The R Project SVN R

///////////////////////////////////////////////////////////////////////////////
//
/// \file       block_buffer_encoder.c
/// \brief      Single-call .xz Block encoder
//
//  Author:     Lasse Collin
//
//  This file has been put into the public domain.
//  You can do whatever you want with this file.
//
///////////////////////////////////////////////////////////////////////////////

#include "block_encoder.h"
#include "filter_encoder.h"
#include "lzma2_encoder.h"
#include "check.h"


/// Estimate the maximum size of the Block Header and Check fields for
/// a Block that uses LZMA2 uncompressed chunks. We could use
/// lzma_block_header_size() but this is simpler.
///
/// Block Header Size + Block Flags + Compressed Size
/// + Uncompressed Size + Filter Flags for LZMA2 + CRC32 + Check
/// and round up to the next multiple of four to take Header Padding
/// into account.
#define HEADERS_BOUND ((1 + 1 + 2 * LZMA_VLI_BYTES_MAX + 3 + 4 \
        + LZMA_CHECK_SIZE_MAX + 3) & ~3)


static lzma_vli
lzma2_bound(lzma_vli uncompressed_size)
{
    // Prevent integer overflow in overhead calculation.
    if (uncompressed_size > COMPRESSED_SIZE_MAX)
        return 0;

    // Calculate the exact overhead of the LZMA2 headers: Round
    // uncompressed_size up to the next multiple of LZMA2_CHUNK_MAX,
    // multiply by the size of per-chunk header, and add one byte for
    // the end marker.
    const lzma_vli overhead = ((uncompressed_size + LZMA2_CHUNK_MAX - 1)
                / LZMA2_CHUNK_MAX)
            * LZMA2_HEADER_UNCOMPRESSED + 1;

    // Catch the possible integer overflow.
    if (COMPRESSED_SIZE_MAX - overhead < uncompressed_size)
        return 0;

    return uncompressed_size + overhead;
}


extern LZMA_API(size_t)
lzma_block_buffer_bound(size_t uncompressed_size)
{
    // For now, if the data doesn't compress, we always use uncompressed
    // chunks of LZMA2. In future we may use Subblock filter too, but
    // but for simplicity we probably will still use the same bound
    // calculation even though Subblock filter would have slightly less
    // overhead.
    lzma_vli lzma2_size = lzma2_bound(uncompressed_size);
    if (lzma2_size == 0)
        return 0;

    // Take Block Padding into account.
    lzma2_size = (lzma2_size + 3) & ~LZMA_VLI_C(3);

#if SIZE_MAX < LZMA_VLI_MAX
    // Catch the possible integer overflow on 32-bit systems. There's no
    // overflow on 64-bit systems, because lzma2_bound() already takes
    // into account the size of the headers in the Block.
    if (SIZE_MAX - HEADERS_BOUND < lzma2_size)
        return 0;
#endif

    return HEADERS_BOUND + lzma2_size;
}


static lzma_ret
block_encode_uncompressed(lzma_block *block, const uint8_t *in, size_t in_size,
        uint8_t *out, size_t *out_pos, size_t out_size)
{
    // TODO: Figure out if the last filter is LZMA2 or Subblock and use
    // that filter to encode the uncompressed chunks.

    // Use LZMA2 uncompressed chunks. We wouldn't need a dictionary at
    // all, but LZMA2 always requires a dictionary, so use the minimum
    // value to minimize memory usage of the decoder.
    lzma_options_lzma lzma2 = {
        .dict_size = LZMA_DICT_SIZE_MIN,
    };

    lzma_filter filters[2];
    filters[0].id = LZMA_FILTER_LZMA2;
    filters[0].options = &lzma2;
    filters[1].id = LZMA_VLI_UNKNOWN;

    // Set the above filter options to *block temporarily so that we can
    // encode the Block Header.
    lzma_filter *filters_orig = block->filters;
    block->filters = filters;

    if (lzma_block_header_size(block) != LZMA_OK) {
        block->filters = filters_orig;
        return LZMA_PROG_ERROR;
    }

    // Check that there's enough output space. The caller has already
    // set block->compressed_size to what lzma2_bound() has returned,
    // so we can reuse that value. We know that compressed_size is a
    // known valid VLI and header_size is a small value so their sum
    // will never overflow.
    assert(block->compressed_size == lzma2_bound(in_size));
    if (out_size - *out_pos
            < block->header_size + block->compressed_size) {
        block->filters = filters_orig;
        return LZMA_BUF_ERROR;
    }

    if (lzma_block_header_encode(block, out + *out_pos) != LZMA_OK) {
        block->filters = filters_orig;
        return LZMA_PROG_ERROR;
    }

    block->filters = filters_orig;
    *out_pos += block->header_size;

    // Encode the data using LZMA2 uncompressed chunks.
    size_t in_pos = 0;
    uint8_t control = 0x01; // Dictionary reset

    while (in_pos < in_size) {
        // Control byte: Indicate uncompressed chunk, of which
        // the first resets the dictionary.
        out[(*out_pos)++] = control;
        control = 0x02; // No dictionary reset

        // Size of the uncompressed chunk
        const size_t copy_size
                = my_min(in_size - in_pos, LZMA2_CHUNK_MAX);
        out[(*out_pos)++] = (copy_size - 1) >> 8;
        out[(*out_pos)++] = (copy_size - 1) & 0xFF;

