Mercurial > dropbear
view libtommath/tommath_private.h @ 1883:f54451afc046
use buf_getptr and m_free on every iteration before m_malloc to insure no memory leaks are happening
author | HansH111 <hans@atbas.org> |
---|---|
date | Tue, 15 Mar 2022 18:57:21 +0000 |
parents | 1051e4eea25a |
children |
line wrap: on
line source
/* LibTomMath, multiple-precision integer library -- Tom St Denis */ /* SPDX-License-Identifier: Unlicense */ #ifndef TOMMATH_PRIV_H_ #define TOMMATH_PRIV_H_ #include "tommath.h" #include "tommath_class.h" /* * Private symbols * --------------- * * On Unix symbols can be marked as hidden if libtommath is compiled * as a shared object. By default, symbols are visible. * As of now, this feature is opt-in via the MP_PRIVATE_SYMBOLS define. * * On Win32 a .def file must be used to specify the exported symbols. */ #if defined (MP_PRIVATE_SYMBOLS) && defined(__GNUC__) && __GNUC__ >= 4 # define MP_PRIVATE __attribute__ ((visibility ("hidden"))) #else # define MP_PRIVATE #endif /* Hardening libtommath * -------------------- * * By default memory is zeroed before calling * MP_FREE to avoid leaking data. This is good * practice in cryptographical applications. * * Note however that memory allocators used * in cryptographical applications can often * be configured by itself to clear memory, * rendering the clearing in tommath unnecessary. * See for example https://github.com/GrapheneOS/hardened_malloc * and the option CONFIG_ZERO_ON_FREE. * * Furthermore there are applications which * value performance more and want this * feature to be disabled. For such applications * define MP_NO_ZERO_ON_FREE during compilation. */ #ifdef MP_NO_ZERO_ON_FREE # define MP_FREE_BUFFER(mem, size) MP_FREE((mem), (size)) # define MP_FREE_DIGITS(mem, digits) MP_FREE((mem), sizeof (mp_digit) * (size_t)(digits)) #else # define MP_FREE_BUFFER(mem, size) \ do { \ size_t fs_ = (size); \ void* fm_ = (mem); \ if (fm_ != NULL) { \ MP_ZERO_BUFFER(fm_, fs_); \ MP_FREE(fm_, fs_); \ } \ } while (0) # define MP_FREE_DIGITS(mem, digits) \ do { \ int fd_ = (digits); \ void* fm_ = (mem); \ if (fm_ != NULL) { \ size_t fs_ = sizeof (mp_digit) * (size_t)fd_; \ MP_ZERO_BUFFER(fm_, fs_); \ MP_FREE(fm_, fs_); \ } \ } while (0) #endif #ifdef MP_USE_MEMSET # include <string.h> # define MP_ZERO_BUFFER(mem, size) memset((mem), 0, (size)) # define MP_ZERO_DIGITS(mem, digits) \ do { \ int zd_ = (digits); \ if (zd_ > 0) { \ memset((mem), 0, sizeof(mp_digit) * (size_t)zd_); \ } \ } while (0) #else # define MP_ZERO_BUFFER(mem, size) \ do { \ size_t zs_ = (size); \ char* zm_ = (char*)(mem); \ while (zs_-- > 0u) { \ *zm_++ = '\0'; \ } \ } while (0) # define MP_ZERO_DIGITS(mem, digits) \ do { \ int zd_ = (digits); \ mp_digit* zm_ = (mem); \ while (zd_-- > 0) { \ *zm_++ = 