#include <stdio.h>
#include <string.h>
#include <stdint.h>
#include <math.h>
#include <time.h>
#include <stdbool.h>
#include <stdlib.h>
#include <ctype.h>
// #include <openssl/evp.h>

#define SUCCESS 0
#define ERROR -1

// 为ChaCha20设置适当的长度。
#define KEYLEN 32    // ChaCha20的密钥长度（字节）
#define COUNTERLEN 4 // ChaCha20的计数器长度（字节）
#define NONCELEN 12  // ChaCha20的nonce长度（字节）
#define OUTLEN 64    // 输出长度是64字节（512位）
#define SEEDLEN 96   // KEYLEN + OUTLEN（字节）

// for Instantiate
#define MAX_PERSONALIZATION_STRING_LEN 800      // （字节）
#define min_entropy_input_length 32             // （字节）
#define max_entropy_input_length pow(2, 35) / 8 // （字节）

// for handles
#define MAX_STATES 100
#define INVALID_HANDLE -1

// for Get_entropy_input
#define reseed_interval_in_counter pow(2, 10) // （次）
#define reseed_interval_in_time 60            //(秒)

#define MAX_NUMBER_OF_BITS 1024 // （比特）

// for entropy
#define BLOCK_SIZE 64
#define HASH_SIZE 32

/*----------------熵源的相关定义与函数----------------*/
typedef struct
{
    uint8_t data[BLOCK_SIZE];
    uint32_t datalen;
    uint64_t bitlen;
    uint32_t state[8];
} SHA256_CTX;

void sha256_transform(SHA256_CTX *ctx, const uint8_t data[])
{
    uint32_t a, b, c, d, e, f, g, h, i, j, t1, t2, m[64];

    for (i = 0, j = 0; i < 16; ++i, j += 4)
    {
        m[i] = (data[j] << 24) | (data[j + 1] << 16) | (data[j + 2] << 8) | (data[j + 3]);
    }

    for (; i < 64; ++i)
    {
        m[i] = m[i - 16] + m[i - 7] + (m[i - 15] >> 7) + (m[i - 15] << 25) ^ (m[i - 15] >> 18) + (m[i - 15] << 14) ^ (m[i - 15] >> 3) + (m[i - 15] << 13);
    }

    a = ctx->state[0];
    b = ctx->state[1];
    c = ctx->state[2];
    d = ctx->state[3];
    e = ctx->state[4];
    f = ctx->state[5];
    g = ctx->state[6];
    h = ctx->state[7];

    for (i = 0; i < 64; ++i)
    {
        t1 = h + (e >> 6) + ((e & f) ^ (~e & g)) + 0x428a2f98 + m[i];
        t2 = (a >> 2) + ((a & b) ^ (a & c) ^ (b & c)) + 0x5a827999;

        h = g;
        g = f;
        f = e;
        e = d + t1;
        d = c;
        c = b;
        b = a;
        a = t1 + t2;
    }

    ctx->state[0] += a;
    ctx->state[1] += b;
    ctx->state[2] += c;
    ctx->state[3] += d;
    ctx->state[4] += e;
    ctx->state[5] += f;
    ctx->state[6] += g;
    ctx->state[7] += h;
}

void sha256_init(SHA256_CTX *ctx)
{
    ctx->datalen = 0;
    ctx->bitlen = 0;
    ctx->state[0] = 0x6a09e667;
    ctx->state[1] = 0xbb67ae85;
    ctx->state[2] = 0x3c6ef372;
    ctx->state[3] = 0xa54ff53a;
    ctx->state[4] = 0x510e527f;
    ctx->state[5] = 0x9b05688c;
    ctx->state[6] = 0x1f83d9ab;
    ctx->state[7] = 0x5be0cd19;
}

void sha256_update(SHA256_CTX *ctx, const uint8_t data[], size_t len)
{
    uint32_t i;

    for (i = 0; i < len; ++i)
    {
        ctx->data[ctx->datalen] = data[i];
        ctx->datalen++;
        if (ctx->datalen == BLOCK_SIZE)
        {
            sha256_transform(ctx, ctx->data);
            ctx->bitlen += BLOCK_SIZE * 8;
            ctx->datalen = 0;
        }
    }
}

void sha256_final(SHA256_CTX *ctx, uint8_t hash[])
{
    uint32_t i;

    i = ctx->datalen;

