It actually works fine. I just forgot one of the typedefs last time.
This gives a roughly 2x improvement on P-256 in clang-cl +
OPENSSL_SMALL, the configuration used by Chrome.
Before:
Did 1302 ECDH P-256 operations in 1015000us (1282.8 ops/sec)
Did 4250 ECDSA P-256 signing operations in 1047000us (4059.2 ops/sec)
Did 1750 ECDSA P-256 verify operations in 1094000us (1599.6 ops/sec)
After:
Did 3250 ECDH P-256 operations in 1078000us (3014.8 ops/sec)
Did 8250 ECDSA P-256 signing operations in 1016000us (8120.1 ops/sec)
Did 3250 ECDSA P-256 verify operations in 1063000us (3057.4 ops/sec)
(These were taken on a VM, so the measurements are extremely noisy, but
this sort of improvement is visible regardless.)
Alas, we do need a little extra bit of fiddling because division does
not work (crbug.com/787617).
Bug: chromium:787617
Update-Note: This removes the MSan uint128_t workaround which does not
appear to be necessary anymore.
Change-Id: I8361314608521e5bdaf0e7eeae7a02c33f55c69f
Reviewed-on: https://boringssl-review.googlesource.com/23984
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>
We haven't supported MSVC 2013 for a while (we may even be able to drop
2015 in not too long). There is also no need to pull in stdalign.h in
C++. alignof and alignas are keywords.
Change-Id: Ib31d8166282592bcb9e1c543e57758ff55746404
Reviewed-on: https://boringssl-review.googlesource.com/20704
Commit-Queue: Steven Valdez <svaldez@google.com>
Reviewed-by: Steven Valdez <svaldez@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
crypto/{asn1,x509,x509v3,pem} were skipped as they are still OpenSSL
style.
Change-Id: I3cd9a60e1cb483a981aca325041f3fbce294247c
Reviewed-on: https://boringssl-review.googlesource.com/19504
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
The ticket encryption key is rotated automatically once every 24 hours,
unless a key has been configured manually (i.e. using
|SSL_CTX_set_tlsext_ticket_keys|) or one of the custom ticket encryption
methods is used.
Change-Id: I0dfff28b33e58e96b3bbf7f94dcd6d2642f37aec
Reviewed-on: https://boringssl-review.googlesource.com/18924
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>
The only place it is used is EC_KEY_{dup,copy} and no one calls that
function on an EC_KEY with ex_data. This aligns with functions like
RSAPublicKey_dup which do not copy ex_data. The logic is also somewhat
subtle in the face of malloc errors (upstream's PR 3323).
In fact, we'd even changed the function pointer signature from upstream,
so BoringSSL-only code is needed to pass this pointer in anyway. (I
haven't switched it to CRYPTO_EX_unused because there are some callers
which pass in an implementation anyway.)
Note, in upstream, the dup hook is also used for SSL_SESSIONs when those
are duplicated (for TLS 1.2 ticket renewal or TLS 1.3 resumption). Our
interpretation is that callers should treat those SSL_SESSIONs
equivalently to newly-established ones. This avoids every consumer
providing a dup hook and simplifies the interface.
(I've gone ahead and removed the TODO(fork). I don't think we'll be able
to change this API. Maybe introduce a new one, but it may not be worth
it? Then again, this API is atrocious... I've never seen anyone use argl
and argp even.)
BUG=21
Change-Id: I6c9e9d5a02347cb229d4c084c1e85125bd741d2b
Reviewed-on: https://boringssl-review.googlesource.com/16344
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>
Instead of a script which generates macros, emit static inlines in
individual header (or C files). This solves a few issues with the
original setup:
- The documentation was off. We match the documentation now.
- The stack macros did not check constness; see some of the fixes in
crypto/x509.
- Type errors did not look like usual type errors.
- Any type which participated in STACK_OF had to be made partially
public. This allows stack types to be defined an internal header or
even an individual file.
- One could not pass sk_FOO_free into something which expects a function
pointer.
Thanks to upstream's 411abf2dd37974a5baa54859c1abcd287b3c1181 for the
idea.
Change-Id: Ie5431390ccad761c17596b0e93941b0d7a68f904
Reviewed-on: https://boringssl-review.googlesource.com/16087
Reviewed-by: Adam Langley <agl@google.com>
Change-Id: Iab7a738a8981de7c56d1585050e78699cb876dab
Reviewed-on: https://boringssl-review.googlesource.com/16467
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Using |size_t| was correct, except for NaCl, which is a 64-bit build
with 32-bit pointers. In that configuration, |size_t| is smaller than
the native word size.
