p_exe.cpp

UPX 源代码

}


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

void PackExe::pack(OutputFile *fo)
{
    unsigned ic;

    if (ih.relocs > MAXRELOCS)
        throwCantPack("too many relocations");
    checkOverlay(ih_overlay);

    // alloc buffers
    relocsize = RSFCRI + 4*ih.relocs;
    ibuf.alloc(ih_imagesize+16+relocsize+2);
    obuf.allocForCompression(ih_imagesize+16+relocsize+2);

    // read image
    fi->seek(ih.headsize16*16,SEEK_SET);
    fi->readx(ibuf,ih_imagesize);

    checkAlreadyPacked(ibuf, UPX_MIN(ih_imagesize, 127u));

    device_driver = get_le32(ibuf) == 0xffffffffu;

    // relocations
    has_9a = false;
    upx_byte *w = ibuf + ih_imagesize;
    if (ih.relocs)
    {
        upx_byte *wr = w + RSFCRI;

        fi->seek(ih.relocoffs,SEEK_SET);
        fi->readx(wr,4*ih.relocs);

        for (ic = 0; ic < ih.relocs; ic++)
        {
            unsigned jc = get_le32(wr+4*ic);
            set_le32(wr+4*ic, ((jc>>16)*16+(jc&0xffff)) & 0xfffff);
        }
        qsort(wr,ih.relocs,4,le32_compare);
        relocsize = optimize_relocs(ibuf, ih_imagesize, wr, ih.relocs, w, &has_9a);
        set_le16(w+relocsize, relocsize+2);
        relocsize += 2;
        if (relocsize > MAXRELOCS)
            throwCantPack("too many relocations");
#if 0
        upx_byte out[9*relocsize/8+1024];
        unsigned in_len = relocsize;
        unsigned out_len = 0;
        ucl_nrv2b_99_compress(w, in_len, out, &out_len, NULL, 9, NULL, NULL);
        printf("reloc compress: %d -> %d\n", in_len, out_len);
#endif
    }
    else
    {
        relocsize = 0;
    }

    // prepare packheader
    ph.u_len = ih_imagesize + relocsize;
    // prepare filter
    Filter ft(ph.level);
    // compress (max_match = 8192)
    upx_compress_config_t cconf; cconf.reset();
    cconf.conf_ucl.max_match = MAXMATCH;
    cconf.conf_lzma.max_num_probs = 1846 + (768 << 4); // ushort: ~28KB stack
    compressWithFilters(&ft, 32, &cconf);

    if (M_IS_NRV2B(ph.method) || M_IS_NRV2D(ph.method) || M_IS_NRV2E(ph.method))
        if (ph.max_run_found + ph.max_match_found > 0x8000)
            throwCantPack("decompressor limit exceeded, send a bugreport");

#ifdef TESTING
    if (opt->debug.debug_level)
    {
        printf("image+relocs %d -> %d\n",ih_imagesize+relocsize,ph.c_len);
        printf("offsets: %d - %d\nmatches: %d - %d\nruns: %d - %d\n",
               0/*ph.min_offset_found*/,ph.max_offset_found,
               0/*ph.min_match_found*/,ph.max_match_found,
               0/*ph.min_run_found*/,ph.max_run_found);
    }
#endif

    int flag = fillExeHeader(&oh);

    const unsigned lsize = getLoaderSize();
    MemBuffer loader(lsize);
    memcpy(loader,getLoader(),lsize);
    //OutputFile::dump("xxloader.dat", loader, lsize);

    // patch loader
    const unsigned packedsize = ph.c_len;
    const unsigned e_len = getLoaderSectionStart("EXECUTPO");
    const unsigned d_len = lsize - e_len;
    assert((e_len&15) == 0);

    const unsigned copysize = (1+packedsize+d_len) & ~1;
    const unsigned firstcopy = copysize%0x10000 ? copysize%0x10000 : 0x10000;

    // set oh.min & oh.max
    ic = ih.min*16 + ih_imagesize;
    if (ic < oh.ss*16u + oh.sp)
        ic = oh.ss*16u + oh.sp;
    oh.min = (ic - (packedsize + lsize)) / 16;
    ic = oh.min + (ih.max - ih.min);
    oh.max = ic < 0xffff && ih.max != 0xffff ? ic : 0xffff;

