Fix some missing CFI bits.
Change-Id: I42114527f0ef8e03079d37a9f466d64a63a313f5
Reviewed-on: https://boringssl-review.googlesource.com/c/34864
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
This was caught by an aarch64 ABI tester. aarch64 has the same
considerations around small arguments as x86_64 does. The aarch64
version of bn_mul_mont does not mask off the upper words of the
argument.
The x86_64 version does, so size_t is, strictly speaking, wrong for
aarch64, but bn_mul_mont already has an implicit size limit to support
its internal alloca, so this doesn't really make things worse than
before.
Change-Id: I39bffc8fdb2287e45a2d1f0d1b4bd5532bbf3868
Reviewed-on: https://boringssl-review.googlesource.com/c/34804
Commit-Queue: David Benjamin <davidben@google.com>
Reviewed-by: Adam Langley <agl@google.com>
As part of this, move the CPU checks to C.
Change-Id: I17b701e1196c1ca116bbd23e0e669cf603ad464d
Reviewed-on: https://boringssl-review.googlesource.com/c/34626
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
It's an assembly function, so types are a little meaningless, but
everything is passed through as BN_ULONG, so be consistent. Also
annotate all the RSAZ prototypes with sizes.
Change-Id: I32e59e896da39e79c30ce9db52652fd645a033b4
Reviewed-on: https://boringssl-review.googlesource.com/c/34625
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
The standard computation model for constant-time code is that memory
access patterns must be independent of secret data.
BN_mod_exp_mont_consttime was previously written to a slightly weaker
model: only cacheline access patterns must be independent of secret
data. It assumed accesses within a cacheline were indistinguishable.
The CacheBleed attack (https://eprint.iacr.org/2016/224.pdf) showed this
assumption was false. Cache lines may be divided into cache banks, and
the researchers were able to measure cache bank contention pre-Haswell.
For Haswell, the researchers note "But, as Haswell does show timing
variations that depend on low address bits [19], it may be vulnerable to
similar attacks."
OpenSSL's fix to CacheBleed was not to adopt the standard constant-time
computation model. Rather, it now assumes accesses within a 16-byte
cache bank are indistinguishable, at least in the C copy_from_prebuf
path. These weaker models failed before with CacheBleed, so avoiding
such assumptions seems prudent. (The [19] citation above notes a false
dependence between memory addresses with a distance of 4k, which may be
what the paper was referring to.) Moreover, the C path is largely unused
on x86_64 (which uses mont5 asm), so it is especially questionable for
the generic C code to make assumptions based on x86_64.
Just walk the entire table in the C implementation. Doing so as-is comes
with a performance hit, but the striped memory layout is, at that point,
useless. We regain the performance loss (and then some) by using a more
natural layout. Benchmarks below.
This CL does not touch the mont5 assembly; I haven't figured out what
it's doing yet.
Pixel 3, aarch64:
Before:
Did 3146 RSA 2048 signing operations in 10009070us (314.3 ops/sec)
Did 447 RSA 4096 signing operations in 10026666us (44.6 ops/sec)
After:
Did 3210 RSA 2048 signing operations in 10010712us (320.7 ops/sec)
Did 456 RSA 4096 signing operations in 10063543us (45.3 ops/sec)
Pixel 3, armv7:
Before:
Did 2688 RSA 2048 signing operations in 10002266us (268.7 ops/sec)
Did 459 RSA 4096 signing operations in 10004785us (45.9 ops/sec)
After:
Did 2709 RSA 2048 signing operations in 10001299us (270.9 ops/sec)
Did 459 RSA 4096 signing operations in 10063737us (45.6 ops/sec)
x86_64 Broadwell, mont5 assembly disabled:
(This configuration is not actually shipped anywhere, but seemed a
useful data point.)
Before:
Did 14274 RSA 2048 signing operations in 10009130us (1426.1 ops/sec)
Did 2448 RSA 4096 signing operations in 10046921us (243.7 ops/sec)
After:
Did 14706 RSA 2048 signing operations in 10037908us (1465.0 ops/sec)
Did 2538 RSA 4096 signing operations in 10059986us (252.3 ops/sec)
Change-Id: If41da911d4281433856a86c6c8eadf99cd33e2d8
Reviewed-on: https://boringssl-review.googlesource.com/c/33268
Reviewed-by: Adam Langley <agl@google.com>
Commit-Queue: David Benjamin <davidben@google.com>
It's a table of BN_ULONGs. No particular need to use unsigned char.
Change-Id: I397883cef9f39fb162c2b0bfbd6a70fe399757a2
Reviewed-on: https://boringssl-review.googlesource.com/c/33267
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
C99 added macros such as PRIu64 to inttypes.h, but it said to exclude them from
C++ unless __STDC_FORMAT_MACROS or __STDC_CONSTANT_MACROS was defined. This
text was never incorporated into any C++ standard and explicitly overruled in
C++11.
