/*-*- mode:c;indent-tabs-mode:nil;c-basic-offset:4;tab-width:8;coding:utf-8 -*-│
│vi: set net ft=c ts=4 sts=4 sw=4 fenc=utf-8                                :vi│
╚──────────────────────────────────────────────────────────────────────────────╝
│                                                                              │
│ Argon2 reference source code package - reference C implementations           │
│                                                                              │
│ Copyright 2015                                                               │
│ Daniel Dinu, Dmitry Khovratovich, Jean-Philippe Aumasson, and Samuel Neves   │
│                                                                              │
│ You may use this work under the terms of a Creative Commons CC0 1.0          │
│ License/Waiver or the Apache Public License 2.0, at your option. The         │
│ terms of these licenses can be found at:                                     │
│                                                                              │
│ - CC0 1.0 Universal : https://creativecommons.org/publicdomain/zero/1.0      │
│ - Apache 2.0        : https://www.apache.org/licenses/LICENSE-2.0            │
│                                                                              │
╚─────────────────────────────────────────────────────────────────────────────*/
#include "libc/mem/mem.h"
#include "third_party/argon2/blake2-impl.h"
#include "third_party/argon2/blake2.h"
#include "third_party/argon2/core.h"

asm(".ident\t\"\\n\\n\
argon2 (CC0 or Apache2)\\n\
Copyright 2016 Daniel Dinu, Dmitry Khovratovich\\n\
Copyright 2016 Jean-Philippe Aumasson, Samuel Neves\"");
/* clang-format off */

int FLAG_clear_internal_memory = 1;

void init_block_value(block *b, uint8_t in) {
    memset(b->v, in, sizeof(b->v));
}

void copy_block(block *dst, const block *src) {
    memcpy(dst->v, src->v, sizeof(uint64_t) * ARGON2_QWORDS_IN_BLOCK);
}

optimizespeed void xor_block(block *dst, const block *src) {
    int i;
    for (i = 0; i < ARGON2_QWORDS_IN_BLOCK; ++i) {
        dst->v[i] ^= src->v[i];
    }
}

static void load_block(block *dst, const void *input) {
    unsigned i;
    for (i = 0; i < ARGON2_QWORDS_IN_BLOCK; ++i) {
        dst->v[i] = load64((const uint8_t *)input + i * sizeof(dst->v[i]));
    }
}

static void store_block(void *output, const block *src) {
    unsigned i;
    for (i = 0; i < ARGON2_QWORDS_IN_BLOCK; ++i) {
        store64((uint8_t *)output + i * sizeof(src->v[i]), src->v[i]);
    }
}

/**
 * Allocates memory to the given pointer, uses the appropriate allocator as
 * specified in the context. Total allocated memory is num*size.
 * @param context argon2_context which specifies the allocator
 * @param memory pointer to the pointer to the memory
 * @param size the size in bytes for each element to be allocated
 * @param num the number of elements to be allocated
 * @return ARGON2_OK if @memory is a valid pointer and memory is allocated
 */
int allocate_memory(const argon2_context *context, uint8_t **memory,
                    size_t num, size_t size) {
    size_t memory_size = num*size;
    if (memory == NULL) {
        return ARGON2_MEMORY_ALLOCATION_ERROR;
    }
    /* 1. Check for multiplication overflow */
    if (size != 0 && memory_size / size != num) {
        return ARGON2_MEMORY_ALLOCATION_ERROR;
    }
    /* 2. Try to allocate with appropriate allocator */
    if (context->allocate_cbk) {
        (context->allocate_cbk)(memory, memory_size);
    } else {
        *memory = malloc(memory_size);
    }
    if (*memory == NULL) {
        return ARGON2_MEMORY_ALLOCATION_ERROR;
    }
    return ARGON2_OK;
}