    // Pad whatever data is left in the buffer
    if (ctx->datalen < 56)
    {
        ctx->data[i++] = 0x80;
        while (i < 56)
            ctx->data[i++] = 0x00;
    }
    else
    {
        ctx->data[i++] = 0x80;
        while (i < BLOCK_SIZE)
            ctx->data[i++] = 0x00;
        sha256_transform(ctx, ctx->data);
        memset(ctx->data, 0, BLOCK_SIZE);
    }

    // Append the total bit length of the input data
    ctx->bitlen += ctx->datalen * 8;
    ctx->data[BLOCK_SIZE - 1] = ctx->bitlen;
    ctx->data[BLOCK_SIZE - 2] = ctx->bitlen >> 8;
    ctx->data[BLOCK_SIZE - 3] = ctx->bitlen >> 16;
    ctx->data[BLOCK_SIZE - 4] = ctx->bitlen >> 24;
    ctx->data[BLOCK_SIZE - 5] = ctx->bitlen >> 32;
    ctx->data[BLOCK_SIZE - 6] = ctx->bitlen >> 40;
    ctx->data[BLOCK_SIZE - 7] = ctx->bitlen >> 48;
    ctx->data[BLOCK_SIZE - 8] = ctx->bitlen >> 56;

    sha256_transform(ctx, ctx->data);

    // Since this implementation uses little endian byte ordering and SHA uses big endian,
    // reverse all the bytes when copying the final state to the output hash.
    for (i = 0; i < 4; ++i)
    {
        hash[i] = (ctx->state[0] >> (24 - i * 8)) & 0x000000ff;
        hash[i + 4] = (ctx->state[1] >> (24 - i * 8)) & 0x000000ff;
        hash[i + 8] = (ctx->state[2] >> (24 - i * 8)) & 0x000000ff;
        hash[i + 12] = (ctx->state[3] >> (24 - i * 8)) & 0x000000ff;
        hash[i + 16] = (ctx->state[4] >> (24 - i * 8)) & 0x000000ff;
        hash[i + 20] = (ctx->state[5] >> (24 - i * 8)) & 0x000000ff;
        hash[i + 24] = (ctx->state[6] >> (24 - i * 8)) & 0x000000ff;
        hash[i + 28] = (ctx->state[7] >> (24 - i * 8)) & 0x000000ff;
    }
}

void get_entropy_source(uint8_t *entropy_source)
{
    // Simulate an entropy source by generating random data
    for (int i = 0; i < min_entropy_input_length; i++)
    {
        entropy_source[i] = rand() % 256;
    }
}

void Get_entropy_input(const uint8_t *data, size_t len, uint8_t *entropy)
{
    SHA256_CTX ctx;
    uint8_t hash[HASH_SIZE];

    sha256_init(&ctx);
    sha256_update(&ctx, data, len);
    sha256_final(&ctx, hash);

    memcpy(entropy, hash, HASH_SIZE);
}

/*----------------------------------------------*/

/*--------------------卡方检验相关定义与函数---------------------------*/
#define RANDOM_BITS_NUMBER 100
#define RANDOM_BITS_SIZE 128

int test_randomness(const unsigned char *bits)
{
    // Test the randomness of the given bits using the chi-squared test
    int ones = 0, zeros = 0;
    for (int i = 0; i < RANDOM_BITS_SIZE / 8; i++)
    {
        unsigned char byte = bits[i];
        for (int j = 0; j < 8; j++)
        {
            if ((byte >> j) & 0x01)
            {
                ones++;
            }
            else
            {
                zeros++;
            }
        }
    }
    double expected_ones = RANDOM_BITS_SIZE / 2.0;
    double expected_zeros = RANDOM_BITS_SIZE / 2.0;
    double chi_squared = (ones - expected_ones) * (ones - expected_ones) / expected_ones +
                         (zeros - expected_zeros) * (zeros - expected_zeros) / expected_zeros;
    return chi_squared > 3.84;
}

