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: NTim Hudson <tjh@openssl.org>
Reviewed-by: NViktor Dukhovni <viktor@openssl.org>
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/7149)

Commit 1c4aa31d 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: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7115)

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: NViktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/7013)

Reviewed-by: NPaul Dale <paul.dale@oracle.com>
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6786)

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
9fb6cb81 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: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6378)

Reviewed-by: NViktor Dukhovni <viktor@openssl.org>
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6469)

TLSv1.3 ignores the alert level, so we should suppress sending of
warning only alerts.

Fixes #6211
Reviewed-by: NAndy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6370)

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: NAndy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6415)

Fix some instances where we weren't checking the error return.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/6373)

Renamed to EVP_PKEY_new_raw_private_key()/EVP_new_raw_public_key() as per
feedback.
Reviewed-by: NRichard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/5520)

Reviewed-by: NRichard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/5520)

As per the latest text in TLSv1.3 draft-26
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/5604)

At the core of things is the concept that each extension is only
defined in certain context(s) -- the ClientHello, EncryptedExtensions,
etc., and sometimes only for a specific protocol or protocol range;
we want to enforce that we only parse or generate extensions in the
context(s) for which they are defined.  There is some subtlety here,
in that the protocol version in use is not known when generating the
ClientHello (but it is known when the ClientHello extensions are
being parsed!), so the SSL_IS_TLS13() macro must be used with caution.
Nonetheless, by making assertions about whether we are acting in a
server role and whether the current context is (not) a ClientHello,
we can consolidate almost all of the logic for determining whether
an extension is permitted in a given protocol message, whether we
are generating or parsing that message.

The only logic that remains separate relates to generating the ClientHello,
as it depends on an external factor (the maximum permitted TLS version) that
is not defined in the parsing context.
Reviewed-by: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2945)

We also default to SHA256 as per the spec if we do not have an explicit
digest defined.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/5554)

They are valid for use in a CertificateRequest message, but we did not
allow it. If a server sent such a message using either of those two
extensions then the handshake would abort.

This corrects that error, but does not add support for actually processing
the extensions. They are simply ignored, and a TODO is inserted to add
support at a later time.

This was found during interoperability testing with btls:
https://gitlab.com/ilari_l/btls

Prompted by these errors I reviewed the complete list of extensions and
compared them with the latest table in draft-24 to confirm there were no
other errors of a similar type. I did not find any.
Reviewed-by: NRich Salz <rsalz@openssl.org>
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/5490)

Coverity was complaining because we checked if s->ctx is NULL and then
later on in the function deref s->ctx anyway. In reality if s->ctx is
NULL then this is an internal error.
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/5334)

The s_client psk_use_session_cb callback has a comment stating that we
should ignore a key that isn't suitable for TLSv1.3. However we were
actually causing the connection to fail. Changing the return value fixes
the issue.

Also related to this is that the early_data extension was not marked as
TLSv1.3 only which it should be.

Fixes #5202
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/5205)

Reviewed-by: NBen Kaduk <kaduk@mit.edu>
Reviewed-by: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4964)

Add SSL_verify_client_post_handshake() for servers to initiate PHA

Add SSL_force_post_handshake_auth() for clients that don't have certificates
initially configured, but use a certificate callback.

Update SSL_CTX_set_verify()/SSL_set_verify() mode:

* Add SSL_VERIFY_POST_HANDSHAKE to postpone client authentication until after
the initial handshake.

* Update SSL_VERIFY_CLIENT_ONCE now only sends out one CertRequest regardless
of when the certificate authentication takes place; either initial handshake,
re-negotiation, or post-handshake authentication.

Add 'RequestPostHandshake' and 'RequirePostHandshake' SSL_CONF options that
add the SSL_VERIFY_POST_HANDSHAKE to the 'Request' and 'Require' options

Add support to s_client:
* Enabled automatically when cert is configured
* Can be forced enabled via -force_pha

Add support to s_server:
* Use 'c' to invoke PHA in s_server
* Remove some dead code

Update documentation

Update unit tests:
* Illegal use of PHA extension
* TLSv1.3 certificate tests

DTLS and TLS behave ever-so-slightly differently. So, when DTLS1.3 is
implemented, it's PHA support state machine may need to be different.
Add a TODO and a #error

Update handshake context to deal with PHA.

The handshake context for TLSv1.3 post-handshake auth is up through the
ClientFinish message, plus the CertificateRequest message. Subsequent
Certificate, CertificateVerify, and Finish messages are based on this
handshake context (not the Certificate message per se, but it's included
after the hash). KeyUpdate, NewSessionTicket, and prior Certificate
Request messages are not included in post-handshake authentication.

After the ClientFinished message is processed, save off the digest state
for future post-handshake authentication. When post-handshake auth occurs,
copy over the saved handshake context into the "main" handshake digest.
This effectively discards the any KeyUpdate or NewSessionTicket messages
and any prior post-handshake authentication.

This, of course, assumes that the ID-22 did not mean to include any
previous post-handshake authentication into the new handshake transcript.
This is implied by section 4.4.1 that lists messages only up to the
first ClientFinished.
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
Reviewed-by: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4964)

The new extension is like signature_algorithms, but only for the
signature *on* the certificate we will present to the peer (the
old signature_algorithms extension is still used for signatures that
we *generate*, i.e., those over TLS data structures).

We do not need to generate this extension, since we are the same
implementation as our X.509 stack and can handle the same types
of signatures, but we need to be prepared to receive it, and use the received
information when selecting what certificate to present.

