upx_rsize_t mem_size(upx_uint64_t element_size, upx_uint64_t n, upx_uint64_t extra1, upx_uint64_t extra2)
{
assert(element_size > 0);
if (element_size > UPX_RSIZE_MAX) throwCantPack("mem_size 1; take care");
if (n > UPX_RSIZE_MAX) throwCantPack("mem_size 2; take care");
if (extra1 > UPX_RSIZE_MAX) throwCantPack("mem_size 3; take care");
if (extra2 > UPX_RSIZE_MAX) throwCantPack("mem_size 4; take care");
upx_uint64_t bytes = element_size * n + extra1 + extra2; // cannot overflow
if (bytes > UPX_RSIZE_MAX) throwCantPack("mem_size 5; take care");
return ACC_ICONV(upx_rsize_t, bytes);
}
bool PackTos::canPack()
{
if (!readFileHeader())
return false;
unsigned char buf[768];
fi->readx(buf, sizeof(buf));
checkAlreadyPacked(buf, sizeof(buf));
if (!checkFileHeader())
throwCantPack("unsupported header flags");
if (file_size < 1024)
throwCantPack("program is too small");
return true;
}
void PackWcle::encodeEntryTable()
{
unsigned count,object,n;
upx_byte *p = ientries;
n = 0;
while (*p)
{
count = *p;
n += count;
if (p[1] == 0) // unused bundle
p += 2;
else if (p[1] == 3) // 32-bit bundle
{
object = get_le16(p+2)-1;
set_le16(p+2,1);
p += 4;
for (; count; count--, p += 5)
set_le32(p+1,IOT(object,my_base_address) + get_le32(p+1));
}
else
throwCantPack("unsupported bundle type in entry table");
}
//if (Opt_debug) printf("%d entries encoded.\n",n);
UNUSED(n);
soentries = ptr_diff(p, ientries) + 1;
oentries = ientries;
ientries = NULL;
}
bool PackCom::canPack()
{
unsigned char buf[128];
fi->readx(buf, sizeof(buf));
if (memcmp(buf,"MZ",2) == 0 || memcmp(buf,"ZM",2) == 0 // .exe
|| memcmp (buf,"\xff\xff\xff\xff",4) == 0) // .sys
return false;
if (!fn_has_ext(fi->getName(),"com"))
return false;
checkAlreadyPacked(buf, sizeof(buf));
if (file_size < 1024)
throwCantPack("file is too small");
if (file_size > 0xFF00)
throwCantPack("file is too big for dos/com");
return true;
}
void PackArmPe::processTls(Interval *) // pass 1
{
if ((sotls = ALIGN_UP(IDSIZE(PEDIR_TLS),4u)) == 0)
return;
// never should happen on wince
throwCantPack("Static TLS entries found. Send a report please.");
}
// common part of canPack(), enhanced by subclasses
bool PackUnix::canPack()
{
if (exetype == 0)
return false;
#if defined(__unix__)
// must be executable by owner
if ((fi->st.st_mode & S_IXUSR) == 0)
throwCantPack("file not executable; try 'chmod +x'");
#endif
if (file_size < 4096)
throwCantPack("file is too small");
// info: currently the header is 36 (32+4) bytes before EOF
unsigned char buf[256];
fi->seek(-(off_t)sizeof(buf), SEEK_END);
fi->readx(buf, sizeof(buf));
checkAlreadyPacked(buf, sizeof(buf));
return true;
}
bool PackTos::checkFileHeader()
{
const unsigned f = ih.fh_flag;
//printf("flags: 0x%x, text: %d, data: %d, bss: %d, sym: %d\n", f, (int)ih.fh_text, (int)ih.fh_data, (int)ih.fh_bss, (int)ih.fh_sym);
if ((ih.fh_text & 1) || (ih.fh_data & 1))
throwCantPack("odd size values in text/data");
if (f & F_OS_SPECIAL)
throwCantPack("I won't pack F_OS_SPECIAL programs");
if ((f & F_PROTMODE) > F_PROT_I)
throwCantPack("invalid protection mode");
if ((f & F_PROTMODE) != F_PROT_P)
{
if (opt->force < 1)
throwCantPack("no private memory protection; use option '-f' to force packing");
}
if (f & F_SHTEXT)
{
if (opt->force < 1)
throwCantPack("shared text segment; use option '-f' to force packing");
}
#if 0
// fh_reserved seems to be unused
if (ih.fh_reserved != 0)
{
if (opt->force < 1)
throwCantPack("reserved header field set; use option '-f' to force packing");
}
#endif
return true;
}
static const
#include "stub/i086-dos16.sys.h"
/*************************************************************************
//
**************************************************************************/
bool PackSys::canPack()
{
unsigned char buf[128];
fi->readx(buf, sizeof(buf));
if (memcmp (buf,"\xff\xff\xff\xff",4) != 0)
return false;
if (!fn_has_ext(fi->getName(),"sys"))
return false;
checkAlreadyPacked(buf, sizeof(buf));
if (file_size < 1024)
throwCantPack("file is too small");
if (file_size > 0x10000)
throwCantPack("file is too big for dos/sys");
return true;
}
void PackVmlinuzI386::readKernel()
{
int klen = decompressKernel();
if (klen <= 0)
throwCantPack("kernel decompression failed");
//OutputFile::dump("kernel.img", ibuf, klen);
// copy the setup boot code
setup_buf.alloc(setup_size);
memcpy(setup_buf, obuf, setup_size);
//OutputFile::dump("setup.img", setup_buf, setup_size);
obuf.dealloc();
obuf.allocForCompression(klen);
ph.u_len = klen;
ph.filter = 0;
}
void PackWcle::readObjectTable()
{
LeFile::readObjectTable();
// temporary copy of the object descriptors
iobject_desc.alloc(objects*sizeof(*iobject_table));
memcpy(iobject_desc,iobject_table,objects*sizeof(*iobject_table));
unsigned ic,jc,virtual_size;
for (ic = jc = virtual_size = 0; ic < objects; ic++)
{
jc += IOT(ic,npages);
IOT(ic,my_base_address) = virtual_size;
virtual_size += (IOT(ic,virtual_size)+mps-1) &~ (mps-1);
}
if (pages != jc)
throwCantPack("bad page number");
}
void PackCom::patchLoader(OutputFile *fo,
