rand_unix.c: Ensure requests to KERN_ARND don't exceed 256 bytes.

Requests for more than 256 bytes will fail.

Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11689)

(cherry picked from commit 7421f085005e0d7a1dd2fe61b991ff23cef91c22)

rand_unix.c: Only enable hack for old FreeBSD versions on FreeBSD

Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11689)

(cherry picked from commit 0c27ce7322e15a7e0733d48aaf9c8e1d1249541f)

rand_unix.c: Include correct headers for sysctl() on NetBSD

This allows sysctl(KERN_ARND) to be detected properly.

Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11689)

(cherry picked from commit e2e4b784e65eaafb133a7db3d344446c43112d41)

Update dgst.pod

Removed "Use the B<pkeyutl> command instead for this." from lines 97-98
Fixes #11633

CLA: trivial

Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/11636)

sslapitest: only compile test when it will be used

The test_ccs_change_cipher() test routine is used only when TLS 1.2
is enabled; to fix the strict-warnings build we should not try to
compile it when TLS 1.2 is disabled, either.

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11458)

(cherry picked from commit fb121631e702352d47049ac8e2e709b11c3318e4)

fuzz/asn1.c: Add missing #include

<openssl/dsa.h> gets included via ts.h...  except when 'no-ts' has been
configured.

Fixes #11597

Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/11640)

(cherry picked from commit 60ebc0ca5a829e2ae939a9ab13658af202b6dfc7)

i2b_PVK_bio: don't set PEM_R_BIO_WRITE_FAILURE in case of success

but in case of an error

CLA: trivial

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11380)

(cherry picked from commit 80b94a5adb461f94629d36db351a051b0a890856)

Fix type cast in SSL_CTX_set1_groups macro

The macro casts the glist parameter to char*
instead of (int *) like the documentation of the function suggest.

Also the function tls1_set_groups that is called from SSL_CTX_ctrl
takes an int * argument. This looks like a copy&paste error from
SSL_CTX_set1_groups_list function.

CLA: trivial
Signed-off-by: Arne Schwabe <arne@rfc2549.org>
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11614)

(cherry picked from commit 7ffce852372799b6cd856b711db21332f0048314)

Fix typo from #10631

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/11595)

(cherry picked from commit 4692e98bdbaebb6f078e89a75c54395839e68b24)

[BN] harden `BN_copy()` against leaks from memory accesses

`BN_copy()` (and indirectly `BN_dup()`) do not propagate the
`BN_FLG_CONSTTIME` flag: the propagation has been turned on and off a
few times in the past years, because in some conditions it has shown
unintended consequences in some code paths.

Without turning the propagation on once more, we can still improve
`BN_copy()` by avoiding to leak `src->top` in case `src` is flagged with
`BN_FLG_CONSTTIME`.
In this case we can instead use `src->dmax` as the number of words
allocated for `dst` and for the `memcpy` operation.

Barring compiler or runtime optimizations, if the caller provides `src`
flagged as const time and preallocated to a public size, no leak should
happen due to the copy operation.

(cherry picked from commit 2d9167ed0b588dacbdd0303fb6041ffe1d8b3a92)

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11127)

[EC] harden EC_KEY against leaks from memory accesses

We should never leak the bit length of the secret scalar in the key,
so we always set the `BN_FLG_CONSTTIME` flag on the internal `BIGNUM`
holding the secret scalar.

This is important also because `BN_dup()` (and `BN_copy()`) do not
propagate the `BN_FLG_CONSTTIME` flag from the source `BIGNUM`, and
this brings an extra risk of inadvertently losing the flag, even when
the called specifically set it.

The propagation has been turned on and off a few times in the past
years because in some conditions has shown unintended consequences in
some code paths, so at the moment we can't fix this in the BN layer.

In `EC_KEY_set_private_key()` we can work around the propagation by
manually setting the flag after `BN_dup()` as we know for sure that
inside the EC module the `BN_FLG_CONSTTIME` is always treated
correctly and should not generate unintended consequences.