Some libc headers followed C99. Notably, glibc prior to 2.18
(https://sourceware.org/bugzilla/show_bug.cgi?id=15366) and old versions of the
Android NDK.
In the NDK, although it was fixed some time ago (API level 20), the NDK used to
use separate headers per API level. Only applications using minSdkVersion >= 20
would get the fix. Starting NDK r14, "unified" headers are available which,
among other things, make the fix available (opt-in) independent of
minSdkVersion. In r15, unified headers are opt-out, and in r16 they are
mandatory.
Try removing these and see if anyone notices. The former is past our five year
watermark. The latter is not and Android has hit
https://boringssl-review.googlesource.com/c/boringssl/+/32686 before, but
unless it is really widespread, it's probably simpler to ask consumers to
define __STDC_CONSTANT_MACROS and __STDC_FORMAT_MACROS globally.
Update-Note: If you see compile failures relating to PRIu64, UINT64_MAX, and
friends, update your glibc or NDK. As a short-term fix, add
__STDC_CONSTANT_MACROS and __STDC_FORMAT_MACROS to your build, but get in touch
so we have a sense of how widespread it is.
Bug: 198
Change-Id: I56cca5f9acdff803de1748254bc45096e4c959c2
Reviewed-on: https://boringssl-review.googlesource.com/c/33146
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 change seems to have stuck, so bring us closer to C/++11 static asserts.
(If we later find we need to support worse toolchains, we can always use
__LINE__ or __COUNTER__ to avoid duplicate typedef names and just punt on
embedding the message into the type name.)
Change-Id: I0e5bb1106405066f07740728e19ebe13cae3e0ee
Reviewed-on: https://boringssl-review.googlesource.com/c/33145
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Since clang-cl uses __udivti3 for __uint128_t division, linking div.obj
fails. Let me make div.c use BN_CAN_DIVIDE_ULLONG to decide using
__uint128_t division instead of BN_ULLONG.
Bug: https://bugs.chromium.org/p/chromium/issues/detail?id=787617
Change-Id: I3ebe245f6b8917d59409591992efbabddea08187
Reviewed-on: https://boringssl-review.googlesource.com/c/32404
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>
The check of `r` instead of `rr` was introduced in change
I298400b988e3bd108d01d6a7c8a5b262ddf81feb.
Change-Id: I4376a81c65856f6457b0a11276176bf35e9c647d
Reviewed-on: https://boringssl-review.googlesource.com/31844
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>
https://boringssl-review.googlesource.com/31085 wasn't right. We already forbid
creating BN_MONT_CTX on negative numbers, which means almost all moduli already
don't work with BN_mod_exp_mont. Only -1 happened to not get rejected, but it
computed the wrong value. Reject it instead.
Update-Note: BN_mod_exp* will no longer work for negative moduli. It already
didn't work for all negative odd moduli other than -1, so rejecting -1 and
negative evens is unlikely to be noticed.
Bug: 71
Change-Id: I7c713d417e2e6512f3e78f402de88540809977e3
Reviewed-on: https://boringssl-review.googlesource.com/31484
Reviewed-by: Adam Langley <agl@google.com>
Historically, OpenSSL's modular exponentiation functions tolerated negative
moduli by ignoring the sign bit. The special case for a modulus of 1 should do
the same. That said, this is ridiculous and the only reason I'm importing this
is BN_abs_is_word(1) is marginally more efficient than BN_is_one() and we
haven't gotten around to enforcing positive moduli yet.
Thanks to Guido Vranken and OSSFuzz for finding this issue and reporting to
OpenSSL.
(Imported from upstream's 235119f015e46a74040b78b10fd6e954f7f07774.)
Change-Id: I526889dfbe2356753aa1e6ecfd3aa3dc3a8cd2b8
Reviewed-on: https://boringssl-review.googlesource.com/31085
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
This imports upstream's be4e1f79f631e49c76d02fe4644b52f907c374b2.
Change-Id: If0c4f066ba0ce540beaddd6a3e2540165d949dd2
Reviewed-on: https://boringssl-review.googlesource.com/31024
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>
Otherwise, if the output BIGNUM was previously negative, we'd incorrectly give
a negative result. Thanks to Guide Vranken for reporting this issue!
Fortunately, this does not appear to come up in any existing caller. This isn't
all that surprising as negative numbers never really come up in cryptography.
Were it not for OpenSSL historically designing a calculator API, we'd just
delete the bit altogether. :-(
Bug: chromium:865924
Change-Id: I28fdc986dfaba3e38435b14ebf07453d537cc60a
Reviewed-on: https://boringssl-review.googlesource.com/29944
Commit-Queue: David Benjamin <davidben@google.com>
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
When building files separately, omitting this causes some #defines to be
missing.