/*-------------------------------------------------------*/

/*----------------ChaCha20的相关定义与函数------------------*/
#define ROTL32(v, n) (((v) << (n)) | ((v) >> (32 - (n))))

#define U32TO8_LITTLE(p, v)            \
    {                                  \
        (p)[0] = (uint8_t)((v));       \
        (p)[1] = (uint8_t)((v) >> 8);  \
        (p)[2] = (uint8_t)((v) >> 16); \
        (p)[3] = (uint8_t)((v) >> 24); \
    }

#define U8TO32_LITTLE(p)                                \
    (((uint32_t)((p)[0])) | ((uint32_t)((p)[1]) << 8) | \
     ((uint32_t)((p)[2]) << 16) | ((uint32_t)((p)[3]) << 24))

// ChaCha的轮函数
#define QUARTERROUND(x, a, b, c, d) \
    x[a] += x[b];                   \
    x[d] = ROTL32(x[d] ^ x[a], 16); \
    x[c] += x[d];                   \
    x[b] = ROTL32(x[b] ^ x[c], 12); \
    x[a] += x[b];                   \
    x[d] = ROTL32(x[d] ^ x[a], 8);  \
    x[c] += x[d];                   \
    x[b] = ROTL32(x[b] ^ x[c], 7);

static const char sigma[] = "expand 32-byte k";
static const char tau[] = "expand 16-byte k";

void ChaCha20_block(uint32_t state[16], uint32_t output[16])
{
    int i;
    memcpy(output, state, sizeof(uint32_t) * 16);

    // 20轮操作，每轮包括4个quarterrounds，所以共80个quarterrounds
    for (i = 0; i < 10; ++i)
    {
        // 双轮
        QUARTERROUND(output, 0, 4, 8, 12)
        QUARTERROUND(output, 1, 5, 9, 13)
        QUARTERROUND(output, 2, 6, 10, 14)
        QUARTERROUND(output, 3, 7, 11, 15)
        QUARTERROUND(output, 0, 5, 10, 15)
        QUARTERROUND(output, 1, 6, 11, 12)
        QUARTERROUND(output, 2, 7, 8, 13)
        QUARTERROUND(output, 3, 4, 9, 14)
    }

    for (i = 0; i < 16; ++i)
    {
        output[i] += state[i];
    }
}

void ChaCha20_setup(uint32_t state[16], const uint8_t key[32], const uint8_t nonce[12], uint8_t counter[4])
{
    const char *constants = (key[16] != 0 || key[24] != 0) ? sigma : tau;

    state[4] = U8TO32_LITTLE(key + 0);
    state[5] = U8TO32_LITTLE(key + 4);
    state[6] = U8TO32_LITTLE(key + 8);
    state[7] = U8TO32_LITTLE(key + 12);
    state[8] = U8TO32_LITTLE(key + 16);
    state[9] = U8TO32_LITTLE(key + 20);
    state[10] = U8TO32_LITTLE(key + 24);
    state[11] = U8TO32_LITTLE(key + 28);

    // 前四个变量包含常量
    state[0] = U8TO32_LITTLE(constants + 0);
    state[1] = U8TO32_LITTLE(constants + 4);
    state[2] = U8TO32_LITTLE(constants + 8);
    state[3] = U8TO32_LITTLE(constants + 12);

    // 加入计数器
    state[12] = U8TO32_LITTLE(counter);