upx_byte *loader, int lsize,
unsigned calls)
{
const int e_len = getLoaderSectionStart("COMCUTPO");
const int d_len = lsize - e_len;
assert(e_len > 0 && e_len < 128);
assert(d_len > 0 && d_len < 256);
const unsigned upper_end = ph.u_len + ph.overlap_overhead + d_len + 0x100;
unsigned stacksize = 0x60;
if (upper_end + stacksize > 0xfffe)
stacksize = 0x56;
if (upper_end + stacksize > 0xfffe)
throwCantPack("file is too big for dos/com");
linker->defineSymbol("calltrick_calls", calls);
linker->defineSymbol("sp_limit", upper_end + stacksize);
linker->defineSymbol("bytes_to_copy", ph.c_len + lsize);
linker->defineSymbol("copy_source", ph.c_len + lsize + 0x100);
linker->defineSymbol("copy_destination", upper_end);
linker->defineSymbol("neg_e_len", 0 - e_len);
linker->defineSymbol("NRV2B160", ph.u_len + ph.overlap_overhead);
relocateLoader();
loader = getLoader();
// some day we could use the relocation stuff for patchPackHeader too
patchPackHeader(loader,e_len);
// write loader + compressed file
fo->write(loader,e_len); // entry
fo->write(obuf,ph.c_len);
fo->write(loader+e_len,d_len); // decompressor
#if 0
printf("%-13s: entry : %8ld bytes\n", getName(), (long) e_len);
printf("%-13s: compressed : %8ld bytes\n", getName(), (long) ph.c_len);
printf("%-13s: decompressor : %8ld bytes\n", getName(), (long) d_len);
#endif
}
// read full kernel into obuf[], gzip-decompress into ibuf[],
// return decompressed size
int PackVmlinuzI386::decompressKernel()
{
// read whole kernel image
obuf.alloc(file_size);
fi->seek(0, SEEK_SET);
fi->readx(obuf, file_size);
{
const upx_byte *base = NULL;
unsigned relocated = 0;
// See startup_32: in linux/arch/i386/boot/compressed/head.S
const upx_byte *p = &obuf[setup_size];
unsigned cpa_0 = 0;
unsigned cpa_1 = 0;
int j;
if (0x205<=h.version) {
cpa_0 = h.kernel_alignment;
cpa_1 = 0u - cpa_0;
} else
for ((p = &obuf[setup_size]), (j= 0); j < 0x200; ++j, ++p) {
if (0==memcmp("\x89\xeb\x81\xc3", p, 4)
&& 0==memcmp("\x81\xe3", 8+ p, 2)) {
// movl %ebp,%ebx
// addl $imm.w,%ebx
// andl $imm.w,%ebx
cpa_0 = 1+ get_te32( 4+ p);
cpa_1 = get_te32(10+ p);
break;
}
}
for ((p = &obuf[setup_size]), (j= 0); j < 0x200; ++j, ++p) {
if (0==memcmp("\x8d\x83", p, 2) // leal d32(%ebx),%eax
&& 0==memcmp("\xff\xe0", 6+ p, 2) // jmp *%eax
) {
relocated = get_te32(2+ p);
}
if (0==memcmp("\xE8\x00\x00\x00\x00\x5D", p, 6)) {
// "call 1f; 1f: pop %ebp" determines actual execution address.
// linux-2.6.21 (spring 2007) and later; upx stub needs work
// unless LOAD_PHYSICAL_ADDR is known.
// Allowed code is: linux-2.6.23/arch/x86/head_32.S 2008-01-01
// call 1f
// 1: popl %ebp
// subl $1b, %ebp # 32-bit immediate
// movl $LOAD_PHYSICAL_ADDR, %ebx
//
if (0==memcmp("\x81\xed", 6+ p, 2) // subl $imm.w,%ebp
&& 0==memcmp("\xbb", 12+ p, 1) ) { // movl $imm.w,%ebx
physical_start = get_te32(13+ p);
} else
if (0==memcmp("\x81\xed", 6+ p, 2) // subl $imm.w,%ebp
&& is_pow2(cpa_0) && (0u-cpa_0)==cpa_1) {
base = (5+ p) - get_te32(8+ p);
config_physical_align = cpa_0;
}
else {
throwCantPack("Unrecognized relocatable kernel");
}
}
// Find "ljmp $__BOOT_CS,$__PHYSICAL_START" if any.
if (0==memcmp("\xEA\x00\x00", p, 3) && 0==(0xf & p[3]) && 0==p[4]) {
/* whole megabyte < 16 MiB */
physical_start = get_te32(1+ p);
break;
}
}
if (base && relocated) {
p = base + relocated;
for (j = 0; j < 0x200; ++j, ++p) {
if (0==memcmp("\x01\x9c\x0b", p, 3) // addl %ebx,d32(%ebx,%ecx)
) {
page_offset = 0u - get_te32(3+ p);
}
if (0==memcmp("\x89\xeb", p, 2) // movl %ebp,%ebx
&& 0==memcmp("\x81\xeb", 2+ p, 2) // subl $imm32,%ebx
) {
physical_start = get_te32(4+ p);
}
}
}
}
checkAlreadyPacked(obuf + setup_size, UPX_MIN(file_size - setup_size, (off_t)1024));
int gzoff = setup_size;
if (0x208<=h.version) {
gzoff += h.payload_offset;
}
for (; gzoff < file_size; gzoff++)
{
// find gzip header (2 bytes magic + 1 byte method "deflated")
int off = find(obuf + gzoff, file_size - gzoff, "\x1F\x8B\x08", 3);
if (off < 0)
break;
gzoff += off;
const int gzlen = (h.version < 0x208) ? (file_size - gzoff) : h.payload_length;
if (gzlen < 256)
break;
// check gzip flag byte
unsigned char flags = obuf[gzoff + 3];
if ((flags & 0xe0) != 0) // reserved bits set
continue;
//printf("found gzip header at offset %d\n", gzoff);
// try to decompress
int klen;
int fd;
off_t fd_pos;
for (;;)
{
klen = -1;
fd = -1;
fd_pos = -1;
// open
fi->seek(gzoff, SEEK_SET);
fd = dup(fi->getFd());
if (fd < 0)
break;
gzFile zf = gzdopen(fd, "rb");
if (zf == NULL)
break;
// estimate gzip-decompressed kernel size & alloc buffer
if (ibuf.getSize() == 0)
ibuf.alloc(gzlen * 3);
// decompress
klen = gzread(zf, ibuf, ibuf.getSize());
fd_pos = lseek(fd, 0, SEEK_CUR);
gzclose(zf);
fd = -1;
if (klen != (int)ibuf.getSize())
break;
// realloc and try again
unsigned s = ibuf.getSize();
ibuf.dealloc();
ibuf.alloc(3 * s / 2);
}
if (fd >= 0)
(void) close(fd);
if (klen <= 0)
continue;
if (klen <= gzlen)
continue;
if (0x208<=h.version && 0==memcmp("\177ELF", ibuf, 4)) {
// Full ELF in theory; for now, try to handle as .bin at physical_start.
