/* This file is part of libbrandt.
 * Copyright (C) 2016 GNUnet e.V.
 *
 * libbrandt is free software: you can redistribute it and/or modify it under
 * the terms of the GNU General Public License as published by the Free Software
 * Foundation, either version 3 of the License, or (at your option) any later
 * version.
 *
 * libbrandt is distributed in the hope that it will be useful, but WITHOUT ANY
 * WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
 * A PARTICULAR PURPOSE.  See the GNU General Public License for more details.
 *
 * You should have received a copy of the GNU General Public License along with
 * libbrandt.  If not, see .
 */
/**
 * @file crypto.c
 * @brief Implementation of the crypto primitives.
 * @author Markus Teich
 */
#include "platform.h"
#include 
#include "crypto.h"
#include "internals.h"
#include "util.h"
#define CURVE "Ed25519"
struct zkp_challenge_dl {
	struct ec_mpi g;
	struct ec_mpi v;
	struct ec_mpi a;
};
struct zkp_challenge_2dle {
	struct ec_mpi g1;
	struct ec_mpi g2;
	struct ec_mpi v;
	struct ec_mpi w;
	struct ec_mpi a;
	struct ec_mpi b;
};
struct zkp_challenge_0og {
	struct ec_mpi g;
	struct ec_mpi alpha;
	struct ec_mpi beta;
	struct ec_mpi a1;
	struct ec_mpi a2;
	struct ec_mpi b1;
	struct ec_mpi b2;
};
static gcry_ctx_t                          ec_ctx;
static gcry_mpi_point_t                    ec_gen;
static gcry_mpi_point_t                    ec_zero;
static gcry_mpi_t                          ec_n;
static struct GNUNET_CRYPTO_EccDlogContext *ec_dlogctx;
/**
 * brandt_crypto_init initializes the crypto system and must be called before
 * any other function from this file.
 *
 * @param[in] dlogctx Pointer to the prepared dlog context.
 */
void
brandt_crypto_init (struct GNUNET_CRYPTO_EccDlogContext *dlogctx)
{
	gcry_error_t rc;
	ec_dlogctx = dlogctx;
	rc = gcry_mpi_ec_new (&ec_ctx, NULL, CURVE);
	brandt_assert_gpgerr (rc);
	ec_gen = gcry_mpi_ec_get_point ("g", ec_ctx, 0);
	brandt_assert (NULL != ec_gen);
	ec_zero = gcry_mpi_point_new (0);
	brandt_assert (NULL != ec_zero);
	gcry_mpi_ec_sub (ec_zero, ec_gen, ec_gen, ec_ctx);
	ec_n = gcry_mpi_ec_get_mpi ("n", ec_ctx, 1);
	brandt_assert (NULL != ec_n);
}
/* --- EC --- */
/**
 * ec_skey_create
 *
 * @param[out] skey where to store the generated secret key. This has to be an
 * already initialized mpi.
 */
void
ec_skey_create (gcry_mpi_t skey)
{
	gcry_mpi_t   ret;
	gcry_sexp_t  s_keyparam;
	gcry_sexp_t  priv_sexp;
	gcry_sexp_t  priv_key;
	gcry_sexp_t  priv_key2;
	gcry_error_t rc;
	rc = gcry_sexp_build (&s_keyparam, NULL, "(genkey(ecc(curve \"" CURVE "\")"
	                      "(flags)))");
	brandt_assert_gpgerr (rc);
	rc = gcry_pk_genkey (&priv_sexp, s_keyparam);
	brandt_assert_gpgerr (rc);
	gcry_sexp_release (s_keyparam);
	priv_key = gcry_sexp_find_token (priv_sexp, "private-key", 11);
	brandt_assert (NULL != priv_key);
	gcry_sexp_release (priv_sexp);
	priv_key2 = gcry_sexp_find_token (priv_key, "d", 1);
	brandt_assert (NULL != priv_key2);
	gcry_sexp_release (priv_key);
	ret = gcry_sexp_nth_mpi (priv_key2, 1, GCRYMPI_FMT_USG);
	brandt_assert (NULL != ret);
	gcry_sexp_release (priv_key2);
	gcry_mpi_snatch (skey, ret);
}
/**
 * ec_keypair_create creates a new keypair by creating a random secret key first
 * and multipyling the base point with it to get the public key.
 *
 * @param[out] pkey where to store the generated public key
 * @param[out] skey where to store the generated secret key. May be NULL if
 * you're not interested in the secret key and just need a random point.
 */
void
ec_keypair_create (gcry_mpi_point_t pkey, gcry_mpi_t skey)
{
	gcry_mpi_t sk;
	brandt_assert (NULL != pkey);
	sk = (NULL == skey) ? gcry_mpi_new (256) : skey;
	ec_skey_create (sk);
	gcry_mpi_ec_mul (pkey, sk, ec_gen, ec_ctx);
	if (NULL == skey)
		gcry_mpi_release (sk);
}
/**
 * ec_keypair_create_base
 *
 * @param[out] pkey where to store the generated public key
 * @param[out] skey where to store the generated secret key
 * @param[in] base which base point should be used to calculate the public key
 */
void
ec_keypair_create_base (gcry_mpi_point_t       pkey,
                        gcry_mpi_t             skey,
                        const gcry_mpi_point_t base)
{
	brandt_assert (NULL != pkey);
	brandt_assert (NULL != skey);
	brandt_assert (NULL != base);
	ec_skey_create (skey);
	gcry_mpi_ec_mul (pkey, skey, base, ec_ctx);
}
/**
 * ec_point_copy creates a copy of one curve point
 *
 * @param[out] dst where to store the copy
 * @param[in] src the input point to be copied
 */
void
ec_point_copy (gcry_mpi_point_t dst, const gcry_mpi_point_t src)
{
	gcry_mpi_t x = gcry_mpi_new (256);
	gcry_mpi_t y = gcry_mpi_new (256);
	gcry_mpi_t z = gcry_mpi_new (256);
	brandt_assert (dst && src);
	gcry_mpi_point_get (x, y, z, src);
	gcry_mpi_point_snatch_set (dst, x, y, z);
}
/**
 * ec_point_cmp compares two curve points
 *
 * @param[in] a the first point
 * @param[in] b the second point
 * @return 0 if @a a and @a b represent the same point on the curve, something
 * else otherwise
 */
int
ec_point_cmp (const gcry_mpi_point_t a, const gcry_mpi_point_t b)
{
	int        ret = 1;
	gcry_mpi_t ax = gcry_mpi_new (256);
	gcry_mpi_t bx = gcry_mpi_new (256);
	gcry_mpi_t ay = gcry_mpi_new (256);
	gcry_mpi_t by = gcry_mpi_new (256);
	brandt_assert (a && b);
	if (!ax || !bx || !ay || !by)
	{
		weprintf ("could not init point in point_cmp");
		return 1;
	}
	if (!gcry_mpi_ec_get_affine (ax, ay, a, ec_ctx) &&
	    !gcry_mpi_ec_get_affine (bx, by, b, ec_ctx))
	{
		ret = gcry_mpi_cmp (ax, bx) || gcry_mpi_cmp (ay, by);
	}
	gcry_mpi_release (ax);
	gcry_mpi_release (bx);
	gcry_mpi_release (ay);
	gcry_mpi_release (by);
	return ret;
}
/**
 * mpi_serialize outputs the given MPI value to the given destination buffer in
 * network byte order. The MPI @a src may not be negative.
 *
 * @param[out] dst where to output to
 * @param[in] src value to write to @a dst
 */
void
mpi_serialize (struct ec_mpi *dst, gcry_mpi_t src)
{
	size_t rsize = 0;
	if (gcry_mpi_get_flag (src, GCRYMPI_FLAG_OPAQUE))
	{   /* Store opaque MPIs left aligned. Used by Ed25519 point compression */
		unsigned int nbits;
		const void   *vp = gcry_mpi_get_opaque (src, &nbits);
		brandt_assert (vp);
		rsize = (nbits + 7) / 8;
		if (rsize > sizeof (struct ec_mpi))
			rsize = sizeof (struct ec_mpi);
		memcpy (dst, vp, rsize);
		if (rsize < sizeof (struct ec_mpi))
			memset (((char *)dst) + rsize, 0, sizeof (struct ec_mpi) - rsize);
	}
	else
	{   /* Store regular MPIs as unsigned ints right aligned into the buffer. */
		char         *cp = (char *)dst;
		gcry_error_t rc;
		rc = gcry_mpi_print (GCRYMPI_FMT_USG, (void *)dst,
		                     sizeof (struct ec_mpi), &rsize, src);
		brandt_assert_gpgerr (rc);
		/* Shift the output to the right, if shorter than available space */
		if (rsize && rsize < sizeof (struct ec_mpi))
		{
			memmove (&cp[sizeof (struct ec_mpi) - rsize], dst, rsize);
			memset (dst, 0, sizeof (struct ec_mpi) - rsize);
		}
	}
}
/**
 * mpi_parse converts src buffer into MPI value.
 * The buffer is interpreted as network byte order, unsigned integer.
 *
 * @param[out] dst where to store MPI value. Must be initialized.
 * @param[in] src raw data source (GCRYMPI_FMT_USG)
 */
void
mpi_parse (gcry_mpi_t dst, const struct ec_mpi *src)
{
	gcry_mpi_t   ret;
	gcry_error_t rc;
	rc = gcry_mpi_scan (&ret,
	                    GCRYMPI_FMT_USG,
	                    src,
	                    sizeof (struct ec_mpi),
	                    NULL);
	brandt_assert_gpgerr (rc);
	gcry_mpi_snatch (dst, ret);
}
/**
 * ec_point_serialize outputs the given curve point to the @a dst buffer.
