Testing symbol presence: also take note of small objects

The S symbol class wasn't checked.

Notified by Sebastian Andrzej Siewior

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

Remove NOEXIST entries

checkpoint before release.

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

Don't leak memory on set_reasons() error path

The set_reasons() function in v3_crld.c leaks a STACK_OF(CONF_VALUE)
object on an error path.

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

Don't leak memory on int X509_PURPOSE_add() error path

The int X509_PURPOSE_add() function was leaking an X509_PURPOSE object
on error.

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

Don't leak memory on X509_TRUST_add() error path

The X509_TRUST_add() function was leaking an X509_TRUST object on error.

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

Don't leak memory on ASN1_item_pack() error path

The ASN1_item_pack() function was leaking an ASN1_STRING object on error
paths.

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

Don't leak memory on ASN1_GENERALIZEDTIME_adj() error path

The ASN1_GENERALIZEDTIME_adj() function leaks an ASN1_GENERALIZEDTIME
object on an error path.

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

Don't leak memory in v2i_POLICY_MAPPINGS() on error path

The v2i_POLICY_MAPPINGS() function leaked ASN1_OBJECT pointers on error
paths.

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

Don't leak memory from notice_section function on error path

The notice_section() function allocates a STACK_OF(CONF_VALUE) but
then fails to free it on an error path.

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

Don't leak memory in v2i_AUTHORITY_KEYID

The v2i_AUTHORITY_KEYID() function can leak memory under an error
condition.

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

Fix memory leak in crl2pkcs7 app

The crl2pkcs7 app leaks a stack of OPENSSL_STRINGs in error paths.

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

Free a temporary buffer used by dsaparam application

The dsaparam application allocates a temporary buffer but then doesn't
free it.

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

Free buffer on error in a2i_ASN1_INTEGER()

The function a2i_ASN1_INTEGER() allocates a buffer |s| but then fails
to free it on error paths.

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

Free memory on error in cms app

The make_receipt_request() function in the cms app can leak memory on
an error condition.

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

Free tempory data on error in ec_wNAF_mul()

The ec_wNAF_mul() function allocates some temporary storage that it
doesn't always free on an error condition.

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

Fix nits in crypto.pod,ssl.pod

After this merge, the only things left (from doc-nit-check) is
74 pages without a "RETURN VALUES" section.

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

Fix various doc nits.

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

Remove/rename some old files.

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

Add final(?) set of copyrights.

Add copyright to missing assembler files.
Add copyrights to missing test/* files.
Add copyrights
Various source and misc files.

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

Fix some RAND bugs

RT2630 -- segfault for int overlow
RT2877 -- check return values in apps/rand
Update CHANGES file for previous "windows rand" changes.

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

Ensure an ASN1_OBJECT is freed in error paths

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

Free allocated password strings on exit

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

Free a BIO_ADDR if DTLSv1_listen return <=0

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

Ensure BIGNUM is freed in an error path

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

Free an X509_CRL in an error path

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

Don't leak X509_OBJECT in an error path

Swap the ordering of some code to avoid a leak in an error path.

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

RT4337: Crash in DES

Salt must be two ASCII characters.  Add tests to check for that,
and a test to test the checks.

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

Fix printing of DH Parameters

The -text argument to dhparam is broken, because the DHparams_print()
function always returns an error. The problem is that always expects a
public or private key to be present, even though that is never the case
with parameters.

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

Add dhparam sanity check and update DH_check documentation

The -check argument to dhparam should never identify any problems if we
have just generated the parameters. Add a sanity check for this and print
an error and fail if necessary.

Also updates the documentation for the -check argument, and the DH_check()
function.

RT#4244

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

Raise an Err when CRYPTO_THREAD_lock_new fails

Add missing error raise call, as it is done everywhere else.
and as CRYPTO_THREAD_lock_new don't do it internally.

