Add SSL_OP64_NO_EXTENDED_MASTER_SECRET, that can be set on either
an SSL or an SSL_CTX. When processing a ClientHello, if this flag
is set, do not indicate that the EMS TLS extension was received in
either the ssl3 object or the SSL_SESSION. Retain most of the
sanity checks between the previous and current session during
session resumption, but weaken the check when the current SSL
object is configured to not use EMS.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/3910)
The original 1.1.1 design was to use SSL_CB_HANDSHAKE_START and
SSL_CB_HANDSHAKE_DONE to signal start/end of a post-handshake message
exchange in TLSv1.3. Unfortunately experience has shown that this confuses
some applications who mistake it for a TLSv1.2 renegotiation. This means
that KeyUpdate messages are not handled properly.
This commit removes the use of SSL_CB_HANDSHAKE_START and
SSL_CB_HANDSHAKE_DONE to signal the start/end of a post-handshake
message exchange. Individual post-handshake messages are still signalled in
the normal way.
This is a potentially breaking change if there are any applications already
written that expect to see these TLSv1.3 events. However, without it,
KeyUpdate is not currently usable for many applications.
Fixes#8069
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/8096)
set_cipher_list() sets TLSv1.2 (and below) ciphers, and its success or
failure should not depend on whether set_ciphersuites() has been used to
setup TLSv1.3 ciphers.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7759)
When computing the end-point shared secret, don't take the
terminating NULL character into account.
Please note that this fix breaks interoperability with older
versions of OpenSSL, which are not fixed.
Fixes#7956
Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7957)
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Paul Yang <yang.yang@baishancloud.com>
Signed-off-by: Ping Yu <ping.yu@intel.com>
Signed-off-by: Steven Linsell <stevenx.linsell@intel.com>
(Merged from https://github.com/openssl/openssl/pull/7573)
This commit erroneously kept the DTLS timer running after the end of the
handshake. This is not correct behaviour and shold be reverted.
This reverts commit f7506416b1.
Fixes#7998
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/8047)
We were setting a limit of SSL3_RT_MAX_PLAIN_LENGTH on the size of the
ClientHello. AFAIK there is nothing in the standards that requires this
limit.
The limit goes all the way back to when support for extensions was first
added for TLSv1.0. It got converted into a WPACKET max size in 1.1.1. Most
likely it was originally added to avoid the complexity of having to grow
the init_buf in the middle of adding extensions. With WPACKET this is
irrelevant since it will grow automatically.
This issue came up when an attempt was made to send a very large
certificate_authorities extension in the ClientHello.
We should just remove the limit.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7424)
1. In addition to overriding the default application name,
one can now also override the configuration file name
and flags passed to CONF_modules_load_file().
2. By default we still keep going when configuration file
processing fails. But, applications that want to be strict
about initialization errors can now make explicit flag
choices via non-null OPENSSL_INIT_SETTINGS that omit the
CONF_MFLAGS_IGNORE_RETURN_CODES flag (which had so far been
both undocumented and unused).
3. In OPENSSL_init_ssl() do not request OPENSSL_INIT_LOAD_CONFIG
if the options already include OPENSSL_INIT_NO_LOAD_CONFIG.
4. Don't set up atexit() handlers when called with INIT_BASE_ONLY.
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7986)
The cryptopro extension is supposed to be unsolicited and appears in the
ServerHello only. Additionally it is unofficial and unregistered - therefore
we should really treat it like any other unknown extension if we see it in
the ClientHello.
Fixes#7747
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7984)
We have a number of instances where there are multiple "init" functions for
a single CRYPTO_ONCE variable, e.g. to load config automatically or to not
load config automatically. Unfortunately the RUN_ONCE mechanism was not
correctly giving the right return value where an alternative init function
was being used.
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7647)
This patch adds support for the Linux TLS Tx socket option.
If the socket option is successful, then the data-path of the TCP socket
is implemented by the kernel.