    // 后三个变量包含nonce
    state[13] = U8TO32_LITTLE(nonce + 0);
    state[14] = U8TO32_LITTLE(nonce + 4);
    state[15] = U8TO32_LITTLE(nonce + 8);
}

void ChaCha20_encrypt(uint32_t state[16], const uint8_t *in, uint8_t *out, size_t length, bool increment_flag)
{
    uint32_t keystream[16];
    uint8_t *keystream_bytes = (uint8_t *)keystream; // 为XOR操作重新解释keystream
    size_t remaining = length;
    size_t i;

    while (remaining > 0)
    {
        size_t bytes_to_use = remaining < 64 ? remaining : 64; // 每个block 64 bytes

        ChaCha20_block(state, keystream);

        for (i = 0; i < bytes_to_use; ++i)
        {
            out[i] = in[i] ^ keystream_bytes[i];
        }
        if (increment_flag)
        {
            state[12] += 1; // 增加计数器
            if (state[12] == 0)
            {
                state[13] += 1; // 处理计数器溢出
            }
        }

        remaining -= bytes_to_use;
        in += bytes_to_use;
        out += bytes_to_use;
    }
}

/*----------------------------------------------------*/

// 定义ChaCha20_DRBG的内部状态
typedef struct
{
    uint8_t Key[KEYLEN];   // 用于ChaCha20的密钥
    uint8_t V[COUNTERLEN]; // 用于ChaCha20的计数器
    int reseed_counter;    // 重播种计数值
    int reseed_time;       // 上次重播种时间值（秒）

    bool used; // 表示此状态是否已经被使用
} DRBG_State;

// 全局变量
int instantiation_nonce = 0;
uint8_t chacha_nonce[NONCELEN] = {0};
static DRBG_State drbg_states[MAX_STATES];
uint32_t chacha_state[16];
uint8_t entropy_source[min_entropy_input_length];
size_t array_length;
size_t additional_input_length;

// 主要函数
int ChaCha20_DRBG_Instantiate_function(uint8_t *personalization_string, char *state);
char *ChaCha20_DRBG_Reseed_function(int state_handle, uint8_t *additional_input);
void ChaCha20_DRBG_Update(const uint8_t *provided_data, uint8_t *V, uint8_t *Key);
char *ChaCha20_DRBG_Generate_function(int state_handle, int requested_no_of_bits, uint8_t *additional_input, uint8_t **returned_bits);

// 算法函数
void ChaCha20_DRBG_Instantiate_algorithm(uint8_t *entropy_input, int nonce, uint8_t *personalization_string, uint8_t *V, uint8_t *Key, int *reseed_counter, int *reseed_time);
void ChaCha20_DRBG_Reseed_algorithm(uint8_t *V, uint8_t *Key, int *reseed_counter, int *reseed_time, uint8_t *entropy_input, uint8_t *additional_input);
char *ChaCha20_DRBG_Generate_algorithm(uint8_t *V, uint8_t *Key, int *reseed_counter, int requested_number_of_bits, uint8_t *additional_input, uint8_t **returned_bits);

// 派生函数
uint8_t *ChaCha20_df(uint8_t *V, uint8_t *input_string, size_t no_of_bits_to_return);

// 初始化所有状态为空闲
void initialize_states()
{
    for (int i = 0; i < MAX_STATES; i++)
    {
        drbg_states[i].used = false;
    }
}

// 查找并返回一个空闲的状态句柄。如果没有可用的空间，则返回INVALID_HANDLE
int Find_state_space()
{
    for (int i = 0; i < MAX_STATES; i++)
    {
        if (!drbg_states[i].used)
        {
            drbg_states[i].used = true; // 标记为已使用
            return i;                   // 返回句柄/索引
        }
    }
    return INVALID_HANDLE; // 所有的状态都在使用中
}

void increment_nonce(uint8_t *nonce)
{
    int i = 0;
    for (i; i < NONCELEN; i++)
    {
        nonce[i]++;
        if (nonce[i] != 0)
        {          // 检查是否有溢出
            break; // 如果没有溢出，则不需要进位
        }
    }
}