    // set extra info
    unsigned char extra_info[9];
    unsigned eisize = 0;
    if (oh.ss != ih.ss)
    {
        set_le16(extra_info+eisize,ih.ss);
        eisize += 2;
        assert((flag & SS) != 0);   // set in fillExeHeader()
    }
    if (oh.sp != ih.sp)
    {
        set_le16(extra_info+eisize,ih.sp);
        eisize += 2;
        assert((flag & SP) != 0);   // set in fillExeHeader()
    }
    if (ih.min != oh.min)
    {
        set_le16(extra_info+eisize,ih.min);
        eisize += 2;
        flag |= MINMEM;
    }
    if (ih.max != oh.max)
    {
        set_le16(extra_info+eisize,ih.max);
        eisize += 2;
        flag |= MAXMEM;
    }
    extra_info[eisize++] = (unsigned char) flag;

    if (M_IS_NRV2B(ph.method) || M_IS_NRV2D(ph.method) || M_IS_NRV2E(ph.method))
        linker->defineSymbol("bx_magic", 0x7FFF + 0x10 * ((packedsize & 15) + 1));

    unsigned decompressor_entry = 1 + (packedsize & 15);
    if (M_IS_LZMA(ph.method))
        decompressor_entry = 0x10;
    linker->defineSymbol("decompressor_entry", decompressor_entry);

    // patch loader
    if (flag & USEJUMP)
    {
        // I use a relocation entry to set the original cs
        unsigned n = getLoaderSectionStart("EXEJUMPF") + 1;
        n += packedsize + 2;
        oh.relocs = 1;
        oh.firstreloc = (n & 0xf) + ((n >> 4) << 16);
    }
    else
    {
        oh.relocs = 0;
        oh.firstreloc = ih.cs * 0x10000 + ih.ip;
    }

    // g++ 3.1 does not like the following line...
//    oh.relocoffs = offsetof(exe_header_t, firstreloc);
    oh.relocoffs = ptr_diff(&oh.firstreloc, &oh);

    linker->defineSymbol("destination_segment", oh.ss - ph.c_len / 16 - e_len / 16);
    linker->defineSymbol("source_segment", e_len / 16 + (copysize - firstcopy) / 16);
    linker->defineSymbol("copy_offset", firstcopy - 2);
    linker->defineSymbol("words_to_copy",firstcopy / 2);

    linker->defineSymbol("exe_stack_sp", oh.sp);
    linker->defineSymbol("exe_stack_ss", oh.ss);
    linker->defineSymbol("interrupt", get_le16(ibuf + 8));
    linker->defineSymbol("attribute", get_le16(ibuf + 4));
    linker->defineSymbol("orig_strategy", get_le16(ibuf + 6));

    const unsigned outputlen = sizeof(oh)+lsize+packedsize+eisize;
    oh.m512 = outputlen & 511;
    oh.p512 = (outputlen + 511) >> 9;

    const char *exeentry = M_IS_LZMA(ph.method) ? "LZMAENTRY" : "EXEENTRY";
    oh.ip = device_driver ? getLoaderSection(exeentry) - 2 : 0;

    defineDecompressorSymbols();
    relocateLoader();
    memcpy(loader, getLoader(), lsize);
    patchPackHeader(loader,e_len);

//fprintf(stderr,"\ne_len=%x d_len=%x c_len=%x oo=%x ulen=%x destp=%x copys=%x images=%x",e_len,d_len,packedsize,ph.overlap_overhead,ph.u_len,destpara,copysize,ih_imagesize);

    // write header + write loader + compressed file
#ifdef TESTING
    if (opt->debug.debug_level)
        printf("\n%d %d %d %d\n",(int)sizeof(oh),e_len,packedsize,d_len);
#endif
    fo->write(&oh,sizeof(oh));
    fo->write(loader,e_len);            // entry
    fo->write(obuf,packedsize);
    fo->write(loader+e_len,d_len);      // decompressor
    fo->write(extra_info,eisize);
    assert(eisize <= 9);
#if 0
    printf("%-13s: program hdr  : %8ld bytes\n", getName(), (long) sizeof(oh));
    printf("%-13s: entry        : %8ld bytes\n", getName(), (long) e_len);
    printf("%-13s: compressed   : %8ld bytes\n", getName(), (long) packedsize);
    printf("%-13s: decompressor : %8ld bytes\n", getName(), (long) d_len);
    printf("%-13s: extra info   : %8ld bytes\n", getName(), (long) eisize);
#endif

    // verify
    verifyOverlappingDecompression();

    // copy the overlay
    copyOverlay(fo, ih_overlay, &obuf);
//fprintf (stderr,"%x %x\n",relocsize,ph.u_len);

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


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

int PackExe::canUnpack()
{
    if (!readFileHeader())
        return false;
    const off_t off = ih.headsize16 * 16;
    fi->seek(off, SEEK_SET);
    bool b = readPackHeader(4096);
    return b && (off + (off_t) ph.c_len <= file_size);
}


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

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

    // read the file
    fi->seek(ih.headsize16*16,SEEK_SET);
    fi->readx(ibuf,ih_imagesize);

    // get compressed data offset
    unsigned e_len = ph.buf_offset + ph.getPackHeaderSize();
    if (ih_imagesize <= e_len + ph.c_len)
        throwCantUnpack("file damaged");

    checkOverlay(ih_overlay);