Change-Id: I235231467d3f51ee0a53325698356aefa72c6a67
Reviewed-on: https://boringssl-review.googlesource.com/28944
Commit-Queue: Adam Langley <agl@google.com>
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>
|alloca| is dangerous and poorly specified, according to any
description of |alloca|. It's also hard for some analysis tools to
reason about.
The code here assumed |alloca| is a macro, which isn't a valid
assumption. Depending on what which headers are included and what
toolchain is being used, |alloca| may or may not be defined as a macro,
and this might change over time if/when toolchains are updated. Or, we
might be doing static analysis and/or dynamic analysis with a different
configuration w.r.t. the availability of |alloca| than production
builds use.
Regardless, the |alloca| code path only kicked in when the inputs are
840 bits or smaller. Since the multi-prime RSA support was removed, for
interesting RSA key sizes the input will be at least 1024 bits and this
code path won't be triggered since powerbufLen will be larger than 3072
bytes in those cases. ECC inversion via Fermat's Little Theorem has its
own constant-time exponentiation so there are no cases where smaller
inputs need to be fast.
The RSAZ code avoids the |OPENSSL_malloc| for 2048-bit RSA keys.
Increasingly the RSAZ code won't be used though, since it will be
skipped over on Broadwell+ CPUs. Generalize the RSAZ stack allocation
to work for non-RSAZ code paths. In order to ensure this doesn't cause
too much stack usage on platforms where RSAZ wasn't already being used,
only do so on x86-64, which already has this large stack size
requirement due to RSAZ.
This change will make it easier to refactor |BN_mod_exp_mont_consttime|
to do that more safely and in a way that's more compatible with various
analysis tools.
This is also a step towards eliminating the |uintptr_t|-based alignment
hack.
Since this change increases the number of times |OPENSSL_free| is
skipped, I've added an explicit |OPENSSL_cleanse| to ensure the
zeroization is done. This should be done regardless of the other changes
here.
Change-Id: I8a161ce2720a26127e85fff7513f394883e50b2e
Reviewed-on: https://boringssl-review.googlesource.com/28584
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Reviewed-by: David Benjamin <davidben@google.com>
Build (and carry) issues are now resolved (as far as we know). Let's try
this again...
Measurements on a Skylake VM (so a little noisy).
Before:
Did 3135 RSA 2048 signing operations in 3015866us (1039.5 ops/sec)
Did 89000 RSA 2048 verify (same key) operations in 3007271us (29594.9 ops/sec)
Did 66000 RSA 2048 verify (fresh key) operations in 3014363us (21895.2 ops/sec)
Did 324 RSA 4096 signing operations in 3004364us (107.8 ops/sec)
Did 23126 RSA 4096 verify (same key) operations in 3003398us (7699.9 ops/sec)
Did 21312 RSA 4096 verify (fresh key) operations in 3017043us (7063.9 ops/sec)
Did 31040 ECDH P-256 operations in 3024273us (10263.6 ops/sec)
Did 91000 ECDSA P-256 signing operations in 3019740us (30135.0 ops/sec)
Did 25678 ECDSA P-256 verify operations in 3046975us (8427.4 ops/sec)
After:
Did 3640 RSA 2048 signing operations in 3035845us (1199.0 ops/sec)
Did 129000 RSA 2048 verify (same key) operations in 3003691us (42947.2 ops/sec)
Did 105000 RSA 2048 verify (fresh key) operations in 3029935us (34654.2 ops/sec)
Did 510 RSA 4096 signing operations in 3014096us (169.2 ops/sec)
Did 38000 RSA 4096 verify (same key) operations in 3092814us (12286.5 ops/sec)
Did 34221 RSA 4096 verify (fresh key) operations in 3003817us (11392.5 ops/sec)
Did 38000 ECDH P-256 operations in 3061758us (12411.2 ops/sec)
Did 116000 ECDSA P-256 signing operations in 3001637us (38645.6 ops/sec)
Did 35100 ECDSA P-256 verify operations in 3023872us (11607.6 ops/sec)
Tested with Intel SDE.
Change-Id: Ib27c0d6012d14274e331ab03f958e5a0c8b7e885
Reviewed-on: https://boringssl-review.googlesource.com/28104
Reviewed-by: Adam Langley <agl@google.com>
This was all new code. There was a request to make this available under
ISC.
Change-Id: Ibabbe6fbf593c2a781aac47a4de7ac378604dbcf
Reviewed-on: https://boringssl-review.googlesource.com/28267
Reviewed-by: Adam Langley <agl@google.com>
Montgomery multiplication post-conditions in some of code paths were
formally non-constant time. Cache access pattern was result-neutral,
but a little bit asymmetric, which might have produced a signal [if
processor reordered load and stores at run-time].
(Imported from upstream's 774ff8fed67e19d4f5f0df2f59050f2737abab2a.)
