Add DSA support to mkcert.sh

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2667)

Add and use function test_pem to work out test filenames.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2667)

Fix a slightly confusing if condition in a2i_ASN1_INTEGER.

Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2640)

Fix a merge error

Fix an error code clash due to a merge

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2665)

Fix test_x509_store

Don't run this test unless 'openssl rehash' works properly.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2664)

Fix symbol shadow

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2663)

Updates following review feedback

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Don't use an enum in the return type for a public API function

We use an int instead. That means SSL_key_update() also should use an int.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Add some KeyUpdate tests

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Add documentation for SSL_key_update() and SSL_get_key_update_type()

This also adds documentation for the pre-existing and related
SSL_renegotiate*() functions.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Fix a shadowed global variable warning

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Updates following review feedback

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Limit the number of KeyUpdate messages we can process

Too many KeyUpdate message could be inicative of a problem (e.g. an
infinite KeyUpdate loop if the peer always responds to a KeyUpdate message
with an "update_requested" KeyUpdate response), or (conceivably) an attack.
Either way we limit the number of KeyUpdate messages we are prepared to
handle.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Actually update the keys when a KeyUpdate message is sent or received

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

If we receive an "update_requested" KeyUpdate then respond with a KeyUpdate

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Add the ability for a server to receive a KeyUpdate message

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Add the ability for s_client to send a KeyUpdate message

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Add a SSL_get_key_update_type() function

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Add the ability for a client to send a KeyUpdate message

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Add the ability for a client to receive a KeyUpdate message

This just receives the message. It doesn't actually update any keys yet.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Add SSL_trace() support for KeyUpdate messages

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Add a capability to s_server to send KeyUpdate messages

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Provide a function to send a KeyUpdate message

This implements the server side KeyUpdate sending capability as well.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2609)

Fix typo in x86_64-mont5.pl CFI directives

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Andy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2655)

Add a test of the X509_STORE / X509_LOOKUP API

Fortunately, "openssl verify" makes good use of that API

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2652)

test/README: clarify test number groups

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2652)

Fix a mem leak in ssl_test_ctx.c

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2650)

Add client auth TLS 1.3 certificate selection tests

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2643)

Add ECDSA client certificates

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2643)

Use tls_choose_sigalg for client auth.

For client auth call tls_choose_sigalg to select the certificate
and signature algorithm. Use the selected algorithm in
tls_construct_cert_verify.

Remove obsolete tls12_get_sigandhash.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2643)

Add client side support to tls_choose_sigalg.

Select appropriate signature algorithm and certificate for client
authentication using tls_choose_sigalg.

A lot of selection logic is very similar except not finding a
certificate is not a fatal error: we just do not present a
certificate.

For TLS 1.2 and earlier we only check the current certificate
is suitable (for compatibility with previous logic) for TLS 1.3
(where there are no compatibility issues) we support multiple
client certificates for different algorithms.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2643)

Change req_check_len error message, it also accepts 20 bytes, but states 'less than' in the error message

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2648)

Use _WIN32 over WIN32 for preprocessor conditional

The intent seems to be that the WIN32 symbol is for things that are a direct
byproduct of being a windows-variant configuration and should be used for
feature en/disablement on windows systems.  Use of the _WIN32 symbol is more
widespread, being used to implement platform portability of more generic code.

We do define WIN32 in some situations in e_os.h, but that is not included
universally.

Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2642)

Update CHANGES and NEWS for new release

Reviewed-by: Richard Levitte <levitte@openssl.org>

Remove an OPENSSL_assert() and replace with a soft assert and check

Following on from CVE-2017-3733, this removes the OPENSSL_assert() check
that failed and replaces it with a soft assert, and an explicit check of
value with an error return if it fails.

Reviewed-by: Richard Levitte <levitte@openssl.org>

Don't change the state of the ETM flags until CCS processing

In 1.1.0 changing the ciphersuite during a renegotiation can result in
a crash leading to a DoS attack. In master this does not occur with TLS
(instead you get an internal error, which is still wrong but not a security
issue) - but the problem still exists in the DTLS code.

The problem is caused by changing the flag indicating whether to use ETM
or not immediately on negotiation of ETM, rather than at CCS. Therefore,
during a renegotiation, if the ETM state is changing (usually due to a
change of ciphersuite), then an error/crash will occur.

Due to the fact that there are separate CCS messages for read and write
we actually now need two flags to determine whether to use ETM or not.

CVE-2017-3733

Reviewed-by: Richard Levitte <levitte@openssl.org>

Provide a test for the Encrypt-Then-Mac renegotiation crash

In 1.1.0 changing the ciphersuite during a renegotiation can result in
a crash leading to a DoS attack. In master this does not occur with TLS
(instead you get an internal error, which is still wrong but not a security
issue) - but the problem still exists in the DTLS code.

