diff options
Diffstat (limited to 'openssl/src/bn.rs')
| -rw-r--r-- | openssl/src/bn.rs | 843 |
1 files changed, 843 insertions, 0 deletions
diff --git a/openssl/src/bn.rs b/openssl/src/bn.rs new file mode 100644 index 00000000..d52be884 --- /dev/null +++ b/openssl/src/bn.rs @@ -0,0 +1,843 @@ +use ffi; +use libc::c_int; +use std::cmp::Ordering; +use std::ffi::CString; +use std::{fmt, ptr}; +use std::ops::{Add, Div, Mul, Neg, Rem, Shl, Shr, Sub, Deref}; + +use {cvt, cvt_p, cvt_n}; +use crypto::CryptoString; +use error::ErrorStack; +use types::{OpenSslType, OpenSslTypeRef}; + +/// Options for the most significant bits of a randomly generated `BigNum`. +pub struct MsbOption(c_int); + +/// The most significant bit of the number may be 0. +pub const MSB_MAYBE_ZERO: MsbOption = MsbOption(-1); + +/// The most significant bit of the number must be 1. +pub const MSB_ONE: MsbOption = MsbOption(0); + +/// The most significant two bits of the number must be 1. +/// +/// The number of bits in the product of two such numbers will always be exactly twice the number +/// of bits in the original numbers. +pub const TWO_MSB_ONE: MsbOption = MsbOption(1); + +type_!(BigNumContext, BigNumContextRef, ffi::BN_CTX, ffi::BN_CTX_free); + +impl BigNumContext { + /// Returns a new `BigNumContext`. + pub fn new() -> Result<BigNumContext, ErrorStack> { + unsafe { cvt_p(ffi::BN_CTX_new()).map(BigNumContext) } + } +} + +impl BigNumRef { + /// Erases the memory used by this `BigNum`, resetting its value to 0. + /// + /// This can be used to destroy sensitive data such as keys when they are no longer needed. + pub fn clear(&mut self) { + unsafe { ffi::BN_clear(self.as_ptr()) } + } + + /// Adds a `u32` to `self`. + pub fn add_word(&mut self, w: u32) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_add_word(self.as_ptr(), w as ffi::BN_ULONG)).map(|_| ()) } + } + + /// Subtracts a `u32` from `self`. + pub fn sub_word(&mut self, w: u32) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_sub_word(self.as_ptr(), w as ffi::BN_ULONG)).map(|_| ()) } + } + + /// Multiplies a `u32` by `self`. + pub fn mul_word(&mut self, w: u32) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_mul_word(self.as_ptr(), w as ffi::BN_ULONG)).map(|_| ()) } + } + + /// Divides `self` by a `u32`, returning the remainder. + pub fn div_word(&mut self, w: u32) -> Result<u64, ErrorStack> { + unsafe { + let r = ffi::BN_div_word(self.as_ptr(), w.into()); + if r == ffi::BN_ULONG::max_value() { + Err(ErrorStack::get()) + } else { + Ok(r.into()) + } + } + } + + /// Returns the result of `self` modulo `w`. + pub fn mod_word(&self, w: u32) -> Result<u64, ErrorStack> { + unsafe { + let r = ffi::BN_mod_word(self.as_ptr(), w.into()); + if r == ffi::BN_ULONG::max_value() { + Err(ErrorStack::get()) + } else { + Ok(r.into()) + } + } + } + + /// Places a cryptographically-secure pseudo-random number nonnegative + /// number less than `self` in `rnd`. + pub fn rand_range(&self, rnd: &mut BigNumRef) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_rand_range(self.as_ptr(), rnd.as_ptr())).map(|_| ()) } + } + + /// The cryptographically weak counterpart to `rand_in_range`. + pub fn