Setting the `BN_FLG_CONSTTIME` flag alone is never enough, we also have
to preallocate the `BIGNUM` internal buffer to a fixed public size big
enough that operations performed during the processing never trigger
a realloc which would leak the size of the scalar through memory
accesses.

Fixed Length
------------

The order of the large prime subgroup of the curve is our choice for
a fixed public size, as that is generally the upper bound for
generating a private key in EC cryptosystems and should fit all valid
secret scalars.

For preallocating the `BIGNUM` storage we look at the number of "words"
required for the internal representation of the order, and we
preallocate 2 extra "words" in case any of the subsequent processing
might temporarily overflow the order length.

Future work
-----------

A separate commit addresses further hardening of `BN_copy()` (and
indirectly `BN_dup()`).

(cherry picked from commit 0401d766afcd022748763f5614188301c9856c6e)

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11127)

[EC] Constify internal EC_KEY pointer usage

A pair of internal functions related to EC_KEY handling could benefit
from declaring `EC_KEY *` variables as `const`, providing clarity for
callers and readers of the code, in addition to enlisting the compiler
in preventing some mistakes.

(cherry picked from commit cd701de96a147260c2290d85af8a0656120a8ff8)

In master `id2_ECParameters` and most of the ASN1 public functions have
been properly constified in their signature.

Unfortunately this has been deemed not doable in a patch release for
1.1.1 as, in subtle ways, this would break API compatibility.
See the discussion at https://github.com/openssl/openssl/pull/9347 for
more details about this.

This constification commit should still be portable w.r.t. our criteria,
as the constification happens only on internal functions.

The fix here is to explicitly discard the const qualifier before the
call to `i2d_ECParameters`, which should be safe anyway because we can
expect `i2d_ECParameters()` to treat the first argument as if it was
const.

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11127)

Add setter equivalents to X509_REQ_get0_signature

Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/10563)

(cherry picked from commit c72e59349f50ee00a1bf8605ada17dfccb8b3b1a)

Prepare for 1.1.1h-dev

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

Prepare for 1.1.1g release

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

Update copyright year

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

Fix NULL dereference in SSL_check_chain() for TLS 1.3

In the tls1_check_sig_alg() helper function, we loop through the list of
"signature_algorithms_cert" values received from the client and attempt
to look up each one in turn in our internal table that maps wire
codepoint to string-form name, digest and/or signature NID, etc., in
order to compare the signature scheme from the peer's list against what
is used to sign the certificates in the certificate chain we're
checking.  Unfortunately, when the peer sends a value that we don't
support, the lookup returns NULL, but we unconditionally dereference the
lookup result for the comparison, leading to an application crash
triggerable by an unauthenticated client.

Since we will not be able to say anything about algorithms we don't
recognize, treat NULL return from lookup as "does not match".

We currently only apply the "signature_algorithm_cert" checks on TLS 1.3
connections, so previous TLS versions are unaffected.  SSL_check_chain()
is not called directly from libssl, but may be used by the application
inside a callback (e.g., client_hello or cert callback) to verify that a
candidate certificate chain will be acceptable to the client.

CVE-2020-1967

Reviewed-by: Matt Caswell <matt@openssl.org>

Add test for CVE-2020-1967

Add to test_sslsigalgs a TLSProxy test that injects a
"signature_algorithms_cert" extension that contains an unallocated
codepoint.

The test currently fails, since s_server segfaults instead of
ignoring the unrecognized value.

Since "signature_algorithms" and "signature_algorithms_cert" are very
similar, also add the analogous test for "signature_algorithms".

Reviewed-by: Matt Caswell <matt@openssl.org>

Update CHANGES and NEWS for release of 1.1.1g

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

Add a CHANGES entry for AES constant time

Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/11411)

Add AES consttime code for no-asm configurations

This adds optional constant time support for AES
when building openssl for no-asm.

