mirror of
https://github.com/Atmosphere-NX/Atmosphere.git
synced 2024-11-29 23:32:23 +00:00
1082 lines
38 KiB
Python
1082 lines
38 KiB
Python
# Copyright 2017 Reswitched Team
|
|
#
|
|
# Permission to use, copy, modify, and/or distribute this software for any purpose with or
|
|
# without fee is hereby granted, provided that the above copyright notice and this permission
|
|
# notice appear in all copies.
|
|
#
|
|
# THE SOFTWARE IS PROVIDED "AS IS" AND THE AUTHOR DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS
|
|
# SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS. IN NO EVENT SHALL
|
|
# THE AUTHOR BE LIABLE FOR ANY SPECIAL, DIRECT, INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY
|
|
# DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF
|
|
# CONTRACT, NEGLIGENCE OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE
|
|
# OR PERFORMANCE OF THIS SOFTWARE.
|
|
|
|
# nxo64.py: IDA loader and library for reading nso/nro files
|
|
|
|
import gzip, math, os, re, struct, sys
|
|
|
|
from io import BytesIO
|
|
from cStringIO import StringIO
|
|
|
|
import lz4.block
|
|
|
|
uncompress = lz4.block.decompress
|
|
|
|
|
|
def get_file_size(f):
|
|
filesize = 0
|
|
try:
|
|
filesize = f.size()
|
|
except:
|
|
ptell = f.tell()
|
|
f.seek(0, 2)
|
|
filesize = f.tell()
|
|
f.seek(ptell)
|
|
return filesize
|
|
|
|
class BinFile(object):
|
|
def __init__(self, li):
|
|
self._f = li
|
|
|
|
def read(self, arg):
|
|
if isinstance(arg, str):
|
|
fmt = '<' + arg
|
|
size = struct.calcsize(fmt)
|
|
raw = self._f.read(size)
|
|
out = struct.unpack(fmt, raw)
|
|
if len(out) == 1:
|
|
return out[0]
|
|
return out
|
|
elif arg is None:
|
|
return self.read_to_end()
|
|
else:
|
|
out = self._f.read(arg)
|
|
if isinstance(arg, (int,long)) and len(out) != arg:
|
|
pass #print 'warning: read of %d bytes got %d bytes' % (arg, len(out))
|
|
return out
|
|
|
|
def read_to_end(self):
|
|
return self.read(self.size()-self.tell())
|
|
|
|
def size(self):
|
|
return get_file_size(self._f)
|
|
|
|
def read_from(self, arg, offset):
|
|
old = self.tell()
|
|
try:
|
|
self.seek(offset)
|
|
out = self.read(arg)
|
|
finally:
|
|
self.seek(old)
|
|
return out
|
|
|
|
def seek(self, off):
|
|
self._f.seek(off)
|
|
|
|
def skip(self, dist):
|
|
self.seek(self.tell()+dist)
|
|
|
|
def close(self):
|
|
self._f.close()
|
|
|
|
def tell(self):
|
|
return self._f.tell()
|
|
|
|
|
|
(DT_NULL, DT_NEEDED, DT_PLTRELSZ, DT_PLTGOT, DT_HASH, DT_STRTAB, DT_SYMTAB, DT_RELA, DT_RELASZ,
|
|
DT_RELAENT, DT_STRSZ, DT_SYMENT, DT_INIT, DT_FINI, DT_SONAME, DT_RPATH, DT_SYMBOLIC, DT_REL,
|
|
DT_RELSZ, DT_RELENT, DT_PLTREL, DT_DEBUG, DT_TEXTREL, DT_JMPREL, DT_BIND_NOW, DT_INIT_ARRAY,
|
|
DT_FINI_ARRAY, DT_INIT_ARRAYSZ, DT_FINI_ARRAYSZ, DT_RUNPATH, DT_FLAGS) = xrange(31)
|
|
|
|
DT_RELRSZ, DT_RELR, DT_RELRENT = 0x23, 0x24, 0x25
|
|
|
|
DT_GNU_HASH = 0x6ffffef5
|
|
DT_VERSYM = 0x6ffffff0
|
|
DT_RELACOUNT = 0x6ffffff9
|
|
DT_RELCOUNT = 0x6ffffffa
|
|
DT_FLAGS_1 = 0x6ffffffb
|
|
DT_VERDEF = 0x6ffffffc
|
|
DT_VERDEFNUM = 0x6ffffffd
|
|
|
|
STT_NOTYPE = 0
|
|
STT_OBJECT = 1
|
|
STT_FUNC = 2
|
|
STT_SECTION = 3
|
|
|
|
STB_LOCAL = 0
|
|
STB_GLOBAL = 1
|
|
STB_WEAK = 2
|
|
|
|
R_ARM_ABS32 = 2
|
|
R_ARM_TLS_DESC = 13
|
|
R_ARM_GLOB_DAT = 21
|
|
R_ARM_JUMP_SLOT = 22
|
|
R_ARM_RELATIVE = 23
|
|
|
|
R_FAKE_RELR = -1
|
|
|
|
R_AARCH64_ABS64 = 257
|
|
R_AARCH64_GLOB_DAT = 1025
|
|
R_AARCH64_JUMP_SLOT = 1026
|
|
R_AARCH64_RELATIVE = 1027
|
|
R_AARCH64_TLSDESC = 1031
|
|
|
|
MULTIPLE_DTS = set([DT_NEEDED])
|
|
|
|
|
|
class Range(object):
|
|
def __init__(self, start, size):
|
|
self.start = start
|
|
self.size = size
|
|
self.end = start+size
|
|
self._inclend = start+size-1
|
|
|
|
def overlaps(self, other):
|
|
return self.start <= other._inclend and other.start <= self._inclend
|
|
|
|
def includes(self, other):
|
|
return other.start >= self.start and other._inclend <= self._inclend
|
|
|
|
def __repr__(self):
|
|
return 'Range(0x%X -> 0x%X)' % (self.start, self.end)
|
|
|
|
|
|
class Segment(object):
|
|
def __init__(self, r, name, kind):
|
|
self.range = r
|
|
self.name = name
|
|
self.kind = kind
|
|
self.sections = []
|
|
|
|