 *
 * @param[out] dst where to write the raw data to
 * @param[in] src curve point to write to @a dst
 */
void
ec_point_serialize (struct ec_mpi *dst, const gcry_mpi_point_t src)
{
	gcry_sexp_t  s;
	gcry_ctx_t   ctx;
	gcry_error_t rc;
	gcry_mpi_t   q;
	brandt_assert (dst);
	rc = gcry_sexp_build (&s, NULL, "(public-key(ecc(curve " CURVE ")))");
	brandt_assert_gpgerr (rc);
	brandt_assert (NULL != s);
	rc = gcry_mpi_ec_new (&ctx, s, NULL);
	brandt_assert_gpgerr (rc);
	gcry_sexp_release (s);
	rc = gcry_mpi_ec_set_point ("q", src, ctx);
	brandt_assert_gpgerr (rc);
	q = gcry_mpi_ec_get_mpi ("q@eddsa", ctx, 0);
	brandt_assert (NULL != q);
	gcry_ctx_release (ctx);
	mpi_serialize (dst, q);
	gcry_mpi_release (q);
}
/**
 * ec_point_parse parses a point on the Ed25519 curve from @a src into @a dst.
 *
 * @param[out] dst where to store the curve point. Must be initialized
 * @param[in] src raw data source
 */
void
ec_point_parse (gcry_mpi_point_t dst, const struct ec_mpi *src)
{
	gcry_sexp_t      s;
	gcry_ctx_t       ctx;
	gcry_mpi_point_t ret;
	gcry_error_t     rc;
	rc = gcry_sexp_build (&s, NULL, "(public-key(ecc(curve " CURVE ")(q %b)))",
	                      sizeof (struct ec_mpi), src);
	brandt_assert_gpgerr (rc);
	rc = gcry_mpi_ec_new (&ctx, s, NULL);
	brandt_assert_gpgerr (rc);
	gcry_sexp_release (s);
	ret = gcry_mpi_ec_get_point ("q", ctx, 0);
	brandt_assert (ret);
	gcry_ctx_release (ctx);
	gcry_mpi_ec_mul (dst, GCRYMPI_CONST_ONE, ret, ec_ctx);
	gcry_mpi_point_release (ret);
}
/**
 * smc_free1 releases all points in @a dst and frees the memory
 *
 * @param[in,out] dst The 1 dimensional array to clean up
 * @param[in] size1 size of the first dimension
 */
void
smc_free1 (gcry_mpi_point_t *dst, uint16_t size1)
{
	if (NULL == dst)
		return;
	for (uint16_t i = 0; i < size1; i++)
		if (NULL != dst[i])
			gcry_mpi_point_release (dst[i]);
	free (dst);
}
/**
 * smc_init1 creates a 1 dimensional array of curve points. Make sure to
 * initialize the values before using them, they are not automatically set to
 * the zero point!
 *
 * @param[in] size1 size of the first dimension
 * @return a pointer to the array or NULL on error.
 * If not used anymore use smc_free2 to reclaim the memory.
 */
gcry_mpi_point_t *
smc_init1 (uint16_t size1)
{
	gcry_mpi_point_t *ret;
	ret = GNUNET_new_array (size1, gcry_mpi_point_t);
	for (uint16_t i = 0; i < size1; i++)
	{
		if (NULL == (ret[i] = gcry_mpi_point_new (0)))
		{
			weprintf ("could not init point in 1 dimensional array. "
			          "out of memory?");
			smc_free1 (ret, size1);
			return NULL;
		}
	}
	return ret;
}
/**
 * smc_free2 releases all points in @a dst and frees the memory
 *
 * @param[in,out] dst The 2 dimensional array to clean up
 * @param[in] size1 size of the first dimension
 * @param[in] size2 size of the second dimension
 */
void
smc_free2 (gcry_mpi_point_t **dst, uint16_t size1, uint16_t size2)
{
	if (NULL == dst)
		return;
	for (uint16_t i = 0; i < size1; i++)
		for (uint16_t j = 0; j < size2; j++)
			if (NULL != dst[i][j])
				gcry_mpi_point_release (dst[i][j]);
	free (dst);
}
/**
 * smc_init2 creates a 2 dimensional array of curve points. Make sure to
 * initialize the values before using them, they are not automatically set to
 * the zero point!
 *
 * @param[in] size1 size of the first dimension
 * @param[in] size2 size of the second dimension
 * @return a pointer to the array or NULL on error.
 * If not used anymore use smc_free2 to reclaim the memory.
 */
gcry_mpi_point_t **
smc_init2 (uint16_t size1, uint16_t size2)
{
	gcry_mpi_point_t **ret;
	gcry_mpi_point_t *data;
	if (NULL == (ret = calloc (size1, sizeof (*ret) + size2 * sizeof (**ret))))
	{
		weprintf ("could not alloc memory for 2 dimensional point array");
		return NULL;
	}
	data = (gcry_mpi_point_t *)&ret[size1];
	for (uint16_t i = 0; i < size1; i++)
	{
		ret[i] = &data[i * size2];
		for (uint16_t j = 0; j < size2; j++)
		{
			if (NULL == (ret[i][j] = gcry_mpi_point_new (0)))
			{
				weprintf ("could not init point in 2 dimensional array. "
				          "out of memory?");
				smc_free2 (ret, size1, size2);
				return NULL;
			}
		}
	}
	return ret;
}
/**
 * smc_free3 releases all points in @a dst and frees the memory
 *
 * @param[in,out] dst The 3 dimensional array to clean up
 * @param[in] size1 size of the first dimension
 * @param[in] size2 size of the second dimension
 * @param[in] size3 size of the third dimension
 */
void
smc_free3 (gcry_mpi_point_t ***dst,
           uint16_t         size1,
           uint16_t         size2,
           uint16_t         size3)
{
	if (NULL == dst)
		return;
	for (uint16_t i = 0; i < size1; i++)
		for (uint16_t j = 0; j < size2; j++)
			for (uint16_t k = 0; k < size3; k++)
				if (NULL != dst[i][j][k])
					gcry_mpi_point_release (dst[i][j][k]);
	free (dst);
}
/**
 * smc_init3 creates a 3 dimensional array of curve points. Make sure to
 * initialize the values before using them, they are not automatically set to
 * the zero point!
 *
 * @param[in] size1 size of the first dimension
 * @param[in] size2 size of the second dimension
 * @param[in] size3 size of the third dimension
 * @return a pointer to the array or NULL on error.
 * If not used anymore use smc_free3 to reclaim the memory.
 */
gcry_mpi_point_t ***
smc_init3 (uint16_t size1, uint16_t size2, uint16_t size3)
{
	gcry_mpi_point_t ***ret;
	gcry_mpi_point_t **layer1;
	gcry_mpi_point_t *layer2;
	if (NULL == (ret = calloc (size1, sizeof (*ret) +
	                           size2 * sizeof (**ret) +
	                           size2 * size3 * sizeof (***ret))))
	{
		weprintf ("could not alloc memory for 3 dimensional point array");
		return NULL;
	}
	layer1 = (gcry_mpi_point_t **)&ret[size1];
	layer2 = (gcry_mpi_point_t *)&layer1[size1 * size2];
	for (uint16_t i = 0; i < size1; i++)
	{
		ret[i] = &layer1[i * size2];
		for (uint16_t j = 0; j < size2; j++)
		{
			layer1[i * size2 + j] = &layer2[(i * size2 + j) * size3];
			for (uint16_t k = 0; k < size3; k++)
			{
				if (NULL == (ret[i][j][k] = gcry_mpi_point_new (0)))
				{
					weprintf ("could not init point in 2 dimensional array. "
					          "out of memory?");
					smc_free3 (ret, size1, size2, size3);
					return NULL;
				}
			}
		}
	}
	return ret;
}
/**
 * smc_sums_partial calculates sums up until the current index and stores them
 * in @a out. \f$\forall i \leq len: out_i=\sum_{h=1}^iin_h\f$
 *
 * @param[out] out Where to store the resulting sums. Points must already be
 * initialized beforehand.
 * @param[in] in Input points.
 * @param[in] len The length of @a out. @a in must be at least @a step times @a
 * len elements long.
 * @param[in] stepi The amount of items to advance in @a in each step. Can be
 * used to sum over multi-dimensional arrays.
 * @param[in] stepo The amount of items to advance in @a out each step. Can be
 * used to store the sum in multi-dimensional arrays.
 */
static void
smc_sums_partial (gcry_mpi_point_t out[],
                  gcry_mpi_point_t in[],
                  uint16_t         len,
                  uint16_t         stepi,
                  uint16_t         stepo)
{
	brandt_assert (NULL != out);
	for (uint16_t i = 0, o = 0; o < len * stepo; i += stepi, o += stepo)
		gcry_mpi_ec_add (out[o], (o ? out[o - stepo] : ec_zero), in[i], ec_ctx);
}
/**
 * smc_sum calculates the sum of all input points.
 * \f$out=\sum_{i=1}^{len}in_i\f$
 *
 * @param[out] out Where to store the result
 * @param[in] in Input points.
 * @param[in] len The amount of summands to use from @a in. @a in must be at
 * least @a step times @a len elements long.
 * @param[in] step The amount of items to advance in @a in each step. Can be
 * used to sum over multi-dimensional arrays.