pseudo_rand_range(&self, rnd: &mut BigNumRef) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_pseudo_rand_range(self.as_ptr(), rnd.as_ptr())).map(|_| ()) } + } + + /// Sets bit `n`. Equivalent to `self |= (1 << n)`. + /// + /// When setting a bit outside of `self`, it is expanded. + pub fn set_bit(&mut self, n: i32) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_set_bit(self.as_ptr(), n.into())).map(|_| ()) } + } + + /// Clears bit `n`, setting it to 0. Equivalent to `self &= ~(1 << n)`. + /// + /// When clearing a bit outside of `self`, an error is returned. + pub fn clear_bit(&mut self, n: i32) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_clear_bit(self.as_ptr(), n.into())).map(|_| ()) } + } + + /// Returns `true` if the `n`th bit of `self` is set to 1, `false` otherwise. + pub fn is_bit_set(&self, n: i32) -> bool { + unsafe { ffi::BN_is_bit_set(self.as_ptr(), n.into()) == 1 } + } + + /// Truncates `self` to the lowest `n` bits. + /// + /// An error occurs if `self` is already shorter than `n` bits. + pub fn mask_bits(&mut self, n: i32) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_mask_bits(self.as_ptr(), n.into())).map(|_| ()) } + } + + /// Places `a << 1` in `self`. + pub fn lshift1(&mut self, a: &BigNumRef) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_lshift1(self.as_ptr(), a.as_ptr())).map(|_| ()) } + } + + /// Places `a >> 1` in `self`. + pub fn rshift1(&mut self, a: &BigNumRef) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_rshift1(self.as_ptr(), a.as_ptr())).map(|_| ()) } + } + + /// Places `a + b` in `self`. + pub fn checked_add(&mut self, a: &BigNumRef, b: &BigNumRef) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_add(self.as_ptr(), a.as_ptr(), b.as_ptr())).map(|_| ()) } + } + + /// Places `a - b` in `self`. + pub fn checked_sub(&mut self, a: &BigNumRef, b: &BigNumRef) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_sub(self.as_ptr(), a.as_ptr(), b.as_ptr())).map(|_| ()) } + } + + /// Places `a << n` in `self`. + pub fn lshift(&mut self, a: &BigNumRef, n: i32) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_lshift(self.as_ptr(), a.as_ptr(), n.into())).map(|_| ()) } + } + + /// Places `a >> n` in `self`. + pub fn rshift(&mut self, a: &BigNumRef, n: i32) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_rshift(self.as_ptr(), a.as_ptr(), n.into())).map(|_| ()) } + } + + pub fn to_owned(&self) -> Result<BigNum, ErrorStack> { + unsafe { cvt_p(ffi::BN_dup(self.as_ptr())).map(|b| BigNum::from_ptr(b)) } + } + + /// Sets the sign of `self`. + pub fn set_negative(&mut self, negative: bool) { + unsafe { ffi::BN_set_negative(self.as_ptr(), negative as c_int) } + } + + /// Compare the absolute values of `self` and `oth`. + /// + /// ``` + /// # use openssl::bn::BigNum; + /// # use std::cmp::Ordering; + /// let s = -BigNum::from_u32(8).unwrap(); + /// let o = BigNum::from_u32(8).unwrap(); + /// + /// assert_eq!