def validate(self, operation, odb_file):
"""" Validate the type fits at the given address. """
vma = operation.vma
# determine the size of this defined data
size = operation.size
# Verify there is not something already defined here
for dd in odb_file.get_structure_list(DefinedData):
if dd.overlaps(vma, size):
raise DefineDataCollisionException(
'Data collides with existing data %s' % dd.var_name)
# Verify there is enough room in the parcel
parcels = ParcelList(odb_file.get_structure_list(Parcel))
parcel = parcels.find_parcel_by_vma(vma)
if not parcel:
raise ValidationError('Cannot define data at bad address 0x%x' %
vma)
elif parcel.is_code:
raise ValidationError('Cannot define data in code region')
if not parcel.contains_vma(vma + size - 1):
raise ValidationError('Not enough room to define data')
def operate(self, odb_file):
self.object_id = odb_file.get_object_id()
parcels = ParcelList(odb_file.get_structure_list(Parcel))
parcel = parcels.find_parcel_by_vma(self.vma)
maxLen = parcel.vma_end - self.vma
ofd = Ofd(odb_file.binary)
rawData = ofd.bfd.raw_data(ofd.bfd.sections[parcel.sec_name], self.vma,
maxLen)
rawData = bytes(rawData)
# instantiate the type to determine its size
type_inst = self.instantiate_type(self.type_kind, self.type_name,
odb_file)
self.size = type_inst.calc_size(rawData)
if (self.size <= 0):
raise DefineDataException("Cannot instantiate empty type")
try:
FitValidator().validate(self, odb_file)
except DefineDataCollisionException as e:
merged = False
for dd in odb_file.get_structure_list(DefinedData):
# find the defined data that overlaps
if dd.overlaps(self.vma, self.size):
# instantiate the class
dd_inst = self.instantiate_type(dd.type_kind, dd.type_name,
odb_file)
# if we can merge
if dd_inst.can_merge(self.type_kind, self.type_name):
# delete the data we're going to merge with
odb_file.execute(UndefineDataOperation(dd.vma))
# run the fit again just to be sure
FitValidator().validate(self, odb_file)
merged = True
if not merged:
raise e
definedData = odb_file.create_item(
DefinedData, {
'type_kind': self.type_kind,
'type_name': self.type_name,
'var_name': self.var_name,
'vma': self.vma,
'size': self.size,
})
odb_file.insert_item(definedData)
def make_code(self, vma):
parcels = ParcelList(self.odb_file.get_structure_list(Parcel))
p = parcels.find_parcel_by_vma(vma)
if p.is_code:
raise Exception("Cannot make code on a code parcel")
self.odb_file.execute(ActiveScanOperation(vma))
def make_data(self, vma):
parcels = ParcelList(self.odb_file.get_structure_list(Parcel))
p = parcels.find_parcel_by_vma(vma)
if not p.is_code:
raise Exception("Cannot make data on non-code parcel")
self.odb_file.execute(SplitParcelOperation(vma))
def convert_to_code(vma, odb_file, analyzer, the_ofd):
parcels = ParcelList(odb_file.get_structure_list(Parcel))
if len(parcels) == 0:
raise Exception('Invalid operation: No parcels found')
parcels_to_adjust = [p for p in parcels if p.vma_start > vma]
p = parcels.find_parcel_by_vma(vma)
if p is None:
raise Exception('Invalid address: 0x%08x' % vma)
# before doing anything, check if the given address starts with at least one valid instruction
if not self._is_valid_instruction(the_ofd, analyzer, vma,
p.vma_end):
raise Exception("Failed to make code: invalid opcode")
bad_code_scanner = BadCodeScanner(analyzer)
function_scanner = FunctionScanner(
analyzer, odb_file.get_structure_list(Symbol))
branch_scanner = BranchScanner(analyzer)
parcel_scanner = ParcelOffsetScanner()
scanners = [
# The parcel scanner must come first here, because we need it to 'push_vma' for the bad instruction, if
# one is encountered
parcel_scanner,
bad_code_scanner,
function_scanner,
branch_scanner,
]