We choose to set this option at the earliest - just after CCS is complete.
Signed-off-by: Boris Pismenny <borisp@mellanox.com>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Paul Yang <yang.yang@baishancloud.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/5253)
Previously, the API version limit was indicated with a numeric version
number. This was "natural" in the pre-3.0.0 because the version was
this simple number.
With 3.0.0, the version is divided into three separate numbers, and
it's only the major number that counts, but we still need to be able
to support pre-3.0.0 version limits.
Therefore, we allow OPENSSL_API_COMPAT to be defined with a pre-3.0.0
style numeric version number or with a simple major number, i.e. can
be defined like this for any application:
-D OPENSSL_API_COMPAT=0x10100000L
-D OPENSSL_API_COMPAT=3
Since the pre-3.0.0 numerical version numbers are high, it's easy to
distinguish between a simple major number and a pre-3.0.0 numerical
version number and to thereby support both forms at the same time.
Internally, we define the following macros depending on the value of
OPENSSL_API_COMPAT:
OPENSSL_API_0_9_8
OPENSSL_API_1_0_0
OPENSSL_API_1_1_0
OPENSSL_API_3
They indicate that functions marked for deprecation in the
corresponding major release shall not be built if defined.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7724)
Fix some issues in tls13_hkdf_expand() which impact the above function
for TLSv1.3. In particular test that we can use the maximum label length
in TLSv1.3.
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7755)
Signed-off-by: Antoine Salon <asalon@vmware.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7738)
If compile OpenSSL with SSL_DEBUG macro, some test cases will cause the
process crashed in the debug code.
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7707)
A missing SSLfatal call can result in an assertion failed error if the
condition gets triggered.
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7594)
Signed-off-by: Antoine Salon <asalon@vmware.com>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7522)
SSL_get_signature_nid() -- local signature algorithm
SSL_get_signature_type_nid() -- local signature algorithm key type
SSL_get_peer_tmp_key() -- Peer key-exchange public key
SSL_get_tmp_key -- local key exchange public key
Aliased pre-existing SSL_get_server_tmp_key(), which was formerly
just for clients, to SSL_get_peer_tmp_key(). Changed internal
calls to use the new name.
Reviewed-by: Matt Caswell <matt@openssl.org>
SSL(_CTX)?_set_client_CA_list() was a server side only function in 1.1.0.
If it was called on the client side then it was ignored. In 1.1.1 it now
makes sense to have a CA list defined for both client and server (the
client now sends it the the TLSv1.3 certificate_authorities extension).
Unfortunately some applications were using the same SSL_CTX for both
clients and servers and this resulted in some client ClientHellos being
excessively large due to the number of certificate authorities being sent.
This commit seperates out the CA list updated by
SSL(_CTX)?_set_client_CA_list() and the more generic
SSL(_CTX)?_set0_CA_list(). This means that SSL(_CTX)?_set_client_CA_list()
still has no effect on the client side. If both CA lists are set then
SSL(_CTX)?_set_client_CA_list() takes priority.
Fixes#7411
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7503)
TLSv1.3 is more restrictive about the curve used. There must be a matching
sig alg defined for that curve. Therefore if we are using some other curve
in our certificate then we should not negotiate TLSv1.3.
Fixes#7435
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7442)
Rather than relying only on mandatory default digests, add a way for
the EVP_PKEY to individually report whether each digest algorithm is
supported.
Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7408)
If the private key says it can only support one specific digest, then
don't ask it to perform a different one.
Fixes: #7348
Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7408)
use_ecc() was always returning 1 because there are default (TLSv1.3)
ciphersuites that use ECC - even if those ciphersuites are disabled by
other options.
Fixes#7471
Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
(Merged from https://github.com/openssl/openssl/pull/7479)
Commit 9ef9088c15 switched the SSL/SSL_CTX
statistics counters to using Thread-Sanitizer-friendly primitives.
