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>
Most of the assembly uses constants from arm_arch.h, but a few references to
ARMV7_NEON don't. Consistently use the macros everywhere.
Signed-off-by: Andy Polyakov <appro@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
and reorganize/harmonize post-conditions.
Additional hardening following on from CVE-2016-0702
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
At the same time remove miniscule bias in final subtraction.
Performance penalty varies from platform to platform, and even with
key length. For rsa2048 sign it was observed to be 4% for Sandy
Bridge and 7% on Broadwell.
CVE-2016-0702
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
Performance penalty is 2% on Linux and 5% on Windows.
CVE-2016-0702
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
Performance penalty varies from platform to platform, and even
key length. For rsa2048 sign it was observed to reach almost 10%.
CVE-2016-0702
Reviewed-by: Richard Levitte <levitte@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
In the BN_hex2bn function the number of hex digits is calculated using
an int value |i|. Later |bn_expand| is called with a value of |i * 4|.
For large values of |i| this can result in |bn_expand| not allocating any
memory because |i * 4| is negative. This leaves ret->d as NULL leading
to a subsequent NULL ptr deref. For very large values of |i|, the
calculation |i * 4| could be a positive value smaller than |i|. In this
case memory is allocated to ret->d, but it is insufficiently sized
leading to heap corruption. A similar issue exists in BN_dec2bn.
This could have security consequences if BN_hex2bn/BN_dec2bn is ever
called by user applications with very large untrusted hex/dec data. This is
anticipated to be a rare occurrence.
All OpenSSL internal usage of this function uses data that is not expected
to be untrusted, e.g. config file data or application command line
arguments. If user developed applications generate config file data based
on untrusted data then it is possible that this could also lead to security
consequences. This is also anticipated to be a rare.
Issue reported by Guido Vranken.
CVE-2016-0797
Reviewed-by: Andy Polyakov <appro@openssl.org>
The entire contents of <internal/bn_conf.h> are unwanted in the UEFI
build because we have to do it differently there. To support building
for both 32-bit and 64-bit platforms without re-running the OpenSSL
Configure script, the EDK2 environment defines THIRTY_TWO_BIT or
SIXTY_FOUR_BIT for itself according to the target platform.
The current setup is broken, though. It checks for OPENSSL_SYS_UEFI but
before it's actually defined, since opensslconf.h hasn't yet been
included.
Let's fix that by including opensslconf.h. And also let's move the
bn_conf.h doesn't even need to *exist* in the UEFI build environment.
This is also GH PR736.
Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Tim Hudson <tjh@openssl.org>
This takes us away from the idea that we know exactly how our static
libraries are going to get used. Instead, we make them available to
build shareable things with, be it other shared libraries or DSOs.
On the other hand, we also have greater control of when the shared
library cflags. They will never be used with object files meant got
binaries, such as apps/openssl or test/test*.
With unified, we take this a bit further and prepare for having to
deal with extra cflags specifically to be used with DSOs (dynamic
engines), libraries and binaries (applications).
Reviewed-by: Rich Salz <rsalz@openssl.org>
All those flags existed because we had all the dependencies versioned
in the repository, and wanted to have it be consistent, no matter what
the local configuration was. Now that the dependencies are gone from
the versioned Makefile.ins, it makes much more sense to use the exact
same flags as when compiling the object files.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Add -DBIO_DEBUG to --strict-warnings.
Remove comments about outdated debugging ifdef guards.
Remove md_rand ifdef guarding an assert; it doesn't seem used.
Remove the conf guards in conf_api since we use OPENSSL_assert, not assert.
For pkcs12 stuff put OPENSSL_ in front of the macro name.
Merge TLS_DEBUG into SSL_DEBUG.
Various things just turned on/off asserts, mainly for checking non-NULL
arguments, which is now removed: camellia, bn_ctx, crypto/modes.
Remove some old debug code, that basically just printed things to stderr:
DEBUG_PRINT_UNKNOWN_CIPHERSUITES, DEBUG_ZLIB, OPENSSL_RI_DEBUG,
RL_DEBUG, RSA_DEBUG, SCRYPT_DEBUG.
