def _generate_cfg(self):
"""
Generate a control flow graph, make sure necessary steps are performed, and return a CFG.
:return: The CFG object
:rtype: angr.analyses.CFG
"""
# TODO
# 1) ida-like cfg
# 2) with some strategies we don't need the cfg, we should be able to apply those strategies even if the cfg fails
l.info("CFG start...")
cfg = self.project.analyses.CFGFast()
l.info("... CFG end")
return cfg
def apply_patches(self, patches):
# deal with stackable patches
# add stackable patches to the one with highest priority
insert_code_patches = [
p for p in patches if isinstance(p, InsertCodePatch)
]
insert_code_patches_dict = defaultdict(list)
for p in insert_code_patches:
insert_code_patches_dict[p.addr].append(p)
insert_code_patches_dict_sorted = defaultdict(list)
for k, v in insert_code_patches_dict.items():
insert_code_patches_dict_sorted[k] = sorted(
v, key=lambda x: -1 * x.priority)
insert_code_patches_stackable = [
p for p in patches
if isinstance(p, InsertCodePatch) and p.stackable
]
for sp in insert_code_patches_stackable:
assert len(sp.dependencies) == 0
if sp.addr in insert_code_patches_dict_sorted:
highest_priority_at_addr = insert_code_patches_dict_sorted[
sp.addr][0]
if highest_priority_at_addr != sp:
highest_priority_at_addr.asm_code += "\n" + sp.asm_code + "\n"
patches.remove(sp)
#deal with AddLabel patches
lpatches = [p for p in patches if (isinstance(p, AddLabelPatch))]
for p in lpatches:
self.name_map[p.name] = p.addr
# check for duplicate labels, it is not very necessary for this backend
# but it is better to behave in the same way of the reassembler backend
relevant_patches = [
p for p in patches
if isinstance(p, (AddCodePatch, AddEntryPointPatch,
InsertCodePatch))
]
all_code = ""
for p in relevant_patches:
if isinstance(p, InsertCodePatch):
code = p.code
else:
code = p.asm_code
all_code += "\n" + code + "\n"
labels = utils.string_to_labels(all_code)
duplicates = set(x for x in labels if labels.count(x) > 1)
if len(duplicates) > 1:
raise DuplicateLabelsException(
"found duplicate assembly labels: %s" % (str(duplicates)))
# for now any added code will be executed by jumping out and back ie CGRex
# apply all add code patches
self.added_code_file_start = len(self.ncontent)
self.name_map.force_insert("ADDED_CODE_START",
(len(self.ncontent) % 0x1000) +
self.added_code_segment)
# 0) RawPatch:
for patch in patches:
if isinstance(patch, RawFilePatch):
self.ncontent = utils.bytes_overwrite(self.ncontent,
patch.data,
patch.file_addr)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
for patch in patches:
if isinstance(patch, RawMemPatch):
self.patch_bin(patch.addr, patch.data)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
for patch in patches:
if isinstance(patch, RemoveInstructionPatch):
if patch.ins_size is None:
size = 4
else:
size = patch.ins_size
self.patch_bin(patch.ins_addr, b"\x60\x00\x00\x00" * int(
(size + 4 - 1) / 4))
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
# 5.5) ReplaceFunctionPatch (preprocessing rodata)
for patch in patches:
if isinstance(patch, ReplaceFunctionPatch):
patches += self.compile_function(patch.asm_code,
entry=patch.addr,
symbols=patch.symbols,
data_only=True,
prefix="_RFP" +
str(patches.index(patch)))
# 1) Add{RO/RW/RWInit}DataPatch
self.added_data_file_start = len(self.ncontent)
curr_data_position = self.name_map["ADDED_DATA_START"]
for patch in patches:
if isinstance(
patch,
(AddRWDataPatch, AddRODataPatch, AddRWInitDataPatch)):
if hasattr(patch, "data"):
final_patch_data = patch.data
else:
final_patch_data = b"\x00" * patch.len
self.added_data += final_patch_data
if patch.name is not None:
self.name_map[patch.name] = curr_data_position
curr_data_position += len(final_patch_data)
self.ncontent = utils.bytes_overwrite(self.ncontent,
final_patch_data)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
self.ncontent = utils.pad_bytes(
self.ncontent, 0x10) # some minimal alignment may be good
self.added_code_file_start = len(self.ncontent)
if self.replace_note_segment:
self.name_map.force_insert(
"ADDED_CODE_START",
int((curr_data_position + 0x10 - 1) / 0x10) * 0x10)
else:
self.name_map.force_insert("ADDED_CODE_START",
(len(self.ncontent) % 0x1000) +
self.added_code_segment)
# 2) AddCodePatch
# resolving symbols
current_symbol_pos = self.get_current_code_position()