/**
 * Frees memory at the given pointer, uses the appropriate deallocator as
 * specified in the context. Also cleans the memory using clear_internal_memory.
 * @param context argon2_context which specifies the deallocator
 * @param memory pointer to buffer to be freed
 * @param size the size in bytes for each element to be deallocated
 * @param num the number of elements to be deallocated
 */
void free_memory(const argon2_context *context, uint8_t *memory,
                 size_t num, size_t size) {
    size_t memory_size = num*size;
    clear_internal_memory(memory, memory_size);
    if (context->free_cbk) {
        (context->free_cbk)(memory, memory_size);
    } else {
        free(memory);
    }
}

/**
 * Function that securely clears the memory if
 * FLAG_clear_internal_memory is set. If the flag isn't set, this
 * function does nothing.
 *
 * @param mem Pointer to the memory
 * @param s Memory size in bytes
 */
void clear_internal_memory(void *v, size_t n) {
  if (FLAG_clear_internal_memory && v) {
    explicit_bzero(v, n);
  }
}

/**
 * XORing the last block of each lane, hashing it, making the tag.
 * Deallocates the memory.
 *
 * @param context current Argon2 context (use only the out parameters from it)
 * @param instance Pointer to current instance of Argon2
 * @pre instance->state must point to necessary amount of memory
 * @pre context->out must point to outlen bytes of memory
 * @pre if context->free_cbk is not NULL, it should point to a function that
 * deallocates memory
 */
void finalize(const argon2_context *context, argon2_instance_t *instance) {
    if (context != NULL && instance != NULL) {
        block blockhash;
        uint32_t l;

        copy_block(&blockhash, instance->memory + instance->lane_length - 1);

        /* XOR the last blocks */
        for (l = 1; l < instance->lanes; ++l) {
            uint32_t last_block_in_lane =
                l * instance->lane_length + (instance->lane_length - 1);
            xor_block(&blockhash, instance->memory + last_block_in_lane);
        }

        /* Hash the result */
        {
            uint8_t blockhash_bytes[ARGON2_BLOCK_SIZE];
            store_block(blockhash_bytes, &blockhash);
            blake2b_long(context->out, context->outlen, blockhash_bytes,
                         ARGON2_BLOCK_SIZE);
            /* clear blockhash and blockhash_bytes */
            clear_internal_memory(blockhash.v, ARGON2_BLOCK_SIZE);
            clear_internal_memory(blockhash_bytes, ARGON2_BLOCK_SIZE);
        }

        free_memory(context, (uint8_t *)instance->memory,
                    instance->memory_blocks, sizeof(block));
    }
}

/**
 * Computes absolute position of reference block in the lane following a skewed
 * distribution and using a pseudo-random value as input
 * @param instance Pointer to the current instance
 * @param position Pointer to the current position
 * @param pseudo_rand 32-bit pseudo-random value used to determine the position
 * @param same_lane Indicates if the block will be taken from the current lane.
 * If so we can reference the current segment
 * @pre All pointers must be valid
 */
uint32_t index_alpha(const argon2_instance_t *instance,
                     const argon2_position_t *position, uint32_t pseudo_rand,
                     int same_lane) {
    /*
     * Pass 0:
     *      This lane : all already finished segments plus already constructed
     * blocks in this segment
     *      Other lanes : all already finished segments
     * Pass 1+:
     *      This lane : (SYNC_POINTS - 1) last segments plus already constructed
     * blocks in this segment
     *      Other lanes : (SYNC_POINTS - 1) last segments
     */
    uint32_t reference_area_size;
    uint64_t relative_position;
    uint32_t start_position, absolute_position;

    if (0 == position->pass) {
        /* First pass */
        if (0 == position->slice) {
            /* First slice */
            reference_area_size =
                position->index - 1; /* all but the previous */
        } else {
            if (same_lane) {
                /* The same lane => add current segment */
                reference_area_size =
                    position->slice * instance->segment_length +
                    position->index - 1;
            } else {
                reference_area_size =
                    position->slice * instance->segment_length +
                    ((position->index == 0) ? (-1) : 0);
            }
        }
    } else {
        /* Second pass */
        if (same_lane) {
            reference_area_size = instance->lane_length -
                                  instance->segment_length + position->index -
                                  1;
        } else {
            reference_area_size = instance->lane_length -
                                  instance->segment_length +
                                  ((position->index == 0) ? (-1) : 0);
        }
    }