Enable with: ./config no-asm -DOPENSSL_AES_CONST_TIME

[extended tests]

Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/11411)

s_time: Allow using -CAfile option as in other commands

The s_time command in difference from all the other similar
commands supported -cafile option instead of -CAfile.
Add the -CAfile option and keep -cafile only for backwards
compatibility.

Fixes #11552

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11555)

INSTALL: document 'no-ui-console' rather than 'no-ui'

The UI interface itself is never disabled, but the console backend may
be. 'no-ui' is a deprecated backward compatibility alias for
'no-ui-console'.

Fixes #11551

Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11554)

Windows: Add type casting in CRYPTO_atomic_add to remove warning

CLA: trivial

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11337)

(cherry picked from commit 7da7b27eec58d1efc7012f002c45ddbdd61a5e79)

BIO_do_accept: correct error return value

`BIO_do_accept` was returning incorrect values when unable to bind to a port.

Fixes #7717
CLA: trivial

Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11505)

(cherry picked from commit 0437435a960123be1ced766d18d715f939698345)

Don't compile commands if disabled

Rather than wrapping whole files in "ifndef OPENSSL_NO_xxx" we handle
the changes in build.info

Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/11518)

Add manpage entry for X509_check_purpose()

Fixes #10263

Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/10301)

(cherry picked from commit eacd30a703b52f5b3e8e70b58f99bd1128458c7c)

Update RAND_METHOD definition in man page

The `add` and `seed` callbacks were changed to return `int` instead of
`void` in b6dcdbfc94c482f6c15ba725754fc9e827e41851 (first included in
tag OpenSSL_1_1_0-pre1).

The `add` callback was changed to take a `double` instead of an `int`
in 853f757ecea74a271a7c5cdee3f3b5fe0d3ae863.

CLA: trivial
Fixes: #10199
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11486)

(cherry picked from commit ae3254287ff87e484c7fd8f757cad1440ee8f5ff)

Fix AES-CTR_DRBG on 1.1.1.

The backport of the timing information leak fix uses u32 which is defined
in crypto/modes/modes_local.h in 1.1.1 and include/crypto/modes.h for 3.0.

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11489)

AES CTR-DRGB: do not leak timing information

Signed-off-by: Patrick Steuer <patrick.steuer@de.ibm.com>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/11147)

(cherry picked from commit 069165d10646a22000c596095cc04d43bbf1f807)

Integer overflow in ASN1_STRING_set.

Addressing a potential integer overflow condition.

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11473)

(cherry picked from commit 96218269f4c2da82f143727fb7697d572c190bc5)

[crypto/ec] blind coordinates in ec_wNAF_mul for robustness

Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
Reviewed-by: Nicola Tuveri <nicola.tuveri@ibm.com>
(Merged from https://github.com/openssl/openssl/pull/11439)

(cherry picked from commit c61ced5ec50fc68707c7cea79f7df1d170f03f13)

Fix the error handling in EC_POINTs_mul

This was pointed out by a false-positive
-fsanitizer warning ;-)

However from the cryptographical POV the
code is wrong:
A point R^0 on the wrong curve
is infinity on the wrong curve.

[extended tests]

Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/11475)

(cherry picked from commit 1eb9b54af7e00fa12196411964ce742ea8677766)

Ensure ECDSA_size always returns >= 0

Fixes #10484

Signed-off-by: William Brawner <me@wbrawner.com>
Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11472)

Add test cases for the non CA certificate with pathlen:0

Accept verification without -x509_strict and reject it with it.

Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11463)

(cherry picked from commit 3cb55fe47c3398b81956e4fe20c4004524d47519)

Set X509_V_ERR_INVALID_EXTENSION error for invalid basic constraints

If we encounter certificate with basic constraints CA:false,
pathlen present and X509_V_FLAG_X509_STRICT is set we set
X509_V_ERR_INVALID_EXTENSION error.

Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11463)

(cherry picked from commit fa86e2ee3533bb7fa9f3c62c38920cf960e9fec0)

Allow certificates with Basic Constraints CA:false, pathlen:0

Do not mark such certificates with EXFLAG_INVALID although they
violate the RFC 5280, they are syntactically correct and
openssl itself can produce such certificates without any errors
with command such as:

openssl x509 -req -signkey private.pem -in csr.pem -out cert.pem \
  -extfile <(echo "basicConstraints=CA:FALSE,pathlen:0")

With the commit ba4356ae4002a04e28642da60c551877eea804f7 the
EXFLAG_INVALID causes openssl to not consider such certificate
even as leaf self-signed certificate which is breaking existing
installations.

Fixes: #11456
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11463)

(cherry picked from commit 428cf5ff83a48d0b51c97476586b2cbd053b6302)

Expand the XTS documentation

Explain that XTS does not support streaming, and that the IV value is the
tweak.

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/11461)

(cherry picked from commit 70d80ef9898ddbe03841efd2df2f526b71829d7f)

[crypto/ec] Ladder tweaks

- Convert to affine coords on ladder entry. This lets us use more efficient
  ladder step formulae.

- Convert to affine coords on ladder exit. This prevents the current code
  awkwardness where conversion happens twice during serialization: first to
  fetch the buffer size, then again to fetch the coords.

- Instead of projectively blinding the input point, blind both accumulators
  independently.

(cherry picked from commit a4a93bbfb0e679eaa249f77c7c4e7e823ca870ef)

Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/11435)

[test] Make sm2_internal_test less fragile to changes in the ec module

Since these are KATs, the trailing randomness consumed by the ec module
does not really matter. So make the fake random buffer circular.

(cherry picked from commit 09736245b174a37abb87fb7ceb55462d940ff2bb)

Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/11435)

Prepare for 1.1.1g-dev

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

Prepare for 1.1.1f release

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

Update copyright year

Reviewed-by: Paul Dale <paul.dale@oracle.com>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11445)

fix generated Makefile for Windows builds made on *nix (1.1.1d)

The fix consists of putting all destination directories
between double-quotes to make the default (and any custom)
prefixes containing spaces to work when doing 'make install'.

Also enable CI test with x86 mingw cross-build.

[extended tests]

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Bernd Edlinger <bernd.edlinger@hotmail.de>
(Merged from https://github.com/openssl/openssl/pull/11434)

Document the revert of the proper reporting of an unexpected EOF

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11400)

Partially revert "Detect EOF while reading in libssl"

This partially reverts commit db943f43a60d1b5b1277e4b5317e8f288e7a0a3a.

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11400)

Use ctx2 instead ctx.

CLA: trivial

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Ben Kaduk <kaduk@mit.edu>
(Merged from https://github.com/openssl/openssl/pull/11372)

(cherry picked from commit 402b00d57921a0c8cd641b190d36bf39ea5fb592)

Fix error handling in x509v3_cache_extensions and related functions

Basically we use EXFLAG_INVALID for all kinds of out of memory and
all kinds of parse errors in x509v3_cache_extensions.

[extended tests]

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/10756)

Coverity: fix two minor NPD issues.
Found by Coverity.

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/8274)

(cherry picked from commit 23dc8feba817560485da00d690d7b7b9e5b15682)

Add a CHANGES entry for BN_generate_prime_ex

BN_generate_prime_ex no longer avoids factors 3..17863 in p-1
when not computing safe primes.

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/9387)

Update documentation of BN_generate_prime_ex

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/9387)

Merge probable_prime_dh_safe with bn_probable_prime_dh

This should avoid half of the trial divisions in probable_prime_dh_safe
and avoid bn_probable_prime_dh generating primes with special properties.

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/9387)

Add a parameter to probable_prime if we look for a safe prime

Currently probable_prime makes sure that p-1 does not have
any prime factors from 3..17863, which is useful for safe primes,
but not necessarily for the general case.