However, it erroneously converted an addition of -1
(for s->session_ctx->stats.sess_accept) to an addition of +1, since that
is the only counter API provided by the internal tsan_assist.h header
until the previous commit. This means that for each accepted (initial)
connection, the session_ctx's counter would get doubly incremented, and the
(switched) ctx's counter would also get incremented.
Restore the counter decrement so that each accepted connection increments
exactly one counter exactly once (in net effect).
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7464)
In TLSv1.2 and below a CertificateRequest is sent after the Certificate
from the server. This means that by the time the client_cert_cb is called
on receipt of the CertificateRequest a call to SSL_get_peer_certificate()
will return the server certificate as expected. In TLSv1.3 a
CertificateRequest is sent before a Certificate message so calling
SSL_get_peer_certificate() returns NULL.
To workaround this we delay calling the client_cert_cb until after we
have processed the CertificateVerify message, when we are doing TLSv1.3.
Fixes#7384
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/7413)
Since 1fb9fdc30 we may attempt to buffer a record from the next epoch
that has already been buffered. Prior to that this never occurred.
We simply ignore a failure to buffer a duplicated record.
Fixes#6902
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7414)
Previously when a ClientHello arrives with a valid cookie using
DTLSv1_listen() we only "peeked" at the message and left it on the
underlying fd. This works fine for single threaded applications but for
multi-threaded apps this does not work since the fd is typically reused for
the server thread, while a new fd is created and connected for the client.
By "peeking" we leave the message on the server fd, and consequently we
think we've received another valid ClientHello and so we create yet another
fd for the client, and so on until we run out of fds.
In this new approach we remove the ClientHello and buffer it in the SSL
object.
Fixes#6934
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/7375)
Rather than using init_buf we use the record layer read and write buffers
in DTLSv1_listen(). These seem more appropriate anyway and will help with
the next commit.
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/7375)
The formula used for this is now
kVarianceBlocks = ((255 + 1 + md_size + md_block_size - 1) / md_block_size) + 1
Notice that md_block_size=64 for SHA256, which results on the
magic constant kVarianceBlocks = 6.
However, md_block_size=128 for SHA384 leading to kVarianceBlocks = 4.
CLA:trivial
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7342)
I was never exported in our shared libraries and no one noticed, and
we don't seem to use it ourselves, so clean it away.
In all likelyhood, this is a remain from the 90's, when it was in
fashion to litter library modules with these kinds of strings.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7340)
If using an old style TLSv1.2 PSK callback then the maximum possible PSK
len is PSK_MAX_PSK_LEN (256) - not 64.
Fixes#7261
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7267)
Historically SSL_CTX_set_ssl_version() has reset the cipher list
to the default. Splitting TLS 1.3 ciphers to be tracked separately
caused a behavior change, in that TLS 1.3 cipher configuration was
preserved across calls to SSL_CTX_set_ssl_version(). To restore commensurate
behavior with the historical behavior, set the ciphersuites to the default as
well as setting the cipher list to the default.
Closes: #7226
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7270)
PR #3783 introduce coded to reset the server side SNI state in
SSL_do_handshake() to ensure any erroneous config time SNI changes are
cleared. Unfortunately SSL_do_handshake() can be called mid-handshake
multiple times so this is the wrong place to do this and can mean that
any SNI data is cleared later on in the handshake too.
Therefore move the code to a more appropriate place.
Fixes#7014
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/7149)
Ideally, SSL_get_servername() would do exactly as it is documented
and return exactly what the client sent (i.e., what we currently
are stashing in the SSL's ext.hostname), without needing to refer
to an SSL_SESSION object. For historical reasons, including the
parsed SNI value from the ClientHello originally being stored in the
SSL_SESSION's ext.hostname field, we have had references to the
SSL_SESSION in this function. We cannot fully excise them due to
the interaction between user-supplied callbacks and TLS 1.2 resumption
flows, where we call all callbacks but the client did not supply an
SNI value. Existing callbacks expect to receive a valid SNI value
in this case, so we must fake one up from the resumed session in
order to avoid breakage.
