Make RSA key exchange code actually constant-time.

Using RSA_PKCS1_PADDING with RSA_private_decrypt is inherently unsafe.
The API requires writing output on success and touching the error queue
on error. Thus, although the padding check itself is constant-time as of
294d1e36c2495ff00e697c9ff622856d3114f14f, and the logic after the
decryption in the SSL code is constant-time as of
adb46dbc6dd7347750df2468c93e8c34bcb93a4b, the API boundary in the middle
still leaks whether the padding check succeeded, giving us our
much-loved Bleichenbacher padding oracle.

Instead, PKCS#1 padding must be handled by the caller which uses
RSA_NO_PADDING, in timing-sensitive code integrated with the
Bleichenbacher mitigation. Removing PKCS#1 padding in constant time is
actually much simpler when the expected length is a constant (and if
it's not a constant, avoiding a padding oracle seems unlikely), so just
do it inline.

Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Rich Salz <rsalz@openssl.org>
GH: #1222

buf2hexstr: properly deal with empty string

It wrote before the start of the string

found by afl

Reviewed-by: Richard Levitte <levitte@openssl.org>
MR: #2994

Fix typo

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1241)

Appveyor: test install as well, via a fake deploy_script

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

Travis: When testing install, install docs as well

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

OpenSSL::Test: Fix directory calculations in __cwd()

We recalculate the location of the directories we keep track of.
However, we did so after having moved to the new directory already, so
the data we did the calculations from were possibly not quite correct.

This change moves the calculations to happen before moving to the new
directory.

This issue is sporadic, and possibly dependent on the platform.

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

More doc cleanup

Add missing entries to NAME section
Add SYNOPSIS lines, remove old NAME entries
Update find-doc-nits; better regexp's for parsing SYNOPSIS sections.
Rename a couple of files to have an API name.
Remove RSA_private_decrypt; it was duplicate content
Update for recent doc additions

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

Avoid using latest clang since repo not available

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

Make arm-xlate.pl set use strict.

It was already nearly clean. Just one undeclared variable.

Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1240)

Add verification of proxy certs to 25-test_verify.t

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

Create some proxy certificates

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

Make it possible to generate proxy certs with test/certs/mkcert.sh

This extends 'req' to take more than one DN component, and to take
them as full DN components and not just CN values.  All other commands
are changed to pass "CN = $cn" instead of just a CN value.

This adds 'genpc', which differs from the other 'gen*' commands by not
calling 'req', and expect the result from 'req' to come through stdin.

Finally, test/certs/setup.sh gets the commands needed to generate a
few proxy certificates.

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

make update

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

Allow proxy certs to be present when verifying a chain

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

Fix proxy certificate pathlength verification

While travelling up the certificate chain, the internal
proxy_path_length must be updated with the pCPathLengthConstraint
value, or verification will not work properly.  This corresponds to
RFC 3820, 4.1.4 (a).

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

Check that the subject name in a proxy cert complies to RFC 3820

The subject name MUST be the same as the issuer name, with a single CN
entry added.

RT#1852

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

RT3925: Remove trailing semi from macro

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

apps/req.c: Increment the right variable when parsing '+'

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

Remove pointless free loop in X509_TRUST_cleanup()

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1194)

Remove pointless free loop in X509_PURPOSE_cleanup()

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1194)

doc and comment fixes

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1233)

crypto/evp/e_aes_cbc_hmac_sha256.c: Remove spurious memset

Reviewed-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1231)

RT3136: Remove space after issuer/subject

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

Make DSA_SIG and ECDSA_SIG getters const.

Reorder arguments to follow convention.

Also allow r/s to be NULL in DSA_SIG_get0, similarly to ECDSA_SIG_get0.

This complements GH1193 which adds non-const setters.

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

rand/randfile.c: remove obsolete commentary.

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

Tests should check validation status directly

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

Test SCT lists

This encompasses what was previously tested.

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

aes/asm/bsaes-armv7.pl: omit redundant stores in XTS subroutines.

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

aes/asm/bsaes-armv7.pl: fix XTS decrypt test failure.

RT#4578

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

Change the RAND_file_name documentation accordingly

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

Change default directory for the .rnd file on Windows and VMS

The previous change for Windows wasn't quite right.  Corrected to use
%HOME%, %USERPROFILE% and %SYSTEMPROFILE%, in that order.

Also adding the default home for VMS, SYS$LOGIN:

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

Add a comment after some #endif at end of apps source code.

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1168)

Useless header include of openssl/rand.h

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1168)

Useless includes

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1168)

Missing NULL check on OBJ_dup result in x509_name_canon

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1168)

Fix an MSVC warning.