Change-Id: I77443fb79242b77e704c34d69f1de9e3162e9538
Reviewed-on: https://boringssl-review.googlesource.com/27987
Reviewed-by: Adam Langley <agl@google.com>
This introduces EC_FELEM, which is analogous to EC_SCALAR. It is used
for EC_POINT's representation in the generic EC_METHOD, as well as
random operations on tuned EC_METHODs that still are implemented
genericly.
Unlike EC_SCALAR, EC_FELEM's exact representation is awkwardly specific
to the EC_METHOD, analogous to how the old values were BIGNUMs but may
or may not have been in Montgomery form. This is kind of a nuisance, but
no more than before. (If p224-64.c were easily convertable to Montgomery
form, we could say |EC_FELEM| is always in Montgomery form. If we
exposed the internal add and double implementations in each of the
curves, we could give |EC_POINT| an |EC_METHOD|-specific representation
and |EC_FELEM| is purely a |EC_GFp_mont_method| type. I'll leave this
for later.)
The generic add and doubling formulas are aligned with the formulas
proved in fiat-crypto. Those only applied to a = -3, so I've proved a
generic one in https://github.com/mit-plv/fiat-crypto/pull/356, in case
someone uses a custom curve. The new formulas are verified,
constant-time, and swap a multiply for a square. As expressed in
fiat-crypto they do use more temporaries, but this seems to be fine with
stack-allocated EC_FELEMs. (We can try to help the compiler later,
but benchamrks below suggest this isn't necessary.)
Unlike BIGNUM, EC_FELEM can be stack-allocated. It also captures the
bounds in the type system and, in particular, that the width is correct,
which will make it easier to select a point in constant-time in the
future. (Indeed the old code did not always have the correct width. Its
point formula involved halving and implemented this in variable time and
variable width.)
Before:
Did 77274 ECDH P-256 operations in 10046087us (7692.0 ops/sec)
Did 5959 ECDH P-384 operations in 10031701us (594.0 ops/sec)
Did 10815 ECDSA P-384 signing operations in 10087892us (1072.1 ops/sec)
Did 8976 ECDSA P-384 verify operations in 10071038us (891.3 ops/sec)
Did 2600 ECDH P-521 operations in 10091688us (257.6 ops/sec)
Did 4590 ECDSA P-521 signing operations in 10055195us (456.5 ops/sec)
Did 3811 ECDSA P-521 verify operations in 10003574us (381.0 ops/sec)
After:
Did 77736 ECDH P-256 operations in 10029858us (7750.5 ops/sec) [+0.8%]
Did 7519 ECDH P-384 operations in 10068076us (746.8 ops/sec) [+25.7%]
Did 13335 ECDSA P-384 signing operations in 10029962us (1329.5 ops/sec) [+24.0%]
Did 11021 ECDSA P-384 verify operations in 10088600us (1092.4 ops/sec) [+22.6%]
Did 2912 ECDH P-521 operations in 10001325us (291.2 ops/sec) [+13.0%]
Did 5150 ECDSA P-521 signing operations in 10027462us (513.6 ops/sec) [+12.5%]
Did 4264 ECDSA P-521 verify operations in 10069694us (423.4 ops/sec) [+11.1%]
This more than pays for removing points_make_affine previously and even
speeds up ECDH P-256 slightly. (The point-on-curve check uses the
generic code.)
Next is to push the stack-allocating up to ec_wNAF_mul, followed by a
constant-time single-point multiplication.
Bug: 239
Change-Id: I44a2dff7c52522e491d0f8cffff64c4ab5cd353c
Reviewed-on: https://boringssl-review.googlesource.com/27668
Reviewed-by: Adam Langley <agl@google.com>
If the caller asked for the base to be treated as secret, we should
provide that. Allowing unbounded inputs is not compatible with being
constant-time.
Additionally, this aligns with the guidance here:
https://github.com/HACS-workshop/spectre-mitigations/blob/master/crypto_guidelines.md#1-do-not-conditionally-choose-between-constant-and-non-constant-time
Update-Note: BN_mod_exp_mont_consttime and BN_mod_exp_mont now require
inputs be fully reduced. I believe current callers tolerate this.
Additionally, due to a quirk of how certain operations were ordered,
using (publicly) zero exponent tolerated a NULL BN_CTX while other
exponents required non-NULL BN_CTX. Non-NULL BN_CTX is now required
uniformly. This is unlikely to cause problems. Any call site where the
exponent is always zero should just be replaced with BN_value_one().
Change-Id: I7c941953ea05f36dc2754facb9f4cf83a6789c61
Reviewed-on: https://boringssl-review.googlesource.com/27665
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Reviewed-by: Steven Valdez <svaldez@google.com>
No sense in adding impossible error cases we need to handle.
Additionally, tighten them a bit and require strong bounds. (I wasn't
sure what we'd need at first and made them unnecessarily general.)