for patch in patches:
if isinstance(patch, AddCodePatch):
if patch.is_c:
code_len = len(
self.compile_c(patch.asm_code,
optimization=patch.optimization,
compiler_flags=patch.compiler_flags))
else:
code_len = len(
self.compile_asm(patch.asm_code, current_symbol_pos))
if patch.name is not None:
self.name_map[patch.name] = current_symbol_pos
current_symbol_pos += code_len
# now compile for real
for patch in patches:
if isinstance(patch, AddCodePatch):
if patch.is_c:
new_code = self.compile_c(
patch.asm_code,
optimization=patch.optimization,
compiler_flags=patch.compiler_flags)
else:
new_code = self.compile_asm(
patch.asm_code, self.get_current_code_position(),
self.name_map)
self.added_code += new_code
self.ncontent = utils.bytes_overwrite(self.ncontent, new_code)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
# 3) AddEntryPointPatch
# basically like AddCodePatch but we detour by changing oep
# and we jump at the end of all of them
# resolving symbols
for patch in patches:
if isinstance(patch, AddEntryPointPatch):
old_oep = self.get_oep()
new_oep = self.get_current_code_position()
# ref: glibc/sysdeps/{ARCH}/start.S
instructions = patch.asm_code
instructions += "\nb {}".format(hex(int(old_oep)))
new_code = self.compile_asm(instructions,
self.get_current_code_position(),
self.name_map)
self.added_code += new_code
self.added_patches.append(patch)
self.ncontent = utils.bytes_overwrite(self.ncontent, new_code)
self.set_oep(new_oep)
l.info("Added patch: %s", str(patch))
# 4) InlinePatch
# we assume the patch never patches the added code
for patch in patches:
if isinstance(patch, InlinePatch):
new_code = self.compile_asm(patch.new_asm,
patch.instruction_addr,
self.name_map)
# Limiting the inline patch to a single block is not necessary
# assert len(new_code) <= self.project.factory.block(patch.instruction_addr, num_inst=patch.num_instr, max_size=).size
file_offset = self.project.loader.main_object.addr_to_offset(
patch.instruction_addr)
self.ncontent = utils.bytes_overwrite(self.ncontent, new_code,
file_offset)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
# 5) InsertCodePatch
# these patches specify an address in some basic block, In general we will move the basic block
# and fix relative offsets
# With this backend heer we can fail applying a patch, in case, resolve dependencies
insert_code_patches = [
p for p in patches if isinstance(p, InsertCodePatch)
]
insert_code_patches = sorted(insert_code_patches,
key=lambda x: -1 * x.priority)
applied_patches = []
while True:
name_list = [
str(p) if (p is None or p.name is None) else p.name
for p in applied_patches
]
l.info("applied_patches is: |%s|", "-".join(name_list))
assert all(a == b
for a, b in zip(applied_patches, insert_code_patches))
for patch in insert_code_patches[len(applied_patches):]:
self.save_state(applied_patches)
try:
l.info("Trying to add patch: %s", str(patch))
if patch.name is not None:
self.name_map[
patch.name] = self.get_current_code_position()
new_code = self.insert_detour(patch)
self.added_code += new_code
self.ncontent = utils.bytes_overwrite(
self.ncontent, new_code)
applied_patches.append(patch)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
except (DetourException, MissingBlockException,
DoubleDetourException) as e:
l.warning(e)
insert_code_patches, removed = self.handle_remove_patch(
insert_code_patches, patch)
#print map(str,removed)
applied_patches = self.restore_state(
applied_patches, removed)
l.warning(
"One patch failed, rolling back InsertCodePatch patches. Failed patch: %s",
str(patch))
break
# TODO: right now rollback goes back to 0 patches, we may want to go back less
# the solution is to save touched_bytes and ncontent indexed by applied patfch
# and go back to the biggest compatible list of patches
else:
break #at this point we applied everything in current insert_code_patches
# TODO symbol name, for now no name_map for InsertCode patches
header_patches = [InsertCodePatch,InlinePatch,AddEntryPointPatch,AddCodePatch, \
AddRWDataPatch,AddRODataPatch,AddRWInitDataPatch]
# 5.5) ReplaceFunctionPatch
for patch in patches:
if isinstance(patch, ReplaceFunctionPatch):
if self.structs.elfclass == 64:
# reloc type not supported (TOC info is in executables but not in object file, but relocs in object file will need TOC info.)