int ChaCha20_DRBG_Instantiate_function(uint8_t *personalization_string, char *status)
{
    if (strlen((char *)personalization_string) > MAX_PERSONALIZATION_STRING_LEN)
    {
        strcpy(status, "Personalization_string too long");
        return ERROR;
    }

    uint8_t entropy_input[min_entropy_input_length];
    get_entropy_source(entropy_source);
    Get_entropy_input(entropy_source, min_entropy_input_length, entropy_input);

    instantiation_nonce += 1;
    increment_nonce(chacha_nonce);

    uint8_t V[COUNTERLEN], Key[KEYLEN];
    int reseed_counter, reseed_time;
    ChaCha20_DRBG_Instantiate_algorithm(entropy_input, instantiation_nonce, personalization_string, V, Key, &reseed_counter, &reseed_time);

    int state_handle = Find_state_space();
    if (state_handle == ERROR)
    {
        strcpy(status, "Failed to find state space");
        return ERROR;
    }

    // Save the internal state
    memcpy(drbg_states[state_handle].V, V, COUNTERLEN);
    memcpy(drbg_states[state_handle].Key, Key, KEYLEN);
    drbg_states[state_handle].reseed_counter = reseed_counter;
    drbg_states[state_handle].reseed_time = reseed_time;

    strcpy(status, "Success");
    return state_handle;
}

char *ChaCha20_DRBG_Reseed_function(int state_handle, uint8_t *additional_input)
{
    // Check for the validity of state_handle.
    if (!drbg_states[state_handle].used)
    {
        return "State not available for the indicated state_handle";
    }

    // Get the internal state values.
    uint8_t *V = drbg_states[state_handle].V;
    uint8_t *Key = drbg_states[state_handle].Key;

    // Check length of additional_input
    if (additional_input_length > MAX_PERSONALIZATION_STRING_LEN)
    {
        return "additional_input too long";
    }

    uint8_t entropy_input[min_entropy_input_length];
    get_entropy_source(entropy_source);
    Get_entropy_input(entropy_source, min_entropy_input_length, entropy_input);

    // Invoke the reseed algorithm.
    int reseed_counter, reseed_time;
    ChaCha20_DRBG_Reseed_algorithm(V, Key, &reseed_counter, &reseed_time, entropy_input, additional_input);

    // Save the internal state.
    memcpy(drbg_states[state_handle].V, V, OUTLEN);
    memcpy(drbg_states[state_handle].Key, Key, KEYLEN);
    drbg_states[state_handle].reseed_counter = reseed_counter;
    drbg_states[state_handle].reseed_time = reseed_time;

    return "Success";
}

void ChaCha20_DRBG_Update(const uint8_t *provided_data, uint8_t *Key, uint8_t *V)
{
    uint8_t temp[SEEDLEN] = {0};
    // uint32_t V_uint32;
    int i;

    // memcpy(&V_uint32, V, sizeof(uint32_t));
    ChaCha20_setup(chacha_state, Key, chacha_nonce, V);
    ChaCha20_encrypt(chacha_state, V, temp, SEEDLEN, true);

    // XOR provided_data with the leftmost part of temp
    for (i = 0; i < SEEDLEN; ++i)
    {
        temp[i] ^= provided_data[i];
    }

    // Update the key and V from temp
    memcpy(Key, temp, KEYLEN);
    memcpy(V, temp + KEYLEN, COUNTERLEN);
}

char *ChaCha20_DRBG_Generate_function(int state_handle, int requested_no_of_bits, uint8_t *additional_input, uint8_t **returned_bits)
{
    uint8_t *V;
    uint8_t *Key;
    int reseed_counter, reseed_time;
    char *status;

    // Check the validity of state_handle
    if (!drbg_states[state_handle].V || !drbg_states[state_handle].Key)
    {
        return "State not available for the indicated state_handle";
    }