    /* 1.2.4. Mapping pseudo_rand to 0..<reference_area_size-1> and produce
     * relative position */
    relative_position = pseudo_rand;
    relative_position = relative_position * relative_position >> 32;
    relative_position = reference_area_size - 1 -
                        (reference_area_size * relative_position >> 32);

    /* 1.2.5 Computing starting position */
    start_position = 0;

    if (0 != position->pass) {
        start_position = (position->slice == ARGON2_SYNC_POINTS - 1)
                             ? 0
                             : (position->slice + 1) * instance->segment_length;
    }

    /* 1.2.6. Computing absolute position */
    absolute_position = (start_position + relative_position) %
                        instance->lane_length; /* absolute position */
    return absolute_position;
}

/* Single-threaded version for p=1 case */
static int fill_memory_blocks_st(argon2_instance_t *instance) {
    uint32_t r, s, l;

    for (r = 0; r < instance->passes; ++r) {
        for (s = 0; s < ARGON2_SYNC_POINTS; ++s) {
            for (l = 0; l < instance->lanes; ++l) {
                argon2_position_t position = {r, l, (uint8_t)s, 0};
                fill_segment(instance, position);
            }
        }
    }
    return ARGON2_OK;
}

#if !defined(ARGON2_NO_THREADS)

#ifdef _WIN32
static unsigned __stdcall fill_segment_thr(void *thread_data)
#else
static void *fill_segment_thr(void *thread_data)
#endif
{
    argon2_thread_data *my_data = thread_data;
    fill_segment(my_data->instance_ptr, my_data->pos);
    argon2_thread_exit();
    return 0;
}

/* Multi-threaded version for p > 1 case */
static int fill_memory_blocks_mt(argon2_instance_t *instance) {
    uint32_t r, s;
    argon2_thread_handle_t *thread = NULL;
    argon2_thread_data *thr_data = NULL;
    int rc = ARGON2_OK;

    /* 1. Allocating space for threads */
    thread = calloc(instance->lanes, sizeof(argon2_thread_handle_t));
    if (thread == NULL) {
        rc = ARGON2_MEMORY_ALLOCATION_ERROR;
        goto fail;
    }

    thr_data = calloc(instance->lanes, sizeof(argon2_thread_data));
    if (thr_data == NULL) {
        rc = ARGON2_MEMORY_ALLOCATION_ERROR;
        goto fail;
    }

    for (r = 0; r < instance->passes; ++r) {
        for (s = 0; s < ARGON2_SYNC_POINTS; ++s) {
            uint32_t l, ll;

            /* 2. Calling threads */
            for (l = 0; l < instance->lanes; ++l) {
                argon2_position_t position;

                /* 2.1 Join a thread if limit is exceeded */
                if (l >= instance->threads) {
                    if (argon2_thread_join(thread[l - instance->threads])) {
                        rc = ARGON2_THREAD_FAIL;
                        goto fail;
                    }
                }

                /* 2.2 Create thread */
                position.pass = r;
                position.lane = l;
                position.slice = (uint8_t)s;
                position.index = 0;
                thr_data[l].instance_ptr =
                    instance; /* preparing the thread input */
                memcpy(&(thr_data[l].pos), &position,
                       sizeof(argon2_position_t));
                if (argon2_thread_create(&thread[l], &fill_segment_thr,
                                         (void *)&thr_data[l])) {
                    /* Wait for already running threads */
                    for (ll = 0; ll < l; ++ll)
                        argon2_thread_join(thread[ll]);
                    rc = ARGON2_THREAD_FAIL;
                    goto fail;
                }

                /* fill_segment(instance, position); */
                /*Non-thread equivalent of the lines above */
            }