(s.ucmp(&o), Ordering::Equal); + /// ``` + pub fn ucmp(&self, oth: &BigNumRef) -> Ordering { + unsafe { ffi::BN_ucmp(self.as_ptr(), oth.as_ptr()).cmp(&0) } + } + + pub fn is_negative(&self) -> bool { + self._is_negative() + } + + #[cfg(ossl10x)] + fn _is_negative(&self) -> bool { + unsafe { (*self.as_ptr()).neg == 1 } + } + + #[cfg(ossl110)] + fn _is_negative(&self) -> bool { + unsafe { ffi::BN_is_negative(self.as_ptr()) == 1 } + } + + /// Returns the number of significant bits in `self`. + pub fn num_bits(&self) -> i32 { + unsafe { ffi::BN_num_bits(self.as_ptr()) as i32 } + } + + /// Returns the size of `self` in bytes. + pub fn num_bytes(&self) -> i32 { + (self.num_bits() + 7) / 8 + } + + /// Generates a cryptographically strong pseudo-random `BigNum`, placing it in `self`. + /// + /// # Parameters + /// + /// * `bits`: Length of the number in bits. + /// * `msb`: The desired properties of the number. + /// * `odd`: If `true`, the generated number will be odd. + pub fn rand(&mut self, bits: i32, msb: MsbOption, odd: bool) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_rand(self.as_ptr(), bits.into(), msb.0, odd as c_int)).map(|_| ()) } + } + + /// The cryptographically weak counterpart to `rand`. + pub fn pseudo_rand(&mut self, bits: i32, msb: MsbOption, odd: bool) -> Result<(), ErrorStack> { + unsafe { + cvt(ffi::BN_pseudo_rand(self.as_ptr(), bits.into(), msb.0, odd as c_int)).map(|_| ()) + } + } + + /// Generates a prime number, placing it in `self`. + /// + /// # Parameters + /// + /// * `bits`: The length of the prime in bits (lower bound). + /// * `safe`: If true, returns a "safe" prime `p` so that `(p-1)/2` is also prime. + /// * `add`/`rem`: If `add` is set to `Some(add)`, `p % add == rem` will hold, where `p` is the + /// generated prime and `rem` is `1` if not specified (`None`). + pub fn generate_prime(&mut self, + bits: i32, + safe: bool, + add: Option<&BigNumRef>, + rem: Option<&BigNumRef>) + -> Result<(), ErrorStack> { + unsafe { + cvt(ffi::BN_generate_prime_ex(self.as_ptr(), + bits as c_int, + safe as c_int, + add.map(|n| n.as_ptr()).unwrap_or(ptr::null_mut()), + rem.map(|n| n.as_ptr()).unwrap_or(ptr::null_mut()), + ptr::null_mut())) + .map(|_| ()) + } + } + + /// Places the result of `a * b` in `self`. + pub fn checked_mul(&mut self, + a: &BigNumRef, + b: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_mul(self.as_ptr(), a.as_ptr(), b.as_ptr(), ctx.as_ptr())).map(|_| ()) } + } + + /// Places the result of `a / b` in `self`. + pub fn checked_div(&mut self, + a: &BigNumRef, + b: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { + cvt(ffi::BN_div(self.as_ptr(), + ptr::null_mut(), + a.as_ptr(), + b.as_ptr(), + ctx.as_ptr())) + .map(|_| ()) + } + } + + /// Places the result of `a % b` in `self`. + pub fn checked_rem(&mut self, + a: &BigNumRef, + b: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { + cvt(ffi::BN_div(ptr::null_mut(), + self.as_ptr(), + a.as_ptr(), + b.as_ptr(), + ctx.as_ptr())) + .map(|_| ()) + } + } + + /// Places the result of `a / b` in `self` and `a % b` in `rem`. + pub fn div_rem(&mut self, + rem: &mut BigNumRef, + a: &BigNumRef, + b: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { + cvt(ffi::BN_div(self.as_ptr(), + rem.as_ptr(), + a.as_ptr(), + b.as_ptr(), + ctx.as_ptr())) + .map(|_| ()) + } + } + + /// Places the result of `a²` in `self`. + pub fn sqr(&mut self, a: &BigNumRef, ctx: &mut