Remove OPENSSL_SSL_DEBUG_BROKEN_PROTOCOL.
Reviewed-by: Richard Levitte <levitte@openssl.org>
It seems that on some platforms, the perlasm scripts call the C
compiler for certain checks. These scripts need the environment
variable CC to have the C compiler command.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Some files in crypto/bn depend on internal/bn_conf.h, and so does
test/bntest. Therefore, we add another inclusion directory.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Add new function BN_bn2binpad() which checks the length of the output
buffer and pads the result with zeroes if necessary.
New functions BN_bn2lebinpad() and BN_lebin2bn() which use little endian
format.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Now that we have the foundation for the "unified" build scheme in
place, we add build.info files. They have been generated from the
Makefiles in the same directories. Things that are platform specific
will appear in later commits.
Reviewed-by: Andy Polyakov <appro@openssl.org>
The "unified" build scheme revolves around small information files,
build.info, which each describe their own bit of everything that needs
to be built, using a mini-language described in Configurations/README.
The information in build.info file contain references to source files
and final result. Object files are not mentioned at all, they are
simply from source files. Because of this, all the *_obj items in
Configurations/*.conf are renamed to *_asm_src and the files listed
in the values are change from object files to their corresponding
source files. For the sake of the other build schemes, Configure
generates corresponding *_obj entries in %target.
Furthermore, the "unified" build scheme supports having a build
directory tree separate from the source directry tree.
All paths in a build.info file is assumed to be relative to its
location, either within the source tree or within the build tree.
Reviewed-by: Andy Polyakov <appro@openssl.org>
There was an unused macro in ssl_locl.h that used an internal
type, so I removed it.
Move bio_st from bio.h to ossl_type.h
Reviewed-by: Andy Polyakov <appro@openssl.org>
The use of the uninitialized buffer in the RNG has no real security
benefits and is only a nuisance when using memory sanitizers.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
This was done by the following
find . -name '*.[ch]' | /tmp/pl
where /tmp/pl is the following three-line script:
print unless $. == 1 && m@/\* .*\.[ch] \*/@;
close ARGV if eof; # Close file to reset $.
And then some hand-editing of other files.
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
Move opensslconf.h.in to include/openssl.
Split off DES,BN,RC4 stuff into separate header file
templates in crypto/include/internal/*_conf.h.in
Reviewed-by: Richard Levitte <levitte@openssl.org>
Rename 'update' to 'generate'. Rather than recurse, just explicitly
call the three generate targets directly.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Remove lint, tags, dclean, tests.
This is prep for a new makedepend scheme.
This is temporary pending unified makefile, and might help it.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Create Makefile's from Makefile.in
Rename Makefile.org to Makefile.in
Rename Makefiles to Makefile.in
Address review feedback from Viktor and Richard
Reviewed-by: Viktor Dukhovni <viktor@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Provide backwards-compatiblity for functions, macros and include
files if OPENSSL_API_COMPAT is either not defined or defined less
than the version number of the release in which the feature was
deprecated.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Some URLs in the source code ended up getting mangled by indent. This fixes
it. Based on a patch supplied by Arnaud Lacombe <al@aerilon.ca>
Reviewed-by: Richard Levitte <levitte@openssl.org>
Commit 2b0180c37f attempted to do this but
only hit one of many BN_mod_exp codepaths. Fix remaining variants and add
a test for each method.
Thanks to Hanno Boeck for reporting this issue.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Dr. Stephen Henson <steve@openssl.org>
The function rsa_ossl_mod_exp uses the function BN_with_flags to create a
temporary copy (local_r1) of a BIGNUM (r1) with modified flags. This
temporary copy shares some state with the original r1. If the state of r1
gets updated then local_r1's state will be stale. This was occurring in the
function so that when local_r1 was freed a call to bn_check_top was made
which failed an assert due to the stale state. To resolve this we must free
local_r1 immediately after we have finished using it and not wait until the
end of the function.