Issue was initially reported here:
MIRONOV, I. Factoring RSA Moduli II.
https://windowsontheory.org/2012/05/17/factoring-rsa-moduli-part-ii/

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/9387)

Prepare for 1.1.1f-dev

Reviewed-by: Paul Yang <kaishen.yy@antfin.com>

Prepare for 1.1.1e release

Reviewed-by: Paul Yang <kaishen.yy@antfin.com>

Update copyright year

Reviewed-by: Paul Yang <kaishen.yy@antfin.com>
(Merged from https://github.com/openssl/openssl/pull/11344)

Update CHANGES for the new release

Reviewed-by: Mark J. Cox <mark@awe.com>
(Merged from https://github.com/openssl/openssl/pull/11342)

Update NEWS for the new release

Reviewed-by: Mark J. Cox <mark@awe.com>
(Merged from https://github.com/openssl/openssl/pull/11342)

sslapitest: don't leak the SSL_CTX pair

We have no need for a new set of SSL_CTXs in test_ccs_change_cipher(), so
just keep using the original ones.  Also, fix a typo in a comment.

[extended tests]

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11336)

(cherry picked from commit b3e6d666e351d45e93d29fe3813245b92a0f5815)

Add test that changes ciphers on CCS

The TLS (pre-1.3) ChangeCipherState message is usually used to indicate
the switch from the unencrypted to encrypted part of the handshake.
However, it can also be used in cases where there is an existing
session (such as during resumption handshakes) or when changing from
one cipher to a different one (such as during renegotiation when the
cipher list offered by the client has changed).  This test serves
to exercise such situations, allowing us to detect whether session
objects are being modified in cases when they must remain immutable
for thread-safety purposes.

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/10943)

(cherry picked from commit 3cd14e5e65011660ad8e3603cf871c8366b565fd)

Code to thread-safety in ChangeCipherState

The server-side ChangeCipherState processing stores the new cipher
in the SSL_SESSION object, so that the new state can be used if
this session gets resumed.  However, writing to the session is only
thread-safe for initial handshakes, as at other times the session
object may be in a shared cache and in use by another thread at the
same time.  Reflect this invariant in the code by only writing to
s->session->cipher when it is currently NULL (we do not cache sessions
with no cipher).  The code prior to this change would never actually
change the (non-NULL) cipher value in a session object, since our
server enforces that (pre-TLS-1.3) resumptions use the exact same
cipher as the initial connection, and non-abbreviated renegotiations
have produced a new session object before we get to this point.
Regardless, include logic to detect such a condition and abort the
handshake if it occurs, to avoid any risk of inadvertently using
the wrong cipher on a connection.

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/10943)

(cherry picked from commit 2e3ec2e1578977fca830a47fd7f521e290540e6d)

Don't write to the session when computing TLS 1.3 keys

TLS 1.3 maintains a separate keys chedule in the SSL object, but
was writing to the 'master_key_length' field in the SSL_SESSION
when generating the per-SSL master_secret.  (The generate_master_secret
SSL3_ENC_METHOD function needs an output variable for the master secret
length, but the TLS 1.3 implementation just uses the output size of
the handshake hash function to get the lengths, so the only natural-looking
thing to use as the output length was the field in the session.
This would potentially involve writing to a SSL_SESSION object that was
in the cache (i.e., resumed) and shared with other threads, though.

The thread-safety impact should be minimal, since TLS 1.3 requires the
hash from the original handshake to be associated with the resumption
PSK and used for the subsequent connection.  This means that (in the
resumption case) the value being written would be the same value that was
previously there, so the only risk would be on architectures that can
produce torn writes/reads for aligned size_t values.

Since the value is essentially ignored anyway, just provide the
address of a local dummy variable to generate_master_secret() instead.