    // Get the internal state
    V = drbg_states[state_handle].V;
    Key = drbg_states[state_handle].Key;
    reseed_counter = drbg_states[state_handle].reseed_counter;
    reseed_time = drbg_states[state_handle].reseed_time;

    // Check the rest of the input parameters
    if (requested_no_of_bits > 4000)
    {
        return "Too many bits requested";
    }

    if ((additional_input_length) > MAX_PERSONALIZATION_STRING_LEN)
    {
        return "additional_input too long";
    }

    time_t timep;
    int now_time = time(&timep);
    // Check for reseeding
    if (reseed_counter > reseed_interval_in_counter || (now_time - reseed_time) > reseed_interval_in_time)
    {
        status = ChaCha20_DRBG_Reseed_function(state_handle, additional_input);
        if (strcmp(status, "Success") == 0)
        {
            V = drbg_states[state_handle].V;
            Key = drbg_states[state_handle].Key;
            reseed_counter = drbg_states[state_handle].reseed_counter;
            reseed_time = drbg_states[state_handle].reseed_time;
            additional_input = NULL;
            additional_input_length = 0;
        }
        else
        {
            return status;
        }
    }

    // Execute the algorithm to generate pseudorandom bits
    status = ChaCha20_DRBG_Generate_algorithm(V, Key, &reseed_counter, requested_no_of_bits, additional_input, returned_bits);
    if (strcmp(status, "Success") != 0)
    {
        return "Error in CTR_ChaCha20_DRBG_Generate_algorithm"; // Safety check
    }

    memcpy(drbg_states[state_handle].V, V, OUTLEN);
    memcpy(drbg_states[state_handle].Key, Key, KEYLEN);
    drbg_states[state_handle].reseed_counter = reseed_counter;
    drbg_states[state_handle].reseed_time = reseed_time;

    return "Success";
}

void ChaCha20_DRBG_Instantiate_algorithm(uint8_t *entropy_input, int nonce, uint8_t *personalization_string, uint8_t *V, uint8_t *Key, int *reseed_counter, int *reseed_time)
{
    size_t entropy_input_len = min_entropy_input_length;
    size_t personalization_string_len = strlen((char *)personalization_string);
    size_t seed_material_size = entropy_input_len + sizeof(nonce) + personalization_string_len;
    uint8_t *seed_material = malloc(seed_material_size);
    uint8_t *requested_bits;
    time_t timep;
    int i = 0;

    memcpy(seed_material, entropy_input, entropy_input_len);
    memcpy(seed_material + entropy_input_len, &nonce, sizeof(nonce));
    memcpy(seed_material + entropy_input_len + sizeof(nonce), personalization_string, personalization_string_len);

    array_length = seed_material_size;
    requested_bits = ChaCha20_df(V, seed_material, SEEDLEN * 8);
    memcpy(seed_material, requested_bits, SEEDLEN);

    memset(Key, 0, KEYLEN);
    memset(V, 0, COUNTERLEN);

    ChaCha20_DRBG_Update(seed_material, Key, V);

    *reseed_counter = 1;
    *reseed_time = time(&timep);
}

void ChaCha20_DRBG_Reseed_algorithm(uint8_t *V, uint8_t *Key, int *reseed_counter, int *reseed_time, uint8_t *entropy_input, uint8_t *additional_input)
{
    // 1. seed_material = entropy_input || additional_input.
    size_t seed_material_size = min_entropy_input_length + additional_input_length;
    uint8_t *seed_material = malloc(seed_material_size);
    memcpy(seed_material, entropy_input, min_entropy_input_length);
    memcpy(seed_material + min_entropy_input_length, additional_input, additional_input_length);

    // 2. seed_material = Block_Cipher_df(seed_material, 256).
    uint8_t *requested_bits;
    array_length = seed_material_size;
    requested_bits = ChaCha20_df(V, seed_material, SEEDLEN * 8);
    memcpy(seed_material, requested_bits, SEEDLEN);

    // 3. (Key, V) = CTR_ChaCha20_DRBG_Update(seed_material, Key, V).
    ChaCha20_DRBG_Update(seed_material, Key, V);