0; \ } \ } while (0) #endif /* Tunable cutoffs * --------------- * * - In the default settings, a cutoff X can be modified at runtime * by adjusting the corresponding X_CUTOFF variable. * * - Tunability of the library can be disabled at compile time * by defining the MP_FIXED_CUTOFFS macro. * * - There is an additional file tommath_cutoffs.h, which defines * the default cutoffs. These can be adjusted manually or by the * autotuner. * */ #ifdef MP_FIXED_CUTOFFS # include "tommath_cutoffs.h" # define MP_KARATSUBA_MUL_CUTOFF MP_DEFAULT_KARATSUBA_MUL_CUTOFF # define MP_KARATSUBA_SQR_CUTOFF MP_DEFAULT_KARATSUBA_SQR_CUTOFF # define MP_TOOM_MUL_CUTOFF MP_DEFAULT_TOOM_MUL_CUTOFF # define MP_TOOM_SQR_CUTOFF MP_DEFAULT_TOOM_SQR_CUTOFF #else # define MP_KARATSUBA_MUL_CUTOFF KARATSUBA_MUL_CUTOFF # define MP_KARATSUBA_SQR_CUTOFF KARATSUBA_SQR_CUTOFF # define MP_TOOM_MUL_CUTOFF TOOM_MUL_CUTOFF # define MP_TOOM_SQR_CUTOFF TOOM_SQR_CUTOFF #endif /* define heap macros */ #ifndef MP_MALLOC /* default to libc stuff */ # include <stdlib.h> # define MP_MALLOC(size) malloc(size) # define MP_REALLOC(mem, oldsize, newsize) realloc((mem), (newsize)) # define MP_CALLOC(nmemb, size) calloc((nmemb), (size)) # define MP_FREE(mem, size) free(mem) #else /* prototypes for our heap functions */ extern void *MP_MALLOC(size_t size); extern void *MP_REALLOC(void *mem, size_t oldsize, size_t newsize); extern void *MP_CALLOC(size_t nmemb, size_t size); extern void MP_FREE(void *mem, size_t size); #endif /* feature detection macro */ #ifdef _MSC_VER /* Prevent false positive: not enough arguments for function-like macro invocation */ #pragma warning(disable: 4003) #endif #define MP_STRINGIZE(x) MP__STRINGIZE(x) #define MP__STRINGIZE(x) ""#x"" #define MP_HAS(x) (sizeof(MP_STRINGIZE(BN_##x##_C)) == 1u) /* TODO: Remove private_mp_word as soon as deprecated mp_word is removed from tommath. */ #undef mp_word typedef private_mp_word mp_word; #define MP_MIN(x, y) (((x) < (y)) ? (x) : (y)) #define MP_MAX(x, y) (((x) > (y)) ? (x) : (y)) /* Static assertion */ #define MP_STATIC_ASSERT(msg, cond) typedef char mp_static_assert_##msg[(cond) ? 1 : -1]; /* ---> Basic Manipulations <--- */ #define MP_IS_ZERO(a) ((a)->used == 0) #define MP_IS_EVEN(a) (((a)->used == 0) || (((a)->dp[0] & 1u) == 0u)) #define MP_IS_ODD(a) (((a)->used > 0) && (((a)->dp[0] & 1u) == 1u)) #define MP_SIZEOF_BITS(type) ((size_t)CHAR_BIT * sizeof(type)) #define MP_MAXFAST (int)(1uL << (MP_SIZEOF_BITS(mp_word) - (2u * (size_t)MP_DIGIT_BIT))) /* TODO: Remove PRIVATE_MP_WARRAY as soon as deprecated MP_WARRAY is removed from tommath.h */ #undef MP_WARRAY #define MP_WARRAY PRIVATE_MP_WARRAY /* TODO: Remove PRIVATE_MP_PREC as soon as deprecated MP_PREC is removed from tommath.h */ #ifdef PRIVATE_MP_PREC # undef