packer.cpp

UPX 源代码

        //patchPackHeader(p, lsize);
        ph.putPackHeader(p);
    }
#endif
}


/*************************************************************************
// Try compression with several methods and filters, choose the best
/  or first working one. Needs buildLoader().
//
// Required inputs:
//   this->ph
//     ulen
//   parm_ft
//     clevel
//     addvalue
//     buf_len (optional)
//
// - updates this->ph
// - updates *ft
// - i_ptr[] is restored to the original unfiltered version
// - o_ptr[] contains the best compressed version
//
// filter_strategy:
//   n:  try the first N filters, use best one
//  -1:  try all filters, use first working one
//  -2:  try only the opt->filter filter
//  -3:  use no filter at all
//
// This has been prepared for generalization into class Packer so that
// opt->all_methods and/or opt->all_filters are available for all
// executable formats.
//
// It will replace the tryFilters() / compress() call sequence.
//
// 2006-02-15: hdr_buf and hdr_u_len are default empty input "header" array
// to fix a 2-pass problem with Elf headers.  As of today there can be
// only one decompression method per executable output file, and that method
// is the one that gives best compression for .text and loader.  However,
// the Elf headers precede .text in the output file, and are written first.
// "--brute" compression often compressed the Elf headers using nrv2b
// but the .text (and loader) with nrv2e.  This often resulted in SIGSEGV
// during decompression.
// The workaround is for hdr_buf and hdr_u_len to describe the Elf headers
// (typically less than 512 bytes) when .text is passed in, and include
// them in the calculation of shortest output.  Then the result
// this->ph.method  will say which [single] method to use for everything.
// The Elf headers are never filtered.  They are short enough (< 512 bytes)
// that compressing them more than once per method (once here when choosing,
// once again just before writing [because compressWithFilters discards])
// is OK because of the simplicity of not having two output arrays.
**************************************************************************/

static int prepareMethods(int *methods, int ph_method, const int *all_methods)
{
    int nmethods = 0;
    if (!opt->all_methods || all_methods == NULL)
    {
        methods[nmethods++] = ph_method;
        return nmethods;
    }
    for (int mm = 0; all_methods[mm] != M_END; ++mm)
    {
        int method = all_methods[mm];
        if (method == M_ULTRA_BRUTE && !opt->ultra_brute)
            break;
        if (method == M_SKIP || method == M_ULTRA_BRUTE)
            continue;
        if (opt->all_methods && !opt->all_methods_use_lzma && M_IS_LZMA(method))
            continue;
        // use this method
        assert(Packer::isValidCompressionMethod(method));
        methods[nmethods++] = method;
    }
    return nmethods;
}


static int prepareFilters(int *filters, int &filter_strategy,
                          const int *all_filters)
{
    int nfilters = 0;

    // setup filter filter_strategy
    if (filter_strategy == 0)
    {
        if (opt->all_filters)
            // choose best from all available filters
            filter_strategy = INT_MAX;
        else if (opt->filter >= 0 && Filter::isValidFilter(opt->filter, all_filters))
            // try opt->filter
            filter_strategy = -2;
        else
            // try the first working filter
            filter_strategy = -1;
    }
    assert(filter_strategy != 0);

    if (filter_strategy == -3)
        goto done;
    if (filter_strategy == -2)
    {
        if (opt->filter >= 0 && Filter::isValidFilter(opt->filter, all_filters))
        {
            filters[nfilters++] = opt->filter;
            goto done;
        }
        filter_strategy = -1;
    }
    assert(filter_strategy >= -1);

    while (all_filters && *all_filters != FT_END)
    {
        int filter_id = *all_filters++;
        if (filter_id == FT_ULTRA_BRUTE && !opt->ultra_brute)
            break;
        if (filter_id == FT_SKIP || filter_id == FT_ULTRA_BRUTE)
            continue;
        if (filter_id == 0)
            continue;
        // use this filter
        assert(Filter::isValidFilter(filter_id));
        filters[nfilters++] = filter_id;
        if (filter_strategy >= 0 && nfilters >= filter_strategy)
            break;
    }

done:
    // filter_strategy now only means "stop after first successful filter"
    filter_strategy = (filter_strategy < 0) ? -1 : 0;
    // make sure that we have a "no filter" fallback
    for (int i = 0; i < nfilters; i++)
        if (filters[i] == 0)
            return nfilters;
    filters[nfilters++] = 0;
    return nfilters;
}


