Instead, use BN_mod_exp_mont_consttime of p - 2. This removes two more
call sites sensitive to BN_FLG_CONSTTIME. We're down to just that last
BN_mod_inverse modulo φ(n). (Sort of. It's actually not sensitive
because even mod inverses always hit the other codepath. Perhaps we
should just leave it alone.)
Note this comes with a slight behavior change. The BN_MONT_CTXs are
initialized a little earlier. If a caller calls RSA_generate_* and then
reaches into the struct to scrap all the fields on it, they'll get
confused. Before, they had to perform an operation on it to get
confused. This is a completely ridiculous thing to do.
Since we do this a lot, this introduces some convenience functions for
doing the Fermat's Little Theorem mod inverse and fixes a leak in the
DSA code should computing kinv hit a malloc error.
BUG=125
Change-Id: Iafcae2fc6fd379d161f015c90ff7050e2282e905
Reviewed-on: https://boringssl-review.googlesource.com/12925
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: Adam Langley <agl@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Get one step closer to removing the dependency on |BN_div| from most
programs. Also get one step closer to a constant-time implementation of
|BN_MONT_CTX_set|; we now "just" need to create a constant-time variant
of |BN_mod_lshift1_quick|.
Note that this version might actually increase the side channel signal,
since the variance in timing in |BN_div| is probably less than the variance
from the many conditional reductions in the new method.
On one Windows x64 machine, the speed of RSA verification using the new
version is not too different from the speed of the old code. However,
|BN_div| is generally slow on Windows x64 so I expect this isn't faster
on all platforms. Regardless, we generally consider ECDSA/EdDSA
signature verification performance to be adaquate and RSA signature
verification is much, much faster even with this change.
For RSA signing the performance is not a significant factor since
performance-sensitive applications will cache the |RSA| structure and
the |RSA| structure will cache the Montgomery contexts.
Change-Id: Ib14f1a35c99b8da435e190342657f6a839381a1a
Reviewed-on: https://boringssl-review.googlesource.com/10520
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: Adam Langley <agl@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
I missed these in the last round.
Change-Id: I9b47216eef87c662728e454670e9e516de71ca21
Reviewed-on: https://boringssl-review.googlesource.com/11740
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Macros need a healthy dose of parentheses to avoid expression-level
misparses. Most of this comes from the clang-tidy CL here:
https://android-review.googlesource.com/c/235696/
Also switch most of the macros to use do { ... } while (0) to avoid all
the excessive comma operators and statement-level misparses.
Change-Id: I4c2ee51e347d2aa8c74a2d82de63838b03bbb0f9
Reviewed-on: https://boringssl-review.googlesource.com/11660
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
Simplify the calculation of the Montgomery constants in
|BN_MONT_CTX_set|, making the inversion constant-time. It should also
be faster by avoiding any use of the |BIGNUM| API in favor of using
only 64-bit arithmetic.
Now it's obvious how it works. /s
Change-Id: I59a1e1c3631f426fbeabd0c752e0de44bcb5fd75
Reviewed-on: https://boringssl-review.googlesource.com/9031
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: Adam Langley <agl@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
There's a __pragma expression which allows this. Android builds us Windows with
MinGW for some reason, so we actually do have to tolerate non-MSVC-compatible
Windows compilers. (Clang for Windows is much more sensible than MinGW and
intentionally mimicks MSVC.)
MinGW doesn't understand MSVC's pragmas and warns a lot. #pragma warning is
safe to suppress, so wrap those to shush them. This also lets us do away with a
few ifdefs.
Change-Id: I1f5a8bec4940d4b2d947c4c1cc9341bc15ec4972
Reviewed-on: https://boringssl-review.googlesource.com/8236
Reviewed-by: Adam Langley <agl@google.com>
If the function returns early due to an error, then the coordinates of the
result will have their |top| value set to a value beyond what has actually
been been written. Fix that, and make it easier to avoid such issues in the
future by refactoring the code.
As a bonus, avoid a false positive MSVC 64-bit opt build "potentially
uninitialized value used" warning.
Change-Id: I8c48deb63163a27f739c8797962414f8ca2588cd
Reviewed-on: https://boringssl-review.googlesource.com/6579
Reviewed-by: David Benjamin <davidben@google.com>
It is only used by |bn_div_rem_words|.
Change-Id: I57627091d8db5890d7fea34d8560897717008646
Reviewed-on: https://boringssl-review.googlesource.com/7128
Reviewed-by: David Benjamin <davidben@google.com>
Create a |bn_div_rem_words| that is used for double-word/single-word
divisions and division-with-remainder. Remove all implementations of
|bn_div_words| except for the implementation needed for 64-bit MSVC.