# split the data parcel
odb_file.execute(SplitParcelOperation(vma))
parcels = ParcelList(odb_file.get_structure_list(Parcel))
p = parcels.find_parcel_by_vma(vma)
ldas_before_scan = p.size # one lda per byte
def callback(addr, rawData, instr, abfd, self):
for scanner in scanners:
scanner.scan_line(addr, rawData, instr, abfd, p)
section = the_ofd.get_section_from_addr(vma)
the_ofd.disassemble(section.name, [],
vma,
p.vma_end,
funcFmtLine=callback,
funcFmtLineArgs={
'self': self,
})
ldas_removed = ldas_before_scan - p.num_ldas
for p in parcels_to_adjust:
p.lda_start -= ldas_removed
for scanner in scanners:
scanner.commit(odb_file)
bad_addr = bad_code_scanner.bad_addr
if bad_addr is not None:
p = parcels.find_parcel_by_vma(bad_addr)
odb_file.execute(SplitParcelOperation(bad_addr))
#
# consolidate the code parcels
#
parcels = ParcelList(odb_file.get_structure_list(Parcel))
transaction = parcels.consolidate()
# remove dead parcels
for p in transaction.to_remove:
odb_file.remove_item(p)
def operate(self, odb_file):
self.object_id = odb_file.get_object_id()
analyzer = get_processor(odb_file.get_arch(), odb_file)
the_ofd = ofd.Ofd(odb_file.get_binary())
def convert_to_code(vma, odb_file, analyzer, the_ofd):
parcels = ParcelList(odb_file.get_structure_list(Parcel))
if len(parcels) == 0:
raise Exception('Invalid operation: No parcels found')
parcels_to_adjust = [p for p in parcels if p.vma_start > vma]
p = parcels.find_parcel_by_vma(vma)
if p is None:
raise Exception('Invalid address: 0x%08x' % vma)
# before doing anything, check if the given address starts with at least one valid instruction
if not self._is_valid_instruction(the_ofd, analyzer, vma,
p.vma_end):
raise Exception("Failed to make code: invalid opcode")
bad_code_scanner = BadCodeScanner(analyzer)
function_scanner = FunctionScanner(
analyzer, odb_file.get_structure_list(Symbol))
branch_scanner = BranchScanner(analyzer)
parcel_scanner = ParcelOffsetScanner()
scanners = [
# The parcel scanner must come first here, because we need it to 'push_vma' for the bad instruction, if
# one is encountered
parcel_scanner,
bad_code_scanner,
function_scanner,
branch_scanner,
]
# split the data parcel
odb_file.execute(SplitParcelOperation(vma))
parcels = ParcelList(odb_file.get_structure_list(Parcel))
p = parcels.find_parcel_by_vma(vma)
ldas_before_scan = p.size # one lda per byte
def callback(addr, rawData, instr, abfd, self):
for scanner in scanners:
scanner.scan_line(addr, rawData, instr, abfd, p)
section = the_ofd.get_section_from_addr(vma)
the_ofd.disassemble(section.name, [],
vma,
p.vma_end,
funcFmtLine=callback,
funcFmtLineArgs={
'self': self,
})
ldas_removed = ldas_before_scan - p.num_ldas
for p in parcels_to_adjust:
p.lda_start -= ldas_removed
for scanner in scanners:
scanner.commit(odb_file)
bad_addr = bad_code_scanner.bad_addr
if bad_addr is not None:
p = parcels.find_parcel_by_vma(bad_addr)
odb_file.execute(SplitParcelOperation(bad_addr))
#
# consolidate the code parcels
#
parcels = ParcelList(odb_file.get_structure_list(Parcel))
transaction = parcels.consolidate()
# remove dead parcels
for p in transaction.to_remove:
odb_file.remove_item(p)
convert_to_code(self.vma, odb_file, analyzer, the_ofd)
# convert newly found functions to code as well
following_new_funcs = True
while following_new_funcs:
funcs = odb_file.get_structure_list(Function)
parcels = ParcelList(odb_file.get_structure_list(Parcel))
# assume we are done unless we find at least one function that is within a data parcel
following_new_funcs = False
for f in funcs:
p = parcels.find_parcel_by_vma(f.vma)
if ( # if this function is at a valid address
(p is not None) and
# and the function is within a data parcel
(not p.is_code) and