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

req command incorrectly displays the bits for an EC key

When the "req" command is used to generate a new EC key using the -newkey
option it will incorrectly display:

 Generating a 2048 bit EC private key

This commit fixes the message to not display the bit length for EC keys
because we don't currently support getting that during generation.

GitHub Issue #1068

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

OpenBSD doesn't have ucontext.h so don't try and include it

On OpenBSD we turn off async capabilities due to no ucontext.h.

RT#4379

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

make update

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

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

Fix: PEM_read_bio_PrivateKey with no-ui / no-stdio

If openssl is compiled with no-ui or no-stdio, then PEM_read_bio_PrivateKey fails if a password but no callback is provided.

The reason is that the premature return in the PEM_def_callback implementation when OPENSSL_NO_STDIO or OPENSSL_NO_UI is defined, comes too early.

This patch moves the ifdef block to the correct place.
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>

Remove null check, per review feedback.  Note this in the docs.

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1135)

Add requested HISTORY section, remove copy/pastos, per review feedback.

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1135)

Add an SSL_SESSION accessor for obtaining the protocol version number, with
accompanying documentation.

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1135)

RT4539: Add section for renamed ciphers.

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

speed.c: fix segfault with unrecognized algorithms

When an unrecognized algorithm is given on command line together with
-async_jobs, speed_main will jump to clean-up and run
ASYNC_cleanup_thread without calling ASYNC_init_thread first.

Example:
openssl speed -async_jobs 4 ras2048

Signed-off-by: Cristian Stoica <cristian.stoica@nxp.com>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1084)

Parameter copy sanity checks.

Don't copy parameters is they're already present in the destination.
Return error if an attempt is made to copy different parameters to
destination. Update documentation.

If key type is not initialised return missing parameters

RT#4149

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

return error in ct_move_scts()

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

Travis: update sanitizer configs

- Use the new enable-ubsan and enable-asan configuration options.
- Separate ubsan and asan runs.
- In addition, run shared ubsan tests to get more coverage.
  (Shared asan tests need a bit more thought to get them working.)

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

Fix strdup macro redefinition

This fixes the following error when the CRT debug heap (crtdbg.h) is used:
e_os.h(476): warning C4005: 'strdup': macro redefinition
C:\Program Files (x86)\Windows Kits\10\Include\10.0.10586.0\ucrt\crtdbg.h(319): note: see previous definition of 'strdup'
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1137)

perl: use the 'if' module to conditionally load File::Glob

Trying to use normal perl conditions to conditionally 'use' a perl
module didn't quite work.  Using the 'if' module to do so does work.

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

Make sure max in fmtstr() doesn't overflow into negativity

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

set RAND_event and RAND_screen to deprecated in 1.1.0 in librypto.num

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

fix deprecation version number in docs

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

update docs with descriptions and deprecation

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

fix return value in docs

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

add removed functions back as deprecated

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

get rid of now empty #if

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

remove winrand.c entirely, nothing seems to reference it

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

cherry pick pr-512 changes

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

get rid of unnecessary include

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

remove RAND_screen and friends

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

cherry pick pr-512 changes

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

OR flags with CRYPT_SILENT to really make sure no UI pops up

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

fix endif comment

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

remove all WINCE ifdefs

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

- remove insane heap walk and kernel loading code; clean up style and calling conventions

Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/1079)

x86_64 assembly pack: tolerate spaces in source directory name.

[as it is now quoting $output is not required, but done just in case]

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

Configure,test/recipes: "pin" glob to File::Glob::glob.

As it turns out default glob's behaviour for quoted argument varies
from version to version, making it impossible to Configure or run
tests in some cases. The reason for quoting globs was to accommodate
source path with spaces in its name, which was treated by default glob
as multiple paths. File::Glob::glob on the other hand doesn't consider
spaces as delimiters and therefore works with unquoted patterns.

[Unfortunaltely File::Glob::glob, being too csh-ly, doesn't work
on VMS, hence the "pinning" is conditional.]