This change adds |crypto_word_t|, an unsigned type with native size and
switches constant-time functions to using it.
Change-Id: Ib275127063d5edbb7c55d413132711b7c74206b0
Reviewed-on: https://boringssl-review.googlesource.com/15325
Reviewed-by: Adam Langley <agl@google.com>
Fork-unsafe buffering was a mode that could be enabled by applications
that were sure that they didn't need to worry about state duplication.
It saved reads to urandom.
Since everything is now going through the CTR-DRBG, we can get the same
effect by simply not reading additional data from urandom in this case.
This change drops the buffering from urandom.c and, instead, implements
fork-unsafe buffering as a mode that skips reading additional data from
urandom, which only happened when RDRAND wasn't available anyway.
Since we expect the power-on self-tests to call into the PRNG, this
change also makes the flag capable of changing at any point by using a
mutex rather than a once. This is split into a separate file so that it
doesn't have to go into the FIPS module—since it uses r/w data that
would be a pain.
Change-Id: I5fd0ead0422e770e35758f080bb1cffa70d0c8da
Reviewed-on: https://boringssl-review.googlesource.com/14924
Reviewed-by: Adam Langley <agl@google.com>
We only need the size_t ones now.
BUG=22
Change-Id: Ie6935656bbc4bd2b602b8fad78effc401c493416
Reviewed-on: https://boringssl-review.googlesource.com/14312
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
These will be used in follow-up commits. The _s names are taken from
upstream, to ease importing code. I've also promoted the CONSTTIME_*
macros from the test. None of them are really necessary except
~0u cannot substitute for CONSTTIME_TRUE_S on 64-bit platforms, so
having the macros seems safer.
Once everything is converted, I expect the unsigned versions can be
removed, so I've made the _8 and _int functions act on size_t rather
than unsigned. The users of these functions basically only believe that
array indices and bytes exist.
BUG=22
Change-Id: I987bfb0c708dc726a6f2afcb05b6619bbd600564
Reviewed-on: https://boringssl-review.googlesource.com/14306
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
I thought I'd rewritten this, but apparently didn't. The old version
dated to a prior iteration which used macros.
Change-Id: Idefbdb2c11700a44dd5b0733b98efec102b10dd2
Reviewed-on: https://boringssl-review.googlesource.com/12968
Reviewed-by: Adam Langley <agl@google.com>
Most C standard library functions are undefined if passed NULL, even
when the corresponding length is zero. This gives them (and, in turn,
all functions which call them) surprising behavior on empty arrays.
Some compilers will miscompile code due to this rule. See also
https://www.imperialviolet.org/2016/06/26/nonnull.html
Add OPENSSL_memcpy, etc., wrappers which avoid this problem.
BUG=23
Change-Id: I95f42b23e92945af0e681264fffaf578e7f8465e
Reviewed-on: https://boringssl-review.googlesource.com/12928
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
This change adds AES and GHASH assembly from upstream, with the aim of
speeding up AES-GCM.
The PPC64LE assembly matches the interface of the ARMv8 assembly so I've
changed the prefix of both sets of asm functions to be the same
("aes_hw_").
Otherwise, the new assmebly files and Perlasm match exactly those from
upstream's c536b6be1a (from their master branch).
Before:
Did 1879000 AES-128-GCM (16 bytes) seal operations in 1000428us (1878196.1 ops/sec): 30.1 MB/s
Did 61000 AES-128-GCM (1350 bytes) seal operations in 1006660us (60596.4 ops/sec): 81.8 MB/s
Did 11000 AES-128-GCM (8192 bytes) seal operations in 1072649us (10255.0 ops/sec): 84.0 MB/s
Did 1665000 AES-256-GCM (16 bytes) seal operations in 1000591us (1664016.6 ops/sec): 26.6 MB/s
Did 52000 AES-256-GCM (1350 bytes) seal operations in 1006971us (51640.0 ops/sec): 69.7 MB/s
Did 8840 AES-256-GCM (8192 bytes) seal operations in 1013294us (8724.0 ops/sec): 71.5 MB/s
After:
Did 4994000 AES-128-GCM (16 bytes) seal operations in 1000017us (4993915.1 ops/sec): 79.9 MB/s
Did 1389000 AES-128-GCM (1350 bytes) seal operations in 1000073us (1388898.6 ops/sec): 1875.0 MB/s
Did 319000 AES-128-GCM (8192 bytes) seal operations in 1000101us (318967.8 ops/sec): 2613.0 MB/s
Did 4668000 AES-256-GCM (16 bytes) seal operations in 1000149us (4667304.6 ops/sec): 74.7 MB/s
Did 1202000 AES-256-GCM (1350 bytes) seal operations in 1000646us (1201224.0 ops/sec): 1621.7 MB/s
Did 269000 AES-256-GCM (8192 bytes) seal operations in 1002804us (268247.8 ops/sec): 2197.5 MB/s
Change-Id: Id848562bd4e1aa79a4683012501dfa5e6c08cfcc
Reviewed-on: https://boringssl-review.googlesource.com/11262
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>
Android uses MinGW for some host tools on Windows. That toolchain
doesn't support the #pragma tricks we use for thread-local destructors,
but does appear to support pthreads.
