io/n2rng/n2rng_provider.c

/*
 * CDDL HEADER START
 *
 * The contents of this file are subject to the terms of the
 * Common Development and Distribution License (the "License").
 * You may not use this file except in compliance with the License.
 *
 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
 * or http://www.opensolaris.org/os/licensing.
 * See the License for the specific language governing permissions
 * and limitations under the License.
 *
 * When distributing Covered Code, include this CDDL HEADER in each
 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
 * If applicable, add the following below this CDDL HEADER, with the
 * fields enclosed by brackets "[]" replaced with your own identifying
 * information: Portions Copyright [yyyy] [name of copyright owner]
 *
 * CDDL HEADER END
 */
/*
 * Copyright 2007 Sun Microsystems, Inc.  All rights reserved.
 * Use is subject to license terms.
 */

#pragma ident	"%Z%%M%	%I%	%E% SMI"

#include <sys/types.h>
#include <sys/sysmacros.h>
#include <sys/modctl.h>
#include <sys/conf.h>
#include <sys/devops.h>
#include <sys/cmn_err.h>
#include <sys/kmem.h>
#include <sys/stat.h>
#include <sys/open.h>
#include <sys/file.h>
#include <sys/cpuvar.h>
#include <sys/disp.h>
#include <sys/hsvc.h>
#include <sys/machsystm.h>
#include <sys/ksynch.h>
#include <sys/hypervisor_api.h>
#include <sys/n2rng.h>
#include <sys/sha1.h>
#include <sys/ddi.h>  /* near end to get min and max macros right */
#include <sys/sunddi.h>

/* n must be a power of 2 */
#define	ROUNDUP(k, n)		(((k) + (n) - 1) & ~((n) - 1))
#define	SHA1BLOCKBITS		512
#define	SHA1BLOCKBYTES		(SHA1BLOCKBITS / 8)
#define	SHA1WORDS		5
#define	SHA1BYTES		(4 * SHA1WORDS)


/*
 * Policy.  ENTROPY_STARVATION is the maximum number of calls each
 * FIPS instance will accept without successfully getting more
 * entropy.  It needs to be large enough to allow RNG operations to
 * not stall because of health checks, etc.  But we don't want it too
 * large.  FIPS 186-2 change 1 (5 October 2001) states that no more
 * that 2,000,000 DSA signatures (done using this algorithm) should be
 * done without reseeding.  We make sure we add 64 bits of entropy at
 * most every 10000 operations, hence we will have stirred in 160 bits
 * of entropy at most once every 30000 operations.  Normally, we stir
 * in 64 bits of entropy for every number generated.
 */
#define	ENTROPY_STARVATION	10000ULL

extern int	n2rng_herr2kerr(uint64_t);


/*
 * Adds val1 and val2 and stores result into sum.  The various input
 * pointers can be exactly aliased.  (They cannot be offset and
 * overlapping, but no one would ever do that.)  Values are big endian
 * by words and native byte order within words.  The return value's
 * 2-bit is 0 if the result is zero, it's 1 bit is carry out.  (This
 * is reused code.  The return code is not used by n2rng.)  Thus,
 * calling with both carryin and complement_val2 ones does a
 * subtraction.  A null sum pointer parameter is allowed.  The
 * subtraction features were required when this code was orginally
 * written so it could do a mod q operation.
 */
static int
add160(uint32_t *sum, uint32_t const *val1, uint32_t const *val2,
    const unsigned carryin, const int complement_val2)
{
	int i;
	uint32_t partialsum;
	uint32_t carry = (carryin > 0);
	uint32_t non_zero = 0;

	for (i = 4; i >= 0; --i) {
		partialsum = val1[i] + (complement_val2 ? ~val2[i] : val2[i]) +
		    carry;
		if (carry) {
			carry = (partialsum <= val1[i]);
		} else {
			carry = (partialsum < val1[i]);
		}
		if (sum) {
			sum[i] = partialsum;
		}
		non_zero |= partialsum;
	}

	return (((non_zero != 0) * 2) | carry);
}


/*
 * Computes a new random value, which is stored in x_j; updates XKEY
 * in the *rs.  XSEED_j is additional input.  In principle, we should
 * protect XKEY, perhaps by putting it on a non-pagable page, but we
 * aways clobber XKEY with fresh entropy just before we use it.  And
 * step 3d irreversibly updates it just after we use it.  The only
 * risk is that if an attacker captured the state while the entropy
 * generator was broken, the attacker could predict future values.
 * There are two cases: 1.  The attack gets root access to a live
 * system.  But there is no defense against that.  2.  The attacker
 * gets access to a crash dump.  But by then no values are being
 * generated.
 *
 * Note that XSEEDj is overwritten with sensitive stuff, and must be
 * zeroed by the caller.  We use two separate symbols (XVAL and
 * XSEEDj) to make each step match the notation in FIPS 186-2.
 */
static void
fips_random_inner(fipsrandomstruct_t *frsp, uint32_t *x_j,
    uint32_t *XSEED_j)
{
	int		i;
	SHA1_CTX	sha1_context;
	/* Alias to preserve terminology from FIPS 186-2 */
#define	XVAL XSEED_j
	/*
	 * K&R section A8.7: If the array has fixed size, the number
	 * of initializers may not exceed the number of members in the
	 * array; if there are fewer, the trailing members are
	 * initialized with 0.
	 */
	static const char	zero[SHA1BLOCKBYTES - SHA1BYTES] = {0};