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

Add more zalloc

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/997)

Fix some missing inits

Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
(Merged from https://github.com/openssl/openssl/pull/997)

poly1305/asm/poly1305-mips.pl: adhere to standard frame layout.

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

SPARC assembly pack: add missing .type directives.

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

Change a call of OPENSSL_strcasecmp to strcasecmp

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

make update

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

Remove internal functions OPENSSL_strcasecmp and OPENSSL_strncasecmp

Their only reason to exist was that they didn't exist in VMS before
version 7.0.  We do not support such old versions any more.

However, for the benefit of systems that don't get strings.h included
by string.h, we include the former in e_os.h.

RT#4458

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

include/openssl/e_os2.h: fix 'noreturn' pre-processor logic.

Newer gcc still recognizes e.g. -std=c9x in which case it wouldn't
have used 'noreturn' at all with original logic.

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

crypto/o_str.c: add _GNU_SOURCE strerror_r case.

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

crypto/o_str.c: strerror_s is provided by specific compiler run-time,
not by OS [as was implied by guarding #if condition].

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

bn/bn_exp.c: explain 'volatile' in MOD_EXP_CTIME_COPY_FROM_PREBUF.

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

bf/build.info: engage assembly module.

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

Avoid calling memcpy with lenght of 0

We can call memcpy() with a pointer 1 past the last allocated byte and length
of 0 and you can argue that that's undefined behaviour.

Reported by tis-interpreter

Reviewed-by: Rich Salz <rsalz@openssl.org>
GH: #1132

Fix ssl_cert_set0_chain invalid pointer

When setting the certificate chain, if a certificate doesn't pass
security checks, then chain may point to a freed STACK_OF(X509)

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

Fix fmtstr for BIO_printf() et al

-   If we have a maximum amount of characters permitted to be printed
    (for example "%.2s", which allows for a maximum of 2 chars), we
    minimize the number of characters from the string to printed to
    that size.
-   If there is space for padding and there is a maximum amount of
    characters to print (for example "%3.2s", which shall give at
    least a 1 space padding), the amount of characters to pad with
    gets added to the maximum so the minimum field size (3 in this
    example) gets filled out.

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

make sure to put quotes around -config argument, in case of spaces

RT#4486

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

perl glob: make sure to put quotes around the pattern, in case of spaces

RT#4486

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

Windows makefile: handle the case with space in source directory

This applies when building out-of-source.

RT#4486

NOTE: we can't do the same for Unix, as Unix make doesn't handle this
type of issue.  Also, directory specs are much less likely to have
spaces on Unix...

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

Configure: handle spaces in source directory spec

RT#4486

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

Allow space in PERL spec (unix only)

Someone wants to configure like this:

    PERL="/usr/bin/env perl" ./config

The end goal is to get that in the #! line of CA.pl and a few other
scripts.  That works well already, but in the Makefile, there were a
few lines looking like this:

    PERL=$(PERL) $(PERL) whatever.pl ...

Those need some quoting.

RT#4311

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

Avoid msys name mangling

If using the msys console then msys attempts to "fix" command line
arguments to convert them from Unix style to Windows style. One of the
things it does is to look for arguments seperated by colons. This it
assumes is a list of file paths, so it replaces the colon with a semi-colon.
This was causing one of our tests to fail when calling the "req" command
line app. We were attempting to create a new DSA key and passing the
argument "dsa:../apps/dsa1024.pem". This is exactly what we intended but
Msys mangles it to "dsa;../apps/dsa1024.pem" and the command fails.
There doesn't seem to be a way to suppress Msys name mangling. Fortunately
we can work around this issue by generating the DSA key in a separate step
by calling "gendsa".

RT#4255

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

Fix intermittent windows failures in TLSProxy tests

When closing down the socket in s_client Windows will close it immediately
even if there is data in the write buffer still waiting to be sent. This
was causing tests to fail in Msys/Mingw builds because TLSProxy doesn't see
the final CloseNotify.