This allows more code to be shared across platforms and also removes
an instance of the dangerous pattern wherein the |div_asm| macro
modified a variable that wasn't passed as a parameter.
Also, document the limitations of the compiler-generated code for the
non-asm code paths more fully. Compilers indeed have not improved in
this respect.
Change-Id: I5a36a2edd7465de406d47d72dcd6bf3e63e5c232
Reviewed-on: https://boringssl-review.googlesource.com/7127
Reviewed-by: David Benjamin <davidben@google.com>
It is always the case that either |BN_ULLONG| is defined or
|BN_UMULT_LOHI| is defined because |BN_ULLONG| is defined everywhere
except 64-bit MSVC, and BN_UMULT_LOHI is defined for 64-bit MSVC.
Change-Id: I85e5d621458562501af1af65d587c0b8d937ba3b
Reviewed-on: https://boringssl-review.googlesource.com/7044
Reviewed-by: David Benjamin <davidben@google.com>
mul.c is the only file that uses these values.
Change-Id: I50a685cbff0f26357229e742f42e014434e9cebe
Reviewed-on: https://boringssl-review.googlesource.com/7061
Reviewed-by: David Benjamin <davidben@google.com>
It is always the case that |BN_ULLONG| is defined or we're building for
64-bit MSVC. Lots of code is trying to handle impossible cases where
neither of those is true.
Change-Id: Ie337adda1dfb453843c6e0999807dfa1afb1ed89
Reviewed-on: https://boringssl-review.googlesource.com/7043
Reviewed-by: David Benjamin <davidben@google.com>
This allows much code to be subsequently simplified and removed.
Change-Id: I0ac256957c6eae9f35a70508bd454cb44f3f8653
Reviewed-on: https://boringssl-review.googlesource.com/7042
Reviewed-by: David Benjamin <davidben@google.com>
BN_LLONG is only ever used in #ifdefs. The actual type is BN_ULLONG. Switch the
ifdefs to check on BN_ULLONG and remove BN_LLONG. Also fix signedness of all
the constants (potentially avoiding undefined behavior in some operations).
Change-Id: I3e7739bbe14c50ea7db04fc507a034a8cb315a5f
Reviewed-on: https://boringssl-review.googlesource.com/6518
Reviewed-by: Adam Langley <agl@google.com>
The comment in |BN_mod_inverse_ex| makes it clear that |BN_BITS2| was
intended. Besides fixing the code to match the comment, remove
the now-unused |BN_BITS| and the already-unused |BN_MASK| to prevent
future confusion of this sort.
On MSVC builds there seems to be very little difference in performance
between the two code paths according to |bssl speed|.
Change-Id: I765b7b3d464e2057b1d7952af25b6deb2724976a
Reviewed-on: https://boringssl-review.googlesource.com/6525
Reviewed-by: Adam Langley <agl@google.com>
Previously, both crypto/dh and crypto/ec defined |TOBN| macros that did
the same thing, but which took their arguments in the opposite order.
This change makes the code consistently use the same macro. It also
makes |STATIC_BIGNUM| available for internal use outside of crypto/bn.
Change-Id: Ide57f6a5b74ea95b3585724c7e1a630c82a864d9
Reviewed-on: https://boringssl-review.googlesource.com/6528
Reviewed-by: Adam Langley <agl@google.com>
BN_bin2bn takes a size_t as it should, but it passes that into bn_wexpand which
takes unsigned. Switch bn_wexpand and bn_expand to take size_t before they
check bounds against INT_MAX.
BIGNUM itself still uses int everywhere and we may want to audit all the
arithmetic at some point. Although I suspect having bn_expand require that the
number of bits fit in an int is sufficient to make everything happy, unless
we're doing interesting arithmetic on the number of bits somewhere.
Change-Id: Id191a4a095adb7c938cde6f5a28bee56644720c6
Reviewed-on: https://boringssl-review.googlesource.com/5680
Reviewed-by: Adam Langley <agl@google.com>
inttypes.h kindly requires a feature macro in C++ on some platforms, due
to a bizarre footnote in C99 (see footnote 191 in section 7.8.1). As
bn.h is a public header, we must leak this wart to the consumer. On
platforms with unfriendly inttypes.h headers, using BN_DEC_FMT1 and
friends now require the feature macro be defined externally.