Change-Id: I21a0afde90a55be2e9a0b8d7288f595252844f5f
Reviewed-on: https://boringssl-review.googlesource.com/27586
Reviewed-by: Adam Langley <alangley@gmail.com>
This is so the *_small functions can assume somewhat more uniform
widths, to simplify their error-handling.
Change-Id: I0420cb237084b253e918c64b0c170a5dfd99ab40
Reviewed-on: https://boringssl-review.googlesource.com/27584
Reviewed-by: Adam Langley <alangley@gmail.com>
https://boringssl-review.googlesource.com/10520 and then later
https://boringssl-review.googlesource.com/25285 made BN_MONT_CTX_set
constant-time, which is necessary for RSA's mont_p and mont_q. However,
due to a typo in the benchmark, they did not correctly measure.
Split BN_MONT_CTX creation into a constant-time and variable-time one.
The constant-time one uses our current algorithm and the latter restores
the original BN_mod codepath.
Should we wish to avoid BN_mod, I have an alternate version lying
around:
First, BN_set_bit + bn_mod_lshift1_consttime as now to count up to 2*R.
Next, observe that 2*R = BN_to_montgomery(2) and R*R =
BN_to_montgomery(R) = BN_to_montgomery(2^r_bits) Also observe that
BN_mod_mul_montgomery only needs n0, not RR. Split the core of
BN_mod_exp_mont into its own function so the caller handles conversion.
Raise 2*R to the r_bits power to get 2^r_bits*R = R*R.
The advantage of that algorithm is that it is still constant-time, so we
only need one BN_MONT_CTX_new. Additionally, it avoids BN_mod which is
otherwise (almost, but the remaining links should be easy to cut) out of
the critical path for correctness. One less operation to worry about.
The disadvantage is that it is gives a 25% (RSA-2048) or 32% (RSA-4096)
slower RSA verification speed. I went with the BN_mod one for the time
being.
Before:
Did 9204 RSA 2048 signing operations in 10052053us (915.6 ops/sec)
Did 326000 RSA 2048 verify (same key) operations in 10028823us (32506.3 ops/sec)
Did 50830 RSA 2048 verify (fresh key) operations in 10033794us (5065.9 ops/sec)
Did 1269 RSA 4096 signing operations in 10019204us (126.7 ops/sec)
Did 88435 RSA 4096 verify (same key) operations in 10031129us (8816.1 ops/sec)
Did 14552 RSA 4096 verify (fresh key) operations in 10053411us (1447.5 ops/sec)
After:
Did 9150 RSA 2048 signing operations in 10022831us (912.9 ops/sec)
Did 322000 RSA 2048 verify (same key) operations in 10028604us (32108.2 ops/sec)
Did 289000 RSA 2048 verify (fresh key) operations in 10017205us (28850.4 ops/sec)
Did 1270 RSA 4096 signing operations in 10072950us (126.1 ops/sec)
Did 87480 RSA 4096 verify (same key) operations in 10036328us (8716.3 ops/sec)
Did 80730 RSA 4096 verify (fresh key) operations in 10073614us (8014.0 ops/sec)
Change-Id: Ie8916d1634ccf8513ceda458fa302f09f3e93c07
Reviewed-on: https://boringssl-review.googlesource.com/27287
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
The first non-zero window (which we can condition on for public
exponents) always multiplies by one. This means we can cut out one
Montgomery multiplication. It also means we never actually need to
initialize r to one, saving another Montgomery multiplication for P-521.
This, in turn, means we don't need the bn_one_to_montgomery optimization
for the public-exponent exponentations, so we can delete
bn_one_to_montgomery_small. (The function does currently promise to
handle p = 0, but this is not actually reachable, so it can just do a
reduction on RR.)
For RSA, where we're not doing many multiplications to begin with,
saving one is noticeable.
Before:
Did 92000 RSA 2048 verify (same key) operations in 3002557us (30640.6 ops/sec)
Did 25165 RSA 4096 verify (same key) operations in 3045046us (8264.2 ops/sec)
After:
Did 100000 RSA 2048 verify (same key) operations in 3002483us (33305.8 ops/sec)
Did 26603 RSA 4096 verify (same key) operations in 3010942us (8835.4 ops/sec)
(Not looking at the fresh key number yet as that still needs to be
fixed.)
Change-Id: I81a025a68d9b0f8eb0f9c6c04ec4eedf0995a345
Reviewed-on: https://boringssl-review.googlesource.com/27286
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>
It's defined to return one in Montgomery form, not a normal one.
(Not that this matters. This function is only used to Fermat's Little
Theorem. Probably it should have been less general, though we'd need to
make new test vectors first.)
Change-Id: Ia8d7588e6a413b25f01280af9aacef0192283771
Reviewed-on: https://boringssl-review.googlesource.com/27285
Reviewed-by: Adam Langley <agl@google.com>
BN_mod_exp_mont is intended to protect the base, but not the exponent.