    // decompress
    decompress(ibuf+e_len,obuf);

    unsigned imagesize = ih_imagesize;
    imagesize--;
    const unsigned char flag = ibuf[imagesize];

    unsigned relocn = 0;
    upx_byte *relocs = obuf + ph.u_len;

    MemBuffer wrkmem;
    if (!(flag & NORELOC))
    {
        relocs -= get_le16(obuf+ph.u_len-2);
        ph.u_len -= 2;

        wrkmem.alloc(4*MAXRELOCS);
        unsigned es = 0, ones = get_le16(relocs);
        const unsigned seghi = get_le16(relocs+2);
        const upx_byte *p = relocs + 4;

        while (ones)
        {
            unsigned di = get_le16(p);
            es += get_le16(p+2);
            bool dorel = true;
            for (p += 4; ones && di < 0x10000; p++)
            {
                if (dorel)
                {
                    set_le16(wrkmem+4*relocn,di);
                    set_le16(wrkmem+2+4*relocn++,es);
                    //printf ("%x\n",es*16+di);
                }
                dorel = true;
                if (*p == 0)
                {
                    const upx_byte *q;
                    for (q = obuf+es*16+di; !(*q == 0x9a && get_le16(q+3) <= seghi); q++)
                        ;
                    di = ptr_diff(q, obuf+es*16) + 3;
                }
                else if (*p == 1)
                {
                    di += 254;
                    if (di < 0x10000)
                        ones--;
                    dorel = false;
                }
                else
                    di += *p;
            }
        }
    }

    // fill new exe header
    memset(&oh,0,sizeof(oh));
    oh.ident = 'M' + 'Z'*256;

    if (relocn)
    {
        oh.relocs = relocn;
        while (relocn & 3)
            set_le32(wrkmem+4*relocn++,0);
    }

    unsigned outputlen = ptr_diff(relocs, obuf) + sizeof(oh) + relocn*4;
    oh.m512 = outputlen & 511;
    oh.p512 = (outputlen + 511) >> 9;
    oh.headsize16 = 2+relocn/4;

    oh.max = ih.max;
    oh.min = ih.min;
    oh.sp = ih.sp;
    oh.ss = ih.ss;

    if (flag & MAXMEM)
        { imagesize -= 2; oh.max = get_le16(ibuf+imagesize); }
    if (flag & MINMEM)
        { imagesize -= 2; oh.min = get_le16(ibuf+imagesize); }
    if (flag & SP)
        { imagesize -= 2; oh.sp = get_le16(ibuf+imagesize); }
    if (flag & SS)
        { imagesize -= 2; oh.ss = get_le16(ibuf+imagesize); }

    unsigned ip = (flag & USEJUMP) ? get_le32(ibuf+imagesize-4) : (unsigned) ih.firstreloc;
    oh.ip = ip & 0xffff;
    oh.cs = ip >> 16;

    oh.relocoffs = sizeof(oh);
    oh.firstreloc = 0;
    if (!fo)
        return;

    // write header + relocations + uncompressed file
    fo->write(&oh,sizeof(oh));
    if (relocn)
        fo->write(wrkmem,relocn*4);
    fo->write(obuf, ptr_diff(relocs, obuf));

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


Linker* PackExe::newLinker() const
{
    return new ElfLinkerX86();
}

/*

memory layout at decompression time
===================================

normal exe
----------

a, at load time

(e - copying code, C - compressed data, d - decompressor+relocator,
 x - not specified, U - uncompressed code+data, R uncompressed relocation)

eeCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCdddd
^ CS:0                                       ^ SS:0

b, after copying

xxxxxxxxxxxxxxxCCCCCCCCCCCCCCCCCCCCCCCCCCCCCCdddd
^ES:DI=0       ^ DS:SI=0                     ^ CS=SS, IP in range 0..0xf

c, after uncompression

UUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUUURRdddd
                                           ^ ES:DI

device driver
-------------

the file has 2 entry points, CS:0 in device driver mode, and
CS:exe_as_device_entry in normal mode. the code in section DEVICEENTRY
sets up the same environment for section EXEENTRY, as it would see in normal
execution mode.

lzma uncompression for normal exes
----------------------------------

(n - nrv2b uncompressor, l - nrv2b compressed lzma + relocator code)

a, at load time

nneelllCCCCCCCCCCCCCCCCCCCCCCCCC

^ CS:0                                       ^ SS:0

b, after nrv2b

nneelllCCCCCCCCCCCCCCCCCCCCCCCCC             dddd
^ CS:0                                       ^ SS:0x10

after this, normal ee code runs

lzma + device driver
--------------------

(D - device driver adapter)

a, at load time

DDnneelllCCCCCCCCCCCCCCCCCCCCCCCCC

*/


/*
vi:ts=4:et
*/