The BIGNUM structure

 typedef struct bignum_st
        {
        int top;      /* index of last used d (most significant word) */
        BN_ULONG *d;  /* pointer to an array of 'BITS2' bit chunks */
        int max;      /* size of the d array */
        int neg;      /* sign */
        } BIGNUM;

The big number is stored in d, a malloc()ed array of BN_ULONGs, least significant first. A BN_ULONG can be either 16, 32 or 64 bits in size (BITS2), depending on the 'number of bits' specified in openssl/bn.h.

max is the size of the d array that has been allocated. top is the 'last' entry being used, so for a value of 4, bn.d[0]=4 and bn.top=1. neg is 1 if the number is negative. When a BIGNUM is 0, the d field can be NULL and top == 0.

Various routines in this library require the use of temporary BIGNUM variables during their execution. Since dynamic memory allocation to create BIGNUMs is rather expensive when used in conjunction with repeated subroutine calls, the BN_CTX structure is used. This structure contains BN_CTX_NUM BIGNUMs, see BN_CTX_start(3).

Low-level arithmetic operations

These functions are implemented in C and for several platforms in assembly language:

bn_mul_words(rp, ap, num, w) operates on the num word arrays rp and ap. It computes ap * w, places the result in rp, and returns the high word (carry).

bn_mul_add_words(rp, ap, num, w) operates on the num word arrays rp and ap. It computes ap * w + rp, places the result in rp, and returns the high word (carry).

bn_sqr_words(rp, ap, n) operates on the num word array ap and the 2*num word array ap. It computes ap * ap word-wise, and places the low and high bytes of the result in rp.

bn_div_words(h, l, d) divides the two word number (h,l) by d and returns the result.

bn_add_words(rp, ap, bp, num) operates on the num word arrays ap, bp and rp. It computes ap + bp, places the result in rp, and returns the high word (carry).

bn_sub_words(rp, ap, bp, num) operates on the num word arrays ap, bp and rp. It computes ap - bp, places the result in rp, and returns the carry (1 if bp > ap, 0 otherwise).

bn_mul_comba4(r, a, b) operates on the 4 word arrays a and b and the 8 word array r. It computes a*b and places the result in r.

bn_mul_comba8(r, a, b) operates on the 8 word arrays a and b and the 16 word array r. It computes a*b and places the result in r.

bn_sqr_comba4(r, a, b) operates on the 4 word arrays a and b and the 8 word array r.

bn_sqr_comba8(r, a, b) operates on the 8 word arrays a and b and the 16 word array r.

The following functions are implemented in C:

bn_cmp_words(a, b, n) operates on the n word arrays a and b. It returns 1, 0 and -1 if a is greater than, equal and less than b.

bn_mul_normal(r, a, na, b, nb) operates on the na word array a, the nb word array b and the na+nb word array r. It computes a*b and places the result in r.

bn_mul_low_normal(r, a, b, n) operates on the n word arrays r, a and b. It computes the n low words of a*b and places the result in r.

bn_mul_recursive(r, a, b, n2, dna, dnb, t) operates on the word arrays a and b of length n2+dna and n2+dnb (dna and dnb are currently allowed to be 0 or negative) and the 2*n2 word arrays r and t. n2 must be a power of 2. It computes a*b and places the result in r.

bn_mul_part_recursive(r, a, b, n, tna, tnb, tmp) operates on the word arrays a and b of length n+tna and n+tnb and the 4*n word arrays r and tmp.

bn_mul_low_recursive(r, a, b, n2, tmp) operates on the n2 word arrays r and tmp and the n2/2 word arrays a and b.

bn_mul_high(r, a, b, l, n2, tmp) operates on the n2 word arrays r, a, b and l (?) and the 3*n2 word array tmp.

BN_mul() calls bn_mul_normal(), or an optimized implementation if the factors have the same size: bn_mul_comba8() is used if they are 8 words long, bn_mul_recursive() if they are larger than BN_MULL_SIZE_NORMAL and the size is an exact multiple of the word size, and bn_mul_part_recursive() for others that are larger than BN_MULL_SIZE_NORMAL.

bn_sqr_normal(r, a, n, tmp) operates on the n word array a and the 2*n word arrays tmp and r.

The implementations use the following macros which, depending on the architecture, may use "long long" C operations or inline assembler. They are defined in bn_lcl.h.

mul(r, a, w, c) computes w*a+c and places the low word of the result in r and the high word in c.

mul_add(r, a, w, c) computes w*a+r+c and places the low word of the result in r and the high word in c.

sqr(r0, r1, a) computes a*a and places the low word of the result in r0 and the high word in r1.

Size changes

bn_expand() ensures that b has enough space for a bits bit number. bn_wexpand() ensures that b has enough space for an n word number. If the number has to be expanded, both macros call bn_expand2(), which allocates a new d array and copies the data. They return NULL on error, b otherwise.

The bn_fix_top() macro reduces a->top to point to the most significant non-zero word when a has shrunk.

Debugging

bn_check_top() verifies that ((a)->top >= 0 && (a)->top E<= (a)->max)>. A violation will cause the program to abort.

bn_print() prints a to stderr. bn_dump() prints n words at d (in reverse order, i.e. most significant word first) to stderr.

bn_set_max() makes a a static number with a max of its current size. This is used by bn_set_low() and bn_set_high() to make r a read-only BIGNUM that contains the n low or high words of a.

If BN_DEBUG is not defined, bn_check_top(), bn_print(), bn_dump() and bn_set_max() are defined as empty macros.