MP_PREC # define MP_PREC PRIVATE_MP_PREC #endif /* Minimum number of available digits in mp_int, MP_PREC >= MP_MIN_PREC */ #define MP_MIN_PREC ((((int)MP_SIZEOF_BITS(long long) + MP_DIGIT_BIT) - 1) / MP_DIGIT_BIT) MP_STATIC_ASSERT(prec_geq_min_prec, MP_PREC >= MP_MIN_PREC) /* random number source */ extern MP_PRIVATE mp_err(*s_mp_rand_source)(void *out, size_t size); /* lowlevel functions, do not call! */ MP_PRIVATE mp_bool s_mp_get_bit(const mp_int *a, unsigned int b); MP_PRIVATE mp_err s_mp_add(const mp_int *a, const mp_int *b, mp_int *c) MP_WUR; MP_PRIVATE mp_err s_mp_sub(const mp_int *a, const mp_int *b, mp_int *c) MP_WUR; MP_PRIVATE mp_err s_mp_mul_digs_fast(const mp_int *a, const mp_int *b, mp_int *c, int digs) MP_WUR; MP_PRIVATE mp_err s_mp_mul_digs(const mp_int *a, const mp_int *b, mp_int *c, int digs) MP_WUR; MP_PRIVATE mp_err s_mp_mul_high_digs_fast(const mp_int *a, const mp_int *b, mp_int *c, int digs) MP_WUR; MP_PRIVATE mp_err s_mp_mul_high_digs(const mp_int *a, const mp_int *b, mp_int *c, int digs) MP_WUR; MP_PRIVATE mp_err s_mp_sqr_fast(const mp_int *a, mp_int *b) MP_WUR; MP_PRIVATE mp_err s_mp_sqr(const mp_int *a, mp_int *b) MP_WUR; MP_PRIVATE mp_err s_mp_balance_mul(const mp_int *a, const mp_int *b, mp_int *c) MP_WUR; MP_PRIVATE mp_err s_mp_karatsuba_mul(const mp_int *a, const mp_int *b, mp_int *c) MP_WUR; MP_PRIVATE mp_err s_mp_toom_mul(const mp_int *a, const mp_int *b, mp_int *c) MP_WUR; MP_PRIVATE mp_err s_mp_karatsuba_sqr(const mp_int *a, mp_int *b) MP_WUR; MP_PRIVATE mp_err s_mp_toom_sqr(const mp_int *a, mp_int *b) MP_WUR; MP_PRIVATE mp_err s_mp_invmod_fast(const mp_int *a, const mp_int *b, mp_int *c) MP_WUR; MP_PRIVATE mp_err s_mp_invmod_slow(const mp_int *a, const mp_int *b, mp_int *c) MP_WUR; MP_PRIVATE mp_err s_mp_montgomery_reduce_fast(mp_int *x, const mp_int *n, mp_digit rho) MP_WUR; MP_PRIVATE mp_err s_mp_exptmod_fast(const mp_int *G, const mp_int *X, const mp_int *P, mp_int *Y, int redmode) MP_WUR; MP_PRIVATE mp_err s_mp_exptmod(const mp_int *G, const mp_int *X, const mp_int *P, mp_int *Y, int redmode) MP_WUR; MP_PRIVATE mp_err s_mp_rand_platform(void *p, size_t n) MP_WUR; MP_PRIVATE mp_err s_mp_prime_random_ex(mp_int *a, int t, int size, int flags, private_mp_prime_callback cb, void *dat); MP_PRIVATE void s_mp_reverse(unsigned char *s, size_t len); MP_PRIVATE mp_err s_mp_prime_is_divisible(const mp_int *a, mp_bool *result); /* TODO: jenkins prng is not thread safe as of now */ MP_PRIVATE mp_err s_mp_rand_jenkins(void *p, size_t n) MP_WUR; MP_PRIVATE void s_mp_rand_jenkins_init(uint64_t seed); extern MP_PRIVATE const char *const mp_s_rmap; extern MP_PRIVATE const uint8_t mp_s_rmap_reverse[]; extern MP_PRIVATE const size_t mp_s_rmap_reverse_sz; extern MP_PRIVATE const mp_digit *s_mp_prime_tab; /* deprecated