# and the function looks to be valid code
self._is_valid_instruction(the_ofd, analyzer, f.vma,
p.vma_end)):
# convert this function to code
convert_to_code(f.vma, odb_file, analyzer, the_ofd)
# take another pass, in case we discovered new functions to disassemble
following_new_funcs = True
# break out of the for loop, but not the while
break
def test_split_to_data_single_inst_in_parcel(self):
odb_file = OdbFile(BinaryFile(
self.get_test_bin_path('code_split_single_instruction/split_parcel_single'), 'elf64-x86-64', 'i386:x86-64'))
odb_file.execute(LoadOperation())
odb_file.execute(PassiveScanOperation())
parcels_before = ParcelList(odb_file.get_structure_list(Parcel))
# only one instruction in .single
p = parcels_before.find_parcel_by_vma(0x400598)
p_text = parcels_before.find_parcel_by_vma(0x400597)
p_fini = parcels_before.find_parcel_by_vma(0x4005a0)
p_got = parcels_before.find_parcel_by_vma(0x600fe0)
num_ldas_before = p.num_ldas
num_text_ldas_before = p_text.num_ldas
num_fini_ldas_before = p_fini.num_ldas
single_lda_before = p.lda_start
fini_lda_before = p_fini.lda_start
text_lda_before = p_text.lda_start
got_lda_before = p_got.lda_start
# number of parcels before the split
self.assertEquals(25, len(parcels_before))
# execute the split operation
odb_file.execute(SplitParcelOperation(0x400598))
# number of parcels after the split (should not have changed)
parcels_after = ParcelList(odb_file.get_structure_list(Parcel))
self.assertEquals(25, len(parcels_after))
# get the parcels
p = parcels_after.find_parcel_by_vma(0x400598)
p_text = parcels_after.find_parcel_by_vma(0x400597)
p_fini = parcels_after.find_parcel_by_vma(0x4005a0)
# make sure they are all unique
self.assertNotEqual(p, p_text)
self.assertNotEqual(p, p_fini)
self.assertNotEqual(p_text, p_fini)
# check num_ldas in p (1 instruction become 7 data bytes)
self.assertEquals(7, p.num_ldas)
# check num_ldas in text didn't change
self.assertEquals(num_text_ldas_before, p_text.num_ldas)
# check num_ldas in fini didn't change
self.assertEquals(num_fini_ldas_before, p_fini.num_ldas)
# make sure vmas line up
self.assertEquals(0x4003d0, p_text.vma_start)
self.assertEquals(0x400598, p_text.vma_end)
self.assertEquals(0x400598, p.vma_start)
self.assertEquals(0x40059f, p.vma_end)
self.assertEquals(0x4005a0, p_fini.vma_start)
self.assertEquals(0x4005ae, p_fini.vma_end)
#
# make sure vma->lda mappings line up
#
# these shouldn't have changed
self.assertEquals(text_lda_before, parcels_after.vma_to_lda(0x4003d0))
self.assertEquals(text_lda_before+num_text_ldas_before-1, parcels_after.vma_to_lda(0x400597))
self.assertEquals(single_lda_before, parcels_after.vma_to_lda(0x400598))
self.assertEquals(True, p_text.is_code)
self.assertEquals(True, p_fini.is_code)
# these changed
self.assertEquals(single_lda_before+1, parcels_after.vma_to_lda(0x400599))
self.assertEquals(single_lda_before+2, parcels_after.vma_to_lda(0x40059a))
self.assertEquals(single_lda_before+3, parcels_after.vma_to_lda(0x40059b))
self.assertEquals(single_lda_before+4, parcels_after.vma_to_lda(0x40059c))
self.assertEquals(single_lda_before+5, parcels_after.vma_to_lda(0x40059d))
self.assertEquals(single_lda_before+6, parcels_after.vma_to_lda(0x40059e))
self.assertEquals(False, p.is_code)
# 1 instruction turned into 7 data bytes, so it shifted the following parcels by 6 ldas
self.assertEquals(fini_lda_before+6, parcels_after.vma_to_lda(0x4005a0))
self.assertEquals(got_lda_before+6, parcels_after.vma_to_lda(0x600fe0))
#
# make sure lda->vma mappings line up
#
self.assertEquals(text_lda_before, parcels_after.vma_to_lda(0x4003d0))
self.assertEquals(text_lda_before+num_text_ldas_before-1, parcels_after.vma_to_lda(0x400597))