// Check for PT_LOAD.p_paddr being ascending and adjacent.
Elf_LE32_Ehdr const *const ehdr = (Elf_LE32_Ehdr const *)(void const *)ibuf;
Elf_LE32_Phdr const *phdr = (Elf_LE32_Phdr const *)(ehdr->e_phoff + (char const *)ehdr);
Elf_LE32_Shdr const *shdr = (Elf_LE32_Shdr const *)(ehdr->e_shoff + (char const *)ehdr);
unsigned hi_paddr = 0, lo_paddr = 0;
unsigned delta_off = 0;
for (unsigned j=0; j < ehdr->e_phnum; ++j, ++phdr) {
if (phdr->PT_LOAD==phdr->p_type) {
unsigned step = (hi_paddr + phdr->p_align - 1) & ~(phdr->p_align - 1);
if (0==hi_paddr) { // first PT_LOAD
if (physical_start!=phdr->p_paddr) {
return 0;
}
delta_off = phdr->p_paddr - phdr->p_offset;
lo_paddr = phdr->p_paddr;
hi_paddr = phdr->p_filesz + phdr->p_paddr;
}
else if (step==phdr->p_paddr
&& delta_off==(phdr->p_paddr - phdr->p_offset)) {
hi_paddr = phdr->p_filesz + phdr->p_paddr;
}
else {
return 0; // Not equivalent to a .bin. Too complex for now.
}
}
}
// FIXME: ascending order is only a convention; might need sorting.
for (unsigned j=1; j < ehdr->e_shnum; ++j) {
if (shdr->SHT_PROGBITS==shdr->sh_type) { // SHT_REL might be intermixed
if (shdr->SHF_EXECINSTR & shdr[j].sh_flags) {
filter_len += shdr[j].sh_size; // FIXME: include sh_addralign
}
else {
break;
}
}
}
memmove(ibuf, (lo_paddr - delta_off) + ibuf, hi_paddr - lo_paddr); // FIXME: set_size
// FIXME: .bss ? Apparently handled by head.S
}
if (opt->force > 0)
return klen;
// some checks
if (fd_pos != file_size)
{
//printf("fd_pos: %ld, file_size: %ld\n", (long)fd_pos, (long)file_size);
// linux-2.6.21.5/arch/i386/boot/compressed/vmlinux.lds
// puts .data.compressed ahead of .text, .rodata, etc;
// so piggy.o need not be last in bzImage. Alas.
//throwCantPack("trailing bytes after kernel image; use option '-f' to force packing");
}
// see /usr/src/linux/arch/i386/kernel/head.S
// 2.4.x: [cli;] cld; mov $...,%eax
if (memcmp(ibuf, "\xFC\xB8", 2) == 0) goto head_ok;
if (memcmp(ibuf, "\xFA\xFC\xB8", 3) == 0) goto head_ok;
// 2.6.21.5 CONFIG_PARAVIRT mov %cs,%eax; test $3,%eax; jne ...;
if (memcmp(ibuf, "\x8c\xc8\xa9\x03\x00\x00\x00\x0f\x85", 9) == 0) goto head_ok;
if (memcmp(ibuf, "\x8c\xc8\xa8\x03\x0f\x85", 6) == 0) goto head_ok;
// 2.6.x: [cli;] cld; lgdt ...
if (memcmp(ibuf, "\xFC\x0F\x01", 3) == 0) goto head_ok;
if (memcmp(ibuf, "\xFA\xFC\x0F\x01", 4) == 0) goto head_ok;
// 2.6.x+grsecurity+strongswan+openwall+trustix: ljmp $0x10,...
if (ibuf[0] == 0xEA && memcmp(ibuf+5, "\x10\x00", 2) == 0) goto head_ok;
// x86_64 2.6.x
if (0xB8==ibuf[0] // mov $...,%eax
&& 0x8E==ibuf[5] && 0xD8==ibuf[6] // mov %eax,%ds
&& 0x0F==ibuf[7] && 0x01==ibuf[8] && 020==(070 & ibuf[9]) // lgdtl
&& 0xB8==ibuf[14] // mov $...,%eax
&& 0x0F==ibuf[19] && 0xA2==ibuf[20] // cpuid
) goto head_ok;
// cmpw $0x207,0x206(%esi) Debian vmlinuz-2.6.24-12-generic
if (0==memcmp("\x66\x81\xbe\x06\x02\x00\x00\x07\x02", ibuf, 9)) goto head_ok;
// testb $0x40,0x211(%esi) Fedora vmlinuz-2.6.25-0.218.rc8.git7.fc9.i686
if (0==memcmp("\xf6\x86\x11\x02\x00\x00\x40", ibuf, 7)) goto head_ok;
// rex.W prefix for x86_64
if (0x48==ibuf[0]) throwCantPack("x86_64 bzImage is not yet supported");
throwCantPack("unrecognized kernel architecture; use option '-f' to force packing");
head_ok:
// FIXME: more checks for special magic bytes in ibuf ???
