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>
|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>
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>
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>
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>
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>
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>
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>
This follows similar lines as the previous cleanups and fixes the
documentation of the preconditions.
And with that, RSA private key operations, provided p and q have the
same bit length, should be constant time, as far as I know. (Though I'm
sure I've missed something.)
bn_cmp_part_words and bn_cmp_words are no longer used and deleted.
Bug: 234
Change-Id: Iceefa39f57e466c214794c69b335c4d2c81f5577
Reviewed-on: https://boringssl-review.googlesource.com/25404
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>
We still need BN_mul and, in particular, bn_mul_recursive will either
require bn_abs_sub_words be generalized or that we add a parallel
bn_abs_sub_part_words, but start with the easy one.
While I'm here, simplify the i and j mess in here. It's patterned after
the multiplication one, but can be much simpler.
Bug: 234
Change-Id: If936099d53304f2512262a1cbffb6c28ae30ccee
Reviewed-on: https://boringssl-review.googlesource.com/25325
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>
This is to be used in constant-time RSA CRT.
Bug: 233
Change-Id: Ibade5792324dc6aba38cab6971d255d41fb5eb91
Reviewed-on: https://boringssl-review.googlesource.com/25286
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>
Use the now constant-time modular arithmetic functions.
Bug: 236
Change-Id: I4567d67bfe62ca82ec295f2233d1a6c9b131e5d2
Reviewed-on: https://boringssl-review.googlesource.com/25285
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>
As the EC code will ultimately want to use these in "words" form by way
of EC_FELEM, and because it's much easier, I've implement these as
low-level words-based functions that require all inputs have the same
width. The BIGNUM versions which RSA and, for now, EC calls are
implemented on top of that.
Unfortunately, doing such things in constant-time and accounting for
undersized inputs requires some scratch space, and these functions don't
take BN_CTX. So I've added internal bn_mod_*_quick_ctx functions that
take a BN_CTX and the old functions now allocate a bit unnecessarily.
RSA only needs lshift (for BN_MONT_CTX) and sub (for CRT), but the
generic EC code wants add as well.
The generic EC code isn't even remotely constant-time, and I hope to
ultimately use stack-allocated EC_FELEMs, so I've made the actual
implementations here implemented in "words", which is much simpler
anyway due to not having to take care of widths.
I've also gone ahead and switched the EC code to these functions,
largely as a test of their performance (an earlier iteration made the EC
code noticeably slower). These operations are otherwise not
performance-critical in RSA.
The conversion from BIGNUM to BIGNUM+BN_CTX should be dropped by the
static linker already, and the unused BIGNUM+BN_CTX functions will fall
off when EC_FELEM happens.
Update-Note: BN_mod_*_quick bounce on malloc a bit now, but they're not
really used externally. The one caller I found was wpa_supplicant
which bounces on malloc already. They appear to be implementing
compressed coordinates by hand? We may be able to convince them to
call EC_POINT_set_compressed_coordinates_GFp.
Bug: 233, 236
Change-Id: I2bf361e9c089e0211b97d95523dbc06f1168e12b
Reviewed-on: https://boringssl-review.googlesource.com/25261
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 fallback functions still themselves leak, but I've left TODOs there.
This only affects BN_mod_mul_montgomery on platforms where we don't use
the bn_mul_mont assembly, but BN_mul additionally affects the final
multiplication in RSA CRT.
Bug: 232
Change-Id: Ia1ae16162c38e10c056b76d6b2afbed67f1a5e16
Reviewed-on: https://boringssl-review.googlesource.com/25260
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>
This has no behavior change, but it has a semantic one. This CL is an
assertion that all BIGNUM functions tolerate non-minimal BIGNUMs now.
Specifically:
- Functions that do not touch top/width are assumed to not care.
- Functions that do touch top/width will be changed by this CL. These
should be checked in review that they tolerate non-minimal BIGNUMs.
Subsequent CLs will start adjusting the widths that BIGNUM functions
output, to fix timing leaks.
Bug: 232
Change-Id: I3a2b41b071f2174452f8d3801bce5c78947bb8f7
Reviewed-on: https://boringssl-review.googlesource.com/25257
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>
Test this by re-running bn_tests.txt tests a lot. For the most part,
this was done by scattering bn_minimal_width or bn_correct_top calls as
needed. We'll incrementally tease apart the functions that need to act
on non-minimal BIGNUMs in constant-time.