This broke the Chromium Android Clang builder:
http://build.chromium.org/p/chromium.linux/builders/Android%20Clang%20Builder%20%28dbg%29/builds/59288
Change-Id: I88275a6788c7babd0eae32cae86f115bfa93a591
Reviewed-on: https://boringssl-review.googlesource.com/4688
Reviewed-by: Adam Langley <agl@google.com>
It seems Android's inttypes.h refuses to define those macros on C++ unless
__STDC_FORMAT_MACROS is set. This unbreaks the roll on Android.
Change-Id: Iad6c971b4789f0302534d9e5022534c6124e0ff0
Reviewed-on: https://boringssl-review.googlesource.com/4202
Reviewed-by: Adam Langley <agl@google.com>
Win64 fires significantly more warnings than Win32. Also some recent
changes made it grumpy.
(We might want to reconsider enabling all of MSVC's warnings. Given the sorts
of warnings some of these are, I'm not sure MSVC's version of -Wall -Werror is
actually tenable. Plus, diverging from the Chromium build, especially before
the bots are ready, is going to break pretty readily.)
Change-Id: If3b8feccf910ceab4a233b0731e7624d7da46f87
Reviewed-on: https://boringssl-review.googlesource.com/3420
Reviewed-by: Adam Langley <agl@google.com>
This is an initial cut at aarch64 support. I have only qemu to test it
however—hopefully hardware will be coming soon.
This also affects 32-bit ARM in that aarch64 chips can run 32-bit code
and we would like to be able to take advantage of the crypto operations
even in 32-bit mode. AES and GHASH should Just Work in this case: the
-armx.pl files can be built for either 32- or 64-bit mode based on the
flavour argument given to the Perl script.
SHA-1 and SHA-256 don't work like this however because they've never
support for multiple implementations, thus BoringSSL built for 32-bit
won't use the SHA instructions on an aarch64 chip.
No dedicated ChaCha20 or Poly1305 support yet.
Change-Id: Ib275bc4894a365c8ec7c42f4e91af6dba3bd686c
Reviewed-on: https://boringssl-review.googlesource.com/2801
Reviewed-by: Adam Langley <agl@google.com>
Upstream (impressively quickly) fixed the missing intrinsic. Switch Windows
clang back to building the same code as MSVC. Also include the intrin.h header
rather than forward-declare the intrinsic. clang only works if the header is
explicitly included. Chromium forcibly includes it to work around these kinds
of issues, but we shouldn't rely on it.
BUG=crbug.com/438382
Change-Id: I0ff6d48e1a3aa455cff99f8dc4c407e88b84d446
Reviewed-on: https://boringssl-review.googlesource.com/2461
Reviewed-by: Adam Langley <agl@google.com>
Windows clang lacks _umul128, but it has inline assembly so just use
that.
Change-Id: I6ff5d2465edc703a4d47ef0efbcea43d6fcc79fa
Reviewed-on: https://boringssl-review.googlesource.com/2454
Reviewed-by: Adam Langley <agl@google.com>
It's never used, upstream or downstream. The 64-bit value is wrong anyway for
LLP64 platforms.
Change-Id: I56afc51f4c17ed3f1c30959b574034f181b5b0c7
Reviewed-on: https://boringssl-review.googlesource.com/2123
Reviewed-by: Adam Langley <agl@google.com>
Android uses these for some conversions from Java formats. The code is
sufficiently bespoke that putting the conversion functions into
BoringSSL doesn't make a lot of sense, but the alternative is to expose
these ones.
Change-Id: If1362bc4a5c44cba4023c909e2ba6488ae019ddb
Apart from the obvious little issues, this also works around a
(seeming) libtool/linker:
a.c defines a symbol:
int kFoo;
b.c uses it:
extern int kFoo;
int f() {
return kFoo;
}
compile them:
$ gcc -c a.c
$ gcc -c b.c
and create a dummy main in order to run it, main.c:
int f();
int main() {
return f();
}
this works as expected:
$ gcc main.c a.o b.o
but, if we make an archive:
$ ar q lib.a a.o b.o
and use that:
$ gcc main.c lib.a
Undefined symbols for architecture x86_64
"_kFoo", referenced from:
_f in lib.a(b.o)
(It doesn't matter what order the .o files are put into the .a)
Linux and Windows don't seem to have this problem.
nm on a.o shows that the symbol is of type "C", which is a "common symbol"[1].
Basically the linker will merge multiple common symbol definitions together.
If ones makes a.c read:
int kFoo = 0;
Then one gets a type "D" symbol - a "data section symbol" and everything works
just fine.
This might actually be a libtool bug instead of an ld bug: Looking at `xxd
lib.a | less`, the __.SYMDEF SORTED index at the beginning of the archive
doesn't contain an entry for kFoo unless initialised.
Change-Id: I4cdad9ba46e9919221c3cbd79637508959359427
Initial fork from f2d678e6e89b6508147086610e985d4e8416e867 (1.0.2 beta).
(This change contains substantial changes from the original and
effectively starts a new history.)