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1168)

Use a STACK_OF(OPENSSL_CSTRING) for const char * stacks

Better than losing the const qualifier.

RT4378

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

OpenBSD has intypes.h

Update e_os2.h so that inttypes.h is included.

RT4378

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

Replace 4 casts with 1

Changing the type of the |str| variable in asn1pars enables us to remove
4 casts with just 1. This silences an OpenBSD warning along the way.

RT4378

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

Improve const correctness for stacks of EVP_MD

EVP_MDs are always const, so stacks of them should be too. This silences
a warning about type punning on OpenBSD.

RT4378

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

constify SRP

Add const qualifiers to lots of SRP stuff. This started out as an effort
to silence some "type-punning" warnings on OpenBSD...but the fix was to
have proper const correctness in SRP.

RT4378

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

Avoid type punning warnings in b_addr.c

RT4378

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

Change default directory for storing the .rnd file on Windows

Previously we would try %RANDFILE%, then %HOME% and finally "C:".
Unfortunately this often ends up being "C:" which the user may not
have write permission for.

Now we try %RANDFILE% first, and then the same set of environment vars
as GetTempFile() uses, i.e. %TMP%, then %TEMP%, %USERPROFILE% and
%SYSTEMROOT%. If all else fails we fall back to %HOME% and only then "C:".

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

Harmonise the different build files

- User targets are now the same and generally do the same things
- configdata.pm depends on exactly the same files on all platforms
- VMS production of shared libraries is simplified
- VMS automatic dependency files get the extension .D rather than .MMS

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

Fix GCC build; make update; fix number re-use

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

Spelling fixes

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1219)

no-ripemd is an alias for no-rmd160

mkdef.pl was failing to understand no-ripemd. This is a deprecated option
which should act as an alias for no-rmd160.

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

Fix build break.

Aggregate local initializers are rarely portable (:

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

Add EVP_PKEY_get0_hmac() function

Before the addition of this function, it was impossible to read the
symmetric key from an EVP_PKEY_HMAC type EVP_PKEY.

Reviewed-by: Emilia Käsper <emilia@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1217)

Revert "RT4526: Call TerminateProcess, not ExitProcess"

This reverts commit 9c1a9ccf65d0ea1912675d3a622fa8e51b524b9e.

TerminateProcess is asynchronous, so the code as written in the above
commit is not correct. It is also probably not needed in the speed
case. Reverting in order to figure out the correct solution.

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

Skip the TLSProxy tests if environmental problems are an issue

On some platforms we can't startup the TLSProxy due to environmental
problems (e.g. network set up on the build machine). These aren't OpenSSL
problems so we shouldn't treat them as test failures. Just visibly
indicate that we are skipping the test.

We only skip the first time we attempt to start up the proxy. If that works
then everything else should do...if not we should probably investigate and
so report as a failure.

This also removes test_networking...there is a danger that this turns into
a test of user's environmental set up rather than OpenSSL.

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

Fix the build and tests following constification of DH, DSA, RSA

Misc fixes following the constification of the DH, DSA and RSA
getters.

Reviewed-by: Stephen Henson <steve@openssl.org>

Change the return type of EVP_EncodeUpdate

Previously EVP_EncodeUpdate returned a void. However there are a couple
of error conditions that can occur. Therefore the return type has been
changed to an int, with 0 indicating error and 1 indicating success.

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

Deal with the consequences of constifying getters

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

Constify the parameter getters for RSA, DSA and DH

Including documentation changes

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

RT4526: Call TerminateProcess, not ExitProcess

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

Constify input buffers of some X509V3 and X509_PURPOSE -related methods

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1215)

Constify X509_TRUST_add method.

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1215)

Constify asn1/asn_mime.c

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1215)

Constify some input buffers in asn1

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1215)

Constify CMS_get0_type input

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1215)

Constify UI

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1215)

Constify PKCS12_create, PKCS12_add_key, PKCS12_add_safe.

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1215)

Constify some conf_mod.c internal methods

Reviewed-by: Kurt Roeckx <kurt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1215)

Change (!seqtt) to (seqtt == NULL)

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

Always check that the value returned by asn1_do_adb() is non-NULL

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

perlasm/x86*.pl: add endbranch instruction.

For further information see "Control-flow Enforcement Technology
Preview" by Intel.

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

modes/asm/ghashp8-ppc.pl: improve performance by 2.7x.

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

aes/asm/aesp8-ppc.pl: implement "tweak chaining".

This is useful in Linux kernel context, in cases data happens
to be fragmented and processing can take multiple calls.

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

test/evptests.txt: add more XTS tests.