// FIXME: more checks for kernel architecture ???
return klen;
}
return 0;
}
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 || (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 || (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 || (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: ~28 KiB 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 || (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();
}
void PackTmt::pack(OutputFile *fo)
{
big_relocs = 0;
Packer::handleStub(fi,fo,adam_offset);
const unsigned usize = ih.imagesize;
const unsigned rsize = ih.relocsize;
ibuf.alloc(usize+rsize+128);
obuf.allocForCompression(usize+rsize+128);
MemBuffer wrkmem;
wrkmem.alloc(rsize+EXTRA_INFO); // relocations
fi->seek(adam_offset+sizeof(ih),SEEK_SET);
fi->readx(ibuf,usize);
fi->readx(wrkmem+4,rsize);
const unsigned overlay = file_size - fi->tell();
if (find_le32(ibuf,128,get_le32("UPX ")) >= 0)
throwAlreadyPacked();
if (rsize == 0)
throwCantPack("file is already compressed with another packer");
checkOverlay(overlay);
unsigned relocsize = 0;
//if (rsize)
{
for (unsigned ic=4; ic<=rsize; ic+=4)
set_le32(wrkmem+ic,get_le32(wrkmem+ic)-4);
relocsize = ptr_diff(optimizeReloc32(wrkmem+4,rsize/4,wrkmem,ibuf,1,&big_relocs), wrkmem);
}
wrkmem[relocsize++] = 0;
set_le32(wrkmem+relocsize,ih.entry); // save original entry point
relocsize += 4;
set_le32(wrkmem+relocsize,relocsize+4);
relocsize += 4;
memcpy(ibuf+usize,wrkmem,relocsize);
// prepare packheader
ph.u_len = usize + relocsize;
// prepare filter
Filter ft(ph.level);
ft.buf_len = usize;
// compress
upx_compress_config_t cconf; cconf.reset();
// limit stack size needed for runtime decompression
cconf.conf_lzma.max_num_probs = 1846 + (768 << 4); // ushort: ~28 KiB stack
compressWithFilters(&ft, 512, &cconf);
const unsigned lsize = getLoaderSize();
const unsigned s_point = getLoaderSection("TMTMAIN1");
int e_len = getLoaderSectionStart("TMTCUTPO");
const unsigned d_len = lsize - e_len;
assert(e_len > 0 && s_point > 0);
// patch loader
linker->defineSymbol("original_entry", ih.entry);
defineDecompressorSymbols();
defineFilterSymbols(&ft);
linker->defineSymbol("bytes_to_copy", ph.c_len + d_len);
linker->defineSymbol("copy_dest", 0u - (ph.u_len + ph.overlap_overhead + d_len - 1));
linker->defineSymbol("copy_source", ph.c_len + lsize - 1);
//fprintf(stderr,"\nelen=%x dlen=%x copy_len=%x copy_to=%x oo=%x jmp_pos=%x ulen=%x c_len=%x \n\n",
// e_len,d_len,copy_len,copy_to,ph.overlap_overhead,jmp_pos,ph.u_len,ph.c_len);
linker->defineSymbol("TMTCUTPO", ph.u_len + ph.overlap_overhead);
relocateLoader();
MemBuffer loader(lsize);
memcpy(loader,getLoader(),lsize);
patchPackHeader(loader,e_len);
memcpy(&oh,&ih,sizeof(oh));
oh.imagesize = ph.c_len + lsize; // new size
oh.entry = s_point; // new entry point
oh.relocsize = 4;
// write loader + compressed file
fo->write(&oh,sizeof(oh));
fo->write(loader,e_len);
fo->write(obuf,ph.c_len);
fo->write(loader+lsize-d_len,d_len); // decompressor
char rel_entry[4];
set_le32(rel_entry,5 + s_point);
fo->write(rel_entry,sizeof (rel_entry));
// verify
verifyOverlappingDecompression();
// copy the overlay
copyOverlay(fo, overlay, &obuf);
// finally check the compression ratio
if (!checkFinalCompressionRatio(fo))
throwNotCompressible();
}
void throwCantUnpack(const char *msg)
{
// UGLY, but makes things easier
throwCantPack(msg);
}
void throwCantPackExact()
{
throwCantPack("option '--exact' does not work with this file");
}
void PackWcle::pack(OutputFile *fo)
{
handleStub(fo);
if (ih.byte_order || ih.word_order
|| ih.exe_format_level
|| ih.cpu_type < 2 || ih.cpu_type > 5
|| ih.target_os != 1
|| ih.module_type != 0x200
|| ih.object_iterate_data_map_offset
|| ih.resource_entries
|| ih.module_directives_entries
|| ih.imported_modules_count
|| ih.object_table_entries > 255)
throwCantPack("watcom/le: unexpected value in header");
readObjectTable();
readPageMap();
readResidentNames();
readEntryTable();
readFixupPageTable();
readFixups();
readImage();
readNonResidentNames();
// if (find_le32(iimage,20,get_le32("UPX ")) >= 0)
if (find_le32(iimage,UPX_MIN(soimage,256u),UPX_MAGIC_LE32) >= 0)
throwAlreadyPacked();
if (ih.init_ss_object != objects)
throwCantPack("the stack is not in the last object");
preprocessFixups();
const unsigned text_size = IOT(ih.init_cs_object-1,npages) * mps;
const unsigned text_vaddr = IOT(ih.init_cs_object-1,my_base_address);
// attach some useful data at the end of preprocessed fixups
ifixups[sofixups++] = (unsigned char) (ih.automatic_data_object & 0xff);
unsigned ic = objects*sizeof(*iobject_table);
memcpy(ifixups+sofixups,iobject_desc,ic);
iobject_desc.dealloc();
sofixups += ic;
set_le32(ifixups+sofixups,ih.init_esp_offset+IOT(ih.init_ss_object-1,my_base_address)); // old stack pointer
set_le32(ifixups+sofixups+4,ih.init_eip_offset+text_vaddr); // real entry point
set_le32(ifixups+sofixups+8,mps*pages); // virtual address of unpacked relocations
ifixups[sofixups+12] = (unsigned char) (unsigned) objects;
sofixups += 13;
// prepare filter
Filter ft(ph.level);
ft.buf_len = text_size;
ft.addvalue = text_vaddr;
// compress
encodeImage(&ft);
const unsigned lsize = getLoaderSize();
neweip = getLoaderSection("WCLEMAIN");
int e_len = getLoaderSectionStart("WCLECUTP");
const unsigned d_len = lsize - e_len;
assert(e_len > 0 && e_len < RESERVED);
memmove(oimage+e_len,oimage+RESERVED,soimage);
soimage += lsize;
opages = (soimage+mps-1)/mps;
oh.bytes_on_last_page = soimage%mps;
encodeObjectTable();
encodeFixups();
encodeFixupPageTable();
encodePageMap();
encodeEntryTable();