This also lets us remove the INIT_ONCE workaround, although that's
removable anyway since Android's MinGW is now new enough.
Change-Id: I8d1573923fdaac880a50d84acbebbf87461c50d2
Reviewed-on: https://boringssl-review.googlesource.com/11125
Reviewed-by: David Benjamin <davidben@google.com>
Reviewed-by: Kenny Root <kroot@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Change-Id: Ie60744761f5aa434a71a998f5ca98a8f8b1c25d5
Reviewed-on: https://boringssl-review.googlesource.com/10447
Reviewed-by: David Benjamin <davidben@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
MSVC 2015 seems to support it just fine.
Change-Id: I9c91c18c260031e6024480d1f57bbb334ed7118c
Reviewed-on: https://boringssl-review.googlesource.com/8501
Reviewed-by: Adam Langley <agl@google.com>
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>
Decrypting is very easy to do in-place, but encrypting in-place is a hassle.
The rules actually were wrong due to record-splitting. The aliasing prefix and
the alignment prefix actually differ by 1. Take it out for now in preparation
for tightening the aliasing rules.
If we decide to do in-place encrypt later, probably it'd be more useful to
return header + in-place ciphertext + trailer. (That, in turn, needs a
scatter/gather thing on the AEAD thanks to TLS 1.3's padding and record type
construction.) We may also wish to rethink how record-splitting works here.
Change-Id: I0187d39c541e76ef933b7c2c193323164fd8a156
Reviewed-on: https://boringssl-review.googlesource.com/8230
Reviewed-by: Adam Langley <agl@google.com>
Now that we no longer support Windows XP, this is available.
Unfortunately, the public header version of CRYPTO_MUTEX means we
still can't easily merge CRYPTO_MUTEX and CRYPTO_STATIC_MUTEX.
BUG=37
Change-Id: If309de3f06e0854c505083b72fd64d1dbb3f4563
Reviewed-on: https://boringssl-review.googlesource.com/8081
Reviewed-by: Adam Langley <agl@google.com>
Windows SRWLOCK requires you call different functions here. Split
them up in preparation for switching Windows from CRITICAL_SECTION.
BUG=37
Change-Id: I7b5c6a98eab9ae5bb0734b805cfa1ff334918f35
Reviewed-on: https://boringssl-review.googlesource.com/8080
Reviewed-by: Adam Langley <agl@google.com>
GCC gets unhappy if we don't initialize the padding.
Change-Id: I084ffee1717d9025dcb10d8f32de0da2339c7f01
Reviewed-on: https://boringssl-review.googlesource.com/7797
Reviewed-by: Steven Valdez <svaldez@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
The copy of mingw-w64 used by Android isn't new enough and is missing half of
the INIT_ONCE definitions. (But not the other half, strangely.) Work around
this for now.
Change-Id: I5c7e89db481f932e03477e50cfb3cbacaeb630e6
Reviewed-on: https://boringssl-review.googlesource.com/7790
Reviewed-by: Adam Langley <agl@google.com>
C and C++ disagree on the sizes of empty structs, which can be rather bad for
structs embedded in public headers. Stick a char in them to avoid issues. (It
doesn't really matter for CRYPTO_STATIC_MUTEX, but it's easier to add a char in
there too.)
Thanks to Andrew Chi for reporting this issue.
Change-Id: Ic54fff710b688decaa94848e9c7e1e73f0c58fd3
Reviewed-on: https://boringssl-review.googlesource.com/7760
Reviewed-by: David Benjamin <davidben@google.com>
Now that we no longer support Windows XP, this function is available. In doing
so, remove the odd run_once_arg_t union and pass in a pointer to a function
pointer which is cleaner and still avoids C's silly rule where function
pointers can't be placed in a void*.
BUG=37
Change-Id: I44888bb3779dacdb660706debd33888ca389ebd5
Reviewed-on: https://boringssl-review.googlesource.com/7613
Reviewed-by: David Benjamin <davidben@google.com>
Many of the compatibility issues are described at
https://msdn.microsoft.com/en-us/library/mt612856.aspx. The macros
that suppressed warnings on a per-function basis no longer work in
Update 1, so replace them with #pragmas. Update 1 warns when |size_t|
arguments to |printf| are casted, so stop doing that casting.