BigNumContextRef) -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_sqr(self.as_ptr(), a.as_ptr(), ctx.as_ptr())).map(|_| ()) } + } + + /// Places the result of `a mod m` in `self`. + pub fn nnmod(&mut self, + a: &BigNumRef, + m: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { + cvt(ffi::BN_nnmod(self.as_ptr(), a.as_ptr(), m.as_ptr(), ctx.as_ptr())).map(|_| ()) + } + } + + /// Places the result of `(a + b) mod m` in `self`. + pub fn mod_add(&mut self, + a: &BigNumRef, + b: &BigNumRef, + m: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { + cvt(ffi::BN_mod_add(self.as_ptr(), a.as_ptr(), b.as_ptr(), m.as_ptr(), ctx.as_ptr())) + .map(|_| ()) + } + } + + /// Places the result of `(a - b) mod m` in `self`. + pub fn mod_sub(&mut self, + a: &BigNumRef, + b: &BigNumRef, + m: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { + cvt(ffi::BN_mod_sub(self.as_ptr(), a.as_ptr(), b.as_ptr(), m.as_ptr(), ctx.as_ptr())) + .map(|_| ()) + } + } + + /// Places the result of `(a * b) mod m` in `self`. + pub fn mod_mul(&mut self, + a: &BigNumRef, + b: &BigNumRef, + m: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { + cvt(ffi::BN_mod_mul(self.as_ptr(), a.as_ptr(), b.as_ptr(), m.as_ptr(), ctx.as_ptr())) + .map(|_| ()) + } + } + + /// Places the result of `a² mod m` in `self`. + pub fn mod_sqr(&mut self, + a: &BigNumRef, + m: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { + cvt(ffi::BN_mod_sqr(self.as_ptr(), a.as_ptr(), m.as_ptr(), ctx.as_ptr())).map(|_| ()) + } + } + + /// Places the result of `a^p` in `self`. + pub fn exp(&mut self, + a: &BigNumRef, + p: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_exp(self.as_ptr(), a.as_ptr(), p.as_ptr(), ctx.as_ptr())).map(|_| ()) } + } + + /// Places the result of `a^p mod m` in `self`. + pub fn mod_exp(&mut self, + a: &BigNumRef, + p: &BigNumRef, + m: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { + cvt(ffi::BN_mod_exp(self.as_ptr(), a.as_ptr(), p.as_ptr(), m.as_ptr(), ctx.as_ptr())) + .map(|_| ()) + } + } + + /// Places the inverse of `a` modulo `n` in `self`. + pub fn mod_inverse(&mut self, + a: &BigNumRef, + n: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { + cvt_p(ffi::BN_mod_inverse(self.as_ptr(), a.as_ptr(), n.as_ptr(), ctx.as_ptr())) + .map(|_| ()) + } + } + + /// Places the greatest common denominator of `a` and `b` in `self`. + pub fn gcd(&mut self, + a: &BigNumRef, + b: &BigNumRef, + ctx: &mut BigNumContextRef) + -> Result<(), ErrorStack> { + unsafe { cvt(ffi::BN_gcd(self.as_ptr(), a.as_ptr(), b.as_ptr(), ctx.as_ptr())).map(|_| ()) } + } + + /// Checks whether `self` is prime. + /// + /// Performs a Miller-Rabin probabilistic primality test with `checks` iterations. + /// + /// Returns `true` if `self` is prime with an error probability of less than `0.25 ^ checks`. + pub fn is_prime(&self, checks: i32, ctx: &mut BigNumContextRef) -> Result<bool, ErrorStack> { + unsafe { + cvt_n(ffi::BN_is_prime_ex(self.as_ptr(), checks.into(), ctx.as_ptr(), ptr::null_mut())) + .map(|r| r != 0) + } + } + + /// Checks whether `self` is prime with optional trial division. + /// + /// If `do_trial_division` is `true`, first performs trial division by a