This problem prompted a review of all BN_with_flag usage within the
codebase. All other usage appears to be correct, although often not
obviously so. This commit refactors things to make it much clearer for
these other uses.
Reviewed-by: Emilia Käsper <emilia@openssl.org>
The problem remained unnoticed so far, because it's never called by default.
You have to craft OPENSSL_ppccap environment variable to trigger the problem.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Rebuild error source files: the new mkerr.pl functionality will now
pick up and translate static function names properly.
Reviewed-by: Richard Levitte <levitte@openssl.org>
BN_with_flags() will read the dest->flags to keep the BN_FLG_MALLOCED but
overwrites everything else.
Signed-off-by: Kurt Roeckx <kurt@roeckx.be>
Reviewed-by: Rich Salz <rsalz@openssl.org>
MR #1231
-Allow mingw debug builds to fail on Travis CI
-Fix Travis email notifications config
-Rename a variable to avoid a bogus warning with old GCC
error: declaration of ``dup'' shadows a global declaration [-Werror=shadow]
-Disable pedantic ms-format warnings with mingw
-Properly define const DH parameters
-Restore --debug flag in Travis CI builds; -d would get incorrectly passed
to ./Configure in mingw debug builds.
Signed-off-by: Rich Salz <rsalz@openssl.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
As some of ARM processors, more specifically Cortex-Mx series, are
Thumb2-only, we need to support Thumb2-only builds even in assembly.
Reviewed-by: Tim Hudson <tjh@openssl.org>
There are many places (nearly 50) where we malloc and then memset.
Add an OPENSSL_zalloc routine to encapsulate that.
(Missed one conversion; thanks Richard)
Also fixes GH328
Reviewed-by: Richard Levitte <levitte@openssl.org>
Don't dereference |d| when |top| is zero. Also test that various BIGNUM methods behave correctly on zero/even inputs.
Follow-up to b11980d79a
Reviewed-by: Rich Salz <rsalz@openssl.org>
The function BN_MONT_CTX_set was assuming that the modulus was non-zero
and therefore that |mod->top| > 0. In an error situation that may not be
the case and could cause a seg fault.
This is a follow on from CVE-2015-1794.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Create bn_free_d utility routine and use it.
Fix RT3950
Also a missing cleanse, from Loganaden Velvindron (loganaden@gmail.com),
who noticed it in a Cloudflare patch.
Reviewed-by: Richard Levitte <levitte@openssl.org>
This leaves behind files with names ending with '.iso-8859-1'. These
should be safe to remove. If something went wrong when re-encoding,
there will be some files with names ending with '.utf8' left behind.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Add secure heap for storage of private keys (when possible).
Add BIO_s_secmem(), CBIGNUM, etc.
Add BIO_CTX_secure_new so all BIGNUM's in the context are secure.
Contributed by Akamai Technologies under the Corporate CLA.
Reviewed-by: Richard Levitte <levitte@openssl.org>
A small rearrangement so the inclusion of rsaz_exp.h would be
unconditional, but what that header defines becomes conditional.
This solves the weirdness where rsaz_exp.h gets in and out of the
dependency list for bn_exp.c, depending on the present architecture.
Reviewed-by: Rich Salz <rsalz@openssl.org>
A BIGNUM can have the value of -0. The function BN_bn2hex fails to account
for this and can allocate a buffer one byte too short in the event of -0
being used, leading to a one byte buffer overrun. All usage within the
OpenSSL library is considered safe. Any security risk is considered
negligible.
With thanks to Mateusz Kocielski (LogicalTrust), Marek Kroemeke and
Filip Palian for discovering and reporting this issue.
Reviewed-by: Tim Hudson <tjh@openssl.org>
bn_get_bits5 was overstepping array boundary by 1 byte. It was exclusively
read overstep and data could not have been used. The only potential problem
would be if array happens to end on the very edge of last accesible page.