def add_section(self, s):
|
|
for i in self.sections:
|
|
assert not i.range.overlaps(s.range), '%r overlaps %r' % (s, i)
|
|
self.sections.append(s)
|
|
|
|
|
|
class Section(object):
|
|
def __init__(self, r, name):
|
|
self.range = r
|
|
self.name = name
|
|
|
|
def __repr__(self):
|
|
return 'Section(%r, %r)' % (self.range, self.name)
|
|
|
|
|
|
def suffixed_name(name, suffix):
|
|
if suffix == 0:
|
|
return name
|
|
return '%s.%d' % (name, suffix)
|
|
|
|
|
|
class SegmentBuilder(object):
|
|
def __init__(self):
|
|
self.segments = []
|
|
self._sections = []
|
|
|
|
def add_segment(self, start, size, name, kind):
|
|
r = Range(start, size)
|
|
for i in self.segments:
|
|
assert not r.overlaps(i.range)
|
|
self.segments.append(Segment(r, name, kind))
|
|
|
|
def add_section(self, name, start, end=None, size=None):
|
|
assert end is None or size is None
|
|
if size is None:
|
|
size = end-start
|
|
if size > 0:
|
|
#assert size > 0
|
|
r = Range(start, size)
|
|
self._sections.append((r, name))
|
|
|
|
def _add_sections_to_segments(self):
|
|
for r, name in self._sections:
|
|
for i in self.segments:
|
|
if i.range.includes(r):
|
|
i.add_section(Section(r, name))
|
|
break
|
|
else:
|
|
assert False, 'no containing segment for %r' % (name,)
|
|
|
|
def flatten(self):
|
|
self._add_sections_to_segments()
|
|
self.segments.sort(key=lambda s: s.range.start)
|
|
parts = []
|
|
for segment in self.segments:
|
|
suffix = 0
|
|
segment.sections.sort(key=lambda s: s.range.start)
|
|
pos = segment.range.start
|
|
for section in segment.sections:
|
|
if pos < section.range.start:
|
|
parts.append((pos, section.range.start, suffixed_name(segment.name, suffix), segment.kind))
|
|
suffix += 1
|
|
pos = section.range.start
|
|
parts.append((section.range.start, section.range.end, section.name, segment.kind))
|
|
pos = section.range.end
|
|
if pos < segment.range.end:
|
|
parts.append((pos, segment.range.end, suffixed_name(segment.name, suffix), segment.kind))
|
|
suffix += 1
|
|
pos = segment.range.end
|
|
return parts
|
|
|
|
|
|
class ElfSym(object):
|
|
def __init__(self, name, info, other, shndx, value, size):
|
|
self.name = name
|
|
self.shndx = shndx
|
|
self.value = value
|
|
self.size = size
|
|
|
|
self.vis = other & 3
|
|
self.type = info & 0xF
|
|
self.bind = info >> 4
|
|
|
|
def __repr__(self):
|
|
return 'Sym(name=%r, shndx=0x%X, value=0x%X, size=0x%X, vis=%r, type=%r, bind=%r)' % (
|
|
self.name, self.shndx, self.value, self.size, self.vis, self.type, self.bind)
|
|
|
|
|
|
class NxoFileBase(object):
|
|
def __init__(self, f, segment_data=None):
|
|
self.binfile = f
|
|
|
|
# read MOD
|
|
self.modoff = f.read_from('I', 4)
|
|
|
|
f.seek(self.modoff)
|
|
if f.read('4s') != 'MOD0':
|
|
raise NxoException('invalid MOD0 magic')
|
|
|
|
self.dynamicoff = self.modoff + f.read('i')
|
|
self.bssoff = self.modoff + f.read('i')
|
|
self.bssend = self.modoff + f.read('i')
|
|
self.unwindoff = self.modoff + f.read('i')
|
|
self.unwindend = self.modoff + f.read('i')
|
|
self.moduleoff = self.modoff + f.read('i')
|
|
|
|
|
|
builder = SegmentBuilder()
|
|
|
|
# read dynamic
|
|
self.armv7 = (f.read_from('Q', self.dynamicoff) > 0xFFFFFFFF or f.read_from('Q', self.dynamicoff+0x10) > 0xFFFFFFFF)
|
|
self.offsize = 4 if self.armv7 else 8
|
|
|
|
f.seek(self.dynamicoff)
|
|
self.dynamic = dynamic = {}
|
|
for i in MULTIPLE_DTS:
|
|
dynamic[i] = []
|
|
for i in xrange((f.size() - self.dynamicoff) / 0x10):
|
|
tag, val = f.read('II' if self.armv7 else 'QQ')
|
|
if tag == DT_NULL:
|
|
break
|
|
if tag in MULTIPLE_DTS:
|
|
dynamic[tag].append(val)
|
|
else:
|
|
dynamic[tag] = val
|
|
self.dynamicsize = f.tell() - self.dynamicoff
|
|
builder.add_section('.dynamic', self.dynamicoff, end=self.dynamicoff + self.dynamicsize)
|
|
builder.add_section('.eh_frame_hdr', self.unwindoff, end=self.unwindend)
|
|
|
|
# read .dynstr
|
|
if DT_STRTAB in dynamic and DT_STRSZ in dynamic:
|
|
f.seek(dynamic[DT_STRTAB])
|
|
self.dynstr = f.read(dynamic[DT_STRSZ])
|
|
else:
|
|
self.dynstr = '\0'
|
|
print 'warning: no dynstr'
|
|
|
|
for startkey, szkey, name in [
|
|
(DT_STRTAB, DT_STRSZ, '.dynstr'),
|
|
(DT_INIT_ARRAY, DT_INIT_ARRAYSZ, '.init_array'),