void Packer::compressWithFilters(upx_bytep i_ptr, unsigned i_len,
                                 upx_bytep o_ptr,
                                 upx_bytep f_ptr, unsigned f_len,
                                 const upx_bytep hdr_ptr, unsigned hdr_len,
                                 Filter *parm_ft,
                                 const unsigned overlap_range,
                                 const upx_compress_config_t *cconf,
                                 int filter_strategy)
{
    parm_ft->buf_len = f_len;
    // struct copies
    const PackHeader orig_ph = this->ph;
          PackHeader best_ph = this->ph;
    const Filter orig_ft = *parm_ft;
          Filter best_ft = *parm_ft;
    //
    best_ph.c_len = i_len;
    best_ph.overlap_overhead = 0;
    unsigned best_ph_lsize = 0;
    unsigned best_hdr_c_len = 0;

    // preconditions
    assert(orig_ph.filter == 0);
    assert(orig_ft.id == 0);

    // prepare methods and filters
    int methods[256];
    int nmethods = prepareMethods(methods, ph.method, getCompressionMethods(M_ALL, ph.level));
    assert(nmethods > 0); assert(nmethods < 256);
    int filters[256];
    int nfilters = prepareFilters(filters, filter_strategy, getFilters());
    assert(nfilters > 0); assert(nfilters < 256);
#if 0
    printf("compressWithFilters: m(%d):", nmethods);
    for (int i = 0; i < nmethods; i++) printf(" %d", methods[i]);
    printf(" f(%d):", nfilters);
    for (int i = 0; i < nfilters; i++) printf(" %d", filters[i]);
    printf("\n");
#endif

    // update total_passes; previous (ui_total_passes > 0) means incremental
    if (uip->ui_total_passes > 0)
        uip->ui_total_passes -= 1;
    if (filter_strategy < 0)
        uip->ui_total_passes += nmethods;
    else
        uip->ui_total_passes += nfilters * nmethods;

    // Working buffer for compressed data. Don't waste memory and allocate as needed.
    upx_bytep o_tmp = o_ptr;
    MemBuffer o_tmp_buf;

    // compress using all methods/filters
    int nfilters_success_total = 0;
    for (int mm = 0; mm < nmethods; mm++) // for all methods
    {
        assert(isValidCompressionMethod(methods[mm]));
        unsigned hdr_c_len = 0;
        if (hdr_ptr != NULL && hdr_len)
        {
            if (nfilters_success_total != 0 && o_tmp == o_ptr)
            {
                // do not overwrite o_ptr
                o_tmp_buf.allocForCompression(UPX_MAX(hdr_len, i_len));
                o_tmp = o_tmp_buf;
            }
            int r = upx_compress(hdr_ptr, hdr_len, o_tmp, &hdr_c_len,
                                 NULL, methods[mm], 10, NULL, NULL);
            if (r != UPX_E_OK)
                throwInternalError("header compression failed");
            if (hdr_c_len >= hdr_len)
                throwInternalError("header compression size increase");
        }
        int nfilters_success_mm = 0;
        for (int ff = 0; ff < nfilters; ff++) // for all filters
        {
            assert(isValidFilter(filters[ff]));
            // get fresh packheader
            ph = orig_ph;
            ph.method = methods[mm];
            ph.filter = filters[ff];
            ph.overlap_overhead = 0;
            // get fresh filter
            Filter ft = orig_ft;
            ft.init(ph.filter, orig_ft.addvalue);
            // filter
            optimizeFilter(&ft, f_ptr, f_len);
            bool success = ft.filter(f_ptr, f_len);
            if (ft.id != 0 && ft.calls == 0)
            {
                // filter did not do anything - no need to call ft.unfilter()
                success = false;
            }
            if (!success)
            {
                // filter failed or was useless
                if (filter_strategy >= 0)
                {
                    // adjust ui passes
                    if (uip->ui_pass >= 0)
                        uip->ui_pass++;
                }
                continue;
            }
            // filter success
#if 0
            printf("filter: id 0x%02x size %6d, calls %5d/%5d/%3d/%5d/%5d, cto 0x%02x\n",
                   ft.id, ft.buf_len, ft.calls, ft.noncalls, ft.wrongcalls, ft.firstcall, ft.lastcall, ft.cto);
#endif
            if (nfilters_success_total != 0 && o_tmp == o_ptr)
            {
                o_tmp_buf.allocForCompression(i_len);
                o_tmp = o_tmp_buf;
            }
            nfilters_success_total++;
            nfilters_success_mm++;
            ph.filter_cto = ft.cto;
            ph.n_mru = ft.n_mru;
            // compress
            if (compress(i_ptr, i_len, o_tmp, cconf))
            {
                unsigned lsize = 0;
                // findOverlapOperhead() might be slow; omit if already too big.
                if (ph.c_len + lsize + hdr_c_len <= best_ph.c_len + best_ph_lsize + best_hdr_c_len)
                {
                    // get results
                    ph.overlap_overhead = findOverlapOverhead(o_tmp, i_ptr, overlap_range);
                    buildLoader(&ft);
                    lsize = getLoaderSize();
                    assert(lsize > 0);
                }
#if 0
                printf("\n%2d %02x: %d +%4d +%3d = %d  (best: %d +%4d +%3d = %d)\n", ph.method, ph.filter,
                       ph.c_len, lsize, hdr_c_len, ph.c_len + lsize + hdr_c_len,
                       best_ph.c_len, best_ph_lsize, best_hdr_c_len, best_ph.c_len + best_ph_lsize + best_hdr_c_len);
#endif
                bool update = false;
                if (ph.c_len + lsize + hdr_c_len < best_ph.c_len + best_ph_lsize + best_hdr_c_len)
                    update = true;
                else if (ph.c_len + lsize + hdr_c_len == best_ph.c_len + best_ph_lsize + best_hdr_c_len)
                {
                    // prefer smaller loaders
                    if (lsize  + hdr_c_len < best_ph_lsize + best_hdr_c_len)
                        update = true;
                    else if (lsize + hdr_c_len == best_ph_lsize + best_hdr_c_len)
                    {
                        // prefer less overlap_overhead
                        if (ph.overlap_overhead < best_ph.overlap_overhead)
                            update = true;
                    }
                }
                if (update)
                {
                    assert((int)ph.overlap_overhead > 0);
                    // update o_ptr[] with best version
                    if (o_tmp != o_ptr)
                        memcpy(o_ptr, o_tmp, ph.c_len);
                    // save compression results
                    best_ph = ph;
                    best_ph_lsize = lsize;
                    best_hdr_c_len = hdr_c_len;
                    best_ft = ft;
                }
            }
            // restore - unfilter with verify
            ft.unfilter(f_ptr, f_len, true);
            if (filter_strategy < 0)
                break;
        }
        assert(nfilters_success_mm > 0);
    }