This commit provides a test for the issue.

CVE-2017-3733

Reviewed-by: Richard Levitte <levitte@openssl.org>

Fix warning

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2644)

Handle negative reply for NNTP STARTTLS in s_client

Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2629)

Properly zero cipher_data for ChaCha20-Poly1305 on cleanup

Fix a typo. Probably this has not been found because EVP_CIPHER_CTX is
smaller than EVP_CHACHA_AEAD_CTX and heap overflow does not occur.

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2294)

crypto/armcap.c: short-circuit processor capability probe in iOS builds.

Capability probing by catching SIGILL appears to be problematic
on iOS. But since Apple universe is "monocultural", it's actually
possible to simply set pre-defined processor capability mask.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2617)

ARMv4 assembly pack: harmonize Thumb-ification of iOS build.

Three modules were left behind in a285992763f3961f69a8d86bf7dfff020a08cef9.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2617)

Set current certificate to selected certificate.

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2637)

sha/asm/*-x86_64.pl: add CFI annotations.

Reviewed-by: Rich Salz <rsalz@openssl.org>

Rework error handling of custom_ext_meth_add towards strong exception safety.

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2636)

Fix merge issue

Causes make update to fail.

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2634)

Skip curve check if sigalg doesn't specify a curve.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2623)

Use CERT_PKEY pointer instead of index

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2623)

Simplify tls_construct_server_key_exchange

Use negotiated signature algorithm and certificate index in
tls_construct_key_exchange instead of recalculating it.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2623)

Use cert_index and sigalg

Now the certificate and signature algorithm is set in one place we
can use it directly insetad of recalculating it. The old functions
ssl_get_server_send_pkey() and ssl_get_server_cert_index() are no
longer required.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2623)

Add sigalg for earlier TLS versions

Update tls_choose_sigalg to choose a signature algorithm for all
versions of TLS not just 1.3.

For TLS 1.2 we choose the highest preference signature algorithm
for the chosen ciphersuite.

For TLS 1.1 and earlier the signature algorithm is determined by
the ciphersuite alone. For RSA we use a special MD5+SHA1 signature
algorithm.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2623)

Change tls_choose_sigalg so it can set errors and alerts.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2623)

add ssl_has_cert

Add inline function ssl_has_cert which checks to see if a certificate and
private key for a given index are not NULL.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2623)

Fix a few typos
[skip ci]

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2571)

Remove obsolete comment

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1613)

Prevents that OPENSSL_gmtime incorrectly signals success if gmtime_r fails, and that struct* tm result's possibly uninitialized content is used

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1613)

Use TLSEXT_KEYNAME_LENGTH in tls_decrypt_ticket.

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2618)

Prevent allocations of size 0 in sh_init, which are not possible with the default OPENSSL_zalloc, but are possible if the user has installed their own allocator using CRYPTO_set_mem_functions. If the 0-allocations succeeds, the secure heap code will later access (at least) the first byte of that space, which is technically an OOB access. This could lead to problems with some custom allocators that only return a valid pointer for subsequent free()-ing, and do not expect that the pointer is actually dereferenced.

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2605)

Add Sieve support (RFC 5804) to s_client ("-starttls sieve")

Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2300)

Add no-ec build

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2626)

Make -xcert work again.

When a certificate is prepended update the list pointer.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2628)

Fix no-ec compilation

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2624)

Remove a double call to ssl3_send_alert()

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Fix a bogus uninit variable warning

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Add a bytestogroup macro

For converting the 2 byte group id into an unsigned int.

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Various style fixes following review feedback

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Update the tls13messages test to add some HRR scenarios

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Update the kex modes tests to check various HRR scenarios

Make sure we get an HRR in the right circumstances based on kex mode.

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Update TLSProxy to know about HelloRetryRequest messages

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Update test counting in checkhandshake.pm

Previously counting the number of tests in checkhandshake.pm took an
initial guess and then modified it based on various known special
cases. That is becoming increasingly untenable, so this changes it to
properly calculate the number of tests we expect to run.

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Update the key_share tests for HelloRetryRequest

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Add trace support for HelloRetryRequest

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Implement support for resumption with a HelloRetryRequest

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Add client side support for parsing Hello Retry Request

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Add server side support for creating the Hello Retry Request message

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

Make the context available to the extensions parse and construction funcs

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2341)

mem leak on error path and error propagation fix

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2559)

aes/asm/*-x86_64.pl: add CFI annotations.

Reviewed-by: Rich Salz <rsalz@openssl.org>

perlasm/x86_64-xlate.pl: recognize even offset(%reg) in cfa_expression.

This is handy when "offset(%reg)" is a perl variable.