Unfortunately, this requires an ugly hack to continue working in
MSVC 2013 as MSVC 2013 doesn't support "%zu". Finally, Update 1 has new
warnings, some of which need to be suppressed.
---
Updated by davidben to give up on suppressing warnings in crypto/x509 and
crypto/x509v3 as those directories aren't changed much from upstream. In each
of these cases, upstream opted just blindly initialize the variable, so do the
same. Also switch C4265 to level 4, per Microsoft's recommendation and work
around a bug in limits.h that happens to get fixed by Google include order
style.
(limits.h is sensitive to whether corecrt.h, pulled in by stddef.h and some
other headers, is included before it. The reason it affected just one file is
we often put the file's header first, which means base.h is pulling in
stddef.h. Relying on this is ugly, but it's no worse than what everything else
is doing and this doesn't seem worth making something as tame as limits.h so
messy to use.)
Change-Id: I02d1f935356899f424d3525d03eca401bfa3e6cd
Reviewed-on: https://boringssl-review.googlesource.com/7480
Reviewed-by: David Benjamin <davidben@google.com>
Casting a pointer-to-non-volatile to pointer-to-volatile can be a no-op
as the compiler only requires volatile semantics when the pointed-to
object is a volatile object and there are no pointers-to-non-volatile
involved. This probably doesn't matter unless building with the MSVC
-volatile:iso flag, and maybe not even then, but it is good practice
anyway.
Change-Id: I94900d3dc61de3b8ce2ddecab2811907a9a7adbf
Reviewed-on: https://boringssl-review.googlesource.com/6973
Reviewed-by: David Benjamin <davidben@google.com>
MSVC doesn't have stdalign.h and so doesn't support |alignas| in C
code. Define |alignas(x)| as a synonym for |__decltype(align(x))|
instead for it.
This also fixes -Wcast-qual warnings in rsaz_exp.c.
Change-Id: Ifce9031724cb93f5a4aa1f567e7af61b272df9d5
Reviewed-on: https://boringssl-review.googlesource.com/6924
Reviewed-by: Adam Langley <agl@google.com>
Reviewed-by: David Benjamin <davidben@google.com>
This callback is never used. The one caller I've ever seen is in Android
code which isn't built with BoringSSL and it was a no-op.
It also doesn't actually make much sense. A callback cannot reasonably
assume that it sees every, say, SSL_CTX created because the index may be
registered after the first SSL_CTX is created. Nor is there any point in
an EX_DATA consumer in one file knowing about an SSL_CTX created in
completely unrelated code.
Replace all the pointers with a typedef to int*. This will ensure code
which passes NULL or 0 continues to compile while breaking code which
passes an actual function.
This simplifies some object creation functions which now needn't worry
about CRYPTO_new_ex_data failing. (Also avoids bouncing on the lock, but
it's taking a read lock, so this doesn't really matter.)
BUG=391192
Change-Id: I02893883c6fa8693682075b7b130aa538a0a1437
Reviewed-on: https://boringssl-review.googlesource.com/6625
Reviewed-by: Adam Langley <agl@google.com>
stdint.h already has macros for this. The spec says that, in C++,
__STDC_CONSTANT_MACROS is needed, so define it for bytestring_test.cc.
Chromium seems to use these macros without trouble, so I'm assuming we
can rely on them.
Change-Id: I56d178689b44d22c6379911bbb93d3b01dd832a3
Reviewed-on: https://boringssl-review.googlesource.com/6510
Reviewed-by: Adam Langley <agl@google.com>
Callers that lack hardware random may obtain a speed improvement by
calling |RAND_enable_fork_unsafe_buffering|, which enables a
thread-local buffer around reads from /dev/urandom.
Change-Id: I46e675d1679b20434dd520c58ece0f888f38a241
Reviewed-on: https://boringssl-review.googlesource.com/5792
Reviewed-by: Adam Langley <agl@google.com>
In the ancient times, before ex_data and OpenSSL, SSLeay supported a
single app_data slot in various types. Later app_data begat ex_data, and
app_data was replaced by compatibility macros to ex_data index zero.
Today, app_data is still in use, but ex_data never reserved index zero
for app_data. This causes some danger where, if the first ex_data
registration did not use NULL callbacks, the registration's callbacks
would collide with app_data.
Instead, add an option to the types with app_data to reserve index zero.
