p_tos.cpp

UPX 源代码

        throwCantPack("program is too small");

    return true;
}


void PackTos::fileInfo()
{
    if (!readFileHeader())
        return;
    con_fprintf(stdout, "    text: %d, data: %d, sym: %d, bss: %d, flags=0x%x\n",
                (int)ih.fh_text, (int)ih.fh_data, (int)ih.fh_sym, (int)ih.fh_bss, (int)ih.fh_flag);
}


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

void PackTos::pack(OutputFile *fo)
{
    unsigned t;
    unsigned nrelocs = 0;
    unsigned relocsize = 0;
    unsigned overlay = 0;

    const unsigned i_text = ih.fh_text;
    const unsigned i_data = ih.fh_data;
    const unsigned i_sym = ih.fh_sym;
    const unsigned i_bss = ih.fh_bss;

    symbols.reset();
    symbols.need_reloc = false;
    // prepare symbols for buildLoader() - worst case
    symbols.loop1.init(65536 + 1);
    symbols.loop2.init((160 - 1) / 4);
    symbols.loop3.init(65536 + 1);
    symbols.up21_d4 = 65536 + 1;
    symbols.up21_a6 = 65536 + 1;
    symbols.up31_base_d4 = 65536 + 1;
    symbols.up31_base_a6 = 65536 + 1;

    // read file
    const unsigned isize = file_size - i_sym;
    ibuf.alloc(isize);
    fi->seek(FH_SIZE, SEEK_SET);
    // read text + data
    t = i_text + i_data;
    fi->readx(ibuf,t);
    // skip symbols
    if (i_sym && opt->exact)
        throwCantPackExact();
    fi->seek(i_sym,SEEK_CUR);
    // read relocations + overlay
    overlay = file_size - (FH_SIZE + i_text + i_data + i_sym);
    fi->readx(ibuf+t,overlay);

#if 0 || defined(TESTING)
    printf("text: %d, data: %d, sym: %d, bss: %d, flags=0x%x\n",
           i_text, i_data, i_sym, i_bss, (int)ih.fh_flag);
    printf("xx1 reloc: %d, overlay: %d, fixup: %d\n", relocsize, overlay, overlay >= 4 ? (int)get_be32(ibuf+t) : -1);
#endif

    // Check relocs (see load_and_reloc() in freemint/sys/memory.c).
    // Must work around TOS bugs and lots of broken programs.
    if (overlay < 4)
    {
        // Bug workaround: Whatever this is, silently keep it in
        // the (unused) relocations for byte-identical unpacking.
        relocsize = overlay;
        overlay = 0;
    }
    else if (get_be32(ibuf+t) == 0)
    {
        // Bug workaround - check the empty fixup before testing fh_reloc.
        relocsize = 4;
        overlay -= 4;
    }
    else if (ih.fh_reloc != 0)
        relocsize = 0;
    else {
        int r = check_relocs(ibuf+t, overlay, t, &nrelocs, &relocsize, &overlay);
        if (r != 0)
            throwCantPack("bad relocation table");
        symbols.need_reloc = true;
    }

#if 0 || defined(TESTING)
    printf("xx2: %d relocs: %d, overlay: %d, t: %d\n", nrelocs, relocsize, overlay, t);
#endif

    checkOverlay(overlay);

    // Append original fileheader.
    t += relocsize;
    ih.fh_sym = 0;                      // we stripped all symbols
    memcpy(ibuf+t, &ih, FH_SIZE);
    t += FH_SIZE;
#if 0 || defined(TESTING)
    printf("xx3 reloc: %d, overlay: %d, t: %d\n", relocsize, overlay, t);
#endif
    assert(t <= isize);

    // Now the data in ibuf[0..t] looks like this:
    //   text + data + relocs + original file header
    // After compression this will become the first part of the
    // data segement. The second part will be the decompressor.

    // alloc buffer (4096 is for decompressor and the various alignments)
    obuf.allocForCompression(t, 4096);

    // prepare packheader
    ph.u_len = t;
    // prepare filter
    Filter ft(ph.level);
    // compress (max_match = 65535)
    upx_compress_config_t cconf; cconf.reset();
    cconf.conf_ucl.max_match = 65535;
    cconf.conf_lzma.max_num_probs = 1846 + (768 << 4); // ushort: ~28KB stack
    compressWithFilters(&ft, 512, &cconf);