Reviewed-by: Rich Salz <rsalz@openssl.org>

ec/asm/ecp_nistz256-x86_64.pl: add CFI directives.

Reviewed-by: Rich Salz <rsalz@openssl.org>

ec/asm/ecp_nistz256-x86_64.pl: fix typo-bug in Win64 SE handler.

Thanks to Jun Sun for spotting this.

Reviewed-by: Rich Salz <rsalz@openssl.org>

Further improvements to ASYNC_WAIT_CTX_clear_fd

Remove call to cleanup function
Use only one loop to find previous element

Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2581)

Remove fd from the list when the engine clears the wait context before pause

This fixes the num of fds added/removed returned by ASYNC_WAIT_CTX_get_changed_fds

Previously, the numbers were not consistent with the fds actually written in
the buffers since the fds that have been both added and removed are explicitly
ignored in the loop.

Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2581)

Add test to show wrong behavior of ASYNC_WAIT_CTX

This happens when a fd is added and then immediately removed from the
ASYNC_WAIT_CTX before pausing the job.

Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2581)

{md5,rc4}/asm/*-x86_64.pl: add CFI annotations.

Reviewed-by: Rich Salz <rsalz@openssl.org>

modes/asm/*-x86_64.pl: add CFI annotations.

Reviewed-by: Rich Salz <rsalz@openssl.org>

DES keys are not 7 days long.

CLA: trivial
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2604)

test_rehash does nothing, have it do something

test/recipes/40-test_rehash.t uses test files from certs/demo, which
doesn't exist any longer.  Have it use PEM files from test/ instead.

Because rehash wants only one certificate or CRL per file, we must
also filter those PEM files to produce test files with a single object
each.

Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2594)

sha/asm/sha1-x86_64.pl: add CFI annotations.

Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Andy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2590)

Replace SSL_PKEY_RSA_ENC, SSL_PKEY_RSA_SIGN

The original intent of SSL_PKEY_RSA_SIGN and SSL_PKEY_RSA_ENC was to
support two different keys for RSA signing and decrypt. However this
was never implemented and we only ever set one key and the other was
always NULL. Replace with single SSL_PKEY_RSA type.

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2587)

Check index >= 0 as 0 is a valid index.

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2587)

perlasm/x86_64-xlate.pl: fix pair of typo-bugs in the new cfi_directive.

.cfi_{start|end}proc and .cfi_def_cfa were not tracked.
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2585)

perlasm/x86_64-xlate.pl: typo fix in comment.

CLA: trivial

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Andy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2584)

Fix copy-pasteism in CFI directives.

I don't think this actually affects anything since the cfi_restore
directives aren't strictly needed anyway. (The old values are still in
memory so either will do.)

CLA: trivial

Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Andy Polyakov <appro@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2582)

Restore EVP_CIPH_FLAG_LENGTH_BITS working properly

EVP_CIPH_FLAG_LENGTH_BITS flag for CFB1 has been broken with the
introduction of the is_partially_overlapping() check that did not take
it into the account (treating number of bits passed as bytes). This
remedies that and allows this flag to work as intended.

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1942)

speed.c: simplify aggregation of ecdh --multi results

CLA: trivial

Signed-off-by: Cristian Stoica <cristian.stoica@nxp.com>
Reviewed-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1377)

speed.c: simplify aggregation of ecdsa --multi results

CLA: trivial

Signed-off-by: Cristian Stoica <cristian.stoica@nxp.com>
Reviewed-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1377)

speed.c: simplify aggregation of dsa --multi results

CLA: trivial

Signed-off-by: Cristian Stoica <cristian.stoica@nxp.com>
Reviewed-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1377)

speed.c: simplify aggregation of rsa --multi results

CLA: trivial

Signed-off-by: Cristian Stoica <cristian.stoica@nxp.com>
Reviewed-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1377)

Don't read uninitialised data for short session IDs.

While it's always safe to read |SSL_MAX_SSL_SESSION_ID_LENGTH| bytes
from an |SSL_SESSION|'s |session_id| array, the hash function would do
so with without considering if all those bytes had been written to.

This change checks |session_id_length| before possibly reading
uninitialised memory. Since the result of the hash function was already
attacker controlled, and since a lookup of a short session ID will
always fail, it doesn't appear that this is anything more than a clean
up.

In particular, |ssl_get_prev_session| uses a stack-allocated placeholder
|SSL_SESSION| as a lookup key, so the |session_id| array may be
uninitialised.

This was originally found with libFuzzer and MSan in
https://boringssl.googlesource.com/boringssl/+/e976e4349d693b4bbb97e1694f45be5a1b22c8c7,
then by Robert Swiecki with honggfuzz and MSan here. Thanks to both.

Reviewed-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/2583)