Otherwise, greatly simplify the implementation and just return the
value in the SSL, as sent by the client.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7115)
Commit 1c4aa31d79 modified the state machine
to clean up stale ext.hostname values from SSL objects in the case when
SNI was not negotiated for the current handshake. This is natural from
the TLS perspective, since this information is an extension that the client
offered but we ignored, and since we ignored it we do not need to keep it
around for anything else.
However, as documented in https://github.com/openssl/openssl/issues/7014 ,
there appear to be some deployed code that relies on retrieving such an
ignored SNI value from the client, after the handshake has completed.
Because the 1.1.1 release is on a stable branch and should preserve the
published ABI, restore the historical behavior by retaining the ext.hostname
value sent by the client, in the SSL structure, for subsequent retrieval.
[extended tests]
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7115)
The is_tls13_capable() function should not return 0 if no certificates
are configured directly because a certificate callback is present.
Fixes#7140
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/7141)
If we've sent a close_notify then we are restricted about what we can do
in response to handshake messages that we receive. However we can sensibly
process NewSessionTicket messages. We can also process a KeyUpdate message
as long as we also ignore any request for us to update our sending keys.
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7114)
If we have selected a ciphersuite using RSA key exchange then we must
not attempt to use an RSA-PSS cert for that.
Fixes#7059
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7099)
Treat a connection using an external PSK like we would a resumption and
send a single NewSessionTicket afterwards.
Fixes#6941
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7097)
If a client sends data to a server and then immediately closes without
waiting to read the NewSessionTickets then the server can receive EPIPE
when trying to write the tickets and never gets the opportunity to read
the data that was sent. Therefore we ignore EPIPE when writing out the
tickets in TLSv1.3
Fixes#6904
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6944)
They add a single item, so the names give a false impression of what
they do, making them hard to remember. Better to give them a somewhat
better name.
Fixes#6930
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6931)
Fixes#6994
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Paul Yang <yang.yang@baishancloud.com>
(Merged from https://github.com/openssl/openssl/pull/7009)
We need to ensure that the min-max version range we use when constructing
the ClientHello is the same range we use when we validate the version
selected by the ServerHello. Otherwise this may appear as a fallback or
downgrade.
Fixes#6964
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7013)
The TLS-1.3 ciphersuites must not be blocked by @SECLEVEL=3 even
though they are not explicitly marked as using DH/ECDH.
Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6959)
We already have SSL_set_post_handshake_auth(). This just adds the SSL_CTX
equivalent.
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6938)
Having post handshake auth automatically switched on breaks some
applications written for TLSv1.2. This changes things so that an explicit
function call is required for a client to indicate support for
post-handshake auth.
Fixes#6933.
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6938)
A client that has fallen back could detect an inappropriate fallback if
the TLSv1.3 downgrade protection sentinels are present.
Fixes#6756
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6894)
At certain points in the handshake we could receive either a plaintext or
an encrypted alert from the client. We should tolerate both where
appropriate.
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6887)
If we sent early_data and then received back an HRR, the enc_write_ctx
was stale resulting in errors if an alert needed to be sent.
Thanks to Quarkslab for reporting this.
In any case it makes little sense to encrypt alerts using the
client_early_traffic_secret, so we add special handling for alerts sent
after early_data. All such alerts are sent in plaintext.
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6887)
Under certain error conditions a call to SSLfatal could accidently be
missed.
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
Reviewed-by: Andy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6872)
In some cases it's about redundant check for return value, in some
cases it's about replacing check for -1 with comparison to 0.
Otherwise compiler might generate redundant check for <-1. [Even
formatting and readability fixes.]
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6860)
In some scenarios the connection could fail without an alert being sent.