BN_sqr was switched to call bn_correct_top at the end, rather than
sample bn_minimal_width, in anticipation of later splitting it into
BN_sqr (for calculators) and BN_sqr_fixed (for BN_mod_mul_montgomery).
BN_div_word also uses bn_correct_top because it calls BN_lshift so
officially shouldn't rely on BN_lshift returning something
minimal-width, though I expect we'd want to split off a BN_lshift_fixed
than change that anyway?
The shifts sample bn_minimal_width rather than bn_correct_top because
they all seem to try to be very clever around the bit width. If we need
constant-time versions of them, we can adjust them later.
Bug: 232
Change-Id: Ie17b39034a713542dbe906cf8954c0c5483c7db7
Reviewed-on: https://boringssl-review.googlesource.com/25255
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>
Give a non-minimal modulus, there are two possible values of R we might
pick: 2^(BN_BITS2 * width) or 2^(BN_BITS2 * bn_minimal_width).
Potentially secret moduli would make the former attractive and things
might even work, but our only secret moduli (RSA) have public bit
widths. It's more cases to test and the usual BIGNUM invariant is that
widths do not affect numerical output.
Thus, settle on minimizing mont->N for now. With the top explicitly made
minimal, computing |lgBigR| is also a little simpler.
This CL also abstracts out the < R check in the RSA code, and implements
it in a width-agnostic way.
Bug: 232
Change-Id: I354643df30530db7866bb7820e34241d7614f3c2
Reviewed-on: https://boringssl-review.googlesource.com/25250
Reviewed-by: Adam Langley <agl@google.com>
This cuts down on a duplicated place where we mess with bn->top. It also
also better abstracts away what determines the value of R.
(I ordered this wrong and rebasing will be annoying. Specifically, the
question is what happens if the modulus is non-minimal. In
https://boringssl-review.googlesource.com/c/boringssl/+/25250/, R will
be determined by the stored width of mont->N, so we want to use mont's
copy of the modulus. Though, one way or another, the important part is
that it's inside the Montgomery abstraction.)
Bug: 232
Change-Id: I74212e094c8a47f396b87982039e49048a130916
Reviewed-on: https://boringssl-review.googlesource.com/25247
Reviewed-by: Adam Langley <agl@google.com>
This makes it easier going to and from non-minimal BIGNUMs and words
without worrying about the widths which are ultimately to become less
friendly.
Bug: 232
Change-Id: Ia57cb29164c560b600573c27b112ad9375a86aad
Reviewed-on: https://boringssl-review.googlesource.com/25245
Reviewed-by: Adam Langley <agl@google.com>
Thanks to Andres Erbsen for extremely helpful suggestions on how finally
plug this long-standing hole!
OpenSSL BIGNUMs are currently minimal-width, which means they cannot be
constant-time. We'll need to either excise BIGNUM from RSA and EC or
somehow fix BIGNUM. EC_SCALAR and later EC_FELEM work will excise it
from EC, but RSA's BIGNUMs are more transparent. Teaching BIGNUM to
handle non-minimal word widths is probably simpler.
The main constraint is BIGNUM's large "calculator" API surface. One
could, in theory, do arbitrary math on RSA components, which means all
public functions must tolerate non-minimal inputs. This is also useful
for EC; https://boringssl-review.googlesource.com/c/boringssl/+/24445 is
silly.
As a first step, fix comparison-type functions that were assuming
minimal BIGNUMs. I've also added bn_resize_words, but it is testing-only
until the rest of the library is fixed.
bn->top is now a loose upper bound we carry around. It does not affect
numerical results, only performance and secrecy. This is a departure
from the original meaning, and compiler help in auditing everything is
nice, so the final change in this series will rename bn->top to
bn->width. Thus these new functions are named per "width", not "top".
Looking further ahead, how are output BIGNUM widths determined? There's
three notions of correctness here:
1. Do I compute the right answer for all widths?
2. Do I handle secret data in constant time?
3. Does my memory usage not balloon absurdly?
For (1), a BIGNUM function must give the same answer for all input
widths. BN_mod_add_quick may assume |a| < |m|, but |a| may still be
wider than |m| by way of leading zeres. The simplest approach is to
write code in a width-agnostic way and rely on functions to accept all
widths. Where functions need to look at bn->d, we'll a few helper
functions to smooth over funny widths.