number of small primes. + /// Then, like `is_prime`, performs a Miller-Rabin probabilistic primality test with `checks` + /// iterations. + /// + /// # Return Value + /// + /// Returns `true` if `self` is prime with an error probability of less than `0.25 ^ checks`. + pub fn is_prime_fasttest(&self, + checks: i32, + ctx: &mut BigNumContextRef, + do_trial_division: bool) + -> Result<bool, ErrorStack> { + unsafe { + cvt_n(ffi::BN_is_prime_fasttest_ex(self.as_ptr(), + checks.into(), + ctx.as_ptr(), + do_trial_division as c_int, + ptr::null_mut())) + .map(|r| r != 0) + } + } + + /// Returns a big-endian byte vector representation of the absolute value of `self`. + /// + /// `self` can be recreated by using `new_from_slice`. + /// + /// ``` + /// # use openssl::bn::BigNum; + /// let s = -BigNum::from_u32(4543).unwrap(); + /// let r = BigNum::from_u32(4543).unwrap(); + /// + /// let s_vec = s.to_vec(); + /// assert_eq!(BigNum::from_slice(&s_vec).unwrap(), r); + /// ``` + pub fn to_vec(&self) -> Vec<u8> { + let size = self.num_bytes() as usize; + let mut v = Vec::with_capacity(size); + unsafe { + ffi::BN_bn2bin(self.as_ptr(), v.as_mut_ptr()); + v.set_len(size); + } + v + } + + /// Returns a decimal string representation of `self`. + /// + /// ``` + /// # use openssl::bn::BigNum; + /// let s = -BigNum::from_u32(12345).unwrap(); + /// + /// assert_eq!(&*s.to_dec_str().unwrap(), "-12345"); + /// ``` + pub fn to_dec_str(&self) -> Result<CryptoString, ErrorStack> { + unsafe { + let buf = try!(cvt_p(ffi::BN_bn2dec(self.as_ptr()))); + Ok(CryptoString::from_null_terminated(buf)) + } + } + + /// Returns a hexadecimal string representation of `self`. + /// + /// ``` + /// # use openssl::bn::BigNum; + /// let s = -BigNum::from_u32(0x99ff).unwrap(); + /// + /// assert_eq!(&*s.to_hex_str().unwrap(), "-99FF"); + /// ``` + pub fn to_hex_str(&self) -> Result<CryptoString, ErrorStack> { + unsafe { + let buf = try!(cvt_p(ffi::BN_bn2hex(self.as_ptr()))); + Ok(CryptoString::from_null_terminated(buf)) + } + } +} + +type_!(BigNum, BigNumRef, ffi::BIGNUM, ffi::BN_free); + +impl BigNum { + /// Creates a new `BigNum` with the value 0. + pub fn new() -> Result<BigNum, ErrorStack> { + unsafe { + ffi::init(); + let v = try!(cvt_p(ffi::BN_new())); + Ok(BigNum::from_ptr(v)) + } + } + + /// Creates a new `BigNum` with the given value. + pub fn from_u32(n: u32) -> Result<BigNum, ErrorStack> { + BigNum::new().and_then(|v| unsafe { + cvt(ffi::BN_set_word(v.as_ptr(), n as ffi::BN_ULONG)).map(|_| v) + }) + } + + /// Creates a `BigNum` from a decimal string. + pub fn from_dec_str(s: &str) -> Result<BigNum, ErrorStack> { + unsafe { + let c_str = CString::new(s.as_bytes()).unwrap(); + let mut bn = ptr::null_mut(); + try!(cvt(ffi::BN_dec2bn(&mut bn, c_str.as_ptr() as *const _))); + Ok(BigNum::from_ptr(bn)) + } + } + + /// Creates a `BigNum` from a hexadecimal string. + pub fn from_hex_str(s: &str) -> Result<BigNum, ErrorStack> { + unsafe { + let c_str = CString::new(s.as_bytes()).unwrap(); + let mut bn = ptr::null_mut(); + try!