This causes a later assertion failure.
Thanks to Quarkslab for reporting this.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/6852)
Ensure that the certificate required alert actually gets sent (and doesn't
get translated into handshake failure in TLSv1.3).
Ensure that proper reason codes are given for the new TLSv1.3 alerts.
Remove an out of date macro for TLS13_AD_END_OF_EARLY_DATA. This is a left
over from an earlier TLSv1.3 draft that is no longer used.
Fixes#6804
Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6809)
Commit 1c4aa31d79 changed how we process
and store SNI information during the handshake, so that a hostname is
only saved in the SSL_SESSION structure if that SNI value has actually
been negotiated. SSL_get_servername() was adjusted to match, with a new
conditional being added to handle the case when the handshake processing
is ongoing, and a different location should be consulted for the offered
SNI value. This was done in an attempt to preserve the historical
behavior of SSL_get_servername(), a function whose behavior only mostly
matches its documentation, and whose documentation is both lacking and
does not necessarily reflect the actual desired behavior for such an
API. Unfortunately, sweeping changes that would bring more sanity to
this space are not possible until OpenSSL 1.2.0, for ABI compatibility
reasons, so we must attempt to maintain the existing behavior to the
extent possible.
The above-mentioned commit did not take into account the behavior
of SSL_get_servername() during resumption handshakes for TLS 1.2 and
prior, where no SNI negotiation is performed. In that case we would
not properly parse the incoming SNI and erroneously return NULL as
the servername, when instead the logical session is associated with
the SNI value cached in the SSL_SESSION. (Note that in some cases an
SNI callback may not need to do anything in a TLS 1.2 or prior resumption
flow, but we are calling the callbacks and did not provide any guidance
that they should no-op if the connection is being resumed, so we must
handle this case in a usable fashion.) Update our behavior accordingly to
return the session's cached value during the handshake, when resuming.
This fixes the boringssl tests.
[extended tests]
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6792)
The API used to set what SNI value to send in the ClientHello
can also be used on server SSL objects, with undocumented and
un-useful behavior. Unfortunately, when generic SSL_METHODs
are used, s->server is still set, prior to the start of the
handshake, so we cannot prevent this nonsensical usage at the
present time. Leave a note to revisit this when ABI-breaking
changes are permitted.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6378)
In particular, adhere to the rule that we must not modify any
property of an SSL_SESSION object once it is (or might be) in
a session cache. Such modifications are thread-unsafe and have
been observed to cause crashes at runtime.
To effect this change, standardize on the property that
SSL_SESSION->ext.hostname is set only when that SNI value
has been negotiated by both parties for use with that session.
For session resumption this is trivially the case, so only new
handshakes are affected.
On the client, the new semantics are that the SSL->ext.hostname is
for storing the value configured by the caller, and this value is
used when constructing the ClientHello. On the server, SSL->ext.hostname
is used to hold the value received from the client. Only if the
SNI negotiation is successful will the hostname be stored into the
session object; the server can do this after it sends the ServerHello,
and the client after it has received and processed the ServerHello.
This obviates the need to remove the hostname from the session object
in case of failed negotiation (a change that was introduced in commit
9fb6cb810b in order to allow TLS 1.3
early data when SNI was present in the ClientHello but not the session
being resumed), which was modifying cached sessions in certain cases.
(In TLS 1.3 we always produce a new SSL_SESSION object for new
connections, even in the case of resumption, so no TLS 1.3 handshakes
were affected.)
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6378)
These tiny functions only read from the input SSL, and we are
about to use them from functions that only have a const SSL* available,
so propagate const a bit further.
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6378)
The spec says that a client MUST set legacy_version to TLSv1.2, and
requires servers to verify that it isn't SSLv3.
Fixes#6600
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6747)
Using the rsa_pss_rsae_sha256 sig alg should imply that the key OID is
rsaEncryption. Similarly rsa_pss_pss_sha256 implies the key OID is
rsassaPss. However we did not check this and incorrectly tolerated a key
OID that did not match the sig alg sent by the peer.