 */
static void
smc_sum (gcry_mpi_point_t out,
         gcry_mpi_point_t in[],
         uint16_t         len,
         uint16_t         step)
{
	brandt_assert (NULL != out);
	ec_point_copy (out, ec_zero);
	for (uint16_t i = 0; i < len * step; i += step)
		gcry_mpi_ec_add (out, out, in[i], ec_ctx);
}
/**
 * smc_gen_keyshare creates the private keyshare and computes the
 * public key share
 *
 * @param[in] ad Pointer to the BRANDT_Auction struct to operate on
 * @param[out] buflen Size of the returned message buffer in bytes
 * @return A buffer containing the public key share
 * which needs to be broadcast
 */
unsigned char *
smc_gen_keyshare (struct BRANDT_Auction *ad, size_t *buflen)
{
	unsigned char   *ret;
	struct proof_dl *proof1;
	struct msg_head *head;
	struct ec_mpi   *pubkey_share;
	brandt_assert (ad && buflen);
	*buflen = (sizeof (*head) +
	           sizeof (*pubkey_share) +
	           sizeof (*proof1));
	ret = GNUNET_new_array (*buflen, unsigned char);
	if (NULL == (ad->y = smc_init1 (ad->n)))
	{
		weprintf ("unable to alloc memory for key shares");
		return NULL;
	}
	head = (struct msg_head *)ret;
	head->prot_version = htonl (0);
	head->msg_type = htonl (msg_init);
	proof1 = (struct proof_dl *)(ret + sizeof (*head) + sizeof (*pubkey_share));
	ad->x = gcry_mpi_new (256);
	ec_skey_create (ad->x);
	smc_zkp_dl (ad->y[ad->i], ad->x, proof1);
	pubkey_share = (struct ec_mpi *)(ret + sizeof (*head));
	ec_point_serialize (pubkey_share, ad->y[ad->i]);
	return ret;
}
int
smc_recv_keyshare (struct BRANDT_Auction *ad,
                   const unsigned char   *buf,
                   size_t                buflen,
                   uint16_t              sender)
{
	int              ret = 0;
	struct proof_dl  *proof1;
	gcry_mpi_point_t y = gcry_mpi_point_new (0);
	brandt_assert (ad && buf);
	if (buflen != (sizeof (struct ec_mpi) + sizeof (*proof1)))
	{
		weprintf ("wrong size of received key share");
		goto quit;
	}
	proof1 = (struct proof_dl *)(buf + sizeof (struct ec_mpi));
	ec_point_parse (y, (struct ec_mpi *)buf);
	if (smc_zkp_dl_check (y, proof1))
	{
		weprintf ("wrong zkp1 for public key share received");
		goto quit;
	}
	ec_point_copy (ad->y[sender], y);
	ret = 1;
quit:
	gcry_mpi_point_release (y);
	return ret;
}
/**
 * smc_encrypt_bid encrypts the own bid with the shared public key and packs it
 * into a message together with proofs of correctnes.
 *
 * @param[in] ad Pointer to the BRANDT_Auction struct to operate on
 * @param[out] buflen Size of the returned message buffer in bytes
 * @return A buffer containing the encrypted bid
 * which needs to be broadcast
 */
unsigned char *
smc_encrypt_bid (struct BRANDT_Auction *ad, size_t *buflen)
{
	unsigned char    *ret;
	unsigned char    *cur;
	struct proof_0og *proof3;
	struct msg_head  *head;
	gcry_mpi_t       r_sum;
	gcry_mpi_t       r_part;
	brandt_assert (ad && buflen);
	*buflen = (sizeof (*head) +              /* msg header */
	           ad->k *                       /* k * (alpha, beta, proof3) */
	           (sizeof (struct ec_mpi) * 2 + /* alpha, beta */
	            sizeof (*proof3)) +
	           sizeof (struct proof_2dle));
	ret = GNUNET_new_array (*buflen, unsigned char);
	if (NULL == (ad->alpha = smc_init2 (ad->n, ad->k)) ||
	    NULL == (ad->beta = smc_init2 (ad->n, ad->k)))
	{
		weprintf ("unable to alloc memory for encrypted bids");
		return NULL;
	}
	head = (struct msg_head *)ret;
	head->prot_version = htonl (0);
	head->msg_type = htonl (msg_bid);
	cur = ret + sizeof (*head);
	ad->Y = gcry_mpi_point_new (0);
	smc_sum (ad->Y, ad->y, ad->n, 1);
	r_sum = gcry_mpi_new (256);
	r_part = gcry_mpi_new (256);
	for (uint16_t j = 0; j < ad->k; j++)
	{
		proof3 = (struct proof_0og *)(cur + 2 * sizeof (struct ec_mpi));
		smc_zkp_0og (j == ad->b,
		             ad->Y,
		             r_part,
		             ad->alpha[ad->i][j],
		             ad->beta[ad->i][j],
		             proof3);
		ec_point_serialize ((struct ec_mpi *)cur, ad->alpha[ad->i][j]);
		ec_point_serialize (&((struct ec_mpi *)cur)[1], ad->beta[ad->i][j]);
		gcry_mpi_addm (r_sum, r_sum, r_part, ec_n);
		cur += 2 * sizeof (struct ec_mpi) + sizeof (struct proof_0og);
	}
	smc_zkp_2dle (NULL, NULL, ad->Y, ec_gen, r_sum, (struct proof_2dle *)cur);
	gcry_mpi_release (r_sum);
	gcry_mpi_release (r_part);
	return ret;
}
int
smc_recv_encrypted_bid (struct BRANDT_Auction *ad,
                        const unsigned char   *buf,
                        size_t                buflen,
                        uint16_t              sender)
{
	int                 ret = 0;
	const unsigned char *cur = buf;
	struct proof_0og    *proof3;
	gcry_mpi_point_t    **ct; /* ciphertexts */
	gcry_mpi_point_t    alpha_sum = gcry_mpi_point_new (0);
	gcry_mpi_point_t    beta_sum = gcry_mpi_point_new (0);
	brandt_assert (ad && buf);
	if (buflen != (ad->k * (sizeof (struct ec_mpi) * 2 + sizeof (*proof3)) +
	               sizeof (struct proof_2dle)) ||
	    NULL == (ct = smc_init2 (2, ad->k)))
	{
		weprintf ("wrong size of received encrypted bid");
		goto quit;
	}
	ec_point_copy (alpha_sum, ec_zero);
	ec_point_copy (beta_sum, ec_zero);
	for (uint16_t j = 0; j < ad->k; j++)
	{
		ec_point_parse (ct[0][j], (struct ec_mpi *)cur);
		ec_point_parse (ct[1][j], &((struct ec_mpi *)cur)[1]);
		proof3 = (struct proof_0og *)(cur + 2 * sizeof (struct ec_mpi));
		if (smc_zkp_0og_check (ad->Y, ct[0][j], ct[1][j], proof3))
		{
			weprintf ("wrong zkp3 for alpha, beta received");
			goto quit;
		}
		gcry_mpi_ec_add (alpha_sum, alpha_sum, ct[0][j], ec_ctx);
		gcry_mpi_ec_add (beta_sum, beta_sum, ct[1][j], ec_ctx);
		cur += 2 * sizeof (struct ec_mpi) + sizeof (struct proof_0og);
	}
	gcry_mpi_ec_sub (alpha_sum, alpha_sum, ec_gen, ec_ctx);
	if (smc_zkp_2dle_check (alpha_sum,
	                        beta_sum,
	                        ad->Y,
	                        ec_gen,
	                        (struct proof_2dle *)cur))
	{
		weprintf ("wrong zkp2 for alpha, beta received");
		goto quit;
	}
	for (uint16_t j = 0; j < ad->k; j++)
	{
		ec_point_copy (ad->alpha[sender][j], ct[0][j]);
		ec_point_copy (ad->beta[sender][j], ct[1][j]);
	}
	smc_free2 (ct, 2, ad->k);
	ret = 1; /* finally success */
quit:
	gcry_mpi_point_release (alpha_sum);
	gcry_mpi_point_release (beta_sum);
	return ret;
}
/**
 * fp_pub_compute_outcome computes the outcome for first price auctions with a
 * public outcome and packs it into a message buffer together with proofs of
 * correctnes.