Accordingly, it shouldn't treat a base of zero as special.
Change-Id: Ib053e8ce65ab1741973a9f9bfeff8c353567439c
Reviewed-on: https://boringssl-review.googlesource.com/27284
Reviewed-by: Adam Langley <agl@google.com>
This is helpful at smaller sizes because the benefits of an unlikely hit
by trival-division are smaller.
The full set of kPrimes eliminates about 94.3% of random numbers. The
first quarter eliminates about 93.2% of them. But the little extra power
of the full set seems to be borderline for RSA 3072 and clearly positive
for RSA 4096.
Did 316 RSA 2048 key-gen operations in 30035598us (10.5 ops/sec)
min: 19423us, median: 80448us, max: 394265us
Change-Id: Iee53f721329674ae7a08fabd85b4f645c24e119d
Reviewed-on: https://boringssl-review.googlesource.com/26944
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Reviewed-by: David Benjamin <davidben@google.com>
This used to work, but I broke it on accident in the recent rewrite.
Change-Id: I06ab5e06eb0c0a6b67ecc97919654e386f3c2198
Reviewed-on: https://boringssl-review.googlesource.com/26984
Commit-Queue: David Benjamin <davidben@google.com>
Commit-Queue: Martin Kreichgauer <martinkr@google.com>
Reviewed-by: Martin Kreichgauer <martinkr@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
ECDSA converts digests to scalars by taking the leftmost n bits, where n
is the number of bits in the group order. This does not necessarily
produce a fully-reduced scalar.
Montgomery multiplication actually tolerates this slightly looser bound,
so we did not bother with the conditional subtraction. However, this
subtraction is free compared to the multiplication, inversion, and base
point multiplication. Simplify things by keeping it fully-reduced.
Change-Id: If49dffefccc21510f40418dc52ea4da7e3ff198f
Reviewed-on: https://boringssl-review.googlesource.com/26968
Reviewed-by: Adam Langley <agl@google.com>
May as well use it. Also avoid an overflow with digest_len if someone
asks to sign a truly enormous digest.
Change-Id: Ia0a53007a496f9c7cadd44b1020ec2774b310936
Reviewed-on: https://boringssl-review.googlesource.com/26966
Reviewed-by: Adam Langley <agl@google.com>
We do this in four different places, with the same long comment, and I'm
about to add yet another one.
Change-Id: If28e3f87ea71020d9b07b92e8947f3848473d99d
Reviewed-on: https://boringssl-review.googlesource.com/26964
Reviewed-by: Adam Langley <agl@google.com>
RSA keygen uses this to pick primes. May as well avoid bouncing on
malloc. (The BIGNUM internally allocates, of course, but that allocation
will be absorbed by BN_CTX in RSA keygen.)
Change-Id: Ie2243a6e48b9c55f777153cbf67ba5c06688c2f1
Reviewed-on: https://boringssl-review.googlesource.com/26887
Reviewed-by: Adam Langley <agl@google.com>
With this, in 0.02% of 1024-bit primes (which is what's used with an RSA
2048 generation), we'll leak that we struggled to generate values less
than the prime. I.e. that there's a greater likelihood of zero bits
after the leading 1 bit in the prime.
But this recovers all the speed loss from making key generation
constant-time, and then some.
Did 273 RSA 2048 key-gen operations in 30023223us (9.1 ops/sec)
min: 23867us, median: 93688us, max: 421466us
Did 66 RSA 3072 key-gen operations in 30041763us (2.2 ops/sec)
min: 117044us, median: 402095us, max: 1096538us
Did 31 RSA 4096 key-gen operations in 31673405us (1.0 ops/sec)
min: 245109us, median: 769480us, max: 2659386us
Change-Id: Id82dedde35f5fbb36b278189c0685a13c7824590
Reviewed-on: https://boringssl-review.googlesource.com/26924
Reviewed-by: Adam Langley <alangley@gmail.com>
This uses the full binary GCD algorithm, where all four of A, B, C, and
D must be retained. (BN_mod_inverse_odd implements the odd number
version which only needs A and C.) It is patterned after the version
in the Handbook of Applied Cryptography, but tweaked so the coefficients
are non-negative and bounded.
Median of 29 RSA keygens: 0m0.225s -> 0m0.220s
(Accuracy beyond 0.1s is questionable.)
Bug: 238
Change-Id: I6dc13524ea7c8ac1072592857880ddf141d87526
Reviewed-on: https://boringssl-review.googlesource.com/26370
Reviewed-by: Adam Langley <alangley@gmail.com>
RSA key generation requires computing a GCD (p-1 and q-1 are relatively
prime with e) and an LCM (the Carmichael totient). I haven't made BN_gcd
itself constant-time here to save having to implement
bn_lshift_secret_shift, since the two necessary operations can be served
by bn_rshift_secret_shift, already added for Rabin-Miller. However, the
guts of BN_gcd are replaced. Otherwise, the new functions are only
connected to tests for now, they'll be used in subsequent CLs.