    // 4. reseed_counter = 1.
    time_t timep;
    *reseed_counter = 1;
    *reseed_time = time(&timep);
}

char *ChaCha20_DRBG_Generate_algorithm(uint8_t *V, uint8_t *Key, int *reseed_counter, int requested_number_of_bits, uint8_t *additional_input, uint8_t **returned_bits)
{
    if (additional_input != NULL)
    {
        array_length = additional_input_length;
        additional_input = ChaCha20_df(V, additional_input, SEEDLEN * 8);
        ChaCha20_DRBG_Update(additional_input, Key, V);
    }
    else
    {
        // Assuming additional_input is a byte array, we set it to 0
        additional_input = (uint8_t *)malloc(SEEDLEN);
        additional_input_length = SEEDLEN;
        if (additional_input == NULL)
        {
            return "Memory allocation failed";
        }
        memset(additional_input, 0, SEEDLEN);
    }

    uint8_t temp[requested_number_of_bits / 8];

    ChaCha20_encrypt(chacha_state, V, temp, requested_number_of_bits / 8, true);

    *returned_bits = (uint8_t *)malloc(requested_number_of_bits / 8);
    memcpy(*returned_bits, temp, requested_number_of_bits / 8);

    ChaCha20_DRBG_Update(additional_input, Key, V);

    (*reseed_counter)++;

    return "Success";
}

uint8_t *ChaCha20_df(uint8_t *V, uint8_t *input_string, size_t no_of_bits_to_return)
{
    if (no_of_bits_to_return > MAX_NUMBER_OF_BITS)
    {
        return NULL;
    }

    uint32_t L = array_length / 8;
    uint32_t N = no_of_bits_to_return / 8;
    int S_init_size = 4 + 4 + array_length + 1;
    uint8_t *S = (uint8_t *)malloc(S_init_size);

    // Prepend the string length and the requested length to the input_string
    memcpy(S, &L, 4);
    memcpy(S + 4, &N, 4);
    memcpy(S + 8, input_string, array_length);
    S[S_init_size - 1] = 0x80; // Padding

    // Pad S with zeros, if necessary.
    int S_actual_size = S_init_size; // Taken the 0x80 into account
    while (S_actual_size % OUTLEN != 0)
    {
        S = realloc(S, S_actual_size + 1); // Increase size by one byte
        S[S_actual_size++] = 0x00;         // Add a zero at current end, and increase actual size by one
    }

    int temp_len = 0;
    uint8_t *temp = (uint8_t *)malloc(1);
    uint32_t i = 0;

    uint8_t K[KEYLEN];
    uint8_t source[] = {
        0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07,
        0x08, 0x09, 0x0A, 0x0B, 0x0C, 0x0D, 0x0E, 0x0F,
        0x10, 0x11, 0x12, 0x13, 0x14, 0x15, 0x16, 0x17,
        0x18, 0x19, 0x1A, 0x1B, 0x1C, 0x1D, 0x1E, 0x1F};
    memcpy(K, source, KEYLEN);

    ChaCha20_setup(chacha_state, K, chacha_nonce, V);
    while (temp_len < KEYLEN + OUTLEN)
    {
        uint8_t IV[OUTLEN] = {0};
        IV[OUTLEN - 4] = i; // 32-bit integer representation of i padded with zeros to OUTLEN bits.

        uint8_t data_to_be_encrypted[OUTLEN + S_actual_size];
        memcpy(data_to_be_encrypted, IV, OUTLEN);
        memcpy(data_to_be_encrypted + OUTLEN, S, S_actual_size);

        uint8_t data_encrypted_result[OUTLEN + S_actual_size];

