker security properties than other data sent over an SSL/TLS
connection. In particular the data does not have forward secrecy. There are also
additional considerations around replay attacks (see L<REPLAY PROTECTION>
below). For these reasons extreme care should be exercised when using early
data. For specific details, consult the TLS 1.3 specification.
When a server receives early data it may opt to immediately respond by sending
application data back to the client. Data sent by the server at this stage is
done before the full handshake has been completed. Specifically the client's
authentication messages have not yet been received, i.e. the client is
unauthenticated at this point and care should be taken when using this
capability.
A server or client can determine whether the full handshake has been completed
or not by calling L<SSL_is_init_finished(3)>.
On the client side, the function SSL_SESSION_get_max_early_data() can be used to
determine if a session established with a server can be used to send early data.
If the session cannot be used then this function will return 0. Otherwise it
will return the maximum number of early data bytes that can be sent.
The function SSL_SESSION_set_max_early_data() sets the maximum number of early
data bytes that can be sent for a session. This would typically be used when
creating a PSK session file (see L<SSL_CTX_set_psk_use_session_callback(3)>). If
using a ticket based PSK then this is set automatically to the value provided by
the server.
A client uses the function SSL_write_early_data() to send early data. This
function is similar to the L<SSL_write_ex(3)> function, but with the following
differences. See L<SSL_write_ex(3)> for information on how to write bytes to
the underlying connection, and how to handle any errors that may arise. This
page describes the differences between SSL_write_early_data() and
L<SSL_write_ex(3)>.
When called by a client, SSL_write_early_data() must be the first IO function
called on a new connection, i.e. it must occur before any calls to
L<SSL_write_ex(3)>, L<SSL_read_ex(3)>, L<SSL_connect(3)>, L<SSL_do_handshake(3)>
or other similar functions. It may be called multiple times to stream data to
the server, but the total number of bytes written must not exceed the value
returned from SSL_SESSION_get_max_early_data(). Once the initial
SSL_write_early_data() call has completed successfully the client may interleave
calls to L<SSL_read_ex(3)> and L<SSL_read(3)> with calls to
SSL_write_early_data() as required.
If SSL_write_early_data() fails you should call L<SSL_get_error(3)> to determine
the correct course of action, as for L<SSL_write_ex(3)>.
When the client no longer wishes to send any more early data then it should
complete the handshake by calling a function such as L<SSL_connect(3)> or
L<SSL_do_handshake(3)>. Alternatively you can call a standard write function
such as L<SSL_write_ex(3)>, which will transparently complete the connection and
write the requested data.
A server may choose to ignore early data that has been sent to it. Once the
connection has been completed you can determine whether the server accepted or
rejected the early data by calling SSL_get_early_data_status(). This will return
SSL_EARLY_DATA_ACCEPTED if the data was accepted, SSL_EARLY_DATA_REJECTED if it
was rejected or SSL_EARLY_DATA_NOT_SENT if no early data was sent. This function
may be called by either the client or the server.
A server uses the SSL_read_early_data() function to receive early data on a
Do not set a nonzero default max_early_data When early data support was first added, this seemed like a good idea, as it would allow applications to just add SSL_read_early_data() calls as needed and have things "Just Work". However, for applications that do not use TLS 1.3 early data, there is a negative side effect. Having a nonzero max_early_data in a SSL_CTX (and thus, SSL objects derived from it) means that when generating a session ticket, tls_construct_stoc_early_data() will indicate to the client that the server supports early data. This is true, in that the implementation of TLS 1.3 (i.e., OpenSSL) does support early data, but does not necessarily indicate that the server application supports early data, when the default value is nonzero. In this case a well-intentioned client would send early data along with its resumption attempt, which would then be ignored by the server application, a waste of network bandwidth. Since, in order to successfully use TLS 1.3 early data, the application must introduce calls to SSL_read_early_data(), it is not much additional burden to require that the application also calls SSL_{CTX_,}set_max_early_data() in order to enable the feature; doing so closes this scenario where early data packets would be sent on the wire but ignored. Update SSL_read_early_data.pod accordingly, and make s_server and our test programs into applications that are compliant with the new requirements on applications that use early data. Fixes #4725 Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/5483)
2018-02-28 20:49:59 +00:00
connection for which early data has been enabled using
SSL_CTX_set_max_early_data() or SSL_set_max_early_data(). As for
SSL_write_early_data(), this must be the first IO function
called on a connection, i.e. it must occur before any calls to
L<SSL_write_ex(3)>, L<SSL_read_ex(3)>, L<SSL_accept(3)>, L<SSL_do_handshake(3)>,
or other similar functions.
SSL_read_early_data() is similar to L<SSL_read_ex(3)> with the following
differences. Refer to L<SSL_read_ex(3)> for full details.
SSL_read_early_data() may return 3 possible values:
=over 4
=item SSL_READ_EARLY_DATA_ERROR
This indicates an IO or some other error occurred. This should be treated in the
same way as a 0 return value from L<SSL_read_ex(3)>.