encodeResidentNames();
encodeNonResidentNames();
// patch loader
ic = (OOT(0,virtual_size) - d_len) &~ 15;
assert(ic > ((ph.u_len + ph.overlap_overhead + 31) &~ 15));
linker->defineSymbol("WCLECUTP", ic);
linker->defineSymbol("original_entry", ih.init_eip_offset + text_vaddr);
linker->defineSymbol("original_stack", ih.init_esp_offset +
IOT(ih.init_ss_object - 1, my_base_address));
linker->defineSymbol("start_of_relocs", mps*pages);
defineDecompressorSymbols();
defineFilterSymbols(&ft);
linker->defineSymbol("filter_buffer_start", text_vaddr);
unsigned jpos = (((ph.c_len + 3) &~ 3) + d_len + 3) / 4;
linker->defineSymbol("words_to_copy", jpos);
linker->defineSymbol("copy_dest", ((ic + d_len + 3) &~ 3) - 4);
linker->defineSymbol("copy_source", e_len + jpos * 4 - 4);
relocateLoader();
MemBuffer loader(lsize);
memcpy(loader, getLoader(), lsize);
patchPackHeader(loader, lsize);
memcpy(oimage, loader, e_len);
memcpy(oimage + soimage - d_len, loader + e_len, d_len);
writeFile(fo, opt->watcom_le.le);
// verify
verifyOverlappingDecompression(oimage + e_len, oimage.getSize() - e_len);
// copy the overlay
const unsigned overlaystart = ih.data_pages_offset + exe_offset
+ getImageSize();
const unsigned overlay = file_size - overlaystart - ih.non_resident_name_table_length;
checkOverlay(overlay);
copyOverlay(fo, overlay, &oimage);
// finally check the compression ratio
if (!checkFinalCompressionRatio(fo))
throwNotCompressible();
}
void PackWcle::preprocessFixups()
{
big_relocs = 0;
unsigned ic,jc;
Array(unsigned, counts, objects+2);
countFixups(counts);
for (ic = jc = 0; ic < objects; ic++)
jc += counts[ic];
if (jc == 0)
{
// FIXME: implement this
throwCantPack("files without relocations are not supported");
}
ByteArray(rl, jc);
ByteArray(srf, counts[objects+0]+1);
ByteArray(slf, counts[objects+1]+1);
upx_byte *selector_fixups = srf;
upx_byte *selfrel_fixups = slf;
unsigned rc = 0;
upx_byte *fix = ifixups;
for (ic = jc = 0; ic < pages; ic++)
{
while ((unsigned)(fix - ifixups) < get_le32(ifpage_table+ic+1))
{
const int fixp2 = get_le16_signed(fix+2);
unsigned value;
switch (*fix)
{
case 2: // selector fixup
if (fixp2 < 0)
{
// cross page selector fixup
dputc('S',stdout);
fix += 5;
break;
}
dputc('s',stdout);
memcpy(selector_fixups,"\x8C\xCB\x66\x89\x9D",5); // mov bx, cs ; mov [xxx+ebp], bx
if (IOT(fix[4]-1,flags) & LEOF_WRITE)
selector_fixups[1] = 0xDB; // ds
set_le32(selector_fixups+5,jc+fixp2);
selector_fixups += 9;
fix += 5;
break;
case 5: // 16-bit offset
if ((unsigned)fixp2 < 4096 && IOT(fix[4]-1,my_base_address) == jc)
dputc('6',stdout);
else
throwCantPack("unsupported 16-bit offset relocation");
fix += (fix[1] & 0x10) ? 9 : 7;
break;
case 6: // 16:32 pointer
if (fixp2 < 0)
{
// cross page pointer fixup
dputc('P',stdout);
fix += (fix[1] & 0x10) ? 9 : 7;
break;
}
dputc('p',stdout);
memcpy(iimage+jc+fixp2,fix+5,(fix[1] & 0x10) ? 4 : 2);
set_le32(rl+4*rc++,jc+fixp2);
set_le32(iimage+jc+fixp2,get_le32(iimage+jc+fixp2)+IOT(fix[4]-1,my_base_address));
memcpy(selector_fixups,"\x8C\xCA\x66\x89\x95",5);
if (IOT(fix[4]-1,flags) & LEOF_WRITE)
selector_fixups[1] = 0xDA; // ds
set_le32(selector_fixups+5,jc+fixp2+4);
selector_fixups += 9;
fix += (fix[1] & 0x10) ? 9 : 7;
break;
case 7: // 32-bit offset
if (fixp2 < 0)
{
fix += (fix[1] & 0x10) ? 9 : 7;
break;
}
//if (memcmp(iimage+jc+fixp2,fix+5,(fix[1] & 0x10) ? 4 : 2))
// throwCantPack("illegal fixup offset");
// work around a pmwunlite bug: remove duplicated fixups
// FIXME: fix the other cases too
if (rc == 0 || get_le32(rl+4*rc-4) != jc+fixp2)
{
set_le32(rl+4*rc++,jc+fixp2);
set_le32(iimage+jc+fixp2,get_le32(iimage+jc+fixp2)+IOT(fix[4]-1,my_base_address));
}
fix += (fix[1] & 0x10) ? 9 : 7;
break;
case 8: // 32-bit self relative fixup
if (fixp2 < 0)
{
// cross page self relative fixup
dputc('R',stdout);
fix += (fix[1] & 0x10) ? 9 : 7;
break;
}
value = get_le32(fix+5);
if (fix[1] == 0)
value &= 0xffff;
set_le32(iimage+jc+fixp2,(value+IOT(fix[4]-1,my_base_address))-jc-fixp2-4);
set_le32(selfrel_fixups,jc+fixp2);
selfrel_fixups += 4;
dputc('r',stdout);
fix += (fix[1] & 0x10) ? 9 : 7;
break;
default:
throwCantPack("unsupported fixup record");
}
}
jc += mps;
}
// resize ifixups if it's too small
if (sofixups < 1000)
{
delete[] ifixups;
ifixups = new upx_byte[1000];
}
fix = optimizeReloc32 (rl,rc,ifixups,iimage,1,&big_relocs);
has_extra_code = srf != selector_fixups;
// FIXME: this could be removed if has_extra_code = false
// but then we'll need a flag
*selector_fixups++ = 0xC3; // ret
memcpy(fix,srf,selector_fixups-srf); // copy selector fixup code
fix += selector_fixups-srf;
memcpy(fix,slf,selfrel_fixups-slf); // copy self-relative fixup positions
fix += selfrel_fixups-slf;
set_le32(fix,0xFFFFFFFFUL);
fix += 4;
sofixups = ptr_diff(fix, ifixups);
}
void throwFilterException()
{
throwCantPack("filter problem");
}
void PackArmPe::pack(OutputFile *fo)
{
// FIXME: we need to think about better support for --exact
if (opt->exact)
throwCantPackExact();
const unsigned objs = ih.objects;
isection = new pe_section_t[objs];
fi->seek(pe_offset+sizeof(ih),SEEK_SET);
fi->readx(isection,sizeof(pe_section_t)*objs);
rvamin = isection[0].vaddr;
infoHeader("[Processing %s, format %s, %d sections]", fn_basename(fi->getName()), getName(), objs);
// check the PE header
// FIXME: add more checks
if (!opt->force && (
(ih.cpu != 0x1c0 && ih.cpu != 0x1c2)
|| (ih.opthdrsize != 0xe0)
|| ((ih.flags & EXECUTABLE) == 0)
|| (ih.subsystem != 9)
|| (ih.entry == 0 /*&& !isdll*/)
|| (ih.ddirsentries != 16)
// || IDSIZE(PEDIR_EXCEPTION) // is this used on arm?