|
|
(DT_FINI_ARRAY, DT_FINI_ARRAYSZ, '.fini_array'),
|
|
(DT_RELA, DT_RELASZ, '.rela.dyn'),
|
|
(DT_REL, DT_RELSZ, '.rel.dyn'),
|
|
(DT_RELR, DT_RELRSZ, '.relr.dyn'),
|
|
(DT_JMPREL, DT_PLTRELSZ, ('.rel.plt' if self.armv7 else '.rela.plt')),
|
|
]:
|
|
if startkey in dynamic and szkey in dynamic:
|
|
builder.add_section(name, dynamic[startkey], size=dynamic[szkey])
|
|
|
|
# TODO
|
|
#build_id = content.find('\x04\x00\x00\x00\x14\x00\x00\x00\x03\x00\x00\x00GNU\x00')
|
|
#if build_id >= 0:
|
|
# builder.add_section('.note.gnu.build-id', build_id, size=0x24)
|
|
#else:
|
|
# build_id = content.index('\x04\x00\x00\x00\x10\x00\x00\x00\x03\x00\x00\x00GNU\x00')
|
|
# if build_id >= 0:
|
|
# builder.add_section('.note.gnu.build-id', build_id, size=0x20)
|
|
|
|
if DT_HASH in dynamic:
|
|
hash_start = dynamic[DT_HASH]
|
|
f.seek(hash_start)
|
|
nbucket, nchain = f.read('II')
|
|
f.skip(nbucket * 4)
|
|
f.skip(nchain * 4)
|
|
hash_end = f.tell()
|
|
builder.add_section('.hash', hash_start, end=hash_end)
|
|
|
|
if DT_GNU_HASH in dynamic:
|
|
gnuhash_start = dynamic[DT_GNU_HASH]
|
|
f.seek(gnuhash_start)
|
|
nbuckets, symoffset, bloom_size, bloom_shift = f.read('IIII')
|
|
f.skip(bloom_size * self.offsize)
|
|
buckets = [f.read('I') for i in range(nbuckets)]
|
|
|
|
max_symix = max(buckets) if buckets else 0
|
|
if max_symix >= symoffset:
|
|
f.skip((max_symix - symoffset) * 4)
|
|
while (f.read('I') & 1) == 0:
|
|
pass
|
|
gnuhash_end = f.tell()
|
|
builder.add_section('.gnu.hash', gnuhash_start, end=gnuhash_end)
|
|
|
|
self.needed = [self.get_dynstr(i) for i in self.dynamic[DT_NEEDED]]
|
|
|
|
# load .dynsym
|
|
self.symbols = symbols = []
|
|
f.seek(dynamic[DT_SYMTAB])
|
|
while True:
|
|
if dynamic[DT_SYMTAB] < dynamic[DT_STRTAB] and f.tell() >= dynamic[DT_STRTAB]:
|
|
break
|
|
if self.armv7:
|
|
st_name, st_value, st_size, st_info, st_other, st_shndx = f.read('IIIBBH')
|
|
else:
|
|
st_name, st_info, st_other, st_shndx, st_value, st_size = f.read('IBBHQQ')
|
|
if st_name > len(self.dynstr):
|
|
break
|
|
symbols.append(ElfSym(self.get_dynstr(st_name), st_info, st_other, st_shndx, st_value, st_size))
|
|
builder.add_section('.dynsym', dynamic[DT_SYMTAB], end=f.tell())
|
|
|
|
self.plt_entries = []
|
|
self.relocations = []
|
|
locations = set()
|
|
if DT_REL in dynamic:
|
|
locations |= self.process_relocations(f, symbols, dynamic[DT_REL], dynamic[DT_RELSZ])
|
|
|
|
if DT_RELA in dynamic:
|
|
locations |= self.process_relocations(f, symbols, dynamic[DT_RELA], dynamic[DT_RELASZ])
|
|
|
|
if DT_RELR in dynamic:
|
|
locations |= self.process_relocations_relr(f, dynamic[DT_RELR], dynamic[DT_RELRSZ])
|
|
|
|
if segment_data is None:
|
|
# infer segment info
|
|
rloc_guess = (dynamic[DT_REL if DT_REL in dynamic else DT_RELA] & ~0xFFF)
|
|
dloc_guess = (min(i for i in locations if i != 0) & ~0xFFF)
|
|
dloc_guess2 = None
|
|
modoff = f.read_from('I', 4)
|
|
if self.modoff != 8:
|
|
search_start = (self.modoff + 0xFFF) & ~0xFFF
|
|
for i in range(search_start, f.size(), 0x1000):
|
|
count = 0
|
|
for j in range(4, 0x1000, 4):
|
|
if f.read_from('I', i - j) != 0:
|
|
break
|
|
count += 1
|
|
if count > 6:
|
|
dloc_guess2 = i
|
|
break
|
|
if dloc_guess2 is not None and dloc_guess2 < dloc_guess:
|
|
dloc_guess = dloc_guess2
|
|
|
|
if segment_data:
|
|
tloc, tsize, rloc, rsize, dloc, dsize = segment_data
|
|
assert rloc_guess == rloc
|
|
assert dloc_guess == dloc
|
|
|
|
self.textoff = 0
|
|
self.textsize = rloc_guess
|
|
self.rodataoff = rloc_guess
|
|
self.rodatasize = dloc_guess - rloc_guess
|
|
self.dataoff = dloc_guess
|
|
else:
|
|
tloc, tsize, rloc, rsize, dloc, dsize = segment_data
|
|
self.textoff = tloc
|
|
self.textsize = tsize
|
|
self.rodataoff = rloc
|
|
self.rodatasize = rsize
|
|
self.dataoff = dloc
|
|
|
|
self.datasize = self.bssoff - self.dataoff
|
|
self.bsssize = self.bssend - self.bssoff
|
|
|
|
plt_got_end = None
|
|
if DT_JMPREL in dynamic:
|
|
pltlocations = self.process_relocations(f, symbols, dynamic[DT_JMPREL], dynamic[DT_PLTRELSZ])
|
|
locations |= pltlocations
|
|
|
|
plt_got_start = min(pltlocations)
|
|
plt_got_end = max(pltlocations) + self.offsize
|
|