 *
 * @param[in] ad Pointer to the BRANDT_Auction struct to operate on
 * @param[out] buflen Size of the returned message buffer in bytes
 * @return A buffer containing the encrypted outcome vectors
 * which needs to be broadcast
 */
unsigned char *
fp_pub_compute_outcome (struct BRANDT_Auction *ad, size_t *buflen)
{
	unsigned char     *ret;
	unsigned char     *cur;
	struct msg_head   *head;
	gcry_mpi_t        coeff = gcry_mpi_copy (GCRYMPI_CONST_ONE);
	gcry_mpi_point_t  tmp = gcry_mpi_point_new (0);
	gcry_mpi_point_t  *tlta1;
	gcry_mpi_point_t  *tltb1;
	gcry_mpi_point_t  **tlta2;
	gcry_mpi_point_t  **tltb2;
	struct ec_mpi     *gamma;
	struct ec_mpi     *delta;
	struct proof_2dle *proof2;
	brandt_assert (ad && buflen);
	*buflen = (sizeof (*head) +
	           ad->k * (sizeof (*gamma) +
	                    sizeof (*delta) +
	                    sizeof (*proof2)));
	ret = GNUNET_new_array (*buflen, unsigned char);
	if (NULL == (ad->gamma2 = smc_init2 (ad->n, ad->k)) ||
	    NULL == (ad->delta2 = smc_init2 (ad->n, ad->k)) ||
	    NULL == (ad->tmpa1 = smc_init1 (ad->k)) ||
	    NULL == (ad->tmpb1 = smc_init1 (ad->k)))
	{
		weprintf ("unable to alloc memory for first price outcome computation");
		return NULL;
	}
	head = (struct msg_head *)ret;
	head->prot_version = htonl (0);
	head->msg_type = htonl (msg_outcome);
	cur = ret + sizeof (*head);
	/* create temporary lookup tables with partial sums */
	tlta1 = smc_init1 (ad->k);
	tltb1 = smc_init1 (ad->k);
	tlta2 = smc_init2 (ad->n, ad->k);
	tltb2 = smc_init2 (ad->n, ad->k);
	/* temporary lookup table for sum of bid vectors */
	for (uint16_t i = 0; i < ad->n; i++)
	{
		smc_sums_partial (tlta2[i], ad->alpha[i], ad->k, 1, 1);
		smc_sums_partial (tltb2[i], ad->beta[i], ad->k, 1, 1);
		for (uint16_t j = 0; j < ad->k; j++)
		{
			gcry_mpi_ec_sub (tlta2[i][j],
			                 tlta2[i][ad->k - 1],
			                 tlta2[i][j],
			                 ec_ctx);
			gcry_mpi_ec_sub (tltb2[i][j],
			                 tltb2[i][ad->k - 1],
			                 tltb2[i][j],
			                 ec_ctx);
		}
		brandt_assert (!ec_point_cmp (ec_zero, tlta2[i][ad->k - 1]));
		brandt_assert (!ec_point_cmp (ec_zero, tltb2[i][ad->k - 1]));
	}
	for (uint16_t j = 0; j < ad->k; j++)
	{
		smc_sum (tlta1[j], &tlta2[0][j], ad->n, ad->k);
		smc_sum (tltb1[j], &tltb2[0][j], ad->n, ad->k);
	}
	smc_free2 (tlta2, ad->n, ad->k);
	smc_free2 (tltb2, ad->n, ad->k);
	brandt_assert (!ec_point_cmp (ec_zero, tlta1[ad->k - 1]));
	brandt_assert (!ec_point_cmp (ec_zero, tltb1[ad->k - 1]));
	/* initialize tmp array with zeroes, since we are calculating a sum */
	for (uint16_t j = 0; j < ad->k; j++)
	{
		ec_point_copy (ad->tmpa1[j], ec_zero);
		ec_point_copy (ad->tmpb1[j], ec_zero);
	}
	/* store the \sum_{i=1}^n2^{i-1}b_i in tmp1 until outcome determination,
	 * since it is needed each time a gamma,delta pair is received from another
	 * bidder */
	for (uint16_t i = 0; i < ad->n; i++)
	{
		for (uint16_t j = 0; j < ad->k; j++)
		{
			gcry_mpi_ec_mul (tmp, coeff, ad->alpha[i][j], ec_ctx);
			gcry_mpi_ec_add (ad->tmpa1[j], ad->tmpa1[j], tmp, ec_ctx);
			gcry_mpi_ec_mul (tmp, coeff, ad->beta[i][j], ec_ctx);
			gcry_mpi_ec_add (ad->tmpb1[j], ad->tmpb1[j], tmp, ec_ctx);
		}
		gcry_mpi_lshift (coeff, coeff, 1);
	}
	for (uint16_t j = 0; j < ad->k; j++)
	{
		gamma = (struct ec_mpi *)cur;
		delta = &((struct ec_mpi *)cur)[1];
		proof2 = (struct proof_2dle *)(cur + 2 * sizeof (struct ec_mpi));
		/* copy unmasked outcome to all other bidder layers so they don't
		 * have to be recomputed to check the ZK proof_2dle's from other
		 * bidders when receiving their outcome messages */
		for (uint16_t a = 0; a < ad->n; a++)
		{
			ec_point_copy (ad->gamma2[a][j], tlta1[j]);
			ec_point_copy (ad->delta2[a][j], tltb1[j]);
		}
		/* apply random masking for losing bidders */
		smc_zkp_2dle (ad->gamma2[ad->i][j],
		              ad->delta2[ad->i][j],
		              tlta1[j],
		              tltb1[j],
		              NULL,
		              proof2);
		ec_point_serialize (gamma, ad->gamma2[ad->i][j]);
		ec_point_serialize (delta, ad->delta2[ad->i][j]);
		/* add winner determination for own gamma,delta */
		gcry_mpi_ec_add (ad->gamma2[ad->i][j],
		                 ad->gamma2[ad->i][j],
		                 ad->tmpa1[j],
		                 ec_ctx);
		gcry_mpi_ec_add (ad->delta2[ad->i][j],
		                 ad->delta2[ad->i][j],
		                 ad->tmpb1[j],
		                 ec_ctx);
		cur += sizeof (*gamma) + sizeof (*delta) + sizeof (*proof2);
	}
	gcry_mpi_release (coeff);
	gcry_mpi_point_release (tmp);
	smc_free1 (tlta1, ad->k);
	smc_free1 (tltb1, ad->k);
	return ret;
}
int
fp_pub_recv_outcome (struct BRANDT_Auction *ad,
                     const unsigned char   *buf,
                     size_t                buflen,
                     uint16_t              sender)
{
	int                 ret = 0;
	const unsigned char *cur = buf;
	struct proof_2dle   *proof2;
	gcry_mpi_point_t    gamma = gcry_mpi_point_new (0);
	gcry_mpi_point_t    delta = gcry_mpi_point_new (0);
	brandt_assert (ad && buf);
	if (buflen != (ad->k * (2 * sizeof (struct ec_mpi) + sizeof (*proof2))))
	{
		weprintf ("wrong size of received outcome");
		goto quit;
	}
	for (uint16_t j = 0; j < ad->k; j++)
	{
		ec_point_parse (gamma, (struct ec_mpi *)cur);
		ec_point_parse (delta, &((struct ec_mpi *)cur)[1]);
		proof2 = (struct proof_2dle *)(cur + 2 * sizeof (struct ec_mpi));
		if (smc_zkp_2dle_check (gamma,
		                        delta,
		                        ad->gamma2[sender][j],
		                        ad->delta2[sender][j],
		                        proof2))
		{
			weprintf ("wrong zkp2 for gamma, delta received");
			goto quit;
		}
		ec_point_copy (ad->gamma2[sender][j], gamma);
		ec_point_copy (ad->delta2[sender][j], delta);
		/* add winner determination summand */
		gcry_mpi_ec_add (ad->gamma2[sender][j],
		                 ad->gamma2[sender][j],
		                 ad->tmpa1[j],
		                 ec_ctx);
		gcry_mpi_ec_add (ad->delta2[sender][j],
		                 ad->delta2[sender][j],
		                 ad->tmpb1[j],
		                 ec_ctx);
		cur += 2 * sizeof (struct ec_mpi) + sizeof (*proof2);
	}
	ret = 1;
quit:
	gcry_mpi_point_release (gamma);
	gcry_mpi_point_release (delta);
	return ret;
}
/**
 * fp_pub_decrypt_outcome decrypts part of the outcome and packs it into a
 * message buffer together with proofs of correctnes.