Reviewed-by: Richard Levitte <levitte@openssl.org>
If BN_rand is called with |bits| set to 1 and |top| set to 1 then a 1 byte
buffer overflow can occur. There are no such instances within the OpenSSL at
the moment.
Thanks to Mateusz Kocielski (LogicalTrust), Marek Kroemeke, Filip Palian for
discovering and reporting this issue.
Reviewed-by: Kurt Roeckx <kurt@openssl.org>
The functions BN_rshift and BN_lshift shift their arguments to the right or
left by a specified number of bits. Unpredicatable results (including
crashes) can occur if a negative number is supplied for the shift value.
Thanks to Mateusz Kocielski (LogicalTrust), Marek Kroemeke and Filip Palian
for discovering and reporting this issue.
Reviewed-by: Kurt Roeckx <kurt@openssl.org>
We had updates of certain header files in both Makefile.org and the
Makefile in the directory the header file lived in. This is error
prone and also sometimes generates slightly different results (usually
just a comment that differs) depending on which way the update was
done.
This removes the file update targets from the top level Makefile, adds
an update: target in all Makefiles and has it depend on the depend: or
local_depend: targets, whichever is appropriate, so we don't get a
double run through the whole file tree.
Reviewed-by: Rich Salz <rsalz@openssl.org>
BLKINIT optimization worked on T4, but for some reason appears "too
aggressive" for T3 triggering intermiitent EC failures. It's not clear
why only EC is affected...
Reviewed-by: Tim Hudson <tjh@openssl.org>
There are header files in crypto/ that are used by a number of crypto/
submodules. Move those to crypto/include/internal and adapt the
affected source code and Makefiles.
The header files that got moved are:
crypto/cryptolib.h
crypto/md32_common.h
Reviewed-by: Rich Salz <rsalz@openssl.org>
Just as with the OPENSSL_malloc calls, consistently use sizeof(*ptr)
for memset and memcpy. Remove needless casts for those functions.
For memset, replace alternative forms of zero with 0.
Reviewed-by: Richard Levitte <levitte@openssl.org>
For a local variable:
TYPE *p;
Allocations like this are "risky":
p = OPENSSL_malloc(sizeof(TYPE));
if the type of p changes, and the malloc call isn't updated, you
could get memory corruption. Instead do this:
p = OPENSSL_malloc(sizeof(*p));
Also fixed a few memset() calls that I noticed while doing this.
Reviewed-by: Richard Levitte <levitte@openssl.org>
This gets BN_.*free:
BN_BLINDING_free BN_CTX_free BN_FLG_FREE BN_GENCB_free
BN_MONT_CTX_free BN_RECP_CTX_free BN_clear_free BN_free BUF_MEM_free
Also fix a call to DSA_SIG_free to ccgost engine and remove some #ifdef'd
dead code in engines/e_ubsec.
Reviewed-by: Richard Levitte <levitte@openssl.org>
Add OPENSSL_clear_free which merges cleanse and free.
(Names was picked to be similar to BN_clear_free, etc.)
Removed OPENSSL_freeFunc macro.
Fixed the small simple ones that are left:
CRYPTO_free CRYPTO_free_locked OPENSSL_free_locked
Reviewed-by: Richard Levitte <levitte@openssl.org>
With no more symlinks, there's no need for those variables, or the links
target. This also goes for all install: and uninstall: targets that do
nothing but copy $(EXHEADER) files, since that's now taken care of by the
top Makefile.
Also, removed METHTEST from test/Makefile. It looks like an old test that's
forgotten...
Reviewed-by: Rich Salz <rsalz@openssl.org>
Rather than making include/openssl/foo.h a symlink to
crypto/foo/foo.h, this change moves the file to include/openssl/foo.h
once and for all.
Likewise, move crypto/foo/footest.c to test/footest.c, instead of
symlinking it there.
Originally-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
In the probable_prime() function we behave slightly different if the number
of bits we are interested in is <= BN_BITS2 (the num of bits in a BN_ULONG).