// || IDSIZE(PEDIR_COPYRIGHT)
))
throwCantPack("unexpected value in PE header (try --force)");
if (IDSIZE(PEDIR_SEC))
IDSIZE(PEDIR_SEC) = IDADDR(PEDIR_SEC) = 0;
// throwCantPack("compressing certificate info is not supported");
if (IDSIZE(PEDIR_COMRT))
throwCantPack(".NET files (win32/net) are not yet supported");
if (isdll)
opt->win32_pe.strip_relocs = false;
else if (opt->win32_pe.strip_relocs < 0)
opt->win32_pe.strip_relocs = (ih.imagebase >= 0x10000);
if (opt->win32_pe.strip_relocs)
{
if (ih.imagebase < 0x10000)
throwCantPack("--strip-relocs is not allowed when imagebase < 0x10000");
else
ih.flags |= RELOCS_STRIPPED;
}
if (memcmp(isection[0].name,"UPX",3) == 0)
throwAlreadyPackedByUPX();
if (!opt->force && IDSIZE(15))
throwCantPack("file is possibly packed/protected (try --force)");
if (ih.entry && ih.entry < rvamin)
throwCantPack("run a virus scanner on this file!");
if (!opt->force && ih.subsystem == 1)
throwCantPack("subsystem 'native' is not supported (try --force)");
if (ih.filealign < 0x200)
throwCantPack("filealign < 0x200 is not yet supported");
handleStub(fi,fo,pe_offset);
const unsigned usize = ih.imagesize;
const unsigned xtrasize = UPX_MAX(ih.datasize, 65536u) + IDSIZE(PEDIR_IMPORT) + IDSIZE(PEDIR_BOUNDIM) + IDSIZE(PEDIR_IAT) + IDSIZE(PEDIR_DELAYIMP) + IDSIZE(PEDIR_RELOC);
ibuf.alloc(usize + xtrasize);
// BOUND IMPORT support. FIXME: is this ok?
fi->seek(0,SEEK_SET);
fi->readx(ibuf,isection[0].rawdataptr);
Interval holes(ibuf);
unsigned ic,jc,overlaystart = 0;
ibuf.clear(0, usize);
for (ic = jc = 0; ic < objs; ic++)
{
if (isection[ic].rawdataptr && overlaystart < isection[ic].rawdataptr + isection[ic].size)
overlaystart = ALIGN_UP(isection[ic].rawdataptr + isection[ic].size,ih.filealign);
if (isection[ic].vsize == 0)
isection[ic].vsize = isection[ic].size;
if ((isection[ic].flags & PEFL_BSS) || isection[ic].rawdataptr == 0
|| (isection[ic].flags & PEFL_INFO))
{
holes.add(isection[ic].vaddr,isection[ic].vsize);
continue;
}
if (isection[ic].vaddr + isection[ic].size > usize)
throwCantPack("section size problem");
if (((isection[ic].flags & (PEFL_WRITE|PEFL_SHARED))
== (PEFL_WRITE|PEFL_SHARED)))
if (!opt->force)
throwCantPack("writable shared sections not supported (try --force)");
if (jc && isection[ic].rawdataptr - jc > ih.filealign)
throwCantPack("superfluous data between sections");
fi->seek(isection[ic].rawdataptr,SEEK_SET);
jc = isection[ic].size;
if (jc > isection[ic].vsize)
jc = isection[ic].vsize;
if (isection[ic].vsize == 0) // hack for some tricky programs - may this break other progs?
jc = isection[ic].vsize = isection[ic].size;
if (isection[ic].vaddr + jc > ibuf.getSize())
throwInternalError("buffer too small 1");
fi->readx(ibuf + isection[ic].vaddr,jc);
jc += isection[ic].rawdataptr;
}
// check for NeoLite
if (find(ibuf + ih.entry, 64+7, "NeoLite", 7) >= 0)
throwCantPack("file is already compressed with another packer");
unsigned overlay = file_size - stripDebug(overlaystart);
if (overlay >= (unsigned) file_size)
{
#if 0
if (overlay < file_size + ih.filealign)
overlay = 0;
else if (!opt->force)
throwNotCompressible("overlay problem (try --force)");
#endif
overlay = 0;
}
checkOverlay(overlay);
Resource res;
Interval tlsiv(ibuf);
Export xport((char*)(unsigned char*)ibuf);
const unsigned dllstrings = processImports();
processTls(&tlsiv); // call before processRelocs!!