 *
 * @param[in] ad Pointer to the BRANDT_Auction struct to operate on
 * @param[out] buflen Size of the returned message buffer in bytes
 * @return A buffer containing the own share of the decrypted outcome
 * which needs to be broadcast
 */
unsigned char *
fp_pub_decrypt_outcome (struct BRANDT_Auction *ad, size_t *buflen)
{
	unsigned char     *ret;
	unsigned char     *cur;
	struct msg_head   *head;
	gcry_mpi_point_t  tmp = gcry_mpi_point_new (0);
	struct ec_mpi     *phi;
	struct proof_2dle *proof2;
	brandt_assert (ad && buflen);
	*buflen = (sizeof (*head) + ad->k * (sizeof (*phi) + sizeof (*proof2)));
	ret = GNUNET_new_array (*buflen, unsigned char);
	if (NULL == (ad->phi2 = smc_init2 (ad->n, ad->k)))
	{
		weprintf ("unable to alloc memory for first price outcome decryption");
		return NULL;
	}
	head = (struct msg_head *)ret;
	head->prot_version = htonl (0);
	head->msg_type = htonl (msg_decrypt);
	cur = ret + sizeof (*head);
	for (uint16_t j = 0; j < ad->k; j++)
	{
		phi = (struct ec_mpi *)cur;
		proof2 = (struct proof_2dle *)(cur + sizeof (*phi));
		smc_sum (tmp, &ad->delta2[0][j], ad->n, ad->k);
		/* copy still encrypted outcome to all other bidder layers so they
		 * don't have to be recomputed to check the ZK proof_2dle's from
		 * other bidders when receiving their outcome decryption messages */
		for (uint16_t a = 0; a < ad->n; a++)
			ec_point_copy (ad->phi2[a][j], tmp);
		/* decrypt outcome component and prove the correct key was used */
		smc_zkp_2dle (ad->phi2[ad->i][j],
		              NULL,
		              tmp,
		              ec_gen,
		              ad->x,
		              proof2);
		ec_point_serialize (phi, ad->phi2[ad->i][j]);
		cur += sizeof (*phi) + sizeof (*proof2);
	}
	gcry_mpi_point_release (tmp);
	return ret;
}
int
fp_pub_recv_decryption (struct BRANDT_Auction *ad,
                        const unsigned char   *buf,
                        size_t                buflen,
                        uint16_t              sender)
{
	int                 ret = 0;
	const unsigned char *cur = buf;
	struct proof_2dle   *proof2;
	gcry_mpi_point_t    phi = gcry_mpi_point_new (0);
	brandt_assert (ad && buf);
	if (buflen != (ad->k * (sizeof (struct ec_mpi) + sizeof (*proof2))))
	{
		weprintf ("wrong size of received outcome decryption");
		goto quit;
	}
	for (uint16_t j = 0; j < ad->k; j++)
	{
		ec_point_parse (phi, (struct ec_mpi *)cur);
		proof2 = (struct proof_2dle *)(cur + sizeof (struct ec_mpi));
		if (smc_zkp_2dle_check (phi,
		                        ad->y[sender],
		                        ad->phi2[sender][j],
		                        ec_gen,
		                        proof2))
		{
			weprintf ("wrong zkp2 for phi, y received");
			goto quit;
		}
		ec_point_copy (ad->phi2[sender][j], phi);
		cur += sizeof (struct ec_mpi) + sizeof (*proof2);
	}
	ret = 1;
quit:
	gcry_mpi_point_release (phi);
	return ret;
}
int32_t
fp_pub_determine_outcome (struct BRANDT_Auction *ad, uint16_t *winner)
{
	int32_t          ret = -1;
	int              dlogi = -1;
	gcry_mpi_t       dlog = gcry_mpi_new (256);
	gcry_mpi_point_t sum_gamma = gcry_mpi_point_new (0);
	gcry_mpi_point_t sum_phi = gcry_mpi_point_new (0);
	brandt_assert (ad);
	for (uint16_t j = ad->k - 1; j >= 0; j--)
	{
		smc_sum (sum_gamma, &ad->gamma2[0][j], ad->n, ad->k);
		smc_sum (sum_phi, &ad->phi2[0][j], ad->n, ad->k);
		gcry_mpi_ec_sub (sum_gamma, sum_gamma, sum_phi, ec_ctx);
		/* first non-zero component determines the price */
		if (ec_point_cmp (sum_gamma, ec_zero))
		{
			ret = j;
			break;
		}
	}
	dlogi = GNUNET_CRYPTO_ecc_dlog (ec_dlogctx, sum_gamma);
	brandt_assert (dlogi > 0);
	gcry_mpi_set_ui (dlog, (unsigned long)dlogi);
	for (uint16_t i = 0; i < ad->n; i++)
	{
		if (gcry_mpi_test_bit (dlog, i))
		{
			if (winner)
				*winner = i;
			break;
		}
	}
	gcry_mpi_release (dlog);
	gcry_mpi_point_release (sum_gamma);
	gcry_mpi_point_release (sum_phi);
	return ret;
}
/**
 * fp_priv_compute_outcome computes encrypted outcome shares and packs them into
 * a message buffer together with proofs of correctnes.
 *
 * @param[in] ad Pointer to the BRANDT_Auction struct to operate on
 * @param[out] buflen Size of the returned message buffer in bytes
 * @return A buffer containing the encrypted outcome vectors
 * which needs to be broadcast
 */
unsigned char *
fp_priv_compute_outcome (struct BRANDT_Auction *ad, size_t *buflen)
{
	unsigned char     *ret;
	unsigned char     *cur;
	struct msg_head   *head;
	gcry_mpi_point_t  tmpa = gcry_mpi_point_new (0);
	gcry_mpi_point_t  tmpb = gcry_mpi_point_new (0);
	gcry_mpi_point_t  *tlta1;
	gcry_mpi_point_t  *tltb1;
	gcry_mpi_point_t  **tlta2;
	gcry_mpi_point_t  **tltb2;
	gcry_mpi_point_t  **tlta3;
	gcry_mpi_point_t  **tltb3;
	struct ec_mpi     *gamma;
	struct ec_mpi     *delta;
	struct proof_2dle *proof2;
	brandt_assert (ad && buflen);
	*buflen = (sizeof (*head) +                /* msg header */
	           ad->n * ad->k *                 /* nk * (gamma, delta, proof2) */
	           (sizeof (*gamma) + sizeof (*delta) + sizeof (*proof2)));
	ret = GNUNET_new_array (*buflen, unsigned char);
	if (NULL == (ad->gamma3 = smc_init3 (ad->n, ad->n, ad->k)) ||
	    NULL == (ad->delta3 = smc_init3 (ad->n, ad->n, ad->k)))
	{
		weprintf ("unable to alloc memory for first price outcome computation");
		return NULL;
	}
	head = (struct msg_head *)ret;
	head->prot_version = htonl (0);
	head->msg_type = htonl (msg_outcome);
	cur = ret + sizeof (*head);
	/* create temporary lookup tables with partial sums */
	tlta1 = smc_init1 (ad->k);
	tltb1 = smc_init1 (ad->k);
	tlta2 = smc_init2 (ad->n, ad->k);
	tltb2 = smc_init2 (ad->n, ad->k);
	tlta3 = smc_init2 (ad->n, ad->k);
	tltb3 = smc_init2 (ad->n, ad->k);
	/* temporary lookup table for first summand (no one has a higher bid) */
	for (uint16_t i = 0; i < ad->n; i++)
	{
		smc_sums_partial (tlta2[i], ad->alpha[i], ad->k, 1, 1);
		smc_sums_partial (tltb2[i], ad->beta[i], ad->k, 1, 1);
		for (uint16_t j = 0; j < ad->k; j++)
		{
			gcry_mpi_ec_sub (tlta3[i][j],
			                 tlta2[i][ad->k - 1],
			                 tlta2[i][j],
			                 ec_ctx);
			gcry_mpi_ec_sub (tltb3[i][j],
			                 tltb2[i][ad->k - 1],
			                 tltb2[i][j],
			                 ec_ctx);
		}
		brandt_assert (!ec_point_cmp (ec_zero, tlta3[i][ad->k - 1]));
		brandt_assert (!ec_point_cmp (ec_zero, tltb3[i][ad->k - 1]));
	}
	for (uint16_t j = 0; j < ad->k; j++)
	{
		smc_sum (tlta1[j], &tlta3[0][j], ad->n, ad->k);
		smc_sum (tltb1[j], &tltb3[0][j], ad->n, ad->k);
	}
	brandt_assert (!ec_point_cmp (ec_zero, tlta1[ad->k - 1]));
	brandt_assert (!ec_point_cmp (ec_zero, tltb1[ad->k - 1]));
	/* \todo: merge into one nested i,j loop and one nested j,i loop? */
	/* temporary lookup table for second summand (my bid is not lower) */
	for (uint16_t i = 0; i < ad->n; i++)
	{
		for (uint16_t j = 0; j < ad->k; j++)
		{
			gcry_mpi_ec_sub (tlta2[i][j], tlta2[i][j], ad->alpha[i][j], ec_ctx);
			gcry_mpi_ec_sub (tltb2[i][j], tltb2[i][j], ad->beta[i][j], ec_ctx);
		}
		brandt_assert (!ec_point_cmp (ec_zero, tlta2[i][0]));
		brandt_assert (!ec_point_cmp (ec_zero, tltb2[i][0]));
	}
	/* temporary lookup table for third summand (no one with a lower index has
	 * the same bid) */
	for (uint16_t j = 0; j < ad->k; j++)
	{
		smc_sums_partial (&tlta3[0][j], &ad->alpha[0][j], ad->n, ad->k, ad->k);
		smc_sums_partial (&tltb3[0][j], &ad->beta[0][j], ad->n, ad->k, ad->k);
		for (uint16_t i = 0; i < ad->n; i++)
		{
			gcry_mpi_ec_sub (tlta3[i][j], tlta3[i][j], ad->alpha[i][j], ec_ctx);
			gcry_mpi_ec_sub (tltb3[i][j], tltb3[i][j], ad->beta[i][j], ec_ctx);
		}
		brandt_assert (!ec_point_cmp (ec_zero, tlta3[0][j]));
		brandt_assert (!ec_point_cmp (ec_zero, tltb3[0][j]));
	}
	for (uint16_t i = 0; i < ad->n; i++)
	{
		for (uint16_t j = 0; j < ad->k; j++)
		{
			gamma = (struct ec_mpi *)cur;
			delta = &((struct ec_mpi *)cur)[1];
			proof2 = (struct proof_2dle *)(cur + 2 * sizeof (struct ec_mpi));
			/* compute inner gamma */
			gcry_mpi_ec_add (tmpa, tlta1[j], tlta2[i][j], ec_ctx);
			gcry_mpi_ec_add (tmpa, tmpa, tlta3[i][j], ec_ctx);
			/* compute inner delta */
			gcry_mpi_ec_add (tmpb, tltb1[j], tltb2[i][j], ec_ctx);
			gcry_mpi_ec_add (tmpb, tmpb, tltb3[i][j], ec_ctx);
			/* copy unmasked outcome to all other bidder layers so they don't
			 * have to be recomputed to check the ZK proof_2dle's from other
			 * bidders when receiving their outcome messages */
			for (uint16_t a = 0; a < ad->n; a++)
			{
				ec_point_copy (ad->gamma3[a][i][j], tmpa);
				ec_point_copy (ad->delta3[a][i][j], tmpb);
			}
			/* apply random masking for losing bidders */
			smc_zkp_2dle (ad->gamma3[ad->i][i][j],
			              ad->delta3[ad->i][i][j],
			              tmpa,
			              tmpb,
			              NULL,
			              proof2);
			ec_point_serialize (gamma, ad->gamma3[ad->i][i][j]);
			ec_point_serialize (delta, ad->delta3[ad->i][i][j]);
			cur += sizeof (*gamma) + sizeof (*delta) + sizeof (*proof2);
		}
	}
	gcry_mpi_point_release (tmpa);
	gcry_mpi_point_release (tmpb);
	smc_free1 (tlta1, ad->k);
	smc_free1 (tltb1, ad->k);
	smc_free2 (tlta2, ad->n, ad->k);
	smc_free2 (tltb2, ad->n, ad->k);
	smc_free2 (tlta3, ad->n, ad->k);
	smc_free2 (tltb3, ad->n, ad->k);
	return ret;
}
int
fp_priv_recv_outcome (struct BRANDT_Auction *ad,
                      const unsigned char   *buf,
                      size_t                buflen,
                      uint16_t              sender)
{
	int                 ret = 0;
	const unsigned char *cur = buf;
	struct proof_2dle   *proof2;
	gcry_mpi_point_t    gamma = gcry_mpi_point_new (0);
	gcry_mpi_point_t    delta = gcry_mpi_point_new (0);
	brandt_assert (ad && buf);
	if (buflen != (ad->n * ad->k *
	               (2 * sizeof (struct ec_mpi) + sizeof (*proof2))))
	{
		weprintf ("wrong size of received outcome");
		goto quit;
	}
	for (uint16_t i = 0; i < ad->n; i++)
	{
		for (uint16_t j = 0; j < ad->k; j++)
		{
			ec_point_parse (gamma, (struct ec_mpi *)cur);
			ec_point_parse (delta, &((struct ec_mpi *)cur)[1]);
			proof2 = (struct proof_2dle *)(cur + 2 * sizeof (struct ec_mpi));
			if (smc_zkp_2dle_check (gamma,
			                        delta,
			                        ad->gamma3[sender][i][j],
			                        ad->delta3[sender][i][j],
			                        proof2))
			{
				weprintf ("wrong zkp2 for gamma, delta received");
				goto quit;
			}
			ec_point_copy (ad->gamma3[sender][i][j], gamma);
			ec_point_copy (ad->delta3[sender][i][j], delta);
			cur += 2 * sizeof (struct ec_mpi) + sizeof (*proof2);
		}
	}
	ret = 1;
quit:
	gcry_mpi_point_release (gamma);
	gcry_mpi_point_release (delta);
	return ret;
}
/**
 * fp_priv_decrypt_outcome decrypts the own shares of the outcome and packs them
 * into a message buffer together with proofs of correctnes.