As part of the calculation we work out a size_limit as follows:
size_limit = (((BN_ULONG)1) << bits) - BN_get_word(rnd) - 1;
There is a problem though if bits == BN_BITS2. Shifting by that much causes
undefined behaviour. I did some tests. On my system BN_BITS2 == 64. So I
set bits to 64 and calculated the result of:
(((BN_ULONG)1) << bits)
I was expecting to get the result 0. I actually got 1! Strangely this...
(((BN_ULONG)0) << BN_BITS2)
...does equal 0! This means that, on my system at least, size_limit will be
off by 1 when bits == BN_BITS2.
This commit fixes the behaviour so that we always get consistent results.
Reviewed-by: Andy Polyakov <appro@openssl.org>
In the event of an error |rr| could be NULL. Therefore don't assume you can
use |rr| in the error handling code.
Reviewed-by: Andy Polyakov <appro@openssl.org>
Remove support for SHA0 and DSS0 (they were broken), and remove
the ability to attempt to build without SHA (it didn't work).
For simplicity, remove the option of not building various SHA algorithms;
you could argue that SHA_224/256/384/512 should be kept, since they're
like crypto algorithms, but I decided to go the other way.
So these options are gone:
GENUINE_DSA OPENSSL_NO_SHA0
OPENSSL_NO_SHA OPENSSL_NO_SHA1
OPENSSL_NO_SHA224 OPENSSL_NO_SHA256
OPENSSL_NO_SHA384 OPENSSL_NO_SHA512
Reviewed-by: Richard Levitte <levitte@openssl.org>
The following compile options (#ifdef's) are removed:
OPENSSL_NO_BIO OPENSSL_NO_BUFFER OPENSSL_NO_CHAIN_VERIFY
OPENSSL_NO_EVP OPENSSL_NO_FIPS_ERR OPENSSL_NO_HASH_COMP
OPENSSL_NO_LHASH OPENSSL_NO_OBJECT OPENSSL_NO_SPEED OPENSSL_NO_STACK
OPENSSL_NO_X509 OPENSSL_NO_X509_VERIFY
This diff is big because of updating the indents on preprocessor lines.
Reviewed-by: Richard Levitte <levitte@openssl.org>
This removes all code surrounded by '#ifdef undef'
One case is left: memmove() replaced by open-coded for loop,
in crypto/stack/stack.c That needs further review.
Also removed a couple of instances of /* dead code */ if I saw them
while doing the main removal.
Reviewed-by: Matt Caswell <matt@openssl.org>
Sometimes it fails to format them very well, and sometimes it corrupts them!
This commit moves some particularly problematic ones.
Reviewed-by: Tim Hudson <tjh@openssl.org>
OPENSSL_NO_RIPEMD160, OPENSSL_NO_RIPEMD merged into OPENSSL_NO_RMD160
OPENSSL_NO_FP_API merged into OPENSSL_NO_STDIO
Two typo's on #endif comments fixed:
OPENSSL_NO_ECB fixed to OPENSSL_NO_OCB
OPENSSL_NO_HW_SureWare fixed to OPENSSL_NO_HW_SUREWARE
Reviewed-by: Richard Levitte <levitte@openssl.org>
BN_init and BN_RECP_CTX_init are deprecated and are not exported
from shared libraries on some platforms (e.g. Windows) convert
bntest to use BN_new and BN_RECP_CTX_new instead.
Reviewed-by: Matt Caswell <matt@openssl.org>
This commit removes NCR, Tandem, Cray.
Regenerates TABLE.
Removes another missing BEOS fluff.
The last platform remaining on this ticket is WIN16.
Reviewed-by: Richard Levitte <levitte@openssl.org>
This facilitates "universal" builds, ones that target multiple
architectures, e.g. ARMv5 through ARMv7. See commentary in
Configure for details.
Reviewed-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Reviewed-by: Matt Caswell <matt@openssl.org>
Invalid zero-padding in the divisor could cause a division by 0.
