Allow OPENSSL_NO_SOCK in e_os.h even for non-Windows/DOS platforms

UEFI needs this too. Don't keep it only in the Windows/DOS ifdef block.

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

ISSUE 43: Add BIO_sock_shutdown

This replaces SHUTDOWN/SHUTDOWN2 with BIO_closesocket.

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

Minor update to includes and documentation for ct_test.c

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

Remove OPENSSL_NO_UNIT_TEST guard from ct_test.c

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

Use s->session->peer instead of calling SSL_get_peer_certificate(s)

Avoids modifying certificate reference count, and thereby avoids locking.

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

Lowercase name of SSL_validate_ct as it is an internal function

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

CT code now calls X509_free() after calling SSL_get_peer_certificate()

Without this, the peer certificate would never be deleted, resulting in
a memory leak.

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

Fixes memory leaks in CT code

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

Fix the build tree include directory for afalg engine

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

Make OpenSSL::Test::setup() a bit more forgiving

It was unexpected that OpenSSL::Test::setup() should be called twice
by the same recipe.  However, that may happen if a recipe combines
OpenSSL::Test and OpenSSL::Test::Simple, which can be a sensible thing
to do.  Therefore, we now allow it.

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

Elide EVP_read_pw_string() and friends for no-ui

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Kurt Roeckx <kurt@openssl.org>

GH768: Minor grammar nits in CRYPTO_get_ex_new_index.pod

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Kurt Roeckx <kurt@openssl.org>

Unified - adapt the generation of padlock assembler to use GENERATE

This gets rid of the BEGINRAW..ENDRAW sections in engines/build.info.

This also moves the assembler generating perl scripts to take the
output file name as last command line argument, where necessary.

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

Unified - adapt the generation of bignum assembler to use GENERATE

This gets rid of the BEGINRAW..ENDRAW sections in crypto/bn/build.info.

This also moves the assembler generating perl scripts to take the
output file name as last command line argument, where necessary.

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

Unified - Add the build.info command OVERRIDE, to avoid build file clashes

Should it be needed because the recipes within a RAW section might
clash with those generated by Configure, it's possible to tell it
not to generate them with the use of OVERRIDES, for example:

    SOURCE[libfoo]=foo.c bar.c

    OVERRIDES=bar.o
    BEGINRAW[Makefile(unix)]
    bar.o: bar.c
        $(CC) $(CFLAGS) -DSPECIAL -c -o $@ $<
    ENDRAW[Makefile(unix)]

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

Unified - Adapt the Unix and VMS templates to support GENERATE

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

Unified - Add the build.info command GENERATE, to generate source files

In some cases, one might want to generate some source files from
others, that's done as follows:

    GENERATE[foo.s]=asm/something.pl $(CFLAGS)
    GENERATE[bar.s]=asm/bar.S

The value of each GENERATE line is a command line or part of it.
Configure places no rules on the command line, except the the first
item muct be the generator file.  It is, however, entirely up to the
build file template to define exactly how those command lines should
be handled, how the output is captured and so on.

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

Add a function to detect if we have async or not

Add the ASYNC_is_capable() function and use it in speed.

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

GH804: Fix unused-result warnings in dasync

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>

documentation and duplicate goto statements

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

move ifdef statements

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

NIST SP800-56A co-factor ECDH KATs

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

Remove use of CRYPTO_LOCK_INIT in init code

Swap the use of CRYPTO_LOCK_INIT in the init code to use the new threading
API mechanism for locking.

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

Swap the init code to use the new Thread API thread locals

The init code was using its own thread local code. Now we have a central
API for it we should use that instead.

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

Swap the init code to use CRYPTO_ONCE

The init code was using its own "once" implementation. Now that we have
the new thread API we should use that instead.

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

make update

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

Update documentation

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

Make PKCS8_PRIV_KEY_INFO opaque.

Make PKCS8_PRIV_KEY_INFO opaque. Several accessor functions already exist
for this structure. Two new ones were added to handle attributes.

The old handling of broken formats has been removed and the corresponding
structures simplified.