 *
 * @param[in] ad Pointer to the BRANDT_Auction struct to operate on
 * @param[out] buflen Size of the returned message buffer in bytes
 * @return A buffer containing the own share of the decrypted outcome
 * which needs to be broadcast
 */
unsigned char *
fp_priv_decrypt_outcome (struct BRANDT_Auction *ad, size_t *buflen)
{
	unsigned char     *ret;
	unsigned char     *cur;
	struct msg_head   *head;
	gcry_mpi_point_t  tmp = gcry_mpi_point_new (0);
	struct ec_mpi     *phi;
	struct proof_2dle *proof2;
	brandt_assert (ad && buflen);
	*buflen = (sizeof (*head) +
	           ad->n * ad->k * (sizeof (*phi) + sizeof (*proof2)));
	ret = GNUNET_new_array (*buflen, unsigned char);
	if (NULL == (ad->phi3 = smc_init3 (ad->n, ad->n, ad->k)))
	{
		weprintf ("unable to alloc memory for first price outcome decryption");
		return NULL;
	}
	head = (struct msg_head *)ret;
	head->prot_version = htonl (0);
	head->msg_type = htonl (msg_decrypt);
	cur = ret + sizeof (*head);
	for (uint16_t i = 0; i < ad->n; i++)
	{
		for (uint16_t j = 0; j < ad->k; j++)
		{
			phi = (struct ec_mpi *)cur;
			proof2 = (struct proof_2dle *)(cur + sizeof (*phi));
			smc_sum (tmp, &ad->delta3[0][i][j], ad->n, ad->n * ad->k);
			/* copy still encrypted outcome to all other bidder layers so they
			 * don't have to be recomputed to check the ZK proof_2dle's from
			 * other bidders when receiving their outcome decryption messages */
			for (uint16_t a = 0; a < ad->n; a++)
				ec_point_copy (ad->phi3[a][i][j], tmp);
			/* decrypt outcome component and prove the correct key was used */
			smc_zkp_2dle (ad->phi3[ad->i][i][j],
			              NULL,
			              tmp,
			              ec_gen,
			              ad->x,
			              proof2);
			ec_point_serialize (phi, ad->phi3[ad->i][i][j]);
			cur += sizeof (*phi) + sizeof (*proof2);
		}
	}
	gcry_mpi_point_release (tmp);
	return ret;
}
int
fp_priv_recv_decryption (struct BRANDT_Auction *ad,
                         const unsigned char   *buf,
                         size_t                buflen,
                         uint16_t              sender)
{
	int                 ret = 0;
	const unsigned char *cur = buf;
	struct proof_2dle   *proof2;
	gcry_mpi_point_t    phi = gcry_mpi_point_new (0);
	brandt_assert (ad && buf);
	if (buflen != (ad->n * ad->k * (sizeof (struct ec_mpi) + sizeof (*proof2))))
	{
		weprintf ("wrong size of received outcome decryption");
		goto quit;
	}
	for (uint16_t i = 0; i < ad->n; i++)
	{
		for (uint16_t j = 0; j < ad->k; j++)
		{
			ec_point_parse (phi, (struct ec_mpi *)cur);
			proof2 = (struct proof_2dle *)(cur + sizeof (struct ec_mpi));
			if (smc_zkp_2dle_check (phi,
			                        ad->y[sender],
			                        ad->phi3[sender][i][j],
			                        ec_gen,
			                        proof2))
			{
				weprintf ("wrong zkp2 for phi, y received");
				goto quit;
			}
			ec_point_copy (ad->phi3[sender][i][j], phi);
			cur += sizeof (struct ec_mpi) + sizeof (*proof2);
		}
	}
	ret = 1;
quit:
	gcry_mpi_point_release (phi);
	return ret;
}
int32_t
fp_priv_determine_outcome (struct BRANDT_Auction *ad)
{
	int32_t          ret = -1;
	gcry_mpi_point_t sum_gamma = gcry_mpi_point_new (0);
	gcry_mpi_point_t sum_phi = gcry_mpi_point_new (0);
	brandt_assert (ad);
	for (uint16_t j = 0; j < ad->k; j++)
	{
		smc_sum (sum_gamma, &ad->gamma3[0][ad->i][j], ad->n, ad->n * ad->k);
		smc_sum (sum_phi, &ad->phi3[0][ad->i][j], ad->n, ad->n * ad->k);
		gcry_mpi_ec_sub (sum_gamma, sum_gamma, sum_phi, ec_ctx);
		if (!ec_point_cmp (sum_gamma, ec_zero))
		{
			if (-1 != ret)
			{
				weprintf ("multiple winning prices detected");
				return -1;
			}
			ret = j;
		}
	}
	gcry_mpi_point_release (sum_gamma);
	gcry_mpi_point_release (sum_phi);
	return ret;
}
/**
 * smc_zkp_dl creates a proof of knowledge of @a x with \f$v = xg\f$ where
 * \f$g\f$ is the base point on Ed25519.
 *
 * @param[out] v output point. Must be known to the verifier.
 * @param[in] x private key. Knowledge of this number is certified in the proof
 * @param[out] proof pointer where to save the output proof structure. Must be
 * shared with the verifier.
 */
void
smc_zkp_dl (gcry_mpi_point_t v,
            const gcry_mpi_t x,
            struct proof_dl  *proof)
{
	struct zkp_challenge_dl challenge;
	gcry_mpi_point_t        a = gcry_mpi_point_new (0);
	gcry_mpi_t              r = gcry_mpi_new (256);
	gcry_mpi_t              c;
	gcry_mpi_t              z = gcry_mpi_new (256);
	/* v = xg */
	gcry_mpi_ec_mul (v, x, ec_gen, ec_ctx);
	/* a = zg */
	ec_keypair_create (a, z);
	/* compute challenge c */
	ec_point_serialize (&challenge.g, ec_gen);
	ec_point_serialize (&challenge.v, v);
	ec_point_serialize (&challenge.a, a);
	GNUNET_CRYPTO_kdf_mod_mpi (&c,
	                           ec_n,
	                           NULL,
	                           0,
	                           &challenge,
	                           sizeof (challenge),
	                           "libbrandt zkp dl");
	/* r = z + cx */
	gcry_mpi_mulm (r, c, x, ec_n);
	gcry_mpi_addm (r, r, z, ec_n);
	ec_point_serialize (&proof->a, a);
	mpi_serialize (&proof->r, r);
	gcry_mpi_point_release (a);
	gcry_mpi_release (r);
	gcry_mpi_release (c);
	gcry_mpi_release (z);
}
/**
 * smc_zkp_dl_check verifies a proof of knowledge of \f$x = ECDL_g(v)\f$ where
 * \f$g\f$ is the base point on Ed25519.
 *
 * @param[in] v input point. Received from the prover.
 * @param[in] proof pointer to the proof structure. Received from the prover.