	/*
	 * Step 3b: XVAL = (XKEY + XSEED_sub_j) mod 2^b.  The mod is
	 * implicit in the 160 bit representation.  Note that XVAL and
	 * XSEED_j are actually the same location.
	 */
	(void) add160(XVAL, frsp->XKEY, XSEED_j, 0, 0);
	/*
	 * Step 3c: x_sub_j = G(t, XVAL) mod q.
	 */
	SHA1Init(&sha1_context);
	SHA1Update(&sha1_context, (unsigned char *)XVAL, SHA1BYTES);
	/*
	 * Filling to 64 bytes is requried by FIPS 186-2 Appendix 3.3.
	 * It also triggers SHA1Transform (the steps a-e of the spec).
	 *
	 * zero is a const char[], but SHA1update does not declare its
	 * second parameter const, even though it does not modify it,
	 * so we cast to suppress a compiler warning.
	 */
	SHA1Update(&sha1_context, (unsigned char *)zero,
	    SHA1BLOCKBYTES - SHA1BYTES);
	/*
	 * The code below directly accesses the state field of
	 * sha1_context, which is of type SHA1_CTX, defined in sha1.h.
	 * This has been deemed acceptable, because that typedef is
	 * Consolidation Private, and n2rng is in the same
	 * consolidation.
	 */
	/* copy out to x_j */
	for (i = 0; i < 5; i++) {
		x_j[i] = sha1_context.state[i];
	}
	/*
	 * Step 3d: XKEY = (1 + XKEY + x_sub_j) mod 2^b.  b=160.  The
	 * mod 2^160 is implicit in the 160 bit representation.  The
	 * one is added via the carry-in flag.
	 */
	(void) add160(frsp->XKEY, frsp->XKEY, x_j, 1, 0);
#undef XVAL
}

int
fips_random(n2rng_t *n2rng, uint8_t *out, size_t nbytes)
{
	int			i;
	fipsrandomstruct_t	*frsp;
	union {
		uint32_t	as32[SHA1WORDS];
		uint64_t	as64[ROUNDUP(SHA1WORDS, 2) >> 1];
	} entropy = {0};
	uint32_t		tempout[SHA1WORDS];


	for (i = 0; i < nbytes; i += SHA1BYTES) {
		/*
		 * Since in the new scheme of things, the RNG latency
		 * will be high on reads after the first, we get just
		 * one word of entropy per call.  And if it fails, we
		 * just go on, but if the number of successive
		 * failures gets too big, we fail.
		 */
		if (n2rng_getentropy(n2rng, (void *)&entropy.as64[1],
		    sizeof (uint64_t))) {
			/* failure case */
			entropy.as64[1] = 0;
		}

		/*
		 * The idea here is that a Niagara2 chip is highly
		 * parallel, with many strands.  If we have just one
		 * instance of the FIPS data, then only one FIPS
		 * computation can happen at a time, serializeing all
		 * the RNG stuff.  So we make N2RNG_FIPS_INSTANCES,
		 * and use them round-robin, with the counter being
		 * n2rng->n_frs.fips_round_robin_j.  We increment the
		 * counter with an atomic op, avoiding having to have
		 * a global muxtex.  The atomic ops are also
		 * significantly faster than mutexes.  The mutex is
		 * put inside the loop, otherwise one thread reading
		 * many blocks could stall all other strands.
		 */
		frsp = &n2rng->n_frs.fipsarray[
		    atomic_inc_32_nv(&n2rng->n_frs.fips_round_robin_j) %
		    N2RNG_FIPS_INSTANCES];

		mutex_enter(&frsp->mtx);

		if (entropy.as64[1] == 0) {
			/*
			 * If we did not get any entropy, entropyword
			 * is zero.  We get a false positive with
			 * probablitity 2^-64.  It's not worth a few
			 * extra stores and tests eliminate the false
			 * positive.
			 */
			if (++frsp->entropyhunger > ENTROPY_STARVATION) {
				mutex_exit(&frsp->mtx);
				cmn_err(CE_WARN, "n2rng: not generating "
				    "entropy");
				return (EIO);
			}
		} else {
			frsp->entropyhunger = 0;
		}

		/* nbytes - i is bytes to go */
		fips_random_inner(frsp, tempout, entropy.as32);
		bcopy(tempout, &out[i], min(nbytes - i,  SHA1BYTES));

		mutex_exit(&frsp->mtx);
	}

	/* Zeroize sensitive information */

	entropy.as64[1] = 0;
	bzero(tempout, SHA1BYTES);

	return (0);
}

/*
 * Initializes one FIPS RNG instance.  Must be called once for each
 * instance.
 */
int
n2rng_fips_random_init(n2rng_t *n2rng, fipsrandomstruct_t *frsp)
{
	/*
	 * All FIPS-approved algorithms will operate as cryptograpic
	 * quality PRNGs even if there is no entropy source.  (In
	 * fact, this the only one that accepts entropy on the fly.)
	 * One motivation for this is that they system keeps on
	 * delivering cryptographic quality random numbers, even if
	 * the entropy source fails.
	 */

	int rv;

	rv = n2rng_getentropy(n2rng, (void *)frsp->XKEY, ROUNDUP(SHA1BYTES, 8));
	if (rv) {
		return (rv);
	}

	mutex_init(&frsp->mtx, NULL, MUTEX_DRIVER, NULL);

	return (0);
}

void
n2rng_fips_random_fini(fipsrandomstruct_t *frsp)
{
	mutex_destroy(&frsp->mtx);
	/*
	 * Zeroise fips data.  Not really necessary, since the
	 * algorithm has backtracking resistance, but do it anyway.
	 */
	bzero(frsp, sizeof (fipsrandomstruct_t));
}