To support LCM, there is also now a constant-time division function.
This does not replace BN_div because bn_div_consttime is some 40x slower
than BN_div. That penalty is fine for RSA keygen because that operation
is not bottlenecked on division, so we prefer simplicity over
performance.
Median of 29 RSA keygens: 0m0.212s -> 0m0.225s
(Accuracy beyond 0.1s is questionable.)
Bug: 238
Change-Id: Idbfbfa6e7f5a3b8782ce227fa130417b3702cf97
Reviewed-on: https://boringssl-review.googlesource.com/26369
Reviewed-by: Adam Langley <alangley@gmail.com>
Expose the constant-time abs_sub functions from the fixed Karatsuba code
in BIGNUM form for RSA to call into. RSA key generation involves
checking if |p - q| is above some lower bound.
BN_sub internally branches on which of p or q is bigger. For any given
iteration, this is not secret---one of p or q is necessarily the larger,
and whether we happened to pick the larger or smaller first is
irrelevant. Accordingly, there is no need to perform the p/q swap at the
end in constant-time.
However, this stage of the algorithm picks p first, sticks with it, and
then computes |p - q| for various q candidates. The distribution of
comparisons leaks information about p. The leak is unlikely to be
problematic, but plug it anyway.
Median of 29 RSA keygens: 0m0.210s -> 0m0.212s
(Accuracy beyond 0.1s is questionable.)
Bug: 238
Change-Id: I024b4e51b364f5ca2bcb419a0393e7be13249aec
Reviewed-on: https://boringssl-review.googlesource.com/26368
Reviewed-by: Adam Langley <alangley@gmail.com>
RSA key generation currently does the GCD check before the primality
test, in hopes of discarding things invalid by other means before
running the expensive primality check.
However, GCD is about to get a bit more expensive to clear the timing
leak, and the trial division part of primality testing is quite fast.
Thus, split that portion out via a new bn_is_obviously_composite and
call it before GCD.
Median of 29 RSA keygens: 0m0.252s -> 0m0.207s
(Accuracy beyond 0.1s is questionable.)
Bug: 238
Change-Id: I3999771fb73cca16797cab9332d14c4ebeb02046
Reviewed-on: https://boringssl-review.googlesource.com/26366
Reviewed-by: Adam Langley <alangley@gmail.com>
I'm not sure why I separated "fixed" and "quick_ctx" names. That's
annoying and doesn't generalize well to, say, adding a bn_div_consttime
function for RSA keygen.
Change-Id: I751d52b30e079de2f0d37a952de380fbf2c1e6b7
Reviewed-on: https://boringssl-review.googlesource.com/26364
Commit-Queue: David Benjamin <davidben@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
Rabin-Miller requires selecting a random number from 2 to |w|-1.
This is done by picking an N-bit number and discarding out-of-range
values. This leaks information about |w|, so apply blinding. Rather than
discard bad values, adjust them to be in range.
Though not uniformly selected, these adjusted values
are still usable as Rabin-Miller checks.
Rabin-Miller is already probabilistic, so we could reach the desired
confidence levels by just suitably increasing the iteration count.
However, to align with FIPS 186-4, we use a more pessimal analysis: we
do not count the non-uniform values towards the iteration count. As a
result, this function is more complex and has more timing risk than
necessary.
We count both total iterations and uniform ones and iterate until we've
reached at least |BN_PRIME_CHECKS_BLINDED| and |iterations|,
respectively. If the latter is large enough, it will be the limiting
factor with high probability and we won't leak information.
Note this blinding does not impact most calls when picking primes
because composites are rejected early. Only the two secret primes see
extra work. So while this does make the BNTest.PrimeChecking test take
about 2x longer to run on debug mode, RSA key generation time is fine.
Another, perhaps simpler, option here would have to run
bn_rand_range_words to the full 100 count, select an arbitrary
successful try, and declare failure of the entire keygen process (as we
do already) if all tries failed. I went with the option in this CL
because I happened to come up with it first, and because the failure
probability decreases much faster. Additionally, the option in this CL
does not affect composite numbers, while the alternate would. This gives
a smaller multiplier on our entropy draw. We also continue to use the
"wasted" work for stronger assurance on primality. FIPS' numbers are
remarkably low, considering the increase has negligible cost.
Thanks to Nathan Benjamin for helping me explore the failure rate as the
target count and blinding count change.
Now we're down to the rest of RSA keygen, which will require all the
operations we've traditionally just avoided in constant-time code!
Median of 29 RSA keygens: 0m0.169s -> 0m0.298s
(Accuracy beyond 0.1s is questionable. The runs at subsequent test- and
rename-only CLs were 0m0.217s, 0m0.245s, 0m0.244s, 0m0.247s.)