self.assertEquals(single_lda_before, parcels_after.vma_to_lda(0x400598))
# these shouldn't have changed
self.assertEquals(0x4003d0, parcels_after.lda_to_vma(text_lda_before))
self.assertEquals(0x400597, parcels_after.lda_to_vma(text_lda_before+num_text_ldas_before-1))
self.assertEquals(0x400598, parcels_after.lda_to_vma(single_lda_before))
# these changed
self.assertEquals(0x400599, parcels_after.lda_to_vma(single_lda_before+1))
self.assertEquals(0x40059a, parcels_after.lda_to_vma(single_lda_before+2))
self.assertEquals(0x40059b, parcels_after.lda_to_vma(single_lda_before+3))
self.assertEquals(0x40059c, parcels_after.lda_to_vma(single_lda_before+4))
self.assertEquals(0x40059d, parcels_after.lda_to_vma(single_lda_before+5))
self.assertEquals(0x40059e, parcels_after.lda_to_vma(single_lda_before+6))
# 1 instruction turned into 7 data bytes, so it shifted the following parcels by 6 ldas
self.assertEquals(0x4005a0, parcels_after.lda_to_vma(fini_lda_before+6))
self.assertEquals(0x600fe0, parcels_after.lda_to_vma(got_lda_before+6))
def test_split_to_data_at_end_of_parcel(self):
odb_file = OdbFile(BinaryFile(self.get_test_bin_path('ls'), 'elf64-x86-64', 'i386:x86-64'))
odb_file.execute(LoadOperation())
odb_file.execute(PassiveScanOperation())
parcels_before = ParcelList(odb_file.get_structure_list(Parcel))
# last instruction in .plt
p = parcels_before.find_parcel_by_vma(0x4029bb)
p3 = parcels_before.find_parcel_by_vma(0x4029c0)
num_p1_ldas_before = p.num_ldas
num_p3_ldas_before = p3.num_ldas
plt_lda_before = parcels_before.vma_to_lda(0x402320)
text_lda_before = parcels_before.vma_to_lda(0x4029c0)
# number of parcels before the split
self.assertEquals(25, len(parcels_before))
# execute the split operation
odb_file.execute(SplitParcelOperation(0x4029bb))
# number of parcels after the split (should only add one more)
parcels_after = ParcelList(odb_file.get_structure_list(Parcel))
self.assertEquals(26, len(parcels_after))
# get the two parcels (plus the .text one following)
p1 = parcels_after.find_parcel_by_vma(0x402320)
p2 = parcels_after.find_parcel_by_vma(0x4029bb)
p3 = parcels_after.find_parcel_by_vma(0x4029c0)
# make sure they are all unique
self.assertNotEqual(p1, p2)
self.assertNotEqual(p1, p3)
self.assertNotEqual(p2, p3)
# check num_ldas in p1 (5 instructions)
self.assertEquals(num_p1_ldas_before - 1, p1.num_ldas)
# check num_ldas in p2 (this one instruction has now become 5 data bytes)
self.assertEquals(5, p2.num_ldas)
# check num_ldas in p3 (should not have changed)
self.assertEquals(num_p3_ldas_before, p3.num_ldas)
# make sure vmas line up
self.assertEquals(0x402320, p1.vma_start)
self.assertEquals(0x4029bb, p1.vma_end)
self.assertEquals(0x4029bb, p2.vma_start)
self.assertEquals(0x4029c0, p2.vma_end)
self.assertEquals(0x4029c0, p3.vma_start)
self.assertEquals(0x412bec, p3.vma_end)
#
# make sure vma->lda mappings line up
#
# these shouldn't have changed
self.assertEquals(plt_lda_before, parcels_after.vma_to_lda(0x402320))
self.assertEquals(plt_lda_before+num_p1_ldas_before-2, parcels_after.vma_to_lda(0x4029b6))
self.assertEquals(plt_lda_before+num_p1_ldas_before-1, parcels_after.vma_to_lda(0x4029bb))
self.assertEquals(True, p1.is_code)
self.assertEquals(True, p3.is_code)
# these changed
self.assertEquals(plt_lda_before+num_p1_ldas_before, parcels_after.vma_to_lda(0x4029bc))
self.assertEquals(plt_lda_before+num_p1_ldas_before+1, parcels_after.vma_to_lda(0x4029bd))
self.assertEquals(plt_lda_before+num_p1_ldas_before+2, parcels_after.vma_to_lda(0x4029be))
self.assertEquals(plt_lda_before+num_p1_ldas_before+3, parcels_after.vma_to_lda(0x4029bf))
self.assertEquals(plt_lda_before+num_p1_ldas_before+4, parcels_after.vma_to_lda(0x4029c0))
self.assertEquals(False, p2.is_code)