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

Add support to ASYNC_WAIT_CTX to speed

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

Remove unnecessary memset() to 0 and check for NULL before OPENSSL_free()

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

Fix the error with RSA and the daysnc engine in async mode.

Move RSA struct in the job local struct.
The change is applied also to other crypto operations (e.g. DSA) to
make things consistent.

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

Add support for async jobs in OpenSSL speed

Summary of the changes:

* Move the calls to the crypto operations inside wrapper functions.
  This is required because ASYNC_start_job takes a function as an argument.

* Add new function run_benchmark() that manages the jobs for all the operations.
  In the POSIX case it uses a select() to receive the events from the engine
  and resume the jobs that are paused, while in the WIN case it uses PeekNamedPipe()

* Add new option argument async_jobs to enable and specify the number of async jobs

Example:
  openssl speed -engine dasync -elapsed -async_jobs 32 rsa2048

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

Rework the default cipherlist.

 - Always prefer forward-secure handshakes.
 - Consistently order ECDSA above RSA.
 - Next, always prefer AEADs to non-AEADs, irrespective of strength.
 - Within AEADs, prefer GCM > CHACHA > CCM for a given strength.
 - Prefer TLS v1.2 ciphers to legacy ciphers.
 - Remove rarely used DSS, IDEA, SEED, CAMELLIA, CCM from the default
   list to reduce ClientHello bloat.

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

Makefile.in: populate [PLIB_]LDFLAG even with $target{} settings.

RT#4373

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

bn/asm/x86[_64]-mont*.pl: complement alloca with page-walking.

Some OSes, *cough*-dows, insist on stack being "wired" to
physical memory in strictly sequential manner, i.e. if stack
allocation spans two pages, then reference to farmost one can
be punishable by SEGV. But page walking can do good even on
other OSes, because it guarantees that villain thread hits
the guard page before it can make damage to innocent one...

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

Restore some mingw builds

"no-pic" builds have in fact been green (and reasonably fast), so
restore them while we figure out why tests without "no-pic" hang.

Reviewed-by: Andy Polyakov <appro@openssl.org>

perlasm/x86_64-xlate.pl: handle binary constants early.

Not all assemblers of "gas" flavour handle binary constants, e.g.
seasoned MacOS Xcode doesn't, so give them a hand.

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

Fix some clang warnings

The af_alg engine and associated test were creating warnings when compiled
with clang. This fixes it.

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

Fix minor errors in the afalg test

The new afalg test should have a copyright date of 2016. Also an
incorrect buffer was being sent to EVP_CipherFinal_ex when
decrypting.

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

Adding afalg test

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

Fix OPENSSL_INIT flags to avoid a clash.

The 0x00010000L OPENSSL_INIT flag appeared twice.

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

Misc afalg build fixes

Misc afalg build fixes as suggested by Richard Levitte for the latest
Configure changes.

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

Rework based on feedback:
1. Cleaned up eventfd handling
2. Reworked socket setup code to allow other algorithms to be added in
   future
3. Fixed compile errors for static build
4. Added error to error stack in all cases of ALG_PERR/ALG_ERR
5. Called afalg_aes_128_cbc() from bind() to avoid race conditions
6. Used MAX_INFLIGHT define in io_getevents system call
7. Coding style fixes

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

ALG: Add AFALG engine

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

Trim Travis config part 2

- Remove Win builds (temporarily). They're slow, allowed to fail,
  and therefore not useful as they are.
- Make the --unified part of the matrix build-only. (This can be
  swapped if --unified becomes the default)
- Only build 'no-engine' once, don't run any tests, but don't allow it
  to fail.

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

Trim the Travis config

- Remove no-asm. We've got to cut something, and this is at least
  partially covered by the sanitizer builds.
- Remove enable-crypto-mdebug from sanitizer
  builds. enable-crypto-mdebug has been shown to catch some static
  initialization bugs that the standard leak sanitizer can't so
  perhaps it has _some_ value; but we shouldn't let the two compete.