Bug: 238
Change-Id: Id6406c3020f2585b86946eb17df64ac42f30ebab
Reviewed-on: https://boringssl-review.googlesource.com/25890
Commit-Queue: Adam Langley <agl@google.com>
CQ-Verified: CQ bot account: commit-bot@chromium.org <commit-bot@chromium.org>
Reviewed-by: Adam Langley <agl@google.com>
(This is actually slightly silly as |a|'s probability distribution falls
off exponentially, but it's easy enough to do right.)
Instead, we run the loop to the end. This is still performant because we
can, as before, return early on composite numbers. Only two calls
actually run to the end. Moreover, running to the end has comparable
cost to BN_mod_exp_mont_consttime.
Median time goes from 0.140s to 0.231s. That cost some, but we're still
faster than the original implementation.
We're down to one more leak, which is that the BN_rand_range_ex call
does not hide |w1|. That one may only be solved probabilistically...
Median of 29 RSA keygens: 0m0.123s -> 0m0.145s
(Accuracy beyond 0.1s is questionable.)
Bug: 238
Change-Id: I4847cb0053118c572d2dd5f855388b5199fa6ce2
Reviewed-on: https://boringssl-review.googlesource.com/25888
Reviewed-by: Adam Langley <agl@google.com>
Compilers use a variant of Barrett reduction to divide by constants,
which conveniently also avoids problematic operations on the secret
numerator. Implement the variant as described here:
http://ridiculousfish.com/blog/posts/labor-of-division-episode-i.html
Repurpose this to implement a constant-time BN_mod_word replacement.
It's even much faster! I've gone ahead and replaced the other
BN_mod_word calls on the primes table.
That should give plenty of budget for the other changes. (I am assuming
that a regression is okay, as RSA keygen is not performance-sensitive,
but that I should avoid anything too dramatic.)
Proof of correctness: https://github.com/davidben/fiat-crypto/blob/barrett/src/Arithmetic/BarrettReduction/RidiculousFish.v
Median of 29 RSA keygens: 0m0.621s -> 0m0.123s
(Accuracy beyond 0.1s is questionable, though this particular
improvement is quite solid.)
Bug: 238
Change-Id: I67fa36ffe522365b13feb503c687b20d91e72932
Reviewed-on: https://boringssl-review.googlesource.com/25887
Reviewed-by: Adam Langley <agl@google.com>
The extra details in Enhanced Rabin-Miller are only used in
RSA_check_key_fips, on the public RSA modulus, which the static linker
will drop in most of our consumers anyway. Implement normal Rabin-Miller
for RSA keygen and use Montgomery reduction so it runs in constant-time.
Note that we only need to avoid leaking information about the input if
it's a large prime. If the number ends up composite, or we find it in
our table of small primes, we can return immediately.
The leaks not addressed by this CL are:
- The difficulty of selecting |b| leaks information about |w|.
- The distribution of whether step 4.4 runs leaks information about w.
- We leak |a| (the largest power of two which divides w) everywhere.
- BN_mod_word in the trial division is not constant-time.
These will be resolved in follow-up changes.
Median of 29 RSA keygens: 0m0.521 -> 0m0.621s
(Accuracy beyond 0.1s is questionable.)
Bug: 238
Change-Id: I0cf0ff22079732a0a3ababfe352bb4327e95b879
Reviewed-on: https://boringssl-review.googlesource.com/25886
Reviewed-by: Adam Langley <agl@google.com>
Primality testing checks for small words in random places.
Median of 29 RSA keygens: 0m0.811s -> 0m0.521s
(Accuracy beyond 0.1s is questionable, and this "speed up" is certainly
noise.)
Bug: 238
Change-Id: Ie5efab7291302a42ac6e283d25da0c094d8577e7
Reviewed-on: https://boringssl-review.googlesource.com/25885
Reviewed-by: Adam Langley <agl@google.com>
There are a number of random subtractions in RSA key generation. Add a
fixed-width version.
Median of 29 RSA keygens: 0m0.859s -> 0m0.811s
(Accuracy beyond 0.1s is questionable.)
Bug: 238
Change-Id: I9fa0771b95a438fd7d2635fd77a332146ccc96d9
Reviewed-on: https://boringssl-review.googlesource.com/25884
Commit-Queue: Adam Langley <agl@google.com>
Reviewed-by: Adam Langley <agl@google.com>
These are composite numbers whose composite witnesses aren't in the
first however many prime numbers, so deterministically checking small
numbers may not work.
We don't check composite witnesses deterministically but these are
probably decent tests. (Not sure how else to find composites with
scarce witnesses, but these seemed decent candidates.)
Change-Id: I23dcb7ba603a64c1f7d1e9a16942e7c29c76da51
Reviewed-on: https://boringssl-review.googlesource.com/26645
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>