# 1 instruction turned into 5 data bytes, so it shifted the following parcels by 4 ldas
self.assertEquals(text_lda_before+4, parcels_after.vma_to_lda(0x4029c0))
#
# make sure lda->vma mappings line up
#
# these shouldn't have changed
self.assertEquals(0x402320, parcels_after.lda_to_vma(plt_lda_before))
self.assertEquals(0x4029b6, parcels_after.lda_to_vma(plt_lda_before+num_p1_ldas_before-2))
self.assertEquals(0x4029bb, parcels_after.lda_to_vma(plt_lda_before+num_p1_ldas_before-1))
# these changed
self.assertEquals(0x4029bc, parcels_after.lda_to_vma(plt_lda_before+num_p1_ldas_before))
self.assertEquals(0x4029bd, parcels_after.lda_to_vma(plt_lda_before+num_p1_ldas_before+1))
self.assertEquals(0x4029be, parcels_after.lda_to_vma(plt_lda_before+num_p1_ldas_before+2))
self.assertEquals(0x4029bf, parcels_after.lda_to_vma(plt_lda_before+num_p1_ldas_before+3))
self.assertEquals(0x4029c0, parcels_after.lda_to_vma(plt_lda_before+num_p1_ldas_before+4))
# 1 instruction turned into 5 data bytes, so it shifted the following parcels by 4 ldas
self.assertEquals(0x4029c0, parcels_after.lda_to_vma(text_lda_before+4))
def test_split_to_data_basic(self):
odb_file = OdbFile(BinaryFile(self.get_test_bin_path('ls'), 'elf64-x86-64', 'i386:x86-64'))
odb_file.execute(LoadOperation())
odb_file.execute(PassiveScanOperation())
parcels_before = ParcelList(odb_file.get_structure_list(Parcel))
p = parcels_before.find_parcel_by_vma(0x4029c0)
num_p1_ldas_before = p.num_ldas
text_lda_before = parcels_before.vma_to_lda(0x4029c0)
fini_lda_before = parcels_before.vma_to_lda(0x412bec)
# number of parcels before the split
self.assertEquals(25, len(parcels_before))
# execute the split operation
odb_file.execute(SplitParcelOperation(0x4029d5))
# number of parcels after the split
parcels_after = ParcelList(odb_file.get_structure_list(Parcel))
self.assertEquals(27, len(parcels_after))
# get the three parcels
p1 = parcels_after.find_parcel_by_vma(0x4029c0)
p2 = parcels_after.find_parcel_by_vma(0x4029d5)
p3 = parcels_after.find_parcel_by_vma(0x4029da)
# make sure they are all unique
self.assertNotEqual(p1, p2)
self.assertNotEqual(p1, p3)
self.assertNotEqual(p2, p3)
# check num_ldas in p1 (5 instructions)
self.assertEquals(5, p1.num_ldas)
# check num_ldas in p2 (this one instruction has now become 5 data bytes)
self.assertEquals(5, p2.num_ldas)
# check num_ldas in p3 (less 5 from p1 and less 1 from p2 - remember it used to only be 1 lda before the split)
self.assertEquals(num_p1_ldas_before - 5 - 1, p3.num_ldas)
# make sure vmas line up
self.assertEquals(0x4029c0, p1.vma_start)
self.assertEquals(0x4029d5, p1.vma_end)
self.assertEquals(0x4029d5, p2.vma_start)
self.assertEquals(0x4029da, p2.vma_end)
self.assertEquals(0x4029da, p3.vma_start)
self.assertEquals(0x412bec, p3.vma_end)
#
# make sure vma->lda mappings line up
#
# these shouldn't have changed
self.assertEquals(text_lda_before, parcels_after.vma_to_lda(0x4029c0))
self.assertEquals(text_lda_before+5, parcels_after.vma_to_lda(0x4029d5))
self.assertEquals(True, p1.is_code)
self.assertEquals(True, p3.is_code)
# these changed
self.assertEquals(text_lda_before+6, parcels_after.vma_to_lda(0x4029d6))
self.assertEquals(text_lda_before+7, parcels_after.vma_to_lda(0x4029d7))
self.assertEquals(text_lda_before+8, parcels_after.vma_to_lda(0x4029d8))
self.assertEquals(text_lda_before+9, parcels_after.vma_to_lda(0x4029d9))
self.assertEquals(text_lda_before+10, parcels_after.vma_to_lda(0x4029da))
self.assertEquals(text_lda_before+11, parcels_after.vma_to_lda(0x4029db))
self.assertEquals(False, p2.is_code)
# 1 instruction turned into 5 data bytes, so it shifted the following parcels by 4 ldas
self.assertEquals(fini_lda_before+4, parcels_after.vma_to_lda(0x412bec))