Fixes#6611
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6732)
If we issue new tickets due to post-handshake authentication there is no
reason to remove previous tickets from the cache. The code that did that
only removed the last session anyway - so if more than one ticket got
issued then those other tickets are still valid.
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6722)
Check that we are either configured for PSK, or that we have a TLSv1.3
capable certificate type. DSA certs can't be used in TLSv1.3 and we
don't (currently) allow GOST ones either (owing to the lack of standard
sig algs).
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6650)
Previoulsy we just had max_early_data which controlled both the value of
max early_data that we advertise in tickets *and* the amount of early_data
that we are willing to receive from clients. This doesn't work too well in
the case where we want to reduce a previously advertised max_early_data
value. In that case clients with old, stale tickets may attempt to send us
more early data than we are willing to receive. Instead of rejecting the
early data we abort the connection if that happens.
To avoid this we introduce a new "recv_max_early_data" value. The old
max_early_data becomes the value that is advertised in tickets while
recv_max_early_data is the maximum we will tolerate from clients.
Fixes#6647
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/6655)
In TLSv1.2 and below we should remove an old session from the client
session cache in the event that we receive a new session ticket from the
server.
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/6601)
This also adds the ability to control this through s_server
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6469)
We need to check the provided SSL_SESSION* for NULL before
attempting to derference it to see if it's a TLS 1.3 session.
Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
(Merged from https://github.com/openssl/openssl/pull/6622)
Currently if you encounter application data while waiting for a
close_notify from the peer, and you have called SSL_shutdown() then
you will get a -1 return (fatal error) and SSL_ERROR_SYSCALL from
SSL_get_error(). This isn't accurate (it should be SSL_ERROR_SSL) and
isn't persistent (you can call SSL_shutdown() again and it might then work).
We change this into a proper fatal error that is persistent.
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
(Merged from https://github.com/openssl/openssl/pull/6340)
In the case where we are shutdown for writing and awaiting a close_notify
back from a subsequent SSL_shutdown() call we skip over handshake data
that is received. This should not be treated as an error - instead it
should be signalled with SSL_ERROR_WANT_READ.
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
(Merged from https://github.com/openssl/openssl/pull/6340)
If we've sent a close_notify and we're waiting for one back we drop
incoming records until we see the close_notify we're looking for. If
SSL_MODE_AUTO_RETRY is on, then we should immediately try and read the
next record.
Fixes#6262
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
(Merged from https://github.com/openssl/openssl/pull/6340)
During anti-replay we cache the ticket anyway, so there is no point in
using a full stateless ticket.
Fixes#6391
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6563)
Implement support for stateful TLSv1.3 tickets, and use them if
SSL_OP_NO_TICKET is set.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6563)
Separate out as a new function the code to write out data which is specific
to a stateless ticket.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6563)
Sessions should be immutable once they are in the cache because they could
be shared with other threads. If you change them then this can cause
corruptions and races
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6561)
These headers are internal and never exposed to a cpp compiler, hence no
need for the preamble.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/6554)
An alpn_selected value containing NUL bytes in it will result in
ext.alpn_selected_len having a larger value than the number of bytes
allocated in ext.alpn_selected.
Issue found by OSS-fuzz.
Reviewed-by: Andy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6507)
Commit 4aa5a5669 accidentally missed off the catch all case of ignoring all
warning alerts that are otherwise unhandled. This breaks the SSLv3 tests
which send a "no certificate" warning alert.
Fixes#6496
[extended tests]
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/6509)
TLSv1.3 ignores the alert level, so we should suppress sending of
warning only alerts.
Fixes#6211
Reviewed-by: Andy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6370)
In TLSv1.3 we should ignore the severity level of an alert according to
the spec.