            /* 3. Joining remaining threads */
            for (l = instance->lanes - instance->threads; l < instance->lanes;
                 ++l) {
                if (argon2_thread_join(thread[l])) {
                    rc = ARGON2_THREAD_FAIL;
                    goto fail;
                }
            }
        }
    }

fail:
    if (thread != NULL) {
        free(thread);
    }
    if (thr_data != NULL) {
        free(thr_data);
    }
    return rc;
}

#endif /* ARGON2_NO_THREADS */

/**
 * Function that fills the entire memory t_cost times based on the first two
 * blocks in each lane
 * @param instance Pointer to the current instance
 * @return ARGON2_OK if successful, @context->state
 */
int fill_memory_blocks(argon2_instance_t *instance) {
	if (instance == NULL || instance->lanes == 0) {
	    return ARGON2_INCORRECT_PARAMETER;
    }
#if defined(ARGON2_NO_THREADS)
    return fill_memory_blocks_st(instance);
#else
    return instance->threads == 1 ?
			fill_memory_blocks_st(instance) : fill_memory_blocks_mt(instance);
#endif
}

/**
 * Function that validates all inputs against predefined restrictions
 * and return an error code.
 *
 * @param context Pointer to current Argon2 context
 * @return ARGON2_OK if everything is all right, otherwise one of error
 *     codes (all defined in argon.h)
 */
int validate_inputs(const argon2_context *context) {
    if (NULL == context) {
        return ARGON2_INCORRECT_PARAMETER;
    }

    if (NULL == context->out) {
        return ARGON2_OUTPUT_PTR_NULL;
    }

    /* Validate output length */
    if (ARGON2_MIN_OUTLEN > context->outlen) {
        return ARGON2_OUTPUT_TOO_SHORT;
    }

    if (ARGON2_MAX_OUTLEN < context->outlen) {
        return ARGON2_OUTPUT_TOO_LONG;
    }

    /* Validate password (required param) */
    if (NULL == context->pwd) {
        if (0 != context->pwdlen) {
            return ARGON2_PWD_PTR_MISMATCH;
        }
    }

    if (ARGON2_MIN_PWD_LENGTH > context->pwdlen) {
      return ARGON2_PWD_TOO_SHORT;
    }

    if (ARGON2_MAX_PWD_LENGTH < context->pwdlen) {
        return ARGON2_PWD_TOO_LONG;
    }

    /* Validate salt (required param) */
    if (NULL == context->salt) {
        if (0 != context->saltlen) {
            return ARGON2_SALT_PTR_MISMATCH;
        }
    }

    if (ARGON2_MIN_SALT_LENGTH > context->saltlen) {
        return ARGON2_SALT_TOO_SHORT;
    }

    if (ARGON2_MAX_SALT_LENGTH < context->saltlen) {
        return ARGON2_SALT_TOO_LONG;
    }

    /* Validate secret (optional param) */
    if (NULL == context->secret) {
        if (0 != context->secretlen) {
            return ARGON2_SECRET_PTR_MISMATCH;
        }
    } else {
        if (ARGON2_MIN_SECRET > context->secretlen) {
            return ARGON2_SECRET_TOO_SHORT;
        }
        if (ARGON2_MAX_SECRET < context->secretlen) {
            return ARGON2_SECRET_TOO_LONG;
        }
    }

    /* Validate associated data (optional param) */
    if (NULL == context->ad) {
        if (0 != context->adlen) {
            return ARGON2_AD_PTR_MISMATCH;
        }
    } else {
        if (ARGON2_MIN_AD_LENGTH > context->adlen) {
            return ARGON2_AD_TOO_SHORT;
        }
        if (ARGON2_MAX_AD_LENGTH < context->adlen) {
            return ARGON2_AD_TOO_LONG;
        }
    }