Reviewed-by: Richard Levitte <levitte@openssl.org>
(cherry picked from commit a43bcd9e96)
The temporary variable causes unused variable warnings in opt mode with clang,
because the subsequent assert is compiled out.
Reviewed-by: Rich Salz <rsalz@openssl.org>
Exported headers shouldn't be included as "foo.h" by code from the same
module, it should only do so for module-internal headers. This is
because the symlinking of exported headers (from include/openssl/foo.h
to crypto/foo/foo.h) is being removed, and the exported headers are
being moved to the include/openssl/ directory instead.
Change-Id: I4c1d80849544713308ddc6999a549848afc25f94
Signed-off-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Rich Salz <rsalz@openssl.org>
OPENSSL_FIPSCANISTER is only set if the fips module is being built
(as opposed to being used). Since the fips module wont be built in
master this is redundant.
Reviewed-by: Tim Hudson <tjh@openssl.org>
These correspond to targets of the same name in test/Makefile that clash when
using the single-makefile build method using GitConfigure and GitMake.
Change-Id: If7e900c75f4341b446608b6916a3d76f202026ea
Signed-off-by: Mike Bland <mbland@acm.org>
Signed-off-by: Geoff Thorpe <geoff@openssl.org>
Reviewed-by: Richard Levitte <levitte@openssl.org>
The trial division and probable prime with coprime tests are disabled
on WIN32 builds because they use internal functions not exported from
the WIN32 DLLs.
Reviewed-by: Emilia Käsper <emilia@openssl.org>
This is actually ok for this function, but initialised to zero anyway if
PURIFY defined.
This does have the impact of masking any *real* unitialised data reads in bn though.
Patch based on approach suggested by Rich Salz.
PR#3415
The lazy-initialisation of BN_MONT_CTX was serialising all threads, as
noted by Daniel Sands and co at Sandia. This was to handle the case that
2 or more threads race to lazy-init the same context, but stunted all
scalability in the case where 2 or more threads are doing unrelated
things! We favour the latter case by punishing the former. The init work
gets done by each thread that finds the context to be uninitialised, and
we then lock the "set" logic after that work is done - the winning
thread's work gets used, the losing threads throw away what they've done.
Signed-off-by: Geoff Thorpe <geoff@openssl.org>
It's not clear whether this inconsistency could lead to an actual
computation error, but it involved a BIGNUM being passed around the
montgomery logic in an inconsistent state. This was found using flags
-DBN_DEBUG -DBN_DEBUG_RAND, and working backwards from this assertion
in 'ectest';
ectest: bn_mul.c:960: BN_mul: Assertion `(_bnum2->top == 0) ||
(_bnum2->d[_bnum2->top - 1] != 0)' failed
Signed-off-by: Geoff Thorpe <geoff@openssl.org>
Fix for the attack described in the paper "Recovering OpenSSL
ECDSA Nonces Using the FLUSH+RELOAD Cache Side-channel Attack"
by Yuval Yarom and Naomi Benger. Details can be obtained from:
http://eprint.iacr.org/2014/140
Thanks to Yuval Yarom and Naomi Benger for discovering this
flaw and to Yuval Yarom for supplying a fix.
(cherry picked from commit 2198be3483)
Conflicts:
CHANGES
Improve RSA sing performance by 20-30% by:
- switching from floating-point to integer conditional moves;
- daisy-chaining sqr-sqr-sqr-sqr-sqr-mul sequences;
- using MONTMUL even during powers table setup;
This change adds the option to calculate (EC)DSA nonces by hashing the
message and private key along with entropy to avoid leaking the private
key if the PRNG fails.
requested size. Fixes OpenSSL #2701.
This change does not address the cases of generating safe primes, or
where the |add| parameter is non-NULL.
Conflicts:
crypto/bn/bn.h
crypto/bn/bn_err.c
eliminating them as dead code.
Both volatile and "memory" are used because of some concern that the compiler
may still cache values across the asm block without it, and because this was
such a painful debugging session that I wanted to ensure that it's never
repeated.