        // The actual data
        assert(*out_pos + copy_size <= out_size);
        memcpy(out + *out_pos, in + in_pos, copy_size);

        in_pos += copy_size;
        *out_pos += copy_size;
    }

    // End marker
    out[(*out_pos)++] = 0x00;
    assert(*out_pos <= out_size);

    return LZMA_OK;
}


static lzma_ret
block_encode_normal(lzma_block *block, lzma_allocator *allocator,
        const uint8_t *in, size_t in_size,
        uint8_t *out, size_t *out_pos, size_t out_size)
{
    // Find out the size of the Block Header.
    block->compressed_size = lzma2_bound(in_size);
    if (block->compressed_size == 0)
        return LZMA_DATA_ERROR;

    block->uncompressed_size = in_size;
    return_if_error(lzma_block_header_size(block));

    // Reserve space for the Block Header and skip it for now.
    if (out_size - *out_pos <= block->header_size)
        return LZMA_BUF_ERROR;

    const size_t out_start = *out_pos;
    *out_pos += block->header_size;

    // Limit out_size so that we stop encoding if the output would grow
    // bigger than what uncompressed Block would be.
    if (out_size - *out_pos > block->compressed_size)
        out_size = *out_pos + block->compressed_size;

    // TODO: In many common cases this could be optimized to use
    // significantly less memory.
    lzma_next_coder raw_encoder = LZMA_NEXT_CODER_INIT;
    lzma_ret ret = lzma_raw_encoder_init(
            &raw_encoder, allocator, block->filters);

    if (ret == LZMA_OK) {
        size_t in_pos = 0;
        ret = raw_encoder.code(raw_encoder.coder, allocator,
                in, &in_pos, in_size, out, out_pos, out_size,
                LZMA_FINISH);
    }

    // NOTE: This needs to be run even if lzma_raw_encoder_init() failed.
    lzma_next_end(&raw_encoder, allocator);

    if (ret == LZMA_STREAM_END) {
        // Compression was successful. Write the Block Header.
        block->compressed_size
                = *out_pos - (out_start + block->header_size);
        ret = lzma_block_header_encode(block, out + out_start);
        if (ret != LZMA_OK)
            ret = LZMA_PROG_ERROR;

    } else if (ret == LZMA_OK) {
        // Output buffer became full.
        ret = LZMA_BUF_ERROR;
    }

    // Reset *out_pos if something went wrong.
    if (ret != LZMA_OK)
        *out_pos = out_start;

    return ret;
}


extern LZMA_API(lzma_ret)
lzma_block_buffer_encode(lzma_block *block, lzma_allocator *allocator,
        const uint8_t *in, size_t in_size,
        uint8_t *out, size_t *out_pos, size_t out_size)
{
    // Sanity checks
    if (block == NULL || block->filters == NULL
            || (in == NULL && in_size != 0) || out == NULL
            || out_pos == NULL || *out_pos > out_size)
        return LZMA_PROG_ERROR;

    // Check the version field.
    if (block->version != 0)
        return LZMA_OPTIONS_ERROR;

    // Size of a Block has to be a multiple of four, so limit the size
    // here already. This way we don't need to check it again when adding
    // Block Padding.
    out_size -= (out_size - *out_pos) & 3;

    // Get the size of the Check field.
    const size_t check_size = lzma_check_size(block->check);
    if (check_size == UINT32_MAX)
        return LZMA_PROG_ERROR;

    // Reserve space for the Check field.
    if (out_size - *out_pos <= check_size)
        return LZMA_BUF_ERROR;

    out_size -= check_size;

    // Do the actual compression.
    const lzma_ret ret = block_encode_normal(block, allocator,
            in, in_size, out, out_pos, out_size);
    if (ret != LZMA_OK) {
        // If the error was something else than output buffer
        // becoming full, return the error now.
        if (ret != LZMA_BUF_ERROR)
            return ret;

        // The data was uncompressible (at least with the options
        // given to us) or the output buffer was too small. Use the
        // uncompressed chunks of LZMA2 to wrap the data into a valid
        // Block. If we haven't been given enough output space, even
        // this may fail.
        return_if_error(block_encode_uncompressed(block, in, in_size,
                out, out_pos, out_size));
    }

    assert(*out_pos <= out_size);

    // Block Padding. No buffer overflow here, because we already adjusted
    // out_size so that (out_size - out_start) is a multiple of four.
    // Thus, if the buffer is full, the loop body can never run.
    for (size_t i = (size_t)(block->compressed_size); i & 3; ++i) {
        assert(*out_pos < out_size);
        out[(*out_pos)++] = 0x00;
    }

    // If there's no Check field, we are done now.
    if (check_size > 0) {
        // Calculate the integrity check. We reserved space for
        // the Check field earlier so we don't need to check for
        // available output space here.
        lzma_check_state check;
        lzma_check_init(&check, block->check);
        lzma_check_update(&check, block->check, in, in_size);
        lzma_check_finish(&check, block->check);

        memcpy(block->raw_check, check.buffer.u8, check_size);
        memcpy(out + *out_pos, check.buffer.u8, check_size);
        *out_pos += check_size;
    }

    return LZMA_OK;
}