Reviewed-by: Andy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6370)
If the remove_session_cb accesses the session's data (for instance,
via SSL_SESSION_get_protocol_version), a potential use after free
can occur. For this, consider the following scenario when adding
a new session via SSL_CTX_add_session:
- The session cache is full
(SSL_CTX_sess_number(ctx) > SSL_CTX_sess_get_cache_size(ctx))
- Only the session cache has a reference to ctx->session_cache_tail
(that is, ctx->session_cache_tail->references == 1)
Since the cache is full, remove_session_lock is called to remove
ctx->session_cache_tail from the cache. That is, it
SSL_SESSION_free()s the session, which free()s the data. Afterwards,
the free()d session is passed to the remove_session_cb. If the callback
accesses the session's data, we have a use after free.
The free before calling the callback behavior was introduced in
commit e4612d02c5 ("Remove sessions
from external cache, even if internal cache not used.").
CLA: trivial
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6222)
We generate the secrets based on the nonce immediately so there is no
need to keep the nonce.
Reviewed-by: Andy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6415)
All tickets on a connection need to have a unique nonce. When this was
originally implemented we only ever sent one ticket on the conneciton so
this didn't matter. We were just using the value 0. Now we can get multiple
tickets to we need to start doing the ticket nonce properly.
Fixes#6387
Reviewed-by: Andy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6415)
The TLSv1.3 spec requires us to use the client application traffic secret
during generation of the Finished message following a post handshake
authentication.
Fixes#6263
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/6297)
Because TLS 1.3 sends more non-application data records some clients run
into problems because they don't expect SSL_read() to return and set
SSL_ERROR_WANT_READ after processing it.
This can cause problems for clients that use blocking I/O and use
select() to see if data is available. It can be cleared using
SSL_CTX_clear_mode().
Reviewed-by: Matt Caswell <matt@openssl.org>
GH: #6260
NULL-check for cipher is redundant, instead check if cipher->name is NULL
While here fix formatting of BIO_printf calls as suggested by Andy Polyakov.
CLA: trivial
Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6282)
In TLSv1.2 and below we always cache new sessions by default on the server
side in the internal cache (even when we're using session tickets). This is
in order to support resumption from a session id.
In TLSv1.3 there is no session id. It is only possible to resume using the
ticket. Therefore, in the default case, there is no point in caching the
session in the internal store.
There is still a reason to call the external cache new session callback
because applications may be using the callbacks just to know about when
sessions are created (and not necessarily implementing a full cache). If
the application also implements the remove session callback then we are
forced to also store it in the internal cache so that we can create
timeout events. Otherwise the external cache could just fill up
indefinitely.
This mostly addresses the issue described in #5628. That issue also proposes
having an option to not create full stateless tickets when using the
internal cache. That aspect hasn't been addressed yet.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/6293)
Also allows the apps to set it.
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/5227)
We send a session ticket automatically in TLSv1.3 at the end of the
handshake. This commit provides the ability to set how many tickets should
be sent. By default this is one.
Fixes#4978
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/5227)
In DTLS if we have buffered a fragment for a zero length message (e.g.
ServerHelloDone) then, when we unbuffered the fragment, we were attempting
to memcpy the contents of the fragment which is zero length and a NULL
pointer. This is undefined behaviour. We should check first whether we
have a zero length fragment.
Fixes a travis issue.
[extended tests]
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6223)
Don't call the decrypt ticket callback if we've already encountered a
fatal error. Do call it if we have an empty ticket present.
Change the return code to have 5 distinct returns codes and separate it
from the input status value.
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6198)
The return value from the ticket_key callback was not properly handled in
TLSv1.3, so that a ticket was *always* renewed even if the callback
requested that it should not be.
Also the ticket decrypt callback was not being called at all in TLSv1.3.
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6198)
When a server call SSL_write_early_data() to write to an unauthenticated
client the buffering BIO is still in place, so we should ensure we flush
the write.
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6198)