raise Exception(
"ReplaceFunctionPatch: PPC64 not yet supported")
for k, v in self.name_map.items():
if k.startswith("_RFP" + str(patches.index(patch))):
patch.symbols[k[len("_RFP" +
str(patches.index(patch))):]] = v
new_code = self.compile_function(
patch.asm_code,
bits=self.structs.elfclass,
little_endian=self.structs.little_endian,
entry=patch.addr,
symbols=patch.symbols)
file_offset = self.project.loader.main_object.addr_to_offset(
patch.addr)
self.ncontent = utils.bytes_overwrite(
self.ncontent, b"\x60\x00\x00\x00" * (patch.size // 4),
file_offset)
if (patch.size >= len(new_code)):
file_offset = self.project.loader.main_object.addr_to_offset(
patch.addr)
self.ncontent = utils.bytes_overwrite(
self.ncontent, new_code, file_offset)
else:
header_patches.append(ReplaceFunctionPatch)
detour_pos = self.get_current_code_position()
offset = self.project.loader.main_object.mapped_base if self.project.loader.main_object.pic else 0
new_code = self.compile_function(
patch.asm_code,
bits=self.structs.elfclass,
little_endian=self.structs.little_endian,
entry=detour_pos + offset,
symbols=patch.symbols)
self.added_code += new_code
self.ncontent = utils.bytes_overwrite(
self.ncontent, new_code)
# compile jmp
jmp_code = self.compile_jmp(patch.addr,
detour_pos + offset)
self.patch_bin(patch.addr, jmp_code)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
if any(isinstance(p,ins) for ins in header_patches for p in self.added_patches) or \
any(isinstance(p,SegmentHeaderPatch) for p in patches):
# either implicitly (because of a patch adding code or data) or explicitly, we need to change segment headers
# 6) SegmentHeaderPatch
segment_header_patches = [
p for p in patches if isinstance(p, SegmentHeaderPatch)
]
if len(segment_header_patches) > 1:
msg = "more than one patch tries to change segment headers: %s", "|".join(
[str(p) for p in segment_header_patches])
raise IncompatiblePatchesException(msg)
if len(segment_header_patches) == 1:
segment_patch = segment_header_patches[0]
segments = segment_patch.segment_headers
l.info("Added patch: %s", str(segment_patch))
else:
segments = self.modded_segments
for patch in [
p for p in patches if isinstance(p, AddSegmentHeaderPatch)
]:
# add after the first
segments = [segments[0]] + [patch.new_segment] + segments[1:]
self.setup_headers(segments)
self.set_added_segment_headers()
l.debug("final symbol table: %s",
repr([(k, hex(v)) for k, v in self.name_map.items()]))
else:
l.info("no patches, the binary will not be touched")
def apply_patches(self, patches):
# deal with stackable patches
# add stackable patches to the one with highest priority
insert_code_patches = [
p for p in patches if isinstance(p, InsertCodePatch)
]
insert_code_patches_dict = defaultdict(list)
for p in insert_code_patches:
insert_code_patches_dict[p.addr].append(p)
insert_code_patches_dict_sorted = defaultdict(list)
for k, v in insert_code_patches_dict.items():
insert_code_patches_dict_sorted[k] = sorted(
v, key=lambda x: -1 * x.priority)
insert_code_patches_stackable = [
p for p in patches
if isinstance(p, InsertCodePatch) and p.stackable
]
for sp in insert_code_patches_stackable:
assert len(sp.dependencies) == 0
if sp.addr in insert_code_patches_dict_sorted:
highest_priority_at_addr = insert_code_patches_dict_sorted[
sp.addr][0]
if highest_priority_at_addr != sp:
highest_priority_at_addr.asm_code += "\n" + sp.asm_code + "\n"
patches.remove(sp)
#deal with AddLabel patches
for patch in patches:
if isinstance(patch, AddLabelPatch):
self.name_map[patch.name] = patch.addr