if DT_PLTGOT in dynamic:
|
|
builder.add_section('.got.plt', dynamic[DT_PLTGOT], end=plt_got_end)
|
|
|
|
if not self.armv7:
|
|
f.seek(0)
|
|
text = f.read(self.textsize)
|
|
last = 12
|
|
while True:
|
|
pos = text.find(struct.pack('<I', 0xD61F0220), last)
|
|
if pos == -1: break
|
|
last = pos+1
|
|
if (pos % 4) != 0: continue
|
|
off = pos - 12
|
|
a, b, c, d = struct.unpack_from('<IIII', text, off)
|
|
if d == 0xD61F0220 and (a & 0x9f00001f) == 0x90000010 and (b & 0xffe003ff) == 0xf9400211:
|
|
base = off & ~0xFFF
|
|
immhi = (a >> 5) & 0x7ffff
|
|
immlo = (a >> 29) & 3
|
|
paddr = base + ((immlo << 12) | (immhi << 14))
|
|
poff = ((b >> 10) & 0xfff) << 3
|
|
target = paddr + poff
|
|
if plt_got_start <= target < plt_got_end:
|
|
self.plt_entries.append((off, target))
|
|
if len(self.plt_entries) > 0:
|
|
builder.add_section('.plt', min(self.plt_entries)[0], end=max(self.plt_entries)[0] + 0x10)
|
|
|
|
# try to find the ".got" which should follow the ".got.plt"
|
|
good = False
|
|
got_start = (plt_got_end if plt_got_end is not None else self.dynamicoff + self.dynamicsize)
|
|
got_end = self.offsize + got_start
|
|
while (got_end in locations or (plt_got_end is None and got_end < dynamic[DT_INIT_ARRAY])) and (DT_INIT_ARRAY not in dynamic or got_end < dynamic[DT_INIT_ARRAY] or dynamic[DT_INIT_ARRAY] < got_start):
|
|
good = True
|
|
got_end += self.offsize
|
|
#print 'got_start got_end %X %X %s %X' % (got_start, got_end, str(got_end in locations), dynamic[DT_INIT_ARRAY])
|
|
|
|
if good:
|
|
self.got_start = got_start
|
|
self.got_end = got_end
|
|
builder.add_section('.got', self.got_start, end=self.got_end)
|
|
|
|
self.eh_table = []
|
|
if not self.armv7:
|
|
f.seek(self.unwindoff)
|
|
version, eh_frame_ptr_enc, fde_count_enc, table_enc = f.read('BBBB')
|
|
if not any(i == 0xff for i in (eh_frame_ptr_enc, fde_count_enc, table_enc)): # DW_EH_PE_omit
|
|
#assert eh_frame_ptr_enc == 0x1B # DW_EH_PE_pcrel | DW_EH_PE_sdata4
|
|
#assert fde_count_enc == 0x03 # DW_EH_PE_absptr | DW_EH_PE_udata4
|
|
#assert table_enc == 0x3B # DW_EH_PE_datarel | DW_EH_PE_sdata4
|
|
if eh_frame_ptr_enc == 0x1B and fde_count_enc == 0x03 and table_enc == 0x3B:
|
|
base_offset = f.tell()
|
|
eh_frame = base_offset + f.read('i')
|
|
|
|
fde_count = f.read('I')
|
|
#assert 8 * fde_count == self.unwindend - f.tell()
|
|
if 8 * fde_count == self.unwindend - f.tell():
|
|
for i in range(fde_count):
|
|
pc = self.unwindoff + f.read('i')
|
|
entry = self.unwindoff + f.read('i')
|
|
self.eh_table.append((pc, entry))
|
|
|
|
# TODO: we miss the last one, but better than nothing
|
|
last_entry = sorted(self.eh_table, key=lambda x: x[1])[-1][1]
|
|
builder.add_section('.eh_frame', eh_frame, end=last_entry)
|
|
|
|
|
|
for off,sz,name,kind in [
|
|
(self.textoff, self.textsize, '.text', 'CODE'),
|
|
(self.rodataoff, self.rodatasize, '.rodata', 'CONST'),
|
|
(self.dataoff, self.datasize, '.data', 'DATA'),
|
|
(self.bssoff, self.bsssize, '.bss', 'BSS'),
|
|
]:
|
|
builder.add_segment(off, sz, name, kind)
|
|
|
|
self.sections = []
|
|
for start, end, name, kind in builder.flatten():
|
|
self.sections.append((start, end, name, kind))
|
|
|
|
self._addr_to_name = None
|
|
self._plt_lookup = None
|
|
|
|
@property
|
|
def addr_to_name(self):
|
|
if self._addr_to_name is None:
|
|
d = {}
|
|
for sym in self.symbols:
|
|
if sym.shndx:
|
|
d[sym.value] = sym.name
|
|
self._addr_to_name = d
|
|
return self._addr_to_name
|
|
|
|
@property
|
|
def plt_lookup(self):
|
|
if self._plt_lookup is None:
|
|
# generate lookups for .plt call redirection
|
|
got_value_lookup = {}
|
|
for offset, r_type, sym, addend in self.relocations:
|
|
if r_type in (R_AARCH64_GLOB_DAT, R_AARCH64_JUMP_SLOT, R_AARCH64_ABS64) and addend == 0 and sym.shndx:
|
|
got_value_lookup[offset] = sym.value
|
|
|
|
self._plt_lookup = {}
|
|
for func, target in self.plt_entries:
|
|
if target in got_value_lookup:
|
|
self._plt_lookup[func] = got_value_lookup[target]
|
|
|
|
return self._plt_lookup
|
|
|
|
def process_relocations(self, f, symbols, offset, size):
|
|
locations = set()
|
|
f.seek(offset)
|
|
relocsize = 8 if self.armv7 else 0x18
|
|
for i in xrange(size / relocsize):