Also switch SSL_get_ex_data_X509_STORE_CTX_idx to always return zero
rather than allocate a new one. It used to be that you used
X509_STORE_CTX_get_app_data. I only found one consumer that we probably
don't care about, but, to be safe and since it's easy, go with the
conservative option. (Although SSL_get_ex_data_X509_STORE_CTX_idx wasn't
guaranteed to alias app_data, in practice it always did. No consumer
ever calls X509_STORE_CTX_get_ex_new_index.)
Change-Id: Ie75b279d60aefd003ffef103f99021c5d696a5e9
Reviewed-on: https://boringssl-review.googlesource.com/5313
Reviewed-by: Adam Langley <agl@google.com>
Support is spotty enough with compiler/library mismatches, and this doesn't
leak to public headers. It's probably simplest to just have consumers supply
it as a build flag.
BUG=491808
Change-Id: I0576a0514a266ee90d644317ae0f49cdddbafd1d
Reviewed-on: https://boringssl-review.googlesource.com/4880
Reviewed-by: Yoshisato Yanagisawa <yyanagisawa@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
Looks like it was the use in type_check.h that was still causing
problems, not that MSVC doesn't short-circuit #if statements.
Change-Id: I574e8dd463c46b0133a989b221a7bb8861b3eed9
6e1f6456 tried to do this, but MSVC doesn't short-circuit #if
statements. So this change tries having the test be in a different #if.
Change-Id: Id0074770c166a2b7cd9ba2c8cd06245a68b77af8
While the compiler on OS X sets the macros as if it supports C11
atomics, stdatomic.h is actually missing.
Change-Id: Ifecaf1c8df6390e6b994663adedc284d9b8130b7
OpenSSL has traditionally done reference counting with |int|s and the
|CRYPTO_add| function. Unless a special callback is installed (rare),
this is implemented by doing the reference count operations under a
lock.
This change adds infrastructure for handling reference counts and uses
atomic operations when C11 support is available.
Change-Id: Ia023ce432319efd00f77a7340da27d16ee4b63c3
Reviewed-on: https://boringssl-review.googlesource.com/4771
Reviewed-by: Adam Langley <agl@google.com>
Trusty doesn't have setjmp.h and nor does it have threads.
Change-Id: I005f7a009a13e6632513be9fab2bbe62294519a4
Reviewed-on: https://boringssl-review.googlesource.com/4660
Reviewed-by: Adam Langley <agl@google.com>
This is used with a platform API, so it should use the corresponding
platform type, saving us the size assert. It's ever defined in an
internal header, so we can freely use windows.h and friends.
Change-Id: Idc979309436adcf54524c835ddc2c98c3870d2e2
Reviewed-on: https://boringssl-review.googlesource.com/4680
Reviewed-by: Adam Langley <agl@google.com>
This reverts commit 68de407b5f. The NaCl fix has
rolled into Chromium.
Change-Id: I9fd6a6ae727c95fa89b8ce27e301f2a748d0acc9
Reviewed-on: https://boringssl-review.googlesource.com/4651
Reviewed-by: Adam Langley <agl@google.com>
Instead, each module defines a static CRYPTO_EX_DATA_CLASS to hold the values.
This makes CRYPTO_cleanup_all_ex_data a no-op as spreading the
CRYPTO_EX_DATA_CLASSes across modules (and across crypto and ssl) makes cleanup
slightly trickier. We can make it do something if needbe, but it's probably not
worth the trouble.
Change-Id: Ib6f6fd39a51d8ba88649f0fa29c66db540610c76
Reviewed-on: https://boringssl-review.googlesource.com/4375
Reviewed-by: Adam Langley <agl@google.com>
No functions for using it were ever added.
Change-Id: Iaee6e5bc8254a740435ccdcdbd715b851d8a0dce
Reviewed-on: https://boringssl-review.googlesource.com/4374
Reviewed-by: Adam Langley <agl@google.com>
No wrappers were ever added and codesearch confirms no one ever added to it
manually. Probably anyone doing complex things with BIOs just made a custom
BIO_METHOD. We can put it back with proper functions if the need ever arises.
Change-Id: Icb5da7ceeb8f1da6d08f4a8854d53dfa75827d9c
Reviewed-on: https://boringssl-review.googlesource.com/4373
Reviewed-by: Adam Langley <agl@google.com>
Callers are required to use the wrappers now. They still need OPENSSL_EXPORT
since crypto and ssl get built separately in the standalone shared library
build.
Change-Id: I61186964e6099b9b589c4cd45b8314dcb2210c89
Reviewed-on: https://boringssl-review.googlesource.com/4372
Reviewed-by: Adam Langley <agl@google.com>