    // postconditions 1)
    assert(nfilters_success_total > 0);
    assert(best_ph.u_len == orig_ph.u_len);
    assert(best_ph.filter == best_ft.id);
    assert(best_ph.filter_cto == best_ft.cto);
    // FIXME  assert(best_ph.n_mru == best_ft.n_mru);

    // copy back results
    this->ph = best_ph;
    *parm_ft = best_ft;

    // finally check compression ratio
    if (best_ph.c_len + best_ph_lsize >= best_ph.u_len)
        throwNotCompressible();
    if (!checkCompressionRatio(best_ph.u_len, best_ph.c_len))
        throwNotCompressible();

    // postconditions 2)
    assert(best_ph.overlap_overhead > 0);

    // convenience
    buildLoader(&best_ft);
}


/*************************************************************************
//
**************************************************************************/

void Packer::compressWithFilters(Filter *ft,
                                 const unsigned overlap_range,
                                 const upx_compress_config_t *cconf,
                                 int filter_strategy)
{
    compressWithFilters(ft, overlap_range, cconf, filter_strategy,
                        0, 0, 0, NULL, 0);
}


void Packer::compressWithFilters(Filter *ft,
                                 const unsigned overlap_range,
                                 const upx_compress_config_t *cconf,
                                 int filter_strategy,
                                 unsigned filter_off,
                                 unsigned ibuf_off,
                                 unsigned obuf_off,
                                 const upx_bytep hdr_ptr, unsigned hdr_len)
{
    ibuf.checkState(); obuf.checkState();

    upx_bytep i_ptr = ibuf + ibuf_off;
    unsigned  i_len = ph.u_len;
    upx_bytep o_ptr = obuf + obuf_off;
    upx_bytep f_ptr = ibuf + filter_off;
    unsigned  f_len = ft->buf_len ? ft->buf_len : i_len;

    assert(f_ptr + f_len <= i_ptr + i_len);

    compressWithFilters(i_ptr, i_len,
                        o_ptr,
                        f_ptr, f_len,
                        hdr_ptr, hdr_len,
                        ft, overlap_range, cconf, filter_strategy);

    ibuf.checkState(); obuf.checkState();
}


/*
vi:ts=4:et:nowrap
*/