    /* Validate memory cost */
    if (ARGON2_MIN_MEMORY > context->m_cost) {
        return ARGON2_MEMORY_TOO_LITTLE;
    }

    if (ARGON2_MAX_MEMORY < context->m_cost) {
        return ARGON2_MEMORY_TOO_MUCH;
    }

    if (context->m_cost < 8 * context->lanes) {
        return ARGON2_MEMORY_TOO_LITTLE;
    }

    /* Validate time cost */
    if (ARGON2_MIN_TIME > context->t_cost) {
        return ARGON2_TIME_TOO_SMALL;
    }

    if (ARGON2_MAX_TIME < context->t_cost) {
        return ARGON2_TIME_TOO_LARGE;
    }

    /* Validate lanes */
    if (ARGON2_MIN_LANES > context->lanes) {
        return ARGON2_LANES_TOO_FEW;
    }

    if (ARGON2_MAX_LANES < context->lanes) {
        return ARGON2_LANES_TOO_MANY;
    }

    /* Validate threads */
    if (ARGON2_MIN_THREADS > context->threads) {
        return ARGON2_THREADS_TOO_FEW;
    }

    if (ARGON2_MAX_THREADS < context->threads) {
        return ARGON2_THREADS_TOO_MANY;
    }

    if (NULL != context->allocate_cbk && NULL == context->free_cbk) {
        return ARGON2_FREE_MEMORY_CBK_NULL;
    }

    if (NULL == context->allocate_cbk && NULL != context->free_cbk) {
        return ARGON2_ALLOCATE_MEMORY_CBK_NULL;
    }

    return ARGON2_OK;
}

/**
 * Function creates first 2 blocks per lane
 * @param instance Pointer to the current instance
 * @param blockhash Pointer to the pre-hashing digest
 * @pre blockhash must point to @a PREHASH_SEED_LENGTH allocated values
 */
void fill_first_blocks(uint8_t *blockhash, const argon2_instance_t *instance) {
    uint32_t l;
    /* Make the first and second block in each lane as G(H0||0||i) or
       G(H0||1||i) */
    uint8_t blockhash_bytes[ARGON2_BLOCK_SIZE];
    for (l = 0; l < instance->lanes; ++l) {

        store32(blockhash + ARGON2_PREHASH_DIGEST_LENGTH, 0);
        store32(blockhash + ARGON2_PREHASH_DIGEST_LENGTH + 4, l);
        blake2b_long(blockhash_bytes, ARGON2_BLOCK_SIZE, blockhash,
                     ARGON2_PREHASH_SEED_LENGTH);
        load_block(&instance->memory[l * instance->lane_length + 0],
                   blockhash_bytes);

        store32(blockhash + ARGON2_PREHASH_DIGEST_LENGTH, 1);
        blake2b_long(blockhash_bytes, ARGON2_BLOCK_SIZE, blockhash,
                     ARGON2_PREHASH_SEED_LENGTH);
        load_block(&instance->memory[l * instance->lane_length + 1],
                   blockhash_bytes);
    }
    clear_internal_memory(blockhash_bytes, ARGON2_BLOCK_SIZE);
}

/**
 * Hashes all the inputs into @a blockhash[PREHASH_DIGEST_LENGTH],
 * clears password and secret if needed
 *
 * @param context Pointer to the Argon2 internal structure containing
 *     memory pointer, and parameters for time and space requirements
 * @param  blockhash Buffer for pre-hashing digest
 * @param  type Argon2 type
 * @pre @a blockhash must have at least @a PREHASH_DIGEST_LENGTH bytes
 *         allocated
 */
void initial_hash(uint8_t *blockhash, argon2_context *context,
                  argon2_type type) {
    blake2b_state BlakeHash;
    uint8_t value[sizeof(uint32_t)];

    if (NULL == context || NULL == blockhash) {
        return;
    }

    blake2b_init(&BlakeHash, ARGON2_PREHASH_DIGEST_LENGTH);

    store32(&value, context->lanes);
    blake2b_update(&BlakeHash, (const uint8_t *)&value, sizeof(value));

    store32(&value, context->outlen);
    blake2b_update(&BlakeHash, (const uint8_t *)&value, sizeof(value));