# check for duplicate labels, it is not very necessary for this backend
# but it is better to behave in the same way of the reassembler backend
relevant_patches = [
p for p in patches
if isinstance(p, (AddCodePatch, InsertCodePatch))
]
all_code = ""
for p in relevant_patches:
if isinstance(p, InsertCodePatch):
code = p.code
else:
code = p.asm_code
all_code += "\n" + code + "\n"
labels = utils.string_to_labels(all_code)
duplicates = set(x for x in labels if labels.count(x) > 1)
if len(duplicates) > 1:
raise DuplicateLabelsException(
"found duplicate assembly labels: %s" % (str(duplicates)))
for patch in patches:
if isinstance(patch, (ReplaceFunctionPatch, AddEntryPointPatch,
AddSegmentHeaderPatch, SegmentHeaderPatch)):
raise NotImplementedError()
# 0) RawPatch:
for patch in patches:
if isinstance(patch, RawFilePatch):
self.ncontent = utils.bytes_overwrite(self.ncontent,
patch.data,
patch.file_addr)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
for patch in patches:
if isinstance(patch, RawMemPatch):
self.patch_bin(patch.addr, patch.data)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
for patch in patches:
if isinstance(patch, RemoveInstructionPatch):
if patch.ins_size is None:
size = 2
else:
size = patch.ins_size
self.patch_bin(patch.ins_addr, b"\x00\x00" * int(
(size + 2 - 1) / 2))
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
# 1) Add{RO/RW/RWInit}DataPatch
curr_data_position = self.name_map["ADDED_DATA_START"]
for patch in patches:
if isinstance(
patch,
(AddRWDataPatch, AddRODataPatch, AddRWInitDataPatch)):
if hasattr(patch, "data"):
final_patch_data = patch.data
else:
final_patch_data = b"\x00" * patch.len
self.added_data += final_patch_data
if patch.name is not None:
self.name_map[patch.name] = curr_data_position
curr_data_position += len(final_patch_data)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
if ((len(self.added_data) + self.added_data_file_start) % 2 == 1):
self.added_data += b"\x00"
self.ncontent = self.insert_bytes(self.ncontent, self.added_data,
self.added_data_file_start)
self.added_code_file_start = self.added_data_file_start + len(
self.added_data)
self.name_map.force_insert(
"ADDED_CODE_START", self.added_code_file_start -
(self.text_section_offset - self.text_section_addr))
# __do_copy_data
# FIXME: not working properly
if len(self.added_data) > 0:
data_start = self.name_map["ADDED_DATA_START"]
data_end = curr_data_position
data_load_start = self.name_map["ADDED_CODE_START"] - len(
self.added_data)
data_start_hi8, data_start_lo8 = data_start >> 8, data_start & 0xFF
data_end_hi8, data_end_lo8 = data_end >> 8, data_end & 0xFF
data_load_start_hi8, data_load_start_lo8 = data_load_start >> 8, data_load_start & 0xFF
do_copy_data_code = '''
ldi r17, %d
ldi r26, %d
ldi r27, %d
ldi r30, %d
ldi r31, %d
rjmp +0x16
lpm r0, z+
st x+, r0
cpi r26, %d
cpc r27, r17
brne 0x2
''' % (data_end_hi8, data_start_lo8, data_start_hi8,
data_load_start_lo8, data_load_start_hi8, data_end_lo8)
# TODO: should not be hardcoded to 0x8c
# we are assuming that 0x8c is end of orginal __do_copy_data and start of __do_clear_bss
patches.insert(
0,
InsertCodePatch(0x8c,
code=do_copy_data_code,
name="__do_copy_data",
priority=1000))
# 2) AddCodePatch
# resolving symbols
current_symbol_pos = self.get_current_code_position()
for patch in patches:
if isinstance(patch, AddCodePatch):
if patch.is_c:
code_len = len(
self.compile_c(patch.asm_code,
optimization=patch.optimization,
compiler_flags=patch.compiler_flags))
else:
code_len = len(
self.compile_asm(patch.asm_code, current_symbol_pos))
if patch.name is not None:
self.name_map[patch.name] = current_symbol_pos