processResources(&res);
processExports(&xport);
processRelocs();
//OutputFile::dump("x1", ibuf, usize);
// some checks for broken linkers - disable filter if necessary
bool allow_filter = true;
if (ih.codebase == ih.database
|| ih.codebase + ih.codesize > ih.imagesize
|| (isection[virta2objnum(ih.codebase,isection,objs)].flags & PEFL_CODE) == 0)
allow_filter = false;
const unsigned oam1 = ih.objectalign - 1;
// FIXME: disabled: the uncompressor would not allocate enough memory
//objs = tryremove(IDADDR(PEDIR_RELOC),objs);
// FIXME: if the last object has a bss then this won't work
// newvsize = (isection[objs-1].vaddr + isection[objs-1].size + oam1) &~ oam1;
// temporary solution:
unsigned newvsize = (isection[objs-1].vaddr + isection[objs-1].vsize + oam1) &~ oam1;
//fprintf(stderr,"newvsize=%x objs=%d\n",newvsize,objs);
if (newvsize + soimport + sorelocs > ibuf.getSize())
throwInternalError("buffer too small 2");
memcpy(ibuf+newvsize,oimport,soimport);
memcpy(ibuf+newvsize+soimport,orelocs,sorelocs);
cimports = newvsize - rvamin; // rva of preprocessed imports
crelocs = cimports + soimport; // rva of preprocessed fixups
ph.u_len = newvsize + soimport + sorelocs;
// some extra data for uncompression support
unsigned s = 0;
upx_byte * const p1 = ibuf + ph.u_len;
memcpy(p1 + s,&ih,sizeof (ih));
s += sizeof (ih);
memcpy(p1 + s,isection,ih.objects * sizeof(*isection));
s += ih.objects * sizeof(*isection);
if (soimport)
{
set_le32(p1 + s,cimports);
set_le32(p1 + s + 4,dllstrings);
s += 8;
}
if (sorelocs)
{
set_le32(p1 + s,crelocs);
p1[s + 4] = (unsigned char) (big_relocs & 6);
s += 5;
}
if (soresources)
{
set_le16(p1 + s,icondir_count);
s += 2;
}
// end of extra data
set_le32(p1 + s,ptr_diff(p1,ibuf) - rvamin);
s += 4;
ph.u_len += s;
obuf.allocForCompression(ph.u_len);
// prepare packheader
ph.u_len -= rvamin;
// prepare filter
Filter ft(ph.level);
ft.buf_len = ih.codesize;
ft.addvalue = ih.codebase - rvamin;
// compress
int filter_strategy = allow_filter ? 0 : -3;
// disable filters for files with broken headers
if (ih.codebase + ih.codesize > ph.u_len)
{
ft.buf_len = 1;
filter_strategy = -3;
}
// limit stack size needed for runtime decompression
upx_compress_config_t cconf; cconf.reset();
cconf.conf_lzma.max_num_probs = 1846 + (768 << 4); // ushort: ~28 KiB stack
compressWithFilters(&ft, 2048, &cconf, filter_strategy,
ih.codebase, rvamin, 0, NULL, 0);
// info: see buildLoader()
newvsize = (ph.u_len + rvamin + ph.overlap_overhead + oam1) &~ oam1;
/*
if (tlsindex && ((newvsize - ph.c_len - 1024 + oam1) &~ oam1) > tlsindex + 4)
tlsindex = 0;
*/
const unsigned lsize = getLoaderSize();
int identsize = 0;
const unsigned codesize = getLoaderSection("IDENTSTR",&identsize);
assert(identsize > 0);
getLoaderSection("UPX1HEAD",(int*)&ic);
identsize += ic;
pe_section_t osection[4];
// section 0 : bss
// 1 : [ident + header] + packed_data + unpacker + tls
// 2 : not compressed data
// 3 : resource data -- wince 5 needs a new section for this
// identsplit - number of ident + (upx header) bytes to put into the PE header
int identsplit = pe_offset + sizeof(osection) + sizeof(oh);
if ((identsplit & 0x1ff) == 0)
identsplit = 0;
else if (((identsplit + identsize) ^ identsplit) < 0x200)
identsplit = identsize;
else
identsplit = ALIGN_GAP(identsplit, 0x200);
ic = identsize - identsplit;
const unsigned c_len = ((ph.c_len + ic) & 15) == 0 ? ph.c_len : ph.c_len + 16 - ((ph.c_len + ic) & 15);
obuf.clear(ph.c_len, c_len - ph.c_len);
const unsigned s1size = ALIGN_UP(ic + c_len + codesize,4u) + sotls;
const unsigned s1addr = (newvsize - (ic + c_len) + oam1) &~ oam1;
const unsigned ncsection = (s1addr + s1size + oam1) &~ oam1;
const unsigned upxsection = s1addr + ic + c_len;
Reloc rel(1024); // new relocations are put here
static const char* symbols_to_relocate[] = {
"ONAM", "BIMP", "BREL", "FIBE", "FIBS", "ENTR", "DST0", "SRC0"
};
for (unsigned s2r = 0; s2r < TABLESIZE(symbols_to_relocate); s2r++)
{
unsigned off = linker->getSymbolOffset(symbols_to_relocate[s2r]);
if (off != 0xdeaddead)
rel.add(off + upxsection, 3);
}
// new PE header
memcpy(&oh,&ih,sizeof(oh));
oh.filealign = 0x200; // identsplit depends on this
memset(osection,0,sizeof(osection));
oh.entry = upxsection;
oh.objects = 4;
oh.chksum = 0;
// fill the data directory
ODADDR(PEDIR_DEBUG) = 0;
ODSIZE(PEDIR_DEBUG) = 0;
ODADDR(PEDIR_IAT) = 0;
ODSIZE(PEDIR_IAT) = 0;
ODADDR(PEDIR_BOUNDIM) = 0;
ODSIZE(PEDIR_BOUNDIM) = 0;
// tls is put into section 1
ic = s1addr + s1size - sotls;
super::processTls(&rel,&tlsiv,ic);
ODADDR(PEDIR_TLS) = sotls ? ic : 0;
ODSIZE(PEDIR_TLS) = sotls ? 0x18 : 0;
ic += sotls;
// these are put into section 2
ic = ncsection;
// wince wants relocation data at the beginning of a section
processRelocs(&rel);
ODADDR(PEDIR_RELOC) = soxrelocs ? ic : 0;
ODSIZE(PEDIR_RELOC) = soxrelocs;
ic += soxrelocs;
processImports(ic, linker->getSymbolOffset("IATT") + upxsection);
ODADDR(PEDIR_IMPORT) = ic;
ODSIZE(PEDIR_IMPORT) = soimpdlls;
ic += soimpdlls;
processExports(&xport,ic);
ODADDR(PEDIR_EXPORT) = soexport ? ic : 0;
ODSIZE(PEDIR_EXPORT) = soexport;
if (!isdll && opt->win32_pe.compress_exports)
{
ODADDR(PEDIR_EXPORT) = IDADDR(PEDIR_EXPORT);
ODSIZE(PEDIR_EXPORT) = IDSIZE(PEDIR_EXPORT);