(cvt(ffi::BN_hex2bn(&mut bn, c_str.as_ptr() as *const _))); + Ok(BigNum::from_ptr(bn)) + } + } + + /// Creates a new `BigNum` from an unsigned, big-endian encoded number of arbitrary length. + /// + /// ``` + /// # use openssl::bn::BigNum; + /// let bignum = BigNum::from_slice(&[0x12, 0x00, 0x34]).unwrap(); + /// + /// assert_eq!(bignum, BigNum::from_u32(0x120034).unwrap()); + /// ``` + pub fn from_slice(n: &[u8]) -> Result<BigNum, ErrorStack> { + unsafe { + assert!(n.len() <= c_int::max_value() as usize); + cvt_p(ffi::BN_bin2bn(n.as_ptr(), n.len() as c_int, ptr::null_mut())) + .map(|p| BigNum::from_ptr(p)) + } + } +} + +impl AsRef<BigNumRef> for BigNum { + fn as_ref(&self) -> &BigNumRef { + self.deref() + } +} + +impl fmt::Debug for BigNumRef { + fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result { + match self.to_dec_str() { + Ok(s) => f.write_str(&s), + Err(e) => Err(e.into()), + } + } +} + +impl fmt::Debug for BigNum { + fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result { + match self.to_dec_str() { + Ok(s) => f.write_str(&s), + Err(e) => Err(e.into()), + } + } +} + +impl fmt::Display for BigNumRef { + fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result { + match self.to_dec_str() { + Ok(s) => f.write_str(&s), + Err(e) => Err(e.into()), + } + } +} + +impl fmt::Display for BigNum { + fn fmt(&self, f: &mut fmt::Formatter) -> fmt::Result { + match self.to_dec_str() { + Ok(s) => f.write_str(&s), + Err(e) => Err(e.into()), + } + } +} + +impl PartialEq<BigNumRef> for BigNumRef { + fn eq(&self, oth: &BigNumRef) -> bool { + self.cmp(oth) == Ordering::Equal + } +} + +impl PartialEq<BigNum> for BigNumRef { + fn eq(&self, oth: &BigNum) -> bool { + self.eq(oth.deref()) + } +} + +impl Eq for BigNumRef {} + +impl PartialEq for BigNum { + fn eq(&self, oth: &BigNum) -> bool { + self.deref().eq(oth) + } +} + +impl PartialEq<BigNumRef> for BigNum { + fn eq(&self, oth: &BigNumRef) -> bool { + self.deref().eq(oth) + } +} + +impl Eq for BigNum {} + +impl PartialOrd<BigNumRef> for BigNumRef { + fn partial_cmp(&self, oth: &BigNumRef) -> Option<Ordering> { + Some(self.cmp(oth)) + } +} + +impl PartialOrd<BigNum> for BigNumRef { + fn partial_cmp(&self, oth: &BigNum) -> Option<Ordering> { + Some(self.cmp(oth.deref())) + } +} + +impl Ord for BigNumRef { + fn cmp(&self, oth: &BigNumRef) -> Ordering { + unsafe { ffi::BN_cmp(self.as_ptr(), oth.as_ptr()).cmp(&0) } + } +} + +impl PartialOrd for BigNum { + fn partial_cmp(&self, oth: &BigNum) -> Option<Ordering> { + self.deref().partial_cmp(oth.deref()) + } +} + +impl PartialOrd<BigNumRef> for BigNum { + fn partial_cmp(&self, oth: &BigNumRef) -> Option<Ordering> { + self.deref().partial_cmp(oth) + } +} + +impl Ord for BigNum { + fn cmp(&self, oth: &BigNum) -> Ordering { + self.deref().cmp(oth.deref()) + } +} + +macro_rules! delegate { + ($t:ident, $m:ident) => { + impl<'a, 'b> $t<&'b BigNum> for &'a BigNumRef { + type Output = BigNum; + + fn $m(self, oth: &BigNum) -> BigNum { + $t::$m(self, oth.deref()) + } + } + + impl<'a, 'b> $t<&'b BigNumRef> for &'a BigNum { + type Output = BigNum; + + fn $m(self, oth: &BigNumRef) -> BigNum { + $t::$m(self.deref(), oth) + } + } + + impl<'a, 'b> $t<&'b BigNum> for &'a BigNum { + type Output = BigNum; + + fn $m(self, oth: &BigNum) -> BigNum { + $t::$m(self.deref(), oth.deref()) + } + } + } +} + +impl<'a, 'b> Add<&'b BigNumRef> for &'a BigNumRef { + type Output = BigNum; + + fn add(self, oth: &BigNumRef) -> BigNum { + let mut r = BigNum::new().unwrap(); + r.checked_add(self, oth).unwrap(); + r + } +} + +delegate!