|
|
# NOTE: currently assumes all armv7 relocs have no addends,
|
|
# and all 64-bit ones do.
|
|
if self.armv7:
|
|
offset, info = f.read('II')
|
|
addend = None
|
|
r_type = info & 0xff
|
|
r_sym = info >> 8
|
|
else:
|
|
offset, info, addend = f.read('QQq')
|
|
r_type = info & 0xffffffff
|
|
r_sym = info >> 32
|
|
|
|
sym = symbols[r_sym] if r_sym != 0 else None
|
|
|
|
if r_type != R_AARCH64_TLSDESC and r_type != R_ARM_TLS_DESC:
|
|
locations.add(offset)
|
|
self.relocations.append((offset, r_type, sym, addend))
|
|
return locations
|
|
|
|
def process_relocations_relr(self, f, offset, size):
|
|
locations = set()
|
|
f.seek(offset)
|
|
relocsize = 8
|
|
for i in xrange(size / relocsize):
|
|
entry = f.read('Q')
|
|
if entry & 1:
|
|
entry >>= 1
|
|
i = 0
|
|
while i < (relocsize * 8) - 1:
|
|
if entry & (1 << i):
|
|
locations.add(where + i * relocsize)
|
|
self.relocations.append((where + i * relocsize, R_FAKE_RELR, None, 0))
|
|
i += 1
|
|
where += relocsize * ((relocsize * 8) - 1)
|
|
else:
|
|
# Where
|
|
where = entry
|
|
locations.add(where)
|
|
self.relocations.append((where, R_FAKE_RELR, None, 0))
|
|
where += relocsize
|
|
return locations
|
|
|
|
def get_dynstr(self, o):
|
|
return self.dynstr[o:self.dynstr.index('\0', o)]
|
|
|
|
def get_path_or_name(self):
|
|
path = None
|
|
for off, end, name, class_ in self.sections:
|
|
if name == '.rodata' and end-off < 0x1000 and end-off > 8:
|
|
id_ = self.binfile.read_from(end-off, off)
|
|
length = struct.unpack_from('<I', id_, 4)[0]
|
|
if length + 8 <= len(id_):
|
|
id_ = id_[8:length + 8]
|
|
return id_
|
|
|
|
self.binfile.seek(self.rodataoff)
|
|
as_string = self.binfile.read(self.rodatasize)
|
|
if path is None:
|
|
strs = re.findall(r'[a-z]:[\\/][ -~]{5,}\.n[rs]s', as_string, flags=re.IGNORECASE)
|
|
if strs:
|
|
return strs[-1]
|
|
|
|
return None
|
|
|
|
def get_name(self):
|
|
name = self.get_path_or_name()
|
|
if name is not None:
|
|
name = name.split('/')[-1].split('\\')[-1]
|
|
if name.lower().endswith(('.nss', '.nrs')):
|
|
name = name[:-4]
|
|
return name
|
|
|
|
class NsoFile(NxoFileBase):
|
|
def __init__(self, fileobj):
|
|
f = BinFile(fileobj)
|
|
|
|
if f.read_from('4s', 0) != 'NSO0':
|
|
raise NxoException('Invalid NSO magic')
|
|
|
|
flags = f.read_from('I', 0xC)
|
|
|
|
toff, tloc, tsize = f.read_from('III', 0x10)
|
|
roff, rloc, rsize = f.read_from('III', 0x20)
|
|
doff, dloc, dsize = f.read_from('III', 0x30)
|
|
|
|
tfilesize, rfilesize, dfilesize = f.read_from('III', 0x60)
|
|
bsssize = f.read_from('I', 0x3C)
|
|
|
|
text = uncompress(f.read_from(tfilesize, toff), uncompressed_size=tsize) if flags & 1 else f.read_from(tfilesize, toff)
|
|
ro = uncompress(f.read_from(rfilesize, roff), uncompressed_size=rsize) if flags & 2 else f.read_from(rfilesize, roff)
|
|
data = uncompress(f.read_from(dfilesize, doff), uncompressed_size=dsize) if flags & 4 else f.read_from(dfilesize, doff)
|
|
|
|
full = text
|
|
if rloc >= len(full):
|
|
full += '\0' * (rloc - len(full))
|
|
else:
|
|
print 'truncating?'
|
|
full = full[:rloc]
|
|
full += ro
|
|
if dloc >= len(full):
|
|
full += '\0' * (dloc - len(full))
|
|
else:
|
|
print 'truncating?'
|
|
full = full[:dloc]
|
|
full += data
|
|
self.full = full
|
|
|
|
super(NsoFile, self).__init__(BinFile(StringIO(full)), (tloc, tsize, rloc, rsize, dloc, dsize))
|
|
|
|
|
|
class NroFile(NxoFileBase):
|
|
def __init__(self, fileobj):
|
|
f = BinFile(fileobj)
|
|
|
|
if f.read_from('4s', 0x10) != 'NRO0':
|
|
raise NxoException('Invalid NRO magic')
|
|
|
|
f.seek(0x20)
|
|
|
|
tloc, tsize = f.read('II')
|
|
rloc, rsize = f.read('II')
|
|
dloc, dsize = f.read('II')
|
|
|
|
# copy f, since all other formats allow the original file to be closed
|
|
f.seek(0)
|
|
full = f.read(f.size())
|
|
|
|
filecopy = BinFile(StringIO(full))
|
|
super(NroFile, self).__init__(filecopy, (tloc, tsize, rloc, rsize, dloc, dsize))
|
|
|
|
def kip1_blz_decompress(compressed):
|
|
compressed_size, init_index, uncompressed_addl_size = struct.unpack('<III', compressed[-0xC:])
|
|
decompressed = compressed[:] + '\x00' * uncompressed_addl_size
|
|
decompressed_size = len(decompressed)
|
|
if not (compressed_size + uncompressed_addl_size):
|
|
return ''
|
|
decompressed = map(ord, decompressed)
|
|
cmp_start = len(compressed) - compressed_size
|
|
cmp_ofs = compressed_size - init_index
|
|
out_ofs = compressed_size + uncompressed_addl_size
|
|
while out_ofs > 0:
|
|
cmp_ofs -= 1
|
|
control = decompressed[cmp_start + cmp_ofs]
|
|
for i in xrange(8):
|
|
if control & 0x80:
|
|
if cmp_ofs < 2 - cmp_start:
|
|
raise ValueError('Compression out of bounds!')
|
|
cmp_ofs -= 2
|
|
segmentoffset = decompressed[cmp_start + cmp_ofs] | (decompressed[cmp_start + cmp_ofs + 1] << 8)
|
|
segmentsize = ((segmentoffset >> 12) & 0xF) + 3
|
|
segmentoffset &= 0x0FFF
|
|
segmentoffset += 2
|
|
if out_ofs < segmentsize - cmp_start:
|
|
raise ValueError('Compression out of bounds!')
|
|
for j in xrange(segmentsize):
|
|
if out_ofs + segmentoffset >= decompressed_size:
|
|
raise ValueError('Compression out of bounds!')
|
|
data = decompressed[cmp_start + out_ofs + segmentoffset]
|
|
out_ofs -= 1
|
|
decompressed[cmp_start + out_ofs] = data
|
|
else:
|
|
if out_ofs < 1 - cmp_start:
|
|
raise ValueError('Compression out of bounds!')