=item SSL_READ_EARLY_DATA_SUCCESS
This indicates that early data was successfully read. This should be treated in
the same way as a 1 return value from L<SSL_read_ex(3)>. You should continue to
call SSL_read_early_data() to read more data.
=item SSL_READ_EARLY_DATA_FINISH
This indicates that no more early data can be read. It may be returned on the
first call to SSL_read_early_data() if the client has not sent any early data,
or if the early data was rejected.
=back
Once the initial SSL_read_early_data() call has completed successfully (i.e. it
has returned SSL_READ_EARLY_DATA_SUCCESS or SSL_READ_EARLY_DATA_FINISH) then the
server may choose to write data immediately to the unauthenticated client using
SSL_write_early_data(). If SSL_read_early_data() returned
SSL_READ_EARLY_DATA_FINISH then in some situations (e.g. if the client only
supports TLSv1.2) the handshake may have already been completed and calls
to SSL_write_early_data() are not allowed. Call L<SSL_is_init_finished(3)> to
determine whether the handshake has completed or not. If the handshake is still
in progress then the server may interleave calls to SSL_write_early_data() with
calls to SSL_read_early_data() as required.
Servers must not call L<SSL_read_ex(3)>, L<SSL_read(3)>, L<SSL_write_ex(3)> or
L<SSL_write(3)> until SSL_read_early_data() has returned with
SSL_READ_EARLY_DATA_FINISH. Once it has done so the connection to the client
still needs to be completed. Complete the connection by calling a function such
as L<SSL_accept(3)> or L<SSL_do_handshake(3)>. Alternatively you can call a
standard read function such as L<SSL_read_ex(3)>, which will transparently
complete the connection and read the requested data. Note that it is an error to
attempt to complete the connection before SSL_read_early_data() has returned
SSL_READ_EARLY_DATA_FINISH.
Only servers may call SSL_read_early_data().
Calls to SSL_read_early_data() may, in certain circumstances, complete the
connection immediately without further need to call a function such as
L<SSL_accept(3)>. This can happen if the client is using a protocol version less
than TLSv1.3. Applications can test for this by calling
L<SSL_is_init_finished(3)>. Alternatively, applications may choose to call
L<SSL_accept(3)> anyway. Such a call will successfully return immediately with no
further action taken.
When a session is created between a server and a client the server will specify
the maximum amount of any early data that it will accept on any future
Do not set a nonzero default max_early_data When early data support was first added, this seemed like a good idea, as it would allow applications to just add SSL_read_early_data() calls as needed and have things "Just Work". However, for applications that do not use TLS 1.3 early data, there is a negative side effect. Having a nonzero max_early_data in a SSL_CTX (and thus, SSL objects derived from it) means that when generating a session ticket, tls_construct_stoc_early_data() will indicate to the client that the server supports early data. This is true, in that the implementation of TLS 1.3 (i.e., OpenSSL) does support early data, but does not necessarily indicate that the server application supports early data, when the default value is nonzero. In this case a well-intentioned client would send early data along with its resumption attempt, which would then be ignored by the server application, a waste of network bandwidth. Since, in order to successfully use TLS 1.3 early data, the application must introduce calls to SSL_read_early_data(), it is not much additional burden to require that the application also calls SSL_{CTX_,}set_max_early_data() in order to enable the feature; doing so closes this scenario where early data packets would be sent on the wire but ignored. Update SSL_read_early_data.pod accordingly, and make s_server and our test programs into applications that are compliant with the new requirements on applications that use early data. Fixes #4725 Reviewed-by: Matt Caswell <matt@openssl.org> (Merged from https://github.com/openssl/openssl/pull/5483)
2018-02-28 20:49:59 +00:00
connection attempt. By default the server does not accept early data; a
server may indicate support for early data by calling
SSL_CTX_set_max_early_data() or
SSL_set_max_early_data() to set it for the whole SSL_CTX or an individual SSL
object respectively. The B<max_early_data> parameter specifies the maximum
amount of early data in bytes that is permitted to be sent on a single
connection. Similarly the SSL_CTX_get_max_early_data() and
SSL_get_max_early_data() functions can be used to obtain the current maximum
early data settings for the SSL_CTX and SSL objects respectively. Generally a
server application will either use both of SSL_read_early_data() and
SSL_CTX_set_max_early_data() (or SSL_set_max_early_data()), or neither of them,
since there is no practical benefit from using only one of them. If the maximum
early data setting for a server is non-zero then replay protection is
automatically enabled (see L</REPLAY PROTECTION> below).
In the event that the current maximum early data setting for the server is
different to that originally specified in a session that a client is resuming
with then the lower of the two values will apply.