For (2), (1) is little cumbersome. Consider constant-time modular
addition. A sane type system would guarantee input widths match. But C
is weak here, and bifurcating the internals is a lot of work. Thus, at
least for now, I do not propose we move RSA's internal computation out
of BIGNUM. (EC_SCALAR/EC_FELEM are valuable for EC because we get to
stack-allocate, curves were already specialized, and EC only has two
types with many operations on those types. None of these apply to RSA.
We've got numbers mod n, mod p, mod q, and their corresponding
exponents, each of which is used for basically one operation.)
Instead, constant-time BIGNUM functions will output non-minimal widths.
This is trivial for BN_bin2bn or modular arithmetic. But for BN_mul,
constant-time[*] would dictate r->top = a->top + b->top. A calculator
repeatedly multiplying by one would then run out of memory. Those we'll
split into a private BN_mul_fixed for crypto, leaving BN_mul for
calculators. BN_mul is just BN_mul_fixed followed by bn_correct_top.
[*] BN_mul is not constant-time for other reasons, but that will be
fixed separately.
Bug: 232
Change-Id: Ide2258ae8c09a9a41bb71d6777908d1c27917069
Reviewed-on: https://boringssl-review.googlesource.com/25244
Reviewed-by: Adam Langley <agl@google.com>
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>
None of the asymmetric crypto we inherented from OpenSSL is
constant-time because of BIGNUM. BIGNUM chops leading zeros off the
front of everything, so we end up leaking information about the first
word, in theory. BIGNUM functions additionally tend to take the full
range of inputs and then call into BN_nnmod at various points.
All our secret values should be acted on in constant-time, but k in
ECDSA is a particularly sensitive value. So, ecdsa_sign_setup, in an
attempt to mitigate the BIGNUM leaks, would add a couple copies of the
order.
This does not work at all. k is used to compute two values: k^-1 and kG.
The first operation when computing k^-1 is to call BN_nnmod if k is out
of range. The entry point to our tuned constant-time curve
implementations is to call BN_nnmod if the scalar has too many bits,
which this causes. The result is both corrections are immediately undone
but cause us to do more variable-time work in the meantime.
Replace all these computations around k with the word-based functions
added in the various preceding CLs. In doing so, replace the BN_mod_mul
calls (which internally call BN_nnmod) with Montgomery reduction. We can
avoid taking k^-1 out of Montgomery form, which combines nicely with
Brian Smith's trick in 3426d10119. Along
the way, we avoid some unnecessary mallocs.
BIGNUM still affects the private key itself, as well as the EC_POINTs.
But this should hopefully be much better now. Also it's 10% faster:
Before:
Did 15000 ECDSA P-224 signing operations in 1069117us (14030.3 ops/sec)
Did 18000 ECDSA P-256 signing operations in 1053908us (17079.3 ops/sec)
Did 1078 ECDSA P-384 signing operations in 1087853us (990.9 ops/sec)
Did 473 ECDSA P-521 signing operations in 1069835us (442.1 ops/sec)
After:
Did 16000 ECDSA P-224 signing operations in 1064799us (15026.3 ops/sec)
Did 19000 ECDSA P-256 signing operations in 1007839us (18852.2 ops/sec)
Did 1078 ECDSA P-384 signing operations in 1079413us (998.7 ops/sec)
Did 484 ECDSA P-521 signing operations in 1083616us (446.7 ops/sec)
Change-Id: I2a25e90fc99dac13c0616d0ea45e125a4bd8cca1
Reviewed-on: https://boringssl-review.googlesource.com/23075
Reviewed-by: Adam Langley <agl@google.com>
These can be used to invert values in ECDSA. Unlike their BIGNUM
counterparts, the caller is responsible for taking values in and out of
Montgomery domain. This will save some work later on in the ECDSA
computation.
Change-Id: Ib7292900a0fdeedce6cb3e9a9123c94863659043
Reviewed-on: https://boringssl-review.googlesource.com/23071
Reviewed-by: Adam Langley <agl@google.com>
These use the square and multiply functions added earlier.