|
|
out_ofs -= 1
|
|
cmp_ofs -= 1
|
|
decompressed[cmp_start + out_ofs] = decompressed[cmp_start + cmp_ofs]
|
|
control <<= 1
|
|
control &= 0xFF
|
|
if not out_ofs:
|
|
break
|
|
return ''.join(map(chr, decompressed))
|
|
|
|
class KipFile(NxoFileBase):
|
|
def __init__(self, fileobj):
|
|
f = BinFile(fileobj)
|
|
|
|
if f.read_from('4s', 0) != 'KIP1':
|
|
raise NxoException('Invalid KIP magic')
|
|
|
|
flags = f.read_from('b', 0x1F)
|
|
|
|
tloc, tsize, tfilesize = f.read_from('III', 0x20)
|
|
rloc, rsize, rfilesize = f.read_from('III', 0x30)
|
|
dloc, dsize, dfilesize = f.read_from('III', 0x40)
|
|
|
|
toff = 0x100
|
|
roff = toff + tfilesize
|
|
doff = roff + rfilesize
|
|
|
|
bsssize = f.read_from('I', 0x18)
|
|
|
|
|
|
text = kip1_blz_decompress(str(f.read_from(tfilesize, toff))) if flags & 1 else f.read_from(tfilesize, toff)
|
|
ro = kip1_blz_decompress(str(f.read_from(rfilesize, roff))) if flags & 2 else f.read_from(rfilesize, roff)
|
|
data = kip1_blz_decompress(str(f.read_from(dfilesize, doff))) if flags & 4 else f.read_from(dfilesize, doff)
|
|
|
|
full = text
|
|
if rloc >= len(full):
|
|
full += '\0' * (rloc - len(full))
|
|
else:
|
|
print 'truncating?'
|
|
full = full[:rloc]
|
|
full += ro
|
|
if dloc >= len(full):
|
|
full += '\0' * (dloc - len(full))
|
|
else:
|
|
print 'truncating?'
|
|
full = full[:dloc]
|
|
full += data
|
|
|
|
super(KipFile, self).__init__(BinFile(StringIO(full)), (tloc, tsize, rloc, rsize, dloc, dsize))
|
|
|
|
class MemoryDumpFile(NxoFileBase):
|
|
def __init__(self, fileobj):
|
|
f = BinFile(fileobj)
|
|
|
|
if f.read_from('4s', f.read_from('I', 4)) != 'MOD0':
|
|
raise NxoException('Invalid MOD0 magic')
|
|
|
|
f.seek(0)
|
|
full = f.read(f.size())
|
|
|
|
filecopy = BinFile(StringIO(full))
|
|
super(MemoryDumpFile, self).__init__(filecopy)
|
|
|
|
class NxoException(Exception):
|
|
pass
|
|
|
|
def looks_like_memory_dump(fileobj):
|
|
fileobj.seek(0)
|
|
header = fileobj.read(8)
|
|
if len(header) < 8:
|
|
return False
|
|
modoff = struct.unpack_from('<I', header, 4)[0]
|
|
if modoff + 0x1c < get_file_size(fileobj):
|
|
fileobj.seek(modoff)
|
|
if fileobj.read(4) == 'MOD0':
|
|
return True
|
|
return False
|
|
|
|
|
|
def load_nxo(fileobj):
|
|
fileobj.seek(0)
|
|
header = fileobj.read(0x14)
|
|
|
|
if header[:4] == 'NSO0':
|
|
return NsoFile(fileobj)
|
|
elif header[:4] == 'KIP1':
|
|
return KipFile(fileobj)
|
|
elif header[0x10:0x14] == 'NRO0':
|
|
return NroFile(fileobj)
|
|
elif looks_like_memory_dump(fileobj):
|
|
return MemoryDumpFile(fileobj)
|
|
|
|
raise NxoException('not an NRO or NSO file')
|
|
|
|
|
|
try:
|
|
import idaapi
|
|
import idc
|
|
except ImportError:
|
|
pass
|
|
else:
|
|
# IDA specific code
|
|
NRO_FORMAT = 'Switch binary (NRO)'
|
|
NSO_FORMAT = 'Switch binary (NSO)'
|
|
KIP_FORMAT = 'Switch binary (KIP)'
|
|
RAW_FORMAT = 'Switch raw binary memory dump'
|
|
SDK_FORMAT = 'Switch SDK binary (NSO, with arm64 .plt call rewriting hack)'
|
|
RAW_SDK_FORMAT = 'Switch SDK (raw memory dump, with arm64 .plt call rewriting hack)'
|
|
EXEFS_FORMAT = 'Switch exefs (multiple files, for use with Mephisto)'
|
|
EXEFS_LOW_FORMAT = 'Switch exefs with 31-bit addressing (multiple files, for use with Mephisto)'
|
|
|
|
OPT_BYPASS_PLT = 'BYPASS_PLT'
|
|
OPT_EXEFS_LOAD = 'EXEFS_LOAD'
|
|
OPT_LOAD_31_BIT = 'LOAD_31_BIT'
|
|
|
|
FORMAT_OPTIONS = {
|
|
NRO_FORMAT: [],
|
|
NSO_FORMAT: [],
|
|
KIP_FORMAT: [],
|
|
RAW_FORMAT: [],
|
|
SDK_FORMAT: [OPT_BYPASS_PLT],
|
|
RAW_SDK_FORMAT: [OPT_BYPASS_PLT],
|
|
EXEFS_FORMAT: [OPT_EXEFS_LOAD],
|
|
EXEFS_LOW_FORMAT: [OPT_EXEFS_LOAD, OPT_LOAD_31_BIT],
|
|
}
|
|
|
|
PRIMARY_EXEFS_NAMES = ['main', 'rtld', 'sdk']
|
|
LOAD_EXEFS_NAMES = PRIMARY_EXEFS_NAMES + ['subsdk%d' % i for i in range(10)]
|
|
ALL_EXEFS_NAMES = LOAD_EXEFS_NAMES + ['main.npdm']
|
|
|
|
def get_load_formats(li, path):
|
|
li.seek(0)
|
|
if li.read(4) == 'NSO0':
|
|
if 'sdk' in os.path.basename(path):
|
|
yield { 'format': SDK_FORMAT, 'options': idaapi.ACCEPT_FIRST }
|
|
yield { 'format': NSO_FORMAT }
|
|
else:
|
|
yield { 'format': NSO_FORMAT, 'options': idaapi.ACCEPT_FIRST }
|
|
yield { 'format': SDK_FORMAT }
|
|
|
|
li.seek(0)
|
|
if li.read(4) == 'KIP1':
|
|
yield { 'format': KIP_FORMAT, 'options': idaapi.ACCEPT_FIRST }
|
|
|
|
li.seek(0x10)
|
|
if li.read(4) == 'NRO0':
|
|
yield { 'format': NRO_FORMAT, 'options': idaapi.ACCEPT_FIRST }