I have experimented with various ways of doing this "properly" (e.g.
shutting down the socket before closing, setting SO_LINGER etc). I can't
seem to find the "magic" formula that will make Windows do this. Inserting
a short 50ms sleep seems to do the trick...but its not very "nice" so I've
inserted a TODO on this item. Perhaps someone else will have better luck
in figuring this out.

RT#4255

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

Fix s_client/s_server waiting for stdin on Windows

On Windows we were using the function _kbhit() to determine whether there
was input waiting in stdin for us to read. Actually all this does is work
out whether there is a keyboard press event waiting to be processed in the
input buffer. This only seems to work in a standard Windows console (not
Msys console) and also doesn't work if you redirect the input from some
other source (as we do in TLSProxy tests). This commit changes things to
work differently depending on whether we are on the Windows console or not.

RT#4255

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

Fix some s_server issues on Windows

In s_server we call BIO_sock_should_retry() to determine the state of the
socket and work out whether we should retry an operation on it or not.
However if you leave it too long to call this then other operations may
have occurred in the meantime which affect the result. Therefore we should
call it early and remember the result for when we need to use it. This fixes
a test problem on Windows.

Another issue with s_server on Windows is that some of output to stdout does
not get displayed immediately. Apparently more liberal use of BIO_flush is
required.

RT#4255

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

Silence some "may be uninitialized when used" warning

Clang was complaining about some variables possibly being uninitialized
when used. The warnings are bogus, but clang can't figure that out. This
silences the warnings.

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

Add a test for printing floating point format specifiers

Previous commits fixed the implementation of the %e and %g format
specifiers as well as other issues. This commit adds a test.

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

Prevent an overflow when trying to print excessively big floats

We convert the integer part of the float to a long. We should check it
fits first.

Issue reported by Guido Vranken.

GitHub Issue #1102

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

Fix implementation of "e" and "g" formats for printing floating points

The previous commit which "fixed" the "e" and "g" floating point formats
just printed them in the same way as "f". This is wrong. This commit
provides the correct formatting.

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

Fix some issues in b_print.c code

Convert assert to OPENSSL_assert(), add some documentation, add the calls
to fmtfp() for the "e" and "g" floating point formats which were missing.

Based on a patch provided by Ger Hobbelt <ger@hobbelt.com>.

RT#2270

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

Update the documentation of BN_hex2bn()

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

fix BN_hex2bn()/BN_dec2bn() memory leak

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

Fix some suspect warnings on Windows

Windows was complaining about a unary minus operator being applied to an
unsigned type. It did seem to go on and do the right thing anyway, but the
code does look a little suspect. This fixes it.

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

The ssl3_digest_cached_records() function does not handle errors properly

The ssl3_digest_cached_records() function was failing to handle errors
that might be returned from EVP_DigestSignInit() and
EVP_DigestSignUpdate().

RT#4180

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

Add rfc822Mailbox to string table.

RT#2369

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

Handle Visual C warning about non-standard function names.

Visual C version from version 2003 complain about certain function
names, for example:

    apps\apps.c(2572) : warning C4996: 'open': The POSIX name for this item is deprecated. Instead, use the ISO C++ conformant name: _open. See online help for details.

This adds preprocessor aliases for them in e_os.h.

Additionally, crypto/conf/conf_lib.c needs to include e_os.h to catch
those aliases.

RT#4488
RT#4489

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

Configure: pull 'which' back.

At earlier point 'which' was replaced with IPC::Cmd::can_run call.
Unfortunately on RPM-based systems it is a separate package and it's
not given that it's installed. Resurrected 'which' provides
poor-man fallback for IPC::Cmd::can_run.

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

Makefile.shared: revert Haiku support commit.

Configurations/50-haiku.conf reuses gnu-shared rules and doesn't
require dedicated targets.

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