 * @return 0 if the proof is correct, something else otherwise
 */
int
smc_zkp_dl_check (const gcry_mpi_point_t v,
                  const struct proof_dl  *proof)
{
	int                     ret;
	struct zkp_challenge_dl challenge;
	gcry_mpi_point_t        a = gcry_mpi_point_new (0);
	gcry_mpi_t              r = gcry_mpi_new (256);
	gcry_mpi_t              c;
	gcry_mpi_point_t        left = gcry_mpi_point_new (0);
	gcry_mpi_point_t        right = gcry_mpi_point_new (0);
	ec_point_parse (a, &proof->a);
	mpi_parse (r, &proof->r);
	/* compute challenge c */
	ec_point_serialize (&challenge.g, ec_gen);
	ec_point_serialize (&challenge.v, v);
	ec_point_serialize (&challenge.a, a);
	GNUNET_CRYPTO_kdf_mod_mpi (&c,
	                           ec_n,
	                           NULL,
	                           0,
	                           &challenge,
	                           sizeof (challenge),
	                           "libbrandt zkp dl");
	/* rg =? a + cv */
	gcry_mpi_ec_mul (left, r, ec_gen, ec_ctx);
	gcry_mpi_ec_mul (right, c, v, ec_ctx);
	gcry_mpi_ec_add (right, a, right, ec_ctx);
	ret = ec_point_cmp (left, right);
	gcry_mpi_point_release (a);
	gcry_mpi_release (r);
	gcry_mpi_release (c);
	gcry_mpi_point_release (left);
	gcry_mpi_point_release (right);
	return ret;
}
/**
 * smc_zkp_2dle creates a proof that two ECDLs are equal without revealing the
 * ECDL. \f$v=xg_1, w=xg_2\f$ are calculated as well and can be returned to the
 * caller if needed.
 *
 * @param[out] v first output point. May be NULL if not needed by the caller.
 * Must be known to the verifier.
 * @param[out] w second output point. May be NULL if not needed by the caller.
 * Must be known to the verifier.
 * @param[in] g1 first base point. Must be known to the verifier.
 * @param[in] g2 second base point. Must be known to the verifier.
 * @param[in] x private number to prove knowledge of. May be NULL if not used by
 * the caller.
 * @param[out] proof pointer where to save the output proof structure. Must be
 * shared with the verifier.
 */
void
smc_zkp_2dle (gcry_mpi_point_t       v,
              gcry_mpi_point_t       w,
              const gcry_mpi_point_t g1,
              const gcry_mpi_point_t g2,
              const gcry_mpi_t       x,
              struct proof_2dle      *proof)
{
	struct zkp_challenge_2dle challenge;
	gcry_mpi_point_t          rv;
	gcry_mpi_point_t          rw;
	gcry_mpi_t                rx;
	gcry_mpi_point_t          a = gcry_mpi_point_new (0);
	gcry_mpi_point_t          b = gcry_mpi_point_new (0);
	gcry_mpi_t                r = gcry_mpi_new (256);
	gcry_mpi_t                c;
	gcry_mpi_t                z = gcry_mpi_new (256);
	rv = (NULL == v) ? gcry_mpi_point_new (0) : v;
	rw = (NULL == w) ? gcry_mpi_point_new (0) : w;
	rx = (NULL == x) ? gcry_mpi_new (256) : x;
	if (NULL == x)
		ec_skey_create (rx);
	/* v = x*g1 */
	gcry_mpi_ec_mul (rv, rx, g1, ec_ctx);
	/* w = x*g2 */
	gcry_mpi_ec_mul (rw, rx, g2, ec_ctx);
	/* a = z*g1 */
	ec_keypair_create_base (a, z, g1);
	/* b = z*g2 */
	gcry_mpi_ec_mul (b, z, g2, ec_ctx);
	/* compute challenge c */
	ec_point_serialize (&challenge.g1, g1);
	ec_point_serialize (&challenge.g2, g2);
	ec_point_serialize (&challenge.v, rv);
	ec_point_serialize (&challenge.w, rw);
	ec_point_serialize (&challenge.a, a);
	ec_point_serialize (&challenge.b, b);
	GNUNET_CRYPTO_kdf_mod_mpi (&c,
	                           ec_n,
	                           NULL,
	                           0,
	                           &challenge,
	                           sizeof (challenge),
	                           "libbrandt zkp 2dle");
	/* r = z + cx */
	gcry_mpi_mulm (r, c, rx, ec_n);
	gcry_mpi_addm (r, r, z, ec_n);
	mpi_serialize (&proof->r, r);
	ec_point_serialize (&proof->a, a);
	ec_point_serialize (&proof->b, b);
	if (NULL == v)
		gcry_mpi_point_release (rv);
	if (NULL == w)
		gcry_mpi_point_release (rw);
	if (NULL == x)
		gcry_mpi_release (rx);
	gcry_mpi_point_release (a);
	gcry_mpi_point_release (b);
	gcry_mpi_release (r);
	gcry_mpi_release (c);
	gcry_mpi_release (z);
}
/**
 * smc_zkp_2dle_check verifies a proof of knowledge of \f$x\f$ with \f$v=xg_1\f$
 * and \f$w=xg_2\f$.
 *
 * @param[in] v first input point.
 * @param[in] w second input point.
 * @param[in] g1 first base point.
 * @param[in] g2 second base point.
 * @param[in] proof pointer to the proof structure. Received from the prover.
 * @return 0 if the proof is correct, something else otherwise
 */
int
smc_zkp_2dle_check (const gcry_mpi_point_t  v,
                    const gcry_mpi_point_t  w,
                    const gcry_mpi_point_t  g1,
                    const gcry_mpi_point_t  g2,
                    const struct proof_2dle *proof)
{
	int                       ret;
	struct zkp_challenge_2dle challenge;
	gcry_mpi_point_t          a = gcry_mpi_point_new (0);
	gcry_mpi_point_t          b = gcry_mpi_point_new (0);
	gcry_mpi_t                r = gcry_mpi_new (256);
	gcry_mpi_t                c;
	gcry_mpi_point_t          left = gcry_mpi_point_new (0);
	gcry_mpi_point_t          right = gcry_mpi_point_new (0);
	mpi_parse (r, &proof->r);
	ec_point_parse (a, &proof->a);
	ec_point_parse (b, &proof->b);
	/* compute challenge c */
	ec_point_serialize (&challenge.g1, g1);
	ec_point_serialize (&challenge.g2, g2);
	ec_point_serialize (&challenge.v, v);
	ec_point_serialize (&challenge.w, w);
	ec_point_serialize (&challenge.a, a);
	ec_point_serialize (&challenge.b, b);
	GNUNET_CRYPTO_kdf_mod_mpi (&c,
	                           ec_n,
	                           NULL,
	                           0,
	                           &challenge,
	                           sizeof (challenge),
	                           "libbrandt zkp 2dle");
	/* r*g1 =? a + cv */
	gcry_mpi_ec_mul (left, r, g1, ec_ctx);
	gcry_mpi_ec_mul (right, c, v, ec_ctx);
	gcry_mpi_ec_add (right, a, right, ec_ctx);
	ret = ec_point_cmp (left, right);
	/* r*g2 =? b + cw */
	gcry_mpi_ec_mul (left, r, g2, ec_ctx);
	gcry_mpi_ec_mul (right, c, w, ec_ctx);
	gcry_mpi_ec_add (right, b, right, ec_ctx);
	ret |= ec_point_cmp (left, right);
	gcry_mpi_point_release (a);
	gcry_mpi_point_release (b);
	gcry_mpi_release (r);
	gcry_mpi_release (c);
	gcry_mpi_point_release (left);
	gcry_mpi_point_release (right);
	return ret;
}
/**
 * smc_zkp_0og encrypts one of two values and creates a proof that the
 * ciphertext decrypts to either one of those two values without revealing which
 * one was encrypted. The two values are the zero point or the base point of the
 * Ed25519 curve. Encryption is done via ElGamal: \f$(\alpha,\beta)=(m+ry,rg)\f$
 * where \f$m\f$ is the value to encrypt, \f$y\f$ is the public key and \f$g\f$
 * is the base point. The nonce \f$r\f$ is generated as well and can be returned
 * to the caller if he needs it (e.g. for another proof).
 *
 * @param[in] m_is_gen if true, the base point is encrypted, else the zero point
 * is encrypted.
 * @param[in] y public key to use for encryption.
 * @param[out] r random number used for encryption. May be NULL if caller
 * doesn't need it.
 * @param[out] alpha first part of the ciphertext output
 * @param[out] beta second part of the ciphertext output
 * @param[out] proof pointer where to save the output proof structure. Must be
 * shared with the verifier.