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/10943)

(cherry picked from commit d74014c4b8740f28a54b562f799ad1e754b517b9)

Fix whitespace nit in ssl_generate_master_secret()

Use a space after a comma.

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/10943)

(cherry picked from commit 1866a0d380fc361d9be2ca0509de0f2281505db5)

doc: fix spelling of TYPE_get_ex_new_index

The generated macros are TYPE_get_ex_new_index() (to match
CRYPTO_get_ex_new_index()), not TYPE_get_new_ex_index(), even though
the latter spelling seems more natural.

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/10943)

(cherry picked from commit fe41c06e69613b1a4814b3e3cdbf460f2678ec99)

Additional updates to SSL_CTX_sess_set_get_cb.pod

Generally modernize the language.

Refer to TLS instead of SSL/TLS, and try to have more consistent
usage of commas and that/which.

Reword some descriptions to avoid implying that a list of potential
reasons for behavior is an exhaustive list.

Clarify how get_session_cb() is only called on servers (i.e., in general,
and that it's given the session ID proposed by the client).

Clarify the semantics of the get_cb()'s "copy" argument.
The behavior seems to have changed in commit
8876bc054802b043a3ec95554b6c5873291770be, though the behavior prior
to that commit was not to leave the reference-count unchanged if
*copy was not written to -- instead, libssl seemed to assume that the
callback already had incremented the reference count.

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/10943)

(cherry picked from commit 06f876837a8ec76b28c42953731a156c0c3700e2)

doc: Update the reference from draft to RFC

CLA: trivial

Signed-off-by: Jakub Jelen <jjelen@redhat.com>
Reviewed-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11299)

(cherry picked from commit c08dea30d4d127412097b39d9974ba6090041a7c)

doc: document that 'openssl rand' is cryptographically secure

(cherry picked from commit 88398d2a358f)

Additionally, remove an outdated paragraph mentioning the .rnd
file, which is obsolete in 1.1.1 since the RANDFILE entry was
removed from openssl.cnf in commit 1fd6afb571e8.

Also borrow some text from 'openssl(1)/Random State Options'
on master (commit a397aca43598) to emphasize that it is not
necessary anymore to restore and save the RNG state using the
'-rand' and '-writerand' options.

Reviewed-by: Paul Dale <paul.dale@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/11251)

Revert "Create a new embeddedSCTs1 that's signed using SHA256"

This reverts commit b98efebeb2d4265bd6638d5947fe365500121e03.

Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11282)

Revert "Stop accepting certificates signed using SHA1 at security level 1"

This reverts commit 68436f0a8964e911eb4f864bc8b31d7ca4d29585.

The OMC did not vote in favour of backporting this to 1.1.1, so this
change should be reverted.

Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11282)

docs: fix typo in SSL functions

CLA: trivial

Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
Reviewed-by: Paul Yang <kaishen.yy@antfin.com>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11253)

(cherry picked from commit 9f44e96e245993c8e7aaa9fadf1d6713c9c60915)

DOC: Make EVP_SignInit.pod conform with man-pages(7)

Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
(cherry picked from commit 03d65ca2095777cf6314ad813eb7de5779c9b93d)

Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/11232)

DOC: New file for EVP_PKEY_size(), EVP_PKEY_bits() and EVP_PKEY_security_bits()

We change the description to be about the key rather than the
signature.  How the key size is related to the signature is explained
in the description of EVP_SignFinal() anyway.

Reviewed-by: Nicola Tuveri <nic.tuv@gmail.com>
(cherry picked from commit 6942a0d6feb8d3dcbbc6a1ec6be9de7ab2df1530)

Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/11232)

apps x509: passing PKCS#11 URL as -signkey

OpenSSL 1.1.0 has extended option checking, and rejects passing a PKCS#11
engine URL to "-signkey" option. The actual code is ready to take it.

Change the option parsing to allow an engine URL to be passed and modify
the manpage accordingly.