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

GH616: Remove dead code

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>

No -fno-common for Darwin

When object files with common block symbols are added to static
libraries on Darwin, those symbols are invisible to the linker that
tries to use them.  Our solution was to use -fno-common when compiling
C source.

Unfortunately, there is assembler code that defines OPENSSL_ia32cap_P
as a common block symbol, unconditionally, and in some cases, there is
no other definition.  -fno-common doesn't help in this case.

However, 'ranlib -c' adds common block symbols to the index of the
static library, which makes them visible to the linker using it, and
that solves the problem we've seen.

The common conclusion is, either use -fno-common or ranlib -c on
Darwin.  Since we have common block symbols unconditionally, choosing
the method for our source is easy.

Reviewed-by: Kurt Roeckx <kurt@openssl.org>

New ordinal files, recreated from scratch

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

Change names of ordinals and libs, libeay => libcrypto and ssleay => libssl

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

Remove the old ordinals

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

use saner default parameters for scrypt

Thanks to Colin Percival for reporting this issue.

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

Disable SSLv3 by default

Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Emilia Kasper <emilia@openssl.org>
MR: #2203

Don't mark the eNULL ciphers as non-default.

They're not part of ALL, so they're not part of COMPLEMENTOFDEFAULT

Reviewed-by: Rich Salz <rsalz@openssl.org>
MR: #2202

Curve25519: fix const-initialization

Clang is permissive of this, but gcc fails.

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

Add ASN.1 ADB callback.

Add support for application supplied any defined by callback. An
application can change the selector value if it wishes. This is
mainly intended for values which are only known at runtime, for
example dynamically created OIDs.

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

GH784: Better variable name

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>

Make formatting consistent in apps/Makefile.in

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

Remove redundant semi-colons from apps/Makefile.in

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

Documentation for new CT s_client flags

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

Documentation for new SSL functions

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

Handle empty log name in "enable_logs" line of the CT log file

e.g. "enabled_logs = foo,,bar"

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

Handle missing "enabled_logs" line in CT log file

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

If a CT log entry in CTLOG_FILE is invalid, skip it and continue loading

Previously, the remaining CT log entries would not be loaded.
Also, CTLOG_STORE_load_file would return 1 even if a log entry was
invalid, resulting in no errors being shown.

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

Extends s_client to allow a basic CT policy to be enabled

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

Change default CT log list filename to "ct_log_list.cnf"

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

Adds CT validation to SSL connections

Disabled by default, but can be enabled by setting the
ct_validation_callback on a SSL or SSL_CTX.

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

Changes to DEFAULT curves

Change the ECC default curve list to be this, in order: x25519, secp256r1,
secp521r1, secp384r1, brainpoolP256r1, brainpoolP384r1, and brainpool512r1.

Reviewed-by: Emilia Käsper <emilia@openssl.org>

Curve25519: avoid undefined behaviour

Appease the sanitizer: avoid left shifts of negative values.

This could've been done entirely with casts to uint and back,
but using masks seemed slightly more readable.

There are also implementation-defined signed right shifts in this
code. Those remain.

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

chacha/asm/chacha-ppc.pl: fix typo.

RT#4365

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

Don't build RC4 ciphersuites into libssl by default

RC4 based ciphersuites in libssl have been disabled by default. They can
be added back by building OpenSSL with the "enable-weak-ssl-ciphers"
Configure option at compile time.

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

Improved HKDF and TLS1-PRF documentation

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

Correction, $disabled{shared} rather than $config{no_shared}

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

Restore the zlib / zlib-dynamic logic

The proper logic is that both zlib and zlib-dynamic are disabled by
default and that enabling zlib-dynamic would enable zlib.  Somewhere
along the way, the logic got changed, zlib-dynamic was enabled by
default and zlib didn't get automatically enabled.

This change restores the original logic.

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

Sanity check PVK file fields.

PVK files with abnormally large length or salt fields can cause an
integer overflow which can result in an OOB read and heap corruption.
However this is an rarely used format and private key files do not
normally come from untrusted sources the security implications not
significant.

Fix by limiting PVK length field to 100K and salt to 10K: these should be
more than enough to cover any files encountered in practice.