Change-Id: I723834f9a227a9983b752504a2d7ce0223c43d24
Reviewed-on: https://boringssl-review.googlesource.com/23070
Reviewed-by: Adam Langley <agl@google.com>
As part of excising BIGNUM from EC scalars, we will need a "words"
version of BN_mod_mul_montgomery. That, in turn, requires BN_sqr and
BN_mul for cases where we don't have bn_mul_mont.
BN_sqr and BN_mul have a lot of logic in there, with the most complex
cases being not even remotely constant time. Fortunately, those only
apply to RSA-sized numbers, not EC-sized numbers. (With the exception, I
believe, of 32-bit P-521 which just barely exceeds the cutoff.) Imposing
a limit also makes it easier to stack-allocate temporaries (BN_CTX
serves a similar purpose in BIGNUM).
Extract bn_mul_small and bn_sqr_small and test them as part of
bn_tests.txt. Later changes will build on these.
If we end up reusing these functions for RSA in the future (though that
would require tending to the egregiously non-constant-time code in the
no-asm build), we probably want to extract a version where there is an
explicit tmp parameter as in bn_sqr_normal rather than the stack bits.
Change-Id: If414981eefe12d6664ab2f5e991a359534aa7532
Reviewed-on: https://boringssl-review.googlesource.com/23068
Reviewed-by: Adam Langley <agl@google.com>
Also replace a pointless call to bn_mul_words with a memset.
Change-Id: Ief30ddab0e84864561b73fe2776bd0477931cf7f
Reviewed-on: https://boringssl-review.googlesource.com/23066
Reviewed-by: Adam Langley <agl@google.com>
This rewrites the internals with a "words" variant that can avoid
bn_correct_top. It still ultimately calls bn_correct_top as the calling
convention is sadly still BIGNUM, but we can lift that calling
convention out incrementally.
Performance seems to be comparable, if not faster.
Before:
Did 85000 ECDSA P-256 signing operations in 5030401us (16897.3 ops/sec)
Did 34278 ECDSA P-256 verify operations in 5048029us (6790.4 ops/sec)
After:
Did 85000 ECDSA P-256 signing operations in 5021057us (16928.7 ops/sec)
Did 34086 ECDSA P-256 verify operations in 5010416us (6803.0 ops/sec)
Change-Id: I1159746dfcc00726dc3f28396076a354556e6e7d
Reviewed-on: https://boringssl-review.googlesource.com/23065
Reviewed-by: Adam Langley <agl@google.com>
Normal shifts do the trick just fine and are less likely to tempt the
compiler into inserting a jump.
Change-Id: Iaa1da1b6f986fd447694fcde8f3525efb9eeaf11
Reviewed-on: https://boringssl-review.googlesource.com/22888
Reviewed-by: Adam Langley <agl@google.com>
This is an OpenSSL thing to support platforms where BN_ULONG is not
actually the size it claims to be. We define BN_ULONG to uint32_t and
uint64_t which are guaranteed by C to implement arithemetic modulo 2^32
and 2^64, respectively. Thus there is no need for any of this.
Change-Id: I098cd4cc050a136b9f2c091dfbc28dd83e01f531
Reviewed-on: https://boringssl-review.googlesource.com/21784
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>
This reverts commit f6942f0d22.
Reason for revert: This doesn't actually work in clang-cl. I
forgot we didn't have the clang-cl try bots enabled! :-( I
believe __asm__ is still okay, but I'll try it by hand
tomorrow.
Original change's description:
> Use uint128_t and __asm__ in clang-cl.
>
> clang-cl does not define __GNUC__ but is still a functioning clang. We
> should be able to use our uint128_t and __asm__ code in it on Windows.
>
> Change-Id: I67310ee68baa0c0c947b2441c265b019ef12af7e
> Reviewed-on: https://boringssl-review.googlesource.com/22184
> 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>
TBR=agl@google.com,davidben@google.com
Change-Id: I5c7e0391cd9c2e8cc0dfde37e174edaf5d17db22
No-Presubmit: true
No-Tree-Checks: true
No-Try: true
Reviewed-on: https://boringssl-review.googlesource.com/22224
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>
clang-cl does not define __GNUC__ but is still a functioning clang. We
should be able to use our uint128_t and __asm__ code in it on Windows.
Change-Id: I67310ee68baa0c0c947b2441c265b019ef12af7e
Reviewed-on: https://boringssl-review.googlesource.com/22184
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>
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>