    //
    // multipass buildLoader()
    //

    // save initial loader
    const unsigned initial_lsize = getLoaderSize();
    unsigned last_lsize = initial_lsize;
    MemBuffer last_loader(last_lsize);
    memcpy(last_loader, getLoader(), last_lsize);

    unsigned o_text, o_data, o_bss;
    unsigned e_len, d_len, d_off;
    for (;;)
    {
        // The decompressed data will now get placed at this offset:
        unsigned offset = (ph.u_len + ph.overlap_overhead) - ph.c_len;

        // get loader
        const unsigned lsize = getLoaderSize();
        e_len = getLoaderSectionStart("CUTPOINT");
        d_len = lsize - e_len;
        assert((e_len & 3) == 0 && (d_len & 1) == 0);

        // compute section sizes
        o_text = e_len;
        o_data = ph.c_len;
        o_bss = i_bss;

        // word align len of compressed data
        while (o_data & 1)
        {
            obuf[o_data++] = 0;
            offset++;
        }

        // append decompressor (part 2 of loader)
        d_off = o_data;
        ////memcpy(obuf + d_off, getLoader() + e_len, d_len); // must be done after relocation
        o_data += d_len;

        // dword align the len of the final data segment
        while (o_data & 3)
        {
            obuf[o_data++] = 0;
            offset++;
        }
        // dword align offset
        while (offset & 3)
            offset++;

        // new bss
        if (i_text + i_data + i_bss > o_text + o_data + o_bss)
            o_bss = (i_text + i_data + i_bss) - (o_text + o_data);

        // dirty bss
        unsigned dirty_bss = (o_data + offset) - (i_text + i_data);
        //printf("real dirty_bss: %d\n", dirty_bss);
        // dword align (or 16 - for speedup when clearing the dirty bss)
        const unsigned dirty_bss_align = opt->small ? 4 : 16;
        while (dirty_bss & (dirty_bss_align - 1))
            dirty_bss++;
        // adjust bss, assert room for some stack
        unsigned stack = 512 + getDecompressorWrkmemSize();
        if (dirty_bss + stack > o_bss)
            o_bss = dirty_bss + stack;

        // dword align the len of the final bss segment
        while (o_bss & 3)
            o_bss++;

        // update symbols for buildLoader()
        if (opt->small)
        {
            symbols.loop1.init(o_data / 4);
            symbols.loop2.init(0);
        }
        else
        {
            symbols.loop1.init(o_data / 160);
            symbols.loop2.init((o_data % 160) / 4);
        }
        symbols.loop3.init(dirty_bss / dirty_bss_align);

        symbols.up21_d4 = o_data + offset;
        symbols.up31_base_d4 = d_off + offset;
        symbols.up21_a6 = symbols.up21_d4 - (i_text + i_data);
        symbols.up31_base_a6 = symbols.up31_base_d4 - (i_text + i_data);
        assert((int)symbols.up21_a6 > 0);
        assert((int)symbols.up31_base_a6 > 0);

        const unsigned c = linker->getSymbolOffset("code_on_stack");
        unsigned d;
        d = linker->getSymbolOffset("flush_cache_rts") - c;
        symbols.flush_cache_rts_offset = d;
        d = linker->getSymbolOffset("clear_dirty_stack_loop") - c;
        symbols.clear_dirty_stack_len = (d + 3) / 4 + 32 - 1;
        d = linker->getSymbolOffset("code_on_stack_end") - c;
        symbols.copy_to_stack_len = d / 2 - 1;

        // now re-build loader
        buildLoader(&ft);
        unsigned new_lsize = getLoaderSize();
        //printf("buildLoader %d %d\n", new_lsize, initial_lsize);
        assert(new_lsize <= initial_lsize);
        if (new_lsize == last_lsize && memcmp(getLoader(), last_loader, last_lsize) == 0)
            break;
        last_lsize = new_lsize;
        memcpy(last_loader, getLoader(), last_lsize);
    }

    //
    // define symbols and reloc
    //

    defineDecompressorSymbols();

    linker->defineSymbol("loop1_count", symbols.loop1.value);
    linker->defineSymbol("loop2_count", symbols.loop2.value);
    linker->defineSymbol("loop3_count", symbols.loop3.value);

    linker->defineSymbol("orig_p_tlen", i_text);
    linker->defineSymbol("orig_p_dlen", i_data);
    linker->defineSymbol("orig_p_blen", i_bss);

    if (symbols.up21_a6 <= 32767)
        linker->defineSymbol("up21_a6", symbols.up21_a6);
    else
        linker->defineSymbol("up21_d4", symbols.up21_d4);

