Commit 6b1bb98f moved the processing of ClientHello extensions into the
state machine post-processing stage. After processing s->init_num is reset
to 0, so by post-processing we cannot rely on its value. Unfortunately we
were using it to handle the PSK extension. This causes the handshake to
fail.

We were using init_num to figure out the length of ClientHello2 so we can
remove it from the handshake_buffer. The handshake_buffer holds the
transcript of all the messages sent so far. For PSK processing though we
only want to add in a partial ClientHello2. This commit changes things so
we just work out where ClientHello2 starts, working forward from the
beginning of handshake_buffer.

Fixes #2983
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2996)

This label for this derivation was incorrectly "derived" or "der" depending
on the pointer size of the build(!). The correct string is "derived secret".

(cherry picked from commit 936dcf272033c1bf59a5e859ec63e2557194f191)
Reviewed-by: NTim Hudson <tjh@openssl.org>
Reviewed-by: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2989)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2918)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2918)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2918)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2918)

Reviewed-by: NRichard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2969)

Reviewed-by: NRichard Levitte <levitte@openssl.org>
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2970)

In OpenSSL 1.1.0 the padding extension MUST be last because it calculates
the length of everything that has been written into the ClientHello to
determine whether it needs to be padded or not. With TLSv1.3 that isn't
possible because the specification requires that the PSK extension is last.
Therefore we need to fix the padding extension to take account of any PSK
extension that will be later added.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2968)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2895)

Choose a new ciphersuite for the HRR. Don't just use the one from the
session.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2895)

Don't include a PSK that does not have the right hash for the selected
ciphersuite following an HRR.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2895)

Draft-19 changes the HRR transcript hash so that the initial ClientHello
is replaced in the transcript with a special synthetic message_hash message
that just contains a hash of ClientHello1 as its message body.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2895)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2895)

As per draft-19
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2895)

The end of early data is now indicated by a new handshake message rather
than an alert.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2895)

TLSv1.3 draft 19 introduces a new pre HKDF-extract Derive-Secret stage.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2895)

tls1_get_curvelist() does not read from its third parameter, so
the assignments prior to function call were dead code and can be removed.
Reviewed-by: NRichard Levitte <levitte@openssl.org>
Reviewed-by: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2952)

Instead of making a positive comparison against the invalid value
that our server would send, make a negative check against the only
value that is not an error.
Reviewed-by: NRichard Levitte <levitte@openssl.org>
Reviewed-by: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2953)

Zero out the length alongside the NULLing of the pointer, to
bring parity between the selected and proposed fields..
Reviewed-by: NRich Salz <rsalz@openssl.org>
Reviewed-by: NRichard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2954)

No need to break out of the loop and repeat the loop termination
condition when we can just return.
Reviewed-by: NKurt Roeckx <kurt@openssl.org>
Reviewed-by: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2949)

Found using various (old-ish) versions of gcc.
Reviewed-by: NRich Salz <rsalz@openssl.org>
Reviewed-by: NRichard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2940)

... in functions dealing with the SSL object rather than the context.
Reviewed-by: NRich Salz <rsalz@openssl.org>
Reviewed-by: NRichard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2870)

that refers to space
deallocated by a call to the free function in tls_decrypt_ticket.
Reviewed-by: NMatt Caswell <matt@openssl.org>
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2897)
(cherry picked from commit 13ed1afa923f4ffb553e389de08f26e9ce84e8a2)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2898)

We need to check that the PSK extension in a ClientHello is the last one.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2896)

Reviewed-by: NRich Salz <rsalz@openssl.org>
Reviewed-by: NMatt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2703)

this change will prevent undefined behavior when src and dst are equal (memcpy), effectively allowing setting length only in both functions.

CLA: trivial
Reviewed-by: NMatt Caswell <matt@openssl.org>
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2750)

The value of SSL3_RT_MAX_ENCRYPTED_LENGTH normally includes the compression
overhead (even if no compression is negotiated for a connection). Except in
a build where no-comp is used the value of SSL3_RT_MAX_ENCRYPTED_LENGTH does
not include the compression overhead.
Reviewed-by: NRichard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2872)

In TLSv1.3 the above messages signal a key change. The spec requires that
the end of these messages must align with a record boundary. We can detect
this by checking for decrypted but as yet unread record data sitting in
OpenSSL buffers at the point where we process the messages.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2875)

Also updates SSL_has_pending() to use it. This actually fixes a bug in
SSL_has_pending() which is supposed to return 1 if we have any processed
or unprocessed data sitting in OpenSSL buffers. However it failed to return
1 if we had processed non-application data pending.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2875)

Also, restore 1.0.2 behavior of looping over all BIO's in the chain.
Thanks to Joseph Bester for finding this and suggesting a fix to the
crash.
Reviewed-by: NTim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2651)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2861)

Reviewed-by: NRich Salz <rsalz@openssl.org>
Reviewed-by: NMatt Caswell <matt@openssl.org>
GH: #2850

If early data is sent to a server, but ALPN is not used then memcmp is
called with a NULL pointer which is undefined behaviour.

Fixes #2841
Reviewed-by: NKurt Roeckx <kurt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2845)

We do not allow the generation of TLSv1.3 cookies. But if we receive one
in an HRR we will echo it back in the ClientHello.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2839)

A leak of an SSL_SESSION object can occur when decoding a psk extension on
an error path when using TLSv1.3
Reviewed-by: NRichard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2843)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2840)

Change tls12_sigalg_allowed() so it is passed a SIGALG_LOOKUP parameter,
this avoids multiple lookups.

When we copy signature algorithms return an error if no valid TLS message
signing algorithm is present. For TLS 1.3 this means we need at least one
signature algorithm other than RSA PKCS#1 or SHA1 both of which can only be
used to sign certificates and not TLS messages.
Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2840)

Reviewed-by: NRich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2840)