From be4e1f79f631e49c76d02fe4644b52f907c374b2 Mon Sep 17 00:00:00 2001 From: Kurt Roeckx Date: Wed, 25 Jul 2018 18:55:16 +0200 Subject: [PATCH] Make number of Miller-Rabin tests for a prime tests depend on the security level of the prime The old numbers where all generated for an 80 bit security level. But the number should depend on security level you want to reach. For bigger primes we want a higher security level and so need to do more tests. Reviewed-by: Richard Levitte Reviewed-by: Matthias St. Pierre Reviewed-by: Paul Dale GH: #6075 Fixes: #6012 (cherry picked from commit feac7a1c8be49fbcb76fcb721ec9f02fdd91030e) --- CHANGES | 7 +++ crypto/bn/bn.h | 87 +++++++++++++++++++++++++------- doc/crypto/BN_generate_prime.pod | 12 ++++- 3 files changed, 87 insertions(+), 19 deletions(-) diff --git a/CHANGES b/CHANGES index 47263c815c..1bf0f0b2a0 100644 --- a/CHANGES +++ b/CHANGES @@ -9,6 +9,13 @@ Changes between 1.0.2o and 1.0.2p [xx XXX xxxx] + *) Change generating and checking of primes so that the error rate of not + being prime depends on the intended use based on the size of the input. + For larger primes this will result in more rounds of Miller-Rabin. + The maximal error rate for primes with more than 1080 bits is lowered + to 2^-128. + [Kurt Roeckx, Annie Yousar] + *) Increase the number of Miller-Rabin rounds for DSA key generating to 64. [Kurt Roeckx] diff --git a/crypto/bn/bn.h b/crypto/bn/bn.h index 633d1b1f60..c056bba5f0 100644 --- a/crypto/bn/bn.h +++ b/crypto/bn/bn.h @@ -375,25 +375,76 @@ int BN_GENCB_call(BN_GENCB *cb, int a, int b); * on the size of the number */ /* - * number of Miller-Rabin iterations for an error rate of less than 2^-80 for - * random 'b'-bit input, b >= 100 (taken from table 4.4 in the Handbook of - * Applied Cryptography [Menezes, van Oorschot, Vanstone; CRC Press 1996]; - * original paper: Damgaard, Landrock, Pomerance: Average case error - * estimates for the strong probable prime test. -- Math. Comp. 61 (1993) - * 177-194) + * BN_prime_checks_for_size() returns the number of Miller-Rabin iterations + * that will be done for checking that a random number is probably prime. The + * error rate for accepting a composite number as prime depends on the size of + * the prime |b|. The error rates used are for calculating an RSA key with 2 primes, + * and so the level is what you would expect for a key of double the size of the + * prime. + * + * This table is generated using the algorithm of FIPS PUB 186-4 + * Digital Signature Standard (DSS), section F.1, page 117. + * (https://dx.doi.org/10.6028/NIST.FIPS.186-4) + * + * The following magma script was used to generate the output: + * securitybits:=125; + * k:=1024; + * for t:=1 to 65 do + * for M:=3 to Floor(2*Sqrt(k-1)-1) do + * S:=0; + * // Sum over m + * for m:=3 to M do + * s:=0; + * // Sum over j + * for j:=2 to m do + * s+:=(RealField(32)!2)^-(j+(k-1)/j); + * end for; + * S+:=2^(m-(m-1)*t)*s; + * end for; + * A:=2^(k-2-M*t); + * B:=8*(Pi(RealField(32))^2-6)/3*2^(k-2)*S; + * pkt:=2.00743*Log(2)*k*2^-k*(A+B); + * seclevel:=Floor(-Log(2,pkt)); + * if seclevel ge securitybits then + * printf "k: %5o, security: %o bits (t: %o, M: %o)\n",k,seclevel,t,M; + * break; + * end if; + * end for; + * if seclevel ge securitybits then break; end if; + * end for; + * + * It can be run online at: + * http://magma.maths.usyd.edu.au/calc + * + * And will output: + * k: 1024, security: 129 bits (t: 6, M: 23) + * + * k is the number of bits of the prime, securitybits is the level we want to + * reach. + * + * prime length | RSA key size | # MR tests | security level + * -------------+--------------|------------+--------------- + * (b) >= 6394 | >= 12788 | 3 | 256 bit + * (b) >= 3747 | >= 7494 | 3 | 192 bit + * (b) >= 1345 | >= 2690 | 4 | 128 bit + * (b) >= 1080 | >= 2160 | 5 | 128 bit + * (b) >= 852 | >= 1704 | 5 | 112 bit + * (b) >= 476 | >= 952 | 5 | 80 bit + * (b) >= 400 | >= 800 | 6 | 80 bit + * (b) >= 347 | >= 694 | 7 | 80 bit + * (b) >= 308 | >= 616 | 8 | 80 bit + * (b) >= 55 | >= 110 | 27 | 64 bit + * (b) >= 6 | >= 12 | 34 | 64 bit */ -# define BN_prime_checks_for_size(b) ((b) >= 1300 ? 2 : \ - (b) >= 850 ? 3 : \ - (b) >= 650 ? 4 : \ - (b) >= 550 ? 5 : \ - (b) >= 450 ? 6 : \ - (b) >= 400 ? 7 : \ - (b) >= 350 ? 8 : \ - (b) >= 300 ? 9 : \ - (b) >= 250 ? 12 : \ - (b) >= 200 ? 15 : \ - (b) >= 150 ? 18 : \ - /* b >= 100 */ 27) + +# define BN_prime_checks_for_size(b) ((b) >= 3747 ? 3 : \ + (b) >= 1345 ? 4 : \ + (b) >= 476 ? 5 : \ + (b) >= 400 ? 6 : \ + (b) >= 347 ? 7 : \ + (b) >= 308 ? 8 : \ + (b) >= 55 ? 27 : \ + /* b >= 6 */ 34) # define BN_num_bytes(a) ((BN_num_bits(a)+7)/8) diff --git a/doc/crypto/BN_generate_prime.pod b/doc/crypto/BN_generate_prime.pod index bf1b5308ad..0079f17a57 100644 --- a/doc/crypto/BN_generate_prime.pod +++ b/doc/crypto/BN_generate_prime.pod @@ -90,7 +90,17 @@ If B, this test is skipped. Both BN_is_prime_ex() and BN_is_prime_fasttest_ex() perform a Miller-Rabin probabilistic primality test with B iterations. If B, a number of iterations is used that -yields a false positive rate of at most 2^-80 for random input. +yields a false positive rate of at most 2^-64 for random input. +The error rate depends on the size of the prime and goes down for bigger primes. +The rate is 2^-80 starting at 308 bits, 2^-112 at 852 bit, 2^-128 at 1080 bits, +2^-192 at 3747 bit and 2^-256 at 6394 bit. + +When the source of the prime is not random or not trusted, the number +of checks needs to be much higher to reach the same level of assurance: +It should equal half of the targeted security level in bits (rounded up to the +next integer if necessary). +For instance, to reach the 128 bit security level, B should be set to +64. If B is not B, B is called after the j-th iteration (j = 0, 1, ...). B is a -- 2.25.1