CLA: trivial

(cherry picked from commit 16d560439d8b1be5082228a87576a8f79b3525ac)

Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11173)

Clarify the usage of EVP_PKEY_get_raw_[private|public]_key()

EVP_PKEY_get_raw_private_key() and EVP_PKEY_get_raw_public_key() expect
the size of the key buffer to be populated in the |*len| parameter on
entry - but the docs made no mention of this.

Fixes #11245

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11254)

(cherry picked from commit f529fc7d53bf4228fae61cb1efe73d97fe3eb35f)

DOC: Fixups of X509_LOOKUP.pod

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11120)

DOC: Add documentation related to X509_LOOKUPs

Most of all, the base X509_LOOKUP functionality is now documented.
Furthermore, the names X509_LOOKUP_METHOD and X509_STORE are added for
reference.

Some functions were moved from X509_LOOKUP_meth_new.pod

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11120)

crypto/ec/curve448/eddsa.c: fix EBCDIC platforms

Signed-off-by: Patrick Steuer <patrick.steuer@de.ibm.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11229)

(cherry picked from commit af7f656cd91d99d62567e2b20c61f07cb4d81d0b)

aes-s390x.pl: fix stg offset caused by typo in perlasm

Signed-off-by: Patrick Steuer <patrick.steuer@de.ibm.com>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11234)

(cherry picked from commit 7b2ce4a6e817e4385ff77fea0c6e349294c7b756)

This works around a gcc-9 crash

It happens when configured with ./config -ftest-coverage

see: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=94029

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11246)

Fix comment placement in ecp_nistp256.ci

CLA: trivial

Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/11175)

(cherry picked from commit c590be6f12d0b725863961e41bc64a81c8cf30d6)

apps x509: restrict CAkeyform option to OPT_FMT_PDE

CAkeyform may be set to PEM, DER or ENGINE, but the current options
are not using the proper optionformat 'E' (OPT_FMT_PDE) for this.

Set the valtype for CAkeyform to 'E' and use OPT_FMT_PDE when extracting
the option value.

This amends bf4006a6f9 ("Fix regression on x509 keyform argument") which
did the same thing for keyform and changed the manpage synopsis entries
for both keyform and CAkeyform but did not change the option section.
Hence, change the option section.

CLA: trivial

Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11172)

VMS: mitigate for the C++ compiler that doesn't understand certain pragmas

This only affects __DECC_INCLUDE_EPILOGUE.H and __DECC_INCLUDE_PROLOGUE.H,
which are used automatically by HP and VSI C/C++ compilers.

Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/11159)

config: Drop linux-alpha-gcc+bwx

Its entry in Configuration/10-main.conf was dropped in commit
7ead0c89185c ("Configure: fold related configurations more aggressively
and clean-up.") probably because all but one of its bn_ops were removed
(RC4_CHAR remained). Benchmarks on an Alpha EV7 indicate that RC4_INT is
better than RC4_CHAR so rather than restoring the configuation, remove
it from config.

CLA: trivial
Bug: https://bugs.gentoo.org/697840

(cherry picked from commit 19ded1a717b6c72c3db241f06787a353f1190755)
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11130)

Teach more BIOs how to handle BIO_CTRL_EOF

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/10882)

Detect EOF while reading in libssl

If we hit an EOF while reading in libssl then we will report an error
back to the application (SSL_ERROR_SYSCALL) but errno will be 0. We add
an error to the stack (which means we instead return SSL_ERROR_SSL) and
therefore give a hint as to what went wrong.

Contains a partial fix for #10880

Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/10882)

Add *.d.tmp files to .gitignore

These are temporary files generated by the build process that should not
be checked in.

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11122)

(cherry picked from commit 7b5108dff4cfde059ca278147a188fb6254603d1)

Handle max_fragment_length overflow for DTLS

Allow for encryption overhead in early DTLS size check
and send overflow if validated record is too long

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/11096)

(cherry picked from commit cc0663f697b05ed121a728241f0502250429802d)

Do not silently truncate files on perlasm errors

If one of the perlasm xlate drivers crashes, OpenSSL's build will
currently swallow the error and silently truncate the output to however
far the driver got. This will hopefully fail to build, but better to
check such things.