#
# make sure lda->vma mappings line up
#
# these shouldn't have changed
self.assertEquals(0x4029c0, parcels_after.lda_to_vma(text_lda_before))
self.assertEquals(0x4029d5, parcels_after.lda_to_vma(text_lda_before+5))
# these changed
self.assertEquals(0x4029d6, parcels_after.lda_to_vma(text_lda_before+6))
self.assertEquals(0x4029d7, parcels_after.lda_to_vma(text_lda_before+7))
self.assertEquals(0x4029d8, parcels_after.lda_to_vma(text_lda_before+8))
self.assertEquals(0x4029d9, parcels_after.lda_to_vma(text_lda_before+9))
self.assertEquals(0x4029da, parcels_after.lda_to_vma(text_lda_before+10))
self.assertEquals(0x4029db, parcels_after.lda_to_vma(text_lda_before+11))
# 1 instruction turned into 5 data bytes, so it shifted the following parcels by 4 ldas
self.assertEquals(0x412bec, parcels_after.lda_to_vma(fini_lda_before+4))
def test_split_to_data_at_start_of_parcel(self):
odb_file = OdbFile(BinaryFile(self.get_test_bin_path('ls'), 'elf64-x86-64', 'i386:x86-64'))
odb_file.execute(LoadOperation())
odb_file.execute(PassiveScanOperation())
parcels_before = ParcelList(odb_file.get_structure_list(Parcel))
# start of .fini section
p = parcels_before.find_parcel_by_vma(0x412bec)
num_p1_ldas_before = p.num_ldas
fini_lda_before = parcels_before.vma_to_lda(0x412bec)
# number of parcels before the split
self.assertEquals(25, len(parcels_before))
# execute the split operation
odb_file.execute(SplitParcelOperation(0x412bec))
# number of parcels after the split (should only add one more)
parcels_after = ParcelList(odb_file.get_structure_list(Parcel))
self.assertEquals(26, len(parcels_after))
# get the two parcels
p1 = parcels_after.find_parcel_by_vma(0x412bec)
p2 = parcels_after.find_parcel_by_vma(0x412bf0)
# make sure they are unique
self.assertNotEqual(p1, p2)
# check num_ldas in p1 (4 data bytes)
self.assertEquals(4, p1.num_ldas)
# check num_ldas in p2
self.assertEquals(2, p2.num_ldas)
# make sure vmas line up
self.assertEquals(0x412bec, p1.vma_start)
self.assertEquals(0x412bf0, p1.vma_end)
self.assertEquals(0x412bf0, p2.vma_start)
self.assertEquals(0x412bf5, p2.vma_end)
#
# make sure vma->lda mappings line up
#
# these shouldn't have changed
self.assertEquals(fini_lda_before, parcels_after.vma_to_lda(0x412bec))
self.assertEquals(True, p2.is_code)
# these changed
self.assertEquals(fini_lda_before+1, parcels_after.vma_to_lda(0x412bed))
self.assertEquals(fini_lda_before+2, parcels_after.vma_to_lda(0x412bee))
self.assertEquals(fini_lda_before+3, parcels_after.vma_to_lda(0x412bef))
self.assertEquals(fini_lda_before+4, parcels_after.vma_to_lda(0x412bf0))
self.assertEquals(fini_lda_before+5, parcels_after.vma_to_lda(0x412bf4))
self.assertEquals(False, p1.is_code)
#
# make sure lda->vma mappings line up
#
# these shouldn't have changed
self.assertEquals(0x412bec, parcels_after.lda_to_vma(fini_lda_before))
# these changed
self.assertEquals(0x412bed, parcels_after.lda_to_vma(fini_lda_before+1))
self.assertEquals(0x412bee, parcels_after.lda_to_vma(fini_lda_before+2))
self.assertEquals(0x412bef, parcels_after.lda_to_vma(fini_lda_before+3))
self.assertEquals(0x412bf0, parcels_after.lda_to_vma(fini_lda_before+4))
self.assertEquals(0x412bf4, parcels_after.lda_to_vma(fini_lda_before+5))
class BranchGraphView(list):
def __init__(self, odb_file, vma):
self.odb_file = odb_file
start = None
end = None
self.parcels = ParcelList(odb_file.get_structure_list(Parcel))
# first, some validation
p = self.parcels.find_parcel_by_vma(vma)
if p:
# initialize to parcel boundaries
start = p.vma_start
end = p.vma_end
# if this is not in a code section
if not p.is_code:
raise Exception(
"Graph view is not available for data sections!")