Issue reported by Guido Vranken.

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

GH773: Possible leak on CRYPTO_THREAD_lock_new failure

Signed-off-by: Rich Salz <rsalz@akamai.com>
Reviewed-by: Matt Caswell <matt@openssl.org>

GH781: openssl.spec: use secure urls

Signed-off-by: Rich Salz <rsalz@akamai.com>
Reviewed-by: Matt Caswell <matt@openssl.org>

GH769: Reuse strndup()

Reviewed-by: Matt Caswell <matt@openssl.org>
Signed-off-by: Rich Salz <rsalz@akamai.com>

GH355: Implement HKDF

This patch implements the HMAC-based Extract-and-Expand Key Derivation
Function (HKDF) as defined in RFC 5869.

It is required to implement the QUIC and TLS 1.3 protocols (among others).

Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Dr. Stephen Henson <steve@openssl.org>

Remove overzealous echoing

This construct in a Makefile is a bit overzealous:

@echo FOO
@FOO

Cleaned up.

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

Add KDF support to pkeyutl. Update documentation.

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

Clean up curve25519 build

- Remove OPENSSL_X25519_X86_64 which never worked, because we don't have
  the assembly.
- Also remove OPENSSL_SMALL (which should have been
  OPENSSL_SMALL_FOOTPRINT) which isn't a priority at the moment.

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

update NEWS

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

fix no-ec build

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

Refactor ClientHello extension parsing

1) Simplify code with better PACKET methods.

2) Make broken SNI parsing explicit. SNI was intended to be extensible
to new name types but RFC 4366 defined the syntax inextensibly, and
OpenSSL has never parsed SNI in a way that would allow adding a new name
type. RFC 6066 fixed the definition but due to broken implementations
being widespread, it appears impossible to ever extend SNI.

3) Annotate resumption behaviour. OpenSSL doesn't currently handle all
extensions correctly upon resumption. Annotate for further clean-up.

4) Send an alert on ALPN protocol mismatch.

Reviewed-by: Kurt Roeckx <kurt@openssl.org>

make update

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

add kdf.h to mkdef.pl

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

Add KDF error codes

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

make update

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

Generalise KDF test in evp_test.c

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

Convert PRF tests to use Ctrl

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

Add Ctrl keyword to KDF test in evp_test

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

Add string ctrl operations to TLS1 PRF, update documentation.

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

Use utility functions for HMAC and CMAC.

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

EVP_PKEY_CTX utility functions.

Utility functions to pass a string or hex string to EVP_PKEY_CTX_ctrl().

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

make errors

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

Handle KDF internally.

Handle KDF in ECDH_compute_key instead of requiring each implementation
support it. This modifies the compute_key method: now it allocates and
populates a buffer containing the shared secret.

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

Document the changes in config settings

Reviewed-by: Andy Polyakov <appro@openssl.org>

Don't copy from %target to %config so much, see %config as a complement

We copied $target{cflags}, $target{defines} and a few more to %config,
just to add to the entries.  Avoid doing so, and let the build templates
deal with combining the two.

There are a few cases where we still fiddle with %target, but that's
acceptable.

Reviewed-by: Andy Polyakov <appro@openssl.org>

Configure - Get rid of the special thread_cflag, replace with thread_scheme

The thread_cflag setting filled a double role, as kinda sorta an
indicator of thread scheme, and as cflags.  Some configs also added
lflags and ex_libs for multithreading regardless of if threading would
be enabled or not.

Instead of this, add threading cflags among in the cflag setting,
threading lflags in the lflag setting and so on if and only if threads
are enabled (which they are by default).

Also, for configs where there are no special cflags for threading (the
VMS configs are of that kind), this makes it possible to still clearly
mention what thread scheme is used.

The exact value of thread scheme is currently ignored except when it's
"(unknown)", and thereby only serves as a flag to tell if we know how
to build for multi-threading in a particular config.  Yet, the
currently used values are "(unknown)", "pthreads", "uithreads" (a.k.a
solaris threads) and "winthreads".

Reviewed-by: Andy Polyakov <appro@openssl.org>