|
|
elif looks_like_memory_dump(li):
|
|
yield { 'format': RAW_FORMAT, 'options': idaapi.ACCEPT_FIRST }
|
|
yield { 'format': RAW_SDK_FORMAT }
|
|
|
|
if os.path.basename(path) in ALL_EXEFS_NAMES:
|
|
dirname = os.path.dirname(path)
|
|
if all(os.path.exists(os.path.join(dirname, i)) for i in PRIMARY_EXEFS_NAMES):
|
|
yield { 'format': EXEFS_FORMAT }
|
|
yield { 'format': EXEFS_LOW_FORMAT }
|
|
|
|
|
|
|
|
# this is mostly just to work with the IDA API
|
|
accept_formats_list = None
|
|
accept_formats_index = 0
|
|
|
|
def accept_file(li, path):
|
|
global accept_formats_list
|
|
global accept_formats_index
|
|
|
|
if accept_formats_list is None:
|
|
accept_formats_list = list(get_load_formats(li, path))
|
|
accept_formats_index = 0
|
|
|
|
if accept_formats_index >= len(accept_formats_list):
|
|
accept_formats_list = None
|
|
accept_formats_index = 0
|
|
return 0
|
|
|
|
ret = accept_formats_list[accept_formats_index]
|
|
if not isinstance(ret, dict):
|
|
ret = { 'format': ret }
|
|
if 'options' not in ret:
|
|
ret['options'] = 1
|
|
if 'processor' not in ret:
|
|
ret['processor'] = 'arm'
|
|
ret['options'] |= 1 | idaapi.ACCEPT_CONTINUE
|
|
|
|
accept_formats_index += 1
|
|
|
|
return ret
|
|
|
|
def ida_make_offset(f, ea):
|
|
if f.armv7:
|
|
idc.MakeDword(ea)
|
|
else:
|
|
idc.MakeQword(ea)
|
|
idc.OpOff(ea, 0, 0)
|
|
|
|
def find_bl_targets(text_start, text_end):
|
|
targets = set()
|
|
for pco in xrange(0, text_end - text_start, 4):
|
|
pc = text_start + pco
|
|
d = Dword(pc)
|
|
if (d & 0xfc000000) == 0x94000000:
|
|
imm = d & 0x3ffffff
|
|
if imm & 0x2000000:
|
|
imm |= ~0x1ffffff
|
|
if 0 <= imm <= 2:
|
|
continue
|
|
target = pc + imm * 4
|
|
if target >= text_start and target < text_end:
|
|
targets.add(target)
|
|
return targets
|
|
|
|
def load_file(li, neflags, fmt):
|
|
idaapi.set_processor_type('arm', SETPROC_ALL|SETPROC_FATAL)
|
|
|
|
options = FORMAT_OPTIONS[fmt]
|
|
|
|
if OPT_EXEFS_LOAD in options:
|
|
ret = load_as_exefs(li, options)
|
|
else:
|
|
ret = load_one_file(li, options, 0)
|
|
|
|
eh_parse = idaapi.find_plugin('eh_parse', True)
|
|
if eh_parse:
|
|
print 'eh_parse ->', idaapi.run_plugin(eh_parse, 0)
|
|
else:
|
|
print 'warning: eh_parse missing'
|
|
|
|
return ret
|
|
|
|
def load_as_exefs(li, options):
|
|
dirname = os.path.dirname(idc.get_input_file_path())
|
|
binaries = LOAD_EXEFS_NAMES
|
|
binaries = [os.path.join(dirname, i) for i in binaries]
|
|
binaries = [i for i in binaries if os.path.exists(i)]
|
|
for idx, fname in enumerate(binaries):
|
|
with open(fname, 'rb') as f:
|
|
if not load_one_file(f, options, idx, os.path.basename(fname)):
|
|
return False
|
|
return True
|
|
|
|
def load_one_file(li, options, idx, basename=None):
|
|
bypass_plt = OPT_BYPASS_PLT in options
|
|
|
|
f = load_nxo(li)
|
|
|
|
if idx == 0:
|
|
if f.armv7:
|
|
idc.SetShortPrm(idc.INF_LFLAGS, idc.GetShortPrm(idc.INF_LFLAGS) | idc.LFLG_PC_FLAT)
|
|
else:
|
|
idc.SetShortPrm(idc.INF_LFLAGS, idc.GetShortPrm(idc.INF_LFLAGS) | idc.LFLG_64BIT)
|
|
|
|
idc.SetCharPrm(idc.INF_DEMNAMES, idaapi.DEMNAM_GCC3)
|
|
idaapi.set_compiler_id(idaapi.COMP_GNU)
|
|
idaapi.add_til2('gnulnx_arm' if f.armv7 else 'gnulnx_arm64', 1)
|
|
# don't create tails
|
|
idc.set_inf_attr(idc.INF_AF, idc.get_inf_attr(idc.INF_AF) & ~idc.AF_FTAIL)
|
|
|
|
if OPT_LOAD_31_BIT in options:
|
|
loadbase = 0x8000000
|
|
step = 0x1000000
|
|
elif f.armv7:
|
|
loadbase = 0x60000000
|
|
step = 0x10000000
|
|
else:
|
|
loadbase = 0x7100000000
|
|
step = 0x100000000
|
|
loadbase += idx * step
|
|
|
|
f.binfile.seek(0)
|
|
as_string = f.binfile.read(f.bssoff)
|
|
idaapi.mem2base(as_string, loadbase)
|
|
|
|
seg_prefix = basename if basename is not None else ''
|
|
for start, end, name, kind in f.sections:
|
|
if name.startswith('.got'):
|
|
kind = 'CONST'
|
|
idaapi.add_segm(0, loadbase+start, loadbase+end, seg_prefix+name, kind)
|
|
segm = idaapi.get_segm_by_name(seg_prefix+name)
|
|
if kind == 'CONST':
|
|
segm.perm = idaapi.SEGPERM_READ
|
|
elif kind == 'CODE':
|
|
segm.perm = idaapi.SEGPERM_READ | idaapi.SEGPERM_EXEC
|
|
elif kind == 'DATA':
|
|
segm.perm = idaapi.SEGPERM_READ | idaapi.SEGPERM_WRITE
|
|
elif kind == 'BSS':
|
|
segm.perm = idaapi.SEGPERM_READ | idaapi.SEGPERM_WRITE
|
|
idaapi.update_segm(segm)
|
|
idaapi.set_segm_addressing(segm, 1 if f.armv7 else 2)