There is a lot of interplay between signature_algorithms_cert and
signature_algorithms, since both affect what certificate we can
use, and thus the resulting signature algorithm used for TLS messages.
So, apply signature_algorithms_cert (if present) as a filter on what
certificates we can consider when choosing a certificate+sigalg
pair.

As part of this addition, we also remove the fallback code that let
keys of type EVP_PKEY_RSA be used to generate RSA-PSS signatures -- the
new rsa_pss_pss_* and rsa_pss_rsae_* signature schemes have pulled
the key type into what is covered by the signature algorithm, so
we should not apply this sort of compatibility workaround.
Reviewed-by: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/5068)

Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/4435)

Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/4435)

This just adds the various extension functions. More changes will be
required to actually use them.
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/4435)

Reviewed-by: NTim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/5038)

Although this is forbidden by all three(!) relevant specifications,
there seem to be multiple server implementations in the wild that
send it.  Since we didn't check for unexpected extensions in any
given message type until TLS 1.3 support was added, our previous
behavior was to silently accept these extensions and pass them over
to the custom extension callback (if any).  In order to avoid
regression of functionality, relax the check for "extension in
unexpected context" for this specific case, but leave the protocol
enforcment mechanism unchanged for other extensions and in other
extension contexts.

Leave a detailed comment to indicate what is going on.
Reviewed-by: NTim Hudson <tjh@openssl.org>
Reviewed-by: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4463)

Reviewed-by: NKurt Roeckx <kurt@roeckx.be>
Reviewed-by: NMatt Caswell <matt@openssl.org>
GH: #4934

Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/4701)

The CCS may be sent at different times based on whether or not we
sent an HRR earlier. In order to make that decision this commit
also updates things to make sure we remember whether an HRR was
used or not.
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/4701)

The new ServerHello format is essentially now the same as the old TLSv1.2
one, but it must additionally include supported_versions. The version
field is fixed at TLSv1.2, and the version negotiation happens solely via
supported_versions.
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/4701)

Follow up from the conversion to use SSLfatal() in the state machine to
clean things up a bit more.

[extended tests]
Reviewed-by: NRichard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4778)

Reviewed-by: NRichard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4778)

Reviewed-by: NRichard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4778)

SNI needs to be consistent before we accept early_data. However a
server may choose to not acknowledge SNI. In that case we have to
expect that a client may send it anyway. We change the consistency
checks so that not acknowledging is treated more a like a "wild card",
accepting any SNI as being consistent.
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/4738)

Based on patch from Tomasz Moń:
https://groups.google.com/forum/#!topic/mailing.openssl.dev/fQxXvCg1uQYReviewed-by: NMatt Caswell <matt@openssl.org>
Reviewed-by: NBernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/1008)

We currently increment the SSL_CTX stats.sess_accept field in
tls_setup_handshake(), which is invoked from the state machine well
before ClientHello processing would have had a chance to switch
the SSL_CTX attached to the SSL object due to a provided SNI value.
However, stats.sess_accept_good is incremented in tls_finish_handshake(),
and uses the s->ctx.stats field (i.e., the new SSL_CTX that was switched
to as a result of SNI processing).  This leads to the confusing
(nonsensical) situation where stats.sess_accept_good is larger than
stats.sess_accept, as the "sess_accept" value was counted on the
s->session_ctx.

In order to provide some more useful numbers, increment
s->ctx.stats.sess_accept after SNI processing if the SNI processing
changed s->ctx to differ from s->session_ctx.  To preserve the
property that any given accept is counted only once, make the
corresponding decrement to s->session_ctx.stats.sess_accept when
doing so.
Reviewed-by: NMatt Caswell <matt@openssl.org>
Reviewed-by: NPaul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/4549)

Previously if a client received an HRR then we would do version negotiation
immediately - because we know we are going to get TLSv1.3. However this
causes a problem when we emit the 2nd ClientHello because we start changing
a whole load of stuff to ommit things that aren't relevant for < TLSv1.3.
The spec requires that the 2nd ClientHello is the same except for changes
required from the HRR. Therefore the simplest thing to do is to defer the
version negotiation until we receive the ServerHello.

Fixes #4292
Reviewed-by: NTim Hudson <tjh@openssl.org>
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/4527)

Fixes #4496
Reviewed-by: NRich Salz <rsalz@openssl.org>
Reviewed-by: NBen Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/4519)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4475)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/4475)

When an SSL's context is swtiched from a ticket-enabled context to
a ticket-disabled context in the servername callback, no session-id
is generated, so the session can't be resumed.

If a servername callback changes the SSL_OP_NO_TICKET option, check
to see if it's changed to disable, and whether a session ticket is
expected (i.e. the client indicated ticket support and the SSL had
tickets enabled at the time), and whether we already have a previous
session (i.e. s->hit is set).

In this case, clear the ticket-expected flag, remove any ticket data
and generate a session-id in the session.

If the SSL hit (resumed) and switched to a ticket-disabled context,
assume that the resumption was via session-id, and don't bother to
update the session.

Before this fix, the updated unit-tests in 06-sni-ticket.conf would
fail test #4 (server1 = SNI, server2 = no SNI).
Reviewed-by: NRich Salz <rsalz@openssl.org>
Reviewed-by: NRichard Levitte <levitte@openssl.org>
Reviewed-by: NMatt Caswell <matt@openssl.org>
Reviewed-by: NPaul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/1529)