current_symbol_pos += code_len
# now compile for real
self.added_code = b""
for patch in patches:
if isinstance(patch, AddCodePatch):
if patch.is_c:
new_code = self.compile_c(
patch.asm_code,
optimization=patch.optimization,
compiler_flags=patch.compiler_flags)
else:
new_code = self.compile_asm(patch.asm_code, self.name_map)
self.added_code += new_code
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
# 4) InlinePatch
# we assume the patch never patches the added code
for patch in patches:
if isinstance(patch, InlinePatch):
new_code = self.compile_asm(patch.new_asm, self.name_map)
file_offset = self.project.loader.main_object.addr_to_offset(
patch.instruction_addr)
self.ncontent = utils.bytes_overwrite(self.ncontent, new_code,
file_offset)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
# 5) InsertCodePatch
# these patches specify an address in some basic block, In general we will move the basic block
# and fix relative offsets
# With this backend heer we can fail applying a patch, in case, resolve dependencies
insert_code_patches = [
p for p in patches if isinstance(p, InsertCodePatch)
]
insert_code_patches = sorted(insert_code_patches,
key=lambda x: -1 * x.priority)
applied_patches = []
while True:
name_list = [
str(p) if (p is None or p.name is None) else p.name
for p in applied_patches
]
l.info("applied_patches is: |%s|", "-".join(name_list))
assert all(a == b
for a, b in zip(applied_patches, insert_code_patches))
for patch in insert_code_patches[len(applied_patches):]:
self.save_state(applied_patches)
try:
l.info("Trying to add patch: %s", str(patch))
if patch.name is not None:
self.name_map[
patch.name] = self.get_current_code_position()
new_code = self.insert_detour(patch)
self.added_code += new_code
applied_patches.append(patch)
self.added_patches.append(patch)
l.info("Added patch: %s", str(patch))
except (DetourException, MissingBlockException,
DoubleDetourException) as e:
l.warning(e)
insert_code_patches, removed = self.handle_remove_patch(
insert_code_patches, patch)
#print map(str,removed)
applied_patches = self.restore_state(
applied_patches, removed)
l.warning(
"One patch failed, rolling back InsertCodePatch patches. Failed patch: %s",
str(patch))
break
# TODO: right now rollback goes back to 0 patches, we may want to go back less
# the solution is to save touched_bytes and ncontent indexed by applied patfch
# and go back to the biggest compatible list of patches
else:
break #at this point we applied everything in current insert_code_patches
# TODO symbol name, for now no name_map for InsertCode patches
self.ncontent = self.insert_bytes(self.ncontent, self.added_code,
self.added_code_file_start)
# Modifiy sections if needed
if (len(self.added_data) + len(self.added_code) > 0):
# update ELF header
current_Ehdr = self.structs.Elf_Ehdr.parse(self.ncontent)
current_Ehdr['e_shoff'] += len(self.added_code) + len(
self.added_data)
self.ncontent = utils.bytes_overwrite(
self.ncontent, self.structs.Elf_Ehdr.build(current_Ehdr), 0)
# update section headers
current_Shdr_index = -1
for section in self.sections:
current_Shdr_index += 1
current_Shdr = section.header
if section.name == ".text":
pass
elif section.name == ".data":
current_Shdr['sh_size'] += len(self.added_code) + len(
self.added_data)
current_Shdr['sh_addr'] = self.text_section_size
else:
current_Shdr['sh_offset'] += len(self.added_code) + len(
self.added_data)
self.ncontent = utils.bytes_overwrite(
self.ncontent, self.structs.Elf_Shdr.build(current_Shdr),
current_Ehdr['e_shoff'] +
current_Ehdr['e_shentsize'] * current_Shdr_index)