|
|
|
|
# do imports
|
|
# TODO: can we make imports show up in "Imports" window?
|
|
undef_count = 0
|
|
for s in f.symbols:
|
|
if not s.shndx and s.name:
|
|
undef_count += 1
|
|
last_ea = max(loadbase + end for start, end, name, kind in f.sections)
|
|
undef_entry_size = 8
|
|
undef_ea = ((last_ea + 0xFFF) & ~0xFFF) + undef_entry_size # plus 8 so we don't end up on the "end" symbol
|
|
|
|
undef_seg = basename + '.UNDEF' if basename is not None else 'UNDEF'
|
|
idaapi.add_segm(0, undef_ea, undef_ea+undef_count*undef_entry_size, undef_seg, 'XTRN')
|
|
segm = idaapi.get_segm_by_name(undef_seg)
|
|
segm.type = idaapi.SEG_XTRN
|
|
idaapi.update_segm(segm)
|
|
for i,s in enumerate(f.symbols):
|
|
if not s.shndx and s.name:
|
|
idc.MakeQword(undef_ea)
|
|
idaapi.do_name_anyway(undef_ea, s.name)
|
|
s.resolved = undef_ea
|
|
undef_ea += undef_entry_size
|
|
elif i != 0:
|
|
assert s.shndx
|
|
s.resolved = loadbase + s.value
|
|
if s.name:
|
|
if s.type == STT_FUNC:
|
|
idaapi.add_entry(s.resolved, s.resolved, s.name, 0)
|
|
else:
|
|
idaapi.do_name_anyway(s.resolved, s.name)
|
|
|
|
else:
|
|
# NULL symbol
|
|
s.resolved = 0
|
|
|
|
funcs = set()
|
|
for s in f.symbols:
|
|
if s.name and s.shndx and s.value:
|
|
if s.type == STT_FUNC:
|
|
funcs.add(loadbase+s.value)
|
|
symend = loadbase+s.value+s.size
|
|
if Dword(symend) != 0:
|
|
funcs.add(symend)
|
|
|
|
got_name_lookup = {}
|
|
for offset, r_type, sym, addend in f.relocations:
|
|
target = offset + loadbase
|
|
if r_type in (R_ARM_GLOB_DAT, R_ARM_JUMP_SLOT, R_ARM_ABS32):
|
|
if not sym:
|
|
print 'error: relocation at %X failed' % target
|
|
else:
|
|
idaapi.put_long(target, sym.resolved)
|
|
elif r_type == R_ARM_RELATIVE:
|
|
idaapi.put_long(target, idaapi.get_long(target) + loadbase)
|
|
elif r_type in (R_AARCH64_GLOB_DAT, R_AARCH64_JUMP_SLOT, R_AARCH64_ABS64):
|
|
idaapi.put_qword(target, sym.resolved + addend)
|
|
if addend == 0:
|
|
got_name_lookup[offset] = sym.name
|
|
elif r_type == R_AARCH64_RELATIVE:
|
|
idaapi.put_qword(target, loadbase + addend)
|
|
if addend < f.textsize:
|
|
funcs.add(loadbase + addend)
|
|
elif r_type == R_FAKE_RELR:
|
|
assert not f.armv7 # TODO
|
|
addend = idaapi.get_qword(target)
|
|
idaapi.put_qword(target, addend + loadbase)
|
|
if addend < f.textsize:
|
|
funcs.add(loadbase + addend)
|
|
else:
|
|
print 'TODO r_type %d' % (r_type,)
|
|
ida_make_offset(f, target)
|
|
|
|
for func, target in f.plt_entries:
|
|
if target in got_name_lookup:
|
|
addr = loadbase + func
|
|
funcs.add(addr)
|
|
idaapi.do_name_anyway(addr, got_name_lookup[target])
|
|
|
|
if not f.armv7:
|
|
funcs |= find_bl_targets(loadbase, loadbase+f.textsize)
|
|
|
|
if bypass_plt:
|
|
plt_lookup = f.plt_lookup
|
|
for pco in xrange(0, f.textsize, 4):
|
|
pc = loadbase + pco
|
|
d = Dword(pc)
|
|
if (d & 0x7c000000) == (0x94000000 & 0x7c000000):
|
|
imm = d & 0x3ffffff
|
|
if imm & 0x2000000:
|
|
imm |= ~0x1ffffff
|
|
if 0 <= imm <= 2:
|
|
continue
|
|
target = (pc + imm * 4) - loadbase
|
|
if target in plt_lookup:
|
|
new_target = plt_lookup[target] + loadbase
|
|
new_instr = (d & ~0x3ffffff) | (((new_target - pc) / 4) & 0x3ffffff)
|
|
idaapi.put_long(pc, new_instr)
|
|
|
|
for pco in xrange(0, f.textsize, 4):
|
|
pc = loadbase + pco
|
|
d = Dword(pc)
|
|
if d == 0x14000001:
|
|
funcs.add(pc + 4)
|
|
|
|
for pc, _ in f.eh_table:
|
|
funcs.add(loadbase + pc)
|
|
|
|
for addr in sorted(funcs, reverse=True):
|
|
idaapi.auto_make_proc(addr)
|
|
|
|
return 1
|