 */
void
smc_zkp_0og (int                    m_is_gen,
             const gcry_mpi_point_t y,
             gcry_mpi_t             r,
             gcry_mpi_point_t       alpha,
             gcry_mpi_point_t       beta,
             struct proof_0og       *proof)
{
	struct zkp_challenge_0og challenge;
	gcry_mpi_point_t         a1 = gcry_mpi_point_new (0);
	gcry_mpi_point_t         a2 = gcry_mpi_point_new (0);
	gcry_mpi_point_t         b1 = gcry_mpi_point_new (0);
	gcry_mpi_point_t         b2 = gcry_mpi_point_new (0);
	gcry_mpi_t               d1 = gcry_mpi_new (256);
	gcry_mpi_t               d2 = gcry_mpi_new (256);
	gcry_mpi_t               r1 = gcry_mpi_new (256);
	gcry_mpi_t               r2 = gcry_mpi_new (256);
	gcry_mpi_t               c;
	gcry_mpi_t               rr;
	gcry_mpi_t               w = gcry_mpi_new (256);
	rr = (NULL == r) ? gcry_mpi_new (256) : r;
	/* beta = r*g */
	ec_keypair_create (beta, rr);
	gcry_mpi_mod (rr, rr, ec_n);
	/* alpha = m + r*y */
	gcry_mpi_ec_mul (alpha, rr, y, ec_ctx);
	gcry_mpi_ec_add (alpha, m_is_gen ? ec_gen : ec_zero, alpha, ec_ctx);
	if (!m_is_gen)
	{   /* m == 0 */
		ec_keypair_create_base (a1, d1, beta);
		gcry_mpi_mod (d1, d1, ec_n);
		ec_keypair_create_base (b1, r1, y);
		gcry_mpi_mod (r1, r1, ec_n);
		/* a1 = r1*g + d1*beta */
		gcry_mpi_ec_mul (a2, r1, ec_gen, ec_ctx);
		gcry_mpi_ec_add (a1, a2, a1, ec_ctx);
		/* b1 = r1*y + d1*(alpha-g) */
		gcry_mpi_ec_sub (b2, alpha, ec_gen, ec_ctx);
		gcry_mpi_ec_mul (a2, d1, b2, ec_ctx);
		gcry_mpi_ec_add (b1, b1, a2, ec_ctx);
		/* a2 = w * g */
		ec_keypair_create_base (a2, w, ec_gen);
		gcry_mpi_mod (w, w, ec_n);
		/* b2 = w * y */
		gcry_mpi_ec_mul (b2, w, y, ec_ctx);
	}
	else
	{   /* m == g */
		ec_keypair_create_base (a2, d2, beta);
		gcry_mpi_mod (d2, d2, ec_n);
		ec_keypair_create_base (b2, r2, y);
		gcry_mpi_mod (r2, r2, ec_n);
		/* a2 = r2*g + d2*beta */
		gcry_mpi_ec_mul (a1, r2, ec_gen, ec_ctx);
		gcry_mpi_ec_add (a2, a1, a2, ec_ctx);
		/* b2 = r2*y + d2*(alpha-0) */
		/* useless subtraction to have same amount of operations as in m == 0 */
		gcry_mpi_ec_sub (b1, alpha, ec_zero, ec_ctx);
		gcry_mpi_ec_mul (a1, d2, b1, ec_ctx);
		gcry_mpi_ec_add (b2, b2, a1, ec_ctx);
		/* a1 = w * g */
		ec_keypair_create_base (a1, w, ec_gen);
		gcry_mpi_mod (w, w, ec_n);
		/* b1 = w * y */
		gcry_mpi_ec_mul (b1, w, y, ec_ctx);
	}
	/* compute challenge c */
	ec_point_serialize (&challenge.g, ec_gen);
	ec_point_serialize (&challenge.alpha, alpha);
	ec_point_serialize (&challenge.beta, beta);
	ec_point_serialize (&challenge.a1, a1);
	ec_point_serialize (&challenge.a2, a2);
	ec_point_serialize (&challenge.b1, b1);
	ec_point_serialize (&challenge.b2, b2);
	GNUNET_CRYPTO_kdf_mod_mpi (&c,
	                           ec_n,
	                           NULL,
	                           0,
	                           &challenge,
	                           sizeof (challenge),
	                           "libbrandt zkp 0og");
	if (!m_is_gen)
	{   /* m == 0 */
		/* d2 = c - d1 */
		gcry_mpi_subm (d2, c, d1, ec_n);
		/* r2 = w - r*d2 */
		gcry_mpi_mulm (r2, rr, d2, ec_n);
		gcry_mpi_subm (r2, w, r2, ec_n);
	}
	else
	{   /* m == g */
		/* d1 = c - d2 */
		gcry_mpi_subm (d1, c, d2, ec_n);
		/* r1 = w - r*d1 */
		gcry_mpi_mulm (r1, rr, d1, ec_n);
		gcry_mpi_subm (r1, w, r1, ec_n);
	}
	ec_point_serialize (&proof->a1, a1);
	ec_point_serialize (&proof->a2, a2);
	ec_point_serialize (&proof->b1, b1);
	ec_point_serialize (&proof->b2, b2);
	mpi_serialize (&proof->d1, d1);
	mpi_serialize (&proof->d2, d2);
	mpi_serialize (&proof->r1, r1);
	mpi_serialize (&proof->r2, r2);
	gcry_mpi_point_release (a1);
	gcry_mpi_point_release (a2);
	gcry_mpi_point_release (b1);
	gcry_mpi_point_release (b2);
	gcry_mpi_release (d1);
	gcry_mpi_release (d2);
	gcry_mpi_release (r1);
	gcry_mpi_release (r2);
	gcry_mpi_release (c);
	if (NULL == r)
		gcry_mpi_release (rr);
	gcry_mpi_release (w);
}
/**
 * smc_zkp_0og_check verifies a proof that \f$(\alpha,\beta\f$ decrypts either
 * to the base point \f$g\f$ or the zero point.
 *
 * @param[in] y the public key used for encryption
 * @param[in] alpha first part of the ciphertext
 * @param[in] beta second part of the ciphertext
 * @param[in] proof pointer to the proof structure. Received from the prover.
 * @return 0 if the proof is correct, something else otherwise
 */
int
smc_zkp_0og_check (const gcry_mpi_point_t y,
                   const gcry_mpi_point_t alpha,
                   const gcry_mpi_point_t beta,
                   const struct proof_0og *proof)
{
	int                      ret;
	struct zkp_challenge_0og challenge;
	gcry_mpi_point_t         a1 = gcry_mpi_point_new (0);
	gcry_mpi_point_t         a2 = gcry_mpi_point_new (0);
	gcry_mpi_point_t         b1 = gcry_mpi_point_new (0);
	gcry_mpi_point_t         b2 = gcry_mpi_point_new (0);
	gcry_mpi_t               d1 = gcry_mpi_new (256);
	gcry_mpi_t               d2 = gcry_mpi_new (256);
	gcry_mpi_t               r1 = gcry_mpi_new (256);
	gcry_mpi_t               r2 = gcry_mpi_new (256);
	gcry_mpi_t               c;
	gcry_mpi_t               sum = gcry_mpi_new (256);
	gcry_mpi_point_t         right = gcry_mpi_point_new (0);
	gcry_mpi_point_t         tmp = gcry_mpi_point_new (0);
	ec_point_parse (a1, &proof->a1);
	ec_point_parse (a2, &proof->a2);
	ec_point_parse (b1, &proof->b1);
	ec_point_parse (b2, &proof->b2);
	mpi_parse (d1, &proof->d1);
	mpi_parse (d2, &proof->d2);
	mpi_parse (r1, &proof->r1);
	mpi_parse (r2, &proof->r2);
	/* compute challenge c */
	ec_point_serialize (&challenge.g, ec_gen);
	ec_point_serialize (&challenge.alpha, alpha);
	ec_point_serialize (&challenge.beta, beta);
	ec_point_serialize (&challenge.a1, a1);
	ec_point_serialize (&challenge.a2, a2);
	ec_point_serialize (&challenge.b1, b1);
	ec_point_serialize (&challenge.b2, b2);
	GNUNET_CRYPTO_kdf_mod_mpi (&c,
	                           ec_n,
	                           NULL,
	                           0,
	                           &challenge,
	                           sizeof (challenge),
	                           "libbrandt zkp 0og");
	/* c == d1 + d2 */
	gcry_mpi_addm (sum, d1, d2, ec_n);
	ret = gcry_mpi_cmp (c, sum);
	/* a1 == r1*g + d1*beta */
	gcry_mpi_ec_mul (tmp, r1, ec_gen, ec_ctx);
	gcry_mpi_ec_mul (right, d1, beta, ec_ctx);
	gcry_mpi_ec_add (right, tmp, right, ec_ctx);
	ret |= ec_point_cmp (a1, right) << 1;
	/* b1 == r1*y + d1*(alpha-g) */
	gcry_mpi_ec_sub (right, alpha, ec_gen, ec_ctx);
	gcry_mpi_ec_mul (tmp, d1, right, ec_ctx);
	gcry_mpi_ec_mul (right, r1, y, ec_ctx);
	gcry_mpi_ec_add (right, right, tmp, ec_ctx);
	ret |= ec_point_cmp (b1, right) << 2;
	/* a2 == r2*g + d2*beta */
	gcry_mpi_ec_mul (tmp, d2, beta, ec_ctx);
	gcry_mpi_ec_mul (right, r2, ec_gen, ec_ctx);
	gcry_mpi_ec_add (right, right, tmp, ec_ctx);
	ret |= ec_point_cmp (a2, right) << 3;
	/* b2 == r2*y + d2*alpha */
	gcry_mpi_ec_mul (tmp, d2, alpha, ec_ctx);
	gcry_mpi_ec_mul (right, r2, y, ec_ctx);
	gcry_mpi_ec_add (right, right, tmp, ec_ctx);
	ret |= ec_point_cmp (b2, right) << 4;
	gcry_mpi_point_release (a1);
	gcry_mpi_point_release (a2);
	gcry_mpi_point_release (b1);
	gcry_mpi_point_release (b2);
	gcry_mpi_release (d1);
	gcry_mpi_release (d2);
	gcry_mpi_release (r1);
	gcry_mpi_release (r2);
	gcry_mpi_release (c);
	gcry_mpi_release (sum);
	gcry_mpi_point_release (right);
	gcry_mpi_point_release (tmp);
	if (ret)
		weprintf ("ret: 0x%x", ret);
	return ret;
}