(Add, add); + +impl<'a, 'b> Sub<&'b BigNumRef> for &'a BigNumRef { + type Output = BigNum; + + fn sub(self, oth: &BigNumRef) -> BigNum { + let mut r = BigNum::new().unwrap(); + r.checked_sub(self, oth).unwrap(); + r + } +} + +delegate!(Sub, sub); + +impl<'a, 'b> Mul<&'b BigNumRef> for &'a BigNumRef { + type Output = BigNum; + + fn mul(self, oth: &BigNumRef) -> BigNum { + let mut ctx = BigNumContext::new().unwrap(); + let mut r = BigNum::new().unwrap(); + r.checked_mul(self, oth, &mut ctx).unwrap(); + r + } +} + +delegate!(Mul, mul); + +impl<'a, 'b> Div<&'b BigNumRef> for &'a BigNumRef { + type Output = BigNum; + + fn div(self, oth: &'b BigNumRef) -> BigNum { + let mut ctx = BigNumContext::new().unwrap(); + let mut r = BigNum::new().unwrap(); + r.checked_div(self, oth, &mut ctx).unwrap(); + r + } +} + +delegate!(Div, div); + +impl<'a, 'b> Rem<&'b BigNumRef> for &'a BigNumRef { + type Output = BigNum; + + fn rem(self, oth: &'b BigNumRef) -> BigNum { + let mut ctx = BigNumContext::new().unwrap(); + let mut r = BigNum::new().unwrap(); + r.checked_rem(self, oth, &mut ctx).unwrap(); + r + } +} + +delegate!(Rem, rem); + +impl<'a> Shl<i32> for &'a BigNumRef { + type Output = BigNum; + + fn shl(self, n: i32) -> BigNum { + let mut r = BigNum::new().unwrap(); + r.lshift(self, n).unwrap(); + r + } +} + +impl<'a> Shl<i32> for &'a BigNum { + type Output = BigNum; + + fn shl(self, n: i32) -> BigNum { + self.deref().shl(n) + } +} + +impl<'a> Shr<i32> for &'a BigNumRef { + type Output = BigNum; + + fn shr(self, n: i32) -> BigNum { + let mut r = BigNum::new().unwrap(); + r.rshift(self, n).unwrap(); + r + } +} + +impl<'a> Shr<i32> for &'a BigNum { + type Output = BigNum; + + fn shr(self, n: i32) -> BigNum { + self.deref().shl(n) + } +} + +impl<'a> Neg for &'a BigNumRef { + type Output = BigNum; + + fn neg(self) -> BigNum { + self.to_owned().unwrap().neg() + } +} + +impl<'a> Neg for &'a BigNum { + type Output = BigNum; + + fn neg(self) -> BigNum { + self.deref().neg() + } +} + +impl Neg for BigNum { + type Output = BigNum; + + fn neg(mut self) -> BigNum { + let negative = self.is_negative(); + self.set_negative(!negative); + self + } +} + +#[cfg(test)] +mod tests { + use bn::{BigNumContext, BigNum}; + + #[test] + fn test_to_from_slice() { + let v0 = BigNum::from_u32(10203004).unwrap(); + let vec = v0.to_vec(); + let v1 = BigNum::from_slice(&vec).unwrap(); + + assert!(v0 == v1); + } + + #[test] + fn test_negation() { + let a = BigNum::from_u32(909829283).unwrap(); + + assert!(!a.is_negative()); + assert!((-a).is_negative()); + } + + #[test] + fn test_prime_numbers() { + let a = BigNum::from_u32(19029017).unwrap(); + let mut p = BigNum::new().unwrap(); + p.generate_prime(128, true, None, Some(&a)).unwrap(); + + let mut ctx = BigNumContext::new().unwrap(); + assert!(p.is_prime(100, &mut ctx).unwrap()); + assert!(p.is_prime_fasttest(100, &mut ctx, true).unwrap()); + } +} |