# else, invalid address
else:
raise Exception(
"Graph view is not available for invalid address 0x%x!" % vma)
if type(odb_file.binary) == BinaryString:
# TODO: Handle nonzero base address
start = 0
end = odb_file.binary.size
functions = sorted(odb_file.get_structure_list(Function),
key=lambda f: f.vma)
for f in functions:
if vma >= f.vma:
start = f.vma
elif vma < f.vma:
end = f.vma
break
# get all the branches within this function
# TODO: Deal with indirect branches (i.e., jmp rax)
branches = filter(
lambda b: b.srcAddr >= start and b.srcAddr < end and b.
targetAddr >= start and b.targetAddr < end,
odb_file.get_structure_list(Branch))
# dict of nodes, keyed by starting address of the block
self._nodes = {}
thisBlock = start
for b in sorted(branches, key=lambda b: b.srcAddr):
# address if the branch is not taken (i.e., the addr of the next instruction)
no = b.srcAddr + b.instrLen
# address if the branch is taken
yes = b.targetAddr
# first node
self._nodes[thisBlock] = BranchGraphNode(thisBlock, b.srcAddr,
no - thisBlock, yes, no)
# advance to next block
thisBlock = no
# make last block
if thisBlock != end:
end_line = self.parcels.vma_to_lda(end - 1)
last_addr = self.parcels.lda_to_vma(end_line)
self._nodes[thisBlock] = BranchGraphNode(thisBlock, last_addr,
end - thisBlock, None,
None)
# for each target 'yes' branch address
# NOTE: The 'no' branch addresses are guaranteed to be nodes by this point
# NOTE: We can't iterate over the nodes with iteritems() since we modify the dictionary in the loop
for addr in [n.yesAddr for n in self._nodes.values()]:
# if this is a new node
if addr not in self._nodes:
# find the node that contains this address (the parent) and split it
for parent in list(self._nodes.values()):
if addr in parent:
newNode = parent.split(addr, self.parcels)
self._nodes[addr] = newNode
'''
{
nodes:
[
{id:X, instructions: [du1, du2, du3, du4 ... }
{id:Y, instructions: [du9, du10, du11 .... }
....
]
links: [
{ from:x, to:y, type: 'taken' }
{ from:y, to: z, type: 'notTaken'}
{ from:o, to: q, type: 'unconditional'}
]
}
'''
def getNodes(self):
dasm = Disassembler(self.odb_file)
# convert _nodes to representation needed by front end
nodes = []
for startAddr, node in self._nodes.items():
startLine = self.parcels.vma_to_lda(startAddr)
endLine = self.parcels.vma_to_lda(startAddr + node.size)
if endLine is None:
endLine = self.parcels.sum_ldas()
dus = dasm.display(startAddr, endLine - startLine, False)
nodes.append({'id': node.id, 'instructions': dus})
return nodes
def getLinks(self):
links = []
# convert _nodes to representation needed by front end
for startAddr, node in self._nodes.items():
try:
if node.noAddr:
links.append({
'from': node.id,
'to': self._nodes[node.noAddr].id,
'type': 'notTaken'
})
links.append({
'from': node.id,
'to': self._nodes[node.yesAddr].id,
'type': 'taken'
})
elif node.yesAddr:
links.append({
'from': node.id,
'to': self._nodes[node.yesAddr].id,
'type': 'unconditional'
})
else:
# else, this is the last node in the function, which has no links
pass
except Exception as e:
# for now, keep going if something goes wrong
# AVD@9/8/15: An exception can happen right now if the last instruction in the function is a
# branch to somewhere in the middle of the function. The problem is that the "no"
# address is not within the function, since it is actually the first instruction of
# the next function. Need to revisit this and see if we can do better than just
# skipping this link.
continue
return links
nodes = property(getNodes)
links = property(getLinks)
def test_basic_active_scan(self):
# we load this ELF as a raw binary, because it has an invalid opcode fairly early on
odb_file = OdbFile(
BinaryFile(self.get_test_bin_path('mkdir'), 'elf64-x86-64',
'i386:x86-64'))
odb_file.execute(LoadOperation())
odb_file.execute(PassiveScanOperation())
parcels_before = ParcelList(odb_file.get_structure_list(Parcel))
p = parcels_before.find_parcel_by_vma(0x4002ce)
gnu_hash_lda_before = parcels_before.vma_to_lda(0x4002cd)
dynsym_lda_before = parcels_before.vma_to_lda(0x4002e0)
dynstr_lda_before = parcels_before.vma_to_lda(0x400a18)
# number of parcels before the split
self.assertEquals(24, len(parcels_before))
# execute the split operation
odb_file.execute(ActiveScanOperation(0x4002ce))
# number of parcels after the split
parcels_after = ParcelList(odb_file.get_structure_list(Parcel))
self.assertEquals(26, len(parcels_after))
# get the three parcels
p1 = parcels_after.find_parcel_by_vma(0x400298)
p2 = parcels_after.find_parcel_by_vma(0x4002ce)
p3 = parcels_after.find_parcel_by_vma(0x4002d5)
# make sure they are all unique
self.assertNotEqual(p1, p2)
self.assertNotEqual(p1, p3)
self.assertNotEqual(p2, p3)
# check num_ldas in p1 (54 bytes)
self.assertEquals(54, p1.num_ldas)
# check num_ldas in p2 (3 instructions)
self.assertEquals(3, p2.num_ldas)
# check num_ldas in p3 (7 bytes)
self.assertEquals(7, p3.num_ldas)
# make sure vmas line up
self.assertEquals(0x400298, p1.vma_start)
self.assertEquals(0x4002ce, p1.vma_end)
self.assertEquals(0x4002ce, p2.vma_start)
self.assertEquals(0x4002d5, p2.vma_end)
self.assertEquals(0x4002d5, p3.vma_start)
self.assertEquals(0x4002dc, p3.vma_end)
#
# make sure vma->lda mappings line up
#
# these shouldn't have changed
self.assertEquals(gnu_hash_lda_before,
parcels_after.vma_to_lda(0x4002cd))
self.assertEquals(gnu_hash_lda_before + 1,
parcels_after.vma_to_lda(0x4002ce))
self.assertEquals(False, p1.is_code)
self.assertEquals(False, p3.is_code)
# these changed
self.assertEquals(gnu_hash_lda_before + 2,
parcels_after.vma_to_lda(0x4002d0))
self.assertEquals(gnu_hash_lda_before + 3,
parcels_after.vma_to_lda(0x4002d2))
self.assertEquals(gnu_hash_lda_before + 4,
parcels_after.vma_to_lda(0x4002d5))
self.assertEquals(gnu_hash_lda_before + 5,
parcels_after.vma_to_lda(0x4002d6))
self.assertEquals(gnu_hash_lda_before + 6,
parcels_after.vma_to_lda(0x4002d7))
self.assertEquals(gnu_hash_lda_before + 7,
parcels_after.vma_to_lda(0x4002d8))
self.assertEquals(gnu_hash_lda_before + 8,
parcels_after.vma_to_lda(0x4002d9))
self.assertEquals(gnu_hash_lda_before + 9,
parcels_after.vma_to_lda(0x4002da))
self.assertEquals(gnu_hash_lda_before + 10,
parcels_after.vma_to_lda(0x4002db))
self.assertEquals(True, p2.is_code)
# 7 bytes turned into 3 instructions, so it shifted the following parcels by -4 ldas
self.assertEquals(dynsym_lda_before - 4,
parcels_after.vma_to_lda(0x4002e0))
self.assertEquals(dynstr_lda_before - 4,
parcels_after.vma_to_lda(0x400a18))
#
# make sure lda->vma mappings line up
#
# these shouldn't have changed
self.assertEquals(0x4002cd,
parcels_after.lda_to_vma(gnu_hash_lda_before))
self.assertEquals(0x4002ce,
parcels_after.lda_to_vma(gnu_hash_lda_before + 1))
# these changed
self.assertEquals(0x4002d0,
parcels_after.lda_to_vma(gnu_hash_lda_before + 2))
self.assertEquals(0x4002d2,
parcels_after.lda_to_vma(gnu_hash_lda_before + 3))
self.assertEquals(0x4002d5,
parcels_after.lda_to_vma(gnu_hash_lda_before + 4))
self.assertEquals(0x4002d6,
parcels_after.lda_to_vma(gnu_hash_lda_before + 5))
self.assertEquals(0x4002d7,
parcels_after.lda_to_vma(gnu_hash_lda_before + 6))
self.assertEquals(0x4002d8,
parcels_after.lda_to_vma(gnu_hash_lda_before + 7))
self.assertEquals(0x4002d9,
parcels_after.lda_to_vma(gnu_hash_lda_before + 8))
self.assertEquals(0x4002da,
parcels_after.lda_to_vma(gnu_hash_lda_before + 9))
self.assertEquals(0x4002db,
parcels_after.lda_to_vma(gnu_hash_lda_before + 10))
# 7 bytes turned into 3 instructions, so it shifted the following parcels by -4 ldas
self.assertEquals(0x4002e0,
parcels_after.lda_to_vma(dynsym_lda_before - 4))
self.assertEquals(0x400a18,
parcels_after.lda_to_vma(dynstr_lda_before - 4))