Handle this by checking for errors when closing STDOUT (which is a pipe
to the xlate driver).

This is the OpenSSL 1.1.1 version of
https://github.com/openssl/openssl/pull/10883 and
https://github.com/openssl/openssl/pull/10930.

Reviewed-by: Mark J. Cox <mark@awe.com>
Reviewed-by: Paul Dale David Benjamin <davidben@google.com>
(Merged from https://github.com/openssl/openssl/pull/10931)

Check that ed25519 and ed448 are allowed by the security level

Signature algorithms not using an MD weren't checked that they're
allowed by the security level.

Reviewed-by: Tomáš Mráz <tmraz@fedoraproject.org>
GH: #11062

Generate new Ed488 certificates

Create a whole chain of Ed488 certificates so that we can use it at security
level 4 (192 bit). We had an 2048 bit RSA (112 bit, level 2) root sign the
Ed488 certificate using SHA256 (128 bit, level 3).

Reviewed-by: Matt Caswell <matt@openssl.org>
GH: #10785
(cherry picked from commit 77c4d3972400adf1bcb76ceea359f5453cc3e8e4)

DOC: document in more detail what a BIO_read_ex() via BIO_f_buffer() does

The BIO_f_buffer() documentation tells in enough detail how it affects
BIO_gets(), but not how it affects BIO_read_ex().  This change
remedies that.

Fixes #10859

Reviewed-by: Tim Hudson <tjh@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/10890)

(cherry picked from commit 9a4fd80ee0ad1833879b6a55c9c4673eeb8446a3)

VMS: Correct error reporting in crypto/rand/rand_vms.c

The future style that's coming with OpenSSL 3.0 was used, we need to
revert that back to "traditional" style.

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11088)

VMS: Adapt descrip.mms template to the changed inclustion dirs

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11088)

Adjust minimal build config in 1.1.1 branch

Backport of improvements from #9982 to 1.1.1 branch.
Adds some more exclusions which were previously missed.

[extended tests]

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11059)

Forgotten GOST2012 support in non-vital places

Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
Reviewed-by: Dmitry Belyavskiy <beldmit@gmail.com>
(Merged from https://github.com/openssl/openssl/pull/11072)

Fix no-tls1_3

The hostname_cb in sslapitest.c was originally only defined if TLSv1.3
was enabled. A recently added test now uses this unconditionally, so we
move the function implementation earlier in the file, and always compile
it in.

Reviewed-by: Richard Levitte <levitte@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/11014)

(cherry picked from commit 104a733df65dfd8c3dd110de9bd56f6ebfc8f2f6)

Appveyor: update to Visual Studio 2017.

Default image was currently "Visual Studio 2015"

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
(Merged from https://github.com/openssl/openssl/pull/10327)

(cherry picked from commit b03de7a9207645c72e22627b10709f15eed211bf)

Avoid leak in error path of PKCS5_PBE_keyivgen

CLA: trivial

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Matthias St. Pierre <Matthias.St.Pierre@ncp-e.com>
Reviewed-by: Tomas Mraz <tmraz@fedoraproject.org>
(Merged from https://github.com/openssl/openssl/pull/10816)

(cherry picked from commit adc9086beb21a91ca59aaf0c619b38b82c223f9b)

Coverity CID 1444960: Error handling issues

Reviewed-by: Matt Caswell <matt@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/8888)

(cherry picked from commit a05bf83c7964bb3928b323fe356b9f70f105036d)

Fix misspelling errors and typos reported by codespell

Fixes #10998

Reviewed-by: Shane Lontis <shane.lontis@oracle.com>
(Merged from https://github.com/openssl/openssl/pull/11000)