}
ic += soexport;
ic = (ic + oam1) &~ oam1;
const unsigned res_start = ic;
if (soresources)
processResources(&res,ic);
ODADDR(PEDIR_RESOURCE) = soresources ? ic : 0;
ODSIZE(PEDIR_RESOURCE) = soresources;
ic += soresources;
const unsigned onam = ncsection + soxrelocs + ih.imagebase;
linker->defineSymbol("start_of_dll_names", onam);
linker->defineSymbol("start_of_imports", ih.imagebase + rvamin + cimports);
linker->defineSymbol("start_of_relocs", crelocs + rvamin + ih.imagebase);
linker->defineSymbol("filter_buffer_end", ih.imagebase + ih.codebase + ih.codesize);
linker->defineSymbol("filter_buffer_start", ih.imagebase + ih.codebase);
linker->defineSymbol("original_entry", ih.entry + ih.imagebase);
linker->defineSymbol("uncompressed_length", ph.u_len);
linker->defineSymbol("start_of_uncompressed", ih.imagebase + rvamin);
linker->defineSymbol("compressed_length", ph.c_len);
linker->defineSymbol("start_of_compressed", ih.imagebase + s1addr + identsize - identsplit);
defineDecompressorSymbols();
relocateLoader();
MemBuffer loader(lsize);
memcpy(loader, getLoader(), lsize);
patchPackHeader(loader, lsize);
// this is computed here, because soxrelocs changes some lines above
const unsigned ncsize = soxrelocs + soimpdlls + soexport;
const unsigned fam1 = oh.filealign - 1;
// fill the sections
strcpy(osection[0].name,"UPX0");
strcpy(osection[1].name,"UPX1");
strcpy(osection[2].name, "UPX2");
strcpy(osection[3].name, ".rsrc");
osection[0].vaddr = rvamin;
osection[1].vaddr = s1addr;
osection[2].vaddr = ncsection;
osection[3].vaddr = res_start;
osection[0].size = 0;
osection[1].size = (s1size + fam1) &~ fam1;
osection[2].size = (ncsize + fam1) &~ fam1;
osection[3].size = (soresources + fam1) &~ fam1;
osection[0].vsize = osection[1].vaddr - osection[0].vaddr;
//osection[1].vsize = (osection[1].size + oam1) &~ oam1;
//osection[2].vsize = (osection[2].size + oam1) &~ oam1;
osection[1].vsize = osection[1].size;
osection[2].vsize = osection[2].size;
osection[3].vsize = osection[3].size;
osection[0].rawdataptr = 0;
osection[1].rawdataptr = (pe_offset + sizeof(oh) + sizeof(osection) + fam1) &~ fam1;
osection[2].rawdataptr = osection[1].rawdataptr + osection[1].size;
osection[3].rawdataptr = osection[2].rawdataptr + osection[2].size;
osection[0].flags = (unsigned) (PEFL_BSS|PEFL_EXEC|PEFL_WRITE|PEFL_READ);
osection[1].flags = (unsigned) (PEFL_DATA|PEFL_EXEC|PEFL_WRITE|PEFL_READ);
osection[2].flags = (unsigned) (PEFL_DATA|PEFL_READ);
osection[3].flags = (unsigned) (PEFL_DATA|PEFL_READ);
oh.imagesize = (osection[3].vaddr + osection[3].vsize + oam1) &~ oam1;
oh.bsssize = osection[0].vsize;
oh.datasize = osection[2].vsize + osection[3].vsize;
oh.database = osection[2].vaddr;
oh.codesize = osection[1].vsize;
oh.codebase = osection[1].vaddr;
oh.headersize = osection[1].rawdataptr;
if (rvamin < osection[0].rawdataptr)
throwCantPack("object alignment too small");
if (opt->win32_pe.strip_relocs && !isdll)
oh.flags |= RELOCS_STRIPPED;
//for (ic = 0; ic < oh.filealign; ic += 4)
// set_le32(ibuf + ic,get_le32("UPX "));
ibuf.clear(0, oh.filealign);
info("Image size change: %u -> %u KiB",
ih.imagesize / 1024, oh.imagesize / 1024);
infoHeader("[Writing compressed file]");
if (soresources == 0)
{
oh.objects = 3;
memset(&osection[3], 0, sizeof(osection[3]));
}
// write loader + compressed file
fo->write(&oh,sizeof(oh));
fo->write(osection,sizeof(osection));
// some alignment
if (identsplit == identsize)
{
unsigned n = osection[1].rawdataptr - fo->getBytesWritten() - identsize;
assert(n <= oh.filealign);
fo->write(ibuf, n);
}
fo->write(loader + codesize,identsize);
infoWriting("loader", fo->getBytesWritten());
fo->write(obuf,c_len);
infoWriting("compressed data", c_len);
fo->write(loader,codesize);
if (opt->debug.dump_stub_loader)
OutputFile::dump(opt->debug.dump_stub_loader, loader, codesize);
if ((ic = fo->getBytesWritten() & 3) != 0)
fo->write(ibuf,4 - ic);
fo->write(otls,sotls);
if ((ic = fo->getBytesWritten() & fam1) != 0)
fo->write(ibuf,oh.filealign - ic);
fo->write(oxrelocs,soxrelocs);
fo->write(oimpdlls,soimpdlls);
fo->write(oexport,soexport);
if ((ic = fo->getBytesWritten() & fam1) != 0)
fo->write(ibuf,oh.filealign - ic);
fo->write(oresources,soresources);
if ((ic = fo->getBytesWritten() & fam1) != 0)
fo->write(ibuf,oh.filealign - ic);
#if 0
printf("%-13s: program hdr : %8ld bytes\n", getName(), (long) sizeof(oh));
printf("%-13s: sections : %8ld bytes\n", getName(), (long) sizeof(osection));
printf("%-13s: ident : %8ld bytes\n", getName(), (long) identsize);
printf("%-13s: compressed : %8ld bytes\n", getName(), (long) c_len);
printf("%-13s: decompressor : %8ld bytes\n", getName(), (long) codesize);
printf("%-13s: tls : %8ld bytes\n", getName(), (long) sotls);
printf("%-13s: resources : %8ld bytes\n", getName(), (long) soresources);
printf("%-13s: imports : %8ld bytes\n", getName(), (long) soimpdlls);
printf("%-13s: exports : %8ld bytes\n", getName(), (long) soexport);
printf("%-13s: relocs : %8ld bytes\n", getName(), (long) soxrelocs);
#endif
// verify
verifyOverlappingDecompression();
// copy the overlay
copyOverlay(fo, overlay, &obuf);
// finally check the compression ratio
if (!checkFinalCompressionRatio(fo))
throwNotCompressible();
}