=head1 NOTES
The whole purpose of early data is to enable a client to start sending data to
the server before a full round trip of network traffic has occurred. Application
developers should ensure they consider optimisation of the underlying TCP socket
to obtain a performant solution. For example Nagle's algorithm is commonly used
by operating systems in an attempt to avoid lots of small TCP packets. In many
scenarios this is beneficial for performance, but it does not work well with the
early data solution as implemented in OpenSSL. In Nagle's algorithm the OS will
buffer outgoing TCP data if a TCP packet has already been sent which we have not
yet received an ACK for from the peer. The buffered data will only be
transmitted if enough data to fill an entire TCP packet is accumulated, or if
the ACK is received from the peer. The initial ClientHello will be sent in the
first TCP packet along with any data from the first call to
SSL_write_early_data(). If the amount of data written will exceed the size of a
single TCP packet, or if there are more calls to SSL_write_early_data() then
that additional data will be sent in subsequent TCP packets which will be
buffered by the OS and not sent until an ACK is received for the first packet
containing the ClientHello. This means the early data is not actually
sent until a complete round trip with the server has occurred which defeats the
objective of early data.
In many operating systems the TCP_NODELAY socket option is available to disable
Nagle's algorithm. If an application opts to disable Nagle's algorithm
consideration should be given to turning it back on again after the handshake is
complete if appropriate.
=head1 REPLAY PROTECTION
When early data is in use the TLS protocol provides no security guarantees that
the same early data was not replayed across multiple connections. As a
mitigation for this issue OpenSSL automatically enables replay protection if the
server is configured with a non-zero max early data value. With replay
protection enabled sessions are forced to be single use only. If a client
attempts to reuse a session ticket more than once, then the second and
subsequent attempts will fall back to a full handshake (and any early data that
was submitted will be ignored). Note that single use tickets are enforced even
if a client does not send any early data.
The replay protection mechanism relies on the internal OpenSSL server session
cache (see L<SSL_CTX_set_session_cache_mode(3)>). By default sessions will be
added to the cache whenever a session ticket is issued. When a client attempts
to resume the session OpenSSL will check for its presence in the internal cache.
If it exists then the resumption is allowed and the session is removed from the
cache. If it does not exist then the resumption is not allowed and a full
handshake will occur.
Note that some applications may maintain an external cache of sessions (see
L<SSL_CTX_sess_set_new_cb(3)> and similar functions). It is the application's
responsibility to ensure that any sessions in the external cache are also
populated in the internal cache and that once removed from the internal cache
they are similarly removed from the external cache. Failing to do this could
result in an application becoming vulnerable to replay attacks. Note that
OpenSSL will lock the internal cache while a session is removed but that lock is
not held when the remove session callback (see L<SSL_CTX_sess_set_remove_cb(3)>)
is called. This could result in a small amount of time where the session has
been removed from the internal cache but is still available in the external
cache. Applications should be designed with this in mind in order to minimise
the possibility of replay attacks.
The OpenSSL replay protection does not apply to external Pre Shared Keys (PSKs)
(e.g. see SSL_CTX_set_psk_find_session_callback(3)). Therefore extreme caution
should be applied when combining external PSKs with early data.
=head1 RETURN VALUES
SSL_write_early_data() returns 1 for success or 0 for failure. In the event of a
failure call L<SSL_get_error(3)> to determine the correct course of action.
SSL_read_early_data() returns SSL_READ_EARLY_DATA_ERROR for failure,
SSL_READ_EARLY_DATA_SUCCESS for success with more data to read and
SSL_READ_EARLY_DATA_FINISH for success with no more to data be read. In the
event of a failure call L<SSL_get_error(3)> to determine the correct course of
action.
SSL_get_max_early_data(), SSL_CTX_get_max_early_data() and
SSL_SESSION_get_max_early_data() return the maximum number of early data bytes
that may be sent.
SSL_set_max_early_data(), SSL_CTX_set_max_early_data() and
SSL_SESSION_set_max_early_data() return 1 for success or 0 for failure.
SSL_get_early_data_status() returns SSL_EARLY_DATA_ACCEPTED if early data was
accepted by the server, SSL_EARLY_DATA_REJECTED if early data was rejected by
the server, or SSL_EARLY_DATA_NOT_SENT if no early data was sent.
=head1 SEE ALSO
L<SSL_get_error(3)>,
L<SSL_write_ex(3)>,
L<SSL_read_ex(3)>,
L<SSL_connect(3)>,
L<SSL_accept(3)>,
L<SSL_do_handshake(3)>,
L<SSL_CTX_set_psk_use_session_callback(3)>,
L<ssl(7)>
=head1 HISTORY
All of the functions described above were added in OpenSSL 1.1.1.
=head1 COPYRIGHT
Copyright 2017-2018 The OpenSSL Project Authors. All Rights Reserved.
Licensed under the OpenSSL license (the "License"). You may not use
this file except in compliance with the License. You can obtain a copy
in the file LICENSE in the source distribution or at
L<https://www.openssl.org/source/license.html>.
=cut