    store32(&value, context->m_cost);
    blake2b_update(&BlakeHash, (const uint8_t *)&value, sizeof(value));

    store32(&value, context->t_cost);
    blake2b_update(&BlakeHash, (const uint8_t *)&value, sizeof(value));

    store32(&value, context->version);
    blake2b_update(&BlakeHash, (const uint8_t *)&value, sizeof(value));

    store32(&value, (uint32_t)type);
    blake2b_update(&BlakeHash, (const uint8_t *)&value, sizeof(value));

    store32(&value, context->pwdlen);
    blake2b_update(&BlakeHash, (const uint8_t *)&value, sizeof(value));

    if (context->pwd != NULL) {
        blake2b_update(&BlakeHash, (const uint8_t *)context->pwd,
                       context->pwdlen);

        if (context->flags & ARGON2_FLAG_CLEAR_PASSWORD) {
            explicit_bzero(context->pwd, context->pwdlen);
            context->pwdlen = 0;
        }
    }

    store32(&value, context->saltlen);
    blake2b_update(&BlakeHash, (const uint8_t *)&value, sizeof(value));

    if (context->salt != NULL) {
        blake2b_update(&BlakeHash, (const uint8_t *)context->salt,
                       context->saltlen);
    }

    store32(&value, context->secretlen);
    blake2b_update(&BlakeHash, (const uint8_t *)&value, sizeof(value));

    if (context->secret != NULL) {
        blake2b_update(&BlakeHash, (const uint8_t *)context->secret,
                       context->secretlen);

        if (context->flags & ARGON2_FLAG_CLEAR_SECRET) {
            explicit_bzero(context->secret, context->secretlen);
            context->secretlen = 0;
        }
    }

    store32(&value, context->adlen);
    blake2b_update(&BlakeHash, (const uint8_t *)&value, sizeof(value));

    if (context->ad != NULL) {
        blake2b_update(&BlakeHash, (const uint8_t *)context->ad,
                       context->adlen);
    }

    blake2b_final(&BlakeHash, blockhash, ARGON2_PREHASH_DIGEST_LENGTH);
}

/**
 * Function allocates memory, hashes the inputs with Blake, and creates
 * first two blocks. Returns the pointer to the main memory with 2
 * blocks per lane initialized.
 *
 * @param  context  Pointer to the Argon2 internal structure containing memory
 *     pointer, and parameters for time and space requirements.
 * @param  instance Current Argon2 instance
 * @return Zero if successful, -1 if memory failed to allocate. @context->state
 *     will be modified if successful.
 */
int initialize(argon2_instance_t *instance, argon2_context *context) {
    uint8_t blockhash[ARGON2_PREHASH_SEED_LENGTH];
    int result = ARGON2_OK;

    if (instance == NULL || context == NULL)
        return ARGON2_INCORRECT_PARAMETER;
    instance->context_ptr = context;

    /* 1. Memory allocation */
    result = allocate_memory(context, (uint8_t **)&(instance->memory),
                             instance->memory_blocks, sizeof(block));
    if (result != ARGON2_OK) {
        return result;
    }

    /* 2. Initial hashing */
    /* H_0 + 8 extra bytes to produce the first blocks */
    /* uint8_t blockhash[ARGON2_PREHASH_SEED_LENGTH]; */
    /* Hashing all inputs */
    initial_hash(blockhash, context, instance->type);
    /* Zeroing 8 extra bytes */
    clear_internal_memory(blockhash + ARGON2_PREHASH_DIGEST_LENGTH,
                          ARGON2_PREHASH_SEED_LENGTH -
                              ARGON2_PREHASH_DIGEST_LENGTH);

    /* 3. Creating first blocks, we always have at least two blocks in a slice
     */
    fill_first_blocks(blockhash, instance);
    /* Clearing the hash */
    clear_internal_memory(blockhash, ARGON2_PREHASH_SEED_LENGTH);

    return ARGON2_OK;
}