functions */ MP_DEPRECATED(s_mp_invmod_fast) mp_err fast_mp_invmod(const mp_int *a, const mp_int *b, mp_int *c); MP_DEPRECATED(s_mp_montgomery_reduce_fast) mp_err fast_mp_montgomery_reduce(mp_int *x, const mp_int *n, mp_digit rho); MP_DEPRECATED(s_mp_mul_digs_fast) mp_err fast_s_mp_mul_digs(const mp_int *a, const mp_int *b, mp_int *c, int digs); MP_DEPRECATED(s_mp_mul_high_digs_fast) mp_err fast_s_mp_mul_high_digs(const mp_int *a, const mp_int *b, mp_int *c, int digs); MP_DEPRECATED(s_mp_sqr_fast) mp_err fast_s_mp_sqr(const mp_int *a, mp_int *b); MP_DEPRECATED(s_mp_balance_mul) mp_err mp_balance_mul(const mp_int *a, const mp_int *b, mp_int *c); MP_DEPRECATED(s_mp_exptmod_fast) mp_err mp_exptmod_fast(const mp_int *G, const mp_int *X, const mp_int *P, mp_int *Y, int redmode); MP_DEPRECATED(s_mp_invmod_slow) mp_err mp_invmod_slow(const mp_int *a, const mp_int *b, mp_int *c); MP_DEPRECATED(s_mp_karatsuba_mul) mp_err mp_karatsuba_mul(const mp_int *a, const mp_int *b, mp_int *c); MP_DEPRECATED(s_mp_karatsuba_sqr) mp_err mp_karatsuba_sqr(const mp_int *a, mp_int *b); MP_DEPRECATED(s_mp_toom_mul) mp_err mp_toom_mul(const mp_int *a, const mp_int *b, mp_int *c); MP_DEPRECATED(s_mp_toom_sqr) mp_err mp_toom_sqr(const mp_int *a, mp_int *b); MP_DEPRECATED(s_mp_reverse) void bn_reverse(unsigned char *s, int len); #define MP_GET_ENDIANNESS(x) \ do{\ int16_t n = 0x1; \ char *p = (char *)&n; \ x = (p[0] == '\x01') ? MP_LITTLE_ENDIAN : MP_BIG_ENDIAN; \ } while (0) /* code-generating macros */ #define MP_SET_UNSIGNED(name, type) \ void name(mp_int * a, type b) \ { \ int i = 0; \ while (b != 0u) { \ a->dp[i++] = ((mp_digit)b & MP_MASK); \ if (MP_SIZEOF_BITS(type) <= MP_DIGIT_BIT) { break; } \ b >>= ((MP_SIZEOF_BITS(type) <= MP_DIGIT_BIT) ? 0 : MP_DIGIT_BIT); \ } \ a->used = i; \ a->sign = MP_ZPOS; \ MP_ZERO_DIGITS(a->dp + a->used, a->alloc - a->used); \ } #define MP_SET_SIGNED(name, uname, type, utype) \ void name(mp_int * a, type b) \ { \ uname(a, (b < 0) ? -(utype)b : (utype)b); \ if (b < 0) { a->sign = MP_NEG; } \ } #define MP_INIT_INT(name , set, type) \ mp_err name(mp_int * a, type b) \ { \ mp_err err; \ if ((err = mp_init(a)) != MP_OKAY) { \ return err; \ } \ set(a, b); \ return MP_OKAY; \ } #define MP_GET_MAG(name, type) \ type name(const mp_int* a) \ { \ unsigned i = MP_MIN((unsigned)a->used, (unsigned)((MP_SIZEOF_BITS(type) + MP_DIGIT_BIT - 1) / MP_DIGIT_BIT)); \ type res = 0u; \ while (i --> 0u) { \ res <<= ((MP_SIZEOF_BITS(type) <= MP_DIGIT_BIT) ? 0 : MP_DIGIT_BIT); \ res |= (type)a->dp[i]; \ if (MP_SIZEOF_BITS(type) <= MP_DIGIT_BIT) { break; } \ } \ return res; \ } #define MP_GET_SIGNED(name, mag, type, utype) \ type name(const mp_int* a) \ { \ utype res = mag(a); \ return (a->sign == MP_NEG) ? (type)-res : (type)res; \ } #endif