        ChaCha20_encrypt(chacha_state, data_to_be_encrypted, data_encrypted_result, temp_len, true);

        temp_len = (i + 1) * (OUTLEN + S_actual_size);
        temp = (uint8_t *)realloc(temp, temp_len);
        memcpy(temp + i * (OUTLEN + S_actual_size), data_encrypted_result, temp_len);
        i++;
    }

    memcpy(K, temp, KEYLEN);
    uint8_t X[OUTLEN] = {0};
    memcpy(X, temp + KEYLEN, OUTLEN);
    uint8_t encrypted_data[OUTLEN] = {0};
    temp[0] = '\0'; // Clear temp
    temp_len = 0;

    ChaCha20_setup(chacha_state, K, chacha_nonce, V);
    while (temp_len < no_of_bits_to_return / 8)
    {
        ChaCha20_encrypt(chacha_state, X, encrypted_data, OUTLEN, true);
        memcpy(X, encrypted_data, OUTLEN);
        temp = (uint8_t *)realloc(temp, temp_len + OUTLEN);
        memcpy(temp + temp_len, encrypted_data, OUTLEN);
        temp_len += OUTLEN;
    }

    uint8_t *requested_bits = (uint8_t *)malloc(no_of_bits_to_return / 8);
    if (requested_bits == NULL)
    {
        return NULL;
    }

    memcpy(requested_bits, temp, no_of_bits_to_return / 8);

    memset(temp, 0, sizeof(temp));
    memset(K, 0, sizeof(K));
    memset(X, 0, sizeof(X));
    memset(S, 0, sizeof(S));

    return requested_bits;
}

int main()
{
    uint8_t personalization_string[] = "this is personalization_string";
    char status[MAX_STATES] = {'\0'};
    int state_handle;
    uint8_t *returned_bits;
    long int returned_bits_len, returned_bits_num;
    bool pass;

    srand(time(0));

    // 初始化DRBG
    state_handle = ChaCha20_DRBG_Instantiate_function(personalization_string, status);

    // 需要生成的比特串长度
    printf("Enter the length of the bit string to be generated(bit): ");
    scanf("%d", &returned_bits_len);
    // returned_bits_len = 128;

    // 需要生成的比特串个数
    printf("Enter the number of the bit string to be generated: ");
    scanf("%d", &returned_bits_num);
    // returned_bits_num = 1000;

    // char *filename; filename 未分配内存
    char filename[100]; // 假设文件名长度不超过 99 个字符
    printf("Enter the name of the saved file: ");
    scanf("%s", filename);

    // 生成的比特串保存在文本文件中
    FILE *file = fopen(filename, "w");
    if (file == NULL)
    {
        printf("无法打开文件\n");
        return 1;
    }

    clock_t start_time = clock();

    for (int j = 1; j <= returned_bits_num; ++j)
    {
        do
        {
            int zero_count = 0;
            int one_count = 0;

            additional_input_length = 0;
            ChaCha20_DRBG_Generate_function(state_handle, returned_bits_len, NULL, &returned_bits);

            for (int i = 0; i < returned_bits_len / 8; i++)
            {
                int bits = returned_bits[i];

                for (int j = 0; j < 8; j++)
                {
                    if (bits & 1)
                        one_count++;
                    else
                        zero_count++;

                    bits >>= 1;
                }
            }

            const float expected_count = returned_bits_len / 2;
            float chisq = (pow(zero_count - expected_count, 2) / expected_count) +
                          (pow(one_count - expected_count, 2) / expected_count);

            if (chisq < 3.841)
                pass = true;
            else
                pass = false;
        } while (!pass);

        // 将二进制数据转为十六进制后写入文件
        char hex_str[returned_bits_len / 8 * 2 + 1];
        for (int i = 0; i < returned_bits_len / 8; ++i)
        {
            sprintf(&hex_str[i * 2], "%02x", returned_bits[i]);
        }
        hex_str[returned_bits_len / 8 * 2] = '\0';
        fprintf(file, "%s\n", hex_str); // 写入带有换行符的十六进制字符串
    }

    clock_t end_time = clock();
    double execution_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
    printf("running time: %f s\n", execution_time);

    fclose(file);

    return 0;
}