    if (symbols.up31_a6 <= 32767)
        linker->defineSymbol("up31_a6", symbols.up31_a6);
    else if (symbols.up31_d4 <= 32767)
        linker->defineSymbol("up31_d4", symbols.up31_d4);
    else if (symbols.up31_a6 <= 65534)
        linker->defineSymbol("up31_a6", symbols.up31_a6 - 32767);
    else
        linker->defineSymbol("up31_d4", symbols.up31_d4);
#if 0
    printf("relocsize = %d\n", relocsize);
    printf("upx21(d4) = %d\n", symbols.up21_d4);
    printf("upx21(a6) = %d\n", symbols.up21_a6);
    printf("upx31(d4) = %d\n", symbols.up31_d4);
    printf("upx31(a6) = %d\n", symbols.up31_a6);
#endif

    linker->defineSymbol("flush_cache_rts_offset", symbols.flush_cache_rts_offset);
    linker->defineSymbol("copy_to_stack_len", symbols.copy_to_stack_len);
    linker->defineSymbol("clear_dirty_stack_len", symbols.clear_dirty_stack_len);

    relocateLoader();

    //
    // write
    //

    // set new file_hdr
    memcpy(&oh, &ih, FH_SIZE);
    if (opt->atari_tos.split_segments)
    {
        oh.fh_text = o_text;
        oh.fh_data = o_data;
    }
    else
    {
        // put everything into the text segment
        oh.fh_text = o_text + o_data;
        oh.fh_data = 0;
    }
    oh.fh_bss = o_bss;
    oh.fh_sym = 0;
    oh.fh_reserved = 0;
    // only keep the following flags:
    oh.fh_flag = ih.fh_flag & (F_FASTLOAD | F_ALTALLOC | F_SMALLTPA | F_ALLOCZERO | F_KEEP);
    // add an empty relocation fixup to workaround a bug in some TOS versions
    oh.fh_reloc = 0;

#if 0 || defined(TESTING)
    printf("old text: %6d, data: %6d, bss: %6d, reloc: %d, overlay: %d\n",
           i_text, i_data, i_bss, relocsize, overlay);
    printf("new text: %6d, data: %6d, bss: %6d, flag=0x%x\n",
           o_text, o_data, o_bss, (int)oh.fh_flag);
    linker->dumpSymbols();
#endif

    // prepare loader
    MemBuffer loader(o_text);
    memcpy(loader, getLoader(), o_text);
    patchPackHeader(loader, o_text);

    // write new file header, loader and compressed file
    fo->write(&oh, FH_SIZE);
    fo->write(loader, o_text);  // entry
    if (opt->debug.dump_stub_loader)
        OutputFile::dump(opt->debug.dump_stub_loader, loader, o_text);
    memcpy(obuf + d_off, getLoader() + e_len, d_len); // copy decompressor
    fo->write(obuf, o_data);    // compressed + decompressor

    // write empty relocation fixup
    fo->write("\x00\x00\x00\x00", 4);

    // verify
    verifyOverlappingDecompression();

    // copy the overlay
    copyOverlay(fo, overlay, &obuf);

    // finally check the compression ratio
    if (!checkFinalCompressionRatio(fo))
        throwNotCompressible();
}


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

int PackTos::canUnpack()
{
    if (!readFileHeader())
        return false;
    if (!readPackHeader(768))
        return false;
    // check header as set by packer
    if ((ih.fh_text & 3) != 0 || (ih.fh_data & 3) != 0 || (ih.fh_bss & 3) != 0
        || ih.fh_sym != 0 || ih.fh_reserved != 0 || ih.fh_reloc > 1)
        throwCantUnpack("program header damaged");
    // generic check
    if (!checkFileHeader())
        throwCantUnpack("unsupported header flags");
    return true;
}


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

void PackTos::unpack(OutputFile *fo)
{
    ibuf.alloc(ph.c_len);
    obuf.allocForUncompression(ph.u_len);

    fi->seek(FH_SIZE + ph.buf_offset + ph.getPackHeaderSize(), SEEK_SET);
    fi->readx(ibuf, ph.c_len);

    // decompress
    decompress(ibuf,obuf);

    // write original header & decompressed file
    if (fo)
    {
        unsigned overlay = file_size - (FH_SIZE + ih.fh_text + ih.fh_data);
        if (ih.fh_reloc == 0 && overlay >= 4)
            overlay -= 4;                           // this is our empty fixup
        checkOverlay(overlay);

        fo->write(obuf+ph.u_len-FH_SIZE, FH_SIZE);  // orig. file_hdr
        fo->write(obuf, ph.u_len-FH_SIZE);          // orig. text+data+relocs

        // copy any overlay
        copyOverlay(fo, overlay, &obuf);
    }
}


/*
vi:ts=4:et
*/