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

evp/e_aes.c: wire hardware-assisted XTS subroutines.

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

aes/asm/aesp8-ppc.pl: add XTS subroutines.

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

(Temporarily) don't download clang package

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

perlasm/x86_64-xlate.pl: add commentary.

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

perlasm/x86_64-xlate.pl: refactor argument parsing loop.

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

Make x86_64-xlate.pl 'use strict' clean.

use strict would have caught a number of historical bugs in the perlasm
code, some in the repository and some found during review. It even found
a fresh masm-only bug (see below).

This required some tweaks. The "single instance is enough" globals got
switched to proper blessed objects rather than relying on symbolic refs.
A few types need $opcode passed in as a result.

The $$line thing is a little bit of a nuisance. There may be a clearer
pattern to use instead.

This even a bug in the masm code.
9b634c9b37afc482a8dc8868e367bdd1b650e507 added logic to make labels
global or function-global based on whether something starts with a $,
seemingly intended to capture the $decor setting of '$L$'. However, it
references $ret which is not defined in label::out. label::out is always
called after label::re, so $ret was always the label itself, so the line
always ran.

I've removed the regular expression so as not to change the behavior of
the script. A number of the assembly files now routinely jump across
functions, so this seems to be the desired behavior now.

GH#1165

Signed-off-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

VMS: use BLDDIR rather than BUILDDIR in descrip.mms.tmpl

Just like in the other build file templates

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

Add a developer target 'build_generated' to rebuild mandatory headers

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

Build file templates: make sure to depend on generated header files

As well as properly generating those that are made from .in files.

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

Move the Configure generated header files to the top build.info

This turns these headers into build file generated ones.

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

Configure: Make it possible to generate mandatory header files

'DEPEND[]=file.h' becomes a special way to say that 'file.h' must be
generated before anything else is built.  It's likely that a number
of source files depend on these header files, this provides a simple
way to make sure they are always generated even it the dependency data
hasn't been added to the build file yet.

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

Fix omitted selector handling.

The selector field could be omitted because it has a DEFAULT value.
In this case *sfld == NULL (sfld can never be NULL). This was not
noticed because this was never used in existing ASN.1 modules.

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

Update fuzz corpora

Reviewed-by: Rich Salz <rsalz@openssl.org>
MR: #2949

Initialize the session_id

ssl_session_hash() always looks at the first 4 bytes, regardless of the length.
A client can send a session id that's shorter, and the callback could also
generate one that's shorter.  So we make sure that the rest of the buffer is
initliazed to 0 so that we always calculate the same hash.

Found by tis-interpreter, also previously reported as RT #2871

Reviewed-by: Rich Salz <rsalz@openssl.org>
MR: #2911

Remove an unused variable assignment

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

RT4562: Fix misleading doc on OPENSSL_config

Also changed the code to use "appname" not "filename"

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

Free memory on an error path

The function a2i_ASN1_STRING can encounter an error after already
allocating a buffer. It wasn't always freeing that buffer on error.

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

Avoid a double-free in crl2pl7

The variable |crtflst| could get double freed in an error path.

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

Add missing break statement

The -psk option processing was falling through to the -srp option
processing in the ciphers app.

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

Fix comment

Fix a comment following commit c2c49969e23605.

RT2388

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

make update

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

Shorten some symbol names

VMS C doesn't allow symbols longer than 31 characters.  We do the
automatic shortening with the library files, but not otherwise (to
make sure to work the VMS C magic).

For consistency, I shortened other similar symbols in the same manner.

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

GH1141: Different fix, preferred by Richard.

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

RT3454: Re-enable nistp-64_gcc_128 on windows

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

Missed some copyright merge

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

Ensure that NULL r and s parameters cannot be set on DSA_SIG/ECDSA_SIGs.

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1193)

Forgot to update the setter argument ordering in the public headers.

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1193)

Reorder the setter arguments to more consistently match that of other APIs,
per review comments.

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1193)

Implement DSA_SIG_set0() and ECDSA_SIG_set0(), for setting signature values.

SSH2 implementations which use DSA_do_verify() and ECDSA_do_verify() are given
the R and S values, and the data to be signed, by the client.  Thus in order
to validate these signatures, SSH2 implementations will digest and sign
the data -- and then pass in properly provisioned DSA_SIG and ECDSA_SIG objects.

Unfortunately, the existing OpenSSL-1.1.0 APIs do not allow for directly setting
those R and S values in these objects, which makes using OpenSSL for such
SSH2 implementations much more difficult.

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Emilia Käsper <emilia@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1193)

Add some missing return value checks

Some misc return value checks

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